drm/i915/mtl: Define MOCS and PAT tables for MTL
cover drm/i915/mtl: Define MOCS and PAT tables for MTL
From: Fei Yang <[email protected]>
The series includes patches needed to enable MTL. Also add new extension for GEM_CREATE uAPI to let user space set cache policy for buffer objects.
v2: addressing review comments and checkpatch warnings
v3: make mtl_ggtt_pte_encode static
Fei Yang (7):
drm/i915/mtl: Set has_llc=0
drm/i915/mtl: Add PTE encode function drm/i915/mtl: workaround coherency issue for Media drm/i915/mtl: end support for set caching ioctl drm/i915: preparation for using PAT index drm/i915: use pat_index instead of cache_level drm/i915: Allow user to set cache at BO creation
Madhumitha Tolakanahalli Pradeep (1):
drm/i915/mtl: Define MOCS and PAT tables for MTL
drivers/gpu/drm/i915/display/intel_dpt.c | 14 ++-- drivers/gpu/drm/i915/gem/i915_gem_create.c | 36 ++++++++ drivers/gpu/drm/i915/gem/i915_gem_domain.c | 30 +++---- .../gpu/drm/i915/gem/i915_gem_execbuffer.c | 10 ++- drivers/gpu/drm/i915/gem/i915_gem_mman.c | 3 +- drivers/gpu/drm/i915/gem/i915_gem_object.c | 67 ++++++++++++++- drivers/gpu/drm/i915/gem/i915_gem_object.h | 8 ++ .../gpu/drm/i915/gem/i915_gem_object_types.h | 26 +++++- drivers/gpu/drm/i915/gem/i915_gem_pages.c | 5 +- drivers/gpu/drm/i915/gem/i915_gem_shmem.c | 9 +- drivers/gpu/drm/i915/gem/i915_gem_shrinker.c | 2 - drivers/gpu/drm/i915/gem/i915_gem_stolen.c | 4 +- drivers/gpu/drm/i915/gem/i915_gem_ttm_move.c | 16 ++-- .../gpu/drm/i915/gem/selftests/huge_pages.c | 2 +- .../drm/i915/gem/selftests/i915_gem_migrate.c | 2 +- .../drm/i915/gem/selftests/i915_gem_mman.c | 2 +- drivers/gpu/drm/i915/gt/gen6_ppgtt.c | 10 ++- drivers/gpu/drm/i915/gt/gen8_ppgtt.c | 76 ++++++++++++----- drivers/gpu/drm/i915/gt/gen8_ppgtt.h | 3 +- drivers/gpu/drm/i915/gt/intel_ggtt.c | 84 +++++++++++++------ drivers/gpu/drm/i915/gt/intel_gt_regs.h | 6 +- drivers/gpu/drm/i915/gt/intel_gtt.c | 47 ++++++++++- drivers/gpu/drm/i915/gt/intel_gtt.h | 38 ++++++--- drivers/gpu/drm/i915/gt/intel_migrate.c | 47 ++++++----- drivers/gpu/drm/i915/gt/intel_migrate.h | 13 ++- drivers/gpu/drm/i915/gt/intel_mocs.c | 76 ++++++++++++++++- drivers/gpu/drm/i915/gt/intel_ppgtt.c | 6 +- drivers/gpu/drm/i915/gt/selftest_migrate.c | 47 ++++++----- drivers/gpu/drm/i915/gt/selftest_mocs.c | 2 +- drivers/gpu/drm/i915/gt/selftest_reset.c | 8 +- drivers/gpu/drm/i915/gt/selftest_timeline.c | 2 +- drivers/gpu/drm/i915/gt/selftest_tlb.c | 4 +- drivers/gpu/drm/i915/gt/uc/intel_gsc_fw.c | 13 +++ drivers/gpu/drm/i915/gt/uc/intel_guc.c | 7 ++ drivers/gpu/drm/i915/gt/uc/intel_guc_ct.c | 6 ++ drivers/gpu/drm/i915/gt/uc/intel_uc_fw.c | 10 ++- drivers/gpu/drm/i915/i915_debugfs.c | 55 +++++++++--- drivers/gpu/drm/i915/i915_gem.c | 16 +++- drivers/gpu/drm/i915/i915_gpu_error.c | 8 +- drivers/gpu/drm/i915/i915_pci.c | 76 +++++++++++++++-- drivers/gpu/drm/i915/i915_vma.c | 16 ++-- drivers/gpu/drm/i915/i915_vma.h | 2 +- drivers/gpu/drm/i915/i915_vma_types.h | 2 - drivers/gpu/drm/i915/intel_device_info.h | 5 ++ drivers/gpu/drm/i915/selftests/i915_gem.c | 5 +- .../gpu/drm/i915/selftests/i915_gem_evict.c | 4 +- drivers/gpu/drm/i915/selftests/i915_gem_gtt.c | 15 ++-- .../drm/i915/selftests/intel_memory_region.c | 4 +- .../gpu/drm/i915/selftests/mock_gem_device.c | 9 ++ drivers/gpu/drm/i915/selftests/mock_gtt.c | 8 +- include/uapi/drm/i915_drm.h | 36 ++++++++ tools/include/uapi/drm/i915_drm.h | 36 ++++++++ 52 files changed, 812 insertions(+), 226 deletions(-)
Signature · 1 line
2.25.1
PATCH 1/8 drm/i915/mtl: Set has_llc=0
From: Fei Yang <[email protected]>
On MTL, LLC is not shared between GT and CPU, set has_llc=0.
--- drivers/gpu/drm/i915/i915_pci.c | 1 + 1 file changed, 1 insertion(+)
drivers/gpu/drm/i915/i915_pci.c +1 -0
@@ -1147,6 +1147,7 @@ static const struct intel_device_info mtl_info = { .has_flat_ccs = 0, .has_gmd_id = 1, .has_guc_deprivilege = 1, + .has_llc = 0, .has_mslice_steering = 0, .has_snoop = 1, .__runtime.memory_regions = REGION_SMEM | REGION_STOLEN_LMEM,
Signature · 1 line
2.25.1
Re: [Intel-gfx] [PATCH 1/8] drm/i915/mtl: Set has_llc=0
We have multiple bugs that requires this and it can be picked up irrespective of this series. I have sent a trybot patch for this and
once that passes, I will push this one.
https://patchwork.freedesktop.org/series/116746/
Nirmoy
Quoted 24 lines · On 4/20/2023 1:00 AM, [email protected] wrote:
From: Fei Yang <[email protected]>
On MTL, LLC is not shared between GT and CPU, set has_llc=0.
Signed-off-by Fei Yang <[email protected]>Reviewed-by Andi Shyti <[email protected]>Reviewed-by Andrzej Hajda <[email protected]>Reviewed-by Nirmoy Das <[email protected]>--- drivers/gpu/drm/i915/i915_pci.c | 1 + 1 file changed, 1 insertion(+)drivers/gpu/drm/i915/i915_pci.c +1
-0@@ -1147,6 +1147,7 @@ static const struct intel_device_info mtl_info = { .has_flat_ccs = 0, .has_gmd_id = 1, .has_guc_deprivilege = 1, + .has_llc = 0, .has_mslice_steering = 0, .has_snoop = 1, .__runtime.memory_regions = REGION_SMEM | REGION_STOLEN_LMEM,
PATCH 2/8 drm/i915/mtl: Define MOCS and PAT tables for MTL
In reply to [email protected]From: Madhumitha Tolakanahalli Pradeep <[email protected]>
On MTL, GT can no longer allocate on LLC - only the CPU can. This, along with addition of support for L4 cache calls for a MOCS/PAT table update.
Also the PAT index registers are multicasted for primary GT, and there is an address jump from index 7 to 8. This patch makes sure that these registers are programmed in the proper way.
BSpec: 44509, 45101, 44235
--- drivers/gpu/drm/i915/gt/intel_gt_regs.h | 6 +- drivers/gpu/drm/i915/gt/intel_gtt.c | 47 ++++++++++++++- drivers/gpu/drm/i915/gt/intel_gtt.h | 20 ++++++- drivers/gpu/drm/i915/gt/intel_mocs.c | 76 +++++++++++++++++++++++-- drivers/gpu/drm/i915/gt/selftest_mocs.c | 2 +- 5 files changed, 143 insertions(+), 8 deletions(-)
drivers/gpu/drm/i915/gt/intel_gt_regs.h +5 -1
@@ -356,7 +356,11 @@#define GEN7_TLB_RD_ADDR _MMIO(0x4700)#define GEN12_PAT_INDEX (index) _MMIO(0x4800 + (index) * 4) -#define XEHP_PAT_INDEX (index) MCR_REG(0x4800 + (index) * 4) +#define _PAT_INDEX (index) _PICK_EVEN_2RANGES(index, 8, \ + 0x4800, 0x4804, \ + 0x4848, 0x484c) +#define XEHP_PAT_INDEX (index) MCR_REG(_PAT_INDEX(index)) +#define XELPMP_PAT_INDEX (index) _MMIO(_PAT_INDEX(index))#define XEHP_TILE0_ADDR_RANGE MCR_REG(0x4900)#define XEHP_TILE_LMEM_RANGE_SHIFT 8
drivers/gpu/drm/i915/gt/intel_gtt.c +46 -1
59 lines not on this page · raw message
drivers/gpu/drm/i915/gt/intel_gtt.h +19 -1
@@ -88,9 +88,18 @@ typedef u64 gen8_pte_t;#define BYT_PTE_SNOOPED_BY_CPU_CACHES REG_BIT(2)#define BYT_PTE_WRITEABLE REG_BIT(1) +#define MTL_PPGTT_PTE_PAT3 BIT_ULL(62)#define GEN12_PPGTT_PTE_LM BIT_ULL(11) +#define GEN12_PPGTT_PTE_PAT2 BIT_ULL(7) +#define GEN12_PPGTT_PTE_NC BIT_ULL(5) +#define GEN12_PPGTT_PTE_PAT1 BIT_ULL(4) +#define GEN12_PPGTT_PTE_PAT0 BIT_ULL(3) -#define GEN12_GGTT_PTE_LM BIT_ULL(1) +#define GEN12_GGTT_PTE_LM BIT_ULL(1) +#define MTL_GGTT_PTE_PAT0 BIT_ULL(52) +#define MTL_GGTT_PTE_PAT1 BIT_ULL(53) +#define GEN12_GGTT_PTE_ADDR_MASK GENMASK_ULL(45, 12) +#define MTL_GGTT_PTE_PAT_MASK GENMASK_ULL(53, 52)#define GEN12_PDE_64K BIT(6)#define GEN12_PTE_PS64 BIT(8)
@@ -147,6 +156,15 @@ typedef u64 gen8_pte_t;#define GEN8_PDE_IPS_64K BIT(11)#define GEN8_PDE_PS_2M BIT(7) +#define MTL_PPAT_L4_CACHE_POLICY_MASK REG_GENMASK(3, 2) +#define MTL_PAT_INDEX_COH_MODE_MASK REG_GENMASK(1, 0) +#define MTL_PPAT_L4_3_UC REG_FIELD_PREP(MTL_PPAT_L4_CACHE_POLICY_MASK, 3) +#define MTL_PPAT_L4_1_WT REG_FIELD_PREP(MTL_PPAT_L4_CACHE_POLICY_MASK, 1) +#define MTL_PPAT_L4_0_WB REG_FIELD_PREP(MTL_PPAT_L4_CACHE_POLICY_MASK, 0) +#define MTL_3_COH_2W REG_FIELD_PREP(MTL_PAT_INDEX_COH_MODE_MASK, 3) +#define MTL_2_COH_1W REG_FIELD_PREP(MTL_PAT_INDEX_COH_MODE_MASK, 2) +#define MTL_0_COH_NON REG_FIELD_PREP(MTL_PAT_INDEX_COH_MODE_MASK, 0) +enum i915_cache_level ;struct drm_i915_gem_object ;
drivers/gpu/drm/i915/gt/intel_mocs.c +72 -4
119 lines not on this page · raw message
drivers/gpu/drm/i915/gt/selftest_mocs.c +1 -1
@@ -137,7 +137,7 @@ static int read_mocs_table(struct i915_request *rq,return 0 ;if (HAS_GLOBAL_MOCS_REGISTERS (rq->engine->i915)) - addr =global_mocs_offset (); + addr =global_mocs_offset (rq->engine->gt);else addr =mocs_offset (rq->engine);
Signature · 1 line
2.25.1
Quoted 122 lines · On Wed, Apr 19, 2023 at 04:00:52PM -0700, [email protected] wrote:
From: Madhumitha Tolakanahalli Pradeep <[email protected]>
On MTL, GT can no longer allocate on LLC - only the CPU can. This, along with addition of support for L4 cache calls for a MOCS/PAT table update.
Also the PAT index registers are multicasted for primary GT, and there is an address jump from index 7 to 8. This patch makes sure that these registers are programmed in the proper way.BSpec: 44509, 45101, 44235
Cc Matt Roper <[email protected]>Cc Lucas De Marchi <[email protected]>Signed-off-by Madhumitha Tolakanahalli Pradeep <[email protected]>Signed-off-by Aravind Iddamsetty <[email protected]>Signed-off-by Nirmoy Das <[email protected]>Signed-off-by Fei Yang <[email protected]>Reviewed-by Andrzej Hajda <[email protected]>Reviewed-by Nirmoy Das <[email protected]>Reviewed-by Andi Shyti <[email protected]>--- drivers/gpu/drm/i915/gt/intel_gt_regs.h | 6 +- drivers/gpu/drm/i915/gt/intel_gtt.c | 47 ++++++++++++++- drivers/gpu/drm/i915/gt/intel_gtt.h | 20 ++++++- drivers/gpu/drm/i915/gt/intel_mocs.c | 76 +++++++++++++++++++++++-- drivers/gpu/drm/i915/gt/selftest_mocs.c | 2 +- 5 files changed, 143 insertions(+), 8 deletions(-)drivers/gpu/drm/i915/gt/intel_gt_regs.h +5
-1@@ -356,7 +356,11 @@#define GEN7_TLB_RD_ADDR _MMIO(0x4700)#define GEN12_PAT_INDEX (index) _MMIO(0x4800 + (index) * 4) -#define XEHP_PAT_INDEX (index) MCR_REG(0x4800 + (index) * 4) +#define _PAT_INDEX (index) _PICK_EVEN_2RANGES(index, 8, \ + 0x4800, 0x4804, \ + 0x4848, 0x484c) +#define XEHP_PAT_INDEX (index) MCR_REG(_PAT_INDEX(index)) +#define XELPMP_PAT_INDEX (index) _MMIO(_PAT_INDEX(index))#define XEHP_TILE0_ADDR_RANGE MCR_REG(0x4900)#define XEHP_TILE_LMEM_RANGE_SHIFT 8drivers/gpu/drm/i915/gt/intel_gtt.c +46
-1@@ -468,6 +468,44 @@ void gtt_write_workarounds(struct intel_gt *gt) } } +static void xelpmp_setup_private_ppat (struct intel_uncore *uncore) +{ +intel_uncore_write (uncore,XELPMP_PAT_INDEX (0 ), +MTL_PPAT_L4_0_WB ); +intel_uncore_write (uncore,XELPMP_PAT_INDEX (1 ), +MTL_PPAT_L4_1_WT ); +intel_uncore_write (uncore,XELPMP_PAT_INDEX (2 ), +MTL_PPAT_L4_3_UC ); +intel_uncore_write (uncore,XELPMP_PAT_INDEX (3 ), +MTL_PPAT_L4_0_WB |MTL_2_COH_1W ); +intel_uncore_write (uncore,XELPMP_PAT_INDEX (4 ), +MTL_PPAT_L4_0_WB |MTL_3_COH_2W ); + +/* +* Remaining PAT entries are left at the hardware-default +* fully-cached setting +*/ +} + +static void xelpg_setup_private_ppat (struct intel_gt *gt) +{ +intel_gt_mcr_multicast_write (gt,XEHP_PAT_INDEX (0 ), +MTL_PPAT_L4_0_WB ); +intel_gt_mcr_multicast_write (gt,XEHP_PAT_INDEX (1 ), +MTL_PPAT_L4_1_WT ); +intel_gt_mcr_multicast_write (gt,XEHP_PAT_INDEX (2 ), +MTL_PPAT_L4_3_UC ); +intel_gt_mcr_multicast_write (gt,XEHP_PAT_INDEX (3 ), +MTL_PPAT_L4_0_WB |MTL_2_COH_1W ); +intel_gt_mcr_multicast_write (gt,XEHP_PAT_INDEX (4 ), +MTL_PPAT_L4_0_WB |MTL_3_COH_2W ); + +/* +* Remaining PAT entries are left at the hardware-default +* fully-cached setting +*/ +} +static void tgl_setup_private_ppat (struct intel_uncore *uncore) {/* TGL doesn't support LLC or AGE settings */ @@ -603,7 +641,14 @@ void setup_private_pat(struct intel_gt *gt)GEM_BUG_ON (GRAPHICS_VER (i915) <8 ); -if (GRAPHICS_VER_FULL (i915) >=IP_VER (12 ,50 )) +if (gt->type ==GT_MEDIA ) { +xelpmp_setup_private_ppat (gt->uncore); +return ; + } + +if (GRAPHICS_VER_FULL (i915) >=IP_VER (12 ,70 )) +xelpg_setup_private_ppat (gt); +else if (GRAPHICS_VER_FULL (i915) >=IP_VER (12 ,50 ))xehp_setup_private_ppat (gt);else if (GRAPHICS_VER (i915) >=12 )tgl_setup_private_ppat (uncore);drivers/gpu/drm/i915/gt/intel_gtt.h +2
-0@@ -88,9 +88,18 @@ typedef u64 gen8_pte_t;#define BYT_PTE_SNOOPED_BY_CPU_CACHES REG_BIT(2)#define BYT_PTE_WRITEABLE REG_BIT(1) +#define MTL_PPGTT_PTE_PAT3 BIT_ULL(62)#define GEN12_PPGTT_PTE_LM BIT_ULL(11) +#define GEN12_PPGTT_PTE_PAT2 BIT_ULL(7)
drivers/gpu/drm/i915/gt/intel_gtt.h typedef u64 gen8_pte_t; · line 102
Quoted 10 lines
drivers/gpu/drm/i915/gt/intel_gtt.h +8
-1+#define GEN12_PPGTT_PTE_NC BIT_ULL(5) +#define GEN12_PPGTT_PTE_PAT1 BIT_ULL(4) +#define GEN12_PPGTT_PTE_PAT0 BIT_ULL(3) -#define GEN12_GGTT_PTE_LM BIT_ULL(1) +#define GEN12_GGTT_PTE_LM BIT_ULL(1) +#define MTL_GGTT_PTE_PAT0 BIT_ULL(52) +#define MTL_GGTT_PTE_PAT1 BIT_ULL(53) +#define GEN12_GGTT_PTE_ADDR_MASK GENMASK_ULL(45, 12) +#define MTL_GGTT_PTE_PAT_MASK GENMASK_ULL(53, 52)
All these PTE bits should probably move to the next patch that deals with PTE encoding. They aren't related to the table settings being defined in this patch.
Quoted 5 lines
drivers/gpu/drm/i915/gt/intel_gtt.h +0
-0#define GEN12_PDE_64K BIT(6)#define GEN12_PTE_PS64 BIT(8)@@ -147,6 +156,15 @@ typedef u64 gen8_pte_t;#define GEN8_PDE_IPS_64K BIT(11)
drivers/gpu/drm/i915/gt/intel_gtt.h typedef u64 gen8_pte_t; · line 166
Quoted 10 lines
drivers/gpu/drm/i915/gt/intel_gtt.h +8
-0#define GEN8_PDE_PS_2M BIT(7) +#define MTL_PPAT_L4_CACHE_POLICY_MASK REG_GENMASK(3, 2) +#define MTL_PAT_INDEX_COH_MODE_MASK REG_GENMASK(1, 0) +#define MTL_PPAT_L4_3_UC REG_FIELD_PREP(MTL_PPAT_L4_CACHE_POLICY_MASK, 3) +#define MTL_PPAT_L4_1_WT REG_FIELD_PREP(MTL_PPAT_L4_CACHE_POLICY_MASK, 1) +#define MTL_PPAT_L4_0_WB REG_FIELD_PREP(MTL_PPAT_L4_CACHE_POLICY_MASK, 0) +#define MTL_3_COH_2W REG_FIELD_PREP(MTL_PAT_INDEX_COH_MODE_MASK, 3) +#define MTL_2_COH_1W REG_FIELD_PREP(MTL_PAT_INDEX_COH_MODE_MASK, 2) +#define MTL_0_COH_NON REG_FIELD_PREP(MTL_PAT_INDEX_COH_MODE_MASK, 0)
This last definition isn't used or needed.
Quoted 114 lines
drivers/gpu/drm/i915/gt/intel_gtt.h +1
-0+enum i915_cache_level ;struct drm_i915_gem_object ;drivers/gpu/drm/i915/gt/intel_mocs.c +69
-1@@ -40,6 +40,10 @@ struct drm_i915_mocs_table {#define LE_COS (value) ((value) << 15)#define LE_SSE (value) ((value) << 17) +/* Defines for the tables (GLOB_MOCS_0 - GLOB_MOCS_16) */ +#define _L4_CACHEABILITY (value) ((value) << 2) +#define IG_PAT (value) ((value) << 8) +/* Defines for the tables (LNCFMOCS0 - LNCFMOCS31) - two entries per word */ #define L3_ESC (value) ((value) << 0)#define L3_SCC (value) ((value) << 1)@@ -50,6 +54,7 @@ struct drm_i915_mocs_table {/* Helper defines */ #define GEN9_NUM_MOCS_ENTRIES 64/* 63-64 are reserved, but configured. */ #define PVC_NUM_MOCS_ENTRIES 3 +#define MTL_NUM_MOCS_ENTRIES 16/* (e)LLC caching options */ /* @@ -73,6 +78,12 @@ struct drm_i915_mocs_table {#define L3_2_RESERVED _L3_CACHEABILITY(2)#define L3_3_WB _L3_CACHEABILITY(3) +/* L4 caching options */ +#define L4_0_WB _L4_CACHEABILITY(0) +#define L4_1_WT _L4_CACHEABILITY(1) +#define L4_2_RESERVED _L4_CACHEABILITY(2) +#define L4_3_UC _L4_CACHEABILITY(3) +#define MOCS_ENTRY (__idx, __control_value, __l3cc_value) \ [__idx] = { \ .control_value = __control_value, \@@ -416,6 +427,57 @@ static const struct drm_i915_mocs_entry pvc_mocs_table[] = { MOCS_ENTRY(2, 0, L3_3_WB), }; +static const struct drm_i915_mocs_entry mtl_mocs_table[] = { +/* Error - Reserved for Non-Use */ +MOCS_ENTRY (0 , +IG_PAT (0 ), +L3_LKUP (1 ) |L3_3_WB ), +/* Cached - L3 + L4 */ +MOCS_ENTRY (1 , +IG_PAT (1 ), +L3_LKUP (1 ) |L3_3_WB ), +/* L4 - GO:L3 */ +MOCS_ENTRY (2 , +IG_PAT (1 ), +L3_LKUP (1 ) |L3_1_UC ), +/* Uncached - GO:L3 */ +MOCS_ENTRY (3 , +IG_PAT (1 ) |L4_3_UC , +L3_LKUP (1 ) |L3_1_UC ), +/* L4 - GO:Mem */ +MOCS_ENTRY (4 , +IG_PAT (1 ), +L3_LKUP (1 ) |L3_GLBGO (1 ) |L3_1_UC ), +/* Uncached - GO:Mem */ +MOCS_ENTRY (5 , +IG_PAT (1 ) |L4_3_UC , +L3_LKUP (1 ) |L3_GLBGO (1 ) |L3_1_UC ), +/* L4 - L3:NoLKUP; GO:L3 */ +MOCS_ENTRY (6 , +IG_PAT (1 ), +L3_1_UC ), +/* Uncached - L3:NoLKUP; GO:L3 */ +MOCS_ENTRY (7 , +IG_PAT (1 ) |L4_3_UC , +L3_1_UC ), +/* L4 - L3:NoLKUP; GO:Mem */ +MOCS_ENTRY (8 , +IG_PAT (1 ), +L3_GLBGO (1 ) |L3_1_UC ), +/* Uncached - L3:NoLKUP; GO:Mem */ +MOCS_ENTRY (9 , +IG_PAT (1 ) |L4_3_UC , +L3_GLBGO (1 ) |L3_1_UC ), +/* Display - L3; L4:WT */ +MOCS_ENTRY (14 , +IG_PAT (1 ) |L4_1_WT , +L3_LKUP (1 ) |L3_3_WB ), +/* CCS - Non-Displayable */ +MOCS_ENTRY (15 , +IG_PAT (1 ), +L3_GLBGO (1 ) |L3_1_UC ), +}; +enum {HAS_GLOBAL_MOCS =BIT (0 ),HAS_ENGINE_MOCS =BIT (1 ),@@ -445,7 +507,13 @@ static unsigned int get_mocs_settings(const struct drm_i915_private *i915,memset (table,0 ,sizeof (struct drm_i915_mocs_table )); table->unused_entries_index =I915_MOCS_PTE ; -if (IS_PONTEVECCHIO (i915)) { +if (IS_METEORLAKE (i915)) { + table->size =ARRAY_SIZE (mtl_mocs_table); + table->table = mtl_mocs_table; + table->n_entries =MTL_NUM_MOCS_ENTRIES ; + table->uc_index =9 ; + table->unused_entries_index =1 ; + }else if (IS_PONTEVECCHIO (i915)) { table->size =ARRAY_SIZE (pvc_mocs_table); table->table = pvc_mocs_table; table->n_entries =PVC_NUM_MOCS_ENTRIES ;
drivers/gpu/drm/i915/gt/intel_mocs.c void intel_mocs_init_engine(struct intel_engine_cs *engine) · line 719
Quoted 8 lines
drivers/gpu/drm/i915/gt/intel_mocs.c +2
-2@@ -646,9 +714,9 @@ void intel_mocs_init_engine(struct intel_engine_cs *engine)init_l3cc_table (engine->gt, &table); } -static u32 global_mocs_offset (void ) +static u32 global_mocs_offset (struct intel_gt *gt) { -return i915_mmio_reg_offset (GEN12_GLOBAL_MOCS (0 )); +return i915_mmio_reg_offset (GEN12_GLOBAL_MOCS (0 )) + gt->uncore->gsi_offset;
The main use of this function is as a parameter to __init_mocs_table(). The value ultimately gets used in intel_uncore_write_fw, which will already apply the GSI offset automatically; the extra addition here is unnecessary.
It seems this is an attempt to work around the secondary usage in a selftest (where the value is encoded into a MI_STORE_REGISTER_MEM_GEN8). Since the selftest is building register offsets into the GPU instructions it is generating, it would probably make more sense to move the GSI translation into the selftest itself so that it's clear what's happening and why.
Fixing the selftest should also probably be done in a separate patch. We can keep this patch focused on the primary goal of providing the new tables documented in the bspec.
Matt
Quoted 28 lines
drivers/gpu/drm/i915/gt/intel_mocs.c +1
-1}void intel_set_mocs_index (struct intel_gt *gt)@@ -671,7 +739,7 @@ void intel_mocs_init(struct intel_gt *gt)*/ flags =get_mocs_settings (gt->i915, &table);if (flags &HAS_GLOBAL_MOCS ) -__init_mocs_table (gt->uncore, &table,global_mocs_offset ()); +__init_mocs_table (gt->uncore, &table,global_mocs_offset (gt));/* * Initialize the L3CC table as part of mocs initalization to make drivers/gpu/drm/i915/gt/selftest_mocs.c +1
-1@@ -137,7 +137,7 @@ static int read_mocs_table(struct i915_request *rq,return 0 ;if (HAS_GLOBAL_MOCS_REGISTERS (rq->engine->i915)) - addr =global_mocs_offset (); + addr =global_mocs_offset (rq->engine->gt);else addr =mocs_offset (rq->engine);Signature · 1 line
2.25.1
Signature · 4 lines
Matt Roper Graphics Software Engineer Linux GPU Platform Enablement Intel Corporation
PATCH 3/8 drm/i915/mtl: Add PTE encode function
In reply to [email protected]From: Fei Yang <[email protected]>
PTE encode functions are platform dependent. This patch implements PTE functions for MTL, and ensures the correct PTE encode function is used by calling pte_encode function pointer instead of the hardcoded gen8 version of PTE encode.
--- drivers/gpu/drm/i915/display/intel_dpt.c | 2 +- drivers/gpu/drm/i915/gt/gen8_ppgtt.c | 45 ++++++++++++++++++++---- drivers/gpu/drm/i915/gt/intel_ggtt.c | 36 +++++++++++++++++-- 3 files changed, 72 insertions(+), 11 deletions(-)
drivers/gpu/drm/i915/display/intel_dpt.c +1 -1
@@ -300,7 +300,7 @@ intel_dpt_create(struct intel_framebuffer *fb) vm->vma_ops.bind_vma = dpt_bind_vma; vm->vma_ops.unbind_vma = dpt_unbind_vma; - vm->pte_encode = gen8_ggtt_pte_encode; + vm->pte_encode = vm->gt->ggtt->vm.pte_encode; dpt->obj = dpt_obj; dpt->obj->is_dpt = true;
drivers/gpu/drm/i915/gt/gen8_ppgtt.c +38 -7
87 lines not on this page · raw message
drivers/gpu/drm/i915/gt/intel_ggtt.c +33 -3
@@ -220,6 +220,33 @@ static void guc_ggtt_invalidate(struct i915_ggtt *ggtt) } } +static u64 mtl_ggtt_pte_encode (dma_addr_t addr, +enum i915_cache_level level, +u32 flags) +{ +gen8_pte_t pte = addr |GEN8_PAGE_PRESENT ; + +WARN_ON_ONCE (addr & ~GEN12_GGTT_PTE_ADDR_MASK ); + +if (flags &PTE_LM ) + pte |=GEN12_GGTT_PTE_LM ; + +switch (level) { +case I915_CACHE_NONE : + pte |=MTL_GGTT_PTE_PAT1 ; +break ; +case I915_CACHE_LLC : +case I915_CACHE_L3_LLC : + pte |=MTL_GGTT_PTE_PAT0 |MTL_GGTT_PTE_PAT1 ; +break ; +case I915_CACHE_WT : + pte |=MTL_GGTT_PTE_PAT0 ; +break ; + } + +return pte; +} +u64 gen8_ggtt_pte_encode (dma_addr_t addr,enum i915_cache_level level,u32 flags)
@@ -247,7 +274,7 @@ static void gen8_ggtt_insert_page(struct i915_address_space *vm,gen8_pte_t __iomem *pte = (gen8_pte_t __iomem *)ggtt->gsm + offset /I915_GTT_PAGE_SIZE ; -gen8_set_pte (pte,gen8_ggtt_pte_encode (addr, level, flags)); +gen8_set_pte (pte, ggtt->vm.pte_encode (addr, level, flags)); ggtt->invalidate (ggtt); }
@@ -257,8 +284,8 @@ static void gen8_ggtt_insert_entries(struct i915_address_space *vm, enum i915_cache_level level, u32 flags) { -const gen8_pte_t pte_encode =gen8_ggtt_pte_encode (0 , level, flags);struct i915_ggtt *ggtt =i915_vm_to_ggtt (vm); +const gen8_pte_t pte_encode = ggtt->vm.pte_encode (0 , level, flags);gen8_pte_t __iomem *gte;gen8_pte_t __iomem *end;struct sgt_iter iter;
@@ -981,7 +1008,10 @@ static int gen8_gmch_probe(struct i915_ggtt *ggtt) ggtt->vm.vma_ops.bind_vma = intel_ggtt_bind_vma; ggtt->vm.vma_ops.unbind_vma = intel_ggtt_unbind_vma; - ggtt->vm.pte_encode = gen8_ggtt_pte_encode; +if (GRAPHICS_VER_FULL (i915) >=IP_VER (12 ,70 )) + ggtt->vm.pte_encode = mtl_ggtt_pte_encode; +else + ggtt->vm.pte_encode = gen8_ggtt_pte_encode;return ggtt_probe_common (ggtt, size); }
Signature · 1 line
2.25.1
Re: [Intel-gfx] [PATCH 3/8] drm/i915/mtl: Add PTE encode function
Quoted 11 lines · On Wed, Apr 19, 2023 at 04:00:53PM -0700, [email protected] wrote:
From: Fei Yang <[email protected]>
PTE encode functions are platform dependent. This patch implements PTE functions for MTL, and ensures the correct PTE encode function is used by calling
pte_encodefunction pointer instead of the hardcoded gen8 version of PTE encode.Signed-off-by Fei Yang <[email protected]>Reviewed-by Andrzej Hajda <[email protected]>Reviewed-by Andi Shyti <[email protected]>Acked-by Nirmoy Das <[email protected]>
Bspec: 45015, 45040
Quoted 28 lines
--- drivers/gpu/drm/i915/display/intel_dpt.c | 2 +- drivers/gpu/drm/i915/gt/gen8_ppgtt.c | 45 ++++++++++++++++++++---- drivers/gpu/drm/i915/gt/intel_ggtt.c | 36 +++++++++++++++++-- 3 files changed, 72 insertions(+), 11 deletions(-)drivers/gpu/drm/i915/display/intel_dpt.c +1
-1@@ -300,7 +300,7 @@ intel_dpt_create(struct intel_framebuffer *fb) vm->vma_ops.bind_vma = dpt_bind_vma; vm->vma_ops.unbind_vma = dpt_unbind_vma; - vm->pte_encode = gen8_ggtt_pte_encode; + vm->pte_encode = vm->gt->ggtt->vm.pte_encode; dpt->obj = dpt_obj; dpt->obj->is_dpt = true;drivers/gpu/drm/i915/gt/gen8_ppgtt.c +1
-0@@ -55,6 +55,34 @@ static u64 gen8_pte_encode(dma_addr_t addr,return pte; } +static u64 mtl_pte_encode(dma_addr_t addr,
drivers/gpu/drm/i915/gt/gen8_ppgtt.c static u64 gen8_pte_encode(dma_addr_t addr, · line 68
Quoted 10 lines
drivers/gpu/drm/i915/gt/gen8_ppgtt.c +10
-0+ enum i915_cache_level level, + u32 flags) +{ +gen8_pte_t pte = addr |GEN8_PAGE_PRESENT |GEN8_PAGE_RW ; + +if (unlikely (flags &PTE_READ_ONLY )) + pte &= ~GEN8_PAGE_RW ; + +if (flags &PTE_LM ) + pte |=GEN12_PPGTT_PTE_LM |GEN12_PPGTT_PTE_NC ;
GEN12_PPGTT_PTE_NC got defined in the previous patch as BIT(5). But according to bspec 45040, bit 5 is ignored in the PTE encoding. What is this trying to do?
Matt
Quoted 151 lines
drivers/gpu/drm/i915/gt/gen8_ppgtt.c +27
-7+ +switch (level) { +case I915_CACHE_NONE : + pte |=GEN12_PPGTT_PTE_PAT1 ; +break ; +case I915_CACHE_LLC : +case I915_CACHE_L3_LLC : + pte |=GEN12_PPGTT_PTE_PAT0 |GEN12_PPGTT_PTE_PAT1 ; +break ; +case I915_CACHE_WT : + pte |=GEN12_PPGTT_PTE_PAT0 ; +break ; + } + +return pte; +} +static void gen8_ppgtt_notify_vgt (struct i915_ppgtt *ppgtt,bool create) {struct drm_i915_private *i915 = ppgtt->vm.i915;@@ -427,7 +455,7 @@ gen8_ppgtt_insert_pte(struct i915_ppgtt *ppgtt, u32 flags) {struct i915_page_directory *pd; -const gen8_pte_t pte_encode =gen8_pte_encode (0 , cache_level, flags); +const gen8_pte_t pte_encode = ppgtt->vm.pte_encode (0 , cache_level, flags);gen8_pte_t *vaddr; pd =i915_pd_entry (pdp,gen8_pd_index (idx,2 ));@@ -580,7 +608,7 @@ static void gen8_ppgtt_insert_huge(struct i915_address_space *vm, enum i915_cache_level cache_level, u32 flags) { -const gen8_pte_t pte_encode =gen8_pte_encode (0 , cache_level, flags); +const gen8_pte_t pte_encode = vm->pte_encode (0 , cache_level, flags);unsigned int rem =sg_dma_len (iter->sg);u64 start = vma_res->start;@@ -743,7 +771,7 @@ static void gen8_ppgtt_insert_entry(struct i915_address_space *vm,GEM_BUG_ON (pt->is_compact); vaddr =px_vaddr (pt); - vaddr[gen8_pd_index (idx,0 )] =gen8_pte_encode (addr, level, flags); + vaddr[gen8_pd_index (idx,0 )] = vm->pte_encode (addr, level, flags);drm_clflush_virt_range (&vaddr[gen8_pd_index (idx,0 )],sizeof (*vaddr)); }@@ -773,7 +801,7 @@ static void __xehpsdv_ppgtt_insert_entry_lm(struct i915_address_space *vm, } vaddr =px_vaddr (pt); - vaddr[gen8_pd_index (idx,0 ) /16 ] =gen8_pte_encode (addr, level, flags); + vaddr[gen8_pd_index (idx,0 ) /16 ] = vm->pte_encode (addr, level, flags); } static void xehpsdv_ppgtt_insert_entry(struct i915_address_space *vm,@@ -820,8 +848,8 @@ static int gen8_init_scratch(struct i915_address_space *vm) pte_flags |=PTE_LM ; vm->scratch[0 ]->encode = -gen8_pte_encode (px_dma (vm->scratch[0 ]), -I915_CACHE_NONE , pte_flags); + vm->pte_encode (px_dma (vm->scratch[0 ]), +I915_CACHE_NONE , pte_flags);for (i =1 ; i <= vm->top; i++) {struct drm_i915_gem_object *obj;@@ -963,7 +991,10 @@ struct i915_ppgtt *gen8_ppgtt_create(struct intel_gt *gt,*/ ppgtt->vm.alloc_scratch_dma = alloc_pt_dma; - ppgtt->vm.pte_encode = gen8_pte_encode; +if (GRAPHICS_VER_FULL (gt->i915) >=IP_VER (12 ,70 )) + ppgtt->vm.pte_encode = mtl_pte_encode; +else + ppgtt->vm.pte_encode = gen8_pte_encode; ppgtt->vm.bind_async_flags =I915_VMA_LOCAL_BIND ; ppgtt->vm.insert_entries = gen8_ppgtt_insert;drivers/gpu/drm/i915/gt/intel_ggtt.c +33
-3@@ -220,6 +220,33 @@ static void guc_ggtt_invalidate(struct i915_ggtt *ggtt) } } +static u64 mtl_ggtt_pte_encode (dma_addr_t addr, +enum i915_cache_level level, +u32 flags) +{ +gen8_pte_t pte = addr |GEN8_PAGE_PRESENT ; + +WARN_ON_ONCE (addr & ~GEN12_GGTT_PTE_ADDR_MASK ); + +if (flags &PTE_LM ) + pte |=GEN12_GGTT_PTE_LM ; + +switch (level) { +case I915_CACHE_NONE : + pte |=MTL_GGTT_PTE_PAT1 ; +break ; +case I915_CACHE_LLC : +case I915_CACHE_L3_LLC : + pte |=MTL_GGTT_PTE_PAT0 |MTL_GGTT_PTE_PAT1 ; +break ; +case I915_CACHE_WT : + pte |=MTL_GGTT_PTE_PAT0 ; +break ; + } + +return pte; +} +u64 gen8_ggtt_pte_encode (dma_addr_t addr,enum i915_cache_level level,u32 flags)@@ -247,7 +274,7 @@ static void gen8_ggtt_insert_page(struct i915_address_space *vm,gen8_pte_t __iomem *pte = (gen8_pte_t __iomem *)ggtt->gsm + offset /I915_GTT_PAGE_SIZE ; -gen8_set_pte (pte,gen8_ggtt_pte_encode (addr, level, flags)); +gen8_set_pte (pte, ggtt->vm.pte_encode (addr, level, flags)); ggtt->invalidate (ggtt); }@@ -257,8 +284,8 @@ static void gen8_ggtt_insert_entries(struct i915_address_space *vm, enum i915_cache_level level, u32 flags) { -const gen8_pte_t pte_encode =gen8_ggtt_pte_encode (0 , level, flags);struct i915_ggtt *ggtt =i915_vm_to_ggtt (vm); +const gen8_pte_t pte_encode = ggtt->vm.pte_encode (0 , level, flags);gen8_pte_t __iomem *gte;gen8_pte_t __iomem *end;struct sgt_iter iter;@@ -981,7 +1008,10 @@ static int gen8_gmch_probe(struct i915_ggtt *ggtt) ggtt->vm.vma_ops.bind_vma = intel_ggtt_bind_vma; ggtt->vm.vma_ops.unbind_vma = intel_ggtt_unbind_vma; - ggtt->vm.pte_encode = gen8_ggtt_pte_encode; +if (GRAPHICS_VER_FULL (i915) >=IP_VER (12 ,70 )) + ggtt->vm.pte_encode = mtl_ggtt_pte_encode; +else + ggtt->vm.pte_encode = gen8_ggtt_pte_encode;return ggtt_probe_common (ggtt, size); }Signature · 1 line
2.25.1
Signature · 4 lines
Matt Roper Graphics Software Engineer Linux GPU Platform Enablement Intel Corporation
Re: [Intel-gfx] [PATCH 3/8] drm/i915/mtl: Add PTE encode function
Quoted 42 lines · On Wed, Apr 19, 2023 at 04:00:53PM -0700, [email protected] wrote:
Quoted 11 lines
From: Fei Yang <[email protected]>
PTE encode functions are platform dependent. This patch implements PTE functions for MTL, and ensures the correct PTE encode function is used by calling
pte_encodefunction pointer instead of the hardcoded gen8 version of PTE encode.Signed-off-by Fei Yang <[email protected]>Reviewed-by Andrzej Hajda <[email protected]>Reviewed-by Andi Shyti <[email protected]>Acked-by Nirmoy Das <[email protected]>Bspec: 45015, 45040
Quoted 27 lines
--- drivers/gpu/drm/i915/display/intel_dpt.c | 2 +- drivers/gpu/drm/i915/gt/gen8_ppgtt.c | 45 ++++++++++++++++++++---- drivers/gpu/drm/i915/gt/intel_ggtt.c | 36 +++++++++++++++++-- 3 files changed, 72 insertions(+), 11 deletions(-)drivers/gpu/drm/i915/display/intel_dpt.c +1
-1@@ -300,7 +300,7 @@ intel_dpt_create(struct intel_framebuffer *fb) vm->vma_ops.bind_vma = dpt_bind_vma; vm->vma_ops.unbind_vma = dpt_unbind_vma; - vm->pte_encode = gen8_ggtt_pte_encode; + vm->pte_encode = vm->gt->ggtt->vm.pte_encode; dpt->obj = dpt_obj; dpt->obj->is_dpt = true;drivers/gpu/drm/i915/gt/gen8_ppgtt.c +0
-0@@ -55,6 +55,34 @@ static u64 gen8_pte_encode(dma_addr_t addr,return pte; }
Quoted 15 lines
drivers/gpu/drm/i915/gt/gen8_ppgtt.c static u64 gen8_pte_encode(dma_addr_t addr, · line 68
Quoted 11 lines
drivers/gpu/drm/i915/gt/gen8_ppgtt.c +11
-0+static u64 mtl_pte_encode (dma_addr_t addr, +enum i915_cache_level level, +u32 flags) +{ +gen8_pte_t pte = addr |GEN8_PAGE_PRESENT |GEN8_PAGE_RW ; + +if (unlikely (flags &PTE_READ_ONLY )) + pte &= ~GEN8_PAGE_RW ; + +if (flags &PTE_LM ) + pte |=GEN12_PPGTT_PTE_LM |GEN12_PPGTT_PTE_NC ;
GEN12_PPGTT_PTE_NCgot defined in the previous patch as BIT(5). But according to bspec 45040, bit 5 is ignored in the PTE encoding. What is this trying to do?
This takes effect only for PTE_LM, doesn't affect MTL. PTE_NC is needed for PVC (use of access counter).
I believe this function was writen based on the one for PVC. And this function did get extended to cover all gen12 in a later patch.
-Fei
Quoted 7 lines
Matt
Quoted 5 lines
drivers/gpu/drm/i915/gt/gen8_ppgtt.c +5
-0+ +switch (level) { +case I915_CACHE_NONE : + pte |=GEN12_PPGTT_PTE_PAT1 ; +break ;
Re: [Intel-gfx] [PATCH 3/8] drm/i915/mtl: Add PTE encode function
Quoted 59 lines · Fei
Quoted 15 lines · On Wed, Apr 19, 2023 at 04:00:53PM -0700, [email protected] wrote:
Quoted 11 lines
From: Fei Yang <[email protected]>
PTE encode functions are platform dependent. This patch implements PTE functions for MTL, and ensures the correct PTE encode function is used by calling
pte_encodefunction pointer instead of the hardcoded gen8 version of PTE encode.Signed-off-by Fei Yang <[email protected]>Reviewed-by Andrzej Hajda <[email protected]>Reviewed-by Andi Shyti <[email protected]>Acked-by Nirmoy Das <[email protected]>Bspec: 45015, 45040
Quoted 7 lines
Review
Quoted 7 lines
---
drivers/gpu/drm/i915/display/intel_dpt.c | 2 +-drivers/gpu/drm/i915/gt/gen8_ppgtt.c| 45 ++++++++++++++++++++----drivers/gpu/drm/i915/gt/intel_ggtt.c| 36 +++++++++++++++++--3 files changed, 72 insertions(+), 11 deletions(-)(quoted diff) +0
-0
b/drivers/gpu/drm/i915/display/intel_dpt.cQuoted 7 lines
Quoted 7 lines
index
b8027392144d..c5eacfdba1a5100644drivers/gpu/drm/i915/display/intel_dpt.c +0
-0@@ -300,7 +300,7 @@ intel_dpt_create(struct intel_framebuffer *fb) vm->vma_ops.bind_vma = dpt_bind_vma; vm->vma_ops.unbind_vma = dpt_unbind_vma;Quoted 6 lines
Review
Quoted 6 lines
drivers/gpu/drm/i915/display/intel_dpt.c +1
-1- vm->pte_encode = gen8_ggtt_pte_encode; + vm->pte_encode = vm->gt->ggtt->vm.pte_encode; dpt->obj = dpt_obj; dpt->obj->is_dpt = true;(quoted diff) +0
-0
b/drivers/gpu/drm/i915/gt/gen8_ppgtt.cQuoted 7 lines
Quoted 7 lines
index
4daaa6f55668..11b91e0453c8100644drivers/gpu/drm/i915/gt/gen8_ppgtt.c +0
-0@@ -55,6 +55,34 @@ static u64 gen8_pte_encode(dma_addr_t addr, return pte; }Quoted 15 lines
drivers/gpu/drm/i915/gt/gen8_ppgtt.c static u64 gen8_pte_encode(dma_addr_t addr, · line 68
Quoted 11 lines
drivers/gpu/drm/i915/gt/gen8_ppgtt.c +11
-0+static u64 mtl_pte_encode (dma_addr_t addr, + enum i915_cache_level level, + u32 flags) +{ + gen8_pte_t pte = addr | GEN8_PAGE_PRESENT | GEN8_PAGE_RW; + + if (unlikely(flags & PTE_READ_ONLY)) + pte &= ~GEN8_PAGE_RW; + + if (flags & PTE_LM) + pte |= GEN12_PPGTT_PTE_LM | GEN12_PPGTT_PTE_NC;
GEN12_PPGTT_PTE_NCgot defined in the previous patch as BIT(5). But according to bspec 45040, bit 5 is ignored in the PTE encoding. What is this trying to do?
Quoted 5 lines
This takes effect only for
PTE_LM, doesn't affect MTL.PTE_NCis needed for PVC (use of access counter).
I believe this function was writen based on the one for PVC. And this function
did get extended to cover all gen12 in a later patch.
Even though MTL doesn't have local memory, PTE_LM is supposed to be used on MTL for access to BAR2 stolen memory.
Matt
Quoted 8 lines
-Fei
> Matt
>
>> +
>> + switch (level) {
>> + case I915_CACHE_NONE:
>> + pte |= GEN12_PPGTT_PTE_PAT1;
>> + break;Signature · 4 lines
Matt Roper Graphics Software Engineer Linux GPU Platform Enablement Intel Corporation
RE: [Intel-gfx] [PATCH 3/8] drm/i915/mtl: Add PTE encode function
Quoted 61 lines · Fei
Quoted 60 lines
Quoted 15 lines · On Wed, Apr 19, 2023 at 04:00:53PM -0700, [email protected] wrote:
Quoted 11 lines
From: Fei Yang <[email protected]>
PTE encode functions are platform dependent. This patch implements PTE functions for MTL, and ensures the correct PTE encode function is used by calling
pte_encodefunction pointer instead of the hardcoded gen8 version of PTE encode.Signed-off-by Fei Yang <[email protected]>Reviewed-by Andrzej Hajda <[email protected]>Reviewed-by Andi Shyti <[email protected]>Acked-by Nirmoy Das <[email protected]>Bspec: 45015, 45040
Quoted 7 lines
Review
Quoted 7 lines
--- drivers/gpu/drm/i915/display/intel_dpt.c | 2 +- drivers/gpu/drm/i915/gt/gen8_ppgtt.c | 45 ++++++++++++++++++++---- drivers/gpu/drm/i915/gt/intel_ggtt.c | 36 +++++++++++++++++-- 3 files changed, 72 insertions(+), 11 deletions(-)(quoted diff) +0
-0
b/drivers/gpu/drm/i915/display/intel_dpt.cQuoted 36 lines
Quoted 22 lines
index
b8027392144d..c5eacfdba1a5100644drivers/gpu/drm/i915/display/intel_dpt.c +1
-1@@ -300,7 +300,7 @@ intel_dpt_create(struct intel_framebuffer *fb) vm->vma_ops.bind_vma = dpt_bind_vma; vm->vma_ops.unbind_vma = dpt_unbind_vma; - vm->pte_encode = gen8_ggtt_pte_encode; + vm->pte_encode = vm->gt->ggtt->vm.pte_encode; dpt->obj = dpt_obj; dpt->obj->is_dpt = true;(quoted diff) +0
-0b/drivers/gpu/drm/i915/gt/gen8_ppgtt.cindex
4daaa6f55668..11b91e0453c8100644drivers/gpu/drm/i915/gt/gen8_ppgtt.c +1
-0@@ -55,6 +55,34 @@ static u64 gen8_pte_encode(dma_addr_t addr,return pte; } +static u64 mtl_pte_encode(dma_addr_t addr,drivers/gpu/drm/i915/gt/gen8_ppgtt.c static u64 gen8_pte_encode(dma_addr_t addr, · line 68
Quoted 10 lines
drivers/gpu/drm/i915/gt/gen8_ppgtt.c +10
-0+ enum i915_cache_level level, + u32 flags) +{ +gen8_pte_t pte = addr |GEN8_PAGE_PRESENT |GEN8_PAGE_RW ; + +if (unlikely (flags &PTE_READ_ONLY )) + pte &= ~GEN8_PAGE_RW ; + +if (flags &PTE_LM ) + pte |=GEN12_PPGTT_PTE_LM |GEN12_PPGTT_PTE_NC ;
GEN12_PPGTT_PTE_NCgot defined in the previous patch as BIT(5). But according to bspec 45040, bit 5 is ignored in the PTE encoding. What is this trying to do?
Quoted 7 lines
Quoted 4 lines
This takes effect only for
PTE_LM, doesn't affect MTL.PTE_NCis needed for PVC (use of access counter).
I believe this function was writen based on the one for PVC. And this function did get extended to cover all gen12 in a later patch.Even though MTL doesn't have local memory,
PTE_LMis supposed to be used on MTL for access to BAR2 stolen memory.
You were right, but I still think this code is fine because this bit is ignored for MTL anyway and it is needed for other platforms with LMEM. Otherwise this code would have some sort of platform checking which is hard to do because we don't have platform info here.
Or we would have to define another PTE encode function for platforms needing PTE_NC just for this one difference, then manage the function pointer correctly.
-Fei
Quoted 10 lines
Matt
Quoted 8 lines
-Fei
Quoted 7 lines
Matt
Quoted 5 lines
drivers/gpu/drm/i915/gt/gen8_ppgtt.c +5
-0+ +switch (level) { +case I915_CACHE_NONE : + pte |=GEN12_PPGTT_PTE_PAT1 ; +break ;
Re: [Intel-gfx] [PATCH 3/8] drm/i915/mtl: Add PTE encode function
Quoted 69 lines · Fei
Quoted 61 lines · Fei
Quoted 60 lines
Quoted 15 lines · On Wed, Apr 19, 2023 at 04:00:53PM -0700, [email protected] wrote:
Quoted 11 lines
From: Fei Yang <[email protected]>
PTE encode functions are platform dependent. This patch implements PTE functions for MTL, and ensures the correct PTE encode function is used by calling
pte_encodefunction pointer instead of the hardcoded gen8 version of PTE encode.Signed-off-by Fei Yang <[email protected]>Reviewed-by Andrzej Hajda <[email protected]>Reviewed-by Andi Shyti <[email protected]>Acked-by Nirmoy Das <[email protected]>Bspec: 45015, 45040
Quoted 7 lines
Review
Quoted 7 lines
--- drivers/gpu/drm/i915/display/intel_dpt.c | 2 +- drivers/gpu/drm/i915/gt/gen8_ppgtt.c | 45 ++++++++++++++++++++---- drivers/gpu/drm/i915/gt/intel_ggtt.c | 36 +++++++++++++++++-- 3 files changed, 72 insertions(+), 11 deletions(-)(quoted diff) +0
-0
b/drivers/gpu/drm/i915/display/intel_dpt.cQuoted 36 lines
Quoted 22 lines
index
b8027392144d..c5eacfdba1a5100644drivers/gpu/drm/i915/display/intel_dpt.c +1
-1@@ -300,7 +300,7 @@ intel_dpt_create(struct intel_framebuffer *fb) vm->vma_ops.bind_vma = dpt_bind_vma; vm->vma_ops.unbind_vma = dpt_unbind_vma; - vm->pte_encode = gen8_ggtt_pte_encode; + vm->pte_encode = vm->gt->ggtt->vm.pte_encode; dpt->obj = dpt_obj; dpt->obj->is_dpt = true;(quoted diff) +0
-0b/drivers/gpu/drm/i915/gt/gen8_ppgtt.cindex
4daaa6f55668..11b91e0453c8100644drivers/gpu/drm/i915/gt/gen8_ppgtt.c +1
-0@@ -55,6 +55,34 @@ static u64 gen8_pte_encode(dma_addr_t addr,return pte; } +static u64 mtl_pte_encode(dma_addr_t addr,drivers/gpu/drm/i915/gt/gen8_ppgtt.c static u64 gen8_pte_encode(dma_addr_t addr, · line 68
Quoted 10 lines
drivers/gpu/drm/i915/gt/gen8_ppgtt.c +10
-0+ enum i915_cache_level level, + u32 flags) +{ +gen8_pte_t pte = addr |GEN8_PAGE_PRESENT |GEN8_PAGE_RW ; + +if (unlikely (flags &PTE_READ_ONLY )) + pte &= ~GEN8_PAGE_RW ; + +if (flags &PTE_LM ) + pte |=GEN12_PPGTT_PTE_LM |GEN12_PPGTT_PTE_NC ;
GEN12_PPGTT_PTE_NCgot defined in the previous patch as BIT(5). But according to bspec 45040, bit 5 is ignored in the PTE encoding. What is this trying to do?Quoted 7 lines
Quoted 4 lines
This takes effect only for
PTE_LM, doesn't affect MTL.PTE_NCis needed for PVC (use of access counter).
I believe this function was writen based on the one for PVC. And this function did get extended to cover all gen12 in a later patch.Even though MTL doesn't have local memory,
PTE_LMis supposed to be used on MTL for access to BAR2 stolen memory.
Quoted 7 lines
You were right, but I still think this code is fine because this bit is ignored for MTL anyway and it is needed for other platforms with LMEM. Otherwise this code would have some sort of platform checking which is hard to do because we don't have platform info here.
Or we would have to define another PTE encode function for platforms needingPTE_NCjust for this one difference, then manage the function pointer correctly.
MTL is the only platform that uses this function right now:
+ if (GRAPHICS_VER_FULL(gt->i915) >= IP_VER(12, 70))
+ ppgtt->vm.pte_encode = mtl_pte_encode;
+ else
+ ppgtt->vm.pte_encode = gen8_pte_encode;
If this is intended for PVC, then you have it in the wrong function to begin with (and it also shouldn't be in a patch labelled "mtl"). If you're trying to future-proof for some post-MTL discrete platform, then such code should be saved until we enable that platform so that it can be properly reviewed.
Matt
Quoted 13 lines
-Fei
Quoted 10 lines
Matt
Quoted 8 lines
-Fei
Quoted 7 lines
Matt
Quoted 5 lines
drivers/gpu/drm/i915/gt/gen8_ppgtt.c +5
-0+ +switch (level) { +case I915_CACHE_NONE : + pte |=GEN12_PPGTT_PTE_PAT1 ; +break ;
Signature · 4 lines
Matt Roper Graphics Software Engineer Linux GPU Platform Enablement Intel Corporation
Re: [Intel-gfx] [PATCH 3/8] drm/i915/mtl: Add PTE encode function
Quoted 85 lines · Fei
Quoted 69 lines
Quoted 61 lines · Fei
Quoted 60 lines
Quoted 15 lines · On Wed, Apr 19, 2023 at 04:00:53PM -0700, [email protected] wrote:
Quoted 11 lines
From: Fei Yang <[email protected]>
PTE encode functions are platform dependent. This patch implements PTE functions for MTL, and ensures the correct PTE encode function is used by calling
pte_encodefunction pointer instead of the hardcoded gen8 version of PTE encode.Signed-off-by Fei Yang <[email protected]>Reviewed-by Andrzej Hajda <[email protected]>Reviewed-by Andi Shyti <[email protected]>Acked-by Nirmoy Das <[email protected]>Bspec: 45015, 45040
Quoted 7 lines
Review
Quoted 7 lines
--- drivers/gpu/drm/i915/display/intel_dpt.c | 2 +- drivers/gpu/drm/i915/gt/gen8_ppgtt.c | 45 ++++++++++++++++++++---- drivers/gpu/drm/i915/gt/intel_ggtt.c | 36 +++++++++++++++++-- 3 files changed, 72 insertions(+), 11 deletions(-)(quoted diff) +0
-0
b/drivers/gpu/drm/i915/display/intel_dpt.cQuoted 36 lines
Quoted 22 lines
index
b8027392144d..c5eacfdba1a5100644drivers/gpu/drm/i915/display/intel_dpt.c +1
-1@@ -300,7 +300,7 @@ intel_dpt_create(struct intel_framebuffer *fb) vm->vma_ops.bind_vma = dpt_bind_vma; vm->vma_ops.unbind_vma = dpt_unbind_vma; - vm->pte_encode = gen8_ggtt_pte_encode; + vm->pte_encode = vm->gt->ggtt->vm.pte_encode; dpt->obj = dpt_obj; dpt->obj->is_dpt = true;(quoted diff) +0
-0b/drivers/gpu/drm/i915/gt/gen8_ppgtt.cindex
4daaa6f55668..11b91e0453c8100644drivers/gpu/drm/i915/gt/gen8_ppgtt.c +1
-0@@ -55,6 +55,34 @@ static u64 gen8_pte_encode(dma_addr_t addr,return pte; } +static u64 mtl_pte_encode(dma_addr_t addr,drivers/gpu/drm/i915/gt/gen8_ppgtt.c static u64 gen8_pte_encode(dma_addr_t addr, · line 68
Quoted 10 lines
drivers/gpu/drm/i915/gt/gen8_ppgtt.c +10
-0+ enum i915_cache_level level, + u32 flags) +{ +gen8_pte_t pte = addr |GEN8_PAGE_PRESENT |GEN8_PAGE_RW ; + +if (unlikely (flags &PTE_READ_ONLY )) + pte &= ~GEN8_PAGE_RW ; + +if (flags &PTE_LM ) + pte |=GEN12_PPGTT_PTE_LM |GEN12_PPGTT_PTE_NC ;
GEN12_PPGTT_PTE_NCgot defined in the previous patch as BIT(5). But according to bspec 45040, bit 5 is ignored in the PTE encoding. What is this trying to do?Quoted 7 lines
Quoted 4 lines
This takes effect only for
PTE_LM, doesn't affect MTL.PTE_NCis needed for PVC (use of access counter).
I believe this function was writen based on the one for PVC. And this function did get extended to cover all gen12 in a later patch.Even though MTL doesn't have local memory,
PTE_LMis supposed to be used on MTL for access to BAR2 stolen memory.Quoted 7 lines
You were right, but I still think this code is fine because this bit is ignored for MTL anyway and it is needed for other platforms with LMEM. Otherwise this code would have some sort of platform checking which is hard to do because we don't have platform info here.
Or we would have to define another PTE encode function for platforms needingPTE_NCjust for this one difference, then manage the function pointer correctly.MTL is the only platform that uses this function right now:
+ if (
GRAPHICS_VER_FULL(gt->i915)>=IP_VER(12, 70))
+ppgtt->vm.pte_encode=mtl_pte_encode;
+ else
+ppgtt->vm.pte_encode=gen8_pte_encode;
Quoted 5 lines
If this is intended for PVC, then you have it in the wrong function to begin with (and it also shouldn't be in a patch labelled "mtl"). If you're trying to future-proof for some post-MTL discrete platform, then such code should be saved until we enable that platform so that it can be properly reviewed.
dropped GEN12_PPGTT_PTE_NC bit in v2 of https://patchwork.freedesktop.org/series/116900/
Quoted 15 lines
Matt
Quoted 13 lines
-Fei
Quoted 10 lines
Matt
Quoted 8 lines
-Fei
Quoted 7 lines
Matt
Quoted 5 lines
drivers/gpu/drm/i915/gt/gen8_ppgtt.c +5
-0+ +switch (level) { +case I915_CACHE_NONE : + pte |=GEN12_PPGTT_PTE_PAT1 ; +break ;
PATCH 4/8 drm/i915/mtl: workaround coherency issue for Media
In reply to [email protected]From: Fei Yang <[email protected]>
This patch implements Wa_22016122933.
In MTL, memory writes initiated by Media tile update the whole cache line even for partial writes. This creates a coherency problem for cacheable memory if both CPU and GPU are writing data to different locations within a single cache line. CTB communication is impacted by this issue because the head and tail pointers are adjacent words within a cache line (see struct guc_ct_buffer_desc), where one is written by GuC and the other by the host. This patch circumvents the issue by making CPU/GPU shared memory uncacheable (WC on CPU side, and PAT index 2 for GPU). Also for CTB which is being updated by both CPU and GuC, mfence instruction is added to make sure the CPU writes are visible to GPU right away (flush the write combining buffer).
While fixing the CTB issue, we noticed some random GSC firmware loading failure because the share buffers are cacheable (WB) on CPU side but uncached on GPU side. To fix these issues we need to map such shared buffers as WC on CPU side. Since such allocations are not all done through GuC allocator, to avoid too many code changes, the i915_coherent_map_type() is now hard coded to return WC for MTL.
BSpec: 45101
--- drivers/gpu/drm/i915/gem/i915_gem_pages.c | 5 ++++- drivers/gpu/drm/i915/gt/uc/intel_gsc_fw.c | 13 +++++++++++++ drivers/gpu/drm/i915/gt/uc/intel_guc.c | 7 +++++++ drivers/gpu/drm/i915/gt/uc/intel_guc_ct.c | 6 ++++++ 4 files changed, 30 insertions(+), 1 deletion(-)
drivers/gpu/drm/i915/gem/i915_gem_pages.c +4 -1
@@ -469,7 +469,10 @@ enum i915_map_type i915_coherent_map_type(struct drm_i915_private *i915, struct drm_i915_gem_object *obj, bool always_coherent) { - if (i915_gem_object_is_lmem(obj)) +/* +* Wa_22016122933: always return I915_MAP_WC for MTL +*/ +if (i915_gem_object_is_lmem (obj) ||IS_METEORLAKE (i915))return I915_MAP_WC ;if (HAS_LLC (i915) || always_coherent)return I915_MAP_WB ;
drivers/gpu/drm/i915/gt/uc/intel_gsc_fw.c +13 -0
@@ -110,6 +110,13 @@ static int gsc_fw_load_prepare(struct intel_gsc_uc *gsc)if (obj->base.size < gsc->fw.size)return -ENOSPC ; +/* +* Wa_22016122933: For MTL the shared memory needs to be mapped +* as WC on CPU side and UC (PAT index 2) on GPU side +*/ +if (IS_METEORLAKE (i915)) +i915_gem_object_set_cache_coherency (obj,I915_CACHE_NONE ); + dst =i915_gem_object_pin_map_unlocked (obj,i915_coherent_map_type (i915, obj, true));if (IS_ERR (dst))
@@ -125,6 +132,12 @@ static int gsc_fw_load_prepare(struct intel_gsc_uc *gsc)memset (dst,0 , obj->base.size);memcpy (dst, src, gsc->fw.size); +/* +* Wa_22016122933: Making sure the data in dst is +* visible to GSC right away +*/ +intel_guc_write_barrier (>->uc.guc); +i915_gem_object_unpin_map (gsc->fw.obj);i915_gem_object_unpin_map (obj);
drivers/gpu/drm/i915/gt/uc/intel_guc.c +7 -0
@@ -744,6 +744,13 @@ struct i915_vma *intel_guc_allocate_vma(struct intel_guc *guc, u32 size)if (IS_ERR (obj))return ERR_CAST (obj); +/* +* Wa_22016122933: For MTL the shared memory needs to be mapped +* as WC on CPU side and UC (PAT index 2) on GPU side +*/ +if (IS_METEORLAKE (gt->i915)) +i915_gem_object_set_cache_coherency (obj,I915_CACHE_NONE ); + vma =i915_vma_instance (obj, >->ggtt->vm,NULL );if (IS_ERR (vma)) goto err;
drivers/gpu/drm/i915/gt/uc/intel_guc_ct.c +6 -0
@@ -902,6 +902,12 @@ static int ct_read(struct intel_guc_ct *ct, struct ct_incoming_msg **msg)/* now update descriptor */ WRITE_ONCE (desc->head, head); +/* +* Wa_22016122933: Making sure the head update is +* visible to GuC right away +*/ +intel_guc_write_barrier (ct_to_guc (ct)); +return available - len; corrupted:
Signature · 1 line
2.25.1
Re: [Intel-gfx] [PATCH 4/8] drm/i915/mtl: workaround coherency issue for Media
Quoted 24 lines · On 20.04.2023 01:00, [email protected] wrote:
From: Fei Yang <[email protected]>
This patch implements
Wa_22016122933.In MTL, memory writes initiated by Media tile update the whole cache line even for partial writes. This creates a coherency problem for cacheable memory if both CPU and GPU are writing data to different locations within a single cache line. CTB communication is impacted by this issue because the head and tail pointers are adjacent words within a cache line (see struct
guc_ct_buffer_desc), where one is written by GuC and the other by the host. This patch circumvents the issue by making CPU/GPU shared memory uncacheable (WC on CPU side, and PAT index 2 for GPU). Also for CTB which is being updated by both CPU and GuC, mfence instruction is added to make sure the CPU writes are visible to GPU right away (flush the write combining buffer).While fixing the CTB issue, we noticed some random GSC firmware loading failure because the share buffers are cacheable (WB) on CPU side but uncached on GPU side. To fix these issues we need to map such shared buffers as WC on CPU side. Since such allocations are not all done through GuC allocator, to avoid too many code changes, the
i915_coherent_map_type()is now hard coded to return WC for MTL.
Quoted 5 lines
BSpec: 45101
Signed-off-by Fei Yang <[email protected]>Reviewed-by Andi Shyti <[email protected]>Acked-by Nirmoy Das <[email protected]>
Regards
Andrzej
Quoted 89 lines
--- drivers/gpu/drm/i915/gem/i915_gem_pages.c | 5 ++++- drivers/gpu/drm/i915/gt/uc/intel_gsc_fw.c | 13 +++++++++++++ drivers/gpu/drm/i915/gt/uc/intel_guc.c | 7 +++++++ drivers/gpu/drm/i915/gt/uc/intel_guc_ct.c | 6 ++++++ 4 files changed, 30 insertions(+), 1 deletion(-)drivers/gpu/drm/i915/gem/i915_gem_pages.c +4
-1@@ -469,7 +469,10 @@ enum i915_map_type i915_coherent_map_type(struct drm_i915_private *i915, struct drm_i915_gem_object *obj, bool always_coherent) { - if (i915_gem_object_is_lmem(obj)) +/* +* Wa_22016122933: always return I915_MAP_WC for MTL +*/ +if (i915_gem_object_is_lmem (obj) ||IS_METEORLAKE (i915))return I915_MAP_WC ;if (HAS_LLC (i915) || always_coherent)return I915_MAP_WB ;drivers/gpu/drm/i915/gt/uc/intel_gsc_fw.c +13
-0@@ -110,6 +110,13 @@ static int gsc_fw_load_prepare(struct intel_gsc_uc *gsc)if (obj->base.size < gsc->fw.size)return -ENOSPC ; +/* +* Wa_22016122933: For MTL the shared memory needs to be mapped +* as WC on CPU side and UC (PAT index 2) on GPU side +*/ +if (IS_METEORLAKE (i915)) +i915_gem_object_set_cache_coherency (obj,I915_CACHE_NONE ); + dst =i915_gem_object_pin_map_unlocked (obj,i915_coherent_map_type (i915, obj, true));if (IS_ERR (dst))@@ -125,6 +132,12 @@ static int gsc_fw_load_prepare(struct intel_gsc_uc *gsc)memset (dst,0 , obj->base.size);memcpy (dst, src, gsc->fw.size); +/* +* Wa_22016122933: Making sure the data in dst is +* visible to GSC right away +*/ +intel_guc_write_barrier (>->uc.guc); +i915_gem_object_unpin_map (gsc->fw.obj);i915_gem_object_unpin_map (obj);drivers/gpu/drm/i915/gt/uc/intel_guc.c +7
-0@@ -744,6 +744,13 @@ struct i915_vma *intel_guc_allocate_vma(struct intel_guc *guc, u32 size)if (IS_ERR (obj))return ERR_CAST (obj); +/* +* Wa_22016122933: For MTL the shared memory needs to be mapped +* as WC on CPU side and UC (PAT index 2) on GPU side +*/ +if (IS_METEORLAKE (gt->i915)) +i915_gem_object_set_cache_coherency (obj,I915_CACHE_NONE ); + vma =i915_vma_instance (obj, >->ggtt->vm,NULL );if (IS_ERR (vma)) goto err;drivers/gpu/drm/i915/gt/uc/intel_guc_ct.c +6
-0@@ -902,6 +902,12 @@ static int ct_read(struct intel_guc_ct *ct, struct ct_incoming_msg **msg)/* now update descriptor */ WRITE_ONCE (desc->head, head); +/* +* Wa_22016122933: Making sure the head update is +* visible to GuC right away +*/ +intel_guc_write_barrier (ct_to_guc (ct)); +return available - len; corrupted:
Re: [Intel-gfx] [PATCH 4/8] drm/i915/mtl: workaround coherency issue for Media
In reply to [email protected]This is a important fix and can be pushed without depending on this series.
I will send this out to mailing list separately for CI.
Regards,
Nirmoy
Quoted 118 lines · On 4/20/2023 1:00 AM, [email protected] wrote:
From: Fei Yang <[email protected]>
This patch implements
Wa_22016122933.In MTL, memory writes initiated by Media tile update the whole cache line even for partial writes. This creates a coherency problem for cacheable memory if both CPU and GPU are writing data to different locations within a single cache line. CTB communication is impacted by this issue because the head and tail pointers are adjacent words within a cache line (see struct
guc_ct_buffer_desc), where one is written by GuC and the other by the host. This patch circumvents the issue by making CPU/GPU shared memory uncacheable (WC on CPU side, and PAT index 2 for GPU). Also for CTB which is being updated by both CPU and GuC, mfence instruction is added to make sure the CPU writes are visible to GPU right away (flush the write combining buffer).While fixing the CTB issue, we noticed some random GSC firmware loading failure because the share buffers are cacheable (WB) on CPU side but uncached on GPU side. To fix these issues we need to map such shared buffers as WC on CPU side. Since such allocations are not all done through GuC allocator, to avoid too many code changes, the
i915_coherent_map_type()is now hard coded to return WC for MTL.BSpec: 45101
Signed-off-by Fei Yang <[email protected]>Reviewed-by Andi Shyti <[email protected]>Acked-by Nirmoy Das <[email protected]>--- drivers/gpu/drm/i915/gem/i915_gem_pages.c | 5 ++++- drivers/gpu/drm/i915/gt/uc/intel_gsc_fw.c | 13 +++++++++++++ drivers/gpu/drm/i915/gt/uc/intel_guc.c | 7 +++++++ drivers/gpu/drm/i915/gt/uc/intel_guc_ct.c | 6 ++++++ 4 files changed, 30 insertions(+), 1 deletion(-)drivers/gpu/drm/i915/gem/i915_gem_pages.c +4
-1@@ -469,7 +469,10 @@ enum i915_map_type i915_coherent_map_type(struct drm_i915_private *i915, struct drm_i915_gem_object *obj, bool always_coherent) { - if (i915_gem_object_is_lmem(obj)) +/* +* Wa_22016122933: always return I915_MAP_WC for MTL +*/ +if (i915_gem_object_is_lmem (obj) ||IS_METEORLAKE (i915))return I915_MAP_WC ;if (HAS_LLC (i915) || always_coherent)return I915_MAP_WB ;drivers/gpu/drm/i915/gt/uc/intel_gsc_fw.c +13
-0@@ -110,6 +110,13 @@ static int gsc_fw_load_prepare(struct intel_gsc_uc *gsc)if (obj->base.size < gsc->fw.size)return -ENOSPC ; +/* +* Wa_22016122933: For MTL the shared memory needs to be mapped +* as WC on CPU side and UC (PAT index 2) on GPU side +*/ +if (IS_METEORLAKE (i915)) +i915_gem_object_set_cache_coherency (obj,I915_CACHE_NONE ); + dst =i915_gem_object_pin_map_unlocked (obj,i915_coherent_map_type (i915, obj, true));if (IS_ERR (dst))@@ -125,6 +132,12 @@ static int gsc_fw_load_prepare(struct intel_gsc_uc *gsc)memset (dst,0 , obj->base.size);memcpy (dst, src, gsc->fw.size); +/* +* Wa_22016122933: Making sure the data in dst is +* visible to GSC right away +*/ +intel_guc_write_barrier (>->uc.guc); +i915_gem_object_unpin_map (gsc->fw.obj);i915_gem_object_unpin_map (obj);drivers/gpu/drm/i915/gt/uc/intel_guc.c +7
-0@@ -744,6 +744,13 @@ struct i915_vma *intel_guc_allocate_vma(struct intel_guc *guc, u32 size)if (IS_ERR (obj))return ERR_CAST (obj); +/* +* Wa_22016122933: For MTL the shared memory needs to be mapped +* as WC on CPU side and UC (PAT index 2) on GPU side +*/ +if (IS_METEORLAKE (gt->i915)) +i915_gem_object_set_cache_coherency (obj,I915_CACHE_NONE ); + vma =i915_vma_instance (obj, >->ggtt->vm,NULL );if (IS_ERR (vma)) goto err;drivers/gpu/drm/i915/gt/uc/intel_guc_ct.c +6
-0@@ -902,6 +902,12 @@ static int ct_read(struct intel_guc_ct *ct, struct ct_incoming_msg **msg)/* now update descriptor */ WRITE_ONCE (desc->head, head); +/* +* Wa_22016122933: Making sure the head update is +* visible to GuC right away +*/ +intel_guc_write_barrier (ct_to_guc (ct)); +return available - len; corrupted:
Re: [Intel-gfx] [PATCH 4/8] drm/i915/mtl: workaround coherency issue for Media
In reply to [email protected]Quoted 4 lines · On Wed, Apr 19, 2023 at 04:00:54PM -0700, [email protected] wrote:
From: Fei Yang <[email protected]>
This patch implements
Wa_22016122933.
Quoted 12 lines
In MTL, memory writes initiated by Media tile update the whole cache line even for partial writes. This creates a coherency problem for cacheable memory if both CPU and GPU are writing data to different locations within a single cache line. CTB communication is impacted by this issue because the head and tail pointers are adjacent words within a cache line (see struct
guc_ct_buffer_desc), where one is written by GuC and the other by the host. This patch circumvents the issue by making CPU/GPU shared memory uncacheable (WC on CPU side, and PAT index 2 for GPU). Also for CTB which is being updated by both CPU and GuC, mfence instruction is added to make sure the CPU writes are visible to GPU right away (flush the write combining buffer).
Is this description accurate? This patch doesn't insert an mfence instruction itself, it just calls intel_guc_write_barrier(). On platforms like MTL that aren't using local memory, that issues a wmb() barrier, which I believe is implemented as an sfence, not mfence. You'd need to be doing a mb() call to get an mfence.
I think in general this level of explanation is unnecessary; you can just give a high-level description indicating that we force the write-combine buffer to be flushed and not give the low-level specifics of what instruction that translates to at the x86 level.
Aside from simplifying the commit message,
Quoted 105 lines
While fixing the CTB issue, we noticed some random GSC firmware loading failure because the share buffers are cacheable (WB) on CPU side but uncached on GPU side. To fix these issues we need to map such shared buffers as WC on CPU side. Since such allocations are not all done through GuC allocator, to avoid too many code changes, the
i915_coherent_map_type()is now hard coded to return WC for MTL.BSpec: 45101
Signed-off-by Fei Yang <[email protected]>Reviewed-by Andi Shyti <[email protected]>Acked-by Nirmoy Das <[email protected]>--- drivers/gpu/drm/i915/gem/i915_gem_pages.c | 5 ++++- drivers/gpu/drm/i915/gt/uc/intel_gsc_fw.c | 13 +++++++++++++ drivers/gpu/drm/i915/gt/uc/intel_guc.c | 7 +++++++ drivers/gpu/drm/i915/gt/uc/intel_guc_ct.c | 6 ++++++ 4 files changed, 30 insertions(+), 1 deletion(-)drivers/gpu/drm/i915/gem/i915_gem_pages.c +4
-1@@ -469,7 +469,10 @@ enum i915_map_type i915_coherent_map_type(struct drm_i915_private *i915, struct drm_i915_gem_object *obj, bool always_coherent) { - if (i915_gem_object_is_lmem(obj)) +/* +* Wa_22016122933: always return I915_MAP_WC for MTL +*/ +if (i915_gem_object_is_lmem (obj) ||IS_METEORLAKE (i915))return I915_MAP_WC ;if (HAS_LLC (i915) || always_coherent)return I915_MAP_WB ;drivers/gpu/drm/i915/gt/uc/intel_gsc_fw.c +13
-0@@ -110,6 +110,13 @@ static int gsc_fw_load_prepare(struct intel_gsc_uc *gsc)if (obj->base.size < gsc->fw.size)return -ENOSPC ; +/* +* Wa_22016122933: For MTL the shared memory needs to be mapped +* as WC on CPU side and UC (PAT index 2) on GPU side +*/ +if (IS_METEORLAKE (i915)) +i915_gem_object_set_cache_coherency (obj,I915_CACHE_NONE ); + dst =i915_gem_object_pin_map_unlocked (obj,i915_coherent_map_type (i915, obj, true));if (IS_ERR (dst))@@ -125,6 +132,12 @@ static int gsc_fw_load_prepare(struct intel_gsc_uc *gsc)memset (dst,0 , obj->base.size);memcpy (dst, src, gsc->fw.size); +/* +* Wa_22016122933: Making sure the data in dst is +* visible to GSC right away +*/ +intel_guc_write_barrier (>->uc.guc); +i915_gem_object_unpin_map (gsc->fw.obj);i915_gem_object_unpin_map (obj);drivers/gpu/drm/i915/gt/uc/intel_guc.c +7
-0@@ -744,6 +744,13 @@ struct i915_vma *intel_guc_allocate_vma(struct intel_guc *guc, u32 size)if (IS_ERR (obj))return ERR_CAST (obj); +/* +* Wa_22016122933: For MTL the shared memory needs to be mapped +* as WC on CPU side and UC (PAT index 2) on GPU side +*/ +if (IS_METEORLAKE (gt->i915)) +i915_gem_object_set_cache_coherency (obj,I915_CACHE_NONE ); + vma =i915_vma_instance (obj, >->ggtt->vm,NULL );if (IS_ERR (vma)) goto err;drivers/gpu/drm/i915/gt/uc/intel_guc_ct.c +6
-0@@ -902,6 +902,12 @@ static int ct_read(struct intel_guc_ct *ct, struct ct_incoming_msg **msg)/* now update descriptor */ WRITE_ONCE (desc->head, head); +/* +* Wa_22016122933: Making sure the head update is +* visible to GuC right away +*/ +intel_guc_write_barrier (ct_to_guc (ct)); +return available - len; corrupted:Signature · 1 line
2.25.1
Signature · 4 lines
Matt Roper Graphics Software Engineer Linux GPU Platform Enablement Intel Corporation
PATCH 5/8 drm/i915/mtl: end support for set caching ioctl
In reply to [email protected]From: Fei Yang <[email protected]>
The design is to keep Buffer Object's caching policy immutable through out its life cycle. This patch ends the support for set caching ioctl from MTL onward. While doing that we also set BO's to be 1-way coherent at creation time because GPU is no longer automatically snooping CPU cache. For UMD's need to fine tune the caching policy for BO's, a follow up patch will extend the GEM_CREATE uAPI to allow UMD's specify caching mode at BO creation time.
--- drivers/gpu/drm/i915/gem/i915_gem_domain.c | 3 +++ drivers/gpu/drm/i915/gem/i915_gem_shmem.c | 9 ++++++++- 2 files changed, 11 insertions(+), 1 deletion(-)
drivers/gpu/drm/i915/gem/i915_gem_domain.c +3 -0
@@ -337,6 +337,9 @@ int i915_gem_set_caching_ioctl(struct drm_device *dev, void *data,if (IS_DGFX (i915))return -ENODEV ; +if (GRAPHICS_VER_FULL (i915) >=IP_VER (12 ,70 )) +return -EOPNOTSUPP ; +switch (args->caching) {case I915_CACHING_NONE : level =I915_CACHE_NONE ;
drivers/gpu/drm/i915/gem/i915_gem_shmem.c +8 -1
@@ -601,7 +601,14 @@ static int shmem_object_init(struct intel_memory_region *mem, obj->write_domain =I915_GEM_DOMAIN_CPU ; obj->read_domains =I915_GEM_DOMAIN_CPU ; -if (HAS_LLC (i915)) +/* +* MTL doesn't snoop CPU cache by default for GPU access (namely +* 1-way coherency). However some UMD's are currently depending on +* that. Make 1-way coherent the default setting for MTL. A follow +* up patch will extend the GEM_CREATE uAPI to allow UMD's specify +* caching mode at BO creation time +*/ +if (HAS_LLC (i915) || (GRAPHICS_VER_FULL (i915) >=IP_VER (12 ,70 )))/* On some devices, we can have the GPU use the LLC (the CPU * cache) for about a 10% performance improvement * compared to uncached. Graphics requests other than
Signature · 1 line
2.25.1
Re: [Intel-gfx] [PATCH 5/8] drm/i915/mtl: end support for set caching ioctl
Quoted 9 lines · On Wed, Apr 19, 2023 at 04:00:55PM -0700, [email protected] wrote:
From: Fei Yang <[email protected]>
The design is to keep Buffer Object's caching policy immutable through out its life cycle. This patch ends the support for set caching ioctl from MTL onward. While doing that we also set BO's to be 1-way coherent at creation time because GPU is no longer automatically snooping CPU cache. For UMD's need to fine tune the caching policy for BO's, a follow up patch will extend the
GEM_CREATEuAPI to allow UMD's specify caching mode at BO creation time.
Nitpick: I don't think "UMD" is a term that anyone really uses outside of Intel. It's probably better to just say "userspace" instead of "UMD" since that's more accurate anyway.
Matt
Quoted 46 lines
Signed-off-by Fei Yang <[email protected]>Reviewed-by Andi Shyti <[email protected]>Reviewed-by Andrzej Hajda <[email protected]>--- drivers/gpu/drm/i915/gem/i915_gem_domain.c | 3 +++ drivers/gpu/drm/i915/gem/i915_gem_shmem.c | 9 ++++++++- 2 files changed, 11 insertions(+), 1 deletion(-)drivers/gpu/drm/i915/gem/i915_gem_domain.c +3
-0@@ -337,6 +337,9 @@ int i915_gem_set_caching_ioctl(struct drm_device *dev, void *data,if (IS_DGFX (i915))return -ENODEV ; +if (GRAPHICS_VER_FULL (i915) >=IP_VER (12 ,70 )) +return -EOPNOTSUPP ; +switch (args->caching) {case I915_CACHING_NONE : level =I915_CACHE_NONE ;drivers/gpu/drm/i915/gem/i915_gem_shmem.c +8
-1@@ -601,7 +601,14 @@ static int shmem_object_init(struct intel_memory_region *mem, obj->write_domain =I915_GEM_DOMAIN_CPU ; obj->read_domains =I915_GEM_DOMAIN_CPU ; -if (HAS_LLC (i915)) +/* +* MTL doesn't snoop CPU cache by default for GPU access (namely +* 1-way coherency). However some UMD's are currently depending on +* that. Make 1-way coherent the default setting for MTL. A follow +* up patch will extend the GEM_CREATE uAPI to allow UMD's specify +* caching mode at BO creation time +*/ +if (HAS_LLC (i915) || (GRAPHICS_VER_FULL (i915) >=IP_VER (12 ,70 )))/* On some devices, we can have the GPU use the LLC (the CPU * cache) for about a 10% performance improvement * compared to uncached. Graphics requests other than Signature · 1 line
2.25.1
Signature · 4 lines
Matt Roper Graphics Software Engineer Linux GPU Platform Enablement Intel Corporation
PATCH 6/8 drm/i915: preparation for using PAT index
In reply to [email protected]From: Fei Yang <[email protected]>
This patch is a preparation for replacing enum i915_cache_level with PAT index. Caching policy for buffer objects is set through the PAT index in PTE, the old i915_cache_level is not sufficient to represent all caching modes supported by the hardware.
Preparing the transition by adding some platform dependent data structures and helper functions to translate the cache_level to pat_index.
cachelevel_to_pat: a platform dependent array mapping cache_level to
pat_index.
max_pat_index: the maximum PAT index supported by the hardware. Needed for
validating the PAT index passed in from user space.
i915_gem_get_pat_index: function to convert cache_level to PAT index.
obj_to_i915(obj): macro moved to header file for wider usage.
I915_MAX_CACHE_LEVEL: upper bound of i915_cache_level for the
convenience of coding.
--- drivers/gpu/drm/i915/gem/i915_gem_object.c | 9 +++ drivers/gpu/drm/i915/gem/i915_gem_object.h | 4 + .../gpu/drm/i915/gem/i915_gem_object_types.h | 1 + drivers/gpu/drm/i915/gem/i915_gem_shrinker.c | 2 - drivers/gpu/drm/i915/gt/gen8_ppgtt.c | 6 ++ drivers/gpu/drm/i915/gt/intel_ggtt.c | 6 ++ drivers/gpu/drm/i915/i915_pci.c | 75 +++++++++++++++++-- drivers/gpu/drm/i915/intel_device_info.h | 5 ++ .../gpu/drm/i915/selftests/mock_gem_device.c | 9 +++ 9 files changed, 107 insertions(+), 10 deletions(-)
drivers/gpu/drm/i915/gem/i915_gem_object.c +9 -0
@@ -45,6 +45,15 @@ static struct kmem_cache *slab_objects;static const struct drm_gem_object_funcs i915_gem_object_funcs; +unsigned int i915_gem_get_pat_index (struct drm_i915_private *i915, +enum i915_cache_level level) +{ +if (drm_WARN_ON (&i915->drm, level >=I915_MAX_CACHE_LEVEL )) +return 0 ; + +return INTEL_INFO (i915)->cachelevel_to_pat[level]; +} +struct drm_i915_gem_object *i915_gem_object_alloc (void ) {struct drm_i915_gem_object *obj;
drivers/gpu/drm/i915/gem/i915_gem_object.h +4 -0
@@ -20,6 +20,8 @@enum intel_region_id ; +#define obj_to_i915 (obj__) to_i915((obj__)->base.dev) +static inline bool i915_gem_object_size_2big (u64 size) {struct drm_i915_gem_object *obj;
@@ -30,6 +32,8 @@ static inline bool i915_gem_object_size_2big(u64 size)return false; } +unsigned int i915_gem_get_pat_index (struct drm_i915_private *i915, +enum i915_cache_level level);void i915_gem_init__objects (struct drm_i915_private *i915);void i915_objects_module_exit (void );
drivers/gpu/drm/i915/gem/i915_gem_object_types.h +1 -0
@@ -194,6 +194,7 @@ enum i915_cache_level {* engine. */ I915_CACHE_WT , + I915_MAX_CACHE_LEVEL, }; enum i915_map_type {
drivers/gpu/drm/i915/gem/i915_gem_shrinker.c +0 -2
@@ -460,8 +460,6 @@ void i915_gem_shrinker_taints_mutex(struct drm_i915_private *i915,fs_reclaim_release (GFP_KERNEL ); } -#define obj_to_i915 (obj__) to_i915((obj__)->base.dev) -/** * i915_gem_object_make_unshrinkable - Hide the object from the shrinker. By * default all object types that support shrinking(see IS_SHRINKABLE), will also
drivers/gpu/drm/i915/gt/gen8_ppgtt.c +6 -0
@@ -78,6 +78,12 @@ static u64 mtl_pte_encode(dma_addr_t addr,case I915_CACHE_WT : pte |=GEN12_PPGTT_PTE_PAT0 ;break ; +default : +/* This should never happen. Added to deal with the compile +* error due to the addition of I915_MAX_CACHE_LEVEL. Will +* be removed by the pat_index patch. +*/ +break ; }return pte;
drivers/gpu/drm/i915/gt/intel_ggtt.c +6 -0
@@ -242,6 +242,12 @@ static u64 mtl_ggtt_pte_encode(dma_addr_t addr,case I915_CACHE_WT : pte |=MTL_GGTT_PTE_PAT0 ;break ; +default : +/* This should never happen. Added to deal with the compile +* error due to the addition of I915_MAX_CACHE_LEVEL. Will +* be removed by the pat_index patch. +*/ +break ; }return pte;
drivers/gpu/drm/i915/i915_pci.c +67 -8
225 lines not on this page · raw message
drivers/gpu/drm/i915/intel_device_info.h +5 -0
@@ -35,6 +35,8 @@#include "gt/intel_context_types.h" #include "gt/intel_sseu.h" +#include "gem/i915_gem_object_types.h" +struct drm_printer ;struct drm_i915_private ;struct intel_gt_definition ;
@@ -308,6 +310,9 @@ struct intel_device_info {* Initial runtime info. Do not access outside of i915_driver_create(). */ const struct intel_runtime_info __runtime; + +u32 cachelevel_to_pat[I915_MAX_CACHE_LEVEL ]; +u32 max_pat_index; };struct intel_driver_caps {
drivers/gpu/drm/i915/selftests/mock_gem_device.c +9 -0
@@ -123,7 +123,9 @@ struct drm_i915_private *mock_gem_device(void)static struct dev_iommu fake_iommu = { .priv = (void *)-1 };#endif struct drm_i915_private *i915; +struct intel_device_info *i915_info;struct pci_dev *pdev; +unsigned int i;int ret; pdev =kzalloc (sizeof (*pdev),GFP_KERNEL );
@@ -180,6 +182,13 @@ struct drm_i915_private *mock_gem_device(void)I915_GTT_PAGE_SIZE_2M ;RUNTIME_INFO (i915)->memory_regions =REGION_SMEM ; + +/* simply use legacy cache level for mock device */ + i915_info = (struct intel_device_info *)INTEL_INFO (i915); + i915_info->max_pat_index =3 ; +for (i =0 ; i <I915_MAX_CACHE_LEVEL ; i++) + i915_info->cachelevel_to_pat[i] = i; +intel_memory_regions_hw_probe (i915);spin_lock_init (&i915->gpu_error.lock);
Signature · 1 line
2.25.1
Re: [Intel-gfx] [PATCH 6/8] drm/i915: preparation for using PAT index
Quoted 65 lines · On 20.04.2023 01:00, [email protected] wrote:
From: Fei Yang <[email protected]>
This patch is a preparation for replacing enum
i915_cache_levelwith PAT index. Caching policy for buffer objects is set through the PAT index in PTE, the oldi915_cache_levelis not sufficient to represent all caching modes supported by the hardware.Preparing the transition by adding some platform dependent data structures and helper functions to translate the
cache_leveltopat_index.
cachelevel_to_pat: a platform dependent array mappingcache_levelto
pat_index.
max_pat_index: the maximum PAT index supported by the hardware. Needed for
validating the PAT index passed in from user space.
i915_gem_get_pat_index: function to convertcache_levelto PAT index.
obj_to_i915(obj): macro moved to header file for wider usage.
I915_MAX_CACHE_LEVEL: upper bound ofi915_cache_levelfor the
convenience of coding.Cc Chris Wilson <[email protected]>Cc Matt Roper <[email protected]>Cc Andi Shyti <[email protected]>Signed-off-by Fei Yang <[email protected]>Reviewed-by Andi Shyti <[email protected]>--- drivers/gpu/drm/i915/gem/i915_gem_object.c | 9 +++ drivers/gpu/drm/i915/gem/i915_gem_object.h | 4 + .../gpu/drm/i915/gem/i915_gem_object_types.h | 1 + drivers/gpu/drm/i915/gem/i915_gem_shrinker.c | 2 - drivers/gpu/drm/i915/gt/gen8_ppgtt.c | 6 ++ drivers/gpu/drm/i915/gt/intel_ggtt.c | 6 ++ drivers/gpu/drm/i915/i915_pci.c | 75 +++++++++++++++++-- drivers/gpu/drm/i915/intel_device_info.h | 5 ++ .../gpu/drm/i915/selftests/mock_gem_device.c | 9 +++ 9 files changed, 107 insertions(+), 10 deletions(-)drivers/gpu/drm/i915/gem/i915_gem_object.c +9
-0@@ -45,6 +45,15 @@ static struct kmem_cache *slab_objects;static const struct drm_gem_object_funcs i915_gem_object_funcs; +unsigned int i915_gem_get_pat_index (struct drm_i915_private *i915, +enum i915_cache_level level) +{ +if (drm_WARN_ON (&i915->drm, level >=I915_MAX_CACHE_LEVEL )) +return 0 ; + +return INTEL_INFO (i915)->cachelevel_to_pat[level]; +} +struct drm_i915_gem_object *i915_gem_object_alloc (void ) {struct drm_i915_gem_object *obj;drivers/gpu/drm/i915/gem/i915_gem_object.h +0
-0
drivers/gpu/drm/i915/gem/i915_gem_object.h line 24
Quoted 5 lines
drivers/gpu/drm/i915/gem/i915_gem_object.h +2
-0@@ -20,6 +20,8 @@enum intel_region_id ; +#define obj_to_i915 (obj__) to_i915((obj__)->base.dev) +
while moved could be replaced by inline, up to you
Quoted 356 lines
drivers/gpu/drm/i915/gem/i915_gem_object.h +2
-0static inline bool i915_gem_object_size_2big (u64 size) {struct drm_i915_gem_object *obj;@@ -30,6 +32,8 @@ static inline bool i915_gem_object_size_2big(u64 size)return false; } +unsigned int i915_gem_get_pat_index (struct drm_i915_private *i915, +enum i915_cache_level level);void i915_gem_init__objects (struct drm_i915_private *i915);void i915_objects_module_exit (void );drivers/gpu/drm/i915/gem/i915_gem_object_types.h +1
-0@@ -194,6 +194,7 @@ enum i915_cache_level {* engine. */ I915_CACHE_WT , + I915_MAX_CACHE_LEVEL, }; enum i915_map_type {drivers/gpu/drm/i915/gem/i915_gem_shrinker.c +0
-2@@ -460,8 +460,6 @@ void i915_gem_shrinker_taints_mutex(struct drm_i915_private *i915,fs_reclaim_release (GFP_KERNEL ); } -#define obj_to_i915 (obj__) to_i915((obj__)->base.dev) -/** * i915_gem_object_make_unshrinkable - Hide the object from the shrinker. By * default all object types that support shrinking(see IS_SHRINKABLE), will also drivers/gpu/drm/i915/gt/gen8_ppgtt.c +6
-0@@ -78,6 +78,12 @@ static u64 mtl_pte_encode(dma_addr_t addr,case I915_CACHE_WT : pte |=GEN12_PPGTT_PTE_PAT0 ;break ; +default : +/* This should never happen. Added to deal with the compile +* error due to the addition of I915_MAX_CACHE_LEVEL. Will +* be removed by the pat_index patch. +*/ +break ; }return pte;drivers/gpu/drm/i915/gt/intel_ggtt.c +6
-0@@ -242,6 +242,12 @@ static u64 mtl_ggtt_pte_encode(dma_addr_t addr,case I915_CACHE_WT : pte |=MTL_GGTT_PTE_PAT0 ;break ; +default : +/* This should never happen. Added to deal with the compile +* error due to the addition of I915_MAX_CACHE_LEVEL. Will +* be removed by the pat_index patch. +*/ +break ; }return pte;drivers/gpu/drm/i915/i915_pci.c +67
-8@@ -30,6 +30,7 @@#include "display/intel_display_driver.h" #include "gt/intel_gt_regs.h" #include "gt/intel_sa_media.h" +#include "gem/i915_gem_object_types.h" #include "i915_driver.h" #include "i915_drv.h" @@ -164,6 +165,38 @@ .gamma_lut_tests = DRM_COLOR_LUT_NON_DECREASING, \ } +#define LEGACY_CACHELEVEL \ + .cachelevel_to_pat = { \ + [I915_CACHE_NONE] = 0, \ + [I915_CACHE_LLC] = 1, \ + [I915_CACHE_L3_LLC] = 2, \ + [I915_CACHE_WT] = 3, \ + } + +#define TGL_CACHELEVEL \ + .cachelevel_to_pat = { \ + [I915_CACHE_NONE] = 3, \ + [I915_CACHE_LLC] = 0, \ + [I915_CACHE_L3_LLC] = 0, \ + [I915_CACHE_WT] = 2, \ + } + +#define PVC_CACHELEVEL \ + .cachelevel_to_pat = { \ + [I915_CACHE_NONE] = 0, \ + [I915_CACHE_LLC] = 3, \ + [I915_CACHE_L3_LLC] = 3, \ + [I915_CACHE_WT] = 2, \ + } + +#define MTL_CACHELEVEL \ + .cachelevel_to_pat = { \ + [I915_CACHE_NONE] = 2, \ + [I915_CACHE_LLC] = 3, \ + [I915_CACHE_L3_LLC] = 3, \ + [I915_CACHE_WT] = 1, \ + } +/* Keep in gen based order, and chronological order within a gen */ #define GEN_DEFAULT_PAGE_SIZES \@@ -189,11 +222,13 @@ .has_snoop = true, \ .has_coherent_ggtt = false, \ .dma_mask_size = 32, \ + .max_pat_index =3 , \I9XX_PIPE_OFFSETS , \I9XX_CURSOR_OFFSETS , \I9XX_COLORS , \GEN_DEFAULT_PAGE_SIZES , \ - GEN_DEFAULT_REGIONS +GEN_DEFAULT_REGIONS , \ + LEGACY_CACHELEVEL#define I845_FEATURES \ GEN(2), \@@ -210,11 +245,13 @@ .has_snoop = true, \ .has_coherent_ggtt = false, \ .dma_mask_size = 32, \ + .max_pat_index =3 , \I845_PIPE_OFFSETS , \I845_CURSOR_OFFSETS , \I845_COLORS , \GEN_DEFAULT_PAGE_SIZES , \ - GEN_DEFAULT_REGIONS +GEN_DEFAULT_REGIONS , \ + LEGACY_CACHELEVELstatic const struct intel_device_info i830_info = {I830_FEATURES ,@@ -249,11 +286,13 @@ static const struct intel_device_info i865g_info = { .has_snoop = true, \ .has_coherent_ggtt = true, \ .dma_mask_size = 32, \ + .max_pat_index =3 , \I9XX_PIPE_OFFSETS , \I9XX_CURSOR_OFFSETS , \I9XX_COLORS , \GEN_DEFAULT_PAGE_SIZES , \ - GEN_DEFAULT_REGIONS +GEN_DEFAULT_REGIONS , \ + LEGACY_CACHELEVELstatic const struct intel_device_info i915g_info = {GEN3_FEATURES ,@@ -341,11 +380,13 @@ static const struct intel_device_info pnv_m_info = { .has_snoop = true, \ .has_coherent_ggtt = true, \ .dma_mask_size = 36, \ + .max_pat_index =3 , \I9XX_PIPE_OFFSETS , \I9XX_CURSOR_OFFSETS , \I9XX_COLORS , \GEN_DEFAULT_PAGE_SIZES , \ - GEN_DEFAULT_REGIONS +GEN_DEFAULT_REGIONS , \ + LEGACY_CACHELEVELstatic const struct intel_device_info i965g_info = {GEN4_FEATURES ,@@ -395,11 +436,13 @@ static const struct intel_device_info gm45_info = {/* ilk does support rc6, but we do not implement [power] contexts */ \ .has_rc6 = 0, \ .dma_mask_size = 36, \ + .max_pat_index =3 , \I9XX_PIPE_OFFSETS , \I9XX_CURSOR_OFFSETS , \ILK_COLORS , \GEN_DEFAULT_PAGE_SIZES , \ -GEN_DEFAULT_REGIONS +GEN_DEFAULT_REGIONS , \ + LEGACY_CACHELEVELstatic const struct intel_device_info ilk_d_info = {GEN5_FEATURES ,@@ -429,13 +472,15 @@ static const struct intel_device_info ilk_m_info = { .has_rc6p = 0, \ .has_rps = true, \ .dma_mask_size = 40, \ + .max_pat_index = 3, \ .__runtime.ppgtt_type = INTEL_PPGTT_ALIASING, \ .__runtime.ppgtt_size =31 , \I9XX_PIPE_OFFSETS , \I9XX_CURSOR_OFFSETS , \ILK_COLORS , \GEN_DEFAULT_PAGE_SIZES , \ - GEN_DEFAULT_REGIONS +GEN_DEFAULT_REGIONS , \ + LEGACY_CACHELEVEL#define SNB_D_PLATFORM \ GEN6_FEATURES, \@@ -482,13 +527,15 @@ static const struct intel_device_info snb_m_gt2_info = { .has_reset_engine = true, \ .has_rps = true, \ .dma_mask_size = 40, \ + .max_pat_index = 3, \ .__runtime.ppgtt_type = INTEL_PPGTT_ALIASING, \ .__runtime.ppgtt_size =31 , \IVB_PIPE_OFFSETS , \IVB_CURSOR_OFFSETS , \IVB_COLORS , \GEN_DEFAULT_PAGE_SIZES , \ - GEN_DEFAULT_REGIONS +GEN_DEFAULT_REGIONS , \ + LEGACY_CACHELEVEL#define IVB_D_PLATFORM \ GEN7_FEATURES, \@@ -542,6 +589,7 @@ static const struct intel_device_info vlv_info = { .display.has_gmch = 1, .display.has_hotplug = 1, .dma_mask_size = 40, + .max_pat_index = 3, .__runtime.ppgtt_type = INTEL_PPGTT_ALIASING, .__runtime.ppgtt_size = 31, .has_snoop = true,@@ -553,6 +601,7 @@ static const struct intel_device_info vlv_info = {I9XX_COLORS ,GEN_DEFAULT_PAGE_SIZES ,GEN_DEFAULT_REGIONS , + LEGACY_CACHELEVEL, };#define G75_FEATURES \@@ -640,6 +689,7 @@ static const struct intel_device_info chv_info = { .has_logical_ring_contexts = 1, .display.has_gmch = 1, .dma_mask_size = 39, + .max_pat_index = 3, .__runtime.ppgtt_type = INTEL_PPGTT_FULL, .__runtime.ppgtt_size = 32, .has_reset_engine = 1,@@ -651,6 +701,7 @@ static const struct intel_device_info chv_info = {CHV_COLORS ,GEN_DEFAULT_PAGE_SIZES ,GEN_DEFAULT_REGIONS , + LEGACY_CACHELEVEL, };#define GEN9_DEFAULT_PAGE_SIZES \@@ -890,9 +941,11 @@ static const struct intel_device_info jsl_info = { [TRANSCODER_DSI_1 ] =TRANSCODER_DSI1_OFFSET , \ }, \TGL_CURSOR_OFFSETS , \ +TGL_CACHELEVEL , \ .has_global_mocs = 1, \ .has_pxp = 1, \ - .display.has_dsb = 1 + .display.has_dsb = 1, \ + .max_pat_index = 3static const struct intel_device_info tgl_info = {GEN12_FEATURES ,@@ -1014,6 +1067,7 @@ static const struct intel_device_info adl_p_info = { .__runtime.graphics.ip.ver = 12, \ .__runtime.graphics.ip.rel =50 , \XE_HP_PAGE_SIZES , \ +TGL_CACHELEVEL , \ .dma_mask_size = 46, \ .has_3d_pipeline = 1, \ .has_64bit_reloc = 1, \@@ -1032,6 +1086,7 @@ static const struct intel_device_info adl_p_info = { .has_reset_engine = 1, \ .has_rps = 1, \ .has_runtime_pm = 1, \ + .max_pat_index = 3, \ .__runtime.ppgtt_size =48 , \ .__runtime.ppgtt_type = INTEL_PPGTT_FULL@@ -1108,11 +1163,13 @@ static const struct intel_device_info pvc_info = {PLATFORM (INTEL_PONTEVECCHIO ),NO_DISPLAY , .has_flat_ccs = 0, + .max_pat_index = 7, .__runtime.platform_engine_mask =BIT (BCS0 ) |BIT (VCS0 ) |BIT (CCS0 ) |BIT (CCS1 ) |BIT (CCS2 ) |BIT (CCS3 ), .require_force_probe = 1, + PVC_CACHELEVEL, };#define XE_LPDP_FEATURES \@@ -1150,9 +1207,11 @@ static const struct intel_device_info mtl_info = { .has_llc = 0, .has_mslice_steering = 0, .has_snoop = 1, + .max_pat_index = 4, .__runtime.memory_regions = REGION_SMEM | REGION_STOLEN_LMEM, .__runtime.platform_engine_mask = BIT(RCS0) | BIT(BCS0) | BIT(CCS0), .require_force_probe = 1, + MTL_CACHELEVEL, };#undef PLATFORMdrivers/gpu/drm/i915/intel_device_info.h +5
-0@@ -35,6 +35,8 @@#include "gt/intel_context_types.h" #include "gt/intel_sseu.h" +#include "gem/i915_gem_object_types.h" +struct drm_printer ;struct drm_i915_private ;struct intel_gt_definition ;@@ -308,6 +310,9 @@ struct intel_device_info {* Initial runtime info. Do not access outside of i915_driver_create(). */ const struct intel_runtime_info __runtime; + +u32 cachelevel_to_pat[I915_MAX_CACHE_LEVEL ]; +u32 max_pat_index; };struct intel_driver_caps {drivers/gpu/drm/i915/selftests/mock_gem_device.c +2
-0@@ -123,7 +123,9 @@ struct drm_i915_private *mock_gem_device(void)static struct dev_iommu fake_iommu = { .priv = (void *)-1 };#endif struct drm_i915_private *i915; +struct intel_device_info *i915_info;struct pci_dev *pdev; +unsigned int i;int ret; pdev =kzalloc (sizeof (*pdev),GFP_KERNEL );
drivers/gpu/drm/i915/selftests/mock_gem_device.c struct drm_i915_private *mock_gem_device(void) · line 190
Quoted 9 lines
drivers/gpu/drm/i915/selftests/mock_gem_device.c +6
-0@@ -180,6 +182,13 @@ struct drm_i915_private *mock_gem_device(void)I915_GTT_PAGE_SIZE_2M ;RUNTIME_INFO (i915)->memory_regions =REGION_SMEM ; + +/* simply use legacy cache level for mock device */ + i915_info = (struct intel_device_info *)INTEL_INFO (i915); + i915_info->max_pat_index =3 ; +for (i =0 ; i <I915_MAX_CACHE_LEVEL ; i++) + i915_info->cachelevel_to_pat[i] = i;
Pity LEGACY_CACHELEVEL cannot be used here.
Regards
Andrzej
Quoted 4 lines
drivers/gpu/drm/i915/selftests/mock_gem_device.c +1
-0+intel_memory_regions_hw_probe (i915);spin_lock_init (&i915->gpu_error.lock);
Re: [Intel-gfx] [PATCH 6/8] drm/i915: preparation for using PAT index
In reply to [email protected]Quoted 7 lines · On Wed, Apr 19, 2023 at 04:00:56PM -0700, [email protected] wrote:
From: Fei Yang <[email protected]>
This patch is a preparation for replacing enum
i915_cache_levelwith PAT index. Caching policy for buffer objects is set through the PAT index in PTE, the oldi915_cache_levelis not sufficient to represent all caching modes supported by the hardware.
Quoted 8 lines
Preparing the transition by adding some platform dependent data structures and helper functions to translate the
cache_leveltopat_index.
cachelevel_to_pat: a platform dependent array mappingcache_levelto
pat_index.
max_pat_index: the maximum PAT index supported by the hardware. Needed for
validating the PAT index passed in from user space.
The description here doesn't quite match how it's being used. For platforms like MTL, the hardware supports PAT indices 0-15. The bspec only gives us values to program for the first 5 of those entries and we leave the rest at their hardware default (fully cached). In the code below, you're setting max_pat_index to the size of the bspec-defined table (i.e., max=4 on MTL). That's fine, but it means the description here ("maximum...supported by hardware") is inaccurate.
Matt
Quoted 427 lines
i915_gem_get_pat_index: function to convertcache_levelto PAT index.
obj_to_i915(obj): macro moved to header file for wider usage.
I915_MAX_CACHE_LEVEL: upper bound ofi915_cache_levelfor the
convenience of coding.Cc Chris Wilson <[email protected]>Cc Matt Roper <[email protected]>Cc Andi Shyti <[email protected]>Signed-off-by Fei Yang <[email protected]>Reviewed-by Andi Shyti <[email protected]>--- drivers/gpu/drm/i915/gem/i915_gem_object.c | 9 +++ drivers/gpu/drm/i915/gem/i915_gem_object.h | 4 + .../gpu/drm/i915/gem/i915_gem_object_types.h | 1 + drivers/gpu/drm/i915/gem/i915_gem_shrinker.c | 2 - drivers/gpu/drm/i915/gt/gen8_ppgtt.c | 6 ++ drivers/gpu/drm/i915/gt/intel_ggtt.c | 6 ++ drivers/gpu/drm/i915/i915_pci.c | 75 +++++++++++++++++-- drivers/gpu/drm/i915/intel_device_info.h | 5 ++ .../gpu/drm/i915/selftests/mock_gem_device.c | 9 +++ 9 files changed, 107 insertions(+), 10 deletions(-)drivers/gpu/drm/i915/gem/i915_gem_object.c +9
-0@@ -45,6 +45,15 @@ static struct kmem_cache *slab_objects;static const struct drm_gem_object_funcs i915_gem_object_funcs; +unsigned int i915_gem_get_pat_index (struct drm_i915_private *i915, +enum i915_cache_level level) +{ +if (drm_WARN_ON (&i915->drm, level >=I915_MAX_CACHE_LEVEL )) +return 0 ; + +return INTEL_INFO (i915)->cachelevel_to_pat[level]; +} +struct drm_i915_gem_object *i915_gem_object_alloc (void ) {struct drm_i915_gem_object *obj;drivers/gpu/drm/i915/gem/i915_gem_object.h +4
-0@@ -20,6 +20,8 @@enum intel_region_id ; +#define obj_to_i915 (obj__) to_i915((obj__)->base.dev) +static inline bool i915_gem_object_size_2big (u64 size) {struct drm_i915_gem_object *obj;@@ -30,6 +32,8 @@ static inline bool i915_gem_object_size_2big(u64 size)return false; } +unsigned int i915_gem_get_pat_index (struct drm_i915_private *i915, +enum i915_cache_level level);void i915_gem_init__objects (struct drm_i915_private *i915);void i915_objects_module_exit (void );drivers/gpu/drm/i915/gem/i915_gem_object_types.h +1
-0@@ -194,6 +194,7 @@ enum i915_cache_level {* engine. */ I915_CACHE_WT , + I915_MAX_CACHE_LEVEL, }; enum i915_map_type {drivers/gpu/drm/i915/gem/i915_gem_shrinker.c +0
-2@@ -460,8 +460,6 @@ void i915_gem_shrinker_taints_mutex(struct drm_i915_private *i915,fs_reclaim_release (GFP_KERNEL ); } -#define obj_to_i915 (obj__) to_i915((obj__)->base.dev) -/** * i915_gem_object_make_unshrinkable - Hide the object from the shrinker. By * default all object types that support shrinking(see IS_SHRINKABLE), will also drivers/gpu/drm/i915/gt/gen8_ppgtt.c +6
-0@@ -78,6 +78,12 @@ static u64 mtl_pte_encode(dma_addr_t addr,case I915_CACHE_WT : pte |=GEN12_PPGTT_PTE_PAT0 ;break ; +default : +/* This should never happen. Added to deal with the compile +* error due to the addition of I915_MAX_CACHE_LEVEL. Will +* be removed by the pat_index patch. +*/ +break ; }return pte;drivers/gpu/drm/i915/gt/intel_ggtt.c +6
-0@@ -242,6 +242,12 @@ static u64 mtl_ggtt_pte_encode(dma_addr_t addr,case I915_CACHE_WT : pte |=MTL_GGTT_PTE_PAT0 ;break ; +default : +/* This should never happen. Added to deal with the compile +* error due to the addition of I915_MAX_CACHE_LEVEL. Will +* be removed by the pat_index patch. +*/ +break ; }return pte;drivers/gpu/drm/i915/i915_pci.c +67
-8@@ -30,6 +30,7 @@#include "display/intel_display_driver.h" #include "gt/intel_gt_regs.h" #include "gt/intel_sa_media.h" +#include "gem/i915_gem_object_types.h" #include "i915_driver.h" #include "i915_drv.h" @@ -164,6 +165,38 @@ .gamma_lut_tests = DRM_COLOR_LUT_NON_DECREASING, \ } +#define LEGACY_CACHELEVEL \ + .cachelevel_to_pat = { \ + [I915_CACHE_NONE] = 0, \ + [I915_CACHE_LLC] = 1, \ + [I915_CACHE_L3_LLC] = 2, \ + [I915_CACHE_WT] = 3, \ + } + +#define TGL_CACHELEVEL \ + .cachelevel_to_pat = { \ + [I915_CACHE_NONE] = 3, \ + [I915_CACHE_LLC] = 0, \ + [I915_CACHE_L3_LLC] = 0, \ + [I915_CACHE_WT] = 2, \ + } + +#define PVC_CACHELEVEL \ + .cachelevel_to_pat = { \ + [I915_CACHE_NONE] = 0, \ + [I915_CACHE_LLC] = 3, \ + [I915_CACHE_L3_LLC] = 3, \ + [I915_CACHE_WT] = 2, \ + } + +#define MTL_CACHELEVEL \ + .cachelevel_to_pat = { \ + [I915_CACHE_NONE] = 2, \ + [I915_CACHE_LLC] = 3, \ + [I915_CACHE_L3_LLC] = 3, \ + [I915_CACHE_WT] = 1, \ + } +/* Keep in gen based order, and chronological order within a gen */ #define GEN_DEFAULT_PAGE_SIZES \@@ -189,11 +222,13 @@ .has_snoop = true, \ .has_coherent_ggtt = false, \ .dma_mask_size = 32, \ + .max_pat_index =3 , \I9XX_PIPE_OFFSETS , \I9XX_CURSOR_OFFSETS , \I9XX_COLORS , \GEN_DEFAULT_PAGE_SIZES , \ - GEN_DEFAULT_REGIONS +GEN_DEFAULT_REGIONS , \ + LEGACY_CACHELEVEL#define I845_FEATURES \ GEN(2), \@@ -210,11 +245,13 @@ .has_snoop = true, \ .has_coherent_ggtt = false, \ .dma_mask_size = 32, \ + .max_pat_index =3 , \I845_PIPE_OFFSETS , \I845_CURSOR_OFFSETS , \I845_COLORS , \GEN_DEFAULT_PAGE_SIZES , \ - GEN_DEFAULT_REGIONS +GEN_DEFAULT_REGIONS , \ + LEGACY_CACHELEVELstatic const struct intel_device_info i830_info = {I830_FEATURES ,@@ -249,11 +286,13 @@ static const struct intel_device_info i865g_info = { .has_snoop = true, \ .has_coherent_ggtt = true, \ .dma_mask_size = 32, \ + .max_pat_index =3 , \I9XX_PIPE_OFFSETS , \I9XX_CURSOR_OFFSETS , \I9XX_COLORS , \GEN_DEFAULT_PAGE_SIZES , \ - GEN_DEFAULT_REGIONS +GEN_DEFAULT_REGIONS , \ + LEGACY_CACHELEVELstatic const struct intel_device_info i915g_info = {GEN3_FEATURES ,@@ -341,11 +380,13 @@ static const struct intel_device_info pnv_m_info = { .has_snoop = true, \ .has_coherent_ggtt = true, \ .dma_mask_size = 36, \ + .max_pat_index =3 , \I9XX_PIPE_OFFSETS , \I9XX_CURSOR_OFFSETS , \I9XX_COLORS , \GEN_DEFAULT_PAGE_SIZES , \ - GEN_DEFAULT_REGIONS +GEN_DEFAULT_REGIONS , \ + LEGACY_CACHELEVELstatic const struct intel_device_info i965g_info = {GEN4_FEATURES ,@@ -395,11 +436,13 @@ static const struct intel_device_info gm45_info = {/* ilk does support rc6, but we do not implement [power] contexts */ \ .has_rc6 = 0, \ .dma_mask_size = 36, \ + .max_pat_index =3 , \I9XX_PIPE_OFFSETS , \I9XX_CURSOR_OFFSETS , \ILK_COLORS , \GEN_DEFAULT_PAGE_SIZES , \ -GEN_DEFAULT_REGIONS +GEN_DEFAULT_REGIONS , \ + LEGACY_CACHELEVELstatic const struct intel_device_info ilk_d_info = {GEN5_FEATURES ,@@ -429,13 +472,15 @@ static const struct intel_device_info ilk_m_info = { .has_rc6p = 0, \ .has_rps = true, \ .dma_mask_size = 40, \ + .max_pat_index = 3, \ .__runtime.ppgtt_type = INTEL_PPGTT_ALIASING, \ .__runtime.ppgtt_size =31 , \I9XX_PIPE_OFFSETS , \I9XX_CURSOR_OFFSETS , \ILK_COLORS , \GEN_DEFAULT_PAGE_SIZES , \ - GEN_DEFAULT_REGIONS +GEN_DEFAULT_REGIONS , \ + LEGACY_CACHELEVEL#define SNB_D_PLATFORM \ GEN6_FEATURES, \@@ -482,13 +527,15 @@ static const struct intel_device_info snb_m_gt2_info = { .has_reset_engine = true, \ .has_rps = true, \ .dma_mask_size = 40, \ + .max_pat_index = 3, \ .__runtime.ppgtt_type = INTEL_PPGTT_ALIASING, \ .__runtime.ppgtt_size =31 , \IVB_PIPE_OFFSETS , \IVB_CURSOR_OFFSETS , \IVB_COLORS , \GEN_DEFAULT_PAGE_SIZES , \ - GEN_DEFAULT_REGIONS +GEN_DEFAULT_REGIONS , \ + LEGACY_CACHELEVEL#define IVB_D_PLATFORM \ GEN7_FEATURES, \@@ -542,6 +589,7 @@ static const struct intel_device_info vlv_info = { .display.has_gmch = 1, .display.has_hotplug = 1, .dma_mask_size = 40, + .max_pat_index = 3, .__runtime.ppgtt_type = INTEL_PPGTT_ALIASING, .__runtime.ppgtt_size = 31, .has_snoop = true,@@ -553,6 +601,7 @@ static const struct intel_device_info vlv_info = {I9XX_COLORS ,GEN_DEFAULT_PAGE_SIZES ,GEN_DEFAULT_REGIONS , + LEGACY_CACHELEVEL, };#define G75_FEATURES \@@ -640,6 +689,7 @@ static const struct intel_device_info chv_info = { .has_logical_ring_contexts = 1, .display.has_gmch = 1, .dma_mask_size = 39, + .max_pat_index = 3, .__runtime.ppgtt_type = INTEL_PPGTT_FULL, .__runtime.ppgtt_size = 32, .has_reset_engine = 1,@@ -651,6 +701,7 @@ static const struct intel_device_info chv_info = {CHV_COLORS ,GEN_DEFAULT_PAGE_SIZES ,GEN_DEFAULT_REGIONS , + LEGACY_CACHELEVEL, };#define GEN9_DEFAULT_PAGE_SIZES \@@ -890,9 +941,11 @@ static const struct intel_device_info jsl_info = { [TRANSCODER_DSI_1 ] =TRANSCODER_DSI1_OFFSET , \ }, \TGL_CURSOR_OFFSETS , \ +TGL_CACHELEVEL , \ .has_global_mocs = 1, \ .has_pxp = 1, \ - .display.has_dsb = 1 + .display.has_dsb = 1, \ + .max_pat_index = 3static const struct intel_device_info tgl_info = {GEN12_FEATURES ,@@ -1014,6 +1067,7 @@ static const struct intel_device_info adl_p_info = { .__runtime.graphics.ip.ver = 12, \ .__runtime.graphics.ip.rel =50 , \XE_HP_PAGE_SIZES , \ +TGL_CACHELEVEL , \ .dma_mask_size = 46, \ .has_3d_pipeline = 1, \ .has_64bit_reloc = 1, \@@ -1032,6 +1086,7 @@ static const struct intel_device_info adl_p_info = { .has_reset_engine = 1, \ .has_rps = 1, \ .has_runtime_pm = 1, \ + .max_pat_index = 3, \ .__runtime.ppgtt_size =48 , \ .__runtime.ppgtt_type = INTEL_PPGTT_FULL@@ -1108,11 +1163,13 @@ static const struct intel_device_info pvc_info = {PLATFORM (INTEL_PONTEVECCHIO ),NO_DISPLAY , .has_flat_ccs = 0, + .max_pat_index = 7, .__runtime.platform_engine_mask =BIT (BCS0 ) |BIT (VCS0 ) |BIT (CCS0 ) |BIT (CCS1 ) |BIT (CCS2 ) |BIT (CCS3 ), .require_force_probe = 1, + PVC_CACHELEVEL, };#define XE_LPDP_FEATURES \@@ -1150,9 +1207,11 @@ static const struct intel_device_info mtl_info = { .has_llc = 0, .has_mslice_steering = 0, .has_snoop = 1, + .max_pat_index = 4, .__runtime.memory_regions = REGION_SMEM | REGION_STOLEN_LMEM, .__runtime.platform_engine_mask = BIT(RCS0) | BIT(BCS0) | BIT(CCS0), .require_force_probe = 1, + MTL_CACHELEVEL, };#undef PLATFORMdrivers/gpu/drm/i915/intel_device_info.h +5
-0@@ -35,6 +35,8 @@#include "gt/intel_context_types.h" #include "gt/intel_sseu.h" +#include "gem/i915_gem_object_types.h" +struct drm_printer ;struct drm_i915_private ;struct intel_gt_definition ;@@ -308,6 +310,9 @@ struct intel_device_info {* Initial runtime info. Do not access outside of i915_driver_create(). */ const struct intel_runtime_info __runtime; + +u32 cachelevel_to_pat[I915_MAX_CACHE_LEVEL ]; +u32 max_pat_index; };struct intel_driver_caps {drivers/gpu/drm/i915/selftests/mock_gem_device.c +9
-0@@ -123,7 +123,9 @@ struct drm_i915_private *mock_gem_device(void)static struct dev_iommu fake_iommu = { .priv = (void *)-1 };#endif struct drm_i915_private *i915; +struct intel_device_info *i915_info;struct pci_dev *pdev; +unsigned int i;int ret; pdev =kzalloc (sizeof (*pdev),GFP_KERNEL );@@ -180,6 +182,13 @@ struct drm_i915_private *mock_gem_device(void)I915_GTT_PAGE_SIZE_2M ;RUNTIME_INFO (i915)->memory_regions =REGION_SMEM ; + +/* simply use legacy cache level for mock device */ + i915_info = (struct intel_device_info *)INTEL_INFO (i915); + i915_info->max_pat_index =3 ; +for (i =0 ; i <I915_MAX_CACHE_LEVEL ; i++) + i915_info->cachelevel_to_pat[i] = i; +intel_memory_regions_hw_probe (i915);spin_lock_init (&i915->gpu_error.lock);Signature · 1 line
2.25.1
Signature · 4 lines
Matt Roper Graphics Software Engineer Linux GPU Platform Enablement Intel Corporation
PATCH 7/8 drm/i915: use pat_index instead of cache_level
In reply to [email protected]From: Fei Yang <[email protected]>
Currently the KMD is using enum i915_cache_level to set caching policy for buffer objects. This is flaky because the PAT index which really controls the caching behavior in PTE has far more levels than what's defined in the enum. In addition, the PAT index is platform dependent, having to translate between i915_cache_level and PAT index is not reliable, and makes the code more complicated.
From UMD's perspective there is also a necessity to set caching policy for performance fine tuning. It's much easier for the UMD to directly use PAT index because the behavior of each PAT index is clearly defined in Bspec. Having the abstracted i915_cache_level sitting in between would only cause more ambiguity.
For these reasons this patch replaces i915_cache_level with PAT index. Also note, the cache_level is not completely removed yet, because the KMD still has the need of creating buffer objects with simple cache settings such as cached, uncached, or writethrough. For such simple cases, using cache_level would help simplify the code.
--- drivers/gpu/drm/i915/display/intel_dpt.c | 12 +-- drivers/gpu/drm/i915/gem/i915_gem_domain.c | 27 ++---- .../gpu/drm/i915/gem/i915_gem_execbuffer.c | 10 ++- drivers/gpu/drm/i915/gem/i915_gem_mman.c | 3 +- drivers/gpu/drm/i915/gem/i915_gem_object.c | 52 +++++++++++- drivers/gpu/drm/i915/gem/i915_gem_object.h | 4 + .../gpu/drm/i915/gem/i915_gem_object_types.h | 25 +++++- drivers/gpu/drm/i915/gem/i915_gem_stolen.c | 4 +- drivers/gpu/drm/i915/gem/i915_gem_ttm_move.c | 16 ++-- .../gpu/drm/i915/gem/selftests/huge_pages.c | 2 +- .../drm/i915/gem/selftests/i915_gem_migrate.c | 2 +- .../drm/i915/gem/selftests/i915_gem_mman.c | 2 +- drivers/gpu/drm/i915/gt/gen6_ppgtt.c | 10 ++- drivers/gpu/drm/i915/gt/gen8_ppgtt.c | 71 ++++++++-------- drivers/gpu/drm/i915/gt/gen8_ppgtt.h | 3 +- drivers/gpu/drm/i915/gt/intel_ggtt.c | 82 +++++++++---------- drivers/gpu/drm/i915/gt/intel_gtt.h | 20 ++--- drivers/gpu/drm/i915/gt/intel_migrate.c | 47 ++++++----- drivers/gpu/drm/i915/gt/intel_migrate.h | 13 ++- drivers/gpu/drm/i915/gt/intel_ppgtt.c | 6 +- drivers/gpu/drm/i915/gt/selftest_migrate.c | 47 ++++++----- drivers/gpu/drm/i915/gt/selftest_reset.c | 8 +- drivers/gpu/drm/i915/gt/selftest_timeline.c | 2 +- drivers/gpu/drm/i915/gt/selftest_tlb.c | 4 +- drivers/gpu/drm/i915/gt/uc/intel_uc_fw.c | 10 ++- drivers/gpu/drm/i915/i915_debugfs.c | 55 ++++++++++--- drivers/gpu/drm/i915/i915_gem.c | 16 +++- drivers/gpu/drm/i915/i915_gpu_error.c | 8 +- drivers/gpu/drm/i915/i915_vma.c | 16 ++-- drivers/gpu/drm/i915/i915_vma.h | 2 +- drivers/gpu/drm/i915/i915_vma_types.h | 2 - drivers/gpu/drm/i915/selftests/i915_gem.c | 5 +- .../gpu/drm/i915/selftests/i915_gem_evict.c | 4 +- drivers/gpu/drm/i915/selftests/i915_gem_gtt.c | 15 ++-- .../drm/i915/selftests/intel_memory_region.c | 4 +- drivers/gpu/drm/i915/selftests/mock_gtt.c | 8 +- 36 files changed, 378 insertions(+), 239 deletions(-)
drivers/gpu/drm/i915/display/intel_dpt.c +6 -6
@@ -43,24 +43,24 @@ static void gen8_set_pte(void __iomem *addr, gen8_pte_t pte)static void dpt_insert_page (struct i915_address_space *vm,dma_addr_t addr,u64 offset, -enum i915_cache_level level, +unsigned int pat_index,u32 flags) {struct i915_dpt *dpt =i915_vm_to_dpt (vm);gen8_pte_t __iomem *base = dpt->iomem;gen8_set_pte (base + offset /I915_GTT_PAGE_SIZE , - vm->pte_encode (addr, level, flags)); + vm->pte_encode (addr, pat_index, flags)); }static void dpt_insert_entries (struct i915_address_space *vm,struct i915_vma_resource *vma_res, -enum i915_cache_level level, +unsigned int pat_index,u32 flags) {struct i915_dpt *dpt =i915_vm_to_dpt (vm);gen8_pte_t __iomem *base = dpt->iomem; -const gen8_pte_t pte_encode = vm->pte_encode (0 , level, flags); +const gen8_pte_t pte_encode = vm->pte_encode (0 , pat_index, flags);struct sgt_iter sgt_iter;dma_addr_t addr;int i;
@@ -83,7 +83,7 @@ static void dpt_clear_range(struct i915_address_space *vm,static void dpt_bind_vma (struct i915_address_space *vm,struct i915_vm_pt_stash *stash,struct i915_vma_resource *vma_res, -enum i915_cache_level cache_level, +unsigned int pat_index,u32 flags) {u32 pte_flags;
@@ -98,7 +98,7 @@ static void dpt_bind_vma(struct i915_address_space *vm,if (vma_res->bi.lmem) pte_flags |=PTE_LM ; - vm->insert_entries (vm, vma_res, cache_level, pte_flags); + vm->insert_entries (vm, vma_res, pat_index, pte_flags); vma_res->page_sizes_gtt =I915_GTT_PAGE_SIZE ;
drivers/gpu/drm/i915/gem/i915_gem_domain.c +9 -18
@@ -27,8 +27,8 @@ static bool gpu_write_needs_clflush(struct drm_i915_gem_object *obj)if (IS_DGFX (i915))return false; -return !(obj->cache_level ==I915_CACHE_NONE || - obj->cache_level ==I915_CACHE_WT ); +return !(i915_gem_object_has_cache_level (obj,I915_CACHE_NONE ) || +i915_gem_object_has_cache_level (obj,I915_CACHE_WT )); }bool i915_gem_cpu_write_needs_clflush (struct drm_i915_gem_object *obj)
@@ -267,7 +267,7 @@ int i915_gem_object_set_cache_level(struct drm_i915_gem_object *obj, {int ret; -if (obj->cache_level == cache_level) +if (i915_gem_object_has_cache_level (obj, cache_level))return 0 ; ret = i915_gem_object_wait(obj,
@@ -278,10 +278,8 @@ int i915_gem_object_set_cache_level(struct drm_i915_gem_object *obj,return ret;/* Always invalidate stale cachelines */ -if (obj->cache_level != cache_level) { -i915_gem_object_set_cache_coherency (obj, cache_level); - obj->cache_dirty = true; - } +i915_gem_object_set_cache_coherency (obj, cache_level); + obj->cache_dirty = true;/* The cache-level will be applied when each vma is rebound. */ return i915_gem_object_unbind(obj,
@@ -306,20 +304,13 @@ int i915_gem_get_caching_ioctl(struct drm_device *dev, void *data, goto out; } -switch (obj->cache_level) { -case I915_CACHE_LLC : -case I915_CACHE_L3_LLC : +if (i915_gem_object_has_cache_level (obj,I915_CACHE_LLC ) || +i915_gem_object_has_cache_level (obj,I915_CACHE_L3_LLC )) args->caching =I915_CACHING_CACHED ; -break ; - -case I915_CACHE_WT : +else if (i915_gem_object_has_cache_level (obj,I915_CACHE_WT )) args->caching =I915_CACHING_DISPLAY ; -break ; - -default : +else args->caching =I915_CACHING_NONE ; -break ; - } out:rcu_read_unlock ();return err;
drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c +6 -4
@@ -642,7 +642,7 @@ static inline int use_cpu_reloc(const struct reloc_cache *cache,return (cache->has_llc || obj->cache_dirty || - obj->cache_level !=I915_CACHE_NONE ); + !i915_gem_object_has_cache_level (obj,I915_CACHE_NONE )); } static int eb_reserve_vma(struct i915_execbuffer *eb,
@@ -1323,8 +1323,10 @@ static void *reloc_iomap(struct i915_vma *batch, offset = cache->node.start;if (drm_mm_node_allocated (&cache->node)) { ggtt->vm.insert_page (&ggtt->vm, -i915_gem_object_get_dma_address (obj, page), - offset,I915_CACHE_NONE ,0 ); +i915_gem_object_get_dma_address (obj, page), + offset, +i915_gem_get_pat_index (ggtt->vm.i915,I915_CACHE_NONE ), +0 ); }else { offset += page <<PAGE_SHIFT ; }
@@ -1464,7 +1466,7 @@ eb_relocate_entry(struct i915_execbuffer *eb,reloc_cache_unmap (&eb->reloc_cache);mutex_lock (&vma->vm->mutex); err =i915_vma_bind (target->vma, - target->vma->obj->cache_level, + target->vma->obj->pat_index,PIN_GLOBAL ,NULL ,NULL );mutex_unlock (&vma->vm->mutex);reloc_cache_remap (&eb->reloc_cache, ev->vma->obj);
drivers/gpu/drm/i915/gem/i915_gem_mman.c +2 -1
@@ -383,7 +383,8 @@ static vm_fault_t vm_fault_gtt(struct vm_fault *vmf) }/* Access to snoopable pages through the GTT is incoherent. */ -if (obj->cache_level !=I915_CACHE_NONE && !HAS_LLC (i915)) { +if (!(i915_gem_object_has_cache_level (obj,I915_CACHE_NONE ) || +HAS_LLC (i915))) { ret = -EFAULT ; goto err_unpin; }
drivers/gpu/drm/i915/gem/i915_gem_object.c +51 -1
70 lines not on this page · raw message
drivers/gpu/drm/i915/gem/i915_gem_object.h +4 -0
@@ -34,6 +34,8 @@ static inline bool i915_gem_object_size_2big(u64 size)unsigned int i915_gem_get_pat_index (struct drm_i915_private *i915,enum i915_cache_level level); +bool i915_gem_object_has_cache_level (const struct drm_i915_gem_object *obj, +enum i915_cache_level lvl);void i915_gem_init__objects (struct drm_i915_private *i915);void i915_objects_module_exit (void );
@@ -764,6 +766,8 @@ bool i915_gem_object_has_unknown_state(struct drm_i915_gem_object *obj);void i915_gem_object_set_cache_coherency (struct drm_i915_gem_object *obj,unsigned int cache_level); +void i915_gem_object_set_pat_index (struct drm_i915_gem_object *obj, +unsigned int pat_index);bool i915_gem_object_can_bypass_llc (struct drm_i915_gem_object *obj);void i915_gem_object_flush_if_display (struct drm_i915_gem_object *obj);void i915_gem_object_flush_if_display_locked (struct drm_i915_gem_object *obj);
drivers/gpu/drm/i915/gem/i915_gem_object_types.h +22 -3
@@ -195,6 +195,7 @@ enum i915_cache_level {*/ I915_CACHE_WT ,I915_MAX_CACHE_LEVEL , + I915_CACHE_INVAL = I915_MAX_CACHE_LEVEL, }; enum i915_map_type {
@@ -358,10 +359,28 @@ struct drm_i915_gem_object {#define I915_BO_FLAG_STRUCT_PAGE BIT(0)/* Object backed by struct pages */ #define I915_BO_FLAG_IOMEM BIT(1)/* Object backed by IO memory */ /** -* @cache_level: The desired GTT caching level. +* @pat_index: The desired PAT index. +* +* See hardware specification for valid PAT indices for each platform. +* This field used to contain a value of enum i915_cache_level. It's +* changed to an unsigned int because PAT indices are being used by +* both UMD and KMD for caching policy control after GEN12. +* For backward compatibility, this field will continue to contain +* value of i915_cache_level for pre-GEN12 platforms so that the PTE +* encode functions for these legacy platforms can stay the same. +* In the meantime platform specific tables are created to translate +* i915_cache_level into pat index, for more details check the macros +* defined i915/i915_pci.c, e.g. PVC_CACHELEVEL. +*/ + unsigned int pat_index:6; +/** +* @cache_level: Indicate whether pat_index is set by UMD * -* See enum i915_cache_level for possible values, along with what -* each does. +* This used to hold desired GTT caching level, but is now replaced by +* pat_index. It's kept here for KMD to tell whether the pat_index is +* set by UMD or converted from enum i915_cache_level. +* This field should be 0 by default, but I915_CACHE_INVAL if the +* pat_index is set by UMD. */ unsigned int cache_level:3;/**
drivers/gpu/drm/i915/gem/i915_gem_stolen.c +3 -1
@@ -565,7 +565,9 @@ static void dbg_poison(struct i915_ggtt *ggtt, ggtt->vm.insert_page (&ggtt->vm, addr, ggtt->error_capture.start, -I915_CACHE_NONE ,0 ); +i915_gem_get_pat_index (ggtt->vm.i915, +I915_CACHE_NONE ), +0 );mb (); s = io_mapping_map_wc(&ggtt->iomap,
drivers/gpu/drm/i915/gem/i915_gem_ttm_move.c +9 -7
@@ -214,7 +214,8 @@ static struct dma_fence *i915_ttm_accel_move(struct ttm_buffer_object *bo,intel_engine_pm_get (to_gt (i915)->migrate.context->engine); ret =intel_context_migrate_clear (to_gt (i915)->migrate.context, deps, - dst_st->sgl, dst_level, + dst_st->sgl, +i915_gem_get_pat_index (i915, dst_level),i915_ttm_gtt_binds_lmem (dst_mem),0 , &rq); }else {
@@ -227,12 +228,13 @@ static struct dma_fence *i915_ttm_accel_move(struct ttm_buffer_object *bo, src_level =i915_ttm_cache_level (i915, bo->resource, src_ttm);intel_engine_pm_get (to_gt (i915)->migrate.context->engine); ret =intel_context_migrate_copy (to_gt (i915)->migrate.context, - deps, src_rsgt->table.sgl, - src_level, -i915_ttm_gtt_binds_lmem (bo->resource), - dst_st->sgl, dst_level, -i915_ttm_gtt_binds_lmem (dst_mem), - &rq); + deps, src_rsgt->table.sgl, +i915_gem_get_pat_index (i915, src_level), +i915_ttm_gtt_binds_lmem (bo->resource), + dst_st->sgl, +i915_gem_get_pat_index (i915, dst_level), +i915_ttm_gtt_binds_lmem (dst_mem), + &rq);i915_refct_sgt_put (src_rsgt); }
drivers/gpu/drm/i915/gem/selftests/huge_pages.c +1 -1
@@ -354,7 +354,7 @@ fake_huge_pages_object(struct drm_i915_private *i915, u64 size, bool single) obj->write_domain =I915_GEM_DOMAIN_CPU ; obj->read_domains =I915_GEM_DOMAIN_CPU ; - obj->cache_level =I915_CACHE_NONE ; + obj->pat_index =i915_gem_get_pat_index (i915,I915_CACHE_NONE );return obj; }
drivers/gpu/drm/i915/gem/selftests/i915_gem_migrate.c +1 -1
@@ -219,7 +219,7 @@ static int __igt_lmem_pages_migrate(struct intel_gt *gt,continue ; err =intel_migrate_clear (>->migrate, &ww, deps, - obj->mm.pages->sgl, obj->cache_level, + obj->mm.pages->sgl, obj->pat_index,i915_gem_object_is_lmem (obj),0xdeadbeaf , &rq);if (rq) {
drivers/gpu/drm/i915/gem/selftests/i915_gem_mman.c +1 -1
@@ -1222,7 +1222,7 @@ static int __igt_mmap_migrate(struct intel_memory_region **placements, } err =intel_context_migrate_clear (to_gt (i915)->migrate.context,NULL , - obj->mm.pages->sgl, obj->cache_level, + obj->mm.pages->sgl, obj->pat_index,i915_gem_object_is_lmem (obj),expand32 (POISON_INUSE ), &rq);i915_gem_object_unpin_pages (obj);
drivers/gpu/drm/i915/gt/gen6_ppgtt.c +6 -4
@@ -109,7 +109,7 @@ static void gen6_ppgtt_clear_range(struct i915_address_space *vm,static void gen6_ppgtt_insert_entries (struct i915_address_space *vm,struct i915_vma_resource *vma_res, -enum i915_cache_level cache_level, +unsigned int pat_index,u32 flags) {struct i915_ppgtt *ppgtt =i915_vm_to_ppgtt (vm);
@@ -117,7 +117,7 @@ static void gen6_ppgtt_insert_entries(struct i915_address_space *vm,unsigned int first_entry = vma_res->start /I915_GTT_PAGE_SIZE ;unsigned int act_pt = first_entry /GEN6_PTES ;unsigned int act_pte = first_entry %GEN6_PTES ; -const u32 pte_encode = vm->pte_encode (0 , cache_level, flags); +const u32 pte_encode = vm->pte_encode (0 , pat_index, flags);struct sgt_dma iter =sgt_dma (vma_res);gen6_pte_t *vaddr;
@@ -227,7 +227,9 @@ static int gen6_ppgtt_init_scratch(struct gen6_ppgtt *ppgtt) vm->scratch[0 ]->encode = vm->pte_encode (px_dma (vm->scratch[0 ]), -I915_CACHE_NONE ,PTE_READ_ONLY ); +i915_gem_get_pat_index (vm->i915, +I915_CACHE_NONE ), +PTE_READ_ONLY ); vm->scratch[1 ] = vm->alloc_pt_dma (vm,I915_GTT_PAGE_SIZE_4K );if (IS_ERR (vm->scratch[1 ])) {
@@ -278,7 +280,7 @@ static void gen6_ppgtt_cleanup(struct i915_address_space *vm)static void pd_vma_bind (struct i915_address_space *vm,struct i915_vm_pt_stash *stash,struct i915_vma_resource *vma_res, -enum i915_cache_level cache_level, +unsigned int pat_index,u32 unused) {struct i915_ggtt *ggtt =i915_vm_to_ggtt (vm);
drivers/gpu/drm/i915/gt/gen8_ppgtt.c +34 -37
174 lines not on this page · raw message
drivers/gpu/drm/i915/gt/gen8_ppgtt.h +1 -2
@@ -10,13 +10,12 @@struct i915_address_space ;struct intel_gt ; -enum i915_cache_level ;struct i915_ppgtt *gen8_ppgtt_create (struct intel_gt *gt,unsigned long lmem_pt_obj_flags);u64 gen8_ggtt_pte_encode (dma_addr_t addr, -enum i915_cache_level level, +unsigned int pat_index,u32 flags);#endif
drivers/gpu/drm/i915/gt/intel_ggtt.c +40 -42
229 lines not on this page · raw message
drivers/gpu/drm/i915/gt/intel_gtt.h +9 -11
@@ -165,8 +165,6 @@ typedef u64 gen8_pte_t;#define MTL_2_COH_1W REG_FIELD_PREP(MTL_PAT_INDEX_COH_MODE_MASK, 2)#define MTL_0_COH_NON REG_FIELD_PREP(MTL_PAT_INDEX_COH_MODE_MASK, 0) -enum i915_cache_level ; -struct drm_i915_gem_object ;struct i915_fence_reg ;struct i915_vma ;
@@ -234,7 +232,7 @@ struct i915_vma_ops {void (*bind_vma)(struct i915_address_space *vm,struct i915_vm_pt_stash *stash,struct i915_vma_resource *vma_res, -enum i915_cache_level cache_level, +unsigned int pat_index,u32 flags);/* * Unmap an object from an address space. This usually consists of
@@ -306,7 +304,7 @@ struct i915_address_space { (*alloc_scratch_dma)(struct i915_address_space *vm, int sz);u64 (*pte_encode)(dma_addr_t addr, -enum i915_cache_level level, +unsigned int pat_index,u32 flags);/* Create a valid PTE */ #define PTE_READ_ONLY BIT(0)#define PTE_LM BIT(1)
@@ -321,20 +319,20 @@ struct i915_address_space {void (*insert_page)(struct i915_address_space *vm,dma_addr_t addr,u64 offset, -enum i915_cache_level cache_level, +unsigned int pat_index,u32 flags);void (*insert_entries)(struct i915_address_space *vm,struct i915_vma_resource *vma_res, -enum i915_cache_level cache_level, +unsigned int pat_index,u32 flags);void (*raw_insert_page)(struct i915_address_space *vm,dma_addr_t addr,u64 offset, -enum i915_cache_level cache_level, +unsigned int pat_index,u32 flags);void (*raw_insert_entries)(struct i915_address_space *vm,struct i915_vma_resource *vma_res, -enum i915_cache_level cache_level, +unsigned int pat_index,u32 flags);void (*cleanup)(struct i915_address_space *vm);
@@ -581,7 +579,7 @@ void ppgtt_init(struct i915_ppgtt *ppgtt, struct intel_gt *gt,void intel_ggtt_bind_vma (struct i915_address_space *vm,struct i915_vm_pt_stash *stash,struct i915_vma_resource *vma_res, -enum i915_cache_level cache_level, +unsigned int pat_index,u32 flags);void intel_ggtt_unbind_vma (struct i915_address_space *vm,struct i915_vma_resource *vma_res);
@@ -639,7 +637,7 @@ void __set_pd_entry(struct i915_page_directory * const pd, const unsigned short idx, struct i915_page_table *pt, - u64 (*encode)(const dma_addr_t, const enum i915_cache_level)); + u64 (*encode)(const dma_addr_t, const unsigned int pat_index));#define set_pd_entry (pd, idx, to) \ __set_pd_entry((pd), (idx), px_pt(to), gen8_pde_encode)
@@ -659,7 +657,7 @@ void gen6_ggtt_invalidate(struct i915_ggtt *ggtt);void ppgtt_bind_vma (struct i915_address_space *vm,struct i915_vm_pt_stash *stash,struct i915_vma_resource *vma_res, -enum i915_cache_level cache_level, +unsigned int pat_index,u32 flags);void ppgtt_unbind_vma (struct i915_address_space *vm,struct i915_vma_resource *vma_res);
drivers/gpu/drm/i915/gt/intel_migrate.c +26 -21
156 lines not on this page · raw message
drivers/gpu/drm/i915/gt/intel_migrate.h +6 -7
@@ -16,7 +16,6 @@ struct i915_request;struct i915_gem_ww_ctx ;struct intel_gt ;struct scatterlist ; -enum i915_cache_level ;int intel_migrate_init (struct intel_migrate *m,struct intel_gt *gt);
@@ -26,20 +25,20 @@ int intel_migrate_copy(struct intel_migrate *m,struct i915_gem_ww_ctx *ww, const struct i915_deps *deps, struct scatterlist *src, - enum i915_cache_level src_cache_level, + unsigned int src_pat_index, bool src_is_lmem, struct scatterlist *dst, - enum i915_cache_level dst_cache_level, + unsigned int dst_pat_index, bool dst_is_lmem, struct i915_request **out);int intel_context_migrate_copy (struct intel_context *ce,const struct i915_deps *deps,struct scatterlist *src, -enum i915_cache_level src_cache_level, +unsigned int src_pat_index,bool src_is_lmem,struct scatterlist *dst, -enum i915_cache_level dst_cache_level, +unsigned int dst_pat_index,bool dst_is_lmem,struct i915_request **out);
@@ -48,7 +47,7 @@ intel_migrate_clear(struct intel_migrate *m,struct i915_gem_ww_ctx *ww, const struct i915_deps *deps, struct scatterlist *sg, - enum i915_cache_level cache_level, + unsigned int pat_index, bool is_lmem, u32 value, struct i915_request **out);
@@ -56,7 +55,7 @@ int intel_context_migrate_clear(struct intel_context *ce, const struct i915_deps *deps, struct scatterlist *sg, - enum i915_cache_level cache_level, + unsigned int pat_index, bool is_lmem, u32 value, struct i915_request **out);
drivers/gpu/drm/i915/gt/intel_ppgtt.c +3 -3
@@ -98,7 +98,7 @@ void __set_pd_entry(struct i915_page_directory * const pd, const unsigned short idx, struct i915_page_table * const to, - u64 (*encode)(const dma_addr_t, const enum i915_cache_level)) + u64 (*encode)(const dma_addr_t, const unsigned int)) { /* Each thread pre-pins the pd, and we may have a thread per pde. */ GEM_BUG_ON(atomic_read(px_used(pd)) > NALLOC * I915_PDES);
@@ -181,7 +181,7 @@ struct i915_ppgtt *i915_ppgtt_create(struct intel_gt *gt,void ppgtt_bind_vma (struct i915_address_space *vm,struct i915_vm_pt_stash *stash,struct i915_vma_resource *vma_res, -enum i915_cache_level cache_level, +unsigned int pat_index,u32 flags) {u32 pte_flags;
@@ -199,7 +199,7 @@ void ppgtt_bind_vma(struct i915_address_space *vm,if (vma_res->bi.lmem) pte_flags |=PTE_LM ; - vm->insert_entries (vm, vma_res, cache_level, pte_flags); + vm->insert_entries (vm, vma_res, pat_index, pte_flags);wmb (); }
drivers/gpu/drm/i915/gt/selftest_migrate.c +25 -22
156 lines not on this page · raw message
drivers/gpu/drm/i915/gt/selftest_reset.c +6 -2
@@ -86,7 +86,9 @@ __igt_reset_stolen(struct intel_gt *gt, ggtt->vm.insert_page (&ggtt->vm, dma, ggtt->error_capture.start, -I915_CACHE_NONE ,0 ); +i915_gem_get_pat_index (gt->i915, +I915_CACHE_NONE ), +0 );mb (); s = io_mapping_map_wc(&ggtt->iomap,
@@ -127,7 +129,9 @@ __igt_reset_stolen(struct intel_gt *gt, ggtt->vm.insert_page (&ggtt->vm, dma, ggtt->error_capture.start, -I915_CACHE_NONE ,0 ); +i915_gem_get_pat_index (gt->i915, +I915_CACHE_NONE ), +0 );mb (); s = io_mapping_map_wc(&ggtt->iomap,
drivers/gpu/drm/i915/gt/selftest_timeline.c +1 -1
@@ -836,7 +836,7 @@ static int setup_watcher(struct hwsp_watcher *w, struct intel_gt *gt,return PTR_ERR (obj);/* keep the same cache settings as timeline */ -i915_gem_object_set_cache_coherency (obj, tl->hwsp_ggtt->obj->cache_level); +i915_gem_object_set_pat_index (obj, tl->hwsp_ggtt->obj->pat_index); w->map =i915_gem_object_pin_map_unlocked (obj,page_unmask_bits (tl->hwsp_ggtt->obj->mm.mapping));if (IS_ERR (w->map)) {
drivers/gpu/drm/i915/gt/selftest_tlb.c +3 -1
@@ -36,6 +36,8 @@ pte_tlbinv(struct intel_context *ce, u64 length, struct rnd_state *prng) { +const unsigned int pat_index = +i915_gem_get_pat_index (ce->vm->i915,I915_CACHE_NONE );struct drm_i915_gem_object *batch;struct drm_mm_node vb_node;struct i915_request *rq;
@@ -155,7 +157,7 @@ pte_tlbinv(struct intel_context *ce,/* Flip the PTE between A and B */ if (i915_gem_object_is_lmem (vb->obj)) pte_flags |=PTE_LM ; - ce->vm->insert_entries (ce->vm, &vb_res,0 , pte_flags); + ce->vm->insert_entries (ce->vm, &vb_res, pat_index, pte_flags);/* Flush the PTE update to concurrent HW */ tlbinv (ce->vm, addr & -length, length);
drivers/gpu/drm/i915/gt/uc/intel_uc_fw.c +8 -2
@@ -890,9 +890,15 @@ static void uc_fw_bind_ggtt(struct intel_uc_fw *uc_fw) pte_flags |=PTE_LM ;if (ggtt->vm.raw_insert_entries) - ggtt->vm.raw_insert_entries (&ggtt->vm, dummy,I915_CACHE_NONE , pte_flags); + ggtt->vm.raw_insert_entries (&ggtt->vm, dummy, +i915_gem_get_pat_index (ggtt->vm.i915, +I915_CACHE_NONE ), + pte_flags);else - ggtt->vm.insert_entries (&ggtt->vm, dummy,I915_CACHE_NONE , pte_flags); + ggtt->vm.insert_entries (&ggtt->vm, dummy, +i915_gem_get_pat_index (ggtt->vm.i915, +I915_CACHE_NONE ), + pte_flags); }static void uc_fw_unbind_ggtt (struct intel_uc_fw *uc_fw)
drivers/gpu/drm/i915/i915_debugfs.c +45 -10
73 lines not on this page · raw message
drivers/gpu/drm/i915/i915_gem.c +12 -4
@@ -420,8 +420,12 @@ i915_gem_gtt_pread(struct drm_i915_gem_object *obj, page_length = remain < page_length ? remain : page_length;if (drm_mm_node_allocated (&node)) { ggtt->vm.insert_page (&ggtt->vm, -i915_gem_object_get_dma_address (obj, offset >>PAGE_SHIFT ), - node.start,I915_CACHE_NONE ,0 ); +i915_gem_object_get_dma_address (obj, + offset >>PAGE_SHIFT ), + node.start, +i915_gem_get_pat_index (i915, +I915_CACHE_NONE ), +0 ); }else { page_base += offset &PAGE_MASK ; }
@@ -598,8 +602,12 @@ i915_gem_gtt_pwrite_fast(struct drm_i915_gem_object *obj,/* flush the write before we modify the GGTT */ intel_gt_flush_ggtt_writes (ggtt->vm.gt); ggtt->vm.insert_page (&ggtt->vm, -i915_gem_object_get_dma_address (obj, offset >>PAGE_SHIFT ), - node.start,I915_CACHE_NONE ,0 ); +i915_gem_object_get_dma_address (obj, + offset >>PAGE_SHIFT ), + node.start, +i915_gem_get_pat_index (i915, +I915_CACHE_NONE ), +0 );wmb ();/* flush modifications to the GGTT (insert_page) */ }else { page_base += offset &PAGE_MASK ;
drivers/gpu/drm/i915/i915_gpu_error.c +6 -2
@@ -1117,10 +1117,14 @@ i915_vma_coredump_create(const struct intel_gt *gt,mutex_lock (&ggtt->error_mutex);if (ggtt->vm.raw_insert_page) ggtt->vm.raw_insert_page (&ggtt->vm, dma, slot, -I915_CACHE_NONE ,0 ); +i915_gem_get_pat_index (gt->i915, +I915_CACHE_NONE ), +0 );else ggtt->vm.insert_page (&ggtt->vm, dma, slot, -I915_CACHE_NONE ,0 ); +i915_gem_get_pat_index (gt->i915, +I915_CACHE_NONE ), +0 );mb (); s =io_mapping_map_wc (&ggtt->iomap, slot,PAGE_SIZE );
drivers/gpu/drm/i915/i915_vma.c +8 -8
@@ -315,7 +315,7 @@ struct i915_vma_work {struct i915_vma_resource *vma_res;struct drm_i915_gem_object *obj;struct i915_sw_dma_fence_cb cb; -enum i915_cache_level cache_level; +unsigned int pat_index;unsigned int flags; };
@@ -334,7 +334,7 @@ static void __vma_bind(struct dma_fence_work *work)return ; vma_res->ops->bind_vma (vma_res->vm, &vw->stash, - vma_res, vw->cache_level, vw->flags); + vma_res, vw->pat_index, vw->flags); }static void __vma_release (struct dma_fence_work *work)
@@ -426,7 +426,7 @@ i915_vma_resource_init_from_vma(struct i915_vma_resource *vma_res,/** * i915_vma_bind - Sets up PTEs for an VMA in it's corresponding address space. * @vma: VMA to map -* @cache_level: mapping cache level +* @pat_index: PAT index to set in PTE * @flags: flags like global or local mapping * @work: preallocated worker for allocating and binding the PTE * @vma_res: pointer to a preallocated vma resource. The resource is either
@@ -437,7 +437,7 @@ i915_vma_resource_init_from_vma(struct i915_vma_resource *vma_res,* Note that DMA addresses are also the only part of the SG table we care about. */ int i915_vma_bind (struct i915_vma *vma, -enum i915_cache_level cache_level, +unsigned int pat_index,u32 flags,struct i915_vma_work *work,struct i915_vma_resource *vma_res)
@@ -507,7 +507,7 @@ int i915_vma_bind(struct i915_vma *vma,struct dma_fence *prev; work->vma_res =i915_vma_resource_get (vma->resource); - work->cache_level = cache_level; + work->pat_index = pat_index; work->flags = bind_flags;/*
@@ -537,7 +537,7 @@ int i915_vma_bind(struct i915_vma *vma,return ret; } - vma->ops->bind_vma (vma->vm,NULL , vma->resource, cache_level, + vma->ops->bind_vma (vma->vm,NULL , vma->resource, pat_index, bind_flags); }
@@ -814,7 +814,7 @@ i915_vma_insert(struct i915_vma *vma, struct i915_gem_ww_ctx *ww, color =0 ;if (i915_vm_has_cache_coloring (vma->vm)) - color = vma->obj->cache_level; + color = vma->obj->pat_index;if (flags &PIN_OFFSET_FIXED ) {u64 offset = flags &PIN_OFFSET_MASK ;
@@ -1518,7 +1518,7 @@ int i915_vma_pin_ww(struct i915_vma *vma, struct i915_gem_ww_ctx *ww,GEM_BUG_ON (!vma->pages); err =i915_vma_bind (vma, - vma->obj->cache_level, + vma->obj->pat_index, flags, work, vma_res); vma_res =NULL ;if (err)
drivers/gpu/drm/i915/i915_vma.h +1 -1
@@ -250,7 +250,7 @@ i915_vma_compare(struct i915_vma *vma,struct i915_vma_work *i915_vma_work (void );int i915_vma_bind (struct i915_vma *vma, -enum i915_cache_level cache_level, +unsigned int pat_index,u32 flags,struct i915_vma_work *work,struct i915_vma_resource *vma_res);
drivers/gpu/drm/i915/i915_vma_types.h +0 -2
@@ -32,8 +32,6 @@#include "gem/i915_gem_object_types.h" -enum i915_cache_level ; -/** * DOC: Global GTT views *
drivers/gpu/drm/i915/selftests/i915_gem.c +4 -1
@@ -57,7 +57,10 @@ static void trash_stolen(struct drm_i915_private *i915)u32 __iomem *s;int x; - ggtt->vm.insert_page (&ggtt->vm, dma, slot,I915_CACHE_NONE ,0 ); + ggtt->vm.insert_page (&ggtt->vm, dma, slot, +i915_gem_get_pat_index (i915, +I915_CACHE_NONE ), +0 ); s =io_mapping_map_atomic_wc (&ggtt->iomap, slot);for (x =0 ; x <PAGE_SIZE /sizeof (u32); x++) {
drivers/gpu/drm/i915/selftests/i915_gem_evict.c +2 -2
@@ -245,7 +245,7 @@ static int igt_evict_for_cache_color(void *arg)struct drm_mm_node target = { .start =I915_GTT_PAGE_SIZE *2 , .size =I915_GTT_PAGE_SIZE , - .color =I915_CACHE_LLC , + .color =i915_gem_get_pat_index (gt->i915,I915_CACHE_LLC ), };struct drm_i915_gem_object *obj;struct i915_vma *vma;
@@ -308,7 +308,7 @@ static int igt_evict_for_cache_color(void *arg)/* Attempt to remove the first *pinned* vma, by removing the (empty) * neighbour -- this should fail. */ - target.color =I915_CACHE_L3_LLC ; + target.color =i915_gem_get_pat_index (gt->i915,I915_CACHE_L3_LLC );mutex_lock (&ggtt->vm.mutex); err =i915_gem_evict_for_node (&ggtt->vm,NULL , &target,0 );
drivers/gpu/drm/i915/selftests/i915_gem_gtt.c +10 -5
@@ -135,7 +135,7 @@ fake_dma_object(struct drm_i915_private *i915, u64 size) obj->write_domain =I915_GEM_DOMAIN_CPU ; obj->read_domains =I915_GEM_DOMAIN_CPU ; - obj->cache_level =I915_CACHE_NONE ; + obj->pat_index =i915_gem_get_pat_index (i915,I915_CACHE_NONE );/* Preallocate the "backing storage" */ if (i915_gem_object_pin_pages_unlocked (obj))
@@ -359,7 +359,9 @@ static int lowlevel_hole(struct i915_address_space *vm,with_intel_runtime_pm (vm->gt->uncore->rpm, wakeref) vm->insert_entries(vm, mock_vma_res, -I915_CACHE_NONE ,0 ); +i915_gem_get_pat_index (vm->i915, +I915_CACHE_NONE ), +0 ); } count = n;
@@ -1377,7 +1379,10 @@ static int igt_ggtt_page(void *arg) ggtt->vm.insert_page (&ggtt->vm,i915_gem_object_get_dma_address (obj,0 ), - offset,I915_CACHE_NONE ,0 ); + offset, +i915_gem_get_pat_index (i915, +I915_CACHE_NONE ), +0 ); } order =i915_random_order (count, &prng);
@@ -1510,7 +1515,7 @@ static int reserve_gtt_with_resource(struct i915_vma *vma, u64 offset)mutex_lock (&vm->mutex); err =i915_gem_gtt_reserve (vm,NULL , &vma->node, obj->base.size, offset, - obj->cache_level, + obj->pat_index,0 );if (!err) {i915_vma_resource_init_from_vma (vma_res, vma);
@@ -1690,7 +1695,7 @@ static int insert_gtt_with_resource(struct i915_vma *vma)mutex_lock (&vm->mutex); err =i915_gem_gtt_insert (vm,NULL , &vma->node, obj->base.size,0 , - obj->cache_level,0 , vm->total,0 ); + obj->pat_index,0 , vm->total,0 );if (!err) {i915_vma_resource_init_from_vma (vma_res, vma); vma->resource = vma_res;
drivers/gpu/drm/i915/selftests/intel_memory_region.c +3 -1
@@ -1070,7 +1070,9 @@ static int igt_lmem_write_cpu(void *arg)/* Put the pages into a known state -- from the gpu for added fun */ intel_engine_pm_get (engine); err =intel_context_migrate_clear (engine->gt->migrate.context,NULL , - obj->mm.pages->sgl,I915_CACHE_NONE , + obj->mm.pages->sgl, +i915_gem_get_pat_index (i915, +I915_CACHE_NONE ), true,0xdeadbeaf , &rq);if (rq) { dma_resv_add_fence(obj->base.resv, &rq->fence,
drivers/gpu/drm/i915/selftests/mock_gtt.c +4 -4
@@ -27,21 +27,21 @@static void mock_insert_page (struct i915_address_space *vm,dma_addr_t addr,u64 offset, -enum i915_cache_level level, +unsigned int pat_index,u32 flags) { }static void mock_insert_entries (struct i915_address_space *vm,struct i915_vma_resource *vma_res, -enum i915_cache_level level,u32 flags) +unsigned int pat_index,u32 flags) { }static void mock_bind_ppgtt (struct i915_address_space *vm,struct i915_vm_pt_stash *stash,struct i915_vma_resource *vma_res, -enum i915_cache_level cache_level, +unsigned int pat_index,u32 flags) {GEM_BUG_ON (flags &I915_VMA_GLOBAL_BIND );
@@ -94,7 +94,7 @@ struct i915_ppgtt *mock_ppgtt(struct drm_i915_private *i915, const char *name)static void mock_bind_ggtt (struct i915_address_space *vm,struct i915_vm_pt_stash *stash,struct i915_vma_resource *vma_res, -enum i915_cache_level cache_level, +unsigned int pat_index,u32 flags) { }
Signature · 1 line
2.25.1
Re: [Intel-gfx] [PATCH 7/8] drm/i915: use pat_index instead of cache_level
Quoted 15 lines · On 20.04.2023 01:00, [email protected] wrote:
From: Fei Yang <[email protected]>
Currently the KMD is using enum
i915_cache_levelto set caching policy for buffer objects. This is flaky because the PAT index which really controls the caching behavior in PTE has far more levels than what's defined in the enum. In addition, the PAT index is platform dependent, having to translate betweeni915_cache_leveland PAT index is not reliable, and makes the code more complicated.From UMD's perspective there is also a necessity to set caching policy for
performance fine tuning. It's much easier for the UMD to directly use PAT index because the behavior of each PAT index is clearly defined in Bspec. Having the abstractedi915_cache_levelsitting in between would only cause more ambiguity.
Quoted 5 lines
For these reasons this patch replaces
i915_cache_levelwith PAT index. Also note, thecache_levelis not completely removed yet, because the KMD still has the need of creating buffer objects with simple cache settings such as cached, uncached, or writethrough. For such simple cases, usingcache_levelwould help simplify the code.
It seems quite fundamental change to me. Does this "not completely removed yet" means that in some future we will not have support for generic cache levels at all? Seems strange to me. Even looking at the number of users of i915_gem_get_pat_index below it seem very unlikely.
And if the support for generic level will stay, maybe it would be better to make usage of it more convienient. All conversion of
f(..., cache_level, ...) to f(..., i915_gem_get_pat_index(i915, cache_level), ...)
looks quite ugly to me.
Maybe extending cache level to support pat index somehow, for example:
enum i915_cache_level {
I915_CACHE_NONE = 0,
I915_CACHE_...,
...
I915_CACHE_1ST_PAT_INDEX = 0x100,
}so real_pat_index = cache_level - I915_CACHE_1ST_PAT_INDEX
and in case of generic level there will be platform dependend conversion to real_pat_index?
I do not know the whole picture so maybe this is all wrong for some reason, just asking :)
Regards
Andrzej
Quoted 211 lines
Cc Chris Wilson <[email protected]>Cc Matt Roper <[email protected]>Signed-off-by Fei Yang <[email protected]>Reviewed-by Andi Shyti <[email protected]>--- drivers/gpu/drm/i915/display/intel_dpt.c | 12 +-- drivers/gpu/drm/i915/gem/i915_gem_domain.c | 27 ++---- .../gpu/drm/i915/gem/i915_gem_execbuffer.c | 10 ++- drivers/gpu/drm/i915/gem/i915_gem_mman.c | 3 +- drivers/gpu/drm/i915/gem/i915_gem_object.c | 52 +++++++++++- drivers/gpu/drm/i915/gem/i915_gem_object.h | 4 + .../gpu/drm/i915/gem/i915_gem_object_types.h | 25 +++++- drivers/gpu/drm/i915/gem/i915_gem_stolen.c | 4 +- drivers/gpu/drm/i915/gem/i915_gem_ttm_move.c | 16 ++-- .../gpu/drm/i915/gem/selftests/huge_pages.c | 2 +- .../drm/i915/gem/selftests/i915_gem_migrate.c | 2 +- .../drm/i915/gem/selftests/i915_gem_mman.c | 2 +- drivers/gpu/drm/i915/gt/gen6_ppgtt.c | 10 ++- drivers/gpu/drm/i915/gt/gen8_ppgtt.c | 71 ++++++++-------- drivers/gpu/drm/i915/gt/gen8_ppgtt.h | 3 +- drivers/gpu/drm/i915/gt/intel_ggtt.c | 82 +++++++++---------- drivers/gpu/drm/i915/gt/intel_gtt.h | 20 ++--- drivers/gpu/drm/i915/gt/intel_migrate.c | 47 ++++++----- drivers/gpu/drm/i915/gt/intel_migrate.h | 13 ++- drivers/gpu/drm/i915/gt/intel_ppgtt.c | 6 +- drivers/gpu/drm/i915/gt/selftest_migrate.c | 47 ++++++----- drivers/gpu/drm/i915/gt/selftest_reset.c | 8 +- drivers/gpu/drm/i915/gt/selftest_timeline.c | 2 +- drivers/gpu/drm/i915/gt/selftest_tlb.c | 4 +- drivers/gpu/drm/i915/gt/uc/intel_uc_fw.c | 10 ++- drivers/gpu/drm/i915/i915_debugfs.c | 55 ++++++++++--- drivers/gpu/drm/i915/i915_gem.c | 16 +++- drivers/gpu/drm/i915/i915_gpu_error.c | 8 +- drivers/gpu/drm/i915/i915_vma.c | 16 ++-- drivers/gpu/drm/i915/i915_vma.h | 2 +- drivers/gpu/drm/i915/i915_vma_types.h | 2 - drivers/gpu/drm/i915/selftests/i915_gem.c | 5 +- .../gpu/drm/i915/selftests/i915_gem_evict.c | 4 +- drivers/gpu/drm/i915/selftests/i915_gem_gtt.c | 15 ++-- .../drm/i915/selftests/intel_memory_region.c | 4 +- drivers/gpu/drm/i915/selftests/mock_gtt.c | 8 +- 36 files changed, 378 insertions(+), 239 deletions(-)drivers/gpu/drm/i915/display/intel_dpt.c +6
-6@@ -43,24 +43,24 @@ static void gen8_set_pte(void __iomem *addr, gen8_pte_t pte)static void dpt_insert_page (struct i915_address_space *vm,dma_addr_t addr,u64 offset, -enum i915_cache_level level, +unsigned int pat_index,u32 flags) {struct i915_dpt *dpt =i915_vm_to_dpt (vm);gen8_pte_t __iomem *base = dpt->iomem;gen8_set_pte (base + offset /I915_GTT_PAGE_SIZE , - vm->pte_encode (addr, level, flags)); + vm->pte_encode (addr, pat_index, flags)); }static void dpt_insert_entries (struct i915_address_space *vm,struct i915_vma_resource *vma_res, -enum i915_cache_level level, +unsigned int pat_index,u32 flags) {struct i915_dpt *dpt =i915_vm_to_dpt (vm);gen8_pte_t __iomem *base = dpt->iomem; -const gen8_pte_t pte_encode = vm->pte_encode (0 , level, flags); +const gen8_pte_t pte_encode = vm->pte_encode (0 , pat_index, flags);struct sgt_iter sgt_iter;dma_addr_t addr;int i;@@ -83,7 +83,7 @@ static void dpt_clear_range(struct i915_address_space *vm,static void dpt_bind_vma (struct i915_address_space *vm,struct i915_vm_pt_stash *stash,struct i915_vma_resource *vma_res, -enum i915_cache_level cache_level, +unsigned int pat_index,u32 flags) {u32 pte_flags;@@ -98,7 +98,7 @@ static void dpt_bind_vma(struct i915_address_space *vm,if (vma_res->bi.lmem) pte_flags |=PTE_LM ; - vm->insert_entries (vm, vma_res, cache_level, pte_flags); + vm->insert_entries (vm, vma_res, pat_index, pte_flags); vma_res->page_sizes_gtt =I915_GTT_PAGE_SIZE ;drivers/gpu/drm/i915/gem/i915_gem_domain.c +9
-18@@ -27,8 +27,8 @@ static bool gpu_write_needs_clflush(struct drm_i915_gem_object *obj)if (IS_DGFX (i915))return false; -return !(obj->cache_level ==I915_CACHE_NONE || - obj->cache_level ==I915_CACHE_WT ); +return !(i915_gem_object_has_cache_level (obj,I915_CACHE_NONE ) || +i915_gem_object_has_cache_level (obj,I915_CACHE_WT )); }bool i915_gem_cpu_write_needs_clflush (struct drm_i915_gem_object *obj)@@ -267,7 +267,7 @@ int i915_gem_object_set_cache_level(struct drm_i915_gem_object *obj, {int ret; -if (obj->cache_level == cache_level) +if (i915_gem_object_has_cache_level (obj, cache_level))return 0 ; ret = i915_gem_object_wait(obj,@@ -278,10 +278,8 @@ int i915_gem_object_set_cache_level(struct drm_i915_gem_object *obj,return ret;/* Always invalidate stale cachelines */ -if (obj->cache_level != cache_level) { -i915_gem_object_set_cache_coherency (obj, cache_level); - obj->cache_dirty = true; - } +i915_gem_object_set_cache_coherency (obj, cache_level); + obj->cache_dirty = true;/* The cache-level will be applied when each vma is rebound. */ return i915_gem_object_unbind(obj,@@ -306,20 +304,13 @@ int i915_gem_get_caching_ioctl(struct drm_device *dev, void *data, goto out; } -switch (obj->cache_level) { -case I915_CACHE_LLC : -case I915_CACHE_L3_LLC : +if (i915_gem_object_has_cache_level (obj,I915_CACHE_LLC ) || +i915_gem_object_has_cache_level (obj,I915_CACHE_L3_LLC )) args->caching =I915_CACHING_CACHED ; -break ; - -case I915_CACHE_WT : +else if (i915_gem_object_has_cache_level (obj,I915_CACHE_WT )) args->caching =I915_CACHING_DISPLAY ; -break ; - -default : +else args->caching =I915_CACHING_NONE ; -break ; - } out:rcu_read_unlock ();return err;drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c +6
-4@@ -642,7 +642,7 @@ static inline int use_cpu_reloc(const struct reloc_cache *cache,return (cache->has_llc || obj->cache_dirty || - obj->cache_level !=I915_CACHE_NONE ); + !i915_gem_object_has_cache_level (obj,I915_CACHE_NONE )); } static int eb_reserve_vma(struct i915_execbuffer *eb,@@ -1323,8 +1323,10 @@ static void *reloc_iomap(struct i915_vma *batch, offset = cache->node.start;if (drm_mm_node_allocated (&cache->node)) { ggtt->vm.insert_page (&ggtt->vm, -i915_gem_object_get_dma_address (obj, page), - offset,I915_CACHE_NONE ,0 ); +i915_gem_object_get_dma_address (obj, page), + offset, +i915_gem_get_pat_index (ggtt->vm.i915,I915_CACHE_NONE ), +0 ); }else { offset += page <<PAGE_SHIFT ; }@@ -1464,7 +1466,7 @@ eb_relocate_entry(struct i915_execbuffer *eb,reloc_cache_unmap (&eb->reloc_cache);mutex_lock (&vma->vm->mutex); err =i915_vma_bind (target->vma, - target->vma->obj->cache_level, + target->vma->obj->pat_index,PIN_GLOBAL ,NULL ,NULL );mutex_unlock (&vma->vm->mutex);reloc_cache_remap (&eb->reloc_cache, ev->vma->obj);drivers/gpu/drm/i915/gem/i915_gem_mman.c +2
-1@@ -383,7 +383,8 @@ static vm_fault_t vm_fault_gtt(struct vm_fault *vmf) }/* Access to snoopable pages through the GTT is incoherent. */ -if (obj->cache_level !=I915_CACHE_NONE && !HAS_LLC (i915)) { +if (!(i915_gem_object_has_cache_level (obj,I915_CACHE_NONE ) || +HAS_LLC (i915))) { ret = -EFAULT ; goto err_unpin; }drivers/gpu/drm/i915/gem/i915_gem_object.c +0
-0
drivers/gpu/drm/i915/gem/i915_gem_object.c unsigned int i915_gem_get_pat_index(struct drm_i915_private *i915, · line 58
Quoted 5 lines
drivers/gpu/drm/i915/gem/i915_gem_object.c +2
-0@@ -54,6 +54,25 @@ unsigned int i915_gem_get_pat_index(struct drm_i915_private *i915,return INTEL_INFO (i915)->cachelevel_to_pat[level]; } +bool i915_gem_object_has_cache_level (const struct drm_i915_gem_object *obj, +enum i915_cache_level lvl)
The name suggest object can have more cache levels, maybe only my impression, up to you.
Quoted 1,709 lines
drivers/gpu/drm/i915/gem/i915_gem_object.c +49
-1+{ +/* +* cache_level == I915_CACHE_INVAL indicates the UMD's have set the +* caching policy through pat_index, in which case the KMD should +* leave the coherency to be managed by user space, simply return +* true here. +*/ +if (obj->cache_level ==I915_CACHE_INVAL ) +return true; + +/* +* Otherwise the pat_index should have been converted from cache_level +* so that the following comparison is valid. +*/ +return obj->pat_index ==i915_gem_get_pat_index (obj_to_i915 (obj), lvl); +} +struct drm_i915_gem_object *i915_gem_object_alloc (void ) {struct drm_i915_gem_object *obj;@@ -133,7 +152,7 @@ void i915_gem_object_set_cache_coherency(struct drm_i915_gem_object *obj, {struct drm_i915_private *i915 =to_i915 (obj->base.dev); - obj->cache_level = cache_level; + obj->pat_index =i915_gem_get_pat_index (i915, cache_level);if (cache_level !=I915_CACHE_NONE ) obj->cache_coherent = (I915_BO_CACHE_COHERENT_FOR_READ |@@ -148,6 +167,37 @@ void i915_gem_object_set_cache_coherency(struct drm_i915_gem_object *obj, !IS_DGFX (i915); } +/** +* i915_gem_object_set_pat_index - set PAT index to be used in PTE encode +* @obj: #drm_i915_gem_object +* @pat_index: PAT index +* +* This is a clone of i915_gem_object_set_cache_coherency taking pat index +* instead of cache_level as its second argument. +*/ +void i915_gem_object_set_pat_index (struct drm_i915_gem_object *obj, +unsigned int pat_index) +{ +struct drm_i915_private *i915 =to_i915 (obj->base.dev); + +if (obj->pat_index == pat_index) +return ; + + obj->pat_index = pat_index; + +if (pat_index !=i915_gem_get_pat_index (i915,I915_CACHE_NONE )) + obj->cache_coherent = (I915_BO_CACHE_COHERENT_FOR_READ | +I915_BO_CACHE_COHERENT_FOR_WRITE ); +else if (HAS_LLC (i915)) + obj->cache_coherent =I915_BO_CACHE_COHERENT_FOR_READ ; +else + obj->cache_coherent =0 ; + + obj->cache_dirty = + !(obj->cache_coherent &I915_BO_CACHE_COHERENT_FOR_WRITE ) && + !IS_DGFX (i915); +} +bool i915_gem_object_can_bypass_llc (struct drm_i915_gem_object *obj) {struct drm_i915_private *i915 =to_i915 (obj->base.dev);drivers/gpu/drm/i915/gem/i915_gem_object.h +4
-0@@ -34,6 +34,8 @@ static inline bool i915_gem_object_size_2big(u64 size)unsigned int i915_gem_get_pat_index (struct drm_i915_private *i915,enum i915_cache_level level); +bool i915_gem_object_has_cache_level (const struct drm_i915_gem_object *obj, +enum i915_cache_level lvl);void i915_gem_init__objects (struct drm_i915_private *i915);void i915_objects_module_exit (void );@@ -764,6 +766,8 @@ bool i915_gem_object_has_unknown_state(struct drm_i915_gem_object *obj);void i915_gem_object_set_cache_coherency (struct drm_i915_gem_object *obj,unsigned int cache_level); +void i915_gem_object_set_pat_index (struct drm_i915_gem_object *obj, +unsigned int pat_index);bool i915_gem_object_can_bypass_llc (struct drm_i915_gem_object *obj);void i915_gem_object_flush_if_display (struct drm_i915_gem_object *obj);void i915_gem_object_flush_if_display_locked (struct drm_i915_gem_object *obj);drivers/gpu/drm/i915/gem/i915_gem_object_types.h +22
-3@@ -195,6 +195,7 @@ enum i915_cache_level {*/ I915_CACHE_WT ,I915_MAX_CACHE_LEVEL , + I915_CACHE_INVAL = I915_MAX_CACHE_LEVEL, }; enum i915_map_type {@@ -358,10 +359,28 @@ struct drm_i915_gem_object {#define I915_BO_FLAG_STRUCT_PAGE BIT(0)/* Object backed by struct pages */ #define I915_BO_FLAG_IOMEM BIT(1)/* Object backed by IO memory */ /** -* @cache_level: The desired GTT caching level. +* @pat_index: The desired PAT index. +* +* See hardware specification for valid PAT indices for each platform. +* This field used to contain a value of enum i915_cache_level. It's +* changed to an unsigned int because PAT indices are being used by +* both UMD and KMD for caching policy control after GEN12. +* For backward compatibility, this field will continue to contain +* value of i915_cache_level for pre-GEN12 platforms so that the PTE +* encode functions for these legacy platforms can stay the same. +* In the meantime platform specific tables are created to translate +* i915_cache_level into pat index, for more details check the macros +* defined i915/i915_pci.c, e.g. PVC_CACHELEVEL. +*/ + unsigned int pat_index:6; +/** +* @cache_level: Indicate whether pat_index is set by UMD * -* See enum i915_cache_level for possible values, along with what -* each does. +* This used to hold desired GTT caching level, but is now replaced by +* pat_index. It's kept here for KMD to tell whether the pat_index is +* set by UMD or converted from enum i915_cache_level. +* This field should be 0 by default, but I915_CACHE_INVAL if the +* pat_index is set by UMD. */ unsigned int cache_level:3;/** drivers/gpu/drm/i915/gem/i915_gem_stolen.c +3
-1@@ -565,7 +565,9 @@ static void dbg_poison(struct i915_ggtt *ggtt, ggtt->vm.insert_page (&ggtt->vm, addr, ggtt->error_capture.start, -I915_CACHE_NONE ,0 ); +i915_gem_get_pat_index (ggtt->vm.i915, +I915_CACHE_NONE ), +0 );mb (); s = io_mapping_map_wc(&ggtt->iomap,drivers/gpu/drm/i915/gem/i915_gem_ttm_move.c +9
-7@@ -214,7 +214,8 @@ static struct dma_fence *i915_ttm_accel_move(struct ttm_buffer_object *bo,intel_engine_pm_get (to_gt (i915)->migrate.context->engine); ret =intel_context_migrate_clear (to_gt (i915)->migrate.context, deps, - dst_st->sgl, dst_level, + dst_st->sgl, +i915_gem_get_pat_index (i915, dst_level),i915_ttm_gtt_binds_lmem (dst_mem),0 , &rq); }else {@@ -227,12 +228,13 @@ static struct dma_fence *i915_ttm_accel_move(struct ttm_buffer_object *bo, src_level =i915_ttm_cache_level (i915, bo->resource, src_ttm);intel_engine_pm_get (to_gt (i915)->migrate.context->engine); ret =intel_context_migrate_copy (to_gt (i915)->migrate.context, - deps, src_rsgt->table.sgl, - src_level, -i915_ttm_gtt_binds_lmem (bo->resource), - dst_st->sgl, dst_level, -i915_ttm_gtt_binds_lmem (dst_mem), - &rq); + deps, src_rsgt->table.sgl, +i915_gem_get_pat_index (i915, src_level), +i915_ttm_gtt_binds_lmem (bo->resource), + dst_st->sgl, +i915_gem_get_pat_index (i915, dst_level), +i915_ttm_gtt_binds_lmem (dst_mem), + &rq);i915_refct_sgt_put (src_rsgt); }drivers/gpu/drm/i915/gem/selftests/huge_pages.c +1
-1@@ -354,7 +354,7 @@ fake_huge_pages_object(struct drm_i915_private *i915, u64 size, bool single) obj->write_domain =I915_GEM_DOMAIN_CPU ; obj->read_domains =I915_GEM_DOMAIN_CPU ; - obj->cache_level =I915_CACHE_NONE ; + obj->pat_index =i915_gem_get_pat_index (i915,I915_CACHE_NONE );return obj; }drivers/gpu/drm/i915/gem/selftests/i915_gem_migrate.c +1
-1@@ -219,7 +219,7 @@ static int __igt_lmem_pages_migrate(struct intel_gt *gt,continue ; err =intel_migrate_clear (>->migrate, &ww, deps, - obj->mm.pages->sgl, obj->cache_level, + obj->mm.pages->sgl, obj->pat_index,i915_gem_object_is_lmem (obj),0xdeadbeaf , &rq);if (rq) {drivers/gpu/drm/i915/gem/selftests/i915_gem_mman.c +1
-1@@ -1222,7 +1222,7 @@ static int __igt_mmap_migrate(struct intel_memory_region **placements, } err =intel_context_migrate_clear (to_gt (i915)->migrate.context,NULL , - obj->mm.pages->sgl, obj->cache_level, + obj->mm.pages->sgl, obj->pat_index,i915_gem_object_is_lmem (obj),expand32 (POISON_INUSE ), &rq);i915_gem_object_unpin_pages (obj);drivers/gpu/drm/i915/gt/gen6_ppgtt.c +6
-4@@ -109,7 +109,7 @@ static void gen6_ppgtt_clear_range(struct i915_address_space *vm,static void gen6_ppgtt_insert_entries (struct i915_address_space *vm,struct i915_vma_resource *vma_res, -enum i915_cache_level cache_level, +unsigned int pat_index,u32 flags) {struct i915_ppgtt *ppgtt =i915_vm_to_ppgtt (vm);@@ -117,7 +117,7 @@ static void gen6_ppgtt_insert_entries(struct i915_address_space *vm,unsigned int first_entry = vma_res->start /I915_GTT_PAGE_SIZE ;unsigned int act_pt = first_entry /GEN6_PTES ;unsigned int act_pte = first_entry %GEN6_PTES ; -const u32 pte_encode = vm->pte_encode (0 , cache_level, flags); +const u32 pte_encode = vm->pte_encode (0 , pat_index, flags);struct sgt_dma iter =sgt_dma (vma_res);gen6_pte_t *vaddr;@@ -227,7 +227,9 @@ static int gen6_ppgtt_init_scratch(struct gen6_ppgtt *ppgtt) vm->scratch[0 ]->encode = vm->pte_encode (px_dma (vm->scratch[0 ]), -I915_CACHE_NONE ,PTE_READ_ONLY ); +i915_gem_get_pat_index (vm->i915, +I915_CACHE_NONE ), +PTE_READ_ONLY ); vm->scratch[1 ] = vm->alloc_pt_dma (vm,I915_GTT_PAGE_SIZE_4K );if (IS_ERR (vm->scratch[1 ])) {@@ -278,7 +280,7 @@ static void gen6_ppgtt_cleanup(struct i915_address_space *vm)static void pd_vma_bind (struct i915_address_space *vm,struct i915_vm_pt_stash *stash,struct i915_vma_resource *vma_res, -enum i915_cache_level cache_level, +unsigned int pat_index,u32 unused) {struct i915_ggtt *ggtt =i915_vm_to_ggtt (vm);drivers/gpu/drm/i915/gt/gen8_ppgtt.c +34
-37@@ -56,7 +56,7 @@ static u64 gen8_pte_encode(dma_addr_t addr, }static u64 mtl_pte_encode (dma_addr_t addr, -enum i915_cache_level level, +unsigned int pat_index,u32 flags) {gen8_pte_t pte = addr |GEN8_PAGE_PRESENT |GEN8_PAGE_RW ;@@ -67,24 +67,17 @@ static u64 mtl_pte_encode(dma_addr_t addr,if (flags &PTE_LM ) pte |=GEN12_PPGTT_PTE_LM |GEN12_PPGTT_PTE_NC ; -switch (level) { -case I915_CACHE_NONE : - pte |=GEN12_PPGTT_PTE_PAT1 ; -break ; -case I915_CACHE_LLC : -case I915_CACHE_L3_LLC : - pte |=GEN12_PPGTT_PTE_PAT0 |GEN12_PPGTT_PTE_PAT1 ; -break ; -case I915_CACHE_WT : +if (pat_index &BIT (0 )) pte |=GEN12_PPGTT_PTE_PAT0 ; -break ; -default : -/* This should never happen. Added to deal with the compile -* error due to the addition of I915_MAX_CACHE_LEVEL. Will -* be removed by the pat_index patch. -*/ -break ; - } + +if (pat_index &BIT (1 )) + pte |=GEN12_PPGTT_PTE_PAT1 ; + +if (pat_index &BIT (2 )) + pte |=GEN12_PPGTT_PTE_PAT2 ; + +if (pat_index &BIT (3 )) + pte |=MTL_PPGTT_PTE_PAT3 ;return pte; }@@ -457,11 +450,11 @@ gen8_ppgtt_insert_pte(struct i915_ppgtt *ppgtt,struct i915_page_directory *pdp, struct sgt_dma *iter, u64 idx, - enum i915_cache_level cache_level, + unsigned int pat_index, u32 flags) { struct i915_page_directory *pd; -const gen8_pte_t pte_encode = ppgtt->vm.pte_encode (0 , cache_level, flags); +const gen8_pte_t pte_encode = ppgtt->vm.pte_encode (0 , pat_index, flags);gen8_pte_t *vaddr; pd =i915_pd_entry (pdp,gen8_pd_index (idx,2 ));@@ -504,10 +497,10 @@ static void xehpsdv_ppgtt_insert_huge(struct i915_address_space *vm, struct i915_vma_resource *vma_res, struct sgt_dma *iter, - enum i915_cache_level cache_level, + unsigned int pat_index, u32 flags) { - const gen8_pte_t pte_encode = vm->pte_encode(0, cache_level, flags); + const gen8_pte_t pte_encode = vm->pte_encode(0, pat_index, flags);unsigned int rem =sg_dma_len (iter->sg);u64 start = vma_res->start;u64 end = start + vma_res->vma_size;@@ -611,10 +604,10 @@ xehpsdv_ppgtt_insert_huge(struct i915_address_space *vm,static void gen8_ppgtt_insert_huge (struct i915_address_space *vm,struct i915_vma_resource *vma_res,struct sgt_dma *iter, -enum i915_cache_level cache_level, +unsigned int pat_index,u32 flags) { -const gen8_pte_t pte_encode = vm->pte_encode (0 , cache_level, flags); +const gen8_pte_t pte_encode = vm->pte_encode (0 , pat_index, flags);unsigned int rem =sg_dma_len (iter->sg);u64 start = vma_res->start;@@ -734,7 +727,7 @@ static void gen8_ppgtt_insert_huge(struct i915_address_space *vm,static void gen8_ppgtt_insert (struct i915_address_space *vm,struct i915_vma_resource *vma_res, -enum i915_cache_level cache_level, +unsigned int pat_index,u32 flags) {struct i915_ppgtt *const ppgtt =i915_vm_to_ppgtt (vm);@@ -742,9 +735,9 @@ static void gen8_ppgtt_insert(struct i915_address_space *vm,if (vma_res->bi.page_sizes.sg >I915_GTT_PAGE_SIZE ) {if (HAS_64K_PAGES (vm->i915)) -xehpsdv_ppgtt_insert_huge (vm, vma_res, &iter, cache_level, flags); +xehpsdv_ppgtt_insert_huge (vm, vma_res, &iter, pat_index, flags);else -gen8_ppgtt_insert_huge (vm, vma_res, &iter, cache_level, flags); +gen8_ppgtt_insert_huge (vm, vma_res, &iter, pat_index, flags); }else {u64 idx = vma_res->start >>GEN8_PTE_SHIFT ;@@ -753,7 +746,7 @@ static void gen8_ppgtt_insert(struct i915_address_space *vm,gen8_pdp_for_page_index (vm, idx); idx =gen8_ppgtt_insert_pte (ppgtt, pdp, &iter, idx, - cache_level, flags); + pat_index, flags); }while (idx); vma_res->page_sizes_gtt =I915_GTT_PAGE_SIZE ;@@ -763,7 +756,7 @@ static void gen8_ppgtt_insert(struct i915_address_space *vm,static void gen8_ppgtt_insert_entry (struct i915_address_space *vm,dma_addr_t addr,u64 offset, -enum i915_cache_level level, +unsigned int pat_index,u32 flags) {u64 idx = offset >>GEN8_PTE_SHIFT ;@@ -777,14 +770,14 @@ static void gen8_ppgtt_insert_entry(struct i915_address_space *vm,GEM_BUG_ON (pt->is_compact); vaddr =px_vaddr (pt); - vaddr[gen8_pd_index (idx,0 )] = vm->pte_encode (addr, level, flags); + vaddr[gen8_pd_index (idx,0 )] = vm->pte_encode (addr, pat_index, flags);drm_clflush_virt_range (&vaddr[gen8_pd_index (idx,0 )],sizeof (*vaddr)); }static void __xehpsdv_ppgtt_insert_entry_lm (struct i915_address_space *vm,dma_addr_t addr,u64 offset, -enum i915_cache_level level, +unsigned int pat_index,u32 flags) {u64 idx = offset >>GEN8_PTE_SHIFT ;@@ -807,20 +800,20 @@ static void __xehpsdv_ppgtt_insert_entry_lm(struct i915_address_space *vm, } vaddr =px_vaddr (pt); - vaddr[gen8_pd_index (idx,0 ) /16 ] = vm->pte_encode (addr, level, flags); + vaddr[gen8_pd_index (idx,0 ) /16 ] = vm->pte_encode (addr, pat_index, flags); }static void xehpsdv_ppgtt_insert_entry (struct i915_address_space *vm,dma_addr_t addr,u64 offset, -enum i915_cache_level level, +unsigned int pat_index,u32 flags) {if (flags &PTE_LM )return __xehpsdv_ppgtt_insert_entry_lm (vm, addr, offset, - level, flags); + pat_index, flags); -return gen8_ppgtt_insert_entry (vm, addr, offset, level, flags); +return gen8_ppgtt_insert_entry (vm, addr, offset, pat_index, flags); }static int gen8_init_scratch (struct i915_address_space *vm)@@ -855,7 +848,9 @@ static int gen8_init_scratch(struct i915_address_space *vm) vm->scratch[0 ]->encode = vm->pte_encode (px_dma (vm->scratch[0 ]), -I915_CACHE_NONE , pte_flags); +i915_gem_get_pat_index (vm->i915, +I915_CACHE_NONE ), + pte_flags);for (i =1 ; i <= vm->top; i++) {struct drm_i915_gem_object *obj;@@ -873,7 +868,9 @@ static int gen8_init_scratch(struct i915_address_space *vm) }fill_px (obj, vm->scratch[i -1 ]->encode); - obj->encode =gen8_pde_encode (px_dma (obj),I915_CACHE_NONE ); + obj->encode =gen8_pde_encode (px_dma (obj), +i915_gem_get_pat_index (vm->i915, +I915_CACHE_NONE )); vm->scratch[i] = obj; }drivers/gpu/drm/i915/gt/gen8_ppgtt.h +1
-2@@ -10,13 +10,12 @@struct i915_address_space ;struct intel_gt ; -enum i915_cache_level ;struct i915_ppgtt *gen8_ppgtt_create (struct intel_gt *gt,unsigned long lmem_pt_obj_flags);u64 gen8_ggtt_pte_encode (dma_addr_t addr, -enum i915_cache_level level, +unsigned int pat_index,u32 flags);#endif drivers/gpu/drm/i915/gt/intel_ggtt.c +40
-42@@ -221,7 +221,7 @@ static void guc_ggtt_invalidate(struct i915_ggtt *ggtt) }static u64 mtl_ggtt_pte_encode (dma_addr_t addr, -enum i915_cache_level level, +unsigned int pat_index,u32 flags) {gen8_pte_t pte = addr |GEN8_PAGE_PRESENT ;@@ -231,30 +231,17 @@ static u64 mtl_ggtt_pte_encode(dma_addr_t addr,if (flags &PTE_LM ) pte |=GEN12_GGTT_PTE_LM ; -switch (level) { -case I915_CACHE_NONE : - pte |=MTL_GGTT_PTE_PAT1 ; -break ; -case I915_CACHE_LLC : -case I915_CACHE_L3_LLC : - pte |=MTL_GGTT_PTE_PAT0 |MTL_GGTT_PTE_PAT1 ; -break ; -case I915_CACHE_WT : +if (pat_index &BIT (0 )) pte |=MTL_GGTT_PTE_PAT0 ; -break ; -default : -/* This should never happen. Added to deal with the compile -* error due to the addition of I915_MAX_CACHE_LEVEL. Will -* be removed by the pat_index patch. -*/ -break ; - } + +if (pat_index &BIT (1 )) + pte |=MTL_GGTT_PTE_PAT1 ;return pte; }u64 gen8_ggtt_pte_encode (dma_addr_t addr, -enum i915_cache_level level, +unsigned int pat_index,u32 flags) {gen8_pte_t pte = addr |GEN8_PAGE_PRESENT ;@@ -273,25 +260,25 @@ static void gen8_set_pte(void __iomem *addr, gen8_pte_t pte)static void gen8_ggtt_insert_page (struct i915_address_space *vm,dma_addr_t addr,u64 offset, -enum i915_cache_level level, +unsigned int pat_index,u32 flags) {struct i915_ggtt *ggtt =i915_vm_to_ggtt (vm);gen8_pte_t __iomem *pte = (gen8_pte_t __iomem *)ggtt->gsm + offset /I915_GTT_PAGE_SIZE ; -gen8_set_pte (pte, ggtt->vm.pte_encode (addr, level, flags)); +gen8_set_pte (pte, ggtt->vm.pte_encode (addr, pat_index, flags)); ggtt->invalidate (ggtt); }static void gen8_ggtt_insert_entries (struct i915_address_space *vm,struct i915_vma_resource *vma_res, -enum i915_cache_level level, +unsigned int pat_index,u32 flags) {struct i915_ggtt *ggtt =i915_vm_to_ggtt (vm); -const gen8_pte_t pte_encode = ggtt->vm.pte_encode (0 , level, flags); +const gen8_pte_t pte_encode = ggtt->vm.pte_encode (0 , pat_index, flags);gen8_pte_t __iomem *gte;gen8_pte_t __iomem *end;struct sgt_iter iter;@@ -348,14 +335,14 @@ static void gen8_ggtt_clear_range(struct i915_address_space *vm,static void gen6_ggtt_insert_page (struct i915_address_space *vm,dma_addr_t addr,u64 offset, -enum i915_cache_level level, +unsigned int pat_index,u32 flags) {struct i915_ggtt *ggtt =i915_vm_to_ggtt (vm);gen6_pte_t __iomem *pte = (gen6_pte_t __iomem *)ggtt->gsm + offset /I915_GTT_PAGE_SIZE ; -iowrite32 (vm->pte_encode (addr, level, flags), pte); +iowrite32 (vm->pte_encode (addr, pat_index, flags), pte); ggtt->invalidate (ggtt); }@@ -368,7 +355,7 @@ static void gen6_ggtt_insert_page(struct i915_address_space *vm,*/ static void gen6_ggtt_insert_entries (struct i915_address_space *vm,struct i915_vma_resource *vma_res, -enum i915_cache_level level, +unsigned int pat_index,u32 flags) {struct i915_ggtt *ggtt =i915_vm_to_ggtt (vm);@@ -385,7 +372,7 @@ static void gen6_ggtt_insert_entries(struct i915_address_space *vm,iowrite32 (vm->scratch[0 ]->encode, gte++); end += (vma_res->node_size + vma_res->guard) /I915_GTT_PAGE_SIZE ;for_each_sgt_daddr (addr, iter, vma_res->bi.pages) -iowrite32 (vm->pte_encode (addr, level, flags), gte++); +iowrite32 (vm->pte_encode (addr, pat_index, flags), gte++);GEM_BUG_ON (gte > end);/* Fill the allocated but "unused" space beyond the end of the buffer */ @@ -420,14 +407,15 @@ struct insert_page {struct i915_address_space *vm;dma_addr_t addr;u64 offset; -enum i915_cache_level level; +unsigned int pat_index; };static int bxt_vtd_ggtt_insert_page__cb (void *_arg) {struct insert_page *arg = _arg; -gen8_ggtt_insert_page (arg->vm, arg->addr, arg->offset, arg->level,0 ); +gen8_ggtt_insert_page (arg->vm, arg->addr, arg->offset, + arg->pat_index,0 );bxt_vtd_ggtt_wa (arg->vm);return 0 ;@@ -436,10 +424,10 @@ static int bxt_vtd_ggtt_insert_page__cb(void *_arg)static void bxt_vtd_ggtt_insert_page__BKL (struct i915_address_space *vm,dma_addr_t addr,u64 offset, -enum i915_cache_level level, +unsigned int pat_index,u32 unused) { -struct insert_page arg = { vm, addr, offset, level }; +struct insert_page arg = { vm, addr, offset, pat_index };stop_machine (bxt_vtd_ggtt_insert_page__cb, &arg,NULL ); }@@ -447,7 +435,7 @@ static void bxt_vtd_ggtt_insert_page__BKL(struct i915_address_space *vm,struct insert_entries {struct i915_address_space *vm;struct i915_vma_resource *vma_res; -enum i915_cache_level level; +unsigned int pat_index;u32 flags; };@@ -455,7 +443,8 @@ static int bxt_vtd_ggtt_insert_entries__cb(void *_arg) {struct insert_entries *arg = _arg; -gen8_ggtt_insert_entries (arg->vm, arg->vma_res, arg->level, arg->flags); +gen8_ggtt_insert_entries (arg->vm, arg->vma_res, + arg->pat_index, arg->flags);bxt_vtd_ggtt_wa (arg->vm);return 0 ;@@ -463,10 +452,10 @@ static int bxt_vtd_ggtt_insert_entries__cb(void *_arg)static void bxt_vtd_ggtt_insert_entries__BKL (struct i915_address_space *vm,struct i915_vma_resource *vma_res, -enum i915_cache_level level, +unsigned int pat_index,u32 flags) { -struct insert_entries arg = { vm, vma_res, level, flags }; +struct insert_entries arg = { vm, vma_res, pat_index, flags };stop_machine (bxt_vtd_ggtt_insert_entries__cb, &arg,NULL ); }@@ -495,7 +484,7 @@ static void gen6_ggtt_clear_range(struct i915_address_space *vm,void intel_ggtt_bind_vma (struct i915_address_space *vm,struct i915_vm_pt_stash *stash,struct i915_vma_resource *vma_res, -enum i915_cache_level cache_level, +unsigned int pat_index,u32 flags) {u32 pte_flags;@@ -512,7 +501,7 @@ void intel_ggtt_bind_vma(struct i915_address_space *vm,if (vma_res->bi.lmem) pte_flags |=PTE_LM ; - vm->insert_entries (vm, vma_res, cache_level, pte_flags); + vm->insert_entries (vm, vma_res, pat_index, pte_flags); vma_res->page_sizes_gtt =I915_GTT_PAGE_SIZE ; }@@ -661,7 +650,7 @@ static int init_ggtt(struct i915_ggtt *ggtt)static void aliasing_gtt_bind_vma (struct i915_address_space *vm,struct i915_vm_pt_stash *stash,struct i915_vma_resource *vma_res, -enum i915_cache_level cache_level, +unsigned int pat_index,u32 flags) {u32 pte_flags;@@ -673,10 +662,10 @@ static void aliasing_gtt_bind_vma(struct i915_address_space *vm,if (flags &I915_VMA_LOCAL_BIND )ppgtt_bind_vma (&i915_vm_to_ggtt (vm)->alias->vm, - stash, vma_res, cache_level, flags); + stash, vma_res, pat_index, flags);if (flags &I915_VMA_GLOBAL_BIND ) - vm->insert_entries (vm, vma_res, cache_level, pte_flags); + vm->insert_entries (vm, vma_res, pat_index, pte_flags); vma_res->bound_flags |= flags; }@@ -933,7 +922,9 @@ static int ggtt_probe_common(struct i915_ggtt *ggtt, u64 size) ggtt->vm.scratch[0 ]->encode = ggtt->vm.pte_encode (px_dma (ggtt->vm.scratch[0 ]), -I915_CACHE_NONE , pte_flags); +i915_gem_get_pat_index (i915, +I915_CACHE_NONE ), + pte_flags);return 0 ; }@@ -1022,6 +1013,11 @@ static int gen8_gmch_probe(struct i915_ggtt *ggtt)return ggtt_probe_common (ggtt, size); } +/* +* For pre-gen8 platforms pat_index is the same as enum i915_cache_level, +* so these PTE encode functions are left with using cache_level. +* See translation table LEGACY_CACHELEVEL. +*/ static u64 snb_pte_encode (dma_addr_t addr,enum i915_cache_level level,u32 flags)@@ -1302,7 +1298,9 @@ bool i915_ggtt_resume_vm(struct i915_address_space *vm)*/ vma->resource->bound_flags =0 ; vma->ops->bind_vma (vm,NULL , vma->resource, - obj ? obj->cache_level :0 , + obj ? obj->pat_index : +i915_gem_get_pat_index (vm->i915, +I915_CACHE_NONE ), was_bound);if (obj) {/* only used during resume => exclusive access */ drivers/gpu/drm/i915/gt/intel_gtt.h +9
-11@@ -165,8 +165,6 @@ typedef u64 gen8_pte_t;#define MTL_2_COH_1W REG_FIELD_PREP(MTL_PAT_INDEX_COH_MODE_MASK, 2)#define MTL_0_COH_NON REG_FIELD_PREP(MTL_PAT_INDEX_COH_MODE_MASK, 0) -enum i915_cache_level ; -struct drm_i915_gem_object ;struct i915_fence_reg ;struct i915_vma ;@@ -234,7 +232,7 @@ struct i915_vma_ops {void (*bind_vma)(struct i915_address_space *vm,struct i915_vm_pt_stash *stash,struct i915_vma_resource *vma_res, -enum i915_cache_level cache_level, +unsigned int pat_index,u32 flags);/* * Unmap an object from an address space. This usually consists of @@ -306,7 +304,7 @@ struct i915_address_space { (*alloc_scratch_dma)(struct i915_address_space *vm, int sz);u64 (*pte_encode)(dma_addr_t addr, -enum i915_cache_level level, +unsigned int pat_index,u32 flags);/* Create a valid PTE */ #define PTE_READ_ONLY BIT(0)#define PTE_LM BIT(1)@@ -321,20 +319,20 @@ struct i915_address_space {void (*insert_page)(struct i915_address_space *vm,dma_addr_t addr,u64 offset, -enum i915_cache_level cache_level, +unsigned int pat_index,u32 flags);void (*insert_entries)(struct i915_address_space *vm,struct i915_vma_resource *vma_res, -enum i915_cache_level cache_level, +unsigned int pat_index,u32 flags);void (*raw_insert_page)(struct i915_address_space *vm,dma_addr_t addr,u64 offset, -enum i915_cache_level cache_level, +unsigned int pat_index,u32 flags);void (*raw_insert_entries)(struct i915_address_space *vm,struct i915_vma_resource *vma_res, -enum i915_cache_level cache_level, +unsigned int pat_index,u32 flags);void (*cleanup)(struct i915_address_space *vm);@@ -581,7 +579,7 @@ void ppgtt_init(struct i915_ppgtt *ppgtt, struct intel_gt *gt,void intel_ggtt_bind_vma (struct i915_address_space *vm,struct i915_vm_pt_stash *stash,struct i915_vma_resource *vma_res, -enum i915_cache_level cache_level, +unsigned int pat_index,u32 flags);void intel_ggtt_unbind_vma (struct i915_address_space *vm,struct i915_vma_resource *vma_res);@@ -639,7 +637,7 @@ void __set_pd_entry(struct i915_page_directory * const pd, const unsigned short idx, struct i915_page_table *pt, - u64 (*encode)(const dma_addr_t, const enum i915_cache_level)); + u64 (*encode)(const dma_addr_t, const unsigned int pat_index));#define set_pd_entry (pd, idx, to) \ __set_pd_entry((pd), (idx), px_pt(to), gen8_pde_encode)@@ -659,7 +657,7 @@ void gen6_ggtt_invalidate(struct i915_ggtt *ggtt);void ppgtt_bind_vma (struct i915_address_space *vm,struct i915_vm_pt_stash *stash,struct i915_vma_resource *vma_res, -enum i915_cache_level cache_level, +unsigned int pat_index,u32 flags);void ppgtt_unbind_vma (struct i915_address_space *vm,struct i915_vma_resource *vma_res);drivers/gpu/drm/i915/gt/intel_migrate.c +26
-21@@ -45,7 +45,9 @@ static void xehpsdv_toggle_pdes(struct i915_address_space *vm,* Insert a dummy PTE into every PT that will map to LMEM to ensure * we have a correctly setup PDE structure for later use. */ - vm->insert_page (vm,0 , d->offset,I915_CACHE_NONE ,PTE_LM ); + vm->insert_page (vm,0 , d->offset, +i915_gem_get_pat_index (vm->i915,I915_CACHE_NONE ), +PTE_LM );GEM_BUG_ON (!pt->is_compact); d->offset +=SZ_2M ; }@@ -63,7 +65,9 @@ static void xehpsdv_insert_pte(struct i915_address_space *vm,* alignment is 64K underneath for the pt, and we are careful * not to access the space in the void. */ - vm->insert_page (vm,px_dma (pt), d->offset,I915_CACHE_NONE ,PTE_LM ); + vm->insert_page (vm,px_dma (pt), d->offset, +i915_gem_get_pat_index (vm->i915,I915_CACHE_NONE ), +PTE_LM ); d->offset +=SZ_64K ; }@@ -73,7 +77,8 @@ static void insert_pte(struct i915_address_space *vm, {struct insert_pte_data *d = data; - vm->insert_page (vm,px_dma (pt), d->offset,I915_CACHE_NONE , + vm->insert_page (vm,px_dma (pt), d->offset, +i915_gem_get_pat_index (vm->i915,I915_CACHE_NONE ),i915_gem_object_is_lmem (pt->base) ?PTE_LM :0 ); d->offset +=PAGE_SIZE ; }@@ -356,13 +361,13 @@ static int max_pte_pkt_size(struct i915_request *rq, int pkt)static int emit_pte (struct i915_request *rq,struct sgt_dma *it, -enum i915_cache_level cache_level, +unsigned int pat_index,bool is_lmem,u64 offset,int length) {bool has_64K_pages =HAS_64K_PAGES (rq->engine->i915); -const u64 encode = rq->context->vm->pte_encode (0 , cache_level, +const u64 encode = rq->context->vm->pte_encode (0 , pat_index, is_lmem ?PTE_LM :0 );struct intel_ring *ring = rq->ring;int pkt, dword_length;@@ -673,17 +678,17 @@ int intel_context_migrate_copy(struct intel_context *ce, const struct i915_deps *deps, struct scatterlist *src, - enum i915_cache_level src_cache_level, + unsigned int src_pat_index, bool src_is_lmem, struct scatterlist *dst, - enum i915_cache_level dst_cache_level, + unsigned int dst_pat_index, bool dst_is_lmem, struct i915_request **out) { struct sgt_dma it_src = sg_sgt(src), it_dst = sg_sgt(dst), it_ccs;struct drm_i915_private *i915 = ce->engine->i915;u64 ccs_bytes_to_cpy =0 , bytes_to_cpy; -enum i915_cache_level ccs_cache_level; +unsigned int ccs_pat_index;u32 src_offset, dst_offset;u8 src_access, dst_access;struct i915_request *rq;@@ -707,12 +712,12 @@ intel_context_migrate_copy(struct intel_context *ce, dst_sz =scatter_list_length (dst);if (src_is_lmem) { it_ccs = it_dst; - ccs_cache_level = dst_cache_level; + ccs_pat_index = dst_pat_index; ccs_is_src = false; }else if (dst_is_lmem) { bytes_to_cpy = dst_sz; it_ccs = it_src; - ccs_cache_level = src_cache_level; + ccs_pat_index = src_pat_index; ccs_is_src = true; }@@ -773,7 +778,7 @@ intel_context_migrate_copy(struct intel_context *ce, src_sz =calculate_chunk_sz (i915, src_is_lmem, bytes_to_cpy, ccs_bytes_to_cpy); - len =emit_pte (rq, &it_src, src_cache_level, src_is_lmem, + len =emit_pte (rq, &it_src, src_pat_index, src_is_lmem, src_offset, src_sz);if (!len) { err = -EINVAL ;@@ -784,7 +789,7 @@ intel_context_migrate_copy(struct intel_context *ce, goto out_rq; } - err =emit_pte (rq, &it_dst, dst_cache_level, dst_is_lmem, + err =emit_pte (rq, &it_dst, dst_pat_index, dst_is_lmem, dst_offset, len);if (err <0 ) goto out_rq;@@ -811,7 +816,7 @@ intel_context_migrate_copy(struct intel_context *ce, goto out_rq; ccs_sz =GET_CCS_BYTES (i915, len); - err =emit_pte (rq, &it_ccs, ccs_cache_level, false, + err =emit_pte (rq, &it_ccs, ccs_pat_index, false, ccs_is_src ? src_offset : dst_offset, ccs_sz);if (err <0 )@@ -979,7 +984,7 @@ int intel_context_migrate_clear(struct intel_context *ce, const struct i915_deps *deps, struct scatterlist *sg, - enum i915_cache_level cache_level, + unsigned int pat_index, bool is_lmem, u32 value, struct i915_request **out)@@ -1027,7 +1032,7 @@ intel_context_migrate_clear(struct intel_context *ce,if (err) goto out_rq; - len =emit_pte (rq, &it, cache_level, is_lmem, offset,CHUNK_SZ ); + len =emit_pte (rq, &it, pat_index, is_lmem, offset,CHUNK_SZ );if (len <=0 ) { err = len; goto out_rq;@@ -1074,10 +1079,10 @@ int intel_migrate_copy(struct intel_migrate *m,struct i915_gem_ww_ctx *ww, const struct i915_deps *deps, struct scatterlist *src, - enum i915_cache_level src_cache_level, + unsigned int src_pat_index, bool src_is_lmem, struct scatterlist *dst, - enum i915_cache_level dst_cache_level, + unsigned int dst_pat_index, bool dst_is_lmem, struct i915_request **out) {@@ -1098,8 +1103,8 @@ int intel_migrate_copy(struct intel_migrate *m, goto out; err =intel_context_migrate_copy (ce, deps, - src, src_cache_level, src_is_lmem, - dst, dst_cache_level, dst_is_lmem, + src, src_pat_index, src_is_lmem, + dst, dst_pat_index, dst_is_lmem, out);intel_context_unpin (ce);@@ -1113,7 +1118,7 @@ intel_migrate_clear(struct intel_migrate *m,struct i915_gem_ww_ctx *ww, const struct i915_deps *deps, struct scatterlist *sg, - enum i915_cache_level cache_level, + unsigned int pat_index, bool is_lmem, u32 value, struct i915_request **out)@@ -1134,7 +1139,7 @@ intel_migrate_clear(struct intel_migrate *m,if (err) goto out; - err =intel_context_migrate_clear (ce, deps, sg, cache_level, + err =intel_context_migrate_clear (ce, deps, sg, pat_index, is_lmem, value, out);intel_context_unpin (ce);drivers/gpu/drm/i915/gt/intel_migrate.h +6
-7@@ -16,7 +16,6 @@ struct i915_request;struct i915_gem_ww_ctx ;struct intel_gt ;struct scatterlist ; -enum i915_cache_level ;int intel_migrate_init (struct intel_migrate *m,struct intel_gt *gt);@@ -26,20 +25,20 @@ int intel_migrate_copy(struct intel_migrate *m,struct i915_gem_ww_ctx *ww, const struct i915_deps *deps, struct scatterlist *src, - enum i915_cache_level src_cache_level, + unsigned int src_pat_index, bool src_is_lmem, struct scatterlist *dst, - enum i915_cache_level dst_cache_level, + unsigned int dst_pat_index, bool dst_is_lmem, struct i915_request **out);int intel_context_migrate_copy (struct intel_context *ce,const struct i915_deps *deps,struct scatterlist *src, -enum i915_cache_level src_cache_level, +unsigned int src_pat_index,bool src_is_lmem,struct scatterlist *dst, -enum i915_cache_level dst_cache_level, +unsigned int dst_pat_index,bool dst_is_lmem,struct i915_request **out);@@ -48,7 +47,7 @@ intel_migrate_clear(struct intel_migrate *m,struct i915_gem_ww_ctx *ww, const struct i915_deps *deps, struct scatterlist *sg, - enum i915_cache_level cache_level, + unsigned int pat_index, bool is_lmem, u32 value, struct i915_request **out);@@ -56,7 +55,7 @@ int intel_context_migrate_clear(struct intel_context *ce, const struct i915_deps *deps, struct scatterlist *sg, - enum i915_cache_level cache_level, + unsigned int pat_index, bool is_lmem, u32 value, struct i915_request **out);drivers/gpu/drm/i915/gt/intel_ppgtt.c +3
-3@@ -98,7 +98,7 @@ void __set_pd_entry(struct i915_page_directory * const pd, const unsigned short idx, struct i915_page_table * const to, - u64 (*encode)(const dma_addr_t, const enum i915_cache_level)) + u64 (*encode)(const dma_addr_t, const unsigned int)) { /* Each thread pre-pins the pd, and we may have a thread per pde. */ GEM_BUG_ON(atomic_read(px_used(pd)) > NALLOC * I915_PDES);@@ -181,7 +181,7 @@ struct i915_ppgtt *i915_ppgtt_create(struct intel_gt *gt,void ppgtt_bind_vma (struct i915_address_space *vm,struct i915_vm_pt_stash *stash,struct i915_vma_resource *vma_res, -enum i915_cache_level cache_level, +unsigned int pat_index,u32 flags) {u32 pte_flags;@@ -199,7 +199,7 @@ void ppgtt_bind_vma(struct i915_address_space *vm,if (vma_res->bi.lmem) pte_flags |=PTE_LM ; - vm->insert_entries (vm, vma_res, cache_level, pte_flags); + vm->insert_entries (vm, vma_res, pat_index, pte_flags);wmb (); }drivers/gpu/drm/i915/gt/selftest_migrate.c +25
-22@@ -137,7 +137,7 @@ static int copy(struct intel_migrate *migrate,static int intel_context_copy_ccs (struct intel_context *ce,const struct i915_deps *deps,struct scatterlist *sg, -enum i915_cache_level cache_level, +unsigned int pat_index,bool write_to_ccs,struct i915_request **out) {@@ -185,7 +185,7 @@ static int intel_context_copy_ccs(struct intel_context *ce,if (err) goto out_rq; - len =emit_pte (rq, &it, cache_level, true, offset,CHUNK_SZ ); + len =emit_pte (rq, &it, pat_index, true, offset,CHUNK_SZ );if (len <=0 ) { err = len; goto out_rq;@@ -223,7 +223,7 @@ intel_migrate_ccs_copy(struct intel_migrate *m,struct i915_gem_ww_ctx *ww, const struct i915_deps *deps, struct scatterlist *sg, - enum i915_cache_level cache_level, + unsigned int pat_index, bool write_to_ccs, struct i915_request **out) {@@ -243,7 +243,7 @@ intel_migrate_ccs_copy(struct intel_migrate *m,if (err) goto out; - err =intel_context_copy_ccs (ce, deps, sg, cache_level, + err =intel_context_copy_ccs (ce, deps, sg, pat_index, write_to_ccs, out);intel_context_unpin (ce);@@ -300,7 +300,7 @@ static int clear(struct intel_migrate *migrate,/* Write the obj data into ccs surface */ err =intel_migrate_ccs_copy (migrate, &ww,NULL , obj->mm.pages->sgl, - obj->cache_level, + obj->pat_index, true, &rq);if (rq && !err) {if (i915_request_wait (rq,0 ,HZ ) <0 ) {@@ -351,7 +351,7 @@ static int clear(struct intel_migrate *migrate, err =intel_migrate_ccs_copy (migrate, &ww,NULL , obj->mm.pages->sgl, - obj->cache_level, + obj->pat_index, false, &rq);if (rq && !err) {if (i915_request_wait (rq,0 ,HZ ) <0 ) {@@ -414,9 +414,9 @@ static int __migrate_copy(struct intel_migrate *migrate, struct i915_request **out) {return intel_migrate_copy (migrate, ww,NULL , - src->mm.pages->sgl, src->cache_level, + src->mm.pages->sgl, src->pat_index,i915_gem_object_is_lmem (src), - dst->mm.pages->sgl, dst->cache_level, + dst->mm.pages->sgl, dst->pat_index,i915_gem_object_is_lmem (dst), out); }@@ -428,9 +428,9 @@ static int __global_copy(struct intel_migrate *migrate, struct i915_request **out) {return intel_context_migrate_copy (migrate->context,NULL , - src->mm.pages->sgl, src->cache_level, + src->mm.pages->sgl, src->pat_index,i915_gem_object_is_lmem (src), - dst->mm.pages->sgl, dst->cache_level, + dst->mm.pages->sgl, dst->pat_index,i915_gem_object_is_lmem (dst), out); }@@ -455,7 +455,7 @@ static int __migrate_clear(struct intel_migrate *migrate, {return intel_migrate_clear (migrate, ww,NULL , obj->mm.pages->sgl, - obj->cache_level, + obj->pat_index,i915_gem_object_is_lmem (obj), value, out); }@@ -468,7 +468,7 @@ static int __global_clear(struct intel_migrate *migrate, {return intel_context_migrate_clear (migrate->context,NULL , obj->mm.pages->sgl, - obj->cache_level, + obj->pat_index,i915_gem_object_is_lmem (obj), value, out); }@@ -648,7 +648,7 @@ static int live_emit_pte_full_ring(void *arg)*/ pr_info ("%s emite_pte ring space=%u\n" , __func__, rq->ring->space); it =sg_sgt (obj->mm.pages->sgl); - len =emit_pte (rq, &it, obj->cache_level, false,0 ,CHUNK_SZ ); + len =emit_pte (rq, &it, obj->pat_index, false,0 ,CHUNK_SZ );if (!len) { err = -EINVAL ; goto out_rq;@@ -844,7 +844,7 @@ static int wrap_ktime_compare(const void *A, const void *B)static int __perf_clear_blt (struct intel_context *ce,struct scatterlist *sg, -enum i915_cache_level cache_level, +unsigned int pat_index,bool is_lmem,size_t sz) {@@ -858,7 +858,7 @@ static int __perf_clear_blt(struct intel_context *ce, t0 =ktime_get (); - err =intel_context_migrate_clear (ce,NULL , sg, cache_level, + err =intel_context_migrate_clear (ce,NULL , sg, pat_index, is_lmem,0 , &rq);if (rq) {if (i915_request_wait (rq,0 ,MAX_SCHEDULE_TIMEOUT ) <0 )@@ -904,7 +904,8 @@ static int perf_clear_blt(void *arg) err =__perf_clear_blt (gt->migrate.context, dst->mm.pages->sgl, -I915_CACHE_NONE , +i915_gem_get_pat_index (gt->i915, +I915_CACHE_NONE ),i915_gem_object_is_lmem (dst), sizes[i]);@@ -919,10 +920,10 @@ static int perf_clear_blt(void *arg)static int __perf_copy_blt (struct intel_context *ce,struct scatterlist *src, -enum i915_cache_level src_cache_level, +unsigned int src_pat_index,bool src_is_lmem,struct scatterlist *dst, -enum i915_cache_level dst_cache_level, +unsigned int dst_pat_index,bool dst_is_lmem,size_t sz) {@@ -937,9 +938,9 @@ static int __perf_copy_blt(struct intel_context *ce, t0 =ktime_get (); err =intel_context_migrate_copy (ce,NULL , - src, src_cache_level, + src, src_pat_index, src_is_lmem, - dst, dst_cache_level, + dst, dst_pat_index, dst_is_lmem, &rq);if (rq) {@@ -994,10 +995,12 @@ static int perf_copy_blt(void *arg) err =__perf_copy_blt (gt->migrate.context, src->mm.pages->sgl, -I915_CACHE_NONE , +i915_gem_get_pat_index (gt->i915, +I915_CACHE_NONE ),i915_gem_object_is_lmem (src), dst->mm.pages->sgl, -I915_CACHE_NONE , +i915_gem_get_pat_index (gt->i915, +I915_CACHE_NONE ),i915_gem_object_is_lmem (dst), sz);drivers/gpu/drm/i915/gt/selftest_reset.c +6
-2@@ -86,7 +86,9 @@ __igt_reset_stolen(struct intel_gt *gt, ggtt->vm.insert_page (&ggtt->vm, dma, ggtt->error_capture.start, -I915_CACHE_NONE ,0 ); +i915_gem_get_pat_index (gt->i915, +I915_CACHE_NONE ), +0 );mb (); s = io_mapping_map_wc(&ggtt->iomap,@@ -127,7 +129,9 @@ __igt_reset_stolen(struct intel_gt *gt, ggtt->vm.insert_page (&ggtt->vm, dma, ggtt->error_capture.start, -I915_CACHE_NONE ,0 ); +i915_gem_get_pat_index (gt->i915, +I915_CACHE_NONE ), +0 );mb (); s = io_mapping_map_wc(&ggtt->iomap,drivers/gpu/drm/i915/gt/selftest_timeline.c +1
-1@@ -836,7 +836,7 @@ static int setup_watcher(struct hwsp_watcher *w, struct intel_gt *gt,return PTR_ERR (obj);/* keep the same cache settings as timeline */ -i915_gem_object_set_cache_coherency (obj, tl->hwsp_ggtt->obj->cache_level); +i915_gem_object_set_pat_index (obj, tl->hwsp_ggtt->obj->pat_index); w->map =i915_gem_object_pin_map_unlocked (obj,page_unmask_bits (tl->hwsp_ggtt->obj->mm.mapping));if (IS_ERR (w->map)) {drivers/gpu/drm/i915/gt/selftest_tlb.c +3
-1@@ -36,6 +36,8 @@ pte_tlbinv(struct intel_context *ce, u64 length, struct rnd_state *prng) { +const unsigned int pat_index = +i915_gem_get_pat_index (ce->vm->i915,I915_CACHE_NONE );struct drm_i915_gem_object *batch;struct drm_mm_node vb_node;struct i915_request *rq;@@ -155,7 +157,7 @@ pte_tlbinv(struct intel_context *ce,/* Flip the PTE between A and B */ if (i915_gem_object_is_lmem (vb->obj)) pte_flags |=PTE_LM ; - ce->vm->insert_entries (ce->vm, &vb_res,0 , pte_flags); + ce->vm->insert_entries (ce->vm, &vb_res, pat_index, pte_flags);/* Flush the PTE update to concurrent HW */ tlbinv (ce->vm, addr & -length, length);drivers/gpu/drm/i915/gt/uc/intel_uc_fw.c +8
-2@@ -890,9 +890,15 @@ static void uc_fw_bind_ggtt(struct intel_uc_fw *uc_fw) pte_flags |=PTE_LM ;if (ggtt->vm.raw_insert_entries) - ggtt->vm.raw_insert_entries (&ggtt->vm, dummy,I915_CACHE_NONE , pte_flags); + ggtt->vm.raw_insert_entries (&ggtt->vm, dummy, +i915_gem_get_pat_index (ggtt->vm.i915, +I915_CACHE_NONE ), + pte_flags);else - ggtt->vm.insert_entries (&ggtt->vm, dummy,I915_CACHE_NONE , pte_flags); + ggtt->vm.insert_entries (&ggtt->vm, dummy, +i915_gem_get_pat_index (ggtt->vm.i915, +I915_CACHE_NONE ), + pte_flags); }static void uc_fw_unbind_ggtt (struct intel_uc_fw *uc_fw)drivers/gpu/drm/i915/i915_debugfs.c +45
-10@@ -139,21 +139,56 @@ static const char *stringify_vma_type(const struct i915_vma *vma)return "ppgtt" ; } -static const char *i915_cache_level_str (struct drm_i915_private *i915,int type) -{ -switch (type) { -case I915_CACHE_NONE :return " uncached" ; -case I915_CACHE_LLC :return HAS_LLC (i915) ?" LLC" :" snooped" ; -case I915_CACHE_L3_LLC :return " L3+LLC" ; -case I915_CACHE_WT :return " WT" ; -default :return "" ; +static const char *i915_cache_level_str (struct drm_i915_gem_object *obj) +{ +struct drm_i915_private *i915 =obj_to_i915 (obj); + +if (IS_METEORLAKE (i915)) { +switch (obj->pat_index) { +case 0 :return " WB" ; +case 1 :return " WT" ; +case 2 :return " UC" ; +case 3 :return " WB (1-Way Coh)" ; +case 4 :return " WB (2-Way Coh)" ; +default :return " not defined" ; + } + }else if (IS_PONTEVECCHIO (i915)) { +switch (obj->pat_index) { +case 0 :return " UC" ; +case 1 :return " WC" ; +case 2 :return " WT" ; +case 3 :return " WB" ; +case 4 :return " WT (CLOS1)" ; +case 5 :return " WB (CLOS1)" ; +case 6 :return " WT (CLOS2)" ; +case 7 :return " WT (CLOS2)" ; +default :return " not defined" ; + } + }else if (GRAPHICS_VER (i915) >=12 ) { +switch (obj->pat_index) { +case 0 :return " WB" ; +case 1 :return " WC" ; +case 2 :return " WT" ; +case 3 :return " UC" ; +default :return " not defined" ; + } + }else { +if (i915_gem_object_has_cache_level (obj,I915_CACHE_NONE )) +return " uncached" ; +else if (i915_gem_object_has_cache_level (obj,I915_CACHE_LLC )) +return HAS_LLC (i915) ?" LLC" :" snooped" ; +else if (i915_gem_object_has_cache_level (obj,I915_CACHE_L3_LLC )) +return " L3+LLC" ; +else if (i915_gem_object_has_cache_level (obj,I915_CACHE_WT )) +return " WT" ; +else +return " not defined" ; } }void i915_debugfs_describe_obj (struct seq_file *m,struct drm_i915_gem_object *obj) { -struct drm_i915_private *dev_priv =to_i915 (obj->base.dev);struct i915_vma *vma;int pin_count =0 ;@@ -165,7 +200,7 @@ i915_debugfs_describe_obj(struct seq_file *m, struct drm_i915_gem_object *obj) obj->base.size /1024 , obj->read_domains, obj->write_domain, -i915_cache_level_str (dev_priv, obj->cache_level), +i915_cache_level_str (obj), obj->mm.dirty ?" dirty" :"" , obj->mm.madv == I915_MADV_DONTNEED ? " purgeable" : "");if (obj->base.name)drivers/gpu/drm/i915/i915_gem.c +12
-4@@ -420,8 +420,12 @@ i915_gem_gtt_pread(struct drm_i915_gem_object *obj, page_length = remain < page_length ? remain : page_length;if (drm_mm_node_allocated (&node)) { ggtt->vm.insert_page (&ggtt->vm, -i915_gem_object_get_dma_address (obj, offset >>PAGE_SHIFT ), - node.start,I915_CACHE_NONE ,0 ); +i915_gem_object_get_dma_address (obj, + offset >>PAGE_SHIFT ), + node.start, +i915_gem_get_pat_index (i915, +I915_CACHE_NONE ), +0 ); }else { page_base += offset &PAGE_MASK ; }@@ -598,8 +602,12 @@ i915_gem_gtt_pwrite_fast(struct drm_i915_gem_object *obj,/* flush the write before we modify the GGTT */ intel_gt_flush_ggtt_writes (ggtt->vm.gt); ggtt->vm.insert_page (&ggtt->vm, -i915_gem_object_get_dma_address (obj, offset >>PAGE_SHIFT ), - node.start,I915_CACHE_NONE ,0 ); +i915_gem_object_get_dma_address (obj, + offset >>PAGE_SHIFT ), + node.start, +i915_gem_get_pat_index (i915, +I915_CACHE_NONE ), +0 );wmb ();/* flush modifications to the GGTT (insert_page) */ }else { page_base += offset &PAGE_MASK ;drivers/gpu/drm/i915/i915_gpu_error.c +6
-2@@ -1117,10 +1117,14 @@ i915_vma_coredump_create(const struct intel_gt *gt,mutex_lock (&ggtt->error_mutex);if (ggtt->vm.raw_insert_page) ggtt->vm.raw_insert_page (&ggtt->vm, dma, slot, -I915_CACHE_NONE ,0 ); +i915_gem_get_pat_index (gt->i915, +I915_CACHE_NONE ), +0 );else ggtt->vm.insert_page (&ggtt->vm, dma, slot, -I915_CACHE_NONE ,0 ); +i915_gem_get_pat_index (gt->i915, +I915_CACHE_NONE ), +0 );mb (); s =io_mapping_map_wc (&ggtt->iomap, slot,PAGE_SIZE );drivers/gpu/drm/i915/i915_vma.c +8
-8@@ -315,7 +315,7 @@ struct i915_vma_work {struct i915_vma_resource *vma_res;struct drm_i915_gem_object *obj;struct i915_sw_dma_fence_cb cb; -enum i915_cache_level cache_level; +unsigned int pat_index;unsigned int flags; };@@ -334,7 +334,7 @@ static void __vma_bind(struct dma_fence_work *work)return ; vma_res->ops->bind_vma (vma_res->vm, &vw->stash, - vma_res, vw->cache_level, vw->flags); + vma_res, vw->pat_index, vw->flags); }static void __vma_release (struct dma_fence_work *work)@@ -426,7 +426,7 @@ i915_vma_resource_init_from_vma(struct i915_vma_resource *vma_res,/** * i915_vma_bind - Sets up PTEs for an VMA in it's corresponding address space. * @vma: VMA to map -* @cache_level: mapping cache level +* @pat_index: PAT index to set in PTE * @flags: flags like global or local mapping * @work: preallocated worker for allocating and binding the PTE * @vma_res: pointer to a preallocated vma resource. The resource is either @@ -437,7 +437,7 @@ i915_vma_resource_init_from_vma(struct i915_vma_resource *vma_res,* Note that DMA addresses are also the only part of the SG table we care about. */ int i915_vma_bind (struct i915_vma *vma, -enum i915_cache_level cache_level, +unsigned int pat_index,u32 flags,struct i915_vma_work *work,struct i915_vma_resource *vma_res)@@ -507,7 +507,7 @@ int i915_vma_bind(struct i915_vma *vma,struct dma_fence *prev; work->vma_res =i915_vma_resource_get (vma->resource); - work->cache_level = cache_level; + work->pat_index = pat_index; work->flags = bind_flags;/* @@ -537,7 +537,7 @@ int i915_vma_bind(struct i915_vma *vma,return ret; } - vma->ops->bind_vma (vma->vm,NULL , vma->resource, cache_level, + vma->ops->bind_vma (vma->vm,NULL , vma->resource, pat_index, bind_flags); }@@ -814,7 +814,7 @@ i915_vma_insert(struct i915_vma *vma, struct i915_gem_ww_ctx *ww, color =0 ;if (i915_vm_has_cache_coloring (vma->vm)) - color = vma->obj->cache_level; + color = vma->obj->pat_index;if (flags &PIN_OFFSET_FIXED ) {u64 offset = flags &PIN_OFFSET_MASK ;@@ -1518,7 +1518,7 @@ int i915_vma_pin_ww(struct i915_vma *vma, struct i915_gem_ww_ctx *ww,GEM_BUG_ON (!vma->pages); err =i915_vma_bind (vma, - vma->obj->cache_level, + vma->obj->pat_index, flags, work, vma_res); vma_res =NULL ;if (err)drivers/gpu/drm/i915/i915_vma.h +1
-1@@ -250,7 +250,7 @@ i915_vma_compare(struct i915_vma *vma,struct i915_vma_work *i915_vma_work (void );int i915_vma_bind (struct i915_vma *vma, -enum i915_cache_level cache_level, +unsigned int pat_index,u32 flags,struct i915_vma_work *work,struct i915_vma_resource *vma_res);drivers/gpu/drm/i915/i915_vma_types.h +0
-2@@ -32,8 +32,6 @@#include "gem/i915_gem_object_types.h" -enum i915_cache_level ; -/** * DOC: Global GTT views * drivers/gpu/drm/i915/selftests/i915_gem.c +4
-1@@ -57,7 +57,10 @@ static void trash_stolen(struct drm_i915_private *i915)u32 __iomem *s;int x; - ggtt->vm.insert_page (&ggtt->vm, dma, slot,I915_CACHE_NONE ,0 ); + ggtt->vm.insert_page (&ggtt->vm, dma, slot, +i915_gem_get_pat_index (i915, +I915_CACHE_NONE ), +0 ); s =io_mapping_map_atomic_wc (&ggtt->iomap, slot);for (x =0 ; x <PAGE_SIZE /sizeof (u32); x++) {drivers/gpu/drm/i915/selftests/i915_gem_evict.c +2
-2@@ -245,7 +245,7 @@ static int igt_evict_for_cache_color(void *arg)struct drm_mm_node target = { .start =I915_GTT_PAGE_SIZE *2 , .size =I915_GTT_PAGE_SIZE , - .color =I915_CACHE_LLC , + .color =i915_gem_get_pat_index (gt->i915,I915_CACHE_LLC ), };struct drm_i915_gem_object *obj;struct i915_vma *vma;@@ -308,7 +308,7 @@ static int igt_evict_for_cache_color(void *arg)/* Attempt to remove the first *pinned* vma, by removing the (empty) * neighbour -- this should fail. */ - target.color =I915_CACHE_L3_LLC ; + target.color =i915_gem_get_pat_index (gt->i915,I915_CACHE_L3_LLC );mutex_lock (&ggtt->vm.mutex); err =i915_gem_evict_for_node (&ggtt->vm,NULL , &target,0 );drivers/gpu/drm/i915/selftests/i915_gem_gtt.c +10
-5@@ -135,7 +135,7 @@ fake_dma_object(struct drm_i915_private *i915, u64 size) obj->write_domain =I915_GEM_DOMAIN_CPU ; obj->read_domains =I915_GEM_DOMAIN_CPU ; - obj->cache_level =I915_CACHE_NONE ; + obj->pat_index =i915_gem_get_pat_index (i915,I915_CACHE_NONE );/* Preallocate the "backing storage" */ if (i915_gem_object_pin_pages_unlocked (obj))@@ -359,7 +359,9 @@ static int lowlevel_hole(struct i915_address_space *vm,with_intel_runtime_pm (vm->gt->uncore->rpm, wakeref) vm->insert_entries(vm, mock_vma_res, -I915_CACHE_NONE ,0 ); +i915_gem_get_pat_index (vm->i915, +I915_CACHE_NONE ), +0 ); } count = n;@@ -1377,7 +1379,10 @@ static int igt_ggtt_page(void *arg) ggtt->vm.insert_page (&ggtt->vm,i915_gem_object_get_dma_address (obj,0 ), - offset,I915_CACHE_NONE ,0 ); + offset, +i915_gem_get_pat_index (i915, +I915_CACHE_NONE ), +0 ); } order =i915_random_order (count, &prng);@@ -1510,7 +1515,7 @@ static int reserve_gtt_with_resource(struct i915_vma *vma, u64 offset)mutex_lock (&vm->mutex); err =i915_gem_gtt_reserve (vm,NULL , &vma->node, obj->base.size, offset, - obj->cache_level, + obj->pat_index,0 );if (!err) {i915_vma_resource_init_from_vma (vma_res, vma);@@ -1690,7 +1695,7 @@ static int insert_gtt_with_resource(struct i915_vma *vma)mutex_lock (&vm->mutex); err =i915_gem_gtt_insert (vm,NULL , &vma->node, obj->base.size,0 , - obj->cache_level,0 , vm->total,0 ); + obj->pat_index,0 , vm->total,0 );if (!err) {i915_vma_resource_init_from_vma (vma_res, vma); vma->resource = vma_res;drivers/gpu/drm/i915/selftests/intel_memory_region.c +3
-1@@ -1070,7 +1070,9 @@ static int igt_lmem_write_cpu(void *arg)/* Put the pages into a known state -- from the gpu for added fun */ intel_engine_pm_get (engine); err =intel_context_migrate_clear (engine->gt->migrate.context,NULL , - obj->mm.pages->sgl,I915_CACHE_NONE , + obj->mm.pages->sgl, +i915_gem_get_pat_index (i915, +I915_CACHE_NONE ), true,0xdeadbeaf , &rq);if (rq) { dma_resv_add_fence(obj->base.resv, &rq->fence,drivers/gpu/drm/i915/selftests/mock_gtt.c +4
-4@@ -27,21 +27,21 @@static void mock_insert_page (struct i915_address_space *vm,dma_addr_t addr,u64 offset, -enum i915_cache_level level, +unsigned int pat_index,u32 flags) { }static void mock_insert_entries (struct i915_address_space *vm,struct i915_vma_resource *vma_res, -enum i915_cache_level level,u32 flags) +unsigned int pat_index,u32 flags) { }static void mock_bind_ppgtt (struct i915_address_space *vm,struct i915_vm_pt_stash *stash,struct i915_vma_resource *vma_res, -enum i915_cache_level cache_level, +unsigned int pat_index,u32 flags) {GEM_BUG_ON (flags &I915_VMA_GLOBAL_BIND );@@ -94,7 +94,7 @@ struct i915_ppgtt *mock_ppgtt(struct drm_i915_private *i915, const char *name)static void mock_bind_ggtt (struct i915_address_space *vm,struct i915_vm_pt_stash *stash,struct i915_vma_resource *vma_res, -enum i915_cache_level cache_level, +unsigned int pat_index,u32 flags) { }
Re: [Intel-gfx] [PATCH 7/8] drm/i915: use pat_index instead of cache_level
Quoted 9 lines · Andrzej Hajda
Quoted 8 lines · On 20.04.2023 01:00, [email protected] wrote:
From: Fei Yang <[email protected]>
Currently the KMD is using enum
i915_cache_levelto set caching policy for buffer objects. This is flaky because the PAT index which really controls the caching behavior in PTE has far more levels than what's defined in the enum. In addition, the PAT index is platform dependent, having to translate betweeni915_cache_leveland PAT index is not reliable, and makes the code more complicated.
How it is flaky and not reliable - yet the series proposed to leave it in place and even claims using cache levels simplifies the code (lower in the commit message). Maybe just the commit message needs work.
Quoted 28 lines
Quoted 11 lines
From UMD's perspective there is also a necessity to set caching policy for performance fine tuning. It's much easier for the UMD to directly use PAT index because the behavior of each PAT index is clearly defined in Bspec. Having the abstracted
i915_cache_levelsitting in between would only cause more ambiguity.For these reasons this patch replaces
i915_cache_levelwith PAT index. Also note, thecache_levelis not completely removed yet, because the KMD still has the need of creating buffer objects with simple cache settings such as cached, uncached, or writethrough. For such simple cases, usingcache_levelwould help simplify the code.It seems quite fundamental change to me. Does this "not completely removed yet" means that in some future we will not have support for generic cache levels at all? Seems strange to me. Even looking at the number of users of
i915_gem_get_pat_indexbelow it seem very unlikely.And if the support for generic level will stay, maybe it would be better to make usage of it more convienient. All conversion of
f(...,cache_level, ...)
to
f(...,i915_gem_get_pat_index(i915, cache_level), ...)
looks quite ugly to me.Maybe extending cache level to support pat index somehow, for example:
enumi915_cache_level{
I915_CACHE_NONE = 0,
I915_CACHE_...,
...
I915_CACHE_1ST_PAT_INDEX =0x100,
}
Quoted 5 lines
so
real_pat_index=cache_level-I915_CACHE_1ST_PAT_INDEXand in case of generic level there will be platform dependend conversion to
real_pat_index?I do not know the whole picture so maybe this is all wrong for some reason, just asking :)
It looks a bit unsightly to me too so yes please, brain storm on whether it can be made more elegant and less intrusive would be appreciated.
Quoted 202 lines
Quoted 202 lines
Cc Chris Wilson <[email protected]>Cc Matt Roper <[email protected]>Signed-off-by Fei Yang <[email protected]>Reviewed-by Andi Shyti <[email protected]>---
drivers/gpu/drm/i915/display/intel_dpt.c | 12 +-- drivers/gpu/drm/i915/gem/i915_gem_domain.c | 27 ++---- .../gpu/drm/i915/gem/i915_gem_execbuffer.c | 10 ++- drivers/gpu/drm/i915/gem/i915_gem_mman.c | 3 +- drivers/gpu/drm/i915/gem/i915_gem_object.c | 52 +++++++++++- drivers/gpu/drm/i915/gem/i915_gem_object.h | 4 + .../gpu/drm/i915/gem/i915_gem_object_types.h | 25 +++++- drivers/gpu/drm/i915/gem/i915_gem_stolen.c | 4 +- drivers/gpu/drm/i915/gem/i915_gem_ttm_move.c | 16 ++-- .../gpu/drm/i915/gem/selftests/huge_pages.c | 2 +-.../drm/i915/gem/selftests/i915_gem_migrate.c| 2 +- .../drm/i915/gem/selftests/i915_gem_mman.c | 2 +-drivers/gpu/drm/i915/gt/gen6_ppgtt.c | 10 ++- drivers/gpu/drm/i915/gt/gen8_ppgtt.c | 71 ++++++++-------- drivers/gpu/drm/i915/gt/gen8_ppgtt.h | 3 +- drivers/gpu/drm/i915/gt/intel_ggtt.c | 82 +++++++++----------drivers/gpu/drm/i915/gt/intel_gtt.h | 20 ++--- drivers/gpu/drm/i915/gt/intel_migrate.c | 47 ++++++----- drivers/gpu/drm/i915/gt/intel_migrate.h | 13 ++- drivers/gpu/drm/i915/gt/intel_ppgtt.c | 6 +- drivers/gpu/drm/i915/gt/selftest_migrate.c | 47 ++++++----- drivers/gpu/drm/i915/gt/selftest_reset.c | 8 +- drivers/gpu/drm/i915/gt/selftest_timeline.c | 2 +- drivers/gpu/drm/i915/gt/selftest_tlb.c | 4 +- drivers/gpu/drm/i915/gt/uc/intel_uc_fw.c | 10 ++- drivers/gpu/drm/i915/i915_debugfs.c | 55 ++++++++++--- drivers/gpu/drm/i915/i915_gem.c | 16 +++- drivers/gpu/drm/i915/i915_gpu_error.c | 8 +- drivers/gpu/drm/i915/i915_vma.c | 16 ++-- drivers/gpu/drm/i915/i915_vma.h | 2 +- drivers/gpu/drm/i915/i915_vma_types.h | 2 - drivers/gpu/drm/i915/selftests/i915_gem.c | 5 +- .../gpu/drm/i915/selftests/i915_gem_evict.c | 4 +-
drivers/gpu/drm/i915/selftests/i915_gem_gtt.c| 15 ++-- .../drm/i915/selftests/intel_memory_region.c | 4 +- drivers/gpu/drm/i915/selftests/mock_gtt.c | 8 +-36 files changed, 378 insertions(+), 239 deletions(-)drivers/gpu/drm/i915/display/intel_dpt.c +5
-5@@ -43,24 +43,24 @@ static void gen8_set_pte(void __iomem *addr, gen8_pte_t pte) static void dpt_insert_page(struct i915_address_space *vm, dma_addr_t addr, u64 offset, - enum i915_cache_level level, + unsigned int pat_index, u32 flags) { struct i915_dpt *dpt = i915_vm_to_dpt(vm); gen8_pte_t __iomem *base = dpt->iomem; gen8_set_pte(base + offset / I915_GTT_PAGE_SIZE, - vm->pte_encode(addr, level, flags)); + vm->pte_encode(addr, pat_index, flags)); } static void dpt_insert_entries(struct i915_address_space *vm, struct i915_vma_resource *vma_res, - enum i915_cache_level level, + unsigned int pat_index, u32 flags) { struct i915_dpt *dpt = i915_vm_to_dpt(vm); gen8_pte_t __iomem *base = dpt->iomem; - const gen8_pte_t pte_encode = vm->pte_encode(0, level, flags); + const gen8_pte_t pte_encode = vm->pte_encode(0, pat_index, flags); struct sgt_iter sgt_iter; dma_addr_t addr; int i;@@ -83,7 +83,7 @@ static void dpt_clear_range(structi915_address_space *vm, static void dpt_bind_vma(struct i915_address_space *vm, struct i915_vm_pt_stash *stash, struct i915_vma_resource *vma_res, - enum i915_cache_level cache_level, + unsigned int pat_index, u32 flags) {u32
pte_flags;drivers/gpu/drm/i915/display/intel_dpt.c +1
-1@@ -98,7 +98,7 @@ static void dpt_bind_vma(struct i915_address_space *vm, if (vma_res->bi.lmem) pte_flags |= PTE_LM; - vm->insert_entries(vm, vma_res, cache_level, pte_flags); + vm->insert_entries(vm, vma_res, pat_index, pte_flags); vma_res->page_sizes_gtt = I915_GTT_PAGE_SIZE;drivers/gpu/drm/i915/gem/i915_gem_domain.c +9
-18@@ -27,8 +27,8 @@ static bool gpu_write_needs_clflush(struct drm_i915_gem_object *obj) if (IS_DGFX(i915)) return false; - return !(obj->cache_level == I915_CACHE_NONE || - obj->cache_level == I915_CACHE_WT); + return !(i915_gem_object_has_cache_level(obj, I915_CACHE_NONE) || + i915_gem_object_has_cache_level(obj, I915_CACHE_WT)); } bool i915_gem_cpu_write_needs_clflush(struct drm_i915_gem_object *obj)@@ -267,7 +267,7 @@ int i915_gem_object_set_cache_level(structdrm_i915_gem_object *obj, { int ret; - if (obj->cache_level == cache_level) + if (i915_gem_object_has_cache_level(obj, cache_level)) return 0; ret = i915_gem_object_wait(obj,@@ -278,10 +278,8 @@ int i915_gem_object_set_cache_level(structdrm_i915_gem_object *obj, return ret; /* Always invalidate stale cachelines */ - if (obj->cache_level != cache_level) { - i915_gem_object_set_cache_coherency(obj, cache_level); - obj->cache_dirty = true; - } + i915_gem_object_set_cache_coherency(obj, cache_level); + obj->cache_dirty = true; /* The cache-level will be applied when each vma is rebound. */ return i915_gem_object_unbind(obj,@@ -306,20 +304,13 @@ int i915_gem_get_caching_ioctl(struct drm_device *dev, void *data, goto out; } - switch (obj->cache_level) { - case I915_CACHE_LLC: - case I915_CACHE_L3_LLC: + if (i915_gem_object_has_cache_level(obj, I915_CACHE_LLC) || + i915_gem_object_has_cache_level(obj, I915_CACHE_L3_LLC)) args->caching = I915_CACHING_CACHED; - break; - - case I915_CACHE_WT: + else if (i915_gem_object_has_cache_level(obj, I915_CACHE_WT)) args->caching = I915_CACHING_DISPLAY; - break; - - default: + else args->caching = I915_CACHING_NONE; - break; - }out:
rcu_read_unlock();
return err;drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c +0
-0@@ -642,7 +642,7 @@ static inline int use_cpu_reloc(const struct reloc_cache *cache,return (
cache->has_llc||
obj->cache_dirty||
-obj->cache_level!=I915_CACHE_NONE);
+ !i915_gem_object_has_cache_level(obj,I915_CACHE_NONE)); }
static int eb_reserve_vma(structi915_execbuffer*eb,drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c +5
-3@@ -1323,8 +1323,10 @@ static void *reloc_iomap(struct i915_vma *batch, offset = cache->node.start; if (drm_mm_node_allocated(&cache->node)) { ggtt->vm.insert_page(&ggtt->vm, - i915_gem_object_get_dma_address(obj, page), - offset, I915_CACHE_NONE, 0); + i915_gem_object_get_dma_address(obj, page), + offset, + i915_gem_get_pat_index(ggtt->vm.i915, I915_CACHE_NONE), + 0); } else { offset += page << PAGE_SHIFT; }@@ -1464,7 +1466,7 @@ eb_relocate_entry(struct i915_execbuffer *eb, reloc_cache_unmap(&eb->reloc_cache); mutex_lock(&vma->vm->mutex); err = i915_vma_bind(target->vma, - target->vma->obj->cache_level, + target->vma->obj->pat_index, PIN_GLOBAL, NULL, NULL); mutex_unlock(&vma->vm->mutex); reloc_cache_remap(&eb->reloc_cache, ev->vma->obj);drivers/gpu/drm/i915/gem/i915_gem_mman.c +2
-1@@ -383,7 +383,8 @@ static vm_fault_t vm_fault_gtt(struct vm_fault *vmf) } /* Access to snoopable pages through the GTT is incoherent. */ - if (obj->cache_level != I915_CACHE_NONE && !HAS_LLC(i915)) { + if (!(i915_gem_object_has_cache_level(obj, I915_CACHE_NONE) || + HAS_LLC(i915))) { ret = -EFAULT; goto err_unpin; }drivers/gpu/drm/i915/gem/i915_gem_object.c +0
-0@@ -54,6 +54,25 @@ unsigned int i915_gem_get_pat_index(struct
Quoted 18 lines
drivers/gpu/drm/i915/gem/i915_gem_object.c unsigned int i915_gem_get_pat_index(struct drm_i915_private *i915, · line 58
Quoted 6 lines
drivers/gpu/drm/i915/gem/i915_gem_object.c +2
-0drm_i915_private *i915, return INTEL_INFO(i915)->cachelevel_to_pat[level]; } +bool i915_gem_object_has_cache_level (const struct drm_i915_gem_object *obj, + enum i915_cache_level lvl)The name suggest object can have more cache levels, maybe only my impression, up to you.
Quoted 9 lines
drivers/gpu/drm/i915/gem/i915_gem_object.c +9
-0+{ + /* + * cache_level == I915_CACHE_INVAL indicates the UMD's have set the + * caching policy through pat_index, in which case the KMD should + * leave the coherency to be managed by user space, simply return + * true here. + */ + if (obj->cache_level == I915_CACHE_INVAL) + return true;
It's a "bit" counter intuitive that answer "has cache level" is yes when cache level is set to invalid!
I worry we don't create an impenetrable code base so I hope this can be improved.
Quoted 94 lines
Quoted 94 lines
drivers/gpu/drm/i915/gem/i915_gem_object.c +3
-0+ + /* +* Otherwise the pat_index should have been converted from
cache_level
+ * so that the following comparison is valid.
+ */
+ returnobj->pat_index==i915_gem_get_pat_index(obj_to_i915(obj), lvl);drivers/gpu/drm/i915/gem/i915_gem_object.c +2
-0+} +struct
drm_i915_gem_object*i915_gem_object_alloc(void){
structdrm_i915_gem_object*obj;drivers/gpu/drm/i915/gem/i915_gem_object.c +1
-1@@ -133,7 +152,7 @@ void i915_gem_object_set_cache_coherency(struct drm_i915_gem_object *obj, { struct drm_i915_private *i915 = to_i915(obj->base.dev); - obj->cache_level = cache_level; + obj->pat_index = i915_gem_get_pat_index(i915, cache_level); if (cache_level != I915_CACHE_NONE)
obj->cache_coherent= (I915_BO_CACHE_COHERENT_FOR_READ|drivers/gpu/drm/i915/gem/i915_gem_object.c +2
-0@@ -148,6 +167,37 @@ void i915_gem_object_set_cache_coherency(structdrm_i915_gem_object *obj, !IS_DGFX(i915); } +/** +* i915_gem_object_set_pat_index - set PAT index to be used in PTE encode
+ * @obj: #drm_i915_gem_object
+ * @pat_index: PAT index
+ *
+ * This is a clone ofi915_gem_object_set_cache_coherencytaking pat index
+ * instead ofcache_levelas its second argument.
+ */drivers/gpu/drm/i915/gem/i915_gem_object.c +23
-0+void i915_gem_object_set_pat_index (struct drm_i915_gem_object *obj, + unsigned int pat_index) +{ + struct drm_i915_private *i915 = to_i915(obj->base.dev); + + if (obj->pat_index == pat_index) + return; + + obj->pat_index = pat_index; + + if (pat_index != i915_gem_get_pat_index(i915, I915_CACHE_NONE)) + obj->cache_coherent = (I915_BO_CACHE_COHERENT_FOR_READ | + I915_BO_CACHE_COHERENT_FOR_WRITE); + else if (HAS_LLC(i915)) + obj->cache_coherent = I915_BO_CACHE_COHERENT_FOR_READ; + else + obj->cache_coherent = 0; + + obj->cache_dirty = + !(obj->cache_coherent & I915_BO_CACHE_COHERENT_FOR_WRITE) && + !IS_DGFX(i915); +} +bool
i915_gem_object_can_bypass_llc(struct drm_i915_gem_object *obj){
structdrm_i915_private*i915 =to_i915(obj->base.dev);drivers/gpu/drm/i915/gem/i915_gem_object.h +4
-0@@ -34,6 +34,8 @@ static inline bool i915_gem_object_size_2big(u64 size) unsigned int i915_gem_get_pat_index(struct drm_i915_private *i915, enum i915_cache_level level); +bool i915_gem_object_has_cache_level (const struct drm_i915_gem_object *obj, + enum i915_cache_level lvl); void i915_gem_init__objects(struct drm_i915_private *i915); void i915_objects_module_exit(void);@@ -764,6 +766,8 @@ bool i915_gem_object_has_unknown_state(struct drm_i915_gem_object *obj); void i915_gem_object_set_cache_coherency(struct drm_i915_gem_object *obj, unsigned int cache_level); +void i915_gem_object_set_pat_index (struct drm_i915_gem_object *obj, + unsigned int pat_index); bool i915_gem_object_can_bypass_llc(struct drm_i915_gem_object *obj); void i915_gem_object_flush_if_display(struct drm_i915_gem_object *obj); void i915_gem_object_flush_if_display_locked(struct drm_i915_gem_object *obj);drivers/gpu/drm/i915/gem/i915_gem_object_types.h +1
-0@@ -195,6 +195,7 @@ enum i915_cache_level {*/ I915_CACHE_WT, I915_MAX_CACHE_LEVEL, + I915_CACHE_INVAL = I915_MAX_CACHE_LEVEL, }; enum i915_map_type {@@ -358,10 +359,28 @@ struct drm_i915_gem_object {
Quoted 8 lines
Quoted 8 lines
drivers/gpu/drm/i915/gem/i915_gem_object_types.h +3
-1#define I915_BO_FLAG_STRUCT_PAGE BIT(0) /* Object backed by struct pages */ #define I915_BO_FLAG_IOMEM BIT(1) /* Object backed by IO memory */ /** -* @cache_level: The desired GTT caching level. +* @pat_index: The desired PAT index. +* +* See hardware specification for valid PAT indices for each platform.
Side note for the last patch in the series - the UAPI blurb next to u32 index needs to at least point to some public PRM which lists the PATs and their configuration I would think. Otherwise it's not fully transparent how to use the feature.
Quoted 1 line
Quoted 1 line
+ * This field used to contain a value of enum
i915_cache_level.
What does this mean? Nothing is changed to unsigned here but just new field added.
It's
Quoted 4 lines
Quoted 4 lines
+ * changed to an unsigned int because PAT indices are being used by + * both UMD and KMD for caching policy control after GEN12. + * For backward compatibility, this field will continue to contain + * value of
i915_cache_levelfor pre-GEN12 platforms so that the PTE
Pat_index:6 is a copy of cache_level:3 pre-Gen12?
But when I look at changes like:
@@ -1302,7 +1298,9 @@ bool i915_ggtt_resume_vm(struct i915_address_space *vm)
*/
vma->resource->bound_flags = 0; vma->ops->bind_vma(vm, NULL, vma->resource,
- obj ? obj->cache_level : 0,
+ obj ? obj->pat_index :
+ i915_gem_get_pat_index(vm->i915,
+ I915_CACHE_NONE),
was_bound);
That suggests it is not a copy but that obj->pat_index is always valid and directly a PAT index.
In which case new cache_level enum to say "use pat instead" may indeed be nicer as Andrzej suggested.
Although it is not clear to me for a glance that we need both. Maybe all in driver object creation can use cache_level but immediately convert to PAT internally and just don't store cache_level? I haven't looked in detail is my disclaimer though.. I guess it may boil down to does i915 ever need to read back cache_level, other than on the top entry points like setting it or so.
Quoted 7 lines
Quoted 7 lines
+ * encode functions for these legacy platforms can stay the same. + * In the meantime platform specific tables are created to translate + *
i915_cache_levelinto pat index, for more details check the macros
+ * definedi915/i915_pci.c, e.g.PVC_CACHELEVEL.
+ */
+ unsigned int pat_index:6;
Existing bitfield takes up 7 bits. I'd check here with pahole if making pat_index a full u8 and changing the existing ones to u8 field:bits maybe ends up better overall.
Quoted 1,189 lines
Quoted 1,189 lines
+ /**
+ * @cache_level: Indicate whetherpat_indexis set by UMD *
- * See enumi915_cache_levelfor possible values, along with what - * each does.
+ * This used to hold desired GTT caching level, but is now replaced by
+ *pat_index. It's kept here for KMD to tell whether thepat_indexis
+ * set by UMD or converted from enumi915_cache_level. + * This field should be 0 by default, butI915_CACHE_INVALif the + *pat_indexis set by UMD.
*/
unsigned int cache_level:3;
/**drivers/gpu/drm/i915/gem/i915_gem_stolen.c +3
-1@@ -565,7 +565,9 @@ static void dbg_poison(struct i915_ggtt *ggtt, ggtt->vm.insert_page(&ggtt->vm, addr, ggtt->error_capture.start, - I915_CACHE_NONE, 0); + i915_gem_get_pat_index(ggtt->vm.i915, + I915_CACHE_NONE), + 0); mb(); s = io_mapping_map_wc(&ggtt->iomap,drivers/gpu/drm/i915/gem/i915_gem_ttm_move.c +9
-7@@ -214,7 +214,8 @@ static struct dma_fence *i915_ttm_accel_move(struct ttm_buffer_object *bo, intel_engine_pm_get(to_gt(i915)->migrate.context->engine); ret = intel_context_migrate_clear(to_gt(i915)->migrate.context, deps, - dst_st->sgl, dst_level, + dst_st->sgl, + i915_gem_get_pat_index(i915, dst_level), i915_ttm_gtt_binds_lmem(dst_mem), 0, &rq); } else {@@ -227,12 +228,13 @@ static struct dma_fence *i915_ttm_accel_move(struct ttm_buffer_object *bo, src_level = i915_ttm_cache_level(i915, bo->resource, src_ttm); intel_engine_pm_get(to_gt(i915)->migrate.context->engine); ret = intel_context_migrate_copy(to_gt(i915)->migrate.context, - deps, src_rsgt->table.sgl, - src_level, - i915_ttm_gtt_binds_lmem(bo->resource), - dst_st->sgl, dst_level, - i915_ttm_gtt_binds_lmem(dst_mem), - &rq); + deps, src_rsgt->table.sgl, + i915_gem_get_pat_index(i915, src_level), + i915_ttm_gtt_binds_lmem(bo->resource), + dst_st->sgl, + i915_gem_get_pat_index(i915, dst_level), + i915_ttm_gtt_binds_lmem(dst_mem), + &rq); i915_refct_sgt_put(src_rsgt); }drivers/gpu/drm/i915/gem/selftests/huge_pages.c +1
-1@@ -354,7 +354,7 @@ fake_huge_pages_object(struct drm_i915_private *i915, u64 size, bool single) obj->write_domain = I915_GEM_DOMAIN_CPU; obj->read_domains = I915_GEM_DOMAIN_CPU; - obj->cache_level = I915_CACHE_NONE; + obj->pat_index = i915_gem_get_pat_index(i915, I915_CACHE_NONE); return obj; }drivers/gpu/drm/i915/gem/selftests/i915_gem_migrate.c +1
-1@@ -219,7 +219,7 @@ static int __igt_lmem_pages_migrate(structintel_gt *gt, continue; err = intel_migrate_clear(>->migrate, &ww, deps, - obj->mm.pages->sgl, obj->cache_level, + obj->mm.pages->sgl, obj->pat_index, i915_gem_object_is_lmem(obj), 0xdeadbeaf, &rq); if (rq) {drivers/gpu/drm/i915/gem/selftests/i915_gem_mman.c +1
-1@@ -1222,7 +1222,7 @@ static int __igt_mmap_migrate(structintel_memory_region **placements, } err = intel_context_migrate_clear(to_gt(i915)->migrate.context, NULL, - obj->mm.pages->sgl, obj->cache_level, + obj->mm.pages->sgl, obj->pat_index, i915_gem_object_is_lmem(obj), expand32(POISON_INUSE), &rq); i915_gem_object_unpin_pages(obj);drivers/gpu/drm/i915/gt/gen6_ppgtt.c +2
-2@@ -109,7 +109,7 @@ static void gen6_ppgtt_clear_range(structi915_address_space *vm, static void gen6_ppgtt_insert_entries(struct i915_address_space *vm, struct i915_vma_resource *vma_res, - enum i915_cache_level cache_level, + unsigned int pat_index, u32 flags) { struct i915_ppgtt *ppgtt = i915_vm_to_ppgtt(vm);@@ -117,7 +117,7 @@ static void gen6_ppgtt_insert_entries(structi915_address_space *vm, unsigned int first_entry = vma_res->start / I915_GTT_PAGE_SIZE; unsigned int act_pt = first_entry / GEN6_PTES; unsigned int act_pte = first_entry % GEN6_PTES; - const u32 pte_encode = vm->pte_encode(0, cache_level, flags); + const u32 pte_encode = vm->pte_encode(0, pat_index, flags); struct sgt_dma iter = sgt_dma(vma_res); gen6_pte_t *vaddr;@@ -227,7 +227,9 @@ static int gen6_ppgtt_init_scratch(struct gen6_ppgtt *ppgtt)vm->scratch[0]->encode = vm->pte_encode(px_dma(vm->scratch[0]),
-I915_CACHE_NONE,PTE_READ_ONLY);
+ i915_gem_get_pat_index(vm->i915,
+I915_CACHE_NONE),
+PTE_READ_ONLY);
vm->scratch[1] =vm->alloc_pt_dma(vm, I915_GTT_PAGE_SIZE_4K); if (IS_ERR(vm->scratch[1])) {drivers/gpu/drm/i915/gt/gen6_ppgtt.c +1
-1@@ -278,7 +280,7 @@ static void gen6_ppgtt_cleanup(struct i915_address_space *vm) static void pd_vma_bind(struct i915_address_space *vm, struct i915_vm_pt_stash *stash, struct i915_vma_resource *vma_res, - enum i915_cache_level cache_level, + unsigned int pat_index, u32 unused) {struct
i915_ggtt*ggtt =i915_vm_to_ggtt(vm);drivers/gpu/drm/i915/gt/gen8_ppgtt.c +13
-20@@ -56,7 +56,7 @@ static u64 gen8_pte_encode(dma_addr_t addr, } static u64 mtl_pte_encode(dma_addr_t addr, - enum i915_cache_level level, + unsigned int pat_index, u32 flags) { gen8_pte_t pte = addr | GEN8_PAGE_PRESENT | GEN8_PAGE_RW;@@ -67,24 +67,17 @@ static u64 mtl_pte_encode(dma_addr_t addr, if (flags & PTE_LM) pte |= GEN12_PPGTT_PTE_LM | GEN12_PPGTT_PTE_NC; - switch (level) { - case I915_CACHE_NONE: - pte |= GEN12_PPGTT_PTE_PAT1; - break; - case I915_CACHE_LLC: - case I915_CACHE_L3_LLC: - pte |= GEN12_PPGTT_PTE_PAT0 | GEN12_PPGTT_PTE_PAT1; - break; - case I915_CACHE_WT: + if (pat_index & BIT(0)) pte |= GEN12_PPGTT_PTE_PAT0; - break; - default: - /* This should never happen. Added to deal with the compile - * error due to the addition of I915_MAX_CACHE_LEVEL. Will - * be removed by the pat_index patch. - */ - break; - } + + if (pat_index & BIT(1)) + pte |= GEN12_PPGTT_PTE_PAT1; + + if (pat_index & BIT(2)) + pte |= GEN12_PPGTT_PTE_PAT2; + + if (pat_index & BIT(3)) + pte |= MTL_PPGTT_PTE_PAT3; return pte; }@@ -457,11 +450,11 @@ gen8_ppgtt_insert_pte(struct i915_ppgtt *ppgtt, struct i915_page_directory *pdp, struct sgt_dma *iter, u64 idx, - enum i915_cache_level cache_level, + unsigned int pat_index, u32 flags) { struct i915_page_directory *pd; - const gen8_pte_t pte_encode = ppgtt->vm.pte_encode(0, cache_level, flags); + const gen8_pte_t pte_encode = ppgtt->vm.pte_encode(0, pat_index, flags); gen8_pte_t *vaddr;pd =
i915_pd_entry(pdp, gen8_pd_index(idx, 2));drivers/gpu/drm/i915/gt/gen8_ppgtt.c +6
-6@@ -504,10 +497,10 @@ static void xehpsdv_ppgtt_insert_huge(struct i915_address_space *vm, struct i915_vma_resource *vma_res, struct sgt_dma *iter, - enum i915_cache_level cache_level, + unsigned int pat_index, u32 flags) { - const gen8_pte_t pte_encode = vm->pte_encode(0, cache_level, flags); + const gen8_pte_t pte_encode = vm->pte_encode(0, pat_index, flags); unsigned int rem = sg_dma_len(iter->sg); u64 start = vma_res->start; u64 end = start + vma_res->vma_size;@@ -611,10 +604,10 @@ xehpsdv_ppgtt_insert_huge(structi915_address_space *vm, static void gen8_ppgtt_insert_huge(struct i915_address_space *vm, struct i915_vma_resource *vma_res, struct sgt_dma *iter, - enum i915_cache_level cache_level, + unsigned int pat_index, u32 flags) { - const gen8_pte_t pte_encode = vm->pte_encode(0, cache_level, flags); + const gen8_pte_t pte_encode = vm->pte_encode(0, pat_index, flags); unsigned int rem = sg_dma_len(iter->sg); u64 start = vma_res->start;@@ -734,7 +727,7 @@ static void gen8_ppgtt_insert_huge(structi915_address_space *vm, static void gen8_ppgtt_insert(struct i915_address_space *vm, struct i915_vma_resource *vma_res, - enum i915_cache_level cache_level, + unsigned int pat_index, u32 flags) { struct i915_ppgtt * const ppgtt = i915_vm_to_ppgtt(vm);@@ -742,9 +735,9 @@ static void gen8_ppgtt_insert(struct i915_address_space *vm, if (vma_res->bi.page_sizes.sg > I915_GTT_PAGE_SIZE) { if (HAS_64K_PAGES(vm->i915)) - xehpsdv_ppgtt_insert_huge(vm, vma_res, &iter, cache_level, flags); + xehpsdv_ppgtt_insert_huge(vm, vma_res, &iter, pat_index, flags);else
-gen8_ppgtt_insert_huge(vm, vma_res, &iter, cache_level, flags);
+gen8_ppgtt_insert_huge(vm, vma_res, &iter, pat_index, flags);
} else {
u64 idx =vma_res->start>>GEN8_PTE_SHIFT;drivers/gpu/drm/i915/gt/gen8_ppgtt.c +2
-2@@ -753,7 +746,7 @@ static void gen8_ppgtt_insert(structi915_address_space *vm, gen8_pdp_for_page_index(vm, idx); idx = gen8_ppgtt_insert_pte(ppgtt, pdp, &iter, idx, - cache_level, flags); + pat_index, flags); } while (idx); vma_res->page_sizes_gtt = I915_GTT_PAGE_SIZE;@@ -763,7 +756,7 @@ static void gen8_ppgtt_insert(structi915_address_space *vm, static void gen8_ppgtt_insert_entry(struct i915_address_space *vm, dma_addr_t addr, u64 offset, - enum i915_cache_level level, + unsigned int pat_index, u32 flags) {u64 idx = offset >>
GEN8_PTE_SHIFT;drivers/gpu/drm/i915/gt/gen8_ppgtt.c +6
-6@@ -777,14 +770,14 @@ static void gen8_ppgtt_insert_entry(structi915_address_space *vm, GEM_BUG_ON(pt->is_compact); vaddr = px_vaddr(pt); - vaddr[gen8_pd_index(idx, 0)] = vm->pte_encode(addr, level, flags); + vaddr[gen8_pd_index(idx, 0)] = vm->pte_encode(addr, pat_index, flags); drm_clflush_virt_range(&vaddr[gen8_pd_index(idx, 0)], sizeof(*vaddr)); } static void __xehpsdv_ppgtt_insert_entry_lm(struct i915_address_space *vm, dma_addr_t addr, u64 offset, - enum i915_cache_level level, + unsigned int pat_index, u32 flags) { u64 idx = offset >> GEN8_PTE_SHIFT;@@ -807,20 +800,20 @@ static void __xehpsdv_ppgtt_insert_entry_lm(struct i915_address_space *vm, } vaddr = px_vaddr(pt); - vaddr[gen8_pd_index(idx, 0) / 16] = vm->pte_encode(addr, level, flags); + vaddr[gen8_pd_index(idx, 0) / 16] = vm->pte_encode(addr, pat_index, flags); } static void xehpsdv_ppgtt_insert_entry(struct i915_address_space *vm, dma_addr_t addr, u64 offset, - enum i915_cache_level level, + unsigned int pat_index, u32 flags) { if (flags & PTE_LM) return __xehpsdv_ppgtt_insert_entry_lm(vm, addr, offset, - level, flags); + pat_index, flags); - return gen8_ppgtt_insert_entry(vm, addr, offset, level, flags); + return gen8_ppgtt_insert_entry(vm, addr, offset, pat_index, flags); } static int gen8_init_scratch(struct i915_address_space *vm)@@ -855,7 +848,9 @@ static int gen8_init_scratch(struct i915_address_space *vm)vm->scratch[0]->encode =
vm->pte_encode(px_dma(vm->scratch[0]), - I915_CACHE_NONE, pte_flags); +i915_gem_get_pat_index(vm->i915, + I915_CACHE_NONE), +pte_flags);
for (i = 1; i <=vm->top; i++) { structdrm_i915_gem_object*obj;drivers/gpu/drm/i915/gt/gen8_ppgtt.c +3
-1@@ -873,7 +868,9 @@ static int gen8_init_scratch(struct i915_address_space *vm) } fill_px(obj, vm->scratch[i - 1]->encode); - obj->encode = gen8_pde_encode(px_dma(obj), I915_CACHE_NONE); + obj->encode = gen8_pde_encode(px_dma(obj), + i915_gem_get_pat_index(vm->i915, + I915_CACHE_NONE)); vm->scratch[i] = obj; }drivers/gpu/drm/i915/gt/gen8_ppgtt.h +1
-2@@ -10,13 +10,12 @@ struct i915_address_space; struct intel_gt; -enum i915_cache_level ; struct i915_ppgtt *gen8_ppgtt_create(struct intel_gt *gt, unsigned long lmem_pt_obj_flags); u64 gen8_ggtt_pte_encode(dma_addr_t addr, - enum i915_cache_level level, + unsigned int pat_index, u32 flags); #endifdrivers/gpu/drm/i915/gt/intel_ggtt.c +7
-20@@ -221,7 +221,7 @@ static void guc_ggtt_invalidate(struct i915_ggtt *ggtt) } static u64 mtl_ggtt_pte_encode(dma_addr_t addr, - enum i915_cache_level level, + unsigned int pat_index, u32 flags) { gen8_pte_t pte = addr | GEN8_PAGE_PRESENT;@@ -231,30 +231,17 @@ static u64 mtl_ggtt_pte_encode(dma_addr_t addr, if (flags & PTE_LM) pte |= GEN12_GGTT_PTE_LM; - switch (level) { - case I915_CACHE_NONE: - pte |= MTL_GGTT_PTE_PAT1; - break; - case I915_CACHE_LLC: - case I915_CACHE_L3_LLC: - pte |= MTL_GGTT_PTE_PAT0 | MTL_GGTT_PTE_PAT1; - break; - case I915_CACHE_WT: + if (pat_index & BIT(0)) pte |= MTL_GGTT_PTE_PAT0; - break; - default: - /* This should never happen. Added to deal with the compile - * error due to the addition of I915_MAX_CACHE_LEVEL. Will - * be removed by the pat_index patch. - */ - break; - } + + if (pat_index & BIT(1)) + pte |= MTL_GGTT_PTE_PAT1; return pte; } u64 gen8_ggtt_pte_encode(dma_addr_t addr, - enum i915_cache_level level, + unsigned int pat_index, u32 flags) { gen8_pte_t pte = addr | GEN8_PAGE_PRESENT;@@ -273,25 +260,25 @@ static void gen8_set_pte(void __iomem *addr, gen8_pte_t pte) static void gen8_ggtt_insert_page(struct i915_address_space *vm, dma_addr_t addr, u64 offset, - enum i915_cache_level level, + unsigned int pat_index, u32 flags) { struct i915_ggtt *ggtt = i915_vm_to_ggtt(vm);
gen8_pte_t__iomem*pte =
(gen8_pte_t__iomem*)ggtt->gsm + offset /I915_GTT_PAGE_SIZE; -gen8_set_pte(pte, ggtt->vm.pte_encode(addr, level, flags)); +gen8_set_pte(pte, ggtt->vm.pte_encode(addr, pat_index, flags));ggtt->invalidate(ggtt);
}
static void gen8_ggtt_insert_entries(structi915_address_space*vm, structi915_vma_resource*vma_res,
- enumi915_cache_levellevel,
+ unsigned intpat_index,
u32 flags)
{
structi915_ggtt*ggtt =i915_vm_to_ggtt(vm);
- constgen8_pte_tpte_encode=ggtt->vm.pte_encode(0, level, flags); + constgen8_pte_tpte_encode=ggtt->vm.pte_encode(0, pat_index, flags);
gen8_pte_t__iomem*gte;
gen8_pte_t__iomem*end;
structsgt_iteriter;drivers/gpu/drm/i915/gt/intel_ggtt.c +1
-1@@ -348,14 +335,14 @@ static void gen8_ggtt_clear_range(structi915_address_space *vm, static void gen6_ggtt_insert_page(struct i915_address_space *vm, dma_addr_t addr, u64 offset, - enum i915_cache_level level, + unsigned int pat_index, u32 flags) { struct i915_ggtt *ggtt = i915_vm_to_ggtt(vm);
gen6_pte_t__iomem*pte =
(gen6_pte_t__iomem*)ggtt->gsm + offset /I915_GTT_PAGE_SIZE; -iowrite32(vm->pte_encode(addr, level, flags), pte);
+iowrite32(vm->pte_encode(addr, pat_index, flags), pte);ggtt->invalidate(ggtt);
}drivers/gpu/drm/i915/gt/intel_ggtt.c +1
-1@@ -368,7 +355,7 @@ static void gen6_ggtt_insert_page(structi915_address_space *vm, */ static void gen6_ggtt_insert_entries(struct i915_address_space *vm, struct i915_vma_resource *vma_res, - enum i915_cache_level level, + unsigned int pat_index, u32 flags) {struct
i915_ggtt*ggtt =i915_vm_to_ggtt(vm);drivers/gpu/drm/i915/gt/intel_ggtt.c +12
-10@@ -385,7 +372,7 @@ static void gen6_ggtt_insert_entries(structi915_address_space *vm, iowrite32(vm->scratch[0]->encode, gte++); end += (vma_res->node_size + vma_res->guard) / I915_GTT_PAGE_SIZE; for_each_sgt_daddr(addr, iter, vma_res->bi.pages) - iowrite32(vm->pte_encode(addr, level, flags), gte++); + iowrite32(vm->pte_encode(addr, pat_index, flags), gte++); GEM_BUG_ON(gte > end); /* Fill the allocated but "unused" space beyond the end of the buffer */@@ -420,14 +407,15 @@ struct insert_page { struct i915_address_space *vm; dma_addr_t addr; u64 offset; - enum i915_cache_level level; + unsigned int pat_index; }; static int bxt_vtd_ggtt_insert_page__cb(void *_arg) { struct insert_page *arg = _arg; - gen8_ggtt_insert_page(arg->vm, arg->addr, arg->offset, arg->level, 0); + gen8_ggtt_insert_page(arg->vm, arg->addr, arg->offset, + arg->pat_index, 0); bxt_vtd_ggtt_wa(arg->vm); return 0;@@ -436,10 +424,10 @@ static int bxt_vtd_ggtt_insert_page__cb(void *_arg) static void bxt_vtd_ggtt_insert_page__BKL(struct i915_address_space *vm, dma_addr_t addr, u64 offset, - enum i915_cache_level level, + unsigned int pat_index, u32 unused) { - struct insert_page arg = { vm, addr, offset, level }; + struct insert_page arg = { vm, addr, offset, pat_index }; stop_machine(bxt_vtd_ggtt_insert_page__cb, &arg, NULL); }@@ -447,7 +435,7 @@ static void bxt_vtd_ggtt_insert_page__BKL(struct i915_address_space *vm, struct insert_entries { struct i915_address_space *vm; struct i915_vma_resource *vma_res; - enum i915_cache_level level; + unsigned int pat_index; u32 flags; };@@ -455,7 +443,8 @@ static int bxt_vtd_ggtt_insert_entries__cb(void *_arg) { struct insert_entries *arg = _arg; - gen8_ggtt_insert_entries(arg->vm, arg->vma_res, arg->level, arg->flags); + gen8_ggtt_insert_entries(arg->vm, arg->vma_res, + arg->pat_index, arg->flags); bxt_vtd_ggtt_wa(arg->vm); return 0;@@ -463,10 +452,10 @@ static int bxt_vtd_ggtt_insert_entries__cb(void *_arg) static void bxt_vtd_ggtt_insert_entries__BKL(struct i915_address_space *vm, struct i915_vma_resource *vma_res, - enum i915_cache_level level, + unsigned int pat_index, u32 flags) { - struct insert_entries arg = { vm, vma_res, level, flags }; + struct insert_entries arg = { vm, vma_res, pat_index, flags }; stop_machine(bxt_vtd_ggtt_insert_entries__cb, &arg, NULL); }@@ -495,7 +484,7 @@ static void gen6_ggtt_clear_range(structi915_address_space *vm, void intel_ggtt_bind_vma(struct i915_address_space *vm, struct i915_vm_pt_stash *stash, struct i915_vma_resource *vma_res, - enum i915_cache_level cache_level, + unsigned int pat_index, u32 flags) {u32
pte_flags;drivers/gpu/drm/i915/gt/intel_ggtt.c +4
-4@@ -512,7 +501,7 @@ void intel_ggtt_bind_vma(struct i915_address_space *vm, if (vma_res->bi.lmem) pte_flags |= PTE_LM; - vm->insert_entries(vm, vma_res, cache_level, pte_flags); + vm->insert_entries(vm, vma_res, pat_index, pte_flags); vma_res->page_sizes_gtt = I915_GTT_PAGE_SIZE; }@@ -661,7 +650,7 @@ static int init_ggtt(struct i915_ggtt *ggtt) static void aliasing_gtt_bind_vma(struct i915_address_space *vm, struct i915_vm_pt_stash *stash, struct i915_vma_resource *vma_res, - enum i915_cache_level cache_level, + unsigned int pat_index, u32 flags) { u32 pte_flags;@@ -673,10 +662,10 @@ static void aliasing_gtt_bind_vma(struct i915_address_space *vm, if (flags & I915_VMA_LOCAL_BIND) ppgtt_bind_vma(&i915_vm_to_ggtt(vm)->alias->vm, - stash, vma_res, cache_level, flags); + stash, vma_res, pat_index, flags); if (flags & I915_VMA_GLOBAL_BIND) - vm->insert_entries(vm, vma_res, cache_level, pte_flags); + vm->insert_entries(vm, vma_res, pat_index, pte_flags); vma_res->bound_flags |= flags; }@@ -933,7 +922,9 @@ static int ggtt_probe_common(struct i915_ggtt *ggtt, u64 size)ggtt->vm.scratch[0]->encode =
ggtt->vm.pte_encode(px_dma(ggtt->vm.scratch[0]), - I915_CACHE_NONE, pte_flags); +i915_gem_get_pat_index(i915, + I915_CACHE_NONE), +pte_flags);
return 0;
}drivers/gpu/drm/i915/gt/intel_ggtt.c +8
-1@@ -1022,6 +1013,11 @@ static int gen8_gmch_probe(struct i915_ggtt *ggtt) return ggtt_probe_common(ggtt, size); } +/* +* For pre-gen8 platforms pat_index is the same as enum i915_cache_level, +* so these PTE encode functions are left with using cache_level. +* See translation table LEGACY_CACHELEVEL. +*/ static u64 snb_pte_encode(dma_addr_t addr, enum i915_cache_level level, u32 flags)@@ -1302,7 +1298,9 @@ bool i915_ggtt_resume_vm(structi915_address_space *vm) */ vma->resource->bound_flags = 0; vma->ops->bind_vma(vm, NULL, vma->resource, - obj ? obj->cache_level : 0, + obj ? obj->pat_index : + i915_gem_get_pat_index(vm->i915, + I915_CACHE_NONE), was_bound);if (obj) { /* only used during resume => exclusive access */
drivers/gpu/drm/i915/gt/intel_gtt.h +2
-4@@ -165,8 +165,6 @@ typedef u64 gen8_pte_t; #define MTL_2_COH_1W REG_FIELD_PREP(MTL_PAT_INDEX_COH_MODE_MASK, 2) #define MTL_0_COH_NON REG_FIELD_PREP(MTL_PAT_INDEX_COH_MODE_MASK, 0) -enum i915_cache_level ; - struct drm_i915_gem_object; struct i915_fence_reg; struct i915_vma;@@ -234,7 +232,7 @@ struct i915_vma_ops { void (*bind_vma)(struct i915_address_space *vm, struct i915_vm_pt_stash *stash, struct i915_vma_resource *vma_res, - enum i915_cache_level cache_level, + unsigned int pat_index, u32 flags); /** Unmap an object from an address space. This usually consists of @@ -306,7 +304,7 @@ struct i915_address_space { (*alloc_scratch_dma)(struct i915_address_space *vm, int sz); u64 (*pte_encode)(dma_addr_t addr, - enum i915_cache_level level, + unsigned int pat_index,u32 flags); /* Create a valid PTE */ #define PTE_READ_ONLY BIT(0)
#define PTE_LM BIT(1)drivers/gpu/drm/i915/gt/intel_gtt.h +5
-5@@ -321,20 +319,20 @@ struct i915_address_space { void (*insert_page)(struct i915_address_space *vm, dma_addr_t addr, u64 offset, - enum i915_cache_level cache_level, + unsigned int pat_index, u32 flags); void (*insert_entries)(struct i915_address_space *vm, struct i915_vma_resource *vma_res, - enum i915_cache_level cache_level, + unsigned int pat_index, u32 flags); void (*raw_insert_page)(struct i915_address_space *vm, dma_addr_t addr, u64 offset, - enum i915_cache_level cache_level, + unsigned int pat_index, u32 flags); void (*raw_insert_entries)(struct i915_address_space *vm, struct i915_vma_resource *vma_res, - enum i915_cache_level cache_level, + unsigned int pat_index, u32 flags); void (*cleanup)(struct i915_address_space *vm);@@ -581,7 +579,7 @@ void ppgtt_init(struct i915_ppgtt *ppgtt, structintel_gt *gt, void intel_ggtt_bind_vma(struct i915_address_space *vm, struct i915_vm_pt_stash *stash, struct i915_vma_resource *vma_res, - enum i915_cache_level cache_level, + unsigned int pat_index, u32 flags); void intel_ggtt_unbind_vma(struct i915_address_space *vm,struct
i915_vma_resource*vma_res);drivers/gpu/drm/i915/gt/intel_gtt.h +1
-1@@ -639,7 +637,7 @@ void __set_pd_entry(struct i915_page_directory * const pd, const unsigned short idx, struct i915_page_table *pt, - u64 (*encode)(const dma_addr_t, const enum i915_cache_level)); + u64 (*encode)(const dma_addr_t, const unsigned int pat_index)); #define set_pd_entry(pd, idx, to) \
__set_pd_entry((pd), (idx), px_pt(to), gen8_pde_encode)drivers/gpu/drm/i915/gt/intel_gtt.h +1
-1@@ -659,7 +657,7 @@ void gen6_ggtt_invalidate(struct i915_ggtt *ggtt); void ppgtt_bind_vma(struct i915_address_space *vm, struct i915_vm_pt_stash *stash, struct i915_vma_resource *vma_res, - enum i915_cache_level cache_level, + unsigned int pat_index, u32 flags); void ppgtt_unbind_vma(struct i915_address_space *vm, struct i915_vma_resource *vma_res);drivers/gpu/drm/i915/gt/intel_migrate.c +3
-1@@ -45,7 +45,9 @@ static void xehpsdv_toggle_pdes(structi915_address_space *vm, * Insert a dummy PTE into every PT that will map to LMEM to ensure * we have a correctly setup PDE structure for later use. */ - vm->insert_page(vm, 0, d->offset, I915_CACHE_NONE, PTE_LM); + vm->insert_page(vm, 0, d->offset, + i915_gem_get_pat_index(vm->i915, I915_CACHE_NONE), + PTE_LM); GEM_BUG_ON(!pt->is_compact); d->offset += SZ_2M;}
drivers/gpu/drm/i915/gt/intel_migrate.c +23
-20@@ -63,7 +65,9 @@ static void xehpsdv_insert_pte(structi915_address_space *vm, * alignment is 64K underneath for the pt, and we are careful * not to access the space in the void. */ - vm->insert_page(vm, px_dma(pt), d->offset, I915_CACHE_NONE, PTE_LM); + vm->insert_page(vm, px_dma(pt), d->offset, + i915_gem_get_pat_index(vm->i915, I915_CACHE_NONE), + PTE_LM); d->offset += SZ_64K; }@@ -73,7 +77,8 @@ static void insert_pte(struct i915_address_space *vm, { struct insert_pte_data *d = data; - vm->insert_page(vm, px_dma(pt), d->offset, I915_CACHE_NONE, + vm->insert_page(vm, px_dma(pt), d->offset, + i915_gem_get_pat_index(vm->i915, I915_CACHE_NONE), i915_gem_object_is_lmem(pt->base) ? PTE_LM : 0); d->offset += PAGE_SIZE; }@@ -356,13 +361,13 @@ static int max_pte_pkt_size(struct i915_request *rq, int pkt) static int emit_pte(struct i915_request *rq, struct sgt_dma *it, - enum i915_cache_level cache_level, + unsigned int pat_index, bool is_lmem, u64 offset, int length) { bool has_64K_pages = HAS_64K_PAGES(rq->engine->i915); - const u64 encode = rq->context->vm->pte_encode(0, cache_level, + const u64 encode = rq->context->vm->pte_encode(0, pat_index, is_lmem ? PTE_LM : 0); struct intel_ring *ring = rq->ring; int pkt, dword_length;@@ -673,17 +678,17 @@ int intel_context_migrate_copy(struct intel_context *ce, const struct i915_deps *deps, struct scatterlist *src, - enum i915_cache_level src_cache_level, + unsigned int src_pat_index, bool src_is_lmem, struct scatterlist *dst, - enum i915_cache_level dst_cache_level, + unsigned int dst_pat_index, bool dst_is_lmem, struct i915_request **out) { struct sgt_dma it_src = sg_sgt(src), it_dst = sg_sgt(dst), it_ccs; struct drm_i915_private *i915 = ce->engine->i915; u64 ccs_bytes_to_cpy = 0, bytes_to_cpy; - enum i915_cache_level ccs_cache_level; + unsigned int ccs_pat_index; u32 src_offset, dst_offset; u8 src_access, dst_access; struct i915_request *rq;@@ -707,12 +712,12 @@ intel_context_migrate_copy(struct intel_context *ce, dst_sz = scatter_list_length(dst); if (src_is_lmem) { it_ccs = it_dst; - ccs_cache_level = dst_cache_level; + ccs_pat_index = dst_pat_index; ccs_is_src = false; } else if (dst_is_lmem) { bytes_to_cpy = dst_sz; it_ccs = it_src; - ccs_cache_level = src_cache_level; + ccs_pat_index = src_pat_index; ccs_is_src = true; }@@ -773,7 +778,7 @@ intel_context_migrate_copy(struct intel_context *ce, src_sz = calculate_chunk_sz(i915, src_is_lmem, bytes_to_cpy, ccs_bytes_to_cpy); - len = emit_pte(rq, &it_src, src_cache_level, src_is_lmem, + len = emit_pte(rq, &it_src, src_pat_index, src_is_lmem, src_offset, src_sz); if (!len) { err = -EINVAL;@@ -784,7 +789,7 @@ intel_context_migrate_copy(struct intel_context *ce, goto out_rq; } - err = emit_pte(rq, &it_dst, dst_cache_level, dst_is_lmem, + err = emit_pte(rq, &it_dst, dst_pat_index, dst_is_lmem, dst_offset, len); if (err < 0) goto out_rq;@@ -811,7 +816,7 @@ intel_context_migrate_copy(struct intel_context *ce, goto out_rq; ccs_sz = GET_CCS_BYTES(i915, len); - err = emit_pte(rq, &it_ccs, ccs_cache_level, false, + err = emit_pte(rq, &it_ccs, ccs_pat_index, false, ccs_is_src ? src_offset : dst_offset, ccs_sz); if (err < 0)@@ -979,7 +984,7 @@ int intel_context_migrate_clear(struct intel_context *ce, const struct i915_deps *deps, struct scatterlist *sg, - enum i915_cache_level cache_level, + unsigned int pat_index, bool is_lmem, u32 value, struct i915_request **out)@@ -1027,7 +1032,7 @@ intel_context_migrate_clear(struct intel_context *ce, if (err) goto out_rq; - len = emit_pte(rq, &it, cache_level, is_lmem, offset, CHUNK_SZ); + len = emit_pte(rq, &it, pat_index, is_lmem, offset, CHUNK_SZ); if (len <= 0) { err = len; goto out_rq;@@ -1074,10 +1079,10 @@ int intel_migrate_copy(struct intel_migrate *m, struct i915_gem_ww_ctx *ww, const struct i915_deps *deps, struct scatterlist *src, - enum i915_cache_level src_cache_level, + unsigned int src_pat_index, bool src_is_lmem, struct scatterlist *dst, - enum i915_cache_level dst_cache_level, + unsigned int dst_pat_index, bool dst_is_lmem, struct i915_request **out) {@@ -1098,8 +1103,8 @@ int intel_migrate_copy(struct intel_migrate *m, goto out; err = intel_context_migrate_copy(ce, deps, - src, src_cache_level, src_is_lmem, - dst, dst_cache_level, dst_is_lmem, + src, src_pat_index, src_is_lmem, + dst, dst_pat_index, dst_is_lmem, out); intel_context_unpin(ce);@@ -1113,7 +1118,7 @@ intel_migrate_clear(struct intel_migrate *m, struct i915_gem_ww_ctx *ww, const struct i915_deps *deps, struct scatterlist *sg, - enum i915_cache_level cache_level, + unsigned int pat_index, bool is_lmem, u32 value, struct i915_request **out)@@ -1134,7 +1139,7 @@ intel_migrate_clear(struct intel_migrate *m, if (err) goto out; - err = intel_context_migrate_clear(ce, deps, sg, cache_level, + err = intel_context_migrate_clear(ce, deps, sg, pat_index, is_lmem, value, out); intel_context_unpin(ce);drivers/gpu/drm/i915/gt/intel_migrate.h +6
-7@@ -16,7 +16,6 @@ struct i915_request; struct i915_gem_ww_ctx; struct intel_gt; struct scatterlist; -enum i915_cache_level ; int intel_migrate_init(struct intel_migrate *m, struct intel_gt *gt);@@ -26,20 +25,20 @@ int intel_migrate_copy(struct intel_migrate *m, struct i915_gem_ww_ctx *ww, const struct i915_deps *deps, struct scatterlist *src, - enum i915_cache_level src_cache_level, + unsigned int src_pat_index, bool src_is_lmem, struct scatterlist *dst, - enum i915_cache_level dst_cache_level, + unsigned int dst_pat_index, bool dst_is_lmem, struct i915_request **out); int intel_context_migrate_copy(struct intel_context *ce, const struct i915_deps *deps, struct scatterlist *src, - enum i915_cache_level src_cache_level, + unsigned int src_pat_index, bool src_is_lmem, struct scatterlist *dst, - enum i915_cache_level dst_cache_level, + unsigned int dst_pat_index, bool dst_is_lmem, struct i915_request **out);@@ -48,7 +47,7 @@ intel_migrate_clear(struct intel_migrate *m, struct i915_gem_ww_ctx *ww, const struct i915_deps *deps, struct scatterlist *sg, - enum i915_cache_level cache_level, + unsigned int pat_index, bool is_lmem, u32 value, struct i915_request **out);@@ -56,7 +55,7 @@ int intel_context_migrate_clear(struct intel_context *ce, const struct i915_deps *deps, struct scatterlist *sg, - enum i915_cache_level cache_level, + unsigned int pat_index, bool is_lmem, u32 value, struct i915_request **out);drivers/gpu/drm/i915/gt/intel_ppgtt.c +2
-2@@ -98,7 +98,7 @@ void __set_pd_entry(struct i915_page_directory * const pd, const unsigned short idx, struct i915_page_table * const to, - u64 (*encode)(const dma_addr_t, const enum i915_cache_level)) + u64 (*encode)(const dma_addr_t, const unsigned int)) { /* Each thread pre-pins the pd, and we may have a thread per pde. */ GEM_BUG_ON(atomic_read(px_used(pd)) > NALLOC * I915_PDES);@@ -181,7 +181,7 @@ struct i915_ppgtt *i915_ppgtt_create(structintel_gt *gt, void ppgtt_bind_vma(struct i915_address_space *vm, struct i915_vm_pt_stash *stash, struct i915_vma_resource *vma_res, - enum i915_cache_level cache_level, + unsigned int pat_index, u32 flags) {u32
pte_flags;drivers/gpu/drm/i915/gt/intel_ppgtt.c +1
-1@@ -199,7 +199,7 @@ void ppgtt_bind_vma(struct i915_address_space *vm, if (vma_res->bi.lmem) pte_flags |= PTE_LM; - vm->insert_entries(vm, vma_res, cache_level, pte_flags); + vm->insert_entries(vm, vma_res, pat_index, pte_flags); wmb(); }drivers/gpu/drm/i915/gt/selftest_migrate.c +8
-8@@ -137,7 +137,7 @@ static int copy(struct intel_migrate *migrate, static int intel_context_copy_ccs(struct intel_context *ce, const struct i915_deps *deps, struct scatterlist *sg, - enum i915_cache_level cache_level, + unsigned int pat_index, bool write_to_ccs, struct i915_request **out) {@@ -185,7 +185,7 @@ static int intel_context_copy_ccs(structintel_context *ce, if (err) goto out_rq; - len = emit_pte(rq, &it, cache_level, true, offset, CHUNK_SZ); + len = emit_pte(rq, &it, pat_index, true, offset, CHUNK_SZ); if (len <= 0) { err = len; goto out_rq;@@ -223,7 +223,7 @@ intel_migrate_ccs_copy(struct intel_migrate *m, struct i915_gem_ww_ctx *ww, const struct i915_deps *deps, struct scatterlist *sg, - enum i915_cache_level cache_level, + unsigned int pat_index, bool write_to_ccs, struct i915_request **out) {@@ -243,7 +243,7 @@ intel_migrate_ccs_copy(struct intel_migrate *m, if (err) goto out; - err = intel_context_copy_ccs(ce, deps, sg, cache_level, + err = intel_context_copy_ccs(ce, deps, sg, pat_index, write_to_ccs, out); intel_context_unpin(ce);@@ -300,7 +300,7 @@ static int clear(struct intel_migrate *migrate, /* Write the obj data into ccs surface */ err = intel_migrate_ccs_copy(migrate, &ww, NULL, obj->mm.pages->sgl, - obj->cache_level, + obj->pat_index, true, &rq); if (rq && !err) { if (i915_request_wait(rq, 0, HZ) < 0) {@@ -351,7 +351,7 @@ static int clear(struct intel_migrate *migrate, err = intel_migrate_ccs_copy(migrate, &ww, NULL, obj->mm.pages->sgl, - obj->cache_level, + obj->pat_index, false, &rq); if (rq && !err) { if (i915_request_wait(rq, 0, HZ) < 0) {@@ -414,9 +414,9 @@ static int __migrate_copy(struct intel_migrate *migrate, struct i915_request **out) { return intel_migrate_copy(migrate, ww, NULL, - src->mm.pages->sgl, src->cache_level, + src->mm.pages->sgl, src->pat_index, i915_gem_object_is_lmem(src), - dst->mm.pages->sgl, dst->cache_level, + dst->mm.pages->sgl, dst->pat_index, i915_gem_object_is_lmem(dst), out);}
drivers/gpu/drm/i915/gt/selftest_migrate.c +2
-2@@ -428,9 +428,9 @@ static int __global_copy(struct intel_migrate *migrate, struct i915_request **out) { return intel_context_migrate_copy(migrate->context, NULL, - src->mm.pages->sgl, src->cache_level, + src->mm.pages->sgl, src->pat_index, i915_gem_object_is_lmem(src), - dst->mm.pages->sgl, dst->cache_level, + dst->mm.pages->sgl, dst->pat_index, i915_gem_object_is_lmem(dst), out);}
drivers/gpu/drm/i915/gt/selftest_migrate.c +1
-1@@ -455,7 +455,7 @@ static int __migrate_clear(struct intel_migrate *migrate, { return intel_migrate_clear(migrate, ww, NULL, obj->mm.pages->sgl, - obj->cache_level, + obj->pat_index, i915_gem_object_is_lmem(obj), value, out);}
drivers/gpu/drm/i915/gt/selftest_migrate.c +1
-1@@ -468,7 +468,7 @@ static int __global_clear(struct intel_migrate *migrate, { return intel_context_migrate_clear(migrate->context, NULL, obj->mm.pages->sgl, - obj->cache_level, + obj->pat_index, i915_gem_object_is_lmem(obj), value, out);}
drivers/gpu/drm/i915/gt/selftest_migrate.c +13
-10@@ -648,7 +648,7 @@ static int live_emit_pte_full_ring(void *arg)*/ pr_info("%s emite_pte ring space=%u\n", __func__, rq->ring->space); it = sg_sgt(obj->mm.pages->sgl); - len = emit_pte(rq, &it, obj->cache_level, false, 0, CHUNK_SZ); + len = emit_pte(rq, &it, obj->pat_index, false, 0, CHUNK_SZ); if (!len) { err = -EINVAL; goto out_rq;@@ -844,7 +844,7 @@ static int wrap_ktime_compare(const void *A, const void *B) static int __perf_clear_blt(struct intel_context *ce, struct scatterlist *sg, - enum i915_cache_level cache_level, + unsigned int pat_index, bool is_lmem, size_t sz) {@@ -858,7 +858,7 @@ static int __perf_clear_blt(struct intel_context *ce, t0 = ktime_get(); - err = intel_context_migrate_clear(ce, NULL, sg, cache_level, + err = intel_context_migrate_clear(ce, NULL, sg, pat_index, is_lmem, 0, &rq); if (rq) { if (i915_request_wait(rq, 0, MAX_SCHEDULE_TIMEOUT) < 0)@@ -904,7 +904,8 @@ static int perf_clear_blt(void *arg) err = __perf_clear_blt(gt->migrate.context, dst->mm.pages->sgl, - I915_CACHE_NONE, + i915_gem_get_pat_index(gt->i915, + I915_CACHE_NONE), i915_gem_object_is_lmem(dst), sizes[i]);@@ -919,10 +920,10 @@ static int perf_clear_blt(void *arg) static int __perf_copy_blt(struct intel_context *ce, struct scatterlist *src, - enum i915_cache_level src_cache_level, + unsigned int src_pat_index, bool src_is_lmem, struct scatterlist *dst, - enum i915_cache_level dst_cache_level, + unsigned int dst_pat_index, bool dst_is_lmem, size_t sz) {@@ -937,9 +938,9 @@ static int __perf_copy_blt(struct intel_context *ce, t0 = ktime_get(); err = intel_context_migrate_copy(ce, NULL, - src, src_cache_level, + src, src_pat_index, src_is_lmem, - dst, dst_cache_level, + dst, dst_pat_index, dst_is_lmem, &rq); if (rq) {@@ -994,10 +995,12 @@ static int perf_copy_blt(void *arg) err = __perf_copy_blt(gt->migrate.context, src->mm.pages->sgl, - I915_CACHE_NONE, + i915_gem_get_pat_index(gt->i915, + I915_CACHE_NONE), i915_gem_object_is_lmem(src), dst->mm.pages->sgl, - I915_CACHE_NONE, + i915_gem_get_pat_index(gt->i915, + I915_CACHE_NONE), i915_gem_object_is_lmem(dst), sz);drivers/gpu/drm/i915/gt/selftest_reset.c +6
-2@@ -86,7 +86,9 @@ __igt_reset_stolen(struct intel_gt *gt, ggtt->vm.insert_page(&ggtt->vm, dma, ggtt->error_capture.start, - I915_CACHE_NONE, 0); + i915_gem_get_pat_index(gt->i915, + I915_CACHE_NONE), + 0); mb(); s = io_mapping_map_wc(&ggtt->iomap,@@ -127,7 +129,9 @@ __igt_reset_stolen(struct intel_gt *gt, ggtt->vm.insert_page(&ggtt->vm, dma, ggtt->error_capture.start, - I915_CACHE_NONE, 0); + i915_gem_get_pat_index(gt->i915, + I915_CACHE_NONE), + 0); mb(); s = io_mapping_map_wc(&ggtt->iomap,drivers/gpu/drm/i915/gt/selftest_timeline.c +1
-1@@ -836,7 +836,7 @@ static int setup_watcher(struct hwsp_watcher *w,struct intel_gt *gt, return PTR_ERR(obj); /* keep the same cache settings as timeline */ - i915_gem_object_set_cache_coherency(obj, tl->hwsp_ggtt->obj->cache_level); + i915_gem_object_set_pat_index(obj, tl->hwsp_ggtt->obj->pat_index); w->map = i915_gem_object_pin_map_unlocked(obj,
page_unmask_bits(tl->hwsp_ggtt->obj->mm.mapping)); if (IS_ERR(w->map)) {drivers/gpu/drm/i915/gt/selftest_tlb.c +3
-1@@ -36,6 +36,8 @@ pte_tlbinv(struct intel_context *ce, u64 length, struct rnd_state *prng) { + const unsigned int pat_index = + i915_gem_get_pat_index(ce->vm->i915, I915_CACHE_NONE); struct drm_i915_gem_object *batch; struct drm_mm_node vb_node; struct i915_request *rq;@@ -155,7 +157,7 @@ pte_tlbinv(struct intel_context *ce, /* Flip the PTE between A and B */ if (i915_gem_object_is_lmem(vb->obj)) pte_flags |= PTE_LM; - ce->vm->insert_entries(ce->vm, &vb_res, 0, pte_flags); + ce->vm->insert_entries(ce->vm, &vb_res, pat_index, pte_flags); /* Flush the PTE update to concurrent HW */ tlbinv(ce->vm, addr & -length, length);drivers/gpu/drm/i915/gt/uc/intel_uc_fw.c +4
-1@@ -890,9 +890,15 @@ static void uc_fw_bind_ggtt(struct intel_uc_fw *uc_fw) pte_flags |= PTE_LM; if (ggtt->vm.raw_insert_entries) - ggtt->vm.raw_insert_entries(&ggtt->vm, dummy, I915_CACHE_NONE, pte_flags); + ggtt->vm.raw_insert_entries(&ggtt->vm, dummy, + i915_gem_get_pat_index(ggtt->vm.i915, + I915_CACHE_NONE), + pte_flags);else
-ggtt->vm.insert_entries(&ggtt->vm, dummy, I915_CACHE_NONE, pte_flags);
+ ggtt->vm.insert_entries(&ggtt->vm, dummy,
+ i915_gem_get_pat_index(ggtt->vm.i915, +I915_CACHE_NONE),
+pte_flags);
}
static voiduc_fw_unbind_ggtt(struct intel_uc_fw *uc_fw)drivers/gpu/drm/i915/i915_debugfs.c +4
-8@@ -139,21 +139,56 @@ static const char *stringify_vma_type(const struct i915_vma *vma) return "ppgtt"; } -static const char *i915_cache_level_str (struct drm_i915_private *i915, int type) -{ - switch (type) { - case I915_CACHE_NONE: return " uncached"; - case I915_CACHE_LLC: return HAS_LLC(i915) ? " LLC" : " snooped"; - case I915_CACHE_L3_LLC: return " L3+LLC"; - case I915_CACHE_WT: return " WT"; - default: return ""; +static const char *i915_cache_level_str (struct drm_i915_gem_object *obj) +{ + struct drm_i915_private *i915 = obj_to_i915(obj); +
Quoted 8 lines
drivers/gpu/drm/i915/i915_debugfs.c static const char *stringify_vma_type(const struct i915_vma *vma) · line 153
Quoted 8 lines
drivers/gpu/drm/i915/i915_debugfs.c +8
-0+ if (IS_METEORLAKE(i915)) { + switch (obj->pat_index) { + case 0: return " WB"; + case 1: return " WT"; + case 2: return " UC"; + case 3: return " WB (1-Way Coh)"; + case 4: return " WB (2-Way Coh)"; + default: return " not defined";
Is not defined possible?
Also, it may be nicer to handle the leading space in the caller.
Quoted 14 lines
Quoted 14 lines
drivers/gpu/drm/i915/i915_debugfs.c +14
-0+ } + } else if (IS_PONTEVECCHIO(i915)) { + switch (obj->pat_index) { + case 0: return " UC"; + case 1: return " WC"; + case 2: return " WT"; + case 3: return " WB"; + case 4: return " WT (CLOS1)"; + case 5: return " WB (CLOS1)"; + case 6: return " WT (CLOS2)"; + case 7: return " WT (CLOS2)"; + default: return " not defined"; + } + } else if (GRAPHICS_VER(i915) >= 12) {
Quoted 8 lines
drivers/gpu/drm/i915/i915_debugfs.c static const char *stringify_vma_type(const struct i915_vma *vma) · line 175
Quoted 8 lines
drivers/gpu/drm/i915/i915_debugfs.c +8
-0+ switch (obj->pat_index) { + case 0: return " WB"; + case 1: return " WC"; + case 2: return " WT"; + case 3: return " UC"; + default: return " not defined"; + } + } else {
Is this correct if a legacy platform used the set pat extension? I don't see that it is disallowed.
Would it simplify things to add a reverse table to device info, so like cachelevel_to_pat, just for pat_index to names? I guess it depends what names PRMs use for PATs on legacy platforms. Is it consistend with the above UC/WC/WB/... or with the below names.
Quoted 2 lines
drivers/gpu/drm/i915/i915_debugfs.c static const char *stringify_vma_type(const struct i915_vma *vma) · line 177
Quoted 2 lines
drivers/gpu/drm/i915/i915_debugfs.c +2
-0+ if (i915_gem_object_has_cache_level(obj, I915_CACHE_NONE)) + return " uncached";
UC for consistency?
Quoted 5 lines
drivers/gpu/drm/i915/i915_debugfs.c static const char *stringify_vma_type(const struct i915_vma *vma) · line 181
Quoted 5 lines
drivers/gpu/drm/i915/i915_debugfs.c +4
-0+ else if (i915_gem_object_has_cache_level(obj, I915_CACHE_LLC)) + return HAS_LLC(i915) ? " LLC" : " snooped"; + else if (i915_gem_object_has_cache_level(obj,I915_CACHE_L3_LLC )) + return " L3+LLC";
Is this correct if !HAS_LLC?
Quoted 4 lines
drivers/gpu/drm/i915/i915_debugfs.c static const char *stringify_vma_type(const struct i915_vma *vma) · line 185
Quoted 4 lines
drivers/gpu/drm/i915/i915_debugfs.c +4
-0+ else if (i915_gem_object_has_cache_level(obj, I915_CACHE_WT)) + return " WT"; + else + return " not defined";
Current code prints nothing for the default switch statement.
But is this even reachable or should it be MISSING_CASE warning?
Quoted 27 lines
Quoted 27 lines
drivers/gpu/drm/i915/i915_debugfs.c +0
-0} }void
i915_debugfs_describe_obj(struct seq_file *m, struct drm_i915_gem_object *obj){
- structdrm_i915_private*dev_priv=to_i915(obj->base.dev); structi915_vma*vma;
intpin_count= 0;drivers/gpu/drm/i915/i915_debugfs.c +1
-1@@ -165,7 +200,7 @@ i915_debugfs_describe_obj(struct seq_file *m, struct drm_i915_gem_object *obj) obj->base.size / 1024, obj->read_domains, obj->write_domain, - i915_cache_level_str(dev_priv, obj->cache_level), + i915_cache_level_str(obj), obj->mm.dirty ? " dirty" : "", obj->mm.madv == I915_MADV_DONTNEED ? " purgeable" : "");if (
obj->base.name)drivers/gpu/drm/i915/i915_gem.c +0
-1@@ -420,8 +420,12 @@ i915_gem_gtt_pread(struct drm_i915_gem_object *obj, page_length = remain < page_length ? remain : page_length; if (drm_mm_node_allocated(&node)) { ggtt->vm.insert_page(&ggtt->vm, - i915_gem_object_get_dma_address(obj, offset
Quoted 7 lines
drivers/gpu/drm/i915/i915_gem.c i915_gem_gtt_pread(struct drm_i915_gem_object *obj, · line 427
Quoted 7 lines
drivers/gpu/drm/i915/i915_gem.c +5
-1>> PAGE_SHIFT), - node.start, I915_CACHE_NONE, 0); + i915_gem_object_get_dma_address(obj, + offset >> PAGE_SHIFT), + node.start, + i915_gem_get_pat_index(i915, + I915_CACHE_NONE),
For the callsites which use const levels you could at least do something like i915->pat_cache_none, or I know the not very popular static inline i915_gem_get_pat_index so it can be evaluated at runtime. Not sure really, throwing out ideas which may be invalid if a more elegant refactoring is possible.
Regards,
Tvrtko
Quoted 275 lines
Quoted 274 lines
drivers/gpu/drm/i915/i915_gem.c +1
-0+ 0); } else { page_base += offset & PAGE_MASK;}
drivers/gpu/drm/i915/i915_gem.c +6
-2@@ -598,8 +602,12 @@ i915_gem_gtt_pwrite_fast(structdrm_i915_gem_object *obj, /* flush the write before we modify the GGTT */ intel_gt_flush_ggtt_writes(ggtt->vm.gt); ggtt->vm.insert_page(&ggtt->vm, - i915_gem_object_get_dma_address(obj, offset >>PAGE_SHIFT ), - node.start, I915_CACHE_NONE, 0); + i915_gem_object_get_dma_address(obj, + offset >> PAGE_SHIFT), + node.start, + i915_gem_get_pat_index(i915, + I915_CACHE_NONE), + 0);
wmb(); /* flush modifications to the GGTT (insert_page) */ } else {
page_base+= offset &PAGE_MASK;drivers/gpu/drm/i915/i915_gpu_error.c +3
-1@@ -1117,10 +1117,14 @@ i915_vma_coredump_create(const struct intel_gt *gt, mutex_lock(&ggtt->error_mutex); if (ggtt->vm.raw_insert_page) ggtt->vm.raw_insert_page(&ggtt->vm, dma, slot, - I915_CACHE_NONE, 0); + i915_gem_get_pat_index(gt->i915, + I915_CACHE_NONE), + 0);else
ggtt->vm.insert_page(&ggtt->vm, dma, slot, - I915_CACHE_NONE, 0);
+ i915_gem_get_pat_index(gt->i915, +I915_CACHE_NONE), + 0);
mb();
s =io_mapping_map_wc(&ggtt->iomap, slot, PAGE_SIZE);drivers/gpu/drm/i915/i915_vma.c +3
-3@@ -315,7 +315,7 @@ struct i915_vma_work { struct i915_vma_resource *vma_res; struct drm_i915_gem_object *obj; struct i915_sw_dma_fence_cb cb; - enum i915_cache_level cache_level; + unsigned int pat_index; unsigned int flags; };@@ -334,7 +334,7 @@ static void __vma_bind(struct dma_fence_work *work) return; vma_res->ops->bind_vma(vma_res->vm, &vw->stash, - vma_res, vw->cache_level, vw->flags); + vma_res, vw->pat_index, vw->flags); } static void __vma_release(struct dma_fence_work *work)@@ -426,7 +426,7 @@ i915_vma_resource_init_from_vma(struct i915_vma_resource *vma_res, /*** i915_vma_bind - Sets up PTEs for an VMA in it's corresponding address space. * @vma: VMA to map -* @cache_level: mapping cache level +* @pat_index: PAT index to set in PTE * @flags: flags like global or local mapping * @work: preallocated worker for allocating and binding the PTE * @vma_res: pointer to a preallocated vma resource. The resource is either
drivers/gpu/drm/i915/i915_vma.c +1
-1@@ -437,7 +437,7 @@ i915_vma_resource_init_from_vma(structi915_vma_resource *vma_res, * Note that DMA addresses are also the only part of the SG table we care about. */ int i915_vma_bind(struct i915_vma *vma, - enum i915_cache_level cache_level, + unsigned int pat_index, u32 flags, struct i915_vma_work *work,struct
i915_vma_resource*vma_res)drivers/gpu/drm/i915/i915_vma.c +4
-4@@ -507,7 +507,7 @@ int i915_vma_bind(struct i915_vma *vma, struct dma_fence *prev; work->vma_res = i915_vma_resource_get(vma->resource); - work->cache_level = cache_level; + work->pat_index = pat_index; work->flags = bind_flags; /*@@ -537,7 +537,7 @@ int i915_vma_bind(struct i915_vma *vma, return ret; } - vma->ops->bind_vma(vma->vm, NULL, vma->resource, cache_level, + vma->ops->bind_vma(vma->vm, NULL, vma->resource, pat_index, bind_flags); }@@ -814,7 +814,7 @@ i915_vma_insert(struct i915_vma *vma, structi915_gem_ww_ctx *ww, color = 0; if (i915_vm_has_cache_coloring(vma->vm)) - color = vma->obj->cache_level; + color = vma->obj->pat_index; if (flags & PIN_OFFSET_FIXED) { u64 offset = flags & PIN_OFFSET_MASK;@@ -1518,7 +1518,7 @@ int i915_vma_pin_ww(struct i915_vma *vma, structi915_gem_ww_ctx *ww, GEM_BUG_ON(!vma->pages); err = i915_vma_bind(vma, - vma->obj->cache_level, + vma->obj->pat_index, flags, work, vma_res); vma_res = NULL; if (err)drivers/gpu/drm/i915/i915_vma.h +1
-1@@ -250,7 +250,7 @@ i915_vma_compare(struct i915_vma *vma, struct i915_vma_work *i915_vma_work(void); int i915_vma_bind(struct i915_vma *vma, - enum i915_cache_level cache_level, + unsigned int pat_index, u32 flags, struct i915_vma_work *work, struct i915_vma_resource *vma_res);drivers/gpu/drm/i915/i915_vma_types.h +0
-2@@ -32,8 +32,6 @@ #include "gem/i915_gem_object_types.h" -enum i915_cache_level ; - /*** DOC: Global GTT views * drivers/gpu/drm/i915/selftests/i915_gem.c +4
-1@@ -57,7 +57,10 @@ static void trash_stolen(struct drm_i915_private *i915) u32 __iomem *s; int x; - ggtt->vm.insert_page(&ggtt->vm, dma, slot, I915_CACHE_NONE, 0); + ggtt->vm.insert_page(&ggtt->vm, dma, slot, + i915_gem_get_pat_index(i915, + I915_CACHE_NONE), + 0); s = io_mapping_map_atomic_wc(&ggtt->iomap, slot); for (x = 0; x < PAGE_SIZE / sizeof(u32); x++) {drivers/gpu/drm/i915/selftests/i915_gem_evict.c +2
-2@@ -245,7 +245,7 @@ static int igt_evict_for_cache_color(void *arg) struct drm_mm_node target = { .start = I915_GTT_PAGE_SIZE * 2, .size = I915_GTT_PAGE_SIZE, - .color = I915_CACHE_LLC, + .color = i915_gem_get_pat_index(gt->i915, I915_CACHE_LLC), }; struct drm_i915_gem_object *obj; struct i915_vma *vma;@@ -308,7 +308,7 @@ static int igt_evict_for_cache_color(void *arg) /* Attempt to remove the first *pinned* vma, by removing the (empty) * neighbour -- this should fail. */ - target.color = I915_CACHE_L3_LLC; + target.color = i915_gem_get_pat_index(gt->i915, I915_CACHE_L3_LLC); mutex_lock(&ggtt->vm.mutex); err = i915_gem_evict_for_node(&ggtt->vm, NULL, &target, 0);drivers/gpu/drm/i915/selftests/i915_gem_gtt.c +9
-4@@ -135,7 +135,7 @@ fake_dma_object(struct drm_i915_private *i915, u64 size) obj->write_domain = I915_GEM_DOMAIN_CPU; obj->read_domains = I915_GEM_DOMAIN_CPU; - obj->cache_level = I915_CACHE_NONE; + obj->pat_index = i915_gem_get_pat_index(i915, I915_CACHE_NONE); /* Preallocate the "backing storage" */ if (i915_gem_object_pin_pages_unlocked(obj))@@ -359,7 +359,9 @@ static int lowlevel_hole(struct i915_address_space *vm, with_intel_runtime_pm(vm->gt->uncore->rpm, wakeref) vm->insert_entries(vm, mock_vma_res, - I915_CACHE_NONE, 0); + i915_gem_get_pat_index(vm->i915, + I915_CACHE_NONE), + 0); } count = n;@@ -1377,7 +1379,10 @@ static int igt_ggtt_page(void *arg) ggtt->vm.insert_page(&ggtt->vm, i915_gem_object_get_dma_address(obj, 0), - offset, I915_CACHE_NONE, 0); + offset, + i915_gem_get_pat_index(i915, + I915_CACHE_NONE), + 0); } order = i915_random_order(count, &prng);@@ -1510,7 +1515,7 @@ static int reserve_gtt_with_resource(struct i915_vma *vma, u64 offset) mutex_lock(&vm->mutex); err = i915_gem_gtt_reserve(vm, NULL, &vma->node, obj->base.size, offset, - obj->cache_level, + obj->pat_index, 0); if (!err) {
i915_vma_resource_init_from_vma(vma_res, vma);drivers/gpu/drm/i915/selftests/i915_gem_gtt.c +1
-1@@ -1690,7 +1695,7 @@ static int insert_gtt_with_resource(struct i915_vma *vma) mutex_lock(&vm->mutex); err = i915_gem_gtt_insert(vm, NULL, &vma->node, obj->base.size, 0, - obj->cache_level, 0, vm->total, 0); + obj->pat_index, 0, vm->total, 0); if (!err) { i915_vma_resource_init_from_vma(vma_res, vma); vma->resource = vma_res;drivers/gpu/drm/i915/selftests/intel_memory_region.c +3
-1@@ -1070,7 +1070,9 @@ static int igt_lmem_write_cpu(void *arg) /* Put the pages into a known state -- from the gpu for added fun */ intel_engine_pm_get(engine); err = intel_context_migrate_clear(engine->gt->migrate.context, NULL, - obj->mm.pages->sgl, I915_CACHE_NONE, + obj->mm.pages->sgl, + i915_gem_get_pat_index(i915, + I915_CACHE_NONE), true, 0xdeadbeaf, &rq); if (rq) { dma_resv_add_fence(obj->base.resv, &rq->fence,drivers/gpu/drm/i915/selftests/mock_gtt.c +4
-4@@ -27,21 +27,21 @@ static void mock_insert_page(struct i915_address_space *vm, dma_addr_t addr, u64 offset, - enum i915_cache_level level, + unsigned int pat_index, u32 flags) { } static void mock_insert_entries(struct i915_address_space *vm, struct i915_vma_resource *vma_res, - enum i915_cache_level level, u32 flags) + unsigned int pat_index, u32 flags) { } static void mock_bind_ppgtt(struct i915_address_space *vm, struct i915_vm_pt_stash *stash, struct i915_vma_resource *vma_res, - enum i915_cache_level cache_level, + unsigned int pat_index, u32 flags) { GEM_BUG_ON(flags & I915_VMA_GLOBAL_BIND);@@ -94,7 +94,7 @@ struct i915_ppgtt *mock_ppgtt(struct drm_i915_private *i915, const char *name) static void mock_bind_ggtt(struct i915_address_space *vm, struct i915_vm_pt_stash *stash, struct i915_vma_resource *vma_res, - enum i915_cache_level cache_level, + unsigned int pat_index, u32 flags) {}
Re: [Intel-gfx] [PATCH 7/8] drm/i915: use pat_index instead of cache_level
Quoted 15 lines · Andrzej Hajda
Quoted 13 lines · On 20.04.2023 01:00, [email protected] wrote:
Quoted 12 lines
From: Fei Yang <[email protected]>
Currently the KMD is using enum
i915_cache_levelto set caching policy for
buffer objects. This is flaky because the PAT index which really controls the caching behavior in PTE has far more levels than what's defined in the
enum. In addition, the PAT index is platform dependent, having to translate
betweeni915_cache_leveland PAT index is not reliable, and makes the code
more complicated.
Quoted 3 lines
How it is flaky and not reliable - yet the series proposed to leave it in place and even claims using cache levels simplifies the code (lower in the commit message). Maybe just the commit message needs work.
If you look into the PTE encode functions, using cache_level there is not even correct. There is no way to map the 4 possible cache levels to all available PAT indices, let alone the number of PAT indices varies from platofrm to platform.
The architectural design allows UMD's to directly set PAT index, this is not possible with the use of cache_level in KMD. PAT index from UMD -> cache_level -> PAT index bits in PTE? Such translation is too much of a headache.
However in the kernel space, cache_level is still a good enough abstraction for KMD objects. Removing that would require code like this,
if (need to set uncached mode) {
if (MTL)
pat_index = 2;
else if (PVC)
pat_index = 0;
else if (GEN12)
pat_index = 3;
}
Quoted 18 lines
Quoted 18 lines
Quoted 17 lines
From UMD's perspective there is also a necessity to set caching
policy for
performance fine tuning. It's much easier for the UMD to directly use PAT index because the behavior of each PAT index is clearly defined in Bspec. Having the abstractedi915_cache_levelsitting in between would only cause
more ambiguity.For these reasons this patch replaces
i915_cache_levelwith PAT index. Also
note, thecache_levelis not completely removed yet, because the KMD still
has the need of creating buffer objects with simple cache settings such as
cached, uncached, or writethrough. For such simple cases, usingcache_level
would help simplify the code.
Quoted 3 lines
Quoted 3 lines
It seems quite fundamental change to me. Does this "not completely removed yet" means that in some future we will not have support for generic cache levels at all?
I think further simplification is possible if there is a consistent PAT table across all platforms. Looking at Xe 2/3, that seems to be the trend.
Quoted 20 lines
Quoted 20 lines
Seems strange to me. Even looking at the
number of users ofi915_gem_get_pat_indexbelow it seem very unlikely.And if the support for generic level will stay, maybe it would be better to make usage of it more convienient. All conversion of
f(..., cache_level, ...) to f(..., i915_gem_get_pat_index(i915, cache_level), ...)looks quite ugly to me.
Maybe extending cache level to support pat index somehow, for example:
enum i915_cache_level { I915_CACHE_NONE = 0, I915_CACHE_..., ... I915_CACHE_1ST_PAT_INDEX = 0x100, }so
real_pat_index=cache_level-I915_CACHE_1ST_PAT_INDEX
Quoted 8 lines
Quoted 5 lines
and in case of generic level there will be platform dependend conversion to
real_pat_index?I do not know the whole picture so maybe this is all wrong for some reason, just asking :)
It looks a bit unsightly to me too so yes please, brain storm on whether it can be made more elegant and less intrusive would be appreciated.
I still think using pat_index without abstraction is better...
Quoted 227 lines
Quoted 208 lines
Quoted 208 lines
Cc Chris Wilson <[email protected]>Cc Matt Roper <[email protected]>Signed-off-by Fei Yang <[email protected]>Reviewed-by Andi Shyti <[email protected]>--- drivers/gpu/drm/i915/display/intel_dpt.c | 12 +-- drivers/gpu/drm/i915/gem/i915_gem_domain.c | 27 ++---- .../gpu/drm/i915/gem/i915_gem_execbuffer.c | 10 ++- drivers/gpu/drm/i915/gem/i915_gem_mman.c | 3 +- drivers/gpu/drm/i915/gem/i915_gem_object.c | 52 +++++++++++- drivers/gpu/drm/i915/gem/i915_gem_object.h | 4 + .../gpu/drm/i915/gem/i915_gem_object_types.h | 25 +++++- drivers/gpu/drm/i915/gem/i915_gem_stolen.c | 4 +- drivers/gpu/drm/i915/gem/i915_gem_ttm_move.c | 16 ++-- .../gpu/drm/i915/gem/selftests/huge_pages.c | 2 +- .../drm/i915/gem/selftests/i915_gem_migrate.c | 2 +- .../drm/i915/gem/selftests/i915_gem_mman.c | 2 +- drivers/gpu/drm/i915/gt/gen6_ppgtt.c | 10 ++- drivers/gpu/drm/i915/gt/gen8_ppgtt.c | 71 ++++++++-------- drivers/gpu/drm/i915/gt/gen8_ppgtt.h | 3 +- drivers/gpu/drm/i915/gt/intel_ggtt.c | 82 +++++++++---------- drivers/gpu/drm/i915/gt/intel_gtt.h | 20 ++--- drivers/gpu/drm/i915/gt/intel_migrate.c | 47 ++++++----- drivers/gpu/drm/i915/gt/intel_migrate.h | 13 ++- drivers/gpu/drm/i915/gt/intel_ppgtt.c | 6 +- drivers/gpu/drm/i915/gt/selftest_migrate.c | 47 ++++++----- drivers/gpu/drm/i915/gt/selftest_reset.c | 8 +- drivers/gpu/drm/i915/gt/selftest_timeline.c | 2 +- drivers/gpu/drm/i915/gt/selftest_tlb.c | 4 +- drivers/gpu/drm/i915/gt/uc/intel_uc_fw.c | 10 ++- drivers/gpu/drm/i915/i915_debugfs.c | 55 ++++++++++--- drivers/gpu/drm/i915/i915_gem.c | 16 +++- drivers/gpu/drm/i915/i915_gpu_error.c | 8 +- drivers/gpu/drm/i915/i915_vma.c | 16 ++-- drivers/gpu/drm/i915/i915_vma.h | 2 +- drivers/gpu/drm/i915/i915_vma_types.h | 2 - drivers/gpu/drm/i915/selftests/i915_gem.c | 5 +- .../gpu/drm/i915/selftests/i915_gem_evict.c | 4 +- drivers/gpu/drm/i915/selftests/i915_gem_gtt.c | 15 ++-- .../drm/i915/selftests/intel_memory_region.c | 4 +- drivers/gpu/drm/i915/selftests/mock_gtt.c | 8 +- 36 files changed, 378 insertions(+), 239 deletions(-)drivers/gpu/drm/i915/display/intel_dpt.c +5
-5@@ -43,24 +43,24 @@ static void gen8_set_pte(void __iomem *addr, gen8_pte_t pte)static void dpt_insert_page (struct i915_address_space *vm,dma_addr_t addr,u64 offset, -enum i915_cache_level level, +unsigned int pat_index,u32 flags) {struct i915_dpt *dpt =i915_vm_to_dpt (vm);gen8_pte_t __iomem *base = dpt->iomem;gen8_set_pte (base + offset /I915_GTT_PAGE_SIZE , - vm->pte_encode (addr, level, flags)); + vm->pte_encode (addr, pat_index, flags)); }static void dpt_insert_entries (struct i915_address_space *vm,struct i915_vma_resource *vma_res, -enum i915_cache_level level, +unsigned int pat_index,u32 flags) {struct i915_dpt *dpt =i915_vm_to_dpt (vm);gen8_pte_t __iomem *base = dpt->iomem; -const gen8_pte_t pte_encode = vm->pte_encode (0 , level, flags); +const gen8_pte_t pte_encode = vm->pte_encode (0 , pat_index, flags);struct sgt_iter sgt_iter;dma_addr_t addr;int i;@@ -83,7 +83,7 @@ static void dpt_clear_range(struct i915_address_space *vm, static void dpt_bind_vma(struct i915_address_space *vm, struct i915_vm_pt_stash *stash, struct i915_vma_resource *vma_res, - enum i915_cache_level cache_level, + unsigned int pat_index, u32 flags) {u32 pte_flags;drivers/gpu/drm/i915/display/intel_dpt.c +1
-1@@ -98,7 +98,7 @@ static void dpt_bind_vma(struct i915_address_space *vm,if (vma_res->bi.lmem) pte_flags |=PTE_LM ; - vm->insert_entries (vm, vma_res, cache_level, pte_flags); + vm->insert_entries (vm, vma_res, pat_index, pte_flags); vma_res->page_sizes_gtt =I915_GTT_PAGE_SIZE ;drivers/gpu/drm/i915/gem/i915_gem_domain.c +9
-18@@ -27,8 +27,8 @@ static bool gpu_write_needs_clflush(struct drm_i915_gem_object *obj)if (IS_DGFX (i915))return false; -return !(obj->cache_level ==I915_CACHE_NONE || - obj->cache_level ==I915_CACHE_WT ); +return !(i915_gem_object_has_cache_level (obj,I915_CACHE_NONE ) || +i915_gem_object_has_cache_level (obj,I915_CACHE_WT )); }bool i915_gem_cpu_write_needs_clflush (struct drm_i915_gem_object *obj)@@ -267,7 +267,7 @@ int i915_gem_object_set_cache_level(structdrm_i915_gem_object *obj, { int ret; -if (obj->cache_level == cache_level) +if (i915_gem_object_has_cache_level (obj, cache_level))return 0 ; ret = i915_gem_object_wait(obj,@@ -278,10 +278,8 @@ int i915_gem_object_set_cache_level(structdrm_i915_gem_object *obj, return ret;/* Always invalidate stale cachelines */ -if (obj->cache_level != cache_level) { -i915_gem_object_set_cache_coherency (obj, cache_level); - obj->cache_dirty = true; - } +i915_gem_object_set_cache_coherency (obj, cache_level); + obj->cache_dirty = true;/* The cache-level will be applied when each vma is rebound. */ return i915_gem_object_unbind(obj,@@ -306,20 +304,13 @@ int i915_gem_get_caching_ioctl(struct drm_device *dev, void *data, goto out; } -switch (obj->cache_level) { -case I915_CACHE_LLC : -case I915_CACHE_L3_LLC : +if (i915_gem_object_has_cache_level (obj,I915_CACHE_LLC ) || +i915_gem_object_has_cache_level (obj,I915_CACHE_L3_LLC )) args->caching =I915_CACHING_CACHED ; -break ; - -case I915_CACHE_WT : +else if (i915_gem_object_has_cache_level (obj,I915_CACHE_WT )) args->caching =I915_CACHING_DISPLAY ; -break ; - -default : +else args->caching =I915_CACHING_NONE ; -break ; - } out:rcu_read_unlock ();return err;drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c +6
-4@@ -642,7 +642,7 @@ static inline int use_cpu_reloc(const struct reloc_cache *cache, return (cache->has_llc || obj->cache_dirty || - obj->cache_level !=I915_CACHE_NONE ); + !i915_gem_object_has_cache_level (obj,I915_CACHE_NONE )); } static int eb_reserve_vma(struct i915_execbuffer *eb,@@ -1323,8 +1323,10 @@ static void *reloc_iomap(struct i915_vma *batch, offset = cache->node.start;if (drm_mm_node_allocated (&cache->node)) { ggtt->vm.insert_page (&ggtt->vm, -i915_gem_object_get_dma_address (obj, page), - offset,I915_CACHE_NONE ,0 ); +i915_gem_object_get_dma_address (obj, page), + offset, +i915_gem_get_pat_index (ggtt->vm.i915,I915_CACHE_NONE ), +0 ); }else { offset += page <<PAGE_SHIFT ; }@@ -1464,7 +1466,7 @@ eb_relocate_entry(struct i915_execbuffer *eb,reloc_cache_unmap (&eb->reloc_cache);mutex_lock (&vma->vm->mutex); err =i915_vma_bind (target->vma, - target->vma->obj->cache_level, + target->vma->obj->pat_index,PIN_GLOBAL ,NULL ,NULL );mutex_unlock (&vma->vm->mutex);reloc_cache_remap (&eb->reloc_cache, ev->vma->obj);drivers/gpu/drm/i915/gem/i915_gem_mman.c +2
-1@@ -383,7 +383,8 @@ static vm_fault_t vm_fault_gtt(struct vm_fault *vmf) }/* Access to snoopable pages through the GTT is incoherent. */ -if (obj->cache_level !=I915_CACHE_NONE && !HAS_LLC (i915)) { +if (!(i915_gem_object_has_cache_level (obj,I915_CACHE_NONE ) || +HAS_LLC (i915))) { ret = -EFAULT ; goto err_unpin; }drivers/gpu/drm/i915/gem/i915_gem_object.c +0
-0@@ -54,6 +54,25 @@ unsigned int i915_gem_get_pat_index(struct drm_i915_private *i915,Quoted 18 lines
drivers/gpu/drm/i915/gem/i915_gem_object.c unsigned int i915_gem_get_pat_index(struct drm_i915_private *i915, · line 58
Quoted 5 lines
drivers/gpu/drm/i915/gem/i915_gem_object.c +2
-0return INTEL_INFO (i915)->cachelevel_to_pat[level]; } +bool i915_gem_object_has_cache_level (const struct drm_i915_gem_object *obj, +enum i915_cache_level lvl)The name suggest object can have more cache levels, maybe only my impression, up to you.
Quoted 9 lines
drivers/gpu/drm/i915/gem/i915_gem_object.c +9
-0+{ +/* +* cache_level == I915_CACHE_INVAL indicates the UMD's have set the +* caching policy through pat_index, in which case the KMD should +* leave the coherency to be managed by user space, simply return +* true here. +*/ +if (obj->cache_level ==I915_CACHE_INVAL ) +return true;
Quoted 2 lines
It's a "bit" counter intuitive that answer "has cache level" is yes when cache level is set to invalid!
The only case for this condition to be true is an object created by UMD through GEM_CREATE with PAT index specified by set_pat extension. In this case the KMD is not supposed to touch the setting.
Quoted 105 lines
I worry we don't create an impenetrable code base so I hope this can be improved.
Quoted 103 lines
Quoted 103 lines
drivers/gpu/drm/i915/gem/i915_gem_object.c +3
-0+ +/* +* Otherwise the pat_index should have been converted from
cache_leveldrivers/gpu/drm/i915/gem/i915_gem_object.c +3
-0+* so that the following comparison is valid. +*/ + return obj->pat_index == i915_gem_get_pat_index(obj_to_i915(obj),lvl);
drivers/gpu/drm/i915/gem/i915_gem_object.c +5
-1+} +struct drm_i915_gem_object *i915_gem_object_alloc (void ) {struct drm_i915_gem_object *obj;@@ -133,7 +152,7 @@ void i915_gem_object_set_cache_coherency(struct drm_i915_gem_object *obj, {struct drm_i915_private *i915 =to_i915 (obj->base.dev); - obj->cache_level = cache_level; + obj->pat_index =i915_gem_get_pat_index (i915, cache_level);if (cache_level !=I915_CACHE_NONE ) obj->cache_coherent = (I915_BO_CACHE_COHERENT_FOR_READ |@@ -148,6 +167,37 @@ void i915_gem_object_set_cache_coherency(struct drm_i915_gem_object *obj, !IS_DGFX (i915); } +/** +* i915_gem_object_set_pat_index - set PAT index to be used in PTE encode
+ * @obj: #drm_i915_gem_object
+ * @pat_index: PAT index
+ *
+ * This is a clone ofi915_gem_object_set_cache_coherencytaking pat index
+ * instead ofcache_levelas its second argument.
+ */drivers/gpu/drm/i915/gem/i915_gem_object.c +23
-0+void i915_gem_object_set_pat_index (struct drm_i915_gem_object *obj, +unsigned int pat_index) +{ +struct drm_i915_private *i915 =to_i915 (obj->base.dev); + +if (obj->pat_index == pat_index) +return ; + + obj->pat_index = pat_index; + +if (pat_index !=i915_gem_get_pat_index (i915,I915_CACHE_NONE )) + obj->cache_coherent = (I915_BO_CACHE_COHERENT_FOR_READ | +I915_BO_CACHE_COHERENT_FOR_WRITE ); +else if (HAS_LLC (i915)) + obj->cache_coherent =I915_BO_CACHE_COHERENT_FOR_READ ; +else + obj->cache_coherent =0 ; + + obj->cache_dirty = + !(obj->cache_coherent &I915_BO_CACHE_COHERENT_FOR_WRITE ) && + !IS_DGFX (i915); +} +bool i915_gem_object_can_bypass_llc (struct drm_i915_gem_object *obj) {struct drm_i915_private *i915 =to_i915 (obj->base.dev);drivers/gpu/drm/i915/gem/i915_gem_object.h +4
-0@@ -34,6 +34,8 @@ static inline bool i915_gem_object_size_2big(u64 size)unsigned int i915_gem_get_pat_index (struct drm_i915_private *i915,enum i915_cache_level level); +bool i915_gem_object_has_cache_level (const struct drm_i915_gem_object *obj, +enum i915_cache_level lvl);void i915_gem_init__objects (struct drm_i915_private *i915);void i915_objects_module_exit (void );@@ -764,6 +766,8 @@ bool i915_gem_object_has_unknown_state(struct drm_i915_gem_object *obj);void i915_gem_object_set_cache_coherency (struct drm_i915_gem_object *obj,unsigned int cache_level); +void i915_gem_object_set_pat_index (struct drm_i915_gem_object *obj, +unsigned int pat_index);bool i915_gem_object_can_bypass_llc (struct drm_i915_gem_object *obj);void i915_gem_object_flush_if_display (struct drm_i915_gem_object *obj);void i915_gem_object_flush_if_display_locked(struct
drm_i915_gem_object*obj);drivers/gpu/drm/i915/gem/i915_gem_object_types.h +1
-0@@ -195,6 +195,7 @@ enum i915_cache_level {*/ I915_CACHE_WT ,I915_MAX_CACHE_LEVEL , + I915_CACHE_INVAL = I915_MAX_CACHE_LEVEL, }; enum i915_map_type {@@ -358,10 +359,28 @@ struct drm_i915_gem_object {#define I915_BO_FLAG_STRUCT_PAGE BIT(0)/* Object backed by struct pages */
#defineI915_BO_FLAG_IOMEMBIT(1) /* Object backed by IO
memory *//**
Quoted 8 lines
Quoted 5 lines
Quoted 5 lines
drivers/gpu/drm/i915/gem/i915_gem_object_types.h +3
-1-* @cache_level: The desired GTT caching level. +* @pat_index: The desired PAT index. +* +* See hardware specification for valid PAT indices for each platform.
Side note for the last patch in the series - the UAPI blurb next to u32 index needs to at least point to some public PRM
That would be the Bspec.
As far as the design goes, the UMD's should be aware of the PAT index and its platform dependent nature. The KMD is doing a sanity check so that the PAT index won't go pass the boundary (e.g. 4 for MTL, 7 for PVC, 3 for TGL/ADL). I believe the UMD's must be maintaining a list for each platform...
Quoted 7 lines
which lists the PATs
and their configuration I would think. Otherwise it's not fully transparent how to use the feature.Quoted 1 line
Quoted 1 line
+ * This field used to contain a value of enum
i915_cache_level.What does this mean? Nothing is changed to unsigned here but just new field added.
This comment needs update. Will do in the next version.
Quoted 7 lines
It's
Quoted 4 lines
drivers/gpu/drm/i915/gem/i915_gem_object_types.h struct drm_i915_gem_object { · line 369
Quoted 4 lines
drivers/gpu/drm/i915/gem/i915_gem_object_types.h +4
-0+* changed to an unsigned int because PAT indices are being used by +* both UMD and KMD for caching policy control after GEN12. +* For backward compatibility, this field will continue to contain +* value of i915_cache_level for pre-GEN12 platforms so that the PTE Pat_index:6 is a copy of cache_level:3 pre-Gen12?
There wasn't PAT index defined in pre-GEN12 platforms, so this is just to say that pat_index would behave the same as the cache_level for these platforms.
Quoted 15 lines
But when I look at changes like:
drivers/gpu/drm/i915/gt/intel_ggtt.c +3
-1@@ -1302,7 +1298,9 @@ bool i915_ggtt_resume_vm(struct i915_address_space *vm)*/ vma->resource->bound_flags =0 ; vma->ops->bind_vma (vm,NULL , vma->resource, - obj ? obj->cache_level :0 , + obj ? obj->pat_index : +i915_gem_get_pat_index (vm->i915, +I915_CACHE_NONE ), was_bound);That suggests it is not a copy but that
obj->pat_indexis always valid and directly a PAT index.
Quoted 2 lines
In which case new
cache_levelenum to say "use pat instead" may indeed be nicer as Andrzej suggested.
I'm not sure I understand the concern here. I915_CACHE_NONE is 0, this is just trying to select between obj->pat_index and pat_index for UC, depending on whether the obj is valid.
Quoted 1,237 lines
Although it is not clear to me for a glance that we need both. Maybe all in driver object creation can use
cache_levelbut immediately convert to PAT internally and just don't storecache_level? I haven't looked in detail is my disclaimer though.. I guess it may boil down to does i915 ever need to read backcache_level, other than on the top entry points like setting it or so.Quoted 7 lines
drivers/gpu/drm/i915/gem/i915_gem_object_types.h struct drm_i915_gem_object { · line 375
Quoted 7 lines
drivers/gpu/drm/i915/gem/i915_gem_object_types.h +3
-0+* encode functions for these legacy platforms can stay the same. +* In the meantime platform specific tables are created to translate +* i915_cache_level into pat index, for more details check the macros
drivers/gpu/drm/i915/gem/i915_gem_object_types.h +3
-0+* defined i915/i915_pci.c, e.g. PVC_CACHELEVEL. +*/ + unsigned int pat_index:6;Existing bitfield takes up 7 bits. I'd check here with pahole if making
pat_indexa full u8 and changing the existing ones to u8 field:bits maybe ends up better overall.Quoted 1,218 lines
Quoted 1,218 lines
drivers/gpu/drm/i915/gem/i915_gem_object_types.h +3
-2+/** +* @cache_level: Indicate whether pat_index is set by UMD * -* See enum i915_cache_level for possible values, along with what -* each does. +* This used to hold desired GTT caching level, but is now replaced by
+ *pat_index. It's kept here for KMD to tell whether thepat_indexisdrivers/gpu/drm/i915/gem/i915_gem_object_types.h +3
-0+* set by UMD or converted from enum i915_cache_level. +* This field should be 0 by default, but I915_CACHE_INVAL if the +* pat_index is set by UMD. */ unsigned int cache_level:3;/** drivers/gpu/drm/i915/gem/i915_gem_stolen.c +3
-1@@ -565,7 +565,9 @@ static void dbg_poison(struct i915_ggtt *ggtt, ggtt->vm.insert_page (&ggtt->vm, addr, ggtt->error_capture.start, -I915_CACHE_NONE ,0 ); +i915_gem_get_pat_index (ggtt->vm.i915, +I915_CACHE_NONE ), +0 );mb (); s = io_mapping_map_wc(&ggtt->iomap,drivers/gpu/drm/i915/gem/i915_gem_ttm_move.c +2
-1@@ -214,7 +214,8 @@ static struct dma_fence *i915_ttm_accel_move(struct ttm_buffer_object *bo, intel_engine_pm_get(to_gt(i915)->migrate.context->engine); ret =intel_context_migrate_clear (to_gt (i915)->migrate.context, deps, - dst_st->sgl, dst_level, + dst_st->sgl, +i915_gem_get_pat_index (i915, dst_level),i915_ttm_gtt_binds_lmem (dst_mem),0 , &rq);} else {drivers/gpu/drm/i915/gem/i915_gem_ttm_move.c +7
-6@@ -227,12 +228,13 @@ static struct dma_fence *i915_ttm_accel_move(struct ttm_buffer_object *bo, src_level = i915_ttm_cache_level(i915, bo->resource, src_ttm);intel_engine_pm_get (to_gt (i915)->migrate.context->engine); ret =intel_context_migrate_copy (to_gt (i915)->migrate.context, - deps, src_rsgt->table.sgl, - src_level, -i915_ttm_gtt_binds_lmem (bo->resource), - dst_st->sgl, dst_level, -i915_ttm_gtt_binds_lmem (dst_mem), - &rq); + deps, src_rsgt->table.sgl, +i915_gem_get_pat_index (i915, src_level), +i915_ttm_gtt_binds_lmem (bo->resource), + dst_st->sgl, +i915_gem_get_pat_index (i915, dst_level), +i915_ttm_gtt_binds_lmem (dst_mem), + &rq);i915_refct_sgt_put (src_rsgt); }drivers/gpu/drm/i915/gem/selftests/huge_pages.c +1
-1@@ -354,7 +354,7 @@ fake_huge_pages_object(struct drm_i915_private *i915, u64 size, bool single) obj->write_domain =I915_GEM_DOMAIN_CPU ; obj->read_domains =I915_GEM_DOMAIN_CPU ; - obj->cache_level =I915_CACHE_NONE ; + obj->pat_index =i915_gem_get_pat_index (i915,I915_CACHE_NONE );return obj; }drivers/gpu/drm/i915/gem/selftests/i915_gem_migrate.c +1
-1@@ -219,7 +219,7 @@ static int __igt_lmem_pages_migrate(structintel_gt *gt, continue; err =intel_migrate_clear (>->migrate, &ww, deps, - obj->mm.pages->sgl, obj->cache_level, + obj->mm.pages->sgl, obj->pat_index,i915_gem_object_is_lmem (obj),0xdeadbeaf , &rq);if (rq) {drivers/gpu/drm/i915/gem/selftests/i915_gem_mman.c +1
-1@@ -1222,7 +1222,7 @@ static int __igt_mmap_migrate(structintel_memory_region **placements, } err =intel_context_migrate_clear (to_gt (i915)->migrate.context,NULL , - obj->mm.pages->sgl, obj->cache_level, + obj->mm.pages->sgl, obj->pat_index,i915_gem_object_is_lmem (obj),expand32 (POISON_INUSE ), &rq);i915_gem_object_unpin_pages(obj);drivers/gpu/drm/i915/gt/gen6_ppgtt.c +6
-4@@ -109,7 +109,7 @@ static void gen6_ppgtt_clear_range(structi915_address_space *vm, static void gen6_ppgtt_insert_entries(struct i915_address_space *vm, struct i915_vma_resource *vma_res, - enum i915_cache_level cache_level, + unsigned int pat_index, u32 flags) { struct i915_ppgtt *ppgtt = i915_vm_to_ppgtt(vm);@@ -117,7 +117,7 @@ static void gen6_ppgtt_insert_entries(structi915_address_space *vm, unsigned int first_entry = vma_res->start / I915_GTT_PAGE_SIZE;unsigned int act_pt = first_entry /GEN6_PTES ;unsigned int act_pte = first_entry %GEN6_PTES ; -const u32 pte_encode = vm->pte_encode (0 , cache_level, flags); +const u32 pte_encode = vm->pte_encode (0 , pat_index, flags);struct sgt_dma iter =sgt_dma (vma_res);gen6_pte_t *vaddr;@@ -227,7 +227,9 @@ static int gen6_ppgtt_init_scratch(struct gen6_ppgtt *ppgtt) vm->scratch[0]->encode = vm->pte_encode (px_dma (vm->scratch[0 ]), -I915_CACHE_NONE ,PTE_READ_ONLY ); +i915_gem_get_pat_index (vm->i915, +I915_CACHE_NONE ), +PTE_READ_ONLY ); vm->scratch[1 ] = vm->alloc_pt_dma (vm,I915_GTT_PAGE_SIZE_4K );if (IS_ERR (vm->scratch[1 ])) {@@ -278,7 +280,7 @@ static void gen6_ppgtt_cleanup(struct i915_address_space *vm)static void pd_vma_bind (struct i915_address_space *vm,struct i915_vm_pt_stash *stash,struct i915_vma_resource *vma_res, -enum i915_cache_level cache_level, +unsigned int pat_index,u32 unused) {struct i915_ggtt *ggtt = i915_vm_to_ggtt(vm);drivers/gpu/drm/i915/gt/gen8_ppgtt.c +13
-20@@ -56,7 +56,7 @@ static u64 gen8_pte_encode(dma_addr_t addr, }static u64 mtl_pte_encode (dma_addr_t addr, -enum i915_cache_level level, +unsigned int pat_index,u32 flags) {gen8_pte_t pte = addr |GEN8_PAGE_PRESENT |GEN8_PAGE_RW ;@@ -67,24 +67,17 @@ static u64 mtl_pte_encode(dma_addr_t addr,if (flags &PTE_LM ) pte |=GEN12_PPGTT_PTE_LM |GEN12_PPGTT_PTE_NC ; -switch (level) { -case I915_CACHE_NONE : - pte |=GEN12_PPGTT_PTE_PAT1 ; -break ; -case I915_CACHE_LLC : -case I915_CACHE_L3_LLC : - pte |=GEN12_PPGTT_PTE_PAT0 |GEN12_PPGTT_PTE_PAT1 ; -break ; -case I915_CACHE_WT : +if (pat_index &BIT (0 )) pte |=GEN12_PPGTT_PTE_PAT0 ; -break ; -default : -/* This should never happen. Added to deal with the compile -* error due to the addition of I915_MAX_CACHE_LEVEL. Will -* be removed by the pat_index patch. -*/ -break ; - } + +if (pat_index &BIT (1 )) + pte |=GEN12_PPGTT_PTE_PAT1 ; + +if (pat_index &BIT (2 )) + pte |=GEN12_PPGTT_PTE_PAT2 ; + +if (pat_index &BIT (3 )) + pte |=MTL_PPGTT_PTE_PAT3 ;return pte; }@@ -457,11 +450,11 @@ gen8_ppgtt_insert_pte(struct i915_ppgtt *ppgtt,struct i915_page_directory *pdp, struct sgt_dma *iter, u64 idx, - enum i915_cache_level cache_level, + unsigned int pat_index, u32 flags) { struct i915_page_directory *pd; -const gen8_pte_t pte_encode = ppgtt->vm.pte_encode (0 , cache_level, flags); +const gen8_pte_t pte_encode = ppgtt->vm.pte_encode (0 , pat_index, flags);gen8_pte_t *vaddr;pd = i915_pd_entry(pdp, gen8_pd_index(idx, 2));drivers/gpu/drm/i915/gt/gen8_ppgtt.c +7
-7@@ -504,10 +497,10 @@ static void xehpsdv_ppgtt_insert_huge(struct i915_address_space *vm, struct i915_vma_resource *vma_res, struct sgt_dma *iter, - enum i915_cache_level cache_level, + unsigned int pat_index, u32 flags) { - const gen8_pte_t pte_encode = vm->pte_encode(0, cache_level, flags); + const gen8_pte_t pte_encode = vm->pte_encode(0, pat_index, flags);unsigned int rem =sg_dma_len (iter->sg);u64 start = vma_res->start;u64 end = start + vma_res->vma_size;@@ -611,10 +604,10 @@ xehpsdv_ppgtt_insert_huge(structi915_address_space *vm, static void gen8_ppgtt_insert_huge(struct i915_address_space *vm, struct i915_vma_resource *vma_res, struct sgt_dma *iter, - enum i915_cache_level cache_level, + unsigned int pat_index, u32 flags) { - const gen8_pte_t pte_encode = vm->pte_encode(0, cache_level, flags); + const gen8_pte_t pte_encode = vm->pte_encode(0, pat_index, flags);unsigned int rem =sg_dma_len (iter->sg);u64 start = vma_res->start;@@ -734,7 +727,7 @@ static void gen8_ppgtt_insert_huge(structi915_address_space *vm, static void gen8_ppgtt_insert(struct i915_address_space *vm, struct i915_vma_resource *vma_res, - enum i915_cache_level cache_level, + unsigned int pat_index, u32 flags) { struct i915_ppgtt * const ppgtt = i915_vm_to_ppgtt(vm);@@ -742,9 +735,9 @@ static void gen8_ppgtt_insert(struct i915_address_space *vm, if (vma_res->bi.page_sizes.sg > I915_GTT_PAGE_SIZE) {if (HAS_64K_PAGES (vm->i915)) -xehpsdv_ppgtt_insert_huge (vm, vma_res, &iter, cache_level, flags); +xehpsdv_ppgtt_insert_huge (vm, vma_res, &iter, pat_index, flags); else - gen8_ppgtt_insert_huge(vm, vma_res, &iter, cache_level, flags); + gen8_ppgtt_insert_huge(vm, vma_res, &iter, pat_index,flags); } else { u64 idx = vma_res->start >> GEN8_PTE_SHIFT;drivers/gpu/drm/i915/gt/gen8_ppgtt.c +2
-2@@ -753,7 +746,7 @@ static void gen8_ppgtt_insert(structi915_address_space *vm,gen8_pdp_for_page_index (vm ,idx ); idx =gen8_ppgtt_insert_pte (ppgtt, pdp, &iter, idx, - cache_level, flags); + pat_index, flags); }while (idx); vma_res->page_sizes_gtt =I915_GTT_PAGE_SIZE ;@@ -763,7 +756,7 @@ static void gen8_ppgtt_insert(struct i915_address_space *vm, static void gen8_ppgtt_insert_entry(struct i915_address_space *vm, dma_addr_t addr, u64 offset, - enum i915_cache_level level, + unsigned int pat_index, u32 flags) {u64 idx = offset >> GEN8_PTE_SHIFT;drivers/gpu/drm/i915/gt/gen8_ppgtt.c +2
-2@@ -777,14 +770,14 @@ static void gen8_ppgtt_insert_entry(struct i915_address_space *vm,GEM_BUG_ON (pt->is_compact); vaddr =px_vaddr (pt); - vaddr[gen8_pd_index (idx,0 )] = vm->pte_encode (addr, level, flags); + vaddr[gen8_pd_index (idx,0 )] = vm->pte_encode (addr, pat_index, flags);drm_clflush_virt_range (&vaddr[gen8_pd_index (idx,0 )],sizeof (*vaddr)); }static void __xehpsdv_ppgtt_insert_entry_lm (struct i915_address_space *vm,dma_addr_t addr,u64 offset, -enum i915_cache_level level, +unsigned int pat_index,u32 flags){ u64 idx = offset >> GEN8_PTE_SHIFT;drivers/gpu/drm/i915/gt/gen8_ppgtt.c +10
-6@@ -807,20 +800,20 @@ static void __xehpsdv_ppgtt_insert_entry_lm(struct i915_address_space *vm, } vaddr =px_vaddr (pt); - vaddr[gen8_pd_index (idx,0 ) /16 ] = vm->pte_encode (addr, level, flags); + vaddr[gen8_pd_index (idx,0 ) /16 ] = vm->pte_encode (addr, pat_index, flags); }static void xehpsdv_ppgtt_insert_entry (struct i915_address_space *vm,dma_addr_t addr,u64 offset, -enum i915_cache_level level, +unsigned int pat_index,u32 flags) {if (flags &PTE_LM )return __xehpsdv_ppgtt_insert_entry_lm (vm, addr, offset, - level, flags); + pat_index, flags); -return gen8_ppgtt_insert_entry (vm, addr, offset, level, flags); +return gen8_ppgtt_insert_entry (vm, addr, offset, pat_index, flags); }static int gen8_init_scratch (struct i915_address_space *vm)@@ -855,7 +848,9 @@ static int gen8_init_scratch(struct i915_address_space *vm) vm->scratch[0]->encode = vm->pte_encode (px_dma (vm->scratch[0 ]), -I915_CACHE_NONE , pte_flags); +i915_gem_get_pat_index (vm->i915, +I915_CACHE_NONE ), + pte_flags);for (i =1 ; i <= vm->top; i++) {struct drm_i915_gem_object *obj;@@ -873,7 +868,9 @@ static int gen8_init_scratch(struct i915_address_space *vm) }fill_px (obj, vm->scratch[i -1 ]->encode); - obj->encode =gen8_pde_encode (px_dma (obj),I915_CACHE_NONE ); + obj->encode =gen8_pde_encode (px_dma (obj), +i915_gem_get_pat_index (vm->i915, +I915_CACHE_NONE )); vm->scratch[i] = obj; }drivers/gpu/drm/i915/gt/gen8_ppgtt.h +1
-2@@ -10,13 +10,12 @@struct i915_address_space ;struct intel_gt ; -enum i915_cache_level ;struct i915_ppgtt *gen8_ppgtt_create (struct intel_gt *gt,unsigned long lmem_pt_obj_flags);u64 gen8_ggtt_pte_encode (dma_addr_t addr, -enum i915_cache_level level, +unsigned int pat_index,u32 flags);#endif drivers/gpu/drm/i915/gt/intel_ggtt.c +13
-26@@ -221,7 +221,7 @@ static void guc_ggtt_invalidate(struct i915_ggtt *ggtt) }static u64 mtl_ggtt_pte_encode (dma_addr_t addr, -enum i915_cache_level level, +unsigned int pat_index,u32 flags) {gen8_pte_t pte = addr |GEN8_PAGE_PRESENT ;@@ -231,30 +231,17 @@ static u64 mtl_ggtt_pte_encode(dma_addr_t addr,if (flags &PTE_LM ) pte |=GEN12_GGTT_PTE_LM ; -switch (level) { -case I915_CACHE_NONE : - pte |=MTL_GGTT_PTE_PAT1 ; -break ; -case I915_CACHE_LLC : -case I915_CACHE_L3_LLC : - pte |=MTL_GGTT_PTE_PAT0 |MTL_GGTT_PTE_PAT1 ; -break ; -case I915_CACHE_WT : +if (pat_index &BIT (0 )) pte |=MTL_GGTT_PTE_PAT0 ; -break ; -default : -/* This should never happen. Added to deal with the compile -* error due to the addition of I915_MAX_CACHE_LEVEL. Will -* be removed by the pat_index patch. -*/ -break ; - } + +if (pat_index &BIT (1 )) + pte |=MTL_GGTT_PTE_PAT1 ;return pte; }u64 gen8_ggtt_pte_encode (dma_addr_t addr, -enum i915_cache_level level, +unsigned int pat_index,u32 flags) {gen8_pte_t pte = addr |GEN8_PAGE_PRESENT ;@@ -273,25 +260,25 @@ static void gen8_set_pte(void __iomem *addr, gen8_pte_t pte)static void gen8_ggtt_insert_page (struct i915_address_space *vm,dma_addr_t addr,u64 offset, -enum i915_cache_level level, +unsigned int pat_index,u32 flags) {struct i915_ggtt *ggtt =i915_vm_to_ggtt (vm);gen8_pte_t __iomem *pte = (gen8_pte_t __iomem *)ggtt->gsm + offset /I915_GTT_PAGE_SIZE ; -gen8_set_pte (pte, ggtt->vm.pte_encode (addr, level, flags)); +gen8_set_pte (pte, ggtt->vm.pte_encode (addr, pat_index, flags)); ggtt->invalidate (ggtt); }static void gen8_ggtt_insert_entries (struct i915_address_space *vm,struct i915_vma_resource *vma_res, -enum i915_cache_level level, +unsigned int pat_index,u32 flags) {struct i915_ggtt *ggtt =i915_vm_to_ggtt (vm); -const gen8_pte_t pte_encode = ggtt->vm.pte_encode (0 , level, flags); +const gen8_pte_t pte_encode = ggtt->vm.pte_encode (0 , pat_index, flags);gen8_pte_t __iomem *gte;gen8_pte_t __iomem *end;struct sgt_iter iter;@@ -348,14 +335,14 @@ static void gen8_ggtt_clear_range(structi915_address_space *vm, static void gen6_ggtt_insert_page(struct i915_address_space *vm, dma_addr_t addr, u64 offset, - enum i915_cache_level level, + unsigned int pat_index, u32 flags) { struct i915_ggtt *ggtt = i915_vm_to_ggtt(vm);gen6_pte_t __iomem *pte = (gen6_pte_t __iomem *)ggtt->gsm + offset /I915_GTT_PAGE_SIZE ; -iowrite32 (vm->pte_encode (addr, level, flags), pte); +iowrite32 (vm->pte_encode (addr, pat_index, flags), pte); ggtt->invalidate (ggtt); }@@ -368,7 +355,7 @@ static void gen6_ggtt_insert_page(structi915_address_space *vm, */ static void gen6_ggtt_insert_entries (struct i915_address_space *vm,struct i915_vma_resource *vma_res, -enum i915_cache_level level, +unsigned int pat_index,u32 flags) {struct i915_ggtt *ggtt = i915_vm_to_ggtt(vm);drivers/gpu/drm/i915/gt/intel_ggtt.c +1
-1@@ -385,7 +372,7 @@ static void gen6_ggtt_insert_entries(struct i915_address_space *vm,iowrite32 (vm->scratch[0 ]->encode, gte++); end += (vma_res->node_size + vma_res->guard) /I915_GTT_PAGE_SIZE ;for_each_sgt_daddr (addr, iter, vma_res->bi.pages) -iowrite32 (vm->pte_encode (addr, level, flags), gte++); +iowrite32 (vm->pte_encode (addr, pat_index, flags), gte++);GEM_BUG_ON (gte > end);/* Fill the allocated but "unused" space beyond the end of the buffer */
drivers/gpu/drm/i915/gt/intel_ggtt.c +11
-9@@ -420,14 +407,15 @@ struct insert_page {struct i915_address_space *vm;dma_addr_t addr;u64 offset; -enum i915_cache_level level; +unsigned int pat_index; };static int bxt_vtd_ggtt_insert_page__cb (void *_arg) {struct insert_page *arg = _arg; -gen8_ggtt_insert_page (arg->vm, arg->addr, arg->offset, arg->level, 0); +gen8_ggtt_insert_page (arg->vm, arg->addr, arg->offset, + arg->pat_index,0 );bxt_vtd_ggtt_wa (arg->vm);return 0 ;@@ -436,10 +424,10 @@ static int bxt_vtd_ggtt_insert_page__cb(void *_arg)static void bxt_vtd_ggtt_insert_page__BKL (struct i915_address_space *vm,dma_addr_t addr,u64 offset, -enum i915_cache_level level, +unsigned int pat_index,u32 unused) { -struct insert_page arg = { vm, addr, offset, level }; +struct insert_page arg = { vm, addr, offset, pat_index };stop_machine (bxt_vtd_ggtt_insert_page__cb, &arg,NULL ); }@@ -447,7 +435,7 @@ static void bxt_vtd_ggtt_insert_page__BKL(struct i915_address_space *vm, struct insert_entries {struct i915_address_space *vm;struct i915_vma_resource *vma_res; -enum i915_cache_level level; +unsigned int pat_index;u32 flags; };@@ -455,7 +443,8 @@ static int bxt_vtd_ggtt_insert_entries__cb(void *_arg) {struct insert_entries *arg = _arg; -gen8_ggtt_insert_entries (arg->vm, arg->vma_res, arg->level, arg->flags); +gen8_ggtt_insert_entries (arg->vm, arg->vma_res, + arg->pat_index, arg->flags);bxt_vtd_ggtt_wa (arg->vm);return 0 ;@@ -463,10 +452,10 @@ static int bxt_vtd_ggtt_insert_entries__cb(void *_arg)static void bxt_vtd_ggtt_insert_entries__BKL (struct i915_address_space *vm,struct i915_vma_resource *vma_res, -enum i915_cache_level level, +unsigned int pat_index,u32 flags) { -struct insert_entries arg = { vm, vma_res, level, flags }; +struct insert_entries arg = { vm, vma_res, pat_index, flags };stop_machine (bxt_vtd_ggtt_insert_entries__cb, &arg,NULL ); }@@ -495,7 +484,7 @@ static void gen6_ggtt_clear_range(struct i915_address_space *vm, void intel_ggtt_bind_vma(struct i915_address_space *vm, struct i915_vm_pt_stash *stash, struct i915_vma_resource *vma_res, - enum i915_cache_level cache_level, + unsigned int pat_index, u32 flags) {u32 pte_flags;drivers/gpu/drm/i915/gt/intel_ggtt.c +15
-6@@ -512,7 +501,7 @@ void intel_ggtt_bind_vma(struct i915_address_space *vm, if (vma_res->bi.lmem) pte_flags |= PTE_LM; - vm->insert_entries (vm, vma_res, cache_level, pte_flags); + vm->insert_entries (vm, vma_res, pat_index, pte_flags); vma_res->page_sizes_gtt =I915_GTT_PAGE_SIZE ; }@@ -661,7 +650,7 @@ static int init_ggtt(struct i915_ggtt *ggtt)static void aliasing_gtt_bind_vma (struct i915_address_space *vm,struct i915_vm_pt_stash *stash,struct i915_vma_resource *vma_res, -enum i915_cache_level cache_level, +unsigned int pat_index,u32 flags) {u32 pte_flags;@@ -673,10 +662,10 @@ static void aliasing_gtt_bind_vma(struct i915_address_space *vm, if (flags & I915_VMA_LOCAL_BIND)ppgtt_bind_vma (&i915_vm_to_ggtt (vm)->alias->vm, - stash, vma_res, cache_level, flags); + stash, vma_res, pat_index, flags);if (flags &I915_VMA_GLOBAL_BIND ) - vm->insert_entries (vm, vma_res, cache_level, pte_flags); + vm->insert_entries (vm, vma_res, pat_index, pte_flags); vma_res->bound_flags |= flags; }@@ -933,7 +922,9 @@ static int ggtt_probe_common(struct i915_ggtt *ggtt, u64 size) ggtt->vm.scratch[0 ]->encode = ggtt->vm.pte_encode (px_dma (ggtt->vm.scratch[0 ]), -I915_CACHE_NONE , pte_flags); +i915_gem_get_pat_index (i915, +I915_CACHE_NONE ), + pte_flags);return 0 ; }@@ -1022,6 +1013,11 @@ static int gen8_gmch_probe(struct i915_ggtt *ggtt)return ggtt_probe_common (ggtt, size); } +/* +* For pre-gen8 platforms pat_index is the same as enum i915_cache_level, +* so these PTE encode functions are left with using cache_level. +* See translation table LEGACY_CACHELEVEL. +*/ static u64 snb_pte_encode (dma_addr_t addr,enum i915_cache_level level,u32 flags)@@ -1302,7 +1298,9 @@ bool i915_ggtt_resume_vm(structi915_address_space *vm) */ vma->resource->bound_flags =0 ; vma->ops->bind_vma (vm,NULL , vma->resource, - obj ? obj->cache_level :0 , + obj ? obj->pat_index : +i915_gem_get_pat_index (vm->i915, +I915_CACHE_NONE ), was_bound);if (obj) {/* only used during resume => exclusive access */ drivers/gpu/drm/i915/gt/intel_gtt.h +7
-9@@ -165,8 +165,6 @@ typedef u64 gen8_pte_t;#define MTL_2_COH_1W REG_FIELD_PREP(MTL_PAT_INDEX_COH_MODE_MASK, 2)#define MTL_0_COH_NON REG_FIELD_PREP(MTL_PAT_INDEX_COH_MODE_MASK, 0) -enum i915_cache_level ; -struct drm_i915_gem_object ;struct i915_fence_reg ;struct i915_vma ;@@ -234,7 +232,7 @@ struct i915_vma_ops {void (*bind_vma)(struct i915_address_space *vm,struct i915_vm_pt_stash *stash,struct i915_vma_resource *vma_res, -enum i915_cache_level cache_level, +unsigned int pat_index,u32 flags);/* * Unmap an object from an address space. This usually consists of @@ -306,7 +304,7 @@ struct i915_address_space { (*alloc_scratch_dma)(struct i915_address_space *vm, int sz);u64 (*pte_encode)(dma_addr_t addr, -enum i915_cache_level level, +unsigned int pat_index,u32 flags);/* Create a valid PTE */ #define PTE_READ_ONLY BIT(0)#define PTE_LM BIT(1)@@ -321,20 +319,20 @@ struct i915_address_space {void (*insert_page)(struct i915_address_space *vm,dma_addr_t addr,u64 offset, -enum i915_cache_level cache_level, +unsigned int pat_index,u32 flags);void (*insert_entries)(struct i915_address_space *vm,struct i915_vma_resource *vma_res, -enum i915_cache_level cache_level, +unsigned int pat_index,u32 flags);void (*raw_insert_page)(struct i915_address_space *vm,dma_addr_t addr,u64 offset, -enum i915_cache_level cache_level, +unsigned int pat_index,u32 flags);void (*raw_insert_entries)(struct i915_address_space *vm,struct i915_vma_resource *vma_res, -enum i915_cache_level cache_level, +unsigned int pat_index,u32 flags);void (*cleanup)(struct i915_address_space *vm);@@ -581,7 +579,7 @@ void ppgtt_init(struct i915_ppgtt *ppgtt, structintel_gt *gt, void intel_ggtt_bind_vma(struct i915_address_space *vm,struct i915_vm_pt_stash *stash,struct i915_vma_resource *vma_res, -enum i915_cache_level cache_level, +unsigned int pat_index,u32 flags); void intel_ggtt_unbind_vma(struct i915_address_space *vm,struct i915_vma_resource *vma_res);drivers/gpu/drm/i915/gt/intel_gtt.h +1
-1@@ -639,7 +637,7 @@ void __set_pd_entry(struct i915_page_directory * const pd, const unsigned short idx, struct i915_page_table *pt, - u64 (*encode)(const dma_addr_t, const enum i915_cache_level)); +u64 (*encode)(const dma_addr_t ,const unsigned int pat_index));#define set_pd_entry (pd, idx, to) \__set_pd_entry((pd), (idx), px_pt(to), gen8_pde_encode)drivers/gpu/drm/i915/gt/intel_gtt.h +1
-1@@ -659,7 +657,7 @@ void gen6_ggtt_invalidate(struct i915_ggtt *ggtt);void ppgtt_bind_vma (struct i915_address_space *vm,struct i915_vm_pt_stash *stash,struct i915_vma_resource *vma_res, -enum i915_cache_level cache_level, +unsigned int pat_index,u32 flags);void ppgtt_unbind_vma (struct i915_address_space *vm,struct i915_vma_resource *vma_res);drivers/gpu/drm/i915/gt/intel_migrate.c +3
-1@@ -45,7 +45,9 @@ static void xehpsdv_toggle_pdes(structi915_address_space *vm, * Insert a dummy PTE into every PT that will map to LMEM to ensure * we have a correctly setup PDE structure for later use. */ - vm->insert_page (vm,0 , d->offset,I915_CACHE_NONE ,PTE_LM ); + vm->insert_page (vm,0 , d->offset, +i915_gem_get_pat_index (vm->i915,I915_CACHE_NONE ), +PTE_LM );GEM_BUG_ON (!pt->is_compact); d->offset +=SZ_2M ;}drivers/gpu/drm/i915/gt/intel_migrate.c +23
-20@@ -63,7 +65,9 @@ static void xehpsdv_insert_pte(structi915_address_space *vm, * alignment is 64K underneath for the pt, and we are careful * not to access the space in the void. */ - vm->insert_page (vm,px_dma (pt), d->offset,I915_CACHE_NONE ,PTE_LM ); + vm->insert_page (vm,px_dma (pt), d->offset, +i915_gem_get_pat_index (vm->i915,I915_CACHE_NONE ), +PTE_LM ); d->offset +=SZ_64K ; }@@ -73,7 +77,8 @@ static void insert_pte(struct i915_address_space *vm, {struct insert_pte_data *d = data; - vm->insert_page (vm,px_dma (pt), d->offset,I915_CACHE_NONE , + vm->insert_page (vm,px_dma (pt), d->offset, +i915_gem_get_pat_index (vm->i915,I915_CACHE_NONE ),i915_gem_object_is_lmem (pt->base) ?PTE_LM :0 ); d->offset +=PAGE_SIZE ; }@@ -356,13 +361,13 @@ static int max_pte_pkt_size(struct i915_request *rq, int pkt)static int emit_pte (struct i915_request *rq,struct sgt_dma *it, -enum i915_cache_level cache_level, +unsigned int pat_index,bool is_lmem,u64 offset,int length) {bool has_64K_pages =HAS_64K_PAGES (rq->engine->i915); -const u64 encode = rq->context->vm->pte_encode (0 , cache_level, +const u64 encode = rq->context->vm->pte_encode (0 , pat_index, is_lmem ?PTE_LM :0 );struct intel_ring *ring = rq->ring;int pkt, dword_length;@@ -673,17 +678,17 @@ int intel_context_migrate_copy(struct intel_context *ce, const struct i915_deps *deps, struct scatterlist *src, - enum i915_cache_level src_cache_level, + unsigned int src_pat_index, bool src_is_lmem, struct scatterlist *dst, - enum i915_cache_level dst_cache_level, + unsigned int dst_pat_index, bool dst_is_lmem, struct i915_request **out) { struct sgt_dma it_src = sg_sgt(src), it_dst = sg_sgt(dst), it_ccs;struct drm_i915_private *i915 = ce->engine->i915;u64 ccs_bytes_to_cpy =0 , bytes_to_cpy; -enum i915_cache_level ccs_cache_level; +unsigned int ccs_pat_index;u32 src_offset, dst_offset;u8 src_access, dst_access;struct i915_request *rq;@@ -707,12 +712,12 @@ intel_context_migrate_copy(struct intel_context *ce, dst_sz =scatter_list_length (dst);if (src_is_lmem) { it_ccs = it_dst; - ccs_cache_level = dst_cache_level; + ccs_pat_index = dst_pat_index; ccs_is_src = false; }else if (dst_is_lmem) { bytes_to_cpy = dst_sz; it_ccs = it_src; - ccs_cache_level = src_cache_level; + ccs_pat_index = src_pat_index; ccs_is_src = true; }@@ -773,7 +778,7 @@ intel_context_migrate_copy(struct intel_context *ce, src_sz =calculate_chunk_sz (i915, src_is_lmem, bytes_to_cpy, ccs_bytes_to_cpy); - len =emit_pte (rq, &it_src, src_cache_level, src_is_lmem, + len =emit_pte (rq, &it_src, src_pat_index, src_is_lmem, src_offset, src_sz);if (!len) { err = -EINVAL ;@@ -784,7 +789,7 @@ intel_context_migrate_copy(struct intel_context *ce, goto out_rq; } - err =emit_pte (rq, &it_dst, dst_cache_level, dst_is_lmem, + err =emit_pte (rq, &it_dst, dst_pat_index, dst_is_lmem, dst_offset, len);if (err <0 ) goto out_rq;@@ -811,7 +816,7 @@ intel_context_migrate_copy(struct intel_context *ce, goto out_rq; ccs_sz =GET_CCS_BYTES (i915, len); - err =emit_pte (rq, &it_ccs, ccs_cache_level, false, + err =emit_pte (rq, &it_ccs, ccs_pat_index, false, ccs_is_src ? src_offset : dst_offset, ccs_sz);if (err <0 )@@ -979,7 +984,7 @@ int intel_context_migrate_clear(struct intel_context *ce, const struct i915_deps *deps, struct scatterlist *sg, - enum i915_cache_level cache_level, + unsigned int pat_index, bool is_lmem, u32 value, struct i915_request **out)@@ -1027,7 +1032,7 @@ intel_context_migrate_clear(struct intel_context *ce, if (err) goto out_rq; - len =emit_pte (rq, &it, cache_level, is_lmem, offset,CHUNK_SZ ); + len =emit_pte (rq, &it, pat_index, is_lmem, offset,CHUNK_SZ );if (len <=0 ) { err = len; goto out_rq;@@ -1074,10 +1079,10 @@ int intel_migrate_copy(struct intel_migrate *m,struct i915_gem_ww_ctx *ww, const struct i915_deps *deps, struct scatterlist *src, - enum i915_cache_level src_cache_level, + unsigned int src_pat_index, bool src_is_lmem, struct scatterlist *dst, - enum i915_cache_level dst_cache_level, + unsigned int dst_pat_index, bool dst_is_lmem, struct i915_request **out) {@@ -1098,8 +1103,8 @@ int intel_migrate_copy(struct intel_migrate *m, goto out; err =intel_context_migrate_copy (ce, deps, - src, src_cache_level, src_is_lmem, - dst, dst_cache_level, dst_is_lmem, + src, src_pat_index, src_is_lmem, + dst, dst_pat_index, dst_is_lmem, out);intel_context_unpin (ce);@@ -1113,7 +1118,7 @@ intel_migrate_clear(struct intel_migrate *m,struct i915_gem_ww_ctx *ww, const struct i915_deps *deps, struct scatterlist *sg, - enum i915_cache_level cache_level, + unsigned int pat_index, bool is_lmem, u32 value, struct i915_request **out)@@ -1134,7 +1139,7 @@ intel_migrate_clear(struct intel_migrate *m,if (err) goto out; - err =intel_context_migrate_clear (ce, deps, sg, cache_level, + err =intel_context_migrate_clear (ce, deps, sg, pat_index, is_lmem, value, out);intel_context_unpin (ce);drivers/gpu/drm/i915/gt/intel_migrate.h +6
-7@@ -16,7 +16,6 @@ struct i915_request;struct i915_gem_ww_ctx ;struct intel_gt ;struct scatterlist ; -enum i915_cache_level ;int intel_migrate_init (struct intel_migrate *m,struct intel_gt *gt);@@ -26,20 +25,20 @@ int intel_migrate_copy(struct intel_migrate *m,struct i915_gem_ww_ctx *ww, const struct i915_deps *deps, struct scatterlist *src, - enum i915_cache_level src_cache_level, + unsigned int src_pat_index, bool src_is_lmem, struct scatterlist *dst, - enum i915_cache_level dst_cache_level, + unsigned int dst_pat_index, bool dst_is_lmem, struct i915_request **out);int intel_context_migrate_copy (struct intel_context *ce,const struct i915_deps *deps,struct scatterlist *src, -enum i915_cache_level src_cache_level, +unsigned int src_pat_index,bool src_is_lmem,struct scatterlist *dst, -enum i915_cache_level dst_cache_level, +unsigned int dst_pat_index,bool dst_is_lmem,struct i915_request **out);@@ -48,7 +47,7 @@ intel_migrate_clear(struct intel_migrate *m,struct i915_gem_ww_ctx *ww, const struct i915_deps *deps, struct scatterlist *sg, - enum i915_cache_level cache_level, + unsigned int pat_index, bool is_lmem, u32 value, struct i915_request **out);@@ -56,7 +55,7 @@ int intel_context_migrate_clear(struct intel_context *ce, const struct i915_deps *deps, struct scatterlist *sg, - enum i915_cache_level cache_level, + unsigned int pat_index, bool is_lmem, u32 value, struct i915_request **out);drivers/gpu/drm/i915/gt/intel_ppgtt.c +1
-1@@ -98,7 +98,7 @@ void __set_pd_entry(struct i915_page_directory * const pd, const unsigned short idx, struct i915_page_table * const to, - u64 (*encode)(const dma_addr_t, const enum i915_cache_level)) + u64 (*encode)(const dma_addr_t, const unsigned int)) {/* Each thread pre-pins the pd, and we may have a thread per pde. */
GEM_BUG_ON(atomic_read(px_used(pd)) > NALLOC * I915_PDES);drivers/gpu/drm/i915/gt/intel_ppgtt.c +1
-1@@ -181,7 +181,7 @@ struct i915_ppgtt *i915_ppgtt_create(struct intel_gt *gt, void ppgtt_bind_vma(struct i915_address_space *vm, struct i915_vm_pt_stash *stash, struct i915_vma_resource *vma_res, - enum i915_cache_level cache_level, + unsigned int pat_index, u32 flags) {u32 pte_flags;drivers/gpu/drm/i915/gt/intel_ppgtt.c +1
-1@@ -199,7 +199,7 @@ void ppgtt_bind_vma(struct i915_address_space *vm,if (vma_res->bi.lmem) pte_flags |=PTE_LM ; - vm->insert_entries (vm, vma_res, cache_level, pte_flags); + vm->insert_entries (vm, vma_res, pat_index, pte_flags);wmb (); }drivers/gpu/drm/i915/gt/selftest_migrate.c +8
-8@@ -137,7 +137,7 @@ static int copy(struct intel_migrate *migrate,static int intel_context_copy_ccs (struct intel_context *ce,const struct i915_deps *deps,struct scatterlist *sg, -enum i915_cache_level cache_level, +unsigned int pat_index,bool write_to_ccs,struct i915_request **out) {@@ -185,7 +185,7 @@ static int intel_context_copy_ccs(structintel_context *ce, if (err) goto out_rq; - len = emit_pte(rq, &it, cache_level, true, offset, CHUNK_SZ); + len = emit_pte(rq, &it, pat_index, true, offset, CHUNK_SZ);if (len <=0 ) { err = len; goto out_rq;@@ -223,7 +223,7 @@ intel_migrate_ccs_copy(struct intel_migrate *m,struct i915_gem_ww_ctx *ww, const struct i915_deps *deps, struct scatterlist *sg, - enum i915_cache_level cache_level, + unsigned int pat_index, bool write_to_ccs, struct i915_request **out) {@@ -243,7 +243,7 @@ intel_migrate_ccs_copy(struct intel_migrate *m,if (err) goto out; - err =intel_context_copy_ccs (ce, deps, sg, cache_level, + err =intel_context_copy_ccs (ce, deps, sg, pat_index, write_to_ccs, out);intel_context_unpin (ce);@@ -300,7 +300,7 @@ static int clear(struct intel_migrate *migrate,/* Write the obj data into ccs surface */ err =intel_migrate_ccs_copy (migrate, &ww,NULL , obj->mm.pages->sgl, - obj->cache_level, + obj->pat_index, true, &rq);if (rq && !err) {if (i915_request_wait (rq,0 ,HZ ) <0 ) {@@ -351,7 +351,7 @@ static int clear(struct intel_migrate *migrate, err =intel_migrate_ccs_copy (migrate, &ww,NULL , obj->mm.pages->sgl, - obj->cache_level, + obj->pat_index, false, &rq);if (rq && !err) {if (i915_request_wait (rq,0 ,HZ ) <0 ) {@@ -414,9 +414,9 @@ static int __migrate_copy(struct intel_migrate *migrate, struct i915_request **out) {return intel_migrate_copy (migrate, ww,NULL , - src->mm.pages->sgl, src->cache_level, + src->mm.pages->sgl, src->pat_index,i915_gem_object_is_lmem (src), - dst->mm.pages->sgl, dst->cache_level, + dst->mm.pages->sgl, dst->pat_index,i915_gem_object_is_lmem (dst), out);}drivers/gpu/drm/i915/gt/selftest_migrate.c +2
-2@@ -428,9 +428,9 @@ static int __global_copy(struct intel_migrate *migrate, struct i915_request **out) {return intel_context_migrate_copy (migrate->context,NULL , - src->mm.pages->sgl, src->cache_level, + src->mm.pages->sgl, src->pat_index,i915_gem_object_is_lmem (src), - dst->mm.pages->sgl, dst->cache_level, + dst->mm.pages->sgl, dst->pat_index,i915_gem_object_is_lmem (dst), out);}drivers/gpu/drm/i915/gt/selftest_migrate.c +1
-1@@ -455,7 +455,7 @@ static int __migrate_clear(struct intel_migrate *migrate, {return intel_migrate_clear (migrate, ww,NULL , obj->mm.pages->sgl, - obj->cache_level, + obj->pat_index,i915_gem_object_is_lmem (obj), value, out);}drivers/gpu/drm/i915/gt/selftest_migrate.c +1
-1@@ -468,7 +468,7 @@ static int __global_clear(struct intel_migrate *migrate, {return intel_context_migrate_clear (migrate->context,NULL , obj->mm.pages->sgl, - obj->cache_level, + obj->pat_index,i915_gem_object_is_lmem (obj), value, out);}drivers/gpu/drm/i915/gt/selftest_migrate.c +13
-10@@ -648,7 +648,7 @@ static int live_emit_pte_full_ring(void *arg)*/ pr_info ("%s emite_pte ring space=%u\n" , __func__, rq->ring->space); it =sg_sgt (obj->mm.pages->sgl); - len =emit_pte (rq, &it, obj->cache_level, false,0 ,CHUNK_SZ ); + len =emit_pte (rq, &it, obj->pat_index, false,0 ,CHUNK_SZ );if (!len) { err = -EINVAL ; goto out_rq;@@ -844,7 +844,7 @@ static int wrap_ktime_compare(const void *A, const void *B)static int __perf_clear_blt (struct intel_context *ce,struct scatterlist *sg, -enum i915_cache_level cache_level, +unsigned int pat_index,bool is_lmem,size_t sz) {@@ -858,7 +858,7 @@ static int __perf_clear_blt(struct intel_context *ce, t0 =ktime_get (); - err =intel_context_migrate_clear (ce,NULL , sg, cache_level, + err =intel_context_migrate_clear (ce,NULL , sg, pat_index, is_lmem,0 , &rq);if (rq) {if (i915_request_wait (rq,0 ,MAX_SCHEDULE_TIMEOUT ) <0 )@@ -904,7 +904,8 @@ static int perf_clear_blt(void *arg) err =__perf_clear_blt (gt->migrate.context, dst->mm.pages->sgl, -I915_CACHE_NONE , +i915_gem_get_pat_index (gt->i915, +I915_CACHE_NONE ),i915_gem_object_is_lmem (dst), sizes[i]);@@ -919,10 +920,10 @@ static int perf_clear_blt(void *arg)static int __perf_copy_blt (struct intel_context *ce,struct scatterlist *src, -enum i915_cache_level src_cache_level, +unsigned int src_pat_index,bool src_is_lmem,struct scatterlist *dst, -enum i915_cache_level dst_cache_level, +unsigned int dst_pat_index,bool dst_is_lmem,size_t sz) {@@ -937,9 +938,9 @@ static int __perf_copy_blt(struct intel_context *ce, t0 =ktime_get (); err =intel_context_migrate_copy (ce,NULL , - src, src_cache_level, + src, src_pat_index, src_is_lmem, - dst, dst_cache_level, + dst, dst_pat_index, dst_is_lmem, &rq);if (rq) {@@ -994,10 +995,12 @@ static int perf_copy_blt(void *arg) err =__perf_copy_blt (gt->migrate.context, src->mm.pages->sgl, -I915_CACHE_NONE , +i915_gem_get_pat_index (gt->i915, +I915_CACHE_NONE ),i915_gem_object_is_lmem (src), dst->mm.pages->sgl, -I915_CACHE_NONE , +i915_gem_get_pat_index (gt->i915, +I915_CACHE_NONE ),i915_gem_object_is_lmem (dst), sz);drivers/gpu/drm/i915/gt/selftest_reset.c +6
-2@@ -86,7 +86,9 @@ __igt_reset_stolen(struct intel_gt *gt, ggtt->vm.insert_page (&ggtt->vm, dma, ggtt->error_capture.start, -I915_CACHE_NONE ,0 ); +i915_gem_get_pat_index (gt->i915, +I915_CACHE_NONE ), +0 );mb (); s = io_mapping_map_wc(&ggtt->iomap,@@ -127,7 +129,9 @@ __igt_reset_stolen(struct intel_gt *gt, ggtt->vm.insert_page (&ggtt->vm, dma, ggtt->error_capture.start, -I915_CACHE_NONE ,0 ); +i915_gem_get_pat_index (gt->i915, +I915_CACHE_NONE ), +0 );mb (); s = io_mapping_map_wc(&ggtt->iomap,drivers/gpu/drm/i915/gt/selftest_timeline.c +1
-1@@ -836,7 +836,7 @@ static int setup_watcher(struct hwsp_watcher *w,struct intel_gt *gt, return PTR_ERR(obj);/* keep the same cache settings as timeline */ -i915_gem_object_set_cache_coherency (obj, tl->hwsp_ggtt->obj->cache_level); +i915_gem_object_set_pat_index (obj, tl->hwsp_ggtt->obj->pat_index); w->map = i915_gem_object_pin_map_unlocked(obj,page_unmask_bits(tl->hwsp_ggtt->obj->mm.mapping)); if (IS_ERR(w->map)) {drivers/gpu/drm/i915/gt/selftest_tlb.c +3
-1@@ -36,6 +36,8 @@ pte_tlbinv(struct intel_context *ce, u64 length, struct rnd_state *prng) { +const unsigned int pat_index = +i915_gem_get_pat_index (ce->vm->i915,I915_CACHE_NONE );struct drm_i915_gem_object *batch;struct drm_mm_node vb_node;struct i915_request *rq;@@ -155,7 +157,7 @@ pte_tlbinv(struct intel_context *ce,/* Flip the PTE between A and B */ if (i915_gem_object_is_lmem (vb->obj)) pte_flags |=PTE_LM ; - ce->vm->insert_entries (ce->vm, &vb_res,0 , pte_flags); + ce->vm->insert_entries (ce->vm, &vb_res, pat_index, pte_flags);/* Flush the PTE update to concurrent HW */ tlbinv (ce->vm, addr & -length, length);drivers/gpu/drm/i915/gt/uc/intel_uc_fw.c +8
-2@@ -890,9 +890,15 @@ static void uc_fw_bind_ggtt(struct intel_uc_fw *uc_fw) pte_flags |=PTE_LM ;if (ggtt->vm.raw_insert_entries) - ggtt->vm.raw_insert_entries (&ggtt->vm, dummy, I915_CACHE_NONE, pte_flags); + ggtt->vm.raw_insert_entries (&ggtt->vm, dummy, +i915_gem_get_pat_index (ggtt->vm.i915, +I915_CACHE_NONE ), + pte_flags); else - ggtt->vm.insert_entries (&ggtt->vm, dummy,I915_CACHE_NONE , pte_flags); + ggtt->vm.insert_entries (&ggtt->vm, dummy, +i915_gem_get_pat_index (ggtt->vm.i915, +I915_CACHE_NONE ), + pte_flags); }static void
uc_fw_unbind_ggtt(struct intel_uc_fw *uc_fw)drivers/gpu/drm/i915/i915_debugfs.c +6
-8@@ -139,21 +139,56 @@ static const char *stringify_vma_type(const struct i915_vma *vma)return "ppgtt" ; } -static const char *i915_cache_level_str (struct drm_i915_private *i915, int type) -{ -switch (type) { -case I915_CACHE_NONE :return " uncached" ; -case I915_CACHE_LLC :return HAS_LLC (i915) ?" LLC" :" snooped" ; -case I915_CACHE_L3_LLC :return " L3+LLC" ; -case I915_CACHE_WT :return " WT" ; -default :return "" ; +static const char *i915_cache_level_str (struct drm_i915_gem_object *obj) +{ +struct drm_i915_private *i915 =obj_to_i915 (obj); + +if (IS_METEORLAKE (i915)) { +switch (obj->pat_index) {
Quoted 8 lines
Quoted 6 lines
drivers/gpu/drm/i915/i915_debugfs.c static const char *stringify_vma_type(const struct i915_vma *vma) · line 153
Quoted 6 lines
drivers/gpu/drm/i915/i915_debugfs.c +6
-0+case 0 :return " WB" ; +case 1 :return " WT" ; +case 2 :return " UC" ; +case 3 :return " WB (1-Way Coh)" ; +case 4 :return " WB (2-Way Coh)" ; +default :return " not defined" ;Is not defined possible?
Not possible because there is a sanity check on pat_index, but I think I got a compilation warning without default, then due to the treat-warning-as-error flag, I actually failed the compilation.
Quoted 1 line
Also, it may be nicer to handle the leading space in the caller.
will check...
Quoted 23 lines
Quoted 14 lines
Quoted 14 lines
drivers/gpu/drm/i915/i915_debugfs.c +14
-0+ } + }else if (IS_PONTEVECCHIO (i915)) { +switch (obj->pat_index) { +case 0 :return " UC" ; +case 1 :return " WC" ; +case 2 :return " WT" ; +case 3 :return " WB" ; +case 4 :return " WT (CLOS1)" ; +case 5 :return " WB (CLOS1)" ; +case 6 :return " WT (CLOS2)" ; +case 7 :return " WT (CLOS2)" ; +default :return " not defined" ; + } + }else if (GRAPHICS_VER (i915) >=12 ) {Quoted 8 lines
drivers/gpu/drm/i915/i915_debugfs.c static const char *stringify_vma_type(const struct i915_vma *vma) · line 175
Quoted 8 lines
drivers/gpu/drm/i915/i915_debugfs.c +8
-0+switch (obj->pat_index) { +case 0 :return " WB" ; +case 1 :return " WC" ; +case 2 :return " WT" ; +case 3 :return " UC" ; +default :return " not defined" ; + } + }else {
Quoted 2 lines
Is this correct if a legacy platform used the set pat extension? I don't see that it is disallowed.
That is allowed. For legacy platforms pat_index is the same as cache_level.
Quoted 4 lines
Would it simplify things to add a reverse table to device info, so like
cachelevel_to_pat, just forpat_indexto names? I guess it depends what names PRMs use for PATs on legacy platforms. Is it consistend with the above UC/WC/WB/... or with the below names.
I was trying to avoid adding tables, but couldn't get away with cachelevel_to_pat translation. Well, I guess having constant tables defined locally here might not be too bad...
Quoted 4 lines
Quoted 2 lines
drivers/gpu/drm/i915/i915_debugfs.c static const char *stringify_vma_type(const struct i915_vma *vma) · line 177
Quoted 2 lines
drivers/gpu/drm/i915/i915_debugfs.c +2
-0+if (i915_gem_object_has_cache_level (obj,I915_CACHE_NONE )) +return " uncached" ;UC for consistency?
Will update.
Quoted 7 lines
Quoted 5 lines
drivers/gpu/drm/i915/i915_debugfs.c static const char *stringify_vma_type(const struct i915_vma *vma) · line 181
Quoted 5 lines
drivers/gpu/drm/i915/i915_debugfs.c +4
-0+else if (i915_gem_object_has_cache_level (obj,I915_CACHE_LLC )) +return HAS_LLC (i915) ?" LLC" :" snooped" ; +else if (i915_gem_object_has_cache_level (obj,I915_CACHE_L3_LLC )) +return " L3+LLC" ;Is this correct if !HAS_LLC?
Nice catch. Will update.
Quoted 8 lines
Quoted 4 lines
drivers/gpu/drm/i915/i915_debugfs.c static const char *stringify_vma_type(const struct i915_vma *vma) · line 185
Quoted 4 lines
drivers/gpu/drm/i915/i915_debugfs.c +4
-0+else if (i915_gem_object_has_cache_level (obj,I915_CACHE_WT )) +return " WT" ; +else +return " not defined" ;Current code prints nothing for the default switch statement.
But is this even reachable or should it be
MISSING_CASEwarning?
hmm... I guess it's possible, will re-examine the code.
Quoted 37 lines
Quoted 29 lines
Quoted 29 lines
drivers/gpu/drm/i915/i915_debugfs.c +0
-1} }void i915_debugfs_describe_obj (struct seq_file *m,struct drm_i915_gem_object *obj) { -struct drm_i915_private *dev_priv =to_i915 (obj->base.dev);struct i915_vma *vma;int pin_count = 0;drivers/gpu/drm/i915/i915_debugfs.c +1
-1@@ -165,7 +200,7 @@ i915_debugfs_describe_obj(struct seq_file *m, struct drm_i915_gem_object *obj) obj->base.size / 1024, obj->read_domains, obj->write_domain, -i915_cache_level_str (dev_priv, obj->cache_level), +i915_cache_level_str (obj), obj->mm.dirty ?" dirty" :"" , obj->mm.madv == I915_MADV_DONTNEED ? " purgeable" : "");if (obj->base.name)drivers/gpu/drm/i915/i915_gem.c +0
-1@@ -420,8 +420,12 @@ i915_gem_gtt_pread(struct drm_i915_gem_object *obj, page_length = remain < page_length ? remain : page_length;if (drm_mm_node_allocated (&node)) { ggtt->vm.insert_page(&ggtt->vm, - i915_gem_object_get_dma_address(obj, offsetQuoted 7 lines
drivers/gpu/drm/i915/i915_gem.c i915_gem_gtt_pread(struct drm_i915_gem_object *obj, · line 427
Quoted 7 lines
drivers/gpu/drm/i915/i915_gem.c +5
-1>> PAGE_SHIFT), - node.start, I915_CACHE_NONE, 0); +i915_gem_object_get_dma_address (obj, + offset >>PAGE_SHIFT ), + node.start, +i915_gem_get_pat_index (i915, + I915_CACHE_NONE),
Quoted 5 lines
For the callsites which use const levels you could at least do something like
i915->pat_cache_none, or I know the not very popular static inlinei915_gem_get_pat_indexso it can be evaluated at runtime. Not sure really, throwing out ideas which may be invalid if a more elegant refactoring is possible.
You meant defining each cache_level in the i915 structure in stead of having a constant table to do the translation?
-Fei
Quoted 294 lines
Regards,
Tvrtko
Quoted 290 lines
Quoted 289 lines
drivers/gpu/drm/i915/i915_gem.c +1
-0+ 0); }else { page_base += offset &PAGE_MASK ;}drivers/gpu/drm/i915/i915_gem.c +6
-2@@ -598,8 +602,12 @@ i915_gem_gtt_pwrite_fast(struct drm_i915_gem_object *obj,/* flush the write before we modify the GGTT */ intel_gt_flush_ggtt_writes (ggtt->vm.gt); ggtt->vm.insert_page(&ggtt->vm, -i915_gem_object_get_dma_address (obj, offset >>PAGE_SHIFT ), - node.start,I915_CACHE_NONE ,0 ); +i915_gem_object_get_dma_address (obj, + offset >>PAGE_SHIFT ), + node.start, +i915_gem_get_pat_index (i915, +I915_CACHE_NONE ), + 0);wmb ();/* flush modifications to the GGTT (insert_page) */ } else { page_base += offset & PAGE_MASK;drivers/gpu/drm/i915/i915_gpu_error.c +6
-2@@ -1117,10 +1117,14 @@ i915_vma_coredump_create(const struct intel_gt *gt,mutex_lock (&ggtt->error_mutex);if (ggtt->vm.raw_insert_page) ggtt->vm.raw_insert_page (&ggtt->vm, dma, slot, -I915_CACHE_NONE ,0 ); +i915_gem_get_pat_index (gt->i915, +I915_CACHE_NONE ), +0 );else ggtt->vm.insert_page (&ggtt->vm, dma, slot, -I915_CACHE_NONE ,0 ); +i915_gem_get_pat_index (gt->i915, +I915_CACHE_NONE ), +0 );mb (); s =io_mapping_map_wc (&ggtt->iomap, slot,PAGE_SIZE );drivers/gpu/drm/i915/i915_vma.c +2
-2@@ -315,7 +315,7 @@ struct i915_vma_work {struct i915_vma_resource *vma_res;struct drm_i915_gem_object *obj;struct i915_sw_dma_fence_cb cb; -enum i915_cache_level cache_level; +unsigned int pat_index;unsigned int flags; };@@ -334,7 +334,7 @@ static void __vma_bind(struct dma_fence_work *work)return ; vma_res->ops->bind_vma (vma_res->vm, &vw->stash, - vma_res, vw->cache_level, vw->flags); + vma_res, vw->pat_index, vw->flags); }static void __vma_release (struct dma_fence_work *work)@@ -426,7 +426,7 @@ i915_vma_resource_init_from_vma(struct i915_vma_resource *vma_res,/** * i915_vma_bind - Sets up PTEs for an VMA in it's corresponding address space.
* @vma: VMA to map
- * @cache_level: mapping cache level
+ * @pat_index: PAT index to set in PTE
* @flags: flags like global or local mapping
* @work: preallocated worker for allocating and binding the PTE
* @vma_res: pointer to a preallocated vma resource. The resource is
eitherdrivers/gpu/drm/i915/i915_vma.c +0
-0@@ -437,7 +437,7 @@ i915_vma_resource_init_from_vma(structi915_vma_resource *vma_res, * Note that DMA addresses are also the only part of the SG table wecare about.
*/
int i915_vma_bind(structi915_vma*vma,drivers/gpu/drm/i915/i915_vma.c +5
-5-enum i915_cache_level cache_level, +unsigned int pat_index, u32 flags, struct i915_vma_work *work, struct i915_vma_resource *vma_res)@@ -507,7 +507,7 @@ int i915_vma_bind(struct i915_vma *vma,struct dma_fence *prev; work->vma_res =i915_vma_resource_get (vma->resource); - work->cache_level = cache_level; + work->pat_index = pat_index; work->flags = bind_flags;/* @@ -537,7 +537,7 @@ int i915_vma_bind(struct i915_vma *vma,return ret; } - vma->ops->bind_vma (vma->vm,NULL , vma->resource, cache_level, + vma->ops->bind_vma (vma->vm,NULL , vma->resource, pat_index, bind_flags); }@@ -814,7 +814,7 @@ i915_vma_insert(struct i915_vma *vma, structi915_gem_ww_ctx *ww, color =0 ;if (i915_vm_has_cache_coloring (vma->vm)) - color = vma->obj->cache_level; + color = vma->obj->pat_index;if (flags &PIN_OFFSET_FIXED ) {u64 offset = flags &PIN_OFFSET_MASK ;@@ -1518,7 +1518,7 @@ int i915_vma_pin_ww(struct i915_vma *vma, struct i915_gem_ww_ctx *ww,GEM_BUG_ON (!vma->pages); err =i915_vma_bind (vma, - vma->obj->cache_level, + vma->obj->pat_index, flags, work, vma_res); vma_res =NULL ;if (err)drivers/gpu/drm/i915/i915_vma.h +1
-1@@ -250,7 +250,7 @@ i915_vma_compare(struct i915_vma *vma,struct i915_vma_work *i915_vma_work (void );int i915_vma_bind (struct i915_vma *vma, -enum i915_cache_level cache_level, +unsigned int pat_index,u32 flags,struct i915_vma_work *work,struct i915_vma_resource *vma_res);drivers/gpu/drm/i915/i915_vma_types.h +0
-2@@ -32,8 +32,6 @@#include "gem/i915_gem_object_types.h" -enum i915_cache_level ; -/** * DOC: Global GTT views * drivers/gpu/drm/i915/selftests/i915_gem.c +4
-1@@ -57,7 +57,10 @@ static void trash_stolen(struct drm_i915_private *i915)u32 __iomem *s;int x; - ggtt->vm.insert_page (&ggtt->vm, dma, slot,I915_CACHE_NONE ,0 ); + ggtt->vm.insert_page (&ggtt->vm, dma, slot, +i915_gem_get_pat_index (i915, +I915_CACHE_NONE ), +0 ); s =io_mapping_map_atomic_wc (&ggtt->iomap, slot);for (x =0 ; x <PAGE_SIZE /sizeof (u32); x++) {drivers/gpu/drm/i915/selftests/i915_gem_evict.c +2
-2@@ -245,7 +245,7 @@ static int igt_evict_for_cache_color(void *arg)struct drm_mm_node target = { .start =I915_GTT_PAGE_SIZE *2 , .size =I915_GTT_PAGE_SIZE , - .color =I915_CACHE_LLC , + .color =i915_gem_get_pat_index (gt->i915,I915_CACHE_LLC ), };struct drm_i915_gem_object *obj;struct i915_vma *vma;@@ -308,7 +308,7 @@ static int igt_evict_for_cache_color(void *arg)/* Attempt to remove the first *pinned* vma, by removing the (empty) * neighbour -- this should fail. */ - target.color =I915_CACHE_L3_LLC ; + target.color =i915_gem_get_pat_index (gt->i915,I915_CACHE_L3_LLC );mutex_lock (&ggtt->vm.mutex); err =i915_gem_evict_for_node (&ggtt->vm,NULL , &target,0 );drivers/gpu/drm/i915/selftests/i915_gem_gtt.c +9
-4@@ -135,7 +135,7 @@ fake_dma_object(struct drm_i915_private *i915, u64 size) obj->write_domain = I915_GEM_DOMAIN_CPU; obj->read_domains =I915_GEM_DOMAIN_CPU ; - obj->cache_level =I915_CACHE_NONE ; + obj->pat_index =i915_gem_get_pat_index (i915,I915_CACHE_NONE );/* Preallocate the "backing storage" */ if (i915_gem_object_pin_pages_unlocked (obj))@@ -359,7 +359,9 @@ static int lowlevel_hole(struct i915_address_space *vm,with_intel_runtime_pm (vm->gt->uncore->rpm, wakeref) vm->insert_entries(vm, mock_vma_res, -I915_CACHE_NONE ,0 ); +i915_gem_get_pat_index (vm->i915, +I915_CACHE_NONE ), +0 ); } count = n;@@ -1377,7 +1379,10 @@ static int igt_ggtt_page(void *arg) ggtt->vm.insert_page (&ggtt->vm,i915_gem_object_get_dma_address (obj,0 ), - offset,I915_CACHE_NONE ,0 ); + offset, +i915_gem_get_pat_index (i915, +I915_CACHE_NONE ), +0 ); } order =i915_random_order (count, &prng);@@ -1510,7 +1515,7 @@ static int reserve_gtt_with_resource(struct i915_vma *vma, u64 offset) mutex_lock(&vm->mutex); err =i915_gem_gtt_reserve (vm,NULL , &vma->node, obj->base.size, offset, - obj->cache_level, + obj->pat_index,0 );if (!err) {i915_vma_resource_init_from_vma(vma_res, vma);drivers/gpu/drm/i915/selftests/i915_gem_gtt.c +1
-1@@ -1690,7 +1695,7 @@ static int insert_gtt_with_resource(struct i915_vma *vma) mutex_lock(&vm->mutex); err =i915_gem_gtt_insert (vm,NULL , &vma->node, obj->base.size,0 , - obj->cache_level,0 , vm->total,0 ); + obj->pat_index,0 , vm->total,0 );if (!err) {i915_vma_resource_init_from_vma (vma_res, vma); vma->resource = vma_res;drivers/gpu/drm/i915/selftests/intel_memory_region.c +0
-0@@ -1070,7 +1070,9 @@ static int igt_lmem_write_cpu(void *arg)/* Put the pages into a known state -- from the gpu for added fun */
intel_engine_pm_get(engine); err = intel_context_migrate_clear(engine->gt->migrate.context,NULL,
drivers/gpu/drm/i915/selftests/intel_memory_region.c +3
-1- obj->mm.pages->sgl,I915_CACHE_NONE , + obj->mm.pages->sgl, +i915_gem_get_pat_index (i915, +I915_CACHE_NONE ), true, 0xdeadbeaf, &rq);if (rq) { dma_resv_add_fence(obj->base.resv, &rq->fence,drivers/gpu/drm/i915/selftests/mock_gtt.c +4
-4@@ -27,21 +27,21 @@static void mock_insert_page (struct i915_address_space *vm,dma_addr_t addr,u64 offset, -enum i915_cache_level level, +unsigned int pat_index,u32 flags) { }static void mock_insert_entries (struct i915_address_space *vm,struct i915_vma_resource *vma_res, -enum i915_cache_level level,u32 flags) +unsigned int pat_index,u32 flags) { }static void mock_bind_ppgtt (struct i915_address_space *vm,struct i915_vm_pt_stash *stash,struct i915_vma_resource *vma_res, -enum i915_cache_level cache_level, +unsigned int pat_index,u32 flags) {GEM_BUG_ON (flags &I915_VMA_GLOBAL_BIND );@@ -94,7 +94,7 @@ struct i915_ppgtt *mock_ppgtt(struct drm_i915_private *i915, const char *name) static void mock_bind_ggtt(struct i915_address_space *vm, struct i915_vm_pt_stash *stash, struct i915_vma_resource *vma_res, - enum i915_cache_level cache_level, + unsigned int pat_index, u32 flags) {}
Re: [Intel-gfx] [PATCH 7/8] drm/i915: use pat_index instead of cache_level
In reply to [email protected]Quoted 25 lines · On 20/04/2023 00:00, [email protected] wrote:
From: Fei Yang <[email protected]>
Currently the KMD is using enum
i915_cache_levelto set caching policy for buffer objects. This is flaky because the PAT index which really controls the caching behavior in PTE has far more levels than what's defined in the enum. In addition, the PAT index is platform dependent, having to translate betweeni915_cache_leveland PAT index is not reliable, and makes the code more complicated.Quoted 1 line
From UMD's perspective there is also a necessity to set caching policy for
performance fine tuning. It's much easier for the UMD to directly use PAT index because the behavior of each PAT index is clearly defined in Bspec. Having the abstracted
i915_cache_levelsitting in between would only cause more ambiguity.For these reasons this patch replaces
i915_cache_levelwith PAT index. Also note, thecache_levelis not completely removed yet, because the KMD still has the need of creating buffer objects with simple cache settings such as cached, uncached, or writethrough. For such simple cases, usingcache_levelwould help simplify the code.Cc Chris Wilson <[email protected]>Cc Matt Roper <[email protected]>Signed-off-by Fei Yang <[email protected]>Reviewed-by Andi Shyti <[email protected]>
[snip]
Quoted 25 lines
drivers/gpu/drm/i915/gem/i915_gem_domain.c +4
-11@@ -306,20 +304,13 @@ int i915_gem_get_caching_ioctl(struct drm_device *dev, void *data, goto out; } -switch (obj->cache_level) { -case I915_CACHE_LLC : -case I915_CACHE_L3_LLC : +if (i915_gem_object_has_cache_level (obj,I915_CACHE_LLC ) || +i915_gem_object_has_cache_level (obj,I915_CACHE_L3_LLC )) args->caching =I915_CACHING_CACHED ; -break ; - -case I915_CACHE_WT : +else if (i915_gem_object_has_cache_level (obj,I915_CACHE_WT )) args->caching =I915_CACHING_DISPLAY ; -break ; - -default : +else args->caching =I915_CACHING_NONE ; -break ; - } out:rcu_read_unlock ();return err;
[snip]
Quoted 9 lines
drivers/gpu/drm/i915/gem/i915_gem_object.c +9
-0+bool i915_gem_object_has_cache_level (const struct drm_i915_gem_object *obj, +enum i915_cache_level lvl) +{ +/* +* cache_level == I915_CACHE_INVAL indicates the UMD's have set the +* caching policy through pat_index, in which case the KMD should +* leave the coherency to be managed by user space, simply return +* true here. +*/
drivers/gpu/drm/i915/gem/i915_gem_object.c unsigned int i915_gem_get_pat_index(struct drm_i915_private *i915, · line 75
Quoted 10 lines
drivers/gpu/drm/i915/gem/i915_gem_object.c +10
-0+if (obj->cache_level ==I915_CACHE_INVAL ) +return true; + +/* +* Otherwise the pat_index should have been converted from cache_level +* so that the following comparison is valid. +*/ +return obj->pat_index ==i915_gem_get_pat_index (obj_to_i915 (obj), lvl); +} +
Isn't i915_gem_get_caching_ioctl always going to report I915_CACHING_CACHED if any PAT index has been set?
Not sure if that is okay or not, or if it only needs mentioning in the commit, I am still reading through it all.
Regards,
Tvrtko
Re: [Intel-gfx] [PATCH 7/8] drm/i915: use pat_index instead of cache_level
In reply to [email protected]Quoted 25 lines · On 20/04/2023 00:00, [email protected] wrote:
From: Fei Yang <[email protected]>
Currently the KMD is using enum
i915_cache_levelto set caching policy for buffer objects. This is flaky because the PAT index which really controls the caching behavior in PTE has far more levels than what's defined in the enum. In addition, the PAT index is platform dependent, having to translate betweeni915_cache_leveland PAT index is not reliable, and makes the code more complicated.Quoted 1 line
From UMD's perspective there is also a necessity to set caching policy for
performance fine tuning. It's much easier for the UMD to directly use PAT index because the behavior of each PAT index is clearly defined in Bspec. Having the abstracted
i915_cache_levelsitting in between would only cause more ambiguity.For these reasons this patch replaces
i915_cache_levelwith PAT index. Also note, thecache_levelis not completely removed yet, because the KMD still has the need of creating buffer objects with simple cache settings such as cached, uncached, or writethrough. For such simple cases, usingcache_levelwould help simplify the code.Cc Chris Wilson <[email protected]>Cc Matt Roper <[email protected]>Signed-off-by Fei Yang <[email protected]>Reviewed-by Andi Shyti <[email protected]>
[snip]
drivers/gpu/drm/i915/gem/i915_gem_domain.c int i915_gem_object_set_cache_level(struct drm_i915_gem_object *obj, · line 271
Quoted 9 lines
bool
i915_gem_cpu_write_needs_clflush(struct drm_i915_gem_object *obj)drivers/gpu/drm/i915/gem/i915_gem_domain.c +1
-1@@ -267,7 +267,7 @@ int i915_gem_object_set_cache_level(struct drm_i915_gem_object *obj, {int ret; -if (obj->cache_level == cache_level) +if (i915_gem_object_has_cache_level (obj, cache_level))return 0 ;
When userspace calls i915_gem_set_caching_ioctl after having set the PAT index explicitly this will make it silently succeed regardless of the cache level passed in, no? Because of:
+bool i915_gem_object_has_cache_level(const struct drm_i915_gem_object *obj,
+ enum i915_cache_level lvl)
+{
+ /*
+ * cache_level == I915_CACHE_INVAL indicates the UMD's have set the
+ * caching policy through pat_index, in which case the KMD should
+ * leave the coherency to be managed by user space, simply return
+ * true here.
+ */
+ if (obj->cache_level == I915_CACHE_INVAL)
+ return true;
I think we need to let it know it is doing it wrong with an error.
Regards,
Tvrtko
RE: [Intel-gfx] [PATCH 7/8] drm/i915: use pat_index instead of cache_level
Quoted 35 lines · On 20/04/2023 00:00, [email protected] wrote:
Quoted 25 lines
From: Fei Yang <[email protected]>
Currently the KMD is using enum
i915_cache_levelto set caching policy for buffer objects. This is flaky because the PAT index which really controls the caching behavior in PTE has far more levels than what's defined in the enum. In addition, the PAT index is platform dependent, having to translate betweeni915_cache_leveland PAT index is not reliable, and makes the code more complicated.From UMD's perspective there is also a necessity to set caching policy for performance fine tuning. It's much easier for the UMD to directly use PAT index because the behavior of each PAT index is clearly defined in Bspec. Having the abstracted
i915_cache_levelsitting in between would only cause more ambiguity.For these reasons this patch replaces
i915_cache_levelwith PAT index. Also note, thecache_levelis not completely removed yet, because the KMD still has the need of creating buffer objects with simple cache settings such as cached, uncached, or writethrough. For such simple cases, usingcache_levelwould help simplify the code.Cc Chris Wilson <[email protected]>Cc Matt Roper <[email protected]>Signed-off-by Fei Yang <[email protected]>Reviewed-by Andi Shyti <[email protected]>[snip]
Quoted 6 lines
bool i915_gem_cpu_write_needs_clflush(struct
drm_i915_gem_object
*obj) @@ -267,7 +267,7 @@ int i915_gem_object_set_cache_level(structdrm_i915_gem_object*obj,{ int ret;
Quoted 5 lines
drivers/gpu/drm/i915/gem/i915_gem_domain.c int i915_gem_object_set_cache_level(struct drm_i915_gem_object *obj, · line 271
Quoted 3 lines
drivers/gpu/drm/i915/gem/i915_gem_domain.c +1
-1-if (obj->cache_level == cache_level) +if (i915_gem_object_has_cache_level (obj, cache_level))return 0 ;When userspace calls
i915_gem_set_caching_ioctl
We are ending the support for set_caching_ioctl.
Quoted 2 lines
after having set the PAT index explicitly this will make it silently succeed regardless of the cache level passed in, no? Because of:
Yes, that's the point. For objects created by userspace with PAT index set, KMD is not supposed to touch the setting.
Quoted 12 lines
drivers/gpu/drm/i915/gem/i915_gem_object.c +11
-0+bool i915_gem_object_has_cache_level (const struct drm_i915_gem_object *obj, +enum i915_cache_level lvl) +{ +/* +* cache_level == I915_CACHE_INVAL indicates the UMD's have set the +* caching policy through pat_index, in which case the KMD should +* leave the coherency to be managed by user space, simply return +* true here. +*/ +if (obj->cache_level ==I915_CACHE_INVAL ) +return true;
Quoted 1 line
I think we need to let it know it is doing it wrong with an error.
This is not an error, by design userspace should know exactly what it's doing.
-Fei
Quoted 3 lines
Regards,
Tvrtko
Re: [Intel-gfx] [PATCH 7/8] drm/i915: use pat_index instead of cache_level
Quoted 35 lines · Fei
Quoted 35 lines · On 20/04/2023 00:00, [email protected] wrote:
Quoted 25 lines
From: Fei Yang <[email protected]>
Currently the KMD is using enum
i915_cache_levelto set caching policy for buffer objects. This is flaky because the PAT index which really controls the caching behavior in PTE has far more levels than what's defined in the enum. In addition, the PAT index is platform dependent, having to translate betweeni915_cache_leveland PAT index is not reliable, and makes the code more complicated.From UMD's perspective there is also a necessity to set caching policy for
performance fine tuning. It's much easier for the UMD to directly use PAT index because the behavior of each PAT index is clearly defined in Bspec. Having the abstractedi915_cache_levelsitting in between would only cause more ambiguity.For these reasons this patch replaces
i915_cache_levelwith PAT index. Also note, thecache_levelis not completely removed yet, because the KMD still has the need of creating buffer objects with simple cache settings such as cached, uncached, or writethrough. For such simple cases, usingcache_levelwould help simplify the code.Cc Chris Wilson <[email protected]>Cc Matt Roper <[email protected]>Signed-off-by Fei Yang <[email protected]>Reviewed-by Andi Shyti <[email protected]>[snip]
Quoted 6 lines
bool i915_gem_cpu_write_needs_clflush(struct
drm_i915_gem_object
*obj) @@ -267,7 +267,7 @@ int i915_gem_object_set_cache_level(structdrm_i915_gem_object*obj,{ int ret;
Quoted 7 lines
Quoted 5 lines
drivers/gpu/drm/i915/gem/i915_gem_domain.c int i915_gem_object_set_cache_level(struct drm_i915_gem_object *obj, · line 271
Quoted 3 lines
drivers/gpu/drm/i915/gem/i915_gem_domain.c +1
-1-if (obj->cache_level == cache_level) +if (i915_gem_object_has_cache_level (obj, cache_level))return 0 ;When userspace calls
i915_gem_set_caching_ioctlWe are ending the support for
set_caching_ioctl.
Not on all platforms.
Quoted 5 lines
Quoted 2 lines
after having set the PAT index explicitly this will make it silently succeed regardless of the cache level passed in, no? Because of:
Yes, that's the point. For objects created by userspace with PAT index set, KMD is not supposed to touch the setting.
Why would that be a reason to lie to it? What would would be the problem with telling it of the mistake?
Quoted 12 lines
Quoted 12 lines
drivers/gpu/drm/i915/gem/i915_gem_object.c +11
-0+bool i915_gem_object_has_cache_level (const struct drm_i915_gem_object *obj, +enum i915_cache_level lvl) +{ +/* +* cache_level == I915_CACHE_INVAL indicates the UMD's have set the +* caching policy through pat_index, in which case the KMD should +* leave the coherency to be managed by user space, simply return +* true here. +*/ +if (obj->cache_level ==I915_CACHE_INVAL ) +return true;
Quoted 3 lines
Quoted 1 line
I think we need to let it know it is doing it wrong with an error.
This is not an error, by design userspace should know exactly what it's doing.
IMO when return values can be misleading that means the API is not great.
I don't see a good reason to lie to both in kernel callers and to userspace (set_caching).
Regards,
Tvrtko
Re: [Intel-gfx] [PATCH 7/8] drm/i915: use pat_index instead of cache_level
In reply to [email protected]Quoted 21 lines · On 20/04/2023 00:00, [email protected] wrote:
From: Fei Yang <[email protected]>
Currently the KMD is using enum
i915_cache_levelto set caching policy for buffer objects. This is flaky because the PAT index which really controls the caching behavior in PTE has far more levels than what's defined in the enum. In addition, the PAT index is platform dependent, having to translate betweeni915_cache_leveland PAT index is not reliable, and makes the code more complicated.Quoted 1 line
From UMD's perspective there is also a necessity to set caching policy for
performance fine tuning. It's much easier for the UMD to directly use PAT index because the behavior of each PAT index is clearly defined in Bspec. Having the abstracted
i915_cache_levelsitting in between would only cause more ambiguity.For these reasons this patch replaces
i915_cache_levelwith PAT index. Also note, thecache_levelis not completely removed yet, because the KMD still has the need of creating buffer objects with simple cache settings such as cached, uncached, or writethrough. For such simple cases, usingcache_levelwould help simplify the code.
Quoted 4 lines
Cc Chris Wilson <[email protected]>Cc Matt Roper <[email protected]>Signed-off-by Fei Yang <[email protected]>Reviewed-by Andi Shyti <[email protected]>
I think have some ideas no how to perhaps make this simpler, please bear with me.
In my mind get/set caching ioctls need to be failing once explicit pat index has been set by userspace. Or at least not return false information.
And I don't like i915_gem_object_has_cache_level and i915_gem_get_pat_index as a refactoring step.
It also seems that the driver has a need to query the caching mode set regardless of the route (of setting).
So how about this.
Three callers which query the caching mode: use_cpu_reloc, vm_fault_gtt, gpu_write_needs_clflush.
We convert them to be like:
i915_gem_object_has_caching_mode(obj, PAT_UC / PAT_WT / ...);
Then apart from the per platform tables for mapping between cache level to pat index, you add tables which map pat index to caching modes (PAT_UC, etc, naming TBD, just enums or bitmasks also TBD, I haven't looked at the bspec to see how exactly it works).
You would use that table in the i915_gem_object_has_caching_mode helper, called from the above three functions instead of obj->cache_level direct comparison.
I am assuming at least for instance cache_level != I915_CACHE_NONE would be equivalent to i915_gem_object_has_caching_mode(obj, PAT_UC), etc.
Same mapping table could also be used in debugfs (i915_cache_level_str) to universally describe any obj->pat_index, with no need to have anything platform dependend there.
In set caching set you always set obj->pat_index and so low level code can always just use that.
Unless I am missing something (possible) I think like that we end up with no i915_gem_get_pat_index sprinkled around and also no confusing i915_gem_object_has_cache_level.
Obj->pat_index would be a single point of truth, while obj->cache_level is just a legacy field for get/set_caching ioctl - not used in the internal driver flows.
We would need an additional field for storing the boolean of whether userspace had overriden the PAT.
Regards,
Tvrtko
Quoted 1,920 lines
--- drivers/gpu/drm/i915/display/intel_dpt.c | 12 +-- drivers/gpu/drm/i915/gem/i915_gem_domain.c | 27 ++---- .../gpu/drm/i915/gem/i915_gem_execbuffer.c | 10 ++- drivers/gpu/drm/i915/gem/i915_gem_mman.c | 3 +- drivers/gpu/drm/i915/gem/i915_gem_object.c | 52 +++++++++++- drivers/gpu/drm/i915/gem/i915_gem_object.h | 4 + .../gpu/drm/i915/gem/i915_gem_object_types.h | 25 +++++- drivers/gpu/drm/i915/gem/i915_gem_stolen.c | 4 +- drivers/gpu/drm/i915/gem/i915_gem_ttm_move.c | 16 ++-- .../gpu/drm/i915/gem/selftests/huge_pages.c | 2 +- .../drm/i915/gem/selftests/i915_gem_migrate.c | 2 +- .../drm/i915/gem/selftests/i915_gem_mman.c | 2 +- drivers/gpu/drm/i915/gt/gen6_ppgtt.c | 10 ++- drivers/gpu/drm/i915/gt/gen8_ppgtt.c | 71 ++++++++-------- drivers/gpu/drm/i915/gt/gen8_ppgtt.h | 3 +- drivers/gpu/drm/i915/gt/intel_ggtt.c | 82 +++++++++---------- drivers/gpu/drm/i915/gt/intel_gtt.h | 20 ++--- drivers/gpu/drm/i915/gt/intel_migrate.c | 47 ++++++----- drivers/gpu/drm/i915/gt/intel_migrate.h | 13 ++- drivers/gpu/drm/i915/gt/intel_ppgtt.c | 6 +- drivers/gpu/drm/i915/gt/selftest_migrate.c | 47 ++++++----- drivers/gpu/drm/i915/gt/selftest_reset.c | 8 +- drivers/gpu/drm/i915/gt/selftest_timeline.c | 2 +- drivers/gpu/drm/i915/gt/selftest_tlb.c | 4 +- drivers/gpu/drm/i915/gt/uc/intel_uc_fw.c | 10 ++- drivers/gpu/drm/i915/i915_debugfs.c | 55 ++++++++++--- drivers/gpu/drm/i915/i915_gem.c | 16 +++- drivers/gpu/drm/i915/i915_gpu_error.c | 8 +- drivers/gpu/drm/i915/i915_vma.c | 16 ++-- drivers/gpu/drm/i915/i915_vma.h | 2 +- drivers/gpu/drm/i915/i915_vma_types.h | 2 - drivers/gpu/drm/i915/selftests/i915_gem.c | 5 +- .../gpu/drm/i915/selftests/i915_gem_evict.c | 4 +- drivers/gpu/drm/i915/selftests/i915_gem_gtt.c | 15 ++-- .../drm/i915/selftests/intel_memory_region.c | 4 +- drivers/gpu/drm/i915/selftests/mock_gtt.c | 8 +- 36 files changed, 378 insertions(+), 239 deletions(-)drivers/gpu/drm/i915/display/intel_dpt.c +6
-6@@ -43,24 +43,24 @@ static void gen8_set_pte(void __iomem *addr, gen8_pte_t pte)static void dpt_insert_page (struct i915_address_space *vm,dma_addr_t addr,u64 offset, -enum i915_cache_level level, +unsigned int pat_index,u32 flags) {struct i915_dpt *dpt =i915_vm_to_dpt (vm);gen8_pte_t __iomem *base = dpt->iomem;gen8_set_pte (base + offset /I915_GTT_PAGE_SIZE , - vm->pte_encode (addr, level, flags)); + vm->pte_encode (addr, pat_index, flags)); }static void dpt_insert_entries (struct i915_address_space *vm,struct i915_vma_resource *vma_res, -enum i915_cache_level level, +unsigned int pat_index,u32 flags) {struct i915_dpt *dpt =i915_vm_to_dpt (vm);gen8_pte_t __iomem *base = dpt->iomem; -const gen8_pte_t pte_encode = vm->pte_encode (0 , level, flags); +const gen8_pte_t pte_encode = vm->pte_encode (0 , pat_index, flags);struct sgt_iter sgt_iter;dma_addr_t addr;int i;@@ -83,7 +83,7 @@ static void dpt_clear_range(struct i915_address_space *vm,static void dpt_bind_vma (struct i915_address_space *vm,struct i915_vm_pt_stash *stash,struct i915_vma_resource *vma_res, -enum i915_cache_level cache_level, +unsigned int pat_index,u32 flags) {u32 pte_flags;@@ -98,7 +98,7 @@ static void dpt_bind_vma(struct i915_address_space *vm,if (vma_res->bi.lmem) pte_flags |=PTE_LM ; - vm->insert_entries (vm, vma_res, cache_level, pte_flags); + vm->insert_entries (vm, vma_res, pat_index, pte_flags); vma_res->page_sizes_gtt =I915_GTT_PAGE_SIZE ;drivers/gpu/drm/i915/gem/i915_gem_domain.c +9
-18@@ -27,8 +27,8 @@ static bool gpu_write_needs_clflush(struct drm_i915_gem_object *obj)if (IS_DGFX (i915))return false; -return !(obj->cache_level ==I915_CACHE_NONE || - obj->cache_level ==I915_CACHE_WT ); +return !(i915_gem_object_has_cache_level (obj,I915_CACHE_NONE ) || +i915_gem_object_has_cache_level (obj,I915_CACHE_WT )); }bool i915_gem_cpu_write_needs_clflush (struct drm_i915_gem_object *obj)@@ -267,7 +267,7 @@ int i915_gem_object_set_cache_level(struct drm_i915_gem_object *obj, {int ret; -if (obj->cache_level == cache_level) +if (i915_gem_object_has_cache_level (obj, cache_level))return 0 ; ret = i915_gem_object_wait(obj,@@ -278,10 +278,8 @@ int i915_gem_object_set_cache_level(struct drm_i915_gem_object *obj,return ret;/* Always invalidate stale cachelines */ -if (obj->cache_level != cache_level) { -i915_gem_object_set_cache_coherency (obj, cache_level); - obj->cache_dirty = true; - } +i915_gem_object_set_cache_coherency (obj, cache_level); + obj->cache_dirty = true;/* The cache-level will be applied when each vma is rebound. */ return i915_gem_object_unbind(obj,@@ -306,20 +304,13 @@ int i915_gem_get_caching_ioctl(struct drm_device *dev, void *data, goto out; } -switch (obj->cache_level) { -case I915_CACHE_LLC : -case I915_CACHE_L3_LLC : +if (i915_gem_object_has_cache_level (obj,I915_CACHE_LLC ) || +i915_gem_object_has_cache_level (obj,I915_CACHE_L3_LLC )) args->caching =I915_CACHING_CACHED ; -break ; - -case I915_CACHE_WT : +else if (i915_gem_object_has_cache_level (obj,I915_CACHE_WT )) args->caching =I915_CACHING_DISPLAY ; -break ; - -default : +else args->caching =I915_CACHING_NONE ; -break ; - } out:rcu_read_unlock ();return err;drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c +6
-4@@ -642,7 +642,7 @@ static inline int use_cpu_reloc(const struct reloc_cache *cache,return (cache->has_llc || obj->cache_dirty || - obj->cache_level !=I915_CACHE_NONE ); + !i915_gem_object_has_cache_level (obj,I915_CACHE_NONE )); } static int eb_reserve_vma(struct i915_execbuffer *eb,@@ -1323,8 +1323,10 @@ static void *reloc_iomap(struct i915_vma *batch, offset = cache->node.start;if (drm_mm_node_allocated (&cache->node)) { ggtt->vm.insert_page (&ggtt->vm, -i915_gem_object_get_dma_address (obj, page), - offset,I915_CACHE_NONE ,0 ); +i915_gem_object_get_dma_address (obj, page), + offset, +i915_gem_get_pat_index (ggtt->vm.i915,I915_CACHE_NONE ), +0 ); }else { offset += page <<PAGE_SHIFT ; }@@ -1464,7 +1466,7 @@ eb_relocate_entry(struct i915_execbuffer *eb,reloc_cache_unmap (&eb->reloc_cache);mutex_lock (&vma->vm->mutex); err =i915_vma_bind (target->vma, - target->vma->obj->cache_level, + target->vma->obj->pat_index,PIN_GLOBAL ,NULL ,NULL );mutex_unlock (&vma->vm->mutex);reloc_cache_remap (&eb->reloc_cache, ev->vma->obj);drivers/gpu/drm/i915/gem/i915_gem_mman.c +2
-1@@ -383,7 +383,8 @@ static vm_fault_t vm_fault_gtt(struct vm_fault *vmf) }/* Access to snoopable pages through the GTT is incoherent. */ -if (obj->cache_level !=I915_CACHE_NONE && !HAS_LLC (i915)) { +if (!(i915_gem_object_has_cache_level (obj,I915_CACHE_NONE ) || +HAS_LLC (i915))) { ret = -EFAULT ; goto err_unpin; }drivers/gpu/drm/i915/gem/i915_gem_object.c +51
-1@@ -54,6 +54,25 @@ unsigned int i915_gem_get_pat_index(struct drm_i915_private *i915,return INTEL_INFO (i915)->cachelevel_to_pat[level]; } +bool i915_gem_object_has_cache_level (const struct drm_i915_gem_object *obj, +enum i915_cache_level lvl) +{ +/* +* cache_level == I915_CACHE_INVAL indicates the UMD's have set the +* caching policy through pat_index, in which case the KMD should +* leave the coherency to be managed by user space, simply return +* true here. +*/ +if (obj->cache_level ==I915_CACHE_INVAL ) +return true; + +/* +* Otherwise the pat_index should have been converted from cache_level +* so that the following comparison is valid. +*/ +return obj->pat_index ==i915_gem_get_pat_index (obj_to_i915 (obj), lvl); +} +struct drm_i915_gem_object *i915_gem_object_alloc (void ) {struct drm_i915_gem_object *obj;@@ -133,7 +152,7 @@ void i915_gem_object_set_cache_coherency(struct drm_i915_gem_object *obj, {struct drm_i915_private *i915 =to_i915 (obj->base.dev); - obj->cache_level = cache_level; + obj->pat_index =i915_gem_get_pat_index (i915, cache_level);if (cache_level !=I915_CACHE_NONE ) obj->cache_coherent = (I915_BO_CACHE_COHERENT_FOR_READ |@@ -148,6 +167,37 @@ void i915_gem_object_set_cache_coherency(struct drm_i915_gem_object *obj, !IS_DGFX (i915); } +/** +* i915_gem_object_set_pat_index - set PAT index to be used in PTE encode +* @obj: #drm_i915_gem_object +* @pat_index: PAT index +* +* This is a clone of i915_gem_object_set_cache_coherency taking pat index +* instead of cache_level as its second argument. +*/ +void i915_gem_object_set_pat_index (struct drm_i915_gem_object *obj, +unsigned int pat_index) +{ +struct drm_i915_private *i915 =to_i915 (obj->base.dev); + +if (obj->pat_index == pat_index) +return ; + + obj->pat_index = pat_index; + +if (pat_index !=i915_gem_get_pat_index (i915,I915_CACHE_NONE )) + obj->cache_coherent = (I915_BO_CACHE_COHERENT_FOR_READ | +I915_BO_CACHE_COHERENT_FOR_WRITE ); +else if (HAS_LLC (i915)) + obj->cache_coherent =I915_BO_CACHE_COHERENT_FOR_READ ; +else + obj->cache_coherent =0 ; + + obj->cache_dirty = + !(obj->cache_coherent &I915_BO_CACHE_COHERENT_FOR_WRITE ) && + !IS_DGFX (i915); +} +bool i915_gem_object_can_bypass_llc (struct drm_i915_gem_object *obj) {struct drm_i915_private *i915 =to_i915 (obj->base.dev);drivers/gpu/drm/i915/gem/i915_gem_object.h +4
-0@@ -34,6 +34,8 @@ static inline bool i915_gem_object_size_2big(u64 size)unsigned int i915_gem_get_pat_index (struct drm_i915_private *i915,enum i915_cache_level level); +bool i915_gem_object_has_cache_level (const struct drm_i915_gem_object *obj, +enum i915_cache_level lvl);void i915_gem_init__objects (struct drm_i915_private *i915);void i915_objects_module_exit (void );@@ -764,6 +766,8 @@ bool i915_gem_object_has_unknown_state(struct drm_i915_gem_object *obj);void i915_gem_object_set_cache_coherency (struct drm_i915_gem_object *obj,unsigned int cache_level); +void i915_gem_object_set_pat_index (struct drm_i915_gem_object *obj, +unsigned int pat_index);bool i915_gem_object_can_bypass_llc (struct drm_i915_gem_object *obj);void i915_gem_object_flush_if_display (struct drm_i915_gem_object *obj);void i915_gem_object_flush_if_display_locked (struct drm_i915_gem_object *obj);drivers/gpu/drm/i915/gem/i915_gem_object_types.h +22
-3@@ -195,6 +195,7 @@ enum i915_cache_level {*/ I915_CACHE_WT ,I915_MAX_CACHE_LEVEL , + I915_CACHE_INVAL = I915_MAX_CACHE_LEVEL, }; enum i915_map_type {@@ -358,10 +359,28 @@ struct drm_i915_gem_object {#define I915_BO_FLAG_STRUCT_PAGE BIT(0)/* Object backed by struct pages */ #define I915_BO_FLAG_IOMEM BIT(1)/* Object backed by IO memory */ /** -* @cache_level: The desired GTT caching level. +* @pat_index: The desired PAT index. +* +* See hardware specification for valid PAT indices for each platform. +* This field used to contain a value of enum i915_cache_level. It's +* changed to an unsigned int because PAT indices are being used by +* both UMD and KMD for caching policy control after GEN12. +* For backward compatibility, this field will continue to contain +* value of i915_cache_level for pre-GEN12 platforms so that the PTE +* encode functions for these legacy platforms can stay the same. +* In the meantime platform specific tables are created to translate +* i915_cache_level into pat index, for more details check the macros +* defined i915/i915_pci.c, e.g. PVC_CACHELEVEL. +*/ + unsigned int pat_index:6; +/** +* @cache_level: Indicate whether pat_index is set by UMD * -* See enum i915_cache_level for possible values, along with what -* each does. +* This used to hold desired GTT caching level, but is now replaced by +* pat_index. It's kept here for KMD to tell whether the pat_index is +* set by UMD or converted from enum i915_cache_level. +* This field should be 0 by default, but I915_CACHE_INVAL if the +* pat_index is set by UMD. */ unsigned int cache_level:3;/** drivers/gpu/drm/i915/gem/i915_gem_stolen.c +3
-1@@ -565,7 +565,9 @@ static void dbg_poison(struct i915_ggtt *ggtt, ggtt->vm.insert_page (&ggtt->vm, addr, ggtt->error_capture.start, -I915_CACHE_NONE ,0 ); +i915_gem_get_pat_index (ggtt->vm.i915, +I915_CACHE_NONE ), +0 );mb (); s = io_mapping_map_wc(&ggtt->iomap,drivers/gpu/drm/i915/gem/i915_gem_ttm_move.c +9
-7@@ -214,7 +214,8 @@ static struct dma_fence *i915_ttm_accel_move(struct ttm_buffer_object *bo,intel_engine_pm_get (to_gt (i915)->migrate.context->engine); ret =intel_context_migrate_clear (to_gt (i915)->migrate.context, deps, - dst_st->sgl, dst_level, + dst_st->sgl, +i915_gem_get_pat_index (i915, dst_level),i915_ttm_gtt_binds_lmem (dst_mem),0 , &rq); }else {@@ -227,12 +228,13 @@ static struct dma_fence *i915_ttm_accel_move(struct ttm_buffer_object *bo, src_level =i915_ttm_cache_level (i915, bo->resource, src_ttm);intel_engine_pm_get (to_gt (i915)->migrate.context->engine); ret =intel_context_migrate_copy (to_gt (i915)->migrate.context, - deps, src_rsgt->table.sgl, - src_level, -i915_ttm_gtt_binds_lmem (bo->resource), - dst_st->sgl, dst_level, -i915_ttm_gtt_binds_lmem (dst_mem), - &rq); + deps, src_rsgt->table.sgl, +i915_gem_get_pat_index (i915, src_level), +i915_ttm_gtt_binds_lmem (bo->resource), + dst_st->sgl, +i915_gem_get_pat_index (i915, dst_level), +i915_ttm_gtt_binds_lmem (dst_mem), + &rq);i915_refct_sgt_put (src_rsgt); }drivers/gpu/drm/i915/gem/selftests/huge_pages.c +1
-1@@ -354,7 +354,7 @@ fake_huge_pages_object(struct drm_i915_private *i915, u64 size, bool single) obj->write_domain =I915_GEM_DOMAIN_CPU ; obj->read_domains =I915_GEM_DOMAIN_CPU ; - obj->cache_level =I915_CACHE_NONE ; + obj->pat_index =i915_gem_get_pat_index (i915,I915_CACHE_NONE );return obj; }drivers/gpu/drm/i915/gem/selftests/i915_gem_migrate.c +1
-1@@ -219,7 +219,7 @@ static int __igt_lmem_pages_migrate(struct intel_gt *gt,continue ; err =intel_migrate_clear (>->migrate, &ww, deps, - obj->mm.pages->sgl, obj->cache_level, + obj->mm.pages->sgl, obj->pat_index,i915_gem_object_is_lmem (obj),0xdeadbeaf , &rq);if (rq) {drivers/gpu/drm/i915/gem/selftests/i915_gem_mman.c +1
-1@@ -1222,7 +1222,7 @@ static int __igt_mmap_migrate(struct intel_memory_region **placements, } err =intel_context_migrate_clear (to_gt (i915)->migrate.context,NULL , - obj->mm.pages->sgl, obj->cache_level, + obj->mm.pages->sgl, obj->pat_index,i915_gem_object_is_lmem (obj),expand32 (POISON_INUSE ), &rq);i915_gem_object_unpin_pages (obj);drivers/gpu/drm/i915/gt/gen6_ppgtt.c +6
-4@@ -109,7 +109,7 @@ static void gen6_ppgtt_clear_range(struct i915_address_space *vm,static void gen6_ppgtt_insert_entries (struct i915_address_space *vm,struct i915_vma_resource *vma_res, -enum i915_cache_level cache_level, +unsigned int pat_index,u32 flags) {struct i915_ppgtt *ppgtt =i915_vm_to_ppgtt (vm);@@ -117,7 +117,7 @@ static void gen6_ppgtt_insert_entries(struct i915_address_space *vm,unsigned int first_entry = vma_res->start /I915_GTT_PAGE_SIZE ;unsigned int act_pt = first_entry /GEN6_PTES ;unsigned int act_pte = first_entry %GEN6_PTES ; -const u32 pte_encode = vm->pte_encode (0 , cache_level, flags); +const u32 pte_encode = vm->pte_encode (0 , pat_index, flags);struct sgt_dma iter =sgt_dma (vma_res);gen6_pte_t *vaddr;@@ -227,7 +227,9 @@ static int gen6_ppgtt_init_scratch(struct gen6_ppgtt *ppgtt) vm->scratch[0 ]->encode = vm->pte_encode (px_dma (vm->scratch[0 ]), -I915_CACHE_NONE ,PTE_READ_ONLY ); +i915_gem_get_pat_index (vm->i915, +I915_CACHE_NONE ), +PTE_READ_ONLY ); vm->scratch[1 ] = vm->alloc_pt_dma (vm,I915_GTT_PAGE_SIZE_4K );if (IS_ERR (vm->scratch[1 ])) {@@ -278,7 +280,7 @@ static void gen6_ppgtt_cleanup(struct i915_address_space *vm)static void pd_vma_bind (struct i915_address_space *vm,struct i915_vm_pt_stash *stash,struct i915_vma_resource *vma_res, -enum i915_cache_level cache_level, +unsigned int pat_index,u32 unused) {struct i915_ggtt *ggtt =i915_vm_to_ggtt (vm);drivers/gpu/drm/i915/gt/gen8_ppgtt.c +34
-37@@ -56,7 +56,7 @@ static u64 gen8_pte_encode(dma_addr_t addr, }static u64 mtl_pte_encode (dma_addr_t addr, -enum i915_cache_level level, +unsigned int pat_index,u32 flags) {gen8_pte_t pte = addr |GEN8_PAGE_PRESENT |GEN8_PAGE_RW ;@@ -67,24 +67,17 @@ static u64 mtl_pte_encode(dma_addr_t addr,if (flags &PTE_LM ) pte |=GEN12_PPGTT_PTE_LM |GEN12_PPGTT_PTE_NC ; -switch (level) { -case I915_CACHE_NONE : - pte |=GEN12_PPGTT_PTE_PAT1 ; -break ; -case I915_CACHE_LLC : -case I915_CACHE_L3_LLC : - pte |=GEN12_PPGTT_PTE_PAT0 |GEN12_PPGTT_PTE_PAT1 ; -break ; -case I915_CACHE_WT : +if (pat_index &BIT (0 )) pte |=GEN12_PPGTT_PTE_PAT0 ; -break ; -default : -/* This should never happen. Added to deal with the compile -* error due to the addition of I915_MAX_CACHE_LEVEL. Will -* be removed by the pat_index patch. -*/ -break ; - } + +if (pat_index &BIT (1 )) + pte |=GEN12_PPGTT_PTE_PAT1 ; + +if (pat_index &BIT (2 )) + pte |=GEN12_PPGTT_PTE_PAT2 ; + +if (pat_index &BIT (3 )) + pte |=MTL_PPGTT_PTE_PAT3 ;return pte; }@@ -457,11 +450,11 @@ gen8_ppgtt_insert_pte(struct i915_ppgtt *ppgtt,struct i915_page_directory *pdp, struct sgt_dma *iter, u64 idx, - enum i915_cache_level cache_level, + unsigned int pat_index, u32 flags) { struct i915_page_directory *pd; -const gen8_pte_t pte_encode = ppgtt->vm.pte_encode (0 , cache_level, flags); +const gen8_pte_t pte_encode = ppgtt->vm.pte_encode (0 , pat_index, flags);gen8_pte_t *vaddr; pd =i915_pd_entry (pdp,gen8_pd_index (idx,2 ));@@ -504,10 +497,10 @@ static void xehpsdv_ppgtt_insert_huge(struct i915_address_space *vm, struct i915_vma_resource *vma_res, struct sgt_dma *iter, - enum i915_cache_level cache_level, + unsigned int pat_index, u32 flags) { - const gen8_pte_t pte_encode = vm->pte_encode(0, cache_level, flags); + const gen8_pte_t pte_encode = vm->pte_encode(0, pat_index, flags);unsigned int rem =sg_dma_len (iter->sg);u64 start = vma_res->start;u64 end = start + vma_res->vma_size;@@ -611,10 +604,10 @@ xehpsdv_ppgtt_insert_huge(struct i915_address_space *vm,static void gen8_ppgtt_insert_huge (struct i915_address_space *vm,struct i915_vma_resource *vma_res,struct sgt_dma *iter, -enum i915_cache_level cache_level, +unsigned int pat_index,u32 flags) { -const gen8_pte_t pte_encode = vm->pte_encode (0 , cache_level, flags); +const gen8_pte_t pte_encode = vm->pte_encode (0 , pat_index, flags);unsigned int rem =sg_dma_len (iter->sg);u64 start = vma_res->start;@@ -734,7 +727,7 @@ static void gen8_ppgtt_insert_huge(struct i915_address_space *vm,static void gen8_ppgtt_insert (struct i915_address_space *vm,struct i915_vma_resource *vma_res, -enum i915_cache_level cache_level, +unsigned int pat_index,u32 flags) {struct i915_ppgtt *const ppgtt =i915_vm_to_ppgtt (vm);@@ -742,9 +735,9 @@ static void gen8_ppgtt_insert(struct i915_address_space *vm,if (vma_res->bi.page_sizes.sg >I915_GTT_PAGE_SIZE ) {if (HAS_64K_PAGES (vm->i915)) -xehpsdv_ppgtt_insert_huge (vm, vma_res, &iter, cache_level, flags); +xehpsdv_ppgtt_insert_huge (vm, vma_res, &iter, pat_index, flags);else -gen8_ppgtt_insert_huge (vm, vma_res, &iter, cache_level, flags); +gen8_ppgtt_insert_huge (vm, vma_res, &iter, pat_index, flags); }else {u64 idx = vma_res->start >>GEN8_PTE_SHIFT ;@@ -753,7 +746,7 @@ static void gen8_ppgtt_insert(struct i915_address_space *vm,gen8_pdp_for_page_index (vm, idx); idx =gen8_ppgtt_insert_pte (ppgtt, pdp, &iter, idx, - cache_level, flags); + pat_index, flags); }while (idx); vma_res->page_sizes_gtt =I915_GTT_PAGE_SIZE ;@@ -763,7 +756,7 @@ static void gen8_ppgtt_insert(struct i915_address_space *vm,static void gen8_ppgtt_insert_entry (struct i915_address_space *vm,dma_addr_t addr,u64 offset, -enum i915_cache_level level, +unsigned int pat_index,u32 flags) {u64 idx = offset >>GEN8_PTE_SHIFT ;@@ -777,14 +770,14 @@ static void gen8_ppgtt_insert_entry(struct i915_address_space *vm,GEM_BUG_ON (pt->is_compact); vaddr =px_vaddr (pt); - vaddr[gen8_pd_index (idx,0 )] = vm->pte_encode (addr, level, flags); + vaddr[gen8_pd_index (idx,0 )] = vm->pte_encode (addr, pat_index, flags);drm_clflush_virt_range (&vaddr[gen8_pd_index (idx,0 )],sizeof (*vaddr)); }static void __xehpsdv_ppgtt_insert_entry_lm (struct i915_address_space *vm,dma_addr_t addr,u64 offset, -enum i915_cache_level level, +unsigned int pat_index,u32 flags) {u64 idx = offset >>GEN8_PTE_SHIFT ;@@ -807,20 +800,20 @@ static void __xehpsdv_ppgtt_insert_entry_lm(struct i915_address_space *vm, } vaddr =px_vaddr (pt); - vaddr[gen8_pd_index (idx,0 ) /16 ] = vm->pte_encode (addr, level, flags); + vaddr[gen8_pd_index (idx,0 ) /16 ] = vm->pte_encode (addr, pat_index, flags); }static void xehpsdv_ppgtt_insert_entry (struct i915_address_space *vm,dma_addr_t addr,u64 offset, -enum i915_cache_level level, +unsigned int pat_index,u32 flags) {if (flags &PTE_LM )return __xehpsdv_ppgtt_insert_entry_lm (vm, addr, offset, - level, flags); + pat_index, flags); -return gen8_ppgtt_insert_entry (vm, addr, offset, level, flags); +return gen8_ppgtt_insert_entry (vm, addr, offset, pat_index, flags); }static int gen8_init_scratch (struct i915_address_space *vm)@@ -855,7 +848,9 @@ static int gen8_init_scratch(struct i915_address_space *vm) vm->scratch[0 ]->encode = vm->pte_encode (px_dma (vm->scratch[0 ]), -I915_CACHE_NONE , pte_flags); +i915_gem_get_pat_index (vm->i915, +I915_CACHE_NONE ), + pte_flags);for (i =1 ; i <= vm->top; i++) {struct drm_i915_gem_object *obj;@@ -873,7 +868,9 @@ static int gen8_init_scratch(struct i915_address_space *vm) }fill_px (obj, vm->scratch[i -1 ]->encode); - obj->encode =gen8_pde_encode (px_dma (obj),I915_CACHE_NONE ); + obj->encode =gen8_pde_encode (px_dma (obj), +i915_gem_get_pat_index (vm->i915, +I915_CACHE_NONE )); vm->scratch[i] = obj; }drivers/gpu/drm/i915/gt/gen8_ppgtt.h +1
-2@@ -10,13 +10,12 @@struct i915_address_space ;struct intel_gt ; -enum i915_cache_level ;struct i915_ppgtt *gen8_ppgtt_create (struct intel_gt *gt,unsigned long lmem_pt_obj_flags);u64 gen8_ggtt_pte_encode (dma_addr_t addr, -enum i915_cache_level level, +unsigned int pat_index,u32 flags);#endif drivers/gpu/drm/i915/gt/intel_ggtt.c +40
-42@@ -221,7 +221,7 @@ static void guc_ggtt_invalidate(struct i915_ggtt *ggtt) }static u64 mtl_ggtt_pte_encode (dma_addr_t addr, -enum i915_cache_level level, +unsigned int pat_index,u32 flags) {gen8_pte_t pte = addr |GEN8_PAGE_PRESENT ;@@ -231,30 +231,17 @@ static u64 mtl_ggtt_pte_encode(dma_addr_t addr,if (flags &PTE_LM ) pte |=GEN12_GGTT_PTE_LM ; -switch (level) { -case I915_CACHE_NONE : - pte |=MTL_GGTT_PTE_PAT1 ; -break ; -case I915_CACHE_LLC : -case I915_CACHE_L3_LLC : - pte |=MTL_GGTT_PTE_PAT0 |MTL_GGTT_PTE_PAT1 ; -break ; -case I915_CACHE_WT : +if (pat_index &BIT (0 )) pte |=MTL_GGTT_PTE_PAT0 ; -break ; -default : -/* This should never happen. Added to deal with the compile -* error due to the addition of I915_MAX_CACHE_LEVEL. Will -* be removed by the pat_index patch. -*/ -break ; - } + +if (pat_index &BIT (1 )) + pte |=MTL_GGTT_PTE_PAT1 ;return pte; }u64 gen8_ggtt_pte_encode (dma_addr_t addr, -enum i915_cache_level level, +unsigned int pat_index,u32 flags) {gen8_pte_t pte = addr |GEN8_PAGE_PRESENT ;@@ -273,25 +260,25 @@ static void gen8_set_pte(void __iomem *addr, gen8_pte_t pte)static void gen8_ggtt_insert_page (struct i915_address_space *vm,dma_addr_t addr,u64 offset, -enum i915_cache_level level, +unsigned int pat_index,u32 flags) {struct i915_ggtt *ggtt =i915_vm_to_ggtt (vm);gen8_pte_t __iomem *pte = (gen8_pte_t __iomem *)ggtt->gsm + offset /I915_GTT_PAGE_SIZE ; -gen8_set_pte (pte, ggtt->vm.pte_encode (addr, level, flags)); +gen8_set_pte (pte, ggtt->vm.pte_encode (addr, pat_index, flags)); ggtt->invalidate (ggtt); }static void gen8_ggtt_insert_entries (struct i915_address_space *vm,struct i915_vma_resource *vma_res, -enum i915_cache_level level, +unsigned int pat_index,u32 flags) {struct i915_ggtt *ggtt =i915_vm_to_ggtt (vm); -const gen8_pte_t pte_encode = ggtt->vm.pte_encode (0 , level, flags); +const gen8_pte_t pte_encode = ggtt->vm.pte_encode (0 , pat_index, flags);gen8_pte_t __iomem *gte;gen8_pte_t __iomem *end;struct sgt_iter iter;@@ -348,14 +335,14 @@ static void gen8_ggtt_clear_range(struct i915_address_space *vm,static void gen6_ggtt_insert_page (struct i915_address_space *vm,dma_addr_t addr,u64 offset, -enum i915_cache_level level, +unsigned int pat_index,u32 flags) {struct i915_ggtt *ggtt =i915_vm_to_ggtt (vm);gen6_pte_t __iomem *pte = (gen6_pte_t __iomem *)ggtt->gsm + offset /I915_GTT_PAGE_SIZE ; -iowrite32 (vm->pte_encode (addr, level, flags), pte); +iowrite32 (vm->pte_encode (addr, pat_index, flags), pte); ggtt->invalidate (ggtt); }@@ -368,7 +355,7 @@ static void gen6_ggtt_insert_page(struct i915_address_space *vm,*/ static void gen6_ggtt_insert_entries (struct i915_address_space *vm,struct i915_vma_resource *vma_res, -enum i915_cache_level level, +unsigned int pat_index,u32 flags) {struct i915_ggtt *ggtt =i915_vm_to_ggtt (vm);@@ -385,7 +372,7 @@ static void gen6_ggtt_insert_entries(struct i915_address_space *vm,iowrite32 (vm->scratch[0 ]->encode, gte++); end += (vma_res->node_size + vma_res->guard) /I915_GTT_PAGE_SIZE ;for_each_sgt_daddr (addr, iter, vma_res->bi.pages) -iowrite32 (vm->pte_encode (addr, level, flags), gte++); +iowrite32 (vm->pte_encode (addr, pat_index, flags), gte++);GEM_BUG_ON (gte > end);/* Fill the allocated but "unused" space beyond the end of the buffer */ @@ -420,14 +407,15 @@ struct insert_page {struct i915_address_space *vm;dma_addr_t addr;u64 offset; -enum i915_cache_level level; +unsigned int pat_index; };static int bxt_vtd_ggtt_insert_page__cb (void *_arg) {struct insert_page *arg = _arg; -gen8_ggtt_insert_page (arg->vm, arg->addr, arg->offset, arg->level,0 ); +gen8_ggtt_insert_page (arg->vm, arg->addr, arg->offset, + arg->pat_index,0 );bxt_vtd_ggtt_wa (arg->vm);return 0 ;@@ -436,10 +424,10 @@ static int bxt_vtd_ggtt_insert_page__cb(void *_arg)static void bxt_vtd_ggtt_insert_page__BKL (struct i915_address_space *vm,dma_addr_t addr,u64 offset, -enum i915_cache_level level, +unsigned int pat_index,u32 unused) { -struct insert_page arg = { vm, addr, offset, level }; +struct insert_page arg = { vm, addr, offset, pat_index };stop_machine (bxt_vtd_ggtt_insert_page__cb, &arg,NULL ); }@@ -447,7 +435,7 @@ static void bxt_vtd_ggtt_insert_page__BKL(struct i915_address_space *vm,struct insert_entries {struct i915_address_space *vm;struct i915_vma_resource *vma_res; -enum i915_cache_level level; +unsigned int pat_index;u32 flags; };@@ -455,7 +443,8 @@ static int bxt_vtd_ggtt_insert_entries__cb(void *_arg) {struct insert_entries *arg = _arg; -gen8_ggtt_insert_entries (arg->vm, arg->vma_res, arg->level, arg->flags); +gen8_ggtt_insert_entries (arg->vm, arg->vma_res, + arg->pat_index, arg->flags);bxt_vtd_ggtt_wa (arg->vm);return 0 ;@@ -463,10 +452,10 @@ static int bxt_vtd_ggtt_insert_entries__cb(void *_arg)static void bxt_vtd_ggtt_insert_entries__BKL (struct i915_address_space *vm,struct i915_vma_resource *vma_res, -enum i915_cache_level level, +unsigned int pat_index,u32 flags) { -struct insert_entries arg = { vm, vma_res, level, flags }; +struct insert_entries arg = { vm, vma_res, pat_index, flags };stop_machine (bxt_vtd_ggtt_insert_entries__cb, &arg,NULL ); }@@ -495,7 +484,7 @@ static void gen6_ggtt_clear_range(struct i915_address_space *vm,void intel_ggtt_bind_vma (struct i915_address_space *vm,struct i915_vm_pt_stash *stash,struct i915_vma_resource *vma_res, -enum i915_cache_level cache_level, +unsigned int pat_index,u32 flags) {u32 pte_flags;@@ -512,7 +501,7 @@ void intel_ggtt_bind_vma(struct i915_address_space *vm,if (vma_res->bi.lmem) pte_flags |=PTE_LM ; - vm->insert_entries (vm, vma_res, cache_level, pte_flags); + vm->insert_entries (vm, vma_res, pat_index, pte_flags); vma_res->page_sizes_gtt =I915_GTT_PAGE_SIZE ; }@@ -661,7 +650,7 @@ static int init_ggtt(struct i915_ggtt *ggtt)static void aliasing_gtt_bind_vma (struct i915_address_space *vm,struct i915_vm_pt_stash *stash,struct i915_vma_resource *vma_res, -enum i915_cache_level cache_level, +unsigned int pat_index,u32 flags) {u32 pte_flags;@@ -673,10 +662,10 @@ static void aliasing_gtt_bind_vma(struct i915_address_space *vm,if (flags &I915_VMA_LOCAL_BIND )ppgtt_bind_vma (&i915_vm_to_ggtt (vm)->alias->vm, - stash, vma_res, cache_level, flags); + stash, vma_res, pat_index, flags);if (flags &I915_VMA_GLOBAL_BIND ) - vm->insert_entries (vm, vma_res, cache_level, pte_flags); + vm->insert_entries (vm, vma_res, pat_index, pte_flags); vma_res->bound_flags |= flags; }@@ -933,7 +922,9 @@ static int ggtt_probe_common(struct i915_ggtt *ggtt, u64 size) ggtt->vm.scratch[0 ]->encode = ggtt->vm.pte_encode (px_dma (ggtt->vm.scratch[0 ]), -I915_CACHE_NONE , pte_flags); +i915_gem_get_pat_index (i915, +I915_CACHE_NONE ), + pte_flags);return 0 ; }@@ -1022,6 +1013,11 @@ static int gen8_gmch_probe(struct i915_ggtt *ggtt)return ggtt_probe_common (ggtt, size); } +/* +* For pre-gen8 platforms pat_index is the same as enum i915_cache_level, +* so these PTE encode functions are left with using cache_level. +* See translation table LEGACY_CACHELEVEL. +*/ static u64 snb_pte_encode (dma_addr_t addr,enum i915_cache_level level,u32 flags)@@ -1302,7 +1298,9 @@ bool i915_ggtt_resume_vm(struct i915_address_space *vm)*/ vma->resource->bound_flags =0 ; vma->ops->bind_vma (vm,NULL , vma->resource, - obj ? obj->cache_level :0 , + obj ? obj->pat_index : +i915_gem_get_pat_index (vm->i915, +I915_CACHE_NONE ), was_bound);if (obj) {/* only used during resume => exclusive access */ drivers/gpu/drm/i915/gt/intel_gtt.h +9
-11@@ -165,8 +165,6 @@ typedef u64 gen8_pte_t;#define MTL_2_COH_1W REG_FIELD_PREP(MTL_PAT_INDEX_COH_MODE_MASK, 2)#define MTL_0_COH_NON REG_FIELD_PREP(MTL_PAT_INDEX_COH_MODE_MASK, 0) -enum i915_cache_level ; -struct drm_i915_gem_object ;struct i915_fence_reg ;struct i915_vma ;@@ -234,7 +232,7 @@ struct i915_vma_ops {void (*bind_vma)(struct i915_address_space *vm,struct i915_vm_pt_stash *stash,struct i915_vma_resource *vma_res, -enum i915_cache_level cache_level, +unsigned int pat_index,u32 flags);/* * Unmap an object from an address space. This usually consists of @@ -306,7 +304,7 @@ struct i915_address_space { (*alloc_scratch_dma)(struct i915_address_space *vm, int sz);u64 (*pte_encode)(dma_addr_t addr, -enum i915_cache_level level, +unsigned int pat_index,u32 flags);/* Create a valid PTE */ #define PTE_READ_ONLY BIT(0)#define PTE_LM BIT(1)@@ -321,20 +319,20 @@ struct i915_address_space {void (*insert_page)(struct i915_address_space *vm,dma_addr_t addr,u64 offset, -enum i915_cache_level cache_level, +unsigned int pat_index,u32 flags);void (*insert_entries)(struct i915_address_space *vm,struct i915_vma_resource *vma_res, -enum i915_cache_level cache_level, +unsigned int pat_index,u32 flags);void (*raw_insert_page)(struct i915_address_space *vm,dma_addr_t addr,u64 offset, -enum i915_cache_level cache_level, +unsigned int pat_index,u32 flags);void (*raw_insert_entries)(struct i915_address_space *vm,struct i915_vma_resource *vma_res, -enum i915_cache_level cache_level, +unsigned int pat_index,u32 flags);void (*cleanup)(struct i915_address_space *vm);@@ -581,7 +579,7 @@ void ppgtt_init(struct i915_ppgtt *ppgtt, struct intel_gt *gt,void intel_ggtt_bind_vma (struct i915_address_space *vm,struct i915_vm_pt_stash *stash,struct i915_vma_resource *vma_res, -enum i915_cache_level cache_level, +unsigned int pat_index,u32 flags);void intel_ggtt_unbind_vma (struct i915_address_space *vm,struct i915_vma_resource *vma_res);@@ -639,7 +637,7 @@ void __set_pd_entry(struct i915_page_directory * const pd, const unsigned short idx, struct i915_page_table *pt, - u64 (*encode)(const dma_addr_t, const enum i915_cache_level)); + u64 (*encode)(const dma_addr_t, const unsigned int pat_index));#define set_pd_entry (pd, idx, to) \ __set_pd_entry((pd), (idx), px_pt(to), gen8_pde_encode)@@ -659,7 +657,7 @@ void gen6_ggtt_invalidate(struct i915_ggtt *ggtt);void ppgtt_bind_vma (struct i915_address_space *vm,struct i915_vm_pt_stash *stash,struct i915_vma_resource *vma_res, -enum i915_cache_level cache_level, +unsigned int pat_index,u32 flags);void ppgtt_unbind_vma (struct i915_address_space *vm,struct i915_vma_resource *vma_res);drivers/gpu/drm/i915/gt/intel_migrate.c +26
-21@@ -45,7 +45,9 @@ static void xehpsdv_toggle_pdes(struct i915_address_space *vm,* Insert a dummy PTE into every PT that will map to LMEM to ensure * we have a correctly setup PDE structure for later use. */ - vm->insert_page (vm,0 , d->offset,I915_CACHE_NONE ,PTE_LM ); + vm->insert_page (vm,0 , d->offset, +i915_gem_get_pat_index (vm->i915,I915_CACHE_NONE ), +PTE_LM );GEM_BUG_ON (!pt->is_compact); d->offset +=SZ_2M ; }@@ -63,7 +65,9 @@ static void xehpsdv_insert_pte(struct i915_address_space *vm,* alignment is 64K underneath for the pt, and we are careful * not to access the space in the void. */ - vm->insert_page (vm,px_dma (pt), d->offset,I915_CACHE_NONE ,PTE_LM ); + vm->insert_page (vm,px_dma (pt), d->offset, +i915_gem_get_pat_index (vm->i915,I915_CACHE_NONE ), +PTE_LM ); d->offset +=SZ_64K ; }@@ -73,7 +77,8 @@ static void insert_pte(struct i915_address_space *vm, {struct insert_pte_data *d = data; - vm->insert_page (vm,px_dma (pt), d->offset,I915_CACHE_NONE , + vm->insert_page (vm,px_dma (pt), d->offset, +i915_gem_get_pat_index (vm->i915,I915_CACHE_NONE ),i915_gem_object_is_lmem (pt->base) ?PTE_LM :0 ); d->offset +=PAGE_SIZE ; }@@ -356,13 +361,13 @@ static int max_pte_pkt_size(struct i915_request *rq, int pkt)static int emit_pte (struct i915_request *rq,struct sgt_dma *it, -enum i915_cache_level cache_level, +unsigned int pat_index,bool is_lmem,u64 offset,int length) {bool has_64K_pages =HAS_64K_PAGES (rq->engine->i915); -const u64 encode = rq->context->vm->pte_encode (0 , cache_level, +const u64 encode = rq->context->vm->pte_encode (0 , pat_index, is_lmem ?PTE_LM :0 );struct intel_ring *ring = rq->ring;int pkt, dword_length;@@ -673,17 +678,17 @@ int intel_context_migrate_copy(struct intel_context *ce, const struct i915_deps *deps, struct scatterlist *src, - enum i915_cache_level src_cache_level, + unsigned int src_pat_index, bool src_is_lmem, struct scatterlist *dst, - enum i915_cache_level dst_cache_level, + unsigned int dst_pat_index, bool dst_is_lmem, struct i915_request **out) { struct sgt_dma it_src = sg_sgt(src), it_dst = sg_sgt(dst), it_ccs;struct drm_i915_private *i915 = ce->engine->i915;u64 ccs_bytes_to_cpy =0 , bytes_to_cpy; -enum i915_cache_level ccs_cache_level; +unsigned int ccs_pat_index;u32 src_offset, dst_offset;u8 src_access, dst_access;struct i915_request *rq;@@ -707,12 +712,12 @@ intel_context_migrate_copy(struct intel_context *ce, dst_sz =scatter_list_length (dst);if (src_is_lmem) { it_ccs = it_dst; - ccs_cache_level = dst_cache_level; + ccs_pat_index = dst_pat_index; ccs_is_src = false; }else if (dst_is_lmem) { bytes_to_cpy = dst_sz; it_ccs = it_src; - ccs_cache_level = src_cache_level; + ccs_pat_index = src_pat_index; ccs_is_src = true; }@@ -773,7 +778,7 @@ intel_context_migrate_copy(struct intel_context *ce, src_sz =calculate_chunk_sz (i915, src_is_lmem, bytes_to_cpy, ccs_bytes_to_cpy); - len =emit_pte (rq, &it_src, src_cache_level, src_is_lmem, + len =emit_pte (rq, &it_src, src_pat_index, src_is_lmem, src_offset, src_sz);if (!len) { err = -EINVAL ;@@ -784,7 +789,7 @@ intel_context_migrate_copy(struct intel_context *ce, goto out_rq; } - err =emit_pte (rq, &it_dst, dst_cache_level, dst_is_lmem, + err =emit_pte (rq, &it_dst, dst_pat_index, dst_is_lmem, dst_offset, len);if (err <0 ) goto out_rq;@@ -811,7 +816,7 @@ intel_context_migrate_copy(struct intel_context *ce, goto out_rq; ccs_sz =GET_CCS_BYTES (i915, len); - err =emit_pte (rq, &it_ccs, ccs_cache_level, false, + err =emit_pte (rq, &it_ccs, ccs_pat_index, false, ccs_is_src ? src_offset : dst_offset, ccs_sz);if (err <0 )@@ -979,7 +984,7 @@ int intel_context_migrate_clear(struct intel_context *ce, const struct i915_deps *deps, struct scatterlist *sg, - enum i915_cache_level cache_level, + unsigned int pat_index, bool is_lmem, u32 value, struct i915_request **out)@@ -1027,7 +1032,7 @@ intel_context_migrate_clear(struct intel_context *ce,if (err) goto out_rq; - len =emit_pte (rq, &it, cache_level, is_lmem, offset,CHUNK_SZ ); + len =emit_pte (rq, &it, pat_index, is_lmem, offset,CHUNK_SZ );if (len <=0 ) { err = len; goto out_rq;@@ -1074,10 +1079,10 @@ int intel_migrate_copy(struct intel_migrate *m,struct i915_gem_ww_ctx *ww, const struct i915_deps *deps, struct scatterlist *src, - enum i915_cache_level src_cache_level, + unsigned int src_pat_index, bool src_is_lmem, struct scatterlist *dst, - enum i915_cache_level dst_cache_level, + unsigned int dst_pat_index, bool dst_is_lmem, struct i915_request **out) {@@ -1098,8 +1103,8 @@ int intel_migrate_copy(struct intel_migrate *m, goto out; err =intel_context_migrate_copy (ce, deps, - src, src_cache_level, src_is_lmem, - dst, dst_cache_level, dst_is_lmem, + src, src_pat_index, src_is_lmem, + dst, dst_pat_index, dst_is_lmem, out);intel_context_unpin (ce);@@ -1113,7 +1118,7 @@ intel_migrate_clear(struct intel_migrate *m,struct i915_gem_ww_ctx *ww, const struct i915_deps *deps, struct scatterlist *sg, - enum i915_cache_level cache_level, + unsigned int pat_index, bool is_lmem, u32 value, struct i915_request **out)@@ -1134,7 +1139,7 @@ intel_migrate_clear(struct intel_migrate *m,if (err) goto out; - err =intel_context_migrate_clear (ce, deps, sg, cache_level, + err =intel_context_migrate_clear (ce, deps, sg, pat_index, is_lmem, value, out);intel_context_unpin (ce);drivers/gpu/drm/i915/gt/intel_migrate.h +6
-7@@ -16,7 +16,6 @@ struct i915_request;struct i915_gem_ww_ctx ;struct intel_gt ;struct scatterlist ; -enum i915_cache_level ;int intel_migrate_init (struct intel_migrate *m,struct intel_gt *gt);@@ -26,20 +25,20 @@ int intel_migrate_copy(struct intel_migrate *m,struct i915_gem_ww_ctx *ww, const struct i915_deps *deps, struct scatterlist *src, - enum i915_cache_level src_cache_level, + unsigned int src_pat_index, bool src_is_lmem, struct scatterlist *dst, - enum i915_cache_level dst_cache_level, + unsigned int dst_pat_index, bool dst_is_lmem, struct i915_request **out);int intel_context_migrate_copy (struct intel_context *ce,const struct i915_deps *deps,struct scatterlist *src, -enum i915_cache_level src_cache_level, +unsigned int src_pat_index,bool src_is_lmem,struct scatterlist *dst, -enum i915_cache_level dst_cache_level, +unsigned int dst_pat_index,bool dst_is_lmem,struct i915_request **out);@@ -48,7 +47,7 @@ intel_migrate_clear(struct intel_migrate *m,struct i915_gem_ww_ctx *ww, const struct i915_deps *deps, struct scatterlist *sg, - enum i915_cache_level cache_level, + unsigned int pat_index, bool is_lmem, u32 value, struct i915_request **out);@@ -56,7 +55,7 @@ int intel_context_migrate_clear(struct intel_context *ce, const struct i915_deps *deps, struct scatterlist *sg, - enum i915_cache_level cache_level, + unsigned int pat_index, bool is_lmem, u32 value, struct i915_request **out);drivers/gpu/drm/i915/gt/intel_ppgtt.c +3
-3@@ -98,7 +98,7 @@ void __set_pd_entry(struct i915_page_directory * const pd, const unsigned short idx, struct i915_page_table * const to, - u64 (*encode)(const dma_addr_t, const enum i915_cache_level)) + u64 (*encode)(const dma_addr_t, const unsigned int)) { /* Each thread pre-pins the pd, and we may have a thread per pde. */ GEM_BUG_ON(atomic_read(px_used(pd)) > NALLOC * I915_PDES);@@ -181,7 +181,7 @@ struct i915_ppgtt *i915_ppgtt_create(struct intel_gt *gt,void ppgtt_bind_vma (struct i915_address_space *vm,struct i915_vm_pt_stash *stash,struct i915_vma_resource *vma_res, -enum i915_cache_level cache_level, +unsigned int pat_index,u32 flags) {u32 pte_flags;@@ -199,7 +199,7 @@ void ppgtt_bind_vma(struct i915_address_space *vm,if (vma_res->bi.lmem) pte_flags |=PTE_LM ; - vm->insert_entries (vm, vma_res, cache_level, pte_flags); + vm->insert_entries (vm, vma_res, pat_index, pte_flags);wmb (); }drivers/gpu/drm/i915/gt/selftest_migrate.c +25
-22@@ -137,7 +137,7 @@ static int copy(struct intel_migrate *migrate,static int intel_context_copy_ccs (struct intel_context *ce,const struct i915_deps *deps,struct scatterlist *sg, -enum i915_cache_level cache_level, +unsigned int pat_index,bool write_to_ccs,struct i915_request **out) {@@ -185,7 +185,7 @@ static int intel_context_copy_ccs(struct intel_context *ce,if (err) goto out_rq; - len =emit_pte (rq, &it, cache_level, true, offset,CHUNK_SZ ); + len =emit_pte (rq, &it, pat_index, true, offset,CHUNK_SZ );if (len <=0 ) { err = len; goto out_rq;@@ -223,7 +223,7 @@ intel_migrate_ccs_copy(struct intel_migrate *m,struct i915_gem_ww_ctx *ww, const struct i915_deps *deps, struct scatterlist *sg, - enum i915_cache_level cache_level, + unsigned int pat_index, bool write_to_ccs, struct i915_request **out) {@@ -243,7 +243,7 @@ intel_migrate_ccs_copy(struct intel_migrate *m,if (err) goto out; - err =intel_context_copy_ccs (ce, deps, sg, cache_level, + err =intel_context_copy_ccs (ce, deps, sg, pat_index, write_to_ccs, out);intel_context_unpin (ce);@@ -300,7 +300,7 @@ static int clear(struct intel_migrate *migrate,/* Write the obj data into ccs surface */ err =intel_migrate_ccs_copy (migrate, &ww,NULL , obj->mm.pages->sgl, - obj->cache_level, + obj->pat_index, true, &rq);if (rq && !err) {if (i915_request_wait (rq,0 ,HZ ) <0 ) {@@ -351,7 +351,7 @@ static int clear(struct intel_migrate *migrate, err =intel_migrate_ccs_copy (migrate, &ww,NULL , obj->mm.pages->sgl, - obj->cache_level, + obj->pat_index, false, &rq);if (rq && !err) {if (i915_request_wait (rq,0 ,HZ ) <0 ) {@@ -414,9 +414,9 @@ static int __migrate_copy(struct intel_migrate *migrate, struct i915_request **out) {return intel_migrate_copy (migrate, ww,NULL , - src->mm.pages->sgl, src->cache_level, + src->mm.pages->sgl, src->pat_index,i915_gem_object_is_lmem (src), - dst->mm.pages->sgl, dst->cache_level, + dst->mm.pages->sgl, dst->pat_index,i915_gem_object_is_lmem (dst), out); }@@ -428,9 +428,9 @@ static int __global_copy(struct intel_migrate *migrate, struct i915_request **out) {return intel_context_migrate_copy (migrate->context,NULL , - src->mm.pages->sgl, src->cache_level, + src->mm.pages->sgl, src->pat_index,i915_gem_object_is_lmem (src), - dst->mm.pages->sgl, dst->cache_level, + dst->mm.pages->sgl, dst->pat_index,i915_gem_object_is_lmem (dst), out); }@@ -455,7 +455,7 @@ static int __migrate_clear(struct intel_migrate *migrate, {return intel_migrate_clear (migrate, ww,NULL , obj->mm.pages->sgl, - obj->cache_level, + obj->pat_index,i915_gem_object_is_lmem (obj), value, out); }@@ -468,7 +468,7 @@ static int __global_clear(struct intel_migrate *migrate, {return intel_context_migrate_clear (migrate->context,NULL , obj->mm.pages->sgl, - obj->cache_level, + obj->pat_index,i915_gem_object_is_lmem (obj), value, out); }@@ -648,7 +648,7 @@ static int live_emit_pte_full_ring(void *arg)*/ pr_info ("%s emite_pte ring space=%u\n" , __func__, rq->ring->space); it =sg_sgt (obj->mm.pages->sgl); - len =emit_pte (rq, &it, obj->cache_level, false,0 ,CHUNK_SZ ); + len =emit_pte (rq, &it, obj->pat_index, false,0 ,CHUNK_SZ );if (!len) { err = -EINVAL ; goto out_rq;@@ -844,7 +844,7 @@ static int wrap_ktime_compare(const void *A, const void *B)static int __perf_clear_blt (struct intel_context *ce,struct scatterlist *sg, -enum i915_cache_level cache_level, +unsigned int pat_index,bool is_lmem,size_t sz) {@@ -858,7 +858,7 @@ static int __perf_clear_blt(struct intel_context *ce, t0 =ktime_get (); - err =intel_context_migrate_clear (ce,NULL , sg, cache_level, + err =intel_context_migrate_clear (ce,NULL , sg, pat_index, is_lmem,0 , &rq);if (rq) {if (i915_request_wait (rq,0 ,MAX_SCHEDULE_TIMEOUT ) <0 )@@ -904,7 +904,8 @@ static int perf_clear_blt(void *arg) err =__perf_clear_blt (gt->migrate.context, dst->mm.pages->sgl, -I915_CACHE_NONE , +i915_gem_get_pat_index (gt->i915, +I915_CACHE_NONE ),i915_gem_object_is_lmem (dst), sizes[i]);@@ -919,10 +920,10 @@ static int perf_clear_blt(void *arg)static int __perf_copy_blt (struct intel_context *ce,struct scatterlist *src, -enum i915_cache_level src_cache_level, +unsigned int src_pat_index,bool src_is_lmem,struct scatterlist *dst, -enum i915_cache_level dst_cache_level, +unsigned int dst_pat_index,bool dst_is_lmem,size_t sz) {@@ -937,9 +938,9 @@ static int __perf_copy_blt(struct intel_context *ce, t0 =ktime_get (); err =intel_context_migrate_copy (ce,NULL , - src, src_cache_level, + src, src_pat_index, src_is_lmem, - dst, dst_cache_level, + dst, dst_pat_index, dst_is_lmem, &rq);if (rq) {@@ -994,10 +995,12 @@ static int perf_copy_blt(void *arg) err =__perf_copy_blt (gt->migrate.context, src->mm.pages->sgl, -I915_CACHE_NONE , +i915_gem_get_pat_index (gt->i915, +I915_CACHE_NONE ),i915_gem_object_is_lmem (src), dst->mm.pages->sgl, -I915_CACHE_NONE , +i915_gem_get_pat_index (gt->i915, +I915_CACHE_NONE ),i915_gem_object_is_lmem (dst), sz);drivers/gpu/drm/i915/gt/selftest_reset.c +6
-2@@ -86,7 +86,9 @@ __igt_reset_stolen(struct intel_gt *gt, ggtt->vm.insert_page (&ggtt->vm, dma, ggtt->error_capture.start, -I915_CACHE_NONE ,0 ); +i915_gem_get_pat_index (gt->i915, +I915_CACHE_NONE ), +0 );mb (); s = io_mapping_map_wc(&ggtt->iomap,@@ -127,7 +129,9 @@ __igt_reset_stolen(struct intel_gt *gt, ggtt->vm.insert_page (&ggtt->vm, dma, ggtt->error_capture.start, -I915_CACHE_NONE ,0 ); +i915_gem_get_pat_index (gt->i915, +I915_CACHE_NONE ), +0 );mb (); s = io_mapping_map_wc(&ggtt->iomap,drivers/gpu/drm/i915/gt/selftest_timeline.c +1
-1@@ -836,7 +836,7 @@ static int setup_watcher(struct hwsp_watcher *w, struct intel_gt *gt,return PTR_ERR (obj);/* keep the same cache settings as timeline */ -i915_gem_object_set_cache_coherency (obj, tl->hwsp_ggtt->obj->cache_level); +i915_gem_object_set_pat_index (obj, tl->hwsp_ggtt->obj->pat_index); w->map =i915_gem_object_pin_map_unlocked (obj,page_unmask_bits (tl->hwsp_ggtt->obj->mm.mapping));if (IS_ERR (w->map)) {drivers/gpu/drm/i915/gt/selftest_tlb.c +3
-1@@ -36,6 +36,8 @@ pte_tlbinv(struct intel_context *ce, u64 length, struct rnd_state *prng) { +const unsigned int pat_index = +i915_gem_get_pat_index (ce->vm->i915,I915_CACHE_NONE );struct drm_i915_gem_object *batch;struct drm_mm_node vb_node;struct i915_request *rq;@@ -155,7 +157,7 @@ pte_tlbinv(struct intel_context *ce,/* Flip the PTE between A and B */ if (i915_gem_object_is_lmem (vb->obj)) pte_flags |=PTE_LM ; - ce->vm->insert_entries (ce->vm, &vb_res,0 , pte_flags); + ce->vm->insert_entries (ce->vm, &vb_res, pat_index, pte_flags);/* Flush the PTE update to concurrent HW */ tlbinv (ce->vm, addr & -length, length);drivers/gpu/drm/i915/gt/uc/intel_uc_fw.c +8
-2@@ -890,9 +890,15 @@ static void uc_fw_bind_ggtt(struct intel_uc_fw *uc_fw) pte_flags |=PTE_LM ;if (ggtt->vm.raw_insert_entries) - ggtt->vm.raw_insert_entries (&ggtt->vm, dummy,I915_CACHE_NONE , pte_flags); + ggtt->vm.raw_insert_entries (&ggtt->vm, dummy, +i915_gem_get_pat_index (ggtt->vm.i915, +I915_CACHE_NONE ), + pte_flags);else - ggtt->vm.insert_entries (&ggtt->vm, dummy,I915_CACHE_NONE , pte_flags); + ggtt->vm.insert_entries (&ggtt->vm, dummy, +i915_gem_get_pat_index (ggtt->vm.i915, +I915_CACHE_NONE ), + pte_flags); }static void uc_fw_unbind_ggtt (struct intel_uc_fw *uc_fw)drivers/gpu/drm/i915/i915_debugfs.c +45
-10@@ -139,21 +139,56 @@ static const char *stringify_vma_type(const struct i915_vma *vma)return "ppgtt" ; } -static const char *i915_cache_level_str (struct drm_i915_private *i915,int type) -{ -switch (type) { -case I915_CACHE_NONE :return " uncached" ; -case I915_CACHE_LLC :return HAS_LLC (i915) ?" LLC" :" snooped" ; -case I915_CACHE_L3_LLC :return " L3+LLC" ; -case I915_CACHE_WT :return " WT" ; -default :return "" ; +static const char *i915_cache_level_str (struct drm_i915_gem_object *obj) +{ +struct drm_i915_private *i915 =obj_to_i915 (obj); + +if (IS_METEORLAKE (i915)) { +switch (obj->pat_index) { +case 0 :return " WB" ; +case 1 :return " WT" ; +case 2 :return " UC" ; +case 3 :return " WB (1-Way Coh)" ; +case 4 :return " WB (2-Way Coh)" ; +default :return " not defined" ; + } + }else if (IS_PONTEVECCHIO (i915)) { +switch (obj->pat_index) { +case 0 :return " UC" ; +case 1 :return " WC" ; +case 2 :return " WT" ; +case 3 :return " WB" ; +case 4 :return " WT (CLOS1)" ; +case 5 :return " WB (CLOS1)" ; +case 6 :return " WT (CLOS2)" ; +case 7 :return " WT (CLOS2)" ; +default :return " not defined" ; + } + }else if (GRAPHICS_VER (i915) >=12 ) { +switch (obj->pat_index) { +case 0 :return " WB" ; +case 1 :return " WC" ; +case 2 :return " WT" ; +case 3 :return " UC" ; +default :return " not defined" ; + } + }else { +if (i915_gem_object_has_cache_level (obj,I915_CACHE_NONE )) +return " uncached" ; +else if (i915_gem_object_has_cache_level (obj,I915_CACHE_LLC )) +return HAS_LLC (i915) ?" LLC" :" snooped" ; +else if (i915_gem_object_has_cache_level (obj,I915_CACHE_L3_LLC )) +return " L3+LLC" ; +else if (i915_gem_object_has_cache_level (obj,I915_CACHE_WT )) +return " WT" ; +else +return " not defined" ; } }void i915_debugfs_describe_obj (struct seq_file *m,struct drm_i915_gem_object *obj) { -struct drm_i915_private *dev_priv =to_i915 (obj->base.dev);struct i915_vma *vma;int pin_count =0 ;@@ -165,7 +200,7 @@ i915_debugfs_describe_obj(struct seq_file *m, struct drm_i915_gem_object *obj) obj->base.size /1024 , obj->read_domains, obj->write_domain, -i915_cache_level_str (dev_priv, obj->cache_level), +i915_cache_level_str (obj), obj->mm.dirty ?" dirty" :"" , obj->mm.madv == I915_MADV_DONTNEED ? " purgeable" : "");if (obj->base.name)drivers/gpu/drm/i915/i915_gem.c +12
-4@@ -420,8 +420,12 @@ i915_gem_gtt_pread(struct drm_i915_gem_object *obj, page_length = remain < page_length ? remain : page_length;if (drm_mm_node_allocated (&node)) { ggtt->vm.insert_page (&ggtt->vm, -i915_gem_object_get_dma_address (obj, offset >>PAGE_SHIFT ), - node.start,I915_CACHE_NONE ,0 ); +i915_gem_object_get_dma_address (obj, + offset >>PAGE_SHIFT ), + node.start, +i915_gem_get_pat_index (i915, +I915_CACHE_NONE ), +0 ); }else { page_base += offset &PAGE_MASK ; }@@ -598,8 +602,12 @@ i915_gem_gtt_pwrite_fast(struct drm_i915_gem_object *obj,/* flush the write before we modify the GGTT */ intel_gt_flush_ggtt_writes (ggtt->vm.gt); ggtt->vm.insert_page (&ggtt->vm, -i915_gem_object_get_dma_address (obj, offset >>PAGE_SHIFT ), - node.start,I915_CACHE_NONE ,0 ); +i915_gem_object_get_dma_address (obj, + offset >>PAGE_SHIFT ), + node.start, +i915_gem_get_pat_index (i915, +I915_CACHE_NONE ), +0 );wmb ();/* flush modifications to the GGTT (insert_page) */ }else { page_base += offset &PAGE_MASK ;drivers/gpu/drm/i915/i915_gpu_error.c +6
-2@@ -1117,10 +1117,14 @@ i915_vma_coredump_create(const struct intel_gt *gt,mutex_lock (&ggtt->error_mutex);if (ggtt->vm.raw_insert_page) ggtt->vm.raw_insert_page (&ggtt->vm, dma, slot, -I915_CACHE_NONE ,0 ); +i915_gem_get_pat_index (gt->i915, +I915_CACHE_NONE ), +0 );else ggtt->vm.insert_page (&ggtt->vm, dma, slot, -I915_CACHE_NONE ,0 ); +i915_gem_get_pat_index (gt->i915, +I915_CACHE_NONE ), +0 );mb (); s =io_mapping_map_wc (&ggtt->iomap, slot,PAGE_SIZE );drivers/gpu/drm/i915/i915_vma.c +8
-8@@ -315,7 +315,7 @@ struct i915_vma_work {struct i915_vma_resource *vma_res;struct drm_i915_gem_object *obj;struct i915_sw_dma_fence_cb cb; -enum i915_cache_level cache_level; +unsigned int pat_index;unsigned int flags; };@@ -334,7 +334,7 @@ static void __vma_bind(struct dma_fence_work *work)return ; vma_res->ops->bind_vma (vma_res->vm, &vw->stash, - vma_res, vw->cache_level, vw->flags); + vma_res, vw->pat_index, vw->flags); }static void __vma_release (struct dma_fence_work *work)@@ -426,7 +426,7 @@ i915_vma_resource_init_from_vma(struct i915_vma_resource *vma_res,/** * i915_vma_bind - Sets up PTEs for an VMA in it's corresponding address space. * @vma: VMA to map -* @cache_level: mapping cache level +* @pat_index: PAT index to set in PTE * @flags: flags like global or local mapping * @work: preallocated worker for allocating and binding the PTE * @vma_res: pointer to a preallocated vma resource. The resource is either @@ -437,7 +437,7 @@ i915_vma_resource_init_from_vma(struct i915_vma_resource *vma_res,* Note that DMA addresses are also the only part of the SG table we care about. */ int i915_vma_bind (struct i915_vma *vma, -enum i915_cache_level cache_level, +unsigned int pat_index,u32 flags,struct i915_vma_work *work,struct i915_vma_resource *vma_res)@@ -507,7 +507,7 @@ int i915_vma_bind(struct i915_vma *vma,struct dma_fence *prev; work->vma_res =i915_vma_resource_get (vma->resource); - work->cache_level = cache_level; + work->pat_index = pat_index; work->flags = bind_flags;/* @@ -537,7 +537,7 @@ int i915_vma_bind(struct i915_vma *vma,return ret; } - vma->ops->bind_vma (vma->vm,NULL , vma->resource, cache_level, + vma->ops->bind_vma (vma->vm,NULL , vma->resource, pat_index, bind_flags); }@@ -814,7 +814,7 @@ i915_vma_insert(struct i915_vma *vma, struct i915_gem_ww_ctx *ww, color =0 ;if (i915_vm_has_cache_coloring (vma->vm)) - color = vma->obj->cache_level; + color = vma->obj->pat_index;if (flags &PIN_OFFSET_FIXED ) {u64 offset = flags &PIN_OFFSET_MASK ;@@ -1518,7 +1518,7 @@ int i915_vma_pin_ww(struct i915_vma *vma, struct i915_gem_ww_ctx *ww,GEM_BUG_ON (!vma->pages); err =i915_vma_bind (vma, - vma->obj->cache_level, + vma->obj->pat_index, flags, work, vma_res); vma_res =NULL ;if (err)drivers/gpu/drm/i915/i915_vma.h +1
-1@@ -250,7 +250,7 @@ i915_vma_compare(struct i915_vma *vma,struct i915_vma_work *i915_vma_work (void );int i915_vma_bind (struct i915_vma *vma, -enum i915_cache_level cache_level, +unsigned int pat_index,u32 flags,struct i915_vma_work *work,struct i915_vma_resource *vma_res);drivers/gpu/drm/i915/i915_vma_types.h +0
-2@@ -32,8 +32,6 @@#include "gem/i915_gem_object_types.h" -enum i915_cache_level ; -/** * DOC: Global GTT views * drivers/gpu/drm/i915/selftests/i915_gem.c +4
-1@@ -57,7 +57,10 @@ static void trash_stolen(struct drm_i915_private *i915)u32 __iomem *s;int x; - ggtt->vm.insert_page (&ggtt->vm, dma, slot,I915_CACHE_NONE ,0 ); + ggtt->vm.insert_page (&ggtt->vm, dma, slot, +i915_gem_get_pat_index (i915, +I915_CACHE_NONE ), +0 ); s =io_mapping_map_atomic_wc (&ggtt->iomap, slot);for (x =0 ; x <PAGE_SIZE /sizeof (u32); x++) {drivers/gpu/drm/i915/selftests/i915_gem_evict.c +2
-2@@ -245,7 +245,7 @@ static int igt_evict_for_cache_color(void *arg)struct drm_mm_node target = { .start =I915_GTT_PAGE_SIZE *2 , .size =I915_GTT_PAGE_SIZE , - .color =I915_CACHE_LLC , + .color =i915_gem_get_pat_index (gt->i915,I915_CACHE_LLC ), };struct drm_i915_gem_object *obj;struct i915_vma *vma;@@ -308,7 +308,7 @@ static int igt_evict_for_cache_color(void *arg)/* Attempt to remove the first *pinned* vma, by removing the (empty) * neighbour -- this should fail. */ - target.color =I915_CACHE_L3_LLC ; + target.color =i915_gem_get_pat_index (gt->i915,I915_CACHE_L3_LLC );mutex_lock (&ggtt->vm.mutex); err =i915_gem_evict_for_node (&ggtt->vm,NULL , &target,0 );drivers/gpu/drm/i915/selftests/i915_gem_gtt.c +10
-5@@ -135,7 +135,7 @@ fake_dma_object(struct drm_i915_private *i915, u64 size) obj->write_domain =I915_GEM_DOMAIN_CPU ; obj->read_domains =I915_GEM_DOMAIN_CPU ; - obj->cache_level =I915_CACHE_NONE ; + obj->pat_index =i915_gem_get_pat_index (i915,I915_CACHE_NONE );/* Preallocate the "backing storage" */ if (i915_gem_object_pin_pages_unlocked (obj))@@ -359,7 +359,9 @@ static int lowlevel_hole(struct i915_address_space *vm,with_intel_runtime_pm (vm->gt->uncore->rpm, wakeref) vm->insert_entries(vm, mock_vma_res, -I915_CACHE_NONE ,0 ); +i915_gem_get_pat_index (vm->i915, +I915_CACHE_NONE ), +0 ); } count = n;@@ -1377,7 +1379,10 @@ static int igt_ggtt_page(void *arg) ggtt->vm.insert_page (&ggtt->vm,i915_gem_object_get_dma_address (obj,0 ), - offset,I915_CACHE_NONE ,0 ); + offset, +i915_gem_get_pat_index (i915, +I915_CACHE_NONE ), +0 ); } order =i915_random_order (count, &prng);@@ -1510,7 +1515,7 @@ static int reserve_gtt_with_resource(struct i915_vma *vma, u64 offset)mutex_lock (&vm->mutex); err =i915_gem_gtt_reserve (vm,NULL , &vma->node, obj->base.size, offset, - obj->cache_level, + obj->pat_index,0 );if (!err) {i915_vma_resource_init_from_vma (vma_res, vma);@@ -1690,7 +1695,7 @@ static int insert_gtt_with_resource(struct i915_vma *vma)mutex_lock (&vm->mutex); err =i915_gem_gtt_insert (vm,NULL , &vma->node, obj->base.size,0 , - obj->cache_level,0 , vm->total,0 ); + obj->pat_index,0 , vm->total,0 );if (!err) {i915_vma_resource_init_from_vma (vma_res, vma); vma->resource = vma_res;drivers/gpu/drm/i915/selftests/intel_memory_region.c +3
-1@@ -1070,7 +1070,9 @@ static int igt_lmem_write_cpu(void *arg)/* Put the pages into a known state -- from the gpu for added fun */ intel_engine_pm_get (engine); err =intel_context_migrate_clear (engine->gt->migrate.context,NULL , - obj->mm.pages->sgl,I915_CACHE_NONE , + obj->mm.pages->sgl, +i915_gem_get_pat_index (i915, +I915_CACHE_NONE ), true,0xdeadbeaf , &rq);if (rq) { dma_resv_add_fence(obj->base.resv, &rq->fence,drivers/gpu/drm/i915/selftests/mock_gtt.c +4
-4@@ -27,21 +27,21 @@static void mock_insert_page (struct i915_address_space *vm,dma_addr_t addr,u64 offset, -enum i915_cache_level level, +unsigned int pat_index,u32 flags) { }static void mock_insert_entries (struct i915_address_space *vm,struct i915_vma_resource *vma_res, -enum i915_cache_level level,u32 flags) +unsigned int pat_index,u32 flags) { }static void mock_bind_ppgtt (struct i915_address_space *vm,struct i915_vm_pt_stash *stash,struct i915_vma_resource *vma_res, -enum i915_cache_level cache_level, +unsigned int pat_index,u32 flags) {GEM_BUG_ON (flags &I915_VMA_GLOBAL_BIND );@@ -94,7 +94,7 @@ struct i915_ppgtt *mock_ppgtt(struct drm_i915_private *i915, const char *name)static void mock_bind_ggtt (struct i915_address_space *vm,struct i915_vm_pt_stash *stash,struct i915_vma_resource *vma_res, -enum i915_cache_level cache_level, +unsigned int pat_index,u32 flags) { }
RE: [Intel-gfx] [PATCH 7/8] drm/i915: use pat_index instead of cache_level
Quoted 27 lines · On 20/04/2023 00:00, [email protected] wrote:
Quoted 21 lines
From: Fei Yang <[email protected]>
Currently the KMD is using enum
i915_cache_levelto set caching policy for buffer objects. This is flaky because the PAT index which really controls the caching behavior in PTE has far more levels than what's defined in the enum. In addition, the PAT index is platform dependent, having to translate betweeni915_cache_leveland PAT index is not reliable, and makes the code more complicated.Quoted 1 line
From UMD's perspective there is also a necessity to set caching policy for
performance fine tuning. It's much easier for the UMD to directly use PAT index because the behavior of each PAT index is clearly defined in Bspec. Having the abstracted
i915_cache_levelsitting in between would only cause more ambiguity.For these reasons this patch replaces
i915_cache_levelwith PAT index. Also note, thecache_levelis not completely removed yet, because the KMD still has the need of creating buffer objects with simple cache settings such as cached, uncached, or writethrough. For such simple cases, usingcache_levelwould help simplify the code.Quoted 4 lines
Cc Chris Wilson <[email protected]>Cc Matt Roper <[email protected]>Signed-off-by Fei Yang <[email protected]>Reviewed-by Andi Shyti <[email protected]>
Quoted 5 lines
I think have some ideas no how to perhaps make this simpler, please bear with me.
In my mind get/set caching ioctls need to be failing once explicit pat index has been set by userspace. Or at least not return false information.
By design we are ending the support for set caching ioctl. The patch is included in this series, "drm/i915/mtl: end support for set caching ioctl"
+ if (GRAPHICS_VER_FULL(i915) >= IP_VER(12, 70))
+ return -EOPNOTSUPP;
+
Quoted 5 lines
And I don't like
i915_gem_object_has_cache_levelandi915_gem_get_pat_indexas a refactoring step.It also seems that the driver has a need to query the caching mode set regardless of the route (of setting).
Only for the objects created by the KMD. For UMD created objects with PAT index set KMD should never touch the setting.
Quoted 8 lines
So how about this.
Three callers which query the caching mode:
use_cpu_reloc,vm_fault_gtt,gpu_write_needs_clflush.We convert them to be like:
i915_gem_object_has_caching_mode(obj, PAT_UC / PAT_WT / ...);
PAT_UC/WT/WB are platform dependent (https://gfxspecs.intel.com/Predator/Home/Index/45101),
performing this check you would have to do something like,
if (MTL)
...
else if (PVC)
...
else if (GEN12)
...
else
...Quoted 11 lines
Then apart from the per platform tables for mapping between cache level to pat index, you add tables which map pat index to caching modes (
PAT_UC, etc, naming TBD, just enums or bitmasks also TBD, I haven't looked at the bspec to see how exactly it works).You would use that table in the
i915_gem_object_has_caching_modehelper, called from the above three functions instead ofobj->cache_leveldirect comparison.I am assuming at least for instance
cache_level!=I915_CACHE_NONEwould be equivalent toi915_gem_object_has_caching_mode(obj, PAT_UC), etc.
So far kernel only needs 4 cache levels defined in enum i915_cache_level, kernel doesn't need to understand all PAT indices. By desgin if the userspace is setting PAT index directly, kernel only needs to pass the setting to PTE.
For objects created by kernel (including objects created by userspace without specifying pat index), there are only 4 options (defined in the cachelevel_to_pat).
For objects created by userspace with PAT index set (GEM_CREATE + set_pat extension), kernel should not touch the setting, just pass it to the PAT index bits in PTE.
That's why I was only checking cache_level. Handling PAT index is much more complicated because of its platform dependent nature and even the number of PAT indices varies from platform to platform. Fortunately kernel doesn't need to understand that.
-Fei
Quoted 21 lines
Same mapping table could also be used in debugfs (
i915_cache_level_str) to universally describe anyobj->pat_index, with no need to have anything platform dependend there.In set caching set you always set
obj->pat_indexand so low level code can always just use that.Unless I am missing something (possible) I think like that we end up with no
i915_gem_get_pat_indexsprinkled around and also no confusingi915_gem_object_has_cache_level.
Obj->pat_indexwould be a single point of truth, whileobj->cache_levelis just a legacy field for get/set_caching ioctl - not used in the internal driver flows.We would need an additional field for storing the boolean of whether userspace had overriden the PAT.
Regards,
Tvrtko
Re: [Intel-gfx] [PATCH 7/8] drm/i915: use pat_index instead of cache_level
Quoted 61 lines · Fei
Quoted 27 lines · On 20/04/2023 00:00, [email protected] wrote:
Quoted 21 lines
From: Fei Yang <[email protected]>
Currently the KMD is using enum
i915_cache_levelto set caching policy for buffer objects. This is flaky because the PAT index which really controls the caching behavior in PTE has far more levels than what's defined in the enum. In addition, the PAT index is platform dependent, having to translate betweeni915_cache_leveland PAT index is not reliable, and makes the code more complicated.Quoted 1 line
From UMD's perspective there is also a necessity to set caching policy for
performance fine tuning. It's much easier for the UMD to directly use PAT index because the behavior of each PAT index is clearly defined in Bspec. Having the abstracted
i915_cache_levelsitting in between would only cause more ambiguity.For these reasons this patch replaces
i915_cache_levelwith PAT index. Also note, thecache_levelis not completely removed yet, because the KMD still has the need of creating buffer objects with simple cache settings such as cached, uncached, or writethrough. For such simple cases, usingcache_levelwould help simplify the code.Quoted 4 lines
Cc Chris Wilson <[email protected]>Cc Matt Roper <[email protected]>Signed-off-by Fei Yang <[email protected]>Reviewed-by Andi Shyti <[email protected]>Quoted 5 lines
I think have some ideas no how to perhaps make this simpler, please bear with me.
In my mind get/set caching ioctls need to be failing once explicit pat index has been set by userspace. Or at least not return false information.
By design we are ending the support for set caching ioctl. The patch is included in this series, "drm/i915/mtl: end support for set caching ioctl"
(quoted diff) +3
-0+ if (GRAPHICS_VER_FULL(i915) >= IP_VER(12, 70)) + return -EOPNOTSUPP; +Quoted 6 lines
And I don't like
i915_gem_object_has_cache_levelandi915_gem_get_pat_indexas a refactoring step.It also seems that the driver has a need to query the caching mode set regardless of the route (of setting).
Only for the objects created by the KMD. For UMD created objects with PAT index set KMD should never touch the setting.
Quoted 8 lines
So how about this.
Three callers which query the caching mode:
use_cpu_reloc,vm_fault_gtt,gpu_write_needs_clflush.We convert them to be like:
i915_gem_object_has_caching_mode(obj, PAT_UC / PAT_WT / ...);PAT_UC/WT/WB are platform dependent (https://gfxspecs.intel.com/Predator/Home/Index/45101),
performing this check you would have to do something like,
Quoted 8 lines
if (MTL)
... else if (PVC) ... else if (GEN12) ... else ...
No, it would be doable with a table as I suggested below. And that table could be re-used for debugfs pretty-printing simplifying that code too.
Quoted 19 lines
Quoted 11 lines
Then apart from the per platform tables for mapping between cache level to pat index, you add tables which map pat index to caching modes (
PAT_UC, etc, naming TBD, just enums or bitmasks also TBD, I haven't looked at the bspec to see how exactly it works).You would use that table in the
i915_gem_object_has_caching_modehelper, called from the above three functions instead ofobj->cache_leveldirect comparison.I am assuming at least for instance
cache_level!=I915_CACHE_NONEwould be equivalent toi915_gem_object_has_caching_mode(obj, PAT_UC), etc.So far kernel only needs 4 cache levels defined in enum
i915_cache_level, kernel doesn't need to understand all PAT indices. By desgin if the userspace is setting PAT index directly, kernel only needs to pass the setting to PTE.For objects created by kernel (including objects created by userspace without specifying pat index), there are only 4 options (defined in the
cachelevel_to_pat).
Quoted 7 lines
For objects created by userspace with PAT index set (
GEM_CREATE+set_patextension), kernel should not touch the setting, just pass it to the PAT index bits in PTE.That's why I was only checking
cache_level. Handling PAT index is much more complicated because of its platform dependent nature and even the number of PAT indices varies from platform to platform. Fortunately kernel doesn't need to understand that.
Yeah but I think you maybe missed the spirit of my proposal - which is to simplify the internal code paths by not having the duality of cache_level-vs-pat almost all the way down. But instead cut it at the top API level.
You have this:
+ .cachelevel_to_pat = { \
+ [I915_CACHE_NONE] = 0, \
+ [I915_CACHE_LLC] = 1, \
+ [I915_CACHE_L3_LLC] = 2, \
+ [I915_CACHE_WT] = 3, \
+ }
I propose to add something like:
.legacy_platform_pat = { /* pat index to driver logical flags */
[0] = PAT_UC,
[1] = PAT_WB | PAT_LLC, /* Just illustrating the principle */
};
i915->platform_pat = &legacy_platform_pat
i915_gem_object_has_caching_mode(obj, PAT_UC)
...
return i915->platform_pat & PAT_UC == PAT_UC /* give or take */
get/set_caching_ioctl
{
...
if (obj->has_pat_index) /* set in the extension only */ return -EINVAL;
debugfs:
i915_show_pat_flags(i915->platform_pat[obj->pat_index]); /* generic! */
etc...
Set obj->pat_index in the extension or set_cache_level _only_. No internal code paths then need to use anything but it. No sprinkling of conversion helpers needed or dubious has_cache_level query.
Regards,
Tvrtko
Quoted 21 lines
Quoted 21 lines
Same mapping table could also be used in debugfs (
i915_cache_level_str) to universally describe anyobj->pat_index, with no need to have anything platform dependend there.In set caching set you always set
obj->pat_indexand so low level code can always just use that.Unless I am missing something (possible) I think like that we end up with no
i915_gem_get_pat_indexsprinkled around and also no confusingi915_gem_object_has_cache_level.
Obj->pat_indexwould be a single point of truth, whileobj->cache_levelis just a legacy field for get/set_caching ioctl - not used in the internal driver flows.We would need an additional field for storing the boolean of whether userspace had overriden the PAT.
Regards,
Tvrtko
PATCH 8/8 drm/i915: Allow user to set cache at BO creation
In reply to [email protected]From: Fei Yang <[email protected]>
To comply with the design that buffer objects shall have immutable cache setting through out their life cycle, {set, get}_caching ioctl's are no longer supported from MTL onward. With that change caching policy can only be set at object creation time. The current code applies a default (platform dependent) cache setting for all objects. However this is not optimal for performance tuning. The patch extends the existing gem_create uAPI to let user set PAT index for the object at creation time.
The new extension is platform independent, so UMD's can switch to using this extension for older platforms as well, while {set, get}_caching are still supported on these legacy paltforms for compatibility reason.
--- drivers/gpu/drm/i915/gem/i915_gem_create.c | 36 ++++++++++++++++++++++ drivers/gpu/drm/i915/gem/i915_gem_object.c | 6 ++++ include/uapi/drm/i915_drm.h | 36 ++++++++++++++++++++++ tools/include/uapi/drm/i915_drm.h | 36 ++++++++++++++++++++++ 4 files changed, 114 insertions(+)
drivers/gpu/drm/i915/gem/i915_gem_create.c +36 -0
@@ -245,6 +245,7 @@ struct create_ext {unsigned int n_placements;unsigned int placement_mask;unsigned long flags; +unsigned int pat_index; }; static void repr_placements(char *buf, size_t size,
@@ -394,11 +395,39 @@ static int ext_set_protected(struct i915_user_extension __user *base, void *datareturn 0 ; } +static int ext_set_pat (struct i915_user_extension __user *base,void *data) +{ +struct create_ext *ext_data = data; +struct drm_i915_private *i915 = ext_data->i915; +struct drm_i915_gem_create_ext_set_pat ext; +unsigned int max_pat_index; + +BUILD_BUG_ON (sizeof (struct drm_i915_gem_create_ext_set_pat ) != + offsetofend(struct drm_i915_gem_create_ext_set_pat, rsvd)); + +if (copy_from_user (&ext, base,sizeof (ext))) +return -EFAULT ; + + max_pat_index =INTEL_INFO (i915)->max_pat_index; + +if (ext.pat_index > max_pat_index) { +drm_dbg (&i915->drm,"PAT index is invalid: %u\n" , + ext.pat_index); +return -EINVAL ; + } + + ext_data->pat_index = ext.pat_index; + +return 0 ; +} +static const i915_user_extension_fn create_extensions[] = { [I915_GEM_CREATE_EXT_MEMORY_REGIONS ] = ext_set_placements, [I915_GEM_CREATE_EXT_PROTECTED_CONTENT ] = ext_set_protected, + [I915_GEM_CREATE_EXT_SET_PAT ] = ext_set_pat, }; +#define PAT_INDEX_NOT_SET 0xffff/** * i915_gem_create_ext_ioctl - Creates a new mm object and returns a handle to it. * @dev: drm device pointer
@@ -418,6 +447,7 @@ i915_gem_create_ext_ioctl(struct drm_device *dev, void *data,if (args->flags & ~I915_GEM_CREATE_EXT_FLAG_NEEDS_CPU_ACCESS )return -EINVAL ; + ext_data.pat_index =PAT_INDEX_NOT_SET ; ret = i915_user_extensions(u64_to_user_ptr(args->extensions), create_extensions, ARRAY_SIZE(create_extensions),
@@ -454,5 +484,11 @@ i915_gem_create_ext_ioctl(struct drm_device *dev, void *data,if (IS_ERR (obj))return PTR_ERR (obj); +if (ext_data.pat_index !=PAT_INDEX_NOT_SET ) { +i915_gem_object_set_pat_index (obj, ext_data.pat_index); +/* Mark pat_index is set by UMD */ + obj->cache_level =I915_CACHE_INVAL ; + } +return i915_gem_publish (obj, file, &args->size, &args->handle); }
drivers/gpu/drm/i915/gem/i915_gem_object.c +6 -0
@@ -209,6 +209,12 @@ bool i915_gem_object_can_bypass_llc(struct drm_i915_gem_object *obj)if (!(obj->flags &I915_BO_ALLOC_USER ))return false; +/* +* Always flush cache for UMD objects at creation time. +*/ +if (obj->cache_level ==I915_CACHE_INVAL ) +return true; +/* * EHL and JSL add the 'Bypass LLC' MOCS entry, which should make it * possible for userspace to bypass the GTT caching bits set by the
include/uapi/drm/i915_drm.h +36 -0
@@ -3630,9 +3630,13 @@ struct drm_i915_gem_create_ext {* * For I915_GEM_CREATE_EXT_PROTECTED_CONTENT usage see * struct drm_i915_gem_create_ext_protected_content. +* +* For I915_GEM_CREATE_EXT_SET_PAT usage see +* struct drm_i915_gem_create_ext_set_pat. */ #define I915_GEM_CREATE_EXT_MEMORY_REGIONS 0#define I915_GEM_CREATE_EXT_PROTECTED_CONTENT 1 +#define I915_GEM_CREATE_EXT_SET_PAT 2__u64 extensions; };
@@ -3747,6 +3751,38 @@ struct drm_i915_gem_create_ext_protected_content {__u32 flags; }; +/** +* struct drm_i915_gem_create_ext_set_pat - The +* I915_GEM_CREATE_EXT_SET_PAT extension. +* +* If this extension is provided, the specified caching policy (PAT index) is +* applied to the buffer object. +* +* Below is an example on how to create an object with specific caching policy: +* +* .. code-block:: C +* +* struct drm_i915_gem_create_ext_set_pat set_pat_ext = { +* .base = { .name = I915_GEM_CREATE_EXT_SET_PAT }, +* .pat_index = 0, +* }; +* struct drm_i915_gem_create_ext create_ext = { +* .size = PAGE_SIZE, +* .extensions = (uintptr_t)&set_pat_ext, +* }; +* +* int err = ioctl(fd, DRM_IOCTL_I915_GEM_CREATE_EXT, &create_ext); +* if (err) ... +*/ +struct drm_i915_gem_create_ext_set_pat { +/** @base: Extension link. See struct i915_user_extension. */ +struct i915_user_extension base; +/** @pat_index: PAT index to be set */ +__u32 pat_index; +/** @rsvd: reserved for future use */ +__u32 rsvd; +}; +/* ID of the protected content session managed by i915 when PXP is active */ #define I915_PROTECTED_CONTENT_DEFAULT_SESSION 0xf
tools/include/uapi/drm/i915_drm.h +36 -0
@@ -3607,9 +3607,13 @@ struct drm_i915_gem_create_ext {* * For I915_GEM_CREATE_EXT_PROTECTED_CONTENT usage see * struct drm_i915_gem_create_ext_protected_content. +* +* For I915_GEM_CREATE_EXT_SET_PAT usage see +* struct drm_i915_gem_create_ext_set_pat. */ #define I915_GEM_CREATE_EXT_MEMORY_REGIONS 0#define I915_GEM_CREATE_EXT_PROTECTED_CONTENT 1 +#define I915_GEM_CREATE_EXT_SET_PAT 2__u64 extensions; };
@@ -3724,6 +3728,38 @@ struct drm_i915_gem_create_ext_protected_content {__u32 flags; }; +/** +* struct drm_i915_gem_create_ext_set_pat - The +* I915_GEM_CREATE_EXT_SET_PAT extension. +* +* If this extension is provided, the specified caching policy (PAT index) is +* applied to the buffer object. +* +* Below is an example on how to create an object with specific caching policy: +* +* .. code-block:: C +* +* struct drm_i915_gem_create_ext_set_pat set_pat_ext = { +* .base = { .name = I915_GEM_CREATE_EXT_SET_PAT }, +* .pat_index = 0, +* }; +* struct drm_i915_gem_create_ext create_ext = { +* .size = PAGE_SIZE, +* .extensions = (uintptr_t)&set_pat_ext, +* }; +* +* int err = ioctl(fd, DRM_IOCTL_I915_GEM_CREATE_EXT, &create_ext); +* if (err) ... +*/ +struct drm_i915_gem_create_ext_set_pat { +/** @base: Extension link. See struct i915_user_extension. */ +struct i915_user_extension base; +/** @pat_index: PAT index to be set */ +__u32 pat_index; +/** @rsvd: reserved for future use */ +__u32 rsvd; +}; +/* ID of the protected content session managed by i915 when PXP is active */ #define I915_PROTECTED_CONTENT_DEFAULT_SESSION 0xf
Signature · 1 line
2.25.1
Re: [PATCH 8/8] drm/i915: Allow user to set cache at BO creation
Hi Fei,
Quoted 12 lines
To comply with the design that buffer objects shall have immutable cache setting through out their life cycle, {set, get}_caching ioctl's are no longer supported from MTL onward. With that change caching policy can only be set at object creation time. The current code applies a default (platform dependent) cache setting for all objects. However this is not optimal for performance tuning. The patch extends the existing
gem_createuAPI to let user set PAT index for the object at creation time.
The new extension is platform independent, so UMD's can switch to using this extension for older platforms as well, while {set, get}_caching are still supported on these legacy paltforms for compatibility reason.
Quoted 5 lines
Cc Chris Wilson <[email protected]>Cc Matt Roper <[email protected]>Cc Andi Shyti <[email protected]>Signed-off-by Fei Yang <[email protected]>Reviewed-by Andi Shyti <[email protected]>
because this is an API change, we need some more information here.
First of all you need to CC the userspace guys that have been working on top of your series and get their ack's.
I also believe that this series has also been tested on a separate repository, would you link it in the commit message?
Thanks,
Andi
Re: [Intel-gfx] [PATCH 8/8] drm/i915: Allow user to set cache at BO creation
Quoted 20 lines · Andi Shyti
Hi Fei,
Quoted 12 lines
To comply with the design that buffer objects shall have immutable cache setting through out their life cycle, {set, get}_caching ioctl's are no longer supported from MTL onward. With that change caching policy can only be set at object creation time. The current code applies a default (platform dependent) cache setting for all objects. However this is not optimal for performance tuning. The patch extends the existing
gem_createuAPI to let user set PAT index for the object at creation time.
The new extension is platform independent, so UMD's can switch to using this extension for older platforms as well, while {set, get}_caching are still supported on these legacy paltforms for compatibility reason.Quoted 5 lines
Cc Chris Wilson <[email protected]>Cc Matt Roper <[email protected]>Cc Andi Shyti <[email protected]>Signed-off-by Fei Yang <[email protected]>Reviewed-by Andi Shyti <[email protected]>
Quoted 5 lines
because this is an API change, we need some more information here.
First of all you need to CC the userspace guys that have been working on top of your series and get their ack's.
Yes, and a link to a Mesa merge request which uses the uapi should be included.
IGTs should be ready to before we can merge. I glanced over igt-dev but did not spot anything.
Regards,
Tvrtko
Quoted 6 lines
I also believe that this series has also been tested on a separate repository, would you link it in the commit message?
Thanks,
Andi
Re: [Intel-gfx] [PATCH 8/8] drm/i915: Allow user to set cache at BO creation
Quoted 21 lines · Andi Shyti
Quoted 20 lines
Hi Fei,
Quoted 17 lines
To comply with the design that buffer objects shall have immutable cache setting through out their life cycle, {set, get}_caching ioctl's are no longer supported from MTL onward. With that change caching policy can only be set at object creation time. The current code applies a default (platform dependent) cache setting for all objects. However this is not optimal for performance tuning. The patch extends the existing
gem_createuAPI to let user set PAT index for the object at creation time.
The new extension is platform independent, so UMD's can switch to using this extension for older platforms as well, while {set, get}_caching are still supported on these legacy paltforms for compatibility reason.Cc Chris Wilson <[email protected]>Cc Matt Roper <[email protected]>Cc Andi Shyti <[email protected]>Signed-off-by Fei Yang <[email protected]>Reviewed-by Andi Shyti <[email protected]>
Quoted 8 lines
Quoted 5 lines
because this is an API change, we need some more information here.
First of all you need to CC the userspace guys that have been working on top of your series and get their ack's.
Yes, and a link to a Mesa merge request which uses the uapi should be included.
Working with Mesa team on this, stay tuned.
Quoted 2 lines
IGTs should be ready to before we can merge. I glanced over igt-dev but did not spot anything.
There is a IGT patch posted to gfx-internal-devel, titled "test/gem_create: exercise gem_create_ext_set_pat"
Quoted 10 lines
Regards,
Tvrtko
Quoted 6 lines
I also believe that this series has also been tested on a separate repository, would you link it in the commit message?
Thanks,
Andi
Re: [Intel-gfx] [PATCH 8/8] drm/i915: Allow user to set cache at BO creation
Hi Fei,
Quoted 27 lines
Quoted 18 lines
Quoted 18 lines
Quoted 17 lines
To comply with the design that buffer objects shall have immutable cache setting through out their life cycle, {set, get}_caching ioctl's are no longer supported from MTL onward. With that change caching policy can only be set at object creation time. The current code applies a default (platform dependent) cache setting for all objects. However this is not optimal for performance tuning. The patch extends the existing
gem_createuAPI to let user set PAT index for the object at creation time.
The new extension is platform independent, so UMD's can switch to using this extension for older platforms as well, while {set, get}_caching are still supported on these legacy paltforms for compatibility reason.Cc Chris Wilson <[email protected]>Cc Matt Roper <[email protected]>Cc Andi Shyti <[email protected]>Signed-off-by Fei Yang <[email protected]>Reviewed-by Andi Shyti <[email protected]>Quoted 8 lines
Quoted 5 lines
because this is an API change, we need some more information here.
First of all you need to CC the userspace guys that have been working on top of your series and get their ack's.
Yes, and a link to a Mesa merge request which uses the uapi should be included.
Quoted 7 lines
Working with Mesa team on this, stay tuned.
Quoted 2 lines
IGTs should be ready to before we can merge. I glanced over igt-dev but did not spot anything.
There is a IGT patch posted to gfx-internal-devel, titled "test/gem_create:
exercisegem_create_ext_set_pat"
Any chance to have it ported upstream? It's OK even if it's not merged (at least on my side) but some public interface testing is needed.
If you do post it upstream you could add in the cover letter:
Test-with: <mail-id>
where the mail-id is referred to the upstream patch of the test.
Andi
Re: [Intel-gfx] [PATCH 8/8] drm/i915: Allow user to set cache at BO creation
In reply to Yang, FeiQuoted 24 lines · Fei
Quoted 24 lines · Andi Shyti
Quoted 23 lines
Hi Fei,
Quoted 12 lines
To comply with the design that buffer objects shall have immutable cache setting through out their life cycle, {set, get}_caching ioctl's are no longer supported from MTL onward. With that change caching policy can only be set at object creation time. The current code applies a default (platform dependent) cache setting for all objects. However this is not optimal for performance tuning. The patch extends the existing
gem_createuAPI to let user set PAT index for the object at creation time.
The new extension is platform independent, so UMD's can switch to using this extension for older platforms as well, while {set, get}_caching are still supported on these legacy paltforms for compatibility reason.Quoted 5 lines
Cc Chris Wilson <[email protected]>Cc Matt Roper <[email protected]>Cc Andi Shyti <[email protected]>Signed-off-by Fei Yang <[email protected]>Reviewed-by Andi Shyti <[email protected]>because this is an API change, we need some more information here.
Quoted 8 lines
Quoted 5 lines
Quoted 2 lines
First of all you need to CC the userspace guys that have been working on top of your series and get their ack's.
Yes, and a link to a Mesa merge request which uses the uapi should be included.
Working with Mesa team on this, stay tuned.
I would like to see the extension detection issue is handled before ack'ing this.
How about a new DRM_I915_QUERY_GEM_CREATE_EXTENSIONS item, that returns a u64 array of usable extension names for DRM_IOCTL_I915_GEM_CREATE_EXT?
A similar DRM_I915_QUERY_GEM_CONTEXT_CREATE_EXTENSIONS could also provide an alternative to Alan's "drm/i915/uapi/pxp: Add a GET_PARAM for PXP", and more easily allow advertising future new extensions for context/buffer creation.
-Jordan
A message that is not in the archive (sent off-list or not yet received): BYAPR11MB2567F03AD43D7E2DE2628D5D9A669@BYAPR11MB2567.namprd11.prod.outlook.com
A message that is not in the archive (sent off-list or not yet received): 168232538771.392286.3227368099155268955@jljusten-skl
Re: [Intel-gfx] [PATCH 8/8] drm/i915: Allow user to set cache at BO creation
[fixed mailing lists addresses]
Quoted 33 lines · Jordan Justen
Quoted 24 lines · Fei
Quoted 12 lines · Fei
Quoted 11 lines
Quoted 10 lines · Andi Shyti
Quoted 6 lines
Hi Fei,
because this is an API change, we need some more information here.
First of all you need to CC the userspace guys that have been working on top of your series and get their ack's.
Yes, and a link to a Mesa merge request which uses the uapi should be included.
Quoted 8 lines
Quoted 2 lines
Working with Mesa team on this, stay tuned.
I would like to see the extension detection issue is handled before ack'ing this.
How about a new
DRM_I915_QUERY_GEM_CREATE_EXTENSIONSitem, that returns a u64 array of usable extension names forDRM_IOCTL_I915_GEM_CREATE_EXT?I agree a query mechanism is necessary, but that should be generic for all uAPI's, not just for
GEM_CREATE.Quoted 7 lines
Quoted 4 lines
A similar
DRM_I915_QUERY_GEM_CONTEXT_CREATE_EXTENSIONScould also provide an alternative to Alan's "drm/i915/uapi/pxp: Add aGET_PARAMfor PXP", and more easily allow advertising future new extensions for context/buffer creation.I think we should have a discussion and come up with a sustainable design for the query uAPI, rather than just put in a quick hack for this.
Quoted 8 lines
I think you are being a bit too quick to dismiss my idea as a quick hack... Nevetheless, I would love to hear an alternate suggestion. Just as long as it's not, "let's figure this out later, because I need to add this feature now".
I don't think "just try to use it and if it fails, I guess it isn't supported" is reasonable. So, if we can't do better, at least add a
GET_PARAM. Yeah, it's a quick hack, but it's better than nothing.
Being able to "list" supported extensions sounds like a reasonable principle, albeit a departure from the design direction to date. Which means there are probably no quick solutions. Also, AFAIU, only PXP context create is the problematic one, right? Everything else is pretty much instant or delayed allocation so super cheap to probe by attempting to use.
If I got that right and given this series is about drm_i915_gem_create_ext I don't think this side discussion should be blocking it.
Furthermore the PXP context create story is even more complicated, in a way that it is not just about querying whether the extension is supported, but the expensive check is something more complicated.
Going back to implementation details for this proposed new feature, one alternative to query could be something like:
drm_i915_gem_create_ext.flags |= I915_GEM_CREATE_EXT_FLAG_PROBE_EXTENSIONS;
That would be somewhat more light weight to implement that the i915_query route. And it appears it would work for all ioctls which support extensions apart for i915_context_param_engines.
Regards,
Tvrtko
Re: [Intel-gfx] [PATCH 8/8] drm/i915: Allow user to set cache at BO creation
Quoted 11 lines · Tvrtko Ursulin
Being able to "list" supported extensions sounds like a reasonable principle, albeit a departure from the design direction to date. Which means there are probably no quick solutions. Also, AFAIU, only PXP context create is the problematic one, right? Everything else is pretty much instant or delayed allocation so super cheap to probe by attempting to use.
If I got that right and given this series is about
drm_i915_gem_create_extI don't think this side discussion should be blocking it.
This still leaves the issue of no reasonable detection mechanism for the extension. If the discussion gets too complicated, then can we add a GET_PARAM for the SET_PAT extension? I'm hoping we could either come up with something better reasonably quickly, or i915/Xe can add a new param for each new extensions until a better approach is available.
Quoted 4 lines
Furthermore the PXP context create story is even more complicated, in a way that it is not just about querying whether the extension is supported, but the expensive check is something more complicated.
Quoted 8 lines
Going back to implementation details for this proposed new feature, one alternative to query could be something like:
drm_i915_gem_create_ext.flags |= I915_GEM_CREATE_EXT_FLAG_PROBE_EXTENSIONS;That would be somewhat more light weight to implement that the
i915_queryroute. And it appears it would work for all ioctls which support extensions apart fori915_context_param_engines.
This seems little better than the "try it, and if it works then it's supported".
I'm not suggesting that userspace should be able to check that scenario x+y+z will work, but more a list of extensions that conceivably could work. Normally this should just a matter of the kernel unconditionally adding the newly implemented extension to the list returned in the query call.
If a GET_PARAM can be made for the PXP case, then it seems like a query item returning CONTEXT_CREATE extensions could conditionally omit that extension just as easily as implementing the proposed new GET_PARAM.
-Jordan
Joonas Lahtinen
maintainerIOCTL feature detection (Was: Re: [Intel-gfx] [PATCH 8/8] drm/i915: Allow user to set cache at BO creation)
(+ Faith and Daniel as they have been involved in previous discussions)
Quoting Jordan Justen (2023-04-24 20:13:00)
Quoted 9 lines · Tvrtko Ursulin
Quoted 8 lines
Being able to "list" supported extensions sounds like a reasonable principle, albeit a departure from the design direction to date. Which means there are probably no quick solutions. Also, AFAIU, only PXP context create is the problematic one, right? Everything else is pretty much instant or delayed allocation so super cheap to probe by attempting to use.
Quoted 6 lines
Quoted 3 lines
If I got that right and given this series is about
drm_i915_gem_create_extI don't think this side discussion should be blocking it.This still leaves the issue of no reasonable detection mechanism for the extension.
I remember this exact discussion being repeated at least a few times in the past 5 years. Previously the conclusion was always that detection by trying to use the extension leads to least amount of uAPI surface. There is also no concern of having mismatch in backporting of the query and the functionality patches.
Why exactly do you think it is more reasonable to do the following?
check_if_extension_query_is_supported();
<check retval>check_if_extension_xyz_is_supported();
<check retval>
VS
create_[context,buffer,whatever]_with_extension(); <check errno> destroy_[context,buffer,whatever]();
For contexts and buffers there's no overhead, and we're keeping the uAPI surface smaller (less bugs, less validation effort). Additionally we support checking combinations of extensions A, B and C without extra effort.
If we're not returning enough clear errnos, then that is something to make sure we do.
Regards, Joonas
Quoted 33 lines
If the discussion gets too complicated, then can we add
aGET_PARAMfor theSET_PATextension? I'm hoping we could either come up with something better reasonably quickly, or i915/Xe can add a new param for each new extensions until a better approach is available.Quoted 4 lines
Furthermore the PXP context create story is even more complicated, in a way that it is not just about querying whether the extension is supported, but the expensive check is something more complicated.
Quoted 8 lines
Going back to implementation details for this proposed new feature, one alternative to query could be something like:
drm_i915_gem_create_ext.flags |= I915_GEM_CREATE_EXT_FLAG_PROBE_EXTENSIONS;That would be somewhat more light weight to implement that the
i915_queryroute. And it appears it would work for all ioctls which support extensions apart fori915_context_param_engines.This seems little better than the "try it, and if it works then it's supported".
I'm not suggesting that userspace should be able to check that scenario x+y+z will work, but more a list of extensions that conceivably could work. Normally this should just a matter of the kernel unconditionally adding the newly implemented extension to the list returned in the query call.
If a
GET_PARAMcan be made for the PXP case, then it seems like a query item returningCONTEXT_CREATEextensions could conditionally omit that extension just as easily as implementing the proposed newGET_PARAM.-Jordan
Re: IOCTL feature detection (Was: Re: [Intel-gfx] [PATCH 8/8] drm/i915: Allow user to set cache at BO creation)
Quoted 1 line · Joonas Lahtinen
(+ Faith and Daniel as they have been involved in previous discussions)
An orthogonal (but losely related) question: Is PXP the only subsystem that has the unique problem of: Uses a delayed worker to complete all dependencies for init.. but they take so long that by the time compositors-mesa-init comes up, creation of PXP context blocks too long and may still likely failing and incorrectly assuming PXP is not supported. (when we don't have a GET_PARAM for it).
I believe HuC has a similiar issue - but doesnt reflect in the UAPI but rather the cmd buffers. We don't have other subsystems that have this problem? (dependency on other startups?)
Re: IOCTL feature detection (Was: Re: [Intel-gfx] [PATCH 8/8] drm/i915: Allow user to set cache at BO creation)
In reply to Joonas LahtinenQuoted 25 lines · Joonas Lahtinen
(+ Faith and Daniel as they have been involved in previous discussions)
Quoting Jordan Justen (2023-04-24 20:13:00)
Quoted 9 lines · Tvrtko Ursulin
Quoted 8 lines
Being able to "list" supported extensions sounds like a reasonable principle, albeit a departure from the design direction to date. Which means there are probably no quick solutions. Also, AFAIU, only PXP context create is the problematic one, right? Everything else is pretty much instant or delayed allocation so super cheap to probe by attempting to use.
Quoted 6 lines
Quoted 3 lines
If I got that right and given this series is about
drm_i915_gem_create_extI don't think this side discussion should be blocking it.This still leaves the issue of no reasonable detection mechanism for the extension.
I remember this exact discussion being repeated at least a few times in the past 5 years. Previously the conclusion was always that detection by trying to use the extension leads to least amount of uAPI surface. There is also no concern of having mismatch in backporting of the query and the functionality patches.
Quoted 6 lines
Why exactly do you think it is more reasonable to do the following?
check_if_extension_query_is_supported();
<check retval>check_if_extension_xyz_is_supported();
<check retval>
As I've mentioned a couple times, I'm asking for query item that returns it all the extensions that conceivably could work.
In theory it could be made a single query item which somehow then enumerates if something is a context extension or bo extension. But, it seems simpler for all if we just have a couple query items returning a simple u64 array for the few classes of extensions.
Quoted 7 lines
VS
create_[context,buffer,whatever]_with_extension(); <check errno> destroy_[context,buffer,whatever]();
For contexts and buffers there's no overhead,
There's no-overhead to creating and destroying things? I think the 8s timeout when trying create a protected content context shows it's not always quite that simple.
Over time userspace will have to continue growing this set of create/destroy tests as new extensions are added. Simply so we can discover what extensions are supported.
This doesn't seem like a very robust way to advertise extensions for an api.
Another point ... say you need to debug why some simple app is failing to start the driver. One tool might be strace. But now you have a bunch of noise of calls from the driver creating and destroying things just to discover what extensions the kernel supports. It would be nice if all this was handled with a query item vs a bunch of create/destroys.
Quoted 2 lines
and we're keeping the uAPI surface smaller (less bugs, less validation effort).
I understand wanting to keep the kernel uapi and implementation simple.
Is it too complicated to return a u64 array for a query item indicating the extensions supported for the various classes of extensions? I think in most cases it'll just be essentially shuffling across an array of u64 items. (Since most known extensions will always be supported by the kernel.)
Quoted 2 lines
Additionally we support checking combinations of extensions A, B and C without extra effort.
Regarding combinations of extensions, is that really something userspace needs to probe? I would think if userspace knows about the possible extensions, then it's on userspace to use combinations appropriately.
But, in detecting extensions, it is possible that, say extension B relies on extension A. Now userspace may have to probe to see if extension A is supported, and then probe for extension B while using extension A. Essentially, probing for an extension could become a bit complicated. Vs the kernel just giving us a u64 array of known extensions...
-Jordan
Re: IOCTL feature detection (Was: Re: [Intel-gfx] [PATCH 8/8] drm/i915: Allow user to set cache at BO creation)
In reply to Joonas LahtinenQuoted 38 lines · Joonas Lahtinen
(+ Faith and Daniel as they have been involved in previous discussions)
Quoting Jordan Justen (2023-04-24 20:13:00)
Quoted 9 lines · Tvrtko Ursulin
Quoted 8 lines
Being able to "list" supported extensions sounds like a reasonable principle, albeit a departure from the design direction to date. Which means there are probably no quick solutions. Also, AFAIU, only PXP context create is the problematic one, right? Everything else is pretty much instant or delayed allocation so super cheap to probe by attempting to use.
Quoted 6 lines
Quoted 3 lines
If I got that right and given this series is about
drm_i915_gem_create_extI don't think this side discussion should be blocking it.This still leaves the issue of no reasonable detection mechanism for the extension.
I remember this exact discussion being repeated at least a few times in the past 5 years. Previously the conclusion was always that detection by trying to use the extension leads to least amount of uAPI surface. There is also no concern of having mismatch in backporting of the query and the functionality patches.
Why exactly do you think it is more reasonable to do the following?
check_if_extension_query_is_supported();
<check retval>check_if_extension_xyz_is_supported();
<check retval>VS
create_[context,buffer,whatever]_with_extension(); <check errno> destroy_[context,buffer,whatever]();
Quoted 7 lines
For contexts and buffers there's no overhead, and we're keeping the uAPI surface smaller (less bugs, less validation effort). Additionally we support checking combinations of extensions A, B and C without extra effort.
If we're not returning enough clear errnos, then that is something to make sure we do.
Joonas asked me to put my thoughts here:
- in general the "feature discovery by trying it out" approach is most robust and hence preferred, but it's also not something that's required when there's good reasons against it
- the more a feature spans drivers/modules, the more it should be discovered by trying it out, e.g. dma-buf fence import/export was a huge discussion, luckily mesa devs figured out how to transparantly fall back at runtime so we didn't end up merging the separate feature flag (I think at least, can't find it). pxp being split across i915/me/fw/who knows what else is kinda similar so I'd heavily lean towards discovery by creating a context
- pxp taking 8s to init a ctx sounds very broken, irrespective of anything else
- in practice there's not really a combinatorial explosion, for a lot of reasons:
- kernel and userspace tend to assume/require implied features where it makes sense, e.g. in kms atomic implies planes and universal planes. mesa has been doing the same
- you cold go all the way to what radeon/amdgpu have done for years with a single incrementing version. Probably not flexible enough for intel.
- every pciid is it's own uapi, so a feature only needs to be tested on platforms where it didn't ship from the start. Also cuts down on runtime discovery a lot
- mesa tends to only support down to current lts kernels and not older, which again cuts the combinations a lot
- I did look through upstream kernel docs for general (driver) uapi recommendations, but there isn't really anything about good taste stuff, just a lot about not screwing up compatibility across kernels/platforms.
tldr; prefer discovery, don't sweat it if not, definitely don't overengineer this with some magic "give me all extensions" thing because that results in guaranteed cross-component backporting pains sooner or later. Or inconsistency, which defeats the point.
Cheers, Daniel
Quoted 35 lines
Regards, Joonas
Quoted 33 lines
If the discussion gets too complicated, then can we add
aGET_PARAMfor theSET_PATextension? I'm hoping we could either come up with something better reasonably quickly, or i915/Xe can add a new param for each new extensions until a better approach is available.Quoted 4 lines
Furthermore the PXP context create story is even more complicated, in a way that it is not just about querying whether the extension is supported, but the expensive check is something more complicated.
Quoted 8 lines
Going back to implementation details for this proposed new feature, one alternative to query could be something like:
drm_i915_gem_create_ext.flags |= I915_GEM_CREATE_EXT_FLAG_PROBE_EXTENSIONS;That would be somewhat more light weight to implement that the
i915_queryroute. And it appears it would work for all ioctls which support extensions apart fori915_context_param_engines.This seems little better than the "try it, and if it works then it's supported".
I'm not suggesting that userspace should be able to check that scenario x+y+z will work, but more a list of extensions that conceivably could work. Normally this should just a matter of the kernel unconditionally adding the newly implemented extension to the list returned in the query call.
If a
GET_PARAMcan be made for the PXP case, then it seems like a query item returningCONTEXT_CREATEextensions could conditionally omit that extension just as easily as implementing the proposed newGET_PARAM.-Jordan
Signature · 3 lines
Daniel Vetter Software Engineer, Intel Corporation http://blog.ffwll.ch
Re: IOCTL feature detection (Was: Re: [Intel-gfx] [PATCH 8/8] drm/i915: Allow user to set cache at BO creation)
Quoted 7 lines · Daniel Vetter
Quoted 5 lines · Joonas Lahtinen
(+ Faith and Daniel as they have been involved in previous discussions) Quoting Jordan Justen (2023-04-24 20:13:00)
Quoted 3 lines · Tvrtko Ursulin
Quoted 2 lines
alan:snip
Quoted 11 lines
- the more a feature spans
drivers/modules, the more it should be discovered by trying it out, e.g. dma-buf fence import/export was a huge discussion, luckily mesa devs figured out how to transparantly fall back at runtime so we didn't end up merging the separate feature flag (I think at least, can't find it). pxp being split across i915/me/fw/who knows what else is kinda similar so I'd heavily lean towards discovery by creating a context- pxp taking 8s to init a ctx sounds very broken, irrespective of anything else
Alan: Please be aware that:
1. the wait-timeout was changed to 1 second sometime back.
2. the I'm not deciding the time-out. I initially wanted to keep it at the same timeout as ADL (250 milisec) - and ask the UMD to retry if user needs it. (as per same ADL behavior). Daniele requested to move it to 8 seconds - but thru review process, we reduced it to 1 second.
3. In anycase, thats just the wait-timeout - and we know it wont succeed until ~6 seconds after i915 (~9 secs after boot). The issue isnt our hardware or i915
- its the component driver load <-- this is what's broken.
Details: PXP context is dependent on gsc-fw load, huc-firmware load, mei-gsc-proxy component driver load + bind, huc-authentication and gsc-proxy-init-handshake. Most of above steps begin rather quickly during i915 driver load - the delay seems to come from a very late mei-gsc-proxy component driver load. In fact the parent mei-me driver is only getting ~6 seconds after i915 init is done. That blocks the gsc-proxy-init-handshake and huc-authentication and lastly PXP.
That said, what is broken is why it takes so long to get the component drivers to come up. NOTE: PXP isnt really doing anything differently in the context creation flow (in terms of time-consuming-steps compared to ADL) besides the extra dependency waits these.
We can actually go back to the original timeout of 250 milisecs like we have in ADL but will fail if MESA calls in too early (but will succeed later) ... or... we can create the GET_PARAMs.
A better idea would be to figure out how to control the driver load order and force mei driver + components to get called right after i915. I was informed there is no way to control this and changes here will likely not be accepted upstream.
++ Daniele - can you chime in?
Take note that ADL has the same issue but for whatever reason, the dependant mei component on ADL loaded much sooner - so it was never an issue that was caught but still existed on ADL time merge (if users customize the kernel + compositor for fastboot it will happen).(i realize I havent tested ADL with the new kernel configs that we use to also boot PXP on MTL - wonder if the new mei configs are causing the delay - i.e. ADL customer could suddenly see this 6 sec delay too. - something i have to check now)
Re: IOCTL feature detection (Was: Re: [Intel-gfx] [PATCH 8/8] drm/i915: Allow user to set cache at BO creation)
Quoted 18 lines · Alan Previn
Quoted 5 lines · Daniel Vetter
Quoted 4 lines · Joonas Lahtinen
(+ Faith and Daniel as they have been involved in previous discussions) Quoting Jordan Justen (2023-04-24 20:13:00)
Quoted 2 lines · Tvrtko Ursulin
Quoted 1 line
alan:snip
Quoted 10 lines
- the more a feature spans
drivers/modules, the more it should be discovered by trying it out, e.g. dma-buf fence import/export was a huge discussion, luckily mesa devs figured out how to transparantly fall back at runtime so we didn't end up merging the separate feature flag (I think at least, can't find it). pxp being split across i915/me/fw/who knows what else is kinda similar so I'd heavily lean towards discovery by creating a context- pxp taking 8s to init a ctx sounds very broken, irrespective of anything else
I think there has been a bit of confusion in regards to this timeout and to where it applies, so let me try to clarify to make sure we're all on the same page (Alan has already explained most of it below, but I'm going to go in a bit more detail and I want to make sure it's all in one place for reference). Before we can do any PXP operation, dependencies need to be satisfied, some of which are outside of i915. For MTL, these are:
GSC FW needs to be loaded (~250 ms)
HuC FW needs to be authenticated for PXP ops (~20 ms)
MEI modules need to be bound (depends on the probe ordering, but usually a few secs)
GSC SW proxy via MEI needs to be established (~500 ms normally, but can take a few seconds on the first boot after a firmware update)
Due to the fact that these can take several seconds in total to complete, to avoid stalling driver load/resume for that long we moved the i915-side operations to a separate worker and we register i915 before they've completed. This means that we can get a PXP context creation call before all the dependencies are in place, in which case we do need to wait and that's where the 8s come from. After all the pieces are in place, a PXP context creation call is much faster (up to ~150 ms, which is the time required to start the PXP session if it is not already running).
The reason why we suggested a dedicated getparam was to avoid requiring early users to wait for all of that to happen just to check the capability. By the time an user actually wants to use PXP, we're likely done with the prep steps (or at least we're far along with them) and therefore the wait will be short.
Quoted 9 lines
Alan: Please be aware that:
1. the wait-timeout was changed to 1 second sometime back.
2. the I'm not deciding the time-out. I initially wanted to keep it at the same timeout as ADL (250 milisec) - and ask the UMD to retry if user needs it. (as per same ADL behavior). Daniele requested to move it to 8 seconds - but thru review process, we reduced it to 1 second.
3. In anycase, thats just the wait-timeout - and we know it wont succeed until ~6 seconds after i915 (~9 secs after boot). The issue isnt our hardware or i915
- its the component driver load <-- this is what's broken.
I think the question here is whether the mei driver is taking a long time to probe or if it is just being probed late. In the latter case, I wouldn't call it broken.
Quoted 13 lines
Details: PXP context is dependent on gsc-fw load, huc-firmware load, mei-gsc-proxy component driver load + bind, huc-authentication and gsc-proxy-init-handshake. Most of above steps begin rather quickly during i915 driver load - the delay seems to come from a very late mei-gsc-proxy component driver load. In fact the parent mei-me driver is only getting ~6 seconds after i915 init is done. That blocks the gsc-proxy-init-handshake and huc-authentication and lastly PXP.
That said, what is broken is why it takes so long to get the component drivers to come up. NOTE: PXP isnt really doing anything differently in the context creation flow (in terms of time-consuming-steps compared to ADL) besides the extra dependency waits these.
Quoted 8 lines
We can actually go back to the original timeout of 250 milisecs like we have in ADL but will fail if MESA calls in too early (but will succeed later) ... or... we can create the
GET_PARAMs.A better idea would be to figure out how to control the driver load order and force mei driver + components to get called right after i915. I was informed there is no way to control this and changes here will likely not be accepted upstream.
we could add a device link to mark i915 as a consumer of mei, but I believe that wouldn't work for 2 reasons
1 - on discrete, mei binds to a child device of i915, so the dependency is reversed
2 - the link might just delay the i915 load to after the mei load, which I'm not sure it is something we want (and at that point we could also just wait for mei to bind from within the i915 load).
Daniele
Quoted 10 lines
++ Daniele - can you chime in?
Take note that ADL has the same issue but for whatever reason, the dependant mei component on ADL loaded much sooner - so it was never an issue that was caught but still existed on ADL time merge (if users customize the kernel + compositor for fastboot it will happen).(i realize I havent tested ADL with the new kernel configs that we use to also boot PXP on MTL - wonder if the new mei configs are causing the delay - i.e. ADL customer could suddenly see this 6 sec delay too. - something i have to check now)
Re: IOCTL feature detection (Was: Re: [Intel-gfx] [PATCH 8/8] drm/i915: Allow user to set cache at BO creation)
In reply to Daniel VetterQuoted 6 lines · Daniel Vetter
Joonas asked me to put my thoughts here:
- in general the "feature discovery by trying it out" approach is most robust and hence preferred, but it's also not something that's required when there's good reasons against it
More robust in what sense?
Userspace has to set up some scenario to use the interface, which hopefully is not too complex. It's difficult to predict what it may look like in the future since we are talking about undefined extensions.
Userspace has to rely on the kernel making creating and destroying this thing essentially free. For I915_GEM_CREATE_EXT_PROTECTED_CONTENT, that did not work out. For I915_GEM_CREATE_EXT_SET_PAT, just wondering, since the PAT indices are platform specific, what might happen if we tried to choose some common index value to detect the extension in a generic manner? Could it potentially lead to unexpected behavior, or maybe just an error. I guess it's really extension specific what kind of issues potentially could arise.
Quoted 4 lines
tldr; prefer discovery, don't sweat it if not, definitely don't overengineer this with some magic "give me all extensions" thing because that results in guaranteed cross-component backporting pains sooner or later. Or inconsistency, which defeats the point.
I guess I don't know the full context of your concerns here.
For returning the gem-create extensions, isn't this just a matter of returning the valid indices to the create_extensions array in i915_gem_create.c? Is that an over-engineered query?
It seems weird that there's a reasonably well defined "extension" mechanism here, but no way for the kernel to just tell us what extensions it knows about.
-Jordan
Patchwork
CI[Intel-gfx] ✗ Fi.CI.SPARSE: warning for drm/i915/mtl: Define MOCS and PAT tables for MTL (rev8)
In reply to [email protected]== Series Details ==
== Series Details ==
Series: drm/i915/mtl: Define MOCS and PAT tables for MTL (rev8)
URL : https://patchwork.freedesktop.org/series/115980/
State : warning
== Summary ==
Error: dim sparse failed
Sparse version: v0.6.2
Fast mode used, each commit won't be checked separately.
Patchwork
CI[Intel-gfx] ✗ Fi.CI.BAT: failure for drm/i915/mtl: Define MOCS and PAT tables for MTL (rev8)
In reply to [email protected]== Series Details ==
== Series Details ==
Series: drm/i915/mtl: Define MOCS and PAT tables for MTL (rev8)
URL : https://patchwork.freedesktop.org/series/115980/
State : failure
== Summary ==
CI Bug Log - changes from CI_DRM_13029 -> Patchwork_115980v8
====================================================
Summary
-------
**FAILURE**
Serious unknown changes coming with Patchwork_115980v8 absolutely need to be verified manually.
If you think the reported changes have nothing to do with the changes introduced in Patchwork_115980v8, please notify your bug team to allow them to document this new failure mode, which will reduce false positives in CI.
External URL: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_115980v8/index.html
Participating hosts (38 -> 37)
------------------------------
Missing (1): fi-snb-2520m
Possible new issues
-------------------
Here are the unknown changes that may have been introduced in Patchwork_115980v8:
### IGT changes ###
#### Possible regressions ####
* igt@i915_selftest@live@execlists:
- fi-apl-guc: [PASS][1] -> [DMESG-FAIL][2] +1 similar issue
[1]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_13029/fi-apl-guc/igt@i915_selftest@[email protected]
[2]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_115980v8/fi-apl-guc/igt@i915_selftest@[email protected]
- fi-glk-j4005: [PASS][3] -> [ABORT][4]
[3]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_13029/fi-glk-j4005/igt@i915_selftest@[email protected]
[4]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_115980v8/fi-glk-j4005/igt@i915_selftest@[email protected]
* igt@i915_selftest@live@gt_engines:
- fi-glk-j4005: [PASS][5] -> [FAIL][6]
[5]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_13029/fi-glk-j4005/igt@i915_selftest@live@gt_engines.html
[6]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_115980v8/fi-glk-j4005/igt@i915_selftest@live@gt_engines.html
* igt@i915_selftest@live@gt_mocs:
- fi-glk-j4005: [PASS][7] -> [DMESG-FAIL][8] +2 similar issues
[7]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_13029/fi-glk-j4005/igt@i915_selftest@live@gt_mocs.html
[8]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_115980v8/fi-glk-j4005/igt@i915_selftest@live@gt_mocs.html
#### Suppressed ####
The following results come from untrusted machines, tests, or statuses. They do not affect the overall result.
* igt@i915_selftest@live@requests:
- {bat-mtlp-8}: NOTRUN -> [ABORT][9]
[9]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_115980v8/bat-mtlp-8/igt@i915_selftest@[email protected]
Known issues
------------
Here are the changes found in Patchwork_115980v8 that come from known issues:
### IGT changes ###
#### Issues hit ####
* igt@gem_exec_suspend@basic-s3@smem:
- bat-rpls-1: NOTRUN -> [ABORT][10] ([i915#6687] / [i915#7978])
[10]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_115980v8/bat-rpls-1/igt@gem_exec_suspend@[email protected]
* igt@i915_selftest@live@gt_heartbeat:
- fi-kbl-soraka: [PASS][11] -> [DMESG-FAIL][12] ([i915#5334] / [i915#7872])
[11]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_13029/fi-kbl-soraka/igt@i915_selftest@live@gt_heartbeat.html
[12]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_115980v8/fi-kbl-soraka/igt@i915_selftest@live@gt_heartbeat.html
* igt@i915_selftest@live@hangcheck:
- bat-dg2-8: [PASS][13] -> [ABORT][14] ([i915#7913] / [i915#7979])
[13]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_13029/bat-dg2-8/igt@i915_selftest@[email protected]
[14]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_115980v8/bat-dg2-8/igt@i915_selftest@[email protected]
* igt@i915_selftest@live@slpc:
- bat-rpls-1: NOTRUN -> [DMESG-FAIL][15] ([i915#6367] / [i915#7996])
[15]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_115980v8/bat-rpls-1/igt@i915_selftest@[email protected]
* igt@kms_pipe_crc_basic@read-crc:
- bat-adlp-9: NOTRUN -> [SKIP][16] ([i915#3546]) +1 similar issue
[16]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_115980v8/bat-adlp-9/igt@[email protected]
#### Possible fixes ####
* igt@i915_selftest@live@gt_mocs:
- {bat-mtlp-8}: [ABORT][17] ([i915#8369]) -> [PASS][18]
[17]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_13029/bat-mtlp-8/igt@i915_selftest@live@gt_mocs.html
[18]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_115980v8/bat-mtlp-8/igt@i915_selftest@live@gt_mocs.html
* igt@i915_selftest@live@gt_pm:
- {bat-mtlp-8}: [DMESG-FAIL][19] ([i915#8370]) -> [PASS][20]
[19]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_13029/bat-mtlp-8/igt@i915_selftest@live@gt_pm.html
[20]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_115980v8/bat-mtlp-8/igt@i915_selftest@live@gt_pm.html
* igt@i915_selftest@live@requests:
- bat-rpls-1: [ABORT][21] ([i915#7911]) -> [PASS][22]
[21]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_13029/bat-rpls-1/igt@i915_selftest@[email protected]
[22]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_115980v8/bat-rpls-1/igt@i915_selftest@[email protected]
* igt@i915_selftest@live@workarounds:
- bat-adlm-1: [DMESG-FAIL][23] -> [PASS][24]
[23]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_13029/bat-adlm-1/igt@i915_selftest@[email protected]
[24]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_115980v8/bat-adlm-1/igt@i915_selftest@[email protected]
* igt@kms_pipe_crc_basic@suspend-read-crc@pipe-a-edp-1:
- bat-rplp-1: [FAIL][25] ([fdo#103375]) -> [PASS][26] +5 similar issues
[25]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_13029/bat-rplp-1/igt@kms_pipe_crc_basic@[email protected]
[26]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_115980v8/bat-rplp-1/igt@kms_pipe_crc_basic@[email protected]
#### Warnings ####
* igt@kms_chamelium_frames@dp-crc-fast:
- fi-kbl-soraka: [INCOMPLETE][27] -> [SKIP][28] ([fdo#109271])
[27]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_13029/fi-kbl-soraka/igt@[email protected]
[28]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_115980v8/fi-kbl-soraka/igt@[email protected]
{name}: This element is suppressed. This means it is ignored when computing
the status of the difference (SUCCESS, WARNING, or FAILURE).
[fdo#103375]: https://bugs.freedesktop.org/show_bug.cgi?id=103375
[fdo#109271]: https://bugs.freedesktop.org/show_bug.cgi?id=109271
[i915#3546]: https://gitlab.freedesktop.org/drm/intel/issues/3546
[i915#5334]: https://gitlab.freedesktop.org/drm/intel/issues/5334
[i915#6367]: https://gitlab.freedesktop.org/drm/intel/issues/6367
[i915#6687]: https://gitlab.freedesktop.org/drm/intel/issues/6687
[i915#7872]: https://gitlab.freedesktop.org/drm/intel/issues/7872
[i915#7911]: https://gitlab.freedesktop.org/drm/intel/issues/7911
[i915#7913]: https://gitlab.freedesktop.org/drm/intel/issues/7913
[i915#7978]: https://gitlab.freedesktop.org/drm/intel/issues/7978
[i915#7979]: https://gitlab.freedesktop.org/drm/intel/issues/7979
[i915#7996]: https://gitlab.freedesktop.org/drm/intel/issues/7996
[i915#8369]: https://gitlab.freedesktop.org/drm/intel/issues/8369
[i915#8370]: https://gitlab.freedesktop.org/drm/intel/issues/8370
Build changes
-------------
* Linux: CI_DRM_13029 -> Patchwork_115980v8
CI-20190529: 20190529
CI_DRM_13029: 5eae4746072c2d127c8bd21c76036072aec806a4 @ git://anongit.freedesktop.org/gfx-ci/linux
IGT_7260: 5a0dab0153d184b4497e5e25305699f76a20b303 @ https://gitlab.freedesktop.org/drm/igt-gpu-tools.git
Patchwork_115980v8: 5eae4746072c2d127c8bd21c76036072aec806a4 @ git://anongit.freedesktop.org/gfx-ci/linux
### Linux commits
e22d629e0ae7 drm/i915: Allow user to set cache at BO creation 204ed6727086 drm/i915: use pat_index instead of cache_level dbbf115ee963 drm/i915: preparation for using PAT index 3d0146b9febb drm/i915/mtl: end support for set caching ioctl fd9c0cec5544 drm/i915/mtl: workaround coherency issue for Media 87e4d08f2809 drm/i915/mtl: Add PTE encode function 32550005f5f4 drm/i915/mtl: Define MOCS and PAT tables for MTL b60ac82926c6 drm/i915/mtl: Set has_llc=0
== Logs ==
For more details see: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_115980v8/index.html
Hi Fei,
Quoted 8 lines · On Wed, Apr 19, 2023 at 04:00:50PM -0700, [email protected] wrote:
From: Fei Yang <[email protected]>
The series includes patches needed to enable MTL. Also add new extension for
GEM_CREATEuAPI to let user space set cache policy for buffer objects.v2: addressing review comments and checkpatch warnings
v3: makemtl_ggtt_pte_encodestatic
This series is a mixture of different series, bug fixes along with PAT table. Maybe a good solution is to split this series in more series, so that CI has the possibility to perform a better per-patch testing.
Nirmoy has already sent the first two patches of this series for testing. The 4th patch can be also taken out from this series.
Andi