drm/amdgpu: fix vm size and block size for VMPT (v3)
Subsystems (2)
PATCH 1/1 drm/amdgpu: fix vm size and block size for VMPT (v3)
From: "Zhang, Jerry" <[email protected]>
v2: set both of them in gmc
v3: move vm size and block size in vm manager
--- drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 8 -------- drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 22 ++++++++++++---------- drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h | 4 +++- drivers/gpu/drm/amd/amdgpu/gfxhub_v1_0.c | 2 +- drivers/gpu/drm/amd/amdgpu/gmc_v6_0.c | 10 ++++++++-- drivers/gpu/drm/amd/amdgpu/gmc_v7_0.c | 9 +++++++-- drivers/gpu/drm/amd/amdgpu/gmc_v8_0.c | 9 +++++++-- drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c | 21 +++++++++++++-------- drivers/gpu/drm/amd/amdgpu/mmhub_v1_0.c | 2 +- 9 files changed, 52 insertions(+), 35 deletions(-)
drivers/gpu/drm/amd/amdgpu/amdgpu_device.c +0 -8
@@ -1041,14 +1041,6 @@ static bool amdgpu_check_pot_argument(int arg)static void amdgpu_get_block_size (struct amdgpu_device *adev) { -/* from AI, asic starts to support multiple level VMPT */ -if (adev->asic_type >=CHIP_VEGA10 ) { -if (amdgpu_vm_block_size !=9 ) -dev_warn (adev->dev, -"Multi-VMPT limits block size to one page!\n" ); - amdgpu_vm_block_size =9 ; -return ; - }/* defines number of bits in page table versus page directory, * a page is 4KB so we have 12 bits offset, minimum 9 bits in the * page table and the remaining bits are in the page directory */
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +12 -10
@@ -100,13 +100,14 @@ static unsigned amdgpu_vm_num_entries(struct amdgpu_device *adev,if (level ==0 )/* For the root directory */ return adev->vm_manager.max_pfn >> - (amdgpu_vm_block_size * adev->vm_manager.num_level); + (adev->vm_manager.block_size * + adev->vm_manager.num_level);else if (level == adev->vm_manager.num_level)/* For the page tables on the leaves */ -return AMDGPU_VM_PTE_COUNT ; +return AMDGPU_VM_PTE_COUNT (adev);else /* Everything in between */ -return 1 << amdgpu_vm_block_size; +return 1 << adev->vm_manager.block_size; }/**
@@ -271,7 +272,7 @@ static int amdgpu_vm_alloc_levels(struct amdgpu_device *adev, unsigned level) {unsigned shift = (adev->vm_manager.num_level - level) * - amdgpu_vm_block_size; + adev->vm_manager.block_size;unsigned pt_idx, from, to;int r;
@@ -970,7 +971,7 @@ static struct amdgpu_bo *amdgpu_vm_get_pt(struct amdgpu_pte_update_params *p,unsigned idx, level = p->adev->vm_manager.num_level;while (entry->entries) { - idx = addr >> (amdgpu_vm_block_size * level--); + idx = addr >> (p->adev->vm_manager.block_size * level--); idx %=amdgpu_bo_size (entry->bo) /8 ; entry = &entry->entries[idx]; }
@@ -997,7 +998,8 @@ static void amdgpu_vm_update_ptes(struct amdgpu_pte_update_params *params, uint64_t start, uint64_t end, uint64_t dst, uint64_t flags) { - const uint64_t mask = AMDGPU_VM_PTE_COUNT - 1; + struct amdgpu_device *adev = params->adev; +const uint64_t mask =AMDGPU_VM_PTE_COUNT (adev) -1 ;uint64_t cur_pe_start, cur_nptes, cur_dst;uint64_t addr;/* next GPU address to be updated */
@@ -1021,7 +1023,7 @@ static void amdgpu_vm_update_ptes(struct amdgpu_pte_update_params *params,if ((addr & ~mask) == (end & ~mask)) nptes = end - addr;else - nptes =AMDGPU_VM_PTE_COUNT - (addr & mask); + nptes =AMDGPU_VM_PTE_COUNT (adev) - (addr & mask); cur_pe_start =amdgpu_bo_gpu_offset (pt); cur_pe_start += (addr & mask) *8 ;
@@ -1049,7 +1051,7 @@ static void amdgpu_vm_update_ptes(struct amdgpu_pte_update_params *params,if ((addr & ~mask) == (end & ~mask)) nptes = end - addr;else - nptes =AMDGPU_VM_PTE_COUNT - (addr & mask); + nptes =AMDGPU_VM_PTE_COUNT (adev) - (addr & mask); next_pe_start =amdgpu_bo_gpu_offset (pt); next_pe_start += (addr & mask) *8 ;
@@ -1196,7 +1198,7 @@ static int amdgpu_vm_bo_update_mapping(struct amdgpu_device *adev,* reserve space for one command every (1 << BLOCK_SIZE) * entries or 2k dwords (whatever is smaller) */ - ncmds = (nptes >>min (amdgpu_vm_block_size,11 )) +1 ; + ncmds = (nptes >>min (adev->vm_manager.block_size,11 )) +1 ;/* padding, etc. */ ndw =64 ;
@@ -2067,7 +2069,7 @@ void amdgpu_vm_bo_invalidate(struct amdgpu_device *adev,int amdgpu_vm_init (struct amdgpu_device *adev,struct amdgpu_vm *vm) {const unsigned align =min (AMDGPU_VM_PTB_ALIGN_SIZE , -AMDGPU_VM_PTE_COUNT *8 ); +AMDGPU_VM_PTE_COUNT (adev) *8 );unsigned ring_instance;struct amdgpu_ring *ring;struct amd_sched_rq *rq;
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h +3 -1
@@ -45,7 +45,7 @@#define AMDGPU_VM_MAX_UPDATE_SIZE 0x3FFFF/* number of entries in page table */ -#define AMDGPU_VM_PTE_COUNT (1 << amdgpu_vm_block_size) +#define AMDGPU_VM_PTE_COUNT (adev) (1 << adev->vm_manager.block_size)/* PTBs (Page Table Blocks) need to be aligned to 32K */ #define AMDGPU_VM_PTB_ALIGN_SIZE 32768
@@ -155,6 +155,8 @@ struct amdgpu_vm_manager {uint64_t max_pfn;uint32_t num_level; +uint64_t vm_size; +uint32_t block_size;/* vram base address for page table entry */ u64 vram_base_offset;/* is vm enabled? */
drivers/gpu/drm/amd/amdgpu/gfxhub_v1_0.c +1 -1
@@ -222,7 +222,7 @@ int gfxhub_v1_0_gart_enable(struct amdgpu_device *adev) EXECUTE_PROTECTION_FAULT_ENABLE_DEFAULT, 1); tmp =REG_SET_FIELD (tmp,VM_CONTEXT1_CNTL ,PAGE_TABLE_BLOCK_SIZE , - amdgpu_vm_block_size -9 ); + adev->vm_manager.block_size -9 );WREG32 (SOC15_REG_OFFSET (GC ,0 , mmVM_CONTEXT1_CNTL) + i, tmp);WREG32 (SOC15_REG_OFFSET (GC ,0 , mmVM_CONTEXT1_PAGE_TABLE_START_ADDR_LO32) + i*2 ,0 );WREG32 (SOC15_REG_OFFSET (GC ,0 , mmVM_CONTEXT1_PAGE_TABLE_START_ADDR_HI32) + i*2 ,0 );
drivers/gpu/drm/amd/amdgpu/gmc_v6_0.c +8 -2
@@ -543,7 +543,8 @@ static int gmc_v6_0_gart_enable(struct amdgpu_device *adev)WREG32 (mmVM_CONTEXT1_CNTL,VM_CONTEXT1_CNTL__ENABLE_CONTEXT_MASK | (1UL <<VM_CONTEXT1_CNTL__PAGE_TABLE_DEPTH__SHIFT ) | - ((amdgpu_vm_block_size -9 ) <<VM_CONTEXT1_CNTL__PAGE_TABLE_BLOCK_SIZE__SHIFT )); + ((adev->vm_manager.block_size -9 ) + <<VM_CONTEXT1_CNTL__PAGE_TABLE_BLOCK_SIZE__SHIFT ));if (amdgpu_vm_fault_stop ==AMDGPU_VM_FAULT_STOP_ALWAYS )gmc_v6_0_set_fault_enable_default (adev, false);else
@@ -848,7 +849,12 @@ static int gmc_v6_0_sw_init(void *handle)if (r)return r; - adev->vm_manager.max_pfn = amdgpu_vm_size <<18 ; + adev->vm_manager.vm_size = amdgpu_vm_size; + adev->vm_manager.block_size = amdgpu_vm_block_size; + adev->vm_manager.max_pfn = adev->vm_manager.vm_size <<18 ; + +DRM_INFO ("vm size is %d GB, block size is %d-bit\n" , + adev->vm_manager.vm_size, adev->vm_manager.block_size); adev->mc.mc_mask =0xffffffffffULL ;
drivers/gpu/drm/amd/amdgpu/gmc_v7_0.c +7 -2
@@ -639,7 +639,7 @@ static int gmc_v7_0_gart_enable(struct amdgpu_device *adev) tmp =REG_SET_FIELD (tmp,VM_CONTEXT1_CNTL ,ENABLE_CONTEXT ,1 ); tmp =REG_SET_FIELD (tmp,VM_CONTEXT1_CNTL ,PAGE_TABLE_DEPTH ,1 ); tmp =REG_SET_FIELD (tmp,VM_CONTEXT1_CNTL ,PAGE_TABLE_BLOCK_SIZE , - amdgpu_vm_block_size -9 ); + adev->vm_manager.block_size -9 );WREG32 (mmVM_CONTEXT1_CNTL, tmp);if (amdgpu_vm_fault_stop ==AMDGPU_VM_FAULT_STOP_ALWAYS )gmc_v7_0_set_fault_enable_default (adev, false);
@@ -998,7 +998,12 @@ static int gmc_v7_0_sw_init(void *handle)* Currently set to 4GB ((1 << 20) 4k pages). * Max GPUVM size for cayman and SI is 40 bits. */ - adev->vm_manager.max_pfn = amdgpu_vm_size <<18 ; + adev->vm_manager.vm_size = amdgpu_vm_size; + adev->vm_manager.block_size = amdgpu_vm_block_size; + adev->vm_manager.max_pfn = adev->vm_manager.vm_size <<18 ; + +DRM_INFO ("vm size is %d GB, block size is %d-bit\n" , + adev->vm_manager.vm_size, adev->vm_manager.block_size);/* Set the internal MC address mask * This is the max address of the GPU's
drivers/gpu/drm/amd/amdgpu/gmc_v8_0.c +7 -2
@@ -848,7 +848,7 @@ static int gmc_v8_0_gart_enable(struct amdgpu_device *adev) tmp =REG_SET_FIELD (tmp,VM_CONTEXT1_CNTL ,WRITE_PROTECTION_FAULT_ENABLE_DEFAULT ,1 ); tmp =REG_SET_FIELD (tmp,VM_CONTEXT1_CNTL ,EXECUTE_PROTECTION_FAULT_ENABLE_DEFAULT ,1 ); tmp =REG_SET_FIELD (tmp,VM_CONTEXT1_CNTL ,PAGE_TABLE_BLOCK_SIZE , - amdgpu_vm_block_size -9 ); + adev->vm_manager.block_size -9 );WREG32 (mmVM_CONTEXT1_CNTL, tmp);if (amdgpu_vm_fault_stop ==AMDGPU_VM_FAULT_STOP_ALWAYS )gmc_v8_0_set_fault_enable_default (adev, false);
@@ -1082,7 +1082,12 @@ static int gmc_v8_0_sw_init(void *handle)* Currently set to 4GB ((1 << 20) 4k pages). * Max GPUVM size for cayman and SI is 40 bits. */ - adev->vm_manager.max_pfn = amdgpu_vm_size <<18 ; + adev->vm_manager.vm_size = amdgpu_vm_size; + adev->vm_manager.block_size = amdgpu_vm_block_size; + adev->vm_manager.max_pfn = adev->vm_manager.vm_size <<18 ; + +DRM_INFO ("vm size is %d GB, block size is %d-bit\n" , + adev->vm_manager.vm_size, adev->vm_manager.block_size);/* Set the internal MC address mask * This is the max address of the GPU's
drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c +13 -8
@@ -543,11 +543,23 @@ static int gmc_v9_0_sw_init(void *handle)if (adev->flags &AMD_IS_APU ) { adev->mc.vram_type =AMDGPU_VRAM_TYPE_UNKNOWN ; + adev->vm_manager.vm_size = amdgpu_vm_size; + adev->vm_manager.block_size = amdgpu_vm_block_size; }else {/* XXX Don't know how to get VRAM type yet. */ adev->mc.vram_type =AMDGPU_VRAM_TYPE_HBM ; +/* +* To fulfill 4-level page support, +* vm size is 256TB (48bit), maximum size of Vega10, +* block size 512 (9bit) +*/ + adev->vm_manager.vm_size =1U <<18 ; + adev->vm_manager.block_size =9 ; } +DRM_INFO ("vm size is %d GB, block size is %d-bit\n" , + adev->vm_manager.vm_size, adev->vm_manager.block_size); +/* This interrupt is VMC page fault.*/ r =amdgpu_irq_add_id (adev,AMDGPU_IH_CLIENTID_VMC ,0 , &adev->mc.vm_fault);
@@ -557,14 +569,7 @@ static int gmc_v9_0_sw_init(void *handle)if (r)return r; -/* Because of four level VMPTs, vm size is at least 512GB. -* The maximum size is 256TB (48bit). -*/ -if (amdgpu_vm_size <512 ) { -DRM_WARN ("VM size is at least 512GB!\n" ); - amdgpu_vm_size =512 ; - } - adev->vm_manager.max_pfn = (uint64_t )amdgpu_vm_size <<18 ; + adev->vm_manager.max_pfn = adev->vm_manager.vm_size <<18 ;/* Set the internal MC address mask * This is the max address of the GPU's
drivers/gpu/drm/amd/amdgpu/mmhub_v1_0.c +1 -1
@@ -242,7 +242,7 @@ int mmhub_v1_0_gart_enable(struct amdgpu_device *adev) EXECUTE_PROTECTION_FAULT_ENABLE_DEFAULT, 1); tmp =REG_SET_FIELD (tmp,VM_CONTEXT1_CNTL ,PAGE_TABLE_BLOCK_SIZE , - amdgpu_vm_block_size -9 ); + adev->vm_manager.block_size -9 );WREG32 (SOC15_REG_OFFSET (MMHUB ,0 , mmVM_CONTEXT1_CNTL) + i, tmp);WREG32 (SOC15_REG_OFFSET (MMHUB ,0 , mmVM_CONTEXT1_PAGE_TABLE_START_ADDR_LO32) + i*2 ,0 );WREG32 (SOC15_REG_OFFSET (MMHUB ,0 , mmVM_CONTEXT1_PAGE_TABLE_START_ADDR_HI32) + i*2 ,0 );
Signature · 6 lines
1.9.1 _______________________________________________ amd-gfx mailing list [email protected] https://lists.freedesktop.org/mailman/listinfo/amd-gfx
A message that is not in the archive (sent off-list or not yet received): 1491014612-12448-1-git-send-email-Jerry.Zhang-5C7GfCeVMHo@public.gmane.org
Deucher, Alexander
Reviewed-by Alex DeucherQuoted 150 lines
-----Original Message-----
From: Junwei Zhang [mailto:[email protected]]
Sent: Friday, March 31, 2017 10:44 PM
To: Deucher, Alexander; Koenig, Christian
Cc: [email protected]; Zhang, Jerry
Subject: [PATCH] drm/amdgpu: fix vm size and block size for VMPT (v3)From: "Zhang, Jerry" <[email protected]>
v2: set both of them in gmc
v3: move vm size and block size in vm managerSigned-off-by Junwei Zhang <[email protected]>--- drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 8 -------- drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 22 ++++++++++++---------(quoted diff) +0
-1- drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h | 4 +++- drivers/gpu/drm/amd/amdgpu/gfxhub_v1_0.c | 2 +- drivers/gpu/drm/amd/amdgpu/gmc_v6_0.c | 10 ++++++++-- drivers/gpu/drm/amd/amdgpu/gmc_v7_0.c | 9 +++++++-- drivers/gpu/drm/amd/amdgpu/gmc_v8_0.c | 9 +++++++-- drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c | 21 +++++++++++++-------- drivers/gpu/drm/amd/amdgpu/mmhub_v1_0.c | 2 +- 9 files changed, 52 insertions(+), 35 deletions(-)drivers/gpu/drm/amd/amdgpu/amdgpu_device.c +0
-8@@ -1041,14 +1041,6 @@ static bool amdgpu_check_pot_argument(int arg)static void amdgpu_get_block_size (struct amdgpu_device *adev) { -/* from AI, asic starts to support multiple level VMPT */ -if (adev->asic_type >=CHIP_VEGA10 ) { -if (amdgpu_vm_block_size !=9 ) -dev_warn (adev->dev, -"Multi-VMPT limits block size to one page!\n"); - amdgpu_vm_block_size =9 ; -return ; - }/* defines number of bits in page table versus page directory, * a page is 4KB so we have 12 bits offset, minimum 9 bits in the drivers/gpu/drm/amd/amdgpu/amdgpu_device.c +0
-0* page table and the remaining bits are in the page directory */ drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +4
-3@@ -100,13 +100,14 @@ static unsigned amdgpu_vm_num_entries(struct amdgpu_device *adev, if (level == 0) /* For the root directory */return adev->vm_manager.max_pfn >> - (amdgpu_vm_block_size * adev- >vm_manager.num_level); + (adev->vm_manager.block_size * + adev->vm_manager.num_level); else if (level == adev->vm_manager.num_level) /* For the page tables on the leaves */ - return AMDGPU_VM_PTE_COUNT; + return AMDGPU_VM_PTE_COUNT(adev); else/* Everything in between */ - return 1 << amdgpu_vm_block_size; + return 1 << adev->vm_manager.block_size; }/**
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +2
-2@@ -271,7 +272,7 @@ static int amdgpu_vm_alloc_levels(structamdgpu_device *adev, unsigned level) { unsigned shift = (adev->vm_manager.num_level - level) * - amdgpu_vm_block_size; + adev->vm_manager.block_size;unsigned pt_idx, from, to;int r;@@ -970,7 +971,7 @@ static struct amdgpu_bo *amdgpu_vm_get_pt(structamdgpu_pte_update_params *p, unsigned idx, level = p->adev->vm_manager.num_level;while (entry->entries) { - idx = addr >> (amdgpu_vm_block_size * level--); + idx = addr >> (p->adev->vm_manager.block_size * level--); idx %=amdgpu_bo_size (entry->bo) /8 ; entry = &entry->entries[idx];drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +2
-1}@@ -997,7 +998,8 @@ static void amdgpu_vm_update_ptes(structamdgpu_pte_update_params *params, uint64_t start, uint64_t end, uint64_t dst, uint64_t flags) { - const uint64_t mask = AMDGPU_VM_PTE_COUNT - 1; + struct amdgpu_device *adev = params->adev; +const uint64_t mask =AMDGPU_VM_PTE_COUNT (adev) -1 ;uint64_t cur_pe_start, cur_nptes, cur_dst;drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +1
-1uint64_t addr;/* next GPU address to be updated */ @@ -1021,7 +1023,7 @@ static void amdgpu_vm_update_ptes(struct amdgpu_pte_update_params *params, if ((addr & ~mask) == (end & ~mask)) nptes = end - addr; else - nptes =AMDGPU_VM_PTE_COUNT - (addr & mask); + nptes =AMDGPU_VM_PTE_COUNT (adev) - (addr & mask); cur_pe_start =amdgpu_bo_gpu_offset (pt);drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +1
-1cur_pe_start += (addr & mask) *8 ;@@ -1049,7 +1051,7 @@ static void amdgpu_vm_update_ptes(struct amdgpu_pte_update_params *params, if ((addr & ~mask) == (end & ~mask)) nptes = end - addr; else - nptes =AMDGPU_VM_PTE_COUNT - (addr & mask); + nptes =AMDGPU_VM_PTE_COUNT (adev) - (addr & mask);drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +1
-1next_pe_start =amdgpu_bo_gpu_offset (pt); next_pe_start += (addr & mask) *8 ;@@ -1196,7 +1198,7 @@ static int amdgpu_vm_bo_update_mapping(structamdgpu_device *adev, * reserve space for one command every (1 << BLOCK_SIZE) * entries or 2k dwords (whatever is smaller) */ - ncmds = (nptes >>min (amdgpu_vm_block_size,11 )) +1 ; + ncmds = (nptes >>min (adev->vm_manager.block_size,11 )) +1 ;/* padding, etc. */ drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +1
-1ndw =64 ;@@ -2067,7 +2069,7 @@ void amdgpu_vm_bo_invalidate(structamdgpu_device *adev, int amdgpu_vm_init(struct amdgpu_device *adev, struct amdgpu_vm *vm) { const unsigned align = min(AMDGPU_VM_PTB_ALIGN_SIZE, -AMDGPU_VM_PTE_COUNT *8 ); +AMDGPU_VM_PTE_COUNT (adev) *8 );unsigned ring_instance;struct amdgpu_ring *ring;drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +0
-0struct amd_sched_rq *rq;drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h +0
-0
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h line 49
Quoted 6 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h +1
-1@@ -45,7 +45,7 @@#define AMDGPU_VM_MAX_UPDATE_SIZE 0x3FFFF/* number of entries in page table */ -#define AMDGPU_VM_PTE_COUNT (1 << amdgpu_vm_block_size) +#define AMDGPU_VM_PTE_COUNT (adev) (1 << adev- >vm_manager.block_size)
For safety, put parens around adev. E.g.,
#define AMDGPU_VM_PTE_COUNT(adev) (1 << (adev)->vm_manager.block_size)
With that fixed:
As a follow on patch, it would be nice to add some reasonable clamping in the gmc files so we can still override the default settings and not cause too much problem with multiple asics in the same system.
Alex
Quoted 192 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h +2
-0/* PTBs (Page Table Blocks) need to be aligned to 32K */ #define AMDGPU_VM_PTB_ALIGN_SIZE 32768@@ -155,6 +155,8 @@ struct amdgpu_vm_manager {uint64_t max_pfn;uint32_t num_level; +uint64_t vm_size; +uint32_t block_size;/* vram base address for page table entry */ u64 vram_base_offset;/* is vm enabled? */ drivers/gpu/drm/amd/amdgpu/gfxhub_v1_0.c +1
-1@@ -222,7 +222,7 @@ int gfxhub_v1_0_gart_enable(struct amdgpu_device *adev) EXECUTE_PROTECTION_FAULT_ENABLE_DEFAULT, 1); tmp =REG_SET_FIELD (tmp,VM_CONTEXT1_CNTL ,PAGE_TABLE_BLOCK_SIZE , - amdgpu_vm_block_size -9 ); + adev->vm_manager.block_size -9 ); WREG32(SOC15_REG_OFFSET(GC, 0,
mmVM_CONTEXT1_CNTL) + i, tmp);
WREG32(SOC15_REG_OFFSET(GC, 0,mmVM_CONTEXT1_PAGE_TABLE_START_ADDR_LO32) + i*2, 0);
WREG32(SOC15_REG_OFFSET(GC, 0,mmVM_CONTEXT1_PAGE_TABLE_START_ADDR_HI32) + i*2, 0);drivers/gpu/drm/amd/amdgpu/gmc_v6_0.c +2
-1@@ -543,7 +543,8 @@ static int gmc_v6_0_gart_enable(struct amdgpu_device *adev) WREG32(mmVM_CONTEXT1_CNTL, VM_CONTEXT1_CNTL__ENABLE_CONTEXT_MASK | (1UL << VM_CONTEXT1_CNTL__PAGE_TABLE_DEPTH__SHIFT) | - ((amdgpu_vm_block_size - 9) << VM_CONTEXT1_CNTL__PAGE_TABLE_BLOCK_SIZE__SHIFT)); + ((adev->vm_manager.block_size - 9) + << VM_CONTEXT1_CNTL__PAGE_TABLE_BLOCK_SIZE__SHIFT)); if (amdgpu_vm_fault_stop ==
AMDGPU_VM_FAULT_STOP_ALWAYS)drivers/gpu/drm/amd/amdgpu/gmc_v6_0.c +6
-1gmc_v6_0_set_fault_enable_default (adev, false); else@@ -848,7 +849,12 @@ static int gmc_v6_0_sw_init(void *handle)if (r)return r; - adev->vm_manager.max_pfn = amdgpu_vm_size <<18 ; + adev->vm_manager.vm_size = amdgpu_vm_size; + adev->vm_manager.block_size = amdgpu_vm_block_size; + adev->vm_manager.max_pfn = adev->vm_manager.vm_size <<18 ; + +DRM_INFO ("vm size is %d GB, block size is %d-bit\n" , + adev->vm_manager.vm_size, adev- >vm_manager.block_size); adev->mc.mc_mask =0xffffffffffULL ;drivers/gpu/drm/amd/amdgpu/gmc_v7_0.c +1
-1@@ -639,7 +639,7 @@ static int gmc_v7_0_gart_enable(struct amdgpu_device *adev) tmp =REG_SET_FIELD (tmp,VM_CONTEXT1_CNTL ,ENABLE_CONTEXT ,1 ); tmp =REG_SET_FIELD (tmp,VM_CONTEXT1_CNTL ,PAGE_TABLE_DEPTH ,1 ); tmp =REG_SET_FIELD (tmp,VM_CONTEXT1_CNTL ,PAGE_TABLE_BLOCK_SIZE , - amdgpu_vm_block_size -9 ); + adev->vm_manager.block_size -9 );WREG32 (mmVM_CONTEXT1_CNTL, tmp); if (amdgpu_vm_fault_stop ==
AMDGPU_VM_FAULT_STOP_ALWAYS)drivers/gpu/drm/amd/amdgpu/gmc_v7_0.c +6
-1gmc_v7_0_set_fault_enable_default (adev, false);@@ -998,7 +998,12 @@ static int gmc_v7_0_sw_init(void *handle)* Currently set to 4GB ((1 << 20) 4k pages). * Max GPUVM size for cayman and SI is 40 bits. */ - adev->vm_manager.max_pfn = amdgpu_vm_size <<18 ; + adev->vm_manager.vm_size = amdgpu_vm_size; + adev->vm_manager.block_size = amdgpu_vm_block_size; + adev->vm_manager.max_pfn = adev->vm_manager.vm_size <<18 ; + +DRM_INFO ("vm size is %d GB, block size is %d-bit\n" , + adev->vm_manager.vm_size, adev- >vm_manager.block_size);/* Set the internal MC address mask drivers/gpu/drm/amd/amdgpu/gmc_v7_0.c +0
-0* This is the max address of the GPU's drivers/gpu/drm/amd/amdgpu/gmc_v8_0.c +1
-1@@ -848,7 +848,7 @@ static int gmc_v8_0_gart_enable(struct amdgpu_device *adev) tmp =REG_SET_FIELD (tmp,VM_CONTEXT1_CNTL ,WRITE_PROTECTION_FAULT_ENABLE_DEFAULT ,1 ); tmp =REG_SET_FIELD (tmp,VM_CONTEXT1_CNTL ,EXECUTE_PROTECTION_FAULT_ENABLE_DEFAULT ,1 ); tmp =REG_SET_FIELD (tmp,VM_CONTEXT1_CNTL ,PAGE_TABLE_BLOCK_SIZE , - amdgpu_vm_block_size -9 ); + adev->vm_manager.block_size -9 );WREG32 (mmVM_CONTEXT1_CNTL, tmp); if (amdgpu_vm_fault_stop ==
AMDGPU_VM_FAULT_STOP_ALWAYS)drivers/gpu/drm/amd/amdgpu/gmc_v8_0.c +6
-1gmc_v8_0_set_fault_enable_default (adev, false);@@ -1082,7 +1082,12 @@ static int gmc_v8_0_sw_init(void *handle)* Currently set to 4GB ((1 << 20) 4k pages). * Max GPUVM size for cayman and SI is 40 bits. */ - adev->vm_manager.max_pfn = amdgpu_vm_size <<18 ; + adev->vm_manager.vm_size = amdgpu_vm_size; + adev->vm_manager.block_size = amdgpu_vm_block_size; + adev->vm_manager.max_pfn = adev->vm_manager.vm_size <<18 ; + +DRM_INFO ("vm size is %d GB, block size is %d-bit\n" , + adev->vm_manager.vm_size, adev- >vm_manager.block_size);/* Set the internal MC address mask drivers/gpu/drm/amd/amdgpu/gmc_v8_0.c +0
-0* This is the max address of the GPU's drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c +12
-0@@ -543,11 +543,23 @@ static int gmc_v9_0_sw_init(void *handle)if (adev->flags &AMD_IS_APU ) { adev->mc.vram_type =AMDGPU_VRAM_TYPE_UNKNOWN ; + adev->vm_manager.vm_size = amdgpu_vm_size; + adev->vm_manager.block_size = amdgpu_vm_block_size; }else {/* XXX Don't know how to get VRAM type yet. */ adev->mc.vram_type =AMDGPU_VRAM_TYPE_HBM ; +/* +* To fulfill 4-level page support, +* vm size is 256TB (48bit), maximum size of Vega10, +* block size 512 (9bit) +*/ + adev->vm_manager.vm_size =1U <<18 ; + adev->vm_manager.block_size =9 ; } +DRM_INFO ("vm size is %d GB, block size is %d-bit\n" , + adev->vm_manager.vm_size, adev- >vm_manager.block_size); +/* This interrupt is VMC page fault.*/ r = amdgpu_irq_add_id(adev, AMDGPU_IH_CLIENTID_VMC, 0,drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c +1
-8&adev->mc.vm_fault);@@ -557,14 +569,7 @@ static int gmc_v9_0_sw_init(void *handle)if (r)return r; -/* Because of four level VMPTs, vm size is at least 512GB. -* The maximum size is 256TB (48bit). -*/ -if (amdgpu_vm_size <512 ) { -DRM_WARN ("VM size is at least 512GB!\n" ); - amdgpu_vm_size =512 ; - } - adev->vm_manager.max_pfn = (uint64_t )amdgpu_vm_size <<18 ; + adev->vm_manager.max_pfn = adev->vm_manager.vm_size <<18 ;/* Set the internal MC address mask * This is the max address of the GPU's drivers/gpu/drm/amd/amdgpu/mmhub_v1_0.c +1
-1@@ -242,7 +242,7 @@ int mmhub_v1_0_gart_enable(struct amdgpu_device *adev) EXECUTE_PROTECTION_FAULT_ENABLE_DEFAULT, 1); tmp =REG_SET_FIELD (tmp,VM_CONTEXT1_CNTL ,PAGE_TABLE_BLOCK_SIZE , - amdgpu_vm_block_size -9 ); + adev->vm_manager.block_size -9 ); WREG32(SOC15_REG_OFFSET(MMHUB, 0,
Quoted 7 lines
mmVM_CONTEXT1_CNTL) + i, tmp);
WREG32(SOC15_REG_OFFSET(MMHUB, 0,mmVM_CONTEXT1_PAGE_TABLE_START_ADDR_LO32) + i*2, 0);
WREG32(SOC15_REG_OFFSET(MMHUB, 0,mmVM_CONTEXT1_PAGE_TABLE_START_ADDR_HI32) + i*2, 0); --
1.9.1
_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
A message that is not in the archive (sent off-list or not yet received): BN6PR12MB1652CF4F5322EEF10C39E636F7080-/b2+HYfkarQqUD6E6FAiowdYzm3356FpvxpqHgZTriW3zl9H0oFU5g@public.gmane.org
Quoted 152 lines · Alexander
Quoted 152 lines
-----Original Message-----
From: Junwei Zhang [mailto:[email protected]]
Sent: Friday, March 31, 2017 10:44 PM
To: Deucher, Alexander; Koenig, Christian
Cc: [email protected]; Zhang, Jerry
Subject: [PATCH] drm/amdgpu: fix vm size and block size for VMPT (v3)From: "Zhang, Jerry" <[email protected]>
v2: set both of them in gmc
v3: move vm size and block size in vm managerSigned-off-by Junwei Zhang <[email protected]>--- drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 8 -------- drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 22 ++++++++++++---------(quoted diff) +0
-1- drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h | 4 +++- drivers/gpu/drm/amd/amdgpu/gfxhub_v1_0.c | 2 +- drivers/gpu/drm/amd/amdgpu/gmc_v6_0.c | 10 ++++++++-- drivers/gpu/drm/amd/amdgpu/gmc_v7_0.c | 9 +++++++-- drivers/gpu/drm/amd/amdgpu/gmc_v8_0.c | 9 +++++++-- drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c | 21 +++++++++++++-------- drivers/gpu/drm/amd/amdgpu/mmhub_v1_0.c | 2 +- 9 files changed, 52 insertions(+), 35 deletions(-)drivers/gpu/drm/amd/amdgpu/amdgpu_device.c +0
-8@@ -1041,14 +1041,6 @@ static bool amdgpu_check_pot_argument(int arg)static void amdgpu_get_block_size (struct amdgpu_device *adev) { -/* from AI, asic starts to support multiple level VMPT */ -if (adev->asic_type >=CHIP_VEGA10 ) { -if (amdgpu_vm_block_size !=9 ) -dev_warn (adev->dev, -"Multi-VMPT limits block size to one page!\n"); - amdgpu_vm_block_size =9 ; -return ; - }/* defines number of bits in page table versus page directory, * a page is 4KB so we have 12 bits offset, minimum 9 bits in the * page table and the remaining bits are in the page directory */
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +4
-3@@ -100,13 +100,14 @@ static unsigned amdgpu_vm_num_entries(struct amdgpu_device *adev, if (level == 0) /* For the root directory */return adev->vm_manager.max_pfn >> - (amdgpu_vm_block_size * adev- >vm_manager.num_level); + (adev->vm_manager.block_size * + adev->vm_manager.num_level); else if (level == adev->vm_manager.num_level) /* For the page tables on the leaves */ - return AMDGPU_VM_PTE_COUNT; + return AMDGPU_VM_PTE_COUNT(adev); else/* Everything in between */ - return 1 << amdgpu_vm_block_size; + return 1 << adev->vm_manager.block_size; }/**
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +2
-2@@ -271,7 +272,7 @@ static int amdgpu_vm_alloc_levels(structamdgpu_device *adev, unsigned level) { unsigned shift = (adev->vm_manager.num_level - level) * - amdgpu_vm_block_size; + adev->vm_manager.block_size;unsigned pt_idx, from, to;int r;@@ -970,7 +971,7 @@ static struct amdgpu_bo *amdgpu_vm_get_pt(structamdgpu_pte_update_params *p, unsigned idx, level = p->adev->vm_manager.num_level;while (entry->entries) { - idx = addr >> (amdgpu_vm_block_size * level--); + idx = addr >> (p->adev->vm_manager.block_size * level--); idx %=amdgpu_bo_size (entry->bo) /8 ; entry = &entry->entries[idx];}drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +2
-1@@ -997,7 +998,8 @@ static void amdgpu_vm_update_ptes(structamdgpu_pte_update_params *params, uint64_t start, uint64_t end, uint64_t dst, uint64_t flags) { - const uint64_t mask = AMDGPU_VM_PTE_COUNT - 1; + struct amdgpu_device *adev = params->adev; +const uint64_t mask =AMDGPU_VM_PTE_COUNT (adev) -1 ;uint64_t cur_pe_start, cur_nptes, cur_dst;
uint64_taddr; /* next GPU address to be updated */drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +1
-1@@ -1021,7 +1023,7 @@ static void amdgpu_vm_update_ptes(struct amdgpu_pte_update_params *params, if ((addr & ~mask) == (end & ~mask)) nptes = end - addr; else - nptes =AMDGPU_VM_PTE_COUNT - (addr & mask); + nptes =AMDGPU_VM_PTE_COUNT (adev) - (addr & mask); cur_pe_start =amdgpu_bo_gpu_offset (pt);cur_pe_start += (addr & mask) * 8;drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +1
-1@@ -1049,7 +1051,7 @@ static void amdgpu_vm_update_ptes(struct amdgpu_pte_update_params *params, if ((addr & ~mask) == (end & ~mask)) nptes = end - addr; else - nptes =AMDGPU_VM_PTE_COUNT - (addr & mask); + nptes =AMDGPU_VM_PTE_COUNT (adev) - (addr & mask);next_pe_start = amdgpu_bo_gpu_offset(pt); next_pe_start += (addr & mask) * 8;drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +1
-1@@ -1196,7 +1198,7 @@ static int amdgpu_vm_bo_update_mapping(structamdgpu_device *adev, * reserve space for one command every (1 << BLOCK_SIZE) * entries or 2k dwords (whatever is smaller) */ - ncmds = (nptes >>min (amdgpu_vm_block_size,11 )) +1 ; + ncmds = (nptes >>min (adev->vm_manager.block_size,11 )) +1 ;/* padding, etc. */ ndw = 64;drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +1
-1@@ -2067,7 +2069,7 @@ void amdgpu_vm_bo_invalidate(structamdgpu_device *adev, int amdgpu_vm_init(struct amdgpu_device *adev, struct amdgpu_vm *vm) { const unsigned align = min(AMDGPU_VM_PTB_ALIGN_SIZE, -AMDGPU_VM_PTE_COUNT *8 ); +AMDGPU_VM_PTE_COUNT (adev) *8 );unsigned ring_instance;struct amdgpu_ring *ring;struct amd_sched_rq *rq;drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h +0
-0@@ -45,7 +45,7 @@#define AMDGPU_VM_MAX_UPDATE_SIZE 0x3FFFF
Quoted 7 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h
Quoted 4 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h +1
-1/* number of entries in page table */ -#define AMDGPU_VM_PTE_COUNT (1 << amdgpu_vm_block_size) +#define AMDGPU_VM_PTE_COUNT (adev) (1 << adev- >vm_manager.block_size)For safety, put parens around adev. E.g.,
#defineAMDGPU_VM_PTE_COUNT(adev)(1 << (adev)->vm_manager.block_size)
Yeah, thanks to point it out.
Quoted 7 lines
With that fixed:
Reviewed-by Alex Deucher <[email protected]>As a follow on patch, it would be nice to add some reasonable clamping in the gmc files so we can still override the default settings and not cause too much problem with multiple asics in the same system.
Could you elaborate it a bit more, about "clamping"?
Jerry
Quoted 189 lines
Alex
Quoted 186 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h +2
-0/* PTBs (Page Table Blocks) need to be aligned to 32K */ #define AMDGPU_VM_PTB_ALIGN_SIZE 32768@@ -155,6 +155,8 @@ struct amdgpu_vm_manager {uint64_t max_pfn;uint32_t num_level; +uint64_t vm_size; +uint32_t block_size;/* vram base address for page table entry */ u64 vram_base_offset;/* is vm enabled? */ drivers/gpu/drm/amd/amdgpu/gfxhub_v1_0.c +1
-1@@ -222,7 +222,7 @@ int gfxhub_v1_0_gart_enable(struct amdgpu_device *adev) EXECUTE_PROTECTION_FAULT_ENABLE_DEFAULT, 1); tmp =REG_SET_FIELD (tmp,VM_CONTEXT1_CNTL ,PAGE_TABLE_BLOCK_SIZE , - amdgpu_vm_block_size -9 ); + adev->vm_manager.block_size -9 ); WREG32(SOC15_REG_OFFSET(GC, 0,mmVM_CONTEXT1_CNTL) + i, tmp); WREG32(SOC15_REG_OFFSET(GC, 0, mmVM_CONTEXT1_PAGE_TABLE_START_ADDR_LO32) + i*2, 0); WREG32(SOC15_REG_OFFSET(GC, 0,
mmVM_CONTEXT1_PAGE_TABLE_START_ADDR_HI32) + i*2, 0);drivers/gpu/drm/amd/amdgpu/gmc_v6_0.c +2
-1@@ -543,7 +543,8 @@ static int gmc_v6_0_gart_enable(struct amdgpu_device *adev) WREG32(mmVM_CONTEXT1_CNTL, VM_CONTEXT1_CNTL__ENABLE_CONTEXT_MASK | (1UL << VM_CONTEXT1_CNTL__PAGE_TABLE_DEPTH__SHIFT) | - ((amdgpu_vm_block_size - 9) << VM_CONTEXT1_CNTL__PAGE_TABLE_BLOCK_SIZE__SHIFT)); + ((adev->vm_manager.block_size - 9) + << VM_CONTEXT1_CNTL__PAGE_TABLE_BLOCK_SIZE__SHIFT)); if (amdgpu_vm_fault_stop ==
AMDGPU_VM_FAULT_STOP_ALWAYS)gmc_v6_0_set_fault_enable_default(adev, false); elsedrivers/gpu/drm/amd/amdgpu/gmc_v6_0.c +6
-1@@ -848,7 +849,12 @@ static int gmc_v6_0_sw_init(void *handle)if (r)return r; - adev->vm_manager.max_pfn = amdgpu_vm_size <<18 ; + adev->vm_manager.vm_size = amdgpu_vm_size; + adev->vm_manager.block_size = amdgpu_vm_block_size; + adev->vm_manager.max_pfn = adev->vm_manager.vm_size <<18 ; + +DRM_INFO ("vm size is %d GB, block size is %d-bit\n" , + adev->vm_manager.vm_size, adev- >vm_manager.block_size); adev->mc.mc_mask =0xffffffffffULL ;drivers/gpu/drm/amd/amdgpu/gmc_v7_0.c +1
-1@@ -639,7 +639,7 @@ static int gmc_v7_0_gart_enable(struct amdgpu_device *adev) tmp =REG_SET_FIELD (tmp,VM_CONTEXT1_CNTL ,ENABLE_CONTEXT ,1 ); tmp =REG_SET_FIELD (tmp,VM_CONTEXT1_CNTL ,PAGE_TABLE_DEPTH ,1 ); tmp =REG_SET_FIELD (tmp,VM_CONTEXT1_CNTL ,PAGE_TABLE_BLOCK_SIZE , - amdgpu_vm_block_size -9 ); + adev->vm_manager.block_size -9 );WREG32 (mmVM_CONTEXT1_CNTL, tmp); if (amdgpu_vm_fault_stop ==
AMDGPU_VM_FAULT_STOP_ALWAYS)gmc_v7_0_set_fault_enable_default(adev, false);drivers/gpu/drm/amd/amdgpu/gmc_v7_0.c +6
-1@@ -998,7 +998,12 @@ static int gmc_v7_0_sw_init(void *handle)* Currently set to 4GB ((1 << 20) 4k pages). * Max GPUVM size for cayman and SI is 40 bits. */ - adev->vm_manager.max_pfn = amdgpu_vm_size <<18 ; + adev->vm_manager.vm_size = amdgpu_vm_size; + adev->vm_manager.block_size = amdgpu_vm_block_size; + adev->vm_manager.max_pfn = adev->vm_manager.vm_size <<18 ; + +DRM_INFO ("vm size is %d GB, block size is %d-bit\n" , + adev->vm_manager.vm_size, adev- >vm_manager.block_size);/* Set the internal MC address mask * This is the max address of the GPU's
drivers/gpu/drm/amd/amdgpu/gmc_v8_0.c +1
-1@@ -848,7 +848,7 @@ static int gmc_v8_0_gart_enable(struct amdgpu_device *adev) tmp =REG_SET_FIELD (tmp,VM_CONTEXT1_CNTL ,WRITE_PROTECTION_FAULT_ENABLE_DEFAULT ,1 ); tmp =REG_SET_FIELD (tmp,VM_CONTEXT1_CNTL ,EXECUTE_PROTECTION_FAULT_ENABLE_DEFAULT ,1 ); tmp =REG_SET_FIELD (tmp,VM_CONTEXT1_CNTL ,PAGE_TABLE_BLOCK_SIZE , - amdgpu_vm_block_size -9 ); + adev->vm_manager.block_size -9 );WREG32 (mmVM_CONTEXT1_CNTL, tmp); if (amdgpu_vm_fault_stop ==
AMDGPU_VM_FAULT_STOP_ALWAYS)gmc_v8_0_set_fault_enable_default(adev, false);drivers/gpu/drm/amd/amdgpu/gmc_v8_0.c +6
-1@@ -1082,7 +1082,12 @@ static int gmc_v8_0_sw_init(void *handle)* Currently set to 4GB ((1 << 20) 4k pages). * Max GPUVM size for cayman and SI is 40 bits. */ - adev->vm_manager.max_pfn = amdgpu_vm_size <<18 ; + adev->vm_manager.vm_size = amdgpu_vm_size; + adev->vm_manager.block_size = amdgpu_vm_block_size; + adev->vm_manager.max_pfn = adev->vm_manager.vm_size <<18 ; + +DRM_INFO ("vm size is %d GB, block size is %d-bit\n" , + adev->vm_manager.vm_size, adev- >vm_manager.block_size);/* Set the internal MC address mask * This is the max address of the GPU's
drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c +12
-0@@ -543,11 +543,23 @@ static int gmc_v9_0_sw_init(void *handle)if (adev->flags &AMD_IS_APU ) { adev->mc.vram_type =AMDGPU_VRAM_TYPE_UNKNOWN ; + adev->vm_manager.vm_size = amdgpu_vm_size; + adev->vm_manager.block_size = amdgpu_vm_block_size; }else {/* XXX Don't know how to get VRAM type yet. */ adev->mc.vram_type =AMDGPU_VRAM_TYPE_HBM ; +/* +* To fulfill 4-level page support, +* vm size is 256TB (48bit), maximum size of Vega10, +* block size 512 (9bit) +*/ + adev->vm_manager.vm_size =1U <<18 ; + adev->vm_manager.block_size =9 ; } +DRM_INFO ("vm size is %d GB, block size is %d-bit\n" , + adev->vm_manager.vm_size, adev- >vm_manager.block_size); +/* This interrupt is VMC page fault.*/ r = amdgpu_irq_add_id(adev, AMDGPU_IH_CLIENTID_VMC, 0,&adev->mc.vm_fault);drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c +1
-8@@ -557,14 +569,7 @@ static int gmc_v9_0_sw_init(void *handle)if (r)return r; -/* Because of four level VMPTs, vm size is at least 512GB. -* The maximum size is 256TB (48bit). -*/ -if (amdgpu_vm_size <512 ) { -DRM_WARN ("VM size is at least 512GB!\n" ); - amdgpu_vm_size =512 ; - } - adev->vm_manager.max_pfn = (uint64_t )amdgpu_vm_size <<18 ; + adev->vm_manager.max_pfn = adev->vm_manager.vm_size <<18 ;/* Set the internal MC address mask * This is the max address of the GPU's drivers/gpu/drm/amd/amdgpu/mmhub_v1_0.c +0
-0@@ -242,7 +242,7 @@ int mmhub_v1_0_gart_enable(struct amdgpu_device *adev)
Quoted 14 lines
Quoted 10 lines
drivers/gpu/drm/amd/amdgpu/mmhub_v1_0.c +1
-1EXECUTE_PROTECTION_FAULT_ENABLE_DEFAULT, 1); tmp =REG_SET_FIELD (tmp,VM_CONTEXT1_CNTL ,PAGE_TABLE_BLOCK_SIZE , - amdgpu_vm_block_size -9 ); + adev->vm_manager.block_size -9 ); WREG32(SOC15_REG_OFFSET(MMHUB, 0,mmVM_CONTEXT1_CNTL) + i, tmp); WREG32(SOC15_REG_OFFSET(MMHUB, 0, mmVM_CONTEXT1_PAGE_TABLE_START_ADDR_LO32) + i*2, 0); WREG32(SOC15_REG_OFFSET(MMHUB, 0,Quoted 3 lines
mmVM_CONTEXT1_PAGE_TABLE_START_ADDR_HI32) + i*2, 0); --
1.9.1
_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
A message that is not in the archive (sent off-list or not yet received): [email protected]
Quoted 170 lines · Jerry (Junwei)
Quoted 153 lines · Alexander
Quoted 153 lines
-----Original Message-----
From: Junwei Zhang [mailto:[email protected]]
Sent: Friday, March 31, 2017 10:44 PM
To: Deucher, Alexander; Koenig, Christian
Cc: [email protected]; Zhang, Jerry
Subject: [PATCH] drm/amdgpu: fix vm size and block size for VMPT (v3)From: "Zhang, Jerry" <[email protected]>
v2: set both of them in gmc
v3: move vm size and block size in vm managerSigned-off-by Junwei Zhang <[email protected]>--- drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 8 -------- drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 22 ++++++++++++---------(quoted diff) +0
-1- drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h | 4 +++- drivers/gpu/drm/amd/amdgpu/gfxhub_v1_0.c | 2 +- drivers/gpu/drm/amd/amdgpu/gmc_v6_0.c | 10 ++++++++-- drivers/gpu/drm/amd/amdgpu/gmc_v7_0.c | 9 +++++++-- drivers/gpu/drm/amd/amdgpu/gmc_v8_0.c | 9 +++++++-- drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c | 21 +++++++++++++-------- drivers/gpu/drm/amd/amdgpu/mmhub_v1_0.c | 2 +- 9 files changed, 52 insertions(+), 35 deletions(-)drivers/gpu/drm/amd/amdgpu/amdgpu_device.c +0
-8@@ -1041,14 +1041,6 @@ static bool amdgpu_check_pot_argument(int arg)static void amdgpu_get_block_size (struct amdgpu_device *adev) { -/* from AI, asic starts to support multiple level VMPT */ -if (adev->asic_type >=CHIP_VEGA10 ) { -if (amdgpu_vm_block_size !=9 ) -dev_warn (adev->dev, -"Multi-VMPT limits block size to one page!\n"); - amdgpu_vm_block_size =9 ; -return ; - }/* defines number of bits in page table versus page directory, * a page is 4KB so we have 12 bits offset, minimum 9 bits in the * page table and the remaining bits are in the page directory */
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +4
-3@@ -100,13 +100,14 @@ static unsigned amdgpu_vm_num_entries(struct amdgpu_device *adev, if (level == 0) /* For the root directory */return adev->vm_manager.max_pfn >> - (amdgpu_vm_block_size * adev- >vm_manager.num_level); + (adev->vm_manager.block_size * + adev->vm_manager.num_level); else if (level == adev->vm_manager.num_level) /* For the page tables on the leaves */ - return AMDGPU_VM_PTE_COUNT; + return AMDGPU_VM_PTE_COUNT(adev); else/* Everything in between */ - return 1 << amdgpu_vm_block_size; + return 1 << adev->vm_manager.block_size; }/**
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +2
-2@@ -271,7 +272,7 @@ static int amdgpu_vm_alloc_levels(structamdgpu_device *adev, unsigned level) { unsigned shift = (adev->vm_manager.num_level - level) * - amdgpu_vm_block_size; + adev->vm_manager.block_size;unsigned pt_idx, from, to;int r;@@ -970,7 +971,7 @@ static struct amdgpu_bo *amdgpu_vm_get_pt(structamdgpu_pte_update_params *p, unsigned idx, level = p->adev->vm_manager.num_level;while (entry->entries) { - idx = addr >> (amdgpu_vm_block_size * level--); + idx = addr >> (p->adev->vm_manager.block_size * level--); idx %=amdgpu_bo_size (entry->bo) /8 ; entry = &entry->entries[idx];}drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +2
-1@@ -997,7 +998,8 @@ static void amdgpu_vm_update_ptes(structamdgpu_pte_update_params *params, uint64_t start, uint64_t end, uint64_t dst, uint64_t flags) { - const uint64_t mask = AMDGPU_VM_PTE_COUNT - 1; + struct amdgpu_device *adev = params->adev; +const uint64_t mask =AMDGPU_VM_PTE_COUNT (adev) -1 ;uint64_t cur_pe_start, cur_nptes, cur_dst;
uint64_taddr; /* next GPU address to be updated */drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +1
-1@@ -1021,7 +1023,7 @@ static void amdgpu_vm_update_ptes(struct amdgpu_pte_update_params *params, if ((addr & ~mask) == (end & ~mask)) nptes = end - addr; else - nptes =AMDGPU_VM_PTE_COUNT - (addr & mask); + nptes =AMDGPU_VM_PTE_COUNT (adev) - (addr & mask); cur_pe_start =amdgpu_bo_gpu_offset (pt);cur_pe_start += (addr & mask) * 8;drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +1
-1@@ -1049,7 +1051,7 @@ static void amdgpu_vm_update_ptes(struct amdgpu_pte_update_params *params, if ((addr & ~mask) == (end & ~mask)) nptes = end - addr; else - nptes =AMDGPU_VM_PTE_COUNT - (addr & mask); + nptes =AMDGPU_VM_PTE_COUNT (adev) - (addr & mask);next_pe_start = amdgpu_bo_gpu_offset(pt); next_pe_start += (addr & mask) * 8;drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +1
-1@@ -1196,7 +1198,7 @@ static int amdgpu_vm_bo_update_mapping(structamdgpu_device *adev, * reserve space for one command every (1 << BLOCK_SIZE) * entries or 2k dwords (whatever is smaller) */ - ncmds = (nptes >>min (amdgpu_vm_block_size,11 )) +1 ; + ncmds = (nptes >>min (adev->vm_manager.block_size,11 )) +1 ;/* padding, etc. */ ndw = 64;drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +1
-1@@ -2067,7 +2069,7 @@ void amdgpu_vm_bo_invalidate(structamdgpu_device *adev, int amdgpu_vm_init(struct amdgpu_device *adev, struct amdgpu_vm *vm) { const unsigned align = min(AMDGPU_VM_PTB_ALIGN_SIZE, -AMDGPU_VM_PTE_COUNT *8 ); +AMDGPU_VM_PTE_COUNT (adev) *8 );unsigned ring_instance;struct amdgpu_ring *ring;struct amd_sched_rq *rq;drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h +0
-0@@ -45,7 +45,7 @@#define AMDGPU_VM_MAX_UPDATE_SIZE 0x3FFFFQuoted 8 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h
Quoted 4 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h +1
-1/* number of entries in page table */ -#define AMDGPU_VM_PTE_COUNT (1 << amdgpu_vm_block_size) +#define AMDGPU_VM_PTE_COUNT (adev) (1 << adev- >vm_manager.block_size)For safety, put parens around adev. E.g.,
#defineAMDGPU_VM_PTE_COUNT(adev)(1 << (adev)->vm_manager.block_size)Yeah, thanks to point it out.
Quoted 4 lines
With that fixed:
Reviewed-by Alex Deucher <[email protected]>
Quoted 8 lines
Quoted 5 lines
As a follow on patch, it would be nice to add some reasonable clamping in the
gmc files so we can still override the default settings and not cause too much
problem with multiple asics in the same system.Could you elaborate it a bit more, about "clamping"?
e.g., setting amdgpu_vm_block_size to -1 (auto) bt default and setting a reasonable default for each gmc if it's set to 0, or overriding if it's not set to -1. That way we can have different defaults on each family and still override for testing/debugging.
Alex
Quoted 207 lines
Jerry
Quoted 189 lines
Alex
Quoted 186 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h +2
-0/* PTBs (Page Table Blocks) need to be aligned to 32K */ #define AMDGPU_VM_PTB_ALIGN_SIZE 32768@@ -155,6 +155,8 @@ struct amdgpu_vm_manager {uint64_t max_pfn;uint32_t num_level; +uint64_t vm_size; +uint32_t block_size;/* vram base address for page table entry */ u64 vram_base_offset;/* is vm enabled? */ drivers/gpu/drm/amd/amdgpu/gfxhub_v1_0.c +1
-1@@ -222,7 +222,7 @@ int gfxhub_v1_0_gart_enable(struct amdgpu_device *adev) EXECUTE_PROTECTION_FAULT_ENABLE_DEFAULT, 1); tmp =REG_SET_FIELD (tmp,VM_CONTEXT1_CNTL ,PAGE_TABLE_BLOCK_SIZE , - amdgpu_vm_block_size -9 ); + adev->vm_manager.block_size -9 ); WREG32(SOC15_REG_OFFSET(GC, 0,mmVM_CONTEXT1_CNTL) + i, tmp); WREG32(SOC15_REG_OFFSET(GC, 0, mmVM_CONTEXT1_PAGE_TABLE_START_ADDR_LO32) + i*2, 0); WREG32(SOC15_REG_OFFSET(GC, 0,
mmVM_CONTEXT1_PAGE_TABLE_START_ADDR_HI32) + i*2, 0);drivers/gpu/drm/amd/amdgpu/gmc_v6_0.c +2
-1@@ -543,7 +543,8 @@ static int gmc_v6_0_gart_enable(struct amdgpu_device *adev) WREG32(mmVM_CONTEXT1_CNTL, VM_CONTEXT1_CNTL__ENABLE_CONTEXT_MASK | (1UL << VM_CONTEXT1_CNTL__PAGE_TABLE_DEPTH__SHIFT) | - ((amdgpu_vm_block_size - 9) << VM_CONTEXT1_CNTL__PAGE_TABLE_BLOCK_SIZE__SHIFT)); + ((adev->vm_manager.block_size - 9) + << VM_CONTEXT1_CNTL__PAGE_TABLE_BLOCK_SIZE__SHIFT)); if (amdgpu_vm_fault_stop ==
AMDGPU_VM_FAULT_STOP_ALWAYS)gmc_v6_0_set_fault_enable_default(adev, false); elsedrivers/gpu/drm/amd/amdgpu/gmc_v6_0.c +6
-1@@ -848,7 +849,12 @@ static int gmc_v6_0_sw_init(void *handle)if (r)return r; - adev->vm_manager.max_pfn = amdgpu_vm_size <<18 ; + adev->vm_manager.vm_size = amdgpu_vm_size; + adev->vm_manager.block_size = amdgpu_vm_block_size; + adev->vm_manager.max_pfn = adev->vm_manager.vm_size <<18 ; + +DRM_INFO ("vm size is %d GB, block size is %d-bit\n" , + adev->vm_manager.vm_size, adev- >vm_manager.block_size); adev->mc.mc_mask =0xffffffffffULL ;drivers/gpu/drm/amd/amdgpu/gmc_v7_0.c +1
-1@@ -639,7 +639,7 @@ static int gmc_v7_0_gart_enable(struct amdgpu_device *adev) tmp =REG_SET_FIELD (tmp,VM_CONTEXT1_CNTL ,ENABLE_CONTEXT ,1 ); tmp =REG_SET_FIELD (tmp,VM_CONTEXT1_CNTL ,PAGE_TABLE_DEPTH ,1 ); tmp =REG_SET_FIELD (tmp,VM_CONTEXT1_CNTL ,PAGE_TABLE_BLOCK_SIZE , - amdgpu_vm_block_size -9 ); + adev->vm_manager.block_size -9 );WREG32 (mmVM_CONTEXT1_CNTL, tmp); if (amdgpu_vm_fault_stop ==
AMDGPU_VM_FAULT_STOP_ALWAYS)gmc_v7_0_set_fault_enable_default(adev, false);drivers/gpu/drm/amd/amdgpu/gmc_v7_0.c +6
-1@@ -998,7 +998,12 @@ static int gmc_v7_0_sw_init(void *handle)* Currently set to 4GB ((1 << 20) 4k pages). * Max GPUVM size for cayman and SI is 40 bits. */ - adev->vm_manager.max_pfn = amdgpu_vm_size <<18 ; + adev->vm_manager.vm_size = amdgpu_vm_size; + adev->vm_manager.block_size = amdgpu_vm_block_size; + adev->vm_manager.max_pfn = adev->vm_manager.vm_size <<18 ; + +DRM_INFO ("vm size is %d GB, block size is %d-bit\n" , + adev->vm_manager.vm_size, adev- >vm_manager.block_size);/* Set the internal MC address mask * This is the max address of the GPU's
drivers/gpu/drm/amd/amdgpu/gmc_v8_0.c +1
-1@@ -848,7 +848,7 @@ static int gmc_v8_0_gart_enable(struct amdgpu_device *adev) tmp =REG_SET_FIELD (tmp,VM_CONTEXT1_CNTL ,WRITE_PROTECTION_FAULT_ENABLE_DEFAULT ,1 ); tmp =REG_SET_FIELD (tmp,VM_CONTEXT1_CNTL ,EXECUTE_PROTECTION_FAULT_ENABLE_DEFAULT ,1 ); tmp =REG_SET_FIELD (tmp,VM_CONTEXT1_CNTL ,PAGE_TABLE_BLOCK_SIZE , - amdgpu_vm_block_size -9 ); + adev->vm_manager.block_size -9 );WREG32 (mmVM_CONTEXT1_CNTL, tmp); if (amdgpu_vm_fault_stop ==
AMDGPU_VM_FAULT_STOP_ALWAYS)gmc_v8_0_set_fault_enable_default(adev, false);drivers/gpu/drm/amd/amdgpu/gmc_v8_0.c +6
-1@@ -1082,7 +1082,12 @@ static int gmc_v8_0_sw_init(void *handle)* Currently set to 4GB ((1 << 20) 4k pages). * Max GPUVM size for cayman and SI is 40 bits. */ - adev->vm_manager.max_pfn = amdgpu_vm_size <<18 ; + adev->vm_manager.vm_size = amdgpu_vm_size; + adev->vm_manager.block_size = amdgpu_vm_block_size; + adev->vm_manager.max_pfn = adev->vm_manager.vm_size <<18 ; + +DRM_INFO ("vm size is %d GB, block size is %d-bit\n" , + adev->vm_manager.vm_size, adev- >vm_manager.block_size);/* Set the internal MC address mask * This is the max address of the GPU's
drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c +12
-0@@ -543,11 +543,23 @@ static int gmc_v9_0_sw_init(void *handle)if (adev->flags &AMD_IS_APU ) { adev->mc.vram_type =AMDGPU_VRAM_TYPE_UNKNOWN ; + adev->vm_manager.vm_size = amdgpu_vm_size; + adev->vm_manager.block_size = amdgpu_vm_block_size; }else {/* XXX Don't know how to get VRAM type yet. */ adev->mc.vram_type =AMDGPU_VRAM_TYPE_HBM ; +/* +* To fulfill 4-level page support, +* vm size is 256TB (48bit), maximum size of Vega10, +* block size 512 (9bit) +*/ + adev->vm_manager.vm_size =1U <<18 ; + adev->vm_manager.block_size =9 ; } +DRM_INFO ("vm size is %d GB, block size is %d-bit\n" , + adev->vm_manager.vm_size, adev- >vm_manager.block_size); +/* This interrupt is VMC page fault.*/ r = amdgpu_irq_add_id(adev, AMDGPU_IH_CLIENTID_VMC, 0,&adev->mc.vm_fault);drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c +1
-8@@ -557,14 +569,7 @@ static int gmc_v9_0_sw_init(void *handle)if (r)return r; -/* Because of four level VMPTs, vm size is at least 512GB. -* The maximum size is 256TB (48bit). -*/ -if (amdgpu_vm_size <512 ) { -DRM_WARN ("VM size is at least 512GB!\n" ); - amdgpu_vm_size =512 ; - } - adev->vm_manager.max_pfn = (uint64_t )amdgpu_vm_size <<18 ; + adev->vm_manager.max_pfn = adev->vm_manager.vm_size <<18 ;/* Set the internal MC address mask * This is the max address of the GPU's drivers/gpu/drm/amd/amdgpu/mmhub_v1_0.c +0
-0@@ -242,7 +242,7 @@ int mmhub_v1_0_gart_enable(struct amdgpu_device *adev)Quoted 15 lines
Quoted 10 lines
drivers/gpu/drm/amd/amdgpu/mmhub_v1_0.c +1
-1EXECUTE_PROTECTION_FAULT_ENABLE_DEFAULT, 1); tmp =REG_SET_FIELD (tmp,VM_CONTEXT1_CNTL ,PAGE_TABLE_BLOCK_SIZE , - amdgpu_vm_block_size -9 ); + adev->vm_manager.block_size -9 ); WREG32(SOC15_REG_OFFSET(MMHUB, 0,mmVM_CONTEXT1_CNTL) + i, tmp); WREG32(SOC15_REG_OFFSET(MMHUB, 0, mmVM_CONTEXT1_PAGE_TABLE_START_ADDR_LO32) + i*2, 0); WREG32(SOC15_REG_OFFSET(MMHUB, 0,Quoted 3 lines
mmVM_CONTEXT1_PAGE_TABLE_START_ADDR_HI32) + i*2, 0); --
1.9.1
Quoted 4 lines
_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
A message that is not in the archive (sent off-list or not yet received): CADnq5_Oae0F4L9UzEKZvzYbo0A1LPPXBvOc6GSKBq4VN3E=B3A-JsoAwUIsXosN+BqQ9rBEUg@public.gmane.org
Quoted 179 lines · Alex Deucher
Quoted 177 lines · Jerry (Junwei)
Quoted 153 lines · Alexander
Quoted 153 lines
-----Original Message-----
From: Junwei Zhang [mailto:[email protected]]
Sent: Friday, March 31, 2017 10:44 PM
To: Deucher, Alexander; Koenig, Christian
Cc: [email protected]; Zhang, Jerry
Subject: [PATCH] drm/amdgpu: fix vm size and block size for VMPT (v3)From: "Zhang, Jerry" <[email protected]>
v2: set both of them in gmc
v3: move vm size and block size in vm managerSigned-off-by Junwei Zhang <[email protected]>--- drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 8 -------- drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 22 ++++++++++++---------(quoted diff) +0
-1- drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h | 4 +++- drivers/gpu/drm/amd/amdgpu/gfxhub_v1_0.c | 2 +- drivers/gpu/drm/amd/amdgpu/gmc_v6_0.c | 10 ++++++++-- drivers/gpu/drm/amd/amdgpu/gmc_v7_0.c | 9 +++++++-- drivers/gpu/drm/amd/amdgpu/gmc_v8_0.c | 9 +++++++-- drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c | 21 +++++++++++++-------- drivers/gpu/drm/amd/amdgpu/mmhub_v1_0.c | 2 +- 9 files changed, 52 insertions(+), 35 deletions(-)drivers/gpu/drm/amd/amdgpu/amdgpu_device.c +0
-8@@ -1041,14 +1041,6 @@ static bool amdgpu_check_pot_argument(int arg)static void amdgpu_get_block_size (struct amdgpu_device *adev) { -/* from AI, asic starts to support multiple level VMPT */ -if (adev->asic_type >=CHIP_VEGA10 ) { -if (amdgpu_vm_block_size !=9 ) -dev_warn (adev->dev, -"Multi-VMPT limits block size to one page!\n"); - amdgpu_vm_block_size =9 ; -return ; - }/* defines number of bits in page table versus page directory, * a page is 4KB so we have 12 bits offset, minimum 9 bits in the * page table and the remaining bits are in the page directory */
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +4
-3@@ -100,13 +100,14 @@ static unsigned amdgpu_vm_num_entries(struct amdgpu_device *adev, if (level == 0) /* For the root directory */return adev->vm_manager.max_pfn >> - (amdgpu_vm_block_size * adev- > vm_manager.num_level); + (adev->vm_manager.block_size * + adev->vm_manager.num_level); else if (level == adev->vm_manager.num_level) /* For the page tables on the leaves */ - return AMDGPU_VM_PTE_COUNT; + return AMDGPU_VM_PTE_COUNT(adev); else/* Everything in between */ - return 1 << amdgpu_vm_block_size; + return 1 << adev->vm_manager.block_size; }/**
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +2
-2@@ -271,7 +272,7 @@ static int amdgpu_vm_alloc_levels(structamdgpu_device *adev, unsigned level) { unsigned shift = (adev->vm_manager.num_level - level) * - amdgpu_vm_block_size; + adev->vm_manager.block_size;unsigned pt_idx, from, to;int r;@@ -970,7 +971,7 @@ static struct amdgpu_bo *amdgpu_vm_get_pt(structamdgpu_pte_update_params *p, unsigned idx, level = p->adev->vm_manager.num_level;while (entry->entries) { - idx = addr >> (amdgpu_vm_block_size * level--); + idx = addr >> (p->adev->vm_manager.block_size * level--); idx %=amdgpu_bo_size (entry->bo) /8 ; entry = &entry->entries[idx];}drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +2
-1@@ -997,7 +998,8 @@ static void amdgpu_vm_update_ptes(structamdgpu_pte_update_params *params, uint64_t start, uint64_t end, uint64_t dst, uint64_t flags) { - const uint64_t mask = AMDGPU_VM_PTE_COUNT - 1; + struct amdgpu_device *adev = params->adev; +const uint64_t mask =AMDGPU_VM_PTE_COUNT (adev) -1 ;uint64_t cur_pe_start, cur_nptes, cur_dst;
uint64_taddr; /* next GPU address to be updated */drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +1
-1@@ -1021,7 +1023,7 @@ static void amdgpu_vm_update_ptes(struct amdgpu_pte_update_params *params, if ((addr & ~mask) == (end & ~mask)) nptes = end - addr; else - nptes =AMDGPU_VM_PTE_COUNT - (addr & mask); + nptes =AMDGPU_VM_PTE_COUNT (adev) - (addr & mask); cur_pe_start =amdgpu_bo_gpu_offset (pt);cur_pe_start += (addr & mask) * 8;drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +1
-1@@ -1049,7 +1051,7 @@ static void amdgpu_vm_update_ptes(struct amdgpu_pte_update_params *params, if ((addr & ~mask) == (end & ~mask)) nptes = end - addr; else - nptes =AMDGPU_VM_PTE_COUNT - (addr & mask); + nptes =AMDGPU_VM_PTE_COUNT (adev) - (addr & mask);next_pe_start = amdgpu_bo_gpu_offset(pt); next_pe_start += (addr & mask) * 8;drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +1
-1@@ -1196,7 +1198,7 @@ static int amdgpu_vm_bo_update_mapping(structamdgpu_device *adev, * reserve space for one command every (1 << BLOCK_SIZE) * entries or 2k dwords (whatever is smaller) */ - ncmds = (nptes >>min (amdgpu_vm_block_size,11 )) +1 ; + ncmds = (nptes >>min (adev->vm_manager.block_size,11 )) +1 ;/* padding, etc. */ ndw = 64;drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +1
-1@@ -2067,7 +2069,7 @@ void amdgpu_vm_bo_invalidate(structamdgpu_device *adev, int amdgpu_vm_init(struct amdgpu_device *adev, struct amdgpu_vm *vm) { const unsigned align = min(AMDGPU_VM_PTB_ALIGN_SIZE, -AMDGPU_VM_PTE_COUNT *8 ); +AMDGPU_VM_PTE_COUNT (adev) *8 );unsigned ring_instance;struct amdgpu_ring *ring;struct amd_sched_rq *rq;drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h +0
-0@@ -45,7 +45,7 @@#define AMDGPU_VM_MAX_UPDATE_SIZE 0x3FFFFQuoted 8 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h line 49
Quoted 4 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h +1
-1/* number of entries in page table */ -#define AMDGPU_VM_PTE_COUNT (1 << amdgpu_vm_block_size) +#define AMDGPU_VM_PTE_COUNT (adev) (1 << adev- > vm_manager.block_size)For safety, put parens around adev. E.g.,
#defineAMDGPU_VM_PTE_COUNT(adev)(1 << (adev)->vm_manager.block_size)Yeah, thanks to point it out.
Quoted 9 lines
With that fixed:
Reviewed-by Alex Deucher <[email protected]>As a follow on patch, it would be nice to add some reasonable clamping in the
gmc files so we can still override the default settings and not cause too much
problem with multiple asics in the same system.
Quoted 6 lines
Quoted 1 line
Could you elaborate it a bit more, about "clamping"?
e.g., setting
amdgpu_vm_block_sizeto -1 (auto) bt default and setting a reasonable default for each gmc if it's set to 0, or overriding if it's not set to -1. That way we can have different defaults on each family and still override for testing/debugging.
Thanks for your explanation.
Now I got the "clamping" :)
Jerry
Quoted 215 lines
Alex
Quoted 211 lines
Jerry
Quoted 189 lines
Alex
Quoted 186 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h +2
-0/* PTBs (Page Table Blocks) need to be aligned to 32K */ #define AMDGPU_VM_PTB_ALIGN_SIZE 32768@@ -155,6 +155,8 @@ struct amdgpu_vm_manager {uint64_t max_pfn;uint32_t num_level; +uint64_t vm_size; +uint32_t block_size;/* vram base address for page table entry */ u64 vram_base_offset;/* is vm enabled? */ drivers/gpu/drm/amd/amdgpu/gfxhub_v1_0.c +1
-1@@ -222,7 +222,7 @@ int gfxhub_v1_0_gart_enable(struct amdgpu_device *adev) EXECUTE_PROTECTION_FAULT_ENABLE_DEFAULT, 1); tmp =REG_SET_FIELD (tmp,VM_CONTEXT1_CNTL ,PAGE_TABLE_BLOCK_SIZE , - amdgpu_vm_block_size -9 ); + adev->vm_manager.block_size -9 ); WREG32(SOC15_REG_OFFSET(GC, 0,mmVM_CONTEXT1_CNTL) + i, tmp); WREG32(SOC15_REG_OFFSET(GC, 0, mmVM_CONTEXT1_PAGE_TABLE_START_ADDR_LO32) + i*2, 0); WREG32(SOC15_REG_OFFSET(GC, 0,
mmVM_CONTEXT1_PAGE_TABLE_START_ADDR_HI32) + i*2, 0);drivers/gpu/drm/amd/amdgpu/gmc_v6_0.c +2
-1@@ -543,7 +543,8 @@ static int gmc_v6_0_gart_enable(struct amdgpu_device *adev) WREG32(mmVM_CONTEXT1_CNTL, VM_CONTEXT1_CNTL__ENABLE_CONTEXT_MASK | (1UL << VM_CONTEXT1_CNTL__PAGE_TABLE_DEPTH__SHIFT) | - ((amdgpu_vm_block_size - 9) << VM_CONTEXT1_CNTL__PAGE_TABLE_BLOCK_SIZE__SHIFT)); + ((adev->vm_manager.block_size - 9) + << VM_CONTEXT1_CNTL__PAGE_TABLE_BLOCK_SIZE__SHIFT)); if (amdgpu_vm_fault_stop ==
AMDGPU_VM_FAULT_STOP_ALWAYS)gmc_v6_0_set_fault_enable_default(adev, false); elsedrivers/gpu/drm/amd/amdgpu/gmc_v6_0.c +6
-1@@ -848,7 +849,12 @@ static int gmc_v6_0_sw_init(void *handle)if (r)return r; - adev->vm_manager.max_pfn = amdgpu_vm_size <<18 ; + adev->vm_manager.vm_size = amdgpu_vm_size; + adev->vm_manager.block_size = amdgpu_vm_block_size; + adev->vm_manager.max_pfn = adev->vm_manager.vm_size <<18 ; + +DRM_INFO ("vm size is %d GB, block size is %d-bit\n" , + adev->vm_manager.vm_size, adev- > vm_manager.block_size); adev->mc.mc_mask =0xffffffffffULL ;drivers/gpu/drm/amd/amdgpu/gmc_v7_0.c +1
-1@@ -639,7 +639,7 @@ static int gmc_v7_0_gart_enable(struct amdgpu_device *adev) tmp =REG_SET_FIELD (tmp,VM_CONTEXT1_CNTL ,ENABLE_CONTEXT ,1 ); tmp =REG_SET_FIELD (tmp,VM_CONTEXT1_CNTL ,PAGE_TABLE_DEPTH ,1 ); tmp =REG_SET_FIELD (tmp,VM_CONTEXT1_CNTL ,PAGE_TABLE_BLOCK_SIZE , - amdgpu_vm_block_size -9 ); + adev->vm_manager.block_size -9 );WREG32 (mmVM_CONTEXT1_CNTL, tmp); if (amdgpu_vm_fault_stop ==
AMDGPU_VM_FAULT_STOP_ALWAYS)gmc_v7_0_set_fault_enable_default(adev, false);drivers/gpu/drm/amd/amdgpu/gmc_v7_0.c +6
-1@@ -998,7 +998,12 @@ static int gmc_v7_0_sw_init(void *handle)* Currently set to 4GB ((1 << 20) 4k pages). * Max GPUVM size for cayman and SI is 40 bits. */ - adev->vm_manager.max_pfn = amdgpu_vm_size <<18 ; + adev->vm_manager.vm_size = amdgpu_vm_size; + adev->vm_manager.block_size = amdgpu_vm_block_size; + adev->vm_manager.max_pfn = adev->vm_manager.vm_size <<18 ; + +DRM_INFO ("vm size is %d GB, block size is %d-bit\n" , + adev->vm_manager.vm_size, adev- > vm_manager.block_size);/* Set the internal MC address mask * This is the max address of the GPU's
drivers/gpu/drm/amd/amdgpu/gmc_v8_0.c +1
-1@@ -848,7 +848,7 @@ static int gmc_v8_0_gart_enable(struct amdgpu_device *adev) tmp =REG_SET_FIELD (tmp,VM_CONTEXT1_CNTL ,WRITE_PROTECTION_FAULT_ENABLE_DEFAULT ,1 ); tmp =REG_SET_FIELD (tmp,VM_CONTEXT1_CNTL ,EXECUTE_PROTECTION_FAULT_ENABLE_DEFAULT ,1 ); tmp =REG_SET_FIELD (tmp,VM_CONTEXT1_CNTL ,PAGE_TABLE_BLOCK_SIZE , - amdgpu_vm_block_size -9 ); + adev->vm_manager.block_size -9 );WREG32 (mmVM_CONTEXT1_CNTL, tmp); if (amdgpu_vm_fault_stop ==
AMDGPU_VM_FAULT_STOP_ALWAYS)gmc_v8_0_set_fault_enable_default(adev, false);drivers/gpu/drm/amd/amdgpu/gmc_v8_0.c +6
-1@@ -1082,7 +1082,12 @@ static int gmc_v8_0_sw_init(void *handle)* Currently set to 4GB ((1 << 20) 4k pages). * Max GPUVM size for cayman and SI is 40 bits. */ - adev->vm_manager.max_pfn = amdgpu_vm_size <<18 ; + adev->vm_manager.vm_size = amdgpu_vm_size; + adev->vm_manager.block_size = amdgpu_vm_block_size; + adev->vm_manager.max_pfn = adev->vm_manager.vm_size <<18 ; + +DRM_INFO ("vm size is %d GB, block size is %d-bit\n" , + adev->vm_manager.vm_size, adev- > vm_manager.block_size);/* Set the internal MC address mask * This is the max address of the GPU's
drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c +12
-0@@ -543,11 +543,23 @@ static int gmc_v9_0_sw_init(void *handle)if (adev->flags &AMD_IS_APU ) { adev->mc.vram_type =AMDGPU_VRAM_TYPE_UNKNOWN ; + adev->vm_manager.vm_size = amdgpu_vm_size; + adev->vm_manager.block_size = amdgpu_vm_block_size; }else {/* XXX Don't know how to get VRAM type yet. */ adev->mc.vram_type =AMDGPU_VRAM_TYPE_HBM ; +/* +* To fulfill 4-level page support, +* vm size is 256TB (48bit), maximum size of Vega10, +* block size 512 (9bit) +*/ + adev->vm_manager.vm_size =1U <<18 ; + adev->vm_manager.block_size =9 ; } +DRM_INFO ("vm size is %d GB, block size is %d-bit\n" , + adev->vm_manager.vm_size, adev- > vm_manager.block_size); +/* This interrupt is VMC page fault.*/ r = amdgpu_irq_add_id(adev, AMDGPU_IH_CLIENTID_VMC, 0,&adev->mc.vm_fault);drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c +1
-8@@ -557,14 +569,7 @@ static int gmc_v9_0_sw_init(void *handle)if (r)return r; -/* Because of four level VMPTs, vm size is at least 512GB. -* The maximum size is 256TB (48bit). -*/ -if (amdgpu_vm_size <512 ) { -DRM_WARN ("VM size is at least 512GB!\n" ); - amdgpu_vm_size =512 ; - } - adev->vm_manager.max_pfn = (uint64_t )amdgpu_vm_size <<18 ; + adev->vm_manager.max_pfn = adev->vm_manager.vm_size <<18 ;/* Set the internal MC address mask * This is the max address of the GPU's drivers/gpu/drm/amd/amdgpu/mmhub_v1_0.c +0
-0@@ -242,7 +242,7 @@ int mmhub_v1_0_gart_enable(struct amdgpu_device *adev)Quoted 15 lines
Quoted 10 lines
drivers/gpu/drm/amd/amdgpu/mmhub_v1_0.c +1
-1EXECUTE_PROTECTION_FAULT_ENABLE_DEFAULT, 1); tmp =REG_SET_FIELD (tmp,VM_CONTEXT1_CNTL ,PAGE_TABLE_BLOCK_SIZE , - amdgpu_vm_block_size -9 ); + adev->vm_manager.block_size -9 ); WREG32(SOC15_REG_OFFSET(MMHUB, 0,mmVM_CONTEXT1_CNTL) + i, tmp); WREG32(SOC15_REG_OFFSET(MMHUB, 0, mmVM_CONTEXT1_PAGE_TABLE_START_ADDR_LO32) + i*2, 0); WREG32(SOC15_REG_OFFSET(MMHUB, 0,Quoted 3 lines
mmVM_CONTEXT1_PAGE_TABLE_START_ADDR_HI32) + i*2, 0); --
1.9.1_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx