drm/xe: Make all xe_ggtt structs private.
Subsystems (3)
v6 cover drm/xe: Make all xe_ggtt structs private.
The previous approach iterated over the entire GGTT, but it's easier to hide even the xe_ggtt_node struct and only provide a xe_ggtt_node_addr() function.
While cleaning up, also remove xe_ggtt_node_alloc()/fini(), this can be handled in xe_ggtt_insert() instead as all the other users are already internal.
This makes the final cleanup the removal of xe_ggtt_node_allocated(), as it can be changed to a simple null check instead.
Maarten Lankhorst (12):
drm/xe: Start using ggtt->start in preparation of balloon removal drm/xe: Convert xe_fb_pin to use a callback for insertion into GGTT drm/xe: Add xe_ggtt_node_addr() to avoid dereferencing xe_ggtt_node drm/xe/display: Avoid dereferencing xe_ggtt_node
drm/xe: Do not dereference ggtt_node in xe_bo.c
drm/xe: Improve xe_gt_sriov_pf_config GGTT handling drm/xe: Privatize xe_ggtt_node
drm/xe: Make xe_ggtt_node offset relative to starting offset drm/xe: Rewrite GGTT VF initialisation
drm/xe: Move struct xe_ggtt to xe_ggtt.c
drm/xe: Make xe_ggtt_node_insert return a node
drm/xe: Remove xe_ggtt_node_allocated
.../gpu/drm/xe/compat-i915-headers/i915_vma.h | 4 +- drivers/gpu/drm/xe/display/xe_fb_pin.c | 114 ++--- drivers/gpu/drm/xe/display/xe_stolen.c | 2 +- drivers/gpu/drm/xe/tests/xe_guc_buf_kunit.c | 8 +- drivers/gpu/drm/xe/xe_bo.c | 6 +- drivers/gpu/drm/xe/xe_bo.h | 8 +- drivers/gpu/drm/xe/xe_device_types.h | 2 - drivers/gpu/drm/xe/xe_ggtt.c | 476 +++++++++--------- drivers/gpu/drm/xe/xe_ggtt.h | 27 +- drivers/gpu/drm/xe/xe_ggtt_types.h | 78 +-- drivers/gpu/drm/xe/xe_gt_sriov_pf_config.c | 53 +- .../gpu/drm/xe/xe_gt_sriov_pf_config_types.h | 8 +- drivers/gpu/drm/xe/xe_gt_sriov_vf.c | 5 +- drivers/gpu/drm/xe/xe_tile.c | 18 + drivers/gpu/drm/xe/xe_tile_sriov_vf.c | 197 +------- drivers/gpu/drm/xe/xe_tile_sriov_vf.h | 4 +- drivers/gpu/drm/xe/xe_tile_sriov_vf_types.h | 4 + 17 files changed, 402 insertions(+), 612 deletions(-)
Signature · 1 line
2.51.0
PATCH v6 1/12 drm/xe: Start using ggtt->start in preparation of balloon removal
Instead of having ggtt->size point to the end of ggtt, have ggtt->size be the actual size of the GGTT, and introduce ggtt->start to point to the beginning of GGTT.
This will allow a massive cleanup of GGTT in case of SRIOV-VF.
--- drivers/gpu/drm/xe/xe_ggtt.c | 82 ++++++++++++++-------- drivers/gpu/drm/xe/xe_ggtt.h | 2 + drivers/gpu/drm/xe/xe_ggtt_types.h | 4 +- drivers/gpu/drm/xe/xe_gt_sriov_pf_config.c | 4 +- 4 files changed, 59 insertions(+), 33 deletions(-)
drivers/gpu/drm/xe/xe_ggtt.c +52 -30
@@ -124,10 +124,32 @@ static void ggtt_update_access_counter(struct xe_ggtt *ggtt) } } +/** +* xe_ggtt_start - Get starting offset of GGTT. +* @ggtt: &xe_ggtt +* +* Returns: Starting offset for this &xe_ggtt. +*/ +u64 xe_ggtt_start (struct xe_ggtt *ggtt) +{ +return ggtt->start; +} + +/** +* xe_ggtt_size - Get size of GGTT. +* @ggtt: &xe_ggtt +* +* Returns: Total usable size of this &xe_ggtt. +*/ +u64 xe_ggtt_size (struct xe_ggtt *ggtt) +{ +return ggtt->size; +} +static void xe_ggtt_set_pte (struct xe_ggtt *ggtt,u64 addr,u64 pte) {xe_tile_assert (ggtt->tile, !(addr &XE_PTE_MASK )); -xe_tile_assert (ggtt->tile, addr < ggtt->size); +xe_tile_assert (ggtt->tile, addr < ggtt->start + ggtt->size);writeq (pte, &ggtt->gsm[addr >>XE_PTE_SHIFT ]); }
@@ -232,16 +254,16 @@ static const struct xe_ggtt_pt_ops xelpg_pt_wa_ops = { .ggtt_set_pte = xe_ggtt_set_pte_and_flush, }; -static void __xe_ggtt_init_early (struct xe_ggtt *ggtt,u32 reserved) +static void __xe_ggtt_init_early (struct xe_ggtt *ggtt,u64 start,u64 size) { -drm_mm_init (&ggtt->mm, reserved, - ggtt->size - reserved); + ggtt->start = start; + ggtt->size = size; +drm_mm_init (&ggtt->mm, start, size); } -int xe_ggtt_init_kunit (struct xe_ggtt *ggtt,u32 reserved,u32 size) +int xe_ggtt_init_kunit (struct xe_ggtt *ggtt,u32 start,u32 size) { - ggtt->size = size; -__xe_ggtt_init_early (ggtt, reserved); +__xe_ggtt_init_early (ggtt, start, size);return 0 ; }EXPORT_SYMBOL_IF_KUNIT (xe_ggtt_init_kunit);
@@ -269,26 +291,32 @@ int xe_ggtt_init_early(struct xe_ggtt *ggtt)struct xe_device *xe =tile_to_xe (ggtt->tile);struct pci_dev *pdev =to_pci_dev (xe->drm.dev);unsigned int gsm_size; +u64 ggtt_start, wopcm =xe_wopcm_size (xe), ggtt_size;int err; -if (IS_SRIOV_VF (xe) ||GRAPHICS_VERx100 (xe) >=1250 ) - gsm_size =SZ_8M ;/* GGTT is expected to be 4GiB */ -else - gsm_size =probe_gsm_size (pdev); - -if (gsm_size ==0 ) { -xe_tile_err (ggtt->tile,"Hardware reported no preallocated GSM\n" ); -return -ENOMEM ; +if (!IS_SRIOV_VF (xe)) { +if (GRAPHICS_VERx100 (xe) >=1250 ) + gsm_size =SZ_8M ;/* GGTT is expected to be 4GiB */ +else + gsm_size =probe_gsm_size (pdev); +if (gsm_size ==0 ) { +xe_tile_err (ggtt->tile,"Hardware reported no preallocated GSM\n" ); +return -ENOMEM ; + } + ggtt_start = wopcm; + ggtt_size = (gsm_size /8 ) * (u64 )XE_PAGE_SIZE - ggtt_start; + }else { +/* GGTT is expected to be 4GiB */ + ggtt_start = wopcm; + ggtt_size =SZ_4G - ggtt_start; } ggtt->gsm = ggtt->tile->mmio.regs +SZ_8M ; - ggtt->size = (gsm_size /8 ) * (u64 )XE_PAGE_SIZE ; -if (IS_DGFX (xe) && xe->info.vram_flags &XE_VRAM_FLAGS_NEED64K ) ggtt->flags |=XE_GGTT_FLAGS_64K ; -if (ggtt->size >GUC_GGTT_TOP ) - ggtt->size =GUC_GGTT_TOP ; +if (ggtt_size + ggtt_start >GUC_GGTT_TOP ) + ggtt_size =GUC_GGTT_TOP - ggtt_start;if (GRAPHICS_VERx100 (xe) >=1270 ) ggtt->pt_ops = (ggtt->tile->media_gt &&
@@ -299,7 +327,7 @@ int xe_ggtt_init_early(struct xe_ggtt *ggtt) ggtt->pt_ops = &xelp_pt_ops; ggtt->wq =alloc_workqueue ("xe-ggtt-wq" ,0 ,WQ_MEM_RECLAIM ); -__xe_ggtt_init_early (ggtt,xe_wopcm_size (xe)); +__xe_ggtt_init_early (ggtt, ggtt_start, ggtt_size); err =drmm_add_action_or_reset (&xe->drm, ggtt_fini_early, ggtt);if (err)
@@ -537,11 +565,9 @@ void xe_ggtt_node_remove_balloon_locked(struct xe_ggtt_node *node)static void xe_ggtt_assert_fit (struct xe_ggtt *ggtt,u64 start,u64 size) {struct xe_tile *tile = ggtt->tile; -struct xe_device *xe =tile_to_xe (tile); -u64 __maybe_unused wopcm =xe_wopcm_size (xe); -xe_tile_assert (tile, start >= wopcm); -xe_tile_assert (tile, start + size < ggtt->size - wopcm); +xe_tile_assert (tile, start >= ggtt->start); +xe_tile_assert (tile, start + size <= ggtt->start + ggtt->size); }/**
@@ -850,14 +876,12 @@ u64 xe_ggtt_largest_hole(struct xe_ggtt *ggtt, u64 alignment, u64 *spare) {const struct drm_mm *mm = &ggtt->mm;const struct drm_mm_node *entry; -u64 hole_min_start =xe_wopcm_size (tile_to_xe (ggtt->tile));u64 hole_start, hole_end, hole_size;u64 max_hole =0 ;mutex_lock (&ggtt->lock); -drm_mm_for_each_hole (entry, mm, hole_start, hole_end) { - hole_start =max (hole_start, hole_min_start); + hole_start =max (hole_start, ggtt->start); hole_start =ALIGN (hole_start, alignment); hole_end =ALIGN_DOWN (hole_end, alignment);if (hole_start >= hole_end)
@@ -950,15 +974,13 @@ u64 xe_ggtt_print_holes(struct xe_ggtt *ggtt, u64 alignment, struct drm_printer {const struct drm_mm *mm = &ggtt->mm;const struct drm_mm_node *entry; -u64 hole_min_start =xe_wopcm_size (tile_to_xe (ggtt->tile));u64 hole_start, hole_end, hole_size;u64 total =0 ;char buf[10 ];mutex_lock (&ggtt->lock); -drm_mm_for_each_hole (entry, mm, hole_start, hole_end) { - hole_start =max (hole_start, hole_min_start); + hole_start =max (hole_start, ggtt->start); hole_start =ALIGN (hole_start, alignment); hole_end =ALIGN_DOWN (hole_end, alignment);if (hole_start >= hole_end)
drivers/gpu/drm/xe/xe_ggtt.h +2 -0
@@ -23,6 +23,8 @@ int xe_ggtt_node_insert_balloon_locked(struct xe_ggtt_node *node, u64 start, u64 size);void xe_ggtt_node_remove_balloon_locked (struct xe_ggtt_node *node);void xe_ggtt_shift_nodes_locked (struct xe_ggtt *ggtt,s64 shift); +u64 xe_ggtt_start (struct xe_ggtt *ggtt); +u64 xe_ggtt_size (struct xe_ggtt *ggtt);int xe_ggtt_node_insert (struct xe_ggtt_node *node,u32 size,u32 align); int xe_ggtt_node_insert_locked(struct xe_ggtt_node *node,
drivers/gpu/drm/xe/xe_ggtt_types.h +3 -1
@@ -22,7 +22,9 @@ struct xe_gt;struct xe_ggtt {/** @tile: Back pointer to tile where this GGTT belongs */ struct xe_tile *tile; -/** @size: Total size of this GGTT */ +/** @start: Start offset of GGTT */ +u64 start; +/** @size: Total usable size of this GGTT */ u64 size;#define XE_GGTT_FLAGS_64K BIT(0)
drivers/gpu/drm/xe/xe_gt_sriov_pf_config.c +2 -2
@@ -343,8 +343,8 @@ static int pf_push_full_vf_config(struct xe_gt *gt, unsigned int vfid)xe_gt_assert (gt, num_dwords <= max_cfg_dwords);if (vfid ==PFID ) { -u64 ggtt_start =xe_wopcm_size (gt_to_xe (gt)); -u64 ggtt_size =gt_to_tile (gt)->mem.ggtt->size - ggtt_start; +u64 ggtt_start =xe_ggtt_start (gt_to_tile (gt)->mem.ggtt); +u64 ggtt_size =xe_ggtt_size (gt_to_tile (gt)->mem.ggtt);/* plain PF config data will never include a real GGTT region */ xe_gt_assert (gt, !encode_config_ggtt (cfg + num_dwords, config, true));
Signature · 1 line
2.51.0
PATCH v6 2/12 drm/xe: Convert xe_fb_pin to use a callback for insertion into GGTT
In reply to Maarten LankhorstThe rotation details belong in xe_fb_pin.c, while the operations involving GGTT belong to xe_ggtt.c. As directly locking xe_ggtt etc results in exposing all of xe_ggtt details anyway, create a special function that allocates a ggtt_node, and allow display to populate it using a callback as a compromise.
--- drivers/gpu/drm/xe/display/xe_fb_pin.c | 111 ++++++++++++------------- drivers/gpu/drm/xe/xe_ggtt.c | 90 ++++++++++++++------ drivers/gpu/drm/xe/xe_ggtt.h | 9 +- drivers/gpu/drm/xe/xe_ggtt_types.h | 8 +- 4 files changed, 130 insertions(+), 88 deletions(-)
drivers/gpu/drm/xe/display/xe_fb_pin.c +52 -59
@@ -171,12 +171,13 @@ static int __xe_pin_fb_vma_dpt(const struct intel_framebuffer *fb, }static void -write_ggtt_rotated (struct xe_bo *bo,struct xe_ggtt *ggtt,u32 *ggtt_ofs,u32 bo_ofs, +write_ggtt_rotated (struct xe_ggtt *ggtt,u32 *ggtt_ofs, +u64 pte_flags, +xe_ggtt_set_pte_fn write_pte, +struct xe_bo *bo,u32 bo_ofs,u32 width,u32 height,u32 src_stride,u32 dst_stride) { -struct xe_device *xe =xe_bo_device (bo);u32 column, row; -u64 pte = ggtt->pt_ops->pte_encode_flags (bo, xe->pat.idx[XE_CACHE_NONE ]);for (column =0 ; column < width; column++) {u32 src_idx = src_stride * (height -1 ) + column + bo_ofs;
@@ -184,7 +185,7 @@ write_ggtt_rotated(struct xe_bo *bo, struct xe_ggtt *ggtt, u32 *ggtt_ofs, u32 bofor (row =0 ; row < height; row++) {u64 addr =xe_bo_addr (bo, src_idx *XE_PAGE_SIZE ,XE_PAGE_SIZE ); - ggtt->pt_ops->ggtt_set_pte (ggtt, *ggtt_ofs, pte | addr); +write_pte (ggtt, *ggtt_ofs, pte_flags | addr); *ggtt_ofs +=XE_PAGE_SIZE ; src_idx -= src_stride; }
@@ -194,6 +195,28 @@ write_ggtt_rotated(struct xe_bo *bo, struct xe_ggtt *ggtt, u32 *ggtt_ofs, u32 bo } } +struct fb_rotate_args { +const struct i915_gtt_view *view; +struct xe_bo *bo; +}; + +static void write_ggtt_rotated_node (struct xe_ggtt *ggtt,struct xe_ggtt_node *node, +u64 pte_flags,xe_ggtt_set_pte_fn write_pte,void *data) +{ +struct fb_rotate_args *args = data; +struct xe_bo *bo = args->bo; +const struct intel_rotation_info *rot_info = &args->view->rotated; +u32 ggtt_ofs = node->base.start; + +for (u32 i =0 ; i <ARRAY_SIZE (rot_info->plane); i++) +write_ggtt_rotated (ggtt, &ggtt_ofs, pte_flags, write_pte, + bo, rot_info->plane[i].offset, + rot_info->plane[i].width, + rot_info->plane[i].height, + rot_info->plane[i].src_stride, + rot_info->plane[i].dst_stride); +} + static int __xe_pin_fb_vma_ggtt(const struct intel_framebuffer *fb, const struct i915_gtt_view *view, struct i915_vma *vma,
@@ -204,70 +227,40 @@ static int __xe_pin_fb_vma_ggtt(const struct intel_framebuffer *fb,struct xe_device *xe =to_xe_device (fb->base.dev);struct xe_tile *tile0 =xe_device_get_root_tile (xe);struct xe_ggtt *ggtt = tile0->mem.ggtt; +u64 pte, size;u32 align; -int ret; - -/* TODO: Consider sharing framebuffer mapping? -* embed i915_vma inside intel_framebuffer -*/ -xe_pm_runtime_get_noresume (xe); - ret =mutex_lock_interruptible (&ggtt->lock); -if (ret) - goto out; +int ret =0 ; align =XE_PAGE_SIZE ; -if (xe_bo_is_vram (bo) && ggtt->flags &XE_GGTT_FLAGS_64K ) - align =max_t (u32, align,SZ_64K ); +if (xe_bo_is_vram (bo) && xe->info.vram_flags &XE_VRAM_FLAGS_NEED64K ) + align =max (align,SZ_64K ); +/* Fast case, preallocated GGTT view? */ if (bo->ggtt_node[tile0->id] && view->type ==I915_GTT_VIEW_NORMAL ) { vma->node = bo->ggtt_node[tile0->id]; - }else if (view->type ==I915_GTT_VIEW_NORMAL ) { - vma->node =xe_ggtt_node_init (ggtt); -if (IS_ERR (vma->node)) { - ret =PTR_ERR (vma->node); - goto out_unlock; - } - - ret =xe_ggtt_node_insert_locked (vma->node,xe_bo_size (bo), align,0 ); -if (ret) { -xe_ggtt_node_fini (vma->node); - goto out_unlock; - } - -xe_ggtt_map_bo (ggtt, vma->node, bo, xe->pat.idx[XE_CACHE_NONE ]); - }else { -u32 i, ggtt_ofs; -const struct intel_rotation_info *rot_info = &view->rotated; - -/* display seems to use tiles instead of bytes here, so convert it back.. */ -u32 size =intel_rotation_info_size (rot_info) *XE_PAGE_SIZE ; - - vma->node =xe_ggtt_node_init (ggtt); -if (IS_ERR (vma->node)) { - ret =PTR_ERR (vma->node); - goto out_unlock; - } - - ret =xe_ggtt_node_insert_locked (vma->node, size, align,0 ); -if (ret) { -xe_ggtt_node_fini (vma->node); - goto out_unlock; - } +return 0 ; + } - ggtt_ofs = vma->node->base.start; +/* TODO: Consider sharing framebuffer mapping? +* embed i915_vma inside intel_framebuffer +*/ +xe_pm_runtime_get_noresume (xe); -for (i =0 ; i <ARRAY_SIZE (rot_info->plane); i++) -write_ggtt_rotated (bo, ggtt, &ggtt_ofs, - rot_info->plane[i].offset, - rot_info->plane[i].width, - rot_info->plane[i].height, - rot_info->plane[i].src_stride, - rot_info->plane[i].dst_stride); - } +if (view->type ==I915_GTT_VIEW_NORMAL ) + size =xe_bo_size (bo); +else +/* display uses tiles instead of bytes here, so convert it back.. */ + size =intel_rotation_info_size (&view->rotated) *XE_PAGE_SIZE ; + + pte =xe_ggtt_encode_pte_flags (ggtt, bo, xe->pat.idx[XE_CACHE_NONE ]); + vma->node =xe_ggtt_node_insert_transform (ggtt, bo, pte, +ALIGN (size, align), align, + view->type ==I915_GTT_VIEW_NORMAL ? +NULL : write_ggtt_rotated_node, + &(struct fb_rotate_args ){view, bo}); +if (IS_ERR (vma->node)) + ret =PTR_ERR (vma->node); -out_unlock: -mutex_unlock (&ggtt->lock); -out:xe_pm_runtime_put (xe);return ret; }
drivers/gpu/drm/xe/xe_ggtt.c +66 -24
132 lines not on this page · raw message
drivers/gpu/drm/xe/xe_ggtt.h +5 -4
@@ -27,12 +27,13 @@ u64 xe_ggtt_start(struct xe_ggtt *ggtt);u64 xe_ggtt_size (struct xe_ggtt *ggtt);int xe_ggtt_node_insert (struct xe_ggtt_node *node,u32 size,u32 align); -int xe_ggtt_node_insert_locked (struct xe_ggtt_node *node, -u32 size,u32 align,u32 mm_flags); +struct xe_ggtt_node * +xe_ggtt_node_insert_transform (struct xe_ggtt *ggtt, +struct xe_bo *bo,u64 pte, +u64 size,u32 align, +xe_ggtt_transform_cb transform,void *arg);void xe_ggtt_node_remove (struct xe_ggtt_node *node,bool invalidate);bool xe_ggtt_node_allocated (const struct xe_ggtt_node *node); -void xe_ggtt_map_bo (struct xe_ggtt *ggtt,struct xe_ggtt_node *node, -struct xe_bo *bo,u16 pat_index);void xe_ggtt_map_bo_unlocked (struct xe_ggtt *ggtt,struct xe_bo *bo);int xe_ggtt_insert_bo (struct xe_ggtt *ggtt,struct xe_bo *bo,struct drm_exec *exec); int xe_ggtt_insert_bo_at(struct xe_ggtt *ggtt, struct xe_bo *bo,
drivers/gpu/drm/xe/xe_ggtt_types.h +7 -1
@@ -71,6 +71,11 @@ struct xe_ggtt_node {bool invalidate_on_remove; }; +typedef void (*xe_ggtt_set_pte_fn )(struct xe_ggtt *ggtt,u64 addr,u64 pte); +typedef void (*xe_ggtt_transform_cb )(struct xe_ggtt *ggtt, +struct xe_ggtt_node *node, +u64 pte_flags, +xe_ggtt_set_pte_fn set_pte,void *arg);/** * struct xe_ggtt_pt_ops - GGTT Page table operations * Which can vary from platform to platform.
@@ -78,8 +83,9 @@ struct xe_ggtt_node {struct xe_ggtt_pt_ops {/** @pte_encode_flags: Encode PTE flags for a given BO */ u64 (*pte_encode_flags)(struct xe_bo *bo,u16 pat_index); +/** @ggtt_set_pte: Directly write into GGTT's PTE */ -void (*ggtt_set_pte)(struct xe_ggtt *ggtt,u64 addr,u64 pte); +xe_ggtt_set_pte_fn ggtt_set_pte; };#endif
Signature · 1 line
2.51.0
PATCH v6 3/12 drm/xe: Add xe_ggtt_node_addr() to avoid dereferencing xe_ggtt_node
In reply to Maarten LankhorstThis function makes it possible to add an offset that is applied to all xe_ggtt_node's, and hides the internals from all its users.
--- drivers/gpu/drm/xe/xe_bo.h | 8 +++++--- drivers/gpu/drm/xe/xe_ggtt.c | 11 +++++++++++ drivers/gpu/drm/xe/xe_ggtt.h | 2 ++ 3 files changed, 18 insertions(+), 3 deletions(-)
drivers/gpu/drm/xe/xe_bo.h +5 -3
@@ -9,6 +9,7 @@#include <drm/ttm/ttm_tt.h> #include "xe_bo_types.h" +#include "xe_ggtt.h" #include "xe_macros.h" #include "xe_validation.h" #include "xe_vm_types.h"
@@ -251,13 +252,14 @@ static inline u32 __xe_bo_ggtt_addr(struct xe_bo *bo, u8 tile_id) {struct xe_ggtt_node *ggtt_node = bo->ggtt_node[tile_id]; +u64 offset;if (XE_WARN_ON (!ggtt_node))return 0 ; -XE_WARN_ON (ggtt_node->base.size >xe_bo_size (bo)); -XE_WARN_ON (ggtt_node->base.start + ggtt_node->base.size > (1ull <<32 )); -return ggtt_node->base.start; + offset =xe_ggtt_node_addr (ggtt_node); +XE_WARN_ON (offset +xe_bo_size (bo) > (1ull <<32 )); +return offset; } static inline u32
drivers/gpu/drm/xe/xe_ggtt.c +11 -0
@@ -1066,3 +1066,14 @@ u64 xe_ggtt_read_pte(struct xe_ggtt *ggtt, u64 offset) {return ioread64 (ggtt->gsm + (offset /XE_PAGE_SIZE )); } + +/** +* xe_ggtt_node_addr - Get @node offset in GGTT. +* @node: &xe_ggtt_node +* +* Get the GGTT offset for allocated node. +*/ +u64 xe_ggtt_node_addr (const struct xe_ggtt_node *node) +{ +return node->base.start; +}
drivers/gpu/drm/xe/xe_ggtt.h +2 -0
@@ -58,4 +58,6 @@ void xe_ggtt_might_lock(struct xe_ggtt *ggtt);u64 xe_ggtt_encode_pte_flags (struct xe_ggtt *ggtt,struct xe_bo *bo,u16 pat_index);u64 xe_ggtt_read_pte (struct xe_ggtt *ggtt,u64 offset); +u64 xe_ggtt_node_addr (const struct xe_ggtt_node *node); +#endif
Signature · 1 line
2.51.0
PATCH v6 4/12 drm/xe/display: Avoid dereferencing xe_ggtt_node
In reply to Maarten LankhorstStart using xe_ggtt_node_addr, and avoid comparing the base offset as vma->node is dynamically allocated.
Also sneak in a xe_bo_size() for stolen, too small to put as separate commit.
--- drivers/gpu/drm/xe/compat-i915-headers/i915_vma.h | 4 ++-- drivers/gpu/drm/xe/display/xe_fb_pin.c | 4 ++-- drivers/gpu/drm/xe/display/xe_stolen.c | 2 +- 3 files changed, 5 insertions(+), 5 deletions(-)
drivers/gpu/drm/xe/compat-i915-headers/i915_vma.h +2 -2
@@ -8,7 +8,7 @@#include <uapi/drm/i915_drm.h> -#include "xe_ggtt_types.h" +#include "xe_ggtt.h" #include <linux/refcount.h>
@@ -32,7 +32,7 @@ struct i915_vma {static inline u32 i915_ggtt_offset (const struct i915_vma *vma) { -return vma->node->base.start; +return xe_ggtt_node_addr (vma->node); }#endif
drivers/gpu/drm/xe/display/xe_fb_pin.c +2 -2
@@ -206,7 +206,7 @@ static void write_ggtt_rotated_node(struct xe_ggtt *ggtt, struct xe_ggtt_node *nstruct fb_rotate_args *args = data;struct xe_bo *bo = args->bo;const struct intel_rotation_info *rot_info = &args->view->rotated; -u32 ggtt_ofs = node->base.start; +u32 ggtt_ofs =xe_ggtt_node_addr (node);for (u32 i =0 ; i <ARRAY_SIZE (rot_info->plane); i++) write_ggtt_rotated(ggtt, &ggtt_ofs, pte_flags, write_pte,
@@ -351,7 +351,7 @@ static void __xe_unpin_fb_vma(struct i915_vma *vma)if (vma->dpt)xe_bo_unpin_map_no_vm (vma->dpt);else if (!xe_ggtt_node_allocated (vma->bo->ggtt_node[tile_id]) || - vma->bo->ggtt_node[tile_id]->base.start != vma->node->base.start) + vma->bo->ggtt_node[tile_id] != vma->node)xe_ggtt_node_remove (vma->node, false);ttm_bo_reserve (&vma->bo->ttm, false, false,NULL );
drivers/gpu/drm/xe/display/xe_stolen.c +1 -1
@@ -100,7 +100,7 @@ u64 i915_gem_stolen_node_address(struct intel_stolen_node *node)u64 i915_gem_stolen_node_size (const struct intel_stolen_node *node) { -return node->bo->ttm.base.size; +return xe_bo_size (node->bo); }struct intel_stolen_node *i915_gem_stolen_node_alloc (struct drm_device *drm)
Signature · 1 line
2.51.0
PATCH v6 5/12 drm/xe: Do not dereference ggtt_node in xe_bo.c
In reply to Maarten LankhorstA careful inspection of __xe_ggtt_insert_bo_at() shows that the ggtt_node can always be seen as inserted from xe_bo.c due to the way error handling is performed.
--- drivers/gpu/drm/xe/xe_bo.c | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-)
drivers/gpu/drm/xe/xe_bo.c +3 -3
@@ -1700,7 +1700,7 @@ static void xe_ttm_bo_destroy(struct ttm_buffer_object *ttm_bo)xe_assert (xe,list_empty (&ttm_bo->base.gpuva.list));for_each_tile (tile, xe, id) -if (bo->ggtt_node[id] && bo->ggtt_node[id]->base.size) +if (bo->ggtt_node[id])xe_ggtt_remove_bo (tile->mem.ggtt, bo);#ifdef CONFIG_PROC_FS
@@ -3581,8 +3581,8 @@ void xe_bo_put(struct xe_bo *bo)might_lock (&bo->client->bos_lock);#endif for_each_tile (tile,xe_bo_device (bo), id) -if (bo->ggtt_node[id] && bo->ggtt_node[id]->ggtt) -xe_ggtt_might_lock (bo->ggtt_node[id]->ggtt); +if (bo->ggtt_node[id]) +xe_ggtt_might_lock (tile->mem.ggtt);drm_gem_object_put (&bo->ttm.base); } }
Signature · 1 line
2.51.0
PATCH v6 6/12 drm/xe: Improve xe_gt_sriov_pf_config GGTT handling
In reply to Maarten LankhorstDo not directly dereference xe_ggtt_node, and add
a member to store the GGTT size.
--- drivers/gpu/drm/xe/xe_gt_sriov_pf_config.c | 15 ++++++++------- drivers/gpu/drm/xe/xe_gt_sriov_pf_config_types.h | 8 ++++++-- 2 files changed, 14 insertions(+), 9 deletions(-)
drivers/gpu/drm/xe/xe_gt_sriov_pf_config.c +8 -7
@@ -264,7 +264,7 @@ static u32 encode_config_ggtt(u32 *cfg, const struct xe_gt_sriov_config *config,if (!xe_ggtt_node_allocated (node))return 0 ; -return encode_ggtt (cfg, node->base.start, node->base.size, details); +return encode_ggtt (cfg,xe_ggtt_node_addr (node), config->ggtt_size, details); }/* Return: number of configuration dwords written */
@@ -495,13 +495,14 @@ static int pf_provision_vf_ggtt(struct xe_gt *gt, unsigned int vfid, u64 size)xe_ggtt_assign (node, vfid);xe_gt_sriov_dbg_verbose (gt,"VF%u assigned GGTT %llx-%llx\n" , - vfid, node->base.start, node->base.start + node->base.size -1 ); + vfid,xe_ggtt_node_addr (node),xe_ggtt_node_addr (node) + size -1 ); - err =pf_distribute_config_ggtt (gt->tile, vfid, node->base.start, node->base.size); + err =pf_distribute_config_ggtt (gt->tile, vfid,xe_ggtt_node_addr (node), size);if (unlikely (err)) goto err; config->ggtt_region = node; + config->ggtt_size = size;return 0 ; err:pf_release_ggtt (tile, node);
@@ -514,7 +515,7 @@ static u64 pf_get_vf_config_ggtt(struct xe_gt *gt, unsigned int vfid)struct xe_ggtt_node *node = config->ggtt_region;xe_gt_assert (gt,xe_gt_is_main_type (gt)); -return xe_ggtt_node_allocated (node) ? node->base.size :0 ; +return xe_ggtt_node_allocated (node) ? config->ggtt_size :0 ; }/**
@@ -2516,11 +2517,11 @@ int xe_gt_sriov_pf_config_print_ggtt(struct xe_gt *gt, struct drm_printer *p)if (!xe_ggtt_node_allocated (config->ggtt_region))continue ; -string_get_size (config->ggtt_region->base.size,1 ,STRING_UNITS_2 , +string_get_size (config->ggtt_size,1 ,STRING_UNITS_2 , buf,sizeof (buf));drm_printf (p,"VF%u:\t%#0llx-%#llx\t(%s)\n" , - n, config->ggtt_region->base.start, - config->ggtt_region->base.start + config->ggtt_region->base.size -1 , + n,xe_ggtt_node_addr (config->ggtt_region), +xe_ggtt_node_addr (config->ggtt_region) + config->ggtt_size -1 , buf); }
drivers/gpu/drm/xe/xe_gt_sriov_pf_config_types.h +6 -2
@@ -17,10 +17,14 @@ struct xe_bo;* Used by the PF driver to maintain per-VF provisioning data. */ struct xe_gt_sriov_config { -/** @ggtt_region: GGTT region assigned to the VF. */ -struct xe_ggtt_node *ggtt_region;/** @lmem_obj: LMEM allocation for use by the VF. */ struct xe_bo *lmem_obj; + +/** @ggtt_region: GGTT region assigned to the VF. */ +struct xe_ggtt_node *ggtt_region; +/** @ggtt_size: Size of GGTT region */ +u64 ggtt_size; +/** @num_ctxs: number of GuC contexts IDs. */ u16 num_ctxs;/** @begin_ctx: start index of GuC context ID range. */
Signature · 1 line
2.51.0
Re: [PATCH v6 06/12] drm/xe: Improve xe_gt_sriov_pf_config GGTT handling
Quoted 70 lines · Maarten Lankhorst
Do not directly dereference
xe_ggtt_node, and add
a member to store the GGTT size.Signed-off-by Maarten Lankhorst <[email protected]>--- drivers/gpu/drm/xe/xe_gt_sriov_pf_config.c | 15 ++++++++------- drivers/gpu/drm/xe/xe_gt_sriov_pf_config_types.h | 8 ++++++-- 2 files changed, 14 insertions(+), 9 deletions(-)drivers/gpu/drm/xe/xe_gt_sriov_pf_config.c +8
-7@@ -264,7 +264,7 @@ static u32 encode_config_ggtt(u32 *cfg, const struct xe_gt_sriov_config *config,if (!xe_ggtt_node_allocated (node))return 0 ; -return encode_ggtt (cfg, node->base.start, node->base.size, details); +return encode_ggtt (cfg,xe_ggtt_node_addr (node), config->ggtt_size, details); }/* Return: number of configuration dwords written */ @@ -495,13 +495,14 @@ static int pf_provision_vf_ggtt(struct xe_gt *gt, unsigned int vfid, u64 size)xe_ggtt_assign (node, vfid);xe_gt_sriov_dbg_verbose (gt,"VF%u assigned GGTT %llx-%llx\n" , - vfid, node->base.start, node->base.start + node->base.size -1 ); + vfid,xe_ggtt_node_addr (node),xe_ggtt_node_addr (node) + size -1 ); - err =pf_distribute_config_ggtt (gt->tile, vfid, node->base.start, node->base.size); + err =pf_distribute_config_ggtt (gt->tile, vfid,xe_ggtt_node_addr (node), size);if (unlikely (err)) goto err; config->ggtt_region = node; + config->ggtt_size = size;return 0 ; err:pf_release_ggtt (tile, node);@@ -514,7 +515,7 @@ static u64 pf_get_vf_config_ggtt(struct xe_gt *gt, unsigned int vfid)struct xe_ggtt_node *node = config->ggtt_region;xe_gt_assert (gt,xe_gt_is_main_type (gt)); -return xe_ggtt_node_allocated (node) ? node->base.size :0 ; +return xe_ggtt_node_allocated (node) ? config->ggtt_size :0 ; }/** @@ -2516,11 +2517,11 @@ int xe_gt_sriov_pf_config_print_ggtt(struct xe_gt *gt, struct drm_printer *p)if (!xe_ggtt_node_allocated (config->ggtt_region))continue ; -string_get_size (config->ggtt_region->base.size,1 ,STRING_UNITS_2 , +string_get_size (config->ggtt_size,1 ,STRING_UNITS_2 , buf,sizeof (buf));drm_printf (p,"VF%u:\t%#0llx-%#llx\t(%s)\n" , - n, config->ggtt_region->base.start, - config->ggtt_region->base.start + config->ggtt_region->base.size -1 , + n,xe_ggtt_node_addr (config->ggtt_region), +xe_ggtt_node_addr (config->ggtt_region) + config->ggtt_size -1 , buf); }drivers/gpu/drm/xe/xe_gt_sriov_pf_config_types.h +0
-0@@ -17,10 +17,14 @@ struct xe_bo;* Used by the PF driver to maintain per-VF provisioning data. */
drivers/gpu/drm/xe/xe_gt_sriov_pf_config_types.h struct xe_bo; · line 26
Quoted 10 lines
drivers/gpu/drm/xe/xe_gt_sriov_pf_config_types.h +5
-2struct xe_gt_sriov_config { -/** @ggtt_region: GGTT region assigned to the VF. */ -struct xe_ggtt_node *ggtt_region;/** @lmem_obj: LMEM allocation for use by the VF. */ struct xe_bo *lmem_obj; + +/** @ggtt_region: GGTT region assigned to the VF. */ +struct xe_ggtt_node *ggtt_region; +/** @ggtt_size: Size of GGTT region */ +u64 ggtt_size;
can't we just add another helper on GGTT side to get the size of the node?
u64 xe_ggtt_node_size(node);
there is one for xe_bo, why can't we have another for xe_ggtt_node ?
Quoted 4 lines
drivers/gpu/drm/xe/xe_gt_sriov_pf_config_types.h +1
-0+/** @num_ctxs: number of GuC contexts IDs. */ u16 num_ctxs;/** @begin_ctx: start index of GuC context ID range. */
PATCH v6 7/12 drm/xe: Privatize xe_ggtt_node
In reply to Maarten LankhorstNothing requires it any more, make the member private.
--- drivers/gpu/drm/xe/xe_ggtt.c | 18 ++++++++++++++++++ drivers/gpu/drm/xe/xe_ggtt_types.h | 19 +------------------ 2 files changed, 19 insertions(+), 18 deletions(-)
drivers/gpu/drm/xe/xe_ggtt.c +18 -0
@@ -34,6 +34,24 @@#include "xe_wa.h" #include "xe_wopcm.h" +/** +* struct xe_ggtt_node - A node in GGTT. +* +* This struct needs to be initialized (only-once) with xe_ggtt_node_init() before any node +* insertion, reservation, or 'ballooning'. +* It will, then, be finalized by either xe_ggtt_node_remove() or xe_ggtt_node_deballoon(). +*/ +struct xe_ggtt_node { +/** @ggtt: Back pointer to xe_ggtt where this region will be inserted at */ +struct xe_ggtt *ggtt; +/** @base: A drm_mm_node */ +struct drm_mm_node base; +/** @delayed_removal_work: The work struct for the delayed removal */ +struct work_struct delayed_removal_work; +/** @invalidate_on_remove: If it needs invalidation upon removal */ +bool invalidate_on_remove; +}; +/** * DOC: Global Graphics Translation Table (GGTT) *
drivers/gpu/drm/xe/xe_ggtt_types.h +1 -18
@@ -11,6 +11,7 @@#include "xe_pt_types.h" struct xe_bo ; +struct xe_ggtt_node ;struct xe_gt ;/**
@@ -53,24 +54,6 @@ struct xe_ggtt {struct workqueue_struct *wq; }; -/** -* struct xe_ggtt_node - A node in GGTT. -* -* This struct needs to be initialized (only-once) with xe_ggtt_node_init() before any node -* insertion, reservation, or 'ballooning'. -* It will, then, be finalized by either xe_ggtt_node_remove() or xe_ggtt_node_deballoon(). -*/ -struct xe_ggtt_node { -/** @ggtt: Back pointer to xe_ggtt where this region will be inserted at */ -struct xe_ggtt *ggtt; -/** @base: A drm_mm_node */ -struct drm_mm_node base; -/** @delayed_removal_work: The work struct for the delayed removal */ -struct work_struct delayed_removal_work; -/** @invalidate_on_remove: If it needs invalidation upon removal */ -bool invalidate_on_remove; -}; -typedef void (*xe_ggtt_set_pte_fn )(struct xe_ggtt *ggtt,u64 addr,u64 pte); typedef void (*xe_ggtt_transform_cb)(struct xe_ggtt *ggtt, struct xe_ggtt_node *node,
Signature · 1 line
2.51.0
PATCH v6 8/12 drm/xe: Make xe_ggtt_node offset relative to starting offset
In reply to Maarten LankhorstThis will make node shifting in the next commit a one-liner.
--- drivers/gpu/drm/xe/xe_ggtt.c | 21 ++++++++++++++------- 1 file changed, 14 insertions(+), 7 deletions(-)
drivers/gpu/drm/xe/xe_ggtt.c +14 -7
@@ -276,7 +276,7 @@ static void __xe_ggtt_init_early(struct xe_ggtt *ggtt, u64 start, u64 size) { ggtt->start = start; ggtt->size = size; -drm_mm_init (&ggtt->mm, start, size); +drm_mm_init (&ggtt->mm,0 , size - start); }int xe_ggtt_init_kunit (struct xe_ggtt *ggtt,u32 start,u32 size)
@@ -375,7 +375,7 @@ static void xe_ggtt_initial_clear(struct xe_ggtt *ggtt)/* Display may have allocated inside ggtt, so be careful with clearing here */ mutex_lock (&ggtt->lock);drm_mm_for_each_hole (hole, &ggtt->mm, start, end) -xe_ggtt_clear (ggtt, start, end - start); +xe_ggtt_clear (ggtt, ggtt->start + start, end - start);xe_ggtt_invalidate (ggtt);mutex_unlock (&ggtt->lock);
@@ -392,7 +392,7 @@ static void ggtt_node_remove(struct xe_ggtt_node *node)mutex_lock (&ggtt->lock);if (bound) -xe_ggtt_clear (ggtt, node->base.start, node->base.size); +xe_ggtt_clear (ggtt,xe_ggtt_node_addr (node), node->base.size);drm_mm_remove_node (&node->base); node->base.size =0 ;mutex_unlock (&ggtt->lock);
@@ -548,13 +548,13 @@ int xe_ggtt_node_insert_balloon_locked(struct xe_ggtt_node *node, u64 start, u64lockdep_assert_held (&ggtt->lock); node->base.color =0 ; - node->base.start = start; + node->base.start = start - ggtt->start; node->base.size = end - start; err =drm_mm_reserve_node (&ggtt->mm, &node->base);if (xe_tile_WARN (ggtt->tile, err,"Failed to balloon GGTT %#llx-%#llx (%pe)\n" , - node->base.start, node->base.start + node->base.size,ERR_PTR (err))) +xe_ggtt_node_addr (node),xe_ggtt_node_addr (node) + node->base.size,ERR_PTR (err)))return err;xe_ggtt_dump_node (ggtt, &node->base,"balloon" );
@@ -731,7 +731,7 @@ static void xe_ggtt_map_bo(struct xe_ggtt *ggtt, struct xe_ggtt_node *node,if (XE_WARN_ON (!node))return ; - start = node->base.start; + start =xe_ggtt_node_addr (node); end = start +xe_bo_size (bo);if (!xe_bo_is_vram (bo) && !xe_bo_is_stolen (bo)) {
@@ -852,6 +852,13 @@ static int __xe_ggtt_insert_bo_at(struct xe_ggtt *ggtt, struct xe_bo *bo, }mutex_lock (&ggtt->lock); +if (start >= ggtt->start) + start -= ggtt->start; +else + start =0 ; + + end -= ggtt->start; + err =drm_mm_insert_node_in_range (&ggtt->mm, &bo->ggtt_node[tile_id]->base,xe_bo_size (bo), alignment,0 , start, end,0 );if (err) {
@@ -1093,5 +1100,5 @@ u64 xe_ggtt_read_pte(struct xe_ggtt *ggtt, u64 offset)*/ u64 xe_ggtt_node_addr (const struct xe_ggtt_node *node) { -return node->base.start; +return node->base.start +READ_ONCE (node->ggtt->start); }
Signature · 1 line
2.51.0
Re: [PATCH v6 08/12] drm/xe: Make xe_ggtt_node offset relative to starting offset
Quoted 12 lines · Maarten Lankhorst
This will make node shifting in the next commit a one-liner.
Signed-off-by Maarten Lankhorst <[email protected]>--- drivers/gpu/drm/xe/xe_ggtt.c | 21 ++++++++++++++------- 1 file changed, 14 insertions(+), 7 deletions(-)drivers/gpu/drm/xe/xe_ggtt.c +0
-0
drivers/gpu/drm/xe/xe_ggtt.c static void __xe_ggtt_init_early(struct xe_ggtt *ggtt, u64 start, u64 size) · line 279
Quoted 5 lines
drivers/gpu/drm/xe/xe_ggtt.c +1
-1@@ -276,7 +276,7 @@ static void __xe_ggtt_init_early(struct xe_ggtt *ggtt, u64 start, u64 size) { ggtt->start = start; ggtt->size = size; -drm_mm_init (&ggtt->mm, start, size); +drm_mm_init (&ggtt->mm,0 , size - start);
that's little confusing
either "size" param should be named "end" or this should be just
drm_mm_init(&ggtt->mm, 0, size);
or am I reading this wrong?
drivers/gpu/drm/xe/xe_ggtt.c static void xe_ggtt_initial_clear(struct xe_ggtt *ggtt) · line 378
Quoted 9 lines
drivers/gpu/drm/xe/xe_ggtt.c +1
-1}int xe_ggtt_init_kunit (struct xe_ggtt *ggtt,u32 start,u32 size)@@ -375,7 +375,7 @@ static void xe_ggtt_initial_clear(struct xe_ggtt *ggtt)/* Display may have allocated inside ggtt, so be careful with clearing here */ mutex_lock (&ggtt->lock);drm_mm_for_each_hole (hole, &ggtt->mm, start, end) -xe_ggtt_clear (ggtt, start, end - start); +xe_ggtt_clear (ggtt, ggtt->start + start, end - start);
since xe_ggtt_clear is static/private then maybe "ggtt->start" shift can be done there?
Quoted 58 lines
drivers/gpu/drm/xe/xe_ggtt.c +12
-5xe_ggtt_invalidate (ggtt);mutex_unlock (&ggtt->lock);@@ -392,7 +392,7 @@ static void ggtt_node_remove(struct xe_ggtt_node *node)mutex_lock (&ggtt->lock);if (bound) -xe_ggtt_clear (ggtt, node->base.start, node->base.size); +xe_ggtt_clear (ggtt,xe_ggtt_node_addr (node), node->base.size);drm_mm_remove_node (&node->base); node->base.size =0 ;mutex_unlock (&ggtt->lock);@@ -548,13 +548,13 @@ int xe_ggtt_node_insert_balloon_locked(struct xe_ggtt_node *node, u64 start, u64lockdep_assert_held (&ggtt->lock); node->base.color =0 ; - node->base.start = start; + node->base.start = start - ggtt->start; node->base.size = end - start; err =drm_mm_reserve_node (&ggtt->mm, &node->base);if (xe_tile_WARN (ggtt->tile, err,"Failed to balloon GGTT %#llx-%#llx (%pe)\n" , - node->base.start, node->base.start + node->base.size,ERR_PTR (err))) +xe_ggtt_node_addr (node),xe_ggtt_node_addr (node) + node->base.size,ERR_PTR (err)))return err;xe_ggtt_dump_node (ggtt, &node->base,"balloon" );@@ -731,7 +731,7 @@ static void xe_ggtt_map_bo(struct xe_ggtt *ggtt, struct xe_ggtt_node *node,if (XE_WARN_ON (!node))return ; - start = node->base.start; + start =xe_ggtt_node_addr (node); end = start +xe_bo_size (bo);if (!xe_bo_is_vram (bo) && !xe_bo_is_stolen (bo)) {@@ -852,6 +852,13 @@ static int __xe_ggtt_insert_bo_at(struct xe_ggtt *ggtt, struct xe_bo *bo, }mutex_lock (&ggtt->lock); +if (start >= ggtt->start) + start -= ggtt->start; +else + start =0 ; + + end -= ggtt->start; + err =drm_mm_insert_node_in_range (&ggtt->mm, &bo->ggtt_node[tile_id]->base,xe_bo_size (bo), alignment,0 , start, end,0 );if (err) {@@ -1093,5 +1100,5 @@ u64 xe_ggtt_read_pte(struct xe_ggtt *ggtt, u64 offset)*/ u64 xe_ggtt_node_addr (const struct xe_ggtt_node *node) { -return node->base.start; +return node->base.start +READ_ONCE (node->ggtt->start); }
PATCH v6 9/12 drm/xe: Rewrite GGTT VF initialisation
In reply to Maarten LankhorstThe previous code was using a complicated system with 2 balloons to set GGTT size and adjust GGTT offset. While it works, it's overly complicated.
A better approach is to set the offset and size when initialising GGTT, this removes the need for adding balloons. The resize function only needs readjust ggtt->start to have GGTT at the new offset.
This removes the need to manipulate the internals of xe_ggtt outside of xe_ggtt, and cleans up a lot of now unneeded code.
--- drivers/gpu/drm/xe/tests/xe_guc_buf_kunit.c | 2 +- drivers/gpu/drm/xe/xe_device_types.h | 2 - drivers/gpu/drm/xe/xe_ggtt.c | 158 +++------------- drivers/gpu/drm/xe/xe_ggtt.h | 5 +- drivers/gpu/drm/xe/xe_ggtt_types.h | 1 + drivers/gpu/drm/xe/xe_gt_sriov_vf.c | 5 +- drivers/gpu/drm/xe/xe_tile.c | 18 ++ drivers/gpu/drm/xe/xe_tile_sriov_vf.c | 197 ++------------------ drivers/gpu/drm/xe/xe_tile_sriov_vf.h | 4 +- drivers/gpu/drm/xe/xe_tile_sriov_vf_types.h | 4 + 10 files changed, 71 insertions(+), 325 deletions(-)
drivers/gpu/drm/xe/tests/xe_guc_buf_kunit.c +1 -1
@@ -67,7 +67,7 @@ static int guc_buf_test_init(struct kunit *test)KUNIT_ASSERT_EQ (test,0 ,xe_ggtt_init_kunit (ggtt,DUT_GGTT_START , -DUT_GGTT_START +DUT_GGTT_SIZE )); +DUT_GGTT_SIZE ));kunit_activate_static_stub (test, xe_managed_bo_create_pin_map, replacement_xe_managed_bo_create_pin_map);
drivers/gpu/drm/xe/xe_device_types.h +0 -2
@@ -192,8 +192,6 @@ struct xe_tile {struct xe_lmtt lmtt; } pf;struct { -/** @sriov.vf.ggtt_balloon: GGTT regions excluded from use. */ -struct xe_ggtt_node *ggtt_balloon[2 ];/** @sriov.vf.self_config: VF configuration data */ struct xe_tile_sriov_vf_selfconfig self_config; } vf;
drivers/gpu/drm/xe/xe_ggtt.c +30 -128
204 lines not on this page · raw message
drivers/gpu/drm/xe/xe_ggtt.h +1 -4
@@ -19,10 +19,7 @@ int xe_ggtt_init(struct xe_ggtt *ggtt);struct xe_ggtt_node *xe_ggtt_node_init (struct xe_ggtt *ggtt);void xe_ggtt_node_fini (struct xe_ggtt_node *node); -int xe_ggtt_node_insert_balloon_locked (struct xe_ggtt_node *node, -u64 start,u64 size); -void xe_ggtt_node_remove_balloon_locked (struct xe_ggtt_node *node); -void xe_ggtt_shift_nodes_locked (struct xe_ggtt *ggtt,s64 shift); +void xe_ggtt_shift_nodes (struct xe_ggtt *ggtt,s64 shift);u64 xe_ggtt_start (struct xe_ggtt *ggtt);u64 xe_ggtt_size (struct xe_ggtt *ggtt);
drivers/gpu/drm/xe/xe_ggtt_types.h +1 -0
@@ -59,6 +59,7 @@ typedef void (*xe_ggtt_transform_cb)(struct xe_ggtt *ggtt,struct xe_ggtt_node *node, u64 pte_flags, xe_ggtt_set_pte_fn set_pte, void *arg); +/** * struct xe_ggtt_pt_ops - GGTT Page table operations * Which can vary from platform to platform.
drivers/gpu/drm/xe/xe_gt_sriov_vf.c +2 -3
@@ -442,7 +442,6 @@ u32 xe_gt_sriov_vf_gmdid(struct xe_gt *gt)static int vf_get_ggtt_info (struct xe_gt *gt) {struct xe_tile *tile =gt_to_tile (gt); -struct xe_ggtt *ggtt = tile->mem.ggtt;struct xe_guc *guc = >->uc.guc;u64 start, size, ggtt_size;s64 shift;
@@ -450,7 +449,7 @@ static int vf_get_ggtt_info(struct xe_gt *gt)xe_gt_assert (gt,IS_SRIOV_VF (gt_to_xe (gt))); -guard (mutex)(&ggtt->lock); +guard (mutex)(&tile->sriov.vf.self_config.ggtt_move_mutex); err =guc_action_query_single_klv64 (guc,GUC_KLV_VF_CFG_GGTT_START_KEY , &start);if (unlikely (err))
@@ -480,7 +479,7 @@ static int vf_get_ggtt_info(struct xe_gt *gt)if (shift && shift != start) {xe_gt_sriov_info (gt,"Shifting GGTT base by %lld to 0x%016llx\n" , shift, start); -xe_tile_sriov_vf_fixup_ggtt_nodes_locked (gt_to_tile (gt), shift); +xe_ggtt_shift_nodes (tile->mem.ggtt, shift); }if (xe_sriov_vf_migration_supported (gt_to_xe (gt))) {
drivers/gpu/drm/xe/xe_tile.c +18 -0
@@ -17,6 +17,7 @@#include "xe_sa.h" #include "xe_svm.h" #include "xe_tile.h" +#include "xe_tile_sriov_vf.h" #include "xe_tile_sysfs.h" #include "xe_ttm_vram_mgr.h" #include "xe_vram.h"
@@ -157,6 +158,12 @@ int xe_tile_init_early(struct xe_tile *tile, struct xe_device *xe, u8 id)if (err)return err; +if (IS_SRIOV_VF (xe)) { + err =xe_tile_sriov_vf_init (tile); +if (err) +return err; + } + tile->primary_gt =xe_gt_alloc (tile);if (IS_ERR (tile->primary_gt))return PTR_ERR (tile->primary_gt);
@@ -201,6 +208,17 @@ int xe_tile_init_noalloc(struct xe_tile *tile)return xe_tile_sysfs_init (tile); } +/** +* xe_tile_init - Initialize the remainder of the tile. +* @tile: The tile to initialize. +* +* This function is used for all tile initialization calls that may allocate memory. +* +* Note that since this is tile initialization, it should not perform any +* GT-specific operations, and thus does not need to hold GT forcewake. +* +* Returns: 0 on success, negative error code on error. +*/ int xe_tile_init (struct xe_tile *tile) {int err;
drivers/gpu/drm/xe/xe_tile_sriov_vf.c +13 -184
243 lines not on this page · raw message
drivers/gpu/drm/xe/xe_tile_sriov_vf.h +1 -3
@@ -10,9 +10,7 @@struct xe_tile ; -int xe_tile_sriov_vf_prepare_ggtt (struct xe_tile *tile); -void xe_tile_sriov_vf_deballoon_ggtt_locked (struct xe_tile *tile); -void xe_tile_sriov_vf_fixup_ggtt_nodes_locked (struct xe_tile *tile,s64 shift); +int xe_tile_sriov_vf_init (struct xe_tile *tile);u64 xe_tile_sriov_vf_ggtt (struct xe_tile *tile);void xe_tile_sriov_vf_ggtt_store (struct xe_tile *tile,u64 ggtt_size);u64 xe_tile_sriov_vf_ggtt_base (struct xe_tile *tile);
drivers/gpu/drm/xe/xe_tile_sriov_vf_types.h +4 -0
@@ -7,11 +7,15 @@#define _XE_TILE_SRIOV_VF_TYPES_H_#include <linux/types.h> +#include <linux/mutex.h> /** * struct xe_tile_sriov_vf_selfconfig - VF configuration data. */ struct xe_tile_sriov_vf_selfconfig { +/** @ggtt_move_mutex: Prevents multiple movements from happening in parallel */ +struct mutex ggtt_move_mutex; +/** @ggtt_base: assigned base offset of the GGTT region. */ u64 ggtt_base;/** @ggtt_size: assigned size of the GGTT region. */
Signature · 1 line
2.51.0
PATCH v6 10/12 drm/xe: Move struct xe_ggtt to xe_ggtt.c
In reply to Maarten LankhorstFrom: Maarten Lankhorst <[email protected]>
No users left outside of xe_ggtt.c, so we can make the struct private.
This prevents us from accidentally touching it before init.
--- drivers/gpu/drm/xe/xe_ggtt.c | 52 +++++++++++++++++++++++++++ drivers/gpu/drm/xe/xe_ggtt.h | 1 + drivers/gpu/drm/xe/xe_ggtt_types.h | 58 ++---------------------------- 3 files changed, 55 insertions(+), 56 deletions(-)
drivers/gpu/drm/xe/xe_ggtt.c +52 -0
@@ -85,6 +85,58 @@ struct xe_ggtt_node {* give us the correct placement for free. */ +/** +* struct xe_ggtt_pt_ops - GGTT Page table operations +* Which can vary from platform to platform. +*/ +struct xe_ggtt_pt_ops { +/** @pte_encode_flags: Encode PTE flags for a given BO */ +u64 (*pte_encode_flags)(struct xe_bo *bo,u16 pat_index); + +/** @ggtt_set_pte: Directly write into GGTT's PTE */ +xe_ggtt_set_pte_fn ggtt_set_pte; +}; + +/** +* struct xe_ggtt - Main GGTT struct +* +* In general, each tile can contains its own Global Graphics Translation Table +* (GGTT) instance. +*/ +struct xe_ggtt { +/** @tile: Back pointer to tile where this GGTT belongs */ +struct xe_tile *tile; +/** @start: Start offset of GGTT */ +u64 start; +/** @size: Total usable size of this GGTT */ +u64 size; + +#define XE_GGTT_FLAGS_64K BIT(0) +/** +* @flags: Flags for this GGTT +* Acceptable flags: +* - %XE_GGTT_FLAGS_64K - if PTE size is 64K. Otherwise, regular is 4K. +*/ +unsigned int flags; +/** @scratch: Internal object allocation used as a scratch page */ +struct xe_bo *scratch; +/** @lock: Mutex lock to protect GGTT data */ +struct mutex lock; +/** +* @gsm: The iomem pointer to the actual location of the translation +* table located in the GSM for easy PTE manipulation +*/ +u64 __iomem *gsm; +/** @pt_ops: Page Table operations per platform */ +const struct xe_ggtt_pt_ops *pt_ops; +/** @mm: The memory manager used to manage individual GGTT allocations */ +struct drm_mm mm; +/** @access_count: counts GGTT writes */ +unsigned int access_count; +/** @wq: Dedicated unordered work queue to process node removals */ +struct workqueue_struct *wq; +}; +static u64 xelp_ggtt_pte_flags (struct xe_bo *bo,u16 pat_index) {u64 pte =XE_PAGE_PRESENT ;
drivers/gpu/drm/xe/xe_ggtt.h +1 -0
@@ -9,6 +9,7 @@#include "xe_ggtt_types.h" struct drm_printer ; +struct xe_bo ;struct xe_tile ;struct drm_exec ;
drivers/gpu/drm/xe/xe_ggtt_types.h +2 -56
71 lines not on this page · raw message
Signature · 1 line
2.51.0
PATCH v6 11/12 drm/xe: Make xe_ggtt_node_insert return a node
In reply to Maarten LankhorstThis extra step is easier to handle inside xe_ggtt.c and makes xe_ggtt_node_allocated a simple null check instead, as the intermediate state 'allocated but not inserted' is no longer used.
Privatize xe_ggtt_node_fini() as it's no longer used outside of xe_ggtt.c
--- drivers/gpu/drm/xe/tests/xe_guc_buf_kunit.c | 6 +- drivers/gpu/drm/xe/xe_ggtt.c | 76 +++++++++------------ drivers/gpu/drm/xe/xe_ggtt.h | 5 +- drivers/gpu/drm/xe/xe_gt_sriov_pf_config.c | 24 +------ 4 files changed, 38 insertions(+), 73 deletions(-)
drivers/gpu/drm/xe/tests/xe_guc_buf_kunit.c +1 -5
@@ -38,12 +38,8 @@ static struct xe_bo *replacement_xe_managed_bo_create_pin_map(struct xe_device *if (flags &XE_BO_FLAG_GGTT ) {struct xe_ggtt *ggtt = tile->mem.ggtt; - bo->ggtt_node[tile->id] =xe_ggtt_node_init (ggtt); + bo->ggtt_node[tile->id] =xe_ggtt_node_insert (ggtt,xe_bo_size (bo),SZ_4K );KUNIT_ASSERT_NOT_ERR_OR_NULL (test, bo->ggtt_node[tile->id]); - -KUNIT_ASSERT_EQ (test,0 , -xe_ggtt_node_insert (bo->ggtt_node[tile->id], -xe_bo_size (bo),SZ_4K )); }return bo;
drivers/gpu/drm/xe/xe_ggtt.c +32 -44
106 lines not on this page · raw message
drivers/gpu/drm/xe/xe_ggtt.h +2 -3
@@ -18,13 +18,12 @@ int xe_ggtt_init_early(struct xe_ggtt *ggtt);int xe_ggtt_init_kunit (struct xe_ggtt *ggtt,u32 reserved,u32 size);int xe_ggtt_init (struct xe_ggtt *ggtt); -struct xe_ggtt_node *xe_ggtt_node_init (struct xe_ggtt *ggtt); -void xe_ggtt_node_fini (struct xe_ggtt_node *node);void xe_ggtt_shift_nodes (struct xe_ggtt *ggtt,s64 shift);u64 xe_ggtt_start (struct xe_ggtt *ggtt);u64 xe_ggtt_size (struct xe_ggtt *ggtt); -int xe_ggtt_node_insert (struct xe_ggtt_node *node,u32 size,u32 align); +struct xe_ggtt_node * +xe_ggtt_node_insert (struct xe_ggtt *ggtt,u32 size,u32 align);struct xe_ggtt_node * xe_ggtt_node_insert_transform(struct xe_ggtt *ggtt, struct xe_bo *bo, u64 pte,
drivers/gpu/drm/xe/xe_gt_sriov_pf_config.c +3 -21
@@ -434,23 +434,9 @@ static int pf_distribute_config_ggtt(struct xe_tile *tile, unsigned int vfid, u6return err ?: err2; } -static void pf_release_ggtt (struct xe_tile *tile,struct xe_ggtt_node *node) -{ -if (xe_ggtt_node_allocated (node)) { -/* -* explicit GGTT PTE assignment to the PF using xe_ggtt_assign() -* is redundant, as PTE will be implicitly re-assigned to PF by -* the xe_ggtt_clear() called by below xe_ggtt_remove_node(). -*/ -xe_ggtt_node_remove (node, false); - }else { -xe_ggtt_node_fini (node); - } -} -static void pf_release_vf_config_ggtt (struct xe_gt *gt,struct xe_gt_sriov_config *config) { -pf_release_ggtt (gt_to_tile (gt), config->ggtt_region); +xe_ggtt_node_remove (config->ggtt_region, false); config->ggtt_region =NULL ; }
@@ -485,14 +471,10 @@ static int pf_provision_vf_ggtt(struct xe_gt *gt, unsigned int vfid, u64 size)if (!size)return 0 ; - node =xe_ggtt_node_init (ggtt); + node =xe_ggtt_node_insert (ggtt, size, alignment);if (IS_ERR (node))return PTR_ERR (node); - err =xe_ggtt_node_insert (node, size, alignment); -if (unlikely (err)) - goto err; -xe_ggtt_assign (node, vfid);xe_gt_sriov_dbg_verbose (gt,"VF%u assigned GGTT %llx-%llx\n" , vfid,xe_ggtt_node_addr (node),xe_ggtt_node_addr (node) + size -1 );
@@ -505,7 +487,7 @@ static int pf_provision_vf_ggtt(struct xe_gt *gt, unsigned int vfid, u64 size) config->ggtt_size = size;return 0 ; err: -pf_release_ggtt (tile, node); +xe_ggtt_node_remove (node, false);return err; }
Signature · 1 line
2.51.0
PATCH v6 12/12 drm/xe: Remove xe_ggtt_node_allocated
In reply to Maarten LankhorstWith the intermediate state gone, no longer useful. Just check against NULL.
--- drivers/gpu/drm/xe/display/xe_fb_pin.c | 3 +-- drivers/gpu/drm/xe/xe_ggtt.c | 14 -------------- drivers/gpu/drm/xe/xe_ggtt.h | 1 - drivers/gpu/drm/xe/xe_gt_sriov_pf_config.c | 12 ++++++------ 4 files changed, 7 insertions(+), 23 deletions(-)
drivers/gpu/drm/xe/display/xe_fb_pin.c +1 -2
@@ -350,8 +350,7 @@ static void __xe_unpin_fb_vma(struct i915_vma *vma)if (vma->dpt)xe_bo_unpin_map_no_vm (vma->dpt); -else if (!xe_ggtt_node_allocated (vma->bo->ggtt_node[tile_id]) || - vma->bo->ggtt_node[tile_id] != vma->node) +else if (vma->bo->ggtt_node[tile_id] != vma->node)xe_ggtt_node_remove (vma->node, false);ttm_bo_reserve (&vma->bo->ttm, false, false,NULL );
drivers/gpu/drm/xe/xe_ggtt.c +0 -14
@@ -643,20 +643,6 @@ struct xe_ggtt_node *xe_ggtt_node_insert(struct xe_ggtt *ggtt, u32 size, u32 alireturn node; } -/** -* xe_ggtt_node_allocated - Check if node is allocated in GGTT -* @node: the &xe_ggtt_node to be inspected -* -* Return: True if allocated, False otherwise. -*/ -bool xe_ggtt_node_allocated (const struct xe_ggtt_node *node) -{ -if (!node || !node->ggtt) -return false; - -return drm_mm_node_allocated (&node->base); -} -/** * xe_ggtt_map_bo - Map the BO into GGTT * @ggtt: the &xe_ggtt where node will be mapped
drivers/gpu/drm/xe/xe_ggtt.h +0 -1
@@ -30,7 +30,6 @@ xe_ggtt_node_insert_transform(struct xe_ggtt *ggtt, u64 size, u32 align, xe_ggtt_transform_cb transform, void *arg);void xe_ggtt_node_remove (struct xe_ggtt_node *node,bool invalidate); -bool xe_ggtt_node_allocated (const struct xe_ggtt_node *node);void xe_ggtt_map_bo_unlocked (struct xe_ggtt *ggtt,struct xe_bo *bo);int xe_ggtt_insert_bo (struct xe_ggtt *ggtt,struct xe_bo *bo,struct drm_exec *exec); int xe_ggtt_insert_bo_at(struct xe_ggtt *ggtt, struct xe_bo *bo,
drivers/gpu/drm/xe/xe_gt_sriov_pf_config.c +6 -6
@@ -261,7 +261,7 @@ static u32 encode_config_ggtt(u32 *cfg, const struct xe_gt_sriov_config *config, {struct xe_ggtt_node *node = config->ggtt_region; -if (!xe_ggtt_node_allocated (node)) +if (!node)return 0 ;return encode_ggtt (cfg,xe_ggtt_node_addr (node), config->ggtt_size, details);
@@ -455,7 +455,7 @@ static int pf_provision_vf_ggtt(struct xe_gt *gt, unsigned int vfid, u64 size) size =round_up (size, alignment); -if (xe_ggtt_node_allocated (config->ggtt_region)) { +if (config->ggtt_region) { err =pf_distribute_config_ggtt (tile, vfid,0 ,0 );if (unlikely (err))return err;
@@ -466,7 +466,7 @@ static int pf_provision_vf_ggtt(struct xe_gt *gt, unsigned int vfid, u64 size)if (unlikely (err))return err; } -xe_gt_assert (gt, !xe_ggtt_node_allocated (config->ggtt_region)); +xe_gt_assert (gt, !config->ggtt_region);if (!size)return 0 ;
@@ -497,7 +497,7 @@ static u64 pf_get_vf_config_ggtt(struct xe_gt *gt, unsigned int vfid)struct xe_ggtt_node *node = config->ggtt_region;xe_gt_assert (gt,xe_gt_is_main_type (gt)); -return xe_ggtt_node_allocated (node) ? config->ggtt_size :0 ; +return node ? config->ggtt_size :0 ; }/**
@@ -2049,7 +2049,7 @@ int xe_gt_sriov_pf_config_release(struct xe_gt *gt, unsigned int vfid, bool forcstatic void pf_sanitize_ggtt (struct xe_ggtt_node *ggtt_region,unsigned int vfid) { -if (xe_ggtt_node_allocated (ggtt_region)) +if (ggtt_region)xe_ggtt_assign (ggtt_region, vfid); }
@@ -2496,7 +2496,7 @@ int xe_gt_sriov_pf_config_print_ggtt(struct xe_gt *gt, struct drm_printer *p)for (n =1 ; n <= total_vfs; n++) { config = >->sriov.pf.vfs[n].config; -if (!xe_ggtt_node_allocated (config->ggtt_region)) +if (!config->ggtt_region)continue ; string_get_size(config->ggtt_size, 1, STRING_UNITS_2,
Signature · 1 line
2.51.0