Add memory page offlining support
Subsystems (4)
v7 cover Add memory page offlining support
Add memory page offlining support
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit
This functionality represents a significant step in making the xe driver gracefully handle hardware memory degradation. By integrating with the DRM Buddy allocator, the driver can permanently "carve out" faulty memory so it isn't reused by subsequent allocations.
This series adds memory page offlining support with following:
1. Link VRAM object with gpu buddy
2. Integrate lockdep for gpu buddy manager
3. Link and track ttm BO's with physical addresses
4. Link LRC BO and its execution Queue
5. Extend BO purge to handle vram pages as well
6. Handle the generated physical address error by reserving addresses 4K page
7. Adds supporting debugfs to automate injection of physcal address error
8. Add buddy block allocation dump for debuggin buddy related issues
9. Add configfs for vram bad page reservation policy
10. Sysfs entry to provide statistics of bad gpu vram pages for user info
v7:
- Improve debugfs warning messages
- Use scope_guard for locking(MattB)
- Adapt addition of queue member of LRC BO(MattB)
- Extend and use xe_ttm_bo_purge API for vram pages(MattB)
- Handle dma_buf_map requests for native and remote(MattB)
- Address if in never initialized block, set block to NULL
- Add lockdep in gpu buddy (MattB)
- Correct allocated_addr_to_block logic (MattA)
V6:
- Add more specific tests to noncritical bo sections
- Handle smooth exit of user created exec queues
- Break code and make purge specific static API
V5:
- Sysfs "max_pages" addition
- Reset block->private NULL post purge
- Remove wedge, return -EIO to system controller will initiate reset
- Add debugfs tests to trigger different test scenarios manually and via igt
- Rename addr_to_tbo to addr_to_block and move under gpu/buddy.c
V4: API reworks, add configfs for policy reservation and apply config everywhere
V3: use res_to_mem_region to avoid use of block->private (MattA)
V2:
- some fixes and clean up on errors
- Added xe_vram_addr_to_region helper to avoid other use of block->private (MattB)
Debugfs shows test of different scenarios,
echo 0 > /sys/kernel/debug/dri/bdf/invalid_addr_vram0 where 0 is below address types to be tested, enum mempage_offline_mode {
MEMPAGE_OFFLINE_UNALLOCATED = 0, MEMPAGE_OFFLINE_USER_ALLOCATED = 1, MEMPAGE_OFFLINE_KERNEL_USER_GGTT_ALLOCATED = 2, MEMPAGE_OFFLINE_KERNEL_USER_PPGTT_ALLOCATED = 3, MEMPAGE_OFFLINE_KERNEL_CRITICAL_ALLOCATED = 4, MEMPAGE_OFFLINE_RESERVED = 5,
};
IGT tests for testing this feature:
https://patchwork.freedesktop.org/patch/714751/
Results of above tests:
Using IGT_SRANDOM=1774610050 for randomisation
Opened device: /dev/dri/card0
Starting subtest: unallocated
Subtest unallocated: SUCCESS (1.834s)
Starting subtest: user-allocated
Subtest user-allocated: SUCCESS (1.832s)
Starting subtest: user-ggtt-allocated
Subtest user-ggtt-allocated: SUCCESS (1.871s)
Starting subtest: user-ppgtt-allocated
Subtest user-ppgtt-allocated: SUCCESS (1.843s)
Starting subtest: critical-allocated
Subtest critical-allocated: SUCCESS (1.824s)
Starting subtest: reserved
Subtest reserved: SUCCESS (0.032s)
Tejas Upadhyay (10):
drm/xe: Link VRAM object with gpu buddy gpu/buddy: Integrate lockdep for gpu buddy manager drm/gpu: Add gpu_buddy_allocated_addr_to_block helper drm/xe: Link LRC BO and its execution Queue drm/xe: Extend BO purge to handle vram pages as well drm/xe: Handle physical memory address error drm/xe/cri: Add debugfs to inject faulty vram address gpu/buddy: Add routine to dump allocated buddy blocks drm/xe/configfs: Add vram bad page reservation policy drm/xe/cri: Add sysfs interface for bad gpu vram pages
drivers/gpu/buddy.c | 114 ++++++- drivers/gpu/drm/drm_buddy.c | 7 +- drivers/gpu/drm/xe/xe_bo.c | 16 +- drivers/gpu/drm/xe/xe_bo.h | 5 +- drivers/gpu/drm/xe/xe_bo_types.h | 3 + drivers/gpu/drm/xe/xe_configfs.c | 64 +++- drivers/gpu/drm/xe/xe_configfs.h | 2 + drivers/gpu/drm/xe/xe_debugfs.c | 171 ++++++++++ drivers/gpu/drm/xe/xe_device_sysfs.c | 7 + drivers/gpu/drm/xe/xe_dma_buf.c | 3 + drivers/gpu/drm/xe/xe_exec_queue.c | 10 +- drivers/gpu/drm/xe/xe_pt.c | 3 +- drivers/gpu/drm/xe/xe_ttm_vram_mgr.c | 371 +++++++++++++++++++++ drivers/gpu/drm/xe/xe_ttm_vram_mgr.h | 2 + drivers/gpu/drm/xe/xe_ttm_vram_mgr_types.h | 32 ++ include/linux/gpu_buddy.h | 44 +++ 16 files changed, 839 insertions(+), 15 deletions(-)
Signature · 1 line
2.52.0
RFC PATCH v7 1/10 drm/xe: Link VRAM object with gpu buddy
Setup to link TTM buffer object inside gpu buddy. This functionality is critical for supporting the memory page offline feature on CRI, where identified faulty pages must be traced back to their originating buffer for safe removal.
V2(MattB): Clear block->private in xe_ttm_vram_mgr_del as well
--- drivers/gpu/drm/xe/xe_ttm_vram_mgr.c | 6 ++++++ 1 file changed, 6 insertions(+)
drivers/gpu/drm/xe/xe_ttm_vram_mgr.c +6 -0
@@ -54,6 +54,7 @@ static int xe_ttm_vram_mgr_new(struct ttm_resource_manager *man,struct xe_ttm_vram_mgr *mgr =to_xe_ttm_vram_mgr (man);struct xe_ttm_vram_mgr_resource *vres;struct gpu_buddy *mm = &mgr->mm; +struct gpu_buddy_block *block;u64 size, min_page_size;unsigned long lpfn;int err;
@@ -138,6 +139,8 @@ static int xe_ttm_vram_mgr_new(struct ttm_resource_manager *man, } mgr->visible_avail -= vres->used_visible_size; +list_for_each_entry (block, &vres->blocks, link) + block->private = tbo;mutex_unlock (&mgr->lock); if (!(vres->base.placement & TTM_PL_FLAG_CONTIGUOUS) &&
@@ -176,8 +179,11 @@ static void xe_ttm_vram_mgr_del(struct ttm_resource_manager *man,to_xe_ttm_vram_mgr_resource (res);struct xe_ttm_vram_mgr *mgr =to_xe_ttm_vram_mgr (man);struct gpu_buddy *mm = &mgr->mm; +struct gpu_buddy_block *block;mutex_lock (&mgr->lock); +list_for_each_entry (block, &vres->blocks, link) + block->private =NULL ;gpu_buddy_free_list (mm, &vres->blocks,0 ); mgr->visible_avail += vres->used_visible_size;mutex_unlock (&mgr->lock);
Signature · 1 line
2.52.0
RFC PATCH v7 2/10 gpu/buddy: Integrate lockdep for gpu buddy manager
In reply to Tejas UpadhyayIntegrating lockdep into the gpu_buddy manager as standard practice for verifying that internal resources are correctly protected by their associated locks.
--- drivers/gpu/buddy.c | 18 ++++++++++-- drivers/gpu/drm/drm_buddy.c | 7 +++-- drivers/gpu/drm/xe/xe_ttm_vram_mgr.c | 3 ++ include/linux/gpu_buddy.h | 41 ++++++++++++++++++++++++++++ 4 files changed, 65 insertions(+), 4 deletions(-)
drivers/gpu/buddy.c +16 -2
@@ -437,6 +437,9 @@ int gpu_buddy_init(struct gpu_buddy *mm, u64 size, u64 chunk_size) root_count++; }while (size); +#ifdef CONFIG_LOCKDEP + mm->lock_dep_map =NULL ; +#endif return 0 ; out_free_roots:
@@ -464,6 +467,7 @@ void gpu_buddy_fini(struct gpu_buddy *mm)unsigned int order;int i; +gpu_buddy_driver_lock_held (mm); size = mm->size;for (i =0 ; i < mm->n_roots; ++i) {
@@ -538,6 +542,7 @@ void gpu_buddy_reset_clear(struct gpu_buddy *mm, bool is_clear)unsigned int order;int i; +gpu_buddy_driver_lock_held (mm); size = mm->size;for (i =0 ; i < mm->n_roots; ++i) { order =ilog2 (size) -ilog2 (mm->chunk_size);
@@ -580,6 +585,7 @@ EXPORT_SYMBOL(gpu_buddy_reset_clear);void gpu_buddy_free_block (struct gpu_buddy *mm,struct gpu_buddy_block *block) { +gpu_buddy_driver_lock_held (mm);BUG_ON (!gpu_buddy_block_is_allocated (block)); mm->avail +=gpu_buddy_block_size (mm, block);if (gpu_buddy_block_is_clear (block))
@@ -633,6 +639,7 @@ void gpu_buddy_free_list(struct gpu_buddy *mm, {bool mark_clear = flags &GPU_BUDDY_CLEARED ; +gpu_buddy_driver_lock_held (mm);__gpu_buddy_free_list (mm, objects, mark_clear, !mark_clear); }EXPORT_SYMBOL (gpu_buddy_free_list);
@@ -1172,6 +1179,8 @@ int gpu_buddy_block_trim(struct gpu_buddy *mm,u64 new_start;int err; +gpu_buddy_driver_lock_held (mm); +if (!list_is_singular (blocks))return -EINVAL ;
@@ -1287,6 +1296,8 @@ int gpu_buddy_alloc_blocks(struct gpu_buddy *mm,unsigned long pages;int err; +gpu_buddy_driver_lock_held (mm); +if (size < mm->chunk_size)return -EINVAL ;
@@ -1458,9 +1469,11 @@ EXPORT_SYMBOL(gpu_buddy_alloc_blocks);void gpu_buddy_block_print (struct gpu_buddy *mm,struct gpu_buddy_block *block) { -u64 start =gpu_buddy_block_offset (block); -u64 size =gpu_buddy_block_size (mm, block); +u64 start, size; +gpu_buddy_driver_lock_held (mm); + start =gpu_buddy_block_offset (block); + size =gpu_buddy_block_size (mm, block);pr_info ("%#018llx-%#018llx: %llu\n" , start, start + size, size); }EXPORT_SYMBOL (gpu_buddy_block_print);
@@ -1475,6 +1488,7 @@ void gpu_buddy_print(struct gpu_buddy *mm) {int order; +gpu_buddy_driver_lock_held (mm);pr_info ("chunk_size: %lluKiB, total: %lluMiB, free: %lluMiB, clear_free: %lluMiB\n" , mm->chunk_size >>10 , mm->size >>20 , mm->avail >>20 , mm->clear_avail >>20 );
drivers/gpu/drm/drm_buddy.c +5 -2
@@ -25,9 +25,11 @@ void drm_buddy_block_print(struct gpu_buddy *mm,struct gpu_buddy_block *block, struct drm_printer *p) { - u64 start = gpu_buddy_block_offset(block); -u64 size =gpu_buddy_block_size (mm, block); + u64 start, size; +gpu_buddy_driver_lock_held (mm); + start =gpu_buddy_block_offset (block); + size =gpu_buddy_block_size (mm, block);drm_printf (p,"%#018llx-%#018llx: %llu\n" , start, start + size, size); }EXPORT_SYMBOL (drm_buddy_block_print);
@@ -42,6 +44,7 @@ void drm_buddy_print(struct gpu_buddy *mm, struct drm_printer *p) {int order; +gpu_buddy_driver_lock_held (mm);drm_printf (p,"chunk_size: %lluKiB, total: %lluMiB, free: %lluMiB, clear_free: %lluMiB\n" , mm->chunk_size >>10 , mm->size >>20 , mm->avail >>20 , mm->clear_avail >>20 );
drivers/gpu/drm/xe/xe_ttm_vram_mgr.c +3 -0
@@ -293,7 +293,9 @@ static void xe_ttm_vram_mgr_fini(struct drm_device *dev, void *arg)WARN_ON_ONCE (mgr->visible_avail != mgr->visible_size); +mutex_lock (&mgr->lock);gpu_buddy_fini (&mgr->mm); +mutex_unlock (&mgr->lock);ttm_resource_manager_cleanup (&mgr->manager);
@@ -328,6 +330,7 @@ int __xe_ttm_vram_mgr_init(struct xe_device *xe, struct xe_ttm_vram_mgr *mgr,if (err)return err; +gpu_buddy_driver_set_lock (&mgr->mm, &mgr->lock);ttm_set_driver_manager (&xe->ttm, mem_type, &mgr->manager);ttm_resource_manager_set_used (&mgr->manager, true);
include/linux/gpu_buddy.h +41 -0
55 lines not on this page · raw message
Signature · 1 line
2.52.0
Matthew Auld
maintainerRe: [RFC PATCH V7 02/10] gpu/buddy: Integrate lockdep for gpu buddy manager
Quoted 77 lines · Tejas Upadhyay
Integrating lockdep into the
gpu_buddymanager as standard practice for verifying that internal resources are correctly protected by their associated locks.Signed-off-by Tejas Upadhyay <[email protected]>--- drivers/gpu/buddy.c | 18 ++++++++++-- drivers/gpu/drm/drm_buddy.c | 7 +++-- drivers/gpu/drm/xe/xe_ttm_vram_mgr.c | 3 ++ include/linux/gpu_buddy.h | 41 ++++++++++++++++++++++++++++ 4 files changed, 65 insertions(+), 4 deletions(-)drivers/gpu/buddy.c +11
-0@@ -437,6 +437,9 @@ int gpu_buddy_init(struct gpu_buddy *mm, u64 size, u64 chunk_size) root_count++; }while (size); +#ifdef CONFIG_LOCKDEP + mm->lock_dep_map =NULL ; +#endif return 0 ; out_free_roots:@@ -464,6 +467,7 @@ void gpu_buddy_fini(struct gpu_buddy *mm)unsigned int order;int i; +gpu_buddy_driver_lock_held (mm); size = mm->size;for (i =0 ; i < mm->n_roots; ++i) {@@ -538,6 +542,7 @@ void gpu_buddy_reset_clear(struct gpu_buddy *mm, bool is_clear)unsigned int order;int i; +gpu_buddy_driver_lock_held (mm); size = mm->size;for (i =0 ; i < mm->n_roots; ++i) { order =ilog2 (size) -ilog2 (mm->chunk_size);@@ -580,6 +585,7 @@ EXPORT_SYMBOL(gpu_buddy_reset_clear);void gpu_buddy_free_block (struct gpu_buddy *mm,struct gpu_buddy_block *block) { +gpu_buddy_driver_lock_held (mm);BUG_ON (!gpu_buddy_block_is_allocated (block)); mm->avail +=gpu_buddy_block_size (mm, block);if (gpu_buddy_block_is_clear (block))@@ -633,6 +639,7 @@ void gpu_buddy_free_list(struct gpu_buddy *mm, {bool mark_clear = flags &GPU_BUDDY_CLEARED ; +gpu_buddy_driver_lock_held (mm);__gpu_buddy_free_list (mm, objects, mark_clear, !mark_clear); }EXPORT_SYMBOL (gpu_buddy_free_list);@@ -1172,6 +1179,8 @@ int gpu_buddy_block_trim(struct gpu_buddy *mm,u64 new_start;int err; +gpu_buddy_driver_lock_held (mm); +if (!list_is_singular (blocks))return -EINVAL ;@@ -1287,6 +1296,8 @@ int gpu_buddy_alloc_blocks(struct gpu_buddy *mm,unsigned long pages;int err; +gpu_buddy_driver_lock_held (mm); +if (size < mm->chunk_size)return -EINVAL ;
drivers/gpu/buddy.c EXPORT_SYMBOL(gpu_buddy_alloc_blocks); · line 1474
Quoted 8 lines
drivers/gpu/buddy.c +2
-2@@ -1458,9 +1469,11 @@ EXPORT_SYMBOL(gpu_buddy_alloc_blocks);void gpu_buddy_block_print (struct gpu_buddy *mm,struct gpu_buddy_block *block) { -u64 start =gpu_buddy_block_offset (block); -u64 size =gpu_buddy_block_size (mm, block); +u64 start, size; +gpu_buddy_driver_lock_held (mm);
I don't think we want this one. The mm interaction is just for immutable state, and the block itself is essentially owned by the caller. Same reason why we don't want annotations for stuff like gpu_buddy_block_offset() etc.
Quoted 18 lines
drivers/gpu/buddy.c +3
-0+ start =gpu_buddy_block_offset (block); + size =gpu_buddy_block_size (mm, block);pr_info ("%#018llx-%#018llx: %llu\n" , start, start + size, size); }EXPORT_SYMBOL (gpu_buddy_block_print);@@ -1475,6 +1488,7 @@ void gpu_buddy_print(struct gpu_buddy *mm) {int order; +gpu_buddy_driver_lock_held (mm);pr_info ("chunk_size: %lluKiB, total: %lluMiB, free: %lluMiB, clear_free: %lluMiB\n" , mm->chunk_size >>10 , mm->size >>20 , mm->avail >>20 , mm->clear_avail >>20 );drivers/gpu/drm/drm_buddy.c +0
-0
drivers/gpu/drm/drm_buddy.c void drm_buddy_block_print(struct gpu_buddy *mm, · line 32
Quoted 10 lines
drivers/gpu/drm/drm_buddy.c +4
-2@@ -25,9 +25,11 @@ void drm_buddy_block_print(struct gpu_buddy *mm,struct gpu_buddy_block *block, struct drm_printer *p) { - u64 start = gpu_buddy_block_offset(block); -u64 size =gpu_buddy_block_size (mm, block); + u64 start, size; +gpu_buddy_driver_lock_held (mm); + start =gpu_buddy_block_offset (block); + size =gpu_buddy_block_size (mm, block);
Same here.
Quoted 16 lines
drivers/gpu/drm/drm_buddy.c +1
-0drm_printf (p,"%#018llx-%#018llx: %llu\n" , start, start + size, size); }EXPORT_SYMBOL (drm_buddy_block_print);@@ -42,6 +44,7 @@ void drm_buddy_print(struct gpu_buddy *mm, struct drm_printer *p) {int order; +gpu_buddy_driver_lock_held (mm);drm_printf (p,"chunk_size: %lluKiB, total: %lluMiB, free: %lluMiB, clear_free: %lluMiB\n" , mm->chunk_size >>10 , mm->size >>20 , mm->avail >>20 , mm->clear_avail >>20 );drivers/gpu/drm/xe/xe_ttm_vram_mgr.c +0
-0
drivers/gpu/drm/xe/xe_ttm_vram_mgr.c static void xe_ttm_vram_mgr_fini(struct drm_device *dev, void *arg) · line 298
Quoted 6 lines
drivers/gpu/drm/xe/xe_ttm_vram_mgr.c +2
-0@@ -293,7 +293,9 @@ static void xe_ttm_vram_mgr_fini(struct drm_device *dev, void *arg)WARN_ON_ONCE (mgr->visible_avail != mgr->visible_size); +mutex_lock (&mgr->lock);gpu_buddy_fini (&mgr->mm); +mutex_unlock (&mgr->lock);
This shouldn't need a lock. Annotation for this one should also be dropped.
Quoted 72 lines
drivers/gpu/drm/xe/xe_ttm_vram_mgr.c +1
-0ttm_resource_manager_cleanup (&mgr->manager);@@ -328,6 +330,7 @@ int __xe_ttm_vram_mgr_init(struct xe_device *xe, struct xe_ttm_vram_mgr *mgr,if (err)return err; +gpu_buddy_driver_set_lock (&mgr->mm, &mgr->lock);ttm_set_driver_manager (&xe->ttm, mem_type, &mgr->manager);ttm_resource_manager_set_used (&mgr->manager, true);include/linux/gpu_buddy.h +41
-0@@ -154,6 +154,7 @@ struct gpu_buddy_block {* @avail: Total free space currently available for allocation in bytes. * @clear_avail: Free space available in the clear tree (zeroed memory) in bytes. * This is a subset of @avail. +* @lock_dep_map: Annotates gpu_buddy API with a driver provided lock. */ struct gpu_buddy {/* private: */ @@ -179,8 +180,48 @@ struct gpu_buddy {u64 size;u64 avail;u64 clear_avail; +#ifdef CONFIG_LOCKDEP +struct lockdep_map *lock_dep_map; +#endif }; +#ifdef CONFIG_LOCKDEP +/** +* gpu_buddy_driver_set_lock() - Set the lock protecting accesses to GPU BUDDY +* @mm: Pointer to GPU buddy structure. +* @lock: the lock used to protect the gpu buddy. The locking primitive +* must contain a dep_map field. +* +* Call this to annotate gpu_buddy APIs which access/modify gpu_buddy manager +*/ +#define gpu_buddy_driver_set_lock (mm, lock) \ + do { \ + struct gpu_buddy *__mm = (mm); \ + if (!WARN(__mm->lock_dep_map, "GPU BUDDY MM lock should be set only once.")) \ + __mm->lock_dep_map = &(lock)->dep_map; \ + } while (0) +#else +#define gpu_buddy_driver_set_lock (mm, lock) do { (void)(mm); (void)(lock); } while (0) +#endif + +#ifdef CONFIG_LOCKDEP +/** +* gpu_buddy_driver_lock_held() - Assert GPU BUDDY manager lock is held +* @mm: Pointer to the GPU BUDDY structure. +* +* Ensure driver lock is held. +*/ +static inline void gpu_buddy_driver_lock_held (struct gpu_buddy *mm) +{ +if ((mm)->lock_dep_map) +lockdep_assert (lock_is_held_type ((mm)->lock_dep_map,0 )); +} +#else +static inline gpu_buddy_driver_lock_held(struct gpu_buddy *mm) +{ +} +#endif +static inline u64 gpu_buddy_block_offset (const struct gpu_buddy_block *block) {
RE: [RFC PATCH V7 02/10] gpu/buddy: Integrate lockdep for gpu buddy manager
Quoted 102 lines
-----Original Message-----
From: Auld, Matthew <[email protected]>
Sent: 16 April 2026 14:25
To: Upadhyay, Tejas <[email protected]>; intel- [email protected]
Cc: Brost, Matthew <[email protected]>; [email protected]; Ghimiray, Himal Prasad <[email protected]>
Subject: Re: [RFC PATCH V7 02/10] gpu/buddy: Integrate lockdep for gpu buddy managerdrivers/gpu/buddy.c int gpu_buddy_init(struct gpu_buddy *mm, u64
Quoted 17 lines · Tejas Upadhyay
Integrating lockdep into the
gpu_buddymanager as standard practice for verifying that internal resources are correctly protected by their associated locks.Signed-off-by Tejas Upadhyay <[email protected]>--- drivers/gpu/buddy.c | 18 ++++++++++-- drivers/gpu/drm/drm_buddy.c | 7 +++-- drivers/gpu/drm/xe/xe_ttm_vram_mgr.c | 3 ++ include/linux/gpu_buddy.h | 41 ++++++++++++++++++++++++++++ 4 files changed, 65 insertions(+), 4 deletions(-)drivers/gpu/buddy.c index renamed from drivers/gpu/buddy.c +0
-0
52686672e99f..53ff85ac2105100644drivers/gpu/buddy.c +0
-0@@ -437,6 +437,9 @@ int gpu_buddy_init(struct gpu_buddy *mm, u64size, u64
chunk_size)Quoted 21 lines
drivers/gpu/buddy.c +4
-0root_count++; }while (size); +#ifdef CONFIG_LOCKDEP + mm->lock_dep_map =NULL ; +#endif return 0 ; out_free_roots:@@ -464,6 +467,7 @@ void gpu_buddy_fini(struct gpu_buddy *mm)unsigned int order;int i; +gpu_buddy_driver_lock_held (mm); size = mm->size; for (i = 0; i < mm->n_roots; ++i) { @@ -538,6 +542,7 @@ void
gpu_buddy_reset_clear(struct gpu_buddy *mm, bool is_clear)
unsigned int order;
int i;drivers/gpu/buddy.c void gpu_buddy_fini(struct gpu_buddy *mm)
Quoted 4 lines
drivers/gpu/buddy.c +1
-0+gpu_buddy_driver_lock_held (mm); size = mm->size;for (i =0 ; i < mm->n_roots; ++i) { order = ilog2(size) - ilog2(mm->chunk_size); @@ -580,6+585,7 @@
drivers/gpu/buddy.c int gpu_buddy_block_trim(struct gpu_buddy
Quoted 17 lines
EXPORT_SYMBOL(gpu_buddy_reset_clear);
void gpu_buddy_free_block(structgpu_buddy*mm,
structgpu_buddy_block*block)
{drivers/gpu/buddy.c +1
-0+gpu_buddy_driver_lock_held (mm);BUG_ON (!gpu_buddy_block_is_allocated (block)); mm->avail +=gpu_buddy_block_size (mm, block); if (gpu_buddy_block_is_clear(block)) @@ -633,6 +639,7 @@ voidgpu_buddy_free_list(struct
gpu_buddy*mm,
{
boolmark_clear= flags &GPU_BUDDY_CLEARED;drivers/gpu/buddy.c +1
-0+gpu_buddy_driver_lock_held (mm);__gpu_buddy_free_list (mm, objects, mark_clear, !mark_clear); }EXPORT_SYMBOL (gpu_buddy_free_list);@@ -1172,6 +1179,8 @@ int gpu_buddy_block_trim(struct gpu_buddy*mm,
drivers/gpu/buddy.c int gpu_buddy_alloc_blocks(struct gpu_buddy
Quoted 9 lines
drivers/gpu/buddy.c +2
-0u64 new_start;int err; +gpu_buddy_driver_lock_held (mm); +if (!list_is_singular (blocks))return -EINVAL ;@@ -1287,6 +1296,8 @@ int gpu_buddy_alloc_blocks(struct gpu_buddy*mm,
Quoted 16 lines
drivers/gpu/buddy.c +3
-2unsigned long pages;int err; +gpu_buddy_driver_lock_held (mm); +if (size < mm->chunk_size)return -EINVAL ;@@ -1458,9 +1469,11 @@ EXPORT_SYMBOL(gpu_buddy_alloc_blocks);void gpu_buddy_block_print (struct gpu_buddy *mm,struct gpu_buddy_block *block) { -u64 start =gpu_buddy_block_offset (block); -u64 size =gpu_buddy_block_size (mm, block); +u64 start, size;
Quoted 6 lines
drivers/gpu/buddy.c EXPORT_SYMBOL(gpu_buddy_alloc_blocks); · line 1474
Quoted 1 line
drivers/gpu/buddy.c +1
-0+gpu_buddy_driver_lock_held (mm);I don't think we want this one. The mm interaction is just for immutable state, and the block itself is essentially owned by the caller. Same reason why we don't want annotations for stuff like
gpu_buddy_block_offset()etc.
Yes, makes sense. I will change it.
Quoted 29 lines
drivers/gpu/buddy.c void gpu_buddy_print(struct gpu_buddy *mm) · line 1492
Quoted 12 lines
drivers/gpu/buddy.c +3
-0+ start =gpu_buddy_block_offset (block); + size =gpu_buddy_block_size (mm, block);pr_info ("%#018llx-%#018llx: %llu\n" , start, start + size, size); }EXPORT_SYMBOL (gpu_buddy_block_print);@@ -1475,6 +1488,7 @@ void gpu_buddy_print(struct gpu_buddy *mm) {int order; +gpu_buddy_driver_lock_held (mm); pr_info("chunk_size: %lluKiB, total: %lluMiB, free: %lluMiB, clear_free:%lluMiB\n",
drivers/gpu/drm/drm_buddy.c void drm_buddy_block_print(struct gpu_buddy
Quoted 8 lines
drivers/gpu/buddy.c +0
-0mm->chunk_size >>10 , mm->size >>20 , mm->avail >>20 , mm->clear_avail >> 20);drivers/gpu/drm/drm_buddy.c +0
-0@@ -25,9 +25,11 @@ void drm_buddy_block_print(struct gpu_buddy*mm,
Quoted 7 lines
drivers/gpu/drm/drm_buddy.c +1
-2struct gpu_buddy_block *block, struct drm_printer *p) { - u64 start = gpu_buddy_block_offset(block); -u64 size =gpu_buddy_block_size (mm, block); + u64 start, size;
Quoted 5 lines
drivers/gpu/drm/drm_buddy.c void drm_buddy_block_print(struct gpu_buddy · line 32
Quoted 3 lines
drivers/gpu/drm/drm_buddy.c +3
-0+gpu_buddy_driver_lock_held (mm); + start =gpu_buddy_block_offset (block); + size =gpu_buddy_block_size (mm, block);Same here.
Yes
Quoted 24 lines
drivers/gpu/drm/drm_buddy.c void drm_buddy_print(struct gpu_buddy *mm, struct
Quoted 5 lines
drivers/gpu/drm/drm_buddy.c +0
-0drm_printf (p,"%#018llx-%#018llx: %llu\n" , start, start + size, size); }EXPORT_SYMBOL (drm_buddy_block_print);@@ -42,6 +44,7 @@ void drm_buddy_print(struct gpu_buddy *mm, struct
drm_printer*p)drivers/gpu/drm/drm_buddy.c void drm_buddy_print(struct gpu_buddy *mm, struct · line 48
Quoted 5 lines
drivers/gpu/drm/drm_buddy.c +1
-0{int order; +gpu_buddy_driver_lock_held (mm); drm_printf(p, "chunk_size: %lluKiB, total: %lluMiB, free: %lluMiB,
clear_free: %lluMiB\n",drivers/gpu/drm/xe/xe_ttm_vram_mgr.c static void xe_ttm_vram_mgr_fini(struct
Quoted 9 lines
drivers/gpu/drm/drm_buddy.c +0
-0mm->chunk_size >>10 , mm->size >>20 , mm->avail >>20 , mm->clear_avail >> 20);drivers/gpu/drm/xe/xe_ttm_vram_mgr.c +0
-0@@ -293,7 +293,9 @@ static void xe_ttm_vram_mgr_fini(struct
drm_deviceQuoted 2 lines
drivers/gpu/drm/xe/xe_ttm_vram_mgr.c +0
-0*dev, void *arg)
Quoted 7 lines
drivers/gpu/drm/xe/xe_ttm_vram_mgr.c static void xe_ttm_vram_mgr_fini(struct drm_device *dev, void *arg) · line 298
Quoted 5 lines
drivers/gpu/drm/xe/xe_ttm_vram_mgr.c +2
-0WARN_ON_ONCE (mgr->visible_avail != mgr->visible_size); +mutex_lock (&mgr->lock);gpu_buddy_fini (&mgr->mm); +mutex_unlock (&mgr->lock);This shouldn't need a lock. Annotation for this one should also be dropped.
Thought was that while we call buddy_fini, there could be in flight access of buddy manager, so some new allocation in progress and thus it will provide protection against it. Please let me know what you think!
Tejas
Quoted 74 lines
drivers/gpu/drm/xe/xe_ttm_vram_mgr.c int __xe_ttm_vram_mgr_init(struct xe_device *xe,
Quoted 5 lines
drivers/gpu/drm/xe/xe_ttm_vram_mgr.c +0
-0ttm_resource_manager_cleanup (&mgr->manager);@@ -328,6 +330,7 @@ int __xe_ttm_vram_mgr_init(struct xe_device *xe,struct
xe_ttm_vram_mgr*mgr,Quoted 12 lines
drivers/gpu/drm/xe/xe_ttm_vram_mgr.c +1
-0if (err)return err; +gpu_buddy_driver_set_lock (&mgr->mm, &mgr->lock);ttm_set_driver_manager (&xe->ttm, mem_type, &mgr->manager);ttm_resource_manager_set_used (&mgr->manager, true);include/linux/gpu_buddy.h +0
-0include/linux/gpu_buddy.h struct gpu_buddy_block { · line 155
Quoted 2 lines
include/linux/gpu_buddy.h +0
-0@@ -154,6 +154,7 @@ struct gpu_buddy_block {* @avail: Total free space currently available for allocation in bytes. * @clear_avail: Free space available in the clear tree (zeroed memory) in bytes.
Quoted 18 lines
include/linux/gpu_buddy.h +8
-0* This is a subset of @avail. +* @lock_dep_map: Annotates gpu_buddy API with a driver provided lock. */ struct gpu_buddy {/* private: */ @@ -179,8 +180,48 @@ struct gpu_buddy {u64 size;u64 avail;u64 clear_avail; +#ifdef CONFIG_LOCKDEP +struct lockdep_map *lock_dep_map; +#endif }; +#ifdef CONFIG_LOCKDEP +/** +* gpu_buddy_driver_set_lock() - Set the lock protecting accesses to +GPU BUDDY include/linux/gpu_buddy.h struct gpu_buddy { · line 201
Quoted 10 lines
include/linux/gpu_buddy.h +10
-0+* @mm: Pointer to GPU buddy structure. +* @lock: the lock used to protect the gpu buddy. The locking +primitive +* must contain a dep_map field. +* +* Call this to annotate gpu_buddy APIs which access/modify gpu_buddy +manager */ #define gpu_buddy_driver_set_lock (mm, lock) \ + do { \ + struct gpu_buddy *__mm = (mm); \ + if (!WARN(__mm->lock_dep_map, "GPU BUDDY MM lockshould be set only once.")) \
Quoted 7 lines
include/linux/gpu_buddy.h +7
-0+ __mm->lock_dep_map = &(lock)->dep_map; \ + } while (0) +#else +#define gpu_buddy_driver_set_lock (mm, lock) do { (void)(mm); +(void)(lock); } while (0) #endif + +#ifdef CONFIG_LOCKDEP include/linux/gpu_buddy.h struct gpu_buddy { · line 217
Quoted 10 lines
include/linux/gpu_buddy.h +10
-0+/** +* gpu_buddy_driver_lock_held() - Assert GPU BUDDY manager lock is +held +* @mm: Pointer to the GPU BUDDY structure. +* +* Ensure driver lock is held. +*/ +static inline void gpu_buddy_driver_lock_held (struct gpu_buddy *mm) { +if ((mm)->lock_dep_map) +lockdep_assert (lock_is_held_type ((mm)->lock_dep_map,0 ));} #else
Quoted 6 lines
include/linux/gpu_buddy.h +3
-0+static inline gpu_buddy_driver_lock_held(struct gpu_buddy *mm) { } +#endif +static inline u64 gpu_buddy_block_offset (const struct gpu_buddy_block *block) {
Matthew Auld
maintainerRe: [RFC PATCH V7 02/10] gpu/buddy: Integrate lockdep for gpu buddy manager
Quoted 174 lines · Tejas
Quoted 102 lines
-----Original Message-----
From: Auld, Matthew <[email protected]>
Sent: 16 April 2026 14:25
To: Upadhyay, Tejas <[email protected]>; intel- [email protected]
Cc: Brost, Matthew <[email protected]>; [email protected]; Ghimiray, Himal Prasad <[email protected]>
Subject: Re: [RFC PATCH V7 02/10] gpu/buddy: Integrate lockdep for gpu buddy managerdrivers/gpu/buddy.c int gpu_buddy_init(struct gpu_buddy *mm, u64
Quoted 17 lines · Tejas Upadhyay
Integrating lockdep into the
gpu_buddymanager as standard practice for verifying that internal resources are correctly protected by their associated locks.Signed-off-by Tejas Upadhyay <[email protected]>--- drivers/gpu/buddy.c | 18 ++++++++++-- drivers/gpu/drm/drm_buddy.c | 7 +++-- drivers/gpu/drm/xe/xe_ttm_vram_mgr.c | 3 ++ include/linux/gpu_buddy.h | 41 ++++++++++++++++++++++++++++ 4 files changed, 65 insertions(+), 4 deletions(-)drivers/gpu/buddy.c index renamed from drivers/gpu/buddy.c +0
-0
52686672e99f..53ff85ac2105100644drivers/gpu/buddy.c +0
-0@@ -437,6 +437,9 @@ int gpu_buddy_init(struct gpu_buddy *mm, u64size, u64
chunk_size)Quoted 21 lines
drivers/gpu/buddy.c +4
-0root_count++; }while (size); +#ifdef CONFIG_LOCKDEP + mm->lock_dep_map =NULL ; +#endif return 0 ; out_free_roots:@@ -464,6 +467,7 @@ void gpu_buddy_fini(struct gpu_buddy *mm)unsigned int order;int i; +gpu_buddy_driver_lock_held (mm); size = mm->size; for (i = 0; i < mm->n_roots; ++i) { @@ -538,6 +542,7 @@ void
gpu_buddy_reset_clear(struct gpu_buddy *mm, bool is_clear)unsigned int order; int i;drivers/gpu/buddy.c void gpu_buddy_fini(struct gpu_buddy *mm)
Quoted 4 lines
drivers/gpu/buddy.c +1
-0+gpu_buddy_driver_lock_held (mm); size = mm->size;for (i =0 ; i < mm->n_roots; ++i) { order = ilog2(size) - ilog2(mm->chunk_size); @@ -580,6+585,7 @@
drivers/gpu/buddy.c int gpu_buddy_block_trim(struct gpu_buddy
Quoted 17 lines
EXPORT_SYMBOL(gpu_buddy_reset_clear);
void gpu_buddy_free_block(structgpu_buddy*mm,
structgpu_buddy_block*block)
{drivers/gpu/buddy.c +1
-0+gpu_buddy_driver_lock_held (mm);BUG_ON (!gpu_buddy_block_is_allocated (block)); mm->avail +=gpu_buddy_block_size (mm, block); if (gpu_buddy_block_is_clear(block)) @@ -633,6 +639,7 @@ voidgpu_buddy_free_list(struct
gpu_buddy*mm,{ bool mark_clear = flags & GPU_BUDDY_CLEARED;drivers/gpu/buddy.c +1
-0+gpu_buddy_driver_lock_held (mm);__gpu_buddy_free_list (mm, objects, mark_clear, !mark_clear); }EXPORT_SYMBOL (gpu_buddy_free_list);@@ -1172,6 +1179,8 @@ int gpu_buddy_block_trim(struct gpu_buddy*mm,
drivers/gpu/buddy.c int gpu_buddy_alloc_blocks(struct gpu_buddy
Quoted 9 lines
drivers/gpu/buddy.c +2
-0u64 new_start;int err; +gpu_buddy_driver_lock_held (mm); +if (!list_is_singular (blocks))return -EINVAL ;@@ -1287,6 +1296,8 @@ int gpu_buddy_alloc_blocks(struct gpu_buddy*mm,
Quoted 16 lines
drivers/gpu/buddy.c +3
-2unsigned long pages;int err; +gpu_buddy_driver_lock_held (mm); +if (size < mm->chunk_size)return -EINVAL ;@@ -1458,9 +1469,11 @@ EXPORT_SYMBOL(gpu_buddy_alloc_blocks);void gpu_buddy_block_print (struct gpu_buddy *mm,struct gpu_buddy_block *block) { -u64 start =gpu_buddy_block_offset (block); -u64 size =gpu_buddy_block_size (mm, block); +u64 start, size;Quoted 6 lines
drivers/gpu/buddy.c EXPORT_SYMBOL(gpu_buddy_alloc_blocks); · line 1474
Quoted 1 line
drivers/gpu/buddy.c +1
-0+gpu_buddy_driver_lock_held (mm);I don't think we want this one. The mm interaction is just for immutable state, and the block itself is essentially owned by the caller. Same reason why we don't want annotations for stuff like
gpu_buddy_block_offset()etc.Yes, makes sense. I will change it.
Quoted 29 lines
drivers/gpu/buddy.c void gpu_buddy_print(struct gpu_buddy *mm) · line 1492
Quoted 12 lines
drivers/gpu/buddy.c +3
-0+ start =gpu_buddy_block_offset (block); + size =gpu_buddy_block_size (mm, block);pr_info ("%#018llx-%#018llx: %llu\n" , start, start + size, size); }EXPORT_SYMBOL (gpu_buddy_block_print);@@ -1475,6 +1488,7 @@ void gpu_buddy_print(struct gpu_buddy *mm) {int order; +gpu_buddy_driver_lock_held (mm); pr_info("chunk_size: %lluKiB, total: %lluMiB, free: %lluMiB, clear_free:%lluMiB\n",
drivers/gpu/drm/drm_buddy.c void drm_buddy_block_print(struct gpu_buddy
Quoted 8 lines
drivers/gpu/buddy.c +0
-0mm->chunk_size >>10 , mm->size >>20 , mm->avail >>20 , mm->clear_avail >> 20);drivers/gpu/drm/drm_buddy.c +0
-0@@ -25,9 +25,11 @@ void drm_buddy_block_print(struct gpu_buddy*mm,
Quoted 7 lines
drivers/gpu/drm/drm_buddy.c +1
-2struct gpu_buddy_block *block, struct drm_printer *p) { - u64 start = gpu_buddy_block_offset(block); -u64 size =gpu_buddy_block_size (mm, block); + u64 start, size;Quoted 5 lines
drivers/gpu/drm/drm_buddy.c void drm_buddy_block_print(struct gpu_buddy · line 32
Quoted 3 lines
drivers/gpu/drm/drm_buddy.c +3
-0+gpu_buddy_driver_lock_held (mm); + start =gpu_buddy_block_offset (block); + size =gpu_buddy_block_size (mm, block);Same here.
Yes
Quoted 26 lines
drivers/gpu/drm/drm_buddy.c void drm_buddy_print(struct gpu_buddy *mm, struct
Quoted 5 lines
drivers/gpu/drm/drm_buddy.c +0
-0drm_printf (p,"%#018llx-%#018llx: %llu\n" , start, start + size, size); }EXPORT_SYMBOL (drm_buddy_block_print);@@ -42,6 +44,7 @@ void drm_buddy_print(struct gpu_buddy *mm, struct
drm_printer*p)drivers/gpu/drm/drm_buddy.c void drm_buddy_print(struct gpu_buddy *mm, struct · line 48
Quoted 5 lines
drivers/gpu/drm/drm_buddy.c +1
-0{int order; +gpu_buddy_driver_lock_held (mm); drm_printf(p, "chunk_size: %lluKiB, total: %lluMiB, free: %lluMiB,
clear_free: %lluMiB\n",drivers/gpu/drm/xe/xe_ttm_vram_mgr.c static void xe_ttm_vram_mgr_fini(struct
Quoted 9 lines
drivers/gpu/drm/drm_buddy.c +0
-0mm->chunk_size >>10 , mm->size >>20 , mm->avail >>20 , mm->clear_avail >> 20);drivers/gpu/drm/xe/xe_ttm_vram_mgr.c +0
-0@@ -293,7 +293,9 @@ static void xe_ttm_vram_mgr_fini(struct
drm_deviceQuoted 4 lines
drivers/gpu/drm/xe/xe_ttm_vram_mgr.c +0
-0*dev, void *arg)WARN_ON_ONCE (mgr->visible_avail != mgr->visible_size);
Quoted 7 lines
Quoted 5 lines
drivers/gpu/drm/xe/xe_ttm_vram_mgr.c static void xe_ttm_vram_mgr_fini(struct drm_device *dev, void *arg) · line 298
Quoted 3 lines
drivers/gpu/drm/xe/xe_ttm_vram_mgr.c +2
-0+mutex_lock (&mgr->lock);gpu_buddy_fini (&mgr->mm); +mutex_unlock (&mgr->lock);This shouldn't need a lock. Annotation for this one should also be dropped.
Thought was that while we call
buddy_fini, there could be in flight access of buddy manager, so some new allocation in progress and thus it will provide protection against it. Please let me know what you think!
The mm stuff is about to get nuked, so if something is still messing with this then we are in big trouble anyway, and grabbing the lock does't really help much. The fini() here should be triggered as we start tearing down the driver.
Quoted 77 lines
Tejas
Quoted 74 lines
drivers/gpu/drm/xe/xe_ttm_vram_mgr.c int __xe_ttm_vram_mgr_init(struct xe_device *xe,
Quoted 5 lines
drivers/gpu/drm/xe/xe_ttm_vram_mgr.c +0
-0ttm_resource_manager_cleanup (&mgr->manager);@@ -328,6 +330,7 @@ int __xe_ttm_vram_mgr_init(struct xe_device *xe,struct
xe_ttm_vram_mgr*mgr,Quoted 12 lines
drivers/gpu/drm/xe/xe_ttm_vram_mgr.c +1
-0if (err)return err; +gpu_buddy_driver_set_lock (&mgr->mm, &mgr->lock);ttm_set_driver_manager (&xe->ttm, mem_type, &mgr->manager);ttm_resource_manager_set_used (&mgr->manager, true);include/linux/gpu_buddy.h +0
-0include/linux/gpu_buddy.h struct gpu_buddy_block { · line 155
Quoted 2 lines
include/linux/gpu_buddy.h +0
-0@@ -154,6 +154,7 @@ struct gpu_buddy_block {* @avail: Total free space currently available for allocation in bytes. * @clear_avail: Free space available in the clear tree (zeroed memory) in bytes.
Quoted 18 lines
include/linux/gpu_buddy.h +8
-0* This is a subset of @avail. +* @lock_dep_map: Annotates gpu_buddy API with a driver provided lock. */ struct gpu_buddy {/* private: */ @@ -179,8 +180,48 @@ struct gpu_buddy {u64 size;u64 avail;u64 clear_avail; +#ifdef CONFIG_LOCKDEP +struct lockdep_map *lock_dep_map; +#endif }; +#ifdef CONFIG_LOCKDEP +/** +* gpu_buddy_driver_set_lock() - Set the lock protecting accesses to +GPU BUDDY include/linux/gpu_buddy.h struct gpu_buddy { · line 201
Quoted 10 lines
include/linux/gpu_buddy.h +10
-0+* @mm: Pointer to GPU buddy structure. +* @lock: the lock used to protect the gpu buddy. The locking +primitive +* must contain a dep_map field. +* +* Call this to annotate gpu_buddy APIs which access/modify gpu_buddy +manager */ #define gpu_buddy_driver_set_lock (mm, lock) \ + do { \ + struct gpu_buddy *__mm = (mm); \ + if (!WARN(__mm->lock_dep_map, "GPU BUDDY MM lockshould be set only once.")) \
Quoted 7 lines
include/linux/gpu_buddy.h +7
-0+ __mm->lock_dep_map = &(lock)->dep_map; \ + } while (0) +#else +#define gpu_buddy_driver_set_lock (mm, lock) do { (void)(mm); +(void)(lock); } while (0) #endif + +#ifdef CONFIG_LOCKDEP include/linux/gpu_buddy.h struct gpu_buddy { · line 217
Quoted 10 lines
include/linux/gpu_buddy.h +10
-0+/** +* gpu_buddy_driver_lock_held() - Assert GPU BUDDY manager lock is +held +* @mm: Pointer to the GPU BUDDY structure. +* +* Ensure driver lock is held. +*/ +static inline void gpu_buddy_driver_lock_held (struct gpu_buddy *mm) { +if ((mm)->lock_dep_map) +lockdep_assert (lock_is_held_type ((mm)->lock_dep_map,0 ));} #else
Quoted 6 lines
include/linux/gpu_buddy.h +3
-0+static inline gpu_buddy_driver_lock_held(struct gpu_buddy *mm) { } +#endif +static inline u64 gpu_buddy_block_offset (const struct gpu_buddy_block *block) {
RE: [RFC PATCH V7 02/10] gpu/buddy: Integrate lockdep for gpu buddy manager
Quoted 199 lines
-----Original Message-----
From: Auld, Matthew <[email protected]>
Sent: 16 April 2026 15:27
To: Upadhyay, Tejas <[email protected]>; intel- [email protected]
Cc: Brost, Matthew <[email protected]>; [email protected]; Ghimiray, Himal Prasad <[email protected]>
Subject: Re: [RFC PATCH V7 02/10] gpu/buddy: Integrate lockdep for gpu buddy managerQuoted 18 lines · Tejas
Quoted 18 lines
-----Original Message-----
From: Auld, Matthew <[email protected]>
Sent: 16 April 2026 14:25
To: Upadhyay, Tejas <[email protected]>; intel- [email protected]
Cc: Brost, Matthew <[email protected]>; [email protected]; Ghimiray, Himal Prasad <[email protected]>
Subject: Re: [RFC PATCH V7 02/10] gpu/buddy: Integrate lockdep for gpu buddy managerQuoted 4 lines · Tejas Upadhyay
Integrating lockdep into the
gpu_buddymanager as standard practice for verifying that internal resources are correctly protected by their associated locks.Quoted 6 lines
Quoted 6 lines
Quoted 6 lines
Signed-off-by Tejas Upadhyay <[email protected]>--- drivers/gpu/buddy.c | 18 ++++++++++-- drivers/gpu/drm/drm_buddy.c | 7 +++-- drivers/gpu/drm/xe/xe_ttm_vram_mgr.c | 3 ++ include/linux/gpu_buddy.h | 41++++++++++++++++++++++++++++
Quoted 88 lines
Quoted 78 lines
drivers/gpu/buddy.c int gpu_buddy_init(struct gpu_buddy *mm, u64
Quoted 7 lines
4 files changed, 65 insertions(+), 4 deletions(-)drivers/gpu/buddy.c index renamed from drivers/gpu/buddy.c +0
-0
52686672e99f..53ff85ac2105100644drivers/gpu/buddy.c +0
-0@@ -437,6 +437,9 @@ int gpu_buddy_init(struct gpu_buddy *mm, u64size, u64
chunk_size)Quoted 21 lines
drivers/gpu/buddy.c +4
-0root_count++; }while (size); +#ifdef CONFIG_LOCKDEP + mm->lock_dep_map =NULL ; +#endif return 0 ; out_free_roots:@@ -464,6 +467,7 @@ void gpu_buddy_fini(struct gpu_buddy *mm)unsigned int order;int i; +gpu_buddy_driver_lock_held (mm); size = mm->size; for (i = 0; i < mm->n_roots; ++i) { @@ -538,6 +542,7 @@ void
gpu_buddy_reset_clear(struct gpu_buddy *mm, bool is_clear)unsigned int order; int i;drivers/gpu/buddy.c void gpu_buddy_fini(struct gpu_buddy *mm)
Quoted 4 lines
drivers/gpu/buddy.c +1
-0+gpu_buddy_driver_lock_held (mm); size = mm->size;for (i =0 ; i < mm->n_roots; ++i) { order = ilog2(size) - ilog2(mm->chunk_size); @@ -580,6+585,7 @@
drivers/gpu/buddy.c int gpu_buddy_block_trim(struct gpu_buddy
Quoted 17 lines
EXPORT_SYMBOL(gpu_buddy_reset_clear);
void gpu_buddy_free_block(structgpu_buddy*mm,
structgpu_buddy_block*block)
{drivers/gpu/buddy.c +1
-0+gpu_buddy_driver_lock_held (mm);BUG_ON (!gpu_buddy_block_is_allocated (block)); mm->avail +=gpu_buddy_block_size (mm, block); if (gpu_buddy_block_is_clear(block)) @@ -633,6 +639,7 @@ voidgpu_buddy_free_list(struct
gpu_buddy*mm,{ bool mark_clear = flags & GPU_BUDDY_CLEARED;drivers/gpu/buddy.c +1
-0+gpu_buddy_driver_lock_held (mm);__gpu_buddy_free_list (mm, objects, mark_clear, !mark_clear); }EXPORT_SYMBOL (gpu_buddy_free_list);@@ -1172,6 +1179,8 @@ int gpu_buddy_block_trim(struct gpu_buddy*mm,
drivers/gpu/buddy.c int gpu_buddy_alloc_blocks(struct gpu_buddy
Quoted 9 lines
drivers/gpu/buddy.c +2
-0u64 new_start;int err; +gpu_buddy_driver_lock_held (mm); +if (!list_is_singular (blocks))return -EINVAL ;@@ -1287,6 +1296,8 @@ int gpu_buddy_alloc_blocks(struct gpu_buddy*mm,
Quoted 16 lines
drivers/gpu/buddy.c +3
-2unsigned long pages;int err; +gpu_buddy_driver_lock_held (mm); +if (size < mm->chunk_size)return -EINVAL ;@@ -1458,9 +1469,11 @@ EXPORT_SYMBOL(gpu_buddy_alloc_blocks);void gpu_buddy_block_print (struct gpu_buddy *mm,struct gpu_buddy_block *block) { -u64 start =gpu_buddy_block_offset (block); -u64 size =gpu_buddy_block_size (mm, block); +u64 start, size;Quoted 6 lines
drivers/gpu/buddy.c EXPORT_SYMBOL(gpu_buddy_alloc_blocks); · line 1474
Quoted 1 line
drivers/gpu/buddy.c +1
-0+gpu_buddy_driver_lock_held (mm);I don't think we want this one. The mm interaction is just for immutable state, and the block itself is essentially owned by the caller. Same reason why we don't want annotations for stuff like
gpu_buddy_block_offset()etc.Yes, makes sense. I will change it.
Quoted 1 line
Quoted 6 lines
Quoted 6 lines
drivers/gpu/buddy.c void gpu_buddy_print(struct gpu_buddy
Quoted 6 lines
drivers/gpu/buddy.c +2
-0+ start =gpu_buddy_block_offset (block); + size =gpu_buddy_block_size (mm, block);pr_info ("%#018llx-%#018llx: %llu\n" , start, start + size, size); }EXPORT_SYMBOL (gpu_buddy_block_print);@@ -1475,6 +1488,7 @@ void gpu_buddy_print(struct gpu_buddy*mm)
Quoted 29 lines
Quoted 23 lines
drivers/gpu/buddy.c void gpu_buddy_print(struct gpu_buddy *mm)
Quoted 5 lines
drivers/gpu/buddy.c +1
-0{int order; +gpu_buddy_driver_lock_held (mm); pr_info("chunk_size: %lluKiB, total: %lluMiB, free: %lluMiB, clear_free:%lluMiB\n",
drivers/gpu/drm/drm_buddy.c void drm_buddy_block_print(struct gpu_buddy
Quoted 9 lines
drivers/gpu/buddy.c +0
-0mm->chunk_size >>10 , mm->size >>20 , mm->avail >>20 , mm->clear_avail >> 20);(quoted diff) +0
-0
b/drivers/gpu/drm/drm_buddy.cindex841f3de5f307..f4ad09b8a36e100644drivers/gpu/drm/drm_buddy.c +0
-0@@ -25,9 +25,11 @@ void drm_buddy_block_print(struct gpu_buddy*mm,
Quoted 7 lines
drivers/gpu/drm/drm_buddy.c +1
-2struct gpu_buddy_block *block, struct drm_printer *p) { - u64 start = gpu_buddy_block_offset(block); -u64 size =gpu_buddy_block_size (mm, block); + u64 start, size;Quoted 5 lines
drivers/gpu/drm/drm_buddy.c void drm_buddy_block_print(struct gpu_buddy · line 32
Quoted 3 lines
drivers/gpu/drm/drm_buddy.c +3
-0+gpu_buddy_driver_lock_held (mm); + start =gpu_buddy_block_offset (block); + size =gpu_buddy_block_size (mm, block);Same here.
Quoted 7 lines
Yes
Quoted 5 lines
drivers/gpu/drm/drm_buddy.c void drm_buddy_print(struct gpu_buddy *mm,
Quoted 5 lines
drivers/gpu/drm/drm_buddy.c +0
-0drm_printf (p,"%#018llx-%#018llx: %llu\n" , start, start + size, size); }EXPORT_SYMBOL (drm_buddy_block_print);@@ -42,6 +44,7 @@ void drm_buddy_print(struct gpu_buddy *mm,struct
Quoted 28 lines
Quoted 27 lines
drm_printer*p)drivers/gpu/drm/drm_buddy.c void drm_buddy_print(struct gpu_buddy *mm, · line 49
Quoted 6 lines
drivers/gpu/drm/drm_buddy.c +1
-0{int order; +gpu_buddy_driver_lock_held (mm); drm_printf(p, "chunk_size: %lluKiB, total: %lluMiB, free: %lluMiB,
clear_free: %lluMiB\n",drivers/gpu/drm/xe/xe_ttm_vram_mgr.c static void xe_ttm_vram_mgr_fini(struct
Quoted 9 lines
drivers/gpu/drm/drm_buddy.c +0
-0mm->chunk_size >> 10, mm->size >> 20, mm->avail >> 20,
mm->clear_avail>> 20);drivers/gpu/drm/xe/xe_ttm_vram_mgr.c +0
-0@@ -293,7 +293,9 @@ static void xe_ttm_vram_mgr_fini(struct
drm_devicedrivers/gpu/drm/xe/xe_ttm_vram_mgr.c static void xe_ttm_vram_mgr_fini(struct drm_device *dev, void *arg) · line 298
Quoted 7 lines
drivers/gpu/drm/xe/xe_ttm_vram_mgr.c +2
-0*dev, void *arg)WARN_ON_ONCE (mgr->visible_avail != mgr->visible_size); +mutex_lock (&mgr->lock);gpu_buddy_fini (&mgr->mm); +mutex_unlock (&mgr->lock);This shouldn't need a lock. Annotation for this one should also be dropped.
Quoted 7 lines
Quoted 1 line
Thought was that while we call
buddy_fini, there could be in flight access ofbuddy manager, so some new allocation in progress and thus it will provide protection against it. Please let me know what you think!
The mm stuff is about to get nuked, so if something is still messing with this then we are in big trouble anyway, and grabbing the lock does't really help much. The
fini()here should be triggered as we start tearing down the driver.
Right, the driver is going away at this stage, but thinking was that at least it will allow in flight things who hold lock to finish before fini proceeds. But I am not too sure though about there will be such situation hit in practical or not. I am ok to remove this.
Tejas
Quoted 79 lines
Quoted 26 lines
Tejas
Quoted 23 lines
drivers/gpu/drm/xe/xe_ttm_vram_mgr.c int __xe_ttm_vram_mgr_init(struct xe_device *xe,
Quoted 5 lines
drivers/gpu/drm/xe/xe_ttm_vram_mgr.c +0
-0ttm_resource_manager_cleanup (&mgr->manager);@@ -328,6 +330,7 @@ int __xe_ttm_vram_mgr_init(struct xe_device *xe,struct
xe_ttm_vram_mgr*mgr,Quoted 12 lines
drivers/gpu/drm/xe/xe_ttm_vram_mgr.c +1
-0if (err)return err; +gpu_buddy_driver_set_lock (&mgr->mm, &mgr->lock);ttm_set_driver_manager (&xe->ttm, mem_type, &mgr->manager);ttm_resource_manager_set_used (&mgr->manager, true);include/linux/gpu_buddy.h +0
-0include/linux/gpu_buddy.h struct gpu_buddy_block { · line 155
Quoted 2 lines
include/linux/gpu_buddy.h +0
-0@@ -154,6 +154,7 @@ struct gpu_buddy_block {* @avail: Total free space currently available for allocation in bytes. * @clear_avail: Free space available in the clear tree (zeroed memory) in bytes.
Quoted 2 lines
include/linux/gpu_buddy.h +1
-0* This is a subset of @avail. +* @lock_dep_map: Annotates gpu_buddy API with a driver provided lock.
Quoted 52 lines
Quoted 51 lines
Quoted 16 lines
include/linux/gpu_buddy.h +7
-0*/ struct gpu_buddy {/* private: */ @@ -179,8 +180,48 @@ struct gpu_buddy {u64 size;u64 avail;u64 clear_avail; +#ifdef CONFIG_LOCKDEP +struct lockdep_map *lock_dep_map; +#endif }; +#ifdef CONFIG_LOCKDEP +/** +* gpu_buddy_driver_set_lock() - Set the lock protecting accesses to +GPU BUDDY include/linux/gpu_buddy.h struct gpu_buddy { · line 201
Quoted 10 lines
include/linux/gpu_buddy.h +10
-0+* @mm: Pointer to GPU buddy structure. +* @lock: the lock used to protect the gpu buddy. The locking +primitive +* must contain a dep_map field. +* +* Call this to annotate gpu_buddy APIs which access/modify gpu_buddy +manager */ #define gpu_buddy_driver_set_lock (mm, lock) \ + do { \ + struct gpu_buddy *__mm = (mm); \ + if (!WARN(__mm->lock_dep_map, "GPU BUDDY MM lockshould be set only once.")) \
Quoted 7 lines
include/linux/gpu_buddy.h +7
-0+ __mm->lock_dep_map = &(lock)->dep_map; \ + } while (0) +#else +#define gpu_buddy_driver_set_lock (mm, lock) do { (void)(mm); +(void)(lock); } while (0) #endif + +#ifdef CONFIG_LOCKDEP include/linux/gpu_buddy.h struct gpu_buddy { · line 217
Quoted 10 lines
include/linux/gpu_buddy.h +10
-0+/** +* gpu_buddy_driver_lock_held() - Assert GPU BUDDY manager lock is +held +* @mm: Pointer to the GPU BUDDY structure. +* +* Ensure driver lock is held. +*/ +static inline void gpu_buddy_driver_lock_held (struct gpu_buddy *mm) { +if ((mm)->lock_dep_map) +lockdep_assert (lock_is_held_type ((mm)->lock_dep_map,0 ));} #else
Quoted 6 lines
include/linux/gpu_buddy.h +3
-0+static inline gpu_buddy_driver_lock_held(struct gpu_buddy *mm) { } +#endif +static inline u64 gpu_buddy_block_offset (const struct gpu_buddy_block *block) {
Matthew Auld
maintainerRe: [RFC PATCH V7 02/10] gpu/buddy: Integrate lockdep for gpu buddy manager
Quoted 200 lines · Tejas
Quoted 200 lines
-----Original Message-----
From: Auld, Matthew <[email protected]>
Sent: 16 April 2026 15:27
To: Upadhyay, Tejas <[email protected]>; intel- [email protected]
Cc: Brost, Matthew <[email protected]>; [email protected]; Ghimiray, Himal Prasad <[email protected]>
Subject: Re: [RFC PATCH V7 02/10] gpu/buddy: Integrate lockdep for gpu buddy managerQuoted 18 lines · Tejas
Quoted 18 lines
-----Original Message-----
From: Auld, Matthew <[email protected]>
Sent: 16 April 2026 14:25
To: Upadhyay, Tejas <[email protected]>; intel- [email protected]
Cc: Brost, Matthew <[email protected]>; [email protected]; Ghimiray, Himal Prasad <[email protected]>
Subject: Re: [RFC PATCH V7 02/10] gpu/buddy: Integrate lockdep for gpu buddy managerQuoted 4 lines · Tejas Upadhyay
Integrating lockdep into the
gpu_buddymanager as standard practice for verifying that internal resources are correctly protected by their associated locks.Quoted 6 lines
Quoted 6 lines
Quoted 6 lines
Signed-off-by Tejas Upadhyay <[email protected]>--- drivers/gpu/buddy.c | 18 ++++++++++-- drivers/gpu/drm/drm_buddy.c | 7 +++-- drivers/gpu/drm/xe/xe_ttm_vram_mgr.c | 3 ++ include/linux/gpu_buddy.h | 41++++++++++++++++++++++++++++
Quoted 88 lines
Quoted 78 lines
drivers/gpu/buddy.c int gpu_buddy_init(struct gpu_buddy *mm, u64
Quoted 7 lines
4 files changed, 65 insertions(+), 4 deletions(-)drivers/gpu/buddy.c index renamed from drivers/gpu/buddy.c +0
-0
52686672e99f..53ff85ac2105100644drivers/gpu/buddy.c +0
-0@@ -437,6 +437,9 @@ int gpu_buddy_init(struct gpu_buddy *mm, u64size, u64
chunk_size)Quoted 21 lines
drivers/gpu/buddy.c +4
-0root_count++; }while (size); +#ifdef CONFIG_LOCKDEP + mm->lock_dep_map =NULL ; +#endif return 0 ; out_free_roots:@@ -464,6 +467,7 @@ void gpu_buddy_fini(struct gpu_buddy *mm)unsigned int order;int i; +gpu_buddy_driver_lock_held (mm); size = mm->size; for (i = 0; i < mm->n_roots; ++i) { @@ -538,6 +542,7 @@ void
gpu_buddy_reset_clear(struct gpu_buddy *mm, bool is_clear)unsigned int order; int i;drivers/gpu/buddy.c void gpu_buddy_fini(struct gpu_buddy *mm)
Quoted 4 lines
drivers/gpu/buddy.c +1
-0+gpu_buddy_driver_lock_held (mm); size = mm->size;for (i =0 ; i < mm->n_roots; ++i) { order = ilog2(size) - ilog2(mm->chunk_size); @@ -580,6+585,7 @@
drivers/gpu/buddy.c int gpu_buddy_block_trim(struct gpu_buddy
Quoted 17 lines
EXPORT_SYMBOL(gpu_buddy_reset_clear);
void gpu_buddy_free_block(structgpu_buddy*mm,
structgpu_buddy_block*block)
{drivers/gpu/buddy.c +1
-0+gpu_buddy_driver_lock_held (mm);BUG_ON (!gpu_buddy_block_is_allocated (block)); mm->avail +=gpu_buddy_block_size (mm, block); if (gpu_buddy_block_is_clear(block)) @@ -633,6 +639,7 @@ voidgpu_buddy_free_list(struct
gpu_buddy*mm,{ bool mark_clear = flags & GPU_BUDDY_CLEARED;drivers/gpu/buddy.c +1
-0+gpu_buddy_driver_lock_held (mm);__gpu_buddy_free_list (mm, objects, mark_clear, !mark_clear); }EXPORT_SYMBOL (gpu_buddy_free_list);@@ -1172,6 +1179,8 @@ int gpu_buddy_block_trim(struct gpu_buddy*mm,
drivers/gpu/buddy.c int gpu_buddy_alloc_blocks(struct gpu_buddy
Quoted 9 lines
drivers/gpu/buddy.c +2
-0u64 new_start;int err; +gpu_buddy_driver_lock_held (mm); +if (!list_is_singular (blocks))return -EINVAL ;@@ -1287,6 +1296,8 @@ int gpu_buddy_alloc_blocks(struct gpu_buddy*mm,
Quoted 16 lines
drivers/gpu/buddy.c +3
-2unsigned long pages;int err; +gpu_buddy_driver_lock_held (mm); +if (size < mm->chunk_size)return -EINVAL ;@@ -1458,9 +1469,11 @@ EXPORT_SYMBOL(gpu_buddy_alloc_blocks);void gpu_buddy_block_print (struct gpu_buddy *mm,struct gpu_buddy_block *block) { -u64 start =gpu_buddy_block_offset (block); -u64 size =gpu_buddy_block_size (mm, block); +u64 start, size;Quoted 6 lines
drivers/gpu/buddy.c EXPORT_SYMBOL(gpu_buddy_alloc_blocks); · line 1474
Quoted 1 line
drivers/gpu/buddy.c +1
-0+gpu_buddy_driver_lock_held (mm);I don't think we want this one. The mm interaction is just for immutable state, and the block itself is essentially owned by the caller. Same reason why we don't want annotations for stuff like
gpu_buddy_block_offset()etc.Yes, makes sense. I will change it.
Quoted 1 line
Quoted 6 lines
Quoted 6 lines
drivers/gpu/buddy.c void gpu_buddy_print(struct gpu_buddy
Quoted 6 lines
drivers/gpu/buddy.c +2
-0+ start =gpu_buddy_block_offset (block); + size =gpu_buddy_block_size (mm, block);pr_info ("%#018llx-%#018llx: %llu\n" , start, start + size, size); }EXPORT_SYMBOL (gpu_buddy_block_print);@@ -1475,6 +1488,7 @@ void gpu_buddy_print(struct gpu_buddy*mm)
Quoted 29 lines
Quoted 23 lines
drivers/gpu/buddy.c void gpu_buddy_print(struct gpu_buddy *mm)
Quoted 5 lines
drivers/gpu/buddy.c +1
-0{int order; +gpu_buddy_driver_lock_held (mm); pr_info("chunk_size: %lluKiB, total: %lluMiB, free: %lluMiB, clear_free:%lluMiB\n",
drivers/gpu/drm/drm_buddy.c void drm_buddy_block_print(struct gpu_buddy
Quoted 9 lines
drivers/gpu/buddy.c +0
-0mm->chunk_size >>10 , mm->size >>20 , mm->avail >>20 , mm->clear_avail >> 20);(quoted diff) +0
-0
b/drivers/gpu/drm/drm_buddy.cindex841f3de5f307..f4ad09b8a36e100644drivers/gpu/drm/drm_buddy.c +0
-0@@ -25,9 +25,11 @@ void drm_buddy_block_print(struct gpu_buddy*mm,
Quoted 7 lines
drivers/gpu/drm/drm_buddy.c +1
-2struct gpu_buddy_block *block, struct drm_printer *p) { - u64 start = gpu_buddy_block_offset(block); -u64 size =gpu_buddy_block_size (mm, block); + u64 start, size;Quoted 5 lines
drivers/gpu/drm/drm_buddy.c void drm_buddy_block_print(struct gpu_buddy · line 32
Quoted 3 lines
drivers/gpu/drm/drm_buddy.c +3
-0+gpu_buddy_driver_lock_held (mm); + start =gpu_buddy_block_offset (block); + size =gpu_buddy_block_size (mm, block);Same here.
Quoted 7 lines
Yes
Quoted 5 lines
drivers/gpu/drm/drm_buddy.c void drm_buddy_print(struct gpu_buddy *mm,
Quoted 5 lines
drivers/gpu/drm/drm_buddy.c +0
-0drm_printf (p,"%#018llx-%#018llx: %llu\n" , start, start + size, size); }EXPORT_SYMBOL (drm_buddy_block_print);@@ -42,6 +44,7 @@ void drm_buddy_print(struct gpu_buddy *mm,struct
Quoted 29 lines
Quoted 20 lines
drm_printer*p)drivers/gpu/drm/drm_buddy.c void drm_buddy_print(struct gpu_buddy *mm, · line 49
Quoted 6 lines
drivers/gpu/drm/drm_buddy.c +1
-0{int order; +gpu_buddy_driver_lock_held (mm); drm_printf(p, "chunk_size: %lluKiB, total: %lluMiB, free: %lluMiB,
clear_free: %lluMiB\n",drivers/gpu/drm/xe/xe_ttm_vram_mgr.c static void xe_ttm_vram_mgr_fini(struct
Quoted 9 lines
drivers/gpu/drm/drm_buddy.c +0
-0mm->chunk_size >> 10, mm->size >> 20, mm->avail >> 20,
mm->clear_avail>> 20);drivers/gpu/drm/xe/xe_ttm_vram_mgr.c +0
-0@@ -293,7 +293,9 @@ static void xe_ttm_vram_mgr_fini(struct
drm_deviceQuoted 2 lines
drivers/gpu/drm/xe/xe_ttm_vram_mgr.c +0
-0*dev, void *arg)Quoted 7 lines
drivers/gpu/drm/xe/xe_ttm_vram_mgr.c static void xe_ttm_vram_mgr_fini(struct drm_device *dev, void *arg) · line 298
Quoted 5 lines
drivers/gpu/drm/xe/xe_ttm_vram_mgr.c +2
-0WARN_ON_ONCE (mgr->visible_avail != mgr->visible_size); +mutex_lock (&mgr->lock);gpu_buddy_fini (&mgr->mm); +mutex_unlock (&mgr->lock);This shouldn't need a lock. Annotation for this one should also be dropped.
Thought was that while we call
buddy_fini, there could be in flight access of
Quoted 8 lines
Quoted 6 lines
buddy manager, so some new allocation in progress and thus it will provide protection against it. Please let me know what you think!
The mm stuff is about to get nuked, so if something is still messing with this then we are in big trouble anyway, and grabbing the lock does't really help much. The
fini()here should be triggered as we start tearing down the driver.Right, the driver is going away at this stage, but thinking was that at least it will allow in flight things who hold lock to finish before fini proceeds. But I am not too sure though about there will be such situation hit in practical or not. I am ok to remove this.
Right, but grabbing the lock won't help much here, I think. As soon you drop it here, the other place probably still goes boom anyway, since the mm state is all gone.
If this is a concern, I think it would need to solved in a different way. We do have various WARN_ON() if we detect still in-use blocks, but essentially the bug is not here, and likely someone just forgot some put() somewhere.
Quoted 82 lines
Tejas
Quoted 79 lines
Quoted 26 lines
Tejas
Quoted 23 lines
drivers/gpu/drm/xe/xe_ttm_vram_mgr.c int __xe_ttm_vram_mgr_init(struct xe_device *xe,
Quoted 5 lines
drivers/gpu/drm/xe/xe_ttm_vram_mgr.c +0
-0ttm_resource_manager_cleanup (&mgr->manager);@@ -328,6 +330,7 @@ int __xe_ttm_vram_mgr_init(struct xe_device *xe,struct
xe_ttm_vram_mgr*mgr,Quoted 12 lines
drivers/gpu/drm/xe/xe_ttm_vram_mgr.c +1
-0if (err)return err; +gpu_buddy_driver_set_lock (&mgr->mm, &mgr->lock);ttm_set_driver_manager (&xe->ttm, mem_type, &mgr->manager);ttm_resource_manager_set_used (&mgr->manager, true);include/linux/gpu_buddy.h +0
-0include/linux/gpu_buddy.h struct gpu_buddy_block { · line 155
Quoted 2 lines
include/linux/gpu_buddy.h +0
-0@@ -154,6 +154,7 @@ struct gpu_buddy_block {* @avail: Total free space currently available for allocation in bytes. * @clear_avail: Free space available in the clear tree (zeroed memory) in bytes.
Quoted 2 lines
include/linux/gpu_buddy.h +1
-0* This is a subset of @avail. +* @lock_dep_map: Annotates gpu_buddy API with a driver provided lock.
Quoted 52 lines
Quoted 51 lines
Quoted 16 lines
include/linux/gpu_buddy.h +7
-0*/ struct gpu_buddy {/* private: */ @@ -179,8 +180,48 @@ struct gpu_buddy {u64 size;u64 avail;u64 clear_avail; +#ifdef CONFIG_LOCKDEP +struct lockdep_map *lock_dep_map; +#endif }; +#ifdef CONFIG_LOCKDEP +/** +* gpu_buddy_driver_set_lock() - Set the lock protecting accesses to +GPU BUDDY include/linux/gpu_buddy.h struct gpu_buddy { · line 201
Quoted 10 lines
include/linux/gpu_buddy.h +10
-0+* @mm: Pointer to GPU buddy structure. +* @lock: the lock used to protect the gpu buddy. The locking +primitive +* must contain a dep_map field. +* +* Call this to annotate gpu_buddy APIs which access/modify gpu_buddy +manager */ #define gpu_buddy_driver_set_lock (mm, lock) \ + do { \ + struct gpu_buddy *__mm = (mm); \ + if (!WARN(__mm->lock_dep_map, "GPU BUDDY MM lockshould be set only once.")) \
Quoted 7 lines
include/linux/gpu_buddy.h +7
-0+ __mm->lock_dep_map = &(lock)->dep_map; \ + } while (0) +#else +#define gpu_buddy_driver_set_lock (mm, lock) do { (void)(mm); +(void)(lock); } while (0) #endif + +#ifdef CONFIG_LOCKDEP include/linux/gpu_buddy.h struct gpu_buddy { · line 217
Quoted 10 lines
include/linux/gpu_buddy.h +10
-0+/** +* gpu_buddy_driver_lock_held() - Assert GPU BUDDY manager lock is +held +* @mm: Pointer to the GPU BUDDY structure. +* +* Ensure driver lock is held. +*/ +static inline void gpu_buddy_driver_lock_held (struct gpu_buddy *mm) { +if ((mm)->lock_dep_map) +lockdep_assert (lock_is_held_type ((mm)->lock_dep_map,0 ));} #else
Quoted 6 lines
include/linux/gpu_buddy.h +3
-0+static inline gpu_buddy_driver_lock_held(struct gpu_buddy *mm) { } +#endif +static inline u64 gpu_buddy_block_offset (const struct gpu_buddy_block *block) {
RE: [RFC PATCH V7 02/10] gpu/buddy: Integrate lockdep for gpu buddy manager
Quoted 240 lines
-----Original Message-----
From: Auld, Matthew <[email protected]>
Sent: 16 April 2026 15:46
To: Upadhyay, Tejas <[email protected]>; intel- [email protected]
Cc: Brost, Matthew <[email protected]>; [email protected]; Ghimiray, Himal Prasad <[email protected]>
Subject: Re: [RFC PATCH V7 02/10] gpu/buddy: Integrate lockdep for gpu buddy managerQuoted 38 lines · Tejas
Quoted 38 lines
-----Original Message-----
From: Auld, Matthew <[email protected]>
Sent: 16 April 2026 15:27
To: Upadhyay, Tejas <[email protected]>; intel- [email protected]
Cc: Brost, Matthew <[email protected]>; [email protected]; Ghimiray, Himal Prasad <[email protected]>
Subject: Re: [RFC PATCH V7 02/10] gpu/buddy: Integrate lockdep for gpu buddy managerQuoted 24 lines · Tejas
Quoted 24 lines
-----Original Message-----
From: Auld, Matthew <[email protected]>
Sent: 16 April 2026 14:25
To: Upadhyay, Tejas <[email protected]>; intel- [email protected]
Cc: Brost, Matthew <[email protected]>; [email protected]; Ghimiray, Himal Prasad <[email protected]>
Subject: Re: [RFC PATCH V7 02/10] gpu/buddy: Integrate lockdep for gpu buddy managerQuoted 10 lines · Tejas Upadhyay
Integrating lockdep into the
gpu_buddymanager as standard practice for verifying that internal resources are correctly protected by their associated locks.Signed-off-by Tejas Upadhyay <[email protected]>--- drivers/gpu/buddy.c | 18 ++++++++++-- drivers/gpu/drm/drm_buddy.c | 7 +++-- drivers/gpu/drm/xe/xe_ttm_vram_mgr.c | 3 ++ include/linux/gpu_buddy.h | 41Quoted 8 lines
Quoted 8 lines
++++++++++++++++++++++++++++
Quoted 7 lines
Quoted 7 lines
drivers/gpu/buddy.c int gpu_buddy_init(struct gpu_buddy *mm,
Quoted 7 lines
4 files changed, 65 insertions(+), 4 deletions(-)drivers/gpu/buddy.c index renamed from drivers/gpu/buddy.c +0
-0
52686672e99f..53ff85ac2105100644drivers/gpu/buddy.c +0
-0@@ -437,6 +437,9 @@ int gpu_buddy_init(struct gpu_buddy *mm,u64
Quoted 22 lines
Quoted 22 lines
Quoted 22 lines
Quoted 22 lines
size, u64
chunk_size)Quoted 21 lines
drivers/gpu/buddy.c +4
-0root_count++; }while (size); +#ifdef CONFIG_LOCKDEP + mm->lock_dep_map =NULL ; +#endif return 0 ; out_free_roots:@@ -464,6 +467,7 @@ void gpu_buddy_fini(struct gpu_buddy *mm)unsigned int order;int i; +gpu_buddy_driver_lock_held (mm); size = mm->size; for (i = 0; i < mm->n_roots; ++i) { @@ -538,6 +542,7 @@ void
gpu_buddy_reset_clear(struct gpu_buddy *mm, bool is_clear)unsigned int order; int i;Quoted 4 lines
Quoted 4 lines
Quoted 4 lines
Quoted 4 lines
drivers/gpu/buddy.c void gpu_buddy_fini(struct gpu_buddy *mm)
Quoted 4 lines
drivers/gpu/buddy.c +1
-0+gpu_buddy_driver_lock_held (mm); size = mm->size;for (i =0 ; i < mm->n_roots; ++i) { order = ilog2(size) - ilog2(mm->chunk_size); @@ -580,6
Quoted 9 lines
Quoted 9 lines
Quoted 9 lines
Quoted 9 lines
+585,7 @@
drivers/gpu/buddy.c EXPORT_SYMBOL(gpu_buddy_reset_clear);
Quoted 8 lines
EXPORT_SYMBOL(gpu_buddy_reset_clear);
void gpu_buddy_free_block(structgpu_buddy*mm,
structgpu_buddy_block*block)
{drivers/gpu/buddy.c +1
-0+gpu_buddy_driver_lock_held (mm);BUG_ON (!gpu_buddy_block_is_allocated (block)); mm->avail +=gpu_buddy_block_size (mm, block); if (gpu_buddy_block_is_clear(block)) @@ -633,6 +639,7 @@void
Quoted 6 lines
Quoted 6 lines
Quoted 6 lines
Quoted 6 lines
drivers/gpu/buddy.c
Quoted 6 lines
gpu_buddy_free_list(struct
gpu_buddy*mm,{ bool mark_clear = flags & GPU_BUDDY_CLEARED;drivers/gpu/buddy.c +1
-0+gpu_buddy_driver_lock_held (mm); __gpu_buddy_free_list(mm, objects, mark_clear,!mark_clear);
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
drivers/gpu/buddy.c int gpu_buddy_block_trim(struct
Quoted 3 lines
} EXPORT_SYMBOL(gpu_buddy_free_list);drivers/gpu/buddy.c +0
-0@@ -1172,6 +1179,8 @@ int gpu_buddy_block_trim(struct
gpu_buddyQuoted 10 lines
Quoted 10 lines
Quoted 10 lines
Quoted 10 lines
*mm,
drivers/gpu/buddy.c int gpu_buddy_alloc_blocks(struct
Quoted 9 lines
drivers/gpu/buddy.c +2
-0u64 new_start;int err; +gpu_buddy_driver_lock_held (mm); +if (!list_is_singular (blocks))return -EINVAL ;@@ -1287,6 +1296,8 @@ int gpu_buddy_alloc_blocks(struct
gpu_buddyQuoted 10 lines
Quoted 10 lines
Quoted 10 lines
Quoted 10 lines
*mm,
drivers/gpu/buddy.c
Quoted 9 lines
drivers/gpu/buddy.c +2
-0unsigned long pages;int err; +gpu_buddy_driver_lock_held (mm); +if (size < mm->chunk_size)return -EINVAL ;@@ -1458,9 +1469,11 @@
EXPORT_SYMBOL(gpu_buddy_alloc_blocks);Quoted 24 lines
Quoted 24 lines
Quoted 17 lines
Quoted 7 lines
Quoted 7 lines
drivers/gpu/buddy.c +1
-2void gpu_buddy_block_print (struct gpu_buddy *mm,struct gpu_buddy_block *block) { -u64 start =gpu_buddy_block_offset (block); -u64 size =gpu_buddy_block_size (mm, block); +u64 start, size;Quoted 6 lines
drivers/gpu/buddy.c line 1474
Quoted 1 line
drivers/gpu/buddy.c +1
-0+gpu_buddy_driver_lock_held (mm);I don't think we want this one. The mm interaction is just for immutable state, and the block itself is essentially owned by the caller. Same reason why we don't want annotations for stuff like
gpu_buddy_block_offset()etc.Yes, makes sense. I will change it.
Quoted 1 line
Quoted 6 lines
Quoted 6 lines
drivers/gpu/buddy.c void gpu_buddy_print(struct gpu_buddy
Quoted 6 lines
drivers/gpu/buddy.c +2
-0+ start =gpu_buddy_block_offset (block); + size =gpu_buddy_block_size (mm, block);pr_info ("%#018llx-%#018llx: %llu\n" , start, start + size, size); }EXPORT_SYMBOL (gpu_buddy_block_print);@@ -1475,6 +1488,7 @@ void gpu_buddy_print(struct gpu_buddy*mm)
Quoted 5 lines
Quoted 5 lines
Quoted 5 lines
Quoted 5 lines
drivers/gpu/buddy.c void gpu_buddy_print(struct gpu_buddy *mm)
Quoted 5 lines
drivers/gpu/buddy.c +1
-0{int order; +gpu_buddy_driver_lock_held (mm); pr_info("chunk_size: %lluKiB, total: %lluMiB, free: %lluMiB,
clear_free:Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
%lluMiB\n",
Quoted 1 line
drivers/gpu/buddy.c +0
-0mm->chunk_size >> 10, mm->size >> 20, mm->avail >>20,
Quoted 22 lines
Quoted 22 lines
Quoted 22 lines
Quoted 21 lines
drivers/gpu/drm/drm_buddy.c void drm_buddy_block_print(struct gpu_buddy
Quoted 8 lines
drivers/gpu/buddy.c +0
-0mm->clear_avail >> 20);(quoted diff) +0
-0
b/drivers/gpu/drm/drm_buddy.cindex841f3de5f307..f4ad09b8a36e100644drivers/gpu/drm/drm_buddy.c +0
-0@@ -25,9 +25,11 @@ void drm_buddy_block_print(struct gpu_buddy*mm,
drivers/gpu/drm/drm_buddy.c void drm_buddy_block_print(struct gpu_buddy *mm, · line 32
Quoted 10 lines
drivers/gpu/drm/drm_buddy.c +4
-2struct gpu_buddy_block *block, struct drm_printer *p) { - u64 start = gpu_buddy_block_offset(block); -u64 size =gpu_buddy_block_size (mm, block); + u64 start, size; +gpu_buddy_driver_lock_held (mm); + start =gpu_buddy_block_offset (block); + size =gpu_buddy_block_size (mm, block);Same here.
Quoted 4 lines
Quoted 4 lines
Quoted 4 lines
Yes
Quoted 2 lines
Quoted 2 lines
drivers/gpu/drm/drm_buddy.c +0
-0drm_printf(p, "%#018llx-%#018llx: %llu\n", start, start + size,size);
Quoted 25 lines
Quoted 25 lines
Quoted 3 lines
Quoted 3 lines
drivers/gpu/drm/drm_buddy.c void drm_buddy_print(struct gpu_buddy *mm,
Quoted 3 lines
drivers/gpu/drm/drm_buddy.c +0
-0}EXPORT_SYMBOL (drm_buddy_block_print);@@ -42,6 +44,7 @@ void drm_buddy_print(struct gpu_buddy *mm,struct
Quoted 9 lines
Quoted 9 lines
drm_printer*p)drivers/gpu/drm/drm_buddy.c void drm_buddy_print(struct gpu_buddy *mm, · line 49
Quoted 6 lines
drivers/gpu/drm/drm_buddy.c +1
-0{int order; +gpu_buddy_driver_lock_held (mm); drm_printf(p, "chunk_size: %lluKiB, total: %lluMiB, free: %lluMiB,
clear_free: %lluMiB\n",Quoted 1 line
drivers/gpu/drm/drm_buddy.c +0
-0mm->chunk_size >>10 , mm->size >>20 , mm->avail20,
Quoted 11 lines
Quoted 11 lines
drivers/gpu/drm/xe/xe_ttm_vram_mgr.c static void xe_ttm_vram_mgr_fini(struct
Quoted 8 lines
mm->clear_avail>> 20);drivers/gpu/drm/xe/xe_ttm_vram_mgr.c +0
-0@@ -293,7 +293,9 @@ static void xe_ttm_vram_mgr_fini(struct
drm_deviceQuoted 2 lines
drivers/gpu/drm/xe/xe_ttm_vram_mgr.c +0
-0*dev, void *arg)Quoted 7 lines
Quoted 7 lines
Quoted 7 lines
Quoted 7 lines
drivers/gpu/drm/xe/xe_ttm_vram_mgr.c static void xe_ttm_vram_mgr_fini(struct drm_device *dev, void *arg) · line 298
Quoted 5 lines
drivers/gpu/drm/xe/xe_ttm_vram_mgr.c +2
-0WARN_ON_ONCE (mgr->visible_avail != mgr->visible_size); +mutex_lock (&mgr->lock);gpu_buddy_fini (&mgr->mm); +mutex_unlock (&mgr->lock);This shouldn't need a lock. Annotation for this one should also be
dropped.
Quoted 9 lines
Quoted 9 lines
Quoted 3 lines
Thought was that while we call
buddy_fini, there could be in flight access ofbuddy manager, so some new allocation in progress and thus it will provide protection against it. Please let me know what you think!
The mm stuff is about to get nuked, so if something is still messing with this then we are in big trouble anyway, and grabbing the lock does't really help much. The
fini()here should be triggered as we starttearing down the driver.
Quoted 2 lines
Right, the driver is going away at this stage, but thinking was that at least it
will allow in flight things who hold lock to finish before fini proceeds. But I am not too sure though about there will be such situation hit in practical or not. I am ok to remove this.
Quoted 8 lines
Right, but grabbing the lock won't help much here, I think. As soon you drop it here, the other place probably still goes boom anyway, since the mm state is all gone.
If this is a concern, I think it would need to solved in a different way. We do have various
WARN_ON()if we detect still in-use blocks, but essentially the bug is not here, and likely someone just forgot someput()somewhere.
Ok, I will remove this locking and its annotation.
Tejas
Quoted 88 lines
Quoted 11 lines
Tejas
Quoted 8 lines
Quoted 8 lines
Tejas
Quoted 5 lines
drivers/gpu/drm/xe/xe_ttm_vram_mgr.c int __xe_ttm_vram_mgr_init(struct xe_device
Quoted 5 lines
drivers/gpu/drm/xe/xe_ttm_vram_mgr.c +0
-0ttm_resource_manager_cleanup (&mgr->manager);@@ -328,6 +330,7 @@ int __xe_ttm_vram_mgr_init(struct xe_device*xe,
Quoted 9 lines
Quoted 6 lines
Quoted 6 lines
Quoted 6 lines
struct
xe_ttm_vram_mgr*mgr,drivers/gpu/drm/xe/xe_ttm_vram_mgr.c int __xe_ttm_vram_mgr_init(struct xe_device *xe, struct xe_ttm_vram_mgr *mgr,
Quoted 5 lines
drivers/gpu/drm/xe/xe_ttm_vram_mgr.c +1
-0if (err)return err; +gpu_buddy_driver_set_lock (&mgr->mm, &mgr->lock); ttm_set_driver_manager(&xe->ttm, mem_type, &mgr-manager);
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
drivers/gpu/drm/xe/xe_ttm_vram_mgr.c +0
-0ttm_resource_manager_set_used (&mgr->manager, true);Quoted 7 lines
Quoted 7 lines
Quoted 7 lines
Quoted 7 lines
include/linux/gpu_buddy.h struct gpu_buddy_block { · line 155
Quoted 7 lines
include/linux/gpu_buddy.h +0
-0@@ -154,6 +154,7 @@ struct gpu_buddy_block {* @avail: Total free space currently available for allocation in bytes. * @clear_avail: Free space available in the clear tree (zeroed memory) in
Quoted 16 lines
Quoted 16 lines
Quoted 3 lines
Quoted 3 lines
bytes.
Quoted 2 lines
include/linux/gpu_buddy.h +1
-0* This is a subset of @avail. +* @lock_dep_map: Annotates gpu_buddy API with a driver provided lock.
Quoted 12 lines
Quoted 12 lines
Quoted 12 lines
include/linux/gpu_buddy.h +3
-0*/ struct gpu_buddy {/* private: */ @@ -179,8 +180,48 @@ struct gpu_buddy {u64 size;u64 avail;u64 clear_avail; +#ifdef CONFIG_LOCKDEP +struct lockdep_map *lock_dep_map; +#endif };Quoted 10 lines
Quoted 10 lines
Quoted 10 lines
Quoted 10 lines
include/linux/gpu_buddy.h struct gpu_buddy {
Quoted 10 lines
include/linux/gpu_buddy.h +10
-0+#ifdef CONFIG_LOCKDEP +/** +* gpu_buddy_driver_set_lock() - Set the lock protecting accesses to +GPU BUDDY +* @mm: Pointer to GPU buddy structure. +* @lock: the lock used to protect the gpu buddy. The locking +primitive +* must contain a dep_map field. +* +* Call this to annotate gpu_buddy APIs which access/modify
gpu_buddyQuoted 5 lines
Quoted 5 lines
Quoted 5 lines
Quoted 5 lines
include/linux/gpu_buddy.h struct gpu_buddy {
Quoted 4 lines
include/linux/gpu_buddy.h +4
-0+manager */ #define gpu_buddy_driver_set_lock (mm, lock) \ + do { \ + struct gpu_buddy *__mm = (mm); \ + if (!WARN(__mm->lock_dep_map, "GPU BUDDY MM lockshould be set only once.")) \
Quoted 15 lines
Quoted 7 lines
Quoted 7 lines
Quoted 7 lines
Quoted 7 lines
include/linux/gpu_buddy.h +7
-0+ __mm->lock_dep_map = &(lock)->dep_map; \ + } while (0) +#else +#define gpu_buddy_driver_set_lock (mm, lock) do { (void)(mm); +(void)(lock); } while (0) #endif + +#ifdef CONFIG_LOCKDEP Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
include/linux/gpu_buddy.h struct gpu_buddy { · line 214
Quoted 8 lines
include/linux/gpu_buddy.h +8
-0+/** +* gpu_buddy_driver_lock_held() - Assert GPU BUDDY manager lock is +held +* @mm: Pointer to the GPU BUDDY structure. +* +* Ensure driver lock is held. +*/ +static inline void gpu_buddy_driver_lock_held (struct gpu_buddy *mm){
Quoted 11 lines
Quoted 10 lines
Quoted 10 lines
Quoted 9 lines
include/linux/gpu_buddy.h struct gpu_buddy { · line 217
Quoted 2 lines
include/linux/gpu_buddy.h +2
-0+if ((mm)->lock_dep_map) +lockdep_assert (lock_is_held_type ((mm)->lock_dep_map,0 ));} #else
Quoted 6 lines
include/linux/gpu_buddy.h +3
-0+static inline gpu_buddy_driver_lock_held(struct gpu_buddy *mm) { } +#endif +static inline u64 gpu_buddy_block_offset (const struct gpu_buddy_block *block) {
RFC PATCH v7 3/10 drm/gpu: Add gpu_buddy_allocated_addr_to_block helper
In reply to Tejas UpadhyayAdd helper with primary purpose is to efficiently trace a specific physical memory address back to its corresponding TTM buffer object.
v2:
- %s/gpu_buddy_addr_to_block/gpu_buddy_allocated_addr_to_block(MattA)
- remove clear->avail and split nodes check(MattA)
- Adapt lockdep(MattB)
--- drivers/gpu/buddy.c | 53 +++++++++++++++++++++++++++++++++++++++ include/linux/gpu_buddy.h | 2 ++ 2 files changed, 55 insertions(+)
drivers/gpu/buddy.c +53 -0
@@ -595,6 +595,59 @@ void gpu_buddy_free_block(struct gpu_buddy *mm, }EXPORT_SYMBOL (gpu_buddy_free_block); +/** +* gpu_buddy_allocated_addr_to_block - given relative address find the allocated block +* +* @mm: GPU buddy manager +* @addr: Relative address +* +* Returns: +* gpu_buddy_block on success, NULL or error code on failure +*/ +struct gpu_buddy_block *gpu_buddy_allocated_addr_to_block (struct gpu_buddy *mm,u64 addr) +{ +struct gpu_buddy_block *block; +LIST_HEAD (dfs); +u64 end; +int i; + +gpu_buddy_driver_lock_held (mm); + + end = addr +SZ_4K -1 ; +for (i =0 ; i < mm->n_roots; ++i) +list_add_tail (&mm->roots[i]->tmp_link, &dfs); + +do { +u64 block_start; +u64 block_end; + + block =list_first_entry_or_null (&dfs, + struct gpu_buddy_block, + tmp_link); +if (!block) +break ; + +list_del (&block->tmp_link); + + block_start =gpu_buddy_block_offset (block); + block_end = block_start +gpu_buddy_block_size (mm, block) -1 ; + +if (!overlaps (addr, end, block_start, block_end)) +continue ; + +if (gpu_buddy_block_is_allocated (block)) +return block; +else if (gpu_buddy_block_is_free (block)) +return NULL ; + +list_add (&block->right->tmp_link, &dfs); +list_add (&block->left->tmp_link, &dfs); + }while (1 ); + +return ERR_PTR (-ENXIO ); +} +EXPORT_SYMBOL (gpu_buddy_allocated_addr_to_block); + static void __gpu_buddy_free_list(struct gpu_buddy *mm, struct list_head *objects, bool mark_clear,
include/linux/gpu_buddy.h +2 -0
@@ -272,6 +272,8 @@ void gpu_buddy_reset_clear(struct gpu_buddy *mm, bool is_clear);void gpu_buddy_free_block (struct gpu_buddy *mm,struct gpu_buddy_block *block); +struct gpu_buddy_block *gpu_buddy_allocated_addr_to_block (struct gpu_buddy *mm,u64 addr); +void gpu_buddy_free_list (struct gpu_buddy *mm,struct list_head *objects,unsigned int flags);
Signature · 1 line
2.52.0
RFC PATCH v7 4/10 drm/xe: Link LRC BO and its execution Queue
In reply to Tejas UpadhyayTo establish a link between an LRC BO (Logical Ring Context Buffer Object) and its corresponding execution Queue in the drm/xe driver, you need to store a back-pointer to the queue within the BO's private data structure. This allows the driver to identify and take corrective action on the specific queue if the LRC BO encounters an error (e.g., memory corruption or eviction issues).
--- drivers/gpu/drm/xe/xe_bo_types.h | 3 +++ drivers/gpu/drm/xe/xe_exec_queue.c | 1 + 2 files changed, 4 insertions(+)
drivers/gpu/drm/xe/xe_bo_types.h +3 -0
@@ -20,6 +20,7 @@struct xe_device ;struct xe_mem_pool_node ;struct xe_vm ; +struct xe_exec_queue ;#define XE_BO_MAX_PLACEMENTS 3
@@ -40,6 +41,8 @@ struct xe_bo {u32 flags;/** @vm: VM this BO is attached to, for extobj this will be NULL */ struct xe_vm *vm; +/** @q: Queue this BO is attached to, mostly for LRC BO, NULL otherwise */ +struct xe_exec_queue *q;/** @tile: Tile this BO is attached to (kernel BO only) */ struct xe_tile *tile;/** @placements: valid placements for this BO */
drivers/gpu/drm/xe/xe_exec_queue.c +1 -0
@@ -386,6 +386,7 @@ static int __xe_exec_queue_init(struct xe_exec_queue *q, u32 exec_queue_flags) goto err_lrc; } + lrc->bo->q = q;xe_exec_queue_set_lrc (q, lrc, i);if (__lrc)
Signature · 1 line
2.52.0
Matthew Brost
maintainerRe: [RFC PATCH V7 04/10] drm/xe: Link LRC BO and its execution Queue
Quoted 40 lines · Tejas Upadhyay
To establish a link between an LRC BO (Logical Ring Context Buffer Object) and its corresponding execution Queue in the drm/xe driver, you need to store a back-pointer to the queue within the BO's private data structure. This allows the driver to identify and take corrective action on the specific queue if the LRC BO encounters an error (e.g., memory corruption or eviction issues).
Signed-off-by Tejas Upadhyay <[email protected]>--- drivers/gpu/drm/xe/xe_bo_types.h | 3 +++ drivers/gpu/drm/xe/xe_exec_queue.c | 1 + 2 files changed, 4 insertions(+)drivers/gpu/drm/xe/xe_bo_types.h +3
-0@@ -20,6 +20,7 @@struct xe_device ;struct xe_mem_pool_node ;struct xe_vm ; +struct xe_exec_queue ;#define XE_BO_MAX_PLACEMENTS 3@@ -40,6 +41,8 @@ struct xe_bo {u32 flags;/** @vm: VM this BO is attached to, for extobj this will be NULL */ struct xe_vm *vm; +/** @q: Queue this BO is attached to, mostly for LRC BO, NULL otherwise */ +struct xe_exec_queue *q;/** @tile: Tile this BO is attached to (kernel BO only) */ struct xe_tile *tile;/** @placements: valid placements for this BO */ drivers/gpu/drm/xe/xe_exec_queue.c +0
-0
drivers/gpu/drm/xe/xe_exec_queue.c static int __xe_exec_queue_init(struct xe_exec_queue *q, u32 exec_queue_flags) · line 389
Quoted 4 lines
drivers/gpu/drm/xe/xe_exec_queue.c +1
-0@@ -386,6 +386,7 @@ static int __xe_exec_queue_init(struct xe_exec_queue *q, u32 exec_queue_flags) goto err_lrc; } + lrc->bo->q = q;
I think we need a comment why this is safe.. Something like:
/*
* The queue ref counts the LRCs, thus it safe for the LRC BO to hold a
* pointer to queue without reference.
*/lrc->bo->q = q;
Multi-q comes into play here... So I think the assignment should actually be:
lrc->bo->q = xe_exec_queue_multi_queue_primary(q);
Then to be paranoid in xe_lrc_finish:
lrc->bo->q = NULL;
Before calling: xe_bo_unpin_map_no_vm(lrc->bo);
Matt
Quoted 6 lines
drivers/gpu/drm/xe/xe_exec_queue.c +0
-0xe_exec_queue_set_lrc (q, lrc, i);if (__lrc)Signature · 1 line
2.52.0
RE: [RFC PATCH V7 04/10] drm/xe: Link LRC BO and its execution Queue
Quoted 71 lines
-----Original Message-----
From: Brost, Matthew <[email protected]>
Sent: 30 April 2026 09:04
To: Upadhyay, Tejas <[email protected]>
Cc: [email protected]; Auld, Matthew <[email protected]>; [email protected]; Ghimiray, Himal Prasad <[email protected]>
Subject: Re: [RFC PATCH V7 04/10] drm/xe: Link LRC BO and its execution QueueQuoted 27 lines · Tejas Upadhyay
To establish a link between an LRC BO (Logical Ring Context Buffer Object) and its corresponding execution Queue in the drm/xe driver, you need to store a back-pointer to the queue within the BO's private data structure. This allows the driver to identify and take corrective action on the specific queue if the LRC BO encounters an error (e.g., memory corruption or eviction issues).
Signed-off-by Tejas Upadhyay <[email protected]>--- drivers/gpu/drm/xe/xe_bo_types.h | 3 +++ drivers/gpu/drm/xe/xe_exec_queue.c | 1 + 2 files changed, 4 insertions(+)drivers/gpu/drm/xe/xe_bo_types.h +1
-0@@ -20,6 +20,7 @@struct xe_device ;struct xe_mem_pool_node ;struct xe_vm ; +struct xe_exec_queue ;#define XE_BO_MAX_PLACEMENTS 3drivers/gpu/drm/xe/xe_bo_types.h struct xe_bo { · line 44
Quoted 4 lines
drivers/gpu/drm/xe/xe_bo_types.h +1
-0@@ -40,6 +41,8 @@ struct xe_bo {u32 flags;/** @vm: VM this BO is attached to, for extobj this will be NULL */ struct xe_vm *vm; +/** @q: Queue this BO is attached to, mostly for LRC BO, NULL otherwise */
drivers/gpu/drm/xe/xe_exec_queue.c static int __xe_exec_queue_init(struct
Quoted 10 lines
drivers/gpu/drm/xe/xe_bo_types.h +1
-0+struct xe_exec_queue *q;/** @tile: Tile this BO is attached to (kernel BO only) */ struct xe_tile *tile; /** @placements: valid placements for this BO */ diff --git
a/drivers/gpu/drm/xe/xe_exec_queue.cb/drivers/gpu/drm/xe/xe_exec_queue.c
index071b8c41df43..632c9603afc1100644drivers/gpu/drm/xe/xe_exec_queue.c +0
-0@@ -386,6 +386,7 @@ static int __xe_exec_queue_init(struct
xe_exec_queue*q, u32exec_queue_flags)drivers/gpu/drm/xe/xe_exec_queue.c static int __xe_exec_queue_init(struct xe_exec_queue *q, u32 exec_queue_flags) · line 389
Quoted 4 lines
drivers/gpu/drm/xe/xe_exec_queue.c +1
-0goto err_lrc; } + lrc->bo->q = q;I think we need a comment why this is safe.. Something like:
/*
* The queue ref counts the LRCs, thus it safe for the LRC BO to hold a
* pointer to queue without reference.
*/lrc->bo->q= q;Multi-q comes into play here... So I think the assignment should actually be:
Quoted 7 lines
lrc->bo->q=xe_exec_queue_multi_queue_primary(q);Then to be paranoid in
xe_lrc_finish:
lrc->bo->q= NULL;Before calling:
xe_bo_unpin_map_no_vm(lrc->bo);
Thanks for comment, sure I need to change that.
Tejas
Quoted 9 lines
Matt
Quoted 6 lines
drivers/gpu/drm/xe/xe_exec_queue.c +0
-1xe_exec_queue_set_lrc (q, lrc, i);if (__lrc) --2.52.0
RFC PATCH v7 5/10 drm/xe: Extend BO purge to handle vram pages as well
In reply to Tejas UpadhyayRecent driver update introduce support for purgeable buffer objects (BOs), extending the API to include VRAM pages to better manage memory pressure and enable memory offlining.
--- drivers/gpu/drm/xe/xe_bo.c | 5 +---- drivers/gpu/drm/xe/xe_bo.h | 1 + 2 files changed, 2 insertions(+), 4 deletions(-)
drivers/gpu/drm/xe/xe_bo.c +1 -4
@@ -903,7 +903,7 @@ void xe_bo_set_purgeable_state(struct xe_bo *bo,* * Return: 0 on success, negative error code on failure */ -static int xe_ttm_bo_purge (struct ttm_buffer_object *ttm_bo,struct ttm_operation_ctx *ctx) +int xe_ttm_bo_purge (struct ttm_buffer_object *ttm_bo,struct ttm_operation_ctx *ctx) {struct xe_bo *bo =ttm_to_xe_bo (ttm_bo);struct ttm_placement place = {};
@@ -911,9 +911,6 @@ static int xe_ttm_bo_purge(struct ttm_buffer_object *ttm_bo, struct ttm_operatioxe_bo_assert_held (bo); -if (!ttm_bo->ttm) -return 0 ; -if (!xe_bo_madv_is_dontneed (bo))return 0 ;
drivers/gpu/drm/xe/xe_bo.h +1 -0
@@ -500,6 +500,7 @@ struct xe_bo_shrink_flags {long xe_bo_shrink (struct ttm_operation_ctx *ctx,struct ttm_buffer_object *bo,const struct xe_bo_shrink_flags flags,unsigned long *scanned); +int xe_ttm_bo_purge (struct ttm_buffer_object *ttm_bo,struct ttm_operation_ctx *ctx);/** * xe_bo_is_mem_type - Whether the bo currently resides in the given
Signature · 1 line
2.52.0
Matthew Brost
maintainerRe: [RFC PATCH V7 05/10] drm/xe: Extend BO purge to handle vram pages as well
Quoted 24 lines · Tejas Upadhyay
Recent driver update introduce support for purgeable buffer objects (BOs), extending the API to include VRAM pages to better manage memory pressure and enable memory offlining.
Signed-off-by Tejas Upadhyay <[email protected]>--- drivers/gpu/drm/xe/xe_bo.c | 5 +---- drivers/gpu/drm/xe/xe_bo.h | 1 + 2 files changed, 2 insertions(+), 4 deletions(-)drivers/gpu/drm/xe/xe_bo.c +1
-1@@ -903,7 +903,7 @@ void xe_bo_set_purgeable_state(struct xe_bo *bo,* * Return: 0 on success, negative error code on failure */ -static int xe_ttm_bo_purge (struct ttm_buffer_object *ttm_bo,struct ttm_operation_ctx *ctx) +int xe_ttm_bo_purge (struct ttm_buffer_object *ttm_bo,struct ttm_operation_ctx *ctx) {struct xe_bo *bo =ttm_to_xe_bo (ttm_bo);struct ttm_placement place = {};
drivers/gpu/drm/xe/xe_bo.c static int xe_ttm_bo_purge(struct ttm_buffer_object *ttm_bo, struct ttm_operatio · line 916
Quoted 6 lines
drivers/gpu/drm/xe/xe_bo.c +0
-3@@ -911,9 +911,6 @@ static int xe_ttm_bo_purge(struct ttm_buffer_object *ttm_bo, struct ttm_operatioxe_bo_assert_held (bo); -if (!ttm_bo->ttm) -return 0 ; -
I think blindly deleting this is fine but I'd run this by Arvind to make sure in normal purge flow this won't do anything bad but afiak in normal purge flows ttm_bo->ttm will never be NULL.
If not then adjust the existing call...
@@ -978,7 +978,8 @@ static int xe_bo_move(struct ttm_buffer_object *ttm_bo, bool evict,
* The move_notify callback will handle invalidation asynchronously.
*/
if (evict && xe_bo_madv_is_dontneed(bo)) {
- ret = xe_ttm_bo_purge(ttm_bo, ctx);
+ if (ttm_bo->ttm)
+ ret = xe_ttm_bo_purge(ttm_bo, ctx);
if (ret)
return ret;Matt
Quoted 18 lines
drivers/gpu/drm/xe/xe_bo.c +0
-0if (!xe_bo_madv_is_dontneed (bo))return 0 ;drivers/gpu/drm/xe/xe_bo.h +1
-0@@ -500,6 +500,7 @@ struct xe_bo_shrink_flags {long xe_bo_shrink (struct ttm_operation_ctx *ctx,struct ttm_buffer_object *bo,const struct xe_bo_shrink_flags flags,unsigned long *scanned); +int xe_ttm_bo_purge (struct ttm_buffer_object *ttm_bo,struct ttm_operation_ctx *ctx);/** * xe_bo_is_mem_type - Whether the bo currently resides in the given Signature · 1 line
2.52.0
RE: [RFC PATCH V7 05/10] drm/xe: Extend BO purge to handle vram pages as well
Quoted 43 lines
-----Original Message-----
From: Brost, Matthew <[email protected]>
Sent: 30 April 2026 09:14
To: Upadhyay, Tejas <[email protected]>
Cc: [email protected]; Auld, Matthew <[email protected]>; [email protected]; Ghimiray, Himal Prasad <[email protected]>
Subject: Re: [RFC PATCH V7 05/10] drm/xe: Extend BO purge to handle vram pages as wellQuoted 30 lines · Tejas Upadhyay
Recent driver update introduce support for purgeable buffer objects (BOs), extending the API to include VRAM pages to better manage memory pressure and enable memory offlining.
Signed-off-by Tejas Upadhyay <[email protected]>--- drivers/gpu/drm/xe/xe_bo.c | 5 +---- drivers/gpu/drm/xe/xe_bo.h | 1 + 2 files changed, 2 insertions(+), 4 deletions(-)drivers/gpu/drm/xe/xe_bo.c +2
-1@@ -903,7 +903,7 @@ void xe_bo_set_purgeable_state(struct xe_bo *bo,* * Return: 0 on success, negative error code on failure */ -static int xe_ttm_bo_purge (struct ttm_buffer_object *ttm_bo,struct ttm_operation_ctx *ctx) +int xe_ttm_bo_purge (struct ttm_buffer_object *ttm_bo,struct +ttm_operation_ctx *ctx) {struct xe_bo *bo =ttm_to_xe_bo (ttm_bo);struct ttm_placement place = {};@@ -911,9 +911,6 @@ static int xe_ttm_bo_purge(struct
ttm_buffer_object*ttm_bo, structttm_operatio
xe_bo_assert_held(bo);
Quoted 7 lines
drivers/gpu/drm/xe/xe_bo.c static int xe_ttm_bo_purge(struct ttm_buffer_object *ttm_bo, struct ttm_operatio · line 916
Quoted 3 lines
drivers/gpu/drm/xe/xe_bo.c +0
-3-if (!ttm_bo->ttm) -return 0 ; -I think blindly deleting this is fine but I'd run this by Arvind to make sure in normal purge flow this won't do anything bad but afiak in normal purge flows
ttm_bo->ttmwill never be NULL.
Thanks, I had communicated with Aravind before sending out this and we validated with discussion that there will never be ttm_bo->ttm NULL. Still I will discuss with him again.
Tejas
Quoted 35 lines
If not then adjust the existing call...
(quoted diff) +2
-1@@ -978,7 +978,8 @@ static int xe_bo_move(struct ttm_buffer_object *ttm_bo, bool evict, * The move_notify callback will handle invalidation asynchronously. */ if (evict && xe_bo_madv_is_dontneed(bo)) { - ret = xe_ttm_bo_purge(ttm_bo, ctx); + if (ttm_bo->ttm) + ret = xe_ttm_bo_purge(ttm_bo, ctx); if (ret) return ret;Matt
Quoted 19 lines
drivers/gpu/drm/xe/xe_bo.c +0
-0if (!xe_bo_madv_is_dontneed (bo))return 0 ;drivers/gpu/drm/xe/xe_bo.h +2
-1@@ -500,6 +500,7 @@ struct xe_bo_shrink_flags { long xe_bo_shrink(struct ttm_operation_ctx *ctx, struct ttm_buffer_object *bo, const struct xe_bo_shrink_flags flags, unsigned long *scanned); +int xe_ttm_bo_purge (struct ttm_buffer_object *ttm_bo,struct +ttm_operation_ctx *ctx);/** * xe_bo_is_mem_type - Whether the bo currently resides in the given -- 2.52.0
Re: [RFC PATCH V7 05/10] drm/xe: Extend BO purge to handle vram pages as well
In reply to Matthew BrostQuoted 26 lines · Matthew Brost
Quoted 25 lines · Tejas Upadhyay
Recent driver update introduce support for purgeable buffer objects (BOs), extending the API to include VRAM pages to better manage memory pressure and enable memory offlining.
Signed-off-by Tejas Upadhyay<[email protected]>--- drivers/gpu/drm/xe/xe_bo.c | 5 +---- drivers/gpu/drm/xe/xe_bo.h | 1 + 2 files changed, 2 insertions(+), 4 deletions(-)drivers/gpu/drm/xe/xe_bo.c +1
-1@@ -903,7 +903,7 @@ void xe_bo_set_purgeable_state(struct xe_bo *bo,* * Return: 0 on success, negative error code on failure */ -static int xe_ttm_bo_purge (struct ttm_buffer_object *ttm_bo,struct ttm_operation_ctx *ctx) +int xe_ttm_bo_purge (struct ttm_buffer_object *ttm_bo,struct ttm_operation_ctx *ctx) {struct xe_bo *bo =ttm_to_xe_bo (ttm_bo);struct ttm_placement place = {};@@ -911,9 +911,6 @@ static int xe_ttm_bo_purge(struct ttm_buffer_object *ttm_bo, struct ttm_operatio
Quoted 8 lines
drivers/gpu/drm/xe/xe_bo.c static int xe_ttm_bo_purge(struct ttm_buffer_object *ttm_bo, struct ttm_operatio · line 916
Quoted 5 lines
drivers/gpu/drm/xe/xe_bo.c +0
-3xe_bo_assert_held (bo); -if (!ttm_bo->ttm) -return 0 ; -I think blindly deleting this is fine but I'd run this by Arvind to make sure in normal purge flow this won't do anything bad but afiak in normal purge flows
ttm_bo->ttmwill never be NULL.
The purge paths (|xe_bo_move(evict=true, DONTNEED)| and |swap_notify|), |ttm_bo->ttm| is always non‑NULL. TTM allocates |ttm_tt| before invoking |->move| on a populated BO. So removing this check is safe.
~Arvind
Quoted 33 lines
If not then adjust the existing call...
(quoted diff) +2
-1@@ -978,7 +978,8 @@ static int xe_bo_move(struct ttm_buffer_object *ttm_bo, bool evict, * The move_notify callback will handle invalidation asynchronously. */ if (evict && xe_bo_madv_is_dontneed(bo)) { - ret = xe_ttm_bo_purge(ttm_bo, ctx); + if (ttm_bo->ttm) + ret = xe_ttm_bo_purge(ttm_bo, ctx); if (ret) return ret;Matt
Quoted 18 lines
drivers/gpu/drm/xe/xe_bo.c +0
-0if (!xe_bo_madv_is_dontneed (bo))return 0 ;drivers/gpu/drm/xe/xe_bo.h +1
-0@@ -500,6 +500,7 @@ struct xe_bo_shrink_flags {long xe_bo_shrink (struct ttm_operation_ctx *ctx,struct ttm_buffer_object *bo,const struct xe_bo_shrink_flags flags,unsigned long *scanned); +int xe_ttm_bo_purge (struct ttm_buffer_object *ttm_bo,struct ttm_operation_ctx *ctx);/** * xe_bo_is_mem_type - Whether the bo currently resides in the given Signature · 1 line
2.52.0
RFC PATCH v7 6/10 drm/xe: Handle physical memory address error
In reply to Tejas UpadhyayThis functionality represents a significant step in making the xe driver gracefully handle hardware memory degradation. By integrating with the DRM Buddy allocator, the driver can permanently "carve out" faulty memory so it isn't reused by subsequent allocations.
Buddy Block Reservation:
----------------------
When a memory address is reported as faulty, the driver instructs the DRM Buddy allocator to reserve a block of the specific page size (typically 4KB). This marks the memory as "dirty/used" indefinitely.
Two-Stage Tracking:
-----------------
Offlined Pages:
Pages that have been successfully isolated and removed from the available memory pool.
Queued Pages:
Addresses that have been flagged as faulty but are currently in use by a process. These are tracked until the associated buffer object (BO) is released or migrated, at which point they move to the "offlined" state.
v7:
- keep vm ref during vm kill and fix some typos
- FW communication code is moved in RAS, keep comment for same V6:
- Use scope_guard for locking(MattB)
- Adapt addition of queue member of LRC BO(MattB)
- Extend and use xe_ttm_bo_purge API for vram pages(MattB)
- Handle dma_buf_map requests for native and remote(MattB)
- Address if in never initialized block, set block to NULL V5:
- Categorise and handle BOs accordingly
- Fix crash found with new debugfs tests
V4:
- Set block->private NULL post bo purge
- Filter out gsm address early on
- Rebase
V3:
-rename api, remove tile dependency and add status of reservation V2:
- Fix mm->avail counter issue
- Remove unused code and handle clean up in case of error
--- drivers/gpu/drm/xe/xe_bo.c | 11 +- drivers/gpu/drm/xe/xe_bo.h | 4 +- drivers/gpu/drm/xe/xe_dma_buf.c | 3 + drivers/gpu/drm/xe/xe_exec_queue.c | 9 +- drivers/gpu/drm/xe/xe_pt.c | 3 +- drivers/gpu/drm/xe/xe_ttm_vram_mgr.c | 273 +++++++++++++++++++++ drivers/gpu/drm/xe/xe_ttm_vram_mgr.h | 1 + drivers/gpu/drm/xe/xe_ttm_vram_mgr_types.h | 28 +++ 8 files changed, 326 insertions(+), 6 deletions(-)
drivers/gpu/drm/xe/xe_bo.c +10 -1
@@ -158,7 +158,16 @@ bool xe_bo_is_vm_bound(struct xe_bo *bo)return !list_empty (&bo->ttm.base.gpuva.list); } -static bool xe_bo_is_user (struct xe_bo *bo) +/** +* xe_bo_is_user - check if BO is user created BO +* @bo: The BO +* +* Check if BO is user created BO. This requires the +* reservation lock for the BO to be held. +* +* Returns: boolean +*/ +bool xe_bo_is_user (struct xe_bo *bo) {return bo->flags &XE_BO_FLAG_USER ; }
drivers/gpu/drm/xe/xe_bo.h +3 -1
@@ -277,7 +277,8 @@ static inline void xe_bo_unpin_map_no_vm(struct xe_bo *bo) {if (likely (bo)) {xe_bo_lock (bo, false); -xe_bo_unpin (bo); +if (!xe_bo_is_purged (bo)) +xe_bo_unpin (bo);xe_bo_unlock (bo);xe_bo_put (bo);
@@ -501,6 +502,7 @@ long xe_bo_shrink(struct ttm_operation_ctx *ctx, struct ttm_buffer_object *bo,const struct xe_bo_shrink_flags flags, unsigned long *scanned);int xe_ttm_bo_purge (struct ttm_buffer_object *ttm_bo,struct ttm_operation_ctx *ctx); +bool xe_bo_is_user (struct xe_bo *bo);/** * xe_bo_is_mem_type - Whether the bo currently resides in the given
drivers/gpu/drm/xe/xe_dma_buf.c +3 -0
@@ -104,6 +104,9 @@ static struct sg_table *xe_dma_buf_map(struct dma_buf_attachment *attach,struct sg_table *sgt;int r =0 ; +if (xe_bo_is_purged (bo)) +return ERR_PTR (-ENOENT ); +if (!attach->peer2peer && !xe_bo_can_migrate (bo,XE_PL_TT ))return ERR_PTR (-EOPNOTSUPP );
drivers/gpu/drm/xe/xe_exec_queue.c +6 -3
@@ -385,7 +385,6 @@ static int __xe_exec_queue_init(struct xe_exec_queue *q, u32 exec_queue_flags) err =PTR_ERR (lrc); goto err_lrc; } - lrc->bo->q = q;xe_exec_queue_set_lrc (q, lrc, i);
@@ -1555,8 +1554,12 @@ void xe_exec_queue_update_run_ticks(struct xe_exec_queue *q)* errors. */ lrc = q->lrc[0 ]; - new_ts =xe_lrc_update_timestamp (lrc, &old_ts); - q->xef->run_ticks[q->class] += (new_ts - old_ts) * q->width; +xe_bo_lock (lrc->bo, false); +if (!xe_bo_is_purged (lrc->bo)) { + new_ts =xe_lrc_update_timestamp (lrc, &old_ts); + q->xef->run_ticks[q->class] += (new_ts - old_ts) * q->width; + } +xe_bo_unlock (lrc->bo);drm_dev_exit (idx); }
drivers/gpu/drm/xe/xe_pt.c +2 -1
@@ -211,7 +211,8 @@ void xe_pt_destroy(struct xe_pt *pt, u32 flags, struct llist_head *deferred)return ;XE_WARN_ON (!list_empty (&pt->bo->ttm.base.gpuva.list)); -xe_bo_unpin (pt->bo); +if (!xe_bo_is_purged (pt->bo)) +xe_bo_unpin (pt->bo);xe_bo_put_deferred (pt->bo, deferred);if (pt->level >0 && pt->num_live) {
drivers/gpu/drm/xe/xe_ttm_vram_mgr.c +273 -0
301 lines not on this page · raw message
drivers/gpu/drm/xe/xe_ttm_vram_mgr.h +1 -0
@@ -31,6 +31,7 @@ u64 xe_ttm_vram_get_cpu_visible_size(struct ttm_resource_manager *man);void xe_ttm_vram_get_used (struct ttm_resource_manager *man,u64 *used,u64 *used_visible); +int xe_ttm_vram_handle_addr_fault (struct xe_device *xe,unsigned long addr);static inline struct xe_ttm_vram_mgr_resource *to_xe_ttm_vram_mgr_resource (struct ttm_resource *res) {
drivers/gpu/drm/xe/xe_ttm_vram_mgr_types.h +28 -0
@@ -19,6 +19,14 @@ struct xe_ttm_vram_mgr {struct ttm_resource_manager manager;/** @mm: DRM buddy allocator which manages the VRAM */ struct gpu_buddy mm; +/** @offlined_pages: List of offlined pages */ +struct list_head offlined_pages; +/** @n_offlined_pages: Number of offlined pages */ +u16 n_offlined_pages; +/** @queued_pages: List of queued pages */ +struct list_head queued_pages; +/** @n_queued_pages: Number of queued pages */ +u16 n_queued_pages;/** @visible_size: Proped size of the CPU visible portion */ u64 visible_size;/** @visible_avail: CPU visible portion still unallocated */
@@ -45,4 +53,24 @@ struct xe_ttm_vram_mgr_resource { unsigned long flags; }; +/** +* struct xe_ttm_vram_offline_resource - Xe TTM VRAM offline resource +*/ +struct xe_ttm_vram_offline_resource { +/** @offlined_link: Link to offlined pages */ +struct list_head offlined_link; +/** @queued_link: Link to queued pages */ +struct list_head queued_link; +/** @blocks: list of DRM buddy blocks */ +struct list_head blocks; +/** @used_visible_size: How many CPU visible bytes this resource is using */ +u64 used_visible_size; +/** @id: The id of an offline resource */ +u16 id; +/** @addr: Address of faulty memory location reported by HW */ +unsigned long addr; +/** @status: reservation status of resource */ +bool status; +}; +#endif
Signature · 1 line
2.52.0
RFC PATCH v7 7/10 drm/xe/cri: Add debugfs to inject faulty vram address
In reply to Tejas UpadhyayAdd debugfs which can help testing feature with manual error injection. Adding a debugfs interface to the drm/xe driver allows manual injection of faulty VRAM addresses, facilitating the testing of the CRI memory page offline feature before it is fully functional. The implementation involves creating a debugfs entry, likely under /sys/kernel/debug/dri/bdf/invalid_addr_vram0,
to accept specific faulty addresses for validation.
For example,
echo 0 > /sys/kernel/debug/dri/bdf/invalid_addr_vram0 where 0 is below address types to be tested, enum mempage_offline_mode {
MEMPAGE_OFFLINE_UNALLOCATED = 0, MEMPAGE_OFFLINE_USER_ALLOCATED = 1, MEMPAGE_OFFLINE_KERNEL_USER_GGTT_ALLOCATED = 2, MEMPAGE_OFFLINE_KERNEL_USER_PPGTT_ALLOCATED = 3, MEMPAGE_OFFLINE_KERNEL_CRITICAL_ALLOCATED = 4, MEMPAGE_OFFLINE_RESERVED = 5
};
v4:
- Use scope_guard around lock, adapt bo->q and enhance warn messages
- %s/gpu_buddy_addr_to_block/gpu_buddy_allocated_addr_to_block v3:
- Add more specific noncritical bo tests
v2:
- Add mode based automated test vs manual address feed
--- drivers/gpu/drm/xe/xe_debugfs.c | 171 +++++++++++++++++++++ drivers/gpu/drm/xe/xe_ttm_vram_mgr_types.h | 2 + 2 files changed, 173 insertions(+)
drivers/gpu/drm/xe/xe_debugfs.c +171 -0
@@ -14,6 +14,7 @@#include "regs/xe_pmt.h" #include "xe_bo.h" #include "xe_device.h" +#include "xe_exec_queue_types.h" #include "xe_force_wake.h" #include "xe_gt.h" #include "xe_gt_debugfs.h"
@@ -21,6 +22,7 @@#include "xe_guc_ads.h" #include "xe_hw_engine.h" #include "xe_mmio.h" +#include "xe_migrate.h" #include "xe_pm.h" #include "xe_psmi.h" #include "xe_pxp_debugfs.h"
@@ -29,6 +31,8 @@#include "xe_sriov_vf.h" #include "xe_step.h" #include "xe_tile_debugfs.h" +#include "xe_ttm_stolen_mgr.h" +#include "xe_ttm_vram_mgr.h" #include "xe_vsec.h" #include "xe_wa.h"
@@ -40,6 +44,14 @@DECLARE_FAULT_ATTR (gt_reset_failure);DECLARE_FAULT_ATTR (inject_csc_hw_error); +enum mempage_offline_mode { +MEMPAGE_OFFLINE_UNALLOCATED =0 , +MEMPAGE_OFFLINE_USER_ALLOCATED =1 , +MEMPAGE_OFFLINE_KERNEL_USER_GGTT_ALLOCATED =2 , +MEMPAGE_OFFLINE_KERNEL_USER_PPGTT_ALLOCATED =3 , +MEMPAGE_OFFLINE_KERNEL_CRITICAL_ALLOCATED =4 , +MEMPAGE_OFFLINE_RESERVED =5 , +};static void read_residency_counter (struct xe_device *xe,struct xe_mmio *mmio,u32 offset,const char *name,struct drm_printer *p)
@@ -544,6 +556,154 @@ static const struct file_operations disable_late_binding_fops = { .write = disable_late_binding_set, }; +static ssize_t addr_fault_reporting_show (struct file *f,char __user *ubuf, +size_t size,loff_t *pos) +{ +struct xe_device *xe =file_inode (f)->i_private; +char buf[32 ]; +int len; + + len =scnprintf (buf,sizeof (buf),"%lld\n" , xe->mem.vram->ttm.offline_mode); + +return simple_read_from_buffer (ubuf, size, pos, buf, len); +} + +static int mempage_exec_offline (struct xe_device *xe,u64 mode) +{ +struct xe_tile *tile =xe_device_get_root_tile (xe); +struct xe_vram_region *vr = tile->mem.vram; +struct ttm_buffer_object *tbo =NULL ; +struct xe_ttm_vram_mgr *vram_mgr; +struct gpu_buddy_block *block; +bool do_offline = false; +struct gpu_buddy *mm; +struct xe_bo *bo; +u64 addr =0x0 ; +int ret =0 ; + + vram_mgr = &vr->ttm; + mm = &vram_mgr->mm; + addr = vr->dpa_base; +while (addr <= vr->dpa_base + vr->actual_physical_size) { + scoped_guard(mutex, &vram_mgr->lock) { + block =gpu_buddy_allocated_addr_to_block (mm, addr); +if (!block && mode ==MEMPAGE_OFFLINE_UNALLOCATED ) + do_offline = true; +if (block &&PTR_ERR (block) != -ENXIO ) { +if (!block->private) { + addr = addr +SZ_4K ; + do_offline = false; +continue ; + } + tbo = block->private; + bo =ttm_to_xe_bo (tbo); +if (bo->ttm.type == ttm_bo_type_device && + bo->flags &XE_BO_FLAG_USER && + bo->flags &XE_BO_FLAG_VRAM_MASK && + mode ==MEMPAGE_OFFLINE_USER_ALLOCATED ) { + do_offline = true; + }else if (bo->q && + mode ==MEMPAGE_OFFLINE_KERNEL_USER_GGTT_ALLOCATED ) { +/* lrc */ +struct xe_vm *migrate_vm; + + migrate_vm =xe_migrate_get_vm (tile->migrate); +if (migrate_vm != bo->q->vm) + do_offline = true; +xe_vm_put (migrate_vm); + }else if (bo->ttm.type == ttm_bo_type_kernel && + bo->flags &XE_BO_FLAG_FORCE_USER_VRAM && + bo->flags &XE_BO_FLAG_PAGETABLE && + mode ==MEMPAGE_OFFLINE_KERNEL_USER_PPGTT_ALLOCATED ) { +/* ppgtt */ + do_offline = true; + }else if (bo->ttm.type == ttm_bo_type_kernel && + !(bo->flags &XE_BO_FLAG_FORCE_USER_VRAM ) && + mode ==MEMPAGE_OFFLINE_KERNEL_CRITICAL_ALLOCATED ) { + do_offline = true; + } + } + } +if (do_offline) { +/* Report fault */ + ret =xe_ttm_vram_handle_addr_fault (xe, addr); +if (ret) { +if ((ret == -EIO ) && + mode ==MEMPAGE_OFFLINE_KERNEL_USER_GGTT_ALLOCATED ) { + addr = addr +SZ_4K ; +if (do_offline) + do_offline = false; +continue ; + } +break ; + } +/* Verify addr + SZ_4K is allocated */ + scoped_guard(mutex, &vram_mgr->lock) { + block =gpu_buddy_allocated_addr_to_block (mm, addr); +if (!block ||PTR_ERR (block) == -ENXIO || block->private) + ret = -EBUSY ; + } +break ; + } + addr = addr +SZ_4K ; +if (do_offline) + do_offline = false; + } +if (!do_offline) +drm_warn (&xe->drm,"no such object, ret:%d\n" , ret); + +return ret; +} + +static ssize_t addr_fault_reporting_set (struct file *f,const char __user *ubuf, +size_t size,loff_t *pos) +{ +struct xe_device *xe =file_inode (f)->i_private; +int ret =0 ; +u64 mode; + + ret =kstrtou64_from_user (ubuf, size,0 , &mode); +if (ret) +return ret; + +switch (mode) { +case MEMPAGE_OFFLINE_UNALLOCATED : +case MEMPAGE_OFFLINE_USER_ALLOCATED : +case MEMPAGE_OFFLINE_KERNEL_USER_GGTT_ALLOCATED : +case MEMPAGE_OFFLINE_KERNEL_USER_PPGTT_ALLOCATED : +case MEMPAGE_OFFLINE_KERNEL_CRITICAL_ALLOCATED : + ret =mempage_exec_offline (xe, mode); +break ; +case MEMPAGE_OFFLINE_RESERVED : +u64 stolen_base; + + stolen_base =xe_ttm_stolen_gpu_offset (xe); + ret =xe_ttm_vram_handle_addr_fault (xe, stolen_base); +break ; +default : + ret = -EINVAL ; +break ; + } + + xe->mem.vram->ttm.offline_mode = mode; +if (!ret || (ret == -EIO && + (mode ==MEMPAGE_OFFLINE_KERNEL_CRITICAL_ALLOCATED || + mode ==MEMPAGE_OFFLINE_RESERVED ))) { +drm_info (&xe->drm,"offline mode %llu passed ret:%d\n" , mode, ret); + }else { +drm_warn (&xe->drm,"offline mode %llu failed, ret:%d\n" , mode, ret); +return ret; + } + +return size; +} + +static const struct file_operations addr_fault_reporting_fops = { + .owner =THIS_MODULE , + .read = addr_fault_reporting_show, + .write = addr_fault_reporting_set, +}; +void xe_debugfs_register (struct xe_device *xe) {struct ttm_device *bdev = &xe->ttm;
@@ -600,6 +760,17 @@ void xe_debugfs_register(struct xe_device *xe)if (man)ttm_resource_manager_create_debugfs (man, root,"stolen_mm" ); +if (xe->info.platform ==XE_CRESCENTISLAND ) { + man =ttm_manager_type (bdev,XE_PL_VRAM0 ); +if (man) { +char name[20 ]; + +snprintf (name,sizeof (name),"invalid_addr_vram%d" ,0 ); +debugfs_create_file (name,0600 , root, xe, + &addr_fault_reporting_fops); + } + } +for_each_tile (tile, xe, tile_id)xe_tile_debugfs_register (tile);
drivers/gpu/drm/xe/xe_ttm_vram_mgr_types.h +2 -0
@@ -37,6 +37,8 @@ struct xe_ttm_vram_mgr {struct mutex lock;/** @mem_type: The TTM memory type */ u32 mem_type; +/** @offline_mode: debugfs hook for setting page offline mode */ +u64 offline_mode; };/**
Signature · 1 line
2.52.0
RFC PATCH v7 8/10 gpu/buddy: Add routine to dump allocated buddy blocks
In reply to Tejas UpadhyayTo implement the ability to see allocated blocks under a specific VRAM instance in the drm driver, new api is introduced. While existing structs often show the free block list, this addition provides a comprehensive view of all currently resident VRAM allocations.
Dump will look like,
[ +0.000003] xe 0000:03:00.0: [drm] 0x00000002f8000000-0x00000002f8800000: 8388608 [ +0.000005] xe 0000:03:00.0: [drm] 0x00000002f8800000-0x00000002f8840000: 262144 [ +0.000004] xe 0000:03:00.0: [drm] 0x00000002f8840000-0x00000002f8860000: 131072 [ +0.000004] xe 0000:03:00.0: [drm] 0x00000002f8860000-0x00000002f8870000: 65536 [ +0.000005] xe 0000:03:00.0: [drm] 0x00000002f9000000-0x00000002f9800000: 8388608 [ +0.000004] xe 0000:03:00.0: [drm] 0x00000002f9800000-0x00000002f9880000: 524288 [ +0.000005] xe 0000:03:00.0: [drm] 0x00000002f9880000-0x00000002f9884000: 16384 [ +0.000004] xe 0000:03:00.0: [drm] 0x00000002f9900000-0x00000002f9980000: 524288 [ +0.000005] xe 0000:03:00.0: [drm] 0x00000002f9980000-0x00000002f9988000: 32768 [ +0.000004] xe 0000:03:00.0: [drm] 0x00000002f9988000-0x00000002f998c000: 16384
v2(MattB):
- Add lockdep assert
--- drivers/gpu/buddy.c | 43 +++++++++++++++++++++++++++++++++++++++ include/linux/gpu_buddy.h | 1 + 2 files changed, 44 insertions(+)
drivers/gpu/buddy.c +43 -0
@@ -10,6 +10,7 @@#include <linux/sizes.h> #include <linux/gpu_buddy.h> +#include <drm/drm_print.h> /** * gpu_buddy_assert - assert a condition in the buddy allocator
@@ -1295,6 +1296,48 @@ int gpu_buddy_block_trim(struct gpu_buddy *mm, }EXPORT_SYMBOL (gpu_buddy_block_trim); +/** +* gpu_buddy_dump_allocated_blocks - print all allocated blocks in drm buddy +* +* @mm: DRM buddy manager to look into +* +* Looks into buddy manager for each block and their status and if allocated +* print allocated block range and size +* +* Returns: +* void +*/ +void gpu_buddy_dump_allocated_blocks (struct gpu_buddy *mm) +{ +struct gpu_buddy_block *block; +LIST_HEAD (dfs); +int i; + +gpu_buddy_driver_lock_held (mm); + +for (i =0 ; i < mm->n_roots; ++i) +list_add_tail (&mm->roots[i]->tmp_link, &dfs); + +do { + block =list_first_entry_or_null (&dfs, + struct gpu_buddy_block, + tmp_link); +if (!block) +break ; + +list_del (&block->tmp_link); + +if (gpu_buddy_block_is_allocated (block)) +gpu_buddy_block_print (mm, block); + +if (gpu_buddy_block_is_split (block)) { +list_add (&block->right->tmp_link, &dfs); +list_add (&block->left->tmp_link, &dfs); + } + }while (1 ); +} +EXPORT_SYMBOL (gpu_buddy_dump_allocated_blocks); +static struct gpu_buddy_block * __gpu_buddy_alloc_blocks(struct gpu_buddy *mm, u64 start, u64 end,
include/linux/gpu_buddy.h +1 -0
@@ -267,6 +267,7 @@ int gpu_buddy_block_trim(struct gpu_buddy *mm,u64 *start, u64 new_size, struct list_head *blocks); +void gpu_buddy_dump_allocated_blocks (struct gpu_buddy *mm);void gpu_buddy_reset_clear (struct gpu_buddy *mm,bool is_clear);
Signature · 1 line
2.52.0
RFC PATCH v7 9/10 drm/xe/configfs: Add vram bad page reservation policy
In reply to Tejas UpadhyayThe interface enables setting the policy for how bad pages are handled in VRAM. This is crucial for maintaining system stability in scenarios where VRAM degradation occurs.
By default policy will be "reserve", which can be changed to "logging" only.
v3:
- All FW communication moved under RAS
v2:
- Add CRI check and rebase
--- drivers/gpu/drm/xe/xe_configfs.c | 64 +++++++++++++++++++++++++++- drivers/gpu/drm/xe/xe_configfs.h | 2 + drivers/gpu/drm/xe/xe_ttm_vram_mgr.c | 10 +++++ 3 files changed, 75 insertions(+), 1 deletion(-)
drivers/gpu/drm/xe/xe_configfs.c +63 -1
@@ -61,7 +61,8 @@* ├── survivability_mode * ├── gt_types_allowed * ├── engines_allowed -* └── enable_psmi +* ├── enable_psmi +* └── bad_page_reservation * * After configuring the attributes as per next section, the device can be * probed with::
@@ -159,6 +160,16 @@* * This attribute can only be set before binding to the device. * +* Bad pages reservation: +* --------------------- +* +* Disable vram bad pages reservation, instead just report it in dmesg. +* Example to disable it:: +* +* # echo 0 > /sys/kernel/config/xe/0000:03:00.0/bad_page_reservation +* +* This attribute can only be set before binding to the device. +* * Context restore BB * ------------------ *
@@ -262,6 +273,7 @@ struct xe_config_group_device {struct wa_bb ctx_restore_mid_bb[XE_ENGINE_CLASS_MAX ];bool survivability_mode;bool enable_psmi; +bool bad_page_reservation;struct {unsigned int max_vfs;bool admin_only_pf;
@@ -281,6 +293,7 @@ static const struct xe_config_device device_defaults = { .engines_allowed = U64_MAX, .survivability_mode = false, .enable_psmi = false, + .bad_page_reservation = true, .sriov = { .max_vfs = XE_DEFAULT_MAX_VFS, .admin_only_pf = XE_DEFAULT_ADMIN_ONLY_PF,
@@ -575,6 +588,32 @@ static ssize_t enable_psmi_store(struct config_item *item, const char *page, sizreturn len; } +static ssize_t bad_page_reservation_show (struct config_item *item,char *page) +{ +struct xe_config_device *dev =to_xe_config_device (item); + +return sprintf (page,"%d\n" , dev->bad_page_reservation); +} + +static ssize_t bad_page_reservation_store (struct config_item *item,const char *page,size_t len) +{ +struct xe_config_group_device *dev =to_xe_config_group_device (item); +bool val; +int ret; + + ret =kstrtobool (page, &val); +if (ret) +return ret; + +guard (mutex)(&dev->lock); +if (is_bound (dev)) +return -EBUSY ; + + dev->config.bad_page_reservation = val; + +return len; +} +static bool wa_bb_read_advance (bool dereference,char **p,const char *append,size_t len,size_t *max_size)
@@ -813,6 +852,7 @@ static ssize_t ctx_restore_post_bb_store(struct config_item *item, CONFIGFS_ATTR(, ctx_restore_mid_bb); CONFIGFS_ATTR(, ctx_restore_post_bb); CONFIGFS_ATTR(, enable_psmi); +CONFIGFS_ATTR(, bad_page_reservation); CONFIGFS_ATTR(, engines_allowed); CONFIGFS_ATTR(, gt_types_allowed); CONFIGFS_ATTR(, survivability_mode);
@@ -821,6 +861,7 @@ static struct configfs_attribute *xe_config_device_attrs[] = { &attr_ctx_restore_mid_bb, &attr_ctx_restore_post_bb, &attr_enable_psmi, + &attr_bad_page_reservation, &attr_engines_allowed, &attr_gt_types_allowed, &attr_survivability_mode,
@@ -1098,6 +1139,7 @@ static void dump_custom_dev_config(struct pci_dev *pdev,PRI_CUSTOM_ATTR ("%llx" , gt_types_allowed);PRI_CUSTOM_ATTR ("%llx" , engines_allowed);PRI_CUSTOM_ATTR ("%d" , enable_psmi); +PRI_CUSTOM_ATTR ("%d" , bad_page_reservation);PRI_CUSTOM_ATTR ("%d" , survivability_mode);PRI_CUSTOM_ATTR ("%u" , sriov.admin_only_pf);
@@ -1225,6 +1267,26 @@ bool xe_configfs_get_psmi_enabled(struct pci_dev *pdev)return ret; } +/** +* xe_configfs_get_bad_page_reservation - get configfs bad_page_reservation setting +* @pdev: pci device +* +* Return: bad_page_reservation setting in configfs +*/ +bool xe_configfs_get_bad_page_reservation (struct pci_dev *pdev) +{ +struct xe_config_group_device *dev =find_xe_config_group_device (pdev); +bool ret; + +if (!dev) +return device_defaults.bad_page_reservation; + + ret = dev->config.bad_page_reservation; +config_group_put (&dev->group); + +return ret; +} +/** * xe_configfs_get_ctx_restore_mid_bb - get configfs ctx_restore_mid_bb setting * @pdev: pci device
drivers/gpu/drm/xe/xe_configfs.h +2 -0
@@ -23,6 +23,7 @@ bool xe_configfs_primary_gt_allowed(struct pci_dev *pdev);bool xe_configfs_media_gt_allowed (struct pci_dev *pdev);u64 xe_configfs_get_engines_allowed (struct pci_dev *pdev);bool xe_configfs_get_psmi_enabled (struct pci_dev *pdev); +bool xe_configfs_get_bad_page_reservation (struct pci_dev *pdev);u32 xe_configfs_get_ctx_restore_mid_bb (struct pci_dev *pdev,enum xe_engine_class class,const u32 **cs);
@@ -42,6 +43,7 @@ static inline bool xe_configfs_primary_gt_allowed(struct pci_dev *pdev) { returnstatic inline bool xe_configfs_media_gt_allowed (struct pci_dev *pdev) {return true; }static inline u64 xe_configfs_get_engines_allowed (struct pci_dev *pdev) {return U64_MAX ; }static inline bool xe_configfs_get_psmi_enabled (struct pci_dev *pdev) {return false; } +static inline bool xe_configfs_get_bad_page_reservation (struct pci_dev *pdev) {return true; }static inline u32 xe_configfs_get_ctx_restore_mid_bb (struct pci_dev *pdev,enum xe_engine_class class,const u32 **cs) {return 0 ; }
drivers/gpu/drm/xe/xe_ttm_vram_mgr.c +10 -0
@@ -12,6 +12,7 @@#include <drm/ttm/ttm_range_manager.h> #include "xe_bo.h" +#include "xe_configfs.h" #include "xe_device.h" #include "xe_exec_queue.h" #include "xe_lrc.h"
@@ -731,6 +732,7 @@ int xe_ttm_vram_handle_addr_fault(struct xe_device *xe, unsigned long addr)struct xe_ttm_vram_mgr *vram_mgr;struct xe_vram_region *vr;struct gpu_buddy *mm; +bool policy;int ret; vr =xe_ttm_vram_addr_to_region (xe, addr);
@@ -745,6 +747,14 @@ int xe_ttm_vram_handle_addr_fault(struct xe_device *xe, unsigned long addr)/* TODO: Check if we already processed faulted address, and if yes return -EEXIST */ + policy =xe_configfs_get_bad_page_reservation (to_pci_dev (xe->drm.dev)); +if (!policy) { +drm_err (&xe->drm,"0x%lx is reported as corrupted address by HW\n" , + addr); +/* Let RAS report to FW to drop addr from SRAM queue */ +return -EOPNOTSUPP ; + } +/* Reserve page at address */ ret =xe_ttm_vram_reserve_page_at_addr (xe, addr, vram_mgr, mm);return ret;
Signature · 1 line
2.52.0
Matthew Auld
maintainerRe: [RFC PATCH V7 09/10] drm/xe/configfs: Add vram bad page reservation policy
Quoted 6 lines · Tejas Upadhyay
The interface enables setting the policy for how bad pages are handled in VRAM. This is crucial for maintaining system stability in scenarios where VRAM degradation occurs.
By default policy will be "reserve", which can be changed to "logging" only.
Might be good give the motivation/why somewhere? At at glance, it's not obvious why we would want logging only.
For example, you set reserve=false, and then probe the device. Would that not skip offlining the known bad pages even for the nice case during probe? Or does this only impact the "runtime" offlining once we make it past probe? But if that is the case then that should be documented clearly, I think.
Quoted 201 lines
v3:
- All FW communication moved under RAS
v2:
- Add CRI check and rebaseSigned-off-by Tejas Upadhyay <[email protected]>--- drivers/gpu/drm/xe/xe_configfs.c | 64 +++++++++++++++++++++++++++- drivers/gpu/drm/xe/xe_configfs.h | 2 + drivers/gpu/drm/xe/xe_ttm_vram_mgr.c | 10 +++++ 3 files changed, 75 insertions(+), 1 deletion(-)drivers/gpu/drm/xe/xe_configfs.c +2
-1@@ -61,7 +61,8 @@* ├── survivability_mode * ├── gt_types_allowed * ├── engines_allowed -* └── enable_psmi +* ├── enable_psmi +* └── bad_page_reservation * * After configuring the attributes as per next section, the device can be
* probed with::drivers/gpu/drm/xe/xe_configfs.c +61
-0@@ -159,6 +160,16 @@* * This attribute can only be set before binding to the device. * +* Bad pages reservation: +* --------------------- +* +* Disable vram bad pages reservation, instead just report it in dmesg. +* Example to disable it:: +* +* # echo 0 > /sys/kernel/config/xe/0000:03:00.0/bad_page_reservation +* +* This attribute can only be set before binding to the device. +* * Context restore BB * ------------------ * @@ -262,6 +273,7 @@ struct xe_config_group_device {struct wa_bb ctx_restore_mid_bb[XE_ENGINE_CLASS_MAX ];bool survivability_mode;bool enable_psmi; +bool bad_page_reservation;struct {unsigned int max_vfs;bool admin_only_pf;@@ -281,6 +293,7 @@ static const struct xe_config_device device_defaults = { .engines_allowed = U64_MAX, .survivability_mode = false, .enable_psmi = false, + .bad_page_reservation = true, .sriov = { .max_vfs = XE_DEFAULT_MAX_VFS, .admin_only_pf = XE_DEFAULT_ADMIN_ONLY_PF,@@ -575,6 +588,32 @@ static ssize_t enable_psmi_store(struct config_item *item, const char *page, sizreturn len; } +static ssize_t bad_page_reservation_show (struct config_item *item,char *page) +{ +struct xe_config_device *dev =to_xe_config_device (item); + +return sprintf (page,"%d\n" , dev->bad_page_reservation); +} + +static ssize_t bad_page_reservation_store (struct config_item *item,const char *page,size_t len) +{ +struct xe_config_group_device *dev =to_xe_config_group_device (item); +bool val; +int ret; + + ret =kstrtobool (page, &val); +if (ret) +return ret; + +guard (mutex)(&dev->lock); +if (is_bound (dev)) +return -EBUSY ; + + dev->config.bad_page_reservation = val; + +return len; +} +static bool wa_bb_read_advance (bool dereference,char **p,const char *append,size_t len,size_t *max_size)@@ -813,6 +852,7 @@ static ssize_t ctx_restore_post_bb_store(struct config_item *item, CONFIGFS_ATTR(, ctx_restore_mid_bb); CONFIGFS_ATTR(, ctx_restore_post_bb); CONFIGFS_ATTR(, enable_psmi); +CONFIGFS_ATTR(, bad_page_reservation); CONFIGFS_ATTR(, engines_allowed); CONFIGFS_ATTR(, gt_types_allowed); CONFIGFS_ATTR(, survivability_mode);@@ -821,6 +861,7 @@ static struct configfs_attribute *xe_config_device_attrs[] = { &attr_ctx_restore_mid_bb, &attr_ctx_restore_post_bb, &attr_enable_psmi, + &attr_bad_page_reservation, &attr_engines_allowed, &attr_gt_types_allowed, &attr_survivability_mode,@@ -1098,6 +1139,7 @@ static void dump_custom_dev_config(struct pci_dev *pdev,PRI_CUSTOM_ATTR ("%llx" , gt_types_allowed);PRI_CUSTOM_ATTR ("%llx" , engines_allowed);PRI_CUSTOM_ATTR ("%d" , enable_psmi); +PRI_CUSTOM_ATTR ("%d" , bad_page_reservation);PRI_CUSTOM_ATTR ("%d" , survivability_mode);PRI_CUSTOM_ATTR ("%u" , sriov.admin_only_pf);@@ -1225,6 +1267,26 @@ bool xe_configfs_get_psmi_enabled(struct pci_dev *pdev)return ret; } +/** +* xe_configfs_get_bad_page_reservation - get configfs bad_page_reservation setting +* @pdev: pci device +* +* Return: bad_page_reservation setting in configfs +*/ +bool xe_configfs_get_bad_page_reservation (struct pci_dev *pdev) +{ +struct xe_config_group_device *dev =find_xe_config_group_device (pdev); +bool ret; + +if (!dev) +return device_defaults.bad_page_reservation; + + ret = dev->config.bad_page_reservation; +config_group_put (&dev->group); + +return ret; +} +/** * xe_configfs_get_ctx_restore_mid_bb - get configfs ctx_restore_mid_bb setting * @pdev: pci device drivers/gpu/drm/xe/xe_configfs.h +2
-0@@ -23,6 +23,7 @@ bool xe_configfs_primary_gt_allowed(struct pci_dev *pdev);bool xe_configfs_media_gt_allowed (struct pci_dev *pdev);u64 xe_configfs_get_engines_allowed (struct pci_dev *pdev);bool xe_configfs_get_psmi_enabled (struct pci_dev *pdev); +bool xe_configfs_get_bad_page_reservation (struct pci_dev *pdev);u32 xe_configfs_get_ctx_restore_mid_bb (struct pci_dev *pdev,enum xe_engine_class class,const u32 **cs);@@ -42,6 +43,7 @@ static inline bool xe_configfs_primary_gt_allowed(struct pci_dev *pdev) { returnstatic inline bool xe_configfs_media_gt_allowed (struct pci_dev *pdev) {return true; }static inline u64 xe_configfs_get_engines_allowed (struct pci_dev *pdev) {return U64_MAX ; }static inline bool xe_configfs_get_psmi_enabled (struct pci_dev *pdev) {return false; } +static inline bool xe_configfs_get_bad_page_reservation (struct pci_dev *pdev) {return true; }static inline u32 xe_configfs_get_ctx_restore_mid_bb (struct pci_dev *pdev,enum xe_engine_class class,const u32 **cs) {return 0 ; }drivers/gpu/drm/xe/xe_ttm_vram_mgr.c +10
-0@@ -12,6 +12,7 @@#include <drm/ttm/ttm_range_manager.h> #include "xe_bo.h" +#include "xe_configfs.h" #include "xe_device.h" #include "xe_exec_queue.h" #include "xe_lrc.h" @@ -731,6 +732,7 @@ int xe_ttm_vram_handle_addr_fault(struct xe_device *xe, unsigned long addr)struct xe_ttm_vram_mgr *vram_mgr;struct xe_vram_region *vr;struct gpu_buddy *mm; +bool policy;int ret; vr =xe_ttm_vram_addr_to_region (xe, addr);@@ -745,6 +747,14 @@ int xe_ttm_vram_handle_addr_fault(struct xe_device *xe, unsigned long addr)/* TODO: Check if we already processed faulted address, and if yes return -EEXIST */ + policy =xe_configfs_get_bad_page_reservation (to_pci_dev (xe->drm.dev)); +if (!policy) { +drm_err (&xe->drm,"0x%lx is reported as corrupted address by HW\n" , + addr); +/* Let RAS report to FW to drop addr from SRAM queue */ +return -EOPNOTSUPP ; + } +/* Reserve page at address */ ret =xe_ttm_vram_reserve_page_at_addr (xe, addr, vram_mgr, mm);return ret;
RFC PATCH v7 10/10 drm/xe/cri: Add sysfs interface for bad gpu vram pages
In reply to Tejas UpadhyayStarting CRI, Include a sysfs interface designed to expose information about bad VRAM pages—those identified as having hardware faults (e.g., ECC errors). This interface allows userspace tools and administrators to monitor the health of the GPU's local memory and track the status of page retirement.To get details on bad gpu vram pages can be found under /sys/bus/pci/devices/bdf/vram_bad_pages.
Where The format is, pfn : gpu page size : flags
flags:
R: reserved, this gpu page is reserved.
P: pending for reserve, this gpu page is marked as bad, will be reserved
in next window of page_reserve.
F: unable to reserve. this gpu page can’t be reserved due to some reasons.
For example if you read using cat /sys/bus/pci/devices/bdf/vram_bad_pages, max_pages : 100000x00000000 : 0x00001000 : R0x00001234 : 0x00001000 : P
v3:
- Move FW communication in RAS code
v2:
- Add max_pages info as per updated design doc
- Rebase
--- drivers/gpu/drm/xe/xe_device_sysfs.c | 7 ++ drivers/gpu/drm/xe/xe_ttm_vram_mgr.c | 79 ++++++++++++++++++++++ drivers/gpu/drm/xe/xe_ttm_vram_mgr.h | 1 + drivers/gpu/drm/xe/xe_ttm_vram_mgr_types.h | 2 + 4 files changed, 89 insertions(+)
drivers/gpu/drm/xe/xe_device_sysfs.c +7 -0
@@ -8,12 +8,14 @@#include <linux/pci.h> #include <linux/sysfs.h> +#include "xe_configfs.h" #include "xe_device.h" #include "xe_device_sysfs.h" #include "xe_mmio.h" #include "xe_pcode_api.h" #include "xe_pcode.h" #include "xe_pm.h" +#include "xe_ttm_vram_mgr.h" /** * DOC: Xe device sysfs
@@ -267,6 +269,7 @@ static const struct attribute_group auto_link_downgrade_attr_group = {int xe_device_sysfs_init (struct xe_device *xe) {struct device *dev = xe->drm.dev; +bool policy;int ret;if (xe->d3cold.capable) {
@@ -285,5 +288,9 @@ int xe_device_sysfs_init(struct xe_device *xe)return ret; } + policy =xe_configfs_get_bad_page_reservation (to_pci_dev (dev)); +if (xe->info.platform ==XE_CRESCENTISLAND && policy) +xe_ttm_vram_sysfs_init (xe); +return 0 ; }
drivers/gpu/drm/xe/xe_ttm_vram_mgr.c +79 -0
82 lines not on this page · raw message
drivers/gpu/drm/xe/xe_ttm_vram_mgr.h +1 -0
@@ -32,6 +32,7 @@ void xe_ttm_vram_get_used(struct ttm_resource_manager *man, u64 *used, u64 *used_visible);int xe_ttm_vram_handle_addr_fault (struct xe_device *xe,unsigned long addr); +int xe_ttm_vram_sysfs_init (struct xe_device *xe);static inline struct xe_ttm_vram_mgr_resource *to_xe_ttm_vram_mgr_resource (struct ttm_resource *res) {
drivers/gpu/drm/xe/xe_ttm_vram_mgr_types.h +2 -0
@@ -39,6 +39,8 @@ struct xe_ttm_vram_mgr {u32 mem_type;/** @offline_mode: debugfs hook for setting page offline mode */ u64 offline_mode; +/** @max_pages: max pages that can be in offline queue retrieved from FW */ +u16 max_pages; };/**
Signature · 1 line
2.52.0
Matthew Auld
maintainerRe: [RFC PATCH V7 10/10] drm/xe/cri: Add sysfs interface for bad gpu vram pages
Quoted 8 lines · Tejas Upadhyay
Starting CRI, Include a sysfs interface designed to expose information about bad VRAM pages—those identified as having hardware faults (e.g., ECC errors). This interface allows userspace tools and administrators to monitor the health of the GPU's local memory and track the status of page retirement.To get details on bad gpu vram pages can be found under
/sys/bus/pci/devices/bdf/vram_bad_pages.Where The format is, pfn : gpu page size : flags
With "gpu page size" this is really just the min block size? gpu-page-size is normally interpreted as GTT page size, which is a different thing. But is that not always 4K here? Since that is the granularity of the addr reservation? Is it useful to print that? Is knowing that pfn x is offlined not enough?
Also what is the story if you have multiple VRAM instances here? There is only one vram_bad_pages file? Would this treat VRAM as one giant unified thing?
Quoted 175 lines
flags:
R: reserved, this gpu page is reserved.
P: pending for reserve, this gpu page is marked as bad, will be reserved
in next window ofpage_reserve.
F: unable to reserve. this gpu page can’t be reserved due to some reasons.For example if you read using cat
/sys/bus/pci/devices/bdf/vram_bad_pages,max_pages: 100000x00000000:0x00001000: R0x00001234:0x00001000: Pv3:
- Move FW communication in RAS code
v2:
- Addmax_pagesinfo as per updated design doc
- RebaseSigned-off-by Tejas Upadhyay <[email protected]>--- drivers/gpu/drm/xe/xe_device_sysfs.c | 7 ++ drivers/gpu/drm/xe/xe_ttm_vram_mgr.c | 79 ++++++++++++++++++++++ drivers/gpu/drm/xe/xe_ttm_vram_mgr.h | 1 + drivers/gpu/drm/xe/xe_ttm_vram_mgr_types.h | 2 + 4 files changed, 89 insertions(+)drivers/gpu/drm/xe/xe_device_sysfs.c +7
-0@@ -8,12 +8,14 @@#include <linux/pci.h> #include <linux/sysfs.h> +#include "xe_configfs.h" #include "xe_device.h" #include "xe_device_sysfs.h" #include "xe_mmio.h" #include "xe_pcode_api.h" #include "xe_pcode.h" #include "xe_pm.h" +#include "xe_ttm_vram_mgr.h" /** * DOC: Xe device sysfs @@ -267,6 +269,7 @@ static const struct attribute_group auto_link_downgrade_attr_group = {int xe_device_sysfs_init (struct xe_device *xe) {struct device *dev = xe->drm.dev; +bool policy;int ret;if (xe->d3cold.capable) {@@ -285,5 +288,9 @@ int xe_device_sysfs_init(struct xe_device *xe)return ret; } + policy =xe_configfs_get_bad_page_reservation (to_pci_dev (dev)); +if (xe->info.platform ==XE_CRESCENTISLAND && policy) +xe_ttm_vram_sysfs_init (xe); +return 0 ; }drivers/gpu/drm/xe/xe_ttm_vram_mgr.c +79
-0@@ -760,3 +760,82 @@ int xe_ttm_vram_handle_addr_fault(struct xe_device *xe, unsigned long addr)return ret; }EXPORT_SYMBOL (xe_ttm_vram_handle_addr_fault); + +static void xe_ttm_vram_dump_bad_pages_info (char *buf,struct xe_ttm_vram_mgr *mgr) +{ +const unsigned int element_size =sizeof ("0xabcdabcd : 0x12345678 : R\n" ) -1 ; +const unsigned int maxpage_size =sizeof ("max_pages: 10000\n" ) -1 ; +struct xe_ttm_vram_offline_resource *pos, *n; +struct gpu_buddy_block *block; +ssize_t s =0 ; + +mutex_lock (&mgr->lock); + s +=scnprintf (&buf[s], maxpage_size +1 ,"max_pages: %d\n" , mgr->max_pages); +list_for_each_entry_safe (pos, n, &mgr->offlined_pages, offlined_link) { + block =list_first_entry (&pos->blocks, + struct gpu_buddy_block, + link); + s +=scnprintf (&buf[s], element_size +1 , +"0x%08llx : 0x%08llx : %1s\n" , +gpu_buddy_block_offset (block) >>PAGE_SHIFT , +gpu_buddy_block_size (&mgr->mm, block), +"R" ); + } +list_for_each_entry_safe (pos, n, &mgr->queued_pages, queued_link) { + block =list_first_entry (&pos->blocks, + struct gpu_buddy_block, + link); + s +=scnprintf (&buf[s], element_size +1 , +"0x%08llx : 0x%08llx : %1s\n" , +gpu_buddy_block_offset (block) >>PAGE_SHIFT , +gpu_buddy_block_size (&mgr->mm, block), + pos->status ?"P" :"F" ); + } +mutex_unlock (&mgr->lock); +} + +static ssize_t vram_bad_pages_show (struct device *dev,struct device_attribute *attr,char *buf) +{ +struct pci_dev *pdev =to_pci_dev (dev); +struct xe_device *xe =pdev_to_xe_device (pdev); +struct ttm_resource_manager *man; +struct xe_ttm_vram_mgr *mgr; + + man =ttm_manager_type (&xe->ttm,XE_PL_VRAM0 ); +if (man) { + mgr =to_xe_ttm_vram_mgr (man); +xe_ttm_vram_dump_bad_pages_info (buf, mgr); + } + +return sysfs_emit (buf,"%s\n" , buf); +} +static DEVICE_ATTR_RO (vram_bad_pages); + +static void xe_ttm_vram_sysfs_fini (void *arg) +{ +struct xe_device *xe = arg; + +device_remove_file (xe->drm.dev, &dev_attr_vram_bad_pages); +} + +/** +* xe_ttm_vram_sysfs_init - Initialize vram sysfs component +* @tile: Xe Tile object +* +* It needs to be initialized after the main tile component is ready +* +* Returns: 0 on success, negative error code on error. +*/ +int xe_ttm_vram_sysfs_init (struct xe_device *xe) +{ +int err; + + err =device_create_file (xe->drm.dev, &dev_attr_vram_bad_pages); +if (err) { +dev_err (xe->drm.dev,"Failed to create vram_bad_pages sysfs file: %d\n" , err); +return 0 ; + } + +return devm_add_action_or_reset (xe->drm.dev, xe_ttm_vram_sysfs_fini, xe); +} +EXPORT_SYMBOL (xe_ttm_vram_sysfs_init);drivers/gpu/drm/xe/xe_ttm_vram_mgr.h +1
-0@@ -32,6 +32,7 @@ void xe_ttm_vram_get_used(struct ttm_resource_manager *man, u64 *used, u64 *used_visible);int xe_ttm_vram_handle_addr_fault (struct xe_device *xe,unsigned long addr); +int xe_ttm_vram_sysfs_init (struct xe_device *xe);static inline struct xe_ttm_vram_mgr_resource *to_xe_ttm_vram_mgr_resource (struct ttm_resource *res) {drivers/gpu/drm/xe/xe_ttm_vram_mgr_types.h +2
-0@@ -39,6 +39,8 @@ struct xe_ttm_vram_mgr {u32 mem_type;/** @offline_mode: debugfs hook for setting page offline mode */ u64 offline_mode; +/** @max_pages: max pages that can be in offline queue retrieved from FW */ +u16 max_pages; };/**
RE: [RFC PATCH V7 10/10] drm/xe/cri: Add sysfs interface for bad gpu vram pages
Quoted 21 lines
-----Original Message-----
From: Auld, Matthew <[email protected]>
Sent: 30 April 2026 19:23
To: Upadhyay, Tejas <[email protected]>; intel- [email protected]
Cc: Brost, Matthew <[email protected]>; [email protected]; Ghimiray, Himal Prasad <[email protected]>
Subject: Re: [RFC PATCH V7 10/10] drm/xe/cri: Add sysfs interface for bad gpu vram pagesQuoted 7 lines · Tejas Upadhyay
Starting CRI, Include a sysfs interface designed to expose information about bad VRAM pages—those identified as having hardware faults (e.g., ECC errors). This interface allows userspace tools and administrators to monitor the health of the GPU's local memory and track the status of page retirement.To get details on bad gpu vram pages can be found under
/sys/bus/pci/devices/bdf/vram_bad_pages.
Quoted 7 lines
Quoted 1 line
Where The format is, pfn : gpu page size : flags
With "gpu page size" this is really just the min block size? gpu-page-size is normally interpreted as GTT page size, which is a different thing. But is that not always 4K here? Since that is the granularity of the addr reservation? Is it useful to print that? Is knowing that pfn x is offlined not enough?
I think you are right, its always 4K. But I will check with Aravind as for what reason design doc is saying format to be "pfn : gpu page size : flags". Adding him in cc.
Quoted 3 lines
Also what is the story if you have multiple VRAM instances here? There is only one
vram_bad_pagesfile? Would this treat VRAM as one giant unified thing?
This is targeted for CRI, thus single VRAM is in reference right now. We can extend in future if need be or may be it will single huge thing only in future for multiple vram.
Tejas
Quoted 188 lines
Quoted 20 lines
flags:
R: reserved, this gpu page is reserved.
P: pending for reserve, this gpu page is marked as bad, will be reserved
in next window ofpage_reserve.
F: unable to reserve. this gpu page can’t be reserved due to some reasons.For example if you read using cat
/sys/bus/pci/devices/bdf/vram_bad_pages,max_pages: 100000x00000000:0x00001000: R0x00001234:0x00001000: Pv3:
- Move FW communication in RAS code
v2:
- Addmax_pagesinfo as per updated design doc
- RebaseQuoted 4 lines
Signed-off-by Tejas Upadhyay <[email protected]>--- drivers/gpu/drm/xe/xe_device_sysfs.c | 7 ++ drivers/gpu/drm/xe/xe_ttm_vram_mgr.c | 79++++++++++++++++++++++
drivers/gpu/drm/xe/xe_device_sysfs.c static const struct attribute_group
Quoted 25 lines
drivers/gpu/drm/xe/xe_ttm_vram_mgr.h | 1 + drivers/gpu/drm/xe/xe_ttm_vram_mgr_types.h | 2 + 4 files changed, 89 insertions(+)drivers/gpu/drm/xe/xe_device_sysfs.c +2
-0@@ -8,12 +8,14 @@#include <linux/pci.h> #include <linux/sysfs.h> +#include "xe_configfs.h" #include "xe_device.h" #include "xe_device_sysfs.h" #include "xe_mmio.h" #include "xe_pcode_api.h" #include "xe_pcode.h" #include "xe_pm.h" +#include "xe_ttm_vram_mgr.h" /** * DOC: Xe device sysfs @@ -267,6 +269,7 @@ static const struct attribute_group
auto_link_downgrade_attr_group= {drivers/gpu/drm/xe/xe_ttm_vram_mgr.c int xe_ttm_vram_handle_addr_fault(struct
Quoted 23 lines
drivers/gpu/drm/xe/xe_device_sysfs.c +5
-0int xe_device_sysfs_init (struct xe_device *xe) {struct device *dev = xe->drm.dev; +bool policy;int ret;if (xe->d3cold.capable) {@@ -285,5 +288,9 @@ int xe_device_sysfs_init(struct xe_device *xe)return ret; } + policy =xe_configfs_get_bad_page_reservation (to_pci_dev (dev)); +if (xe->info.platform ==XE_CRESCENTISLAND && policy) +xe_ttm_vram_sysfs_init (xe); +return 0 ; }drivers/gpu/drm/xe/xe_ttm_vram_mgr.c +0
-0@@ -760,3 +760,82 @@ int xe_ttm_vram_handle_addr_fault(struct
xe_device*xe, unsigned long addr)drivers/gpu/drm/xe/xe_ttm_vram_mgr.c int xe_ttm_vram_handle_addr_fault(struct xe_device *xe, unsigned long addr)
Quoted 7 lines
drivers/gpu/drm/xe/xe_ttm_vram_mgr.c +4
-0return ret; }EXPORT_SYMBOL (xe_ttm_vram_handle_addr_fault); + +static void xe_ttm_vram_dump_bad_pages_info (char *buf,struct +xe_ttm_vram_mgr *mgr) { + const unsigned int element_size = sizeof("0xabcdabcd : 0x12345678 :R\n") - 1;
drivers/gpu/drm/xe/xe_ttm_vram_mgr.c
Quoted 9 lines
drivers/gpu/drm/xe/xe_ttm_vram_mgr.c +7
-0+const unsigned int maxpage_size =sizeof ("max_pages: 10000\n" ) -1 ; +struct xe_ttm_vram_offline_resource *pos, *n; +struct gpu_buddy_block *block; +ssize_t s =0 ; + +mutex_lock (&mgr->lock); + s += scnprintf(&buf[s], maxpage_size + 1, "max_pages: %d\n", mgr-
max_pages);drivers/gpu/drm/xe/xe_ttm_vram_mgr.c +1
-0+list_for_each_entry_safe (pos, n, &mgr->offlined_pages, offlined_link){
drivers/gpu/drm/xe/xe_ttm_vram_mgr.c int xe_ttm_vram_handle_addr_fault(struct xe_device *xe, unsigned long addr)
Quoted 10 lines
drivers/gpu/drm/xe/xe_ttm_vram_mgr.c +10
-0+ block =list_first_entry (&pos->blocks, + struct gpu_buddy_block, + link); + s +=scnprintf (&buf[s], element_size +1 , +"0x%08llx : 0x%08llx : %1s\n" , +gpu_buddy_block_offset (block) >>PAGE_SHIFT , +gpu_buddy_block_size (&mgr->mm, block), +"R" ); + } +list_for_each_entry_safe (pos, n, &mgr->queued_pages, queued_link){
Quoted 37 lines
drivers/gpu/drm/xe/xe_ttm_vram_mgr.c +37
-0+ block =list_first_entry (&pos->blocks, + struct gpu_buddy_block, + link); + s +=scnprintf (&buf[s], element_size +1 , +"0x%08llx : 0x%08llx : %1s\n" , +gpu_buddy_block_offset (block) >>PAGE_SHIFT , +gpu_buddy_block_size (&mgr->mm, block), + pos->status ?"P" :"F" ); + } +mutex_unlock (&mgr->lock); +} + +static ssize_t vram_bad_pages_show (struct device *dev,struct +device_attribute *attr,char *buf) { +struct pci_dev *pdev =to_pci_dev (dev); +struct xe_device *xe =pdev_to_xe_device (pdev); +struct ttm_resource_manager *man; +struct xe_ttm_vram_mgr *mgr; + + man =ttm_manager_type (&xe->ttm,XE_PL_VRAM0 ); +if (man) { + mgr =to_xe_ttm_vram_mgr (man); +xe_ttm_vram_dump_bad_pages_info (buf, mgr); + } + +return sysfs_emit (buf,"%s\n" , buf); }static +DEVICE_ATTR_RO (vram_bad_pages); + +static void xe_ttm_vram_sysfs_fini (void *arg) { +struct xe_device *xe = arg; + +device_remove_file (xe->drm.dev, &dev_attr_vram_bad_pages); } + +/** +* xe_ttm_vram_sysfs_init - Initialize vram sysfs component +* @tile: Xe Tile object +* drivers/gpu/drm/xe/xe_ttm_vram_mgr.c int xe_ttm_vram_handle_addr_fault(struct xe_device *xe, unsigned long addr)
Quoted 10 lines
drivers/gpu/drm/xe/xe_ttm_vram_mgr.c +10
-0+* It needs to be initialized after the main tile component is ready +* +* Returns: 0 on success, negative error code on error. +*/ +int xe_ttm_vram_sysfs_init (struct xe_device *xe) { +int err; + + err =device_create_file (xe->drm.dev, &dev_attr_vram_bad_pages); +if (err) { + dev_err(xe->drm.dev, "Failed to create vram_bad_pages sysfsfile: %d\n", err);
drivers/gpu/drm/xe/xe_ttm_vram_mgr.c int xe_ttm_vram_handle_addr_fault(struct xe_device *xe, unsigned long addr)
Quoted 4 lines
drivers/gpu/drm/xe/xe_ttm_vram_mgr.c +4
-0+return 0 ; + } + + return devm_add_action_or_reset(xe->drm.dev,
xe_ttm_vram_sysfs_fini,drivers/gpu/drm/xe/xe_ttm_vram_mgr.h void xe_ttm_vram_get_used(struct
Quoted 7 lines
drivers/gpu/drm/xe/xe_ttm_vram_mgr.c +1
-0+xe); } EXPORT_SYMBOL(xe_ttm_vram_sysfs_init);drivers/gpu/drm/xe/xe_ttm_vram_mgr.h +0
-0@@ -32,6 +32,7 @@ void xe_ttm_vram_get_used(struct
ttm_resource_manager*man,Quoted 14 lines
drivers/gpu/drm/xe/xe_ttm_vram_mgr.h +1
-0u64 *used, u64 *used_visible);int xe_ttm_vram_handle_addr_fault (struct xe_device *xe,unsigned long addr); +int xe_ttm_vram_sysfs_init (struct xe_device *xe);static inline struct xe_ttm_vram_mgr_resource *to_xe_ttm_vram_mgr_resource (struct ttm_resource *res){drivers/gpu/drm/xe/xe_ttm_vram_mgr_types.h +0
-0drivers/gpu/drm/xe/xe_ttm_vram_mgr_types.h struct xe_ttm_vram_mgr { · line 42
Quoted 4 lines
drivers/gpu/drm/xe/xe_ttm_vram_mgr_types.h +1
-0@@ -39,6 +39,8 @@ struct xe_ttm_vram_mgr {u32 mem_type;/** @offline_mode: debugfs hook for setting page offline mode */ u64 offline_mode; +/** @max_pages: max pages that can be in offline queue retrieved from FW */
Quoted 4 lines
drivers/gpu/drm/xe/xe_ttm_vram_mgr_types.h +1
-0+u16 max_pages; };/**
Matthew Auld
maintainerRe: [RFC PATCH V7 10/10] drm/xe/cri: Add sysfs interface for bad gpu vram pages
Quoted 29 lines · Tejas
Quoted 21 lines
-----Original Message-----
From: Auld, Matthew <[email protected]>
Sent: 30 April 2026 19:23
To: Upadhyay, Tejas <[email protected]>; intel- [email protected]
Cc: Brost, Matthew <[email protected]>; [email protected]; Ghimiray, Himal Prasad <[email protected]>
Subject: Re: [RFC PATCH V7 10/10] drm/xe/cri: Add sysfs interface for bad gpu vram pagesQuoted 7 lines · Tejas Upadhyay
Starting CRI, Include a sysfs interface designed to expose information about bad VRAM pages—those identified as having hardware faults (e.g., ECC errors). This interface allows userspace tools and administrators to monitor the health of the GPU's local memory and track the status of page retirement.To get details on bad gpu vram pages can be found under
/sys/bus/pci/devices/bdf/vram_bad_pages.Quoted 7 lines
Quoted 1 line
Where The format is, pfn : gpu page size : flags
With "gpu page size" this is really just the min block size? gpu-page-size is normally interpreted as GTT page size, which is a different thing. But is that not always 4K here? Since that is the granularity of the addr reservation? Is it useful to print that? Is knowing that pfn x is offlined not enough?
Quoted 7 lines
I think you are right, its always 4K. But I will check with Aravind as for what reason design doc is saying format to be "pfn : gpu page size : flags". Adding him in cc.
Quoted 3 lines
Also what is the story if you have multiple VRAM instances here? There is only one
vram_bad_pagesfile? Would this treat VRAM as one giant unified thing?This is targeted for CRI, thus single VRAM is in reference right now. We can extend in future if need be or may be it will single huge thing only in future for multiple vram.
Since this is uapi, I think would be good to be forward looking and make sure we don't need to complety re-design if now asked to support multiple VRAM instances. I think one flat unfied thing listing all the pfns, should work, but I guess depends on the consumer of this and how they are going to use this.
Quoted 191 lines
Tejas
Quoted 188 lines
Quoted 20 lines
flags:
R: reserved, this gpu page is reserved.
P: pending for reserve, this gpu page is marked as bad, will be reserved
in next window ofpage_reserve.
F: unable to reserve. this gpu page can’t be reserved due to some reasons.For example if you read using cat
/sys/bus/pci/devices/bdf/vram_bad_pages,max_pages: 100000x00000000:0x00001000: R0x00001234:0x00001000: Pv3:
- Move FW communication in RAS code
v2:
- Addmax_pagesinfo as per updated design doc
- RebaseQuoted 4 lines
Signed-off-by Tejas Upadhyay <[email protected]>--- drivers/gpu/drm/xe/xe_device_sysfs.c | 7 ++ drivers/gpu/drm/xe/xe_ttm_vram_mgr.c | 79++++++++++++++++++++++
drivers/gpu/drm/xe/xe_device_sysfs.c static const struct attribute_group
Quoted 25 lines
drivers/gpu/drm/xe/xe_ttm_vram_mgr.h | 1 + drivers/gpu/drm/xe/xe_ttm_vram_mgr_types.h | 2 + 4 files changed, 89 insertions(+)drivers/gpu/drm/xe/xe_device_sysfs.c +2
-0@@ -8,12 +8,14 @@#include <linux/pci.h> #include <linux/sysfs.h> +#include "xe_configfs.h" #include "xe_device.h" #include "xe_device_sysfs.h" #include "xe_mmio.h" #include "xe_pcode_api.h" #include "xe_pcode.h" #include "xe_pm.h" +#include "xe_ttm_vram_mgr.h" /** * DOC: Xe device sysfs @@ -267,6 +269,7 @@ static const struct attribute_group
auto_link_downgrade_attr_group= {drivers/gpu/drm/xe/xe_ttm_vram_mgr.c int xe_ttm_vram_handle_addr_fault(struct
Quoted 23 lines
drivers/gpu/drm/xe/xe_device_sysfs.c +5
-0int xe_device_sysfs_init (struct xe_device *xe) {struct device *dev = xe->drm.dev; +bool policy;int ret;if (xe->d3cold.capable) {@@ -285,5 +288,9 @@ int xe_device_sysfs_init(struct xe_device *xe)return ret; } + policy =xe_configfs_get_bad_page_reservation (to_pci_dev (dev)); +if (xe->info.platform ==XE_CRESCENTISLAND && policy) +xe_ttm_vram_sysfs_init (xe); +return 0 ; }drivers/gpu/drm/xe/xe_ttm_vram_mgr.c +0
-0@@ -760,3 +760,82 @@ int xe_ttm_vram_handle_addr_fault(struct
xe_device*xe, unsigned long addr)drivers/gpu/drm/xe/xe_ttm_vram_mgr.c int xe_ttm_vram_handle_addr_fault(struct xe_device *xe, unsigned long addr)
Quoted 7 lines
drivers/gpu/drm/xe/xe_ttm_vram_mgr.c +4
-0return ret; }EXPORT_SYMBOL (xe_ttm_vram_handle_addr_fault); + +static void xe_ttm_vram_dump_bad_pages_info (char *buf,struct +xe_ttm_vram_mgr *mgr) { + const unsigned int element_size = sizeof("0xabcdabcd : 0x12345678 :R\n") - 1;
drivers/gpu/drm/xe/xe_ttm_vram_mgr.c
Quoted 9 lines
drivers/gpu/drm/xe/xe_ttm_vram_mgr.c +7
-0+const unsigned int maxpage_size =sizeof ("max_pages: 10000\n" ) -1 ; +struct xe_ttm_vram_offline_resource *pos, *n; +struct gpu_buddy_block *block; +ssize_t s =0 ; + +mutex_lock (&mgr->lock); + s += scnprintf(&buf[s], maxpage_size + 1, "max_pages: %d\n", mgr-
max_pages);drivers/gpu/drm/xe/xe_ttm_vram_mgr.c +1
-0+list_for_each_entry_safe (pos, n, &mgr->offlined_pages, offlined_link){
drivers/gpu/drm/xe/xe_ttm_vram_mgr.c int xe_ttm_vram_handle_addr_fault(struct xe_device *xe, unsigned long addr)
Quoted 10 lines
drivers/gpu/drm/xe/xe_ttm_vram_mgr.c +10
-0+ block =list_first_entry (&pos->blocks, + struct gpu_buddy_block, + link); + s +=scnprintf (&buf[s], element_size +1 , +"0x%08llx : 0x%08llx : %1s\n" , +gpu_buddy_block_offset (block) >>PAGE_SHIFT , +gpu_buddy_block_size (&mgr->mm, block), +"R" ); + } +list_for_each_entry_safe (pos, n, &mgr->queued_pages, queued_link){
Quoted 37 lines
drivers/gpu/drm/xe/xe_ttm_vram_mgr.c +37
-0+ block =list_first_entry (&pos->blocks, + struct gpu_buddy_block, + link); + s +=scnprintf (&buf[s], element_size +1 , +"0x%08llx : 0x%08llx : %1s\n" , +gpu_buddy_block_offset (block) >>PAGE_SHIFT , +gpu_buddy_block_size (&mgr->mm, block), + pos->status ?"P" :"F" ); + } +mutex_unlock (&mgr->lock); +} + +static ssize_t vram_bad_pages_show (struct device *dev,struct +device_attribute *attr,char *buf) { +struct pci_dev *pdev =to_pci_dev (dev); +struct xe_device *xe =pdev_to_xe_device (pdev); +struct ttm_resource_manager *man; +struct xe_ttm_vram_mgr *mgr; + + man =ttm_manager_type (&xe->ttm,XE_PL_VRAM0 ); +if (man) { + mgr =to_xe_ttm_vram_mgr (man); +xe_ttm_vram_dump_bad_pages_info (buf, mgr); + } + +return sysfs_emit (buf,"%s\n" , buf); }static +DEVICE_ATTR_RO (vram_bad_pages); + +static void xe_ttm_vram_sysfs_fini (void *arg) { +struct xe_device *xe = arg; + +device_remove_file (xe->drm.dev, &dev_attr_vram_bad_pages); } + +/** +* xe_ttm_vram_sysfs_init - Initialize vram sysfs component +* @tile: Xe Tile object +* drivers/gpu/drm/xe/xe_ttm_vram_mgr.c int xe_ttm_vram_handle_addr_fault(struct xe_device *xe, unsigned long addr)
Quoted 10 lines
drivers/gpu/drm/xe/xe_ttm_vram_mgr.c +10
-0+* It needs to be initialized after the main tile component is ready +* +* Returns: 0 on success, negative error code on error. +*/ +int xe_ttm_vram_sysfs_init (struct xe_device *xe) { +int err; + + err =device_create_file (xe->drm.dev, &dev_attr_vram_bad_pages); +if (err) { + dev_err(xe->drm.dev, "Failed to create vram_bad_pages sysfsfile: %d\n", err);
drivers/gpu/drm/xe/xe_ttm_vram_mgr.c int xe_ttm_vram_handle_addr_fault(struct xe_device *xe, unsigned long addr)
Quoted 4 lines
drivers/gpu/drm/xe/xe_ttm_vram_mgr.c +4
-0+return 0 ; + } + + return devm_add_action_or_reset(xe->drm.dev,
xe_ttm_vram_sysfs_fini,drivers/gpu/drm/xe/xe_ttm_vram_mgr.h void xe_ttm_vram_get_used(struct
Quoted 7 lines
drivers/gpu/drm/xe/xe_ttm_vram_mgr.c +1
-0+xe); } EXPORT_SYMBOL(xe_ttm_vram_sysfs_init);drivers/gpu/drm/xe/xe_ttm_vram_mgr.h +0
-0@@ -32,6 +32,7 @@ void xe_ttm_vram_get_used(struct
ttm_resource_manager*man,Quoted 14 lines
drivers/gpu/drm/xe/xe_ttm_vram_mgr.h +1
-0u64 *used, u64 *used_visible);int xe_ttm_vram_handle_addr_fault (struct xe_device *xe,unsigned long addr); +int xe_ttm_vram_sysfs_init (struct xe_device *xe);static inline struct xe_ttm_vram_mgr_resource *to_xe_ttm_vram_mgr_resource (struct ttm_resource *res){drivers/gpu/drm/xe/xe_ttm_vram_mgr_types.h +0
-0drivers/gpu/drm/xe/xe_ttm_vram_mgr_types.h struct xe_ttm_vram_mgr { · line 42
Quoted 4 lines
drivers/gpu/drm/xe/xe_ttm_vram_mgr_types.h +1
-0@@ -39,6 +39,8 @@ struct xe_ttm_vram_mgr {u32 mem_type;/** @offline_mode: debugfs hook for setting page offline mode */ u64 offline_mode; +/** @max_pages: max pages that can be in offline queue retrieved from FW */
Quoted 4 lines
drivers/gpu/drm/xe/xe_ttm_vram_mgr_types.h +1
-0+u16 max_pages; };/**
RE: [RFC PATCH V7 10/10] drm/xe/cri: Add sysfs interface for bad gpu vram pages
Quoted 52 lines
-----Original Message-----
From: Auld, Matthew <[email protected]>
Sent: 05 May 2026 14:14
To: Upadhyay, Tejas <[email protected]>; intel- [email protected]
Cc: Brost, Matthew <[email protected]>; [email protected]; Ghimiray, Himal Prasad <[email protected]>
Subject: Re: [RFC PATCH V7 10/10] drm/xe/cri: Add sysfs interface for bad gpu vram pagesQuoted 23 lines · Tejas
Quoted 23 lines
-----Original Message-----
From: Auld, Matthew <[email protected]>
Sent: 30 April 2026 19:23
To: Upadhyay, Tejas <[email protected]>; intel- [email protected]
Cc: Brost, Matthew <[email protected]>; [email protected]; Ghimiray, Himal Prasad <[email protected]>
Subject: Re: [RFC PATCH V7 10/10] drm/xe/cri: Add sysfs interface for bad gpu vram pagesQuoted 8 lines · Tejas Upadhyay
Starting CRI, Include a sysfs interface designed to expose information about bad VRAM pages—those identified as having hardware faults (e.g., ECC errors). This interface allows userspace tools and administrators to monitor the health of the GPU's local memory and track the status of page retirement.To get details on bad gpu vram pages can be found under
/sys/bus/pci/devices/bdf/vram_bad_pages.Where The format is, pfn : gpu page size : flags
Quoted 7 lines
Quoted 5 lines
With "gpu page size" this is really just the min block size? gpu-page-size is normally interpreted as GTT page size, which is a different thing. But is that not always 4K here? Since that is the granularity of the addr reservation? Is it useful to print that? Is knowing that pfn x is offlined not enough?
I think you are right, its always 4K. But I will check with Aravind as for what
reason design doc is saying format to be "pfn : gpu page size : flags". Adding him in cc.
Quoted 4 lines
Quoted 4 lines
Also what is the story if you have multiple VRAM instances here? There is only one
vram_bad_pagesfile? Would this treat VRAM as one giantunified thing?
Quoted 1 line
Quoted 8 lines
Quoted 1 line
This is targeted for CRI, thus single VRAM is in reference right now. We can
extend in future if need be or may be it will single huge thing only in future for multiple vram.
Since this is uapi, I think would be good to be forward looking and make sure we don't need to complety re-design if now asked to support multiple VRAM instances. I think one flat unfied thing listing all the pfns, should work, but I guess depends on the consumer of this and how they are going to use this.
Right, lets double check with Aravind.
Tejas
Quoted 194 lines
Quoted 194 lines
Tejas
Quoted 190 lines
Quoted 20 lines
flags:
R: reserved, this gpu page is reserved.
P: pending for reserve, this gpu page is marked as bad, will be reserved
in next window ofpage_reserve.
F: unable to reserve. this gpu page can’t be reserved due to some reasons.For example if you read using cat
/sys/bus/pci/devices/bdf/vram_bad_pages,max_pages: 100000x00000000:0x00001000: R0x00001234:0x00001000: Pv3:
- Move FW communication in RAS code
v2:
- Addmax_pagesinfo as per updated design doc
- RebaseQuoted 4 lines
Signed-off-by Tejas Upadhyay <[email protected]>--- drivers/gpu/drm/xe/xe_device_sysfs.c | 7 ++ drivers/gpu/drm/xe/xe_ttm_vram_mgr.c | 79++++++++++++++++++++++
drivers/gpu/drm/xe/xe_device_sysfs.c static const struct attribute_group
Quoted 25 lines
drivers/gpu/drm/xe/xe_ttm_vram_mgr.h | 1 + drivers/gpu/drm/xe/xe_ttm_vram_mgr_types.h | 2 + 4 files changed, 89 insertions(+)drivers/gpu/drm/xe/xe_device_sysfs.c +2
-0@@ -8,12 +8,14 @@#include <linux/pci.h> #include <linux/sysfs.h> +#include "xe_configfs.h" #include "xe_device.h" #include "xe_device_sysfs.h" #include "xe_mmio.h" #include "xe_pcode_api.h" #include "xe_pcode.h" #include "xe_pm.h" +#include "xe_ttm_vram_mgr.h" /** * DOC: Xe device sysfs @@ -267,6 +269,7 @@ static const struct attribute_group
auto_link_downgrade_attr_group= {drivers/gpu/drm/xe/xe_ttm_vram_mgr.c int xe_ttm_vram_handle_addr_fault(struct
Quoted 23 lines
drivers/gpu/drm/xe/xe_device_sysfs.c +5
-0int xe_device_sysfs_init (struct xe_device *xe) {struct device *dev = xe->drm.dev; +bool policy;int ret;if (xe->d3cold.capable) {@@ -285,5 +288,9 @@ int xe_device_sysfs_init(struct xe_device *xe)return ret; } + policy =xe_configfs_get_bad_page_reservation (to_pci_dev (dev)); +if (xe->info.platform ==XE_CRESCENTISLAND && policy) +xe_ttm_vram_sysfs_init (xe); +return 0 ; }drivers/gpu/drm/xe/xe_ttm_vram_mgr.c +0
-0@@ -760,3 +760,82 @@ int xe_ttm_vram_handle_addr_fault(struct
xe_device*xe, unsigned long addr)drivers/gpu/drm/xe/xe_ttm_vram_mgr.c int xe_ttm_vram_handle_addr_fault(struct xe_device *xe, unsigned long addr)
Quoted 7 lines
drivers/gpu/drm/xe/xe_ttm_vram_mgr.c +4
-0return ret; }EXPORT_SYMBOL (xe_ttm_vram_handle_addr_fault); + +static void xe_ttm_vram_dump_bad_pages_info (char *buf,struct +xe_ttm_vram_mgr *mgr) { + const unsigned int element_size = sizeof("0xabcdabcd : 0x12345678 :R\n") - 1;
drivers/gpu/drm/xe/xe_ttm_vram_mgr.c
Quoted 10 lines
drivers/gpu/drm/xe/xe_ttm_vram_mgr.c +7
-0+const unsigned int maxpage_size =sizeof ("max_pages: 10000\n" ) -1 ; +struct xe_ttm_vram_offline_resource *pos, *n; +struct gpu_buddy_block *block; +ssize_t s =0 ; + +mutex_lock (&mgr->lock); + s += scnprintf(&buf[s], maxpage_size + 1, "max_pages: %d\n", mgr-
max_pages);drivers/gpu/drm/xe/xe_ttm_vram_mgr.c +2
-0+list_for_each_entry_safe (pos, n, &mgr->offlined_pages, +offlined_link){
drivers/gpu/drm/xe/xe_ttm_vram_mgr.c int xe_ttm_vram_handle_addr_fault(struct xe_device *xe, unsigned long addr)
Quoted 10 lines
drivers/gpu/drm/xe/xe_ttm_vram_mgr.c +10
-0+ block =list_first_entry (&pos->blocks, + struct gpu_buddy_block, + link); + s +=scnprintf (&buf[s], element_size +1 , +"0x%08llx : 0x%08llx : %1s\n" , +gpu_buddy_block_offset (block) >>PAGE_SHIFT , +gpu_buddy_block_size (&mgr->mm, block), +"R" ); + } +list_for_each_entry_safe (pos, n, &mgr->queued_pages, queued_link){
Quoted 38 lines
drivers/gpu/drm/xe/xe_ttm_vram_mgr.c +38
-0+ block =list_first_entry (&pos->blocks, + struct gpu_buddy_block, + link); + s +=scnprintf (&buf[s], element_size +1 , +"0x%08llx : 0x%08llx : %1s\n" , +gpu_buddy_block_offset (block) >>PAGE_SHIFT , +gpu_buddy_block_size (&mgr->mm, block), + pos->status ?"P" :"F" ); + } +mutex_unlock (&mgr->lock); +} + +static ssize_t vram_bad_pages_show (struct device *dev,struct +device_attribute *attr,char *buf) { +struct pci_dev *pdev =to_pci_dev (dev); +struct xe_device *xe =pdev_to_xe_device (pdev); +struct ttm_resource_manager *man; +struct xe_ttm_vram_mgr *mgr; + + man =ttm_manager_type (&xe->ttm,XE_PL_VRAM0 ); +if (man) { + mgr =to_xe_ttm_vram_mgr (man); +xe_ttm_vram_dump_bad_pages_info (buf, mgr); + } + +return sysfs_emit (buf,"%s\n" , buf); }static +DEVICE_ATTR_RO (vram_bad_pages); + +static void xe_ttm_vram_sysfs_fini (void *arg) { +struct xe_device *xe = arg; + +device_remove_file (xe->drm.dev, &dev_attr_vram_bad_pages); } + +/** +* xe_ttm_vram_sysfs_init - Initialize vram sysfs component +* @tile: Xe Tile object +* +* It needs to be initialized after the main tile component is drivers/gpu/drm/xe/xe_ttm_vram_mgr.c int xe_ttm_vram_handle_addr_fault(struct xe_device *xe, unsigned long addr)
Quoted 10 lines
drivers/gpu/drm/xe/xe_ttm_vram_mgr.c +10
-0+ready +* +* Returns: 0 on success, negative error code on error. +*/ +int xe_ttm_vram_sysfs_init (struct xe_device *xe) { +int err; + + err =device_create_file (xe->drm.dev, &dev_attr_vram_bad_pages); +if (err) { + dev_err(xe->drm.dev, "Failed to create vram_bad_pages sysfsfile: %d\n", err);
drivers/gpu/drm/xe/xe_ttm_vram_mgr.c int xe_ttm_vram_handle_addr_fault(struct xe_device *xe, unsigned long addr)
Quoted 4 lines
drivers/gpu/drm/xe/xe_ttm_vram_mgr.c +4
-0+return 0 ; + } + + return devm_add_action_or_reset(xe->drm.dev,
xe_ttm_vram_sysfs_fini,drivers/gpu/drm/xe/xe_ttm_vram_mgr.h void xe_ttm_vram_get_used(struct
Quoted 7 lines
drivers/gpu/drm/xe/xe_ttm_vram_mgr.c +1
-0+xe); } EXPORT_SYMBOL(xe_ttm_vram_sysfs_init);drivers/gpu/drm/xe/xe_ttm_vram_mgr.h +0
-0@@ -32,6 +32,7 @@ void xe_ttm_vram_get_used(struct
ttm_resource_manager*man,Quoted 14 lines
drivers/gpu/drm/xe/xe_ttm_vram_mgr.h +1
-0u64 *used, u64 *used_visible);int xe_ttm_vram_handle_addr_fault (struct xe_device *xe,unsigned long addr); +int xe_ttm_vram_sysfs_init (struct xe_device *xe);static inline struct xe_ttm_vram_mgr_resource *to_xe_ttm_vram_mgr_resource (struct ttm_resource *res){drivers/gpu/drm/xe/xe_ttm_vram_mgr_types.h +0
-0drivers/gpu/drm/xe/xe_ttm_vram_mgr_types.h struct xe_ttm_vram_mgr { · line 42
Quoted 4 lines
drivers/gpu/drm/xe/xe_ttm_vram_mgr_types.h +1
-0@@ -39,6 +39,8 @@ struct xe_ttm_vram_mgr {u32 mem_type;/** @offline_mode: debugfs hook for setting page offline mode */ u64 offline_mode; +/** @max_pages: max pages that can be in offline queue retrieved from FW */
Quoted 4 lines
drivers/gpu/drm/xe/xe_ttm_vram_mgr_types.h +1
-0+u16 max_pages; };/**
RE: [RFC PATCH V7 10/10] drm/xe/cri: Add sysfs interface for bad gpu vram pages
In reply to Matthew AuldQuoted 52 lines
-----Original Message-----
From: Auld, Matthew <[email protected]>
Sent: 05 May 2026 14:14
To: Upadhyay, Tejas <[email protected]>; intel- [email protected]
Cc: Brost, Matthew <[email protected]>; [email protected]; Ghimiray, Himal Prasad <[email protected]>
Subject: Re: [RFC PATCH V7 10/10] drm/xe/cri: Add sysfs interface for bad gpu vram pagesQuoted 23 lines · Tejas
Quoted 23 lines
-----Original Message-----
From: Auld, Matthew <[email protected]>
Sent: 30 April 2026 19:23
To: Upadhyay, Tejas <[email protected]>; intel- [email protected]
Cc: Brost, Matthew <[email protected]>; [email protected]; Ghimiray, Himal Prasad <[email protected]>
Subject: Re: [RFC PATCH V7 10/10] drm/xe/cri: Add sysfs interface for bad gpu vram pagesQuoted 8 lines · Tejas Upadhyay
Starting CRI, Include a sysfs interface designed to expose information about bad VRAM pages—those identified as having hardware faults (e.g., ECC errors). This interface allows userspace tools and administrators to monitor the health of the GPU's local memory and track the status of page retirement.To get details on bad gpu vram pages can be found under
/sys/bus/pci/devices/bdf/vram_bad_pages.Where The format is, pfn : gpu page size : flags
Quoted 7 lines
Quoted 5 lines
With "gpu page size" this is really just the min block size? gpu-page-size is normally interpreted as GTT page size, which is a different thing. But is that not always 4K here? Since that is the granularity of the addr reservation? Is it useful to print that? Is knowing that pfn x is offlined not enough?
I think you are right, its always 4K. But I will check with Aravind as for what
reason design doc is saying format to be "pfn : gpu page size : flags". Adding him in cc.
Quoted 4 lines
Quoted 4 lines
Also what is the story if you have multiple VRAM instances here? There is only one
vram_bad_pagesfile? Would this treat VRAM as one giantunified thing?
Quoted 1 line
Quoted 8 lines
Quoted 1 line
This is targeted for CRI, thus single VRAM is in reference right now. We can
extend in future if need be or may be it will single huge thing only in future for multiple vram.
Since this is uapi, I think would be good to be forward looking and make sure we don't need to complety re-design if now asked to support multiple VRAM instances. I think one flat unfied thing listing all the pfns, should work, but I guess depends on the consumer of this and how they are going to use this.
I discussed with Aravind, it is intended for one big thing only not per VRAM, no such requirement. Also the gpu_page_size could be 4K now but for future compatibility, like we had in PVC for some cases we had 64K page size, we kept it in format.
Tejas
Quoted 194 lines
Quoted 194 lines
Tejas
Quoted 190 lines
Quoted 20 lines
flags:
R: reserved, this gpu page is reserved.
P: pending for reserve, this gpu page is marked as bad, will be reserved
in next window ofpage_reserve.
F: unable to reserve. this gpu page can’t be reserved due to some reasons.For example if you read using cat
/sys/bus/pci/devices/bdf/vram_bad_pages,max_pages: 100000x00000000:0x00001000: R0x00001234:0x00001000: Pv3:
- Move FW communication in RAS code
v2:
- Addmax_pagesinfo as per updated design doc
- RebaseQuoted 4 lines
Signed-off-by Tejas Upadhyay <[email protected]>--- drivers/gpu/drm/xe/xe_device_sysfs.c | 7 ++ drivers/gpu/drm/xe/xe_ttm_vram_mgr.c | 79++++++++++++++++++++++
drivers/gpu/drm/xe/xe_device_sysfs.c static const struct attribute_group
Quoted 25 lines
drivers/gpu/drm/xe/xe_ttm_vram_mgr.h | 1 + drivers/gpu/drm/xe/xe_ttm_vram_mgr_types.h | 2 + 4 files changed, 89 insertions(+)drivers/gpu/drm/xe/xe_device_sysfs.c +2
-0@@ -8,12 +8,14 @@#include <linux/pci.h> #include <linux/sysfs.h> +#include "xe_configfs.h" #include "xe_device.h" #include "xe_device_sysfs.h" #include "xe_mmio.h" #include "xe_pcode_api.h" #include "xe_pcode.h" #include "xe_pm.h" +#include "xe_ttm_vram_mgr.h" /** * DOC: Xe device sysfs @@ -267,6 +269,7 @@ static const struct attribute_group
auto_link_downgrade_attr_group= {drivers/gpu/drm/xe/xe_ttm_vram_mgr.c int xe_ttm_vram_handle_addr_fault(struct
Quoted 23 lines
drivers/gpu/drm/xe/xe_device_sysfs.c +5
-0int xe_device_sysfs_init (struct xe_device *xe) {struct device *dev = xe->drm.dev; +bool policy;int ret;if (xe->d3cold.capable) {@@ -285,5 +288,9 @@ int xe_device_sysfs_init(struct xe_device *xe)return ret; } + policy =xe_configfs_get_bad_page_reservation (to_pci_dev (dev)); +if (xe->info.platform ==XE_CRESCENTISLAND && policy) +xe_ttm_vram_sysfs_init (xe); +return 0 ; }drivers/gpu/drm/xe/xe_ttm_vram_mgr.c +0
-0@@ -760,3 +760,82 @@ int xe_ttm_vram_handle_addr_fault(struct
xe_device*xe, unsigned long addr)drivers/gpu/drm/xe/xe_ttm_vram_mgr.c int xe_ttm_vram_handle_addr_fault(struct xe_device *xe, unsigned long addr)
Quoted 7 lines
drivers/gpu/drm/xe/xe_ttm_vram_mgr.c +4
-0return ret; }EXPORT_SYMBOL (xe_ttm_vram_handle_addr_fault); + +static void xe_ttm_vram_dump_bad_pages_info (char *buf,struct +xe_ttm_vram_mgr *mgr) { + const unsigned int element_size = sizeof("0xabcdabcd : 0x12345678 :R\n") - 1;
drivers/gpu/drm/xe/xe_ttm_vram_mgr.c
Quoted 10 lines
drivers/gpu/drm/xe/xe_ttm_vram_mgr.c +7
-0+const unsigned int maxpage_size =sizeof ("max_pages: 10000\n" ) -1 ; +struct xe_ttm_vram_offline_resource *pos, *n; +struct gpu_buddy_block *block; +ssize_t s =0 ; + +mutex_lock (&mgr->lock); + s += scnprintf(&buf[s], maxpage_size + 1, "max_pages: %d\n", mgr-
max_pages);drivers/gpu/drm/xe/xe_ttm_vram_mgr.c +2
-0+list_for_each_entry_safe (pos, n, &mgr->offlined_pages, +offlined_link){
drivers/gpu/drm/xe/xe_ttm_vram_mgr.c int xe_ttm_vram_handle_addr_fault(struct xe_device *xe, unsigned long addr)
Quoted 10 lines
drivers/gpu/drm/xe/xe_ttm_vram_mgr.c +10
-0+ block =list_first_entry (&pos->blocks, + struct gpu_buddy_block, + link); + s +=scnprintf (&buf[s], element_size +1 , +"0x%08llx : 0x%08llx : %1s\n" , +gpu_buddy_block_offset (block) >>PAGE_SHIFT , +gpu_buddy_block_size (&mgr->mm, block), +"R" ); + } +list_for_each_entry_safe (pos, n, &mgr->queued_pages, queued_link){
Quoted 38 lines
drivers/gpu/drm/xe/xe_ttm_vram_mgr.c +38
-0+ block =list_first_entry (&pos->blocks, + struct gpu_buddy_block, + link); + s +=scnprintf (&buf[s], element_size +1 , +"0x%08llx : 0x%08llx : %1s\n" , +gpu_buddy_block_offset (block) >>PAGE_SHIFT , +gpu_buddy_block_size (&mgr->mm, block), + pos->status ?"P" :"F" ); + } +mutex_unlock (&mgr->lock); +} + +static ssize_t vram_bad_pages_show (struct device *dev,struct +device_attribute *attr,char *buf) { +struct pci_dev *pdev =to_pci_dev (dev); +struct xe_device *xe =pdev_to_xe_device (pdev); +struct ttm_resource_manager *man; +struct xe_ttm_vram_mgr *mgr; + + man =ttm_manager_type (&xe->ttm,XE_PL_VRAM0 ); +if (man) { + mgr =to_xe_ttm_vram_mgr (man); +xe_ttm_vram_dump_bad_pages_info (buf, mgr); + } + +return sysfs_emit (buf,"%s\n" , buf); }static +DEVICE_ATTR_RO (vram_bad_pages); + +static void xe_ttm_vram_sysfs_fini (void *arg) { +struct xe_device *xe = arg; + +device_remove_file (xe->drm.dev, &dev_attr_vram_bad_pages); } + +/** +* xe_ttm_vram_sysfs_init - Initialize vram sysfs component +* @tile: Xe Tile object +* +* It needs to be initialized after the main tile component is drivers/gpu/drm/xe/xe_ttm_vram_mgr.c int xe_ttm_vram_handle_addr_fault(struct xe_device *xe, unsigned long addr)
Quoted 10 lines
drivers/gpu/drm/xe/xe_ttm_vram_mgr.c +10
-0+ready +* +* Returns: 0 on success, negative error code on error. +*/ +int xe_ttm_vram_sysfs_init (struct xe_device *xe) { +int err; + + err =device_create_file (xe->drm.dev, &dev_attr_vram_bad_pages); +if (err) { + dev_err(xe->drm.dev, "Failed to create vram_bad_pages sysfsfile: %d\n", err);
drivers/gpu/drm/xe/xe_ttm_vram_mgr.c int xe_ttm_vram_handle_addr_fault(struct xe_device *xe, unsigned long addr)
Quoted 4 lines
drivers/gpu/drm/xe/xe_ttm_vram_mgr.c +4
-0+return 0 ; + } + + return devm_add_action_or_reset(xe->drm.dev,
xe_ttm_vram_sysfs_fini,drivers/gpu/drm/xe/xe_ttm_vram_mgr.h void xe_ttm_vram_get_used(struct
Quoted 7 lines
drivers/gpu/drm/xe/xe_ttm_vram_mgr.c +1
-0+xe); } EXPORT_SYMBOL(xe_ttm_vram_sysfs_init);drivers/gpu/drm/xe/xe_ttm_vram_mgr.h +0
-0@@ -32,6 +32,7 @@ void xe_ttm_vram_get_used(struct
ttm_resource_manager*man,Quoted 14 lines
drivers/gpu/drm/xe/xe_ttm_vram_mgr.h +1
-0u64 *used, u64 *used_visible);int xe_ttm_vram_handle_addr_fault (struct xe_device *xe,unsigned long addr); +int xe_ttm_vram_sysfs_init (struct xe_device *xe);static inline struct xe_ttm_vram_mgr_resource *to_xe_ttm_vram_mgr_resource (struct ttm_resource *res){drivers/gpu/drm/xe/xe_ttm_vram_mgr_types.h +0
-0drivers/gpu/drm/xe/xe_ttm_vram_mgr_types.h struct xe_ttm_vram_mgr { · line 42
Quoted 4 lines
drivers/gpu/drm/xe/xe_ttm_vram_mgr_types.h +1
-0@@ -39,6 +39,8 @@ struct xe_ttm_vram_mgr {u32 mem_type;/** @offline_mode: debugfs hook for setting page offline mode */ u64 offline_mode; +/** @max_pages: max pages that can be in offline queue retrieved from FW */
Quoted 4 lines
drivers/gpu/drm/xe/xe_ttm_vram_mgr_types.h +1
-0+u16 max_pages; };/**
Matthew Auld
maintainerRe: [RFC PATCH V7 10/10] drm/xe/cri: Add sysfs interface for bad gpu vram pages
Quoted 56 lines · Tejas
Quoted 56 lines
-----Original Message-----
From: Auld, Matthew <[email protected]>
Sent: 05 May 2026 14:14
To: Upadhyay, Tejas <[email protected]>; intel- [email protected]
Cc: Brost, Matthew <[email protected]>; [email protected]; Ghimiray, Himal Prasad <[email protected]>
Subject: Re: [RFC PATCH V7 10/10] drm/xe/cri: Add sysfs interface for bad gpu vram pagesQuoted 23 lines · Tejas
Quoted 23 lines
-----Original Message-----
From: Auld, Matthew <[email protected]>
Sent: 30 April 2026 19:23
To: Upadhyay, Tejas <[email protected]>; intel- [email protected]
Cc: Brost, Matthew <[email protected]>; [email protected]; Ghimiray, Himal Prasad <[email protected]>
Subject: Re: [RFC PATCH V7 10/10] drm/xe/cri: Add sysfs interface for bad gpu vram pagesQuoted 8 lines · Tejas Upadhyay
Starting CRI, Include a sysfs interface designed to expose information about bad VRAM pages—those identified as having hardware faults (e.g., ECC errors). This interface allows userspace tools and administrators to monitor the health of the GPU's local memory and track the status of page retirement.To get details on bad gpu vram pages can be found under
/sys/bus/pci/devices/bdf/vram_bad_pages.Where The format is, pfn : gpu page size : flags
Quoted 7 lines
Quoted 5 lines
With "gpu page size" this is really just the min block size? gpu-page-size is normally interpreted as GTT page size, which is a different thing. But is that not always 4K here? Since that is the granularity of the addr reservation? Is it useful to print that? Is knowing that pfn x is offlined not enough?
I think you are right, its always 4K. But I will check with Aravind as for what
reason design doc is saying format to be "pfn : gpu page size : flags". Adding him in cc.
Quoted 4 lines
Quoted 4 lines
Also what is the story if you have multiple VRAM instances here? There is only one
vram_bad_pagesfile? Would this treat VRAM as one giantunified thing?
Quoted 2 lines
This is targeted for CRI, thus single VRAM is in reference right now. We can
extend in future if need be or may be it will single huge thing only in future for multiple vram.
Quoted 6 lines
Quoted 4 lines
Since this is uapi, I think would be good to be forward looking and make sure we don't need to complety re-design if now asked to support multiple VRAM instances. I think one flat unfied thing listing all the pfns, should work, but I guess depends on the consumer of this and how they are going to use this.
I discussed with Aravind, it is intended for one big thing only not per VRAM, no such requirement. Also the
gpu_page_sizecould be 4K now but for future compatibility, like we had in PVC for some cases we had 64K page size, we kept it in format.
The PVC 64K page (and some early BMG), was a GTT page thing, so from allocator pov you could still allocate 4K, and I think we did exactly that for memory that was never touched via GTT. If the FW gives 64K addresses then I think makes sense.
Also just realised we might get some funny behaviour with CPU PAGE_SIZE != 4K, since allocator chunk_size will currently match the CPU PAGE_SIZE. Driver doesn't yet support that fully, but might in the future.
Perhaps just "PFN size: <>" at the top, instead of repeating on every line? AFAICT it should always be the same, on that machine?
Quoted 197 lines
Tejas
Quoted 194 lines
Quoted 194 lines
Tejas
Quoted 190 lines
Quoted 20 lines
flags:
R: reserved, this gpu page is reserved.
P: pending for reserve, this gpu page is marked as bad, will be reserved
in next window ofpage_reserve.
F: unable to reserve. this gpu page can’t be reserved due to some reasons.For example if you read using cat
/sys/bus/pci/devices/bdf/vram_bad_pages,max_pages: 100000x00000000:0x00001000: R0x00001234:0x00001000: Pv3:
- Move FW communication in RAS code
v2:
- Addmax_pagesinfo as per updated design doc
- RebaseQuoted 4 lines
Signed-off-by Tejas Upadhyay <[email protected]>--- drivers/gpu/drm/xe/xe_device_sysfs.c | 7 ++ drivers/gpu/drm/xe/xe_ttm_vram_mgr.c | 79++++++++++++++++++++++
drivers/gpu/drm/xe/xe_device_sysfs.c static const struct attribute_group
Quoted 25 lines
drivers/gpu/drm/xe/xe_ttm_vram_mgr.h | 1 + drivers/gpu/drm/xe/xe_ttm_vram_mgr_types.h | 2 + 4 files changed, 89 insertions(+)drivers/gpu/drm/xe/xe_device_sysfs.c +2
-0@@ -8,12 +8,14 @@#include <linux/pci.h> #include <linux/sysfs.h> +#include "xe_configfs.h" #include "xe_device.h" #include "xe_device_sysfs.h" #include "xe_mmio.h" #include "xe_pcode_api.h" #include "xe_pcode.h" #include "xe_pm.h" +#include "xe_ttm_vram_mgr.h" /** * DOC: Xe device sysfs @@ -267,6 +269,7 @@ static const struct attribute_group
auto_link_downgrade_attr_group= {drivers/gpu/drm/xe/xe_ttm_vram_mgr.c int xe_ttm_vram_handle_addr_fault(struct
Quoted 23 lines
drivers/gpu/drm/xe/xe_device_sysfs.c +5
-0int xe_device_sysfs_init (struct xe_device *xe) {struct device *dev = xe->drm.dev; +bool policy;int ret;if (xe->d3cold.capable) {@@ -285,5 +288,9 @@ int xe_device_sysfs_init(struct xe_device *xe)return ret; } + policy =xe_configfs_get_bad_page_reservation (to_pci_dev (dev)); +if (xe->info.platform ==XE_CRESCENTISLAND && policy) +xe_ttm_vram_sysfs_init (xe); +return 0 ; }drivers/gpu/drm/xe/xe_ttm_vram_mgr.c +0
-0@@ -760,3 +760,82 @@ int xe_ttm_vram_handle_addr_fault(struct
xe_device*xe, unsigned long addr)drivers/gpu/drm/xe/xe_ttm_vram_mgr.c int xe_ttm_vram_handle_addr_fault(struct xe_device *xe, unsigned long addr)
Quoted 7 lines
drivers/gpu/drm/xe/xe_ttm_vram_mgr.c +4
-0return ret; }EXPORT_SYMBOL (xe_ttm_vram_handle_addr_fault); + +static void xe_ttm_vram_dump_bad_pages_info (char *buf,struct +xe_ttm_vram_mgr *mgr) { + const unsigned int element_size = sizeof("0xabcdabcd : 0x12345678 :R\n") - 1;
drivers/gpu/drm/xe/xe_ttm_vram_mgr.c
Quoted 10 lines
drivers/gpu/drm/xe/xe_ttm_vram_mgr.c +7
-0+const unsigned int maxpage_size =sizeof ("max_pages: 10000\n" ) -1 ; +struct xe_ttm_vram_offline_resource *pos, *n; +struct gpu_buddy_block *block; +ssize_t s =0 ; + +mutex_lock (&mgr->lock); + s += scnprintf(&buf[s], maxpage_size + 1, "max_pages: %d\n", mgr-
max_pages);drivers/gpu/drm/xe/xe_ttm_vram_mgr.c +2
-0+list_for_each_entry_safe (pos, n, &mgr->offlined_pages, +offlined_link){
drivers/gpu/drm/xe/xe_ttm_vram_mgr.c int xe_ttm_vram_handle_addr_fault(struct xe_device *xe, unsigned long addr)
Quoted 10 lines
drivers/gpu/drm/xe/xe_ttm_vram_mgr.c +10
-0+ block =list_first_entry (&pos->blocks, + struct gpu_buddy_block, + link); + s +=scnprintf (&buf[s], element_size +1 , +"0x%08llx : 0x%08llx : %1s\n" , +gpu_buddy_block_offset (block) >>PAGE_SHIFT , +gpu_buddy_block_size (&mgr->mm, block), +"R" ); + } +list_for_each_entry_safe (pos, n, &mgr->queued_pages, queued_link){
Quoted 38 lines
drivers/gpu/drm/xe/xe_ttm_vram_mgr.c +38
-0+ block =list_first_entry (&pos->blocks, + struct gpu_buddy_block, + link); + s +=scnprintf (&buf[s], element_size +1 , +"0x%08llx : 0x%08llx : %1s\n" , +gpu_buddy_block_offset (block) >>PAGE_SHIFT , +gpu_buddy_block_size (&mgr->mm, block), + pos->status ?"P" :"F" ); + } +mutex_unlock (&mgr->lock); +} + +static ssize_t vram_bad_pages_show (struct device *dev,struct +device_attribute *attr,char *buf) { +struct pci_dev *pdev =to_pci_dev (dev); +struct xe_device *xe =pdev_to_xe_device (pdev); +struct ttm_resource_manager *man; +struct xe_ttm_vram_mgr *mgr; + + man =ttm_manager_type (&xe->ttm,XE_PL_VRAM0 ); +if (man) { + mgr =to_xe_ttm_vram_mgr (man); +xe_ttm_vram_dump_bad_pages_info (buf, mgr); + } + +return sysfs_emit (buf,"%s\n" , buf); }static +DEVICE_ATTR_RO (vram_bad_pages); + +static void xe_ttm_vram_sysfs_fini (void *arg) { +struct xe_device *xe = arg; + +device_remove_file (xe->drm.dev, &dev_attr_vram_bad_pages); } + +/** +* xe_ttm_vram_sysfs_init - Initialize vram sysfs component +* @tile: Xe Tile object +* +* It needs to be initialized after the main tile component is drivers/gpu/drm/xe/xe_ttm_vram_mgr.c int xe_ttm_vram_handle_addr_fault(struct xe_device *xe, unsigned long addr)
Quoted 10 lines
drivers/gpu/drm/xe/xe_ttm_vram_mgr.c +10
-0+ready +* +* Returns: 0 on success, negative error code on error. +*/ +int xe_ttm_vram_sysfs_init (struct xe_device *xe) { +int err; + + err =device_create_file (xe->drm.dev, &dev_attr_vram_bad_pages); +if (err) { + dev_err(xe->drm.dev, "Failed to create vram_bad_pages sysfsfile: %d\n", err);
drivers/gpu/drm/xe/xe_ttm_vram_mgr.c int xe_ttm_vram_handle_addr_fault(struct xe_device *xe, unsigned long addr)
Quoted 4 lines
drivers/gpu/drm/xe/xe_ttm_vram_mgr.c +4
-0+return 0 ; + } + + return devm_add_action_or_reset(xe->drm.dev,
xe_ttm_vram_sysfs_fini,drivers/gpu/drm/xe/xe_ttm_vram_mgr.h void xe_ttm_vram_get_used(struct
Quoted 7 lines
drivers/gpu/drm/xe/xe_ttm_vram_mgr.c +1
-0+xe); } EXPORT_SYMBOL(xe_ttm_vram_sysfs_init);drivers/gpu/drm/xe/xe_ttm_vram_mgr.h +0
-0@@ -32,6 +32,7 @@ void xe_ttm_vram_get_used(struct
ttm_resource_manager*man,Quoted 14 lines
drivers/gpu/drm/xe/xe_ttm_vram_mgr.h +1
-0u64 *used, u64 *used_visible);int xe_ttm_vram_handle_addr_fault (struct xe_device *xe,unsigned long addr); +int xe_ttm_vram_sysfs_init (struct xe_device *xe);static inline struct xe_ttm_vram_mgr_resource *to_xe_ttm_vram_mgr_resource (struct ttm_resource *res){drivers/gpu/drm/xe/xe_ttm_vram_mgr_types.h +0
-0drivers/gpu/drm/xe/xe_ttm_vram_mgr_types.h struct xe_ttm_vram_mgr { · line 42
Quoted 4 lines
drivers/gpu/drm/xe/xe_ttm_vram_mgr_types.h +1
-0@@ -39,6 +39,8 @@ struct xe_ttm_vram_mgr {u32 mem_type;/** @offline_mode: debugfs hook for setting page offline mode */ u64 offline_mode; +/** @max_pages: max pages that can be in offline queue retrieved from FW */
Quoted 4 lines
drivers/gpu/drm/xe/xe_ttm_vram_mgr_types.h +1
-0+u16 max_pages; };/**
RE: [RFC PATCH V7 10/10] drm/xe/cri: Add sysfs interface for bad gpu vram pages
Quoted 91 lines
-----Original Message-----
From: Auld, Matthew <[email protected]>
Sent: 07 May 2026 15:20
To: Upadhyay, Tejas <[email protected]>; intel- [email protected]
Cc: Brost, Matthew <[email protected]>; [email protected]; Ghimiray, Himal Prasad <[email protected]>; Iddamsetty, Aravind <[email protected]>
Subject: Re: [RFC PATCH V7 10/10] drm/xe/cri: Add sysfs interface for bad gpu vram pagesQuoted 27 lines · Tejas
Quoted 27 lines
-----Original Message-----
From: Auld, Matthew <[email protected]>
Sent: 05 May 2026 14:14
To: Upadhyay, Tejas <[email protected]>; intel- [email protected]
Cc: Brost, Matthew <[email protected]>; [email protected]; Ghimiray, Himal Prasad <[email protected]>
Subject: Re: [RFC PATCH V7 10/10] drm/xe/cri: Add sysfs interface for bad gpu vram pagesQuoted 13 lines · Tejas
Quoted 13 lines
-----Original Message-----
From: Auld, Matthew <[email protected]>
Sent: 30 April 2026 19:23
To: Upadhyay, Tejas <[email protected]>; intel- [email protected]
Cc: Brost, Matthew <[email protected]>; [email protected]; Ghimiray, Himal Prasad <[email protected]>
Subject: Re: [RFC PATCH V7 10/10] drm/xe/cri: Add sysfs interface for bad gpu vram pagesQuoted 7 lines
Quoted 7 lines
Quoted 7 lines
Quoted 7 lines · Tejas Upadhyay
Quoted 6 lines
Starting CRI, Include a sysfs interface designed to expose information about bad VRAM pages—those identified as having hardware faults (e.g., ECC errors). This interface allows userspace tools and administrators to monitor the health of the GPU's local memory and track the status of page retirement.To get details on bad gpu vram pages can be found under
/sys/bus/pci/devices/bdf/vram_bad_pages.Quoted 22 lines
Quoted 22 lines
Quoted 11 lines
Quoted 8 lines
Quoted 2 lines
Where The format is, pfn : gpu page size : flags
With "gpu page size" this is really just the min block size? gpu-page-size is normally interpreted as GTT page size, which is a different thing. But is that not always 4K here? Since that is the granularity of the addr reservation? Is it useful to print that? Is knowing that pfn x is offlined not enough?
I think you are right, its always 4K. But I will check with Aravind as for what
reason design doc is saying format to be "pfn : gpu page size :
flags". Adding him in cc.Quoted 5 lines
Quoted 5 lines
Also what is the story if you have multiple VRAM instances here? There is only one
vram_bad_pagesfile? Would this treat VRAM as one giantunified thing?
Quoted 3 lines
This is targeted for CRI, thus single VRAM is in reference right now. We can
Quoted 7 lines
Quoted 7 lines
extend in future if need be or may be it will single huge thing only in future for multiple vram.
Since this is uapi, I think would be good to be forward looking and make sure we don't need to complety re-design if now asked to support multiple VRAM instances. I think one flat unfied thing listing all the pfns, should work, but I guess depends on the consumer of this and
how they are going to use this.
Quoted 2 lines
I discussed with Aravind, it is intended for one big thing only not per VRAM,
no such requirement. Also the
gpu_page_sizecould be 4K now but for future compatibility, like we had in PVC for some cases we had 64K page size, we kept it in format.The PVC 64K page (and some early BMG), was a GTT page thing, so from allocator pov you could still allocate 4K, and I think we did exactly that for memory that was never touched via GTT. If the FW gives 64K addresses then I think makes sense.
Quoted 6 lines
Also just realised we might get some funny behaviour with CPU
PAGE_SIZE!= 4K, since allocatorchunk_sizewill currently match the CPUPAGE_SIZE. Driver doesn't yet support that fully, but might in the future.Perhaps just "PFN size: <>" at the top, instead of repeating on every line? AFAICT it should always be the same, on that machine?
Yeah it makes sense, FW will give 4K aligned addresses always atleast right now. Thoughts on Matt's proposal @Iddamsetty, Aravind?
Tejas
Quoted 202 lines
Quoted 13 lines
Tejas
Quoted 10 lines
Quoted 10 lines
Tejas
Quoted 7 lines
Quoted 7 lines
flags:
R: reserved, this gpu page is reserved.
P: pending for reserve, this gpu page is marked as bad, will be reserved
in next window ofpage_reserve.
F: unable to reserve. this gpu page can’t be reserved due to somereasons.
Quoted 164 lines
Quoted 164 lines
Quoted 164 lines
Quoted 164 lines
Quoted 13 lines
For example if you read using cat
/sys/bus/pci/devices/bdf/vram_bad_pages,max_pages: 100000x00000000:0x00001000: R0x00001234:0x00001000: Pv3:
- Move FW communication in RAS code
v2:
- Addmax_pagesinfo as per updated design doc
- RebaseQuoted 4 lines
Signed-off-by Tejas Upadhyay <[email protected]>--- drivers/gpu/drm/xe/xe_device_sysfs.c | 7 ++ drivers/gpu/drm/xe/xe_ttm_vram_mgr.c | 79++++++++++++++++++++++
drivers/gpu/drm/xe/xe_device_sysfs.c static const struct attribute_group
Quoted 25 lines
drivers/gpu/drm/xe/xe_ttm_vram_mgr.h | 1 + drivers/gpu/drm/xe/xe_ttm_vram_mgr_types.h | 2 + 4 files changed, 89 insertions(+)drivers/gpu/drm/xe/xe_device_sysfs.c +2
-0@@ -8,12 +8,14 @@#include <linux/pci.h> #include <linux/sysfs.h> +#include "xe_configfs.h" #include "xe_device.h" #include "xe_device_sysfs.h" #include "xe_mmio.h" #include "xe_pcode_api.h" #include "xe_pcode.h" #include "xe_pm.h" +#include "xe_ttm_vram_mgr.h" /** * DOC: Xe device sysfs @@ -267,6 +269,7 @@ static const struct attribute_group
auto_link_downgrade_attr_group= {drivers/gpu/drm/xe/xe_ttm_vram_mgr.c int xe_ttm_vram_handle_addr_fault(struct
Quoted 23 lines
drivers/gpu/drm/xe/xe_device_sysfs.c +5
-0int xe_device_sysfs_init (struct xe_device *xe) {struct device *dev = xe->drm.dev; +bool policy;int ret;if (xe->d3cold.capable) {@@ -285,5 +288,9 @@ int xe_device_sysfs_init(struct xe_device *xe)return ret; } + policy =xe_configfs_get_bad_page_reservation (to_pci_dev (dev)); +if (xe->info.platform ==XE_CRESCENTISLAND && policy) +xe_ttm_vram_sysfs_init (xe); +return 0 ; }drivers/gpu/drm/xe/xe_ttm_vram_mgr.c +0
-0@@ -760,3 +760,82 @@ int xe_ttm_vram_handle_addr_fault(struct
xe_device*xe, unsigned long addr)drivers/gpu/drm/xe/xe_ttm_vram_mgr.c int xe_ttm_vram_handle_addr_fault(struct xe_device *xe, unsigned long addr)
Quoted 7 lines
drivers/gpu/drm/xe/xe_ttm_vram_mgr.c +4
-0return ret; }EXPORT_SYMBOL (xe_ttm_vram_handle_addr_fault); + +static void xe_ttm_vram_dump_bad_pages_info (char *buf,struct +xe_ttm_vram_mgr *mgr) { + const unsigned int element_size = sizeof("0xabcdabcd : 0x12345678 :R\n") - 1;
drivers/gpu/drm/xe/xe_ttm_vram_mgr.c
Quoted 11 lines
drivers/gpu/drm/xe/xe_ttm_vram_mgr.c +8
-0+const unsigned int maxpage_size =sizeof ("max_pages: 10000\n" ) -1 ; +struct xe_ttm_vram_offline_resource *pos, *n; +struct gpu_buddy_block *block; +ssize_t s =0 ; + +mutex_lock (&mgr->lock); + s += scnprintf(&buf[s], maxpage_size + 1, "max_pages: %d\n", +mgr-
max_pages);drivers/gpu/drm/xe/xe_ttm_vram_mgr.c +2
-0+list_for_each_entry_safe (pos, n, &mgr->offlined_pages, +offlined_link){
drivers/gpu/drm/xe/xe_ttm_vram_mgr.c int xe_ttm_vram_handle_addr_fault(struct xe_device *xe, unsigned long addr)
Quoted 11 lines
drivers/gpu/drm/xe/xe_ttm_vram_mgr.c +11
-0+ block =list_first_entry (&pos->blocks, + struct gpu_buddy_block, + link); + s +=scnprintf (&buf[s], element_size +1 , +"0x%08llx : 0x%08llx : %1s\n" , +gpu_buddy_block_offset (block) >>PAGE_SHIFT , +gpu_buddy_block_size (&mgr->mm, block), +"R" ); + } +list_for_each_entry_safe (pos, n, &mgr->queued_pages, +queued_link){
Quoted 38 lines
drivers/gpu/drm/xe/xe_ttm_vram_mgr.c +38
-0+ block =list_first_entry (&pos->blocks, + struct gpu_buddy_block, + link); + s +=scnprintf (&buf[s], element_size +1 , +"0x%08llx : 0x%08llx : %1s\n" , +gpu_buddy_block_offset (block) >>PAGE_SHIFT , +gpu_buddy_block_size (&mgr->mm, block), + pos->status ?"P" :"F" ); + } +mutex_unlock (&mgr->lock); +} + +static ssize_t vram_bad_pages_show (struct device *dev,struct +device_attribute *attr,char *buf) { +struct pci_dev *pdev =to_pci_dev (dev); +struct xe_device *xe =pdev_to_xe_device (pdev); +struct ttm_resource_manager *man; +struct xe_ttm_vram_mgr *mgr; + + man =ttm_manager_type (&xe->ttm,XE_PL_VRAM0 ); +if (man) { + mgr =to_xe_ttm_vram_mgr (man); +xe_ttm_vram_dump_bad_pages_info (buf, mgr); + } + +return sysfs_emit (buf,"%s\n" , buf); }static +DEVICE_ATTR_RO (vram_bad_pages); + +static void xe_ttm_vram_sysfs_fini (void *arg) { +struct xe_device *xe = arg; + +device_remove_file (xe->drm.dev, &dev_attr_vram_bad_pages); } + +/** +* xe_ttm_vram_sysfs_init - Initialize vram sysfs component +* @tile: Xe Tile object +* +* It needs to be initialized after the main tile component is drivers/gpu/drm/xe/xe_ttm_vram_mgr.c int xe_ttm_vram_handle_addr_fault(struct xe_device *xe, unsigned long addr)
Quoted 10 lines
drivers/gpu/drm/xe/xe_ttm_vram_mgr.c +10
-0+ready +* +* Returns: 0 on success, negative error code on error. +*/ +int xe_ttm_vram_sysfs_init (struct xe_device *xe) { +int err; + + err =device_create_file (xe->drm.dev, &dev_attr_vram_bad_pages); +if (err) { + dev_err(xe->drm.dev, "Failed to create vram_bad_pages sysfsfile: %d\n", err);
drivers/gpu/drm/xe/xe_ttm_vram_mgr.c int xe_ttm_vram_handle_addr_fault(struct xe_device *xe, unsigned long addr)
Quoted 4 lines
drivers/gpu/drm/xe/xe_ttm_vram_mgr.c +4
-0+return 0 ; + } + + return devm_add_action_or_reset(xe->drm.dev,
xe_ttm_vram_sysfs_fini,drivers/gpu/drm/xe/xe_ttm_vram_mgr.h void xe_ttm_vram_get_used(struct
Quoted 7 lines
drivers/gpu/drm/xe/xe_ttm_vram_mgr.c +1
-0+xe); } EXPORT_SYMBOL(xe_ttm_vram_sysfs_init);drivers/gpu/drm/xe/xe_ttm_vram_mgr.h +0
-0@@ -32,6 +32,7 @@ void xe_ttm_vram_get_used(struct
ttm_resource_manager*man,Quoted 2 lines
drivers/gpu/drm/xe/xe_ttm_vram_mgr.h +0
-0u64 *used, u64 *used_visible);Quoted 14 lines
Quoted 6 lines
Quoted 6 lines
Quoted 6 lines
Quoted 6 lines
drivers/gpu/drm/xe/xe_ttm_vram_mgr.h +1
-0int xe_ttm_vram_handle_addr_fault (struct xe_device *xe,unsigned long addr); +int xe_ttm_vram_sysfs_init (struct xe_device *xe);static inline struct xe_ttm_vram_mgr_resource *to_xe_ttm_vram_mgr_resource (struct ttm_resource *res){Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
drivers/gpu/drm/xe/xe_ttm_vram_mgr_types.h struct xe_ttm_vram_mgr { · line 40
Quoted 8 lines
drivers/gpu/drm/xe/xe_ttm_vram_mgr_types.h +0
-0@@ -39,6 +39,8 @@ struct xe_ttm_vram_mgr {u32 mem_type;/** @offline_mode: debugfs hook for setting page offline mode */
Quoted 9 lines
Quoted 8 lines
Quoted 8 lines
Quoted 7 lines
drivers/gpu/drm/xe/xe_ttm_vram_mgr_types.h struct xe_ttm_vram_mgr { · line 42
Quoted 2 lines
drivers/gpu/drm/xe/xe_ttm_vram_mgr_types.h +1
-0u64 offline_mode; +/** @max_pages: max pages that can be in offline queue retrieved from FW */
Quoted 4 lines
drivers/gpu/drm/xe/xe_ttm_vram_mgr_types.h +1
-0+u16 max_pages; };/**
Patchwork
CI✗ CI.checkpatch: warning for Add memory page offlining support (rev8)
In reply to Tejas Upadhyay== Series Details ==
== Series Details ==
Series: Add memory page offlining support (rev8)
URL : https://patchwork.freedesktop.org/series/161473/
State : warning
== Summary ==
+ KERNEL=/kernel
+ git clone https://gitlab.freedesktop.org/drm/maintainer-tools mt Cloning into 'mt'...
warning: redirecting to https://gitlab.freedesktop.org/drm/maintainer-tools.git/
+ git -C mt rev-list -n1 origin/master 1f57ba1afceae32108bd24770069f764d940a0e4
+ cd /kernel
+ git config --global --add safe.directory /kernel
+ git log -n1
commit cddb2434942e3dc976b828dca2b17d73fb7d6b68
Author: Tejas Upadhyay <[email protected]> Date: Thu Apr 16 13:19:59 2026 +0530
drm/xe/cri: Add sysfs interface for bad gpu vram pages
Starting CRI, Include a sysfs interface designed to expose information about bad VRAM pages—those identified as having hardware faults (e.g., ECC errors). This interface allows userspace tools and administrators to monitor the health of the GPU's local memory and track the status of page retirement.To get details on bad gpu vram pages can be found under /sys/bus/pci/devices/bdf/vram_bad_pages.
Where The format is, pfn : gpu page size : flags
flags:
R: reserved, this gpu page is reserved.
P: pending for reserve, this gpu page is marked as bad, will be reserved
in next window of page_reserve.
F: unable to reserve. this gpu page can’t be reserved due to some reasons.
For example if you read using cat /sys/bus/pci/devices/bdf/vram_bad_pages, max_pages : 10000
0x00000000 : 0x00001000 : R
0x00001234 : 0x00001000 : P
v3:
- Move FW communication in RAS code v2:
- Add max_pages info as per updated design doc
- Rebase
+ /mt/dim checkpatch 7e772296e93b337ca5e12fced2ac12716a159d28 drm-intel7ad853ed2074 drm/xe: Link VRAM object with gpu buddy9be10af07d2b gpu/buddy: Integrate lockdep for gpu buddy managercfbf17fee984 drm/gpu: Add gpu_buddy_allocated_addr_to_block helper8e7bb33194ec drm/xe: Link LRC BO and its execution Queue86665af845df drm/xe: Extend BO purge to handle vram pages as wellc6ca0861ed2a drm/xe: Handle physical memory address error
-:13: ERROR:BAD_COMMIT_SEPARATOR: Invalid commit separator - some tools may have problems applying this #13:
----------------------
-:20: ERROR:BAD_COMMIT_SEPARATOR: Invalid commit separator - some tools may have problems applying this #20:
-----------------
total: 2 errors, 0 warnings, 0 checks, 418 lines checked0aa34b8cb1c5 drm/xe/cri: Add debugfs to inject faulty vram addressb73669f195ce gpu/buddy: Add routine to dump allocated buddy blocks
-:13: WARNING:COMMIT_LOG_LONG_LINE: Prefer a maximum 75 chars per line (possible unwrapped commit description?) #13:
[ +0.000003] xe 0000:03:00.0: [drm] 0x00000002f8000000-0x00000002f8800000: 8388608
total: 0 errors, 1 warnings, 0 checks, 62 lines checkeda50fc40a0ce1 drm/xe/configfs: Add vram bad page reservation policycddb2434942e drm/xe/cri: Add sysfs interface for bad gpu vram pages
-:22: WARNING:COMMIT_LOG_LONG_LINE: Prefer a maximum 75 chars per line (possible unwrapped commit description?) #22:
F: unable to reserve. this gpu page can’t be reserved due to some reasons.
total: 0 errors, 1 warnings, 0 checks, 127 lines checked
Patchwork
CI✗ CI.KUnit: failure for Add memory page offlining support (rev8)
In reply to Tejas Upadhyay== Series Details ==
== Series Details ==
Series: Add memory page offlining support (rev8)
URL : https://patchwork.freedesktop.org/series/161473/
State : failure
== Summary ==
+ trap cleanup EXIT
+ /kernel/tools/testing/kunit/kunit.py run --kunitconfig /kernel/drivers/gpu/drm/xe/.kunitconfig [07:56:07] Configuring KUnit Kernel ...
Generating .config ...
Populating config with:
$ make ARCH=um O=.kunit olddefconfig [07:56:11] Building KUnit Kernel ... Populating config with: $ make ARCH=um O=.kunit olddefconfig Building with: $ make all compile_commands.json scripts_gdb ARCH=um O=.kunit --jobs=48 [07:56:42] Starting KUnit Kernel (1/1)... [07:56:42] ============================================================ Running tests with: $ .kunit/linux kunit.enable=1 mem=1G console=tty kunit_shutdown=halt [07:56:42] ================== guc_buf (11 subtests) =================== [07:56:42] [PASSED] test_smallest [07:56:42] [PASSED] test_largest [07:56:42] [PASSED] test_granular [07:56:42] [PASSED] test_unique [07:56:42] [PASSED] test_overlap [07:56:42] [PASSED] test_reusable [07:56:42] [PASSED] test_too_big [07:56:42] [PASSED] test_flush [07:56:42] [PASSED] test_lookup [07:56:42] [PASSED] test_data [07:56:42] [PASSED] test_class [07:56:42] ===================== [PASSED] guc_buf ===================== [07:56:42] =================== guc_dbm (7 subtests) =================== [07:56:42] [PASSED] test_empty [07:56:42] [PASSED] test_default [07:56:42] ======================== test_size ======================== [07:56:42] [PASSED] 4 [07:56:42] [PASSED] 8 [07:56:42] [PASSED] 32 [07:56:42] [PASSED] 256 [07:56:42] ==================== [PASSED] test_size ==================== [07:56:42] ======================= test_reuse ======================== [07:56:42] [PASSED] 4 [07:56:42] [PASSED] 8 [07:56:42] [PASSED] 32 [07:56:42] [PASSED] 256 [07:56:42] =================== [PASSED] test_reuse ==================== [07:56:42] =================== test_range_overlap ==================== [07:56:42] [PASSED] 4 [07:56:42] [PASSED] 8 [07:56:42] [PASSED] 32 [07:56:42] [PASSED] 256 [07:56:42] =============== [PASSED] test_range_overlap ================ [07:56:42] =================== test_range_compact ==================== [07:56:42] [PASSED] 4 [07:56:42] [PASSED] 8 [07:56:42] [PASSED] 32 [07:56:42] [PASSED] 256 [07:56:42] =============== [PASSED] test_range_compact ================ [07:56:42] ==================== test_range_spare ===================== [07:56:42] [PASSED] 4 [07:56:42] [PASSED] 8 [07:56:42] [PASSED] 32 [07:56:42] [PASSED] 256 [07:56:42] ================ [PASSED] test_range_spare ================= [07:56:42] ===================== [PASSED] guc_dbm ===================== [07:56:42] =================== guc_idm (6 subtests) =================== [07:56:42] [PASSED] bad_init [07:56:42] [PASSED] no_init [07:56:42] [PASSED] init_fini [07:56:42] [PASSED] check_used [07:56:42] [PASSED] check_quota [07:56:42] [PASSED] check_all [07:56:42] ===================== [PASSED] guc_idm ===================== [07:56:42] ================== no_relay (3 subtests) =================== [07:56:42] [PASSED] xe_drops_guc2pf_if_not_ready [07:56:42] [PASSED] xe_drops_guc2vf_if_not_ready [07:56:42] [PASSED] xe_rejects_send_if_not_ready [07:56:42] ==================== [PASSED] no_relay ===================== [07:56:42] ================== pf_relay (14 subtests) ================== [07:56:42] [PASSED] pf_rejects_guc2pf_too_short [07:56:42] [PASSED] pf_rejects_guc2pf_too_long [07:56:42] [PASSED] pf_rejects_guc2pf_no_payload [07:56:42] [PASSED] pf_fails_no_payload [07:56:42] [PASSED] pf_fails_bad_origin [07:56:42] [PASSED] pf_fails_bad_type [07:56:42] [PASSED] pf_txn_reports_error [07:56:42] [PASSED] pf_txn_sends_pf2guc [07:56:42] [PASSED] pf_sends_pf2guc [07:56:42] [SKIPPED] pf_loopback_nop [07:56:42] [SKIPPED] pf_loopback_echo [07:56:42] [SKIPPED] pf_loopback_fail [07:56:42] [SKIPPED] pf_loopback_busy [07:56:42] [SKIPPED] pf_loopback_retry [07:56:42] ==================== [PASSED] pf_relay ===================== [07:56:42] ================== vf_relay (3 subtests) =================== [07:56:42] [PASSED] vf_rejects_guc2vf_too_short [07:56:42] [PASSED] vf_rejects_guc2vf_too_long [07:56:42] [PASSED] vf_rejects_guc2vf_no_payload [07:56:42] ==================== [PASSED] vf_relay ===================== [07:56:42] ================ pf_gt_config (9 subtests) ================= [07:56:42] [PASSED] fair_contexts_1vf [07:56:42] [PASSED] fair_doorbells_1vf [07:56:42] [PASSED] fair_ggtt_1vf [07:56:42] ====================== fair_vram_1vf ====================== [07:56:42] [PASSED] 3.50 GiB [07:56:42] [PASSED] 11.5 GiB [07:56:42] [PASSED] 15.5 GiB [07:56:42] [PASSED] 31.5 GiB [07:56:42] [PASSED] 63.5 GiB [07:56:42] [PASSED] 1.91 GiB [07:56:42] ================== [PASSED] fair_vram_1vf ================== [07:56:42] ================ fair_vram_1vf_admin_only ================= [07:56:42] [PASSED] 3.50 GiB [07:56:42] [PASSED] 11.5 GiB [07:56:42] [PASSED] 15.5 GiB [07:56:42] [PASSED] 31.5 GiB [07:56:42] [PASSED] 63.5 GiB [07:56:42] [PASSED] 1.91 GiB [07:56:42] ============ [PASSED] fair_vram_1vf_admin_only ============= [07:56:42] ====================== fair_contexts ====================== [07:56:42] [PASSED] 1 VF [07:56:42] [PASSED] 2 VFs [07:56:42] [PASSED] 3 VFs [07:56:42] [PASSED] 4 VFs [07:56:42] [PASSED] 5 VFs [07:56:42] [PASSED] 6 VFs [07:56:42] [PASSED] 7 VFs [07:56:42] [PASSED] 8 VFs [07:56:42] [PASSED] 9 VFs [07:56:42] [PASSED] 10 VFs [07:56:42] [PASSED] 11 VFs [07:56:42] [PASSED] 12 VFs [07:56:42] [PASSED] 13 VFs [07:56:42] [PASSED] 14 VFs [07:56:42] [PASSED] 15 VFs [07:56:42] [PASSED] 16 VFs [07:56:42] [PASSED] 17 VFs [07:56:42] [PASSED] 18 VFs [07:56:42] [PASSED] 19 VFs [07:56:42] [PASSED] 20 VFs [07:56:42] [PASSED] 21 VFs [07:56:42] [PASSED] 22 VFs [07:56:42] [PASSED] 23 VFs [07:56:42] [PASSED] 24 VFs [07:56:42] [PASSED] 25 VFs [07:56:42] [PASSED] 26 VFs [07:56:42] [PASSED] 27 VFs [07:56:42] [PASSED] 28 VFs [07:56:42] [PASSED] 29 VFs [07:56:42] [PASSED] 30 VFs [07:56:42] [PASSED] 31 VFs [07:56:42] [PASSED] 32 VFs [07:56:42] [PASSED] 33 VFs [07:56:42] [PASSED] 34 VFs [07:56:42] [PASSED] 35 VFs [07:56:42] [PASSED] 36 VFs [07:56:42] [PASSED] 37 VFs [07:56:42] [PASSED] 38 VFs [07:56:42] [PASSED] 39 VFs [07:56:42] [PASSED] 40 VFs [07:56:42] [PASSED] 41 VFs [07:56:42] [PASSED] 42 VFs [07:56:42] [PASSED] 43 VFs [07:56:42] [PASSED] 44 VFs [07:56:42] [PASSED] 45 VFs [07:56:42] [PASSED] 46 VFs [07:56:42] [PASSED] 47 VFs [07:56:42] [PASSED] 48 VFs [07:56:42] [PASSED] 49 VFs [07:56:42] [PASSED] 50 VFs [07:56:42] [PASSED] 51 VFs [07:56:42] [PASSED] 52 VFs [07:56:42] [PASSED] 53 VFs [07:56:42] [PASSED] 54 VFs [07:56:42] [PASSED] 55 VFs [07:56:42] [PASSED] 56 VFs [07:56:42] [PASSED] 57 VFs [07:56:42] [PASSED] 58 VFs [07:56:42] [PASSED] 59 VFs [07:56:42] [PASSED] 60 VFs [07:56:42] [PASSED] 61 VFs [07:56:42] [PASSED] 62 VFs [07:56:42] [PASSED] 63 VFs [07:56:42] ================== [PASSED] fair_contexts ================== [07:56:42] ===================== fair_doorbells ====================== [07:56:42] [PASSED] 1 VF [07:56:42] [PASSED] 2 VFs [07:56:42] [PASSED] 3 VFs [07:56:42] [PASSED] 4 VFs [07:56:42] [PASSED] 5 VFs [07:56:42] [PASSED] 6 VFs [07:56:42] [PASSED] 7 VFs [07:56:42] [PASSED] 8 VFs [07:56:42] [PASSED] 9 VFs [07:56:42] [PASSED] 10 VFs [07:56:42] [PASSED] 11 VFs [07:56:42] [PASSED] 12 VFs [07:56:42] [PASSED] 13 VFs [07:56:42] [PASSED] 14 VFs [07:56:42] [PASSED] 15 VFs [07:56:42] [PASSED] 16 VFs [07:56:42] [PASSED] 17 VFs [07:56:42] [PASSED] 18 VFs [07:56:42] [PASSED] 19 VFs [07:56:42] [PASSED] 20 VFs [07:56:42] [PASSED] 21 VFs [07:56:42] [PASSED] 22 VFs [07:56:42] [PASSED] 23 VFs [07:56:42] [PASSED] 24 VFs [07:56:42] [PASSED] 25 VFs [07:56:42] [PASSED] 26 VFs [07:56:42] [PASSED] 27 VFs [07:56:42] [PASSED] 28 VFs [07:56:42] [PASSED] 29 VFs [07:56:42] [PASSED] 30 VFs [07:56:42] [PASSED] 31 VFs [07:56:42] [PASSED] 32 VFs [07:56:42] [PASSED] 33 VFs [07:56:42] [PASSED] 34 VFs [07:56:42] [PASSED] 35 VFs [07:56:42] [PASSED] 36 VFs [07:56:42] [PASSED] 37 VFs [07:56:42] [PASSED] 38 VFs [07:56:42] [PASSED] 39 VFs [07:56:42] [PASSED] 40 VFs [07:56:42] [PASSED] 41 VFs [07:56:42] [PASSED] 42 VFs [07:56:42] [PASSED] 43 VFs [07:56:42] [PASSED] 44 VFs [07:56:42] [PASSED] 45 VFs [07:56:42] [PASSED] 46 VFs [07:56:42] [PASSED] 47 VFs [07:56:42] [PASSED] 48 VFs [07:56:42] [PASSED] 49 VFs [07:56:42] [PASSED] 50 VFs [07:56:42] [PASSED] 51 VFs [07:56:42] [PASSED] 52 VFs [07:56:42] [PASSED] 53 VFs [07:56:42] [PASSED] 54 VFs [07:56:42] [PASSED] 55 VFs [07:56:42] [PASSED] 56 VFs [07:56:42] [PASSED] 57 VFs [07:56:42] [PASSED] 58 VFs [07:56:42] [PASSED] 59 VFs [07:56:42] [PASSED] 60 VFs [07:56:42] [PASSED] 61 VFs [07:56:42] [PASSED] 62 VFs [07:56:42] [PASSED] 63 VFs [07:56:42] ================= [PASSED] fair_doorbells ================== [07:56:42] ======================== fair_ggtt ======================== [07:56:42] [PASSED] 1 VF [07:56:42] [PASSED] 2 VFs [07:56:42] [PASSED] 3 VFs [07:56:42] [PASSED] 4 VFs [07:56:42] [PASSED] 5 VFs [07:56:42] [PASSED] 6 VFs [07:56:42] [PASSED] 7 VFs [07:56:42] [PASSED] 8 VFs [07:56:42] [PASSED] 9 VFs [07:56:42] [PASSED] 10 VFs [07:56:42] [PASSED] 11 VFs [07:56:42] [PASSED] 12 VFs [07:56:42] [PASSED] 13 VFs [07:56:42] [PASSED] 14 VFs [07:56:42] [PASSED] 15 VFs [07:56:42] [PASSED] 16 VFs [07:56:42] [PASSED] 17 VFs [07:56:42] [PASSED] 18 VFs [07:56:42] [PASSED] 19 VFs [07:56:42] [PASSED] 20 VFs [07:56:42] [PASSED] 21 VFs [07:56:42] [PASSED] 22 VFs [07:56:42] [PASSED] 23 VFs [07:56:42] [PASSED] 24 VFs [07:56:42] [PASSED] 25 VFs [07:56:42] [PASSED] 26 VFs [07:56:42] [PASSED] 27 VFs [07:56:42] [PASSED] 28 VFs [07:56:42] [PASSED] 29 VFs [07:56:42] [PASSED] 30 VFs [07:56:42] [PASSED] 31 VFs [07:56:42] [PASSED] 32 VFs [07:56:42] [PASSED] 33 VFs [07:56:42] [PASSED] 34 VFs [07:56:42] [PASSED] 35 VFs [07:56:42] [PASSED] 36 VFs [07:56:42] [PASSED] 37 VFs [07:56:42] [PASSED] 38 VFs [07:56:42] [PASSED] 39 VFs [07:56:42] [PASSED] 40 VFs [07:56:42] [PASSED] 41 VFs [07:56:42] [PASSED] 42 VFs [07:56:42] [PASSED] 43 VFs [07:56:42] [PASSED] 44 VFs [07:56:42] [PASSED] 45 VFs [07:56:42] [PASSED] 46 VFs [07:56:42] [PASSED] 47 VFs [07:56:42] [PASSED] 48 VFs [07:56:42] [PASSED] 49 VFs [07:56:42] [PASSED] 50 VFs [07:56:42] [PASSED] 51 VFs [07:56:42] [PASSED] 52 VFs [07:56:42] [PASSED] 53 VFs [07:56:42] [PASSED] 54 VFs [07:56:42] [PASSED] 55 VFs [07:56:42] [PASSED] 56 VFs [07:56:42] [PASSED] 57 VFs [07:56:42] [PASSED] 58 VFs [07:56:42] [PASSED] 59 VFs [07:56:42] [PASSED] 60 VFs [07:56:42] [PASSED] 61 VFs [07:56:42] [PASSED] 62 VFs [07:56:42] [PASSED] 63 VFs [07:56:42] ==================== [PASSED] fair_ggtt ==================== [07:56:42] ======================== fair_vram ======================== [07:56:42] [PASSED] 1 VF [07:56:42] [PASSED] 2 VFs [07:56:42] [PASSED] 3 VFs [07:56:42] [PASSED] 4 VFs [07:56:42] [PASSED] 5 VFs [07:56:42] [PASSED] 6 VFs [07:56:42] [PASSED] 7 VFs [07:56:42] [PASSED] 8 VFs [07:56:42] [PASSED] 9 VFs [07:56:42] [PASSED] 10 VFs [07:56:42] [PASSED] 11 VFs [07:56:42] [PASSED] 12 VFs [07:56:42] [PASSED] 13 VFs [07:56:42] [PASSED] 14 VFs [07:56:42] [PASSED] 15 VFs [07:56:42] [PASSED] 16 VFs [07:56:42] [PASSED] 17 VFs [07:56:42] [PASSED] 18 VFs [07:56:42] [PASSED] 19 VFs [07:56:42] [PASSED] 20 VFs [07:56:42] [PASSED] 21 VFs [07:56:42] [PASSED] 22 VFs [07:56:42] [PASSED] 23 VFs [07:56:42] [PASSED] 24 VFs [07:56:42] [PASSED] 25 VFs [07:56:42] [PASSED] 26 VFs [07:56:42] [PASSED] 27 VFs [07:56:42] [PASSED] 28 VFs [07:56:42] [PASSED] 29 VFs [07:56:42] [PASSED] 30 VFs [07:56:42] [PASSED] 31 VFs [07:56:42] [PASSED] 32 VFs [07:56:42] [PASSED] 33 VFs [07:56:42] [PASSED] 34 VFs [07:56:42] [PASSED] 35 VFs [07:56:42] [PASSED] 36 VFs [07:56:42] [PASSED] 37 VFs [07:56:42] [PASSED] 38 VFs [07:56:42] [PASSED] 39 VFs [07:56:42] [PASSED] 40 VFs [07:56:42] [PASSED] 41 VFs [07:56:42] [PASSED] 42 VFs [07:56:42] [PASSED] 43 VFs [07:56:42] [PASSED] 44 VFs [07:56:42] [PASSED] 45 VFs [07:56:42] [PASSED] 46 VFs [07:56:42] [PASSED] 47 VFs [07:56:42] [PASSED] 48 VFs [07:56:42] [PASSED] 49 VFs [07:56:42] [PASSED] 50 VFs [07:56:42] [PASSED] 51 VFs [07:56:42] [PASSED] 52 VFs [07:56:42] [PASSED] 53 VFs [07:56:43] [PASSED] 54 VFs [07:56:43] [PASSED] 55 VFs [07:56:43] [PASSED] 56 VFs [07:56:43] [PASSED] 57 VFs [07:56:43] [PASSED] 58 VFs [07:56:43] [PASSED] 59 VFs [07:56:43] [PASSED] 60 VFs [07:56:43] [PASSED] 61 VFs [07:56:43] [PASSED] 62 VFs [07:56:43] [PASSED] 63 VFs [07:56:43] ==================== [PASSED] fair_vram ==================== [07:56:43] ================== [PASSED] pf_gt_config =================== [07:56:43] ===================== lmtt (1 subtest) ===================== [07:56:43] ======================== test_ops ========================= [07:56:43] [PASSED] 2-level [07:56:43] [PASSED] multi-level [07:56:43] ==================== [PASSED] test_ops ===================== [07:56:43] ====================== [PASSED] lmtt ======================= [07:56:43] ================= pf_service (11 subtests) ================= [07:56:43] [PASSED] pf_negotiate_any [07:56:43] [PASSED] pf_negotiate_base_match [07:56:43] [PASSED] pf_negotiate_base_newer [07:56:43] [PASSED] pf_negotiate_base_next [07:56:43] [SKIPPED] pf_negotiate_base_older [07:56:43] [PASSED] pf_negotiate_base_prev [07:56:43] [PASSED] pf_negotiate_latest_match [07:56:43] [PASSED] pf_negotiate_latest_newer [07:56:43] [PASSED] pf_negotiate_latest_next [07:56:43] [SKIPPED] pf_negotiate_latest_older [07:56:43] [SKIPPED] pf_negotiate_latest_prev [07:56:43] =================== [PASSED] pf_service ==================== [07:56:43] ================= xe_guc_g2g (2 subtests) ================== [07:56:43] ============== xe_live_guc_g2g_kunit_default ============== [07:56:43] ========= [SKIPPED] xe_live_guc_g2g_kunit_default ========== [07:56:43] ============== xe_live_guc_g2g_kunit_allmem =============== [07:56:43] ========== [SKIPPED] xe_live_guc_g2g_kunit_allmem ========== [07:56:43] =================== [SKIPPED] xe_guc_g2g =================== [07:56:43] =================== xe_mocs (2 subtests) =================== [07:56:43] ================ xe_live_mocs_kernel_kunit ================ [07:56:43] =========== [SKIPPED] xe_live_mocs_kernel_kunit ============ [07:56:43] ================ xe_live_mocs_reset_kunit ================= [07:56:43] ============ [SKIPPED] xe_live_mocs_reset_kunit ============ [07:56:43] ==================== [SKIPPED] xe_mocs ===================== [07:56:43] ================= xe_migrate (2 subtests) ================== [07:56:43] ================= xe_migrate_sanity_kunit ================= [07:56:43] ============ [SKIPPED] xe_migrate_sanity_kunit ============= [07:56:43] ================== xe_validate_ccs_kunit ================== [07:56:43] ============= [SKIPPED] xe_validate_ccs_kunit ============== [07:56:43] =================== [SKIPPED] xe_migrate =================== [07:56:43] ================== xe_dma_buf (1 subtest) ================== [07:56:43] ==================== xe_dma_buf_kunit ===================== [07:56:43] ================ [SKIPPED] xe_dma_buf_kunit ================ [07:56:43] =================== [SKIPPED] xe_dma_buf =================== [07:56:43] ================= xe_bo_shrink (1 subtest) ================= [07:56:43] =================== xe_bo_shrink_kunit ==================== [07:56:43] =============== [SKIPPED] xe_bo_shrink_kunit =============== [07:56:43] ================== [SKIPPED] xe_bo_shrink ================== [07:56:43] ==================== xe_bo (2 subtests) ==================== [07:56:43] ================== xe_ccs_migrate_kunit =================== [07:56:43] ============== [SKIPPED] xe_ccs_migrate_kunit ============== [07:56:43] ==================== xe_bo_evict_kunit ==================== [07:56:43] =============== [SKIPPED] xe_bo_evict_kunit ================ [07:56:43] ===================== [SKIPPED] xe_bo ====================== [07:56:43] ==================== args (13 subtests) ==================== [07:56:43] [PASSED] count_args_test [07:56:43] [PASSED] call_args_example [07:56:43] [PASSED] call_args_test [07:56:43] [PASSED] drop_first_arg_example [07:56:43] [PASSED] drop_first_arg_test [07:56:43] [PASSED] first_arg_example [07:56:43] [PASSED] first_arg_test [07:56:43] [PASSED] last_arg_example [07:56:43] [PASSED] last_arg_test [07:56:43] [PASSED] pick_arg_example [07:56:43] [PASSED] if_args_example [07:56:43] [PASSED] if_args_test [07:56:43] [PASSED] sep_comma_example [07:56:43] ====================== [PASSED] args ======================= [07:56:43] =================== xe_pci (3 subtests) ==================== [07:56:43] ==================== check_graphics_ip ==================== [07:56:43] [PASSED] 12.00 Xe_LP [07:56:43] [PASSED] 12.10 Xe_LP+ [07:56:43] [PASSED] 12.55 Xe_HPG [07:56:43] [PASSED] 12.60 Xe_HPC [07:56:43] [PASSED] 12.70 Xe_LPG [07:56:43] [PASSED] 12.71 Xe_LPG [07:56:43] [PASSED] 12.74 Xe_LPG+ [07:56:43] [PASSED] 20.01 Xe2_HPG [07:56:43] [PASSED] 20.02 Xe2_HPG [07:56:43] [PASSED] 20.04 Xe2_LPG [07:56:43] [PASSED] 30.00 Xe3_LPG [07:56:43] [PASSED] 30.01 Xe3_LPG [07:56:43] [PASSED] 30.03 Xe3_LPG [07:56:43] [PASSED] 30.04 Xe3_LPG [07:56:43] [PASSED] 30.05 Xe3_LPG [07:56:43] [PASSED] 35.10 Xe3p_LPG [07:56:43] [PASSED] 35.11 Xe3p_XPC [07:56:43] ================ [PASSED] check_graphics_ip ================ [07:56:43] ===================== check_media_ip ====================== [07:56:43] [PASSED] 12.00 Xe_M [07:56:43] [PASSED] 12.55 Xe_HPM [07:56:43] [PASSED] 13.00 Xe_LPM+ [07:56:43] [PASSED] 13.01 Xe2_HPM [07:56:43] [PASSED] 20.00 Xe2_LPM [07:56:43] [PASSED] 30.00 Xe3_LPM [07:56:43] [PASSED] 30.02 Xe3_LPM [07:56:43] [PASSED] 35.00 Xe3p_LPM [07:56:43] [PASSED] 35.03 Xe3p_HPM [07:56:43] ================= [PASSED] check_media_ip ================== [07:56:43] =================== check_platform_desc =================== [07:56:43] [PASSED] 0x9A60 (TIGERLAKE) [07:56:43] [PASSED] 0x9A68 (TIGERLAKE) [07:56:43] [PASSED] 0x9A70 (TIGERLAKE) [07:56:43] [PASSED] 0x9A40 (TIGERLAKE) [07:56:43] [PASSED] 0x9A49 (TIGERLAKE) [07:56:43] [PASSED] 0x9A59 (TIGERLAKE) [07:56:43] [PASSED] 0x9A78 (TIGERLAKE) [07:56:43] [PASSED] 0x9AC0 (TIGERLAKE) [07:56:43] [PASSED] 0x9AC9 (TIGERLAKE) [07:56:43] [PASSED] 0x9AD9 (TIGERLAKE) [07:56:43] [PASSED] 0x9AF8 (TIGERLAKE) [07:56:43] [PASSED] 0x4C80 (ROCKETLAKE) [07:56:43] [PASSED] 0x4C8A (ROCKETLAKE) [07:56:43] [PASSED] 0x4C8B (ROCKETLAKE) [07:56:43] [PASSED] 0x4C8C (ROCKETLAKE) [07:56:43] [PASSED] 0x4C90 (ROCKETLAKE) [07:56:43] [PASSED] 0x4C9A (ROCKETLAKE) [07:56:43] [PASSED] 0x4680 (ALDERLAKE_S) [07:56:43] [PASSED] 0x4682 (ALDERLAKE_S) [07:56:43] [PASSED] 0x4688 (ALDERLAKE_S) [07:56:43] [PASSED] 0x468A (ALDERLAKE_S) [07:56:43] [PASSED] 0x468B (ALDERLAKE_S) [07:56:43] [PASSED] 0x4690 (ALDERLAKE_S) [07:56:43] [PASSED] 0x4692 (ALDERLAKE_S) [07:56:43] [PASSED] 0x4693 (ALDERLAKE_S) [07:56:43] [PASSED] 0x46A0 (ALDERLAKE_P) [07:56:43] [PASSED] 0x46A1 (ALDERLAKE_P) [07:56:43] [PASSED] 0x46A2 (ALDERLAKE_P) [07:56:43] [PASSED] 0x46A3 (ALDERLAKE_P) [07:56:43] [PASSED] 0x46A6 (ALDERLAKE_P) [07:56:43] [PASSED] 0x46A8 (ALDERLAKE_P) [07:56:43] [PASSED] 0x46AA (ALDERLAKE_P) [07:56:43] [PASSED] 0x462A (ALDERLAKE_P) [07:56:43] [PASSED] 0x4626 (ALDERLAKE_P) [07:56:43] [PASSED] 0x4628 (ALDERLAKE_P) [07:56:43] [PASSED] 0x46B0 (ALDERLAKE_P) [07:56:43] [PASSED] 0x46B1 (ALDERLAKE_P) [07:56:43] [PASSED] 0x46B2 (ALDERLAKE_P) [07:56:43] [PASSED] 0x46B3 (ALDERLAKE_P) [07:56:43] [PASSED] 0x46C0 (ALDERLAKE_P) [07:56:43] [PASSED] 0x46C1 (ALDERLAKE_P) [07:56:43] [PASSED] 0x46C2 (ALDERLAKE_P) [07:56:43] [PASSED] 0x46C3 (ALDERLAKE_P) [07:56:43] [PASSED] 0x46D0 (ALDERLAKE_N) [07:56:43] [PASSED] 0x46D1 (ALDERLAKE_N) [07:56:43] [PASSED] 0x46D2 (ALDERLAKE_N) [07:56:43] [PASSED] 0x46D3 (ALDERLAKE_N) [07:56:43] [PASSED] 0x46D4 (ALDERLAKE_N) [07:56:43] [PASSED] 0xA721 (ALDERLAKE_P) [07:56:43] [PASSED] 0xA7A1 (ALDERLAKE_P) [07:56:43] [PASSED] 0xA7A9 (ALDERLAKE_P) [07:56:43] [PASSED] 0xA7AC (ALDERLAKE_P) [07:56:43] [PASSED] 0xA7AD (ALDERLAKE_P) [07:56:43] [PASSED] 0xA720 (ALDERLAKE_P) [07:56:43] [PASSED] 0xA7A0 (ALDERLAKE_P) [07:56:43] [PASSED] 0xA7A8 (ALDERLAKE_P) [07:56:43] [PASSED] 0xA7AA (ALDERLAKE_P) [07:56:43] [PASSED] 0xA7AB (ALDERLAKE_P) [07:56:43] [PASSED] 0xA780 (ALDERLAKE_S) [07:56:43] [PASSED] 0xA781 (ALDERLAKE_S) [07:56:43] [PASSED] 0xA782 (ALDERLAKE_S) [07:56:43] [PASSED] 0xA783 (ALDERLAKE_S) [07:56:43] [PASSED] 0xA788 (ALDERLAKE_S) [07:56:43] [PASSED] 0xA789 (ALDERLAKE_S) [07:56:43] [PASSED] 0xA78A (ALDERLAKE_S) [07:56:43] [PASSED] 0xA78B (ALDERLAKE_S) [07:56:43] [PASSED] 0x4905 (DG1) [07:56:43] [PASSED] 0x4906 (DG1) [07:56:43] [PASSED] 0x4907 (DG1) [07:56:43] [PASSED] 0x4908 (DG1) [07:56:43] [PASSED] 0x4909 (DG1) [07:56:43] [PASSED] 0x56C0 (DG2) [07:56:43] [PASSED] 0x56C2 (DG2) [07:56:43] [PASSED] 0x56C1 (DG2) [07:56:43] [PASSED] 0x7D51 (METEORLAKE) [07:56:43] [PASSED] 0x7DD1 (METEORLAKE) [07:56:43] [PASSED] 0x7D41 (METEORLAKE) [07:56:43] [PASSED] 0x7D67 (METEORLAKE) [07:56:43] [PASSED] 0xB640 (METEORLAKE) [07:56:43] [PASSED] 0x56A0 (DG2) [07:56:43] [PASSED] 0x56A1 (DG2) [07:56:43] [PASSED] 0x56A2 (DG2) [07:56:43] [PASSED] 0x56BE (DG2) [07:56:43] [PASSED] 0x56BF (DG2) [07:56:43] [PASSED] 0x5690 (DG2) [07:56:43] [PASSED] 0x5691 (DG2) [07:56:43] [PASSED] 0x5692 (DG2) [07:56:43] [PASSED] 0x56A5 (DG2) [07:56:43] [PASSED] 0x56A6 (DG2) [07:56:43] [PASSED] 0x56B0 (DG2) [07:56:43] [PASSED] 0x56B1 (DG2) [07:56:43] [PASSED] 0x56BA (DG2) [07:56:43] [PASSED] 0x56BB (DG2) [07:56:43] [PASSED] 0x56BC (DG2) [07:56:43] [PASSED] 0x56BD (DG2) [07:56:43] [PASSED] 0x5693 (DG2) [07:56:43] [PASSED] 0x5694 (DG2) [07:56:43] [PASSED] 0x5695 (DG2) [07:56:43] [PASSED] 0x56A3 (DG2) [07:56:43] [PASSED] 0x56A4 (DG2) [07:56:43] [PASSED] 0x56B2 (DG2) [07:56:43] [PASSED] 0x56B3 (DG2) [07:56:43] [PASSED] 0x5696 (DG2) [07:56:43] [PASSED] 0x5697 (DG2) [07:56:43] [PASSED] 0xB69 (PVC) [07:56:43] [PASSED] 0xB6E (PVC) [07:56:43] [PASSED] 0xBD4 (PVC) [07:56:43] [PASSED] 0xBD5 (PVC) [07:56:43] [PASSED] 0xBD6 (PVC) [07:56:43] [PASSED] 0xBD7 (PVC) [07:56:43] [PASSED] 0xBD8 (PVC) [07:56:43] [PASSED] 0xBD9 (PVC) [07:56:43] [PASSED] 0xBDA (PVC) [07:56:43] [PASSED] 0xBDB (PVC) [07:56:43] [PASSED] 0xBE0 (PVC) [07:56:43] [PASSED] 0xBE1 (PVC) [07:56:43] [PASSED] 0xBE5 (PVC) [07:56:43] [PASSED] 0x7D40 (METEORLAKE) [07:56:43] [PASSED] 0x7D45 (METEORLAKE) [07:56:43] [PASSED] 0x7D55 (METEORLAKE) [07:56:43] [PASSED] 0x7D60 (METEORLAKE) [07:56:43] [PASSED] 0x7DD5 (METEORLAKE) [07:56:43] [PASSED] 0x6420 (LUNARLAKE) [07:56:43] [PASSED] 0x64A0 (LUNARLAKE) [07:56:43] [PASSED] 0x64B0 (LUNARLAKE) [07:56:43] [PASSED] 0xE202 (BATTLEMAGE) [07:56:43] [PASSED] 0xE209 (BATTLEMAGE) [07:56:43] [PASSED] 0xE20B (BATTLEMAGE) [07:56:43] [PASSED] 0xE20C (BATTLEMAGE) [07:56:43] [PASSED] 0xE20D (BATTLEMAGE) [07:56:43] [PASSED] 0xE210 (BATTLEMAGE) [07:56:43] [PASSED] 0xE211 (BATTLEMAGE) [07:56:43] [PASSED] 0xE212 (BATTLEMAGE) [07:56:43] [PASSED] 0xE216 (BATTLEMAGE) [07:56:43] [PASSED] 0xE220 (BATTLEMAGE) [07:56:43] [PASSED] 0xE221 (BATTLEMAGE) [07:56:43] [PASSED] 0xE222 (BATTLEMAGE) [07:56:43] [PASSED] 0xE223 (BATTLEMAGE) [07:56:43] [PASSED] 0xB080 (PANTHERLAKE) [07:56:43] [PASSED] 0xB081 (PANTHERLAKE) [07:56:43] [PASSED] 0xB082 (PANTHERLAKE) [07:56:43] [PASSED] 0xB083 (PANTHERLAKE) [07:56:43] [PASSED] 0xB084 (PANTHERLAKE) [07:56:43] [PASSED] 0xB085 (PANTHERLAKE) [07:56:43] [PASSED] 0xB086 (PANTHERLAKE) [07:56:43] [PASSED] 0xB087 (PANTHERLAKE) [07:56:43] [PASSED] 0xB08F (PANTHERLAKE) [07:56:43] [PASSED] 0xB090 (PANTHERLAKE) [07:56:43] [PASSED] 0xB0A0 (PANTHERLAKE) [07:56:43] [PASSED] 0xB0B0 (PANTHERLAKE) [07:56:43] [PASSED] 0xFD80 (PANTHERLAKE) [07:56:43] [PASSED] 0xFD81 (PANTHERLAKE) [07:56:43] [PASSED] 0xD740 (NOVALAKE_S) [07:56:43] [PASSED] 0xD741 (NOVALAKE_S) [07:56:43] [PASSED] 0xD742 (NOVALAKE_S) [07:56:43] [PASSED] 0xD743 (NOVALAKE_S) [07:56:43] [PASSED] 0xD744 (NOVALAKE_S) [07:56:43] [PASSED] 0xD745 (NOVALAKE_S) [07:56:43] [PASSED] 0x674C (CRESCENTISLAND) [07:56:43] [PASSED] 0xD750 (NOVALAKE_P) [07:56:43] [PASSED] 0xD751 (NOVALAKE_P) [07:56:43] [PASSED] 0xD752 (NOVALAKE_P) [07:56:43] [PASSED] 0xD753 (NOVALAKE_P) [07:56:43] [PASSED] 0xD754 (NOVALAKE_P) [07:56:43] [PASSED] 0xD755 (NOVALAKE_P) [07:56:43] [PASSED] 0xD756 (NOVALAKE_P) [07:56:43] [PASSED] 0xD757 (NOVALAKE_P) [07:56:43] [PASSED] 0xD75F (NOVALAKE_P) [07:56:43] =============== [PASSED] check_platform_desc =============== [07:56:43] ===================== [PASSED] xe_pci ====================== [07:56:43] =================== xe_rtp (2 subtests) ==================== [07:56:43] =============== xe_rtp_process_to_sr_tests ================ [07:56:43] [PASSED] coalesce-same-reg [07:56:43] [PASSED] no-match-no-add [07:56:43] [PASSED] match-or [07:56:43] [PASSED] match-or-xfail [07:56:43] [PASSED] no-match-no-add-multiple-rules [07:56:43] [PASSED] two-regs-two-entries [07:56:43] [PASSED] clr-one-set-other [07:56:43] [PASSED] set-field [07:56:43] [PASSED] conflict-duplicate stty: 'standard input': Inappropriate ioctl for device [07:56:43] [PASSED] conflict-not-disjoint [07:56:43] [PASSED] conflict-reg-type [07:56:43] =========== [PASSED] xe_rtp_process_to_sr_tests ============ [07:56:43] ================== xe_rtp_process_tests =================== [07:56:43] [PASSED] active1 [07:56:43] [PASSED] active2 [07:56:43] [PASSED] active-inactive [07:56:43] [PASSED] inactive-active [07:56:43] [PASSED] inactive-1st_or_active-inactive [07:56:43] [PASSED] inactive-2nd_or_active-inactive [07:56:43] [PASSED] inactive-last_or_active-inactive [07:56:43] [PASSED] inactive-no_or_active-inactive [07:56:43] ============== [PASSED] xe_rtp_process_tests =============== [07:56:43] ===================== [PASSED] xe_rtp ====================== [07:56:43] ==================== xe_wa (1 subtest) ===================== [07:56:43] ======================== xe_wa_gt ========================= [07:56:43] [PASSED] TIGERLAKE B0 [07:56:43] [PASSED] DG1 A0 [07:56:43] [PASSED] DG1 B0 [07:56:43] [PASSED] ALDERLAKE_S A0 [07:56:43] [PASSED] ALDERLAKE_S B0 [07:56:43] [PASSED] ALDERLAKE_S C0 [07:56:43] [PASSED] ALDERLAKE_S D0 [07:56:43] [PASSED] ALDERLAKE_P A0 [07:56:43] [PASSED] ALDERLAKE_P B0 [07:56:43] [PASSED] ALDERLAKE_P C0 [07:56:43] [PASSED] ALDERLAKE_S RPLS D0 [07:56:43] [PASSED] ALDERLAKE_P RPLU E0 [07:56:43] [PASSED] DG2 G10 C0 [07:56:43] [PASSED] DG2 G11 B1 [07:56:43] [PASSED] DG2 G12 A1 [07:56:43] [PASSED] METEORLAKE 12.70(Xe_LPG) A0 13.00(Xe_LPM+) A0 [07:56:43] [PASSED] METEORLAKE 12.71(Xe_LPG) A0 13.00(Xe_LPM+) A0 [07:56:43] [PASSED] METEORLAKE 12.74(Xe_LPG+) A0 13.00(Xe_LPM+) A0 [07:56:43] [PASSED] LUNARLAKE 20.04(Xe2_LPG) A0 20.00(Xe2_LPM) A0 [07:56:43] [PASSED] LUNARLAKE 20.04(Xe2_LPG) B0 20.00(Xe2_LPM) A0 [07:56:43] [PASSED] BATTLEMAGE 20.01(Xe2_HPG) A0 13.01(Xe2_HPM) A1 [07:56:43] [PASSED] PANTHERLAKE 30.00(Xe3_LPG) A0 30.00(Xe3_LPM) A0 [07:56:43] ==================== [PASSED] xe_wa_gt ===================== [07:56:43] ====================== [PASSED] xe_wa ====================== [07:56:43] ============================================================ [07:56:43] Testing complete. Ran 597 tests: passed: 579, skipped: 18 [07:56:43] Elapsed time: 36.067s total, 4.298s configuring, 31.152s building, 0.608s running
+ /kernel/tools/testing/kunit/kunit.py run --kunitconfig /kernel/drivers/gpu/drm/tests/.kunitconfig ERROR:root:In file included from ../drivers/gpu/buddy.c:12:../include/linux/gpu_buddy.h:220:15: error: return type defaults to ‘int’ [-Werror=implicit-int]
220 | static inline gpu_buddy_driver_lock_held(struct gpu_buddy *mm)
| ^~~~~~~~~~~~~~~~~~~~~~~~~~
../include/linux/gpu_buddy.h: In function ‘gpu_buddy_driver_lock_held’:
../include/linux/gpu_buddy.h:222:1: error: control reaches end of non-void function [-Werror=return-type]
222 | }
| ^cc1: some warnings being treated as errors
make[5]: *** [../scripts/Makefile.build:289: drivers/gpu/buddy.o] Error 1 make[5]: *** Waiting for unfinished jobs....
In file included from ../drivers/gpu/drm/drm_buddy.c:13:../include/linux/gpu_buddy.h:220:15: error: return type defaults to ‘int’ [-Werror=implicit-int]
220 | static inline gpu_buddy_driver_lock_held(struct gpu_buddy *mm)
| ^~~~~~~~~~~~~~~~~~~~~~~~~~
../include/linux/gpu_buddy.h: In function ‘gpu_buddy_driver_lock_held’:
../include/linux/gpu_buddy.h:222:1: error: control reaches end of non-void function [-Werror=return-type]
222 | }
| ^cc1: some warnings being treated as errors
make[6]: *** [../scripts/Makefile.build:289: drivers/gpu/drm/drm_buddy.o] Error 1 make[6]: *** Waiting for unfinished jobs....
make[5]: *** [../scripts/Makefile.build:549: drivers/gpu/drm] Error 2 make[4]: *** [../scripts/Makefile.build:549: drivers/gpu] Error 2 make[3]: *** [../scripts/Makefile.build:549: drivers] Error 2 make[3]: *** Waiting for unfinished jobs....
make[2]: *** [/kernel/Makefile:2105: .] Error 2
make[1]: *** [/kernel/Makefile:248: __sub-make] Error 2
make: *** [Makefile:248: __sub-make] Error 2
[07:56:43] Configuring KUnit Kernel ...
Regenerating .config ...
Populating config with:
$ make ARCH=um O=.kunit olddefconfig [07:56:44] Building KUnit Kernel ... Populating config with: $ make ARCH=um O=.kunit olddefconfig Building with: $ make all compile_commands.json scripts_gdb ARCH=um O=.kunit --jobs=48 + cleanup ++ stat -c %u:%g /kernel + chown -R 1003:1003 /kernel