Skip to contentSkip to conversation

drm/amdkfd: Add SVM range mapped_to_gpu flag

PATCH In review Philip Yang 3 participants 8 messages Apr 20, 2022 – Apr 25, 2022

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

PATCH 1/2 drm/amdkfd: Add SVM range mapped_to_gpu flag

To avoid unnecessary unmap SVM range from GPUs if range is not mapped on GPUs when migrating the range. This flag will also be used to flush TLB when updating the existing mapping on GPUs.

It is protected by prange->migrate_mutex and mmap read lock in MMU notifier callback.

---
 drivers/gpu/drm/amd/amdkfd/kfd_svm.c | 17 ++++++++++++++++-
 drivers/gpu/drm/amd/amdkfd/kfd_svm.h |  1 +
 2 files changed, 17 insertions(+), 1 deletion(-)
drivers/gpu/drm/amd/amdkfd/kfd_svm.c +16 -1
index 5afe216cf099..6be1695f3a09 100644
@@ -951,6 +951,7 @@ svm_range_split_adjust(struct svm_range *new, struct svm_range *old,
 	new->prefetch_loc = old->prefetch_loc;
 	new->actual_loc = old->actual_loc;
 	new->granularity = old->granularity;
+	new->mapped_to_gpu = old->mapped_to_gpu;
 	bitmap_copy(new->bitmap_access, old->bitmap_access, MAX_GPU_INSTANCE);
 	bitmap_copy(new->bitmap_aip, old->bitmap_aip, MAX_GPU_INSTANCE);
 
@@ -1204,6 +1205,17 @@ svm_range_unmap_from_gpus(struct svm_range *prange, unsigned long start,
 	uint32_t gpuidx;
 	int r = 0;
 
+	if (!prange->mapped_to_gpu) {
+		pr_debug("prange 0x%p [0x%lx 0x%lx] not mapped to GPU\n",
+			 prange, prange->start, prange->last);
+		return 0;
+	}
+
+	if (prange->start == start && prange->last == last) {
+		pr_debug("unmap svms 0x%p prange 0x%p\n", prange->svms, prange);
+		prange->mapped_to_gpu = false;
+	}
+
 	bitmap_or(bitmap, prange->bitmap_access, prange->bitmap_aip,
 		  MAX_GPU_INSTANCE);
 	p = container_of(prange->svms, struct kfd_process, svms);
@@ -1590,8 +1602,10 @@ static int svm_range_validate_and_map(struct mm_struct *mm,
 		addr = next;
 	}
 
-	if (addr == end)
+	if (addr == end) {
 		prange->validated_once = true;
+		prange->mapped_to_gpu = true;
+	}
 
 unreserve_out:
 	svm_range_unreserve_bos(&ctx);
@@ -1822,6 +1836,7 @@ static struct svm_range *svm_range_clone(struct svm_range *old)
 	new->prefetch_loc = old->prefetch_loc;
 	new->actual_loc = old->actual_loc;
 	new->granularity = old->granularity;
+	new->mapped_to_gpu = old->mapped_to_gpu;
 	bitmap_copy(new->bitmap_access, old->bitmap_access, MAX_GPU_INSTANCE);
 	bitmap_copy(new->bitmap_aip, old->bitmap_aip, MAX_GPU_INSTANCE);
 
drivers/gpu/drm/amd/amdkfd/kfd_svm.h +1 -0
index 66c77f00ac3e..2d54147b4dda 100644
@@ -133,6 +133,7 @@ struct svm_range {
 	DECLARE_BITMAP(bitmap_access, MAX_GPU_INSTANCE);
 	DECLARE_BITMAP(bitmap_aip, MAX_GPU_INSTANCE);
 	bool				validated_once;
+	bool				mapped_to_gpu;
 };
 
 static inline void svm_range_lock(struct svm_range *prange)
Signature · 1 line
2.35.1
#1 lore raw

PATCH 2/2 drm/amdkfd: Update mapping if range attributes changed

Change SVM range mapping flags or access attributes don't trigger migration, if range is already mapped on GPUs we should update GPU mapping, and pass flush_tlb flag to amdgpu vm.

---
 drivers/gpu/drm/amd/amdkfd/kfd_svm.c | 41 ++++++++++++++++++----------
 1 file changed, 27 insertions(+), 14 deletions(-)
drivers/gpu/drm/amd/amdkfd/kfd_svm.c +27 -14
index 6be1695f3a09..0a9bdadf875e 100644
@@ -686,7 +686,8 @@ svm_range_check_attr(struct kfd_process *p,
 
 static void
 svm_range_apply_attrs(struct kfd_process *p, struct svm_range *prange,
-		      uint32_t nattr, struct kfd_ioctl_svm_attribute *attrs)
+		      uint32_t nattr, struct kfd_ioctl_svm_attribute *attrs,
+		      bool *update_mapping)
 {
 	uint32_t i;
 	int gpuidx;
@@ -702,6 +703,7 @@ svm_range_apply_attrs(struct kfd_process *p, struct svm_range *prange,
 		case KFD_IOCTL_SVM_ATTR_ACCESS:
 		case KFD_IOCTL_SVM_ATTR_ACCESS_IN_PLACE:
 		case KFD_IOCTL_SVM_ATTR_NO_ACCESS:
+			*update_mapping = true;
 			gpuidx = kfd_process_gpuidx_from_gpuid(p,
 							       attrs[i].value);
 			if (attrs[i].type == KFD_IOCTL_SVM_ATTR_NO_ACCESS) {
@@ -716,9 +718,11 @@ svm_range_apply_attrs(struct kfd_process *p, struct svm_range *prange,
 			}
 			break;
 		case KFD_IOCTL_SVM_ATTR_SET_FLAGS:
+			*update_mapping = true;
 			prange->flags |= attrs[i].value;
 			break;
 		case KFD_IOCTL_SVM_ATTR_CLR_FLAGS:
+			*update_mapping = true;
 			prange->flags &= ~attrs[i].value;
 			break;
 		case KFD_IOCTL_SVM_ATTR_GRANULARITY:
@@ -1253,7 +1257,7 @@ static int
 svm_range_map_to_gpu(struct kfd_process_device *pdd, struct svm_range *prange,
 		     unsigned long offset, unsigned long npages, bool readonly,
 		     dma_addr_t *dma_addr, struct amdgpu_device *bo_adev,
-		     struct dma_fence **fence)
+		     struct dma_fence **fence, bool flush_tlb)
 {
 	struct amdgpu_device *adev = pdd->dev->adev;
 	struct amdgpu_vm *vm = drm_priv_to_vm(pdd->drm_priv);
@@ -1291,7 +1295,7 @@ svm_range_map_to_gpu(struct kfd_process_device *pdd, struct svm_range *prange,
 			 (last_domain == SVM_RANGE_VRAM_DOMAIN) ? 1 : 0,
 			 pte_flags);
 
-		r = amdgpu_vm_update_range(adev, vm, false, false, false, NULL,
+		r = amdgpu_vm_update_range(adev, vm, false, false, flush_tlb, NULL,
 					   last_start, prange->start + i,
 					   pte_flags,
 					   last_start - prange->start,
@@ -1325,7 +1329,7 @@ svm_range_map_to_gpu(struct kfd_process_device *pdd, struct svm_range *prange,
 static int
 svm_range_map_to_gpus(struct svm_range *prange, unsigned long offset,
 		      unsigned long npages, bool readonly,
-		      unsigned long *bitmap, bool wait)
+		      unsigned long *bitmap, bool wait, bool flush_tlb)
 {
 	struct kfd_process_device *pdd;
 	struct amdgpu_device *bo_adev;
@@ -1360,7 +1364,8 @@ svm_range_map_to_gpus(struct svm_range *prange, unsigned long offset,
 
 		r = svm_range_map_to_gpu(pdd, prange, offset, npages, readonly,
 					 prange->dma_addr[gpuidx],
-					 bo_adev, wait ? &fence : NULL);
+					 bo_adev, wait ? &fence : NULL,
+					 flush_tlb);
 		if (r)
 			break;
 
@@ -1481,8 +1486,8 @@ static void *kfd_svm_page_owner(struct kfd_process *p, int32_t gpuidx)
  * 5. Release page table (and SVM BO) reservation
  */
 static int svm_range_validate_and_map(struct mm_struct *mm,
-				      struct svm_range *prange,
-				      int32_t gpuidx, bool intr, bool wait)
+				      struct svm_range *prange, int32_t gpuidx,
+				      bool intr, bool wait, bool flush_tlb)
 {
 	struct svm_validate_context ctx;
 	unsigned long start, end, addr;
@@ -1521,8 +1526,12 @@ static int svm_range_validate_and_map(struct mm_struct *mm,
 			  prange->bitmap_aip, MAX_GPU_INSTANCE);
 	}
 
-	if (bitmap_empty(ctx.bitmap, MAX_GPU_INSTANCE))
-		return 0;
+	if (bitmap_empty(ctx.bitmap, MAX_GPU_INSTANCE)) {
+		if (!prange->mapped_to_gpu)
+			return 0;
+
+		bitmap_copy(ctx.bitmap, prange->bitmap_access, MAX_GPU_INSTANCE);
+	}
 
 	if (prange->actual_loc && !prange->ttm_res) {
 		/* This should never happen. actual_loc gets set by
@@ -1594,7 +1603,7 @@ static int svm_range_validate_and_map(struct mm_struct *mm,
 		}
 
 		r = svm_range_map_to_gpus(prange, offset, npages, readonly,
-					  ctx.bitmap, wait);
+					  ctx.bitmap, wait, flush_tlb);
 
 unlock_out:
 		svm_range_unlock(prange);
@@ -1690,7 +1699,7 @@ static void svm_range_restore_work(struct work_struct *work)
 		mutex_lock(&prange->migrate_mutex);
 
 		r = svm_range_validate_and_map(mm, prange, MAX_GPU_INSTANCE,
-					       false, true);
+					       false, true, false);
 		if (r)
 			pr_debug("failed %d to map 0x%lx to gpus\n", r,
 				 prange->start);
@@ -2828,7 +2837,7 @@ svm_range_restore_pages(struct amdgpu_device *adev, unsigned int pasid,
 		}
 	}
 
-	r = svm_range_validate_and_map(mm, prange, gpuidx, false, false);
+	r = svm_range_validate_and_map(mm, prange, gpuidx, false, false, false);
 	if (r)
 		pr_debug("failed %d to map svms 0x%p [0x%lx 0x%lx] to gpus\n",
 			 r, svms, prange->start, prange->last);
@@ -3241,6 +3250,8 @@ svm_range_set_attr(struct kfd_process *p, struct mm_struct *mm,
 	struct svm_range_list *svms;
 	struct svm_range *prange;
 	struct svm_range *next;
+	bool update_mapping = false;
+	bool flush_tlb;
 	int r = 0;
 
 	pr_debug("pasid 0x%x svms 0x%p [0x%llx 0x%llx] pages 0x%llx\n",
@@ -3279,7 +3290,7 @@ svm_range_set_attr(struct kfd_process *p, struct mm_struct *mm,
 		svm_range_add_notifier_locked(mm, prange);
 	}
 	list_for_each_entry(prange, &update_list, update_list) {
-		svm_range_apply_attrs(p, prange, nattr, attrs);
+		svm_range_apply_attrs(p, prange, nattr, attrs, &update_mapping);
 		/* TODO: unmap ranges from GPU that lost access */
 	}
 	list_for_each_entry_safe(prange, next, &remove_list, update_list) {
@@ -3312,8 +3323,10 @@ svm_range_set_attr(struct kfd_process *p, struct mm_struct *mm,
 			continue;
 		}
 
+		flush_tlb = !migrated && update_mapping && prange->mapped_to_gpu;
+
 		r = svm_range_validate_and_map(mm, prange, MAX_GPU_INSTANCE,
-					       true, true);
+					       true, true, flush_tlb);
 		if (r)
 			pr_debug("failed %d to map svm range\n", r);
 
Signature · 1 line
2.35.1
maintainer
#2 lore raw

Re: [PATCH 2/2] drm/amdkfd: Update mapping if range attributes changed

Am 2022-04-19 um 20:47 schrieb Philip Yang:

Quoted 152 lines

Change SVM range mapping flags or access attributes don't trigger migration, if range is already mapped on GPUs we should update GPU mapping, and pass flush_tlb flag to amdgpu vm.

---
  drivers/gpu/drm/amd/amdkfd/kfd_svm.c | 41 ++++++++++++++++++----------
  1 file changed, 27 insertions(+), 14 deletions(-)
drivers/gpu/drm/amd/amdkfd/kfd_svm.c +24 -13
index 6be1695f3a09..0a9bdadf875e 100644
@@ -686,7 +686,8 @@ svm_range_check_attr(struct kfd_process *p,
 
  static void
  svm_range_apply_attrs(struct kfd_process *p, struct svm_range *prange,
-		      uint32_t nattr, struct kfd_ioctl_svm_attribute *attrs)
+		      uint32_t nattr, struct kfd_ioctl_svm_attribute *attrs,
+		      bool *update_mapping)
  {
  	uint32_t i;
  	int gpuidx;
@@ -702,6 +703,7 @@ svm_range_apply_attrs(struct kfd_process *p, struct svm_range *prange,
  		case KFD_IOCTL_SVM_ATTR_ACCESS:
  		case KFD_IOCTL_SVM_ATTR_ACCESS_IN_PLACE:
  		case KFD_IOCTL_SVM_ATTR_NO_ACCESS:
+			*update_mapping = true;
  			gpuidx = kfd_process_gpuidx_from_gpuid(p,
  							       attrs[i].value);
  			if (attrs[i].type == KFD_IOCTL_SVM_ATTR_NO_ACCESS) {
@@ -716,9 +718,11 @@ svm_range_apply_attrs(struct kfd_process *p, struct svm_range *prange,
  			}
  			break;
  		case KFD_IOCTL_SVM_ATTR_SET_FLAGS:
+			*update_mapping = true;
  			prange->flags |= attrs[i].value;
  			break;
  		case KFD_IOCTL_SVM_ATTR_CLR_FLAGS:
+			*update_mapping = true;
  			prange->flags &= ~attrs[i].value;
  			break;
  		case KFD_IOCTL_SVM_ATTR_GRANULARITY:
@@ -1253,7 +1257,7 @@ static int
  svm_range_map_to_gpu(struct kfd_process_device *pdd, struct svm_range *prange,
  		     unsigned long offset, unsigned long npages, bool readonly,
  		     dma_addr_t *dma_addr, struct amdgpu_device *bo_adev,
-		     struct dma_fence **fence)
+		     struct dma_fence **fence, bool flush_tlb)
  {
  	struct amdgpu_device *adev = pdd->dev->adev;
  	struct amdgpu_vm *vm = drm_priv_to_vm(pdd->drm_priv);
@@ -1291,7 +1295,7 @@ svm_range_map_to_gpu(struct kfd_process_device *pdd, struct svm_range *prange,
  			 (last_domain == SVM_RANGE_VRAM_DOMAIN) ? 1 : 0,
  			 pte_flags);
 
-		r = amdgpu_vm_update_range(adev, vm, false, false, false, NULL,
+		r = amdgpu_vm_update_range(adev, vm, false, false, flush_tlb, NULL,
  					   last_start, prange->start + i,
  					   pte_flags,
  					   last_start - prange->start,
@@ -1325,7 +1329,7 @@ svm_range_map_to_gpu(struct kfd_process_device *pdd, struct svm_range *prange,
  static int
  svm_range_map_to_gpus(struct svm_range *prange, unsigned long offset,
  		      unsigned long npages, bool readonly,
-		      unsigned long *bitmap, bool wait)
+		      unsigned long *bitmap, bool wait, bool flush_tlb)
  {
  	struct kfd_process_device *pdd;
  	struct amdgpu_device *bo_adev;
@@ -1360,7 +1364,8 @@ svm_range_map_to_gpus(struct svm_range *prange, unsigned long offset,
 
  		r = svm_range_map_to_gpu(pdd, prange, offset, npages, readonly,
  					 prange->dma_addr[gpuidx],
-					 bo_adev, wait ? &fence : NULL);
+					 bo_adev, wait ? &fence : NULL,
+					 flush_tlb);
  		if (r)
  			break;
 
@@ -1481,8 +1486,8 @@ static void *kfd_svm_page_owner(struct kfd_process *p, int32_t gpuidx)
   * 5. Release page table (and SVM BO) reservation
   */
  static int svm_range_validate_and_map(struct mm_struct *mm,
-				      struct svm_range *prange,
-				      int32_t gpuidx, bool intr, bool wait)
+				      struct svm_range *prange, int32_t gpuidx,
+				      bool intr, bool wait, bool flush_tlb)
  {
  	struct svm_validate_context ctx;
  	unsigned long start, end, addr;
@@ -1521,8 +1526,12 @@ static int svm_range_validate_and_map(struct mm_struct *mm,
  			  prange->bitmap_aip, MAX_GPU_INSTANCE);
  	}
 
-	if (bitmap_empty(ctx.bitmap, MAX_GPU_INSTANCE))
-		return 0;
+	if (bitmap_empty(ctx.bitmap, MAX_GPU_INSTANCE)) {
+		if (!prange->mapped_to_gpu)
+			return 0;
+
+		bitmap_copy(ctx.bitmap, prange->bitmap_access, MAX_GPU_INSTANCE);
+	}
 
  	if (prange->actual_loc && !prange->ttm_res) {
  		/* This should never happen. actual_loc gets set by
@@ -1594,7 +1603,7 @@ static int svm_range_validate_and_map(struct mm_struct *mm,
  		}
 
  		r = svm_range_map_to_gpus(prange, offset, npages, readonly,
-					  ctx.bitmap, wait);
+					  ctx.bitmap, wait, flush_tlb);
 
  unlock_out:
  		svm_range_unlock(prange);
@@ -1690,7 +1699,7 @@ static void svm_range_restore_work(struct work_struct *work)
  		mutex_lock(&prange->migrate_mutex);
 
  		r = svm_range_validate_and_map(mm, prange, MAX_GPU_INSTANCE,
-					       false, true);
+					       false, true, false);
  		if (r)
  			pr_debug("failed %d to map 0x%lx to gpus\n", r,
  				 prange->start);
@@ -2828,7 +2837,7 @@ svm_range_restore_pages(struct amdgpu_device *adev, unsigned int pasid,
  		}
  	}
 
-	r = svm_range_validate_and_map(mm, prange, gpuidx, false, false);
+	r = svm_range_validate_and_map(mm, prange, gpuidx, false, false, false);
  	if (r)
  		pr_debug("failed %d to map svms 0x%p [0x%lx 0x%lx] to gpus\n",
  			 r, svms, prange->start, prange->last);
@@ -3241,6 +3250,8 @@ svm_range_set_attr(struct kfd_process *p, struct mm_struct *mm,
  	struct svm_range_list *svms;
  	struct svm_range *prange;
  	struct svm_range *next;
+	bool update_mapping = false;
+	bool flush_tlb;
  	int r = 0;
 
  	pr_debug("pasid 0x%x svms 0x%p [0x%llx 0x%llx] pages 0x%llx\n",
@@ -3279,7 +3290,7 @@ svm_range_set_attr(struct kfd_process *p, struct mm_struct *mm,
  		svm_range_add_notifier_locked(mm, prange);
  	}
  	list_for_each_entry(prange, &update_list, update_list) {
-		svm_range_apply_attrs(p, prange, nattr, attrs);
+		svm_range_apply_attrs(p, prange, nattr, attrs, &update_mapping);
  		/* TODO: unmap ranges from GPU that lost access */
  	}
  	list_for_each_entry_safe(prange, next, &remove_list, update_list) {

drivers/gpu/drm/amd/amdkfd/kfd_svm.c svm_range_set_attr(struct kfd_process *p, struct mm_struct *mm, · line 3327

Quoted 5 lines
drivers/gpu/drm/amd/amdkfd/kfd_svm.c +2 -0
@@ -3312,8 +3323,10 @@ svm_range_set_attr(struct kfd_process *p, struct mm_struct *mm,
  			continue;
  		}
 
+		flush_tlb = !migrated && update_mapping && prange->mapped_to_gpu;
+

Can we skip the validate_and_map if update_mapping is false?

Other than that, the series looks good to me.

Regards,
  Felix

Quoted 6 lines
drivers/gpu/drm/amd/amdkfd/kfd_svm.c +1 -1
  		r = svm_range_validate_and_map(mm, prange, MAX_GPU_INSTANCE,
-					       true, true);
+					       true, true, flush_tlb);
  		if (r)
  			pr_debug("failed %d to map svm range\n", r);
 
#3 lore raw

Re: [PATCH 2/2] drm/amdkfd: Update mapping if range attributes changed

On 2022-04-21 16:17, Felix Kuehling wrote:
Am 2022-04-19 um 20:47 schrieb Philip Yang:
Change SVM range mapping flags or access attributes don't trigger migration, if range is already mapped on GPUs we should update GPU mapping, and pass flush_tlb flag to amdgpu vm.

---
  drivers/gpu/drm/amd/amdkfd/kfd_svm.c | 41 ++++++++++++++++++----------

  1 file changed, 27 insertions(+), 14 deletions(-)
drivers/gpu/drm/amd/amdkfd/kfd_svm.c +0 -0
index 6be1695f3a09..0a9bdadf875e 100644
@@ -686,7 +686,8 @@ svm_range_check_attr(struct kfd_process *p,

    static void
  svm_range_apply_attrs(struct kfd_process *p, struct svm_range *prange,
-              uint32_t nattr, struct kfd_ioctl_svm_attribute *attrs)
+              uint32_t nattr, struct kfd_ioctl_svm_attribute *attrs,
+              bool *update_mapping)
  {
      uint32_t i;
      int gpuidx;
@@ -702,6 +703,7 @@ svm_range_apply_attrs(struct kfd_process *p, struct svm_range *prange, case KFD_IOCTL_SVM_ATTR_ACCESS:
          case KFD_IOCTL_SVM_ATTR_ACCESS_IN_PLACE:
          case KFD_IOCTL_SVM_ATTR_NO_ACCESS:
+            *update_mapping = true;
              gpuidx = kfd_process_gpuidx_from_gpuid(p,
                                     attrs[i].value);
              if (attrs[i].type == KFD_IOCTL_SVM_ATTR_NO_ACCESS) {
@@ -716,9 +718,11 @@ svm_range_apply_attrs(struct kfd_process *p, struct svm_range *prange, }
              break;
          case KFD_IOCTL_SVM_ATTR_SET_FLAGS:
+            *update_mapping = true;
              prange->flags |= attrs[i].value;
              break;
          case KFD_IOCTL_SVM_ATTR_CLR_FLAGS:
+            *update_mapping = true;
              prange->flags &= ~attrs[i].value;
              break;
          case KFD_IOCTL_SVM_ATTR_GRANULARITY:
@@ -1253,7 +1257,7 @@ static int
  svm_range_map_to_gpu(struct kfd_process_device *pdd, struct svm_range *prange, unsigned long offset, unsigned long npages, bool readonly,
               dma_addr_t *dma_addr, struct amdgpu_device *bo_adev,
-             struct dma_fence **fence)
+             struct dma_fence **fence, bool flush_tlb)
  {
      struct amdgpu_device *adev = pdd->dev->adev;
      struct amdgpu_vm *vm = drm_priv_to_vm(pdd->drm_priv);
@@ -1291,7 +1295,7 @@ svm_range_map_to_gpu(struct kfd_process_device *pdd, struct svm_range *prange, (last_domain == SVM_RANGE_VRAM_DOMAIN) ? 1 : 0,
               pte_flags);
  -        r = amdgpu_vm_update_range(adev, vm, false, false, false, NULL,
+        r = amdgpu_vm_update_range(adev, vm, false, false, flush_tlb, NULL, last_start, prange->start + i,
                         pte_flags,
                         last_start - prange->start,
@@ -1325,7 +1329,7 @@ svm_range_map_to_gpu(struct kfd_process_device *pdd, struct svm_range *prange, static int
  svm_range_map_to_gpus(struct svm_range *prange, unsigned long offset,
                unsigned long npages, bool readonly,
-              unsigned long *bitmap, bool wait)
+              unsigned long *bitmap, bool wait, bool flush_tlb)
  {
      struct kfd_process_device *pdd;
      struct amdgpu_device *bo_adev;
@@ -1360,7 +1364,8 @@ svm_range_map_to_gpus(struct svm_range *prange, unsigned long offset, r = svm_range_map_to_gpu(pdd, prange, offset, npages, readonly, prange->dma_addr[gpuidx],
-                     bo_adev, wait ? &fence : NULL);
+                     bo_adev, wait ? &fence : NULL,
+                     flush_tlb);
          if (r)
              break;
  @@ -1481,8 +1486,8 @@ static void *kfd_svm_page_owner(struct kfd_process *p, int32_t gpuidx)
   * 5. Release page table (and SVM BO) reservation
   */
  static int svm_range_validate_and_map(struct mm_struct *mm,
-                      struct svm_range *prange,
-                      int32_t gpuidx, bool intr, bool wait)
+                      struct svm_range *prange, int32_t gpuidx,
+                      bool intr, bool wait, bool flush_tlb)
  {
      struct svm_validate_context ctx;
      unsigned long start, end, addr;
@@ -1521,8 +1526,12 @@ static int svm_range_validate_and_map(struct mm_struct *mm, prange->bitmap_aip, MAX_GPU_INSTANCE);
      }
  -    if (bitmap_empty(ctx.bitmap, MAX_GPU_INSTANCE))
-        return 0;
+    if (bitmap_empty(ctx.bitmap, MAX_GPU_INSTANCE)) {
+        if (!prange->mapped_to_gpu)
+            return 0;
+
+        bitmap_copy(ctx.bitmap, prange->bitmap_access, MAX_GPU_INSTANCE);
+    }
        if (prange->actual_loc && !prange->ttm_res) {
          /* This should never happen. actual_loc gets set by
@@ -1594,7 +1603,7 @@ static int svm_range_validate_and_map(struct mm_struct *mm,
          }
            r = svm_range_map_to_gpus(prange, offset, npages, readonly,
-                      ctx.bitmap, wait);
+                      ctx.bitmap, wait, flush_tlb);
    unlock_out:
          svm_range_unlock(prange);
@@ -1690,7 +1699,7 @@ static void svm_range_restore_work(struct work_struct *work) mutex_lock(&prange->migrate_mutex);
            r = svm_range_validate_and_map(mm, prange, MAX_GPU_INSTANCE,
-                           false, true);
+                           false, true, false);
          if (r)
              pr_debug("failed %d to map 0x%lx to gpus\n", r,
                   prange->start);
@@ -2828,7 +2837,7 @@ svm_range_restore_pages(struct amdgpu_device *adev, unsigned int pasid, }
      }
  -    r = svm_range_validate_and_map(mm, prange, gpuidx, false, false);
+    r = svm_range_validate_and_map(mm, prange, gpuidx, false, false, false);
      if (r)
          pr_debug("failed %d to map svms 0x%p [0x%lx 0x%lx] to gpus\n",
               r, svms, prange->start, prange->last);
@@ -3241,6 +3250,8 @@ svm_range_set_attr(struct kfd_process *p, struct mm_struct *mm, struct svm_range_list *svms;
      struct svm_range *prange;
      struct svm_range *next;
+    bool update_mapping = false;
+    bool flush_tlb;
      int r = 0;
        pr_debug("pasid 0x%x svms 0x%p [0x%llx 0x%llx] pages 0x%llx\n",
@@ -3279,7 +3290,7 @@ svm_range_set_attr(struct kfd_process *p, struct mm_struct *mm, svm_range_add_notifier_locked(mm, prange);
      }
      list_for_each_entry(prange, &update_list, update_list) {
-        svm_range_apply_attrs(p, prange, nattr, attrs);
+        svm_range_apply_attrs(p, prange, nattr, attrs, &update_mapping);
          /* TODO: unmap ranges from GPU that lost access */
      }
      list_for_each_entry_safe(prange, next, &remove_list, update_list) {
@@ -3312,8 +3323,10 @@ svm_range_set_attr(struct kfd_process *p, struct mm_struct *mm, continue;
          }
  +        flush_tlb = !migrated && update_mapping && prange->mapped_to_gpu;
+

Can we skip the validate_and_map if update_mapping is false?

Yes, if changing range perferred_loc or migration granularity, don't need update mapping, we can skip validate_and_map, will send out v2 patch.
Thanks,
Philip

Other than that, the series looks good to me.

Regards,
  Felix

          r = svm_range_validate_and_map(mm, prange, MAX_GPU_INSTANCE, -                           true, true);
+                           true, true, flush_tlb);
          if (r)
              pr_debug("failed %d to map svm range\n", r);

#4 lore raw

PATCH v2 2/2 drm/amdkfd: Update mapping if range attributes changed

In reply to Philip Yang

Change SVM range mapping flags or access attributes don't trigger migration, if range is already mapped on GPUs we should update GPU mapping and pass flush_tlb flag true to amdgpu vm.

Change SVM range preferred_loc or migration granularity don't need update GPU mapping, skip the validate_and_map.

---
 drivers/gpu/drm/amd/amdkfd/kfd_svm.c | 46 +++++++++++++++++++---------
 1 file changed, 32 insertions(+), 14 deletions(-)
drivers/gpu/drm/amd/amdkfd/kfd_svm.c +32 -14
index 8a077cd066a1..e740384df9c7 100644
@@ -686,7 +686,8 @@ svm_range_check_attr(struct kfd_process *p,
 
 static void
 svm_range_apply_attrs(struct kfd_process *p, struct svm_range *prange,
-		      uint32_t nattr, struct kfd_ioctl_svm_attribute *attrs)
+		      uint32_t nattr, struct kfd_ioctl_svm_attribute *attrs,
+		      bool *update_mapping)
 {
 	uint32_t i;
 	int gpuidx;
@@ -702,6 +703,7 @@ svm_range_apply_attrs(struct kfd_process *p, struct svm_range *prange,
 		case KFD_IOCTL_SVM_ATTR_ACCESS:
 		case KFD_IOCTL_SVM_ATTR_ACCESS_IN_PLACE:
 		case KFD_IOCTL_SVM_ATTR_NO_ACCESS:
+			*update_mapping = true;
 			gpuidx = kfd_process_gpuidx_from_gpuid(p,
 							       attrs[i].value);
 			if (attrs[i].type == KFD_IOCTL_SVM_ATTR_NO_ACCESS) {
@@ -716,9 +718,11 @@ svm_range_apply_attrs(struct kfd_process *p, struct svm_range *prange,
 			}
 			break;
 		case KFD_IOCTL_SVM_ATTR_SET_FLAGS:
+			*update_mapping = true;
 			prange->flags |= attrs[i].value;
 			break;
 		case KFD_IOCTL_SVM_ATTR_CLR_FLAGS:
+			*update_mapping = true;
 			prange->flags &= ~attrs[i].value;
 			break;
 		case KFD_IOCTL_SVM_ATTR_GRANULARITY:
@@ -1254,7 +1258,7 @@ static int
 svm_range_map_to_gpu(struct kfd_process_device *pdd, struct svm_range *prange,
 		     unsigned long offset, unsigned long npages, bool readonly,
 		     dma_addr_t *dma_addr, struct amdgpu_device *bo_adev,
-		     struct dma_fence **fence)
+		     struct dma_fence **fence, bool flush_tlb)
 {
 	struct amdgpu_device *adev = pdd->dev->adev;
 	struct amdgpu_vm *vm = drm_priv_to_vm(pdd->drm_priv);
@@ -1292,7 +1296,7 @@ svm_range_map_to_gpu(struct kfd_process_device *pdd, struct svm_range *prange,
 			 (last_domain == SVM_RANGE_VRAM_DOMAIN) ? 1 : 0,
 			 pte_flags);
 
-		r = amdgpu_vm_update_range(adev, vm, false, false, false, NULL,
+		r = amdgpu_vm_update_range(adev, vm, false, false, flush_tlb, NULL,
 					   last_start, prange->start + i,
 					   pte_flags,
 					   last_start - prange->start,
@@ -1326,7 +1330,7 @@ svm_range_map_to_gpu(struct kfd_process_device *pdd, struct svm_range *prange,
 static int
 svm_range_map_to_gpus(struct svm_range *prange, unsigned long offset,
 		      unsigned long npages, bool readonly,
-		      unsigned long *bitmap, bool wait)
+		      unsigned long *bitmap, bool wait, bool flush_tlb)
 {
 	struct kfd_process_device *pdd;
 	struct amdgpu_device *bo_adev;
@@ -1361,7 +1365,8 @@ svm_range_map_to_gpus(struct svm_range *prange, unsigned long offset,
 
 		r = svm_range_map_to_gpu(pdd, prange, offset, npages, readonly,
 					 prange->dma_addr[gpuidx],
-					 bo_adev, wait ? &fence : NULL);
+					 bo_adev, wait ? &fence : NULL,
+					 flush_tlb);
 		if (r)
 			break;
 
@@ -1482,8 +1487,8 @@ static void *kfd_svm_page_owner(struct kfd_process *p, int32_t gpuidx)
  * 5. Release page table (and SVM BO) reservation
  */
 static int svm_range_validate_and_map(struct mm_struct *mm,
-				      struct svm_range *prange,
-				      int32_t gpuidx, bool intr, bool wait)
+				      struct svm_range *prange, int32_t gpuidx,
+				      bool intr, bool wait, bool flush_tlb)
 {
 	struct svm_validate_context ctx;
 	unsigned long start, end, addr;
@@ -1522,8 +1527,12 @@ static int svm_range_validate_and_map(struct mm_struct *mm,
 			  prange->bitmap_aip, MAX_GPU_INSTANCE);
 	}
 
-	if (bitmap_empty(ctx.bitmap, MAX_GPU_INSTANCE))
-		return 0;
+	if (bitmap_empty(ctx.bitmap, MAX_GPU_INSTANCE)) {
+		if (!prange->mapped_to_gpu)
+			return 0;
+
+		bitmap_copy(ctx.bitmap, prange->bitmap_access, MAX_GPU_INSTANCE);
+	}
 
 	if (prange->actual_loc && !prange->ttm_res) {
 		/* This should never happen. actual_loc gets set by
@@ -1595,7 +1604,7 @@ static int svm_range_validate_and_map(struct mm_struct *mm,
 		}
 
 		r = svm_range_map_to_gpus(prange, offset, npages, readonly,
-					  ctx.bitmap, wait);
+					  ctx.bitmap, wait, flush_tlb);
 
 unlock_out:
 		svm_range_unlock(prange);
@@ -1691,7 +1700,7 @@ static void svm_range_restore_work(struct work_struct *work)
 		mutex_lock(&prange->migrate_mutex);
 
 		r = svm_range_validate_and_map(mm, prange, MAX_GPU_INSTANCE,
-					       false, true);
+					       false, true, false);
 		if (r)
 			pr_debug("failed %d to map 0x%lx to gpus\n", r,
 				 prange->start);
@@ -2847,7 +2856,7 @@ svm_range_restore_pages(struct amdgpu_device *adev, unsigned int pasid,
 		}
 	}
 
-	r = svm_range_validate_and_map(mm, prange, gpuidx, false, false);
+	r = svm_range_validate_and_map(mm, prange, gpuidx, false, false, false);
 	if (r)
 		pr_debug("failed %d to map svms 0x%p [0x%lx 0x%lx] to gpus\n",
 			 r, svms, prange->start, prange->last);
@@ -3264,6 +3273,8 @@ svm_range_set_attr(struct kfd_process *p, struct mm_struct *mm,
 	struct svm_range_list *svms;
 	struct svm_range *prange;
 	struct svm_range *next;
+	bool update_mapping = false;
+	bool flush_tlb;
 	int r = 0;
 
 	pr_debug("pasid 0x%x svms 0x%p [0x%llx 0x%llx] pages 0x%llx\n",
@@ -3302,7 +3313,7 @@ svm_range_set_attr(struct kfd_process *p, struct mm_struct *mm,
 		svm_range_add_notifier_locked(mm, prange);
 	}
 	list_for_each_entry(prange, &update_list, update_list) {
-		svm_range_apply_attrs(p, prange, nattr, attrs);
+		svm_range_apply_attrs(p, prange, nattr, attrs, &update_mapping);
 		/* TODO: unmap ranges from GPU that lost access */
 	}
 	list_for_each_entry_safe(prange, next, &remove_list, update_list) {
@@ -3335,8 +3346,15 @@ svm_range_set_attr(struct kfd_process *p, struct mm_struct *mm,
 			continue;
 		}
 
+		if (!migrated && !update_mapping) {
+			mutex_unlock(&prange->migrate_mutex);
+			continue;
+		}
+
+		flush_tlb = !migrated && update_mapping && prange->mapped_to_gpu;
+
 		r = svm_range_validate_and_map(mm, prange, MAX_GPU_INSTANCE,
-					       true, true);
+					       true, true, flush_tlb);
 		if (r)
 			pr_debug("failed %d to map svm range\n", r);
 
Signature · 1 line
2.35.1
maintainer
#5 lore raw

Re: [PATCH v2 2/2] drm/amdkfd: Update mapping if range attributes changed

Quoted 155 lines · Philip Yang

Change SVM range mapping flags or access attributes don't trigger migration, if range is already mapped on GPUs we should update GPU mapping and pass flush_tlb flag true to amdgpu vm.

Change SVM range preferred_loc or migration granularity don't need update GPU mapping, skip the validate_and_map.

---
  drivers/gpu/drm/amd/amdkfd/kfd_svm.c | 46 +++++++++++++++++++---------
  1 file changed, 32 insertions(+), 14 deletions(-)
drivers/gpu/drm/amd/amdkfd/kfd_svm.c +24 -13
index 8a077cd066a1..e740384df9c7 100644
@@ -686,7 +686,8 @@ svm_range_check_attr(struct kfd_process *p,
 
  static void
  svm_range_apply_attrs(struct kfd_process *p, struct svm_range *prange,
-		      uint32_t nattr, struct kfd_ioctl_svm_attribute *attrs)
+		      uint32_t nattr, struct kfd_ioctl_svm_attribute *attrs,
+		      bool *update_mapping)
  {
  	uint32_t i;
  	int gpuidx;
@@ -702,6 +703,7 @@ svm_range_apply_attrs(struct kfd_process *p, struct svm_range *prange,
  		case KFD_IOCTL_SVM_ATTR_ACCESS:
  		case KFD_IOCTL_SVM_ATTR_ACCESS_IN_PLACE:
  		case KFD_IOCTL_SVM_ATTR_NO_ACCESS:
+			*update_mapping = true;
  			gpuidx = kfd_process_gpuidx_from_gpuid(p,
  							       attrs[i].value);
  			if (attrs[i].type == KFD_IOCTL_SVM_ATTR_NO_ACCESS) {
@@ -716,9 +718,11 @@ svm_range_apply_attrs(struct kfd_process *p, struct svm_range *prange,
  			}
  			break;
  		case KFD_IOCTL_SVM_ATTR_SET_FLAGS:
+			*update_mapping = true;
  			prange->flags |= attrs[i].value;
  			break;
  		case KFD_IOCTL_SVM_ATTR_CLR_FLAGS:
+			*update_mapping = true;
  			prange->flags &= ~attrs[i].value;
  			break;
  		case KFD_IOCTL_SVM_ATTR_GRANULARITY:
@@ -1254,7 +1258,7 @@ static int
  svm_range_map_to_gpu(struct kfd_process_device *pdd, struct svm_range *prange,
  		     unsigned long offset, unsigned long npages, bool readonly,
  		     dma_addr_t *dma_addr, struct amdgpu_device *bo_adev,
-		     struct dma_fence **fence)
+		     struct dma_fence **fence, bool flush_tlb)
  {
  	struct amdgpu_device *adev = pdd->dev->adev;
  	struct amdgpu_vm *vm = drm_priv_to_vm(pdd->drm_priv);
@@ -1292,7 +1296,7 @@ svm_range_map_to_gpu(struct kfd_process_device *pdd, struct svm_range *prange,
  			 (last_domain == SVM_RANGE_VRAM_DOMAIN) ? 1 : 0,
  			 pte_flags);
 
-		r = amdgpu_vm_update_range(adev, vm, false, false, false, NULL,
+		r = amdgpu_vm_update_range(adev, vm, false, false, flush_tlb, NULL,
  					   last_start, prange->start + i,
  					   pte_flags,
  					   last_start - prange->start,
@@ -1326,7 +1330,7 @@ svm_range_map_to_gpu(struct kfd_process_device *pdd, struct svm_range *prange,
  static int
  svm_range_map_to_gpus(struct svm_range *prange, unsigned long offset,
  		      unsigned long npages, bool readonly,
-		      unsigned long *bitmap, bool wait)
+		      unsigned long *bitmap, bool wait, bool flush_tlb)
  {
  	struct kfd_process_device *pdd;
  	struct amdgpu_device *bo_adev;
@@ -1361,7 +1365,8 @@ svm_range_map_to_gpus(struct svm_range *prange, unsigned long offset,
 
  		r = svm_range_map_to_gpu(pdd, prange, offset, npages, readonly,
  					 prange->dma_addr[gpuidx],
-					 bo_adev, wait ? &fence : NULL);
+					 bo_adev, wait ? &fence : NULL,
+					 flush_tlb);
  		if (r)
  			break;
 
@@ -1482,8 +1487,8 @@ static void *kfd_svm_page_owner(struct kfd_process *p, int32_t gpuidx)
   * 5. Release page table (and SVM BO) reservation
   */
  static int svm_range_validate_and_map(struct mm_struct *mm,
-				      struct svm_range *prange,
-				      int32_t gpuidx, bool intr, bool wait)
+				      struct svm_range *prange, int32_t gpuidx,
+				      bool intr, bool wait, bool flush_tlb)
  {
  	struct svm_validate_context ctx;
  	unsigned long start, end, addr;
@@ -1522,8 +1527,12 @@ static int svm_range_validate_and_map(struct mm_struct *mm,
  			  prange->bitmap_aip, MAX_GPU_INSTANCE);
  	}
 
-	if (bitmap_empty(ctx.bitmap, MAX_GPU_INSTANCE))
-		return 0;
+	if (bitmap_empty(ctx.bitmap, MAX_GPU_INSTANCE)) {
+		if (!prange->mapped_to_gpu)
+			return 0;
+
+		bitmap_copy(ctx.bitmap, prange->bitmap_access, MAX_GPU_INSTANCE);
+	}
 
  	if (prange->actual_loc && !prange->ttm_res) {
  		/* This should never happen. actual_loc gets set by
@@ -1595,7 +1604,7 @@ static int svm_range_validate_and_map(struct mm_struct *mm,
  		}
 
  		r = svm_range_map_to_gpus(prange, offset, npages, readonly,
-					  ctx.bitmap, wait);
+					  ctx.bitmap, wait, flush_tlb);
 
  unlock_out:
  		svm_range_unlock(prange);
@@ -1691,7 +1700,7 @@ static void svm_range_restore_work(struct work_struct *work)
  		mutex_lock(&prange->migrate_mutex);
 
  		r = svm_range_validate_and_map(mm, prange, MAX_GPU_INSTANCE,
-					       false, true);
+					       false, true, false);
  		if (r)
  			pr_debug("failed %d to map 0x%lx to gpus\n", r,
  				 prange->start);
@@ -2847,7 +2856,7 @@ svm_range_restore_pages(struct amdgpu_device *adev, unsigned int pasid,
  		}
  	}
 
-	r = svm_range_validate_and_map(mm, prange, gpuidx, false, false);
+	r = svm_range_validate_and_map(mm, prange, gpuidx, false, false, false);
  	if (r)
  		pr_debug("failed %d to map svms 0x%p [0x%lx 0x%lx] to gpus\n",
  			 r, svms, prange->start, prange->last);
@@ -3264,6 +3273,8 @@ svm_range_set_attr(struct kfd_process *p, struct mm_struct *mm,
  	struct svm_range_list *svms;
  	struct svm_range *prange;
  	struct svm_range *next;
+	bool update_mapping = false;
+	bool flush_tlb;
  	int r = 0;
 
  	pr_debug("pasid 0x%x svms 0x%p [0x%llx 0x%llx] pages 0x%llx\n",
@@ -3302,7 +3313,7 @@ svm_range_set_attr(struct kfd_process *p, struct mm_struct *mm,
  		svm_range_add_notifier_locked(mm, prange);
  	}
  	list_for_each_entry(prange, &update_list, update_list) {
-		svm_range_apply_attrs(p, prange, nattr, attrs);
+		svm_range_apply_attrs(p, prange, nattr, attrs, &update_mapping);
  		/* TODO: unmap ranges from GPU that lost access */
  	}
  	list_for_each_entry_safe(prange, next, &remove_list, update_list) {

drivers/gpu/drm/amd/amdkfd/kfd_svm.c svm_range_set_attr(struct kfd_process *p, struct mm_struct *mm, · line 3355

Quoted 10 lines
drivers/gpu/drm/amd/amdkfd/kfd_svm.c +7 -0
@@ -3335,8 +3346,15 @@ svm_range_set_attr(struct kfd_process *p, struct mm_struct *mm,
  			continue;
  		}
 
+		if (!migrated && !update_mapping) {
+			mutex_unlock(&prange->migrate_mutex);
+			continue;
+		}
+
+		flush_tlb = !migrated && update_mapping && prange->mapped_to_gpu;
+

Please check that this doesn't break the XNACK off case. If a new range is created, and that does not trigger a migration or any of the conditions that set update_mapping, we still need to make sure the GPU mapping is created so that a GPU access to the new range won't fault.

Regards,
  Felix

Quoted 6 lines
drivers/gpu/drm/amd/amdkfd/kfd_svm.c +1 -1
  		r = svm_range_validate_and_map(mm, prange, MAX_GPU_INSTANCE,
-					       true, true);
+					       true, true, flush_tlb);
  		if (r)
  			pr_debug("failed %d to map svm range\n", r);
 
maintainer Reviewed-by
#6 lore raw

Re: [PATCH v2 2/2] drm/amdkfd: Update mapping if range attributes changed

Am 2022-04-22 um 18:05 schrieb Felix Kuehling:

Quoted 165 lines · Philip Yang
Quoted 164 lines

Change SVM range mapping flags or access attributes don't trigger migration, if range is already mapped on GPUs we should update GPU mapping and pass flush_tlb flag true to amdgpu vm.

Change SVM range preferred_loc or migration granularity don't need update GPU mapping, skip the validate_and_map.

---
  drivers/gpu/drm/amd/amdkfd/kfd_svm.c | 46 +++++++++++++++++++---------

  1 file changed, 32 insertions(+), 14 deletions(-)
drivers/gpu/drm/amd/amdkfd/kfd_svm.c +0 -0
index 8a077cd066a1..e740384df9c7 100644
@@ -686,7 +686,8 @@ svm_range_check_attr(struct kfd_process *p,

    static void
  svm_range_apply_attrs(struct kfd_process *p, struct svm_range *prange, -              uint32_t nattr, struct kfd_ioctl_svm_attribute *attrs) +              uint32_t nattr, struct kfd_ioctl_svm_attribute *attrs, +              bool *update_mapping)
  {
      uint32_t i;
      int gpuidx;

drivers/gpu/drm/amd/amdkfd/kfd_svm.c +0 -0
@@ -702,6 +703,7 @@ svm_range_apply_attrs(struct kfd_process *p, 
 struct svm_range *prange,

          case KFD_IOCTL_SVM_ATTR_ACCESS:
          case KFD_IOCTL_SVM_ATTR_ACCESS_IN_PLACE:
          case KFD_IOCTL_SVM_ATTR_NO_ACCESS:
+            *update_mapping = true;
              gpuidx = kfd_process_gpuidx_from_gpuid(p, attrs[i].value); if (attrs[i].type == KFD_IOCTL_SVM_ATTR_NO_ACCESS) {

drivers/gpu/drm/amd/amdkfd/kfd_svm.c +0 -0
@@ -716,9 +718,11 @@ svm_range_apply_attrs(struct kfd_process *p, 
 struct svm_range *prange,
               }
               break;

          case KFD_IOCTL_SVM_ATTR_SET_FLAGS:
+            *update_mapping = true;
              prange->flags |= attrs[i].value;
              break;
          case KFD_IOCTL_SVM_ATTR_CLR_FLAGS:
+            *update_mapping = true;
              prange->flags &= ~attrs[i].value;
              break;
          case KFD_IOCTL_SVM_ATTR_GRANULARITY:

drivers/gpu/drm/amd/amdkfd/kfd_svm.c +2 -1
@@ -1254,7 +1258,7 @@ static int
   svm_range_map_to_gpu(struct kfd_process_device *pdd, struct svm_range *prange,
                unsigned long offset, unsigned long npages, bool readonly,
                dma_addr_t *dma_addr, struct amdgpu_device *bo_adev,
-             struct dma_fence **fence)
+             struct dma_fence **fence, bool flush_tlb)
   {
       struct amdgpu_device *adev = pdd->dev->adev;
       struct amdgpu_vm *vm = drm_priv_to_vm(pdd->drm_priv);
@@ -1292,7 +1296,7 @@ svm_range_map_to_gpu(struct kfd_process_device 
 *pdd, struct svm_range *prange,
                (last_domain == SVM_RANGE_VRAM_DOMAIN) ? 1 : 0,
                pte_flags);
   -        r = amdgpu_vm_update_range(adev, vm, false, false, false, NULL,
+        r = amdgpu_vm_update_range(adev, vm, false, false, 
 flush_tlb, NULL,
                          last_start, prange->start + i,

                         pte_flags, last_start - prange->start,

drivers/gpu/drm/amd/amdkfd/kfd_svm.c +0 -0
@@ -1326,7 +1330,7 @@ svm_range_map_to_gpu(struct kfd_process_device 
 *pdd, struct svm_range *prange,

  static int
  svm_range_map_to_gpus(struct svm_range *prange, unsigned long offset, unsigned long npages, bool readonly,
-              unsigned long *bitmap, bool wait)
+              unsigned long *bitmap, bool wait, bool flush_tlb) {
      struct kfd_process_device *pdd;
      struct amdgpu_device *bo_adev;

drivers/gpu/drm/amd/amdkfd/kfd_svm.c +2 -1
@@ -1361,7 +1365,8 @@ svm_range_map_to_gpus(struct svm_range *prange, 
 unsigned long offset,
             r = svm_range_map_to_gpu(pdd, prange, offset, npages, readonly,
                        prange->dma_addr[gpuidx],
-                     bo_adev, wait ? &fence : NULL);
+                     bo_adev, wait ? &fence : NULL,
+                     flush_tlb);
           if (r)
               break;
   @@ -1482,8 +1487,8 @@ static void *kfd_svm_page_owner(struct kfd_process *p, int32_t gpuidx)

   * 5. Release page table (and SVM BO) reservation
   */
  static int svm_range_validate_and_map(struct mm_struct *mm, -                      struct svm_range *prange,
-                      int32_t gpuidx, bool intr, bool wait) +                      struct svm_range *prange, int32_t gpuidx, +                      bool intr, bool wait, bool flush_tlb) {
      struct svm_validate_context ctx;
      unsigned long start, end, addr;

drivers/gpu/drm/amd/amdkfd/kfd_svm.c +6 -1
@@ -1522,8 +1527,12 @@ static int svm_range_validate_and_map(struct 
 mm_struct *mm,
                 prange->bitmap_aip, MAX_GPU_INSTANCE);
       }
   -    if (bitmap_empty(ctx.bitmap, MAX_GPU_INSTANCE))
-        return 0;
+    if (bitmap_empty(ctx.bitmap, MAX_GPU_INSTANCE)) {
+        if (!prange->mapped_to_gpu)
+            return 0;
+
+        bitmap_copy(ctx.bitmap, prange->bitmap_access, 
 MAX_GPU_INSTANCE);
+    }
         if (prange->actual_loc && !prange->ttm_res) {

          /* This should never happen. actual_loc gets set by

drivers/gpu/drm/amd/amdkfd/kfd_svm.c +1 -1
@@ -1595,7 +1604,7 @@ static int svm_range_validate_and_map(struct 
 mm_struct *mm,
           }
             r = svm_range_map_to_gpus(prange, offset, npages, readonly,
-                      ctx.bitmap, wait);
+                      ctx.bitmap, wait, flush_tlb);

    unlock_out:
          svm_range_unlock(prange);

drivers/gpu/drm/amd/amdkfd/kfd_svm.c +2 -1
@@ -1691,7 +1700,7 @@ static void svm_range_restore_work(struct 
 work_struct *work)
           mutex_lock(&prange->migrate_mutex);
             r = svm_range_validate_and_map(mm, prange, MAX_GPU_INSTANCE,
-                           false, true);
+                           false, true, false);
           if (r)
               pr_debug("failed %d to map 0x%lx to gpus\n", r,
                    prange->start);
@@ -2847,7 +2856,7 @@ svm_range_restore_pages(struct amdgpu_device 
 *adev, unsigned int pasid,
           }
       }
   -    r = svm_range_validate_and_map(mm, prange, gpuidx, false, false);
+    r = svm_range_validate_and_map(mm, prange, gpuidx, false, false, 
 false);
       if (r)

          pr_debug("failed %d to map svms 0x%p [0x%lx 0x%lx] to gpus\n", r, svms, prange->start, prange->last);

drivers/gpu/drm/amd/amdkfd/kfd_svm.c +3 -1
@@ -3264,6 +3273,8 @@ svm_range_set_attr(struct kfd_process *p, 
 struct mm_struct *mm,
       struct svm_range_list *svms;
       struct svm_range *prange;
       struct svm_range *next;
+    bool update_mapping = false;
+    bool flush_tlb;
       int r = 0;
         pr_debug("pasid 0x%x svms 0x%p [0x%llx 0x%llx] pages 0x%llx\n",
@@ -3302,7 +3313,7 @@ svm_range_set_attr(struct kfd_process *p, 
 struct mm_struct *mm,
           svm_range_add_notifier_locked(mm, prange);
       }
       list_for_each_entry(prange, &update_list, update_list) {
-        svm_range_apply_attrs(p, prange, nattr, attrs);
+        svm_range_apply_attrs(p, prange, nattr, attrs, 
 &update_mapping);
           /* TODO: unmap ranges from GPU that lost access */

      }
      list_for_each_entry_safe(prange, next, &remove_list, update_list) {

drivers/gpu/drm/amd/amdkfd/kfd_svm.c +1 -0
@@ -3335,8 +3346,15 @@ svm_range_set_attr(struct kfd_process *p, 
 struct mm_struct *mm,
               continue;
           }
   +        if (!migrated && !update_mapping) {
+            mutex_unlock(&prange->migrate_mutex);
Quoted 8 lines

drivers/gpu/drm/amd/amdkfd/kfd_svm.c svm_range_set_attr(struct kfd_process *p, struct mm_struct *mm, · line 3355

Quoted 6 lines
drivers/gpu/drm/amd/amdkfd/kfd_svm.c +5 -0
+            continue;
+        }
+
+        flush_tlb = !migrated && update_mapping && 
 prange->mapped_to_gpu;
+

Please check that this doesn't break the XNACK off case. If a new range is created, and that does not trigger a migration or any of the conditions that set update_mapping, we still need to make sure the GPU mapping is created so that a GPU access to the new range won't fault.

Thanks for following up offline. For the record, with XNACK off, the default is "not accessible". So to create a GPU mapping, the application has to set one of the accessibility attributes, which will trigger a page table update with your patch.

The patch is

Quoted 10 lines

Regards,
  Felix

Quoted 5 lines
drivers/gpu/drm/amd/amdkfd/kfd_svm.c +1 -1
           r = svm_range_validate_and_map(mm, prange, MAX_GPU_INSTANCE,
-                           true, true);
+                           true, true, flush_tlb);
           if (r)

              pr_debug("failed %d to map svm range\n", r);

maintainer Reviewed-by on v1 1/2
#7 lore raw
In reply to Philip Yang

Am 2022-04-19 um 20:47 schrieb Philip Yang:

Quoted 8 lines

To avoid unnecessary unmap SVM range from GPUs if range is not mapped on GPUs when migrating the range. This flag will also be used to flush TLB when updating the existing mapping on GPUs.

It is protected by prange->migrate_mutex and mmap read lock in MMU notifier callback.

Quoted 67 lines
---
  drivers/gpu/drm/amd/amdkfd/kfd_svm.c | 17 ++++++++++++++++-
  drivers/gpu/drm/amd/amdkfd/kfd_svm.h |  1 +
  2 files changed, 17 insertions(+), 1 deletion(-)
drivers/gpu/drm/amd/amdkfd/kfd_svm.c +16 -1
index 5afe216cf099..6be1695f3a09 100644
@@ -951,6 +951,7 @@ svm_range_split_adjust(struct svm_range *new, struct svm_range *old,
  	new->prefetch_loc = old->prefetch_loc;
  	new->actual_loc = old->actual_loc;
  	new->granularity = old->granularity;
+	new->mapped_to_gpu = old->mapped_to_gpu;
  	bitmap_copy(new->bitmap_access, old->bitmap_access, MAX_GPU_INSTANCE);
  	bitmap_copy(new->bitmap_aip, old->bitmap_aip, MAX_GPU_INSTANCE);
 
@@ -1204,6 +1205,17 @@ svm_range_unmap_from_gpus(struct svm_range *prange, unsigned long start,
  	uint32_t gpuidx;
  	int r = 0;
 
+	if (!prange->mapped_to_gpu) {
+		pr_debug("prange 0x%p [0x%lx 0x%lx] not mapped to GPU\n",
+			 prange, prange->start, prange->last);
+		return 0;
+	}
+
+	if (prange->start == start && prange->last == last) {
+		pr_debug("unmap svms 0x%p prange 0x%p\n", prange->svms, prange);
+		prange->mapped_to_gpu = false;
+	}
+
  	bitmap_or(bitmap, prange->bitmap_access, prange->bitmap_aip,
  		  MAX_GPU_INSTANCE);
  	p = container_of(prange->svms, struct kfd_process, svms);
@@ -1590,8 +1602,10 @@ static int svm_range_validate_and_map(struct mm_struct *mm,
  		addr = next;
  	}
 
-	if (addr == end)
+	if (addr == end) {
  		prange->validated_once = true;
+		prange->mapped_to_gpu = true;
+	}
 
  unreserve_out:
  	svm_range_unreserve_bos(&ctx);
@@ -1822,6 +1836,7 @@ static struct svm_range *svm_range_clone(struct svm_range *old)
  	new->prefetch_loc = old->prefetch_loc;
  	new->actual_loc = old->actual_loc;
  	new->granularity = old->granularity;
+	new->mapped_to_gpu = old->mapped_to_gpu;
  	bitmap_copy(new->bitmap_access, old->bitmap_access, MAX_GPU_INSTANCE);
  	bitmap_copy(new->bitmap_aip, old->bitmap_aip, MAX_GPU_INSTANCE);
 
drivers/gpu/drm/amd/amdkfd/kfd_svm.h +1 -0
index 66c77f00ac3e..2d54147b4dda 100644
@@ -133,6 +133,7 @@ struct svm_range {
  	DECLARE_BITMAP(bitmap_access, MAX_GPU_INSTANCE);
  	DECLARE_BITMAP(bitmap_aip, MAX_GPU_INSTANCE);
  	bool				validated_once;
+	bool				mapped_to_gpu;
  };
 
  static inline void svm_range_lock(struct svm_range *prange)