[PATCH 1/5] drm/amdkfd: device pgmap owner at the svm migrate init

[PATCH 1/5] drm/amdkfd: device pgmap owner at the svm migrate init

[Alex Sierra]

pgmap owner member at the svm migrate init could be referenced
to either adev or hive, depending on device topology.

Signed-off-by: Alex Sierra <alex.sierra@amd.com>
---
 drivers/gpu/drm/amd/amdkfd/kfd_migrate.c | 6 +++---
 drivers/gpu/drm/amd/amdkfd/kfd_svm.h     | 3 +++
 2 files changed, 6 insertions(+), 3 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_migrate.c b/drivers/gpu/drm/amd/amdkfd/kfd_migrate.c
index bc2fc528ad58..02340f36bd3d 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_migrate.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_migrate.c
@@ -426,7 +426,7 @@ svm_migrate_vma_to_vram(struct amdgpu_device *adev, struct svm_range *prange,
 	migrate.start = start;
 	migrate.end = end;
 	migrate.flags = MIGRATE_VMA_SELECT_SYSTEM;
-	migrate.pgmap_owner = adev;
+	migrate.pgmap_owner = SVM_ADEV_PGMAP_OWNER(adev);
 
 	size = 2 * sizeof(*migrate.src) + sizeof(uint64_t) + sizeof(dma_addr_t);
 	size *= npages;
@@ -641,7 +641,7 @@ svm_migrate_vma_to_ram(struct amdgpu_device *adev, struct svm_range *prange,
 	migrate.start = start;
 	migrate.end = end;
 	migrate.flags = MIGRATE_VMA_SELECT_DEVICE_PRIVATE;
-	migrate.pgmap_owner = adev;
+	migrate.pgmap_owner = SVM_ADEV_PGMAP_OWNER(adev);
 
 	size = 2 * sizeof(*migrate.src) + sizeof(uint64_t) + sizeof(dma_addr_t);
 	size *= npages;
@@ -907,7 +907,7 @@ int svm_migrate_init(struct amdgpu_device *adev)
 	pgmap->range.start = res->start;
 	pgmap->range.end = res->end;
 	pgmap->ops = &svm_migrate_pgmap_ops;
-	pgmap->owner = adev;
+	pgmap->owner = SVM_ADEV_PGMAP_OWNER(adev);
 	pgmap->flags = MIGRATE_VMA_SELECT_DEVICE_PRIVATE;
 	r = devm_memremap_pages(adev->dev, pgmap);
 	if (IS_ERR(r)) {
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_svm.h b/drivers/gpu/drm/amd/amdkfd/kfd_svm.h
index 573f984b81fe..4297250f259d 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_svm.h
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_svm.h
@@ -35,6 +35,9 @@
 #include "amdgpu.h"
 #include "kfd_priv.h"
 
+#define SVM_ADEV_PGMAP_OWNER(adev)\
+			((adev)->hive ? (void *)(adev)->hive : (void *)(adev))
+
 struct svm_range_bo {
 	struct amdgpu_bo		*bo;
 	struct kref			kref;
-- 
2.17.1

_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

[PATCH 2/5] drm/amdkfd: add owner ref param to get hmm pages

[Alex Sierra]

The parameter is used in the dev_private_owner to decide if device
pages in the range require to be migrated back to system memory, based
if they are or not in the same memory domain.
In this case, this reference could come from the same memory domain
with devices connected to the same hive.

Signed-off-by: Alex Sierra <alex.sierra@amd.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_mn.c  | 3 ++-
 drivers/gpu/drm/amd/amdgpu/amdgpu_mn.h  | 2 +-
 drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c | 2 +-
 drivers/gpu/drm/amd/amdkfd/kfd_svm.c    | 4 ++--
 4 files changed, 6 insertions(+), 5 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_mn.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_mn.c
index 2741c28ff1b5..378c238c2099 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_mn.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_mn.c
@@ -160,7 +160,7 @@ int amdgpu_hmm_range_get_pages(struct mmu_interval_notifier *notifier,
 			       struct mm_struct *mm, struct page **pages,
 			       uint64_t start, uint64_t npages,
 			       struct hmm_range **phmm_range, bool readonly,
-			       bool mmap_locked)
+			       bool mmap_locked, void *owner)
 {
 	struct hmm_range *hmm_range;
 	unsigned long timeout;
@@ -185,6 +185,7 @@ int amdgpu_hmm_range_get_pages(struct mmu_interval_notifier *notifier,
 	hmm_range->hmm_pfns = pfns;
 	hmm_range->start = start;
 	hmm_range->end = start + npages * PAGE_SIZE;
+	hmm_range->dev_private_owner = owner;
 
 	/* Assuming 512MB takes maxmium 1 second to fault page address */
 	timeout = max(npages >> 17, 1ULL) * HMM_RANGE_DEFAULT_TIMEOUT;
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_mn.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_mn.h
index 7f7d37a457c3..14a3c1864085 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_mn.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_mn.h
@@ -34,7 +34,7 @@ int amdgpu_hmm_range_get_pages(struct mmu_interval_notifier *notifier,
 			       struct mm_struct *mm, struct page **pages,
 			       uint64_t start, uint64_t npages,
 			       struct hmm_range **phmm_range, bool readonly,
-			       bool mmap_locked);
+			       bool mmap_locked, void *owner);
 int amdgpu_hmm_range_get_pages_done(struct hmm_range *hmm_range);
 
 #if defined(CONFIG_HMM_MIRROR)
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
index 7e7d8330d64b..c13f7fbfc070 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
@@ -709,7 +709,7 @@ int amdgpu_ttm_tt_get_user_pages(struct amdgpu_bo *bo, struct page **pages)
 	readonly = amdgpu_ttm_tt_is_readonly(ttm);
 	r = amdgpu_hmm_range_get_pages(&bo->notifier, mm, pages, start,
 				       ttm->num_pages, &gtt->range, readonly,
-				       false);
+				       false, NULL);
 out_putmm:
 	mmput(mm);
 
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_svm.c b/drivers/gpu/drm/amd/amdkfd/kfd_svm.c
index a165e51c4a1c..540496d27c07 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_svm.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_svm.c
@@ -1386,7 +1386,7 @@ static int svm_range_validate_and_map(struct mm_struct *mm,
 		r = amdgpu_hmm_range_get_pages(&prange->notifier, mm, NULL,
 					       prange->start << PAGE_SHIFT,
 					       prange->npages, &hmm_range,
-					       false, true);
+					       false, true, NULL);
 		if (r) {
 			pr_debug("failed %d to get svm range pages\n", r);
 			goto unreserve_out;
@@ -2653,7 +2653,7 @@ void svm_range_prefault(struct svm_range *prange, struct mm_struct *mm)
 	r = amdgpu_hmm_range_get_pages(&prange->notifier, mm, NULL,
 				       prange->start << PAGE_SHIFT,
 				       prange->npages, &hmm_range,
-				       false, true);
+				       false, true, NULL);
 	if (!r) {
 		amdgpu_hmm_range_get_pages_done(hmm_range);
 		prange->validated_once = true;
-- 
2.17.1

_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

[PATCH 3/5] drm/amdkfd: set owner ref to svm range prefault

[Alex Sierra]

svm_range_prefault is called right before migrations to VRAM,
to make sure pages are resident in system memory before the migration.
With partial migrations, this reference is used by hmm range get pages
to avoid migrating pages that are already in the same VRAM domain.

Signed-off-by: Alex Sierra <alex.sierra@amd.com>
---
 drivers/gpu/drm/amd/amdkfd/kfd_migrate.c | 2 +-
 drivers/gpu/drm/amd/amdkfd/kfd_svm.c     | 5 +++--
 drivers/gpu/drm/amd/amdkfd/kfd_svm.h     | 3 ++-
 3 files changed, 6 insertions(+), 4 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_migrate.c b/drivers/gpu/drm/amd/amdkfd/kfd_migrate.c
index 02340f36bd3d..231ba60f130f 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_migrate.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_migrate.c
@@ -512,7 +512,7 @@ svm_migrate_ram_to_vram(struct svm_range *prange, uint32_t best_loc,
 		 prange->start, prange->last, best_loc);
 
 	/* FIXME: workaround for page locking bug with invalid pages */
-	svm_range_prefault(prange, mm);
+	svm_range_prefault(prange, mm, SVM_ADEV_PGMAP_OWNER(adev));
 
 	start = prange->start << PAGE_SHIFT;
 	end = (prange->last + 1) << PAGE_SHIFT;
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_svm.c b/drivers/gpu/drm/amd/amdkfd/kfd_svm.c
index 540496d27c07..f8537940ab22 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_svm.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_svm.c
@@ -2642,7 +2642,8 @@ svm_range_best_prefetch_location(struct svm_range *prange)
 /* FIXME: This is a workaround for page locking bug when some pages are
  * invalid during migration to VRAM
  */
-void svm_range_prefault(struct svm_range *prange, struct mm_struct *mm)
+void svm_range_prefault(struct svm_range *prange, struct mm_struct *mm,
+			void *owner)
 {
 	struct hmm_range *hmm_range;
 	int r;
@@ -2653,7 +2654,7 @@ void svm_range_prefault(struct svm_range *prange, struct mm_struct *mm)
 	r = amdgpu_hmm_range_get_pages(&prange->notifier, mm, NULL,
 				       prange->start << PAGE_SHIFT,
 				       prange->npages, &hmm_range,
-				       false, true, NULL);
+				       false, true, owner);
 	if (!r) {
 		amdgpu_hmm_range_get_pages_done(hmm_range);
 		prange->validated_once = true;
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_svm.h b/drivers/gpu/drm/amd/amdkfd/kfd_svm.h
index 4297250f259d..08542fe39303 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_svm.h
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_svm.h
@@ -176,7 +176,8 @@ void schedule_deferred_list_work(struct svm_range_list *svms);
 void svm_range_dma_unmap(struct device *dev, dma_addr_t *dma_addr,
 			 unsigned long offset, unsigned long npages);
 void svm_range_free_dma_mappings(struct svm_range *prange);
-void svm_range_prefault(struct svm_range *prange, struct mm_struct *mm);
+void svm_range_prefault(struct svm_range *prange, struct mm_struct *mm,
+			void *owner);
 
 #else
 
-- 
2.17.1

_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

[PATCH 4/5] drm/amdgpu: get owner ref in validate and map

[Alex Sierra]

Get the proper owner reference for amdgpu_hmm_range_get_pages function.
This is useful for partial migrations. To avoid migrating back to
system memory, VRAM pages, that are accessible by all devices in the
same memory domain.
Ex. multiple devices in the same hive.

Signed-off-by: Alex Sierra <alex.sierra@amd.com>
---
 drivers/gpu/drm/amd/amdkfd/kfd_svm.c | 25 ++++++++++++++++++++++++-
 1 file changed, 24 insertions(+), 1 deletion(-)

diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_svm.c b/drivers/gpu/drm/amd/amdkfd/kfd_svm.c
index f8537940ab22..940165f4437d 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_svm.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_svm.c
@@ -1307,6 +1307,17 @@ static void svm_range_unreserve_bos(struct svm_validate_context *ctx)
 	ttm_eu_backoff_reservation(&ctx->ticket, &ctx->validate_list);
 }
 
+static void *kfd_svm_page_owner(struct kfd_process *p, int32_t gpuidx)
+{
+	struct kfd_process_device *pdd;
+	struct amdgpu_device *adev;
+
+	pdd = kfd_process_device_from_gpuidx(p, gpuidx);
+	adev = (struct amdgpu_device *)pdd->dev->kgd;
+
+	return SVM_ADEV_PGMAP_OWNER(adev);
+}
+
 /*
  * Validation+GPU mapping with concurrent invalidation (MMU notifiers)
  *
@@ -1337,6 +1348,9 @@ static int svm_range_validate_and_map(struct mm_struct *mm,
 {
 	struct svm_validate_context ctx;
 	struct hmm_range *hmm_range;
+	struct kfd_process *p;
+	void *owner;
+	int32_t idx;
 	int r = 0;
 
 	ctx.process = container_of(prange->svms, struct kfd_process, svms);
@@ -1383,10 +1397,19 @@ static int svm_range_validate_and_map(struct mm_struct *mm,
 	svm_range_reserve_bos(&ctx);
 
 	if (!prange->actual_loc) {
+		p = container_of(prange->svms, struct kfd_process, svms);
+		owner = kfd_svm_page_owner(p, find_first_bit(ctx.bitmap,
+							MAX_GPU_INSTANCE));
+		for_each_set_bit(idx, ctx.bitmap, MAX_GPU_INSTANCE) {
+			if (kfd_svm_page_owner(p, idx) != owner) {
+				owner = NULL;
+				break;
+			}
+		}
 		r = amdgpu_hmm_range_get_pages(&prange->notifier, mm, NULL,
 					       prange->start << PAGE_SHIFT,
 					       prange->npages, &hmm_range,
-					       false, true, NULL);
+					       false, true, owner);
 		if (r) {
 			pr_debug("failed %d to get svm range pages\n", r);
 			goto unreserve_out;
-- 
2.17.1

_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

[PATCH 5/5] drm/amdkfd: classify and map mixed svm range pages in GPU

[Alex Sierra]

[Why]
svm ranges can have mixed pages from device or system memory.
A good example is, after a prange has been allocated in VRAM and a
copy-on-write is triggered by a fork. This invalidates some pages
inside the prange. Endding up in mixed pages.

[How]
By classifying each page inside a prange, based on its type. Device or
system memory, during dma mapping call. If page corresponds
to VRAM domain, a flag is set to its dma_addr entry for each GPU.
Then, at the GPU page table mapping. All group of contiguous pages within
the same type are mapped with their proper pte flags.

Signed-off-by: Alex Sierra <alex.sierra@amd.com>
---
 drivers/gpu/drm/amd/amdkfd/kfd_svm.c | 51 ++++++++++++++++++----------
 drivers/gpu/drm/amd/amdkfd/kfd_svm.h |  1 +
 2 files changed, 35 insertions(+), 17 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_svm.c b/drivers/gpu/drm/amd/amdkfd/kfd_svm.c
index 940165f4437d..1440e4e555f1 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_svm.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_svm.c
@@ -141,6 +141,12 @@ svm_range_dma_map_dev(struct device *dev, dma_addr_t **dma_addr,
 			dma_unmap_page(dev, addr[i], PAGE_SIZE, dir);
 
 		page = hmm_pfn_to_page(hmm_pfns[i]);
+		if (is_zone_device_page(page)) {
+			addr[i] = hmm_pfns[i] << PAGE_SHIFT;
+			addr[i] |= SVM_RANGE_VRAM_DOMAIN;
+			pr_debug("vram address detected: 0x%llx\n", addr[i] >> PAGE_SHIFT);
+			continue;
+		}
 		addr[i] = dma_map_page(dev, page, 0, PAGE_SIZE, dir);
 		r = dma_mapping_error(dev, addr[i]);
 		if (r) {
@@ -1131,32 +1137,43 @@ svm_range_map_to_gpu(struct amdgpu_device *adev, struct amdgpu_vm *vm,
 		     struct amdgpu_device *bo_adev, struct dma_fence **fence)
 {
 	struct amdgpu_bo_va bo_va;
+	struct ttm_resource *ttm_res;
 	uint64_t pte_flags;
+	unsigned long last_start;
+	int last_domain;
 	int r = 0;
+	int64_t i;
 
 	pr_debug("svms 0x%p [0x%lx 0x%lx]\n", prange->svms, prange->start,
 		 prange->last);
 
-	if (prange->svm_bo && prange->ttm_res) {
+	if (prange->svm_bo && prange->ttm_res)
 		bo_va.is_xgmi = amdgpu_xgmi_same_hive(adev, bo_adev);
-		prange->mapping.bo_va = &bo_va;
-	}
 
-	prange->mapping.start = prange->start;
-	prange->mapping.last = prange->last;
-	prange->mapping.offset = prange->offset;
-	pte_flags = svm_range_get_pte_flags(adev, prange);
+	last_start = prange->start;
+	for (i = 0; i < prange->npages; i++) {
+		last_domain = dma_addr[i] & SVM_RANGE_VRAM_DOMAIN;
+		if ((prange->start + i) < prange->last &&
+		    last_domain == (dma_addr[i + 1] & SVM_RANGE_VRAM_DOMAIN))
+			continue;
 
-	r = amdgpu_vm_bo_update_mapping(adev, bo_adev, vm, false, false, NULL,
-					prange->mapping.start,
-					prange->mapping.last, pte_flags,
-					prange->mapping.offset,
-					prange->ttm_res ?
-						prange->ttm_res->mm_node : NULL,
-					dma_addr, &vm->last_update);
-	if (r) {
-		pr_debug("failed %d to map to gpu 0x%lx\n", r, prange->start);
-		goto out;
+		pr_debug("Mapping range [0x%lx 0x%llx] on domain: %s\n",
+			 last_start, prange->start + i, last_domain ? "GPU" : "CPU");
+		ttm_res = last_domain ? prange->ttm_res : NULL;
+		pte_flags = svm_range_get_pte_flags(adev, prange);
+		r = amdgpu_vm_bo_update_mapping(adev, bo_adev, vm, false, false, NULL,
+						last_start,
+						prange->start + i, pte_flags,
+						prange->offset +
+						((last_start - prange->start) << PAGE_SHIFT),
+						ttm_res ? ttm_res->mm_node : NULL,
+						ttm_res ? NULL : dma_addr,
+						&vm->last_update);
+		if (r) {
+			pr_debug("failed %d to map to gpu 0x%lx\n", r, prange->start);
+			goto out;
+		}
+		last_start += prange->start + i + 1;
 	}
 
 	r = amdgpu_vm_update_pdes(adev, vm, false);
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_svm.h b/drivers/gpu/drm/amd/amdkfd/kfd_svm.h
index 08542fe39303..e68aa51322df 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_svm.h
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_svm.h
@@ -35,6 +35,7 @@
 #include "amdgpu.h"
 #include "kfd_priv.h"
 
+#define SVM_RANGE_VRAM_DOMAIN (1UL << 0)
 #define SVM_ADEV_PGMAP_OWNER(adev)\
 			((adev)->hive ? (void *)(adev)->hive : (void *)(adev))
 
-- 
2.17.1

_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

Re: [PATCH 5/5] drm/amdkfd: classify and map mixed svm range pages in GPU

[Felix Kuehling]

Am 2021-05-06 um 11:36 p.m. schrieb Alex Sierra:
> [Why]
> svm ranges can have mixed pages from device or system memory.
> A good example is, after a prange has been allocated in VRAM and a
> copy-on-write is triggered by a fork. This invalidates some pages
> inside the prange. Endding up in mixed pages.
>
> [How]
> By classifying each page inside a prange, based on its type. Device or
> system memory, during dma mapping call. If page corresponds
> to VRAM domain, a flag is set to its dma_addr entry for each GPU.
> Then, at the GPU page table mapping. All group of contiguous pages within
> the same type are mapped with their proper pte flags.
>
> Signed-off-by: Alex Sierra <alex.sierra@amd.com>
> ---
>  drivers/gpu/drm/amd/amdkfd/kfd_svm.c | 51 ++++++++++++++++++----------
>  drivers/gpu/drm/amd/amdkfd/kfd_svm.h |  1 +
>  2 files changed, 35 insertions(+), 17 deletions(-)
>
> diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_svm.c b/drivers/gpu/drm/amd/amdkfd/kfd_svm.c
> index 940165f4437d..1440e4e555f1 100644
> --- a/drivers/gpu/drm/amd/amdkfd/kfd_svm.c
> +++ b/drivers/gpu/drm/amd/amdkfd/kfd_svm.c
> @@ -141,6 +141,12 @@ svm_range_dma_map_dev(struct device *dev, dma_addr_t **dma_addr,
>  			dma_unmap_page(dev, addr[i], PAGE_SIZE, dir);
>  
>  		page = hmm_pfn_to_page(hmm_pfns[i]);
> +		if (is_zone_device_page(page)) {
> +			addr[i] = hmm_pfns[i] << PAGE_SHIFT;
> +			addr[i] |= SVM_RANGE_VRAM_DOMAIN;
> +			pr_debug("vram address detected: 0x%llx\n", addr[i] >> PAGE_SHIFT);
> +			continue;
> +		}
>  		addr[i] = dma_map_page(dev, page, 0, PAGE_SIZE, dir);
>  		r = dma_mapping_error(dev, addr[i]);
>  		if (r) {
> @@ -1131,32 +1137,43 @@ svm_range_map_to_gpu(struct amdgpu_device *adev, struct amdgpu_vm *vm,
>  		     struct amdgpu_device *bo_adev, struct dma_fence **fence)
>  {
>  	struct amdgpu_bo_va bo_va;
> +	struct ttm_resource *ttm_res;
>  	uint64_t pte_flags;
> +	unsigned long last_start;
> +	int last_domain;
>  	int r = 0;
> +	int64_t i;
>  
>  	pr_debug("svms 0x%p [0x%lx 0x%lx]\n", prange->svms, prange->start,
>  		 prange->last);
>  
> -	if (prange->svm_bo && prange->ttm_res) {
> +	if (prange->svm_bo && prange->ttm_res)
>  		bo_va.is_xgmi = amdgpu_xgmi_same_hive(adev, bo_adev);
> -		prange->mapping.bo_va = &bo_va;
> -	}
>  
> -	prange->mapping.start = prange->start;
> -	prange->mapping.last = prange->last;
> -	prange->mapping.offset = prange->offset;
> -	pte_flags = svm_range_get_pte_flags(adev, prange);
> +	last_start = prange->start;
> +	for (i = 0; i < prange->npages; i++) {
> +		last_domain = dma_addr[i] & SVM_RANGE_VRAM_DOMAIN;
> +		if ((prange->start + i) < prange->last &&
> +		    last_domain == (dma_addr[i + 1] & SVM_RANGE_VRAM_DOMAIN))
> +			continue;
>  
> -	r = amdgpu_vm_bo_update_mapping(adev, bo_adev, vm, false, false, NULL,
> -					prange->mapping.start,
> -					prange->mapping.last, pte_flags,
> -					prange->mapping.offset,
> -					prange->ttm_res ?
> -						prange->ttm_res->mm_node : NULL,
> -					dma_addr, &vm->last_update);
> -	if (r) {
> -		pr_debug("failed %d to map to gpu 0x%lx\n", r, prange->start);
> -		goto out;
> +		pr_debug("Mapping range [0x%lx 0x%llx] on domain: %s\n",
> +			 last_start, prange->start + i, last_domain ? "GPU" : "CPU");
> +		ttm_res = last_domain ? prange->ttm_res : NULL;
> +		pte_flags = svm_range_get_pte_flags(adev, prange);
> +		r = amdgpu_vm_bo_update_mapping(adev, bo_adev, vm, false, false, NULL,
> +						last_start,
> +						prange->start + i, pte_flags,
> +						prange->offset +
> +						((last_start - prange->start) << PAGE_SHIFT),
> +						ttm_res ? ttm_res->mm_node : NULL,
> +						ttm_res ? NULL : dma_addr,
> +						&vm->last_update);
> +		if (r) {
> +			pr_debug("failed %d to map to gpu 0x%lx\n", r, prange->start);
> +			goto out;
> +		}
> +		last_start += prange->start + i + 1;

This looks wrong. It's either

    last_start += i + 1;

or

    last_start = prange->start + i + 1;

With that fixed, the series is
Reviewed-by: Felix Kuehling <Felix.Kuehling@amd.com>


>  	}
>  
>  	r = amdgpu_vm_update_pdes(adev, vm, false);
> diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_svm.h b/drivers/gpu/drm/amd/amdkfd/kfd_svm.h
> index 08542fe39303..e68aa51322df 100644
> --- a/drivers/gpu/drm/amd/amdkfd/kfd_svm.h
> +++ b/drivers/gpu/drm/amd/amdkfd/kfd_svm.h
> @@ -35,6 +35,7 @@
>  #include "amdgpu.h"
>  #include "kfd_priv.h"
>  
> +#define SVM_RANGE_VRAM_DOMAIN (1UL << 0)
>  #define SVM_ADEV_PGMAP_OWNER(adev)\
>  			((adev)->hive ? (void *)(adev)->hive : (void *)(adev))
>  
_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx