Blob Blame History Raw
From 3d3c596d0363c675391d3cf547189d2ebe0e82c2 Mon Sep 17 00:00:00 2001
From: Philip Yang <Philip.Yang@amd.com>
Date: Fri, 11 Mar 2022 15:22:15 -0500
Subject: drm/amdkfd: evict svm bo worker handle error
Git-commit: 9527b9caf82b040e794fba2667f6d8a85a793bb7
Patch-mainline: v5.18-rc1
References: jsc#PED-1166 jsc#PED-1168 jsc#PED-1170 jsc#PED-1218 jsc#PED-1220 jsc#PED-1222 jsc#PED-1223 jsc#PED-1225

Migrate vram to ram may fail to find the vma if process is exiting
and vma is removed, evict svm bo worker sets prange->svm_bo to NULL
and warn svm_bo ref count != 1 only if migrating vram to ram
successfully.

Signed-off-by: Philip Yang <Philip.Yang@amd.com>
Reviewed-by: Felix Kuehling <Felix.Kuehling@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Acked-by: Patrik Jakobsson <pjakobsson@suse.de>
---
 drivers/gpu/drm/amd/amdkfd/kfd_migrate.c | 29 +++++++++++++++++++-----
 drivers/gpu/drm/amd/amdkfd/kfd_svm.c     | 21 ++++++++++-------
 2 files changed, 36 insertions(+), 14 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_migrate.c b/drivers/gpu/drm/amd/amdkfd/kfd_migrate.c
index de36765013f8..7e3a7fcb9fe6 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_migrate.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_migrate.c
@@ -638,6 +638,22 @@ svm_migrate_copy_to_ram(struct amdgpu_device *adev, struct svm_range *prange,
 	return r;
 }
 
+/**
+ * svm_migrate_vma_to_ram - migrate range inside one vma from device to system
+ *
+ * @adev: amdgpu device to migrate from
+ * @prange: svm range structure
+ * @vma: vm_area_struct that range [start, end] belongs to
+ * @start: range start virtual address in pages
+ * @end: range end virtual address in pages
+ *
+ * Context: Process context, caller hold mmap read lock, prange->migrate_mutex
+ *
+ * Return:
+ *   0 - success with all pages migrated
+ *   negative values - indicate error
+ *   positive values - partial migration, number of pages not migrated
+ */
 static long
 svm_migrate_vma_to_ram(struct amdgpu_device *adev, struct svm_range *prange,
 		       struct vm_area_struct *vma, uint64_t start, uint64_t end)
@@ -709,8 +725,6 @@ svm_migrate_vma_to_ram(struct amdgpu_device *adev, struct svm_range *prange,
 		pdd = svm_range_get_pdd_by_adev(prange, adev);
 		if (pdd)
 			WRITE_ONCE(pdd->page_out, pdd->page_out + cpages);
-
-		return upages;
 	}
 	return r ? r : upages;
 }
@@ -720,7 +734,7 @@ svm_migrate_vma_to_ram(struct amdgpu_device *adev, struct svm_range *prange,
  * @prange: range structure
  * @mm: process mm, use current->mm if NULL
  *
- * Context: Process context, caller hold mmap read lock, svms lock, prange lock
+ * Context: Process context, caller hold mmap read lock, prange->migrate_mutex
  *
  * Return:
  * 0 - OK, otherwise error code
@@ -759,13 +773,16 @@ int svm_migrate_vram_to_ram(struct svm_range *prange, struct mm_struct *mm)
 		unsigned long next;
 
 		vma = find_vma(mm, addr);
-		if (!vma || addr < vma->vm_start)
+		if (!vma || addr < vma->vm_start) {
+			pr_debug("failed to find vma for prange %p\n", prange);
+			r = -EFAULT;
 			break;
+		}
 
 		next = min(vma->vm_end, end);
 		r = svm_migrate_vma_to_ram(adev, prange, vma, addr, next);
 		if (r < 0) {
-			pr_debug("failed %ld to migrate\n", r);
+			pr_debug("failed %ld to migrate prange %p\n", r, prange);
 			break;
 		} else {
 			upages += r;
@@ -773,7 +790,7 @@ int svm_migrate_vram_to_ram(struct svm_range *prange, struct mm_struct *mm)
 		addr = next;
 	}
 
-	if (!upages) {
+	if (r >= 0 && !upages) {
 		svm_range_vram_node_free(prange);
 		prange->actual_loc = 0;
 	}
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_svm.c b/drivers/gpu/drm/amd/amdkfd/kfd_svm.c
index 509d915cbe69..3b8856b4cece 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_svm.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_svm.c
@@ -3155,6 +3155,7 @@ static void svm_range_evict_svm_bo_worker(struct work_struct *work)
 	struct svm_range_bo *svm_bo;
 	struct kfd_process *p;
 	struct mm_struct *mm;
+	int r = 0;
 
 	svm_bo = container_of(work, struct svm_range_bo, eviction_work);
 	if (!svm_bo_ref_unless_zero(svm_bo))
@@ -3170,7 +3171,7 @@ static void svm_range_evict_svm_bo_worker(struct work_struct *work)
 
 	mmap_read_lock(mm);
 	spin_lock(&svm_bo->list_lock);
-	while (!list_empty(&svm_bo->range_list)) {
+	while (!list_empty(&svm_bo->range_list) && !r) {
 		struct svm_range *prange =
 				list_first_entry(&svm_bo->range_list,
 						struct svm_range, svm_bo_list);
@@ -3184,15 +3185,18 @@ static void svm_range_evict_svm_bo_worker(struct work_struct *work)
 
 		mutex_lock(&prange->migrate_mutex);
 		do {
-			svm_migrate_vram_to_ram(prange,
+			r = svm_migrate_vram_to_ram(prange,
 						svm_bo->eviction_fence->mm);
-		} while (prange->actual_loc && --retries);
-		WARN(prange->actual_loc, "Migration failed during eviction");
+		} while (!r && prange->actual_loc && --retries);
 
-		mutex_lock(&prange->lock);
-		prange->svm_bo = NULL;
-		mutex_unlock(&prange->lock);
+		if (!r && prange->actual_loc)
+			pr_info_once("Migration failed during eviction");
 
+		if (!prange->actual_loc) {
+			mutex_lock(&prange->lock);
+			prange->svm_bo = NULL;
+			mutex_unlock(&prange->lock);
+		}
 		mutex_unlock(&prange->migrate_mutex);
 
 		spin_lock(&svm_bo->list_lock);
@@ -3201,10 +3205,11 @@ static void svm_range_evict_svm_bo_worker(struct work_struct *work)
 	mmap_read_unlock(mm);
 
 	dma_fence_signal(&svm_bo->eviction_fence->base);
+
 	/* This is the last reference to svm_bo, after svm_range_vram_node_free
 	 * has been called in svm_migrate_vram_to_ram
 	 */
-	WARN_ONCE(kref_read(&svm_bo->kref) != 1, "This was not the last reference\n");
+	WARN_ONCE(!r && kref_read(&svm_bo->kref) != 1, "This was not the last reference\n");
 	svm_range_bo_unref(svm_bo);
 }
 
-- 
2.38.1