[PATCH 1/3] drm/amdgpu: remove acc_size from reserve/unreserve mem

* [PATCH 1/3] drm/amdgpu: remove acc_size from reserve/unreserve mem
@ 2022-07-12  1:56 Alex Sierra
  2022-07-12  1:56 ` [PATCH 2/3] drm/amdkfd: track unified memory reservation with xnack off Alex Sierra
  2022-07-12  1:56 ` [PATCH 3/3] drm/amdgpu: add debugfs for kfd system and ttm mem used Alex Sierra
  0 siblings, 2 replies; 8+ messages in thread
From: Alex Sierra @ 2022-07-12  1:56 UTC (permalink / raw)
  To: amd-gfx; +Cc: Alex Sierra, Philip Yang, Felix Kuehling, Christian König

TTM used to track the "acc_size" of all BOs internally. We needed to
keep track of it in our memory reservation to avoid TTM running out
of memory in its own accounting. However, that "acc_size" accounting
has since been removed from TTM. Therefore we don't really need to
track it any more.

Signed-off-by: Alex Sierra <alex.sierra@amd.com>
Reviewed-by: Philip Yang <philip.yang@amd.com>
Reviewed-by: Felix Kuehling <Felix.Kuehling@amd.com>
Acked-by: Christian König <christian.koenig@amd.com>
---
 .../gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c  | 60 ++++++-------------
 1 file changed, 17 insertions(+), 43 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c
index 4effee12a4ac..2bc36ff0aa0f 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c
@@ -115,21 +115,12 @@ void amdgpu_amdkfd_reserve_system_mem(uint64_t size)
  * compromise that should work in most cases without reserving too
  * much memory for page tables unnecessarily (factor 16K, >> 14).
  */
-#define ESTIMATE_PT_SIZE(mem_size) max(((mem_size) >> 14), AMDGPU_VM_RESERVED_VRAM)
-
-static size_t amdgpu_amdkfd_acc_size(uint64_t size)
-{
-	size >>= PAGE_SHIFT;
-	size *= sizeof(dma_addr_t) + sizeof(void *);
 
-	return __roundup_pow_of_two(sizeof(struct amdgpu_bo)) +
-		__roundup_pow_of_two(sizeof(struct ttm_tt)) +
-		PAGE_ALIGN(size);
-}
+#define ESTIMATE_PT_SIZE(mem_size) max(((mem_size) >> 14), AMDGPU_VM_RESERVED_VRAM)
 
 /**
  * amdgpu_amdkfd_reserve_mem_limit() - Decrease available memory by size
- * of buffer including any reserved for control structures
+ * of buffer.
  *
  * @adev: Device to which allocated BO belongs to
  * @size: Size of buffer, in bytes, encapsulated by B0. This should be
@@ -143,19 +134,16 @@ static int amdgpu_amdkfd_reserve_mem_limit(struct amdgpu_device *adev,
 {
 	uint64_t reserved_for_pt =
 		ESTIMATE_PT_SIZE(amdgpu_amdkfd_total_mem_size);
-	size_t acc_size, system_mem_needed, ttm_mem_needed, vram_needed;
+	size_t system_mem_needed, ttm_mem_needed, vram_needed;
 	int ret = 0;
 
-	acc_size = amdgpu_amdkfd_acc_size(size);
-
+	system_mem_needed = 0;
+	ttm_mem_needed = 0;
 	vram_needed = 0;
 	if (alloc_flag & KFD_IOC_ALLOC_MEM_FLAGS_GTT) {
-		system_mem_needed = acc_size + size;
-		ttm_mem_needed = acc_size + size;
+		system_mem_needed = size;
+		ttm_mem_needed = size;
 	} else if (alloc_flag & KFD_IOC_ALLOC_MEM_FLAGS_VRAM) {
-		system_mem_needed = acc_size;
-		ttm_mem_needed = acc_size;
-
 		/*
 		 * Conservatively round up the allocation requirement to 2 MB
 		 * to avoid fragmentation caused by 4K allocations in the tail
@@ -163,14 +151,10 @@ static int amdgpu_amdkfd_reserve_mem_limit(struct amdgpu_device *adev,
 		 */
 		vram_needed = ALIGN(size, VRAM_ALLOCATION_ALIGN);
 	} else if (alloc_flag & KFD_IOC_ALLOC_MEM_FLAGS_USERPTR) {
-		system_mem_needed = acc_size + size;
-		ttm_mem_needed = acc_size;
-	} else if (alloc_flag &
-		   (KFD_IOC_ALLOC_MEM_FLAGS_DOORBELL |
-		    KFD_IOC_ALLOC_MEM_FLAGS_MMIO_REMAP)) {
-		system_mem_needed = acc_size;
-		ttm_mem_needed = acc_size;
-	} else {
+		system_mem_needed = size;
+	} else if (!(alloc_flag &
+				(KFD_IOC_ALLOC_MEM_FLAGS_DOORBELL |
+				 KFD_IOC_ALLOC_MEM_FLAGS_MMIO_REMAP))) {
 		pr_err("%s: Invalid BO type %#x\n", __func__, alloc_flag);
 		return -ENOMEM;
 	}
@@ -208,28 +192,18 @@ static int amdgpu_amdkfd_reserve_mem_limit(struct amdgpu_device *adev,
 static void unreserve_mem_limit(struct amdgpu_device *adev,
 		uint64_t size, u32 alloc_flag)
 {
-	size_t acc_size;
-
-	acc_size = amdgpu_amdkfd_acc_size(size);
-
 	spin_lock(&kfd_mem_limit.mem_limit_lock);
 
 	if (alloc_flag & KFD_IOC_ALLOC_MEM_FLAGS_GTT) {
-		kfd_mem_limit.system_mem_used -= (acc_size + size);
-		kfd_mem_limit.ttm_mem_used -= (acc_size + size);
+		kfd_mem_limit.system_mem_used -= size;
+		kfd_mem_limit.ttm_mem_used -= size;
 	} else if (alloc_flag & KFD_IOC_ALLOC_MEM_FLAGS_VRAM) {
-		kfd_mem_limit.system_mem_used -= acc_size;
-		kfd_mem_limit.ttm_mem_used -= acc_size;
 		adev->kfd.vram_used -= ALIGN(size, VRAM_ALLOCATION_ALIGN);
 	} else if (alloc_flag & KFD_IOC_ALLOC_MEM_FLAGS_USERPTR) {
-		kfd_mem_limit.system_mem_used -= (acc_size + size);
-		kfd_mem_limit.ttm_mem_used -= acc_size;
-	} else if (alloc_flag &
-		   (KFD_IOC_ALLOC_MEM_FLAGS_DOORBELL |
-		    KFD_IOC_ALLOC_MEM_FLAGS_MMIO_REMAP)) {
-		kfd_mem_limit.system_mem_used -= acc_size;
-		kfd_mem_limit.ttm_mem_used -= acc_size;
-	} else {
+		kfd_mem_limit.system_mem_used -= size;
+	} else if (!(alloc_flag &
+				(KFD_IOC_ALLOC_MEM_FLAGS_DOORBELL |
+				 KFD_IOC_ALLOC_MEM_FLAGS_MMIO_REMAP))) {
 		pr_err("%s: Invalid BO type %#x\n", __func__, alloc_flag);
 		goto release;
 	}
-- 
2.32.0


^ permalink raw reply related	[flat|nested] 8+ messages in thread