drm/amdgpu: Remove rounding from vram allocation path

Rounding up allocations in the allocation path caused test regressions,
so now just round in the availability path.

Signed-off-by: Daniel Phillips <daniel.phillips@amd.com>
Reviewed-by: Felix Kuehling <Felix.Kuehling@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c
index a699134a1e8cf5dd8c92fd0ddfe2f1a449eb3b7b..2d2ef633c5618362d5dc7df9c1b32fdee3514c7f 100644 (file)
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c
@@ -40,10 +40,10 @@
 #define AMDGPU_USERPTR_RESTORE_DELAY_MS 1
 
 /*
- * Align VRAM allocations to 2MB to avoid fragmentation caused by 4K allocations in the tail 2MB
+ * Align VRAM availability to 2MB to avoid fragmentation caused by 4K allocations in the tail 2MB
  * BO chunk
  */
-#define VRAM_ALLOCATION_ALIGN (1 << 21)
+#define VRAM_AVAILABLITY_ALIGN (1 << 21)
 
 /* Impose limit on how much memory KFD can use */
 static struct {
@@ -149,7 +149,7 @@ int amdgpu_amdkfd_reserve_mem_limit(struct amdgpu_device *adev,
                 * to avoid fragmentation caused by 4K allocations in the tail
                 * 2M BO chunk.
                 */
-               vram_needed = ALIGN(size, VRAM_ALLOCATION_ALIGN);
+               vram_needed = size;
        } else if (alloc_flag & KFD_IOC_ALLOC_MEM_FLAGS_USERPTR) {
                system_mem_needed = size;
        } else if (!(alloc_flag &
@@ -204,7 +204,7 @@ void amdgpu_amdkfd_unreserve_mem_limit(struct amdgpu_device *adev,
                WARN_ONCE(!adev,
                          "adev reference can't be null when alloc mem flags vram is set");
                if (adev)
-                       adev->kfd.vram_used -= ALIGN(size, VRAM_ALLOCATION_ALIGN);
+                       adev->kfd.vram_used -= size;
        } else if (alloc_flag & KFD_IOC_ALLOC_MEM_FLAGS_USERPTR) {
                kfd_mem_limit.system_mem_used -= size;
        } else if (!(alloc_flag &
@@ -1608,7 +1608,6 @@ size_t amdgpu_amdkfd_get_available_memory(struct amdgpu_device *adev)
        uint64_t reserved_for_pt =
                ESTIMATE_PT_SIZE(amdgpu_amdkfd_total_mem_size);
        size_t available;
-
        spin_lock(&kfd_mem_limit.mem_limit_lock);
        available = adev->gmc.real_vram_size
                - adev->kfd.vram_used
@@ -1616,7 +1615,7 @@ size_t amdgpu_amdkfd_get_available_memory(struct amdgpu_device *adev)
                - reserved_for_pt;
        spin_unlock(&kfd_mem_limit.mem_limit_lock);
 
-       return ALIGN_DOWN(available, VRAM_ALLOCATION_ALIGN);
+       return ALIGN_DOWN(available, VRAM_AVAILABLITY_ALIGN);
 }
 
 int amdgpu_amdkfd_gpuvm_alloc_memory_of_gpu(