On 11/14/2022 12:45, Eric Huang wrote:
It is to resolve a regression, which fails to allocate
VRAM due to no free memory in application, the reason
is we add check of vram_pin_size for memory limit, and
application is pinning the memory for Peerdirect, KFD
should not count it in memory limit. So removing
vram_pin_size will resolve it.
Any idea when the regression was introduced? Could you narrow it down
to a commit?
If so, it would be great to include a "Fixes" tag so that this could
also backport to relevant stable kernels that have the regression.
Signed-off-by: Eric Huang <jinhuieric.huang@xxxxxxx>
Reviewed-by: Felix Kuehling <Felix.Kuehling@xxxxxxx>
---
drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c | 4 +---
1 file changed, 1 insertion(+), 3 deletions(-)
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c
index db772942f7a6..fb1bb593312e 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c
@@ -172,9 +172,7 @@ int amdgpu_amdkfd_reserve_mem_limit(struct amdgpu_device *adev,
(kfd_mem_limit.ttm_mem_used + ttm_mem_needed >
kfd_mem_limit.max_ttm_mem_limit) ||
(adev && adev->kfd.vram_used + vram_needed >
- adev->gmc.real_vram_size -
- atomic64_read(&adev->vram_pin_size) -
- reserved_for_pt)) {
+ adev->gmc.real_vram_size - reserved_for_pt)) {
ret = -ENOMEM;
goto release;
}
