diff options
| author | Alex Sierra <alex.sierra@amd.com> | 2022-05-17 17:43:06 -0500 |
|---|---|---|
| committer | Alex Deucher <alexander.deucher@amd.com> | 2022-07-28 16:05:16 -0400 |
| commit | f9af3c16bfe19d145cf0588afa06d7f1070cbe2d (patch) | |
| tree | 36c90cf5d364103744b1b23f6fb86af68b512860 /drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c | |
| parent | 06ac561fb0edf868f7b292fb4a3c8ffbbb1e14bb (diff) | |
| download | lwn-f9af3c16bfe19d145cf0588afa06d7f1070cbe2d.tar.gz lwn-f9af3c16bfe19d145cf0588afa06d7f1070cbe2d.zip | |
drm/amdkfd: track unified memory reservation with xnack off
[WHY]
Unified memory with xnack off should be tracked, as userptr mappings
and legacy allocations do. To avoid oversuscribe system memory when
xnack off.
[How]
Exposing functions reserve_mem_limit and unreserve_mem_limit to SVM
API and call them on every prange creation and free.
Signed-off-by: Alex Sierra <alex.sierra@amd.com>
Reviewed-by: Felix Kuehling <Felix.Kuehling@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Diffstat (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c')
| -rw-r--r-- | drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c | 23 |
1 files changed, 14 insertions, 9 deletions
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c index 08997092e7f7..1d4982471f8d 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c @@ -129,7 +129,7 @@ void amdgpu_amdkfd_reserve_system_mem(uint64_t size) * * Return: returns -ENOMEM in case of error, ZERO otherwise */ -static int amdgpu_amdkfd_reserve_mem_limit(struct amdgpu_device *adev, +int amdgpu_amdkfd_reserve_mem_limit(struct amdgpu_device *adev, uint64_t size, u32 alloc_flag) { uint64_t reserved_for_pt = @@ -169,7 +169,7 @@ static int amdgpu_amdkfd_reserve_mem_limit(struct amdgpu_device *adev, kfd_mem_limit.max_system_mem_limit && !no_system_mem_limit) || (kfd_mem_limit.ttm_mem_used + ttm_mem_needed > kfd_mem_limit.max_ttm_mem_limit) || - (adev->kfd.vram_used + vram_needed > + (adev && adev->kfd.vram_used + vram_needed > adev->gmc.real_vram_size - atomic64_read(&adev->vram_pin_size) - reserved_for_pt)) { @@ -180,7 +180,10 @@ static int amdgpu_amdkfd_reserve_mem_limit(struct amdgpu_device *adev, /* Update memory accounting by decreasing available system * memory, TTM memory and GPU memory as computed above */ - adev->kfd.vram_used += vram_needed; + WARN_ONCE(vram_needed && !adev, + "adev reference can't be null when vram is used"); + if (adev) + adev->kfd.vram_used += vram_needed; kfd_mem_limit.system_mem_used += system_mem_needed; kfd_mem_limit.ttm_mem_used += ttm_mem_needed; @@ -189,7 +192,7 @@ release: return ret; } -static void unreserve_mem_limit(struct amdgpu_device *adev, +void amdgpu_amdkfd_unreserve_mem_limit(struct amdgpu_device *adev, uint64_t size, u32 alloc_flag) { spin_lock(&kfd_mem_limit.mem_limit_lock); @@ -198,7 +201,10 @@ static void unreserve_mem_limit(struct amdgpu_device *adev, kfd_mem_limit.system_mem_used -= size; kfd_mem_limit.ttm_mem_used -= size; } else if (alloc_flag & KFD_IOC_ALLOC_MEM_FLAGS_VRAM) { - adev->kfd.vram_used -= ALIGN(size, VRAM_ALLOCATION_ALIGN); + 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); } else if (alloc_flag & KFD_IOC_ALLOC_MEM_FLAGS_USERPTR) { kfd_mem_limit.system_mem_used -= size; } else if (!(alloc_flag & @@ -207,8 +213,7 @@ static void unreserve_mem_limit(struct amdgpu_device *adev, pr_err("%s: Invalid BO type %#x\n", __func__, alloc_flag); goto release; } - - WARN_ONCE(adev->kfd.vram_used < 0, + WARN_ONCE(adev && adev->kfd.vram_used < 0, "KFD VRAM memory accounting unbalanced"); WARN_ONCE(kfd_mem_limit.ttm_mem_used < 0, "KFD TTM memory accounting unbalanced"); @@ -225,7 +230,7 @@ void amdgpu_amdkfd_release_notify(struct amdgpu_bo *bo) u32 alloc_flags = bo->kfd_bo->alloc_flags; u64 size = amdgpu_bo_size(bo); - unreserve_mem_limit(adev, size, alloc_flags); + amdgpu_amdkfd_unreserve_mem_limit(adev, size, alloc_flags); kfree(bo->kfd_bo); } @@ -1749,7 +1754,7 @@ err_node_allow: /* Don't unreserve system mem limit twice */ goto err_reserve_limit; err_bo_create: - unreserve_mem_limit(adev, size, flags); + amdgpu_amdkfd_unreserve_mem_limit(adev, size, flags); err_reserve_limit: mutex_destroy(&(*mem)->lock); if (gobj) |