git/linux-cryptodev-2.6
1
0
Fork 0
mirror of https://git.kernel.org/pub/scm/linux/kernel/git/herbert/cryptodev-2.6.git synced 2026-04-18 11:33:36 -04:00
Code Issues Packages Projects Releases Wiki Activity

drm/amdkfd: add a new flag to manage where VRAM allocations go

Browse Source 
On big and small APUs we send KFD VRAM allocations to GTT
since the carve out is either non-existent or relatively
small.  However, if someone sets the carve out size to be
relatively large, we may end up using GTT rather than VRAM.

No change of logic with this patch, but it allows the
driver to determine which logic to use based on the
carve out size in the future.

Reviewed-by: Mario Limonciello <mario.limonciello@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
This commit is contained in:
Alex Deucher
2025-01-30 15:12:58 -05:00
parent cc0e91a755
commit 8b0d068e7d
7 changed files with 22 additions and 14 deletions
Download Patch File Download Diff File Expand all files Collapse all files

16
drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c
View File

@@ -197,7 +197,7 @@ int amdgpu_amdkfd_reserve_mem_limit(struct amdgpu_device *adev,
return -EINVAL;
vram_size = KFD_XCP_MEMORY_SIZE(adev, xcp_id);
if (adev->flags & AMD_IS_APU) {
if (adev->apu_prefer_gtt) {
system_mem_needed = size;
ttm_mem_needed = size;
}
@@ -234,7 +234,7 @@ int amdgpu_amdkfd_reserve_mem_limit(struct amdgpu_device *adev,
if (adev && xcp_id >= 0) {
adev->kfd.vram_used[xcp_id] += vram_needed;
adev->kfd.vram_used_aligned[xcp_id] +=
(adev->flags & AMD_IS_APU) ?
adev->apu_prefer_gtt ?
vram_needed :
ALIGN(vram_needed, VRAM_AVAILABLITY_ALIGN);
}
@@ -262,7 +262,7 @@ void amdgpu_amdkfd_unreserve_mem_limit(struct amdgpu_device *adev,
if (adev) {
adev->kfd.vram_used[xcp_id] -= size;
if (adev->flags & AMD_IS_APU) {
if (adev->apu_prefer_gtt) {
adev->kfd.vram_used_aligned[xcp_id] -= size;
kfd_mem_limit.system_mem_used -= size;
kfd_mem_limit.ttm_mem_used -= size;
@@ -890,7 +890,7 @@ static int kfd_mem_attach(struct amdgpu_device *adev, struct kgd_mem *mem,
* if peer device has large BAR. In contrast, access over xGMI is
* allowed for both small and large BAR configurations of peer device
*/
if ((adev != bo_adev && !(adev->flags & AMD_IS_APU)) &&
if ((adev != bo_adev && !adev->apu_prefer_gtt) &&
((mem->domain == AMDGPU_GEM_DOMAIN_VRAM) ||
(mem->alloc_flags & KFD_IOC_ALLOC_MEM_FLAGS_DOORBELL) ||
(mem->alloc_flags & KFD_IOC_ALLOC_MEM_FLAGS_MMIO_REMAP))) {
@@ -1667,7 +1667,7 @@ size_t amdgpu_amdkfd_get_available_memory(struct amdgpu_device *adev,
- reserved_for_pt
- reserved_for_ras;
if (adev->flags & AMD_IS_APU) {
if (adev->apu_prefer_gtt) {
system_mem_available = no_system_mem_limit ?
kfd_mem_limit.max_system_mem_limit :
kfd_mem_limit.max_system_mem_limit -
@@ -1715,7 +1715,7 @@ int amdgpu_amdkfd_gpuvm_alloc_memory_of_gpu(
if (flags & KFD_IOC_ALLOC_MEM_FLAGS_VRAM) {
domain = alloc_domain = AMDGPU_GEM_DOMAIN_VRAM;
if (adev->flags & AMD_IS_APU) {
if (adev->apu_prefer_gtt) {
domain = AMDGPU_GEM_DOMAIN_GTT;
alloc_domain = AMDGPU_GEM_DOMAIN_GTT;
alloc_flags = 0;
@@ -1966,7 +1966,7 @@ int amdgpu_amdkfd_gpuvm_free_memory_of_gpu(
if (size) {
if (!is_imported &&
(mem->bo->preferred_domains == AMDGPU_GEM_DOMAIN_VRAM ||
((adev->flags & AMD_IS_APU) &&
(adev->apu_prefer_gtt &&
mem->bo->preferred_domains == AMDGPU_GEM_DOMAIN_GTT)))
*size = bo_size;
else
@@ -2393,7 +2393,7 @@ static int import_obj_create(struct amdgpu_device *adev,
(*mem)->bo = bo;
(*mem)->va = va;
(*mem)->domain = (bo->preferred_domains & AMDGPU_GEM_DOMAIN_VRAM) &&
!(adev->flags & AMD_IS_APU) ?
!adev->apu_prefer_gtt ?
AMDGPU_GEM_DOMAIN_VRAM : AMDGPU_GEM_DOMAIN_GTT;
(*mem)->mapped_to_gpu_memory = 0;