diff options
| author |  Xiaogang Chen <xiaogang.chen@amd.com>
| 2023-10-04 16:20:58 -0500 |
|---|---|---|
| committer |  Alex Deucher <alexander.deucher@amd.com>
| 2023-10-05 17:58:59 -0400 |
| commit | dc427a473e5d119232ddb27530920d9796cdea70 (patch) | |
| tree | 9c3b35c793cc047ee2c778498494d308d41ea54a /drivers/gpu/drm/amd/amdkfd/kfd_svm.h | |
| parent | drm/amd/display: Simplify the per-CPU usage. (diff) | |
| download | linux-dc427a473e5d119232ddb27530920d9796cdea70.tar.gz linux-dc427a473e5d119232ddb27530920d9796cdea70.tar.bz2 linux-dc427a473e5d119232ddb27530920d9796cdea70.zip | |
drm/amdkfd: Use partial migrations in GPU page faults
This patch implements partial migration in gpu page fault according to migration
granularity(default 2MB) and not split svm range in cpu page fault handling.
A svm range may include pages from both system ram and vram of one gpu now.
These chagnes are expected to improve migration performance and reduce mmu
callback and TLB flush workloads.
Signed-off-by: Xiaogang Chen <xiaogang.chen@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/amdkfd/kfd_svm.h')
| -rw-r--r-- | drivers/gpu/drm/amd/amdkfd/kfd_svm.h | 6 |
1 files changed, 5 insertions, 1 deletions
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_svm.h b/drivers/gpu/drm/amd/amdkfd/kfd_svm.h index c528df1d0ba2..be11ba0c4289 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_svm.h +++ b/drivers/gpu/drm/amd/amdkfd/kfd_svm.h @@ -78,6 +78,7 @@ struct svm_work_list_item { * @update_list:link list node used to add to update_list * @mapping: bo_va mapping structure to create and update GPU page table * @npages: number of pages + * @vram_pages: vram pages number in this svm_range * @dma_addr: dma mapping address on each GPU for system memory physical page * @ttm_res: vram ttm resource map * @offset: range start offset within mm_nodes @@ -88,7 +89,9 @@ struct svm_work_list_item { * @flags: flags defined as KFD_IOCTL_SVM_FLAG_* * @perferred_loc: perferred location, 0 for CPU, or GPU id * @perfetch_loc: last prefetch location, 0 for CPU, or GPU id - * @actual_loc: the actual location, 0 for CPU, or GPU id + * @actual_loc: this svm_range location. 0: all pages are from sys ram; + * GPU id: this svm_range may include vram pages from GPU with + * id actual_loc. * @granularity:migration granularity, log2 num pages * @invalid: not 0 means cpu page table is invalidated * @validate_timestamp: system timestamp when range is validated @@ -112,6 +115,7 @@ struct svm_range { struct list_head list; struct list_head update_list; uint64_t npages; + uint64_t vram_pages; dma_addr_t *dma_addr[MAX_GPU_INSTANCE]; struct ttm_resource *ttm_res; uint64_t offset; |