drm/amdgpu: move GTT allocation from gmc_sw_init to gmc_hw_init(V2)
Subsystems (2)
PATCH v2 1/1 drm/amdgpu: move GTT allocation from gmc_sw_init to gmc_hw_init(V2)
The below patch causes system hang for harvested ASICs. d015e9861e55 drm/amdgpu: improve debug VRAM access performance using sdma
The root cause is that GTT buffer should be allocated after GC SA harvest programming completed.
For harvested AISC, the GC SA harvest process(see utcl2_harvest) is programmed in gmc_v10_0_hw_init function. This is a hardware programming. Therefore should be located in hw init. Hence need to move GTT allocation from gmc_v*_0_sw_init to gmc_v*_0_hw_init.
V2: expand to all gmc_v*_0_hw_init functions.
--- drivers/gpu/drm/amd/amdgpu/amdgpu_gmc.c | 21 +++++++++++++++++++++ drivers/gpu/drm/amd/amdgpu/amdgpu_gmc.h | 3 +++ drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c | 8 -------- drivers/gpu/drm/amd/amdgpu/gmc_v10_0.c | 4 +++- drivers/gpu/drm/amd/amdgpu/gmc_v6_0.c | 13 +++++++++---- drivers/gpu/drm/amd/amdgpu/gmc_v7_0.c | 13 +++++++++---- drivers/gpu/drm/amd/amdgpu/gmc_v8_0.c | 13 +++++++++---- drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c | 13 +++++++++---- 8 files changed, 63 insertions(+), 25 deletions(-)
drivers/gpu/drm/amd/amdgpu/amdgpu_gmc.c +21 -0
@@ -876,3 +876,24 @@ int amdgpu_gmc_vram_checking(struct amdgpu_device *adev)return 0 ; } + +int amdgpu_gmc_allocate_sdma_access_gtt (struct amdgpu_device *adev) +{ +int ret; + + ret =amdgpu_bo_create_kernel (adev,PAGE_SIZE ,PAGE_SIZE , +AMDGPU_GEM_DOMAIN_GTT , + &adev->mman.sdma_access_bo,NULL , + &adev->mman.sdma_access_ptr); + +if (ret) +DRM_WARN ("Debug VRAM access will use slowpath MM access\n" ); + +return ret; +} + +void amdgpu_gmc_free_sdma_access_gtt (struct amdgpu_device *adev) +{ +amdgpu_bo_free_kernel (&adev->mman.sdma_access_bo,NULL , + &adev->mman.sdma_access_ptr); +}
drivers/gpu/drm/amd/amdgpu/amdgpu_gmc.h +3 -0
@@ -338,4 +338,7 @@ uint64_t amdgpu_gmc_vram_mc2pa(struct amdgpu_device *adev, uint64_t mc_addr);uint64_t amdgpu_gmc_vram_pa (struct amdgpu_device *adev,struct amdgpu_bo *bo);uint64_t amdgpu_gmc_vram_cpu_pa (struct amdgpu_device *adev,struct amdgpu_bo *bo);int amdgpu_gmc_vram_checking (struct amdgpu_device *adev); +int amdgpu_gmc_allocate_sdma_access_gtt (struct amdgpu_device *adev); +void amdgpu_gmc_free_sdma_access_gtt (struct amdgpu_device *adev); +#endif
drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c +0 -8
@@ -1823,12 +1823,6 @@ int amdgpu_ttm_init(struct amdgpu_device *adev)return r; } -if (amdgpu_bo_create_kernel (adev,PAGE_SIZE ,PAGE_SIZE , -AMDGPU_GEM_DOMAIN_GTT , - &adev->mman.sdma_access_bo,NULL , - &adev->mman.sdma_access_ptr)) -DRM_WARN ("Debug VRAM access will use slowpath MM access\n" ); -return 0 ; }
@@ -1850,8 +1844,6 @@ void amdgpu_ttm_fini(struct amdgpu_device *adev)if (adev->mman.stolen_reserved_size)amdgpu_bo_free_kernel (&adev->mman.stolen_reserved_memory,NULL ,NULL ); -amdgpu_bo_free_kernel (&adev->mman.sdma_access_bo,NULL , - &adev->mman.sdma_access_ptr);amdgpu_ttm_fw_reserve_vram_fini (adev);if (drm_dev_enter (adev_to_drm (adev), &idx)) {
drivers/gpu/drm/amd/amdgpu/gmc_v10_0.c +3 -1
@@ -1062,7 +1062,7 @@ static int gmc_v10_0_hw_init(void *handle)if (adev->umc.funcs && adev->umc.funcs->init_registers) adev->umc.funcs->init_registers (adev); -return 0 ; +return amdgpu_gmc_allocate_sdma_access_gtt (adev); }/**
@@ -1082,6 +1082,8 @@ static int gmc_v10_0_hw_fini(void *handle) {struct amdgpu_device *adev = (struct amdgpu_device *)handle; +amdgpu_gmc_free_sdma_access_gtt (adev); +gmc_v10_0_gart_disable (adev);if (amdgpu_sriov_vf (adev)) {
drivers/gpu/drm/amd/amdgpu/gmc_v6_0.c +9 -4
@@ -919,16 +919,21 @@ static int gmc_v6_0_hw_init(void *handle)if (r)return r; -if (amdgpu_emu_mode ==1 ) -return amdgpu_gmc_vram_checking (adev); -else -return r; +if (amdgpu_emu_mode ==1 ) { + r =amdgpu_gmc_vram_checking (adev); +if (r) +return r; + } + +return amdgpu_gmc_allocate_sdma_access_gtt (adev); }static int gmc_v6_0_hw_fini (void *handle) {struct amdgpu_device *adev = (struct amdgpu_device *)handle; +amdgpu_gmc_free_sdma_access_gtt (adev); +amdgpu_irq_put (adev, &adev->gmc.vm_fault,0 );gmc_v6_0_gart_disable (adev);
drivers/gpu/drm/amd/amdgpu/gmc_v7_0.c +9 -4
@@ -1107,16 +1107,21 @@ static int gmc_v7_0_hw_init(void *handle)if (r)return r; -if (amdgpu_emu_mode ==1 ) -return amdgpu_gmc_vram_checking (adev); -else -return r; +if (amdgpu_emu_mode ==1 ) { + r =amdgpu_gmc_vram_checking (adev); +if (r) +return r; + } + +return amdgpu_gmc_allocate_sdma_access_gtt (adev); }static int gmc_v7_0_hw_fini (void *handle) {struct amdgpu_device *adev = (struct amdgpu_device *)handle; +amdgpu_gmc_free_sdma_access_gtt (adev); +amdgpu_irq_put (adev, &adev->gmc.vm_fault,0 );gmc_v7_0_gart_disable (adev);
drivers/gpu/drm/amd/amdgpu/gmc_v8_0.c +9 -4
@@ -1238,16 +1238,21 @@ static int gmc_v8_0_hw_init(void *handle)if (r)return r; -if (amdgpu_emu_mode ==1 ) -return amdgpu_gmc_vram_checking (adev); -else -return r; +if (amdgpu_emu_mode ==1 ) { + r =amdgpu_gmc_vram_checking (adev); +if (r) +return r; + } + +return amdgpu_gmc_allocate_sdma_access_gtt (adev); }static int gmc_v8_0_hw_fini (void *handle) {struct amdgpu_device *adev = (struct amdgpu_device *)handle; +amdgpu_gmc_free_sdma_access_gtt (adev); +amdgpu_irq_put (adev, &adev->gmc.vm_fault,0 );gmc_v8_0_gart_disable (adev);
drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c +9 -4
@@ -1852,10 +1852,13 @@ static int gmc_v9_0_hw_init(void *handle)if (r)return r; -if (amdgpu_emu_mode ==1 ) -return amdgpu_gmc_vram_checking (adev); -else -return r; +if (amdgpu_emu_mode ==1 ) { + r =amdgpu_gmc_vram_checking (adev); +if (r) +return r; + } + +return amdgpu_gmc_allocate_sdma_access_gtt (adev); }/**
@@ -1875,6 +1878,8 @@ static int gmc_v9_0_hw_fini(void *handle) {struct amdgpu_device *adev = (struct amdgpu_device *)handle; +amdgpu_gmc_free_sdma_access_gtt (adev); +gmc_v9_0_gart_disable (adev);if (amdgpu_sriov_vf (adev)) {
Signature · 1 line
2.25.1
Am 27.01.22 um 08:52 schrieb Aaron Liu:
Quoted 5 lines
The below patch causes system hang for harvested ASICs.
d015e9861e55drm/amdgpu: improve debug VRAM access performance using sdmaThe root cause is that GTT buffer should be allocated after GC SA harvest programming completed.
That doesn't make much sense. We intentionally modified the code so that this is possible.
What is the rational that we need to program the HW before allocating system memory?
Regards,
Christian.
Quoted 230 lines
For harvested AISC, the GC SA harvest
process(see utcl2_harvest)is programmed ingmc_v10_0_hw_initfunction. This is a hardware programming. Therefore should be located in hw init. Hence need to move GTT allocation from gmc_v*_0_sw_init to gmc_v*_0_hw_init.V2: expand to all gmc_v*_0_hw_init functions.
Signed-off-by Aaron Liu <[email protected]>--- drivers/gpu/drm/amd/amdgpu/amdgpu_gmc.c | 21 +++++++++++++++++++++ drivers/gpu/drm/amd/amdgpu/amdgpu_gmc.h | 3 +++ drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c | 8 -------- drivers/gpu/drm/amd/amdgpu/gmc_v10_0.c | 4 +++- drivers/gpu/drm/amd/amdgpu/gmc_v6_0.c | 13 +++++++++---- drivers/gpu/drm/amd/amdgpu/gmc_v7_0.c | 13 +++++++++---- drivers/gpu/drm/amd/amdgpu/gmc_v8_0.c | 13 +++++++++---- drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c | 13 +++++++++---- 8 files changed, 63 insertions(+), 25 deletions(-)drivers/gpu/drm/amd/amdgpu/amdgpu_gmc.c +21
-0@@ -876,3 +876,24 @@ int amdgpu_gmc_vram_checking(struct amdgpu_device *adev)return 0 ; } + +int amdgpu_gmc_allocate_sdma_access_gtt (struct amdgpu_device *adev) +{ +int ret; + + ret =amdgpu_bo_create_kernel (adev,PAGE_SIZE ,PAGE_SIZE , +AMDGPU_GEM_DOMAIN_GTT , + &adev->mman.sdma_access_bo,NULL , + &adev->mman.sdma_access_ptr); + +if (ret) +DRM_WARN ("Debug VRAM access will use slowpath MM access\n" ); + +return ret; +} + +void amdgpu_gmc_free_sdma_access_gtt (struct amdgpu_device *adev) +{ +amdgpu_bo_free_kernel (&adev->mman.sdma_access_bo,NULL , + &adev->mman.sdma_access_ptr); +}drivers/gpu/drm/amd/amdgpu/amdgpu_gmc.h +3
-0@@ -338,4 +338,7 @@ uint64_t amdgpu_gmc_vram_mc2pa(struct amdgpu_device *adev, uint64_t mc_addr);uint64_t amdgpu_gmc_vram_pa (struct amdgpu_device *adev,struct amdgpu_bo *bo);uint64_t amdgpu_gmc_vram_cpu_pa (struct amdgpu_device *adev,struct amdgpu_bo *bo);int amdgpu_gmc_vram_checking (struct amdgpu_device *adev); +int amdgpu_gmc_allocate_sdma_access_gtt (struct amdgpu_device *adev); +void amdgpu_gmc_free_sdma_access_gtt (struct amdgpu_device *adev); +#endif drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c +0
-8@@ -1823,12 +1823,6 @@ int amdgpu_ttm_init(struct amdgpu_device *adev)return r; } -if (amdgpu_bo_create_kernel (adev,PAGE_SIZE ,PAGE_SIZE , -AMDGPU_GEM_DOMAIN_GTT , - &adev->mman.sdma_access_bo,NULL , - &adev->mman.sdma_access_ptr)) -DRM_WARN ("Debug VRAM access will use slowpath MM access\n" ); -return 0 ; }@@ -1850,8 +1844,6 @@ void amdgpu_ttm_fini(struct amdgpu_device *adev)if (adev->mman.stolen_reserved_size)amdgpu_bo_free_kernel (&adev->mman.stolen_reserved_memory,NULL ,NULL ); -amdgpu_bo_free_kernel (&adev->mman.sdma_access_bo,NULL , - &adev->mman.sdma_access_ptr);amdgpu_ttm_fw_reserve_vram_fini (adev);if (drm_dev_enter (adev_to_drm (adev), &idx)) {drivers/gpu/drm/amd/amdgpu/gmc_v10_0.c +3
-1@@ -1062,7 +1062,7 @@ static int gmc_v10_0_hw_init(void *handle)if (adev->umc.funcs && adev->umc.funcs->init_registers) adev->umc.funcs->init_registers (adev); -return 0 ; +return amdgpu_gmc_allocate_sdma_access_gtt (adev); }/** @@ -1082,6 +1082,8 @@ static int gmc_v10_0_hw_fini(void *handle) {struct amdgpu_device *adev = (struct amdgpu_device *)handle; +amdgpu_gmc_free_sdma_access_gtt (adev); +gmc_v10_0_gart_disable (adev);if (amdgpu_sriov_vf (adev)) {drivers/gpu/drm/amd/amdgpu/gmc_v6_0.c +9
-4@@ -919,16 +919,21 @@ static int gmc_v6_0_hw_init(void *handle)if (r)return r; -if (amdgpu_emu_mode ==1 ) -return amdgpu_gmc_vram_checking (adev); -else -return r; +if (amdgpu_emu_mode ==1 ) { + r =amdgpu_gmc_vram_checking (adev); +if (r) +return r; + } + +return amdgpu_gmc_allocate_sdma_access_gtt (adev); }static int gmc_v6_0_hw_fini (void *handle) {struct amdgpu_device *adev = (struct amdgpu_device *)handle; +amdgpu_gmc_free_sdma_access_gtt (adev); +amdgpu_irq_put (adev, &adev->gmc.vm_fault,0 );gmc_v6_0_gart_disable (adev);drivers/gpu/drm/amd/amdgpu/gmc_v7_0.c +9
-4@@ -1107,16 +1107,21 @@ static int gmc_v7_0_hw_init(void *handle)if (r)return r; -if (amdgpu_emu_mode ==1 ) -return amdgpu_gmc_vram_checking (adev); -else -return r; +if (amdgpu_emu_mode ==1 ) { + r =amdgpu_gmc_vram_checking (adev); +if (r) +return r; + } + +return amdgpu_gmc_allocate_sdma_access_gtt (adev); }static int gmc_v7_0_hw_fini (void *handle) {struct amdgpu_device *adev = (struct amdgpu_device *)handle; +amdgpu_gmc_free_sdma_access_gtt (adev); +amdgpu_irq_put (adev, &adev->gmc.vm_fault,0 );gmc_v7_0_gart_disable (adev);drivers/gpu/drm/amd/amdgpu/gmc_v8_0.c +9
-4@@ -1238,16 +1238,21 @@ static int gmc_v8_0_hw_init(void *handle)if (r)return r; -if (amdgpu_emu_mode ==1 ) -return amdgpu_gmc_vram_checking (adev); -else -return r; +if (amdgpu_emu_mode ==1 ) { + r =amdgpu_gmc_vram_checking (adev); +if (r) +return r; + } + +return amdgpu_gmc_allocate_sdma_access_gtt (adev); }static int gmc_v8_0_hw_fini (void *handle) {struct amdgpu_device *adev = (struct amdgpu_device *)handle; +amdgpu_gmc_free_sdma_access_gtt (adev); +amdgpu_irq_put (adev, &adev->gmc.vm_fault,0 );gmc_v8_0_gart_disable (adev);drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c +9
-4@@ -1852,10 +1852,13 @@ static int gmc_v9_0_hw_init(void *handle)if (r)return r; -if (amdgpu_emu_mode ==1 ) -return amdgpu_gmc_vram_checking (adev); -else -return r; +if (amdgpu_emu_mode ==1 ) { + r =amdgpu_gmc_vram_checking (adev); +if (r) +return r; + } + +return amdgpu_gmc_allocate_sdma_access_gtt (adev); }/** @@ -1875,6 +1878,8 @@ static int gmc_v9_0_hw_fini(void *handle) {struct amdgpu_device *adev = (struct amdgpu_device *)handle; +amdgpu_gmc_free_sdma_access_gtt (adev); +gmc_v9_0_gart_disable (adev);if (amdgpu_sriov_vf (adev)) {