On 2018-03-07 04:16 AM, Christian König wrote:
Am 06.03.2018 um 22:29 schrieb Felix Kuehling:
NAK.
For KFD we need the ability to create a BO from an SG list that doesn't come from another BO. We use this for mapping pages from the doorbell aperture into GPUVM for GPU self-dispatch.
You can still do this, see amdgpu_gem_prime_import_sg_table.
Just set the BO type to SG and then setting the SG table after creating it:
ret = amdgpu_bo_create(adev, attach->dmabuf->size, PAGE_SIZE, AMDGPU_GEM_DOMAIN_CPU, 0, ttm_bo_type_sg, resv, &bo); if (ret) goto error; bo->tbo.sg = sg; bo->tbo.ttm->sg = sg;
Then validate the result into the GTT domain to actually use the SG table.
OK, I'll try implementing that before we merge your change.
BTW: How do you create an SG table for the doorbell?
KFD manages the doorbell aperture and assigns one or two pages of doorbells to each process. KFD knows the bus address of the doorbells and passes it to amdgpu, which creates the SG like this:
static struct sg_table *create_doorbell_sg(uint64_t addr, uint32_t size) { struct sg_table *sg = kmalloc(sizeof(*sg), GFP_KERNEL);
if (!sg) return NULL; if (sg_alloc_table(sg, 1, GFP_KERNEL)) { kfree(sg); return NULL; } sg->sgl->dma_address = addr; sg->sgl->length = size; #ifdef CONFIG_NEED_SG_DMA_LENGTH sg->sgl->dma_length = size; #endif return sg; }
Regards, Felix
Regards, Christian.
If you remove this now, I'll need to add it back in some form in a month or two when I get to that part of upstreaming KFD.
There may be other ways to implement this. Currently we need to create a BO for anything we want to map into a GPUVM page table, because the amdgpu_vm code is based around BOs. If there was a way to map physical addresses into GPUVM without creating a buffer object, that would work too.
Regards, Felix
On 2018-03-06 09:43 AM, Christian König wrote:
Drop the "kernel" and sg parameter and give the BO type to create explicit to amdgpu_bo_create instead of figuring it out from the parameters.
Signed-off-by: Christian König christian.koenig@amd.com
drivers/gpu/drm/amd/amdgpu/amdgpu.h | 2 +- drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c | 5 +-- drivers/gpu/drm/amd/amdgpu/amdgpu_benchmark.c | 8 ++--- drivers/gpu/drm/amd/amdgpu/amdgpu_gart.c | 7 ++-- drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c | 6 ++-- drivers/gpu/drm/amd/amdgpu/amdgpu_object.c | 46 +++++++++++---------------- drivers/gpu/drm/amd/amdgpu/amdgpu_object.h | 11 +++---- drivers/gpu/drm/amd/amdgpu/amdgpu_prime.c | 7 ++-- drivers/gpu/drm/amd/amdgpu/amdgpu_test.c | 11 +++---- drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c | 11 ++++--- drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 8 ++--- 11 files changed, 58 insertions(+), 64 deletions(-)
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h index 292c7e72820c..b1116b773516 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h @@ -441,7 +441,7 @@ struct amdgpu_sa_bo { void amdgpu_gem_force_release(struct amdgpu_device *adev); int amdgpu_gem_object_create(struct amdgpu_device *adev, unsigned long size, int alignment, u32 initial_domain, - u64 flags, bool kernel, + u64 flags, enum ttm_bo_type type, struct reservation_object *resv, struct drm_gem_object **obj); diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c index 450426dbed92..7f096ed6e83d 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c @@ -215,8 +215,9 @@ int alloc_gtt_mem(struct kgd_dev *kgd, size_t size, if ((*mem) == NULL) return -ENOMEM; - r = amdgpu_bo_create(adev, size, PAGE_SIZE, true, AMDGPU_GEM_DOMAIN_GTT, - AMDGPU_GEM_CREATE_CPU_GTT_USWC, NULL, NULL, &(*mem)->bo); + r = amdgpu_bo_create(adev, size, PAGE_SIZE, AMDGPU_GEM_DOMAIN_GTT, + AMDGPU_GEM_CREATE_CPU_GTT_USWC, ttm_bo_type_kernel, + NULL, &(*mem)->bo); if (r) { dev_err(adev->dev, "failed to allocate BO for amdkfd (%d)\n", r); diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_benchmark.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_benchmark.c index 2fb299afc12b..02b849be083b 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_benchmark.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_benchmark.c @@ -80,8 +80,8 @@ static void amdgpu_benchmark_move(struct amdgpu_device *adev, unsigned size, int time; n = AMDGPU_BENCHMARK_ITERATIONS; - r = amdgpu_bo_create(adev, size, PAGE_SIZE, true, sdomain, 0, NULL, - NULL, &sobj); + r = amdgpu_bo_create(adev, size, PAGE_SIZE,sdomain, 0, + ttm_bo_type_kernel, NULL, &sobj); if (r) { goto out_cleanup; } @@ -93,8 +93,8 @@ static void amdgpu_benchmark_move(struct amdgpu_device *adev, unsigned size, if (r) { goto out_cleanup; } - r = amdgpu_bo_create(adev, size, PAGE_SIZE, true, ddomain, 0, NULL, - NULL, &dobj); + r = amdgpu_bo_create(adev, size, PAGE_SIZE, ddomain, 0, + ttm_bo_type_kernel, NULL, &dobj); if (r) { goto out_cleanup; } diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gart.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_gart.c index dc8d9f3216fa..cf0f186c6092 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gart.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gart.c @@ -113,11 +113,12 @@ int amdgpu_gart_table_vram_alloc(struct amdgpu_device *adev) int r; if (adev->gart.robj == NULL) { - r = amdgpu_bo_create(adev, adev->gart.table_size, - PAGE_SIZE, true, AMDGPU_GEM_DOMAIN_VRAM, + r = amdgpu_bo_create(adev, adev->gart.table_size, PAGE_SIZE, + AMDGPU_GEM_DOMAIN_VRAM, AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED | AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS, - NULL, NULL, &adev->gart.robj); + ttm_bo_type_kernel, NULL, + &adev->gart.robj); if (r) { return r; } diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c index 55a840ae6d68..bb9b21266b67 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c @@ -45,7 +45,7 @@ void amdgpu_gem_object_free(struct drm_gem_object *gobj) int amdgpu_gem_object_create(struct amdgpu_device *adev, unsigned long size, int alignment, u32 initial_domain, - u64 flags, bool kernel, + u64 flags, enum ttm_bo_type type, struct reservation_object *resv, struct drm_gem_object **obj) { @@ -59,8 +59,8 @@ int amdgpu_gem_object_create(struct amdgpu_device *adev, unsigned long size, } retry: - r = amdgpu_bo_create(adev, size, alignment, kernel, initial_domain, - flags, NULL, resv, &bo); + r = amdgpu_bo_create(adev, size, alignment, initial_domain, + flags, type, resv, &bo); if (r) { if (r != -ERESTARTSYS) { if (flags & AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED) { diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c index 216799ccb545..8f5027e8241a 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c @@ -187,10 +187,10 @@ int amdgpu_bo_create_reserved(struct amdgpu_device *adev, int r; if (!*bo_ptr) { - r = amdgpu_bo_create(adev, size, align, true, domain, + r = amdgpu_bo_create(adev, size, align, domain, AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED | AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS, - NULL, NULL, bo_ptr); + ttm_bo_type_kernel, NULL, bo_ptr); if (r) { dev_err(adev->dev, "(%d) failed to allocate kernel bo\n", r); @@ -331,21 +331,19 @@ static bool amdgpu_bo_validate_size(struct amdgpu_device *adev, return false; } -static int amdgpu_bo_do_create(struct amdgpu_device *adev, - unsigned long size, int byte_align, - bool kernel, u32 domain, u64 flags, - struct sg_table *sg, +static int amdgpu_bo_do_create(struct amdgpu_device *adev, unsigned long size, + int byte_align, u32 domain, + u64 flags, enum ttm_bo_type type, struct reservation_object *resv, struct amdgpu_bo **bo_ptr) { struct ttm_operation_ctx ctx = { - .interruptible = !kernel, + .interruptible = (type != ttm_bo_type_kernel), .no_wait_gpu = false, .resv = resv, .flags = TTM_OPT_FLAG_ALLOW_RES_EVICT }; struct amdgpu_bo *bo; - enum ttm_bo_type type; unsigned long page_align; size_t acc_size; int r; @@ -356,13 +354,6 @@ static int amdgpu_bo_do_create(struct amdgpu_device *adev, if (!amdgpu_bo_validate_size(adev, size, domain)) return -ENOMEM; - if (kernel) { - type = ttm_bo_type_kernel; - } else if (sg) { - type = ttm_bo_type_sg; - } else { - type = ttm_bo_type_device; - } *bo_ptr = NULL; acc_size = ttm_bo_dma_acc_size(&adev->mman.bdev, size, @@ -381,7 +372,8 @@ static int amdgpu_bo_do_create(struct amdgpu_device *adev, AMDGPU_GEM_DOMAIN_GWS | AMDGPU_GEM_DOMAIN_OA); bo->allowed_domains = bo->preferred_domains; - if (!kernel && bo->allowed_domains == AMDGPU_GEM_DOMAIN_VRAM) + if (type != ttm_bo_type_kernel && + bo->allowed_domains == AMDGPU_GEM_DOMAIN_VRAM) bo->allowed_domains |= AMDGPU_GEM_DOMAIN_GTT; bo->flags = flags; @@ -419,7 +411,7 @@ static int amdgpu_bo_do_create(struct amdgpu_device *adev, r = ttm_bo_init_reserved(&adev->mman.bdev, &bo->tbo, size, type, &bo->placement, page_align, &ctx, acc_size, - sg, resv, &amdgpu_ttm_bo_destroy); + NULL, resv, &amdgpu_ttm_bo_destroy); if (unlikely(r != 0)) return r; @@ -431,7 +423,7 @@ static int amdgpu_bo_do_create(struct amdgpu_device *adev, else amdgpu_cs_report_moved_bytes(adev, ctx.bytes_moved, 0); - if (kernel) + if (type == ttm_bo_type_kernel) bo->tbo.priority = 1; if (flags & AMDGPU_GEM_CREATE_VRAM_CLEARED && @@ -475,12 +467,11 @@ static int amdgpu_bo_create_shadow(struct amdgpu_device *adev, if (bo->shadow) return 0; - r = amdgpu_bo_do_create(adev, size, byte_align, true, - AMDGPU_GEM_DOMAIN_GTT, + r = amdgpu_bo_do_create(adev, size, byte_align, AMDGPU_GEM_DOMAIN_GTT, AMDGPU_GEM_CREATE_CPU_GTT_USWC | AMDGPU_GEM_CREATE_SHADOW, - NULL, bo->tbo.resv, - &bo->shadow); + ttm_bo_type_kernel, + bo->tbo.resv, &bo->shadow); if (!r) { bo->shadow->parent = amdgpu_bo_ref(bo); mutex_lock(&adev->shadow_list_lock); @@ -491,18 +482,17 @@ static int amdgpu_bo_create_shadow(struct amdgpu_device *adev, return r; } -int amdgpu_bo_create(struct amdgpu_device *adev, - unsigned long size, int byte_align, - bool kernel, u32 domain, u64 flags, - struct sg_table *sg, +int amdgpu_bo_create(struct amdgpu_device *adev, unsigned long size, + int byte_align, u32 domain, + u64 flags, enum ttm_bo_type type, struct reservation_object *resv, struct amdgpu_bo **bo_ptr) { uint64_t parent_flags = flags & ~AMDGPU_GEM_CREATE_SHADOW; int r; - r = amdgpu_bo_do_create(adev, size, byte_align, kernel, domain, - parent_flags, sg, resv, bo_ptr); + r = amdgpu_bo_do_create(adev, size, byte_align, domain, + parent_flags, type, resv, bo_ptr); if (r) return r; diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h index 1cef944ef98d..deb10ecd89a1 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h @@ -201,12 +201,11 @@ static inline bool amdgpu_bo_explicit_sync(struct amdgpu_bo *bo) return bo->flags & AMDGPU_GEM_CREATE_EXPLICIT_SYNC; } -int amdgpu_bo_create(struct amdgpu_device *adev, - unsigned long size, int byte_align, - bool kernel, u32 domain, u64 flags, - struct sg_table *sg, - struct reservation_object *resv, - struct amdgpu_bo **bo_ptr); +int amdgpu_bo_create(struct amdgpu_device *adev, unsigned long size, + int byte_align, u32 domain, + u64 flags, enum ttm_bo_type type, + struct reservation_object *resv, + struct amdgpu_bo **bo_ptr); int amdgpu_bo_create_reserved(struct amdgpu_device *adev, unsigned long size, int align, u32 domain, struct amdgpu_bo **bo_ptr, diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_prime.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_prime.c index 44230165e181..1c9991738477 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_prime.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_prime.c @@ -105,11 +105,14 @@ amdgpu_gem_prime_import_sg_table(struct drm_device *dev, int ret; ww_mutex_lock(&resv->lock, NULL); - ret = amdgpu_bo_create(adev, attach->dmabuf->size, PAGE_SIZE, false, - AMDGPU_GEM_DOMAIN_CPU, 0, sg, resv, &bo); + ret = amdgpu_bo_create(adev, attach->dmabuf->size, PAGE_SIZE, + AMDGPU_GEM_DOMAIN_CPU, 0, ttm_bo_type_sg, + resv, &bo); if (ret) goto error; + bo->tbo.sg = sg; + bo->tbo.ttm->sg = sg; bo->allowed_domains = AMDGPU_GEM_DOMAIN_GTT; bo->preferred_domains = AMDGPU_GEM_DOMAIN_GTT; if (attach->dmabuf->ops != &amdgpu_dmabuf_ops) diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_test.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_test.c index f3d81b6fb499..2dbe87591f81 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_test.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_test.c @@ -59,9 +59,8 @@ static void amdgpu_do_test_moves(struct amdgpu_device *adev) goto out_cleanup; } - r = amdgpu_bo_create(adev, size, PAGE_SIZE, true, - AMDGPU_GEM_DOMAIN_VRAM, 0, - NULL, NULL, &vram_obj); + r = amdgpu_bo_create(adev, size, PAGE_SIZE, AMDGPU_GEM_DOMAIN_VRAM, 0, + ttm_bo_type_kernel, NULL, &vram_obj); if (r) { DRM_ERROR("Failed to create VRAM object\n"); goto out_cleanup; @@ -80,9 +79,9 @@ static void amdgpu_do_test_moves(struct amdgpu_device *adev) void **vram_start, **vram_end; struct dma_fence *fence = NULL; - r = amdgpu_bo_create(adev, size, PAGE_SIZE, true, - AMDGPU_GEM_DOMAIN_GTT, 0, NULL, - NULL, gtt_obj + i); + r = amdgpu_bo_create(adev, size, PAGE_SIZE, + AMDGPU_GEM_DOMAIN_GTT, 0, + ttm_bo_type_kernel, NULL, gtt_obj + i); if (r) { DRM_ERROR("Failed to create GTT object %d\n", i); goto out_lclean; diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c index eebc9f77b611..1ec54629f277 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c @@ -1317,11 +1317,12 @@ static int amdgpu_ttm_fw_reserve_vram_init(struct amdgpu_device *adev) if (adev->fw_vram_usage.size > 0 && adev->fw_vram_usage.size <= vram_size) { - r = amdgpu_bo_create(adev, adev->fw_vram_usage.size, - PAGE_SIZE, true, AMDGPU_GEM_DOMAIN_VRAM, - AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED | - AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS, NULL, NULL, - &adev->fw_vram_usage.reserved_bo); + r = amdgpu_bo_create(adev, adev->fw_vram_usage.size, PAGE_SIZE, + AMDGPU_GEM_DOMAIN_VRAM, + AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED | + AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS, + ttm_bo_type_kernel, NULL, + &adev->fw_vram_usage.reserved_bo); if (r) goto error_create; diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c index 0b237e027cab..24474294c92a 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c @@ -413,9 +413,9 @@ static int amdgpu_vm_alloc_levels(struct amdgpu_device *adev, if (!entry->base.bo) { r = amdgpu_bo_create(adev, amdgpu_vm_bo_size(adev, level), - AMDGPU_GPU_PAGE_SIZE, true, + AMDGPU_GPU_PAGE_SIZE, AMDGPU_GEM_DOMAIN_VRAM, flags, - NULL, resv, &pt); + ttm_bo_type_kernel, resv, &pt); if (r) return r; @@ -2409,8 +2409,8 @@ int amdgpu_vm_init(struct amdgpu_device *adev, struct amdgpu_vm *vm, AMDGPU_GEM_CREATE_SHADOW); size = amdgpu_vm_bo_size(adev, adev->vm_manager.root_level); - r = amdgpu_bo_create(adev, size, align, true, AMDGPU_GEM_DOMAIN_VRAM, - flags, NULL, NULL, &vm->root.base.bo); + r = amdgpu_bo_create(adev, size, align, AMDGPU_GEM_DOMAIN_VRAM, flags, + ttm_bo_type_kernel, NULL, &vm->root.base.bo); if (r) goto error_free_sched_entity;