Linaro-mm-sig May 2025

linaro-mm-sig@lists.linaro.org

27 participants
102 discussions

[PATCH AUTOSEL 6.14 628/642] drm/gem: Internally test import_attach for imported objects

by Sasha Levin

From: Thomas Zimmermann <tzimmermann(a)suse.de> [ Upstream commit 8260731ccad0451207b45844bb66eb161a209218 ] Test struct drm_gem_object.import_attach to detect imported objects. During object clenanup, the dma_buf field might be NULL. Testing it in an object's free callback then incorrectly does a cleanup as for native objects. Happens for calls to drm_mode_destroy_dumb_ioctl() that clears the dma_buf field in drm_gem_object_exported_dma_buf_free(). v3: - only test for import_attach (Boris) v2: - use import_attach.dmabuf instead of dma_buf (Christian) Signed-off-by: Thomas Zimmermann <tzimmermann(a)suse.de> Fixes: b57aa47d39e9 ("drm/gem: Test for imported GEM buffers with helper") Reported-by: Andy Yan <andyshrk(a)163.com> Closes: https://lore.kernel.org/dri-devel/38d09d34.4354.196379aa560.Coremail.andysh… Tested-by: Andy Yan <andyshrk(a)163.com> Cc: Thomas Zimmermann <tzimmermann(a)suse.de> Cc: Anusha Srivatsa <asrivats(a)redhat.com> Cc: Christian König <christian.koenig(a)amd.com> Cc: Maarten Lankhorst <maarten.lankhorst(a)linux.intel.com> Cc: Maxime Ripard <mripard(a)kernel.org> Cc: David Airlie <airlied(a)gmail.com> Cc: Simona Vetter <simona(a)ffwll.ch> Cc: Sumit Semwal <sumit.semwal(a)linaro.org> Cc: "Christian König" <christian.koenig(a)amd.com> Cc: dri-devel(a)lists.freedesktop.org Cc: linux-media(a)vger.kernel.org Cc: linaro-mm-sig(a)lists.linaro.org Reviewed-by: Boris Brezillon <boris.brezillon(a)collabora.com> Reviewed-by: Simona Vetter <simona.vetter(a)ffwll.ch> Link: https://lore.kernel.org/r/20250416065820.26076-1-tzimmermann@suse.de Signed-off-by: Sasha Levin <sashal(a)kernel.org> --- include/drm/drm_gem.h | 3 +-- 1 file changed, 1 insertion(+), 2 deletions(-) diff --git a/include/drm/drm_gem.h b/include/drm/drm_gem.h index 2bf893eabb4b2..bcd54020d6ba5 100644 --- a/include/drm/drm_gem.h +++ b/include/drm/drm_gem.h @@ -585,8 +585,7 @@ static inline bool drm_gem_object_is_shared_for_memory_stats(struct drm_gem_obje */ static inline bool drm_gem_is_imported(const struct drm_gem_object *obj) { - /* The dma-buf's priv field points to the original GEM object. */ - return obj->dma_buf && (obj->dma_buf->priv != obj); + return !!obj->import_attach; } #ifdef CONFIG_LOCKDEP -- 2.39.5

2 weeks, 4 days

[PATCH AUTOSEL 6.14 355/642] drm/amdgpu: remove all KFD fences from the BO on release

by Sasha Levin

From: Christian König <christian.koenig(a)amd.com> [ Upstream commit cb0de06d1b0afb2d0c600ad748069f5ce27730ec ] Remove all KFD BOs from the private dma_resv object. This prevents the KFD from being evict unecessarily when an exported BO is released. Signed-off-by: Christian König <christian.koenig(a)amd.com> Signed-off-by: James Zhu <James.Zhu(a)amd.com> Reviewed-by: Felix Kuehling <felix.kuehling(a)amd.com> Reviewed-and-tested-by: James Zhu <James.Zhu(a)amd.com> Signed-off-by: Alex Deucher <alexander.deucher(a)amd.com> Signed-off-by: Sasha Levin <sashal(a)kernel.org> --- drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h | 5 +- .../gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c | 52 ++++++++----------- drivers/gpu/drm/amd/amdgpu/amdgpu_object.c | 38 ++++++++------ 3 files changed, 47 insertions(+), 48 deletions(-) diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h index 8af67f18500a7..2f48dc5747aa2 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h @@ -192,7 +192,7 @@ int kfd_debugfs_kfd_mem_limits(struct seq_file *m, void *data); #if IS_ENABLED(CONFIG_HSA_AMD) bool amdkfd_fence_check_mm(struct dma_fence *f, struct mm_struct *mm); struct amdgpu_amdkfd_fence *to_amdgpu_amdkfd_fence(struct dma_fence *f); -int amdgpu_amdkfd_remove_fence_on_pt_pd_bos(struct amdgpu_bo *bo); +void amdgpu_amdkfd_remove_all_eviction_fences(struct amdgpu_bo *bo); int amdgpu_amdkfd_evict_userptr(struct mmu_interval_notifier *mni, unsigned long cur_seq, struct kgd_mem *mem); int amdgpu_amdkfd_bo_validate_and_fence(struct amdgpu_bo *bo, @@ -212,9 +212,8 @@ struct amdgpu_amdkfd_fence *to_amdgpu_amdkfd_fence(struct dma_fence *f) } static inline -int amdgpu_amdkfd_remove_fence_on_pt_pd_bos(struct amdgpu_bo *bo) +void amdgpu_amdkfd_remove_all_eviction_fences(struct amdgpu_bo *bo) { - return 0; } static inline diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c index 70224b9f54f2f..c0aaa72b6c210 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c @@ -370,40 +370,32 @@ static int amdgpu_amdkfd_remove_eviction_fence(struct amdgpu_bo *bo, return 0; } -int amdgpu_amdkfd_remove_fence_on_pt_pd_bos(struct amdgpu_bo *bo) +/** + * amdgpu_amdkfd_remove_all_eviction_fences - Remove all eviction fences + * @bo: the BO where to remove the evictions fences from. + * + * This functions should only be used on release when all references to the BO + * are already dropped. We remove the eviction fence from the private copy of + * the dma_resv object here since that is what is used during release to + * determine of the BO is idle or not. + */ +void amdgpu_amdkfd_remove_all_eviction_fences(struct amdgpu_bo *bo) { - struct amdgpu_bo *root = bo; - struct amdgpu_vm_bo_base *vm_bo; - struct amdgpu_vm *vm; - struct amdkfd_process_info *info; - struct amdgpu_amdkfd_fence *ef; - int ret; - - /* we can always get vm_bo from root PD bo.*/ - while (root->parent) - root = root->parent; + struct dma_resv *resv = &bo->tbo.base._resv; + struct dma_fence *fence, *stub; + struct dma_resv_iter cursor; - vm_bo = root->vm_bo; - if (!vm_bo) - return 0; + dma_resv_assert_held(resv); - vm = vm_bo->vm; - if (!vm) - return 0; - - info = vm->process_info; - if (!info || !info->eviction_fence) - return 0; - - ef = container_of(dma_fence_get(&info->eviction_fence->base), - struct amdgpu_amdkfd_fence, base); - - BUG_ON(!dma_resv_trylock(bo->tbo.base.resv)); - ret = amdgpu_amdkfd_remove_eviction_fence(bo, ef); - dma_resv_unlock(bo->tbo.base.resv); + stub = dma_fence_get_stub(); + dma_resv_for_each_fence(&cursor, resv, DMA_RESV_USAGE_BOOKKEEP, fence) { + if (!to_amdgpu_amdkfd_fence(fence)) + continue; - dma_fence_put(&ef->base); - return ret; + dma_resv_replace_fences(resv, fence->context, stub, + DMA_RESV_USAGE_BOOKKEEP); + } + dma_fence_put(stub); } static int amdgpu_amdkfd_bo_validate(struct amdgpu_bo *bo, uint32_t domain, diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c index 00752e3f9d8ab..0b9987781f762 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c @@ -1295,28 +1295,36 @@ void amdgpu_bo_release_notify(struct ttm_buffer_object *bo) if (abo->kfd_bo) amdgpu_amdkfd_release_notify(abo); - /* We only remove the fence if the resv has individualized. */ - WARN_ON_ONCE(bo->type == ttm_bo_type_kernel - && bo->base.resv != &bo->base._resv); - if (bo->base.resv == &bo->base._resv) - amdgpu_amdkfd_remove_fence_on_pt_pd_bos(abo); + /* + * We lock the private dma_resv object here and since the BO is about to + * be released nobody else should have a pointer to it. + * So when this locking here fails something is wrong with the reference + * counting. + */ + if (WARN_ON_ONCE(!dma_resv_trylock(&bo->base._resv))) + return; + + amdgpu_amdkfd_remove_all_eviction_fences(abo); if (!bo->resource || bo->resource->mem_type != TTM_PL_VRAM || !(abo->flags & AMDGPU_GEM_CREATE_VRAM_WIPE_ON_RELEASE) || adev->in_suspend || drm_dev_is_unplugged(adev_to_drm(adev))) - return; + goto out; - if (WARN_ON_ONCE(!dma_resv_trylock(bo->base.resv))) - return; + r = dma_resv_reserve_fences(&bo->base._resv, 1); + if (r) + goto out; - r = amdgpu_fill_buffer(abo, 0, bo->base.resv, &fence, true); - if (!WARN_ON(r)) { - amdgpu_vram_mgr_set_cleared(bo->resource); - amdgpu_bo_fence(abo, fence, false); - dma_fence_put(fence); - } + r = amdgpu_fill_buffer(abo, 0, &bo->base._resv, &fence, true); + if (WARN_ON(r)) + goto out; + + amdgpu_vram_mgr_set_cleared(bo->resource); + dma_resv_add_fence(&bo->base._resv, fence, DMA_RESV_USAGE_KERNEL); + dma_fence_put(fence); - dma_resv_unlock(bo->base.resv); +out: + dma_resv_unlock(&bo->base._resv); } /** -- 2.39.5

2 weeks, 4 days

[PATCH AUTOSEL 6.14 241/642] drm/gem: Test for imported GEM buffers with helper

by Sasha Levin

From: Thomas Zimmermann <tzimmermann(a)suse.de> [ Upstream commit b57aa47d39e94dc47403a745e2024664e544078c ] Add drm_gem_is_imported() that tests if a GEM object's buffer has been imported. Update the GEM code accordingly. GEM code usually tests for imports if import_attach has been set in struct drm_gem_object. But attaching a dma-buf on import requires a DMA-capable importer device, which is not the case for many serial busses like USB or I2C. The new helper tests if a GEM object's dma-buf has been created from the GEM object. Signed-off-by: Thomas Zimmermann <tzimmermann(a)suse.de> Reviewed-by: Anusha Srivatsa <asrivats(a)redhat.com> Reviewed-by: Christian König <christian.koenig(a)amd.com> Link: https://patchwork.freedesktop.org/patch/msgid/20250226172457.217725-2-tzimm… Signed-off-by: Sasha Levin <sashal(a)kernel.org> --- drivers/gpu/drm/drm_gem.c | 4 ++-- include/drm/drm_gem.h | 14 ++++++++++++++ 2 files changed, 16 insertions(+), 2 deletions(-) diff --git a/drivers/gpu/drm/drm_gem.c b/drivers/gpu/drm/drm_gem.c index ee811764c3df4..c6240bab3fa55 100644 --- a/drivers/gpu/drm/drm_gem.c +++ b/drivers/gpu/drm/drm_gem.c @@ -348,7 +348,7 @@ int drm_gem_dumb_map_offset(struct drm_file *file, struct drm_device *dev, return -ENOENT; /* Don't allow imported objects to be mapped */ - if (obj->import_attach) { + if (drm_gem_is_imported(obj)) { ret = -EINVAL; goto out; } @@ -1178,7 +1178,7 @@ void drm_gem_print_info(struct drm_printer *p, unsigned int indent, drm_vma_node_start(&obj->vma_node)); drm_printf_indent(p, indent, "size=%zu\n", obj->size); drm_printf_indent(p, indent, "imported=%s\n", - str_yes_no(obj->import_attach)); + str_yes_no(drm_gem_is_imported(obj))); if (obj->funcs->print_info) obj->funcs->print_info(p, indent, obj); diff --git a/include/drm/drm_gem.h b/include/drm/drm_gem.h index fdae947682cd0..2bf893eabb4b2 100644 --- a/include/drm/drm_gem.h +++ b/include/drm/drm_gem.h @@ -35,6 +35,7 @@ */ #include <linux/kref.h> +#include <linux/dma-buf.h> #include <linux/dma-resv.h> #include <linux/list.h> #include <linux/mutex.h> @@ -575,6 +576,19 @@ static inline bool drm_gem_object_is_shared_for_memory_stats(struct drm_gem_obje return (obj->handle_count > 1) || obj->dma_buf; } +/** + * drm_gem_is_imported() - Tests if GEM object's buffer has been imported + * @obj: the GEM object + * + * Returns: + * True if the GEM object's buffer has been imported, false otherwise + */ +static inline bool drm_gem_is_imported(const struct drm_gem_object *obj) +{ + /* The dma-buf's priv field points to the original GEM object. */ + return obj->dma_buf && (obj->dma_buf->priv != obj); +} + #ifdef CONFIG_LOCKDEP /** * drm_gem_gpuva_set_lock() - Set the lock protecting accesses to the gpuva list. -- 2.39.5

2 weeks, 4 days

[PATCH AUTOSEL 6.14 061/642] drm/amdgpu: use GFP_NOWAIT for memory allocations

by Sasha Levin

From: Christian König <christian.koenig(a)amd.com> [ Upstream commit 16590745b571c07869ef8958e0bbe44ab6f08d1f ] In the critical submission path memory allocations can't wait for reclaim since that can potentially wait for submissions to finish. Finally clean that up and mark most memory allocations in the critical path with GFP_NOWAIT. The only exception left is the dma_fence_array() used when no VMID is available, but that will be cleaned up later on. Signed-off-by: Christian König <christian.koenig(a)amd.com> Acked-by: Srinivasan Shanmugam <srinivasan.shanmugam(a)amd.com> Signed-off-by: Alex Deucher <alexander.deucher(a)amd.com> Signed-off-by: Sasha Levin <sashal(a)kernel.org> --- .../gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c | 8 ++++---- drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c | 18 +++++++++++------- drivers/gpu/drm/amd/amdgpu/amdgpu_ids.c | 11 +++++++---- drivers/gpu/drm/amd/amdgpu/amdgpu_mes.c | 4 ++-- drivers/gpu/drm/amd/amdgpu/amdgpu_sync.c | 11 ++++++----- drivers/gpu/drm/amd/amdgpu/amdgpu_sync.h | 3 ++- 6 files changed, 32 insertions(+), 23 deletions(-) diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c index 1e998f972c308..70224b9f54f2f 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c @@ -499,7 +499,7 @@ static int vm_update_pds(struct amdgpu_vm *vm, struct amdgpu_sync *sync) if (ret) return ret; - return amdgpu_sync_fence(sync, vm->last_update); + return amdgpu_sync_fence(sync, vm->last_update, GFP_KERNEL); } static uint64_t get_pte_flags(struct amdgpu_device *adev, struct kgd_mem *mem) @@ -1263,7 +1263,7 @@ static int unmap_bo_from_gpuvm(struct kgd_mem *mem, (void)amdgpu_vm_clear_freed(adev, vm, &bo_va->last_pt_update); - (void)amdgpu_sync_fence(sync, bo_va->last_pt_update); + (void)amdgpu_sync_fence(sync, bo_va->last_pt_update, GFP_KERNEL); return 0; } @@ -1287,7 +1287,7 @@ static int update_gpuvm_pte(struct kgd_mem *mem, return ret; } - return amdgpu_sync_fence(sync, bo_va->last_pt_update); + return amdgpu_sync_fence(sync, bo_va->last_pt_update, GFP_KERNEL); } static int map_bo_to_gpuvm(struct kgd_mem *mem, @@ -2969,7 +2969,7 @@ int amdgpu_amdkfd_gpuvm_restore_process_bos(void *info, struct dma_fence __rcu * } dma_resv_for_each_fence(&cursor, bo->tbo.base.resv, DMA_RESV_USAGE_KERNEL, fence) { - ret = amdgpu_sync_fence(&sync_obj, fence); + ret = amdgpu_sync_fence(&sync_obj, fence, GFP_KERNEL); if (ret) { pr_debug("Memory eviction: Sync BO fence failed. Try again\n"); goto validate_map_fail; diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c index 5cc5f59e30184..4a5b406601fa2 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c @@ -428,7 +428,7 @@ static int amdgpu_cs_p2_dependencies(struct amdgpu_cs_parser *p, dma_fence_put(old); } - r = amdgpu_sync_fence(&p->sync, fence); + r = amdgpu_sync_fence(&p->sync, fence, GFP_KERNEL); dma_fence_put(fence); if (r) return r; @@ -450,7 +450,7 @@ static int amdgpu_syncobj_lookup_and_add(struct amdgpu_cs_parser *p, return r; } - r = amdgpu_sync_fence(&p->sync, fence); + r = amdgpu_sync_fence(&p->sync, fence, GFP_KERNEL); dma_fence_put(fence); return r; } @@ -1124,7 +1124,8 @@ static int amdgpu_cs_vm_handling(struct amdgpu_cs_parser *p) if (r) return r; - r = amdgpu_sync_fence(&p->sync, fpriv->prt_va->last_pt_update); + r = amdgpu_sync_fence(&p->sync, fpriv->prt_va->last_pt_update, + GFP_KERNEL); if (r) return r; @@ -1135,7 +1136,8 @@ static int amdgpu_cs_vm_handling(struct amdgpu_cs_parser *p) if (r) return r; - r = amdgpu_sync_fence(&p->sync, bo_va->last_pt_update); + r = amdgpu_sync_fence(&p->sync, bo_va->last_pt_update, + GFP_KERNEL); if (r) return r; } @@ -1154,7 +1156,8 @@ static int amdgpu_cs_vm_handling(struct amdgpu_cs_parser *p) if (r) return r; - r = amdgpu_sync_fence(&p->sync, bo_va->last_pt_update); + r = amdgpu_sync_fence(&p->sync, bo_va->last_pt_update, + GFP_KERNEL); if (r) return r; } @@ -1167,7 +1170,7 @@ static int amdgpu_cs_vm_handling(struct amdgpu_cs_parser *p) if (r) return r; - r = amdgpu_sync_fence(&p->sync, vm->last_update); + r = amdgpu_sync_fence(&p->sync, vm->last_update, GFP_KERNEL); if (r) return r; @@ -1248,7 +1251,8 @@ static int amdgpu_cs_sync_rings(struct amdgpu_cs_parser *p) continue; } - r = amdgpu_sync_fence(&p->gang_leader->explicit_sync, fence); + r = amdgpu_sync_fence(&p->gang_leader->explicit_sync, fence, + GFP_KERNEL); dma_fence_put(fence); if (r) return r; diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ids.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ids.c index 9008b7388e897..92ab821afc06a 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ids.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ids.c @@ -209,7 +209,7 @@ static int amdgpu_vmid_grab_idle(struct amdgpu_ring *ring, return 0; } - fences = kmalloc_array(id_mgr->num_ids, sizeof(void *), GFP_KERNEL); + fences = kmalloc_array(id_mgr->num_ids, sizeof(void *), GFP_NOWAIT); if (!fences) return -ENOMEM; @@ -313,7 +313,8 @@ static int amdgpu_vmid_grab_reserved(struct amdgpu_vm *vm, /* Good we can use this VMID. Remember this submission as * user of the VMID. */ - r = amdgpu_sync_fence(&(*id)->active, &job->base.s_fence->finished); + r = amdgpu_sync_fence(&(*id)->active, &job->base.s_fence->finished, + GFP_NOWAIT); if (r) return r; @@ -372,7 +373,8 @@ static int amdgpu_vmid_grab_used(struct amdgpu_vm *vm, * user of the VMID. */ r = amdgpu_sync_fence(&(*id)->active, - &job->base.s_fence->finished); + &job->base.s_fence->finished, + GFP_NOWAIT); if (r) return r; @@ -424,7 +426,8 @@ int amdgpu_vmid_grab(struct amdgpu_vm *vm, struct amdgpu_ring *ring, /* Remember this submission as user of the VMID */ r = amdgpu_sync_fence(&id->active, - &job->base.s_fence->finished); + &job->base.s_fence->finished, + GFP_NOWAIT); if (r) goto error; diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_mes.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_mes.c index 6fa20980a0b15..e4251d0691c9c 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_mes.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_mes.c @@ -1335,14 +1335,14 @@ int amdgpu_mes_ctx_map_meta_data(struct amdgpu_device *adev, DRM_ERROR("failed to do vm_bo_update on meta data\n"); goto error_del_bo_va; } - amdgpu_sync_fence(&sync, bo_va->last_pt_update); + amdgpu_sync_fence(&sync, bo_va->last_pt_update, GFP_KERNEL); r = amdgpu_vm_update_pdes(adev, vm, false); if (r) { DRM_ERROR("failed to update pdes on meta data\n"); goto error_del_bo_va; } - amdgpu_sync_fence(&sync, vm->last_update); + amdgpu_sync_fence(&sync, vm->last_update, GFP_KERNEL); amdgpu_sync_wait(&sync, false); drm_exec_fini(&exec); diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_sync.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_sync.c index d75715b3f1870..34fc742fda91d 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_sync.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_sync.c @@ -152,7 +152,8 @@ static bool amdgpu_sync_add_later(struct amdgpu_sync *sync, struct dma_fence *f) * * Add the fence to the sync object. */ -int amdgpu_sync_fence(struct amdgpu_sync *sync, struct dma_fence *f) +int amdgpu_sync_fence(struct amdgpu_sync *sync, struct dma_fence *f, + gfp_t flags) { struct amdgpu_sync_entry *e; @@ -162,7 +163,7 @@ int amdgpu_sync_fence(struct amdgpu_sync *sync, struct dma_fence *f) if (amdgpu_sync_add_later(sync, f)) return 0; - e = kmem_cache_alloc(amdgpu_sync_slab, GFP_KERNEL); + e = kmem_cache_alloc(amdgpu_sync_slab, flags); if (!e) return -ENOMEM; @@ -249,7 +250,7 @@ int amdgpu_sync_resv(struct amdgpu_device *adev, struct amdgpu_sync *sync, struct dma_fence *tmp = dma_fence_chain_contained(f); if (amdgpu_sync_test_fence(adev, mode, owner, tmp)) { - r = amdgpu_sync_fence(sync, f); + r = amdgpu_sync_fence(sync, f, GFP_KERNEL); dma_fence_put(f); if (r) return r; @@ -281,7 +282,7 @@ int amdgpu_sync_kfd(struct amdgpu_sync *sync, struct dma_resv *resv) if (fence_owner != AMDGPU_FENCE_OWNER_KFD) continue; - r = amdgpu_sync_fence(sync, f); + r = amdgpu_sync_fence(sync, f, GFP_KERNEL); if (r) break; } @@ -388,7 +389,7 @@ int amdgpu_sync_clone(struct amdgpu_sync *source, struct amdgpu_sync *clone) hash_for_each_safe(source->fences, i, tmp, e, node) { f = e->fence; if (!dma_fence_is_signaled(f)) { - r = amdgpu_sync_fence(clone, f); + r = amdgpu_sync_fence(clone, f, GFP_KERNEL); if (r) return r; } else { diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_sync.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_sync.h index a91a8eaf808b1..51eb4382c91eb 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_sync.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_sync.h @@ -47,7 +47,8 @@ struct amdgpu_sync { }; void amdgpu_sync_create(struct amdgpu_sync *sync); -int amdgpu_sync_fence(struct amdgpu_sync *sync, struct dma_fence *f); +int amdgpu_sync_fence(struct amdgpu_sync *sync, struct dma_fence *f, + gfp_t flags); int amdgpu_sync_resv(struct amdgpu_device *adev, struct amdgpu_sync *sync, struct dma_resv *resv, enum amdgpu_sync_mode mode, void *owner); -- 2.39.5

2 weeks, 4 days

[PATCH AUTOSEL 6.14 060/642] drm/amdgpu: rework how isolation is enforced v2

by Sasha Levin

From: Christian König <christian.koenig(a)amd.com> [ Upstream commit bd22e44ad415ac22e3a4f9a983d2a085f6cb4427 ] Limiting the number of available VMIDs to enforce isolation causes some issues with gang submit and applying certain HW workarounds which require multiple VMIDs to work correctly. So instead start to track all submissions to the relevant engines in a per partition data structure and use the dma_fences of the submissions to enforce isolation similar to what a VMID limit does. v2: use ~0l for jobs without isolation to distinct it from kernel submissions which uses NULL for the owner. Add some warning when we are OOM. Signed-off-by: Christian König <christian.koenig(a)amd.com> Acked-by: Srinivasan Shanmugam <srinivasan.shanmugam(a)amd.com> Signed-off-by: Alex Deucher <alexander.deucher(a)amd.com> Signed-off-by: Sasha Levin <sashal(a)kernel.org> --- drivers/gpu/drm/amd/amdgpu/amdgpu.h | 13 ++- drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 98 +++++++++++++++++++++- drivers/gpu/drm/amd/amdgpu/amdgpu_ids.c | 43 ++++------ drivers/gpu/drm/amd/amdgpu/amdgpu_job.c | 16 +++- drivers/gpu/drm/amd/amdgpu/amdgpu_sync.c | 19 +++++ drivers/gpu/drm/amd/amdgpu/amdgpu_sync.h | 1 + 6 files changed, 155 insertions(+), 35 deletions(-) diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h index 98f0c12df12bc..9a61f5fe3245a 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h @@ -1187,9 +1187,15 @@ struct amdgpu_device { bool debug_enable_ras_aca; bool debug_exp_resets; - bool enforce_isolation[MAX_XCP]; - /* Added this mutex for cleaner shader isolation between GFX and compute processes */ + /* Protection for the following isolation structure */ struct mutex enforce_isolation_mutex; + bool enforce_isolation[MAX_XCP]; + struct amdgpu_isolation { + void *owner; + struct dma_fence *spearhead; + struct amdgpu_sync active; + struct amdgpu_sync prev; + } isolation[MAX_XCP]; struct amdgpu_init_level *init_lvl; }; @@ -1470,6 +1476,9 @@ void amdgpu_device_pcie_port_wreg(struct amdgpu_device *adev, struct dma_fence *amdgpu_device_get_gang(struct amdgpu_device *adev); struct dma_fence *amdgpu_device_switch_gang(struct amdgpu_device *adev, struct dma_fence *gang); +struct dma_fence *amdgpu_device_enforce_isolation(struct amdgpu_device *adev, + struct amdgpu_ring *ring, + struct amdgpu_job *job); bool amdgpu_device_has_display_hardware(struct amdgpu_device *adev); ssize_t amdgpu_get_soft_full_reset_mask(struct amdgpu_ring *ring); ssize_t amdgpu_show_reset_mask(char *buf, uint32_t supported_reset); diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c index 71e8a76180ad6..e298b48488c22 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c @@ -4232,6 +4232,11 @@ int amdgpu_device_init(struct amdgpu_device *adev, mutex_init(&adev->gfx.reset_sem_mutex); /* Initialize the mutex for cleaner shader isolation between GFX and compute processes */ mutex_init(&adev->enforce_isolation_mutex); + for (i = 0; i < MAX_XCP; ++i) { + adev->isolation[i].spearhead = dma_fence_get_stub(); + amdgpu_sync_create(&adev->isolation[i].active); + amdgpu_sync_create(&adev->isolation[i].prev); + } mutex_init(&adev->gfx.kfd_sch_mutex); amdgpu_device_init_apu_flags(adev); @@ -4731,7 +4736,7 @@ void amdgpu_device_fini_hw(struct amdgpu_device *adev) void amdgpu_device_fini_sw(struct amdgpu_device *adev) { - int idx; + int i, idx; bool px; amdgpu_device_ip_fini(adev); @@ -4739,6 +4744,11 @@ void amdgpu_device_fini_sw(struct amdgpu_device *adev) amdgpu_ucode_release(&adev->firmware.gpu_info_fw); adev->accel_working = false; dma_fence_put(rcu_dereference_protected(adev->gang_submit, true)); + for (i = 0; i < MAX_XCP; ++i) { + dma_fence_put(adev->isolation[i].spearhead); + amdgpu_sync_free(&adev->isolation[i].active); + amdgpu_sync_free(&adev->isolation[i].prev); + } amdgpu_reset_fini(adev); @@ -6875,6 +6885,92 @@ struct dma_fence *amdgpu_device_switch_gang(struct amdgpu_device *adev, return NULL; } +/** + * amdgpu_device_enforce_isolation - enforce HW isolation + * @adev: the amdgpu device pointer + * @ring: the HW ring the job is supposed to run on + * @job: the job which is about to be pushed to the HW ring + * + * Makes sure that only one client at a time can use the GFX block. + * Returns: The dependency to wait on before the job can be pushed to the HW. + * The function is called multiple times until NULL is returned. + */ +struct dma_fence *amdgpu_device_enforce_isolation(struct amdgpu_device *adev, + struct amdgpu_ring *ring, + struct amdgpu_job *job) +{ + struct amdgpu_isolation *isolation = &adev->isolation[ring->xcp_id]; + struct drm_sched_fence *f = job->base.s_fence; + struct dma_fence *dep; + void *owner; + int r; + + /* + * For now enforce isolation only for the GFX block since we only need + * the cleaner shader on those rings. + */ + if (ring->funcs->type != AMDGPU_RING_TYPE_GFX && + ring->funcs->type != AMDGPU_RING_TYPE_COMPUTE) + return NULL; + + /* + * All submissions where enforce isolation is false are handled as if + * they come from a single client. Use ~0l as the owner to distinct it + * from kernel submissions where the owner is NULL. + */ + owner = job->enforce_isolation ? f->owner : (void *)~0l; + + mutex_lock(&adev->enforce_isolation_mutex); + + /* + * The "spearhead" submission is the first one which changes the + * ownership to its client. We always need to wait for it to be + * pushed to the HW before proceeding with anything. + */ + if (&f->scheduled != isolation->spearhead && + !dma_fence_is_signaled(isolation->spearhead)) { + dep = isolation->spearhead; + goto out_grab_ref; + } + + if (isolation->owner != owner) { + + /* + * Wait for any gang to be assembled before switching to a + * different owner or otherwise we could deadlock the + * submissions. + */ + if (!job->gang_submit) { + dep = amdgpu_device_get_gang(adev); + if (!dma_fence_is_signaled(dep)) + goto out_return_dep; + dma_fence_put(dep); + } + + dma_fence_put(isolation->spearhead); + isolation->spearhead = dma_fence_get(&f->scheduled); + amdgpu_sync_move(&isolation->active, &isolation->prev); + isolation->owner = owner; + } + + /* + * Specifying the ring here helps to pipeline submissions even when + * isolation is enabled. If that is not desired for testing NULL can be + * used instead of the ring to enforce a CPU round trip while switching + * between clients. + */ + dep = amdgpu_sync_peek_fence(&isolation->prev, ring); + r = amdgpu_sync_fence(&isolation->active, &f->finished, GFP_NOWAIT); + if (r) + DRM_WARN("OOM tracking isolation\n"); + +out_grab_ref: + dma_fence_get(dep); +out_return_dep: + mutex_unlock(&adev->enforce_isolation_mutex); + return dep; +} + bool amdgpu_device_has_display_hardware(struct amdgpu_device *adev) { switch (adev->asic_type) { diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ids.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ids.c index 8e712a11aba5d..9008b7388e897 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ids.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ids.c @@ -287,40 +287,27 @@ static int amdgpu_vmid_grab_reserved(struct amdgpu_vm *vm, (*id)->flushed_updates < updates || !(*id)->last_flush || ((*id)->last_flush->context != fence_context && - !dma_fence_is_signaled((*id)->last_flush))) { + !dma_fence_is_signaled((*id)->last_flush))) + needs_flush = true; + + if ((*id)->owner != vm->immediate.fence_context || + (!adev->vm_manager.concurrent_flush && needs_flush)) { struct dma_fence *tmp; - /* Wait for the gang to be assembled before using a - * reserved VMID or otherwise the gang could deadlock. + /* Don't use per engine and per process VMID at the + * same time */ - tmp = amdgpu_device_get_gang(adev); - if (!dma_fence_is_signaled(tmp) && tmp != job->gang_submit) { + if (adev->vm_manager.concurrent_flush) + ring = NULL; + + /* to prevent one context starved by another context */ + (*id)->pd_gpu_addr = 0; + tmp = amdgpu_sync_peek_fence(&(*id)->active, ring); + if (tmp) { *id = NULL; - *fence = tmp; + *fence = dma_fence_get(tmp); return 0; } - dma_fence_put(tmp); - - /* Make sure the id is owned by the gang before proceeding */ - if (!job->gang_submit || - (*id)->owner != vm->immediate.fence_context) { - - /* Don't use per engine and per process VMID at the - * same time - */ - if (adev->vm_manager.concurrent_flush) - ring = NULL; - - /* to prevent one context starved by another context */ - (*id)->pd_gpu_addr = 0; - tmp = amdgpu_sync_peek_fence(&(*id)->active, ring); - if (tmp) { - *id = NULL; - *fence = dma_fence_get(tmp); - return 0; - } - } - needs_flush = true; } /* Good we can use this VMID. Remember this submission as diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c index 100f044759435..685c61a05af85 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c @@ -342,17 +342,24 @@ amdgpu_job_prepare_job(struct drm_sched_job *sched_job, { struct amdgpu_ring *ring = to_amdgpu_ring(s_entity->rq->sched); struct amdgpu_job *job = to_amdgpu_job(sched_job); - struct dma_fence *fence = NULL; + struct dma_fence *fence; int r; r = drm_sched_entity_error(s_entity); if (r) goto error; - if (job->gang_submit) + if (job->gang_submit) { fence = amdgpu_device_switch_gang(ring->adev, job->gang_submit); + if (fence) + return fence; + } + + fence = amdgpu_device_enforce_isolation(ring->adev, ring, job); + if (fence) + return fence; - if (!fence && job->vm && !job->vmid) { + if (job->vm && !job->vmid) { r = amdgpu_vmid_grab(job->vm, ring, job, &fence); if (r) { dev_err(ring->adev->dev, "Error getting VM ID (%d)\n", r); @@ -365,9 +372,10 @@ amdgpu_job_prepare_job(struct drm_sched_job *sched_job, */ if (!fence) job->vm = NULL; + return fence; } - return fence; + return NULL; error: dma_fence_set_error(&job->base.s_fence->finished, r); diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_sync.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_sync.c index c586ab4c911bf..d75715b3f1870 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_sync.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_sync.c @@ -399,6 +399,25 @@ int amdgpu_sync_clone(struct amdgpu_sync *source, struct amdgpu_sync *clone) return 0; } +/** + * amdgpu_sync_move - move all fences from src to dst + * + * @src: source of the fences, empty after function + * @dst: destination for the fences + * + * Moves all fences from source to destination. All fences in destination are + * freed and source is empty after the function call. + */ +void amdgpu_sync_move(struct amdgpu_sync *src, struct amdgpu_sync *dst) +{ + unsigned int i; + + amdgpu_sync_free(dst); + + for (i = 0; i < HASH_SIZE(src->fences); ++i) + hlist_move_list(&src->fences[i], &dst->fences[i]); +} + /** * amdgpu_sync_push_to_job - push fences into job * @sync: sync object to get the fences from diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_sync.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_sync.h index e3272dce798d7..a91a8eaf808b1 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_sync.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_sync.h @@ -56,6 +56,7 @@ struct dma_fence *amdgpu_sync_peek_fence(struct amdgpu_sync *sync, struct amdgpu_ring *ring); struct dma_fence *amdgpu_sync_get_fence(struct amdgpu_sync *sync); int amdgpu_sync_clone(struct amdgpu_sync *source, struct amdgpu_sync *clone); +void amdgpu_sync_move(struct amdgpu_sync *src, struct amdgpu_sync *dst); int amdgpu_sync_push_to_job(struct amdgpu_sync *sync, struct amdgpu_job *job); int amdgpu_sync_wait(struct amdgpu_sync *sync, bool intr); void amdgpu_sync_free(struct amdgpu_sync *sync); -- 2.39.5

2 weeks, 4 days

[PATCH AUTOSEL 6.14 059/642] drm/amdgpu: rework how the cleaner shader is emitted v3

by Sasha Levin

From: Christian König <christian.koenig(a)amd.com> [ Upstream commit b7fbcd77bb467d09ba14cb4ec3b121dc85bb3100 ] Instead of emitting the cleaner shader for every job which has the enforce_isolation flag set only emit it for the first submission from every client. v2: add missing NULL check v3: fix another NULL pointer deref Signed-off-by: Christian König <christian.koenig(a)amd.com> Acked-by: Srinivasan Shanmugam <srinivasan.shanmugam(a)amd.com> Signed-off-by: Alex Deucher <alexander.deucher(a)amd.com> Signed-off-by: Sasha Levin <sashal(a)kernel.org> --- drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 27 ++++++++++++++++++++------ 1 file changed, 21 insertions(+), 6 deletions(-) diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c index 22aa4a8f11891..f0d675c0fc69c 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c @@ -754,6 +754,7 @@ int amdgpu_vm_flush(struct amdgpu_ring *ring, struct amdgpu_job *job, bool need_pipe_sync) { struct amdgpu_device *adev = ring->adev; + struct amdgpu_isolation *isolation = &adev->isolation[ring->xcp_id]; unsigned vmhub = ring->vm_hub; struct amdgpu_vmid_mgr *id_mgr = &adev->vm_manager.id_mgr[vmhub]; struct amdgpu_vmid *id = &id_mgr->ids[job->vmid]; @@ -761,8 +762,9 @@ int amdgpu_vm_flush(struct amdgpu_ring *ring, struct amdgpu_job *job, bool gds_switch_needed = ring->funcs->emit_gds_switch && job->gds_switch_needed; bool vm_flush_needed = job->vm_needs_flush; - struct dma_fence *fence = NULL; + bool cleaner_shader_needed = false; bool pasid_mapping_needed = false; + struct dma_fence *fence = NULL; unsigned int patch; int r; @@ -785,8 +787,12 @@ int amdgpu_vm_flush(struct amdgpu_ring *ring, struct amdgpu_job *job, pasid_mapping_needed &= adev->gmc.gmc_funcs->emit_pasid_mapping && ring->funcs->emit_wreg; + cleaner_shader_needed = adev->gfx.enable_cleaner_shader && + ring->funcs->emit_cleaner_shader && job->base.s_fence && + &job->base.s_fence->scheduled == isolation->spearhead; + if (!vm_flush_needed && !gds_switch_needed && !need_pipe_sync && - !(job->enforce_isolation && !job->vmid)) + !cleaner_shader_needed) return 0; amdgpu_ring_ib_begin(ring); @@ -797,9 +803,7 @@ int amdgpu_vm_flush(struct amdgpu_ring *ring, struct amdgpu_job *job, if (need_pipe_sync) amdgpu_ring_emit_pipeline_sync(ring); - if (adev->gfx.enable_cleaner_shader && - ring->funcs->emit_cleaner_shader && - job->enforce_isolation) + if (cleaner_shader_needed) ring->funcs->emit_cleaner_shader(ring); if (vm_flush_needed) { @@ -821,7 +825,7 @@ int amdgpu_vm_flush(struct amdgpu_ring *ring, struct amdgpu_job *job, job->oa_size); } - if (vm_flush_needed || pasid_mapping_needed) { + if (vm_flush_needed || pasid_mapping_needed || cleaner_shader_needed) { r = amdgpu_fence_emit(ring, &fence, NULL, 0); if (r) return r; @@ -843,6 +847,17 @@ int amdgpu_vm_flush(struct amdgpu_ring *ring, struct amdgpu_job *job, id->pasid_mapping = dma_fence_get(fence); mutex_unlock(&id_mgr->lock); } + + /* + * Make sure that all other submissions wait for the cleaner shader to + * finish before we push them to the HW. + */ + if (cleaner_shader_needed) { + mutex_lock(&adev->enforce_isolation_mutex); + dma_fence_put(isolation->spearhead); + isolation->spearhead = dma_fence_get(fence); + mutex_unlock(&adev->enforce_isolation_mutex); + } dma_fence_put(fence); amdgpu_ring_patch_cond_exec(ring, patch); -- 2.39.5

2 weeks, 4 days

[PATCH v2 0/6] Replace CONFIG_DMABUF_SYSFS_STATS with BPF

by T.J. Mercier

Until CONFIG_DMABUF_SYSFS_STATS was added [1] it was only possible to perform per-buffer accounting with debugfs which is not suitable for production environments. Eventually we discovered the overhead with per-buffer sysfs file creation/removal was significantly impacting allocation and free times, and exacerbated kernfs lock contention. [2] dma_buf_stats_setup() is responsible for 39% of single-page buffer creation duration, or 74% of single-page dma_buf_export() duration when stressing dmabuf allocations and frees. I prototyped a change from per-buffer to per-exporter statistics with a RCU protected list of exporter allocations that accommodates most (but not all) of our use-cases and avoids almost all of the sysfs overhead. While that adds less overhead than per-buffer sysfs, and less even than the maintenance of the dmabuf debugfs_list, it's still *additional* overhead on top of the debugfs_list and doesn't give us per-buffer info. This series uses the existing dmabuf debugfs_list to implement a BPF dmabuf iterator, which adds no overhead to buffer allocation/free and provides per-buffer info. The list has been moved outside of CONFIG_DEBUG_FS scope so that it is always populated. The BPF program loaded by userspace that extracts per-buffer information gets to define its own interface which avoids the lack of ABI stability with debugfs. As this is a replacement for our use of CONFIG_DMABUF_SYSFS_STATS, the last patch is a RFC for removing it from the kernel. Please see my suggestion there regarding the timeline for that. [1] https://lore.kernel.org/linux-media/20201210044400.1080308-1-hridya@google.… [2] https://lore.kernel.org/all/20220516171315.2400578-1-tjmercier@google.com v1: https://lore.kernel.org/all/20250414225227.3642618-1-tjmercier@google.com v1 -> v2: Make the DMA buffer list independent of CONFIG_DEBUG_FS per Christian König Add CONFIG_DMA_SHARED_BUFFER check to kernel/bpf/Makefile per kernel test robot Use BTF_ID_LIST_SINGLE instead of BTF_ID_LIST_GLOBAL_SINGLE per Song Liu Fixup comment style, mixing code/declarations, and use ASSERT_OK_FD in selftest per Song Liu Add BPF_ITER_RESCHED feature to bpf_dmabuf_reg_info per Alexei Starovoitov Add open-coded iterator and selftest per Alexei Starovoitov Add a second test buffer from the system dmabuf heap to selftests Use the BPF program we'll use in production for selftest per Alexei Starovoitov https://r.android.com/c/platform/system/bpfprogs/+/3616123/2/dmabufIter.c https://r.android.com/c/platform/system/memory/libmeminfo/+/3614259/1/libdm… T.J. Mercier (6): dma-buf: Rename and expose debugfs symbols bpf: Add dmabuf iterator bpf: Add open coded dmabuf iterator selftests/bpf: Add test for dmabuf_iter selftests/bpf: Add test for open coded dmabuf_iter RFC: dma-buf: Remove DMA-BUF statistics .../ABI/testing/sysfs-kernel-dmabuf-buffers | 24 -- Documentation/driver-api/dma-buf.rst | 5 - drivers/dma-buf/Kconfig | 15 - drivers/dma-buf/Makefile | 1 - drivers/dma-buf/dma-buf-sysfs-stats.c | 202 -------------- drivers/dma-buf/dma-buf-sysfs-stats.h | 35 --- drivers/dma-buf/dma-buf.c | 58 +--- include/linux/dma-buf.h | 6 +- kernel/bpf/Makefile | 3 + kernel/bpf/dmabuf_iter.c | 177 ++++++++++++ kernel/bpf/helpers.c | 5 + .../testing/selftests/bpf/bpf_experimental.h | 5 + tools/testing/selftests/bpf/config | 3 + .../selftests/bpf/prog_tests/dmabuf_iter.c | 258 ++++++++++++++++++ .../testing/selftests/bpf/progs/dmabuf_iter.c | 91 ++++++ 15 files changed, 561 insertions(+), 327 deletions(-) delete mode 100644 Documentation/ABI/testing/sysfs-kernel-dmabuf-buffers delete mode 100644 drivers/dma-buf/dma-buf-sysfs-stats.c delete mode 100644 drivers/dma-buf/dma-buf-sysfs-stats.h create mode 100644 kernel/bpf/dmabuf_iter.c create mode 100644 tools/testing/selftests/bpf/prog_tests/dmabuf_iter.c create mode 100644 tools/testing/selftests/bpf/progs/dmabuf_iter.c base-commit: 0af2f6be1b4281385b618cb86ad946eded089ac8 -- 2.49.0.906.g1f30a19c02-goog

2 weeks, 4 days

[PATCH v4 00/33] drm/msm: sparse / "VM_BIND" support

by Rob Clark

From: Rob Clark <robdclark(a)chromium.org> Conversion to DRM GPU VA Manager[1], and adding support for Vulkan Sparse Memory[2] in the form of: 1. A new VM_BIND submitqueue type for executing VM MSM_SUBMIT_BO_OP_MAP/ MAP_NULL/UNMAP commands 2. A new VM_BIND ioctl to allow submitting batches of one or more MAP/MAP_NULL/UNMAP commands to a VM_BIND submitqueue I did not implement support for synchronous VM_BIND commands. Since userspace could just immediately wait for the `SUBMIT` to complete, I don't think we need this extra complexity in the kernel. Synchronous/immediate VM_BIND operations could be implemented with a 2nd VM_BIND submitqueue. The corresponding mesa MR: https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/32533 Changes in v4: - Replace selftests_running flag with IO_PGTABLE_QUIRK_NO_WARN_ON [Robin Murphy] - Rework msm_gem_vm_sm_step_remap() for cases that orig_vma is evicted to solve some crashes - Block when drm_file is closed until pending VM_BIND ops complete, before tearing down the VM's scheduler, to solve some memory leaks. - Link to v3: https://lore.kernel.org/all/20250428205619.227835-1-robdclark@gmail.com/ Changes in v3: - Switched to separate VM_BIND ioctl. This makes the UABI a bit cleaner, but OTOH the userspace code was cleaner when the end result of either type of VkQueue lead to the same ioctl. So I'm a bit on the fence. - Switched to doing the gpuvm bookkeeping synchronously, and only deferring the pgtable updates. This avoids needing to hold any resv locks in the fence signaling path, resolving the last shrinker related lockdep complaints. OTOH it means userspace can trigger invalid pgtable updates with multiple VM_BIND queues. In this case, we ensure that unmaps happen completely (to prevent userspace from using this to access free'd pages), mark the context as unusable, and move on with life. - Link to v2: https://lore.kernel.org/all/20250319145425.51935-1-robdclark@gmail.com/ Changes in v2: - Dropped Bibek Kumar Patro's arm-smmu patches[3], which have since been merged. - Pre-allocate all the things, and drop HACK patch which disabled shrinker. This includes ensuring that vm_bo objects are allocated up front, pre- allocating VMA objects, and pre-allocating pages used for pgtable updates. The latter utilizes io_pgtable_cfg callbacks for pgtable alloc/free, that were initially added for panthor. - Add back support for BO dumping for devcoredump. - Link to v1 (RFC): https://lore.kernel.org/dri-devel/20241207161651.410556-1-robdclark@gmail.c… [1] https://www.kernel.org/doc/html/next/gpu/drm-mm.html#drm-gpuvm [2] https://docs.vulkan.org/spec/latest/chapters/sparsemem.html [3] https://patchwork.kernel.org/project/linux-arm-kernel/list/?series=909700 Rob Clark (33): drm/gpuvm: Don't require obj lock in destructor path drm/gpuvm: Allow VAs to hold soft reference to BOs iommu/io-pgtable-arm: Add quirk to quiet WARN_ON() drm/msm: Rename msm_file_private -> msm_context drm/msm: Improve msm_context comments drm/msm: Rename msm_gem_address_space -> msm_gem_vm drm/msm: Remove vram carveout support drm/msm: Collapse vma allocation and initialization drm/msm: Collapse vma close and delete drm/msm: Don't close VMAs on purge drm/msm: drm_gpuvm conversion drm/msm: Convert vm locking drm/msm: Use drm_gpuvm types more drm/msm: Split out helper to get iommu prot flags drm/msm: Add mmu support for non-zero offset drm/msm: Add PRR support drm/msm: Rename msm_gem_vma_purge() -> _unmap() drm/msm: Lazily create context VM drm/msm: Add opt-in for VM_BIND drm/msm: Mark VM as unusable on GPU hangs drm/msm: Add _NO_SHARE flag drm/msm: Crashdump prep for sparse mappings drm/msm: rd dumping prep for sparse mappings drm/msm: Crashdec support for sparse drm/msm: rd dumping support for sparse drm/msm: Extract out syncobj helpers drm/msm: Use DMA_RESV_USAGE_BOOKKEEP/KERNEL drm/msm: Add VM_BIND submitqueue drm/msm: Support IO_PGTABLE_QUIRK_NO_WARN_ON drm/msm: Support pgtable preallocation drm/msm: Split out map/unmap ops drm/msm: Add VM_BIND ioctl drm/msm: Bump UAPI version drivers/gpu/drm/drm_gpuvm.c | 15 +- drivers/gpu/drm/msm/Kconfig | 1 + drivers/gpu/drm/msm/Makefile | 1 + drivers/gpu/drm/msm/adreno/a2xx_gpu.c | 25 +- drivers/gpu/drm/msm/adreno/a2xx_gpummu.c | 5 +- drivers/gpu/drm/msm/adreno/a3xx_gpu.c | 17 +- drivers/gpu/drm/msm/adreno/a4xx_gpu.c | 17 +- drivers/gpu/drm/msm/adreno/a5xx_debugfs.c | 4 +- drivers/gpu/drm/msm/adreno/a5xx_gpu.c | 22 +- drivers/gpu/drm/msm/adreno/a5xx_power.c | 2 +- drivers/gpu/drm/msm/adreno/a5xx_preempt.c | 10 +- drivers/gpu/drm/msm/adreno/a6xx_gmu.c | 32 +- drivers/gpu/drm/msm/adreno/a6xx_gmu.h | 2 +- drivers/gpu/drm/msm/adreno/a6xx_gpu.c | 49 +- drivers/gpu/drm/msm/adreno/a6xx_gpu_state.c | 6 +- drivers/gpu/drm/msm/adreno/a6xx_preempt.c | 10 +- drivers/gpu/drm/msm/adreno/adreno_device.c | 4 - drivers/gpu/drm/msm/adreno/adreno_gpu.c | 88 +- drivers/gpu/drm/msm/adreno/adreno_gpu.h | 23 +- .../drm/msm/disp/dpu1/dpu_encoder_phys_wb.c | 14 +- drivers/gpu/drm/msm/disp/dpu1/dpu_formats.c | 18 +- drivers/gpu/drm/msm/disp/dpu1/dpu_formats.h | 2 +- drivers/gpu/drm/msm/disp/dpu1/dpu_kms.c | 18 +- drivers/gpu/drm/msm/disp/dpu1/dpu_plane.c | 14 +- drivers/gpu/drm/msm/disp/dpu1/dpu_plane.h | 4 +- drivers/gpu/drm/msm/disp/mdp4/mdp4_crtc.c | 6 +- drivers/gpu/drm/msm/disp/mdp4/mdp4_kms.c | 28 +- drivers/gpu/drm/msm/disp/mdp4/mdp4_plane.c | 12 +- drivers/gpu/drm/msm/disp/mdp5/mdp5_crtc.c | 4 +- drivers/gpu/drm/msm/disp/mdp5/mdp5_kms.c | 19 +- drivers/gpu/drm/msm/disp/mdp5/mdp5_plane.c | 12 +- drivers/gpu/drm/msm/dsi/dsi_host.c | 14 +- drivers/gpu/drm/msm/msm_drv.c | 183 +-- drivers/gpu/drm/msm/msm_drv.h | 35 +- drivers/gpu/drm/msm/msm_fb.c | 18 +- drivers/gpu/drm/msm/msm_fbdev.c | 2 +- drivers/gpu/drm/msm/msm_gem.c | 489 +++---- drivers/gpu/drm/msm/msm_gem.h | 217 ++- drivers/gpu/drm/msm/msm_gem_prime.c | 15 + drivers/gpu/drm/msm/msm_gem_shrinker.c | 4 +- drivers/gpu/drm/msm/msm_gem_submit.c | 295 ++-- drivers/gpu/drm/msm/msm_gem_vma.c | 1288 +++++++++++++++-- drivers/gpu/drm/msm/msm_gpu.c | 171 ++- drivers/gpu/drm/msm/msm_gpu.h | 132 +- drivers/gpu/drm/msm/msm_iommu.c | 298 +++- drivers/gpu/drm/msm/msm_kms.c | 18 +- drivers/gpu/drm/msm/msm_kms.h | 2 +- drivers/gpu/drm/msm/msm_mmu.h | 38 +- drivers/gpu/drm/msm/msm_rd.c | 62 +- drivers/gpu/drm/msm/msm_ringbuffer.c | 4 +- drivers/gpu/drm/msm/msm_submitqueue.c | 96 +- drivers/gpu/drm/msm/msm_syncobj.c | 172 +++ drivers/gpu/drm/msm/msm_syncobj.h | 37 + drivers/iommu/io-pgtable-arm.c | 27 +- include/drm/drm_gpuvm.h | 12 +- include/linux/io-pgtable.h | 8 + include/uapi/drm/msm_drm.h | 149 +- 57 files changed, 3043 insertions(+), 1227 deletions(-) create mode 100644 drivers/gpu/drm/msm/msm_syncobj.c create mode 100644 drivers/gpu/drm/msm/msm_syncobj.h -- 2.49.0

2 weeks, 4 days

Re: [PATCH RESEND] drm/msm: fix a potential memory leak issue in submit_create()

by Rob Clark

On Wed, Apr 23, 2025 at 8:28 PM Haoxiang Li <haoxiang_li2024(a)163.com> wrote: > > The memory allocated by msm_fence_alloc() actually is the > container of msm_fence_alloc()'s return value. Thus, just > free its return value is not enough. > Add a helper 'msm_fence_free()' in msm_fence.h/msm_fence.c > to do the complete job. > > Fixes: f94e6a51e17c ("drm/msm: Pre-allocate hw_fence") > Cc: stable(a)vger.kernel.org > Signed-off-by: Haoxiang Li <haoxiang_li2024(a)163.com> > --- > drivers/gpu/drm/msm/msm_fence.c | 7 +++++++ > drivers/gpu/drm/msm/msm_fence.h | 1 + > drivers/gpu/drm/msm/msm_gem_submit.c | 2 +- > 3 files changed, 9 insertions(+), 1 deletion(-) > > diff --git a/drivers/gpu/drm/msm/msm_fence.c b/drivers/gpu/drm/msm/msm_fence.c > index d41e5a6bbee0..72641e6a627d 100644 > --- a/drivers/gpu/drm/msm/msm_fence.c > +++ b/drivers/gpu/drm/msm/msm_fence.c > @@ -183,6 +183,13 @@ msm_fence_alloc(void) > return &f->base; > } > > +void msm_fence_free(struct dma_fence *fence) > +{ > + struct msm_fence *f = to_msm_fence(fence); > + > + kfree(f); > +} > + > void > msm_fence_init(struct dma_fence *fence, struct msm_fence_context *fctx) > { > diff --git a/drivers/gpu/drm/msm/msm_fence.h b/drivers/gpu/drm/msm/msm_fence.h > index 148196375a0b..635c68629070 100644 > --- a/drivers/gpu/drm/msm/msm_fence.h > +++ b/drivers/gpu/drm/msm/msm_fence.h > @@ -82,6 +82,7 @@ bool msm_fence_completed(struct msm_fence_context *fctx, uint32_t fence); > void msm_update_fence(struct msm_fence_context *fctx, uint32_t fence); > > struct dma_fence * msm_fence_alloc(void); > +void msm_fence_free(struct dma_fence *fence); > void msm_fence_init(struct dma_fence *fence, struct msm_fence_context *fctx); > > static inline bool > diff --git a/drivers/gpu/drm/msm/msm_gem_submit.c b/drivers/gpu/drm/msm/msm_gem_submit.c > index 3e9aa2cc38ef..213baa5bca5e 100644 > --- a/drivers/gpu/drm/msm/msm_gem_submit.c > +++ b/drivers/gpu/drm/msm/msm_gem_submit.c > @@ -56,7 +56,7 @@ static struct msm_gem_submit *submit_create(struct drm_device *dev, > > ret = drm_sched_job_init(&submit->base, queue->entity, 1, queue); > if (ret) { > - kfree(submit->hw_fence); > + msm_fence_free(submit->hw_fence); `struct dma_fence base` is the first field in `struct msm_fence`, so to_msm_fence() is just a pointer cast. Ie. it is fine to pass it to kfree() as-is BR, -R > kfree(submit); > return ERR_PTR(ret); > } > -- > 2.25.1 >

2 weeks, 5 days

Re: [PATCH 2/3] drm/prime: Support importing DMA-BUF without sg_table

by kernel test robot

Hi, kernel test robot noticed the following build warnings: [auto build test WARNING on jic23-iio/togreg] [also build test WARNING on char-misc/char-misc-testing char-misc/char-misc-next char-misc/char-misc-linus usb/usb-testing usb/usb-next usb/usb-linus xen-tip/linux-next linus/master v6.15-rc4] [cannot apply to tegra/for-next drm-xe/drm-xe-next rmk-arm/drm-armada-devel rmk-arm/drm-armada-fixes next-20250501] [If your patch is applied to the wrong git tree, kindly drop us a note. And when submitting patch, we suggest to use '--base' as documented in https://git-scm.com/docs/git-format-patch#_base_tree_information] url: https://github.com/intel-lab-lkp/linux/commits/oushixiong1025-163-com/drm-p… base: https://git.kernel.org/pub/scm/linux/kernel/git/jic23/iio.git togreg patch link: https://lore.kernel.org/r/20250430085658.540746-2-oushixiong1025%40163.com patch subject: [PATCH 2/3] drm/prime: Support importing DMA-BUF without sg_table config: arc-randconfig-002-20250501 (https://download.01.org/0day-ci/archive/20250502/202505022224.FCDQ8TCB-lkp@…) compiler: arc-linux-gcc (GCC) 14.2.0 reproduce (this is a W=1 build): (https://download.01.org/0day-ci/archive/20250502/202505022224.FCDQ8TCB-lkp@…) If you fix the issue in a separate patch/commit (i.e. not just a new version of the same patch/commit), kindly add following tags | Reported-by: kernel test robot <lkp(a)intel.com> | Closes: https://lore.kernel.org/oe-kbuild-all/202505022224.FCDQ8TCB-lkp@intel.com/ All warnings (new ones prefixed by >>): >> drivers/gpu/drm/drm_prime.c:925:24: warning: no previous prototype for 'drm_gem_prime_import_dev_skip_map' [-Wmissing-prototypes] 925 | struct drm_gem_object *drm_gem_prime_import_dev_skip_map(struct drm_device *dev, | ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ vim +/drm_gem_prime_import_dev_skip_map +925 drivers/gpu/drm/drm_prime.c 913 914 /** 915 * drm_gem_prime_import_dev_skip_map - core implementation of the import callback 916 * @dev: drm_device to import into 917 * @dma_buf: dma-buf object to import 918 * @attach_dev: struct device to dma_buf attach 919 * 920 * This function exports a dma-buf without get it's scatter/gather table. 921 * 922 * Drivers who need to get an scatter/gather table for objects need to call 923 * drm_gem_prime_import_dev() instead. 924 */ > 925 struct drm_gem_object *drm_gem_prime_import_dev_skip_map(struct drm_device *dev, 926 struct dma_buf *dma_buf, 927 struct device *attach_dev) 928 { 929 struct dma_buf_attachment *attach; 930 struct drm_gem_object *obj; 931 int ret; 932 933 if (dma_buf->ops == &drm_gem_prime_dmabuf_ops) { 934 obj = dma_buf->priv; 935 if (obj->dev == dev) { 936 /* 937 * Importing dmabuf exported from our own gem increases 938 * refcount on gem itself instead of f_count of dmabuf. 939 */ 940 drm_gem_object_get(obj); 941 return obj; 942 } 943 } 944 945 attach = dma_buf_attach(dma_buf, attach_dev, true); 946 if (IS_ERR(attach)) 947 return ERR_CAST(attach); 948 949 get_dma_buf(dma_buf); 950 951 obj = dev->driver->gem_prime_import_attachment(dev, attach); 952 if (IS_ERR(obj)) { 953 ret = PTR_ERR(obj); 954 goto fail_detach; 955 } 956 957 obj->import_attach = attach; 958 obj->resv = dma_buf->resv; 959 960 return obj; 961 962 fail_detach: 963 dma_buf_detach(dma_buf, attach); 964 dma_buf_put(dma_buf); 965 966 return ERR_PTR(ret); 967 } 968 EXPORT_SYMBOL(drm_gem_prime_import_dev_skip_map); 969 -- 0-DAY CI Kernel Test Service https://github.com/intel/lkp-tests/wiki

3 weeks

2025

2024

2023

2022

2021

2020

2019

2018

2017

2016

2015

2014

2013

2012

2011

Linaro-mm-sig May 2025