[RFC,v3,18/21] drm/ttm: Convert ttm vm to using drm_exec

Message ID	20240521071639.77614-19-thomas.hellstrom@linux.intel.com (mailing list archive)
State	New, archived
Headers	show Return-Path: <dri-devel-bounces@lists.freedesktop.org> From: =?utf-8?q?Thomas_Hellstr=C3=B6m?= <thomas.hellstrom@linux.intel.com> To: intel-xe@lists.freedesktop.org Cc: =?utf-8?q?Thomas_Hellstr=C3=B6m?= <thomas.hellstrom@linux.intel.com>, =?utf-8?q?Christian_K=C3=B6nig?= <christian.koenig@amd.com>, Somalapuram Amaranath <Amaranath.Somalapuram@amd.com>, Matthew Brost <matthew.brost@intel.com>, dri-devel@lists.freedesktop.org Subject: [RFC PATCH v3 18/21] drm/ttm: Convert ttm vm to using drm_exec Date: Tue, 21 May 2024 09:16:36 +0200 Message-ID: <20240521071639.77614-19-thomas.hellstrom@linux.intel.com> In-Reply-To: <20240521071639.77614-1-thomas.hellstrom@linux.intel.com> References: <20240521071639.77614-1-thomas.hellstrom@linux.intel.com> MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Precedence: list Errors-To: dri-devel-bounces@lists.freedesktop.org Sender: "dri-devel" <dri-devel-bounces@lists.freedesktop.org>
Series	TTM shrinker helpers and xe buffer object shrinker \| expand [v3,00/21] TTM shrinker helpers and xe buffer object shrinker [v3,01/21] drm/ttm: Allow TTM LRU list nodes of different types [v3,02/21] drm/ttm: Slightly clean up LRU list iteration [v3,03/21] drm/ttm: Use LRU hitches [v3,04/21] drm/ttm, drm/amdgpu, drm/xe: Consider hitch moves within bulk sublist moves [v3,05/21] drm/ttm: Provide a generic LRU walker helper [v3,06/21] drm/ttm: Use the LRU walker helper for swapping [v3,07/21] drm/ttm: Use the LRU walker for eviction [v3,08/21] drm/ttm: Add a virtual base class for graphics memory backup [v3,09/21] drm/ttm/pool: Provide a helper to shrink pages [v3,10/21] drm/ttm: Use fault-injection to test error paths [v3,11/21] drm/ttm, drm/xe: Add a shrinker for xe bos [RFC,v3,12/21] dma-buf/dma-resv: Introduce dma_resv_trylock_ctx() [RFC,v3,13/21] drm/exec: Rework contended locking [RFC,v3,14/21] drm/exec: Introduce a drm_exec_trylock_obj() function [RFC,v3,15/21] drm/exec: Add a snapshot capability [RFC,v3,16/21] drm/exec: Introduce an evict mode [RFC,v3,17/21] drm/ttm: Support drm_exec locking for eviction and swapping [RFC,v3,18/21] drm/ttm: Convert ttm vm to using drm_exec [RFC,v3,19/21] drm/xe: Use drm_exec for fault locking [RFC,v3,20/21] drm/ttm: Use drm_exec_trylock for bo initialization [RFC,v3,21/21] drm/xe: Initial support for drm exec locking during validate

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c index 17e16c971e21..22d61cdb0d88 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c @@ -52,7 +52,7 @@ static vm_fault_t amdgpu_gem_fault(struct vm_fault *vmf) vm_fault_t ret; int idx; - ret = ttm_bo_vm_reserve(bo, vmf); + ret = ttm_bo_vm_reserve(bo, vmf, NULL); if (ret) return ret; @@ -64,7 +64,7 @@ static vm_fault_t amdgpu_gem_fault(struct vm_fault *vmf) } ret = ttm_bo_vm_fault_reserved(vmf, vmf->vma->vm_page_prot, - TTM_BO_VM_NUM_PREFAULT); + TTM_BO_VM_NUM_PREFAULT, NULL); drm_dev_exit(idx); } else { diff --git a/drivers/gpu/drm/i915/gem/i915_gem_ttm.c b/drivers/gpu/drm/i915/gem/i915_gem_ttm.c index e6f177183c0f..c66e2b54c9a2 100644 --- a/drivers/gpu/drm/i915/gem/i915_gem_ttm.c +++ b/drivers/gpu/drm/i915/gem/i915_gem_ttm.c @@ -1046,7 +1046,7 @@ static vm_fault_t vm_fault_ttm(struct vm_fault *vmf) area->vm_flags & VM_WRITE)) return VM_FAULT_SIGBUS; - ret = ttm_bo_vm_reserve(bo, vmf); + ret = ttm_bo_vm_reserve(bo, vmf, NULL); if (ret) return ret; @@ -1108,7 +1108,7 @@ static vm_fault_t vm_fault_ttm(struct vm_fault *vmf) if (drm_dev_enter(dev, &idx)) { ret = ttm_bo_vm_fault_reserved(vmf, vmf->vma->vm_page_prot, - TTM_BO_VM_NUM_PREFAULT); + TTM_BO_VM_NUM_PREFAULT, NULL); drm_dev_exit(idx); } else { ret = ttm_bo_vm_dummy_page(vmf, vmf->vma->vm_page_prot); diff --git a/drivers/gpu/drm/nouveau/nouveau_gem.c b/drivers/gpu/drm/nouveau/nouveau_gem.c index 5a887d67dc0e..bc6901955508 100644 --- a/drivers/gpu/drm/nouveau/nouveau_gem.c +++ b/drivers/gpu/drm/nouveau/nouveau_gem.c @@ -46,7 +46,7 @@ static vm_fault_t nouveau_ttm_fault(struct vm_fault *vmf) pgprot_t prot; vm_fault_t ret; - ret = ttm_bo_vm_reserve(bo, vmf); + ret = ttm_bo_vm_reserve(bo, vmf, NULL); if (ret) return ret; @@ -56,7 +56,7 @@ static vm_fault_t nouveau_ttm_fault(struct vm_fault *vmf) nouveau_bo_del_io_reserve_lru(bo); prot = vm_get_page_prot(vma->vm_flags); - ret = ttm_bo_vm_fault_reserved(vmf, prot, TTM_BO_VM_NUM_PREFAULT); + ret = ttm_bo_vm_fault_reserved(vmf, prot, TTM_BO_VM_NUM_PREFAULT, NULL); nouveau_bo_add_io_reserve_lru(bo); if (ret == VM_FAULT_RETRY && !(vmf->flags & FAULT_FLAG_RETRY_NOWAIT)) return ret; diff --git a/drivers/gpu/drm/radeon/radeon_gem.c b/drivers/gpu/drm/radeon/radeon_gem.c index 2ef201a072f1..f29761b7ca97 100644 --- a/drivers/gpu/drm/radeon/radeon_gem.c +++ b/drivers/gpu/drm/radeon/radeon_gem.c @@ -54,7 +54,7 @@ static vm_fault_t radeon_gem_fault(struct vm_fault *vmf) down_read(&rdev->pm.mclk_lock); - ret = ttm_bo_vm_reserve(bo, vmf); + ret = ttm_bo_vm_reserve(bo, vmf, NULL); if (ret) goto unlock_mclk; @@ -63,7 +63,7 @@ static vm_fault_t radeon_gem_fault(struct vm_fault *vmf) goto unlock_resv; ret = ttm_bo_vm_fault_reserved(vmf, vmf->vma->vm_page_prot, - TTM_BO_VM_NUM_PREFAULT); + TTM_BO_VM_NUM_PREFAULT, NULL); if (ret == VM_FAULT_RETRY && !(vmf->flags & FAULT_FLAG_RETRY_NOWAIT)) goto unlock_mclk; diff --git a/drivers/gpu/drm/ttm/ttm_bo_vm.c b/drivers/gpu/drm/ttm/ttm_bo_vm.c index 4212b8c91dd4..74daa910d0b7 100644 --- a/drivers/gpu/drm/ttm/ttm_bo_vm.c +++ b/drivers/gpu/drm/ttm/ttm_bo_vm.c @@ -31,6 +31,8 @@ #define pr_fmt(fmt) "[TTM] " fmt +#include <drm/drm_exec.h> + #include <drm/ttm/ttm_bo.h> #include <drm/ttm/ttm_placement.h> #include <drm/ttm/ttm_tt.h> @@ -39,7 +41,8 @@ #include <drm/drm_managed.h> static vm_fault_t ttm_bo_vm_fault_idle(struct ttm_buffer_object *bo, - struct vm_fault *vmf) + struct vm_fault *vmf, + struct drm_exec *exec) { long err = 0; @@ -63,7 +66,10 @@ static vm_fault_t ttm_bo_vm_fault_idle(struct ttm_buffer_object *bo, (void)dma_resv_wait_timeout(bo->base.resv, DMA_RESV_USAGE_KERNEL, true, MAX_SCHEDULE_TIMEOUT); - dma_resv_unlock(bo->base.resv); + if (exec) + drm_exec_unlock_obj(exec, &bo->base); + else + dma_resv_unlock(bo->base.resv); ttm_bo_put(bo); return VM_FAULT_RETRY; } @@ -96,6 +102,7 @@ static unsigned long ttm_bo_io_mem_pfn(struct ttm_buffer_object *bo, * ttm_bo_vm_reserve - Reserve a buffer object in a retryable vm callback * @bo: The buffer object * @vmf: The fault structure handed to the callback + * @exec: The drm_exec locking transaction context. May be NULL. * * vm callbacks like fault() and *_mkwrite() allow for the mmap_lock to be dropped * during long waits, and after the wait the callback will be restarted. This @@ -114,15 +121,16 @@ static unsigned long ttm_bo_io_mem_pfn(struct ttm_buffer_object *bo, * VM_FAULT_NOPAGE if blocking wait and retrying was not allowed. */ vm_fault_t ttm_bo_vm_reserve(struct ttm_buffer_object *bo, - struct vm_fault *vmf) + struct vm_fault *vmf, struct drm_exec *exec) { - /* - * Work around locking order reversal in fault / nopfn - * between mmap_lock and bo_reserve: Perform a trylock operation - * for reserve, and if it fails, retry the fault after waiting - * for the buffer to become unreserved. - */ - if (unlikely(!dma_resv_trylock(bo->base.resv))) { + int ret; + + if (exec) + ret = drm_exec_trylock_obj(exec, &bo->base); + else + ret = dma_resv_trylock(bo->base.resv) ? 0 : -EBUSY; + + if (unlikely(ret == -EBUSY)) { /* * If the fault allows retry and this is the first * fault attempt, we try to release the mmap_lock @@ -132,16 +140,26 @@ vm_fault_t ttm_bo_vm_reserve(struct ttm_buffer_object *bo, if (!(vmf->flags & FAULT_FLAG_RETRY_NOWAIT)) { ttm_bo_get(bo); mmap_read_unlock(vmf->vma->vm_mm); - if (!dma_resv_lock_interruptible(bo->base.resv, - NULL)) - dma_resv_unlock(bo->base.resv); + if (exec) { + ret = drm_exec_lock_obj(exec, &bo->base); + if (!ret) + drm_exec_unlock_obj(exec, &bo->base); + } else { + if (!dma_resv_lock_interruptible(bo->base.resv, + NULL)) + dma_resv_unlock(bo->base.resv); + } ttm_bo_put(bo); } return VM_FAULT_RETRY; } - if (dma_resv_lock_interruptible(bo->base.resv, NULL)) + if (exec) + ret = drm_exec_lock_obj(exec, &bo->base); + else + ret = dma_resv_lock_interruptible(bo->base.resv, NULL); + if (ret) return VM_FAULT_NOPAGE; } @@ -151,7 +169,10 @@ vm_fault_t ttm_bo_vm_reserve(struct ttm_buffer_object *bo, */ if (bo->ttm && (bo->ttm->page_flags & TTM_TT_FLAG_EXTERNAL)) { if (!(bo->ttm->page_flags & TTM_TT_FLAG_EXTERNAL_MAPPABLE)) { - dma_resv_unlock(bo->base.resv); + if (exec) + drm_exec_unlock_obj(exec, &bo->base); + else + dma_resv_unlock(bo->base.resv); return VM_FAULT_SIGBUS; } } @@ -167,6 +188,7 @@ EXPORT_SYMBOL(ttm_bo_vm_reserve); * @num_prefault: Maximum number of prefault pages. The caller may want to * specify this based on madvice settings and the size of the GPU object * backed by the memory. + * @exec: The struct drm_exec locking transaction context. May be NULL. * * This function inserts one or more page table entries pointing to the * memory backing the buffer object, and then returns a return code @@ -180,7 +202,8 @@ EXPORT_SYMBOL(ttm_bo_vm_reserve); */ vm_fault_t ttm_bo_vm_fault_reserved(struct vm_fault *vmf, pgprot_t prot, - pgoff_t num_prefault) + pgoff_t num_prefault, + struct drm_exec *exec) { struct vm_area_struct *vma = vmf->vma; struct ttm_buffer_object *bo = vma->vm_private_data; @@ -199,7 +222,7 @@ vm_fault_t ttm_bo_vm_fault_reserved(struct vm_fault *vmf, * Wait for buffer data in transit, due to a pipelined * move. */ - ret = ttm_bo_vm_fault_idle(bo, vmf); + ret = ttm_bo_vm_fault_idle(bo, vmf, exec); if (unlikely(ret != 0)) return ret; @@ -220,7 +243,8 @@ vm_fault_t ttm_bo_vm_fault_reserved(struct vm_fault *vmf, struct ttm_operation_ctx ctx = { .interruptible = true, .no_wait_gpu = false, - .force_alloc = true + .force_alloc = true, + .exec = exec, }; ttm = bo->ttm; @@ -324,25 +348,34 @@ vm_fault_t ttm_bo_vm_fault(struct vm_fault *vmf) pgprot_t prot; struct ttm_buffer_object *bo = vma->vm_private_data; struct drm_device *ddev = bo->base.dev; + struct drm_exec exec; vm_fault_t ret; - int idx; - - ret = ttm_bo_vm_reserve(bo, vmf); - if (ret) - return ret; - - prot = vma->vm_page_prot; - if (drm_dev_enter(ddev, &idx)) { - ret = ttm_bo_vm_fault_reserved(vmf, prot, TTM_BO_VM_NUM_PREFAULT); - drm_dev_exit(idx); - } else { - ret = ttm_bo_vm_dummy_page(vmf, prot); + int idx, err; + + drm_exec_init(&exec, DRM_EXEC_INTERRUPTIBLE_WAIT, 16); + drm_exec_until_all_locked(&exec) { + ret = ttm_bo_vm_reserve(bo, vmf, &exec); + err = drm_exec_retry_on_contention(&exec, 0); + if (err) + ret = VM_FAULT_NOPAGE; + if (ret) + goto out; + + prot = vma->vm_page_prot; + if (drm_dev_enter(ddev, &idx)) { + ret = ttm_bo_vm_fault_reserved(vmf, prot, TTM_BO_VM_NUM_PREFAULT, + &exec); + drm_dev_exit(idx); + err = drm_exec_retry_on_contention(&exec, 0); + if (err) + ret = VM_FAULT_NOPAGE; + } else { + ret = ttm_bo_vm_dummy_page(vmf, prot); + } } - if (ret == VM_FAULT_RETRY && !(vmf->flags & FAULT_FLAG_RETRY_NOWAIT)) - return ret; - - dma_resv_unlock(bo->base.resv); +out: + drm_exec_fini(&exec); return ret; } EXPORT_SYMBOL(ttm_bo_vm_fault); diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_page_dirty.c b/drivers/gpu/drm/vmwgfx/vmwgfx_page_dirty.c index 74ff2812d66a..fc275afd000c 100644 --- a/drivers/gpu/drm/vmwgfx/vmwgfx_page_dirty.c +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_page_dirty.c @@ -388,7 +388,7 @@ vm_fault_t vmw_bo_vm_mkwrite(struct vm_fault *vmf) */ save_flags = vmf->flags; vmf->flags &= ~FAULT_FLAG_ALLOW_RETRY; - ret = ttm_bo_vm_reserve(bo, vmf); + ret = ttm_bo_vm_reserve(bo, vmf, NULL); vmf->flags = save_flags; if (ret) return ret; @@ -423,7 +423,7 @@ vm_fault_t vmw_bo_vm_fault(struct vm_fault *vmf) pgprot_t prot; vm_fault_t ret; - ret = ttm_bo_vm_reserve(bo, vmf); + ret = ttm_bo_vm_reserve(bo, vmf, NULL); if (ret) return ret; @@ -457,7 +457,7 @@ vm_fault_t vmw_bo_vm_fault(struct vm_fault *vmf) else prot = vm_get_page_prot(vma->vm_flags); - ret = ttm_bo_vm_fault_reserved(vmf, prot, num_prefault); + ret = ttm_bo_vm_fault_reserved(vmf, prot, num_prefault, NULL); if (ret == VM_FAULT_RETRY && !(vmf->flags & FAULT_FLAG_RETRY_NOWAIT)) return ret; diff --git a/drivers/gpu/drm/xe/xe_bo.c b/drivers/gpu/drm/xe/xe_bo.c index 9a0ca2cab7b6..3c56858e0751 100644 --- a/drivers/gpu/drm/xe/xe_bo.c +++ b/drivers/gpu/drm/xe/xe_bo.c @@ -1223,7 +1223,7 @@ static vm_fault_t xe_gem_fault(struct vm_fault *vmf) if (needs_rpm) xe_pm_runtime_get(xe); - ret = ttm_bo_vm_reserve(tbo, vmf); + ret = ttm_bo_vm_reserve(tbo, vmf, NULL); if (ret) goto out; @@ -1231,7 +1231,8 @@ static vm_fault_t xe_gem_fault(struct vm_fault *vmf) trace_xe_bo_cpu_fault(bo); ret = ttm_bo_vm_fault_reserved(vmf, vmf->vma->vm_page_prot, - TTM_BO_VM_NUM_PREFAULT); + TTM_BO_VM_NUM_PREFAULT, + NULL); drm_dev_exit(idx); } else { ret = ttm_bo_vm_dummy_page(vmf, vmf->vma->vm_page_prot); diff --git a/include/drm/ttm/ttm_bo.h b/include/drm/ttm/ttm_bo.h index 1c9f4880abb9..d749de3aa936 100644 --- a/include/drm/ttm/ttm_bo.h +++ b/include/drm/ttm/ttm_bo.h @@ -427,10 +427,12 @@ int ttm_bo_evict_first(struct ttm_device *bdev, struct ttm_resource_manager *man, struct ttm_operation_ctx *ctx); vm_fault_t ttm_bo_vm_reserve(struct ttm_buffer_object *bo, - struct vm_fault *vmf); + struct vm_fault *vmf, + struct drm_exec *exec); vm_fault_t ttm_bo_vm_fault_reserved(struct vm_fault *vmf, pgprot_t prot, - pgoff_t num_prefault); + pgoff_t num_prefault, + struct drm_exec *exec); vm_fault_t ttm_bo_vm_fault(struct vm_fault *vmf); void ttm_bo_vm_open(struct vm_area_struct *vma); void ttm_bo_vm_close(struct vm_area_struct *vma);

[RFC,v3,18/21] drm/ttm: Convert ttm vm to using drm_exec

Commit Message

Patch