[PATCH 12/16] drm/xe: Adjust to "drm/gpuvm: add common dma-resv per struct drm_gpuvm"
Rodrigo Vivi
rodrigo.vivi at intel.com
Thu Dec 7 19:08:24 UTC 2023
On Thu, Dec 07, 2023 at 03:11:52PM +0100, Thomas Hellström wrote:
This will be fixup on multiple places, but I will take care of those.
The end result looks clean and right
Reviewed-by: Rodrigo Vivi <rodrigo.vivi at intel.com>
> Signed-off-by: Thomas Hellström <thomas.hellstrom at linux.intel.com>
> ---
> drivers/gpu/drm/xe/xe_bo.c | 17 +++++---
> drivers/gpu/drm/xe/xe_bo.h | 11 ++++-
> drivers/gpu/drm/xe/xe_exec.c | 4 +-
> drivers/gpu/drm/xe/xe_migrate.c | 4 +-
> drivers/gpu/drm/xe/xe_pt.c | 6 +--
> drivers/gpu/drm/xe/xe_vm.c | 72 ++++++++++++++++----------------
> drivers/gpu/drm/xe/xe_vm.h | 21 ++++++++--
> drivers/gpu/drm/xe/xe_vm_types.h | 6 ---
> 8 files changed, 83 insertions(+), 58 deletions(-)
>
> diff --git a/drivers/gpu/drm/xe/xe_bo.c b/drivers/gpu/drm/xe/xe_bo.c
> index 72dc4a4eed4e..ad9d8793db3e 100644
> --- a/drivers/gpu/drm/xe/xe_bo.c
> +++ b/drivers/gpu/drm/xe/xe_bo.c
> @@ -519,9 +519,9 @@ static int xe_bo_trigger_rebind(struct xe_device *xe, struct xe_bo *bo,
> * that we indeed have it locked, put the vma an the
> * vm's notifier.rebind_list instead and scoop later.
> */
> - if (dma_resv_trylock(&vm->resv))
> + if (dma_resv_trylock(xe_vm_resv(vm)))
> vm_resv_locked = true;
> - else if (ctx->resv != &vm->resv) {
> + else if (ctx->resv != xe_vm_resv(vm)) {
> spin_lock(&vm->notifier.list_lock);
> if (!(vma->gpuva.flags & XE_VMA_DESTROYED))
> list_move_tail(&vma->notifier.rebind_link,
> @@ -538,7 +538,7 @@ static int xe_bo_trigger_rebind(struct xe_device *xe, struct xe_bo *bo,
> &vm->rebind_list);
>
> if (vm_resv_locked)
> - dma_resv_unlock(&vm->resv);
> + dma_resv_unlock(xe_vm_resv(vm));
> }
> }
>
> @@ -1398,7 +1398,7 @@ __xe_bo_create_locked(struct xe_device *xe,
> }
> }
>
> - bo = ___xe_bo_create_locked(xe, bo, tile, vm ? &vm->resv : NULL,
> + bo = ___xe_bo_create_locked(xe, bo, tile, vm ? xe_vm_resv(vm) : NULL,
> vm && !xe_vm_in_fault_mode(vm) &&
> flags & XE_BO_CREATE_USER_BIT ?
> &vm->lru_bulk_move : NULL, size,
> @@ -1406,6 +1406,13 @@ __xe_bo_create_locked(struct xe_device *xe,
> if (IS_ERR(bo))
> return bo;
>
> + /*
> + * Note that instead of taking a reference no the drm_gpuvm_resv_bo(),
> + * to ensure the shared resv doesn't disappear under the bo, the bo
> + * will keep a reference to the vm, and avoid circular references
> + * by having all the vm's bo refereferences released at vm close
> + * time.
> + */
> if (vm && xe_bo_is_user(bo))
> xe_vm_get(vm);
> bo->vm = vm;
> @@ -1772,7 +1779,7 @@ int xe_bo_validate(struct xe_bo *bo, struct xe_vm *vm, bool allow_res_evict)
> xe_vm_assert_held(vm);
>
> ctx.allow_res_evict = allow_res_evict;
> - ctx.resv = &vm->resv;
> + ctx.resv = xe_vm_resv(vm);
> }
>
> return ttm_bo_validate(&bo->ttm, &bo->placement, &ctx);
> diff --git a/drivers/gpu/drm/xe/xe_bo.h b/drivers/gpu/drm/xe/xe_bo.h
> index 098ccab7fa1e..9b1279aca127 100644
> --- a/drivers/gpu/drm/xe/xe_bo.h
> +++ b/drivers/gpu/drm/xe/xe_bo.h
> @@ -11,6 +11,15 @@
> #include "xe_bo_types.h"
> #include "xe_macros.h"
> #include "xe_vm_types.h"
> +#include "xe_vm.h"
> +
> +/**
> + * xe_vm_assert_held(vm) - Assert that the vm's reservation object is held.
> + * @vm: The vm
> + */
> +#define xe_vm_assert_held(vm) dma_resv_assert_held(xe_vm_resv(vm))
> +
> +
>
> #define XE_DEFAULT_GTT_SIZE_MB 3072ULL /* 3GB by default */
>
> @@ -168,7 +177,7 @@ void xe_bo_unlock(struct xe_bo *bo);
> static inline void xe_bo_unlock_vm_held(struct xe_bo *bo)
> {
> if (bo) {
> - XE_WARN_ON(bo->vm && bo->ttm.base.resv != &bo->vm->resv);
> + XE_WARN_ON(bo->vm && bo->ttm.base.resv != xe_vm_resv(bo->vm));
> if (bo->vm)
> xe_vm_assert_held(bo->vm);
> else
> diff --git a/drivers/gpu/drm/xe/xe_exec.c b/drivers/gpu/drm/xe/xe_exec.c
> index 347239f28170..5ec37df33afe 100644
> --- a/drivers/gpu/drm/xe/xe_exec.c
> +++ b/drivers/gpu/drm/xe/xe_exec.c
> @@ -281,7 +281,7 @@ int xe_exec_ioctl(struct drm_device *dev, void *data, struct drm_file *file)
> /* Wait behind munmap style rebinds */
> if (!xe_vm_in_lr_mode(vm)) {
> err = drm_sched_job_add_resv_dependencies(&job->drm,
> - &vm->resv,
> + xe_vm_resv(vm),
> DMA_RESV_USAGE_KERNEL);
> if (err)
> goto err_put_job;
> @@ -309,7 +309,7 @@ int xe_exec_ioctl(struct drm_device *dev, void *data, struct drm_file *file)
> xe_sched_job_arm(job);
> if (!xe_vm_in_lr_mode(vm)) {
> /* Block userptr invalidations / BO eviction */
> - dma_resv_add_fence(&vm->resv,
> + dma_resv_add_fence(xe_vm_resv(vm),
> &job->drm.s_fence->finished,
> DMA_RESV_USAGE_BOOKKEEP);
>
> diff --git a/drivers/gpu/drm/xe/xe_migrate.c b/drivers/gpu/drm/xe/xe_migrate.c
> index e8b567708ac0..a25697cdc2cc 100644
> --- a/drivers/gpu/drm/xe/xe_migrate.c
> +++ b/drivers/gpu/drm/xe/xe_migrate.c
> @@ -1136,7 +1136,7 @@ xe_migrate_update_pgtables_cpu(struct xe_migrate *m,
> DMA_RESV_USAGE_KERNEL))
> return ERR_PTR(-ETIME);
>
> - if (wait_vm && !dma_resv_test_signaled(&vm->resv,
> + if (wait_vm && !dma_resv_test_signaled(xe_vm_resv(vm),
> DMA_RESV_USAGE_BOOKKEEP))
> return ERR_PTR(-ETIME);
>
> @@ -1345,7 +1345,7 @@ xe_migrate_update_pgtables(struct xe_migrate *m,
> * trigger preempts before moving forward
> */
> if (first_munmap_rebind) {
> - err = job_add_deps(job, &vm->resv,
> + err = job_add_deps(job, xe_vm_resv(vm),
> DMA_RESV_USAGE_BOOKKEEP);
> if (err)
> goto err_job;
> diff --git a/drivers/gpu/drm/xe/xe_pt.c b/drivers/gpu/drm/xe/xe_pt.c
> index 35bd7940a571..3b485313804a 100644
> --- a/drivers/gpu/drm/xe/xe_pt.c
> +++ b/drivers/gpu/drm/xe/xe_pt.c
> @@ -866,7 +866,7 @@ static void xe_pt_commit_locks_assert(struct xe_vma *vma)
> else if (!xe_vma_is_null(vma))
> dma_resv_assert_held(xe_vma_bo(vma)->ttm.base.resv);
>
> - dma_resv_assert_held(&vm->resv);
> + xe_vm_assert_held(vm);
> }
>
> static void xe_pt_commit_bind(struct xe_vma *vma,
> @@ -1328,7 +1328,7 @@ __xe_pt_bind_vma(struct xe_tile *tile, struct xe_vma *vma, struct xe_exec_queue
> }
>
> /* add shared fence now for pagetable delayed destroy */
> - dma_resv_add_fence(&vm->resv, fence, !rebind &&
> + dma_resv_add_fence(xe_vm_resv(vm), fence, !rebind &&
> last_munmap_rebind ?
> DMA_RESV_USAGE_KERNEL :
> DMA_RESV_USAGE_BOOKKEEP);
> @@ -1665,7 +1665,7 @@ __xe_pt_unbind_vma(struct xe_tile *tile, struct xe_vma *vma, struct xe_exec_queu
> fence = &ifence->base.base;
>
> /* add shared fence now for pagetable delayed destroy */
> - dma_resv_add_fence(&vm->resv, fence,
> + dma_resv_add_fence(xe_vm_resv(vm), fence,
> DMA_RESV_USAGE_BOOKKEEP);
>
> /* This fence will be installed by caller when doing eviction */
> diff --git a/drivers/gpu/drm/xe/xe_vm.c b/drivers/gpu/drm/xe/xe_vm.c
> index e09050f16f07..9a090f21f9af 100644
> --- a/drivers/gpu/drm/xe/xe_vm.c
> +++ b/drivers/gpu/drm/xe/xe_vm.c
> @@ -39,6 +39,11 @@
>
> #define TEST_VM_ASYNC_OPS_ERROR
>
> +static struct drm_gem_object *xe_vm_obj(struct xe_vm *vm)
> +{
> + return vm->gpuvm.r_obj;
> +}
> +
> /**
> * xe_vma_userptr_check_repin() - Advisory check for repin needed
> * @vma: The userptr vma
> @@ -323,7 +328,7 @@ static void resume_and_reinstall_preempt_fences(struct xe_vm *vm)
> list_for_each_entry(q, &vm->preempt.exec_queues, compute.link) {
> q->ops->resume(q);
>
> - dma_resv_add_fence(&vm->resv, q->compute.pfence,
> + dma_resv_add_fence(xe_vm_resv(vm), q->compute.pfence,
> DMA_RESV_USAGE_BOOKKEEP);
> xe_vm_fence_all_extobjs(vm, q->compute.pfence,
> DMA_RESV_USAGE_BOOKKEEP);
> @@ -361,7 +366,7 @@ int xe_vm_add_compute_exec_queue(struct xe_vm *vm, struct xe_exec_queue *q)
>
> down_read(&vm->userptr.notifier_lock);
>
> - dma_resv_add_fence(&vm->resv, pfence,
> + dma_resv_add_fence(xe_vm_resv(vm), pfence,
> DMA_RESV_USAGE_BOOKKEEP);
>
> xe_vm_fence_all_extobjs(vm, pfence, DMA_RESV_USAGE_BOOKKEEP);
> @@ -447,8 +452,7 @@ int xe_vm_lock_dma_resv(struct xe_vm *vm, struct drm_exec *exec,
> lockdep_assert_held(&vm->lock);
>
> if (lock_vm) {
> - err = drm_exec_prepare_obj(exec, &xe_vm_ttm_bo(vm)->base,
> - num_shared);
> + err = drm_exec_prepare_obj(exec, xe_vm_obj(vm), num_shared);
> if (err)
> return err;
> }
> @@ -544,7 +548,7 @@ static int xe_preempt_work_begin(struct drm_exec *exec, struct xe_vm *vm,
> * 1 fence for each preempt fence plus a fence for each tile from a
> * possible rebind
> */
> - err = drm_exec_prepare_obj(exec, &xe_vm_ttm_bo(vm)->base,
> + err = drm_exec_prepare_obj(exec, xe_vm_obj(vm),
> vm->preempt.num_exec_queues +
> vm->xe->info.tile_count);
> if (err)
> @@ -643,7 +647,7 @@ static void preempt_rebind_work_func(struct work_struct *w)
> }
>
> /* Wait on munmap style VM unbinds */
> - wait = dma_resv_wait_timeout(&vm->resv,
> + wait = dma_resv_wait_timeout(xe_vm_resv(vm),
> DMA_RESV_USAGE_KERNEL,
> false, MAX_SCHEDULE_TIMEOUT);
> if (wait <= 0) {
> @@ -738,13 +742,13 @@ static bool vma_userptr_invalidate(struct mmu_interval_notifier *mni,
> * unbinds to complete, and those are attached as BOOKMARK fences
> * to the vm.
> */
> - dma_resv_iter_begin(&cursor, &vm->resv,
> + dma_resv_iter_begin(&cursor, xe_vm_resv(vm),
> DMA_RESV_USAGE_BOOKKEEP);
> dma_resv_for_each_fence_unlocked(&cursor, fence)
> dma_fence_enable_sw_signaling(fence);
> dma_resv_iter_end(&cursor);
>
> - err = dma_resv_wait_timeout(&vm->resv,
> + err = dma_resv_wait_timeout(xe_vm_resv(vm),
> DMA_RESV_USAGE_BOOKKEEP,
> false, MAX_SCHEDULE_TIMEOUT);
> XE_WARN_ON(err <= 0);
> @@ -793,14 +797,14 @@ int xe_vm_userptr_pin(struct xe_vm *vm)
> }
>
> /* Take lock and move to rebind_list for rebinding. */
> - err = dma_resv_lock_interruptible(&vm->resv, NULL);
> + err = dma_resv_lock_interruptible(xe_vm_resv(vm), NULL);
> if (err)
> goto out_err;
>
> list_for_each_entry_safe(vma, next, &tmp_evict, combined_links.userptr)
> list_move_tail(&vma->combined_links.rebind, &vm->rebind_list);
>
> - dma_resv_unlock(&vm->resv);
> + dma_resv_unlock(xe_vm_resv(vm));
>
> return 0;
>
> @@ -1116,7 +1120,7 @@ int xe_vm_prepare_vma(struct drm_exec *exec, struct xe_vma *vma,
> int err;
>
> XE_WARN_ON(!vm);
> - err = drm_exec_prepare_obj(exec, &xe_vm_ttm_bo(vm)->base, num_shared);
> + err = drm_exec_prepare_obj(exec, xe_vm_obj(vm), num_shared);
> if (!err && bo && !bo->vm)
> err = drm_exec_prepare_obj(exec, &bo->ttm.base, num_shared);
>
> @@ -1331,6 +1335,7 @@ static void vm_destroy_work_func(struct work_struct *w);
>
> struct xe_vm *xe_vm_create(struct xe_device *xe, u32 flags)
> {
> + struct drm_gem_object *vm_resv_obj;
> struct xe_vm *vm;
> int err, i = 0, number_tiles = 0;
> struct xe_tile *tile;
> @@ -1342,7 +1347,6 @@ struct xe_vm *xe_vm_create(struct xe_device *xe, u32 flags)
>
> vm->xe = xe;
> kref_init(&vm->refcount);
> - dma_resv_init(&vm->resv);
>
> vm->size = 1ull << xe->info.va_bits;
>
> @@ -1375,12 +1379,21 @@ struct xe_vm *xe_vm_create(struct xe_device *xe, u32 flags)
> if (!(flags & XE_VM_FLAG_MIGRATION))
> xe_device_mem_access_get(xe);
>
> - err = dma_resv_lock_interruptible(&vm->resv, NULL);
> + vm_resv_obj = drm_gpuvm_resv_object_alloc(&xe->drm);
> + if (!vm_resv_obj) {
> + err = -ENOMEM;
> + goto err_no_resv;
> + }
> +
> + drm_gpuvm_init(&vm->gpuvm, "Xe VM", &xe->drm, vm_resv_obj, 0, vm->size,
> + 0, 0, &gpuvm_ops);
> +
> + drm_gem_object_put(vm_resv_obj);
> +
> + err = dma_resv_lock_interruptible(xe_vm_resv(vm), NULL);
> if (err)
> goto err_put;
>
> - drm_gpuvm_init(&vm->gpuvm, "Xe VM", 0, vm->size, 0, 0,
> - &gpuvm_ops);
> if (IS_DGFX(xe) && xe->info.vram_flags & XE_VRAM_FLAGS_NEED64K)
> vm->flags |= XE_VM_FLAG_64K;
>
> @@ -1422,7 +1435,7 @@ struct xe_vm *xe_vm_create(struct xe_device *xe, u32 flags)
>
> xe_pt_populate_empty(tile, vm, vm->pt_root[id]);
> }
> - dma_resv_unlock(&vm->resv);
> + dma_resv_unlock(xe_vm_resv(vm));
>
> /* Kernel migration VM shouldn't have a circular loop.. */
> if (!(flags & XE_VM_FLAG_MIGRATION)) {
> @@ -1483,10 +1496,10 @@ struct xe_vm *xe_vm_create(struct xe_device *xe, u32 flags)
> if (vm->pt_root[id])
> xe_pt_destroy(vm->pt_root[id], vm->flags, NULL);
> }
> - dma_resv_unlock(&vm->resv);
> - drm_gpuvm_destroy(&vm->gpuvm);
> + dma_resv_unlock(xe_vm_resv(vm));
> err_put:
> - dma_resv_fini(&vm->resv);
> + drm_gpuvm_destroy(&vm->gpuvm);
> +err_no_resv:
> for_each_tile(tile, xe, id)
> xe_range_fence_tree_fini(&vm->rftree[id]);
> kfree(vm);
> @@ -1590,8 +1603,6 @@ void xe_vm_close_and_put(struct xe_vm *vm)
> xe_assert(xe, list_empty(&vm->extobj.list));
> up_write(&vm->lock);
>
> - drm_gpuvm_destroy(&vm->gpuvm);
> -
> mutex_lock(&xe->usm.lock);
> if (vm->flags & XE_VM_FLAG_FAULT_MODE)
> xe->usm.num_vm_in_fault_mode--;
> @@ -1644,7 +1655,7 @@ static void vm_destroy_work_func(struct work_struct *w)
>
> trace_xe_vm_free(vm);
> dma_fence_put(vm->rebind_fence);
> - dma_resv_fini(&vm->resv);
> + drm_gpuvm_destroy(&vm->gpuvm);
> kfree(vm);
> }
>
> @@ -2092,15 +2103,6 @@ static int xe_vm_prefetch(struct xe_vm *vm, struct xe_vma *vma,
> }
> }
>
> -struct ttm_buffer_object *xe_vm_ttm_bo(struct xe_vm *vm)
> -{
> - int idx = vm->flags & XE_VM_FLAG_MIGRATION ?
> - XE_VM_FLAG_TILE_ID(vm->flags) : 0;
> -
> - /* Safe to use index 0 as all BO in the VM share a single dma-resv lock */
> - return &vm->pt_root[idx]->bo->ttm;
> -}
> -
> static void prep_vma_destroy(struct xe_vm *vm, struct xe_vma *vma,
> bool post_commit)
> {
> @@ -3205,9 +3207,9 @@ int xe_vm_bind_ioctl(struct drm_device *dev, void *data, struct drm_file *file)
> int xe_vm_lock(struct xe_vm *vm, bool intr)
> {
> if (intr)
> - return dma_resv_lock_interruptible(&vm->resv, NULL);
> + return dma_resv_lock_interruptible(xe_vm_resv(vm), NULL);
>
> - return dma_resv_lock(&vm->resv, NULL);
> + return dma_resv_lock(xe_vm_resv(vm), NULL);
> }
>
> /**
> @@ -3218,7 +3220,7 @@ int xe_vm_lock(struct xe_vm *vm, bool intr)
> */
> void xe_vm_unlock(struct xe_vm *vm)
> {
> - dma_resv_unlock(&vm->resv);
> + dma_resv_unlock(xe_vm_resv(vm));
> }
>
> /**
> @@ -3250,7 +3252,7 @@ int xe_vm_invalidate_vma(struct xe_vma *vma)
> WARN_ON_ONCE(!mmu_interval_check_retry
> (&vma->userptr.notifier,
> vma->userptr.notifier_seq));
> - WARN_ON_ONCE(!dma_resv_test_signaled(&xe_vma_vm(vma)->resv,
> + WARN_ON_ONCE(!dma_resv_test_signaled(xe_vm_resv(xe_vma_vm(vma)),
> DMA_RESV_USAGE_BOOKKEEP));
>
> } else {
> diff --git a/drivers/gpu/drm/xe/xe_vm.h b/drivers/gpu/drm/xe/xe_vm.h
> index 9a0ae19c47b7..e4b5cb8a0f08 100644
> --- a/drivers/gpu/drm/xe/xe_vm.h
> +++ b/drivers/gpu/drm/xe/xe_vm.h
> @@ -139,8 +139,6 @@ static inline bool xe_vma_is_userptr(struct xe_vma *vma)
> return xe_vma_has_no_bo(vma) && !xe_vma_is_null(vma);
> }
>
> -#define xe_vm_assert_held(vm) dma_resv_assert_held(&(vm)->resv)
> -
> u64 xe_vm_pdp4_descriptor(struct xe_vm *vm, struct xe_tile *tile);
>
> int xe_vm_create_ioctl(struct drm_device *dev, void *data,
> @@ -182,8 +180,6 @@ int xe_vm_invalidate_vma(struct xe_vma *vma);
>
> extern struct ttm_device_funcs xe_ttm_funcs;
>
> -struct ttm_buffer_object *xe_vm_ttm_bo(struct xe_vm *vm);
> -
> static inline void xe_vm_queue_rebind_worker(struct xe_vm *vm)
> {
> xe_assert(vm->xe, xe_vm_in_preempt_fence_mode(vm));
> @@ -224,6 +220,23 @@ int xe_analyze_vm(struct drm_printer *p, struct xe_vm *vm, int gt_id);
> int xe_vm_prepare_vma(struct drm_exec *exec, struct xe_vma *vma,
> unsigned int num_shared);
>
> +/**
> + * xe_vm_resv() - Return's the vm's reservation object
> + * @vm: The vm
> + *
> + * Return: Pointer to the vm's reservation object.
> + */
> +static inline struct dma_resv *xe_vm_resv(struct xe_vm *vm)
> +{
> + return drm_gpuvm_resv(&vm->gpuvm);
> +}
> +
> +/**
> + * xe_vm_assert_held(vm) - Assert that the vm's reservation object is held.
> + * @vm: The vm
> + */
> +#define xe_vm_assert_held(vm) dma_resv_assert_held(xe_vm_resv(vm))
> +
> #if IS_ENABLED(CONFIG_DRM_XE_DEBUG_VM)
> #define vm_dbg drm_dbg
> #else
> diff --git a/drivers/gpu/drm/xe/xe_vm_types.h b/drivers/gpu/drm/xe/xe_vm_types.h
> index 23abdfd8622f..4e540d013702 100644
> --- a/drivers/gpu/drm/xe/xe_vm_types.h
> +++ b/drivers/gpu/drm/xe/xe_vm_types.h
> @@ -136,8 +136,6 @@ struct xe_vma {
>
> struct xe_device;
>
> -#define xe_vm_assert_held(vm) dma_resv_assert_held(&(vm)->resv)
> -
> struct xe_vm {
> /** @gpuvm: base GPUVM used to track VMAs */
> struct drm_gpuvm gpuvm;
> @@ -149,9 +147,6 @@ struct xe_vm {
> /* exec queue used for (un)binding vma's */
> struct xe_exec_queue *q[XE_MAX_TILES_PER_DEVICE];
>
> - /** Protects @rebind_list and the page-table structures */
> - struct dma_resv resv;
> -
> /** @lru_bulk_move: Bulk LRU move list for this VM's BOs */
> struct ttm_lru_bulk_move lru_bulk_move;
>
> @@ -424,5 +419,4 @@ struct xe_vma_op {
> struct xe_vma_op_prefetch prefetch;
> };
> };
> -
> #endif
> --
> 2.42.0
>
More information about the Intel-xe
mailing list