[PATCH 1/2] drm/xe: Use DRM GPUVM helpers for external- and evicted objects
Matthew Brost
matthew.brost at intel.com
Mon Dec 11 23:08:17 UTC 2023
On Sat, Dec 09, 2023 at 03:49:16PM +0100, Thomas Hellström wrote:
> Adapt to the DRM_GPUVM helpers moving removing a lot of complicated
> driver-specific code.
>
> For now this uses fine-grained locking for the evict list and external
> object list, which may incur a slight performance penalty in some
> situations.
>
> Signed-off-by: Thomas Hellström <thomas.hellstrom at linux.intel.com>
> ---
> drivers/gpu/drm/xe/xe_bo.c | 63 +++----
> drivers/gpu/drm/xe/xe_exec.c | 72 ++------
> drivers/gpu/drm/xe/xe_vm.c | 292 +++++++------------------------
> drivers/gpu/drm/xe/xe_vm.h | 13 +-
> drivers/gpu/drm/xe/xe_vm_types.h | 67 ++-----
> 5 files changed, 120 insertions(+), 387 deletions(-)
>
> diff --git a/drivers/gpu/drm/xe/xe_bo.c b/drivers/gpu/drm/xe/xe_bo.c
> index 7e25c8b7a01a..bd79b322e6ac 100644
> --- a/drivers/gpu/drm/xe/xe_bo.c
> +++ b/drivers/gpu/drm/xe/xe_bo.c
> @@ -468,9 +468,9 @@ static int xe_bo_trigger_rebind(struct xe_device *xe, struct xe_bo *bo,
> {
> struct dma_resv_iter cursor;
> struct dma_fence *fence;
> - struct drm_gpuva *gpuva;
> struct drm_gem_object *obj = &bo->ttm.base;
> struct drm_gpuvm_bo *vm_bo;
> + bool idle = false;
> int ret = 0;
>
> dma_resv_assert_held(bo->ttm.base.resv);
> @@ -484,14 +484,15 @@ static int xe_bo_trigger_rebind(struct xe_device *xe, struct xe_bo *bo,
> }
>
> drm_gem_for_each_gpuvm_bo(vm_bo, obj) {
> - drm_gpuvm_bo_for_each_va(gpuva, vm_bo) {
> - struct xe_vma *vma = gpuva_to_vma(gpuva);
> - struct xe_vm *vm = xe_vma_vm(vma);
> + struct xe_vm *vm = gpuvm_to_vm(vm_bo->vm);
> + struct drm_gpuva *gpuva;
>
> - trace_xe_vma_evict(vma);
> + if (!xe_vm_in_fault_mode(vm)) {
> + drm_gpuvm_bo_evict(vm_bo, true);
> + continue;
> + }
>
> - if (xe_vm_in_fault_mode(vm)) {
> - /* Wait for pending binds / unbinds. */
> + if (!idle) {
> long timeout;
>
> if (ctx->no_wait_gpu &&
> @@ -503,45 +504,21 @@ static int xe_bo_trigger_rebind(struct xe_device *xe, struct xe_bo *bo,
> DMA_RESV_USAGE_BOOKKEEP,
> ctx->interruptible,
> MAX_SCHEDULE_TIMEOUT);
> - if (timeout > 0) {
> - ret = xe_vm_invalidate_vma(vma);
> - XE_WARN_ON(ret);
> - } else if (!timeout) {
> - ret = -ETIME;
> - } else {
> - ret = timeout;
> - }
> -
> - } else {
> - bool vm_resv_locked = false;
> + if (!timeout)
> + return -ETIME;
> + if (timeout < 0)
> + return timeout;
>
> - /*
> - * We need to put the vma on the vm's rebind_list,
> - * but need the vm resv to do so. If we can't verify
> - * that we indeed have it locked, put the vma an the
> - * vm's notifier.rebind_list instead and scoop later.
> - */
> - if (dma_resv_trylock(xe_vm_resv(vm)))
> - vm_resv_locked = true;
> - 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,
> - &vm->notifier.rebind_list);
> - spin_unlock(&vm->notifier.list_lock);
> - continue;
> - }
> + idle = true;
> + }
>
> - xe_vm_assert_held(vm);
> - if (vma->tile_present &&
> - !(vma->gpuva.flags & XE_VMA_DESTROYED) &&
> - list_empty(&vma->combined_links.rebind))
> - list_add_tail(&vma->combined_links.rebind,
> - &vm->rebind_list);
> + drm_gpuvm_bo_for_each_va(gpuva, vm_bo) {
> + struct xe_vma *vma = gpuva_to_vma(gpuva);
>
> - if (vm_resv_locked)
> - dma_resv_unlock(xe_vm_resv(vm));
> - }
> + trace_xe_vma_evict(vma);
> + ret = xe_vm_invalidate_vma(vma);
> + if (XE_WARN_ON(ret))
> + return ret;
> }
> }
>
> diff --git a/drivers/gpu/drm/xe/xe_exec.c b/drivers/gpu/drm/xe/xe_exec.c
> index 5ec37df33afe..1a4abfedd6f8 100644
> --- a/drivers/gpu/drm/xe/xe_exec.c
> +++ b/drivers/gpu/drm/xe/xe_exec.c
> @@ -94,40 +94,9 @@
> * Unlock all
> */
>
> -static int xe_exec_begin(struct drm_exec *exec, struct xe_vm *vm)
> +static int xe_exec_fn(struct drm_gpuvm_exec *vm_exec)
> {
> - struct xe_vma *vma;
> - LIST_HEAD(dups);
> - int err = 0;
> -
> - if (xe_vm_in_lr_mode(vm))
> - return 0;
> -
This is a change of behavior in this patch (i.e. now in LR mode during
execs the dma-resv locks are taken BOs validated). Is this intentional?
If so, what is the reasoning?
Matt
> - /*
> - * 1 fence for job from exec plus a fence for each tile from a possible
> - * rebind
> - */
> - err = xe_vm_lock_dma_resv(vm, exec, 1 + vm->xe->info.tile_count, true);
> - if (err)
> - return err;
> -
> - /*
> - * Validate BOs that have been evicted (i.e. make sure the
> - * BOs have valid placements possibly moving an evicted BO back
> - * to a location where the GPU can access it).
> - */
> - list_for_each_entry(vma, &vm->rebind_list, combined_links.rebind) {
> - xe_assert(vm->xe, !xe_vma_is_null(vma));
> -
> - if (xe_vma_is_userptr(vma))
> - continue;
> -
> - err = xe_bo_validate(xe_vma_bo(vma), vm, false);
> - if (err)
> - break;
> - }
> -
> - return err;
> + return drm_gpuvm_validate(vm_exec->vm, &vm_exec->exec);
> }
>
> int xe_exec_ioctl(struct drm_device *dev, void *data, struct drm_file *file)
> @@ -140,7 +109,8 @@ int xe_exec_ioctl(struct drm_device *dev, void *data, struct drm_file *file)
> struct xe_exec_queue *q;
> struct xe_sync_entry *syncs = NULL;
> u64 addresses[XE_HW_ENGINE_MAX_INSTANCE];
> - struct drm_exec exec;
> + struct drm_gpuvm_exec vm_exec = {.extra.fn = xe_exec_fn};
> + struct drm_exec *exec = &vm_exec.exec;
> u32 i, num_syncs = 0;
> struct xe_sched_job *job;
> struct dma_fence *rebind_fence;
> @@ -216,16 +186,14 @@ int xe_exec_ioctl(struct drm_device *dev, void *data, struct drm_file *file)
> goto err_unlock_list;
> }
>
> - drm_exec_init(&exec, DRM_EXEC_INTERRUPTIBLE_WAIT);
> - drm_exec_until_all_locked(&exec) {
> - err = xe_exec_begin(&exec, vm);
> - drm_exec_retry_on_contention(&exec);
> - if (err && xe_vm_validate_should_retry(&exec, err, &end)) {
> + vm_exec.vm = &vm->gpuvm;
> + vm_exec.num_fences = 1 + vm->xe->info.tile_count;
> + vm_exec.flags = DRM_EXEC_INTERRUPTIBLE_WAIT;
> + err = drm_gpuvm_exec_lock(&vm_exec);
> + if (err) {
> + if (xe_vm_validate_should_retry(exec, err, &end))
> err = -EAGAIN;
> - goto err_unlock_list;
> - }
> - if (err)
> - goto err_exec;
> + goto err_unlock_list;
> }
>
> if (xe_vm_is_closed_or_banned(q->vm)) {
> @@ -307,19 +275,9 @@ int xe_exec_ioctl(struct drm_device *dev, void *data, struct drm_file *file)
> * the job and let the DRM scheduler / backend clean up the job.
> */
> xe_sched_job_arm(job);
> - if (!xe_vm_in_lr_mode(vm)) {
> - /* Block userptr invalidations / BO eviction */
> - dma_resv_add_fence(xe_vm_resv(vm),
> - &job->drm.s_fence->finished,
> - DMA_RESV_USAGE_BOOKKEEP);
> -
> - /*
> - * Make implicit sync work across drivers, assuming all external
> - * BOs are written as we don't pass in a read / write list.
> - */
> - xe_vm_fence_all_extobjs(vm, &job->drm.s_fence->finished,
> - DMA_RESV_USAGE_WRITE);
> - }
> + if (!xe_vm_in_lr_mode(vm))
> + drm_gpuvm_resv_add_fence(&vm->gpuvm, exec, &job->drm.s_fence->finished,
> + DMA_RESV_USAGE_BOOKKEEP, DMA_RESV_USAGE_WRITE);
>
> for (i = 0; i < num_syncs; i++)
> xe_sync_entry_signal(&syncs[i], job,
> @@ -343,7 +301,7 @@ int xe_exec_ioctl(struct drm_device *dev, void *data, struct drm_file *file)
> if (err)
> xe_sched_job_put(job);
> err_exec:
> - drm_exec_fini(&exec);
> + drm_exec_fini(exec);
> err_unlock_list:
> if (write_locked)
> up_write(&vm->lock);
> diff --git a/drivers/gpu/drm/xe/xe_vm.c b/drivers/gpu/drm/xe/xe_vm.c
> index 11667529e40b..2c35395ff5d4 100644
> --- a/drivers/gpu/drm/xe/xe_vm.c
> +++ b/drivers/gpu/drm/xe/xe_vm.c
> @@ -299,26 +299,8 @@ static int add_preempt_fences(struct xe_vm *vm, struct xe_bo *bo)
> return err;
> }
>
> -/**
> - * xe_vm_fence_all_extobjs() - Add a fence to vm's external objects' resv
> - * @vm: The vm.
> - * @fence: The fence to add.
> - * @usage: The resv usage for the fence.
> - *
> - * Loops over all of the vm's external object bindings and adds a @fence
> - * with the given @usage to all of the external object's reservation
> - * objects.
> - */
> -void xe_vm_fence_all_extobjs(struct xe_vm *vm, struct dma_fence *fence,
> - enum dma_resv_usage usage)
> -{
> - struct xe_vma *vma;
> -
> - list_for_each_entry(vma, &vm->extobj.list, extobj.link)
> - dma_resv_add_fence(xe_vma_bo(vma)->ttm.base.resv, fence, usage);
> -}
> -
> -static void resume_and_reinstall_preempt_fences(struct xe_vm *vm)
> +static void resume_and_reinstall_preempt_fences(struct xe_vm *vm,
> + struct drm_exec *exec)
> {
> struct xe_exec_queue *q;
>
> @@ -328,16 +310,19 @@ 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(xe_vm_resv(vm), q->compute.pfence,
> - DMA_RESV_USAGE_BOOKKEEP);
> - xe_vm_fence_all_extobjs(vm, q->compute.pfence,
> - DMA_RESV_USAGE_BOOKKEEP);
> + drm_gpuvm_resv_add_fence(&vm->gpuvm, exec, q->compute.pfence,
> + DMA_RESV_USAGE_BOOKKEEP, DMA_RESV_USAGE_BOOKKEEP);
> }
> }
>
> int xe_vm_add_compute_exec_queue(struct xe_vm *vm, struct xe_exec_queue *q)
> {
> - struct drm_exec exec;
> + struct drm_gpuvm_exec vm_exec = {
> + .vm = &vm->gpuvm,
> + .flags = DRM_EXEC_INTERRUPTIBLE_WAIT,
> + .num_fences = 1,
> + };
> + struct drm_exec *exec = &vm_exec.exec;
> struct dma_fence *pfence;
> int err;
> bool wait;
> @@ -345,13 +330,9 @@ int xe_vm_add_compute_exec_queue(struct xe_vm *vm, struct xe_exec_queue *q)
> xe_assert(vm->xe, xe_vm_in_preempt_fence_mode(vm));
>
> down_write(&vm->lock);
> - drm_exec_init(&exec, DRM_EXEC_INTERRUPTIBLE_WAIT);
> - drm_exec_until_all_locked(&exec) {
> - err = xe_vm_lock_dma_resv(vm, &exec, 1, true);
> - drm_exec_retry_on_contention(&exec);
> - if (err)
> - goto out_unlock;
> - }
> + err = drm_gpuvm_exec_lock(&vm_exec);
> + if (err)
> + return err;
>
> pfence = xe_preempt_fence_create(q, q->compute.context,
> ++q->compute.seqno);
> @@ -366,10 +347,8 @@ 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(xe_vm_resv(vm), pfence,
> - DMA_RESV_USAGE_BOOKKEEP);
> -
> - xe_vm_fence_all_extobjs(vm, pfence, DMA_RESV_USAGE_BOOKKEEP);
> + drm_gpuvm_resv_add_fence(&vm->gpuvm, exec, pfence,
> + DMA_RESV_USAGE_BOOKKEEP, DMA_RESV_USAGE_BOOKKEEP);
>
> /*
> * Check to see if a preemption on VM is in flight or userptr
> @@ -383,7 +362,7 @@ int xe_vm_add_compute_exec_queue(struct xe_vm *vm, struct xe_exec_queue *q)
> up_read(&vm->userptr.notifier_lock);
>
> out_unlock:
> - drm_exec_fini(&exec);
> + drm_exec_fini(exec);
> up_write(&vm->lock);
>
> return err;
> @@ -429,55 +408,6 @@ int __xe_vm_userptr_needs_repin(struct xe_vm *vm)
> list_empty(&vm->userptr.invalidated)) ? 0 : -EAGAIN;
> }
>
> -/**
> - * xe_vm_lock_dma_resv() - Lock the vm dma_resv object and the dma_resv
> - * objects of the vm's external buffer objects.
> - * @vm: The vm.
> - * @exec: Pointer to a struct drm_exec locking context.
> - * @num_shared: Number of dma-fence slots to reserve in the locked objects.
> - * @lock_vm: Lock also the vm's dma_resv.
> - *
> - * Locks the vm dma-resv objects and all the dma-resv objects of the
> - * buffer objects on the vm external object list.
> - *
> - * Return: 0 on success, Negative error code on error. In particular if
> - * @intr is set to true, -EINTR or -ERESTARTSYS may be returned.
> - */
> -int xe_vm_lock_dma_resv(struct xe_vm *vm, struct drm_exec *exec,
> - unsigned int num_shared, bool lock_vm)
> -{
> - struct xe_vma *vma, *next;
> - int err = 0;
> -
> - lockdep_assert_held(&vm->lock);
> -
> - if (lock_vm) {
> - err = drm_exec_prepare_obj(exec, xe_vm_obj(vm), num_shared);
> - if (err)
> - return err;
> - }
> -
> - list_for_each_entry(vma, &vm->extobj.list, extobj.link) {
> - err = drm_exec_prepare_obj(exec, &xe_vma_bo(vma)->ttm.base, num_shared);
> - if (err)
> - return err;
> - }
> -
> - spin_lock(&vm->notifier.list_lock);
> - list_for_each_entry_safe(vma, next, &vm->notifier.rebind_list,
> - notifier.rebind_link) {
> - xe_bo_assert_held(xe_vma_bo(vma));
> -
> - list_del_init(&vma->notifier.rebind_link);
> - if (vma->tile_present && !(vma->gpuva.flags & XE_VMA_DESTROYED))
> - list_move_tail(&vma->combined_links.rebind,
> - &vm->rebind_list);
> - }
> - spin_unlock(&vm->notifier.list_lock);
> -
> - return 0;
> -}
> -
> #define XE_VM_REBIND_RETRY_TIMEOUT_MS 1000
>
> static void xe_vm_kill(struct xe_vm *vm)
> @@ -526,30 +456,39 @@ bool xe_vm_validate_should_retry(struct drm_exec *exec, int err, ktime_t *end)
> if (!ktime_before(cur, *end))
> return false;
>
> - /*
> - * We would like to keep the ticket here with
> - * drm_exec_unlock_all(), but WW mutex asserts currently
> - * stop us from that. In any case this function could go away
> - * with proper TTM -EDEADLK handling.
> - */
> - drm_exec_fini(exec);
> -
> msleep(20);
> return true;
> }
>
> +static int xe_gpuvm_validate(struct drm_gpuvm_bo *vm_bo, struct drm_exec *exec)
> +{
> + struct xe_vm *vm = gpuvm_to_vm(vm_bo->vm);
> + struct drm_gpuva *gpuva;
> + int ret;
> +
> + lockdep_assert_held(&vm->lock);
> + drm_gpuvm_bo_for_each_va(gpuva, vm_bo)
> + list_move_tail(&gpuva_to_vma(gpuva)->combined_links.rebind,
> + &vm->rebind_list);
> +
> + ret = xe_bo_validate(gem_to_xe_bo(vm_bo->obj), vm, false);
> + if (ret)
> + return ret;
> +
> + vm_bo->evicted = false;
> + return 0;
> +}
> +
> static int xe_preempt_work_begin(struct drm_exec *exec, struct xe_vm *vm,
> bool *done)
> {
> - struct xe_vma *vma;
> int err;
>
> /*
> * 1 fence for each preempt fence plus a fence for each tile from a
> * possible rebind
> */
> - err = drm_exec_prepare_obj(exec, xe_vm_obj(vm),
> - vm->preempt.num_exec_queues +
> + err = drm_gpuvm_prepare_vm(&vm->gpuvm, exec, vm->preempt.num_exec_queues +
> vm->xe->info.tile_count);
> if (err)
> return err;
> @@ -565,7 +504,7 @@ static int xe_preempt_work_begin(struct drm_exec *exec, struct xe_vm *vm,
> return 0;
> }
>
> - err = xe_vm_lock_dma_resv(vm, exec, vm->preempt.num_exec_queues, false);
> + err = drm_gpuvm_prepare_objects(&vm->gpuvm, exec, vm->preempt.num_exec_queues);
> if (err)
> return err;
>
> @@ -573,17 +512,7 @@ static int xe_preempt_work_begin(struct drm_exec *exec, struct xe_vm *vm,
> if (err)
> return err;
>
> - list_for_each_entry(vma, &vm->rebind_list, combined_links.rebind) {
> - if (xe_vma_has_no_bo(vma) ||
> - vma->gpuva.flags & XE_VMA_DESTROYED)
> - continue;
> -
> - err = xe_bo_validate(xe_vma_bo(vma), vm, false);
> - if (err)
> - break;
> - }
> -
> - return err;
> + return drm_gpuvm_validate(&vm->gpuvm, exec);
> }
>
> static void preempt_rebind_work_func(struct work_struct *w)
> @@ -623,12 +552,13 @@ static void preempt_rebind_work_func(struct work_struct *w)
>
> err = xe_preempt_work_begin(&exec, vm, &done);
> drm_exec_retry_on_contention(&exec);
> - if (err && xe_vm_validate_should_retry(&exec, err, &end)) {
> - err = -EAGAIN;
> + if (err || done) {
> + drm_exec_fini(&exec);
> + if (err && xe_vm_validate_should_retry(&exec, err, &end))
> + err = -EAGAIN;
> +
> goto out_unlock_outer;
> }
> - if (err || done)
> - goto out_unlock;
> }
>
> err = alloc_preempt_fences(vm, &preempt_fences, &fence_count);
> @@ -675,7 +605,7 @@ static void preempt_rebind_work_func(struct work_struct *w)
>
> /* Point of no return. */
> arm_preempt_fences(vm, &preempt_fences);
> - resume_and_reinstall_preempt_fences(vm);
> + resume_and_reinstall_preempt_fences(vm, &exec);
> up_read(&vm->userptr.notifier_lock);
>
> out_unlock:
> @@ -780,9 +710,8 @@ int xe_vm_userptr_pin(struct xe_vm *vm)
> list_for_each_entry_safe(vma, next, &vm->userptr.invalidated,
> userptr.invalidate_link) {
> list_del_init(&vma->userptr.invalidate_link);
> - if (list_empty(&vma->combined_links.userptr))
> - list_move_tail(&vma->combined_links.userptr,
> - &vm->userptr.repin_list);
> + list_move_tail(&vma->combined_links.userptr,
> + &vm->userptr.repin_list);
> }
> spin_unlock(&vm->userptr.invalidated_lock);
>
> @@ -791,27 +720,12 @@ int xe_vm_userptr_pin(struct xe_vm *vm)
> combined_links.userptr) {
> err = xe_vma_userptr_pin_pages(vma);
> if (err < 0)
> - goto out_err;
> + return err;
>
> - list_move_tail(&vma->combined_links.userptr, &tmp_evict);
> + list_move_tail(&vma->combined_links.userptr, &vm->rebind_list);
> }
>
> - /* Take lock and move to rebind_list for rebinding. */
> - 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(xe_vm_resv(vm));
> -
> return 0;
> -
> -out_err:
> - list_splice_tail(&tmp_evict, &vm->userptr.repin_list);
> -
> - return err;
> }
>
> /**
> @@ -890,8 +804,6 @@ static struct xe_vma *xe_vma_create(struct xe_vm *vm,
> }
>
> INIT_LIST_HEAD(&vma->combined_links.rebind);
> - INIT_LIST_HEAD(&vma->notifier.rebind_link);
> - INIT_LIST_HEAD(&vma->extobj.link);
>
> INIT_LIST_HEAD(&vma->gpuva.gem.entry);
> vma->gpuva.vm = &vm->gpuvm;
> @@ -921,6 +833,7 @@ static struct xe_vma *xe_vma_create(struct xe_vm *vm,
> return ERR_CAST(vm_bo);
> }
>
> + drm_gpuvm_bo_extobj_add(vm_bo);
> drm_gem_object_get(&bo->ttm.base);
> vma->gpuva.gem.obj = &bo->ttm.base;
> vma->gpuva.gem.offset = bo_offset_or_userptr;
> @@ -953,16 +866,6 @@ static struct xe_vma *xe_vma_create(struct xe_vm *vm,
> return vma;
> }
>
> -static bool vm_remove_extobj(struct xe_vma *vma)
> -{
> - if (!list_empty(&vma->extobj.link)) {
> - xe_vma_vm(vma)->extobj.entries--;
> - list_del_init(&vma->extobj.link);
> - return true;
> - }
> - return false;
> -}
> -
> static void xe_vma_destroy_late(struct xe_vma *vma)
> {
> struct xe_vm *vm = xe_vma_vm(vma);
> @@ -1003,60 +906,6 @@ static void vma_destroy_work_func(struct work_struct *w)
> xe_vma_destroy_late(vma);
> }
>
> -static struct xe_vma *
> -bo_has_vm_references_locked(struct xe_bo *bo, struct xe_vm *vm,
> - struct xe_vma *ignore)
> -{
> - struct drm_gpuvm_bo *vm_bo;
> - struct drm_gpuva *va;
> - struct drm_gem_object *obj = &bo->ttm.base;
> -
> - xe_bo_assert_held(bo);
> -
> - drm_gem_for_each_gpuvm_bo(vm_bo, obj) {
> - drm_gpuvm_bo_for_each_va(va, vm_bo) {
> - struct xe_vma *vma = gpuva_to_vma(va);
> -
> - if (vma != ignore && xe_vma_vm(vma) == vm)
> - return vma;
> - }
> - }
> -
> - return NULL;
> -}
> -
> -static bool bo_has_vm_references(struct xe_bo *bo, struct xe_vm *vm,
> - struct xe_vma *ignore)
> -{
> - bool ret;
> -
> - xe_bo_lock(bo, false);
> - ret = !!bo_has_vm_references_locked(bo, vm, ignore);
> - xe_bo_unlock(bo);
> -
> - return ret;
> -}
> -
> -static void __vm_insert_extobj(struct xe_vm *vm, struct xe_vma *vma)
> -{
> - lockdep_assert_held_write(&vm->lock);
> -
> - list_add(&vma->extobj.link, &vm->extobj.list);
> - vm->extobj.entries++;
> -}
> -
> -static void vm_insert_extobj(struct xe_vm *vm, struct xe_vma *vma)
> -{
> - struct xe_bo *bo = xe_vma_bo(vma);
> -
> - lockdep_assert_held_write(&vm->lock);
> -
> - if (bo_has_vm_references(bo, vm, vma))
> - return;
> -
> - __vm_insert_extobj(vm, vma);
> -}
> -
> static void vma_destroy_cb(struct dma_fence *fence,
> struct dma_fence_cb *cb)
> {
> @@ -1082,20 +931,7 @@ static void xe_vma_destroy(struct xe_vma *vma, struct dma_fence *fence)
> } else if (!xe_vma_is_null(vma)) {
> xe_bo_assert_held(xe_vma_bo(vma));
>
> - spin_lock(&vm->notifier.list_lock);
> - list_del(&vma->notifier.rebind_link);
> - spin_unlock(&vm->notifier.list_lock);
> -
> drm_gpuva_unlink(&vma->gpuva);
> -
> - if (!xe_vma_bo(vma)->vm && vm_remove_extobj(vma)) {
> - struct xe_vma *other;
> -
> - other = bo_has_vm_references_locked(xe_vma_bo(vma), vm, NULL);
> -
> - if (other)
> - __vm_insert_extobj(vm, other);
> - }
> }
>
> xe_vm_assert_held(vm);
> @@ -1213,6 +1049,7 @@ static void xe_vm_free(struct drm_gpuvm *gpuvm);
>
> static struct drm_gpuvm_ops gpuvm_ops = {
> .op_alloc = xe_vm_op_alloc,
> + .vm_bo_validate = xe_gpuvm_validate,
> .vm_free = xe_vm_free,
> };
>
> @@ -1375,9 +1212,6 @@ struct xe_vm *xe_vm_create(struct xe_device *xe, u32 flags)
> init_rwsem(&vm->userptr.notifier_lock);
> spin_lock_init(&vm->userptr.invalidated_lock);
>
> - INIT_LIST_HEAD(&vm->notifier.rebind_list);
> - spin_lock_init(&vm->notifier.list_lock);
> -
> INIT_WORK(&vm->destroy_work, vm_destroy_work_func);
>
> INIT_LIST_HEAD(&vm->preempt.exec_queues);
> @@ -1386,8 +1220,6 @@ struct xe_vm *xe_vm_create(struct xe_device *xe, u32 flags)
> for_each_tile(tile, xe, id)
> xe_range_fence_tree_init(&vm->rftree[id]);
>
> - INIT_LIST_HEAD(&vm->extobj.list);
> -
> vm->pt_ops = &xelp_pt_ops;
>
> if (!(flags & XE_VM_FLAG_MIGRATION))
> @@ -1603,7 +1435,6 @@ void xe_vm_close_and_put(struct xe_vm *vm)
> xe_vma_destroy_unlocked(vma);
> }
>
> - xe_assert(xe, list_empty(&vm->extobj.list));
> up_write(&vm->lock);
>
> mutex_lock(&xe->usm.lock);
> @@ -2245,22 +2076,36 @@ static struct xe_vma *new_vma(struct xe_vm *vm, struct drm_gpuva_op_map *op,
> bool read_only, bool is_null, u16 pat_index)
> {
> struct xe_bo *bo = op->gem.obj ? gem_to_xe_bo(op->gem.obj) : NULL;
> + struct drm_exec exec;
> struct xe_vma *vma;
> int err;
>
> lockdep_assert_held_write(&vm->lock);
>
> if (bo) {
> - err = xe_bo_lock(bo, true);
> - if (err)
> - return ERR_PTR(err);
> + drm_exec_init(&exec, DRM_EXEC_INTERRUPTIBLE_WAIT);
> + drm_exec_until_all_locked(&exec) {
> + err = 0;
> + if (!bo->vm) {
> + err = drm_exec_lock_obj(&exec, xe_vm_obj(vm));
> + drm_exec_retry_on_contention(&exec);
> + }
> + if (!err) {
> + err = drm_exec_lock_obj(&exec, &bo->ttm.base);
> + drm_exec_retry_on_contention(&exec);
> + }
> + if (err) {
> + drm_exec_fini(&exec);
> + return ERR_PTR(err);
> + }
> + }
> }
> vma = xe_vma_create(vm, bo, op->gem.offset,
> op->va.addr, op->va.addr +
> op->va.range - 1, read_only, is_null,
> pat_index);
> if (bo)
> - xe_bo_unlock(bo);
> + drm_exec_fini(&exec);
>
> if (xe_vma_is_userptr(vma)) {
> err = xe_vma_userptr_pin_pages(vma);
> @@ -2270,7 +2115,6 @@ static struct xe_vma *new_vma(struct xe_vm *vm, struct drm_gpuva_op_map *op,
> return ERR_PTR(err);
> }
> } else if (!xe_vma_has_no_bo(vma) && !bo->vm) {
> - vm_insert_extobj(vm, vma);
> err = add_preempt_fences(vm, bo);
> if (err) {
> prep_vma_destroy(vm, vma, false);
> diff --git a/drivers/gpu/drm/xe/xe_vm.h b/drivers/gpu/drm/xe/xe_vm.h
> index 12bb5d79487f..306cda064c36 100644
> --- a/drivers/gpu/drm/xe/xe_vm.h
> +++ b/drivers/gpu/drm/xe/xe_vm.h
> @@ -63,9 +63,14 @@ static inline bool xe_vm_is_closed_or_banned(struct xe_vm *vm)
> struct xe_vma *
> xe_vm_find_overlapping_vma(struct xe_vm *vm, u64 start, u64 range);
>
> +static inline struct xe_vm *gpuvm_to_vm(struct drm_gpuvm *gpuvm)
> +{
> + return container_of(gpuvm, struct xe_vm, gpuvm);
> +}
> +
> static inline struct xe_vm *gpuva_to_vm(struct drm_gpuva *gpuva)
> {
> - return container_of(gpuva->vm, struct xe_vm, gpuvm);
> + return gpuvm_to_vm(gpuva->vm);
> }
>
> static inline struct xe_vma *gpuva_to_vma(struct drm_gpuva *gpuva)
> @@ -208,12 +213,6 @@ int xe_vma_userptr_check_repin(struct xe_vma *vma);
>
> bool xe_vm_validate_should_retry(struct drm_exec *exec, int err, ktime_t *end);
>
> -int xe_vm_lock_dma_resv(struct xe_vm *vm, struct drm_exec *exec,
> - unsigned int num_shared, bool lock_vm);
> -
> -void xe_vm_fence_all_extobjs(struct xe_vm *vm, struct dma_fence *fence,
> - enum dma_resv_usage usage);
> -
> 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,
> diff --git a/drivers/gpu/drm/xe/xe_vm_types.h b/drivers/gpu/drm/xe/xe_vm_types.h
> index e70ec6b2fabe..e4922d58d8f0 100644
> --- a/drivers/gpu/drm/xe/xe_vm_types.h
> +++ b/drivers/gpu/drm/xe/xe_vm_types.h
> @@ -62,26 +62,17 @@ struct xe_vma {
> /** @gpuva: Base GPUVA object */
> struct drm_gpuva gpuva;
>
> - /** @combined_links: links into lists which are mutually exclusive */
> + /**
> + * @combined_links: links into lists which are mutually exclusive.
> + * Locking: vm lock in write mode OR vm lock in read mode and the vm's
> + * resv.
> + */
> union {
> - /**
> - * @userptr: link into VM repin list if userptr. Protected by
> - * vm->lock in write mode.
> - */
> + /** @userptr: link into VM repin list if userptr. */
> struct list_head userptr;
> - /**
> - * @rebind: link into VM if this VMA needs rebinding, and
> - * if it's a bo (not userptr) needs validation after a possible
> - * eviction. Protected by the vm's resv lock and typically
> - * vm->lock is also held in write mode. The only place where
> - * vm->lock isn't held is the BO eviction path which has
> - * mutually exclusive execution with userptr.
> - */
> + /** @rebind: link into VM if this VMA needs rebinding. */
> struct list_head rebind;
> - /**
> - * @destroy: link to contested list when VM is being closed.
> - * Protected by vm->lock in write mode and vm's resv lock.
> - */
> + /** @destroy: link to contested list when VM is being closed. */
> struct list_head destroy;
> } combined_links;
>
> @@ -115,18 +106,6 @@ struct xe_vma {
> */
> u16 pat_index;
>
> - struct {
> - struct list_head rebind_link;
> - } notifier;
> -
> - struct {
> - /**
> - * @extobj.link: Link into vm's external object list.
> - * protected by the vm lock.
> - */
> - struct list_head link;
> - } extobj;
> -
> /**
> * @userptr: user pointer state, only allocated for VMAs that are
> * user pointers
> @@ -181,9 +160,9 @@ struct xe_vm {
> struct rw_semaphore lock;
>
> /**
> - * @rebind_list: list of VMAs that need rebinding, and if they are
> - * bos (not userptr), need validation after a possible eviction. The
> - * list is protected by @resv.
> + * @rebind_list: list of VMAs that need rebinding. Protected by the
> + * vm->lock in write mode, OR (the vm->lock in read mode and the
> + * vm resv).
> */
> struct list_head rebind_list;
>
> @@ -203,14 +182,6 @@ struct xe_vm {
> */
> struct xe_range_fence_tree rftree[XE_MAX_TILES_PER_DEVICE];
>
> - /** @extobj: bookkeeping for external objects. Protected by the vm lock */
> - struct {
> - /** @enties: number of external BOs attached this VM */
> - u32 entries;
> - /** @list: list of vmas with external bos attached */
> - struct list_head list;
> - } extobj;
> -
> /** @async_ops: async VM operations (bind / unbinds) */
> struct {
> /** @list: list of pending async VM ops */
> @@ -300,22 +271,6 @@ struct xe_vm {
> struct xe_vma *last_fault_vma;
> } usm;
>
> - /**
> - * @notifier: Lists and locks for temporary usage within notifiers where
> - * we either can't grab the vm lock or the vm resv.
> - */
> - struct {
> - /** @notifier.list_lock: lock protecting @rebind_list */
> - spinlock_t list_lock;
> - /**
> - * @notifier.rebind_list: list of vmas that we want to put on the
> - * main @rebind_list. This list is protected for writing by both
> - * notifier.list_lock, and the resv of the bo the vma points to,
> - * and for reading by the notifier.list_lock only.
> - */
> - struct list_head rebind_list;
> - } notifier;
> -
> /** @error_capture: allow to track errors */
> struct {
> /** @capture_once: capture only one error per VM */
> --
> 2.42.0
>
More information about the Intel-xe
mailing list