diff options
Diffstat (limited to 'drivers/gpu/drm/xe/xe_vm.c')
| -rw-r--r-- | drivers/gpu/drm/xe/xe_vm.c | 3435 |
1 files changed, 2211 insertions, 1224 deletions
diff --git a/drivers/gpu/drm/xe/xe_vm.c b/drivers/gpu/drm/xe/xe_vm.c index 4aa3943e6f29..7cac646bdf1c 100644 --- a/drivers/gpu/drm/xe/xe_vm.c +++ b/drivers/gpu/drm/xe/xe_vm.c @@ -8,11 +8,11 @@ #include <linux/dma-fence-array.h> #include <linux/nospec.h> +#include <drm/drm_drv.h> #include <drm/drm_exec.h> #include <drm/drm_print.h> -#include <drm/ttm/ttm_execbuf_util.h> #include <drm/ttm/ttm_tt.h> -#include <drm/xe_drm.h> +#include <uapi/drm/xe_drm.h> #include <linux/ascii85.h> #include <linux/delay.h> #include <linux/kthread.h> @@ -27,18 +27,20 @@ #include "xe_device.h" #include "xe_drm_client.h" #include "xe_exec_queue.h" -#include "xe_gt_pagefault.h" -#include "xe_gt_tlb_invalidation.h" #include "xe_migrate.h" #include "xe_pat.h" #include "xe_pm.h" #include "xe_preempt_fence.h" #include "xe_pt.h" +#include "xe_pxp.h" #include "xe_res_cursor.h" +#include "xe_sriov_vf.h" +#include "xe_svm.h" #include "xe_sync.h" -#include "xe_trace.h" +#include "xe_tile.h" +#include "xe_tlb_inval.h" +#include "xe_trace_bo.h" #include "xe_wa.h" -#include "xe_hmm.h" static struct drm_gem_object *xe_vm_obj(struct xe_vm *vm) { @@ -46,34 +48,17 @@ static struct drm_gem_object *xe_vm_obj(struct xe_vm *vm) } /** - * xe_vma_userptr_check_repin() - Advisory check for repin needed - * @uvma: The userptr vma + * xe_vm_drm_exec_lock() - Lock the vm's resv with a drm_exec transaction + * @vm: The vm whose resv is to be locked. + * @exec: The drm_exec transaction. * - * Check if the userptr vma has been invalidated since last successful - * repin. The check is advisory only and can the function can be called - * without the vm->userptr.notifier_lock held. There is no guarantee that the - * vma userptr will remain valid after a lockless check, so typically - * the call needs to be followed by a proper check under the notifier_lock. + * Helper to lock the vm's resv as part of a drm_exec transaction. * - * Return: 0 if userptr vma is valid, -EAGAIN otherwise; repin recommended. + * Return: %0 on success. See drm_exec_lock_obj() for error codes. */ -int xe_vma_userptr_check_repin(struct xe_userptr_vma *uvma) +int xe_vm_drm_exec_lock(struct xe_vm *vm, struct drm_exec *exec) { - return mmu_interval_check_retry(&uvma->userptr.notifier, - uvma->userptr.notifier_seq) ? - -EAGAIN : 0; -} - -int xe_vma_userptr_pin_pages(struct xe_userptr_vma *uvma) -{ - struct xe_vma *vma = &uvma->vma; - struct xe_vm *vm = xe_vma_vm(vma); - struct xe_device *xe = vm->xe; - - lockdep_assert_held(&vm->lock); - xe_assert(xe, xe_vma_is_userptr(vma)); - - return xe_hmm_userptr_populate_range(uvma, false); + return drm_exec_lock_obj(exec, xe_vm_obj(vm)); } static bool preempt_fences_waiting(struct xe_vm *vm) @@ -83,10 +68,10 @@ static bool preempt_fences_waiting(struct xe_vm *vm) lockdep_assert_held(&vm->lock); xe_vm_assert_held(vm); - list_for_each_entry(q, &vm->preempt.exec_queues, compute.link) { - if (!q->compute.pfence || - (q->compute.pfence && test_bit(DMA_FENCE_FLAG_ENABLE_SIGNAL_BIT, - &q->compute.pfence->flags))) { + list_for_each_entry(q, &vm->preempt.exec_queues, lr.link) { + if (!q->lr.pfence || + test_bit(DMA_FENCE_FLAG_ENABLE_SIGNAL_BIT, + &q->lr.pfence->flags)) { return true; } } @@ -126,17 +111,29 @@ static int alloc_preempt_fences(struct xe_vm *vm, struct list_head *list, static int wait_for_existing_preempt_fences(struct xe_vm *vm) { struct xe_exec_queue *q; + bool vf_migration = IS_SRIOV_VF(vm->xe) && + xe_sriov_vf_migration_supported(vm->xe); + signed long wait_time = vf_migration ? HZ / 5 : MAX_SCHEDULE_TIMEOUT; xe_vm_assert_held(vm); - list_for_each_entry(q, &vm->preempt.exec_queues, compute.link) { - if (q->compute.pfence) { - long timeout = dma_fence_wait(q->compute.pfence, false); + list_for_each_entry(q, &vm->preempt.exec_queues, lr.link) { + if (q->lr.pfence) { + long timeout; - if (timeout < 0) + timeout = dma_fence_wait_timeout(q->lr.pfence, false, + wait_time); + if (!timeout) { + xe_assert(vm->xe, vf_migration); + return -EAGAIN; + } + + /* Only -ETIME on fence indicates VM needs to be killed */ + if (timeout < 0 || q->lr.pfence->error == -ETIME) return -ETIME; - dma_fence_put(q->compute.pfence); - q->compute.pfence = NULL; + + dma_fence_put(q->lr.pfence); + q->lr.pfence = NULL; } } @@ -148,7 +145,7 @@ static bool xe_vm_is_idle(struct xe_vm *vm) struct xe_exec_queue *q; xe_vm_assert_held(vm); - list_for_each_entry(q, &vm->preempt.exec_queues, compute.link) { + list_for_each_entry(q, &vm->preempt.exec_queues, lr.link) { if (!xe_exec_queue_is_idle(q)) return false; } @@ -161,17 +158,17 @@ static void arm_preempt_fences(struct xe_vm *vm, struct list_head *list) struct list_head *link; struct xe_exec_queue *q; - list_for_each_entry(q, &vm->preempt.exec_queues, compute.link) { + list_for_each_entry(q, &vm->preempt.exec_queues, lr.link) { struct dma_fence *fence; link = list->next; xe_assert(vm->xe, link != list); fence = xe_preempt_fence_arm(to_preempt_fence_from_link(link), - q, q->compute.context, - ++q->compute.seqno); - dma_fence_put(q->compute.pfence); - q->compute.pfence = fence; + q, q->lr.context, + ++q->lr.seqno); + dma_fence_put(q->lr.pfence); + q->lr.pfence = fence; } } @@ -180,27 +177,23 @@ static int add_preempt_fences(struct xe_vm *vm, struct xe_bo *bo) struct xe_exec_queue *q; int err; + xe_bo_assert_held(bo); + if (!vm->preempt.num_exec_queues) return 0; - err = xe_bo_lock(bo, true); - if (err) - return err; - err = dma_resv_reserve_fences(bo->ttm.base.resv, vm->preempt.num_exec_queues); if (err) - goto out_unlock; + return err; - list_for_each_entry(q, &vm->preempt.exec_queues, compute.link) - if (q->compute.pfence) { + list_for_each_entry(q, &vm->preempt.exec_queues, lr.link) + if (q->lr.pfence) { dma_resv_add_fence(bo->ttm.base.resv, - q->compute.pfence, + q->lr.pfence, DMA_RESV_USAGE_BOOKKEEP); } -out_unlock: - xe_bo_unlock(bo); - return err; + return 0; } static void resume_and_reinstall_preempt_fences(struct xe_vm *vm, @@ -211,10 +204,10 @@ static void resume_and_reinstall_preempt_fences(struct xe_vm *vm, lockdep_assert_held(&vm->lock); xe_vm_assert_held(vm); - list_for_each_entry(q, &vm->preempt.exec_queues, compute.link) { + list_for_each_entry(q, &vm->preempt.exec_queues, lr.link) { q->ops->resume(q); - drm_gpuvm_resv_add_fence(&vm->gpuvm, exec, q->compute.pfence, + drm_gpuvm_resv_add_fence(&vm->gpuvm, exec, q->lr.pfence, DMA_RESV_USAGE_BOOKKEEP, DMA_RESV_USAGE_BOOKKEEP); } } @@ -227,6 +220,7 @@ int xe_vm_add_compute_exec_queue(struct xe_vm *vm, struct xe_exec_queue *q) .num_fences = 1, }; struct drm_exec *exec = &vm_exec.exec; + struct xe_validation_ctx ctx; struct dma_fence *pfence; int err; bool wait; @@ -234,22 +228,22 @@ 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); - err = drm_gpuvm_exec_lock(&vm_exec); + err = xe_validation_exec_lock(&ctx, &vm_exec, &vm->xe->val); if (err) goto out_up_write; - pfence = xe_preempt_fence_create(q, q->compute.context, - ++q->compute.seqno); - if (!pfence) { - err = -ENOMEM; + pfence = xe_preempt_fence_create(q, q->lr.context, + ++q->lr.seqno); + if (IS_ERR(pfence)) { + err = PTR_ERR(pfence); goto out_fini; } - list_add(&q->compute.link, &vm->preempt.exec_queues); + list_add(&q->lr.link, &vm->preempt.exec_queues); ++vm->preempt.num_exec_queues; - q->compute.pfence = pfence; + q->lr.pfence = pfence; - down_read(&vm->userptr.notifier_lock); + xe_svm_notifier_lock(vm); drm_gpuvm_resv_add_fence(&vm->gpuvm, exec, pfence, DMA_RESV_USAGE_BOOKKEEP, DMA_RESV_USAGE_BOOKKEEP); @@ -263,20 +257,23 @@ int xe_vm_add_compute_exec_queue(struct xe_vm *vm, struct xe_exec_queue *q) if (wait) dma_fence_enable_sw_signaling(pfence); - up_read(&vm->userptr.notifier_lock); + xe_svm_notifier_unlock(vm); out_fini: - drm_exec_fini(exec); + xe_validation_ctx_fini(&ctx); out_up_write: up_write(&vm->lock); return err; } +ALLOW_ERROR_INJECTION(xe_vm_add_compute_exec_queue, ERRNO); /** * xe_vm_remove_compute_exec_queue() - Remove compute exec queue from VM * @vm: The VM. * @q: The exec_queue + * + * Note that this function might be called multiple times on the same queue. */ void xe_vm_remove_compute_exec_queue(struct xe_vm *vm, struct xe_exec_queue *q) { @@ -284,85 +281,47 @@ void xe_vm_remove_compute_exec_queue(struct xe_vm *vm, struct xe_exec_queue *q) return; down_write(&vm->lock); - list_del(&q->compute.link); - --vm->preempt.num_exec_queues; - if (q->compute.pfence) { - dma_fence_enable_sw_signaling(q->compute.pfence); - dma_fence_put(q->compute.pfence); - q->compute.pfence = NULL; + if (!list_empty(&q->lr.link)) { + list_del_init(&q->lr.link); + --vm->preempt.num_exec_queues; + } + if (q->lr.pfence) { + dma_fence_enable_sw_signaling(q->lr.pfence); + dma_fence_put(q->lr.pfence); + q->lr.pfence = NULL; } up_write(&vm->lock); } +#define XE_VM_REBIND_RETRY_TIMEOUT_MS 1000 + /** - * __xe_vm_userptr_needs_repin() - Check whether the VM does have userptrs - * that need repinning. + * xe_vm_kill() - VM Kill * @vm: The VM. + * @unlocked: Flag indicates the VM's dma-resv is not held * - * This function checks for whether the VM has userptrs that need repinning, - * and provides a release-type barrier on the userptr.notifier_lock after - * checking. - * - * Return: 0 if there are no userptrs needing repinning, -EAGAIN if there are. + * Kill the VM by setting banned flag indicated VM is no longer available for + * use. If in preempt fence mode, also kill all exec queue attached to the VM. */ -int __xe_vm_userptr_needs_repin(struct xe_vm *vm) -{ - lockdep_assert_held_read(&vm->userptr.notifier_lock); - - return (list_empty(&vm->userptr.repin_list) && - list_empty(&vm->userptr.invalidated)) ? 0 : -EAGAIN; -} - -#define XE_VM_REBIND_RETRY_TIMEOUT_MS 1000 - -static void xe_vm_kill(struct xe_vm *vm) +void xe_vm_kill(struct xe_vm *vm, bool unlocked) { struct xe_exec_queue *q; lockdep_assert_held(&vm->lock); - xe_vm_lock(vm, false); + if (unlocked) + xe_vm_lock(vm, false); + vm->flags |= XE_VM_FLAG_BANNED; trace_xe_vm_kill(vm); - list_for_each_entry(q, &vm->preempt.exec_queues, compute.link) + list_for_each_entry(q, &vm->preempt.exec_queues, lr.link) q->ops->kill(q); - xe_vm_unlock(vm); - - /* TODO: Inform user the VM is banned */ -} - -/** - * xe_vm_validate_should_retry() - Whether to retry after a validate error. - * @exec: The drm_exec object used for locking before validation. - * @err: The error returned from ttm_bo_validate(). - * @end: A ktime_t cookie that should be set to 0 before first use and - * that should be reused on subsequent calls. - * - * With multiple active VMs, under memory pressure, it is possible that - * ttm_bo_validate() run into -EDEADLK and in such case returns -ENOMEM. - * Until ttm properly handles locking in such scenarios, best thing the - * driver can do is retry with a timeout. Check if that is necessary, and - * if so unlock the drm_exec's objects while keeping the ticket to prepare - * for a rerun. - * - * Return: true if a retry after drm_exec_init() is recommended; - * false otherwise. - */ -bool xe_vm_validate_should_retry(struct drm_exec *exec, int err, ktime_t *end) -{ - ktime_t cur; - - if (err != -ENOMEM) - return false; - cur = ktime_get(); - *end = *end ? : ktime_add_ms(cur, XE_VM_REBIND_RETRY_TIMEOUT_MS); - if (!ktime_before(cur, *end)) - return false; + if (unlocked) + xe_vm_unlock(vm); - msleep(20); - return true; + /* TODO: Inform user the VM is banned */ } static int xe_gpuvm_validate(struct drm_gpuvm_bo *vm_bo, struct drm_exec *exec) @@ -376,7 +335,10 @@ static int xe_gpuvm_validate(struct drm_gpuvm_bo *vm_bo, struct drm_exec *exec) 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 (!try_wait_for_completion(&vm->xe->pm_block)) + return -EAGAIN; + + ret = xe_bo_validate(gem_to_xe_bo(vm_bo->obj), vm, false, exec); if (ret) return ret; @@ -462,13 +424,40 @@ static int xe_preempt_work_begin(struct drm_exec *exec, struct xe_vm *vm, return xe_vm_validate_rebind(vm, exec, vm->preempt.num_exec_queues); } +static bool vm_suspend_rebind_worker(struct xe_vm *vm) +{ + struct xe_device *xe = vm->xe; + bool ret = false; + + mutex_lock(&xe->rebind_resume_lock); + if (!try_wait_for_completion(&vm->xe->pm_block)) { + ret = true; + list_move_tail(&vm->preempt.pm_activate_link, &xe->rebind_resume_list); + } + mutex_unlock(&xe->rebind_resume_lock); + + return ret; +} + +/** + * xe_vm_resume_rebind_worker() - Resume the rebind worker. + * @vm: The vm whose preempt worker to resume. + * + * Resume a preempt worker that was previously suspended by + * vm_suspend_rebind_worker(). + */ +void xe_vm_resume_rebind_worker(struct xe_vm *vm) +{ + queue_work(vm->xe->ordered_wq, &vm->preempt.rebind_work); +} + static void preempt_rebind_work_func(struct work_struct *w) { struct xe_vm *vm = container_of(w, struct xe_vm, preempt.rebind_work); + struct xe_validation_ctx ctx; struct drm_exec exec; unsigned int fence_count = 0; LIST_HEAD(preempt_fences); - ktime_t end = 0; int err = 0; long wait; int __maybe_unused tries = 0; @@ -485,24 +474,32 @@ static void preempt_rebind_work_func(struct work_struct *w) } retry: + if (!try_wait_for_completion(&vm->xe->pm_block) && vm_suspend_rebind_worker(vm)) { + up_write(&vm->lock); + /* We don't actually block but don't make progress. */ + xe_pm_might_block_on_suspend(); + return; + } + if (xe_vm_userptr_check_repin(vm)) { err = xe_vm_userptr_pin(vm); if (err) goto out_unlock_outer; } - drm_exec_init(&exec, DRM_EXEC_INTERRUPTIBLE_WAIT, 0); + err = xe_validation_ctx_init(&ctx, &vm->xe->val, &exec, + (struct xe_val_flags) {.interruptible = true}); + if (err) + goto out_unlock_outer; drm_exec_until_all_locked(&exec) { bool done = false; err = xe_preempt_work_begin(&exec, vm, &done); drm_exec_retry_on_contention(&exec); + xe_validation_retry_on_oom(&ctx, &err); if (err || done) { - drm_exec_fini(&exec); - if (err && xe_vm_validate_should_retry(&exec, err, &end)) - err = -EAGAIN; - + xe_validation_ctx_fini(&ctx); goto out_unlock_outer; } } @@ -511,7 +508,9 @@ retry: if (err) goto out_unlock; + xe_vm_set_validation_exec(vm, &exec); err = xe_vm_rebind(vm, true); + xe_vm_set_validation_exec(vm, NULL); if (err) goto out_unlock; @@ -529,9 +528,9 @@ retry: (!(__tries)++ || __xe_vm_userptr_needs_repin(__vm)) : \ __xe_vm_userptr_needs_repin(__vm)) - down_read(&vm->userptr.notifier_lock); + xe_svm_notifier_lock(vm); if (retry_required(tries, vm)) { - up_read(&vm->userptr.notifier_lock); + xe_svm_notifier_unlock(vm); err = -EAGAIN; goto out_unlock; } @@ -545,19 +544,32 @@ retry: /* Point of no return. */ arm_preempt_fences(vm, &preempt_fences); resume_and_reinstall_preempt_fences(vm, &exec); - up_read(&vm->userptr.notifier_lock); + xe_svm_notifier_unlock(vm); out_unlock: - drm_exec_fini(&exec); + xe_validation_ctx_fini(&ctx); out_unlock_outer: if (err == -EAGAIN) { trace_xe_vm_rebind_worker_retry(vm); + + /* + * We can't block in workers on a VF which supports migration + * given this can block the VF post-migration workers from + * getting scheduled. + */ + if (IS_SRIOV_VF(vm->xe) && + xe_sriov_vf_migration_supported(vm->xe)) { + up_write(&vm->lock); + xe_vm_queue_rebind_worker(vm); + return; + } + goto retry; } if (err) { drm_warn(&vm->xe->drm, "VM worker error: %d\n", err); - xe_vm_kill(vm); + xe_vm_kill(vm, true); } up_write(&vm->lock); @@ -566,181 +578,385 @@ out_unlock_outer: trace_xe_vm_rebind_worker_exit(vm); } -static bool vma_userptr_invalidate(struct mmu_interval_notifier *mni, - const struct mmu_notifier_range *range, - unsigned long cur_seq) +static int xe_vma_ops_alloc(struct xe_vma_ops *vops, bool array_of_binds) { - struct xe_userptr *userptr = container_of(mni, typeof(*userptr), notifier); - struct xe_userptr_vma *uvma = container_of(userptr, typeof(*uvma), userptr); - struct xe_vma *vma = &uvma->vma; - struct xe_vm *vm = xe_vma_vm(vma); - struct dma_resv_iter cursor; - struct dma_fence *fence; - long err; + int i; - xe_assert(vm->xe, xe_vma_is_userptr(vma)); - trace_xe_vma_userptr_invalidate(vma); + for (i = 0; i < XE_MAX_TILES_PER_DEVICE; ++i) { + if (!vops->pt_update_ops[i].num_ops) + continue; - if (!mmu_notifier_range_blockable(range)) - return false; + vops->pt_update_ops[i].ops = + kmalloc_array(vops->pt_update_ops[i].num_ops, + sizeof(*vops->pt_update_ops[i].ops), + GFP_KERNEL | __GFP_RETRY_MAYFAIL | __GFP_NOWARN); + if (!vops->pt_update_ops[i].ops) + return array_of_binds ? -ENOBUFS : -ENOMEM; + } - vm_dbg(&xe_vma_vm(vma)->xe->drm, - "NOTIFIER: addr=0x%016llx, range=0x%016llx", - xe_vma_start(vma), xe_vma_size(vma)); + return 0; +} +ALLOW_ERROR_INJECTION(xe_vma_ops_alloc, ERRNO); - down_write(&vm->userptr.notifier_lock); - mmu_interval_set_seq(mni, cur_seq); +static void xe_vma_svm_prefetch_op_fini(struct xe_vma_op *op) +{ + struct xe_vma *vma; - /* No need to stop gpu access if the userptr is not yet bound. */ - if (!userptr->initial_bind) { - up_write(&vm->userptr.notifier_lock); - return true; - } + vma = gpuva_to_vma(op->base.prefetch.va); - /* - * Tell exec and rebind worker they need to repin and rebind this - * userptr. - */ - if (!xe_vm_in_fault_mode(vm) && - !(vma->gpuva.flags & XE_VMA_DESTROYED) && vma->tile_present) { - spin_lock(&vm->userptr.invalidated_lock); - list_move_tail(&userptr->invalidate_link, - &vm->userptr.invalidated); - spin_unlock(&vm->userptr.invalidated_lock); - } + if (op->base.op == DRM_GPUVA_OP_PREFETCH && xe_vma_is_cpu_addr_mirror(vma)) + xa_destroy(&op->prefetch_range.range); +} - up_write(&vm->userptr.notifier_lock); +static void xe_vma_svm_prefetch_ops_fini(struct xe_vma_ops *vops) +{ + struct xe_vma_op *op; - /* - * Preempt fences turn into schedule disables, pipeline these. - * Note that even in fault mode, we need to wait for binds and - * unbinds to complete, and those are attached as BOOKMARK fences - * to the vm. - */ - 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); + if (!(vops->flags & XE_VMA_OPS_FLAG_HAS_SVM_PREFETCH)) + return; - err = dma_resv_wait_timeout(xe_vm_resv(vm), - DMA_RESV_USAGE_BOOKKEEP, - false, MAX_SCHEDULE_TIMEOUT); - XE_WARN_ON(err <= 0); + list_for_each_entry(op, &vops->list, link) + xe_vma_svm_prefetch_op_fini(op); +} - if (xe_vm_in_fault_mode(vm)) { - err = xe_vm_invalidate_vma(vma); - XE_WARN_ON(err); - } +static void xe_vma_ops_fini(struct xe_vma_ops *vops) +{ + int i; - trace_xe_vma_userptr_invalidate_complete(vma); + xe_vma_svm_prefetch_ops_fini(vops); - return true; + for (i = 0; i < XE_MAX_TILES_PER_DEVICE; ++i) + kfree(vops->pt_update_ops[i].ops); } -static const struct mmu_interval_notifier_ops vma_userptr_notifier_ops = { - .invalidate = vma_userptr_invalidate, -}; - -int xe_vm_userptr_pin(struct xe_vm *vm) +static void xe_vma_ops_incr_pt_update_ops(struct xe_vma_ops *vops, u8 tile_mask, int inc_val) { - struct xe_userptr_vma *uvma, *next; - int err = 0; - LIST_HEAD(tmp_evict); + int i; - xe_assert(vm->xe, !xe_vm_in_fault_mode(vm)); - lockdep_assert_held_write(&vm->lock); + if (!inc_val) + return; - /* Collect invalidated userptrs */ - spin_lock(&vm->userptr.invalidated_lock); - list_for_each_entry_safe(uvma, next, &vm->userptr.invalidated, - userptr.invalidate_link) { - list_del_init(&uvma->userptr.invalidate_link); - list_move_tail(&uvma->userptr.repin_link, - &vm->userptr.repin_list); - } - spin_unlock(&vm->userptr.invalidated_lock); - - /* Pin and move to temporary list */ - list_for_each_entry_safe(uvma, next, &vm->userptr.repin_list, - userptr.repin_link) { - err = xe_vma_userptr_pin_pages(uvma); - if (err == -EFAULT) { - list_del_init(&uvma->userptr.repin_link); - - /* Wait for pending binds */ - xe_vm_lock(vm, false); - dma_resv_wait_timeout(xe_vm_resv(vm), - DMA_RESV_USAGE_BOOKKEEP, - false, MAX_SCHEDULE_TIMEOUT); - - err = xe_vm_invalidate_vma(&uvma->vma); - xe_vm_unlock(vm); - if (err) - return err; - } else { - if (err < 0) - return err; + for (i = 0; i < XE_MAX_TILES_PER_DEVICE; ++i) + if (BIT(i) & tile_mask) + vops->pt_update_ops[i].num_ops += inc_val; +} - list_del_init(&uvma->userptr.repin_link); - list_move_tail(&uvma->vma.combined_links.rebind, - &vm->rebind_list); - } - } +#define XE_VMA_CREATE_MASK ( \ + XE_VMA_READ_ONLY | \ + XE_VMA_DUMPABLE | \ + XE_VMA_SYSTEM_ALLOCATOR | \ + DRM_GPUVA_SPARSE | \ + XE_VMA_MADV_AUTORESET) - return 0; +static void xe_vm_populate_rebind(struct xe_vma_op *op, struct xe_vma *vma, + u8 tile_mask) +{ + INIT_LIST_HEAD(&op->link); + op->tile_mask = tile_mask; + op->base.op = DRM_GPUVA_OP_MAP; + op->base.map.va.addr = vma->gpuva.va.addr; + op->base.map.va.range = vma->gpuva.va.range; + op->base.map.gem.obj = vma->gpuva.gem.obj; + op->base.map.gem.offset = vma->gpuva.gem.offset; + op->map.vma = vma; + op->map.immediate = true; + op->map.vma_flags = vma->gpuva.flags & XE_VMA_CREATE_MASK; } -/** - * xe_vm_userptr_check_repin() - Check whether the VM might have userptrs - * that need repinning. - * @vm: The VM. - * - * This function does an advisory check for whether the VM has userptrs that - * need repinning. - * - * Return: 0 if there are no indications of userptrs needing repinning, - * -EAGAIN if there are. - */ -int xe_vm_userptr_check_repin(struct xe_vm *vm) +static int xe_vm_ops_add_rebind(struct xe_vma_ops *vops, struct xe_vma *vma, + u8 tile_mask) { - return (list_empty_careful(&vm->userptr.repin_list) && - list_empty_careful(&vm->userptr.invalidated)) ? 0 : -EAGAIN; + struct xe_vma_op *op; + + op = kzalloc(sizeof(*op), GFP_KERNEL); + if (!op) + return -ENOMEM; + + xe_vm_populate_rebind(op, vma, tile_mask); + list_add_tail(&op->link, &vops->list); + xe_vma_ops_incr_pt_update_ops(vops, tile_mask, 1); + + return 0; } -static struct dma_fence * -xe_vm_bind_vma(struct xe_vma *vma, struct xe_exec_queue *q, - struct xe_sync_entry *syncs, u32 num_syncs, - bool first_op, bool last_op); +static struct dma_fence *ops_execute(struct xe_vm *vm, + struct xe_vma_ops *vops); +static void xe_vma_ops_init(struct xe_vma_ops *vops, struct xe_vm *vm, + struct xe_exec_queue *q, + struct xe_sync_entry *syncs, u32 num_syncs); int xe_vm_rebind(struct xe_vm *vm, bool rebind_worker) { struct dma_fence *fence; struct xe_vma *vma, *next; + struct xe_vma_ops vops; + struct xe_vma_op *op, *next_op; + int err, i; lockdep_assert_held(&vm->lock); - if (xe_vm_in_lr_mode(vm) && !rebind_worker) + if ((xe_vm_in_lr_mode(vm) && !rebind_worker) || + list_empty(&vm->rebind_list)) return 0; + xe_vma_ops_init(&vops, vm, NULL, NULL, 0); + for (i = 0; i < XE_MAX_TILES_PER_DEVICE; ++i) + vops.pt_update_ops[i].wait_vm_bookkeep = true; + xe_vm_assert_held(vm); - list_for_each_entry_safe(vma, next, &vm->rebind_list, - combined_links.rebind) { + list_for_each_entry(vma, &vm->rebind_list, combined_links.rebind) { xe_assert(vm->xe, vma->tile_present); - list_del_init(&vma->combined_links.rebind); if (rebind_worker) trace_xe_vma_rebind_worker(vma); else trace_xe_vma_rebind_exec(vma); - fence = xe_vm_bind_vma(vma, NULL, NULL, 0, false, false); - if (IS_ERR(fence)) - return PTR_ERR(fence); + + err = xe_vm_ops_add_rebind(&vops, vma, + vma->tile_present); + if (err) + goto free_ops; + } + + err = xe_vma_ops_alloc(&vops, false); + if (err) + goto free_ops; + + fence = ops_execute(vm, &vops); + if (IS_ERR(fence)) { + err = PTR_ERR(fence); + } else { dma_fence_put(fence); + list_for_each_entry_safe(vma, next, &vm->rebind_list, + combined_links.rebind) + list_del_init(&vma->combined_links.rebind); + } +free_ops: + list_for_each_entry_safe(op, next_op, &vops.list, link) { + list_del(&op->link); + kfree(op); } + xe_vma_ops_fini(&vops); + + return err; +} + +struct dma_fence *xe_vma_rebind(struct xe_vm *vm, struct xe_vma *vma, u8 tile_mask) +{ + struct dma_fence *fence = NULL; + struct xe_vma_ops vops; + struct xe_vma_op *op, *next_op; + struct xe_tile *tile; + u8 id; + int err; + + lockdep_assert_held(&vm->lock); + xe_vm_assert_held(vm); + xe_assert(vm->xe, xe_vm_in_fault_mode(vm)); + + xe_vma_ops_init(&vops, vm, NULL, NULL, 0); + vops.flags |= XE_VMA_OPS_FLAG_SKIP_TLB_WAIT; + for_each_tile(tile, vm->xe, id) { + vops.pt_update_ops[id].wait_vm_bookkeep = true; + vops.pt_update_ops[tile->id].q = + xe_migrate_exec_queue(tile->migrate); + } + + err = xe_vm_ops_add_rebind(&vops, vma, tile_mask); + if (err) + return ERR_PTR(err); + + err = xe_vma_ops_alloc(&vops, false); + if (err) { + fence = ERR_PTR(err); + goto free_ops; + } + + fence = ops_execute(vm, &vops); + +free_ops: + list_for_each_entry_safe(op, next_op, &vops.list, link) { + list_del(&op->link); + kfree(op); + } + xe_vma_ops_fini(&vops); + + return fence; +} + +static void xe_vm_populate_range_rebind(struct xe_vma_op *op, + struct xe_vma *vma, + struct xe_svm_range *range, + u8 tile_mask) +{ + INIT_LIST_HEAD(&op->link); + op->tile_mask = tile_mask; + op->base.op = DRM_GPUVA_OP_DRIVER; + op->subop = XE_VMA_SUBOP_MAP_RANGE; + op->map_range.vma = vma; + op->map_range.range = range; +} + +static int +xe_vm_ops_add_range_rebind(struct xe_vma_ops *vops, + struct xe_vma *vma, + struct xe_svm_range *range, + u8 tile_mask) +{ + struct xe_vma_op *op; + + op = kzalloc(sizeof(*op), GFP_KERNEL); + if (!op) + return -ENOMEM; + + xe_vm_populate_range_rebind(op, vma, range, tile_mask); + list_add_tail(&op->link, &vops->list); + xe_vma_ops_incr_pt_update_ops(vops, tile_mask, 1); + + return 0; +} + +/** + * xe_vm_range_rebind() - VM range (re)bind + * @vm: The VM which the range belongs to. + * @vma: The VMA which the range belongs to. + * @range: SVM range to rebind. + * @tile_mask: Tile mask to bind the range to. + * + * (re)bind SVM range setting up GPU page tables for the range. + * + * Return: dma fence for rebind to signal completion on success, ERR_PTR on + * failure + */ +struct dma_fence *xe_vm_range_rebind(struct xe_vm *vm, + struct xe_vma *vma, + struct xe_svm_range *range, + u8 tile_mask) +{ + struct dma_fence *fence = NULL; + struct xe_vma_ops vops; + struct xe_vma_op *op, *next_op; + struct xe_tile *tile; + u8 id; + int err; + + lockdep_assert_held(&vm->lock); + xe_vm_assert_held(vm); + xe_assert(vm->xe, xe_vm_in_fault_mode(vm)); + xe_assert(vm->xe, xe_vma_is_cpu_addr_mirror(vma)); + + xe_vma_ops_init(&vops, vm, NULL, NULL, 0); + vops.flags |= XE_VMA_OPS_FLAG_SKIP_TLB_WAIT; + for_each_tile(tile, vm->xe, id) { + vops.pt_update_ops[id].wait_vm_bookkeep = true; + vops.pt_update_ops[tile->id].q = + xe_migrate_exec_queue(tile->migrate); + } + + err = xe_vm_ops_add_range_rebind(&vops, vma, range, tile_mask); + if (err) + return ERR_PTR(err); + + err = xe_vma_ops_alloc(&vops, false); + if (err) { + fence = ERR_PTR(err); + goto free_ops; + } + + fence = ops_execute(vm, &vops); + +free_ops: + list_for_each_entry_safe(op, next_op, &vops.list, link) { + list_del(&op->link); + kfree(op); + } + xe_vma_ops_fini(&vops); + + return fence; +} + +static void xe_vm_populate_range_unbind(struct xe_vma_op *op, + struct xe_svm_range *range) +{ + INIT_LIST_HEAD(&op->link); + op->tile_mask = range->tile_present; + op->base.op = DRM_GPUVA_OP_DRIVER; + op->subop = XE_VMA_SUBOP_UNMAP_RANGE; + op->unmap_range.range = range; +} + +static int +xe_vm_ops_add_range_unbind(struct xe_vma_ops *vops, + struct xe_svm_range *range) +{ + struct xe_vma_op *op; + + op = kzalloc(sizeof(*op), GFP_KERNEL); + if (!op) + return -ENOMEM; + + xe_vm_populate_range_unbind(op, range); + list_add_tail(&op->link, &vops->list); + xe_vma_ops_incr_pt_update_ops(vops, range->tile_present, 1); return 0; } +/** + * xe_vm_range_unbind() - VM range unbind + * @vm: The VM which the range belongs to. + * @range: SVM range to rebind. + * + * Unbind SVM range removing the GPU page tables for the range. + * + * Return: dma fence for unbind to signal completion on success, ERR_PTR on + * failure + */ +struct dma_fence *xe_vm_range_unbind(struct xe_vm *vm, + struct xe_svm_range *range) +{ + struct dma_fence *fence = NULL; + struct xe_vma_ops vops; + struct xe_vma_op *op, *next_op; + struct xe_tile *tile; + u8 id; + int err; + + lockdep_assert_held(&vm->lock); + xe_vm_assert_held(vm); + xe_assert(vm->xe, xe_vm_in_fault_mode(vm)); + + if (!range->tile_present) + return dma_fence_get_stub(); + + xe_vma_ops_init(&vops, vm, NULL, NULL, 0); + for_each_tile(tile, vm->xe, id) { + vops.pt_update_ops[id].wait_vm_bookkeep = true; + vops.pt_update_ops[tile->id].q = + xe_migrate_exec_queue(tile->migrate); + } + + err = xe_vm_ops_add_range_unbind(&vops, range); + if (err) + return ERR_PTR(err); + + err = xe_vma_ops_alloc(&vops, false); + if (err) { + fence = ERR_PTR(err); + goto free_ops; + } + + fence = ops_execute(vm, &vops); + +free_ops: + list_for_each_entry_safe(op, next_op, &vops.list, link) { + list_del(&op->link); + kfree(op); + } + xe_vma_ops_fini(&vops); + + return fence; +} + static void xe_vma_free(struct xe_vma *vma) { if (xe_vma_is_userptr(vma)) @@ -749,22 +965,18 @@ static void xe_vma_free(struct xe_vma *vma) kfree(vma); } -#define VMA_CREATE_FLAG_READ_ONLY BIT(0) -#define VMA_CREATE_FLAG_IS_NULL BIT(1) -#define VMA_CREATE_FLAG_DUMPABLE BIT(2) - static struct xe_vma *xe_vma_create(struct xe_vm *vm, struct xe_bo *bo, u64 bo_offset_or_userptr, u64 start, u64 end, - u16 pat_index, unsigned int flags) + struct xe_vma_mem_attr *attr, + unsigned int flags) { struct xe_vma *vma; struct xe_tile *tile; u8 id; - bool read_only = (flags & VMA_CREATE_FLAG_READ_ONLY); - bool is_null = (flags & VMA_CREATE_FLAG_IS_NULL); - bool dumpable = (flags & VMA_CREATE_FLAG_DUMPABLE); + bool is_null = (flags & DRM_GPUVA_SPARSE); + bool is_cpu_addr_mirror = (flags & XE_VMA_SYSTEM_ALLOCATOR); xe_assert(vm->xe, start < end); xe_assert(vm->xe, end < vm->size); @@ -773,7 +985,7 @@ static struct xe_vma *xe_vma_create(struct xe_vm *vm, * Allocate and ensure that the xe_vma_is_userptr() return * matches what was allocated. */ - if (!bo && !is_null) { + if (!bo && !is_null && !is_cpu_addr_mirror) { struct xe_userptr_vma *uvma = kzalloc(sizeof(*uvma), GFP_KERNEL); if (!uvma) @@ -785,8 +997,6 @@ static struct xe_vma *xe_vma_create(struct xe_vm *vm, if (!vma) return ERR_PTR(-ENOMEM); - if (is_null) - vma->gpuva.flags |= DRM_GPUVA_SPARSE; if (bo) vma->gpuva.gem.obj = &bo->ttm.base; } @@ -797,18 +1007,15 @@ static struct xe_vma *xe_vma_create(struct xe_vm *vm, vma->gpuva.vm = &vm->gpuvm; vma->gpuva.va.addr = start; vma->gpuva.va.range = end - start + 1; - if (read_only) - vma->gpuva.flags |= XE_VMA_READ_ONLY; - if (dumpable) - vma->gpuva.flags |= XE_VMA_DUMPABLE; + vma->gpuva.flags = flags; for_each_tile(tile, vm->xe, id) vma->tile_mask |= 0x1 << id; - if (GRAPHICS_VER(vm->xe) >= 20 || vm->xe->info.platform == XE_PVC) + if (vm->xe->info.has_atomic_enable_pte_bit) vma->gpuva.flags |= XE_VMA_ATOMIC_PTE_BIT; - vma->pat_index = pat_index; + vma->attr = *attr; if (bo) { struct drm_gpuvm_bo *vm_bo; @@ -827,25 +1034,18 @@ static struct xe_vma *xe_vma_create(struct xe_vm *vm, drm_gpuva_link(&vma->gpuva, vm_bo); drm_gpuvm_bo_put(vm_bo); } else /* userptr or null */ { - if (!is_null) { - struct xe_userptr *userptr = &to_userptr_vma(vma)->userptr; + if (!is_null && !is_cpu_addr_mirror) { + struct xe_userptr_vma *uvma = to_userptr_vma(vma); u64 size = end - start + 1; int err; - INIT_LIST_HEAD(&userptr->invalidate_link); - INIT_LIST_HEAD(&userptr->repin_link); vma->gpuva.gem.offset = bo_offset_or_userptr; - err = mmu_interval_notifier_insert(&userptr->notifier, - current->mm, - xe_vma_userptr(vma), size, - &vma_userptr_notifier_ops); + err = xe_userptr_setup(uvma, xe_vma_userptr(vma), size); if (err) { xe_vma_free(vma); return ERR_PTR(err); } - - userptr->notifier_seq = LONG_MAX; } xe_vm_get(vm); @@ -865,19 +1065,10 @@ static void xe_vma_destroy_late(struct xe_vma *vma) if (xe_vma_is_userptr(vma)) { struct xe_userptr_vma *uvma = to_userptr_vma(vma); - struct xe_userptr *userptr = &uvma->userptr; - if (userptr->sg) - xe_hmm_userptr_free_sg(uvma); - - /* - * Since userptr pages are not pinned, we can't remove - * the notifer until we're sure the GPU is not accessing - * them anymore - */ - mmu_interval_notifier_remove(&userptr->notifier); + xe_userptr_remove(uvma); xe_vm_put(vm); - } else if (xe_vma_is_null(vma)) { + } else if (xe_vma_is_null(vma) || xe_vma_is_cpu_addr_mirror(vma)) { xe_vm_put(vm); } else { xe_bo_put(xe_vma_bo(vma)); @@ -912,11 +1103,8 @@ static void xe_vma_destroy(struct xe_vma *vma, struct dma_fence *fence) if (xe_vma_is_userptr(vma)) { xe_assert(vm->xe, vma->gpuva.flags & XE_VMA_DESTROYED); - - spin_lock(&vm->userptr.invalidated_lock); - list_del(&to_userptr_vma(vma)->userptr.invalidate_link); - spin_unlock(&vm->userptr.invalidated_lock); - } else if (!xe_vma_is_null(vma)) { + xe_userptr_destroy(to_userptr_vma(vma)); + } else if (!xe_vma_is_null(vma) && !xe_vma_is_cpu_addr_mirror(vma)) { xe_bo_assert_held(xe_vma_bo(vma)); drm_gpuva_unlink(&vma->gpuva); @@ -963,20 +1151,19 @@ int xe_vm_lock_vma(struct drm_exec *exec, struct xe_vma *vma) static void xe_vma_destroy_unlocked(struct xe_vma *vma) { + struct xe_device *xe = xe_vma_vm(vma)->xe; + struct xe_validation_ctx ctx; struct drm_exec exec; - int err; + int err = 0; - drm_exec_init(&exec, 0, 0); - drm_exec_until_all_locked(&exec) { + xe_validation_guard(&ctx, &xe->val, &exec, (struct xe_val_flags) {}, err) { err = xe_vm_lock_vma(&exec, vma); drm_exec_retry_on_contention(&exec); if (XE_WARN_ON(err)) break; + xe_vma_destroy(vma, NULL); } - - xe_vma_destroy(vma, NULL); - - drm_exec_fini(&exec); + xe_assert(xe, !err); } struct xe_vma * @@ -1043,7 +1230,7 @@ static const struct drm_gpuvm_ops gpuvm_ops = { .vm_free = xe_vm_free, }; -static u64 pde_encode_pat_index(struct xe_device *xe, u16 pat_index) +static u64 pde_encode_pat_index(u16 pat_index) { u64 pte = 0; @@ -1056,8 +1243,7 @@ static u64 pde_encode_pat_index(struct xe_device *xe, u16 pat_index) return pte; } -static u64 pte_encode_pat_index(struct xe_device *xe, u16 pat_index, - u32 pt_level) +static u64 pte_encode_pat_index(u16 pat_index, u32 pt_level) { u64 pte = 0; @@ -1095,15 +1281,39 @@ static u64 pte_encode_ps(u32 pt_level) return 0; } -static u64 xelp_pde_encode_bo(struct xe_bo *bo, u64 bo_offset, - const u16 pat_index) +static u16 pde_pat_index(struct xe_bo *bo) { struct xe_device *xe = xe_bo_device(bo); + u16 pat_index; + + /* + * We only have two bits to encode the PAT index in non-leaf nodes, but + * these only point to other paging structures so we only need a minimal + * selection of options. The user PAT index is only for encoding leaf + * nodes, where we have use of more bits to do the encoding. The + * non-leaf nodes are instead under driver control so the chosen index + * here should be distinct from the user PAT index. Also the + * corresponding coherency of the PAT index should be tied to the + * allocation type of the page table (or at least we should pick + * something which is always safe). + */ + if (!xe_bo_is_vram(bo) && bo->ttm.ttm->caching == ttm_cached) + pat_index = xe->pat.idx[XE_CACHE_WB]; + else + pat_index = xe->pat.idx[XE_CACHE_NONE]; + + xe_assert(xe, pat_index <= 3); + + return pat_index; +} + +static u64 xelp_pde_encode_bo(struct xe_bo *bo, u64 bo_offset) +{ u64 pde; pde = xe_bo_addr(bo, bo_offset, XE_PAGE_SIZE); pde |= XE_PAGE_PRESENT | XE_PAGE_RW; - pde |= pde_encode_pat_index(xe, pat_index); + pde |= pde_encode_pat_index(pde_pat_index(bo)); return pde; } @@ -1111,12 +1321,11 @@ static u64 xelp_pde_encode_bo(struct xe_bo *bo, u64 bo_offset, static u64 xelp_pte_encode_bo(struct xe_bo *bo, u64 bo_offset, u16 pat_index, u32 pt_level) { - struct xe_device *xe = xe_bo_device(bo); u64 pte; pte = xe_bo_addr(bo, bo_offset, XE_PAGE_SIZE); pte |= XE_PAGE_PRESENT | XE_PAGE_RW; - pte |= pte_encode_pat_index(xe, pat_index, pt_level); + pte |= pte_encode_pat_index(pat_index, pt_level); pte |= pte_encode_ps(pt_level); if (xe_bo_is_vram(bo) || xe_bo_is_stolen_devmem(bo)) @@ -1128,14 +1337,12 @@ static u64 xelp_pte_encode_bo(struct xe_bo *bo, u64 bo_offset, static u64 xelp_pte_encode_vma(u64 pte, struct xe_vma *vma, u16 pat_index, u32 pt_level) { - struct xe_device *xe = xe_vma_vm(vma)->xe; - pte |= XE_PAGE_PRESENT; if (likely(!xe_vma_read_only(vma))) pte |= XE_PAGE_RW; - pte |= pte_encode_pat_index(xe, pat_index, pt_level); + pte |= pte_encode_pat_index(pat_index, pt_level); pte |= pte_encode_ps(pt_level); if (unlikely(xe_vma_is_null(vma))) @@ -1155,7 +1362,7 @@ static u64 xelp_pte_encode_addr(struct xe_device *xe, u64 addr, pte = addr; pte |= XE_PAGE_PRESENT | XE_PAGE_RW; - pte |= pte_encode_pat_index(xe, pat_index, pt_level); + pte |= pte_encode_pat_index(pat_index, pt_level); pte |= pte_encode_ps(pt_level); if (devmem) @@ -1173,12 +1380,15 @@ static const struct xe_pt_ops xelp_pt_ops = { .pde_encode_bo = xelp_pde_encode_bo, }; +static void vm_destroy_work_func(struct work_struct *w); + /** * xe_vm_create_scratch() - Setup a scratch memory pagetable tree for the * given tile and vm. * @xe: xe device. * @tile: tile to set up for. * @vm: vm to set up for. + * @exec: The struct drm_exec object used to lock the vm resv. * * Sets up a pagetable tree with one page-table per level and a single * leaf PTE. All pagetable entries point to the single page-table or, @@ -1188,21 +1398,25 @@ static const struct xe_pt_ops xelp_pt_ops = { * Return: 0 on success, negative error code on error. */ static int xe_vm_create_scratch(struct xe_device *xe, struct xe_tile *tile, - struct xe_vm *vm) + struct xe_vm *vm, struct drm_exec *exec) { u8 id = tile->id; int i; for (i = MAX_HUGEPTE_LEVEL; i < vm->pt_root[id]->level; i++) { - vm->scratch_pt[id][i] = xe_pt_create(vm, tile, i); - if (IS_ERR(vm->scratch_pt[id][i])) - return PTR_ERR(vm->scratch_pt[id][i]); + vm->scratch_pt[id][i] = xe_pt_create(vm, tile, i, exec); + if (IS_ERR(vm->scratch_pt[id][i])) { + int err = PTR_ERR(vm->scratch_pt[id][i]); + vm->scratch_pt[id][i] = NULL; + return err; + } xe_pt_populate_empty(tile, vm, vm->scratch_pt[id][i]); } return 0; } +ALLOW_ERROR_INJECTION(xe_vm_create_scratch, ERRNO); static void xe_vm_free_scratch(struct xe_vm *vm) { @@ -1224,14 +1438,37 @@ static void xe_vm_free_scratch(struct xe_vm *vm) } } -struct xe_vm *xe_vm_create(struct xe_device *xe, u32 flags) +static void xe_vm_pt_destroy(struct xe_vm *vm) +{ + struct xe_tile *tile; + u8 id; + + xe_vm_assert_held(vm); + + for_each_tile(tile, vm->xe, id) { + if (vm->pt_root[id]) { + xe_pt_destroy(vm->pt_root[id], vm->flags, NULL); + vm->pt_root[id] = NULL; + } + } +} + +struct xe_vm *xe_vm_create(struct xe_device *xe, u32 flags, struct xe_file *xef) { struct drm_gem_object *vm_resv_obj; + struct xe_validation_ctx ctx; + struct drm_exec exec; struct xe_vm *vm; - int err, number_tiles = 0; + int err; struct xe_tile *tile; u8 id; + /* + * Since the GSCCS is not user-accessible, we don't expect a GSC VM to + * ever be in faulting mode. + */ + xe_assert(xe, !((flags & XE_VM_FLAG_GSC) && (flags & XE_VM_FLAG_FAULT_MODE))); + vm = kzalloc(sizeof(*vm), GFP_KERNEL); if (!vm) return ERR_PTR(-ENOMEM); @@ -1239,19 +1476,37 @@ struct xe_vm *xe_vm_create(struct xe_device *xe, u32 flags) vm->xe = xe; vm->size = 1ull << xe->info.va_bits; - vm->flags = flags; - init_rwsem(&vm->lock); + if (xef) + vm->xef = xe_file_get(xef); + /** + * GSC VMs are kernel-owned, only used for PXP ops and can sometimes be + * manipulated under the PXP mutex. However, the PXP mutex can be taken + * under a user-VM lock when the PXP session is started at exec_queue + * creation time. Those are different VMs and therefore there is no risk + * of deadlock, but we need to tell lockdep that this is the case or it + * will print a warning. + */ + if (flags & XE_VM_FLAG_GSC) { + static struct lock_class_key gsc_vm_key; + + __init_rwsem(&vm->lock, "gsc_vm", &gsc_vm_key); + } else { + init_rwsem(&vm->lock); + } mutex_init(&vm->snap_mutex); INIT_LIST_HEAD(&vm->rebind_list); INIT_LIST_HEAD(&vm->userptr.repin_list); INIT_LIST_HEAD(&vm->userptr.invalidated); - init_rwsem(&vm->userptr.notifier_lock); spin_lock_init(&vm->userptr.invalidated_lock); + ttm_lru_bulk_move_init(&vm->lru_bulk_move); + + INIT_WORK(&vm->destroy_work, vm_destroy_work_func); + INIT_LIST_HEAD(&vm->preempt.exec_queues); vm->preempt.min_run_period_ms = 10; /* FIXME: Wire up to uAPI */ @@ -1260,13 +1515,26 @@ struct xe_vm *xe_vm_create(struct xe_device *xe, u32 flags) vm->pt_ops = &xelp_pt_ops; - if (!(flags & XE_VM_FLAG_MIGRATION)) + /* + * Long-running workloads are not protected by the scheduler references. + * By design, run_job for long-running workloads returns NULL and the + * scheduler drops all the references of it, hence protecting the VM + * for this case is necessary. + */ + if (flags & XE_VM_FLAG_LR_MODE) { + INIT_WORK(&vm->preempt.rebind_work, preempt_rebind_work_func); xe_pm_runtime_get_noresume(xe); + INIT_LIST_HEAD(&vm->preempt.pm_activate_link); + } + + err = xe_svm_init(vm); + if (err) + goto err_no_resv; vm_resv_obj = drm_gpuvm_resv_object_alloc(&xe->drm); if (!vm_resv_obj) { err = -ENOMEM; - goto err_no_resv; + goto err_svm_fini; } drm_gpuvm_init(&vm->gpuvm, "Xe VM", DRM_GPUVM_RESV_PROTECTED, &xe->drm, @@ -1274,112 +1542,167 @@ struct xe_vm *xe_vm_create(struct xe_device *xe, u32 flags) drm_gem_object_put(vm_resv_obj); - err = dma_resv_lock_interruptible(xe_vm_resv(vm), NULL); - if (err) - goto err_close; - - if (IS_DGFX(xe) && xe->info.vram_flags & XE_VRAM_FLAGS_NEED64K) - vm->flags |= XE_VM_FLAG_64K; - - for_each_tile(tile, xe, id) { - if (flags & XE_VM_FLAG_MIGRATION && - tile->id != XE_VM_FLAG_TILE_ID(flags)) - continue; + err = 0; + xe_validation_guard(&ctx, &xe->val, &exec, (struct xe_val_flags) {.interruptible = true}, + err) { + err = xe_vm_drm_exec_lock(vm, &exec); + drm_exec_retry_on_contention(&exec); - vm->pt_root[id] = xe_pt_create(vm, tile, xe->info.vm_max_level); - if (IS_ERR(vm->pt_root[id])) { - err = PTR_ERR(vm->pt_root[id]); - vm->pt_root[id] = NULL; - goto err_unlock_close; - } - } + if (IS_DGFX(xe) && xe->info.vram_flags & XE_VRAM_FLAGS_NEED64K) + vm->flags |= XE_VM_FLAG_64K; - if (xe_vm_has_scratch(vm)) { for_each_tile(tile, xe, id) { - if (!vm->pt_root[id]) + if (flags & XE_VM_FLAG_MIGRATION && + tile->id != XE_VM_FLAG_TILE_ID(flags)) continue; - err = xe_vm_create_scratch(xe, tile, vm); + vm->pt_root[id] = xe_pt_create(vm, tile, xe->info.vm_max_level, + &exec); + if (IS_ERR(vm->pt_root[id])) { + err = PTR_ERR(vm->pt_root[id]); + vm->pt_root[id] = NULL; + xe_vm_pt_destroy(vm); + drm_exec_retry_on_contention(&exec); + xe_validation_retry_on_oom(&ctx, &err); + break; + } + } + if (err) + break; + + if (xe_vm_has_scratch(vm)) { + for_each_tile(tile, xe, id) { + if (!vm->pt_root[id]) + continue; + + err = xe_vm_create_scratch(xe, tile, vm, &exec); + if (err) { + xe_vm_free_scratch(vm); + xe_vm_pt_destroy(vm); + drm_exec_retry_on_contention(&exec); + xe_validation_retry_on_oom(&ctx, &err); + break; + } + } if (err) - goto err_unlock_close; + break; + vm->batch_invalidate_tlb = true; } - vm->batch_invalidate_tlb = true; - } - if (vm->flags & XE_VM_FLAG_LR_MODE) { - INIT_WORK(&vm->preempt.rebind_work, preempt_rebind_work_func); - vm->batch_invalidate_tlb = false; - } + if (vm->flags & XE_VM_FLAG_LR_MODE) { + INIT_WORK(&vm->preempt.rebind_work, preempt_rebind_work_func); + vm->batch_invalidate_tlb = false; + } - /* Fill pt_root after allocating scratch tables */ - for_each_tile(tile, xe, id) { - if (!vm->pt_root[id]) - continue; + /* Fill pt_root after allocating scratch tables */ + for_each_tile(tile, xe, id) { + if (!vm->pt_root[id]) + continue; - xe_pt_populate_empty(tile, vm, vm->pt_root[id]); + xe_pt_populate_empty(tile, vm, vm->pt_root[id]); + } } - dma_resv_unlock(xe_vm_resv(vm)); + if (err) + goto err_close; /* Kernel migration VM shouldn't have a circular loop.. */ if (!(flags & XE_VM_FLAG_MIGRATION)) { for_each_tile(tile, xe, id) { - struct xe_gt *gt = tile->primary_gt; - struct xe_vm *migrate_vm; struct xe_exec_queue *q; u32 create_flags = EXEC_QUEUE_FLAG_VM; if (!vm->pt_root[id]) continue; - migrate_vm = xe_migrate_get_vm(tile->migrate); - q = xe_exec_queue_create_class(xe, gt, migrate_vm, - XE_ENGINE_CLASS_COPY, - create_flags); - xe_vm_put(migrate_vm); + q = xe_exec_queue_create_bind(xe, tile, create_flags, 0); if (IS_ERR(q)) { err = PTR_ERR(q); goto err_close; } vm->q[id] = q; - number_tiles++; } } - if (number_tiles > 1) - vm->composite_fence_ctx = dma_fence_context_alloc(1); + if (xef && xe->info.has_asid) { + u32 asid; + + down_write(&xe->usm.lock); + err = xa_alloc_cyclic(&xe->usm.asid_to_vm, &asid, vm, + XA_LIMIT(1, XE_MAX_ASID - 1), + &xe->usm.next_asid, GFP_KERNEL); + up_write(&xe->usm.lock); + if (err < 0) + goto err_close; - mutex_lock(&xe->usm.lock); - if (flags & XE_VM_FLAG_FAULT_MODE) - xe->usm.num_vm_in_fault_mode++; - else if (!(flags & XE_VM_FLAG_MIGRATION)) - xe->usm.num_vm_in_non_fault_mode++; - mutex_unlock(&xe->usm.lock); + vm->usm.asid = asid; + } trace_xe_vm_create(vm); return vm; -err_unlock_close: - dma_resv_unlock(xe_vm_resv(vm)); err_close: xe_vm_close_and_put(vm); return ERR_PTR(err); +err_svm_fini: + if (flags & XE_VM_FLAG_FAULT_MODE) { + vm->size = 0; /* close the vm */ + xe_svm_fini(vm); + } err_no_resv: mutex_destroy(&vm->snap_mutex); for_each_tile(tile, xe, id) xe_range_fence_tree_fini(&vm->rftree[id]); + ttm_lru_bulk_move_fini(&xe->ttm, &vm->lru_bulk_move); + if (vm->xef) + xe_file_put(vm->xef); kfree(vm); - if (!(flags & XE_VM_FLAG_MIGRATION)) + if (flags & XE_VM_FLAG_LR_MODE) xe_pm_runtime_put(xe); return ERR_PTR(err); } static void xe_vm_close(struct xe_vm *vm) { + struct xe_device *xe = vm->xe; + bool bound; + int idx; + + bound = drm_dev_enter(&xe->drm, &idx); + down_write(&vm->lock); + if (xe_vm_in_fault_mode(vm)) + xe_svm_notifier_lock(vm); + vm->size = 0; + + if (!((vm->flags & XE_VM_FLAG_MIGRATION))) { + struct xe_tile *tile; + struct xe_gt *gt; + u8 id; + + /* Wait for pending binds */ + dma_resv_wait_timeout(xe_vm_resv(vm), + DMA_RESV_USAGE_BOOKKEEP, + false, MAX_SCHEDULE_TIMEOUT); + + if (bound) { + for_each_tile(tile, xe, id) + if (vm->pt_root[id]) + xe_pt_clear(xe, vm->pt_root[id]); + + for_each_gt(gt, xe, id) + xe_tlb_inval_vm(>->tlb_inval, vm); + } + } + + if (xe_vm_in_fault_mode(vm)) + xe_svm_notifier_unlock(vm); up_write(&vm->lock); + + if (bound) + drm_dev_exit(idx); } void xe_vm_close_and_put(struct xe_vm *vm) @@ -1394,13 +1717,24 @@ void xe_vm_close_and_put(struct xe_vm *vm) xe_assert(xe, !vm->preempt.num_exec_queues); xe_vm_close(vm); - if (xe_vm_in_preempt_fence_mode(vm)) + if (xe_vm_in_preempt_fence_mode(vm)) { + mutex_lock(&xe->rebind_resume_lock); + list_del_init(&vm->preempt.pm_activate_link); + mutex_unlock(&xe->rebind_resume_lock); flush_work(&vm->preempt.rebind_work); + } + if (xe_vm_in_fault_mode(vm)) + xe_svm_close(vm); down_write(&vm->lock); for_each_tile(tile, xe, id) { - if (vm->q[id]) + if (vm->q[id]) { + int i; + xe_exec_queue_last_fence_put(vm->q[id], vm); + for_each_tlb_inval(i) + xe_exec_queue_tlb_inval_last_fence_put(vm->q[id], vm, i); + } } up_write(&vm->lock); @@ -1418,9 +1752,9 @@ void xe_vm_close_and_put(struct xe_vm *vm) vma = gpuva_to_vma(gpuva); if (xe_vma_has_no_bo(vma)) { - down_read(&vm->userptr.notifier_lock); + xe_svm_notifier_lock(vm); vma->gpuva.flags |= XE_VMA_DESTROYED; - up_read(&vm->userptr.notifier_lock); + xe_svm_notifier_unlock(vm); } xe_vm_remove_vma(vm, vma); @@ -1444,13 +1778,7 @@ void xe_vm_close_and_put(struct xe_vm *vm) * destroy the pagetables immediately. */ xe_vm_free_scratch(vm); - - for_each_tile(tile, xe, id) { - if (vm->pt_root[id]) { - xe_pt_destroy(vm->pt_root[id], vm->flags, NULL); - vm->pt_root[id] = NULL; - } - } + xe_vm_pt_destroy(vm); xe_vm_unlock(vm); /* @@ -1464,14 +1792,11 @@ void xe_vm_close_and_put(struct xe_vm *vm) xe_vma_destroy_unlocked(vma); } - up_write(&vm->lock); + xe_svm_fini(vm); - mutex_lock(&xe->usm.lock); - if (vm->flags & XE_VM_FLAG_FAULT_MODE) - xe->usm.num_vm_in_fault_mode--; - else if (!(vm->flags & XE_VM_FLAG_MIGRATION)) - xe->usm.num_vm_in_non_fault_mode--; + up_write(&vm->lock); + down_write(&xe->usm.lock); if (vm->usm.asid) { void *lookup; @@ -1481,7 +1806,7 @@ void xe_vm_close_and_put(struct xe_vm *vm) lookup = xa_erase(&xe->usm.asid_to_vm, vm->usm.asid); xe_assert(xe, lookup == vm); } - mutex_unlock(&xe->usm.lock); + up_write(&xe->usm.lock); for_each_tile(tile, xe, id) xe_range_fence_tree_fini(&vm->rftree[id]); @@ -1489,9 +1814,10 @@ void xe_vm_close_and_put(struct xe_vm *vm) xe_vm_put(vm); } -static void xe_vm_free(struct drm_gpuvm *gpuvm) +static void vm_destroy_work_func(struct work_struct *w) { - struct xe_vm *vm = container_of(gpuvm, struct xe_vm, gpuvm); + struct xe_vm *vm = + container_of(w, struct xe_vm, destroy_work); struct xe_device *xe = vm->xe; struct xe_tile *tile; u8 id; @@ -1504,16 +1830,30 @@ static void xe_vm_free(struct drm_gpuvm *gpuvm) mutex_destroy(&vm->snap_mutex); - if (!(vm->flags & XE_VM_FLAG_MIGRATION)) + if (vm->flags & XE_VM_FLAG_LR_MODE) xe_pm_runtime_put(xe); for_each_tile(tile, xe, id) XE_WARN_ON(vm->pt_root[id]); trace_xe_vm_free(vm); + + ttm_lru_bulk_move_fini(&xe->ttm, &vm->lru_bulk_move); + + if (vm->xef) + xe_file_put(vm->xef); + kfree(vm); } +static void xe_vm_free(struct drm_gpuvm *gpuvm) +{ + struct xe_vm *vm = container_of(gpuvm, struct xe_vm, gpuvm); + + /* To destroy the VM we need to be able to sleep */ + queue_work(system_unbound_wq, &vm->destroy_work); +} + struct xe_vm *xe_vm_lookup(struct xe_file *xef, u32 id) { struct xe_vm *vm; @@ -1529,8 +1869,7 @@ struct xe_vm *xe_vm_lookup(struct xe_file *xef, u32 id) u64 xe_vm_pdp4_descriptor(struct xe_vm *vm, struct xe_tile *tile) { - return vm->pt_ops->pde_encode_bo(vm->pt_root[tile->id]->bo, 0, - tile_to_xe(tile)->pat.idx[XE_CACHE_WB]); + return vm->pt_ops->pde_encode_bo(vm->pt_root[tile->id]->bo, 0); } static struct xe_exec_queue * @@ -1539,167 +1878,6 @@ to_wait_exec_queue(struct xe_vm *vm, struct xe_exec_queue *q) return q ? q : vm->q[0]; } -static struct dma_fence * -xe_vm_unbind_vma(struct xe_vma *vma, struct xe_exec_queue *q, - struct xe_sync_entry *syncs, u32 num_syncs, - bool first_op, bool last_op) -{ - struct xe_vm *vm = xe_vma_vm(vma); - struct xe_exec_queue *wait_exec_queue = to_wait_exec_queue(vm, q); - struct xe_tile *tile; - struct dma_fence *fence = NULL; - struct dma_fence **fences = NULL; - struct dma_fence_array *cf = NULL; - int cur_fence = 0, i; - int number_tiles = hweight8(vma->tile_present); - int err; - u8 id; - - trace_xe_vma_unbind(vma); - - if (vma->ufence) { - struct xe_user_fence * const f = vma->ufence; - - if (!xe_sync_ufence_get_status(f)) - return ERR_PTR(-EBUSY); - - vma->ufence = NULL; - xe_sync_ufence_put(f); - } - - if (number_tiles > 1) { - fences = kmalloc_array(number_tiles, sizeof(*fences), - GFP_KERNEL); - if (!fences) - return ERR_PTR(-ENOMEM); - } - - for_each_tile(tile, vm->xe, id) { - if (!(vma->tile_present & BIT(id))) - goto next; - - fence = __xe_pt_unbind_vma(tile, vma, q ? q : vm->q[id], - first_op ? syncs : NULL, - first_op ? num_syncs : 0); - if (IS_ERR(fence)) { - err = PTR_ERR(fence); - goto err_fences; - } - - if (fences) - fences[cur_fence++] = fence; - -next: - if (q && vm->pt_root[id] && !list_empty(&q->multi_gt_list)) - q = list_next_entry(q, multi_gt_list); - } - - if (fences) { - cf = dma_fence_array_create(number_tiles, fences, - vm->composite_fence_ctx, - vm->composite_fence_seqno++, - false); - if (!cf) { - --vm->composite_fence_seqno; - err = -ENOMEM; - goto err_fences; - } - } - - fence = cf ? &cf->base : !fence ? - xe_exec_queue_last_fence_get(wait_exec_queue, vm) : fence; - if (last_op) { - for (i = 0; i < num_syncs; i++) - xe_sync_entry_signal(&syncs[i], fence); - } - - return fence; - -err_fences: - if (fences) { - while (cur_fence) - dma_fence_put(fences[--cur_fence]); - kfree(fences); - } - - return ERR_PTR(err); -} - -static struct dma_fence * -xe_vm_bind_vma(struct xe_vma *vma, struct xe_exec_queue *q, - struct xe_sync_entry *syncs, u32 num_syncs, - bool first_op, bool last_op) -{ - struct xe_tile *tile; - struct dma_fence *fence; - struct dma_fence **fences = NULL; - struct dma_fence_array *cf = NULL; - struct xe_vm *vm = xe_vma_vm(vma); - int cur_fence = 0, i; - int number_tiles = hweight8(vma->tile_mask); - int err; - u8 id; - - trace_xe_vma_bind(vma); - - if (number_tiles > 1) { - fences = kmalloc_array(number_tiles, sizeof(*fences), - GFP_KERNEL); - if (!fences) - return ERR_PTR(-ENOMEM); - } - - for_each_tile(tile, vm->xe, id) { - if (!(vma->tile_mask & BIT(id))) - goto next; - - fence = __xe_pt_bind_vma(tile, vma, q ? q : vm->q[id], - first_op ? syncs : NULL, - first_op ? num_syncs : 0, - vma->tile_present & BIT(id)); - if (IS_ERR(fence)) { - err = PTR_ERR(fence); - goto err_fences; - } - - if (fences) - fences[cur_fence++] = fence; - -next: - if (q && vm->pt_root[id] && !list_empty(&q->multi_gt_list)) - q = list_next_entry(q, multi_gt_list); - } - - if (fences) { - cf = dma_fence_array_create(number_tiles, fences, - vm->composite_fence_ctx, - vm->composite_fence_seqno++, - false); - if (!cf) { - --vm->composite_fence_seqno; - err = -ENOMEM; - goto err_fences; - } - } - - if (last_op) { - for (i = 0; i < num_syncs; i++) - xe_sync_entry_signal(&syncs[i], - cf ? &cf->base : fence); - } - - return cf ? &cf->base : fence; - -err_fences: - if (fences) { - while (cur_fence) - dma_fence_put(fences[--cur_fence]); - kfree(fences); - } - - return ERR_PTR(err); -} - static struct xe_user_fence * find_ufence_get(struct xe_sync_entry *syncs, u32 num_syncs) { @@ -1715,89 +1893,6 @@ find_ufence_get(struct xe_sync_entry *syncs, u32 num_syncs) return NULL; } -static int __xe_vm_bind(struct xe_vm *vm, struct xe_vma *vma, - struct xe_exec_queue *q, struct xe_sync_entry *syncs, - u32 num_syncs, bool immediate, bool first_op, - bool last_op) -{ - struct dma_fence *fence; - struct xe_exec_queue *wait_exec_queue = to_wait_exec_queue(vm, q); - struct xe_user_fence *ufence; - - xe_vm_assert_held(vm); - - ufence = find_ufence_get(syncs, num_syncs); - if (vma->ufence && ufence) - xe_sync_ufence_put(vma->ufence); - - vma->ufence = ufence ?: vma->ufence; - - if (immediate) { - fence = xe_vm_bind_vma(vma, q, syncs, num_syncs, first_op, - last_op); - if (IS_ERR(fence)) - return PTR_ERR(fence); - } else { - int i; - - xe_assert(vm->xe, xe_vm_in_fault_mode(vm)); - - fence = xe_exec_queue_last_fence_get(wait_exec_queue, vm); - if (last_op) { - for (i = 0; i < num_syncs; i++) - xe_sync_entry_signal(&syncs[i], fence); - } - } - - if (last_op) - xe_exec_queue_last_fence_set(wait_exec_queue, vm, fence); - dma_fence_put(fence); - - return 0; -} - -static int xe_vm_bind(struct xe_vm *vm, struct xe_vma *vma, struct xe_exec_queue *q, - struct xe_bo *bo, struct xe_sync_entry *syncs, - u32 num_syncs, bool immediate, bool first_op, - bool last_op) -{ - int err; - - xe_vm_assert_held(vm); - xe_bo_assert_held(bo); - - if (bo && immediate) { - err = xe_bo_validate(bo, vm, true); - if (err) - return err; - } - - return __xe_vm_bind(vm, vma, q, syncs, num_syncs, immediate, first_op, - last_op); -} - -static int xe_vm_unbind(struct xe_vm *vm, struct xe_vma *vma, - struct xe_exec_queue *q, struct xe_sync_entry *syncs, - u32 num_syncs, bool first_op, bool last_op) -{ - struct dma_fence *fence; - struct xe_exec_queue *wait_exec_queue = to_wait_exec_queue(vm, q); - - xe_vm_assert_held(vm); - xe_bo_assert_held(xe_vma_bo(vma)); - - fence = xe_vm_unbind_vma(vma, q, syncs, num_syncs, first_op, last_op); - if (IS_ERR(fence)) - return PTR_ERR(fence); - - xe_vma_destroy(vma, fence); - if (last_op) - xe_exec_queue_last_fence_set(wait_exec_queue, vm, fence); - dma_fence_put(fence); - - return 0; -} - #define ALL_DRM_XE_VM_CREATE_FLAGS (DRM_XE_VM_CREATE_FLAG_SCRATCH_PAGE | \ DRM_XE_VM_CREATE_FLAG_LR_MODE | \ DRM_XE_VM_CREATE_FLAG_FAULT_MODE) @@ -1808,16 +1903,16 @@ int xe_vm_create_ioctl(struct drm_device *dev, void *data, struct xe_device *xe = to_xe_device(dev); struct xe_file *xef = to_xe_file(file); struct drm_xe_vm_create *args = data; - struct xe_tile *tile; + struct xe_gt *wa_gt = xe_root_mmio_gt(xe); struct xe_vm *vm; - u32 id, asid; + u32 id; int err; u32 flags = 0; if (XE_IOCTL_DBG(xe, args->extensions)) return -EINVAL; - if (XE_WA(xe_root_mmio_gt(xe), 14016763929)) + if (wa_gt && XE_GT_WA(wa_gt, 22014953428)) args->flags |= DRM_XE_VM_CREATE_FLAG_SCRATCH_PAGE; if (XE_IOCTL_DBG(xe, args->flags & DRM_XE_VM_CREATE_FLAG_FAULT_MODE && @@ -1831,24 +1926,14 @@ int xe_vm_create_ioctl(struct drm_device *dev, void *data, return -EINVAL; if (XE_IOCTL_DBG(xe, args->flags & DRM_XE_VM_CREATE_FLAG_SCRATCH_PAGE && - args->flags & DRM_XE_VM_CREATE_FLAG_FAULT_MODE)) + args->flags & DRM_XE_VM_CREATE_FLAG_FAULT_MODE && + !xe->info.needs_scratch)) return -EINVAL; if (XE_IOCTL_DBG(xe, !(args->flags & DRM_XE_VM_CREATE_FLAG_LR_MODE) && args->flags & DRM_XE_VM_CREATE_FLAG_FAULT_MODE)) return -EINVAL; - if (XE_IOCTL_DBG(xe, args->flags & DRM_XE_VM_CREATE_FLAG_FAULT_MODE && - xe_device_in_non_fault_mode(xe))) - return -EINVAL; - - if (XE_IOCTL_DBG(xe, !(args->flags & DRM_XE_VM_CREATE_FLAG_FAULT_MODE) && - xe_device_in_fault_mode(xe))) - return -EINVAL; - - if (XE_IOCTL_DBG(xe, args->extensions)) - return -EINVAL; - if (args->flags & DRM_XE_VM_CREATE_FLAG_SCRATCH_PAGE) flags |= XE_VM_FLAG_SCRATCH_PAGE; if (args->flags & DRM_XE_VM_CREATE_FLAG_LR_MODE) @@ -1856,47 +1941,24 @@ int xe_vm_create_ioctl(struct drm_device *dev, void *data, if (args->flags & DRM_XE_VM_CREATE_FLAG_FAULT_MODE) flags |= XE_VM_FLAG_FAULT_MODE; - vm = xe_vm_create(xe, flags); + vm = xe_vm_create(xe, flags, xef); if (IS_ERR(vm)) return PTR_ERR(vm); - mutex_lock(&xef->vm.lock); +#if IS_ENABLED(CONFIG_DRM_XE_DEBUG_MEM) + /* Warning: Security issue - never enable by default */ + args->reserved[0] = xe_bo_main_addr(vm->pt_root[0]->bo, XE_PAGE_SIZE); +#endif + + /* user id alloc must always be last in ioctl to prevent UAF */ err = xa_alloc(&xef->vm.xa, &id, vm, xa_limit_32b, GFP_KERNEL); - mutex_unlock(&xef->vm.lock); if (err) goto err_close_and_put; - if (xe->info.has_asid) { - mutex_lock(&xe->usm.lock); - err = xa_alloc_cyclic(&xe->usm.asid_to_vm, &asid, vm, - XA_LIMIT(1, XE_MAX_ASID - 1), - &xe->usm.next_asid, GFP_KERNEL); - mutex_unlock(&xe->usm.lock); - if (err < 0) - goto err_free_id; - - vm->usm.asid = asid; - } - args->vm_id = id; - vm->xef = xef; - - /* Record BO memory for VM pagetable created against client */ - for_each_tile(tile, xe, id) - if (vm->pt_root[id]) - xe_drm_client_add_bo(vm->xef->client, vm->pt_root[id]->bo); - -#if IS_ENABLED(CONFIG_DRM_XE_DEBUG_MEM) - /* Warning: Security issue - never enable by default */ - args->reserved[0] = xe_bo_main_addr(vm->pt_root[0]->bo, XE_PAGE_SIZE); -#endif return 0; -err_free_id: - mutex_lock(&xef->vm.lock); - xa_erase(&xef->vm.xa, id); - mutex_unlock(&xef->vm.lock); err_close_and_put: xe_vm_close_and_put(vm); @@ -1932,55 +1994,151 @@ int xe_vm_destroy_ioctl(struct drm_device *dev, void *data, return err; } -static const u32 region_to_mem_type[] = { - XE_PL_TT, - XE_PL_VRAM0, - XE_PL_VRAM1, -}; +static int xe_vm_query_vmas(struct xe_vm *vm, u64 start, u64 end) +{ + struct drm_gpuva *gpuva; + u32 num_vmas = 0; + + lockdep_assert_held(&vm->lock); + drm_gpuvm_for_each_va_range(gpuva, &vm->gpuvm, start, end) + num_vmas++; + + return num_vmas; +} -static int xe_vm_prefetch(struct xe_vm *vm, struct xe_vma *vma, - struct xe_exec_queue *q, u32 region, - struct xe_sync_entry *syncs, u32 num_syncs, - bool first_op, bool last_op) +static int get_mem_attrs(struct xe_vm *vm, u32 *num_vmas, u64 start, + u64 end, struct drm_xe_mem_range_attr *attrs) { - struct xe_exec_queue *wait_exec_queue = to_wait_exec_queue(vm, q); - int err; + struct drm_gpuva *gpuva; + int i = 0; - xe_assert(vm->xe, region < ARRAY_SIZE(region_to_mem_type)); + lockdep_assert_held(&vm->lock); - if (!xe_vma_has_no_bo(vma)) { - err = xe_bo_migrate(xe_vma_bo(vma), region_to_mem_type[region]); - if (err) - return err; + drm_gpuvm_for_each_va_range(gpuva, &vm->gpuvm, start, end) { + struct xe_vma *vma = gpuva_to_vma(gpuva); + + if (i == *num_vmas) + return -ENOSPC; + + attrs[i].start = xe_vma_start(vma); + attrs[i].end = xe_vma_end(vma); + attrs[i].atomic.val = vma->attr.atomic_access; + attrs[i].pat_index.val = vma->attr.pat_index; + attrs[i].preferred_mem_loc.devmem_fd = vma->attr.preferred_loc.devmem_fd; + attrs[i].preferred_mem_loc.migration_policy = + vma->attr.preferred_loc.migration_policy; + + i++; } - if (vma->tile_mask != (vma->tile_present & ~vma->tile_invalidated)) { - return xe_vm_bind(vm, vma, q, xe_vma_bo(vma), syncs, num_syncs, - true, first_op, last_op); - } else { - int i; + *num_vmas = i; + return 0; +} - /* Nothing to do, signal fences now */ - if (last_op) { - for (i = 0; i < num_syncs; i++) { - struct dma_fence *fence = - xe_exec_queue_last_fence_get(wait_exec_queue, vm); +int xe_vm_query_vmas_attrs_ioctl(struct drm_device *dev, void *data, struct drm_file *file) +{ + struct xe_device *xe = to_xe_device(dev); + struct xe_file *xef = to_xe_file(file); + struct drm_xe_mem_range_attr *mem_attrs; + struct drm_xe_vm_query_mem_range_attr *args = data; + u64 __user *attrs_user = u64_to_user_ptr(args->vector_of_mem_attr); + struct xe_vm *vm; + int err = 0; - xe_sync_entry_signal(&syncs[i], fence); - dma_fence_put(fence); - } - } + if (XE_IOCTL_DBG(xe, + ((args->num_mem_ranges == 0 && + (attrs_user || args->sizeof_mem_range_attr != 0)) || + (args->num_mem_ranges > 0 && + (!attrs_user || + args->sizeof_mem_range_attr != + sizeof(struct drm_xe_mem_range_attr)))))) + return -EINVAL; - return 0; + vm = xe_vm_lookup(xef, args->vm_id); + if (XE_IOCTL_DBG(xe, !vm)) + return -EINVAL; + + err = down_read_interruptible(&vm->lock); + if (err) + goto put_vm; + + attrs_user = u64_to_user_ptr(args->vector_of_mem_attr); + + if (args->num_mem_ranges == 0 && !attrs_user) { + args->num_mem_ranges = xe_vm_query_vmas(vm, args->start, args->start + args->range); + args->sizeof_mem_range_attr = sizeof(struct drm_xe_mem_range_attr); + goto unlock_vm; } + + mem_attrs = kvmalloc_array(args->num_mem_ranges, args->sizeof_mem_range_attr, + GFP_KERNEL | __GFP_ACCOUNT | + __GFP_RETRY_MAYFAIL | __GFP_NOWARN); + if (!mem_attrs) { + err = args->num_mem_ranges > 1 ? -ENOBUFS : -ENOMEM; + goto unlock_vm; + } + + memset(mem_attrs, 0, args->num_mem_ranges * args->sizeof_mem_range_attr); + err = get_mem_attrs(vm, &args->num_mem_ranges, args->start, + args->start + args->range, mem_attrs); + if (err) + goto free_mem_attrs; + + err = copy_to_user(attrs_user, mem_attrs, + args->sizeof_mem_range_attr * args->num_mem_ranges); + if (err) + err = -EFAULT; + +free_mem_attrs: + kvfree(mem_attrs); +unlock_vm: + up_read(&vm->lock); +put_vm: + xe_vm_put(vm); + return err; +} + +static bool vma_matches(struct xe_vma *vma, u64 page_addr) +{ + if (page_addr > xe_vma_end(vma) - 1 || + page_addr + SZ_4K - 1 < xe_vma_start(vma)) + return false; + + return true; +} + +/** + * xe_vm_find_vma_by_addr() - Find a VMA by its address + * + * @vm: the xe_vm the vma belongs to + * @page_addr: address to look up + */ +struct xe_vma *xe_vm_find_vma_by_addr(struct xe_vm *vm, u64 page_addr) +{ + struct xe_vma *vma = NULL; + + if (vm->usm.last_fault_vma) { /* Fast lookup */ + if (vma_matches(vm->usm.last_fault_vma, page_addr)) + vma = vm->usm.last_fault_vma; + } + if (!vma) + vma = xe_vm_find_overlapping_vma(vm, page_addr, SZ_4K); + + return vma; } +static const u32 region_to_mem_type[] = { + XE_PL_TT, + XE_PL_VRAM0, + XE_PL_VRAM1, +}; + static void prep_vma_destroy(struct xe_vm *vm, struct xe_vma *vma, bool post_commit) { - down_read(&vm->userptr.notifier_lock); + xe_svm_notifier_lock(vm); vma->gpuva.flags |= XE_VMA_DESTROYED; - up_read(&vm->userptr.notifier_lock); + xe_svm_notifier_unlock(vm); if (post_commit) xe_vm_remove_vma(vm, vma); } @@ -2035,13 +2193,39 @@ static void print_op(struct xe_device *xe, struct drm_gpuva_op *op) } #endif +static bool __xe_vm_needs_clear_scratch_pages(struct xe_vm *vm, u32 bind_flags) +{ + if (!xe_vm_in_fault_mode(vm)) + return false; + + if (!xe_vm_has_scratch(vm)) + return false; + + if (bind_flags & DRM_XE_VM_BIND_FLAG_IMMEDIATE) + return false; + + return true; +} + +static void xe_svm_prefetch_gpuva_ops_fini(struct drm_gpuva_ops *ops) +{ + struct drm_gpuva_op *__op; + + drm_gpuva_for_each_op(__op, ops) { + struct xe_vma_op *op = gpuva_op_to_vma_op(__op); + + xe_vma_svm_prefetch_op_fini(op); + } +} + /* * Create operations list from IOCTL arguments, setup operations fields so parse * and commit steps are decoupled from IOCTL arguments. This step can fail. */ static struct drm_gpuva_ops * -vm_bind_ioctl_ops_create(struct xe_vm *vm, struct xe_bo *bo, - u64 bo_offset_or_userptr, u64 addr, u64 range, +vm_bind_ioctl_ops_create(struct xe_vm *vm, struct xe_vma_ops *vops, + struct xe_bo *bo, u64 bo_offset_or_userptr, + u64 addr, u64 range, u32 operation, u32 flags, u32 prefetch_region, u16 pat_index) { @@ -2049,6 +2233,7 @@ vm_bind_ioctl_ops_create(struct xe_vm *vm, struct xe_bo *bo, struct drm_gpuva_ops *ops; struct drm_gpuva_op *__op; struct drm_gpuvm_bo *vm_bo; + u64 range_end = addr + range; int err; lockdep_assert_held_write(&vm->lock); @@ -2060,10 +2245,17 @@ vm_bind_ioctl_ops_create(struct xe_vm *vm, struct xe_bo *bo, switch (operation) { case DRM_XE_VM_BIND_OP_MAP: - case DRM_XE_VM_BIND_OP_MAP_USERPTR: - ops = drm_gpuvm_sm_map_ops_create(&vm->gpuvm, addr, range, - obj, bo_offset_or_userptr); + case DRM_XE_VM_BIND_OP_MAP_USERPTR: { + struct drm_gpuvm_map_req map_req = { + .map.va.addr = addr, + .map.va.range = range, + .map.gem.obj = obj, + .map.gem.offset = bo_offset_or_userptr, + }; + + ops = drm_gpuvm_sm_map_ops_create(&vm->gpuvm, &map_req); break; + } case DRM_XE_VM_BIND_OP_UNMAP: ops = drm_gpuvm_sm_unmap_ops_create(&vm->gpuvm, addr, range); break; @@ -2100,35 +2292,128 @@ vm_bind_ioctl_ops_create(struct xe_vm *vm, struct xe_bo *bo, if (__op->op == DRM_GPUVA_OP_MAP) { op->map.immediate = flags & DRM_XE_VM_BIND_FLAG_IMMEDIATE; - op->map.read_only = - flags & DRM_XE_VM_BIND_FLAG_READONLY; - op->map.is_null = flags & DRM_XE_VM_BIND_FLAG_NULL; - op->map.dumpable = flags & DRM_XE_VM_BIND_FLAG_DUMPABLE; + if (flags & DRM_XE_VM_BIND_FLAG_READONLY) + op->map.vma_flags |= XE_VMA_READ_ONLY; + if (flags & DRM_XE_VM_BIND_FLAG_NULL) + op->map.vma_flags |= DRM_GPUVA_SPARSE; + if (flags & DRM_XE_VM_BIND_FLAG_CPU_ADDR_MIRROR) + op->map.vma_flags |= XE_VMA_SYSTEM_ALLOCATOR; + if (flags & DRM_XE_VM_BIND_FLAG_DUMPABLE) + op->map.vma_flags |= XE_VMA_DUMPABLE; + if (flags & DRM_XE_VM_BIND_FLAG_MADVISE_AUTORESET) + op->map.vma_flags |= XE_VMA_MADV_AUTORESET; op->map.pat_index = pat_index; + op->map.invalidate_on_bind = + __xe_vm_needs_clear_scratch_pages(vm, flags); } else if (__op->op == DRM_GPUVA_OP_PREFETCH) { - op->prefetch.region = prefetch_region; - } + struct xe_vma *vma = gpuva_to_vma(op->base.prefetch.va); + struct xe_tile *tile; + struct xe_svm_range *svm_range; + struct drm_gpusvm_ctx ctx = {}; + struct drm_pagemap *dpagemap; + u8 id, tile_mask = 0; + u32 i; + + if (!xe_vma_is_cpu_addr_mirror(vma)) { + op->prefetch.region = prefetch_region; + break; + } + + ctx.read_only = xe_vma_read_only(vma); + ctx.devmem_possible = IS_DGFX(vm->xe) && + IS_ENABLED(CONFIG_DRM_XE_PAGEMAP); + + for_each_tile(tile, vm->xe, id) + tile_mask |= 0x1 << id; + + xa_init_flags(&op->prefetch_range.range, XA_FLAGS_ALLOC); + op->prefetch_range.ranges_count = 0; + tile = NULL; + + if (prefetch_region == DRM_XE_CONSULT_MEM_ADVISE_PREF_LOC) { + dpagemap = xe_vma_resolve_pagemap(vma, + xe_device_get_root_tile(vm->xe)); + /* + * TODO: Once multigpu support is enabled will need + * something to dereference tile from dpagemap. + */ + if (dpagemap) + tile = xe_device_get_root_tile(vm->xe); + } else if (prefetch_region) { + tile = &vm->xe->tiles[region_to_mem_type[prefetch_region] - + XE_PL_VRAM0]; + } + + op->prefetch_range.tile = tile; +alloc_next_range: + svm_range = xe_svm_range_find_or_insert(vm, addr, vma, &ctx); + + if (PTR_ERR(svm_range) == -ENOENT) { + u64 ret = xe_svm_find_vma_start(vm, addr, range_end, vma); + + addr = ret == ULONG_MAX ? 0 : ret; + if (addr) + goto alloc_next_range; + else + goto print_op_label; + } + + if (IS_ERR(svm_range)) { + err = PTR_ERR(svm_range); + goto unwind_prefetch_ops; + } + + if (xe_svm_range_validate(vm, svm_range, tile_mask, !!tile)) { + xe_svm_range_debug(svm_range, "PREFETCH - RANGE IS VALID"); + goto check_next_range; + } + err = xa_alloc(&op->prefetch_range.range, + &i, svm_range, xa_limit_32b, + GFP_KERNEL); + + if (err) + goto unwind_prefetch_ops; + + op->prefetch_range.ranges_count++; + vops->flags |= XE_VMA_OPS_FLAG_HAS_SVM_PREFETCH; + xe_svm_range_debug(svm_range, "PREFETCH - RANGE CREATED"); +check_next_range: + if (range_end > xe_svm_range_end(svm_range) && + xe_svm_range_end(svm_range) < xe_vma_end(vma)) { + addr = xe_svm_range_end(svm_range); + goto alloc_next_range; + } + } +print_op_label: print_op(vm->xe, __op); } return ops; + +unwind_prefetch_ops: + xe_svm_prefetch_gpuva_ops_fini(ops); + drm_gpuva_ops_free(&vm->gpuvm, ops); + return ERR_PTR(err); } +ALLOW_ERROR_INJECTION(vm_bind_ioctl_ops_create, ERRNO); + static struct xe_vma *new_vma(struct xe_vm *vm, struct drm_gpuva_op_map *op, - u16 pat_index, unsigned int flags) + struct xe_vma_mem_attr *attr, unsigned int flags) { struct xe_bo *bo = op->gem.obj ? gem_to_xe_bo(op->gem.obj) : NULL; + struct xe_validation_ctx ctx; struct drm_exec exec; struct xe_vma *vma; - int err; + int err = 0; lockdep_assert_held_write(&vm->lock); if (bo) { - drm_exec_init(&exec, DRM_EXEC_INTERRUPTIBLE_WAIT, 0); - drm_exec_until_all_locked(&exec) { - err = 0; + err = 0; + xe_validation_guard(&ctx, &vm->xe->val, &exec, + (struct xe_val_flags) {.interruptible = true}, err) { if (!bo->vm) { err = drm_exec_lock_obj(&exec, xe_vm_obj(vm)); drm_exec_retry_on_contention(&exec); @@ -2137,32 +2422,39 @@ static struct xe_vma *new_vma(struct xe_vm *vm, struct drm_gpuva_op_map *op, err = drm_exec_lock_obj(&exec, &bo->ttm.base); drm_exec_retry_on_contention(&exec); } - if (err) { - drm_exec_fini(&exec); + if (err) return ERR_PTR(err); - } - } - } - vma = xe_vma_create(vm, bo, op->gem.offset, - op->va.addr, op->va.addr + - op->va.range - 1, pat_index, flags); - if (bo) - drm_exec_fini(&exec); - if (xe_vma_is_userptr(vma)) { - err = xe_vma_userptr_pin_pages(to_userptr_vma(vma)); - if (err) { - prep_vma_destroy(vm, vma, false); - xe_vma_destroy_unlocked(vma); - return ERR_PTR(err); + vma = xe_vma_create(vm, bo, op->gem.offset, + op->va.addr, op->va.addr + + op->va.range - 1, attr, flags); + if (IS_ERR(vma)) + return vma; + + if (!bo->vm) { + err = add_preempt_fences(vm, bo); + if (err) { + prep_vma_destroy(vm, vma, false); + xe_vma_destroy(vma, NULL); + } + } } - } else if (!xe_vma_has_no_bo(vma) && !bo->vm) { - err = add_preempt_fences(vm, bo); - if (err) { - prep_vma_destroy(vm, vma, false); - xe_vma_destroy_unlocked(vma); + if (err) return ERR_PTR(err); - } + } else { + vma = xe_vma_create(vm, NULL, op->gem.offset, + op->va.addr, op->va.addr + + op->va.range - 1, attr, flags); + if (IS_ERR(vma)) + return vma; + + if (xe_vma_is_userptr(vma)) + err = xe_vma_userptr_pin_pages(to_userptr_vma(vma)); + } + if (err) { + prep_vma_destroy(vm, vma, false); + xe_vma_destroy_unlocked(vma); + vma = ERR_PTR(err); } return vma; @@ -2242,7 +2534,7 @@ static int xe_vma_op_commit(struct xe_vm *vm, struct xe_vma_op *op) } } - /* Adjust for partial unbind after removin VMA from VM */ + /* Adjust for partial unbind after removing VMA from VM */ if (!err) { op->base.remap.unmap->va->va.addr = op->remap.start; op->base.remap.unmap->va->va.range = op->remap.range; @@ -2263,75 +2555,109 @@ static int xe_vma_op_commit(struct xe_vm *vm, struct xe_vma_op *op) return err; } +/** + * xe_vma_has_default_mem_attrs - Check if a VMA has default memory attributes + * @vma: Pointer to the xe_vma structure to check + * + * This function determines whether the given VMA (Virtual Memory Area) + * has its memory attributes set to their default values. Specifically, + * it checks the following conditions: + * + * - `atomic_access` is `DRM_XE_VMA_ATOMIC_UNDEFINED` + * - `pat_index` is equal to `default_pat_index` + * - `preferred_loc.devmem_fd` is `DRM_XE_PREFERRED_LOC_DEFAULT_DEVICE` + * - `preferred_loc.migration_policy` is `DRM_XE_MIGRATE_ALL_PAGES` + * + * Return: true if all attributes are at their default values, false otherwise. + */ +bool xe_vma_has_default_mem_attrs(struct xe_vma *vma) +{ + return (vma->attr.atomic_access == DRM_XE_ATOMIC_UNDEFINED && + vma->attr.pat_index == vma->attr.default_pat_index && + vma->attr.preferred_loc.devmem_fd == DRM_XE_PREFERRED_LOC_DEFAULT_DEVICE && + vma->attr.preferred_loc.migration_policy == DRM_XE_MIGRATE_ALL_PAGES); +} -static int vm_bind_ioctl_ops_parse(struct xe_vm *vm, struct xe_exec_queue *q, - struct drm_gpuva_ops *ops, - struct xe_sync_entry *syncs, u32 num_syncs, - struct list_head *ops_list, bool last) +static int vm_bind_ioctl_ops_parse(struct xe_vm *vm, struct drm_gpuva_ops *ops, + struct xe_vma_ops *vops) { struct xe_device *xe = vm->xe; - struct xe_vma_op *last_op = NULL; struct drm_gpuva_op *__op; + struct xe_tile *tile; + u8 id, tile_mask = 0; int err = 0; lockdep_assert_held_write(&vm->lock); + for_each_tile(tile, vm->xe, id) + tile_mask |= 0x1 << id; + drm_gpuva_for_each_op(__op, ops) { struct xe_vma_op *op = gpuva_op_to_vma_op(__op); struct xe_vma *vma; - bool first = list_empty(ops_list); unsigned int flags = 0; INIT_LIST_HEAD(&op->link); - list_add_tail(&op->link, ops_list); - - if (first) { - op->flags |= XE_VMA_OP_FIRST; - op->num_syncs = num_syncs; - op->syncs = syncs; - } - - op->q = q; + list_add_tail(&op->link, &vops->list); + op->tile_mask = tile_mask; switch (op->base.op) { case DRM_GPUVA_OP_MAP: { - flags |= op->map.read_only ? - VMA_CREATE_FLAG_READ_ONLY : 0; - flags |= op->map.is_null ? - VMA_CREATE_FLAG_IS_NULL : 0; - flags |= op->map.dumpable ? - VMA_CREATE_FLAG_DUMPABLE : 0; - - vma = new_vma(vm, &op->base.map, op->map.pat_index, + struct xe_vma_mem_attr default_attr = { + .preferred_loc = { + .devmem_fd = DRM_XE_PREFERRED_LOC_DEFAULT_DEVICE, + .migration_policy = DRM_XE_MIGRATE_ALL_PAGES, + }, + .atomic_access = DRM_XE_ATOMIC_UNDEFINED, + .default_pat_index = op->map.pat_index, + .pat_index = op->map.pat_index, + }; + + flags |= op->map.vma_flags & XE_VMA_CREATE_MASK; + + vma = new_vma(vm, &op->base.map, &default_attr, flags); if (IS_ERR(vma)) return PTR_ERR(vma); op->map.vma = vma; + if (((op->map.immediate || !xe_vm_in_fault_mode(vm)) && + !(op->map.vma_flags & XE_VMA_SYSTEM_ALLOCATOR)) || + op->map.invalidate_on_bind) + xe_vma_ops_incr_pt_update_ops(vops, + op->tile_mask, 1); break; } case DRM_GPUVA_OP_REMAP: { struct xe_vma *old = gpuva_to_vma(op->base.remap.unmap->va); + bool skip = xe_vma_is_cpu_addr_mirror(old); + u64 start = xe_vma_start(old), end = xe_vma_end(old); + int num_remap_ops = 0; + + if (op->base.remap.prev) + start = op->base.remap.prev->va.addr + + op->base.remap.prev->va.range; + if (op->base.remap.next) + end = op->base.remap.next->va.addr; + + if (xe_vma_is_cpu_addr_mirror(old) && + xe_svm_has_mapping(vm, start, end)) { + if (vops->flags & XE_VMA_OPS_FLAG_MADVISE) + xe_svm_unmap_address_range(vm, start, end); + else + return -EBUSY; + } op->remap.start = xe_vma_start(old); op->remap.range = xe_vma_size(old); + flags |= op->base.remap.unmap->va->flags & XE_VMA_CREATE_MASK; if (op->base.remap.prev) { - flags |= op->base.remap.unmap->va->flags & - XE_VMA_READ_ONLY ? - VMA_CREATE_FLAG_READ_ONLY : 0; - flags |= op->base.remap.unmap->va->flags & - DRM_GPUVA_SPARSE ? - VMA_CREATE_FLAG_IS_NULL : 0; - flags |= op->base.remap.unmap->va->flags & - XE_VMA_DUMPABLE ? - VMA_CREATE_FLAG_DUMPABLE : 0; - vma = new_vma(vm, op->base.remap.prev, - old->pat_index, flags); + &old->attr, flags); if (IS_ERR(vma)) return PTR_ERR(vma); @@ -2341,9 +2667,10 @@ static int vm_bind_ioctl_ops_parse(struct xe_vm *vm, struct xe_exec_queue *q, * Userptr creates a new SG mapping so * we must also rebind. */ - op->remap.skip_prev = !xe_vma_is_userptr(old) && + op->remap.skip_prev = skip || + (!xe_vma_is_userptr(old) && IS_ALIGNED(xe_vma_end(vma), - xe_vma_max_pte_size(old)); + xe_vma_max_pte_size(old))); if (op->remap.skip_prev) { xe_vma_set_pte_size(vma, xe_vma_max_pte_size(old)); op->remap.range -= @@ -2353,22 +2680,14 @@ static int vm_bind_ioctl_ops_parse(struct xe_vm *vm, struct xe_exec_queue *q, vm_dbg(&xe->drm, "REMAP:SKIP_PREV: addr=0x%016llx, range=0x%016llx", (ULL)op->remap.start, (ULL)op->remap.range); + } else { + num_remap_ops++; } } if (op->base.remap.next) { - flags |= op->base.remap.unmap->va->flags & - XE_VMA_READ_ONLY ? - VMA_CREATE_FLAG_READ_ONLY : 0; - flags |= op->base.remap.unmap->va->flags & - DRM_GPUVA_SPARSE ? - VMA_CREATE_FLAG_IS_NULL : 0; - flags |= op->base.remap.unmap->va->flags & - XE_VMA_DUMPABLE ? - VMA_CREATE_FLAG_DUMPABLE : 0; - vma = new_vma(vm, op->base.remap.next, - old->pat_index, flags); + &old->attr, flags); if (IS_ERR(vma)) return PTR_ERR(vma); @@ -2378,9 +2697,10 @@ static int vm_bind_ioctl_ops_parse(struct xe_vm *vm, struct xe_exec_queue *q, * Userptr creates a new SG mapping so * we must also rebind. */ - op->remap.skip_next = !xe_vma_is_userptr(old) && + op->remap.skip_next = skip || + (!xe_vma_is_userptr(old) && IS_ALIGNED(xe_vma_start(vma), - xe_vma_max_pte_size(old)); + xe_vma_max_pte_size(old))); if (op->remap.skip_next) { xe_vma_set_pte_size(vma, xe_vma_max_pte_size(old)); op->remap.range -= @@ -2389,343 +2709,608 @@ static int vm_bind_ioctl_ops_parse(struct xe_vm *vm, struct xe_exec_queue *q, vm_dbg(&xe->drm, "REMAP:SKIP_NEXT: addr=0x%016llx, range=0x%016llx", (ULL)op->remap.start, (ULL)op->remap.range); + } else { + num_remap_ops++; } } + if (!skip) + num_remap_ops++; + + xe_vma_ops_incr_pt_update_ops(vops, op->tile_mask, num_remap_ops); break; } case DRM_GPUVA_OP_UNMAP: + vma = gpuva_to_vma(op->base.unmap.va); + + if (xe_vma_is_cpu_addr_mirror(vma) && + xe_svm_has_mapping(vm, xe_vma_start(vma), + xe_vma_end(vma))) + return -EBUSY; + + if (!xe_vma_is_cpu_addr_mirror(vma)) + xe_vma_ops_incr_pt_update_ops(vops, op->tile_mask, 1); + break; case DRM_GPUVA_OP_PREFETCH: - /* Nothing to do */ + vma = gpuva_to_vma(op->base.prefetch.va); + + if (xe_vma_is_userptr(vma)) { + err = xe_vma_userptr_pin_pages(to_userptr_vma(vma)); + if (err) + return err; + } + + if (xe_vma_is_cpu_addr_mirror(vma)) + xe_vma_ops_incr_pt_update_ops(vops, op->tile_mask, + op->prefetch_range.ranges_count); + else + xe_vma_ops_incr_pt_update_ops(vops, op->tile_mask, 1); + break; default: drm_warn(&vm->xe->drm, "NOT POSSIBLE"); } - last_op = op; - err = xe_vma_op_commit(vm, op); if (err) return err; } - /* FIXME: Unhandled corner case */ - XE_WARN_ON(!last_op && last && !list_empty(ops_list)); - - if (!last_op) - return 0; - - last_op->ops = ops; - if (last) { - last_op->flags |= XE_VMA_OP_LAST; - last_op->num_syncs = num_syncs; - last_op->syncs = syncs; - } - return 0; } -static int op_execute(struct drm_exec *exec, struct xe_vm *vm, - struct xe_vma *vma, struct xe_vma_op *op) +static void xe_vma_op_unwind(struct xe_vm *vm, struct xe_vma_op *op, + bool post_commit, bool prev_post_commit, + bool next_post_commit) { - int err; - lockdep_assert_held_write(&vm->lock); - err = xe_vm_lock_vma(exec, vma); - if (err) - return err; - - xe_vm_assert_held(vm); - xe_bo_assert_held(xe_vma_bo(vma)); - switch (op->base.op) { case DRM_GPUVA_OP_MAP: - err = xe_vm_bind(vm, vma, op->q, xe_vma_bo(vma), - op->syncs, op->num_syncs, - op->map.immediate || !xe_vm_in_fault_mode(vm), - op->flags & XE_VMA_OP_FIRST, - op->flags & XE_VMA_OP_LAST); + if (op->map.vma) { + prep_vma_destroy(vm, op->map.vma, post_commit); + xe_vma_destroy_unlocked(op->map.vma); + } break; - case DRM_GPUVA_OP_REMAP: + case DRM_GPUVA_OP_UNMAP: { - bool prev = !!op->remap.prev; - bool next = !!op->remap.next; - - if (!op->remap.unmap_done) { - if (prev || next) - vma->gpuva.flags |= XE_VMA_FIRST_REBIND; - err = xe_vm_unbind(vm, vma, op->q, op->syncs, - op->num_syncs, - op->flags & XE_VMA_OP_FIRST, - op->flags & XE_VMA_OP_LAST && - !prev && !next); - if (err) - break; - op->remap.unmap_done = true; - } + struct xe_vma *vma = gpuva_to_vma(op->base.unmap.va); - if (prev) { - op->remap.prev->gpuva.flags |= XE_VMA_LAST_REBIND; - err = xe_vm_bind(vm, op->remap.prev, op->q, - xe_vma_bo(op->remap.prev), op->syncs, - op->num_syncs, true, false, - op->flags & XE_VMA_OP_LAST && !next); - op->remap.prev->gpuva.flags &= ~XE_VMA_LAST_REBIND; - if (err) - break; - op->remap.prev = NULL; + if (vma) { + xe_svm_notifier_lock(vm); + vma->gpuva.flags &= ~XE_VMA_DESTROYED; + xe_svm_notifier_unlock(vm); + if (post_commit) + xe_vm_insert_vma(vm, vma); } + break; + } + case DRM_GPUVA_OP_REMAP: + { + struct xe_vma *vma = gpuva_to_vma(op->base.remap.unmap->va); - if (next) { - op->remap.next->gpuva.flags |= XE_VMA_LAST_REBIND; - err = xe_vm_bind(vm, op->remap.next, op->q, - xe_vma_bo(op->remap.next), - op->syncs, op->num_syncs, - true, false, - op->flags & XE_VMA_OP_LAST); - op->remap.next->gpuva.flags &= ~XE_VMA_LAST_REBIND; - if (err) - break; - op->remap.next = NULL; + if (op->remap.prev) { + prep_vma_destroy(vm, op->remap.prev, prev_post_commit); + xe_vma_destroy_unlocked(op->remap.prev); + } + if (op->remap.next) { + prep_vma_destroy(vm, op->remap.next, next_post_commit); + xe_vma_destroy_unlocked(op->remap.next); + } + if (vma) { + xe_svm_notifier_lock(vm); + vma->gpuva.flags &= ~XE_VMA_DESTROYED; + xe_svm_notifier_unlock(vm); + if (post_commit) + xe_vm_insert_vma(vm, vma); } - break; } - case DRM_GPUVA_OP_UNMAP: - err = xe_vm_unbind(vm, vma, op->q, op->syncs, - op->num_syncs, op->flags & XE_VMA_OP_FIRST, - op->flags & XE_VMA_OP_LAST); - break; case DRM_GPUVA_OP_PREFETCH: - err = xe_vm_prefetch(vm, vma, op->q, op->prefetch.region, - op->syncs, op->num_syncs, - op->flags & XE_VMA_OP_FIRST, - op->flags & XE_VMA_OP_LAST); + /* Nothing to do */ break; default: drm_warn(&vm->xe->drm, "NOT POSSIBLE"); } +} - if (err) - trace_xe_vma_fail(vma); +static void vm_bind_ioctl_ops_unwind(struct xe_vm *vm, + struct drm_gpuva_ops **ops, + int num_ops_list) +{ + int i; + + for (i = num_ops_list - 1; i >= 0; --i) { + struct drm_gpuva_ops *__ops = ops[i]; + struct drm_gpuva_op *__op; + + if (!__ops) + continue; + + drm_gpuva_for_each_op_reverse(__op, __ops) { + struct xe_vma_op *op = gpuva_op_to_vma_op(__op); + + xe_vma_op_unwind(vm, op, + op->flags & XE_VMA_OP_COMMITTED, + op->flags & XE_VMA_OP_PREV_COMMITTED, + op->flags & XE_VMA_OP_NEXT_COMMITTED); + } + } +} + +static int vma_lock_and_validate(struct drm_exec *exec, struct xe_vma *vma, + bool res_evict, bool validate) +{ + struct xe_bo *bo = xe_vma_bo(vma); + struct xe_vm *vm = xe_vma_vm(vma); + int err = 0; + + if (bo) { + if (!bo->vm) + err = drm_exec_lock_obj(exec, &bo->ttm.base); + if (!err && validate) + err = xe_bo_validate(bo, vm, + !xe_vm_in_preempt_fence_mode(vm) && + res_evict, exec); + } return err; } -static int __xe_vma_op_execute(struct xe_vm *vm, struct xe_vma *vma, - struct xe_vma_op *op) +static int check_ufence(struct xe_vma *vma) { - struct drm_exec exec; - int err; + if (vma->ufence) { + struct xe_user_fence * const f = vma->ufence; -retry_userptr: - drm_exec_init(&exec, DRM_EXEC_INTERRUPTIBLE_WAIT, 0); - drm_exec_until_all_locked(&exec) { - err = op_execute(&exec, vm, vma, op); - drm_exec_retry_on_contention(&exec); - if (err) - break; + if (!xe_sync_ufence_get_status(f)) + return -EBUSY; + + vma->ufence = NULL; + xe_sync_ufence_put(f); } - drm_exec_fini(&exec); - if (err == -EAGAIN) { - lockdep_assert_held_write(&vm->lock); + return 0; +} - if (op->base.op == DRM_GPUVA_OP_REMAP) { - if (!op->remap.unmap_done) - vma = gpuva_to_vma(op->base.remap.unmap->va); - else if (op->remap.prev) - vma = op->remap.prev; - else - vma = op->remap.next; - } +static int prefetch_ranges(struct xe_vm *vm, struct xe_vma_op *op) +{ + bool devmem_possible = IS_DGFX(vm->xe) && IS_ENABLED(CONFIG_DRM_XE_PAGEMAP); + struct xe_vma *vma = gpuva_to_vma(op->base.prefetch.va); + struct xe_tile *tile = op->prefetch_range.tile; + int err = 0; - if (xe_vma_is_userptr(vma)) { - err = xe_vma_userptr_pin_pages(to_userptr_vma(vma)); - if (!err) - goto retry_userptr; + struct xe_svm_range *svm_range; + struct drm_gpusvm_ctx ctx = {}; + unsigned long i; + + if (!xe_vma_is_cpu_addr_mirror(vma)) + return 0; + + ctx.read_only = xe_vma_read_only(vma); + ctx.devmem_possible = devmem_possible; + ctx.check_pages_threshold = devmem_possible ? SZ_64K : 0; + ctx.device_private_page_owner = xe_svm_devm_owner(vm->xe); - trace_xe_vma_fail(vma); + /* TODO: Threading the migration */ + xa_for_each(&op->prefetch_range.range, i, svm_range) { + if (!tile) + xe_svm_range_migrate_to_smem(vm, svm_range); + + if (xe_svm_range_needs_migrate_to_vram(svm_range, vma, !!tile)) { + err = xe_svm_alloc_vram(tile, svm_range, &ctx); + if (err) { + drm_dbg(&vm->xe->drm, "VRAM allocation failed, retry from userspace, asid=%u, gpusvm=%p, errno=%pe\n", + vm->usm.asid, &vm->svm.gpusvm, ERR_PTR(err)); + return -ENODATA; + } + xe_svm_range_debug(svm_range, "PREFETCH - RANGE MIGRATED TO VRAM"); } + + err = xe_svm_range_get_pages(vm, svm_range, &ctx); + if (err) { + drm_dbg(&vm->xe->drm, "Get pages failed, asid=%u, gpusvm=%p, errno=%pe\n", + vm->usm.asid, &vm->svm.gpusvm, ERR_PTR(err)); + if (err == -EOPNOTSUPP || err == -EFAULT || err == -EPERM) + err = -ENODATA; + return err; + } + xe_svm_range_debug(svm_range, "PREFETCH - RANGE GET PAGES DONE"); } return err; } -static int xe_vma_op_execute(struct xe_vm *vm, struct xe_vma_op *op) +static int op_lock_and_prep(struct drm_exec *exec, struct xe_vm *vm, + struct xe_vma_ops *vops, struct xe_vma_op *op) { - int ret = 0; + int err = 0; + bool res_evict; - lockdep_assert_held_write(&vm->lock); + /* + * We only allow evicting a BO within the VM if it is not part of an + * array of binds, as an array of binds can evict another BO within the + * bind. + */ + res_evict = !(vops->flags & XE_VMA_OPS_ARRAY_OF_BINDS); switch (op->base.op) { case DRM_GPUVA_OP_MAP: - ret = __xe_vma_op_execute(vm, op->map.vma, op); + if (!op->map.invalidate_on_bind) + err = vma_lock_and_validate(exec, op->map.vma, + res_evict, + !xe_vm_in_fault_mode(vm) || + op->map.immediate); break; case DRM_GPUVA_OP_REMAP: - { - struct xe_vma *vma; - - if (!op->remap.unmap_done) - vma = gpuva_to_vma(op->base.remap.unmap->va); - else if (op->remap.prev) - vma = op->remap.prev; - else - vma = op->remap.next; + err = check_ufence(gpuva_to_vma(op->base.remap.unmap->va)); + if (err) + break; - ret = __xe_vma_op_execute(vm, vma, op); + err = vma_lock_and_validate(exec, + gpuva_to_vma(op->base.remap.unmap->va), + res_evict, false); + if (!err && op->remap.prev) + err = vma_lock_and_validate(exec, op->remap.prev, + res_evict, true); + if (!err && op->remap.next) + err = vma_lock_and_validate(exec, op->remap.next, + res_evict, true); break; - } case DRM_GPUVA_OP_UNMAP: - ret = __xe_vma_op_execute(vm, gpuva_to_vma(op->base.unmap.va), - op); + err = check_ufence(gpuva_to_vma(op->base.unmap.va)); + if (err) + break; + + err = vma_lock_and_validate(exec, + gpuva_to_vma(op->base.unmap.va), + res_evict, false); break; case DRM_GPUVA_OP_PREFETCH: - ret = __xe_vma_op_execute(vm, - gpuva_to_vma(op->base.prefetch.va), - op); + { + struct xe_vma *vma = gpuva_to_vma(op->base.prefetch.va); + u32 region; + + if (!xe_vma_is_cpu_addr_mirror(vma)) { + region = op->prefetch.region; + xe_assert(vm->xe, region == DRM_XE_CONSULT_MEM_ADVISE_PREF_LOC || + region <= ARRAY_SIZE(region_to_mem_type)); + } + + err = vma_lock_and_validate(exec, + gpuva_to_vma(op->base.prefetch.va), + res_evict, false); + if (!err && !xe_vma_has_no_bo(vma)) + err = xe_bo_migrate(xe_vma_bo(vma), + region_to_mem_type[region], + NULL, + exec); break; + } default: drm_warn(&vm->xe->drm, "NOT POSSIBLE"); } - return ret; + return err; } -static void xe_vma_op_cleanup(struct xe_vm *vm, struct xe_vma_op *op) +static int vm_bind_ioctl_ops_prefetch_ranges(struct xe_vm *vm, struct xe_vma_ops *vops) { - bool last = op->flags & XE_VMA_OP_LAST; + struct xe_vma_op *op; + int err; - if (last) { - while (op->num_syncs--) - xe_sync_entry_cleanup(&op->syncs[op->num_syncs]); - kfree(op->syncs); - if (op->q) - xe_exec_queue_put(op->q); + if (!(vops->flags & XE_VMA_OPS_FLAG_HAS_SVM_PREFETCH)) + return 0; + + list_for_each_entry(op, &vops->list, link) { + if (op->base.op == DRM_GPUVA_OP_PREFETCH) { + err = prefetch_ranges(vm, op); + if (err) + return err; + } } - if (!list_empty(&op->link)) - list_del(&op->link); - if (op->ops) - drm_gpuva_ops_free(&vm->gpuvm, op->ops); - if (last) - xe_vm_put(vm); + + return 0; } -static void xe_vma_op_unwind(struct xe_vm *vm, struct xe_vma_op *op, - bool post_commit, bool prev_post_commit, - bool next_post_commit) +static int vm_bind_ioctl_ops_lock_and_prep(struct drm_exec *exec, + struct xe_vm *vm, + struct xe_vma_ops *vops) { - lockdep_assert_held_write(&vm->lock); + struct xe_vma_op *op; + int err; + err = drm_exec_lock_obj(exec, xe_vm_obj(vm)); + if (err) + return err; + + list_for_each_entry(op, &vops->list, link) { + err = op_lock_and_prep(exec, vm, vops, op); + if (err) + return err; + } + +#ifdef TEST_VM_OPS_ERROR + if (vops->inject_error && + vm->xe->vm_inject_error_position == FORCE_OP_ERROR_LOCK) + return -ENOSPC; +#endif + + return 0; +} + +static void op_trace(struct xe_vma_op *op) +{ switch (op->base.op) { case DRM_GPUVA_OP_MAP: - if (op->map.vma) { - prep_vma_destroy(vm, op->map.vma, post_commit); - xe_vma_destroy_unlocked(op->map.vma); - } + trace_xe_vma_bind(op->map.vma); + break; + case DRM_GPUVA_OP_REMAP: + trace_xe_vma_unbind(gpuva_to_vma(op->base.remap.unmap->va)); + if (op->remap.prev) + trace_xe_vma_bind(op->remap.prev); + if (op->remap.next) + trace_xe_vma_bind(op->remap.next); break; case DRM_GPUVA_OP_UNMAP: - { - struct xe_vma *vma = gpuva_to_vma(op->base.unmap.va); - - if (vma) { - down_read(&vm->userptr.notifier_lock); - vma->gpuva.flags &= ~XE_VMA_DESTROYED; - up_read(&vm->userptr.notifier_lock); - if (post_commit) - xe_vm_insert_vma(vm, vma); - } + trace_xe_vma_unbind(gpuva_to_vma(op->base.unmap.va)); + break; + case DRM_GPUVA_OP_PREFETCH: + trace_xe_vma_bind(gpuva_to_vma(op->base.prefetch.va)); + break; + case DRM_GPUVA_OP_DRIVER: break; + default: + XE_WARN_ON("NOT POSSIBLE"); } - case DRM_GPUVA_OP_REMAP: - { - struct xe_vma *vma = gpuva_to_vma(op->base.remap.unmap->va); +} - if (op->remap.prev) { - prep_vma_destroy(vm, op->remap.prev, prev_post_commit); - xe_vma_destroy_unlocked(op->remap.prev); - } - if (op->remap.next) { - prep_vma_destroy(vm, op->remap.next, next_post_commit); - xe_vma_destroy_unlocked(op->remap.next); +static void trace_xe_vm_ops_execute(struct xe_vma_ops *vops) +{ + struct xe_vma_op *op; + + list_for_each_entry(op, &vops->list, link) + op_trace(op); +} + +static int vm_ops_setup_tile_args(struct xe_vm *vm, struct xe_vma_ops *vops) +{ + struct xe_exec_queue *q = vops->q; + struct xe_tile *tile; + int number_tiles = 0; + u8 id; + + for_each_tile(tile, vm->xe, id) { + if (vops->pt_update_ops[id].num_ops) + ++number_tiles; + + if (vops->pt_update_ops[id].q) + continue; + + if (q) { + vops->pt_update_ops[id].q = q; + if (vm->pt_root[id] && !list_empty(&q->multi_gt_list)) + q = list_next_entry(q, multi_gt_list); + } else { + vops->pt_update_ops[id].q = vm->q[id]; } - if (vma) { - down_read(&vm->userptr.notifier_lock); - vma->gpuva.flags &= ~XE_VMA_DESTROYED; - up_read(&vm->userptr.notifier_lock); - if (post_commit) - xe_vm_insert_vma(vm, vma); + } + + return number_tiles; +} + +static struct dma_fence *ops_execute(struct xe_vm *vm, + struct xe_vma_ops *vops) +{ + struct xe_tile *tile; + struct dma_fence *fence = NULL; + struct dma_fence **fences = NULL; + struct dma_fence_array *cf = NULL; + int number_tiles = 0, current_fence = 0, n_fence = 0, err; + u8 id; + + number_tiles = vm_ops_setup_tile_args(vm, vops); + if (number_tiles == 0) + return ERR_PTR(-ENODATA); + + if (vops->flags & XE_VMA_OPS_FLAG_SKIP_TLB_WAIT) { + for_each_tile(tile, vm->xe, id) + ++n_fence; + } else { + for_each_tile(tile, vm->xe, id) + n_fence += (1 + XE_MAX_GT_PER_TILE); + } + + fences = kmalloc_array(n_fence, sizeof(*fences), GFP_KERNEL); + if (!fences) { + fence = ERR_PTR(-ENOMEM); + goto err_trace; + } + + cf = dma_fence_array_alloc(n_fence); + if (!cf) { + fence = ERR_PTR(-ENOMEM); + goto err_out; + } + + for_each_tile(tile, vm->xe, id) { + if (!vops->pt_update_ops[id].num_ops) + continue; + + err = xe_pt_update_ops_prepare(tile, vops); + if (err) { + fence = ERR_PTR(err); + goto err_out; } - break; } + + trace_xe_vm_ops_execute(vops); + + for_each_tile(tile, vm->xe, id) { + struct xe_exec_queue *q = vops->pt_update_ops[tile->id].q; + int i; + + fence = NULL; + if (!vops->pt_update_ops[id].num_ops) + goto collect_fences; + + fence = xe_pt_update_ops_run(tile, vops); + if (IS_ERR(fence)) + goto err_out; + +collect_fences: + fences[current_fence++] = fence ?: dma_fence_get_stub(); + if (vops->flags & XE_VMA_OPS_FLAG_SKIP_TLB_WAIT) + continue; + + xe_migrate_job_lock(tile->migrate, q); + for_each_tlb_inval(i) + fences[current_fence++] = + xe_exec_queue_tlb_inval_last_fence_get(q, vm, i); + xe_migrate_job_unlock(tile->migrate, q); + } + + xe_assert(vm->xe, current_fence == n_fence); + dma_fence_array_init(cf, n_fence, fences, dma_fence_context_alloc(1), + 1, false); + fence = &cf->base; + + for_each_tile(tile, vm->xe, id) { + if (!vops->pt_update_ops[id].num_ops) + continue; + + xe_pt_update_ops_fini(tile, vops); + } + + return fence; + +err_out: + for_each_tile(tile, vm->xe, id) { + if (!vops->pt_update_ops[id].num_ops) + continue; + + xe_pt_update_ops_abort(tile, vops); + } + while (current_fence) + dma_fence_put(fences[--current_fence]); + kfree(fences); + kfree(cf); + +err_trace: + trace_xe_vm_ops_fail(vm); + return fence; +} + +static void vma_add_ufence(struct xe_vma *vma, struct xe_user_fence *ufence) +{ + if (vma->ufence) + xe_sync_ufence_put(vma->ufence); + vma->ufence = __xe_sync_ufence_get(ufence); +} + +static void op_add_ufence(struct xe_vm *vm, struct xe_vma_op *op, + struct xe_user_fence *ufence) +{ + switch (op->base.op) { + case DRM_GPUVA_OP_MAP: + vma_add_ufence(op->map.vma, ufence); + break; + case DRM_GPUVA_OP_REMAP: + if (op->remap.prev) + vma_add_ufence(op->remap.prev, ufence); + if (op->remap.next) + vma_add_ufence(op->remap.next, ufence); + break; + case DRM_GPUVA_OP_UNMAP: + break; case DRM_GPUVA_OP_PREFETCH: - /* Nothing to do */ + vma_add_ufence(gpuva_to_vma(op->base.prefetch.va), ufence); break; default: drm_warn(&vm->xe->drm, "NOT POSSIBLE"); } } -static void vm_bind_ioctl_ops_unwind(struct xe_vm *vm, - struct drm_gpuva_ops **ops, - int num_ops_list) +static void vm_bind_ioctl_ops_fini(struct xe_vm *vm, struct xe_vma_ops *vops, + struct dma_fence *fence) { + struct xe_user_fence *ufence; + struct xe_vma_op *op; int i; - for (i = num_ops_list - 1; i >= 0; --i) { - struct drm_gpuva_ops *__ops = ops[i]; - struct drm_gpuva_op *__op; - - if (!__ops) - continue; + ufence = find_ufence_get(vops->syncs, vops->num_syncs); + list_for_each_entry(op, &vops->list, link) { + if (ufence) + op_add_ufence(vm, op, ufence); - drm_gpuva_for_each_op_reverse(__op, __ops) { - struct xe_vma_op *op = gpuva_op_to_vma_op(__op); - - xe_vma_op_unwind(vm, op, - op->flags & XE_VMA_OP_COMMITTED, - op->flags & XE_VMA_OP_PREV_COMMITTED, - op->flags & XE_VMA_OP_NEXT_COMMITTED); - } - - drm_gpuva_ops_free(&vm->gpuvm, __ops); + if (op->base.op == DRM_GPUVA_OP_UNMAP) + xe_vma_destroy(gpuva_to_vma(op->base.unmap.va), fence); + else if (op->base.op == DRM_GPUVA_OP_REMAP) + xe_vma_destroy(gpuva_to_vma(op->base.remap.unmap->va), + fence); + } + if (ufence) + xe_sync_ufence_put(ufence); + if (fence) { + for (i = 0; i < vops->num_syncs; i++) + xe_sync_entry_signal(vops->syncs + i, fence); } } -static int vm_bind_ioctl_ops_execute(struct xe_vm *vm, - struct list_head *ops_list) +static struct dma_fence *vm_bind_ioctl_ops_execute(struct xe_vm *vm, + struct xe_vma_ops *vops) { - struct xe_vma_op *op, *next; - int err; + struct xe_validation_ctx ctx; + struct drm_exec exec; + struct dma_fence *fence; + int err = 0; lockdep_assert_held_write(&vm->lock); - list_for_each_entry_safe(op, next, ops_list, link) { - err = xe_vma_op_execute(vm, op); - if (err) { - drm_warn(&vm->xe->drm, "VM op(%d) failed with %d", - op->base.op, err); - /* - * FIXME: Killing VM rather than proper error handling - */ - xe_vm_kill(vm); - return -ENOSPC; + xe_validation_guard(&ctx, &vm->xe->val, &exec, + ((struct xe_val_flags) { + .interruptible = true, + .exec_ignore_duplicates = true, + }), err) { + err = vm_bind_ioctl_ops_lock_and_prep(&exec, vm, vops); + drm_exec_retry_on_contention(&exec); + xe_validation_retry_on_oom(&ctx, &err); + if (err) + return ERR_PTR(err); + + xe_vm_set_validation_exec(vm, &exec); + fence = ops_execute(vm, vops); + xe_vm_set_validation_exec(vm, NULL); + if (IS_ERR(fence)) { + if (PTR_ERR(fence) == -ENODATA) + vm_bind_ioctl_ops_fini(vm, vops, NULL); + return fence; } - xe_vma_op_cleanup(vm, op); + + vm_bind_ioctl_ops_fini(vm, vops, fence); } - return 0; + return err ? ERR_PTR(err) : fence; } +ALLOW_ERROR_INJECTION(vm_bind_ioctl_ops_execute, ERRNO); -#define SUPPORTED_FLAGS \ +#define SUPPORTED_FLAGS_STUB \ (DRM_XE_VM_BIND_FLAG_READONLY | \ DRM_XE_VM_BIND_FLAG_IMMEDIATE | \ DRM_XE_VM_BIND_FLAG_NULL | \ - DRM_XE_VM_BIND_FLAG_DUMPABLE) + DRM_XE_VM_BIND_FLAG_DUMPABLE | \ + DRM_XE_VM_BIND_FLAG_CHECK_PXP | \ + DRM_XE_VM_BIND_FLAG_CPU_ADDR_MIRROR | \ + DRM_XE_VM_BIND_FLAG_MADVISE_AUTORESET) + +#ifdef TEST_VM_OPS_ERROR +#define SUPPORTED_FLAGS (SUPPORTED_FLAGS_STUB | FORCE_OP_ERROR) +#else +#define SUPPORTED_FLAGS SUPPORTED_FLAGS_STUB +#endif + #define XE_64K_PAGE_MASK 0xffffull #define ALL_DRM_XE_SYNCS_FLAGS (DRM_XE_SYNCS_FLAG_WAIT_FOR_OP) -static int vm_bind_ioctl_check_args(struct xe_device *xe, +static int vm_bind_ioctl_check_args(struct xe_device *xe, struct xe_vm *vm, struct drm_xe_vm_bind *args, struct drm_xe_vm_bind_op **bind_ops) { @@ -2745,13 +3330,14 @@ static int vm_bind_ioctl_check_args(struct xe_device *xe, *bind_ops = kvmalloc_array(args->num_binds, sizeof(struct drm_xe_vm_bind_op), - GFP_KERNEL | __GFP_ACCOUNT); + GFP_KERNEL | __GFP_ACCOUNT | + __GFP_RETRY_MAYFAIL | __GFP_NOWARN); if (!*bind_ops) - return -ENOMEM; + return args->num_binds > 1 ? -ENOBUFS : -ENOMEM; - err = __copy_from_user(*bind_ops, bind_user, - sizeof(struct drm_xe_vm_bind_op) * - args->num_binds); + err = copy_from_user(*bind_ops, bind_user, + sizeof(struct drm_xe_vm_bind_op) * + args->num_binds); if (XE_IOCTL_DBG(xe, err)) { err = -EFAULT; goto free_bind_ops; @@ -2769,9 +3355,18 @@ static int vm_bind_ioctl_check_args(struct xe_device *xe, u64 obj_offset = (*bind_ops)[i].obj_offset; u32 prefetch_region = (*bind_ops)[i].prefetch_mem_region_instance; bool is_null = flags & DRM_XE_VM_BIND_FLAG_NULL; + bool is_cpu_addr_mirror = flags & + DRM_XE_VM_BIND_FLAG_CPU_ADDR_MIRROR; u16 pat_index = (*bind_ops)[i].pat_index; u16 coh_mode; + if (XE_IOCTL_DBG(xe, is_cpu_addr_mirror && + (!xe_vm_in_fault_mode(vm) || + !IS_ENABLED(CONFIG_DRM_XE_GPUSVM)))) { + err = -EINVAL; + goto free_bind_ops; + } + if (XE_IOCTL_DBG(xe, pat_index >= xe->pat.n_entries)) { err = -EINVAL; goto free_bind_ops; @@ -2792,13 +3387,14 @@ static int vm_bind_ioctl_check_args(struct xe_device *xe, if (XE_IOCTL_DBG(xe, op > DRM_XE_VM_BIND_OP_PREFETCH) || XE_IOCTL_DBG(xe, flags & ~SUPPORTED_FLAGS) || - XE_IOCTL_DBG(xe, obj && is_null) || - XE_IOCTL_DBG(xe, obj_offset && is_null) || + XE_IOCTL_DBG(xe, obj && (is_null || is_cpu_addr_mirror)) || + XE_IOCTL_DBG(xe, obj_offset && (is_null || + is_cpu_addr_mirror)) || XE_IOCTL_DBG(xe, op != DRM_XE_VM_BIND_OP_MAP && - is_null) || + (is_null || is_cpu_addr_mirror)) || XE_IOCTL_DBG(xe, !obj && op == DRM_XE_VM_BIND_OP_MAP && - !is_null) || + !is_null && !is_cpu_addr_mirror) || XE_IOCTL_DBG(xe, !obj && op == DRM_XE_VM_BIND_OP_UNMAP_ALL) || XE_IOCTL_DBG(xe, addr && @@ -2809,14 +3405,20 @@ static int vm_bind_ioctl_check_args(struct xe_device *xe, op == DRM_XE_VM_BIND_OP_MAP_USERPTR) || XE_IOCTL_DBG(xe, coh_mode == XE_COH_NONE && op == DRM_XE_VM_BIND_OP_MAP_USERPTR) || + XE_IOCTL_DBG(xe, op == DRM_XE_VM_BIND_OP_MAP_USERPTR && + !IS_ENABLED(CONFIG_DRM_GPUSVM)) || XE_IOCTL_DBG(xe, obj && op == DRM_XE_VM_BIND_OP_PREFETCH) || XE_IOCTL_DBG(xe, prefetch_region && op != DRM_XE_VM_BIND_OP_PREFETCH) || - XE_IOCTL_DBG(xe, !(BIT(prefetch_region) & - xe->info.mem_region_mask)) || + XE_IOCTL_DBG(xe, (prefetch_region != DRM_XE_CONSULT_MEM_ADVISE_PREF_LOC && + /* Guard against undefined shift in BIT(prefetch_region) */ + (prefetch_region >= (sizeof(xe->info.mem_region_mask) * 8) || + !(BIT(prefetch_region) & xe->info.mem_region_mask)))) || XE_IOCTL_DBG(xe, obj && - op == DRM_XE_VM_BIND_OP_UNMAP)) { + op == DRM_XE_VM_BIND_OP_UNMAP) || + XE_IOCTL_DBG(xe, (flags & DRM_XE_VM_BIND_FLAG_MADVISE_AUTORESET) && + (!is_cpu_addr_mirror || op != DRM_XE_VM_BIND_OP_MAP))) { err = -EINVAL; goto free_bind_ops; } @@ -2836,6 +3438,7 @@ static int vm_bind_ioctl_check_args(struct xe_device *xe, free_bind_ops: if (args->num_binds > 1) kvfree(*bind_ops); + *bind_ops = NULL; return err; } @@ -2844,24 +3447,92 @@ static int vm_bind_ioctl_signal_fences(struct xe_vm *vm, struct xe_sync_entry *syncs, int num_syncs) { - struct dma_fence *fence; + struct dma_fence *fence = NULL; int i, err = 0; - fence = xe_sync_in_fence_get(syncs, num_syncs, - to_wait_exec_queue(vm, q), vm); - if (IS_ERR(fence)) - return PTR_ERR(fence); + if (num_syncs) { + fence = xe_sync_in_fence_get(syncs, num_syncs, + to_wait_exec_queue(vm, q), vm); + if (IS_ERR(fence)) + return PTR_ERR(fence); - for (i = 0; i < num_syncs; i++) - xe_sync_entry_signal(&syncs[i], fence); + for (i = 0; i < num_syncs; i++) + xe_sync_entry_signal(&syncs[i], fence); + } - xe_exec_queue_last_fence_set(to_wait_exec_queue(vm, q), vm, - fence); dma_fence_put(fence); return err; } +static void xe_vma_ops_init(struct xe_vma_ops *vops, struct xe_vm *vm, + struct xe_exec_queue *q, + struct xe_sync_entry *syncs, u32 num_syncs) +{ + memset(vops, 0, sizeof(*vops)); + INIT_LIST_HEAD(&vops->list); + vops->vm = vm; + vops->q = q; + vops->syncs = syncs; + vops->num_syncs = num_syncs; + vops->flags = 0; +} + +static int xe_vm_bind_ioctl_validate_bo(struct xe_device *xe, struct xe_bo *bo, + u64 addr, u64 range, u64 obj_offset, + u16 pat_index, u32 op, u32 bind_flags) +{ + u16 coh_mode; + + if (XE_IOCTL_DBG(xe, range > xe_bo_size(bo)) || + XE_IOCTL_DBG(xe, obj_offset > + xe_bo_size(bo) - range)) { + return -EINVAL; + } + + /* + * Some platforms require 64k VM_BIND alignment, + * specifically those with XE_VRAM_FLAGS_NEED64K. + * + * Other platforms may have BO's set to 64k physical placement, + * but can be mapped at 4k offsets anyway. This check is only + * there for the former case. + */ + if ((bo->flags & XE_BO_FLAG_INTERNAL_64K) && + (xe->info.vram_flags & XE_VRAM_FLAGS_NEED64K)) { + if (XE_IOCTL_DBG(xe, obj_offset & + XE_64K_PAGE_MASK) || + XE_IOCTL_DBG(xe, addr & XE_64K_PAGE_MASK) || + XE_IOCTL_DBG(xe, range & XE_64K_PAGE_MASK)) { + return -EINVAL; + } + } + + coh_mode = xe_pat_index_get_coh_mode(xe, pat_index); + if (bo->cpu_caching) { + if (XE_IOCTL_DBG(xe, coh_mode == XE_COH_NONE && + bo->cpu_caching == DRM_XE_GEM_CPU_CACHING_WB)) { + return -EINVAL; + } + } else if (XE_IOCTL_DBG(xe, coh_mode == XE_COH_NONE)) { + /* + * Imported dma-buf from a different device should + * require 1way or 2way coherency since we don't know + * how it was mapped on the CPU. Just assume is it + * potentially cached on CPU side. + */ + return -EINVAL; + } + + /* If a BO is protected it can only be mapped if the key is still valid */ + if ((bind_flags & DRM_XE_VM_BIND_FLAG_CHECK_PXP) && xe_bo_is_protected(bo) && + op != DRM_XE_VM_BIND_OP_UNMAP && op != DRM_XE_VM_BIND_OP_UNMAP_ALL) + if (XE_IOCTL_DBG(xe, xe_pxp_bo_key_check(xe->pxp, bo) != 0)) + return -ENOEXEC; + + return 0; +} + int xe_vm_bind_ioctl(struct drm_device *dev, void *data, struct drm_file *file) { struct xe_device *xe = to_xe_device(dev); @@ -2874,20 +3545,25 @@ int xe_vm_bind_ioctl(struct drm_device *dev, void *data, struct drm_file *file) struct xe_exec_queue *q = NULL; u32 num_syncs, num_ufence = 0; struct xe_sync_entry *syncs = NULL; - struct drm_xe_vm_bind_op *bind_ops; - LIST_HEAD(ops_list); + struct drm_xe_vm_bind_op *bind_ops = NULL; + struct xe_vma_ops vops; + struct dma_fence *fence; int err; int i; - err = vm_bind_ioctl_check_args(xe, args, &bind_ops); + vm = xe_vm_lookup(xef, args->vm_id); + if (XE_IOCTL_DBG(xe, !vm)) + return -EINVAL; + + err = vm_bind_ioctl_check_args(xe, vm, args, &bind_ops); if (err) - return err; + goto put_vm; if (args->exec_queue_id) { q = xe_exec_queue_lookup(xef, args->exec_queue_id); if (XE_IOCTL_DBG(xe, !q)) { err = -ENOENT; - goto free_objs; + goto free_bind_ops; } if (XE_IOCTL_DBG(xe, !(q->flags & EXEC_QUEUE_FLAG_VM))) { @@ -2896,15 +3572,12 @@ int xe_vm_bind_ioctl(struct drm_device *dev, void *data, struct drm_file *file) } } - vm = xe_vm_lookup(xef, args->vm_id); - if (XE_IOCTL_DBG(xe, !vm)) { - err = -EINVAL; - goto put_exec_queue; - } + /* Ensure all UNMAPs visible */ + xe_svm_flush(vm); err = down_write_killable(&vm->lock); if (err) - goto put_vm; + goto put_exec_queue; if (XE_IOCTL_DBG(xe, xe_vm_is_closed_or_banned(vm))) { err = -ENOENT; @@ -2924,17 +3597,19 @@ int xe_vm_bind_ioctl(struct drm_device *dev, void *data, struct drm_file *file) if (args->num_binds) { bos = kvcalloc(args->num_binds, sizeof(*bos), - GFP_KERNEL | __GFP_ACCOUNT); + GFP_KERNEL | __GFP_ACCOUNT | + __GFP_RETRY_MAYFAIL | __GFP_NOWARN); if (!bos) { err = -ENOMEM; goto release_vm_lock; } ops = kvcalloc(args->num_binds, sizeof(*ops), - GFP_KERNEL | __GFP_ACCOUNT); + GFP_KERNEL | __GFP_ACCOUNT | + __GFP_RETRY_MAYFAIL | __GFP_NOWARN); if (!ops) { err = -ENOMEM; - goto release_vm_lock; + goto free_bos; } } @@ -2945,7 +3620,8 @@ int xe_vm_bind_ioctl(struct drm_device *dev, void *data, struct drm_file *file) u32 obj = bind_ops[i].obj; u64 obj_offset = bind_ops[i].obj_offset; u16 pat_index = bind_ops[i].pat_index; - u16 coh_mode; + u32 op = bind_ops[i].op; + u32 bind_flags = bind_ops[i].flags; if (!obj) continue; @@ -2957,40 +3633,11 @@ int xe_vm_bind_ioctl(struct drm_device *dev, void *data, struct drm_file *file) } bos[i] = gem_to_xe_bo(gem_obj); - if (XE_IOCTL_DBG(xe, range > bos[i]->size) || - XE_IOCTL_DBG(xe, obj_offset > - bos[i]->size - range)) { - err = -EINVAL; - goto put_obj; - } - - if (bos[i]->flags & XE_BO_FLAG_INTERNAL_64K) { - if (XE_IOCTL_DBG(xe, obj_offset & - XE_64K_PAGE_MASK) || - XE_IOCTL_DBG(xe, addr & XE_64K_PAGE_MASK) || - XE_IOCTL_DBG(xe, range & XE_64K_PAGE_MASK)) { - err = -EINVAL; - goto put_obj; - } - } - - coh_mode = xe_pat_index_get_coh_mode(xe, pat_index); - if (bos[i]->cpu_caching) { - if (XE_IOCTL_DBG(xe, coh_mode == XE_COH_NONE && - bos[i]->cpu_caching == DRM_XE_GEM_CPU_CACHING_WB)) { - err = -EINVAL; - goto put_obj; - } - } else if (XE_IOCTL_DBG(xe, coh_mode == XE_COH_NONE)) { - /* - * Imported dma-buf from a different device should - * require 1way or 2way coherency since we don't know - * how it was mapped on the CPU. Just assume is it - * potentially cached on CPU side. - */ - err = -EINVAL; + err = xe_vm_bind_ioctl_validate_bo(xe, bos[i], addr, range, + obj_offset, pat_index, op, + bind_flags); + if (err) goto put_obj; - } } if (args->num_syncs) { @@ -3003,8 +3650,12 @@ int xe_vm_bind_ioctl(struct drm_device *dev, void *data, struct drm_file *file) syncs_user = u64_to_user_ptr(args->syncs); for (num_syncs = 0; num_syncs < args->num_syncs; num_syncs++) { + struct xe_exec_queue *__q = q ?: vm->q[0]; + err = xe_sync_entry_parse(xe, xef, &syncs[num_syncs], &syncs_user[num_syncs], + __q->ufence_syncobj, + ++__q->ufence_timeline_value, (xe_vm_in_lr_mode(vm) ? SYNC_PARSE_FLAG_LR_MODE : 0) | (!args->num_binds ? @@ -3026,6 +3677,9 @@ int xe_vm_bind_ioctl(struct drm_device *dev, void *data, struct drm_file *file) goto free_syncs; } + xe_vma_ops_init(&vops, vm, q, syncs, num_syncs); + if (args->num_binds > 1) + vops.flags |= XE_VMA_OPS_ARRAY_OF_BINDS; for (i = 0; i < args->num_binds; ++i) { u64 range = bind_ops[i].range; u64 addr = bind_ops[i].addr; @@ -3035,7 +3689,7 @@ int xe_vm_bind_ioctl(struct drm_device *dev, void *data, struct drm_file *file) u32 prefetch_region = bind_ops[i].prefetch_mem_region_instance; u16 pat_index = bind_ops[i].pat_index; - ops[i] = vm_bind_ioctl_ops_create(vm, bos[i], obj_offset, + ops[i] = vm_bind_ioctl_ops_create(vm, &vops, bos[i], obj_offset, addr, range, op, flags, prefetch_region, pat_index); if (IS_ERR(ops[i])) { @@ -3044,43 +3698,47 @@ int xe_vm_bind_ioctl(struct drm_device *dev, void *data, struct drm_file *file) goto unwind_ops; } - err = vm_bind_ioctl_ops_parse(vm, q, ops[i], syncs, num_syncs, - &ops_list, - i == args->num_binds - 1); + err = vm_bind_ioctl_ops_parse(vm, ops[i], &vops); if (err) goto unwind_ops; + +#ifdef TEST_VM_OPS_ERROR + if (flags & FORCE_OP_ERROR) { + vops.inject_error = true; + vm->xe->vm_inject_error_position = + (vm->xe->vm_inject_error_position + 1) % + FORCE_OP_ERROR_COUNT; + } +#endif } /* Nothing to do */ - if (list_empty(&ops_list)) { + if (list_empty(&vops.list)) { err = -ENODATA; goto unwind_ops; } - xe_vm_get(vm); - if (q) - xe_exec_queue_get(q); - - err = vm_bind_ioctl_ops_execute(vm, &ops_list); - - up_write(&vm->lock); - - if (q) - xe_exec_queue_put(q); - xe_vm_put(vm); - - for (i = 0; bos && i < args->num_binds; ++i) - xe_bo_put(bos[i]); + err = xe_vma_ops_alloc(&vops, args->num_binds > 1); + if (err) + goto unwind_ops; - kvfree(bos); - kvfree(ops); - if (args->num_binds > 1) - kvfree(bind_ops); + err = vm_bind_ioctl_ops_prefetch_ranges(vm, &vops); + if (err) + goto unwind_ops; - return err; + fence = vm_bind_ioctl_ops_execute(vm, &vops); + if (IS_ERR(fence)) + err = PTR_ERR(fence); + else + dma_fence_put(fence); unwind_ops: - vm_bind_ioctl_ops_unwind(vm, ops, args->num_binds); + if (err && err != -ENODATA) + vm_bind_ioctl_ops_unwind(vm, ops, args->num_binds); + xe_vma_ops_fini(&vops); + for (i = args->num_binds - 1; i >= 0; --i) + if (ops[i]) + drm_gpuva_ops_free(&vm->gpuvm, ops[i]); free_syncs: if (err == -ENODATA) err = vm_bind_ioctl_signal_fences(vm, q, syncs, num_syncs); @@ -3091,22 +3749,99 @@ free_syncs: put_obj: for (i = 0; i < args->num_binds; ++i) xe_bo_put(bos[i]); + + kvfree(ops); +free_bos: + kvfree(bos); release_vm_lock: up_write(&vm->lock); -put_vm: - xe_vm_put(vm); put_exec_queue: if (q) xe_exec_queue_put(q); -free_objs: - kvfree(bos); - kvfree(ops); +free_bind_ops: if (args->num_binds > 1) kvfree(bind_ops); +put_vm: + xe_vm_put(vm); return err; } /** + * xe_vm_bind_kernel_bo - bind a kernel BO to a VM + * @vm: VM to bind the BO to + * @bo: BO to bind + * @q: exec queue to use for the bind (optional) + * @addr: address at which to bind the BO + * @cache_lvl: PAT cache level to use + * + * Execute a VM bind map operation on a kernel-owned BO to bind it into a + * kernel-owned VM. + * + * Returns a dma_fence to track the binding completion if the job to do so was + * successfully submitted, an error pointer otherwise. + */ +struct dma_fence *xe_vm_bind_kernel_bo(struct xe_vm *vm, struct xe_bo *bo, + struct xe_exec_queue *q, u64 addr, + enum xe_cache_level cache_lvl) +{ + struct xe_vma_ops vops; + struct drm_gpuva_ops *ops = NULL; + struct dma_fence *fence; + int err; + + xe_bo_get(bo); + xe_vm_get(vm); + if (q) + xe_exec_queue_get(q); + + down_write(&vm->lock); + + xe_vma_ops_init(&vops, vm, q, NULL, 0); + + ops = vm_bind_ioctl_ops_create(vm, &vops, bo, 0, addr, xe_bo_size(bo), + DRM_XE_VM_BIND_OP_MAP, 0, 0, + vm->xe->pat.idx[cache_lvl]); + if (IS_ERR(ops)) { + err = PTR_ERR(ops); + goto release_vm_lock; + } + + err = vm_bind_ioctl_ops_parse(vm, ops, &vops); + if (err) + goto release_vm_lock; + + xe_assert(vm->xe, !list_empty(&vops.list)); + + err = xe_vma_ops_alloc(&vops, false); + if (err) + goto unwind_ops; + + fence = vm_bind_ioctl_ops_execute(vm, &vops); + if (IS_ERR(fence)) + err = PTR_ERR(fence); + +unwind_ops: + if (err && err != -ENODATA) + vm_bind_ioctl_ops_unwind(vm, &ops, 1); + + xe_vma_ops_fini(&vops); + drm_gpuva_ops_free(&vm->gpuvm, ops); + +release_vm_lock: + up_write(&vm->lock); + + if (q) + xe_exec_queue_put(q); + xe_vm_put(vm); + xe_bo_put(bo); + + if (err) + fence = ERR_PTR(err); + + return fence; +} + +/** * xe_vm_lock() - Lock the vm's dma_resv object * @vm: The struct xe_vm whose lock is to be locked * @intr: Whether to perform any wait interruptible @@ -3117,10 +3852,14 @@ free_objs: */ int xe_vm_lock(struct xe_vm *vm, bool intr) { + int ret; + if (intr) - return dma_resv_lock_interruptible(xe_vm_resv(vm), NULL); + ret = dma_resv_lock_interruptible(xe_vm_resv(vm), NULL); + else + ret = dma_resv_lock(xe_vm_resv(vm), NULL); - return dma_resv_lock(xe_vm_resv(vm), NULL); + return ret; } /** @@ -3135,6 +3874,66 @@ void xe_vm_unlock(struct xe_vm *vm) } /** + * xe_vm_range_tilemask_tlb_inval - Issue a TLB invalidation on this tilemask for an + * address range + * @vm: The VM + * @start: start address + * @end: end address + * @tile_mask: mask for which gt's issue tlb invalidation + * + * Issue a range based TLB invalidation for gt's in tilemask + * + * Returns 0 for success, negative error code otherwise. + */ +int xe_vm_range_tilemask_tlb_inval(struct xe_vm *vm, u64 start, + u64 end, u8 tile_mask) +{ + struct xe_tlb_inval_fence + fence[XE_MAX_TILES_PER_DEVICE * XE_MAX_GT_PER_TILE]; + struct xe_tile *tile; + u32 fence_id = 0; + u8 id; + int err; + + if (!tile_mask) + return 0; + + for_each_tile(tile, vm->xe, id) { + if (!(tile_mask & BIT(id))) + continue; + + xe_tlb_inval_fence_init(&tile->primary_gt->tlb_inval, + &fence[fence_id], true); + + err = xe_tlb_inval_range(&tile->primary_gt->tlb_inval, + &fence[fence_id], start, end, + vm->usm.asid); + if (err) + goto wait; + ++fence_id; + + if (!tile->media_gt) + continue; + + xe_tlb_inval_fence_init(&tile->media_gt->tlb_inval, + &fence[fence_id], true); + + err = xe_tlb_inval_range(&tile->media_gt->tlb_inval, + &fence[fence_id], start, end, + vm->usm.asid); + if (err) + goto wait; + ++fence_id; + } + +wait: + for (id = 0; id < fence_id; ++id) + xe_tlb_inval_fence_wait(&fence[id]); + + return err; +} + +/** * xe_vm_invalidate_vma - invalidate GPU mappings for VMA without a lock * @vma: VMA to invalidate * @@ -3147,26 +3946,34 @@ void xe_vm_unlock(struct xe_vm *vm) int xe_vm_invalidate_vma(struct xe_vma *vma) { struct xe_device *xe = xe_vma_vm(vma)->xe; + struct xe_vm *vm = xe_vma_vm(vma); struct xe_tile *tile; - u32 tile_needs_invalidate = 0; - int seqno[XE_MAX_TILES_PER_DEVICE]; + u8 tile_mask = 0; + int ret = 0; u8 id; - int ret; xe_assert(xe, !xe_vma_is_null(vma)); + xe_assert(xe, !xe_vma_is_cpu_addr_mirror(vma)); trace_xe_vma_invalidate(vma); - vm_dbg(&xe_vma_vm(vma)->xe->drm, + vm_dbg(&vm->xe->drm, "INVALIDATE: addr=0x%016llx, range=0x%016llx", xe_vma_start(vma), xe_vma_size(vma)); - /* Check that we don't race with page-table updates */ + /* + * Check that we don't race with page-table updates, tile_invalidated + * update is safe + */ if (IS_ENABLED(CONFIG_PROVE_LOCKING)) { if (xe_vma_is_userptr(vma)) { + lockdep_assert(lockdep_is_held_type(&vm->svm.gpusvm.notifier_lock, 0) || + (lockdep_is_held_type(&vm->svm.gpusvm.notifier_lock, 1) && + lockdep_is_held(&xe_vm_resv(vm)->lock.base))); + WARN_ON_ONCE(!mmu_interval_check_retry (&to_userptr_vma(vma)->userptr.notifier, - to_userptr_vma(vma)->userptr.notifier_seq)); - WARN_ON_ONCE(!dma_resv_test_signaled(xe_vm_resv(xe_vma_vm(vma)), + to_userptr_vma(vma)->userptr.pages.notifier_seq)); + WARN_ON_ONCE(!dma_resv_test_signaled(xe_vm_resv(vm), DMA_RESV_USAGE_BOOKKEEP)); } else { @@ -3174,80 +3981,48 @@ int xe_vm_invalidate_vma(struct xe_vma *vma) } } - for_each_tile(tile, xe, id) { - if (xe_pt_zap_ptes(tile, vma)) { - tile_needs_invalidate |= BIT(id); - xe_device_wmb(xe); - /* - * FIXME: We potentially need to invalidate multiple - * GTs within the tile - */ - seqno[id] = xe_gt_tlb_invalidation_vma(tile->primary_gt, NULL, vma); - if (seqno[id] < 0) - return seqno[id]; - } - } + for_each_tile(tile, xe, id) + if (xe_pt_zap_ptes(tile, vma)) + tile_mask |= BIT(id); - for_each_tile(tile, xe, id) { - if (tile_needs_invalidate & BIT(id)) { - ret = xe_gt_tlb_invalidation_wait(tile->primary_gt, seqno[id]); - if (ret < 0) - return ret; - } - } + xe_device_wmb(xe); - vma->tile_invalidated = vma->tile_mask; + ret = xe_vm_range_tilemask_tlb_inval(xe_vma_vm(vma), xe_vma_start(vma), + xe_vma_end(vma), tile_mask); - return 0; + /* WRITE_ONCE pairs with READ_ONCE in xe_vm_has_valid_gpu_mapping() */ + WRITE_ONCE(vma->tile_invalidated, vma->tile_mask); + + return ret; } -int xe_analyze_vm(struct drm_printer *p, struct xe_vm *vm, int gt_id) +int xe_vm_validate_protected(struct xe_vm *vm) { struct drm_gpuva *gpuva; - bool is_vram; - uint64_t addr; + int err = 0; - if (!down_read_trylock(&vm->lock)) { - drm_printf(p, " Failed to acquire VM lock to dump capture"); - return 0; - } - if (vm->pt_root[gt_id]) { - addr = xe_bo_addr(vm->pt_root[gt_id]->bo, 0, XE_PAGE_SIZE); - is_vram = xe_bo_is_vram(vm->pt_root[gt_id]->bo); - drm_printf(p, " VM root: A:0x%llx %s\n", addr, - is_vram ? "VRAM" : "SYS"); - } + if (!vm) + return -ENODEV; + + mutex_lock(&vm->snap_mutex); drm_gpuvm_for_each_va(gpuva, &vm->gpuvm) { struct xe_vma *vma = gpuva_to_vma(gpuva); - bool is_userptr = xe_vma_is_userptr(vma); - bool is_null = xe_vma_is_null(vma); - - if (is_null) { - addr = 0; - } else if (is_userptr) { - struct sg_table *sg = to_userptr_vma(vma)->userptr.sg; - struct xe_res_cursor cur; - - if (sg) { - xe_res_first_sg(sg, 0, XE_PAGE_SIZE, &cur); - addr = xe_res_dma(&cur); - } else { - addr = 0; - } - } else { - addr = __xe_bo_addr(xe_vma_bo(vma), 0, XE_PAGE_SIZE); - is_vram = xe_bo_is_vram(xe_vma_bo(vma)); + struct xe_bo *bo = vma->gpuva.gem.obj ? + gem_to_xe_bo(vma->gpuva.gem.obj) : NULL; + + if (!bo) + continue; + + if (xe_bo_is_protected(bo)) { + err = xe_pxp_bo_key_check(vm->xe->pxp, bo); + if (err) + break; } - drm_printf(p, " [%016llx-%016llx] S:0x%016llx A:%016llx %s\n", - xe_vma_start(vma), xe_vma_end(vma) - 1, - xe_vma_size(vma), - addr, is_null ? "NULL" : is_userptr ? "USR" : - is_vram ? "VRAM" : "SYS"); } - up_read(&vm->lock); - return 0; + mutex_unlock(&vm->snap_mutex); + return err; } struct xe_vm_snapshot { @@ -3326,7 +4101,6 @@ void xe_vm_snapshot_capture_delayed(struct xe_vm_snapshot *snap) for (int i = 0; i < snap->num_snaps; i++) { struct xe_bo *bo = snap->snap[i].bo; - struct iosys_map src; int err; if (IS_ERR(snap->snap[i].data)) @@ -3339,16 +4113,8 @@ void xe_vm_snapshot_capture_delayed(struct xe_vm_snapshot *snap) } if (bo) { - dma_resv_lock(bo->ttm.base.resv, NULL); - err = ttm_bo_vmap(&bo->ttm, &src); - if (!err) { - xe_map_memcpy_from(xe_bo_device(bo), - snap->snap[i].data, - &src, snap->snap[i].bo_ofs, - snap->snap[i].len); - ttm_bo_vunmap(&bo->ttm, &src); - } - dma_resv_unlock(bo->ttm.base.resv); + err = xe_bo_read(bo, snap->snap[i].bo_ofs, + snap->snap[i].data, snap->snap[i].len); } else { void __user *userptr = (void __user *)(size_t)snap->snap[i].bo_ofs; @@ -3402,6 +4168,9 @@ void xe_vm_snapshot_print(struct xe_vm_snapshot *snap, struct drm_printer *p) } drm_puts(p, "\n"); + + if (drm_coredump_printer_is_full(p)) + return; } } @@ -3421,3 +4190,221 @@ void xe_vm_snapshot_free(struct xe_vm_snapshot *snap) } kvfree(snap); } + +/** + * xe_vma_need_vram_for_atomic - Check if VMA needs VRAM migration for atomic operations + * @xe: Pointer to the Xe device structure + * @vma: Pointer to the virtual memory area (VMA) structure + * @is_atomic: In pagefault path and atomic operation + * + * This function determines whether the given VMA needs to be migrated to + * VRAM in order to do atomic GPU operation. + * + * Return: + * 1 - Migration to VRAM is required + * 0 - Migration is not required + * -EACCES - Invalid access for atomic memory attr + * + */ +int xe_vma_need_vram_for_atomic(struct xe_device *xe, struct xe_vma *vma, bool is_atomic) +{ + u32 atomic_access = xe_vma_bo(vma) ? xe_vma_bo(vma)->attr.atomic_access : + vma->attr.atomic_access; + + if (!IS_DGFX(xe) || !is_atomic) + return false; + + /* + * NOTE: The checks implemented here are platform-specific. For + * instance, on a device supporting CXL atomics, these would ideally + * work universally without additional handling. + */ + switch (atomic_access) { + case DRM_XE_ATOMIC_DEVICE: + return !xe->info.has_device_atomics_on_smem; + + case DRM_XE_ATOMIC_CPU: + return -EACCES; + + case DRM_XE_ATOMIC_UNDEFINED: + case DRM_XE_ATOMIC_GLOBAL: + default: + return 1; + } +} + +static int xe_vm_alloc_vma(struct xe_vm *vm, + struct drm_gpuvm_map_req *map_req, + bool is_madvise) +{ + struct xe_vma_ops vops; + struct drm_gpuva_ops *ops = NULL; + struct drm_gpuva_op *__op; + unsigned int vma_flags = 0; + bool remap_op = false; + struct xe_vma_mem_attr tmp_attr; + u16 default_pat; + int err; + + lockdep_assert_held_write(&vm->lock); + + if (is_madvise) + ops = drm_gpuvm_madvise_ops_create(&vm->gpuvm, map_req); + else + ops = drm_gpuvm_sm_map_ops_create(&vm->gpuvm, map_req); + + if (IS_ERR(ops)) + return PTR_ERR(ops); + + if (list_empty(&ops->list)) { + err = 0; + goto free_ops; + } + + drm_gpuva_for_each_op(__op, ops) { + struct xe_vma_op *op = gpuva_op_to_vma_op(__op); + struct xe_vma *vma = NULL; + + if (!is_madvise) { + if (__op->op == DRM_GPUVA_OP_UNMAP) { + vma = gpuva_to_vma(op->base.unmap.va); + XE_WARN_ON(!xe_vma_has_default_mem_attrs(vma)); + default_pat = vma->attr.default_pat_index; + vma_flags = vma->gpuva.flags; + } + + if (__op->op == DRM_GPUVA_OP_REMAP) { + vma = gpuva_to_vma(op->base.remap.unmap->va); + default_pat = vma->attr.default_pat_index; + vma_flags = vma->gpuva.flags; + } + + if (__op->op == DRM_GPUVA_OP_MAP) { + op->map.vma_flags |= vma_flags & XE_VMA_CREATE_MASK; + op->map.pat_index = default_pat; + } + } else { + if (__op->op == DRM_GPUVA_OP_REMAP) { + vma = gpuva_to_vma(op->base.remap.unmap->va); + xe_assert(vm->xe, !remap_op); + xe_assert(vm->xe, xe_vma_has_no_bo(vma)); + remap_op = true; + vma_flags = vma->gpuva.flags; + } + + if (__op->op == DRM_GPUVA_OP_MAP) { + xe_assert(vm->xe, remap_op); + remap_op = false; + /* + * In case of madvise ops DRM_GPUVA_OP_MAP is + * always after DRM_GPUVA_OP_REMAP, so ensure + * to propagate the flags from the vma we're + * unmapping. + */ + op->map.vma_flags |= vma_flags & XE_VMA_CREATE_MASK; + } + } + print_op(vm->xe, __op); + } + + xe_vma_ops_init(&vops, vm, NULL, NULL, 0); + + if (is_madvise) + vops.flags |= XE_VMA_OPS_FLAG_MADVISE; + + err = vm_bind_ioctl_ops_parse(vm, ops, &vops); + if (err) + goto unwind_ops; + + xe_vm_lock(vm, false); + + drm_gpuva_for_each_op(__op, ops) { + struct xe_vma_op *op = gpuva_op_to_vma_op(__op); + struct xe_vma *vma; + + if (__op->op == DRM_GPUVA_OP_UNMAP) { + vma = gpuva_to_vma(op->base.unmap.va); + /* There should be no unmap for madvise */ + if (is_madvise) + XE_WARN_ON("UNEXPECTED UNMAP"); + + xe_vma_destroy(vma, NULL); + } else if (__op->op == DRM_GPUVA_OP_REMAP) { + vma = gpuva_to_vma(op->base.remap.unmap->va); + /* In case of madvise ops Store attributes for REMAP UNMAPPED + * VMA, so they can be assigned to newly MAP created vma. + */ + if (is_madvise) + tmp_attr = vma->attr; + + xe_vma_destroy(gpuva_to_vma(op->base.remap.unmap->va), NULL); + } else if (__op->op == DRM_GPUVA_OP_MAP) { + vma = op->map.vma; + /* In case of madvise call, MAP will always be followed by REMAP. + * Therefore temp_attr will always have sane values, making it safe to + * copy them to new vma. + */ + if (is_madvise) + vma->attr = tmp_attr; + } + } + + xe_vm_unlock(vm); + drm_gpuva_ops_free(&vm->gpuvm, ops); + return 0; + +unwind_ops: + vm_bind_ioctl_ops_unwind(vm, &ops, 1); +free_ops: + drm_gpuva_ops_free(&vm->gpuvm, ops); + return err; +} + +/** + * xe_vm_alloc_madvise_vma - Allocate VMA's with madvise ops + * @vm: Pointer to the xe_vm structure + * @start: Starting input address + * @range: Size of the input range + * + * This function splits existing vma to create new vma for user provided input range + * + * Return: 0 if success + */ +int xe_vm_alloc_madvise_vma(struct xe_vm *vm, uint64_t start, uint64_t range) +{ + struct drm_gpuvm_map_req map_req = { + .map.va.addr = start, + .map.va.range = range, + }; + + lockdep_assert_held_write(&vm->lock); + + vm_dbg(&vm->xe->drm, "MADVISE_OPS_CREATE: addr=0x%016llx, size=0x%016llx", start, range); + + return xe_vm_alloc_vma(vm, &map_req, true); +} + +/** + * xe_vm_alloc_cpu_addr_mirror_vma - Allocate CPU addr mirror vma + * @vm: Pointer to the xe_vm structure + * @start: Starting input address + * @range: Size of the input range + * + * This function splits/merges existing vma to create new vma for user provided input range + * + * Return: 0 if success + */ +int xe_vm_alloc_cpu_addr_mirror_vma(struct xe_vm *vm, uint64_t start, uint64_t range) +{ + struct drm_gpuvm_map_req map_req = { + .map.va.addr = start, + .map.va.range = range, + }; + + lockdep_assert_held_write(&vm->lock); + + vm_dbg(&vm->xe->drm, "CPU_ADDR_MIRROR_VMA_OPS_CREATE: addr=0x%016llx, size=0x%016llx", + start, range); + + return xe_vm_alloc_vma(vm, &map_req, false); +} |