unsigned int fence_count = 0;
LIST_HEAD(preempt_fences);
ktime_t end = 0;
- int err;
+ int err = 0;
long wait;
int __maybe_unused tries = 0;
}
retry:
- if (vm->async_ops.error)
- goto out_unlock_outer;
-
- /*
- * Extreme corner where we exit a VM error state with a munmap style VM
- * unbind inflight which requires a rebind. In this case the rebind
- * needs to install some fences into the dma-resv slots. The worker to
- * do this queued, let that worker make progress by dropping vm->lock
- * and trying this again.
- */
- if (vm->async_ops.munmap_rebind_inflight) {
- up_write(&vm->lock);
- flush_work(&vm->async_ops.work);
- goto retry;
- }
-
if (xe_vm_userptr_check_repin(vm)) {
err = xe_vm_userptr_pin(vm);
if (err)
return pte;
}
-static u64 pte_encode_pat_index(struct xe_device *xe, u16 pat_index)
+static u64 pte_encode_pat_index(struct xe_device *xe, u16 pat_index,
+ u32 pt_level)
{
u64 pte = 0;
if (pat_index & BIT(1))
pte |= XE_PPGTT_PTE_PAT1;
- if (pat_index & BIT(2))
- pte |= XE_PPGTT_PTE_PAT2;
+ if (pat_index & BIT(2)) {
+ if (pt_level)
+ pte |= XE_PPGTT_PDE_PDPE_PAT2;
+ else
+ pte |= XE_PPGTT_PTE_PAT2;
+ }
if (pat_index & BIT(3))
pte |= XELPG_PPGTT_PTE_PAT3;
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);
+ pte |= pte_encode_pat_index(xe, pat_index, pt_level);
pte |= pte_encode_ps(pt_level);
if (xe_bo_is_vram(bo) || xe_bo_is_stolen_devmem(bo))
if (likely(!xe_vma_read_only(vma)))
pte |= XE_PAGE_RW;
- pte |= pte_encode_pat_index(xe, pat_index);
+ pte |= pte_encode_pat_index(xe, pat_index, pt_level);
pte |= pte_encode_ps(pt_level);
if (unlikely(xe_vma_is_null(vma)))
pte = addr;
pte |= XE_PAGE_PRESENT | XE_PAGE_RW;
- pte |= pte_encode_pat_index(xe, pat_index);
+ pte |= pte_encode_pat_index(xe, pat_index, pt_level);
pte |= pte_encode_ps(pt_level);
if (devmem)
.pde_encode_bo = xelp_pde_encode_bo,
};
-static void xe_vma_op_work_func(struct work_struct *w);
static void vm_destroy_work_func(struct work_struct *w);
struct xe_vm *xe_vm_create(struct xe_device *xe, u32 flags)
INIT_LIST_HEAD(&vm->notifier.rebind_list);
spin_lock_init(&vm->notifier.list_lock);
- INIT_LIST_HEAD(&vm->async_ops.pending);
- INIT_WORK(&vm->async_ops.work, xe_vma_op_work_func);
- spin_lock_init(&vm->async_ops.lock);
-
INIT_WORK(&vm->destroy_work, vm_destroy_work_func);
INIT_LIST_HEAD(&vm->preempt.exec_queues);
vm->batch_invalidate_tlb = false;
}
- if (flags & XE_VM_FLAG_ASYNC_BIND_OPS) {
- vm->async_ops.fence.context = dma_fence_context_alloc(1);
- vm->flags |= XE_VM_FLAG_ASYNC_BIND_OPS;
- }
-
/* Fill pt_root after allocating scratch tables */
for_each_tile(tile, xe, id) {
if (!vm->pt_root[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 |
+ ((flags & XE_VM_FLAG_ASYNC_DEFAULT) ?
+ EXEC_QUEUE_FLAG_VM_ASYNC : 0);
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,
- EXEC_QUEUE_FLAG_VM);
+ create_flags);
xe_vm_put(migrate_vm);
if (IS_ERR(q)) {
err = PTR_ERR(q);
return ERR_PTR(err);
}
-static void flush_async_ops(struct xe_vm *vm)
-{
- queue_work(system_unbound_wq, &vm->async_ops.work);
- flush_work(&vm->async_ops.work);
-}
-
static void xe_vm_close(struct xe_vm *vm)
{
down_write(&vm->lock);
xe_assert(xe, !vm->preempt.num_exec_queues);
xe_vm_close(vm);
- flush_async_ops(vm);
if (xe_vm_in_compute_mode(vm))
flush_work(&vm->preempt.rebind_work);
+ down_write(&vm->lock);
+ for_each_tile(tile, xe, id) {
+ if (vm->q[id])
+ xe_exec_queue_last_fence_put(vm->q[id], vm);
+ }
+ up_write(&vm->lock);
+
for_each_tile(tile, xe, id) {
if (vm->q[id]) {
xe_exec_queue_kill(vm->q[id]);
tile_to_xe(tile)->pat.idx[XE_CACHE_WB]);
}
+static struct xe_exec_queue *
+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;
- struct xe_vm *vm = xe_vma_vm(vma);
int cur_fence = 0, i;
int number_tiles = hweight8(vma->tile_present);
int err;
}
}
+ 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], NULL,
- cf ? &cf->base : fence);
+ xe_sync_entry_signal(&syncs[i], NULL, fence);
}
- return cf ? &cf->base : !fence ? dma_fence_get_stub() : fence;
+ return fence;
err_fences:
if (fences) {
- while (cur_fence) {
- /* FIXME: Rewind the previous binds? */
+ while (cur_fence)
dma_fence_put(fences[--cur_fence]);
- }
kfree(fences);
}
err_fences:
if (fences) {
- while (cur_fence) {
- /* FIXME: Rewind the previous binds? */
+ while (cur_fence)
dma_fence_put(fences[--cur_fence]);
- }
kfree(fences);
}
return ERR_PTR(err);
}
-struct async_op_fence {
- struct dma_fence fence;
- struct dma_fence *wait_fence;
- struct dma_fence_cb cb;
- struct xe_vm *vm;
- wait_queue_head_t wq;
- bool started;
-};
-
-static const char *async_op_fence_get_driver_name(struct dma_fence *dma_fence)
-{
- return "xe";
-}
-
-static const char *
-async_op_fence_get_timeline_name(struct dma_fence *dma_fence)
+static bool xe_vm_sync_mode(struct xe_vm *vm, struct xe_exec_queue *q)
{
- return "async_op_fence";
-}
-
-static const struct dma_fence_ops async_op_fence_ops = {
- .get_driver_name = async_op_fence_get_driver_name,
- .get_timeline_name = async_op_fence_get_timeline_name,
-};
-
-static void async_op_fence_cb(struct dma_fence *fence, struct dma_fence_cb *cb)
-{
- struct async_op_fence *afence =
- container_of(cb, struct async_op_fence, cb);
-
- afence->fence.error = afence->wait_fence->error;
- dma_fence_signal(&afence->fence);
- xe_vm_put(afence->vm);
- dma_fence_put(afence->wait_fence);
- dma_fence_put(&afence->fence);
-}
-
-static void add_async_op_fence_cb(struct xe_vm *vm,
- struct dma_fence *fence,
- struct async_op_fence *afence)
-{
- int ret;
-
- if (!xe_vm_no_dma_fences(vm)) {
- afence->started = true;
- smp_wmb();
- wake_up_all(&afence->wq);
- }
-
- afence->wait_fence = dma_fence_get(fence);
- afence->vm = xe_vm_get(vm);
- dma_fence_get(&afence->fence);
- ret = dma_fence_add_callback(fence, &afence->cb, async_op_fence_cb);
- if (ret == -ENOENT) {
- afence->fence.error = afence->wait_fence->error;
- dma_fence_signal(&afence->fence);
- }
- if (ret) {
- xe_vm_put(vm);
- dma_fence_put(afence->wait_fence);
- dma_fence_put(&afence->fence);
- }
- XE_WARN_ON(ret && ret != -ENOENT);
-}
-
-int xe_vm_async_fence_wait_start(struct dma_fence *fence)
-{
- if (fence->ops == &async_op_fence_ops) {
- struct async_op_fence *afence =
- container_of(fence, struct async_op_fence, fence);
-
- xe_assert(afence->vm->xe, !xe_vm_no_dma_fences(afence->vm));
-
- smp_rmb();
- return wait_event_interruptible(afence->wq, afence->started);
- }
-
- return 0;
+ return q ? !(q->flags & EXEC_QUEUE_FLAG_VM_ASYNC) :
+ !(vm->flags & XE_VM_FLAG_ASYNC_DEFAULT);
}
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, struct async_op_fence *afence,
- bool immediate, bool first_op, bool last_op)
+ 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);
xe_vm_assert_held(vm);
xe_assert(vm->xe, xe_vm_in_fault_mode(vm));
- fence = dma_fence_get_stub();
+ 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], NULL, fence);
}
}
- if (afence)
- add_async_op_fence_cb(vm, fence, afence);
+ if (last_op)
+ xe_exec_queue_last_fence_set(wait_exec_queue, vm, fence);
+ if (last_op && xe_vm_sync_mode(vm, q))
+ dma_fence_wait(fence, true);
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, struct async_op_fence *afence,
- bool immediate, bool first_op, bool last_op)
+ u32 num_syncs, bool immediate, bool first_op,
+ bool last_op)
{
int err;
return err;
}
- return __xe_vm_bind(vm, vma, q, syncs, num_syncs, afence, immediate,
- first_op, last_op);
+ 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, struct async_op_fence *afence,
- bool first_op, bool last_op)
+ 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);
- if (afence)
- add_async_op_fence_cb(vm, fence, afence);
xe_vma_destroy(vma, fence);
+ if (last_op)
+ xe_exec_queue_last_fence_set(wait_exec_queue, vm, fence);
+ if (last_op && xe_vm_sync_mode(vm, q))
+ dma_fence_wait(fence, true);
dma_fence_put(fence);
return 0;
}
-#define ALL_DRM_XE_VM_CREATE_FLAGS (DRM_XE_VM_CREATE_SCRATCH_PAGE | \
- DRM_XE_VM_CREATE_COMPUTE_MODE | \
- DRM_XE_VM_CREATE_ASYNC_BIND_OPS | \
- DRM_XE_VM_CREATE_FAULT_MODE)
+#define ALL_DRM_XE_VM_CREATE_FLAGS (DRM_XE_VM_CREATE_FLAG_SCRATCH_PAGE | \
+ DRM_XE_VM_CREATE_FLAG_COMPUTE_MODE | \
+ DRM_XE_VM_CREATE_FLAG_ASYNC_DEFAULT | \
+ DRM_XE_VM_CREATE_FLAG_FAULT_MODE)
int xe_vm_create_ioctl(struct drm_device *dev, void *data,
struct drm_file *file)
return -EINVAL;
if (XE_WA(xe_root_mmio_gt(xe), 14016763929))
- args->flags |= DRM_XE_VM_CREATE_SCRATCH_PAGE;
+ args->flags |= DRM_XE_VM_CREATE_FLAG_SCRATCH_PAGE;
- if (XE_IOCTL_DBG(xe, args->flags & DRM_XE_VM_CREATE_FAULT_MODE &&
+ if (XE_IOCTL_DBG(xe, args->flags & DRM_XE_VM_CREATE_FLAG_FAULT_MODE &&
!xe->info.supports_usm))
return -EINVAL;
if (XE_IOCTL_DBG(xe, args->flags & ~ALL_DRM_XE_VM_CREATE_FLAGS))
return -EINVAL;
- if (XE_IOCTL_DBG(xe, args->flags & DRM_XE_VM_CREATE_SCRATCH_PAGE &&
- args->flags & DRM_XE_VM_CREATE_FAULT_MODE))
+ if (XE_IOCTL_DBG(xe, args->flags & DRM_XE_VM_CREATE_FLAG_SCRATCH_PAGE &&
+ args->flags & DRM_XE_VM_CREATE_FLAG_FAULT_MODE))
return -EINVAL;
- if (XE_IOCTL_DBG(xe, args->flags & DRM_XE_VM_CREATE_COMPUTE_MODE &&
- args->flags & DRM_XE_VM_CREATE_FAULT_MODE))
+ if (XE_IOCTL_DBG(xe, args->flags & DRM_XE_VM_CREATE_FLAG_COMPUTE_MODE &&
+ args->flags & DRM_XE_VM_CREATE_FLAG_FAULT_MODE))
return -EINVAL;
- if (XE_IOCTL_DBG(xe, args->flags & DRM_XE_VM_CREATE_FAULT_MODE &&
+ 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_FAULT_MODE) &&
+ if (XE_IOCTL_DBG(xe, !(args->flags & DRM_XE_VM_CREATE_FLAG_FAULT_MODE) &&
xe_device_in_fault_mode(xe)))
return -EINVAL;
- if (args->flags & DRM_XE_VM_CREATE_SCRATCH_PAGE)
+ 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_COMPUTE_MODE)
+ if (args->flags & DRM_XE_VM_CREATE_FLAG_COMPUTE_MODE)
flags |= XE_VM_FLAG_COMPUTE_MODE;
- if (args->flags & DRM_XE_VM_CREATE_ASYNC_BIND_OPS)
- flags |= XE_VM_FLAG_ASYNC_BIND_OPS;
- if (args->flags & DRM_XE_VM_CREATE_FAULT_MODE)
+ if (args->flags & DRM_XE_VM_CREATE_FLAG_ASYNC_DEFAULT)
+ flags |= XE_VM_FLAG_ASYNC_DEFAULT;
+ if (args->flags & DRM_XE_VM_CREATE_FLAG_FAULT_MODE)
flags |= XE_VM_FLAG_FAULT_MODE;
vm = xe_vm_create(xe, flags);
if (xe->info.has_asid) {
mutex_lock(&xe->usm.lock);
err = xa_alloc_cyclic(&xe->usm.asid_to_vm, &asid, vm,
- XA_LIMIT(0, XE_MAX_ASID - 1),
+ XA_LIMIT(1, XE_MAX_ASID - 1),
&xe->usm.next_asid, GFP_KERNEL);
mutex_unlock(&xe->usm.lock);
- if (err) {
+ if (err < 0) {
xe_vm_close_and_put(vm);
return err;
}
+ err = 0;
vm->usm.asid = asid;
}
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,
- struct async_op_fence *afence, bool first_op,
- bool last_op)
+ bool first_op, bool last_op)
{
+ struct xe_exec_queue *wait_exec_queue = to_wait_exec_queue(vm, q);
int err;
xe_assert(vm->xe, region <= ARRAY_SIZE(region_to_mem_type));
if (vma->tile_mask != (vma->tile_present & ~vma->usm.tile_invalidated)) {
return xe_vm_bind(vm, vma, q, xe_vma_bo(vma), syncs, num_syncs,
- afence, true, first_op, last_op);
+ true, first_op, last_op);
} else {
int i;
/* Nothing to do, signal fences now */
if (last_op) {
- for (i = 0; i < num_syncs; i++)
- xe_sync_entry_signal(&syncs[i], NULL,
- dma_fence_get_stub());
- }
- if (afence)
- dma_fence_signal(&afence->fence);
- return 0;
- }
-}
+ for (i = 0; i < num_syncs; i++) {
+ struct dma_fence *fence =
+ xe_exec_queue_last_fence_get(wait_exec_queue, vm);
-static void vm_set_async_error(struct xe_vm *vm, int err)
-{
- lockdep_assert_held(&vm->lock);
- vm->async_ops.error = err;
-}
-
-static int vm_bind_ioctl_lookup_vma(struct xe_vm *vm, struct xe_bo *bo,
- u64 addr, u64 range, u32 op, u32 flags)
-{
- struct xe_device *xe = vm->xe;
- struct xe_vma *vma;
- bool async = !!(flags & XE_VM_BIND_FLAG_ASYNC);
-
- lockdep_assert_held(&vm->lock);
+ xe_sync_entry_signal(&syncs[i], NULL, fence);
+ }
+ }
- switch (op) {
- case XE_VM_BIND_OP_MAP:
- case XE_VM_BIND_OP_MAP_USERPTR:
- vma = xe_vm_find_overlapping_vma(vm, addr, range);
- if (XE_IOCTL_DBG(xe, vma && !async))
- return -EBUSY;
- break;
- case XE_VM_BIND_OP_UNMAP:
- case XE_VM_BIND_OP_PREFETCH:
- vma = xe_vm_find_overlapping_vma(vm, addr, range);
- if (XE_IOCTL_DBG(xe, !vma))
- /* Not an actual error, IOCTL cleans up returns and 0 */
- return -ENODATA;
- if (XE_IOCTL_DBG(xe, (xe_vma_start(vma) != addr ||
- xe_vma_end(vma) != addr + range) && !async))
- return -EINVAL;
- break;
- case XE_VM_BIND_OP_UNMAP_ALL:
- if (XE_IOCTL_DBG(xe, list_empty(&bo->ttm.base.gpuva.list)))
- /* Not an actual error, IOCTL cleans up returns and 0 */
- return -ENODATA;
- break;
- default:
- drm_warn(&xe->drm, "NOT POSSIBLE");
- return -EINVAL;
+ return 0;
}
-
- return 0;
}
static void prep_vma_destroy(struct xe_vm *vm, struct xe_vma *vma,
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,
- u32 operation, u32 flags, u8 tile_mask, u32 region)
+ u32 operation, u32 flags, u8 tile_mask,
+ u32 prefetch_region)
{
struct drm_gem_object *obj = bo ? &bo->ttm.base : NULL;
struct drm_gpuva_ops *ops;
(ULL)bo_offset_or_userptr);
switch (operation) {
- case XE_VM_BIND_OP_MAP:
- case XE_VM_BIND_OP_MAP_USERPTR:
+ 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);
- if (IS_ERR(ops))
- return ops;
-
- drm_gpuva_for_each_op(__op, ops) {
- struct xe_vma_op *op = gpuva_op_to_vma_op(__op);
-
- op->tile_mask = tile_mask;
- op->map.immediate =
- flags & XE_VM_BIND_FLAG_IMMEDIATE;
- op->map.read_only =
- flags & XE_VM_BIND_FLAG_READONLY;
- op->map.is_null = flags & XE_VM_BIND_FLAG_NULL;
- }
break;
- case XE_VM_BIND_OP_UNMAP:
+ case DRM_XE_VM_BIND_OP_UNMAP:
ops = drm_gpuvm_sm_unmap_ops_create(&vm->gpuvm, addr, range);
- if (IS_ERR(ops))
- return ops;
-
- drm_gpuva_for_each_op(__op, ops) {
- struct xe_vma_op *op = gpuva_op_to_vma_op(__op);
-
- op->tile_mask = tile_mask;
- }
break;
- case XE_VM_BIND_OP_PREFETCH:
+ case DRM_XE_VM_BIND_OP_PREFETCH:
ops = drm_gpuvm_prefetch_ops_create(&vm->gpuvm, addr, range);
- if (IS_ERR(ops))
- return ops;
-
- drm_gpuva_for_each_op(__op, ops) {
- struct xe_vma_op *op = gpuva_op_to_vma_op(__op);
-
- op->tile_mask = tile_mask;
- op->prefetch.region = region;
- }
break;
- case XE_VM_BIND_OP_UNMAP_ALL:
+ case DRM_XE_VM_BIND_OP_UNMAP_ALL:
xe_assert(vm->xe, bo);
err = xe_bo_lock(bo, true);
ops = drm_gpuvm_bo_unmap_ops_create(vm_bo);
drm_gpuvm_bo_put(vm_bo);
xe_bo_unlock(bo);
- if (IS_ERR(ops))
- return ops;
-
- drm_gpuva_for_each_op(__op, ops) {
- struct xe_vma_op *op = gpuva_op_to_vma_op(__op);
-
- op->tile_mask = tile_mask;
- }
break;
default:
drm_warn(&vm->xe->drm, "NOT POSSIBLE");
ops = ERR_PTR(-EINVAL);
}
+ if (IS_ERR(ops))
+ return ops;
#ifdef TEST_VM_ASYNC_OPS_ERROR
if (operation & FORCE_ASYNC_OP_ERROR) {
}
#endif
- if (!IS_ERR(ops))
- drm_gpuva_for_each_op(__op, ops)
- print_op(vm->xe, __op);
+ drm_gpuva_for_each_op(__op, ops) {
+ struct xe_vma_op *op = gpuva_op_to_vma_op(__op);
+
+ op->tile_mask = tile_mask;
+ 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;
+ } else if (__op->op == DRM_GPUVA_OP_PREFETCH) {
+ op->prefetch.region = prefetch_region;
+ }
+
+ print_op(vm->xe, __op);
+ }
return ops;
}
op->flags |= XE_VMA_OP_COMMITTED;
break;
case DRM_GPUVA_OP_REMAP:
+ {
+ u8 tile_present =
+ gpuva_to_vma(op->base.remap.unmap->va)->tile_present;
+
prep_vma_destroy(vm, gpuva_to_vma(op->base.remap.unmap->va),
true);
op->flags |= XE_VMA_OP_COMMITTED;
err |= xe_vm_insert_vma(vm, op->remap.prev);
if (!err)
op->flags |= XE_VMA_OP_PREV_COMMITTED;
- if (!err && op->remap.skip_prev)
+ if (!err && op->remap.skip_prev) {
+ op->remap.prev->tile_present =
+ tile_present;
op->remap.prev = NULL;
+ }
}
if (op->remap.next) {
err |= xe_vm_insert_vma(vm, op->remap.next);
if (!err)
op->flags |= XE_VMA_OP_NEXT_COMMITTED;
- if (!err && op->remap.skip_next)
+ if (!err && op->remap.skip_next) {
+ op->remap.next->tile_present =
+ tile_present;
op->remap.next = NULL;
+ }
}
/* Adjust for partial unbind after removin VMA from VM */
op->base.remap.unmap->va->va.range = op->remap.range;
}
break;
+ }
case DRM_GPUVA_OP_UNMAP:
prep_vma_destroy(vm, gpuva_to_vma(op->base.unmap.va), true);
op->flags |= XE_VMA_OP_COMMITTED;
bool async)
{
struct xe_vma_op *last_op = NULL;
- struct async_op_fence *fence = NULL;
struct drm_gpuva_op *__op;
int err = 0;
lockdep_assert_held_write(&vm->lock);
- if (last && num_syncs && async) {
- u64 seqno;
-
- fence = kmalloc(sizeof(*fence), GFP_KERNEL);
- if (!fence)
- return -ENOMEM;
-
- seqno = q ? ++q->bind.fence_seqno : ++vm->async_ops.fence.seqno;
- dma_fence_init(&fence->fence, &async_op_fence_ops,
- &vm->async_ops.lock, q ? q->bind.fence_ctx :
- vm->async_ops.fence.context, seqno);
-
- if (!xe_vm_no_dma_fences(vm)) {
- fence->vm = vm;
- fence->started = false;
- init_waitqueue_head(&fence->wq);
- }
- }
-
drm_gpuva_for_each_op(__op, ops) {
struct xe_vma_op *op = gpuva_op_to_vma_op(__op);
bool first = list_empty(ops_list);
- xe_assert(vm->xe, first || async);
-
INIT_LIST_HEAD(&op->link);
list_add_tail(&op->link, ops_list);
vma = new_vma(vm, &op->base.map,
op->tile_mask, op->map.read_only,
op->map.is_null);
- if (IS_ERR(vma)) {
- err = PTR_ERR(vma);
- goto free_fence;
- }
+ if (IS_ERR(vma))
+ return PTR_ERR(vma);
op->map.vma = vma;
break;
vma = new_vma(vm, op->base.remap.prev,
op->tile_mask, read_only,
is_null);
- if (IS_ERR(vma)) {
- err = PTR_ERR(vma);
- goto free_fence;
- }
+ if (IS_ERR(vma))
+ return PTR_ERR(vma);
op->remap.prev = vma;
vma = new_vma(vm, op->base.remap.next,
op->tile_mask, read_only,
is_null);
- if (IS_ERR(vma)) {
- err = PTR_ERR(vma);
- goto free_fence;
- }
+ if (IS_ERR(vma))
+ return PTR_ERR(vma);
op->remap.next = vma;
err = xe_vma_op_commit(vm, op);
if (err)
- goto free_fence;
+ return err;
}
/* FIXME: Unhandled corner case */
XE_WARN_ON(!last_op && last && !list_empty(ops_list));
if (!last_op)
- goto free_fence;
+ 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;
- last_op->fence = fence;
}
return 0;
-
-free_fence:
- kfree(fence);
- return err;
}
static int op_execute(struct drm_exec *exec, struct xe_vm *vm,
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->fence,
+ 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);
bool next = !!op->remap.next;
if (!op->remap.unmap_done) {
- if (prev || next) {
- vm->async_ops.munmap_rebind_inflight = true;
+ if (prev || next)
vma->gpuva.flags |= XE_VMA_FIRST_REBIND;
- }
err = xe_vm_unbind(vm, vma, op->q, op->syncs,
op->num_syncs,
- !prev && !next ? op->fence : NULL,
op->flags & XE_VMA_OP_FIRST,
- op->flags & XE_VMA_OP_LAST && !prev &&
- !next);
+ op->flags & XE_VMA_OP_LAST &&
+ !prev && !next);
if (err)
break;
op->remap.unmap_done = true;
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,
- !next ? op->fence : NULL, true, false,
+ op->num_syncs, true, false,
op->flags & XE_VMA_OP_LAST && !next);
op->remap.prev->gpuva.flags &= ~XE_VMA_LAST_REBIND;
if (err)
err = xe_vm_bind(vm, op->remap.next, op->q,
xe_vma_bo(op->remap.next),
op->syncs, op->num_syncs,
- op->fence, true, false,
+ true, false,
op->flags & XE_VMA_OP_LAST);
op->remap.next->gpuva.flags &= ~XE_VMA_LAST_REBIND;
if (err)
break;
op->remap.next = NULL;
}
- vm->async_ops.munmap_rebind_inflight = false;
break;
}
case DRM_GPUVA_OP_UNMAP:
err = xe_vm_unbind(vm, vma, op->q, op->syncs,
- op->num_syncs, op->fence,
- op->flags & XE_VMA_OP_FIRST,
+ 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->fence,
+ op->syncs, op->num_syncs,
op->flags & XE_VMA_OP_FIRST,
op->flags & XE_VMA_OP_LAST);
break;
kfree(op->syncs);
if (op->q)
xe_exec_queue_put(op->q);
- if (op->fence)
- dma_fence_put(&op->fence->fence);
}
- if (!list_empty(&op->link)) {
- spin_lock_irq(&vm->async_ops.lock);
+ if (!list_empty(&op->link))
list_del(&op->link);
- spin_unlock_irq(&vm->async_ops.lock);
- }
if (op->ops)
drm_gpuva_ops_free(&vm->gpuvm, op->ops);
if (last)
}
}
-static struct xe_vma_op *next_vma_op(struct xe_vm *vm)
-{
- return list_first_entry_or_null(&vm->async_ops.pending,
- struct xe_vma_op, link);
-}
-
-static void xe_vma_op_work_func(struct work_struct *w)
-{
- struct xe_vm *vm = container_of(w, struct xe_vm, async_ops.work);
-
- for (;;) {
- struct xe_vma_op *op;
- int err;
-
- if (vm->async_ops.error && !xe_vm_is_closed(vm))
- break;
-
- spin_lock_irq(&vm->async_ops.lock);
- op = next_vma_op(vm);
- spin_unlock_irq(&vm->async_ops.lock);
-
- if (!op)
- break;
-
- if (!xe_vm_is_closed(vm)) {
- down_write(&vm->lock);
- err = xe_vma_op_execute(vm, op);
- if (err) {
- drm_warn(&vm->xe->drm,
- "Async VM op(%d) failed with %d",
- op->base.op, err);
- vm_set_async_error(vm, err);
- up_write(&vm->lock);
-
- break;
- }
- up_write(&vm->lock);
- } else {
- struct xe_vma *vma;
-
- switch (op->base.op) {
- case DRM_GPUVA_OP_REMAP:
- vma = gpuva_to_vma(op->base.remap.unmap->va);
- trace_xe_vma_flush(vma);
-
- down_write(&vm->lock);
- xe_vma_destroy_unlocked(vma);
- up_write(&vm->lock);
- break;
- case DRM_GPUVA_OP_UNMAP:
- vma = gpuva_to_vma(op->base.unmap.va);
- trace_xe_vma_flush(vma);
-
- down_write(&vm->lock);
- xe_vma_destroy_unlocked(vma);
- up_write(&vm->lock);
- break;
- default:
- /* Nothing to do */
- break;
- }
-
- if (op->fence && !test_bit(DMA_FENCE_FLAG_SIGNALED_BIT,
- &op->fence->fence.flags)) {
- if (!xe_vm_no_dma_fences(vm)) {
- op->fence->started = true;
- wake_up_all(&op->fence->wq);
- }
- dma_fence_signal(&op->fence->fence);
- }
- }
-
- xe_vma_op_cleanup(vm, op);
- }
-}
-
-static int vm_bind_ioctl_ops_execute(struct xe_vm *vm,
- struct list_head *ops_list, bool async)
-{
- struct xe_vma_op *op, *last_op, *next;
- int err;
-
- lockdep_assert_held_write(&vm->lock);
-
- last_op = list_last_entry(ops_list, struct xe_vma_op, link);
-
- if (!async) {
- err = xe_vma_op_execute(vm, last_op);
- if (err)
- goto unwind;
- xe_vma_op_cleanup(vm, last_op);
- } else {
- int i;
- bool installed = false;
-
- for (i = 0; i < last_op->num_syncs; i++)
- installed |= xe_sync_entry_signal(&last_op->syncs[i],
- NULL,
- &last_op->fence->fence);
- if (!installed && last_op->fence)
- dma_fence_signal(&last_op->fence->fence);
-
- spin_lock_irq(&vm->async_ops.lock);
- list_splice_tail(ops_list, &vm->async_ops.pending);
- spin_unlock_irq(&vm->async_ops.lock);
-
- if (!vm->async_ops.error)
- queue_work(system_unbound_wq, &vm->async_ops.work);
- }
-
- return 0;
-
-unwind:
- list_for_each_entry_reverse(op, ops_list, link)
- 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);
- list_for_each_entry_safe(op, next, ops_list, link)
- xe_vma_op_cleanup(vm, op);
-
- return err;
-}
-
static void vm_bind_ioctl_ops_unwind(struct xe_vm *vm,
struct drm_gpuva_ops **ops,
int num_ops_list)
}
}
+static int vm_bind_ioctl_ops_execute(struct xe_vm *vm,
+ struct list_head *ops_list)
+{
+ struct xe_vma_op *op, *next;
+ int err;
+
+ 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_vma_op_cleanup(vm, op);
+ }
+
+ return 0;
+}
+
#ifdef TEST_VM_ASYNC_OPS_ERROR
#define SUPPORTED_FLAGS \
- (FORCE_ASYNC_OP_ERROR | XE_VM_BIND_FLAG_ASYNC | \
- XE_VM_BIND_FLAG_READONLY | XE_VM_BIND_FLAG_IMMEDIATE | \
- XE_VM_BIND_FLAG_NULL | 0xffff)
+ (FORCE_ASYNC_OP_ERROR | DRM_XE_VM_BIND_FLAG_ASYNC | \
+ DRM_XE_VM_BIND_FLAG_READONLY | DRM_XE_VM_BIND_FLAG_IMMEDIATE | \
+ DRM_XE_VM_BIND_FLAG_NULL | 0xffff)
#else
#define SUPPORTED_FLAGS \
- (XE_VM_BIND_FLAG_ASYNC | XE_VM_BIND_FLAG_READONLY | \
- XE_VM_BIND_FLAG_IMMEDIATE | XE_VM_BIND_FLAG_NULL | 0xffff)
+ (DRM_XE_VM_BIND_FLAG_ASYNC | DRM_XE_VM_BIND_FLAG_READONLY | \
+ DRM_XE_VM_BIND_FLAG_IMMEDIATE | DRM_XE_VM_BIND_FLAG_NULL | \
+ 0xffff)
#endif
#define XE_64K_PAGE_MASK 0xffffull
u32 flags = (*bind_ops)[i].flags;
u32 obj = (*bind_ops)[i].obj;
u64 obj_offset = (*bind_ops)[i].obj_offset;
- u32 region = (*bind_ops)[i].region;
- bool is_null = flags & XE_VM_BIND_FLAG_NULL;
+ u32 prefetch_region = (*bind_ops)[i].prefetch_mem_region_instance;
+ bool is_null = flags & DRM_XE_VM_BIND_FLAG_NULL;
if (i == 0) {
- *async = !!(flags & XE_VM_BIND_FLAG_ASYNC);
- } else if (XE_IOCTL_DBG(xe, !*async) ||
- XE_IOCTL_DBG(xe, !(flags & XE_VM_BIND_FLAG_ASYNC)) ||
- XE_IOCTL_DBG(xe, op == XE_VM_BIND_OP_RESTART)) {
- err = -EINVAL;
- goto free_bind_ops;
- }
-
- if (XE_IOCTL_DBG(xe, !*async &&
- op == XE_VM_BIND_OP_UNMAP_ALL)) {
- err = -EINVAL;
- goto free_bind_ops;
- }
-
- if (XE_IOCTL_DBG(xe, !*async &&
- op == XE_VM_BIND_OP_PREFETCH)) {
+ *async = !!(flags & DRM_XE_VM_BIND_FLAG_ASYNC);
+ if (XE_IOCTL_DBG(xe, !*async && args->num_syncs)) {
+ err = -EINVAL;
+ goto free_bind_ops;
+ }
+ } else if (XE_IOCTL_DBG(xe, *async !=
+ !!(flags & DRM_XE_VM_BIND_FLAG_ASYNC))) {
err = -EINVAL;
goto free_bind_ops;
}
- if (XE_IOCTL_DBG(xe, op > XE_VM_BIND_OP_PREFETCH) ||
+ 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, op != XE_VM_BIND_OP_MAP &&
+ XE_IOCTL_DBG(xe, op != DRM_XE_VM_BIND_OP_MAP &&
is_null) ||
XE_IOCTL_DBG(xe, !obj &&
- op == XE_VM_BIND_OP_MAP &&
+ op == DRM_XE_VM_BIND_OP_MAP &&
!is_null) ||
XE_IOCTL_DBG(xe, !obj &&
- op == XE_VM_BIND_OP_UNMAP_ALL) ||
+ op == DRM_XE_VM_BIND_OP_UNMAP_ALL) ||
XE_IOCTL_DBG(xe, addr &&
- op == XE_VM_BIND_OP_UNMAP_ALL) ||
+ op == DRM_XE_VM_BIND_OP_UNMAP_ALL) ||
XE_IOCTL_DBG(xe, range &&
- op == XE_VM_BIND_OP_UNMAP_ALL) ||
+ op == DRM_XE_VM_BIND_OP_UNMAP_ALL) ||
XE_IOCTL_DBG(xe, obj &&
- op == XE_VM_BIND_OP_MAP_USERPTR) ||
+ op == DRM_XE_VM_BIND_OP_MAP_USERPTR) ||
XE_IOCTL_DBG(xe, obj &&
- op == XE_VM_BIND_OP_PREFETCH) ||
- XE_IOCTL_DBG(xe, region &&
- op != XE_VM_BIND_OP_PREFETCH) ||
- XE_IOCTL_DBG(xe, !(BIT(region) &
+ 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, obj &&
- op == XE_VM_BIND_OP_UNMAP)) {
+ op == DRM_XE_VM_BIND_OP_UNMAP)) {
err = -EINVAL;
goto free_bind_ops;
}
if (XE_IOCTL_DBG(xe, obj_offset & ~PAGE_MASK) ||
XE_IOCTL_DBG(xe, addr & ~PAGE_MASK) ||
XE_IOCTL_DBG(xe, range & ~PAGE_MASK) ||
- XE_IOCTL_DBG(xe, !range && op !=
- XE_VM_BIND_OP_RESTART &&
- op != XE_VM_BIND_OP_UNMAP_ALL)) {
+ XE_IOCTL_DBG(xe, !range &&
+ op != DRM_XE_VM_BIND_OP_UNMAP_ALL)) {
err = -EINVAL;
goto free_bind_ops;
}
err = -EINVAL;
goto put_exec_queue;
}
+
+ if (XE_IOCTL_DBG(xe, async !=
+ !!(q->flags & EXEC_QUEUE_FLAG_VM_ASYNC))) {
+ err = -EINVAL;
+ goto put_exec_queue;
+ }
}
vm = xe_vm_lookup(xef, args->vm_id);
goto put_exec_queue;
}
+ if (!args->exec_queue_id) {
+ if (XE_IOCTL_DBG(xe, async !=
+ !!(vm->flags & XE_VM_FLAG_ASYNC_DEFAULT))) {
+ err = -EINVAL;
+ goto put_vm;
+ }
+ }
+
err = down_write_killable(&vm->lock);
if (err)
goto put_vm;
goto release_vm_lock;
}
- if (bind_ops[0].op == XE_VM_BIND_OP_RESTART) {
- if (XE_IOCTL_DBG(xe, !(vm->flags & XE_VM_FLAG_ASYNC_BIND_OPS)))
- err = -EOPNOTSUPP;
- if (XE_IOCTL_DBG(xe, !err && args->num_syncs))
- err = EINVAL;
- if (XE_IOCTL_DBG(xe, !err && !vm->async_ops.error))
- err = -EPROTO;
-
- if (!err) {
- trace_xe_vm_restart(vm);
- vm_set_async_error(vm, 0);
-
- queue_work(system_unbound_wq, &vm->async_ops.work);
-
- /* Rebinds may have been blocked, give worker a kick */
- if (xe_vm_in_compute_mode(vm))
- xe_vm_queue_rebind_worker(vm);
- }
-
- goto release_vm_lock;
- }
-
- if (XE_IOCTL_DBG(xe, !vm->async_ops.error &&
- async != !!(vm->flags & XE_VM_FLAG_ASYNC_BIND_OPS))) {
- err = -EOPNOTSUPP;
- goto release_vm_lock;
- }
-
for (i = 0; i < args->num_binds; ++i) {
u64 range = bind_ops[i].range;
u64 addr = bind_ops[i].addr;
goto free_syncs;
}
- /* Do some error checking first to make the unwind easier */
- for (i = 0; i < args->num_binds; ++i) {
- u64 range = bind_ops[i].range;
- u64 addr = bind_ops[i].addr;
- u32 op = bind_ops[i].op;
- u32 flags = bind_ops[i].flags;
-
- err = vm_bind_ioctl_lookup_vma(vm, bos[i], addr, range, op, flags);
- if (err)
- goto free_syncs;
- }
-
for (i = 0; i < args->num_binds; ++i) {
u64 range = bind_ops[i].range;
u64 addr = bind_ops[i].addr;
u32 flags = bind_ops[i].flags;
u64 obj_offset = bind_ops[i].obj_offset;
u8 tile_mask = bind_ops[i].tile_mask;
- u32 region = bind_ops[i].region;
+ u32 prefetch_region = bind_ops[i].prefetch_mem_region_instance;
ops[i] = vm_bind_ioctl_ops_create(vm, bos[i], obj_offset,
addr, range, op, flags,
- tile_mask, region);
+ tile_mask, prefetch_region);
if (IS_ERR(ops[i])) {
err = PTR_ERR(ops[i]);
ops[i] = NULL;
goto unwind_ops;
}
- err = vm_bind_ioctl_ops_execute(vm, &ops_list, async);
+ xe_vm_get(vm);
+ if (q)
+ xe_exec_queue_get(q);
+
+ err = vm_bind_ioctl_ops_execute(vm, &ops_list);
+
up_write(&vm->lock);
- for (i = 0; i < args->num_binds; ++i)
+ 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]);
kfree(bos);
unwind_ops:
vm_bind_ioctl_ops_unwind(vm, ops, args->num_binds);
free_syncs:
- for (i = 0; err == -ENODATA && i < num_syncs; i++)
- xe_sync_entry_signal(&syncs[i], NULL, dma_fence_get_stub());
+ for (i = 0; err == -ENODATA && i < num_syncs; i++) {
+ struct dma_fence *fence =
+ xe_exec_queue_last_fence_get(to_wait_exec_queue(vm, q), vm);
+
+ xe_sync_entry_signal(&syncs[i], NULL, fence);
+ }
while (num_syncs--)
xe_sync_entry_cleanup(&syncs[num_syncs]);
projects / linux-2.6-microblaze.git / blobdiff