[PATCH v5 4/7] drm/xe: Convert multiple bind ops into single job

Matthew Auld matthew.auld at intel.com
Wed Jun 26 14:03:41 UTC 2024


On 26/06/2024 01:39, Matthew Brost wrote:
> This aligns with the uAPI of an array of binds or single bind that
> results in multiple GPUVA ops to be considered a single atomic
> operations.
> 
> The implemenation is roughly:
> - xe_vma_ops is a list of xe_vma_op (GPUVA op)
> - each xe_vma_op resolves to 0-3 PT ops
> - xe_vma_ops creates a single job
> - if at any point during binding a failure occurs, xe_vma_ops contains
>    the information necessary unwind the PT and VMA (GPUVA) state
> 
> v2:
>   - add missing dma-resv slot reservation (CI, testing)
> v4:
>   - Fix TLB invalidation (Paulo)
>   - Add missing xe_sched_job_last_fence_add/test_dep check (Inspection)
> v5:
>   - Invert i, j usage (Matthew Auld)
>   - Add helper to test and add job dep (Matthew Auld)
>   - Return on anything but -ETIME for cpu bind (Matthew Auld)
>   - Return -ENOBUFS if suballoc of BB fails due to size (Matthew Auld)
>   - s/do/Do (Matthew Auld)
>   - Add missing comma (Matthew Auld)
>   - Do not assign return value to xe_range_fence_insert (Matthew Auld)
> 
> Cc: Thomas Hellström <thomas.hellstrom at linux.intel.com>
> Signed-off-by: Matthew Brost <matthew.brost at intel.com>
> ---
>   drivers/gpu/drm/xe/xe_bo_types.h |    2 +
>   drivers/gpu/drm/xe/xe_migrate.c  |  306 ++++-----
>   drivers/gpu/drm/xe/xe_migrate.h  |   32 +-
>   drivers/gpu/drm/xe/xe_pt.c       | 1102 +++++++++++++++++++-----------
>   drivers/gpu/drm/xe/xe_pt.h       |   14 +-
>   drivers/gpu/drm/xe/xe_pt_types.h |   36 +
>   drivers/gpu/drm/xe/xe_vm.c       |  519 +++-----------
>   drivers/gpu/drm/xe/xe_vm.h       |    2 +
>   drivers/gpu/drm/xe/xe_vm_types.h |   45 +-
>   9 files changed, 1036 insertions(+), 1022 deletions(-)
> 
> diff --git a/drivers/gpu/drm/xe/xe_bo_types.h b/drivers/gpu/drm/xe/xe_bo_types.h
> index 86422e113d39..02d68873558a 100644
> --- a/drivers/gpu/drm/xe/xe_bo_types.h
> +++ b/drivers/gpu/drm/xe/xe_bo_types.h
> @@ -58,6 +58,8 @@ struct xe_bo {
>   #endif
>   	/** @freed: List node for delayed put. */
>   	struct llist_node freed;
> +	/** @update_index: Update index if PT BO */
> +	int update_index;
>   	/** @created: Whether the bo has passed initial creation */
>   	bool created;
>   
> diff --git a/drivers/gpu/drm/xe/xe_migrate.c b/drivers/gpu/drm/xe/xe_migrate.c
> index af62783d34ac..160bfcd510ae 100644
> --- a/drivers/gpu/drm/xe/xe_migrate.c
> +++ b/drivers/gpu/drm/xe/xe_migrate.c
> @@ -1125,6 +1125,7 @@ struct dma_fence *xe_migrate_clear(struct xe_migrate *m,
>   }
>   
>   static void write_pgtable(struct xe_tile *tile, struct xe_bb *bb, u64 ppgtt_ofs,
> +			  const struct xe_vm_pgtable_update_op *pt_op,
>   			  const struct xe_vm_pgtable_update *update,
>   			  struct xe_migrate_pt_update *pt_update)
>   {
> @@ -1159,8 +1160,12 @@ static void write_pgtable(struct xe_tile *tile, struct xe_bb *bb, u64 ppgtt_ofs,
>   		bb->cs[bb->len++] = MI_STORE_DATA_IMM | MI_SDI_NUM_QW(chunk);
>   		bb->cs[bb->len++] = lower_32_bits(addr);
>   		bb->cs[bb->len++] = upper_32_bits(addr);
> -		ops->populate(pt_update, tile, NULL, bb->cs + bb->len, ofs, chunk,
> -			      update);
> +		if (pt_op->bind)
> +			ops->populate(pt_update, tile, NULL, bb->cs + bb->len,
> +				      ofs, chunk, update);
> +		else
> +			ops->clear(pt_update, tile, NULL, bb->cs + bb->len,
> +				   ofs, chunk, update);
>   
>   		bb->len += chunk * 2;
>   		ofs += chunk;
> @@ -1185,114 +1190,58 @@ struct migrate_test_params {
>   
>   static struct dma_fence *
>   xe_migrate_update_pgtables_cpu(struct xe_migrate *m,
> -			       struct xe_vm *vm, struct xe_bo *bo,
> -			       const struct  xe_vm_pgtable_update *updates,
> -			       u32 num_updates, bool wait_vm,
>   			       struct xe_migrate_pt_update *pt_update)
>   {
>   	XE_TEST_DECLARE(struct migrate_test_params *test =
>   			to_migrate_test_params
>   			(xe_cur_kunit_priv(XE_TEST_LIVE_MIGRATE));)
>   	const struct xe_migrate_pt_update_ops *ops = pt_update->ops;
> -	struct dma_fence *fence;
> +	struct xe_vm *vm = pt_update->vops->vm;
> +	struct xe_vm_pgtable_update_ops *pt_update_ops =
> +		&pt_update->vops->pt_update_ops[pt_update->tile_id];
>   	int err;
> -	u32 i;
> +	u32 i, j;
>   
>   	if (XE_TEST_ONLY(test && test->force_gpu))
>   		return ERR_PTR(-ETIME);
>   
> -	if (bo && !dma_resv_test_signaled(bo->ttm.base.resv,
> -					  DMA_RESV_USAGE_KERNEL))
> -		return ERR_PTR(-ETIME);
> -
> -	if (wait_vm && !dma_resv_test_signaled(xe_vm_resv(vm),
> -					       DMA_RESV_USAGE_BOOKKEEP))
> -		return ERR_PTR(-ETIME);
> -
>   	if (ops->pre_commit) {
>   		pt_update->job = NULL;
>   		err = ops->pre_commit(pt_update);
>   		if (err)
>   			return ERR_PTR(err);
>   	}
> -	for (i = 0; i < num_updates; i++) {
> -		const struct xe_vm_pgtable_update *update = &updates[i];
> -
> -		ops->populate(pt_update, m->tile, &update->pt_bo->vmap, NULL,
> -			      update->ofs, update->qwords, update);
> -	}
> -
> -	if (vm) {
> -		trace_xe_vm_cpu_bind(vm);
> -		xe_device_wmb(vm->xe);
> -	}
> -
> -	fence = dma_fence_get_stub();
> -
> -	return fence;
> -}
> -
> -static bool no_in_syncs(struct xe_vm *vm, struct xe_exec_queue *q,
> -			struct xe_sync_entry *syncs, u32 num_syncs)
> -{
> -	struct dma_fence *fence;
> -	int i;
>   
> -	for (i = 0; i < num_syncs; i++) {
> -		fence = syncs[i].fence;
> -
> -		if (fence && !test_bit(DMA_FENCE_FLAG_SIGNALED_BIT,
> -				       &fence->flags))
> -			return false;
> -	}
> -	if (q) {
> -		fence = xe_exec_queue_last_fence_get(q, vm);
> -		if (!test_bit(DMA_FENCE_FLAG_SIGNALED_BIT, &fence->flags)) {
> -			dma_fence_put(fence);
> -			return false;
> +	for (i = 0; i < pt_update_ops->num_ops; ++i) {
> +		const struct xe_vm_pgtable_update_op *pt_op =
> +			&pt_update_ops->ops[i];
> +
> +		for (j = 0; j < pt_op->num_entries; j++) {
> +			const struct xe_vm_pgtable_update *update =
> +				&pt_op->entries[j];
> +
> +			if (pt_op->bind)
> +				ops->populate(pt_update, m->tile,
> +					      &update->pt_bo->vmap, NULL,
> +					      update->ofs, update->qwords,
> +					      update);
> +			else
> +				ops->clear(pt_update, m->tile,
> +					   &update->pt_bo->vmap, NULL,
> +					   update->ofs, update->qwords, update);
>   		}
> -		dma_fence_put(fence);
>   	}
>   
> -	return true;
> +	trace_xe_vm_cpu_bind(vm);
> +	xe_device_wmb(vm->xe);
> +
> +	return dma_fence_get_stub();
>   }
>   
> -/**
> - * xe_migrate_update_pgtables() - Pipelined page-table update
> - * @m: The migrate context.
> - * @vm: The vm we'll be updating.
> - * @bo: The bo whose dma-resv we will await before updating, or NULL if userptr.
> - * @q: The exec queue to be used for the update or NULL if the default
> - * migration engine is to be used.
> - * @updates: An array of update descriptors.
> - * @num_updates: Number of descriptors in @updates.
> - * @syncs: Array of xe_sync_entry to await before updating. Note that waits
> - * will block the engine timeline.
> - * @num_syncs: Number of entries in @syncs.
> - * @pt_update: Pointer to a struct xe_migrate_pt_update, which contains
> - * pointers to callback functions and, if subclassed, private arguments to
> - * those.
> - *
> - * Perform a pipelined page-table update. The update descriptors are typically
> - * built under the same lock critical section as a call to this function. If
> - * using the default engine for the updates, they will be performed in the
> - * order they grab the job_mutex. If different engines are used, external
> - * synchronization is needed for overlapping updates to maintain page-table
> - * consistency. Note that the meaing of "overlapping" is that the updates
> - * touch the same page-table, which might be a higher-level page-directory.
> - * If no pipelining is needed, then updates may be performed by the cpu.
> - *
> - * Return: A dma_fence that, when signaled, indicates the update completion.
> - */
> -struct dma_fence *
> -xe_migrate_update_pgtables(struct xe_migrate *m,
> -			   struct xe_vm *vm,
> -			   struct xe_bo *bo,
> -			   struct xe_exec_queue *q,
> -			   const struct xe_vm_pgtable_update *updates,
> -			   u32 num_updates,
> -			   struct xe_sync_entry *syncs, u32 num_syncs,
> -			   struct xe_migrate_pt_update *pt_update)
> +static struct dma_fence *
> +__xe_migrate_update_pgtables(struct xe_migrate *m,
> +			     struct xe_migrate_pt_update *pt_update,
> +			     struct xe_vm_pgtable_update_ops *pt_update_ops)
>   {
>   	const struct xe_migrate_pt_update_ops *ops = pt_update->ops;
>   	struct xe_tile *tile = m->tile;
> @@ -1301,59 +1250,53 @@ xe_migrate_update_pgtables(struct xe_migrate *m,
>   	struct xe_sched_job *job;
>   	struct dma_fence *fence;
>   	struct drm_suballoc *sa_bo = NULL;
> -	struct xe_vma *vma = pt_update->vma;
>   	struct xe_bb *bb;
> -	u32 i, batch_size, ppgtt_ofs, update_idx, page_ofs = 0;
> +	u32 i, j, batch_size = 0, ppgtt_ofs, update_idx, page_ofs = 0;
> +	u32 num_updates = 0, current_update = 0;
>   	u64 addr;
>   	int err = 0;
> -	bool usm = !q && xe->info.has_usm;
> -	bool first_munmap_rebind = vma &&
> -		vma->gpuva.flags & XE_VMA_FIRST_REBIND;
> -	struct xe_exec_queue *q_override = !q ? m->q : q;
> -	u16 pat_index = xe->pat.idx[XE_CACHE_WB];
> +	bool is_migrate = pt_update_ops->q == m->q;
> +	bool usm = is_migrate && xe->info.has_usm;
> +
> +	for (i = 0; i < pt_update_ops->num_ops; ++i) {
> +		struct xe_vm_pgtable_update_op *pt_op = &pt_update_ops->ops[i];
> +		struct xe_vm_pgtable_update *updates = pt_op->entries;
>   
> -	/* Use the CPU if no in syncs and engine is idle */
> -	if (no_in_syncs(vm, q, syncs, num_syncs) && xe_exec_queue_is_idle(q_override)) {
> -		fence =  xe_migrate_update_pgtables_cpu(m, vm, bo, updates,
> -							num_updates,
> -							first_munmap_rebind,
> -							pt_update);
> -		if (!IS_ERR(fence) || fence == ERR_PTR(-EAGAIN))
> -			return fence;
> +		num_updates += pt_op->num_entries;
> +		for (j = 0; j < pt_op->num_entries; ++j) {
> +			u32 num_cmds = DIV_ROUND_UP(updates[j].qwords, 0x1ff);

Why 0x1ff here? Should it not be MAX_PTE_PER_SDI? There is a failure in 
CI here: 
https://intel-gfx-ci.01.org/tree/intel-xe/xe-pw-133034v5/shard-lnl-8/igt@xe_vm@mmap-style-bind-either-side-partial-split-page-hammer.html

Wondering if this is the cause? i.e we think we can fit more stores per 
MI_STORE_DATA_IMM, since write_pgtable() below is using MAX_PTE_PER_SDI 
and not 0x1ff?

> +
> +			/* align noop + MI_STORE_DATA_IMM cmd prefix */
> +			batch_size += 4 * num_cmds + updates[j].qwords * 2;
> +		}
>   	}
>   
>   	/* fixed + PTE entries */
>   	if (IS_DGFX(xe))
> -		batch_size = 2;
> +		batch_size += 2;
>   	else
> -		batch_size = 6 + num_updates * 2;
> -
> -	for (i = 0; i < num_updates; i++) {
> -		u32 num_cmds = DIV_ROUND_UP(updates[i].qwords, MAX_PTE_PER_SDI);
> +		batch_size += 6 + num_updates * 2;
>   
> -		/* align noop + MI_STORE_DATA_IMM cmd prefix */
> -		batch_size += 4 * num_cmds + updates[i].qwords * 2;
> -	}
> -
> -	/*
> -	 * XXX: Create temp bo to copy from, if batch_size becomes too big?
> -	 *
> -	 * Worst case: Sum(2 * (each lower level page size) + (top level page size))
> -	 * Should be reasonably bound..
> -	 */
> -	xe_tile_assert(tile, batch_size < SZ_128K);
> +	bb = xe_bb_new(gt, batch_size, usm);

So do we now validate that batch_size fits within the total pool size 
somewhere, to avoid triggering the warning in drm_suballoc_new? Did your 
igt not trigger the warning? Also the below check is not too late?

> +	if (IS_ERR(bb)) {
> +		/*
> +		 * BB to large, return -ENOBUFS indicating user should split
> +		 * array of binds into smaller chunks.
> +		 */
> +		if (PTR_ERR(bb) == -EINVAL)
> +			return ERR_PTR(-ENOBUFS);
>   
> -	bb = xe_bb_new(gt, batch_size, !q && xe->info.has_usm);
> -	if (IS_ERR(bb))
>   		return ERR_CAST(bb);
> +	}
>   
>   	/* For sysmem PTE's, need to map them in our hole.. */
>   	if (!IS_DGFX(xe)) {
>   		ppgtt_ofs = NUM_KERNEL_PDE - 1;
> -		if (q) {
> -			xe_tile_assert(tile, num_updates <= NUM_VMUSA_WRITES_PER_UNIT);
> +		if (!is_migrate) {
> +			u32 num_units = DIV_ROUND_UP(num_updates,
> +						     NUM_VMUSA_WRITES_PER_UNIT);
>   
> -			sa_bo = drm_suballoc_new(&m->vm_update_sa, 1,
> +			sa_bo = drm_suballoc_new(&m->vm_update_sa, num_units,
>   						 GFP_KERNEL, true, 0);

And maybe here also?

>   			if (IS_ERR(sa_bo)) {
>   				err = PTR_ERR(sa_bo);
> @@ -1373,14 +1316,26 @@ xe_migrate_update_pgtables(struct xe_migrate *m,
>   		bb->cs[bb->len++] = ppgtt_ofs * XE_PAGE_SIZE + page_ofs;
>   		bb->cs[bb->len++] = 0; /* upper_32_bits */

Off camera this is doing:

b->cs[bb->len++] = MI_STORE_DATA_IMM | MI_SDI_NUM_QW(num_updates);

What are certain that num_updates still can be done with single command 
here?  Do we not need to break it up like we do with write_pgtable? If 
it's not an issue I think would be good to add an assert to ensure it fits?

<snip>


More information about the Intel-xe mailing list