drivers/gpu/drm/i915/gvt/scheduler.c

/*
 * Copyright(c) 2011-2016 Intel Corporation. All rights reserved.
 *
 * Permission is hereby granted, free of charge, to any person obtaining a
 * copy of this software and associated documentation files (the "Software"),
 * to deal in the Software without restriction, including without limitation
 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
 * and/or sell copies of the Software, and to permit persons to whom the
 * Software is furnished to do so, subject to the following conditions:
 *
 * The above copyright notice and this permission notice (including the next
 * paragraph) shall be included in all copies or substantial portions of the
 * Software.
 *
 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
 * SOFTWARE.
 *
 * Authors:
 *    Zhi Wang <zhi.a.wang@intel.com>
 *
 * Contributors:
 *    Ping Gao <ping.a.gao@intel.com>
 *    Tina Zhang <tina.zhang@intel.com>
 *    Chanbin Du <changbin.du@intel.com>
 *    Min He <min.he@intel.com>
 *    Bing Niu <bing.niu@intel.com>
 *    Zhenyu Wang <zhenyuw@linux.intel.com>
 *
 */

#include <linux/kthread.h>

#include "i915_drv.h"
#include "gvt.h"

#define RING_CTX_OFF(x) \
	offsetof(struct execlist_ring_context, x)

static void set_context_pdp_root_pointer(
		struct execlist_ring_context *ring_context,
		u32 pdp[8])
{
	struct execlist_mmio_pair *pdp_pair = &ring_context->pdp3_UDW;
	int i;

	for (i = 0; i < 8; i++)
		pdp_pair[i].val = pdp[7 - i];
}

static int populate_shadow_context(struct intel_vgpu_workload *workload)
{
	struct intel_vgpu *vgpu = workload->vgpu;
	struct intel_gvt *gvt = vgpu->gvt;
	int ring_id = workload->ring_id;
	struct i915_gem_context *shadow_ctx = vgpu->submission.shadow_ctx;
	struct drm_i915_gem_object *ctx_obj =
		shadow_ctx->engine[ring_id].state->obj;
	struct execlist_ring_context *shadow_ring_context;
	struct page *page;
	void *dst;
	unsigned long context_gpa, context_page_num;
	int i;

	gvt_dbg_sched("ring id %d workload lrca %x", ring_id,
			workload->ctx_desc.lrca);

	context_page_num = gvt->dev_priv->engine[ring_id]->context_size;

	context_page_num = context_page_num >> PAGE_SHIFT;

	if (IS_BROADWELL(gvt->dev_priv) && ring_id == RCS)
		context_page_num = 19;

	i = 2;

	while (i < context_page_num) {
		context_gpa = intel_vgpu_gma_to_gpa(vgpu->gtt.ggtt_mm,
				(u32)((workload->ctx_desc.lrca + i) <<
				GTT_PAGE_SHIFT));
		if (context_gpa == INTEL_GVT_INVALID_ADDR) {
			gvt_vgpu_err("Invalid guest context descriptor\n");
			return -EFAULT;
		}

		page = i915_gem_object_get_page(ctx_obj, LRC_HEADER_PAGES + i);
		dst = kmap(page);
		intel_gvt_hypervisor_read_gpa(vgpu, context_gpa, dst,
				GTT_PAGE_SIZE);
		kunmap(page);
		i++;
	}

	page = i915_gem_object_get_page(ctx_obj, LRC_STATE_PN);
	shadow_ring_context = kmap(page);

#define COPY_REG(name) \
	intel_gvt_hypervisor_read_gpa(vgpu, workload->ring_context_gpa \
		+ RING_CTX_OFF(name.val), &shadow_ring_context->name.val, 4)

	COPY_REG(ctx_ctrl);
	COPY_REG(ctx_timestamp);

	if (ring_id == RCS) {
		COPY_REG(bb_per_ctx_ptr);
		COPY_REG(rcs_indirect_ctx);
		COPY_REG(rcs_indirect_ctx_offset);
	}
#undef COPY_REG

	set_context_pdp_root_pointer(shadow_ring_context,
				     workload->shadow_mm->shadow_page_table);

	intel_gvt_hypervisor_read_gpa(vgpu,
			workload->ring_context_gpa +
			sizeof(*shadow_ring_context),
			(void *)shadow_ring_context +
			sizeof(*shadow_ring_context),
			GTT_PAGE_SIZE - sizeof(*shadow_ring_context));

	kunmap(page);
	return 0;
}

static inline bool is_gvt_request(struct drm_i915_gem_request *req)
{
	return i915_gem_context_force_single_submission(req->ctx);
}

static int shadow_context_status_change(struct notifier_block *nb,
		unsigned long action, void *data)
{
	struct drm_i915_gem_request *req = (struct drm_i915_gem_request *)data;
	struct intel_gvt *gvt = container_of(nb, struct intel_gvt,
				shadow_ctx_notifier_block[req->engine->id]);
	struct intel_gvt_workload_scheduler *scheduler = &gvt->scheduler;
	enum intel_engine_id ring_id = req->engine->id;
	struct intel_vgpu_workload *workload;

	if (!is_gvt_request(req)) {
		spin_lock_bh(&scheduler->mmio_context_lock);
		if (action == INTEL_CONTEXT_SCHEDULE_IN &&
		    scheduler->engine_owner[ring_id]) {
			/* Switch ring from vGPU to host. */
			intel_gvt_switch_mmio(scheduler->engine_owner[ring_id],
					      NULL, ring_id);
			scheduler->engine_owner[ring_id] = NULL;
		}
		spin_unlock_bh(&scheduler->mmio_context_lock);

		return NOTIFY_OK;
	}

	workload = scheduler->current_workload[ring_id];
	if (unlikely(!workload))
		return NOTIFY_OK;

	switch (action) {
	case INTEL_CONTEXT_SCHEDULE_IN:
		spin_lock_bh(&scheduler->mmio_context_lock);
		if (workload->vgpu != scheduler->engine_owner[ring_id]) {
			/* Switch ring from host to vGPU or vGPU to vGPU. */
			intel_gvt_switch_mmio(scheduler->engine_owner[ring_id],
					      workload->vgpu, ring_id);
			scheduler->engine_owner[ring_id] = workload->vgpu;
		} else
			gvt_dbg_sched("skip ring %d mmio switch for vgpu%d\n",
				      ring_id, workload->vgpu->id);
		spin_unlock_bh(&scheduler->mmio_context_lock);
		atomic_set(&workload->shadow_ctx_active, 1);
		break;
	case INTEL_CONTEXT_SCHEDULE_OUT:
	case INTEL_CONTEXT_SCHEDULE_PREEMPTED:
		atomic_set(&workload->shadow_ctx_active, 0);
		break;
	default:
		WARN_ON(1);
		return NOTIFY_OK;
	}
	wake_up(&workload->shadow_ctx_status_wq);
	return NOTIFY_OK;
}

static void shadow_context_descriptor_update(struct i915_gem_context *ctx,
		struct intel_engine_cs *engine)
{
	struct intel_context *ce = &ctx->engine[engine->id];
	u64 desc = 0;

	desc = ce->lrc_desc;

	/* Update bits 0-11 of the context descriptor which includes flags
	 * like GEN8_CTX_* cached in desc_template
	 */
	desc &= U64_MAX << 12;
	desc |= ctx->desc_template & ((1ULL << 12) - 1);

	ce->lrc_desc = desc;
}

static int copy_workload_to_ring_buffer(struct intel_vgpu_workload *workload)
{
	struct intel_vgpu *vgpu = workload->vgpu;
	void *shadow_ring_buffer_va;
	u32 *cs;

	/* allocate shadow ring buffer */
	cs = intel_ring_begin(workload->req, workload->rb_len / sizeof(u32));
	if (IS_ERR(cs)) {
		gvt_vgpu_err("fail to alloc size =%ld shadow  ring buffer\n",
			workload->rb_len);
		return PTR_ERR(cs);
	}

	shadow_ring_buffer_va = workload->shadow_ring_buffer_va;

	/* get shadow ring buffer va */
	workload->shadow_ring_buffer_va = cs;

	memcpy(cs, shadow_ring_buffer_va,
			workload->rb_len);

	cs += workload->rb_len / sizeof(u32);
	intel_ring_advance(workload->req, cs);

	return 0;
}

void release_shadow_wa_ctx(struct intel_shadow_wa_ctx *wa_ctx)
{
	if (!wa_ctx->indirect_ctx.obj)
		return;

	i915_gem_object_unpin_map(wa_ctx->indirect_ctx.obj);
	i915_gem_object_put(wa_ctx->indirect_ctx.obj);
}

/**
 * intel_gvt_scan_and_shadow_workload - audit the workload by scanning and
 * shadow it as well, include ringbuffer,wa_ctx and ctx.
 * @workload: an abstract entity for each execlist submission.
 *
 * This function is called before the workload submitting to i915, to make
 * sure the content of the workload is valid.
 */
int intel_gvt_scan_and_shadow_workload(struct intel_vgpu_workload *workload)
{
	struct intel_vgpu *vgpu = workload->vgpu;
	struct intel_vgpu_submission *s = &vgpu->submission;
	struct i915_gem_context *shadow_ctx = s->shadow_ctx;
	struct drm_i915_private *dev_priv = vgpu->gvt->dev_priv;
	int ring_id = workload->ring_id;
	struct intel_engine_cs *engine = dev_priv->engine[ring_id];
	struct drm_i915_gem_request *rq;
	struct intel_ring *ring;
	int ret;

	lockdep_assert_held(&dev_priv->drm.struct_mutex);

	if (workload->shadowed)
		return 0;

	shadow_ctx->desc_template &= ~(0x3 << GEN8_CTX_ADDRESSING_MODE_SHIFT);
	shadow_ctx->desc_template |= workload->ctx_desc.addressing_mode <<
				    GEN8_CTX_ADDRESSING_MODE_SHIFT;

	if (!test_and_set_bit(ring_id, s->shadow_ctx_desc_updated))
		shadow_context_descriptor_update(shadow_ctx,
					dev_priv->engine[ring_id]);

	ret = intel_gvt_scan_and_shadow_ringbuffer(workload);
	if (ret)
		goto err_scan;

	if ((workload->ring_id == RCS) &&
	    (workload->wa_ctx.indirect_ctx.size != 0)) {
		ret = intel_gvt_scan_and_shadow_wa_ctx(&workload->wa_ctx);
		if (ret)
			goto err_scan;
	}

	/* pin shadow context by gvt even the shadow context will be pinned
	 * when i915 alloc request. That is because gvt will update the guest
	 * context from shadow context when workload is completed, and at that
	 * moment, i915 may already unpined the shadow context to make the
	 * shadow_ctx pages invalid. So gvt need to pin itself. After update
	 * the guest context, gvt can unpin the shadow_ctx safely.
	 */
	ring = engine->context_pin(engine, shadow_ctx);
	if (IS_ERR(ring)) {
		ret = PTR_ERR(ring);
		gvt_vgpu_err("fail to pin shadow context\n");
		goto err_shadow;
	}

	ret = populate_shadow_context(workload);
	if (ret)
		goto err_unpin;

	rq = i915_gem_request_alloc(dev_priv->engine[ring_id], shadow_ctx);
	if (IS_ERR(rq)) {
		gvt_vgpu_err("fail to allocate gem request\n");
		ret = PTR_ERR(rq);
		goto err_unpin;
	}

	gvt_dbg_sched("ring id %d get i915 gem request %p\n", ring_id, rq);

	workload->req = i915_gem_request_get(rq);
	ret = copy_workload_to_ring_buffer(workload);
	if (ret)
		goto err_unpin;
	workload->shadowed = true;
	return 0;

err_unpin:
	engine->context_unpin(engine, shadow_ctx);
err_shadow:
	release_shadow_wa_ctx(&workload->wa_ctx);
err_scan:
	return ret;
}

static void release_shadow_batch_buffer(struct intel_vgpu_workload *workload);

static int prepare_shadow_batch_buffer(struct intel_vgpu_workload *workload)
{
	struct intel_gvt *gvt = workload->vgpu->gvt;
	const int gmadr_bytes = gvt->device_info.gmadr_bytes_in_cmd;
	struct intel_vgpu_shadow_bb *bb;
	int ret;

	list_for_each_entry(bb, &workload->shadow_bb, list) {
		bb->vma = i915_gem_object_ggtt_pin(bb->obj, NULL, 0, 0, 0);
		if (IS_ERR(bb->vma)) {
			ret = PTR_ERR(bb->vma);
			goto err;
		}

		/* relocate shadow batch buffer */
		bb->bb_start_cmd_va[1] = i915_ggtt_offset(bb->vma);
		if (gmadr_bytes == 8)
			bb->bb_start_cmd_va[2] = 0;

		/* No one is going to touch shadow bb from now on. */
		if (bb->clflush & CLFLUSH_AFTER) {
			drm_clflush_virt_range(bb->va, bb->obj->base.size);
			bb->clflush &= ~CLFLUSH_AFTER;
		}

		ret = i915_gem_object_set_to_gtt_domain(bb->obj, false);
		if (ret)
			goto err;

		i915_gem_obj_finish_shmem_access(bb->obj);
		bb->accessing = false;

		i915_vma_move_to_active(bb->vma, workload->req, 0);
	}
	return 0;
err:
	release_shadow_batch_buffer(workload);
	return ret;
}

static int update_wa_ctx_2_shadow_ctx(struct intel_shadow_wa_ctx *wa_ctx)
{
	struct intel_vgpu_workload *workload = container_of(wa_ctx,
					struct intel_vgpu_workload,
					wa_ctx);
	int ring_id = workload->ring_id;
	struct intel_vgpu_submission *s = &workload->vgpu->submission;
	struct i915_gem_context *shadow_ctx = s->shadow_ctx;
	struct drm_i915_gem_object *ctx_obj =
		shadow_ctx->engine[ring_id].state->obj;
	struct execlist_ring_context *shadow_ring_context;
	struct page *page;

	page = i915_gem_object_get_page(ctx_obj, LRC_STATE_PN);
	shadow_ring_context = kmap_atomic(page);

	shadow_ring_context->bb_per_ctx_ptr.val =
		(shadow_ring_context->bb_per_ctx_ptr.val &
		(~PER_CTX_ADDR_MASK)) | wa_ctx->per_ctx.shadow_gma;
	shadow_ring_context->rcs_indirect_ctx.val =
		(shadow_ring_context->rcs_indirect_ctx.val &
		(~INDIRECT_CTX_ADDR_MASK)) | wa_ctx->indirect_ctx.shadow_gma;

	kunmap_atomic(shadow_ring_context);
	return 0;
}

static int prepare_shadow_wa_ctx(struct intel_shadow_wa_ctx *wa_ctx)
{
	struct i915_vma *vma;
	unsigned char *per_ctx_va =
		(unsigned char *)wa_ctx->indirect_ctx.shadow_va +
		wa_ctx->indirect_ctx.size;

	if (wa_ctx->indirect_ctx.size == 0)
		return 0;

	vma = i915_gem_object_ggtt_pin(wa_ctx->indirect_ctx.obj, NULL,
				       0, CACHELINE_BYTES, 0);
	if (IS_ERR(vma))
		return PTR_ERR(vma);

	/* FIXME: we are not tracking our pinned VMA leaving it
	 * up to the core to fix up the stray pin_count upon
	 * free.
	 */

	wa_ctx->indirect_ctx.shadow_gma = i915_ggtt_offset(vma);

	wa_ctx->per_ctx.shadow_gma = *((unsigned int *)per_ctx_va + 1);
	memset(per_ctx_va, 0, CACHELINE_BYTES);

	update_wa_ctx_2_shadow_ctx(wa_ctx);
	return 0;
}

static void release_shadow_batch_buffer(struct intel_vgpu_workload *workload)
{
	struct intel_vgpu *vgpu = workload->vgpu;
	struct drm_i915_private *dev_priv = vgpu->gvt->dev_priv;
	struct intel_vgpu_shadow_bb *bb, *pos;

	if (list_empty(&workload->shadow_bb))
		return;

	bb = list_first_entry(&workload->shadow_bb,
			struct intel_vgpu_shadow_bb, list);

	mutex_lock(&dev_priv->drm.struct_mutex);

	list_for_each_entry_safe(bb, pos, &workload->shadow_bb, list) {
		if (bb->obj) {
			if (bb->accessing)
				i915_gem_obj_finish_shmem_access(bb->obj);

			if (bb->va && !IS_ERR(bb->va))
				i915_gem_object_unpin_map(bb->obj);

			if (bb->vma && !IS_ERR(bb->vma)) {
				i915_vma_unpin(bb->vma);
				i915_vma_close(bb->vma);
			}
			__i915_gem_object_release_unless_active(bb->obj);
		}
		list_del(&bb->list);
		kfree(bb);
	}

	mutex_unlock(&dev_priv->drm.struct_mutex);
}

static int prepare_workload(struct intel_vgpu_workload *workload)
{
	struct intel_vgpu *vgpu = workload->vgpu;
	int ret = 0;

	ret = intel_vgpu_pin_mm(workload->shadow_mm);
	if (ret) {
		gvt_vgpu_err("fail to vgpu pin mm\n");
		return ret;
	}

	ret = intel_vgpu_sync_oos_pages(workload->vgpu);
	if (ret) {
		gvt_vgpu_err("fail to vgpu sync oos pages\n");
		goto err_unpin_mm;
	}

	ret = intel_vgpu_flush_post_shadow(workload->vgpu);
	if (ret) {
		gvt_vgpu_err("fail to flush post shadow\n");
		goto err_unpin_mm;
	}

	ret = prepare_shadow_batch_buffer(workload);
	if (ret) {
		gvt_vgpu_err("fail to prepare_shadow_batch_buffer\n");
		goto err_unpin_mm;
	}

	ret = prepare_shadow_wa_ctx(&workload->wa_ctx);
	if (ret) {
		gvt_vgpu_err("fail to prepare_shadow_wa_ctx\n");
		goto err_shadow_batch;
	}

	if (workload->prepare) {
		ret = workload->prepare(workload);
		if (ret)
			goto err_shadow_wa_ctx;
	}

	return 0;
err_shadow_wa_ctx:
	release_shadow_wa_ctx(&workload->wa_ctx);
err_shadow_batch:
	release_shadow_batch_buffer(workload);
err_unpin_mm:
	intel_vgpu_unpin_mm(workload->shadow_mm);
	return ret;
}

static int dispatch_workload(struct intel_vgpu_workload *workload)
{
	struct intel_vgpu *vgpu = workload->vgpu;
	struct intel_vgpu_submission *s = &vgpu->submission;
	struct i915_gem_context *shadow_ctx = s->shadow_ctx;
	struct drm_i915_private *dev_priv = vgpu->gvt->dev_priv;
	int ring_id = workload->ring_id;
	struct intel_engine_cs *engine = dev_priv->engine[ring_id];
	int ret = 0;

	gvt_dbg_sched("ring id %d prepare to dispatch workload %p\n",
		ring_id, workload);

	mutex_lock(&dev_priv->drm.struct_mutex);

	ret = intel_gvt_scan_and_shadow_workload(workload);
	if (ret)
		goto out;

	ret = prepare_workload(workload);
	if (ret) {
		engine->context_unpin(engine, shadow_ctx);
		goto out;
	}

out:
	if (ret)
		workload->status = ret;

	if (!IS_ERR_OR_NULL(workload->req)) {
		gvt_dbg_sched("ring id %d submit workload to i915 %p\n",
				ring_id, workload->req);
		i915_add_request(workload->req);
		workload->dispatched = true;
	}

	mutex_unlock(&dev_priv->drm.struct_mutex);
	return ret;
}

static struct intel_vgpu_workload *pick_next_workload(
		struct intel_gvt *gvt, int ring_id)
{
	struct intel_gvt_workload_scheduler *scheduler = &gvt->scheduler;
	struct intel_vgpu_workload *workload = NULL;

	mutex_lock(&gvt->lock);

	/*
	 * no current vgpu / will be scheduled out / no workload
	 * bail out
	 */
	if (!scheduler->current_vgpu) {
		gvt_dbg_sched("ring id %d stop - no current vgpu\n", ring_id);
		goto out;
	}

	if (scheduler->need_reschedule) {
		gvt_dbg_sched("ring id %d stop - will reschedule\n", ring_id);
		goto out;
	}

	if (list_empty(workload_q_head(scheduler->current_vgpu, ring_id)))
		goto out;

	/*
	 * still have current workload, maybe the workload disptacher
	 * fail to submit it for some reason, resubmit it.
	 */
	if (scheduler->current_workload[ring_id]) {
		workload = scheduler->current_workload[ring_id];
		gvt_dbg_sched("ring id %d still have current workload %p\n",
				ring_id, workload);
		goto out;
	}

	/*
	 * pick a workload as current workload
	 * once current workload is set, schedule policy routines
	 * will wait the current workload is finished when trying to
	 * schedule out a vgpu.
	 */
	scheduler->current_workload[ring_id] = container_of(
			workload_q_head(scheduler->current_vgpu, ring_id)->next,
			struct intel_vgpu_workload, list);

	workload = scheduler->current_workload[ring_id];

	gvt_dbg_sched("ring id %d pick new workload %p\n", ring_id, workload);

	atomic_inc(&workload->vgpu->submission.running_workload_num);
out:
	mutex_unlock(&gvt->lock);
	return workload;
}

static void update_guest_context(struct intel_vgpu_workload *workload)
{
	struct intel_vgpu *vgpu = workload->vgpu;
	struct intel_gvt *gvt = vgpu->gvt;
	struct intel_vgpu_submission *s = &vgpu->submission;
	struct i915_gem_context *shadow_ctx = s->shadow_ctx;
	int ring_id = workload->ring_id;
	struct drm_i915_gem_object *ctx_obj =
		shadow_ctx->engine[ring_id].state->obj;
	struct execlist_ring_context *shadow_ring_context;
	struct page *page;
	void *src;
	unsigned long context_gpa, context_page_num;
	int i;

	gvt_dbg_sched("ring id %d workload lrca %x\n", ring_id,
			workload->ctx_desc.lrca);

	context_page_num = gvt->dev_priv->engine[ring_id]->context_size;

	context_page_num = context_page_num >> PAGE_SHIFT;

	if (IS_BROADWELL(gvt->dev_priv) && ring_id == RCS)
		context_page_num = 19;

	i = 2;

	while (i < context_page_num) {
		context_gpa = intel_vgpu_gma_to_gpa(vgpu->gtt.ggtt_mm,
				(u32)((workload->ctx_desc.lrca + i) <<
					GTT_PAGE_SHIFT));
		if (context_gpa == INTEL_GVT_INVALID_ADDR) {
			gvt_vgpu_err("invalid guest context descriptor\n");
			return;
		}

		page = i915_gem_object_get_page(ctx_obj, LRC_HEADER_PAGES + i);
		src = kmap(page);
		intel_gvt_hypervisor_write_gpa(vgpu, context_gpa, src,
				GTT_PAGE_SIZE);
		kunmap(page);
		i++;
	}

	intel_gvt_hypervisor_write_gpa(vgpu, workload->ring_context_gpa +
		RING_CTX_OFF(ring_header.val), &workload->rb_tail, 4);

	page = i915_gem_object_get_page(ctx_obj, LRC_STATE_PN);
	shadow_ring_context = kmap(page);

#define COPY_REG(name) \
	intel_gvt_hypervisor_write_gpa(vgpu, workload->ring_context_gpa + \
		RING_CTX_OFF(name.val), &shadow_ring_context->name.val, 4)

	COPY_REG(ctx_ctrl);
	COPY_REG(ctx_timestamp);

#undef COPY_REG

	intel_gvt_hypervisor_write_gpa(vgpu,
			workload->ring_context_gpa +
			sizeof(*shadow_ring_context),
			(void *)shadow_ring_context +
			sizeof(*shadow_ring_context),
			GTT_PAGE_SIZE - sizeof(*shadow_ring_context));

	kunmap(page);
}

static void clean_workloads(struct intel_vgpu *vgpu, unsigned long engine_mask)
{
	struct intel_vgpu_submission *s = &vgpu->submission;
	struct drm_i915_private *dev_priv = vgpu->gvt->dev_priv;
	struct intel_engine_cs *engine;
	struct intel_vgpu_workload *pos, *n;
	unsigned int tmp;

	/* free the unsubmited workloads in the queues. */
	for_each_engine_masked(engine, dev_priv, engine_mask, tmp) {
		list_for_each_entry_safe(pos, n,
			&s->workload_q_head[engine->id], list) {
			list_del_init(&pos->list);
			intel_vgpu_destroy_workload(pos);
		}
		clear_bit(engine->id, s->shadow_ctx_desc_updated);
	}
}

static void complete_current_workload(struct intel_gvt *gvt, int ring_id)
{
	struct intel_gvt_workload_scheduler *scheduler = &gvt->scheduler;
	struct intel_vgpu_workload *workload =
		scheduler->current_workload[ring_id];
	struct intel_vgpu *vgpu = workload->vgpu;
	struct intel_vgpu_submission *s = &vgpu->submission;
	int event;

	mutex_lock(&gvt->lock);

	/* For the workload w/ request, needs to wait for the context
	 * switch to make sure request is completed.
	 * For the workload w/o request, directly complete the workload.
	 */
	if (workload->req) {
		struct drm_i915_private *dev_priv =
			workload->vgpu->gvt->dev_priv;
		struct intel_engine_cs *engine =
			dev_priv->engine[workload->ring_id];
		wait_event(workload->shadow_ctx_status_wq,
			   !atomic_read(&workload->shadow_ctx_active));

		/* If this request caused GPU hang, req->fence.error will
		 * be set to -EIO. Use -EIO to set workload status so
		 * that when this request caused GPU hang, didn't trigger
		 * context switch interrupt to guest.
		 */
		if (likely(workload->status == -EINPROGRESS)) {
			if (workload->req->fence.error == -EIO)
				workload->status = -EIO;
			else
				workload->status = 0;
		}

		i915_gem_request_put(fetch_and_zero(&workload->req));

		if (!workload->status && !(vgpu->resetting_eng &
					   ENGINE_MASK(ring_id))) {
			update_guest_context(workload);

			for_each_set_bit(event, workload->pending_events,
					 INTEL_GVT_EVENT_MAX)
				intel_vgpu_trigger_virtual_event(vgpu, event);
		}
		mutex_lock(&dev_priv->drm.struct_mutex);
		/* unpin shadow ctx as the shadow_ctx update is done */
		engine->context_unpin(engine, s->shadow_ctx);
		mutex_unlock(&dev_priv->drm.struct_mutex);
	}

	gvt_dbg_sched("ring id %d complete workload %p status %d\n",
			ring_id, workload, workload->status);

	scheduler->current_workload[ring_id] = NULL;

	list_del_init(&workload->list);

	if (!workload->status) {
		release_shadow_batch_buffer(workload);
		release_shadow_wa_ctx(&workload->wa_ctx);
	}

	if (workload->status || (vgpu->resetting_eng & ENGINE_MASK(ring_id))) {
		/* if workload->status is not successful means HW GPU
		 * has occurred GPU hang or something wrong with i915/GVT,
		 * and GVT won't inject context switch interrupt to guest.
		 * So this error is a vGPU hang actually to the guest.
		 * According to this we should emunlate a vGPU hang. If
		 * there are pending workloads which are already submitted
		 * from guest, we should clean them up like HW GPU does.
		 *
		 * if it is in middle of engine resetting, the pending
		 * workloads won't be submitted to HW GPU and will be
		 * cleaned up during the resetting process later, so doing
		 * the workload clean up here doesn't have any impact.
		 **/
		clean_workloads(vgpu, ENGINE_MASK(ring_id));
	}

	workload->complete(workload);

	atomic_dec(&s->running_workload_num);
	wake_up(&scheduler->workload_complete_wq);

	if (gvt->scheduler.need_reschedule)
		intel_gvt_request_service(gvt, INTEL_GVT_REQUEST_EVENT_SCHED);

	mutex_unlock(&gvt->lock);
}

struct workload_thread_param {
	struct intel_gvt *gvt;
	int ring_id;
};

static int workload_thread(void *priv)
{
	struct workload_thread_param *p = (struct workload_thread_param *)priv;
	struct intel_gvt *gvt = p->gvt;
	int ring_id = p->ring_id;
	struct intel_gvt_workload_scheduler *scheduler = &gvt->scheduler;
	struct intel_vgpu_workload *workload = NULL;
	struct intel_vgpu *vgpu = NULL;
	int ret;
	bool need_force_wake = IS_SKYLAKE(gvt->dev_priv)
			|| IS_KABYLAKE(gvt->dev_priv);
	DEFINE_WAIT_FUNC(wait, woken_wake_function);

	kfree(p);

	gvt_dbg_core("workload thread for ring %d started\n", ring_id);

	while (!kthread_should_stop()) {
		add_wait_queue(&scheduler->waitq[ring_id], &wait);
		do {
			workload = pick_next_workload(gvt, ring_id);
			if (workload)
				break;
			wait_woken(&wait, TASK_INTERRUPTIBLE,
				   MAX_SCHEDULE_TIMEOUT);
		} while (!kthread_should_stop());
		remove_wait_queue(&scheduler->waitq[ring_id], &wait);

		if (!workload)
			break;

		gvt_dbg_sched("ring id %d next workload %p vgpu %d\n",
				workload->ring_id, workload,
				workload->vgpu->id);

		intel_runtime_pm_get(gvt->dev_priv);

		gvt_dbg_sched("ring id %d will dispatch workload %p\n",
				workload->ring_id, workload);

		if (need_force_wake)
			intel_uncore_forcewake_get(gvt->dev_priv,
					FORCEWAKE_ALL);

		mutex_lock(&gvt->lock);
		ret = dispatch_workload(workload);
		mutex_unlock(&gvt->lock);

		if (ret) {
			vgpu = workload->vgpu;
			gvt_vgpu_err("fail to dispatch workload, skip\n");
			goto complete;
		}

		gvt_dbg_sched("ring id %d wait workload %p\n",
				workload->ring_id, workload);
		i915_wait_request(workload->req, 0, MAX_SCHEDULE_TIMEOUT);

complete:
		gvt_dbg_sched("will complete workload %p, status: %d\n",
				workload, workload->status);

		complete_current_workload(gvt, ring_id);

		if (need_force_wake)
			intel_uncore_forcewake_put(gvt->dev_priv,
					FORCEWAKE_ALL);

		intel_runtime_pm_put(gvt->dev_priv);
		if (ret && (vgpu_is_vm_unhealthy(ret)))
			enter_failsafe_mode(vgpu, GVT_FAILSAFE_GUEST_ERR);
	}
	return 0;
}

void intel_gvt_wait_vgpu_idle(struct intel_vgpu *vgpu)
{
	struct intel_vgpu_submission *s = &vgpu->submission;
	struct intel_gvt *gvt = vgpu->gvt;
	struct intel_gvt_workload_scheduler *scheduler = &gvt->scheduler;

	if (atomic_read(&s->running_workload_num)) {
		gvt_dbg_sched("wait vgpu idle\n");

		wait_event(scheduler->workload_complete_wq,
				!atomic_read(&s->running_workload_num));
	}
}

void intel_gvt_clean_workload_scheduler(struct intel_gvt *gvt)
{
	struct intel_gvt_workload_scheduler *scheduler = &gvt->scheduler;
	struct intel_engine_cs *engine;
	enum intel_engine_id i;

	gvt_dbg_core("clean workload scheduler\n");

	for_each_engine(engine, gvt->dev_priv, i) {
		atomic_notifier_chain_unregister(
					&engine->context_status_notifier,
					&gvt->shadow_ctx_notifier_block[i]);
		kthread_stop(scheduler->thread[i]);
	}
}

int intel_gvt_init_workload_scheduler(struct intel_gvt *gvt)
{
	struct intel_gvt_workload_scheduler *scheduler = &gvt->scheduler;
	struct workload_thread_param *param = NULL;
	struct intel_engine_cs *engine;
	enum intel_engine_id i;
	int ret;

	gvt_dbg_core("init workload scheduler\n");

	init_waitqueue_head(&scheduler->workload_complete_wq);

	for_each_engine(engine, gvt->dev_priv, i) {
		init_waitqueue_head(&scheduler->waitq[i]);

		param = kzalloc(sizeof(*param), GFP_KERNEL);
		if (!param) {
			ret = -ENOMEM;
			goto err;
		}

		param->gvt = gvt;
		param->ring_id = i;

		scheduler->thread[i] = kthread_run(workload_thread, param,
			"gvt workload %d", i);
		if (IS_ERR(scheduler->thread[i])) {
			gvt_err("fail to create workload thread\n");
			ret = PTR_ERR(scheduler->thread[i]);
			goto err;
		}

		gvt->shadow_ctx_notifier_block[i].notifier_call =
					shadow_context_status_change;
		atomic_notifier_chain_register(&engine->context_status_notifier,
					&gvt->shadow_ctx_notifier_block[i]);
	}
	return 0;
err:
	intel_gvt_clean_workload_scheduler(gvt);
	kfree(param);
	param = NULL;
	return ret;
}

/**
 * intel_vgpu_clean_submission - free submission-related resource for vGPU
 * @vgpu: a vGPU
 *
 * This function is called when a vGPU is being destroyed.
 *
 */
void intel_vgpu_clean_submission(struct intel_vgpu *vgpu)
{
	struct intel_vgpu_submission *s = &vgpu->submission;

	intel_vgpu_select_submission_ops(vgpu, 0);
	i915_gem_context_put(s->shadow_ctx);
	kmem_cache_destroy(s->workloads);
}


/**
 * intel_vgpu_reset_submission - reset submission-related resource for vGPU
 * @vgpu: a vGPU
 * @engine_mask: engines expected to be reset
 *
 * This function is called when a vGPU is being destroyed.
 *
 */
void intel_vgpu_reset_submission(struct intel_vgpu *vgpu,
		unsigned long engine_mask)
{
	struct intel_vgpu_submission *s = &vgpu->submission;

	if (!s->active)
		return;

	clean_workloads(vgpu, engine_mask);
	s->ops->reset(vgpu, engine_mask);
}

/**
 * intel_vgpu_setup_submission - setup submission-related resource for vGPU
 * @vgpu: a vGPU
 *
 * This function is called when a vGPU is being created.
 *
 * Returns:
 * Zero on success, negative error code if failed.
 *
 */
int intel_vgpu_setup_submission(struct intel_vgpu *vgpu)
{
	struct intel_vgpu_submission *s = &vgpu->submission;
	enum intel_engine_id i;
	struct intel_engine_cs *engine;
	int ret;

	s->shadow_ctx = i915_gem_context_create_gvt(
			&vgpu->gvt->dev_priv->drm);
	if (IS_ERR(s->shadow_ctx))
		return PTR_ERR(s->shadow_ctx);

	bitmap_zero(s->shadow_ctx_desc_updated, I915_NUM_ENGINES);

	s->workloads = kmem_cache_create("gvt-g_vgpu_workload",
			sizeof(struct intel_vgpu_workload), 0,
			SLAB_HWCACHE_ALIGN,
			NULL);

	if (!s->workloads) {
		ret = -ENOMEM;
		goto out_shadow_ctx;
	}

	for_each_engine(engine, vgpu->gvt->dev_priv, i)
		INIT_LIST_HEAD(&s->workload_q_head[i]);

	atomic_set(&s->running_workload_num, 0);
	bitmap_zero(s->tlb_handle_pending, I915_NUM_ENGINES);

	return 0;

out_shadow_ctx:
	i915_gem_context_put(s->shadow_ctx);
	return ret;
}

/**
 * intel_vgpu_select_submission_ops - select virtual submission interface
 * @vgpu: a vGPU
 * @interface: expected vGPU virtual submission interface
 *
 * This function is called when guest configures submission interface.
 *
 * Returns:
 * Zero on success, negative error code if failed.
 *
 */
int intel_vgpu_select_submission_ops(struct intel_vgpu *vgpu,
				     unsigned int interface)
{
	struct intel_vgpu_submission *s = &vgpu->submission;
	const struct intel_vgpu_submission_ops *ops[] = {
		[INTEL_VGPU_EXECLIST_SUBMISSION] =
			&intel_vgpu_execlist_submission_ops,
	};
	int ret;

	if (WARN_ON(interface >= ARRAY_SIZE(ops)))
		return -EINVAL;

	if (s->active) {
		s->ops->clean(vgpu);
		s->active = false;
		gvt_dbg_core("vgpu%d: de-select ops [ %s ] \n",
				vgpu->id, s->ops->name);
	}

	if (interface == 0) {
		s->ops = NULL;
		s->virtual_submission_interface = 0;
		gvt_dbg_core("vgpu%d: no submission ops\n", vgpu->id);
		return 0;
	}

	ret = ops[interface]->init(vgpu);
	if (ret)
		return ret;

	s->ops = ops[interface];
	s->virtual_submission_interface = interface;
	s->active = true;

	gvt_dbg_core("vgpu%d: activate ops [ %s ]\n",
			vgpu->id, s->ops->name);

	return 0;
}

/**
 * intel_vgpu_destroy_workload - destroy a vGPU workload
 * @vgpu: a vGPU
 *
 * This function is called when destroy a vGPU workload.
 *
 */
void intel_vgpu_destroy_workload(struct intel_vgpu_workload *workload)
{
	struct intel_vgpu_submission *s = &workload->vgpu->submission;

	if (workload->shadow_mm)
		intel_gvt_mm_unreference(workload->shadow_mm);

	kmem_cache_free(s->workloads, workload);
}

static struct intel_vgpu_workload *
alloc_workload(struct intel_vgpu *vgpu)
{
	struct intel_vgpu_submission *s = &vgpu->submission;
	struct intel_vgpu_workload *workload;

	workload = kmem_cache_zalloc(s->workloads, GFP_KERNEL);
	if (!workload)
		return ERR_PTR(-ENOMEM);

	INIT_LIST_HEAD(&workload->list);
	INIT_LIST_HEAD(&workload->shadow_bb);

	init_waitqueue_head(&workload->shadow_ctx_status_wq);
	atomic_set(&workload->shadow_ctx_active, 0);

	workload->status = -EINPROGRESS;
	workload->shadowed = false;
	workload->vgpu = vgpu;

	return workload;
}

#define RING_CTX_OFF(x) \
	offsetof(struct execlist_ring_context, x)

static void read_guest_pdps(struct intel_vgpu *vgpu,
		u64 ring_context_gpa, u32 pdp[8])
{
	u64 gpa;
	int i;

	gpa = ring_context_gpa + RING_CTX_OFF(pdp3_UDW.val);

	for (i = 0; i < 8; i++)
		intel_gvt_hypervisor_read_gpa(vgpu,
				gpa + i * 8, &pdp[7 - i], 4);
}

static int prepare_mm(struct intel_vgpu_workload *workload)
{
	struct execlist_ctx_descriptor_format *desc = &workload->ctx_desc;
	struct intel_vgpu_mm *mm;
	struct intel_vgpu *vgpu = workload->vgpu;
	int page_table_level;
	u32 pdp[8];

	if (desc->addressing_mode == 1) { /* legacy 32-bit */
		page_table_level = 3;
	} else if (desc->addressing_mode == 3) { /* legacy 64 bit */
		page_table_level = 4;
	} else {
		gvt_vgpu_err("Advanced Context mode(SVM) is not supported!\n");
		return -EINVAL;
	}

	read_guest_pdps(workload->vgpu, workload->ring_context_gpa, pdp);

	mm = intel_vgpu_find_ppgtt_mm(workload->vgpu, page_table_level, pdp);
	if (mm) {
		intel_gvt_mm_reference(mm);
	} else {

		mm = intel_vgpu_create_mm(workload->vgpu, INTEL_GVT_MM_PPGTT,
				pdp, page_table_level, 0);
		if (IS_ERR(mm)) {
			gvt_vgpu_err("fail to create mm object.\n");
			return PTR_ERR(mm);
		}
	}
	workload->shadow_mm = mm;
	return 0;
}

#define same_context(a, b) (((a)->context_id == (b)->context_id) && \
		((a)->lrca == (b)->lrca))

#define get_last_workload(q) \
	(list_empty(q) ? NULL : container_of(q->prev, \
	struct intel_vgpu_workload, list))
/**
 * intel_vgpu_create_workload - create a vGPU workload
 * @vgpu: a vGPU
 * @desc: a guest context descriptor
 *
 * This function is called when creating a vGPU workload.
 *
 * Returns:
 * struct intel_vgpu_workload * on success, negative error code in
 * pointer if failed.
 *
 */
struct intel_vgpu_workload *
intel_vgpu_create_workload(struct intel_vgpu *vgpu, int ring_id,
			   struct execlist_ctx_descriptor_format *desc)
{
	struct intel_vgpu_submission *s = &vgpu->submission;
	struct list_head *q = workload_q_head(vgpu, ring_id);
	struct intel_vgpu_workload *last_workload = get_last_workload(q);
	struct intel_vgpu_workload *workload = NULL;
	struct drm_i915_private *dev_priv = vgpu->gvt->dev_priv;
	u64 ring_context_gpa;
	u32 head, tail, start, ctl, ctx_ctl, per_ctx, indirect_ctx;
	int ret;

	ring_context_gpa = intel_vgpu_gma_to_gpa(vgpu->gtt.ggtt_mm,
			(u32)((desc->lrca + 1) << GTT_PAGE_SHIFT));
	if (ring_context_gpa == INTEL_GVT_INVALID_ADDR) {
		gvt_vgpu_err("invalid guest context LRCA: %x\n", desc->lrca);
		return ERR_PTR(-EINVAL);
	}

	intel_gvt_hypervisor_read_gpa(vgpu, ring_context_gpa +
			RING_CTX_OFF(ring_header.val), &head, 4);

	intel_gvt_hypervisor_read_gpa(vgpu, ring_context_gpa +
			RING_CTX_OFF(ring_tail.val), &tail, 4);

	head &= RB_HEAD_OFF_MASK;
	tail &= RB_TAIL_OFF_MASK;

	if (last_workload && same_context(&last_workload->ctx_desc, desc)) {
		gvt_dbg_el("ring id %d cur workload == last\n", ring_id);
		gvt_dbg_el("ctx head %x real head %lx\n", head,
				last_workload->rb_tail);
		/*
		 * cannot use guest context head pointer here,
		 * as it might not be updated at this time
		 */
		head = last_workload->rb_tail;
	}

	gvt_dbg_el("ring id %d begin a new workload\n", ring_id);

	/* record some ring buffer register values for scan and shadow */
	intel_gvt_hypervisor_read_gpa(vgpu, ring_context_gpa +
			RING_CTX_OFF(rb_start.val), &start, 4);
	intel_gvt_hypervisor_read_gpa(vgpu, ring_context_gpa +
			RING_CTX_OFF(rb_ctrl.val), &ctl, 4);
	intel_gvt_hypervisor_read_gpa(vgpu, ring_context_gpa +
			RING_CTX_OFF(ctx_ctrl.val), &ctx_ctl, 4);

	workload = alloc_workload(vgpu);
	if (IS_ERR(workload))
		return workload;

	workload->ring_id = ring_id;
	workload->ctx_desc = *desc;
	workload->ring_context_gpa = ring_context_gpa;
	workload->rb_head = head;
	workload->rb_tail = tail;
	workload->rb_start = start;
	workload->rb_ctl = ctl;

	if (ring_id == RCS) {
		intel_gvt_hypervisor_read_gpa(vgpu, ring_context_gpa +
			RING_CTX_OFF(bb_per_ctx_ptr.val), &per_ctx, 4);
		intel_gvt_hypervisor_read_gpa(vgpu, ring_context_gpa +
			RING_CTX_OFF(rcs_indirect_ctx.val), &indirect_ctx, 4);

		workload->wa_ctx.indirect_ctx.guest_gma =
			indirect_ctx & INDIRECT_CTX_ADDR_MASK;
		workload->wa_ctx.indirect_ctx.size =
			(indirect_ctx & INDIRECT_CTX_SIZE_MASK) *
			CACHELINE_BYTES;
		workload->wa_ctx.per_ctx.guest_gma =
			per_ctx & PER_CTX_ADDR_MASK;
		workload->wa_ctx.per_ctx.valid = per_ctx & 1;
	}

	gvt_dbg_el("workload %p ring id %d head %x tail %x start %x ctl %x\n",
			workload, ring_id, head, tail, start, ctl);

	ret = prepare_mm(workload);
	if (ret) {
		kmem_cache_free(s->workloads, workload);
		return ERR_PTR(ret);
	}

	/* Only scan and shadow the first workload in the queue
	 * as there is only one pre-allocated buf-obj for shadow.
	 */
	if (list_empty(workload_q_head(vgpu, ring_id))) {
		intel_runtime_pm_get(dev_priv);
		mutex_lock(&dev_priv->drm.struct_mutex);
		ret = intel_gvt_scan_and_shadow_workload(workload);
		mutex_unlock(&dev_priv->drm.struct_mutex);
		intel_runtime_pm_put(dev_priv);
	}

	if (ret && (vgpu_is_vm_unhealthy(ret))) {
		enter_failsafe_mode(vgpu, GVT_FAILSAFE_GUEST_ERR);
		intel_vgpu_destroy_workload(workload);
		return ERR_PTR(ret);
	}

	return workload;
}