scheduler.c 20.1 KB
Newer Older
Z
Zhi Wang 已提交
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37
/*
 * 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>

38 39 40
#include "i915_drv.h"
#include "gvt.h"

Z
Zhi Wang 已提交
41 42 43
#define RING_CTX_OFF(x) \
	offsetof(struct execlist_ring_context, x)

44 45
static void set_context_pdp_root_pointer(
		struct execlist_ring_context *ring_context,
Z
Zhi Wang 已提交
46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71
		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 = workload->vgpu->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);

72
	context_page_num = gvt->dev_priv->engine[ring_id]->context_size;
Z
Zhi Wang 已提交
73 74 75 76 77 78 79 80 81 82 83 84 85

	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) {
86
			gvt_vgpu_err("Invalid guest context descriptor\n");
Z
Zhi Wang 已提交
87 88 89
			return -EINVAL;
		}

90
		page = i915_gem_object_get_page(ctx_obj, LRC_HEADER_PAGES + i);
91
		dst = kmap(page);
Z
Zhi Wang 已提交
92 93
		intel_gvt_hypervisor_read_gpa(vgpu, context_gpa, dst,
				GTT_PAGE_SIZE);
94
		kunmap(page);
Z
Zhi Wang 已提交
95 96 97 98
		i++;
	}

	page = i915_gem_object_get_page(ctx_obj, LRC_STATE_PN);
99
	shadow_ring_context = kmap(page);
Z
Zhi Wang 已提交
100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124

#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));

125
	kunmap(page);
Z
Zhi Wang 已提交
126 127 128
	return 0;
}

129 130 131 132 133
static inline bool is_gvt_request(struct drm_i915_gem_request *req)
{
	return i915_gem_context_force_single_submission(req->ctx);
}

Z
Zhi Wang 已提交
134 135 136
static int shadow_context_status_change(struct notifier_block *nb,
		unsigned long action, void *data)
{
137 138 139 140
	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;
141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156
	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;
	}
Z
Zhi Wang 已提交
157

158 159
	workload = scheduler->current_workload[ring_id];
	if (unlikely(!workload))
160 161
		return NOTIFY_OK;

Z
Zhi Wang 已提交
162 163
	switch (action) {
	case INTEL_CONTEXT_SCHEDULE_IN:
164 165 166 167 168 169 170 171 172 173
		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);
Z
Zhi Wang 已提交
174 175 176
		atomic_set(&workload->shadow_ctx_active, 1);
		break;
	case INTEL_CONTEXT_SCHEDULE_OUT:
177
	case INTEL_CONTEXT_SCHEDULE_PREEMPTED:
Z
Zhi Wang 已提交
178 179 180 181 182 183 184 185 186 187
		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;
}

188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204
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;
}

205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232
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;
}

233 234 235 236 237 238 239 240 241
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);
}

242 243 244 245 246 247 248 249 250
/**
 * 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)
Z
Zhi Wang 已提交
251 252 253 254
{
	int ring_id = workload->ring_id;
	struct i915_gem_context *shadow_ctx = workload->vgpu->shadow_ctx;
	struct drm_i915_private *dev_priv = workload->vgpu->gvt->dev_priv;
255
	struct intel_engine_cs *engine = dev_priv->engine[ring_id];
256
	struct drm_i915_gem_request *rq;
257
	struct intel_vgpu *vgpu = workload->vgpu;
258
	struct intel_ring *ring;
Z
Zhi Wang 已提交
259 260
	int ret;

261 262
	lockdep_assert_held(&dev_priv->drm.struct_mutex);

263 264
	if (workload->shadowed)
		return 0;
Z
Zhi Wang 已提交
265

266 267
	shadow_ctx->desc_template &= ~(0x3 << GEN8_CTX_ADDRESSING_MODE_SHIFT);
	shadow_ctx->desc_template |= workload->ctx_desc.addressing_mode <<
Z
Zhi Wang 已提交
268 269
				    GEN8_CTX_ADDRESSING_MODE_SHIFT;

270 271 272
	if (!test_and_set_bit(ring_id, vgpu->shadow_ctx_desc_updated))
		shadow_context_descriptor_update(shadow_ctx,
					dev_priv->engine[ring_id]);
273

274
	ret = intel_gvt_scan_and_shadow_ringbuffer(workload);
Z
Zhi Wang 已提交
275
	if (ret)
276
		goto err_scan;
Z
Zhi Wang 已提交
277

278 279 280 281
	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)
282
			goto err_scan;
283
	}
Z
Zhi Wang 已提交
284

285 286 287 288 289 290 291 292 293 294 295
	/* 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");
296
		goto err_shadow;
297 298
	}

Z
Zhi Wang 已提交
299 300
	ret = populate_shadow_context(workload);
	if (ret)
301
		goto err_unpin;
Z
Zhi Wang 已提交
302

303 304 305 306
	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);
307
		goto err_unpin;
308 309 310 311 312 313 314
	}

	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)
315
		goto err_unpin;
316
	workload->shadowed = true;
317
	return 0;
318

319 320 321 322 323
err_unpin:
	engine->context_unpin(engine, shadow_ctx);
err_shadow:
	release_shadow_wa_ctx(&workload->wa_ctx);
err_scan:
324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340
	return ret;
}

static int dispatch_workload(struct intel_vgpu_workload *workload)
{
	int ring_id = workload->ring_id;
	struct i915_gem_context *shadow_ctx = workload->vgpu->shadow_ctx;
	struct drm_i915_private *dev_priv = workload->vgpu->gvt->dev_priv;
	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);
Z
Zhi Wang 已提交
341
	if (ret)
342
		goto out;
Z
Zhi Wang 已提交
343 344 345

	if (workload->prepare) {
		ret = workload->prepare(workload);
346 347
		if (ret) {
			engine->context_unpin(engine, shadow_ctx);
348
			goto out;
349
		}
Z
Zhi Wang 已提交
350 351
	}

352 353 354
out:
	if (ret)
		workload->status = ret;
355

356 357 358 359 360 361
	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;
	}
362

363
	mutex_unlock(&dev_priv->drm.struct_mutex);
Z
Zhi Wang 已提交
364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388
	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;
	}

389
	if (list_empty(workload_q_head(scheduler->current_vgpu, ring_id)))
Z
Zhi Wang 已提交
390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439
		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->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;
	int ring_id = workload->ring_id;
	struct i915_gem_context *shadow_ctx = workload->vgpu->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 *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);

440
	context_page_num = gvt->dev_priv->engine[ring_id]->context_size;
Z
Zhi Wang 已提交
441 442 443 444 445 446 447 448 449 450 451 452 453

	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) {
454
			gvt_vgpu_err("invalid guest context descriptor\n");
Z
Zhi Wang 已提交
455 456 457
			return;
		}

458
		page = i915_gem_object_get_page(ctx_obj, LRC_HEADER_PAGES + i);
459
		src = kmap(page);
Z
Zhi Wang 已提交
460 461
		intel_gvt_hypervisor_write_gpa(vgpu, context_gpa, src,
				GTT_PAGE_SIZE);
462
		kunmap(page);
Z
Zhi Wang 已提交
463 464 465 466 467 468 469
		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);
470
	shadow_ring_context = kmap(page);
Z
Zhi Wang 已提交
471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487

#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));

488
	kunmap(page);
Z
Zhi Wang 已提交
489 490 491 492 493 494
}

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;
495
	struct intel_vgpu *vgpu;
Z
Zhi Wang 已提交
496
	int event;
Z
Zhi Wang 已提交
497 498 499 500

	mutex_lock(&gvt->lock);

	workload = scheduler->current_workload[ring_id];
501
	vgpu = workload->vgpu;
Z
Zhi Wang 已提交
502

503 504 505 506 507
	/* 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) {
508 509 510 511
		struct drm_i915_private *dev_priv =
			workload->vgpu->gvt->dev_priv;
		struct intel_engine_cs *engine =
			dev_priv->engine[workload->ring_id];
Z
Zhi Wang 已提交
512 513 514
		wait_event(workload->shadow_ctx_status_wq,
			   !atomic_read(&workload->shadow_ctx_active));

515 516 517 518 519 520 521 522 523 524 525 526
		/* 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;
		}

527
		i915_gem_request_put(fetch_and_zero(&workload->req));
Z
Zhi Wang 已提交
528

529 530
		if (!workload->status && !(vgpu->resetting_eng &
					   ENGINE_MASK(ring_id))) {
531
			update_guest_context(workload);
Z
Zhi Wang 已提交
532

533 534 535 536
			for_each_set_bit(event, workload->pending_events,
					 INTEL_GVT_EVENT_MAX)
				intel_vgpu_trigger_virtual_event(vgpu, event);
		}
537 538 539 540
		mutex_lock(&dev_priv->drm.struct_mutex);
		/* unpin shadow ctx as the shadow_ctx update is done */
		engine->context_unpin(engine, workload->vgpu->shadow_ctx);
		mutex_unlock(&dev_priv->drm.struct_mutex);
Z
Zhi Wang 已提交
541 542 543 544 545 546 547 548 549 550
	}

	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);
	workload->complete(workload);

551
	atomic_dec(&vgpu->running_workload_num);
Z
Zhi Wang 已提交
552
	wake_up(&scheduler->workload_complete_wq);
553 554 555 556

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

Z
Zhi Wang 已提交
557 558 559 560 561 562 563 564 565 566 567 568 569 570 571
	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;
572
	struct intel_vgpu *vgpu = NULL;
Z
Zhi Wang 已提交
573
	int ret;
574 575
	bool need_force_wake = IS_SKYLAKE(gvt->dev_priv)
			|| IS_KABYLAKE(gvt->dev_priv);
576
	DEFINE_WAIT_FUNC(wait, woken_wake_function);
Z
Zhi Wang 已提交
577 578 579 580 581 582

	kfree(p);

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

	while (!kthread_should_stop()) {
583 584 585 586 587 588 589 590 591 592 593
		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)
Z
Zhi Wang 已提交
594 595 596 597 598 599 600 601 602 603 604 605 606 607 608
			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);

609
		mutex_lock(&gvt->lock);
Z
Zhi Wang 已提交
610
		ret = dispatch_workload(workload);
611
		mutex_unlock(&gvt->lock);
612

Z
Zhi Wang 已提交
613
		if (ret) {
614 615
			vgpu = workload->vgpu;
			gvt_vgpu_err("fail to dispatch workload, skip\n");
Z
Zhi Wang 已提交
616 617 618 619 620
			goto complete;
		}

		gvt_dbg_sched("ring id %d wait workload %p\n",
				workload->ring_id, workload);
621
		i915_wait_request(workload->req, 0, MAX_SCHEDULE_TIMEOUT);
Z
Zhi Wang 已提交
622 623

complete:
624
		gvt_dbg_sched("will complete workload %p, status: %d\n",
Z
Zhi Wang 已提交
625 626
				workload, workload->status);

627 628
		complete_current_workload(gvt, ring_id);

Z
Zhi Wang 已提交
629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653
		if (need_force_wake)
			intel_uncore_forcewake_put(gvt->dev_priv,
					FORCEWAKE_ALL);

		intel_runtime_pm_put(gvt->dev_priv);
	}
	return 0;
}

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

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

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

void intel_gvt_clean_workload_scheduler(struct intel_gvt *gvt)
{
	struct intel_gvt_workload_scheduler *scheduler = &gvt->scheduler;
654 655
	struct intel_engine_cs *engine;
	enum intel_engine_id i;
Z
Zhi Wang 已提交
656 657 658

	gvt_dbg_core("clean workload scheduler\n");

659 660 661 662 663
	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]);
Z
Zhi Wang 已提交
664 665 666 667 668 669 670
	}
}

int intel_gvt_init_workload_scheduler(struct intel_gvt *gvt)
{
	struct intel_gvt_workload_scheduler *scheduler = &gvt->scheduler;
	struct workload_thread_param *param = NULL;
671 672
	struct intel_engine_cs *engine;
	enum intel_engine_id i;
Z
Zhi Wang 已提交
673 674 675 676 677 678
	int ret;

	gvt_dbg_core("init workload scheduler\n");

	init_waitqueue_head(&scheduler->workload_complete_wq);

679
	for_each_engine(engine, gvt->dev_priv, i) {
Z
Zhi Wang 已提交
680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697
		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;
		}
698 699 700 701 702

		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]);
Z
Zhi Wang 已提交
703 704 705 706 707 708 709 710 711 712 713
	}
	return 0;
err:
	intel_gvt_clean_workload_scheduler(gvt);
	kfree(param);
	param = NULL;
	return ret;
}

void intel_vgpu_clean_gvt_context(struct intel_vgpu *vgpu)
{
714
	i915_gem_context_put(vgpu->shadow_ctx);
Z
Zhi Wang 已提交
715 716 717 718 719 720 721 722 723 724 725
}

int intel_vgpu_init_gvt_context(struct intel_vgpu *vgpu)
{
	atomic_set(&vgpu->running_workload_num, 0);

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

726 727
	bitmap_zero(vgpu->shadow_ctx_desc_updated, I915_NUM_ENGINES);

Z
Zhi Wang 已提交
728 729
	return 0;
}