i915_gem_context.c 23.2 KB
Newer Older
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 38 39 40 41 42 43 44 45 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 72 73 74 75
/*
 * Copyright © 2011-2012 Intel Corporation
 *
 * 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:
 *    Ben Widawsky <ben@bwidawsk.net>
 *
 */

/*
 * This file implements HW context support. On gen5+ a HW context consists of an
 * opaque GPU object which is referenced at times of context saves and restores.
 * With RC6 enabled, the context is also referenced as the GPU enters and exists
 * from RC6 (GPU has it's own internal power context, except on gen5). Though
 * something like a context does exist for the media ring, the code only
 * supports contexts for the render ring.
 *
 * In software, there is a distinction between contexts created by the user,
 * and the default HW context. The default HW context is used by GPU clients
 * that do not request setup of their own hardware context. The default
 * context's state is never restored to help prevent programming errors. This
 * would happen if a client ran and piggy-backed off another clients GPU state.
 * The default context only exists to give the GPU some offset to load as the
 * current to invoke a save of the context we actually care about. In fact, the
 * code could likely be constructed, albeit in a more complicated fashion, to
 * never use the default context, though that limits the driver's ability to
 * swap out, and/or destroy other contexts.
 *
 * All other contexts are created as a request by the GPU client. These contexts
 * store GPU state, and thus allow GPU clients to not re-emit state (and
 * potentially query certain state) at any time. The kernel driver makes
 * certain that the appropriate commands are inserted.
 *
 * The context life cycle is semi-complicated in that context BOs may live
 * longer than the context itself because of the way the hardware, and object
 * tracking works. Below is a very crude representation of the state machine
 * describing the context life.
 *                                         refcount     pincount     active
 * S0: initial state                          0            0           0
 * S1: context created                        1            0           0
 * S2: context is currently running           2            1           X
 * S3: GPU referenced, but not current        2            0           1
 * S4: context is current, but destroyed      1            1           0
 * S5: like S3, but destroyed                 1            0           1
 *
 * The most common (but not all) transitions:
 * S0->S1: client creates a context
 * S1->S2: client submits execbuf with context
 * S2->S3: other clients submits execbuf with context
 * S3->S1: context object was retired
 * S3->S2: clients submits another execbuf
 * S2->S4: context destroy called with current context
 * S3->S5->S0: destroy path
 * S4->S5->S0: destroy path on current context
 *
 * There are two confusing terms used above:
 *  The "current context" means the context which is currently running on the
D
Damien Lespiau 已提交
76
 *  GPU. The GPU has loaded its state already and has stored away the gtt
77 78 79 80 81 82 83 84 85 86 87
 *  offset of the BO. The GPU is not actively referencing the data at this
 *  offset, but it will on the next context switch. The only way to avoid this
 *  is to do a GPU reset.
 *
 *  An "active context' is one which was previously the "current context" and is
 *  on the active list waiting for the next context switch to occur. Until this
 *  happens, the object must remain at the same gtt offset. It is therefore
 *  possible to destroy a context, but it is still active.
 *
 */

88 89
#include <drm/drmP.h>
#include <drm/i915_drm.h>
90 91
#include "i915_drv.h"

92 93 94 95
/* This is a HW constraint. The value below is the largest known requirement
 * I've seen in a spec to date, and that was a workaround for a non-shipping
 * part. It should be safe to decrease this, but it's more future proof as is.
 */
B
Ben Widawsky 已提交
96 97
#define GEN6_CONTEXT_ALIGN (64<<10)
#define GEN7_CONTEXT_ALIGN 4096
98

99 100
static int do_switch(struct intel_ring_buffer *ring,
		     struct i915_hw_context *to);
101

B
Ben Widawsky 已提交
102 103 104 105 106 107 108 109
static size_t get_context_alignment(struct drm_device *dev)
{
	if (IS_GEN6(dev))
		return GEN6_CONTEXT_ALIGN;

	return GEN7_CONTEXT_ALIGN;
}

110 111 112 113 114 115 116 117 118 119 120 121
static int get_context_size(struct drm_device *dev)
{
	struct drm_i915_private *dev_priv = dev->dev_private;
	int ret;
	u32 reg;

	switch (INTEL_INFO(dev)->gen) {
	case 6:
		reg = I915_READ(CXT_SIZE);
		ret = GEN6_CXT_TOTAL_SIZE(reg) * 64;
		break;
	case 7:
B
Ben Widawsky 已提交
122
		reg = I915_READ(GEN7_CXT_SIZE);
B
Ben Widawsky 已提交
123
		if (IS_HASWELL(dev))
124
			ret = HSW_CXT_TOTAL_SIZE;
B
Ben Widawsky 已提交
125 126
		else
			ret = GEN7_CXT_TOTAL_SIZE(reg) * 64;
127
		break;
B
Ben Widawsky 已提交
128 129 130
	case 8:
		ret = GEN8_CXT_TOTAL_SIZE;
		break;
131 132 133 134 135 136 137
	default:
		BUG();
	}

	return ret;
}

138
void i915_gem_context_free(struct kref *ctx_ref)
139
{
140 141
	struct i915_hw_context *ctx = container_of(ctx_ref,
						   typeof(*ctx), ref);
B
Ben Widawsky 已提交
142
	struct i915_hw_ppgtt *ppgtt = NULL;
143

B
Ben Widawsky 已提交
144 145
	/* We refcount even the aliasing PPGTT to keep the code symmetric */
	if (USES_ALIASING_PPGTT(ctx->obj->base.dev))
146
		ppgtt = ctx_to_ppgtt(ctx);
B
Ben Widawsky 已提交
147 148 149

	/* XXX: Free up the object before tearing down the address space, in
	 * case we're bound in the PPGTT */
150
	drm_gem_object_unreference(&ctx->obj->base);
B
Ben Widawsky 已提交
151 152 153 154

	if (ppgtt)
		kref_put(&ppgtt->ref, ppgtt_release);
	list_del(&ctx->link);
155 156 157
	kfree(ctx);
}

158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176
static struct i915_hw_ppgtt *
create_vm_for_ctx(struct drm_device *dev, struct i915_hw_context *ctx)
{
	struct i915_hw_ppgtt *ppgtt;
	int ret;

	ppgtt = kzalloc(sizeof(*ppgtt), GFP_KERNEL);
	if (!ppgtt)
		return ERR_PTR(-ENOMEM);

	ret = i915_gem_init_ppgtt(dev, ppgtt);
	if (ret) {
		kfree(ppgtt);
		return ERR_PTR(ret);
	}

	return ppgtt;
}

177
static struct i915_hw_context *
178
__create_hw_context(struct drm_device *dev,
179
		  struct drm_i915_file_private *file_priv)
180 181
{
	struct drm_i915_private *dev_priv = dev->dev_private;
182
	struct i915_hw_context *ctx;
T
Tejun Heo 已提交
183
	int ret;
184

185
	ctx = kzalloc(sizeof(*ctx), GFP_KERNEL);
186 187
	if (ctx == NULL)
		return ERR_PTR(-ENOMEM);
188

189
	kref_init(&ctx->ref);
190
	ctx->obj = i915_gem_alloc_object(dev, dev_priv->hw_context_size);
191
	INIT_LIST_HEAD(&ctx->link);
192 193
	if (ctx->obj == NULL) {
		kfree(ctx);
194
		DRM_DEBUG_DRIVER("Context object allocated failed\n");
195
		return ERR_PTR(-ENOMEM);
196 197
	}

198 199
	if (INTEL_INFO(dev)->gen >= 7) {
		ret = i915_gem_object_set_cache_level(ctx->obj,
200
						      I915_CACHE_L3_LLC);
B
Ben Widawsky 已提交
201 202
		/* Failure shouldn't ever happen this early */
		if (WARN_ON(ret))
203 204 205
			goto err_out;
	}

206
	list_add_tail(&ctx->link, &dev_priv->context_list);
207 208 209

	/* Default context will never have a file_priv */
	if (file_priv == NULL)
210
		return ctx;
211

212
	ret = idr_alloc(&file_priv->context_idr, ctx, DEFAULT_CONTEXT_ID, 0,
T
Tejun Heo 已提交
213 214
			GFP_KERNEL);
	if (ret < 0)
215
		goto err_out;
216 217

	ctx->file_priv = file_priv;
T
Tejun Heo 已提交
218
	ctx->id = ret;
219 220 221 222
	/* NB: Mark all slices as needing a remap so that when the context first
	 * loads it will restore whatever remap state already exists. If there
	 * is no remap info, it will be a NOP. */
	ctx->remap_slice = (1 << NUM_L3_SLICES(dev)) - 1;
223

224
	return ctx;
225 226

err_out:
227
	i915_gem_context_unreference(ctx);
228
	return ERR_PTR(ret);
229 230
}

231 232
static inline bool is_default_context(struct i915_hw_context *ctx)
{
233
	return (ctx->id == DEFAULT_CONTEXT_ID);
234 235
}

236 237 238 239 240
/**
 * The default context needs to exist per ring that uses contexts. It stores the
 * context state of the GPU for applications that don't utilize HW contexts, as
 * well as an idle case.
 */
241
static struct i915_hw_context *
242 243 244
i915_gem_create_context(struct drm_device *dev,
			struct drm_i915_file_private *file_priv,
			bool create_vm)
245
{
246
	const bool is_global_default_ctx = file_priv == NULL;
247
	struct drm_i915_private *dev_priv = dev->dev_private;
248
	struct i915_hw_context *ctx;
249
	int ret = 0;
250

B
Ben Widawsky 已提交
251
	BUG_ON(!mutex_is_locked(&dev->struct_mutex));
252

253
	ctx = __create_hw_context(dev, file_priv);
254
	if (IS_ERR(ctx))
255
		return ctx;
256

257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273
	if (is_global_default_ctx) {
		/* We may need to do things with the shrinker which
		 * require us to immediately switch back to the default
		 * context. This can cause a problem as pinning the
		 * default context also requires GTT space which may not
		 * be available. To avoid this we always pin the default
		 * context.
		 */
		ret = i915_gem_obj_ggtt_pin(ctx->obj,
					    get_context_alignment(dev),
					    false, false);
		if (ret) {
			DRM_DEBUG_DRIVER("Couldn't pin %d\n", ret);
			goto err_destroy;
		}
	}

274 275 276 277
	if (create_vm) {
		struct i915_hw_ppgtt *ppgtt = create_vm_for_ctx(dev, ctx);

		if (IS_ERR_OR_NULL(ppgtt)) {
278 279
			DRM_DEBUG_DRIVER("PPGTT setup failed (%ld)\n",
					 PTR_ERR(ppgtt));
280
			ret = PTR_ERR(ppgtt);
281
			goto err_unpin;
282 283 284 285 286
		} else
			ctx->vm = &ppgtt->base;

		/* This case is reserved for the global default context and
		 * should only happen once. */
287
		if (is_global_default_ctx) {
288 289
			if (WARN_ON(dev_priv->mm.aliasing_ppgtt)) {
				ret = -EEXIST;
290
				goto err_unpin;
291 292 293 294 295 296 297 298
			}

			dev_priv->mm.aliasing_ppgtt = ppgtt;
		}
	} else if (USES_ALIASING_PPGTT(dev)) {
		/* For platforms which only have aliasing PPGTT, we fake the
		 * address space and refcounting. */
		ctx->vm = &dev_priv->mm.aliasing_ppgtt->base;
299 300
		kref_get(&dev_priv->mm.aliasing_ppgtt->ref);
	} else
301 302
		ctx->vm = &dev_priv->gtt.base;

303
	return ctx;
304

305 306 307
err_unpin:
	if (is_global_default_ctx)
		i915_gem_object_ggtt_unpin(ctx->obj);
308
err_destroy:
309
	i915_gem_context_unreference(ctx);
310
	return ERR_PTR(ret);
311 312
}

313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355
void i915_gem_context_reset(struct drm_device *dev)
{
	struct drm_i915_private *dev_priv = dev->dev_private;
	struct intel_ring_buffer *ring;
	int i;

	if (!HAS_HW_CONTEXTS(dev))
		return;

	/* Prevent the hardware from restoring the last context (which hung) on
	 * the next switch */
	for (i = 0; i < I915_NUM_RINGS; i++) {
		struct i915_hw_context *dctx;
		if (!(INTEL_INFO(dev)->ring_mask & (1<<i)))
			continue;

		/* Do a fake switch to the default context */
		ring = &dev_priv->ring[i];
		dctx = ring->default_context;
		if (WARN_ON(!dctx))
			continue;

		if (!ring->last_context)
			continue;

		if (ring->last_context == dctx)
			continue;

		if (i == RCS) {
			WARN_ON(i915_gem_obj_ggtt_pin(dctx->obj,
						      get_context_alignment(dev),
						      false, false));
			/* Fake a finish/inactive */
			dctx->obj->base.write_domain = 0;
			dctx->obj->active = 0;
		}

		i915_gem_context_unreference(ring->last_context);
		i915_gem_context_reference(dctx);
		ring->last_context = dctx;
	}
}

356
int i915_gem_context_init(struct drm_device *dev)
357 358
{
	struct drm_i915_private *dev_priv = dev->dev_private;
359
	struct intel_ring_buffer *ring;
360
	int i;
361

362 363
	if (!HAS_HW_CONTEXTS(dev))
		return 0;
364

365 366 367
	/* Init should only be called once per module load. Eventually the
	 * restriction on the context_disabled check can be loosened. */
	if (WARN_ON(dev_priv->ring[RCS].default_context))
368
		return 0;
369

370
	dev_priv->hw_context_size = round_up(get_context_size(dev), 4096);
371

372
	if (dev_priv->hw_context_size > (1<<20)) {
B
Ben Widawsky 已提交
373
		DRM_DEBUG_DRIVER("Disabling HW Contexts; invalid size\n");
374
		return -E2BIG;
375 376
	}

377
	dev_priv->ring[RCS].default_context =
378
		i915_gem_create_context(dev, NULL, USES_ALIASING_PPGTT(dev));
379 380 381 382 383

	if (IS_ERR_OR_NULL(dev_priv->ring[RCS].default_context)) {
		DRM_DEBUG_DRIVER("Disabling HW Contexts; create failed %ld\n",
				 PTR_ERR(dev_priv->ring[RCS].default_context));
		return PTR_ERR(dev_priv->ring[RCS].default_context);
384 385
	}

386 387 388 389 390 391 392 393 394 395
	for (i = RCS + 1; i < I915_NUM_RINGS; i++) {
		if (!(INTEL_INFO(dev)->ring_mask & (1<<i)))
			continue;

		ring = &dev_priv->ring[i];

		/* NB: RCS will hold a ref for all rings */
		ring->default_context = dev_priv->ring[RCS].default_context;
	}

396
	DRM_DEBUG_DRIVER("HW context support initialized\n");
397
	return 0;
398 399 400 401 402
}

void i915_gem_context_fini(struct drm_device *dev)
{
	struct drm_i915_private *dev_priv = dev->dev_private;
403
	struct i915_hw_context *dctx = dev_priv->ring[RCS].default_context;
404
	int i;
405

406
	if (!HAS_HW_CONTEXTS(dev))
407
		return;
408

409 410 411 412 413
	/* The only known way to stop the gpu from accessing the hw context is
	 * to reset it. Do this as the very last operation to avoid confusing
	 * other code, leading to spurious errors. */
	intel_gpu_reset(dev);

414 415 416 417 418 419
	/* When default context is created and switched to, base object refcount
	 * will be 2 (+1 from object creation and +1 from do_switch()).
	 * i915_gem_context_fini() will be called after gpu_idle() has switched
	 * to default context. So we need to unreference the base object once
	 * to offset the do_switch part, so that i915_gem_context_unreference()
	 * can then free the base object correctly. */
B
Ben Widawsky 已提交
420 421 422 423
	WARN_ON(!dev_priv->ring[RCS].last_context);
	if (dev_priv->ring[RCS].last_context == dctx) {
		/* Fake switch to NULL context */
		WARN_ON(dctx->obj->active);
B
Ben Widawsky 已提交
424
		i915_gem_object_ggtt_unpin(dctx->obj);
B
Ben Widawsky 已提交
425
		i915_gem_context_unreference(dctx);
426 427 428 429 430 431 432 433 434 435 436 437
		dev_priv->ring[RCS].last_context = NULL;
	}

	for (i = 0; i < I915_NUM_RINGS; i++) {
		struct intel_ring_buffer *ring = &dev_priv->ring[i];
		if (!(INTEL_INFO(dev)->ring_mask & (1<<i)))
			continue;

		if (ring->last_context)
			i915_gem_context_unreference(ring->last_context);

		ring->default_context = NULL;
438
		ring->last_context = NULL;
B
Ben Widawsky 已提交
439 440
	}

B
Ben Widawsky 已提交
441
	i915_gem_object_ggtt_unpin(dctx->obj);
442
	i915_gem_context_unreference(dctx);
443
	dev_priv->mm.aliasing_ppgtt = NULL;
444 445
}

446 447 448 449 450 451 452 453
int i915_gem_context_enable(struct drm_i915_private *dev_priv)
{
	struct intel_ring_buffer *ring;
	int ret, i;

	if (!HAS_HW_CONTEXTS(dev_priv->dev))
		return 0;

454 455 456 457 458 459 460
	/* This is the only place the aliasing PPGTT gets enabled, which means
	 * it has to happen before we bail on reset */
	if (dev_priv->mm.aliasing_ppgtt) {
		struct i915_hw_ppgtt *ppgtt = dev_priv->mm.aliasing_ppgtt;
		ppgtt->enable(ppgtt);
	}

461 462 463 464 465
	/* FIXME: We should make this work, even in reset */
	if (i915_reset_in_progress(&dev_priv->gpu_error))
		return 0;

	BUG_ON(!dev_priv->ring[RCS].default_context);
466

467 468 469 470 471 472 473 474 475
	for_each_ring(ring, dev_priv, i) {
		ret = do_switch(ring, ring->default_context);
		if (ret)
			return ret;
	}

	return 0;
}

476 477
static int context_idr_cleanup(int id, void *p, void *data)
{
478
	struct i915_hw_context *ctx = p;
479

480 481 482
	/* Ignore the default context because close will handle it */
	if (is_default_context(ctx))
		return 0;
483

484
	i915_gem_context_unreference(ctx);
485
	return 0;
486 487
}

488 489 490
int i915_gem_context_open(struct drm_device *dev, struct drm_file *file)
{
	struct drm_i915_file_private *file_priv = file->driver_priv;
491
	struct drm_i915_private *dev_priv = dev->dev_private;
492

493 494 495 496 497
	if (!HAS_HW_CONTEXTS(dev)) {
		/* Cheat for hang stats */
		file_priv->private_default_ctx =
			kzalloc(sizeof(struct i915_hw_context), GFP_KERNEL);
		file_priv->private_default_ctx->vm = &dev_priv->gtt.base;
498
		return 0;
499
	}
500 501 502

	idr_init(&file_priv->context_idr);

503 504
	mutex_lock(&dev->struct_mutex);
	file_priv->private_default_ctx =
505
		i915_gem_create_context(dev, file_priv, USES_FULL_PPGTT(dev));
506 507 508 509 510 511 512
	mutex_unlock(&dev->struct_mutex);

	if (IS_ERR(file_priv->private_default_ctx)) {
		idr_destroy(&file_priv->context_idr);
		return PTR_ERR(file_priv->private_default_ctx);
	}

513 514 515
	return 0;
}

516 517
void i915_gem_context_close(struct drm_device *dev, struct drm_file *file)
{
518
	struct drm_i915_file_private *file_priv = file->driver_priv;
519

520 521
	if (!HAS_HW_CONTEXTS(dev)) {
		kfree(file_priv->private_default_ctx);
522
		return;
523
	}
524

525
	idr_for_each(&file_priv->context_idr, context_idr_cleanup, NULL);
526
	i915_gem_context_unreference(file_priv->private_default_ctx);
527 528 529
	idr_destroy(&file_priv->context_idr);
}

530
struct i915_hw_context *
531 532
i915_gem_context_get(struct drm_i915_file_private *file_priv, u32 id)
{
533 534
	struct i915_hw_context *ctx;

535 536 537
	if (!HAS_HW_CONTEXTS(file_priv->dev_priv->dev))
		return file_priv->private_default_ctx;

538 539 540 541 542
	ctx = (struct i915_hw_context *)idr_find(&file_priv->context_idr, id);
	if (!ctx)
		return ERR_PTR(-ENOENT);

	return ctx;
543
}
544 545 546 547 548 549 550 551

static inline int
mi_set_context(struct intel_ring_buffer *ring,
	       struct i915_hw_context *new_context,
	       u32 hw_flags)
{
	int ret;

552 553 554 555 556 557
	/* w/a: If Flush TLB Invalidation Mode is enabled, driver must do a TLB
	 * invalidation prior to MI_SET_CONTEXT. On GEN6 we don't set the value
	 * explicitly, so we rely on the value at ring init, stored in
	 * itlb_before_ctx_switch.
	 */
	if (IS_GEN6(ring->dev) && ring->itlb_before_ctx_switch) {
558
		ret = ring->flush(ring, I915_GEM_GPU_DOMAINS, 0);
559 560 561 562
		if (ret)
			return ret;
	}

563
	ret = intel_ring_begin(ring, 6);
564 565 566
	if (ret)
		return ret;

567
	/* WaProgramMiArbOnOffAroundMiSetContext:ivb,vlv,hsw */
568 569 570 571 572
	if (IS_GEN7(ring->dev))
		intel_ring_emit(ring, MI_ARB_ON_OFF | MI_ARB_DISABLE);
	else
		intel_ring_emit(ring, MI_NOOP);

573 574
	intel_ring_emit(ring, MI_NOOP);
	intel_ring_emit(ring, MI_SET_CONTEXT);
575
	intel_ring_emit(ring, i915_gem_obj_ggtt_offset(new_context->obj) |
576 577 578 579
			MI_MM_SPACE_GTT |
			MI_SAVE_EXT_STATE_EN |
			MI_RESTORE_EXT_STATE_EN |
			hw_flags);
580 581 582 583
	/*
	 * w/a: MI_SET_CONTEXT must always be followed by MI_NOOP
	 * WaMiSetContext_Hang:snb,ivb,vlv
	 */
584 585
	intel_ring_emit(ring, MI_NOOP);

586 587 588 589 590
	if (IS_GEN7(ring->dev))
		intel_ring_emit(ring, MI_ARB_ON_OFF | MI_ARB_ENABLE);
	else
		intel_ring_emit(ring, MI_NOOP);

591 592 593 594 595
	intel_ring_advance(ring);

	return ret;
}

596 597
static int do_switch(struct intel_ring_buffer *ring,
		     struct i915_hw_context *to)
598
{
599
	struct drm_i915_private *dev_priv = ring->dev->dev_private;
600
	struct i915_hw_context *from = ring->last_context;
601
	struct i915_hw_ppgtt *ppgtt = ctx_to_ppgtt(to);
602
	u32 hw_flags = 0;
603
	int ret, i;
604

605 606 607 608
	if (from != NULL && ring == &dev_priv->ring[RCS]) {
		BUG_ON(from->obj == NULL);
		BUG_ON(!i915_gem_obj_is_pinned(from->obj));
	}
609

610
	if (from == to && from->last_ring == ring && !to->remap_slice)
611 612
		return 0;

613 614 615 616 617 618 619
	/* Trying to pin first makes error handling easier. */
	if (ring == &dev_priv->ring[RCS]) {
		ret = i915_gem_obj_ggtt_pin(to->obj,
					    get_context_alignment(ring->dev),
					    false, false);
		if (ret)
			return ret;
620 621
	}

622 623 624 625 626 627 628
	/*
	 * Pin can switch back to the default context if we end up calling into
	 * evict_everything - as a last ditch gtt defrag effort that also
	 * switches to the default context. Hence we need to reload from here.
	 */
	from = ring->last_context;

629 630 631 632 633 634 635 636 637 638 639 640
	if (USES_FULL_PPGTT(ring->dev)) {
		ret = ppgtt->switch_mm(ppgtt, ring, false);
		if (ret)
			goto unpin_out;
	}

	if (ring != &dev_priv->ring[RCS]) {
		if (from)
			i915_gem_context_unreference(from);
		goto done;
	}

641 642
	/*
	 * Clear this page out of any CPU caches for coherent swap-in/out. Note
643 644 645
	 * that thanks to write = false in this call and us not setting any gpu
	 * write domains when putting a context object onto the active list
	 * (when switching away from it), this won't block.
646 647 648
	 *
	 * XXX: We need a real interface to do this instead of trickery.
	 */
649
	ret = i915_gem_object_set_to_gtt_domain(to->obj, false);
650 651
	if (ret)
		goto unpin_out;
652

653 654 655 656 657
	if (!to->obj->has_global_gtt_mapping) {
		struct i915_vma *vma = i915_gem_obj_to_vma(to->obj,
							   &dev_priv->gtt.base);
		vma->bind_vma(vma, to->obj->cache_level, GLOBAL_BIND);
	}
658

659 660 661 662
	if (!to->is_initialized || is_default_context(to))
		hw_flags |= MI_RESTORE_INHIBIT;

	ret = mi_set_context(ring, to, hw_flags);
663 664
	if (ret)
		goto unpin_out;
665

666 667 668 669 670 671 672 673 674 675 676 677
	for (i = 0; i < MAX_L3_SLICES; i++) {
		if (!(to->remap_slice & (1<<i)))
			continue;

		ret = i915_gem_l3_remap(ring, i);
		/* If it failed, try again next round */
		if (ret)
			DRM_DEBUG_DRIVER("L3 remapping failed\n");
		else
			to->remap_slice &= ~(1<<i);
	}

678 679 680 681 682 683
	/* The backing object for the context is done after switching to the
	 * *next* context. Therefore we cannot retire the previous context until
	 * the next context has already started running. In fact, the below code
	 * is a bit suboptimal because the retiring can occur simply after the
	 * MI_SET_CONTEXT instead of when the next seqno has completed.
	 */
684 685
	if (from != NULL) {
		from->obj->base.read_domains = I915_GEM_DOMAIN_INSTRUCTION;
B
Ben Widawsky 已提交
686
		i915_vma_move_to_active(i915_gem_obj_to_ggtt(from->obj), ring);
687 688 689 690 691 692 693
		/* As long as MI_SET_CONTEXT is serializing, ie. it flushes the
		 * whole damn pipeline, we don't need to explicitly mark the
		 * object dirty. The only exception is that the context must be
		 * correct in case the object gets swapped out. Ideally we'd be
		 * able to defer doing this until we know the object would be
		 * swapped, but there is no way to do that yet.
		 */
694 695 696
		from->obj->dirty = 1;
		BUG_ON(from->obj->ring != ring);

697
		/* obj is kept alive until the next request by its active ref */
B
Ben Widawsky 已提交
698
		i915_gem_object_ggtt_unpin(from->obj);
699
		i915_gem_context_unreference(from);
700 701
	}

702 703
	to->is_initialized = true;

704
done:
705 706
	i915_gem_context_reference(to);
	ring->last_context = to;
707
	to->last_ring = ring;
708 709

	return 0;
710 711 712 713 714

unpin_out:
	if (ring->id == RCS)
		i915_gem_object_ggtt_unpin(to->obj);
	return ret;
715 716 717 718 719 720 721 722 723 724 725 726 727 728 729
}

/**
 * i915_switch_context() - perform a GPU context switch.
 * @ring: ring for which we'll execute the context switch
 * @file_priv: file_priv associated with the context, may be NULL
 * @id: context id number
 *
 * The context life cycle is simple. The context refcount is incremented and
 * decremented by 1 and create and destroy. If the context is in use by the GPU,
 * it will have a refoucnt > 1. This allows us to destroy the context abstract
 * object while letting the normal object tracking destroy the backing BO.
 */
int i915_switch_context(struct intel_ring_buffer *ring,
			struct drm_file *file,
730
			struct i915_hw_context *to)
731 732 733
{
	struct drm_i915_private *dev_priv = ring->dev->dev_private;

734 735
	WARN_ON(!mutex_is_locked(&dev_priv->dev->struct_mutex));

736
	BUG_ON(file && to == NULL);
737

738 739 740 741
	/* We have the fake context, but don't supports switching. */
	if (!HAS_HW_CONTEXTS(ring->dev))
		return 0;

742
	return do_switch(ring, to);
743
}
744 745 746 747 748 749 750 751 752 753 754 755

int i915_gem_context_create_ioctl(struct drm_device *dev, void *data,
				  struct drm_file *file)
{
	struct drm_i915_gem_context_create *args = data;
	struct drm_i915_file_private *file_priv = file->driver_priv;
	struct i915_hw_context *ctx;
	int ret;

	if (!(dev->driver->driver_features & DRIVER_GEM))
		return -ENODEV;

756
	if (!HAS_HW_CONTEXTS(dev))
757 758
		return -ENODEV;

759 760 761 762
	ret = i915_mutex_lock_interruptible(dev);
	if (ret)
		return ret;

763
	ctx = i915_gem_create_context(dev, file_priv, USES_FULL_PPGTT(dev));
764
	mutex_unlock(&dev->struct_mutex);
765 766
	if (IS_ERR(ctx))
		return PTR_ERR(ctx);
767 768 769 770

	args->ctx_id = ctx->id;
	DRM_DEBUG_DRIVER("HW context %d created\n", args->ctx_id);

771
	return 0;
772 773 774 775 776 777 778 779 780 781 782 783 784
}

int i915_gem_context_destroy_ioctl(struct drm_device *dev, void *data,
				   struct drm_file *file)
{
	struct drm_i915_gem_context_destroy *args = data;
	struct drm_i915_file_private *file_priv = file->driver_priv;
	struct i915_hw_context *ctx;
	int ret;

	if (!(dev->driver->driver_features & DRIVER_GEM))
		return -ENODEV;

785
	if (args->ctx_id == DEFAULT_CONTEXT_ID)
786
		return -ENOENT;
787

788 789 790 791 792
	ret = i915_mutex_lock_interruptible(dev);
	if (ret)
		return ret;

	ctx = i915_gem_context_get(file_priv, args->ctx_id);
793
	if (IS_ERR(ctx)) {
794
		mutex_unlock(&dev->struct_mutex);
795
		return PTR_ERR(ctx);
796 797
	}

798 799
	idr_remove(&ctx->file_priv->context_idr, ctx->id);
	i915_gem_context_unreference(ctx);
800 801 802 803 804
	mutex_unlock(&dev->struct_mutex);

	DRM_DEBUG_DRIVER("HW context %d destroyed\n", args->ctx_id);
	return 0;
}