i915_gem_context.c 22.9 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 struct i915_hw_context *
i915_gem_context_get(struct drm_i915_file_private *file_priv, u32 id);
101 102
static int do_switch(struct intel_ring_buffer *ring,
		     struct i915_hw_context *to);
103

B
Ben Widawsky 已提交
104 105 106 107 108 109 110 111
static size_t get_context_alignment(struct drm_device *dev)
{
	if (IS_GEN6(dev))
		return GEN6_CONTEXT_ALIGN;

	return GEN7_CONTEXT_ALIGN;
}

112 113 114 115 116 117 118 119 120 121 122 123
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
124
		reg = I915_READ(GEN7_CXT_SIZE);
B
Ben Widawsky 已提交
125
		if (IS_HASWELL(dev))
126
			ret = HSW_CXT_TOTAL_SIZE;
B
Ben Widawsky 已提交
127 128
		else
			ret = GEN7_CXT_TOTAL_SIZE(reg) * 64;
129
		break;
B
Ben Widawsky 已提交
130 131 132
	case 8:
		ret = GEN8_CXT_TOTAL_SIZE;
		break;
133 134 135 136 137 138 139
	default:
		BUG();
	}

	return ret;
}

140
void i915_gem_context_free(struct kref *ctx_ref)
141
{
142 143
	struct i915_hw_context *ctx = container_of(ctx_ref,
						   typeof(*ctx), ref);
B
Ben Widawsky 已提交
144
	struct i915_hw_ppgtt *ppgtt = NULL;
145

B
Ben Widawsky 已提交
146 147
	/* We refcount even the aliasing PPGTT to keep the code symmetric */
	if (USES_ALIASING_PPGTT(ctx->obj->base.dev))
148
		ppgtt = ctx_to_ppgtt(ctx);
B
Ben Widawsky 已提交
149 150 151

	/* XXX: Free up the object before tearing down the address space, in
	 * case we're bound in the PPGTT */
152
	drm_gem_object_unreference(&ctx->obj->base);
B
Ben Widawsky 已提交
153 154 155 156

	if (ppgtt)
		kref_put(&ppgtt->ref, ppgtt_release);
	list_del(&ctx->link);
157 158 159
	kfree(ctx);
}

160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178
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;
}

179
static struct i915_hw_context *
180
__create_hw_context(struct drm_device *dev,
181
		  struct drm_i915_file_private *file_priv)
182 183
{
	struct drm_i915_private *dev_priv = dev->dev_private;
184
	struct i915_hw_context *ctx;
T
Tejun Heo 已提交
185
	int ret;
186

187
	ctx = kzalloc(sizeof(*ctx), GFP_KERNEL);
188 189
	if (ctx == NULL)
		return ERR_PTR(-ENOMEM);
190

191
	kref_init(&ctx->ref);
192
	ctx->obj = i915_gem_alloc_object(dev, dev_priv->hw_context_size);
193
	INIT_LIST_HEAD(&ctx->link);
194 195
	if (ctx->obj == NULL) {
		kfree(ctx);
196
		DRM_DEBUG_DRIVER("Context object allocated failed\n");
197
		return ERR_PTR(-ENOMEM);
198 199
	}

200 201
	if (INTEL_INFO(dev)->gen >= 7) {
		ret = i915_gem_object_set_cache_level(ctx->obj,
202
						      I915_CACHE_L3_LLC);
B
Ben Widawsky 已提交
203 204
		/* Failure shouldn't ever happen this early */
		if (WARN_ON(ret))
205 206 207
			goto err_out;
	}

208
	list_add_tail(&ctx->link, &dev_priv->context_list);
209 210 211

	/* Default context will never have a file_priv */
	if (file_priv == NULL)
212
		return ctx;
213

214
	ret = idr_alloc(&file_priv->context_idr, ctx, DEFAULT_CONTEXT_ID, 0,
T
Tejun Heo 已提交
215 216
			GFP_KERNEL);
	if (ret < 0)
217
		goto err_out;
218 219

	ctx->file_priv = file_priv;
T
Tejun Heo 已提交
220
	ctx->id = ret;
221 222 223 224
	/* 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;
225

226
	return ctx;
227 228

err_out:
229
	i915_gem_context_unreference(ctx);
230
	return ERR_PTR(ret);
231 232
}

233 234
static inline bool is_default_context(struct i915_hw_context *ctx)
{
235
	return (ctx->id == DEFAULT_CONTEXT_ID);
236 237
}

238 239 240 241 242
/**
 * 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.
 */
243
static struct i915_hw_context *
244 245 246
i915_gem_create_context(struct drm_device *dev,
			struct drm_i915_file_private *file_priv,
			bool create_vm)
247
{
248
	struct drm_i915_private *dev_priv = dev->dev_private;
249
	struct i915_hw_context *ctx;
250
	int ret = 0;
251

B
Ben Widawsky 已提交
252
	BUG_ON(!mutex_is_locked(&dev->struct_mutex));
253

254
	ctx = __create_hw_context(dev, file_priv);
255
	if (IS_ERR(ctx))
256
		return ctx;
257

258 259 260 261
	if (create_vm) {
		struct i915_hw_ppgtt *ppgtt = create_vm_for_ctx(dev, ctx);

		if (IS_ERR_OR_NULL(ppgtt)) {
262 263
			DRM_DEBUG_DRIVER("PPGTT setup failed (%ld)\n",
					 PTR_ERR(ppgtt));
264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297
			ret = PTR_ERR(ppgtt);
			goto err_destroy;
		} else
			ctx->vm = &ppgtt->base;

		/* This case is reserved for the global default context and
		 * should only happen once. */
		if (!file_priv) {
			if (WARN_ON(dev_priv->mm.aliasing_ppgtt)) {
				ret = -EEXIST;
				goto err_destroy;
			}

			dev_priv->mm.aliasing_ppgtt = ppgtt;

			/* 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;
			}
		}
	} else if (USES_ALIASING_PPGTT(dev)) {
		/* For platforms which only have aliasing PPGTT, we fake the
		 * address space and refcounting. */
		kref_get(&dev_priv->mm.aliasing_ppgtt->ref);
B
Ben Widawsky 已提交
298
	}
299

300 301 302 303 304 305
	/* TODO: Until full ppgtt... */
	if (USES_ALIASING_PPGTT(dev))
		ctx->vm = &dev_priv->mm.aliasing_ppgtt->base;
	else
		ctx->vm = &dev_priv->gtt.base;

306
	return ctx;
307 308

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
struct i915_ctx_hang_stats *
489
i915_gem_context_get_hang_stats(struct drm_device *dev,
490 491 492
				struct drm_file *file,
				u32 id)
{
493
	struct i915_hw_context *ctx;
494

495
	ctx = i915_gem_context_get(file->driver_priv, id);
496
	if (ctx == NULL)
497 498
		return ERR_PTR(-ENOENT);

499
	return &ctx->hang_stats;
500 501
}

502 503 504
int i915_gem_context_open(struct drm_device *dev, struct drm_file *file)
{
	struct drm_i915_file_private *file_priv = file->driver_priv;
505
	struct drm_i915_private *dev_priv = dev->dev_private;
506

507 508 509 510 511
	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;
512
		return 0;
513
	}
514 515 516

	idr_init(&file_priv->context_idr);

517 518 519 520 521 522 523 524 525 526
	mutex_lock(&dev->struct_mutex);
	file_priv->private_default_ctx =
		i915_gem_create_context(dev, file_priv, false);
	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);
	}

527 528 529
	return 0;
}

530 531
void i915_gem_context_close(struct drm_device *dev, struct drm_file *file)
{
532
	struct drm_i915_file_private *file_priv = file->driver_priv;
533

534 535
	if (!HAS_HW_CONTEXTS(dev)) {
		kfree(file_priv->private_default_ctx);
536
		return;
537
	}
538

539
	mutex_lock(&dev->struct_mutex);
540
	idr_for_each(&file_priv->context_idr, context_idr_cleanup, NULL);
541
	i915_gem_context_unreference(file_priv->private_default_ctx);
542 543 544 545
	idr_destroy(&file_priv->context_idr);
	mutex_unlock(&dev->struct_mutex);
}

546
static struct i915_hw_context *
547 548 549
i915_gem_context_get(struct drm_i915_file_private *file_priv, u32 id)
{
	return (struct i915_hw_context *)idr_find(&file_priv->context_idr, id);
550
}
551 552 553 554 555 556 557 558

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

559 560 561 562 563 564
	/* 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) {
565
		ret = ring->flush(ring, I915_GEM_GPU_DOMAINS, 0);
566 567 568 569
		if (ret)
			return ret;
	}

570
	ret = intel_ring_begin(ring, 6);
571 572 573
	if (ret)
		return ret;

574
	/* WaProgramMiArbOnOffAroundMiSetContext:ivb,vlv,hsw */
575 576 577 578 579
	if (IS_GEN7(ring->dev))
		intel_ring_emit(ring, MI_ARB_ON_OFF | MI_ARB_DISABLE);
	else
		intel_ring_emit(ring, MI_NOOP);

580 581
	intel_ring_emit(ring, MI_NOOP);
	intel_ring_emit(ring, MI_SET_CONTEXT);
582
	intel_ring_emit(ring, i915_gem_obj_ggtt_offset(new_context->obj) |
583 584 585 586 587 588 589
			MI_MM_SPACE_GTT |
			MI_SAVE_EXT_STATE_EN |
			MI_RESTORE_EXT_STATE_EN |
			hw_flags);
	/* w/a: MI_SET_CONTEXT must always be followed by MI_NOOP */
	intel_ring_emit(ring, MI_NOOP);

590 591 592 593 594
	if (IS_GEN7(ring->dev))
		intel_ring_emit(ring, MI_ARB_ON_OFF | MI_ARB_ENABLE);
	else
		intel_ring_emit(ring, MI_NOOP);

595 596 597 598 599
	intel_ring_advance(ring);

	return ret;
}

600 601
static int do_switch(struct intel_ring_buffer *ring,
		     struct i915_hw_context *to)
602
{
603
	struct drm_i915_private *dev_priv = ring->dev->dev_private;
604
	struct i915_hw_context *from = ring->last_context;
605
	u32 hw_flags = 0;
606
	int ret, i;
607

608 609 610 611
	if (from != NULL && ring == &dev_priv->ring[RCS]) {
		BUG_ON(from->obj == NULL);
		BUG_ON(!i915_gem_obj_is_pinned(from->obj));
	}
612

613
	if (from == to && from->last_ring == ring && !to->remap_slice)
614 615
		return 0;

616 617 618 619 620 621
	if (ring != &dev_priv->ring[RCS]) {
		if (from)
			i915_gem_context_unreference(from);
		goto done;
	}

B
Ben Widawsky 已提交
622 623
	ret = i915_gem_obj_ggtt_pin(to->obj, get_context_alignment(ring->dev),
				    false, false);
624 625 626
	if (ret)
		return ret;

627 628 629 630 631 632 633
	/* Clear this page out of any CPU caches for coherent swap-in/out. Note
	 * 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.
	 * XXX: We need a real interface to do this instead of trickery. */
	ret = i915_gem_object_set_to_gtt_domain(to->obj, false);
	if (ret) {
B
Ben Widawsky 已提交
634
		i915_gem_object_ggtt_unpin(to->obj);
635 636 637
		return ret;
	}

638 639 640 641 642
	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);
	}
643

644 645 646 647 648
	if (!to->is_initialized || is_default_context(to))
		hw_flags |= MI_RESTORE_INHIBIT;

	ret = mi_set_context(ring, to, hw_flags);
	if (ret) {
B
Ben Widawsky 已提交
649
		i915_gem_object_ggtt_unpin(to->obj);
650 651 652
		return ret;
	}

653 654 655 656 657 658 659 660 661 662 663 664
	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);
	}

665 666 667 668 669 670
	/* 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.
	 */
671 672
	if (from != NULL) {
		from->obj->base.read_domains = I915_GEM_DOMAIN_INSTRUCTION;
B
Ben Widawsky 已提交
673
		i915_vma_move_to_active(i915_gem_obj_to_ggtt(from->obj), ring);
674 675 676 677 678 679 680
		/* 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.
		 */
681 682 683
		from->obj->dirty = 1;
		BUG_ON(from->obj->ring != ring);

684
		/* obj is kept alive until the next request by its active ref */
B
Ben Widawsky 已提交
685
		i915_gem_object_ggtt_unpin(from->obj);
686
		i915_gem_context_unreference(from);
687 688
	}

689
done:
690 691
	i915_gem_context_reference(to);
	ring->last_context = to;
692
	to->is_initialized = true;
693
	to->last_ring = ring;
694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715

	return 0;
}

/**
 * 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,
			int to_id)
{
	struct drm_i915_private *dev_priv = ring->dev->dev_private;
	struct i915_hw_context *to;

716 717 718
	WARN_ON(!mutex_is_locked(&dev_priv->dev->struct_mutex));

	if (file == NULL)
719
		to = ring->default_context;
720
	else
721
		to = i915_gem_context_get(file->driver_priv, to_id);
722 723 724

	if (to == NULL)
		return -ENOENT;
725

726 727 728 729
	/* We have the fake context, but don't supports switching. */
	if (!HAS_HW_CONTEXTS(ring->dev))
		return 0;

730
	return do_switch(ring, to);
731
}
732 733 734 735 736 737 738 739 740 741 742 743

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;

744
	if (!HAS_HW_CONTEXTS(dev))
745 746
		return -ENODEV;

747 748 749 750
	ret = i915_mutex_lock_interruptible(dev);
	if (ret)
		return ret;

751
	ctx = i915_gem_create_context(dev, file_priv, false);
752
	mutex_unlock(&dev->struct_mutex);
753 754
	if (IS_ERR(ctx))
		return PTR_ERR(ctx);
755 756 757 758

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

759
	return 0;
760 761 762 763 764 765 766 767 768 769 770 771 772
}

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;

773 774 775
	if (args->ctx_id == DEFAULT_CONTEXT_ID)
		return -EPERM;

776 777 778 779 780 781 782
	ret = i915_mutex_lock_interruptible(dev);
	if (ret)
		return ret;

	ctx = i915_gem_context_get(file_priv, args->ctx_id);
	if (!ctx) {
		mutex_unlock(&dev->struct_mutex);
783
		return -ENOENT;
784 785
	}

786 787
	idr_remove(&ctx->file_priv->context_idr, ctx->id);
	i915_gem_context_unreference(ctx);
788 789 790 791 792
	mutex_unlock(&dev->struct_mutex);

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