amdgpu_fence.c 22.3 KB
Newer Older
A
Alex Deucher 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 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 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128
/*
 * Copyright 2009 Jerome Glisse.
 * 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, sub license, 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 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 NON-INFRINGEMENT. IN NO EVENT SHALL
 * THE COPYRIGHT HOLDERS, AUTHORS AND/OR ITS SUPPLIERS 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.
 *
 * The above copyright notice and this permission notice (including the
 * next paragraph) shall be included in all copies or substantial portions
 * of the Software.
 *
 */
/*
 * Authors:
 *    Jerome Glisse <glisse@freedesktop.org>
 *    Dave Airlie
 */
#include <linux/seq_file.h>
#include <linux/atomic.h>
#include <linux/wait.h>
#include <linux/kref.h>
#include <linux/slab.h>
#include <linux/firmware.h>
#include <drm/drmP.h>
#include "amdgpu.h"
#include "amdgpu_trace.h"

/*
 * Fences
 * Fences mark an event in the GPUs pipeline and are used
 * for GPU/CPU synchronization.  When the fence is written,
 * it is expected that all buffers associated with that fence
 * are no longer in use by the associated ring on the GPU and
 * that the the relevant GPU caches have been flushed.
 */

/**
 * amdgpu_fence_write - write a fence value
 *
 * @ring: ring the fence is associated with
 * @seq: sequence number to write
 *
 * Writes a fence value to memory (all asics).
 */
static void amdgpu_fence_write(struct amdgpu_ring *ring, u32 seq)
{
	struct amdgpu_fence_driver *drv = &ring->fence_drv;

	if (drv->cpu_addr)
		*drv->cpu_addr = cpu_to_le32(seq);
}

/**
 * amdgpu_fence_read - read a fence value
 *
 * @ring: ring the fence is associated with
 *
 * Reads a fence value from memory (all asics).
 * Returns the value of the fence read from memory.
 */
static u32 amdgpu_fence_read(struct amdgpu_ring *ring)
{
	struct amdgpu_fence_driver *drv = &ring->fence_drv;
	u32 seq = 0;

	if (drv->cpu_addr)
		seq = le32_to_cpu(*drv->cpu_addr);
	else
		seq = lower_32_bits(atomic64_read(&drv->last_seq));

	return seq;
}

/**
 * amdgpu_fence_schedule_check - schedule lockup check
 *
 * @ring: pointer to struct amdgpu_ring
 *
 * Queues a delayed work item to check for lockups.
 */
static void amdgpu_fence_schedule_check(struct amdgpu_ring *ring)
{
	/*
	 * Do not reset the timer here with mod_delayed_work,
	 * this can livelock in an interaction with TTM delayed destroy.
	 */
	queue_delayed_work(system_power_efficient_wq,
		&ring->fence_drv.lockup_work,
		AMDGPU_FENCE_JIFFIES_TIMEOUT);
}

/**
 * amdgpu_fence_emit - emit a fence on the requested ring
 *
 * @ring: ring the fence is associated with
 * @owner: creator of the fence
 * @fence: amdgpu fence object
 *
 * Emits a fence command on the requested ring (all asics).
 * Returns 0 on success, -ENOMEM on failure.
 */
int amdgpu_fence_emit(struct amdgpu_ring *ring, void *owner,
		      struct amdgpu_fence **fence)
{
	struct amdgpu_device *adev = ring->adev;

	/* we are protected by the ring emission mutex */
	*fence = kmalloc(sizeof(struct amdgpu_fence), GFP_KERNEL);
	if ((*fence) == NULL) {
		return -ENOMEM;
	}
	(*fence)->seq = ++ring->fence_drv.sync_seq[ring->idx];
	(*fence)->ring = ring;
	(*fence)->owner = owner;
	fence_init(&(*fence)->base, &amdgpu_fence_ops,
129 130
		&ring->fence_drv.fence_queue.lock,
		adev->fence_context + ring->idx,
A
Alex Deucher 已提交
131
		(*fence)->seq);
132 133 134
	amdgpu_ring_emit_fence(ring, ring->fence_drv.gpu_addr,
			       (*fence)->seq,
			       AMDGPU_FENCE_FLAG_INT);
A
Alex Deucher 已提交
135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160
	trace_amdgpu_fence_emit(ring->adev->ddev, ring->idx, (*fence)->seq);
	return 0;
}

/**
 * amdgpu_fence_activity - check for fence activity
 *
 * @ring: pointer to struct amdgpu_ring
 *
 * Checks the current fence value and calculates the last
 * signalled fence value. Returns true if activity occured
 * on the ring, and the fence_queue should be waken up.
 */
static bool amdgpu_fence_activity(struct amdgpu_ring *ring)
{
	uint64_t seq, last_seq, last_emitted;
	unsigned count_loop = 0;
	bool wake = false;

	/* Note there is a scenario here for an infinite loop but it's
	 * very unlikely to happen. For it to happen, the current polling
	 * process need to be interrupted by another process and another
	 * process needs to update the last_seq btw the atomic read and
	 * xchg of the current process.
	 *
	 * More over for this to go in infinite loop there need to be
161
	 * continuously new fence signaled ie amdgpu_fence_read needs
A
Alex Deucher 已提交
162 163 164 165 166
	 * to return a different value each time for both the currently
	 * polling process and the other process that xchg the last_seq
	 * btw atomic read and xchg of the current process. And the
	 * value the other process set as last seq must be higher than
	 * the seq value we just read. Which means that current process
167
	 * need to be interrupted after amdgpu_fence_read and before
A
Alex Deucher 已提交
168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226
	 * atomic xchg.
	 *
	 * To be even more safe we count the number of time we loop and
	 * we bail after 10 loop just accepting the fact that we might
	 * have temporarly set the last_seq not to the true real last
	 * seq but to an older one.
	 */
	last_seq = atomic64_read(&ring->fence_drv.last_seq);
	do {
		last_emitted = ring->fence_drv.sync_seq[ring->idx];
		seq = amdgpu_fence_read(ring);
		seq |= last_seq & 0xffffffff00000000LL;
		if (seq < last_seq) {
			seq &= 0xffffffff;
			seq |= last_emitted & 0xffffffff00000000LL;
		}

		if (seq <= last_seq || seq > last_emitted) {
			break;
		}
		/* If we loop over we don't want to return without
		 * checking if a fence is signaled as it means that the
		 * seq we just read is different from the previous on.
		 */
		wake = true;
		last_seq = seq;
		if ((count_loop++) > 10) {
			/* We looped over too many time leave with the
			 * fact that we might have set an older fence
			 * seq then the current real last seq as signaled
			 * by the hw.
			 */
			break;
		}
	} while (atomic64_xchg(&ring->fence_drv.last_seq, seq) > seq);

	if (seq < last_emitted)
		amdgpu_fence_schedule_check(ring);

	return wake;
}

/**
 * amdgpu_fence_check_lockup - check for hardware lockup
 *
 * @work: delayed work item
 *
 * Checks for fence activity and if there is none probe
 * the hardware if a lockup occured.
 */
static void amdgpu_fence_check_lockup(struct work_struct *work)
{
	struct amdgpu_fence_driver *fence_drv;
	struct amdgpu_ring *ring;

	fence_drv = container_of(work, struct amdgpu_fence_driver,
				lockup_work.work);
	ring = fence_drv->ring;

227
	if (amdgpu_fence_activity(ring))
228
		wake_up_all(&ring->fence_drv.fence_queue);
A
Alex Deucher 已提交
229 230 231 232 233 234 235 236 237 238 239 240 241
}

/**
 * amdgpu_fence_process - process a fence
 *
 * @adev: amdgpu_device pointer
 * @ring: ring index the fence is associated with
 *
 * Checks the current fence value and wakes the fence queue
 * if the sequence number has increased (all asics).
 */
void amdgpu_fence_process(struct amdgpu_ring *ring)
{
242
	if (amdgpu_fence_activity(ring))
243
		wake_up_all(&ring->fence_drv.fence_queue);
A
Alex Deucher 已提交
244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271
}

/**
 * amdgpu_fence_seq_signaled - check if a fence sequence number has signaled
 *
 * @ring: ring the fence is associated with
 * @seq: sequence number
 *
 * Check if the last signaled fence sequnce number is >= the requested
 * sequence number (all asics).
 * Returns true if the fence has signaled (current fence value
 * is >= requested value) or false if it has not (current fence
 * value is < the requested value.  Helper function for
 * amdgpu_fence_signaled().
 */
static bool amdgpu_fence_seq_signaled(struct amdgpu_ring *ring, u64 seq)
{
	if (atomic64_read(&ring->fence_drv.last_seq) >= seq)
		return true;

	/* poll new last sequence at least once */
	amdgpu_fence_process(ring);
	if (atomic64_read(&ring->fence_drv.last_seq) >= seq)
		return true;

	return false;
}

272 273 274 275 276 277
/*
 * amdgpu_ring_wait_seq_timeout - wait for seq of the specific ring to signal
 * @ring: ring to wait on for the seq number
 * @seq: seq number wait for
 *
 * return value:
278 279
 * 0: seq signaled, and gpu not hang
 * -EDEADL: GPU hang detected
280
 * -EINVAL: some paramter is not valid
A
Alex Deucher 已提交
281
 */
282
static int amdgpu_fence_ring_wait_seq(struct amdgpu_ring *ring, uint64_t seq)
A
Alex Deucher 已提交
283
{
284
	bool signaled = false;
A
Alex Deucher 已提交
285

286 287 288
	BUG_ON(!ring);
	if (seq > ring->fence_drv.sync_seq[ring->idx])
		return -EINVAL;
A
Alex Deucher 已提交
289

290
	if (atomic64_read(&ring->fence_drv.last_seq) >= seq)
291
		return 0;
A
Alex Deucher 已提交
292

293
	amdgpu_fence_schedule_check(ring);
294
	wait_event(ring->fence_drv.fence_queue, (
295
		   (signaled = amdgpu_fence_seq_signaled(ring, seq))));
A
Alex Deucher 已提交
296

297 298 299 300
	if (signaled)
		return 0;
	else
		return -EDEADLK;
A
Alex Deucher 已提交
301 302 303 304 305 306 307 308 309 310 311 312 313 314
}

/**
 * amdgpu_fence_wait_next - wait for the next fence to signal
 *
 * @adev: amdgpu device pointer
 * @ring: ring index the fence is associated with
 *
 * Wait for the next fence on the requested ring to signal (all asics).
 * Returns 0 if the next fence has passed, error for all other cases.
 * Caller must hold ring lock.
 */
int amdgpu_fence_wait_next(struct amdgpu_ring *ring)
{
315
	uint64_t seq = atomic64_read(&ring->fence_drv.last_seq) + 1ULL;
316

317
	if (seq >= ring->fence_drv.sync_seq[ring->idx])
A
Alex Deucher 已提交
318
		return -ENOENT;
319

320
	return amdgpu_fence_ring_wait_seq(ring, seq);
A
Alex Deucher 已提交
321 322 323 324 325 326 327 328 329 330 331 332 333 334
}

/**
 * amdgpu_fence_wait_empty - wait for all fences to signal
 *
 * @adev: amdgpu device pointer
 * @ring: ring index the fence is associated with
 *
 * Wait for all fences on the requested ring to signal (all asics).
 * Returns 0 if the fences have passed, error for all other cases.
 * Caller must hold ring lock.
 */
int amdgpu_fence_wait_empty(struct amdgpu_ring *ring)
{
335
	uint64_t seq = ring->fence_drv.sync_seq[ring->idx];
336

337
	if (!seq)
A
Alex Deucher 已提交
338 339
		return 0;

340
	return amdgpu_fence_ring_wait_seq(ring, seq);
A
Alex Deucher 已提交
341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460
}

/**
 * amdgpu_fence_count_emitted - get the count of emitted fences
 *
 * @ring: ring the fence is associated with
 *
 * Get the number of fences emitted on the requested ring (all asics).
 * Returns the number of emitted fences on the ring.  Used by the
 * dynpm code to ring track activity.
 */
unsigned amdgpu_fence_count_emitted(struct amdgpu_ring *ring)
{
	uint64_t emitted;

	/* We are not protected by ring lock when reading the last sequence
	 * but it's ok to report slightly wrong fence count here.
	 */
	amdgpu_fence_process(ring);
	emitted = ring->fence_drv.sync_seq[ring->idx]
		- atomic64_read(&ring->fence_drv.last_seq);
	/* to avoid 32bits warp around */
	if (emitted > 0x10000000)
		emitted = 0x10000000;

	return (unsigned)emitted;
}

/**
 * amdgpu_fence_need_sync - do we need a semaphore
 *
 * @fence: amdgpu fence object
 * @dst_ring: which ring to check against
 *
 * Check if the fence needs to be synced against another ring
 * (all asics).  If so, we need to emit a semaphore.
 * Returns true if we need to sync with another ring, false if
 * not.
 */
bool amdgpu_fence_need_sync(struct amdgpu_fence *fence,
			    struct amdgpu_ring *dst_ring)
{
	struct amdgpu_fence_driver *fdrv;

	if (!fence)
		return false;

	if (fence->ring == dst_ring)
		return false;

	/* we are protected by the ring mutex */
	fdrv = &dst_ring->fence_drv;
	if (fence->seq <= fdrv->sync_seq[fence->ring->idx])
		return false;

	return true;
}

/**
 * amdgpu_fence_note_sync - record the sync point
 *
 * @fence: amdgpu fence object
 * @dst_ring: which ring to check against
 *
 * Note the sequence number at which point the fence will
 * be synced with the requested ring (all asics).
 */
void amdgpu_fence_note_sync(struct amdgpu_fence *fence,
			    struct amdgpu_ring *dst_ring)
{
	struct amdgpu_fence_driver *dst, *src;
	unsigned i;

	if (!fence)
		return;

	if (fence->ring == dst_ring)
		return;

	/* we are protected by the ring mutex */
	src = &fence->ring->fence_drv;
	dst = &dst_ring->fence_drv;
	for (i = 0; i < AMDGPU_MAX_RINGS; ++i) {
		if (i == dst_ring->idx)
			continue;

		dst->sync_seq[i] = max(dst->sync_seq[i], src->sync_seq[i]);
	}
}

/**
 * amdgpu_fence_driver_start_ring - make the fence driver
 * ready for use on the requested ring.
 *
 * @ring: ring to start the fence driver on
 * @irq_src: interrupt source to use for this ring
 * @irq_type: interrupt type to use for this ring
 *
 * Make the fence driver ready for processing (all asics).
 * Not all asics have all rings, so each asic will only
 * start the fence driver on the rings it has.
 * Returns 0 for success, errors for failure.
 */
int amdgpu_fence_driver_start_ring(struct amdgpu_ring *ring,
				   struct amdgpu_irq_src *irq_src,
				   unsigned irq_type)
{
	struct amdgpu_device *adev = ring->adev;
	uint64_t index;

	if (ring != &adev->uvd.ring) {
		ring->fence_drv.cpu_addr = &adev->wb.wb[ring->fence_offs];
		ring->fence_drv.gpu_addr = adev->wb.gpu_addr + (ring->fence_offs * 4);
	} else {
		/* put fence directly behind firmware */
		index = ALIGN(adev->uvd.fw->size, 8);
		ring->fence_drv.cpu_addr = adev->uvd.cpu_addr + index;
		ring->fence_drv.gpu_addr = adev->uvd.gpu_addr + index;
	}
	amdgpu_fence_write(ring, atomic64_read(&ring->fence_drv.last_seq));
461 462
	amdgpu_irq_get(adev, irq_src, irq_type);

A
Alex Deucher 已提交
463 464
	ring->fence_drv.irq_src = irq_src;
	ring->fence_drv.irq_type = irq_type;
465 466
	ring->fence_drv.initialized = true;

A
Alex Deucher 已提交
467 468 469 470 471 472 473 474 475 476 477 478 479 480 481
	dev_info(adev->dev, "fence driver on ring %d use gpu addr 0x%016llx, "
		 "cpu addr 0x%p\n", ring->idx,
		 ring->fence_drv.gpu_addr, ring->fence_drv.cpu_addr);
	return 0;
}

/**
 * amdgpu_fence_driver_init_ring - init the fence driver
 * for the requested ring.
 *
 * @ring: ring to init the fence driver on
 *
 * Init the fence driver for the requested ring (all asics).
 * Helper function for amdgpu_fence_driver_init().
 */
482
int amdgpu_fence_driver_init_ring(struct amdgpu_ring *ring)
A
Alex Deucher 已提交
483
{
484
	int i, r;
A
Alex Deucher 已提交
485 486 487 488 489 490 491 492 493 494 495 496

	ring->fence_drv.cpu_addr = NULL;
	ring->fence_drv.gpu_addr = 0;
	for (i = 0; i < AMDGPU_MAX_RINGS; ++i)
		ring->fence_drv.sync_seq[i] = 0;

	atomic64_set(&ring->fence_drv.last_seq, 0);
	ring->fence_drv.initialized = false;

	INIT_DELAYED_WORK(&ring->fence_drv.lockup_work,
			amdgpu_fence_check_lockup);
	ring->fence_drv.ring = ring;
497

498 499
	init_waitqueue_head(&ring->fence_drv.fence_queue);

500
	if (amdgpu_enable_scheduler) {
501 502 503 504 505 506 507 508 509 510 511
		long timeout = msecs_to_jiffies(amdgpu_lockup_timeout);
		if (timeout == 0) {
			/*
			 * FIXME:
			 * Delayed workqueue cannot use it directly,
			 * so the scheduler will not use delayed workqueue if
			 * MAX_SCHEDULE_TIMEOUT is set.
			 * Currently keep it simple and silly.
			 */
			timeout = MAX_SCHEDULE_TIMEOUT;
		}
512
		r = amd_sched_init(&ring->sched, &amdgpu_sched_ops,
513 514
				   amdgpu_sched_hw_submission,
				   timeout, ring->name);
515 516 517 518 519
		if (r) {
			DRM_ERROR("Failed to create scheduler on ring %s.\n",
				  ring->name);
			return r;
		}
520
	}
521 522

	return 0;
A
Alex Deucher 已提交
523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566
}

/**
 * amdgpu_fence_driver_init - init the fence driver
 * for all possible rings.
 *
 * @adev: amdgpu device pointer
 *
 * Init the fence driver for all possible rings (all asics).
 * Not all asics have all rings, so each asic will only
 * start the fence driver on the rings it has using
 * amdgpu_fence_driver_start_ring().
 * Returns 0 for success.
 */
int amdgpu_fence_driver_init(struct amdgpu_device *adev)
{
	if (amdgpu_debugfs_fence_init(adev))
		dev_err(adev->dev, "fence debugfs file creation failed\n");

	return 0;
}

/**
 * amdgpu_fence_driver_fini - tear down the fence driver
 * for all possible rings.
 *
 * @adev: amdgpu device pointer
 *
 * Tear down the fence driver for all possible rings (all asics).
 */
void amdgpu_fence_driver_fini(struct amdgpu_device *adev)
{
	int i, r;

	mutex_lock(&adev->ring_lock);
	for (i = 0; i < AMDGPU_MAX_RINGS; i++) {
		struct amdgpu_ring *ring = adev->rings[i];
		if (!ring || !ring->fence_drv.initialized)
			continue;
		r = amdgpu_fence_wait_empty(ring);
		if (r) {
			/* no need to trigger GPU reset as we are unloading */
			amdgpu_fence_driver_force_completion(adev);
		}
567
		wake_up_all(&ring->fence_drv.fence_queue);
568 569
		amdgpu_irq_put(adev, ring->fence_drv.irq_src,
			       ring->fence_drv.irq_type);
570
		amd_sched_fini(&ring->sched);
A
Alex Deucher 已提交
571 572 573 574 575
		ring->fence_drv.initialized = false;
	}
	mutex_unlock(&adev->ring_lock);
}

576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636
/**
 * amdgpu_fence_driver_suspend - suspend the fence driver
 * for all possible rings.
 *
 * @adev: amdgpu device pointer
 *
 * Suspend the fence driver for all possible rings (all asics).
 */
void amdgpu_fence_driver_suspend(struct amdgpu_device *adev)
{
	int i, r;

	mutex_lock(&adev->ring_lock);
	for (i = 0; i < AMDGPU_MAX_RINGS; i++) {
		struct amdgpu_ring *ring = adev->rings[i];
		if (!ring || !ring->fence_drv.initialized)
			continue;

		/* wait for gpu to finish processing current batch */
		r = amdgpu_fence_wait_empty(ring);
		if (r) {
			/* delay GPU reset to resume */
			amdgpu_fence_driver_force_completion(adev);
		}

		/* disable the interrupt */
		amdgpu_irq_put(adev, ring->fence_drv.irq_src,
			       ring->fence_drv.irq_type);
	}
	mutex_unlock(&adev->ring_lock);
}

/**
 * amdgpu_fence_driver_resume - resume the fence driver
 * for all possible rings.
 *
 * @adev: amdgpu device pointer
 *
 * Resume the fence driver for all possible rings (all asics).
 * Not all asics have all rings, so each asic will only
 * start the fence driver on the rings it has using
 * amdgpu_fence_driver_start_ring().
 * Returns 0 for success.
 */
void amdgpu_fence_driver_resume(struct amdgpu_device *adev)
{
	int i;

	mutex_lock(&adev->ring_lock);
	for (i = 0; i < AMDGPU_MAX_RINGS; i++) {
		struct amdgpu_ring *ring = adev->rings[i];
		if (!ring || !ring->fence_drv.initialized)
			continue;

		/* enable the interrupt */
		amdgpu_irq_get(adev, ring->fence_drv.irq_src,
			       ring->fence_drv.irq_type);
	}
	mutex_unlock(&adev->ring_lock);
}

A
Alex Deucher 已提交
637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657
/**
 * amdgpu_fence_driver_force_completion - force all fence waiter to complete
 *
 * @adev: amdgpu device pointer
 *
 * In case of GPU reset failure make sure no process keep waiting on fence
 * that will never complete.
 */
void amdgpu_fence_driver_force_completion(struct amdgpu_device *adev)
{
	int i;

	for (i = 0; i < AMDGPU_MAX_RINGS; i++) {
		struct amdgpu_ring *ring = adev->rings[i];
		if (!ring || !ring->fence_drv.initialized)
			continue;

		amdgpu_fence_write(ring, ring->fence_drv.sync_seq[i]);
	}
}

658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766
/*
 * Common fence implementation
 */

static const char *amdgpu_fence_get_driver_name(struct fence *fence)
{
	return "amdgpu";
}

static const char *amdgpu_fence_get_timeline_name(struct fence *f)
{
	struct amdgpu_fence *fence = to_amdgpu_fence(f);
	return (const char *)fence->ring->name;
}

/**
 * amdgpu_fence_is_signaled - test if fence is signaled
 *
 * @f: fence to test
 *
 * Test the fence sequence number if it is already signaled. If it isn't
 * signaled start fence processing. Returns True if the fence is signaled.
 */
static bool amdgpu_fence_is_signaled(struct fence *f)
{
	struct amdgpu_fence *fence = to_amdgpu_fence(f);
	struct amdgpu_ring *ring = fence->ring;

	if (atomic64_read(&ring->fence_drv.last_seq) >= fence->seq)
		return true;

	amdgpu_fence_process(ring);

	if (atomic64_read(&ring->fence_drv.last_seq) >= fence->seq)
		return true;

	return false;
}

/**
 * amdgpu_fence_check_signaled - callback from fence_queue
 *
 * this function is called with fence_queue lock held, which is also used
 * for the fence locking itself, so unlocked variants are used for
 * fence_signal, and remove_wait_queue.
 */
static int amdgpu_fence_check_signaled(wait_queue_t *wait, unsigned mode, int flags, void *key)
{
	struct amdgpu_fence *fence;
	struct amdgpu_device *adev;
	u64 seq;
	int ret;

	fence = container_of(wait, struct amdgpu_fence, fence_wake);
	adev = fence->ring->adev;

	/*
	 * We cannot use amdgpu_fence_process here because we're already
	 * in the waitqueue, in a call from wake_up_all.
	 */
	seq = atomic64_read(&fence->ring->fence_drv.last_seq);
	if (seq >= fence->seq) {
		ret = fence_signal_locked(&fence->base);
		if (!ret)
			FENCE_TRACE(&fence->base, "signaled from irq context\n");
		else
			FENCE_TRACE(&fence->base, "was already signaled\n");

		__remove_wait_queue(&fence->ring->fence_drv.fence_queue, &fence->fence_wake);
		fence_put(&fence->base);
	} else
		FENCE_TRACE(&fence->base, "pending\n");
	return 0;
}

/**
 * amdgpu_fence_enable_signaling - enable signalling on fence
 * @fence: fence
 *
 * This function is called with fence_queue lock held, and adds a callback
 * to fence_queue that checks if this fence is signaled, and if so it
 * signals the fence and removes itself.
 */
static bool amdgpu_fence_enable_signaling(struct fence *f)
{
	struct amdgpu_fence *fence = to_amdgpu_fence(f);
	struct amdgpu_ring *ring = fence->ring;

	if (atomic64_read(&ring->fence_drv.last_seq) >= fence->seq)
		return false;

	fence->fence_wake.flags = 0;
	fence->fence_wake.private = NULL;
	fence->fence_wake.func = amdgpu_fence_check_signaled;
	__add_wait_queue(&ring->fence_drv.fence_queue, &fence->fence_wake);
	fence_get(f);
	amdgpu_fence_schedule_check(ring);
	FENCE_TRACE(&fence->base, "armed on ring %i!\n", ring->idx);
	return true;
}

const struct fence_ops amdgpu_fence_ops = {
	.get_driver_name = amdgpu_fence_get_driver_name,
	.get_timeline_name = amdgpu_fence_get_timeline_name,
	.enable_signaling = amdgpu_fence_enable_signaling,
	.signaled = amdgpu_fence_is_signaled,
	.wait = fence_default_wait,
	.release = NULL,
};
A
Alex Deucher 已提交
767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785

/*
 * Fence debugfs
 */
#if defined(CONFIG_DEBUG_FS)
static int amdgpu_debugfs_fence_info(struct seq_file *m, void *data)
{
	struct drm_info_node *node = (struct drm_info_node *)m->private;
	struct drm_device *dev = node->minor->dev;
	struct amdgpu_device *adev = dev->dev_private;
	int i, j;

	for (i = 0; i < AMDGPU_MAX_RINGS; ++i) {
		struct amdgpu_ring *ring = adev->rings[i];
		if (!ring || !ring->fence_drv.initialized)
			continue;

		amdgpu_fence_process(ring);

786
		seq_printf(m, "--- ring %d (%s) ---\n", i, ring->name);
A
Alex Deucher 已提交
787 788 789 790 791 792 793
		seq_printf(m, "Last signaled fence 0x%016llx\n",
			   (unsigned long long)atomic64_read(&ring->fence_drv.last_seq));
		seq_printf(m, "Last emitted        0x%016llx\n",
			   ring->fence_drv.sync_seq[i]);

		for (j = 0; j < AMDGPU_MAX_RINGS; ++j) {
			struct amdgpu_ring *other = adev->rings[j];
794 795
			if (i != j && other && other->fence_drv.initialized &&
			    ring->fence_drv.sync_seq[j])
A
Alex Deucher 已提交
796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816
				seq_printf(m, "Last sync to ring %d 0x%016llx\n",
					   j, ring->fence_drv.sync_seq[j]);
		}
	}
	return 0;
}

static struct drm_info_list amdgpu_debugfs_fence_list[] = {
	{"amdgpu_fence_info", &amdgpu_debugfs_fence_info, 0, NULL},
};
#endif

int amdgpu_debugfs_fence_init(struct amdgpu_device *adev)
{
#if defined(CONFIG_DEBUG_FS)
	return amdgpu_debugfs_add_files(adev, amdgpu_debugfs_fence_list, 1);
#else
	return 0;
#endif
}