kfd_process.c 20.1 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
/*
 * Copyright 2014 Advanced Micro Devices, Inc.
 *
 * 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 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 COPYRIGHT HOLDER(S) OR AUTHOR(S) 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.
 */

#include <linux/mutex.h>
#include <linux/log2.h>
#include <linux/sched.h>
26
#include <linux/sched/mm.h>
27
#include <linux/sched/task.h>
28
#include <linux/slab.h>
29
#include <linux/amd-iommu.h>
30
#include <linux/notifier.h>
31
#include <linux/compat.h>
F
Felix Kuehling 已提交
32
#include <linux/mman.h>
33

34 35 36
struct mm_struct;

#include "kfd_priv.h"
37
#include "kfd_device_queue_manager.h"
38
#include "kfd_dbgmgr.h"
39
#include "kfd_iommu.h"
40 41 42 43 44

/*
 * List of struct kfd_process (field kfd_process).
 * Unique/indexed by mm_struct*
 */
45
DEFINE_HASHTABLE(kfd_processes_table, KFD_PROCESS_TABLE_SIZE);
46 47
static DEFINE_MUTEX(kfd_processes_mutex);

48
DEFINE_SRCU(kfd_processes_srcu);
49 50 51 52

static struct workqueue_struct *kfd_process_wq;

static struct kfd_process *find_process(const struct task_struct *thread);
53
static void kfd_process_ref_release(struct kref *ref);
54 55
static struct kfd_process *create_process(const struct task_struct *thread,
					struct file *filep);
F
Felix Kuehling 已提交
56 57
static int kfd_process_init_cwsr(struct kfd_process *p, struct file *filep);

58 59 60
static void evict_process_worker(struct work_struct *work);
static void restore_process_worker(struct work_struct *work);

61 62 63 64

void kfd_process_create_wq(void)
{
	if (!kfd_process_wq)
65
		kfd_process_wq = alloc_workqueue("kfd_process_wq", 0, 0);
66 67 68 69 70 71 72 73 74 75
}

void kfd_process_destroy_wq(void)
{
	if (kfd_process_wq) {
		destroy_workqueue(kfd_process_wq);
		kfd_process_wq = NULL;
	}
}

F
Felix Kuehling 已提交
76
struct kfd_process *kfd_create_process(struct file *filep)
77 78
{
	struct kfd_process *process;
F
Felix Kuehling 已提交
79
	struct task_struct *thread = current;
80

81
	if (!thread->mm)
82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97
		return ERR_PTR(-EINVAL);

	/* Only the pthreads threading model is supported. */
	if (thread->group_leader->mm != thread->mm)
		return ERR_PTR(-EINVAL);

	/*
	 * take kfd processes mutex before starting of process creation
	 * so there won't be a case where two threads of the same process
	 * create two kfd_process structures
	 */
	mutex_lock(&kfd_processes_mutex);

	/* A prior open of /dev/kfd could have already created the process. */
	process = find_process(thread);
	if (process)
98
		pr_debug("Process already found\n");
99 100
	else
		process = create_process(thread, filep);
101 102 103 104 105 106 107 108 109 110

	mutex_unlock(&kfd_processes_mutex);

	return process;
}

struct kfd_process *kfd_get_process(const struct task_struct *thread)
{
	struct kfd_process *process;

111
	if (!thread->mm)
112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146
		return ERR_PTR(-EINVAL);

	/* Only the pthreads threading model is supported. */
	if (thread->group_leader->mm != thread->mm)
		return ERR_PTR(-EINVAL);

	process = find_process(thread);

	return process;
}

static struct kfd_process *find_process_by_mm(const struct mm_struct *mm)
{
	struct kfd_process *process;

	hash_for_each_possible_rcu(kfd_processes_table, process,
					kfd_processes, (uintptr_t)mm)
		if (process->mm == mm)
			return process;

	return NULL;
}

static struct kfd_process *find_process(const struct task_struct *thread)
{
	struct kfd_process *p;
	int idx;

	idx = srcu_read_lock(&kfd_processes_srcu);
	p = find_process_by_mm(thread->mm);
	srcu_read_unlock(&kfd_processes_srcu, idx);

	return p;
}

147 148 149 150 151
void kfd_unref_process(struct kfd_process *p)
{
	kref_put(&p->ref, kfd_process_ref_release);
}

152
static void kfd_process_destroy_pdds(struct kfd_process *p)
153 154 155 156
{
	struct kfd_process_device *pdd, *temp;

	list_for_each_entry_safe(pdd, temp, &p->per_device_data,
157 158
				 per_device_list) {
		pr_debug("Releasing pdd (topology id %d) for process (pasid %d)\n",
159 160
				pdd->dev->id, p->pasid);

161 162 163 164
		if (pdd->vm)
			pdd->dev->kfd2kgd->destroy_process_vm(
				pdd->dev->kgd, pdd->vm);

165
		list_del(&pdd->per_device_list);
F
Felix Kuehling 已提交
166 167 168 169 170

		if (pdd->qpd.cwsr_kaddr)
			free_pages((unsigned long)pdd->qpd.cwsr_kaddr,
				get_order(KFD_CWSR_TBA_TMA_SIZE));

171 172
		kfree(pdd);
	}
173 174 175 176 177 178 179 180 181 182 183 184
}

/* No process locking is needed in this function, because the process
 * is not findable any more. We must assume that no other thread is
 * using it any more, otherwise we couldn't safely free the process
 * structure in the end.
 */
static void kfd_process_wq_release(struct work_struct *work)
{
	struct kfd_process *p = container_of(work, struct kfd_process,
					     release_work);

185
	kfd_iommu_unbind_process(p);
186 187

	kfd_process_destroy_pdds(p);
188
	dma_fence_put(p->ef);
189

190 191
	kfd_event_free_process(p);

192
	kfd_pasid_free(p->pasid);
193
	kfd_free_process_doorbells(p);
194 195 196

	mutex_destroy(&p->mutex);

197 198
	put_task_struct(p->lead_thread);

199 200 201
	kfree(p);
}

202
static void kfd_process_ref_release(struct kref *ref)
203
{
204
	struct kfd_process *p = container_of(ref, struct kfd_process, ref);
205

206 207 208
	INIT_WORK(&p->release_work, kfd_process_wq_release);
	queue_work(kfd_process_wq, &p->release_work);
}
209

210 211 212
static void kfd_process_destroy_delayed(struct rcu_head *rcu)
{
	struct kfd_process *p = container_of(rcu, struct kfd_process, rcu);
213

214
	kfd_unref_process(p);
215 216 217 218 219 220
}

static void kfd_process_notifier_release(struct mmu_notifier *mn,
					struct mm_struct *mm)
{
	struct kfd_process *p;
221
	struct kfd_process_device *pdd = NULL;
222 223 224 225 226 227

	/*
	 * The kfd_process structure can not be free because the
	 * mmu_notifier srcu is read locked
	 */
	p = container_of(mn, struct kfd_process, mmu_notifier);
228 229
	if (WARN_ON(p->mm != mm))
		return;
230 231 232 233 234 235

	mutex_lock(&kfd_processes_mutex);
	hash_del_rcu(&p->kfd_processes);
	mutex_unlock(&kfd_processes_mutex);
	synchronize_srcu(&kfd_processes_srcu);

236 237 238
	cancel_delayed_work_sync(&p->eviction_work);
	cancel_delayed_work_sync(&p->restore_work);

239 240
	mutex_lock(&p->mutex);

241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257
	/* Iterate over all process device data structures and if the
	 * pdd is in debug mode, we should first force unregistration,
	 * then we will be able to destroy the queues
	 */
	list_for_each_entry(pdd, &p->per_device_data, per_device_list) {
		struct kfd_dev *dev = pdd->dev;

		mutex_lock(kfd_get_dbgmgr_mutex());
		if (dev && dev->dbgmgr && dev->dbgmgr->pasid == p->pasid) {
			if (!kfd_dbgmgr_unregister(dev->dbgmgr, p)) {
				kfd_dbgmgr_destroy(dev->dbgmgr);
				dev->dbgmgr = NULL;
			}
		}
		mutex_unlock(kfd_get_dbgmgr_mutex());
	}

258
	kfd_process_dequeue_from_all_devices(p);
259 260
	pqm_uninit(&p->pqm);

261 262 263
	/* Indicate to other users that MM is no longer valid */
	p->mm = NULL;

264 265
	mutex_unlock(&p->mutex);

266
	mmu_notifier_unregister_no_release(&p->mmu_notifier, mm);
267 268 269 270 271 272 273
	mmu_notifier_call_srcu(&p->rcu, &kfd_process_destroy_delayed);
}

static const struct mmu_notifier_ops kfd_process_mmu_notifier_ops = {
	.release = kfd_process_notifier_release,
};

F
Felix Kuehling 已提交
274 275 276
static int kfd_process_init_cwsr(struct kfd_process *p, struct file *filep)
{
	unsigned long  offset;
277
	struct kfd_process_device *pdd = NULL;
F
Felix Kuehling 已提交
278 279 280
	struct kfd_dev *dev = NULL;
	struct qcm_process_device *qpd = NULL;

281
	list_for_each_entry(pdd, &p->per_device_data, per_device_list) {
F
Felix Kuehling 已提交
282 283 284 285 286 287 288 289 290 291
		dev = pdd->dev;
		qpd = &pdd->qpd;
		if (!dev->cwsr_enabled || qpd->cwsr_kaddr)
			continue;
		offset = (dev->id | KFD_MMAP_RESERVED_MEM_MASK) << PAGE_SHIFT;
		qpd->tba_addr = (int64_t)vm_mmap(filep, 0,
			KFD_CWSR_TBA_TMA_SIZE, PROT_READ | PROT_EXEC,
			MAP_SHARED, offset);

		if (IS_ERR_VALUE(qpd->tba_addr)) {
292 293 294
			int err = qpd->tba_addr;

			pr_err("Failure to set tba address. error %d.\n", err);
F
Felix Kuehling 已提交
295 296
			qpd->tba_addr = 0;
			qpd->cwsr_kaddr = NULL;
297
			return err;
F
Felix Kuehling 已提交
298 299 300 301 302 303 304 305
		}

		memcpy(qpd->cwsr_kaddr, dev->cwsr_isa, dev->cwsr_isa_size);

		qpd->tma_addr = qpd->tba_addr + KFD_CWSR_TMA_OFFSET;
		pr_debug("set tba :0x%llx, tma:0x%llx, cwsr_kaddr:%p for pqm.\n",
			qpd->tba_addr, qpd->tma_addr, qpd->cwsr_kaddr);
	}
306 307

	return 0;
F
Felix Kuehling 已提交
308 309
}

310 311
static struct kfd_process *create_process(const struct task_struct *thread,
					struct file *filep)
312 313 314 315 316 317 318 319 320 321 322 323 324
{
	struct kfd_process *process;
	int err = -ENOMEM;

	process = kzalloc(sizeof(*process), GFP_KERNEL);

	if (!process)
		goto err_alloc_process;

	process->pasid = kfd_pasid_alloc();
	if (process->pasid == 0)
		goto err_alloc_pasid;

325 326 327
	if (kfd_alloc_process_doorbells(process) < 0)
		goto err_alloc_doorbells;

328 329
	kref_init(&process->ref);

330 331 332 333 334 335
	mutex_init(&process->mutex);

	process->mm = thread->mm;

	/* register notifier */
	process->mmu_notifier.ops = &kfd_process_mmu_notifier_ops;
336
	err = mmu_notifier_register(&process->mmu_notifier, process->mm);
337 338 339 340 341 342 343
	if (err)
		goto err_mmu_notifier;

	hash_add_rcu(kfd_processes_table, &process->kfd_processes,
			(uintptr_t)process->mm);

	process->lead_thread = thread->group_leader;
344
	get_task_struct(process->lead_thread);
345 346 347

	INIT_LIST_HEAD(&process->per_device_data);

348 349
	kfd_event_init_process(process);

350 351 352 353
	err = pqm_init(&process->pqm, process);
	if (err != 0)
		goto err_process_pqm_init;

354
	/* init process apertures*/
355
	process->is_32bit_user_mode = in_compat_syscall();
356 357
	err = kfd_init_apertures(process);
	if (err != 0)
358
		goto err_init_apertures;
359

360 361 362 363
	INIT_DELAYED_WORK(&process->eviction_work, evict_process_worker);
	INIT_DELAYED_WORK(&process->restore_work, restore_process_worker);
	process->last_restore_timestamp = get_jiffies_64();

364 365 366 367
	err = kfd_process_init_cwsr(process, filep);
	if (err)
		goto err_init_cwsr;

368 369
	return process;

370 371
err_init_cwsr:
	kfd_process_destroy_pdds(process);
372
err_init_apertures:
373
	pqm_uninit(&process->pqm);
374 375 376 377
err_process_pqm_init:
	hash_del_rcu(&process->kfd_processes);
	synchronize_rcu();
	mmu_notifier_unregister_no_release(&process->mmu_notifier, process->mm);
378
err_mmu_notifier:
379
	mutex_destroy(&process->mutex);
380 381
	kfd_free_process_doorbells(process);
err_alloc_doorbells:
382 383 384 385 386 387 388 389
	kfd_pasid_free(process->pasid);
err_alloc_pasid:
	kfree(process);
err_alloc_process:
	return ERR_PTR(err);
}

struct kfd_process_device *kfd_get_process_device_data(struct kfd_dev *dev,
390
							struct kfd_process *p)
391 392 393 394 395
{
	struct kfd_process_device *pdd = NULL;

	list_for_each_entry(pdd, &p->per_device_data, per_device_list)
		if (pdd->dev == dev)
396
			return pdd;
397

398
	return NULL;
399 400 401 402 403 404 405 406
}

struct kfd_process_device *kfd_create_process_device_data(struct kfd_dev *dev,
							struct kfd_process *p)
{
	struct kfd_process_device *pdd = NULL;

	pdd = kzalloc(sizeof(*pdd), GFP_KERNEL);
407 408 409 410 411 412 413 414
	if (!pdd)
		return NULL;

	pdd->dev = dev;
	INIT_LIST_HEAD(&pdd->qpd.queues_list);
	INIT_LIST_HEAD(&pdd->qpd.priv_queue_list);
	pdd->qpd.dqm = dev->dqm;
	pdd->qpd.pqm = &p->pqm;
415
	pdd->qpd.evicted = 0;
416 417 418 419
	pdd->process = p;
	pdd->bound = PDD_UNBOUND;
	pdd->already_dequeued = false;
	list_add(&pdd->per_device_list, &p->per_device_data);
420

421 422 423 424 425 426
	/* Create the GPUVM context for this specific device */
	if (dev->kfd2kgd->create_process_vm(dev->kgd, &pdd->vm,
					    &p->kgd_process_info, &p->ef)) {
		pr_err("Failed to create process VM object\n");
		goto err_create_pdd;
	}
427
	return pdd;
428 429 430 431 432

err_create_pdd:
	list_del(&pdd->per_device_list);
	kfree(pdd);
	return NULL;
433 434 435 436 437 438 439 440 441 442 443 444
}

/*
 * Direct the IOMMU to bind the process (specifically the pasid->mm)
 * to the device.
 * Unbinding occurs when the process dies or the device is removed.
 *
 * Assumes that the process lock is held.
 */
struct kfd_process_device *kfd_bind_process_to_device(struct kfd_dev *dev,
							struct kfd_process *p)
{
445
	struct kfd_process_device *pdd;
446
	int err;
447

448 449 450
	pdd = kfd_get_process_device_data(dev, p);
	if (!pdd) {
		pr_err("Process device data doesn't exist\n");
451
		return ERR_PTR(-ENOMEM);
452
	}
453

454 455
	err = kfd_iommu_bind_process_to_device(pdd);
	if (err)
456 457
		return ERR_PTR(err);

458 459 460
	return pdd;
}

461 462
struct kfd_process_device *kfd_get_first_process_device_data(
						struct kfd_process *p)
463 464 465 466 467 468
{
	return list_first_entry(&p->per_device_data,
				struct kfd_process_device,
				per_device_list);
}

469 470
struct kfd_process_device *kfd_get_next_process_device_data(
						struct kfd_process *p,
471 472 473 474 475 476 477 478 479 480 481
						struct kfd_process_device *pdd)
{
	if (list_is_last(&pdd->per_device_list, &p->per_device_data))
		return NULL;
	return list_next_entry(pdd, per_device_list);
}

bool kfd_has_process_device_data(struct kfd_process *p)
{
	return !(list_empty(&p->per_device_data));
}
482

483
/* This increments the process->ref counter. */
484 485
struct kfd_process *kfd_lookup_process_by_pasid(unsigned int pasid)
{
486
	struct kfd_process *p, *ret_p = NULL;
487 488 489 490 491 492
	unsigned int temp;

	int idx = srcu_read_lock(&kfd_processes_srcu);

	hash_for_each_rcu(kfd_processes_table, temp, p, kfd_processes) {
		if (p->pasid == pasid) {
493
			kref_get(&p->ref);
494
			ret_p = p;
495 496 497 498 499 500
			break;
		}
	}

	srcu_read_unlock(&kfd_processes_srcu, idx);

501
	return ret_p;
502
}
F
Felix Kuehling 已提交
503

504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705
/* This increments the process->ref counter. */
struct kfd_process *kfd_lookup_process_by_mm(const struct mm_struct *mm)
{
	struct kfd_process *p;

	int idx = srcu_read_lock(&kfd_processes_srcu);

	p = find_process_by_mm(mm);
	if (p)
		kref_get(&p->ref);

	srcu_read_unlock(&kfd_processes_srcu, idx);

	return p;
}

/* process_evict_queues - Evict all user queues of a process
 *
 * Eviction is reference-counted per process-device. This means multiple
 * evictions from different sources can be nested safely.
 */
static int process_evict_queues(struct kfd_process *p)
{
	struct kfd_process_device *pdd;
	int r = 0;
	unsigned int n_evicted = 0;

	list_for_each_entry(pdd, &p->per_device_data, per_device_list) {
		r = pdd->dev->dqm->ops.evict_process_queues(pdd->dev->dqm,
							    &pdd->qpd);
		if (r) {
			pr_err("Failed to evict process queues\n");
			goto fail;
		}
		n_evicted++;
	}

	return r;

fail:
	/* To keep state consistent, roll back partial eviction by
	 * restoring queues
	 */
	list_for_each_entry(pdd, &p->per_device_data, per_device_list) {
		if (n_evicted == 0)
			break;
		if (pdd->dev->dqm->ops.restore_process_queues(pdd->dev->dqm,
							      &pdd->qpd))
			pr_err("Failed to restore queues\n");

		n_evicted--;
	}

	return r;
}

/* process_restore_queues - Restore all user queues of a process */
static  int process_restore_queues(struct kfd_process *p)
{
	struct kfd_process_device *pdd;
	int r, ret = 0;

	list_for_each_entry(pdd, &p->per_device_data, per_device_list) {
		r = pdd->dev->dqm->ops.restore_process_queues(pdd->dev->dqm,
							      &pdd->qpd);
		if (r) {
			pr_err("Failed to restore process queues\n");
			if (!ret)
				ret = r;
		}
	}

	return ret;
}

static void evict_process_worker(struct work_struct *work)
{
	int ret;
	struct kfd_process *p;
	struct delayed_work *dwork;

	dwork = to_delayed_work(work);

	/* Process termination destroys this worker thread. So during the
	 * lifetime of this thread, kfd_process p will be valid
	 */
	p = container_of(dwork, struct kfd_process, eviction_work);
	WARN_ONCE(p->last_eviction_seqno != p->ef->seqno,
		  "Eviction fence mismatch\n");

	/* Narrow window of overlap between restore and evict work
	 * item is possible. Once amdgpu_amdkfd_gpuvm_restore_process_bos
	 * unreserves KFD BOs, it is possible to evicted again. But
	 * restore has few more steps of finish. So lets wait for any
	 * previous restore work to complete
	 */
	flush_delayed_work(&p->restore_work);

	pr_debug("Started evicting pasid %d\n", p->pasid);
	ret = process_evict_queues(p);
	if (!ret) {
		dma_fence_signal(p->ef);
		dma_fence_put(p->ef);
		p->ef = NULL;
		schedule_delayed_work(&p->restore_work,
				msecs_to_jiffies(PROCESS_RESTORE_TIME_MS));

		pr_debug("Finished evicting pasid %d\n", p->pasid);
	} else
		pr_err("Failed to evict queues of pasid %d\n", p->pasid);
}

static void restore_process_worker(struct work_struct *work)
{
	struct delayed_work *dwork;
	struct kfd_process *p;
	struct kfd_process_device *pdd;
	int ret = 0;

	dwork = to_delayed_work(work);

	/* Process termination destroys this worker thread. So during the
	 * lifetime of this thread, kfd_process p will be valid
	 */
	p = container_of(dwork, struct kfd_process, restore_work);

	/* Call restore_process_bos on the first KGD device. This function
	 * takes care of restoring the whole process including other devices.
	 * Restore can fail if enough memory is not available. If so,
	 * reschedule again.
	 */
	pdd = list_first_entry(&p->per_device_data,
			       struct kfd_process_device,
			       per_device_list);

	pr_debug("Started restoring pasid %d\n", p->pasid);

	/* Setting last_restore_timestamp before successful restoration.
	 * Otherwise this would have to be set by KGD (restore_process_bos)
	 * before KFD BOs are unreserved. If not, the process can be evicted
	 * again before the timestamp is set.
	 * If restore fails, the timestamp will be set again in the next
	 * attempt. This would mean that the minimum GPU quanta would be
	 * PROCESS_ACTIVE_TIME_MS - (time to execute the following two
	 * functions)
	 */

	p->last_restore_timestamp = get_jiffies_64();
	ret = pdd->dev->kfd2kgd->restore_process_bos(p->kgd_process_info,
						     &p->ef);
	if (ret) {
		pr_debug("Failed to restore BOs of pasid %d, retry after %d ms\n",
			 p->pasid, PROCESS_BACK_OFF_TIME_MS);
		ret = schedule_delayed_work(&p->restore_work,
				msecs_to_jiffies(PROCESS_BACK_OFF_TIME_MS));
		WARN(!ret, "reschedule restore work failed\n");
		return;
	}

	ret = process_restore_queues(p);
	if (!ret)
		pr_debug("Finished restoring pasid %d\n", p->pasid);
	else
		pr_err("Failed to restore queues of pasid %d\n", p->pasid);
}

void kfd_suspend_all_processes(void)
{
	struct kfd_process *p;
	unsigned int temp;
	int idx = srcu_read_lock(&kfd_processes_srcu);

	hash_for_each_rcu(kfd_processes_table, temp, p, kfd_processes) {
		cancel_delayed_work_sync(&p->eviction_work);
		cancel_delayed_work_sync(&p->restore_work);

		if (process_evict_queues(p))
			pr_err("Failed to suspend process %d\n", p->pasid);
		dma_fence_signal(p->ef);
		dma_fence_put(p->ef);
		p->ef = NULL;
	}
	srcu_read_unlock(&kfd_processes_srcu, idx);
}

int kfd_resume_all_processes(void)
{
	struct kfd_process *p;
	unsigned int temp;
	int ret = 0, idx = srcu_read_lock(&kfd_processes_srcu);

	hash_for_each_rcu(kfd_processes_table, temp, p, kfd_processes) {
		if (!schedule_delayed_work(&p->restore_work, 0)) {
			pr_err("Restore process %d failed during resume\n",
			       p->pasid);
			ret = -EFAULT;
		}
	}
	srcu_read_unlock(&kfd_processes_srcu, idx);
	return ret;
}

F
Felix Kuehling 已提交
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
int kfd_reserved_mem_mmap(struct kfd_process *process,
			  struct vm_area_struct *vma)
{
	struct kfd_dev *dev = kfd_device_by_id(vma->vm_pgoff);
	struct kfd_process_device *pdd;
	struct qcm_process_device *qpd;

	if (!dev)
		return -EINVAL;
	if ((vma->vm_end - vma->vm_start) != KFD_CWSR_TBA_TMA_SIZE) {
		pr_err("Incorrect CWSR mapping size.\n");
		return -EINVAL;
	}

	pdd = kfd_get_process_device_data(dev, process);
	if (!pdd)
		return -EINVAL;
	qpd = &pdd->qpd;

	qpd->cwsr_kaddr = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
					get_order(KFD_CWSR_TBA_TMA_SIZE));
	if (!qpd->cwsr_kaddr) {
		pr_err("Error allocating per process CWSR buffer.\n");
		return -ENOMEM;
	}

	vma->vm_flags |= VM_IO | VM_DONTCOPY | VM_DONTEXPAND
		| VM_NORESERVE | VM_DONTDUMP | VM_PFNMAP;
	/* Mapping pages to user process */
	return remap_pfn_range(vma, vma->vm_start,
			       PFN_DOWN(__pa(qpd->cwsr_kaddr)),
			       KFD_CWSR_TBA_TMA_SIZE, vma->vm_page_prot);
}
739

740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755
void kfd_flush_tlb(struct kfd_process_device *pdd)
{
	struct kfd_dev *dev = pdd->dev;
	const struct kfd2kgd_calls *f2g = dev->kfd2kgd;

	if (dev->dqm->sched_policy == KFD_SCHED_POLICY_NO_HWS) {
		/* Nothing to flush until a VMID is assigned, which
		 * only happens when the first queue is created.
		 */
		if (pdd->qpd.vmid)
			f2g->invalidate_tlbs_vmid(dev->kgd, pdd->qpd.vmid);
	} else {
		f2g->invalidate_tlbs(dev->kgd, pdd->process->pasid);
	}
}

756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783
#if defined(CONFIG_DEBUG_FS)

int kfd_debugfs_mqds_by_process(struct seq_file *m, void *data)
{
	struct kfd_process *p;
	unsigned int temp;
	int r = 0;

	int idx = srcu_read_lock(&kfd_processes_srcu);

	hash_for_each_rcu(kfd_processes_table, temp, p, kfd_processes) {
		seq_printf(m, "Process %d PASID %d:\n",
			   p->lead_thread->tgid, p->pasid);

		mutex_lock(&p->mutex);
		r = pqm_debugfs_mqds(m, &p->pqm);
		mutex_unlock(&p->mutex);

		if (r)
			break;
	}

	srcu_read_unlock(&kfd_processes_srcu, idx);

	return r;
}

#endif