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
129
130
131
132
133
134
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
161
162
163
164
165
166
167
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
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
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
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
298
299
300
301
302
303
304
305
306
307
308
309
310
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
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
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
|
/*
* Copyright 2015 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/kthread.h>
#include <linux/wait.h>
#include <linux/sched.h>
#include <drm/drmP.h>
#include "gpu_scheduler.h"
#define CREATE_TRACE_POINTS
#include "gpu_sched_trace.h"
static struct amd_sched_job *
amd_sched_entity_pop_job(struct amd_sched_entity *entity);
static void amd_sched_wakeup(struct amd_gpu_scheduler *sched);
struct kmem_cache *sched_fence_slab;
atomic_t sched_fence_slab_ref = ATOMIC_INIT(0);
/* Initialize a given run queue struct */
static void amd_sched_rq_init(struct amd_sched_rq *rq)
{
spin_lock_init(&rq->lock);
INIT_LIST_HEAD(&rq->entities);
rq->current_entity = NULL;
}
static void amd_sched_rq_add_entity(struct amd_sched_rq *rq,
struct amd_sched_entity *entity)
{
spin_lock(&rq->lock);
list_add_tail(&entity->list, &rq->entities);
spin_unlock(&rq->lock);
}
static void amd_sched_rq_remove_entity(struct amd_sched_rq *rq,
struct amd_sched_entity *entity)
{
spin_lock(&rq->lock);
list_del_init(&entity->list);
if (rq->current_entity == entity)
rq->current_entity = NULL;
spin_unlock(&rq->lock);
}
/**
* Select next job from a specified run queue with round robin policy.
* Return NULL if nothing available.
*/
static struct amd_sched_job *
amd_sched_rq_select_job(struct amd_sched_rq *rq)
{
struct amd_sched_entity *entity;
struct amd_sched_job *sched_job;
spin_lock(&rq->lock);
entity = rq->current_entity;
if (entity) {
list_for_each_entry_continue(entity, &rq->entities, list) {
sched_job = amd_sched_entity_pop_job(entity);
if (sched_job) {
rq->current_entity = entity;
spin_unlock(&rq->lock);
return sched_job;
}
}
}
list_for_each_entry(entity, &rq->entities, list) {
sched_job = amd_sched_entity_pop_job(entity);
if (sched_job) {
rq->current_entity = entity;
spin_unlock(&rq->lock);
return sched_job;
}
if (entity == rq->current_entity)
break;
}
spin_unlock(&rq->lock);
return NULL;
}
/**
* Init a context entity used by scheduler when submit to HW ring.
*
* @sched The pointer to the scheduler
* @entity The pointer to a valid amd_sched_entity
* @rq The run queue this entity belongs
* @kernel If this is an entity for the kernel
* @jobs The max number of jobs in the job queue
*
* return 0 if succeed. negative error code on failure
*/
int amd_sched_entity_init(struct amd_gpu_scheduler *sched,
struct amd_sched_entity *entity,
struct amd_sched_rq *rq,
uint32_t jobs)
{
int r;
if (!(sched && entity && rq))
return -EINVAL;
memset(entity, 0, sizeof(struct amd_sched_entity));
INIT_LIST_HEAD(&entity->list);
entity->rq = rq;
entity->sched = sched;
spin_lock_init(&entity->queue_lock);
r = kfifo_alloc(&entity->job_queue, jobs * sizeof(void *), GFP_KERNEL);
if (r)
return r;
atomic_set(&entity->fence_seq, 0);
entity->fence_context = fence_context_alloc(1);
/* Add the entity to the run queue */
amd_sched_rq_add_entity(rq, entity);
return 0;
}
/**
* Query if entity is initialized
*
* @sched Pointer to scheduler instance
* @entity The pointer to a valid scheduler entity
*
* return true if entity is initialized, false otherwise
*/
static bool amd_sched_entity_is_initialized(struct amd_gpu_scheduler *sched,
struct amd_sched_entity *entity)
{
return entity->sched == sched &&
entity->rq != NULL;
}
/**
* Check if entity is idle
*
* @entity The pointer to a valid scheduler entity
*
* Return true if entity don't has any unscheduled jobs.
*/
static bool amd_sched_entity_is_idle(struct amd_sched_entity *entity)
{
rmb();
if (kfifo_is_empty(&entity->job_queue))
return true;
return false;
}
/**
* Destroy a context entity
*
* @sched Pointer to scheduler instance
* @entity The pointer to a valid scheduler entity
*
* Cleanup and free the allocated resources.
*/
void amd_sched_entity_fini(struct amd_gpu_scheduler *sched,
struct amd_sched_entity *entity)
{
struct amd_sched_rq *rq = entity->rq;
if (!amd_sched_entity_is_initialized(sched, entity))
return;
/**
* The client will not queue more IBs during this fini, consume existing
* queued IBs
*/
wait_event(sched->job_scheduled, amd_sched_entity_is_idle(entity));
amd_sched_rq_remove_entity(rq, entity);
kfifo_free(&entity->job_queue);
}
static void amd_sched_entity_wakeup(struct fence *f, struct fence_cb *cb)
{
struct amd_sched_entity *entity =
container_of(cb, struct amd_sched_entity, cb);
entity->dependency = NULL;
fence_put(f);
amd_sched_wakeup(entity->sched);
}
static struct amd_sched_job *
amd_sched_entity_pop_job(struct amd_sched_entity *entity)
{
struct amd_gpu_scheduler *sched = entity->sched;
struct amd_sched_job *sched_job;
if (ACCESS_ONCE(entity->dependency))
return NULL;
if (!kfifo_out_peek(&entity->job_queue, &sched_job, sizeof(sched_job)))
return NULL;
while ((entity->dependency = sched->ops->dependency(sched_job))) {
if (entity->dependency->context == entity->fence_context) {
/* We can ignore fences from ourself */
fence_put(entity->dependency);
continue;
}
if (fence_add_callback(entity->dependency, &entity->cb,
amd_sched_entity_wakeup))
fence_put(entity->dependency);
else
return NULL;
}
return sched_job;
}
/**
* Helper to submit a job to the job queue
*
* @sched_job The pointer to job required to submit
*
* Returns true if we could submit the job.
*/
static bool amd_sched_entity_in(struct amd_sched_job *sched_job)
{
struct amd_sched_entity *entity = sched_job->s_entity;
bool added, first = false;
spin_lock(&entity->queue_lock);
added = kfifo_in(&entity->job_queue, &sched_job,
sizeof(sched_job)) == sizeof(sched_job);
if (added && kfifo_len(&entity->job_queue) == sizeof(sched_job))
first = true;
spin_unlock(&entity->queue_lock);
/* first job wakes up scheduler */
if (first)
amd_sched_wakeup(sched_job->sched);
return added;
}
/**
* Submit a job to the job queue
*
* @sched_job The pointer to job required to submit
*
* Returns 0 for success, negative error code otherwise.
*/
int amd_sched_entity_push_job(struct amd_sched_job *sched_job)
{
struct amd_sched_entity *entity = sched_job->s_entity;
struct amd_sched_fence *fence = amd_sched_fence_create(
entity, sched_job->owner);
if (!fence)
return -ENOMEM;
fence_get(&fence->base);
sched_job->s_fence = fence;
wait_event(entity->sched->job_scheduled,
amd_sched_entity_in(sched_job));
trace_amd_sched_job(sched_job);
return 0;
}
/**
* Return ture if we can push more jobs to the hw.
*/
static bool amd_sched_ready(struct amd_gpu_scheduler *sched)
{
return atomic_read(&sched->hw_rq_count) <
sched->hw_submission_limit;
}
/**
* Wake up the scheduler when it is ready
*/
static void amd_sched_wakeup(struct amd_gpu_scheduler *sched)
{
if (amd_sched_ready(sched))
wake_up_interruptible(&sched->wake_up_worker);
}
/**
* Select next to run
*/
static struct amd_sched_job *
amd_sched_select_job(struct amd_gpu_scheduler *sched)
{
struct amd_sched_job *sched_job;
if (!amd_sched_ready(sched))
return NULL;
/* Kernel run queue has higher priority than normal run queue*/
sched_job = amd_sched_rq_select_job(&sched->kernel_rq);
if (sched_job == NULL)
sched_job = amd_sched_rq_select_job(&sched->sched_rq);
return sched_job;
}
static void amd_sched_process_job(struct fence *f, struct fence_cb *cb)
{
struct amd_sched_fence *s_fence =
container_of(cb, struct amd_sched_fence, cb);
struct amd_gpu_scheduler *sched = s_fence->sched;
unsigned long flags;
atomic_dec(&sched->hw_rq_count);
amd_sched_fence_signal(s_fence);
if (sched->timeout != MAX_SCHEDULE_TIMEOUT) {
cancel_delayed_work(&s_fence->dwork);
spin_lock_irqsave(&sched->fence_list_lock, flags);
list_del_init(&s_fence->list);
spin_unlock_irqrestore(&sched->fence_list_lock, flags);
}
fence_put(&s_fence->base);
wake_up_interruptible(&sched->wake_up_worker);
}
static void amd_sched_fence_work_func(struct work_struct *work)
{
struct amd_sched_fence *s_fence =
container_of(work, struct amd_sched_fence, dwork.work);
struct amd_gpu_scheduler *sched = s_fence->sched;
struct amd_sched_fence *entity, *tmp;
unsigned long flags;
DRM_ERROR("[%s] scheduler is timeout!\n", sched->name);
/* Clean all pending fences */
spin_lock_irqsave(&sched->fence_list_lock, flags);
list_for_each_entry_safe(entity, tmp, &sched->fence_list, list) {
DRM_ERROR(" fence no %d\n", entity->base.seqno);
cancel_delayed_work(&entity->dwork);
list_del_init(&entity->list);
fence_put(&entity->base);
}
spin_unlock_irqrestore(&sched->fence_list_lock, flags);
}
static int amd_sched_main(void *param)
{
struct sched_param sparam = {.sched_priority = 1};
struct amd_gpu_scheduler *sched = (struct amd_gpu_scheduler *)param;
int r, count;
spin_lock_init(&sched->fence_list_lock);
INIT_LIST_HEAD(&sched->fence_list);
sched_setscheduler(current, SCHED_FIFO, &sparam);
while (!kthread_should_stop()) {
struct amd_sched_entity *entity;
struct amd_sched_fence *s_fence;
struct amd_sched_job *sched_job;
struct fence *fence;
unsigned long flags;
wait_event_interruptible(sched->wake_up_worker,
kthread_should_stop() ||
(sched_job = amd_sched_select_job(sched)));
if (!sched_job)
continue;
entity = sched_job->s_entity;
s_fence = sched_job->s_fence;
if (sched->timeout != MAX_SCHEDULE_TIMEOUT) {
INIT_DELAYED_WORK(&s_fence->dwork, amd_sched_fence_work_func);
schedule_delayed_work(&s_fence->dwork, sched->timeout);
spin_lock_irqsave(&sched->fence_list_lock, flags);
list_add_tail(&s_fence->list, &sched->fence_list);
spin_unlock_irqrestore(&sched->fence_list_lock, flags);
}
atomic_inc(&sched->hw_rq_count);
fence = sched->ops->run_job(sched_job);
if (fence) {
r = fence_add_callback(fence, &s_fence->cb,
amd_sched_process_job);
if (r == -ENOENT)
amd_sched_process_job(fence, &s_fence->cb);
else if (r)
DRM_ERROR("fence add callback failed (%d)\n", r);
fence_put(fence);
} else {
DRM_ERROR("Failed to run job!\n");
amd_sched_process_job(NULL, &s_fence->cb);
}
count = kfifo_out(&entity->job_queue, &sched_job,
sizeof(sched_job));
WARN_ON(count != sizeof(sched_job));
wake_up(&sched->job_scheduled);
}
return 0;
}
/**
* Init a gpu scheduler instance
*
* @sched The pointer to the scheduler
* @ops The backend operations for this scheduler.
* @hw_submissions Number of hw submissions to do.
* @name Name used for debugging
*
* Return 0 on success, otherwise error code.
*/
int amd_sched_init(struct amd_gpu_scheduler *sched,
struct amd_sched_backend_ops *ops,
unsigned hw_submission, long timeout, const char *name)
{
sched->ops = ops;
sched->hw_submission_limit = hw_submission;
sched->name = name;
sched->timeout = timeout;
amd_sched_rq_init(&sched->sched_rq);
amd_sched_rq_init(&sched->kernel_rq);
init_waitqueue_head(&sched->wake_up_worker);
init_waitqueue_head(&sched->job_scheduled);
atomic_set(&sched->hw_rq_count, 0);
if (atomic_inc_return(&sched_fence_slab_ref) == 1) {
sched_fence_slab = kmem_cache_create(
"amd_sched_fence", sizeof(struct amd_sched_fence), 0,
SLAB_HWCACHE_ALIGN, NULL);
if (!sched_fence_slab)
return -ENOMEM;
}
/* Each scheduler will run on a seperate kernel thread */
sched->thread = kthread_run(amd_sched_main, sched, sched->name);
if (IS_ERR(sched->thread)) {
DRM_ERROR("Failed to create scheduler for %s.\n", name);
return PTR_ERR(sched->thread);
}
return 0;
}
/**
* Destroy a gpu scheduler
*
* @sched The pointer to the scheduler
*/
void amd_sched_fini(struct amd_gpu_scheduler *sched)
{
if (sched->thread)
kthread_stop(sched->thread);
if (atomic_dec_and_test(&sched_fence_slab_ref))
kmem_cache_destroy(sched_fence_slab);
}
|