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
486
487
488
489
490
|
// SPDX-License-Identifier: GPL-2.0+
/* Copyright (C) 2018 Broadcom */
/**
* DOC: Broadcom V3D scheduling
*
* The shared DRM GPU scheduler is used to coordinate submitting jobs
* to the hardware. Each DRM fd (roughly a client process) gets its
* own scheduler entity, which will process jobs in order. The GPU
* scheduler will round-robin between clients to submit the next job.
*
* For simplicity, and in order to keep latency low for interactive
* jobs when bulk background jobs are queued up, we submit a new job
* to the HW only when it has completed the last one, instead of
* filling up the CT[01]Q FIFOs with jobs. Similarly, we use
* drm_sched_job_add_dependency() to manage the dependency between bin and
* render, instead of having the clients submit jobs using the HW's
* semaphores to interlock between them.
*/
#include <linux/sched/clock.h>
#include <linux/kthread.h>
#include "v3d_drv.h"
#include "v3d_regs.h"
#include "v3d_trace.h"
static struct v3d_job *
to_v3d_job(struct drm_sched_job *sched_job)
{
return container_of(sched_job, struct v3d_job, base);
}
static struct v3d_bin_job *
to_bin_job(struct drm_sched_job *sched_job)
{
return container_of(sched_job, struct v3d_bin_job, base.base);
}
static struct v3d_render_job *
to_render_job(struct drm_sched_job *sched_job)
{
return container_of(sched_job, struct v3d_render_job, base.base);
}
static struct v3d_tfu_job *
to_tfu_job(struct drm_sched_job *sched_job)
{
return container_of(sched_job, struct v3d_tfu_job, base.base);
}
static struct v3d_csd_job *
to_csd_job(struct drm_sched_job *sched_job)
{
return container_of(sched_job, struct v3d_csd_job, base.base);
}
static void
v3d_sched_job_free(struct drm_sched_job *sched_job)
{
struct v3d_job *job = to_v3d_job(sched_job);
v3d_job_cleanup(job);
}
static void
v3d_switch_perfmon(struct v3d_dev *v3d, struct v3d_job *job)
{
if (job->perfmon != v3d->active_perfmon)
v3d_perfmon_stop(v3d, v3d->active_perfmon, true);
if (job->perfmon && v3d->active_perfmon != job->perfmon)
v3d_perfmon_start(v3d, job->perfmon);
}
static struct dma_fence *v3d_bin_job_run(struct drm_sched_job *sched_job)
{
struct v3d_bin_job *job = to_bin_job(sched_job);
struct v3d_dev *v3d = job->base.v3d;
struct v3d_file_priv *file = job->base.file->driver_priv;
struct drm_device *dev = &v3d->drm;
struct dma_fence *fence;
unsigned long irqflags;
if (unlikely(job->base.base.s_fence->finished.error))
return NULL;
/* Lock required around bin_job update vs
* v3d_overflow_mem_work().
*/
spin_lock_irqsave(&v3d->job_lock, irqflags);
v3d->bin_job = job;
/* Clear out the overflow allocation, so we don't
* reuse the overflow attached to a previous job.
*/
V3D_CORE_WRITE(0, V3D_PTB_BPOS, 0);
spin_unlock_irqrestore(&v3d->job_lock, irqflags);
v3d_invalidate_caches(v3d);
fence = v3d_fence_create(v3d, V3D_BIN);
if (IS_ERR(fence))
return NULL;
if (job->base.irq_fence)
dma_fence_put(job->base.irq_fence);
job->base.irq_fence = dma_fence_get(fence);
trace_v3d_submit_cl(dev, false, to_v3d_fence(fence)->seqno,
job->start, job->end);
file->start_ns[V3D_BIN] = local_clock();
v3d->queue[V3D_BIN].start_ns = file->start_ns[V3D_BIN];
v3d_switch_perfmon(v3d, &job->base);
/* Set the current and end address of the control list.
* Writing the end register is what starts the job.
*/
if (job->qma) {
V3D_CORE_WRITE(0, V3D_CLE_CT0QMA, job->qma);
V3D_CORE_WRITE(0, V3D_CLE_CT0QMS, job->qms);
}
if (job->qts) {
V3D_CORE_WRITE(0, V3D_CLE_CT0QTS,
V3D_CLE_CT0QTS_ENABLE |
job->qts);
}
V3D_CORE_WRITE(0, V3D_CLE_CT0QBA, job->start);
V3D_CORE_WRITE(0, V3D_CLE_CT0QEA, job->end);
return fence;
}
static struct dma_fence *v3d_render_job_run(struct drm_sched_job *sched_job)
{
struct v3d_render_job *job = to_render_job(sched_job);
struct v3d_dev *v3d = job->base.v3d;
struct v3d_file_priv *file = job->base.file->driver_priv;
struct drm_device *dev = &v3d->drm;
struct dma_fence *fence;
if (unlikely(job->base.base.s_fence->finished.error))
return NULL;
v3d->render_job = job;
/* Can we avoid this flush? We need to be careful of
* scheduling, though -- imagine job0 rendering to texture and
* job1 reading, and them being executed as bin0, bin1,
* render0, render1, so that render1's flush at bin time
* wasn't enough.
*/
v3d_invalidate_caches(v3d);
fence = v3d_fence_create(v3d, V3D_RENDER);
if (IS_ERR(fence))
return NULL;
if (job->base.irq_fence)
dma_fence_put(job->base.irq_fence);
job->base.irq_fence = dma_fence_get(fence);
trace_v3d_submit_cl(dev, true, to_v3d_fence(fence)->seqno,
job->start, job->end);
file->start_ns[V3D_RENDER] = local_clock();
v3d->queue[V3D_RENDER].start_ns = file->start_ns[V3D_RENDER];
v3d_switch_perfmon(v3d, &job->base);
/* XXX: Set the QCFG */
/* Set the current and end address of the control list.
* Writing the end register is what starts the job.
*/
V3D_CORE_WRITE(0, V3D_CLE_CT1QBA, job->start);
V3D_CORE_WRITE(0, V3D_CLE_CT1QEA, job->end);
return fence;
}
static struct dma_fence *
v3d_tfu_job_run(struct drm_sched_job *sched_job)
{
struct v3d_tfu_job *job = to_tfu_job(sched_job);
struct v3d_dev *v3d = job->base.v3d;
struct v3d_file_priv *file = job->base.file->driver_priv;
struct drm_device *dev = &v3d->drm;
struct dma_fence *fence;
fence = v3d_fence_create(v3d, V3D_TFU);
if (IS_ERR(fence))
return NULL;
v3d->tfu_job = job;
if (job->base.irq_fence)
dma_fence_put(job->base.irq_fence);
job->base.irq_fence = dma_fence_get(fence);
trace_v3d_submit_tfu(dev, to_v3d_fence(fence)->seqno);
file->start_ns[V3D_TFU] = local_clock();
v3d->queue[V3D_TFU].start_ns = file->start_ns[V3D_TFU];
V3D_WRITE(V3D_TFU_IIA(v3d->ver), job->args.iia);
V3D_WRITE(V3D_TFU_IIS(v3d->ver), job->args.iis);
V3D_WRITE(V3D_TFU_ICA(v3d->ver), job->args.ica);
V3D_WRITE(V3D_TFU_IUA(v3d->ver), job->args.iua);
V3D_WRITE(V3D_TFU_IOA(v3d->ver), job->args.ioa);
if (v3d->ver >= 71)
V3D_WRITE(V3D_V7_TFU_IOC, job->args.v71.ioc);
V3D_WRITE(V3D_TFU_IOS(v3d->ver), job->args.ios);
V3D_WRITE(V3D_TFU_COEF0(v3d->ver), job->args.coef[0]);
if (v3d->ver >= 71 || (job->args.coef[0] & V3D_TFU_COEF0_USECOEF)) {
V3D_WRITE(V3D_TFU_COEF1(v3d->ver), job->args.coef[1]);
V3D_WRITE(V3D_TFU_COEF2(v3d->ver), job->args.coef[2]);
V3D_WRITE(V3D_TFU_COEF3(v3d->ver), job->args.coef[3]);
}
/* ICFG kicks off the job. */
V3D_WRITE(V3D_TFU_ICFG(v3d->ver), job->args.icfg | V3D_TFU_ICFG_IOC);
return fence;
}
static struct dma_fence *
v3d_csd_job_run(struct drm_sched_job *sched_job)
{
struct v3d_csd_job *job = to_csd_job(sched_job);
struct v3d_dev *v3d = job->base.v3d;
struct v3d_file_priv *file = job->base.file->driver_priv;
struct drm_device *dev = &v3d->drm;
struct dma_fence *fence;
int i, csd_cfg0_reg, csd_cfg_reg_count;
v3d->csd_job = job;
v3d_invalidate_caches(v3d);
fence = v3d_fence_create(v3d, V3D_CSD);
if (IS_ERR(fence))
return NULL;
if (job->base.irq_fence)
dma_fence_put(job->base.irq_fence);
job->base.irq_fence = dma_fence_get(fence);
trace_v3d_submit_csd(dev, to_v3d_fence(fence)->seqno);
file->start_ns[V3D_CSD] = local_clock();
v3d->queue[V3D_CSD].start_ns = file->start_ns[V3D_CSD];
v3d_switch_perfmon(v3d, &job->base);
csd_cfg0_reg = V3D_CSD_QUEUED_CFG0(v3d->ver);
csd_cfg_reg_count = v3d->ver < 71 ? 6 : 7;
for (i = 1; i <= csd_cfg_reg_count; i++)
V3D_CORE_WRITE(0, csd_cfg0_reg + 4 * i, job->args.cfg[i]);
/* CFG0 write kicks off the job. */
V3D_CORE_WRITE(0, csd_cfg0_reg, job->args.cfg[0]);
return fence;
}
static struct dma_fence *
v3d_cache_clean_job_run(struct drm_sched_job *sched_job)
{
struct v3d_job *job = to_v3d_job(sched_job);
struct v3d_dev *v3d = job->v3d;
struct v3d_file_priv *file = job->file->driver_priv;
u64 runtime;
file->start_ns[V3D_CACHE_CLEAN] = local_clock();
v3d->queue[V3D_CACHE_CLEAN].start_ns = file->start_ns[V3D_CACHE_CLEAN];
v3d_clean_caches(v3d);
runtime = local_clock() - file->start_ns[V3D_CACHE_CLEAN];
file->enabled_ns[V3D_CACHE_CLEAN] += runtime;
v3d->queue[V3D_CACHE_CLEAN].enabled_ns += runtime;
file->jobs_sent[V3D_CACHE_CLEAN]++;
v3d->queue[V3D_CACHE_CLEAN].jobs_sent++;
file->start_ns[V3D_CACHE_CLEAN] = 0;
v3d->queue[V3D_CACHE_CLEAN].start_ns = 0;
return NULL;
}
static enum drm_gpu_sched_stat
v3d_gpu_reset_for_timeout(struct v3d_dev *v3d, struct drm_sched_job *sched_job)
{
enum v3d_queue q;
mutex_lock(&v3d->reset_lock);
/* block scheduler */
for (q = 0; q < V3D_MAX_QUEUES; q++)
drm_sched_stop(&v3d->queue[q].sched, sched_job);
if (sched_job)
drm_sched_increase_karma(sched_job);
/* get the GPU back into the init state */
v3d_reset(v3d);
for (q = 0; q < V3D_MAX_QUEUES; q++)
drm_sched_resubmit_jobs(&v3d->queue[q].sched);
/* Unblock schedulers and restart their jobs. */
for (q = 0; q < V3D_MAX_QUEUES; q++) {
drm_sched_start(&v3d->queue[q].sched, true);
}
mutex_unlock(&v3d->reset_lock);
return DRM_GPU_SCHED_STAT_NOMINAL;
}
/* If the current address or return address have changed, then the GPU
* has probably made progress and we should delay the reset. This
* could fail if the GPU got in an infinite loop in the CL, but that
* is pretty unlikely outside of an i-g-t testcase.
*/
static enum drm_gpu_sched_stat
v3d_cl_job_timedout(struct drm_sched_job *sched_job, enum v3d_queue q,
u32 *timedout_ctca, u32 *timedout_ctra)
{
struct v3d_job *job = to_v3d_job(sched_job);
struct v3d_dev *v3d = job->v3d;
u32 ctca = V3D_CORE_READ(0, V3D_CLE_CTNCA(q));
u32 ctra = V3D_CORE_READ(0, V3D_CLE_CTNRA(q));
if (*timedout_ctca != ctca || *timedout_ctra != ctra) {
*timedout_ctca = ctca;
*timedout_ctra = ctra;
return DRM_GPU_SCHED_STAT_NOMINAL;
}
return v3d_gpu_reset_for_timeout(v3d, sched_job);
}
static enum drm_gpu_sched_stat
v3d_bin_job_timedout(struct drm_sched_job *sched_job)
{
struct v3d_bin_job *job = to_bin_job(sched_job);
return v3d_cl_job_timedout(sched_job, V3D_BIN,
&job->timedout_ctca, &job->timedout_ctra);
}
static enum drm_gpu_sched_stat
v3d_render_job_timedout(struct drm_sched_job *sched_job)
{
struct v3d_render_job *job = to_render_job(sched_job);
return v3d_cl_job_timedout(sched_job, V3D_RENDER,
&job->timedout_ctca, &job->timedout_ctra);
}
static enum drm_gpu_sched_stat
v3d_generic_job_timedout(struct drm_sched_job *sched_job)
{
struct v3d_job *job = to_v3d_job(sched_job);
return v3d_gpu_reset_for_timeout(job->v3d, sched_job);
}
static enum drm_gpu_sched_stat
v3d_csd_job_timedout(struct drm_sched_job *sched_job)
{
struct v3d_csd_job *job = to_csd_job(sched_job);
struct v3d_dev *v3d = job->base.v3d;
u32 batches = V3D_CORE_READ(0, V3D_CSD_CURRENT_CFG4(v3d->ver));
/* If we've made progress, skip reset and let the timer get
* rearmed.
*/
if (job->timedout_batches != batches) {
job->timedout_batches = batches;
return DRM_GPU_SCHED_STAT_NOMINAL;
}
return v3d_gpu_reset_for_timeout(v3d, sched_job);
}
static const struct drm_sched_backend_ops v3d_bin_sched_ops = {
.run_job = v3d_bin_job_run,
.timedout_job = v3d_bin_job_timedout,
.free_job = v3d_sched_job_free,
};
static const struct drm_sched_backend_ops v3d_render_sched_ops = {
.run_job = v3d_render_job_run,
.timedout_job = v3d_render_job_timedout,
.free_job = v3d_sched_job_free,
};
static const struct drm_sched_backend_ops v3d_tfu_sched_ops = {
.run_job = v3d_tfu_job_run,
.timedout_job = v3d_generic_job_timedout,
.free_job = v3d_sched_job_free,
};
static const struct drm_sched_backend_ops v3d_csd_sched_ops = {
.run_job = v3d_csd_job_run,
.timedout_job = v3d_csd_job_timedout,
.free_job = v3d_sched_job_free
};
static const struct drm_sched_backend_ops v3d_cache_clean_sched_ops = {
.run_job = v3d_cache_clean_job_run,
.timedout_job = v3d_generic_job_timedout,
.free_job = v3d_sched_job_free
};
int
v3d_sched_init(struct v3d_dev *v3d)
{
int hw_jobs_limit = 1;
int job_hang_limit = 0;
int hang_limit_ms = 500;
int ret;
ret = drm_sched_init(&v3d->queue[V3D_BIN].sched,
&v3d_bin_sched_ops, NULL,
DRM_SCHED_PRIORITY_COUNT,
hw_jobs_limit, job_hang_limit,
msecs_to_jiffies(hang_limit_ms), NULL,
NULL, "v3d_bin", v3d->drm.dev);
if (ret)
return ret;
ret = drm_sched_init(&v3d->queue[V3D_RENDER].sched,
&v3d_render_sched_ops, NULL,
DRM_SCHED_PRIORITY_COUNT,
hw_jobs_limit, job_hang_limit,
msecs_to_jiffies(hang_limit_ms), NULL,
NULL, "v3d_render", v3d->drm.dev);
if (ret)
goto fail;
ret = drm_sched_init(&v3d->queue[V3D_TFU].sched,
&v3d_tfu_sched_ops, NULL,
DRM_SCHED_PRIORITY_COUNT,
hw_jobs_limit, job_hang_limit,
msecs_to_jiffies(hang_limit_ms), NULL,
NULL, "v3d_tfu", v3d->drm.dev);
if (ret)
goto fail;
if (v3d_has_csd(v3d)) {
ret = drm_sched_init(&v3d->queue[V3D_CSD].sched,
&v3d_csd_sched_ops, NULL,
DRM_SCHED_PRIORITY_COUNT,
hw_jobs_limit, job_hang_limit,
msecs_to_jiffies(hang_limit_ms), NULL,
NULL, "v3d_csd", v3d->drm.dev);
if (ret)
goto fail;
ret = drm_sched_init(&v3d->queue[V3D_CACHE_CLEAN].sched,
&v3d_cache_clean_sched_ops, NULL,
DRM_SCHED_PRIORITY_COUNT,
hw_jobs_limit, job_hang_limit,
msecs_to_jiffies(hang_limit_ms), NULL,
NULL, "v3d_cache_clean", v3d->drm.dev);
if (ret)
goto fail;
}
return 0;
fail:
v3d_sched_fini(v3d);
return ret;
}
void
v3d_sched_fini(struct v3d_dev *v3d)
{
enum v3d_queue q;
for (q = 0; q < V3D_MAX_QUEUES; q++) {
if (v3d->queue[q].sched.ready)
drm_sched_fini(&v3d->queue[q].sched);
}
}
|