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#include <linux/seq_file.h>
30#include <linux/slab.h>
31#include <linux/uaccess.h>
32#include <linux/debugfs.h>
33
34#include <drm/amdgpu_drm.h>
35#include "amdgpu.h"
36#include "atom.h"
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51static int amdgpu_debugfs_ring_init(struct amdgpu_device *adev,
52 struct amdgpu_ring *ring);
53static void amdgpu_debugfs_ring_fini(struct amdgpu_ring *ring);
54
55
56
57
58
59
60
61
62
63
64
65int amdgpu_ring_alloc(struct amdgpu_ring *ring, unsigned ndw)
66{
67
68
69 ndw = (ndw + ring->funcs->align_mask) & ~ring->funcs->align_mask;
70
71
72
73
74 if (WARN_ON_ONCE(ndw > ring->max_dw))
75 return -ENOMEM;
76
77 ring->count_dw = ndw;
78 ring->wptr_old = ring->wptr;
79
80 if (ring->funcs->begin_use)
81 ring->funcs->begin_use(ring);
82
83 return 0;
84}
85
86
87
88
89
90
91
92
93void amdgpu_ring_insert_nop(struct amdgpu_ring *ring, uint32_t count)
94{
95 int i;
96
97 for (i = 0; i < count; i++)
98 amdgpu_ring_write(ring, ring->funcs->nop);
99}
100
101
102
103
104
105
106
107
108void amdgpu_ring_generic_pad_ib(struct amdgpu_ring *ring, struct amdgpu_ib *ib)
109{
110 while (ib->length_dw & ring->funcs->align_mask)
111 ib->ptr[ib->length_dw++] = ring->funcs->nop;
112}
113
114
115
116
117
118
119
120
121
122
123
124void amdgpu_ring_commit(struct amdgpu_ring *ring)
125{
126 uint32_t count;
127
128
129 count = ring->funcs->align_mask + 1 -
130 (ring->wptr & ring->funcs->align_mask);
131 count %= ring->funcs->align_mask + 1;
132 ring->funcs->insert_nop(ring, count);
133
134 mb();
135 amdgpu_ring_set_wptr(ring);
136
137 if (ring->funcs->end_use)
138 ring->funcs->end_use(ring);
139}
140
141
142
143
144
145
146
147
148void amdgpu_ring_undo(struct amdgpu_ring *ring)
149{
150 ring->wptr = ring->wptr_old;
151
152 if (ring->funcs->end_use)
153 ring->funcs->end_use(ring);
154}
155
156
157
158
159
160
161
162
163
164void amdgpu_ring_priority_put(struct amdgpu_ring *ring,
165 enum drm_sched_priority priority)
166{
167 int i;
168
169 if (!ring->funcs->set_priority)
170 return;
171
172 if (atomic_dec_return(&ring->num_jobs[priority]) > 0)
173 return;
174
175
176 if (priority == DRM_SCHED_PRIORITY_NORMAL)
177 return;
178
179 mutex_lock(&ring->priority_mutex);
180
181 if (ring->priority > priority)
182 goto out_unlock;
183
184
185 for (i = priority; i >= DRM_SCHED_PRIORITY_MIN; i--) {
186 if (i == DRM_SCHED_PRIORITY_NORMAL
187 || atomic_read(&ring->num_jobs[i])) {
188 ring->priority = i;
189 ring->funcs->set_priority(ring, i);
190 break;
191 }
192 }
193
194out_unlock:
195 mutex_unlock(&ring->priority_mutex);
196}
197
198
199
200
201
202
203
204
205
206void amdgpu_ring_priority_get(struct amdgpu_ring *ring,
207 enum drm_sched_priority priority)
208{
209 if (!ring->funcs->set_priority)
210 return;
211
212 if (atomic_inc_return(&ring->num_jobs[priority]) <= 0)
213 return;
214
215 mutex_lock(&ring->priority_mutex);
216 if (priority <= ring->priority)
217 goto out_unlock;
218
219 ring->priority = priority;
220 ring->funcs->set_priority(ring, priority);
221
222out_unlock:
223 mutex_unlock(&ring->priority_mutex);
224}
225
226
227
228
229
230
231
232
233
234
235
236
237int amdgpu_ring_init(struct amdgpu_device *adev, struct amdgpu_ring *ring,
238 unsigned max_dw, struct amdgpu_irq_src *irq_src,
239 unsigned irq_type)
240{
241 int r, i;
242 int sched_hw_submission = amdgpu_sched_hw_submission;
243
244
245
246
247
248
249
250 if (ring->funcs->type == AMDGPU_RING_TYPE_KIQ)
251 sched_hw_submission = max(sched_hw_submission, 256);
252 else if (ring == &adev->sdma.instance[0].page)
253 sched_hw_submission = 256;
254
255 if (ring->adev == NULL) {
256 if (adev->num_rings >= AMDGPU_MAX_RINGS)
257 return -EINVAL;
258
259 ring->adev = adev;
260 ring->idx = adev->num_rings++;
261 adev->rings[ring->idx] = ring;
262 r = amdgpu_fence_driver_init_ring(ring, sched_hw_submission);
263 if (r)
264 return r;
265 }
266
267 r = amdgpu_device_wb_get(adev, &ring->rptr_offs);
268 if (r) {
269 dev_err(adev->dev, "(%d) ring rptr_offs wb alloc failed\n", r);
270 return r;
271 }
272
273 r = amdgpu_device_wb_get(adev, &ring->wptr_offs);
274 if (r) {
275 dev_err(adev->dev, "(%d) ring wptr_offs wb alloc failed\n", r);
276 return r;
277 }
278
279 r = amdgpu_device_wb_get(adev, &ring->fence_offs);
280 if (r) {
281 dev_err(adev->dev, "(%d) ring fence_offs wb alloc failed\n", r);
282 return r;
283 }
284
285 r = amdgpu_device_wb_get(adev, &ring->trail_fence_offs);
286 if (r) {
287 dev_err(adev->dev,
288 "(%d) ring trail_fence_offs wb alloc failed\n", r);
289 return r;
290 }
291 ring->trail_fence_gpu_addr =
292 adev->wb.gpu_addr + (ring->trail_fence_offs * 4);
293 ring->trail_fence_cpu_addr = &adev->wb.wb[ring->trail_fence_offs];
294
295 r = amdgpu_device_wb_get(adev, &ring->cond_exe_offs);
296 if (r) {
297 dev_err(adev->dev, "(%d) ring cond_exec_polling wb alloc failed\n", r);
298 return r;
299 }
300 ring->cond_exe_gpu_addr = adev->wb.gpu_addr + (ring->cond_exe_offs * 4);
301 ring->cond_exe_cpu_addr = &adev->wb.wb[ring->cond_exe_offs];
302
303 *ring->cond_exe_cpu_addr = 1;
304
305 r = amdgpu_fence_driver_start_ring(ring, irq_src, irq_type);
306 if (r) {
307 dev_err(adev->dev, "failed initializing fences (%d).\n", r);
308 return r;
309 }
310
311 ring->ring_size = roundup_pow_of_two(max_dw * 4 * sched_hw_submission);
312
313 ring->buf_mask = (ring->ring_size / 4) - 1;
314 ring->ptr_mask = ring->funcs->support_64bit_ptrs ?
315 0xffffffffffffffff : ring->buf_mask;
316
317 if (ring->ring_obj == NULL) {
318 r = amdgpu_bo_create_kernel(adev, ring->ring_size + ring->funcs->extra_dw, PAGE_SIZE,
319 AMDGPU_GEM_DOMAIN_GTT,
320 &ring->ring_obj,
321 &ring->gpu_addr,
322 (void **)&ring->ring);
323 if (r) {
324 dev_err(adev->dev, "(%d) ring create failed\n", r);
325 return r;
326 }
327 amdgpu_ring_clear_ring(ring);
328 }
329
330 ring->max_dw = max_dw;
331 ring->priority = DRM_SCHED_PRIORITY_NORMAL;
332 mutex_init(&ring->priority_mutex);
333
334 for (i = 0; i < DRM_SCHED_PRIORITY_MAX; ++i)
335 atomic_set(&ring->num_jobs[i], 0);
336
337 if (amdgpu_debugfs_ring_init(adev, ring)) {
338 DRM_ERROR("Failed to register debugfs file for rings !\n");
339 }
340
341 return 0;
342}
343
344
345
346
347
348
349
350
351
352void amdgpu_ring_fini(struct amdgpu_ring *ring)
353{
354 ring->sched.ready = false;
355
356
357 if (!(ring->adev) || !(ring->adev->rings[ring->idx]))
358 return;
359
360 amdgpu_device_wb_free(ring->adev, ring->rptr_offs);
361 amdgpu_device_wb_free(ring->adev, ring->wptr_offs);
362
363 amdgpu_device_wb_free(ring->adev, ring->cond_exe_offs);
364 amdgpu_device_wb_free(ring->adev, ring->fence_offs);
365
366 amdgpu_bo_free_kernel(&ring->ring_obj,
367 &ring->gpu_addr,
368 (void **)&ring->ring);
369
370 amdgpu_debugfs_ring_fini(ring);
371
372 dma_fence_put(ring->vmid_wait);
373 ring->vmid_wait = NULL;
374 ring->me = 0;
375
376 ring->adev->rings[ring->idx] = NULL;
377}
378
379
380
381
382
383
384
385
386
387
388
389
390
391void amdgpu_ring_emit_reg_write_reg_wait_helper(struct amdgpu_ring *ring,
392 uint32_t reg0, uint32_t reg1,
393 uint32_t ref, uint32_t mask)
394{
395 amdgpu_ring_emit_wreg(ring, reg0, ref);
396 amdgpu_ring_emit_reg_wait(ring, reg1, mask, mask);
397}
398
399
400
401
402
403
404
405
406
407
408bool amdgpu_ring_soft_recovery(struct amdgpu_ring *ring, unsigned int vmid,
409 struct dma_fence *fence)
410{
411 ktime_t deadline = ktime_add_us(ktime_get(), 10000);
412
413 if (amdgpu_sriov_vf(ring->adev) || !ring->funcs->soft_recovery || !fence)
414 return false;
415
416 atomic_inc(&ring->adev->gpu_reset_counter);
417 while (!dma_fence_is_signaled(fence) &&
418 ktime_to_ns(ktime_sub(deadline, ktime_get())) > 0)
419 ring->funcs->soft_recovery(ring, vmid);
420
421 return dma_fence_is_signaled(fence);
422}
423
424
425
426
427#if defined(CONFIG_DEBUG_FS)
428
429
430
431
432
433
434
435
436static ssize_t amdgpu_debugfs_ring_read(struct file *f, char __user *buf,
437 size_t size, loff_t *pos)
438{
439 struct amdgpu_ring *ring = file_inode(f)->i_private;
440 int r, i;
441 uint32_t value, result, early[3];
442
443 if (*pos & 3 || size & 3)
444 return -EINVAL;
445
446 result = 0;
447
448 if (*pos < 12) {
449 early[0] = amdgpu_ring_get_rptr(ring) & ring->buf_mask;
450 early[1] = amdgpu_ring_get_wptr(ring) & ring->buf_mask;
451 early[2] = ring->wptr & ring->buf_mask;
452 for (i = *pos / 4; i < 3 && size; i++) {
453 r = put_user(early[i], (uint32_t *)buf);
454 if (r)
455 return r;
456 buf += 4;
457 result += 4;
458 size -= 4;
459 *pos += 4;
460 }
461 }
462
463 while (size) {
464 if (*pos >= (ring->ring_size + 12))
465 return result;
466
467 value = ring->ring[(*pos - 12)/4];
468 r = put_user(value, (uint32_t*)buf);
469 if (r)
470 return r;
471 buf += 4;
472 result += 4;
473 size -= 4;
474 *pos += 4;
475 }
476
477 return result;
478}
479
480static const struct file_operations amdgpu_debugfs_ring_fops = {
481 .owner = THIS_MODULE,
482 .read = amdgpu_debugfs_ring_read,
483 .llseek = default_llseek
484};
485
486#endif
487
488static int amdgpu_debugfs_ring_init(struct amdgpu_device *adev,
489 struct amdgpu_ring *ring)
490{
491#if defined(CONFIG_DEBUG_FS)
492 struct drm_minor *minor = adev->ddev->primary;
493 struct dentry *ent, *root = minor->debugfs_root;
494 char name[32];
495
496 sprintf(name, "amdgpu_ring_%s", ring->name);
497
498 ent = debugfs_create_file(name,
499 S_IFREG | S_IRUGO, root,
500 ring, &amdgpu_debugfs_ring_fops);
501 if (!ent)
502 return -ENOMEM;
503
504 i_size_write(ent->d_inode, ring->ring_size + 12);
505 ring->ent = ent;
506#endif
507 return 0;
508}
509
510static void amdgpu_debugfs_ring_fini(struct amdgpu_ring *ring)
511{
512#if defined(CONFIG_DEBUG_FS)
513 debugfs_remove(ring->ent);
514#endif
515}
516
517
518
519
520
521
522
523
524
525
526int amdgpu_ring_test_helper(struct amdgpu_ring *ring)
527{
528 struct amdgpu_device *adev = ring->adev;
529 int r;
530
531 r = amdgpu_ring_test_ring(ring);
532 if (r)
533 DRM_DEV_ERROR(adev->dev, "ring %s test failed (%d)\n",
534 ring->name, r);
535 else
536 DRM_DEV_DEBUG(adev->dev, "ring test on %s succeeded\n",
537 ring->name);
538
539 ring->sched.ready = !r;
540 return r;
541}
542