1 /*
2  * Copyright 2017 Advanced Micro Devices, Inc.
3  *
4  * Permission is hereby granted, free of charge, to any person obtaining a
5  * copy of this software and associated documentation files (the "Software"),
6  * to deal in the Software without restriction, including without limitation
7  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8  * and/or sell copies of the Software, and to permit persons to whom the
9  * Software is furnished to do so, subject to the following conditions:
10  *
11  * The above copyright notice and this permission notice shall be included in
12  * all copies or substantial portions of the Software.
13  *
14  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
15  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
16  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
17  * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
18  * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
19  * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
20  * OTHER DEALINGS IN THE SOFTWARE.
21  *
22  */
23 
24 #if !defined(_AMDGPU_TRACE_H) || defined(TRACE_HEADER_MULTI_READ)
25 #define _AMDGPU_TRACE_H_
26 
27 #include <linux/stringify.h>
28 #include <linux/types.h>
29 #include <linux/tracepoint.h>
30 
31 #undef TRACE_SYSTEM
32 #define TRACE_SYSTEM amdgpu
33 #define TRACE_INCLUDE_FILE amdgpu_trace
34 
35 #define AMDGPU_JOB_GET_TIMELINE_NAME(job) \
36 	 job->base.s_fence->finished.ops->get_timeline_name(&job->base.s_fence->finished)
37 
38 TRACE_EVENT(amdgpu_device_rreg,
39 	    TP_PROTO(unsigned did, uint32_t reg, uint32_t value),
40 	    TP_ARGS(did, reg, value),
41 	    TP_STRUCT__entry(
42 				__field(unsigned, did)
43 				__field(uint32_t, reg)
44 				__field(uint32_t, value)
45 			    ),
46 	    TP_fast_assign(
47 			   __entry->did = did;
48 			   __entry->reg = reg;
49 			   __entry->value = value;
50 			   ),
51 	    TP_printk("0x%04lx, 0x%08lx, 0x%08lx",
52 		      (unsigned long)__entry->did,
53 		      (unsigned long)__entry->reg,
54 		      (unsigned long)__entry->value)
55 );
56 
57 TRACE_EVENT(amdgpu_device_wreg,
58 	    TP_PROTO(unsigned did, uint32_t reg, uint32_t value),
59 	    TP_ARGS(did, reg, value),
60 	    TP_STRUCT__entry(
61 				__field(unsigned, did)
62 				__field(uint32_t, reg)
63 				__field(uint32_t, value)
64 			    ),
65 	    TP_fast_assign(
66 			   __entry->did = did;
67 			   __entry->reg = reg;
68 			   __entry->value = value;
69 			   ),
70 	    TP_printk("0x%04lx, 0x%08lx, 0x%08lx",
71 		      (unsigned long)__entry->did,
72 		      (unsigned long)__entry->reg,
73 		      (unsigned long)__entry->value)
74 );
75 
76 TRACE_EVENT(amdgpu_iv,
77 	    TP_PROTO(unsigned ih, struct amdgpu_iv_entry *iv),
78 	    TP_ARGS(ih, iv),
79 	    TP_STRUCT__entry(
80 			     __field(unsigned, ih)
81 			     __field(unsigned, client_id)
82 			     __field(unsigned, src_id)
83 			     __field(unsigned, ring_id)
84 			     __field(unsigned, vmid)
85 			     __field(unsigned, vmid_src)
86 			     __field(uint64_t, timestamp)
87 			     __field(unsigned, timestamp_src)
88 			     __field(unsigned, pasid)
89 			     __array(unsigned, src_data, 4)
90 			    ),
91 	    TP_fast_assign(
92 			   __entry->ih = ih;
93 			   __entry->client_id = iv->client_id;
94 			   __entry->src_id = iv->src_id;
95 			   __entry->ring_id = iv->ring_id;
96 			   __entry->vmid = iv->vmid;
97 			   __entry->vmid_src = iv->vmid_src;
98 			   __entry->timestamp = iv->timestamp;
99 			   __entry->timestamp_src = iv->timestamp_src;
100 			   __entry->pasid = iv->pasid;
101 			   __entry->src_data[0] = iv->src_data[0];
102 			   __entry->src_data[1] = iv->src_data[1];
103 			   __entry->src_data[2] = iv->src_data[2];
104 			   __entry->src_data[3] = iv->src_data[3];
105 			   ),
106 	    TP_printk("ih:%u client_id:%u src_id:%u ring:%u vmid:%u "
107 		      "timestamp: %llu pasid:%u src_data: %08x %08x %08x %08x",
108 		      __entry->ih, __entry->client_id, __entry->src_id,
109 		      __entry->ring_id, __entry->vmid,
110 		      __entry->timestamp, __entry->pasid,
111 		      __entry->src_data[0], __entry->src_data[1],
112 		      __entry->src_data[2], __entry->src_data[3])
113 );
114 
115 
116 TRACE_EVENT(amdgpu_bo_create,
117 	    TP_PROTO(struct amdgpu_bo *bo),
118 	    TP_ARGS(bo),
119 	    TP_STRUCT__entry(
120 			     __field(struct amdgpu_bo *, bo)
121 			     __field(u32, pages)
122 			     __field(u32, type)
123 			     __field(u32, prefer)
124 			     __field(u32, allow)
125 			     __field(u32, visible)
126 			     ),
127 
128 	    TP_fast_assign(
129 			   __entry->bo = bo;
130 			   __entry->pages = bo->tbo.num_pages;
131 			   __entry->type = bo->tbo.mem.mem_type;
132 			   __entry->prefer = bo->preferred_domains;
133 			   __entry->allow = bo->allowed_domains;
134 			   __entry->visible = bo->flags;
135 			   ),
136 
137 	    TP_printk("bo=%p, pages=%u, type=%d, preferred=%d, allowed=%d, visible=%d",
138 		       __entry->bo, __entry->pages, __entry->type,
139 		       __entry->prefer, __entry->allow, __entry->visible)
140 );
141 
142 TRACE_EVENT(amdgpu_cs,
143 	    TP_PROTO(struct amdgpu_cs_parser *p, int i),
144 	    TP_ARGS(p, i),
145 	    TP_STRUCT__entry(
146 			     __field(struct amdgpu_bo_list *, bo_list)
147 			     __field(u32, ring)
148 			     __field(u32, dw)
149 			     __field(u32, fences)
150 			     ),
151 
152 	    TP_fast_assign(
153 			   __entry->bo_list = p->bo_list;
154 			   __entry->ring = to_amdgpu_ring(p->entity->rq->sched)->idx;
155 			   __entry->dw = p->job->ibs[i].length_dw;
156 			   __entry->fences = amdgpu_fence_count_emitted(
157 				to_amdgpu_ring(p->entity->rq->sched));
158 			   ),
159 	    TP_printk("bo_list=%p, ring=%u, dw=%u, fences=%u",
160 		      __entry->bo_list, __entry->ring, __entry->dw,
161 		      __entry->fences)
162 );
163 
164 TRACE_EVENT(amdgpu_cs_ioctl,
165 	    TP_PROTO(struct amdgpu_job *job),
166 	    TP_ARGS(job),
167 	    TP_STRUCT__entry(
168 			     __field(uint64_t, sched_job_id)
169 			     __string(timeline, AMDGPU_JOB_GET_TIMELINE_NAME(job))
170 			     __field(unsigned int, context)
171 			     __field(unsigned int, seqno)
172 			     __field(struct dma_fence *, fence)
173 			     __string(ring, to_amdgpu_ring(job->base.sched)->name)
174 			     __field(u32, num_ibs)
175 			     ),
176 
177 	    TP_fast_assign(
178 			   __entry->sched_job_id = job->base.id;
179 			   __assign_str(timeline, AMDGPU_JOB_GET_TIMELINE_NAME(job))
180 			   __entry->context = job->base.s_fence->finished.context;
181 			   __entry->seqno = job->base.s_fence->finished.seqno;
182 			   __assign_str(ring, to_amdgpu_ring(job->base.sched)->name)
183 			   __entry->num_ibs = job->num_ibs;
184 			   ),
185 	    TP_printk("sched_job=%llu, timeline=%s, context=%u, seqno=%u, ring_name=%s, num_ibs=%u",
186 		      __entry->sched_job_id, __get_str(timeline), __entry->context,
187 		      __entry->seqno, __get_str(ring), __entry->num_ibs)
188 );
189 
190 TRACE_EVENT(amdgpu_sched_run_job,
191 	    TP_PROTO(struct amdgpu_job *job),
192 	    TP_ARGS(job),
193 	    TP_STRUCT__entry(
194 			     __field(uint64_t, sched_job_id)
195 			     __string(timeline, AMDGPU_JOB_GET_TIMELINE_NAME(job))
196 			     __field(unsigned int, context)
197 			     __field(unsigned int, seqno)
198 			     __string(ring, to_amdgpu_ring(job->base.sched)->name)
199 			     __field(u32, num_ibs)
200 			     ),
201 
202 	    TP_fast_assign(
203 			   __entry->sched_job_id = job->base.id;
204 			   __assign_str(timeline, AMDGPU_JOB_GET_TIMELINE_NAME(job))
205 			   __entry->context = job->base.s_fence->finished.context;
206 			   __entry->seqno = job->base.s_fence->finished.seqno;
207 			   __assign_str(ring, to_amdgpu_ring(job->base.sched)->name)
208 			   __entry->num_ibs = job->num_ibs;
209 			   ),
210 	    TP_printk("sched_job=%llu, timeline=%s, context=%u, seqno=%u, ring_name=%s, num_ibs=%u",
211 		      __entry->sched_job_id, __get_str(timeline), __entry->context,
212 		      __entry->seqno, __get_str(ring), __entry->num_ibs)
213 );
214 
215 
216 TRACE_EVENT(amdgpu_vm_grab_id,
217 	    TP_PROTO(struct amdgpu_vm *vm, struct amdgpu_ring *ring,
218 		     struct amdgpu_job *job),
219 	    TP_ARGS(vm, ring, job),
220 	    TP_STRUCT__entry(
221 			     __field(u32, pasid)
222 			     __string(ring, ring->name)
223 			     __field(u32, ring)
224 			     __field(u32, vmid)
225 			     __field(u32, vm_hub)
226 			     __field(u64, pd_addr)
227 			     __field(u32, needs_flush)
228 			     ),
229 
230 	    TP_fast_assign(
231 			   __entry->pasid = vm->pasid;
232 			   __assign_str(ring, ring->name)
233 			   __entry->vmid = job->vmid;
234 			   __entry->vm_hub = ring->funcs->vmhub,
235 			   __entry->pd_addr = job->vm_pd_addr;
236 			   __entry->needs_flush = job->vm_needs_flush;
237 			   ),
238 	    TP_printk("pasid=%d, ring=%s, id=%u, hub=%u, pd_addr=%010Lx needs_flush=%u",
239 		      __entry->pasid, __get_str(ring), __entry->vmid,
240 		      __entry->vm_hub, __entry->pd_addr, __entry->needs_flush)
241 );
242 
243 TRACE_EVENT(amdgpu_vm_bo_map,
244 	    TP_PROTO(struct amdgpu_bo_va *bo_va,
245 		     struct amdgpu_bo_va_mapping *mapping),
246 	    TP_ARGS(bo_va, mapping),
247 	    TP_STRUCT__entry(
248 			     __field(struct amdgpu_bo *, bo)
249 			     __field(long, start)
250 			     __field(long, last)
251 			     __field(u64, offset)
252 			     __field(u64, flags)
253 			     ),
254 
255 	    TP_fast_assign(
256 			   __entry->bo = bo_va ? bo_va->base.bo : NULL;
257 			   __entry->start = mapping->start;
258 			   __entry->last = mapping->last;
259 			   __entry->offset = mapping->offset;
260 			   __entry->flags = mapping->flags;
261 			   ),
262 	    TP_printk("bo=%p, start=%lx, last=%lx, offset=%010llx, flags=%llx",
263 		      __entry->bo, __entry->start, __entry->last,
264 		      __entry->offset, __entry->flags)
265 );
266 
267 TRACE_EVENT(amdgpu_vm_bo_unmap,
268 	    TP_PROTO(struct amdgpu_bo_va *bo_va,
269 		     struct amdgpu_bo_va_mapping *mapping),
270 	    TP_ARGS(bo_va, mapping),
271 	    TP_STRUCT__entry(
272 			     __field(struct amdgpu_bo *, bo)
273 			     __field(long, start)
274 			     __field(long, last)
275 			     __field(u64, offset)
276 			     __field(u64, flags)
277 			     ),
278 
279 	    TP_fast_assign(
280 			   __entry->bo = bo_va ? bo_va->base.bo : NULL;
281 			   __entry->start = mapping->start;
282 			   __entry->last = mapping->last;
283 			   __entry->offset = mapping->offset;
284 			   __entry->flags = mapping->flags;
285 			   ),
286 	    TP_printk("bo=%p, start=%lx, last=%lx, offset=%010llx, flags=%llx",
287 		      __entry->bo, __entry->start, __entry->last,
288 		      __entry->offset, __entry->flags)
289 );
290 
291 DECLARE_EVENT_CLASS(amdgpu_vm_mapping,
292 	    TP_PROTO(struct amdgpu_bo_va_mapping *mapping),
293 	    TP_ARGS(mapping),
294 	    TP_STRUCT__entry(
295 			     __field(u64, soffset)
296 			     __field(u64, eoffset)
297 			     __field(u64, flags)
298 			     ),
299 
300 	    TP_fast_assign(
301 			   __entry->soffset = mapping->start;
302 			   __entry->eoffset = mapping->last + 1;
303 			   __entry->flags = mapping->flags;
304 			   ),
305 	    TP_printk("soffs=%010llx, eoffs=%010llx, flags=%llx",
306 		      __entry->soffset, __entry->eoffset, __entry->flags)
307 );
308 
309 DEFINE_EVENT(amdgpu_vm_mapping, amdgpu_vm_bo_update,
310 	    TP_PROTO(struct amdgpu_bo_va_mapping *mapping),
311 	    TP_ARGS(mapping)
312 );
313 
314 DEFINE_EVENT(amdgpu_vm_mapping, amdgpu_vm_bo_mapping,
315 	    TP_PROTO(struct amdgpu_bo_va_mapping *mapping),
316 	    TP_ARGS(mapping)
317 );
318 
319 DEFINE_EVENT(amdgpu_vm_mapping, amdgpu_vm_bo_cs,
320 	    TP_PROTO(struct amdgpu_bo_va_mapping *mapping),
321 	    TP_ARGS(mapping)
322 );
323 
324 TRACE_EVENT(amdgpu_vm_update_ptes,
325 	    TP_PROTO(struct amdgpu_vm_update_params *p,
326 		     uint64_t start, uint64_t end,
327 		     unsigned int nptes, uint64_t dst,
328 		     uint64_t incr, uint64_t flags,
329 		     pid_t pid, uint64_t vm_ctx),
330 	TP_ARGS(p, start, end, nptes, dst, incr, flags, pid, vm_ctx),
331 	TP_STRUCT__entry(
332 			 __field(u64, start)
333 			 __field(u64, end)
334 			 __field(u64, flags)
335 			 __field(unsigned int, nptes)
336 			 __field(u64, incr)
337 			 __field(pid_t, pid)
338 			 __field(u64, vm_ctx)
339 			 __dynamic_array(u64, dst, nptes)
340 	),
341 
342 	TP_fast_assign(
343 			unsigned int i;
344 
345 			__entry->start = start;
346 			__entry->end = end;
347 			__entry->flags = flags;
348 			__entry->incr = incr;
349 			__entry->nptes = nptes;
350 			__entry->pid = pid;
351 			__entry->vm_ctx = vm_ctx;
352 			for (i = 0; i < nptes; ++i) {
353 				u64 addr = p->pages_addr ? amdgpu_vm_map_gart(
354 					p->pages_addr, dst) : dst;
355 
356 				((u64 *)__get_dynamic_array(dst))[i] = addr;
357 				dst += incr;
358 			}
359 	),
360 	TP_printk("pid:%u vm_ctx:0x%llx start:0x%010llx end:0x%010llx,"
361 		  " flags:0x%llx, incr:%llu, dst:\n%s", __entry->pid,
362 		  __entry->vm_ctx, __entry->start, __entry->end,
363 		  __entry->flags, __entry->incr,  __print_array(
364 		  __get_dynamic_array(dst), __entry->nptes, 8))
365 );
366 
367 TRACE_EVENT(amdgpu_vm_set_ptes,
368 	    TP_PROTO(uint64_t pe, uint64_t addr, unsigned count,
369 		     uint32_t incr, uint64_t flags, bool direct),
370 	    TP_ARGS(pe, addr, count, incr, flags, direct),
371 	    TP_STRUCT__entry(
372 			     __field(u64, pe)
373 			     __field(u64, addr)
374 			     __field(u32, count)
375 			     __field(u32, incr)
376 			     __field(u64, flags)
377 			     __field(bool, direct)
378 			     ),
379 
380 	    TP_fast_assign(
381 			   __entry->pe = pe;
382 			   __entry->addr = addr;
383 			   __entry->count = count;
384 			   __entry->incr = incr;
385 			   __entry->flags = flags;
386 			   __entry->direct = direct;
387 			   ),
388 	    TP_printk("pe=%010Lx, addr=%010Lx, incr=%u, flags=%llx, count=%u, "
389 		      "direct=%d", __entry->pe, __entry->addr, __entry->incr,
390 		      __entry->flags, __entry->count, __entry->direct)
391 );
392 
393 TRACE_EVENT(amdgpu_vm_copy_ptes,
394 	    TP_PROTO(uint64_t pe, uint64_t src, unsigned count, bool direct),
395 	    TP_ARGS(pe, src, count, direct),
396 	    TP_STRUCT__entry(
397 			     __field(u64, pe)
398 			     __field(u64, src)
399 			     __field(u32, count)
400 			     __field(bool, direct)
401 			     ),
402 
403 	    TP_fast_assign(
404 			   __entry->pe = pe;
405 			   __entry->src = src;
406 			   __entry->count = count;
407 			   __entry->direct = direct;
408 			   ),
409 	    TP_printk("pe=%010Lx, src=%010Lx, count=%u, direct=%d",
410 		      __entry->pe, __entry->src, __entry->count,
411 		      __entry->direct)
412 );
413 
414 TRACE_EVENT(amdgpu_vm_flush,
415 	    TP_PROTO(struct amdgpu_ring *ring, unsigned vmid,
416 		     uint64_t pd_addr),
417 	    TP_ARGS(ring, vmid, pd_addr),
418 	    TP_STRUCT__entry(
419 			     __string(ring, ring->name)
420 			     __field(u32, vmid)
421 			     __field(u32, vm_hub)
422 			     __field(u64, pd_addr)
423 			     ),
424 
425 	    TP_fast_assign(
426 			   __assign_str(ring, ring->name)
427 			   __entry->vmid = vmid;
428 			   __entry->vm_hub = ring->funcs->vmhub;
429 			   __entry->pd_addr = pd_addr;
430 			   ),
431 	    TP_printk("ring=%s, id=%u, hub=%u, pd_addr=%010Lx",
432 		      __get_str(ring), __entry->vmid,
433 		      __entry->vm_hub,__entry->pd_addr)
434 );
435 
436 DECLARE_EVENT_CLASS(amdgpu_pasid,
437 	    TP_PROTO(unsigned pasid),
438 	    TP_ARGS(pasid),
439 	    TP_STRUCT__entry(
440 			     __field(unsigned, pasid)
441 			     ),
442 	    TP_fast_assign(
443 			   __entry->pasid = pasid;
444 			   ),
445 	    TP_printk("pasid=%u", __entry->pasid)
446 );
447 
448 DEFINE_EVENT(amdgpu_pasid, amdgpu_pasid_allocated,
449 	    TP_PROTO(unsigned pasid),
450 	    TP_ARGS(pasid)
451 );
452 
453 DEFINE_EVENT(amdgpu_pasid, amdgpu_pasid_freed,
454 	    TP_PROTO(unsigned pasid),
455 	    TP_ARGS(pasid)
456 );
457 
458 TRACE_EVENT(amdgpu_bo_list_set,
459 	    TP_PROTO(struct amdgpu_bo_list *list, struct amdgpu_bo *bo),
460 	    TP_ARGS(list, bo),
461 	    TP_STRUCT__entry(
462 			     __field(struct amdgpu_bo_list *, list)
463 			     __field(struct amdgpu_bo *, bo)
464 			     __field(u64, bo_size)
465 			     ),
466 
467 	    TP_fast_assign(
468 			   __entry->list = list;
469 			   __entry->bo = bo;
470 			   __entry->bo_size = amdgpu_bo_size(bo);
471 			   ),
472 	    TP_printk("list=%p, bo=%p, bo_size=%Ld",
473 		      __entry->list,
474 		      __entry->bo,
475 		      __entry->bo_size)
476 );
477 
478 TRACE_EVENT(amdgpu_cs_bo_status,
479 	    TP_PROTO(uint64_t total_bo, uint64_t total_size),
480 	    TP_ARGS(total_bo, total_size),
481 	    TP_STRUCT__entry(
482 			__field(u64, total_bo)
483 			__field(u64, total_size)
484 			),
485 
486 	    TP_fast_assign(
487 			__entry->total_bo = total_bo;
488 			__entry->total_size = total_size;
489 			),
490 	    TP_printk("total_bo_size=%Ld, total_bo_count=%Ld",
491 			__entry->total_bo, __entry->total_size)
492 );
493 
494 TRACE_EVENT(amdgpu_bo_move,
495 	    TP_PROTO(struct amdgpu_bo* bo, uint32_t new_placement, uint32_t old_placement),
496 	    TP_ARGS(bo, new_placement, old_placement),
497 	    TP_STRUCT__entry(
498 			__field(struct amdgpu_bo *, bo)
499 			__field(u64, bo_size)
500 			__field(u32, new_placement)
501 			__field(u32, old_placement)
502 			),
503 
504 	    TP_fast_assign(
505 			__entry->bo      = bo;
506 			__entry->bo_size = amdgpu_bo_size(bo);
507 			__entry->new_placement = new_placement;
508 			__entry->old_placement = old_placement;
509 			),
510 	    TP_printk("bo=%p, from=%d, to=%d, size=%Ld",
511 			__entry->bo, __entry->old_placement,
512 			__entry->new_placement, __entry->bo_size)
513 );
514 
515 TRACE_EVENT(amdgpu_ib_pipe_sync,
516 	    TP_PROTO(struct amdgpu_job *sched_job, struct dma_fence *fence),
517 	    TP_ARGS(sched_job, fence),
518 	    TP_STRUCT__entry(
519 			     __string(ring, sched_job->base.sched->name)
520 			     __field(uint64_t, id)
521 			     __field(struct dma_fence *, fence)
522 			     __field(uint64_t, ctx)
523 			     __field(unsigned, seqno)
524 			     ),
525 
526 	    TP_fast_assign(
527 			   __assign_str(ring, sched_job->base.sched->name)
528 			   __entry->id = sched_job->base.id;
529 			   __entry->fence = fence;
530 			   __entry->ctx = fence->context;
531 			   __entry->seqno = fence->seqno;
532 			   ),
533 	    TP_printk("job ring=%s, id=%llu, need pipe sync to fence=%p, context=%llu, seq=%u",
534 		      __get_str(ring), __entry->id,
535 		      __entry->fence, __entry->ctx,
536 		      __entry->seqno)
537 );
538 
539 #undef AMDGPU_JOB_GET_TIMELINE_NAME
540 #endif
541 
542 /* This part must be outside protection */
543 #undef TRACE_INCLUDE_PATH
544 #define TRACE_INCLUDE_PATH ../../drivers/gpu/drm/amd/amdgpu
545 #include <trace/define_trace.h>
546