]> git.ipfire.org Git - thirdparty/linux.git/blob - drivers/gpu/drm/i915/i915_gpu_error.c
io_uring: reset -EBUSY error when io sq thread is waken up
[thirdparty/linux.git] / drivers / gpu / drm / i915 / i915_gpu_error.c
1 /*
2 * Copyright (c) 2008 Intel Corporation
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 (including the next
12 * paragraph) shall be included in all copies or substantial portions of the
13 * Software.
14 *
15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
18 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
20 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
21 * IN THE SOFTWARE.
22 *
23 * Authors:
24 * Eric Anholt <eric@anholt.net>
25 * Keith Packard <keithp@keithp.com>
26 * Mika Kuoppala <mika.kuoppala@intel.com>
27 *
28 */
29
30 #include <linux/ascii85.h>
31 #include <linux/nmi.h>
32 #include <linux/pagevec.h>
33 #include <linux/scatterlist.h>
34 #include <linux/utsname.h>
35 #include <linux/zlib.h>
36
37 #include <drm/drm_print.h>
38
39 #include "display/intel_atomic.h"
40 #include "display/intel_csr.h"
41 #include "display/intel_overlay.h"
42
43 #include "gem/i915_gem_context.h"
44 #include "gem/i915_gem_lmem.h"
45 #include "gt/intel_gt_pm.h"
46
47 #include "i915_drv.h"
48 #include "i915_gpu_error.h"
49 #include "i915_memcpy.h"
50 #include "i915_scatterlist.h"
51
52 #define ALLOW_FAIL (GFP_KERNEL | __GFP_RETRY_MAYFAIL | __GFP_NOWARN)
53 #define ATOMIC_MAYFAIL (GFP_ATOMIC | __GFP_NOWARN)
54
55 static void __sg_set_buf(struct scatterlist *sg,
56 void *addr, unsigned int len, loff_t it)
57 {
58 sg->page_link = (unsigned long)virt_to_page(addr);
59 sg->offset = offset_in_page(addr);
60 sg->length = len;
61 sg->dma_address = it;
62 }
63
64 static bool __i915_error_grow(struct drm_i915_error_state_buf *e, size_t len)
65 {
66 if (!len)
67 return false;
68
69 if (e->bytes + len + 1 <= e->size)
70 return true;
71
72 if (e->bytes) {
73 __sg_set_buf(e->cur++, e->buf, e->bytes, e->iter);
74 e->iter += e->bytes;
75 e->buf = NULL;
76 e->bytes = 0;
77 }
78
79 if (e->cur == e->end) {
80 struct scatterlist *sgl;
81
82 sgl = (typeof(sgl))__get_free_page(ALLOW_FAIL);
83 if (!sgl) {
84 e->err = -ENOMEM;
85 return false;
86 }
87
88 if (e->cur) {
89 e->cur->offset = 0;
90 e->cur->length = 0;
91 e->cur->page_link =
92 (unsigned long)sgl | SG_CHAIN;
93 } else {
94 e->sgl = sgl;
95 }
96
97 e->cur = sgl;
98 e->end = sgl + SG_MAX_SINGLE_ALLOC - 1;
99 }
100
101 e->size = ALIGN(len + 1, SZ_64K);
102 e->buf = kmalloc(e->size, ALLOW_FAIL);
103 if (!e->buf) {
104 e->size = PAGE_ALIGN(len + 1);
105 e->buf = kmalloc(e->size, GFP_KERNEL);
106 }
107 if (!e->buf) {
108 e->err = -ENOMEM;
109 return false;
110 }
111
112 return true;
113 }
114
115 __printf(2, 0)
116 static void i915_error_vprintf(struct drm_i915_error_state_buf *e,
117 const char *fmt, va_list args)
118 {
119 va_list ap;
120 int len;
121
122 if (e->err)
123 return;
124
125 va_copy(ap, args);
126 len = vsnprintf(NULL, 0, fmt, ap);
127 va_end(ap);
128 if (len <= 0) {
129 e->err = len;
130 return;
131 }
132
133 if (!__i915_error_grow(e, len))
134 return;
135
136 GEM_BUG_ON(e->bytes >= e->size);
137 len = vscnprintf(e->buf + e->bytes, e->size - e->bytes, fmt, args);
138 if (len < 0) {
139 e->err = len;
140 return;
141 }
142 e->bytes += len;
143 }
144
145 static void i915_error_puts(struct drm_i915_error_state_buf *e, const char *str)
146 {
147 unsigned len;
148
149 if (e->err || !str)
150 return;
151
152 len = strlen(str);
153 if (!__i915_error_grow(e, len))
154 return;
155
156 GEM_BUG_ON(e->bytes + len > e->size);
157 memcpy(e->buf + e->bytes, str, len);
158 e->bytes += len;
159 }
160
161 #define err_printf(e, ...) i915_error_printf(e, __VA_ARGS__)
162 #define err_puts(e, s) i915_error_puts(e, s)
163
164 static void __i915_printfn_error(struct drm_printer *p, struct va_format *vaf)
165 {
166 i915_error_vprintf(p->arg, vaf->fmt, *vaf->va);
167 }
168
169 static inline struct drm_printer
170 i915_error_printer(struct drm_i915_error_state_buf *e)
171 {
172 struct drm_printer p = {
173 .printfn = __i915_printfn_error,
174 .arg = e,
175 };
176 return p;
177 }
178
179 /* single threaded page allocator with a reserved stash for emergencies */
180 static void pool_fini(struct pagevec *pv)
181 {
182 pagevec_release(pv);
183 }
184
185 static int pool_refill(struct pagevec *pv, gfp_t gfp)
186 {
187 while (pagevec_space(pv)) {
188 struct page *p;
189
190 p = alloc_page(gfp);
191 if (!p)
192 return -ENOMEM;
193
194 pagevec_add(pv, p);
195 }
196
197 return 0;
198 }
199
200 static int pool_init(struct pagevec *pv, gfp_t gfp)
201 {
202 int err;
203
204 pagevec_init(pv);
205
206 err = pool_refill(pv, gfp);
207 if (err)
208 pool_fini(pv);
209
210 return err;
211 }
212
213 static void *pool_alloc(struct pagevec *pv, gfp_t gfp)
214 {
215 struct page *p;
216
217 p = alloc_page(gfp);
218 if (!p && pagevec_count(pv))
219 p = pv->pages[--pv->nr];
220
221 return p ? page_address(p) : NULL;
222 }
223
224 static void pool_free(struct pagevec *pv, void *addr)
225 {
226 struct page *p = virt_to_page(addr);
227
228 if (pagevec_space(pv))
229 pagevec_add(pv, p);
230 else
231 __free_page(p);
232 }
233
234 #ifdef CONFIG_DRM_I915_COMPRESS_ERROR
235
236 struct i915_vma_compress {
237 struct pagevec pool;
238 struct z_stream_s zstream;
239 void *tmp;
240 };
241
242 static bool compress_init(struct i915_vma_compress *c)
243 {
244 struct z_stream_s *zstream = &c->zstream;
245
246 if (pool_init(&c->pool, ALLOW_FAIL))
247 return false;
248
249 zstream->workspace =
250 kmalloc(zlib_deflate_workspacesize(MAX_WBITS, MAX_MEM_LEVEL),
251 ALLOW_FAIL);
252 if (!zstream->workspace) {
253 pool_fini(&c->pool);
254 return false;
255 }
256
257 c->tmp = NULL;
258 if (i915_has_memcpy_from_wc())
259 c->tmp = pool_alloc(&c->pool, ALLOW_FAIL);
260
261 return true;
262 }
263
264 static bool compress_start(struct i915_vma_compress *c)
265 {
266 struct z_stream_s *zstream = &c->zstream;
267 void *workspace = zstream->workspace;
268
269 memset(zstream, 0, sizeof(*zstream));
270 zstream->workspace = workspace;
271
272 return zlib_deflateInit(zstream, Z_DEFAULT_COMPRESSION) == Z_OK;
273 }
274
275 static void *compress_next_page(struct i915_vma_compress *c,
276 struct i915_vma_coredump *dst)
277 {
278 void *page;
279
280 if (dst->page_count >= dst->num_pages)
281 return ERR_PTR(-ENOSPC);
282
283 page = pool_alloc(&c->pool, ALLOW_FAIL);
284 if (!page)
285 return ERR_PTR(-ENOMEM);
286
287 return dst->pages[dst->page_count++] = page;
288 }
289
290 static int compress_page(struct i915_vma_compress *c,
291 void *src,
292 struct i915_vma_coredump *dst,
293 bool wc)
294 {
295 struct z_stream_s *zstream = &c->zstream;
296
297 zstream->next_in = src;
298 if (wc && c->tmp && i915_memcpy_from_wc(c->tmp, src, PAGE_SIZE))
299 zstream->next_in = c->tmp;
300 zstream->avail_in = PAGE_SIZE;
301
302 do {
303 if (zstream->avail_out == 0) {
304 zstream->next_out = compress_next_page(c, dst);
305 if (IS_ERR(zstream->next_out))
306 return PTR_ERR(zstream->next_out);
307
308 zstream->avail_out = PAGE_SIZE;
309 }
310
311 if (zlib_deflate(zstream, Z_NO_FLUSH) != Z_OK)
312 return -EIO;
313 } while (zstream->avail_in);
314
315 /* Fallback to uncompressed if we increase size? */
316 if (0 && zstream->total_out > zstream->total_in)
317 return -E2BIG;
318
319 return 0;
320 }
321
322 static int compress_flush(struct i915_vma_compress *c,
323 struct i915_vma_coredump *dst)
324 {
325 struct z_stream_s *zstream = &c->zstream;
326
327 do {
328 switch (zlib_deflate(zstream, Z_FINISH)) {
329 case Z_OK: /* more space requested */
330 zstream->next_out = compress_next_page(c, dst);
331 if (IS_ERR(zstream->next_out))
332 return PTR_ERR(zstream->next_out);
333
334 zstream->avail_out = PAGE_SIZE;
335 break;
336
337 case Z_STREAM_END:
338 goto end;
339
340 default: /* any error */
341 return -EIO;
342 }
343 } while (1);
344
345 end:
346 memset(zstream->next_out, 0, zstream->avail_out);
347 dst->unused = zstream->avail_out;
348 return 0;
349 }
350
351 static void compress_finish(struct i915_vma_compress *c)
352 {
353 zlib_deflateEnd(&c->zstream);
354 }
355
356 static void compress_fini(struct i915_vma_compress *c)
357 {
358 kfree(c->zstream.workspace);
359 if (c->tmp)
360 pool_free(&c->pool, c->tmp);
361 pool_fini(&c->pool);
362 }
363
364 static void err_compression_marker(struct drm_i915_error_state_buf *m)
365 {
366 err_puts(m, ":");
367 }
368
369 #else
370
371 struct i915_vma_compress {
372 struct pagevec pool;
373 };
374
375 static bool compress_init(struct i915_vma_compress *c)
376 {
377 return pool_init(&c->pool, ALLOW_FAIL) == 0;
378 }
379
380 static bool compress_start(struct i915_vma_compress *c)
381 {
382 return true;
383 }
384
385 static int compress_page(struct i915_vma_compress *c,
386 void *src,
387 struct i915_vma_coredump *dst,
388 bool wc)
389 {
390 void *ptr;
391
392 ptr = pool_alloc(&c->pool, ALLOW_FAIL);
393 if (!ptr)
394 return -ENOMEM;
395
396 if (!(wc && i915_memcpy_from_wc(ptr, src, PAGE_SIZE)))
397 memcpy(ptr, src, PAGE_SIZE);
398 dst->pages[dst->page_count++] = ptr;
399
400 return 0;
401 }
402
403 static int compress_flush(struct i915_vma_compress *c,
404 struct i915_vma_coredump *dst)
405 {
406 return 0;
407 }
408
409 static void compress_finish(struct i915_vma_compress *c)
410 {
411 }
412
413 static void compress_fini(struct i915_vma_compress *c)
414 {
415 pool_fini(&c->pool);
416 }
417
418 static void err_compression_marker(struct drm_i915_error_state_buf *m)
419 {
420 err_puts(m, "~");
421 }
422
423 #endif
424
425 static void error_print_instdone(struct drm_i915_error_state_buf *m,
426 const struct intel_engine_coredump *ee)
427 {
428 const struct sseu_dev_info *sseu = &RUNTIME_INFO(m->i915)->sseu;
429 int slice;
430 int subslice;
431
432 err_printf(m, " INSTDONE: 0x%08x\n",
433 ee->instdone.instdone);
434
435 if (ee->engine->class != RENDER_CLASS || INTEL_GEN(m->i915) <= 3)
436 return;
437
438 err_printf(m, " SC_INSTDONE: 0x%08x\n",
439 ee->instdone.slice_common);
440
441 if (INTEL_GEN(m->i915) <= 6)
442 return;
443
444 for_each_instdone_slice_subslice(m->i915, sseu, slice, subslice)
445 err_printf(m, " SAMPLER_INSTDONE[%d][%d]: 0x%08x\n",
446 slice, subslice,
447 ee->instdone.sampler[slice][subslice]);
448
449 for_each_instdone_slice_subslice(m->i915, sseu, slice, subslice)
450 err_printf(m, " ROW_INSTDONE[%d][%d]: 0x%08x\n",
451 slice, subslice,
452 ee->instdone.row[slice][subslice]);
453
454 if (INTEL_GEN(m->i915) < 12)
455 return;
456
457 err_printf(m, " SC_INSTDONE_EXTRA: 0x%08x\n",
458 ee->instdone.slice_common_extra[0]);
459 err_printf(m, " SC_INSTDONE_EXTRA2: 0x%08x\n",
460 ee->instdone.slice_common_extra[1]);
461 }
462
463 static void error_print_request(struct drm_i915_error_state_buf *m,
464 const char *prefix,
465 const struct i915_request_coredump *erq)
466 {
467 if (!erq->seqno)
468 return;
469
470 err_printf(m, "%s pid %d, seqno %8x:%08x%s%s, prio %d, start %08x, head %08x, tail %08x\n",
471 prefix, erq->pid, erq->context, erq->seqno,
472 test_bit(DMA_FENCE_FLAG_SIGNALED_BIT,
473 &erq->flags) ? "!" : "",
474 test_bit(DMA_FENCE_FLAG_ENABLE_SIGNAL_BIT,
475 &erq->flags) ? "+" : "",
476 erq->sched_attr.priority,
477 erq->start, erq->head, erq->tail);
478 }
479
480 static void error_print_context(struct drm_i915_error_state_buf *m,
481 const char *header,
482 const struct i915_gem_context_coredump *ctx)
483 {
484 const u32 period = RUNTIME_INFO(m->i915)->cs_timestamp_period_ns;
485
486 err_printf(m, "%s%s[%d] prio %d, guilty %d active %d, runtime total %lluns, avg %lluns\n",
487 header, ctx->comm, ctx->pid, ctx->sched_attr.priority,
488 ctx->guilty, ctx->active,
489 ctx->total_runtime * period,
490 mul_u32_u32(ctx->avg_runtime, period));
491 }
492
493 static struct i915_vma_coredump *
494 __find_vma(struct i915_vma_coredump *vma, const char *name)
495 {
496 while (vma) {
497 if (strcmp(vma->name, name) == 0)
498 return vma;
499 vma = vma->next;
500 }
501
502 return NULL;
503 }
504
505 static struct i915_vma_coredump *
506 find_batch(const struct intel_engine_coredump *ee)
507 {
508 return __find_vma(ee->vma, "batch");
509 }
510
511 static void error_print_engine(struct drm_i915_error_state_buf *m,
512 const struct intel_engine_coredump *ee)
513 {
514 struct i915_vma_coredump *batch;
515 int n;
516
517 err_printf(m, "%s command stream:\n", ee->engine->name);
518 err_printf(m, " CCID: 0x%08x\n", ee->ccid);
519 err_printf(m, " START: 0x%08x\n", ee->start);
520 err_printf(m, " HEAD: 0x%08x [0x%08x]\n", ee->head, ee->rq_head);
521 err_printf(m, " TAIL: 0x%08x [0x%08x, 0x%08x]\n",
522 ee->tail, ee->rq_post, ee->rq_tail);
523 err_printf(m, " CTL: 0x%08x\n", ee->ctl);
524 err_printf(m, " MODE: 0x%08x\n", ee->mode);
525 err_printf(m, " HWS: 0x%08x\n", ee->hws);
526 err_printf(m, " ACTHD: 0x%08x %08x\n",
527 (u32)(ee->acthd>>32), (u32)ee->acthd);
528 err_printf(m, " IPEIR: 0x%08x\n", ee->ipeir);
529 err_printf(m, " IPEHR: 0x%08x\n", ee->ipehr);
530 err_printf(m, " ESR: 0x%08x\n", ee->esr);
531
532 error_print_instdone(m, ee);
533
534 batch = find_batch(ee);
535 if (batch) {
536 u64 start = batch->gtt_offset;
537 u64 end = start + batch->gtt_size;
538
539 err_printf(m, " batch: [0x%08x_%08x, 0x%08x_%08x]\n",
540 upper_32_bits(start), lower_32_bits(start),
541 upper_32_bits(end), lower_32_bits(end));
542 }
543 if (INTEL_GEN(m->i915) >= 4) {
544 err_printf(m, " BBADDR: 0x%08x_%08x\n",
545 (u32)(ee->bbaddr>>32), (u32)ee->bbaddr);
546 err_printf(m, " BB_STATE: 0x%08x\n", ee->bbstate);
547 err_printf(m, " INSTPS: 0x%08x\n", ee->instps);
548 }
549 err_printf(m, " INSTPM: 0x%08x\n", ee->instpm);
550 err_printf(m, " FADDR: 0x%08x %08x\n", upper_32_bits(ee->faddr),
551 lower_32_bits(ee->faddr));
552 if (INTEL_GEN(m->i915) >= 6) {
553 err_printf(m, " RC PSMI: 0x%08x\n", ee->rc_psmi);
554 err_printf(m, " FAULT_REG: 0x%08x\n", ee->fault_reg);
555 }
556 if (HAS_PPGTT(m->i915)) {
557 err_printf(m, " GFX_MODE: 0x%08x\n", ee->vm_info.gfx_mode);
558
559 if (INTEL_GEN(m->i915) >= 8) {
560 int i;
561 for (i = 0; i < 4; i++)
562 err_printf(m, " PDP%d: 0x%016llx\n",
563 i, ee->vm_info.pdp[i]);
564 } else {
565 err_printf(m, " PP_DIR_BASE: 0x%08x\n",
566 ee->vm_info.pp_dir_base);
567 }
568 }
569 err_printf(m, " engine reset count: %u\n", ee->reset_count);
570
571 for (n = 0; n < ee->num_ports; n++) {
572 err_printf(m, " ELSP[%d]:", n);
573 error_print_request(m, " ", &ee->execlist[n]);
574 }
575
576 error_print_context(m, " Active context: ", &ee->context);
577 }
578
579 void i915_error_printf(struct drm_i915_error_state_buf *e, const char *f, ...)
580 {
581 va_list args;
582
583 va_start(args, f);
584 i915_error_vprintf(e, f, args);
585 va_end(args);
586 }
587
588 static void print_error_vma(struct drm_i915_error_state_buf *m,
589 const struct intel_engine_cs *engine,
590 const struct i915_vma_coredump *vma)
591 {
592 char out[ASCII85_BUFSZ];
593 int page;
594
595 if (!vma)
596 return;
597
598 err_printf(m, "%s --- %s = 0x%08x %08x\n",
599 engine ? engine->name : "global", vma->name,
600 upper_32_bits(vma->gtt_offset),
601 lower_32_bits(vma->gtt_offset));
602
603 if (vma->gtt_page_sizes > I915_GTT_PAGE_SIZE_4K)
604 err_printf(m, "gtt_page_sizes = 0x%08x\n", vma->gtt_page_sizes);
605
606 err_compression_marker(m);
607 for (page = 0; page < vma->page_count; page++) {
608 int i, len;
609
610 len = PAGE_SIZE;
611 if (page == vma->page_count - 1)
612 len -= vma->unused;
613 len = ascii85_encode_len(len);
614
615 for (i = 0; i < len; i++)
616 err_puts(m, ascii85_encode(vma->pages[page][i], out));
617 }
618 err_puts(m, "\n");
619 }
620
621 static void err_print_capabilities(struct drm_i915_error_state_buf *m,
622 const struct intel_device_info *info,
623 const struct intel_runtime_info *runtime,
624 const struct intel_driver_caps *caps)
625 {
626 struct drm_printer p = i915_error_printer(m);
627
628 intel_device_info_print_static(info, &p);
629 intel_device_info_print_runtime(runtime, &p);
630 intel_device_info_print_topology(&runtime->sseu, &p);
631 intel_driver_caps_print(caps, &p);
632 }
633
634 static void err_print_params(struct drm_i915_error_state_buf *m,
635 const struct i915_params *params)
636 {
637 struct drm_printer p = i915_error_printer(m);
638
639 i915_params_dump(params, &p);
640 }
641
642 static void err_print_pciid(struct drm_i915_error_state_buf *m,
643 struct drm_i915_private *i915)
644 {
645 struct pci_dev *pdev = i915->drm.pdev;
646
647 err_printf(m, "PCI ID: 0x%04x\n", pdev->device);
648 err_printf(m, "PCI Revision: 0x%02x\n", pdev->revision);
649 err_printf(m, "PCI Subsystem: %04x:%04x\n",
650 pdev->subsystem_vendor,
651 pdev->subsystem_device);
652 }
653
654 static void err_print_uc(struct drm_i915_error_state_buf *m,
655 const struct intel_uc_coredump *error_uc)
656 {
657 struct drm_printer p = i915_error_printer(m);
658
659 intel_uc_fw_dump(&error_uc->guc_fw, &p);
660 intel_uc_fw_dump(&error_uc->huc_fw, &p);
661 print_error_vma(m, NULL, error_uc->guc_log);
662 }
663
664 static void err_free_sgl(struct scatterlist *sgl)
665 {
666 while (sgl) {
667 struct scatterlist *sg;
668
669 for (sg = sgl; !sg_is_chain(sg); sg++) {
670 kfree(sg_virt(sg));
671 if (sg_is_last(sg))
672 break;
673 }
674
675 sg = sg_is_last(sg) ? NULL : sg_chain_ptr(sg);
676 free_page((unsigned long)sgl);
677 sgl = sg;
678 }
679 }
680
681 static void err_print_gt(struct drm_i915_error_state_buf *m,
682 struct intel_gt_coredump *gt)
683 {
684 const struct intel_engine_coredump *ee;
685 int i;
686
687 err_printf(m, "GT awake: %s\n", yesno(gt->awake));
688 err_printf(m, "EIR: 0x%08x\n", gt->eir);
689 err_printf(m, "IER: 0x%08x\n", gt->ier);
690 for (i = 0; i < gt->ngtier; i++)
691 err_printf(m, "GTIER[%d]: 0x%08x\n", i, gt->gtier[i]);
692 err_printf(m, "PGTBL_ER: 0x%08x\n", gt->pgtbl_er);
693 err_printf(m, "FORCEWAKE: 0x%08x\n", gt->forcewake);
694 err_printf(m, "DERRMR: 0x%08x\n", gt->derrmr);
695
696 for (i = 0; i < gt->nfence; i++)
697 err_printf(m, " fence[%d] = %08llx\n", i, gt->fence[i]);
698
699 if (IS_GEN_RANGE(m->i915, 6, 11)) {
700 err_printf(m, "ERROR: 0x%08x\n", gt->error);
701 err_printf(m, "DONE_REG: 0x%08x\n", gt->done_reg);
702 }
703
704 if (INTEL_GEN(m->i915) >= 8)
705 err_printf(m, "FAULT_TLB_DATA: 0x%08x 0x%08x\n",
706 gt->fault_data1, gt->fault_data0);
707
708 if (IS_GEN(m->i915, 7))
709 err_printf(m, "ERR_INT: 0x%08x\n", gt->err_int);
710
711 if (IS_GEN_RANGE(m->i915, 8, 11))
712 err_printf(m, "GTT_CACHE_EN: 0x%08x\n", gt->gtt_cache);
713
714 if (IS_GEN(m->i915, 12))
715 err_printf(m, "AUX_ERR_DBG: 0x%08x\n", gt->aux_err);
716
717 if (INTEL_GEN(m->i915) >= 12) {
718 int i;
719
720 for (i = 0; i < GEN12_SFC_DONE_MAX; i++)
721 err_printf(m, " SFC_DONE[%d]: 0x%08x\n", i,
722 gt->sfc_done[i]);
723
724 err_printf(m, " GAM_DONE: 0x%08x\n", gt->gam_done);
725 }
726
727 for (ee = gt->engine; ee; ee = ee->next) {
728 const struct i915_vma_coredump *vma;
729
730 error_print_engine(m, ee);
731 for (vma = ee->vma; vma; vma = vma->next)
732 print_error_vma(m, ee->engine, vma);
733 }
734
735 if (gt->uc)
736 err_print_uc(m, gt->uc);
737 }
738
739 static void __err_print_to_sgl(struct drm_i915_error_state_buf *m,
740 struct i915_gpu_coredump *error)
741 {
742 const struct intel_engine_coredump *ee;
743 struct timespec64 ts;
744
745 if (*error->error_msg)
746 err_printf(m, "%s\n", error->error_msg);
747 err_printf(m, "Kernel: %s %s\n",
748 init_utsname()->release,
749 init_utsname()->machine);
750 err_printf(m, "Driver: %s\n", DRIVER_DATE);
751 ts = ktime_to_timespec64(error->time);
752 err_printf(m, "Time: %lld s %ld us\n",
753 (s64)ts.tv_sec, ts.tv_nsec / NSEC_PER_USEC);
754 ts = ktime_to_timespec64(error->boottime);
755 err_printf(m, "Boottime: %lld s %ld us\n",
756 (s64)ts.tv_sec, ts.tv_nsec / NSEC_PER_USEC);
757 ts = ktime_to_timespec64(error->uptime);
758 err_printf(m, "Uptime: %lld s %ld us\n",
759 (s64)ts.tv_sec, ts.tv_nsec / NSEC_PER_USEC);
760 err_printf(m, "Capture: %lu jiffies; %d ms ago\n",
761 error->capture, jiffies_to_msecs(jiffies - error->capture));
762
763 for (ee = error->gt ? error->gt->engine : NULL; ee; ee = ee->next)
764 err_printf(m, "Active process (on ring %s): %s [%d]\n",
765 ee->engine->name,
766 ee->context.comm,
767 ee->context.pid);
768
769 err_printf(m, "Reset count: %u\n", error->reset_count);
770 err_printf(m, "Suspend count: %u\n", error->suspend_count);
771 err_printf(m, "Platform: %s\n", intel_platform_name(error->device_info.platform));
772 err_printf(m, "Subplatform: 0x%x\n",
773 intel_subplatform(&error->runtime_info,
774 error->device_info.platform));
775 err_print_pciid(m, m->i915);
776
777 err_printf(m, "IOMMU enabled?: %d\n", error->iommu);
778
779 if (HAS_CSR(m->i915)) {
780 struct intel_csr *csr = &m->i915->csr;
781
782 err_printf(m, "DMC loaded: %s\n",
783 yesno(csr->dmc_payload != NULL));
784 err_printf(m, "DMC fw version: %d.%d\n",
785 CSR_VERSION_MAJOR(csr->version),
786 CSR_VERSION_MINOR(csr->version));
787 }
788
789 err_printf(m, "RPM wakelock: %s\n", yesno(error->wakelock));
790 err_printf(m, "PM suspended: %s\n", yesno(error->suspended));
791
792 if (error->gt)
793 err_print_gt(m, error->gt);
794
795 if (error->overlay)
796 intel_overlay_print_error_state(m, error->overlay);
797
798 if (error->display)
799 intel_display_print_error_state(m, error->display);
800
801 err_print_capabilities(m, &error->device_info, &error->runtime_info,
802 &error->driver_caps);
803 err_print_params(m, &error->params);
804 }
805
806 static int err_print_to_sgl(struct i915_gpu_coredump *error)
807 {
808 struct drm_i915_error_state_buf m;
809
810 if (IS_ERR(error))
811 return PTR_ERR(error);
812
813 if (READ_ONCE(error->sgl))
814 return 0;
815
816 memset(&m, 0, sizeof(m));
817 m.i915 = error->i915;
818
819 __err_print_to_sgl(&m, error);
820
821 if (m.buf) {
822 __sg_set_buf(m.cur++, m.buf, m.bytes, m.iter);
823 m.bytes = 0;
824 m.buf = NULL;
825 }
826 if (m.cur) {
827 GEM_BUG_ON(m.end < m.cur);
828 sg_mark_end(m.cur - 1);
829 }
830 GEM_BUG_ON(m.sgl && !m.cur);
831
832 if (m.err) {
833 err_free_sgl(m.sgl);
834 return m.err;
835 }
836
837 if (cmpxchg(&error->sgl, NULL, m.sgl))
838 err_free_sgl(m.sgl);
839
840 return 0;
841 }
842
843 ssize_t i915_gpu_coredump_copy_to_buffer(struct i915_gpu_coredump *error,
844 char *buf, loff_t off, size_t rem)
845 {
846 struct scatterlist *sg;
847 size_t count;
848 loff_t pos;
849 int err;
850
851 if (!error || !rem)
852 return 0;
853
854 err = err_print_to_sgl(error);
855 if (err)
856 return err;
857
858 sg = READ_ONCE(error->fit);
859 if (!sg || off < sg->dma_address)
860 sg = error->sgl;
861 if (!sg)
862 return 0;
863
864 pos = sg->dma_address;
865 count = 0;
866 do {
867 size_t len, start;
868
869 if (sg_is_chain(sg)) {
870 sg = sg_chain_ptr(sg);
871 GEM_BUG_ON(sg_is_chain(sg));
872 }
873
874 len = sg->length;
875 if (pos + len <= off) {
876 pos += len;
877 continue;
878 }
879
880 start = sg->offset;
881 if (pos < off) {
882 GEM_BUG_ON(off - pos > len);
883 len -= off - pos;
884 start += off - pos;
885 pos = off;
886 }
887
888 len = min(len, rem);
889 GEM_BUG_ON(!len || len > sg->length);
890
891 memcpy(buf, page_address(sg_page(sg)) + start, len);
892
893 count += len;
894 pos += len;
895
896 buf += len;
897 rem -= len;
898 if (!rem) {
899 WRITE_ONCE(error->fit, sg);
900 break;
901 }
902 } while (!sg_is_last(sg++));
903
904 return count;
905 }
906
907 static void i915_vma_coredump_free(struct i915_vma_coredump *vma)
908 {
909 while (vma) {
910 struct i915_vma_coredump *next = vma->next;
911 int page;
912
913 for (page = 0; page < vma->page_count; page++)
914 free_page((unsigned long)vma->pages[page]);
915
916 kfree(vma);
917 vma = next;
918 }
919 }
920
921 static void cleanup_params(struct i915_gpu_coredump *error)
922 {
923 i915_params_free(&error->params);
924 }
925
926 static void cleanup_uc(struct intel_uc_coredump *uc)
927 {
928 kfree(uc->guc_fw.path);
929 kfree(uc->huc_fw.path);
930 i915_vma_coredump_free(uc->guc_log);
931
932 kfree(uc);
933 }
934
935 static void cleanup_gt(struct intel_gt_coredump *gt)
936 {
937 while (gt->engine) {
938 struct intel_engine_coredump *ee = gt->engine;
939
940 gt->engine = ee->next;
941
942 i915_vma_coredump_free(ee->vma);
943 kfree(ee);
944 }
945
946 if (gt->uc)
947 cleanup_uc(gt->uc);
948
949 kfree(gt);
950 }
951
952 void __i915_gpu_coredump_free(struct kref *error_ref)
953 {
954 struct i915_gpu_coredump *error =
955 container_of(error_ref, typeof(*error), ref);
956
957 while (error->gt) {
958 struct intel_gt_coredump *gt = error->gt;
959
960 error->gt = gt->next;
961 cleanup_gt(gt);
962 }
963
964 kfree(error->overlay);
965 kfree(error->display);
966
967 cleanup_params(error);
968
969 err_free_sgl(error->sgl);
970 kfree(error);
971 }
972
973 static struct i915_vma_coredump *
974 i915_vma_coredump_create(const struct intel_gt *gt,
975 const struct i915_vma *vma,
976 const char *name,
977 struct i915_vma_compress *compress)
978 {
979 struct i915_ggtt *ggtt = gt->ggtt;
980 const u64 slot = ggtt->error_capture.start;
981 struct i915_vma_coredump *dst;
982 unsigned long num_pages;
983 struct sgt_iter iter;
984 int ret;
985
986 might_sleep();
987
988 if (!vma || !vma->pages || !compress)
989 return NULL;
990
991 num_pages = min_t(u64, vma->size, vma->obj->base.size) >> PAGE_SHIFT;
992 num_pages = DIV_ROUND_UP(10 * num_pages, 8); /* worstcase zlib growth */
993 dst = kmalloc(sizeof(*dst) + num_pages * sizeof(u32 *), ALLOW_FAIL);
994 if (!dst)
995 return NULL;
996
997 if (!compress_start(compress)) {
998 kfree(dst);
999 return NULL;
1000 }
1001
1002 strcpy(dst->name, name);
1003 dst->next = NULL;
1004
1005 dst->gtt_offset = vma->node.start;
1006 dst->gtt_size = vma->node.size;
1007 dst->gtt_page_sizes = vma->page_sizes.gtt;
1008 dst->num_pages = num_pages;
1009 dst->page_count = 0;
1010 dst->unused = 0;
1011
1012 ret = -EINVAL;
1013 if (drm_mm_node_allocated(&ggtt->error_capture)) {
1014 void __iomem *s;
1015 dma_addr_t dma;
1016
1017 for_each_sgt_daddr(dma, iter, vma->pages) {
1018 ggtt->vm.insert_page(&ggtt->vm, dma, slot,
1019 I915_CACHE_NONE, 0);
1020 mb();
1021
1022 s = io_mapping_map_wc(&ggtt->iomap, slot, PAGE_SIZE);
1023 ret = compress_page(compress,
1024 (void __force *)s, dst,
1025 true);
1026 io_mapping_unmap(s);
1027 if (ret)
1028 break;
1029 }
1030 } else if (i915_gem_object_is_lmem(vma->obj)) {
1031 struct intel_memory_region *mem = vma->obj->mm.region;
1032 dma_addr_t dma;
1033
1034 for_each_sgt_daddr(dma, iter, vma->pages) {
1035 void __iomem *s;
1036
1037 s = io_mapping_map_wc(&mem->iomap, dma, PAGE_SIZE);
1038 ret = compress_page(compress,
1039 (void __force *)s, dst,
1040 true);
1041 io_mapping_unmap(s);
1042 if (ret)
1043 break;
1044 }
1045 } else {
1046 struct page *page;
1047
1048 for_each_sgt_page(page, iter, vma->pages) {
1049 void *s;
1050
1051 drm_clflush_pages(&page, 1);
1052
1053 s = kmap(page);
1054 ret = compress_page(compress, s, dst, false);
1055 kunmap(page);
1056
1057 drm_clflush_pages(&page, 1);
1058
1059 if (ret)
1060 break;
1061 }
1062 }
1063
1064 if (ret || compress_flush(compress, dst)) {
1065 while (dst->page_count--)
1066 pool_free(&compress->pool, dst->pages[dst->page_count]);
1067 kfree(dst);
1068 dst = NULL;
1069 }
1070 compress_finish(compress);
1071
1072 return dst;
1073 }
1074
1075 static void gt_record_fences(struct intel_gt_coredump *gt)
1076 {
1077 struct i915_ggtt *ggtt = gt->_gt->ggtt;
1078 struct intel_uncore *uncore = gt->_gt->uncore;
1079 int i;
1080
1081 if (INTEL_GEN(uncore->i915) >= 6) {
1082 for (i = 0; i < ggtt->num_fences; i++)
1083 gt->fence[i] =
1084 intel_uncore_read64(uncore,
1085 FENCE_REG_GEN6_LO(i));
1086 } else if (INTEL_GEN(uncore->i915) >= 4) {
1087 for (i = 0; i < ggtt->num_fences; i++)
1088 gt->fence[i] =
1089 intel_uncore_read64(uncore,
1090 FENCE_REG_965_LO(i));
1091 } else {
1092 for (i = 0; i < ggtt->num_fences; i++)
1093 gt->fence[i] =
1094 intel_uncore_read(uncore, FENCE_REG(i));
1095 }
1096 gt->nfence = i;
1097 }
1098
1099 static void engine_record_registers(struct intel_engine_coredump *ee)
1100 {
1101 const struct intel_engine_cs *engine = ee->engine;
1102 struct drm_i915_private *i915 = engine->i915;
1103
1104 if (INTEL_GEN(i915) >= 6) {
1105 ee->rc_psmi = ENGINE_READ(engine, RING_PSMI_CTL);
1106
1107 if (INTEL_GEN(i915) >= 12)
1108 ee->fault_reg = intel_uncore_read(engine->uncore,
1109 GEN12_RING_FAULT_REG);
1110 else if (INTEL_GEN(i915) >= 8)
1111 ee->fault_reg = intel_uncore_read(engine->uncore,
1112 GEN8_RING_FAULT_REG);
1113 else
1114 ee->fault_reg = GEN6_RING_FAULT_REG_READ(engine);
1115 }
1116
1117 if (INTEL_GEN(i915) >= 4) {
1118 ee->esr = ENGINE_READ(engine, RING_ESR);
1119 ee->faddr = ENGINE_READ(engine, RING_DMA_FADD);
1120 ee->ipeir = ENGINE_READ(engine, RING_IPEIR);
1121 ee->ipehr = ENGINE_READ(engine, RING_IPEHR);
1122 ee->instps = ENGINE_READ(engine, RING_INSTPS);
1123 ee->bbaddr = ENGINE_READ(engine, RING_BBADDR);
1124 ee->ccid = ENGINE_READ(engine, CCID);
1125 if (INTEL_GEN(i915) >= 8) {
1126 ee->faddr |= (u64)ENGINE_READ(engine, RING_DMA_FADD_UDW) << 32;
1127 ee->bbaddr |= (u64)ENGINE_READ(engine, RING_BBADDR_UDW) << 32;
1128 }
1129 ee->bbstate = ENGINE_READ(engine, RING_BBSTATE);
1130 } else {
1131 ee->faddr = ENGINE_READ(engine, DMA_FADD_I8XX);
1132 ee->ipeir = ENGINE_READ(engine, IPEIR);
1133 ee->ipehr = ENGINE_READ(engine, IPEHR);
1134 }
1135
1136 intel_engine_get_instdone(engine, &ee->instdone);
1137
1138 ee->instpm = ENGINE_READ(engine, RING_INSTPM);
1139 ee->acthd = intel_engine_get_active_head(engine);
1140 ee->start = ENGINE_READ(engine, RING_START);
1141 ee->head = ENGINE_READ(engine, RING_HEAD);
1142 ee->tail = ENGINE_READ(engine, RING_TAIL);
1143 ee->ctl = ENGINE_READ(engine, RING_CTL);
1144 if (INTEL_GEN(i915) > 2)
1145 ee->mode = ENGINE_READ(engine, RING_MI_MODE);
1146
1147 if (!HWS_NEEDS_PHYSICAL(i915)) {
1148 i915_reg_t mmio;
1149
1150 if (IS_GEN(i915, 7)) {
1151 switch (engine->id) {
1152 default:
1153 MISSING_CASE(engine->id);
1154 /* fall through */
1155 case RCS0:
1156 mmio = RENDER_HWS_PGA_GEN7;
1157 break;
1158 case BCS0:
1159 mmio = BLT_HWS_PGA_GEN7;
1160 break;
1161 case VCS0:
1162 mmio = BSD_HWS_PGA_GEN7;
1163 break;
1164 case VECS0:
1165 mmio = VEBOX_HWS_PGA_GEN7;
1166 break;
1167 }
1168 } else if (IS_GEN(engine->i915, 6)) {
1169 mmio = RING_HWS_PGA_GEN6(engine->mmio_base);
1170 } else {
1171 /* XXX: gen8 returns to sanity */
1172 mmio = RING_HWS_PGA(engine->mmio_base);
1173 }
1174
1175 ee->hws = intel_uncore_read(engine->uncore, mmio);
1176 }
1177
1178 ee->reset_count = i915_reset_engine_count(&i915->gpu_error, engine);
1179
1180 if (HAS_PPGTT(i915)) {
1181 int i;
1182
1183 ee->vm_info.gfx_mode = ENGINE_READ(engine, RING_MODE_GEN7);
1184
1185 if (IS_GEN(i915, 6)) {
1186 ee->vm_info.pp_dir_base =
1187 ENGINE_READ(engine, RING_PP_DIR_BASE_READ);
1188 } else if (IS_GEN(i915, 7)) {
1189 ee->vm_info.pp_dir_base =
1190 ENGINE_READ(engine, RING_PP_DIR_BASE);
1191 } else if (INTEL_GEN(i915) >= 8) {
1192 u32 base = engine->mmio_base;
1193
1194 for (i = 0; i < 4; i++) {
1195 ee->vm_info.pdp[i] =
1196 intel_uncore_read(engine->uncore,
1197 GEN8_RING_PDP_UDW(base, i));
1198 ee->vm_info.pdp[i] <<= 32;
1199 ee->vm_info.pdp[i] |=
1200 intel_uncore_read(engine->uncore,
1201 GEN8_RING_PDP_LDW(base, i));
1202 }
1203 }
1204 }
1205 }
1206
1207 static void record_request(const struct i915_request *request,
1208 struct i915_request_coredump *erq)
1209 {
1210 const struct i915_gem_context *ctx;
1211
1212 erq->flags = request->fence.flags;
1213 erq->context = request->fence.context;
1214 erq->seqno = request->fence.seqno;
1215 erq->sched_attr = request->sched.attr;
1216 erq->start = i915_ggtt_offset(request->ring->vma);
1217 erq->head = request->head;
1218 erq->tail = request->tail;
1219
1220 erq->pid = 0;
1221 rcu_read_lock();
1222 ctx = rcu_dereference(request->context->gem_context);
1223 if (ctx)
1224 erq->pid = pid_nr(ctx->pid);
1225 rcu_read_unlock();
1226 }
1227
1228 static void engine_record_execlists(struct intel_engine_coredump *ee)
1229 {
1230 const struct intel_engine_execlists * const el = &ee->engine->execlists;
1231 struct i915_request * const *port = el->active;
1232 unsigned int n = 0;
1233
1234 while (*port)
1235 record_request(*port++, &ee->execlist[n++]);
1236
1237 ee->num_ports = n;
1238 }
1239
1240 static bool record_context(struct i915_gem_context_coredump *e,
1241 const struct i915_request *rq)
1242 {
1243 struct i915_gem_context *ctx;
1244 struct task_struct *task;
1245 bool simulated;
1246
1247 rcu_read_lock();
1248 ctx = rcu_dereference(rq->context->gem_context);
1249 if (ctx && !kref_get_unless_zero(&ctx->ref))
1250 ctx = NULL;
1251 rcu_read_unlock();
1252 if (!ctx)
1253 return true;
1254
1255 rcu_read_lock();
1256 task = pid_task(ctx->pid, PIDTYPE_PID);
1257 if (task) {
1258 strcpy(e->comm, task->comm);
1259 e->pid = task->pid;
1260 }
1261 rcu_read_unlock();
1262
1263 e->sched_attr = ctx->sched;
1264 e->guilty = atomic_read(&ctx->guilty_count);
1265 e->active = atomic_read(&ctx->active_count);
1266
1267 e->total_runtime = rq->context->runtime.total;
1268 e->avg_runtime = ewma_runtime_read(&rq->context->runtime.avg);
1269
1270 simulated = i915_gem_context_no_error_capture(ctx);
1271
1272 i915_gem_context_put(ctx);
1273 return simulated;
1274 }
1275
1276 struct intel_engine_capture_vma {
1277 struct intel_engine_capture_vma *next;
1278 struct i915_vma *vma;
1279 char name[16];
1280 };
1281
1282 static struct intel_engine_capture_vma *
1283 capture_vma(struct intel_engine_capture_vma *next,
1284 struct i915_vma *vma,
1285 const char *name,
1286 gfp_t gfp)
1287 {
1288 struct intel_engine_capture_vma *c;
1289
1290 if (!vma)
1291 return next;
1292
1293 c = kmalloc(sizeof(*c), gfp);
1294 if (!c)
1295 return next;
1296
1297 if (!i915_active_acquire_if_busy(&vma->active)) {
1298 kfree(c);
1299 return next;
1300 }
1301
1302 strcpy(c->name, name);
1303 c->vma = i915_vma_get(vma);
1304
1305 c->next = next;
1306 return c;
1307 }
1308
1309 static struct intel_engine_capture_vma *
1310 capture_user(struct intel_engine_capture_vma *capture,
1311 const struct i915_request *rq,
1312 gfp_t gfp)
1313 {
1314 struct i915_capture_list *c;
1315
1316 for (c = rq->capture_list; c; c = c->next)
1317 capture = capture_vma(capture, c->vma, "user", gfp);
1318
1319 return capture;
1320 }
1321
1322 static struct i915_vma_coredump *
1323 capture_object(const struct intel_gt *gt,
1324 struct drm_i915_gem_object *obj,
1325 const char *name,
1326 struct i915_vma_compress *compress)
1327 {
1328 if (obj && i915_gem_object_has_pages(obj)) {
1329 struct i915_vma fake = {
1330 .node = { .start = U64_MAX, .size = obj->base.size },
1331 .size = obj->base.size,
1332 .pages = obj->mm.pages,
1333 .obj = obj,
1334 };
1335
1336 return i915_vma_coredump_create(gt, &fake, name, compress);
1337 } else {
1338 return NULL;
1339 }
1340 }
1341
1342 static void add_vma(struct intel_engine_coredump *ee,
1343 struct i915_vma_coredump *vma)
1344 {
1345 if (vma) {
1346 vma->next = ee->vma;
1347 ee->vma = vma;
1348 }
1349 }
1350
1351 struct intel_engine_coredump *
1352 intel_engine_coredump_alloc(struct intel_engine_cs *engine, gfp_t gfp)
1353 {
1354 struct intel_engine_coredump *ee;
1355
1356 ee = kzalloc(sizeof(*ee), gfp);
1357 if (!ee)
1358 return NULL;
1359
1360 ee->engine = engine;
1361
1362 engine_record_registers(ee);
1363 engine_record_execlists(ee);
1364
1365 return ee;
1366 }
1367
1368 struct intel_engine_capture_vma *
1369 intel_engine_coredump_add_request(struct intel_engine_coredump *ee,
1370 struct i915_request *rq,
1371 gfp_t gfp)
1372 {
1373 struct intel_engine_capture_vma *vma = NULL;
1374
1375 ee->simulated |= record_context(&ee->context, rq);
1376 if (ee->simulated)
1377 return NULL;
1378
1379 /*
1380 * We need to copy these to an anonymous buffer
1381 * as the simplest method to avoid being overwritten
1382 * by userspace.
1383 */
1384 vma = capture_vma(vma, rq->batch, "batch", gfp);
1385 vma = capture_user(vma, rq, gfp);
1386 vma = capture_vma(vma, rq->ring->vma, "ring", gfp);
1387 vma = capture_vma(vma, rq->context->state, "HW context", gfp);
1388
1389 ee->rq_head = rq->head;
1390 ee->rq_post = rq->postfix;
1391 ee->rq_tail = rq->tail;
1392
1393 return vma;
1394 }
1395
1396 void
1397 intel_engine_coredump_add_vma(struct intel_engine_coredump *ee,
1398 struct intel_engine_capture_vma *capture,
1399 struct i915_vma_compress *compress)
1400 {
1401 const struct intel_engine_cs *engine = ee->engine;
1402
1403 while (capture) {
1404 struct intel_engine_capture_vma *this = capture;
1405 struct i915_vma *vma = this->vma;
1406
1407 add_vma(ee,
1408 i915_vma_coredump_create(engine->gt,
1409 vma, this->name,
1410 compress));
1411
1412 i915_active_release(&vma->active);
1413 i915_vma_put(vma);
1414
1415 capture = this->next;
1416 kfree(this);
1417 }
1418
1419 add_vma(ee,
1420 i915_vma_coredump_create(engine->gt,
1421 engine->status_page.vma,
1422 "HW Status",
1423 compress));
1424
1425 add_vma(ee,
1426 i915_vma_coredump_create(engine->gt,
1427 engine->wa_ctx.vma,
1428 "WA context",
1429 compress));
1430
1431 add_vma(ee,
1432 capture_object(engine->gt,
1433 engine->default_state,
1434 "NULL context",
1435 compress));
1436 }
1437
1438 static struct intel_engine_coredump *
1439 capture_engine(struct intel_engine_cs *engine,
1440 struct i915_vma_compress *compress)
1441 {
1442 struct intel_engine_capture_vma *capture = NULL;
1443 struct intel_engine_coredump *ee;
1444 struct i915_request *rq;
1445 unsigned long flags;
1446
1447 ee = intel_engine_coredump_alloc(engine, GFP_KERNEL);
1448 if (!ee)
1449 return NULL;
1450
1451 spin_lock_irqsave(&engine->active.lock, flags);
1452 rq = intel_engine_find_active_request(engine);
1453 if (rq)
1454 capture = intel_engine_coredump_add_request(ee, rq,
1455 ATOMIC_MAYFAIL);
1456 spin_unlock_irqrestore(&engine->active.lock, flags);
1457 if (!capture) {
1458 kfree(ee);
1459 return NULL;
1460 }
1461
1462 intel_engine_coredump_add_vma(ee, capture, compress);
1463
1464 return ee;
1465 }
1466
1467 static void
1468 gt_record_engines(struct intel_gt_coredump *gt,
1469 struct i915_vma_compress *compress)
1470 {
1471 struct intel_engine_cs *engine;
1472 enum intel_engine_id id;
1473
1474 for_each_engine(engine, gt->_gt, id) {
1475 struct intel_engine_coredump *ee;
1476
1477 /* Refill our page pool before entering atomic section */
1478 pool_refill(&compress->pool, ALLOW_FAIL);
1479
1480 ee = capture_engine(engine, compress);
1481 if (!ee)
1482 continue;
1483
1484 gt->simulated |= ee->simulated;
1485 if (ee->simulated) {
1486 kfree(ee);
1487 continue;
1488 }
1489
1490 ee->next = gt->engine;
1491 gt->engine = ee;
1492 }
1493 }
1494
1495 static struct intel_uc_coredump *
1496 gt_record_uc(struct intel_gt_coredump *gt,
1497 struct i915_vma_compress *compress)
1498 {
1499 const struct intel_uc *uc = &gt->_gt->uc;
1500 struct intel_uc_coredump *error_uc;
1501
1502 error_uc = kzalloc(sizeof(*error_uc), ALLOW_FAIL);
1503 if (!error_uc)
1504 return NULL;
1505
1506 memcpy(&error_uc->guc_fw, &uc->guc.fw, sizeof(uc->guc.fw));
1507 memcpy(&error_uc->huc_fw, &uc->huc.fw, sizeof(uc->huc.fw));
1508
1509 /* Non-default firmware paths will be specified by the modparam.
1510 * As modparams are generally accesible from the userspace make
1511 * explicit copies of the firmware paths.
1512 */
1513 error_uc->guc_fw.path = kstrdup(uc->guc.fw.path, ALLOW_FAIL);
1514 error_uc->huc_fw.path = kstrdup(uc->huc.fw.path, ALLOW_FAIL);
1515 error_uc->guc_log =
1516 i915_vma_coredump_create(gt->_gt,
1517 uc->guc.log.vma, "GuC log buffer",
1518 compress);
1519
1520 return error_uc;
1521 }
1522
1523 static void gt_capture_prepare(struct intel_gt_coredump *gt)
1524 {
1525 struct i915_ggtt *ggtt = gt->_gt->ggtt;
1526
1527 mutex_lock(&ggtt->error_mutex);
1528 }
1529
1530 static void gt_capture_finish(struct intel_gt_coredump *gt)
1531 {
1532 struct i915_ggtt *ggtt = gt->_gt->ggtt;
1533
1534 if (drm_mm_node_allocated(&ggtt->error_capture))
1535 ggtt->vm.clear_range(&ggtt->vm,
1536 ggtt->error_capture.start,
1537 PAGE_SIZE);
1538
1539 mutex_unlock(&ggtt->error_mutex);
1540 }
1541
1542 /* Capture all registers which don't fit into another category. */
1543 static void gt_record_regs(struct intel_gt_coredump *gt)
1544 {
1545 struct intel_uncore *uncore = gt->_gt->uncore;
1546 struct drm_i915_private *i915 = uncore->i915;
1547 int i;
1548
1549 /*
1550 * General organization
1551 * 1. Registers specific to a single generation
1552 * 2. Registers which belong to multiple generations
1553 * 3. Feature specific registers.
1554 * 4. Everything else
1555 * Please try to follow the order.
1556 */
1557
1558 /* 1: Registers specific to a single generation */
1559 if (IS_VALLEYVIEW(i915)) {
1560 gt->gtier[0] = intel_uncore_read(uncore, GTIER);
1561 gt->ier = intel_uncore_read(uncore, VLV_IER);
1562 gt->forcewake = intel_uncore_read_fw(uncore, FORCEWAKE_VLV);
1563 }
1564
1565 if (IS_GEN(i915, 7))
1566 gt->err_int = intel_uncore_read(uncore, GEN7_ERR_INT);
1567
1568 if (INTEL_GEN(i915) >= 12) {
1569 gt->fault_data0 = intel_uncore_read(uncore,
1570 GEN12_FAULT_TLB_DATA0);
1571 gt->fault_data1 = intel_uncore_read(uncore,
1572 GEN12_FAULT_TLB_DATA1);
1573 } else if (INTEL_GEN(i915) >= 8) {
1574 gt->fault_data0 = intel_uncore_read(uncore,
1575 GEN8_FAULT_TLB_DATA0);
1576 gt->fault_data1 = intel_uncore_read(uncore,
1577 GEN8_FAULT_TLB_DATA1);
1578 }
1579
1580 if (IS_GEN(i915, 6)) {
1581 gt->forcewake = intel_uncore_read_fw(uncore, FORCEWAKE);
1582 gt->gab_ctl = intel_uncore_read(uncore, GAB_CTL);
1583 gt->gfx_mode = intel_uncore_read(uncore, GFX_MODE);
1584 }
1585
1586 /* 2: Registers which belong to multiple generations */
1587 if (INTEL_GEN(i915) >= 7)
1588 gt->forcewake = intel_uncore_read_fw(uncore, FORCEWAKE_MT);
1589
1590 if (INTEL_GEN(i915) >= 6) {
1591 gt->derrmr = intel_uncore_read(uncore, DERRMR);
1592 if (INTEL_GEN(i915) < 12) {
1593 gt->error = intel_uncore_read(uncore, ERROR_GEN6);
1594 gt->done_reg = intel_uncore_read(uncore, DONE_REG);
1595 }
1596 }
1597
1598 /* 3: Feature specific registers */
1599 if (IS_GEN_RANGE(i915, 6, 7)) {
1600 gt->gam_ecochk = intel_uncore_read(uncore, GAM_ECOCHK);
1601 gt->gac_eco = intel_uncore_read(uncore, GAC_ECO_BITS);
1602 }
1603
1604 if (IS_GEN_RANGE(i915, 8, 11))
1605 gt->gtt_cache = intel_uncore_read(uncore, HSW_GTT_CACHE_EN);
1606
1607 if (IS_GEN(i915, 12))
1608 gt->aux_err = intel_uncore_read(uncore, GEN12_AUX_ERR_DBG);
1609
1610 if (INTEL_GEN(i915) >= 12) {
1611 for (i = 0; i < GEN12_SFC_DONE_MAX; i++) {
1612 gt->sfc_done[i] =
1613 intel_uncore_read(uncore, GEN12_SFC_DONE(i));
1614 }
1615
1616 gt->gam_done = intel_uncore_read(uncore, GEN12_GAM_DONE);
1617 }
1618
1619 /* 4: Everything else */
1620 if (INTEL_GEN(i915) >= 11) {
1621 gt->ier = intel_uncore_read(uncore, GEN8_DE_MISC_IER);
1622 gt->gtier[0] =
1623 intel_uncore_read(uncore,
1624 GEN11_RENDER_COPY_INTR_ENABLE);
1625 gt->gtier[1] =
1626 intel_uncore_read(uncore, GEN11_VCS_VECS_INTR_ENABLE);
1627 gt->gtier[2] =
1628 intel_uncore_read(uncore, GEN11_GUC_SG_INTR_ENABLE);
1629 gt->gtier[3] =
1630 intel_uncore_read(uncore,
1631 GEN11_GPM_WGBOXPERF_INTR_ENABLE);
1632 gt->gtier[4] =
1633 intel_uncore_read(uncore,
1634 GEN11_CRYPTO_RSVD_INTR_ENABLE);
1635 gt->gtier[5] =
1636 intel_uncore_read(uncore,
1637 GEN11_GUNIT_CSME_INTR_ENABLE);
1638 gt->ngtier = 6;
1639 } else if (INTEL_GEN(i915) >= 8) {
1640 gt->ier = intel_uncore_read(uncore, GEN8_DE_MISC_IER);
1641 for (i = 0; i < 4; i++)
1642 gt->gtier[i] =
1643 intel_uncore_read(uncore, GEN8_GT_IER(i));
1644 gt->ngtier = 4;
1645 } else if (HAS_PCH_SPLIT(i915)) {
1646 gt->ier = intel_uncore_read(uncore, DEIER);
1647 gt->gtier[0] = intel_uncore_read(uncore, GTIER);
1648 gt->ngtier = 1;
1649 } else if (IS_GEN(i915, 2)) {
1650 gt->ier = intel_uncore_read16(uncore, GEN2_IER);
1651 } else if (!IS_VALLEYVIEW(i915)) {
1652 gt->ier = intel_uncore_read(uncore, GEN2_IER);
1653 }
1654 gt->eir = intel_uncore_read(uncore, EIR);
1655 gt->pgtbl_er = intel_uncore_read(uncore, PGTBL_ER);
1656 }
1657
1658 /*
1659 * Generate a semi-unique error code. The code is not meant to have meaning, The
1660 * code's only purpose is to try to prevent false duplicated bug reports by
1661 * grossly estimating a GPU error state.
1662 *
1663 * TODO Ideally, hashing the batchbuffer would be a very nice way to determine
1664 * the hang if we could strip the GTT offset information from it.
1665 *
1666 * It's only a small step better than a random number in its current form.
1667 */
1668 static u32 generate_ecode(const struct intel_engine_coredump *ee)
1669 {
1670 /*
1671 * IPEHR would be an ideal way to detect errors, as it's the gross
1672 * measure of "the command that hung." However, has some very common
1673 * synchronization commands which almost always appear in the case
1674 * strictly a client bug. Use instdone to differentiate those some.
1675 */
1676 return ee ? ee->ipehr ^ ee->instdone.instdone : 0;
1677 }
1678
1679 static const char *error_msg(struct i915_gpu_coredump *error)
1680 {
1681 struct intel_engine_coredump *first = NULL;
1682 struct intel_gt_coredump *gt;
1683 intel_engine_mask_t engines;
1684 int len;
1685
1686 engines = 0;
1687 for (gt = error->gt; gt; gt = gt->next) {
1688 struct intel_engine_coredump *cs;
1689
1690 if (gt->engine && !first)
1691 first = gt->engine;
1692
1693 for (cs = gt->engine; cs; cs = cs->next)
1694 engines |= cs->engine->mask;
1695 }
1696
1697 len = scnprintf(error->error_msg, sizeof(error->error_msg),
1698 "GPU HANG: ecode %d:%x:%08x",
1699 INTEL_GEN(error->i915), engines,
1700 generate_ecode(first));
1701 if (first && first->context.pid) {
1702 /* Just show the first executing process, more is confusing */
1703 len += scnprintf(error->error_msg + len,
1704 sizeof(error->error_msg) - len,
1705 ", in %s [%d]",
1706 first->context.comm, first->context.pid);
1707 }
1708
1709 return error->error_msg;
1710 }
1711
1712 static void capture_gen(struct i915_gpu_coredump *error)
1713 {
1714 struct drm_i915_private *i915 = error->i915;
1715
1716 error->wakelock = atomic_read(&i915->runtime_pm.wakeref_count);
1717 error->suspended = i915->runtime_pm.suspended;
1718
1719 error->iommu = -1;
1720 #ifdef CONFIG_INTEL_IOMMU
1721 error->iommu = intel_iommu_gfx_mapped;
1722 #endif
1723 error->reset_count = i915_reset_count(&i915->gpu_error);
1724 error->suspend_count = i915->suspend_count;
1725
1726 i915_params_copy(&error->params, &i915_modparams);
1727 memcpy(&error->device_info,
1728 INTEL_INFO(i915),
1729 sizeof(error->device_info));
1730 memcpy(&error->runtime_info,
1731 RUNTIME_INFO(i915),
1732 sizeof(error->runtime_info));
1733 error->driver_caps = i915->caps;
1734 }
1735
1736 struct i915_gpu_coredump *
1737 i915_gpu_coredump_alloc(struct drm_i915_private *i915, gfp_t gfp)
1738 {
1739 struct i915_gpu_coredump *error;
1740
1741 if (!i915_modparams.error_capture)
1742 return NULL;
1743
1744 error = kzalloc(sizeof(*error), gfp);
1745 if (!error)
1746 return NULL;
1747
1748 kref_init(&error->ref);
1749 error->i915 = i915;
1750
1751 error->time = ktime_get_real();
1752 error->boottime = ktime_get_boottime();
1753 error->uptime = ktime_sub(ktime_get(), i915->gt.last_init_time);
1754 error->capture = jiffies;
1755
1756 capture_gen(error);
1757
1758 return error;
1759 }
1760
1761 #define DAY_AS_SECONDS(x) (24 * 60 * 60 * (x))
1762
1763 struct intel_gt_coredump *
1764 intel_gt_coredump_alloc(struct intel_gt *gt, gfp_t gfp)
1765 {
1766 struct intel_gt_coredump *gc;
1767
1768 gc = kzalloc(sizeof(*gc), gfp);
1769 if (!gc)
1770 return NULL;
1771
1772 gc->_gt = gt;
1773 gc->awake = intel_gt_pm_is_awake(gt);
1774
1775 gt_record_regs(gc);
1776 gt_record_fences(gc);
1777
1778 return gc;
1779 }
1780
1781 struct i915_vma_compress *
1782 i915_vma_capture_prepare(struct intel_gt_coredump *gt)
1783 {
1784 struct i915_vma_compress *compress;
1785
1786 compress = kmalloc(sizeof(*compress), ALLOW_FAIL);
1787 if (!compress)
1788 return NULL;
1789
1790 if (!compress_init(compress)) {
1791 kfree(compress);
1792 return NULL;
1793 }
1794
1795 gt_capture_prepare(gt);
1796
1797 return compress;
1798 }
1799
1800 void i915_vma_capture_finish(struct intel_gt_coredump *gt,
1801 struct i915_vma_compress *compress)
1802 {
1803 if (!compress)
1804 return;
1805
1806 gt_capture_finish(gt);
1807
1808 compress_fini(compress);
1809 kfree(compress);
1810 }
1811
1812 struct i915_gpu_coredump *i915_gpu_coredump(struct drm_i915_private *i915)
1813 {
1814 struct i915_gpu_coredump *error;
1815
1816 /* Check if GPU capture has been disabled */
1817 error = READ_ONCE(i915->gpu_error.first_error);
1818 if (IS_ERR(error))
1819 return error;
1820
1821 error = i915_gpu_coredump_alloc(i915, ALLOW_FAIL);
1822 if (!error)
1823 return ERR_PTR(-ENOMEM);
1824
1825 error->gt = intel_gt_coredump_alloc(&i915->gt, ALLOW_FAIL);
1826 if (error->gt) {
1827 struct i915_vma_compress *compress;
1828
1829 compress = i915_vma_capture_prepare(error->gt);
1830 if (!compress) {
1831 kfree(error->gt);
1832 kfree(error);
1833 return ERR_PTR(-ENOMEM);
1834 }
1835
1836 gt_record_engines(error->gt, compress);
1837
1838 if (INTEL_INFO(i915)->has_gt_uc)
1839 error->gt->uc = gt_record_uc(error->gt, compress);
1840
1841 i915_vma_capture_finish(error->gt, compress);
1842
1843 error->simulated |= error->gt->simulated;
1844 }
1845
1846 error->overlay = intel_overlay_capture_error_state(i915);
1847 error->display = intel_display_capture_error_state(i915);
1848
1849 return error;
1850 }
1851
1852 void i915_error_state_store(struct i915_gpu_coredump *error)
1853 {
1854 struct drm_i915_private *i915;
1855 static bool warned;
1856
1857 if (IS_ERR_OR_NULL(error))
1858 return;
1859
1860 i915 = error->i915;
1861 dev_info(i915->drm.dev, "%s\n", error_msg(error));
1862
1863 if (error->simulated ||
1864 cmpxchg(&i915->gpu_error.first_error, NULL, error))
1865 return;
1866
1867 i915_gpu_coredump_get(error);
1868
1869 if (!xchg(&warned, true) &&
1870 ktime_get_real_seconds() - DRIVER_TIMESTAMP < DAY_AS_SECONDS(180)) {
1871 pr_info("GPU hangs can indicate a bug anywhere in the entire gfx stack, including userspace.\n");
1872 pr_info("Please file a _new_ bug report at https://gitlab.freedesktop.org/drm/intel/issues/new.\n");
1873 pr_info("Please see https://gitlab.freedesktop.org/drm/intel/-/wikis/How-to-file-i915-bugs for details.\n");
1874 pr_info("drm/i915 developers can then reassign to the right component if it's not a kernel issue.\n");
1875 pr_info("The GPU crash dump is required to analyze GPU hangs, so please always attach it.\n");
1876 pr_info("GPU crash dump saved to /sys/class/drm/card%d/error\n",
1877 i915->drm.primary->index);
1878 }
1879 }
1880
1881 /**
1882 * i915_capture_error_state - capture an error record for later analysis
1883 * @i915: i915 device
1884 *
1885 * Should be called when an error is detected (either a hang or an error
1886 * interrupt) to capture error state from the time of the error. Fills
1887 * out a structure which becomes available in debugfs for user level tools
1888 * to pick up.
1889 */
1890 void i915_capture_error_state(struct drm_i915_private *i915)
1891 {
1892 struct i915_gpu_coredump *error;
1893
1894 error = i915_gpu_coredump(i915);
1895 if (IS_ERR(error)) {
1896 cmpxchg(&i915->gpu_error.first_error, NULL, error);
1897 return;
1898 }
1899
1900 i915_error_state_store(error);
1901 i915_gpu_coredump_put(error);
1902 }
1903
1904 struct i915_gpu_coredump *
1905 i915_first_error_state(struct drm_i915_private *i915)
1906 {
1907 struct i915_gpu_coredump *error;
1908
1909 spin_lock_irq(&i915->gpu_error.lock);
1910 error = i915->gpu_error.first_error;
1911 if (!IS_ERR_OR_NULL(error))
1912 i915_gpu_coredump_get(error);
1913 spin_unlock_irq(&i915->gpu_error.lock);
1914
1915 return error;
1916 }
1917
1918 void i915_reset_error_state(struct drm_i915_private *i915)
1919 {
1920 struct i915_gpu_coredump *error;
1921
1922 spin_lock_irq(&i915->gpu_error.lock);
1923 error = i915->gpu_error.first_error;
1924 if (error != ERR_PTR(-ENODEV)) /* if disabled, always disabled */
1925 i915->gpu_error.first_error = NULL;
1926 spin_unlock_irq(&i915->gpu_error.lock);
1927
1928 if (!IS_ERR_OR_NULL(error))
1929 i915_gpu_coredump_put(error);
1930 }
1931
1932 void i915_disable_error_state(struct drm_i915_private *i915, int err)
1933 {
1934 spin_lock_irq(&i915->gpu_error.lock);
1935 if (!i915->gpu_error.first_error)
1936 i915->gpu_error.first_error = ERR_PTR(err);
1937 spin_unlock_irq(&i915->gpu_error.lock);
1938 }