vt: vt_ioctl: fix VT_DISALLOCATE freeing in-use virtual console
[linux/fpc-iii.git] / drivers / gpu / drm / i915 / i915_gpu_error.c
bloba262a64f562565d80642cf887788ac261991336e
1 /*
2 * Copyright (c) 2008 Intel Corporation
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:
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.
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.
23 * Authors:
24 * Eric Anholt <eric@anholt.net>
25 * Keith Packard <keithp@keithp.com>
26 * Mika Kuoppala <mika.kuoppala@intel.com>
30 #include <generated/utsrelease.h>
31 #include <linux/stop_machine.h>
32 #include <linux/zlib.h>
33 #include <drm/drm_print.h>
34 #include <linux/ascii85.h>
36 #include "i915_gpu_error.h"
37 #include "i915_drv.h"
39 static inline const struct intel_engine_cs *
40 engine_lookup(const struct drm_i915_private *i915, unsigned int id)
42 if (id >= I915_NUM_ENGINES)
43 return NULL;
45 return i915->engine[id];
48 static inline const char *
49 __engine_name(const struct intel_engine_cs *engine)
51 return engine ? engine->name : "";
54 static const char *
55 engine_name(const struct drm_i915_private *i915, unsigned int id)
57 return __engine_name(engine_lookup(i915, id));
60 static const char *tiling_flag(int tiling)
62 switch (tiling) {
63 default:
64 case I915_TILING_NONE: return "";
65 case I915_TILING_X: return " X";
66 case I915_TILING_Y: return " Y";
70 static const char *dirty_flag(int dirty)
72 return dirty ? " dirty" : "";
75 static const char *purgeable_flag(int purgeable)
77 return purgeable ? " purgeable" : "";
80 static bool __i915_error_ok(struct drm_i915_error_state_buf *e)
83 if (!e->err && WARN(e->bytes > (e->size - 1), "overflow")) {
84 e->err = -ENOSPC;
85 return false;
88 if (e->bytes == e->size - 1 || e->err)
89 return false;
91 return true;
94 static bool __i915_error_seek(struct drm_i915_error_state_buf *e,
95 unsigned len)
97 if (e->pos + len <= e->start) {
98 e->pos += len;
99 return false;
102 /* First vsnprintf needs to fit in its entirety for memmove */
103 if (len >= e->size) {
104 e->err = -EIO;
105 return false;
108 return true;
111 static void __i915_error_advance(struct drm_i915_error_state_buf *e,
112 unsigned len)
114 /* If this is first printf in this window, adjust it so that
115 * start position matches start of the buffer
118 if (e->pos < e->start) {
119 const size_t off = e->start - e->pos;
121 /* Should not happen but be paranoid */
122 if (off > len || e->bytes) {
123 e->err = -EIO;
124 return;
127 memmove(e->buf, e->buf + off, len - off);
128 e->bytes = len - off;
129 e->pos = e->start;
130 return;
133 e->bytes += len;
134 e->pos += len;
137 __printf(2, 0)
138 static void i915_error_vprintf(struct drm_i915_error_state_buf *e,
139 const char *f, va_list args)
141 unsigned len;
143 if (!__i915_error_ok(e))
144 return;
146 /* Seek the first printf which is hits start position */
147 if (e->pos < e->start) {
148 va_list tmp;
150 va_copy(tmp, args);
151 len = vsnprintf(NULL, 0, f, tmp);
152 va_end(tmp);
154 if (!__i915_error_seek(e, len))
155 return;
158 len = vsnprintf(e->buf + e->bytes, e->size - e->bytes, f, args);
159 if (len >= e->size - e->bytes)
160 len = e->size - e->bytes - 1;
162 __i915_error_advance(e, len);
165 static void i915_error_puts(struct drm_i915_error_state_buf *e,
166 const char *str)
168 unsigned len;
170 if (!__i915_error_ok(e))
171 return;
173 len = strlen(str);
175 /* Seek the first printf which is hits start position */
176 if (e->pos < e->start) {
177 if (!__i915_error_seek(e, len))
178 return;
181 if (len >= e->size - e->bytes)
182 len = e->size - e->bytes - 1;
183 memcpy(e->buf + e->bytes, str, len);
185 __i915_error_advance(e, len);
188 #define err_printf(e, ...) i915_error_printf(e, __VA_ARGS__)
189 #define err_puts(e, s) i915_error_puts(e, s)
191 static void __i915_printfn_error(struct drm_printer *p, struct va_format *vaf)
193 i915_error_vprintf(p->arg, vaf->fmt, *vaf->va);
196 static inline struct drm_printer
197 i915_error_printer(struct drm_i915_error_state_buf *e)
199 struct drm_printer p = {
200 .printfn = __i915_printfn_error,
201 .arg = e,
203 return p;
206 #ifdef CONFIG_DRM_I915_COMPRESS_ERROR
208 struct compress {
209 struct z_stream_s zstream;
210 void *tmp;
213 static bool compress_init(struct compress *c)
215 struct z_stream_s *zstream = memset(&c->zstream, 0, sizeof(c->zstream));
217 zstream->workspace =
218 kmalloc(zlib_deflate_workspacesize(MAX_WBITS, MAX_MEM_LEVEL),
219 GFP_ATOMIC | __GFP_NOWARN);
220 if (!zstream->workspace)
221 return false;
223 if (zlib_deflateInit(zstream, Z_DEFAULT_COMPRESSION) != Z_OK) {
224 kfree(zstream->workspace);
225 return false;
228 c->tmp = NULL;
229 if (i915_has_memcpy_from_wc())
230 c->tmp = (void *)__get_free_page(GFP_ATOMIC | __GFP_NOWARN);
232 return true;
235 static void *compress_next_page(struct drm_i915_error_object *dst)
237 unsigned long page;
239 if (dst->page_count >= dst->num_pages)
240 return ERR_PTR(-ENOSPC);
242 page = __get_free_page(GFP_ATOMIC | __GFP_NOWARN);
243 if (!page)
244 return ERR_PTR(-ENOMEM);
246 return dst->pages[dst->page_count++] = (void *)page;
249 static int compress_page(struct compress *c,
250 void *src,
251 struct drm_i915_error_object *dst)
253 struct z_stream_s *zstream = &c->zstream;
255 zstream->next_in = src;
256 if (c->tmp && i915_memcpy_from_wc(c->tmp, src, PAGE_SIZE))
257 zstream->next_in = c->tmp;
258 zstream->avail_in = PAGE_SIZE;
260 do {
261 if (zstream->avail_out == 0) {
262 zstream->next_out = compress_next_page(dst);
263 if (IS_ERR(zstream->next_out))
264 return PTR_ERR(zstream->next_out);
266 zstream->avail_out = PAGE_SIZE;
269 if (zlib_deflate(zstream, Z_NO_FLUSH) != Z_OK)
270 return -EIO;
271 } while (zstream->avail_in);
273 /* Fallback to uncompressed if we increase size? */
274 if (0 && zstream->total_out > zstream->total_in)
275 return -E2BIG;
277 return 0;
280 static int compress_flush(struct compress *c,
281 struct drm_i915_error_object *dst)
283 struct z_stream_s *zstream = &c->zstream;
285 do {
286 switch (zlib_deflate(zstream, Z_FINISH)) {
287 case Z_OK: /* more space requested */
288 zstream->next_out = compress_next_page(dst);
289 if (IS_ERR(zstream->next_out))
290 return PTR_ERR(zstream->next_out);
292 zstream->avail_out = PAGE_SIZE;
293 break;
295 case Z_STREAM_END:
296 goto end;
298 default: /* any error */
299 return -EIO;
301 } while (1);
303 end:
304 memset(zstream->next_out, 0, zstream->avail_out);
305 dst->unused = zstream->avail_out;
306 return 0;
309 static void compress_fini(struct compress *c,
310 struct drm_i915_error_object *dst)
312 struct z_stream_s *zstream = &c->zstream;
314 zlib_deflateEnd(zstream);
315 kfree(zstream->workspace);
316 if (c->tmp)
317 free_page((unsigned long)c->tmp);
320 static void err_compression_marker(struct drm_i915_error_state_buf *m)
322 err_puts(m, ":");
325 #else
327 struct compress {
330 static bool compress_init(struct compress *c)
332 return true;
335 static int compress_page(struct compress *c,
336 void *src,
337 struct drm_i915_error_object *dst)
339 unsigned long page;
340 void *ptr;
342 page = __get_free_page(GFP_ATOMIC | __GFP_NOWARN);
343 if (!page)
344 return -ENOMEM;
346 ptr = (void *)page;
347 if (!i915_memcpy_from_wc(ptr, src, PAGE_SIZE))
348 memcpy(ptr, src, PAGE_SIZE);
349 dst->pages[dst->page_count++] = ptr;
351 return 0;
354 static int compress_flush(struct compress *c,
355 struct drm_i915_error_object *dst)
357 return 0;
360 static void compress_fini(struct compress *c,
361 struct drm_i915_error_object *dst)
365 static void err_compression_marker(struct drm_i915_error_state_buf *m)
367 err_puts(m, "~");
370 #endif
372 static void print_error_buffers(struct drm_i915_error_state_buf *m,
373 const char *name,
374 struct drm_i915_error_buffer *err,
375 int count)
377 err_printf(m, "%s [%d]:\n", name, count);
379 while (count--) {
380 err_printf(m, " %08x_%08x %8u %02x %02x %02x",
381 upper_32_bits(err->gtt_offset),
382 lower_32_bits(err->gtt_offset),
383 err->size,
384 err->read_domains,
385 err->write_domain,
386 err->wseqno);
387 err_puts(m, tiling_flag(err->tiling));
388 err_puts(m, dirty_flag(err->dirty));
389 err_puts(m, purgeable_flag(err->purgeable));
390 err_puts(m, err->userptr ? " userptr" : "");
391 err_puts(m, err->engine != -1 ? " " : "");
392 err_puts(m, engine_name(m->i915, err->engine));
393 err_puts(m, i915_cache_level_str(m->i915, err->cache_level));
395 if (err->name)
396 err_printf(m, " (name: %d)", err->name);
397 if (err->fence_reg != I915_FENCE_REG_NONE)
398 err_printf(m, " (fence: %d)", err->fence_reg);
400 err_puts(m, "\n");
401 err++;
405 static void error_print_instdone(struct drm_i915_error_state_buf *m,
406 const struct drm_i915_error_engine *ee)
408 int slice;
409 int subslice;
411 err_printf(m, " INSTDONE: 0x%08x\n",
412 ee->instdone.instdone);
414 if (ee->engine_id != RCS || INTEL_GEN(m->i915) <= 3)
415 return;
417 err_printf(m, " SC_INSTDONE: 0x%08x\n",
418 ee->instdone.slice_common);
420 if (INTEL_GEN(m->i915) <= 6)
421 return;
423 for_each_instdone_slice_subslice(m->i915, slice, subslice)
424 err_printf(m, " SAMPLER_INSTDONE[%d][%d]: 0x%08x\n",
425 slice, subslice,
426 ee->instdone.sampler[slice][subslice]);
428 for_each_instdone_slice_subslice(m->i915, slice, subslice)
429 err_printf(m, " ROW_INSTDONE[%d][%d]: 0x%08x\n",
430 slice, subslice,
431 ee->instdone.row[slice][subslice]);
434 static const char *bannable(const struct drm_i915_error_context *ctx)
436 return ctx->bannable ? "" : " (unbannable)";
439 static void error_print_request(struct drm_i915_error_state_buf *m,
440 const char *prefix,
441 const struct drm_i915_error_request *erq,
442 const unsigned long epoch)
444 if (!erq->seqno)
445 return;
447 err_printf(m, "%s pid %d, ban score %d, seqno %8x:%08x, prio %d, emitted %dms, start %08x, head %08x, tail %08x\n",
448 prefix, erq->pid, erq->ban_score,
449 erq->context, erq->seqno, erq->sched_attr.priority,
450 jiffies_to_msecs(erq->jiffies - epoch),
451 erq->start, erq->head, erq->tail);
454 static void error_print_context(struct drm_i915_error_state_buf *m,
455 const char *header,
456 const struct drm_i915_error_context *ctx)
458 err_printf(m, "%s%s[%d] user_handle %d hw_id %d, prio %d, ban score %d%s guilty %d active %d\n",
459 header, ctx->comm, ctx->pid, ctx->handle, ctx->hw_id,
460 ctx->sched_attr.priority, ctx->ban_score, bannable(ctx),
461 ctx->guilty, ctx->active);
464 static void error_print_engine(struct drm_i915_error_state_buf *m,
465 const struct drm_i915_error_engine *ee,
466 const unsigned long epoch)
468 int n;
470 err_printf(m, "%s command stream:\n",
471 engine_name(m->i915, ee->engine_id));
472 err_printf(m, " IDLE?: %s\n", yesno(ee->idle));
473 err_printf(m, " START: 0x%08x\n", ee->start);
474 err_printf(m, " HEAD: 0x%08x [0x%08x]\n", ee->head, ee->rq_head);
475 err_printf(m, " TAIL: 0x%08x [0x%08x, 0x%08x]\n",
476 ee->tail, ee->rq_post, ee->rq_tail);
477 err_printf(m, " CTL: 0x%08x\n", ee->ctl);
478 err_printf(m, " MODE: 0x%08x\n", ee->mode);
479 err_printf(m, " HWS: 0x%08x\n", ee->hws);
480 err_printf(m, " ACTHD: 0x%08x %08x\n",
481 (u32)(ee->acthd>>32), (u32)ee->acthd);
482 err_printf(m, " IPEIR: 0x%08x\n", ee->ipeir);
483 err_printf(m, " IPEHR: 0x%08x\n", ee->ipehr);
485 error_print_instdone(m, ee);
487 if (ee->batchbuffer) {
488 u64 start = ee->batchbuffer->gtt_offset;
489 u64 end = start + ee->batchbuffer->gtt_size;
491 err_printf(m, " batch: [0x%08x_%08x, 0x%08x_%08x]\n",
492 upper_32_bits(start), lower_32_bits(start),
493 upper_32_bits(end), lower_32_bits(end));
495 if (INTEL_GEN(m->i915) >= 4) {
496 err_printf(m, " BBADDR: 0x%08x_%08x\n",
497 (u32)(ee->bbaddr>>32), (u32)ee->bbaddr);
498 err_printf(m, " BB_STATE: 0x%08x\n", ee->bbstate);
499 err_printf(m, " INSTPS: 0x%08x\n", ee->instps);
501 err_printf(m, " INSTPM: 0x%08x\n", ee->instpm);
502 err_printf(m, " FADDR: 0x%08x %08x\n", upper_32_bits(ee->faddr),
503 lower_32_bits(ee->faddr));
504 if (INTEL_GEN(m->i915) >= 6) {
505 err_printf(m, " RC PSMI: 0x%08x\n", ee->rc_psmi);
506 err_printf(m, " FAULT_REG: 0x%08x\n", ee->fault_reg);
507 err_printf(m, " SYNC_0: 0x%08x\n",
508 ee->semaphore_mboxes[0]);
509 err_printf(m, " SYNC_1: 0x%08x\n",
510 ee->semaphore_mboxes[1]);
511 if (HAS_VEBOX(m->i915))
512 err_printf(m, " SYNC_2: 0x%08x\n",
513 ee->semaphore_mboxes[2]);
515 if (USES_PPGTT(m->i915)) {
516 err_printf(m, " GFX_MODE: 0x%08x\n", ee->vm_info.gfx_mode);
518 if (INTEL_GEN(m->i915) >= 8) {
519 int i;
520 for (i = 0; i < 4; i++)
521 err_printf(m, " PDP%d: 0x%016llx\n",
522 i, ee->vm_info.pdp[i]);
523 } else {
524 err_printf(m, " PP_DIR_BASE: 0x%08x\n",
525 ee->vm_info.pp_dir_base);
528 err_printf(m, " seqno: 0x%08x\n", ee->seqno);
529 err_printf(m, " last_seqno: 0x%08x\n", ee->last_seqno);
530 err_printf(m, " waiting: %s\n", yesno(ee->waiting));
531 err_printf(m, " ring->head: 0x%08x\n", ee->cpu_ring_head);
532 err_printf(m, " ring->tail: 0x%08x\n", ee->cpu_ring_tail);
533 err_printf(m, " hangcheck stall: %s\n", yesno(ee->hangcheck_stalled));
534 err_printf(m, " hangcheck action: %s\n",
535 hangcheck_action_to_str(ee->hangcheck_action));
536 err_printf(m, " hangcheck action timestamp: %dms (%lu%s)\n",
537 jiffies_to_msecs(ee->hangcheck_timestamp - epoch),
538 ee->hangcheck_timestamp,
539 ee->hangcheck_timestamp == epoch ? "; epoch" : "");
540 err_printf(m, " engine reset count: %u\n", ee->reset_count);
542 for (n = 0; n < ee->num_ports; n++) {
543 err_printf(m, " ELSP[%d]:", n);
544 error_print_request(m, " ", &ee->execlist[n], epoch);
547 error_print_context(m, " Active context: ", &ee->context);
550 void i915_error_printf(struct drm_i915_error_state_buf *e, const char *f, ...)
552 va_list args;
554 va_start(args, f);
555 i915_error_vprintf(e, f, args);
556 va_end(args);
559 static void print_error_obj(struct drm_i915_error_state_buf *m,
560 struct intel_engine_cs *engine,
561 const char *name,
562 struct drm_i915_error_object *obj)
564 char out[ASCII85_BUFSZ];
565 int page;
567 if (!obj)
568 return;
570 if (name) {
571 err_printf(m, "%s --- %s = 0x%08x %08x\n",
572 engine ? engine->name : "global", name,
573 upper_32_bits(obj->gtt_offset),
574 lower_32_bits(obj->gtt_offset));
577 err_compression_marker(m);
578 for (page = 0; page < obj->page_count; page++) {
579 int i, len;
581 len = PAGE_SIZE;
582 if (page == obj->page_count - 1)
583 len -= obj->unused;
584 len = ascii85_encode_len(len);
586 for (i = 0; i < len; i++)
587 err_puts(m, ascii85_encode(obj->pages[page][i], out));
589 err_puts(m, "\n");
592 static void err_print_capabilities(struct drm_i915_error_state_buf *m,
593 const struct intel_device_info *info,
594 const struct intel_driver_caps *caps)
596 struct drm_printer p = i915_error_printer(m);
598 intel_device_info_dump_flags(info, &p);
599 intel_driver_caps_print(caps, &p);
600 intel_device_info_dump_topology(&info->sseu, &p);
603 static void err_print_params(struct drm_i915_error_state_buf *m,
604 const struct i915_params *params)
606 struct drm_printer p = i915_error_printer(m);
608 i915_params_dump(params, &p);
611 static void err_print_pciid(struct drm_i915_error_state_buf *m,
612 struct drm_i915_private *i915)
614 struct pci_dev *pdev = i915->drm.pdev;
616 err_printf(m, "PCI ID: 0x%04x\n", pdev->device);
617 err_printf(m, "PCI Revision: 0x%02x\n", pdev->revision);
618 err_printf(m, "PCI Subsystem: %04x:%04x\n",
619 pdev->subsystem_vendor,
620 pdev->subsystem_device);
623 static void err_print_uc(struct drm_i915_error_state_buf *m,
624 const struct i915_error_uc *error_uc)
626 struct drm_printer p = i915_error_printer(m);
627 const struct i915_gpu_state *error =
628 container_of(error_uc, typeof(*error), uc);
630 if (!error->device_info.has_guc)
631 return;
633 intel_uc_fw_dump(&error_uc->guc_fw, &p);
634 intel_uc_fw_dump(&error_uc->huc_fw, &p);
635 print_error_obj(m, NULL, "GuC log buffer", error_uc->guc_log);
638 int i915_error_state_to_str(struct drm_i915_error_state_buf *m,
639 const struct i915_gpu_state *error)
641 struct drm_i915_private *dev_priv = m->i915;
642 struct drm_i915_error_object *obj;
643 struct timespec64 ts;
644 int i, j;
646 if (!error) {
647 err_printf(m, "No error state collected\n");
648 return 0;
651 if (*error->error_msg)
652 err_printf(m, "%s\n", error->error_msg);
653 err_printf(m, "Kernel: " UTS_RELEASE "\n");
654 ts = ktime_to_timespec64(error->time);
655 err_printf(m, "Time: %lld s %ld us\n",
656 (s64)ts.tv_sec, ts.tv_nsec / NSEC_PER_USEC);
657 ts = ktime_to_timespec64(error->boottime);
658 err_printf(m, "Boottime: %lld s %ld us\n",
659 (s64)ts.tv_sec, ts.tv_nsec / NSEC_PER_USEC);
660 ts = ktime_to_timespec64(error->uptime);
661 err_printf(m, "Uptime: %lld s %ld us\n",
662 (s64)ts.tv_sec, ts.tv_nsec / NSEC_PER_USEC);
663 err_printf(m, "Epoch: %lu jiffies (%u HZ)\n", error->epoch, HZ);
664 err_printf(m, "Capture: %lu jiffies; %d ms ago, %d ms after epoch\n",
665 error->capture,
666 jiffies_to_msecs(jiffies - error->capture),
667 jiffies_to_msecs(error->capture - error->epoch));
669 for (i = 0; i < ARRAY_SIZE(error->engine); i++) {
670 if (error->engine[i].hangcheck_stalled &&
671 error->engine[i].context.pid) {
672 err_printf(m, "Active process (on ring %s): %s [%d], score %d%s\n",
673 engine_name(m->i915, i),
674 error->engine[i].context.comm,
675 error->engine[i].context.pid,
676 error->engine[i].context.ban_score,
677 bannable(&error->engine[i].context));
680 err_printf(m, "Reset count: %u\n", error->reset_count);
681 err_printf(m, "Suspend count: %u\n", error->suspend_count);
682 err_printf(m, "Platform: %s\n", intel_platform_name(error->device_info.platform));
683 err_print_pciid(m, error->i915);
685 err_printf(m, "IOMMU enabled?: %d\n", error->iommu);
687 if (HAS_CSR(dev_priv)) {
688 struct intel_csr *csr = &dev_priv->csr;
690 err_printf(m, "DMC loaded: %s\n",
691 yesno(csr->dmc_payload != NULL));
692 err_printf(m, "DMC fw version: %d.%d\n",
693 CSR_VERSION_MAJOR(csr->version),
694 CSR_VERSION_MINOR(csr->version));
697 err_printf(m, "GT awake: %s\n", yesno(error->awake));
698 err_printf(m, "RPM wakelock: %s\n", yesno(error->wakelock));
699 err_printf(m, "PM suspended: %s\n", yesno(error->suspended));
700 err_printf(m, "EIR: 0x%08x\n", error->eir);
701 err_printf(m, "IER: 0x%08x\n", error->ier);
702 for (i = 0; i < error->ngtier; i++)
703 err_printf(m, "GTIER[%d]: 0x%08x\n", i, error->gtier[i]);
704 err_printf(m, "PGTBL_ER: 0x%08x\n", error->pgtbl_er);
705 err_printf(m, "FORCEWAKE: 0x%08x\n", error->forcewake);
706 err_printf(m, "DERRMR: 0x%08x\n", error->derrmr);
707 err_printf(m, "CCID: 0x%08x\n", error->ccid);
708 err_printf(m, "Missed interrupts: 0x%08lx\n", dev_priv->gpu_error.missed_irq_rings);
710 for (i = 0; i < error->nfence; i++)
711 err_printf(m, " fence[%d] = %08llx\n", i, error->fence[i]);
713 if (INTEL_GEN(dev_priv) >= 6) {
714 err_printf(m, "ERROR: 0x%08x\n", error->error);
716 if (INTEL_GEN(dev_priv) >= 8)
717 err_printf(m, "FAULT_TLB_DATA: 0x%08x 0x%08x\n",
718 error->fault_data1, error->fault_data0);
720 err_printf(m, "DONE_REG: 0x%08x\n", error->done_reg);
723 if (IS_GEN7(dev_priv))
724 err_printf(m, "ERR_INT: 0x%08x\n", error->err_int);
726 for (i = 0; i < ARRAY_SIZE(error->engine); i++) {
727 if (error->engine[i].engine_id != -1)
728 error_print_engine(m, &error->engine[i], error->epoch);
731 for (i = 0; i < ARRAY_SIZE(error->active_vm); i++) {
732 char buf[128];
733 int len, first = 1;
735 if (!error->active_vm[i])
736 break;
738 len = scnprintf(buf, sizeof(buf), "Active (");
739 for (j = 0; j < ARRAY_SIZE(error->engine); j++) {
740 if (error->engine[j].vm != error->active_vm[i])
741 continue;
743 len += scnprintf(buf + len, sizeof(buf), "%s%s",
744 first ? "" : ", ",
745 dev_priv->engine[j]->name);
746 first = 0;
748 scnprintf(buf + len, sizeof(buf), ")");
749 print_error_buffers(m, buf,
750 error->active_bo[i],
751 error->active_bo_count[i]);
754 print_error_buffers(m, "Pinned (global)",
755 error->pinned_bo,
756 error->pinned_bo_count);
758 for (i = 0; i < ARRAY_SIZE(error->engine); i++) {
759 const struct drm_i915_error_engine *ee = &error->engine[i];
761 obj = ee->batchbuffer;
762 if (obj) {
763 err_puts(m, dev_priv->engine[i]->name);
764 if (ee->context.pid)
765 err_printf(m, " (submitted by %s [%d], ctx %d [%d], score %d%s)",
766 ee->context.comm,
767 ee->context.pid,
768 ee->context.handle,
769 ee->context.hw_id,
770 ee->context.ban_score,
771 bannable(&ee->context));
772 err_printf(m, " --- gtt_offset = 0x%08x %08x\n",
773 upper_32_bits(obj->gtt_offset),
774 lower_32_bits(obj->gtt_offset));
775 print_error_obj(m, dev_priv->engine[i], NULL, obj);
778 for (j = 0; j < ee->user_bo_count; j++)
779 print_error_obj(m, dev_priv->engine[i],
780 "user", ee->user_bo[j]);
782 if (ee->num_requests) {
783 err_printf(m, "%s --- %d requests\n",
784 dev_priv->engine[i]->name,
785 ee->num_requests);
786 for (j = 0; j < ee->num_requests; j++)
787 error_print_request(m, " ",
788 &ee->requests[j],
789 error->epoch);
792 if (IS_ERR(ee->waiters)) {
793 err_printf(m, "%s --- ? waiters [unable to acquire spinlock]\n",
794 dev_priv->engine[i]->name);
795 } else if (ee->num_waiters) {
796 err_printf(m, "%s --- %d waiters\n",
797 dev_priv->engine[i]->name,
798 ee->num_waiters);
799 for (j = 0; j < ee->num_waiters; j++) {
800 err_printf(m, " seqno 0x%08x for %s [%d]\n",
801 ee->waiters[j].seqno,
802 ee->waiters[j].comm,
803 ee->waiters[j].pid);
807 print_error_obj(m, dev_priv->engine[i],
808 "ringbuffer", ee->ringbuffer);
810 print_error_obj(m, dev_priv->engine[i],
811 "HW Status", ee->hws_page);
813 print_error_obj(m, dev_priv->engine[i],
814 "HW context", ee->ctx);
816 print_error_obj(m, dev_priv->engine[i],
817 "WA context", ee->wa_ctx);
819 print_error_obj(m, dev_priv->engine[i],
820 "WA batchbuffer", ee->wa_batchbuffer);
822 print_error_obj(m, dev_priv->engine[i],
823 "NULL context", ee->default_state);
826 if (error->overlay)
827 intel_overlay_print_error_state(m, error->overlay);
829 if (error->display)
830 intel_display_print_error_state(m, error->display);
832 err_print_capabilities(m, &error->device_info, &error->driver_caps);
833 err_print_params(m, &error->params);
834 err_print_uc(m, &error->uc);
836 if (m->bytes == 0 && m->err)
837 return m->err;
839 return 0;
842 int i915_error_state_buf_init(struct drm_i915_error_state_buf *ebuf,
843 struct drm_i915_private *i915,
844 size_t count, loff_t pos)
846 memset(ebuf, 0, sizeof(*ebuf));
847 ebuf->i915 = i915;
849 /* We need to have enough room to store any i915_error_state printf
850 * so that we can move it to start position.
852 ebuf->size = count + 1 > PAGE_SIZE ? count + 1 : PAGE_SIZE;
853 ebuf->buf = kmalloc(ebuf->size,
854 GFP_KERNEL | __GFP_NORETRY | __GFP_NOWARN);
856 if (ebuf->buf == NULL) {
857 ebuf->size = PAGE_SIZE;
858 ebuf->buf = kmalloc(ebuf->size, GFP_KERNEL);
861 if (ebuf->buf == NULL) {
862 ebuf->size = 128;
863 ebuf->buf = kmalloc(ebuf->size, GFP_KERNEL);
866 if (ebuf->buf == NULL)
867 return -ENOMEM;
869 ebuf->start = pos;
871 return 0;
874 static void i915_error_object_free(struct drm_i915_error_object *obj)
876 int page;
878 if (obj == NULL)
879 return;
881 for (page = 0; page < obj->page_count; page++)
882 free_page((unsigned long)obj->pages[page]);
884 kfree(obj);
887 static __always_inline void free_param(const char *type, void *x)
889 if (!__builtin_strcmp(type, "char *"))
890 kfree(*(void **)x);
893 static void cleanup_params(struct i915_gpu_state *error)
895 #define FREE(T, x, ...) free_param(#T, &error->params.x);
896 I915_PARAMS_FOR_EACH(FREE);
897 #undef FREE
900 static void cleanup_uc_state(struct i915_gpu_state *error)
902 struct i915_error_uc *error_uc = &error->uc;
904 kfree(error_uc->guc_fw.path);
905 kfree(error_uc->huc_fw.path);
906 i915_error_object_free(error_uc->guc_log);
909 void __i915_gpu_state_free(struct kref *error_ref)
911 struct i915_gpu_state *error =
912 container_of(error_ref, typeof(*error), ref);
913 long i, j;
915 for (i = 0; i < ARRAY_SIZE(error->engine); i++) {
916 struct drm_i915_error_engine *ee = &error->engine[i];
918 for (j = 0; j < ee->user_bo_count; j++)
919 i915_error_object_free(ee->user_bo[j]);
920 kfree(ee->user_bo);
922 i915_error_object_free(ee->batchbuffer);
923 i915_error_object_free(ee->wa_batchbuffer);
924 i915_error_object_free(ee->ringbuffer);
925 i915_error_object_free(ee->hws_page);
926 i915_error_object_free(ee->ctx);
927 i915_error_object_free(ee->wa_ctx);
929 kfree(ee->requests);
930 if (!IS_ERR_OR_NULL(ee->waiters))
931 kfree(ee->waiters);
934 for (i = 0; i < ARRAY_SIZE(error->active_bo); i++)
935 kfree(error->active_bo[i]);
936 kfree(error->pinned_bo);
938 kfree(error->overlay);
939 kfree(error->display);
941 cleanup_params(error);
942 cleanup_uc_state(error);
944 kfree(error);
947 static struct drm_i915_error_object *
948 i915_error_object_create(struct drm_i915_private *i915,
949 struct i915_vma *vma)
951 struct i915_ggtt *ggtt = &i915->ggtt;
952 const u64 slot = ggtt->error_capture.start;
953 struct drm_i915_error_object *dst;
954 struct compress compress;
955 unsigned long num_pages;
956 struct sgt_iter iter;
957 dma_addr_t dma;
958 int ret;
960 if (!vma)
961 return NULL;
963 num_pages = min_t(u64, vma->size, vma->obj->base.size) >> PAGE_SHIFT;
964 num_pages = DIV_ROUND_UP(10 * num_pages, 8); /* worstcase zlib growth */
965 dst = kmalloc(sizeof(*dst) + num_pages * sizeof(u32 *),
966 GFP_ATOMIC | __GFP_NOWARN);
967 if (!dst)
968 return NULL;
970 dst->gtt_offset = vma->node.start;
971 dst->gtt_size = vma->node.size;
972 dst->num_pages = num_pages;
973 dst->page_count = 0;
974 dst->unused = 0;
976 if (!compress_init(&compress)) {
977 kfree(dst);
978 return NULL;
981 ret = -EINVAL;
982 for_each_sgt_dma(dma, iter, vma->pages) {
983 void __iomem *s;
985 ggtt->vm.insert_page(&ggtt->vm, dma, slot, I915_CACHE_NONE, 0);
987 s = io_mapping_map_atomic_wc(&ggtt->iomap, slot);
988 ret = compress_page(&compress, (void __force *)s, dst);
989 io_mapping_unmap_atomic(s);
990 if (ret)
991 break;
994 if (ret || compress_flush(&compress, dst)) {
995 while (dst->page_count--)
996 free_page((unsigned long)dst->pages[dst->page_count]);
997 kfree(dst);
998 dst = NULL;
1001 compress_fini(&compress, dst);
1002 ggtt->vm.clear_range(&ggtt->vm, slot, PAGE_SIZE);
1003 return dst;
1006 /* The error capture is special as tries to run underneath the normal
1007 * locking rules - so we use the raw version of the i915_gem_active lookup.
1009 static inline uint32_t
1010 __active_get_seqno(struct i915_gem_active *active)
1012 struct i915_request *request;
1014 request = __i915_gem_active_peek(active);
1015 return request ? request->global_seqno : 0;
1018 static inline int
1019 __active_get_engine_id(struct i915_gem_active *active)
1021 struct i915_request *request;
1023 request = __i915_gem_active_peek(active);
1024 return request ? request->engine->id : -1;
1027 static void capture_bo(struct drm_i915_error_buffer *err,
1028 struct i915_vma *vma)
1030 struct drm_i915_gem_object *obj = vma->obj;
1032 err->size = obj->base.size;
1033 err->name = obj->base.name;
1035 err->wseqno = __active_get_seqno(&obj->frontbuffer_write);
1036 err->engine = __active_get_engine_id(&obj->frontbuffer_write);
1038 err->gtt_offset = vma->node.start;
1039 err->read_domains = obj->read_domains;
1040 err->write_domain = obj->write_domain;
1041 err->fence_reg = vma->fence ? vma->fence->id : -1;
1042 err->tiling = i915_gem_object_get_tiling(obj);
1043 err->dirty = obj->mm.dirty;
1044 err->purgeable = obj->mm.madv != I915_MADV_WILLNEED;
1045 err->userptr = obj->userptr.mm != NULL;
1046 err->cache_level = obj->cache_level;
1049 static u32 capture_error_bo(struct drm_i915_error_buffer *err,
1050 int count, struct list_head *head,
1051 bool pinned_only)
1053 struct i915_vma *vma;
1054 int i = 0;
1056 list_for_each_entry(vma, head, vm_link) {
1057 if (!vma->obj)
1058 continue;
1060 if (pinned_only && !i915_vma_is_pinned(vma))
1061 continue;
1063 capture_bo(err++, vma);
1064 if (++i == count)
1065 break;
1068 return i;
1071 /* Generate a semi-unique error code. The code is not meant to have meaning, The
1072 * code's only purpose is to try to prevent false duplicated bug reports by
1073 * grossly estimating a GPU error state.
1075 * TODO Ideally, hashing the batchbuffer would be a very nice way to determine
1076 * the hang if we could strip the GTT offset information from it.
1078 * It's only a small step better than a random number in its current form.
1080 static uint32_t i915_error_generate_code(struct drm_i915_private *dev_priv,
1081 struct i915_gpu_state *error,
1082 int *engine_id)
1084 uint32_t error_code = 0;
1085 int i;
1087 /* IPEHR would be an ideal way to detect errors, as it's the gross
1088 * measure of "the command that hung." However, has some very common
1089 * synchronization commands which almost always appear in the case
1090 * strictly a client bug. Use instdone to differentiate those some.
1092 for (i = 0; i < I915_NUM_ENGINES; i++) {
1093 if (error->engine[i].hangcheck_stalled) {
1094 if (engine_id)
1095 *engine_id = i;
1097 return error->engine[i].ipehr ^
1098 error->engine[i].instdone.instdone;
1102 return error_code;
1105 static void gem_record_fences(struct i915_gpu_state *error)
1107 struct drm_i915_private *dev_priv = error->i915;
1108 int i;
1110 if (INTEL_GEN(dev_priv) >= 6) {
1111 for (i = 0; i < dev_priv->num_fence_regs; i++)
1112 error->fence[i] = I915_READ64(FENCE_REG_GEN6_LO(i));
1113 } else if (INTEL_GEN(dev_priv) >= 4) {
1114 for (i = 0; i < dev_priv->num_fence_regs; i++)
1115 error->fence[i] = I915_READ64(FENCE_REG_965_LO(i));
1116 } else {
1117 for (i = 0; i < dev_priv->num_fence_regs; i++)
1118 error->fence[i] = I915_READ(FENCE_REG(i));
1120 error->nfence = i;
1123 static void gen6_record_semaphore_state(struct intel_engine_cs *engine,
1124 struct drm_i915_error_engine *ee)
1126 struct drm_i915_private *dev_priv = engine->i915;
1128 ee->semaphore_mboxes[0] = I915_READ(RING_SYNC_0(engine->mmio_base));
1129 ee->semaphore_mboxes[1] = I915_READ(RING_SYNC_1(engine->mmio_base));
1130 if (HAS_VEBOX(dev_priv))
1131 ee->semaphore_mboxes[2] =
1132 I915_READ(RING_SYNC_2(engine->mmio_base));
1135 static void error_record_engine_waiters(struct intel_engine_cs *engine,
1136 struct drm_i915_error_engine *ee)
1138 struct intel_breadcrumbs *b = &engine->breadcrumbs;
1139 struct drm_i915_error_waiter *waiter;
1140 struct rb_node *rb;
1141 int count;
1143 ee->num_waiters = 0;
1144 ee->waiters = NULL;
1146 if (RB_EMPTY_ROOT(&b->waiters))
1147 return;
1149 if (!spin_trylock_irq(&b->rb_lock)) {
1150 ee->waiters = ERR_PTR(-EDEADLK);
1151 return;
1154 count = 0;
1155 for (rb = rb_first(&b->waiters); rb != NULL; rb = rb_next(rb))
1156 count++;
1157 spin_unlock_irq(&b->rb_lock);
1159 waiter = NULL;
1160 if (count)
1161 waiter = kmalloc_array(count,
1162 sizeof(struct drm_i915_error_waiter),
1163 GFP_ATOMIC);
1164 if (!waiter)
1165 return;
1167 if (!spin_trylock_irq(&b->rb_lock)) {
1168 kfree(waiter);
1169 ee->waiters = ERR_PTR(-EDEADLK);
1170 return;
1173 ee->waiters = waiter;
1174 for (rb = rb_first(&b->waiters); rb; rb = rb_next(rb)) {
1175 struct intel_wait *w = rb_entry(rb, typeof(*w), node);
1177 strcpy(waiter->comm, w->tsk->comm);
1178 waiter->pid = w->tsk->pid;
1179 waiter->seqno = w->seqno;
1180 waiter++;
1182 if (++ee->num_waiters == count)
1183 break;
1185 spin_unlock_irq(&b->rb_lock);
1188 static void error_record_engine_registers(struct i915_gpu_state *error,
1189 struct intel_engine_cs *engine,
1190 struct drm_i915_error_engine *ee)
1192 struct drm_i915_private *dev_priv = engine->i915;
1194 if (INTEL_GEN(dev_priv) >= 6) {
1195 ee->rc_psmi = I915_READ(RING_PSMI_CTL(engine->mmio_base));
1196 if (INTEL_GEN(dev_priv) >= 8) {
1197 ee->fault_reg = I915_READ(GEN8_RING_FAULT_REG);
1198 } else {
1199 gen6_record_semaphore_state(engine, ee);
1200 ee->fault_reg = I915_READ(RING_FAULT_REG(engine));
1204 if (INTEL_GEN(dev_priv) >= 4) {
1205 ee->faddr = I915_READ(RING_DMA_FADD(engine->mmio_base));
1206 ee->ipeir = I915_READ(RING_IPEIR(engine->mmio_base));
1207 ee->ipehr = I915_READ(RING_IPEHR(engine->mmio_base));
1208 ee->instps = I915_READ(RING_INSTPS(engine->mmio_base));
1209 ee->bbaddr = I915_READ(RING_BBADDR(engine->mmio_base));
1210 if (INTEL_GEN(dev_priv) >= 8) {
1211 ee->faddr |= (u64) I915_READ(RING_DMA_FADD_UDW(engine->mmio_base)) << 32;
1212 ee->bbaddr |= (u64) I915_READ(RING_BBADDR_UDW(engine->mmio_base)) << 32;
1214 ee->bbstate = I915_READ(RING_BBSTATE(engine->mmio_base));
1215 } else {
1216 ee->faddr = I915_READ(DMA_FADD_I8XX);
1217 ee->ipeir = I915_READ(IPEIR);
1218 ee->ipehr = I915_READ(IPEHR);
1221 intel_engine_get_instdone(engine, &ee->instdone);
1223 ee->waiting = intel_engine_has_waiter(engine);
1224 ee->instpm = I915_READ(RING_INSTPM(engine->mmio_base));
1225 ee->acthd = intel_engine_get_active_head(engine);
1226 ee->seqno = intel_engine_get_seqno(engine);
1227 ee->last_seqno = intel_engine_last_submit(engine);
1228 ee->start = I915_READ_START(engine);
1229 ee->head = I915_READ_HEAD(engine);
1230 ee->tail = I915_READ_TAIL(engine);
1231 ee->ctl = I915_READ_CTL(engine);
1232 if (INTEL_GEN(dev_priv) > 2)
1233 ee->mode = I915_READ_MODE(engine);
1235 if (!HWS_NEEDS_PHYSICAL(dev_priv)) {
1236 i915_reg_t mmio;
1238 if (IS_GEN7(dev_priv)) {
1239 switch (engine->id) {
1240 default:
1241 case RCS:
1242 mmio = RENDER_HWS_PGA_GEN7;
1243 break;
1244 case BCS:
1245 mmio = BLT_HWS_PGA_GEN7;
1246 break;
1247 case VCS:
1248 mmio = BSD_HWS_PGA_GEN7;
1249 break;
1250 case VECS:
1251 mmio = VEBOX_HWS_PGA_GEN7;
1252 break;
1254 } else if (IS_GEN6(engine->i915)) {
1255 mmio = RING_HWS_PGA_GEN6(engine->mmio_base);
1256 } else {
1257 /* XXX: gen8 returns to sanity */
1258 mmio = RING_HWS_PGA(engine->mmio_base);
1261 ee->hws = I915_READ(mmio);
1264 ee->idle = intel_engine_is_idle(engine);
1265 ee->hangcheck_timestamp = engine->hangcheck.action_timestamp;
1266 ee->hangcheck_action = engine->hangcheck.action;
1267 ee->hangcheck_stalled = engine->hangcheck.stalled;
1268 ee->reset_count = i915_reset_engine_count(&dev_priv->gpu_error,
1269 engine);
1271 if (USES_PPGTT(dev_priv)) {
1272 int i;
1274 ee->vm_info.gfx_mode = I915_READ(RING_MODE_GEN7(engine));
1276 if (IS_GEN6(dev_priv))
1277 ee->vm_info.pp_dir_base =
1278 I915_READ(RING_PP_DIR_BASE_READ(engine));
1279 else if (IS_GEN7(dev_priv))
1280 ee->vm_info.pp_dir_base =
1281 I915_READ(RING_PP_DIR_BASE(engine));
1282 else if (INTEL_GEN(dev_priv) >= 8)
1283 for (i = 0; i < 4; i++) {
1284 ee->vm_info.pdp[i] =
1285 I915_READ(GEN8_RING_PDP_UDW(engine, i));
1286 ee->vm_info.pdp[i] <<= 32;
1287 ee->vm_info.pdp[i] |=
1288 I915_READ(GEN8_RING_PDP_LDW(engine, i));
1293 static void record_request(struct i915_request *request,
1294 struct drm_i915_error_request *erq)
1296 struct i915_gem_context *ctx = request->gem_context;
1298 erq->context = ctx->hw_id;
1299 erq->sched_attr = request->sched.attr;
1300 erq->ban_score = atomic_read(&ctx->ban_score);
1301 erq->seqno = request->global_seqno;
1302 erq->jiffies = request->emitted_jiffies;
1303 erq->start = i915_ggtt_offset(request->ring->vma);
1304 erq->head = request->head;
1305 erq->tail = request->tail;
1307 rcu_read_lock();
1308 erq->pid = ctx->pid ? pid_nr(ctx->pid) : 0;
1309 rcu_read_unlock();
1312 static void engine_record_requests(struct intel_engine_cs *engine,
1313 struct i915_request *first,
1314 struct drm_i915_error_engine *ee)
1316 struct i915_request *request;
1317 int count;
1319 count = 0;
1320 request = first;
1321 list_for_each_entry_from(request, &engine->timeline.requests, link)
1322 count++;
1323 if (!count)
1324 return;
1326 ee->requests = kcalloc(count, sizeof(*ee->requests), GFP_ATOMIC);
1327 if (!ee->requests)
1328 return;
1330 ee->num_requests = count;
1332 count = 0;
1333 request = first;
1334 list_for_each_entry_from(request, &engine->timeline.requests, link) {
1335 if (count >= ee->num_requests) {
1337 * If the ring request list was changed in
1338 * between the point where the error request
1339 * list was created and dimensioned and this
1340 * point then just exit early to avoid crashes.
1342 * We don't need to communicate that the
1343 * request list changed state during error
1344 * state capture and that the error state is
1345 * slightly incorrect as a consequence since we
1346 * are typically only interested in the request
1347 * list state at the point of error state
1348 * capture, not in any changes happening during
1349 * the capture.
1351 break;
1354 record_request(request, &ee->requests[count++]);
1356 ee->num_requests = count;
1359 static void error_record_engine_execlists(struct intel_engine_cs *engine,
1360 struct drm_i915_error_engine *ee)
1362 const struct intel_engine_execlists * const execlists = &engine->execlists;
1363 unsigned int n;
1365 for (n = 0; n < execlists_num_ports(execlists); n++) {
1366 struct i915_request *rq = port_request(&execlists->port[n]);
1368 if (!rq)
1369 break;
1371 record_request(rq, &ee->execlist[n]);
1374 ee->num_ports = n;
1377 static void record_context(struct drm_i915_error_context *e,
1378 struct i915_gem_context *ctx)
1380 if (ctx->pid) {
1381 struct task_struct *task;
1383 rcu_read_lock();
1384 task = pid_task(ctx->pid, PIDTYPE_PID);
1385 if (task) {
1386 strcpy(e->comm, task->comm);
1387 e->pid = task->pid;
1389 rcu_read_unlock();
1392 e->handle = ctx->user_handle;
1393 e->hw_id = ctx->hw_id;
1394 e->sched_attr = ctx->sched;
1395 e->ban_score = atomic_read(&ctx->ban_score);
1396 e->bannable = i915_gem_context_is_bannable(ctx);
1397 e->guilty = atomic_read(&ctx->guilty_count);
1398 e->active = atomic_read(&ctx->active_count);
1401 static void request_record_user_bo(struct i915_request *request,
1402 struct drm_i915_error_engine *ee)
1404 struct i915_capture_list *c;
1405 struct drm_i915_error_object **bo;
1406 long count;
1408 count = 0;
1409 for (c = request->capture_list; c; c = c->next)
1410 count++;
1412 bo = NULL;
1413 if (count)
1414 bo = kcalloc(count, sizeof(*bo), GFP_ATOMIC);
1415 if (!bo)
1416 return;
1418 count = 0;
1419 for (c = request->capture_list; c; c = c->next) {
1420 bo[count] = i915_error_object_create(request->i915, c->vma);
1421 if (!bo[count])
1422 break;
1423 count++;
1426 ee->user_bo = bo;
1427 ee->user_bo_count = count;
1430 static struct drm_i915_error_object *
1431 capture_object(struct drm_i915_private *dev_priv,
1432 struct drm_i915_gem_object *obj)
1434 if (obj && i915_gem_object_has_pages(obj)) {
1435 struct i915_vma fake = {
1436 .node = { .start = U64_MAX, .size = obj->base.size },
1437 .size = obj->base.size,
1438 .pages = obj->mm.pages,
1439 .obj = obj,
1442 return i915_error_object_create(dev_priv, &fake);
1443 } else {
1444 return NULL;
1448 static void gem_record_rings(struct i915_gpu_state *error)
1450 struct drm_i915_private *i915 = error->i915;
1451 struct i915_ggtt *ggtt = &i915->ggtt;
1452 int i;
1454 for (i = 0; i < I915_NUM_ENGINES; i++) {
1455 struct intel_engine_cs *engine = i915->engine[i];
1456 struct drm_i915_error_engine *ee = &error->engine[i];
1457 struct i915_request *request;
1459 ee->engine_id = -1;
1461 if (!engine)
1462 continue;
1464 ee->engine_id = i;
1466 error_record_engine_registers(error, engine, ee);
1467 error_record_engine_waiters(engine, ee);
1468 error_record_engine_execlists(engine, ee);
1470 request = i915_gem_find_active_request(engine);
1471 if (request) {
1472 struct i915_gem_context *ctx = request->gem_context;
1473 struct intel_ring *ring;
1475 ee->vm = ctx->ppgtt ? &ctx->ppgtt->vm : &ggtt->vm;
1477 record_context(&ee->context, ctx);
1479 /* We need to copy these to an anonymous buffer
1480 * as the simplest method to avoid being overwritten
1481 * by userspace.
1483 ee->batchbuffer =
1484 i915_error_object_create(i915, request->batch);
1486 if (HAS_BROKEN_CS_TLB(i915))
1487 ee->wa_batchbuffer =
1488 i915_error_object_create(i915,
1489 engine->scratch);
1490 request_record_user_bo(request, ee);
1492 ee->ctx =
1493 i915_error_object_create(i915,
1494 request->hw_context->state);
1496 error->simulated |=
1497 i915_gem_context_no_error_capture(ctx);
1499 ee->rq_head = request->head;
1500 ee->rq_post = request->postfix;
1501 ee->rq_tail = request->tail;
1503 ring = request->ring;
1504 ee->cpu_ring_head = ring->head;
1505 ee->cpu_ring_tail = ring->tail;
1506 ee->ringbuffer =
1507 i915_error_object_create(i915, ring->vma);
1509 engine_record_requests(engine, request, ee);
1512 ee->hws_page =
1513 i915_error_object_create(i915,
1514 engine->status_page.vma);
1516 ee->wa_ctx = i915_error_object_create(i915, engine->wa_ctx.vma);
1518 ee->default_state = capture_object(i915, engine->default_state);
1522 static void gem_capture_vm(struct i915_gpu_state *error,
1523 struct i915_address_space *vm,
1524 int idx)
1526 struct drm_i915_error_buffer *active_bo;
1527 struct i915_vma *vma;
1528 int count;
1530 count = 0;
1531 list_for_each_entry(vma, &vm->active_list, vm_link)
1532 count++;
1534 active_bo = NULL;
1535 if (count)
1536 active_bo = kcalloc(count, sizeof(*active_bo), GFP_ATOMIC);
1537 if (active_bo)
1538 count = capture_error_bo(active_bo, count, &vm->active_list, false);
1539 else
1540 count = 0;
1542 error->active_vm[idx] = vm;
1543 error->active_bo[idx] = active_bo;
1544 error->active_bo_count[idx] = count;
1547 static void capture_active_buffers(struct i915_gpu_state *error)
1549 int cnt = 0, i, j;
1551 BUILD_BUG_ON(ARRAY_SIZE(error->engine) > ARRAY_SIZE(error->active_bo));
1552 BUILD_BUG_ON(ARRAY_SIZE(error->active_bo) != ARRAY_SIZE(error->active_vm));
1553 BUILD_BUG_ON(ARRAY_SIZE(error->active_bo) != ARRAY_SIZE(error->active_bo_count));
1555 /* Scan each engine looking for unique active contexts/vm */
1556 for (i = 0; i < ARRAY_SIZE(error->engine); i++) {
1557 struct drm_i915_error_engine *ee = &error->engine[i];
1558 bool found;
1560 if (!ee->vm)
1561 continue;
1563 found = false;
1564 for (j = 0; j < i && !found; j++)
1565 found = error->engine[j].vm == ee->vm;
1566 if (!found)
1567 gem_capture_vm(error, ee->vm, cnt++);
1571 static void capture_pinned_buffers(struct i915_gpu_state *error)
1573 struct i915_address_space *vm = &error->i915->ggtt.vm;
1574 struct drm_i915_error_buffer *bo;
1575 struct i915_vma *vma;
1576 int count_inactive, count_active;
1578 count_inactive = 0;
1579 list_for_each_entry(vma, &vm->inactive_list, vm_link)
1580 count_inactive++;
1582 count_active = 0;
1583 list_for_each_entry(vma, &vm->active_list, vm_link)
1584 count_active++;
1586 bo = NULL;
1587 if (count_inactive + count_active)
1588 bo = kcalloc(count_inactive + count_active,
1589 sizeof(*bo), GFP_ATOMIC);
1590 if (!bo)
1591 return;
1593 count_inactive = capture_error_bo(bo, count_inactive,
1594 &vm->active_list, true);
1595 count_active = capture_error_bo(bo + count_inactive, count_active,
1596 &vm->inactive_list, true);
1597 error->pinned_bo_count = count_inactive + count_active;
1598 error->pinned_bo = bo;
1601 static void capture_uc_state(struct i915_gpu_state *error)
1603 struct drm_i915_private *i915 = error->i915;
1604 struct i915_error_uc *error_uc = &error->uc;
1606 /* Capturing uC state won't be useful if there is no GuC */
1607 if (!error->device_info.has_guc)
1608 return;
1610 error_uc->guc_fw = i915->guc.fw;
1611 error_uc->huc_fw = i915->huc.fw;
1613 /* Non-default firmware paths will be specified by the modparam.
1614 * As modparams are generally accesible from the userspace make
1615 * explicit copies of the firmware paths.
1617 error_uc->guc_fw.path = kstrdup(i915->guc.fw.path, GFP_ATOMIC);
1618 error_uc->huc_fw.path = kstrdup(i915->huc.fw.path, GFP_ATOMIC);
1619 error_uc->guc_log = i915_error_object_create(i915, i915->guc.log.vma);
1622 /* Capture all registers which don't fit into another category. */
1623 static void capture_reg_state(struct i915_gpu_state *error)
1625 struct drm_i915_private *dev_priv = error->i915;
1626 int i;
1628 /* General organization
1629 * 1. Registers specific to a single generation
1630 * 2. Registers which belong to multiple generations
1631 * 3. Feature specific registers.
1632 * 4. Everything else
1633 * Please try to follow the order.
1636 /* 1: Registers specific to a single generation */
1637 if (IS_VALLEYVIEW(dev_priv)) {
1638 error->gtier[0] = I915_READ(GTIER);
1639 error->ier = I915_READ(VLV_IER);
1640 error->forcewake = I915_READ_FW(FORCEWAKE_VLV);
1643 if (IS_GEN7(dev_priv))
1644 error->err_int = I915_READ(GEN7_ERR_INT);
1646 if (INTEL_GEN(dev_priv) >= 8) {
1647 error->fault_data0 = I915_READ(GEN8_FAULT_TLB_DATA0);
1648 error->fault_data1 = I915_READ(GEN8_FAULT_TLB_DATA1);
1651 if (IS_GEN6(dev_priv)) {
1652 error->forcewake = I915_READ_FW(FORCEWAKE);
1653 error->gab_ctl = I915_READ(GAB_CTL);
1654 error->gfx_mode = I915_READ(GFX_MODE);
1657 /* 2: Registers which belong to multiple generations */
1658 if (INTEL_GEN(dev_priv) >= 7)
1659 error->forcewake = I915_READ_FW(FORCEWAKE_MT);
1661 if (INTEL_GEN(dev_priv) >= 6) {
1662 error->derrmr = I915_READ(DERRMR);
1663 error->error = I915_READ(ERROR_GEN6);
1664 error->done_reg = I915_READ(DONE_REG);
1667 if (INTEL_GEN(dev_priv) >= 5)
1668 error->ccid = I915_READ(CCID);
1670 /* 3: Feature specific registers */
1671 if (IS_GEN6(dev_priv) || IS_GEN7(dev_priv)) {
1672 error->gam_ecochk = I915_READ(GAM_ECOCHK);
1673 error->gac_eco = I915_READ(GAC_ECO_BITS);
1676 /* 4: Everything else */
1677 if (INTEL_GEN(dev_priv) >= 11) {
1678 error->ier = I915_READ(GEN8_DE_MISC_IER);
1679 error->gtier[0] = I915_READ(GEN11_RENDER_COPY_INTR_ENABLE);
1680 error->gtier[1] = I915_READ(GEN11_VCS_VECS_INTR_ENABLE);
1681 error->gtier[2] = I915_READ(GEN11_GUC_SG_INTR_ENABLE);
1682 error->gtier[3] = I915_READ(GEN11_GPM_WGBOXPERF_INTR_ENABLE);
1683 error->gtier[4] = I915_READ(GEN11_CRYPTO_RSVD_INTR_ENABLE);
1684 error->gtier[5] = I915_READ(GEN11_GUNIT_CSME_INTR_ENABLE);
1685 error->ngtier = 6;
1686 } else if (INTEL_GEN(dev_priv) >= 8) {
1687 error->ier = I915_READ(GEN8_DE_MISC_IER);
1688 for (i = 0; i < 4; i++)
1689 error->gtier[i] = I915_READ(GEN8_GT_IER(i));
1690 error->ngtier = 4;
1691 } else if (HAS_PCH_SPLIT(dev_priv)) {
1692 error->ier = I915_READ(DEIER);
1693 error->gtier[0] = I915_READ(GTIER);
1694 error->ngtier = 1;
1695 } else if (IS_GEN2(dev_priv)) {
1696 error->ier = I915_READ16(IER);
1697 } else if (!IS_VALLEYVIEW(dev_priv)) {
1698 error->ier = I915_READ(IER);
1700 error->eir = I915_READ(EIR);
1701 error->pgtbl_er = I915_READ(PGTBL_ER);
1704 static void i915_error_capture_msg(struct drm_i915_private *dev_priv,
1705 struct i915_gpu_state *error,
1706 u32 engine_mask,
1707 const char *error_msg)
1709 u32 ecode;
1710 int engine_id = -1, len;
1712 ecode = i915_error_generate_code(dev_priv, error, &engine_id);
1714 len = scnprintf(error->error_msg, sizeof(error->error_msg),
1715 "GPU HANG: ecode %d:%d:0x%08x",
1716 INTEL_GEN(dev_priv), engine_id, ecode);
1718 if (engine_id != -1 && error->engine[engine_id].context.pid)
1719 len += scnprintf(error->error_msg + len,
1720 sizeof(error->error_msg) - len,
1721 ", in %s [%d]",
1722 error->engine[engine_id].context.comm,
1723 error->engine[engine_id].context.pid);
1725 scnprintf(error->error_msg + len, sizeof(error->error_msg) - len,
1726 ", reason: %s, action: %s",
1727 error_msg,
1728 engine_mask ? "reset" : "continue");
1731 static void capture_gen_state(struct i915_gpu_state *error)
1733 struct drm_i915_private *i915 = error->i915;
1735 error->awake = i915->gt.awake;
1736 error->wakelock = atomic_read(&i915->runtime_pm.wakeref_count);
1737 error->suspended = i915->runtime_pm.suspended;
1739 error->iommu = -1;
1740 #ifdef CONFIG_INTEL_IOMMU
1741 error->iommu = intel_iommu_gfx_mapped;
1742 #endif
1743 error->reset_count = i915_reset_count(&i915->gpu_error);
1744 error->suspend_count = i915->suspend_count;
1746 memcpy(&error->device_info,
1747 INTEL_INFO(i915),
1748 sizeof(error->device_info));
1749 error->driver_caps = i915->caps;
1752 static __always_inline void dup_param(const char *type, void *x)
1754 if (!__builtin_strcmp(type, "char *"))
1755 *(void **)x = kstrdup(*(void **)x, GFP_ATOMIC);
1758 static void capture_params(struct i915_gpu_state *error)
1760 error->params = i915_modparams;
1761 #define DUP(T, x, ...) dup_param(#T, &error->params.x);
1762 I915_PARAMS_FOR_EACH(DUP);
1763 #undef DUP
1766 static unsigned long capture_find_epoch(const struct i915_gpu_state *error)
1768 unsigned long epoch = error->capture;
1769 int i;
1771 for (i = 0; i < ARRAY_SIZE(error->engine); i++) {
1772 const struct drm_i915_error_engine *ee = &error->engine[i];
1774 if (ee->hangcheck_stalled &&
1775 time_before(ee->hangcheck_timestamp, epoch))
1776 epoch = ee->hangcheck_timestamp;
1779 return epoch;
1782 static int capture(void *data)
1784 struct i915_gpu_state *error = data;
1786 error->time = ktime_get_real();
1787 error->boottime = ktime_get_boottime();
1788 error->uptime = ktime_sub(ktime_get(),
1789 error->i915->gt.last_init_time);
1790 error->capture = jiffies;
1792 capture_params(error);
1793 capture_gen_state(error);
1794 capture_uc_state(error);
1795 capture_reg_state(error);
1796 gem_record_fences(error);
1797 gem_record_rings(error);
1798 capture_active_buffers(error);
1799 capture_pinned_buffers(error);
1801 error->overlay = intel_overlay_capture_error_state(error->i915);
1802 error->display = intel_display_capture_error_state(error->i915);
1804 error->epoch = capture_find_epoch(error);
1806 return 0;
1809 #define DAY_AS_SECONDS(x) (24 * 60 * 60 * (x))
1811 struct i915_gpu_state *
1812 i915_capture_gpu_state(struct drm_i915_private *i915)
1814 struct i915_gpu_state *error;
1816 error = kzalloc(sizeof(*error), GFP_ATOMIC);
1817 if (!error)
1818 return NULL;
1820 kref_init(&error->ref);
1821 error->i915 = i915;
1823 stop_machine(capture, error, NULL);
1825 return error;
1829 * i915_capture_error_state - capture an error record for later analysis
1830 * @i915: i915 device
1831 * @engine_mask: the mask of engines triggering the hang
1832 * @error_msg: a message to insert into the error capture header
1834 * Should be called when an error is detected (either a hang or an error
1835 * interrupt) to capture error state from the time of the error. Fills
1836 * out a structure which becomes available in debugfs for user level tools
1837 * to pick up.
1839 void i915_capture_error_state(struct drm_i915_private *i915,
1840 u32 engine_mask,
1841 const char *error_msg)
1843 static bool warned;
1844 struct i915_gpu_state *error;
1845 unsigned long flags;
1847 if (!i915_modparams.error_capture)
1848 return;
1850 if (READ_ONCE(i915->gpu_error.first_error))
1851 return;
1853 error = i915_capture_gpu_state(i915);
1854 if (!error) {
1855 DRM_DEBUG_DRIVER("out of memory, not capturing error state\n");
1856 return;
1859 i915_error_capture_msg(i915, error, engine_mask, error_msg);
1860 DRM_INFO("%s\n", error->error_msg);
1862 if (!error->simulated) {
1863 spin_lock_irqsave(&i915->gpu_error.lock, flags);
1864 if (!i915->gpu_error.first_error) {
1865 i915->gpu_error.first_error = error;
1866 error = NULL;
1868 spin_unlock_irqrestore(&i915->gpu_error.lock, flags);
1871 if (error) {
1872 __i915_gpu_state_free(&error->ref);
1873 return;
1876 if (!warned &&
1877 ktime_get_real_seconds() - DRIVER_TIMESTAMP < DAY_AS_SECONDS(180)) {
1878 DRM_INFO("GPU hangs can indicate a bug anywhere in the entire gfx stack, including userspace.\n");
1879 DRM_INFO("Please file a _new_ bug report on bugs.freedesktop.org against DRI -> DRM/Intel\n");
1880 DRM_INFO("drm/i915 developers can then reassign to the right component if it's not a kernel issue.\n");
1881 DRM_INFO("The gpu crash dump is required to analyze gpu hangs, so please always attach it.\n");
1882 DRM_INFO("GPU crash dump saved to /sys/class/drm/card%d/error\n",
1883 i915->drm.primary->index);
1884 warned = true;
1888 struct i915_gpu_state *
1889 i915_first_error_state(struct drm_i915_private *i915)
1891 struct i915_gpu_state *error;
1893 spin_lock_irq(&i915->gpu_error.lock);
1894 error = i915->gpu_error.first_error;
1895 if (error)
1896 i915_gpu_state_get(error);
1897 spin_unlock_irq(&i915->gpu_error.lock);
1899 return error;
1902 void i915_reset_error_state(struct drm_i915_private *i915)
1904 struct i915_gpu_state *error;
1906 spin_lock_irq(&i915->gpu_error.lock);
1907 error = i915->gpu_error.first_error;
1908 i915->gpu_error.first_error = NULL;
1909 spin_unlock_irq(&i915->gpu_error.lock);
1911 i915_gpu_state_put(error);