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
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
24 * Eric Anholt <eric@anholt.net>
25 * Keith Packard <keithp@keithp.com>
26 * Mika Kuoppala <mika.kuoppala@intel.com>
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>
37 #include <drm/drm_print.h>
39 #include "display/intel_atomic.h"
40 #include "display/intel_overlay.h"
42 #include "gem/i915_gem_context.h"
45 #include "i915_gpu_error.h"
46 #include "i915_memcpy.h"
47 #include "i915_scatterlist.h"
48 #include "intel_csr.h"
50 #define ALLOW_FAIL (GFP_KERNEL | __GFP_RETRY_MAYFAIL | __GFP_NOWARN)
51 #define ATOMIC_MAYFAIL (GFP_ATOMIC | __GFP_NOWARN)
53 static void __sg_set_buf(struct scatterlist *sg,
54 void *addr, unsigned int len, loff_t it)
56 sg->page_link = (unsigned long)virt_to_page(addr);
57 sg->offset = offset_in_page(addr);
62 static bool __i915_error_grow(struct drm_i915_error_state_buf *e, size_t len)
67 if (e->bytes + len + 1 <= e->size)
71 __sg_set_buf(e->cur++, e->buf, e->bytes, e->iter);
77 if (e->cur == e->end) {
78 struct scatterlist *sgl;
80 sgl = (typeof(sgl))__get_free_page(ALLOW_FAIL);
90 (unsigned long)sgl | SG_CHAIN;
96 e->end = sgl + SG_MAX_SINGLE_ALLOC - 1;
99 e->size = ALIGN(len + 1, SZ_64K);
100 e->buf = kmalloc(e->size, ALLOW_FAIL);
102 e->size = PAGE_ALIGN(len + 1);
103 e->buf = kmalloc(e->size, GFP_KERNEL);
114 static void i915_error_vprintf(struct drm_i915_error_state_buf *e,
115 const char *fmt, va_list args)
124 len = vsnprintf(NULL, 0, fmt, ap);
131 if (!__i915_error_grow(e, len))
134 GEM_BUG_ON(e->bytes >= e->size);
135 len = vscnprintf(e->buf + e->bytes, e->size - e->bytes, fmt, args);
143 static void i915_error_puts(struct drm_i915_error_state_buf *e, const char *str)
151 if (!__i915_error_grow(e, len))
154 GEM_BUG_ON(e->bytes + len > e->size);
155 memcpy(e->buf + e->bytes, str, len);
159 #define err_printf(e, ...) i915_error_printf(e, __VA_ARGS__)
160 #define err_puts(e, s) i915_error_puts(e, s)
162 static void __i915_printfn_error(struct drm_printer *p, struct va_format *vaf)
164 i915_error_vprintf(p->arg, vaf->fmt, *vaf->va);
167 static inline struct drm_printer
168 i915_error_printer(struct drm_i915_error_state_buf *e)
170 struct drm_printer p = {
171 .printfn = __i915_printfn_error,
177 /* single threaded page allocator with a reserved stash for emergencies */
178 static void pool_fini(struct pagevec *pv)
183 static int pool_refill(struct pagevec *pv, gfp_t gfp)
185 while (pagevec_space(pv)) {
198 static int pool_init(struct pagevec *pv, gfp_t gfp)
204 err = pool_refill(pv, gfp);
211 static void *pool_alloc(struct pagevec *pv, gfp_t gfp)
216 if (!p && pagevec_count(pv))
217 p = pv->pages[--pv->nr];
219 return p ? page_address(p) : NULL;
222 static void pool_free(struct pagevec *pv, void *addr)
224 struct page *p = virt_to_page(addr);
226 if (pagevec_space(pv))
232 #ifdef CONFIG_DRM_I915_COMPRESS_ERROR
236 struct z_stream_s zstream;
240 static bool compress_init(struct compress *c)
242 struct z_stream_s *zstream = &c->zstream;
244 if (pool_init(&c->pool, ALLOW_FAIL))
248 kmalloc(zlib_deflate_workspacesize(MAX_WBITS, MAX_MEM_LEVEL),
250 if (!zstream->workspace) {
256 if (i915_has_memcpy_from_wc())
257 c->tmp = pool_alloc(&c->pool, ALLOW_FAIL);
262 static bool compress_start(struct compress *c)
264 struct z_stream_s *zstream = &c->zstream;
265 void *workspace = zstream->workspace;
267 memset(zstream, 0, sizeof(*zstream));
268 zstream->workspace = workspace;
270 return zlib_deflateInit(zstream, Z_DEFAULT_COMPRESSION) == Z_OK;
273 static void *compress_next_page(struct compress *c,
274 struct drm_i915_error_object *dst)
278 if (dst->page_count >= dst->num_pages)
279 return ERR_PTR(-ENOSPC);
281 page = pool_alloc(&c->pool, ALLOW_FAIL);
283 return ERR_PTR(-ENOMEM);
285 return dst->pages[dst->page_count++] = page;
288 static int compress_page(struct compress *c,
290 struct drm_i915_error_object *dst)
292 struct z_stream_s *zstream = &c->zstream;
294 zstream->next_in = src;
295 if (c->tmp && i915_memcpy_from_wc(c->tmp, src, PAGE_SIZE))
296 zstream->next_in = c->tmp;
297 zstream->avail_in = PAGE_SIZE;
300 if (zstream->avail_out == 0) {
301 zstream->next_out = compress_next_page(c, dst);
302 if (IS_ERR(zstream->next_out))
303 return PTR_ERR(zstream->next_out);
305 zstream->avail_out = PAGE_SIZE;
308 if (zlib_deflate(zstream, Z_NO_FLUSH) != Z_OK)
310 } while (zstream->avail_in);
312 /* Fallback to uncompressed if we increase size? */
313 if (0 && zstream->total_out > zstream->total_in)
319 static int compress_flush(struct compress *c,
320 struct drm_i915_error_object *dst)
322 struct z_stream_s *zstream = &c->zstream;
325 switch (zlib_deflate(zstream, Z_FINISH)) {
326 case Z_OK: /* more space requested */
327 zstream->next_out = compress_next_page(c, dst);
328 if (IS_ERR(zstream->next_out))
329 return PTR_ERR(zstream->next_out);
331 zstream->avail_out = PAGE_SIZE;
337 default: /* any error */
343 memset(zstream->next_out, 0, zstream->avail_out);
344 dst->unused = zstream->avail_out;
348 static void compress_finish(struct compress *c)
350 zlib_deflateEnd(&c->zstream);
353 static void compress_fini(struct compress *c)
355 kfree(c->zstream.workspace);
357 pool_free(&c->pool, c->tmp);
361 static void err_compression_marker(struct drm_i915_error_state_buf *m)
372 static bool compress_init(struct compress *c)
374 return pool_init(&c->pool, ALLOW_FAIL) == 0;
377 static bool compress_start(struct compress *c)
382 static int compress_page(struct compress *c,
384 struct drm_i915_error_object *dst)
388 ptr = pool_alloc(&c->pool, ALLOW_FAIL);
392 if (!i915_memcpy_from_wc(ptr, src, PAGE_SIZE))
393 memcpy(ptr, src, PAGE_SIZE);
394 dst->pages[dst->page_count++] = ptr;
399 static int compress_flush(struct compress *c,
400 struct drm_i915_error_object *dst)
405 static void compress_finish(struct compress *c)
409 static void compress_fini(struct compress *c)
414 static void err_compression_marker(struct drm_i915_error_state_buf *m)
421 static void error_print_instdone(struct drm_i915_error_state_buf *m,
422 const struct drm_i915_error_engine *ee)
424 const struct sseu_dev_info *sseu = &RUNTIME_INFO(m->i915)->sseu;
428 err_printf(m, " INSTDONE: 0x%08x\n",
429 ee->instdone.instdone);
431 if (ee->engine->class != RENDER_CLASS || INTEL_GEN(m->i915) <= 3)
434 err_printf(m, " SC_INSTDONE: 0x%08x\n",
435 ee->instdone.slice_common);
437 if (INTEL_GEN(m->i915) <= 6)
440 for_each_instdone_slice_subslice(m->i915, sseu, slice, subslice)
441 err_printf(m, " SAMPLER_INSTDONE[%d][%d]: 0x%08x\n",
443 ee->instdone.sampler[slice][subslice]);
445 for_each_instdone_slice_subslice(m->i915, sseu, slice, subslice)
446 err_printf(m, " ROW_INSTDONE[%d][%d]: 0x%08x\n",
448 ee->instdone.row[slice][subslice]);
451 static void error_print_request(struct drm_i915_error_state_buf *m,
453 const struct drm_i915_error_request *erq,
454 const unsigned long epoch)
459 err_printf(m, "%s pid %d, seqno %8x:%08x%s%s, prio %d, emitted %dms, start %08x, head %08x, tail %08x\n",
460 prefix, erq->pid, erq->context, erq->seqno,
461 test_bit(DMA_FENCE_FLAG_SIGNALED_BIT,
462 &erq->flags) ? "!" : "",
463 test_bit(DMA_FENCE_FLAG_ENABLE_SIGNAL_BIT,
464 &erq->flags) ? "+" : "",
465 erq->sched_attr.priority,
466 jiffies_to_msecs(erq->jiffies - epoch),
467 erq->start, erq->head, erq->tail);
470 static void error_print_context(struct drm_i915_error_state_buf *m,
472 const struct drm_i915_error_context *ctx)
474 err_printf(m, "%s%s[%d] hw_id %d, prio %d, guilty %d active %d\n",
475 header, ctx->comm, ctx->pid, ctx->hw_id,
476 ctx->sched_attr.priority, ctx->guilty, ctx->active);
479 static void error_print_engine(struct drm_i915_error_state_buf *m,
480 const struct drm_i915_error_engine *ee,
481 const unsigned long epoch)
485 err_printf(m, "%s command stream:\n", ee->engine->name);
486 err_printf(m, " IDLE?: %s\n", yesno(ee->idle));
487 err_printf(m, " START: 0x%08x\n", ee->start);
488 err_printf(m, " HEAD: 0x%08x [0x%08x]\n", ee->head, ee->rq_head);
489 err_printf(m, " TAIL: 0x%08x [0x%08x, 0x%08x]\n",
490 ee->tail, ee->rq_post, ee->rq_tail);
491 err_printf(m, " CTL: 0x%08x\n", ee->ctl);
492 err_printf(m, " MODE: 0x%08x\n", ee->mode);
493 err_printf(m, " HWS: 0x%08x\n", ee->hws);
494 err_printf(m, " ACTHD: 0x%08x %08x\n",
495 (u32)(ee->acthd>>32), (u32)ee->acthd);
496 err_printf(m, " IPEIR: 0x%08x\n", ee->ipeir);
497 err_printf(m, " IPEHR: 0x%08x\n", ee->ipehr);
499 error_print_instdone(m, ee);
501 if (ee->batchbuffer) {
502 u64 start = ee->batchbuffer->gtt_offset;
503 u64 end = start + ee->batchbuffer->gtt_size;
505 err_printf(m, " batch: [0x%08x_%08x, 0x%08x_%08x]\n",
506 upper_32_bits(start), lower_32_bits(start),
507 upper_32_bits(end), lower_32_bits(end));
509 if (INTEL_GEN(m->i915) >= 4) {
510 err_printf(m, " BBADDR: 0x%08x_%08x\n",
511 (u32)(ee->bbaddr>>32), (u32)ee->bbaddr);
512 err_printf(m, " BB_STATE: 0x%08x\n", ee->bbstate);
513 err_printf(m, " INSTPS: 0x%08x\n", ee->instps);
515 err_printf(m, " INSTPM: 0x%08x\n", ee->instpm);
516 err_printf(m, " FADDR: 0x%08x %08x\n", upper_32_bits(ee->faddr),
517 lower_32_bits(ee->faddr));
518 if (INTEL_GEN(m->i915) >= 6) {
519 err_printf(m, " RC PSMI: 0x%08x\n", ee->rc_psmi);
520 err_printf(m, " FAULT_REG: 0x%08x\n", ee->fault_reg);
522 if (HAS_PPGTT(m->i915)) {
523 err_printf(m, " GFX_MODE: 0x%08x\n", ee->vm_info.gfx_mode);
525 if (INTEL_GEN(m->i915) >= 8) {
527 for (i = 0; i < 4; i++)
528 err_printf(m, " PDP%d: 0x%016llx\n",
529 i, ee->vm_info.pdp[i]);
531 err_printf(m, " PP_DIR_BASE: 0x%08x\n",
532 ee->vm_info.pp_dir_base);
535 err_printf(m, " ring->head: 0x%08x\n", ee->cpu_ring_head);
536 err_printf(m, " ring->tail: 0x%08x\n", ee->cpu_ring_tail);
537 err_printf(m, " hangcheck timestamp: %dms (%lu%s)\n",
538 jiffies_to_msecs(ee->hangcheck_timestamp - epoch),
539 ee->hangcheck_timestamp,
540 ee->hangcheck_timestamp == epoch ? "; epoch" : "");
541 err_printf(m, " engine reset count: %u\n", ee->reset_count);
543 for (n = 0; n < ee->num_ports; n++) {
544 err_printf(m, " ELSP[%d]:", n);
545 error_print_request(m, " ", &ee->execlist[n], epoch);
548 error_print_context(m, " Active context: ", &ee->context);
551 void i915_error_printf(struct drm_i915_error_state_buf *e, const char *f, ...)
556 i915_error_vprintf(e, f, args);
560 static void print_error_obj(struct drm_i915_error_state_buf *m,
561 const struct intel_engine_cs *engine,
563 const struct drm_i915_error_object *obj)
565 char out[ASCII85_BUFSZ];
572 err_printf(m, "%s --- %s = 0x%08x %08x\n",
573 engine ? engine->name : "global", name,
574 upper_32_bits(obj->gtt_offset),
575 lower_32_bits(obj->gtt_offset));
578 err_compression_marker(m);
579 for (page = 0; page < obj->page_count; page++) {
583 if (page == obj->page_count - 1)
585 len = ascii85_encode_len(len);
587 for (i = 0; i < len; i++)
588 err_puts(m, ascii85_encode(obj->pages[page][i], out));
593 static void err_print_capabilities(struct drm_i915_error_state_buf *m,
594 const struct intel_device_info *info,
595 const struct intel_runtime_info *runtime,
596 const struct intel_driver_caps *caps)
598 struct drm_printer p = i915_error_printer(m);
600 intel_device_info_dump_flags(info, &p);
601 intel_driver_caps_print(caps, &p);
602 intel_device_info_dump_topology(&runtime->sseu, &p);
605 static void err_print_params(struct drm_i915_error_state_buf *m,
606 const struct i915_params *params)
608 struct drm_printer p = i915_error_printer(m);
610 i915_params_dump(params, &p);
613 static void err_print_pciid(struct drm_i915_error_state_buf *m,
614 struct drm_i915_private *i915)
616 struct pci_dev *pdev = i915->drm.pdev;
618 err_printf(m, "PCI ID: 0x%04x\n", pdev->device);
619 err_printf(m, "PCI Revision: 0x%02x\n", pdev->revision);
620 err_printf(m, "PCI Subsystem: %04x:%04x\n",
621 pdev->subsystem_vendor,
622 pdev->subsystem_device);
625 static void err_print_uc(struct drm_i915_error_state_buf *m,
626 const struct i915_error_uc *error_uc)
628 struct drm_printer p = i915_error_printer(m);
629 const struct i915_gpu_state *error =
630 container_of(error_uc, typeof(*error), uc);
632 if (!error->device_info.has_gt_uc)
635 intel_uc_fw_dump(&error_uc->guc_fw, &p);
636 intel_uc_fw_dump(&error_uc->huc_fw, &p);
637 print_error_obj(m, NULL, "GuC log buffer", error_uc->guc_log);
640 static void err_free_sgl(struct scatterlist *sgl)
643 struct scatterlist *sg;
645 for (sg = sgl; !sg_is_chain(sg); sg++) {
651 sg = sg_is_last(sg) ? NULL : sg_chain_ptr(sg);
652 free_page((unsigned long)sgl);
657 static void __err_print_to_sgl(struct drm_i915_error_state_buf *m,
658 struct i915_gpu_state *error)
660 const struct drm_i915_error_engine *ee;
661 struct timespec64 ts;
664 if (*error->error_msg)
665 err_printf(m, "%s\n", error->error_msg);
666 err_printf(m, "Kernel: %s %s\n",
667 init_utsname()->release,
668 init_utsname()->machine);
669 err_printf(m, "Driver: %s\n", DRIVER_DATE);
670 ts = ktime_to_timespec64(error->time);
671 err_printf(m, "Time: %lld s %ld us\n",
672 (s64)ts.tv_sec, ts.tv_nsec / NSEC_PER_USEC);
673 ts = ktime_to_timespec64(error->boottime);
674 err_printf(m, "Boottime: %lld s %ld us\n",
675 (s64)ts.tv_sec, ts.tv_nsec / NSEC_PER_USEC);
676 ts = ktime_to_timespec64(error->uptime);
677 err_printf(m, "Uptime: %lld s %ld us\n",
678 (s64)ts.tv_sec, ts.tv_nsec / NSEC_PER_USEC);
679 err_printf(m, "Epoch: %lu jiffies (%u HZ)\n", error->epoch, HZ);
680 err_printf(m, "Capture: %lu jiffies; %d ms ago, %d ms after epoch\n",
682 jiffies_to_msecs(jiffies - error->capture),
683 jiffies_to_msecs(error->capture - error->epoch));
685 for (ee = error->engine; ee; ee = ee->next)
686 err_printf(m, "Active process (on ring %s): %s [%d]\n",
691 err_printf(m, "Reset count: %u\n", error->reset_count);
692 err_printf(m, "Suspend count: %u\n", error->suspend_count);
693 err_printf(m, "Platform: %s\n", intel_platform_name(error->device_info.platform));
694 err_printf(m, "Subplatform: 0x%x\n",
695 intel_subplatform(&error->runtime_info,
696 error->device_info.platform));
697 err_print_pciid(m, m->i915);
699 err_printf(m, "IOMMU enabled?: %d\n", error->iommu);
701 if (HAS_CSR(m->i915)) {
702 struct intel_csr *csr = &m->i915->csr;
704 err_printf(m, "DMC loaded: %s\n",
705 yesno(csr->dmc_payload != NULL));
706 err_printf(m, "DMC fw version: %d.%d\n",
707 CSR_VERSION_MAJOR(csr->version),
708 CSR_VERSION_MINOR(csr->version));
711 err_printf(m, "GT awake: %s\n", yesno(error->awake));
712 err_printf(m, "RPM wakelock: %s\n", yesno(error->wakelock));
713 err_printf(m, "PM suspended: %s\n", yesno(error->suspended));
714 err_printf(m, "EIR: 0x%08x\n", error->eir);
715 err_printf(m, "IER: 0x%08x\n", error->ier);
716 for (i = 0; i < error->ngtier; i++)
717 err_printf(m, "GTIER[%d]: 0x%08x\n", i, error->gtier[i]);
718 err_printf(m, "PGTBL_ER: 0x%08x\n", error->pgtbl_er);
719 err_printf(m, "FORCEWAKE: 0x%08x\n", error->forcewake);
720 err_printf(m, "DERRMR: 0x%08x\n", error->derrmr);
721 err_printf(m, "CCID: 0x%08x\n", error->ccid);
723 for (i = 0; i < error->nfence; i++)
724 err_printf(m, " fence[%d] = %08llx\n", i, error->fence[i]);
726 if (IS_GEN_RANGE(m->i915, 6, 11)) {
727 err_printf(m, "ERROR: 0x%08x\n", error->error);
728 err_printf(m, "DONE_REG: 0x%08x\n", error->done_reg);
731 if (INTEL_GEN(m->i915) >= 8)
732 err_printf(m, "FAULT_TLB_DATA: 0x%08x 0x%08x\n",
733 error->fault_data1, error->fault_data0);
735 if (IS_GEN(m->i915, 7))
736 err_printf(m, "ERR_INT: 0x%08x\n", error->err_int);
738 for (ee = error->engine; ee; ee = ee->next)
739 error_print_engine(m, ee, error->epoch);
741 for (ee = error->engine; ee; ee = ee->next) {
742 const struct drm_i915_error_object *obj;
744 obj = ee->batchbuffer;
746 err_puts(m, ee->engine->name);
748 err_printf(m, " (submitted by %s [%d])",
751 err_printf(m, " --- gtt_offset = 0x%08x %08x\n",
752 upper_32_bits(obj->gtt_offset),
753 lower_32_bits(obj->gtt_offset));
754 print_error_obj(m, ee->engine, NULL, obj);
757 for (j = 0; j < ee->user_bo_count; j++)
758 print_error_obj(m, ee->engine, "user", ee->user_bo[j]);
760 if (ee->num_requests) {
761 err_printf(m, "%s --- %d requests\n",
764 for (j = 0; j < ee->num_requests; j++)
765 error_print_request(m, " ",
770 print_error_obj(m, ee->engine, "ringbuffer", ee->ringbuffer);
771 print_error_obj(m, ee->engine, "HW Status", ee->hws_page);
772 print_error_obj(m, ee->engine, "HW context", ee->ctx);
773 print_error_obj(m, ee->engine, "WA context", ee->wa_ctx);
774 print_error_obj(m, ee->engine,
775 "WA batchbuffer", ee->wa_batchbuffer);
776 print_error_obj(m, ee->engine,
777 "NULL context", ee->default_state);
781 intel_overlay_print_error_state(m, error->overlay);
784 intel_display_print_error_state(m, error->display);
786 err_print_capabilities(m, &error->device_info, &error->runtime_info,
787 &error->driver_caps);
788 err_print_params(m, &error->params);
789 err_print_uc(m, &error->uc);
792 static int err_print_to_sgl(struct i915_gpu_state *error)
794 struct drm_i915_error_state_buf m;
797 return PTR_ERR(error);
799 if (READ_ONCE(error->sgl))
802 memset(&m, 0, sizeof(m));
803 m.i915 = error->i915;
805 __err_print_to_sgl(&m, error);
808 __sg_set_buf(m.cur++, m.buf, m.bytes, m.iter);
813 GEM_BUG_ON(m.end < m.cur);
814 sg_mark_end(m.cur - 1);
816 GEM_BUG_ON(m.sgl && !m.cur);
823 if (cmpxchg(&error->sgl, NULL, m.sgl))
829 ssize_t i915_gpu_state_copy_to_buffer(struct i915_gpu_state *error,
830 char *buf, loff_t off, size_t rem)
832 struct scatterlist *sg;
840 err = err_print_to_sgl(error);
844 sg = READ_ONCE(error->fit);
845 if (!sg || off < sg->dma_address)
850 pos = sg->dma_address;
855 if (sg_is_chain(sg)) {
856 sg = sg_chain_ptr(sg);
857 GEM_BUG_ON(sg_is_chain(sg));
861 if (pos + len <= off) {
868 GEM_BUG_ON(off - pos > len);
875 GEM_BUG_ON(!len || len > sg->length);
877 memcpy(buf, page_address(sg_page(sg)) + start, len);
885 WRITE_ONCE(error->fit, sg);
888 } while (!sg_is_last(sg++));
893 static void i915_error_object_free(struct drm_i915_error_object *obj)
900 for (page = 0; page < obj->page_count; page++)
901 free_page((unsigned long)obj->pages[page]);
907 static void cleanup_params(struct i915_gpu_state *error)
909 i915_params_free(&error->params);
912 static void cleanup_uc_state(struct i915_gpu_state *error)
914 struct i915_error_uc *error_uc = &error->uc;
916 kfree(error_uc->guc_fw.path);
917 kfree(error_uc->huc_fw.path);
918 i915_error_object_free(error_uc->guc_log);
921 void __i915_gpu_state_free(struct kref *error_ref)
923 struct i915_gpu_state *error =
924 container_of(error_ref, typeof(*error), ref);
927 while (error->engine) {
928 struct drm_i915_error_engine *ee = error->engine;
930 error->engine = ee->next;
932 for (i = 0; i < ee->user_bo_count; i++)
933 i915_error_object_free(ee->user_bo[i]);
936 i915_error_object_free(ee->batchbuffer);
937 i915_error_object_free(ee->wa_batchbuffer);
938 i915_error_object_free(ee->ringbuffer);
939 i915_error_object_free(ee->hws_page);
940 i915_error_object_free(ee->ctx);
941 i915_error_object_free(ee->wa_ctx);
947 kfree(error->overlay);
948 kfree(error->display);
950 cleanup_params(error);
951 cleanup_uc_state(error);
953 err_free_sgl(error->sgl);
957 static struct drm_i915_error_object *
958 i915_error_object_create(struct drm_i915_private *i915,
959 struct i915_vma *vma,
960 struct compress *compress)
962 struct i915_ggtt *ggtt = &i915->ggtt;
963 const u64 slot = ggtt->error_capture.start;
964 struct drm_i915_error_object *dst;
965 unsigned long num_pages;
966 struct sgt_iter iter;
972 if (!vma || !vma->pages)
975 num_pages = min_t(u64, vma->size, vma->obj->base.size) >> PAGE_SHIFT;
976 num_pages = DIV_ROUND_UP(10 * num_pages, 8); /* worstcase zlib growth */
977 dst = kmalloc(sizeof(*dst) + num_pages * sizeof(u32 *), ALLOW_FAIL);
981 if (!compress_start(compress)) {
986 dst->gtt_offset = vma->node.start;
987 dst->gtt_size = vma->node.size;
988 dst->num_pages = num_pages;
993 for_each_sgt_daddr(dma, iter, vma->pages) {
996 ggtt->vm.insert_page(&ggtt->vm, dma, slot, I915_CACHE_NONE, 0);
998 s = io_mapping_map_wc(&ggtt->iomap, slot, PAGE_SIZE);
999 ret = compress_page(compress, (void __force *)s, dst);
1000 io_mapping_unmap(s);
1005 if (ret || compress_flush(compress, dst)) {
1006 while (dst->page_count--)
1007 pool_free(&compress->pool, dst->pages[dst->page_count]);
1011 compress_finish(compress);
1017 * Generate a semi-unique error code. The code is not meant to have meaning, The
1018 * code's only purpose is to try to prevent false duplicated bug reports by
1019 * grossly estimating a GPU error state.
1021 * TODO Ideally, hashing the batchbuffer would be a very nice way to determine
1022 * the hang if we could strip the GTT offset information from it.
1024 * It's only a small step better than a random number in its current form.
1026 static u32 i915_error_generate_code(struct i915_gpu_state *error)
1028 const struct drm_i915_error_engine *ee = error->engine;
1031 * IPEHR would be an ideal way to detect errors, as it's the gross
1032 * measure of "the command that hung." However, has some very common
1033 * synchronization commands which almost always appear in the case
1034 * strictly a client bug. Use instdone to differentiate those some.
1036 return ee ? ee->ipehr ^ ee->instdone.instdone : 0;
1039 static void gem_record_fences(struct i915_gpu_state *error)
1041 struct drm_i915_private *dev_priv = error->i915;
1042 struct intel_uncore *uncore = &dev_priv->uncore;
1045 if (INTEL_GEN(dev_priv) >= 6) {
1046 for (i = 0; i < dev_priv->ggtt.num_fences; i++)
1048 intel_uncore_read64(uncore,
1049 FENCE_REG_GEN6_LO(i));
1050 } else if (INTEL_GEN(dev_priv) >= 4) {
1051 for (i = 0; i < dev_priv->ggtt.num_fences; i++)
1053 intel_uncore_read64(uncore,
1054 FENCE_REG_965_LO(i));
1056 for (i = 0; i < dev_priv->ggtt.num_fences; i++)
1058 intel_uncore_read(uncore, FENCE_REG(i));
1063 static void error_record_engine_registers(struct i915_gpu_state *error,
1064 struct intel_engine_cs *engine,
1065 struct drm_i915_error_engine *ee)
1067 struct drm_i915_private *dev_priv = engine->i915;
1069 if (INTEL_GEN(dev_priv) >= 6) {
1070 ee->rc_psmi = ENGINE_READ(engine, RING_PSMI_CTL);
1072 if (INTEL_GEN(dev_priv) >= 12)
1073 ee->fault_reg = I915_READ(GEN12_RING_FAULT_REG);
1074 else if (INTEL_GEN(dev_priv) >= 8)
1075 ee->fault_reg = I915_READ(GEN8_RING_FAULT_REG);
1077 ee->fault_reg = GEN6_RING_FAULT_REG_READ(engine);
1080 if (INTEL_GEN(dev_priv) >= 4) {
1081 ee->faddr = ENGINE_READ(engine, RING_DMA_FADD);
1082 ee->ipeir = ENGINE_READ(engine, RING_IPEIR);
1083 ee->ipehr = ENGINE_READ(engine, RING_IPEHR);
1084 ee->instps = ENGINE_READ(engine, RING_INSTPS);
1085 ee->bbaddr = ENGINE_READ(engine, RING_BBADDR);
1086 if (INTEL_GEN(dev_priv) >= 8) {
1087 ee->faddr |= (u64)ENGINE_READ(engine, RING_DMA_FADD_UDW) << 32;
1088 ee->bbaddr |= (u64)ENGINE_READ(engine, RING_BBADDR_UDW) << 32;
1090 ee->bbstate = ENGINE_READ(engine, RING_BBSTATE);
1092 ee->faddr = ENGINE_READ(engine, DMA_FADD_I8XX);
1093 ee->ipeir = ENGINE_READ(engine, IPEIR);
1094 ee->ipehr = ENGINE_READ(engine, IPEHR);
1097 intel_engine_get_instdone(engine, &ee->instdone);
1099 ee->instpm = ENGINE_READ(engine, RING_INSTPM);
1100 ee->acthd = intel_engine_get_active_head(engine);
1101 ee->start = ENGINE_READ(engine, RING_START);
1102 ee->head = ENGINE_READ(engine, RING_HEAD);
1103 ee->tail = ENGINE_READ(engine, RING_TAIL);
1104 ee->ctl = ENGINE_READ(engine, RING_CTL);
1105 if (INTEL_GEN(dev_priv) > 2)
1106 ee->mode = ENGINE_READ(engine, RING_MI_MODE);
1108 if (!HWS_NEEDS_PHYSICAL(dev_priv)) {
1111 if (IS_GEN(dev_priv, 7)) {
1112 switch (engine->id) {
1114 MISSING_CASE(engine->id);
1117 mmio = RENDER_HWS_PGA_GEN7;
1120 mmio = BLT_HWS_PGA_GEN7;
1123 mmio = BSD_HWS_PGA_GEN7;
1126 mmio = VEBOX_HWS_PGA_GEN7;
1129 } else if (IS_GEN(engine->i915, 6)) {
1130 mmio = RING_HWS_PGA_GEN6(engine->mmio_base);
1132 /* XXX: gen8 returns to sanity */
1133 mmio = RING_HWS_PGA(engine->mmio_base);
1136 ee->hws = I915_READ(mmio);
1139 ee->idle = intel_engine_is_idle(engine);
1141 ee->hangcheck_timestamp = engine->hangcheck.action_timestamp;
1142 ee->reset_count = i915_reset_engine_count(&dev_priv->gpu_error,
1145 if (HAS_PPGTT(dev_priv)) {
1148 ee->vm_info.gfx_mode = ENGINE_READ(engine, RING_MODE_GEN7);
1150 if (IS_GEN(dev_priv, 6)) {
1151 ee->vm_info.pp_dir_base =
1152 ENGINE_READ(engine, RING_PP_DIR_BASE_READ);
1153 } else if (IS_GEN(dev_priv, 7)) {
1154 ee->vm_info.pp_dir_base =
1155 ENGINE_READ(engine, RING_PP_DIR_BASE);
1156 } else if (INTEL_GEN(dev_priv) >= 8) {
1157 u32 base = engine->mmio_base;
1159 for (i = 0; i < 4; i++) {
1160 ee->vm_info.pdp[i] =
1161 I915_READ(GEN8_RING_PDP_UDW(base, i));
1162 ee->vm_info.pdp[i] <<= 32;
1163 ee->vm_info.pdp[i] |=
1164 I915_READ(GEN8_RING_PDP_LDW(base, i));
1170 static void record_request(const struct i915_request *request,
1171 struct drm_i915_error_request *erq)
1173 const struct i915_gem_context *ctx = request->gem_context;
1175 erq->flags = request->fence.flags;
1176 erq->context = request->fence.context;
1177 erq->seqno = request->fence.seqno;
1178 erq->sched_attr = request->sched.attr;
1179 erq->jiffies = request->emitted_jiffies;
1180 erq->start = i915_ggtt_offset(request->ring->vma);
1181 erq->head = request->head;
1182 erq->tail = request->tail;
1185 erq->pid = ctx->pid ? pid_nr(ctx->pid) : 0;
1189 static void engine_record_requests(struct intel_engine_cs *engine,
1190 struct i915_request *first,
1191 struct drm_i915_error_engine *ee)
1193 struct i915_request *request;
1198 list_for_each_entry_from(request, &engine->active.requests, sched.link)
1203 ee->requests = kcalloc(count, sizeof(*ee->requests), ATOMIC_MAYFAIL);
1207 ee->num_requests = count;
1211 list_for_each_entry_from(request,
1212 &engine->active.requests, sched.link) {
1213 if (count >= ee->num_requests) {
1215 * If the ring request list was changed in
1216 * between the point where the error request
1217 * list was created and dimensioned and this
1218 * point then just exit early to avoid crashes.
1220 * We don't need to communicate that the
1221 * request list changed state during error
1222 * state capture and that the error state is
1223 * slightly incorrect as a consequence since we
1224 * are typically only interested in the request
1225 * list state at the point of error state
1226 * capture, not in any changes happening during
1232 record_request(request, &ee->requests[count++]);
1234 ee->num_requests = count;
1237 static void error_record_engine_execlists(const struct intel_engine_cs *engine,
1238 struct drm_i915_error_engine *ee)
1240 const struct intel_engine_execlists * const execlists = &engine->execlists;
1241 struct i915_request * const *port = execlists->active;
1245 record_request(*port++, &ee->execlist[n++]);
1250 static bool record_context(struct drm_i915_error_context *e,
1251 const struct i915_request *rq)
1253 const struct i915_gem_context *ctx = rq->gem_context;
1256 struct task_struct *task;
1259 task = pid_task(ctx->pid, PIDTYPE_PID);
1261 strcpy(e->comm, task->comm);
1267 e->hw_id = ctx->hw_id;
1268 e->sched_attr = ctx->sched;
1269 e->guilty = atomic_read(&ctx->guilty_count);
1270 e->active = atomic_read(&ctx->active_count);
1272 return i915_gem_context_no_error_capture(ctx);
1275 struct capture_vma {
1276 struct capture_vma *next;
1280 static struct capture_vma *
1281 capture_vma(struct capture_vma *next,
1282 struct i915_vma *vma,
1283 struct drm_i915_error_object **out)
1285 struct capture_vma *c;
1291 c = kmalloc(sizeof(*c), ATOMIC_MAYFAIL);
1295 if (!i915_active_trygrab(&vma->active)) {
1300 c->slot = (void **)out;
1301 *c->slot = i915_vma_get(vma);
1307 static struct capture_vma *
1308 request_record_user_bo(struct i915_request *request,
1309 struct drm_i915_error_engine *ee,
1310 struct capture_vma *capture)
1312 struct i915_capture_list *c;
1313 struct drm_i915_error_object **bo;
1317 for (c = request->capture_list; c; c = c->next)
1322 bo = kmalloc_array(max, sizeof(*bo), ATOMIC_MAYFAIL);
1324 /* If we can't capture everything, try to capture something. */
1325 max = min_t(long, max, PAGE_SIZE / sizeof(*bo));
1326 bo = kmalloc_array(max, sizeof(*bo), ATOMIC_MAYFAIL);
1332 for (c = request->capture_list; c; c = c->next) {
1333 capture = capture_vma(capture, c->vma, &bo[count]);
1339 ee->user_bo_count = count;
1344 static struct drm_i915_error_object *
1345 capture_object(struct drm_i915_private *dev_priv,
1346 struct drm_i915_gem_object *obj,
1347 struct compress *compress)
1349 if (obj && i915_gem_object_has_pages(obj)) {
1350 struct i915_vma fake = {
1351 .node = { .start = U64_MAX, .size = obj->base.size },
1352 .size = obj->base.size,
1353 .pages = obj->mm.pages,
1357 return i915_error_object_create(dev_priv, &fake, compress);
1364 gem_record_rings(struct i915_gpu_state *error, struct compress *compress)
1366 struct drm_i915_private *i915 = error->i915;
1367 struct intel_engine_cs *engine;
1368 struct drm_i915_error_engine *ee;
1370 ee = kzalloc(sizeof(*ee), GFP_KERNEL);
1374 for_each_uabi_engine(engine, i915) {
1375 struct capture_vma *capture = NULL;
1376 struct i915_request *request;
1377 unsigned long flags;
1379 /* Refill our page pool before entering atomic section */
1380 pool_refill(&compress->pool, ALLOW_FAIL);
1382 spin_lock_irqsave(&engine->active.lock, flags);
1383 request = intel_engine_find_active_request(engine);
1385 spin_unlock_irqrestore(&engine->active.lock, flags);
1389 error->simulated |= record_context(&ee->context, request);
1392 * We need to copy these to an anonymous buffer
1393 * as the simplest method to avoid being overwritten
1396 capture = capture_vma(capture,
1400 if (HAS_BROKEN_CS_TLB(i915))
1401 capture = capture_vma(capture,
1402 engine->gt->scratch,
1403 &ee->wa_batchbuffer);
1405 capture = request_record_user_bo(request, ee, capture);
1407 capture = capture_vma(capture,
1408 request->hw_context->state,
1411 capture = capture_vma(capture,
1415 ee->cpu_ring_head = request->ring->head;
1416 ee->cpu_ring_tail = request->ring->tail;
1418 ee->rq_head = request->head;
1419 ee->rq_post = request->postfix;
1420 ee->rq_tail = request->tail;
1422 engine_record_requests(engine, request, ee);
1423 spin_unlock_irqrestore(&engine->active.lock, flags);
1425 error_record_engine_registers(error, engine, ee);
1426 error_record_engine_execlists(engine, ee);
1429 struct capture_vma *this = capture;
1430 struct i915_vma *vma = *this->slot;
1433 i915_error_object_create(i915, vma, compress);
1435 i915_active_ungrab(&vma->active);
1438 capture = this->next;
1443 i915_error_object_create(i915,
1444 engine->status_page.vma,
1448 i915_error_object_create(i915,
1453 capture_object(i915, engine->default_state, compress);
1455 ee->engine = engine;
1457 ee->next = error->engine;
1460 ee = kzalloc(sizeof(*ee), GFP_KERNEL);
1469 capture_uc_state(struct i915_gpu_state *error, struct compress *compress)
1471 struct drm_i915_private *i915 = error->i915;
1472 struct i915_error_uc *error_uc = &error->uc;
1473 struct intel_uc *uc = &i915->gt.uc;
1475 /* Capturing uC state won't be useful if there is no GuC */
1476 if (!error->device_info.has_gt_uc)
1479 memcpy(&error_uc->guc_fw, &uc->guc.fw, sizeof(uc->guc.fw));
1480 memcpy(&error_uc->huc_fw, &uc->huc.fw, sizeof(uc->huc.fw));
1482 /* Non-default firmware paths will be specified by the modparam.
1483 * As modparams are generally accesible from the userspace make
1484 * explicit copies of the firmware paths.
1486 error_uc->guc_fw.path = kstrdup(uc->guc.fw.path, ALLOW_FAIL);
1487 error_uc->huc_fw.path = kstrdup(uc->huc.fw.path, ALLOW_FAIL);
1488 error_uc->guc_log = i915_error_object_create(i915,
1493 /* Capture all registers which don't fit into another category. */
1494 static void capture_reg_state(struct i915_gpu_state *error)
1496 struct drm_i915_private *i915 = error->i915;
1497 struct intel_uncore *uncore = &i915->uncore;
1500 /* General organization
1501 * 1. Registers specific to a single generation
1502 * 2. Registers which belong to multiple generations
1503 * 3. Feature specific registers.
1504 * 4. Everything else
1505 * Please try to follow the order.
1508 /* 1: Registers specific to a single generation */
1509 if (IS_VALLEYVIEW(i915)) {
1510 error->gtier[0] = intel_uncore_read(uncore, GTIER);
1511 error->ier = intel_uncore_read(uncore, VLV_IER);
1512 error->forcewake = intel_uncore_read_fw(uncore, FORCEWAKE_VLV);
1515 if (IS_GEN(i915, 7))
1516 error->err_int = intel_uncore_read(uncore, GEN7_ERR_INT);
1518 if (INTEL_GEN(i915) >= 12) {
1519 error->fault_data0 = intel_uncore_read(uncore,
1520 GEN12_FAULT_TLB_DATA0);
1521 error->fault_data1 = intel_uncore_read(uncore,
1522 GEN12_FAULT_TLB_DATA1);
1523 } else if (INTEL_GEN(i915) >= 8) {
1524 error->fault_data0 = intel_uncore_read(uncore,
1525 GEN8_FAULT_TLB_DATA0);
1526 error->fault_data1 = intel_uncore_read(uncore,
1527 GEN8_FAULT_TLB_DATA1);
1530 if (IS_GEN(i915, 6)) {
1531 error->forcewake = intel_uncore_read_fw(uncore, FORCEWAKE);
1532 error->gab_ctl = intel_uncore_read(uncore, GAB_CTL);
1533 error->gfx_mode = intel_uncore_read(uncore, GFX_MODE);
1536 /* 2: Registers which belong to multiple generations */
1537 if (INTEL_GEN(i915) >= 7)
1538 error->forcewake = intel_uncore_read_fw(uncore, FORCEWAKE_MT);
1540 if (INTEL_GEN(i915) >= 6) {
1541 error->derrmr = intel_uncore_read(uncore, DERRMR);
1542 if (INTEL_GEN(i915) < 12) {
1543 error->error = intel_uncore_read(uncore, ERROR_GEN6);
1544 error->done_reg = intel_uncore_read(uncore, DONE_REG);
1548 if (INTEL_GEN(i915) >= 5)
1549 error->ccid = intel_uncore_read(uncore, CCID(RENDER_RING_BASE));
1551 /* 3: Feature specific registers */
1552 if (IS_GEN_RANGE(i915, 6, 7)) {
1553 error->gam_ecochk = intel_uncore_read(uncore, GAM_ECOCHK);
1554 error->gac_eco = intel_uncore_read(uncore, GAC_ECO_BITS);
1557 /* 4: Everything else */
1558 if (INTEL_GEN(i915) >= 11) {
1559 error->ier = intel_uncore_read(uncore, GEN8_DE_MISC_IER);
1561 intel_uncore_read(uncore,
1562 GEN11_RENDER_COPY_INTR_ENABLE);
1564 intel_uncore_read(uncore, GEN11_VCS_VECS_INTR_ENABLE);
1566 intel_uncore_read(uncore, GEN11_GUC_SG_INTR_ENABLE);
1568 intel_uncore_read(uncore,
1569 GEN11_GPM_WGBOXPERF_INTR_ENABLE);
1571 intel_uncore_read(uncore,
1572 GEN11_CRYPTO_RSVD_INTR_ENABLE);
1574 intel_uncore_read(uncore,
1575 GEN11_GUNIT_CSME_INTR_ENABLE);
1577 } else if (INTEL_GEN(i915) >= 8) {
1578 error->ier = intel_uncore_read(uncore, GEN8_DE_MISC_IER);
1579 for (i = 0; i < 4; i++)
1580 error->gtier[i] = intel_uncore_read(uncore,
1583 } else if (HAS_PCH_SPLIT(i915)) {
1584 error->ier = intel_uncore_read(uncore, DEIER);
1585 error->gtier[0] = intel_uncore_read(uncore, GTIER);
1587 } else if (IS_GEN(i915, 2)) {
1588 error->ier = intel_uncore_read16(uncore, GEN2_IER);
1589 } else if (!IS_VALLEYVIEW(i915)) {
1590 error->ier = intel_uncore_read(uncore, GEN2_IER);
1592 error->eir = intel_uncore_read(uncore, EIR);
1593 error->pgtbl_er = intel_uncore_read(uncore, PGTBL_ER);
1597 error_msg(struct i915_gpu_state *error,
1598 intel_engine_mask_t engines, const char *msg)
1602 len = scnprintf(error->error_msg, sizeof(error->error_msg),
1603 "GPU HANG: ecode %d:%x:0x%08x",
1604 INTEL_GEN(error->i915), engines,
1605 i915_error_generate_code(error));
1606 if (error->engine) {
1607 /* Just show the first executing process, more is confusing */
1608 len += scnprintf(error->error_msg + len,
1609 sizeof(error->error_msg) - len,
1611 error->engine->context.comm,
1612 error->engine->context.pid);
1615 len += scnprintf(error->error_msg + len,
1616 sizeof(error->error_msg) - len,
1619 return error->error_msg;
1622 static void capture_gen_state(struct i915_gpu_state *error)
1624 struct drm_i915_private *i915 = error->i915;
1626 error->awake = i915->gt.awake;
1627 error->wakelock = atomic_read(&i915->runtime_pm.wakeref_count);
1628 error->suspended = i915->runtime_pm.suspended;
1631 #ifdef CONFIG_INTEL_IOMMU
1632 error->iommu = intel_iommu_gfx_mapped;
1634 error->reset_count = i915_reset_count(&i915->gpu_error);
1635 error->suspend_count = i915->suspend_count;
1637 memcpy(&error->device_info,
1639 sizeof(error->device_info));
1640 memcpy(&error->runtime_info,
1642 sizeof(error->runtime_info));
1643 error->driver_caps = i915->caps;
1646 static void capture_params(struct i915_gpu_state *error)
1648 i915_params_copy(&error->params, &i915_modparams);
1651 static unsigned long capture_find_epoch(const struct i915_gpu_state *error)
1653 const struct drm_i915_error_engine *ee;
1654 unsigned long epoch = error->capture;
1656 for (ee = error->engine; ee; ee = ee->next) {
1657 if (ee->hangcheck_timestamp &&
1658 time_before(ee->hangcheck_timestamp, epoch))
1659 epoch = ee->hangcheck_timestamp;
1665 static void capture_finish(struct i915_gpu_state *error)
1667 struct i915_ggtt *ggtt = &error->i915->ggtt;
1668 const u64 slot = ggtt->error_capture.start;
1670 ggtt->vm.clear_range(&ggtt->vm, slot, PAGE_SIZE);
1673 #define DAY_AS_SECONDS(x) (24 * 60 * 60 * (x))
1675 struct i915_gpu_state *
1676 i915_capture_gpu_state(struct drm_i915_private *i915)
1678 struct i915_gpu_state *error;
1679 struct compress compress;
1681 /* Check if GPU capture has been disabled */
1682 error = READ_ONCE(i915->gpu_error.first_error);
1686 error = kzalloc(sizeof(*error), ALLOW_FAIL);
1688 i915_disable_error_state(i915, -ENOMEM);
1689 return ERR_PTR(-ENOMEM);
1692 if (!compress_init(&compress)) {
1694 i915_disable_error_state(i915, -ENOMEM);
1695 return ERR_PTR(-ENOMEM);
1698 kref_init(&error->ref);
1701 error->time = ktime_get_real();
1702 error->boottime = ktime_get_boottime();
1703 error->uptime = ktime_sub(ktime_get(), i915->gt.last_init_time);
1704 error->capture = jiffies;
1706 capture_params(error);
1707 capture_gen_state(error);
1708 capture_uc_state(error, &compress);
1709 capture_reg_state(error);
1710 gem_record_fences(error);
1711 gem_record_rings(error, &compress);
1713 error->overlay = intel_overlay_capture_error_state(i915);
1714 error->display = intel_display_capture_error_state(i915);
1716 error->epoch = capture_find_epoch(error);
1718 capture_finish(error);
1719 compress_fini(&compress);
1725 * i915_capture_error_state - capture an error record for later analysis
1726 * @i915: i915 device
1727 * @engine_mask: the mask of engines triggering the hang
1728 * @msg: a message to insert into the error capture header
1730 * Should be called when an error is detected (either a hang or an error
1731 * interrupt) to capture error state from the time of the error. Fills
1732 * out a structure which becomes available in debugfs for user level tools
1735 void i915_capture_error_state(struct drm_i915_private *i915,
1736 intel_engine_mask_t engine_mask,
1740 struct i915_gpu_state *error;
1741 unsigned long flags;
1743 if (!i915_modparams.error_capture)
1746 if (READ_ONCE(i915->gpu_error.first_error))
1749 error = i915_capture_gpu_state(i915);
1753 dev_info(i915->drm.dev, "%s\n", error_msg(error, engine_mask, msg));
1755 if (!error->simulated) {
1756 spin_lock_irqsave(&i915->gpu_error.lock, flags);
1757 if (!i915->gpu_error.first_error) {
1758 i915->gpu_error.first_error = error;
1761 spin_unlock_irqrestore(&i915->gpu_error.lock, flags);
1765 __i915_gpu_state_free(&error->ref);
1769 if (!xchg(&warned, true) &&
1770 ktime_get_real_seconds() - DRIVER_TIMESTAMP < DAY_AS_SECONDS(180)) {
1771 pr_info("GPU hangs can indicate a bug anywhere in the entire gfx stack, including userspace.\n");
1772 pr_info("Please file a _new_ bug report on bugs.freedesktop.org against DRI -> DRM/Intel\n");
1773 pr_info("drm/i915 developers can then reassign to the right component if it's not a kernel issue.\n");
1774 pr_info("The GPU crash dump is required to analyze GPU hangs, so please always attach it.\n");
1775 pr_info("GPU crash dump saved to /sys/class/drm/card%d/error\n",
1776 i915->drm.primary->index);
1780 struct i915_gpu_state *
1781 i915_first_error_state(struct drm_i915_private *i915)
1783 struct i915_gpu_state *error;
1785 spin_lock_irq(&i915->gpu_error.lock);
1786 error = i915->gpu_error.first_error;
1787 if (!IS_ERR_OR_NULL(error))
1788 i915_gpu_state_get(error);
1789 spin_unlock_irq(&i915->gpu_error.lock);
1794 void i915_reset_error_state(struct drm_i915_private *i915)
1796 struct i915_gpu_state *error;
1798 spin_lock_irq(&i915->gpu_error.lock);
1799 error = i915->gpu_error.first_error;
1800 if (error != ERR_PTR(-ENODEV)) /* if disabled, always disabled */
1801 i915->gpu_error.first_error = NULL;
1802 spin_unlock_irq(&i915->gpu_error.lock);
1804 if (!IS_ERR_OR_NULL(error))
1805 i915_gpu_state_put(error);
1808 void i915_disable_error_state(struct drm_i915_private *i915, int err)
1810 spin_lock_irq(&i915->gpu_error.lock);
1811 if (!i915->gpu_error.first_error)
1812 i915->gpu_error.first_error = ERR_PTR(err);
1813 spin_unlock_irq(&i915->gpu_error.lock);