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_csr.h"
41 #include "display/intel_overlay.h"
43 #include "gem/i915_gem_context.h"
44 #include "gem/i915_gem_lmem.h"
45 #include "gt/intel_gt.h"
46 #include "gt/intel_gt_pm.h"
49 #include "i915_gpu_error.h"
50 #include "i915_memcpy.h"
51 #include "i915_scatterlist.h"
53 #define ALLOW_FAIL (GFP_KERNEL | __GFP_RETRY_MAYFAIL | __GFP_NOWARN)
54 #define ATOMIC_MAYFAIL (GFP_ATOMIC | __GFP_NOWARN)
56 static void __sg_set_buf(struct scatterlist *sg,
57 void *addr, unsigned int len, loff_t it)
59 sg->page_link = (unsigned long)virt_to_page(addr);
60 sg->offset = offset_in_page(addr);
65 static bool __i915_error_grow(struct drm_i915_error_state_buf *e, size_t len)
70 if (e->bytes + len + 1 <= e->size)
74 __sg_set_buf(e->cur++, e->buf, e->bytes, e->iter);
80 if (e->cur == e->end) {
81 struct scatterlist *sgl;
83 sgl = (typeof(sgl))__get_free_page(ALLOW_FAIL);
93 (unsigned long)sgl | SG_CHAIN;
99 e->end = sgl + SG_MAX_SINGLE_ALLOC - 1;
102 e->size = ALIGN(len + 1, SZ_64K);
103 e->buf = kmalloc(e->size, ALLOW_FAIL);
105 e->size = PAGE_ALIGN(len + 1);
106 e->buf = kmalloc(e->size, GFP_KERNEL);
117 static void i915_error_vprintf(struct drm_i915_error_state_buf *e,
118 const char *fmt, va_list args)
127 len = vsnprintf(NULL, 0, fmt, ap);
134 if (!__i915_error_grow(e, len))
137 GEM_BUG_ON(e->bytes >= e->size);
138 len = vscnprintf(e->buf + e->bytes, e->size - e->bytes, fmt, args);
146 static void i915_error_puts(struct drm_i915_error_state_buf *e, const char *str)
154 if (!__i915_error_grow(e, len))
157 GEM_BUG_ON(e->bytes + len > e->size);
158 memcpy(e->buf + e->bytes, str, len);
162 #define err_printf(e, ...) i915_error_printf(e, __VA_ARGS__)
163 #define err_puts(e, s) i915_error_puts(e, s)
165 static void __i915_printfn_error(struct drm_printer *p, struct va_format *vaf)
167 i915_error_vprintf(p->arg, vaf->fmt, *vaf->va);
170 static inline struct drm_printer
171 i915_error_printer(struct drm_i915_error_state_buf *e)
173 struct drm_printer p = {
174 .printfn = __i915_printfn_error,
180 /* single threaded page allocator with a reserved stash for emergencies */
181 static void pool_fini(struct pagevec *pv)
186 static int pool_refill(struct pagevec *pv, gfp_t gfp)
188 while (pagevec_space(pv)) {
201 static int pool_init(struct pagevec *pv, gfp_t gfp)
207 err = pool_refill(pv, gfp);
214 static void *pool_alloc(struct pagevec *pv, gfp_t gfp)
219 if (!p && pagevec_count(pv))
220 p = pv->pages[--pv->nr];
222 return p ? page_address(p) : NULL;
225 static void pool_free(struct pagevec *pv, void *addr)
227 struct page *p = virt_to_page(addr);
229 if (pagevec_space(pv))
235 #ifdef CONFIG_DRM_I915_COMPRESS_ERROR
237 struct i915_vma_compress {
239 struct z_stream_s zstream;
243 static bool compress_init(struct i915_vma_compress *c)
245 struct z_stream_s *zstream = &c->zstream;
247 if (pool_init(&c->pool, ALLOW_FAIL))
251 kmalloc(zlib_deflate_workspacesize(MAX_WBITS, MAX_MEM_LEVEL),
253 if (!zstream->workspace) {
259 if (i915_has_memcpy_from_wc())
260 c->tmp = pool_alloc(&c->pool, ALLOW_FAIL);
265 static bool compress_start(struct i915_vma_compress *c)
267 struct z_stream_s *zstream = &c->zstream;
268 void *workspace = zstream->workspace;
270 memset(zstream, 0, sizeof(*zstream));
271 zstream->workspace = workspace;
273 return zlib_deflateInit(zstream, Z_DEFAULT_COMPRESSION) == Z_OK;
276 static void *compress_next_page(struct i915_vma_compress *c,
277 struct i915_vma_coredump *dst)
281 if (dst->page_count >= dst->num_pages)
282 return ERR_PTR(-ENOSPC);
284 page = pool_alloc(&c->pool, ALLOW_FAIL);
286 return ERR_PTR(-ENOMEM);
288 return dst->pages[dst->page_count++] = page;
291 static int compress_page(struct i915_vma_compress *c,
293 struct i915_vma_coredump *dst,
296 struct z_stream_s *zstream = &c->zstream;
298 zstream->next_in = src;
299 if (wc && c->tmp && i915_memcpy_from_wc(c->tmp, src, PAGE_SIZE))
300 zstream->next_in = c->tmp;
301 zstream->avail_in = PAGE_SIZE;
304 if (zstream->avail_out == 0) {
305 zstream->next_out = compress_next_page(c, dst);
306 if (IS_ERR(zstream->next_out))
307 return PTR_ERR(zstream->next_out);
309 zstream->avail_out = PAGE_SIZE;
312 if (zlib_deflate(zstream, Z_NO_FLUSH) != Z_OK)
314 } while (zstream->avail_in);
316 /* Fallback to uncompressed if we increase size? */
317 if (0 && zstream->total_out > zstream->total_in)
323 static int compress_flush(struct i915_vma_compress *c,
324 struct i915_vma_coredump *dst)
326 struct z_stream_s *zstream = &c->zstream;
329 switch (zlib_deflate(zstream, Z_FINISH)) {
330 case Z_OK: /* more space requested */
331 zstream->next_out = compress_next_page(c, dst);
332 if (IS_ERR(zstream->next_out))
333 return PTR_ERR(zstream->next_out);
335 zstream->avail_out = PAGE_SIZE;
341 default: /* any error */
347 memset(zstream->next_out, 0, zstream->avail_out);
348 dst->unused = zstream->avail_out;
352 static void compress_finish(struct i915_vma_compress *c)
354 zlib_deflateEnd(&c->zstream);
357 static void compress_fini(struct i915_vma_compress *c)
359 kfree(c->zstream.workspace);
361 pool_free(&c->pool, c->tmp);
365 static void err_compression_marker(struct drm_i915_error_state_buf *m)
372 struct i915_vma_compress {
376 static bool compress_init(struct i915_vma_compress *c)
378 return pool_init(&c->pool, ALLOW_FAIL) == 0;
381 static bool compress_start(struct i915_vma_compress *c)
386 static int compress_page(struct i915_vma_compress *c,
388 struct i915_vma_coredump *dst,
393 ptr = pool_alloc(&c->pool, ALLOW_FAIL);
397 if (!(wc && i915_memcpy_from_wc(ptr, src, PAGE_SIZE)))
398 memcpy(ptr, src, PAGE_SIZE);
399 dst->pages[dst->page_count++] = ptr;
404 static int compress_flush(struct i915_vma_compress *c,
405 struct i915_vma_coredump *dst)
410 static void compress_finish(struct i915_vma_compress *c)
414 static void compress_fini(struct i915_vma_compress *c)
419 static void err_compression_marker(struct drm_i915_error_state_buf *m)
426 static void error_print_instdone(struct drm_i915_error_state_buf *m,
427 const struct intel_engine_coredump *ee)
429 const struct sseu_dev_info *sseu = &ee->engine->gt->info.sseu;
433 err_printf(m, " INSTDONE: 0x%08x\n",
434 ee->instdone.instdone);
436 if (ee->engine->class != RENDER_CLASS || INTEL_GEN(m->i915) <= 3)
439 err_printf(m, " SC_INSTDONE: 0x%08x\n",
440 ee->instdone.slice_common);
442 if (INTEL_GEN(m->i915) <= 6)
445 for_each_instdone_slice_subslice(m->i915, sseu, slice, subslice)
446 err_printf(m, " SAMPLER_INSTDONE[%d][%d]: 0x%08x\n",
448 ee->instdone.sampler[slice][subslice]);
450 for_each_instdone_slice_subslice(m->i915, sseu, slice, subslice)
451 err_printf(m, " ROW_INSTDONE[%d][%d]: 0x%08x\n",
453 ee->instdone.row[slice][subslice]);
455 if (INTEL_GEN(m->i915) < 12)
458 err_printf(m, " SC_INSTDONE_EXTRA: 0x%08x\n",
459 ee->instdone.slice_common_extra[0]);
460 err_printf(m, " SC_INSTDONE_EXTRA2: 0x%08x\n",
461 ee->instdone.slice_common_extra[1]);
464 static void error_print_request(struct drm_i915_error_state_buf *m,
466 const struct i915_request_coredump *erq)
471 err_printf(m, "%s pid %d, seqno %8x:%08x%s%s, prio %d, head %08x, tail %08x\n",
472 prefix, erq->pid, erq->context, erq->seqno,
473 test_bit(DMA_FENCE_FLAG_SIGNALED_BIT,
474 &erq->flags) ? "!" : "",
475 test_bit(DMA_FENCE_FLAG_ENABLE_SIGNAL_BIT,
476 &erq->flags) ? "+" : "",
477 erq->sched_attr.priority,
478 erq->head, erq->tail);
481 static void error_print_context(struct drm_i915_error_state_buf *m,
483 const struct i915_gem_context_coredump *ctx)
485 const u32 period = RUNTIME_INFO(m->i915)->cs_timestamp_period_ns;
487 err_printf(m, "%s%s[%d] prio %d, guilty %d active %d, runtime total %lluns, avg %lluns\n",
488 header, ctx->comm, ctx->pid, ctx->sched_attr.priority,
489 ctx->guilty, ctx->active,
490 ctx->total_runtime * period,
491 mul_u32_u32(ctx->avg_runtime, period));
494 static struct i915_vma_coredump *
495 __find_vma(struct i915_vma_coredump *vma, const char *name)
498 if (strcmp(vma->name, name) == 0)
506 static struct i915_vma_coredump *
507 find_batch(const struct intel_engine_coredump *ee)
509 return __find_vma(ee->vma, "batch");
512 static void error_print_engine(struct drm_i915_error_state_buf *m,
513 const struct intel_engine_coredump *ee)
515 struct i915_vma_coredump *batch;
518 err_printf(m, "%s command stream:\n", ee->engine->name);
519 err_printf(m, " CCID: 0x%08x\n", ee->ccid);
520 err_printf(m, " START: 0x%08x\n", ee->start);
521 err_printf(m, " HEAD: 0x%08x [0x%08x]\n", ee->head, ee->rq_head);
522 err_printf(m, " TAIL: 0x%08x [0x%08x, 0x%08x]\n",
523 ee->tail, ee->rq_post, ee->rq_tail);
524 err_printf(m, " CTL: 0x%08x\n", ee->ctl);
525 err_printf(m, " MODE: 0x%08x\n", ee->mode);
526 err_printf(m, " HWS: 0x%08x\n", ee->hws);
527 err_printf(m, " ACTHD: 0x%08x %08x\n",
528 (u32)(ee->acthd>>32), (u32)ee->acthd);
529 err_printf(m, " IPEIR: 0x%08x\n", ee->ipeir);
530 err_printf(m, " IPEHR: 0x%08x\n", ee->ipehr);
531 err_printf(m, " ESR: 0x%08x\n", ee->esr);
533 error_print_instdone(m, ee);
535 batch = find_batch(ee);
537 u64 start = batch->gtt_offset;
538 u64 end = start + batch->gtt_size;
540 err_printf(m, " batch: [0x%08x_%08x, 0x%08x_%08x]\n",
541 upper_32_bits(start), lower_32_bits(start),
542 upper_32_bits(end), lower_32_bits(end));
544 if (INTEL_GEN(m->i915) >= 4) {
545 err_printf(m, " BBADDR: 0x%08x_%08x\n",
546 (u32)(ee->bbaddr>>32), (u32)ee->bbaddr);
547 err_printf(m, " BB_STATE: 0x%08x\n", ee->bbstate);
548 err_printf(m, " INSTPS: 0x%08x\n", ee->instps);
550 err_printf(m, " INSTPM: 0x%08x\n", ee->instpm);
551 err_printf(m, " FADDR: 0x%08x %08x\n", upper_32_bits(ee->faddr),
552 lower_32_bits(ee->faddr));
553 if (INTEL_GEN(m->i915) >= 6) {
554 err_printf(m, " RC PSMI: 0x%08x\n", ee->rc_psmi);
555 err_printf(m, " FAULT_REG: 0x%08x\n", ee->fault_reg);
557 if (HAS_PPGTT(m->i915)) {
558 err_printf(m, " GFX_MODE: 0x%08x\n", ee->vm_info.gfx_mode);
560 if (INTEL_GEN(m->i915) >= 8) {
562 for (i = 0; i < 4; i++)
563 err_printf(m, " PDP%d: 0x%016llx\n",
564 i, ee->vm_info.pdp[i]);
566 err_printf(m, " PP_DIR_BASE: 0x%08x\n",
567 ee->vm_info.pp_dir_base);
570 err_printf(m, " engine reset count: %u\n", ee->reset_count);
572 for (n = 0; n < ee->num_ports; n++) {
573 err_printf(m, " ELSP[%d]:", n);
574 error_print_request(m, " ", &ee->execlist[n]);
577 error_print_context(m, " Active context: ", &ee->context);
580 void i915_error_printf(struct drm_i915_error_state_buf *e, const char *f, ...)
585 i915_error_vprintf(e, f, args);
589 static void print_error_vma(struct drm_i915_error_state_buf *m,
590 const struct intel_engine_cs *engine,
591 const struct i915_vma_coredump *vma)
593 char out[ASCII85_BUFSZ];
599 err_printf(m, "%s --- %s = 0x%08x %08x\n",
600 engine ? engine->name : "global", vma->name,
601 upper_32_bits(vma->gtt_offset),
602 lower_32_bits(vma->gtt_offset));
604 if (vma->gtt_page_sizes > I915_GTT_PAGE_SIZE_4K)
605 err_printf(m, "gtt_page_sizes = 0x%08x\n", vma->gtt_page_sizes);
607 err_compression_marker(m);
608 for (page = 0; page < vma->page_count; page++) {
612 if (page == vma->page_count - 1)
614 len = ascii85_encode_len(len);
616 for (i = 0; i < len; i++)
617 err_puts(m, ascii85_encode(vma->pages[page][i], out));
622 static void err_print_capabilities(struct drm_i915_error_state_buf *m,
623 struct i915_gpu_coredump *error)
625 struct drm_printer p = i915_error_printer(m);
627 intel_device_info_print_static(&error->device_info, &p);
628 intel_device_info_print_runtime(&error->runtime_info, &p);
629 intel_driver_caps_print(&error->driver_caps, &p);
632 static void err_print_params(struct drm_i915_error_state_buf *m,
633 const struct i915_params *params)
635 struct drm_printer p = i915_error_printer(m);
637 i915_params_dump(params, &p);
640 static void err_print_pciid(struct drm_i915_error_state_buf *m,
641 struct drm_i915_private *i915)
643 struct pci_dev *pdev = i915->drm.pdev;
645 err_printf(m, "PCI ID: 0x%04x\n", pdev->device);
646 err_printf(m, "PCI Revision: 0x%02x\n", pdev->revision);
647 err_printf(m, "PCI Subsystem: %04x:%04x\n",
648 pdev->subsystem_vendor,
649 pdev->subsystem_device);
652 static void err_print_uc(struct drm_i915_error_state_buf *m,
653 const struct intel_uc_coredump *error_uc)
655 struct drm_printer p = i915_error_printer(m);
657 intel_uc_fw_dump(&error_uc->guc_fw, &p);
658 intel_uc_fw_dump(&error_uc->huc_fw, &p);
659 print_error_vma(m, NULL, error_uc->guc_log);
662 static void err_free_sgl(struct scatterlist *sgl)
665 struct scatterlist *sg;
667 for (sg = sgl; !sg_is_chain(sg); sg++) {
673 sg = sg_is_last(sg) ? NULL : sg_chain_ptr(sg);
674 free_page((unsigned long)sgl);
679 static void err_print_gt_info(struct drm_i915_error_state_buf *m,
680 struct intel_gt_coredump *gt)
682 struct drm_printer p = i915_error_printer(m);
684 intel_gt_info_print(>->info, &p);
685 intel_sseu_print_topology(>->info.sseu, &p);
688 static void err_print_gt(struct drm_i915_error_state_buf *m,
689 struct intel_gt_coredump *gt)
691 const struct intel_engine_coredump *ee;
694 err_printf(m, "GT awake: %s\n", yesno(gt->awake));
695 err_printf(m, "EIR: 0x%08x\n", gt->eir);
696 err_printf(m, "IER: 0x%08x\n", gt->ier);
697 for (i = 0; i < gt->ngtier; i++)
698 err_printf(m, "GTIER[%d]: 0x%08x\n", i, gt->gtier[i]);
699 err_printf(m, "PGTBL_ER: 0x%08x\n", gt->pgtbl_er);
700 err_printf(m, "FORCEWAKE: 0x%08x\n", gt->forcewake);
701 err_printf(m, "DERRMR: 0x%08x\n", gt->derrmr);
703 for (i = 0; i < gt->nfence; i++)
704 err_printf(m, " fence[%d] = %08llx\n", i, gt->fence[i]);
706 if (IS_GEN_RANGE(m->i915, 6, 11)) {
707 err_printf(m, "ERROR: 0x%08x\n", gt->error);
708 err_printf(m, "DONE_REG: 0x%08x\n", gt->done_reg);
711 if (INTEL_GEN(m->i915) >= 8)
712 err_printf(m, "FAULT_TLB_DATA: 0x%08x 0x%08x\n",
713 gt->fault_data1, gt->fault_data0);
715 if (IS_GEN(m->i915, 7))
716 err_printf(m, "ERR_INT: 0x%08x\n", gt->err_int);
718 if (IS_GEN_RANGE(m->i915, 8, 11))
719 err_printf(m, "GTT_CACHE_EN: 0x%08x\n", gt->gtt_cache);
721 if (IS_GEN(m->i915, 12))
722 err_printf(m, "AUX_ERR_DBG: 0x%08x\n", gt->aux_err);
724 if (INTEL_GEN(m->i915) >= 12) {
727 for (i = 0; i < GEN12_SFC_DONE_MAX; i++)
728 err_printf(m, " SFC_DONE[%d]: 0x%08x\n", i,
731 err_printf(m, " GAM_DONE: 0x%08x\n", gt->gam_done);
734 for (ee = gt->engine; ee; ee = ee->next) {
735 const struct i915_vma_coredump *vma;
737 error_print_engine(m, ee);
738 for (vma = ee->vma; vma; vma = vma->next)
739 print_error_vma(m, ee->engine, vma);
743 err_print_uc(m, gt->uc);
745 err_print_gt_info(m, gt);
748 static void __err_print_to_sgl(struct drm_i915_error_state_buf *m,
749 struct i915_gpu_coredump *error)
751 const struct intel_engine_coredump *ee;
752 struct timespec64 ts;
754 if (*error->error_msg)
755 err_printf(m, "%s\n", error->error_msg);
756 err_printf(m, "Kernel: %s %s\n",
757 init_utsname()->release,
758 init_utsname()->machine);
759 err_printf(m, "Driver: %s\n", DRIVER_DATE);
760 ts = ktime_to_timespec64(error->time);
761 err_printf(m, "Time: %lld s %ld us\n",
762 (s64)ts.tv_sec, ts.tv_nsec / NSEC_PER_USEC);
763 ts = ktime_to_timespec64(error->boottime);
764 err_printf(m, "Boottime: %lld s %ld us\n",
765 (s64)ts.tv_sec, ts.tv_nsec / NSEC_PER_USEC);
766 ts = ktime_to_timespec64(error->uptime);
767 err_printf(m, "Uptime: %lld s %ld us\n",
768 (s64)ts.tv_sec, ts.tv_nsec / NSEC_PER_USEC);
769 err_printf(m, "Capture: %lu jiffies; %d ms ago\n",
770 error->capture, jiffies_to_msecs(jiffies - error->capture));
772 for (ee = error->gt ? error->gt->engine : NULL; ee; ee = ee->next)
773 err_printf(m, "Active process (on ring %s): %s [%d]\n",
778 err_printf(m, "Reset count: %u\n", error->reset_count);
779 err_printf(m, "Suspend count: %u\n", error->suspend_count);
780 err_printf(m, "Platform: %s\n", intel_platform_name(error->device_info.platform));
781 err_printf(m, "Subplatform: 0x%x\n",
782 intel_subplatform(&error->runtime_info,
783 error->device_info.platform));
784 err_print_pciid(m, m->i915);
786 err_printf(m, "IOMMU enabled?: %d\n", error->iommu);
788 if (HAS_CSR(m->i915)) {
789 struct intel_csr *csr = &m->i915->csr;
791 err_printf(m, "DMC loaded: %s\n",
792 yesno(csr->dmc_payload != NULL));
793 err_printf(m, "DMC fw version: %d.%d\n",
794 CSR_VERSION_MAJOR(csr->version),
795 CSR_VERSION_MINOR(csr->version));
798 err_printf(m, "RPM wakelock: %s\n", yesno(error->wakelock));
799 err_printf(m, "PM suspended: %s\n", yesno(error->suspended));
802 err_print_gt(m, error->gt);
805 intel_overlay_print_error_state(m, error->overlay);
808 intel_display_print_error_state(m, error->display);
810 err_print_capabilities(m, error);
811 err_print_params(m, &error->params);
814 static int err_print_to_sgl(struct i915_gpu_coredump *error)
816 struct drm_i915_error_state_buf m;
819 return PTR_ERR(error);
821 if (READ_ONCE(error->sgl))
824 memset(&m, 0, sizeof(m));
825 m.i915 = error->i915;
827 __err_print_to_sgl(&m, error);
830 __sg_set_buf(m.cur++, m.buf, m.bytes, m.iter);
835 GEM_BUG_ON(m.end < m.cur);
836 sg_mark_end(m.cur - 1);
838 GEM_BUG_ON(m.sgl && !m.cur);
845 if (cmpxchg(&error->sgl, NULL, m.sgl))
851 ssize_t i915_gpu_coredump_copy_to_buffer(struct i915_gpu_coredump *error,
852 char *buf, loff_t off, size_t rem)
854 struct scatterlist *sg;
862 err = err_print_to_sgl(error);
866 sg = READ_ONCE(error->fit);
867 if (!sg || off < sg->dma_address)
872 pos = sg->dma_address;
877 if (sg_is_chain(sg)) {
878 sg = sg_chain_ptr(sg);
879 GEM_BUG_ON(sg_is_chain(sg));
883 if (pos + len <= off) {
890 GEM_BUG_ON(off - pos > len);
897 GEM_BUG_ON(!len || len > sg->length);
899 memcpy(buf, page_address(sg_page(sg)) + start, len);
907 WRITE_ONCE(error->fit, sg);
910 } while (!sg_is_last(sg++));
915 static void i915_vma_coredump_free(struct i915_vma_coredump *vma)
918 struct i915_vma_coredump *next = vma->next;
921 for (page = 0; page < vma->page_count; page++)
922 free_page((unsigned long)vma->pages[page]);
929 static void cleanup_params(struct i915_gpu_coredump *error)
931 i915_params_free(&error->params);
934 static void cleanup_uc(struct intel_uc_coredump *uc)
936 kfree(uc->guc_fw.path);
937 kfree(uc->huc_fw.path);
938 i915_vma_coredump_free(uc->guc_log);
943 static void cleanup_gt(struct intel_gt_coredump *gt)
946 struct intel_engine_coredump *ee = gt->engine;
948 gt->engine = ee->next;
950 i915_vma_coredump_free(ee->vma);
960 void __i915_gpu_coredump_free(struct kref *error_ref)
962 struct i915_gpu_coredump *error =
963 container_of(error_ref, typeof(*error), ref);
966 struct intel_gt_coredump *gt = error->gt;
968 error->gt = gt->next;
972 kfree(error->overlay);
973 kfree(error->display);
975 cleanup_params(error);
977 err_free_sgl(error->sgl);
981 static struct i915_vma_coredump *
982 i915_vma_coredump_create(const struct intel_gt *gt,
983 const struct i915_vma *vma,
985 struct i915_vma_compress *compress)
987 struct i915_ggtt *ggtt = gt->ggtt;
988 const u64 slot = ggtt->error_capture.start;
989 struct i915_vma_coredump *dst;
990 unsigned long num_pages;
991 struct sgt_iter iter;
996 if (!vma || !vma->pages || !compress)
999 num_pages = min_t(u64, vma->size, vma->obj->base.size) >> PAGE_SHIFT;
1000 num_pages = DIV_ROUND_UP(10 * num_pages, 8); /* worstcase zlib growth */
1001 dst = kmalloc(sizeof(*dst) + num_pages * sizeof(u32 *), ALLOW_FAIL);
1005 if (!compress_start(compress)) {
1010 strcpy(dst->name, name);
1013 dst->gtt_offset = vma->node.start;
1014 dst->gtt_size = vma->node.size;
1015 dst->gtt_page_sizes = vma->page_sizes.gtt;
1016 dst->num_pages = num_pages;
1017 dst->page_count = 0;
1021 if (drm_mm_node_allocated(&ggtt->error_capture)) {
1025 for_each_sgt_daddr(dma, iter, vma->pages) {
1026 ggtt->vm.insert_page(&ggtt->vm, dma, slot,
1027 I915_CACHE_NONE, 0);
1030 s = io_mapping_map_wc(&ggtt->iomap, slot, PAGE_SIZE);
1031 ret = compress_page(compress,
1032 (void __force *)s, dst,
1034 io_mapping_unmap(s);
1038 } else if (i915_gem_object_is_lmem(vma->obj)) {
1039 struct intel_memory_region *mem = vma->obj->mm.region;
1042 for_each_sgt_daddr(dma, iter, vma->pages) {
1045 s = io_mapping_map_wc(&mem->iomap, dma, PAGE_SIZE);
1046 ret = compress_page(compress,
1047 (void __force *)s, dst,
1049 io_mapping_unmap(s);
1056 for_each_sgt_page(page, iter, vma->pages) {
1059 drm_clflush_pages(&page, 1);
1062 ret = compress_page(compress, s, dst, false);
1065 drm_clflush_pages(&page, 1);
1072 if (ret || compress_flush(compress, dst)) {
1073 while (dst->page_count--)
1074 pool_free(&compress->pool, dst->pages[dst->page_count]);
1078 compress_finish(compress);
1083 static void gt_record_fences(struct intel_gt_coredump *gt)
1085 struct i915_ggtt *ggtt = gt->_gt->ggtt;
1086 struct intel_uncore *uncore = gt->_gt->uncore;
1089 if (INTEL_GEN(uncore->i915) >= 6) {
1090 for (i = 0; i < ggtt->num_fences; i++)
1092 intel_uncore_read64(uncore,
1093 FENCE_REG_GEN6_LO(i));
1094 } else if (INTEL_GEN(uncore->i915) >= 4) {
1095 for (i = 0; i < ggtt->num_fences; i++)
1097 intel_uncore_read64(uncore,
1098 FENCE_REG_965_LO(i));
1100 for (i = 0; i < ggtt->num_fences; i++)
1102 intel_uncore_read(uncore, FENCE_REG(i));
1107 static void engine_record_registers(struct intel_engine_coredump *ee)
1109 const struct intel_engine_cs *engine = ee->engine;
1110 struct drm_i915_private *i915 = engine->i915;
1112 if (INTEL_GEN(i915) >= 6) {
1113 ee->rc_psmi = ENGINE_READ(engine, RING_PSMI_CTL);
1115 if (INTEL_GEN(i915) >= 12)
1116 ee->fault_reg = intel_uncore_read(engine->uncore,
1117 GEN12_RING_FAULT_REG);
1118 else if (INTEL_GEN(i915) >= 8)
1119 ee->fault_reg = intel_uncore_read(engine->uncore,
1120 GEN8_RING_FAULT_REG);
1122 ee->fault_reg = GEN6_RING_FAULT_REG_READ(engine);
1125 if (INTEL_GEN(i915) >= 4) {
1126 ee->esr = ENGINE_READ(engine, RING_ESR);
1127 ee->faddr = ENGINE_READ(engine, RING_DMA_FADD);
1128 ee->ipeir = ENGINE_READ(engine, RING_IPEIR);
1129 ee->ipehr = ENGINE_READ(engine, RING_IPEHR);
1130 ee->instps = ENGINE_READ(engine, RING_INSTPS);
1131 ee->bbaddr = ENGINE_READ(engine, RING_BBADDR);
1132 ee->ccid = ENGINE_READ(engine, CCID);
1133 if (INTEL_GEN(i915) >= 8) {
1134 ee->faddr |= (u64)ENGINE_READ(engine, RING_DMA_FADD_UDW) << 32;
1135 ee->bbaddr |= (u64)ENGINE_READ(engine, RING_BBADDR_UDW) << 32;
1137 ee->bbstate = ENGINE_READ(engine, RING_BBSTATE);
1139 ee->faddr = ENGINE_READ(engine, DMA_FADD_I8XX);
1140 ee->ipeir = ENGINE_READ(engine, IPEIR);
1141 ee->ipehr = ENGINE_READ(engine, IPEHR);
1144 intel_engine_get_instdone(engine, &ee->instdone);
1146 ee->instpm = ENGINE_READ(engine, RING_INSTPM);
1147 ee->acthd = intel_engine_get_active_head(engine);
1148 ee->start = ENGINE_READ(engine, RING_START);
1149 ee->head = ENGINE_READ(engine, RING_HEAD);
1150 ee->tail = ENGINE_READ(engine, RING_TAIL);
1151 ee->ctl = ENGINE_READ(engine, RING_CTL);
1152 if (INTEL_GEN(i915) > 2)
1153 ee->mode = ENGINE_READ(engine, RING_MI_MODE);
1155 if (!HWS_NEEDS_PHYSICAL(i915)) {
1158 if (IS_GEN(i915, 7)) {
1159 switch (engine->id) {
1161 MISSING_CASE(engine->id);
1164 mmio = RENDER_HWS_PGA_GEN7;
1167 mmio = BLT_HWS_PGA_GEN7;
1170 mmio = BSD_HWS_PGA_GEN7;
1173 mmio = VEBOX_HWS_PGA_GEN7;
1176 } else if (IS_GEN(engine->i915, 6)) {
1177 mmio = RING_HWS_PGA_GEN6(engine->mmio_base);
1179 /* XXX: gen8 returns to sanity */
1180 mmio = RING_HWS_PGA(engine->mmio_base);
1183 ee->hws = intel_uncore_read(engine->uncore, mmio);
1186 ee->reset_count = i915_reset_engine_count(&i915->gpu_error, engine);
1188 if (HAS_PPGTT(i915)) {
1191 ee->vm_info.gfx_mode = ENGINE_READ(engine, RING_MODE_GEN7);
1193 if (IS_GEN(i915, 6)) {
1194 ee->vm_info.pp_dir_base =
1195 ENGINE_READ(engine, RING_PP_DIR_BASE_READ);
1196 } else if (IS_GEN(i915, 7)) {
1197 ee->vm_info.pp_dir_base =
1198 ENGINE_READ(engine, RING_PP_DIR_BASE);
1199 } else if (INTEL_GEN(i915) >= 8) {
1200 u32 base = engine->mmio_base;
1202 for (i = 0; i < 4; i++) {
1203 ee->vm_info.pdp[i] =
1204 intel_uncore_read(engine->uncore,
1205 GEN8_RING_PDP_UDW(base, i));
1206 ee->vm_info.pdp[i] <<= 32;
1207 ee->vm_info.pdp[i] |=
1208 intel_uncore_read(engine->uncore,
1209 GEN8_RING_PDP_LDW(base, i));
1215 static void record_request(const struct i915_request *request,
1216 struct i915_request_coredump *erq)
1218 erq->flags = request->fence.flags;
1219 erq->context = request->fence.context;
1220 erq->seqno = request->fence.seqno;
1221 erq->sched_attr = request->sched.attr;
1222 erq->head = request->head;
1223 erq->tail = request->tail;
1227 if (!intel_context_is_closed(request->context)) {
1228 const struct i915_gem_context *ctx;
1230 ctx = rcu_dereference(request->context->gem_context);
1232 erq->pid = pid_nr(ctx->pid);
1237 static void engine_record_execlists(struct intel_engine_coredump *ee)
1239 const struct intel_engine_execlists * const el = &ee->engine->execlists;
1240 struct i915_request * const *port = el->active;
1244 record_request(*port++, &ee->execlist[n++]);
1249 static bool record_context(struct i915_gem_context_coredump *e,
1250 const struct i915_request *rq)
1252 struct i915_gem_context *ctx;
1253 struct task_struct *task;
1257 ctx = rcu_dereference(rq->context->gem_context);
1258 if (ctx && !kref_get_unless_zero(&ctx->ref))
1265 task = pid_task(ctx->pid, PIDTYPE_PID);
1267 strcpy(e->comm, task->comm);
1272 e->sched_attr = ctx->sched;
1273 e->guilty = atomic_read(&ctx->guilty_count);
1274 e->active = atomic_read(&ctx->active_count);
1276 e->total_runtime = rq->context->runtime.total;
1277 e->avg_runtime = ewma_runtime_read(&rq->context->runtime.avg);
1279 simulated = i915_gem_context_no_error_capture(ctx);
1281 i915_gem_context_put(ctx);
1285 struct intel_engine_capture_vma {
1286 struct intel_engine_capture_vma *next;
1287 struct i915_vma *vma;
1291 static struct intel_engine_capture_vma *
1292 capture_vma(struct intel_engine_capture_vma *next,
1293 struct i915_vma *vma,
1297 struct intel_engine_capture_vma *c;
1302 c = kmalloc(sizeof(*c), gfp);
1306 if (!i915_active_acquire_if_busy(&vma->active)) {
1311 strcpy(c->name, name);
1312 c->vma = i915_vma_get(vma);
1318 static struct intel_engine_capture_vma *
1319 capture_user(struct intel_engine_capture_vma *capture,
1320 const struct i915_request *rq,
1323 struct i915_capture_list *c;
1325 for (c = rq->capture_list; c; c = c->next)
1326 capture = capture_vma(capture, c->vma, "user", gfp);
1331 static void add_vma(struct intel_engine_coredump *ee,
1332 struct i915_vma_coredump *vma)
1335 vma->next = ee->vma;
1340 struct intel_engine_coredump *
1341 intel_engine_coredump_alloc(struct intel_engine_cs *engine, gfp_t gfp)
1343 struct intel_engine_coredump *ee;
1345 ee = kzalloc(sizeof(*ee), gfp);
1349 ee->engine = engine;
1351 engine_record_registers(ee);
1352 engine_record_execlists(ee);
1357 struct intel_engine_capture_vma *
1358 intel_engine_coredump_add_request(struct intel_engine_coredump *ee,
1359 struct i915_request *rq,
1362 struct intel_engine_capture_vma *vma = NULL;
1364 ee->simulated |= record_context(&ee->context, rq);
1369 * We need to copy these to an anonymous buffer
1370 * as the simplest method to avoid being overwritten
1373 vma = capture_vma(vma, rq->batch, "batch", gfp);
1374 vma = capture_user(vma, rq, gfp);
1375 vma = capture_vma(vma, rq->ring->vma, "ring", gfp);
1376 vma = capture_vma(vma, rq->context->state, "HW context", gfp);
1378 ee->rq_head = rq->head;
1379 ee->rq_post = rq->postfix;
1380 ee->rq_tail = rq->tail;
1386 intel_engine_coredump_add_vma(struct intel_engine_coredump *ee,
1387 struct intel_engine_capture_vma *capture,
1388 struct i915_vma_compress *compress)
1390 const struct intel_engine_cs *engine = ee->engine;
1393 struct intel_engine_capture_vma *this = capture;
1394 struct i915_vma *vma = this->vma;
1397 i915_vma_coredump_create(engine->gt,
1401 i915_active_release(&vma->active);
1404 capture = this->next;
1409 i915_vma_coredump_create(engine->gt,
1410 engine->status_page.vma,
1415 i915_vma_coredump_create(engine->gt,
1421 static struct intel_engine_coredump *
1422 capture_engine(struct intel_engine_cs *engine,
1423 struct i915_vma_compress *compress)
1425 struct intel_engine_capture_vma *capture = NULL;
1426 struct intel_engine_coredump *ee;
1427 struct i915_request *rq;
1428 unsigned long flags;
1430 ee = intel_engine_coredump_alloc(engine, GFP_KERNEL);
1434 spin_lock_irqsave(&engine->active.lock, flags);
1435 rq = intel_engine_find_active_request(engine);
1437 capture = intel_engine_coredump_add_request(ee, rq,
1439 spin_unlock_irqrestore(&engine->active.lock, flags);
1445 intel_engine_coredump_add_vma(ee, capture, compress);
1451 gt_record_engines(struct intel_gt_coredump *gt,
1452 struct i915_vma_compress *compress)
1454 struct intel_engine_cs *engine;
1455 enum intel_engine_id id;
1457 for_each_engine(engine, gt->_gt, id) {
1458 struct intel_engine_coredump *ee;
1460 /* Refill our page pool before entering atomic section */
1461 pool_refill(&compress->pool, ALLOW_FAIL);
1463 ee = capture_engine(engine, compress);
1467 gt->simulated |= ee->simulated;
1468 if (ee->simulated) {
1473 ee->next = gt->engine;
1478 static struct intel_uc_coredump *
1479 gt_record_uc(struct intel_gt_coredump *gt,
1480 struct i915_vma_compress *compress)
1482 const struct intel_uc *uc = >->_gt->uc;
1483 struct intel_uc_coredump *error_uc;
1485 error_uc = kzalloc(sizeof(*error_uc), ALLOW_FAIL);
1489 memcpy(&error_uc->guc_fw, &uc->guc.fw, sizeof(uc->guc.fw));
1490 memcpy(&error_uc->huc_fw, &uc->huc.fw, sizeof(uc->huc.fw));
1492 /* Non-default firmware paths will be specified by the modparam.
1493 * As modparams are generally accesible from the userspace make
1494 * explicit copies of the firmware paths.
1496 error_uc->guc_fw.path = kstrdup(uc->guc.fw.path, ALLOW_FAIL);
1497 error_uc->huc_fw.path = kstrdup(uc->huc.fw.path, ALLOW_FAIL);
1499 i915_vma_coredump_create(gt->_gt,
1500 uc->guc.log.vma, "GuC log buffer",
1506 static void gt_capture_prepare(struct intel_gt_coredump *gt)
1508 struct i915_ggtt *ggtt = gt->_gt->ggtt;
1510 mutex_lock(&ggtt->error_mutex);
1513 static void gt_capture_finish(struct intel_gt_coredump *gt)
1515 struct i915_ggtt *ggtt = gt->_gt->ggtt;
1517 if (drm_mm_node_allocated(&ggtt->error_capture))
1518 ggtt->vm.clear_range(&ggtt->vm,
1519 ggtt->error_capture.start,
1522 mutex_unlock(&ggtt->error_mutex);
1525 /* Capture all registers which don't fit into another category. */
1526 static void gt_record_regs(struct intel_gt_coredump *gt)
1528 struct intel_uncore *uncore = gt->_gt->uncore;
1529 struct drm_i915_private *i915 = uncore->i915;
1533 * General organization
1534 * 1. Registers specific to a single generation
1535 * 2. Registers which belong to multiple generations
1536 * 3. Feature specific registers.
1537 * 4. Everything else
1538 * Please try to follow the order.
1541 /* 1: Registers specific to a single generation */
1542 if (IS_VALLEYVIEW(i915)) {
1543 gt->gtier[0] = intel_uncore_read(uncore, GTIER);
1544 gt->ier = intel_uncore_read(uncore, VLV_IER);
1545 gt->forcewake = intel_uncore_read_fw(uncore, FORCEWAKE_VLV);
1548 if (IS_GEN(i915, 7))
1549 gt->err_int = intel_uncore_read(uncore, GEN7_ERR_INT);
1551 if (INTEL_GEN(i915) >= 12) {
1552 gt->fault_data0 = intel_uncore_read(uncore,
1553 GEN12_FAULT_TLB_DATA0);
1554 gt->fault_data1 = intel_uncore_read(uncore,
1555 GEN12_FAULT_TLB_DATA1);
1556 } else if (INTEL_GEN(i915) >= 8) {
1557 gt->fault_data0 = intel_uncore_read(uncore,
1558 GEN8_FAULT_TLB_DATA0);
1559 gt->fault_data1 = intel_uncore_read(uncore,
1560 GEN8_FAULT_TLB_DATA1);
1563 if (IS_GEN(i915, 6)) {
1564 gt->forcewake = intel_uncore_read_fw(uncore, FORCEWAKE);
1565 gt->gab_ctl = intel_uncore_read(uncore, GAB_CTL);
1566 gt->gfx_mode = intel_uncore_read(uncore, GFX_MODE);
1569 /* 2: Registers which belong to multiple generations */
1570 if (INTEL_GEN(i915) >= 7)
1571 gt->forcewake = intel_uncore_read_fw(uncore, FORCEWAKE_MT);
1573 if (INTEL_GEN(i915) >= 6) {
1574 gt->derrmr = intel_uncore_read(uncore, DERRMR);
1575 if (INTEL_GEN(i915) < 12) {
1576 gt->error = intel_uncore_read(uncore, ERROR_GEN6);
1577 gt->done_reg = intel_uncore_read(uncore, DONE_REG);
1581 /* 3: Feature specific registers */
1582 if (IS_GEN_RANGE(i915, 6, 7)) {
1583 gt->gam_ecochk = intel_uncore_read(uncore, GAM_ECOCHK);
1584 gt->gac_eco = intel_uncore_read(uncore, GAC_ECO_BITS);
1587 if (IS_GEN_RANGE(i915, 8, 11))
1588 gt->gtt_cache = intel_uncore_read(uncore, HSW_GTT_CACHE_EN);
1590 if (IS_GEN(i915, 12))
1591 gt->aux_err = intel_uncore_read(uncore, GEN12_AUX_ERR_DBG);
1593 if (INTEL_GEN(i915) >= 12) {
1594 for (i = 0; i < GEN12_SFC_DONE_MAX; i++) {
1596 intel_uncore_read(uncore, GEN12_SFC_DONE(i));
1599 gt->gam_done = intel_uncore_read(uncore, GEN12_GAM_DONE);
1602 /* 4: Everything else */
1603 if (INTEL_GEN(i915) >= 11) {
1604 gt->ier = intel_uncore_read(uncore, GEN8_DE_MISC_IER);
1606 intel_uncore_read(uncore,
1607 GEN11_RENDER_COPY_INTR_ENABLE);
1609 intel_uncore_read(uncore, GEN11_VCS_VECS_INTR_ENABLE);
1611 intel_uncore_read(uncore, GEN11_GUC_SG_INTR_ENABLE);
1613 intel_uncore_read(uncore,
1614 GEN11_GPM_WGBOXPERF_INTR_ENABLE);
1616 intel_uncore_read(uncore,
1617 GEN11_CRYPTO_RSVD_INTR_ENABLE);
1619 intel_uncore_read(uncore,
1620 GEN11_GUNIT_CSME_INTR_ENABLE);
1622 } else if (INTEL_GEN(i915) >= 8) {
1623 gt->ier = intel_uncore_read(uncore, GEN8_DE_MISC_IER);
1624 for (i = 0; i < 4; i++)
1626 intel_uncore_read(uncore, GEN8_GT_IER(i));
1628 } else if (HAS_PCH_SPLIT(i915)) {
1629 gt->ier = intel_uncore_read(uncore, DEIER);
1630 gt->gtier[0] = intel_uncore_read(uncore, GTIER);
1632 } else if (IS_GEN(i915, 2)) {
1633 gt->ier = intel_uncore_read16(uncore, GEN2_IER);
1634 } else if (!IS_VALLEYVIEW(i915)) {
1635 gt->ier = intel_uncore_read(uncore, GEN2_IER);
1637 gt->eir = intel_uncore_read(uncore, EIR);
1638 gt->pgtbl_er = intel_uncore_read(uncore, PGTBL_ER);
1641 static void gt_record_info(struct intel_gt_coredump *gt)
1643 memcpy(>->info, >->_gt->info, sizeof(struct intel_gt_info));
1647 * Generate a semi-unique error code. The code is not meant to have meaning, The
1648 * code's only purpose is to try to prevent false duplicated bug reports by
1649 * grossly estimating a GPU error state.
1651 * TODO Ideally, hashing the batchbuffer would be a very nice way to determine
1652 * the hang if we could strip the GTT offset information from it.
1654 * It's only a small step better than a random number in its current form.
1656 static u32 generate_ecode(const struct intel_engine_coredump *ee)
1659 * IPEHR would be an ideal way to detect errors, as it's the gross
1660 * measure of "the command that hung." However, has some very common
1661 * synchronization commands which almost always appear in the case
1662 * strictly a client bug. Use instdone to differentiate those some.
1664 return ee ? ee->ipehr ^ ee->instdone.instdone : 0;
1667 static const char *error_msg(struct i915_gpu_coredump *error)
1669 struct intel_engine_coredump *first = NULL;
1670 struct intel_gt_coredump *gt;
1671 intel_engine_mask_t engines;
1675 for (gt = error->gt; gt; gt = gt->next) {
1676 struct intel_engine_coredump *cs;
1678 if (gt->engine && !first)
1681 for (cs = gt->engine; cs; cs = cs->next)
1682 engines |= cs->engine->mask;
1685 len = scnprintf(error->error_msg, sizeof(error->error_msg),
1686 "GPU HANG: ecode %d:%x:%08x",
1687 INTEL_GEN(error->i915), engines,
1688 generate_ecode(first));
1689 if (first && first->context.pid) {
1690 /* Just show the first executing process, more is confusing */
1691 len += scnprintf(error->error_msg + len,
1692 sizeof(error->error_msg) - len,
1694 first->context.comm, first->context.pid);
1697 return error->error_msg;
1700 static void capture_gen(struct i915_gpu_coredump *error)
1702 struct drm_i915_private *i915 = error->i915;
1704 error->wakelock = atomic_read(&i915->runtime_pm.wakeref_count);
1705 error->suspended = i915->runtime_pm.suspended;
1708 #ifdef CONFIG_INTEL_IOMMU
1709 error->iommu = intel_iommu_gfx_mapped;
1711 error->reset_count = i915_reset_count(&i915->gpu_error);
1712 error->suspend_count = i915->suspend_count;
1714 i915_params_copy(&error->params, &i915->params);
1715 memcpy(&error->device_info,
1717 sizeof(error->device_info));
1718 memcpy(&error->runtime_info,
1720 sizeof(error->runtime_info));
1721 error->driver_caps = i915->caps;
1724 struct i915_gpu_coredump *
1725 i915_gpu_coredump_alloc(struct drm_i915_private *i915, gfp_t gfp)
1727 struct i915_gpu_coredump *error;
1729 if (!i915->params.error_capture)
1732 error = kzalloc(sizeof(*error), gfp);
1736 kref_init(&error->ref);
1739 error->time = ktime_get_real();
1740 error->boottime = ktime_get_boottime();
1741 error->uptime = ktime_sub(ktime_get(), i915->gt.last_init_time);
1742 error->capture = jiffies;
1749 #define DAY_AS_SECONDS(x) (24 * 60 * 60 * (x))
1751 struct intel_gt_coredump *
1752 intel_gt_coredump_alloc(struct intel_gt *gt, gfp_t gfp)
1754 struct intel_gt_coredump *gc;
1756 gc = kzalloc(sizeof(*gc), gfp);
1761 gc->awake = intel_gt_pm_is_awake(gt);
1764 gt_record_fences(gc);
1769 struct i915_vma_compress *
1770 i915_vma_capture_prepare(struct intel_gt_coredump *gt)
1772 struct i915_vma_compress *compress;
1774 compress = kmalloc(sizeof(*compress), ALLOW_FAIL);
1778 if (!compress_init(compress)) {
1783 gt_capture_prepare(gt);
1788 void i915_vma_capture_finish(struct intel_gt_coredump *gt,
1789 struct i915_vma_compress *compress)
1794 gt_capture_finish(gt);
1796 compress_fini(compress);
1800 struct i915_gpu_coredump *i915_gpu_coredump(struct drm_i915_private *i915)
1802 struct i915_gpu_coredump *error;
1804 /* Check if GPU capture has been disabled */
1805 error = READ_ONCE(i915->gpu_error.first_error);
1809 error = i915_gpu_coredump_alloc(i915, ALLOW_FAIL);
1811 return ERR_PTR(-ENOMEM);
1813 error->gt = intel_gt_coredump_alloc(&i915->gt, ALLOW_FAIL);
1815 struct i915_vma_compress *compress;
1817 compress = i915_vma_capture_prepare(error->gt);
1821 return ERR_PTR(-ENOMEM);
1824 gt_record_info(error->gt);
1825 gt_record_engines(error->gt, compress);
1827 if (INTEL_INFO(i915)->has_gt_uc)
1828 error->gt->uc = gt_record_uc(error->gt, compress);
1830 i915_vma_capture_finish(error->gt, compress);
1832 error->simulated |= error->gt->simulated;
1835 error->overlay = intel_overlay_capture_error_state(i915);
1836 error->display = intel_display_capture_error_state(i915);
1841 void i915_error_state_store(struct i915_gpu_coredump *error)
1843 struct drm_i915_private *i915;
1846 if (IS_ERR_OR_NULL(error))
1850 drm_info(&i915->drm, "%s\n", error_msg(error));
1852 if (error->simulated ||
1853 cmpxchg(&i915->gpu_error.first_error, NULL, error))
1856 i915_gpu_coredump_get(error);
1858 if (!xchg(&warned, true) &&
1859 ktime_get_real_seconds() - DRIVER_TIMESTAMP < DAY_AS_SECONDS(180)) {
1860 pr_info("GPU hangs can indicate a bug anywhere in the entire gfx stack, including userspace.\n");
1861 pr_info("Please file a _new_ bug report at https://gitlab.freedesktop.org/drm/intel/issues/new.\n");
1862 pr_info("Please see https://gitlab.freedesktop.org/drm/intel/-/wikis/How-to-file-i915-bugs for details.\n");
1863 pr_info("drm/i915 developers can then reassign to the right component if it's not a kernel issue.\n");
1864 pr_info("The GPU crash dump is required to analyze GPU hangs, so please always attach it.\n");
1865 pr_info("GPU crash dump saved to /sys/class/drm/card%d/error\n",
1866 i915->drm.primary->index);
1871 * i915_capture_error_state - capture an error record for later analysis
1872 * @i915: i915 device
1874 * Should be called when an error is detected (either a hang or an error
1875 * interrupt) to capture error state from the time of the error. Fills
1876 * out a structure which becomes available in debugfs for user level tools
1879 void i915_capture_error_state(struct drm_i915_private *i915)
1881 struct i915_gpu_coredump *error;
1883 error = i915_gpu_coredump(i915);
1884 if (IS_ERR(error)) {
1885 cmpxchg(&i915->gpu_error.first_error, NULL, error);
1889 i915_error_state_store(error);
1890 i915_gpu_coredump_put(error);
1893 struct i915_gpu_coredump *
1894 i915_first_error_state(struct drm_i915_private *i915)
1896 struct i915_gpu_coredump *error;
1898 spin_lock_irq(&i915->gpu_error.lock);
1899 error = i915->gpu_error.first_error;
1900 if (!IS_ERR_OR_NULL(error))
1901 i915_gpu_coredump_get(error);
1902 spin_unlock_irq(&i915->gpu_error.lock);
1907 void i915_reset_error_state(struct drm_i915_private *i915)
1909 struct i915_gpu_coredump *error;
1911 spin_lock_irq(&i915->gpu_error.lock);
1912 error = i915->gpu_error.first_error;
1913 if (error != ERR_PTR(-ENODEV)) /* if disabled, always disabled */
1914 i915->gpu_error.first_error = NULL;
1915 spin_unlock_irq(&i915->gpu_error.lock);
1917 if (!IS_ERR_OR_NULL(error))
1918 i915_gpu_coredump_put(error);
1921 void i915_disable_error_state(struct drm_i915_private *i915, int err)
1923 spin_lock_irq(&i915->gpu_error.lock);
1924 if (!i915->gpu_error.first_error)
1925 i915->gpu_error.first_error = ERR_PTR(err);
1926 spin_unlock_irq(&i915->gpu_error.lock);