1 /*
2  * Copyright (c) 2008 Intel Corporation
3  *
4  * Permission is hereby granted, free of charge, to any person obtaining a
5  * copy of this software and associated documentation files (the "Software"),
6  * to deal in the Software without restriction, including without limitation
7  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8  * and/or sell copies of the Software, and to permit persons to whom the
9  * Software is furnished to do so, subject to the following conditions:
10  *
11  * The above copyright notice and this permission notice (including the next
12  * paragraph) shall be included in all copies or substantial portions of the
13  * Software.
14  *
15  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
18  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
20  * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
21  * IN THE SOFTWARE.
22  *
23  * Authors:
24  *    Eric Anholt <eric@anholt.net>
25  *    Keith Packard <keithp@keithp.com>
26  *    Mika Kuoppala <mika.kuoppala@intel.com>
27  *
28  */
29 
30 #include <linux/ascii85.h>
31 #include <linux/nmi.h>
32 #include <linux/pagevec.h>
33 #include <linux/scatterlist.h>
34 #include <linux/utsname.h>
35 #include <linux/zlib.h>
36 
37 #include <drm/drm_cache.h>
38 #include <drm/drm_print.h>
39 
40 #include "display/intel_dmc.h"
41 #include "display/intel_overlay.h"
42 
43 #include "gem/i915_gem_context.h"
44 #include "gem/i915_gem_lmem.h"
45 #include "gt/intel_engine_regs.h"
46 #include "gt/intel_gt.h"
47 #include "gt/intel_gt_pm.h"
48 #include "gt/intel_gt_regs.h"
49 
50 #include "i915_driver.h"
51 #include "i915_drv.h"
52 #include "i915_gpu_error.h"
53 #include "i915_memcpy.h"
54 #include "i915_scatterlist.h"
55 
56 #define ALLOW_FAIL (__GFP_KSWAPD_RECLAIM | __GFP_RETRY_MAYFAIL | __GFP_NOWARN)
57 #define ATOMIC_MAYFAIL (GFP_ATOMIC | __GFP_NOWARN)
58 
59 static void __sg_set_buf(struct scatterlist *sg,
60 			 void *addr, unsigned int len, loff_t it)
61 {
62 	sg->page_link = (unsigned long)virt_to_page(addr);
63 	sg->offset = offset_in_page(addr);
64 	sg->length = len;
65 	sg->dma_address = it;
66 }
67 
68 static bool __i915_error_grow(struct drm_i915_error_state_buf *e, size_t len)
69 {
70 	if (!len)
71 		return false;
72 
73 	if (e->bytes + len + 1 <= e->size)
74 		return true;
75 
76 	if (e->bytes) {
77 		__sg_set_buf(e->cur++, e->buf, e->bytes, e->iter);
78 		e->iter += e->bytes;
79 		e->buf = NULL;
80 		e->bytes = 0;
81 	}
82 
83 	if (e->cur == e->end) {
84 		struct scatterlist *sgl;
85 
86 		sgl = (typeof(sgl))__get_free_page(ALLOW_FAIL);
87 		if (!sgl) {
88 			e->err = -ENOMEM;
89 			return false;
90 		}
91 
92 		if (e->cur) {
93 			e->cur->offset = 0;
94 			e->cur->length = 0;
95 			e->cur->page_link =
96 				(unsigned long)sgl | SG_CHAIN;
97 		} else {
98 			e->sgl = sgl;
99 		}
100 
101 		e->cur = sgl;
102 		e->end = sgl + SG_MAX_SINGLE_ALLOC - 1;
103 	}
104 
105 	e->size = ALIGN(len + 1, SZ_64K);
106 	e->buf = kmalloc(e->size, ALLOW_FAIL);
107 	if (!e->buf) {
108 		e->size = PAGE_ALIGN(len + 1);
109 		e->buf = kmalloc(e->size, GFP_KERNEL);
110 	}
111 	if (!e->buf) {
112 		e->err = -ENOMEM;
113 		return false;
114 	}
115 
116 	return true;
117 }
118 
119 __printf(2, 0)
120 static void i915_error_vprintf(struct drm_i915_error_state_buf *e,
121 			       const char *fmt, va_list args)
122 {
123 	va_list ap;
124 	int len;
125 
126 	if (e->err)
127 		return;
128 
129 	va_copy(ap, args);
130 	len = vsnprintf(NULL, 0, fmt, ap);
131 	va_end(ap);
132 	if (len <= 0) {
133 		e->err = len;
134 		return;
135 	}
136 
137 	if (!__i915_error_grow(e, len))
138 		return;
139 
140 	GEM_BUG_ON(e->bytes >= e->size);
141 	len = vscnprintf(e->buf + e->bytes, e->size - e->bytes, fmt, args);
142 	if (len < 0) {
143 		e->err = len;
144 		return;
145 	}
146 	e->bytes += len;
147 }
148 
149 static void i915_error_puts(struct drm_i915_error_state_buf *e, const char *str)
150 {
151 	unsigned len;
152 
153 	if (e->err || !str)
154 		return;
155 
156 	len = strlen(str);
157 	if (!__i915_error_grow(e, len))
158 		return;
159 
160 	GEM_BUG_ON(e->bytes + len > e->size);
161 	memcpy(e->buf + e->bytes, str, len);
162 	e->bytes += len;
163 }
164 
165 #define err_printf(e, ...) i915_error_printf(e, __VA_ARGS__)
166 #define err_puts(e, s) i915_error_puts(e, s)
167 
168 static void __i915_printfn_error(struct drm_printer *p, struct va_format *vaf)
169 {
170 	i915_error_vprintf(p->arg, vaf->fmt, *vaf->va);
171 }
172 
173 static inline struct drm_printer
174 i915_error_printer(struct drm_i915_error_state_buf *e)
175 {
176 	struct drm_printer p = {
177 		.printfn = __i915_printfn_error,
178 		.arg = e,
179 	};
180 	return p;
181 }
182 
183 /* single threaded page allocator with a reserved stash for emergencies */
184 static void pool_fini(struct pagevec *pv)
185 {
186 	pagevec_release(pv);
187 }
188 
189 static int pool_refill(struct pagevec *pv, gfp_t gfp)
190 {
191 	while (pagevec_space(pv)) {
192 		struct page *p;
193 
194 		p = alloc_page(gfp);
195 		if (!p)
196 			return -ENOMEM;
197 
198 		pagevec_add(pv, p);
199 	}
200 
201 	return 0;
202 }
203 
204 static int pool_init(struct pagevec *pv, gfp_t gfp)
205 {
206 	int err;
207 
208 	pagevec_init(pv);
209 
210 	err = pool_refill(pv, gfp);
211 	if (err)
212 		pool_fini(pv);
213 
214 	return err;
215 }
216 
217 static void *pool_alloc(struct pagevec *pv, gfp_t gfp)
218 {
219 	struct page *p;
220 
221 	p = alloc_page(gfp);
222 	if (!p && pagevec_count(pv))
223 		p = pv->pages[--pv->nr];
224 
225 	return p ? page_address(p) : NULL;
226 }
227 
228 static void pool_free(struct pagevec *pv, void *addr)
229 {
230 	struct page *p = virt_to_page(addr);
231 
232 	if (pagevec_space(pv))
233 		pagevec_add(pv, p);
234 	else
235 		__free_page(p);
236 }
237 
238 #ifdef CONFIG_DRM_I915_COMPRESS_ERROR
239 
240 struct i915_vma_compress {
241 	struct pagevec pool;
242 	struct z_stream_s zstream;
243 	void *tmp;
244 };
245 
246 static bool compress_init(struct i915_vma_compress *c)
247 {
248 	struct z_stream_s *zstream = &c->zstream;
249 
250 	if (pool_init(&c->pool, ALLOW_FAIL))
251 		return false;
252 
253 	zstream->workspace =
254 		kmalloc(zlib_deflate_workspacesize(MAX_WBITS, MAX_MEM_LEVEL),
255 			ALLOW_FAIL);
256 	if (!zstream->workspace) {
257 		pool_fini(&c->pool);
258 		return false;
259 	}
260 
261 	c->tmp = NULL;
262 	if (i915_has_memcpy_from_wc())
263 		c->tmp = pool_alloc(&c->pool, ALLOW_FAIL);
264 
265 	return true;
266 }
267 
268 static bool compress_start(struct i915_vma_compress *c)
269 {
270 	struct z_stream_s *zstream = &c->zstream;
271 	void *workspace = zstream->workspace;
272 
273 	memset(zstream, 0, sizeof(*zstream));
274 	zstream->workspace = workspace;
275 
276 	return zlib_deflateInit(zstream, Z_DEFAULT_COMPRESSION) == Z_OK;
277 }
278 
279 static void *compress_next_page(struct i915_vma_compress *c,
280 				struct i915_vma_coredump *dst)
281 {
282 	void *page_addr;
283 	struct page *page;
284 
285 	page_addr = pool_alloc(&c->pool, ALLOW_FAIL);
286 	if (!page_addr)
287 		return ERR_PTR(-ENOMEM);
288 
289 	page = virt_to_page(page_addr);
290 	list_add_tail(&page->lru, &dst->page_list);
291 	return page_addr;
292 }
293 
294 static int compress_page(struct i915_vma_compress *c,
295 			 void *src,
296 			 struct i915_vma_coredump *dst,
297 			 bool wc)
298 {
299 	struct z_stream_s *zstream = &c->zstream;
300 
301 	zstream->next_in = src;
302 	if (wc && c->tmp && i915_memcpy_from_wc(c->tmp, src, PAGE_SIZE))
303 		zstream->next_in = c->tmp;
304 	zstream->avail_in = PAGE_SIZE;
305 
306 	do {
307 		if (zstream->avail_out == 0) {
308 			zstream->next_out = compress_next_page(c, dst);
309 			if (IS_ERR(zstream->next_out))
310 				return PTR_ERR(zstream->next_out);
311 
312 			zstream->avail_out = PAGE_SIZE;
313 		}
314 
315 		if (zlib_deflate(zstream, Z_NO_FLUSH) != Z_OK)
316 			return -EIO;
317 
318 		cond_resched();
319 	} while (zstream->avail_in);
320 
321 	/* Fallback to uncompressed if we increase size? */
322 	if (0 && zstream->total_out > zstream->total_in)
323 		return -E2BIG;
324 
325 	return 0;
326 }
327 
328 static int compress_flush(struct i915_vma_compress *c,
329 			  struct i915_vma_coredump *dst)
330 {
331 	struct z_stream_s *zstream = &c->zstream;
332 
333 	do {
334 		switch (zlib_deflate(zstream, Z_FINISH)) {
335 		case Z_OK: /* more space requested */
336 			zstream->next_out = compress_next_page(c, dst);
337 			if (IS_ERR(zstream->next_out))
338 				return PTR_ERR(zstream->next_out);
339 
340 			zstream->avail_out = PAGE_SIZE;
341 			break;
342 
343 		case Z_STREAM_END:
344 			goto end;
345 
346 		default: /* any error */
347 			return -EIO;
348 		}
349 	} while (1);
350 
351 end:
352 	memset(zstream->next_out, 0, zstream->avail_out);
353 	dst->unused = zstream->avail_out;
354 	return 0;
355 }
356 
357 static void compress_finish(struct i915_vma_compress *c)
358 {
359 	zlib_deflateEnd(&c->zstream);
360 }
361 
362 static void compress_fini(struct i915_vma_compress *c)
363 {
364 	kfree(c->zstream.workspace);
365 	if (c->tmp)
366 		pool_free(&c->pool, c->tmp);
367 	pool_fini(&c->pool);
368 }
369 
370 static void err_compression_marker(struct drm_i915_error_state_buf *m)
371 {
372 	err_puts(m, ":");
373 }
374 
375 #else
376 
377 struct i915_vma_compress {
378 	struct pagevec pool;
379 };
380 
381 static bool compress_init(struct i915_vma_compress *c)
382 {
383 	return pool_init(&c->pool, ALLOW_FAIL) == 0;
384 }
385 
386 static bool compress_start(struct i915_vma_compress *c)
387 {
388 	return true;
389 }
390 
391 static int compress_page(struct i915_vma_compress *c,
392 			 void *src,
393 			 struct i915_vma_coredump *dst,
394 			 bool wc)
395 {
396 	void *ptr;
397 
398 	ptr = pool_alloc(&c->pool, ALLOW_FAIL);
399 	if (!ptr)
400 		return -ENOMEM;
401 
402 	if (!(wc && i915_memcpy_from_wc(ptr, src, PAGE_SIZE)))
403 		memcpy(ptr, src, PAGE_SIZE);
404 	list_add_tail(&virt_to_page(ptr)->lru, &dst->page_list);
405 	cond_resched();
406 
407 	return 0;
408 }
409 
410 static int compress_flush(struct i915_vma_compress *c,
411 			  struct i915_vma_coredump *dst)
412 {
413 	return 0;
414 }
415 
416 static void compress_finish(struct i915_vma_compress *c)
417 {
418 }
419 
420 static void compress_fini(struct i915_vma_compress *c)
421 {
422 	pool_fini(&c->pool);
423 }
424 
425 static void err_compression_marker(struct drm_i915_error_state_buf *m)
426 {
427 	err_puts(m, "~");
428 }
429 
430 #endif
431 
432 static void error_print_instdone(struct drm_i915_error_state_buf *m,
433 				 const struct intel_engine_coredump *ee)
434 {
435 	const struct sseu_dev_info *sseu = &ee->engine->gt->info.sseu;
436 	int slice;
437 	int subslice;
438 	int iter;
439 
440 	err_printf(m, "  INSTDONE: 0x%08x\n",
441 		   ee->instdone.instdone);
442 
443 	if (ee->engine->class != RENDER_CLASS || GRAPHICS_VER(m->i915) <= 3)
444 		return;
445 
446 	err_printf(m, "  SC_INSTDONE: 0x%08x\n",
447 		   ee->instdone.slice_common);
448 
449 	if (GRAPHICS_VER(m->i915) <= 6)
450 		return;
451 
452 	if (GRAPHICS_VER_FULL(m->i915) >= IP_VER(12, 50)) {
453 		for_each_instdone_gslice_dss_xehp(m->i915, sseu, iter, slice, subslice)
454 			err_printf(m, "  SAMPLER_INSTDONE[%d][%d]: 0x%08x\n",
455 				   slice, subslice,
456 				   ee->instdone.sampler[slice][subslice]);
457 
458 		for_each_instdone_gslice_dss_xehp(m->i915, sseu, iter, slice, subslice)
459 			err_printf(m, "  ROW_INSTDONE[%d][%d]: 0x%08x\n",
460 				   slice, subslice,
461 				   ee->instdone.row[slice][subslice]);
462 	} else {
463 		for_each_instdone_slice_subslice(m->i915, sseu, slice, subslice)
464 			err_printf(m, "  SAMPLER_INSTDONE[%d][%d]: 0x%08x\n",
465 				   slice, subslice,
466 				   ee->instdone.sampler[slice][subslice]);
467 
468 		for_each_instdone_slice_subslice(m->i915, sseu, slice, subslice)
469 			err_printf(m, "  ROW_INSTDONE[%d][%d]: 0x%08x\n",
470 				   slice, subslice,
471 				   ee->instdone.row[slice][subslice]);
472 	}
473 
474 	if (GRAPHICS_VER(m->i915) < 12)
475 		return;
476 
477 	if (GRAPHICS_VER_FULL(m->i915) >= IP_VER(12, 55)) {
478 		for_each_instdone_gslice_dss_xehp(m->i915, sseu, iter, slice, subslice)
479 			err_printf(m, "  GEOM_SVGUNIT_INSTDONE[%d][%d]: 0x%08x\n",
480 				   slice, subslice,
481 				   ee->instdone.geom_svg[slice][subslice]);
482 	}
483 
484 	err_printf(m, "  SC_INSTDONE_EXTRA: 0x%08x\n",
485 		   ee->instdone.slice_common_extra[0]);
486 	err_printf(m, "  SC_INSTDONE_EXTRA2: 0x%08x\n",
487 		   ee->instdone.slice_common_extra[1]);
488 }
489 
490 static void error_print_request(struct drm_i915_error_state_buf *m,
491 				const char *prefix,
492 				const struct i915_request_coredump *erq)
493 {
494 	if (!erq->seqno)
495 		return;
496 
497 	err_printf(m, "%s pid %d, seqno %8x:%08x%s%s, prio %d, head %08x, tail %08x\n",
498 		   prefix, erq->pid, erq->context, erq->seqno,
499 		   test_bit(DMA_FENCE_FLAG_SIGNALED_BIT,
500 			    &erq->flags) ? "!" : "",
501 		   test_bit(DMA_FENCE_FLAG_ENABLE_SIGNAL_BIT,
502 			    &erq->flags) ? "+" : "",
503 		   erq->sched_attr.priority,
504 		   erq->head, erq->tail);
505 }
506 
507 static void error_print_context(struct drm_i915_error_state_buf *m,
508 				const char *header,
509 				const struct i915_gem_context_coredump *ctx)
510 {
511 	const u32 period = to_gt(m->i915)->clock_period_ns;
512 
513 	err_printf(m, "%s%s[%d] prio %d, guilty %d active %d, runtime total %lluns, avg %lluns\n",
514 		   header, ctx->comm, ctx->pid, ctx->sched_attr.priority,
515 		   ctx->guilty, ctx->active,
516 		   ctx->total_runtime * period,
517 		   mul_u32_u32(ctx->avg_runtime, period));
518 }
519 
520 static struct i915_vma_coredump *
521 __find_vma(struct i915_vma_coredump *vma, const char *name)
522 {
523 	while (vma) {
524 		if (strcmp(vma->name, name) == 0)
525 			return vma;
526 		vma = vma->next;
527 	}
528 
529 	return NULL;
530 }
531 
532 static struct i915_vma_coredump *
533 find_batch(const struct intel_engine_coredump *ee)
534 {
535 	return __find_vma(ee->vma, "batch");
536 }
537 
538 static void error_print_engine(struct drm_i915_error_state_buf *m,
539 			       const struct intel_engine_coredump *ee)
540 {
541 	struct i915_vma_coredump *batch;
542 	int n;
543 
544 	err_printf(m, "%s command stream:\n", ee->engine->name);
545 	err_printf(m, "  CCID:  0x%08x\n", ee->ccid);
546 	err_printf(m, "  START: 0x%08x\n", ee->start);
547 	err_printf(m, "  HEAD:  0x%08x [0x%08x]\n", ee->head, ee->rq_head);
548 	err_printf(m, "  TAIL:  0x%08x [0x%08x, 0x%08x]\n",
549 		   ee->tail, ee->rq_post, ee->rq_tail);
550 	err_printf(m, "  CTL:   0x%08x\n", ee->ctl);
551 	err_printf(m, "  MODE:  0x%08x\n", ee->mode);
552 	err_printf(m, "  HWS:   0x%08x\n", ee->hws);
553 	err_printf(m, "  ACTHD: 0x%08x %08x\n",
554 		   (u32)(ee->acthd>>32), (u32)ee->acthd);
555 	err_printf(m, "  IPEIR: 0x%08x\n", ee->ipeir);
556 	err_printf(m, "  IPEHR: 0x%08x\n", ee->ipehr);
557 	err_printf(m, "  ESR:   0x%08x\n", ee->esr);
558 
559 	error_print_instdone(m, ee);
560 
561 	batch = find_batch(ee);
562 	if (batch) {
563 		u64 start = batch->gtt_offset;
564 		u64 end = start + batch->gtt_size;
565 
566 		err_printf(m, "  batch: [0x%08x_%08x, 0x%08x_%08x]\n",
567 			   upper_32_bits(start), lower_32_bits(start),
568 			   upper_32_bits(end), lower_32_bits(end));
569 	}
570 	if (GRAPHICS_VER(m->i915) >= 4) {
571 		err_printf(m, "  BBADDR: 0x%08x_%08x\n",
572 			   (u32)(ee->bbaddr>>32), (u32)ee->bbaddr);
573 		err_printf(m, "  BB_STATE: 0x%08x\n", ee->bbstate);
574 		err_printf(m, "  INSTPS: 0x%08x\n", ee->instps);
575 	}
576 	err_printf(m, "  INSTPM: 0x%08x\n", ee->instpm);
577 	err_printf(m, "  FADDR: 0x%08x %08x\n", upper_32_bits(ee->faddr),
578 		   lower_32_bits(ee->faddr));
579 	if (GRAPHICS_VER(m->i915) >= 6) {
580 		err_printf(m, "  RC PSMI: 0x%08x\n", ee->rc_psmi);
581 		err_printf(m, "  FAULT_REG: 0x%08x\n", ee->fault_reg);
582 	}
583 	if (HAS_PPGTT(m->i915)) {
584 		err_printf(m, "  GFX_MODE: 0x%08x\n", ee->vm_info.gfx_mode);
585 
586 		if (GRAPHICS_VER(m->i915) >= 8) {
587 			int i;
588 			for (i = 0; i < 4; i++)
589 				err_printf(m, "  PDP%d: 0x%016llx\n",
590 					   i, ee->vm_info.pdp[i]);
591 		} else {
592 			err_printf(m, "  PP_DIR_BASE: 0x%08x\n",
593 				   ee->vm_info.pp_dir_base);
594 		}
595 	}
596 	err_printf(m, "  hung: %u\n", ee->hung);
597 	err_printf(m, "  engine reset count: %u\n", ee->reset_count);
598 
599 	for (n = 0; n < ee->num_ports; n++) {
600 		err_printf(m, "  ELSP[%d]:", n);
601 		error_print_request(m, " ", &ee->execlist[n]);
602 	}
603 
604 	error_print_context(m, "  Active context: ", &ee->context);
605 }
606 
607 void i915_error_printf(struct drm_i915_error_state_buf *e, const char *f, ...)
608 {
609 	va_list args;
610 
611 	va_start(args, f);
612 	i915_error_vprintf(e, f, args);
613 	va_end(args);
614 }
615 
616 static void print_error_vma(struct drm_i915_error_state_buf *m,
617 			    const struct intel_engine_cs *engine,
618 			    const struct i915_vma_coredump *vma)
619 {
620 	char out[ASCII85_BUFSZ];
621 	struct page *page;
622 
623 	if (!vma)
624 		return;
625 
626 	err_printf(m, "%s --- %s = 0x%08x %08x\n",
627 		   engine ? engine->name : "global", vma->name,
628 		   upper_32_bits(vma->gtt_offset),
629 		   lower_32_bits(vma->gtt_offset));
630 
631 	if (vma->gtt_page_sizes > I915_GTT_PAGE_SIZE_4K)
632 		err_printf(m, "gtt_page_sizes = 0x%08x\n", vma->gtt_page_sizes);
633 
634 	err_compression_marker(m);
635 	list_for_each_entry(page, &vma->page_list, lru) {
636 		int i, len;
637 		const u32 *addr = page_address(page);
638 
639 		len = PAGE_SIZE;
640 		if (page == list_last_entry(&vma->page_list, typeof(*page), lru))
641 			len -= vma->unused;
642 		len = ascii85_encode_len(len);
643 
644 		for (i = 0; i < len; i++)
645 			err_puts(m, ascii85_encode(addr[i], out));
646 	}
647 	err_puts(m, "\n");
648 }
649 
650 static void err_print_capabilities(struct drm_i915_error_state_buf *m,
651 				   struct i915_gpu_coredump *error)
652 {
653 	struct drm_printer p = i915_error_printer(m);
654 
655 	intel_device_info_print_static(&error->device_info, &p);
656 	intel_device_info_print_runtime(&error->runtime_info, &p);
657 	intel_driver_caps_print(&error->driver_caps, &p);
658 }
659 
660 static void err_print_params(struct drm_i915_error_state_buf *m,
661 			     const struct i915_params *params)
662 {
663 	struct drm_printer p = i915_error_printer(m);
664 
665 	i915_params_dump(params, &p);
666 }
667 
668 static void err_print_pciid(struct drm_i915_error_state_buf *m,
669 			    struct drm_i915_private *i915)
670 {
671 	struct pci_dev *pdev = to_pci_dev(i915->drm.dev);
672 
673 	err_printf(m, "PCI ID: 0x%04x\n", pdev->device);
674 	err_printf(m, "PCI Revision: 0x%02x\n", pdev->revision);
675 	err_printf(m, "PCI Subsystem: %04x:%04x\n",
676 		   pdev->subsystem_vendor,
677 		   pdev->subsystem_device);
678 }
679 
680 static void err_print_uc(struct drm_i915_error_state_buf *m,
681 			 const struct intel_uc_coredump *error_uc)
682 {
683 	struct drm_printer p = i915_error_printer(m);
684 
685 	intel_uc_fw_dump(&error_uc->guc_fw, &p);
686 	intel_uc_fw_dump(&error_uc->huc_fw, &p);
687 	print_error_vma(m, NULL, error_uc->guc_log);
688 }
689 
690 static void err_free_sgl(struct scatterlist *sgl)
691 {
692 	while (sgl) {
693 		struct scatterlist *sg;
694 
695 		for (sg = sgl; !sg_is_chain(sg); sg++) {
696 			kfree(sg_virt(sg));
697 			if (sg_is_last(sg))
698 				break;
699 		}
700 
701 		sg = sg_is_last(sg) ? NULL : sg_chain_ptr(sg);
702 		free_page((unsigned long)sgl);
703 		sgl = sg;
704 	}
705 }
706 
707 static void err_print_gt_info(struct drm_i915_error_state_buf *m,
708 			      struct intel_gt_coredump *gt)
709 {
710 	struct drm_printer p = i915_error_printer(m);
711 
712 	intel_gt_info_print(&gt->info, &p);
713 	intel_sseu_print_topology(&gt->info.sseu, &p);
714 }
715 
716 static void err_print_gt(struct drm_i915_error_state_buf *m,
717 			 struct intel_gt_coredump *gt)
718 {
719 	const struct intel_engine_coredump *ee;
720 	int i;
721 
722 	err_printf(m, "GT awake: %s\n", yesno(gt->awake));
723 	err_printf(m, "EIR: 0x%08x\n", gt->eir);
724 	err_printf(m, "IER: 0x%08x\n", gt->ier);
725 	for (i = 0; i < gt->ngtier; i++)
726 		err_printf(m, "GTIER[%d]: 0x%08x\n", i, gt->gtier[i]);
727 	err_printf(m, "PGTBL_ER: 0x%08x\n", gt->pgtbl_er);
728 	err_printf(m, "FORCEWAKE: 0x%08x\n", gt->forcewake);
729 	err_printf(m, "DERRMR: 0x%08x\n", gt->derrmr);
730 
731 	for (i = 0; i < gt->nfence; i++)
732 		err_printf(m, "  fence[%d] = %08llx\n", i, gt->fence[i]);
733 
734 	if (IS_GRAPHICS_VER(m->i915, 6, 11)) {
735 		err_printf(m, "ERROR: 0x%08x\n", gt->error);
736 		err_printf(m, "DONE_REG: 0x%08x\n", gt->done_reg);
737 	}
738 
739 	if (GRAPHICS_VER(m->i915) >= 8)
740 		err_printf(m, "FAULT_TLB_DATA: 0x%08x 0x%08x\n",
741 			   gt->fault_data1, gt->fault_data0);
742 
743 	if (GRAPHICS_VER(m->i915) == 7)
744 		err_printf(m, "ERR_INT: 0x%08x\n", gt->err_int);
745 
746 	if (IS_GRAPHICS_VER(m->i915, 8, 11))
747 		err_printf(m, "GTT_CACHE_EN: 0x%08x\n", gt->gtt_cache);
748 
749 	if (GRAPHICS_VER(m->i915) == 12)
750 		err_printf(m, "AUX_ERR_DBG: 0x%08x\n", gt->aux_err);
751 
752 	if (GRAPHICS_VER(m->i915) >= 12) {
753 		int i;
754 
755 		for (i = 0; i < GEN12_SFC_DONE_MAX; i++) {
756 			/*
757 			 * SFC_DONE resides in the VD forcewake domain, so it
758 			 * only exists if the corresponding VCS engine is
759 			 * present.
760 			 */
761 			if ((gt->_gt->info.sfc_mask & BIT(i)) == 0 ||
762 			    !HAS_ENGINE(gt->_gt, _VCS(i * 2)))
763 				continue;
764 
765 			err_printf(m, "  SFC_DONE[%d]: 0x%08x\n", i,
766 				   gt->sfc_done[i]);
767 		}
768 
769 		err_printf(m, "  GAM_DONE: 0x%08x\n", gt->gam_done);
770 	}
771 
772 	for (ee = gt->engine; ee; ee = ee->next) {
773 		const struct i915_vma_coredump *vma;
774 
775 		error_print_engine(m, ee);
776 		for (vma = ee->vma; vma; vma = vma->next)
777 			print_error_vma(m, ee->engine, vma);
778 	}
779 
780 	if (gt->uc)
781 		err_print_uc(m, gt->uc);
782 
783 	err_print_gt_info(m, gt);
784 }
785 
786 static void __err_print_to_sgl(struct drm_i915_error_state_buf *m,
787 			       struct i915_gpu_coredump *error)
788 {
789 	const struct intel_engine_coredump *ee;
790 	struct timespec64 ts;
791 
792 	if (*error->error_msg)
793 		err_printf(m, "%s\n", error->error_msg);
794 	err_printf(m, "Kernel: %s %s\n",
795 		   init_utsname()->release,
796 		   init_utsname()->machine);
797 	err_printf(m, "Driver: %s\n", DRIVER_DATE);
798 	ts = ktime_to_timespec64(error->time);
799 	err_printf(m, "Time: %lld s %ld us\n",
800 		   (s64)ts.tv_sec, ts.tv_nsec / NSEC_PER_USEC);
801 	ts = ktime_to_timespec64(error->boottime);
802 	err_printf(m, "Boottime: %lld s %ld us\n",
803 		   (s64)ts.tv_sec, ts.tv_nsec / NSEC_PER_USEC);
804 	ts = ktime_to_timespec64(error->uptime);
805 	err_printf(m, "Uptime: %lld s %ld us\n",
806 		   (s64)ts.tv_sec, ts.tv_nsec / NSEC_PER_USEC);
807 	err_printf(m, "Capture: %lu jiffies; %d ms ago\n",
808 		   error->capture, jiffies_to_msecs(jiffies - error->capture));
809 
810 	for (ee = error->gt ? error->gt->engine : NULL; ee; ee = ee->next)
811 		err_printf(m, "Active process (on ring %s): %s [%d]\n",
812 			   ee->engine->name,
813 			   ee->context.comm,
814 			   ee->context.pid);
815 
816 	err_printf(m, "Reset count: %u\n", error->reset_count);
817 	err_printf(m, "Suspend count: %u\n", error->suspend_count);
818 	err_printf(m, "Platform: %s\n", intel_platform_name(error->device_info.platform));
819 	err_printf(m, "Subplatform: 0x%x\n",
820 		   intel_subplatform(&error->runtime_info,
821 				     error->device_info.platform));
822 	err_print_pciid(m, m->i915);
823 
824 	err_printf(m, "IOMMU enabled?: %d\n", error->iommu);
825 
826 	if (HAS_DMC(m->i915)) {
827 		struct intel_dmc *dmc = &m->i915->dmc;
828 
829 		err_printf(m, "DMC loaded: %s\n",
830 			   yesno(intel_dmc_has_payload(m->i915) != 0));
831 		err_printf(m, "DMC fw version: %d.%d\n",
832 			   DMC_VERSION_MAJOR(dmc->version),
833 			   DMC_VERSION_MINOR(dmc->version));
834 	}
835 
836 	err_printf(m, "RPM wakelock: %s\n", yesno(error->wakelock));
837 	err_printf(m, "PM suspended: %s\n", yesno(error->suspended));
838 
839 	if (error->gt)
840 		err_print_gt(m, error->gt);
841 
842 	if (error->overlay)
843 		intel_overlay_print_error_state(m, error->overlay);
844 
845 	err_print_capabilities(m, error);
846 	err_print_params(m, &error->params);
847 }
848 
849 static int err_print_to_sgl(struct i915_gpu_coredump *error)
850 {
851 	struct drm_i915_error_state_buf m;
852 
853 	if (IS_ERR(error))
854 		return PTR_ERR(error);
855 
856 	if (READ_ONCE(error->sgl))
857 		return 0;
858 
859 	memset(&m, 0, sizeof(m));
860 	m.i915 = error->i915;
861 
862 	__err_print_to_sgl(&m, error);
863 
864 	if (m.buf) {
865 		__sg_set_buf(m.cur++, m.buf, m.bytes, m.iter);
866 		m.bytes = 0;
867 		m.buf = NULL;
868 	}
869 	if (m.cur) {
870 		GEM_BUG_ON(m.end < m.cur);
871 		sg_mark_end(m.cur - 1);
872 	}
873 	GEM_BUG_ON(m.sgl && !m.cur);
874 
875 	if (m.err) {
876 		err_free_sgl(m.sgl);
877 		return m.err;
878 	}
879 
880 	if (cmpxchg(&error->sgl, NULL, m.sgl))
881 		err_free_sgl(m.sgl);
882 
883 	return 0;
884 }
885 
886 ssize_t i915_gpu_coredump_copy_to_buffer(struct i915_gpu_coredump *error,
887 					 char *buf, loff_t off, size_t rem)
888 {
889 	struct scatterlist *sg;
890 	size_t count;
891 	loff_t pos;
892 	int err;
893 
894 	if (!error || !rem)
895 		return 0;
896 
897 	err = err_print_to_sgl(error);
898 	if (err)
899 		return err;
900 
901 	sg = READ_ONCE(error->fit);
902 	if (!sg || off < sg->dma_address)
903 		sg = error->sgl;
904 	if (!sg)
905 		return 0;
906 
907 	pos = sg->dma_address;
908 	count = 0;
909 	do {
910 		size_t len, start;
911 
912 		if (sg_is_chain(sg)) {
913 			sg = sg_chain_ptr(sg);
914 			GEM_BUG_ON(sg_is_chain(sg));
915 		}
916 
917 		len = sg->length;
918 		if (pos + len <= off) {
919 			pos += len;
920 			continue;
921 		}
922 
923 		start = sg->offset;
924 		if (pos < off) {
925 			GEM_BUG_ON(off - pos > len);
926 			len -= off - pos;
927 			start += off - pos;
928 			pos = off;
929 		}
930 
931 		len = min(len, rem);
932 		GEM_BUG_ON(!len || len > sg->length);
933 
934 		memcpy(buf, page_address(sg_page(sg)) + start, len);
935 
936 		count += len;
937 		pos += len;
938 
939 		buf += len;
940 		rem -= len;
941 		if (!rem) {
942 			WRITE_ONCE(error->fit, sg);
943 			break;
944 		}
945 	} while (!sg_is_last(sg++));
946 
947 	return count;
948 }
949 
950 static void i915_vma_coredump_free(struct i915_vma_coredump *vma)
951 {
952 	while (vma) {
953 		struct i915_vma_coredump *next = vma->next;
954 		struct page *page, *n;
955 
956 		list_for_each_entry_safe(page, n, &vma->page_list, lru) {
957 			list_del_init(&page->lru);
958 			__free_page(page);
959 		}
960 
961 		kfree(vma);
962 		vma = next;
963 	}
964 }
965 
966 static void cleanup_params(struct i915_gpu_coredump *error)
967 {
968 	i915_params_free(&error->params);
969 }
970 
971 static void cleanup_uc(struct intel_uc_coredump *uc)
972 {
973 	kfree(uc->guc_fw.path);
974 	kfree(uc->huc_fw.path);
975 	i915_vma_coredump_free(uc->guc_log);
976 
977 	kfree(uc);
978 }
979 
980 static void cleanup_gt(struct intel_gt_coredump *gt)
981 {
982 	while (gt->engine) {
983 		struct intel_engine_coredump *ee = gt->engine;
984 
985 		gt->engine = ee->next;
986 
987 		i915_vma_coredump_free(ee->vma);
988 		kfree(ee);
989 	}
990 
991 	if (gt->uc)
992 		cleanup_uc(gt->uc);
993 
994 	kfree(gt);
995 }
996 
997 void __i915_gpu_coredump_free(struct kref *error_ref)
998 {
999 	struct i915_gpu_coredump *error =
1000 		container_of(error_ref, typeof(*error), ref);
1001 
1002 	while (error->gt) {
1003 		struct intel_gt_coredump *gt = error->gt;
1004 
1005 		error->gt = gt->next;
1006 		cleanup_gt(gt);
1007 	}
1008 
1009 	kfree(error->overlay);
1010 
1011 	cleanup_params(error);
1012 
1013 	err_free_sgl(error->sgl);
1014 	kfree(error);
1015 }
1016 
1017 static struct i915_vma_coredump *
1018 i915_vma_coredump_create(const struct intel_gt *gt,
1019 			 const struct i915_vma_resource *vma_res,
1020 			 struct i915_vma_compress *compress,
1021 			 const char *name)
1022 
1023 {
1024 	struct i915_ggtt *ggtt = gt->ggtt;
1025 	const u64 slot = ggtt->error_capture.start;
1026 	struct i915_vma_coredump *dst;
1027 	struct sgt_iter iter;
1028 	int ret;
1029 
1030 	might_sleep();
1031 
1032 	if (!vma_res || !vma_res->bi.pages || !compress)
1033 		return NULL;
1034 
1035 	dst = kmalloc(sizeof(*dst), ALLOW_FAIL);
1036 	if (!dst)
1037 		return NULL;
1038 
1039 	if (!compress_start(compress)) {
1040 		kfree(dst);
1041 		return NULL;
1042 	}
1043 
1044 	INIT_LIST_HEAD(&dst->page_list);
1045 	strcpy(dst->name, name);
1046 	dst->next = NULL;
1047 
1048 	dst->gtt_offset = vma_res->start;
1049 	dst->gtt_size = vma_res->node_size;
1050 	dst->gtt_page_sizes = vma_res->page_sizes_gtt;
1051 	dst->unused = 0;
1052 
1053 	ret = -EINVAL;
1054 	if (drm_mm_node_allocated(&ggtt->error_capture)) {
1055 		void __iomem *s;
1056 		dma_addr_t dma;
1057 
1058 		for_each_sgt_daddr(dma, iter, vma_res->bi.pages) {
1059 			mutex_lock(&ggtt->error_mutex);
1060 			ggtt->vm.insert_page(&ggtt->vm, dma, slot,
1061 					     I915_CACHE_NONE, 0);
1062 			mb();
1063 
1064 			s = io_mapping_map_wc(&ggtt->iomap, slot, PAGE_SIZE);
1065 			ret = compress_page(compress,
1066 					    (void  __force *)s, dst,
1067 					    true);
1068 			io_mapping_unmap(s);
1069 
1070 			mb();
1071 			ggtt->vm.clear_range(&ggtt->vm, slot, PAGE_SIZE);
1072 			mutex_unlock(&ggtt->error_mutex);
1073 			if (ret)
1074 				break;
1075 		}
1076 	} else if (vma_res->bi.lmem) {
1077 		struct intel_memory_region *mem = vma_res->mr;
1078 		dma_addr_t dma;
1079 
1080 		for_each_sgt_daddr(dma, iter, vma_res->bi.pages) {
1081 			void __iomem *s;
1082 
1083 			s = io_mapping_map_wc(&mem->iomap,
1084 					      dma - mem->region.start,
1085 					      PAGE_SIZE);
1086 			ret = compress_page(compress,
1087 					    (void __force *)s, dst,
1088 					    true);
1089 			io_mapping_unmap(s);
1090 			if (ret)
1091 				break;
1092 		}
1093 	} else {
1094 		struct page *page;
1095 
1096 		for_each_sgt_page(page, iter, vma_res->bi.pages) {
1097 			void *s;
1098 
1099 			drm_clflush_pages(&page, 1);
1100 
1101 			s = kmap(page);
1102 			ret = compress_page(compress, s, dst, false);
1103 			kunmap(page);
1104 
1105 			drm_clflush_pages(&page, 1);
1106 
1107 			if (ret)
1108 				break;
1109 		}
1110 	}
1111 
1112 	if (ret || compress_flush(compress, dst)) {
1113 		struct page *page, *n;
1114 
1115 		list_for_each_entry_safe_reverse(page, n, &dst->page_list, lru) {
1116 			list_del_init(&page->lru);
1117 			pool_free(&compress->pool, page_address(page));
1118 		}
1119 
1120 		kfree(dst);
1121 		dst = NULL;
1122 	}
1123 	compress_finish(compress);
1124 
1125 	return dst;
1126 }
1127 
1128 static void gt_record_fences(struct intel_gt_coredump *gt)
1129 {
1130 	struct i915_ggtt *ggtt = gt->_gt->ggtt;
1131 	struct intel_uncore *uncore = gt->_gt->uncore;
1132 	int i;
1133 
1134 	if (GRAPHICS_VER(uncore->i915) >= 6) {
1135 		for (i = 0; i < ggtt->num_fences; i++)
1136 			gt->fence[i] =
1137 				intel_uncore_read64(uncore,
1138 						    FENCE_REG_GEN6_LO(i));
1139 	} else if (GRAPHICS_VER(uncore->i915) >= 4) {
1140 		for (i = 0; i < ggtt->num_fences; i++)
1141 			gt->fence[i] =
1142 				intel_uncore_read64(uncore,
1143 						    FENCE_REG_965_LO(i));
1144 	} else {
1145 		for (i = 0; i < ggtt->num_fences; i++)
1146 			gt->fence[i] =
1147 				intel_uncore_read(uncore, FENCE_REG(i));
1148 	}
1149 	gt->nfence = i;
1150 }
1151 
1152 static void engine_record_registers(struct intel_engine_coredump *ee)
1153 {
1154 	const struct intel_engine_cs *engine = ee->engine;
1155 	struct drm_i915_private *i915 = engine->i915;
1156 
1157 	if (GRAPHICS_VER(i915) >= 6) {
1158 		ee->rc_psmi = ENGINE_READ(engine, RING_PSMI_CTL);
1159 
1160 		if (GRAPHICS_VER(i915) >= 12)
1161 			ee->fault_reg = intel_uncore_read(engine->uncore,
1162 							  GEN12_RING_FAULT_REG);
1163 		else if (GRAPHICS_VER(i915) >= 8)
1164 			ee->fault_reg = intel_uncore_read(engine->uncore,
1165 							  GEN8_RING_FAULT_REG);
1166 		else
1167 			ee->fault_reg = GEN6_RING_FAULT_REG_READ(engine);
1168 	}
1169 
1170 	if (GRAPHICS_VER(i915) >= 4) {
1171 		ee->esr = ENGINE_READ(engine, RING_ESR);
1172 		ee->faddr = ENGINE_READ(engine, RING_DMA_FADD);
1173 		ee->ipeir = ENGINE_READ(engine, RING_IPEIR);
1174 		ee->ipehr = ENGINE_READ(engine, RING_IPEHR);
1175 		ee->instps = ENGINE_READ(engine, RING_INSTPS);
1176 		ee->bbaddr = ENGINE_READ(engine, RING_BBADDR);
1177 		ee->ccid = ENGINE_READ(engine, CCID);
1178 		if (GRAPHICS_VER(i915) >= 8) {
1179 			ee->faddr |= (u64)ENGINE_READ(engine, RING_DMA_FADD_UDW) << 32;
1180 			ee->bbaddr |= (u64)ENGINE_READ(engine, RING_BBADDR_UDW) << 32;
1181 		}
1182 		ee->bbstate = ENGINE_READ(engine, RING_BBSTATE);
1183 	} else {
1184 		ee->faddr = ENGINE_READ(engine, DMA_FADD_I8XX);
1185 		ee->ipeir = ENGINE_READ(engine, IPEIR);
1186 		ee->ipehr = ENGINE_READ(engine, IPEHR);
1187 	}
1188 
1189 	intel_engine_get_instdone(engine, &ee->instdone);
1190 
1191 	ee->instpm = ENGINE_READ(engine, RING_INSTPM);
1192 	ee->acthd = intel_engine_get_active_head(engine);
1193 	ee->start = ENGINE_READ(engine, RING_START);
1194 	ee->head = ENGINE_READ(engine, RING_HEAD);
1195 	ee->tail = ENGINE_READ(engine, RING_TAIL);
1196 	ee->ctl = ENGINE_READ(engine, RING_CTL);
1197 	if (GRAPHICS_VER(i915) > 2)
1198 		ee->mode = ENGINE_READ(engine, RING_MI_MODE);
1199 
1200 	if (!HWS_NEEDS_PHYSICAL(i915)) {
1201 		i915_reg_t mmio;
1202 
1203 		if (GRAPHICS_VER(i915) == 7) {
1204 			switch (engine->id) {
1205 			default:
1206 				MISSING_CASE(engine->id);
1207 				fallthrough;
1208 			case RCS0:
1209 				mmio = RENDER_HWS_PGA_GEN7;
1210 				break;
1211 			case BCS0:
1212 				mmio = BLT_HWS_PGA_GEN7;
1213 				break;
1214 			case VCS0:
1215 				mmio = BSD_HWS_PGA_GEN7;
1216 				break;
1217 			case VECS0:
1218 				mmio = VEBOX_HWS_PGA_GEN7;
1219 				break;
1220 			}
1221 		} else if (GRAPHICS_VER(engine->i915) == 6) {
1222 			mmio = RING_HWS_PGA_GEN6(engine->mmio_base);
1223 		} else {
1224 			/* XXX: gen8 returns to sanity */
1225 			mmio = RING_HWS_PGA(engine->mmio_base);
1226 		}
1227 
1228 		ee->hws = intel_uncore_read(engine->uncore, mmio);
1229 	}
1230 
1231 	ee->reset_count = i915_reset_engine_count(&i915->gpu_error, engine);
1232 
1233 	if (HAS_PPGTT(i915)) {
1234 		int i;
1235 
1236 		ee->vm_info.gfx_mode = ENGINE_READ(engine, RING_MODE_GEN7);
1237 
1238 		if (GRAPHICS_VER(i915) == 6) {
1239 			ee->vm_info.pp_dir_base =
1240 				ENGINE_READ(engine, RING_PP_DIR_BASE_READ);
1241 		} else if (GRAPHICS_VER(i915) == 7) {
1242 			ee->vm_info.pp_dir_base =
1243 				ENGINE_READ(engine, RING_PP_DIR_BASE);
1244 		} else if (GRAPHICS_VER(i915) >= 8) {
1245 			u32 base = engine->mmio_base;
1246 
1247 			for (i = 0; i < 4; i++) {
1248 				ee->vm_info.pdp[i] =
1249 					intel_uncore_read(engine->uncore,
1250 							  GEN8_RING_PDP_UDW(base, i));
1251 				ee->vm_info.pdp[i] <<= 32;
1252 				ee->vm_info.pdp[i] |=
1253 					intel_uncore_read(engine->uncore,
1254 							  GEN8_RING_PDP_LDW(base, i));
1255 			}
1256 		}
1257 	}
1258 }
1259 
1260 static void record_request(const struct i915_request *request,
1261 			   struct i915_request_coredump *erq)
1262 {
1263 	erq->flags = request->fence.flags;
1264 	erq->context = request->fence.context;
1265 	erq->seqno = request->fence.seqno;
1266 	erq->sched_attr = request->sched.attr;
1267 	erq->head = request->head;
1268 	erq->tail = request->tail;
1269 
1270 	erq->pid = 0;
1271 	rcu_read_lock();
1272 	if (!intel_context_is_closed(request->context)) {
1273 		const struct i915_gem_context *ctx;
1274 
1275 		ctx = rcu_dereference(request->context->gem_context);
1276 		if (ctx)
1277 			erq->pid = pid_nr(ctx->pid);
1278 	}
1279 	rcu_read_unlock();
1280 }
1281 
1282 static void engine_record_execlists(struct intel_engine_coredump *ee)
1283 {
1284 	const struct intel_engine_execlists * const el = &ee->engine->execlists;
1285 	struct i915_request * const *port = el->active;
1286 	unsigned int n = 0;
1287 
1288 	while (*port)
1289 		record_request(*port++, &ee->execlist[n++]);
1290 
1291 	ee->num_ports = n;
1292 }
1293 
1294 static bool record_context(struct i915_gem_context_coredump *e,
1295 			   const struct i915_request *rq)
1296 {
1297 	struct i915_gem_context *ctx;
1298 	struct task_struct *task;
1299 	bool simulated;
1300 
1301 	rcu_read_lock();
1302 	ctx = rcu_dereference(rq->context->gem_context);
1303 	if (ctx && !kref_get_unless_zero(&ctx->ref))
1304 		ctx = NULL;
1305 	rcu_read_unlock();
1306 	if (!ctx)
1307 		return true;
1308 
1309 	rcu_read_lock();
1310 	task = pid_task(ctx->pid, PIDTYPE_PID);
1311 	if (task) {
1312 		strcpy(e->comm, task->comm);
1313 		e->pid = task->pid;
1314 	}
1315 	rcu_read_unlock();
1316 
1317 	e->sched_attr = ctx->sched;
1318 	e->guilty = atomic_read(&ctx->guilty_count);
1319 	e->active = atomic_read(&ctx->active_count);
1320 
1321 	e->total_runtime = rq->context->runtime.total;
1322 	e->avg_runtime = ewma_runtime_read(&rq->context->runtime.avg);
1323 
1324 	simulated = i915_gem_context_no_error_capture(ctx);
1325 
1326 	i915_gem_context_put(ctx);
1327 	return simulated;
1328 }
1329 
1330 struct intel_engine_capture_vma {
1331 	struct intel_engine_capture_vma *next;
1332 	struct i915_vma_resource *vma_res;
1333 	char name[16];
1334 	bool lockdep_cookie;
1335 };
1336 
1337 static struct intel_engine_capture_vma *
1338 capture_vma_snapshot(struct intel_engine_capture_vma *next,
1339 		     struct i915_vma_resource *vma_res,
1340 		     gfp_t gfp, const char *name)
1341 {
1342 	struct intel_engine_capture_vma *c;
1343 
1344 	if (!vma_res)
1345 		return next;
1346 
1347 	c = kmalloc(sizeof(*c), gfp);
1348 	if (!c)
1349 		return next;
1350 
1351 	if (!i915_vma_resource_hold(vma_res, &c->lockdep_cookie)) {
1352 		kfree(c);
1353 		return next;
1354 	}
1355 
1356 	strcpy(c->name, name);
1357 	c->vma_res = i915_vma_resource_get(vma_res);
1358 
1359 	c->next = next;
1360 	return c;
1361 }
1362 
1363 static struct intel_engine_capture_vma *
1364 capture_vma(struct intel_engine_capture_vma *next,
1365 	    struct i915_vma *vma,
1366 	    const char *name,
1367 	    gfp_t gfp)
1368 {
1369 	if (!vma)
1370 		return next;
1371 
1372 	/*
1373 	 * If the vma isn't pinned, then the vma should be snapshotted
1374 	 * to a struct i915_vma_snapshot at command submission time.
1375 	 * Not here.
1376 	 */
1377 	if (GEM_WARN_ON(!i915_vma_is_pinned(vma)))
1378 		return next;
1379 
1380 	next = capture_vma_snapshot(next, vma->resource, gfp, name);
1381 
1382 	return next;
1383 }
1384 
1385 static struct intel_engine_capture_vma *
1386 capture_user(struct intel_engine_capture_vma *capture,
1387 	     const struct i915_request *rq,
1388 	     gfp_t gfp)
1389 {
1390 	struct i915_capture_list *c;
1391 
1392 	for (c = rq->capture_list; c; c = c->next)
1393 		capture = capture_vma_snapshot(capture, c->vma_res, gfp,
1394 					       "user");
1395 
1396 	return capture;
1397 }
1398 
1399 static void add_vma(struct intel_engine_coredump *ee,
1400 		    struct i915_vma_coredump *vma)
1401 {
1402 	if (vma) {
1403 		vma->next = ee->vma;
1404 		ee->vma = vma;
1405 	}
1406 }
1407 
1408 static struct i915_vma_coredump *
1409 create_vma_coredump(const struct intel_gt *gt, struct i915_vma *vma,
1410 		    const char *name, struct i915_vma_compress *compress)
1411 {
1412 	struct i915_vma_coredump *ret = NULL;
1413 	struct i915_vma_resource *vma_res;
1414 	bool lockdep_cookie;
1415 
1416 	if (!vma)
1417 		return NULL;
1418 
1419 	vma_res = vma->resource;
1420 
1421 	if (i915_vma_resource_hold(vma_res, &lockdep_cookie)) {
1422 		ret = i915_vma_coredump_create(gt, vma_res, compress, name);
1423 		i915_vma_resource_unhold(vma_res, lockdep_cookie);
1424 	}
1425 
1426 	return ret;
1427 }
1428 
1429 static void add_vma_coredump(struct intel_engine_coredump *ee,
1430 			     const struct intel_gt *gt,
1431 			     struct i915_vma *vma,
1432 			     const char *name,
1433 			     struct i915_vma_compress *compress)
1434 {
1435 	add_vma(ee, create_vma_coredump(gt, vma, name, compress));
1436 }
1437 
1438 struct intel_engine_coredump *
1439 intel_engine_coredump_alloc(struct intel_engine_cs *engine, gfp_t gfp)
1440 {
1441 	struct intel_engine_coredump *ee;
1442 
1443 	ee = kzalloc(sizeof(*ee), gfp);
1444 	if (!ee)
1445 		return NULL;
1446 
1447 	ee->engine = engine;
1448 
1449 	engine_record_registers(ee);
1450 	engine_record_execlists(ee);
1451 
1452 	return ee;
1453 }
1454 
1455 struct intel_engine_capture_vma *
1456 intel_engine_coredump_add_request(struct intel_engine_coredump *ee,
1457 				  struct i915_request *rq,
1458 				  gfp_t gfp)
1459 {
1460 	struct intel_engine_capture_vma *vma = NULL;
1461 
1462 	ee->simulated |= record_context(&ee->context, rq);
1463 	if (ee->simulated)
1464 		return NULL;
1465 
1466 	/*
1467 	 * We need to copy these to an anonymous buffer
1468 	 * as the simplest method to avoid being overwritten
1469 	 * by userspace.
1470 	 */
1471 	vma = capture_vma_snapshot(vma, rq->batch_res, gfp, "batch");
1472 	vma = capture_user(vma, rq, gfp);
1473 	vma = capture_vma(vma, rq->ring->vma, "ring", gfp);
1474 	vma = capture_vma(vma, rq->context->state, "HW context", gfp);
1475 
1476 	ee->rq_head = rq->head;
1477 	ee->rq_post = rq->postfix;
1478 	ee->rq_tail = rq->tail;
1479 
1480 	return vma;
1481 }
1482 
1483 void
1484 intel_engine_coredump_add_vma(struct intel_engine_coredump *ee,
1485 			      struct intel_engine_capture_vma *capture,
1486 			      struct i915_vma_compress *compress)
1487 {
1488 	const struct intel_engine_cs *engine = ee->engine;
1489 
1490 	while (capture) {
1491 		struct intel_engine_capture_vma *this = capture;
1492 		struct i915_vma_resource *vma_res = this->vma_res;
1493 
1494 		add_vma(ee,
1495 			i915_vma_coredump_create(engine->gt, vma_res,
1496 						 compress, this->name));
1497 
1498 		i915_vma_resource_unhold(vma_res, this->lockdep_cookie);
1499 		i915_vma_resource_put(vma_res);
1500 
1501 		capture = this->next;
1502 		kfree(this);
1503 	}
1504 
1505 	add_vma_coredump(ee, engine->gt, engine->status_page.vma,
1506 			 "HW Status", compress);
1507 
1508 	add_vma_coredump(ee, engine->gt, engine->wa_ctx.vma,
1509 			 "WA context", compress);
1510 }
1511 
1512 static struct intel_engine_coredump *
1513 capture_engine(struct intel_engine_cs *engine,
1514 	       struct i915_vma_compress *compress)
1515 {
1516 	struct intel_engine_capture_vma *capture = NULL;
1517 	struct intel_engine_coredump *ee;
1518 	struct intel_context *ce;
1519 	struct i915_request *rq = NULL;
1520 	unsigned long flags;
1521 
1522 	ee = intel_engine_coredump_alloc(engine, ALLOW_FAIL);
1523 	if (!ee)
1524 		return NULL;
1525 
1526 	ce = intel_engine_get_hung_context(engine);
1527 	if (ce) {
1528 		intel_engine_clear_hung_context(engine);
1529 		rq = intel_context_find_active_request(ce);
1530 		if (!rq || !i915_request_started(rq))
1531 			goto no_request_capture;
1532 	} else {
1533 		/*
1534 		 * Getting here with GuC enabled means it is a forced error capture
1535 		 * with no actual hang. So, no need to attempt the execlist search.
1536 		 */
1537 		if (!intel_uc_uses_guc_submission(&engine->gt->uc)) {
1538 			spin_lock_irqsave(&engine->sched_engine->lock, flags);
1539 			rq = intel_engine_execlist_find_hung_request(engine);
1540 			spin_unlock_irqrestore(&engine->sched_engine->lock,
1541 					       flags);
1542 		}
1543 	}
1544 	if (rq)
1545 		rq = i915_request_get_rcu(rq);
1546 
1547 	if (!rq)
1548 		goto no_request_capture;
1549 
1550 	capture = intel_engine_coredump_add_request(ee, rq, ATOMIC_MAYFAIL);
1551 	if (!capture) {
1552 		i915_request_put(rq);
1553 		goto no_request_capture;
1554 	}
1555 
1556 	intel_engine_coredump_add_vma(ee, capture, compress);
1557 	i915_request_put(rq);
1558 
1559 	return ee;
1560 
1561 no_request_capture:
1562 	kfree(ee);
1563 	return NULL;
1564 }
1565 
1566 static void
1567 gt_record_engines(struct intel_gt_coredump *gt,
1568 		  intel_engine_mask_t engine_mask,
1569 		  struct i915_vma_compress *compress)
1570 {
1571 	struct intel_engine_cs *engine;
1572 	enum intel_engine_id id;
1573 
1574 	for_each_engine(engine, gt->_gt, id) {
1575 		struct intel_engine_coredump *ee;
1576 
1577 		/* Refill our page pool before entering atomic section */
1578 		pool_refill(&compress->pool, ALLOW_FAIL);
1579 
1580 		ee = capture_engine(engine, compress);
1581 		if (!ee)
1582 			continue;
1583 
1584 		ee->hung = engine->mask & engine_mask;
1585 
1586 		gt->simulated |= ee->simulated;
1587 		if (ee->simulated) {
1588 			kfree(ee);
1589 			continue;
1590 		}
1591 
1592 		ee->next = gt->engine;
1593 		gt->engine = ee;
1594 	}
1595 }
1596 
1597 static struct intel_uc_coredump *
1598 gt_record_uc(struct intel_gt_coredump *gt,
1599 	     struct i915_vma_compress *compress)
1600 {
1601 	const struct intel_uc *uc = &gt->_gt->uc;
1602 	struct intel_uc_coredump *error_uc;
1603 
1604 	error_uc = kzalloc(sizeof(*error_uc), ALLOW_FAIL);
1605 	if (!error_uc)
1606 		return NULL;
1607 
1608 	memcpy(&error_uc->guc_fw, &uc->guc.fw, sizeof(uc->guc.fw));
1609 	memcpy(&error_uc->huc_fw, &uc->huc.fw, sizeof(uc->huc.fw));
1610 
1611 	/* Non-default firmware paths will be specified by the modparam.
1612 	 * As modparams are generally accesible from the userspace make
1613 	 * explicit copies of the firmware paths.
1614 	 */
1615 	error_uc->guc_fw.path = kstrdup(uc->guc.fw.path, ALLOW_FAIL);
1616 	error_uc->huc_fw.path = kstrdup(uc->huc.fw.path, ALLOW_FAIL);
1617 	error_uc->guc_log = create_vma_coredump(gt->_gt, uc->guc.log.vma,
1618 						"GuC log buffer", compress);
1619 
1620 	return error_uc;
1621 }
1622 
1623 /* Capture all registers which don't fit into another category. */
1624 static void gt_record_regs(struct intel_gt_coredump *gt)
1625 {
1626 	struct intel_uncore *uncore = gt->_gt->uncore;
1627 	struct drm_i915_private *i915 = uncore->i915;
1628 	int i;
1629 
1630 	/*
1631 	 * General organization
1632 	 * 1. Registers specific to a single generation
1633 	 * 2. Registers which belong to multiple generations
1634 	 * 3. Feature specific registers.
1635 	 * 4. Everything else
1636 	 * Please try to follow the order.
1637 	 */
1638 
1639 	/* 1: Registers specific to a single generation */
1640 	if (IS_VALLEYVIEW(i915)) {
1641 		gt->gtier[0] = intel_uncore_read(uncore, GTIER);
1642 		gt->ier = intel_uncore_read(uncore, VLV_IER);
1643 		gt->forcewake = intel_uncore_read_fw(uncore, FORCEWAKE_VLV);
1644 	}
1645 
1646 	if (GRAPHICS_VER(i915) == 7)
1647 		gt->err_int = intel_uncore_read(uncore, GEN7_ERR_INT);
1648 
1649 	if (GRAPHICS_VER(i915) >= 12) {
1650 		gt->fault_data0 = intel_uncore_read(uncore,
1651 						    GEN12_FAULT_TLB_DATA0);
1652 		gt->fault_data1 = intel_uncore_read(uncore,
1653 						    GEN12_FAULT_TLB_DATA1);
1654 	} else if (GRAPHICS_VER(i915) >= 8) {
1655 		gt->fault_data0 = intel_uncore_read(uncore,
1656 						    GEN8_FAULT_TLB_DATA0);
1657 		gt->fault_data1 = intel_uncore_read(uncore,
1658 						    GEN8_FAULT_TLB_DATA1);
1659 	}
1660 
1661 	if (GRAPHICS_VER(i915) == 6) {
1662 		gt->forcewake = intel_uncore_read_fw(uncore, FORCEWAKE);
1663 		gt->gab_ctl = intel_uncore_read(uncore, GAB_CTL);
1664 		gt->gfx_mode = intel_uncore_read(uncore, GFX_MODE);
1665 	}
1666 
1667 	/* 2: Registers which belong to multiple generations */
1668 	if (GRAPHICS_VER(i915) >= 7)
1669 		gt->forcewake = intel_uncore_read_fw(uncore, FORCEWAKE_MT);
1670 
1671 	if (GRAPHICS_VER(i915) >= 6) {
1672 		gt->derrmr = intel_uncore_read(uncore, DERRMR);
1673 		if (GRAPHICS_VER(i915) < 12) {
1674 			gt->error = intel_uncore_read(uncore, ERROR_GEN6);
1675 			gt->done_reg = intel_uncore_read(uncore, DONE_REG);
1676 		}
1677 	}
1678 
1679 	/* 3: Feature specific registers */
1680 	if (IS_GRAPHICS_VER(i915, 6, 7)) {
1681 		gt->gam_ecochk = intel_uncore_read(uncore, GAM_ECOCHK);
1682 		gt->gac_eco = intel_uncore_read(uncore, GAC_ECO_BITS);
1683 	}
1684 
1685 	if (IS_GRAPHICS_VER(i915, 8, 11))
1686 		gt->gtt_cache = intel_uncore_read(uncore, HSW_GTT_CACHE_EN);
1687 
1688 	if (GRAPHICS_VER(i915) == 12)
1689 		gt->aux_err = intel_uncore_read(uncore, GEN12_AUX_ERR_DBG);
1690 
1691 	if (GRAPHICS_VER(i915) >= 12) {
1692 		for (i = 0; i < GEN12_SFC_DONE_MAX; i++) {
1693 			/*
1694 			 * SFC_DONE resides in the VD forcewake domain, so it
1695 			 * only exists if the corresponding VCS engine is
1696 			 * present.
1697 			 */
1698 			if ((gt->_gt->info.sfc_mask & BIT(i)) == 0 ||
1699 			    !HAS_ENGINE(gt->_gt, _VCS(i * 2)))
1700 				continue;
1701 
1702 			gt->sfc_done[i] =
1703 				intel_uncore_read(uncore, GEN12_SFC_DONE(i));
1704 		}
1705 
1706 		gt->gam_done = intel_uncore_read(uncore, GEN12_GAM_DONE);
1707 	}
1708 
1709 	/* 4: Everything else */
1710 	if (GRAPHICS_VER(i915) >= 11) {
1711 		gt->ier = intel_uncore_read(uncore, GEN8_DE_MISC_IER);
1712 		gt->gtier[0] =
1713 			intel_uncore_read(uncore,
1714 					  GEN11_RENDER_COPY_INTR_ENABLE);
1715 		gt->gtier[1] =
1716 			intel_uncore_read(uncore, GEN11_VCS_VECS_INTR_ENABLE);
1717 		gt->gtier[2] =
1718 			intel_uncore_read(uncore, GEN11_GUC_SG_INTR_ENABLE);
1719 		gt->gtier[3] =
1720 			intel_uncore_read(uncore,
1721 					  GEN11_GPM_WGBOXPERF_INTR_ENABLE);
1722 		gt->gtier[4] =
1723 			intel_uncore_read(uncore,
1724 					  GEN11_CRYPTO_RSVD_INTR_ENABLE);
1725 		gt->gtier[5] =
1726 			intel_uncore_read(uncore,
1727 					  GEN11_GUNIT_CSME_INTR_ENABLE);
1728 		gt->ngtier = 6;
1729 	} else if (GRAPHICS_VER(i915) >= 8) {
1730 		gt->ier = intel_uncore_read(uncore, GEN8_DE_MISC_IER);
1731 		for (i = 0; i < 4; i++)
1732 			gt->gtier[i] =
1733 				intel_uncore_read(uncore, GEN8_GT_IER(i));
1734 		gt->ngtier = 4;
1735 	} else if (HAS_PCH_SPLIT(i915)) {
1736 		gt->ier = intel_uncore_read(uncore, DEIER);
1737 		gt->gtier[0] = intel_uncore_read(uncore, GTIER);
1738 		gt->ngtier = 1;
1739 	} else if (GRAPHICS_VER(i915) == 2) {
1740 		gt->ier = intel_uncore_read16(uncore, GEN2_IER);
1741 	} else if (!IS_VALLEYVIEW(i915)) {
1742 		gt->ier = intel_uncore_read(uncore, GEN2_IER);
1743 	}
1744 	gt->eir = intel_uncore_read(uncore, EIR);
1745 	gt->pgtbl_er = intel_uncore_read(uncore, PGTBL_ER);
1746 }
1747 
1748 static void gt_record_info(struct intel_gt_coredump *gt)
1749 {
1750 	memcpy(&gt->info, &gt->_gt->info, sizeof(struct intel_gt_info));
1751 }
1752 
1753 /*
1754  * Generate a semi-unique error code. The code is not meant to have meaning, The
1755  * code's only purpose is to try to prevent false duplicated bug reports by
1756  * grossly estimating a GPU error state.
1757  *
1758  * TODO Ideally, hashing the batchbuffer would be a very nice way to determine
1759  * the hang if we could strip the GTT offset information from it.
1760  *
1761  * It's only a small step better than a random number in its current form.
1762  */
1763 static u32 generate_ecode(const struct intel_engine_coredump *ee)
1764 {
1765 	/*
1766 	 * IPEHR would be an ideal way to detect errors, as it's the gross
1767 	 * measure of "the command that hung." However, has some very common
1768 	 * synchronization commands which almost always appear in the case
1769 	 * strictly a client bug. Use instdone to differentiate those some.
1770 	 */
1771 	return ee ? ee->ipehr ^ ee->instdone.instdone : 0;
1772 }
1773 
1774 static const char *error_msg(struct i915_gpu_coredump *error)
1775 {
1776 	struct intel_engine_coredump *first = NULL;
1777 	unsigned int hung_classes = 0;
1778 	struct intel_gt_coredump *gt;
1779 	int len;
1780 
1781 	for (gt = error->gt; gt; gt = gt->next) {
1782 		struct intel_engine_coredump *cs;
1783 
1784 		for (cs = gt->engine; cs; cs = cs->next) {
1785 			if (cs->hung) {
1786 				hung_classes |= BIT(cs->engine->uabi_class);
1787 				if (!first)
1788 					first = cs;
1789 			}
1790 		}
1791 	}
1792 
1793 	len = scnprintf(error->error_msg, sizeof(error->error_msg),
1794 			"GPU HANG: ecode %d:%x:%08x",
1795 			GRAPHICS_VER(error->i915), hung_classes,
1796 			generate_ecode(first));
1797 	if (first && first->context.pid) {
1798 		/* Just show the first executing process, more is confusing */
1799 		len += scnprintf(error->error_msg + len,
1800 				 sizeof(error->error_msg) - len,
1801 				 ", in %s [%d]",
1802 				 first->context.comm, first->context.pid);
1803 	}
1804 
1805 	return error->error_msg;
1806 }
1807 
1808 static void capture_gen(struct i915_gpu_coredump *error)
1809 {
1810 	struct drm_i915_private *i915 = error->i915;
1811 
1812 	error->wakelock = atomic_read(&i915->runtime_pm.wakeref_count);
1813 	error->suspended = i915->runtime_pm.suspended;
1814 
1815 	error->iommu = intel_vtd_active(i915);
1816 	error->reset_count = i915_reset_count(&i915->gpu_error);
1817 	error->suspend_count = i915->suspend_count;
1818 
1819 	i915_params_copy(&error->params, &i915->params);
1820 	memcpy(&error->device_info,
1821 	       INTEL_INFO(i915),
1822 	       sizeof(error->device_info));
1823 	memcpy(&error->runtime_info,
1824 	       RUNTIME_INFO(i915),
1825 	       sizeof(error->runtime_info));
1826 	error->driver_caps = i915->caps;
1827 }
1828 
1829 struct i915_gpu_coredump *
1830 i915_gpu_coredump_alloc(struct drm_i915_private *i915, gfp_t gfp)
1831 {
1832 	struct i915_gpu_coredump *error;
1833 
1834 	if (!i915->params.error_capture)
1835 		return NULL;
1836 
1837 	error = kzalloc(sizeof(*error), gfp);
1838 	if (!error)
1839 		return NULL;
1840 
1841 	kref_init(&error->ref);
1842 	error->i915 = i915;
1843 
1844 	error->time = ktime_get_real();
1845 	error->boottime = ktime_get_boottime();
1846 	error->uptime = ktime_sub(ktime_get(), to_gt(i915)->last_init_time);
1847 	error->capture = jiffies;
1848 
1849 	capture_gen(error);
1850 
1851 	return error;
1852 }
1853 
1854 #define DAY_AS_SECONDS(x) (24 * 60 * 60 * (x))
1855 
1856 struct intel_gt_coredump *
1857 intel_gt_coredump_alloc(struct intel_gt *gt, gfp_t gfp)
1858 {
1859 	struct intel_gt_coredump *gc;
1860 
1861 	gc = kzalloc(sizeof(*gc), gfp);
1862 	if (!gc)
1863 		return NULL;
1864 
1865 	gc->_gt = gt;
1866 	gc->awake = intel_gt_pm_is_awake(gt);
1867 
1868 	gt_record_regs(gc);
1869 	gt_record_fences(gc);
1870 
1871 	return gc;
1872 }
1873 
1874 struct i915_vma_compress *
1875 i915_vma_capture_prepare(struct intel_gt_coredump *gt)
1876 {
1877 	struct i915_vma_compress *compress;
1878 
1879 	compress = kmalloc(sizeof(*compress), ALLOW_FAIL);
1880 	if (!compress)
1881 		return NULL;
1882 
1883 	if (!compress_init(compress)) {
1884 		kfree(compress);
1885 		return NULL;
1886 	}
1887 
1888 	return compress;
1889 }
1890 
1891 void i915_vma_capture_finish(struct intel_gt_coredump *gt,
1892 			     struct i915_vma_compress *compress)
1893 {
1894 	if (!compress)
1895 		return;
1896 
1897 	compress_fini(compress);
1898 	kfree(compress);
1899 }
1900 
1901 static struct i915_gpu_coredump *
1902 __i915_gpu_coredump(struct intel_gt *gt, intel_engine_mask_t engine_mask)
1903 {
1904 	struct drm_i915_private *i915 = gt->i915;
1905 	struct i915_gpu_coredump *error;
1906 
1907 	/* Check if GPU capture has been disabled */
1908 	error = READ_ONCE(i915->gpu_error.first_error);
1909 	if (IS_ERR(error))
1910 		return error;
1911 
1912 	error = i915_gpu_coredump_alloc(i915, ALLOW_FAIL);
1913 	if (!error)
1914 		return ERR_PTR(-ENOMEM);
1915 
1916 	error->gt = intel_gt_coredump_alloc(gt, ALLOW_FAIL);
1917 	if (error->gt) {
1918 		struct i915_vma_compress *compress;
1919 
1920 		compress = i915_vma_capture_prepare(error->gt);
1921 		if (!compress) {
1922 			kfree(error->gt);
1923 			kfree(error);
1924 			return ERR_PTR(-ENOMEM);
1925 		}
1926 
1927 		gt_record_info(error->gt);
1928 		gt_record_engines(error->gt, engine_mask, compress);
1929 
1930 		if (INTEL_INFO(i915)->has_gt_uc)
1931 			error->gt->uc = gt_record_uc(error->gt, compress);
1932 
1933 		i915_vma_capture_finish(error->gt, compress);
1934 
1935 		error->simulated |= error->gt->simulated;
1936 	}
1937 
1938 	error->overlay = intel_overlay_capture_error_state(i915);
1939 
1940 	return error;
1941 }
1942 
1943 struct i915_gpu_coredump *
1944 i915_gpu_coredump(struct intel_gt *gt, intel_engine_mask_t engine_mask)
1945 {
1946 	static DEFINE_MUTEX(capture_mutex);
1947 	int ret = mutex_lock_interruptible(&capture_mutex);
1948 	struct i915_gpu_coredump *dump;
1949 
1950 	if (ret)
1951 		return ERR_PTR(ret);
1952 
1953 	dump = __i915_gpu_coredump(gt, engine_mask);
1954 	mutex_unlock(&capture_mutex);
1955 
1956 	return dump;
1957 }
1958 
1959 void i915_error_state_store(struct i915_gpu_coredump *error)
1960 {
1961 	struct drm_i915_private *i915;
1962 	static bool warned;
1963 
1964 	if (IS_ERR_OR_NULL(error))
1965 		return;
1966 
1967 	i915 = error->i915;
1968 	drm_info(&i915->drm, "%s\n", error_msg(error));
1969 
1970 	if (error->simulated ||
1971 	    cmpxchg(&i915->gpu_error.first_error, NULL, error))
1972 		return;
1973 
1974 	i915_gpu_coredump_get(error);
1975 
1976 	if (!xchg(&warned, true) &&
1977 	    ktime_get_real_seconds() - DRIVER_TIMESTAMP < DAY_AS_SECONDS(180)) {
1978 		pr_info("GPU hangs can indicate a bug anywhere in the entire gfx stack, including userspace.\n");
1979 		pr_info("Please file a _new_ bug report at https://gitlab.freedesktop.org/drm/intel/issues/new.\n");
1980 		pr_info("Please see https://gitlab.freedesktop.org/drm/intel/-/wikis/How-to-file-i915-bugs for details.\n");
1981 		pr_info("drm/i915 developers can then reassign to the right component if it's not a kernel issue.\n");
1982 		pr_info("The GPU crash dump is required to analyze GPU hangs, so please always attach it.\n");
1983 		pr_info("GPU crash dump saved to /sys/class/drm/card%d/error\n",
1984 			i915->drm.primary->index);
1985 	}
1986 }
1987 
1988 /**
1989  * i915_capture_error_state - capture an error record for later analysis
1990  * @gt: intel_gt which originated the hang
1991  * @engine_mask: hung engines
1992  *
1993  *
1994  * Should be called when an error is detected (either a hang or an error
1995  * interrupt) to capture error state from the time of the error.  Fills
1996  * out a structure which becomes available in debugfs for user level tools
1997  * to pick up.
1998  */
1999 void i915_capture_error_state(struct intel_gt *gt,
2000 			      intel_engine_mask_t engine_mask)
2001 {
2002 	struct i915_gpu_coredump *error;
2003 
2004 	error = i915_gpu_coredump(gt, engine_mask);
2005 	if (IS_ERR(error)) {
2006 		cmpxchg(&gt->i915->gpu_error.first_error, NULL, error);
2007 		return;
2008 	}
2009 
2010 	i915_error_state_store(error);
2011 	i915_gpu_coredump_put(error);
2012 }
2013 
2014 struct i915_gpu_coredump *
2015 i915_first_error_state(struct drm_i915_private *i915)
2016 {
2017 	struct i915_gpu_coredump *error;
2018 
2019 	spin_lock_irq(&i915->gpu_error.lock);
2020 	error = i915->gpu_error.first_error;
2021 	if (!IS_ERR_OR_NULL(error))
2022 		i915_gpu_coredump_get(error);
2023 	spin_unlock_irq(&i915->gpu_error.lock);
2024 
2025 	return error;
2026 }
2027 
2028 void i915_reset_error_state(struct drm_i915_private *i915)
2029 {
2030 	struct i915_gpu_coredump *error;
2031 
2032 	spin_lock_irq(&i915->gpu_error.lock);
2033 	error = i915->gpu_error.first_error;
2034 	if (error != ERR_PTR(-ENODEV)) /* if disabled, always disabled */
2035 		i915->gpu_error.first_error = NULL;
2036 	spin_unlock_irq(&i915->gpu_error.lock);
2037 
2038 	if (!IS_ERR_OR_NULL(error))
2039 		i915_gpu_coredump_put(error);
2040 }
2041 
2042 void i915_disable_error_state(struct drm_i915_private *i915, int err)
2043 {
2044 	spin_lock_irq(&i915->gpu_error.lock);
2045 	if (!i915->gpu_error.first_error)
2046 		i915->gpu_error.first_error = ERR_PTR(err);
2047 	spin_unlock_irq(&i915->gpu_error.lock);
2048 }
2049