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/highmem.h>
32 #include <linux/nmi.h>
33 #include <linux/pagevec.h>
34 #include <linux/scatterlist.h>
35 #include <linux/string_helpers.h>
36 #include <linux/utsname.h>
37 #include <linux/zlib.h>
38 
39 #include <drm/drm_cache.h>
40 #include <drm/drm_print.h>
41 
42 #include "display/intel_dmc.h"
43 #include "display/intel_overlay.h"
44 
45 #include "gem/i915_gem_context.h"
46 #include "gem/i915_gem_lmem.h"
47 #include "gt/intel_engine_regs.h"
48 #include "gt/intel_gt.h"
49 #include "gt/intel_gt_mcr.h"
50 #include "gt/intel_gt_pm.h"
51 #include "gt/intel_gt_regs.h"
52 #include "gt/uc/intel_guc_capture.h"
53 
54 #include "i915_driver.h"
55 #include "i915_drv.h"
56 #include "i915_gpu_error.h"
57 #include "i915_memcpy.h"
58 #include "i915_scatterlist.h"
59 #include "i915_utils.h"
60 
61 #define ALLOW_FAIL (__GFP_KSWAPD_RECLAIM | __GFP_RETRY_MAYFAIL | __GFP_NOWARN)
62 #define ATOMIC_MAYFAIL (GFP_ATOMIC | __GFP_NOWARN)
63 
64 static void __sg_set_buf(struct scatterlist *sg,
65 			 void *addr, unsigned int len, loff_t it)
66 {
67 	sg->page_link = (unsigned long)virt_to_page(addr);
68 	sg->offset = offset_in_page(addr);
69 	sg->length = len;
70 	sg->dma_address = it;
71 }
72 
73 static bool __i915_error_grow(struct drm_i915_error_state_buf *e, size_t len)
74 {
75 	if (!len)
76 		return false;
77 
78 	if (e->bytes + len + 1 <= e->size)
79 		return true;
80 
81 	if (e->bytes) {
82 		__sg_set_buf(e->cur++, e->buf, e->bytes, e->iter);
83 		e->iter += e->bytes;
84 		e->buf = NULL;
85 		e->bytes = 0;
86 	}
87 
88 	if (e->cur == e->end) {
89 		struct scatterlist *sgl;
90 
91 		sgl = (typeof(sgl))__get_free_page(ALLOW_FAIL);
92 		if (!sgl) {
93 			e->err = -ENOMEM;
94 			return false;
95 		}
96 
97 		if (e->cur) {
98 			e->cur->offset = 0;
99 			e->cur->length = 0;
100 			e->cur->page_link =
101 				(unsigned long)sgl | SG_CHAIN;
102 		} else {
103 			e->sgl = sgl;
104 		}
105 
106 		e->cur = sgl;
107 		e->end = sgl + SG_MAX_SINGLE_ALLOC - 1;
108 	}
109 
110 	e->size = ALIGN(len + 1, SZ_64K);
111 	e->buf = kmalloc(e->size, ALLOW_FAIL);
112 	if (!e->buf) {
113 		e->size = PAGE_ALIGN(len + 1);
114 		e->buf = kmalloc(e->size, GFP_KERNEL);
115 	}
116 	if (!e->buf) {
117 		e->err = -ENOMEM;
118 		return false;
119 	}
120 
121 	return true;
122 }
123 
124 __printf(2, 0)
125 static void i915_error_vprintf(struct drm_i915_error_state_buf *e,
126 			       const char *fmt, va_list args)
127 {
128 	va_list ap;
129 	int len;
130 
131 	if (e->err)
132 		return;
133 
134 	va_copy(ap, args);
135 	len = vsnprintf(NULL, 0, fmt, ap);
136 	va_end(ap);
137 	if (len <= 0) {
138 		e->err = len;
139 		return;
140 	}
141 
142 	if (!__i915_error_grow(e, len))
143 		return;
144 
145 	GEM_BUG_ON(e->bytes >= e->size);
146 	len = vscnprintf(e->buf + e->bytes, e->size - e->bytes, fmt, args);
147 	if (len < 0) {
148 		e->err = len;
149 		return;
150 	}
151 	e->bytes += len;
152 }
153 
154 static void i915_error_puts(struct drm_i915_error_state_buf *e, const char *str)
155 {
156 	unsigned len;
157 
158 	if (e->err || !str)
159 		return;
160 
161 	len = strlen(str);
162 	if (!__i915_error_grow(e, len))
163 		return;
164 
165 	GEM_BUG_ON(e->bytes + len > e->size);
166 	memcpy(e->buf + e->bytes, str, len);
167 	e->bytes += len;
168 }
169 
170 #define err_printf(e, ...) i915_error_printf(e, __VA_ARGS__)
171 #define err_puts(e, s) i915_error_puts(e, s)
172 
173 static void __i915_printfn_error(struct drm_printer *p, struct va_format *vaf)
174 {
175 	i915_error_vprintf(p->arg, vaf->fmt, *vaf->va);
176 }
177 
178 static inline struct drm_printer
179 i915_error_printer(struct drm_i915_error_state_buf *e)
180 {
181 	struct drm_printer p = {
182 		.printfn = __i915_printfn_error,
183 		.arg = e,
184 	};
185 	return p;
186 }
187 
188 /* single threaded page allocator with a reserved stash for emergencies */
189 static void pool_fini(struct pagevec *pv)
190 {
191 	pagevec_release(pv);
192 }
193 
194 static int pool_refill(struct pagevec *pv, gfp_t gfp)
195 {
196 	while (pagevec_space(pv)) {
197 		struct page *p;
198 
199 		p = alloc_page(gfp);
200 		if (!p)
201 			return -ENOMEM;
202 
203 		pagevec_add(pv, p);
204 	}
205 
206 	return 0;
207 }
208 
209 static int pool_init(struct pagevec *pv, gfp_t gfp)
210 {
211 	int err;
212 
213 	pagevec_init(pv);
214 
215 	err = pool_refill(pv, gfp);
216 	if (err)
217 		pool_fini(pv);
218 
219 	return err;
220 }
221 
222 static void *pool_alloc(struct pagevec *pv, gfp_t gfp)
223 {
224 	struct page *p;
225 
226 	p = alloc_page(gfp);
227 	if (!p && pagevec_count(pv))
228 		p = pv->pages[--pv->nr];
229 
230 	return p ? page_address(p) : NULL;
231 }
232 
233 static void pool_free(struct pagevec *pv, void *addr)
234 {
235 	struct page *p = virt_to_page(addr);
236 
237 	if (pagevec_space(pv))
238 		pagevec_add(pv, p);
239 	else
240 		__free_page(p);
241 }
242 
243 #ifdef CONFIG_DRM_I915_COMPRESS_ERROR
244 
245 struct i915_vma_compress {
246 	struct pagevec pool;
247 	struct z_stream_s zstream;
248 	void *tmp;
249 };
250 
251 static bool compress_init(struct i915_vma_compress *c)
252 {
253 	struct z_stream_s *zstream = &c->zstream;
254 
255 	if (pool_init(&c->pool, ALLOW_FAIL))
256 		return false;
257 
258 	zstream->workspace =
259 		kmalloc(zlib_deflate_workspacesize(MAX_WBITS, MAX_MEM_LEVEL),
260 			ALLOW_FAIL);
261 	if (!zstream->workspace) {
262 		pool_fini(&c->pool);
263 		return false;
264 	}
265 
266 	c->tmp = NULL;
267 	if (i915_has_memcpy_from_wc())
268 		c->tmp = pool_alloc(&c->pool, ALLOW_FAIL);
269 
270 	return true;
271 }
272 
273 static bool compress_start(struct i915_vma_compress *c)
274 {
275 	struct z_stream_s *zstream = &c->zstream;
276 	void *workspace = zstream->workspace;
277 
278 	memset(zstream, 0, sizeof(*zstream));
279 	zstream->workspace = workspace;
280 
281 	return zlib_deflateInit(zstream, Z_DEFAULT_COMPRESSION) == Z_OK;
282 }
283 
284 static void *compress_next_page(struct i915_vma_compress *c,
285 				struct i915_vma_coredump *dst)
286 {
287 	void *page_addr;
288 	struct page *page;
289 
290 	page_addr = pool_alloc(&c->pool, ALLOW_FAIL);
291 	if (!page_addr)
292 		return ERR_PTR(-ENOMEM);
293 
294 	page = virt_to_page(page_addr);
295 	list_add_tail(&page->lru, &dst->page_list);
296 	return page_addr;
297 }
298 
299 static int compress_page(struct i915_vma_compress *c,
300 			 void *src,
301 			 struct i915_vma_coredump *dst,
302 			 bool wc)
303 {
304 	struct z_stream_s *zstream = &c->zstream;
305 
306 	zstream->next_in = src;
307 	if (wc && c->tmp && i915_memcpy_from_wc(c->tmp, src, PAGE_SIZE))
308 		zstream->next_in = c->tmp;
309 	zstream->avail_in = PAGE_SIZE;
310 
311 	do {
312 		if (zstream->avail_out == 0) {
313 			zstream->next_out = compress_next_page(c, dst);
314 			if (IS_ERR(zstream->next_out))
315 				return PTR_ERR(zstream->next_out);
316 
317 			zstream->avail_out = PAGE_SIZE;
318 		}
319 
320 		if (zlib_deflate(zstream, Z_NO_FLUSH) != Z_OK)
321 			return -EIO;
322 
323 		cond_resched();
324 	} while (zstream->avail_in);
325 
326 	/* Fallback to uncompressed if we increase size? */
327 	if (0 && zstream->total_out > zstream->total_in)
328 		return -E2BIG;
329 
330 	return 0;
331 }
332 
333 static int compress_flush(struct i915_vma_compress *c,
334 			  struct i915_vma_coredump *dst)
335 {
336 	struct z_stream_s *zstream = &c->zstream;
337 
338 	do {
339 		switch (zlib_deflate(zstream, Z_FINISH)) {
340 		case Z_OK: /* more space requested */
341 			zstream->next_out = compress_next_page(c, dst);
342 			if (IS_ERR(zstream->next_out))
343 				return PTR_ERR(zstream->next_out);
344 
345 			zstream->avail_out = PAGE_SIZE;
346 			break;
347 
348 		case Z_STREAM_END:
349 			goto end;
350 
351 		default: /* any error */
352 			return -EIO;
353 		}
354 	} while (1);
355 
356 end:
357 	memset(zstream->next_out, 0, zstream->avail_out);
358 	dst->unused = zstream->avail_out;
359 	return 0;
360 }
361 
362 static void compress_finish(struct i915_vma_compress *c)
363 {
364 	zlib_deflateEnd(&c->zstream);
365 }
366 
367 static void compress_fini(struct i915_vma_compress *c)
368 {
369 	kfree(c->zstream.workspace);
370 	if (c->tmp)
371 		pool_free(&c->pool, c->tmp);
372 	pool_fini(&c->pool);
373 }
374 
375 static void err_compression_marker(struct drm_i915_error_state_buf *m)
376 {
377 	err_puts(m, ":");
378 }
379 
380 #else
381 
382 struct i915_vma_compress {
383 	struct pagevec pool;
384 };
385 
386 static bool compress_init(struct i915_vma_compress *c)
387 {
388 	return pool_init(&c->pool, ALLOW_FAIL) == 0;
389 }
390 
391 static bool compress_start(struct i915_vma_compress *c)
392 {
393 	return true;
394 }
395 
396 static int compress_page(struct i915_vma_compress *c,
397 			 void *src,
398 			 struct i915_vma_coredump *dst,
399 			 bool wc)
400 {
401 	void *ptr;
402 
403 	ptr = pool_alloc(&c->pool, ALLOW_FAIL);
404 	if (!ptr)
405 		return -ENOMEM;
406 
407 	if (!(wc && i915_memcpy_from_wc(ptr, src, PAGE_SIZE)))
408 		memcpy(ptr, src, PAGE_SIZE);
409 	list_add_tail(&virt_to_page(ptr)->lru, &dst->page_list);
410 	cond_resched();
411 
412 	return 0;
413 }
414 
415 static int compress_flush(struct i915_vma_compress *c,
416 			  struct i915_vma_coredump *dst)
417 {
418 	return 0;
419 }
420 
421 static void compress_finish(struct i915_vma_compress *c)
422 {
423 }
424 
425 static void compress_fini(struct i915_vma_compress *c)
426 {
427 	pool_fini(&c->pool);
428 }
429 
430 static void err_compression_marker(struct drm_i915_error_state_buf *m)
431 {
432 	err_puts(m, "~");
433 }
434 
435 #endif
436 
437 static void error_print_instdone(struct drm_i915_error_state_buf *m,
438 				 const struct intel_engine_coredump *ee)
439 {
440 	int slice;
441 	int subslice;
442 	int iter;
443 
444 	err_printf(m, "  INSTDONE: 0x%08x\n",
445 		   ee->instdone.instdone);
446 
447 	if (ee->engine->class != RENDER_CLASS || GRAPHICS_VER(m->i915) <= 3)
448 		return;
449 
450 	err_printf(m, "  SC_INSTDONE: 0x%08x\n",
451 		   ee->instdone.slice_common);
452 
453 	if (GRAPHICS_VER(m->i915) <= 6)
454 		return;
455 
456 	for_each_ss_steering(iter, ee->engine->gt, slice, subslice)
457 		err_printf(m, "  SAMPLER_INSTDONE[%d][%d]: 0x%08x\n",
458 			   slice, subslice,
459 			   ee->instdone.sampler[slice][subslice]);
460 
461 	for_each_ss_steering(iter, ee->engine->gt, slice, subslice)
462 		err_printf(m, "  ROW_INSTDONE[%d][%d]: 0x%08x\n",
463 			   slice, subslice,
464 			   ee->instdone.row[slice][subslice]);
465 
466 	if (GRAPHICS_VER(m->i915) < 12)
467 		return;
468 
469 	if (GRAPHICS_VER_FULL(m->i915) >= IP_VER(12, 55)) {
470 		for_each_ss_steering(iter, ee->engine->gt, slice, subslice)
471 			err_printf(m, "  GEOM_SVGUNIT_INSTDONE[%d][%d]: 0x%08x\n",
472 				   slice, subslice,
473 				   ee->instdone.geom_svg[slice][subslice]);
474 	}
475 
476 	err_printf(m, "  SC_INSTDONE_EXTRA: 0x%08x\n",
477 		   ee->instdone.slice_common_extra[0]);
478 	err_printf(m, "  SC_INSTDONE_EXTRA2: 0x%08x\n",
479 		   ee->instdone.slice_common_extra[1]);
480 }
481 
482 static void error_print_request(struct drm_i915_error_state_buf *m,
483 				const char *prefix,
484 				const struct i915_request_coredump *erq)
485 {
486 	if (!erq->seqno)
487 		return;
488 
489 	err_printf(m, "%s pid %d, seqno %8x:%08x%s%s, prio %d, head %08x, tail %08x\n",
490 		   prefix, erq->pid, erq->context, erq->seqno,
491 		   test_bit(DMA_FENCE_FLAG_SIGNALED_BIT,
492 			    &erq->flags) ? "!" : "",
493 		   test_bit(DMA_FENCE_FLAG_ENABLE_SIGNAL_BIT,
494 			    &erq->flags) ? "+" : "",
495 		   erq->sched_attr.priority,
496 		   erq->head, erq->tail);
497 }
498 
499 static void error_print_context(struct drm_i915_error_state_buf *m,
500 				const char *header,
501 				const struct i915_gem_context_coredump *ctx)
502 {
503 	err_printf(m, "%s%s[%d] prio %d, guilty %d active %d, runtime total %lluns, avg %lluns\n",
504 		   header, ctx->comm, ctx->pid, ctx->sched_attr.priority,
505 		   ctx->guilty, ctx->active,
506 		   ctx->total_runtime, ctx->avg_runtime);
507 }
508 
509 static struct i915_vma_coredump *
510 __find_vma(struct i915_vma_coredump *vma, const char *name)
511 {
512 	while (vma) {
513 		if (strcmp(vma->name, name) == 0)
514 			return vma;
515 		vma = vma->next;
516 	}
517 
518 	return NULL;
519 }
520 
521 struct i915_vma_coredump *
522 intel_gpu_error_find_batch(const struct intel_engine_coredump *ee)
523 {
524 	return __find_vma(ee->vma, "batch");
525 }
526 
527 static void error_print_engine(struct drm_i915_error_state_buf *m,
528 			       const struct intel_engine_coredump *ee)
529 {
530 	struct i915_vma_coredump *batch;
531 	int n;
532 
533 	err_printf(m, "%s command stream:\n", ee->engine->name);
534 	err_printf(m, "  CCID:  0x%08x\n", ee->ccid);
535 	err_printf(m, "  START: 0x%08x\n", ee->start);
536 	err_printf(m, "  HEAD:  0x%08x [0x%08x]\n", ee->head, ee->rq_head);
537 	err_printf(m, "  TAIL:  0x%08x [0x%08x, 0x%08x]\n",
538 		   ee->tail, ee->rq_post, ee->rq_tail);
539 	err_printf(m, "  CTL:   0x%08x\n", ee->ctl);
540 	err_printf(m, "  MODE:  0x%08x\n", ee->mode);
541 	err_printf(m, "  HWS:   0x%08x\n", ee->hws);
542 	err_printf(m, "  ACTHD: 0x%08x %08x\n",
543 		   (u32)(ee->acthd>>32), (u32)ee->acthd);
544 	err_printf(m, "  IPEIR: 0x%08x\n", ee->ipeir);
545 	err_printf(m, "  IPEHR: 0x%08x\n", ee->ipehr);
546 	err_printf(m, "  ESR:   0x%08x\n", ee->esr);
547 
548 	error_print_instdone(m, ee);
549 
550 	batch = intel_gpu_error_find_batch(ee);
551 	if (batch) {
552 		u64 start = batch->gtt_offset;
553 		u64 end = start + batch->gtt_size;
554 
555 		err_printf(m, "  batch: [0x%08x_%08x, 0x%08x_%08x]\n",
556 			   upper_32_bits(start), lower_32_bits(start),
557 			   upper_32_bits(end), lower_32_bits(end));
558 	}
559 	if (GRAPHICS_VER(m->i915) >= 4) {
560 		err_printf(m, "  BBADDR: 0x%08x_%08x\n",
561 			   (u32)(ee->bbaddr>>32), (u32)ee->bbaddr);
562 		err_printf(m, "  BB_STATE: 0x%08x\n", ee->bbstate);
563 		err_printf(m, "  INSTPS: 0x%08x\n", ee->instps);
564 	}
565 	err_printf(m, "  INSTPM: 0x%08x\n", ee->instpm);
566 	err_printf(m, "  FADDR: 0x%08x %08x\n", upper_32_bits(ee->faddr),
567 		   lower_32_bits(ee->faddr));
568 	if (GRAPHICS_VER(m->i915) >= 6) {
569 		err_printf(m, "  RC PSMI: 0x%08x\n", ee->rc_psmi);
570 		err_printf(m, "  FAULT_REG: 0x%08x\n", ee->fault_reg);
571 	}
572 	if (GRAPHICS_VER(m->i915) >= 11) {
573 		err_printf(m, "  NOPID: 0x%08x\n", ee->nopid);
574 		err_printf(m, "  EXCC: 0x%08x\n", ee->excc);
575 		err_printf(m, "  CMD_CCTL: 0x%08x\n", ee->cmd_cctl);
576 		err_printf(m, "  CSCMDOP: 0x%08x\n", ee->cscmdop);
577 		err_printf(m, "  CTX_SR_CTL: 0x%08x\n", ee->ctx_sr_ctl);
578 		err_printf(m, "  DMA_FADDR_HI: 0x%08x\n", ee->dma_faddr_hi);
579 		err_printf(m, "  DMA_FADDR_LO: 0x%08x\n", ee->dma_faddr_lo);
580 	}
581 	if (HAS_PPGTT(m->i915)) {
582 		err_printf(m, "  GFX_MODE: 0x%08x\n", ee->vm_info.gfx_mode);
583 
584 		if (GRAPHICS_VER(m->i915) >= 8) {
585 			int i;
586 			for (i = 0; i < 4; i++)
587 				err_printf(m, "  PDP%d: 0x%016llx\n",
588 					   i, ee->vm_info.pdp[i]);
589 		} else {
590 			err_printf(m, "  PP_DIR_BASE: 0x%08x\n",
591 				   ee->vm_info.pp_dir_base);
592 		}
593 	}
594 
595 	for (n = 0; n < ee->num_ports; n++) {
596 		err_printf(m, "  ELSP[%d]:", n);
597 		error_print_request(m, " ", &ee->execlist[n]);
598 	}
599 }
600 
601 void i915_error_printf(struct drm_i915_error_state_buf *e, const char *f, ...)
602 {
603 	va_list args;
604 
605 	va_start(args, f);
606 	i915_error_vprintf(e, f, args);
607 	va_end(args);
608 }
609 
610 void intel_gpu_error_print_vma(struct drm_i915_error_state_buf *m,
611 			       const struct intel_engine_cs *engine,
612 			       const struct i915_vma_coredump *vma)
613 {
614 	char out[ASCII85_BUFSZ];
615 	struct page *page;
616 
617 	if (!vma)
618 		return;
619 
620 	err_printf(m, "%s --- %s = 0x%08x %08x\n",
621 		   engine ? engine->name : "global", vma->name,
622 		   upper_32_bits(vma->gtt_offset),
623 		   lower_32_bits(vma->gtt_offset));
624 
625 	if (vma->gtt_page_sizes > I915_GTT_PAGE_SIZE_4K)
626 		err_printf(m, "gtt_page_sizes = 0x%08x\n", vma->gtt_page_sizes);
627 
628 	err_compression_marker(m);
629 	list_for_each_entry(page, &vma->page_list, lru) {
630 		int i, len;
631 		const u32 *addr = page_address(page);
632 
633 		len = PAGE_SIZE;
634 		if (page == list_last_entry(&vma->page_list, typeof(*page), lru))
635 			len -= vma->unused;
636 		len = ascii85_encode_len(len);
637 
638 		for (i = 0; i < len; i++)
639 			err_puts(m, ascii85_encode(addr[i], out));
640 	}
641 	err_puts(m, "\n");
642 }
643 
644 static void err_print_capabilities(struct drm_i915_error_state_buf *m,
645 				   struct i915_gpu_coredump *error)
646 {
647 	struct drm_printer p = i915_error_printer(m);
648 
649 	intel_device_info_print_static(&error->device_info, &p);
650 	intel_device_info_print_runtime(&error->runtime_info, &p);
651 	intel_driver_caps_print(&error->driver_caps, &p);
652 }
653 
654 static void err_print_params(struct drm_i915_error_state_buf *m,
655 			     const struct i915_params *params)
656 {
657 	struct drm_printer p = i915_error_printer(m);
658 
659 	i915_params_dump(params, &p);
660 }
661 
662 static void err_print_pciid(struct drm_i915_error_state_buf *m,
663 			    struct drm_i915_private *i915)
664 {
665 	struct pci_dev *pdev = to_pci_dev(i915->drm.dev);
666 
667 	err_printf(m, "PCI ID: 0x%04x\n", pdev->device);
668 	err_printf(m, "PCI Revision: 0x%02x\n", pdev->revision);
669 	err_printf(m, "PCI Subsystem: %04x:%04x\n",
670 		   pdev->subsystem_vendor,
671 		   pdev->subsystem_device);
672 }
673 
674 static void err_print_guc_ctb(struct drm_i915_error_state_buf *m,
675 			      const char *name,
676 			      const struct intel_ctb_coredump *ctb)
677 {
678 	if (!ctb->size)
679 		return;
680 
681 	err_printf(m, "GuC %s CTB: raw: 0x%08X, 0x%08X/%08X, cached: 0x%08X/%08X, desc = 0x%08X, buf = 0x%08X x 0x%08X\n",
682 		   name, ctb->raw_status, ctb->raw_head, ctb->raw_tail,
683 		   ctb->head, ctb->tail, ctb->desc_offset, ctb->cmds_offset, ctb->size);
684 }
685 
686 static void err_print_uc(struct drm_i915_error_state_buf *m,
687 			 const struct intel_uc_coredump *error_uc)
688 {
689 	struct drm_printer p = i915_error_printer(m);
690 
691 	intel_uc_fw_dump(&error_uc->guc_fw, &p);
692 	intel_uc_fw_dump(&error_uc->huc_fw, &p);
693 	err_printf(m, "GuC timestamp: 0x%08x\n", error_uc->guc.timestamp);
694 	intel_gpu_error_print_vma(m, NULL, error_uc->guc.vma_log);
695 	err_printf(m, "GuC CTB fence: %d\n", error_uc->guc.last_fence);
696 	err_print_guc_ctb(m, "Send", error_uc->guc.ctb + 0);
697 	err_print_guc_ctb(m, "Recv", error_uc->guc.ctb + 1);
698 	intel_gpu_error_print_vma(m, NULL, error_uc->guc.vma_ctb);
699 }
700 
701 static void err_free_sgl(struct scatterlist *sgl)
702 {
703 	while (sgl) {
704 		struct scatterlist *sg;
705 
706 		for (sg = sgl; !sg_is_chain(sg); sg++) {
707 			kfree(sg_virt(sg));
708 			if (sg_is_last(sg))
709 				break;
710 		}
711 
712 		sg = sg_is_last(sg) ? NULL : sg_chain_ptr(sg);
713 		free_page((unsigned long)sgl);
714 		sgl = sg;
715 	}
716 }
717 
718 static void err_print_gt_info(struct drm_i915_error_state_buf *m,
719 			      struct intel_gt_coredump *gt)
720 {
721 	struct drm_printer p = i915_error_printer(m);
722 
723 	intel_gt_info_print(&gt->info, &p);
724 	intel_sseu_print_topology(gt->_gt->i915, &gt->info.sseu, &p);
725 }
726 
727 static void err_print_gt_display(struct drm_i915_error_state_buf *m,
728 				 struct intel_gt_coredump *gt)
729 {
730 	err_printf(m, "IER: 0x%08x\n", gt->ier);
731 	err_printf(m, "DERRMR: 0x%08x\n", gt->derrmr);
732 }
733 
734 static void err_print_gt_global_nonguc(struct drm_i915_error_state_buf *m,
735 				       struct intel_gt_coredump *gt)
736 {
737 	int i;
738 
739 	err_printf(m, "GT awake: %s\n", str_yes_no(gt->awake));
740 	err_printf(m, "CS timestamp frequency: %u Hz, %d ns\n",
741 		   gt->clock_frequency, gt->clock_period_ns);
742 	err_printf(m, "EIR: 0x%08x\n", gt->eir);
743 	err_printf(m, "PGTBL_ER: 0x%08x\n", gt->pgtbl_er);
744 
745 	for (i = 0; i < gt->ngtier; i++)
746 		err_printf(m, "GTIER[%d]: 0x%08x\n", i, gt->gtier[i]);
747 }
748 
749 static void err_print_gt_global(struct drm_i915_error_state_buf *m,
750 				struct intel_gt_coredump *gt)
751 {
752 	err_printf(m, "FORCEWAKE: 0x%08x\n", gt->forcewake);
753 
754 	if (IS_GRAPHICS_VER(m->i915, 6, 11)) {
755 		err_printf(m, "ERROR: 0x%08x\n", gt->error);
756 		err_printf(m, "DONE_REG: 0x%08x\n", gt->done_reg);
757 	}
758 
759 	if (GRAPHICS_VER(m->i915) >= 8)
760 		err_printf(m, "FAULT_TLB_DATA: 0x%08x 0x%08x\n",
761 			   gt->fault_data1, gt->fault_data0);
762 
763 	if (GRAPHICS_VER(m->i915) == 7)
764 		err_printf(m, "ERR_INT: 0x%08x\n", gt->err_int);
765 
766 	if (IS_GRAPHICS_VER(m->i915, 8, 11))
767 		err_printf(m, "GTT_CACHE_EN: 0x%08x\n", gt->gtt_cache);
768 
769 	if (GRAPHICS_VER(m->i915) == 12)
770 		err_printf(m, "AUX_ERR_DBG: 0x%08x\n", gt->aux_err);
771 
772 	if (GRAPHICS_VER(m->i915) >= 12) {
773 		int i;
774 
775 		for (i = 0; i < I915_MAX_SFC; i++) {
776 			/*
777 			 * SFC_DONE resides in the VD forcewake domain, so it
778 			 * only exists if the corresponding VCS engine is
779 			 * present.
780 			 */
781 			if ((gt->_gt->info.sfc_mask & BIT(i)) == 0 ||
782 			    !HAS_ENGINE(gt->_gt, _VCS(i * 2)))
783 				continue;
784 
785 			err_printf(m, "  SFC_DONE[%d]: 0x%08x\n", i,
786 				   gt->sfc_done[i]);
787 		}
788 
789 		err_printf(m, "  GAM_DONE: 0x%08x\n", gt->gam_done);
790 	}
791 }
792 
793 static void err_print_gt_fences(struct drm_i915_error_state_buf *m,
794 				struct intel_gt_coredump *gt)
795 {
796 	int i;
797 
798 	for (i = 0; i < gt->nfence; i++)
799 		err_printf(m, "  fence[%d] = %08llx\n", i, gt->fence[i]);
800 }
801 
802 static void err_print_gt_engines(struct drm_i915_error_state_buf *m,
803 				 struct intel_gt_coredump *gt)
804 {
805 	const struct intel_engine_coredump *ee;
806 
807 	for (ee = gt->engine; ee; ee = ee->next) {
808 		const struct i915_vma_coredump *vma;
809 
810 		if (ee->guc_capture_node)
811 			intel_guc_capture_print_engine_node(m, ee);
812 		else
813 			error_print_engine(m, ee);
814 
815 		err_printf(m, "  hung: %u\n", ee->hung);
816 		err_printf(m, "  engine reset count: %u\n", ee->reset_count);
817 		error_print_context(m, "  Active context: ", &ee->context);
818 
819 		for (vma = ee->vma; vma; vma = vma->next)
820 			intel_gpu_error_print_vma(m, ee->engine, vma);
821 	}
822 
823 }
824 
825 static void __err_print_to_sgl(struct drm_i915_error_state_buf *m,
826 			       struct i915_gpu_coredump *error)
827 {
828 	const struct intel_engine_coredump *ee;
829 	struct timespec64 ts;
830 
831 	if (*error->error_msg)
832 		err_printf(m, "%s\n", error->error_msg);
833 	err_printf(m, "Kernel: %s %s\n",
834 		   init_utsname()->release,
835 		   init_utsname()->machine);
836 	err_printf(m, "Driver: %s\n", DRIVER_DATE);
837 	ts = ktime_to_timespec64(error->time);
838 	err_printf(m, "Time: %lld s %ld us\n",
839 		   (s64)ts.tv_sec, ts.tv_nsec / NSEC_PER_USEC);
840 	ts = ktime_to_timespec64(error->boottime);
841 	err_printf(m, "Boottime: %lld s %ld us\n",
842 		   (s64)ts.tv_sec, ts.tv_nsec / NSEC_PER_USEC);
843 	ts = ktime_to_timespec64(error->uptime);
844 	err_printf(m, "Uptime: %lld s %ld us\n",
845 		   (s64)ts.tv_sec, ts.tv_nsec / NSEC_PER_USEC);
846 	err_printf(m, "Capture: %lu jiffies; %d ms ago\n",
847 		   error->capture, jiffies_to_msecs(jiffies - error->capture));
848 
849 	for (ee = error->gt ? error->gt->engine : NULL; ee; ee = ee->next)
850 		err_printf(m, "Active process (on ring %s): %s [%d]\n",
851 			   ee->engine->name,
852 			   ee->context.comm,
853 			   ee->context.pid);
854 
855 	err_printf(m, "Reset count: %u\n", error->reset_count);
856 	err_printf(m, "Suspend count: %u\n", error->suspend_count);
857 	err_printf(m, "Platform: %s\n", intel_platform_name(error->device_info.platform));
858 	err_printf(m, "Subplatform: 0x%x\n",
859 		   intel_subplatform(&error->runtime_info,
860 				     error->device_info.platform));
861 	err_print_pciid(m, m->i915);
862 
863 	err_printf(m, "IOMMU enabled?: %d\n", error->iommu);
864 
865 	intel_dmc_print_error_state(m, m->i915);
866 
867 	err_printf(m, "RPM wakelock: %s\n", str_yes_no(error->wakelock));
868 	err_printf(m, "PM suspended: %s\n", str_yes_no(error->suspended));
869 
870 	if (error->gt) {
871 		bool print_guc_capture = false;
872 
873 		if (error->gt->uc && error->gt->uc->guc.is_guc_capture)
874 			print_guc_capture = true;
875 
876 		err_print_gt_display(m, error->gt);
877 		err_print_gt_global_nonguc(m, error->gt);
878 		err_print_gt_fences(m, error->gt);
879 
880 		/*
881 		 * GuC dumped global, eng-class and eng-instance registers together
882 		 * as part of engine state dump so we print in err_print_gt_engines
883 		 */
884 		if (!print_guc_capture)
885 			err_print_gt_global(m, error->gt);
886 
887 		err_print_gt_engines(m, error->gt);
888 
889 		if (error->gt->uc)
890 			err_print_uc(m, error->gt->uc);
891 
892 		err_print_gt_info(m, error->gt);
893 	}
894 
895 	if (error->overlay)
896 		intel_overlay_print_error_state(m, error->overlay);
897 
898 	err_print_capabilities(m, error);
899 	err_print_params(m, &error->params);
900 }
901 
902 static int err_print_to_sgl(struct i915_gpu_coredump *error)
903 {
904 	struct drm_i915_error_state_buf m;
905 
906 	if (IS_ERR(error))
907 		return PTR_ERR(error);
908 
909 	if (READ_ONCE(error->sgl))
910 		return 0;
911 
912 	memset(&m, 0, sizeof(m));
913 	m.i915 = error->i915;
914 
915 	__err_print_to_sgl(&m, error);
916 
917 	if (m.buf) {
918 		__sg_set_buf(m.cur++, m.buf, m.bytes, m.iter);
919 		m.bytes = 0;
920 		m.buf = NULL;
921 	}
922 	if (m.cur) {
923 		GEM_BUG_ON(m.end < m.cur);
924 		sg_mark_end(m.cur - 1);
925 	}
926 	GEM_BUG_ON(m.sgl && !m.cur);
927 
928 	if (m.err) {
929 		err_free_sgl(m.sgl);
930 		return m.err;
931 	}
932 
933 	if (cmpxchg(&error->sgl, NULL, m.sgl))
934 		err_free_sgl(m.sgl);
935 
936 	return 0;
937 }
938 
939 ssize_t i915_gpu_coredump_copy_to_buffer(struct i915_gpu_coredump *error,
940 					 char *buf, loff_t off, size_t rem)
941 {
942 	struct scatterlist *sg;
943 	size_t count;
944 	loff_t pos;
945 	int err;
946 
947 	if (!error || !rem)
948 		return 0;
949 
950 	err = err_print_to_sgl(error);
951 	if (err)
952 		return err;
953 
954 	sg = READ_ONCE(error->fit);
955 	if (!sg || off < sg->dma_address)
956 		sg = error->sgl;
957 	if (!sg)
958 		return 0;
959 
960 	pos = sg->dma_address;
961 	count = 0;
962 	do {
963 		size_t len, start;
964 
965 		if (sg_is_chain(sg)) {
966 			sg = sg_chain_ptr(sg);
967 			GEM_BUG_ON(sg_is_chain(sg));
968 		}
969 
970 		len = sg->length;
971 		if (pos + len <= off) {
972 			pos += len;
973 			continue;
974 		}
975 
976 		start = sg->offset;
977 		if (pos < off) {
978 			GEM_BUG_ON(off - pos > len);
979 			len -= off - pos;
980 			start += off - pos;
981 			pos = off;
982 		}
983 
984 		len = min(len, rem);
985 		GEM_BUG_ON(!len || len > sg->length);
986 
987 		memcpy(buf, page_address(sg_page(sg)) + start, len);
988 
989 		count += len;
990 		pos += len;
991 
992 		buf += len;
993 		rem -= len;
994 		if (!rem) {
995 			WRITE_ONCE(error->fit, sg);
996 			break;
997 		}
998 	} while (!sg_is_last(sg++));
999 
1000 	return count;
1001 }
1002 
1003 static void i915_vma_coredump_free(struct i915_vma_coredump *vma)
1004 {
1005 	while (vma) {
1006 		struct i915_vma_coredump *next = vma->next;
1007 		struct page *page, *n;
1008 
1009 		list_for_each_entry_safe(page, n, &vma->page_list, lru) {
1010 			list_del_init(&page->lru);
1011 			__free_page(page);
1012 		}
1013 
1014 		kfree(vma);
1015 		vma = next;
1016 	}
1017 }
1018 
1019 static void cleanup_params(struct i915_gpu_coredump *error)
1020 {
1021 	i915_params_free(&error->params);
1022 }
1023 
1024 static void cleanup_uc(struct intel_uc_coredump *uc)
1025 {
1026 	kfree(uc->guc_fw.file_selected.path);
1027 	kfree(uc->huc_fw.file_selected.path);
1028 	kfree(uc->guc_fw.file_wanted.path);
1029 	kfree(uc->huc_fw.file_wanted.path);
1030 	i915_vma_coredump_free(uc->guc.vma_log);
1031 	i915_vma_coredump_free(uc->guc.vma_ctb);
1032 
1033 	kfree(uc);
1034 }
1035 
1036 static void cleanup_gt(struct intel_gt_coredump *gt)
1037 {
1038 	while (gt->engine) {
1039 		struct intel_engine_coredump *ee = gt->engine;
1040 
1041 		gt->engine = ee->next;
1042 
1043 		i915_vma_coredump_free(ee->vma);
1044 		intel_guc_capture_free_node(ee);
1045 		kfree(ee);
1046 	}
1047 
1048 	if (gt->uc)
1049 		cleanup_uc(gt->uc);
1050 
1051 	kfree(gt);
1052 }
1053 
1054 void __i915_gpu_coredump_free(struct kref *error_ref)
1055 {
1056 	struct i915_gpu_coredump *error =
1057 		container_of(error_ref, typeof(*error), ref);
1058 
1059 	while (error->gt) {
1060 		struct intel_gt_coredump *gt = error->gt;
1061 
1062 		error->gt = gt->next;
1063 		cleanup_gt(gt);
1064 	}
1065 
1066 	kfree(error->overlay);
1067 
1068 	cleanup_params(error);
1069 
1070 	err_free_sgl(error->sgl);
1071 	kfree(error);
1072 }
1073 
1074 static struct i915_vma_coredump *
1075 i915_vma_coredump_create(const struct intel_gt *gt,
1076 			 const struct i915_vma_resource *vma_res,
1077 			 struct i915_vma_compress *compress,
1078 			 const char *name)
1079 
1080 {
1081 	struct i915_ggtt *ggtt = gt->ggtt;
1082 	const u64 slot = ggtt->error_capture.start;
1083 	struct i915_vma_coredump *dst;
1084 	struct sgt_iter iter;
1085 	int ret;
1086 
1087 	might_sleep();
1088 
1089 	if (!vma_res || !vma_res->bi.pages || !compress)
1090 		return NULL;
1091 
1092 	dst = kmalloc(sizeof(*dst), ALLOW_FAIL);
1093 	if (!dst)
1094 		return NULL;
1095 
1096 	if (!compress_start(compress)) {
1097 		kfree(dst);
1098 		return NULL;
1099 	}
1100 
1101 	INIT_LIST_HEAD(&dst->page_list);
1102 	strcpy(dst->name, name);
1103 	dst->next = NULL;
1104 
1105 	dst->gtt_offset = vma_res->start;
1106 	dst->gtt_size = vma_res->node_size;
1107 	dst->gtt_page_sizes = vma_res->page_sizes_gtt;
1108 	dst->unused = 0;
1109 
1110 	ret = -EINVAL;
1111 	if (drm_mm_node_allocated(&ggtt->error_capture)) {
1112 		void __iomem *s;
1113 		dma_addr_t dma;
1114 
1115 		for_each_sgt_daddr(dma, iter, vma_res->bi.pages) {
1116 			mutex_lock(&ggtt->error_mutex);
1117 			if (ggtt->vm.raw_insert_page)
1118 				ggtt->vm.raw_insert_page(&ggtt->vm, dma, slot,
1119 							 I915_CACHE_NONE, 0);
1120 			else
1121 				ggtt->vm.insert_page(&ggtt->vm, dma, slot,
1122 						     I915_CACHE_NONE, 0);
1123 			mb();
1124 
1125 			s = io_mapping_map_wc(&ggtt->iomap, slot, PAGE_SIZE);
1126 			ret = compress_page(compress,
1127 					    (void  __force *)s, dst,
1128 					    true);
1129 			io_mapping_unmap(s);
1130 
1131 			mb();
1132 			ggtt->vm.clear_range(&ggtt->vm, slot, PAGE_SIZE);
1133 			mutex_unlock(&ggtt->error_mutex);
1134 			if (ret)
1135 				break;
1136 		}
1137 	} else if (vma_res->bi.lmem) {
1138 		struct intel_memory_region *mem = vma_res->mr;
1139 		dma_addr_t dma;
1140 
1141 		for_each_sgt_daddr(dma, iter, vma_res->bi.pages) {
1142 			dma_addr_t offset = dma - mem->region.start;
1143 			void __iomem *s;
1144 
1145 			if (offset + PAGE_SIZE > mem->io_size) {
1146 				ret = -EINVAL;
1147 				break;
1148 			}
1149 
1150 			s = io_mapping_map_wc(&mem->iomap, offset, PAGE_SIZE);
1151 			ret = compress_page(compress,
1152 					    (void __force *)s, dst,
1153 					    true);
1154 			io_mapping_unmap(s);
1155 			if (ret)
1156 				break;
1157 		}
1158 	} else {
1159 		struct page *page;
1160 
1161 		for_each_sgt_page(page, iter, vma_res->bi.pages) {
1162 			void *s;
1163 
1164 			drm_clflush_pages(&page, 1);
1165 
1166 			s = kmap(page);
1167 			ret = compress_page(compress, s, dst, false);
1168 			kunmap(page);
1169 
1170 			drm_clflush_pages(&page, 1);
1171 
1172 			if (ret)
1173 				break;
1174 		}
1175 	}
1176 
1177 	if (ret || compress_flush(compress, dst)) {
1178 		struct page *page, *n;
1179 
1180 		list_for_each_entry_safe_reverse(page, n, &dst->page_list, lru) {
1181 			list_del_init(&page->lru);
1182 			pool_free(&compress->pool, page_address(page));
1183 		}
1184 
1185 		kfree(dst);
1186 		dst = NULL;
1187 	}
1188 	compress_finish(compress);
1189 
1190 	return dst;
1191 }
1192 
1193 static void gt_record_fences(struct intel_gt_coredump *gt)
1194 {
1195 	struct i915_ggtt *ggtt = gt->_gt->ggtt;
1196 	struct intel_uncore *uncore = gt->_gt->uncore;
1197 	int i;
1198 
1199 	if (GRAPHICS_VER(uncore->i915) >= 6) {
1200 		for (i = 0; i < ggtt->num_fences; i++)
1201 			gt->fence[i] =
1202 				intel_uncore_read64(uncore,
1203 						    FENCE_REG_GEN6_LO(i));
1204 	} else if (GRAPHICS_VER(uncore->i915) >= 4) {
1205 		for (i = 0; i < ggtt->num_fences; i++)
1206 			gt->fence[i] =
1207 				intel_uncore_read64(uncore,
1208 						    FENCE_REG_965_LO(i));
1209 	} else {
1210 		for (i = 0; i < ggtt->num_fences; i++)
1211 			gt->fence[i] =
1212 				intel_uncore_read(uncore, FENCE_REG(i));
1213 	}
1214 	gt->nfence = i;
1215 }
1216 
1217 static void engine_record_registers(struct intel_engine_coredump *ee)
1218 {
1219 	const struct intel_engine_cs *engine = ee->engine;
1220 	struct drm_i915_private *i915 = engine->i915;
1221 
1222 	if (GRAPHICS_VER(i915) >= 6) {
1223 		ee->rc_psmi = ENGINE_READ(engine, RING_PSMI_CTL);
1224 
1225 		if (GRAPHICS_VER(i915) >= 12)
1226 			ee->fault_reg = intel_uncore_read(engine->uncore,
1227 							  GEN12_RING_FAULT_REG);
1228 		else if (GRAPHICS_VER(i915) >= 8)
1229 			ee->fault_reg = intel_uncore_read(engine->uncore,
1230 							  GEN8_RING_FAULT_REG);
1231 		else
1232 			ee->fault_reg = GEN6_RING_FAULT_REG_READ(engine);
1233 	}
1234 
1235 	if (GRAPHICS_VER(i915) >= 4) {
1236 		ee->esr = ENGINE_READ(engine, RING_ESR);
1237 		ee->faddr = ENGINE_READ(engine, RING_DMA_FADD);
1238 		ee->ipeir = ENGINE_READ(engine, RING_IPEIR);
1239 		ee->ipehr = ENGINE_READ(engine, RING_IPEHR);
1240 		ee->instps = ENGINE_READ(engine, RING_INSTPS);
1241 		ee->bbaddr = ENGINE_READ(engine, RING_BBADDR);
1242 		ee->ccid = ENGINE_READ(engine, CCID);
1243 		if (GRAPHICS_VER(i915) >= 8) {
1244 			ee->faddr |= (u64)ENGINE_READ(engine, RING_DMA_FADD_UDW) << 32;
1245 			ee->bbaddr |= (u64)ENGINE_READ(engine, RING_BBADDR_UDW) << 32;
1246 		}
1247 		ee->bbstate = ENGINE_READ(engine, RING_BBSTATE);
1248 	} else {
1249 		ee->faddr = ENGINE_READ(engine, DMA_FADD_I8XX);
1250 		ee->ipeir = ENGINE_READ(engine, IPEIR);
1251 		ee->ipehr = ENGINE_READ(engine, IPEHR);
1252 	}
1253 
1254 	if (GRAPHICS_VER(i915) >= 11) {
1255 		ee->cmd_cctl = ENGINE_READ(engine, RING_CMD_CCTL);
1256 		ee->cscmdop = ENGINE_READ(engine, RING_CSCMDOP);
1257 		ee->ctx_sr_ctl = ENGINE_READ(engine, RING_CTX_SR_CTL);
1258 		ee->dma_faddr_hi = ENGINE_READ(engine, RING_DMA_FADD_UDW);
1259 		ee->dma_faddr_lo = ENGINE_READ(engine, RING_DMA_FADD);
1260 		ee->nopid = ENGINE_READ(engine, RING_NOPID);
1261 		ee->excc = ENGINE_READ(engine, RING_EXCC);
1262 	}
1263 
1264 	intel_engine_get_instdone(engine, &ee->instdone);
1265 
1266 	ee->instpm = ENGINE_READ(engine, RING_INSTPM);
1267 	ee->acthd = intel_engine_get_active_head(engine);
1268 	ee->start = ENGINE_READ(engine, RING_START);
1269 	ee->head = ENGINE_READ(engine, RING_HEAD);
1270 	ee->tail = ENGINE_READ(engine, RING_TAIL);
1271 	ee->ctl = ENGINE_READ(engine, RING_CTL);
1272 	if (GRAPHICS_VER(i915) > 2)
1273 		ee->mode = ENGINE_READ(engine, RING_MI_MODE);
1274 
1275 	if (!HWS_NEEDS_PHYSICAL(i915)) {
1276 		i915_reg_t mmio;
1277 
1278 		if (GRAPHICS_VER(i915) == 7) {
1279 			switch (engine->id) {
1280 			default:
1281 				MISSING_CASE(engine->id);
1282 				fallthrough;
1283 			case RCS0:
1284 				mmio = RENDER_HWS_PGA_GEN7;
1285 				break;
1286 			case BCS0:
1287 				mmio = BLT_HWS_PGA_GEN7;
1288 				break;
1289 			case VCS0:
1290 				mmio = BSD_HWS_PGA_GEN7;
1291 				break;
1292 			case VECS0:
1293 				mmio = VEBOX_HWS_PGA_GEN7;
1294 				break;
1295 			}
1296 		} else if (GRAPHICS_VER(engine->i915) == 6) {
1297 			mmio = RING_HWS_PGA_GEN6(engine->mmio_base);
1298 		} else {
1299 			/* XXX: gen8 returns to sanity */
1300 			mmio = RING_HWS_PGA(engine->mmio_base);
1301 		}
1302 
1303 		ee->hws = intel_uncore_read(engine->uncore, mmio);
1304 	}
1305 
1306 	ee->reset_count = i915_reset_engine_count(&i915->gpu_error, engine);
1307 
1308 	if (HAS_PPGTT(i915)) {
1309 		int i;
1310 
1311 		ee->vm_info.gfx_mode = ENGINE_READ(engine, RING_MODE_GEN7);
1312 
1313 		if (GRAPHICS_VER(i915) == 6) {
1314 			ee->vm_info.pp_dir_base =
1315 				ENGINE_READ(engine, RING_PP_DIR_BASE_READ);
1316 		} else if (GRAPHICS_VER(i915) == 7) {
1317 			ee->vm_info.pp_dir_base =
1318 				ENGINE_READ(engine, RING_PP_DIR_BASE);
1319 		} else if (GRAPHICS_VER(i915) >= 8) {
1320 			u32 base = engine->mmio_base;
1321 
1322 			for (i = 0; i < 4; i++) {
1323 				ee->vm_info.pdp[i] =
1324 					intel_uncore_read(engine->uncore,
1325 							  GEN8_RING_PDP_UDW(base, i));
1326 				ee->vm_info.pdp[i] <<= 32;
1327 				ee->vm_info.pdp[i] |=
1328 					intel_uncore_read(engine->uncore,
1329 							  GEN8_RING_PDP_LDW(base, i));
1330 			}
1331 		}
1332 	}
1333 }
1334 
1335 static void record_request(const struct i915_request *request,
1336 			   struct i915_request_coredump *erq)
1337 {
1338 	erq->flags = request->fence.flags;
1339 	erq->context = request->fence.context;
1340 	erq->seqno = request->fence.seqno;
1341 	erq->sched_attr = request->sched.attr;
1342 	erq->head = request->head;
1343 	erq->tail = request->tail;
1344 
1345 	erq->pid = 0;
1346 	rcu_read_lock();
1347 	if (!intel_context_is_closed(request->context)) {
1348 		const struct i915_gem_context *ctx;
1349 
1350 		ctx = rcu_dereference(request->context->gem_context);
1351 		if (ctx)
1352 			erq->pid = pid_nr(ctx->pid);
1353 	}
1354 	rcu_read_unlock();
1355 }
1356 
1357 static void engine_record_execlists(struct intel_engine_coredump *ee)
1358 {
1359 	const struct intel_engine_execlists * const el = &ee->engine->execlists;
1360 	struct i915_request * const *port = el->active;
1361 	unsigned int n = 0;
1362 
1363 	while (*port)
1364 		record_request(*port++, &ee->execlist[n++]);
1365 
1366 	ee->num_ports = n;
1367 }
1368 
1369 static bool record_context(struct i915_gem_context_coredump *e,
1370 			   const struct i915_request *rq)
1371 {
1372 	struct i915_gem_context *ctx;
1373 	struct task_struct *task;
1374 	bool simulated;
1375 
1376 	rcu_read_lock();
1377 	ctx = rcu_dereference(rq->context->gem_context);
1378 	if (ctx && !kref_get_unless_zero(&ctx->ref))
1379 		ctx = NULL;
1380 	rcu_read_unlock();
1381 	if (!ctx)
1382 		return true;
1383 
1384 	rcu_read_lock();
1385 	task = pid_task(ctx->pid, PIDTYPE_PID);
1386 	if (task) {
1387 		strcpy(e->comm, task->comm);
1388 		e->pid = task->pid;
1389 	}
1390 	rcu_read_unlock();
1391 
1392 	e->sched_attr = ctx->sched;
1393 	e->guilty = atomic_read(&ctx->guilty_count);
1394 	e->active = atomic_read(&ctx->active_count);
1395 
1396 	e->total_runtime = intel_context_get_total_runtime_ns(rq->context);
1397 	e->avg_runtime = intel_context_get_avg_runtime_ns(rq->context);
1398 
1399 	simulated = i915_gem_context_no_error_capture(ctx);
1400 
1401 	i915_gem_context_put(ctx);
1402 	return simulated;
1403 }
1404 
1405 struct intel_engine_capture_vma {
1406 	struct intel_engine_capture_vma *next;
1407 	struct i915_vma_resource *vma_res;
1408 	char name[16];
1409 	bool lockdep_cookie;
1410 };
1411 
1412 static struct intel_engine_capture_vma *
1413 capture_vma_snapshot(struct intel_engine_capture_vma *next,
1414 		     struct i915_vma_resource *vma_res,
1415 		     gfp_t gfp, const char *name)
1416 {
1417 	struct intel_engine_capture_vma *c;
1418 
1419 	if (!vma_res)
1420 		return next;
1421 
1422 	c = kmalloc(sizeof(*c), gfp);
1423 	if (!c)
1424 		return next;
1425 
1426 	if (!i915_vma_resource_hold(vma_res, &c->lockdep_cookie)) {
1427 		kfree(c);
1428 		return next;
1429 	}
1430 
1431 	strcpy(c->name, name);
1432 	c->vma_res = i915_vma_resource_get(vma_res);
1433 
1434 	c->next = next;
1435 	return c;
1436 }
1437 
1438 static struct intel_engine_capture_vma *
1439 capture_vma(struct intel_engine_capture_vma *next,
1440 	    struct i915_vma *vma,
1441 	    const char *name,
1442 	    gfp_t gfp)
1443 {
1444 	if (!vma)
1445 		return next;
1446 
1447 	/*
1448 	 * If the vma isn't pinned, then the vma should be snapshotted
1449 	 * to a struct i915_vma_snapshot at command submission time.
1450 	 * Not here.
1451 	 */
1452 	if (GEM_WARN_ON(!i915_vma_is_pinned(vma)))
1453 		return next;
1454 
1455 	next = capture_vma_snapshot(next, vma->resource, gfp, name);
1456 
1457 	return next;
1458 }
1459 
1460 static struct intel_engine_capture_vma *
1461 capture_user(struct intel_engine_capture_vma *capture,
1462 	     const struct i915_request *rq,
1463 	     gfp_t gfp)
1464 {
1465 	struct i915_capture_list *c;
1466 
1467 	for (c = rq->capture_list; c; c = c->next)
1468 		capture = capture_vma_snapshot(capture, c->vma_res, gfp,
1469 					       "user");
1470 
1471 	return capture;
1472 }
1473 
1474 static void add_vma(struct intel_engine_coredump *ee,
1475 		    struct i915_vma_coredump *vma)
1476 {
1477 	if (vma) {
1478 		vma->next = ee->vma;
1479 		ee->vma = vma;
1480 	}
1481 }
1482 
1483 static struct i915_vma_coredump *
1484 create_vma_coredump(const struct intel_gt *gt, struct i915_vma *vma,
1485 		    const char *name, struct i915_vma_compress *compress)
1486 {
1487 	struct i915_vma_coredump *ret = NULL;
1488 	struct i915_vma_resource *vma_res;
1489 	bool lockdep_cookie;
1490 
1491 	if (!vma)
1492 		return NULL;
1493 
1494 	vma_res = vma->resource;
1495 
1496 	if (i915_vma_resource_hold(vma_res, &lockdep_cookie)) {
1497 		ret = i915_vma_coredump_create(gt, vma_res, compress, name);
1498 		i915_vma_resource_unhold(vma_res, lockdep_cookie);
1499 	}
1500 
1501 	return ret;
1502 }
1503 
1504 static void add_vma_coredump(struct intel_engine_coredump *ee,
1505 			     const struct intel_gt *gt,
1506 			     struct i915_vma *vma,
1507 			     const char *name,
1508 			     struct i915_vma_compress *compress)
1509 {
1510 	add_vma(ee, create_vma_coredump(gt, vma, name, compress));
1511 }
1512 
1513 struct intel_engine_coredump *
1514 intel_engine_coredump_alloc(struct intel_engine_cs *engine, gfp_t gfp, u32 dump_flags)
1515 {
1516 	struct intel_engine_coredump *ee;
1517 
1518 	ee = kzalloc(sizeof(*ee), gfp);
1519 	if (!ee)
1520 		return NULL;
1521 
1522 	ee->engine = engine;
1523 
1524 	if (!(dump_flags & CORE_DUMP_FLAG_IS_GUC_CAPTURE)) {
1525 		engine_record_registers(ee);
1526 		engine_record_execlists(ee);
1527 	}
1528 
1529 	return ee;
1530 }
1531 
1532 struct intel_engine_capture_vma *
1533 intel_engine_coredump_add_request(struct intel_engine_coredump *ee,
1534 				  struct i915_request *rq,
1535 				  gfp_t gfp)
1536 {
1537 	struct intel_engine_capture_vma *vma = NULL;
1538 
1539 	ee->simulated |= record_context(&ee->context, rq);
1540 	if (ee->simulated)
1541 		return NULL;
1542 
1543 	/*
1544 	 * We need to copy these to an anonymous buffer
1545 	 * as the simplest method to avoid being overwritten
1546 	 * by userspace.
1547 	 */
1548 	vma = capture_vma_snapshot(vma, rq->batch_res, gfp, "batch");
1549 	vma = capture_user(vma, rq, gfp);
1550 	vma = capture_vma(vma, rq->ring->vma, "ring", gfp);
1551 	vma = capture_vma(vma, rq->context->state, "HW context", gfp);
1552 
1553 	ee->rq_head = rq->head;
1554 	ee->rq_post = rq->postfix;
1555 	ee->rq_tail = rq->tail;
1556 
1557 	return vma;
1558 }
1559 
1560 void
1561 intel_engine_coredump_add_vma(struct intel_engine_coredump *ee,
1562 			      struct intel_engine_capture_vma *capture,
1563 			      struct i915_vma_compress *compress)
1564 {
1565 	const struct intel_engine_cs *engine = ee->engine;
1566 
1567 	while (capture) {
1568 		struct intel_engine_capture_vma *this = capture;
1569 		struct i915_vma_resource *vma_res = this->vma_res;
1570 
1571 		add_vma(ee,
1572 			i915_vma_coredump_create(engine->gt, vma_res,
1573 						 compress, this->name));
1574 
1575 		i915_vma_resource_unhold(vma_res, this->lockdep_cookie);
1576 		i915_vma_resource_put(vma_res);
1577 
1578 		capture = this->next;
1579 		kfree(this);
1580 	}
1581 
1582 	add_vma_coredump(ee, engine->gt, engine->status_page.vma,
1583 			 "HW Status", compress);
1584 
1585 	add_vma_coredump(ee, engine->gt, engine->wa_ctx.vma,
1586 			 "WA context", compress);
1587 }
1588 
1589 static struct intel_engine_coredump *
1590 capture_engine(struct intel_engine_cs *engine,
1591 	       struct i915_vma_compress *compress,
1592 	       u32 dump_flags)
1593 {
1594 	struct intel_engine_capture_vma *capture = NULL;
1595 	struct intel_engine_coredump *ee;
1596 	struct intel_context *ce;
1597 	struct i915_request *rq = NULL;
1598 	unsigned long flags;
1599 
1600 	ee = intel_engine_coredump_alloc(engine, ALLOW_FAIL, dump_flags);
1601 	if (!ee)
1602 		return NULL;
1603 
1604 	ce = intel_engine_get_hung_context(engine);
1605 	if (ce) {
1606 		intel_engine_clear_hung_context(engine);
1607 		rq = intel_context_find_active_request(ce);
1608 		if (!rq || !i915_request_started(rq))
1609 			goto no_request_capture;
1610 	} else {
1611 		/*
1612 		 * Getting here with GuC enabled means it is a forced error capture
1613 		 * with no actual hang. So, no need to attempt the execlist search.
1614 		 */
1615 		if (!intel_uc_uses_guc_submission(&engine->gt->uc)) {
1616 			spin_lock_irqsave(&engine->sched_engine->lock, flags);
1617 			rq = intel_engine_execlist_find_hung_request(engine);
1618 			spin_unlock_irqrestore(&engine->sched_engine->lock,
1619 					       flags);
1620 		}
1621 	}
1622 	if (rq)
1623 		rq = i915_request_get_rcu(rq);
1624 
1625 	if (!rq)
1626 		goto no_request_capture;
1627 
1628 	capture = intel_engine_coredump_add_request(ee, rq, ATOMIC_MAYFAIL);
1629 	if (!capture) {
1630 		i915_request_put(rq);
1631 		goto no_request_capture;
1632 	}
1633 	if (dump_flags & CORE_DUMP_FLAG_IS_GUC_CAPTURE)
1634 		intel_guc_capture_get_matching_node(engine->gt, ee, ce);
1635 
1636 	intel_engine_coredump_add_vma(ee, capture, compress);
1637 	i915_request_put(rq);
1638 
1639 	return ee;
1640 
1641 no_request_capture:
1642 	kfree(ee);
1643 	return NULL;
1644 }
1645 
1646 static void
1647 gt_record_engines(struct intel_gt_coredump *gt,
1648 		  intel_engine_mask_t engine_mask,
1649 		  struct i915_vma_compress *compress,
1650 		  u32 dump_flags)
1651 {
1652 	struct intel_engine_cs *engine;
1653 	enum intel_engine_id id;
1654 
1655 	for_each_engine(engine, gt->_gt, id) {
1656 		struct intel_engine_coredump *ee;
1657 
1658 		/* Refill our page pool before entering atomic section */
1659 		pool_refill(&compress->pool, ALLOW_FAIL);
1660 
1661 		ee = capture_engine(engine, compress, dump_flags);
1662 		if (!ee)
1663 			continue;
1664 
1665 		ee->hung = engine->mask & engine_mask;
1666 
1667 		gt->simulated |= ee->simulated;
1668 		if (ee->simulated) {
1669 			if (dump_flags & CORE_DUMP_FLAG_IS_GUC_CAPTURE)
1670 				intel_guc_capture_free_node(ee);
1671 			kfree(ee);
1672 			continue;
1673 		}
1674 
1675 		ee->next = gt->engine;
1676 		gt->engine = ee;
1677 	}
1678 }
1679 
1680 static void gt_record_guc_ctb(struct intel_ctb_coredump *saved,
1681 			      const struct intel_guc_ct_buffer *ctb,
1682 			      const void *blob_ptr, struct intel_guc *guc)
1683 {
1684 	if (!ctb || !ctb->desc)
1685 		return;
1686 
1687 	saved->raw_status = ctb->desc->status;
1688 	saved->raw_head = ctb->desc->head;
1689 	saved->raw_tail = ctb->desc->tail;
1690 	saved->head = ctb->head;
1691 	saved->tail = ctb->tail;
1692 	saved->size = ctb->size;
1693 	saved->desc_offset = ((void *)ctb->desc) - blob_ptr;
1694 	saved->cmds_offset = ((void *)ctb->cmds) - blob_ptr;
1695 }
1696 
1697 static struct intel_uc_coredump *
1698 gt_record_uc(struct intel_gt_coredump *gt,
1699 	     struct i915_vma_compress *compress)
1700 {
1701 	const struct intel_uc *uc = &gt->_gt->uc;
1702 	struct intel_uc_coredump *error_uc;
1703 
1704 	error_uc = kzalloc(sizeof(*error_uc), ALLOW_FAIL);
1705 	if (!error_uc)
1706 		return NULL;
1707 
1708 	memcpy(&error_uc->guc_fw, &uc->guc.fw, sizeof(uc->guc.fw));
1709 	memcpy(&error_uc->huc_fw, &uc->huc.fw, sizeof(uc->huc.fw));
1710 
1711 	error_uc->guc_fw.file_selected.path = kstrdup(uc->guc.fw.file_selected.path, ALLOW_FAIL);
1712 	error_uc->huc_fw.file_selected.path = kstrdup(uc->huc.fw.file_selected.path, ALLOW_FAIL);
1713 	error_uc->guc_fw.file_wanted.path = kstrdup(uc->guc.fw.file_wanted.path, ALLOW_FAIL);
1714 	error_uc->huc_fw.file_wanted.path = kstrdup(uc->huc.fw.file_wanted.path, ALLOW_FAIL);
1715 
1716 	/*
1717 	 * Save the GuC log and include a timestamp reference for converting the
1718 	 * log times to system times (in conjunction with the error->boottime and
1719 	 * gt->clock_frequency fields saved elsewhere).
1720 	 */
1721 	error_uc->guc.timestamp = intel_uncore_read(gt->_gt->uncore, GUCPMTIMESTAMP);
1722 	error_uc->guc.vma_log = create_vma_coredump(gt->_gt, uc->guc.log.vma,
1723 						    "GuC log buffer", compress);
1724 	error_uc->guc.vma_ctb = create_vma_coredump(gt->_gt, uc->guc.ct.vma,
1725 						    "GuC CT buffer", compress);
1726 	error_uc->guc.last_fence = uc->guc.ct.requests.last_fence;
1727 	gt_record_guc_ctb(error_uc->guc.ctb + 0, &uc->guc.ct.ctbs.send,
1728 			  uc->guc.ct.ctbs.send.desc, (struct intel_guc *)&uc->guc);
1729 	gt_record_guc_ctb(error_uc->guc.ctb + 1, &uc->guc.ct.ctbs.recv,
1730 			  uc->guc.ct.ctbs.send.desc, (struct intel_guc *)&uc->guc);
1731 
1732 	return error_uc;
1733 }
1734 
1735 /* Capture display registers. */
1736 static void gt_record_display_regs(struct intel_gt_coredump *gt)
1737 {
1738 	struct intel_uncore *uncore = gt->_gt->uncore;
1739 	struct drm_i915_private *i915 = uncore->i915;
1740 
1741 	if (GRAPHICS_VER(i915) >= 6)
1742 		gt->derrmr = intel_uncore_read(uncore, DERRMR);
1743 
1744 	if (GRAPHICS_VER(i915) >= 8)
1745 		gt->ier = intel_uncore_read(uncore, GEN8_DE_MISC_IER);
1746 	else if (IS_VALLEYVIEW(i915))
1747 		gt->ier = intel_uncore_read(uncore, VLV_IER);
1748 	else if (HAS_PCH_SPLIT(i915))
1749 		gt->ier = intel_uncore_read(uncore, DEIER);
1750 	else if (GRAPHICS_VER(i915) == 2)
1751 		gt->ier = intel_uncore_read16(uncore, GEN2_IER);
1752 	else
1753 		gt->ier = intel_uncore_read(uncore, GEN2_IER);
1754 }
1755 
1756 /* Capture all other registers that GuC doesn't capture. */
1757 static void gt_record_global_nonguc_regs(struct intel_gt_coredump *gt)
1758 {
1759 	struct intel_uncore *uncore = gt->_gt->uncore;
1760 	struct drm_i915_private *i915 = uncore->i915;
1761 	int i;
1762 
1763 	if (IS_VALLEYVIEW(i915)) {
1764 		gt->gtier[0] = intel_uncore_read(uncore, GTIER);
1765 		gt->ngtier = 1;
1766 	} else if (GRAPHICS_VER(i915) >= 11) {
1767 		gt->gtier[0] =
1768 			intel_uncore_read(uncore,
1769 					  GEN11_RENDER_COPY_INTR_ENABLE);
1770 		gt->gtier[1] =
1771 			intel_uncore_read(uncore, GEN11_VCS_VECS_INTR_ENABLE);
1772 		gt->gtier[2] =
1773 			intel_uncore_read(uncore, GEN11_GUC_SG_INTR_ENABLE);
1774 		gt->gtier[3] =
1775 			intel_uncore_read(uncore,
1776 					  GEN11_GPM_WGBOXPERF_INTR_ENABLE);
1777 		gt->gtier[4] =
1778 			intel_uncore_read(uncore,
1779 					  GEN11_CRYPTO_RSVD_INTR_ENABLE);
1780 		gt->gtier[5] =
1781 			intel_uncore_read(uncore,
1782 					  GEN11_GUNIT_CSME_INTR_ENABLE);
1783 		gt->ngtier = 6;
1784 	} else if (GRAPHICS_VER(i915) >= 8) {
1785 		for (i = 0; i < 4; i++)
1786 			gt->gtier[i] =
1787 				intel_uncore_read(uncore, GEN8_GT_IER(i));
1788 		gt->ngtier = 4;
1789 	} else if (HAS_PCH_SPLIT(i915)) {
1790 		gt->gtier[0] = intel_uncore_read(uncore, GTIER);
1791 		gt->ngtier = 1;
1792 	}
1793 
1794 	gt->eir = intel_uncore_read(uncore, EIR);
1795 	gt->pgtbl_er = intel_uncore_read(uncore, PGTBL_ER);
1796 }
1797 
1798 /*
1799  * Capture all registers that relate to workload submission.
1800  * NOTE: In GuC submission, when GuC resets an engine, it can dump these for us
1801  */
1802 static void gt_record_global_regs(struct intel_gt_coredump *gt)
1803 {
1804 	struct intel_uncore *uncore = gt->_gt->uncore;
1805 	struct drm_i915_private *i915 = uncore->i915;
1806 	int i;
1807 
1808 	/*
1809 	 * General organization
1810 	 * 1. Registers specific to a single generation
1811 	 * 2. Registers which belong to multiple generations
1812 	 * 3. Feature specific registers.
1813 	 * 4. Everything else
1814 	 * Please try to follow the order.
1815 	 */
1816 
1817 	/* 1: Registers specific to a single generation */
1818 	if (IS_VALLEYVIEW(i915))
1819 		gt->forcewake = intel_uncore_read_fw(uncore, FORCEWAKE_VLV);
1820 
1821 	if (GRAPHICS_VER(i915) == 7)
1822 		gt->err_int = intel_uncore_read(uncore, GEN7_ERR_INT);
1823 
1824 	if (GRAPHICS_VER(i915) >= 12) {
1825 		gt->fault_data0 = intel_uncore_read(uncore,
1826 						    GEN12_FAULT_TLB_DATA0);
1827 		gt->fault_data1 = intel_uncore_read(uncore,
1828 						    GEN12_FAULT_TLB_DATA1);
1829 	} else if (GRAPHICS_VER(i915) >= 8) {
1830 		gt->fault_data0 = intel_uncore_read(uncore,
1831 						    GEN8_FAULT_TLB_DATA0);
1832 		gt->fault_data1 = intel_uncore_read(uncore,
1833 						    GEN8_FAULT_TLB_DATA1);
1834 	}
1835 
1836 	if (GRAPHICS_VER(i915) == 6) {
1837 		gt->forcewake = intel_uncore_read_fw(uncore, FORCEWAKE);
1838 		gt->gab_ctl = intel_uncore_read(uncore, GAB_CTL);
1839 		gt->gfx_mode = intel_uncore_read(uncore, GFX_MODE);
1840 	}
1841 
1842 	/* 2: Registers which belong to multiple generations */
1843 	if (GRAPHICS_VER(i915) >= 7)
1844 		gt->forcewake = intel_uncore_read_fw(uncore, FORCEWAKE_MT);
1845 
1846 	if (GRAPHICS_VER(i915) >= 6) {
1847 		if (GRAPHICS_VER(i915) < 12) {
1848 			gt->error = intel_uncore_read(uncore, ERROR_GEN6);
1849 			gt->done_reg = intel_uncore_read(uncore, DONE_REG);
1850 		}
1851 	}
1852 
1853 	/* 3: Feature specific registers */
1854 	if (IS_GRAPHICS_VER(i915, 6, 7)) {
1855 		gt->gam_ecochk = intel_uncore_read(uncore, GAM_ECOCHK);
1856 		gt->gac_eco = intel_uncore_read(uncore, GAC_ECO_BITS);
1857 	}
1858 
1859 	if (IS_GRAPHICS_VER(i915, 8, 11))
1860 		gt->gtt_cache = intel_uncore_read(uncore, HSW_GTT_CACHE_EN);
1861 
1862 	if (GRAPHICS_VER(i915) == 12)
1863 		gt->aux_err = intel_uncore_read(uncore, GEN12_AUX_ERR_DBG);
1864 
1865 	if (GRAPHICS_VER(i915) >= 12) {
1866 		for (i = 0; i < I915_MAX_SFC; i++) {
1867 			/*
1868 			 * SFC_DONE resides in the VD forcewake domain, so it
1869 			 * only exists if the corresponding VCS engine is
1870 			 * present.
1871 			 */
1872 			if ((gt->_gt->info.sfc_mask & BIT(i)) == 0 ||
1873 			    !HAS_ENGINE(gt->_gt, _VCS(i * 2)))
1874 				continue;
1875 
1876 			gt->sfc_done[i] =
1877 				intel_uncore_read(uncore, GEN12_SFC_DONE(i));
1878 		}
1879 
1880 		gt->gam_done = intel_uncore_read(uncore, GEN12_GAM_DONE);
1881 	}
1882 }
1883 
1884 static void gt_record_info(struct intel_gt_coredump *gt)
1885 {
1886 	memcpy(&gt->info, &gt->_gt->info, sizeof(struct intel_gt_info));
1887 	gt->clock_frequency = gt->_gt->clock_frequency;
1888 	gt->clock_period_ns = gt->_gt->clock_period_ns;
1889 }
1890 
1891 /*
1892  * Generate a semi-unique error code. The code is not meant to have meaning, The
1893  * code's only purpose is to try to prevent false duplicated bug reports by
1894  * grossly estimating a GPU error state.
1895  *
1896  * TODO Ideally, hashing the batchbuffer would be a very nice way to determine
1897  * the hang if we could strip the GTT offset information from it.
1898  *
1899  * It's only a small step better than a random number in its current form.
1900  */
1901 static u32 generate_ecode(const struct intel_engine_coredump *ee)
1902 {
1903 	/*
1904 	 * IPEHR would be an ideal way to detect errors, as it's the gross
1905 	 * measure of "the command that hung." However, has some very common
1906 	 * synchronization commands which almost always appear in the case
1907 	 * strictly a client bug. Use instdone to differentiate those some.
1908 	 */
1909 	return ee ? ee->ipehr ^ ee->instdone.instdone : 0;
1910 }
1911 
1912 static const char *error_msg(struct i915_gpu_coredump *error)
1913 {
1914 	struct intel_engine_coredump *first = NULL;
1915 	unsigned int hung_classes = 0;
1916 	struct intel_gt_coredump *gt;
1917 	int len;
1918 
1919 	for (gt = error->gt; gt; gt = gt->next) {
1920 		struct intel_engine_coredump *cs;
1921 
1922 		for (cs = gt->engine; cs; cs = cs->next) {
1923 			if (cs->hung) {
1924 				hung_classes |= BIT(cs->engine->uabi_class);
1925 				if (!first)
1926 					first = cs;
1927 			}
1928 		}
1929 	}
1930 
1931 	len = scnprintf(error->error_msg, sizeof(error->error_msg),
1932 			"GPU HANG: ecode %d:%x:%08x",
1933 			GRAPHICS_VER(error->i915), hung_classes,
1934 			generate_ecode(first));
1935 	if (first && first->context.pid) {
1936 		/* Just show the first executing process, more is confusing */
1937 		len += scnprintf(error->error_msg + len,
1938 				 sizeof(error->error_msg) - len,
1939 				 ", in %s [%d]",
1940 				 first->context.comm, first->context.pid);
1941 	}
1942 
1943 	return error->error_msg;
1944 }
1945 
1946 static void capture_gen(struct i915_gpu_coredump *error)
1947 {
1948 	struct drm_i915_private *i915 = error->i915;
1949 
1950 	error->wakelock = atomic_read(&i915->runtime_pm.wakeref_count);
1951 	error->suspended = i915->runtime_pm.suspended;
1952 
1953 	error->iommu = i915_vtd_active(i915);
1954 	error->reset_count = i915_reset_count(&i915->gpu_error);
1955 	error->suspend_count = i915->suspend_count;
1956 
1957 	i915_params_copy(&error->params, &i915->params);
1958 	memcpy(&error->device_info,
1959 	       INTEL_INFO(i915),
1960 	       sizeof(error->device_info));
1961 	memcpy(&error->runtime_info,
1962 	       RUNTIME_INFO(i915),
1963 	       sizeof(error->runtime_info));
1964 	error->driver_caps = i915->caps;
1965 }
1966 
1967 struct i915_gpu_coredump *
1968 i915_gpu_coredump_alloc(struct drm_i915_private *i915, gfp_t gfp)
1969 {
1970 	struct i915_gpu_coredump *error;
1971 
1972 	if (!i915->params.error_capture)
1973 		return NULL;
1974 
1975 	error = kzalloc(sizeof(*error), gfp);
1976 	if (!error)
1977 		return NULL;
1978 
1979 	kref_init(&error->ref);
1980 	error->i915 = i915;
1981 
1982 	error->time = ktime_get_real();
1983 	error->boottime = ktime_get_boottime();
1984 	error->uptime = ktime_sub(ktime_get(), to_gt(i915)->last_init_time);
1985 	error->capture = jiffies;
1986 
1987 	capture_gen(error);
1988 
1989 	return error;
1990 }
1991 
1992 #define DAY_AS_SECONDS(x) (24 * 60 * 60 * (x))
1993 
1994 struct intel_gt_coredump *
1995 intel_gt_coredump_alloc(struct intel_gt *gt, gfp_t gfp, u32 dump_flags)
1996 {
1997 	struct intel_gt_coredump *gc;
1998 
1999 	gc = kzalloc(sizeof(*gc), gfp);
2000 	if (!gc)
2001 		return NULL;
2002 
2003 	gc->_gt = gt;
2004 	gc->awake = intel_gt_pm_is_awake(gt);
2005 
2006 	gt_record_display_regs(gc);
2007 	gt_record_global_nonguc_regs(gc);
2008 
2009 	/*
2010 	 * GuC dumps global, eng-class and eng-instance registers
2011 	 * (that can change as part of engine state during execution)
2012 	 * before an engine is reset due to a hung context.
2013 	 * GuC captures and reports all three groups of registers
2014 	 * together as a single set before the engine is reset.
2015 	 * Thus, if GuC triggered the context reset we retrieve
2016 	 * the register values as part of gt_record_engines.
2017 	 */
2018 	if (!(dump_flags & CORE_DUMP_FLAG_IS_GUC_CAPTURE))
2019 		gt_record_global_regs(gc);
2020 
2021 	gt_record_fences(gc);
2022 
2023 	return gc;
2024 }
2025 
2026 struct i915_vma_compress *
2027 i915_vma_capture_prepare(struct intel_gt_coredump *gt)
2028 {
2029 	struct i915_vma_compress *compress;
2030 
2031 	compress = kmalloc(sizeof(*compress), ALLOW_FAIL);
2032 	if (!compress)
2033 		return NULL;
2034 
2035 	if (!compress_init(compress)) {
2036 		kfree(compress);
2037 		return NULL;
2038 	}
2039 
2040 	return compress;
2041 }
2042 
2043 void i915_vma_capture_finish(struct intel_gt_coredump *gt,
2044 			     struct i915_vma_compress *compress)
2045 {
2046 	if (!compress)
2047 		return;
2048 
2049 	compress_fini(compress);
2050 	kfree(compress);
2051 }
2052 
2053 static struct i915_gpu_coredump *
2054 __i915_gpu_coredump(struct intel_gt *gt, intel_engine_mask_t engine_mask, u32 dump_flags)
2055 {
2056 	struct drm_i915_private *i915 = gt->i915;
2057 	struct i915_gpu_coredump *error;
2058 
2059 	/* Check if GPU capture has been disabled */
2060 	error = READ_ONCE(i915->gpu_error.first_error);
2061 	if (IS_ERR(error))
2062 		return error;
2063 
2064 	error = i915_gpu_coredump_alloc(i915, ALLOW_FAIL);
2065 	if (!error)
2066 		return ERR_PTR(-ENOMEM);
2067 
2068 	error->gt = intel_gt_coredump_alloc(gt, ALLOW_FAIL, dump_flags);
2069 	if (error->gt) {
2070 		struct i915_vma_compress *compress;
2071 
2072 		compress = i915_vma_capture_prepare(error->gt);
2073 		if (!compress) {
2074 			kfree(error->gt);
2075 			kfree(error);
2076 			return ERR_PTR(-ENOMEM);
2077 		}
2078 
2079 		if (INTEL_INFO(i915)->has_gt_uc) {
2080 			error->gt->uc = gt_record_uc(error->gt, compress);
2081 			if (error->gt->uc) {
2082 				if (dump_flags & CORE_DUMP_FLAG_IS_GUC_CAPTURE)
2083 					error->gt->uc->guc.is_guc_capture = true;
2084 				else
2085 					GEM_BUG_ON(error->gt->uc->guc.is_guc_capture);
2086 			}
2087 		}
2088 
2089 		gt_record_info(error->gt);
2090 		gt_record_engines(error->gt, engine_mask, compress, dump_flags);
2091 
2092 
2093 		i915_vma_capture_finish(error->gt, compress);
2094 
2095 		error->simulated |= error->gt->simulated;
2096 	}
2097 
2098 	error->overlay = intel_overlay_capture_error_state(i915);
2099 
2100 	return error;
2101 }
2102 
2103 struct i915_gpu_coredump *
2104 i915_gpu_coredump(struct intel_gt *gt, intel_engine_mask_t engine_mask, u32 dump_flags)
2105 {
2106 	static DEFINE_MUTEX(capture_mutex);
2107 	int ret = mutex_lock_interruptible(&capture_mutex);
2108 	struct i915_gpu_coredump *dump;
2109 
2110 	if (ret)
2111 		return ERR_PTR(ret);
2112 
2113 	dump = __i915_gpu_coredump(gt, engine_mask, dump_flags);
2114 	mutex_unlock(&capture_mutex);
2115 
2116 	return dump;
2117 }
2118 
2119 void i915_error_state_store(struct i915_gpu_coredump *error)
2120 {
2121 	struct drm_i915_private *i915;
2122 	static bool warned;
2123 
2124 	if (IS_ERR_OR_NULL(error))
2125 		return;
2126 
2127 	i915 = error->i915;
2128 	drm_info(&i915->drm, "%s\n", error_msg(error));
2129 
2130 	if (error->simulated ||
2131 	    cmpxchg(&i915->gpu_error.first_error, NULL, error))
2132 		return;
2133 
2134 	i915_gpu_coredump_get(error);
2135 
2136 	if (!xchg(&warned, true) &&
2137 	    ktime_get_real_seconds() - DRIVER_TIMESTAMP < DAY_AS_SECONDS(180)) {
2138 		pr_info("GPU hangs can indicate a bug anywhere in the entire gfx stack, including userspace.\n");
2139 		pr_info("Please file a _new_ bug report at https://gitlab.freedesktop.org/drm/intel/issues/new.\n");
2140 		pr_info("Please see https://gitlab.freedesktop.org/drm/intel/-/wikis/How-to-file-i915-bugs for details.\n");
2141 		pr_info("drm/i915 developers can then reassign to the right component if it's not a kernel issue.\n");
2142 		pr_info("The GPU crash dump is required to analyze GPU hangs, so please always attach it.\n");
2143 		pr_info("GPU crash dump saved to /sys/class/drm/card%d/error\n",
2144 			i915->drm.primary->index);
2145 	}
2146 }
2147 
2148 /**
2149  * i915_capture_error_state - capture an error record for later analysis
2150  * @gt: intel_gt which originated the hang
2151  * @engine_mask: hung engines
2152  *
2153  *
2154  * Should be called when an error is detected (either a hang or an error
2155  * interrupt) to capture error state from the time of the error.  Fills
2156  * out a structure which becomes available in debugfs for user level tools
2157  * to pick up.
2158  */
2159 void i915_capture_error_state(struct intel_gt *gt,
2160 			      intel_engine_mask_t engine_mask, u32 dump_flags)
2161 {
2162 	struct i915_gpu_coredump *error;
2163 
2164 	error = i915_gpu_coredump(gt, engine_mask, dump_flags);
2165 	if (IS_ERR(error)) {
2166 		cmpxchg(&gt->i915->gpu_error.first_error, NULL, error);
2167 		return;
2168 	}
2169 
2170 	i915_error_state_store(error);
2171 	i915_gpu_coredump_put(error);
2172 }
2173 
2174 struct i915_gpu_coredump *
2175 i915_first_error_state(struct drm_i915_private *i915)
2176 {
2177 	struct i915_gpu_coredump *error;
2178 
2179 	spin_lock_irq(&i915->gpu_error.lock);
2180 	error = i915->gpu_error.first_error;
2181 	if (!IS_ERR_OR_NULL(error))
2182 		i915_gpu_coredump_get(error);
2183 	spin_unlock_irq(&i915->gpu_error.lock);
2184 
2185 	return error;
2186 }
2187 
2188 void i915_reset_error_state(struct drm_i915_private *i915)
2189 {
2190 	struct i915_gpu_coredump *error;
2191 
2192 	spin_lock_irq(&i915->gpu_error.lock);
2193 	error = i915->gpu_error.first_error;
2194 	if (error != ERR_PTR(-ENODEV)) /* if disabled, always disabled */
2195 		i915->gpu_error.first_error = NULL;
2196 	spin_unlock_irq(&i915->gpu_error.lock);
2197 
2198 	if (!IS_ERR_OR_NULL(error))
2199 		i915_gpu_coredump_put(error);
2200 }
2201 
2202 void i915_disable_error_state(struct drm_i915_private *i915, int err)
2203 {
2204 	spin_lock_irq(&i915->gpu_error.lock);
2205 	if (!i915->gpu_error.first_error)
2206 		i915->gpu_error.first_error = ERR_PTR(err);
2207 	spin_unlock_irq(&i915->gpu_error.lock);
2208 }
2209