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_print.h>
38 
39 #include "display/intel_atomic.h"
40 #include "display/intel_csr.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_gt.h"
46 #include "gt/intel_gt_pm.h"
47 
48 #include "i915_drv.h"
49 #include "i915_gpu_error.h"
50 #include "i915_memcpy.h"
51 #include "i915_scatterlist.h"
52 
53 #define ALLOW_FAIL (GFP_KERNEL | __GFP_RETRY_MAYFAIL | __GFP_NOWARN)
54 #define ATOMIC_MAYFAIL (GFP_ATOMIC | __GFP_NOWARN)
55 
56 static void __sg_set_buf(struct scatterlist *sg,
57 			 void *addr, unsigned int len, loff_t it)
58 {
59 	sg->page_link = (unsigned long)virt_to_page(addr);
60 	sg->offset = offset_in_page(addr);
61 	sg->length = len;
62 	sg->dma_address = it;
63 }
64 
65 static bool __i915_error_grow(struct drm_i915_error_state_buf *e, size_t len)
66 {
67 	if (!len)
68 		return false;
69 
70 	if (e->bytes + len + 1 <= e->size)
71 		return true;
72 
73 	if (e->bytes) {
74 		__sg_set_buf(e->cur++, e->buf, e->bytes, e->iter);
75 		e->iter += e->bytes;
76 		e->buf = NULL;
77 		e->bytes = 0;
78 	}
79 
80 	if (e->cur == e->end) {
81 		struct scatterlist *sgl;
82 
83 		sgl = (typeof(sgl))__get_free_page(ALLOW_FAIL);
84 		if (!sgl) {
85 			e->err = -ENOMEM;
86 			return false;
87 		}
88 
89 		if (e->cur) {
90 			e->cur->offset = 0;
91 			e->cur->length = 0;
92 			e->cur->page_link =
93 				(unsigned long)sgl | SG_CHAIN;
94 		} else {
95 			e->sgl = sgl;
96 		}
97 
98 		e->cur = sgl;
99 		e->end = sgl + SG_MAX_SINGLE_ALLOC - 1;
100 	}
101 
102 	e->size = ALIGN(len + 1, SZ_64K);
103 	e->buf = kmalloc(e->size, ALLOW_FAIL);
104 	if (!e->buf) {
105 		e->size = PAGE_ALIGN(len + 1);
106 		e->buf = kmalloc(e->size, GFP_KERNEL);
107 	}
108 	if (!e->buf) {
109 		e->err = -ENOMEM;
110 		return false;
111 	}
112 
113 	return true;
114 }
115 
116 __printf(2, 0)
117 static void i915_error_vprintf(struct drm_i915_error_state_buf *e,
118 			       const char *fmt, va_list args)
119 {
120 	va_list ap;
121 	int len;
122 
123 	if (e->err)
124 		return;
125 
126 	va_copy(ap, args);
127 	len = vsnprintf(NULL, 0, fmt, ap);
128 	va_end(ap);
129 	if (len <= 0) {
130 		e->err = len;
131 		return;
132 	}
133 
134 	if (!__i915_error_grow(e, len))
135 		return;
136 
137 	GEM_BUG_ON(e->bytes >= e->size);
138 	len = vscnprintf(e->buf + e->bytes, e->size - e->bytes, fmt, args);
139 	if (len < 0) {
140 		e->err = len;
141 		return;
142 	}
143 	e->bytes += len;
144 }
145 
146 static void i915_error_puts(struct drm_i915_error_state_buf *e, const char *str)
147 {
148 	unsigned len;
149 
150 	if (e->err || !str)
151 		return;
152 
153 	len = strlen(str);
154 	if (!__i915_error_grow(e, len))
155 		return;
156 
157 	GEM_BUG_ON(e->bytes + len > e->size);
158 	memcpy(e->buf + e->bytes, str, len);
159 	e->bytes += len;
160 }
161 
162 #define err_printf(e, ...) i915_error_printf(e, __VA_ARGS__)
163 #define err_puts(e, s) i915_error_puts(e, s)
164 
165 static void __i915_printfn_error(struct drm_printer *p, struct va_format *vaf)
166 {
167 	i915_error_vprintf(p->arg, vaf->fmt, *vaf->va);
168 }
169 
170 static inline struct drm_printer
171 i915_error_printer(struct drm_i915_error_state_buf *e)
172 {
173 	struct drm_printer p = {
174 		.printfn = __i915_printfn_error,
175 		.arg = e,
176 	};
177 	return p;
178 }
179 
180 /* single threaded page allocator with a reserved stash for emergencies */
181 static void pool_fini(struct pagevec *pv)
182 {
183 	pagevec_release(pv);
184 }
185 
186 static int pool_refill(struct pagevec *pv, gfp_t gfp)
187 {
188 	while (pagevec_space(pv)) {
189 		struct page *p;
190 
191 		p = alloc_page(gfp);
192 		if (!p)
193 			return -ENOMEM;
194 
195 		pagevec_add(pv, p);
196 	}
197 
198 	return 0;
199 }
200 
201 static int pool_init(struct pagevec *pv, gfp_t gfp)
202 {
203 	int err;
204 
205 	pagevec_init(pv);
206 
207 	err = pool_refill(pv, gfp);
208 	if (err)
209 		pool_fini(pv);
210 
211 	return err;
212 }
213 
214 static void *pool_alloc(struct pagevec *pv, gfp_t gfp)
215 {
216 	struct page *p;
217 
218 	p = alloc_page(gfp);
219 	if (!p && pagevec_count(pv))
220 		p = pv->pages[--pv->nr];
221 
222 	return p ? page_address(p) : NULL;
223 }
224 
225 static void pool_free(struct pagevec *pv, void *addr)
226 {
227 	struct page *p = virt_to_page(addr);
228 
229 	if (pagevec_space(pv))
230 		pagevec_add(pv, p);
231 	else
232 		__free_page(p);
233 }
234 
235 #ifdef CONFIG_DRM_I915_COMPRESS_ERROR
236 
237 struct i915_vma_compress {
238 	struct pagevec pool;
239 	struct z_stream_s zstream;
240 	void *tmp;
241 };
242 
243 static bool compress_init(struct i915_vma_compress *c)
244 {
245 	struct z_stream_s *zstream = &c->zstream;
246 
247 	if (pool_init(&c->pool, ALLOW_FAIL))
248 		return false;
249 
250 	zstream->workspace =
251 		kmalloc(zlib_deflate_workspacesize(MAX_WBITS, MAX_MEM_LEVEL),
252 			ALLOW_FAIL);
253 	if (!zstream->workspace) {
254 		pool_fini(&c->pool);
255 		return false;
256 	}
257 
258 	c->tmp = NULL;
259 	if (i915_has_memcpy_from_wc())
260 		c->tmp = pool_alloc(&c->pool, ALLOW_FAIL);
261 
262 	return true;
263 }
264 
265 static bool compress_start(struct i915_vma_compress *c)
266 {
267 	struct z_stream_s *zstream = &c->zstream;
268 	void *workspace = zstream->workspace;
269 
270 	memset(zstream, 0, sizeof(*zstream));
271 	zstream->workspace = workspace;
272 
273 	return zlib_deflateInit(zstream, Z_DEFAULT_COMPRESSION) == Z_OK;
274 }
275 
276 static void *compress_next_page(struct i915_vma_compress *c,
277 				struct i915_vma_coredump *dst)
278 {
279 	void *page;
280 
281 	if (dst->page_count >= dst->num_pages)
282 		return ERR_PTR(-ENOSPC);
283 
284 	page = pool_alloc(&c->pool, ALLOW_FAIL);
285 	if (!page)
286 		return ERR_PTR(-ENOMEM);
287 
288 	return dst->pages[dst->page_count++] = page;
289 }
290 
291 static int compress_page(struct i915_vma_compress *c,
292 			 void *src,
293 			 struct i915_vma_coredump *dst,
294 			 bool wc)
295 {
296 	struct z_stream_s *zstream = &c->zstream;
297 
298 	zstream->next_in = src;
299 	if (wc && c->tmp && i915_memcpy_from_wc(c->tmp, src, PAGE_SIZE))
300 		zstream->next_in = c->tmp;
301 	zstream->avail_in = PAGE_SIZE;
302 
303 	do {
304 		if (zstream->avail_out == 0) {
305 			zstream->next_out = compress_next_page(c, dst);
306 			if (IS_ERR(zstream->next_out))
307 				return PTR_ERR(zstream->next_out);
308 
309 			zstream->avail_out = PAGE_SIZE;
310 		}
311 
312 		if (zlib_deflate(zstream, Z_NO_FLUSH) != Z_OK)
313 			return -EIO;
314 	} while (zstream->avail_in);
315 
316 	/* Fallback to uncompressed if we increase size? */
317 	if (0 && zstream->total_out > zstream->total_in)
318 		return -E2BIG;
319 
320 	return 0;
321 }
322 
323 static int compress_flush(struct i915_vma_compress *c,
324 			  struct i915_vma_coredump *dst)
325 {
326 	struct z_stream_s *zstream = &c->zstream;
327 
328 	do {
329 		switch (zlib_deflate(zstream, Z_FINISH)) {
330 		case Z_OK: /* more space requested */
331 			zstream->next_out = compress_next_page(c, dst);
332 			if (IS_ERR(zstream->next_out))
333 				return PTR_ERR(zstream->next_out);
334 
335 			zstream->avail_out = PAGE_SIZE;
336 			break;
337 
338 		case Z_STREAM_END:
339 			goto end;
340 
341 		default: /* any error */
342 			return -EIO;
343 		}
344 	} while (1);
345 
346 end:
347 	memset(zstream->next_out, 0, zstream->avail_out);
348 	dst->unused = zstream->avail_out;
349 	return 0;
350 }
351 
352 static void compress_finish(struct i915_vma_compress *c)
353 {
354 	zlib_deflateEnd(&c->zstream);
355 }
356 
357 static void compress_fini(struct i915_vma_compress *c)
358 {
359 	kfree(c->zstream.workspace);
360 	if (c->tmp)
361 		pool_free(&c->pool, c->tmp);
362 	pool_fini(&c->pool);
363 }
364 
365 static void err_compression_marker(struct drm_i915_error_state_buf *m)
366 {
367 	err_puts(m, ":");
368 }
369 
370 #else
371 
372 struct i915_vma_compress {
373 	struct pagevec pool;
374 };
375 
376 static bool compress_init(struct i915_vma_compress *c)
377 {
378 	return pool_init(&c->pool, ALLOW_FAIL) == 0;
379 }
380 
381 static bool compress_start(struct i915_vma_compress *c)
382 {
383 	return true;
384 }
385 
386 static int compress_page(struct i915_vma_compress *c,
387 			 void *src,
388 			 struct i915_vma_coredump *dst,
389 			 bool wc)
390 {
391 	void *ptr;
392 
393 	ptr = pool_alloc(&c->pool, ALLOW_FAIL);
394 	if (!ptr)
395 		return -ENOMEM;
396 
397 	if (!(wc && i915_memcpy_from_wc(ptr, src, PAGE_SIZE)))
398 		memcpy(ptr, src, PAGE_SIZE);
399 	dst->pages[dst->page_count++] = ptr;
400 
401 	return 0;
402 }
403 
404 static int compress_flush(struct i915_vma_compress *c,
405 			  struct i915_vma_coredump *dst)
406 {
407 	return 0;
408 }
409 
410 static void compress_finish(struct i915_vma_compress *c)
411 {
412 }
413 
414 static void compress_fini(struct i915_vma_compress *c)
415 {
416 	pool_fini(&c->pool);
417 }
418 
419 static void err_compression_marker(struct drm_i915_error_state_buf *m)
420 {
421 	err_puts(m, "~");
422 }
423 
424 #endif
425 
426 static void error_print_instdone(struct drm_i915_error_state_buf *m,
427 				 const struct intel_engine_coredump *ee)
428 {
429 	const struct sseu_dev_info *sseu = &ee->engine->gt->info.sseu;
430 	int slice;
431 	int subslice;
432 
433 	err_printf(m, "  INSTDONE: 0x%08x\n",
434 		   ee->instdone.instdone);
435 
436 	if (ee->engine->class != RENDER_CLASS || INTEL_GEN(m->i915) <= 3)
437 		return;
438 
439 	err_printf(m, "  SC_INSTDONE: 0x%08x\n",
440 		   ee->instdone.slice_common);
441 
442 	if (INTEL_GEN(m->i915) <= 6)
443 		return;
444 
445 	for_each_instdone_slice_subslice(m->i915, sseu, slice, subslice)
446 		err_printf(m, "  SAMPLER_INSTDONE[%d][%d]: 0x%08x\n",
447 			   slice, subslice,
448 			   ee->instdone.sampler[slice][subslice]);
449 
450 	for_each_instdone_slice_subslice(m->i915, sseu, slice, subslice)
451 		err_printf(m, "  ROW_INSTDONE[%d][%d]: 0x%08x\n",
452 			   slice, subslice,
453 			   ee->instdone.row[slice][subslice]);
454 
455 	if (INTEL_GEN(m->i915) < 12)
456 		return;
457 
458 	err_printf(m, "  SC_INSTDONE_EXTRA: 0x%08x\n",
459 		   ee->instdone.slice_common_extra[0]);
460 	err_printf(m, "  SC_INSTDONE_EXTRA2: 0x%08x\n",
461 		   ee->instdone.slice_common_extra[1]);
462 }
463 
464 static void error_print_request(struct drm_i915_error_state_buf *m,
465 				const char *prefix,
466 				const struct i915_request_coredump *erq)
467 {
468 	if (!erq->seqno)
469 		return;
470 
471 	err_printf(m, "%s pid %d, seqno %8x:%08x%s%s, prio %d, head %08x, tail %08x\n",
472 		   prefix, erq->pid, erq->context, erq->seqno,
473 		   test_bit(DMA_FENCE_FLAG_SIGNALED_BIT,
474 			    &erq->flags) ? "!" : "",
475 		   test_bit(DMA_FENCE_FLAG_ENABLE_SIGNAL_BIT,
476 			    &erq->flags) ? "+" : "",
477 		   erq->sched_attr.priority,
478 		   erq->head, erq->tail);
479 }
480 
481 static void error_print_context(struct drm_i915_error_state_buf *m,
482 				const char *header,
483 				const struct i915_gem_context_coredump *ctx)
484 {
485 	const u32 period = RUNTIME_INFO(m->i915)->cs_timestamp_period_ns;
486 
487 	err_printf(m, "%s%s[%d] prio %d, guilty %d active %d, runtime total %lluns, avg %lluns\n",
488 		   header, ctx->comm, ctx->pid, ctx->sched_attr.priority,
489 		   ctx->guilty, ctx->active,
490 		   ctx->total_runtime * period,
491 		   mul_u32_u32(ctx->avg_runtime, period));
492 }
493 
494 static struct i915_vma_coredump *
495 __find_vma(struct i915_vma_coredump *vma, const char *name)
496 {
497 	while (vma) {
498 		if (strcmp(vma->name, name) == 0)
499 			return vma;
500 		vma = vma->next;
501 	}
502 
503 	return NULL;
504 }
505 
506 static struct i915_vma_coredump *
507 find_batch(const struct intel_engine_coredump *ee)
508 {
509 	return __find_vma(ee->vma, "batch");
510 }
511 
512 static void error_print_engine(struct drm_i915_error_state_buf *m,
513 			       const struct intel_engine_coredump *ee)
514 {
515 	struct i915_vma_coredump *batch;
516 	int n;
517 
518 	err_printf(m, "%s command stream:\n", ee->engine->name);
519 	err_printf(m, "  CCID:  0x%08x\n", ee->ccid);
520 	err_printf(m, "  START: 0x%08x\n", ee->start);
521 	err_printf(m, "  HEAD:  0x%08x [0x%08x]\n", ee->head, ee->rq_head);
522 	err_printf(m, "  TAIL:  0x%08x [0x%08x, 0x%08x]\n",
523 		   ee->tail, ee->rq_post, ee->rq_tail);
524 	err_printf(m, "  CTL:   0x%08x\n", ee->ctl);
525 	err_printf(m, "  MODE:  0x%08x\n", ee->mode);
526 	err_printf(m, "  HWS:   0x%08x\n", ee->hws);
527 	err_printf(m, "  ACTHD: 0x%08x %08x\n",
528 		   (u32)(ee->acthd>>32), (u32)ee->acthd);
529 	err_printf(m, "  IPEIR: 0x%08x\n", ee->ipeir);
530 	err_printf(m, "  IPEHR: 0x%08x\n", ee->ipehr);
531 	err_printf(m, "  ESR:   0x%08x\n", ee->esr);
532 
533 	error_print_instdone(m, ee);
534 
535 	batch = find_batch(ee);
536 	if (batch) {
537 		u64 start = batch->gtt_offset;
538 		u64 end = start + batch->gtt_size;
539 
540 		err_printf(m, "  batch: [0x%08x_%08x, 0x%08x_%08x]\n",
541 			   upper_32_bits(start), lower_32_bits(start),
542 			   upper_32_bits(end), lower_32_bits(end));
543 	}
544 	if (INTEL_GEN(m->i915) >= 4) {
545 		err_printf(m, "  BBADDR: 0x%08x_%08x\n",
546 			   (u32)(ee->bbaddr>>32), (u32)ee->bbaddr);
547 		err_printf(m, "  BB_STATE: 0x%08x\n", ee->bbstate);
548 		err_printf(m, "  INSTPS: 0x%08x\n", ee->instps);
549 	}
550 	err_printf(m, "  INSTPM: 0x%08x\n", ee->instpm);
551 	err_printf(m, "  FADDR: 0x%08x %08x\n", upper_32_bits(ee->faddr),
552 		   lower_32_bits(ee->faddr));
553 	if (INTEL_GEN(m->i915) >= 6) {
554 		err_printf(m, "  RC PSMI: 0x%08x\n", ee->rc_psmi);
555 		err_printf(m, "  FAULT_REG: 0x%08x\n", ee->fault_reg);
556 	}
557 	if (HAS_PPGTT(m->i915)) {
558 		err_printf(m, "  GFX_MODE: 0x%08x\n", ee->vm_info.gfx_mode);
559 
560 		if (INTEL_GEN(m->i915) >= 8) {
561 			int i;
562 			for (i = 0; i < 4; i++)
563 				err_printf(m, "  PDP%d: 0x%016llx\n",
564 					   i, ee->vm_info.pdp[i]);
565 		} else {
566 			err_printf(m, "  PP_DIR_BASE: 0x%08x\n",
567 				   ee->vm_info.pp_dir_base);
568 		}
569 	}
570 	err_printf(m, "  engine reset count: %u\n", ee->reset_count);
571 
572 	for (n = 0; n < ee->num_ports; n++) {
573 		err_printf(m, "  ELSP[%d]:", n);
574 		error_print_request(m, " ", &ee->execlist[n]);
575 	}
576 
577 	error_print_context(m, "  Active context: ", &ee->context);
578 }
579 
580 void i915_error_printf(struct drm_i915_error_state_buf *e, const char *f, ...)
581 {
582 	va_list args;
583 
584 	va_start(args, f);
585 	i915_error_vprintf(e, f, args);
586 	va_end(args);
587 }
588 
589 static void print_error_vma(struct drm_i915_error_state_buf *m,
590 			    const struct intel_engine_cs *engine,
591 			    const struct i915_vma_coredump *vma)
592 {
593 	char out[ASCII85_BUFSZ];
594 	int page;
595 
596 	if (!vma)
597 		return;
598 
599 	err_printf(m, "%s --- %s = 0x%08x %08x\n",
600 		   engine ? engine->name : "global", vma->name,
601 		   upper_32_bits(vma->gtt_offset),
602 		   lower_32_bits(vma->gtt_offset));
603 
604 	if (vma->gtt_page_sizes > I915_GTT_PAGE_SIZE_4K)
605 		err_printf(m, "gtt_page_sizes = 0x%08x\n", vma->gtt_page_sizes);
606 
607 	err_compression_marker(m);
608 	for (page = 0; page < vma->page_count; page++) {
609 		int i, len;
610 
611 		len = PAGE_SIZE;
612 		if (page == vma->page_count - 1)
613 			len -= vma->unused;
614 		len = ascii85_encode_len(len);
615 
616 		for (i = 0; i < len; i++)
617 			err_puts(m, ascii85_encode(vma->pages[page][i], out));
618 	}
619 	err_puts(m, "\n");
620 }
621 
622 static void err_print_capabilities(struct drm_i915_error_state_buf *m,
623 				   struct i915_gpu_coredump *error)
624 {
625 	struct drm_printer p = i915_error_printer(m);
626 
627 	intel_device_info_print_static(&error->device_info, &p);
628 	intel_device_info_print_runtime(&error->runtime_info, &p);
629 	intel_driver_caps_print(&error->driver_caps, &p);
630 }
631 
632 static void err_print_params(struct drm_i915_error_state_buf *m,
633 			     const struct i915_params *params)
634 {
635 	struct drm_printer p = i915_error_printer(m);
636 
637 	i915_params_dump(params, &p);
638 }
639 
640 static void err_print_pciid(struct drm_i915_error_state_buf *m,
641 			    struct drm_i915_private *i915)
642 {
643 	struct pci_dev *pdev = i915->drm.pdev;
644 
645 	err_printf(m, "PCI ID: 0x%04x\n", pdev->device);
646 	err_printf(m, "PCI Revision: 0x%02x\n", pdev->revision);
647 	err_printf(m, "PCI Subsystem: %04x:%04x\n",
648 		   pdev->subsystem_vendor,
649 		   pdev->subsystem_device);
650 }
651 
652 static void err_print_uc(struct drm_i915_error_state_buf *m,
653 			 const struct intel_uc_coredump *error_uc)
654 {
655 	struct drm_printer p = i915_error_printer(m);
656 
657 	intel_uc_fw_dump(&error_uc->guc_fw, &p);
658 	intel_uc_fw_dump(&error_uc->huc_fw, &p);
659 	print_error_vma(m, NULL, error_uc->guc_log);
660 }
661 
662 static void err_free_sgl(struct scatterlist *sgl)
663 {
664 	while (sgl) {
665 		struct scatterlist *sg;
666 
667 		for (sg = sgl; !sg_is_chain(sg); sg++) {
668 			kfree(sg_virt(sg));
669 			if (sg_is_last(sg))
670 				break;
671 		}
672 
673 		sg = sg_is_last(sg) ? NULL : sg_chain_ptr(sg);
674 		free_page((unsigned long)sgl);
675 		sgl = sg;
676 	}
677 }
678 
679 static void err_print_gt_info(struct drm_i915_error_state_buf *m,
680 			      struct intel_gt_coredump *gt)
681 {
682 	struct drm_printer p = i915_error_printer(m);
683 
684 	intel_gt_info_print(&gt->info, &p);
685 	intel_sseu_print_topology(&gt->info.sseu, &p);
686 }
687 
688 static void err_print_gt(struct drm_i915_error_state_buf *m,
689 			 struct intel_gt_coredump *gt)
690 {
691 	const struct intel_engine_coredump *ee;
692 	int i;
693 
694 	err_printf(m, "GT awake: %s\n", yesno(gt->awake));
695 	err_printf(m, "EIR: 0x%08x\n", gt->eir);
696 	err_printf(m, "IER: 0x%08x\n", gt->ier);
697 	for (i = 0; i < gt->ngtier; i++)
698 		err_printf(m, "GTIER[%d]: 0x%08x\n", i, gt->gtier[i]);
699 	err_printf(m, "PGTBL_ER: 0x%08x\n", gt->pgtbl_er);
700 	err_printf(m, "FORCEWAKE: 0x%08x\n", gt->forcewake);
701 	err_printf(m, "DERRMR: 0x%08x\n", gt->derrmr);
702 
703 	for (i = 0; i < gt->nfence; i++)
704 		err_printf(m, "  fence[%d] = %08llx\n", i, gt->fence[i]);
705 
706 	if (IS_GEN_RANGE(m->i915, 6, 11)) {
707 		err_printf(m, "ERROR: 0x%08x\n", gt->error);
708 		err_printf(m, "DONE_REG: 0x%08x\n", gt->done_reg);
709 	}
710 
711 	if (INTEL_GEN(m->i915) >= 8)
712 		err_printf(m, "FAULT_TLB_DATA: 0x%08x 0x%08x\n",
713 			   gt->fault_data1, gt->fault_data0);
714 
715 	if (IS_GEN(m->i915, 7))
716 		err_printf(m, "ERR_INT: 0x%08x\n", gt->err_int);
717 
718 	if (IS_GEN_RANGE(m->i915, 8, 11))
719 		err_printf(m, "GTT_CACHE_EN: 0x%08x\n", gt->gtt_cache);
720 
721 	if (IS_GEN(m->i915, 12))
722 		err_printf(m, "AUX_ERR_DBG: 0x%08x\n", gt->aux_err);
723 
724 	if (INTEL_GEN(m->i915) >= 12) {
725 		int i;
726 
727 		for (i = 0; i < GEN12_SFC_DONE_MAX; i++)
728 			err_printf(m, "  SFC_DONE[%d]: 0x%08x\n", i,
729 				   gt->sfc_done[i]);
730 
731 		err_printf(m, "  GAM_DONE: 0x%08x\n", gt->gam_done);
732 	}
733 
734 	for (ee = gt->engine; ee; ee = ee->next) {
735 		const struct i915_vma_coredump *vma;
736 
737 		error_print_engine(m, ee);
738 		for (vma = ee->vma; vma; vma = vma->next)
739 			print_error_vma(m, ee->engine, vma);
740 	}
741 
742 	if (gt->uc)
743 		err_print_uc(m, gt->uc);
744 
745 	err_print_gt_info(m, gt);
746 }
747 
748 static void __err_print_to_sgl(struct drm_i915_error_state_buf *m,
749 			       struct i915_gpu_coredump *error)
750 {
751 	const struct intel_engine_coredump *ee;
752 	struct timespec64 ts;
753 
754 	if (*error->error_msg)
755 		err_printf(m, "%s\n", error->error_msg);
756 	err_printf(m, "Kernel: %s %s\n",
757 		   init_utsname()->release,
758 		   init_utsname()->machine);
759 	err_printf(m, "Driver: %s\n", DRIVER_DATE);
760 	ts = ktime_to_timespec64(error->time);
761 	err_printf(m, "Time: %lld s %ld us\n",
762 		   (s64)ts.tv_sec, ts.tv_nsec / NSEC_PER_USEC);
763 	ts = ktime_to_timespec64(error->boottime);
764 	err_printf(m, "Boottime: %lld s %ld us\n",
765 		   (s64)ts.tv_sec, ts.tv_nsec / NSEC_PER_USEC);
766 	ts = ktime_to_timespec64(error->uptime);
767 	err_printf(m, "Uptime: %lld s %ld us\n",
768 		   (s64)ts.tv_sec, ts.tv_nsec / NSEC_PER_USEC);
769 	err_printf(m, "Capture: %lu jiffies; %d ms ago\n",
770 		   error->capture, jiffies_to_msecs(jiffies - error->capture));
771 
772 	for (ee = error->gt ? error->gt->engine : NULL; ee; ee = ee->next)
773 		err_printf(m, "Active process (on ring %s): %s [%d]\n",
774 			   ee->engine->name,
775 			   ee->context.comm,
776 			   ee->context.pid);
777 
778 	err_printf(m, "Reset count: %u\n", error->reset_count);
779 	err_printf(m, "Suspend count: %u\n", error->suspend_count);
780 	err_printf(m, "Platform: %s\n", intel_platform_name(error->device_info.platform));
781 	err_printf(m, "Subplatform: 0x%x\n",
782 		   intel_subplatform(&error->runtime_info,
783 				     error->device_info.platform));
784 	err_print_pciid(m, m->i915);
785 
786 	err_printf(m, "IOMMU enabled?: %d\n", error->iommu);
787 
788 	if (HAS_CSR(m->i915)) {
789 		struct intel_csr *csr = &m->i915->csr;
790 
791 		err_printf(m, "DMC loaded: %s\n",
792 			   yesno(csr->dmc_payload != NULL));
793 		err_printf(m, "DMC fw version: %d.%d\n",
794 			   CSR_VERSION_MAJOR(csr->version),
795 			   CSR_VERSION_MINOR(csr->version));
796 	}
797 
798 	err_printf(m, "RPM wakelock: %s\n", yesno(error->wakelock));
799 	err_printf(m, "PM suspended: %s\n", yesno(error->suspended));
800 
801 	if (error->gt)
802 		err_print_gt(m, error->gt);
803 
804 	if (error->overlay)
805 		intel_overlay_print_error_state(m, error->overlay);
806 
807 	if (error->display)
808 		intel_display_print_error_state(m, error->display);
809 
810 	err_print_capabilities(m, error);
811 	err_print_params(m, &error->params);
812 }
813 
814 static int err_print_to_sgl(struct i915_gpu_coredump *error)
815 {
816 	struct drm_i915_error_state_buf m;
817 
818 	if (IS_ERR(error))
819 		return PTR_ERR(error);
820 
821 	if (READ_ONCE(error->sgl))
822 		return 0;
823 
824 	memset(&m, 0, sizeof(m));
825 	m.i915 = error->i915;
826 
827 	__err_print_to_sgl(&m, error);
828 
829 	if (m.buf) {
830 		__sg_set_buf(m.cur++, m.buf, m.bytes, m.iter);
831 		m.bytes = 0;
832 		m.buf = NULL;
833 	}
834 	if (m.cur) {
835 		GEM_BUG_ON(m.end < m.cur);
836 		sg_mark_end(m.cur - 1);
837 	}
838 	GEM_BUG_ON(m.sgl && !m.cur);
839 
840 	if (m.err) {
841 		err_free_sgl(m.sgl);
842 		return m.err;
843 	}
844 
845 	if (cmpxchg(&error->sgl, NULL, m.sgl))
846 		err_free_sgl(m.sgl);
847 
848 	return 0;
849 }
850 
851 ssize_t i915_gpu_coredump_copy_to_buffer(struct i915_gpu_coredump *error,
852 					 char *buf, loff_t off, size_t rem)
853 {
854 	struct scatterlist *sg;
855 	size_t count;
856 	loff_t pos;
857 	int err;
858 
859 	if (!error || !rem)
860 		return 0;
861 
862 	err = err_print_to_sgl(error);
863 	if (err)
864 		return err;
865 
866 	sg = READ_ONCE(error->fit);
867 	if (!sg || off < sg->dma_address)
868 		sg = error->sgl;
869 	if (!sg)
870 		return 0;
871 
872 	pos = sg->dma_address;
873 	count = 0;
874 	do {
875 		size_t len, start;
876 
877 		if (sg_is_chain(sg)) {
878 			sg = sg_chain_ptr(sg);
879 			GEM_BUG_ON(sg_is_chain(sg));
880 		}
881 
882 		len = sg->length;
883 		if (pos + len <= off) {
884 			pos += len;
885 			continue;
886 		}
887 
888 		start = sg->offset;
889 		if (pos < off) {
890 			GEM_BUG_ON(off - pos > len);
891 			len -= off - pos;
892 			start += off - pos;
893 			pos = off;
894 		}
895 
896 		len = min(len, rem);
897 		GEM_BUG_ON(!len || len > sg->length);
898 
899 		memcpy(buf, page_address(sg_page(sg)) + start, len);
900 
901 		count += len;
902 		pos += len;
903 
904 		buf += len;
905 		rem -= len;
906 		if (!rem) {
907 			WRITE_ONCE(error->fit, sg);
908 			break;
909 		}
910 	} while (!sg_is_last(sg++));
911 
912 	return count;
913 }
914 
915 static void i915_vma_coredump_free(struct i915_vma_coredump *vma)
916 {
917 	while (vma) {
918 		struct i915_vma_coredump *next = vma->next;
919 		int page;
920 
921 		for (page = 0; page < vma->page_count; page++)
922 			free_page((unsigned long)vma->pages[page]);
923 
924 		kfree(vma);
925 		vma = next;
926 	}
927 }
928 
929 static void cleanup_params(struct i915_gpu_coredump *error)
930 {
931 	i915_params_free(&error->params);
932 }
933 
934 static void cleanup_uc(struct intel_uc_coredump *uc)
935 {
936 	kfree(uc->guc_fw.path);
937 	kfree(uc->huc_fw.path);
938 	i915_vma_coredump_free(uc->guc_log);
939 
940 	kfree(uc);
941 }
942 
943 static void cleanup_gt(struct intel_gt_coredump *gt)
944 {
945 	while (gt->engine) {
946 		struct intel_engine_coredump *ee = gt->engine;
947 
948 		gt->engine = ee->next;
949 
950 		i915_vma_coredump_free(ee->vma);
951 		kfree(ee);
952 	}
953 
954 	if (gt->uc)
955 		cleanup_uc(gt->uc);
956 
957 	kfree(gt);
958 }
959 
960 void __i915_gpu_coredump_free(struct kref *error_ref)
961 {
962 	struct i915_gpu_coredump *error =
963 		container_of(error_ref, typeof(*error), ref);
964 
965 	while (error->gt) {
966 		struct intel_gt_coredump *gt = error->gt;
967 
968 		error->gt = gt->next;
969 		cleanup_gt(gt);
970 	}
971 
972 	kfree(error->overlay);
973 	kfree(error->display);
974 
975 	cleanup_params(error);
976 
977 	err_free_sgl(error->sgl);
978 	kfree(error);
979 }
980 
981 static struct i915_vma_coredump *
982 i915_vma_coredump_create(const struct intel_gt *gt,
983 			 const struct i915_vma *vma,
984 			 const char *name,
985 			 struct i915_vma_compress *compress)
986 {
987 	struct i915_ggtt *ggtt = gt->ggtt;
988 	const u64 slot = ggtt->error_capture.start;
989 	struct i915_vma_coredump *dst;
990 	unsigned long num_pages;
991 	struct sgt_iter iter;
992 	int ret;
993 
994 	might_sleep();
995 
996 	if (!vma || !vma->pages || !compress)
997 		return NULL;
998 
999 	num_pages = min_t(u64, vma->size, vma->obj->base.size) >> PAGE_SHIFT;
1000 	num_pages = DIV_ROUND_UP(10 * num_pages, 8); /* worstcase zlib growth */
1001 	dst = kmalloc(sizeof(*dst) + num_pages * sizeof(u32 *), ALLOW_FAIL);
1002 	if (!dst)
1003 		return NULL;
1004 
1005 	if (!compress_start(compress)) {
1006 		kfree(dst);
1007 		return NULL;
1008 	}
1009 
1010 	strcpy(dst->name, name);
1011 	dst->next = NULL;
1012 
1013 	dst->gtt_offset = vma->node.start;
1014 	dst->gtt_size = vma->node.size;
1015 	dst->gtt_page_sizes = vma->page_sizes.gtt;
1016 	dst->num_pages = num_pages;
1017 	dst->page_count = 0;
1018 	dst->unused = 0;
1019 
1020 	ret = -EINVAL;
1021 	if (drm_mm_node_allocated(&ggtt->error_capture)) {
1022 		void __iomem *s;
1023 		dma_addr_t dma;
1024 
1025 		for_each_sgt_daddr(dma, iter, vma->pages) {
1026 			ggtt->vm.insert_page(&ggtt->vm, dma, slot,
1027 					     I915_CACHE_NONE, 0);
1028 			mb();
1029 
1030 			s = io_mapping_map_wc(&ggtt->iomap, slot, PAGE_SIZE);
1031 			ret = compress_page(compress,
1032 					    (void  __force *)s, dst,
1033 					    true);
1034 			io_mapping_unmap(s);
1035 			if (ret)
1036 				break;
1037 		}
1038 	} else if (i915_gem_object_is_lmem(vma->obj)) {
1039 		struct intel_memory_region *mem = vma->obj->mm.region;
1040 		dma_addr_t dma;
1041 
1042 		for_each_sgt_daddr(dma, iter, vma->pages) {
1043 			void __iomem *s;
1044 
1045 			s = io_mapping_map_wc(&mem->iomap, dma, PAGE_SIZE);
1046 			ret = compress_page(compress,
1047 					    (void __force *)s, dst,
1048 					    true);
1049 			io_mapping_unmap(s);
1050 			if (ret)
1051 				break;
1052 		}
1053 	} else {
1054 		struct page *page;
1055 
1056 		for_each_sgt_page(page, iter, vma->pages) {
1057 			void *s;
1058 
1059 			drm_clflush_pages(&page, 1);
1060 
1061 			s = kmap(page);
1062 			ret = compress_page(compress, s, dst, false);
1063 			kunmap(page);
1064 
1065 			drm_clflush_pages(&page, 1);
1066 
1067 			if (ret)
1068 				break;
1069 		}
1070 	}
1071 
1072 	if (ret || compress_flush(compress, dst)) {
1073 		while (dst->page_count--)
1074 			pool_free(&compress->pool, dst->pages[dst->page_count]);
1075 		kfree(dst);
1076 		dst = NULL;
1077 	}
1078 	compress_finish(compress);
1079 
1080 	return dst;
1081 }
1082 
1083 static void gt_record_fences(struct intel_gt_coredump *gt)
1084 {
1085 	struct i915_ggtt *ggtt = gt->_gt->ggtt;
1086 	struct intel_uncore *uncore = gt->_gt->uncore;
1087 	int i;
1088 
1089 	if (INTEL_GEN(uncore->i915) >= 6) {
1090 		for (i = 0; i < ggtt->num_fences; i++)
1091 			gt->fence[i] =
1092 				intel_uncore_read64(uncore,
1093 						    FENCE_REG_GEN6_LO(i));
1094 	} else if (INTEL_GEN(uncore->i915) >= 4) {
1095 		for (i = 0; i < ggtt->num_fences; i++)
1096 			gt->fence[i] =
1097 				intel_uncore_read64(uncore,
1098 						    FENCE_REG_965_LO(i));
1099 	} else {
1100 		for (i = 0; i < ggtt->num_fences; i++)
1101 			gt->fence[i] =
1102 				intel_uncore_read(uncore, FENCE_REG(i));
1103 	}
1104 	gt->nfence = i;
1105 }
1106 
1107 static void engine_record_registers(struct intel_engine_coredump *ee)
1108 {
1109 	const struct intel_engine_cs *engine = ee->engine;
1110 	struct drm_i915_private *i915 = engine->i915;
1111 
1112 	if (INTEL_GEN(i915) >= 6) {
1113 		ee->rc_psmi = ENGINE_READ(engine, RING_PSMI_CTL);
1114 
1115 		if (INTEL_GEN(i915) >= 12)
1116 			ee->fault_reg = intel_uncore_read(engine->uncore,
1117 							  GEN12_RING_FAULT_REG);
1118 		else if (INTEL_GEN(i915) >= 8)
1119 			ee->fault_reg = intel_uncore_read(engine->uncore,
1120 							  GEN8_RING_FAULT_REG);
1121 		else
1122 			ee->fault_reg = GEN6_RING_FAULT_REG_READ(engine);
1123 	}
1124 
1125 	if (INTEL_GEN(i915) >= 4) {
1126 		ee->esr = ENGINE_READ(engine, RING_ESR);
1127 		ee->faddr = ENGINE_READ(engine, RING_DMA_FADD);
1128 		ee->ipeir = ENGINE_READ(engine, RING_IPEIR);
1129 		ee->ipehr = ENGINE_READ(engine, RING_IPEHR);
1130 		ee->instps = ENGINE_READ(engine, RING_INSTPS);
1131 		ee->bbaddr = ENGINE_READ(engine, RING_BBADDR);
1132 		ee->ccid = ENGINE_READ(engine, CCID);
1133 		if (INTEL_GEN(i915) >= 8) {
1134 			ee->faddr |= (u64)ENGINE_READ(engine, RING_DMA_FADD_UDW) << 32;
1135 			ee->bbaddr |= (u64)ENGINE_READ(engine, RING_BBADDR_UDW) << 32;
1136 		}
1137 		ee->bbstate = ENGINE_READ(engine, RING_BBSTATE);
1138 	} else {
1139 		ee->faddr = ENGINE_READ(engine, DMA_FADD_I8XX);
1140 		ee->ipeir = ENGINE_READ(engine, IPEIR);
1141 		ee->ipehr = ENGINE_READ(engine, IPEHR);
1142 	}
1143 
1144 	intel_engine_get_instdone(engine, &ee->instdone);
1145 
1146 	ee->instpm = ENGINE_READ(engine, RING_INSTPM);
1147 	ee->acthd = intel_engine_get_active_head(engine);
1148 	ee->start = ENGINE_READ(engine, RING_START);
1149 	ee->head = ENGINE_READ(engine, RING_HEAD);
1150 	ee->tail = ENGINE_READ(engine, RING_TAIL);
1151 	ee->ctl = ENGINE_READ(engine, RING_CTL);
1152 	if (INTEL_GEN(i915) > 2)
1153 		ee->mode = ENGINE_READ(engine, RING_MI_MODE);
1154 
1155 	if (!HWS_NEEDS_PHYSICAL(i915)) {
1156 		i915_reg_t mmio;
1157 
1158 		if (IS_GEN(i915, 7)) {
1159 			switch (engine->id) {
1160 			default:
1161 				MISSING_CASE(engine->id);
1162 				/* fall through */
1163 			case RCS0:
1164 				mmio = RENDER_HWS_PGA_GEN7;
1165 				break;
1166 			case BCS0:
1167 				mmio = BLT_HWS_PGA_GEN7;
1168 				break;
1169 			case VCS0:
1170 				mmio = BSD_HWS_PGA_GEN7;
1171 				break;
1172 			case VECS0:
1173 				mmio = VEBOX_HWS_PGA_GEN7;
1174 				break;
1175 			}
1176 		} else if (IS_GEN(engine->i915, 6)) {
1177 			mmio = RING_HWS_PGA_GEN6(engine->mmio_base);
1178 		} else {
1179 			/* XXX: gen8 returns to sanity */
1180 			mmio = RING_HWS_PGA(engine->mmio_base);
1181 		}
1182 
1183 		ee->hws = intel_uncore_read(engine->uncore, mmio);
1184 	}
1185 
1186 	ee->reset_count = i915_reset_engine_count(&i915->gpu_error, engine);
1187 
1188 	if (HAS_PPGTT(i915)) {
1189 		int i;
1190 
1191 		ee->vm_info.gfx_mode = ENGINE_READ(engine, RING_MODE_GEN7);
1192 
1193 		if (IS_GEN(i915, 6)) {
1194 			ee->vm_info.pp_dir_base =
1195 				ENGINE_READ(engine, RING_PP_DIR_BASE_READ);
1196 		} else if (IS_GEN(i915, 7)) {
1197 			ee->vm_info.pp_dir_base =
1198 				ENGINE_READ(engine, RING_PP_DIR_BASE);
1199 		} else if (INTEL_GEN(i915) >= 8) {
1200 			u32 base = engine->mmio_base;
1201 
1202 			for (i = 0; i < 4; i++) {
1203 				ee->vm_info.pdp[i] =
1204 					intel_uncore_read(engine->uncore,
1205 							  GEN8_RING_PDP_UDW(base, i));
1206 				ee->vm_info.pdp[i] <<= 32;
1207 				ee->vm_info.pdp[i] |=
1208 					intel_uncore_read(engine->uncore,
1209 							  GEN8_RING_PDP_LDW(base, i));
1210 			}
1211 		}
1212 	}
1213 }
1214 
1215 static void record_request(const struct i915_request *request,
1216 			   struct i915_request_coredump *erq)
1217 {
1218 	erq->flags = request->fence.flags;
1219 	erq->context = request->fence.context;
1220 	erq->seqno = request->fence.seqno;
1221 	erq->sched_attr = request->sched.attr;
1222 	erq->head = request->head;
1223 	erq->tail = request->tail;
1224 
1225 	erq->pid = 0;
1226 	rcu_read_lock();
1227 	if (!intel_context_is_closed(request->context)) {
1228 		const struct i915_gem_context *ctx;
1229 
1230 		ctx = rcu_dereference(request->context->gem_context);
1231 		if (ctx)
1232 			erq->pid = pid_nr(ctx->pid);
1233 	}
1234 	rcu_read_unlock();
1235 }
1236 
1237 static void engine_record_execlists(struct intel_engine_coredump *ee)
1238 {
1239 	const struct intel_engine_execlists * const el = &ee->engine->execlists;
1240 	struct i915_request * const *port = el->active;
1241 	unsigned int n = 0;
1242 
1243 	while (*port)
1244 		record_request(*port++, &ee->execlist[n++]);
1245 
1246 	ee->num_ports = n;
1247 }
1248 
1249 static bool record_context(struct i915_gem_context_coredump *e,
1250 			   const struct i915_request *rq)
1251 {
1252 	struct i915_gem_context *ctx;
1253 	struct task_struct *task;
1254 	bool simulated;
1255 
1256 	rcu_read_lock();
1257 	ctx = rcu_dereference(rq->context->gem_context);
1258 	if (ctx && !kref_get_unless_zero(&ctx->ref))
1259 		ctx = NULL;
1260 	rcu_read_unlock();
1261 	if (!ctx)
1262 		return true;
1263 
1264 	rcu_read_lock();
1265 	task = pid_task(ctx->pid, PIDTYPE_PID);
1266 	if (task) {
1267 		strcpy(e->comm, task->comm);
1268 		e->pid = task->pid;
1269 	}
1270 	rcu_read_unlock();
1271 
1272 	e->sched_attr = ctx->sched;
1273 	e->guilty = atomic_read(&ctx->guilty_count);
1274 	e->active = atomic_read(&ctx->active_count);
1275 
1276 	e->total_runtime = rq->context->runtime.total;
1277 	e->avg_runtime = ewma_runtime_read(&rq->context->runtime.avg);
1278 
1279 	simulated = i915_gem_context_no_error_capture(ctx);
1280 
1281 	i915_gem_context_put(ctx);
1282 	return simulated;
1283 }
1284 
1285 struct intel_engine_capture_vma {
1286 	struct intel_engine_capture_vma *next;
1287 	struct i915_vma *vma;
1288 	char name[16];
1289 };
1290 
1291 static struct intel_engine_capture_vma *
1292 capture_vma(struct intel_engine_capture_vma *next,
1293 	    struct i915_vma *vma,
1294 	    const char *name,
1295 	    gfp_t gfp)
1296 {
1297 	struct intel_engine_capture_vma *c;
1298 
1299 	if (!vma)
1300 		return next;
1301 
1302 	c = kmalloc(sizeof(*c), gfp);
1303 	if (!c)
1304 		return next;
1305 
1306 	if (!i915_active_acquire_if_busy(&vma->active)) {
1307 		kfree(c);
1308 		return next;
1309 	}
1310 
1311 	strcpy(c->name, name);
1312 	c->vma = i915_vma_get(vma);
1313 
1314 	c->next = next;
1315 	return c;
1316 }
1317 
1318 static struct intel_engine_capture_vma *
1319 capture_user(struct intel_engine_capture_vma *capture,
1320 	     const struct i915_request *rq,
1321 	     gfp_t gfp)
1322 {
1323 	struct i915_capture_list *c;
1324 
1325 	for (c = rq->capture_list; c; c = c->next)
1326 		capture = capture_vma(capture, c->vma, "user", gfp);
1327 
1328 	return capture;
1329 }
1330 
1331 static void add_vma(struct intel_engine_coredump *ee,
1332 		    struct i915_vma_coredump *vma)
1333 {
1334 	if (vma) {
1335 		vma->next = ee->vma;
1336 		ee->vma = vma;
1337 	}
1338 }
1339 
1340 struct intel_engine_coredump *
1341 intel_engine_coredump_alloc(struct intel_engine_cs *engine, gfp_t gfp)
1342 {
1343 	struct intel_engine_coredump *ee;
1344 
1345 	ee = kzalloc(sizeof(*ee), gfp);
1346 	if (!ee)
1347 		return NULL;
1348 
1349 	ee->engine = engine;
1350 
1351 	engine_record_registers(ee);
1352 	engine_record_execlists(ee);
1353 
1354 	return ee;
1355 }
1356 
1357 struct intel_engine_capture_vma *
1358 intel_engine_coredump_add_request(struct intel_engine_coredump *ee,
1359 				  struct i915_request *rq,
1360 				  gfp_t gfp)
1361 {
1362 	struct intel_engine_capture_vma *vma = NULL;
1363 
1364 	ee->simulated |= record_context(&ee->context, rq);
1365 	if (ee->simulated)
1366 		return NULL;
1367 
1368 	/*
1369 	 * We need to copy these to an anonymous buffer
1370 	 * as the simplest method to avoid being overwritten
1371 	 * by userspace.
1372 	 */
1373 	vma = capture_vma(vma, rq->batch, "batch", gfp);
1374 	vma = capture_user(vma, rq, gfp);
1375 	vma = capture_vma(vma, rq->ring->vma, "ring", gfp);
1376 	vma = capture_vma(vma, rq->context->state, "HW context", gfp);
1377 
1378 	ee->rq_head = rq->head;
1379 	ee->rq_post = rq->postfix;
1380 	ee->rq_tail = rq->tail;
1381 
1382 	return vma;
1383 }
1384 
1385 void
1386 intel_engine_coredump_add_vma(struct intel_engine_coredump *ee,
1387 			      struct intel_engine_capture_vma *capture,
1388 			      struct i915_vma_compress *compress)
1389 {
1390 	const struct intel_engine_cs *engine = ee->engine;
1391 
1392 	while (capture) {
1393 		struct intel_engine_capture_vma *this = capture;
1394 		struct i915_vma *vma = this->vma;
1395 
1396 		add_vma(ee,
1397 			i915_vma_coredump_create(engine->gt,
1398 						 vma, this->name,
1399 						 compress));
1400 
1401 		i915_active_release(&vma->active);
1402 		i915_vma_put(vma);
1403 
1404 		capture = this->next;
1405 		kfree(this);
1406 	}
1407 
1408 	add_vma(ee,
1409 		i915_vma_coredump_create(engine->gt,
1410 					 engine->status_page.vma,
1411 					 "HW Status",
1412 					 compress));
1413 
1414 	add_vma(ee,
1415 		i915_vma_coredump_create(engine->gt,
1416 					 engine->wa_ctx.vma,
1417 					 "WA context",
1418 					 compress));
1419 }
1420 
1421 static struct intel_engine_coredump *
1422 capture_engine(struct intel_engine_cs *engine,
1423 	       struct i915_vma_compress *compress)
1424 {
1425 	struct intel_engine_capture_vma *capture = NULL;
1426 	struct intel_engine_coredump *ee;
1427 	struct i915_request *rq;
1428 	unsigned long flags;
1429 
1430 	ee = intel_engine_coredump_alloc(engine, GFP_KERNEL);
1431 	if (!ee)
1432 		return NULL;
1433 
1434 	spin_lock_irqsave(&engine->active.lock, flags);
1435 	rq = intel_engine_find_active_request(engine);
1436 	if (rq)
1437 		capture = intel_engine_coredump_add_request(ee, rq,
1438 							    ATOMIC_MAYFAIL);
1439 	spin_unlock_irqrestore(&engine->active.lock, flags);
1440 	if (!capture) {
1441 		kfree(ee);
1442 		return NULL;
1443 	}
1444 
1445 	intel_engine_coredump_add_vma(ee, capture, compress);
1446 
1447 	return ee;
1448 }
1449 
1450 static void
1451 gt_record_engines(struct intel_gt_coredump *gt,
1452 		  struct i915_vma_compress *compress)
1453 {
1454 	struct intel_engine_cs *engine;
1455 	enum intel_engine_id id;
1456 
1457 	for_each_engine(engine, gt->_gt, id) {
1458 		struct intel_engine_coredump *ee;
1459 
1460 		/* Refill our page pool before entering atomic section */
1461 		pool_refill(&compress->pool, ALLOW_FAIL);
1462 
1463 		ee = capture_engine(engine, compress);
1464 		if (!ee)
1465 			continue;
1466 
1467 		gt->simulated |= ee->simulated;
1468 		if (ee->simulated) {
1469 			kfree(ee);
1470 			continue;
1471 		}
1472 
1473 		ee->next = gt->engine;
1474 		gt->engine = ee;
1475 	}
1476 }
1477 
1478 static struct intel_uc_coredump *
1479 gt_record_uc(struct intel_gt_coredump *gt,
1480 	     struct i915_vma_compress *compress)
1481 {
1482 	const struct intel_uc *uc = &gt->_gt->uc;
1483 	struct intel_uc_coredump *error_uc;
1484 
1485 	error_uc = kzalloc(sizeof(*error_uc), ALLOW_FAIL);
1486 	if (!error_uc)
1487 		return NULL;
1488 
1489 	memcpy(&error_uc->guc_fw, &uc->guc.fw, sizeof(uc->guc.fw));
1490 	memcpy(&error_uc->huc_fw, &uc->huc.fw, sizeof(uc->huc.fw));
1491 
1492 	/* Non-default firmware paths will be specified by the modparam.
1493 	 * As modparams are generally accesible from the userspace make
1494 	 * explicit copies of the firmware paths.
1495 	 */
1496 	error_uc->guc_fw.path = kstrdup(uc->guc.fw.path, ALLOW_FAIL);
1497 	error_uc->huc_fw.path = kstrdup(uc->huc.fw.path, ALLOW_FAIL);
1498 	error_uc->guc_log =
1499 		i915_vma_coredump_create(gt->_gt,
1500 					 uc->guc.log.vma, "GuC log buffer",
1501 					 compress);
1502 
1503 	return error_uc;
1504 }
1505 
1506 static void gt_capture_prepare(struct intel_gt_coredump *gt)
1507 {
1508 	struct i915_ggtt *ggtt = gt->_gt->ggtt;
1509 
1510 	mutex_lock(&ggtt->error_mutex);
1511 }
1512 
1513 static void gt_capture_finish(struct intel_gt_coredump *gt)
1514 {
1515 	struct i915_ggtt *ggtt = gt->_gt->ggtt;
1516 
1517 	if (drm_mm_node_allocated(&ggtt->error_capture))
1518 		ggtt->vm.clear_range(&ggtt->vm,
1519 				     ggtt->error_capture.start,
1520 				     PAGE_SIZE);
1521 
1522 	mutex_unlock(&ggtt->error_mutex);
1523 }
1524 
1525 /* Capture all registers which don't fit into another category. */
1526 static void gt_record_regs(struct intel_gt_coredump *gt)
1527 {
1528 	struct intel_uncore *uncore = gt->_gt->uncore;
1529 	struct drm_i915_private *i915 = uncore->i915;
1530 	int i;
1531 
1532 	/*
1533 	 * General organization
1534 	 * 1. Registers specific to a single generation
1535 	 * 2. Registers which belong to multiple generations
1536 	 * 3. Feature specific registers.
1537 	 * 4. Everything else
1538 	 * Please try to follow the order.
1539 	 */
1540 
1541 	/* 1: Registers specific to a single generation */
1542 	if (IS_VALLEYVIEW(i915)) {
1543 		gt->gtier[0] = intel_uncore_read(uncore, GTIER);
1544 		gt->ier = intel_uncore_read(uncore, VLV_IER);
1545 		gt->forcewake = intel_uncore_read_fw(uncore, FORCEWAKE_VLV);
1546 	}
1547 
1548 	if (IS_GEN(i915, 7))
1549 		gt->err_int = intel_uncore_read(uncore, GEN7_ERR_INT);
1550 
1551 	if (INTEL_GEN(i915) >= 12) {
1552 		gt->fault_data0 = intel_uncore_read(uncore,
1553 						    GEN12_FAULT_TLB_DATA0);
1554 		gt->fault_data1 = intel_uncore_read(uncore,
1555 						    GEN12_FAULT_TLB_DATA1);
1556 	} else if (INTEL_GEN(i915) >= 8) {
1557 		gt->fault_data0 = intel_uncore_read(uncore,
1558 						    GEN8_FAULT_TLB_DATA0);
1559 		gt->fault_data1 = intel_uncore_read(uncore,
1560 						    GEN8_FAULT_TLB_DATA1);
1561 	}
1562 
1563 	if (IS_GEN(i915, 6)) {
1564 		gt->forcewake = intel_uncore_read_fw(uncore, FORCEWAKE);
1565 		gt->gab_ctl = intel_uncore_read(uncore, GAB_CTL);
1566 		gt->gfx_mode = intel_uncore_read(uncore, GFX_MODE);
1567 	}
1568 
1569 	/* 2: Registers which belong to multiple generations */
1570 	if (INTEL_GEN(i915) >= 7)
1571 		gt->forcewake = intel_uncore_read_fw(uncore, FORCEWAKE_MT);
1572 
1573 	if (INTEL_GEN(i915) >= 6) {
1574 		gt->derrmr = intel_uncore_read(uncore, DERRMR);
1575 		if (INTEL_GEN(i915) < 12) {
1576 			gt->error = intel_uncore_read(uncore, ERROR_GEN6);
1577 			gt->done_reg = intel_uncore_read(uncore, DONE_REG);
1578 		}
1579 	}
1580 
1581 	/* 3: Feature specific registers */
1582 	if (IS_GEN_RANGE(i915, 6, 7)) {
1583 		gt->gam_ecochk = intel_uncore_read(uncore, GAM_ECOCHK);
1584 		gt->gac_eco = intel_uncore_read(uncore, GAC_ECO_BITS);
1585 	}
1586 
1587 	if (IS_GEN_RANGE(i915, 8, 11))
1588 		gt->gtt_cache = intel_uncore_read(uncore, HSW_GTT_CACHE_EN);
1589 
1590 	if (IS_GEN(i915, 12))
1591 		gt->aux_err = intel_uncore_read(uncore, GEN12_AUX_ERR_DBG);
1592 
1593 	if (INTEL_GEN(i915) >= 12) {
1594 		for (i = 0; i < GEN12_SFC_DONE_MAX; i++) {
1595 			gt->sfc_done[i] =
1596 				intel_uncore_read(uncore, GEN12_SFC_DONE(i));
1597 		}
1598 
1599 		gt->gam_done = intel_uncore_read(uncore, GEN12_GAM_DONE);
1600 	}
1601 
1602 	/* 4: Everything else */
1603 	if (INTEL_GEN(i915) >= 11) {
1604 		gt->ier = intel_uncore_read(uncore, GEN8_DE_MISC_IER);
1605 		gt->gtier[0] =
1606 			intel_uncore_read(uncore,
1607 					  GEN11_RENDER_COPY_INTR_ENABLE);
1608 		gt->gtier[1] =
1609 			intel_uncore_read(uncore, GEN11_VCS_VECS_INTR_ENABLE);
1610 		gt->gtier[2] =
1611 			intel_uncore_read(uncore, GEN11_GUC_SG_INTR_ENABLE);
1612 		gt->gtier[3] =
1613 			intel_uncore_read(uncore,
1614 					  GEN11_GPM_WGBOXPERF_INTR_ENABLE);
1615 		gt->gtier[4] =
1616 			intel_uncore_read(uncore,
1617 					  GEN11_CRYPTO_RSVD_INTR_ENABLE);
1618 		gt->gtier[5] =
1619 			intel_uncore_read(uncore,
1620 					  GEN11_GUNIT_CSME_INTR_ENABLE);
1621 		gt->ngtier = 6;
1622 	} else if (INTEL_GEN(i915) >= 8) {
1623 		gt->ier = intel_uncore_read(uncore, GEN8_DE_MISC_IER);
1624 		for (i = 0; i < 4; i++)
1625 			gt->gtier[i] =
1626 				intel_uncore_read(uncore, GEN8_GT_IER(i));
1627 		gt->ngtier = 4;
1628 	} else if (HAS_PCH_SPLIT(i915)) {
1629 		gt->ier = intel_uncore_read(uncore, DEIER);
1630 		gt->gtier[0] = intel_uncore_read(uncore, GTIER);
1631 		gt->ngtier = 1;
1632 	} else if (IS_GEN(i915, 2)) {
1633 		gt->ier = intel_uncore_read16(uncore, GEN2_IER);
1634 	} else if (!IS_VALLEYVIEW(i915)) {
1635 		gt->ier = intel_uncore_read(uncore, GEN2_IER);
1636 	}
1637 	gt->eir = intel_uncore_read(uncore, EIR);
1638 	gt->pgtbl_er = intel_uncore_read(uncore, PGTBL_ER);
1639 }
1640 
1641 static void gt_record_info(struct intel_gt_coredump *gt)
1642 {
1643 	memcpy(&gt->info, &gt->_gt->info, sizeof(struct intel_gt_info));
1644 }
1645 
1646 /*
1647  * Generate a semi-unique error code. The code is not meant to have meaning, The
1648  * code's only purpose is to try to prevent false duplicated bug reports by
1649  * grossly estimating a GPU error state.
1650  *
1651  * TODO Ideally, hashing the batchbuffer would be a very nice way to determine
1652  * the hang if we could strip the GTT offset information from it.
1653  *
1654  * It's only a small step better than a random number in its current form.
1655  */
1656 static u32 generate_ecode(const struct intel_engine_coredump *ee)
1657 {
1658 	/*
1659 	 * IPEHR would be an ideal way to detect errors, as it's the gross
1660 	 * measure of "the command that hung." However, has some very common
1661 	 * synchronization commands which almost always appear in the case
1662 	 * strictly a client bug. Use instdone to differentiate those some.
1663 	 */
1664 	return ee ? ee->ipehr ^ ee->instdone.instdone : 0;
1665 }
1666 
1667 static const char *error_msg(struct i915_gpu_coredump *error)
1668 {
1669 	struct intel_engine_coredump *first = NULL;
1670 	struct intel_gt_coredump *gt;
1671 	intel_engine_mask_t engines;
1672 	int len;
1673 
1674 	engines = 0;
1675 	for (gt = error->gt; gt; gt = gt->next) {
1676 		struct intel_engine_coredump *cs;
1677 
1678 		if (gt->engine && !first)
1679 			first = gt->engine;
1680 
1681 		for (cs = gt->engine; cs; cs = cs->next)
1682 			engines |= cs->engine->mask;
1683 	}
1684 
1685 	len = scnprintf(error->error_msg, sizeof(error->error_msg),
1686 			"GPU HANG: ecode %d:%x:%08x",
1687 			INTEL_GEN(error->i915), engines,
1688 			generate_ecode(first));
1689 	if (first && first->context.pid) {
1690 		/* Just show the first executing process, more is confusing */
1691 		len += scnprintf(error->error_msg + len,
1692 				 sizeof(error->error_msg) - len,
1693 				 ", in %s [%d]",
1694 				 first->context.comm, first->context.pid);
1695 	}
1696 
1697 	return error->error_msg;
1698 }
1699 
1700 static void capture_gen(struct i915_gpu_coredump *error)
1701 {
1702 	struct drm_i915_private *i915 = error->i915;
1703 
1704 	error->wakelock = atomic_read(&i915->runtime_pm.wakeref_count);
1705 	error->suspended = i915->runtime_pm.suspended;
1706 
1707 	error->iommu = -1;
1708 #ifdef CONFIG_INTEL_IOMMU
1709 	error->iommu = intel_iommu_gfx_mapped;
1710 #endif
1711 	error->reset_count = i915_reset_count(&i915->gpu_error);
1712 	error->suspend_count = i915->suspend_count;
1713 
1714 	i915_params_copy(&error->params, &i915->params);
1715 	memcpy(&error->device_info,
1716 	       INTEL_INFO(i915),
1717 	       sizeof(error->device_info));
1718 	memcpy(&error->runtime_info,
1719 	       RUNTIME_INFO(i915),
1720 	       sizeof(error->runtime_info));
1721 	error->driver_caps = i915->caps;
1722 }
1723 
1724 struct i915_gpu_coredump *
1725 i915_gpu_coredump_alloc(struct drm_i915_private *i915, gfp_t gfp)
1726 {
1727 	struct i915_gpu_coredump *error;
1728 
1729 	if (!i915->params.error_capture)
1730 		return NULL;
1731 
1732 	error = kzalloc(sizeof(*error), gfp);
1733 	if (!error)
1734 		return NULL;
1735 
1736 	kref_init(&error->ref);
1737 	error->i915 = i915;
1738 
1739 	error->time = ktime_get_real();
1740 	error->boottime = ktime_get_boottime();
1741 	error->uptime = ktime_sub(ktime_get(), i915->gt.last_init_time);
1742 	error->capture = jiffies;
1743 
1744 	capture_gen(error);
1745 
1746 	return error;
1747 }
1748 
1749 #define DAY_AS_SECONDS(x) (24 * 60 * 60 * (x))
1750 
1751 struct intel_gt_coredump *
1752 intel_gt_coredump_alloc(struct intel_gt *gt, gfp_t gfp)
1753 {
1754 	struct intel_gt_coredump *gc;
1755 
1756 	gc = kzalloc(sizeof(*gc), gfp);
1757 	if (!gc)
1758 		return NULL;
1759 
1760 	gc->_gt = gt;
1761 	gc->awake = intel_gt_pm_is_awake(gt);
1762 
1763 	gt_record_regs(gc);
1764 	gt_record_fences(gc);
1765 
1766 	return gc;
1767 }
1768 
1769 struct i915_vma_compress *
1770 i915_vma_capture_prepare(struct intel_gt_coredump *gt)
1771 {
1772 	struct i915_vma_compress *compress;
1773 
1774 	compress = kmalloc(sizeof(*compress), ALLOW_FAIL);
1775 	if (!compress)
1776 		return NULL;
1777 
1778 	if (!compress_init(compress)) {
1779 		kfree(compress);
1780 		return NULL;
1781 	}
1782 
1783 	gt_capture_prepare(gt);
1784 
1785 	return compress;
1786 }
1787 
1788 void i915_vma_capture_finish(struct intel_gt_coredump *gt,
1789 			     struct i915_vma_compress *compress)
1790 {
1791 	if (!compress)
1792 		return;
1793 
1794 	gt_capture_finish(gt);
1795 
1796 	compress_fini(compress);
1797 	kfree(compress);
1798 }
1799 
1800 struct i915_gpu_coredump *i915_gpu_coredump(struct drm_i915_private *i915)
1801 {
1802 	struct i915_gpu_coredump *error;
1803 
1804 	/* Check if GPU capture has been disabled */
1805 	error = READ_ONCE(i915->gpu_error.first_error);
1806 	if (IS_ERR(error))
1807 		return error;
1808 
1809 	error = i915_gpu_coredump_alloc(i915, ALLOW_FAIL);
1810 	if (!error)
1811 		return ERR_PTR(-ENOMEM);
1812 
1813 	error->gt = intel_gt_coredump_alloc(&i915->gt, ALLOW_FAIL);
1814 	if (error->gt) {
1815 		struct i915_vma_compress *compress;
1816 
1817 		compress = i915_vma_capture_prepare(error->gt);
1818 		if (!compress) {
1819 			kfree(error->gt);
1820 			kfree(error);
1821 			return ERR_PTR(-ENOMEM);
1822 		}
1823 
1824 		gt_record_info(error->gt);
1825 		gt_record_engines(error->gt, compress);
1826 
1827 		if (INTEL_INFO(i915)->has_gt_uc)
1828 			error->gt->uc = gt_record_uc(error->gt, compress);
1829 
1830 		i915_vma_capture_finish(error->gt, compress);
1831 
1832 		error->simulated |= error->gt->simulated;
1833 	}
1834 
1835 	error->overlay = intel_overlay_capture_error_state(i915);
1836 	error->display = intel_display_capture_error_state(i915);
1837 
1838 	return error;
1839 }
1840 
1841 void i915_error_state_store(struct i915_gpu_coredump *error)
1842 {
1843 	struct drm_i915_private *i915;
1844 	static bool warned;
1845 
1846 	if (IS_ERR_OR_NULL(error))
1847 		return;
1848 
1849 	i915 = error->i915;
1850 	drm_info(&i915->drm, "%s\n", error_msg(error));
1851 
1852 	if (error->simulated ||
1853 	    cmpxchg(&i915->gpu_error.first_error, NULL, error))
1854 		return;
1855 
1856 	i915_gpu_coredump_get(error);
1857 
1858 	if (!xchg(&warned, true) &&
1859 	    ktime_get_real_seconds() - DRIVER_TIMESTAMP < DAY_AS_SECONDS(180)) {
1860 		pr_info("GPU hangs can indicate a bug anywhere in the entire gfx stack, including userspace.\n");
1861 		pr_info("Please file a _new_ bug report at https://gitlab.freedesktop.org/drm/intel/issues/new.\n");
1862 		pr_info("Please see https://gitlab.freedesktop.org/drm/intel/-/wikis/How-to-file-i915-bugs for details.\n");
1863 		pr_info("drm/i915 developers can then reassign to the right component if it's not a kernel issue.\n");
1864 		pr_info("The GPU crash dump is required to analyze GPU hangs, so please always attach it.\n");
1865 		pr_info("GPU crash dump saved to /sys/class/drm/card%d/error\n",
1866 			i915->drm.primary->index);
1867 	}
1868 }
1869 
1870 /**
1871  * i915_capture_error_state - capture an error record for later analysis
1872  * @i915: i915 device
1873  *
1874  * Should be called when an error is detected (either a hang or an error
1875  * interrupt) to capture error state from the time of the error.  Fills
1876  * out a structure which becomes available in debugfs for user level tools
1877  * to pick up.
1878  */
1879 void i915_capture_error_state(struct drm_i915_private *i915)
1880 {
1881 	struct i915_gpu_coredump *error;
1882 
1883 	error = i915_gpu_coredump(i915);
1884 	if (IS_ERR(error)) {
1885 		cmpxchg(&i915->gpu_error.first_error, NULL, error);
1886 		return;
1887 	}
1888 
1889 	i915_error_state_store(error);
1890 	i915_gpu_coredump_put(error);
1891 }
1892 
1893 struct i915_gpu_coredump *
1894 i915_first_error_state(struct drm_i915_private *i915)
1895 {
1896 	struct i915_gpu_coredump *error;
1897 
1898 	spin_lock_irq(&i915->gpu_error.lock);
1899 	error = i915->gpu_error.first_error;
1900 	if (!IS_ERR_OR_NULL(error))
1901 		i915_gpu_coredump_get(error);
1902 	spin_unlock_irq(&i915->gpu_error.lock);
1903 
1904 	return error;
1905 }
1906 
1907 void i915_reset_error_state(struct drm_i915_private *i915)
1908 {
1909 	struct i915_gpu_coredump *error;
1910 
1911 	spin_lock_irq(&i915->gpu_error.lock);
1912 	error = i915->gpu_error.first_error;
1913 	if (error != ERR_PTR(-ENODEV)) /* if disabled, always disabled */
1914 		i915->gpu_error.first_error = NULL;
1915 	spin_unlock_irq(&i915->gpu_error.lock);
1916 
1917 	if (!IS_ERR_OR_NULL(error))
1918 		i915_gpu_coredump_put(error);
1919 }
1920 
1921 void i915_disable_error_state(struct drm_i915_private *i915, int err)
1922 {
1923 	spin_lock_irq(&i915->gpu_error.lock);
1924 	if (!i915->gpu_error.first_error)
1925 		i915->gpu_error.first_error = ERR_PTR(err);
1926 	spin_unlock_irq(&i915->gpu_error.lock);
1927 }
1928