Linux Audio

Check our new training course

Loading...
v4.6
   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 <generated/utsrelease.h>
 
 
 
 
 
 
 
 
 
 
 
 
 
  31#include "i915_drv.h"
 
 
 
 
  32
  33static const char *ring_str(int ring)
 
 
 
 
  34{
  35	switch (ring) {
  36	case RCS: return "render";
  37	case VCS: return "bsd";
  38	case BCS: return "blt";
  39	case VECS: return "vebox";
  40	case VCS2: return "bsd2";
  41	default: return "";
  42	}
  43}
  44
  45static const char *pin_flag(int pinned)
  46{
  47	if (pinned > 0)
  48		return " P";
  49	else if (pinned < 0)
  50		return " p";
  51	else
  52		return "";
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
  53}
  54
  55static const char *tiling_flag(int tiling)
 
 
  56{
  57	switch (tiling) {
  58	default:
  59	case I915_TILING_NONE: return "";
  60	case I915_TILING_X: return " X";
  61	case I915_TILING_Y: return " Y";
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
  62	}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
  63}
  64
  65static const char *dirty_flag(int dirty)
 
  66{
  67	return dirty ? " dirty" : "";
 
 
 
 
  68}
  69
  70static const char *purgeable_flag(int purgeable)
 
  71{
  72	return purgeable ? " purgeable" : "";
  73}
  74
  75static bool __i915_error_ok(struct drm_i915_error_state_buf *e)
  76{
 
 
  77
  78	if (!e->err && WARN(e->bytes > (e->size - 1), "overflow")) {
  79		e->err = -ENOSPC;
  80		return false;
 
 
  81	}
  82
  83	if (e->bytes == e->size - 1 || e->err)
  84		return false;
  85
  86	return true;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
  87}
  88
  89static bool __i915_error_seek(struct drm_i915_error_state_buf *e,
  90			      unsigned len)
 
 
 
 
 
 
 
  91{
  92	if (e->pos + len <= e->start) {
  93		e->pos += len;
 
  94		return false;
  95	}
  96
  97	/* First vsnprintf needs to fit in its entirety for memmove */
  98	if (len >= e->size) {
  99		e->err = -EIO;
 
 
 100		return false;
 101	}
 102
 
 
 
 
 103	return true;
 104}
 105
 106static void __i915_error_advance(struct drm_i915_error_state_buf *e,
 107				 unsigned len)
 108{
 109	/* If this is first printf in this window, adjust it so that
 110	 * start position matches start of the buffer
 111	 */
 112
 113	if (e->pos < e->start) {
 114		const size_t off = e->start - e->pos;
 115
 116		/* Should not happen but be paranoid */
 117		if (off > len || e->bytes) {
 118			e->err = -EIO;
 119			return;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 120		}
 121
 122		memmove(e->buf, e->buf + off, len - off);
 123		e->bytes = len - off;
 124		e->pos = e->start;
 125		return;
 126	}
 
 
 127
 128	e->bytes += len;
 129	e->pos += len;
 130}
 131
 132static void i915_error_vprintf(struct drm_i915_error_state_buf *e,
 133			       const char *f, va_list args)
 134{
 135	unsigned len;
 136
 137	if (!__i915_error_ok(e))
 138		return;
 
 
 
 
 139
 140	/* Seek the first printf which is hits start position */
 141	if (e->pos < e->start) {
 142		va_list tmp;
 143
 144		va_copy(tmp, args);
 145		len = vsnprintf(NULL, 0, f, tmp);
 146		va_end(tmp);
 147
 148		if (!__i915_error_seek(e, len))
 149			return;
 150	}
 151
 152	len = vsnprintf(e->buf + e->bytes, e->size - e->bytes, f, args);
 153	if (len >= e->size - e->bytes)
 154		len = e->size - e->bytes - 1;
 
 155
 156	__i915_error_advance(e, len);
 
 
 
 157}
 158
 159static void i915_error_puts(struct drm_i915_error_state_buf *e,
 160			    const char *str)
 161{
 162	unsigned len;
 
 163
 164	if (!__i915_error_ok(e))
 165		return;
 
 
 
 
 
 166
 167	len = strlen(str);
 
 
 
 168
 169	/* Seek the first printf which is hits start position */
 170	if (e->pos < e->start) {
 171		if (!__i915_error_seek(e, len))
 172			return;
 173	}
 174
 175	if (len >= e->size - e->bytes)
 176		len = e->size - e->bytes - 1;
 177	memcpy(e->buf + e->bytes, str, len);
 178
 179	__i915_error_advance(e, len);
 
 
 180}
 181
 182#define err_printf(e, ...) i915_error_printf(e, __VA_ARGS__)
 183#define err_puts(e, s) i915_error_puts(e, s)
 
 
 184
 185static void print_error_buffers(struct drm_i915_error_state_buf *m,
 186				const char *name,
 187				struct drm_i915_error_buffer *err,
 188				int count)
 189{
 190	int i;
 
 
 
 
 191
 192	err_printf(m, "  %s [%d]:\n", name, count);
 
 
 193
 194	while (count--) {
 195		err_printf(m, "    %08x_%08x %8u %02x %02x [ ",
 196			   upper_32_bits(err->gtt_offset),
 197			   lower_32_bits(err->gtt_offset),
 198			   err->size,
 199			   err->read_domains,
 200			   err->write_domain);
 201		for (i = 0; i < I915_NUM_RINGS; i++)
 202			err_printf(m, "%02x ", err->rseqno[i]);
 203
 204		err_printf(m, "] %02x", err->wseqno);
 205		err_puts(m, pin_flag(err->pinned));
 206		err_puts(m, tiling_flag(err->tiling));
 207		err_puts(m, dirty_flag(err->dirty));
 208		err_puts(m, purgeable_flag(err->purgeable));
 209		err_puts(m, err->userptr ? " userptr" : "");
 210		err_puts(m, err->ring != -1 ? " " : "");
 211		err_puts(m, ring_str(err->ring));
 212		err_puts(m, i915_cache_level_str(m->i915, err->cache_level));
 213
 214		if (err->name)
 215			err_printf(m, " (name: %d)", err->name);
 216		if (err->fence_reg != I915_FENCE_REG_NONE)
 217			err_printf(m, " (fence: %d)", err->fence_reg);
 218
 219		err_puts(m, "\n");
 220		err++;
 221	}
 222}
 223
 224static const char *hangcheck_action_to_str(enum intel_ring_hangcheck_action a)
 225{
 226	switch (a) {
 227	case HANGCHECK_IDLE:
 228		return "idle";
 229	case HANGCHECK_WAIT:
 230		return "wait";
 231	case HANGCHECK_ACTIVE:
 232		return "active";
 233	case HANGCHECK_ACTIVE_LOOP:
 234		return "active (loop)";
 235	case HANGCHECK_KICK:
 236		return "kick";
 237	case HANGCHECK_HUNG:
 238		return "hung";
 239	}
 240
 241	return "unknown";
 242}
 243
 244static void i915_ring_error_state(struct drm_i915_error_state_buf *m,
 245				  struct drm_device *dev,
 246				  struct drm_i915_error_state *error,
 247				  int ring_idx)
 248{
 249	struct drm_i915_error_ring *ring = &error->ring[ring_idx];
 
 
 
 
 
 
 
 
 
 
 
 
 250
 251	if (!ring->valid)
 252		return;
 253
 254	err_printf(m, "%s command stream:\n", ring_str(ring_idx));
 255	err_printf(m, "  START: 0x%08x\n", ring->start);
 256	err_printf(m, "  HEAD:  0x%08x\n", ring->head);
 257	err_printf(m, "  TAIL:  0x%08x\n", ring->tail);
 258	err_printf(m, "  CTL:   0x%08x\n", ring->ctl);
 259	err_printf(m, "  HWS:   0x%08x\n", ring->hws);
 260	err_printf(m, "  ACTHD: 0x%08x %08x\n", (u32)(ring->acthd>>32), (u32)ring->acthd);
 261	err_printf(m, "  IPEIR: 0x%08x\n", ring->ipeir);
 262	err_printf(m, "  IPEHR: 0x%08x\n", ring->ipehr);
 263	err_printf(m, "  INSTDONE: 0x%08x\n", ring->instdone);
 264	if (INTEL_INFO(dev)->gen >= 4) {
 265		err_printf(m, "  BBADDR: 0x%08x %08x\n", (u32)(ring->bbaddr>>32), (u32)ring->bbaddr);
 266		err_printf(m, "  BB_STATE: 0x%08x\n", ring->bbstate);
 267		err_printf(m, "  INSTPS: 0x%08x\n", ring->instps);
 268	}
 269	err_printf(m, "  INSTPM: 0x%08x\n", ring->instpm);
 270	err_printf(m, "  FADDR: 0x%08x %08x\n", upper_32_bits(ring->faddr),
 271		   lower_32_bits(ring->faddr));
 272	if (INTEL_INFO(dev)->gen >= 6) {
 273		err_printf(m, "  RC PSMI: 0x%08x\n", ring->rc_psmi);
 274		err_printf(m, "  FAULT_REG: 0x%08x\n", ring->fault_reg);
 275		err_printf(m, "  SYNC_0: 0x%08x [last synced 0x%08x]\n",
 276			   ring->semaphore_mboxes[0],
 277			   ring->semaphore_seqno[0]);
 278		err_printf(m, "  SYNC_1: 0x%08x [last synced 0x%08x]\n",
 279			   ring->semaphore_mboxes[1],
 280			   ring->semaphore_seqno[1]);
 281		if (HAS_VEBOX(dev)) {
 282			err_printf(m, "  SYNC_2: 0x%08x [last synced 0x%08x]\n",
 283				   ring->semaphore_mboxes[2],
 284				   ring->semaphore_seqno[2]);
 285		}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 286	}
 287	if (USES_PPGTT(dev)) {
 288		err_printf(m, "  GFX_MODE: 0x%08x\n", ring->vm_info.gfx_mode);
 289
 290		if (INTEL_INFO(dev)->gen >= 8) {
 291			int i;
 292			for (i = 0; i < 4; i++)
 293				err_printf(m, "  PDP%d: 0x%016llx\n",
 294					   i, ring->vm_info.pdp[i]);
 295		} else {
 296			err_printf(m, "  PP_DIR_BASE: 0x%08x\n",
 297				   ring->vm_info.pp_dir_base);
 298		}
 299	}
 300	err_printf(m, "  seqno: 0x%08x\n", ring->seqno);
 301	err_printf(m, "  waiting: %s\n", yesno(ring->waiting));
 302	err_printf(m, "  ring->head: 0x%08x\n", ring->cpu_ring_head);
 303	err_printf(m, "  ring->tail: 0x%08x\n", ring->cpu_ring_tail);
 304	err_printf(m, "  hangcheck: %s [%d]\n",
 305		   hangcheck_action_to_str(ring->hangcheck_action),
 306		   ring->hangcheck_score);
 
 
 
 
 
 
 
 307}
 308
 309void i915_error_printf(struct drm_i915_error_state_buf *e, const char *f, ...)
 310{
 311	va_list args;
 312
 313	va_start(args, f);
 314	i915_error_vprintf(e, f, args);
 315	va_end(args);
 316}
 317
 318static void print_error_obj(struct drm_i915_error_state_buf *m,
 319			    struct drm_i915_error_object *obj)
 
 
 320{
 321	int page, offset, elt;
 
 322
 323	for (page = offset = 0; page < obj->page_count; page++) {
 324		for (elt = 0; elt < PAGE_SIZE/4; elt++) {
 325			err_printf(m, "%08x :  %08x\n", offset,
 326				   obj->pages[page][elt]);
 327			offset += 4;
 328		}
 
 
 329	}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 330}
 331
 332int i915_error_state_to_str(struct drm_i915_error_state_buf *m,
 333			    const struct i915_error_state_file_priv *error_priv)
 
 
 334{
 335	struct drm_device *dev = error_priv->dev;
 336	struct drm_i915_private *dev_priv = dev->dev_private;
 337	struct drm_i915_error_state *error = error_priv->error;
 338	struct drm_i915_error_object *obj;
 339	int i, j, offset, elt;
 340	int max_hangcheck_score;
 341
 342	if (!error) {
 343		err_printf(m, "no error state collected\n");
 344		goto out;
 345	}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 346
 347	err_printf(m, "%s\n", error->error_msg);
 348	err_printf(m, "Time: %ld s %ld us\n", error->time.tv_sec,
 349		   error->time.tv_usec);
 350	err_printf(m, "Kernel: " UTS_RELEASE "\n");
 351	max_hangcheck_score = 0;
 352	for (i = 0; i < ARRAY_SIZE(error->ring); i++) {
 353		if (error->ring[i].hangcheck_score > max_hangcheck_score)
 354			max_hangcheck_score = error->ring[i].hangcheck_score;
 355	}
 356	for (i = 0; i < ARRAY_SIZE(error->ring); i++) {
 357		if (error->ring[i].hangcheck_score == max_hangcheck_score &&
 358		    error->ring[i].pid != -1) {
 359			err_printf(m, "Active process (on ring %s): %s [%d]\n",
 360				   ring_str(i),
 361				   error->ring[i].comm,
 362				   error->ring[i].pid);
 363		}
 
 
 
 
 364	}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 365	err_printf(m, "Reset count: %u\n", error->reset_count);
 366	err_printf(m, "Suspend count: %u\n", error->suspend_count);
 367	err_printf(m, "PCI ID: 0x%04x\n", dev->pdev->device);
 368	err_printf(m, "PCI Revision: 0x%02x\n", dev->pdev->revision);
 369	err_printf(m, "PCI Subsystem: %04x:%04x\n",
 370		   dev->pdev->subsystem_vendor,
 371		   dev->pdev->subsystem_device);
 
 372	err_printf(m, "IOMMU enabled?: %d\n", error->iommu);
 373
 374	if (HAS_CSR(dev)) {
 375		struct intel_csr *csr = &dev_priv->csr;
 376
 377		err_printf(m, "DMC loaded: %s\n",
 378			   yesno(csr->dmc_payload != NULL));
 379		err_printf(m, "DMC fw version: %d.%d\n",
 380			   CSR_VERSION_MAJOR(csr->version),
 381			   CSR_VERSION_MINOR(csr->version));
 382	}
 383
 
 
 
 384	err_printf(m, "EIR: 0x%08x\n", error->eir);
 385	err_printf(m, "IER: 0x%08x\n", error->ier);
 386	if (INTEL_INFO(dev)->gen >= 8) {
 387		for (i = 0; i < 4; i++)
 388			err_printf(m, "GTIER gt %d: 0x%08x\n", i,
 389				   error->gtier[i]);
 390	} else if (HAS_PCH_SPLIT(dev) || IS_VALLEYVIEW(dev))
 391		err_printf(m, "GTIER: 0x%08x\n", error->gtier[0]);
 392	err_printf(m, "PGTBL_ER: 0x%08x\n", error->pgtbl_er);
 393	err_printf(m, "FORCEWAKE: 0x%08x\n", error->forcewake);
 394	err_printf(m, "DERRMR: 0x%08x\n", error->derrmr);
 395	err_printf(m, "CCID: 0x%08x\n", error->ccid);
 396	err_printf(m, "Missed interrupts: 0x%08lx\n", dev_priv->gpu_error.missed_irq_rings);
 397
 398	for (i = 0; i < dev_priv->num_fence_regs; i++)
 399		err_printf(m, "  fence[%d] = %08llx\n", i, error->fence[i]);
 400
 401	for (i = 0; i < ARRAY_SIZE(error->extra_instdone); i++)
 402		err_printf(m, "  INSTDONE_%d: 0x%08x\n", i,
 403			   error->extra_instdone[i]);
 404
 405	if (INTEL_INFO(dev)->gen >= 6) {
 406		err_printf(m, "ERROR: 0x%08x\n", error->error);
 407
 408		if (INTEL_INFO(dev)->gen >= 8)
 409			err_printf(m, "FAULT_TLB_DATA: 0x%08x 0x%08x\n",
 410				   error->fault_data1, error->fault_data0);
 411
 412		err_printf(m, "DONE_REG: 0x%08x\n", error->done_reg);
 413	}
 414
 415	if (INTEL_INFO(dev)->gen == 7)
 416		err_printf(m, "ERR_INT: 0x%08x\n", error->err_int);
 417
 418	for (i = 0; i < ARRAY_SIZE(error->ring); i++)
 419		i915_ring_error_state(m, dev, error, i);
 420
 421	for (i = 0; i < error->vm_count; i++) {
 422		err_printf(m, "vm[%d]\n", i);
 423
 424		print_error_buffers(m, "Active",
 425				    error->active_bo[i],
 426				    error->active_bo_count[i]);
 427
 428		print_error_buffers(m, "Pinned",
 429				    error->pinned_bo[i],
 430				    error->pinned_bo_count[i]);
 431	}
 432
 433	for (i = 0; i < ARRAY_SIZE(error->ring); i++) {
 434		obj = error->ring[i].batchbuffer;
 435		if (obj) {
 436			err_puts(m, dev_priv->ring[i].name);
 437			if (error->ring[i].pid != -1)
 438				err_printf(m, " (submitted by %s [%d])",
 439					   error->ring[i].comm,
 440					   error->ring[i].pid);
 441			err_printf(m, " --- gtt_offset = 0x%08x %08x\n",
 442				   upper_32_bits(obj->gtt_offset),
 443				   lower_32_bits(obj->gtt_offset));
 444			print_error_obj(m, obj);
 445		}
 446
 447		obj = error->ring[i].wa_batchbuffer;
 448		if (obj) {
 449			err_printf(m, "%s (w/a) --- gtt_offset = 0x%08x\n",
 450				   dev_priv->ring[i].name,
 451				   lower_32_bits(obj->gtt_offset));
 452			print_error_obj(m, obj);
 453		}
 454
 455		if (error->ring[i].num_requests) {
 456			err_printf(m, "%s --- %d requests\n",
 457				   dev_priv->ring[i].name,
 458				   error->ring[i].num_requests);
 459			for (j = 0; j < error->ring[i].num_requests; j++) {
 460				err_printf(m, "  seqno 0x%08x, emitted %ld, tail 0x%08x\n",
 461					   error->ring[i].requests[j].seqno,
 462					   error->ring[i].requests[j].jiffies,
 463					   error->ring[i].requests[j].tail);
 464			}
 465		}
 466
 467		if ((obj = error->ring[i].ringbuffer)) {
 468			err_printf(m, "%s --- ringbuffer = 0x%08x\n",
 469				   dev_priv->ring[i].name,
 470				   lower_32_bits(obj->gtt_offset));
 471			print_error_obj(m, obj);
 472		}
 473
 474		if ((obj = error->ring[i].hws_page)) {
 475			u64 hws_offset = obj->gtt_offset;
 476			u32 *hws_page = &obj->pages[0][0];
 477
 478			if (i915.enable_execlists) {
 479				hws_offset += LRC_PPHWSP_PN * PAGE_SIZE;
 480				hws_page = &obj->pages[LRC_PPHWSP_PN][0];
 481			}
 482			err_printf(m, "%s --- HW Status = 0x%08llx\n",
 483				   dev_priv->ring[i].name, hws_offset);
 484			offset = 0;
 485			for (elt = 0; elt < PAGE_SIZE/16; elt += 4) {
 486				err_printf(m, "[%04x] %08x %08x %08x %08x\n",
 487					   offset,
 488					   hws_page[elt],
 489					   hws_page[elt+1],
 490					   hws_page[elt+2],
 491					   hws_page[elt+3]);
 492					offset += 16;
 493			}
 494		}
 495
 496		if ((obj = error->ring[i].ctx)) {
 497			err_printf(m, "%s --- HW Context = 0x%08x\n",
 498				   dev_priv->ring[i].name,
 499				   lower_32_bits(obj->gtt_offset));
 500			print_error_obj(m, obj);
 501		}
 502	}
 503
 504	if ((obj = error->semaphore_obj)) {
 505		err_printf(m, "Semaphore page = 0x%08x\n",
 506			   lower_32_bits(obj->gtt_offset));
 507		for (elt = 0; elt < PAGE_SIZE/16; elt += 4) {
 508			err_printf(m, "[%04x] %08x %08x %08x %08x\n",
 509				   elt * 4,
 510				   obj->pages[0][elt],
 511				   obj->pages[0][elt+1],
 512				   obj->pages[0][elt+2],
 513				   obj->pages[0][elt+3]);
 514		}
 515	}
 516
 517	if (error->overlay)
 518		intel_overlay_print_error_state(m, error->overlay);
 519
 520	if (error->display)
 521		intel_display_print_error_state(m, dev, error->display);
 522
 523out:
 524	if (m->bytes == 0 && m->err)
 525		return m->err;
 526
 527	return 0;
 
 
 
 528}
 529
 530int i915_error_state_buf_init(struct drm_i915_error_state_buf *ebuf,
 531			      struct drm_i915_private *i915,
 532			      size_t count, loff_t pos)
 533{
 534	memset(ebuf, 0, sizeof(*ebuf));
 535	ebuf->i915 = i915;
 536
 537	/* We need to have enough room to store any i915_error_state printf
 538	 * so that we can move it to start position.
 539	 */
 540	ebuf->size = count + 1 > PAGE_SIZE ? count + 1 : PAGE_SIZE;
 541	ebuf->buf = kmalloc(ebuf->size,
 542				GFP_TEMPORARY | __GFP_NORETRY | __GFP_NOWARN);
 543
 544	if (ebuf->buf == NULL) {
 545		ebuf->size = PAGE_SIZE;
 546		ebuf->buf = kmalloc(ebuf->size, GFP_TEMPORARY);
 547	}
 
 
 
 548
 549	if (ebuf->buf == NULL) {
 550		ebuf->size = 128;
 551		ebuf->buf = kmalloc(ebuf->size, GFP_TEMPORARY);
 
 552	}
 
 
 
 
 
 553
 554	if (ebuf->buf == NULL)
 555		return -ENOMEM;
 
 
 556
 557	ebuf->start = pos;
 
 558
 559	return 0;
 560}
 561
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 562static void i915_error_object_free(struct drm_i915_error_object *obj)
 563{
 564	int page;
 565
 566	if (obj == NULL)
 567		return;
 568
 569	for (page = 0; page < obj->page_count; page++)
 570		kfree(obj->pages[page]);
 571
 572	kfree(obj);
 573}
 574
 575static void i915_error_state_free(struct kref *error_ref)
 576{
 577	struct drm_i915_error_state *error = container_of(error_ref,
 578							  typeof(*error), ref);
 579	int i;
 580
 581	for (i = 0; i < ARRAY_SIZE(error->ring); i++) {
 582		i915_error_object_free(error->ring[i].batchbuffer);
 583		i915_error_object_free(error->ring[i].wa_batchbuffer);
 584		i915_error_object_free(error->ring[i].ringbuffer);
 585		i915_error_object_free(error->ring[i].hws_page);
 586		i915_error_object_free(error->ring[i].ctx);
 587		kfree(error->ring[i].requests);
 588	}
 589
 590	i915_error_object_free(error->semaphore_obj);
 591
 592	for (i = 0; i < error->vm_count; i++)
 593		kfree(error->active_bo[i]);
 594
 595	kfree(error->active_bo);
 596	kfree(error->active_bo_count);
 597	kfree(error->pinned_bo);
 598	kfree(error->pinned_bo_count);
 599	kfree(error->overlay);
 600	kfree(error->display);
 601	kfree(error);
 602}
 603
 604static struct drm_i915_error_object *
 605i915_error_object_create(struct drm_i915_private *dev_priv,
 606			 struct drm_i915_gem_object *src,
 607			 struct i915_address_space *vm)
 608{
 609	struct drm_i915_error_object *dst;
 610	struct i915_vma *vma = NULL;
 611	int num_pages;
 612	bool use_ggtt;
 613	int i = 0;
 614	u64 reloc_offset;
 615
 616	if (src == NULL || src->pages == NULL)
 617		return NULL;
 618
 619	num_pages = src->base.size >> PAGE_SHIFT;
 
 
 
 620
 621	dst = kmalloc(sizeof(*dst) + num_pages * sizeof(u32 *), GFP_ATOMIC);
 622	if (dst == NULL)
 623		return NULL;
 
 
 624
 625	if (i915_gem_obj_bound(src, vm))
 626		dst->gtt_offset = i915_gem_obj_offset(src, vm);
 627	else
 628		dst->gtt_offset = -1;
 629
 630	reloc_offset = dst->gtt_offset;
 631	if (i915_is_ggtt(vm))
 632		vma = i915_gem_obj_to_ggtt(src);
 633	use_ggtt = (src->cache_level == I915_CACHE_NONE &&
 634		   vma && (vma->bound & GLOBAL_BIND) &&
 635		   reloc_offset + num_pages * PAGE_SIZE <= dev_priv->gtt.mappable_end);
 636
 637	/* Cannot access stolen address directly, try to use the aperture */
 638	if (src->stolen) {
 639		use_ggtt = true;
 640
 641		if (!(vma && vma->bound & GLOBAL_BIND))
 642			goto unwind;
 
 
 
 
 643
 644		reloc_offset = i915_gem_obj_ggtt_offset(src);
 645		if (reloc_offset + num_pages * PAGE_SIZE > dev_priv->gtt.mappable_end)
 646			goto unwind;
 647	}
 648
 649	/* Cannot access snooped pages through the aperture */
 650	if (use_ggtt && src->cache_level != I915_CACHE_NONE && !HAS_LLC(dev_priv->dev))
 651		goto unwind;
 652
 653	dst->page_count = num_pages;
 654	while (num_pages--) {
 655		unsigned long flags;
 656		void *d;
 657
 658		d = kmalloc(PAGE_SIZE, GFP_ATOMIC);
 659		if (d == NULL)
 660			goto unwind;
 661
 662		local_irq_save(flags);
 663		if (use_ggtt) {
 664			void __iomem *s;
 665
 666			/* Simply ignore tiling or any overlapping fence.
 667			 * It's part of the error state, and this hopefully
 668			 * captures what the GPU read.
 669			 */
 670
 671			s = io_mapping_map_atomic_wc(dev_priv->gtt.mappable,
 672						     reloc_offset);
 673			memcpy_fromio(d, s, PAGE_SIZE);
 674			io_mapping_unmap_atomic(s);
 675		} else {
 676			struct page *page;
 677			void *s;
 678
 679			page = i915_gem_object_get_page(src, i);
 680
 681			drm_clflush_pages(&page, 1);
 682
 683			s = kmap_atomic(page);
 684			memcpy(d, s, PAGE_SIZE);
 685			kunmap_atomic(s);
 686
 687			drm_clflush_pages(&page, 1);
 688		}
 689		local_irq_restore(flags);
 690
 691		dst->pages[i++] = d;
 692		reloc_offset += PAGE_SIZE;
 693	}
 694
 695	return dst;
 
 696
 697unwind:
 698	while (i--)
 699		kfree(dst->pages[i]);
 700	kfree(dst);
 701	return NULL;
 702}
 703#define i915_error_ggtt_object_create(dev_priv, src) \
 704	i915_error_object_create((dev_priv), (src), &(dev_priv)->gtt.base)
 705
 706static void capture_bo(struct drm_i915_error_buffer *err,
 707		       struct i915_vma *vma)
 
 
 708{
 709	struct drm_i915_gem_object *obj = vma->obj;
 710	int i;
 
 
 
 
 
 711
 712	err->size = obj->base.size;
 713	err->name = obj->base.name;
 714	for (i = 0; i < I915_NUM_RINGS; i++)
 715		err->rseqno[i] = i915_gem_request_get_seqno(obj->last_read_req[i]);
 716	err->wseqno = i915_gem_request_get_seqno(obj->last_write_req);
 717	err->gtt_offset = vma->node.start;
 718	err->read_domains = obj->base.read_domains;
 719	err->write_domain = obj->base.write_domain;
 720	err->fence_reg = obj->fence_reg;
 721	err->pinned = 0;
 722	if (i915_gem_obj_is_pinned(obj))
 723		err->pinned = 1;
 724	err->tiling = obj->tiling_mode;
 725	err->dirty = obj->dirty;
 726	err->purgeable = obj->madv != I915_MADV_WILLNEED;
 727	err->userptr = obj->userptr.mm != NULL;
 728	err->ring = obj->last_write_req ?
 729			i915_gem_request_get_ring(obj->last_write_req)->id : -1;
 730	err->cache_level = obj->cache_level;
 731}
 732
 733static u32 capture_active_bo(struct drm_i915_error_buffer *err,
 734			     int count, struct list_head *head)
 735{
 736	struct i915_vma *vma;
 737	int i = 0;
 738
 739	list_for_each_entry(vma, head, vm_link) {
 740		capture_bo(err++, vma);
 741		if (++i == count)
 742			break;
 743	}
 744
 745	return i;
 746}
 747
 748static u32 capture_pinned_bo(struct drm_i915_error_buffer *err,
 749			     int count, struct list_head *head,
 750			     struct i915_address_space *vm)
 751{
 752	struct drm_i915_gem_object *obj;
 753	struct drm_i915_error_buffer * const first = err;
 754	struct drm_i915_error_buffer * const last = err + count;
 755
 756	list_for_each_entry(obj, head, global_list) {
 757		struct i915_vma *vma;
 
 
 758
 759		if (err == last)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 760			break;
 
 761
 762		list_for_each_entry(vma, &obj->vma_list, obj_link)
 763			if (vma->vm == vm && vma->pin_count > 0)
 764				capture_bo(err++, vma);
 
 
 765	}
 
 766
 767	return err - first;
 768}
 769
 770/* Generate a semi-unique error code. The code is not meant to have meaning, The
 
 771 * code's only purpose is to try to prevent false duplicated bug reports by
 772 * grossly estimating a GPU error state.
 773 *
 774 * TODO Ideally, hashing the batchbuffer would be a very nice way to determine
 775 * the hang if we could strip the GTT offset information from it.
 776 *
 777 * It's only a small step better than a random number in its current form.
 778 */
 779static uint32_t i915_error_generate_code(struct drm_i915_private *dev_priv,
 780					 struct drm_i915_error_state *error,
 781					 int *ring_id)
 782{
 783	uint32_t error_code = 0;
 784	int i;
 785
 786	/* IPEHR would be an ideal way to detect errors, as it's the gross
 
 787	 * measure of "the command that hung." However, has some very common
 788	 * synchronization commands which almost always appear in the case
 789	 * strictly a client bug. Use instdone to differentiate those some.
 790	 */
 791	for (i = 0; i < I915_NUM_RINGS; i++) {
 792		if (error->ring[i].hangcheck_action == HANGCHECK_HUNG) {
 793			if (ring_id)
 794				*ring_id = i;
 795
 796			return error->ring[i].ipehr ^ error->ring[i].instdone;
 797		}
 798	}
 799
 800	return error_code;
 801}
 802
 803static void i915_gem_record_fences(struct drm_device *dev,
 804				   struct drm_i915_error_state *error)
 805{
 806	struct drm_i915_private *dev_priv = dev->dev_private;
 807	int i;
 808
 809	if (IS_GEN3(dev) || IS_GEN2(dev)) {
 810		for (i = 0; i < dev_priv->num_fence_regs; i++)
 811			error->fence[i] = I915_READ(FENCE_REG(i));
 812	} else if (IS_GEN5(dev) || IS_GEN4(dev)) {
 813		for (i = 0; i < dev_priv->num_fence_regs; i++)
 814			error->fence[i] = I915_READ64(FENCE_REG_965_LO(i));
 815	} else if (INTEL_INFO(dev)->gen >= 6) {
 816		for (i = 0; i < dev_priv->num_fence_regs; i++)
 817			error->fence[i] = I915_READ64(FENCE_REG_GEN6_LO(i));
 818	}
 819}
 820
 821
 822static void gen8_record_semaphore_state(struct drm_i915_private *dev_priv,
 823					struct drm_i915_error_state *error,
 824					struct intel_engine_cs *ring,
 825					struct drm_i915_error_ring *ering)
 826{
 827	struct intel_engine_cs *to;
 
 828	int i;
 829
 830	if (!i915_semaphore_is_enabled(dev_priv->dev))
 831		return;
 832
 833	if (!error->semaphore_obj)
 834		error->semaphore_obj =
 835			i915_error_ggtt_object_create(dev_priv,
 836						      dev_priv->semaphore_obj);
 837
 838	for_each_ring(to, dev_priv, i) {
 839		int idx;
 840		u16 signal_offset;
 841		u32 *tmp;
 842
 843		if (ring == to)
 844			continue;
 845
 846		signal_offset = (GEN8_SIGNAL_OFFSET(ring, i) & (PAGE_SIZE - 1))
 847				/ 4;
 848		tmp = error->semaphore_obj->pages[0];
 849		idx = intel_ring_sync_index(ring, to);
 850
 851		ering->semaphore_mboxes[idx] = tmp[signal_offset];
 852		ering->semaphore_seqno[idx] = ring->semaphore.sync_seqno[idx];
 853	}
 
 854}
 855
 856static void gen6_record_semaphore_state(struct drm_i915_private *dev_priv,
 857					struct intel_engine_cs *ring,
 858					struct drm_i915_error_ring *ering)
 859{
 860	ering->semaphore_mboxes[0] = I915_READ(RING_SYNC_0(ring->mmio_base));
 861	ering->semaphore_mboxes[1] = I915_READ(RING_SYNC_1(ring->mmio_base));
 862	ering->semaphore_seqno[0] = ring->semaphore.sync_seqno[0];
 863	ering->semaphore_seqno[1] = ring->semaphore.sync_seqno[1];
 864
 865	if (HAS_VEBOX(dev_priv->dev)) {
 866		ering->semaphore_mboxes[2] =
 867			I915_READ(RING_SYNC_2(ring->mmio_base));
 868		ering->semaphore_seqno[2] = ring->semaphore.sync_seqno[2];
 869	}
 870}
 871
 872static void i915_record_ring_state(struct drm_device *dev,
 873				   struct drm_i915_error_state *error,
 874				   struct intel_engine_cs *ring,
 875				   struct drm_i915_error_ring *ering)
 876{
 877	struct drm_i915_private *dev_priv = dev->dev_private;
 878
 879	if (INTEL_INFO(dev)->gen >= 6) {
 880		ering->rc_psmi = I915_READ(RING_PSMI_CTL(ring->mmio_base));
 881		ering->fault_reg = I915_READ(RING_FAULT_REG(ring));
 882		if (INTEL_INFO(dev)->gen >= 8)
 883			gen8_record_semaphore_state(dev_priv, error, ring, ering);
 884		else
 885			gen6_record_semaphore_state(dev_priv, ring, ering);
 886	}
 887
 888	if (INTEL_INFO(dev)->gen >= 4) {
 889		ering->faddr = I915_READ(RING_DMA_FADD(ring->mmio_base));
 890		ering->ipeir = I915_READ(RING_IPEIR(ring->mmio_base));
 891		ering->ipehr = I915_READ(RING_IPEHR(ring->mmio_base));
 892		ering->instdone = I915_READ(RING_INSTDONE(ring->mmio_base));
 893		ering->instps = I915_READ(RING_INSTPS(ring->mmio_base));
 894		ering->bbaddr = I915_READ(RING_BBADDR(ring->mmio_base));
 895		if (INTEL_INFO(dev)->gen >= 8) {
 896			ering->faddr |= (u64) I915_READ(RING_DMA_FADD_UDW(ring->mmio_base)) << 32;
 897			ering->bbaddr |= (u64) I915_READ(RING_BBADDR_UDW(ring->mmio_base)) << 32;
 898		}
 899		ering->bbstate = I915_READ(RING_BBSTATE(ring->mmio_base));
 900	} else {
 901		ering->faddr = I915_READ(DMA_FADD_I8XX);
 902		ering->ipeir = I915_READ(IPEIR);
 903		ering->ipehr = I915_READ(IPEHR);
 904		ering->instdone = I915_READ(GEN2_INSTDONE);
 905	}
 906
 907	ering->waiting = waitqueue_active(&ring->irq_queue);
 908	ering->instpm = I915_READ(RING_INSTPM(ring->mmio_base));
 909	ering->seqno = ring->get_seqno(ring, false);
 910	ering->acthd = intel_ring_get_active_head(ring);
 911	ering->start = I915_READ_START(ring);
 912	ering->head = I915_READ_HEAD(ring);
 913	ering->tail = I915_READ_TAIL(ring);
 914	ering->ctl = I915_READ_CTL(ring);
 
 915
 916	if (I915_NEED_GFX_HWS(dev)) {
 917		i915_reg_t mmio;
 918
 919		if (IS_GEN7(dev)) {
 920			switch (ring->id) {
 921			default:
 922			case RCS:
 
 
 923				mmio = RENDER_HWS_PGA_GEN7;
 924				break;
 925			case BCS:
 926				mmio = BLT_HWS_PGA_GEN7;
 927				break;
 928			case VCS:
 929				mmio = BSD_HWS_PGA_GEN7;
 930				break;
 931			case VECS:
 932				mmio = VEBOX_HWS_PGA_GEN7;
 933				break;
 934			}
 935		} else if (IS_GEN6(ring->dev)) {
 936			mmio = RING_HWS_PGA_GEN6(ring->mmio_base);
 937		} else {
 938			/* XXX: gen8 returns to sanity */
 939			mmio = RING_HWS_PGA(ring->mmio_base);
 940		}
 941
 942		ering->hws = I915_READ(mmio);
 943	}
 944
 945	ering->hangcheck_score = ring->hangcheck.score;
 946	ering->hangcheck_action = ring->hangcheck.action;
 
 
 
 947
 948	if (USES_PPGTT(dev)) {
 949		int i;
 950
 951		ering->vm_info.gfx_mode = I915_READ(RING_MODE_GEN7(ring));
 
 
 
 
 
 
 
 
 
 952
 953		if (IS_GEN6(dev))
 954			ering->vm_info.pp_dir_base =
 955				I915_READ(RING_PP_DIR_BASE_READ(ring));
 956		else if (IS_GEN7(dev))
 957			ering->vm_info.pp_dir_base =
 958				I915_READ(RING_PP_DIR_BASE(ring));
 959		else if (INTEL_INFO(dev)->gen >= 8)
 960			for (i = 0; i < 4; i++) {
 961				ering->vm_info.pdp[i] =
 962					I915_READ(GEN8_RING_PDP_UDW(ring, i));
 963				ering->vm_info.pdp[i] <<= 32;
 964				ering->vm_info.pdp[i] |=
 965					I915_READ(GEN8_RING_PDP_LDW(ring, i));
 966			}
 
 967	}
 968}
 969
 970
 971static void i915_gem_record_active_context(struct intel_engine_cs *ring,
 972					   struct drm_i915_error_state *error,
 973					   struct drm_i915_error_ring *ering)
 974{
 975	struct drm_i915_private *dev_priv = ring->dev->dev_private;
 976	struct drm_i915_gem_object *obj;
 977
 978	/* Currently render ring is the only HW context user */
 979	if (ring->id != RCS || !error->ccid)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 980		return;
 981
 982	list_for_each_entry(obj, &dev_priv->mm.bound_list, global_list) {
 983		if (!i915_gem_obj_ggtt_bound(obj))
 984			continue;
 985
 986		if ((error->ccid & PAGE_MASK) == i915_gem_obj_ggtt_offset(obj)) {
 987			ering->ctx = i915_error_ggtt_object_create(dev_priv, obj);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 988			break;
 989		}
 
 
 990	}
 
 991}
 992
 993static void i915_gem_record_rings(struct drm_device *dev,
 994				  struct drm_i915_error_state *error)
 995{
 996	struct drm_i915_private *dev_priv = dev->dev_private;
 997	struct drm_i915_gem_request *request;
 998	int i, count;
 999
1000	for (i = 0; i < I915_NUM_RINGS; i++) {
1001		struct intel_engine_cs *ring = &dev_priv->ring[i];
1002		struct intel_ringbuffer *rbuf;
1003
1004		error->ring[i].pid = -1;
 
1005
1006		if (ring->dev == NULL)
1007			continue;
 
 
1008
1009		error->ring[i].valid = true;
 
1010
1011		i915_record_ring_state(dev, error, ring, &error->ring[i]);
 
 
 
 
 
 
 
1012
1013		request = i915_gem_find_active_request(ring);
1014		if (request) {
1015			struct i915_address_space *vm;
 
1016
1017			vm = request->ctx && request->ctx->ppgtt ?
1018				&request->ctx->ppgtt->base :
1019				&dev_priv->gtt.base;
1020
1021			/* We need to copy these to an anonymous buffer
1022			 * as the simplest method to avoid being overwritten
1023			 * by userspace.
1024			 */
1025			error->ring[i].batchbuffer =
1026				i915_error_object_create(dev_priv,
1027							 request->batch_obj,
1028							 vm);
1029
1030			if (HAS_BROKEN_CS_TLB(dev_priv->dev))
1031				error->ring[i].wa_batchbuffer =
1032					i915_error_ggtt_object_create(dev_priv,
1033							     ring->scratch.obj);
1034
1035			if (request->pid) {
1036				struct task_struct *task;
1037
1038				rcu_read_lock();
1039				task = pid_task(request->pid, PIDTYPE_PID);
1040				if (task) {
1041					strcpy(error->ring[i].comm, task->comm);
1042					error->ring[i].pid = task->pid;
1043				}
1044				rcu_read_unlock();
1045			}
1046		}
1047
1048		if (i915.enable_execlists) {
1049			/* TODO: This is only a small fix to keep basic error
1050			 * capture working, but we need to add more information
1051			 * for it to be useful (e.g. dump the context being
1052			 * executed).
1053			 */
1054			if (request)
1055				rbuf = request->ctx->engine[ring->id].ringbuf;
1056			else
1057				rbuf = dev_priv->kernel_context->engine[ring->id].ringbuf;
1058		} else
1059			rbuf = ring->buffer;
1060
1061		error->ring[i].cpu_ring_head = rbuf->head;
1062		error->ring[i].cpu_ring_tail = rbuf->tail;
1063
1064		error->ring[i].ringbuffer =
1065			i915_error_ggtt_object_create(dev_priv, rbuf->obj);
1066
1067		error->ring[i].hws_page =
1068			i915_error_ggtt_object_create(dev_priv, ring->status_page.obj);
1069
1070		i915_gem_record_active_context(ring, error, &error->ring[i]);
1071
1072		count = 0;
1073		list_for_each_entry(request, &ring->request_list, list)
1074			count++;
1075
1076		error->ring[i].num_requests = count;
1077		error->ring[i].requests =
1078			kcalloc(count, sizeof(*error->ring[i].requests),
1079				GFP_ATOMIC);
1080		if (error->ring[i].requests == NULL) {
1081			error->ring[i].num_requests = 0;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1082			continue;
1083		}
1084
1085		count = 0;
1086		list_for_each_entry(request, &ring->request_list, list) {
1087			struct drm_i915_error_request *erq;
1088
1089			if (count >= error->ring[i].num_requests) {
1090				/*
1091				 * If the ring request list was changed in
1092				 * between the point where the error request
1093				 * list was created and dimensioned and this
1094				 * point then just exit early to avoid crashes.
1095				 *
1096				 * We don't need to communicate that the
1097				 * request list changed state during error
1098				 * state capture and that the error state is
1099				 * slightly incorrect as a consequence since we
1100				 * are typically only interested in the request
1101				 * list state at the point of error state
1102				 * capture, not in any changes happening during
1103				 * the capture.
1104				 */
1105				break;
1106			}
 
 
 
 
 
 
 
 
 
 
1107
1108			erq = &error->ring[i].requests[count++];
1109			erq->seqno = request->seqno;
1110			erq->jiffies = request->emitted_jiffies;
1111			erq->tail = request->postfix;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1112		}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1113	}
 
 
1114}
1115
1116/* FIXME: Since pin count/bound list is global, we duplicate what we capture per
1117 * VM.
1118 */
1119static void i915_gem_capture_vm(struct drm_i915_private *dev_priv,
1120				struct drm_i915_error_state *error,
1121				struct i915_address_space *vm,
1122				const int ndx)
1123{
1124	struct drm_i915_error_buffer *active_bo = NULL, *pinned_bo = NULL;
1125	struct drm_i915_gem_object *obj;
1126	struct i915_vma *vma;
1127	int i;
1128
1129	i = 0;
1130	list_for_each_entry(vma, &vm->active_list, vm_link)
1131		i++;
1132	error->active_bo_count[ndx] = i;
1133
1134	list_for_each_entry(obj, &dev_priv->mm.bound_list, global_list) {
1135		list_for_each_entry(vma, &obj->vma_list, obj_link)
1136			if (vma->vm == vm && vma->pin_count > 0)
1137				i++;
1138	}
1139	error->pinned_bo_count[ndx] = i - error->active_bo_count[ndx];
1140
1141	if (i) {
1142		active_bo = kcalloc(i, sizeof(*active_bo), GFP_ATOMIC);
1143		if (active_bo)
1144			pinned_bo = active_bo + error->active_bo_count[ndx];
1145	}
1146
1147	if (active_bo)
1148		error->active_bo_count[ndx] =
1149			capture_active_bo(active_bo,
1150					  error->active_bo_count[ndx],
1151					  &vm->active_list);
1152
1153	if (pinned_bo)
1154		error->pinned_bo_count[ndx] =
1155			capture_pinned_bo(pinned_bo,
1156					  error->pinned_bo_count[ndx],
1157					  &dev_priv->mm.bound_list, vm);
1158	error->active_bo[ndx] = active_bo;
1159	error->pinned_bo[ndx] = pinned_bo;
1160}
1161
1162static void i915_gem_capture_buffers(struct drm_i915_private *dev_priv,
1163				     struct drm_i915_error_state *error)
1164{
1165	struct i915_address_space *vm;
1166	int cnt = 0, i = 0;
1167
1168	list_for_each_entry(vm, &dev_priv->vm_list, global_link)
1169		cnt++;
1170
1171	error->active_bo = kcalloc(cnt, sizeof(*error->active_bo), GFP_ATOMIC);
1172	error->pinned_bo = kcalloc(cnt, sizeof(*error->pinned_bo), GFP_ATOMIC);
1173	error->active_bo_count = kcalloc(cnt, sizeof(*error->active_bo_count),
1174					 GFP_ATOMIC);
1175	error->pinned_bo_count = kcalloc(cnt, sizeof(*error->pinned_bo_count),
1176					 GFP_ATOMIC);
1177
1178	if (error->active_bo == NULL ||
1179	    error->pinned_bo == NULL ||
1180	    error->active_bo_count == NULL ||
1181	    error->pinned_bo_count == NULL) {
1182		kfree(error->active_bo);
1183		kfree(error->active_bo_count);
1184		kfree(error->pinned_bo);
1185		kfree(error->pinned_bo_count);
1186
1187		error->active_bo = NULL;
1188		error->active_bo_count = NULL;
1189		error->pinned_bo = NULL;
1190		error->pinned_bo_count = NULL;
1191	} else {
1192		list_for_each_entry(vm, &dev_priv->vm_list, global_link)
1193			i915_gem_capture_vm(dev_priv, error, vm, i++);
1194
1195		error->vm_count = cnt;
1196	}
 
 
 
 
 
 
 
 
 
 
1197}
1198
1199/* Capture all registers which don't fit into another category. */
1200static void i915_capture_reg_state(struct drm_i915_private *dev_priv,
1201				   struct drm_i915_error_state *error)
1202{
1203	struct drm_device *dev = dev_priv->dev;
 
1204	int i;
1205
1206	/* General organization
1207	 * 1. Registers specific to a single generation
1208	 * 2. Registers which belong to multiple generations
1209	 * 3. Feature specific registers.
1210	 * 4. Everything else
1211	 * Please try to follow the order.
1212	 */
1213
1214	/* 1: Registers specific to a single generation */
1215	if (IS_VALLEYVIEW(dev)) {
1216		error->gtier[0] = I915_READ(GTIER);
1217		error->ier = I915_READ(VLV_IER);
1218		error->forcewake = I915_READ_FW(FORCEWAKE_VLV);
1219	}
1220
1221	if (IS_GEN7(dev))
1222		error->err_int = I915_READ(GEN7_ERR_INT);
1223
1224	if (INTEL_INFO(dev)->gen >= 8) {
1225		error->fault_data0 = I915_READ(GEN8_FAULT_TLB_DATA0);
1226		error->fault_data1 = I915_READ(GEN8_FAULT_TLB_DATA1);
1227	}
1228
1229	if (IS_GEN6(dev)) {
1230		error->forcewake = I915_READ_FW(FORCEWAKE);
1231		error->gab_ctl = I915_READ(GAB_CTL);
1232		error->gfx_mode = I915_READ(GFX_MODE);
 
 
 
 
 
 
 
1233	}
1234
1235	/* 2: Registers which belong to multiple generations */
1236	if (INTEL_INFO(dev)->gen >= 7)
1237		error->forcewake = I915_READ_FW(FORCEWAKE_MT);
1238
1239	if (INTEL_INFO(dev)->gen >= 6) {
1240		error->derrmr = I915_READ(DERRMR);
1241		error->error = I915_READ(ERROR_GEN6);
1242		error->done_reg = I915_READ(DONE_REG);
 
 
1243	}
1244
 
 
 
1245	/* 3: Feature specific registers */
1246	if (IS_GEN6(dev) || IS_GEN7(dev)) {
1247		error->gam_ecochk = I915_READ(GAM_ECOCHK);
1248		error->gac_eco = I915_READ(GAC_ECO_BITS);
1249	}
1250
1251	/* 4: Everything else */
1252	if (HAS_HW_CONTEXTS(dev))
1253		error->ccid = I915_READ(CCID);
1254
1255	if (INTEL_INFO(dev)->gen >= 8) {
1256		error->ier = I915_READ(GEN8_DE_MISC_IER);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1257		for (i = 0; i < 4; i++)
1258			error->gtier[i] = I915_READ(GEN8_GT_IER(i));
1259	} else if (HAS_PCH_SPLIT(dev)) {
1260		error->ier = I915_READ(DEIER);
1261		error->gtier[0] = I915_READ(GTIER);
1262	} else if (IS_GEN2(dev)) {
1263		error->ier = I915_READ16(IER);
1264	} else if (!IS_VALLEYVIEW(dev)) {
1265		error->ier = I915_READ(IER);
1266	}
1267	error->eir = I915_READ(EIR);
1268	error->pgtbl_er = I915_READ(PGTBL_ER);
1269
1270	i915_get_extra_instdone(dev, error->extra_instdone);
1271}
1272
1273static void i915_error_capture_msg(struct drm_device *dev,
1274				   struct drm_i915_error_state *error,
1275				   bool wedged,
1276				   const char *error_msg)
1277{
1278	struct drm_i915_private *dev_priv = dev->dev_private;
1279	u32 ecode;
1280	int ring_id = -1, len;
1281
1282	ecode = i915_error_generate_code(dev_priv, error, &ring_id);
1283
1284	len = scnprintf(error->error_msg, sizeof(error->error_msg),
1285			"GPU HANG: ecode %d:%d:0x%08x",
1286			INTEL_INFO(dev)->gen, ring_id, ecode);
1287
1288	if (ring_id != -1 && error->ring[ring_id].pid != -1)
 
1289		len += scnprintf(error->error_msg + len,
1290				 sizeof(error->error_msg) - len,
1291				 ", in %s [%d]",
1292				 error->ring[ring_id].comm,
1293				 error->ring[ring_id].pid);
 
 
 
 
 
1294
1295	scnprintf(error->error_msg + len, sizeof(error->error_msg) - len,
1296		  ", reason: %s, action: %s",
1297		  error_msg,
1298		  wedged ? "reset" : "continue");
1299}
1300
1301static void i915_capture_gen_state(struct drm_i915_private *dev_priv,
1302				   struct drm_i915_error_state *error)
1303{
 
 
 
 
 
 
1304	error->iommu = -1;
1305#ifdef CONFIG_INTEL_IOMMU
1306	error->iommu = intel_iommu_gfx_mapped;
1307#endif
1308	error->reset_count = i915_reset_count(&dev_priv->gpu_error);
1309	error->suspend_count = dev_priv->suspend_count;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1310}
1311
1312/**
1313 * i915_capture_error_state - capture an error record for later analysis
1314 * @dev: drm device
 
 
1315 *
1316 * Should be called when an error is detected (either a hang or an error
1317 * interrupt) to capture error state from the time of the error.  Fills
1318 * out a structure which becomes available in debugfs for user level tools
1319 * to pick up.
1320 */
1321void i915_capture_error_state(struct drm_device *dev, bool wedged,
1322			      const char *error_msg)
 
1323{
1324	static bool warned;
1325	struct drm_i915_private *dev_priv = dev->dev_private;
1326	struct drm_i915_error_state *error;
1327	unsigned long flags;
1328
1329	/* Account for pipe specific data like PIPE*STAT */
1330	error = kzalloc(sizeof(*error), GFP_ATOMIC);
1331	if (!error) {
1332		DRM_DEBUG_DRIVER("out of memory, not capturing error state\n");
1333		return;
1334	}
1335
1336	kref_init(&error->ref);
 
 
 
 
 
 
 
1337
1338	i915_capture_gen_state(dev_priv, error);
1339	i915_capture_reg_state(dev_priv, error);
1340	i915_gem_capture_buffers(dev_priv, error);
1341	i915_gem_record_fences(dev, error);
1342	i915_gem_record_rings(dev, error);
1343
1344	do_gettimeofday(&error->time);
1345
1346	error->overlay = intel_overlay_capture_error_state(dev);
1347	error->display = intel_display_capture_error_state(dev);
1348
1349	i915_error_capture_msg(dev, error, wedged, error_msg);
1350	DRM_INFO("%s\n", error->error_msg);
1351
1352	spin_lock_irqsave(&dev_priv->gpu_error.lock, flags);
1353	if (dev_priv->gpu_error.first_error == NULL) {
1354		dev_priv->gpu_error.first_error = error;
1355		error = NULL;
1356	}
1357	spin_unlock_irqrestore(&dev_priv->gpu_error.lock, flags);
1358
1359	if (error) {
1360		i915_error_state_free(&error->ref);
1361		return;
1362	}
1363
1364	if (!warned) {
1365		DRM_INFO("GPU hangs can indicate a bug anywhere in the entire gfx stack, including userspace.\n");
1366		DRM_INFO("Please file a _new_ bug report on bugs.freedesktop.org against DRI -> DRM/Intel\n");
1367		DRM_INFO("drm/i915 developers can then reassign to the right component if it's not a kernel issue.\n");
1368		DRM_INFO("The gpu crash dump is required to analyze gpu hangs, so please always attach it.\n");
1369		DRM_INFO("GPU crash dump saved to /sys/class/drm/card%d/error\n", dev->primary->index);
1370		warned = true;
 
1371	}
1372}
1373
1374void i915_error_state_get(struct drm_device *dev,
1375			  struct i915_error_state_file_priv *error_priv)
1376{
1377	struct drm_i915_private *dev_priv = dev->dev_private;
1378
1379	spin_lock_irq(&dev_priv->gpu_error.lock);
1380	error_priv->error = dev_priv->gpu_error.first_error;
1381	if (error_priv->error)
1382		kref_get(&error_priv->error->ref);
1383	spin_unlock_irq(&dev_priv->gpu_error.lock);
1384
1385}
 
 
 
 
1386
1387void i915_error_state_put(struct i915_error_state_file_priv *error_priv)
1388{
1389	if (error_priv->error)
1390		kref_put(&error_priv->error->ref, i915_error_state_free);
1391}
1392
1393void i915_destroy_error_state(struct drm_device *dev)
1394{
1395	struct drm_i915_private *dev_priv = dev->dev_private;
1396	struct drm_i915_error_state *error;
1397
1398	spin_lock_irq(&dev_priv->gpu_error.lock);
1399	error = dev_priv->gpu_error.first_error;
1400	dev_priv->gpu_error.first_error = NULL;
1401	spin_unlock_irq(&dev_priv->gpu_error.lock);
 
1402
1403	if (error)
1404		kref_put(&error->ref, i915_error_state_free);
1405}
1406
1407const char *i915_cache_level_str(struct drm_i915_private *i915, int type)
1408{
1409	switch (type) {
1410	case I915_CACHE_NONE: return " uncached";
1411	case I915_CACHE_LLC: return HAS_LLC(i915) ? " LLC" : " snooped";
1412	case I915_CACHE_L3_LLC: return " L3+LLC";
1413	case I915_CACHE_WT: return " WT";
1414	default: return "";
1415	}
1416}
1417
1418/* NB: please notice the memset */
1419void i915_get_extra_instdone(struct drm_device *dev, uint32_t *instdone)
1420{
1421	struct drm_i915_private *dev_priv = dev->dev_private;
1422	memset(instdone, 0, sizeof(*instdone) * I915_NUM_INSTDONE_REG);
1423
1424	if (IS_GEN2(dev) || IS_GEN3(dev))
1425		instdone[0] = I915_READ(GEN2_INSTDONE);
1426	else if (IS_GEN4(dev) || IS_GEN5(dev) || IS_GEN6(dev)) {
1427		instdone[0] = I915_READ(RING_INSTDONE(RENDER_RING_BASE));
1428		instdone[1] = I915_READ(GEN4_INSTDONE1);
1429	} else if (INTEL_INFO(dev)->gen >= 7) {
1430		instdone[0] = I915_READ(RING_INSTDONE(RENDER_RING_BASE));
1431		instdone[1] = I915_READ(GEN7_SC_INSTDONE);
1432		instdone[2] = I915_READ(GEN7_SAMPLER_INSTDONE);
1433		instdone[3] = I915_READ(GEN7_ROW_INSTDONE);
1434	}
1435}
v5.4
   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_overlay.h"
  41
  42#include "gem/i915_gem_context.h"
  43
  44#include "i915_drv.h"
  45#include "i915_gpu_error.h"
  46#include "i915_memcpy.h"
  47#include "i915_scatterlist.h"
  48#include "intel_csr.h"
  49
  50#define ALLOW_FAIL (GFP_KERNEL | __GFP_RETRY_MAYFAIL | __GFP_NOWARN)
  51#define ATOMIC_MAYFAIL (GFP_ATOMIC | __GFP_NOWARN)
  52
  53static void __sg_set_buf(struct scatterlist *sg,
  54			 void *addr, unsigned int len, loff_t it)
  55{
  56	sg->page_link = (unsigned long)virt_to_page(addr);
  57	sg->offset = offset_in_page(addr);
  58	sg->length = len;
  59	sg->dma_address = it;
 
 
 
 
  60}
  61
  62static bool __i915_error_grow(struct drm_i915_error_state_buf *e, size_t len)
  63{
  64	if (!len)
  65		return false;
  66
  67	if (e->bytes + len + 1 <= e->size)
  68		return true;
  69
  70	if (e->bytes) {
  71		__sg_set_buf(e->cur++, e->buf, e->bytes, e->iter);
  72		e->iter += e->bytes;
  73		e->buf = NULL;
  74		e->bytes = 0;
  75	}
  76
  77	if (e->cur == e->end) {
  78		struct scatterlist *sgl;
  79
  80		sgl = (typeof(sgl))__get_free_page(ALLOW_FAIL);
  81		if (!sgl) {
  82			e->err = -ENOMEM;
  83			return false;
  84		}
  85
  86		if (e->cur) {
  87			e->cur->offset = 0;
  88			e->cur->length = 0;
  89			e->cur->page_link =
  90				(unsigned long)sgl | SG_CHAIN;
  91		} else {
  92			e->sgl = sgl;
  93		}
  94
  95		e->cur = sgl;
  96		e->end = sgl + SG_MAX_SINGLE_ALLOC - 1;
  97	}
  98
  99	e->size = ALIGN(len + 1, SZ_64K);
 100	e->buf = kmalloc(e->size, ALLOW_FAIL);
 101	if (!e->buf) {
 102		e->size = PAGE_ALIGN(len + 1);
 103		e->buf = kmalloc(e->size, GFP_KERNEL);
 104	}
 105	if (!e->buf) {
 106		e->err = -ENOMEM;
 107		return false;
 108	}
 109
 110	return true;
 111}
 112
 113__printf(2, 0)
 114static void i915_error_vprintf(struct drm_i915_error_state_buf *e,
 115			       const char *fmt, va_list args)
 116{
 117	va_list ap;
 118	int len;
 119
 120	if (e->err)
 121		return;
 122
 123	va_copy(ap, args);
 124	len = vsnprintf(NULL, 0, fmt, ap);
 125	va_end(ap);
 126	if (len <= 0) {
 127		e->err = len;
 128		return;
 129	}
 130
 131	if (!__i915_error_grow(e, len))
 132		return;
 133
 134	GEM_BUG_ON(e->bytes >= e->size);
 135	len = vscnprintf(e->buf + e->bytes, e->size - e->bytes, fmt, args);
 136	if (len < 0) {
 137		e->err = len;
 138		return;
 139	}
 140	e->bytes += len;
 141}
 142
 143static void i915_error_puts(struct drm_i915_error_state_buf *e, const char *str)
 144{
 145	unsigned len;
 146
 147	if (e->err || !str)
 148		return;
 149
 150	len = strlen(str);
 151	if (!__i915_error_grow(e, len))
 152		return;
 153
 154	GEM_BUG_ON(e->bytes + len > e->size);
 155	memcpy(e->buf + e->bytes, str, len);
 156	e->bytes += len;
 157}
 158
 159#define err_printf(e, ...) i915_error_printf(e, __VA_ARGS__)
 160#define err_puts(e, s) i915_error_puts(e, s)
 161
 162static void __i915_printfn_error(struct drm_printer *p, struct va_format *vaf)
 163{
 164	i915_error_vprintf(p->arg, vaf->fmt, *vaf->va);
 165}
 166
 167static inline struct drm_printer
 168i915_error_printer(struct drm_i915_error_state_buf *e)
 169{
 170	struct drm_printer p = {
 171		.printfn = __i915_printfn_error,
 172		.arg = e,
 173	};
 174	return p;
 175}
 176
 177/* single threaded page allocator with a reserved stash for emergencies */
 178static void pool_fini(struct pagevec *pv)
 179{
 180	pagevec_release(pv);
 181}
 182
 183static int pool_refill(struct pagevec *pv, gfp_t gfp)
 184{
 185	while (pagevec_space(pv)) {
 186		struct page *p;
 187
 188		p = alloc_page(gfp);
 189		if (!p)
 190			return -ENOMEM;
 191
 192		pagevec_add(pv, p);
 193	}
 194
 195	return 0;
 196}
 197
 198static int pool_init(struct pagevec *pv, gfp_t gfp)
 199{
 200	int err;
 201
 202	pagevec_init(pv);
 203
 204	err = pool_refill(pv, gfp);
 205	if (err)
 206		pool_fini(pv);
 207
 208	return err;
 209}
 210
 211static void *pool_alloc(struct pagevec *pv, gfp_t gfp)
 212{
 213	struct page *p;
 214
 215	p = alloc_page(gfp);
 216	if (!p && pagevec_count(pv))
 217		p = pv->pages[--pv->nr];
 218
 219	return p ? page_address(p) : NULL;
 220}
 221
 222static void pool_free(struct pagevec *pv, void *addr)
 223{
 224	struct page *p = virt_to_page(addr);
 225
 226	if (pagevec_space(pv))
 227		pagevec_add(pv, p);
 228	else
 229		__free_page(p);
 230}
 231
 232#ifdef CONFIG_DRM_I915_COMPRESS_ERROR
 233
 234struct compress {
 235	struct pagevec pool;
 236	struct z_stream_s zstream;
 237	void *tmp;
 238};
 239
 240static bool compress_init(struct compress *c)
 241{
 242	struct z_stream_s *zstream = &c->zstream;
 243
 244	if (pool_init(&c->pool, ALLOW_FAIL))
 245		return false;
 
 246
 247	zstream->workspace =
 248		kmalloc(zlib_deflate_workspacesize(MAX_WBITS, MAX_MEM_LEVEL),
 249			ALLOW_FAIL);
 250	if (!zstream->workspace) {
 251		pool_fini(&c->pool);
 252		return false;
 253	}
 254
 255	c->tmp = NULL;
 256	if (i915_has_memcpy_from_wc())
 257		c->tmp = pool_alloc(&c->pool, ALLOW_FAIL);
 258
 259	return true;
 260}
 261
 262static bool compress_start(struct compress *c)
 
 263{
 264	struct z_stream_s *zstream = &c->zstream;
 265	void *workspace = zstream->workspace;
 
 266
 267	memset(zstream, 0, sizeof(*zstream));
 268	zstream->workspace = workspace;
 269
 270	return zlib_deflateInit(zstream, Z_DEFAULT_COMPRESSION) == Z_OK;
 271}
 272
 273static void *compress_next_page(struct compress *c,
 274				struct drm_i915_error_object *dst)
 275{
 276	void *page;
 277
 278	if (dst->page_count >= dst->num_pages)
 279		return ERR_PTR(-ENOSPC);
 280
 281	page = pool_alloc(&c->pool, ALLOW_FAIL);
 282	if (!page)
 283		return ERR_PTR(-ENOMEM);
 284
 285	return dst->pages[dst->page_count++] = page;
 286}
 287
 288static int compress_page(struct compress *c,
 289			 void *src,
 290			 struct drm_i915_error_object *dst)
 291{
 292	struct z_stream_s *zstream = &c->zstream;
 293
 294	zstream->next_in = src;
 295	if (c->tmp && i915_memcpy_from_wc(c->tmp, src, PAGE_SIZE))
 296		zstream->next_in = c->tmp;
 297	zstream->avail_in = PAGE_SIZE;
 298
 299	do {
 300		if (zstream->avail_out == 0) {
 301			zstream->next_out = compress_next_page(c, dst);
 302			if (IS_ERR(zstream->next_out))
 303				return PTR_ERR(zstream->next_out);
 304
 305			zstream->avail_out = PAGE_SIZE;
 306		}
 307
 308		if (zlib_deflate(zstream, Z_NO_FLUSH) != Z_OK)
 309			return -EIO;
 310	} while (zstream->avail_in);
 311
 312	/* Fallback to uncompressed if we increase size? */
 313	if (0 && zstream->total_out > zstream->total_in)
 314		return -E2BIG;
 315
 316	return 0;
 
 317}
 318
 319static int compress_flush(struct compress *c,
 320			  struct drm_i915_error_object *dst)
 321{
 322	struct z_stream_s *zstream = &c->zstream;
 323
 324	do {
 325		switch (zlib_deflate(zstream, Z_FINISH)) {
 326		case Z_OK: /* more space requested */
 327			zstream->next_out = compress_next_page(c, dst);
 328			if (IS_ERR(zstream->next_out))
 329				return PTR_ERR(zstream->next_out);
 330
 331			zstream->avail_out = PAGE_SIZE;
 332			break;
 
 
 
 
 
 333
 334		case Z_STREAM_END:
 335			goto end;
 
 336
 337		default: /* any error */
 338			return -EIO;
 339		}
 340	} while (1);
 341
 342end:
 343	memset(zstream->next_out, 0, zstream->avail_out);
 344	dst->unused = zstream->avail_out;
 345	return 0;
 346}
 347
 348static void compress_finish(struct compress *c)
 
 349{
 350	zlib_deflateEnd(&c->zstream);
 351}
 352
 353static void compress_fini(struct compress *c)
 354{
 355	kfree(c->zstream.workspace);
 356	if (c->tmp)
 357		pool_free(&c->pool, c->tmp);
 358	pool_fini(&c->pool);
 359}
 360
 361static void err_compression_marker(struct drm_i915_error_state_buf *m)
 362{
 363	err_puts(m, ":");
 364}
 365
 366#else
 
 
 
 
 367
 368struct compress {
 369	struct pagevec pool;
 370};
 371
 372static bool compress_init(struct compress *c)
 373{
 374	return pool_init(&c->pool, ALLOW_FAIL) == 0;
 375}
 376
 377static bool compress_start(struct compress *c)
 378{
 379	return true;
 380}
 381
 382static int compress_page(struct compress *c,
 383			 void *src,
 384			 struct drm_i915_error_object *dst)
 
 385{
 386	void *ptr;
 387
 388	ptr = pool_alloc(&c->pool, ALLOW_FAIL);
 389	if (!ptr)
 390		return -ENOMEM;
 391
 392	if (!i915_memcpy_from_wc(ptr, src, PAGE_SIZE))
 393		memcpy(ptr, src, PAGE_SIZE);
 394	dst->pages[dst->page_count++] = ptr;
 395
 396	return 0;
 397}
 398
 399static int compress_flush(struct compress *c,
 400			  struct drm_i915_error_object *dst)
 401{
 402	return 0;
 403}
 404
 405static void compress_finish(struct compress *c)
 406{
 407}
 408
 409static void compress_fini(struct compress *c)
 410{
 411	pool_fini(&c->pool);
 
 
 
 
 
 
 
 
 
 
 
 
 412}
 413
 414static void err_compression_marker(struct drm_i915_error_state_buf *m)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 415{
 416	err_puts(m, "~");
 417}
 418
 419#endif
 420
 421static void error_print_instdone(struct drm_i915_error_state_buf *m,
 422				 const struct drm_i915_error_engine *ee)
 423{
 424	int slice;
 425	int subslice;
 426
 427	err_printf(m, "  INSTDONE: 0x%08x\n",
 428		   ee->instdone.instdone);
 429
 430	if (ee->engine->class != RENDER_CLASS || INTEL_GEN(m->i915) <= 3)
 431		return;
 432
 433	err_printf(m, "  SC_INSTDONE: 0x%08x\n",
 434		   ee->instdone.slice_common);
 435
 436	if (INTEL_GEN(m->i915) <= 6)
 437		return;
 438
 439	for_each_instdone_slice_subslice(m->i915, slice, subslice)
 440		err_printf(m, "  SAMPLER_INSTDONE[%d][%d]: 0x%08x\n",
 441			   slice, subslice,
 442			   ee->instdone.sampler[slice][subslice]);
 443
 444	for_each_instdone_slice_subslice(m->i915, slice, subslice)
 445		err_printf(m, "  ROW_INSTDONE[%d][%d]: 0x%08x\n",
 446			   slice, subslice,
 447			   ee->instdone.row[slice][subslice]);
 448}
 449
 450static void error_print_request(struct drm_i915_error_state_buf *m,
 451				const char *prefix,
 452				const struct drm_i915_error_request *erq,
 453				const unsigned long epoch)
 454{
 455	if (!erq->seqno)
 456		return;
 457
 458	err_printf(m, "%s pid %d, seqno %8x:%08x%s%s, prio %d, emitted %dms, start %08x, head %08x, tail %08x\n",
 459		   prefix, erq->pid, erq->context, erq->seqno,
 460		   test_bit(DMA_FENCE_FLAG_SIGNALED_BIT,
 461			    &erq->flags) ? "!" : "",
 462		   test_bit(DMA_FENCE_FLAG_ENABLE_SIGNAL_BIT,
 463			    &erq->flags) ? "+" : "",
 464		   erq->sched_attr.priority,
 465		   jiffies_to_msecs(erq->jiffies - epoch),
 466		   erq->start, erq->head, erq->tail);
 467}
 468
 469static void error_print_context(struct drm_i915_error_state_buf *m,
 470				const char *header,
 471				const struct drm_i915_error_context *ctx)
 472{
 473	err_printf(m, "%s%s[%d] hw_id %d, prio %d, guilty %d active %d\n",
 474		   header, ctx->comm, ctx->pid, ctx->hw_id,
 475		   ctx->sched_attr.priority, ctx->guilty, ctx->active);
 476}
 477
 478static void error_print_engine(struct drm_i915_error_state_buf *m,
 479			       const struct drm_i915_error_engine *ee,
 480			       const unsigned long epoch)
 481{
 482	int n;
 483
 484	err_printf(m, "%s command stream:\n", ee->engine->name);
 485	err_printf(m, "  IDLE?: %s\n", yesno(ee->idle));
 486	err_printf(m, "  START: 0x%08x\n", ee->start);
 487	err_printf(m, "  HEAD:  0x%08x [0x%08x]\n", ee->head, ee->rq_head);
 488	err_printf(m, "  TAIL:  0x%08x [0x%08x, 0x%08x]\n",
 489		   ee->tail, ee->rq_post, ee->rq_tail);
 490	err_printf(m, "  CTL:   0x%08x\n", ee->ctl);
 491	err_printf(m, "  MODE:  0x%08x\n", ee->mode);
 492	err_printf(m, "  HWS:   0x%08x\n", ee->hws);
 493	err_printf(m, "  ACTHD: 0x%08x %08x\n",
 494		   (u32)(ee->acthd>>32), (u32)ee->acthd);
 495	err_printf(m, "  IPEIR: 0x%08x\n", ee->ipeir);
 496	err_printf(m, "  IPEHR: 0x%08x\n", ee->ipehr);
 497
 498	error_print_instdone(m, ee);
 499
 500	if (ee->batchbuffer) {
 501		u64 start = ee->batchbuffer->gtt_offset;
 502		u64 end = start + ee->batchbuffer->gtt_size;
 503
 504		err_printf(m, "  batch: [0x%08x_%08x, 0x%08x_%08x]\n",
 505			   upper_32_bits(start), lower_32_bits(start),
 506			   upper_32_bits(end), lower_32_bits(end));
 507	}
 508	if (INTEL_GEN(m->i915) >= 4) {
 509		err_printf(m, "  BBADDR: 0x%08x_%08x\n",
 510			   (u32)(ee->bbaddr>>32), (u32)ee->bbaddr);
 511		err_printf(m, "  BB_STATE: 0x%08x\n", ee->bbstate);
 512		err_printf(m, "  INSTPS: 0x%08x\n", ee->instps);
 513	}
 514	err_printf(m, "  INSTPM: 0x%08x\n", ee->instpm);
 515	err_printf(m, "  FADDR: 0x%08x %08x\n", upper_32_bits(ee->faddr),
 516		   lower_32_bits(ee->faddr));
 517	if (INTEL_GEN(m->i915) >= 6) {
 518		err_printf(m, "  RC PSMI: 0x%08x\n", ee->rc_psmi);
 519		err_printf(m, "  FAULT_REG: 0x%08x\n", ee->fault_reg);
 520	}
 521	if (HAS_PPGTT(m->i915)) {
 522		err_printf(m, "  GFX_MODE: 0x%08x\n", ee->vm_info.gfx_mode);
 523
 524		if (INTEL_GEN(m->i915) >= 8) {
 525			int i;
 526			for (i = 0; i < 4; i++)
 527				err_printf(m, "  PDP%d: 0x%016llx\n",
 528					   i, ee->vm_info.pdp[i]);
 529		} else {
 530			err_printf(m, "  PP_DIR_BASE: 0x%08x\n",
 531				   ee->vm_info.pp_dir_base);
 532		}
 533	}
 534	err_printf(m, "  ring->head: 0x%08x\n", ee->cpu_ring_head);
 535	err_printf(m, "  ring->tail: 0x%08x\n", ee->cpu_ring_tail);
 536	err_printf(m, "  hangcheck timestamp: %dms (%lu%s)\n",
 537		   jiffies_to_msecs(ee->hangcheck_timestamp - epoch),
 538		   ee->hangcheck_timestamp,
 539		   ee->hangcheck_timestamp == epoch ? "; epoch" : "");
 540	err_printf(m, "  engine reset count: %u\n", ee->reset_count);
 541
 542	for (n = 0; n < ee->num_ports; n++) {
 543		err_printf(m, "  ELSP[%d]:", n);
 544		error_print_request(m, " ", &ee->execlist[n], epoch);
 545	}
 546
 547	error_print_context(m, "  Active context: ", &ee->context);
 548}
 549
 550void i915_error_printf(struct drm_i915_error_state_buf *e, const char *f, ...)
 551{
 552	va_list args;
 553
 554	va_start(args, f);
 555	i915_error_vprintf(e, f, args);
 556	va_end(args);
 557}
 558
 559static void print_error_obj(struct drm_i915_error_state_buf *m,
 560			    const struct intel_engine_cs *engine,
 561			    const char *name,
 562			    const struct drm_i915_error_object *obj)
 563{
 564	char out[ASCII85_BUFSZ];
 565	int page;
 566
 567	if (!obj)
 568		return;
 569
 570	if (name) {
 571		err_printf(m, "%s --- %s = 0x%08x %08x\n",
 572			   engine ? engine->name : "global", name,
 573			   upper_32_bits(obj->gtt_offset),
 574			   lower_32_bits(obj->gtt_offset));
 575	}
 576
 577	err_compression_marker(m);
 578	for (page = 0; page < obj->page_count; page++) {
 579		int i, len;
 580
 581		len = PAGE_SIZE;
 582		if (page == obj->page_count - 1)
 583			len -= obj->unused;
 584		len = ascii85_encode_len(len);
 585
 586		for (i = 0; i < len; i++)
 587			err_puts(m, ascii85_encode(obj->pages[page][i], out));
 588	}
 589	err_puts(m, "\n");
 590}
 591
 592static void err_print_capabilities(struct drm_i915_error_state_buf *m,
 593				   const struct intel_device_info *info,
 594				   const struct intel_runtime_info *runtime,
 595				   const struct intel_driver_caps *caps)
 596{
 597	struct drm_printer p = i915_error_printer(m);
 
 
 
 
 
 598
 599	intel_device_info_dump_flags(info, &p);
 600	intel_driver_caps_print(caps, &p);
 601	intel_device_info_dump_topology(&runtime->sseu, &p);
 602}
 603
 604static void err_print_params(struct drm_i915_error_state_buf *m,
 605			     const struct i915_params *params)
 606{
 607	struct drm_printer p = i915_error_printer(m);
 608
 609	i915_params_dump(params, &p);
 610}
 611
 612static void err_print_pciid(struct drm_i915_error_state_buf *m,
 613			    struct drm_i915_private *i915)
 614{
 615	struct pci_dev *pdev = i915->drm.pdev;
 616
 617	err_printf(m, "PCI ID: 0x%04x\n", pdev->device);
 618	err_printf(m, "PCI Revision: 0x%02x\n", pdev->revision);
 619	err_printf(m, "PCI Subsystem: %04x:%04x\n",
 620		   pdev->subsystem_vendor,
 621		   pdev->subsystem_device);
 622}
 623
 624static void err_print_uc(struct drm_i915_error_state_buf *m,
 625			 const struct i915_error_uc *error_uc)
 626{
 627	struct drm_printer p = i915_error_printer(m);
 628	const struct i915_gpu_state *error =
 629		container_of(error_uc, typeof(*error), uc);
 630
 631	if (!error->device_info.has_gt_uc)
 632		return;
 633
 634	intel_uc_fw_dump(&error_uc->guc_fw, &p);
 635	intel_uc_fw_dump(&error_uc->huc_fw, &p);
 636	print_error_obj(m, NULL, "GuC log buffer", error_uc->guc_log);
 637}
 638
 639static void err_free_sgl(struct scatterlist *sgl)
 640{
 641	while (sgl) {
 642		struct scatterlist *sg;
 643
 644		for (sg = sgl; !sg_is_chain(sg); sg++) {
 645			kfree(sg_virt(sg));
 646			if (sg_is_last(sg))
 647				break;
 
 
 
 
 
 
 
 
 
 
 
 
 648		}
 649
 650		sg = sg_is_last(sg) ? NULL : sg_chain_ptr(sg);
 651		free_page((unsigned long)sgl);
 652		sgl = sg;
 653	}
 654}
 655
 656static void __err_print_to_sgl(struct drm_i915_error_state_buf *m,
 657			       struct i915_gpu_state *error)
 658{
 659	const struct drm_i915_error_engine *ee;
 660	struct timespec64 ts;
 661	int i, j;
 662
 663	if (*error->error_msg)
 664		err_printf(m, "%s\n", error->error_msg);
 665	err_printf(m, "Kernel: %s %s\n",
 666		   init_utsname()->release,
 667		   init_utsname()->machine);
 668	err_printf(m, "Driver: %s\n", DRIVER_DATE);
 669	ts = ktime_to_timespec64(error->time);
 670	err_printf(m, "Time: %lld s %ld us\n",
 671		   (s64)ts.tv_sec, ts.tv_nsec / NSEC_PER_USEC);
 672	ts = ktime_to_timespec64(error->boottime);
 673	err_printf(m, "Boottime: %lld s %ld us\n",
 674		   (s64)ts.tv_sec, ts.tv_nsec / NSEC_PER_USEC);
 675	ts = ktime_to_timespec64(error->uptime);
 676	err_printf(m, "Uptime: %lld s %ld us\n",
 677		   (s64)ts.tv_sec, ts.tv_nsec / NSEC_PER_USEC);
 678	err_printf(m, "Epoch: %lu jiffies (%u HZ)\n", error->epoch, HZ);
 679	err_printf(m, "Capture: %lu jiffies; %d ms ago, %d ms after epoch\n",
 680		   error->capture,
 681		   jiffies_to_msecs(jiffies - error->capture),
 682		   jiffies_to_msecs(error->capture - error->epoch));
 683
 684	for (ee = error->engine; ee; ee = ee->next)
 685		err_printf(m, "Active process (on ring %s): %s [%d]\n",
 686			   ee->engine->name,
 687			   ee->context.comm,
 688			   ee->context.pid);
 689
 690	err_printf(m, "Reset count: %u\n", error->reset_count);
 691	err_printf(m, "Suspend count: %u\n", error->suspend_count);
 692	err_printf(m, "Platform: %s\n", intel_platform_name(error->device_info.platform));
 693	err_printf(m, "Subplatform: 0x%x\n",
 694		   intel_subplatform(&error->runtime_info,
 695				     error->device_info.platform));
 696	err_print_pciid(m, m->i915);
 697
 698	err_printf(m, "IOMMU enabled?: %d\n", error->iommu);
 699
 700	if (HAS_CSR(m->i915)) {
 701		struct intel_csr *csr = &m->i915->csr;
 702
 703		err_printf(m, "DMC loaded: %s\n",
 704			   yesno(csr->dmc_payload != NULL));
 705		err_printf(m, "DMC fw version: %d.%d\n",
 706			   CSR_VERSION_MAJOR(csr->version),
 707			   CSR_VERSION_MINOR(csr->version));
 708	}
 709
 710	err_printf(m, "GT awake: %s\n", yesno(error->awake));
 711	err_printf(m, "RPM wakelock: %s\n", yesno(error->wakelock));
 712	err_printf(m, "PM suspended: %s\n", yesno(error->suspended));
 713	err_printf(m, "EIR: 0x%08x\n", error->eir);
 714	err_printf(m, "IER: 0x%08x\n", error->ier);
 715	for (i = 0; i < error->ngtier; i++)
 716		err_printf(m, "GTIER[%d]: 0x%08x\n", i, error->gtier[i]);
 
 
 
 
 717	err_printf(m, "PGTBL_ER: 0x%08x\n", error->pgtbl_er);
 718	err_printf(m, "FORCEWAKE: 0x%08x\n", error->forcewake);
 719	err_printf(m, "DERRMR: 0x%08x\n", error->derrmr);
 720	err_printf(m, "CCID: 0x%08x\n", error->ccid);
 
 721
 722	for (i = 0; i < error->nfence; i++)
 723		err_printf(m, "  fence[%d] = %08llx\n", i, error->fence[i]);
 724
 725	if (IS_GEN_RANGE(m->i915, 6, 11)) {
 
 
 
 
 726		err_printf(m, "ERROR: 0x%08x\n", error->error);
 
 
 
 
 
 727		err_printf(m, "DONE_REG: 0x%08x\n", error->done_reg);
 728	}
 729
 730	if (INTEL_GEN(m->i915) >= 8)
 731		err_printf(m, "FAULT_TLB_DATA: 0x%08x 0x%08x\n",
 732			   error->fault_data1, error->fault_data0);
 
 
 733
 734	if (IS_GEN(m->i915, 7))
 735		err_printf(m, "ERR_INT: 0x%08x\n", error->err_int);
 736
 737	for (ee = error->engine; ee; ee = ee->next)
 738		error_print_engine(m, ee, error->epoch);
 
 739
 740	for (ee = error->engine; ee; ee = ee->next) {
 741		const struct drm_i915_error_object *obj;
 
 
 742
 743		obj = ee->batchbuffer;
 
 744		if (obj) {
 745			err_puts(m, ee->engine->name);
 746			if (ee->context.pid)
 747				err_printf(m, " (submitted by %s [%d])",
 748					   ee->context.comm,
 749					   ee->context.pid);
 750			err_printf(m, " --- gtt_offset = 0x%08x %08x\n",
 751				   upper_32_bits(obj->gtt_offset),
 752				   lower_32_bits(obj->gtt_offset));
 753			print_error_obj(m, ee->engine, NULL, obj);
 754		}
 755
 756		for (j = 0; j < ee->user_bo_count; j++)
 757			print_error_obj(m, ee->engine, "user", ee->user_bo[j]);
 
 
 
 
 
 758
 759		if (ee->num_requests) {
 760			err_printf(m, "%s --- %d requests\n",
 761				   ee->engine->name,
 762				   ee->num_requests);
 763			for (j = 0; j < ee->num_requests; j++)
 764				error_print_request(m, " ",
 765						    &ee->requests[j],
 766						    error->epoch);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 767		}
 768
 769		print_error_obj(m, ee->engine, "ringbuffer", ee->ringbuffer);
 770		print_error_obj(m, ee->engine, "HW Status", ee->hws_page);
 771		print_error_obj(m, ee->engine, "HW context", ee->ctx);
 772		print_error_obj(m, ee->engine, "WA context", ee->wa_ctx);
 773		print_error_obj(m, ee->engine,
 774				"WA batchbuffer", ee->wa_batchbuffer);
 775		print_error_obj(m, ee->engine,
 776				"NULL context", ee->default_state);
 
 
 
 
 
 
 
 
 
 
 
 777	}
 778
 779	if (error->overlay)
 780		intel_overlay_print_error_state(m, error->overlay);
 781
 782	if (error->display)
 783		intel_display_print_error_state(m, error->display);
 
 
 
 
 784
 785	err_print_capabilities(m, &error->device_info, &error->runtime_info,
 786			       &error->driver_caps);
 787	err_print_params(m, &error->params);
 788	err_print_uc(m, &error->uc);
 789}
 790
 791static int err_print_to_sgl(struct i915_gpu_state *error)
 
 
 792{
 793	struct drm_i915_error_state_buf m;
 
 794
 795	if (IS_ERR(error))
 796		return PTR_ERR(error);
 
 
 
 
 797
 798	if (READ_ONCE(error->sgl))
 799		return 0;
 800
 801	memset(&m, 0, sizeof(m));
 802	m.i915 = error->i915;
 803
 804	__err_print_to_sgl(&m, error);
 805
 806	if (m.buf) {
 807		__sg_set_buf(m.cur++, m.buf, m.bytes, m.iter);
 808		m.bytes = 0;
 809		m.buf = NULL;
 810	}
 811	if (m.cur) {
 812		GEM_BUG_ON(m.end < m.cur);
 813		sg_mark_end(m.cur - 1);
 814	}
 815	GEM_BUG_ON(m.sgl && !m.cur);
 816
 817	if (m.err) {
 818		err_free_sgl(m.sgl);
 819		return m.err;
 820	}
 821
 822	if (cmpxchg(&error->sgl, NULL, m.sgl))
 823		err_free_sgl(m.sgl);
 824
 825	return 0;
 826}
 827
 828ssize_t i915_gpu_state_copy_to_buffer(struct i915_gpu_state *error,
 829				      char *buf, loff_t off, size_t rem)
 830{
 831	struct scatterlist *sg;
 832	size_t count;
 833	loff_t pos;
 834	int err;
 835
 836	if (!error || !rem)
 837		return 0;
 838
 839	err = err_print_to_sgl(error);
 840	if (err)
 841		return err;
 842
 843	sg = READ_ONCE(error->fit);
 844	if (!sg || off < sg->dma_address)
 845		sg = error->sgl;
 846	if (!sg)
 847		return 0;
 848
 849	pos = sg->dma_address;
 850	count = 0;
 851	do {
 852		size_t len, start;
 853
 854		if (sg_is_chain(sg)) {
 855			sg = sg_chain_ptr(sg);
 856			GEM_BUG_ON(sg_is_chain(sg));
 857		}
 858
 859		len = sg->length;
 860		if (pos + len <= off) {
 861			pos += len;
 862			continue;
 863		}
 864
 865		start = sg->offset;
 866		if (pos < off) {
 867			GEM_BUG_ON(off - pos > len);
 868			len -= off - pos;
 869			start += off - pos;
 870			pos = off;
 871		}
 872
 873		len = min(len, rem);
 874		GEM_BUG_ON(!len || len > sg->length);
 875
 876		memcpy(buf, page_address(sg_page(sg)) + start, len);
 877
 878		count += len;
 879		pos += len;
 880
 881		buf += len;
 882		rem -= len;
 883		if (!rem) {
 884			WRITE_ONCE(error->fit, sg);
 885			break;
 886		}
 887	} while (!sg_is_last(sg++));
 888
 889	return count;
 890}
 891
 892static void i915_error_object_free(struct drm_i915_error_object *obj)
 893{
 894	int page;
 895
 896	if (obj == NULL)
 897		return;
 898
 899	for (page = 0; page < obj->page_count; page++)
 900		free_page((unsigned long)obj->pages[page]);
 901
 902	kfree(obj);
 903}
 904
 
 
 
 
 
 905
 906static void cleanup_params(struct i915_gpu_state *error)
 907{
 908	i915_params_free(&error->params);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 909}
 910
 911static void cleanup_uc_state(struct i915_gpu_state *error)
 
 
 
 912{
 913	struct i915_error_uc *error_uc = &error->uc;
 
 
 
 
 
 
 
 
 914
 915	kfree(error_uc->guc_fw.path);
 916	kfree(error_uc->huc_fw.path);
 917	i915_error_object_free(error_uc->guc_log);
 918}
 919
 920void __i915_gpu_state_free(struct kref *error_ref)
 921{
 922	struct i915_gpu_state *error =
 923		container_of(error_ref, typeof(*error), ref);
 924	long i;
 925
 926	while (error->engine) {
 927		struct drm_i915_error_engine *ee = error->engine;
 
 
 928
 929		error->engine = ee->next;
 
 
 
 
 
 930
 931		for (i = 0; i < ee->user_bo_count; i++)
 932			i915_error_object_free(ee->user_bo[i]);
 933		kfree(ee->user_bo);
 934
 935		i915_error_object_free(ee->batchbuffer);
 936		i915_error_object_free(ee->wa_batchbuffer);
 937		i915_error_object_free(ee->ringbuffer);
 938		i915_error_object_free(ee->hws_page);
 939		i915_error_object_free(ee->ctx);
 940		i915_error_object_free(ee->wa_ctx);
 941
 942		kfree(ee->requests);
 943		kfree(ee);
 
 944	}
 945
 946	kfree(error->overlay);
 947	kfree(error->display);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 948
 949	cleanup_params(error);
 950	cleanup_uc_state(error);
 951
 952	err_free_sgl(error->sgl);
 953	kfree(error);
 
 
 
 954}
 
 
 955
 956static struct drm_i915_error_object *
 957i915_error_object_create(struct drm_i915_private *i915,
 958			 struct i915_vma *vma,
 959			 struct compress *compress)
 960{
 961	struct i915_ggtt *ggtt = &i915->ggtt;
 962	const u64 slot = ggtt->error_capture.start;
 963	struct drm_i915_error_object *dst;
 964	unsigned long num_pages;
 965	struct sgt_iter iter;
 966	dma_addr_t dma;
 967	int ret;
 968
 969	might_sleep();
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 970
 971	if (!vma || !vma->pages)
 972		return NULL;
 973
 974	num_pages = min_t(u64, vma->size, vma->obj->base.size) >> PAGE_SHIFT;
 975	num_pages = DIV_ROUND_UP(10 * num_pages, 8); /* worstcase zlib growth */
 976	dst = kmalloc(sizeof(*dst) + num_pages * sizeof(u32 *), ALLOW_FAIL);
 977	if (!dst)
 978		return NULL;
 
 
 979
 980	if (!compress_start(compress)) {
 981		kfree(dst);
 982		return NULL;
 983	}
 984
 985	dst->gtt_offset = vma->node.start;
 986	dst->gtt_size = vma->node.size;
 987	dst->num_pages = num_pages;
 988	dst->page_count = 0;
 989	dst->unused = 0;
 990
 991	ret = -EINVAL;
 992	for_each_sgt_dma(dma, iter, vma->pages) {
 993		void __iomem *s;
 994
 995		ggtt->vm.insert_page(&ggtt->vm, dma, slot, I915_CACHE_NONE, 0);
 996
 997		s = io_mapping_map_wc(&ggtt->iomap, slot, PAGE_SIZE);
 998		ret = compress_page(compress, (void  __force *)s, dst);
 999		io_mapping_unmap(s);
1000		if (ret)
1001			break;
1002	}
1003
1004	if (ret || compress_flush(compress, dst)) {
1005		while (dst->page_count--)
1006			pool_free(&compress->pool, dst->pages[dst->page_count]);
1007		kfree(dst);
1008		dst = NULL;
1009	}
1010	compress_finish(compress);
1011
1012	return dst;
1013}
1014
1015/*
1016 * Generate a semi-unique error code. The code is not meant to have meaning, The
1017 * code's only purpose is to try to prevent false duplicated bug reports by
1018 * grossly estimating a GPU error state.
1019 *
1020 * TODO Ideally, hashing the batchbuffer would be a very nice way to determine
1021 * the hang if we could strip the GTT offset information from it.
1022 *
1023 * It's only a small step better than a random number in its current form.
1024 */
1025static u32 i915_error_generate_code(struct i915_gpu_state *error)
 
 
1026{
1027	const struct drm_i915_error_engine *ee = error->engine;
 
1028
1029	/*
1030	 * IPEHR would be an ideal way to detect errors, as it's the gross
1031	 * measure of "the command that hung." However, has some very common
1032	 * synchronization commands which almost always appear in the case
1033	 * strictly a client bug. Use instdone to differentiate those some.
1034	 */
1035	return ee ? ee->ipehr ^ ee->instdone.instdone : 0;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1036}
1037
1038static void gem_record_fences(struct i915_gpu_state *error)
 
 
 
 
1039{
1040	struct drm_i915_private *dev_priv = error->i915;
1041	struct intel_uncore *uncore = &dev_priv->uncore;
1042	int i;
1043
1044	if (INTEL_GEN(dev_priv) >= 6) {
1045		for (i = 0; i < dev_priv->ggtt.num_fences; i++)
1046			error->fence[i] =
1047				intel_uncore_read64(uncore,
1048						    FENCE_REG_GEN6_LO(i));
1049	} else if (INTEL_GEN(dev_priv) >= 4) {
1050		for (i = 0; i < dev_priv->ggtt.num_fences; i++)
1051			error->fence[i] =
1052				intel_uncore_read64(uncore,
1053						    FENCE_REG_965_LO(i));
1054	} else {
1055		for (i = 0; i < dev_priv->ggtt.num_fences; i++)
1056			error->fence[i] =
1057				intel_uncore_read(uncore, FENCE_REG(i));
 
 
 
 
 
 
 
 
 
1058	}
1059	error->nfence = i;
1060}
1061
1062static void error_record_engine_registers(struct i915_gpu_state *error,
1063					  struct intel_engine_cs *engine,
1064					  struct drm_i915_error_engine *ee)
1065{
1066	struct drm_i915_private *dev_priv = engine->i915;
 
 
 
 
 
 
 
 
 
 
1067
1068	if (INTEL_GEN(dev_priv) >= 6) {
1069		ee->rc_psmi = ENGINE_READ(engine, RING_PSMI_CTL);
 
 
 
 
1070
1071		if (INTEL_GEN(dev_priv) >= 12)
1072			ee->fault_reg = I915_READ(GEN12_RING_FAULT_REG);
1073		else if (INTEL_GEN(dev_priv) >= 8)
1074			ee->fault_reg = I915_READ(GEN8_RING_FAULT_REG);
 
1075		else
1076			ee->fault_reg = GEN6_RING_FAULT_REG_READ(engine);
1077	}
1078
1079	if (INTEL_GEN(dev_priv) >= 4) {
1080		ee->faddr = ENGINE_READ(engine, RING_DMA_FADD);
1081		ee->ipeir = ENGINE_READ(engine, RING_IPEIR);
1082		ee->ipehr = ENGINE_READ(engine, RING_IPEHR);
1083		ee->instps = ENGINE_READ(engine, RING_INSTPS);
1084		ee->bbaddr = ENGINE_READ(engine, RING_BBADDR);
1085		if (INTEL_GEN(dev_priv) >= 8) {
1086			ee->faddr |= (u64)ENGINE_READ(engine, RING_DMA_FADD_UDW) << 32;
1087			ee->bbaddr |= (u64)ENGINE_READ(engine, RING_BBADDR_UDW) << 32;
 
1088		}
1089		ee->bbstate = ENGINE_READ(engine, RING_BBSTATE);
1090	} else {
1091		ee->faddr = ENGINE_READ(engine, DMA_FADD_I8XX);
1092		ee->ipeir = ENGINE_READ(engine, IPEIR);
1093		ee->ipehr = ENGINE_READ(engine, IPEHR);
1094	}
1095
1096	intel_engine_get_instdone(engine, &ee->instdone);
1097
1098	ee->instpm = ENGINE_READ(engine, RING_INSTPM);
1099	ee->acthd = intel_engine_get_active_head(engine);
1100	ee->start = ENGINE_READ(engine, RING_START);
1101	ee->head = ENGINE_READ(engine, RING_HEAD);
1102	ee->tail = ENGINE_READ(engine, RING_TAIL);
1103	ee->ctl = ENGINE_READ(engine, RING_CTL);
1104	if (INTEL_GEN(dev_priv) > 2)
1105		ee->mode = ENGINE_READ(engine, RING_MI_MODE);
1106
1107	if (!HWS_NEEDS_PHYSICAL(dev_priv)) {
1108		i915_reg_t mmio;
1109
1110		if (IS_GEN(dev_priv, 7)) {
1111			switch (engine->id) {
1112			default:
1113				MISSING_CASE(engine->id);
1114				/* fall through */
1115			case RCS0:
1116				mmio = RENDER_HWS_PGA_GEN7;
1117				break;
1118			case BCS0:
1119				mmio = BLT_HWS_PGA_GEN7;
1120				break;
1121			case VCS0:
1122				mmio = BSD_HWS_PGA_GEN7;
1123				break;
1124			case VECS0:
1125				mmio = VEBOX_HWS_PGA_GEN7;
1126				break;
1127			}
1128		} else if (IS_GEN(engine->i915, 6)) {
1129			mmio = RING_HWS_PGA_GEN6(engine->mmio_base);
1130		} else {
1131			/* XXX: gen8 returns to sanity */
1132			mmio = RING_HWS_PGA(engine->mmio_base);
1133		}
1134
1135		ee->hws = I915_READ(mmio);
1136	}
1137
1138	ee->idle = intel_engine_is_idle(engine);
1139	if (!ee->idle)
1140		ee->hangcheck_timestamp = engine->hangcheck.action_timestamp;
1141	ee->reset_count = i915_reset_engine_count(&dev_priv->gpu_error,
1142						  engine);
1143
1144	if (HAS_PPGTT(dev_priv)) {
1145		int i;
1146
1147		ee->vm_info.gfx_mode = ENGINE_READ(engine, RING_MODE_GEN7);
1148
1149		if (IS_GEN(dev_priv, 6)) {
1150			ee->vm_info.pp_dir_base =
1151				ENGINE_READ(engine, RING_PP_DIR_BASE_READ);
1152		} else if (IS_GEN(dev_priv, 7)) {
1153			ee->vm_info.pp_dir_base =
1154				ENGINE_READ(engine, RING_PP_DIR_BASE);
1155		} else if (INTEL_GEN(dev_priv) >= 8) {
1156			u32 base = engine->mmio_base;
1157
 
 
 
 
 
 
 
1158			for (i = 0; i < 4; i++) {
1159				ee->vm_info.pdp[i] =
1160					I915_READ(GEN8_RING_PDP_UDW(base, i));
1161				ee->vm_info.pdp[i] <<= 32;
1162				ee->vm_info.pdp[i] |=
1163					I915_READ(GEN8_RING_PDP_LDW(base, i));
1164			}
1165		}
1166	}
1167}
1168
1169static void record_request(const struct i915_request *request,
1170			   struct drm_i915_error_request *erq)
 
 
1171{
1172	const struct i915_gem_context *ctx = request->gem_context;
 
1173
1174	erq->flags = request->fence.flags;
1175	erq->context = request->fence.context;
1176	erq->seqno = request->fence.seqno;
1177	erq->sched_attr = request->sched.attr;
1178	erq->jiffies = request->emitted_jiffies;
1179	erq->start = i915_ggtt_offset(request->ring->vma);
1180	erq->head = request->head;
1181	erq->tail = request->tail;
1182
1183	rcu_read_lock();
1184	erq->pid = ctx->pid ? pid_nr(ctx->pid) : 0;
1185	rcu_read_unlock();
1186}
1187
1188static void engine_record_requests(struct intel_engine_cs *engine,
1189				   struct i915_request *first,
1190				   struct drm_i915_error_engine *ee)
1191{
1192	struct i915_request *request;
1193	int count;
1194
1195	count = 0;
1196	request = first;
1197	list_for_each_entry_from(request, &engine->active.requests, sched.link)
1198		count++;
1199	if (!count)
1200		return;
1201
1202	ee->requests = kcalloc(count, sizeof(*ee->requests), ATOMIC_MAYFAIL);
1203	if (!ee->requests)
1204		return;
1205
1206	ee->num_requests = count;
1207
1208	count = 0;
1209	request = first;
1210	list_for_each_entry_from(request,
1211				 &engine->active.requests, sched.link) {
1212		if (count >= ee->num_requests) {
1213			/*
1214			 * If the ring request list was changed in
1215			 * between the point where the error request
1216			 * list was created and dimensioned and this
1217			 * point then just exit early to avoid crashes.
1218			 *
1219			 * We don't need to communicate that the
1220			 * request list changed state during error
1221			 * state capture and that the error state is
1222			 * slightly incorrect as a consequence since we
1223			 * are typically only interested in the request
1224			 * list state at the point of error state
1225			 * capture, not in any changes happening during
1226			 * the capture.
1227			 */
1228			break;
1229		}
1230
1231		record_request(request, &ee->requests[count++]);
1232	}
1233	ee->num_requests = count;
1234}
1235
1236static void error_record_engine_execlists(const struct intel_engine_cs *engine,
1237					  struct drm_i915_error_engine *ee)
1238{
1239	const struct intel_engine_execlists * const execlists = &engine->execlists;
1240	struct i915_request * const *port = execlists->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
1249static bool record_context(struct drm_i915_error_context *e,
1250			   const struct i915_request *rq)
1251{
1252	const struct i915_gem_context *ctx = rq->gem_context;
1253
1254	if (ctx->pid) {
1255		struct task_struct *task;
1256
1257		rcu_read_lock();
1258		task = pid_task(ctx->pid, PIDTYPE_PID);
1259		if (task) {
1260			strcpy(e->comm, task->comm);
1261			e->pid = task->pid;
1262		}
1263		rcu_read_unlock();
1264	}
1265
1266	e->hw_id = ctx->hw_id;
1267	e->sched_attr = ctx->sched;
1268	e->guilty = atomic_read(&ctx->guilty_count);
1269	e->active = atomic_read(&ctx->active_count);
1270
1271	return i915_gem_context_no_error_capture(ctx);
1272}
 
1273
1274struct capture_vma {
1275	struct capture_vma *next;
1276	void **slot;
1277};
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1278
1279static struct capture_vma *
1280capture_vma(struct capture_vma *next,
1281	    struct i915_vma *vma,
1282	    struct drm_i915_error_object **out)
1283{
1284	struct capture_vma *c;
1285
1286	*out = NULL;
1287	if (!vma)
1288		return next;
1289
1290	c = kmalloc(sizeof(*c), ATOMIC_MAYFAIL);
1291	if (!c)
1292		return next;
1293
1294	if (!i915_active_trygrab(&vma->active)) {
1295		kfree(c);
1296		return next;
1297	}
1298
1299	c->slot = (void **)out;
1300	*c->slot = i915_vma_get(vma);
1301
1302	c->next = next;
1303	return c;
1304}
1305
1306static struct capture_vma *
1307request_record_user_bo(struct i915_request *request,
1308		       struct drm_i915_error_engine *ee,
1309		       struct capture_vma *capture)
1310{
1311	struct i915_capture_list *c;
1312	struct drm_i915_error_object **bo;
1313	long count, max;
1314
1315	max = 0;
1316	for (c = request->capture_list; c; c = c->next)
1317		max++;
1318	if (!max)
1319		return capture;
1320
1321	bo = kmalloc_array(max, sizeof(*bo), ATOMIC_MAYFAIL);
1322	if (!bo) {
1323		/* If we can't capture everything, try to capture something. */
1324		max = min_t(long, max, PAGE_SIZE / sizeof(*bo));
1325		bo = kmalloc_array(max, sizeof(*bo), ATOMIC_MAYFAIL);
1326	}
1327	if (!bo)
1328		return capture;
1329
1330	count = 0;
1331	for (c = request->capture_list; c; c = c->next) {
1332		capture = capture_vma(capture, c->vma, &bo[count]);
1333		if (++count == max)
1334			break;
1335	}
1336
1337	ee->user_bo = bo;
1338	ee->user_bo_count = count;
1339
1340	return capture;
1341}
1342
1343static struct drm_i915_error_object *
1344capture_object(struct drm_i915_private *dev_priv,
1345	       struct drm_i915_gem_object *obj,
1346	       struct compress *compress)
1347{
1348	if (obj && i915_gem_object_has_pages(obj)) {
1349		struct i915_vma fake = {
1350			.node = { .start = U64_MAX, .size = obj->base.size },
1351			.size = obj->base.size,
1352			.pages = obj->mm.pages,
1353			.obj = obj,
1354		};
1355
1356		return i915_error_object_create(dev_priv, &fake, compress);
1357	} else {
1358		return NULL;
1359	}
1360}
1361
1362static void
1363gem_record_rings(struct i915_gpu_state *error, struct compress *compress)
1364{
1365	struct drm_i915_private *i915 = error->i915;
1366	struct intel_engine_cs *engine;
1367	struct drm_i915_error_engine *ee;
1368
1369	ee = kzalloc(sizeof(*ee), GFP_KERNEL);
1370	if (!ee)
1371		return;
1372
1373	for_each_uabi_engine(engine, i915) {
1374		struct capture_vma *capture = NULL;
1375		struct i915_request *request;
1376		unsigned long flags;
1377
1378		/* Refill our page pool before entering atomic section */
1379		pool_refill(&compress->pool, ALLOW_FAIL);
1380
1381		spin_lock_irqsave(&engine->active.lock, flags);
1382		request = intel_engine_find_active_request(engine);
1383		if (!request) {
1384			spin_unlock_irqrestore(&engine->active.lock, flags);
1385			continue;
1386		}
1387
1388		error->simulated |= record_context(&ee->context, request);
1389
1390		/*
1391		 * We need to copy these to an anonymous buffer
1392		 * as the simplest method to avoid being overwritten
1393		 * by userspace.
1394		 */
1395		capture = capture_vma(capture,
1396				      request->batch,
1397				      &ee->batchbuffer);
1398
1399		if (HAS_BROKEN_CS_TLB(i915))
1400			capture = capture_vma(capture,
1401					      engine->gt->scratch,
1402					      &ee->wa_batchbuffer);
1403
1404		capture = request_record_user_bo(request, ee, capture);
1405
1406		capture = capture_vma(capture,
1407				      request->hw_context->state,
1408				      &ee->ctx);
1409
1410		capture = capture_vma(capture,
1411				      request->ring->vma,
1412				      &ee->ringbuffer);
1413
1414		ee->cpu_ring_head = request->ring->head;
1415		ee->cpu_ring_tail = request->ring->tail;
1416
1417		ee->rq_head = request->head;
1418		ee->rq_post = request->postfix;
1419		ee->rq_tail = request->tail;
1420
1421		engine_record_requests(engine, request, ee);
1422		spin_unlock_irqrestore(&engine->active.lock, flags);
1423
1424		error_record_engine_registers(error, engine, ee);
1425		error_record_engine_execlists(engine, ee);
1426
1427		while (capture) {
1428			struct capture_vma *this = capture;
1429			struct i915_vma *vma = *this->slot;
1430
1431			*this->slot =
1432				i915_error_object_create(i915, vma, compress);
1433
1434			i915_active_ungrab(&vma->active);
1435			i915_vma_put(vma);
1436
1437			capture = this->next;
1438			kfree(this);
1439		}
1440
1441		ee->hws_page =
1442			i915_error_object_create(i915,
1443						 engine->status_page.vma,
1444						 compress);
1445
1446		ee->wa_ctx =
1447			i915_error_object_create(i915,
1448						 engine->wa_ctx.vma,
1449						 compress);
1450
1451		ee->default_state =
1452			capture_object(i915, engine->default_state, compress);
1453
1454		ee->engine = engine;
1455
1456		ee->next = error->engine;
1457		error->engine = ee;
1458
1459		ee = kzalloc(sizeof(*ee), GFP_KERNEL);
1460		if (!ee)
1461			return;
1462	}
1463
1464	kfree(ee);
1465}
1466
1467static void
1468capture_uc_state(struct i915_gpu_state *error, struct compress *compress)
1469{
1470	struct drm_i915_private *i915 = error->i915;
1471	struct i915_error_uc *error_uc = &error->uc;
1472	struct intel_uc *uc = &i915->gt.uc;
 
 
 
 
 
 
1473
1474	/* Capturing uC state won't be useful if there is no GuC */
1475	if (!error->device_info.has_gt_uc)
1476		return;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1477
1478	memcpy(&error_uc->guc_fw, &uc->guc.fw, sizeof(uc->guc.fw));
1479	memcpy(&error_uc->huc_fw, &uc->huc.fw, sizeof(uc->huc.fw));
1480
1481	/* Non-default firmware paths will be specified by the modparam.
1482	 * As modparams are generally accesible from the userspace make
1483	 * explicit copies of the firmware paths.
1484	 */
1485	error_uc->guc_fw.path = kstrdup(uc->guc.fw.path, ALLOW_FAIL);
1486	error_uc->huc_fw.path = kstrdup(uc->huc.fw.path, ALLOW_FAIL);
1487	error_uc->guc_log = i915_error_object_create(i915,
1488						     uc->guc.log.vma,
1489						     compress);
1490}
1491
1492/* Capture all registers which don't fit into another category. */
1493static void capture_reg_state(struct i915_gpu_state *error)
 
1494{
1495	struct drm_i915_private *i915 = error->i915;
1496	struct intel_uncore *uncore = &i915->uncore;
1497	int i;
1498
1499	/* General organization
1500	 * 1. Registers specific to a single generation
1501	 * 2. Registers which belong to multiple generations
1502	 * 3. Feature specific registers.
1503	 * 4. Everything else
1504	 * Please try to follow the order.
1505	 */
1506
1507	/* 1: Registers specific to a single generation */
1508	if (IS_VALLEYVIEW(i915)) {
1509		error->gtier[0] = intel_uncore_read(uncore, GTIER);
1510		error->ier = intel_uncore_read(uncore, VLV_IER);
1511		error->forcewake = intel_uncore_read_fw(uncore, FORCEWAKE_VLV);
1512	}
1513
1514	if (IS_GEN(i915, 7))
1515		error->err_int = intel_uncore_read(uncore, GEN7_ERR_INT);
1516
1517	if (INTEL_GEN(i915) >= 12) {
1518		error->fault_data0 = intel_uncore_read(uncore,
1519						       GEN12_FAULT_TLB_DATA0);
1520		error->fault_data1 = intel_uncore_read(uncore,
1521						       GEN12_FAULT_TLB_DATA1);
1522	} else if (INTEL_GEN(i915) >= 8) {
1523		error->fault_data0 = intel_uncore_read(uncore,
1524						       GEN8_FAULT_TLB_DATA0);
1525		error->fault_data1 = intel_uncore_read(uncore,
1526						       GEN8_FAULT_TLB_DATA1);
1527	}
1528
1529	if (IS_GEN(i915, 6)) {
1530		error->forcewake = intel_uncore_read_fw(uncore, FORCEWAKE);
1531		error->gab_ctl = intel_uncore_read(uncore, GAB_CTL);
1532		error->gfx_mode = intel_uncore_read(uncore, GFX_MODE);
1533	}
1534
1535	/* 2: Registers which belong to multiple generations */
1536	if (INTEL_GEN(i915) >= 7)
1537		error->forcewake = intel_uncore_read_fw(uncore, FORCEWAKE_MT);
1538
1539	if (INTEL_GEN(i915) >= 6) {
1540		error->derrmr = intel_uncore_read(uncore, DERRMR);
1541		if (INTEL_GEN(i915) < 12) {
1542			error->error = intel_uncore_read(uncore, ERROR_GEN6);
1543			error->done_reg = intel_uncore_read(uncore, DONE_REG);
1544		}
1545	}
1546
1547	if (INTEL_GEN(i915) >= 5)
1548		error->ccid = intel_uncore_read(uncore, CCID(RENDER_RING_BASE));
1549
1550	/* 3: Feature specific registers */
1551	if (IS_GEN_RANGE(i915, 6, 7)) {
1552		error->gam_ecochk = intel_uncore_read(uncore, GAM_ECOCHK);
1553		error->gac_eco = intel_uncore_read(uncore, GAC_ECO_BITS);
1554	}
1555
1556	/* 4: Everything else */
1557	if (INTEL_GEN(i915) >= 11) {
1558		error->ier = intel_uncore_read(uncore, GEN8_DE_MISC_IER);
1559		error->gtier[0] =
1560			intel_uncore_read(uncore,
1561					  GEN11_RENDER_COPY_INTR_ENABLE);
1562		error->gtier[1] =
1563			intel_uncore_read(uncore, GEN11_VCS_VECS_INTR_ENABLE);
1564		error->gtier[2] =
1565			intel_uncore_read(uncore, GEN11_GUC_SG_INTR_ENABLE);
1566		error->gtier[3] =
1567			intel_uncore_read(uncore,
1568					  GEN11_GPM_WGBOXPERF_INTR_ENABLE);
1569		error->gtier[4] =
1570			intel_uncore_read(uncore,
1571					  GEN11_CRYPTO_RSVD_INTR_ENABLE);
1572		error->gtier[5] =
1573			intel_uncore_read(uncore,
1574					  GEN11_GUNIT_CSME_INTR_ENABLE);
1575		error->ngtier = 6;
1576	} else if (INTEL_GEN(i915) >= 8) {
1577		error->ier = intel_uncore_read(uncore, GEN8_DE_MISC_IER);
1578		for (i = 0; i < 4; i++)
1579			error->gtier[i] = intel_uncore_read(uncore,
1580							    GEN8_GT_IER(i));
1581		error->ngtier = 4;
1582	} else if (HAS_PCH_SPLIT(i915)) {
1583		error->ier = intel_uncore_read(uncore, DEIER);
1584		error->gtier[0] = intel_uncore_read(uncore, GTIER);
1585		error->ngtier = 1;
1586	} else if (IS_GEN(i915, 2)) {
1587		error->ier = intel_uncore_read16(uncore, GEN2_IER);
1588	} else if (!IS_VALLEYVIEW(i915)) {
1589		error->ier = intel_uncore_read(uncore, GEN2_IER);
1590	}
1591	error->eir = intel_uncore_read(uncore, EIR);
1592	error->pgtbl_er = intel_uncore_read(uncore, PGTBL_ER);
1593}
1594
1595static const char *
1596error_msg(struct i915_gpu_state *error,
1597	  intel_engine_mask_t engines, const char *msg)
1598{
1599	int len;
 
 
 
 
1600
1601	len = scnprintf(error->error_msg, sizeof(error->error_msg),
1602			"GPU HANG: ecode %d:%x:0x%08x",
1603			INTEL_GEN(error->i915), engines,
1604			i915_error_generate_code(error));
1605	if (error->engine) {
1606		/* Just show the first executing process, more is confusing */
1607		len += scnprintf(error->error_msg + len,
1608				 sizeof(error->error_msg) - len,
1609				 ", in %s [%d]",
1610				 error->engine->context.comm,
1611				 error->engine->context.pid);
1612	}
1613	if (msg)
1614		len += scnprintf(error->error_msg + len,
1615				 sizeof(error->error_msg) - len,
1616				 ", %s", msg);
1617
1618	return error->error_msg;
 
 
 
1619}
1620
1621static void capture_gen_state(struct i915_gpu_state *error)
 
1622{
1623	struct drm_i915_private *i915 = error->i915;
1624
1625	error->awake = i915->gt.awake;
1626	error->wakelock = atomic_read(&i915->runtime_pm.wakeref_count);
1627	error->suspended = i915->runtime_pm.suspended;
1628
1629	error->iommu = -1;
1630#ifdef CONFIG_INTEL_IOMMU
1631	error->iommu = intel_iommu_gfx_mapped;
1632#endif
1633	error->reset_count = i915_reset_count(&i915->gpu_error);
1634	error->suspend_count = i915->suspend_count;
1635
1636	memcpy(&error->device_info,
1637	       INTEL_INFO(i915),
1638	       sizeof(error->device_info));
1639	memcpy(&error->runtime_info,
1640	       RUNTIME_INFO(i915),
1641	       sizeof(error->runtime_info));
1642	error->driver_caps = i915->caps;
1643}
1644
1645static void capture_params(struct i915_gpu_state *error)
1646{
1647	i915_params_copy(&error->params, &i915_modparams);
1648}
1649
1650static unsigned long capture_find_epoch(const struct i915_gpu_state *error)
1651{
1652	const struct drm_i915_error_engine *ee;
1653	unsigned long epoch = error->capture;
1654
1655	for (ee = error->engine; ee; ee = ee->next) {
1656		if (ee->hangcheck_timestamp &&
1657		    time_before(ee->hangcheck_timestamp, epoch))
1658			epoch = ee->hangcheck_timestamp;
1659	}
1660
1661	return epoch;
1662}
1663
1664static void capture_finish(struct i915_gpu_state *error)
1665{
1666	struct i915_ggtt *ggtt = &error->i915->ggtt;
1667	const u64 slot = ggtt->error_capture.start;
1668
1669	ggtt->vm.clear_range(&ggtt->vm, slot, PAGE_SIZE);
1670}
1671
1672#define DAY_AS_SECONDS(x) (24 * 60 * 60 * (x))
1673
1674struct i915_gpu_state *
1675i915_capture_gpu_state(struct drm_i915_private *i915)
1676{
1677	struct i915_gpu_state *error;
1678	struct compress compress;
1679
1680	/* Check if GPU capture has been disabled */
1681	error = READ_ONCE(i915->gpu_error.first_error);
1682	if (IS_ERR(error))
1683		return error;
1684
1685	error = kzalloc(sizeof(*error), ALLOW_FAIL);
1686	if (!error) {
1687		i915_disable_error_state(i915, -ENOMEM);
1688		return ERR_PTR(-ENOMEM);
1689	}
1690
1691	if (!compress_init(&compress)) {
1692		kfree(error);
1693		i915_disable_error_state(i915, -ENOMEM);
1694		return ERR_PTR(-ENOMEM);
1695	}
1696
1697	kref_init(&error->ref);
1698	error->i915 = i915;
1699
1700	error->time = ktime_get_real();
1701	error->boottime = ktime_get_boottime();
1702	error->uptime = ktime_sub(ktime_get(), i915->gt.last_init_time);
1703	error->capture = jiffies;
1704
1705	capture_params(error);
1706	capture_gen_state(error);
1707	capture_uc_state(error, &compress);
1708	capture_reg_state(error);
1709	gem_record_fences(error);
1710	gem_record_rings(error, &compress);
1711
1712	error->overlay = intel_overlay_capture_error_state(i915);
1713	error->display = intel_display_capture_error_state(i915);
1714
1715	error->epoch = capture_find_epoch(error);
1716
1717	capture_finish(error);
1718	compress_fini(&compress);
1719
1720	return error;
1721}
1722
1723/**
1724 * i915_capture_error_state - capture an error record for later analysis
1725 * @i915: i915 device
1726 * @engine_mask: the mask of engines triggering the hang
1727 * @msg: a message to insert into the error capture header
1728 *
1729 * Should be called when an error is detected (either a hang or an error
1730 * interrupt) to capture error state from the time of the error.  Fills
1731 * out a structure which becomes available in debugfs for user level tools
1732 * to pick up.
1733 */
1734void i915_capture_error_state(struct drm_i915_private *i915,
1735			      intel_engine_mask_t engine_mask,
1736			      const char *msg)
1737{
1738	static bool warned;
1739	struct i915_gpu_state *error;
 
1740	unsigned long flags;
1741
1742	if (!i915_modparams.error_capture)
 
 
 
1743		return;
 
1744
1745	if (READ_ONCE(i915->gpu_error.first_error))
1746		return;
1747
1748	error = i915_capture_gpu_state(i915);
1749	if (IS_ERR(error))
1750		return;
1751
1752	dev_info(i915->drm.dev, "%s\n", error_msg(error, engine_mask, msg));
1753
1754	if (!error->simulated) {
1755		spin_lock_irqsave(&i915->gpu_error.lock, flags);
1756		if (!i915->gpu_error.first_error) {
1757			i915->gpu_error.first_error = error;
1758			error = NULL;
1759		}
1760		spin_unlock_irqrestore(&i915->gpu_error.lock, flags);
 
 
 
 
 
 
 
 
 
 
 
1761	}
 
1762
1763	if (error) {
1764		__i915_gpu_state_free(&error->ref);
1765		return;
1766	}
1767
1768	if (!xchg(&warned, true) &&
1769	    ktime_get_real_seconds() - DRIVER_TIMESTAMP < DAY_AS_SECONDS(180)) {
1770		pr_info("GPU hangs can indicate a bug anywhere in the entire gfx stack, including userspace.\n");
1771		pr_info("Please file a _new_ bug report on bugs.freedesktop.org against DRI -> DRM/Intel\n");
1772		pr_info("drm/i915 developers can then reassign to the right component if it's not a kernel issue.\n");
1773		pr_info("The GPU crash dump is required to analyze GPU hangs, so please always attach it.\n");
1774		pr_info("GPU crash dump saved to /sys/class/drm/card%d/error\n",
1775			i915->drm.primary->index);
1776	}
1777}
1778
1779struct i915_gpu_state *
1780i915_first_error_state(struct drm_i915_private *i915)
1781{
1782	struct i915_gpu_state *error;
 
 
 
 
 
 
1783
1784	spin_lock_irq(&i915->gpu_error.lock);
1785	error = i915->gpu_error.first_error;
1786	if (!IS_ERR_OR_NULL(error))
1787		i915_gpu_state_get(error);
1788	spin_unlock_irq(&i915->gpu_error.lock);
1789
1790	return error;
 
 
 
1791}
1792
1793void i915_reset_error_state(struct drm_i915_private *i915)
1794{
1795	struct i915_gpu_state *error;
 
1796
1797	spin_lock_irq(&i915->gpu_error.lock);
1798	error = i915->gpu_error.first_error;
1799	if (error != ERR_PTR(-ENODEV)) /* if disabled, always disabled */
1800		i915->gpu_error.first_error = NULL;
1801	spin_unlock_irq(&i915->gpu_error.lock);
1802
1803	if (!IS_ERR_OR_NULL(error))
1804		i915_gpu_state_put(error);
1805}
1806
1807void i915_disable_error_state(struct drm_i915_private *i915, int err)
1808{
1809	spin_lock_irq(&i915->gpu_error.lock);
1810	if (!i915->gpu_error.first_error)
1811		i915->gpu_error.first_error = ERR_PTR(err);
1812	spin_unlock_irq(&i915->gpu_error.lock);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1813}