src/intel/tools/intel_dump_gpu.c

   1 /*
   2  * Copyright © 2015 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
  24 #include <stdlib.h>
  25 #include <stdio.h>
  26 #include <string.h>
  27 #include <stdint.h>
  28 #include <stdbool.h>
  29 #include <signal.h>
  30 #include <stdarg.h>
  31 #include <fcntl.h>
  32 #include <sys/types.h>
  33 #include <sys/sysmacros.h>
  34 #include <sys/stat.h>
  35 #include <sys/ioctl.h>
  36 #include <unistd.h>
  37 #include <errno.h>
  38 #include <sys/mman.h>
  39 #include <dlfcn.h>
  40 #include <i915_drm.h>
  41 #include <inttypes.h>
  42
  43 #include "intel_aub.h"
  44 #include "aub_write.h"
  45
  46 #include "dev/gen_device_info.h"
  47 #include "util/macros.h"
  48
  49 static int close_init_helper(int fd);
  50 static int ioctl_init_helper(int fd, unsigned long request, ...);
  51
  52 static int (*libc_close)(int fd) = close_init_helper;
  53 static int (*libc_ioctl)(int fd, unsigned long request, ...) = ioctl_init_helper;
  54
  55 static int drm_fd = -1;
  56 static char *output_filename = NULL;
  57 static FILE *output_file = NULL;
  58 static int verbose = 0;
  59 static bool device_override;
  60
  61 #define MAX_BO_COUNT 64 * 1024
  62
  63 struct bo {
  64    uint32_t size;
  65    uint64_t offset;
  66    void *map;
  67 };
  68
  69 static struct bo *bos;
  70
  71 #define DRM_MAJOR 226
  72
  73 /* We set bit 0 in the map pointer for userptr BOs so we know not to
  74  * munmap them on DRM_IOCTL_GEM_CLOSE.
  75  */
  76 #define USERPTR_FLAG 1
  77 #define IS_USERPTR(p) ((uintptr_t) (p) & USERPTR_FLAG)
  78 #define GET_PTR(p) ( (void *) ((uintptr_t) p & ~(uintptr_t) 1) )
  79
  80 static void __attribute__ ((format(__printf__, 2, 3)))
  81 fail_if(int cond, const char *format, ...)
  82 {
  83    va_list args;
  84
  85    if (!cond)
  86       return;
  87
  88    va_start(args, format);
  89    fprintf(stderr, "intel_dump_gpu: ");
  90    vfprintf(stderr, format, args);
  91    va_end(args);
  92
  93    raise(SIGTRAP);
  94 }
  95
  96 static struct bo *
  97 get_bo(uint32_t handle)
  98 {
  99    struct bo *bo;
 100
 101    fail_if(handle >= MAX_BO_COUNT, "bo handle too large\n");
 102    bo = &bos[handle];
 103
 104    return bo;
 105 }
 106
 107 static inline uint32_t
 108 align_u32(uint32_t v, uint32_t a)
 109 {
 110    return (v + a - 1) & ~(a - 1);
 111 }
 112
 113 static struct gen_device_info devinfo = {0};
 114 static uint32_t device = 0;
 115 static struct aub_file aub_file;
 116
 117 static void *
 118 relocate_bo(struct bo *bo, const struct drm_i915_gem_execbuffer2 *execbuffer2,
 119             const struct drm_i915_gem_exec_object2 *obj)
 120 {
 121    const struct drm_i915_gem_exec_object2 *exec_objects =
 122       (struct drm_i915_gem_exec_object2 *) (uintptr_t) execbuffer2->buffers_ptr;
 123    const struct drm_i915_gem_relocation_entry *relocs =
 124       (const struct drm_i915_gem_relocation_entry *) (uintptr_t) obj->relocs_ptr;
 125    void *relocated;
 126    int handle;
 127
 128    relocated = malloc(bo->size);
 129    fail_if(relocated == NULL, "out of memory\n");
 130    memcpy(relocated, GET_PTR(bo->map), bo->size);
 131    for (size_t i = 0; i < obj->relocation_count; i++) {
 132       fail_if(relocs[i].offset >= bo->size, "reloc outside bo\n");
 133
 134       if (execbuffer2->flags & I915_EXEC_HANDLE_LUT)
 135          handle = exec_objects[relocs[i].target_handle].handle;
 136       else
 137          handle = relocs[i].target_handle;
 138
 139       aub_write_reloc(&devinfo, ((char *)relocated) + relocs[i].offset,
 140                       get_bo(handle)->offset + relocs[i].delta);
 141    }
 142
 143    return relocated;
 144 }
 145
 146 static int
 147 gem_ioctl(int fd, unsigned long request, void *argp)
 148 {
 149    int ret;
 150
 151    do {
 152       ret = libc_ioctl(fd, request, argp);
 153    } while (ret == -1 && (errno == EINTR || errno == EAGAIN));
 154
 155    return ret;
 156 }
 157
 158 static void *
 159 gem_mmap(int fd, uint32_t handle, uint64_t offset, uint64_t size)
 160 {
 161    struct drm_i915_gem_mmap mmap = {
 162       .handle = handle,
 163       .offset = offset,
 164       .size = size
 165    };
 166
 167    if (gem_ioctl(fd, DRM_IOCTL_I915_GEM_MMAP, &mmap) == -1)
 168       return MAP_FAILED;
 169
 170    return (void *)(uintptr_t) mmap.addr_ptr;
 171 }
 172
 173 static int
 174 gem_get_param(int fd, uint32_t param)
 175 {
 176    int value;
 177    drm_i915_getparam_t gp = {
 178       .param = param,
 179       .value = &value
 180    };
 181
 182    if (gem_ioctl(fd, DRM_IOCTL_I915_GETPARAM, &gp) == -1)
 183       return 0;
 184
 185    return value;
 186 }
 187
 188 static void
 189 dump_execbuffer2(int fd, struct drm_i915_gem_execbuffer2 *execbuffer2)
 190 {
 191    struct drm_i915_gem_exec_object2 *exec_objects =
 192       (struct drm_i915_gem_exec_object2 *) (uintptr_t) execbuffer2->buffers_ptr;
 193    uint32_t ring_flag = execbuffer2->flags & I915_EXEC_RING_MASK;
 194    uint32_t offset;
 195    struct drm_i915_gem_exec_object2 *obj;
 196    struct bo *bo, *batch_bo;
 197    int batch_index;
 198    void *data;
 199
 200    /* We can't do this at open time as we're not yet authenticated. */
 201    if (device == 0) {
 202       device = gem_get_param(fd, I915_PARAM_CHIPSET_ID);
 203       fail_if(device == 0 || devinfo.gen == 0, "failed to identify chipset\n");
 204    }
 205    if (devinfo.gen == 0) {
 206       fail_if(!gen_get_device_info(device, &devinfo),
 207               "failed to identify chipset=0x%x\n", device);
 208
 209       aub_file_init(&aub_file, output_file, device);
 210       if (verbose == 2)
 211          aub_file.verbose_log_file = stdout;
 212       aub_write_header(&aub_file, program_invocation_short_name);
 213
 214       if (verbose)
 215          printf("[running, output file %s, chipset id 0x%04x, gen %d]\n",
 216                 output_filename, device, devinfo.gen);
 217    }
 218
 219    if (aub_use_execlists(&aub_file))
 220       offset = 0x1000;
 221    else
 222       offset = aub_gtt_size(&aub_file);
 223
 224    if (verbose)
 225       printf("Dumping execbuffer2:\n");
 226
 227    for (uint32_t i = 0; i < execbuffer2->buffer_count; i++) {
 228       obj = &exec_objects[i];
 229       bo = get_bo(obj->handle);
 230
 231       /* If bo->size == 0, this means they passed us an invalid
 232        * buffer.  The kernel will reject it and so should we.
 233        */
 234       if (bo->size == 0) {
 235          if (verbose)
 236             printf("BO #%d is invalid!\n", obj->handle);
 237          return;
 238       }
 239
 240       if (obj->flags & EXEC_OBJECT_PINNED) {
 241          bo->offset = obj->offset;
 242          if (verbose)
 243             printf("BO #%d (%dB) pinned @ 0x%lx\n",
 244                    obj->handle, bo->size, bo->offset);
 245       } else {
 246          if (obj->alignment != 0)
 247             offset = align_u32(offset, obj->alignment);
 248          bo->offset = offset;
 249          if (verbose)
 250             printf("BO #%d (%dB) @ 0x%lx\n", obj->handle,
 251                    bo->size, bo->offset);
 252          offset = align_u32(offset + bo->size + 4095, 4096);
 253       }
 254
 255       if (bo->map == NULL && bo->size > 0)
 256          bo->map = gem_mmap(fd, obj->handle, 0, bo->size);
 257       fail_if(bo->map == MAP_FAILED, "bo mmap failed\n");
 258
 259       if (aub_use_execlists(&aub_file))
 260          aub_map_ppgtt(&aub_file, bo->offset, bo->size);
 261    }
 262
 263    batch_index = (execbuffer2->flags & I915_EXEC_BATCH_FIRST) ? 0 :
 264       execbuffer2->buffer_count - 1;
 265    batch_bo = get_bo(exec_objects[batch_index].handle);
 266    for (uint32_t i = 0; i < execbuffer2->buffer_count; i++) {
 267       obj = &exec_objects[i];
 268       bo = get_bo(obj->handle);
 269
 270       if (obj->relocation_count > 0)
 271          data = relocate_bo(bo, execbuffer2, obj);
 272       else
 273          data = bo->map;
 274
 275       if (bo == batch_bo) {
 276          aub_write_trace_block(&aub_file, AUB_TRACE_TYPE_BATCH,
 277                                GET_PTR(data), bo->size, bo->offset);
 278       } else {
 279          aub_write_trace_block(&aub_file, AUB_TRACE_TYPE_NOTYPE,
 280                                GET_PTR(data), bo->size, bo->offset);
 281       }
 282
 283       if (data != bo->map)
 284          free(data);
 285    }
 286
 287    aub_write_exec(&aub_file,
 288                   batch_bo->offset + execbuffer2->batch_start_offset,
 289                   offset, ring_flag);
 290
 291    if (device_override &&
 292        (execbuffer2->flags & I915_EXEC_FENCE_ARRAY) != 0) {
 293       struct drm_i915_gem_exec_fence *fences =
 294          (void*)(uintptr_t)execbuffer2->cliprects_ptr;
 295       for (uint32_t i = 0; i < execbuffer2->num_cliprects; i++) {
 296          if ((fences[i].flags & I915_EXEC_FENCE_SIGNAL) != 0) {
 297             struct drm_syncobj_array arg = {
 298                .handles = (uintptr_t)&fences[i].handle,
 299                .count_handles = 1,
 300                .pad = 0,
 301             };
 302             libc_ioctl(fd, DRM_IOCTL_SYNCOBJ_SIGNAL, &arg);
 303          }
 304       }
 305    }
 306 }
 307
 308 static void
 309 add_new_bo(int handle, uint64_t size, void *map)
 310 {
 311    struct bo *bo = &bos[handle];
 312
 313    fail_if(handle >= MAX_BO_COUNT, "bo handle out of range\n");
 314    fail_if(size == 0, "bo size is invalid\n");
 315
 316    bo->size = size;
 317    bo->map = map;
 318 }
 319
 320 static void
 321 remove_bo(int handle)
 322 {
 323    struct bo *bo = get_bo(handle);
 324
 325    if (bo->map && !IS_USERPTR(bo->map))
 326       munmap(bo->map, bo->size);
 327    bo->size = 0;
 328    bo->map = NULL;
 329 }
 330
 331 __attribute__ ((visibility ("default"))) int
 332 close(int fd)
 333 {
 334    if (fd == drm_fd)
 335       drm_fd = -1;
 336
 337    return libc_close(fd);
 338 }
 339
 340 static void
 341 maybe_init(void)
 342 {
 343    static bool initialized = false;
 344    FILE *config;
 345    char *key, *value;
 346
 347    if (initialized)
 348       return;
 349
 350    initialized = true;
 351
 352    config = fopen(getenv("INTEL_DUMP_GPU_CONFIG"), "r");
 353    while (fscanf(config, "%m[^=]=%m[^\n]\n", &key, &value) != EOF) {
 354       if (!strcmp(key, "verbose")) {
 355          if (!strcmp(value, "1")) {
 356             verbose = 1;
 357          } else if (!strcmp(value, "2")) {
 358             verbose = 2;
 359          }
 360       } else if (!strcmp(key, "device")) {
 361          fail_if(device != 0, "Device/Platform override specified multiple times.");
 362          fail_if(sscanf(value, "%i", &device) != 1,
 363                  "failed to parse device id '%s'",
 364                  value);
 365          device_override = true;
 366       } else if (!strcmp(key, "platform")) {
 367          fail_if(device != 0, "Device/Platform override specified multiple times.");
 368          device = gen_device_name_to_pci_device_id(value);
 369          fail_if(device == -1, "Unknown platform '%s'", value);
 370          device_override = true;
 371       } else if (!strcmp(key, "file")) {
 372          output_filename = strdup(value);
 373          output_file = fopen(output_filename, "w+");
 374          fail_if(output_file == NULL,
 375                  "failed to open file '%s'\n",
 376                  output_filename);
 377       } else {
 378          fprintf(stderr, "unknown option '%s'\n", key);
 379       }
 380
 381       free(key);
 382       free(value);
 383    }
 384    fclose(config);
 385
 386    bos = calloc(MAX_BO_COUNT, sizeof(bos[0]));
 387    fail_if(bos == NULL, "out of memory\n");
 388 }
 389
 390 __attribute__ ((visibility ("default"))) int
 391 ioctl(int fd, unsigned long request, ...)
 392 {
 393    va_list args;
 394    void *argp;
 395    int ret;
 396    struct stat buf;
 397
 398    va_start(args, request);
 399    argp = va_arg(args, void *);
 400    va_end(args);
 401
 402    if (_IOC_TYPE(request) == DRM_IOCTL_BASE &&
 403        drm_fd != fd && fstat(fd, &buf) == 0 &&
 404        (buf.st_mode & S_IFMT) == S_IFCHR && major(buf.st_rdev) == DRM_MAJOR) {
 405       drm_fd = fd;
 406       if (verbose)
 407          printf("[intercept drm ioctl on fd %d]\n", fd);
 408    }
 409
 410    if (fd == drm_fd) {
 411       maybe_init();
 412
 413       switch (request) {
 414       case DRM_IOCTL_I915_GETPARAM: {
 415          struct drm_i915_getparam *getparam = argp;
 416
 417          if (device_override && getparam->param == I915_PARAM_CHIPSET_ID) {
 418             *getparam->value = device;
 419             return 0;
 420          }
 421
 422          ret = libc_ioctl(fd, request, argp);
 423
 424          /* If the application looks up chipset_id
 425           * (they typically do), we'll piggy-back on
 426           * their ioctl and store the id for later
 427           * use. */
 428          if (ret == 0 && getparam->param == I915_PARAM_CHIPSET_ID)
 429             device = *getparam->value;
 430
 431          return ret;
 432       }
 433
 434       case DRM_IOCTL_I915_GEM_EXECBUFFER: {
 435          static bool once;
 436          if (!once) {
 437             fprintf(stderr,
 438                     "application uses DRM_IOCTL_I915_GEM_EXECBUFFER, not handled\n");
 439             once = true;
 440          }
 441          return libc_ioctl(fd, request, argp);
 442       }
 443
 444       case DRM_IOCTL_I915_GEM_EXECBUFFER2:
 445       case DRM_IOCTL_I915_GEM_EXECBUFFER2_WR: {
 446          dump_execbuffer2(fd, argp);
 447          if (device_override)
 448             return 0;
 449
 450          return libc_ioctl(fd, request, argp);
 451       }
 452
 453       case DRM_IOCTL_I915_GEM_CREATE: {
 454          struct drm_i915_gem_create *create = argp;
 455
 456          ret = libc_ioctl(fd, request, argp);
 457          if (ret == 0)
 458             add_new_bo(create->handle, create->size, NULL);
 459
 460          return ret;
 461       }
 462
 463       case DRM_IOCTL_I915_GEM_USERPTR: {
 464          struct drm_i915_gem_userptr *userptr = argp;
 465
 466          ret = libc_ioctl(fd, request, argp);
 467          if (ret == 0)
 468             add_new_bo(userptr->handle, userptr->user_size,
 469                        (void *) (uintptr_t) (userptr->user_ptr | USERPTR_FLAG));
 470          return ret;
 471       }
 472
 473       case DRM_IOCTL_GEM_CLOSE: {
 474          struct drm_gem_close *close = argp;
 475
 476          remove_bo(close->handle);
 477
 478          return libc_ioctl(fd, request, argp);
 479       }
 480
 481       case DRM_IOCTL_GEM_OPEN: {
 482          struct drm_gem_open *open = argp;
 483
 484          ret = libc_ioctl(fd, request, argp);
 485          if (ret == 0)
 486             add_new_bo(open->handle, open->size, NULL);
 487
 488          return ret;
 489       }
 490
 491       case DRM_IOCTL_PRIME_FD_TO_HANDLE: {
 492          struct drm_prime_handle *prime = argp;
 493
 494          ret = libc_ioctl(fd, request, argp);
 495          if (ret == 0) {
 496             off_t size;
 497
 498             size = lseek(prime->fd, 0, SEEK_END);
 499             fail_if(size == -1, "failed to get prime bo size\n");
 500             add_new_bo(prime->handle, size, NULL);
 501          }
 502
 503          return ret;
 504       }
 505
 506       default:
 507          return libc_ioctl(fd, request, argp);
 508       }
 509    } else {
 510       return libc_ioctl(fd, request, argp);
 511    }
 512 }
 513
 514 static void
 515 init(void)
 516 {
 517    libc_close = dlsym(RTLD_NEXT, "close");
 518    libc_ioctl = dlsym(RTLD_NEXT, "ioctl");
 519    fail_if(libc_close == NULL || libc_ioctl == NULL,
 520            "failed to get libc ioctl or close\n");
 521 }
 522
 523 static int
 524 close_init_helper(int fd)
 525 {
 526    init();
 527    return libc_close(fd);
 528 }
 529
 530 static int
 531 ioctl_init_helper(int fd, unsigned long request, ...)
 532 {
 533    va_list args;
 534    void *argp;
 535
 536    va_start(args, request);
 537    argp = va_arg(args, void *);
 538    va_end(args);
 539
 540    init();
 541    return libc_ioctl(fd, request, argp);
 542 }
 543
 544 static void __attribute__ ((destructor))
 545 fini(void)
 546 {
 547    free(output_filename);
 548    aub_file_finish(&aub_file);
 549    free(bos);
 550 }