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_FD_COUNT 64
  62 #define MAX_BO_COUNT 64 * 1024
  63
  64 struct bo {
  65    uint32_t size;
  66    uint64_t offset;
  67    void *map;
  68 };
  69
  70 static struct bo *bos;
  71
  72 #define DRM_MAJOR 226
  73
  74 /* We set bit 0 in the map pointer for userptr BOs so we know not to
  75  * munmap them on DRM_IOCTL_GEM_CLOSE.
  76  */
  77 #define USERPTR_FLAG 1
  78 #define IS_USERPTR(p) ((uintptr_t) (p) & USERPTR_FLAG)
  79 #define GET_PTR(p) ( (void *) ((uintptr_t) p & ~(uintptr_t) 1) )
  80
  81 static void __attribute__ ((format(__printf__, 2, 3)))
  82 fail_if(int cond, const char *format, ...)
  83 {
  84    va_list args;
  85
  86    if (!cond)
  87       return;
  88
  89    va_start(args, format);
  90    fprintf(stderr, "intel_dump_gpu: ");
  91    vfprintf(stderr, format, args);
  92    va_end(args);
  93
  94    raise(SIGTRAP);
  95 }
  96
  97 static struct bo *
  98 get_bo(unsigned fd, uint32_t handle)
  99 {
 100    struct bo *bo;
 101
 102    fail_if(handle >= MAX_BO_COUNT, "bo handle too large\n");
 103    fail_if(fd >= MAX_FD_COUNT, "bo fd too large\n");
 104    bo = &bos[handle + fd * MAX_BO_COUNT];
 105
 106    return bo;
 107 }
 108
 109 static inline uint32_t
 110 align_u32(uint32_t v, uint32_t a)
 111 {
 112    return (v + a - 1) & ~(a - 1);
 113 }
 114
 115 static struct gen_device_info devinfo = {0};
 116 static uint32_t device = 0;
 117 static struct aub_file aub_file;
 118
 119 static void *
 120 relocate_bo(int fd, struct bo *bo, const struct drm_i915_gem_execbuffer2 *execbuffer2,
 121             const struct drm_i915_gem_exec_object2 *obj)
 122 {
 123    const struct drm_i915_gem_exec_object2 *exec_objects =
 124       (struct drm_i915_gem_exec_object2 *) (uintptr_t) execbuffer2->buffers_ptr;
 125    const struct drm_i915_gem_relocation_entry *relocs =
 126       (const struct drm_i915_gem_relocation_entry *) (uintptr_t) obj->relocs_ptr;
 127    void *relocated;
 128    int handle;
 129
 130    relocated = malloc(bo->size);
 131    fail_if(relocated == NULL, "out of memory\n");
 132    memcpy(relocated, GET_PTR(bo->map), bo->size);
 133    for (size_t i = 0; i < obj->relocation_count; i++) {
 134       fail_if(relocs[i].offset >= bo->size, "reloc outside bo\n");
 135
 136       if (execbuffer2->flags & I915_EXEC_HANDLE_LUT)
 137          handle = exec_objects[relocs[i].target_handle].handle;
 138       else
 139          handle = relocs[i].target_handle;
 140
 141       aub_write_reloc(&devinfo, ((char *)relocated) + relocs[i].offset,
 142                       get_bo(fd, handle)->offset + relocs[i].delta);
 143    }
 144
 145    return relocated;
 146 }
 147
 148 static int
 149 gem_ioctl(int fd, unsigned long request, void *argp)
 150 {
 151    int ret;
 152
 153    do {
 154       ret = libc_ioctl(fd, request, argp);
 155    } while (ret == -1 && (errno == EINTR || errno == EAGAIN));
 156
 157    return ret;
 158 }
 159
 160 static void *
 161 gem_mmap(int fd, uint32_t handle, uint64_t offset, uint64_t size)
 162 {
 163    struct drm_i915_gem_mmap mmap = {
 164       .handle = handle,
 165       .offset = offset,
 166       .size = size
 167    };
 168
 169    if (gem_ioctl(fd, DRM_IOCTL_I915_GEM_MMAP, &mmap) == -1)
 170       return MAP_FAILED;
 171
 172    return (void *)(uintptr_t) mmap.addr_ptr;
 173 }
 174
 175 static int
 176 gem_get_param(int fd, uint32_t param)
 177 {
 178    int value;
 179    drm_i915_getparam_t gp = {
 180       .param = param,
 181       .value = &value
 182    };
 183
 184    if (gem_ioctl(fd, DRM_IOCTL_I915_GETPARAM, &gp) == -1)
 185       return 0;
 186
 187    return value;
 188 }
 189
 190 static void
 191 dump_execbuffer2(int fd, struct drm_i915_gem_execbuffer2 *execbuffer2)
 192 {
 193    struct drm_i915_gem_exec_object2 *exec_objects =
 194       (struct drm_i915_gem_exec_object2 *) (uintptr_t) execbuffer2->buffers_ptr;
 195    uint32_t ring_flag = execbuffer2->flags & I915_EXEC_RING_MASK;
 196    uint32_t offset;
 197    struct drm_i915_gem_exec_object2 *obj;
 198    struct bo *bo, *batch_bo;
 199    int batch_index;
 200    void *data;
 201
 202    /* We can't do this at open time as we're not yet authenticated. */
 203    if (device == 0) {
 204       device = gem_get_param(fd, I915_PARAM_CHIPSET_ID);
 205       fail_if(device == 0 || devinfo.gen == 0, "failed to identify chipset\n");
 206    }
 207    if (devinfo.gen == 0) {
 208       fail_if(!gen_get_device_info(device, &devinfo),
 209               "failed to identify chipset=0x%x\n", device);
 210
 211       aub_file_init(&aub_file, output_file, device);
 212       if (verbose == 2)
 213          aub_file.verbose_log_file = stdout;
 214       aub_write_header(&aub_file, program_invocation_short_name);
 215
 216       if (verbose)
 217          printf("[running, output file %s, chipset id 0x%04x, gen %d]\n",
 218                 output_filename, device, devinfo.gen);
 219    }
 220
 221    if (aub_use_execlists(&aub_file))
 222       offset = 0x1000;
 223    else
 224       offset = aub_gtt_size(&aub_file);
 225
 226    if (verbose)
 227       printf("Dumping execbuffer2:\n");
 228
 229    for (uint32_t i = 0; i < execbuffer2->buffer_count; i++) {
 230       obj = &exec_objects[i];
 231       bo = get_bo(fd, obj->handle);
 232
 233       /* If bo->size == 0, this means they passed us an invalid
 234        * buffer.  The kernel will reject it and so should we.
 235        */
 236       if (bo->size == 0) {
 237          if (verbose)
 238             printf("BO #%d is invalid!\n", obj->handle);
 239          return;
 240       }
 241
 242       if (obj->flags & EXEC_OBJECT_PINNED) {
 243          bo->offset = obj->offset;
 244          if (verbose)
 245             printf("BO #%d (%dB) pinned @ 0x%lx\n",
 246                    obj->handle, bo->size, bo->offset);
 247       } else {
 248          if (obj->alignment != 0)
 249             offset = align_u32(offset, obj->alignment);
 250          bo->offset = offset;
 251          if (verbose)
 252             printf("BO #%d (%dB) @ 0x%lx\n", obj->handle,
 253                    bo->size, bo->offset);
 254          offset = align_u32(offset + bo->size + 4095, 4096);
 255       }
 256
 257       if (bo->map == NULL && bo->size > 0)
 258          bo->map = gem_mmap(fd, obj->handle, 0, bo->size);
 259       fail_if(bo->map == MAP_FAILED, "bo mmap failed\n");
 260
 261       if (aub_use_execlists(&aub_file))
 262          aub_map_ppgtt(&aub_file, bo->offset, bo->size);
 263    }
 264
 265    batch_index = (execbuffer2->flags & I915_EXEC_BATCH_FIRST) ? 0 :
 266       execbuffer2->buffer_count - 1;
 267    batch_bo = get_bo(fd, exec_objects[batch_index].handle);
 268    for (uint32_t i = 0; i < execbuffer2->buffer_count; i++) {
 269       obj = &exec_objects[i];
 270       bo = get_bo(fd, obj->handle);
 271
 272       if (obj->relocation_count > 0)
 273          data = relocate_bo(fd, bo, execbuffer2, obj);
 274       else
 275          data = bo->map;
 276
 277       if (bo == batch_bo) {
 278          aub_write_trace_block(&aub_file, AUB_TRACE_TYPE_BATCH,
 279                                GET_PTR(data), bo->size, bo->offset);
 280       } else {
 281          aub_write_trace_block(&aub_file, AUB_TRACE_TYPE_NOTYPE,
 282                                GET_PTR(data), bo->size, bo->offset);
 283       }
 284
 285       if (data != bo->map)
 286          free(data);
 287    }
 288
 289    aub_write_exec(&aub_file,
 290                   batch_bo->offset + execbuffer2->batch_start_offset,
 291                   offset, ring_flag);
 292
 293    if (device_override &&
 294        (execbuffer2->flags & I915_EXEC_FENCE_ARRAY) != 0) {
 295       struct drm_i915_gem_exec_fence *fences =
 296          (void*)(uintptr_t)execbuffer2->cliprects_ptr;
 297       for (uint32_t i = 0; i < execbuffer2->num_cliprects; i++) {
 298          if ((fences[i].flags & I915_EXEC_FENCE_SIGNAL) != 0) {
 299             struct drm_syncobj_array arg = {
 300                .handles = (uintptr_t)&fences[i].handle,
 301                .count_handles = 1,
 302                .pad = 0,
 303             };
 304             libc_ioctl(fd, DRM_IOCTL_SYNCOBJ_SIGNAL, &arg);
 305          }
 306       }
 307    }
 308 }
 309
 310 static void
 311 add_new_bo(unsigned fd, int handle, uint64_t size, void *map)
 312 {
 313    struct bo *bo = &bos[handle + fd * MAX_BO_COUNT];
 314
 315    fail_if(handle >= MAX_BO_COUNT, "bo handle out of range\n");
 316    fail_if(fd >= MAX_FD_COUNT, "bo fd out of range\n");
 317    fail_if(size == 0, "bo size is invalid\n");
 318
 319    bo->size = size;
 320    bo->map = map;
 321 }
 322
 323 static void
 324 remove_bo(int fd, int handle)
 325 {
 326    struct bo *bo = get_bo(fd, handle);
 327
 328    if (bo->map && !IS_USERPTR(bo->map))
 329       munmap(bo->map, bo->size);
 330    bo->size = 0;
 331    bo->map = NULL;
 332 }
 333
 334 __attribute__ ((visibility ("default"))) int
 335 close(int fd)
 336 {
 337    if (fd == drm_fd)
 338       drm_fd = -1;
 339
 340    return libc_close(fd);
 341 }
 342
 343 static void
 344 maybe_init(void)
 345 {
 346    static bool initialized = false;
 347    FILE *config;
 348    char *key, *value;
 349
 350    if (initialized)
 351       return;
 352
 353    initialized = true;
 354
 355    config = fopen(getenv("INTEL_DUMP_GPU_CONFIG"), "r");
 356    while (fscanf(config, "%m[^=]=%m[^\n]\n", &key, &value) != EOF) {
 357       if (!strcmp(key, "verbose")) {
 358          if (!strcmp(value, "1")) {
 359             verbose = 1;
 360          } else if (!strcmp(value, "2")) {
 361             verbose = 2;
 362          }
 363       } else if (!strcmp(key, "device")) {
 364          fail_if(device != 0, "Device/Platform override specified multiple times.");
 365          fail_if(sscanf(value, "%i", &device) != 1,
 366                  "failed to parse device id '%s'",
 367                  value);
 368          device_override = true;
 369       } else if (!strcmp(key, "platform")) {
 370          fail_if(device != 0, "Device/Platform override specified multiple times.");
 371          device = gen_device_name_to_pci_device_id(value);
 372          fail_if(device == -1, "Unknown platform '%s'", value);
 373          device_override = true;
 374       } else if (!strcmp(key, "file")) {
 375          output_filename = strdup(value);
 376          output_file = fopen(output_filename, "w+");
 377          fail_if(output_file == NULL,
 378                  "failed to open file '%s'\n",
 379                  output_filename);
 380       } else {
 381          fprintf(stderr, "unknown option '%s'\n", key);
 382       }
 383
 384       free(key);
 385       free(value);
 386    }
 387    fclose(config);
 388
 389    bos = calloc(MAX_FD_COUNT * MAX_BO_COUNT, sizeof(bos[0]));
 390    fail_if(bos == NULL, "out of memory\n");
 391 }
 392
 393 __attribute__ ((visibility ("default"))) int
 394 ioctl(int fd, unsigned long request, ...)
 395 {
 396    va_list args;
 397    void *argp;
 398    int ret;
 399    struct stat buf;
 400
 401    va_start(args, request);
 402    argp = va_arg(args, void *);
 403    va_end(args);
 404
 405    if (_IOC_TYPE(request) == DRM_IOCTL_BASE &&
 406        drm_fd != fd && fstat(fd, &buf) == 0 &&
 407        (buf.st_mode & S_IFMT) == S_IFCHR && major(buf.st_rdev) == DRM_MAJOR) {
 408       drm_fd = fd;
 409       if (verbose)
 410          printf("[intercept drm ioctl on fd %d]\n", fd);
 411    }
 412
 413    if (fd == drm_fd) {
 414       maybe_init();
 415
 416       switch (request) {
 417       case DRM_IOCTL_I915_GETPARAM: {
 418          struct drm_i915_getparam *getparam = argp;
 419
 420          if (device_override && getparam->param == I915_PARAM_CHIPSET_ID) {
 421             *getparam->value = device;
 422             return 0;
 423          }
 424
 425          ret = libc_ioctl(fd, request, argp);
 426
 427          /* If the application looks up chipset_id
 428           * (they typically do), we'll piggy-back on
 429           * their ioctl and store the id for later
 430           * use. */
 431          if (ret == 0 && getparam->param == I915_PARAM_CHIPSET_ID)
 432             device = *getparam->value;
 433
 434          return ret;
 435       }
 436
 437       case DRM_IOCTL_I915_GEM_EXECBUFFER: {
 438          static bool once;
 439          if (!once) {
 440             fprintf(stderr,
 441                     "application uses DRM_IOCTL_I915_GEM_EXECBUFFER, not handled\n");
 442             once = true;
 443          }
 444          return libc_ioctl(fd, request, argp);
 445       }
 446
 447       case DRM_IOCTL_I915_GEM_EXECBUFFER2:
 448       case DRM_IOCTL_I915_GEM_EXECBUFFER2_WR: {
 449          dump_execbuffer2(fd, argp);
 450          if (device_override)
 451             return 0;
 452
 453          return libc_ioctl(fd, request, argp);
 454       }
 455
 456       case DRM_IOCTL_I915_GEM_CREATE: {
 457          struct drm_i915_gem_create *create = argp;
 458
 459          ret = libc_ioctl(fd, request, argp);
 460          if (ret == 0)
 461             add_new_bo(fd, create->handle, create->size, NULL);
 462
 463          return ret;
 464       }
 465
 466       case DRM_IOCTL_I915_GEM_USERPTR: {
 467          struct drm_i915_gem_userptr *userptr = argp;
 468
 469          ret = libc_ioctl(fd, request, argp);
 470          if (ret == 0)
 471             add_new_bo(fd, userptr->handle, userptr->user_size,
 472                        (void *) (uintptr_t) (userptr->user_ptr | USERPTR_FLAG));
 473
 474          return ret;
 475       }
 476
 477       case DRM_IOCTL_GEM_CLOSE: {
 478          struct drm_gem_close *close = argp;
 479
 480          remove_bo(fd, close->handle);
 481
 482          return libc_ioctl(fd, request, argp);
 483       }
 484
 485       case DRM_IOCTL_GEM_OPEN: {
 486          struct drm_gem_open *open = argp;
 487
 488          ret = libc_ioctl(fd, request, argp);
 489          if (ret == 0)
 490             add_new_bo(fd, open->handle, open->size, NULL);
 491
 492          return ret;
 493       }
 494
 495       case DRM_IOCTL_PRIME_FD_TO_HANDLE: {
 496          struct drm_prime_handle *prime = argp;
 497
 498          ret = libc_ioctl(fd, request, argp);
 499          if (ret == 0) {
 500             off_t size;
 501
 502             size = lseek(prime->fd, 0, SEEK_END);
 503             fail_if(size == -1, "failed to get prime bo size\n");
 504             add_new_bo(fd, prime->handle, size, NULL);
 505
 506          }
 507
 508          return ret;
 509       }
 510
 511       default:
 512          return libc_ioctl(fd, request, argp);
 513       }
 514    } else {
 515       return libc_ioctl(fd, request, argp);
 516    }
 517 }
 518
 519 static void
 520 init(void)
 521 {
 522    libc_close = dlsym(RTLD_NEXT, "close");
 523    libc_ioctl = dlsym(RTLD_NEXT, "ioctl");
 524    fail_if(libc_close == NULL || libc_ioctl == NULL,
 525            "failed to get libc ioctl or close\n");
 526 }
 527
 528 static int
 529 close_init_helper(int fd)
 530 {
 531    init();
 532    return libc_close(fd);
 533 }
 534
 535 static int
 536 ioctl_init_helper(int fd, unsigned long request, ...)
 537 {
 538    va_list args;
 539    void *argp;
 540
 541    va_start(args, request);
 542    argp = va_arg(args, void *);
 543    va_end(args);
 544
 545    init();
 546    return libc_ioctl(fd, request, argp);
 547 }
 548
 549 static void __attribute__ ((destructor))
 550 fini(void)
 551 {
 552    free(output_filename);
 553    aub_file_finish(&aub_file);
 554    free(bos);
 555 }