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 "drm-uapi/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 enum drm_i915_gem_engine_class
 176 engine_class_from_ring_flag(uint32_t ring_flag)
 177 {
 178    switch (ring_flag) {
 179    case I915_EXEC_DEFAULT:
 180    case I915_EXEC_RENDER:
 181       return I915_ENGINE_CLASS_RENDER;
 182    case I915_EXEC_BSD:
 183       return I915_ENGINE_CLASS_VIDEO;
 184    case I915_EXEC_BLT:
 185       return I915_ENGINE_CLASS_COPY;
 186    case I915_EXEC_VEBOX:
 187       return I915_ENGINE_CLASS_VIDEO_ENHANCE;
 188    default:
 189       return I915_ENGINE_CLASS_INVALID;
 190    }
 191 }
 192
 193 static void
 194 dump_execbuffer2(int fd, struct drm_i915_gem_execbuffer2 *execbuffer2)
 195 {
 196    struct drm_i915_gem_exec_object2 *exec_objects =
 197       (struct drm_i915_gem_exec_object2 *) (uintptr_t) execbuffer2->buffers_ptr;
 198    uint32_t ring_flag = execbuffer2->flags & I915_EXEC_RING_MASK;
 199    uint32_t offset;
 200    struct drm_i915_gem_exec_object2 *obj;
 201    struct bo *bo, *batch_bo;
 202    int batch_index;
 203    void *data;
 204
 205    /* We can't do this at open time as we're not yet authenticated. */
 206    if (device == 0) {
 207       fail_if(!gen_get_device_info_from_fd(fd, &devinfo),
 208               "failed to identify chipset.\n");
 209       device = devinfo.chipset_id;
 210       aub_file_init(&aub_file, output_file,
 211                     verbose == 2 ? stdout : NULL,
 212                     device, program_invocation_short_name);
 213       aub_write_default_setup(&aub_file);
 214
 215       if (verbose)
 216          printf("[running, output file %s, chipset id 0x%04x, gen %d]\n",
 217                 output_filename, device, devinfo.gen);
 218    }
 219
 220    if (aub_use_execlists(&aub_file))
 221       offset = 0x1000;
 222    else
 223       offset = aub_gtt_size(&aub_file);
 224
 225    if (verbose)
 226       printf("Dumping execbuffer2:\n");
 227
 228    for (uint32_t i = 0; i < execbuffer2->buffer_count; i++) {
 229       obj = &exec_objects[i];
 230       bo = get_bo(fd, obj->handle);
 231
 232       /* If bo->size == 0, this means they passed us an invalid
 233        * buffer.  The kernel will reject it and so should we.
 234        */
 235       if (bo->size == 0) {
 236          if (verbose)
 237             printf("BO #%d is invalid!\n", obj->handle);
 238          return;
 239       }
 240
 241       if (obj->flags & EXEC_OBJECT_PINNED) {
 242          bo->offset = obj->offset;
 243          if (verbose)
 244             printf("BO #%d (%dB) pinned @ 0x%" PRIx64 "\n",
 245                    obj->handle, bo->size, bo->offset);
 246       } else {
 247          if (obj->alignment != 0)
 248             offset = align_u32(offset, obj->alignment);
 249          bo->offset = offset;
 250          if (verbose)
 251             printf("BO #%d (%dB) @ 0x%" PRIx64 "\n", obj->handle,
 252                    bo->size, bo->offset);
 253          offset = align_u32(offset + bo->size + 4095, 4096);
 254       }
 255
 256       if (bo->map == NULL && bo->size > 0)
 257          bo->map = gem_mmap(fd, obj->handle, 0, bo->size);
 258       fail_if(bo->map == MAP_FAILED, "bo mmap failed\n");
 259
 260       if (aub_use_execlists(&aub_file))
 261          aub_map_ppgtt(&aub_file, bo->offset, bo->size);
 262    }
 263
 264    batch_index = (execbuffer2->flags & I915_EXEC_BATCH_FIRST) ? 0 :
 265       execbuffer2->buffer_count - 1;
 266    batch_bo = get_bo(fd, exec_objects[batch_index].handle);
 267    for (uint32_t i = 0; i < execbuffer2->buffer_count; i++) {
 268       obj = &exec_objects[i];
 269       bo = get_bo(fd, obj->handle);
 270
 271       if (obj->relocation_count > 0)
 272          data = relocate_bo(fd, bo, execbuffer2, obj);
 273       else
 274          data = bo->map;
 275
 276       if (bo == batch_bo) {
 277          aub_write_trace_block(&aub_file, AUB_TRACE_TYPE_BATCH,
 278                                GET_PTR(data), bo->size, bo->offset);
 279       } else {
 280          aub_write_trace_block(&aub_file, AUB_TRACE_TYPE_NOTYPE,
 281                                GET_PTR(data), bo->size, bo->offset);
 282       }
 283
 284       if (data != bo->map)
 285          free(data);
 286    }
 287
 288    aub_write_exec(&aub_file,
 289                   batch_bo->offset + execbuffer2->batch_start_offset,
 290                   offset, engine_class_from_ring_flag(ring_flag));
 291
 292    if (device_override &&
 293        (execbuffer2->flags & I915_EXEC_FENCE_ARRAY) != 0) {
 294       struct drm_i915_gem_exec_fence *fences =
 295          (void*)(uintptr_t)execbuffer2->cliprects_ptr;
 296       for (uint32_t i = 0; i < execbuffer2->num_cliprects; i++) {
 297          if ((fences[i].flags & I915_EXEC_FENCE_SIGNAL) != 0) {
 298             struct drm_syncobj_array arg = {
 299                .handles = (uintptr_t)&fences[i].handle,
 300                .count_handles = 1,
 301                .pad = 0,
 302             };
 303             libc_ioctl(fd, DRM_IOCTL_SYNCOBJ_SIGNAL, &arg);
 304          }
 305       }
 306    }
 307 }
 308
 309 static void
 310 add_new_bo(unsigned fd, int handle, uint64_t size, void *map)
 311 {
 312    struct bo *bo = &bos[handle + fd * MAX_BO_COUNT];
 313
 314    fail_if(handle >= MAX_BO_COUNT, "bo handle out of range\n");
 315    fail_if(fd >= MAX_FD_COUNT, "bo fd out of range\n");
 316    fail_if(size == 0, "bo size is invalid\n");
 317
 318    bo->size = size;
 319    bo->map = map;
 320 }
 321
 322 static void
 323 remove_bo(int fd, int handle)
 324 {
 325    struct bo *bo = get_bo(fd, handle);
 326
 327    if (bo->map && !IS_USERPTR(bo->map))
 328       munmap(bo->map, bo->size);
 329    bo->size = 0;
 330    bo->map = NULL;
 331 }
 332
 333 __attribute__ ((visibility ("default"))) int
 334 close(int fd)
 335 {
 336    if (fd == drm_fd)
 337       drm_fd = -1;
 338
 339    return libc_close(fd);
 340 }
 341
 342 static void
 343 maybe_init(void)
 344 {
 345    static bool initialized = false;
 346    FILE *config;
 347    char *key, *value;
 348
 349    if (initialized)
 350       return;
 351
 352    initialized = true;
 353
 354    config = fopen(getenv("INTEL_DUMP_GPU_CONFIG"), "r");
 355    while (fscanf(config, "%m[^=]=%m[^\n]\n", &key, &value) != EOF) {
 356       if (!strcmp(key, "verbose")) {
 357          if (!strcmp(value, "1")) {
 358             verbose = 1;
 359          } else if (!strcmp(value, "2")) {
 360             verbose = 2;
 361          }
 362       } else if (!strcmp(key, "device")) {
 363          fail_if(device != 0, "Device/Platform override specified multiple times.");
 364          fail_if(sscanf(value, "%i", &device) != 1,
 365                  "failed to parse device id '%s'",
 366                  value);
 367          device_override = true;
 368       } else if (!strcmp(key, "platform")) {
 369          fail_if(device != 0, "Device/Platform override specified multiple times.");
 370          device = gen_device_name_to_pci_device_id(value);
 371          fail_if(device == -1, "Unknown platform '%s'", value);
 372          device_override = true;
 373       } else if (!strcmp(key, "file")) {
 374          output_filename = strdup(value);
 375          output_file = fopen(output_filename, "w+");
 376          fail_if(output_file == NULL,
 377                  "failed to open file '%s'\n",
 378                  output_filename);
 379       } else {
 380          fprintf(stderr, "unknown option '%s'\n", key);
 381       }
 382
 383       free(key);
 384       free(value);
 385    }
 386    fclose(config);
 387
 388    bos = calloc(MAX_FD_COUNT * MAX_BO_COUNT, sizeof(bos[0]));
 389    fail_if(bos == NULL, "out of memory\n");
 390 }
 391
 392 __attribute__ ((visibility ("default"))) int
 393 ioctl(int fd, unsigned long request, ...)
 394 {
 395    va_list args;
 396    void *argp;
 397    int ret;
 398    struct stat buf;
 399
 400    va_start(args, request);
 401    argp = va_arg(args, void *);
 402    va_end(args);
 403
 404    if (_IOC_TYPE(request) == DRM_IOCTL_BASE &&
 405        drm_fd != fd && fstat(fd, &buf) == 0 &&
 406        (buf.st_mode & S_IFMT) == S_IFCHR && major(buf.st_rdev) == DRM_MAJOR) {
 407       drm_fd = fd;
 408       if (verbose)
 409          printf("[intercept drm ioctl on fd %d]\n", fd);
 410    }
 411
 412    if (fd == drm_fd) {
 413       maybe_init();
 414
 415       switch (request) {
 416       case DRM_IOCTL_I915_GETPARAM: {
 417          struct drm_i915_getparam *getparam = argp;
 418
 419          if (device_override && getparam->param == I915_PARAM_CHIPSET_ID) {
 420             *getparam->value = device;
 421             return 0;
 422          }
 423
 424          ret = libc_ioctl(fd, request, argp);
 425
 426          /* If the application looks up chipset_id
 427           * (they typically do), we'll piggy-back on
 428           * their ioctl and store the id for later
 429           * use. */
 430          if (ret == 0 && getparam->param == I915_PARAM_CHIPSET_ID)
 431             device = *getparam->value;
 432
 433          return ret;
 434       }
 435
 436       case DRM_IOCTL_I915_GEM_EXECBUFFER: {
 437          static bool once;
 438          if (!once) {
 439             fprintf(stderr,
 440                     "application uses DRM_IOCTL_I915_GEM_EXECBUFFER, not handled\n");
 441             once = true;
 442          }
 443          return libc_ioctl(fd, request, argp);
 444       }
 445
 446       case DRM_IOCTL_I915_GEM_EXECBUFFER2:
 447       case DRM_IOCTL_I915_GEM_EXECBUFFER2_WR: {
 448          dump_execbuffer2(fd, argp);
 449          if (device_override)
 450             return 0;
 451
 452          return libc_ioctl(fd, request, argp);
 453       }
 454
 455       case DRM_IOCTL_I915_GEM_CREATE: {
 456          struct drm_i915_gem_create *create = argp;
 457
 458          ret = libc_ioctl(fd, request, argp);
 459          if (ret == 0)
 460             add_new_bo(fd, create->handle, create->size, NULL);
 461
 462          return ret;
 463       }
 464
 465       case DRM_IOCTL_I915_GEM_USERPTR: {
 466          struct drm_i915_gem_userptr *userptr = argp;
 467
 468          ret = libc_ioctl(fd, request, argp);
 469          if (ret == 0)
 470             add_new_bo(fd, userptr->handle, userptr->user_size,
 471                        (void *) (uintptr_t) (userptr->user_ptr | USERPTR_FLAG));
 472
 473          return ret;
 474       }
 475
 476       case DRM_IOCTL_GEM_CLOSE: {
 477          struct drm_gem_close *close = argp;
 478
 479          remove_bo(fd, close->handle);
 480
 481          return libc_ioctl(fd, request, argp);
 482       }
 483
 484       case DRM_IOCTL_GEM_OPEN: {
 485          struct drm_gem_open *open = argp;
 486
 487          ret = libc_ioctl(fd, request, argp);
 488          if (ret == 0)
 489             add_new_bo(fd, open->handle, open->size, NULL);
 490
 491          return ret;
 492       }
 493
 494       case DRM_IOCTL_PRIME_FD_TO_HANDLE: {
 495          struct drm_prime_handle *prime = argp;
 496
 497          ret = libc_ioctl(fd, request, argp);
 498          if (ret == 0) {
 499             off_t size;
 500
 501             size = lseek(prime->fd, 0, SEEK_END);
 502             fail_if(size == -1, "failed to get prime bo size\n");
 503             add_new_bo(fd, prime->handle, size, NULL);
 504
 505          }
 506
 507          return ret;
 508       }
 509
 510       default:
 511          return libc_ioctl(fd, request, argp);
 512       }
 513    } else {
 514       return libc_ioctl(fd, request, argp);
 515    }
 516 }
 517
 518 static void
 519 init(void)
 520 {
 521    libc_close = dlsym(RTLD_NEXT, "close");
 522    libc_ioctl = dlsym(RTLD_NEXT, "ioctl");
 523    fail_if(libc_close == NULL || libc_ioctl == NULL,
 524            "failed to get libc ioctl or close\n");
 525 }
 526
 527 static int
 528 close_init_helper(int fd)
 529 {
 530    init();
 531    return libc_close(fd);
 532 }
 533
 534 static int
 535 ioctl_init_helper(int fd, unsigned long request, ...)
 536 {
 537    va_list args;
 538    void *argp;
 539
 540    va_start(args, request);
 541    argp = va_arg(args, void *);
 542    va_end(args);
 543
 544    init();
 545    return libc_ioctl(fd, request, argp);
 546 }
 547
 548 static void __attribute__ ((destructor))
 549 fini(void)
 550 {
 551    if (devinfo.gen != 0) {
 552       free(output_filename);
 553       aub_file_finish(&aub_file);
 554       free(bos);
 555    }
 556 }