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 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 enum drm_i915_gem_engine_class
 191 engine_class_from_ring_flag(uint32_t ring_flag)
 192 {
 193    switch (ring_flag) {
 194    case I915_EXEC_DEFAULT:
 195    case I915_EXEC_RENDER:
 196       return I915_ENGINE_CLASS_RENDER;
 197    case I915_EXEC_BSD:
 198       return I915_ENGINE_CLASS_VIDEO;
 199    case I915_EXEC_BLT:
 200       return I915_ENGINE_CLASS_COPY;
 201    case I915_EXEC_VEBOX:
 202       return I915_ENGINE_CLASS_VIDEO_ENHANCE;
 203    default:
 204       return I915_ENGINE_CLASS_INVALID;
 205    }
 206 }
 207
 208 static void
 209 dump_execbuffer2(int fd, struct drm_i915_gem_execbuffer2 *execbuffer2)
 210 {
 211    struct drm_i915_gem_exec_object2 *exec_objects =
 212       (struct drm_i915_gem_exec_object2 *) (uintptr_t) execbuffer2->buffers_ptr;
 213    uint32_t ring_flag = execbuffer2->flags & I915_EXEC_RING_MASK;
 214    uint32_t offset;
 215    struct drm_i915_gem_exec_object2 *obj;
 216    struct bo *bo, *batch_bo;
 217    int batch_index;
 218    void *data;
 219
 220    /* We can't do this at open time as we're not yet authenticated. */
 221    if (device == 0) {
 222       device = gem_get_param(fd, I915_PARAM_CHIPSET_ID);
 223       fail_if(device == 0 || devinfo.gen == 0, "failed to identify chipset\n");
 224    }
 225    if (devinfo.gen == 0) {
 226       fail_if(!gen_get_device_info(device, &devinfo),
 227               "failed to identify chipset=0x%x\n", device);
 228
 229       aub_file_init(&aub_file, output_file,
 230                     verbose == 2 ? stdout : NULL,
 231                     device, program_invocation_short_name);
 232       aub_write_default_setup(&aub_file);
 233
 234       if (verbose)
 235          printf("[running, output file %s, chipset id 0x%04x, gen %d]\n",
 236                 output_filename, device, devinfo.gen);
 237    }
 238
 239    if (aub_use_execlists(&aub_file))
 240       offset = 0x1000;
 241    else
 242       offset = aub_gtt_size(&aub_file);
 243
 244    if (verbose)
 245       printf("Dumping execbuffer2:\n");
 246
 247    for (uint32_t i = 0; i < execbuffer2->buffer_count; i++) {
 248       obj = &exec_objects[i];
 249       bo = get_bo(fd, obj->handle);
 250
 251       /* If bo->size == 0, this means they passed us an invalid
 252        * buffer.  The kernel will reject it and so should we.
 253        */
 254       if (bo->size == 0) {
 255          if (verbose)
 256             printf("BO #%d is invalid!\n", obj->handle);
 257          return;
 258       }
 259
 260       if (obj->flags & EXEC_OBJECT_PINNED) {
 261          bo->offset = obj->offset;
 262          if (verbose)
 263             printf("BO #%d (%dB) pinned @ 0x%" PRIx64 "\n",
 264                    obj->handle, bo->size, bo->offset);
 265       } else {
 266          if (obj->alignment != 0)
 267             offset = align_u32(offset, obj->alignment);
 268          bo->offset = offset;
 269          if (verbose)
 270             printf("BO #%d (%dB) @ 0x%" PRIx64 "\n", obj->handle,
 271                    bo->size, bo->offset);
 272          offset = align_u32(offset + bo->size + 4095, 4096);
 273       }
 274
 275       if (bo->map == NULL && bo->size > 0)
 276          bo->map = gem_mmap(fd, obj->handle, 0, bo->size);
 277       fail_if(bo->map == MAP_FAILED, "bo mmap failed\n");
 278
 279       if (aub_use_execlists(&aub_file))
 280          aub_map_ppgtt(&aub_file, bo->offset, bo->size);
 281    }
 282
 283    batch_index = (execbuffer2->flags & I915_EXEC_BATCH_FIRST) ? 0 :
 284       execbuffer2->buffer_count - 1;
 285    batch_bo = get_bo(fd, exec_objects[batch_index].handle);
 286    for (uint32_t i = 0; i < execbuffer2->buffer_count; i++) {
 287       obj = &exec_objects[i];
 288       bo = get_bo(fd, obj->handle);
 289
 290       if (obj->relocation_count > 0)
 291          data = relocate_bo(fd, bo, execbuffer2, obj);
 292       else
 293          data = bo->map;
 294
 295       if (bo == batch_bo) {
 296          aub_write_trace_block(&aub_file, AUB_TRACE_TYPE_BATCH,
 297                                GET_PTR(data), bo->size, bo->offset);
 298       } else {
 299          aub_write_trace_block(&aub_file, AUB_TRACE_TYPE_NOTYPE,
 300                                GET_PTR(data), bo->size, bo->offset);
 301       }
 302
 303       if (data != bo->map)
 304          free(data);
 305    }
 306
 307    aub_write_exec(&aub_file,
 308                   batch_bo->offset + execbuffer2->batch_start_offset,
 309                   offset, engine_class_from_ring_flag(ring_flag));
 310
 311    if (device_override &&
 312        (execbuffer2->flags & I915_EXEC_FENCE_ARRAY) != 0) {
 313       struct drm_i915_gem_exec_fence *fences =
 314          (void*)(uintptr_t)execbuffer2->cliprects_ptr;
 315       for (uint32_t i = 0; i < execbuffer2->num_cliprects; i++) {
 316          if ((fences[i].flags & I915_EXEC_FENCE_SIGNAL) != 0) {
 317             struct drm_syncobj_array arg = {
 318                .handles = (uintptr_t)&fences[i].handle,
 319                .count_handles = 1,
 320                .pad = 0,
 321             };
 322             libc_ioctl(fd, DRM_IOCTL_SYNCOBJ_SIGNAL, &arg);
 323          }
 324       }
 325    }
 326 }
 327
 328 static void
 329 add_new_bo(unsigned fd, int handle, uint64_t size, void *map)
 330 {
 331    struct bo *bo = &bos[handle + fd * MAX_BO_COUNT];
 332
 333    fail_if(handle >= MAX_BO_COUNT, "bo handle out of range\n");
 334    fail_if(fd >= MAX_FD_COUNT, "bo fd out of range\n");
 335    fail_if(size == 0, "bo size is invalid\n");
 336
 337    bo->size = size;
 338    bo->map = map;
 339 }
 340
 341 static void
 342 remove_bo(int fd, int handle)
 343 {
 344    struct bo *bo = get_bo(fd, handle);
 345
 346    if (bo->map && !IS_USERPTR(bo->map))
 347       munmap(bo->map, bo->size);
 348    bo->size = 0;
 349    bo->map = NULL;
 350 }
 351
 352 __attribute__ ((visibility ("default"))) int
 353 close(int fd)
 354 {
 355    if (fd == drm_fd)
 356       drm_fd = -1;
 357
 358    return libc_close(fd);
 359 }
 360
 361 static void
 362 maybe_init(void)
 363 {
 364    static bool initialized = false;
 365    FILE *config;
 366    char *key, *value;
 367
 368    if (initialized)
 369       return;
 370
 371    initialized = true;
 372
 373    config = fopen(getenv("INTEL_DUMP_GPU_CONFIG"), "r");
 374    while (fscanf(config, "%m[^=]=%m[^\n]\n", &key, &value) != EOF) {
 375       if (!strcmp(key, "verbose")) {
 376          if (!strcmp(value, "1")) {
 377             verbose = 1;
 378          } else if (!strcmp(value, "2")) {
 379             verbose = 2;
 380          }
 381       } else if (!strcmp(key, "device")) {
 382          fail_if(device != 0, "Device/Platform override specified multiple times.");
 383          fail_if(sscanf(value, "%i", &device) != 1,
 384                  "failed to parse device id '%s'",
 385                  value);
 386          device_override = true;
 387       } else if (!strcmp(key, "platform")) {
 388          fail_if(device != 0, "Device/Platform override specified multiple times.");
 389          device = gen_device_name_to_pci_device_id(value);
 390          fail_if(device == -1, "Unknown platform '%s'", value);
 391          device_override = true;
 392       } else if (!strcmp(key, "file")) {
 393          output_filename = strdup(value);
 394          output_file = fopen(output_filename, "w+");
 395          fail_if(output_file == NULL,
 396                  "failed to open file '%s'\n",
 397                  output_filename);
 398       } else {
 399          fprintf(stderr, "unknown option '%s'\n", key);
 400       }
 401
 402       free(key);
 403       free(value);
 404    }
 405    fclose(config);
 406
 407    bos = calloc(MAX_FD_COUNT * MAX_BO_COUNT, sizeof(bos[0]));
 408    fail_if(bos == NULL, "out of memory\n");
 409 }
 410
 411 __attribute__ ((visibility ("default"))) int
 412 ioctl(int fd, unsigned long request, ...)
 413 {
 414    va_list args;
 415    void *argp;
 416    int ret;
 417    struct stat buf;
 418
 419    va_start(args, request);
 420    argp = va_arg(args, void *);
 421    va_end(args);
 422
 423    if (_IOC_TYPE(request) == DRM_IOCTL_BASE &&
 424        drm_fd != fd && fstat(fd, &buf) == 0 &&
 425        (buf.st_mode & S_IFMT) == S_IFCHR && major(buf.st_rdev) == DRM_MAJOR) {
 426       drm_fd = fd;
 427       if (verbose)
 428          printf("[intercept drm ioctl on fd %d]\n", fd);
 429    }
 430
 431    if (fd == drm_fd) {
 432       maybe_init();
 433
 434       switch (request) {
 435       case DRM_IOCTL_I915_GETPARAM: {
 436          struct drm_i915_getparam *getparam = argp;
 437
 438          if (device_override && getparam->param == I915_PARAM_CHIPSET_ID) {
 439             *getparam->value = device;
 440             return 0;
 441          }
 442
 443          ret = libc_ioctl(fd, request, argp);
 444
 445          /* If the application looks up chipset_id
 446           * (they typically do), we'll piggy-back on
 447           * their ioctl and store the id for later
 448           * use. */
 449          if (ret == 0 && getparam->param == I915_PARAM_CHIPSET_ID)
 450             device = *getparam->value;
 451
 452          return ret;
 453       }
 454
 455       case DRM_IOCTL_I915_GEM_EXECBUFFER: {
 456          static bool once;
 457          if (!once) {
 458             fprintf(stderr,
 459                     "application uses DRM_IOCTL_I915_GEM_EXECBUFFER, not handled\n");
 460             once = true;
 461          }
 462          return libc_ioctl(fd, request, argp);
 463       }
 464
 465       case DRM_IOCTL_I915_GEM_EXECBUFFER2:
 466       case DRM_IOCTL_I915_GEM_EXECBUFFER2_WR: {
 467          dump_execbuffer2(fd, argp);
 468          if (device_override)
 469             return 0;
 470
 471          return libc_ioctl(fd, request, argp);
 472       }
 473
 474       case DRM_IOCTL_I915_GEM_CREATE: {
 475          struct drm_i915_gem_create *create = argp;
 476
 477          ret = libc_ioctl(fd, request, argp);
 478          if (ret == 0)
 479             add_new_bo(fd, create->handle, create->size, NULL);
 480
 481          return ret;
 482       }
 483
 484       case DRM_IOCTL_I915_GEM_USERPTR: {
 485          struct drm_i915_gem_userptr *userptr = argp;
 486
 487          ret = libc_ioctl(fd, request, argp);
 488          if (ret == 0)
 489             add_new_bo(fd, userptr->handle, userptr->user_size,
 490                        (void *) (uintptr_t) (userptr->user_ptr | USERPTR_FLAG));
 491
 492          return ret;
 493       }
 494
 495       case DRM_IOCTL_GEM_CLOSE: {
 496          struct drm_gem_close *close = argp;
 497
 498          remove_bo(fd, close->handle);
 499
 500          return libc_ioctl(fd, request, argp);
 501       }
 502
 503       case DRM_IOCTL_GEM_OPEN: {
 504          struct drm_gem_open *open = argp;
 505
 506          ret = libc_ioctl(fd, request, argp);
 507          if (ret == 0)
 508             add_new_bo(fd, open->handle, open->size, NULL);
 509
 510          return ret;
 511       }
 512
 513       case DRM_IOCTL_PRIME_FD_TO_HANDLE: {
 514          struct drm_prime_handle *prime = argp;
 515
 516          ret = libc_ioctl(fd, request, argp);
 517          if (ret == 0) {
 518             off_t size;
 519
 520             size = lseek(prime->fd, 0, SEEK_END);
 521             fail_if(size == -1, "failed to get prime bo size\n");
 522             add_new_bo(fd, prime->handle, size, NULL);
 523
 524          }
 525
 526          return ret;
 527       }
 528
 529       default:
 530          return libc_ioctl(fd, request, argp);
 531       }
 532    } else {
 533       return libc_ioctl(fd, request, argp);
 534    }
 535 }
 536
 537 static void
 538 init(void)
 539 {
 540    libc_close = dlsym(RTLD_NEXT, "close");
 541    libc_ioctl = dlsym(RTLD_NEXT, "ioctl");
 542    fail_if(libc_close == NULL || libc_ioctl == NULL,
 543            "failed to get libc ioctl or close\n");
 544 }
 545
 546 static int
 547 close_init_helper(int fd)
 548 {
 549    init();
 550    return libc_close(fd);
 551 }
 552
 553 static int
 554 ioctl_init_helper(int fd, unsigned long request, ...)
 555 {
 556    va_list args;
 557    void *argp;
 558
 559    va_start(args, request);
 560    argp = va_arg(args, void *);
 561    va_end(args);
 562
 563    init();
 564    return libc_ioctl(fd, request, argp);
 565 }
 566
 567 static void __attribute__ ((destructor))
 568 fini(void)
 569 {
 570    free(output_filename);
 571    aub_file_finish(&aub_file);
 572    free(bos);
 573 }