tests/gem_tiled_partial_pwrite_pread.c

   1 /*
   2  * Copyright © 2011 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  *    Daniel Vetter <daniel.vetter@ffwll.ch>
  25  *
  26  */
  27
  28 #include <stdlib.h>
  29 #include <stdio.h>
  30 #include <string.h>
  31 #include <fcntl.h>
  32 #include <inttypes.h>
  33 #include <errno.h>
  34 #include <sys/stat.h>
  35 #include <sys/time.h>
  36
  37 #include <drm.h>
  38
  39 #include "ioctl_wrappers.h"
  40 #include "drmtest.h"
  41 #include "intel_chipset.h"
  42 #include "intel_io.h"
  43 #include "igt_aux.h"
  44
  45 /*
  46  * Testcase: pwrite/pread consistency when touching partial cachelines
  47  *
  48  * Some fancy new pwrite/pread optimizations clflush in-line while
  49  * reading/writing. Check whether all required clflushes happen.
  50  *
  51  * Unfortunately really old mesa used unaligned pread/pwrite for s/w fallback
  52  * rendering, so we need to check whether this works on tiled buffers, too.
  53  *
  54  */
  55
  56 static drm_intel_bufmgr *bufmgr;
  57 struct intel_batchbuffer *batch;
  58
  59 drm_intel_bo *scratch_bo;
  60 drm_intel_bo *staging_bo;
  61 drm_intel_bo *tiled_staging_bo;
  62 unsigned long scratch_pitch;
  63 #define BO_SIZE (32*4096)
  64 uint32_t devid;
  65 uint64_t mappable_gtt_limit;
  66 int fd;
  67
  68 static void
  69 copy_bo(drm_intel_bo *src, int src_tiled,
  70         drm_intel_bo *dst, int dst_tiled)
  71 {
  72         unsigned long dst_pitch = scratch_pitch;
  73         unsigned long src_pitch = scratch_pitch;
  74         uint32_t cmd_bits = 0;
  75
  76         /* dst is tiled ... */
  77         if (intel_gen(devid) >= 4 && dst_tiled) {
  78                 dst_pitch /= 4;
  79                 cmd_bits |= XY_SRC_COPY_BLT_DST_TILED;
  80         }
  81
  82         if (intel_gen(devid) >= 4 && dst_tiled) {
  83                 src_pitch /= 4;
  84                 cmd_bits |= XY_SRC_COPY_BLT_SRC_TILED;
  85         }
  86
  87         BLIT_COPY_BATCH_START(devid, cmd_bits);
  88         OUT_BATCH((3 << 24) | /* 32 bits */
  89                   (0xcc << 16) | /* copy ROP */
  90                   dst_pitch);
  91         OUT_BATCH(0 << 16 | 0);
  92         OUT_BATCH(BO_SIZE/scratch_pitch << 16 | 1024);
  93         OUT_RELOC_FENCED(dst, I915_GEM_DOMAIN_RENDER, I915_GEM_DOMAIN_RENDER, 0);
  94         BLIT_RELOC_UDW(devid);
  95         OUT_BATCH(0 << 16 | 0);
  96         OUT_BATCH(src_pitch);
  97         OUT_RELOC_FENCED(src, I915_GEM_DOMAIN_RENDER, 0, 0);
  98         BLIT_RELOC_UDW(devid);
  99         ADVANCE_BATCH();
 100
 101         intel_batchbuffer_flush(batch);
 102 }
 103
 104 static void
 105 blt_bo_fill(drm_intel_bo *tmp_bo, drm_intel_bo *bo, int val)
 106 {
 107         uint8_t *gtt_ptr;
 108         int i;
 109
 110         drm_intel_gem_bo_map_gtt(tmp_bo);
 111         gtt_ptr = tmp_bo->virtual;
 112
 113         for (i = 0; i < BO_SIZE; i++)
 114                 gtt_ptr[i] = val;
 115
 116         drm_intel_gem_bo_unmap_gtt(tmp_bo);
 117
 118         if (bo->offset < mappable_gtt_limit &&
 119             (IS_G33(devid) || intel_gen(devid) >= 4))
 120                 igt_trash_aperture();
 121
 122         copy_bo(tmp_bo, 0, bo, 1);
 123 }
 124
 125 #define MAX_BLT_SIZE 128
 126 #define ROUNDS 200
 127 uint8_t tmp[BO_SIZE];
 128 uint8_t compare_tmp[BO_SIZE];
 129
 130 static void test_partial_reads(void)
 131 {
 132         int i, j;
 133
 134         printf("checking partial reads\n");
 135         for (i = 0; i < ROUNDS; i++) {
 136                 int start, len;
 137                 int val = i % 256;
 138
 139                 blt_bo_fill(staging_bo, scratch_bo, i);
 140
 141                 start = random() % BO_SIZE;
 142                 len = random() % (BO_SIZE-start) + 1;
 143
 144                 drm_intel_bo_get_subdata(scratch_bo, start, len, tmp);
 145                 for (j = 0; j < len; j++) {
 146                         igt_assert_f(tmp[j] == val,
 147                                      "mismatch at %i, got: %i, expected: %i\n",
 148                                      start + j, tmp[j], val);
 149                 }
 150
 151                 igt_progress("partial reads test: ", i, ROUNDS);
 152         }
 153 }
 154
 155 static void test_partial_writes(void)
 156 {
 157         int i, j;
 158
 159         printf("checking partial writes\n");
 160         for (i = 0; i < ROUNDS; i++) {
 161                 int start, len;
 162                 int val = i % 256;
 163
 164                 blt_bo_fill(staging_bo, scratch_bo, i);
 165
 166                 start = random() % BO_SIZE;
 167                 len = random() % (BO_SIZE-start) + 1;
 168
 169                 memset(tmp, i + 63, BO_SIZE);
 170
 171                 drm_intel_bo_subdata(scratch_bo, start, len, tmp);
 172
 173                 copy_bo(scratch_bo, 1, tiled_staging_bo, 1);
 174                 drm_intel_bo_get_subdata(tiled_staging_bo, 0, BO_SIZE,
 175                                          compare_tmp);
 176
 177                 for (j = 0; j < start; j++) {
 178                         igt_assert_f(compare_tmp[j] == val,
 179                                      "mismatch at %i, got: %i, expected: %i\n",
 180                                      j, tmp[j], val);
 181                 }
 182                 for (; j < start + len; j++) {
 183                         igt_assert_f(compare_tmp[j] == tmp[0],
 184                                      "mismatch at %i, got: %i, expected: %i\n",
 185                                      j, tmp[j], i);
 186                 }
 187                 for (; j < BO_SIZE; j++) {
 188                         igt_assert_f(compare_tmp[j] == val,
 189                                      "mismatch at %i, got: %i, expected: %i\n",
 190                                      j, tmp[j], val);
 191                 }
 192                 drm_intel_gem_bo_unmap_gtt(staging_bo);
 193
 194                 igt_progress("partial writes test: ", i, ROUNDS);
 195         }
 196 }
 197
 198 static void test_partial_read_writes(void)
 199 {
 200         int i, j;
 201
 202         printf("checking partial writes after partial reads\n");
 203         for (i = 0; i < ROUNDS; i++) {
 204                 int start, len;
 205                 int val = i % 256;
 206
 207                 blt_bo_fill(staging_bo, scratch_bo, i);
 208
 209                 /* partial read */
 210                 start = random() % BO_SIZE;
 211                 len = random() % (BO_SIZE-start) + 1;
 212
 213                 drm_intel_bo_get_subdata(scratch_bo, start, len, tmp);
 214                 for (j = 0; j < len; j++) {
 215                         igt_assert_f(tmp[j] == val,
 216                                      "mismatch in read at %i, got: %i, expected: %i\n",
 217                                      start + j, tmp[j], val);
 218                 }
 219
 220                 /* Change contents through gtt to make the pread cachelines
 221                  * stale. */
 222                 val = (i + 17) % 256;
 223                 blt_bo_fill(staging_bo, scratch_bo, val);
 224
 225                 /* partial write */
 226                 start = random() % BO_SIZE;
 227                 len = random() % (BO_SIZE-start) + 1;
 228
 229                 memset(tmp, i + 63, BO_SIZE);
 230
 231                 drm_intel_bo_subdata(scratch_bo, start, len, tmp);
 232
 233                 copy_bo(scratch_bo, 1, tiled_staging_bo, 1);
 234                 drm_intel_bo_get_subdata(tiled_staging_bo, 0, BO_SIZE,
 235                                          compare_tmp);
 236
 237                 for (j = 0; j < start; j++) {
 238                         igt_assert_f(compare_tmp[j] == val,
 239                                      "mismatch at %i, got: %i, expected: %i\n",
 240                                      j, tmp[j], val);
 241                 }
 242                 for (; j < start + len; j++) {
 243                         igt_assert_f(compare_tmp[j] == tmp[0],
 244                                      "mismatch at %i, got: %i, expected: %i\n",
 245                                      j, tmp[j], tmp[0]);
 246                 }
 247                 for (; j < BO_SIZE; j++) {
 248                         igt_assert_f(compare_tmp[j] == val,
 249                                      "mismatch at %i, got: %i, expected: %i\n",
 250                                      j, tmp[j], val);
 251                 }
 252                 drm_intel_gem_bo_unmap_gtt(staging_bo);
 253
 254                 igt_progress("partial read/writes test: ", i, ROUNDS);
 255         }
 256 }
 257
 258 igt_main
 259 {
 260         uint32_t tiling_mode = I915_TILING_X;
 261
 262         igt_skip_on_simulation();
 263
 264         srandom(0xdeadbeef);
 265
 266         igt_fixture {
 267                 fd = drm_open_any();
 268
 269                 bufmgr = drm_intel_bufmgr_gem_init(fd, 4096);
 270                 //drm_intel_bufmgr_gem_enable_reuse(bufmgr);
 271                 devid = intel_get_drm_devid(fd);
 272                 batch = intel_batchbuffer_alloc(bufmgr, devid);
 273
 274                 /* overallocate the buffers we're actually using because */
 275                 scratch_bo = drm_intel_bo_alloc_tiled(bufmgr, "scratch bo", 1024,
 276                                                       BO_SIZE/4096, 4,
 277                                                       &tiling_mode, &scratch_pitch, 0);
 278                 igt_assert(tiling_mode == I915_TILING_X);
 279                 igt_assert(scratch_pitch == 4096);
 280                 staging_bo = drm_intel_bo_alloc(bufmgr, "staging bo", BO_SIZE, 4096);
 281                 tiled_staging_bo = drm_intel_bo_alloc_tiled(bufmgr, "scratch bo", 1024,
 282                                                             BO_SIZE/4096, 4,
 283                                                             &tiling_mode,
 284                                                             &scratch_pitch, 0);
 285
 286                 igt_init_aperture_trashers(bufmgr);
 287                 mappable_gtt_limit = gem_mappable_aperture_size();
 288         }
 289
 290         igt_subtest("reads")
 291                 test_partial_reads();
 292
 293         igt_subtest("writes")
 294                 test_partial_writes();
 295
 296         igt_subtest("writes-after-reads")
 297                 test_partial_read_writes();
 298
 299         igt_fixture {
 300                 igt_cleanup_aperture_trashers();
 301                 drm_intel_bufmgr_destroy(bufmgr);
 302
 303                 close(fd);
 304         }
 305 }