1 /************************************************************************** 2 * 3 * Copyright 2006 VMware, Inc. 4 * All Rights Reserved. 5 * 6 * Permission is hereby granted, free of charge, to any person obtaining a 7 * copy of this software and associated documentation files (the 8 * "Software"), to deal in the Software without restriction, including 9 * without limitation the rights to use, copy, modify, merge, publish, 10 * distribute, sub license, and/or sell copies of the Software, and to 11 * permit persons to whom the Software is furnished to do so, subject to 12 * the following conditions: 13 * 14 * The above copyright notice and this permission notice (including the 15 * next paragraph) shall be included in all copies or substantial portions 16 * of the Software. 17 * 18 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS 19 * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF 20 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. 21 * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR 22 * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, 23 * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE 24 * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. 25 * 26 **************************************************************************/ 27 28 /* Provide additional functionality on top of bufmgr buffers: 29 * - 2d semantics and blit operations 30 * - refcounting of buffers for multiple images in a buffer. 31 * - refcounting of buffer mappings. 32 * - some logic for moving the buffers to the best memory pools for 33 * given operations. 34 * 35 * Most of this is to make it easier to implement the fixed-layout 36 * mipmap tree required by intel hardware in the face of GL's 37 * programming interface where each image can be specifed in random 38 * order and it isn't clear what layout the tree should have until the 39 * last moment. 40 */ 41 42 #include <sys/ioctl.h> 43 #include <errno.h> 44 45 #include "main/hash.h" 46 #include "intel_context.h" 47 #include "intel_regions.h" 48 #include "intel_blit.h" 49 #include "intel_buffer_objects.h" 50 #include "intel_bufmgr.h" 51 #include "intel_batchbuffer.h" 52 53 #define FILE_DEBUG_FLAG DEBUG_REGION 54 55 /* This should be set to the maximum backtrace size desired. 56 * Set it to 0 to disable backtrace debugging. 57 */ 58 #define DEBUG_BACKTRACE_SIZE 0 59 60 #if DEBUG_BACKTRACE_SIZE == 0 61 /* Use the standard debug output */ 62 #define _DBG(...) DBG(__VA_ARGS__) 63 #else 64 /* Use backtracing debug output */ 65 #define _DBG(...) {debug_backtrace(); DBG(__VA_ARGS__);} 66 67 /* Backtracing debug support */ 68 #include <execinfo.h> 69 70 static void 71 debug_backtrace(void) 72 { 73 void *trace[DEBUG_BACKTRACE_SIZE]; 74 char **strings = NULL; 75 int traceSize; 76 register int i; 77 78 traceSize = backtrace(trace, DEBUG_BACKTRACE_SIZE); 79 strings = backtrace_symbols(trace, traceSize); 80 if (strings == NULL) { 81 DBG("no backtrace:"); 82 return; 83 } 84 85 /* Spit out all the strings with a colon separator. Ignore 86 * the first, since we don't really care about the call 87 * to debug_backtrace() itself. Skip until the final "/" in 88 * the trace to avoid really long lines. 89 */ 90 for (i = 1; i < traceSize; i++) { 91 char *p = strings[i], *slash = strings[i]; 92 while (*p) { 93 if (*p++ == '/') { 94 slash = p; 95 } 96 } 97 98 DBG("%s:", slash); 99 } 100 101 /* Free up the memory, and we're done */ 102 free(strings); 103 } 104 105 #endif 106 107 static struct intel_region * 108 intel_region_alloc_internal(struct intel_screen *screen, 109 GLuint cpp, 110 GLuint width, GLuint height, GLuint pitch, 111 uint32_t tiling, drm_intel_bo *buffer) 112 { 113 struct intel_region *region; 114 115 region = calloc(sizeof(*region), 1); 116 if (region == NULL) 117 return region; 118 119 region->cpp = cpp; 120 region->width = width; 121 region->height = height; 122 region->pitch = pitch; 123 region->refcount = 1; 124 region->bo = buffer; 125 region->tiling = tiling; 126 127 _DBG("%s <-- %p\n", __func__, region); 128 return region; 129 } 130 131 struct intel_region * 132 intel_region_alloc(struct intel_screen *screen, 133 uint32_t tiling, 134 GLuint cpp, GLuint width, GLuint height, 135 bool expect_accelerated_upload) 136 { 137 drm_intel_bo *buffer; 138 unsigned long flags = 0; 139 unsigned long aligned_pitch; 140 struct intel_region *region; 141 142 if (expect_accelerated_upload) 143 flags |= BO_ALLOC_FOR_RENDER; 144 145 buffer = drm_intel_bo_alloc_tiled(screen->bufmgr, "region", 146 width, height, cpp, 147 &tiling, &aligned_pitch, flags); 148 if (buffer == NULL) 149 return NULL; 150 151 region = intel_region_alloc_internal(screen, cpp, width, height, 152 aligned_pitch, tiling, buffer); 153 if (region == NULL) { 154 drm_intel_bo_unreference(buffer); 155 return NULL; 156 } 157 158 return region; 159 } 160 161 bool 162 intel_region_flink(struct intel_region *region, uint32_t *name) 163 { 164 if (region->name == 0) { 165 if (drm_intel_bo_flink(region->bo, ®ion->name)) 166 return false; 167 } 168 169 *name = region->name; 170 171 return true; 172 } 173 174 struct intel_region * 175 intel_region_alloc_for_handle(struct intel_screen *screen, 176 GLuint cpp, 177 GLuint width, GLuint height, GLuint pitch, 178 GLuint handle, const char *name) 179 { 180 struct intel_region *region; 181 drm_intel_bo *buffer; 182 int ret; 183 uint32_t bit_6_swizzle, tiling; 184 185 buffer = intel_bo_gem_create_from_name(screen->bufmgr, name, handle); 186 if (buffer == NULL) 187 return NULL; 188 ret = drm_intel_bo_get_tiling(buffer, &tiling, &bit_6_swizzle); 189 if (ret != 0) { 190 fprintf(stderr, "Couldn't get tiling of buffer %d (%s): %s\n", 191 handle, name, strerror(-ret)); 192 drm_intel_bo_unreference(buffer); 193 return NULL; 194 } 195 196 region = intel_region_alloc_internal(screen, cpp, 197 width, height, pitch, tiling, buffer); 198 if (region == NULL) { 199 drm_intel_bo_unreference(buffer); 200 return NULL; 201 } 202 203 region->name = handle; 204 205 return region; 206 } 207 208 struct intel_region * 209 intel_region_alloc_for_fd(struct intel_screen *screen, 210 GLuint cpp, 211 GLuint width, GLuint height, GLuint pitch, 212 GLuint size, 213 int fd, const char *name) 214 { 215 struct intel_region *region; 216 drm_intel_bo *buffer; 217 int ret; 218 uint32_t bit_6_swizzle, tiling; 219 220 buffer = drm_intel_bo_gem_create_from_prime(screen->bufmgr, fd, size); 221 if (buffer == NULL) 222 return NULL; 223 ret = drm_intel_bo_get_tiling(buffer, &tiling, &bit_6_swizzle); 224 if (ret != 0) { 225 fprintf(stderr, "Couldn't get tiling of buffer (%s): %s\n", 226 name, strerror(-ret)); 227 drm_intel_bo_unreference(buffer); 228 return NULL; 229 } 230 231 region = intel_region_alloc_internal(screen, cpp, 232 width, height, pitch, tiling, buffer); 233 if (region == NULL) { 234 drm_intel_bo_unreference(buffer); 235 return NULL; 236 } 237 238 return region; 239 } 240 241 void 242 intel_region_reference(struct intel_region **dst, struct intel_region *src) 243 { 244 _DBG("%s: %p(%d) -> %p(%d)\n", __func__, 245 *dst, *dst ? (*dst)->refcount : 0, src, src ? src->refcount : 0); 246 247 if (src != *dst) { 248 if (*dst) 249 intel_region_release(dst); 250 251 if (src) 252 src->refcount++; 253 *dst = src; 254 } 255 } 256 257 void 258 intel_region_release(struct intel_region **region_handle) 259 { 260 struct intel_region *region = *region_handle; 261 262 if (region == NULL) { 263 _DBG("%s NULL\n", __func__); 264 return; 265 } 266 267 _DBG("%s %p %d\n", __func__, region, region->refcount - 1); 268 269 assert(region->refcount > 0); 270 region->refcount--; 271 272 if (region->refcount == 0) { 273 drm_intel_bo_unreference(region->bo); 274 275 free(region); 276 } 277 *region_handle = NULL; 278 } 279 280 /** 281 * This function computes masks that may be used to select the bits of the X 282 * and Y coordinates that indicate the offset within a tile. If the region is 283 * untiled, the masks are set to 0. 284 */ 285 void 286 intel_region_get_tile_masks(struct intel_region *region, 287 uint32_t *mask_x, uint32_t *mask_y, 288 bool map_stencil_as_y_tiled) 289 { 290 int cpp = region->cpp; 291 uint32_t tiling = region->tiling; 292 293 if (map_stencil_as_y_tiled) 294 tiling = I915_TILING_Y; 295 296 switch (tiling) { 297 default: 298 assert(false); 299 case I915_TILING_NONE: 300 *mask_x = *mask_y = 0; 301 break; 302 case I915_TILING_X: 303 *mask_x = 512 / cpp - 1; 304 *mask_y = 7; 305 break; 306 case I915_TILING_Y: 307 *mask_x = 128 / cpp - 1; 308 *mask_y = 31; 309 break; 310 } 311 } 312 313 /** 314 * Compute the offset (in bytes) from the start of the region to the given x 315 * and y coordinate. For tiled regions, caller must ensure that x and y are 316 * multiples of the tile size. 317 */ 318 uint32_t 319 intel_region_get_aligned_offset(struct intel_region *region, uint32_t x, 320 uint32_t y, bool map_stencil_as_y_tiled) 321 { 322 int cpp = region->cpp; 323 uint32_t pitch = region->pitch; 324 uint32_t tiling = region->tiling; 325 326 if (map_stencil_as_y_tiled) { 327 tiling = I915_TILING_Y; 328 329 /* When mapping a W-tiled stencil buffer as Y-tiled, each 64-high W-tile 330 * gets transformed into a 32-high Y-tile. Accordingly, the pitch of 331 * the resulting region is twice the pitch of the original region, since 332 * each row in the Y-tiled view corresponds to two rows in the actual 333 * W-tiled surface. So we need to correct the pitch before computing 334 * the offsets. 335 */ 336 pitch *= 2; 337 } 338 339 switch (tiling) { 340 default: 341 assert(false); 342 case I915_TILING_NONE: 343 return y * pitch + x * cpp; 344 case I915_TILING_X: 345 assert((x % (512 / cpp)) == 0); 346 assert((y % 8) == 0); 347 return y * pitch + x / (512 / cpp) * 4096; 348 case I915_TILING_Y: 349 assert((x % (128 / cpp)) == 0); 350 assert((y % 32) == 0); 351 return y * pitch + x / (128 / cpp) * 4096; 352 } 353 } 354