src/gallium/winsys/r600/drm/radeon_ctx.c

   1 /*
   2  * Copyright 2010 Jerome Glisse <glisse@freedesktop.org>
   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  * on the rights to use, copy, modify, merge, publish, distribute, sub
   8  * license, and/or sell copies of the Software, and to permit persons to whom
   9  * the 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 NON-INFRINGEMENT. IN NO EVENT SHALL
  18  * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
  19  * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
  20  * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
  21  * USE OR OTHER DEALINGS IN THE SOFTWARE.
  22  *
  23  * Authors:
  24  *      Jerome Glisse
  25  */
  26 #include <stdio.h>
  27 #include <stdlib.h>
  28 #include <string.h>
  29 #include "radeon_priv.h"
  30 #include "radeon_drm.h"
  31 #include "bof.h"
  32
  33 static int radeon_ctx_set_bo_new(struct radeon_ctx *ctx, struct radeon_ws_bo *bo)
  34 {
  35         if (ctx->nbo >= RADEON_CTX_MAX_PM4)
  36                 return -EBUSY;
  37         radeon_ws_bo_reference(ctx->radeon, &ctx->bo[ctx->nbo], bo);
  38         ctx->nbo++;
  39         return 0;
  40 }
  41
  42 static struct radeon_ws_bo *radeon_ctx_get_bo(struct radeon_ctx *ctx, unsigned reloc)
  43 {
  44         struct radeon_cs_reloc *greloc;
  45         unsigned i;
  46         struct radeon_ws_bo *bo;
  47
  48         greloc = (void *)(((u8 *)ctx->reloc) + reloc * 4);
  49         for (i = 0; i < ctx->nbo; i++) {
  50                 if (ctx->bo[i]->bo->handle == greloc->handle) {
  51                         radeon_ws_bo_reference(ctx->radeon, &bo, ctx->bo[i]);
  52                         return bo;
  53                 }
  54         }
  55         fprintf(stderr, "%s no bo for reloc[%d 0x%08X] %d\n", __func__, reloc, greloc->handle, ctx->nbo);
  56         return NULL;
  57 }
  58
  59 static void radeon_ctx_get_placement(struct radeon_ctx *ctx, unsigned reloc, u32 *placement)
  60 {
  61         struct radeon_cs_reloc *greloc;
  62         unsigned i;
  63
  64         placement[0] = 0;
  65         placement[1] = 0;
  66         greloc = (void *)(((u8 *)ctx->reloc) + reloc * 4);
  67         for (i = 0; i < ctx->nbo; i++) {
  68                 if (ctx->bo[i]->bo->handle == greloc->handle) {
  69                         placement[0] = greloc->read_domain | greloc->write_domain;
  70                         placement[1] = placement[0];
  71                         return;
  72                 }
  73         }
  74 }
  75
  76 void radeon_ctx_clear(struct radeon_ctx *ctx)
  77 {
  78         for (int i = 0; i < ctx->nbo; i++) {
  79                 radeon_ws_bo_reference(ctx->radeon, &ctx->bo[i], NULL);
  80         }
  81         ctx->ndwords = RADEON_CTX_MAX_PM4;
  82         ctx->cdwords = 0;
  83         ctx->nreloc = 0;
  84         ctx->nbo = 0;
  85 }
  86
  87 struct radeon_ctx *radeon_ctx_init(struct radeon *radeon)
  88 {
  89         struct radeon_ctx *ctx;
  90         if (radeon == NULL)
  91                 return NULL;
  92         ctx = calloc(1, sizeof(struct radeon_ctx));
  93         ctx->radeon = radeon_incref(radeon);
  94         radeon_ctx_clear(ctx);
  95         ctx->pm4 = malloc(RADEON_CTX_MAX_PM4 * 4);
  96         if (ctx->pm4 == NULL) {
  97                 radeon_ctx_fini(ctx);
  98                 return NULL;
  99         }
 100         ctx->reloc = malloc(sizeof(struct radeon_cs_reloc) * RADEON_CTX_MAX_PM4);
 101         if (ctx->reloc == NULL) {
 102                 radeon_ctx_fini(ctx);
 103                 return NULL;
 104         }
 105         ctx->bo = malloc(sizeof(void *) * RADEON_CTX_MAX_PM4);
 106         if (ctx->bo == NULL) {
 107                 radeon_ctx_fini(ctx);
 108                 return NULL;
 109         }
 110         return ctx;
 111 }
 112
 113 void radeon_ctx_fini(struct radeon_ctx *ctx)
 114 {
 115         unsigned i;
 116
 117         if (ctx == NULL)
 118                 return;
 119
 120         for (i = 0; i < ctx->nbo; i++) {
 121                 radeon_ws_bo_reference(ctx->radeon, &ctx->bo[i], NULL);
 122         }
 123         ctx->radeon = radeon_decref(ctx->radeon);
 124         free(ctx->bo);
 125         free(ctx->pm4);
 126         free(ctx->reloc);
 127         free(ctx);
 128 }
 129
 130 static int radeon_ctx_state_bo(struct radeon_ctx *ctx, struct radeon_state *state)
 131 {
 132         unsigned i, j;
 133         int r;
 134
 135         if (state == NULL)
 136                 return 0;
 137         for (i = 0; i < state->nbo; i++) {
 138                 for (j = 0; j < ctx->nbo; j++) {
 139                         if (state->bo[i] == ctx->bo[j])
 140                                 break;
 141                 }
 142                 if (j == ctx->nbo) {
 143                         r = radeon_ctx_set_bo_new(ctx, state->bo[i]);
 144                         if (r)
 145                                 return r;
 146                 }
 147         }
 148         return 0;
 149 }
 150
 151
 152 int radeon_ctx_submit(struct radeon_ctx *ctx)
 153 {
 154         struct drm_radeon_cs drmib;
 155         struct drm_radeon_cs_chunk chunks[2];
 156         uint64_t chunk_array[2];
 157         int r = 0;
 158
 159         if (!ctx->cdwords)
 160                 return 0;
 161 #if 0
 162         for (r = 0; r < ctx->cdwords; r++) {
 163                 fprintf(stderr, "0x%08X\n", ctx->pm4[r]);
 164         }
 165 #endif
 166         drmib.num_chunks = 2;
 167         drmib.chunks = (uint64_t)(uintptr_t)chunk_array;
 168         chunks[0].chunk_id = RADEON_CHUNK_ID_IB;
 169         chunks[0].length_dw = ctx->cdwords;
 170         chunks[0].chunk_data = (uint64_t)(uintptr_t)ctx->pm4;
 171         chunks[1].chunk_id = RADEON_CHUNK_ID_RELOCS;
 172         chunks[1].length_dw = ctx->nreloc * sizeof(struct radeon_cs_reloc) / 4;
 173         chunks[1].chunk_data = (uint64_t)(uintptr_t)ctx->reloc;
 174         chunk_array[0] = (uint64_t)(uintptr_t)&chunks[0];
 175         chunk_array[1] = (uint64_t)(uintptr_t)&chunks[1];
 176 #if 1
 177         r = drmCommandWriteRead(ctx->radeon->fd, DRM_RADEON_CS, &drmib,
 178                                 sizeof(struct drm_radeon_cs));
 179 #endif
 180         return r;
 181 }
 182
 183 static int radeon_ctx_reloc(struct radeon_ctx *ctx, struct radeon_ws_bo *bo,
 184                         unsigned id, unsigned *placement)
 185 {
 186         unsigned i;
 187
 188         for (i = 0; i < ctx->nreloc; i++) {
 189                 if (ctx->reloc[i].handle == bo->bo->handle) {
 190                         ctx->pm4[id] = i * sizeof(struct radeon_cs_reloc) / 4;
 191                         return 0;
 192                 }
 193         }
 194         if (ctx->nreloc >= RADEON_CTX_MAX_PM4) {
 195                 return -EBUSY;
 196         }
 197         ctx->reloc[ctx->nreloc].handle = bo->bo->handle;
 198         ctx->reloc[ctx->nreloc].read_domain = placement[0] | placement [1];
 199         ctx->reloc[ctx->nreloc].write_domain = placement[0] | placement [1];
 200         ctx->reloc[ctx->nreloc].flags = 0;
 201         ctx->pm4[id] = ctx->nreloc * sizeof(struct radeon_cs_reloc) / 4;
 202         ctx->nreloc++;
 203         return 0;
 204 }
 205
 206 static int radeon_ctx_state_schedule(struct radeon_ctx *ctx, struct radeon_state *state)
 207 {
 208         unsigned i, rid, bid, cid;
 209         int r;
 210
 211         if (state == NULL)
 212                 return 0;
 213         if (state->cpm4 > ctx->ndwords) {
 214                 return -EBUSY;
 215         }
 216         memcpy(&ctx->pm4[ctx->cdwords], state->pm4, state->cpm4 * 4);
 217         for (i = 0; i < state->nreloc; i++) {
 218                 rid = state->reloc_pm4_id[i];
 219                 bid = state->reloc_bo_id[i];
 220                 cid = ctx->cdwords + rid;
 221                 r = radeon_ctx_reloc(ctx, state->bo[bid], cid,
 222                                         &state->placement[bid * 2]);
 223                 if (r) {
 224                         fprintf(stderr, "%s state %d failed to reloc\n", __func__, state->stype->stype);
 225                         return r;
 226                 }
 227         }
 228         ctx->cdwords += state->cpm4;
 229         ctx->ndwords -= state->cpm4;
 230         return 0;
 231 }
 232
 233 int radeon_ctx_set_query_state(struct radeon_ctx *ctx, struct radeon_state *state)
 234 {
 235         int r = 0;
 236
 237         /* !!! ONLY ACCEPT QUERY STATE HERE !!! */
 238         r = radeon_state_pm4(state);
 239         if (r)
 240                 return r;
 241         /* BEGIN/END query are balanced in the same cs so account for END
 242          * END query when scheduling BEGIN query
 243          */
 244         switch (state->stype->stype) {
 245         case R600_STATE_QUERY_BEGIN:
 246                 /* is there enough place for begin & end */
 247                 if ((state->cpm4 * 2) > ctx->ndwords)
 248                         return -EBUSY;
 249                 ctx->ndwords -= state->cpm4;
 250                 break;
 251         case R600_STATE_QUERY_END:
 252                 ctx->ndwords += state->cpm4;
 253                 break;
 254         default:
 255                 return -EINVAL;
 256         }
 257         return radeon_ctx_state_schedule(ctx, state);
 258 }
 259
 260 int radeon_ctx_set_draw(struct radeon_ctx *ctx, struct radeon_draw *draw)
 261 {
 262         unsigned previous_cdwords;
 263         int r = 0;
 264         int i;
 265
 266         for (i = 0; i < ctx->radeon->max_states; i++) {
 267                 r = radeon_ctx_state_bo(ctx, draw->state[i]);
 268                 if (r)
 269                         return r;
 270         }
 271         previous_cdwords = ctx->cdwords;
 272         for (i = 0; i < ctx->radeon->max_states; i++) {
 273                 if (draw->state[i]) {
 274                         r = radeon_ctx_state_schedule(ctx, draw->state[i]);
 275                         if (r) {
 276                                 ctx->cdwords = previous_cdwords;
 277                                 return r;
 278                         }
 279                 }
 280         }
 281
 282         return 0;
 283 }
 284
 285 #if 0
 286 int radeon_ctx_pm4(struct radeon_ctx *ctx)
 287 {
 288         unsigned i;
 289         int r;
 290
 291         free(ctx->pm4);
 292         ctx->cpm4 = 0;
 293         ctx->pm4 = malloc(ctx->draw_cpm4 * 4);
 294         if (ctx->pm4 == NULL)
 295                 return -EINVAL;
 296         for (i = 0, ctx->id = 0; i < ctx->nstate; i++) {
 297         }
 298         if (ctx->id != ctx->draw_cpm4) {
 299                 fprintf(stderr, "%s miss predicted pm4 size %d for %d\n",
 300                         __func__, ctx->draw_cpm4, ctx->id);
 301                 return -EINVAL;
 302         }
 303         ctx->cpm4 = ctx->draw_cpm4;
 304         return 0;
 305 }
 306 #endif
 307
 308 void radeon_ctx_dump_bof(struct radeon_ctx *ctx, const char *file)
 309 {
 310         bof_t *bcs, *blob, *array, *bo, *size, *handle, *device_id, *root;
 311         unsigned i;
 312         void *data;
 313
 314         root = device_id = bcs = blob = array = bo = size = handle = NULL;
 315         root = bof_object();
 316         if (root == NULL)
 317                 goto out_err;
 318         device_id = bof_int32(ctx->radeon->device);
 319         if (device_id == NULL)
 320                 return;
 321         if (bof_object_set(root, "device_id", device_id))
 322                 goto out_err;
 323         bof_decref(device_id);
 324         device_id = NULL;
 325         /* dump relocs */
 326         blob = bof_blob(ctx->nreloc * 16, ctx->reloc);
 327         if (blob == NULL)
 328                 goto out_err;
 329         if (bof_object_set(root, "reloc", blob))
 330                 goto out_err;
 331         bof_decref(blob);
 332         blob = NULL;
 333         /* dump cs */
 334         blob = bof_blob(ctx->cdwords * 4, ctx->pm4);
 335         if (blob == NULL)
 336                 goto out_err;
 337         if (bof_object_set(root, "pm4", blob))
 338                 goto out_err;
 339         bof_decref(blob);
 340         blob = NULL;
 341         /* dump bo */
 342         array = bof_array();
 343         if (array == NULL)
 344                 goto out_err;
 345         for (i = 0; i < ctx->nbo; i++) {
 346                 bo = bof_object();
 347                 if (bo == NULL)
 348                         goto out_err;
 349                 size = bof_int32(ctx->bo[i]->bo->size);
 350                 if (size == NULL)
 351                         goto out_err;
 352                 if (bof_object_set(bo, "size", size))
 353                         goto out_err;
 354                 bof_decref(size);
 355                 size = NULL;
 356                 handle = bof_int32(ctx->bo[i]->bo->handle);
 357                 if (handle == NULL)
 358                         goto out_err;
 359                 if (bof_object_set(bo, "handle", handle))
 360                         goto out_err;
 361                 bof_decref(handle);
 362                 handle = NULL;
 363                 data = radeon_ws_bo_map(ctx->radeon, ctx->bo[i], 0, NULL);
 364                 blob = bof_blob(ctx->bo[i]->bo->size, data);
 365                 radeon_ws_bo_unmap(ctx->radeon, ctx->bo[i]);
 366                 if (blob == NULL)
 367                         goto out_err;
 368                 if (bof_object_set(bo, "data", blob))
 369                         goto out_err;
 370                 bof_decref(blob);
 371                 blob = NULL;
 372                 if (bof_array_append(array, bo))
 373                         goto out_err;
 374                 bof_decref(bo);
 375                 bo = NULL;
 376         }
 377         if (bof_object_set(root, "bo", array))
 378                 goto out_err;
 379         bof_dump_file(root, file);
 380 out_err:
 381         bof_decref(blob);
 382         bof_decref(array);
 383         bof_decref(bo);
 384         bof_decref(size);
 385         bof_decref(handle);
 386         bof_decref(device_id);
 387         bof_decref(root);
 388 }