drivers/gpu/drm/radeon/radeon_cs.c

   1 /*
   2  * Copyright 2008 Jerome Glisse.
   3  * All Rights Reserved.
   4  *
   5  * Permission is hereby granted, free of charge, to any person obtaining a
   6  * copy of this software and associated documentation files (the "Software"),
   7  * to deal in the Software without restriction, including without limitation
   8  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
   9  * and/or sell copies of the Software, and to permit persons to whom the
  10  * Software is furnished to do so, subject to the following conditions:
  11  *
  12  * The above copyright notice and this permission notice (including the next
  13  * paragraph) shall be included in all copies or substantial portions of the
  14  * Software.
  15  *
  16  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  17  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  18  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
  19  * PRECISION INSIGHT AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, DAMAGES OR
  20  * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
  21  * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
  22  * DEALINGS IN THE SOFTWARE.
  23  *
  24  * Authors:
  25  *    Jerome Glisse <glisse@freedesktop.org>
  26  */
  27 #include "drmP.h"
  28 #include "radeon_drm.h"
  29 #include "radeon_reg.h"
  30 #include "radeon.h"
  31
  32 void r100_cs_dump_packet(struct radeon_cs_parser *p,
  33                          struct radeon_cs_packet *pkt);
  34
  35 int radeon_cs_parser_relocs(struct radeon_cs_parser *p)
  36 {
  37         struct drm_device *ddev = p->rdev->ddev;
  38         struct radeon_cs_chunk *chunk;
  39         unsigned i, j;
  40         bool duplicate;
  41
  42         if (p->chunk_relocs_idx == -1) {
  43                 return 0;
  44         }
  45         chunk = &p->chunks[p->chunk_relocs_idx];
  46         /* FIXME: we assume that each relocs use 4 dwords */
  47         p->nrelocs = chunk->length_dw / 4;
  48         p->relocs_ptr = kcalloc(p->nrelocs, sizeof(void *), GFP_KERNEL);
  49         if (p->relocs_ptr == NULL) {
  50                 return -ENOMEM;
  51         }
  52         p->relocs = kcalloc(p->nrelocs, sizeof(struct radeon_cs_reloc), GFP_KERNEL);
  53         if (p->relocs == NULL) {
  54                 return -ENOMEM;
  55         }
  56         for (i = 0; i < p->nrelocs; i++) {
  57                 struct drm_radeon_cs_reloc *r;
  58
  59                 duplicate = false;
  60                 r = (struct drm_radeon_cs_reloc *)&chunk->kdata[i*4];
  61                 for (j = 0; j < p->nrelocs; j++) {
  62                         if (r->handle == p->relocs[j].handle) {
  63                                 p->relocs_ptr[i] = &p->relocs[j];
  64                                 duplicate = true;
  65                                 break;
  66                         }
  67                 }
  68                 if (!duplicate) {
  69                         p->relocs[i].gobj = drm_gem_object_lookup(ddev,
  70                                                                   p->filp,
  71                                                                   r->handle);
  72                         if (p->relocs[i].gobj == NULL) {
  73                                 DRM_ERROR("gem object lookup failed 0x%x\n",
  74                                           r->handle);
  75                                 return -ENOENT;
  76                         }
  77                         p->relocs_ptr[i] = &p->relocs[i];
  78                         p->relocs[i].robj = gem_to_radeon_bo(p->relocs[i].gobj);
  79                         p->relocs[i].lobj.bo = p->relocs[i].robj;
  80                         p->relocs[i].lobj.wdomain = r->write_domain;
  81                         p->relocs[i].lobj.rdomain = r->read_domains;
  82                         p->relocs[i].lobj.tv.bo = &p->relocs[i].robj->tbo;
  83                         p->relocs[i].handle = r->handle;
  84                         p->relocs[i].flags = r->flags;
  85                         radeon_bo_list_add_object(&p->relocs[i].lobj,
  86                                                   &p->validated);
  87                 }
  88         }
  89         return radeon_bo_list_validate(&p->validated);
  90 }
  91
  92 int radeon_cs_parser_init(struct radeon_cs_parser *p, void *data)
  93 {
  94         struct drm_radeon_cs *cs = data;
  95         uint64_t *chunk_array_ptr;
  96         unsigned size, i, flags = 0;
  97
  98         INIT_LIST_HEAD(&p->validated);
  99
 100         if (!cs->num_chunks) {
 101                 return 0;
 102         }
 103
 104         /* get chunks */
 105         p->idx = 0;
 106         p->chunk_ib_idx = -1;
 107         p->chunk_relocs_idx = -1;
 108         p->chunks_array = kcalloc(cs->num_chunks, sizeof(uint64_t), GFP_KERNEL);
 109         if (p->chunks_array == NULL) {
 110                 return -ENOMEM;
 111         }
 112         chunk_array_ptr = (uint64_t *)(unsigned long)(cs->chunks);
 113         if (DRM_COPY_FROM_USER(p->chunks_array, chunk_array_ptr,
 114                                sizeof(uint64_t)*cs->num_chunks)) {
 115                 return -EFAULT;
 116         }
 117         p->nchunks = cs->num_chunks;
 118         p->chunks = kcalloc(p->nchunks, sizeof(struct radeon_cs_chunk), GFP_KERNEL);
 119         if (p->chunks == NULL) {
 120                 return -ENOMEM;
 121         }
 122         for (i = 0; i < p->nchunks; i++) {
 123                 struct drm_radeon_cs_chunk __user **chunk_ptr = NULL;
 124                 struct drm_radeon_cs_chunk user_chunk;
 125                 uint32_t __user *cdata;
 126
 127                 chunk_ptr = (void __user*)(unsigned long)p->chunks_array[i];
 128                 if (DRM_COPY_FROM_USER(&user_chunk, chunk_ptr,
 129                                        sizeof(struct drm_radeon_cs_chunk))) {
 130                         return -EFAULT;
 131                 }
 132                 p->chunks[i].length_dw = user_chunk.length_dw;
 133                 p->chunks[i].kdata = NULL;
 134                 p->chunks[i].chunk_id = user_chunk.chunk_id;
 135
 136                 if (p->chunks[i].chunk_id == RADEON_CHUNK_ID_RELOCS) {
 137                         p->chunk_relocs_idx = i;
 138                 }
 139                 if (p->chunks[i].chunk_id == RADEON_CHUNK_ID_IB) {
 140                         p->chunk_ib_idx = i;
 141                         /* zero length IB isn't useful */
 142                         if (p->chunks[i].length_dw == 0)
 143                                 return -EINVAL;
 144                 }
 145                 if (p->chunks[i].chunk_id == RADEON_CHUNK_ID_FLAGS &&
 146                     !p->chunks[i].length_dw) {
 147                         return -EINVAL;
 148                 }
 149
 150                 p->chunks[i].length_dw = user_chunk.length_dw;
 151                 p->chunks[i].user_ptr = (void __user *)(unsigned long)user_chunk.chunk_data;
 152
 153                 cdata = (uint32_t *)(unsigned long)user_chunk.chunk_data;
 154                 if (p->chunks[i].chunk_id != RADEON_CHUNK_ID_IB) {
 155                         size = p->chunks[i].length_dw * sizeof(uint32_t);
 156                         p->chunks[i].kdata = kmalloc(size, GFP_KERNEL);
 157                         if (p->chunks[i].kdata == NULL) {
 158                                 return -ENOMEM;
 159                         }
 160                         if (DRM_COPY_FROM_USER(p->chunks[i].kdata,
 161                                                p->chunks[i].user_ptr, size)) {
 162                                 return -EFAULT;
 163                         }
 164                         if (p->chunks[i].chunk_id == RADEON_CHUNK_ID_FLAGS) {
 165                                 flags = p->chunks[i].kdata[0];
 166                         }
 167                 } else {
 168                         p->chunks[i].kpage[0] = kmalloc(PAGE_SIZE, GFP_KERNEL);
 169                         p->chunks[i].kpage[1] = kmalloc(PAGE_SIZE, GFP_KERNEL);
 170                         if (p->chunks[i].kpage[0] == NULL || p->chunks[i].kpage[1] == NULL) {
 171                                 kfree(p->chunks[i].kpage[0]);
 172                                 kfree(p->chunks[i].kpage[1]);
 173                                 p->chunks[i].kpage[0] = NULL;
 174                                 p->chunks[i].kpage[1] = NULL;
 175                                 return -ENOMEM;
 176                         }
 177                         p->chunks[i].kpage_idx[0] = -1;
 178                         p->chunks[i].kpage_idx[1] = -1;
 179                         p->chunks[i].last_copied_page = -1;
 180                         p->chunks[i].last_page_index = ((p->chunks[i].length_dw * 4) - 1) / PAGE_SIZE;
 181                 }
 182         }
 183         if (p->chunks[p->chunk_ib_idx].length_dw > (16 * 1024)) {
 184                 DRM_ERROR("cs IB too big: %d\n",
 185                           p->chunks[p->chunk_ib_idx].length_dw);
 186                 return -EINVAL;
 187         }
 188
 189         p->keep_tiling_flags = (flags & RADEON_CS_KEEP_TILING_FLAGS) != 0;
 190         return 0;
 191 }
 192
 193 /**
 194  * cs_parser_fini() - clean parser states
 195  * @parser:     parser structure holding parsing context.
 196  * @error:      error number
 197  *
 198  * If error is set than unvalidate buffer, otherwise just free memory
 199  * used by parsing context.
 200  **/
 201 static void radeon_cs_parser_fini(struct radeon_cs_parser *parser, int error)
 202 {
 203         unsigned i;
 204
 205
 206         if (!error && parser->ib)
 207                 ttm_eu_fence_buffer_objects(&parser->validated,
 208                                             parser->ib->fence);
 209         else
 210                 ttm_eu_backoff_reservation(&parser->validated);
 211
 212         if (parser->relocs != NULL) {
 213                 for (i = 0; i < parser->nrelocs; i++) {
 214                         if (parser->relocs[i].gobj)
 215                                 drm_gem_object_unreference_unlocked(parser->relocs[i].gobj);
 216                 }
 217         }
 218         kfree(parser->track);
 219         kfree(parser->relocs);
 220         kfree(parser->relocs_ptr);
 221         for (i = 0; i < parser->nchunks; i++) {
 222                 kfree(parser->chunks[i].kdata);
 223                 kfree(parser->chunks[i].kpage[0]);
 224                 kfree(parser->chunks[i].kpage[1]);
 225         }
 226         kfree(parser->chunks);
 227         kfree(parser->chunks_array);
 228         radeon_ib_free(parser->rdev, &parser->ib);
 229 }
 230
 231 int radeon_cs_ioctl(struct drm_device *dev, void *data, struct drm_file *filp)
 232 {
 233         struct radeon_device *rdev = dev->dev_private;
 234         struct radeon_cs_parser parser;
 235         struct radeon_cs_chunk *ib_chunk;
 236         int r;
 237
 238         radeon_mutex_lock(&rdev->cs_mutex);
 239         /* initialize parser */
 240         memset(&parser, 0, sizeof(struct radeon_cs_parser));
 241         parser.filp = filp;
 242         parser.rdev = rdev;
 243         parser.dev = rdev->dev;
 244         parser.family = rdev->family;
 245         r = radeon_cs_parser_init(&parser, data);
 246         if (r) {
 247                 DRM_ERROR("Failed to initialize parser !\n");
 248                 radeon_cs_parser_fini(&parser, r);
 249                 radeon_mutex_unlock(&rdev->cs_mutex);
 250                 return r;
 251         }
 252         r =  radeon_ib_get(rdev, &parser.ib);
 253         if (r) {
 254                 DRM_ERROR("Failed to get ib !\n");
 255                 radeon_cs_parser_fini(&parser, r);
 256                 radeon_mutex_unlock(&rdev->cs_mutex);
 257                 return r;
 258         }
 259         r = radeon_cs_parser_relocs(&parser);
 260         if (r) {
 261                 if (r != -ERESTARTSYS)
 262                         DRM_ERROR("Failed to parse relocation %d!\n", r);
 263                 radeon_cs_parser_fini(&parser, r);
 264                 radeon_mutex_unlock(&rdev->cs_mutex);
 265                 return r;
 266         }
 267         /* Copy the packet into the IB, the parser will read from the
 268          * input memory (cached) and write to the IB (which can be
 269          * uncached). */
 270         ib_chunk = &parser.chunks[parser.chunk_ib_idx];
 271         parser.ib->length_dw = ib_chunk->length_dw;
 272         r = radeon_cs_parse(&parser);
 273         if (r || parser.parser_error) {
 274                 DRM_ERROR("Invalid command stream !\n");
 275                 radeon_cs_parser_fini(&parser, r);
 276                 radeon_mutex_unlock(&rdev->cs_mutex);
 277                 return r;
 278         }
 279         r = radeon_cs_finish_pages(&parser);
 280         if (r) {
 281                 DRM_ERROR("Invalid command stream !\n");
 282                 radeon_cs_parser_fini(&parser, r);
 283                 radeon_mutex_unlock(&rdev->cs_mutex);
 284                 return r;
 285         }
 286         r = radeon_ib_schedule(rdev, parser.ib);
 287         if (r) {
 288                 DRM_ERROR("Failed to schedule IB !\n");
 289         }
 290         radeon_cs_parser_fini(&parser, r);
 291         radeon_mutex_unlock(&rdev->cs_mutex);
 292         return r;
 293 }
 294
 295 int radeon_cs_finish_pages(struct radeon_cs_parser *p)
 296 {
 297         struct radeon_cs_chunk *ibc = &p->chunks[p->chunk_ib_idx];
 298         int i;
 299         int size = PAGE_SIZE;
 300
 301         for (i = ibc->last_copied_page + 1; i <= ibc->last_page_index; i++) {
 302                 if (i == ibc->last_page_index) {
 303                         size = (ibc->length_dw * 4) % PAGE_SIZE;
 304                         if (size == 0)
 305                                 size = PAGE_SIZE;
 306                 }
 307
 308                 if (DRM_COPY_FROM_USER(p->ib->ptr + (i * (PAGE_SIZE/4)),
 309                                        ibc->user_ptr + (i * PAGE_SIZE),
 310                                        size))
 311                         return -EFAULT;
 312         }
 313         return 0;
 314 }
 315
 316 int radeon_cs_update_pages(struct radeon_cs_parser *p, int pg_idx)
 317 {
 318         int new_page;
 319         struct radeon_cs_chunk *ibc = &p->chunks[p->chunk_ib_idx];
 320         int i;
 321         int size = PAGE_SIZE;
 322
 323         for (i = ibc->last_copied_page + 1; i < pg_idx; i++) {
 324                 if (DRM_COPY_FROM_USER(p->ib->ptr + (i * (PAGE_SIZE/4)),
 325                                        ibc->user_ptr + (i * PAGE_SIZE),
 326                                        PAGE_SIZE)) {
 327                         p->parser_error = -EFAULT;
 328                         return 0;
 329                 }
 330         }
 331
 332         new_page = ibc->kpage_idx[0] < ibc->kpage_idx[1] ? 0 : 1;
 333
 334         if (pg_idx == ibc->last_page_index) {
 335                 size = (ibc->length_dw * 4) % PAGE_SIZE;
 336                         if (size == 0)
 337                                 size = PAGE_SIZE;
 338         }
 339
 340         if (DRM_COPY_FROM_USER(ibc->kpage[new_page],
 341                                ibc->user_ptr + (pg_idx * PAGE_SIZE),
 342                                size)) {
 343                 p->parser_error = -EFAULT;
 344                 return 0;
 345         }
 346
 347         /* copy to IB here */
 348         memcpy((void *)(p->ib->ptr+(pg_idx*(PAGE_SIZE/4))), ibc->kpage[new_page], size);
 349
 350         ibc->last_copied_page = pg_idx;
 351         ibc->kpage_idx[new_page] = pg_idx;
 352
 353         return new_page;
 354 }