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                         if (r->read_domains)
  84                                 p->relocs[i].lobj.tv.usage |= TTM_USAGE_READ;
  85                         if (r->write_domain)
  86                                 p->relocs[i].lobj.tv.usage |= TTM_USAGE_WRITE;
  87                         p->relocs[i].handle = r->handle;
  88                         p->relocs[i].flags = r->flags;
  89                         radeon_bo_list_add_object(&p->relocs[i].lobj,
  90                                                   &p->validated);
  91                 }
  92         }
  93         return radeon_bo_list_validate(&p->validated);
  94 }
  95
  96 int radeon_cs_parser_init(struct radeon_cs_parser *p, void *data)
  97 {
  98         struct drm_radeon_cs *cs = data;
  99         uint64_t *chunk_array_ptr;
 100         unsigned size, i;
 101
 102         if (!cs->num_chunks) {
 103                 return 0;
 104         }
 105         /* get chunks */
 106         INIT_LIST_HEAD(&p->validated);
 107         p->idx = 0;
 108         p->chunk_ib_idx = -1;
 109         p->chunk_relocs_idx = -1;
 110         p->chunks_array = kcalloc(cs->num_chunks, sizeof(uint64_t), GFP_KERNEL);
 111         if (p->chunks_array == NULL) {
 112                 return -ENOMEM;
 113         }
 114         chunk_array_ptr = (uint64_t *)(unsigned long)(cs->chunks);
 115         if (DRM_COPY_FROM_USER(p->chunks_array, chunk_array_ptr,
 116                                sizeof(uint64_t)*cs->num_chunks)) {
 117                 return -EFAULT;
 118         }
 119         p->nchunks = cs->num_chunks;
 120         p->chunks = kcalloc(p->nchunks, sizeof(struct radeon_cs_chunk), GFP_KERNEL);
 121         if (p->chunks == NULL) {
 122                 return -ENOMEM;
 123         }
 124         for (i = 0; i < p->nchunks; i++) {
 125                 struct drm_radeon_cs_chunk __user **chunk_ptr = NULL;
 126                 struct drm_radeon_cs_chunk user_chunk;
 127                 uint32_t __user *cdata;
 128
 129                 chunk_ptr = (void __user*)(unsigned long)p->chunks_array[i];
 130                 if (DRM_COPY_FROM_USER(&user_chunk, chunk_ptr,
 131                                        sizeof(struct drm_radeon_cs_chunk))) {
 132                         return -EFAULT;
 133                 }
 134                 p->chunks[i].length_dw = user_chunk.length_dw;
 135                 p->chunks[i].kdata = NULL;
 136                 p->chunks[i].chunk_id = user_chunk.chunk_id;
 137
 138                 if (p->chunks[i].chunk_id == RADEON_CHUNK_ID_RELOCS) {
 139                         p->chunk_relocs_idx = i;
 140                 }
 141                 if (p->chunks[i].chunk_id == RADEON_CHUNK_ID_IB) {
 142                         p->chunk_ib_idx = i;
 143                         /* zero length IB isn't useful */
 144                         if (p->chunks[i].length_dw == 0)
 145                                 return -EINVAL;
 146                 }
 147
 148                 p->chunks[i].length_dw = user_chunk.length_dw;
 149                 p->chunks[i].user_ptr = (void __user *)(unsigned long)user_chunk.chunk_data;
 150
 151                 cdata = (uint32_t *)(unsigned long)user_chunk.chunk_data;
 152                 if (p->chunks[i].chunk_id != RADEON_CHUNK_ID_IB) {
 153                         size = p->chunks[i].length_dw * sizeof(uint32_t);
 154                         p->chunks[i].kdata = kmalloc(size, GFP_KERNEL);
 155                         if (p->chunks[i].kdata == NULL) {
 156                                 return -ENOMEM;
 157                         }
 158                         if (DRM_COPY_FROM_USER(p->chunks[i].kdata,
 159                                                p->chunks[i].user_ptr, size)) {
 160                                 return -EFAULT;
 161                         }
 162                 } else {
 163                         p->chunks[i].kpage[0] = kmalloc(PAGE_SIZE, GFP_KERNEL);
 164                         p->chunks[i].kpage[1] = kmalloc(PAGE_SIZE, GFP_KERNEL);
 165                         if (p->chunks[i].kpage[0] == NULL || p->chunks[i].kpage[1] == NULL) {
 166                                 kfree(p->chunks[i].kpage[0]);
 167                                 kfree(p->chunks[i].kpage[1]);
 168                                 return -ENOMEM;
 169                         }
 170                         p->chunks[i].kpage_idx[0] = -1;
 171                         p->chunks[i].kpage_idx[1] = -1;
 172                         p->chunks[i].last_copied_page = -1;
 173                         p->chunks[i].last_page_index = ((p->chunks[i].length_dw * 4) - 1) / PAGE_SIZE;
 174                 }
 175         }
 176         if (p->chunks[p->chunk_ib_idx].length_dw > (16 * 1024)) {
 177                 DRM_ERROR("cs IB too big: %d\n",
 178                           p->chunks[p->chunk_ib_idx].length_dw);
 179                 return -EINVAL;
 180         }
 181         return 0;
 182 }
 183
 184 /**
 185  * cs_parser_fini() - clean parser states
 186  * @parser:     parser structure holding parsing context.
 187  * @error:      error number
 188  *
 189  * If error is set than unvalidate buffer, otherwise just free memory
 190  * used by parsing context.
 191  **/
 192 static void radeon_cs_parser_fini(struct radeon_cs_parser *parser, int error)
 193 {
 194         unsigned i;
 195
 196
 197         if (!error && parser->ib)
 198                 ttm_eu_fence_buffer_objects(&parser->validated,
 199                                             parser->ib->fence);
 200         else
 201                 ttm_eu_backoff_reservation(&parser->validated);
 202
 203         if (parser->relocs != NULL) {
 204                 for (i = 0; i < parser->nrelocs; i++) {
 205                         if (parser->relocs[i].gobj)
 206                                 drm_gem_object_unreference_unlocked(parser->relocs[i].gobj);
 207                 }
 208         }
 209         kfree(parser->track);
 210         kfree(parser->relocs);
 211         kfree(parser->relocs_ptr);
 212         for (i = 0; i < parser->nchunks; i++) {
 213                 kfree(parser->chunks[i].kdata);
 214                 kfree(parser->chunks[i].kpage[0]);
 215                 kfree(parser->chunks[i].kpage[1]);
 216         }
 217         kfree(parser->chunks);
 218         kfree(parser->chunks_array);
 219         radeon_ib_free(parser->rdev, &parser->ib);
 220 }
 221
 222 int radeon_cs_ioctl(struct drm_device *dev, void *data, struct drm_file *filp)
 223 {
 224         struct radeon_device *rdev = dev->dev_private;
 225         struct radeon_cs_parser parser;
 226         struct radeon_cs_chunk *ib_chunk;
 227         int r;
 228
 229         mutex_lock(&rdev->cs_mutex);
 230         /* initialize parser */
 231         memset(&parser, 0, sizeof(struct radeon_cs_parser));
 232         parser.filp = filp;
 233         parser.rdev = rdev;
 234         parser.dev = rdev->dev;
 235         parser.family = rdev->family;
 236         r = radeon_cs_parser_init(&parser, data);
 237         if (r) {
 238                 DRM_ERROR("Failed to initialize parser !\n");
 239                 radeon_cs_parser_fini(&parser, r);
 240                 mutex_unlock(&rdev->cs_mutex);
 241                 return r;
 242         }
 243         r =  radeon_ib_get(rdev, &parser.ib);
 244         if (r) {
 245                 DRM_ERROR("Failed to get ib !\n");
 246                 radeon_cs_parser_fini(&parser, r);
 247                 mutex_unlock(&rdev->cs_mutex);
 248                 return r;
 249         }
 250         r = radeon_cs_parser_relocs(&parser);
 251         if (r) {
 252                 if (r != -ERESTARTSYS)
 253                         DRM_ERROR("Failed to parse relocation %d!\n", r);
 254                 radeon_cs_parser_fini(&parser, r);
 255                 mutex_unlock(&rdev->cs_mutex);
 256                 return r;
 257         }
 258         /* Copy the packet into the IB, the parser will read from the
 259          * input memory (cached) and write to the IB (which can be
 260          * uncached). */
 261         ib_chunk = &parser.chunks[parser.chunk_ib_idx];
 262         parser.ib->length_dw = ib_chunk->length_dw;
 263         r = radeon_cs_parse(&parser);
 264         if (r || parser.parser_error) {
 265                 DRM_ERROR("Invalid command stream !\n");
 266                 radeon_cs_parser_fini(&parser, r);
 267                 mutex_unlock(&rdev->cs_mutex);
 268                 return r;
 269         }
 270         r = radeon_cs_finish_pages(&parser);
 271         if (r) {
 272                 DRM_ERROR("Invalid command stream !\n");
 273                 radeon_cs_parser_fini(&parser, r);
 274                 mutex_unlock(&rdev->cs_mutex);
 275                 return r;
 276         }
 277         r = radeon_ib_schedule(rdev, parser.ib);
 278         if (r) {
 279                 DRM_ERROR("Failed to schedule IB !\n");
 280         }
 281         radeon_cs_parser_fini(&parser, r);
 282         mutex_unlock(&rdev->cs_mutex);
 283         return r;
 284 }
 285
 286 int radeon_cs_finish_pages(struct radeon_cs_parser *p)
 287 {
 288         struct radeon_cs_chunk *ibc = &p->chunks[p->chunk_ib_idx];
 289         int i;
 290         int size = PAGE_SIZE;
 291
 292         for (i = ibc->last_copied_page + 1; i <= ibc->last_page_index; i++) {
 293                 if (i == ibc->last_page_index) {
 294                         size = (ibc->length_dw * 4) % PAGE_SIZE;
 295                         if (size == 0)
 296                                 size = PAGE_SIZE;
 297                 }
 298
 299                 if (DRM_COPY_FROM_USER(p->ib->ptr + (i * (PAGE_SIZE/4)),
 300                                        ibc->user_ptr + (i * PAGE_SIZE),
 301                                        size))
 302                         return -EFAULT;
 303         }
 304         return 0;
 305 }
 306
 307 int radeon_cs_update_pages(struct radeon_cs_parser *p, int pg_idx)
 308 {
 309         int new_page;
 310         struct radeon_cs_chunk *ibc = &p->chunks[p->chunk_ib_idx];
 311         int i;
 312         int size = PAGE_SIZE;
 313
 314         for (i = ibc->last_copied_page + 1; i < pg_idx; i++) {
 315                 if (DRM_COPY_FROM_USER(p->ib->ptr + (i * (PAGE_SIZE/4)),
 316                                        ibc->user_ptr + (i * PAGE_SIZE),
 317                                        PAGE_SIZE)) {
 318                         p->parser_error = -EFAULT;
 319                         return 0;
 320                 }
 321         }
 322
 323         new_page = ibc->kpage_idx[0] < ibc->kpage_idx[1] ? 0 : 1;
 324
 325         if (pg_idx == ibc->last_page_index) {
 326                 size = (ibc->length_dw * 4) % PAGE_SIZE;
 327                         if (size == 0)
 328                                 size = PAGE_SIZE;
 329         }
 330
 331         if (DRM_COPY_FROM_USER(ibc->kpage[new_page],
 332                                ibc->user_ptr + (pg_idx * PAGE_SIZE),
 333                                size)) {
 334                 p->parser_error = -EFAULT;
 335                 return 0;
 336         }
 337
 338         /* copy to IB here */
 339         memcpy((void *)(p->ib->ptr+(pg_idx*(PAGE_SIZE/4))), ibc->kpage[new_page], size);
 340
 341         ibc->last_copied_page = pg_idx;
 342         ibc->kpage_idx[new_page] = pg_idx;
 343
 344         return new_page;
 345 }