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 -EINVAL;
  76                         }
  77                         p->relocs_ptr[i] = &p->relocs[i];
  78                         p->relocs[i].robj = p->relocs[i].gobj->driver_private;
  79                         p->relocs[i].lobj.bo = p->relocs[i].robj;
  80                         p->relocs[i].lobj.rdomain = r->read_domains;
  81                         p->relocs[i].lobj.wdomain = r->write_domain;
  82                         p->relocs[i].handle = r->handle;
  83                         p->relocs[i].flags = r->flags;
  84                         INIT_LIST_HEAD(&p->relocs[i].lobj.list);
  85                         radeon_bo_list_add_object(&p->relocs[i].lobj,
  86                                                 &p->validated);
  87                 }
  88         }
  89         return radeon_bo_list_validate(&p->validated, p->ib->fence);
  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;
  97
  98         if (!cs->num_chunks) {
  99                 return 0;
 100         }
 101         /* get chunks */
 102         INIT_LIST_HEAD(&p->validated);
 103         p->idx = 0;
 104         p->chunk_ib_idx = -1;
 105         p->chunk_relocs_idx = -1;
 106         p->chunks_array = kcalloc(cs->num_chunks, sizeof(uint64_t), GFP_KERNEL);
 107         if (p->chunks_array == NULL) {
 108                 return -ENOMEM;
 109         }
 110         chunk_array_ptr = (uint64_t *)(unsigned long)(cs->chunks);
 111         if (DRM_COPY_FROM_USER(p->chunks_array, chunk_array_ptr,
 112                                sizeof(uint64_t)*cs->num_chunks)) {
 113                 return -EFAULT;
 114         }
 115         p->nchunks = cs->num_chunks;
 116         p->chunks = kcalloc(p->nchunks, sizeof(struct radeon_cs_chunk), GFP_KERNEL);
 117         if (p->chunks == NULL) {
 118                 return -ENOMEM;
 119         }
 120         for (i = 0; i < p->nchunks; i++) {
 121                 struct drm_radeon_cs_chunk __user **chunk_ptr = NULL;
 122                 struct drm_radeon_cs_chunk user_chunk;
 123                 uint32_t __user *cdata;
 124
 125                 chunk_ptr = (void __user*)(unsigned long)p->chunks_array[i];
 126                 if (DRM_COPY_FROM_USER(&user_chunk, chunk_ptr,
 127                                        sizeof(struct drm_radeon_cs_chunk))) {
 128                         return -EFAULT;
 129                 }
 130                 p->chunks[i].length_dw = user_chunk.length_dw;
 131                 p->chunks[i].kdata = NULL;
 132                 p->chunks[i].chunk_id = user_chunk.chunk_id;
 133
 134                 if (p->chunks[i].chunk_id == RADEON_CHUNK_ID_RELOCS) {
 135                         p->chunk_relocs_idx = i;
 136                 }
 137                 if (p->chunks[i].chunk_id == RADEON_CHUNK_ID_IB) {
 138                         p->chunk_ib_idx = i;
 139                         /* zero length IB isn't useful */
 140                         if (p->chunks[i].length_dw == 0)
 141                                 return -EINVAL;
 142                 }
 143
 144                 p->chunks[i].length_dw = user_chunk.length_dw;
 145                 p->chunks[i].user_ptr = (void __user *)(unsigned long)user_chunk.chunk_data;
 146
 147                 cdata = (uint32_t *)(unsigned long)user_chunk.chunk_data;
 148                 if (p->chunks[i].chunk_id != RADEON_CHUNK_ID_IB) {
 149                         size = p->chunks[i].length_dw * sizeof(uint32_t);
 150                         p->chunks[i].kdata = kmalloc(size, GFP_KERNEL);
 151                         if (p->chunks[i].kdata == NULL) {
 152                                 return -ENOMEM;
 153                         }
 154                         if (DRM_COPY_FROM_USER(p->chunks[i].kdata,
 155                                                p->chunks[i].user_ptr, size)) {
 156                                 return -EFAULT;
 157                         }
 158                 } else {
 159                         p->chunks[i].kpage[0] = kmalloc(PAGE_SIZE, GFP_KERNEL);
 160                         p->chunks[i].kpage[1] = kmalloc(PAGE_SIZE, GFP_KERNEL);
 161                         if (p->chunks[i].kpage[0] == NULL || p->chunks[i].kpage[1] == NULL) {
 162                                 kfree(p->chunks[i].kpage[0]);
 163                                 kfree(p->chunks[i].kpage[1]);
 164                                 return -ENOMEM;
 165                         }
 166                         p->chunks[i].kpage_idx[0] = -1;
 167                         p->chunks[i].kpage_idx[1] = -1;
 168                         p->chunks[i].last_copied_page = -1;
 169                         p->chunks[i].last_page_index = ((p->chunks[i].length_dw * 4) - 1) / PAGE_SIZE;
 170                 }
 171         }
 172         if (p->chunks[p->chunk_ib_idx].length_dw > (16 * 1024)) {
 173                 DRM_ERROR("cs IB too big: %d\n",
 174                           p->chunks[p->chunk_ib_idx].length_dw);
 175                 return -EINVAL;
 176         }
 177         return 0;
 178 }
 179
 180 /**
 181  * cs_parser_fini() - clean parser states
 182  * @parser:     parser structure holding parsing context.
 183  * @error:      error number
 184  *
 185  * If error is set than unvalidate buffer, otherwise just free memory
 186  * used by parsing context.
 187  **/
 188 static void radeon_cs_parser_fini(struct radeon_cs_parser *parser, int error)
 189 {
 190         unsigned i;
 191
 192         if (error) {
 193                 radeon_bo_list_unvalidate(&parser->validated,
 194                                                 parser->ib->fence);
 195         } else {
 196                 radeon_bo_list_unreserve(&parser->validated);
 197         }
 198         for (i = 0; i < parser->nrelocs; i++) {
 199                 if (parser->relocs[i].gobj) {
 200                         mutex_lock(&parser->rdev->ddev->struct_mutex);
 201                         drm_gem_object_unreference(parser->relocs[i].gobj);
 202                         mutex_unlock(&parser->rdev->ddev->struct_mutex);
 203                 }
 204         }
 205         kfree(parser->track);
 206         kfree(parser->relocs);
 207         kfree(parser->relocs_ptr);
 208         for (i = 0; i < parser->nchunks; i++) {
 209                 kfree(parser->chunks[i].kdata);
 210                 kfree(parser->chunks[i].kpage[0]);
 211                 kfree(parser->chunks[i].kpage[1]);
 212         }
 213         kfree(parser->chunks);
 214         kfree(parser->chunks_array);
 215         radeon_ib_free(parser->rdev, &parser->ib);
 216 }
 217
 218 int radeon_cs_ioctl(struct drm_device *dev, void *data, struct drm_file *filp)
 219 {
 220         struct radeon_device *rdev = dev->dev_private;
 221         struct radeon_cs_parser parser;
 222         struct radeon_cs_chunk *ib_chunk;
 223         int r;
 224
 225         mutex_lock(&rdev->cs_mutex);
 226         if (rdev->gpu_lockup) {
 227                 mutex_unlock(&rdev->cs_mutex);
 228                 return -EINVAL;
 229         }
 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         r = radeon_cs_parser_init(&parser, data);
 236         if (r) {
 237                 DRM_ERROR("Failed to initialize parser !\n");
 238                 radeon_cs_parser_fini(&parser, r);
 239                 mutex_unlock(&rdev->cs_mutex);
 240                 return r;
 241         }
 242         r =  radeon_ib_get(rdev, &parser.ib);
 243         if (r) {
 244                 DRM_ERROR("Failed to get ib !\n");
 245                 radeon_cs_parser_fini(&parser, r);
 246                 mutex_unlock(&rdev->cs_mutex);
 247                 return r;
 248         }
 249         r = radeon_cs_parser_relocs(&parser);
 250         if (r) {
 251                 DRM_ERROR("Failed to parse relocation !\n");
 252                 radeon_cs_parser_fini(&parser, r);
 253                 mutex_unlock(&rdev->cs_mutex);
 254                 return r;
 255         }
 256         /* Copy the packet into the IB, the parser will read from the
 257          * input memory (cached) and write to the IB (which can be
 258          * uncached). */
 259         ib_chunk = &parser.chunks[parser.chunk_ib_idx];
 260         parser.ib->length_dw = ib_chunk->length_dw;
 261         r = radeon_cs_parse(&parser);
 262         if (r || parser.parser_error) {
 263                 DRM_ERROR("Invalid command stream !\n");
 264                 radeon_cs_parser_fini(&parser, r);
 265                 mutex_unlock(&rdev->cs_mutex);
 266                 return r;
 267         }
 268         r = radeon_cs_finish_pages(&parser);
 269         if (r) {
 270                 DRM_ERROR("Invalid command stream !\n");
 271                 radeon_cs_parser_fini(&parser, r);
 272                 mutex_unlock(&rdev->cs_mutex);
 273                 return r;
 274         }
 275         r = radeon_ib_schedule(rdev, parser.ib);
 276         if (r) {
 277                 DRM_ERROR("Faild to schedule IB !\n");
 278         }
 279         radeon_cs_parser_fini(&parser, r);
 280         mutex_unlock(&rdev->cs_mutex);
 281         return r;
 282 }
 283
 284 int radeon_cs_finish_pages(struct radeon_cs_parser *p)
 285 {
 286         struct radeon_cs_chunk *ibc = &p->chunks[p->chunk_ib_idx];
 287         int i;
 288         int size = PAGE_SIZE;
 289
 290         for (i = ibc->last_copied_page + 1; i <= ibc->last_page_index; i++) {
 291                 if (i == ibc->last_page_index) {
 292                         size = (ibc->length_dw * 4) % PAGE_SIZE;
 293                         if (size == 0)
 294                                 size = PAGE_SIZE;
 295                 }
 296
 297                 if (DRM_COPY_FROM_USER(p->ib->ptr + (i * (PAGE_SIZE/4)),
 298                                        ibc->user_ptr + (i * PAGE_SIZE),
 299                                        size))
 300                         return -EFAULT;
 301         }
 302         return 0;
 303 }
 304
 305 int radeon_cs_update_pages(struct radeon_cs_parser *p, int pg_idx)
 306 {
 307         int new_page;
 308         struct radeon_cs_chunk *ibc = &p->chunks[p->chunk_ib_idx];
 309         int i;
 310         int size = PAGE_SIZE;
 311
 312         for (i = ibc->last_copied_page + 1; i < pg_idx; i++) {
 313                 if (DRM_COPY_FROM_USER(p->ib->ptr + (i * (PAGE_SIZE/4)),
 314                                        ibc->user_ptr + (i * PAGE_SIZE),
 315                                        PAGE_SIZE)) {
 316                         p->parser_error = -EFAULT;
 317                         return 0;
 318                 }
 319         }
 320
 321         new_page = ibc->kpage_idx[0] < ibc->kpage_idx[1] ? 0 : 1;
 322
 323         if (pg_idx == ibc->last_page_index) {
 324                 size = (ibc->length_dw * 4) % PAGE_SIZE;
 325                         if (size == 0)
 326                                 size = PAGE_SIZE;
 327         }
 328
 329         if (DRM_COPY_FROM_USER(ibc->kpage[new_page],
 330                                ibc->user_ptr + (pg_idx * PAGE_SIZE),
 331                                size)) {
 332                 p->parser_error = -EFAULT;
 333                 return 0;
 334         }
 335
 336         /* copy to IB here */
 337         memcpy((void *)(p->ib->ptr+(pg_idx*(PAGE_SIZE/4))), ibc->kpage[new_page], size);
 338
 339         ibc->last_copied_page = pg_idx;
 340         ibc->kpage_idx[new_page] = pg_idx;
 341
 342         return new_page;
 343 }