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.robj = 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_object_list_add_object(&p->relocs[i].lobj,
  86                                                       &p->validated);
  87                 }
  88         }
  89         return radeon_object_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].chunk_id = user_chunk.chunk_id;
 131                 if (p->chunks[i].chunk_id == RADEON_CHUNK_ID_RELOCS) {
 132                         p->chunk_relocs_idx = i;
 133                 }
 134                 if (p->chunks[i].chunk_id == RADEON_CHUNK_ID_IB) {
 135                         p->chunk_ib_idx = i;
 136                 }
 137                 p->chunks[i].length_dw = user_chunk.length_dw;
 138                 cdata = (uint32_t *)(unsigned long)user_chunk.chunk_data;
 139
 140                 p->chunks[i].kdata = NULL;
 141                 size = p->chunks[i].length_dw * sizeof(uint32_t);
 142                 p->chunks[i].kdata = kzalloc(size, GFP_KERNEL);
 143                 if (p->chunks[i].kdata == NULL) {
 144                         return -ENOMEM;
 145                 }
 146                 if (DRM_COPY_FROM_USER(p->chunks[i].kdata, cdata, size)) {
 147                         return -EFAULT;
 148                 }
 149         }
 150         if (p->chunks[p->chunk_ib_idx].length_dw > (16 * 1024)) {
 151                 DRM_ERROR("cs IB too big: %d\n",
 152                           p->chunks[p->chunk_ib_idx].length_dw);
 153                 return -EINVAL;
 154         }
 155         return 0;
 156 }
 157
 158 /**
 159  * cs_parser_fini() - clean parser states
 160  * @parser:     parser structure holding parsing context.
 161  * @error:      error number
 162  *
 163  * If error is set than unvalidate buffer, otherwise just free memory
 164  * used by parsing context.
 165  **/
 166 static void radeon_cs_parser_fini(struct radeon_cs_parser *parser, int error)
 167 {
 168         unsigned i;
 169
 170         if (error) {
 171                 radeon_object_list_unvalidate(&parser->validated);
 172         } else {
 173                 radeon_object_list_clean(&parser->validated);
 174         }
 175         for (i = 0; i < parser->nrelocs; i++) {
 176                 if (parser->relocs[i].gobj) {
 177                         mutex_lock(&parser->rdev->ddev->struct_mutex);
 178                         drm_gem_object_unreference(parser->relocs[i].gobj);
 179                         mutex_unlock(&parser->rdev->ddev->struct_mutex);
 180                 }
 181         }
 182         kfree(parser->relocs);
 183         kfree(parser->relocs_ptr);
 184         for (i = 0; i < parser->nchunks; i++) {
 185                 kfree(parser->chunks[i].kdata);
 186         }
 187         kfree(parser->chunks);
 188         kfree(parser->chunks_array);
 189         radeon_ib_free(parser->rdev, &parser->ib);
 190 }
 191
 192 int radeon_cs_ioctl(struct drm_device *dev, void *data, struct drm_file *filp)
 193 {
 194         struct radeon_device *rdev = dev->dev_private;
 195         struct radeon_cs_parser parser;
 196         struct radeon_cs_chunk *ib_chunk;
 197         int r;
 198
 199         mutex_lock(&rdev->cs_mutex);
 200         if (rdev->gpu_lockup) {
 201                 mutex_unlock(&rdev->cs_mutex);
 202                 return -EINVAL;
 203         }
 204         /* initialize parser */
 205         memset(&parser, 0, sizeof(struct radeon_cs_parser));
 206         parser.filp = filp;
 207         parser.rdev = rdev;
 208         r = radeon_cs_parser_init(&parser, data);
 209         if (r) {
 210                 DRM_ERROR("Failed to initialize parser !\n");
 211                 radeon_cs_parser_fini(&parser, r);
 212                 mutex_unlock(&rdev->cs_mutex);
 213                 return r;
 214         }
 215         r =  radeon_ib_get(rdev, &parser.ib);
 216         if (r) {
 217                 DRM_ERROR("Failed to get ib !\n");
 218                 radeon_cs_parser_fini(&parser, r);
 219                 mutex_unlock(&rdev->cs_mutex);
 220                 return r;
 221         }
 222         r = radeon_cs_parser_relocs(&parser);
 223         if (r) {
 224                 DRM_ERROR("Failed to parse relocation !\n");
 225                 radeon_cs_parser_fini(&parser, r);
 226                 mutex_unlock(&rdev->cs_mutex);
 227                 return r;
 228         }
 229         /* Copy the packet into the IB, the parser will read from the
 230          * input memory (cached) and write to the IB (which can be
 231          * uncached). */
 232         ib_chunk = &parser.chunks[parser.chunk_ib_idx];
 233         parser.ib->length_dw = ib_chunk->length_dw;
 234         memcpy((void *)parser.ib->ptr, ib_chunk->kdata, ib_chunk->length_dw*4);
 235         r = radeon_cs_parse(&parser);
 236         if (r) {
 237                 DRM_ERROR("Invalid command stream !\n");
 238                 radeon_cs_parser_fini(&parser, r);
 239                 mutex_unlock(&rdev->cs_mutex);
 240                 return r;
 241         }
 242         r = radeon_ib_schedule(rdev, parser.ib);
 243         if (r) {
 244                 DRM_ERROR("Faild to schedule IB !\n");
 245         }
 246         radeon_cs_parser_fini(&parser, r);
 247         mutex_unlock(&rdev->cs_mutex);
 248         return r;
 249 }