drivers/gpu/drm/amd/amdgpu/amdgpu_bo_list.c

   1 /*
   2  * Copyright 2015 Advanced Micro Devices, Inc.
   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
   7  * "Software"), to deal in the Software without restriction, including
   8  * without limitation the rights to use, copy, modify, merge, publish,
   9  * distribute, sub license, and/or sell copies of the Software, and to
  10  * permit persons to whom the Software is furnished to do so, subject to
  11  * the following conditions:
  12  *
  13  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  14  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  15  * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
  16  * THE COPYRIGHT HOLDERS, AUTHORS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM,
  17  * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
  18  * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
  19  * USE OR OTHER DEALINGS IN THE SOFTWARE.
  20  *
  21  * The above copyright notice and this permission notice (including the
  22  * next paragraph) shall be included in all copies or substantial portions
  23  * of the Software.
  24  *
  25  */
  26 /*
  27  * Authors:
  28  *    Christian König <deathsimple@vodafone.de>
  29  */
  30
  31 #include <drm/drmP.h>
  32 #include "amdgpu.h"
  33 #include "amdgpu_trace.h"
  34
  35 #define AMDGPU_BO_LIST_MAX_PRIORITY     32u
  36 #define AMDGPU_BO_LIST_NUM_BUCKETS      (AMDGPU_BO_LIST_MAX_PRIORITY + 1)
  37
  38 static int amdgpu_bo_list_create(struct amdgpu_fpriv *fpriv,
  39                                  struct amdgpu_bo_list **result,
  40                                  int *id)
  41 {
  42         int r;
  43
  44         *result = kzalloc(sizeof(struct amdgpu_bo_list), GFP_KERNEL);
  45         if (!*result)
  46                 return -ENOMEM;
  47
  48         mutex_lock(&fpriv->bo_list_lock);
  49         r = idr_alloc(&fpriv->bo_list_handles, *result,
  50                       1, 0, GFP_KERNEL);
  51         if (r < 0) {
  52                 mutex_unlock(&fpriv->bo_list_lock);
  53                 kfree(*result);
  54                 return r;
  55         }
  56         *id = r;
  57
  58         mutex_init(&(*result)->lock);
  59         (*result)->num_entries = 0;
  60         (*result)->array = NULL;
  61
  62         mutex_lock(&(*result)->lock);
  63         mutex_unlock(&fpriv->bo_list_lock);
  64
  65         return 0;
  66 }
  67
  68 static void amdgpu_bo_list_destroy(struct amdgpu_fpriv *fpriv, int id)
  69 {
  70         struct amdgpu_bo_list *list;
  71
  72         mutex_lock(&fpriv->bo_list_lock);
  73         list = idr_find(&fpriv->bo_list_handles, id);
  74         if (list) {
  75                 mutex_lock(&list->lock);
  76                 idr_remove(&fpriv->bo_list_handles, id);
  77                 mutex_unlock(&list->lock);
  78                 amdgpu_bo_list_free(list);
  79         }
  80         mutex_unlock(&fpriv->bo_list_lock);
  81 }
  82
  83 static int amdgpu_bo_list_set(struct amdgpu_device *adev,
  84                                      struct drm_file *filp,
  85                                      struct amdgpu_bo_list *list,
  86                                      struct drm_amdgpu_bo_list_entry *info,
  87                                      unsigned num_entries)
  88 {
  89         struct amdgpu_bo_list_entry *array;
  90         struct amdgpu_bo *gds_obj = adev->gds.gds_gfx_bo;
  91         struct amdgpu_bo *gws_obj = adev->gds.gws_gfx_bo;
  92         struct amdgpu_bo *oa_obj = adev->gds.oa_gfx_bo;
  93
  94         unsigned last_entry = 0, first_userptr = num_entries;
  95         unsigned i;
  96         int r;
  97         unsigned long total_size = 0;
  98
  99         array = drm_malloc_ab(num_entries, sizeof(struct amdgpu_bo_list_entry));
 100         if (!array)
 101                 return -ENOMEM;
 102         memset(array, 0, num_entries * sizeof(struct amdgpu_bo_list_entry));
 103
 104         for (i = 0; i < num_entries; ++i) {
 105                 struct amdgpu_bo_list_entry *entry;
 106                 struct drm_gem_object *gobj;
 107                 struct amdgpu_bo *bo;
 108                 struct mm_struct *usermm;
 109
 110                 gobj = drm_gem_object_lookup(filp, info[i].bo_handle);
 111                 if (!gobj) {
 112                         r = -ENOENT;
 113                         goto error_free;
 114                 }
 115
 116                 bo = amdgpu_bo_ref(gem_to_amdgpu_bo(gobj));
 117                 drm_gem_object_unreference_unlocked(gobj);
 118
 119                 usermm = amdgpu_ttm_tt_get_usermm(bo->tbo.ttm);
 120                 if (usermm) {
 121                         if (usermm != current->mm) {
 122                                 amdgpu_bo_unref(&bo);
 123                                 r = -EPERM;
 124                                 goto error_free;
 125                         }
 126                         entry = &array[--first_userptr];
 127                 } else {
 128                         entry = &array[last_entry++];
 129                 }
 130
 131                 entry->robj = bo;
 132                 entry->priority = min(info[i].bo_priority,
 133                                       AMDGPU_BO_LIST_MAX_PRIORITY);
 134                 entry->tv.bo = &entry->robj->tbo;
 135                 entry->tv.shared = !entry->robj->prime_shared_count;
 136
 137                 if (entry->robj->prefered_domains == AMDGPU_GEM_DOMAIN_GDS)
 138                         gds_obj = entry->robj;
 139                 if (entry->robj->prefered_domains == AMDGPU_GEM_DOMAIN_GWS)
 140                         gws_obj = entry->robj;
 141                 if (entry->robj->prefered_domains == AMDGPU_GEM_DOMAIN_OA)
 142                         oa_obj = entry->robj;
 143
 144                 total_size += amdgpu_bo_size(entry->robj);
 145                 trace_amdgpu_bo_list_set(list, entry->robj);
 146         }
 147
 148         for (i = 0; i < list->num_entries; ++i)
 149                 amdgpu_bo_unref(&list->array[i].robj);
 150
 151         drm_free_large(list->array);
 152
 153         list->gds_obj = gds_obj;
 154         list->gws_obj = gws_obj;
 155         list->oa_obj = oa_obj;
 156         list->first_userptr = first_userptr;
 157         list->array = array;
 158         list->num_entries = num_entries;
 159
 160         trace_amdgpu_cs_bo_status(list->num_entries, total_size);
 161         return 0;
 162
 163 error_free:
 164         while (i--)
 165                 amdgpu_bo_unref(&array[i].robj);
 166         drm_free_large(array);
 167         return r;
 168 }
 169
 170 struct amdgpu_bo_list *
 171 amdgpu_bo_list_get(struct amdgpu_fpriv *fpriv, int id)
 172 {
 173         struct amdgpu_bo_list *result;
 174
 175         mutex_lock(&fpriv->bo_list_lock);
 176         result = idr_find(&fpriv->bo_list_handles, id);
 177         if (result)
 178                 mutex_lock(&result->lock);
 179         mutex_unlock(&fpriv->bo_list_lock);
 180         return result;
 181 }
 182
 183 void amdgpu_bo_list_get_list(struct amdgpu_bo_list *list,
 184                              struct list_head *validated)
 185 {
 186         /* This is based on the bucket sort with O(n) time complexity.
 187          * An item with priority "i" is added to bucket[i]. The lists are then
 188          * concatenated in descending order.
 189          */
 190         struct list_head bucket[AMDGPU_BO_LIST_NUM_BUCKETS];
 191         unsigned i;
 192
 193         for (i = 0; i < AMDGPU_BO_LIST_NUM_BUCKETS; i++)
 194                 INIT_LIST_HEAD(&bucket[i]);
 195
 196         /* Since buffers which appear sooner in the relocation list are
 197          * likely to be used more often than buffers which appear later
 198          * in the list, the sort mustn't change the ordering of buffers
 199          * with the same priority, i.e. it must be stable.
 200          */
 201         for (i = 0; i < list->num_entries; i++) {
 202                 unsigned priority = list->array[i].priority;
 203
 204                 list_add_tail(&list->array[i].tv.head,
 205                               &bucket[priority]);
 206                 list->array[i].user_pages = NULL;
 207         }
 208
 209         /* Connect the sorted buckets in the output list. */
 210         for (i = 0; i < AMDGPU_BO_LIST_NUM_BUCKETS; i++)
 211                 list_splice(&bucket[i], validated);
 212 }
 213
 214 void amdgpu_bo_list_put(struct amdgpu_bo_list *list)
 215 {
 216         mutex_unlock(&list->lock);
 217 }
 218
 219 void amdgpu_bo_list_free(struct amdgpu_bo_list *list)
 220 {
 221         unsigned i;
 222
 223         for (i = 0; i < list->num_entries; ++i)
 224                 amdgpu_bo_unref(&list->array[i].robj);
 225
 226         mutex_destroy(&list->lock);
 227         drm_free_large(list->array);
 228         kfree(list);
 229 }
 230
 231 int amdgpu_bo_list_ioctl(struct drm_device *dev, void *data,
 232                                 struct drm_file *filp)
 233 {
 234         const uint32_t info_size = sizeof(struct drm_amdgpu_bo_list_entry);
 235
 236         struct amdgpu_device *adev = dev->dev_private;
 237         struct amdgpu_fpriv *fpriv = filp->driver_priv;
 238         union drm_amdgpu_bo_list *args = data;
 239         uint32_t handle = args->in.list_handle;
 240         const void __user *uptr = (const void*)(long)args->in.bo_info_ptr;
 241
 242         struct drm_amdgpu_bo_list_entry *info;
 243         struct amdgpu_bo_list *list;
 244
 245         int r;
 246
 247         info = drm_malloc_ab(args->in.bo_number,
 248                              sizeof(struct drm_amdgpu_bo_list_entry));
 249         if (!info)
 250                 return -ENOMEM;
 251
 252         /* copy the handle array from userspace to a kernel buffer */
 253         r = -EFAULT;
 254         if (likely(info_size == args->in.bo_info_size)) {
 255                 unsigned long bytes = args->in.bo_number *
 256                         args->in.bo_info_size;
 257
 258                 if (copy_from_user(info, uptr, bytes))
 259                         goto error_free;
 260
 261         } else {
 262                 unsigned long bytes = min(args->in.bo_info_size, info_size);
 263                 unsigned i;
 264
 265                 memset(info, 0, args->in.bo_number * info_size);
 266                 for (i = 0; i < args->in.bo_number; ++i) {
 267                         if (copy_from_user(&info[i], uptr, bytes))
 268                                 goto error_free;
 269
 270                         uptr += args->in.bo_info_size;
 271                 }
 272         }
 273
 274         switch (args->in.operation) {
 275         case AMDGPU_BO_LIST_OP_CREATE:
 276                 r = amdgpu_bo_list_create(fpriv, &list, &handle);
 277                 if (r)
 278                         goto error_free;
 279
 280                 r = amdgpu_bo_list_set(adev, filp, list, info,
 281                                               args->in.bo_number);
 282                 amdgpu_bo_list_put(list);
 283                 if (r)
 284                         goto error_free;
 285
 286                 break;
 287
 288         case AMDGPU_BO_LIST_OP_DESTROY:
 289                 amdgpu_bo_list_destroy(fpriv, handle);
 290                 handle = 0;
 291                 break;
 292
 293         case AMDGPU_BO_LIST_OP_UPDATE:
 294                 r = -ENOENT;
 295                 list = amdgpu_bo_list_get(fpriv, handle);
 296                 if (!list)
 297                         goto error_free;
 298
 299                 r = amdgpu_bo_list_set(adev, filp, list, info,
 300                                               args->in.bo_number);
 301                 amdgpu_bo_list_put(list);
 302                 if (r)
 303                         goto error_free;
 304
 305                 break;
 306
 307         default:
 308                 r = -EINVAL;
 309                 goto error_free;
 310         }
 311
 312         memset(args, 0, sizeof(*args));
 313         args->out.list_handle = handle;
 314         drm_free_large(info);
 315
 316         return 0;
 317
 318 error_free:
 319         drm_free_large(info);
 320         return r;
 321 }