drivers/gpu/drm/radeon/radeon_sync.c

   1 /*
   2  * Copyright 2014 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 <christian.koenig@amd.com>
  29  */
  30
  31 #include <drm/drmP.h>
  32 #include "radeon.h"
  33 #include "radeon_trace.h"
  34
  35 /**
  36  * radeon_sync_create - zero init sync object
  37  *
  38  * @sync: sync object to initialize
  39  *
  40  * Just clear the sync object for now.
  41  */
  42 void radeon_sync_create(struct radeon_sync *sync)
  43 {
  44         unsigned i;
  45
  46         for (i = 0; i < RADEON_NUM_SYNCS; ++i)
  47                 sync->semaphores[i] = NULL;
  48
  49         for (i = 0; i < RADEON_NUM_RINGS; ++i)
  50                 sync->sync_to[i] = NULL;
  51
  52         sync->last_vm_update = NULL;
  53 }
  54
  55 /**
  56  * radeon_sync_fence - use the semaphore to sync to a fence
  57  *
  58  * @sync: sync object to add fence to
  59  * @fence: fence to sync to
  60  *
  61  * Sync to the fence using the semaphore objects
  62  */
  63 void radeon_sync_fence(struct radeon_sync *sync,
  64                        struct radeon_fence *fence)
  65 {
  66         struct radeon_fence *other;
  67
  68         if (!fence)
  69                 return;
  70
  71         other = sync->sync_to[fence->ring];
  72         sync->sync_to[fence->ring] = radeon_fence_later(fence, other);
  73
  74         if (fence->is_vm_update) {
  75                 other = sync->last_vm_update;
  76                 sync->last_vm_update = radeon_fence_later(fence, other);
  77         }
  78 }
  79
  80 /**
  81  * radeon_sync_resv - use the semaphores to sync to a reservation object
  82  *
  83  * @sync: sync object to add fences from reservation object to
  84  * @resv: reservation object with embedded fence
  85  * @shared: true if we should only sync to the exclusive fence
  86  *
  87  * Sync to the fence using the semaphore objects
  88  */
  89 int radeon_sync_resv(struct radeon_device *rdev,
  90                      struct radeon_sync *sync,
  91                      struct reservation_object *resv,
  92                      bool shared)
  93 {
  94         struct reservation_object_list *flist;
  95         struct dma_fence *f;
  96         struct radeon_fence *fence;
  97         unsigned i;
  98         int r = 0;
  99
 100         /* always sync to the exclusive fence */
 101         f = reservation_object_get_excl(resv);
 102         fence = f ? to_radeon_fence(f) : NULL;
 103         if (fence && fence->rdev == rdev)
 104                 radeon_sync_fence(sync, fence);
 105         else if (f)
 106                 r = dma_fence_wait(f, true);
 107
 108         flist = reservation_object_get_list(resv);
 109         if (shared || !flist || r)
 110                 return r;
 111
 112         for (i = 0; i < flist->shared_count; ++i) {
 113                 f = rcu_dereference_protected(flist->shared[i],
 114                                               reservation_object_held(resv));
 115                 fence = to_radeon_fence(f);
 116                 if (fence && fence->rdev == rdev)
 117                         radeon_sync_fence(sync, fence);
 118                 else
 119                         r = dma_fence_wait(f, true);
 120
 121                 if (r)
 122                         break;
 123         }
 124         return r;
 125 }
 126
 127 /**
 128  * radeon_sync_rings - sync ring to all registered fences
 129  *
 130  * @rdev: radeon_device pointer
 131  * @sync: sync object to use
 132  * @ring: ring that needs sync
 133  *
 134  * Ensure that all registered fences are signaled before letting
 135  * the ring continue. The caller must hold the ring lock.
 136  */
 137 int radeon_sync_rings(struct radeon_device *rdev,
 138                       struct radeon_sync *sync,
 139                       int ring)
 140 {
 141         unsigned count = 0;
 142         int i, r;
 143
 144         for (i = 0; i < RADEON_NUM_RINGS; ++i) {
 145                 struct radeon_fence *fence = sync->sync_to[i];
 146                 struct radeon_semaphore *semaphore;
 147
 148                 /* check if we really need to sync */
 149                 if (!radeon_fence_need_sync(fence, ring))
 150                         continue;
 151
 152                 /* prevent GPU deadlocks */
 153                 if (!rdev->ring[i].ready) {
 154                         dev_err(rdev->dev, "Syncing to a disabled ring!");
 155                         return -EINVAL;
 156                 }
 157
 158                 if (count >= RADEON_NUM_SYNCS) {
 159                         /* not enough room, wait manually */
 160                         r = radeon_fence_wait(fence, false);
 161                         if (r)
 162                                 return r;
 163                         continue;
 164                 }
 165                 r = radeon_semaphore_create(rdev, &semaphore);
 166                 if (r)
 167                         return r;
 168
 169                 sync->semaphores[count++] = semaphore;
 170
 171                 /* allocate enough space for sync command */
 172                 r = radeon_ring_alloc(rdev, &rdev->ring[i], 16);
 173                 if (r)
 174                         return r;
 175
 176                 /* emit the signal semaphore */
 177                 if (!radeon_semaphore_emit_signal(rdev, i, semaphore)) {
 178                         /* signaling wasn't successful wait manually */
 179                         radeon_ring_undo(&rdev->ring[i]);
 180                         r = radeon_fence_wait(fence, false);
 181                         if (r)
 182                                 return r;
 183                         continue;
 184                 }
 185
 186                 /* we assume caller has already allocated space on waiters ring */
 187                 if (!radeon_semaphore_emit_wait(rdev, ring, semaphore)) {
 188                         /* waiting wasn't successful wait manually */
 189                         radeon_ring_undo(&rdev->ring[i]);
 190                         r = radeon_fence_wait(fence, false);
 191                         if (r)
 192                                 return r;
 193                         continue;
 194                 }
 195
 196                 radeon_ring_commit(rdev, &rdev->ring[i], false);
 197                 radeon_fence_note_sync(fence, ring);
 198         }
 199
 200         return 0;
 201 }
 202
 203 /**
 204  * radeon_sync_free - free the sync object
 205  *
 206  * @rdev: radeon_device pointer
 207  * @sync: sync object to use
 208  * @fence: fence to use for the free
 209  *
 210  * Free the sync object by freeing all semaphores in it.
 211  */
 212 void radeon_sync_free(struct radeon_device *rdev,
 213                       struct radeon_sync *sync,
 214                       struct radeon_fence *fence)
 215 {
 216         unsigned i;
 217
 218         for (i = 0; i < RADEON_NUM_SYNCS; ++i)
 219                 radeon_semaphore_free(rdev, &sync->semaphores[i], fence);
 220 }