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