1 /*
2  * Copyright 2015 Advanced Micro Devices, Inc.
3  *
4  * Permission is hereby granted, free of charge, to any person obtaining a
5  * copy of this software and associated documentation files (the "Software"),
6  * to deal in the Software without restriction, including without limitation
7  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8  * and/or sell copies of the Software, and to permit persons to whom the
9  * Software is furnished to do so, subject to the following conditions:
10  *
11  * The above copyright notice and this permission notice shall be included in
12  * all copies or substantial portions of the Software.
13  *
14  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
15  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
16  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
17  * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
18  * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
19  * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
20  * OTHER DEALINGS IN THE SOFTWARE.
21  *
22  */
23 
24 #include <linux/kthread.h>
25 #include <linux/wait.h>
26 #include <linux/sched.h>
27 #include <drm/drmP.h>
28 #include <drm/gpu_scheduler.h>
29 
30 static struct kmem_cache *sched_fence_slab;
31 
32 static int __init drm_sched_fence_slab_init(void)
33 {
34 	sched_fence_slab = kmem_cache_create(
35 		"drm_sched_fence", sizeof(struct drm_sched_fence), 0,
36 		SLAB_HWCACHE_ALIGN, NULL);
37 	if (!sched_fence_slab)
38 		return -ENOMEM;
39 
40 	return 0;
41 }
42 
43 static void __exit drm_sched_fence_slab_fini(void)
44 {
45 	rcu_barrier();
46 	kmem_cache_destroy(sched_fence_slab);
47 }
48 
49 void drm_sched_fence_scheduled(struct drm_sched_fence *fence)
50 {
51 	int ret = dma_fence_signal(&fence->scheduled);
52 
53 	if (!ret)
54 		DMA_FENCE_TRACE(&fence->scheduled,
55 				"signaled from irq context\n");
56 	else
57 		DMA_FENCE_TRACE(&fence->scheduled,
58 				"was already signaled\n");
59 }
60 
61 void drm_sched_fence_finished(struct drm_sched_fence *fence)
62 {
63 	int ret = dma_fence_signal(&fence->finished);
64 
65 	if (!ret)
66 		DMA_FENCE_TRACE(&fence->finished,
67 				"signaled from irq context\n");
68 	else
69 		DMA_FENCE_TRACE(&fence->finished,
70 				"was already signaled\n");
71 }
72 
73 static const char *drm_sched_fence_get_driver_name(struct dma_fence *fence)
74 {
75 	return "drm_sched";
76 }
77 
78 static const char *drm_sched_fence_get_timeline_name(struct dma_fence *f)
79 {
80 	struct drm_sched_fence *fence = to_drm_sched_fence(f);
81 	return (const char *)fence->sched->name;
82 }
83 
84 static bool drm_sched_fence_enable_signaling(struct dma_fence *f)
85 {
86 	return true;
87 }
88 
89 /**
90  * amd_sched_fence_free - free up the fence memory
91  *
92  * @rcu: RCU callback head
93  *
94  * Free up the fence memory after the RCU grace period.
95  */
96 static void drm_sched_fence_free(struct rcu_head *rcu)
97 {
98 	struct dma_fence *f = container_of(rcu, struct dma_fence, rcu);
99 	struct drm_sched_fence *fence = to_drm_sched_fence(f);
100 
101 	dma_fence_put(fence->parent);
102 	kmem_cache_free(sched_fence_slab, fence);
103 }
104 
105 /**
106  * amd_sched_fence_release_scheduled - callback that fence can be freed
107  *
108  * @fence: fence
109  *
110  * This function is called when the reference count becomes zero.
111  * It just RCU schedules freeing up the fence.
112  */
113 static void drm_sched_fence_release_scheduled(struct dma_fence *f)
114 {
115 	struct drm_sched_fence *fence = to_drm_sched_fence(f);
116 
117 	call_rcu(&fence->finished.rcu, drm_sched_fence_free);
118 }
119 
120 /**
121  * amd_sched_fence_release_finished - drop extra reference
122  *
123  * @f: fence
124  *
125  * Drop the extra reference from the scheduled fence to the base fence.
126  */
127 static void drm_sched_fence_release_finished(struct dma_fence *f)
128 {
129 	struct drm_sched_fence *fence = to_drm_sched_fence(f);
130 
131 	dma_fence_put(&fence->scheduled);
132 }
133 
134 const struct dma_fence_ops drm_sched_fence_ops_scheduled = {
135 	.get_driver_name = drm_sched_fence_get_driver_name,
136 	.get_timeline_name = drm_sched_fence_get_timeline_name,
137 	.enable_signaling = drm_sched_fence_enable_signaling,
138 	.signaled = NULL,
139 	.wait = dma_fence_default_wait,
140 	.release = drm_sched_fence_release_scheduled,
141 };
142 
143 const struct dma_fence_ops drm_sched_fence_ops_finished = {
144 	.get_driver_name = drm_sched_fence_get_driver_name,
145 	.get_timeline_name = drm_sched_fence_get_timeline_name,
146 	.enable_signaling = drm_sched_fence_enable_signaling,
147 	.signaled = NULL,
148 	.wait = dma_fence_default_wait,
149 	.release = drm_sched_fence_release_finished,
150 };
151 
152 struct drm_sched_fence *to_drm_sched_fence(struct dma_fence *f)
153 {
154 	if (f->ops == &drm_sched_fence_ops_scheduled)
155 		return container_of(f, struct drm_sched_fence, scheduled);
156 
157 	if (f->ops == &drm_sched_fence_ops_finished)
158 		return container_of(f, struct drm_sched_fence, finished);
159 
160 	return NULL;
161 }
162 EXPORT_SYMBOL(to_drm_sched_fence);
163 
164 struct drm_sched_fence *drm_sched_fence_create(struct drm_sched_entity *entity,
165 					       void *owner)
166 {
167 	struct drm_sched_fence *fence = NULL;
168 	unsigned seq;
169 
170 	fence = kmem_cache_zalloc(sched_fence_slab, GFP_KERNEL);
171 	if (fence == NULL)
172 		return NULL;
173 
174 	fence->owner = owner;
175 	fence->sched = entity->sched;
176 	spin_lock_init(&fence->lock);
177 
178 	seq = atomic_inc_return(&entity->fence_seq);
179 	dma_fence_init(&fence->scheduled, &drm_sched_fence_ops_scheduled,
180 		       &fence->lock, entity->fence_context, seq);
181 	dma_fence_init(&fence->finished, &drm_sched_fence_ops_finished,
182 		       &fence->lock, entity->fence_context + 1, seq);
183 
184 	return fence;
185 }
186 
187 module_init(drm_sched_fence_slab_init);
188 module_exit(drm_sched_fence_slab_fini);
189 
190 MODULE_DESCRIPTION("DRM GPU scheduler");
191 MODULE_LICENSE("GPL and additional rights");
192