xref: /linux/drivers/gpu/drm/i915/i915_active.h (revision 17cfcb68af3bc7d5e8ae08779b1853310a2949f3)
1 /*
2  * SPDX-License-Identifier: MIT
3  *
4  * Copyright © 2019 Intel Corporation
5  */
6 
7 #ifndef _I915_ACTIVE_H_
8 #define _I915_ACTIVE_H_
9 
10 #include <linux/lockdep.h>
11 
12 #include "i915_active_types.h"
13 #include "i915_request.h"
14 
15 struct i915_request;
16 struct intel_engine_cs;
17 struct intel_timeline;
18 
19 /*
20  * We treat requests as fences. This is not be to confused with our
21  * "fence registers" but pipeline synchronisation objects ala GL_ARB_sync.
22  * We use the fences to synchronize access from the CPU with activity on the
23  * GPU, for example, we should not rewrite an object's PTE whilst the GPU
24  * is reading them. We also track fences at a higher level to provide
25  * implicit synchronisation around GEM objects, e.g. set-domain will wait
26  * for outstanding GPU rendering before marking the object ready for CPU
27  * access, or a pageflip will wait until the GPU is complete before showing
28  * the frame on the scanout.
29  *
30  * In order to use a fence, the object must track the fence it needs to
31  * serialise with. For example, GEM objects want to track both read and
32  * write access so that we can perform concurrent read operations between
33  * the CPU and GPU engines, as well as waiting for all rendering to
34  * complete, or waiting for the last GPU user of a "fence register". The
35  * object then embeds a #i915_active_fence to track the most recent (in
36  * retirement order) request relevant for the desired mode of access.
37  * The #i915_active_fence is updated with i915_active_fence_set() to
38  * track the most recent fence request, typically this is done as part of
39  * i915_vma_move_to_active().
40  *
41  * When the #i915_active_fence completes (is retired), it will
42  * signal its completion to the owner through a callback as well as mark
43  * itself as idle (i915_active_fence.request == NULL). The owner
44  * can then perform any action, such as delayed freeing of an active
45  * resource including itself.
46  */
47 
48 void i915_active_noop(struct dma_fence *fence, struct dma_fence_cb *cb);
49 
50 /**
51  * __i915_active_fence_init - prepares the activity tracker for use
52  * @active - the active tracker
53  * @fence - initial fence to track, can be NULL
54  * @func - a callback when then the tracker is retired (becomes idle),
55  *         can be NULL
56  *
57  * i915_active_fence_init() prepares the embedded @active struct for use as
58  * an activity tracker, that is for tracking the last known active fence
59  * associated with it. When the last fence becomes idle, when it is retired
60  * after completion, the optional callback @func is invoked.
61  */
62 static inline void
63 __i915_active_fence_init(struct i915_active_fence *active,
64 			 struct mutex *lock,
65 			 void *fence,
66 			 dma_fence_func_t fn)
67 {
68 	RCU_INIT_POINTER(active->fence, fence);
69 	active->cb.func = fn ?: i915_active_noop;
70 #if IS_ENABLED(CONFIG_DRM_I915_DEBUG_GEM)
71 	active->lock = lock;
72 #endif
73 }
74 
75 #define INIT_ACTIVE_FENCE(A, LOCK) \
76 	__i915_active_fence_init((A), (LOCK), NULL, NULL)
77 
78 struct dma_fence *
79 __i915_active_fence_set(struct i915_active_fence *active,
80 			struct dma_fence *fence);
81 
82 /**
83  * i915_active_fence_set - updates the tracker to watch the current fence
84  * @active - the active tracker
85  * @rq - the request to watch
86  *
87  * i915_active_fence_set() watches the given @rq for completion. While
88  * that @rq is busy, the @active reports busy. When that @rq is signaled
89  * (or else retired) the @active tracker is updated to report idle.
90  */
91 int __must_check
92 i915_active_fence_set(struct i915_active_fence *active,
93 		      struct i915_request *rq);
94 /**
95  * i915_active_fence_get - return a reference to the active fence
96  * @active - the active tracker
97  *
98  * i915_active_fence_get() returns a reference to the active fence,
99  * or NULL if the active tracker is idle. The reference is obtained under RCU,
100  * so no locking is required by the caller.
101  *
102  * The reference should be freed with dma_fence_put().
103  */
104 static inline struct dma_fence *
105 i915_active_fence_get(struct i915_active_fence *active)
106 {
107 	struct dma_fence *fence;
108 
109 	rcu_read_lock();
110 	fence = dma_fence_get_rcu_safe(&active->fence);
111 	rcu_read_unlock();
112 
113 	return fence;
114 }
115 
116 /**
117  * i915_active_fence_isset - report whether the active tracker is assigned
118  * @active - the active tracker
119  *
120  * i915_active_fence_isset() returns true if the active tracker is currently
121  * assigned to a fence. Due to the lazy retiring, that fence may be idle
122  * and this may report stale information.
123  */
124 static inline bool
125 i915_active_fence_isset(const struct i915_active_fence *active)
126 {
127 	return rcu_access_pointer(active->fence);
128 }
129 
130 static inline void
131 i915_active_fence_cb(struct dma_fence *fence, struct dma_fence_cb *cb)
132 {
133 	struct i915_active_fence *active =
134 		container_of(cb, typeof(*active), cb);
135 
136 	RCU_INIT_POINTER(active->fence, NULL);
137 }
138 
139 /*
140  * GPU activity tracking
141  *
142  * Each set of commands submitted to the GPU compromises a single request that
143  * signals a fence upon completion. struct i915_request combines the
144  * command submission, scheduling and fence signaling roles. If we want to see
145  * if a particular task is complete, we need to grab the fence (struct
146  * i915_request) for that task and check or wait for it to be signaled. More
147  * often though we want to track the status of a bunch of tasks, for example
148  * to wait for the GPU to finish accessing some memory across a variety of
149  * different command pipelines from different clients. We could choose to
150  * track every single request associated with the task, but knowing that
151  * each request belongs to an ordered timeline (later requests within a
152  * timeline must wait for earlier requests), we need only track the
153  * latest request in each timeline to determine the overall status of the
154  * task.
155  *
156  * struct i915_active provides this tracking across timelines. It builds a
157  * composite shared-fence, and is updated as new work is submitted to the task,
158  * forming a snapshot of the current status. It should be embedded into the
159  * different resources that need to track their associated GPU activity to
160  * provide a callback when that GPU activity has ceased, or otherwise to
161  * provide a serialisation point either for request submission or for CPU
162  * synchronisation.
163  */
164 
165 void __i915_active_init(struct i915_active *ref,
166 			int (*active)(struct i915_active *ref),
167 			void (*retire)(struct i915_active *ref),
168 			struct lock_class_key *key);
169 #define i915_active_init(ref, active, retire) do {		\
170 	static struct lock_class_key __key;				\
171 									\
172 	__i915_active_init(ref, active, retire, &__key);		\
173 } while (0)
174 
175 int i915_active_ref(struct i915_active *ref,
176 		    struct intel_timeline *tl,
177 		    struct dma_fence *fence);
178 
179 static inline int
180 i915_active_add_request(struct i915_active *ref, struct i915_request *rq)
181 {
182 	return i915_active_ref(ref, i915_request_timeline(rq), &rq->fence);
183 }
184 
185 void i915_active_set_exclusive(struct i915_active *ref, struct dma_fence *f);
186 
187 static inline bool i915_active_has_exclusive(struct i915_active *ref)
188 {
189 	return rcu_access_pointer(ref->excl.fence);
190 }
191 
192 int i915_active_wait(struct i915_active *ref);
193 
194 int i915_request_await_active(struct i915_request *rq, struct i915_active *ref);
195 
196 int i915_active_acquire(struct i915_active *ref);
197 bool i915_active_acquire_if_busy(struct i915_active *ref);
198 void i915_active_release(struct i915_active *ref);
199 
200 static inline bool
201 i915_active_is_idle(const struct i915_active *ref)
202 {
203 	return !atomic_read(&ref->count);
204 }
205 
206 #if IS_ENABLED(CONFIG_DRM_I915_DEBUG_GEM)
207 void i915_active_fini(struct i915_active *ref);
208 #else
209 static inline void i915_active_fini(struct i915_active *ref) { }
210 #endif
211 
212 int i915_active_acquire_preallocate_barrier(struct i915_active *ref,
213 					    struct intel_engine_cs *engine);
214 void i915_active_acquire_barrier(struct i915_active *ref);
215 void i915_request_add_active_barriers(struct i915_request *rq);
216 
217 #endif /* _I915_ACTIVE_H_ */
218