1 /*
2  * Copyright © 2014 Intel Corporation
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 (including the next
12  * paragraph) shall be included in all copies or substantial portions of the
13  * Software.
14  *
15  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
18  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
20  * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
21  * DEALINGS IN THE SOFTWARE.
22  *
23  * Authors:
24  *	Daniel Vetter <daniel.vetter@ffwll.ch>
25  */
26 
27 /**
28  * DOC: frontbuffer tracking
29  *
30  * Many features require us to track changes to the currently active
31  * frontbuffer, especially rendering targeted at the frontbuffer.
32  *
33  * To be able to do so we track frontbuffers using a bitmask for all possible
34  * frontbuffer slots through intel_frontbuffer_track(). The functions in this
35  * file are then called when the contents of the frontbuffer are invalidated,
36  * when frontbuffer rendering has stopped again to flush out all the changes
37  * and when the frontbuffer is exchanged with a flip. Subsystems interested in
38  * frontbuffer changes (e.g. PSR, FBC, DRRS) should directly put their callbacks
39  * into the relevant places and filter for the frontbuffer slots that they are
40  * interested int.
41  *
42  * On a high level there are two types of powersaving features. The first one
43  * work like a special cache (FBC and PSR) and are interested when they should
44  * stop caching and when to restart caching. This is done by placing callbacks
45  * into the invalidate and the flush functions: At invalidate the caching must
46  * be stopped and at flush time it can be restarted. And maybe they need to know
47  * when the frontbuffer changes (e.g. when the hw doesn't initiate an invalidate
48  * and flush on its own) which can be achieved with placing callbacks into the
49  * flip functions.
50  *
51  * The other type of display power saving feature only cares about busyness
52  * (e.g. DRRS). In that case all three (invalidate, flush and flip) indicate
53  * busyness. There is no direct way to detect idleness. Instead an idle timer
54  * work delayed work should be started from the flush and flip functions and
55  * cancelled as soon as busyness is detected.
56  */
57 
58 #include "i915_drv.h"
59 #include "intel_display_trace.h"
60 #include "intel_display_types.h"
61 #include "intel_dp.h"
62 #include "intel_drrs.h"
63 #include "intel_fbc.h"
64 #include "intel_frontbuffer.h"
65 #include "intel_psr.h"
66 
67 /**
68  * frontbuffer_flush - flush frontbuffer
69  * @i915: i915 device
70  * @frontbuffer_bits: frontbuffer plane tracking bits
71  * @origin: which operation caused the flush
72  *
73  * This function gets called every time rendering on the given planes has
74  * completed and frontbuffer caching can be started again. Flushes will get
75  * delayed if they're blocked by some outstanding asynchronous rendering.
76  *
77  * Can be called without any locks held.
78  */
79 static void frontbuffer_flush(struct drm_i915_private *i915,
80 			      unsigned int frontbuffer_bits,
81 			      enum fb_op_origin origin)
82 {
83 	/* Delay flushing when rings are still busy.*/
84 	spin_lock(&i915->fb_tracking.lock);
85 	frontbuffer_bits &= ~i915->fb_tracking.busy_bits;
86 	spin_unlock(&i915->fb_tracking.lock);
87 
88 	if (!frontbuffer_bits)
89 		return;
90 
91 	trace_intel_frontbuffer_flush(frontbuffer_bits, origin);
92 
93 	might_sleep();
94 	intel_drrs_flush(i915, frontbuffer_bits);
95 	intel_psr_flush(i915, frontbuffer_bits, origin);
96 	intel_fbc_flush(i915, frontbuffer_bits, origin);
97 }
98 
99 /**
100  * intel_frontbuffer_flip_prepare - prepare asynchronous frontbuffer flip
101  * @i915: i915 device
102  * @frontbuffer_bits: frontbuffer plane tracking bits
103  *
104  * This function gets called after scheduling a flip on @obj. The actual
105  * frontbuffer flushing will be delayed until completion is signalled with
106  * intel_frontbuffer_flip_complete. If an invalidate happens in between this
107  * flush will be cancelled.
108  *
109  * Can be called without any locks held.
110  */
111 void intel_frontbuffer_flip_prepare(struct drm_i915_private *i915,
112 				    unsigned frontbuffer_bits)
113 {
114 	spin_lock(&i915->fb_tracking.lock);
115 	i915->fb_tracking.flip_bits |= frontbuffer_bits;
116 	/* Remove stale busy bits due to the old buffer. */
117 	i915->fb_tracking.busy_bits &= ~frontbuffer_bits;
118 	spin_unlock(&i915->fb_tracking.lock);
119 }
120 
121 /**
122  * intel_frontbuffer_flip_complete - complete asynchronous frontbuffer flip
123  * @i915: i915 device
124  * @frontbuffer_bits: frontbuffer plane tracking bits
125  *
126  * This function gets called after the flip has been latched and will complete
127  * on the next vblank. It will execute the flush if it hasn't been cancelled yet.
128  *
129  * Can be called without any locks held.
130  */
131 void intel_frontbuffer_flip_complete(struct drm_i915_private *i915,
132 				     unsigned frontbuffer_bits)
133 {
134 	spin_lock(&i915->fb_tracking.lock);
135 	/* Mask any cancelled flips. */
136 	frontbuffer_bits &= i915->fb_tracking.flip_bits;
137 	i915->fb_tracking.flip_bits &= ~frontbuffer_bits;
138 	spin_unlock(&i915->fb_tracking.lock);
139 
140 	if (frontbuffer_bits)
141 		frontbuffer_flush(i915, frontbuffer_bits, ORIGIN_FLIP);
142 }
143 
144 /**
145  * intel_frontbuffer_flip - synchronous frontbuffer flip
146  * @i915: i915 device
147  * @frontbuffer_bits: frontbuffer plane tracking bits
148  *
149  * This function gets called after scheduling a flip on @obj. This is for
150  * synchronous plane updates which will happen on the next vblank and which will
151  * not get delayed by pending gpu rendering.
152  *
153  * Can be called without any locks held.
154  */
155 void intel_frontbuffer_flip(struct drm_i915_private *i915,
156 			    unsigned frontbuffer_bits)
157 {
158 	spin_lock(&i915->fb_tracking.lock);
159 	/* Remove stale busy bits due to the old buffer. */
160 	i915->fb_tracking.busy_bits &= ~frontbuffer_bits;
161 	spin_unlock(&i915->fb_tracking.lock);
162 
163 	frontbuffer_flush(i915, frontbuffer_bits, ORIGIN_FLIP);
164 }
165 
166 void __intel_fb_invalidate(struct intel_frontbuffer *front,
167 			   enum fb_op_origin origin,
168 			   unsigned int frontbuffer_bits)
169 {
170 	struct drm_i915_private *i915 = to_i915(front->obj->base.dev);
171 
172 	if (origin == ORIGIN_CS) {
173 		spin_lock(&i915->fb_tracking.lock);
174 		i915->fb_tracking.busy_bits |= frontbuffer_bits;
175 		i915->fb_tracking.flip_bits &= ~frontbuffer_bits;
176 		spin_unlock(&i915->fb_tracking.lock);
177 	}
178 
179 	trace_intel_frontbuffer_invalidate(frontbuffer_bits, origin);
180 
181 	might_sleep();
182 	intel_psr_invalidate(i915, frontbuffer_bits, origin);
183 	intel_drrs_invalidate(i915, frontbuffer_bits);
184 	intel_fbc_invalidate(i915, frontbuffer_bits, origin);
185 }
186 
187 void __intel_fb_flush(struct intel_frontbuffer *front,
188 		      enum fb_op_origin origin,
189 		      unsigned int frontbuffer_bits)
190 {
191 	struct drm_i915_private *i915 = to_i915(front->obj->base.dev);
192 
193 	if (origin == ORIGIN_CS) {
194 		spin_lock(&i915->fb_tracking.lock);
195 		/* Filter out new bits since rendering started. */
196 		frontbuffer_bits &= i915->fb_tracking.busy_bits;
197 		i915->fb_tracking.busy_bits &= ~frontbuffer_bits;
198 		spin_unlock(&i915->fb_tracking.lock);
199 	}
200 
201 	if (frontbuffer_bits)
202 		frontbuffer_flush(i915, frontbuffer_bits, origin);
203 }
204 
205 static int frontbuffer_active(struct i915_active *ref)
206 {
207 	struct intel_frontbuffer *front =
208 		container_of(ref, typeof(*front), write);
209 
210 	kref_get(&front->ref);
211 	return 0;
212 }
213 
214 static void frontbuffer_retire(struct i915_active *ref)
215 {
216 	struct intel_frontbuffer *front =
217 		container_of(ref, typeof(*front), write);
218 
219 	intel_frontbuffer_flush(front, ORIGIN_CS);
220 	intel_frontbuffer_put(front);
221 }
222 
223 static void frontbuffer_release(struct kref *ref)
224 	__releases(&to_i915(front->obj->base.dev)->fb_tracking.lock)
225 {
226 	struct intel_frontbuffer *front =
227 		container_of(ref, typeof(*front), ref);
228 	struct drm_i915_gem_object *obj = front->obj;
229 	struct i915_vma *vma;
230 
231 	drm_WARN_ON(obj->base.dev, atomic_read(&front->bits));
232 
233 	spin_lock(&obj->vma.lock);
234 	for_each_ggtt_vma(vma, obj) {
235 		i915_vma_clear_scanout(vma);
236 		vma->display_alignment = I915_GTT_MIN_ALIGNMENT;
237 	}
238 	spin_unlock(&obj->vma.lock);
239 
240 	RCU_INIT_POINTER(obj->frontbuffer, NULL);
241 	spin_unlock(&to_i915(obj->base.dev)->fb_tracking.lock);
242 
243 	i915_active_fini(&front->write);
244 
245 	i915_gem_object_put(obj);
246 	kfree_rcu(front, rcu);
247 }
248 
249 struct intel_frontbuffer *
250 intel_frontbuffer_get(struct drm_i915_gem_object *obj)
251 {
252 	struct drm_i915_private *i915 = to_i915(obj->base.dev);
253 	struct intel_frontbuffer *front;
254 
255 	front = __intel_frontbuffer_get(obj);
256 	if (front)
257 		return front;
258 
259 	front = kmalloc(sizeof(*front), GFP_KERNEL);
260 	if (!front)
261 		return NULL;
262 
263 	front->obj = obj;
264 	kref_init(&front->ref);
265 	atomic_set(&front->bits, 0);
266 	i915_active_init(&front->write,
267 			 frontbuffer_active,
268 			 frontbuffer_retire,
269 			 I915_ACTIVE_RETIRE_SLEEPS);
270 
271 	spin_lock(&i915->fb_tracking.lock);
272 	if (rcu_access_pointer(obj->frontbuffer)) {
273 		kfree(front);
274 		front = rcu_dereference_protected(obj->frontbuffer, true);
275 		kref_get(&front->ref);
276 	} else {
277 		i915_gem_object_get(obj);
278 		rcu_assign_pointer(obj->frontbuffer, front);
279 	}
280 	spin_unlock(&i915->fb_tracking.lock);
281 
282 	return front;
283 }
284 
285 void intel_frontbuffer_put(struct intel_frontbuffer *front)
286 {
287 	kref_put_lock(&front->ref,
288 		      frontbuffer_release,
289 		      &to_i915(front->obj->base.dev)->fb_tracking.lock);
290 }
291 
292 /**
293  * intel_frontbuffer_track - update frontbuffer tracking
294  * @old: current buffer for the frontbuffer slots
295  * @new: new buffer for the frontbuffer slots
296  * @frontbuffer_bits: bitmask of frontbuffer slots
297  *
298  * This updates the frontbuffer tracking bits @frontbuffer_bits by clearing them
299  * from @old and setting them in @new. Both @old and @new can be NULL.
300  */
301 void intel_frontbuffer_track(struct intel_frontbuffer *old,
302 			     struct intel_frontbuffer *new,
303 			     unsigned int frontbuffer_bits)
304 {
305 	/*
306 	 * Control of individual bits within the mask are guarded by
307 	 * the owning plane->mutex, i.e. we can never see concurrent
308 	 * manipulation of individual bits. But since the bitfield as a whole
309 	 * is updated using RMW, we need to use atomics in order to update
310 	 * the bits.
311 	 */
312 	BUILD_BUG_ON(INTEL_FRONTBUFFER_BITS_PER_PIPE * I915_MAX_PIPES >
313 		     BITS_PER_TYPE(atomic_t));
314 
315 	if (old) {
316 		drm_WARN_ON(old->obj->base.dev,
317 			    !(atomic_read(&old->bits) & frontbuffer_bits));
318 		atomic_andnot(frontbuffer_bits, &old->bits);
319 	}
320 
321 	if (new) {
322 		drm_WARN_ON(new->obj->base.dev,
323 			    atomic_read(&new->bits) & frontbuffer_bits);
324 		atomic_or(frontbuffer_bits, &new->bits);
325 	}
326 }
327