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 "gem/i915_gem_object_frontbuffer.h"
59 #include "i915_active.h"
60 #include "i915_drv.h"
61 #include "intel_display_trace.h"
62 #include "intel_display_types.h"
63 #include "intel_dp.h"
64 #include "intel_drrs.h"
65 #include "intel_fbc.h"
66 #include "intel_frontbuffer.h"
67 #include "intel_psr.h"
68 
69 /**
70  * frontbuffer_flush - flush frontbuffer
71  * @i915: i915 device
72  * @frontbuffer_bits: frontbuffer plane tracking bits
73  * @origin: which operation caused the flush
74  *
75  * This function gets called every time rendering on the given planes has
76  * completed and frontbuffer caching can be started again. Flushes will get
77  * delayed if they're blocked by some outstanding asynchronous rendering.
78  *
79  * Can be called without any locks held.
80  */
81 static void frontbuffer_flush(struct drm_i915_private *i915,
82 			      unsigned int frontbuffer_bits,
83 			      enum fb_op_origin origin)
84 {
85 	/* Delay flushing when rings are still busy.*/
86 	spin_lock(&i915->display.fb_tracking.lock);
87 	frontbuffer_bits &= ~i915->display.fb_tracking.busy_bits;
88 	spin_unlock(&i915->display.fb_tracking.lock);
89 
90 	if (!frontbuffer_bits)
91 		return;
92 
93 	trace_intel_frontbuffer_flush(i915, frontbuffer_bits, origin);
94 
95 	might_sleep();
96 	intel_drrs_flush(i915, frontbuffer_bits);
97 	intel_psr_flush(i915, frontbuffer_bits, origin);
98 	intel_fbc_flush(i915, frontbuffer_bits, origin);
99 }
100 
101 /**
102  * intel_frontbuffer_flip_prepare - prepare asynchronous frontbuffer flip
103  * @i915: i915 device
104  * @frontbuffer_bits: frontbuffer plane tracking bits
105  *
106  * This function gets called after scheduling a flip on @obj. The actual
107  * frontbuffer flushing will be delayed until completion is signalled with
108  * intel_frontbuffer_flip_complete. If an invalidate happens in between this
109  * flush will be cancelled.
110  *
111  * Can be called without any locks held.
112  */
113 void intel_frontbuffer_flip_prepare(struct drm_i915_private *i915,
114 				    unsigned frontbuffer_bits)
115 {
116 	spin_lock(&i915->display.fb_tracking.lock);
117 	i915->display.fb_tracking.flip_bits |= frontbuffer_bits;
118 	/* Remove stale busy bits due to the old buffer. */
119 	i915->display.fb_tracking.busy_bits &= ~frontbuffer_bits;
120 	spin_unlock(&i915->display.fb_tracking.lock);
121 }
122 
123 /**
124  * intel_frontbuffer_flip_complete - complete asynchronous frontbuffer flip
125  * @i915: i915 device
126  * @frontbuffer_bits: frontbuffer plane tracking bits
127  *
128  * This function gets called after the flip has been latched and will complete
129  * on the next vblank. It will execute the flush if it hasn't been cancelled yet.
130  *
131  * Can be called without any locks held.
132  */
133 void intel_frontbuffer_flip_complete(struct drm_i915_private *i915,
134 				     unsigned frontbuffer_bits)
135 {
136 	spin_lock(&i915->display.fb_tracking.lock);
137 	/* Mask any cancelled flips. */
138 	frontbuffer_bits &= i915->display.fb_tracking.flip_bits;
139 	i915->display.fb_tracking.flip_bits &= ~frontbuffer_bits;
140 	spin_unlock(&i915->display.fb_tracking.lock);
141 
142 	if (frontbuffer_bits)
143 		frontbuffer_flush(i915, frontbuffer_bits, ORIGIN_FLIP);
144 }
145 
146 /**
147  * intel_frontbuffer_flip - synchronous frontbuffer flip
148  * @i915: i915 device
149  * @frontbuffer_bits: frontbuffer plane tracking bits
150  *
151  * This function gets called after scheduling a flip on @obj. This is for
152  * synchronous plane updates which will happen on the next vblank and which will
153  * not get delayed by pending gpu rendering.
154  *
155  * Can be called without any locks held.
156  */
157 void intel_frontbuffer_flip(struct drm_i915_private *i915,
158 			    unsigned frontbuffer_bits)
159 {
160 	spin_lock(&i915->display.fb_tracking.lock);
161 	/* Remove stale busy bits due to the old buffer. */
162 	i915->display.fb_tracking.busy_bits &= ~frontbuffer_bits;
163 	spin_unlock(&i915->display.fb_tracking.lock);
164 
165 	frontbuffer_flush(i915, frontbuffer_bits, ORIGIN_FLIP);
166 }
167 
168 void __intel_fb_invalidate(struct intel_frontbuffer *front,
169 			   enum fb_op_origin origin,
170 			   unsigned int frontbuffer_bits)
171 {
172 	struct drm_i915_private *i915 = intel_bo_to_i915(front->obj);
173 
174 	if (origin == ORIGIN_CS) {
175 		spin_lock(&i915->display.fb_tracking.lock);
176 		i915->display.fb_tracking.busy_bits |= frontbuffer_bits;
177 		i915->display.fb_tracking.flip_bits &= ~frontbuffer_bits;
178 		spin_unlock(&i915->display.fb_tracking.lock);
179 	}
180 
181 	trace_intel_frontbuffer_invalidate(i915, frontbuffer_bits, origin);
182 
183 	might_sleep();
184 	intel_psr_invalidate(i915, frontbuffer_bits, origin);
185 	intel_drrs_invalidate(i915, frontbuffer_bits);
186 	intel_fbc_invalidate(i915, frontbuffer_bits, origin);
187 }
188 
189 void __intel_fb_flush(struct intel_frontbuffer *front,
190 		      enum fb_op_origin origin,
191 		      unsigned int frontbuffer_bits)
192 {
193 	struct drm_i915_private *i915 = intel_bo_to_i915(front->obj);
194 
195 	if (origin == ORIGIN_CS) {
196 		spin_lock(&i915->display.fb_tracking.lock);
197 		/* Filter out new bits since rendering started. */
198 		frontbuffer_bits &= i915->display.fb_tracking.busy_bits;
199 		i915->display.fb_tracking.busy_bits &= ~frontbuffer_bits;
200 		spin_unlock(&i915->display.fb_tracking.lock);
201 	}
202 
203 	if (frontbuffer_bits)
204 		frontbuffer_flush(i915, frontbuffer_bits, origin);
205 }
206 
207 static void intel_frontbuffer_flush_work(struct work_struct *work)
208 {
209 	struct intel_frontbuffer *front =
210 		container_of(work, struct intel_frontbuffer, flush_work);
211 
212 	i915_gem_object_flush_if_display(front->obj);
213 	intel_frontbuffer_flush(front, ORIGIN_DIRTYFB);
214 	intel_frontbuffer_put(front);
215 }
216 
217 /**
218  * intel_frontbuffer_queue_flush - queue flushing frontbuffer object
219  * @front: GEM object to flush
220  *
221  * This function is targeted for our dirty callback for queueing flush when
222  * dma fence is signales
223  */
224 void intel_frontbuffer_queue_flush(struct intel_frontbuffer *front)
225 {
226 	if (!front)
227 		return;
228 
229 	kref_get(&front->ref);
230 	if (!schedule_work(&front->flush_work))
231 		intel_frontbuffer_put(front);
232 }
233 
234 static int frontbuffer_active(struct i915_active *ref)
235 {
236 	struct intel_frontbuffer *front =
237 		container_of(ref, typeof(*front), write);
238 
239 	kref_get(&front->ref);
240 	return 0;
241 }
242 
243 static void frontbuffer_retire(struct i915_active *ref)
244 {
245 	struct intel_frontbuffer *front =
246 		container_of(ref, typeof(*front), write);
247 
248 	intel_frontbuffer_flush(front, ORIGIN_CS);
249 	intel_frontbuffer_put(front);
250 }
251 
252 static void frontbuffer_release(struct kref *ref)
253 	__releases(&intel_bo_to_i915(front->obj)->display.fb_tracking.lock)
254 {
255 	struct intel_frontbuffer *ret, *front =
256 		container_of(ref, typeof(*front), ref);
257 	struct drm_i915_gem_object *obj = front->obj;
258 
259 	drm_WARN_ON(&intel_bo_to_i915(obj)->drm, atomic_read(&front->bits));
260 
261 	i915_ggtt_clear_scanout(obj);
262 
263 	ret = i915_gem_object_set_frontbuffer(obj, NULL);
264 	drm_WARN_ON(&intel_bo_to_i915(obj)->drm, ret);
265 	spin_unlock(&intel_bo_to_i915(obj)->display.fb_tracking.lock);
266 
267 	i915_active_fini(&front->write);
268 	kfree_rcu(front, rcu);
269 }
270 
271 struct intel_frontbuffer *
272 intel_frontbuffer_get(struct drm_i915_gem_object *obj)
273 {
274 	struct drm_i915_private *i915 = intel_bo_to_i915(obj);
275 	struct intel_frontbuffer *front, *cur;
276 
277 	front = i915_gem_object_get_frontbuffer(obj);
278 	if (front)
279 		return front;
280 
281 	front = kmalloc(sizeof(*front), GFP_KERNEL);
282 	if (!front)
283 		return NULL;
284 
285 	front->obj = obj;
286 	kref_init(&front->ref);
287 	atomic_set(&front->bits, 0);
288 	i915_active_init(&front->write,
289 			 frontbuffer_active,
290 			 frontbuffer_retire,
291 			 I915_ACTIVE_RETIRE_SLEEPS);
292 	INIT_WORK(&front->flush_work, intel_frontbuffer_flush_work);
293 
294 	spin_lock(&i915->display.fb_tracking.lock);
295 	cur = i915_gem_object_set_frontbuffer(obj, front);
296 	spin_unlock(&i915->display.fb_tracking.lock);
297 	if (cur != front)
298 		kfree(front);
299 	return cur;
300 }
301 
302 void intel_frontbuffer_put(struct intel_frontbuffer *front)
303 {
304 	kref_put_lock(&front->ref,
305 		      frontbuffer_release,
306 		      &intel_bo_to_i915(front->obj)->display.fb_tracking.lock);
307 }
308 
309 /**
310  * intel_frontbuffer_track - update frontbuffer tracking
311  * @old: current buffer for the frontbuffer slots
312  * @new: new buffer for the frontbuffer slots
313  * @frontbuffer_bits: bitmask of frontbuffer slots
314  *
315  * This updates the frontbuffer tracking bits @frontbuffer_bits by clearing them
316  * from @old and setting them in @new. Both @old and @new can be NULL.
317  */
318 void intel_frontbuffer_track(struct intel_frontbuffer *old,
319 			     struct intel_frontbuffer *new,
320 			     unsigned int frontbuffer_bits)
321 {
322 	/*
323 	 * Control of individual bits within the mask are guarded by
324 	 * the owning plane->mutex, i.e. we can never see concurrent
325 	 * manipulation of individual bits. But since the bitfield as a whole
326 	 * is updated using RMW, we need to use atomics in order to update
327 	 * the bits.
328 	 */
329 	BUILD_BUG_ON(INTEL_FRONTBUFFER_BITS_PER_PIPE * I915_MAX_PIPES >
330 		     BITS_PER_TYPE(atomic_t));
331 	BUILD_BUG_ON(INTEL_FRONTBUFFER_BITS_PER_PIPE * I915_MAX_PIPES > 32);
332 	BUILD_BUG_ON(I915_MAX_PLANES > INTEL_FRONTBUFFER_BITS_PER_PIPE);
333 
334 	if (old) {
335 		drm_WARN_ON(&intel_bo_to_i915(old->obj)->drm,
336 			    !(atomic_read(&old->bits) & frontbuffer_bits));
337 		atomic_andnot(frontbuffer_bits, &old->bits);
338 	}
339 
340 	if (new) {
341 		drm_WARN_ON(&intel_bo_to_i915(new->obj)->drm,
342 			    atomic_read(&new->bits) & frontbuffer_bits);
343 		atomic_or(frontbuffer_bits, &new->bits);
344 	}
345 }
346