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 "display/intel_dp.h"
59 
60 #include "i915_drv.h"
61 #include "i915_trace.h"
62 #include "intel_display_types.h"
63 #include "intel_fbc.h"
64 #include "intel_frontbuffer.h"
65 #include "intel_drrs.h"
66 #include "intel_psr.h"
67 
68 /**
69  * frontbuffer_flush - flush frontbuffer
70  * @i915: i915 device
71  * @frontbuffer_bits: frontbuffer plane tracking bits
72  * @origin: which operation caused the flush
73  *
74  * This function gets called every time rendering on the given planes has
75  * completed and frontbuffer caching can be started again. Flushes will get
76  * delayed if they're blocked by some outstanding asynchronous rendering.
77  *
78  * Can be called without any locks held.
79  */
80 static void frontbuffer_flush(struct drm_i915_private *i915,
81 			      unsigned int frontbuffer_bits,
82 			      enum fb_op_origin origin)
83 {
84 	/* Delay flushing when rings are still busy.*/
85 	spin_lock(&i915->fb_tracking.lock);
86 	frontbuffer_bits &= ~i915->fb_tracking.busy_bits;
87 	spin_unlock(&i915->fb_tracking.lock);
88 
89 	if (!frontbuffer_bits)
90 		return;
91 
92 	trace_intel_frontbuffer_flush(frontbuffer_bits, origin);
93 
94 	might_sleep();
95 	intel_drrs_flush(i915, frontbuffer_bits);
96 	intel_psr_flush(i915, frontbuffer_bits, origin);
97 	intel_fbc_flush(i915, frontbuffer_bits, origin);
98 }
99 
100 /**
101  * intel_frontbuffer_flip_prepare - prepare asynchronous frontbuffer flip
102  * @i915: i915 device
103  * @frontbuffer_bits: frontbuffer plane tracking bits
104  *
105  * This function gets called after scheduling a flip on @obj. The actual
106  * frontbuffer flushing will be delayed until completion is signalled with
107  * intel_frontbuffer_flip_complete. If an invalidate happens in between this
108  * flush will be cancelled.
109  *
110  * Can be called without any locks held.
111  */
112 void intel_frontbuffer_flip_prepare(struct drm_i915_private *i915,
113 				    unsigned frontbuffer_bits)
114 {
115 	spin_lock(&i915->fb_tracking.lock);
116 	i915->fb_tracking.flip_bits |= frontbuffer_bits;
117 	/* Remove stale busy bits due to the old buffer. */
118 	i915->fb_tracking.busy_bits &= ~frontbuffer_bits;
119 	spin_unlock(&i915->fb_tracking.lock);
120 }
121 
122 /**
123  * intel_frontbuffer_flip_complete - complete asynchronous frontbuffer flip
124  * @i915: i915 device
125  * @frontbuffer_bits: frontbuffer plane tracking bits
126  *
127  * This function gets called after the flip has been latched and will complete
128  * on the next vblank. It will execute the flush if it hasn't been cancelled yet.
129  *
130  * Can be called without any locks held.
131  */
132 void intel_frontbuffer_flip_complete(struct drm_i915_private *i915,
133 				     unsigned frontbuffer_bits)
134 {
135 	spin_lock(&i915->fb_tracking.lock);
136 	/* Mask any cancelled flips. */
137 	frontbuffer_bits &= i915->fb_tracking.flip_bits;
138 	i915->fb_tracking.flip_bits &= ~frontbuffer_bits;
139 	spin_unlock(&i915->fb_tracking.lock);
140 
141 	if (frontbuffer_bits)
142 		frontbuffer_flush(i915, frontbuffer_bits, ORIGIN_FLIP);
143 }
144 
145 /**
146  * intel_frontbuffer_flip - synchronous frontbuffer flip
147  * @i915: i915 device
148  * @frontbuffer_bits: frontbuffer plane tracking bits
149  *
150  * This function gets called after scheduling a flip on @obj. This is for
151  * synchronous plane updates which will happen on the next vblank and which will
152  * not get delayed by pending gpu rendering.
153  *
154  * Can be called without any locks held.
155  */
156 void intel_frontbuffer_flip(struct drm_i915_private *i915,
157 			    unsigned frontbuffer_bits)
158 {
159 	spin_lock(&i915->fb_tracking.lock);
160 	/* Remove stale busy bits due to the old buffer. */
161 	i915->fb_tracking.busy_bits &= ~frontbuffer_bits;
162 	spin_unlock(&i915->fb_tracking.lock);
163 
164 	frontbuffer_flush(i915, frontbuffer_bits, ORIGIN_FLIP);
165 }
166 
167 void __intel_fb_invalidate(struct intel_frontbuffer *front,
168 			   enum fb_op_origin origin,
169 			   unsigned int frontbuffer_bits)
170 {
171 	struct drm_i915_private *i915 = to_i915(front->obj->base.dev);
172 
173 	if (origin == ORIGIN_CS) {
174 		spin_lock(&i915->fb_tracking.lock);
175 		i915->fb_tracking.busy_bits |= frontbuffer_bits;
176 		i915->fb_tracking.flip_bits &= ~frontbuffer_bits;
177 		spin_unlock(&i915->fb_tracking.lock);
178 	}
179 
180 	trace_intel_frontbuffer_invalidate(frontbuffer_bits, origin);
181 
182 	might_sleep();
183 	intel_psr_invalidate(i915, frontbuffer_bits, origin);
184 	intel_drrs_invalidate(i915, frontbuffer_bits);
185 	intel_fbc_invalidate(i915, frontbuffer_bits, origin);
186 }
187 
188 void __intel_fb_flush(struct intel_frontbuffer *front,
189 		      enum fb_op_origin origin,
190 		      unsigned int frontbuffer_bits)
191 {
192 	struct drm_i915_private *i915 = to_i915(front->obj->base.dev);
193 
194 	if (origin == ORIGIN_CS) {
195 		spin_lock(&i915->fb_tracking.lock);
196 		/* Filter out new bits since rendering started. */
197 		frontbuffer_bits &= i915->fb_tracking.busy_bits;
198 		i915->fb_tracking.busy_bits &= ~frontbuffer_bits;
199 		spin_unlock(&i915->fb_tracking.lock);
200 	}
201 
202 	if (frontbuffer_bits)
203 		frontbuffer_flush(i915, frontbuffer_bits, origin);
204 }
205 
206 static int frontbuffer_active(struct i915_active *ref)
207 {
208 	struct intel_frontbuffer *front =
209 		container_of(ref, typeof(*front), write);
210 
211 	kref_get(&front->ref);
212 	return 0;
213 }
214 
215 static void frontbuffer_retire(struct i915_active *ref)
216 {
217 	struct intel_frontbuffer *front =
218 		container_of(ref, typeof(*front), write);
219 
220 	intel_frontbuffer_flush(front, ORIGIN_CS);
221 	intel_frontbuffer_put(front);
222 }
223 
224 static void frontbuffer_release(struct kref *ref)
225 	__releases(&to_i915(front->obj->base.dev)->fb_tracking.lock)
226 {
227 	struct intel_frontbuffer *front =
228 		container_of(ref, typeof(*front), ref);
229 	struct drm_i915_gem_object *obj = front->obj;
230 	struct i915_vma *vma;
231 
232 	drm_WARN_ON(obj->base.dev, atomic_read(&front->bits));
233 
234 	spin_lock(&obj->vma.lock);
235 	for_each_ggtt_vma(vma, obj) {
236 		i915_vma_clear_scanout(vma);
237 		vma->display_alignment = I915_GTT_MIN_ALIGNMENT;
238 	}
239 	spin_unlock(&obj->vma.lock);
240 
241 	RCU_INIT_POINTER(obj->frontbuffer, NULL);
242 	spin_unlock(&to_i915(obj->base.dev)->fb_tracking.lock);
243 
244 	i915_active_fini(&front->write);
245 
246 	i915_gem_object_put(obj);
247 	kfree_rcu(front, rcu);
248 }
249 
250 struct intel_frontbuffer *
251 intel_frontbuffer_get(struct drm_i915_gem_object *obj)
252 {
253 	struct drm_i915_private *i915 = to_i915(obj->base.dev);
254 	struct intel_frontbuffer *front;
255 
256 	front = __intel_frontbuffer_get(obj);
257 	if (front)
258 		return front;
259 
260 	front = kmalloc(sizeof(*front), GFP_KERNEL);
261 	if (!front)
262 		return NULL;
263 
264 	front->obj = obj;
265 	kref_init(&front->ref);
266 	atomic_set(&front->bits, 0);
267 	i915_active_init(&front->write,
268 			 frontbuffer_active,
269 			 frontbuffer_retire,
270 			 I915_ACTIVE_RETIRE_SLEEPS);
271 
272 	spin_lock(&i915->fb_tracking.lock);
273 	if (rcu_access_pointer(obj->frontbuffer)) {
274 		kfree(front);
275 		front = rcu_dereference_protected(obj->frontbuffer, true);
276 		kref_get(&front->ref);
277 	} else {
278 		i915_gem_object_get(obj);
279 		rcu_assign_pointer(obj->frontbuffer, front);
280 	}
281 	spin_unlock(&i915->fb_tracking.lock);
282 
283 	return front;
284 }
285 
286 void intel_frontbuffer_put(struct intel_frontbuffer *front)
287 {
288 	kref_put_lock(&front->ref,
289 		      frontbuffer_release,
290 		      &to_i915(front->obj->base.dev)->fb_tracking.lock);
291 }
292 
293 /**
294  * intel_frontbuffer_track - update frontbuffer tracking
295  * @old: current buffer for the frontbuffer slots
296  * @new: new buffer for the frontbuffer slots
297  * @frontbuffer_bits: bitmask of frontbuffer slots
298  *
299  * This updates the frontbuffer tracking bits @frontbuffer_bits by clearing them
300  * from @old and setting them in @new. Both @old and @new can be NULL.
301  */
302 void intel_frontbuffer_track(struct intel_frontbuffer *old,
303 			     struct intel_frontbuffer *new,
304 			     unsigned int frontbuffer_bits)
305 {
306 	/*
307 	 * Control of individual bits within the mask are guarded by
308 	 * the owning plane->mutex, i.e. we can never see concurrent
309 	 * manipulation of individual bits. But since the bitfield as a whole
310 	 * is updated using RMW, we need to use atomics in order to update
311 	 * the bits.
312 	 */
313 	BUILD_BUG_ON(INTEL_FRONTBUFFER_BITS_PER_PIPE * I915_MAX_PIPES >
314 		     BITS_PER_TYPE(atomic_t));
315 
316 	if (old) {
317 		drm_WARN_ON(old->obj->base.dev,
318 			    !(atomic_read(&old->bits) & frontbuffer_bits));
319 		atomic_andnot(frontbuffer_bits, &old->bits);
320 	}
321 
322 	if (new) {
323 		drm_WARN_ON(new->obj->base.dev,
324 			    atomic_read(&new->bits) & frontbuffer_bits);
325 		atomic_or(frontbuffer_bits, &new->bits);
326 	}
327 }
328