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 <drm/drm_gem.h>
59 #include <drm/drm_print.h>
60
61 #include "intel_bo.h"
62 #include "intel_display_trace.h"
63 #include "intel_display_types.h"
64 #include "intel_dp.h"
65 #include "intel_drrs.h"
66 #include "intel_fbc.h"
67 #include "intel_frontbuffer.h"
68 #include "intel_psr.h"
69 #include "intel_tdf.h"
70
71 /**
72 * frontbuffer_flush - flush frontbuffer
73 * @display: display device
74 * @frontbuffer_bits: frontbuffer plane tracking bits
75 * @origin: which operation caused the flush
76 *
77 * This function gets called every time rendering on the given planes has
78 * completed and frontbuffer caching can be started again. Flushes will get
79 * delayed if they're blocked by some outstanding asynchronous rendering.
80 *
81 * Can be called without any locks held.
82 */
frontbuffer_flush(struct intel_display * display,unsigned int frontbuffer_bits,enum fb_op_origin origin)83 static void frontbuffer_flush(struct intel_display *display,
84 unsigned int frontbuffer_bits,
85 enum fb_op_origin origin)
86 {
87 /* Delay flushing when rings are still busy.*/
88 spin_lock(&display->fb_tracking.lock);
89 frontbuffer_bits &= ~display->fb_tracking.busy_bits;
90 spin_unlock(&display->fb_tracking.lock);
91
92 if (!frontbuffer_bits)
93 return;
94
95 trace_intel_frontbuffer_flush(display, frontbuffer_bits, origin);
96
97 might_sleep();
98 intel_td_flush(display);
99 intel_drrs_flush(display, frontbuffer_bits);
100 intel_psr_flush(display, frontbuffer_bits, origin);
101 intel_fbc_flush(display, frontbuffer_bits, origin);
102 }
103
104 /**
105 * intel_frontbuffer_flip - synchronous frontbuffer flip
106 * @display: display device
107 * @frontbuffer_bits: frontbuffer plane tracking bits
108 *
109 * This function gets called after scheduling a flip on @obj. This is for
110 * synchronous plane updates which will happen on the next vblank and which will
111 * not get delayed by pending gpu rendering.
112 *
113 * Can be called without any locks held.
114 */
intel_frontbuffer_flip(struct intel_display * display,unsigned frontbuffer_bits)115 void intel_frontbuffer_flip(struct intel_display *display,
116 unsigned frontbuffer_bits)
117 {
118 spin_lock(&display->fb_tracking.lock);
119 /* Remove stale busy bits due to the old buffer. */
120 display->fb_tracking.busy_bits &= ~frontbuffer_bits;
121 spin_unlock(&display->fb_tracking.lock);
122
123 frontbuffer_flush(display, frontbuffer_bits, ORIGIN_FLIP);
124 }
125
__intel_fb_invalidate(struct intel_frontbuffer * front,enum fb_op_origin origin,unsigned int frontbuffer_bits)126 void __intel_fb_invalidate(struct intel_frontbuffer *front,
127 enum fb_op_origin origin,
128 unsigned int frontbuffer_bits)
129 {
130 struct intel_display *display = front->display;
131
132 if (origin == ORIGIN_CS) {
133 spin_lock(&display->fb_tracking.lock);
134 display->fb_tracking.busy_bits |= frontbuffer_bits;
135 spin_unlock(&display->fb_tracking.lock);
136 }
137
138 trace_intel_frontbuffer_invalidate(display, frontbuffer_bits, origin);
139
140 might_sleep();
141 intel_psr_invalidate(display, frontbuffer_bits, origin);
142 intel_drrs_invalidate(display, frontbuffer_bits);
143 intel_fbc_invalidate(display, frontbuffer_bits, origin);
144 }
145
__intel_fb_flush(struct intel_frontbuffer * front,enum fb_op_origin origin,unsigned int frontbuffer_bits)146 void __intel_fb_flush(struct intel_frontbuffer *front,
147 enum fb_op_origin origin,
148 unsigned int frontbuffer_bits)
149 {
150 struct intel_display *display = front->display;
151
152 if (origin == ORIGIN_DIRTYFB)
153 intel_bo_frontbuffer_flush_for_display(front);
154
155 if (origin == ORIGIN_CS) {
156 spin_lock(&display->fb_tracking.lock);
157 /* Filter out new bits since rendering started. */
158 frontbuffer_bits &= display->fb_tracking.busy_bits;
159 display->fb_tracking.busy_bits &= ~frontbuffer_bits;
160 spin_unlock(&display->fb_tracking.lock);
161 }
162
163 if (frontbuffer_bits)
164 frontbuffer_flush(display, frontbuffer_bits, origin);
165 }
166
intel_frontbuffer_ref(struct intel_frontbuffer * front)167 static void intel_frontbuffer_ref(struct intel_frontbuffer *front)
168 {
169 intel_bo_frontbuffer_ref(front);
170 }
171
intel_frontbuffer_flush_work(struct work_struct * work)172 static void intel_frontbuffer_flush_work(struct work_struct *work)
173 {
174 struct intel_frontbuffer *front =
175 container_of(work, struct intel_frontbuffer, flush_work);
176
177 intel_frontbuffer_flush(front, ORIGIN_DIRTYFB);
178 intel_frontbuffer_put(front);
179 }
180
181 /**
182 * intel_frontbuffer_queue_flush - queue flushing frontbuffer object
183 * @front: GEM object to flush
184 *
185 * This function is targeted for our dirty callback for queueing flush when
186 * dma fence is signals
187 */
intel_frontbuffer_queue_flush(struct intel_frontbuffer * front)188 void intel_frontbuffer_queue_flush(struct intel_frontbuffer *front)
189 {
190 if (!front)
191 return;
192
193 intel_frontbuffer_ref(front);
194 if (!schedule_work(&front->flush_work))
195 intel_frontbuffer_put(front);
196 }
197
intel_frontbuffer_init(struct intel_frontbuffer * front,struct drm_device * drm)198 void intel_frontbuffer_init(struct intel_frontbuffer *front, struct drm_device *drm)
199 {
200 front->display = to_intel_display(drm);
201 atomic_set(&front->bits, 0);
202 INIT_WORK(&front->flush_work, intel_frontbuffer_flush_work);
203 }
204
intel_frontbuffer_fini(struct intel_frontbuffer * front)205 void intel_frontbuffer_fini(struct intel_frontbuffer *front)
206 {
207 drm_WARN_ON(front->display->drm, atomic_read(&front->bits));
208 }
209
intel_frontbuffer_get(struct drm_gem_object * obj)210 struct intel_frontbuffer *intel_frontbuffer_get(struct drm_gem_object *obj)
211 {
212 return intel_bo_frontbuffer_get(obj);
213 }
214
intel_frontbuffer_put(struct intel_frontbuffer * front)215 void intel_frontbuffer_put(struct intel_frontbuffer *front)
216 {
217 intel_bo_frontbuffer_put(front);
218 }
219
220 /**
221 * intel_frontbuffer_track - update frontbuffer tracking
222 * @old: current buffer for the frontbuffer slots
223 * @new: new buffer for the frontbuffer slots
224 * @frontbuffer_bits: bitmask of frontbuffer slots
225 *
226 * This updates the frontbuffer tracking bits @frontbuffer_bits by clearing them
227 * from @old and setting them in @new. Both @old and @new can be NULL.
228 */
intel_frontbuffer_track(struct intel_frontbuffer * old,struct intel_frontbuffer * new,unsigned int frontbuffer_bits)229 void intel_frontbuffer_track(struct intel_frontbuffer *old,
230 struct intel_frontbuffer *new,
231 unsigned int frontbuffer_bits)
232 {
233 /*
234 * Control of individual bits within the mask are guarded by
235 * the owning plane->mutex, i.e. we can never see concurrent
236 * manipulation of individual bits. But since the bitfield as a whole
237 * is updated using RMW, we need to use atomics in order to update
238 * the bits.
239 */
240 BUILD_BUG_ON(INTEL_FRONTBUFFER_BITS_PER_PIPE * I915_MAX_PIPES >
241 BITS_PER_TYPE(atomic_t));
242 BUILD_BUG_ON(INTEL_FRONTBUFFER_BITS_PER_PIPE * I915_MAX_PIPES > 32);
243 BUILD_BUG_ON(I915_MAX_PLANES > INTEL_FRONTBUFFER_BITS_PER_PIPE);
244
245 if (old) {
246 drm_WARN_ON(old->display->drm,
247 !(atomic_read(&old->bits) & frontbuffer_bits));
248 atomic_andnot(frontbuffer_bits, &old->bits);
249 }
250
251 if (new) {
252 drm_WARN_ON(new->display->drm,
253 atomic_read(&new->bits) & frontbuffer_bits);
254 atomic_or(frontbuffer_bits, &new->bits);
255 }
256 }
257