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
60 #include "i915_active.h"
61 #include "i915_drv.h"
62 #include "intel_bo.h"
63 #include "intel_display_trace.h"
64 #include "intel_display_types.h"
65 #include "intel_dp.h"
66 #include "intel_drrs.h"
67 #include "intel_fbc.h"
68 #include "intel_frontbuffer.h"
69 #include "intel_psr.h"
70 #include "intel_tdf.h"
71
72 /**
73 * frontbuffer_flush - flush frontbuffer
74 * @i915: i915 device
75 * @frontbuffer_bits: frontbuffer plane tracking bits
76 * @origin: which operation caused the flush
77 *
78 * This function gets called every time rendering on the given planes has
79 * completed and frontbuffer caching can be started again. Flushes will get
80 * delayed if they're blocked by some outstanding asynchronous rendering.
81 *
82 * Can be called without any locks held.
83 */
frontbuffer_flush(struct drm_i915_private * i915,unsigned int frontbuffer_bits,enum fb_op_origin origin)84 static void frontbuffer_flush(struct drm_i915_private *i915,
85 unsigned int frontbuffer_bits,
86 enum fb_op_origin origin)
87 {
88 struct intel_display *display = &i915->display;
89
90 /* Delay flushing when rings are still busy.*/
91 spin_lock(&i915->display.fb_tracking.lock);
92 frontbuffer_bits &= ~i915->display.fb_tracking.busy_bits;
93 spin_unlock(&i915->display.fb_tracking.lock);
94
95 if (!frontbuffer_bits)
96 return;
97
98 trace_intel_frontbuffer_flush(display, frontbuffer_bits, origin);
99
100 might_sleep();
101 intel_td_flush(i915);
102 intel_drrs_flush(i915, frontbuffer_bits);
103 intel_psr_flush(display, frontbuffer_bits, origin);
104 intel_fbc_flush(i915, frontbuffer_bits, origin);
105 }
106
107 /**
108 * intel_frontbuffer_flip_prepare - prepare asynchronous frontbuffer flip
109 * @i915: i915 device
110 * @frontbuffer_bits: frontbuffer plane tracking bits
111 *
112 * This function gets called after scheduling a flip on @obj. The actual
113 * frontbuffer flushing will be delayed until completion is signalled with
114 * intel_frontbuffer_flip_complete. If an invalidate happens in between this
115 * flush will be cancelled.
116 *
117 * Can be called without any locks held.
118 */
intel_frontbuffer_flip_prepare(struct drm_i915_private * i915,unsigned frontbuffer_bits)119 void intel_frontbuffer_flip_prepare(struct drm_i915_private *i915,
120 unsigned frontbuffer_bits)
121 {
122 spin_lock(&i915->display.fb_tracking.lock);
123 i915->display.fb_tracking.flip_bits |= frontbuffer_bits;
124 /* Remove stale busy bits due to the old buffer. */
125 i915->display.fb_tracking.busy_bits &= ~frontbuffer_bits;
126 spin_unlock(&i915->display.fb_tracking.lock);
127 }
128
129 /**
130 * intel_frontbuffer_flip_complete - complete asynchronous frontbuffer flip
131 * @i915: i915 device
132 * @frontbuffer_bits: frontbuffer plane tracking bits
133 *
134 * This function gets called after the flip has been latched and will complete
135 * on the next vblank. It will execute the flush if it hasn't been cancelled yet.
136 *
137 * Can be called without any locks held.
138 */
intel_frontbuffer_flip_complete(struct drm_i915_private * i915,unsigned frontbuffer_bits)139 void intel_frontbuffer_flip_complete(struct drm_i915_private *i915,
140 unsigned frontbuffer_bits)
141 {
142 spin_lock(&i915->display.fb_tracking.lock);
143 /* Mask any cancelled flips. */
144 frontbuffer_bits &= i915->display.fb_tracking.flip_bits;
145 i915->display.fb_tracking.flip_bits &= ~frontbuffer_bits;
146 spin_unlock(&i915->display.fb_tracking.lock);
147
148 if (frontbuffer_bits)
149 frontbuffer_flush(i915, frontbuffer_bits, ORIGIN_FLIP);
150 }
151
152 /**
153 * intel_frontbuffer_flip - synchronous frontbuffer flip
154 * @i915: i915 device
155 * @frontbuffer_bits: frontbuffer plane tracking bits
156 *
157 * This function gets called after scheduling a flip on @obj. This is for
158 * synchronous plane updates which will happen on the next vblank and which will
159 * not get delayed by pending gpu rendering.
160 *
161 * Can be called without any locks held.
162 */
intel_frontbuffer_flip(struct drm_i915_private * i915,unsigned frontbuffer_bits)163 void intel_frontbuffer_flip(struct drm_i915_private *i915,
164 unsigned frontbuffer_bits)
165 {
166 spin_lock(&i915->display.fb_tracking.lock);
167 /* Remove stale busy bits due to the old buffer. */
168 i915->display.fb_tracking.busy_bits &= ~frontbuffer_bits;
169 spin_unlock(&i915->display.fb_tracking.lock);
170
171 frontbuffer_flush(i915, frontbuffer_bits, ORIGIN_FLIP);
172 }
173
__intel_fb_invalidate(struct intel_frontbuffer * front,enum fb_op_origin origin,unsigned int frontbuffer_bits)174 void __intel_fb_invalidate(struct intel_frontbuffer *front,
175 enum fb_op_origin origin,
176 unsigned int frontbuffer_bits)
177 {
178 struct intel_display *display = to_intel_display(front->obj->dev);
179 struct drm_i915_private *i915 = to_i915(display->drm);
180
181 if (origin == ORIGIN_CS) {
182 spin_lock(&display->fb_tracking.lock);
183 display->fb_tracking.busy_bits |= frontbuffer_bits;
184 display->fb_tracking.flip_bits &= ~frontbuffer_bits;
185 spin_unlock(&display->fb_tracking.lock);
186 }
187
188 trace_intel_frontbuffer_invalidate(display, frontbuffer_bits, origin);
189
190 might_sleep();
191 intel_psr_invalidate(display, frontbuffer_bits, origin);
192 intel_drrs_invalidate(i915, frontbuffer_bits);
193 intel_fbc_invalidate(i915, frontbuffer_bits, origin);
194 }
195
__intel_fb_flush(struct intel_frontbuffer * front,enum fb_op_origin origin,unsigned int frontbuffer_bits)196 void __intel_fb_flush(struct intel_frontbuffer *front,
197 enum fb_op_origin origin,
198 unsigned int frontbuffer_bits)
199 {
200 struct intel_display *display = to_intel_display(front->obj->dev);
201 struct drm_i915_private *i915 = to_i915(display->drm);
202
203 if (origin == ORIGIN_CS) {
204 spin_lock(&display->fb_tracking.lock);
205 /* Filter out new bits since rendering started. */
206 frontbuffer_bits &= display->fb_tracking.busy_bits;
207 display->fb_tracking.busy_bits &= ~frontbuffer_bits;
208 spin_unlock(&display->fb_tracking.lock);
209 }
210
211 if (frontbuffer_bits)
212 frontbuffer_flush(i915, frontbuffer_bits, origin);
213 }
214
intel_frontbuffer_flush_work(struct work_struct * work)215 static void intel_frontbuffer_flush_work(struct work_struct *work)
216 {
217 struct intel_frontbuffer *front =
218 container_of(work, struct intel_frontbuffer, flush_work);
219
220 intel_bo_flush_if_display(front->obj);
221 intel_frontbuffer_flush(front, ORIGIN_DIRTYFB);
222 intel_frontbuffer_put(front);
223 }
224
225 /**
226 * intel_frontbuffer_queue_flush - queue flushing frontbuffer object
227 * @front: GEM object to flush
228 *
229 * This function is targeted for our dirty callback for queueing flush when
230 * dma fence is signales
231 */
intel_frontbuffer_queue_flush(struct intel_frontbuffer * front)232 void intel_frontbuffer_queue_flush(struct intel_frontbuffer *front)
233 {
234 if (!front)
235 return;
236
237 kref_get(&front->ref);
238 if (!schedule_work(&front->flush_work))
239 intel_frontbuffer_put(front);
240 }
241
frontbuffer_active(struct i915_active * ref)242 static int frontbuffer_active(struct i915_active *ref)
243 {
244 struct intel_frontbuffer *front =
245 container_of(ref, typeof(*front), write);
246
247 kref_get(&front->ref);
248 return 0;
249 }
250
frontbuffer_retire(struct i915_active * ref)251 static void frontbuffer_retire(struct i915_active *ref)
252 {
253 struct intel_frontbuffer *front =
254 container_of(ref, typeof(*front), write);
255
256 intel_frontbuffer_flush(front, ORIGIN_CS);
257 intel_frontbuffer_put(front);
258 }
259
frontbuffer_release(struct kref * ref)260 static void frontbuffer_release(struct kref *ref)
261 __releases(&to_intel_display(front->obj->dev)->fb_tracking.lock)
262 {
263 struct intel_frontbuffer *ret, *front =
264 container_of(ref, typeof(*front), ref);
265 struct drm_gem_object *obj = front->obj;
266 struct intel_display *display = to_intel_display(obj->dev);
267
268 drm_WARN_ON(display->drm, atomic_read(&front->bits));
269
270 i915_ggtt_clear_scanout(to_intel_bo(obj));
271
272 ret = intel_bo_set_frontbuffer(obj, NULL);
273 drm_WARN_ON(display->drm, ret);
274 spin_unlock(&display->fb_tracking.lock);
275
276 i915_active_fini(&front->write);
277 kfree_rcu(front, rcu);
278 }
279
280 struct intel_frontbuffer *
intel_frontbuffer_get(struct drm_gem_object * obj)281 intel_frontbuffer_get(struct drm_gem_object *obj)
282 {
283 struct drm_i915_private *i915 = to_i915(obj->dev);
284 struct intel_frontbuffer *front, *cur;
285
286 front = intel_bo_get_frontbuffer(obj);
287 if (front)
288 return front;
289
290 front = kmalloc(sizeof(*front), GFP_KERNEL);
291 if (!front)
292 return NULL;
293
294 front->obj = obj;
295 kref_init(&front->ref);
296 atomic_set(&front->bits, 0);
297 i915_active_init(&front->write,
298 frontbuffer_active,
299 frontbuffer_retire,
300 I915_ACTIVE_RETIRE_SLEEPS);
301 INIT_WORK(&front->flush_work, intel_frontbuffer_flush_work);
302
303 spin_lock(&i915->display.fb_tracking.lock);
304 cur = intel_bo_set_frontbuffer(obj, front);
305 spin_unlock(&i915->display.fb_tracking.lock);
306 if (cur != front)
307 kfree(front);
308 return cur;
309 }
310
intel_frontbuffer_put(struct intel_frontbuffer * front)311 void intel_frontbuffer_put(struct intel_frontbuffer *front)
312 {
313 kref_put_lock(&front->ref,
314 frontbuffer_release,
315 &to_intel_display(front->obj->dev)->fb_tracking.lock);
316 }
317
318 /**
319 * intel_frontbuffer_track - update frontbuffer tracking
320 * @old: current buffer for the frontbuffer slots
321 * @new: new buffer for the frontbuffer slots
322 * @frontbuffer_bits: bitmask of frontbuffer slots
323 *
324 * This updates the frontbuffer tracking bits @frontbuffer_bits by clearing them
325 * from @old and setting them in @new. Both @old and @new can be NULL.
326 */
intel_frontbuffer_track(struct intel_frontbuffer * old,struct intel_frontbuffer * new,unsigned int frontbuffer_bits)327 void intel_frontbuffer_track(struct intel_frontbuffer *old,
328 struct intel_frontbuffer *new,
329 unsigned int frontbuffer_bits)
330 {
331 /*
332 * Control of individual bits within the mask are guarded by
333 * the owning plane->mutex, i.e. we can never see concurrent
334 * manipulation of individual bits. But since the bitfield as a whole
335 * is updated using RMW, we need to use atomics in order to update
336 * the bits.
337 */
338 BUILD_BUG_ON(INTEL_FRONTBUFFER_BITS_PER_PIPE * I915_MAX_PIPES >
339 BITS_PER_TYPE(atomic_t));
340 BUILD_BUG_ON(INTEL_FRONTBUFFER_BITS_PER_PIPE * I915_MAX_PIPES > 32);
341 BUILD_BUG_ON(I915_MAX_PLANES > INTEL_FRONTBUFFER_BITS_PER_PIPE);
342
343 if (old) {
344 struct intel_display *display = to_intel_display(old->obj->dev);
345
346 drm_WARN_ON(display->drm,
347 !(atomic_read(&old->bits) & frontbuffer_bits));
348 atomic_andnot(frontbuffer_bits, &old->bits);
349 }
350
351 if (new) {
352 struct intel_display *display = to_intel_display(new->obj->dev);
353
354 drm_WARN_ON(display->drm,
355 atomic_read(&new->bits) & frontbuffer_bits);
356 atomic_or(frontbuffer_bits, &new->bits);
357 }
358 }
359