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 "intel_display_types.h"
 62#include "intel_fbc.h"
 63#include "intel_frontbuffer.h"
 64#include "intel_psr.h"
 65
 66/**
 67 * frontbuffer_flush - flush frontbuffer
 68 * @i915: i915 device
 69 * @frontbuffer_bits: frontbuffer plane tracking bits
 70 * @origin: which operation caused the flush
 71 *
 72 * This function gets called every time rendering on the given planes has
 73 * completed and frontbuffer caching can be started again. Flushes will get
 74 * delayed if they're blocked by some outstanding asynchronous rendering.
 75 *
 76 * Can be called without any locks held.
 77 */
 78static void frontbuffer_flush(struct drm_i915_private *i915,
 79			      unsigned int frontbuffer_bits,
 80			      enum fb_op_origin origin)
 81{
 82	/* Delay flushing when rings are still busy.*/
 83	spin_lock(&i915->fb_tracking.lock);
 84	frontbuffer_bits &= ~i915->fb_tracking.busy_bits;
 85	spin_unlock(&i915->fb_tracking.lock);
 86
 87	if (!frontbuffer_bits)
 88		return;
 89
 90	might_sleep();
 91	intel_edp_drrs_flush(i915, frontbuffer_bits);
 92	intel_psr_flush(i915, frontbuffer_bits, origin);
 93	intel_fbc_flush(i915, frontbuffer_bits, origin);
 94}
 95
 96/**
 97 * intel_frontbuffer_flip_prepare - prepare asynchronous frontbuffer flip
 98 * @i915: i915 device
 99 * @frontbuffer_bits: frontbuffer plane tracking bits
100 *
101 * This function gets called after scheduling a flip on @obj. The actual
102 * frontbuffer flushing will be delayed until completion is signalled with
103 * intel_frontbuffer_flip_complete. If an invalidate happens in between this
104 * flush will be cancelled.
105 *
106 * Can be called without any locks held.
107 */
108void intel_frontbuffer_flip_prepare(struct drm_i915_private *i915,
109				    unsigned frontbuffer_bits)
110{
111	spin_lock(&i915->fb_tracking.lock);
112	i915->fb_tracking.flip_bits |= frontbuffer_bits;
113	/* Remove stale busy bits due to the old buffer. */
114	i915->fb_tracking.busy_bits &= ~frontbuffer_bits;
115	spin_unlock(&i915->fb_tracking.lock);
116}
117
118/**
119 * intel_frontbuffer_flip_complete - complete asynchronous frontbuffer flip
120 * @i915: i915 device
121 * @frontbuffer_bits: frontbuffer plane tracking bits
122 *
123 * This function gets called after the flip has been latched and will complete
124 * on the next vblank. It will execute the flush if it hasn't been cancelled yet.
125 *
126 * Can be called without any locks held.
127 */
128void intel_frontbuffer_flip_complete(struct drm_i915_private *i915,
129				     unsigned frontbuffer_bits)
130{
131	spin_lock(&i915->fb_tracking.lock);
132	/* Mask any cancelled flips. */
133	frontbuffer_bits &= i915->fb_tracking.flip_bits;
134	i915->fb_tracking.flip_bits &= ~frontbuffer_bits;
135	spin_unlock(&i915->fb_tracking.lock);
136
137	if (frontbuffer_bits)
138		frontbuffer_flush(i915, frontbuffer_bits, ORIGIN_FLIP);
139}
140
141/**
142 * intel_frontbuffer_flip - synchronous frontbuffer flip
143 * @i915: i915 device
144 * @frontbuffer_bits: frontbuffer plane tracking bits
145 *
146 * This function gets called after scheduling a flip on @obj. This is for
147 * synchronous plane updates which will happen on the next vblank and which will
148 * not get delayed by pending gpu rendering.
149 *
150 * Can be called without any locks held.
151 */
152void intel_frontbuffer_flip(struct drm_i915_private *i915,
153			    unsigned frontbuffer_bits)
154{
155	spin_lock(&i915->fb_tracking.lock);
156	/* Remove stale busy bits due to the old buffer. */
157	i915->fb_tracking.busy_bits &= ~frontbuffer_bits;
158	spin_unlock(&i915->fb_tracking.lock);
159
160	frontbuffer_flush(i915, frontbuffer_bits, ORIGIN_FLIP);
161}
162
163void __intel_fb_invalidate(struct intel_frontbuffer *front,
164			   enum fb_op_origin origin,
165			   unsigned int frontbuffer_bits)
166{
167	struct drm_i915_private *i915 = to_i915(front->obj->base.dev);
168
169	if (origin == ORIGIN_CS) {
170		spin_lock(&i915->fb_tracking.lock);
171		i915->fb_tracking.busy_bits |= frontbuffer_bits;
172		i915->fb_tracking.flip_bits &= ~frontbuffer_bits;
173		spin_unlock(&i915->fb_tracking.lock);
174	}
175
176	might_sleep();
177	intel_psr_invalidate(i915, frontbuffer_bits, origin);
178	intel_edp_drrs_invalidate(i915, frontbuffer_bits);
179	intel_fbc_invalidate(i915, frontbuffer_bits, origin);
180}
181
182void __intel_fb_flush(struct intel_frontbuffer *front,
183		      enum fb_op_origin origin,
184		      unsigned int frontbuffer_bits)
185{
186	struct drm_i915_private *i915 = to_i915(front->obj->base.dev);
187
188	if (origin == ORIGIN_CS) {
189		spin_lock(&i915->fb_tracking.lock);
190		/* Filter out new bits since rendering started. */
191		frontbuffer_bits &= i915->fb_tracking.busy_bits;
192		i915->fb_tracking.busy_bits &= ~frontbuffer_bits;
193		spin_unlock(&i915->fb_tracking.lock);
194	}
195
196	if (frontbuffer_bits)
197		frontbuffer_flush(i915, frontbuffer_bits, origin);
198}
199
200static int frontbuffer_active(struct i915_active *ref)
201{
202	struct intel_frontbuffer *front =
203		container_of(ref, typeof(*front), write);
204
205	kref_get(&front->ref);
206	return 0;
207}
208
209__i915_active_call
210static void frontbuffer_retire(struct i915_active *ref)
211{
212	struct intel_frontbuffer *front =
213		container_of(ref, typeof(*front), write);
214
215	intel_frontbuffer_flush(front, ORIGIN_CS);
216	intel_frontbuffer_put(front);
217}
218
219static void frontbuffer_release(struct kref *ref)
220	__releases(&to_i915(front->obj->base.dev)->fb_tracking.lock)
221{
222	struct intel_frontbuffer *front =
223		container_of(ref, typeof(*front), ref);
224	struct drm_i915_gem_object *obj = front->obj;
225	struct i915_vma *vma;
226
227	spin_lock(&obj->vma.lock);
228	for_each_ggtt_vma(vma, obj)
229		vma->display_alignment = I915_GTT_MIN_ALIGNMENT;
230	spin_unlock(&obj->vma.lock);
231
232	RCU_INIT_POINTER(obj->frontbuffer, NULL);
233	spin_unlock(&to_i915(obj->base.dev)->fb_tracking.lock);
234
235	i915_gem_object_put(obj);
236	kfree_rcu(front, rcu);
237}
238
239struct intel_frontbuffer *
240intel_frontbuffer_get(struct drm_i915_gem_object *obj)
241{
242	struct drm_i915_private *i915 = to_i915(obj->base.dev);
243	struct intel_frontbuffer *front;
244
245	front = __intel_frontbuffer_get(obj);
246	if (front)
247		return front;
248
249	front = kmalloc(sizeof(*front), GFP_KERNEL);
250	if (!front)
251		return NULL;
252
253	front->obj = obj;
254	kref_init(&front->ref);
255	atomic_set(&front->bits, 0);
256	i915_active_init(&front->write,
257			 frontbuffer_active,
258			 i915_active_may_sleep(frontbuffer_retire));
259
260	spin_lock(&i915->fb_tracking.lock);
261	if (rcu_access_pointer(obj->frontbuffer)) {
262		kfree(front);
263		front = rcu_dereference_protected(obj->frontbuffer, true);
264		kref_get(&front->ref);
265	} else {
266		i915_gem_object_get(obj);
267		rcu_assign_pointer(obj->frontbuffer, front);
268	}
269	spin_unlock(&i915->fb_tracking.lock);
270
271	return front;
272}
273
274void intel_frontbuffer_put(struct intel_frontbuffer *front)
275{
276	kref_put_lock(&front->ref,
277		      frontbuffer_release,
278		      &to_i915(front->obj->base.dev)->fb_tracking.lock);
279}
280
281/**
282 * intel_frontbuffer_track - update frontbuffer tracking
283 * @old: current buffer for the frontbuffer slots
284 * @new: new buffer for the frontbuffer slots
285 * @frontbuffer_bits: bitmask of frontbuffer slots
286 *
287 * This updates the frontbuffer tracking bits @frontbuffer_bits by clearing them
288 * from @old and setting them in @new. Both @old and @new can be NULL.
289 */
290void intel_frontbuffer_track(struct intel_frontbuffer *old,
291			     struct intel_frontbuffer *new,
292			     unsigned int frontbuffer_bits)
293{
294	/*
295	 * Control of individual bits within the mask are guarded by
296	 * the owning plane->mutex, i.e. we can never see concurrent
297	 * manipulation of individual bits. But since the bitfield as a whole
298	 * is updated using RMW, we need to use atomics in order to update
299	 * the bits.
300	 */
301	BUILD_BUG_ON(INTEL_FRONTBUFFER_BITS_PER_PIPE * I915_MAX_PIPES >
302		     BITS_PER_TYPE(atomic_t));
303
304	if (old) {
305		drm_WARN_ON(old->obj->base.dev,
306			    !(atomic_read(&old->bits) & frontbuffer_bits));
307		atomic_andnot(frontbuffer_bits, &old->bits);
308	}
309
310	if (new) {
311		drm_WARN_ON(new->obj->base.dev,
312			    atomic_read(&new->bits) & frontbuffer_bits);
313		atomic_or(frontbuffer_bits, &new->bits);
314	}
315}