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 #include "intel_tdf.h" 69 70 /** 71 * frontbuffer_flush - flush frontbuffer 72 * @i915: i915 device 73 * @frontbuffer_bits: frontbuffer plane tracking bits 74 * @origin: which operation caused the flush 75 * 76 * This function gets called every time rendering on the given planes has 77 * completed and frontbuffer caching can be started again. Flushes will get 78 * delayed if they're blocked by some outstanding asynchronous rendering. 79 * 80 * Can be called without any locks held. 81 */ 82 static void frontbuffer_flush(struct drm_i915_private *i915, 83 unsigned int frontbuffer_bits, 84 enum fb_op_origin origin) 85 { 86 struct intel_display *display = &i915->display; 87 88 /* Delay flushing when rings are still busy.*/ 89 spin_lock(&i915->display.fb_tracking.lock); 90 frontbuffer_bits &= ~i915->display.fb_tracking.busy_bits; 91 spin_unlock(&i915->display.fb_tracking.lock); 92 93 if (!frontbuffer_bits) 94 return; 95 96 trace_intel_frontbuffer_flush(i915, frontbuffer_bits, origin); 97 98 might_sleep(); 99 intel_td_flush(i915); 100 intel_drrs_flush(i915, frontbuffer_bits); 101 intel_psr_flush(display, frontbuffer_bits, origin); 102 intel_fbc_flush(i915, frontbuffer_bits, origin); 103 } 104 105 /** 106 * intel_frontbuffer_flip_prepare - prepare asynchronous frontbuffer flip 107 * @i915: i915 device 108 * @frontbuffer_bits: frontbuffer plane tracking bits 109 * 110 * This function gets called after scheduling a flip on @obj. The actual 111 * frontbuffer flushing will be delayed until completion is signalled with 112 * intel_frontbuffer_flip_complete. If an invalidate happens in between this 113 * flush will be cancelled. 114 * 115 * Can be called without any locks held. 116 */ 117 void intel_frontbuffer_flip_prepare(struct drm_i915_private *i915, 118 unsigned frontbuffer_bits) 119 { 120 spin_lock(&i915->display.fb_tracking.lock); 121 i915->display.fb_tracking.flip_bits |= frontbuffer_bits; 122 /* Remove stale busy bits due to the old buffer. */ 123 i915->display.fb_tracking.busy_bits &= ~frontbuffer_bits; 124 spin_unlock(&i915->display.fb_tracking.lock); 125 } 126 127 /** 128 * intel_frontbuffer_flip_complete - complete asynchronous frontbuffer flip 129 * @i915: i915 device 130 * @frontbuffer_bits: frontbuffer plane tracking bits 131 * 132 * This function gets called after the flip has been latched and will complete 133 * on the next vblank. It will execute the flush if it hasn't been cancelled yet. 134 * 135 * Can be called without any locks held. 136 */ 137 void intel_frontbuffer_flip_complete(struct drm_i915_private *i915, 138 unsigned frontbuffer_bits) 139 { 140 spin_lock(&i915->display.fb_tracking.lock); 141 /* Mask any cancelled flips. */ 142 frontbuffer_bits &= i915->display.fb_tracking.flip_bits; 143 i915->display.fb_tracking.flip_bits &= ~frontbuffer_bits; 144 spin_unlock(&i915->display.fb_tracking.lock); 145 146 if (frontbuffer_bits) 147 frontbuffer_flush(i915, frontbuffer_bits, ORIGIN_FLIP); 148 } 149 150 /** 151 * intel_frontbuffer_flip - synchronous frontbuffer flip 152 * @i915: i915 device 153 * @frontbuffer_bits: frontbuffer plane tracking bits 154 * 155 * This function gets called after scheduling a flip on @obj. This is for 156 * synchronous plane updates which will happen on the next vblank and which will 157 * not get delayed by pending gpu rendering. 158 * 159 * Can be called without any locks held. 160 */ 161 void intel_frontbuffer_flip(struct drm_i915_private *i915, 162 unsigned frontbuffer_bits) 163 { 164 spin_lock(&i915->display.fb_tracking.lock); 165 /* Remove stale busy bits due to the old buffer. */ 166 i915->display.fb_tracking.busy_bits &= ~frontbuffer_bits; 167 spin_unlock(&i915->display.fb_tracking.lock); 168 169 frontbuffer_flush(i915, frontbuffer_bits, ORIGIN_FLIP); 170 } 171 172 void __intel_fb_invalidate(struct intel_frontbuffer *front, 173 enum fb_op_origin origin, 174 unsigned int frontbuffer_bits) 175 { 176 struct drm_i915_private *i915 = intel_bo_to_i915(front->obj); 177 struct intel_display *display = &i915->display; 178 179 if (origin == ORIGIN_CS) { 180 spin_lock(&i915->display.fb_tracking.lock); 181 i915->display.fb_tracking.busy_bits |= frontbuffer_bits; 182 i915->display.fb_tracking.flip_bits &= ~frontbuffer_bits; 183 spin_unlock(&i915->display.fb_tracking.lock); 184 } 185 186 trace_intel_frontbuffer_invalidate(i915, frontbuffer_bits, origin); 187 188 might_sleep(); 189 intel_psr_invalidate(display, frontbuffer_bits, origin); 190 intel_drrs_invalidate(i915, frontbuffer_bits); 191 intel_fbc_invalidate(i915, frontbuffer_bits, origin); 192 } 193 194 void __intel_fb_flush(struct intel_frontbuffer *front, 195 enum fb_op_origin origin, 196 unsigned int frontbuffer_bits) 197 { 198 struct drm_i915_private *i915 = intel_bo_to_i915(front->obj); 199 200 if (origin == ORIGIN_CS) { 201 spin_lock(&i915->display.fb_tracking.lock); 202 /* Filter out new bits since rendering started. */ 203 frontbuffer_bits &= i915->display.fb_tracking.busy_bits; 204 i915->display.fb_tracking.busy_bits &= ~frontbuffer_bits; 205 spin_unlock(&i915->display.fb_tracking.lock); 206 } 207 208 if (frontbuffer_bits) 209 frontbuffer_flush(i915, frontbuffer_bits, origin); 210 } 211 212 static void intel_frontbuffer_flush_work(struct work_struct *work) 213 { 214 struct intel_frontbuffer *front = 215 container_of(work, struct intel_frontbuffer, flush_work); 216 217 i915_gem_object_flush_if_display(front->obj); 218 intel_frontbuffer_flush(front, ORIGIN_DIRTYFB); 219 intel_frontbuffer_put(front); 220 } 221 222 /** 223 * intel_frontbuffer_queue_flush - queue flushing frontbuffer object 224 * @front: GEM object to flush 225 * 226 * This function is targeted for our dirty callback for queueing flush when 227 * dma fence is signales 228 */ 229 void intel_frontbuffer_queue_flush(struct intel_frontbuffer *front) 230 { 231 if (!front) 232 return; 233 234 kref_get(&front->ref); 235 if (!schedule_work(&front->flush_work)) 236 intel_frontbuffer_put(front); 237 } 238 239 static int frontbuffer_active(struct i915_active *ref) 240 { 241 struct intel_frontbuffer *front = 242 container_of(ref, typeof(*front), write); 243 244 kref_get(&front->ref); 245 return 0; 246 } 247 248 static void frontbuffer_retire(struct i915_active *ref) 249 { 250 struct intel_frontbuffer *front = 251 container_of(ref, typeof(*front), write); 252 253 intel_frontbuffer_flush(front, ORIGIN_CS); 254 intel_frontbuffer_put(front); 255 } 256 257 static void frontbuffer_release(struct kref *ref) 258 __releases(&intel_bo_to_i915(front->obj)->display.fb_tracking.lock) 259 { 260 struct intel_frontbuffer *ret, *front = 261 container_of(ref, typeof(*front), ref); 262 struct drm_i915_gem_object *obj = front->obj; 263 264 drm_WARN_ON(&intel_bo_to_i915(obj)->drm, atomic_read(&front->bits)); 265 266 i915_ggtt_clear_scanout(obj); 267 268 ret = i915_gem_object_set_frontbuffer(obj, NULL); 269 drm_WARN_ON(&intel_bo_to_i915(obj)->drm, ret); 270 spin_unlock(&intel_bo_to_i915(obj)->display.fb_tracking.lock); 271 272 i915_active_fini(&front->write); 273 kfree_rcu(front, rcu); 274 } 275 276 struct intel_frontbuffer * 277 intel_frontbuffer_get(struct drm_i915_gem_object *obj) 278 { 279 struct drm_i915_private *i915 = intel_bo_to_i915(obj); 280 struct intel_frontbuffer *front, *cur; 281 282 front = i915_gem_object_get_frontbuffer(obj); 283 if (front) 284 return front; 285 286 front = kmalloc(sizeof(*front), GFP_KERNEL); 287 if (!front) 288 return NULL; 289 290 front->obj = obj; 291 kref_init(&front->ref); 292 atomic_set(&front->bits, 0); 293 i915_active_init(&front->write, 294 frontbuffer_active, 295 frontbuffer_retire, 296 I915_ACTIVE_RETIRE_SLEEPS); 297 INIT_WORK(&front->flush_work, intel_frontbuffer_flush_work); 298 299 spin_lock(&i915->display.fb_tracking.lock); 300 cur = i915_gem_object_set_frontbuffer(obj, front); 301 spin_unlock(&i915->display.fb_tracking.lock); 302 if (cur != front) 303 kfree(front); 304 return cur; 305 } 306 307 void intel_frontbuffer_put(struct intel_frontbuffer *front) 308 { 309 kref_put_lock(&front->ref, 310 frontbuffer_release, 311 &intel_bo_to_i915(front->obj)->display.fb_tracking.lock); 312 } 313 314 /** 315 * intel_frontbuffer_track - update frontbuffer tracking 316 * @old: current buffer for the frontbuffer slots 317 * @new: new buffer for the frontbuffer slots 318 * @frontbuffer_bits: bitmask of frontbuffer slots 319 * 320 * This updates the frontbuffer tracking bits @frontbuffer_bits by clearing them 321 * from @old and setting them in @new. Both @old and @new can be NULL. 322 */ 323 void intel_frontbuffer_track(struct intel_frontbuffer *old, 324 struct intel_frontbuffer *new, 325 unsigned int frontbuffer_bits) 326 { 327 /* 328 * Control of individual bits within the mask are guarded by 329 * the owning plane->mutex, i.e. we can never see concurrent 330 * manipulation of individual bits. But since the bitfield as a whole 331 * is updated using RMW, we need to use atomics in order to update 332 * the bits. 333 */ 334 BUILD_BUG_ON(INTEL_FRONTBUFFER_BITS_PER_PIPE * I915_MAX_PIPES > 335 BITS_PER_TYPE(atomic_t)); 336 BUILD_BUG_ON(INTEL_FRONTBUFFER_BITS_PER_PIPE * I915_MAX_PIPES > 32); 337 BUILD_BUG_ON(I915_MAX_PLANES > INTEL_FRONTBUFFER_BITS_PER_PIPE); 338 339 if (old) { 340 drm_WARN_ON(&intel_bo_to_i915(old->obj)->drm, 341 !(atomic_read(&old->bits) & frontbuffer_bits)); 342 atomic_andnot(frontbuffer_bits, &old->bits); 343 } 344 345 if (new) { 346 drm_WARN_ON(&intel_bo_to_i915(new->obj)->drm, 347 atomic_read(&new->bits) & frontbuffer_bits); 348 atomic_or(frontbuffer_bits, &new->bits); 349 } 350 } 351