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
24 /**
25 * DOC: Frame Buffer Compression (FBC)
26 *
27 * FBC tries to save memory bandwidth (and so power consumption) by
28 * compressing the amount of memory used by the display. It is total
29 * transparent to user space and completely handled in the kernel.
30 *
31 * The benefits of FBC are mostly visible with solid backgrounds and
32 * variation-less patterns. It comes from keeping the memory footprint small
33 * and having fewer memory pages opened and accessed for refreshing the display.
34 *
35 * i915 is responsible to reserve stolen memory for FBC and configure its
36 * offset on proper registers. The hardware takes care of all
37 * compress/decompress. However there are many known cases where we have to
38 * forcibly disable it to allow proper screen updates.
39 */
40
41 #include <drm/drm_fourcc.h>
42
43 #include "i915_drv.h"
44 #include "i915_trace.h"
45 #include "i915_vgpu.h"
46 #include "intel_display_types.h"
47 #include "intel_fbc.h"
48 #include "intel_frontbuffer.h"
49
50 /*
51 * For SKL+, the plane source size used by the hardware is based on the value we
52 * write to the PLANE_SIZE register. For BDW-, the hardware looks at the value
53 * we wrote to PIPESRC.
54 */
intel_fbc_get_plane_source_size(const struct intel_fbc_state_cache * cache,int * width,int * height)55 static void intel_fbc_get_plane_source_size(const struct intel_fbc_state_cache *cache,
56 int *width, int *height)
57 {
58 if (width)
59 *width = cache->plane.src_w;
60 if (height)
61 *height = cache->plane.src_h;
62 }
63
intel_fbc_calculate_cfb_size(struct drm_i915_private * dev_priv,const struct intel_fbc_state_cache * cache)64 static int intel_fbc_calculate_cfb_size(struct drm_i915_private *dev_priv,
65 const struct intel_fbc_state_cache *cache)
66 {
67 int lines;
68
69 intel_fbc_get_plane_source_size(cache, NULL, &lines);
70 if (IS_GEN(dev_priv, 7))
71 lines = min(lines, 2048);
72 else if (INTEL_GEN(dev_priv) >= 8)
73 lines = min(lines, 2560);
74
75 /* Hardware needs the full buffer stride, not just the active area. */
76 return lines * cache->fb.stride;
77 }
78
i8xx_fbc_deactivate(struct drm_i915_private * dev_priv)79 static void i8xx_fbc_deactivate(struct drm_i915_private *dev_priv)
80 {
81 u32 fbc_ctl;
82
83 /* Disable compression */
84 fbc_ctl = intel_de_read(dev_priv, FBC_CONTROL);
85 if ((fbc_ctl & FBC_CTL_EN) == 0)
86 return;
87
88 fbc_ctl &= ~FBC_CTL_EN;
89 intel_de_write(dev_priv, FBC_CONTROL, fbc_ctl);
90
91 /* Wait for compressing bit to clear */
92 if (intel_de_wait_for_clear(dev_priv, FBC_STATUS,
93 FBC_STAT_COMPRESSING, 10)) {
94 drm_dbg_kms(&dev_priv->drm, "FBC idle timed out\n");
95 return;
96 }
97 }
98
i8xx_fbc_activate(struct drm_i915_private * dev_priv)99 static void i8xx_fbc_activate(struct drm_i915_private *dev_priv)
100 {
101 struct intel_fbc_reg_params *params = &dev_priv->fbc.params;
102 int cfb_pitch;
103 int i;
104 u32 fbc_ctl;
105
106 /* Note: fbc.threshold == 1 for i8xx */
107 cfb_pitch = params->cfb_size / FBC_LL_SIZE;
108 if (params->fb.stride < cfb_pitch)
109 cfb_pitch = params->fb.stride;
110
111 /* FBC_CTL wants 32B or 64B units */
112 if (IS_GEN(dev_priv, 2))
113 cfb_pitch = (cfb_pitch / 32) - 1;
114 else
115 cfb_pitch = (cfb_pitch / 64) - 1;
116
117 /* Clear old tags */
118 for (i = 0; i < (FBC_LL_SIZE / 32) + 1; i++)
119 intel_de_write(dev_priv, FBC_TAG(i), 0);
120
121 if (IS_GEN(dev_priv, 4)) {
122 u32 fbc_ctl2;
123
124 /* Set it up... */
125 fbc_ctl2 = FBC_CTL_FENCE_DBL | FBC_CTL_IDLE_IMM;
126 fbc_ctl2 |= FBC_CTL_PLANE(params->crtc.i9xx_plane);
127 if (params->fence_id >= 0)
128 fbc_ctl2 |= FBC_CTL_CPU_FENCE;
129 intel_de_write(dev_priv, FBC_CONTROL2, fbc_ctl2);
130 intel_de_write(dev_priv, FBC_FENCE_OFF,
131 params->fence_y_offset);
132 }
133
134 /* enable it... */
135 fbc_ctl = FBC_CTL_INTERVAL(params->interval);
136 fbc_ctl |= FBC_CTL_EN | FBC_CTL_PERIODIC;
137 if (IS_I945GM(dev_priv))
138 fbc_ctl |= FBC_CTL_C3_IDLE; /* 945 needs special SR handling */
139 fbc_ctl |= FBC_CTL_STRIDE(cfb_pitch & 0xff);
140 if (params->fence_id >= 0)
141 fbc_ctl |= FBC_CTL_FENCENO(params->fence_id);
142 intel_de_write(dev_priv, FBC_CONTROL, fbc_ctl);
143 }
144
i8xx_fbc_is_active(struct drm_i915_private * dev_priv)145 static bool i8xx_fbc_is_active(struct drm_i915_private *dev_priv)
146 {
147 return intel_de_read(dev_priv, FBC_CONTROL) & FBC_CTL_EN;
148 }
149
g4x_fbc_activate(struct drm_i915_private * dev_priv)150 static void g4x_fbc_activate(struct drm_i915_private *dev_priv)
151 {
152 struct intel_fbc_reg_params *params = &dev_priv->fbc.params;
153 u32 dpfc_ctl;
154
155 dpfc_ctl = DPFC_CTL_PLANE(params->crtc.i9xx_plane) | DPFC_SR_EN;
156 if (params->fb.format->cpp[0] == 2)
157 dpfc_ctl |= DPFC_CTL_LIMIT_2X;
158 else
159 dpfc_ctl |= DPFC_CTL_LIMIT_1X;
160
161 if (params->fence_id >= 0) {
162 dpfc_ctl |= DPFC_CTL_FENCE_EN | params->fence_id;
163 intel_de_write(dev_priv, DPFC_FENCE_YOFF,
164 params->fence_y_offset);
165 } else {
166 intel_de_write(dev_priv, DPFC_FENCE_YOFF, 0);
167 }
168
169 /* enable it... */
170 intel_de_write(dev_priv, DPFC_CONTROL, dpfc_ctl | DPFC_CTL_EN);
171 }
172
g4x_fbc_deactivate(struct drm_i915_private * dev_priv)173 static void g4x_fbc_deactivate(struct drm_i915_private *dev_priv)
174 {
175 u32 dpfc_ctl;
176
177 /* Disable compression */
178 dpfc_ctl = intel_de_read(dev_priv, DPFC_CONTROL);
179 if (dpfc_ctl & DPFC_CTL_EN) {
180 dpfc_ctl &= ~DPFC_CTL_EN;
181 intel_de_write(dev_priv, DPFC_CONTROL, dpfc_ctl);
182 }
183 }
184
g4x_fbc_is_active(struct drm_i915_private * dev_priv)185 static bool g4x_fbc_is_active(struct drm_i915_private *dev_priv)
186 {
187 return intel_de_read(dev_priv, DPFC_CONTROL) & DPFC_CTL_EN;
188 }
189
i8xx_fbc_recompress(struct drm_i915_private * dev_priv)190 static void i8xx_fbc_recompress(struct drm_i915_private *dev_priv)
191 {
192 struct intel_fbc_reg_params *params = &dev_priv->fbc.params;
193 enum i9xx_plane_id i9xx_plane = params->crtc.i9xx_plane;
194
195 spin_lock_irq(&dev_priv->uncore.lock);
196 intel_de_write_fw(dev_priv, DSPADDR(i9xx_plane),
197 intel_de_read_fw(dev_priv, DSPADDR(i9xx_plane)));
198 spin_unlock_irq(&dev_priv->uncore.lock);
199 }
200
i965_fbc_recompress(struct drm_i915_private * dev_priv)201 static void i965_fbc_recompress(struct drm_i915_private *dev_priv)
202 {
203 struct intel_fbc_reg_params *params = &dev_priv->fbc.params;
204 enum i9xx_plane_id i9xx_plane = params->crtc.i9xx_plane;
205
206 spin_lock_irq(&dev_priv->uncore.lock);
207 intel_de_write_fw(dev_priv, DSPSURF(i9xx_plane),
208 intel_de_read_fw(dev_priv, DSPSURF(i9xx_plane)));
209 spin_unlock_irq(&dev_priv->uncore.lock);
210 }
211
212 /* This function forces a CFB recompression through the nuke operation. */
snb_fbc_recompress(struct drm_i915_private * dev_priv)213 static void snb_fbc_recompress(struct drm_i915_private *dev_priv)
214 {
215 struct intel_fbc *fbc = &dev_priv->fbc;
216
217 trace_intel_fbc_nuke(fbc->crtc);
218
219 intel_de_write(dev_priv, MSG_FBC_REND_STATE, FBC_REND_NUKE);
220 intel_de_posting_read(dev_priv, MSG_FBC_REND_STATE);
221 }
222
intel_fbc_recompress(struct drm_i915_private * dev_priv)223 static void intel_fbc_recompress(struct drm_i915_private *dev_priv)
224 {
225 if (INTEL_GEN(dev_priv) >= 6)
226 snb_fbc_recompress(dev_priv);
227 else if (INTEL_GEN(dev_priv) >= 4)
228 i965_fbc_recompress(dev_priv);
229 else
230 i8xx_fbc_recompress(dev_priv);
231 }
232
ilk_fbc_activate(struct drm_i915_private * dev_priv)233 static void ilk_fbc_activate(struct drm_i915_private *dev_priv)
234 {
235 struct intel_fbc_reg_params *params = &dev_priv->fbc.params;
236 u32 dpfc_ctl;
237 int threshold = dev_priv->fbc.threshold;
238
239 dpfc_ctl = DPFC_CTL_PLANE(params->crtc.i9xx_plane);
240 if (params->fb.format->cpp[0] == 2)
241 threshold++;
242
243 switch (threshold) {
244 case 4:
245 case 3:
246 dpfc_ctl |= DPFC_CTL_LIMIT_4X;
247 break;
248 case 2:
249 dpfc_ctl |= DPFC_CTL_LIMIT_2X;
250 break;
251 case 1:
252 dpfc_ctl |= DPFC_CTL_LIMIT_1X;
253 break;
254 }
255
256 if (params->fence_id >= 0) {
257 dpfc_ctl |= DPFC_CTL_FENCE_EN;
258 if (IS_GEN(dev_priv, 5))
259 dpfc_ctl |= params->fence_id;
260 if (IS_GEN(dev_priv, 6)) {
261 intel_de_write(dev_priv, SNB_DPFC_CTL_SA,
262 SNB_CPU_FENCE_ENABLE | params->fence_id);
263 intel_de_write(dev_priv, DPFC_CPU_FENCE_OFFSET,
264 params->fence_y_offset);
265 }
266 } else {
267 if (IS_GEN(dev_priv, 6)) {
268 intel_de_write(dev_priv, SNB_DPFC_CTL_SA, 0);
269 intel_de_write(dev_priv, DPFC_CPU_FENCE_OFFSET, 0);
270 }
271 }
272
273 intel_de_write(dev_priv, ILK_DPFC_FENCE_YOFF,
274 params->fence_y_offset);
275 /* enable it... */
276 intel_de_write(dev_priv, ILK_DPFC_CONTROL, dpfc_ctl | DPFC_CTL_EN);
277
278 intel_fbc_recompress(dev_priv);
279 }
280
ilk_fbc_deactivate(struct drm_i915_private * dev_priv)281 static void ilk_fbc_deactivate(struct drm_i915_private *dev_priv)
282 {
283 u32 dpfc_ctl;
284
285 /* Disable compression */
286 dpfc_ctl = intel_de_read(dev_priv, ILK_DPFC_CONTROL);
287 if (dpfc_ctl & DPFC_CTL_EN) {
288 dpfc_ctl &= ~DPFC_CTL_EN;
289 intel_de_write(dev_priv, ILK_DPFC_CONTROL, dpfc_ctl);
290 }
291 }
292
ilk_fbc_is_active(struct drm_i915_private * dev_priv)293 static bool ilk_fbc_is_active(struct drm_i915_private *dev_priv)
294 {
295 return intel_de_read(dev_priv, ILK_DPFC_CONTROL) & DPFC_CTL_EN;
296 }
297
gen7_fbc_activate(struct drm_i915_private * dev_priv)298 static void gen7_fbc_activate(struct drm_i915_private *dev_priv)
299 {
300 struct intel_fbc_reg_params *params = &dev_priv->fbc.params;
301 u32 dpfc_ctl;
302 int threshold = dev_priv->fbc.threshold;
303
304 /* Display WA #0529: skl, kbl, bxt. */
305 if (IS_GEN9_BC(dev_priv) || IS_BROXTON(dev_priv)) {
306 u32 val = intel_de_read(dev_priv, CHICKEN_MISC_4);
307
308 val &= ~(FBC_STRIDE_OVERRIDE | FBC_STRIDE_MASK);
309
310 if (params->gen9_wa_cfb_stride)
311 val |= FBC_STRIDE_OVERRIDE | params->gen9_wa_cfb_stride;
312
313 intel_de_write(dev_priv, CHICKEN_MISC_4, val);
314 }
315
316 dpfc_ctl = 0;
317 if (IS_IVYBRIDGE(dev_priv))
318 dpfc_ctl |= IVB_DPFC_CTL_PLANE(params->crtc.i9xx_plane);
319
320 if (params->fb.format->cpp[0] == 2)
321 threshold++;
322
323 switch (threshold) {
324 case 4:
325 case 3:
326 dpfc_ctl |= DPFC_CTL_LIMIT_4X;
327 break;
328 case 2:
329 dpfc_ctl |= DPFC_CTL_LIMIT_2X;
330 break;
331 case 1:
332 dpfc_ctl |= DPFC_CTL_LIMIT_1X;
333 break;
334 }
335
336 if (params->fence_id >= 0) {
337 dpfc_ctl |= IVB_DPFC_CTL_FENCE_EN;
338 intel_de_write(dev_priv, SNB_DPFC_CTL_SA,
339 SNB_CPU_FENCE_ENABLE | params->fence_id);
340 intel_de_write(dev_priv, DPFC_CPU_FENCE_OFFSET,
341 params->fence_y_offset);
342 } else if (dev_priv->ggtt.num_fences) {
343 intel_de_write(dev_priv, SNB_DPFC_CTL_SA, 0);
344 intel_de_write(dev_priv, DPFC_CPU_FENCE_OFFSET, 0);
345 }
346
347 if (dev_priv->fbc.false_color)
348 dpfc_ctl |= FBC_CTL_FALSE_COLOR;
349
350 intel_de_write(dev_priv, ILK_DPFC_CONTROL, dpfc_ctl | DPFC_CTL_EN);
351
352 intel_fbc_recompress(dev_priv);
353 }
354
intel_fbc_hw_is_active(struct drm_i915_private * dev_priv)355 static bool intel_fbc_hw_is_active(struct drm_i915_private *dev_priv)
356 {
357 if (INTEL_GEN(dev_priv) >= 5)
358 return ilk_fbc_is_active(dev_priv);
359 else if (IS_GM45(dev_priv))
360 return g4x_fbc_is_active(dev_priv);
361 else
362 return i8xx_fbc_is_active(dev_priv);
363 }
364
intel_fbc_hw_activate(struct drm_i915_private * dev_priv)365 static void intel_fbc_hw_activate(struct drm_i915_private *dev_priv)
366 {
367 struct intel_fbc *fbc = &dev_priv->fbc;
368
369 trace_intel_fbc_activate(fbc->crtc);
370
371 fbc->active = true;
372 fbc->activated = true;
373
374 if (INTEL_GEN(dev_priv) >= 7)
375 gen7_fbc_activate(dev_priv);
376 else if (INTEL_GEN(dev_priv) >= 5)
377 ilk_fbc_activate(dev_priv);
378 else if (IS_GM45(dev_priv))
379 g4x_fbc_activate(dev_priv);
380 else
381 i8xx_fbc_activate(dev_priv);
382 }
383
intel_fbc_hw_deactivate(struct drm_i915_private * dev_priv)384 static void intel_fbc_hw_deactivate(struct drm_i915_private *dev_priv)
385 {
386 struct intel_fbc *fbc = &dev_priv->fbc;
387
388 trace_intel_fbc_deactivate(fbc->crtc);
389
390 fbc->active = false;
391
392 if (INTEL_GEN(dev_priv) >= 5)
393 ilk_fbc_deactivate(dev_priv);
394 else if (IS_GM45(dev_priv))
395 g4x_fbc_deactivate(dev_priv);
396 else
397 i8xx_fbc_deactivate(dev_priv);
398 }
399
400 /**
401 * intel_fbc_is_active - Is FBC active?
402 * @dev_priv: i915 device instance
403 *
404 * This function is used to verify the current state of FBC.
405 *
406 * FIXME: This should be tracked in the plane config eventually
407 * instead of queried at runtime for most callers.
408 */
intel_fbc_is_active(struct drm_i915_private * dev_priv)409 bool intel_fbc_is_active(struct drm_i915_private *dev_priv)
410 {
411 return dev_priv->fbc.active;
412 }
413
intel_fbc_deactivate(struct drm_i915_private * dev_priv,const char * reason)414 static void intel_fbc_deactivate(struct drm_i915_private *dev_priv,
415 const char *reason)
416 {
417 struct intel_fbc *fbc = &dev_priv->fbc;
418
419 drm_WARN_ON(&dev_priv->drm, !mutex_is_locked(&fbc->lock));
420
421 if (fbc->active)
422 intel_fbc_hw_deactivate(dev_priv);
423
424 fbc->no_fbc_reason = reason;
425 }
426
intel_fbc_cfb_base_max(struct drm_i915_private * i915)427 static u64 intel_fbc_cfb_base_max(struct drm_i915_private *i915)
428 {
429 if (INTEL_GEN(i915) >= 5 || IS_G4X(i915))
430 return BIT_ULL(28);
431 else
432 return BIT_ULL(32);
433 }
434
find_compression_threshold(struct drm_i915_private * dev_priv,struct drm_mm_node * node,unsigned int size,unsigned int fb_cpp)435 static int find_compression_threshold(struct drm_i915_private *dev_priv,
436 struct drm_mm_node *node,
437 unsigned int size,
438 unsigned int fb_cpp)
439 {
440 int compression_threshold = 1;
441 int ret;
442 u64 end;
443
444 /* The FBC hardware for BDW/SKL doesn't have access to the stolen
445 * reserved range size, so it always assumes the maximum (8mb) is used.
446 * If we enable FBC using a CFB on that memory range we'll get FIFO
447 * underruns, even if that range is not reserved by the BIOS. */
448 if (IS_BROADWELL(dev_priv) || IS_GEN9_BC(dev_priv))
449 end = resource_size(&dev_priv->dsm) - 8 * 1024 * 1024;
450 else
451 end = U64_MAX;
452
453 end = min(end, intel_fbc_cfb_base_max(dev_priv));
454
455 /* HACK: This code depends on what we will do in *_enable_fbc. If that
456 * code changes, this code needs to change as well.
457 *
458 * The enable_fbc code will attempt to use one of our 2 compression
459 * thresholds, therefore, in that case, we only have 1 resort.
460 */
461
462 /* Try to over-allocate to reduce reallocations and fragmentation. */
463 ret = i915_gem_stolen_insert_node_in_range(dev_priv, node, size <<= 1,
464 4096, 0, end);
465 if (ret == 0)
466 return compression_threshold;
467
468 again:
469 /* HW's ability to limit the CFB is 1:4 */
470 if (compression_threshold > 4 ||
471 (fb_cpp == 2 && compression_threshold == 2))
472 return 0;
473
474 ret = i915_gem_stolen_insert_node_in_range(dev_priv, node, size >>= 1,
475 4096, 0, end);
476 if (ret && INTEL_GEN(dev_priv) <= 4) {
477 return 0;
478 } else if (ret) {
479 compression_threshold <<= 1;
480 goto again;
481 } else {
482 return compression_threshold;
483 }
484 }
485
intel_fbc_alloc_cfb(struct drm_i915_private * dev_priv,unsigned int size,unsigned int fb_cpp)486 static int intel_fbc_alloc_cfb(struct drm_i915_private *dev_priv,
487 unsigned int size, unsigned int fb_cpp)
488 {
489 struct intel_fbc *fbc = &dev_priv->fbc;
490 struct drm_mm_node *compressed_llb;
491 int ret;
492
493 drm_WARN_ON(&dev_priv->drm,
494 drm_mm_node_allocated(&fbc->compressed_fb));
495
496 ret = find_compression_threshold(dev_priv, &fbc->compressed_fb,
497 size, fb_cpp);
498 if (!ret)
499 goto err_llb;
500 else if (ret > 1) {
501 drm_info_once(&dev_priv->drm,
502 "Reducing the compressed framebuffer size. This may lead to less power savings than a non-reduced-size. Try to increase stolen memory size if available in BIOS.\n");
503 }
504
505 fbc->threshold = ret;
506
507 if (INTEL_GEN(dev_priv) >= 5)
508 intel_de_write(dev_priv, ILK_DPFC_CB_BASE,
509 fbc->compressed_fb.start);
510 else if (IS_GM45(dev_priv)) {
511 intel_de_write(dev_priv, DPFC_CB_BASE,
512 fbc->compressed_fb.start);
513 } else {
514 compressed_llb = kzalloc(sizeof(*compressed_llb), GFP_KERNEL);
515 if (!compressed_llb)
516 goto err_fb;
517
518 ret = i915_gem_stolen_insert_node(dev_priv, compressed_llb,
519 4096, 4096);
520 if (ret)
521 goto err_fb;
522
523 fbc->compressed_llb = compressed_llb;
524
525 GEM_BUG_ON(range_overflows_end_t(u64, dev_priv->dsm.start,
526 fbc->compressed_fb.start,
527 U32_MAX));
528 GEM_BUG_ON(range_overflows_end_t(u64, dev_priv->dsm.start,
529 fbc->compressed_llb->start,
530 U32_MAX));
531 intel_de_write(dev_priv, FBC_CFB_BASE,
532 dev_priv->dsm.start + fbc->compressed_fb.start);
533 intel_de_write(dev_priv, FBC_LL_BASE,
534 dev_priv->dsm.start + compressed_llb->start);
535 }
536
537 drm_dbg_kms(&dev_priv->drm,
538 "reserved %llu bytes of contiguous stolen space for FBC, threshold: %d\n",
539 fbc->compressed_fb.size, fbc->threshold);
540
541 return 0;
542
543 err_fb:
544 kfree(compressed_llb);
545 i915_gem_stolen_remove_node(dev_priv, &fbc->compressed_fb);
546 err_llb:
547 if (drm_mm_initialized(&dev_priv->mm.stolen))
548 drm_info_once(&dev_priv->drm, "not enough stolen space for compressed buffer (need %d more bytes), disabling. Hint: you may be able to increase stolen memory size in the BIOS to avoid this.\n", size);
549 return -ENOSPC;
550 }
551
__intel_fbc_cleanup_cfb(struct drm_i915_private * dev_priv)552 static void __intel_fbc_cleanup_cfb(struct drm_i915_private *dev_priv)
553 {
554 struct intel_fbc *fbc = &dev_priv->fbc;
555
556 if (WARN_ON(intel_fbc_hw_is_active(dev_priv)))
557 return;
558
559 if (!drm_mm_node_allocated(&fbc->compressed_fb))
560 return;
561
562 if (fbc->compressed_llb) {
563 i915_gem_stolen_remove_node(dev_priv, fbc->compressed_llb);
564 kfree(fbc->compressed_llb);
565 }
566
567 i915_gem_stolen_remove_node(dev_priv, &fbc->compressed_fb);
568 }
569
intel_fbc_cleanup_cfb(struct drm_i915_private * dev_priv)570 void intel_fbc_cleanup_cfb(struct drm_i915_private *dev_priv)
571 {
572 struct intel_fbc *fbc = &dev_priv->fbc;
573
574 if (!HAS_FBC(dev_priv))
575 return;
576
577 mutex_lock(&fbc->lock);
578 __intel_fbc_cleanup_cfb(dev_priv);
579 mutex_unlock(&fbc->lock);
580 }
581
stride_is_valid(struct drm_i915_private * dev_priv,u64 modifier,unsigned int stride)582 static bool stride_is_valid(struct drm_i915_private *dev_priv,
583 u64 modifier, unsigned int stride)
584 {
585 /* This should have been caught earlier. */
586 if (drm_WARN_ON_ONCE(&dev_priv->drm, (stride & (64 - 1)) != 0))
587 return false;
588
589 /* Below are the additional FBC restrictions. */
590 if (stride < 512)
591 return false;
592
593 if (IS_GEN(dev_priv, 2) || IS_GEN(dev_priv, 3))
594 return stride == 4096 || stride == 8192;
595
596 if (IS_GEN(dev_priv, 4) && !IS_G4X(dev_priv) && stride < 2048)
597 return false;
598
599 /* Display WA #1105: skl,bxt,kbl,cfl,glk */
600 if (IS_GEN(dev_priv, 9) &&
601 modifier == DRM_FORMAT_MOD_LINEAR && stride & 511)
602 return false;
603
604 if (stride > 16384)
605 return false;
606
607 return true;
608 }
609
pixel_format_is_valid(struct drm_i915_private * dev_priv,u32 pixel_format)610 static bool pixel_format_is_valid(struct drm_i915_private *dev_priv,
611 u32 pixel_format)
612 {
613 switch (pixel_format) {
614 case DRM_FORMAT_XRGB8888:
615 case DRM_FORMAT_XBGR8888:
616 return true;
617 case DRM_FORMAT_XRGB1555:
618 case DRM_FORMAT_RGB565:
619 /* 16bpp not supported on gen2 */
620 if (IS_GEN(dev_priv, 2))
621 return false;
622 /* WaFbcOnly1to1Ratio:ctg */
623 if (IS_G4X(dev_priv))
624 return false;
625 return true;
626 default:
627 return false;
628 }
629 }
630
rotation_is_valid(struct drm_i915_private * dev_priv,u32 pixel_format,unsigned int rotation)631 static bool rotation_is_valid(struct drm_i915_private *dev_priv,
632 u32 pixel_format, unsigned int rotation)
633 {
634 if (INTEL_GEN(dev_priv) >= 9 && pixel_format == DRM_FORMAT_RGB565 &&
635 drm_rotation_90_or_270(rotation))
636 return false;
637 else if (INTEL_GEN(dev_priv) <= 4 && !IS_G4X(dev_priv) &&
638 rotation != DRM_MODE_ROTATE_0)
639 return false;
640
641 return true;
642 }
643
644 /*
645 * For some reason, the hardware tracking starts looking at whatever we
646 * programmed as the display plane base address register. It does not look at
647 * the X and Y offset registers. That's why we include the src x/y offsets
648 * instead of just looking at the plane size.
649 */
intel_fbc_hw_tracking_covers_screen(struct intel_crtc * crtc)650 static bool intel_fbc_hw_tracking_covers_screen(struct intel_crtc *crtc)
651 {
652 struct drm_i915_private *dev_priv = to_i915(crtc->base.dev);
653 struct intel_fbc *fbc = &dev_priv->fbc;
654 unsigned int effective_w, effective_h, max_w, max_h;
655
656 if (INTEL_GEN(dev_priv) >= 10 || IS_GEMINILAKE(dev_priv)) {
657 max_w = 5120;
658 max_h = 4096;
659 } else if (INTEL_GEN(dev_priv) >= 8 || IS_HASWELL(dev_priv)) {
660 max_w = 4096;
661 max_h = 4096;
662 } else if (IS_G4X(dev_priv) || INTEL_GEN(dev_priv) >= 5) {
663 max_w = 4096;
664 max_h = 2048;
665 } else {
666 max_w = 2048;
667 max_h = 1536;
668 }
669
670 intel_fbc_get_plane_source_size(&fbc->state_cache, &effective_w,
671 &effective_h);
672 effective_w += fbc->state_cache.plane.adjusted_x;
673 effective_h += fbc->state_cache.plane.adjusted_y;
674
675 return effective_w <= max_w && effective_h <= max_h;
676 }
677
tiling_is_valid(struct drm_i915_private * dev_priv,uint64_t modifier)678 static bool tiling_is_valid(struct drm_i915_private *dev_priv,
679 uint64_t modifier)
680 {
681 switch (modifier) {
682 case DRM_FORMAT_MOD_LINEAR:
683 if (INTEL_GEN(dev_priv) >= 9)
684 return true;
685 return false;
686 case I915_FORMAT_MOD_X_TILED:
687 case I915_FORMAT_MOD_Y_TILED:
688 return true;
689 default:
690 return false;
691 }
692 }
693
intel_fbc_update_state_cache(struct intel_crtc * crtc,const struct intel_crtc_state * crtc_state,const struct intel_plane_state * plane_state)694 static void intel_fbc_update_state_cache(struct intel_crtc *crtc,
695 const struct intel_crtc_state *crtc_state,
696 const struct intel_plane_state *plane_state)
697 {
698 struct drm_i915_private *dev_priv = to_i915(crtc->base.dev);
699 struct intel_fbc *fbc = &dev_priv->fbc;
700 struct intel_fbc_state_cache *cache = &fbc->state_cache;
701 struct drm_framebuffer *fb = plane_state->hw.fb;
702
703 cache->plane.visible = plane_state->uapi.visible;
704 if (!cache->plane.visible)
705 return;
706
707 cache->crtc.mode_flags = crtc_state->hw.adjusted_mode.flags;
708 if (IS_HASWELL(dev_priv) || IS_BROADWELL(dev_priv))
709 cache->crtc.hsw_bdw_pixel_rate = crtc_state->pixel_rate;
710
711 cache->plane.rotation = plane_state->hw.rotation;
712 /*
713 * Src coordinates are already rotated by 270 degrees for
714 * the 90/270 degree plane rotation cases (to match the
715 * GTT mapping), hence no need to account for rotation here.
716 */
717 cache->plane.src_w = drm_rect_width(&plane_state->uapi.src) >> 16;
718 cache->plane.src_h = drm_rect_height(&plane_state->uapi.src) >> 16;
719 cache->plane.adjusted_x = plane_state->color_plane[0].x;
720 cache->plane.adjusted_y = plane_state->color_plane[0].y;
721
722 cache->plane.pixel_blend_mode = plane_state->hw.pixel_blend_mode;
723
724 cache->fb.format = fb->format;
725 cache->fb.modifier = fb->modifier;
726
727 /* FIXME is this correct? */
728 cache->fb.stride = plane_state->color_plane[0].stride;
729 if (drm_rotation_90_or_270(plane_state->hw.rotation))
730 cache->fb.stride *= fb->format->cpp[0];
731
732 /* FBC1 compression interval: arbitrary choice of 1 second */
733 cache->interval = drm_mode_vrefresh(&crtc_state->hw.adjusted_mode);
734
735 cache->fence_y_offset = intel_plane_fence_y_offset(plane_state);
736
737 drm_WARN_ON(&dev_priv->drm, plane_state->flags & PLANE_HAS_FENCE &&
738 !plane_state->vma->fence);
739
740 if (plane_state->flags & PLANE_HAS_FENCE &&
741 plane_state->vma->fence)
742 cache->fence_id = plane_state->vma->fence->id;
743 else
744 cache->fence_id = -1;
745 }
746
intel_fbc_cfb_size_changed(struct drm_i915_private * dev_priv)747 static bool intel_fbc_cfb_size_changed(struct drm_i915_private *dev_priv)
748 {
749 struct intel_fbc *fbc = &dev_priv->fbc;
750
751 return intel_fbc_calculate_cfb_size(dev_priv, &fbc->state_cache) >
752 fbc->compressed_fb.size * fbc->threshold;
753 }
754
intel_fbc_gen9_wa_cfb_stride(struct drm_i915_private * dev_priv)755 static u16 intel_fbc_gen9_wa_cfb_stride(struct drm_i915_private *dev_priv)
756 {
757 struct intel_fbc *fbc = &dev_priv->fbc;
758 struct intel_fbc_state_cache *cache = &fbc->state_cache;
759
760 if ((IS_GEN9_BC(dev_priv) || IS_BROXTON(dev_priv)) &&
761 cache->fb.modifier != I915_FORMAT_MOD_X_TILED)
762 return DIV_ROUND_UP(cache->plane.src_w, 32 * fbc->threshold) * 8;
763 else
764 return 0;
765 }
766
intel_fbc_gen9_wa_cfb_stride_changed(struct drm_i915_private * dev_priv)767 static bool intel_fbc_gen9_wa_cfb_stride_changed(struct drm_i915_private *dev_priv)
768 {
769 struct intel_fbc *fbc = &dev_priv->fbc;
770
771 return fbc->params.gen9_wa_cfb_stride != intel_fbc_gen9_wa_cfb_stride(dev_priv);
772 }
773
intel_fbc_can_enable(struct drm_i915_private * dev_priv)774 static bool intel_fbc_can_enable(struct drm_i915_private *dev_priv)
775 {
776 struct intel_fbc *fbc = &dev_priv->fbc;
777
778 if (intel_vgpu_active(dev_priv)) {
779 fbc->no_fbc_reason = "VGPU is active";
780 return false;
781 }
782
783 if (!dev_priv->params.enable_fbc) {
784 fbc->no_fbc_reason = "disabled per module param or by default";
785 return false;
786 }
787
788 if (fbc->underrun_detected) {
789 fbc->no_fbc_reason = "underrun detected";
790 return false;
791 }
792
793 return true;
794 }
795
intel_fbc_can_activate(struct intel_crtc * crtc)796 static bool intel_fbc_can_activate(struct intel_crtc *crtc)
797 {
798 struct drm_i915_private *dev_priv = to_i915(crtc->base.dev);
799 struct intel_fbc *fbc = &dev_priv->fbc;
800 struct intel_fbc_state_cache *cache = &fbc->state_cache;
801
802 if (!intel_fbc_can_enable(dev_priv))
803 return false;
804
805 if (!cache->plane.visible) {
806 fbc->no_fbc_reason = "primary plane not visible";
807 return false;
808 }
809
810 /* We don't need to use a state cache here since this information is
811 * global for all CRTC.
812 */
813 if (fbc->underrun_detected) {
814 fbc->no_fbc_reason = "underrun detected";
815 return false;
816 }
817
818 if (cache->crtc.mode_flags & DRM_MODE_FLAG_INTERLACE) {
819 fbc->no_fbc_reason = "incompatible mode";
820 return false;
821 }
822
823 if (!intel_fbc_hw_tracking_covers_screen(crtc)) {
824 fbc->no_fbc_reason = "mode too large for compression";
825 return false;
826 }
827
828 /* The use of a CPU fence is one of two ways to detect writes by the
829 * CPU to the scanout and trigger updates to the FBC.
830 *
831 * The other method is by software tracking (see
832 * intel_fbc_invalidate/flush()), it will manually notify FBC and nuke
833 * the current compressed buffer and recompress it.
834 *
835 * Note that is possible for a tiled surface to be unmappable (and
836 * so have no fence associated with it) due to aperture constraints
837 * at the time of pinning.
838 *
839 * FIXME with 90/270 degree rotation we should use the fence on
840 * the normal GTT view (the rotated view doesn't even have a
841 * fence). Would need changes to the FBC fence Y offset as well.
842 * For now this will effectively disable FBC with 90/270 degree
843 * rotation.
844 */
845 if (INTEL_GEN(dev_priv) < 9 && cache->fence_id < 0) {
846 fbc->no_fbc_reason = "framebuffer not tiled or fenced";
847 return false;
848 }
849
850 if (!pixel_format_is_valid(dev_priv, cache->fb.format->format)) {
851 fbc->no_fbc_reason = "pixel format is invalid";
852 return false;
853 }
854
855 if (!rotation_is_valid(dev_priv, cache->fb.format->format,
856 cache->plane.rotation)) {
857 fbc->no_fbc_reason = "rotation unsupported";
858 return false;
859 }
860
861 if (!tiling_is_valid(dev_priv, cache->fb.modifier)) {
862 fbc->no_fbc_reason = "tiling unsupported";
863 return false;
864 }
865
866 if (!stride_is_valid(dev_priv, cache->fb.modifier, cache->fb.stride)) {
867 fbc->no_fbc_reason = "framebuffer stride not supported";
868 return false;
869 }
870
871 if (cache->plane.pixel_blend_mode != DRM_MODE_BLEND_PIXEL_NONE &&
872 cache->fb.format->has_alpha) {
873 fbc->no_fbc_reason = "per-pixel alpha blending is incompatible with FBC";
874 return false;
875 }
876
877 /* WaFbcExceedCdClockThreshold:hsw,bdw */
878 if ((IS_HASWELL(dev_priv) || IS_BROADWELL(dev_priv)) &&
879 cache->crtc.hsw_bdw_pixel_rate >= dev_priv->cdclk.hw.cdclk * 95 / 100) {
880 fbc->no_fbc_reason = "pixel rate is too big";
881 return false;
882 }
883
884 /* It is possible for the required CFB size change without a
885 * crtc->disable + crtc->enable since it is possible to change the
886 * stride without triggering a full modeset. Since we try to
887 * over-allocate the CFB, there's a chance we may keep FBC enabled even
888 * if this happens, but if we exceed the current CFB size we'll have to
889 * disable FBC. Notice that it would be possible to disable FBC, wait
890 * for a frame, free the stolen node, then try to reenable FBC in case
891 * we didn't get any invalidate/deactivate calls, but this would require
892 * a lot of tracking just for a specific case. If we conclude it's an
893 * important case, we can implement it later. */
894 if (intel_fbc_cfb_size_changed(dev_priv)) {
895 fbc->no_fbc_reason = "CFB requirements changed";
896 return false;
897 }
898
899 /*
900 * Work around a problem on GEN9+ HW, where enabling FBC on a plane
901 * having a Y offset that isn't divisible by 4 causes FIFO underrun
902 * and screen flicker.
903 */
904 if (INTEL_GEN(dev_priv) >= 9 &&
905 (fbc->state_cache.plane.adjusted_y & 3)) {
906 fbc->no_fbc_reason = "plane Y offset is misaligned";
907 return false;
908 }
909
910 return true;
911 }
912
intel_fbc_get_reg_params(struct intel_crtc * crtc,struct intel_fbc_reg_params * params)913 static void intel_fbc_get_reg_params(struct intel_crtc *crtc,
914 struct intel_fbc_reg_params *params)
915 {
916 struct drm_i915_private *dev_priv = to_i915(crtc->base.dev);
917 struct intel_fbc *fbc = &dev_priv->fbc;
918 struct intel_fbc_state_cache *cache = &fbc->state_cache;
919
920 /* Since all our fields are integer types, use memset here so the
921 * comparison function can rely on memcmp because the padding will be
922 * zero. */
923 memset(params, 0, sizeof(*params));
924
925 params->fence_id = cache->fence_id;
926 params->fence_y_offset = cache->fence_y_offset;
927
928 params->interval = cache->interval;
929
930 params->crtc.pipe = crtc->pipe;
931 params->crtc.i9xx_plane = to_intel_plane(crtc->base.primary)->i9xx_plane;
932
933 params->fb.format = cache->fb.format;
934 params->fb.modifier = cache->fb.modifier;
935 params->fb.stride = cache->fb.stride;
936
937 params->cfb_size = intel_fbc_calculate_cfb_size(dev_priv, cache);
938
939 params->gen9_wa_cfb_stride = cache->gen9_wa_cfb_stride;
940
941 params->plane_visible = cache->plane.visible;
942 }
943
intel_fbc_can_flip_nuke(const struct intel_crtc_state * crtc_state)944 static bool intel_fbc_can_flip_nuke(const struct intel_crtc_state *crtc_state)
945 {
946 struct intel_crtc *crtc = to_intel_crtc(crtc_state->uapi.crtc);
947 struct drm_i915_private *dev_priv = to_i915(crtc->base.dev);
948 const struct intel_fbc *fbc = &dev_priv->fbc;
949 const struct intel_fbc_state_cache *cache = &fbc->state_cache;
950 const struct intel_fbc_reg_params *params = &fbc->params;
951
952 if (drm_atomic_crtc_needs_modeset(&crtc_state->uapi))
953 return false;
954
955 if (!params->plane_visible)
956 return false;
957
958 if (!intel_fbc_can_activate(crtc))
959 return false;
960
961 if (params->fb.format != cache->fb.format)
962 return false;
963
964 if (params->fb.modifier != cache->fb.modifier)
965 return false;
966
967 if (params->fb.stride != cache->fb.stride)
968 return false;
969
970 if (params->cfb_size != intel_fbc_calculate_cfb_size(dev_priv, cache))
971 return false;
972
973 if (params->gen9_wa_cfb_stride != cache->gen9_wa_cfb_stride)
974 return false;
975
976 return true;
977 }
978
intel_fbc_pre_update(struct intel_atomic_state * state,struct intel_crtc * crtc)979 bool intel_fbc_pre_update(struct intel_atomic_state *state,
980 struct intel_crtc *crtc)
981 {
982 struct intel_plane *plane = to_intel_plane(crtc->base.primary);
983 const struct intel_crtc_state *crtc_state =
984 intel_atomic_get_new_crtc_state(state, crtc);
985 const struct intel_plane_state *plane_state =
986 intel_atomic_get_new_plane_state(state, plane);
987 struct drm_i915_private *dev_priv = to_i915(crtc->base.dev);
988 struct intel_fbc *fbc = &dev_priv->fbc;
989 const char *reason = "update pending";
990 bool need_vblank_wait = false;
991
992 if (!plane->has_fbc || !plane_state)
993 return need_vblank_wait;
994
995 mutex_lock(&fbc->lock);
996
997 if (fbc->crtc != crtc)
998 goto unlock;
999
1000 intel_fbc_update_state_cache(crtc, crtc_state, plane_state);
1001 fbc->flip_pending = true;
1002
1003 if (!intel_fbc_can_flip_nuke(crtc_state)) {
1004 intel_fbc_deactivate(dev_priv, reason);
1005
1006 /*
1007 * Display WA #1198: glk+
1008 * Need an extra vblank wait between FBC disable and most plane
1009 * updates. Bspec says this is only needed for plane disable, but
1010 * that is not true. Touching most plane registers will cause the
1011 * corruption to appear. Also SKL/derivatives do not seem to be
1012 * affected.
1013 *
1014 * TODO: could optimize this a bit by sampling the frame
1015 * counter when we disable FBC (if it was already done earlier)
1016 * and skipping the extra vblank wait before the plane update
1017 * if at least one frame has already passed.
1018 */
1019 if (fbc->activated &&
1020 (INTEL_GEN(dev_priv) >= 10 || IS_GEMINILAKE(dev_priv)))
1021 need_vblank_wait = true;
1022 fbc->activated = false;
1023 }
1024 unlock:
1025 mutex_unlock(&fbc->lock);
1026
1027 return need_vblank_wait;
1028 }
1029
1030 /**
1031 * __intel_fbc_disable - disable FBC
1032 * @dev_priv: i915 device instance
1033 *
1034 * This is the low level function that actually disables FBC. Callers should
1035 * grab the FBC lock.
1036 */
__intel_fbc_disable(struct drm_i915_private * dev_priv)1037 static void __intel_fbc_disable(struct drm_i915_private *dev_priv)
1038 {
1039 struct intel_fbc *fbc = &dev_priv->fbc;
1040 struct intel_crtc *crtc = fbc->crtc;
1041
1042 drm_WARN_ON(&dev_priv->drm, !mutex_is_locked(&fbc->lock));
1043 drm_WARN_ON(&dev_priv->drm, !fbc->crtc);
1044 drm_WARN_ON(&dev_priv->drm, fbc->active);
1045
1046 drm_dbg_kms(&dev_priv->drm, "Disabling FBC on pipe %c\n",
1047 pipe_name(crtc->pipe));
1048
1049 __intel_fbc_cleanup_cfb(dev_priv);
1050
1051 fbc->crtc = NULL;
1052 }
1053
__intel_fbc_post_update(struct intel_crtc * crtc)1054 static void __intel_fbc_post_update(struct intel_crtc *crtc)
1055 {
1056 struct drm_i915_private *dev_priv = to_i915(crtc->base.dev);
1057 struct intel_fbc *fbc = &dev_priv->fbc;
1058
1059 drm_WARN_ON(&dev_priv->drm, !mutex_is_locked(&fbc->lock));
1060
1061 if (fbc->crtc != crtc)
1062 return;
1063
1064 fbc->flip_pending = false;
1065
1066 if (!dev_priv->params.enable_fbc) {
1067 intel_fbc_deactivate(dev_priv, "disabled at runtime per module param");
1068 __intel_fbc_disable(dev_priv);
1069
1070 return;
1071 }
1072
1073 intel_fbc_get_reg_params(crtc, &fbc->params);
1074
1075 if (!intel_fbc_can_activate(crtc))
1076 return;
1077
1078 if (!fbc->busy_bits)
1079 intel_fbc_hw_activate(dev_priv);
1080 else
1081 intel_fbc_deactivate(dev_priv, "frontbuffer write");
1082 }
1083
intel_fbc_post_update(struct intel_atomic_state * state,struct intel_crtc * crtc)1084 void intel_fbc_post_update(struct intel_atomic_state *state,
1085 struct intel_crtc *crtc)
1086 {
1087 struct drm_i915_private *dev_priv = to_i915(crtc->base.dev);
1088 struct intel_plane *plane = to_intel_plane(crtc->base.primary);
1089 const struct intel_plane_state *plane_state =
1090 intel_atomic_get_new_plane_state(state, plane);
1091 struct intel_fbc *fbc = &dev_priv->fbc;
1092
1093 if (!plane->has_fbc || !plane_state)
1094 return;
1095
1096 mutex_lock(&fbc->lock);
1097 __intel_fbc_post_update(crtc);
1098 mutex_unlock(&fbc->lock);
1099 }
1100
intel_fbc_get_frontbuffer_bit(struct intel_fbc * fbc)1101 static unsigned int intel_fbc_get_frontbuffer_bit(struct intel_fbc *fbc)
1102 {
1103 if (fbc->crtc)
1104 return to_intel_plane(fbc->crtc->base.primary)->frontbuffer_bit;
1105 else
1106 return fbc->possible_framebuffer_bits;
1107 }
1108
intel_fbc_invalidate(struct drm_i915_private * dev_priv,unsigned int frontbuffer_bits,enum fb_op_origin origin)1109 void intel_fbc_invalidate(struct drm_i915_private *dev_priv,
1110 unsigned int frontbuffer_bits,
1111 enum fb_op_origin origin)
1112 {
1113 struct intel_fbc *fbc = &dev_priv->fbc;
1114
1115 if (!HAS_FBC(dev_priv))
1116 return;
1117
1118 if (origin == ORIGIN_GTT || origin == ORIGIN_FLIP)
1119 return;
1120
1121 mutex_lock(&fbc->lock);
1122
1123 fbc->busy_bits |= intel_fbc_get_frontbuffer_bit(fbc) & frontbuffer_bits;
1124
1125 if (fbc->crtc && fbc->busy_bits)
1126 intel_fbc_deactivate(dev_priv, "frontbuffer write");
1127
1128 mutex_unlock(&fbc->lock);
1129 }
1130
intel_fbc_flush(struct drm_i915_private * dev_priv,unsigned int frontbuffer_bits,enum fb_op_origin origin)1131 void intel_fbc_flush(struct drm_i915_private *dev_priv,
1132 unsigned int frontbuffer_bits, enum fb_op_origin origin)
1133 {
1134 struct intel_fbc *fbc = &dev_priv->fbc;
1135
1136 if (!HAS_FBC(dev_priv))
1137 return;
1138
1139 /*
1140 * GTT tracking does not nuke the entire cfb
1141 * so don't clear busy_bits set for some other
1142 * reason.
1143 */
1144 if (origin == ORIGIN_GTT)
1145 return;
1146
1147 mutex_lock(&fbc->lock);
1148
1149 fbc->busy_bits &= ~frontbuffer_bits;
1150
1151 if (origin == ORIGIN_FLIP)
1152 goto out;
1153
1154 if (!fbc->busy_bits && fbc->crtc &&
1155 (frontbuffer_bits & intel_fbc_get_frontbuffer_bit(fbc))) {
1156 if (fbc->active)
1157 intel_fbc_recompress(dev_priv);
1158 else if (!fbc->flip_pending)
1159 __intel_fbc_post_update(fbc->crtc);
1160 }
1161
1162 out:
1163 mutex_unlock(&fbc->lock);
1164 }
1165
1166 /**
1167 * intel_fbc_choose_crtc - select a CRTC to enable FBC on
1168 * @dev_priv: i915 device instance
1169 * @state: the atomic state structure
1170 *
1171 * This function looks at the proposed state for CRTCs and planes, then chooses
1172 * which pipe is going to have FBC by setting intel_crtc_state->enable_fbc to
1173 * true.
1174 *
1175 * Later, intel_fbc_enable is going to look for state->enable_fbc and then maybe
1176 * enable FBC for the chosen CRTC. If it does, it will set dev_priv->fbc.crtc.
1177 */
intel_fbc_choose_crtc(struct drm_i915_private * dev_priv,struct intel_atomic_state * state)1178 void intel_fbc_choose_crtc(struct drm_i915_private *dev_priv,
1179 struct intel_atomic_state *state)
1180 {
1181 struct intel_fbc *fbc = &dev_priv->fbc;
1182 struct intel_plane *plane;
1183 struct intel_plane_state *plane_state;
1184 bool crtc_chosen = false;
1185 int i;
1186
1187 mutex_lock(&fbc->lock);
1188
1189 /* Does this atomic commit involve the CRTC currently tied to FBC? */
1190 if (fbc->crtc &&
1191 !intel_atomic_get_new_crtc_state(state, fbc->crtc))
1192 goto out;
1193
1194 if (!intel_fbc_can_enable(dev_priv))
1195 goto out;
1196
1197 /* Simply choose the first CRTC that is compatible and has a visible
1198 * plane. We could go for fancier schemes such as checking the plane
1199 * size, but this would just affect the few platforms that don't tie FBC
1200 * to pipe or plane A. */
1201 for_each_new_intel_plane_in_state(state, plane, plane_state, i) {
1202 struct intel_crtc_state *crtc_state;
1203 struct intel_crtc *crtc = to_intel_crtc(plane_state->hw.crtc);
1204
1205 if (!plane->has_fbc)
1206 continue;
1207
1208 if (!plane_state->uapi.visible)
1209 continue;
1210
1211 crtc_state = intel_atomic_get_new_crtc_state(state, crtc);
1212
1213 crtc_state->enable_fbc = true;
1214 crtc_chosen = true;
1215 break;
1216 }
1217
1218 if (!crtc_chosen)
1219 fbc->no_fbc_reason = "no suitable CRTC for FBC";
1220
1221 out:
1222 mutex_unlock(&fbc->lock);
1223 }
1224
1225 /**
1226 * intel_fbc_enable: tries to enable FBC on the CRTC
1227 * @crtc: the CRTC
1228 * @state: corresponding &drm_crtc_state for @crtc
1229 *
1230 * This function checks if the given CRTC was chosen for FBC, then enables it if
1231 * possible. Notice that it doesn't activate FBC. It is valid to call
1232 * intel_fbc_enable multiple times for the same pipe without an
1233 * intel_fbc_disable in the middle, as long as it is deactivated.
1234 */
intel_fbc_enable(struct intel_atomic_state * state,struct intel_crtc * crtc)1235 void intel_fbc_enable(struct intel_atomic_state *state,
1236 struct intel_crtc *crtc)
1237 {
1238 struct drm_i915_private *dev_priv = to_i915(crtc->base.dev);
1239 struct intel_plane *plane = to_intel_plane(crtc->base.primary);
1240 const struct intel_crtc_state *crtc_state =
1241 intel_atomic_get_new_crtc_state(state, crtc);
1242 const struct intel_plane_state *plane_state =
1243 intel_atomic_get_new_plane_state(state, plane);
1244 struct intel_fbc *fbc = &dev_priv->fbc;
1245 struct intel_fbc_state_cache *cache = &fbc->state_cache;
1246
1247 if (!plane->has_fbc || !plane_state)
1248 return;
1249
1250 mutex_lock(&fbc->lock);
1251
1252 if (fbc->crtc) {
1253 if (fbc->crtc != crtc ||
1254 (!intel_fbc_cfb_size_changed(dev_priv) &&
1255 !intel_fbc_gen9_wa_cfb_stride_changed(dev_priv)))
1256 goto out;
1257
1258 __intel_fbc_disable(dev_priv);
1259 }
1260
1261 drm_WARN_ON(&dev_priv->drm, fbc->active);
1262
1263 intel_fbc_update_state_cache(crtc, crtc_state, plane_state);
1264
1265 /* FIXME crtc_state->enable_fbc lies :( */
1266 if (!cache->plane.visible)
1267 goto out;
1268
1269 if (intel_fbc_alloc_cfb(dev_priv,
1270 intel_fbc_calculate_cfb_size(dev_priv, cache),
1271 plane_state->hw.fb->format->cpp[0])) {
1272 cache->plane.visible = false;
1273 fbc->no_fbc_reason = "not enough stolen memory";
1274 goto out;
1275 }
1276
1277 cache->gen9_wa_cfb_stride = intel_fbc_gen9_wa_cfb_stride(dev_priv);
1278
1279 drm_dbg_kms(&dev_priv->drm, "Enabling FBC on pipe %c\n",
1280 pipe_name(crtc->pipe));
1281 fbc->no_fbc_reason = "FBC enabled but not active yet\n";
1282
1283 fbc->crtc = crtc;
1284 out:
1285 mutex_unlock(&fbc->lock);
1286 }
1287
1288 /**
1289 * intel_fbc_disable - disable FBC if it's associated with crtc
1290 * @crtc: the CRTC
1291 *
1292 * This function disables FBC if it's associated with the provided CRTC.
1293 */
intel_fbc_disable(struct intel_crtc * crtc)1294 void intel_fbc_disable(struct intel_crtc *crtc)
1295 {
1296 struct drm_i915_private *dev_priv = to_i915(crtc->base.dev);
1297 struct intel_plane *plane = to_intel_plane(crtc->base.primary);
1298 struct intel_fbc *fbc = &dev_priv->fbc;
1299
1300 if (!plane->has_fbc)
1301 return;
1302
1303 mutex_lock(&fbc->lock);
1304 if (fbc->crtc == crtc)
1305 __intel_fbc_disable(dev_priv);
1306 mutex_unlock(&fbc->lock);
1307 }
1308
1309 /**
1310 * intel_fbc_global_disable - globally disable FBC
1311 * @dev_priv: i915 device instance
1312 *
1313 * This function disables FBC regardless of which CRTC is associated with it.
1314 */
intel_fbc_global_disable(struct drm_i915_private * dev_priv)1315 void intel_fbc_global_disable(struct drm_i915_private *dev_priv)
1316 {
1317 struct intel_fbc *fbc = &dev_priv->fbc;
1318
1319 if (!HAS_FBC(dev_priv))
1320 return;
1321
1322 mutex_lock(&fbc->lock);
1323 if (fbc->crtc) {
1324 drm_WARN_ON(&dev_priv->drm, fbc->crtc->active);
1325 __intel_fbc_disable(dev_priv);
1326 }
1327 mutex_unlock(&fbc->lock);
1328 }
1329
intel_fbc_underrun_work_fn(struct work_struct * work)1330 static void intel_fbc_underrun_work_fn(struct work_struct *work)
1331 {
1332 struct drm_i915_private *dev_priv =
1333 container_of(work, struct drm_i915_private, fbc.underrun_work);
1334 struct intel_fbc *fbc = &dev_priv->fbc;
1335
1336 mutex_lock(&fbc->lock);
1337
1338 /* Maybe we were scheduled twice. */
1339 if (fbc->underrun_detected || !fbc->crtc)
1340 goto out;
1341
1342 drm_dbg_kms(&dev_priv->drm, "Disabling FBC due to FIFO underrun.\n");
1343 fbc->underrun_detected = true;
1344
1345 intel_fbc_deactivate(dev_priv, "FIFO underrun");
1346 out:
1347 mutex_unlock(&fbc->lock);
1348 }
1349
1350 /*
1351 * intel_fbc_reset_underrun - reset FBC fifo underrun status.
1352 * @dev_priv: i915 device instance
1353 *
1354 * See intel_fbc_handle_fifo_underrun_irq(). For automated testing we
1355 * want to re-enable FBC after an underrun to increase test coverage.
1356 */
intel_fbc_reset_underrun(struct drm_i915_private * dev_priv)1357 int intel_fbc_reset_underrun(struct drm_i915_private *dev_priv)
1358 {
1359 int ret;
1360
1361 cancel_work_sync(&dev_priv->fbc.underrun_work);
1362
1363 ret = mutex_lock_interruptible(&dev_priv->fbc.lock);
1364 if (ret)
1365 return ret;
1366
1367 if (dev_priv->fbc.underrun_detected) {
1368 drm_dbg_kms(&dev_priv->drm,
1369 "Re-allowing FBC after fifo underrun\n");
1370 dev_priv->fbc.no_fbc_reason = "FIFO underrun cleared";
1371 }
1372
1373 dev_priv->fbc.underrun_detected = false;
1374 mutex_unlock(&dev_priv->fbc.lock);
1375
1376 return 0;
1377 }
1378
1379 /**
1380 * intel_fbc_handle_fifo_underrun_irq - disable FBC when we get a FIFO underrun
1381 * @dev_priv: i915 device instance
1382 *
1383 * Without FBC, most underruns are harmless and don't really cause too many
1384 * problems, except for an annoying message on dmesg. With FBC, underruns can
1385 * become black screens or even worse, especially when paired with bad
1386 * watermarks. So in order for us to be on the safe side, completely disable FBC
1387 * in case we ever detect a FIFO underrun on any pipe. An underrun on any pipe
1388 * already suggests that watermarks may be bad, so try to be as safe as
1389 * possible.
1390 *
1391 * This function is called from the IRQ handler.
1392 */
intel_fbc_handle_fifo_underrun_irq(struct drm_i915_private * dev_priv)1393 void intel_fbc_handle_fifo_underrun_irq(struct drm_i915_private *dev_priv)
1394 {
1395 struct intel_fbc *fbc = &dev_priv->fbc;
1396
1397 if (!HAS_FBC(dev_priv))
1398 return;
1399
1400 /* There's no guarantee that underrun_detected won't be set to true
1401 * right after this check and before the work is scheduled, but that's
1402 * not a problem since we'll check it again under the work function
1403 * while FBC is locked. This check here is just to prevent us from
1404 * unnecessarily scheduling the work, and it relies on the fact that we
1405 * never switch underrun_detect back to false after it's true. */
1406 if (READ_ONCE(fbc->underrun_detected))
1407 return;
1408
1409 schedule_work(&fbc->underrun_work);
1410 }
1411
1412 /*
1413 * The DDX driver changes its behavior depending on the value it reads from
1414 * i915.enable_fbc, so sanitize it by translating the default value into either
1415 * 0 or 1 in order to allow it to know what's going on.
1416 *
1417 * Notice that this is done at driver initialization and we still allow user
1418 * space to change the value during runtime without sanitizing it again. IGT
1419 * relies on being able to change i915.enable_fbc at runtime.
1420 */
intel_sanitize_fbc_option(struct drm_i915_private * dev_priv)1421 static int intel_sanitize_fbc_option(struct drm_i915_private *dev_priv)
1422 {
1423 if (dev_priv->params.enable_fbc >= 0)
1424 return !!dev_priv->params.enable_fbc;
1425
1426 if (!HAS_FBC(dev_priv))
1427 return 0;
1428
1429 /*
1430 * Fbc is causing random underruns in CI execution on TGL platforms.
1431 * Disabling the same while the problem is being debugged and analyzed.
1432 */
1433 if (IS_TIGERLAKE(dev_priv))
1434 return 0;
1435
1436 if (IS_BROADWELL(dev_priv) || INTEL_GEN(dev_priv) >= 9)
1437 return 1;
1438
1439 return 0;
1440 }
1441
need_fbc_vtd_wa(struct drm_i915_private * dev_priv)1442 static bool need_fbc_vtd_wa(struct drm_i915_private *dev_priv)
1443 {
1444 /* WaFbcTurnOffFbcWhenHyperVisorIsUsed:skl,bxt */
1445 if (intel_vtd_active() &&
1446 (IS_SKYLAKE(dev_priv) || IS_BROXTON(dev_priv))) {
1447 drm_info(&dev_priv->drm,
1448 "Disabling framebuffer compression (FBC) to prevent screen flicker with VT-d enabled\n");
1449 return true;
1450 }
1451
1452 return false;
1453 }
1454
1455 /**
1456 * intel_fbc_init - Initialize FBC
1457 * @dev_priv: the i915 device
1458 *
1459 * This function might be called during PM init process.
1460 */
intel_fbc_init(struct drm_i915_private * dev_priv)1461 void intel_fbc_init(struct drm_i915_private *dev_priv)
1462 {
1463 struct intel_fbc *fbc = &dev_priv->fbc;
1464
1465 INIT_WORK(&fbc->underrun_work, intel_fbc_underrun_work_fn);
1466 mutex_init(&fbc->lock);
1467 fbc->active = false;
1468
1469 if (!drm_mm_initialized(&dev_priv->mm.stolen))
1470 mkwrite_device_info(dev_priv)->display.has_fbc = false;
1471
1472 if (need_fbc_vtd_wa(dev_priv))
1473 mkwrite_device_info(dev_priv)->display.has_fbc = false;
1474
1475 dev_priv->params.enable_fbc = intel_sanitize_fbc_option(dev_priv);
1476 drm_dbg_kms(&dev_priv->drm, "Sanitized enable_fbc value: %d\n",
1477 dev_priv->params.enable_fbc);
1478
1479 if (!HAS_FBC(dev_priv)) {
1480 fbc->no_fbc_reason = "unsupported by this chipset";
1481 return;
1482 }
1483
1484 /* We still don't have any sort of hardware state readout for FBC, so
1485 * deactivate it in case the BIOS activated it to make sure software
1486 * matches the hardware state. */
1487 if (intel_fbc_hw_is_active(dev_priv))
1488 intel_fbc_hw_deactivate(dev_priv);
1489 }
1490