Merge drm/drm-next into drm-misc-next
[linux-2.6-block.git] / drivers / gpu / drm / vc4 / vc4_plane.c
1 /*
2  * Copyright (C) 2015 Broadcom
3  *
4  * This program is free software; you can redistribute it and/or modify
5  * it under the terms of the GNU General Public License version 2 as
6  * published by the Free Software Foundation.
7  */
8
9 /**
10  * DOC: VC4 plane module
11  *
12  * Each DRM plane is a layer of pixels being scanned out by the HVS.
13  *
14  * At atomic modeset check time, we compute the HVS display element
15  * state that would be necessary for displaying the plane (giving us a
16  * chance to figure out if a plane configuration is invalid), then at
17  * atomic flush time the CRTC will ask us to write our element state
18  * into the region of the HVS that it has allocated for us.
19  */
20
21 #include <drm/drm_atomic.h>
22 #include <drm/drm_atomic_helper.h>
23 #include <drm/drm_fb_cma_helper.h>
24 #include <drm/drm_plane_helper.h>
25
26 #include "uapi/drm/vc4_drm.h"
27 #include "vc4_drv.h"
28 #include "vc4_regs.h"
29
30 static const struct hvs_format {
31         u32 drm; /* DRM_FORMAT_* */
32         u32 hvs; /* HVS_FORMAT_* */
33         u32 pixel_order;
34 } hvs_formats[] = {
35         {
36                 .drm = DRM_FORMAT_XRGB8888, .hvs = HVS_PIXEL_FORMAT_RGBA8888,
37                 .pixel_order = HVS_PIXEL_ORDER_ABGR,
38         },
39         {
40                 .drm = DRM_FORMAT_ARGB8888, .hvs = HVS_PIXEL_FORMAT_RGBA8888,
41                 .pixel_order = HVS_PIXEL_ORDER_ABGR,
42         },
43         {
44                 .drm = DRM_FORMAT_ABGR8888, .hvs = HVS_PIXEL_FORMAT_RGBA8888,
45                 .pixel_order = HVS_PIXEL_ORDER_ARGB,
46         },
47         {
48                 .drm = DRM_FORMAT_XBGR8888, .hvs = HVS_PIXEL_FORMAT_RGBA8888,
49                 .pixel_order = HVS_PIXEL_ORDER_ARGB,
50         },
51         {
52                 .drm = DRM_FORMAT_RGB565, .hvs = HVS_PIXEL_FORMAT_RGB565,
53                 .pixel_order = HVS_PIXEL_ORDER_XRGB,
54         },
55         {
56                 .drm = DRM_FORMAT_BGR565, .hvs = HVS_PIXEL_FORMAT_RGB565,
57                 .pixel_order = HVS_PIXEL_ORDER_XBGR,
58         },
59         {
60                 .drm = DRM_FORMAT_ARGB1555, .hvs = HVS_PIXEL_FORMAT_RGBA5551,
61                 .pixel_order = HVS_PIXEL_ORDER_ABGR,
62         },
63         {
64                 .drm = DRM_FORMAT_XRGB1555, .hvs = HVS_PIXEL_FORMAT_RGBA5551,
65                 .pixel_order = HVS_PIXEL_ORDER_ABGR,
66         },
67         {
68                 .drm = DRM_FORMAT_RGB888, .hvs = HVS_PIXEL_FORMAT_RGB888,
69                 .pixel_order = HVS_PIXEL_ORDER_XRGB,
70         },
71         {
72                 .drm = DRM_FORMAT_BGR888, .hvs = HVS_PIXEL_FORMAT_RGB888,
73                 .pixel_order = HVS_PIXEL_ORDER_XBGR,
74         },
75         {
76                 .drm = DRM_FORMAT_YUV422,
77                 .hvs = HVS_PIXEL_FORMAT_YCBCR_YUV422_3PLANE,
78                 .pixel_order = HVS_PIXEL_ORDER_XYCBCR,
79         },
80         {
81                 .drm = DRM_FORMAT_YVU422,
82                 .hvs = HVS_PIXEL_FORMAT_YCBCR_YUV422_3PLANE,
83                 .pixel_order = HVS_PIXEL_ORDER_XYCRCB,
84         },
85         {
86                 .drm = DRM_FORMAT_YUV420,
87                 .hvs = HVS_PIXEL_FORMAT_YCBCR_YUV420_3PLANE,
88                 .pixel_order = HVS_PIXEL_ORDER_XYCBCR,
89         },
90         {
91                 .drm = DRM_FORMAT_YVU420,
92                 .hvs = HVS_PIXEL_FORMAT_YCBCR_YUV420_3PLANE,
93                 .pixel_order = HVS_PIXEL_ORDER_XYCRCB,
94         },
95         {
96                 .drm = DRM_FORMAT_NV12,
97                 .hvs = HVS_PIXEL_FORMAT_YCBCR_YUV420_2PLANE,
98                 .pixel_order = HVS_PIXEL_ORDER_XYCBCR,
99         },
100         {
101                 .drm = DRM_FORMAT_NV21,
102                 .hvs = HVS_PIXEL_FORMAT_YCBCR_YUV420_2PLANE,
103                 .pixel_order = HVS_PIXEL_ORDER_XYCRCB,
104         },
105         {
106                 .drm = DRM_FORMAT_NV16,
107                 .hvs = HVS_PIXEL_FORMAT_YCBCR_YUV422_2PLANE,
108                 .pixel_order = HVS_PIXEL_ORDER_XYCBCR,
109         },
110         {
111                 .drm = DRM_FORMAT_NV61,
112                 .hvs = HVS_PIXEL_FORMAT_YCBCR_YUV422_2PLANE,
113                 .pixel_order = HVS_PIXEL_ORDER_XYCRCB,
114         },
115 };
116
117 static const struct hvs_format *vc4_get_hvs_format(u32 drm_format)
118 {
119         unsigned i;
120
121         for (i = 0; i < ARRAY_SIZE(hvs_formats); i++) {
122                 if (hvs_formats[i].drm == drm_format)
123                         return &hvs_formats[i];
124         }
125
126         return NULL;
127 }
128
129 static enum vc4_scaling_mode vc4_get_scaling_mode(u32 src, u32 dst)
130 {
131         if (dst > src)
132                 return VC4_SCALING_PPF;
133         else if (dst < src)
134                 return VC4_SCALING_TPZ;
135         else
136                 return VC4_SCALING_NONE;
137 }
138
139 static bool plane_enabled(struct drm_plane_state *state)
140 {
141         return state->fb && state->crtc;
142 }
143
144 static struct drm_plane_state *vc4_plane_duplicate_state(struct drm_plane *plane)
145 {
146         struct vc4_plane_state *vc4_state;
147
148         if (WARN_ON(!plane->state))
149                 return NULL;
150
151         vc4_state = kmemdup(plane->state, sizeof(*vc4_state), GFP_KERNEL);
152         if (!vc4_state)
153                 return NULL;
154
155         memset(&vc4_state->lbm, 0, sizeof(vc4_state->lbm));
156
157         __drm_atomic_helper_plane_duplicate_state(plane, &vc4_state->base);
158
159         if (vc4_state->dlist) {
160                 vc4_state->dlist = kmemdup(vc4_state->dlist,
161                                            vc4_state->dlist_count * 4,
162                                            GFP_KERNEL);
163                 if (!vc4_state->dlist) {
164                         kfree(vc4_state);
165                         return NULL;
166                 }
167                 vc4_state->dlist_size = vc4_state->dlist_count;
168         }
169
170         return &vc4_state->base;
171 }
172
173 static void vc4_plane_destroy_state(struct drm_plane *plane,
174                                     struct drm_plane_state *state)
175 {
176         struct vc4_dev *vc4 = to_vc4_dev(plane->dev);
177         struct vc4_plane_state *vc4_state = to_vc4_plane_state(state);
178
179         if (vc4_state->lbm.allocated) {
180                 unsigned long irqflags;
181
182                 spin_lock_irqsave(&vc4->hvs->mm_lock, irqflags);
183                 drm_mm_remove_node(&vc4_state->lbm);
184                 spin_unlock_irqrestore(&vc4->hvs->mm_lock, irqflags);
185         }
186
187         kfree(vc4_state->dlist);
188         __drm_atomic_helper_plane_destroy_state(&vc4_state->base);
189         kfree(state);
190 }
191
192 /* Called during init to allocate the plane's atomic state. */
193 static void vc4_plane_reset(struct drm_plane *plane)
194 {
195         struct vc4_plane_state *vc4_state;
196
197         WARN_ON(plane->state);
198
199         vc4_state = kzalloc(sizeof(*vc4_state), GFP_KERNEL);
200         if (!vc4_state)
201                 return;
202
203         __drm_atomic_helper_plane_reset(plane, &vc4_state->base);
204 }
205
206 static void vc4_dlist_write(struct vc4_plane_state *vc4_state, u32 val)
207 {
208         if (vc4_state->dlist_count == vc4_state->dlist_size) {
209                 u32 new_size = max(4u, vc4_state->dlist_count * 2);
210                 u32 *new_dlist = kmalloc_array(new_size, 4, GFP_KERNEL);
211
212                 if (!new_dlist)
213                         return;
214                 memcpy(new_dlist, vc4_state->dlist, vc4_state->dlist_count * 4);
215
216                 kfree(vc4_state->dlist);
217                 vc4_state->dlist = new_dlist;
218                 vc4_state->dlist_size = new_size;
219         }
220
221         vc4_state->dlist[vc4_state->dlist_count++] = val;
222 }
223
224 /* Returns the scl0/scl1 field based on whether the dimensions need to
225  * be up/down/non-scaled.
226  *
227  * This is a replication of a table from the spec.
228  */
229 static u32 vc4_get_scl_field(struct drm_plane_state *state, int plane)
230 {
231         struct vc4_plane_state *vc4_state = to_vc4_plane_state(state);
232
233         switch (vc4_state->x_scaling[plane] << 2 | vc4_state->y_scaling[plane]) {
234         case VC4_SCALING_PPF << 2 | VC4_SCALING_PPF:
235                 return SCALER_CTL0_SCL_H_PPF_V_PPF;
236         case VC4_SCALING_TPZ << 2 | VC4_SCALING_PPF:
237                 return SCALER_CTL0_SCL_H_TPZ_V_PPF;
238         case VC4_SCALING_PPF << 2 | VC4_SCALING_TPZ:
239                 return SCALER_CTL0_SCL_H_PPF_V_TPZ;
240         case VC4_SCALING_TPZ << 2 | VC4_SCALING_TPZ:
241                 return SCALER_CTL0_SCL_H_TPZ_V_TPZ;
242         case VC4_SCALING_PPF << 2 | VC4_SCALING_NONE:
243                 return SCALER_CTL0_SCL_H_PPF_V_NONE;
244         case VC4_SCALING_NONE << 2 | VC4_SCALING_PPF:
245                 return SCALER_CTL0_SCL_H_NONE_V_PPF;
246         case VC4_SCALING_NONE << 2 | VC4_SCALING_TPZ:
247                 return SCALER_CTL0_SCL_H_NONE_V_TPZ;
248         case VC4_SCALING_TPZ << 2 | VC4_SCALING_NONE:
249                 return SCALER_CTL0_SCL_H_TPZ_V_NONE;
250         default:
251         case VC4_SCALING_NONE << 2 | VC4_SCALING_NONE:
252                 /* The unity case is independently handled by
253                  * SCALER_CTL0_UNITY.
254                  */
255                 return 0;
256         }
257 }
258
259 static int vc4_plane_setup_clipping_and_scaling(struct drm_plane_state *state)
260 {
261         struct drm_plane *plane = state->plane;
262         struct vc4_plane_state *vc4_state = to_vc4_plane_state(state);
263         struct drm_framebuffer *fb = state->fb;
264         struct drm_gem_cma_object *bo = drm_fb_cma_get_gem_obj(fb, 0);
265         u32 subpixel_src_mask = (1 << 16) - 1;
266         u32 format = fb->format->format;
267         int num_planes = fb->format->num_planes;
268         u32 h_subsample = 1;
269         u32 v_subsample = 1;
270         int i;
271
272         for (i = 0; i < num_planes; i++)
273                 vc4_state->offsets[i] = bo->paddr + fb->offsets[i];
274
275         /* We don't support subpixel source positioning for scaling. */
276         if ((state->src_x & subpixel_src_mask) ||
277             (state->src_y & subpixel_src_mask) ||
278             (state->src_w & subpixel_src_mask) ||
279             (state->src_h & subpixel_src_mask)) {
280                 return -EINVAL;
281         }
282
283         vc4_state->src_x = state->src_x >> 16;
284         vc4_state->src_y = state->src_y >> 16;
285         vc4_state->src_w[0] = state->src_w >> 16;
286         vc4_state->src_h[0] = state->src_h >> 16;
287
288         vc4_state->crtc_x = state->crtc_x;
289         vc4_state->crtc_y = state->crtc_y;
290         vc4_state->crtc_w = state->crtc_w;
291         vc4_state->crtc_h = state->crtc_h;
292
293         vc4_state->x_scaling[0] = vc4_get_scaling_mode(vc4_state->src_w[0],
294                                                        vc4_state->crtc_w);
295         vc4_state->y_scaling[0] = vc4_get_scaling_mode(vc4_state->src_h[0],
296                                                        vc4_state->crtc_h);
297
298         if (num_planes > 1) {
299                 vc4_state->is_yuv = true;
300
301                 h_subsample = drm_format_horz_chroma_subsampling(format);
302                 v_subsample = drm_format_vert_chroma_subsampling(format);
303                 vc4_state->src_w[1] = vc4_state->src_w[0] / h_subsample;
304                 vc4_state->src_h[1] = vc4_state->src_h[0] / v_subsample;
305
306                 vc4_state->x_scaling[1] =
307                         vc4_get_scaling_mode(vc4_state->src_w[1],
308                                              vc4_state->crtc_w);
309                 vc4_state->y_scaling[1] =
310                         vc4_get_scaling_mode(vc4_state->src_h[1],
311                                              vc4_state->crtc_h);
312
313                 /* YUV conversion requires that scaling be enabled,
314                  * even on a plane that's otherwise 1:1.  Choose TPZ
315                  * for simplicity.
316                  */
317                 if (vc4_state->x_scaling[0] == VC4_SCALING_NONE)
318                         vc4_state->x_scaling[0] = VC4_SCALING_TPZ;
319                 if (vc4_state->y_scaling[0] == VC4_SCALING_NONE)
320                         vc4_state->y_scaling[0] = VC4_SCALING_TPZ;
321         } else {
322                 vc4_state->x_scaling[1] = VC4_SCALING_NONE;
323                 vc4_state->y_scaling[1] = VC4_SCALING_NONE;
324         }
325
326         vc4_state->is_unity = (vc4_state->x_scaling[0] == VC4_SCALING_NONE &&
327                                vc4_state->y_scaling[0] == VC4_SCALING_NONE &&
328                                vc4_state->x_scaling[1] == VC4_SCALING_NONE &&
329                                vc4_state->y_scaling[1] == VC4_SCALING_NONE);
330
331         /* No configuring scaling on the cursor plane, since it gets
332            non-vblank-synced updates, and scaling requires requires
333            LBM changes which have to be vblank-synced.
334          */
335         if (plane->type == DRM_PLANE_TYPE_CURSOR && !vc4_state->is_unity)
336                 return -EINVAL;
337
338         /* Clamp the on-screen start x/y to 0.  The hardware doesn't
339          * support negative y, and negative x wastes bandwidth.
340          */
341         if (vc4_state->crtc_x < 0) {
342                 for (i = 0; i < num_planes; i++) {
343                         u32 cpp = fb->format->cpp[i];
344                         u32 subs = ((i == 0) ? 1 : h_subsample);
345
346                         vc4_state->offsets[i] += (cpp *
347                                                   (-vc4_state->crtc_x) / subs);
348                 }
349                 vc4_state->src_w[0] += vc4_state->crtc_x;
350                 vc4_state->src_w[1] += vc4_state->crtc_x / h_subsample;
351                 vc4_state->crtc_x = 0;
352         }
353
354         if (vc4_state->crtc_y < 0) {
355                 for (i = 0; i < num_planes; i++) {
356                         u32 subs = ((i == 0) ? 1 : v_subsample);
357
358                         vc4_state->offsets[i] += (fb->pitches[i] *
359                                                   (-vc4_state->crtc_y) / subs);
360                 }
361                 vc4_state->src_h[0] += vc4_state->crtc_y;
362                 vc4_state->src_h[1] += vc4_state->crtc_y / v_subsample;
363                 vc4_state->crtc_y = 0;
364         }
365
366         return 0;
367 }
368
369 static void vc4_write_tpz(struct vc4_plane_state *vc4_state, u32 src, u32 dst)
370 {
371         u32 scale, recip;
372
373         scale = (1 << 16) * src / dst;
374
375         /* The specs note that while the reciprocal would be defined
376          * as (1<<32)/scale, ~0 is close enough.
377          */
378         recip = ~0 / scale;
379
380         vc4_dlist_write(vc4_state,
381                         VC4_SET_FIELD(scale, SCALER_TPZ0_SCALE) |
382                         VC4_SET_FIELD(0, SCALER_TPZ0_IPHASE));
383         vc4_dlist_write(vc4_state,
384                         VC4_SET_FIELD(recip, SCALER_TPZ1_RECIP));
385 }
386
387 static void vc4_write_ppf(struct vc4_plane_state *vc4_state, u32 src, u32 dst)
388 {
389         u32 scale = (1 << 16) * src / dst;
390
391         vc4_dlist_write(vc4_state,
392                         SCALER_PPF_AGC |
393                         VC4_SET_FIELD(scale, SCALER_PPF_SCALE) |
394                         VC4_SET_FIELD(0, SCALER_PPF_IPHASE));
395 }
396
397 static u32 vc4_lbm_size(struct drm_plane_state *state)
398 {
399         struct vc4_plane_state *vc4_state = to_vc4_plane_state(state);
400         /* This is the worst case number.  One of the two sizes will
401          * be used depending on the scaling configuration.
402          */
403         u32 pix_per_line = max(vc4_state->src_w[0], (u32)vc4_state->crtc_w);
404         u32 lbm;
405
406         if (!vc4_state->is_yuv) {
407                 if (vc4_state->is_unity)
408                         return 0;
409                 else if (vc4_state->y_scaling[0] == VC4_SCALING_TPZ)
410                         lbm = pix_per_line * 8;
411                 else {
412                         /* In special cases, this multiplier might be 12. */
413                         lbm = pix_per_line * 16;
414                 }
415         } else {
416                 /* There are cases for this going down to a multiplier
417                  * of 2, but according to the firmware source, the
418                  * table in the docs is somewhat wrong.
419                  */
420                 lbm = pix_per_line * 16;
421         }
422
423         lbm = roundup(lbm, 32);
424
425         return lbm;
426 }
427
428 static void vc4_write_scaling_parameters(struct drm_plane_state *state,
429                                          int channel)
430 {
431         struct vc4_plane_state *vc4_state = to_vc4_plane_state(state);
432
433         /* Ch0 H-PPF Word 0: Scaling Parameters */
434         if (vc4_state->x_scaling[channel] == VC4_SCALING_PPF) {
435                 vc4_write_ppf(vc4_state,
436                               vc4_state->src_w[channel], vc4_state->crtc_w);
437         }
438
439         /* Ch0 V-PPF Words 0-1: Scaling Parameters, Context */
440         if (vc4_state->y_scaling[channel] == VC4_SCALING_PPF) {
441                 vc4_write_ppf(vc4_state,
442                               vc4_state->src_h[channel], vc4_state->crtc_h);
443                 vc4_dlist_write(vc4_state, 0xc0c0c0c0);
444         }
445
446         /* Ch0 H-TPZ Words 0-1: Scaling Parameters, Recip */
447         if (vc4_state->x_scaling[channel] == VC4_SCALING_TPZ) {
448                 vc4_write_tpz(vc4_state,
449                               vc4_state->src_w[channel], vc4_state->crtc_w);
450         }
451
452         /* Ch0 V-TPZ Words 0-2: Scaling Parameters, Recip, Context */
453         if (vc4_state->y_scaling[channel] == VC4_SCALING_TPZ) {
454                 vc4_write_tpz(vc4_state,
455                               vc4_state->src_h[channel], vc4_state->crtc_h);
456                 vc4_dlist_write(vc4_state, 0xc0c0c0c0);
457         }
458 }
459
460 /* Writes out a full display list for an active plane to the plane's
461  * private dlist state.
462  */
463 static int vc4_plane_mode_set(struct drm_plane *plane,
464                               struct drm_plane_state *state)
465 {
466         struct vc4_dev *vc4 = to_vc4_dev(plane->dev);
467         struct vc4_plane_state *vc4_state = to_vc4_plane_state(state);
468         struct drm_framebuffer *fb = state->fb;
469         u32 ctl0_offset = vc4_state->dlist_count;
470         const struct hvs_format *format = vc4_get_hvs_format(fb->format->format);
471         u64 base_format_mod = fourcc_mod_broadcom_mod(fb->modifier);
472         int num_planes = drm_format_num_planes(format->drm);
473         bool mix_plane_alpha;
474         bool covers_screen;
475         u32 scl0, scl1, pitch0;
476         u32 lbm_size, tiling;
477         unsigned long irqflags;
478         u32 hvs_format = format->hvs;
479         int ret, i;
480
481         ret = vc4_plane_setup_clipping_and_scaling(state);
482         if (ret)
483                 return ret;
484
485         /* Allocate the LBM memory that the HVS will use for temporary
486          * storage due to our scaling/format conversion.
487          */
488         lbm_size = vc4_lbm_size(state);
489         if (lbm_size) {
490                 if (!vc4_state->lbm.allocated) {
491                         spin_lock_irqsave(&vc4->hvs->mm_lock, irqflags);
492                         ret = drm_mm_insert_node_generic(&vc4->hvs->lbm_mm,
493                                                          &vc4_state->lbm,
494                                                          lbm_size, 32, 0, 0);
495                         spin_unlock_irqrestore(&vc4->hvs->mm_lock, irqflags);
496                 } else {
497                         WARN_ON_ONCE(lbm_size != vc4_state->lbm.size);
498                 }
499         }
500
501         if (ret)
502                 return ret;
503
504         /* SCL1 is used for Cb/Cr scaling of planar formats.  For RGB
505          * and 4:4:4, scl1 should be set to scl0 so both channels of
506          * the scaler do the same thing.  For YUV, the Y plane needs
507          * to be put in channel 1 and Cb/Cr in channel 0, so we swap
508          * the scl fields here.
509          */
510         if (num_planes == 1) {
511                 scl0 = vc4_get_scl_field(state, 0);
512                 scl1 = scl0;
513         } else {
514                 scl0 = vc4_get_scl_field(state, 1);
515                 scl1 = vc4_get_scl_field(state, 0);
516         }
517
518         switch (base_format_mod) {
519         case DRM_FORMAT_MOD_LINEAR:
520                 tiling = SCALER_CTL0_TILING_LINEAR;
521                 pitch0 = VC4_SET_FIELD(fb->pitches[0], SCALER_SRC_PITCH);
522                 break;
523
524         case DRM_FORMAT_MOD_BROADCOM_VC4_T_TILED: {
525                 /* For T-tiled, the FB pitch is "how many bytes from
526                  * one row to the next, such that pitch * tile_h ==
527                  * tile_size * tiles_per_row."
528                  */
529                 u32 tile_size_shift = 12; /* T tiles are 4kb */
530                 u32 tile_h_shift = 5; /* 16 and 32bpp are 32 pixels high */
531                 u32 tiles_w = fb->pitches[0] >> (tile_size_shift - tile_h_shift);
532
533                 tiling = SCALER_CTL0_TILING_256B_OR_T;
534
535                 pitch0 = (VC4_SET_FIELD(0, SCALER_PITCH0_TILE_Y_OFFSET) |
536                           VC4_SET_FIELD(0, SCALER_PITCH0_TILE_WIDTH_L) |
537                           VC4_SET_FIELD(tiles_w, SCALER_PITCH0_TILE_WIDTH_R));
538                 break;
539         }
540
541         case DRM_FORMAT_MOD_BROADCOM_SAND64:
542         case DRM_FORMAT_MOD_BROADCOM_SAND128:
543         case DRM_FORMAT_MOD_BROADCOM_SAND256: {
544                 uint32_t param = fourcc_mod_broadcom_param(fb->modifier);
545
546                 /* Column-based NV12 or RGBA.
547                  */
548                 if (fb->format->num_planes > 1) {
549                         if (hvs_format != HVS_PIXEL_FORMAT_YCBCR_YUV420_2PLANE) {
550                                 DRM_DEBUG_KMS("SAND format only valid for NV12/21");
551                                 return -EINVAL;
552                         }
553                         hvs_format = HVS_PIXEL_FORMAT_H264;
554                 } else {
555                         if (base_format_mod == DRM_FORMAT_MOD_BROADCOM_SAND256) {
556                                 DRM_DEBUG_KMS("SAND256 format only valid for H.264");
557                                 return -EINVAL;
558                         }
559                 }
560
561                 switch (base_format_mod) {
562                 case DRM_FORMAT_MOD_BROADCOM_SAND64:
563                         tiling = SCALER_CTL0_TILING_64B;
564                         break;
565                 case DRM_FORMAT_MOD_BROADCOM_SAND128:
566                         tiling = SCALER_CTL0_TILING_128B;
567                         break;
568                 case DRM_FORMAT_MOD_BROADCOM_SAND256:
569                         tiling = SCALER_CTL0_TILING_256B_OR_T;
570                         break;
571                 default:
572                         break;
573                 }
574
575                 if (param > SCALER_TILE_HEIGHT_MASK) {
576                         DRM_DEBUG_KMS("SAND height too large (%d)\n", param);
577                         return -EINVAL;
578                 }
579
580                 pitch0 = VC4_SET_FIELD(param, SCALER_TILE_HEIGHT);
581                 break;
582         }
583
584         default:
585                 DRM_DEBUG_KMS("Unsupported FB tiling flag 0x%16llx",
586                               (long long)fb->modifier);
587                 return -EINVAL;
588         }
589
590         /* Control word */
591         vc4_dlist_write(vc4_state,
592                         SCALER_CTL0_VALID |
593                         VC4_SET_FIELD(SCALER_CTL0_RGBA_EXPAND_ROUND, SCALER_CTL0_RGBA_EXPAND) |
594                         (format->pixel_order << SCALER_CTL0_ORDER_SHIFT) |
595                         (hvs_format << SCALER_CTL0_PIXEL_FORMAT_SHIFT) |
596                         VC4_SET_FIELD(tiling, SCALER_CTL0_TILING) |
597                         (vc4_state->is_unity ? SCALER_CTL0_UNITY : 0) |
598                         VC4_SET_FIELD(scl0, SCALER_CTL0_SCL0) |
599                         VC4_SET_FIELD(scl1, SCALER_CTL0_SCL1));
600
601         /* Position Word 0: Image Positions and Alpha Value */
602         vc4_state->pos0_offset = vc4_state->dlist_count;
603         vc4_dlist_write(vc4_state,
604                         VC4_SET_FIELD(state->alpha >> 8, SCALER_POS0_FIXED_ALPHA) |
605                         VC4_SET_FIELD(vc4_state->crtc_x, SCALER_POS0_START_X) |
606                         VC4_SET_FIELD(vc4_state->crtc_y, SCALER_POS0_START_Y));
607
608         /* Position Word 1: Scaled Image Dimensions. */
609         if (!vc4_state->is_unity) {
610                 vc4_dlist_write(vc4_state,
611                                 VC4_SET_FIELD(vc4_state->crtc_w,
612                                               SCALER_POS1_SCL_WIDTH) |
613                                 VC4_SET_FIELD(vc4_state->crtc_h,
614                                               SCALER_POS1_SCL_HEIGHT));
615         }
616
617         /* Don't waste cycles mixing with plane alpha if the set alpha
618          * is opaque or there is no per-pixel alpha information.
619          * In any case we use the alpha property value as the fixed alpha.
620          */
621         mix_plane_alpha = state->alpha != DRM_BLEND_ALPHA_OPAQUE &&
622                           fb->format->has_alpha;
623
624         /* Position Word 2: Source Image Size, Alpha */
625         vc4_state->pos2_offset = vc4_state->dlist_count;
626         vc4_dlist_write(vc4_state,
627                         VC4_SET_FIELD(fb->format->has_alpha ?
628                                       SCALER_POS2_ALPHA_MODE_PIPELINE :
629                                       SCALER_POS2_ALPHA_MODE_FIXED,
630                                       SCALER_POS2_ALPHA_MODE) |
631                         (mix_plane_alpha ? SCALER_POS2_ALPHA_MIX : 0) |
632                         (fb->format->has_alpha ? SCALER_POS2_ALPHA_PREMULT : 0) |
633                         VC4_SET_FIELD(vc4_state->src_w[0], SCALER_POS2_WIDTH) |
634                         VC4_SET_FIELD(vc4_state->src_h[0], SCALER_POS2_HEIGHT));
635
636         /* Position Word 3: Context.  Written by the HVS. */
637         vc4_dlist_write(vc4_state, 0xc0c0c0c0);
638
639
640         /* Pointer Word 0/1/2: RGB / Y / Cb / Cr Pointers
641          *
642          * The pointers may be any byte address.
643          */
644         vc4_state->ptr0_offset = vc4_state->dlist_count;
645         for (i = 0; i < num_planes; i++)
646                 vc4_dlist_write(vc4_state, vc4_state->offsets[i]);
647
648         /* Pointer Context Word 0/1/2: Written by the HVS */
649         for (i = 0; i < num_planes; i++)
650                 vc4_dlist_write(vc4_state, 0xc0c0c0c0);
651
652         /* Pitch word 0 */
653         vc4_dlist_write(vc4_state, pitch0);
654
655         /* Pitch word 1/2 */
656         for (i = 1; i < num_planes; i++) {
657                 if (hvs_format != HVS_PIXEL_FORMAT_H264) {
658                         vc4_dlist_write(vc4_state,
659                                         VC4_SET_FIELD(fb->pitches[i],
660                                                       SCALER_SRC_PITCH));
661                 } else {
662                         vc4_dlist_write(vc4_state, pitch0);
663                 }
664         }
665
666         /* Colorspace conversion words */
667         if (vc4_state->is_yuv) {
668                 vc4_dlist_write(vc4_state, SCALER_CSC0_ITR_R_601_5);
669                 vc4_dlist_write(vc4_state, SCALER_CSC1_ITR_R_601_5);
670                 vc4_dlist_write(vc4_state, SCALER_CSC2_ITR_R_601_5);
671         }
672
673         if (!vc4_state->is_unity) {
674                 /* LBM Base Address. */
675                 if (vc4_state->y_scaling[0] != VC4_SCALING_NONE ||
676                     vc4_state->y_scaling[1] != VC4_SCALING_NONE) {
677                         vc4_dlist_write(vc4_state, vc4_state->lbm.start);
678                 }
679
680                 if (num_planes > 1) {
681                         /* Emit Cb/Cr as channel 0 and Y as channel
682                          * 1. This matches how we set up scl0/scl1
683                          * above.
684                          */
685                         vc4_write_scaling_parameters(state, 1);
686                 }
687                 vc4_write_scaling_parameters(state, 0);
688
689                 /* If any PPF setup was done, then all the kernel
690                  * pointers get uploaded.
691                  */
692                 if (vc4_state->x_scaling[0] == VC4_SCALING_PPF ||
693                     vc4_state->y_scaling[0] == VC4_SCALING_PPF ||
694                     vc4_state->x_scaling[1] == VC4_SCALING_PPF ||
695                     vc4_state->y_scaling[1] == VC4_SCALING_PPF) {
696                         u32 kernel = VC4_SET_FIELD(vc4->hvs->mitchell_netravali_filter.start,
697                                                    SCALER_PPF_KERNEL_OFFSET);
698
699                         /* HPPF plane 0 */
700                         vc4_dlist_write(vc4_state, kernel);
701                         /* VPPF plane 0 */
702                         vc4_dlist_write(vc4_state, kernel);
703                         /* HPPF plane 1 */
704                         vc4_dlist_write(vc4_state, kernel);
705                         /* VPPF plane 1 */
706                         vc4_dlist_write(vc4_state, kernel);
707                 }
708         }
709
710         vc4_state->dlist[ctl0_offset] |=
711                 VC4_SET_FIELD(vc4_state->dlist_count, SCALER_CTL0_SIZE);
712
713         /* crtc_* are already clipped coordinates. */
714         covers_screen = vc4_state->crtc_x == 0 && vc4_state->crtc_y == 0 &&
715                         vc4_state->crtc_w == state->crtc->mode.hdisplay &&
716                         vc4_state->crtc_h == state->crtc->mode.vdisplay;
717         /* Background fill might be necessary when the plane has per-pixel
718          * alpha content or a non-opaque plane alpha and could blend from the
719          * background or does not cover the entire screen.
720          */
721         vc4_state->needs_bg_fill = fb->format->has_alpha || !covers_screen ||
722                                    state->alpha != DRM_BLEND_ALPHA_OPAQUE;
723
724         return 0;
725 }
726
727 /* If a modeset involves changing the setup of a plane, the atomic
728  * infrastructure will call this to validate a proposed plane setup.
729  * However, if a plane isn't getting updated, this (and the
730  * corresponding vc4_plane_atomic_update) won't get called.  Thus, we
731  * compute the dlist here and have all active plane dlists get updated
732  * in the CRTC's flush.
733  */
734 static int vc4_plane_atomic_check(struct drm_plane *plane,
735                                   struct drm_plane_state *state)
736 {
737         struct vc4_plane_state *vc4_state = to_vc4_plane_state(state);
738
739         vc4_state->dlist_count = 0;
740
741         if (plane_enabled(state))
742                 return vc4_plane_mode_set(plane, state);
743         else
744                 return 0;
745 }
746
747 static void vc4_plane_atomic_update(struct drm_plane *plane,
748                                     struct drm_plane_state *old_state)
749 {
750         /* No contents here.  Since we don't know where in the CRTC's
751          * dlist we should be stored, our dlist is uploaded to the
752          * hardware with vc4_plane_write_dlist() at CRTC atomic_flush
753          * time.
754          */
755 }
756
757 u32 vc4_plane_write_dlist(struct drm_plane *plane, u32 __iomem *dlist)
758 {
759         struct vc4_plane_state *vc4_state = to_vc4_plane_state(plane->state);
760         int i;
761
762         vc4_state->hw_dlist = dlist;
763
764         /* Can't memcpy_toio() because it needs to be 32-bit writes. */
765         for (i = 0; i < vc4_state->dlist_count; i++)
766                 writel(vc4_state->dlist[i], &dlist[i]);
767
768         return vc4_state->dlist_count;
769 }
770
771 u32 vc4_plane_dlist_size(const struct drm_plane_state *state)
772 {
773         const struct vc4_plane_state *vc4_state =
774                 container_of(state, typeof(*vc4_state), base);
775
776         return vc4_state->dlist_count;
777 }
778
779 /* Updates the plane to immediately (well, once the FIFO needs
780  * refilling) scan out from at a new framebuffer.
781  */
782 void vc4_plane_async_set_fb(struct drm_plane *plane, struct drm_framebuffer *fb)
783 {
784         struct vc4_plane_state *vc4_state = to_vc4_plane_state(plane->state);
785         struct drm_gem_cma_object *bo = drm_fb_cma_get_gem_obj(fb, 0);
786         uint32_t addr;
787
788         /* We're skipping the address adjustment for negative origin,
789          * because this is only called on the primary plane.
790          */
791         WARN_ON_ONCE(plane->state->crtc_x < 0 || plane->state->crtc_y < 0);
792         addr = bo->paddr + fb->offsets[0];
793
794         /* Write the new address into the hardware immediately.  The
795          * scanout will start from this address as soon as the FIFO
796          * needs to refill with pixels.
797          */
798         writel(addr, &vc4_state->hw_dlist[vc4_state->ptr0_offset]);
799
800         /* Also update the CPU-side dlist copy, so that any later
801          * atomic updates that don't do a new modeset on our plane
802          * also use our updated address.
803          */
804         vc4_state->dlist[vc4_state->ptr0_offset] = addr;
805 }
806
807 static void vc4_plane_atomic_async_update(struct drm_plane *plane,
808                                           struct drm_plane_state *state)
809 {
810         struct vc4_plane_state *vc4_state = to_vc4_plane_state(plane->state);
811
812         if (plane->state->fb != state->fb) {
813                 vc4_plane_async_set_fb(plane, state->fb);
814                 drm_atomic_set_fb_for_plane(plane->state, state->fb);
815         }
816
817         /* Set the cursor's position on the screen.  This is the
818          * expected change from the drm_mode_cursor_universal()
819          * helper.
820          */
821         plane->state->crtc_x = state->crtc_x;
822         plane->state->crtc_y = state->crtc_y;
823
824         /* Allow changing the start position within the cursor BO, if
825          * that matters.
826          */
827         plane->state->src_x = state->src_x;
828         plane->state->src_y = state->src_y;
829
830         /* Update the display list based on the new crtc_x/y. */
831         vc4_plane_atomic_check(plane, plane->state);
832
833         /* Note that we can't just call vc4_plane_write_dlist()
834          * because that would smash the context data that the HVS is
835          * currently using.
836          */
837         writel(vc4_state->dlist[vc4_state->pos0_offset],
838                &vc4_state->hw_dlist[vc4_state->pos0_offset]);
839         writel(vc4_state->dlist[vc4_state->pos2_offset],
840                &vc4_state->hw_dlist[vc4_state->pos2_offset]);
841         writel(vc4_state->dlist[vc4_state->ptr0_offset],
842                &vc4_state->hw_dlist[vc4_state->ptr0_offset]);
843 }
844
845 static int vc4_plane_atomic_async_check(struct drm_plane *plane,
846                                         struct drm_plane_state *state)
847 {
848         /* No configuring new scaling in the fast path. */
849         if (plane->state->crtc_w != state->crtc_w ||
850             plane->state->crtc_h != state->crtc_h ||
851             plane->state->src_w != state->src_w ||
852             plane->state->src_h != state->src_h)
853                 return -EINVAL;
854
855         return 0;
856 }
857
858 static int vc4_prepare_fb(struct drm_plane *plane,
859                           struct drm_plane_state *state)
860 {
861         struct vc4_bo *bo;
862         struct dma_fence *fence;
863         int ret;
864
865         if (!state->fb)
866                 return 0;
867
868         bo = to_vc4_bo(&drm_fb_cma_get_gem_obj(state->fb, 0)->base);
869
870         fence = reservation_object_get_excl_rcu(bo->resv);
871         drm_atomic_set_fence_for_plane(state, fence);
872
873         if (plane->state->fb == state->fb)
874                 return 0;
875
876         ret = vc4_bo_inc_usecnt(bo);
877         if (ret)
878                 return ret;
879
880         return 0;
881 }
882
883 static void vc4_cleanup_fb(struct drm_plane *plane,
884                            struct drm_plane_state *state)
885 {
886         struct vc4_bo *bo;
887
888         if (plane->state->fb == state->fb || !state->fb)
889                 return;
890
891         bo = to_vc4_bo(&drm_fb_cma_get_gem_obj(state->fb, 0)->base);
892         vc4_bo_dec_usecnt(bo);
893 }
894
895 static const struct drm_plane_helper_funcs vc4_plane_helper_funcs = {
896         .atomic_check = vc4_plane_atomic_check,
897         .atomic_update = vc4_plane_atomic_update,
898         .prepare_fb = vc4_prepare_fb,
899         .cleanup_fb = vc4_cleanup_fb,
900         .atomic_async_check = vc4_plane_atomic_async_check,
901         .atomic_async_update = vc4_plane_atomic_async_update,
902 };
903
904 static void vc4_plane_destroy(struct drm_plane *plane)
905 {
906         drm_plane_helper_disable(plane, NULL);
907         drm_plane_cleanup(plane);
908 }
909
910 static bool vc4_format_mod_supported(struct drm_plane *plane,
911                                      uint32_t format,
912                                      uint64_t modifier)
913 {
914         /* Support T_TILING for RGB formats only. */
915         switch (format) {
916         case DRM_FORMAT_XRGB8888:
917         case DRM_FORMAT_ARGB8888:
918         case DRM_FORMAT_ABGR8888:
919         case DRM_FORMAT_XBGR8888:
920         case DRM_FORMAT_RGB565:
921         case DRM_FORMAT_BGR565:
922         case DRM_FORMAT_ARGB1555:
923         case DRM_FORMAT_XRGB1555:
924                 switch (fourcc_mod_broadcom_mod(modifier)) {
925                 case DRM_FORMAT_MOD_LINEAR:
926                 case DRM_FORMAT_MOD_BROADCOM_VC4_T_TILED:
927                 case DRM_FORMAT_MOD_BROADCOM_SAND64:
928                 case DRM_FORMAT_MOD_BROADCOM_SAND128:
929                         return true;
930                 default:
931                         return false;
932                 }
933         case DRM_FORMAT_NV12:
934         case DRM_FORMAT_NV21:
935                 switch (fourcc_mod_broadcom_mod(modifier)) {
936                 case DRM_FORMAT_MOD_LINEAR:
937                 case DRM_FORMAT_MOD_BROADCOM_SAND64:
938                 case DRM_FORMAT_MOD_BROADCOM_SAND128:
939                 case DRM_FORMAT_MOD_BROADCOM_SAND256:
940                         return true;
941                 default:
942                         return false;
943                 }
944         case DRM_FORMAT_YUV422:
945         case DRM_FORMAT_YVU422:
946         case DRM_FORMAT_YUV420:
947         case DRM_FORMAT_YVU420:
948         case DRM_FORMAT_NV16:
949         case DRM_FORMAT_NV61:
950         default:
951                 return (modifier == DRM_FORMAT_MOD_LINEAR);
952         }
953 }
954
955 static const struct drm_plane_funcs vc4_plane_funcs = {
956         .update_plane = drm_atomic_helper_update_plane,
957         .disable_plane = drm_atomic_helper_disable_plane,
958         .destroy = vc4_plane_destroy,
959         .set_property = NULL,
960         .reset = vc4_plane_reset,
961         .atomic_duplicate_state = vc4_plane_duplicate_state,
962         .atomic_destroy_state = vc4_plane_destroy_state,
963         .format_mod_supported = vc4_format_mod_supported,
964 };
965
966 struct drm_plane *vc4_plane_init(struct drm_device *dev,
967                                  enum drm_plane_type type)
968 {
969         struct drm_plane *plane = NULL;
970         struct vc4_plane *vc4_plane;
971         u32 formats[ARRAY_SIZE(hvs_formats)];
972         u32 num_formats = 0;
973         int ret = 0;
974         unsigned i;
975         static const uint64_t modifiers[] = {
976                 DRM_FORMAT_MOD_BROADCOM_VC4_T_TILED,
977                 DRM_FORMAT_MOD_BROADCOM_SAND128,
978                 DRM_FORMAT_MOD_BROADCOM_SAND64,
979                 DRM_FORMAT_MOD_BROADCOM_SAND256,
980                 DRM_FORMAT_MOD_LINEAR,
981                 DRM_FORMAT_MOD_INVALID
982         };
983
984         vc4_plane = devm_kzalloc(dev->dev, sizeof(*vc4_plane),
985                                  GFP_KERNEL);
986         if (!vc4_plane)
987                 return ERR_PTR(-ENOMEM);
988
989         for (i = 0; i < ARRAY_SIZE(hvs_formats); i++) {
990                 /* Don't allow YUV in cursor planes, since that means
991                  * tuning on the scaler, which we don't allow for the
992                  * cursor.
993                  */
994                 if (type != DRM_PLANE_TYPE_CURSOR ||
995                     hvs_formats[i].hvs < HVS_PIXEL_FORMAT_YCBCR_YUV420_3PLANE) {
996                         formats[num_formats++] = hvs_formats[i].drm;
997                 }
998         }
999         plane = &vc4_plane->base;
1000         ret = drm_universal_plane_init(dev, plane, 0,
1001                                        &vc4_plane_funcs,
1002                                        formats, num_formats,
1003                                        modifiers, type, NULL);
1004
1005         drm_plane_helper_add(plane, &vc4_plane_helper_funcs);
1006
1007         drm_plane_create_alpha_property(plane);
1008
1009         return plane;
1010 }