ab39315c9078bb40967b6e2bcd50f3f6d2e18be0
[platform/kernel/linux-rpi.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         plane->state = &vc4_state->base;
204         plane->state->alpha = DRM_BLEND_ALPHA_OPAQUE;
205         vc4_state->base.plane = plane;
206 }
207
208 static void vc4_dlist_write(struct vc4_plane_state *vc4_state, u32 val)
209 {
210         if (vc4_state->dlist_count == vc4_state->dlist_size) {
211                 u32 new_size = max(4u, vc4_state->dlist_count * 2);
212                 u32 *new_dlist = kmalloc_array(new_size, 4, GFP_KERNEL);
213
214                 if (!new_dlist)
215                         return;
216                 memcpy(new_dlist, vc4_state->dlist, vc4_state->dlist_count * 4);
217
218                 kfree(vc4_state->dlist);
219                 vc4_state->dlist = new_dlist;
220                 vc4_state->dlist_size = new_size;
221         }
222
223         vc4_state->dlist[vc4_state->dlist_count++] = val;
224 }
225
226 /* Returns the scl0/scl1 field based on whether the dimensions need to
227  * be up/down/non-scaled.
228  *
229  * This is a replication of a table from the spec.
230  */
231 static u32 vc4_get_scl_field(struct drm_plane_state *state, int plane)
232 {
233         struct vc4_plane_state *vc4_state = to_vc4_plane_state(state);
234
235         switch (vc4_state->x_scaling[plane] << 2 | vc4_state->y_scaling[plane]) {
236         case VC4_SCALING_PPF << 2 | VC4_SCALING_PPF:
237                 return SCALER_CTL0_SCL_H_PPF_V_PPF;
238         case VC4_SCALING_TPZ << 2 | VC4_SCALING_PPF:
239                 return SCALER_CTL0_SCL_H_TPZ_V_PPF;
240         case VC4_SCALING_PPF << 2 | VC4_SCALING_TPZ:
241                 return SCALER_CTL0_SCL_H_PPF_V_TPZ;
242         case VC4_SCALING_TPZ << 2 | VC4_SCALING_TPZ:
243                 return SCALER_CTL0_SCL_H_TPZ_V_TPZ;
244         case VC4_SCALING_PPF << 2 | VC4_SCALING_NONE:
245                 return SCALER_CTL0_SCL_H_PPF_V_NONE;
246         case VC4_SCALING_NONE << 2 | VC4_SCALING_PPF:
247                 return SCALER_CTL0_SCL_H_NONE_V_PPF;
248         case VC4_SCALING_NONE << 2 | VC4_SCALING_TPZ:
249                 return SCALER_CTL0_SCL_H_NONE_V_TPZ;
250         case VC4_SCALING_TPZ << 2 | VC4_SCALING_NONE:
251                 return SCALER_CTL0_SCL_H_TPZ_V_NONE;
252         default:
253         case VC4_SCALING_NONE << 2 | VC4_SCALING_NONE:
254                 /* The unity case is independently handled by
255                  * SCALER_CTL0_UNITY.
256                  */
257                 return 0;
258         }
259 }
260
261 static int vc4_plane_setup_clipping_and_scaling(struct drm_plane_state *state)
262 {
263         struct drm_plane *plane = state->plane;
264         struct vc4_plane_state *vc4_state = to_vc4_plane_state(state);
265         struct drm_framebuffer *fb = state->fb;
266         struct drm_gem_cma_object *bo = drm_fb_cma_get_gem_obj(fb, 0);
267         u32 subpixel_src_mask = (1 << 16) - 1;
268         u32 format = fb->format->format;
269         int num_planes = fb->format->num_planes;
270         u32 h_subsample = 1;
271         u32 v_subsample = 1;
272         int i;
273
274         for (i = 0; i < num_planes; i++)
275                 vc4_state->offsets[i] = bo->paddr + fb->offsets[i];
276
277         /* We don't support subpixel source positioning for scaling. */
278         if ((state->src_x & subpixel_src_mask) ||
279             (state->src_y & subpixel_src_mask) ||
280             (state->src_w & subpixel_src_mask) ||
281             (state->src_h & subpixel_src_mask)) {
282                 return -EINVAL;
283         }
284
285         vc4_state->src_x = state->src_x >> 16;
286         vc4_state->src_y = state->src_y >> 16;
287         vc4_state->src_w[0] = state->src_w >> 16;
288         vc4_state->src_h[0] = state->src_h >> 16;
289
290         vc4_state->crtc_x = state->crtc_x;
291         vc4_state->crtc_y = state->crtc_y;
292         vc4_state->crtc_w = state->crtc_w;
293         vc4_state->crtc_h = state->crtc_h;
294
295         vc4_state->x_scaling[0] = vc4_get_scaling_mode(vc4_state->src_w[0],
296                                                        vc4_state->crtc_w);
297         vc4_state->y_scaling[0] = vc4_get_scaling_mode(vc4_state->src_h[0],
298                                                        vc4_state->crtc_h);
299
300         vc4_state->is_unity = (vc4_state->x_scaling[0] == VC4_SCALING_NONE &&
301                                vc4_state->y_scaling[0] == VC4_SCALING_NONE);
302
303         if (num_planes > 1) {
304                 vc4_state->is_yuv = true;
305
306                 h_subsample = drm_format_horz_chroma_subsampling(format);
307                 v_subsample = drm_format_vert_chroma_subsampling(format);
308                 vc4_state->src_w[1] = vc4_state->src_w[0] / h_subsample;
309                 vc4_state->src_h[1] = vc4_state->src_h[0] / v_subsample;
310
311                 vc4_state->x_scaling[1] =
312                         vc4_get_scaling_mode(vc4_state->src_w[1],
313                                              vc4_state->crtc_w);
314                 vc4_state->y_scaling[1] =
315                         vc4_get_scaling_mode(vc4_state->src_h[1],
316                                              vc4_state->crtc_h);
317
318                 /* YUV conversion requires that horizontal scaling be enabled
319                  * on the UV plane even if vc4_get_scaling_mode() returned
320                  * VC4_SCALING_NONE (which can happen when the down-scaling
321                  * ratio is 0.5). Let's force it to VC4_SCALING_PPF in this
322                  * case.
323                  */
324                 if (vc4_state->x_scaling[1] == VC4_SCALING_NONE)
325                         vc4_state->x_scaling[1] = VC4_SCALING_PPF;
326         } else {
327                 vc4_state->is_yuv = false;
328                 vc4_state->x_scaling[1] = VC4_SCALING_NONE;
329                 vc4_state->y_scaling[1] = VC4_SCALING_NONE;
330         }
331
332         /* No configuring scaling on the cursor plane, since it gets
333            non-vblank-synced updates, and scaling requires requires
334            LBM changes which have to be vblank-synced.
335          */
336         if (plane->type == DRM_PLANE_TYPE_CURSOR && !vc4_state->is_unity)
337                 return -EINVAL;
338
339         /* Clamp the on-screen start x/y to 0.  The hardware doesn't
340          * support negative y, and negative x wastes bandwidth.
341          */
342         if (vc4_state->crtc_x < 0) {
343                 for (i = 0; i < num_planes; i++) {
344                         u32 cpp = fb->format->cpp[i];
345                         u32 subs = ((i == 0) ? 1 : h_subsample);
346
347                         vc4_state->offsets[i] += (cpp *
348                                                   (-vc4_state->crtc_x) / subs);
349                 }
350                 vc4_state->src_w[0] += vc4_state->crtc_x;
351                 vc4_state->src_w[1] += vc4_state->crtc_x / h_subsample;
352                 vc4_state->crtc_x = 0;
353         }
354
355         if (vc4_state->crtc_y < 0) {
356                 for (i = 0; i < num_planes; i++) {
357                         u32 subs = ((i == 0) ? 1 : v_subsample);
358
359                         vc4_state->offsets[i] += (fb->pitches[i] *
360                                                   (-vc4_state->crtc_y) / subs);
361                 }
362                 vc4_state->src_h[0] += vc4_state->crtc_y;
363                 vc4_state->src_h[1] += vc4_state->crtc_y / v_subsample;
364                 vc4_state->crtc_y = 0;
365         }
366
367         return 0;
368 }
369
370 static void vc4_write_tpz(struct vc4_plane_state *vc4_state, u32 src, u32 dst)
371 {
372         u32 scale, recip;
373
374         scale = (1 << 16) * src / dst;
375
376         /* The specs note that while the reciprocal would be defined
377          * as (1<<32)/scale, ~0 is close enough.
378          */
379         recip = ~0 / scale;
380
381         vc4_dlist_write(vc4_state,
382                         VC4_SET_FIELD(scale, SCALER_TPZ0_SCALE) |
383                         VC4_SET_FIELD(0, SCALER_TPZ0_IPHASE));
384         vc4_dlist_write(vc4_state,
385                         VC4_SET_FIELD(recip, SCALER_TPZ1_RECIP));
386 }
387
388 static void vc4_write_ppf(struct vc4_plane_state *vc4_state, u32 src, u32 dst)
389 {
390         u32 scale = (1 << 16) * src / dst;
391
392         vc4_dlist_write(vc4_state,
393                         SCALER_PPF_AGC |
394                         VC4_SET_FIELD(scale, SCALER_PPF_SCALE) |
395                         VC4_SET_FIELD(0, SCALER_PPF_IPHASE));
396 }
397
398 static u32 vc4_lbm_size(struct drm_plane_state *state)
399 {
400         struct vc4_plane_state *vc4_state = to_vc4_plane_state(state);
401         /* This is the worst case number.  One of the two sizes will
402          * be used depending on the scaling configuration.
403          */
404         u32 pix_per_line = max(vc4_state->src_w[0], (u32)vc4_state->crtc_w);
405         u32 lbm;
406
407         if (!vc4_state->is_yuv) {
408                 if (vc4_state->is_unity)
409                         return 0;
410                 else if (vc4_state->y_scaling[0] == VC4_SCALING_TPZ)
411                         lbm = pix_per_line * 8;
412                 else {
413                         /* In special cases, this multiplier might be 12. */
414                         lbm = pix_per_line * 16;
415                 }
416         } else {
417                 /* There are cases for this going down to a multiplier
418                  * of 2, but according to the firmware source, the
419                  * table in the docs is somewhat wrong.
420                  */
421                 lbm = pix_per_line * 16;
422         }
423
424         lbm = roundup(lbm, 32);
425
426         return lbm;
427 }
428
429 static void vc4_write_scaling_parameters(struct drm_plane_state *state,
430                                          int channel)
431 {
432         struct vc4_plane_state *vc4_state = to_vc4_plane_state(state);
433
434         /* Ch0 H-PPF Word 0: Scaling Parameters */
435         if (vc4_state->x_scaling[channel] == VC4_SCALING_PPF) {
436                 vc4_write_ppf(vc4_state,
437                               vc4_state->src_w[channel], vc4_state->crtc_w);
438         }
439
440         /* Ch0 V-PPF Words 0-1: Scaling Parameters, Context */
441         if (vc4_state->y_scaling[channel] == VC4_SCALING_PPF) {
442                 vc4_write_ppf(vc4_state,
443                               vc4_state->src_h[channel], vc4_state->crtc_h);
444                 vc4_dlist_write(vc4_state, 0xc0c0c0c0);
445         }
446
447         /* Ch0 H-TPZ Words 0-1: Scaling Parameters, Recip */
448         if (vc4_state->x_scaling[channel] == VC4_SCALING_TPZ) {
449                 vc4_write_tpz(vc4_state,
450                               vc4_state->src_w[channel], vc4_state->crtc_w);
451         }
452
453         /* Ch0 V-TPZ Words 0-2: Scaling Parameters, Recip, Context */
454         if (vc4_state->y_scaling[channel] == VC4_SCALING_TPZ) {
455                 vc4_write_tpz(vc4_state,
456                               vc4_state->src_h[channel], vc4_state->crtc_h);
457                 vc4_dlist_write(vc4_state, 0xc0c0c0c0);
458         }
459 }
460
461 /* Writes out a full display list for an active plane to the plane's
462  * private dlist state.
463  */
464 static int vc4_plane_mode_set(struct drm_plane *plane,
465                               struct drm_plane_state *state)
466 {
467         struct vc4_dev *vc4 = to_vc4_dev(plane->dev);
468         struct vc4_plane_state *vc4_state = to_vc4_plane_state(state);
469         struct drm_framebuffer *fb = state->fb;
470         u32 ctl0_offset = vc4_state->dlist_count;
471         const struct hvs_format *format = vc4_get_hvs_format(fb->format->format);
472         u64 base_format_mod = fourcc_mod_broadcom_mod(fb->modifier);
473         int num_planes = drm_format_num_planes(format->drm);
474         bool mix_plane_alpha;
475         bool covers_screen;
476         u32 scl0, scl1, pitch0;
477         u32 lbm_size, tiling;
478         unsigned long irqflags;
479         u32 hvs_format = format->hvs;
480         int ret, i;
481
482         ret = vc4_plane_setup_clipping_and_scaling(state);
483         if (ret)
484                 return ret;
485
486         /* Allocate the LBM memory that the HVS will use for temporary
487          * storage due to our scaling/format conversion.
488          */
489         lbm_size = vc4_lbm_size(state);
490         if (lbm_size) {
491                 if (!vc4_state->lbm.allocated) {
492                         spin_lock_irqsave(&vc4->hvs->mm_lock, irqflags);
493                         ret = drm_mm_insert_node_generic(&vc4->hvs->lbm_mm,
494                                                          &vc4_state->lbm,
495                                                          lbm_size, 32, 0, 0);
496                         spin_unlock_irqrestore(&vc4->hvs->mm_lock, irqflags);
497                 } else {
498                         WARN_ON_ONCE(lbm_size != vc4_state->lbm.size);
499                 }
500         }
501
502         if (ret)
503                 return ret;
504
505         /* SCL1 is used for Cb/Cr scaling of planar formats.  For RGB
506          * and 4:4:4, scl1 should be set to scl0 so both channels of
507          * the scaler do the same thing.  For YUV, the Y plane needs
508          * to be put in channel 1 and Cb/Cr in channel 0, so we swap
509          * the scl fields here.
510          */
511         if (num_planes == 1) {
512                 scl0 = vc4_get_scl_field(state, 0);
513                 scl1 = scl0;
514         } else {
515                 scl0 = vc4_get_scl_field(state, 1);
516                 scl1 = vc4_get_scl_field(state, 0);
517         }
518
519         switch (base_format_mod) {
520         case DRM_FORMAT_MOD_LINEAR:
521                 tiling = SCALER_CTL0_TILING_LINEAR;
522                 pitch0 = VC4_SET_FIELD(fb->pitches[0], SCALER_SRC_PITCH);
523                 break;
524
525         case DRM_FORMAT_MOD_BROADCOM_VC4_T_TILED: {
526                 /* For T-tiled, the FB pitch is "how many bytes from
527                  * one row to the next, such that pitch * tile_h ==
528                  * tile_size * tiles_per_row."
529                  */
530                 u32 tile_size_shift = 12; /* T tiles are 4kb */
531                 u32 tile_h_shift = 5; /* 16 and 32bpp are 32 pixels high */
532                 u32 tiles_w = fb->pitches[0] >> (tile_size_shift - tile_h_shift);
533
534                 tiling = SCALER_CTL0_TILING_256B_OR_T;
535
536                 pitch0 = (VC4_SET_FIELD(0, SCALER_PITCH0_TILE_Y_OFFSET) |
537                           VC4_SET_FIELD(0, SCALER_PITCH0_TILE_WIDTH_L) |
538                           VC4_SET_FIELD(tiles_w, SCALER_PITCH0_TILE_WIDTH_R));
539                 break;
540         }
541
542         case DRM_FORMAT_MOD_BROADCOM_SAND64:
543         case DRM_FORMAT_MOD_BROADCOM_SAND128:
544         case DRM_FORMAT_MOD_BROADCOM_SAND256: {
545                 uint32_t param = fourcc_mod_broadcom_param(fb->modifier);
546
547                 /* Column-based NV12 or RGBA.
548                  */
549                 if (fb->format->num_planes > 1) {
550                         if (hvs_format != HVS_PIXEL_FORMAT_YCBCR_YUV420_2PLANE) {
551                                 DRM_DEBUG_KMS("SAND format only valid for NV12/21");
552                                 return -EINVAL;
553                         }
554                         hvs_format = HVS_PIXEL_FORMAT_H264;
555                 } else {
556                         if (base_format_mod == DRM_FORMAT_MOD_BROADCOM_SAND256) {
557                                 DRM_DEBUG_KMS("SAND256 format only valid for H.264");
558                                 return -EINVAL;
559                         }
560                 }
561
562                 switch (base_format_mod) {
563                 case DRM_FORMAT_MOD_BROADCOM_SAND64:
564                         tiling = SCALER_CTL0_TILING_64B;
565                         break;
566                 case DRM_FORMAT_MOD_BROADCOM_SAND128:
567                         tiling = SCALER_CTL0_TILING_128B;
568                         break;
569                 case DRM_FORMAT_MOD_BROADCOM_SAND256:
570                         tiling = SCALER_CTL0_TILING_256B_OR_T;
571                         break;
572                 default:
573                         break;
574                 }
575
576                 if (param > SCALER_TILE_HEIGHT_MASK) {
577                         DRM_DEBUG_KMS("SAND height too large (%d)\n", param);
578                         return -EINVAL;
579                 }
580
581                 pitch0 = VC4_SET_FIELD(param, SCALER_TILE_HEIGHT);
582                 break;
583         }
584
585         default:
586                 DRM_DEBUG_KMS("Unsupported FB tiling flag 0x%16llx",
587                               (long long)fb->modifier);
588                 return -EINVAL;
589         }
590
591         /* Control word */
592         vc4_dlist_write(vc4_state,
593                         SCALER_CTL0_VALID |
594                         VC4_SET_FIELD(SCALER_CTL0_RGBA_EXPAND_ROUND, SCALER_CTL0_RGBA_EXPAND) |
595                         (format->pixel_order << SCALER_CTL0_ORDER_SHIFT) |
596                         (hvs_format << SCALER_CTL0_PIXEL_FORMAT_SHIFT) |
597                         VC4_SET_FIELD(tiling, SCALER_CTL0_TILING) |
598                         (vc4_state->is_unity ? SCALER_CTL0_UNITY : 0) |
599                         VC4_SET_FIELD(scl0, SCALER_CTL0_SCL0) |
600                         VC4_SET_FIELD(scl1, SCALER_CTL0_SCL1));
601
602         /* Position Word 0: Image Positions and Alpha Value */
603         vc4_state->pos0_offset = vc4_state->dlist_count;
604         vc4_dlist_write(vc4_state,
605                         VC4_SET_FIELD(state->alpha >> 8, SCALER_POS0_FIXED_ALPHA) |
606                         VC4_SET_FIELD(vc4_state->crtc_x, SCALER_POS0_START_X) |
607                         VC4_SET_FIELD(vc4_state->crtc_y, SCALER_POS0_START_Y));
608
609         /* Position Word 1: Scaled Image Dimensions. */
610         if (!vc4_state->is_unity) {
611                 vc4_dlist_write(vc4_state,
612                                 VC4_SET_FIELD(vc4_state->crtc_w,
613                                               SCALER_POS1_SCL_WIDTH) |
614                                 VC4_SET_FIELD(vc4_state->crtc_h,
615                                               SCALER_POS1_SCL_HEIGHT));
616         }
617
618         /* Don't waste cycles mixing with plane alpha if the set alpha
619          * is opaque or there is no per-pixel alpha information.
620          * In any case we use the alpha property value as the fixed alpha.
621          */
622         mix_plane_alpha = state->alpha != DRM_BLEND_ALPHA_OPAQUE &&
623                           fb->format->has_alpha;
624
625         /* Position Word 2: Source Image Size, Alpha */
626         vc4_state->pos2_offset = vc4_state->dlist_count;
627         vc4_dlist_write(vc4_state,
628                         VC4_SET_FIELD(fb->format->has_alpha ?
629                                       SCALER_POS2_ALPHA_MODE_PIPELINE :
630                                       SCALER_POS2_ALPHA_MODE_FIXED,
631                                       SCALER_POS2_ALPHA_MODE) |
632                         (mix_plane_alpha ? SCALER_POS2_ALPHA_MIX : 0) |
633                         (fb->format->has_alpha ? SCALER_POS2_ALPHA_PREMULT : 0) |
634                         VC4_SET_FIELD(vc4_state->src_w[0], SCALER_POS2_WIDTH) |
635                         VC4_SET_FIELD(vc4_state->src_h[0], SCALER_POS2_HEIGHT));
636
637         /* Position Word 3: Context.  Written by the HVS. */
638         vc4_dlist_write(vc4_state, 0xc0c0c0c0);
639
640
641         /* Pointer Word 0/1/2: RGB / Y / Cb / Cr Pointers
642          *
643          * The pointers may be any byte address.
644          */
645         vc4_state->ptr0_offset = vc4_state->dlist_count;
646         for (i = 0; i < num_planes; i++)
647                 vc4_dlist_write(vc4_state, vc4_state->offsets[i]);
648
649         /* Pointer Context Word 0/1/2: Written by the HVS */
650         for (i = 0; i < num_planes; i++)
651                 vc4_dlist_write(vc4_state, 0xc0c0c0c0);
652
653         /* Pitch word 0 */
654         vc4_dlist_write(vc4_state, pitch0);
655
656         /* Pitch word 1/2 */
657         for (i = 1; i < num_planes; i++) {
658                 if (hvs_format != HVS_PIXEL_FORMAT_H264) {
659                         vc4_dlist_write(vc4_state,
660                                         VC4_SET_FIELD(fb->pitches[i],
661                                                       SCALER_SRC_PITCH));
662                 } else {
663                         vc4_dlist_write(vc4_state, pitch0);
664                 }
665         }
666
667         /* Colorspace conversion words */
668         if (vc4_state->is_yuv) {
669                 vc4_dlist_write(vc4_state, SCALER_CSC0_ITR_R_601_5);
670                 vc4_dlist_write(vc4_state, SCALER_CSC1_ITR_R_601_5);
671                 vc4_dlist_write(vc4_state, SCALER_CSC2_ITR_R_601_5);
672         }
673
674         if (vc4_state->x_scaling[0] != VC4_SCALING_NONE ||
675             vc4_state->x_scaling[1] != VC4_SCALING_NONE ||
676             vc4_state->y_scaling[0] != VC4_SCALING_NONE ||
677             vc4_state->y_scaling[1] != VC4_SCALING_NONE) {
678                 /* LBM Base Address. */
679                 if (vc4_state->y_scaling[0] != VC4_SCALING_NONE ||
680                     vc4_state->y_scaling[1] != VC4_SCALING_NONE) {
681                         vc4_dlist_write(vc4_state, vc4_state->lbm.start);
682                 }
683
684                 if (num_planes > 1) {
685                         /* Emit Cb/Cr as channel 0 and Y as channel
686                          * 1. This matches how we set up scl0/scl1
687                          * above.
688                          */
689                         vc4_write_scaling_parameters(state, 1);
690                 }
691                 vc4_write_scaling_parameters(state, 0);
692
693                 /* If any PPF setup was done, then all the kernel
694                  * pointers get uploaded.
695                  */
696                 if (vc4_state->x_scaling[0] == VC4_SCALING_PPF ||
697                     vc4_state->y_scaling[0] == VC4_SCALING_PPF ||
698                     vc4_state->x_scaling[1] == VC4_SCALING_PPF ||
699                     vc4_state->y_scaling[1] == VC4_SCALING_PPF) {
700                         u32 kernel = VC4_SET_FIELD(vc4->hvs->mitchell_netravali_filter.start,
701                                                    SCALER_PPF_KERNEL_OFFSET);
702
703                         /* HPPF plane 0 */
704                         vc4_dlist_write(vc4_state, kernel);
705                         /* VPPF plane 0 */
706                         vc4_dlist_write(vc4_state, kernel);
707                         /* HPPF plane 1 */
708                         vc4_dlist_write(vc4_state, kernel);
709                         /* VPPF plane 1 */
710                         vc4_dlist_write(vc4_state, kernel);
711                 }
712         }
713
714         vc4_state->dlist[ctl0_offset] |=
715                 VC4_SET_FIELD(vc4_state->dlist_count, SCALER_CTL0_SIZE);
716
717         /* crtc_* are already clipped coordinates. */
718         covers_screen = vc4_state->crtc_x == 0 && vc4_state->crtc_y == 0 &&
719                         vc4_state->crtc_w == state->crtc->mode.hdisplay &&
720                         vc4_state->crtc_h == state->crtc->mode.vdisplay;
721         /* Background fill might be necessary when the plane has per-pixel
722          * alpha content or a non-opaque plane alpha and could blend from the
723          * background or does not cover the entire screen.
724          */
725         vc4_state->needs_bg_fill = fb->format->has_alpha || !covers_screen ||
726                                    state->alpha != DRM_BLEND_ALPHA_OPAQUE;
727
728         return 0;
729 }
730
731 /* If a modeset involves changing the setup of a plane, the atomic
732  * infrastructure will call this to validate a proposed plane setup.
733  * However, if a plane isn't getting updated, this (and the
734  * corresponding vc4_plane_atomic_update) won't get called.  Thus, we
735  * compute the dlist here and have all active plane dlists get updated
736  * in the CRTC's flush.
737  */
738 static int vc4_plane_atomic_check(struct drm_plane *plane,
739                                   struct drm_plane_state *state)
740 {
741         struct vc4_plane_state *vc4_state = to_vc4_plane_state(state);
742
743         vc4_state->dlist_count = 0;
744
745         if (plane_enabled(state))
746                 return vc4_plane_mode_set(plane, state);
747         else
748                 return 0;
749 }
750
751 static void vc4_plane_atomic_update(struct drm_plane *plane,
752                                     struct drm_plane_state *old_state)
753 {
754         /* No contents here.  Since we don't know where in the CRTC's
755          * dlist we should be stored, our dlist is uploaded to the
756          * hardware with vc4_plane_write_dlist() at CRTC atomic_flush
757          * time.
758          */
759 }
760
761 u32 vc4_plane_write_dlist(struct drm_plane *plane, u32 __iomem *dlist)
762 {
763         struct vc4_plane_state *vc4_state = to_vc4_plane_state(plane->state);
764         int i;
765
766         vc4_state->hw_dlist = dlist;
767
768         /* Can't memcpy_toio() because it needs to be 32-bit writes. */
769         for (i = 0; i < vc4_state->dlist_count; i++)
770                 writel(vc4_state->dlist[i], &dlist[i]);
771
772         return vc4_state->dlist_count;
773 }
774
775 u32 vc4_plane_dlist_size(const struct drm_plane_state *state)
776 {
777         const struct vc4_plane_state *vc4_state =
778                 container_of(state, typeof(*vc4_state), base);
779
780         return vc4_state->dlist_count;
781 }
782
783 /* Updates the plane to immediately (well, once the FIFO needs
784  * refilling) scan out from at a new framebuffer.
785  */
786 void vc4_plane_async_set_fb(struct drm_plane *plane, struct drm_framebuffer *fb)
787 {
788         struct vc4_plane_state *vc4_state = to_vc4_plane_state(plane->state);
789         struct drm_gem_cma_object *bo = drm_fb_cma_get_gem_obj(fb, 0);
790         uint32_t addr;
791
792         /* We're skipping the address adjustment for negative origin,
793          * because this is only called on the primary plane.
794          */
795         WARN_ON_ONCE(plane->state->crtc_x < 0 || plane->state->crtc_y < 0);
796         addr = bo->paddr + fb->offsets[0];
797
798         /* Write the new address into the hardware immediately.  The
799          * scanout will start from this address as soon as the FIFO
800          * needs to refill with pixels.
801          */
802         writel(addr, &vc4_state->hw_dlist[vc4_state->ptr0_offset]);
803
804         /* Also update the CPU-side dlist copy, so that any later
805          * atomic updates that don't do a new modeset on our plane
806          * also use our updated address.
807          */
808         vc4_state->dlist[vc4_state->ptr0_offset] = addr;
809 }
810
811 static void vc4_plane_atomic_async_update(struct drm_plane *plane,
812                                           struct drm_plane_state *state)
813 {
814         struct vc4_plane_state *vc4_state = to_vc4_plane_state(plane->state);
815
816         if (plane->state->fb != state->fb) {
817                 vc4_plane_async_set_fb(plane, state->fb);
818                 drm_atomic_set_fb_for_plane(plane->state, state->fb);
819         }
820
821         /* Set the cursor's position on the screen.  This is the
822          * expected change from the drm_mode_cursor_universal()
823          * helper.
824          */
825         plane->state->crtc_x = state->crtc_x;
826         plane->state->crtc_y = state->crtc_y;
827
828         /* Allow changing the start position within the cursor BO, if
829          * that matters.
830          */
831         plane->state->src_x = state->src_x;
832         plane->state->src_y = state->src_y;
833
834         /* Update the display list based on the new crtc_x/y. */
835         vc4_plane_atomic_check(plane, plane->state);
836
837         /* Note that we can't just call vc4_plane_write_dlist()
838          * because that would smash the context data that the HVS is
839          * currently using.
840          */
841         writel(vc4_state->dlist[vc4_state->pos0_offset],
842                &vc4_state->hw_dlist[vc4_state->pos0_offset]);
843         writel(vc4_state->dlist[vc4_state->pos2_offset],
844                &vc4_state->hw_dlist[vc4_state->pos2_offset]);
845         writel(vc4_state->dlist[vc4_state->ptr0_offset],
846                &vc4_state->hw_dlist[vc4_state->ptr0_offset]);
847 }
848
849 static int vc4_plane_atomic_async_check(struct drm_plane *plane,
850                                         struct drm_plane_state *state)
851 {
852         /* No configuring new scaling in the fast path. */
853         if (plane->state->crtc_w != state->crtc_w ||
854             plane->state->crtc_h != state->crtc_h ||
855             plane->state->src_w != state->src_w ||
856             plane->state->src_h != state->src_h)
857                 return -EINVAL;
858
859         return 0;
860 }
861
862 static int vc4_prepare_fb(struct drm_plane *plane,
863                           struct drm_plane_state *state)
864 {
865         struct vc4_bo *bo;
866         struct dma_fence *fence;
867         int ret;
868
869         if (!state->fb)
870                 return 0;
871
872         bo = to_vc4_bo(&drm_fb_cma_get_gem_obj(state->fb, 0)->base);
873
874         fence = reservation_object_get_excl_rcu(bo->resv);
875         drm_atomic_set_fence_for_plane(state, fence);
876
877         if (plane->state->fb == state->fb)
878                 return 0;
879
880         ret = vc4_bo_inc_usecnt(bo);
881         if (ret)
882                 return ret;
883
884         return 0;
885 }
886
887 static void vc4_cleanup_fb(struct drm_plane *plane,
888                            struct drm_plane_state *state)
889 {
890         struct vc4_bo *bo;
891
892         if (plane->state->fb == state->fb || !state->fb)
893                 return;
894
895         bo = to_vc4_bo(&drm_fb_cma_get_gem_obj(state->fb, 0)->base);
896         vc4_bo_dec_usecnt(bo);
897 }
898
899 static const struct drm_plane_helper_funcs vc4_plane_helper_funcs = {
900         .atomic_check = vc4_plane_atomic_check,
901         .atomic_update = vc4_plane_atomic_update,
902         .prepare_fb = vc4_prepare_fb,
903         .cleanup_fb = vc4_cleanup_fb,
904         .atomic_async_check = vc4_plane_atomic_async_check,
905         .atomic_async_update = vc4_plane_atomic_async_update,
906 };
907
908 static void vc4_plane_destroy(struct drm_plane *plane)
909 {
910         drm_plane_helper_disable(plane, NULL);
911         drm_plane_cleanup(plane);
912 }
913
914 static bool vc4_format_mod_supported(struct drm_plane *plane,
915                                      uint32_t format,
916                                      uint64_t modifier)
917 {
918         /* Support T_TILING for RGB formats only. */
919         switch (format) {
920         case DRM_FORMAT_XRGB8888:
921         case DRM_FORMAT_ARGB8888:
922         case DRM_FORMAT_ABGR8888:
923         case DRM_FORMAT_XBGR8888:
924         case DRM_FORMAT_RGB565:
925         case DRM_FORMAT_BGR565:
926         case DRM_FORMAT_ARGB1555:
927         case DRM_FORMAT_XRGB1555:
928                 switch (fourcc_mod_broadcom_mod(modifier)) {
929                 case DRM_FORMAT_MOD_LINEAR:
930                 case DRM_FORMAT_MOD_BROADCOM_VC4_T_TILED:
931                 case DRM_FORMAT_MOD_BROADCOM_SAND64:
932                 case DRM_FORMAT_MOD_BROADCOM_SAND128:
933                         return true;
934                 default:
935                         return false;
936                 }
937         case DRM_FORMAT_NV12:
938         case DRM_FORMAT_NV21:
939                 switch (fourcc_mod_broadcom_mod(modifier)) {
940                 case DRM_FORMAT_MOD_LINEAR:
941                 case DRM_FORMAT_MOD_BROADCOM_SAND64:
942                 case DRM_FORMAT_MOD_BROADCOM_SAND128:
943                 case DRM_FORMAT_MOD_BROADCOM_SAND256:
944                         return true;
945                 default:
946                         return false;
947                 }
948         case DRM_FORMAT_YUV422:
949         case DRM_FORMAT_YVU422:
950         case DRM_FORMAT_YUV420:
951         case DRM_FORMAT_YVU420:
952         case DRM_FORMAT_NV16:
953         case DRM_FORMAT_NV61:
954         default:
955                 return (modifier == DRM_FORMAT_MOD_LINEAR);
956         }
957 }
958
959 static const struct drm_plane_funcs vc4_plane_funcs = {
960         .update_plane = drm_atomic_helper_update_plane,
961         .disable_plane = drm_atomic_helper_disable_plane,
962         .destroy = vc4_plane_destroy,
963         .set_property = NULL,
964         .reset = vc4_plane_reset,
965         .atomic_duplicate_state = vc4_plane_duplicate_state,
966         .atomic_destroy_state = vc4_plane_destroy_state,
967         .format_mod_supported = vc4_format_mod_supported,
968 };
969
970 struct drm_plane *vc4_plane_init(struct drm_device *dev,
971                                  enum drm_plane_type type)
972 {
973         struct drm_plane *plane = NULL;
974         struct vc4_plane *vc4_plane;
975         u32 formats[ARRAY_SIZE(hvs_formats)];
976         u32 num_formats = 0;
977         int ret = 0;
978         unsigned i;
979         static const uint64_t modifiers[] = {
980                 DRM_FORMAT_MOD_BROADCOM_VC4_T_TILED,
981                 DRM_FORMAT_MOD_BROADCOM_SAND128,
982                 DRM_FORMAT_MOD_BROADCOM_SAND64,
983                 DRM_FORMAT_MOD_BROADCOM_SAND256,
984                 DRM_FORMAT_MOD_LINEAR,
985                 DRM_FORMAT_MOD_INVALID
986         };
987
988         vc4_plane = devm_kzalloc(dev->dev, sizeof(*vc4_plane),
989                                  GFP_KERNEL);
990         if (!vc4_plane)
991                 return ERR_PTR(-ENOMEM);
992
993         for (i = 0; i < ARRAY_SIZE(hvs_formats); i++) {
994                 /* Don't allow YUV in cursor planes, since that means
995                  * tuning on the scaler, which we don't allow for the
996                  * cursor.
997                  */
998                 if (type != DRM_PLANE_TYPE_CURSOR ||
999                     hvs_formats[i].hvs < HVS_PIXEL_FORMAT_YCBCR_YUV420_3PLANE) {
1000                         formats[num_formats++] = hvs_formats[i].drm;
1001                 }
1002         }
1003         plane = &vc4_plane->base;
1004         ret = drm_universal_plane_init(dev, plane, 0,
1005                                        &vc4_plane_funcs,
1006                                        formats, num_formats,
1007                                        modifiers, type, NULL);
1008
1009         drm_plane_helper_add(plane, &vc4_plane_helper_funcs);
1010
1011         drm_plane_create_alpha_property(plane);
1012
1013         return plane;
1014 }