]> asedeno.scripts.mit.edu Git - linux.git/blob - drivers/gpu/drm/vc4/vc4_plane.c
c06bc73cc845840bc316661e6b3615faac56ed69
[linux.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 #include <drm/drm_atomic_uapi.h>
26
27 #include "uapi/drm/vc4_drm.h"
28 #include "vc4_drv.h"
29 #include "vc4_regs.h"
30
31 static const struct hvs_format {
32         u32 drm; /* DRM_FORMAT_* */
33         u32 hvs; /* HVS_FORMAT_* */
34         u32 pixel_order;
35 } hvs_formats[] = {
36         {
37                 .drm = DRM_FORMAT_XRGB8888, .hvs = HVS_PIXEL_FORMAT_RGBA8888,
38                 .pixel_order = HVS_PIXEL_ORDER_ABGR,
39         },
40         {
41                 .drm = DRM_FORMAT_ARGB8888, .hvs = HVS_PIXEL_FORMAT_RGBA8888,
42                 .pixel_order = HVS_PIXEL_ORDER_ABGR,
43         },
44         {
45                 .drm = DRM_FORMAT_ABGR8888, .hvs = HVS_PIXEL_FORMAT_RGBA8888,
46                 .pixel_order = HVS_PIXEL_ORDER_ARGB,
47         },
48         {
49                 .drm = DRM_FORMAT_XBGR8888, .hvs = HVS_PIXEL_FORMAT_RGBA8888,
50                 .pixel_order = HVS_PIXEL_ORDER_ARGB,
51         },
52         {
53                 .drm = DRM_FORMAT_RGB565, .hvs = HVS_PIXEL_FORMAT_RGB565,
54                 .pixel_order = HVS_PIXEL_ORDER_XRGB,
55         },
56         {
57                 .drm = DRM_FORMAT_BGR565, .hvs = HVS_PIXEL_FORMAT_RGB565,
58                 .pixel_order = HVS_PIXEL_ORDER_XBGR,
59         },
60         {
61                 .drm = DRM_FORMAT_ARGB1555, .hvs = HVS_PIXEL_FORMAT_RGBA5551,
62                 .pixel_order = HVS_PIXEL_ORDER_ABGR,
63         },
64         {
65                 .drm = DRM_FORMAT_XRGB1555, .hvs = HVS_PIXEL_FORMAT_RGBA5551,
66                 .pixel_order = HVS_PIXEL_ORDER_ABGR,
67         },
68         {
69                 .drm = DRM_FORMAT_RGB888, .hvs = HVS_PIXEL_FORMAT_RGB888,
70                 .pixel_order = HVS_PIXEL_ORDER_XRGB,
71         },
72         {
73                 .drm = DRM_FORMAT_BGR888, .hvs = HVS_PIXEL_FORMAT_RGB888,
74                 .pixel_order = HVS_PIXEL_ORDER_XBGR,
75         },
76         {
77                 .drm = DRM_FORMAT_YUV422,
78                 .hvs = HVS_PIXEL_FORMAT_YCBCR_YUV422_3PLANE,
79                 .pixel_order = HVS_PIXEL_ORDER_XYCBCR,
80         },
81         {
82                 .drm = DRM_FORMAT_YVU422,
83                 .hvs = HVS_PIXEL_FORMAT_YCBCR_YUV422_3PLANE,
84                 .pixel_order = HVS_PIXEL_ORDER_XYCRCB,
85         },
86         {
87                 .drm = DRM_FORMAT_YUV420,
88                 .hvs = HVS_PIXEL_FORMAT_YCBCR_YUV420_3PLANE,
89                 .pixel_order = HVS_PIXEL_ORDER_XYCBCR,
90         },
91         {
92                 .drm = DRM_FORMAT_YVU420,
93                 .hvs = HVS_PIXEL_FORMAT_YCBCR_YUV420_3PLANE,
94                 .pixel_order = HVS_PIXEL_ORDER_XYCRCB,
95         },
96         {
97                 .drm = DRM_FORMAT_NV12,
98                 .hvs = HVS_PIXEL_FORMAT_YCBCR_YUV420_2PLANE,
99                 .pixel_order = HVS_PIXEL_ORDER_XYCBCR,
100         },
101         {
102                 .drm = DRM_FORMAT_NV21,
103                 .hvs = HVS_PIXEL_FORMAT_YCBCR_YUV420_2PLANE,
104                 .pixel_order = HVS_PIXEL_ORDER_XYCRCB,
105         },
106         {
107                 .drm = DRM_FORMAT_NV16,
108                 .hvs = HVS_PIXEL_FORMAT_YCBCR_YUV422_2PLANE,
109                 .pixel_order = HVS_PIXEL_ORDER_XYCBCR,
110         },
111         {
112                 .drm = DRM_FORMAT_NV61,
113                 .hvs = HVS_PIXEL_FORMAT_YCBCR_YUV422_2PLANE,
114                 .pixel_order = HVS_PIXEL_ORDER_XYCRCB,
115         },
116 };
117
118 static const struct hvs_format *vc4_get_hvs_format(u32 drm_format)
119 {
120         unsigned i;
121
122         for (i = 0; i < ARRAY_SIZE(hvs_formats); i++) {
123                 if (hvs_formats[i].drm == drm_format)
124                         return &hvs_formats[i];
125         }
126
127         return NULL;
128 }
129
130 static enum vc4_scaling_mode vc4_get_scaling_mode(u32 src, u32 dst)
131 {
132         if (dst > src)
133                 return VC4_SCALING_PPF;
134         else if (dst < src)
135                 return VC4_SCALING_TPZ;
136         else
137                 return VC4_SCALING_NONE;
138 }
139
140 static bool plane_enabled(struct drm_plane_state *state)
141 {
142         return state->fb && state->crtc;
143 }
144
145 static struct drm_plane_state *vc4_plane_duplicate_state(struct drm_plane *plane)
146 {
147         struct vc4_plane_state *vc4_state;
148
149         if (WARN_ON(!plane->state))
150                 return NULL;
151
152         vc4_state = kmemdup(plane->state, sizeof(*vc4_state), GFP_KERNEL);
153         if (!vc4_state)
154                 return NULL;
155
156         memset(&vc4_state->lbm, 0, sizeof(vc4_state->lbm));
157
158         __drm_atomic_helper_plane_duplicate_state(plane, &vc4_state->base);
159
160         if (vc4_state->dlist) {
161                 vc4_state->dlist = kmemdup(vc4_state->dlist,
162                                            vc4_state->dlist_count * 4,
163                                            GFP_KERNEL);
164                 if (!vc4_state->dlist) {
165                         kfree(vc4_state);
166                         return NULL;
167                 }
168                 vc4_state->dlist_size = vc4_state->dlist_count;
169         }
170
171         return &vc4_state->base;
172 }
173
174 static void vc4_plane_destroy_state(struct drm_plane *plane,
175                                     struct drm_plane_state *state)
176 {
177         struct vc4_dev *vc4 = to_vc4_dev(plane->dev);
178         struct vc4_plane_state *vc4_state = to_vc4_plane_state(state);
179
180         if (vc4_state->lbm.allocated) {
181                 unsigned long irqflags;
182
183                 spin_lock_irqsave(&vc4->hvs->mm_lock, irqflags);
184                 drm_mm_remove_node(&vc4_state->lbm);
185                 spin_unlock_irqrestore(&vc4->hvs->mm_lock, irqflags);
186         }
187
188         kfree(vc4_state->dlist);
189         __drm_atomic_helper_plane_destroy_state(&vc4_state->base);
190         kfree(state);
191 }
192
193 /* Called during init to allocate the plane's atomic state. */
194 static void vc4_plane_reset(struct drm_plane *plane)
195 {
196         struct vc4_plane_state *vc4_state;
197
198         WARN_ON(plane->state);
199
200         vc4_state = kzalloc(sizeof(*vc4_state), GFP_KERNEL);
201         if (!vc4_state)
202                 return;
203
204         __drm_atomic_helper_plane_reset(plane, &vc4_state->base);
205 }
206
207 static void vc4_dlist_write(struct vc4_plane_state *vc4_state, u32 val)
208 {
209         if (vc4_state->dlist_count == vc4_state->dlist_size) {
210                 u32 new_size = max(4u, vc4_state->dlist_count * 2);
211                 u32 *new_dlist = kmalloc_array(new_size, 4, GFP_KERNEL);
212
213                 if (!new_dlist)
214                         return;
215                 memcpy(new_dlist, vc4_state->dlist, vc4_state->dlist_count * 4);
216
217                 kfree(vc4_state->dlist);
218                 vc4_state->dlist = new_dlist;
219                 vc4_state->dlist_size = new_size;
220         }
221
222         vc4_state->dlist[vc4_state->dlist_count++] = val;
223 }
224
225 /* Returns the scl0/scl1 field based on whether the dimensions need to
226  * be up/down/non-scaled.
227  *
228  * This is a replication of a table from the spec.
229  */
230 static u32 vc4_get_scl_field(struct drm_plane_state *state, int plane)
231 {
232         struct vc4_plane_state *vc4_state = to_vc4_plane_state(state);
233
234         switch (vc4_state->x_scaling[plane] << 2 | vc4_state->y_scaling[plane]) {
235         case VC4_SCALING_PPF << 2 | VC4_SCALING_PPF:
236                 return SCALER_CTL0_SCL_H_PPF_V_PPF;
237         case VC4_SCALING_TPZ << 2 | VC4_SCALING_PPF:
238                 return SCALER_CTL0_SCL_H_TPZ_V_PPF;
239         case VC4_SCALING_PPF << 2 | VC4_SCALING_TPZ:
240                 return SCALER_CTL0_SCL_H_PPF_V_TPZ;
241         case VC4_SCALING_TPZ << 2 | VC4_SCALING_TPZ:
242                 return SCALER_CTL0_SCL_H_TPZ_V_TPZ;
243         case VC4_SCALING_PPF << 2 | VC4_SCALING_NONE:
244                 return SCALER_CTL0_SCL_H_PPF_V_NONE;
245         case VC4_SCALING_NONE << 2 | VC4_SCALING_PPF:
246                 return SCALER_CTL0_SCL_H_NONE_V_PPF;
247         case VC4_SCALING_NONE << 2 | VC4_SCALING_TPZ:
248                 return SCALER_CTL0_SCL_H_NONE_V_TPZ;
249         case VC4_SCALING_TPZ << 2 | VC4_SCALING_NONE:
250                 return SCALER_CTL0_SCL_H_TPZ_V_NONE;
251         default:
252         case VC4_SCALING_NONE << 2 | VC4_SCALING_NONE:
253                 /* The unity case is independently handled by
254                  * SCALER_CTL0_UNITY.
255                  */
256                 return 0;
257         }
258 }
259
260 static int vc4_plane_setup_clipping_and_scaling(struct drm_plane_state *state)
261 {
262         struct drm_plane *plane = state->plane;
263         struct vc4_plane_state *vc4_state = to_vc4_plane_state(state);
264         struct drm_framebuffer *fb = state->fb;
265         struct drm_gem_cma_object *bo = drm_fb_cma_get_gem_obj(fb, 0);
266         u32 subpixel_src_mask = (1 << 16) - 1;
267         u32 format = fb->format->format;
268         int num_planes = fb->format->num_planes;
269         int min_scale = 1, max_scale = INT_MAX;
270         struct drm_crtc_state *crtc_state;
271         u32 h_subsample, v_subsample;
272         int i, ret;
273
274         crtc_state = drm_atomic_get_existing_crtc_state(state->state,
275                                                         state->crtc);
276         if (!crtc_state) {
277                 DRM_DEBUG_KMS("Invalid crtc state\n");
278                 return -EINVAL;
279         }
280
281         /* No configuring scaling on the cursor plane, since it gets
282          * non-vblank-synced updates, and scaling requires LBM changes which
283          * have to be vblank-synced.
284          */
285         if (plane->type == DRM_PLANE_TYPE_CURSOR) {
286                 min_scale = DRM_PLANE_HELPER_NO_SCALING;
287                 max_scale = DRM_PLANE_HELPER_NO_SCALING;
288         } else {
289                 min_scale = 1;
290                 max_scale = INT_MAX;
291         }
292
293         ret = drm_atomic_helper_check_plane_state(state, crtc_state,
294                                                   min_scale, max_scale,
295                                                   true, true);
296         if (ret)
297                 return ret;
298
299         h_subsample = drm_format_horz_chroma_subsampling(format);
300         v_subsample = drm_format_vert_chroma_subsampling(format);
301
302         for (i = 0; i < num_planes; i++)
303                 vc4_state->offsets[i] = bo->paddr + fb->offsets[i];
304
305         /* We don't support subpixel source positioning for scaling. */
306         if ((state->src.x1 & subpixel_src_mask) ||
307             (state->src.x2 & subpixel_src_mask) ||
308             (state->src.y1 & subpixel_src_mask) ||
309             (state->src.y2 & subpixel_src_mask)) {
310                 return -EINVAL;
311         }
312
313         vc4_state->src_x = state->src.x1 >> 16;
314         vc4_state->src_y = state->src.y1 >> 16;
315         vc4_state->src_w[0] = (state->src.x2 - state->src.x1) >> 16;
316         vc4_state->src_h[0] = (state->src.y2 - state->src.y1) >> 16;
317
318         vc4_state->crtc_x = state->dst.x1;
319         vc4_state->crtc_y = state->dst.y1;
320         vc4_state->crtc_w = state->dst.x2 - state->dst.x1;
321         vc4_state->crtc_h = state->dst.y2 - state->dst.y1;
322
323         vc4_state->x_scaling[0] = vc4_get_scaling_mode(vc4_state->src_w[0],
324                                                        vc4_state->crtc_w);
325         vc4_state->y_scaling[0] = vc4_get_scaling_mode(vc4_state->src_h[0],
326                                                        vc4_state->crtc_h);
327
328         vc4_state->is_unity = (vc4_state->x_scaling[0] == VC4_SCALING_NONE &&
329                                vc4_state->y_scaling[0] == VC4_SCALING_NONE);
330
331         if (num_planes > 1) {
332                 vc4_state->is_yuv = true;
333
334                 vc4_state->src_w[1] = vc4_state->src_w[0] / h_subsample;
335                 vc4_state->src_h[1] = vc4_state->src_h[0] / v_subsample;
336
337                 vc4_state->x_scaling[1] =
338                         vc4_get_scaling_mode(vc4_state->src_w[1],
339                                              vc4_state->crtc_w);
340                 vc4_state->y_scaling[1] =
341                         vc4_get_scaling_mode(vc4_state->src_h[1],
342                                              vc4_state->crtc_h);
343
344                 /* YUV conversion requires that horizontal scaling be enabled,
345                  * even on a plane that's otherwise 1:1. Looks like only PPF
346                  * works in that case, so let's pick that one.
347                  */
348                 if (vc4_state->is_unity)
349                         vc4_state->x_scaling[0] = VC4_SCALING_PPF;
350         } else {
351                 vc4_state->x_scaling[1] = VC4_SCALING_NONE;
352                 vc4_state->y_scaling[1] = VC4_SCALING_NONE;
353         }
354
355         /* Adjust the base pointer to the first pixel to be scanned out. */
356         for (i = 0; i < num_planes; i++) {
357                 vc4_state->offsets[i] += (vc4_state->src_y /
358                                           (i ? v_subsample : 1)) *
359                                          fb->pitches[i];
360                 vc4_state->offsets[i] += (vc4_state->src_x /
361                                           (i ? h_subsample : 1)) *
362                                          fb->format->cpp[i];
363         }
364
365         return 0;
366 }
367
368 static void vc4_write_tpz(struct vc4_plane_state *vc4_state, u32 src, u32 dst)
369 {
370         u32 scale, recip;
371
372         scale = (1 << 16) * src / dst;
373
374         /* The specs note that while the reciprocal would be defined
375          * as (1<<32)/scale, ~0 is close enough.
376          */
377         recip = ~0 / scale;
378
379         vc4_dlist_write(vc4_state,
380                         VC4_SET_FIELD(scale, SCALER_TPZ0_SCALE) |
381                         VC4_SET_FIELD(0, SCALER_TPZ0_IPHASE));
382         vc4_dlist_write(vc4_state,
383                         VC4_SET_FIELD(recip, SCALER_TPZ1_RECIP));
384 }
385
386 static void vc4_write_ppf(struct vc4_plane_state *vc4_state, u32 src, u32 dst)
387 {
388         u32 scale = (1 << 16) * src / dst;
389
390         vc4_dlist_write(vc4_state,
391                         SCALER_PPF_AGC |
392                         VC4_SET_FIELD(scale, SCALER_PPF_SCALE) |
393                         VC4_SET_FIELD(0, SCALER_PPF_IPHASE));
394 }
395
396 static u32 vc4_lbm_size(struct drm_plane_state *state)
397 {
398         struct vc4_plane_state *vc4_state = to_vc4_plane_state(state);
399         /* This is the worst case number.  One of the two sizes will
400          * be used depending on the scaling configuration.
401          */
402         u32 pix_per_line = max(vc4_state->src_w[0], (u32)vc4_state->crtc_w);
403         u32 lbm;
404
405         if (!vc4_state->is_yuv) {
406                 if (vc4_state->is_unity)
407                         return 0;
408                 else if (vc4_state->y_scaling[0] == VC4_SCALING_TPZ)
409                         lbm = pix_per_line * 8;
410                 else {
411                         /* In special cases, this multiplier might be 12. */
412                         lbm = pix_per_line * 16;
413                 }
414         } else {
415                 /* There are cases for this going down to a multiplier
416                  * of 2, but according to the firmware source, the
417                  * table in the docs is somewhat wrong.
418                  */
419                 lbm = pix_per_line * 16;
420         }
421
422         lbm = roundup(lbm, 32);
423
424         return lbm;
425 }
426
427 static void vc4_write_scaling_parameters(struct drm_plane_state *state,
428                                          int channel)
429 {
430         struct vc4_plane_state *vc4_state = to_vc4_plane_state(state);
431
432         /* Ch0 H-PPF Word 0: Scaling Parameters */
433         if (vc4_state->x_scaling[channel] == VC4_SCALING_PPF) {
434                 vc4_write_ppf(vc4_state,
435                               vc4_state->src_w[channel], vc4_state->crtc_w);
436         }
437
438         /* Ch0 V-PPF Words 0-1: Scaling Parameters, Context */
439         if (vc4_state->y_scaling[channel] == VC4_SCALING_PPF) {
440                 vc4_write_ppf(vc4_state,
441                               vc4_state->src_h[channel], vc4_state->crtc_h);
442                 vc4_dlist_write(vc4_state, 0xc0c0c0c0);
443         }
444
445         /* Ch0 H-TPZ Words 0-1: Scaling Parameters, Recip */
446         if (vc4_state->x_scaling[channel] == VC4_SCALING_TPZ) {
447                 vc4_write_tpz(vc4_state,
448                               vc4_state->src_w[channel], vc4_state->crtc_w);
449         }
450
451         /* Ch0 V-TPZ Words 0-2: Scaling Parameters, Recip, Context */
452         if (vc4_state->y_scaling[channel] == VC4_SCALING_TPZ) {
453                 vc4_write_tpz(vc4_state,
454                               vc4_state->src_h[channel], vc4_state->crtc_h);
455                 vc4_dlist_write(vc4_state, 0xc0c0c0c0);
456         }
457 }
458
459 /* Writes out a full display list for an active plane to the plane's
460  * private dlist state.
461  */
462 static int vc4_plane_mode_set(struct drm_plane *plane,
463                               struct drm_plane_state *state)
464 {
465         struct vc4_dev *vc4 = to_vc4_dev(plane->dev);
466         struct vc4_plane_state *vc4_state = to_vc4_plane_state(state);
467         struct drm_framebuffer *fb = state->fb;
468         u32 ctl0_offset = vc4_state->dlist_count;
469         const struct hvs_format *format = vc4_get_hvs_format(fb->format->format);
470         u64 base_format_mod = fourcc_mod_broadcom_mod(fb->modifier);
471         int num_planes = drm_format_num_planes(format->drm);
472         bool mix_plane_alpha;
473         bool covers_screen;
474         u32 scl0, scl1, pitch0;
475         u32 lbm_size, tiling;
476         unsigned long irqflags;
477         u32 hvs_format = format->hvs;
478         int ret, i;
479
480         ret = vc4_plane_setup_clipping_and_scaling(state);
481         if (ret)
482                 return ret;
483
484         /* Allocate the LBM memory that the HVS will use for temporary
485          * storage due to our scaling/format conversion.
486          */
487         lbm_size = vc4_lbm_size(state);
488         if (lbm_size) {
489                 if (!vc4_state->lbm.allocated) {
490                         spin_lock_irqsave(&vc4->hvs->mm_lock, irqflags);
491                         ret = drm_mm_insert_node_generic(&vc4->hvs->lbm_mm,
492                                                          &vc4_state->lbm,
493                                                          lbm_size, 32, 0, 0);
494                         spin_unlock_irqrestore(&vc4->hvs->mm_lock, irqflags);
495                 } else {
496                         WARN_ON_ONCE(lbm_size != vc4_state->lbm.size);
497                 }
498         }
499
500         if (ret)
501                 return ret;
502
503         /* SCL1 is used for Cb/Cr scaling of planar formats.  For RGB
504          * and 4:4:4, scl1 should be set to scl0 so both channels of
505          * the scaler do the same thing.  For YUV, the Y plane needs
506          * to be put in channel 1 and Cb/Cr in channel 0, so we swap
507          * the scl fields here.
508          */
509         if (num_planes == 1) {
510                 scl0 = vc4_get_scl_field(state, 0);
511                 scl1 = scl0;
512         } else {
513                 scl0 = vc4_get_scl_field(state, 1);
514                 scl1 = vc4_get_scl_field(state, 0);
515         }
516
517         switch (base_format_mod) {
518         case DRM_FORMAT_MOD_LINEAR:
519                 tiling = SCALER_CTL0_TILING_LINEAR;
520                 pitch0 = VC4_SET_FIELD(fb->pitches[0], SCALER_SRC_PITCH);
521                 break;
522
523         case DRM_FORMAT_MOD_BROADCOM_VC4_T_TILED: {
524                 /* For T-tiled, the FB pitch is "how many bytes from
525                  * one row to the next, such that pitch * tile_h ==
526                  * tile_size * tiles_per_row."
527                  */
528                 u32 tile_size_shift = 12; /* T tiles are 4kb */
529                 u32 tile_h_shift = 5; /* 16 and 32bpp are 32 pixels high */
530                 u32 tiles_w = fb->pitches[0] >> (tile_size_shift - tile_h_shift);
531
532                 tiling = SCALER_CTL0_TILING_256B_OR_T;
533
534                 pitch0 = (VC4_SET_FIELD(0, SCALER_PITCH0_TILE_Y_OFFSET) |
535                           VC4_SET_FIELD(0, SCALER_PITCH0_TILE_WIDTH_L) |
536                           VC4_SET_FIELD(tiles_w, SCALER_PITCH0_TILE_WIDTH_R));
537                 break;
538         }
539
540         case DRM_FORMAT_MOD_BROADCOM_SAND64:
541         case DRM_FORMAT_MOD_BROADCOM_SAND128:
542         case DRM_FORMAT_MOD_BROADCOM_SAND256: {
543                 uint32_t param = fourcc_mod_broadcom_param(fb->modifier);
544
545                 /* Column-based NV12 or RGBA.
546                  */
547                 if (fb->format->num_planes > 1) {
548                         if (hvs_format != HVS_PIXEL_FORMAT_YCBCR_YUV420_2PLANE) {
549                                 DRM_DEBUG_KMS("SAND format only valid for NV12/21");
550                                 return -EINVAL;
551                         }
552                         hvs_format = HVS_PIXEL_FORMAT_H264;
553                 } else {
554                         if (base_format_mod == DRM_FORMAT_MOD_BROADCOM_SAND256) {
555                                 DRM_DEBUG_KMS("SAND256 format only valid for H.264");
556                                 return -EINVAL;
557                         }
558                 }
559
560                 switch (base_format_mod) {
561                 case DRM_FORMAT_MOD_BROADCOM_SAND64:
562                         tiling = SCALER_CTL0_TILING_64B;
563                         break;
564                 case DRM_FORMAT_MOD_BROADCOM_SAND128:
565                         tiling = SCALER_CTL0_TILING_128B;
566                         break;
567                 case DRM_FORMAT_MOD_BROADCOM_SAND256:
568                         tiling = SCALER_CTL0_TILING_256B_OR_T;
569                         break;
570                 default:
571                         break;
572                 }
573
574                 if (param > SCALER_TILE_HEIGHT_MASK) {
575                         DRM_DEBUG_KMS("SAND height too large (%d)\n", param);
576                         return -EINVAL;
577                 }
578
579                 pitch0 = VC4_SET_FIELD(param, SCALER_TILE_HEIGHT);
580                 break;
581         }
582
583         default:
584                 DRM_DEBUG_KMS("Unsupported FB tiling flag 0x%16llx",
585                               (long long)fb->modifier);
586                 return -EINVAL;
587         }
588
589         /* Control word */
590         vc4_dlist_write(vc4_state,
591                         SCALER_CTL0_VALID |
592                         VC4_SET_FIELD(SCALER_CTL0_RGBA_EXPAND_ROUND, SCALER_CTL0_RGBA_EXPAND) |
593                         (format->pixel_order << SCALER_CTL0_ORDER_SHIFT) |
594                         (hvs_format << SCALER_CTL0_PIXEL_FORMAT_SHIFT) |
595                         VC4_SET_FIELD(tiling, SCALER_CTL0_TILING) |
596                         (vc4_state->is_unity ? SCALER_CTL0_UNITY : 0) |
597                         VC4_SET_FIELD(scl0, SCALER_CTL0_SCL0) |
598                         VC4_SET_FIELD(scl1, SCALER_CTL0_SCL1));
599
600         /* Position Word 0: Image Positions and Alpha Value */
601         vc4_state->pos0_offset = vc4_state->dlist_count;
602         vc4_dlist_write(vc4_state,
603                         VC4_SET_FIELD(state->alpha >> 8, SCALER_POS0_FIXED_ALPHA) |
604                         VC4_SET_FIELD(vc4_state->crtc_x, SCALER_POS0_START_X) |
605                         VC4_SET_FIELD(vc4_state->crtc_y, SCALER_POS0_START_Y));
606
607         /* Position Word 1: Scaled Image Dimensions. */
608         if (!vc4_state->is_unity) {
609                 vc4_dlist_write(vc4_state,
610                                 VC4_SET_FIELD(vc4_state->crtc_w,
611                                               SCALER_POS1_SCL_WIDTH) |
612                                 VC4_SET_FIELD(vc4_state->crtc_h,
613                                               SCALER_POS1_SCL_HEIGHT));
614         }
615
616         /* Don't waste cycles mixing with plane alpha if the set alpha
617          * is opaque or there is no per-pixel alpha information.
618          * In any case we use the alpha property value as the fixed alpha.
619          */
620         mix_plane_alpha = state->alpha != DRM_BLEND_ALPHA_OPAQUE &&
621                           fb->format->has_alpha;
622
623         /* Position Word 2: Source Image Size, Alpha */
624         vc4_state->pos2_offset = vc4_state->dlist_count;
625         vc4_dlist_write(vc4_state,
626                         VC4_SET_FIELD(fb->format->has_alpha ?
627                                       SCALER_POS2_ALPHA_MODE_PIPELINE :
628                                       SCALER_POS2_ALPHA_MODE_FIXED,
629                                       SCALER_POS2_ALPHA_MODE) |
630                         (mix_plane_alpha ? SCALER_POS2_ALPHA_MIX : 0) |
631                         (fb->format->has_alpha ? SCALER_POS2_ALPHA_PREMULT : 0) |
632                         VC4_SET_FIELD(vc4_state->src_w[0], SCALER_POS2_WIDTH) |
633                         VC4_SET_FIELD(vc4_state->src_h[0], SCALER_POS2_HEIGHT));
634
635         /* Position Word 3: Context.  Written by the HVS. */
636         vc4_dlist_write(vc4_state, 0xc0c0c0c0);
637
638
639         /* Pointer Word 0/1/2: RGB / Y / Cb / Cr Pointers
640          *
641          * The pointers may be any byte address.
642          */
643         vc4_state->ptr0_offset = vc4_state->dlist_count;
644         for (i = 0; i < num_planes; i++)
645                 vc4_dlist_write(vc4_state, vc4_state->offsets[i]);
646
647         /* Pointer Context Word 0/1/2: Written by the HVS */
648         for (i = 0; i < num_planes; i++)
649                 vc4_dlist_write(vc4_state, 0xc0c0c0c0);
650
651         /* Pitch word 0 */
652         vc4_dlist_write(vc4_state, pitch0);
653
654         /* Pitch word 1/2 */
655         for (i = 1; i < num_planes; i++) {
656                 if (hvs_format != HVS_PIXEL_FORMAT_H264) {
657                         vc4_dlist_write(vc4_state,
658                                         VC4_SET_FIELD(fb->pitches[i],
659                                                       SCALER_SRC_PITCH));
660                 } else {
661                         vc4_dlist_write(vc4_state, pitch0);
662                 }
663         }
664
665         /* Colorspace conversion words */
666         if (vc4_state->is_yuv) {
667                 vc4_dlist_write(vc4_state, SCALER_CSC0_ITR_R_601_5);
668                 vc4_dlist_write(vc4_state, SCALER_CSC1_ITR_R_601_5);
669                 vc4_dlist_write(vc4_state, SCALER_CSC2_ITR_R_601_5);
670         }
671
672         if (vc4_state->x_scaling[0] != VC4_SCALING_NONE ||
673             vc4_state->x_scaling[1] != VC4_SCALING_NONE ||
674             vc4_state->y_scaling[0] != VC4_SCALING_NONE ||
675             vc4_state->y_scaling[1] != VC4_SCALING_NONE) {
676                 /* LBM Base Address. */
677                 if (vc4_state->y_scaling[0] != VC4_SCALING_NONE ||
678                     vc4_state->y_scaling[1] != VC4_SCALING_NONE) {
679                         vc4_dlist_write(vc4_state, vc4_state->lbm.start);
680                 }
681
682                 if (num_planes > 1) {
683                         /* Emit Cb/Cr as channel 0 and Y as channel
684                          * 1. This matches how we set up scl0/scl1
685                          * above.
686                          */
687                         vc4_write_scaling_parameters(state, 1);
688                 }
689                 vc4_write_scaling_parameters(state, 0);
690
691                 /* If any PPF setup was done, then all the kernel
692                  * pointers get uploaded.
693                  */
694                 if (vc4_state->x_scaling[0] == VC4_SCALING_PPF ||
695                     vc4_state->y_scaling[0] == VC4_SCALING_PPF ||
696                     vc4_state->x_scaling[1] == VC4_SCALING_PPF ||
697                     vc4_state->y_scaling[1] == VC4_SCALING_PPF) {
698                         u32 kernel = VC4_SET_FIELD(vc4->hvs->mitchell_netravali_filter.start,
699                                                    SCALER_PPF_KERNEL_OFFSET);
700
701                         /* HPPF plane 0 */
702                         vc4_dlist_write(vc4_state, kernel);
703                         /* VPPF plane 0 */
704                         vc4_dlist_write(vc4_state, kernel);
705                         /* HPPF plane 1 */
706                         vc4_dlist_write(vc4_state, kernel);
707                         /* VPPF plane 1 */
708                         vc4_dlist_write(vc4_state, kernel);
709                 }
710         }
711
712         vc4_state->dlist[ctl0_offset] |=
713                 VC4_SET_FIELD(vc4_state->dlist_count, SCALER_CTL0_SIZE);
714
715         /* crtc_* are already clipped coordinates. */
716         covers_screen = vc4_state->crtc_x == 0 && vc4_state->crtc_y == 0 &&
717                         vc4_state->crtc_w == state->crtc->mode.hdisplay &&
718                         vc4_state->crtc_h == state->crtc->mode.vdisplay;
719         /* Background fill might be necessary when the plane has per-pixel
720          * alpha content or a non-opaque plane alpha and could blend from the
721          * background or does not cover the entire screen.
722          */
723         vc4_state->needs_bg_fill = fb->format->has_alpha || !covers_screen ||
724                                    state->alpha != DRM_BLEND_ALPHA_OPAQUE;
725
726         return 0;
727 }
728
729 /* If a modeset involves changing the setup of a plane, the atomic
730  * infrastructure will call this to validate a proposed plane setup.
731  * However, if a plane isn't getting updated, this (and the
732  * corresponding vc4_plane_atomic_update) won't get called.  Thus, we
733  * compute the dlist here and have all active plane dlists get updated
734  * in the CRTC's flush.
735  */
736 static int vc4_plane_atomic_check(struct drm_plane *plane,
737                                   struct drm_plane_state *state)
738 {
739         struct vc4_plane_state *vc4_state = to_vc4_plane_state(state);
740
741         vc4_state->dlist_count = 0;
742
743         if (plane_enabled(state))
744                 return vc4_plane_mode_set(plane, state);
745         else
746                 return 0;
747 }
748
749 static void vc4_plane_atomic_update(struct drm_plane *plane,
750                                     struct drm_plane_state *old_state)
751 {
752         /* No contents here.  Since we don't know where in the CRTC's
753          * dlist we should be stored, our dlist is uploaded to the
754          * hardware with vc4_plane_write_dlist() at CRTC atomic_flush
755          * time.
756          */
757 }
758
759 u32 vc4_plane_write_dlist(struct drm_plane *plane, u32 __iomem *dlist)
760 {
761         struct vc4_plane_state *vc4_state = to_vc4_plane_state(plane->state);
762         int i;
763
764         vc4_state->hw_dlist = dlist;
765
766         /* Can't memcpy_toio() because it needs to be 32-bit writes. */
767         for (i = 0; i < vc4_state->dlist_count; i++)
768                 writel(vc4_state->dlist[i], &dlist[i]);
769
770         return vc4_state->dlist_count;
771 }
772
773 u32 vc4_plane_dlist_size(const struct drm_plane_state *state)
774 {
775         const struct vc4_plane_state *vc4_state =
776                 container_of(state, typeof(*vc4_state), base);
777
778         return vc4_state->dlist_count;
779 }
780
781 /* Updates the plane to immediately (well, once the FIFO needs
782  * refilling) scan out from at a new framebuffer.
783  */
784 void vc4_plane_async_set_fb(struct drm_plane *plane, struct drm_framebuffer *fb)
785 {
786         struct vc4_plane_state *vc4_state = to_vc4_plane_state(plane->state);
787         struct drm_gem_cma_object *bo = drm_fb_cma_get_gem_obj(fb, 0);
788         uint32_t addr;
789
790         /* We're skipping the address adjustment for negative origin,
791          * because this is only called on the primary plane.
792          */
793         WARN_ON_ONCE(plane->state->crtc_x < 0 || plane->state->crtc_y < 0);
794         addr = bo->paddr + fb->offsets[0];
795
796         /* Write the new address into the hardware immediately.  The
797          * scanout will start from this address as soon as the FIFO
798          * needs to refill with pixels.
799          */
800         writel(addr, &vc4_state->hw_dlist[vc4_state->ptr0_offset]);
801
802         /* Also update the CPU-side dlist copy, so that any later
803          * atomic updates that don't do a new modeset on our plane
804          * also use our updated address.
805          */
806         vc4_state->dlist[vc4_state->ptr0_offset] = addr;
807 }
808
809 static void vc4_plane_atomic_async_update(struct drm_plane *plane,
810                                           struct drm_plane_state *state)
811 {
812         struct vc4_plane_state *vc4_state = to_vc4_plane_state(plane->state);
813
814         if (plane->state->fb != state->fb) {
815                 vc4_plane_async_set_fb(plane, state->fb);
816                 drm_atomic_set_fb_for_plane(plane->state, state->fb);
817         }
818
819         /* Set the cursor's position on the screen.  This is the
820          * expected change from the drm_mode_cursor_universal()
821          * helper.
822          */
823         plane->state->crtc_x = state->crtc_x;
824         plane->state->crtc_y = state->crtc_y;
825
826         /* Allow changing the start position within the cursor BO, if
827          * that matters.
828          */
829         plane->state->src_x = state->src_x;
830         plane->state->src_y = state->src_y;
831
832         /* Update the display list based on the new crtc_x/y. */
833         vc4_plane_atomic_check(plane, plane->state);
834
835         /* Note that we can't just call vc4_plane_write_dlist()
836          * because that would smash the context data that the HVS is
837          * currently using.
838          */
839         writel(vc4_state->dlist[vc4_state->pos0_offset],
840                &vc4_state->hw_dlist[vc4_state->pos0_offset]);
841         writel(vc4_state->dlist[vc4_state->pos2_offset],
842                &vc4_state->hw_dlist[vc4_state->pos2_offset]);
843         writel(vc4_state->dlist[vc4_state->ptr0_offset],
844                &vc4_state->hw_dlist[vc4_state->ptr0_offset]);
845 }
846
847 static int vc4_plane_atomic_async_check(struct drm_plane *plane,
848                                         struct drm_plane_state *state)
849 {
850         /* No configuring new scaling in the fast path. */
851         if (plane->state->crtc_w != state->crtc_w ||
852             plane->state->crtc_h != state->crtc_h ||
853             plane->state->src_w != state->src_w ||
854             plane->state->src_h != state->src_h)
855                 return -EINVAL;
856
857         return 0;
858 }
859
860 static int vc4_prepare_fb(struct drm_plane *plane,
861                           struct drm_plane_state *state)
862 {
863         struct vc4_bo *bo;
864         struct dma_fence *fence;
865         int ret;
866
867         if (!state->fb)
868                 return 0;
869
870         bo = to_vc4_bo(&drm_fb_cma_get_gem_obj(state->fb, 0)->base);
871
872         fence = reservation_object_get_excl_rcu(bo->resv);
873         drm_atomic_set_fence_for_plane(state, fence);
874
875         if (plane->state->fb == state->fb)
876                 return 0;
877
878         ret = vc4_bo_inc_usecnt(bo);
879         if (ret)
880                 return ret;
881
882         return 0;
883 }
884
885 static void vc4_cleanup_fb(struct drm_plane *plane,
886                            struct drm_plane_state *state)
887 {
888         struct vc4_bo *bo;
889
890         if (plane->state->fb == state->fb || !state->fb)
891                 return;
892
893         bo = to_vc4_bo(&drm_fb_cma_get_gem_obj(state->fb, 0)->base);
894         vc4_bo_dec_usecnt(bo);
895 }
896
897 static const struct drm_plane_helper_funcs vc4_plane_helper_funcs = {
898         .atomic_check = vc4_plane_atomic_check,
899         .atomic_update = vc4_plane_atomic_update,
900         .prepare_fb = vc4_prepare_fb,
901         .cleanup_fb = vc4_cleanup_fb,
902         .atomic_async_check = vc4_plane_atomic_async_check,
903         .atomic_async_update = vc4_plane_atomic_async_update,
904 };
905
906 static void vc4_plane_destroy(struct drm_plane *plane)
907 {
908         drm_plane_cleanup(plane);
909 }
910
911 static bool vc4_format_mod_supported(struct drm_plane *plane,
912                                      uint32_t format,
913                                      uint64_t modifier)
914 {
915         /* Support T_TILING for RGB formats only. */
916         switch (format) {
917         case DRM_FORMAT_XRGB8888:
918         case DRM_FORMAT_ARGB8888:
919         case DRM_FORMAT_ABGR8888:
920         case DRM_FORMAT_XBGR8888:
921         case DRM_FORMAT_RGB565:
922         case DRM_FORMAT_BGR565:
923         case DRM_FORMAT_ARGB1555:
924         case DRM_FORMAT_XRGB1555:
925                 switch (fourcc_mod_broadcom_mod(modifier)) {
926                 case DRM_FORMAT_MOD_LINEAR:
927                 case DRM_FORMAT_MOD_BROADCOM_VC4_T_TILED:
928                 case DRM_FORMAT_MOD_BROADCOM_SAND64:
929                 case DRM_FORMAT_MOD_BROADCOM_SAND128:
930                         return true;
931                 default:
932                         return false;
933                 }
934         case DRM_FORMAT_NV12:
935         case DRM_FORMAT_NV21:
936                 switch (fourcc_mod_broadcom_mod(modifier)) {
937                 case DRM_FORMAT_MOD_LINEAR:
938                 case DRM_FORMAT_MOD_BROADCOM_SAND64:
939                 case DRM_FORMAT_MOD_BROADCOM_SAND128:
940                 case DRM_FORMAT_MOD_BROADCOM_SAND256:
941                         return true;
942                 default:
943                         return false;
944                 }
945         case DRM_FORMAT_YUV422:
946         case DRM_FORMAT_YVU422:
947         case DRM_FORMAT_YUV420:
948         case DRM_FORMAT_YVU420:
949         case DRM_FORMAT_NV16:
950         case DRM_FORMAT_NV61:
951         default:
952                 return (modifier == DRM_FORMAT_MOD_LINEAR);
953         }
954 }
955
956 static const struct drm_plane_funcs vc4_plane_funcs = {
957         .update_plane = drm_atomic_helper_update_plane,
958         .disable_plane = drm_atomic_helper_disable_plane,
959         .destroy = vc4_plane_destroy,
960         .set_property = NULL,
961         .reset = vc4_plane_reset,
962         .atomic_duplicate_state = vc4_plane_duplicate_state,
963         .atomic_destroy_state = vc4_plane_destroy_state,
964         .format_mod_supported = vc4_format_mod_supported,
965 };
966
967 struct drm_plane *vc4_plane_init(struct drm_device *dev,
968                                  enum drm_plane_type type)
969 {
970         struct drm_plane *plane = NULL;
971         struct vc4_plane *vc4_plane;
972         u32 formats[ARRAY_SIZE(hvs_formats)];
973         u32 num_formats = 0;
974         int ret = 0;
975         unsigned i;
976         static const uint64_t modifiers[] = {
977                 DRM_FORMAT_MOD_BROADCOM_VC4_T_TILED,
978                 DRM_FORMAT_MOD_BROADCOM_SAND128,
979                 DRM_FORMAT_MOD_BROADCOM_SAND64,
980                 DRM_FORMAT_MOD_BROADCOM_SAND256,
981                 DRM_FORMAT_MOD_LINEAR,
982                 DRM_FORMAT_MOD_INVALID
983         };
984
985         vc4_plane = devm_kzalloc(dev->dev, sizeof(*vc4_plane),
986                                  GFP_KERNEL);
987         if (!vc4_plane)
988                 return ERR_PTR(-ENOMEM);
989
990         for (i = 0; i < ARRAY_SIZE(hvs_formats); i++) {
991                 /* Don't allow YUV in cursor planes, since that means
992                  * tuning on the scaler, which we don't allow for the
993                  * cursor.
994                  */
995                 if (type != DRM_PLANE_TYPE_CURSOR ||
996                     hvs_formats[i].hvs < HVS_PIXEL_FORMAT_YCBCR_YUV420_3PLANE) {
997                         formats[num_formats++] = hvs_formats[i].drm;
998                 }
999         }
1000         plane = &vc4_plane->base;
1001         ret = drm_universal_plane_init(dev, plane, 0,
1002                                        &vc4_plane_funcs,
1003                                        formats, num_formats,
1004                                        modifiers, type, NULL);
1005
1006         drm_plane_helper_add(plane, &vc4_plane_helper_funcs);
1007
1008         drm_plane_create_alpha_property(plane);
1009
1010         return plane;
1011 }