]> Git Repo - J-linux.git/blob - drivers/gpu/drm/vc4/vc4_plane.c
Merge tag 'drm-misc-next-2018-06-21' of git://anongit.freedesktop.org/drm/drm-misc...
[J-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
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         if (num_planes > 1) {
301                 vc4_state->is_yuv = true;
302
303                 h_subsample = drm_format_horz_chroma_subsampling(format);
304                 v_subsample = drm_format_vert_chroma_subsampling(format);
305                 vc4_state->src_w[1] = vc4_state->src_w[0] / h_subsample;
306                 vc4_state->src_h[1] = vc4_state->src_h[0] / v_subsample;
307
308                 vc4_state->x_scaling[1] =
309                         vc4_get_scaling_mode(vc4_state->src_w[1],
310                                              vc4_state->crtc_w);
311                 vc4_state->y_scaling[1] =
312                         vc4_get_scaling_mode(vc4_state->src_h[1],
313                                              vc4_state->crtc_h);
314
315                 /* YUV conversion requires that scaling be enabled,
316                  * even on a plane that's otherwise 1:1.  Choose TPZ
317                  * for simplicity.
318                  */
319                 if (vc4_state->x_scaling[0] == VC4_SCALING_NONE)
320                         vc4_state->x_scaling[0] = VC4_SCALING_TPZ;
321                 if (vc4_state->y_scaling[0] == VC4_SCALING_NONE)
322                         vc4_state->y_scaling[0] = VC4_SCALING_TPZ;
323         }
324
325         vc4_state->is_unity = (vc4_state->x_scaling[0] == VC4_SCALING_NONE &&
326                                vc4_state->y_scaling[0] == VC4_SCALING_NONE &&
327                                vc4_state->x_scaling[1] == VC4_SCALING_NONE &&
328                                vc4_state->y_scaling[1] == VC4_SCALING_NONE);
329
330         /* No configuring scaling on the cursor plane, since it gets
331            non-vblank-synced updates, and scaling requires requires
332            LBM changes which have to be vblank-synced.
333          */
334         if (plane->type == DRM_PLANE_TYPE_CURSOR && !vc4_state->is_unity)
335                 return -EINVAL;
336
337         /* Clamp the on-screen start x/y to 0.  The hardware doesn't
338          * support negative y, and negative x wastes bandwidth.
339          */
340         if (vc4_state->crtc_x < 0) {
341                 for (i = 0; i < num_planes; i++) {
342                         u32 cpp = fb->format->cpp[i];
343                         u32 subs = ((i == 0) ? 1 : h_subsample);
344
345                         vc4_state->offsets[i] += (cpp *
346                                                   (-vc4_state->crtc_x) / subs);
347                 }
348                 vc4_state->src_w[0] += vc4_state->crtc_x;
349                 vc4_state->src_w[1] += vc4_state->crtc_x / h_subsample;
350                 vc4_state->crtc_x = 0;
351         }
352
353         if (vc4_state->crtc_y < 0) {
354                 for (i = 0; i < num_planes; i++) {
355                         u32 subs = ((i == 0) ? 1 : v_subsample);
356
357                         vc4_state->offsets[i] += (fb->pitches[i] *
358                                                   (-vc4_state->crtc_y) / subs);
359                 }
360                 vc4_state->src_h[0] += vc4_state->crtc_y;
361                 vc4_state->src_h[1] += vc4_state->crtc_y / v_subsample;
362                 vc4_state->crtc_y = 0;
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->is_unity) {
673                 /* LBM Base Address. */
674                 if (vc4_state->y_scaling[0] != VC4_SCALING_NONE ||
675                     vc4_state->y_scaling[1] != VC4_SCALING_NONE) {
676                         vc4_dlist_write(vc4_state, vc4_state->lbm.start);
677                 }
678
679                 if (num_planes > 1) {
680                         /* Emit Cb/Cr as channel 0 and Y as channel
681                          * 1. This matches how we set up scl0/scl1
682                          * above.
683                          */
684                         vc4_write_scaling_parameters(state, 1);
685                 }
686                 vc4_write_scaling_parameters(state, 0);
687
688                 /* If any PPF setup was done, then all the kernel
689                  * pointers get uploaded.
690                  */
691                 if (vc4_state->x_scaling[0] == VC4_SCALING_PPF ||
692                     vc4_state->y_scaling[0] == VC4_SCALING_PPF ||
693                     vc4_state->x_scaling[1] == VC4_SCALING_PPF ||
694                     vc4_state->y_scaling[1] == VC4_SCALING_PPF) {
695                         u32 kernel = VC4_SET_FIELD(vc4->hvs->mitchell_netravali_filter.start,
696                                                    SCALER_PPF_KERNEL_OFFSET);
697
698                         /* HPPF plane 0 */
699                         vc4_dlist_write(vc4_state, kernel);
700                         /* VPPF plane 0 */
701                         vc4_dlist_write(vc4_state, kernel);
702                         /* HPPF plane 1 */
703                         vc4_dlist_write(vc4_state, kernel);
704                         /* VPPF plane 1 */
705                         vc4_dlist_write(vc4_state, kernel);
706                 }
707         }
708
709         vc4_state->dlist[ctl0_offset] |=
710                 VC4_SET_FIELD(vc4_state->dlist_count, SCALER_CTL0_SIZE);
711
712         /* crtc_* are already clipped coordinates. */
713         covers_screen = vc4_state->crtc_x == 0 && vc4_state->crtc_y == 0 &&
714                         vc4_state->crtc_w == state->crtc->mode.hdisplay &&
715                         vc4_state->crtc_h == state->crtc->mode.vdisplay;
716         /* Background fill might be necessary when the plane has per-pixel
717          * alpha content or a non-opaque plane alpha and could blend from the
718          * background or does not cover the entire screen.
719          */
720         vc4_state->needs_bg_fill = fb->format->has_alpha || !covers_screen ||
721                                    state->alpha != DRM_BLEND_ALPHA_OPAQUE;
722
723         return 0;
724 }
725
726 /* If a modeset involves changing the setup of a plane, the atomic
727  * infrastructure will call this to validate a proposed plane setup.
728  * However, if a plane isn't getting updated, this (and the
729  * corresponding vc4_plane_atomic_update) won't get called.  Thus, we
730  * compute the dlist here and have all active plane dlists get updated
731  * in the CRTC's flush.
732  */
733 static int vc4_plane_atomic_check(struct drm_plane *plane,
734                                   struct drm_plane_state *state)
735 {
736         struct vc4_plane_state *vc4_state = to_vc4_plane_state(state);
737
738         vc4_state->dlist_count = 0;
739
740         if (plane_enabled(state))
741                 return vc4_plane_mode_set(plane, state);
742         else
743                 return 0;
744 }
745
746 static void vc4_plane_atomic_update(struct drm_plane *plane,
747                                     struct drm_plane_state *old_state)
748 {
749         /* No contents here.  Since we don't know where in the CRTC's
750          * dlist we should be stored, our dlist is uploaded to the
751          * hardware with vc4_plane_write_dlist() at CRTC atomic_flush
752          * time.
753          */
754 }
755
756 u32 vc4_plane_write_dlist(struct drm_plane *plane, u32 __iomem *dlist)
757 {
758         struct vc4_plane_state *vc4_state = to_vc4_plane_state(plane->state);
759         int i;
760
761         vc4_state->hw_dlist = dlist;
762
763         /* Can't memcpy_toio() because it needs to be 32-bit writes. */
764         for (i = 0; i < vc4_state->dlist_count; i++)
765                 writel(vc4_state->dlist[i], &dlist[i]);
766
767         return vc4_state->dlist_count;
768 }
769
770 u32 vc4_plane_dlist_size(const struct drm_plane_state *state)
771 {
772         const struct vc4_plane_state *vc4_state =
773                 container_of(state, typeof(*vc4_state), base);
774
775         return vc4_state->dlist_count;
776 }
777
778 /* Updates the plane to immediately (well, once the FIFO needs
779  * refilling) scan out from at a new framebuffer.
780  */
781 void vc4_plane_async_set_fb(struct drm_plane *plane, struct drm_framebuffer *fb)
782 {
783         struct vc4_plane_state *vc4_state = to_vc4_plane_state(plane->state);
784         struct drm_gem_cma_object *bo = drm_fb_cma_get_gem_obj(fb, 0);
785         uint32_t addr;
786
787         /* We're skipping the address adjustment for negative origin,
788          * because this is only called on the primary plane.
789          */
790         WARN_ON_ONCE(plane->state->crtc_x < 0 || plane->state->crtc_y < 0);
791         addr = bo->paddr + fb->offsets[0];
792
793         /* Write the new address into the hardware immediately.  The
794          * scanout will start from this address as soon as the FIFO
795          * needs to refill with pixels.
796          */
797         writel(addr, &vc4_state->hw_dlist[vc4_state->ptr0_offset]);
798
799         /* Also update the CPU-side dlist copy, so that any later
800          * atomic updates that don't do a new modeset on our plane
801          * also use our updated address.
802          */
803         vc4_state->dlist[vc4_state->ptr0_offset] = addr;
804 }
805
806 static void vc4_plane_atomic_async_update(struct drm_plane *plane,
807                                           struct drm_plane_state *state)
808 {
809         struct vc4_plane_state *vc4_state = to_vc4_plane_state(plane->state);
810
811         if (plane->state->fb != state->fb) {
812                 vc4_plane_async_set_fb(plane, state->fb);
813                 drm_atomic_set_fb_for_plane(plane->state, state->fb);
814         }
815
816         /* Set the cursor's position on the screen.  This is the
817          * expected change from the drm_mode_cursor_universal()
818          * helper.
819          */
820         plane->state->crtc_x = state->crtc_x;
821         plane->state->crtc_y = state->crtc_y;
822
823         /* Allow changing the start position within the cursor BO, if
824          * that matters.
825          */
826         plane->state->src_x = state->src_x;
827         plane->state->src_y = state->src_y;
828
829         /* Update the display list based on the new crtc_x/y. */
830         vc4_plane_atomic_check(plane, plane->state);
831
832         /* Note that we can't just call vc4_plane_write_dlist()
833          * because that would smash the context data that the HVS is
834          * currently using.
835          */
836         writel(vc4_state->dlist[vc4_state->pos0_offset],
837                &vc4_state->hw_dlist[vc4_state->pos0_offset]);
838         writel(vc4_state->dlist[vc4_state->pos2_offset],
839                &vc4_state->hw_dlist[vc4_state->pos2_offset]);
840         writel(vc4_state->dlist[vc4_state->ptr0_offset],
841                &vc4_state->hw_dlist[vc4_state->ptr0_offset]);
842 }
843
844 static int vc4_plane_atomic_async_check(struct drm_plane *plane,
845                                         struct drm_plane_state *state)
846 {
847         /* No configuring new scaling in the fast path. */
848         if (plane->state->crtc_w != state->crtc_w ||
849             plane->state->crtc_h != state->crtc_h ||
850             plane->state->src_w != state->src_w ||
851             plane->state->src_h != state->src_h)
852                 return -EINVAL;
853
854         return 0;
855 }
856
857 static int vc4_prepare_fb(struct drm_plane *plane,
858                           struct drm_plane_state *state)
859 {
860         struct vc4_bo *bo;
861         struct dma_fence *fence;
862         int ret;
863
864         if (!state->fb)
865                 return 0;
866
867         bo = to_vc4_bo(&drm_fb_cma_get_gem_obj(state->fb, 0)->base);
868
869         fence = reservation_object_get_excl_rcu(bo->resv);
870         drm_atomic_set_fence_for_plane(state, fence);
871
872         if (plane->state->fb == state->fb)
873                 return 0;
874
875         ret = vc4_bo_inc_usecnt(bo);
876         if (ret)
877                 return ret;
878
879         return 0;
880 }
881
882 static void vc4_cleanup_fb(struct drm_plane *plane,
883                            struct drm_plane_state *state)
884 {
885         struct vc4_bo *bo;
886
887         if (plane->state->fb == state->fb || !state->fb)
888                 return;
889
890         bo = to_vc4_bo(&drm_fb_cma_get_gem_obj(state->fb, 0)->base);
891         vc4_bo_dec_usecnt(bo);
892 }
893
894 static const struct drm_plane_helper_funcs vc4_plane_helper_funcs = {
895         .atomic_check = vc4_plane_atomic_check,
896         .atomic_update = vc4_plane_atomic_update,
897         .prepare_fb = vc4_prepare_fb,
898         .cleanup_fb = vc4_cleanup_fb,
899         .atomic_async_check = vc4_plane_atomic_async_check,
900         .atomic_async_update = vc4_plane_atomic_async_update,
901 };
902
903 static void vc4_plane_destroy(struct drm_plane *plane)
904 {
905         drm_plane_helper_disable(plane);
906         drm_plane_cleanup(plane);
907 }
908
909 static bool vc4_format_mod_supported(struct drm_plane *plane,
910                                      uint32_t format,
911                                      uint64_t modifier)
912 {
913         /* Support T_TILING for RGB formats only. */
914         switch (format) {
915         case DRM_FORMAT_XRGB8888:
916         case DRM_FORMAT_ARGB8888:
917         case DRM_FORMAT_ABGR8888:
918         case DRM_FORMAT_XBGR8888:
919         case DRM_FORMAT_RGB565:
920         case DRM_FORMAT_BGR565:
921         case DRM_FORMAT_ARGB1555:
922         case DRM_FORMAT_XRGB1555:
923                 switch (fourcc_mod_broadcom_mod(modifier)) {
924                 case DRM_FORMAT_MOD_LINEAR:
925                 case DRM_FORMAT_MOD_BROADCOM_VC4_T_TILED:
926                 case DRM_FORMAT_MOD_BROADCOM_SAND64:
927                 case DRM_FORMAT_MOD_BROADCOM_SAND128:
928                         return true;
929                 default:
930                         return false;
931                 }
932         case DRM_FORMAT_NV12:
933         case DRM_FORMAT_NV21:
934                 switch (fourcc_mod_broadcom_mod(modifier)) {
935                 case DRM_FORMAT_MOD_LINEAR:
936                 case DRM_FORMAT_MOD_BROADCOM_SAND64:
937                 case DRM_FORMAT_MOD_BROADCOM_SAND128:
938                 case DRM_FORMAT_MOD_BROADCOM_SAND256:
939                         return true;
940                 default:
941                         return false;
942                 }
943         case DRM_FORMAT_YUV422:
944         case DRM_FORMAT_YVU422:
945         case DRM_FORMAT_YUV420:
946         case DRM_FORMAT_YVU420:
947         case DRM_FORMAT_NV16:
948         case DRM_FORMAT_NV61:
949         default:
950                 return (modifier == DRM_FORMAT_MOD_LINEAR);
951         }
952 }
953
954 static const struct drm_plane_funcs vc4_plane_funcs = {
955         .update_plane = drm_atomic_helper_update_plane,
956         .disable_plane = drm_atomic_helper_disable_plane,
957         .destroy = vc4_plane_destroy,
958         .set_property = NULL,
959         .reset = vc4_plane_reset,
960         .atomic_duplicate_state = vc4_plane_duplicate_state,
961         .atomic_destroy_state = vc4_plane_destroy_state,
962         .format_mod_supported = vc4_format_mod_supported,
963 };
964
965 struct drm_plane *vc4_plane_init(struct drm_device *dev,
966                                  enum drm_plane_type type)
967 {
968         struct drm_plane *plane = NULL;
969         struct vc4_plane *vc4_plane;
970         u32 formats[ARRAY_SIZE(hvs_formats)];
971         u32 num_formats = 0;
972         int ret = 0;
973         unsigned i;
974         static const uint64_t modifiers[] = {
975                 DRM_FORMAT_MOD_BROADCOM_VC4_T_TILED,
976                 DRM_FORMAT_MOD_BROADCOM_SAND128,
977                 DRM_FORMAT_MOD_BROADCOM_SAND64,
978                 DRM_FORMAT_MOD_BROADCOM_SAND256,
979                 DRM_FORMAT_MOD_LINEAR,
980                 DRM_FORMAT_MOD_INVALID
981         };
982
983         vc4_plane = devm_kzalloc(dev->dev, sizeof(*vc4_plane),
984                                  GFP_KERNEL);
985         if (!vc4_plane)
986                 return ERR_PTR(-ENOMEM);
987
988         for (i = 0; i < ARRAY_SIZE(hvs_formats); i++) {
989                 /* Don't allow YUV in cursor planes, since that means
990                  * tuning on the scaler, which we don't allow for the
991                  * cursor.
992                  */
993                 if (type != DRM_PLANE_TYPE_CURSOR ||
994                     hvs_formats[i].hvs < HVS_PIXEL_FORMAT_YCBCR_YUV420_3PLANE) {
995                         formats[num_formats++] = hvs_formats[i].drm;
996                 }
997         }
998         plane = &vc4_plane->base;
999         ret = drm_universal_plane_init(dev, plane, 0,
1000                                        &vc4_plane_funcs,
1001                                        formats, num_formats,
1002                                        modifiers, type, NULL);
1003
1004         drm_plane_helper_add(plane, &vc4_plane_helper_funcs);
1005
1006         drm_plane_create_alpha_property(plane);
1007
1008         return plane;
1009 }
This page took 0.104643 seconds and 4 git commands to generate.