Merge tag 'amlogic-dt-2' of https://git.kernel.org/pub/scm/linux/kernel/git/khilman...
[linux-2.6-block.git] / drivers / gpu / drm / vc4 / vc4_plane.c
1 /*
2  * Copyright (C) 2015 Broadcom
3  *
4  * This program is free software; you can redistribute it and/or modify
5  * it under the terms of the GNU General Public License version 2 as
6  * published by the Free Software Foundation.
7  */
8
9 /**
10  * DOC: VC4 plane module
11  *
12  * Each DRM plane is a layer of pixels being scanned out by the HVS.
13  *
14  * At atomic modeset check time, we compute the HVS display element
15  * state that would be necessary for displaying the plane (giving us a
16  * chance to figure out if a plane configuration is invalid), then at
17  * atomic flush time the CRTC will ask us to write our element state
18  * into the region of the HVS that it has allocated for us.
19  */
20
21 #include <drm/drm_atomic.h>
22 #include <drm/drm_atomic_helper.h>
23 #include <drm/drm_fb_cma_helper.h>
24 #include <drm/drm_plane_helper.h>
25
26 #include "uapi/drm/vc4_drm.h"
27 #include "vc4_drv.h"
28 #include "vc4_regs.h"
29
30 static const struct hvs_format {
31         u32 drm; /* DRM_FORMAT_* */
32         u32 hvs; /* HVS_FORMAT_* */
33         u32 pixel_order;
34 } hvs_formats[] = {
35         {
36                 .drm = DRM_FORMAT_XRGB8888, .hvs = HVS_PIXEL_FORMAT_RGBA8888,
37                 .pixel_order = HVS_PIXEL_ORDER_ABGR,
38         },
39         {
40                 .drm = DRM_FORMAT_ARGB8888, .hvs = HVS_PIXEL_FORMAT_RGBA8888,
41                 .pixel_order = HVS_PIXEL_ORDER_ABGR,
42         },
43         {
44                 .drm = DRM_FORMAT_ABGR8888, .hvs = HVS_PIXEL_FORMAT_RGBA8888,
45                 .pixel_order = HVS_PIXEL_ORDER_ARGB,
46         },
47         {
48                 .drm = DRM_FORMAT_XBGR8888, .hvs = HVS_PIXEL_FORMAT_RGBA8888,
49                 .pixel_order = HVS_PIXEL_ORDER_ARGB,
50         },
51         {
52                 .drm = DRM_FORMAT_RGB565, .hvs = HVS_PIXEL_FORMAT_RGB565,
53                 .pixel_order = HVS_PIXEL_ORDER_XRGB,
54         },
55         {
56                 .drm = DRM_FORMAT_BGR565, .hvs = HVS_PIXEL_FORMAT_RGB565,
57                 .pixel_order = HVS_PIXEL_ORDER_XBGR,
58         },
59         {
60                 .drm = DRM_FORMAT_ARGB1555, .hvs = HVS_PIXEL_FORMAT_RGBA5551,
61                 .pixel_order = HVS_PIXEL_ORDER_ABGR,
62         },
63         {
64                 .drm = DRM_FORMAT_XRGB1555, .hvs = HVS_PIXEL_FORMAT_RGBA5551,
65                 .pixel_order = HVS_PIXEL_ORDER_ABGR,
66         },
67         {
68                 .drm = DRM_FORMAT_RGB888, .hvs = HVS_PIXEL_FORMAT_RGB888,
69                 .pixel_order = HVS_PIXEL_ORDER_XRGB,
70         },
71         {
72                 .drm = DRM_FORMAT_BGR888, .hvs = HVS_PIXEL_FORMAT_RGB888,
73                 .pixel_order = HVS_PIXEL_ORDER_XBGR,
74         },
75         {
76                 .drm = DRM_FORMAT_YUV422,
77                 .hvs = HVS_PIXEL_FORMAT_YCBCR_YUV422_3PLANE,
78                 .pixel_order = HVS_PIXEL_ORDER_XYCBCR,
79         },
80         {
81                 .drm = DRM_FORMAT_YVU422,
82                 .hvs = HVS_PIXEL_FORMAT_YCBCR_YUV422_3PLANE,
83                 .pixel_order = HVS_PIXEL_ORDER_XYCRCB,
84         },
85         {
86                 .drm = DRM_FORMAT_YUV420,
87                 .hvs = HVS_PIXEL_FORMAT_YCBCR_YUV420_3PLANE,
88                 .pixel_order = HVS_PIXEL_ORDER_XYCBCR,
89         },
90         {
91                 .drm = DRM_FORMAT_YVU420,
92                 .hvs = HVS_PIXEL_FORMAT_YCBCR_YUV420_3PLANE,
93                 .pixel_order = HVS_PIXEL_ORDER_XYCRCB,
94         },
95         {
96                 .drm = DRM_FORMAT_NV12,
97                 .hvs = HVS_PIXEL_FORMAT_YCBCR_YUV420_2PLANE,
98                 .pixel_order = HVS_PIXEL_ORDER_XYCBCR,
99         },
100         {
101                 .drm = DRM_FORMAT_NV21,
102                 .hvs = HVS_PIXEL_FORMAT_YCBCR_YUV420_2PLANE,
103                 .pixel_order = HVS_PIXEL_ORDER_XYCRCB,
104         },
105         {
106                 .drm = DRM_FORMAT_NV16,
107                 .hvs = HVS_PIXEL_FORMAT_YCBCR_YUV422_2PLANE,
108                 .pixel_order = HVS_PIXEL_ORDER_XYCBCR,
109         },
110         {
111                 .drm = DRM_FORMAT_NV61,
112                 .hvs = HVS_PIXEL_FORMAT_YCBCR_YUV422_2PLANE,
113                 .pixel_order = HVS_PIXEL_ORDER_XYCRCB,
114         },
115 };
116
117 static const struct hvs_format *vc4_get_hvs_format(u32 drm_format)
118 {
119         unsigned i;
120
121         for (i = 0; i < ARRAY_SIZE(hvs_formats); i++) {
122                 if (hvs_formats[i].drm == drm_format)
123                         return &hvs_formats[i];
124         }
125
126         return NULL;
127 }
128
129 static enum vc4_scaling_mode vc4_get_scaling_mode(u32 src, u32 dst)
130 {
131         if (dst > src)
132                 return VC4_SCALING_PPF;
133         else if (dst < src)
134                 return VC4_SCALING_TPZ;
135         else
136                 return VC4_SCALING_NONE;
137 }
138
139 static bool plane_enabled(struct drm_plane_state *state)
140 {
141         return state->fb && state->crtc;
142 }
143
144 static struct drm_plane_state *vc4_plane_duplicate_state(struct drm_plane *plane)
145 {
146         struct vc4_plane_state *vc4_state;
147
148         if (WARN_ON(!plane->state))
149                 return NULL;
150
151         vc4_state = kmemdup(plane->state, sizeof(*vc4_state), GFP_KERNEL);
152         if (!vc4_state)
153                 return NULL;
154
155         memset(&vc4_state->lbm, 0, sizeof(vc4_state->lbm));
156
157         __drm_atomic_helper_plane_duplicate_state(plane, &vc4_state->base);
158
159         if (vc4_state->dlist) {
160                 vc4_state->dlist = kmemdup(vc4_state->dlist,
161                                            vc4_state->dlist_count * 4,
162                                            GFP_KERNEL);
163                 if (!vc4_state->dlist) {
164                         kfree(vc4_state);
165                         return NULL;
166                 }
167                 vc4_state->dlist_size = vc4_state->dlist_count;
168         }
169
170         return &vc4_state->base;
171 }
172
173 static void vc4_plane_destroy_state(struct drm_plane *plane,
174                                     struct drm_plane_state *state)
175 {
176         struct vc4_dev *vc4 = to_vc4_dev(plane->dev);
177         struct vc4_plane_state *vc4_state = to_vc4_plane_state(state);
178
179         if (vc4_state->lbm.allocated) {
180                 unsigned long irqflags;
181
182                 spin_lock_irqsave(&vc4->hvs->mm_lock, irqflags);
183                 drm_mm_remove_node(&vc4_state->lbm);
184                 spin_unlock_irqrestore(&vc4->hvs->mm_lock, irqflags);
185         }
186
187         kfree(vc4_state->dlist);
188         __drm_atomic_helper_plane_destroy_state(&vc4_state->base);
189         kfree(state);
190 }
191
192 /* Called during init to allocate the plane's atomic state. */
193 static void vc4_plane_reset(struct drm_plane *plane)
194 {
195         struct vc4_plane_state *vc4_state;
196
197         WARN_ON(plane->state);
198
199         vc4_state = kzalloc(sizeof(*vc4_state), GFP_KERNEL);
200         if (!vc4_state)
201                 return;
202
203         plane->state = &vc4_state->base;
204         vc4_state->base.plane = plane;
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(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         u32 h_subsample = 1;
270         u32 v_subsample = 1;
271         int i;
272
273         for (i = 0; i < num_planes; i++)
274                 vc4_state->offsets[i] = bo->paddr + fb->offsets[i];
275
276         /* We don't support subpixel source positioning for scaling. */
277         if ((state->src_x & subpixel_src_mask) ||
278             (state->src_y & subpixel_src_mask) ||
279             (state->src_w & subpixel_src_mask) ||
280             (state->src_h & subpixel_src_mask)) {
281                 return -EINVAL;
282         }
283
284         vc4_state->src_x = state->src_x >> 16;
285         vc4_state->src_y = state->src_y >> 16;
286         vc4_state->src_w[0] = state->src_w >> 16;
287         vc4_state->src_h[0] = state->src_h >> 16;
288
289         vc4_state->crtc_x = state->crtc_x;
290         vc4_state->crtc_y = state->crtc_y;
291         vc4_state->crtc_w = state->crtc_w;
292         vc4_state->crtc_h = state->crtc_h;
293
294         vc4_state->x_scaling[0] = vc4_get_scaling_mode(vc4_state->src_w[0],
295                                                        vc4_state->crtc_w);
296         vc4_state->y_scaling[0] = vc4_get_scaling_mode(vc4_state->src_h[0],
297                                                        vc4_state->crtc_h);
298
299         if (num_planes > 1) {
300                 vc4_state->is_yuv = true;
301
302                 h_subsample = drm_format_horz_chroma_subsampling(format);
303                 v_subsample = drm_format_vert_chroma_subsampling(format);
304                 vc4_state->src_w[1] = vc4_state->src_w[0] / h_subsample;
305                 vc4_state->src_h[1] = vc4_state->src_h[0] / v_subsample;
306
307                 vc4_state->x_scaling[1] =
308                         vc4_get_scaling_mode(vc4_state->src_w[1],
309                                              vc4_state->crtc_w);
310                 vc4_state->y_scaling[1] =
311                         vc4_get_scaling_mode(vc4_state->src_h[1],
312                                              vc4_state->crtc_h);
313
314                 /* YUV conversion requires that scaling be enabled,
315                  * even on a plane that's otherwise 1:1.  Choose TPZ
316                  * for simplicity.
317                  */
318                 if (vc4_state->x_scaling[0] == VC4_SCALING_NONE)
319                         vc4_state->x_scaling[0] = VC4_SCALING_TPZ;
320                 if (vc4_state->y_scaling[0] == VC4_SCALING_NONE)
321                         vc4_state->y_scaling[0] = VC4_SCALING_TPZ;
322         }
323
324         vc4_state->is_unity = (vc4_state->x_scaling[0] == VC4_SCALING_NONE &&
325                                vc4_state->y_scaling[0] == VC4_SCALING_NONE &&
326                                vc4_state->x_scaling[1] == VC4_SCALING_NONE &&
327                                vc4_state->y_scaling[1] == VC4_SCALING_NONE);
328
329         /* No configuring scaling on the cursor plane, since it gets
330            non-vblank-synced updates, and scaling requires requires
331            LBM changes which have to be vblank-synced.
332          */
333         if (plane->type == DRM_PLANE_TYPE_CURSOR && !vc4_state->is_unity)
334                 return -EINVAL;
335
336         /* Clamp the on-screen start x/y to 0.  The hardware doesn't
337          * support negative y, and negative x wastes bandwidth.
338          */
339         if (vc4_state->crtc_x < 0) {
340                 for (i = 0; i < num_planes; i++) {
341                         u32 cpp = fb->format->cpp[i];
342                         u32 subs = ((i == 0) ? 1 : h_subsample);
343
344                         vc4_state->offsets[i] += (cpp *
345                                                   (-vc4_state->crtc_x) / subs);
346                 }
347                 vc4_state->src_w[0] += vc4_state->crtc_x;
348                 vc4_state->src_w[1] += vc4_state->crtc_x / h_subsample;
349                 vc4_state->crtc_x = 0;
350         }
351
352         if (vc4_state->crtc_y < 0) {
353                 for (i = 0; i < num_planes; i++) {
354                         u32 subs = ((i == 0) ? 1 : v_subsample);
355
356                         vc4_state->offsets[i] += (fb->pitches[i] *
357                                                   (-vc4_state->crtc_y) / subs);
358                 }
359                 vc4_state->src_h[0] += vc4_state->crtc_y;
360                 vc4_state->src_h[1] += vc4_state->crtc_y / v_subsample;
361                 vc4_state->crtc_y = 0;
362         }
363
364         return 0;
365 }
366
367 static void vc4_write_tpz(struct vc4_plane_state *vc4_state, u32 src, u32 dst)
368 {
369         u32 scale, recip;
370
371         scale = (1 << 16) * src / dst;
372
373         /* The specs note that while the reciprocal would be defined
374          * as (1<<32)/scale, ~0 is close enough.
375          */
376         recip = ~0 / scale;
377
378         vc4_dlist_write(vc4_state,
379                         VC4_SET_FIELD(scale, SCALER_TPZ0_SCALE) |
380                         VC4_SET_FIELD(0, SCALER_TPZ0_IPHASE));
381         vc4_dlist_write(vc4_state,
382                         VC4_SET_FIELD(recip, SCALER_TPZ1_RECIP));
383 }
384
385 static void vc4_write_ppf(struct vc4_plane_state *vc4_state, u32 src, u32 dst)
386 {
387         u32 scale = (1 << 16) * src / dst;
388
389         vc4_dlist_write(vc4_state,
390                         SCALER_PPF_AGC |
391                         VC4_SET_FIELD(scale, SCALER_PPF_SCALE) |
392                         VC4_SET_FIELD(0, SCALER_PPF_IPHASE));
393 }
394
395 static u32 vc4_lbm_size(struct drm_plane_state *state)
396 {
397         struct vc4_plane_state *vc4_state = to_vc4_plane_state(state);
398         /* This is the worst case number.  One of the two sizes will
399          * be used depending on the scaling configuration.
400          */
401         u32 pix_per_line = max(vc4_state->src_w[0], (u32)vc4_state->crtc_w);
402         u32 lbm;
403
404         if (!vc4_state->is_yuv) {
405                 if (vc4_state->is_unity)
406                         return 0;
407                 else if (vc4_state->y_scaling[0] == VC4_SCALING_TPZ)
408                         lbm = pix_per_line * 8;
409                 else {
410                         /* In special cases, this multiplier might be 12. */
411                         lbm = pix_per_line * 16;
412                 }
413         } else {
414                 /* There are cases for this going down to a multiplier
415                  * of 2, but according to the firmware source, the
416                  * table in the docs is somewhat wrong.
417                  */
418                 lbm = pix_per_line * 16;
419         }
420
421         lbm = roundup(lbm, 32);
422
423         return lbm;
424 }
425
426 static void vc4_write_scaling_parameters(struct drm_plane_state *state,
427                                          int channel)
428 {
429         struct vc4_plane_state *vc4_state = to_vc4_plane_state(state);
430
431         /* Ch0 H-PPF Word 0: Scaling Parameters */
432         if (vc4_state->x_scaling[channel] == VC4_SCALING_PPF) {
433                 vc4_write_ppf(vc4_state,
434                               vc4_state->src_w[channel], vc4_state->crtc_w);
435         }
436
437         /* Ch0 V-PPF Words 0-1: Scaling Parameters, Context */
438         if (vc4_state->y_scaling[channel] == VC4_SCALING_PPF) {
439                 vc4_write_ppf(vc4_state,
440                               vc4_state->src_h[channel], vc4_state->crtc_h);
441                 vc4_dlist_write(vc4_state, 0xc0c0c0c0);
442         }
443
444         /* Ch0 H-TPZ Words 0-1: Scaling Parameters, Recip */
445         if (vc4_state->x_scaling[channel] == VC4_SCALING_TPZ) {
446                 vc4_write_tpz(vc4_state,
447                               vc4_state->src_w[channel], vc4_state->crtc_w);
448         }
449
450         /* Ch0 V-TPZ Words 0-2: Scaling Parameters, Recip, Context */
451         if (vc4_state->y_scaling[channel] == VC4_SCALING_TPZ) {
452                 vc4_write_tpz(vc4_state,
453                               vc4_state->src_h[channel], vc4_state->crtc_h);
454                 vc4_dlist_write(vc4_state, 0xc0c0c0c0);
455         }
456 }
457
458 /* Writes out a full display list for an active plane to the plane's
459  * private dlist state.
460  */
461 static int vc4_plane_mode_set(struct drm_plane *plane,
462                               struct drm_plane_state *state)
463 {
464         struct vc4_dev *vc4 = to_vc4_dev(plane->dev);
465         struct vc4_plane_state *vc4_state = to_vc4_plane_state(state);
466         struct drm_framebuffer *fb = state->fb;
467         u32 ctl0_offset = vc4_state->dlist_count;
468         const struct hvs_format *format = vc4_get_hvs_format(fb->format->format);
469         int num_planes = drm_format_num_planes(format->drm);
470         bool covers_screen;
471         u32 scl0, scl1, pitch0;
472         u32 lbm_size, tiling;
473         unsigned long irqflags;
474         int ret, i;
475
476         ret = vc4_plane_setup_clipping_and_scaling(state);
477         if (ret)
478                 return ret;
479
480         /* Allocate the LBM memory that the HVS will use for temporary
481          * storage due to our scaling/format conversion.
482          */
483         lbm_size = vc4_lbm_size(state);
484         if (lbm_size) {
485                 if (!vc4_state->lbm.allocated) {
486                         spin_lock_irqsave(&vc4->hvs->mm_lock, irqflags);
487                         ret = drm_mm_insert_node_generic(&vc4->hvs->lbm_mm,
488                                                          &vc4_state->lbm,
489                                                          lbm_size, 32, 0, 0);
490                         spin_unlock_irqrestore(&vc4->hvs->mm_lock, irqflags);
491                 } else {
492                         WARN_ON_ONCE(lbm_size != vc4_state->lbm.size);
493                 }
494         }
495
496         if (ret)
497                 return ret;
498
499         /* SCL1 is used for Cb/Cr scaling of planar formats.  For RGB
500          * and 4:4:4, scl1 should be set to scl0 so both channels of
501          * the scaler do the same thing.  For YUV, the Y plane needs
502          * to be put in channel 1 and Cb/Cr in channel 0, so we swap
503          * the scl fields here.
504          */
505         if (num_planes == 1) {
506                 scl0 = vc4_get_scl_field(state, 0);
507                 scl1 = scl0;
508         } else {
509                 scl0 = vc4_get_scl_field(state, 1);
510                 scl1 = vc4_get_scl_field(state, 0);
511         }
512
513         switch (fb->modifier) {
514         case DRM_FORMAT_MOD_LINEAR:
515                 tiling = SCALER_CTL0_TILING_LINEAR;
516                 pitch0 = VC4_SET_FIELD(fb->pitches[0], SCALER_SRC_PITCH);
517                 break;
518
519         case DRM_FORMAT_MOD_BROADCOM_VC4_T_TILED: {
520                 /* For T-tiled, the FB pitch is "how many bytes from
521                  * one row to the next, such that pitch * tile_h ==
522                  * tile_size * tiles_per_row."
523                  */
524                 u32 tile_size_shift = 12; /* T tiles are 4kb */
525                 u32 tile_h_shift = 5; /* 16 and 32bpp are 32 pixels high */
526                 u32 tiles_w = fb->pitches[0] >> (tile_size_shift - tile_h_shift);
527
528                 tiling = SCALER_CTL0_TILING_256B_OR_T;
529
530                 pitch0 = (VC4_SET_FIELD(0, SCALER_PITCH0_TILE_Y_OFFSET) |
531                           VC4_SET_FIELD(0, SCALER_PITCH0_TILE_WIDTH_L) |
532                           VC4_SET_FIELD(tiles_w, SCALER_PITCH0_TILE_WIDTH_R));
533                 break;
534         }
535
536         default:
537                 DRM_DEBUG_KMS("Unsupported FB tiling flag 0x%16llx",
538                               (long long)fb->modifier);
539                 return -EINVAL;
540         }
541
542         /* Control word */
543         vc4_dlist_write(vc4_state,
544                         SCALER_CTL0_VALID |
545                         (format->pixel_order << SCALER_CTL0_ORDER_SHIFT) |
546                         (format->hvs << SCALER_CTL0_PIXEL_FORMAT_SHIFT) |
547                         VC4_SET_FIELD(tiling, SCALER_CTL0_TILING) |
548                         (vc4_state->is_unity ? SCALER_CTL0_UNITY : 0) |
549                         VC4_SET_FIELD(scl0, SCALER_CTL0_SCL0) |
550                         VC4_SET_FIELD(scl1, SCALER_CTL0_SCL1));
551
552         /* Position Word 0: Image Positions and Alpha Value */
553         vc4_state->pos0_offset = vc4_state->dlist_count;
554         vc4_dlist_write(vc4_state,
555                         VC4_SET_FIELD(0xff, SCALER_POS0_FIXED_ALPHA) |
556                         VC4_SET_FIELD(vc4_state->crtc_x, SCALER_POS0_START_X) |
557                         VC4_SET_FIELD(vc4_state->crtc_y, SCALER_POS0_START_Y));
558
559         /* Position Word 1: Scaled Image Dimensions. */
560         if (!vc4_state->is_unity) {
561                 vc4_dlist_write(vc4_state,
562                                 VC4_SET_FIELD(vc4_state->crtc_w,
563                                               SCALER_POS1_SCL_WIDTH) |
564                                 VC4_SET_FIELD(vc4_state->crtc_h,
565                                               SCALER_POS1_SCL_HEIGHT));
566         }
567
568         /* Position Word 2: Source Image Size, Alpha */
569         vc4_state->pos2_offset = vc4_state->dlist_count;
570         vc4_dlist_write(vc4_state,
571                         VC4_SET_FIELD(fb->format->has_alpha ?
572                                       SCALER_POS2_ALPHA_MODE_PIPELINE :
573                                       SCALER_POS2_ALPHA_MODE_FIXED,
574                                       SCALER_POS2_ALPHA_MODE) |
575                         (fb->format->has_alpha ? SCALER_POS2_ALPHA_PREMULT : 0) |
576                         VC4_SET_FIELD(vc4_state->src_w[0], SCALER_POS2_WIDTH) |
577                         VC4_SET_FIELD(vc4_state->src_h[0], SCALER_POS2_HEIGHT));
578
579         /* Position Word 3: Context.  Written by the HVS. */
580         vc4_dlist_write(vc4_state, 0xc0c0c0c0);
581
582
583         /* Pointer Word 0/1/2: RGB / Y / Cb / Cr Pointers
584          *
585          * The pointers may be any byte address.
586          */
587         vc4_state->ptr0_offset = vc4_state->dlist_count;
588         for (i = 0; i < num_planes; i++)
589                 vc4_dlist_write(vc4_state, vc4_state->offsets[i]);
590
591         /* Pointer Context Word 0/1/2: Written by the HVS */
592         for (i = 0; i < num_planes; i++)
593                 vc4_dlist_write(vc4_state, 0xc0c0c0c0);
594
595         /* Pitch word 0 */
596         vc4_dlist_write(vc4_state, pitch0);
597
598         /* Pitch word 1/2 */
599         for (i = 1; i < num_planes; i++) {
600                 vc4_dlist_write(vc4_state,
601                                 VC4_SET_FIELD(fb->pitches[i], SCALER_SRC_PITCH));
602         }
603
604         /* Colorspace conversion words */
605         if (vc4_state->is_yuv) {
606                 vc4_dlist_write(vc4_state, SCALER_CSC0_ITR_R_601_5);
607                 vc4_dlist_write(vc4_state, SCALER_CSC1_ITR_R_601_5);
608                 vc4_dlist_write(vc4_state, SCALER_CSC2_ITR_R_601_5);
609         }
610
611         if (!vc4_state->is_unity) {
612                 /* LBM Base Address. */
613                 if (vc4_state->y_scaling[0] != VC4_SCALING_NONE ||
614                     vc4_state->y_scaling[1] != VC4_SCALING_NONE) {
615                         vc4_dlist_write(vc4_state, vc4_state->lbm.start);
616                 }
617
618                 if (num_planes > 1) {
619                         /* Emit Cb/Cr as channel 0 and Y as channel
620                          * 1. This matches how we set up scl0/scl1
621                          * above.
622                          */
623                         vc4_write_scaling_parameters(state, 1);
624                 }
625                 vc4_write_scaling_parameters(state, 0);
626
627                 /* If any PPF setup was done, then all the kernel
628                  * pointers get uploaded.
629                  */
630                 if (vc4_state->x_scaling[0] == VC4_SCALING_PPF ||
631                     vc4_state->y_scaling[0] == VC4_SCALING_PPF ||
632                     vc4_state->x_scaling[1] == VC4_SCALING_PPF ||
633                     vc4_state->y_scaling[1] == VC4_SCALING_PPF) {
634                         u32 kernel = VC4_SET_FIELD(vc4->hvs->mitchell_netravali_filter.start,
635                                                    SCALER_PPF_KERNEL_OFFSET);
636
637                         /* HPPF plane 0 */
638                         vc4_dlist_write(vc4_state, kernel);
639                         /* VPPF plane 0 */
640                         vc4_dlist_write(vc4_state, kernel);
641                         /* HPPF plane 1 */
642                         vc4_dlist_write(vc4_state, kernel);
643                         /* VPPF plane 1 */
644                         vc4_dlist_write(vc4_state, kernel);
645                 }
646         }
647
648         vc4_state->dlist[ctl0_offset] |=
649                 VC4_SET_FIELD(vc4_state->dlist_count, SCALER_CTL0_SIZE);
650
651         /* crtc_* are already clipped coordinates. */
652         covers_screen = vc4_state->crtc_x == 0 && vc4_state->crtc_y == 0 &&
653                         vc4_state->crtc_w == state->crtc->mode.hdisplay &&
654                         vc4_state->crtc_h == state->crtc->mode.vdisplay;
655         /* Background fill might be necessary when the plane has per-pixel
656          * alpha content and blends from the background or does not cover
657          * the entire screen.
658          */
659         vc4_state->needs_bg_fill = fb->format->has_alpha || !covers_screen;
660
661         return 0;
662 }
663
664 /* If a modeset involves changing the setup of a plane, the atomic
665  * infrastructure will call this to validate a proposed plane setup.
666  * However, if a plane isn't getting updated, this (and the
667  * corresponding vc4_plane_atomic_update) won't get called.  Thus, we
668  * compute the dlist here and have all active plane dlists get updated
669  * in the CRTC's flush.
670  */
671 static int vc4_plane_atomic_check(struct drm_plane *plane,
672                                   struct drm_plane_state *state)
673 {
674         struct vc4_plane_state *vc4_state = to_vc4_plane_state(state);
675
676         vc4_state->dlist_count = 0;
677
678         if (plane_enabled(state))
679                 return vc4_plane_mode_set(plane, state);
680         else
681                 return 0;
682 }
683
684 static void vc4_plane_atomic_update(struct drm_plane *plane,
685                                     struct drm_plane_state *old_state)
686 {
687         /* No contents here.  Since we don't know where in the CRTC's
688          * dlist we should be stored, our dlist is uploaded to the
689          * hardware with vc4_plane_write_dlist() at CRTC atomic_flush
690          * time.
691          */
692 }
693
694 u32 vc4_plane_write_dlist(struct drm_plane *plane, u32 __iomem *dlist)
695 {
696         struct vc4_plane_state *vc4_state = to_vc4_plane_state(plane->state);
697         int i;
698
699         vc4_state->hw_dlist = dlist;
700
701         /* Can't memcpy_toio() because it needs to be 32-bit writes. */
702         for (i = 0; i < vc4_state->dlist_count; i++)
703                 writel(vc4_state->dlist[i], &dlist[i]);
704
705         return vc4_state->dlist_count;
706 }
707
708 u32 vc4_plane_dlist_size(const struct drm_plane_state *state)
709 {
710         const struct vc4_plane_state *vc4_state =
711                 container_of(state, typeof(*vc4_state), base);
712
713         return vc4_state->dlist_count;
714 }
715
716 /* Updates the plane to immediately (well, once the FIFO needs
717  * refilling) scan out from at a new framebuffer.
718  */
719 void vc4_plane_async_set_fb(struct drm_plane *plane, struct drm_framebuffer *fb)
720 {
721         struct vc4_plane_state *vc4_state = to_vc4_plane_state(plane->state);
722         struct drm_gem_cma_object *bo = drm_fb_cma_get_gem_obj(fb, 0);
723         uint32_t addr;
724
725         /* We're skipping the address adjustment for negative origin,
726          * because this is only called on the primary plane.
727          */
728         WARN_ON_ONCE(plane->state->crtc_x < 0 || plane->state->crtc_y < 0);
729         addr = bo->paddr + fb->offsets[0];
730
731         /* Write the new address into the hardware immediately.  The
732          * scanout will start from this address as soon as the FIFO
733          * needs to refill with pixels.
734          */
735         writel(addr, &vc4_state->hw_dlist[vc4_state->ptr0_offset]);
736
737         /* Also update the CPU-side dlist copy, so that any later
738          * atomic updates that don't do a new modeset on our plane
739          * also use our updated address.
740          */
741         vc4_state->dlist[vc4_state->ptr0_offset] = addr;
742 }
743
744 static int vc4_prepare_fb(struct drm_plane *plane,
745                           struct drm_plane_state *state)
746 {
747         struct vc4_bo *bo;
748         struct dma_fence *fence;
749         int ret;
750
751         if ((plane->state->fb == state->fb) || !state->fb)
752                 return 0;
753
754         bo = to_vc4_bo(&drm_fb_cma_get_gem_obj(state->fb, 0)->base);
755
756         ret = vc4_bo_inc_usecnt(bo);
757         if (ret)
758                 return ret;
759
760         fence = reservation_object_get_excl_rcu(bo->resv);
761         drm_atomic_set_fence_for_plane(state, fence);
762
763         return 0;
764 }
765
766 static void vc4_cleanup_fb(struct drm_plane *plane,
767                            struct drm_plane_state *state)
768 {
769         struct vc4_bo *bo;
770
771         if (plane->state->fb == state->fb || !state->fb)
772                 return;
773
774         bo = to_vc4_bo(&drm_fb_cma_get_gem_obj(state->fb, 0)->base);
775         vc4_bo_dec_usecnt(bo);
776 }
777
778 static const struct drm_plane_helper_funcs vc4_plane_helper_funcs = {
779         .atomic_check = vc4_plane_atomic_check,
780         .atomic_update = vc4_plane_atomic_update,
781         .prepare_fb = vc4_prepare_fb,
782         .cleanup_fb = vc4_cleanup_fb,
783 };
784
785 static void vc4_plane_destroy(struct drm_plane *plane)
786 {
787         drm_plane_helper_disable(plane);
788         drm_plane_cleanup(plane);
789 }
790
791 /* Implements immediate (non-vblank-synced) updates of the cursor
792  * position, or falls back to the atomic helper otherwise.
793  */
794 static int
795 vc4_update_plane(struct drm_plane *plane,
796                  struct drm_crtc *crtc,
797                  struct drm_framebuffer *fb,
798                  int crtc_x, int crtc_y,
799                  unsigned int crtc_w, unsigned int crtc_h,
800                  uint32_t src_x, uint32_t src_y,
801                  uint32_t src_w, uint32_t src_h,
802                  struct drm_modeset_acquire_ctx *ctx)
803 {
804         struct drm_plane_state *plane_state;
805         struct vc4_plane_state *vc4_state;
806
807         if (plane != crtc->cursor)
808                 goto out;
809
810         plane_state = plane->state;
811         vc4_state = to_vc4_plane_state(plane_state);
812
813         if (!plane_state)
814                 goto out;
815
816         /* No configuring new scaling in the fast path. */
817         if (crtc_w != plane_state->crtc_w ||
818             crtc_h != plane_state->crtc_h ||
819             src_w != plane_state->src_w ||
820             src_h != plane_state->src_h) {
821                 goto out;
822         }
823
824         if (fb != plane_state->fb) {
825                 drm_atomic_set_fb_for_plane(plane->state, fb);
826                 vc4_plane_async_set_fb(plane, fb);
827         }
828
829         /* Set the cursor's position on the screen.  This is the
830          * expected change from the drm_mode_cursor_universal()
831          * helper.
832          */
833         plane_state->crtc_x = crtc_x;
834         plane_state->crtc_y = crtc_y;
835
836         /* Allow changing the start position within the cursor BO, if
837          * that matters.
838          */
839         plane_state->src_x = src_x;
840         plane_state->src_y = src_y;
841
842         /* Update the display list based on the new crtc_x/y. */
843         vc4_plane_atomic_check(plane, plane_state);
844
845         /* Note that we can't just call vc4_plane_write_dlist()
846          * because that would smash the context data that the HVS is
847          * currently using.
848          */
849         writel(vc4_state->dlist[vc4_state->pos0_offset],
850                &vc4_state->hw_dlist[vc4_state->pos0_offset]);
851         writel(vc4_state->dlist[vc4_state->pos2_offset],
852                &vc4_state->hw_dlist[vc4_state->pos2_offset]);
853         writel(vc4_state->dlist[vc4_state->ptr0_offset],
854                &vc4_state->hw_dlist[vc4_state->ptr0_offset]);
855
856         return 0;
857
858 out:
859         return drm_atomic_helper_update_plane(plane, crtc, fb,
860                                               crtc_x, crtc_y,
861                                               crtc_w, crtc_h,
862                                               src_x, src_y,
863                                               src_w, src_h,
864                                               ctx);
865 }
866
867 static bool vc4_format_mod_supported(struct drm_plane *plane,
868                                      uint32_t format,
869                                      uint64_t modifier)
870 {
871         /* Support T_TILING for RGB formats only. */
872         switch (format) {
873         case DRM_FORMAT_XRGB8888:
874         case DRM_FORMAT_ARGB8888:
875         case DRM_FORMAT_ABGR8888:
876         case DRM_FORMAT_XBGR8888:
877         case DRM_FORMAT_RGB565:
878         case DRM_FORMAT_BGR565:
879         case DRM_FORMAT_ARGB1555:
880         case DRM_FORMAT_XRGB1555:
881                 return true;
882         case DRM_FORMAT_YUV422:
883         case DRM_FORMAT_YVU422:
884         case DRM_FORMAT_YUV420:
885         case DRM_FORMAT_YVU420:
886         case DRM_FORMAT_NV12:
887         case DRM_FORMAT_NV16:
888         default:
889                 return (modifier == DRM_FORMAT_MOD_LINEAR);
890         }
891 }
892
893 static const struct drm_plane_funcs vc4_plane_funcs = {
894         .update_plane = vc4_update_plane,
895         .disable_plane = drm_atomic_helper_disable_plane,
896         .destroy = vc4_plane_destroy,
897         .set_property = NULL,
898         .reset = vc4_plane_reset,
899         .atomic_duplicate_state = vc4_plane_duplicate_state,
900         .atomic_destroy_state = vc4_plane_destroy_state,
901         .format_mod_supported = vc4_format_mod_supported,
902 };
903
904 struct drm_plane *vc4_plane_init(struct drm_device *dev,
905                                  enum drm_plane_type type)
906 {
907         struct drm_plane *plane = NULL;
908         struct vc4_plane *vc4_plane;
909         u32 formats[ARRAY_SIZE(hvs_formats)];
910         u32 num_formats = 0;
911         int ret = 0;
912         unsigned i;
913         static const uint64_t modifiers[] = {
914                 DRM_FORMAT_MOD_BROADCOM_VC4_T_TILED,
915                 DRM_FORMAT_MOD_LINEAR,
916                 DRM_FORMAT_MOD_INVALID
917         };
918
919         vc4_plane = devm_kzalloc(dev->dev, sizeof(*vc4_plane),
920                                  GFP_KERNEL);
921         if (!vc4_plane)
922                 return ERR_PTR(-ENOMEM);
923
924         for (i = 0; i < ARRAY_SIZE(hvs_formats); i++) {
925                 /* Don't allow YUV in cursor planes, since that means
926                  * tuning on the scaler, which we don't allow for the
927                  * cursor.
928                  */
929                 if (type != DRM_PLANE_TYPE_CURSOR ||
930                     hvs_formats[i].hvs < HVS_PIXEL_FORMAT_YCBCR_YUV420_3PLANE) {
931                         formats[num_formats++] = hvs_formats[i].drm;
932                 }
933         }
934         plane = &vc4_plane->base;
935         ret = drm_universal_plane_init(dev, plane, 0,
936                                        &vc4_plane_funcs,
937                                        formats, num_formats,
938                                        modifiers, type, NULL);
939
940         drm_plane_helper_add(plane, &vc4_plane_helper_funcs);
941
942         return plane;
943 }