2 * Copyright (C) 2015 Broadcom
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.
10 * DOC: VC4 plane module
12 * Each DRM plane is a layer of pixels being scanned out by the HVS.
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.
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>
27 #include "uapi/drm/vc4_drm.h"
31 static const struct hvs_format {
32 u32 drm; /* DRM_FORMAT_* */
33 u32 hvs; /* HVS_FORMAT_* */
37 .drm = DRM_FORMAT_XRGB8888, .hvs = HVS_PIXEL_FORMAT_RGBA8888,
38 .pixel_order = HVS_PIXEL_ORDER_ABGR,
41 .drm = DRM_FORMAT_ARGB8888, .hvs = HVS_PIXEL_FORMAT_RGBA8888,
42 .pixel_order = HVS_PIXEL_ORDER_ABGR,
45 .drm = DRM_FORMAT_ABGR8888, .hvs = HVS_PIXEL_FORMAT_RGBA8888,
46 .pixel_order = HVS_PIXEL_ORDER_ARGB,
49 .drm = DRM_FORMAT_XBGR8888, .hvs = HVS_PIXEL_FORMAT_RGBA8888,
50 .pixel_order = HVS_PIXEL_ORDER_ARGB,
53 .drm = DRM_FORMAT_RGB565, .hvs = HVS_PIXEL_FORMAT_RGB565,
54 .pixel_order = HVS_PIXEL_ORDER_XRGB,
57 .drm = DRM_FORMAT_BGR565, .hvs = HVS_PIXEL_FORMAT_RGB565,
58 .pixel_order = HVS_PIXEL_ORDER_XBGR,
61 .drm = DRM_FORMAT_ARGB1555, .hvs = HVS_PIXEL_FORMAT_RGBA5551,
62 .pixel_order = HVS_PIXEL_ORDER_ABGR,
65 .drm = DRM_FORMAT_XRGB1555, .hvs = HVS_PIXEL_FORMAT_RGBA5551,
66 .pixel_order = HVS_PIXEL_ORDER_ABGR,
69 .drm = DRM_FORMAT_RGB888, .hvs = HVS_PIXEL_FORMAT_RGB888,
70 .pixel_order = HVS_PIXEL_ORDER_XRGB,
73 .drm = DRM_FORMAT_BGR888, .hvs = HVS_PIXEL_FORMAT_RGB888,
74 .pixel_order = HVS_PIXEL_ORDER_XBGR,
77 .drm = DRM_FORMAT_YUV422,
78 .hvs = HVS_PIXEL_FORMAT_YCBCR_YUV422_3PLANE,
79 .pixel_order = HVS_PIXEL_ORDER_XYCBCR,
82 .drm = DRM_FORMAT_YVU422,
83 .hvs = HVS_PIXEL_FORMAT_YCBCR_YUV422_3PLANE,
84 .pixel_order = HVS_PIXEL_ORDER_XYCRCB,
87 .drm = DRM_FORMAT_YUV420,
88 .hvs = HVS_PIXEL_FORMAT_YCBCR_YUV420_3PLANE,
89 .pixel_order = HVS_PIXEL_ORDER_XYCBCR,
92 .drm = DRM_FORMAT_YVU420,
93 .hvs = HVS_PIXEL_FORMAT_YCBCR_YUV420_3PLANE,
94 .pixel_order = HVS_PIXEL_ORDER_XYCRCB,
97 .drm = DRM_FORMAT_NV12,
98 .hvs = HVS_PIXEL_FORMAT_YCBCR_YUV420_2PLANE,
99 .pixel_order = HVS_PIXEL_ORDER_XYCBCR,
102 .drm = DRM_FORMAT_NV21,
103 .hvs = HVS_PIXEL_FORMAT_YCBCR_YUV420_2PLANE,
104 .pixel_order = HVS_PIXEL_ORDER_XYCRCB,
107 .drm = DRM_FORMAT_NV16,
108 .hvs = HVS_PIXEL_FORMAT_YCBCR_YUV422_2PLANE,
109 .pixel_order = HVS_PIXEL_ORDER_XYCBCR,
112 .drm = DRM_FORMAT_NV61,
113 .hvs = HVS_PIXEL_FORMAT_YCBCR_YUV422_2PLANE,
114 .pixel_order = HVS_PIXEL_ORDER_XYCRCB,
118 static const struct hvs_format *vc4_get_hvs_format(u32 drm_format)
122 for (i = 0; i < ARRAY_SIZE(hvs_formats); i++) {
123 if (hvs_formats[i].drm == drm_format)
124 return &hvs_formats[i];
130 static enum vc4_scaling_mode vc4_get_scaling_mode(u32 src, u32 dst)
133 return VC4_SCALING_NONE;
134 if (3 * dst >= 2 * src)
135 return VC4_SCALING_PPF;
137 return VC4_SCALING_TPZ;
140 static bool plane_enabled(struct drm_plane_state *state)
142 return state->fb && state->crtc;
145 static struct drm_plane_state *vc4_plane_duplicate_state(struct drm_plane *plane)
147 struct vc4_plane_state *vc4_state;
149 if (WARN_ON(!plane->state))
152 vc4_state = kmemdup(plane->state, sizeof(*vc4_state), GFP_KERNEL);
156 memset(&vc4_state->lbm, 0, sizeof(vc4_state->lbm));
158 __drm_atomic_helper_plane_duplicate_state(plane, &vc4_state->base);
160 if (vc4_state->dlist) {
161 vc4_state->dlist = kmemdup(vc4_state->dlist,
162 vc4_state->dlist_count * 4,
164 if (!vc4_state->dlist) {
168 vc4_state->dlist_size = vc4_state->dlist_count;
171 return &vc4_state->base;
174 static void vc4_plane_destroy_state(struct drm_plane *plane,
175 struct drm_plane_state *state)
177 struct vc4_dev *vc4 = to_vc4_dev(plane->dev);
178 struct vc4_plane_state *vc4_state = to_vc4_plane_state(state);
180 if (vc4_state->lbm.allocated) {
181 unsigned long irqflags;
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);
188 kfree(vc4_state->dlist);
189 __drm_atomic_helper_plane_destroy_state(&vc4_state->base);
193 /* Called during init to allocate the plane's atomic state. */
194 static void vc4_plane_reset(struct drm_plane *plane)
196 struct vc4_plane_state *vc4_state;
198 WARN_ON(plane->state);
200 vc4_state = kzalloc(sizeof(*vc4_state), GFP_KERNEL);
204 __drm_atomic_helper_plane_reset(plane, &vc4_state->base);
207 static void vc4_dlist_write(struct vc4_plane_state *vc4_state, u32 val)
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);
215 memcpy(new_dlist, vc4_state->dlist, vc4_state->dlist_count * 4);
217 kfree(vc4_state->dlist);
218 vc4_state->dlist = new_dlist;
219 vc4_state->dlist_size = new_size;
222 vc4_state->dlist[vc4_state->dlist_count++] = val;
225 /* Returns the scl0/scl1 field based on whether the dimensions need to
226 * be up/down/non-scaled.
228 * This is a replication of a table from the spec.
230 static u32 vc4_get_scl_field(struct drm_plane_state *state, int plane)
232 struct vc4_plane_state *vc4_state = to_vc4_plane_state(state);
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;
252 case VC4_SCALING_NONE << 2 | VC4_SCALING_NONE:
253 /* The unity case is independently handled by
260 static int vc4_plane_setup_clipping_and_scaling(struct drm_plane_state *state)
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;
274 crtc_state = drm_atomic_get_existing_crtc_state(state->state,
277 DRM_DEBUG_KMS("Invalid crtc state\n");
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.
285 if (plane->type == DRM_PLANE_TYPE_CURSOR) {
286 min_scale = DRM_PLANE_HELPER_NO_SCALING;
287 max_scale = DRM_PLANE_HELPER_NO_SCALING;
293 ret = drm_atomic_helper_check_plane_state(state, crtc_state,
294 min_scale, max_scale,
299 h_subsample = drm_format_horz_chroma_subsampling(format);
300 v_subsample = drm_format_vert_chroma_subsampling(format);
302 for (i = 0; i < num_planes; i++)
303 vc4_state->offsets[i] = bo->paddr + fb->offsets[i];
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)) {
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;
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;
323 vc4_state->x_scaling[0] = vc4_get_scaling_mode(vc4_state->src_w[0],
325 vc4_state->y_scaling[0] = vc4_get_scaling_mode(vc4_state->src_h[0],
328 vc4_state->is_unity = (vc4_state->x_scaling[0] == VC4_SCALING_NONE &&
329 vc4_state->y_scaling[0] == VC4_SCALING_NONE);
331 if (num_planes > 1) {
332 vc4_state->is_yuv = true;
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;
337 vc4_state->x_scaling[1] =
338 vc4_get_scaling_mode(vc4_state->src_w[1],
340 vc4_state->y_scaling[1] =
341 vc4_get_scaling_mode(vc4_state->src_h[1],
344 /* YUV conversion requires that horizontal scaling be enabled
345 * on the UV plane even if vc4_get_scaling_mode() returned
346 * VC4_SCALING_NONE (which can happen when the down-scaling
347 * ratio is 0.5). Let's force it to VC4_SCALING_PPF in this
350 if (vc4_state->x_scaling[1] == VC4_SCALING_NONE)
351 vc4_state->x_scaling[1] = VC4_SCALING_PPF;
353 vc4_state->is_yuv = false;
354 vc4_state->x_scaling[1] = VC4_SCALING_NONE;
355 vc4_state->y_scaling[1] = VC4_SCALING_NONE;
361 static void vc4_write_tpz(struct vc4_plane_state *vc4_state, u32 src, u32 dst)
365 scale = (1 << 16) * src / dst;
367 /* The specs note that while the reciprocal would be defined
368 * as (1<<32)/scale, ~0 is close enough.
372 vc4_dlist_write(vc4_state,
373 VC4_SET_FIELD(scale, SCALER_TPZ0_SCALE) |
374 VC4_SET_FIELD(0, SCALER_TPZ0_IPHASE));
375 vc4_dlist_write(vc4_state,
376 VC4_SET_FIELD(recip, SCALER_TPZ1_RECIP));
379 static void vc4_write_ppf(struct vc4_plane_state *vc4_state, u32 src, u32 dst)
381 u32 scale = (1 << 16) * src / dst;
383 vc4_dlist_write(vc4_state,
385 VC4_SET_FIELD(scale, SCALER_PPF_SCALE) |
386 VC4_SET_FIELD(0, SCALER_PPF_IPHASE));
389 static u32 vc4_lbm_size(struct drm_plane_state *state)
391 struct vc4_plane_state *vc4_state = to_vc4_plane_state(state);
392 /* This is the worst case number. One of the two sizes will
393 * be used depending on the scaling configuration.
395 u32 pix_per_line = max(vc4_state->src_w[0], (u32)vc4_state->crtc_w);
398 if (!vc4_state->is_yuv) {
399 if (vc4_state->is_unity)
401 else if (vc4_state->y_scaling[0] == VC4_SCALING_TPZ)
402 lbm = pix_per_line * 8;
404 /* In special cases, this multiplier might be 12. */
405 lbm = pix_per_line * 16;
408 /* There are cases for this going down to a multiplier
409 * of 2, but according to the firmware source, the
410 * table in the docs is somewhat wrong.
412 lbm = pix_per_line * 16;
415 lbm = roundup(lbm, 32);
420 static void vc4_write_scaling_parameters(struct drm_plane_state *state,
423 struct vc4_plane_state *vc4_state = to_vc4_plane_state(state);
425 /* Ch0 H-PPF Word 0: Scaling Parameters */
426 if (vc4_state->x_scaling[channel] == VC4_SCALING_PPF) {
427 vc4_write_ppf(vc4_state,
428 vc4_state->src_w[channel], vc4_state->crtc_w);
431 /* Ch0 V-PPF Words 0-1: Scaling Parameters, Context */
432 if (vc4_state->y_scaling[channel] == VC4_SCALING_PPF) {
433 vc4_write_ppf(vc4_state,
434 vc4_state->src_h[channel], vc4_state->crtc_h);
435 vc4_dlist_write(vc4_state, 0xc0c0c0c0);
438 /* Ch0 H-TPZ Words 0-1: Scaling Parameters, Recip */
439 if (vc4_state->x_scaling[channel] == VC4_SCALING_TPZ) {
440 vc4_write_tpz(vc4_state,
441 vc4_state->src_w[channel], vc4_state->crtc_w);
444 /* Ch0 V-TPZ Words 0-2: Scaling Parameters, Recip, Context */
445 if (vc4_state->y_scaling[channel] == VC4_SCALING_TPZ) {
446 vc4_write_tpz(vc4_state,
447 vc4_state->src_h[channel], vc4_state->crtc_h);
448 vc4_dlist_write(vc4_state, 0xc0c0c0c0);
452 /* Writes out a full display list for an active plane to the plane's
453 * private dlist state.
455 static int vc4_plane_mode_set(struct drm_plane *plane,
456 struct drm_plane_state *state)
458 struct vc4_dev *vc4 = to_vc4_dev(plane->dev);
459 struct vc4_plane_state *vc4_state = to_vc4_plane_state(state);
460 struct drm_framebuffer *fb = state->fb;
461 u32 ctl0_offset = vc4_state->dlist_count;
462 const struct hvs_format *format = vc4_get_hvs_format(fb->format->format);
463 u64 base_format_mod = fourcc_mod_broadcom_mod(fb->modifier);
464 int num_planes = drm_format_num_planes(format->drm);
465 u32 h_subsample, v_subsample;
466 bool mix_plane_alpha;
468 u32 scl0, scl1, pitch0;
469 u32 lbm_size, tiling;
470 unsigned long irqflags;
471 u32 hvs_format = format->hvs;
474 ret = vc4_plane_setup_clipping_and_scaling(state);
478 /* Allocate the LBM memory that the HVS will use for temporary
479 * storage due to our scaling/format conversion.
481 lbm_size = vc4_lbm_size(state);
483 if (!vc4_state->lbm.allocated) {
484 spin_lock_irqsave(&vc4->hvs->mm_lock, irqflags);
485 ret = drm_mm_insert_node_generic(&vc4->hvs->lbm_mm,
488 spin_unlock_irqrestore(&vc4->hvs->mm_lock, irqflags);
490 WARN_ON_ONCE(lbm_size != vc4_state->lbm.size);
497 /* SCL1 is used for Cb/Cr scaling of planar formats. For RGB
498 * and 4:4:4, scl1 should be set to scl0 so both channels of
499 * the scaler do the same thing. For YUV, the Y plane needs
500 * to be put in channel 1 and Cb/Cr in channel 0, so we swap
501 * the scl fields here.
503 if (num_planes == 1) {
504 scl0 = vc4_get_scl_field(state, 0);
507 scl0 = vc4_get_scl_field(state, 1);
508 scl1 = vc4_get_scl_field(state, 0);
511 h_subsample = drm_format_horz_chroma_subsampling(format->drm);
512 v_subsample = drm_format_vert_chroma_subsampling(format->drm);
514 switch (base_format_mod) {
515 case DRM_FORMAT_MOD_LINEAR:
516 tiling = SCALER_CTL0_TILING_LINEAR;
517 pitch0 = VC4_SET_FIELD(fb->pitches[0], SCALER_SRC_PITCH);
519 /* Adjust the base pointer to the first pixel to be scanned
522 for (i = 0; i < num_planes; i++) {
523 vc4_state->offsets[i] += vc4_state->src_y /
524 (i ? v_subsample : 1) *
526 vc4_state->offsets[i] += vc4_state->src_x /
527 (i ? h_subsample : 1) *
533 case DRM_FORMAT_MOD_BROADCOM_VC4_T_TILED: {
534 u32 tile_size_shift = 12; /* T tiles are 4kb */
535 /* Whole-tile offsets, mostly for setting the pitch. */
536 u32 tile_w_shift = fb->format->cpp[0] == 2 ? 6 : 5;
537 u32 tile_h_shift = 5; /* 16 and 32bpp are 32 pixels high */
538 u32 tile_w_mask = (1 << tile_w_shift) - 1;
539 /* The height mask on 32-bit-per-pixel tiles is 63, i.e. twice
540 * the height (in pixels) of a 4k tile.
542 u32 tile_h_mask = (2 << tile_h_shift) - 1;
543 /* For T-tiled, the FB pitch is "how many bytes from one row to
544 * the next, such that
546 * pitch * tile_h == tile_size * tiles_per_row
548 u32 tiles_w = fb->pitches[0] >> (tile_size_shift - tile_h_shift);
549 u32 tiles_l = vc4_state->src_x >> tile_w_shift;
550 u32 tiles_r = tiles_w - tiles_l;
551 u32 tiles_t = vc4_state->src_y >> tile_h_shift;
552 /* Intra-tile offsets, which modify the base address (the
553 * SCALER_PITCH0_TILE_Y_OFFSET tells HVS how to walk from that
556 u32 tile_y = (vc4_state->src_y >> 4) & 1;
557 u32 subtile_y = (vc4_state->src_y >> 2) & 3;
558 u32 utile_y = vc4_state->src_y & 3;
559 u32 x_off = vc4_state->src_x & tile_w_mask;
560 u32 y_off = vc4_state->src_y & tile_h_mask;
562 tiling = SCALER_CTL0_TILING_256B_OR_T;
563 pitch0 = (VC4_SET_FIELD(x_off, SCALER_PITCH0_SINK_PIX) |
564 VC4_SET_FIELD(y_off, SCALER_PITCH0_TILE_Y_OFFSET) |
565 VC4_SET_FIELD(tiles_l, SCALER_PITCH0_TILE_WIDTH_L) |
566 VC4_SET_FIELD(tiles_r, SCALER_PITCH0_TILE_WIDTH_R));
567 vc4_state->offsets[0] += tiles_t * (tiles_w << tile_size_shift);
568 vc4_state->offsets[0] += subtile_y << 8;
569 vc4_state->offsets[0] += utile_y << 4;
571 /* Rows of tiles alternate left-to-right and right-to-left. */
573 pitch0 |= SCALER_PITCH0_TILE_INITIAL_LINE_DIR;
574 vc4_state->offsets[0] += (tiles_w - tiles_l) <<
576 vc4_state->offsets[0] -= (1 + !tile_y) << 10;
578 vc4_state->offsets[0] += tiles_l << tile_size_shift;
579 vc4_state->offsets[0] += tile_y << 10;
585 case DRM_FORMAT_MOD_BROADCOM_SAND64:
586 case DRM_FORMAT_MOD_BROADCOM_SAND128:
587 case DRM_FORMAT_MOD_BROADCOM_SAND256: {
588 uint32_t param = fourcc_mod_broadcom_param(fb->modifier);
590 /* Column-based NV12 or RGBA.
592 if (fb->format->num_planes > 1) {
593 if (hvs_format != HVS_PIXEL_FORMAT_YCBCR_YUV420_2PLANE) {
594 DRM_DEBUG_KMS("SAND format only valid for NV12/21");
597 hvs_format = HVS_PIXEL_FORMAT_H264;
599 if (base_format_mod == DRM_FORMAT_MOD_BROADCOM_SAND256) {
600 DRM_DEBUG_KMS("SAND256 format only valid for H.264");
605 switch (base_format_mod) {
606 case DRM_FORMAT_MOD_BROADCOM_SAND64:
607 tiling = SCALER_CTL0_TILING_64B;
609 case DRM_FORMAT_MOD_BROADCOM_SAND128:
610 tiling = SCALER_CTL0_TILING_128B;
612 case DRM_FORMAT_MOD_BROADCOM_SAND256:
613 tiling = SCALER_CTL0_TILING_256B_OR_T;
619 if (param > SCALER_TILE_HEIGHT_MASK) {
620 DRM_DEBUG_KMS("SAND height too large (%d)\n", param);
624 pitch0 = VC4_SET_FIELD(param, SCALER_TILE_HEIGHT);
629 DRM_DEBUG_KMS("Unsupported FB tiling flag 0x%16llx",
630 (long long)fb->modifier);
635 vc4_dlist_write(vc4_state,
637 VC4_SET_FIELD(SCALER_CTL0_RGBA_EXPAND_ROUND, SCALER_CTL0_RGBA_EXPAND) |
638 (format->pixel_order << SCALER_CTL0_ORDER_SHIFT) |
639 (hvs_format << SCALER_CTL0_PIXEL_FORMAT_SHIFT) |
640 VC4_SET_FIELD(tiling, SCALER_CTL0_TILING) |
641 (vc4_state->is_unity ? SCALER_CTL0_UNITY : 0) |
642 VC4_SET_FIELD(scl0, SCALER_CTL0_SCL0) |
643 VC4_SET_FIELD(scl1, SCALER_CTL0_SCL1));
645 /* Position Word 0: Image Positions and Alpha Value */
646 vc4_state->pos0_offset = vc4_state->dlist_count;
647 vc4_dlist_write(vc4_state,
648 VC4_SET_FIELD(state->alpha >> 8, SCALER_POS0_FIXED_ALPHA) |
649 VC4_SET_FIELD(vc4_state->crtc_x, SCALER_POS0_START_X) |
650 VC4_SET_FIELD(vc4_state->crtc_y, SCALER_POS0_START_Y));
652 /* Position Word 1: Scaled Image Dimensions. */
653 if (!vc4_state->is_unity) {
654 vc4_dlist_write(vc4_state,
655 VC4_SET_FIELD(vc4_state->crtc_w,
656 SCALER_POS1_SCL_WIDTH) |
657 VC4_SET_FIELD(vc4_state->crtc_h,
658 SCALER_POS1_SCL_HEIGHT));
661 /* Don't waste cycles mixing with plane alpha if the set alpha
662 * is opaque or there is no per-pixel alpha information.
663 * In any case we use the alpha property value as the fixed alpha.
665 mix_plane_alpha = state->alpha != DRM_BLEND_ALPHA_OPAQUE &&
666 fb->format->has_alpha;
668 /* Position Word 2: Source Image Size, Alpha */
669 vc4_state->pos2_offset = vc4_state->dlist_count;
670 vc4_dlist_write(vc4_state,
671 VC4_SET_FIELD(fb->format->has_alpha ?
672 SCALER_POS2_ALPHA_MODE_PIPELINE :
673 SCALER_POS2_ALPHA_MODE_FIXED,
674 SCALER_POS2_ALPHA_MODE) |
675 (mix_plane_alpha ? SCALER_POS2_ALPHA_MIX : 0) |
676 (fb->format->has_alpha ? SCALER_POS2_ALPHA_PREMULT : 0) |
677 VC4_SET_FIELD(vc4_state->src_w[0], SCALER_POS2_WIDTH) |
678 VC4_SET_FIELD(vc4_state->src_h[0], SCALER_POS2_HEIGHT));
680 /* Position Word 3: Context. Written by the HVS. */
681 vc4_dlist_write(vc4_state, 0xc0c0c0c0);
684 /* Pointer Word 0/1/2: RGB / Y / Cb / Cr Pointers
686 * The pointers may be any byte address.
688 vc4_state->ptr0_offset = vc4_state->dlist_count;
689 for (i = 0; i < num_planes; i++)
690 vc4_dlist_write(vc4_state, vc4_state->offsets[i]);
692 /* Pointer Context Word 0/1/2: Written by the HVS */
693 for (i = 0; i < num_planes; i++)
694 vc4_dlist_write(vc4_state, 0xc0c0c0c0);
697 vc4_dlist_write(vc4_state, pitch0);
700 for (i = 1; i < num_planes; i++) {
701 if (hvs_format != HVS_PIXEL_FORMAT_H264) {
702 vc4_dlist_write(vc4_state,
703 VC4_SET_FIELD(fb->pitches[i],
706 vc4_dlist_write(vc4_state, pitch0);
710 /* Colorspace conversion words */
711 if (vc4_state->is_yuv) {
712 vc4_dlist_write(vc4_state, SCALER_CSC0_ITR_R_601_5);
713 vc4_dlist_write(vc4_state, SCALER_CSC1_ITR_R_601_5);
714 vc4_dlist_write(vc4_state, SCALER_CSC2_ITR_R_601_5);
717 if (vc4_state->x_scaling[0] != VC4_SCALING_NONE ||
718 vc4_state->x_scaling[1] != VC4_SCALING_NONE ||
719 vc4_state->y_scaling[0] != VC4_SCALING_NONE ||
720 vc4_state->y_scaling[1] != VC4_SCALING_NONE) {
721 /* LBM Base Address. */
722 if (vc4_state->y_scaling[0] != VC4_SCALING_NONE ||
723 vc4_state->y_scaling[1] != VC4_SCALING_NONE) {
724 vc4_dlist_write(vc4_state, vc4_state->lbm.start);
727 if (num_planes > 1) {
728 /* Emit Cb/Cr as channel 0 and Y as channel
729 * 1. This matches how we set up scl0/scl1
732 vc4_write_scaling_parameters(state, 1);
734 vc4_write_scaling_parameters(state, 0);
736 /* If any PPF setup was done, then all the kernel
737 * pointers get uploaded.
739 if (vc4_state->x_scaling[0] == VC4_SCALING_PPF ||
740 vc4_state->y_scaling[0] == VC4_SCALING_PPF ||
741 vc4_state->x_scaling[1] == VC4_SCALING_PPF ||
742 vc4_state->y_scaling[1] == VC4_SCALING_PPF) {
743 u32 kernel = VC4_SET_FIELD(vc4->hvs->mitchell_netravali_filter.start,
744 SCALER_PPF_KERNEL_OFFSET);
747 vc4_dlist_write(vc4_state, kernel);
749 vc4_dlist_write(vc4_state, kernel);
751 vc4_dlist_write(vc4_state, kernel);
753 vc4_dlist_write(vc4_state, kernel);
757 vc4_state->dlist[ctl0_offset] |=
758 VC4_SET_FIELD(vc4_state->dlist_count, SCALER_CTL0_SIZE);
760 /* crtc_* are already clipped coordinates. */
761 covers_screen = vc4_state->crtc_x == 0 && vc4_state->crtc_y == 0 &&
762 vc4_state->crtc_w == state->crtc->mode.hdisplay &&
763 vc4_state->crtc_h == state->crtc->mode.vdisplay;
764 /* Background fill might be necessary when the plane has per-pixel
765 * alpha content or a non-opaque plane alpha and could blend from the
766 * background or does not cover the entire screen.
768 vc4_state->needs_bg_fill = fb->format->has_alpha || !covers_screen ||
769 state->alpha != DRM_BLEND_ALPHA_OPAQUE;
774 /* If a modeset involves changing the setup of a plane, the atomic
775 * infrastructure will call this to validate a proposed plane setup.
776 * However, if a plane isn't getting updated, this (and the
777 * corresponding vc4_plane_atomic_update) won't get called. Thus, we
778 * compute the dlist here and have all active plane dlists get updated
779 * in the CRTC's flush.
781 static int vc4_plane_atomic_check(struct drm_plane *plane,
782 struct drm_plane_state *state)
784 struct vc4_plane_state *vc4_state = to_vc4_plane_state(state);
786 vc4_state->dlist_count = 0;
788 if (plane_enabled(state))
789 return vc4_plane_mode_set(plane, state);
794 static void vc4_plane_atomic_update(struct drm_plane *plane,
795 struct drm_plane_state *old_state)
797 /* No contents here. Since we don't know where in the CRTC's
798 * dlist we should be stored, our dlist is uploaded to the
799 * hardware with vc4_plane_write_dlist() at CRTC atomic_flush
804 u32 vc4_plane_write_dlist(struct drm_plane *plane, u32 __iomem *dlist)
806 struct vc4_plane_state *vc4_state = to_vc4_plane_state(plane->state);
809 vc4_state->hw_dlist = dlist;
811 /* Can't memcpy_toio() because it needs to be 32-bit writes. */
812 for (i = 0; i < vc4_state->dlist_count; i++)
813 writel(vc4_state->dlist[i], &dlist[i]);
815 return vc4_state->dlist_count;
818 u32 vc4_plane_dlist_size(const struct drm_plane_state *state)
820 const struct vc4_plane_state *vc4_state =
821 container_of(state, typeof(*vc4_state), base);
823 return vc4_state->dlist_count;
826 /* Updates the plane to immediately (well, once the FIFO needs
827 * refilling) scan out from at a new framebuffer.
829 void vc4_plane_async_set_fb(struct drm_plane *plane, struct drm_framebuffer *fb)
831 struct vc4_plane_state *vc4_state = to_vc4_plane_state(plane->state);
832 struct drm_gem_cma_object *bo = drm_fb_cma_get_gem_obj(fb, 0);
835 /* We're skipping the address adjustment for negative origin,
836 * because this is only called on the primary plane.
838 WARN_ON_ONCE(plane->state->crtc_x < 0 || plane->state->crtc_y < 0);
839 addr = bo->paddr + fb->offsets[0];
841 /* Write the new address into the hardware immediately. The
842 * scanout will start from this address as soon as the FIFO
843 * needs to refill with pixels.
845 writel(addr, &vc4_state->hw_dlist[vc4_state->ptr0_offset]);
847 /* Also update the CPU-side dlist copy, so that any later
848 * atomic updates that don't do a new modeset on our plane
849 * also use our updated address.
851 vc4_state->dlist[vc4_state->ptr0_offset] = addr;
854 static void vc4_plane_atomic_async_update(struct drm_plane *plane,
855 struct drm_plane_state *state)
857 struct vc4_plane_state *vc4_state, *new_vc4_state;
859 if (plane->state->fb != state->fb) {
860 vc4_plane_async_set_fb(plane, state->fb);
861 drm_atomic_set_fb_for_plane(plane->state, state->fb);
864 /* Set the cursor's position on the screen. This is the
865 * expected change from the drm_mode_cursor_universal()
868 plane->state->crtc_x = state->crtc_x;
869 plane->state->crtc_y = state->crtc_y;
871 /* Allow changing the start position within the cursor BO, if
874 plane->state->src_x = state->src_x;
875 plane->state->src_y = state->src_y;
877 /* Update the display list based on the new crtc_x/y. */
878 vc4_plane_atomic_check(plane, state);
880 new_vc4_state = to_vc4_plane_state(state);
881 vc4_state = to_vc4_plane_state(plane->state);
883 /* Update the current vc4_state pos0, pos2 and ptr0 dlist entries. */
884 vc4_state->dlist[vc4_state->pos0_offset] =
885 new_vc4_state->dlist[vc4_state->pos0_offset];
886 vc4_state->dlist[vc4_state->pos2_offset] =
887 new_vc4_state->dlist[vc4_state->pos2_offset];
888 vc4_state->dlist[vc4_state->ptr0_offset] =
889 new_vc4_state->dlist[vc4_state->ptr0_offset];
891 /* Note that we can't just call vc4_plane_write_dlist()
892 * because that would smash the context data that the HVS is
895 writel(vc4_state->dlist[vc4_state->pos0_offset],
896 &vc4_state->hw_dlist[vc4_state->pos0_offset]);
897 writel(vc4_state->dlist[vc4_state->pos2_offset],
898 &vc4_state->hw_dlist[vc4_state->pos2_offset]);
899 writel(vc4_state->dlist[vc4_state->ptr0_offset],
900 &vc4_state->hw_dlist[vc4_state->ptr0_offset]);
903 static int vc4_plane_atomic_async_check(struct drm_plane *plane,
904 struct drm_plane_state *state)
906 /* No configuring new scaling in the fast path. */
907 if (plane->state->crtc_w != state->crtc_w ||
908 plane->state->crtc_h != state->crtc_h ||
909 plane->state->src_w != state->src_w ||
910 plane->state->src_h != state->src_h)
916 static int vc4_prepare_fb(struct drm_plane *plane,
917 struct drm_plane_state *state)
920 struct dma_fence *fence;
926 bo = to_vc4_bo(&drm_fb_cma_get_gem_obj(state->fb, 0)->base);
928 fence = reservation_object_get_excl_rcu(bo->resv);
929 drm_atomic_set_fence_for_plane(state, fence);
931 if (plane->state->fb == state->fb)
934 ret = vc4_bo_inc_usecnt(bo);
941 static void vc4_cleanup_fb(struct drm_plane *plane,
942 struct drm_plane_state *state)
946 if (plane->state->fb == state->fb || !state->fb)
949 bo = to_vc4_bo(&drm_fb_cma_get_gem_obj(state->fb, 0)->base);
950 vc4_bo_dec_usecnt(bo);
953 static const struct drm_plane_helper_funcs vc4_plane_helper_funcs = {
954 .atomic_check = vc4_plane_atomic_check,
955 .atomic_update = vc4_plane_atomic_update,
956 .prepare_fb = vc4_prepare_fb,
957 .cleanup_fb = vc4_cleanup_fb,
958 .atomic_async_check = vc4_plane_atomic_async_check,
959 .atomic_async_update = vc4_plane_atomic_async_update,
962 static void vc4_plane_destroy(struct drm_plane *plane)
964 drm_plane_cleanup(plane);
967 static bool vc4_format_mod_supported(struct drm_plane *plane,
971 /* Support T_TILING for RGB formats only. */
973 case DRM_FORMAT_XRGB8888:
974 case DRM_FORMAT_ARGB8888:
975 case DRM_FORMAT_ABGR8888:
976 case DRM_FORMAT_XBGR8888:
977 case DRM_FORMAT_RGB565:
978 case DRM_FORMAT_BGR565:
979 case DRM_FORMAT_ARGB1555:
980 case DRM_FORMAT_XRGB1555:
981 switch (fourcc_mod_broadcom_mod(modifier)) {
982 case DRM_FORMAT_MOD_LINEAR:
983 case DRM_FORMAT_MOD_BROADCOM_VC4_T_TILED:
984 case DRM_FORMAT_MOD_BROADCOM_SAND64:
985 case DRM_FORMAT_MOD_BROADCOM_SAND128:
990 case DRM_FORMAT_NV12:
991 case DRM_FORMAT_NV21:
992 switch (fourcc_mod_broadcom_mod(modifier)) {
993 case DRM_FORMAT_MOD_LINEAR:
994 case DRM_FORMAT_MOD_BROADCOM_SAND64:
995 case DRM_FORMAT_MOD_BROADCOM_SAND128:
996 case DRM_FORMAT_MOD_BROADCOM_SAND256:
1001 case DRM_FORMAT_YUV422:
1002 case DRM_FORMAT_YVU422:
1003 case DRM_FORMAT_YUV420:
1004 case DRM_FORMAT_YVU420:
1005 case DRM_FORMAT_NV16:
1006 case DRM_FORMAT_NV61:
1008 return (modifier == DRM_FORMAT_MOD_LINEAR);
1012 static const struct drm_plane_funcs vc4_plane_funcs = {
1013 .update_plane = drm_atomic_helper_update_plane,
1014 .disable_plane = drm_atomic_helper_disable_plane,
1015 .destroy = vc4_plane_destroy,
1016 .set_property = NULL,
1017 .reset = vc4_plane_reset,
1018 .atomic_duplicate_state = vc4_plane_duplicate_state,
1019 .atomic_destroy_state = vc4_plane_destroy_state,
1020 .format_mod_supported = vc4_format_mod_supported,
1023 struct drm_plane *vc4_plane_init(struct drm_device *dev,
1024 enum drm_plane_type type)
1026 struct drm_plane *plane = NULL;
1027 struct vc4_plane *vc4_plane;
1028 u32 formats[ARRAY_SIZE(hvs_formats)];
1029 u32 num_formats = 0;
1032 static const uint64_t modifiers[] = {
1033 DRM_FORMAT_MOD_BROADCOM_VC4_T_TILED,
1034 DRM_FORMAT_MOD_BROADCOM_SAND128,
1035 DRM_FORMAT_MOD_BROADCOM_SAND64,
1036 DRM_FORMAT_MOD_BROADCOM_SAND256,
1037 DRM_FORMAT_MOD_LINEAR,
1038 DRM_FORMAT_MOD_INVALID
1041 vc4_plane = devm_kzalloc(dev->dev, sizeof(*vc4_plane),
1044 return ERR_PTR(-ENOMEM);
1046 for (i = 0; i < ARRAY_SIZE(hvs_formats); i++) {
1047 /* Don't allow YUV in cursor planes, since that means
1048 * tuning on the scaler, which we don't allow for the
1051 if (type != DRM_PLANE_TYPE_CURSOR ||
1052 hvs_formats[i].hvs < HVS_PIXEL_FORMAT_YCBCR_YUV420_3PLANE) {
1053 formats[num_formats++] = hvs_formats[i].drm;
1056 plane = &vc4_plane->base;
1057 ret = drm_universal_plane_init(dev, plane, 0,
1059 formats, num_formats,
1060 modifiers, type, NULL);
1062 drm_plane_helper_add(plane, &vc4_plane_helper_funcs);
1064 drm_plane_create_alpha_property(plane);