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_PPF;
135 return VC4_SCALING_TPZ;
137 return VC4_SCALING_NONE;
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;
273 for (i = 0; i < num_planes; i++)
274 vc4_state->offsets[i] = bo->paddr + fb->offsets[i];
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)) {
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;
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;
294 vc4_state->x_scaling[0] = vc4_get_scaling_mode(vc4_state->src_w[0],
296 vc4_state->y_scaling[0] = vc4_get_scaling_mode(vc4_state->src_h[0],
299 vc4_state->is_unity = (vc4_state->x_scaling[0] == VC4_SCALING_NONE &&
300 vc4_state->y_scaling[0] == VC4_SCALING_NONE);
302 if (num_planes > 1) {
303 vc4_state->is_yuv = true;
305 h_subsample = drm_format_horz_chroma_subsampling(format);
306 v_subsample = drm_format_vert_chroma_subsampling(format);
307 vc4_state->src_w[1] = vc4_state->src_w[0] / h_subsample;
308 vc4_state->src_h[1] = vc4_state->src_h[0] / v_subsample;
310 vc4_state->x_scaling[1] =
311 vc4_get_scaling_mode(vc4_state->src_w[1],
313 vc4_state->y_scaling[1] =
314 vc4_get_scaling_mode(vc4_state->src_h[1],
317 /* YUV conversion requires that horizontal scaling be enabled,
318 * even on a plane that's otherwise 1:1. Looks like only PPF
319 * works in that case, so let's pick that one.
321 if (vc4_state->is_unity)
322 vc4_state->x_scaling[0] = VC4_SCALING_PPF;
324 vc4_state->x_scaling[1] = VC4_SCALING_NONE;
325 vc4_state->y_scaling[1] = VC4_SCALING_NONE;
328 /* No configuring scaling on the cursor plane, since it gets
329 non-vblank-synced updates, and scaling requires requires
330 LBM changes which have to be vblank-synced.
332 if (plane->type == DRM_PLANE_TYPE_CURSOR && !vc4_state->is_unity)
335 /* Clamp the on-screen start x/y to 0. The hardware doesn't
336 * support negative y, and negative x wastes bandwidth.
338 if (vc4_state->crtc_x < 0) {
339 for (i = 0; i < num_planes; i++) {
340 u32 cpp = fb->format->cpp[i];
341 u32 subs = ((i == 0) ? 1 : h_subsample);
343 vc4_state->offsets[i] += (cpp *
344 (-vc4_state->crtc_x) / subs);
346 vc4_state->src_w[0] += vc4_state->crtc_x;
347 vc4_state->src_w[1] += vc4_state->crtc_x / h_subsample;
348 vc4_state->crtc_x = 0;
351 if (vc4_state->crtc_y < 0) {
352 for (i = 0; i < num_planes; i++) {
353 u32 subs = ((i == 0) ? 1 : v_subsample);
355 vc4_state->offsets[i] += (fb->pitches[i] *
356 (-vc4_state->crtc_y) / subs);
358 vc4_state->src_h[0] += vc4_state->crtc_y;
359 vc4_state->src_h[1] += vc4_state->crtc_y / v_subsample;
360 vc4_state->crtc_y = 0;
366 static void vc4_write_tpz(struct vc4_plane_state *vc4_state, u32 src, u32 dst)
370 scale = (1 << 16) * src / dst;
372 /* The specs note that while the reciprocal would be defined
373 * as (1<<32)/scale, ~0 is close enough.
377 vc4_dlist_write(vc4_state,
378 VC4_SET_FIELD(scale, SCALER_TPZ0_SCALE) |
379 VC4_SET_FIELD(0, SCALER_TPZ0_IPHASE));
380 vc4_dlist_write(vc4_state,
381 VC4_SET_FIELD(recip, SCALER_TPZ1_RECIP));
384 static void vc4_write_ppf(struct vc4_plane_state *vc4_state, u32 src, u32 dst)
386 u32 scale = (1 << 16) * src / dst;
388 vc4_dlist_write(vc4_state,
390 VC4_SET_FIELD(scale, SCALER_PPF_SCALE) |
391 VC4_SET_FIELD(0, SCALER_PPF_IPHASE));
394 static u32 vc4_lbm_size(struct drm_plane_state *state)
396 struct vc4_plane_state *vc4_state = to_vc4_plane_state(state);
397 /* This is the worst case number. One of the two sizes will
398 * be used depending on the scaling configuration.
400 u32 pix_per_line = max(vc4_state->src_w[0], (u32)vc4_state->crtc_w);
403 if (!vc4_state->is_yuv) {
404 if (vc4_state->is_unity)
406 else if (vc4_state->y_scaling[0] == VC4_SCALING_TPZ)
407 lbm = pix_per_line * 8;
409 /* In special cases, this multiplier might be 12. */
410 lbm = pix_per_line * 16;
413 /* There are cases for this going down to a multiplier
414 * of 2, but according to the firmware source, the
415 * table in the docs is somewhat wrong.
417 lbm = pix_per_line * 16;
420 lbm = roundup(lbm, 32);
425 static void vc4_write_scaling_parameters(struct drm_plane_state *state,
428 struct vc4_plane_state *vc4_state = to_vc4_plane_state(state);
430 /* Ch0 H-PPF Word 0: Scaling Parameters */
431 if (vc4_state->x_scaling[channel] == VC4_SCALING_PPF) {
432 vc4_write_ppf(vc4_state,
433 vc4_state->src_w[channel], vc4_state->crtc_w);
436 /* Ch0 V-PPF Words 0-1: Scaling Parameters, Context */
437 if (vc4_state->y_scaling[channel] == VC4_SCALING_PPF) {
438 vc4_write_ppf(vc4_state,
439 vc4_state->src_h[channel], vc4_state->crtc_h);
440 vc4_dlist_write(vc4_state, 0xc0c0c0c0);
443 /* Ch0 H-TPZ Words 0-1: Scaling Parameters, Recip */
444 if (vc4_state->x_scaling[channel] == VC4_SCALING_TPZ) {
445 vc4_write_tpz(vc4_state,
446 vc4_state->src_w[channel], vc4_state->crtc_w);
449 /* Ch0 V-TPZ Words 0-2: Scaling Parameters, Recip, Context */
450 if (vc4_state->y_scaling[channel] == VC4_SCALING_TPZ) {
451 vc4_write_tpz(vc4_state,
452 vc4_state->src_h[channel], vc4_state->crtc_h);
453 vc4_dlist_write(vc4_state, 0xc0c0c0c0);
457 /* Writes out a full display list for an active plane to the plane's
458 * private dlist state.
460 static int vc4_plane_mode_set(struct drm_plane *plane,
461 struct drm_plane_state *state)
463 struct vc4_dev *vc4 = to_vc4_dev(plane->dev);
464 struct vc4_plane_state *vc4_state = to_vc4_plane_state(state);
465 struct drm_framebuffer *fb = state->fb;
466 u32 ctl0_offset = vc4_state->dlist_count;
467 const struct hvs_format *format = vc4_get_hvs_format(fb->format->format);
468 u64 base_format_mod = fourcc_mod_broadcom_mod(fb->modifier);
469 int num_planes = drm_format_num_planes(format->drm);
470 bool mix_plane_alpha;
472 u32 scl0, scl1, pitch0;
473 u32 lbm_size, tiling;
474 unsigned long irqflags;
475 u32 hvs_format = format->hvs;
478 ret = vc4_plane_setup_clipping_and_scaling(state);
482 /* Allocate the LBM memory that the HVS will use for temporary
483 * storage due to our scaling/format conversion.
485 lbm_size = vc4_lbm_size(state);
487 if (!vc4_state->lbm.allocated) {
488 spin_lock_irqsave(&vc4->hvs->mm_lock, irqflags);
489 ret = drm_mm_insert_node_generic(&vc4->hvs->lbm_mm,
492 spin_unlock_irqrestore(&vc4->hvs->mm_lock, irqflags);
494 WARN_ON_ONCE(lbm_size != vc4_state->lbm.size);
501 /* SCL1 is used for Cb/Cr scaling of planar formats. For RGB
502 * and 4:4:4, scl1 should be set to scl0 so both channels of
503 * the scaler do the same thing. For YUV, the Y plane needs
504 * to be put in channel 1 and Cb/Cr in channel 0, so we swap
505 * the scl fields here.
507 if (num_planes == 1) {
508 scl0 = vc4_get_scl_field(state, 0);
511 scl0 = vc4_get_scl_field(state, 1);
512 scl1 = vc4_get_scl_field(state, 0);
515 switch (base_format_mod) {
516 case DRM_FORMAT_MOD_LINEAR:
517 tiling = SCALER_CTL0_TILING_LINEAR;
518 pitch0 = VC4_SET_FIELD(fb->pitches[0], SCALER_SRC_PITCH);
521 case DRM_FORMAT_MOD_BROADCOM_VC4_T_TILED: {
522 /* For T-tiled, the FB pitch is "how many bytes from
523 * one row to the next, such that pitch * tile_h ==
524 * tile_size * tiles_per_row."
526 u32 tile_size_shift = 12; /* T tiles are 4kb */
527 u32 tile_h_shift = 5; /* 16 and 32bpp are 32 pixels high */
528 u32 tiles_w = fb->pitches[0] >> (tile_size_shift - tile_h_shift);
530 tiling = SCALER_CTL0_TILING_256B_OR_T;
532 pitch0 = (VC4_SET_FIELD(0, SCALER_PITCH0_TILE_Y_OFFSET) |
533 VC4_SET_FIELD(0, SCALER_PITCH0_TILE_WIDTH_L) |
534 VC4_SET_FIELD(tiles_w, SCALER_PITCH0_TILE_WIDTH_R));
538 case DRM_FORMAT_MOD_BROADCOM_SAND64:
539 case DRM_FORMAT_MOD_BROADCOM_SAND128:
540 case DRM_FORMAT_MOD_BROADCOM_SAND256: {
541 uint32_t param = fourcc_mod_broadcom_param(fb->modifier);
543 /* Column-based NV12 or RGBA.
545 if (fb->format->num_planes > 1) {
546 if (hvs_format != HVS_PIXEL_FORMAT_YCBCR_YUV420_2PLANE) {
547 DRM_DEBUG_KMS("SAND format only valid for NV12/21");
550 hvs_format = HVS_PIXEL_FORMAT_H264;
552 if (base_format_mod == DRM_FORMAT_MOD_BROADCOM_SAND256) {
553 DRM_DEBUG_KMS("SAND256 format only valid for H.264");
558 switch (base_format_mod) {
559 case DRM_FORMAT_MOD_BROADCOM_SAND64:
560 tiling = SCALER_CTL0_TILING_64B;
562 case DRM_FORMAT_MOD_BROADCOM_SAND128:
563 tiling = SCALER_CTL0_TILING_128B;
565 case DRM_FORMAT_MOD_BROADCOM_SAND256:
566 tiling = SCALER_CTL0_TILING_256B_OR_T;
572 if (param > SCALER_TILE_HEIGHT_MASK) {
573 DRM_DEBUG_KMS("SAND height too large (%d)\n", param);
577 pitch0 = VC4_SET_FIELD(param, SCALER_TILE_HEIGHT);
582 DRM_DEBUG_KMS("Unsupported FB tiling flag 0x%16llx",
583 (long long)fb->modifier);
588 vc4_dlist_write(vc4_state,
590 VC4_SET_FIELD(SCALER_CTL0_RGBA_EXPAND_ROUND, SCALER_CTL0_RGBA_EXPAND) |
591 (format->pixel_order << SCALER_CTL0_ORDER_SHIFT) |
592 (hvs_format << SCALER_CTL0_PIXEL_FORMAT_SHIFT) |
593 VC4_SET_FIELD(tiling, SCALER_CTL0_TILING) |
594 (vc4_state->is_unity ? SCALER_CTL0_UNITY : 0) |
595 VC4_SET_FIELD(scl0, SCALER_CTL0_SCL0) |
596 VC4_SET_FIELD(scl1, SCALER_CTL0_SCL1));
598 /* Position Word 0: Image Positions and Alpha Value */
599 vc4_state->pos0_offset = vc4_state->dlist_count;
600 vc4_dlist_write(vc4_state,
601 VC4_SET_FIELD(state->alpha >> 8, SCALER_POS0_FIXED_ALPHA) |
602 VC4_SET_FIELD(vc4_state->crtc_x, SCALER_POS0_START_X) |
603 VC4_SET_FIELD(vc4_state->crtc_y, SCALER_POS0_START_Y));
605 /* Position Word 1: Scaled Image Dimensions. */
606 if (!vc4_state->is_unity) {
607 vc4_dlist_write(vc4_state,
608 VC4_SET_FIELD(vc4_state->crtc_w,
609 SCALER_POS1_SCL_WIDTH) |
610 VC4_SET_FIELD(vc4_state->crtc_h,
611 SCALER_POS1_SCL_HEIGHT));
614 /* Don't waste cycles mixing with plane alpha if the set alpha
615 * is opaque or there is no per-pixel alpha information.
616 * In any case we use the alpha property value as the fixed alpha.
618 mix_plane_alpha = state->alpha != DRM_BLEND_ALPHA_OPAQUE &&
619 fb->format->has_alpha;
621 /* Position Word 2: Source Image Size, Alpha */
622 vc4_state->pos2_offset = vc4_state->dlist_count;
623 vc4_dlist_write(vc4_state,
624 VC4_SET_FIELD(fb->format->has_alpha ?
625 SCALER_POS2_ALPHA_MODE_PIPELINE :
626 SCALER_POS2_ALPHA_MODE_FIXED,
627 SCALER_POS2_ALPHA_MODE) |
628 (mix_plane_alpha ? SCALER_POS2_ALPHA_MIX : 0) |
629 (fb->format->has_alpha ? SCALER_POS2_ALPHA_PREMULT : 0) |
630 VC4_SET_FIELD(vc4_state->src_w[0], SCALER_POS2_WIDTH) |
631 VC4_SET_FIELD(vc4_state->src_h[0], SCALER_POS2_HEIGHT));
633 /* Position Word 3: Context. Written by the HVS. */
634 vc4_dlist_write(vc4_state, 0xc0c0c0c0);
637 /* Pointer Word 0/1/2: RGB / Y / Cb / Cr Pointers
639 * The pointers may be any byte address.
641 vc4_state->ptr0_offset = vc4_state->dlist_count;
642 for (i = 0; i < num_planes; i++)
643 vc4_dlist_write(vc4_state, vc4_state->offsets[i]);
645 /* Pointer Context Word 0/1/2: Written by the HVS */
646 for (i = 0; i < num_planes; i++)
647 vc4_dlist_write(vc4_state, 0xc0c0c0c0);
650 vc4_dlist_write(vc4_state, pitch0);
653 for (i = 1; i < num_planes; i++) {
654 if (hvs_format != HVS_PIXEL_FORMAT_H264) {
655 vc4_dlist_write(vc4_state,
656 VC4_SET_FIELD(fb->pitches[i],
659 vc4_dlist_write(vc4_state, pitch0);
663 /* Colorspace conversion words */
664 if (vc4_state->is_yuv) {
665 vc4_dlist_write(vc4_state, SCALER_CSC0_ITR_R_601_5);
666 vc4_dlist_write(vc4_state, SCALER_CSC1_ITR_R_601_5);
667 vc4_dlist_write(vc4_state, SCALER_CSC2_ITR_R_601_5);
670 if (vc4_state->x_scaling[0] != VC4_SCALING_NONE ||
671 vc4_state->x_scaling[1] != VC4_SCALING_NONE ||
672 vc4_state->y_scaling[0] != VC4_SCALING_NONE ||
673 vc4_state->y_scaling[1] != VC4_SCALING_NONE) {
674 /* LBM Base Address. */
675 if (vc4_state->y_scaling[0] != VC4_SCALING_NONE ||
676 vc4_state->y_scaling[1] != VC4_SCALING_NONE) {
677 vc4_dlist_write(vc4_state, vc4_state->lbm.start);
680 if (num_planes > 1) {
681 /* Emit Cb/Cr as channel 0 and Y as channel
682 * 1. This matches how we set up scl0/scl1
685 vc4_write_scaling_parameters(state, 1);
687 vc4_write_scaling_parameters(state, 0);
689 /* If any PPF setup was done, then all the kernel
690 * pointers get uploaded.
692 if (vc4_state->x_scaling[0] == VC4_SCALING_PPF ||
693 vc4_state->y_scaling[0] == VC4_SCALING_PPF ||
694 vc4_state->x_scaling[1] == VC4_SCALING_PPF ||
695 vc4_state->y_scaling[1] == VC4_SCALING_PPF) {
696 u32 kernel = VC4_SET_FIELD(vc4->hvs->mitchell_netravali_filter.start,
697 SCALER_PPF_KERNEL_OFFSET);
700 vc4_dlist_write(vc4_state, kernel);
702 vc4_dlist_write(vc4_state, kernel);
704 vc4_dlist_write(vc4_state, kernel);
706 vc4_dlist_write(vc4_state, kernel);
710 vc4_state->dlist[ctl0_offset] |=
711 VC4_SET_FIELD(vc4_state->dlist_count, SCALER_CTL0_SIZE);
713 /* crtc_* are already clipped coordinates. */
714 covers_screen = vc4_state->crtc_x == 0 && vc4_state->crtc_y == 0 &&
715 vc4_state->crtc_w == state->crtc->mode.hdisplay &&
716 vc4_state->crtc_h == state->crtc->mode.vdisplay;
717 /* Background fill might be necessary when the plane has per-pixel
718 * alpha content or a non-opaque plane alpha and could blend from the
719 * background or does not cover the entire screen.
721 vc4_state->needs_bg_fill = fb->format->has_alpha || !covers_screen ||
722 state->alpha != DRM_BLEND_ALPHA_OPAQUE;
727 /* If a modeset involves changing the setup of a plane, the atomic
728 * infrastructure will call this to validate a proposed plane setup.
729 * However, if a plane isn't getting updated, this (and the
730 * corresponding vc4_plane_atomic_update) won't get called. Thus, we
731 * compute the dlist here and have all active plane dlists get updated
732 * in the CRTC's flush.
734 static int vc4_plane_atomic_check(struct drm_plane *plane,
735 struct drm_plane_state *state)
737 struct vc4_plane_state *vc4_state = to_vc4_plane_state(state);
739 vc4_state->dlist_count = 0;
741 if (plane_enabled(state))
742 return vc4_plane_mode_set(plane, state);
747 static void vc4_plane_atomic_update(struct drm_plane *plane,
748 struct drm_plane_state *old_state)
750 /* No contents here. Since we don't know where in the CRTC's
751 * dlist we should be stored, our dlist is uploaded to the
752 * hardware with vc4_plane_write_dlist() at CRTC atomic_flush
757 u32 vc4_plane_write_dlist(struct drm_plane *plane, u32 __iomem *dlist)
759 struct vc4_plane_state *vc4_state = to_vc4_plane_state(plane->state);
762 vc4_state->hw_dlist = dlist;
764 /* Can't memcpy_toio() because it needs to be 32-bit writes. */
765 for (i = 0; i < vc4_state->dlist_count; i++)
766 writel(vc4_state->dlist[i], &dlist[i]);
768 return vc4_state->dlist_count;
771 u32 vc4_plane_dlist_size(const struct drm_plane_state *state)
773 const struct vc4_plane_state *vc4_state =
774 container_of(state, typeof(*vc4_state), base);
776 return vc4_state->dlist_count;
779 /* Updates the plane to immediately (well, once the FIFO needs
780 * refilling) scan out from at a new framebuffer.
782 void vc4_plane_async_set_fb(struct drm_plane *plane, struct drm_framebuffer *fb)
784 struct vc4_plane_state *vc4_state = to_vc4_plane_state(plane->state);
785 struct drm_gem_cma_object *bo = drm_fb_cma_get_gem_obj(fb, 0);
788 /* We're skipping the address adjustment for negative origin,
789 * because this is only called on the primary plane.
791 WARN_ON_ONCE(plane->state->crtc_x < 0 || plane->state->crtc_y < 0);
792 addr = bo->paddr + fb->offsets[0];
794 /* Write the new address into the hardware immediately. The
795 * scanout will start from this address as soon as the FIFO
796 * needs to refill with pixels.
798 writel(addr, &vc4_state->hw_dlist[vc4_state->ptr0_offset]);
800 /* Also update the CPU-side dlist copy, so that any later
801 * atomic updates that don't do a new modeset on our plane
802 * also use our updated address.
804 vc4_state->dlist[vc4_state->ptr0_offset] = addr;
807 static void vc4_plane_atomic_async_update(struct drm_plane *plane,
808 struct drm_plane_state *state)
810 struct vc4_plane_state *vc4_state = to_vc4_plane_state(plane->state);
812 if (plane->state->fb != state->fb) {
813 vc4_plane_async_set_fb(plane, state->fb);
814 drm_atomic_set_fb_for_plane(plane->state, state->fb);
817 /* Set the cursor's position on the screen. This is the
818 * expected change from the drm_mode_cursor_universal()
821 plane->state->crtc_x = state->crtc_x;
822 plane->state->crtc_y = state->crtc_y;
824 /* Allow changing the start position within the cursor BO, if
827 plane->state->src_x = state->src_x;
828 plane->state->src_y = state->src_y;
830 /* Update the display list based on the new crtc_x/y. */
831 vc4_plane_atomic_check(plane, plane->state);
833 /* Note that we can't just call vc4_plane_write_dlist()
834 * because that would smash the context data that the HVS is
837 writel(vc4_state->dlist[vc4_state->pos0_offset],
838 &vc4_state->hw_dlist[vc4_state->pos0_offset]);
839 writel(vc4_state->dlist[vc4_state->pos2_offset],
840 &vc4_state->hw_dlist[vc4_state->pos2_offset]);
841 writel(vc4_state->dlist[vc4_state->ptr0_offset],
842 &vc4_state->hw_dlist[vc4_state->ptr0_offset]);
845 static int vc4_plane_atomic_async_check(struct drm_plane *plane,
846 struct drm_plane_state *state)
848 /* No configuring new scaling in the fast path. */
849 if (plane->state->crtc_w != state->crtc_w ||
850 plane->state->crtc_h != state->crtc_h ||
851 plane->state->src_w != state->src_w ||
852 plane->state->src_h != state->src_h)
858 static int vc4_prepare_fb(struct drm_plane *plane,
859 struct drm_plane_state *state)
862 struct dma_fence *fence;
868 bo = to_vc4_bo(&drm_fb_cma_get_gem_obj(state->fb, 0)->base);
870 fence = reservation_object_get_excl_rcu(bo->resv);
871 drm_atomic_set_fence_for_plane(state, fence);
873 if (plane->state->fb == state->fb)
876 ret = vc4_bo_inc_usecnt(bo);
883 static void vc4_cleanup_fb(struct drm_plane *plane,
884 struct drm_plane_state *state)
888 if (plane->state->fb == state->fb || !state->fb)
891 bo = to_vc4_bo(&drm_fb_cma_get_gem_obj(state->fb, 0)->base);
892 vc4_bo_dec_usecnt(bo);
895 static const struct drm_plane_helper_funcs vc4_plane_helper_funcs = {
896 .atomic_check = vc4_plane_atomic_check,
897 .atomic_update = vc4_plane_atomic_update,
898 .prepare_fb = vc4_prepare_fb,
899 .cleanup_fb = vc4_cleanup_fb,
900 .atomic_async_check = vc4_plane_atomic_async_check,
901 .atomic_async_update = vc4_plane_atomic_async_update,
904 static void vc4_plane_destroy(struct drm_plane *plane)
906 drm_plane_helper_disable(plane, NULL);
907 drm_plane_cleanup(plane);
910 static bool vc4_format_mod_supported(struct drm_plane *plane,
914 /* Support T_TILING for RGB formats only. */
916 case DRM_FORMAT_XRGB8888:
917 case DRM_FORMAT_ARGB8888:
918 case DRM_FORMAT_ABGR8888:
919 case DRM_FORMAT_XBGR8888:
920 case DRM_FORMAT_RGB565:
921 case DRM_FORMAT_BGR565:
922 case DRM_FORMAT_ARGB1555:
923 case DRM_FORMAT_XRGB1555:
924 switch (fourcc_mod_broadcom_mod(modifier)) {
925 case DRM_FORMAT_MOD_LINEAR:
926 case DRM_FORMAT_MOD_BROADCOM_VC4_T_TILED:
927 case DRM_FORMAT_MOD_BROADCOM_SAND64:
928 case DRM_FORMAT_MOD_BROADCOM_SAND128:
933 case DRM_FORMAT_NV12:
934 case DRM_FORMAT_NV21:
935 switch (fourcc_mod_broadcom_mod(modifier)) {
936 case DRM_FORMAT_MOD_LINEAR:
937 case DRM_FORMAT_MOD_BROADCOM_SAND64:
938 case DRM_FORMAT_MOD_BROADCOM_SAND128:
939 case DRM_FORMAT_MOD_BROADCOM_SAND256:
944 case DRM_FORMAT_YUV422:
945 case DRM_FORMAT_YVU422:
946 case DRM_FORMAT_YUV420:
947 case DRM_FORMAT_YVU420:
948 case DRM_FORMAT_NV16:
949 case DRM_FORMAT_NV61:
951 return (modifier == DRM_FORMAT_MOD_LINEAR);
955 static const struct drm_plane_funcs vc4_plane_funcs = {
956 .update_plane = drm_atomic_helper_update_plane,
957 .disable_plane = drm_atomic_helper_disable_plane,
958 .destroy = vc4_plane_destroy,
959 .set_property = NULL,
960 .reset = vc4_plane_reset,
961 .atomic_duplicate_state = vc4_plane_duplicate_state,
962 .atomic_destroy_state = vc4_plane_destroy_state,
963 .format_mod_supported = vc4_format_mod_supported,
966 struct drm_plane *vc4_plane_init(struct drm_device *dev,
967 enum drm_plane_type type)
969 struct drm_plane *plane = NULL;
970 struct vc4_plane *vc4_plane;
971 u32 formats[ARRAY_SIZE(hvs_formats)];
975 static const uint64_t modifiers[] = {
976 DRM_FORMAT_MOD_BROADCOM_VC4_T_TILED,
977 DRM_FORMAT_MOD_BROADCOM_SAND128,
978 DRM_FORMAT_MOD_BROADCOM_SAND64,
979 DRM_FORMAT_MOD_BROADCOM_SAND256,
980 DRM_FORMAT_MOD_LINEAR,
981 DRM_FORMAT_MOD_INVALID
984 vc4_plane = devm_kzalloc(dev->dev, sizeof(*vc4_plane),
987 return ERR_PTR(-ENOMEM);
989 for (i = 0; i < ARRAY_SIZE(hvs_formats); i++) {
990 /* Don't allow YUV in cursor planes, since that means
991 * tuning on the scaler, which we don't allow for the
994 if (type != DRM_PLANE_TYPE_CURSOR ||
995 hvs_formats[i].hvs < HVS_PIXEL_FORMAT_YCBCR_YUV420_3PLANE) {
996 formats[num_formats++] = hvs_formats[i].drm;
999 plane = &vc4_plane->base;
1000 ret = drm_universal_plane_init(dev, plane, 0,
1002 formats, num_formats,
1003 modifiers, type, NULL);
1005 drm_plane_helper_add(plane, &vc4_plane_helper_funcs);
1007 drm_plane_create_alpha_property(plane);