Linux Audio

Check our new training course

Loading...
Note: File does not exist in v3.1.
   1// SPDX-License-Identifier: GPL-2.0-only
   2/*
   3 * Copyright (C) 2015 Broadcom
   4 */
   5
   6/**
   7 * DOC: VC4 plane module
   8 *
   9 * Each DRM plane is a layer of pixels being scanned out by the HVS.
  10 *
  11 * At atomic modeset check time, we compute the HVS display element
  12 * state that would be necessary for displaying the plane (giving us a
  13 * chance to figure out if a plane configuration is invalid), then at
  14 * atomic flush time the CRTC will ask us to write our element state
  15 * into the region of the HVS that it has allocated for us.
  16 */
  17
  18#include <drm/drm_atomic.h>
  19#include <drm/drm_atomic_helper.h>
  20#include <drm/drm_atomic_uapi.h>
  21#include <drm/drm_blend.h>
  22#include <drm/drm_drv.h>
  23#include <drm/drm_fb_dma_helper.h>
  24#include <drm/drm_fourcc.h>
  25#include <drm/drm_framebuffer.h>
  26#include <drm/drm_gem_atomic_helper.h>
  27
  28#include "uapi/drm/vc4_drm.h"
  29
  30#include "vc4_drv.h"
  31#include "vc4_regs.h"
  32
  33static const struct hvs_format {
  34	u32 drm; /* DRM_FORMAT_* */
  35	u32 hvs; /* HVS_FORMAT_* */
  36	u32 pixel_order;
  37	u32 pixel_order_hvs5;
  38	bool hvs5_only;
  39} hvs_formats[] = {
  40	{
  41		.drm = DRM_FORMAT_XRGB8888,
  42		.hvs = HVS_PIXEL_FORMAT_RGBA8888,
  43		.pixel_order = HVS_PIXEL_ORDER_ABGR,
  44		.pixel_order_hvs5 = HVS_PIXEL_ORDER_ARGB,
  45	},
  46	{
  47		.drm = DRM_FORMAT_ARGB8888,
  48		.hvs = HVS_PIXEL_FORMAT_RGBA8888,
  49		.pixel_order = HVS_PIXEL_ORDER_ABGR,
  50		.pixel_order_hvs5 = HVS_PIXEL_ORDER_ARGB,
  51	},
  52	{
  53		.drm = DRM_FORMAT_ABGR8888,
  54		.hvs = HVS_PIXEL_FORMAT_RGBA8888,
  55		.pixel_order = HVS_PIXEL_ORDER_ARGB,
  56		.pixel_order_hvs5 = HVS_PIXEL_ORDER_ABGR,
  57	},
  58	{
  59		.drm = DRM_FORMAT_XBGR8888,
  60		.hvs = HVS_PIXEL_FORMAT_RGBA8888,
  61		.pixel_order = HVS_PIXEL_ORDER_ARGB,
  62		.pixel_order_hvs5 = HVS_PIXEL_ORDER_ABGR,
  63	},
  64	{
  65		.drm = DRM_FORMAT_RGB565,
  66		.hvs = HVS_PIXEL_FORMAT_RGB565,
  67		.pixel_order = HVS_PIXEL_ORDER_XRGB,
  68	},
  69	{
  70		.drm = DRM_FORMAT_BGR565,
  71		.hvs = HVS_PIXEL_FORMAT_RGB565,
  72		.pixel_order = HVS_PIXEL_ORDER_XBGR,
  73	},
  74	{
  75		.drm = DRM_FORMAT_ARGB1555,
  76		.hvs = HVS_PIXEL_FORMAT_RGBA5551,
  77		.pixel_order = HVS_PIXEL_ORDER_ABGR,
  78	},
  79	{
  80		.drm = DRM_FORMAT_XRGB1555,
  81		.hvs = HVS_PIXEL_FORMAT_RGBA5551,
  82		.pixel_order = HVS_PIXEL_ORDER_ABGR,
  83	},
  84	{
  85		.drm = DRM_FORMAT_RGB888,
  86		.hvs = HVS_PIXEL_FORMAT_RGB888,
  87		.pixel_order = HVS_PIXEL_ORDER_XRGB,
  88	},
  89	{
  90		.drm = DRM_FORMAT_BGR888,
  91		.hvs = HVS_PIXEL_FORMAT_RGB888,
  92		.pixel_order = HVS_PIXEL_ORDER_XBGR,
  93	},
  94	{
  95		.drm = DRM_FORMAT_YUV422,
  96		.hvs = HVS_PIXEL_FORMAT_YCBCR_YUV422_3PLANE,
  97		.pixel_order = HVS_PIXEL_ORDER_XYCBCR,
  98	},
  99	{
 100		.drm = DRM_FORMAT_YVU422,
 101		.hvs = HVS_PIXEL_FORMAT_YCBCR_YUV422_3PLANE,
 102		.pixel_order = HVS_PIXEL_ORDER_XYCRCB,
 103	},
 104	{
 105		.drm = DRM_FORMAT_YUV420,
 106		.hvs = HVS_PIXEL_FORMAT_YCBCR_YUV420_3PLANE,
 107		.pixel_order = HVS_PIXEL_ORDER_XYCBCR,
 108	},
 109	{
 110		.drm = DRM_FORMAT_YVU420,
 111		.hvs = HVS_PIXEL_FORMAT_YCBCR_YUV420_3PLANE,
 112		.pixel_order = HVS_PIXEL_ORDER_XYCRCB,
 113	},
 114	{
 115		.drm = DRM_FORMAT_NV12,
 116		.hvs = HVS_PIXEL_FORMAT_YCBCR_YUV420_2PLANE,
 117		.pixel_order = HVS_PIXEL_ORDER_XYCBCR,
 118	},
 119	{
 120		.drm = DRM_FORMAT_NV21,
 121		.hvs = HVS_PIXEL_FORMAT_YCBCR_YUV420_2PLANE,
 122		.pixel_order = HVS_PIXEL_ORDER_XYCRCB,
 123	},
 124	{
 125		.drm = DRM_FORMAT_NV16,
 126		.hvs = HVS_PIXEL_FORMAT_YCBCR_YUV422_2PLANE,
 127		.pixel_order = HVS_PIXEL_ORDER_XYCBCR,
 128	},
 129	{
 130		.drm = DRM_FORMAT_NV61,
 131		.hvs = HVS_PIXEL_FORMAT_YCBCR_YUV422_2PLANE,
 132		.pixel_order = HVS_PIXEL_ORDER_XYCRCB,
 133	},
 134	{
 135		.drm = DRM_FORMAT_P030,
 136		.hvs = HVS_PIXEL_FORMAT_YCBCR_10BIT,
 137		.pixel_order = HVS_PIXEL_ORDER_XYCBCR,
 138		.hvs5_only = true,
 139	},
 140};
 141
 142static const struct hvs_format *vc4_get_hvs_format(u32 drm_format)
 143{
 144	unsigned i;
 145
 146	for (i = 0; i < ARRAY_SIZE(hvs_formats); i++) {
 147		if (hvs_formats[i].drm == drm_format)
 148			return &hvs_formats[i];
 149	}
 150
 151	return NULL;
 152}
 153
 154static enum vc4_scaling_mode vc4_get_scaling_mode(u32 src, u32 dst)
 155{
 156	if (dst == src)
 157		return VC4_SCALING_NONE;
 158	if (3 * dst >= 2 * src)
 159		return VC4_SCALING_PPF;
 160	else
 161		return VC4_SCALING_TPZ;
 162}
 163
 164static bool plane_enabled(struct drm_plane_state *state)
 165{
 166	return state->fb && !WARN_ON(!state->crtc);
 167}
 168
 169static struct drm_plane_state *vc4_plane_duplicate_state(struct drm_plane *plane)
 170{
 171	struct vc4_plane_state *vc4_state;
 172
 173	if (WARN_ON(!plane->state))
 174		return NULL;
 175
 176	vc4_state = kmemdup(plane->state, sizeof(*vc4_state), GFP_KERNEL);
 177	if (!vc4_state)
 178		return NULL;
 179
 180	memset(&vc4_state->lbm, 0, sizeof(vc4_state->lbm));
 181	vc4_state->dlist_initialized = 0;
 182
 183	__drm_atomic_helper_plane_duplicate_state(plane, &vc4_state->base);
 184
 185	if (vc4_state->dlist) {
 186		vc4_state->dlist = kmemdup(vc4_state->dlist,
 187					   vc4_state->dlist_count * 4,
 188					   GFP_KERNEL);
 189		if (!vc4_state->dlist) {
 190			kfree(vc4_state);
 191			return NULL;
 192		}
 193		vc4_state->dlist_size = vc4_state->dlist_count;
 194	}
 195
 196	return &vc4_state->base;
 197}
 198
 199static void vc4_plane_destroy_state(struct drm_plane *plane,
 200				    struct drm_plane_state *state)
 201{
 202	struct vc4_dev *vc4 = to_vc4_dev(plane->dev);
 203	struct vc4_plane_state *vc4_state = to_vc4_plane_state(state);
 204
 205	if (drm_mm_node_allocated(&vc4_state->lbm)) {
 206		unsigned long irqflags;
 207
 208		spin_lock_irqsave(&vc4->hvs->mm_lock, irqflags);
 209		drm_mm_remove_node(&vc4_state->lbm);
 210		spin_unlock_irqrestore(&vc4->hvs->mm_lock, irqflags);
 211	}
 212
 213	kfree(vc4_state->dlist);
 214	__drm_atomic_helper_plane_destroy_state(&vc4_state->base);
 215	kfree(state);
 216}
 217
 218/* Called during init to allocate the plane's atomic state. */
 219static void vc4_plane_reset(struct drm_plane *plane)
 220{
 221	struct vc4_plane_state *vc4_state;
 222
 223	WARN_ON(plane->state);
 224
 225	vc4_state = kzalloc(sizeof(*vc4_state), GFP_KERNEL);
 226	if (!vc4_state)
 227		return;
 228
 229	__drm_atomic_helper_plane_reset(plane, &vc4_state->base);
 230}
 231
 232static void vc4_dlist_counter_increment(struct vc4_plane_state *vc4_state)
 233{
 234	if (vc4_state->dlist_count == vc4_state->dlist_size) {
 235		u32 new_size = max(4u, vc4_state->dlist_count * 2);
 236		u32 *new_dlist = kmalloc_array(new_size, 4, GFP_KERNEL);
 237
 238		if (!new_dlist)
 239			return;
 240		memcpy(new_dlist, vc4_state->dlist, vc4_state->dlist_count * 4);
 241
 242		kfree(vc4_state->dlist);
 243		vc4_state->dlist = new_dlist;
 244		vc4_state->dlist_size = new_size;
 245	}
 246
 247	vc4_state->dlist_count++;
 248}
 249
 250static void vc4_dlist_write(struct vc4_plane_state *vc4_state, u32 val)
 251{
 252	unsigned int idx = vc4_state->dlist_count;
 253
 254	vc4_dlist_counter_increment(vc4_state);
 255	vc4_state->dlist[idx] = val;
 256}
 257
 258/* Returns the scl0/scl1 field based on whether the dimensions need to
 259 * be up/down/non-scaled.
 260 *
 261 * This is a replication of a table from the spec.
 262 */
 263static u32 vc4_get_scl_field(struct drm_plane_state *state, int plane)
 264{
 265	struct vc4_plane_state *vc4_state = to_vc4_plane_state(state);
 266
 267	switch (vc4_state->x_scaling[plane] << 2 | vc4_state->y_scaling[plane]) {
 268	case VC4_SCALING_PPF << 2 | VC4_SCALING_PPF:
 269		return SCALER_CTL0_SCL_H_PPF_V_PPF;
 270	case VC4_SCALING_TPZ << 2 | VC4_SCALING_PPF:
 271		return SCALER_CTL0_SCL_H_TPZ_V_PPF;
 272	case VC4_SCALING_PPF << 2 | VC4_SCALING_TPZ:
 273		return SCALER_CTL0_SCL_H_PPF_V_TPZ;
 274	case VC4_SCALING_TPZ << 2 | VC4_SCALING_TPZ:
 275		return SCALER_CTL0_SCL_H_TPZ_V_TPZ;
 276	case VC4_SCALING_PPF << 2 | VC4_SCALING_NONE:
 277		return SCALER_CTL0_SCL_H_PPF_V_NONE;
 278	case VC4_SCALING_NONE << 2 | VC4_SCALING_PPF:
 279		return SCALER_CTL0_SCL_H_NONE_V_PPF;
 280	case VC4_SCALING_NONE << 2 | VC4_SCALING_TPZ:
 281		return SCALER_CTL0_SCL_H_NONE_V_TPZ;
 282	case VC4_SCALING_TPZ << 2 | VC4_SCALING_NONE:
 283		return SCALER_CTL0_SCL_H_TPZ_V_NONE;
 284	default:
 285	case VC4_SCALING_NONE << 2 | VC4_SCALING_NONE:
 286		/* The unity case is independently handled by
 287		 * SCALER_CTL0_UNITY.
 288		 */
 289		return 0;
 290	}
 291}
 292
 293static int vc4_plane_margins_adj(struct drm_plane_state *pstate)
 294{
 295	struct vc4_plane_state *vc4_pstate = to_vc4_plane_state(pstate);
 296	unsigned int left, right, top, bottom, adjhdisplay, adjvdisplay;
 297	struct drm_crtc_state *crtc_state;
 298
 299	crtc_state = drm_atomic_get_new_crtc_state(pstate->state,
 300						   pstate->crtc);
 301
 302	vc4_crtc_get_margins(crtc_state, &left, &right, &top, &bottom);
 303	if (!left && !right && !top && !bottom)
 304		return 0;
 305
 306	if (left + right >= crtc_state->mode.hdisplay ||
 307	    top + bottom >= crtc_state->mode.vdisplay)
 308		return -EINVAL;
 309
 310	adjhdisplay = crtc_state->mode.hdisplay - (left + right);
 311	vc4_pstate->crtc_x = DIV_ROUND_CLOSEST(vc4_pstate->crtc_x *
 312					       adjhdisplay,
 313					       crtc_state->mode.hdisplay);
 314	vc4_pstate->crtc_x += left;
 315	if (vc4_pstate->crtc_x > crtc_state->mode.hdisplay - right)
 316		vc4_pstate->crtc_x = crtc_state->mode.hdisplay - right;
 317
 318	adjvdisplay = crtc_state->mode.vdisplay - (top + bottom);
 319	vc4_pstate->crtc_y = DIV_ROUND_CLOSEST(vc4_pstate->crtc_y *
 320					       adjvdisplay,
 321					       crtc_state->mode.vdisplay);
 322	vc4_pstate->crtc_y += top;
 323	if (vc4_pstate->crtc_y > crtc_state->mode.vdisplay - bottom)
 324		vc4_pstate->crtc_y = crtc_state->mode.vdisplay - bottom;
 325
 326	vc4_pstate->crtc_w = DIV_ROUND_CLOSEST(vc4_pstate->crtc_w *
 327					       adjhdisplay,
 328					       crtc_state->mode.hdisplay);
 329	vc4_pstate->crtc_h = DIV_ROUND_CLOSEST(vc4_pstate->crtc_h *
 330					       adjvdisplay,
 331					       crtc_state->mode.vdisplay);
 332
 333	if (!vc4_pstate->crtc_w || !vc4_pstate->crtc_h)
 334		return -EINVAL;
 335
 336	return 0;
 337}
 338
 339static int vc4_plane_setup_clipping_and_scaling(struct drm_plane_state *state)
 340{
 341	struct vc4_plane_state *vc4_state = to_vc4_plane_state(state);
 342	struct drm_framebuffer *fb = state->fb;
 343	struct drm_gem_dma_object *bo;
 344	int num_planes = fb->format->num_planes;
 345	struct drm_crtc_state *crtc_state;
 346	u32 h_subsample = fb->format->hsub;
 347	u32 v_subsample = fb->format->vsub;
 348	int i, ret;
 349
 350	crtc_state = drm_atomic_get_existing_crtc_state(state->state,
 351							state->crtc);
 352	if (!crtc_state) {
 353		DRM_DEBUG_KMS("Invalid crtc state\n");
 354		return -EINVAL;
 355	}
 356
 357	ret = drm_atomic_helper_check_plane_state(state, crtc_state, 1,
 358						  INT_MAX, true, true);
 359	if (ret)
 360		return ret;
 361
 362	for (i = 0; i < num_planes; i++) {
 363		bo = drm_fb_dma_get_gem_obj(fb, i);
 364		vc4_state->offsets[i] = bo->dma_addr + fb->offsets[i];
 365	}
 366
 367	/*
 368	 * We don't support subpixel source positioning for scaling,
 369	 * but fractional coordinates can be generated by clipping
 370	 * so just round for now
 371	 */
 372	vc4_state->src_x = DIV_ROUND_CLOSEST(state->src.x1, 1 << 16);
 373	vc4_state->src_y = DIV_ROUND_CLOSEST(state->src.y1, 1 << 16);
 374	vc4_state->src_w[0] = DIV_ROUND_CLOSEST(state->src.x2, 1 << 16) - vc4_state->src_x;
 375	vc4_state->src_h[0] = DIV_ROUND_CLOSEST(state->src.y2, 1 << 16) - vc4_state->src_y;
 376
 377	vc4_state->crtc_x = state->dst.x1;
 378	vc4_state->crtc_y = state->dst.y1;
 379	vc4_state->crtc_w = state->dst.x2 - state->dst.x1;
 380	vc4_state->crtc_h = state->dst.y2 - state->dst.y1;
 381
 382	ret = vc4_plane_margins_adj(state);
 383	if (ret)
 384		return ret;
 385
 386	vc4_state->x_scaling[0] = vc4_get_scaling_mode(vc4_state->src_w[0],
 387						       vc4_state->crtc_w);
 388	vc4_state->y_scaling[0] = vc4_get_scaling_mode(vc4_state->src_h[0],
 389						       vc4_state->crtc_h);
 390
 391	vc4_state->is_unity = (vc4_state->x_scaling[0] == VC4_SCALING_NONE &&
 392			       vc4_state->y_scaling[0] == VC4_SCALING_NONE);
 393
 394	if (num_planes > 1) {
 395		vc4_state->is_yuv = true;
 396
 397		vc4_state->src_w[1] = vc4_state->src_w[0] / h_subsample;
 398		vc4_state->src_h[1] = vc4_state->src_h[0] / v_subsample;
 399
 400		vc4_state->x_scaling[1] =
 401			vc4_get_scaling_mode(vc4_state->src_w[1],
 402					     vc4_state->crtc_w);
 403		vc4_state->y_scaling[1] =
 404			vc4_get_scaling_mode(vc4_state->src_h[1],
 405					     vc4_state->crtc_h);
 406
 407		/* YUV conversion requires that horizontal scaling be enabled
 408		 * on the UV plane even if vc4_get_scaling_mode() returned
 409		 * VC4_SCALING_NONE (which can happen when the down-scaling
 410		 * ratio is 0.5). Let's force it to VC4_SCALING_PPF in this
 411		 * case.
 412		 */
 413		if (vc4_state->x_scaling[1] == VC4_SCALING_NONE)
 414			vc4_state->x_scaling[1] = VC4_SCALING_PPF;
 415	} else {
 416		vc4_state->is_yuv = false;
 417		vc4_state->x_scaling[1] = VC4_SCALING_NONE;
 418		vc4_state->y_scaling[1] = VC4_SCALING_NONE;
 419	}
 420
 421	return 0;
 422}
 423
 424static void vc4_write_tpz(struct vc4_plane_state *vc4_state, u32 src, u32 dst)
 425{
 426	u32 scale, recip;
 427
 428	scale = (1 << 16) * src / dst;
 429
 430	/* The specs note that while the reciprocal would be defined
 431	 * as (1<<32)/scale, ~0 is close enough.
 432	 */
 433	recip = ~0 / scale;
 434
 435	vc4_dlist_write(vc4_state,
 436			VC4_SET_FIELD(scale, SCALER_TPZ0_SCALE) |
 437			VC4_SET_FIELD(0, SCALER_TPZ0_IPHASE));
 438	vc4_dlist_write(vc4_state,
 439			VC4_SET_FIELD(recip, SCALER_TPZ1_RECIP));
 440}
 441
 442static void vc4_write_ppf(struct vc4_plane_state *vc4_state, u32 src, u32 dst)
 443{
 444	u32 scale = (1 << 16) * src / dst;
 445
 446	vc4_dlist_write(vc4_state,
 447			SCALER_PPF_AGC |
 448			VC4_SET_FIELD(scale, SCALER_PPF_SCALE) |
 449			VC4_SET_FIELD(0, SCALER_PPF_IPHASE));
 450}
 451
 452static u32 vc4_lbm_size(struct drm_plane_state *state)
 453{
 454	struct vc4_plane_state *vc4_state = to_vc4_plane_state(state);
 455	struct vc4_dev *vc4 = to_vc4_dev(state->plane->dev);
 456	u32 pix_per_line;
 457	u32 lbm;
 458
 459	/* LBM is not needed when there's no vertical scaling. */
 460	if (vc4_state->y_scaling[0] == VC4_SCALING_NONE &&
 461	    vc4_state->y_scaling[1] == VC4_SCALING_NONE)
 462		return 0;
 463
 464	/*
 465	 * This can be further optimized in the RGB/YUV444 case if the PPF
 466	 * decimation factor is between 0.5 and 1.0 by using crtc_w.
 467	 *
 468	 * It's not an issue though, since in that case since src_w[0] is going
 469	 * to be greater than or equal to crtc_w.
 470	 */
 471	if (vc4_state->x_scaling[0] == VC4_SCALING_TPZ)
 472		pix_per_line = vc4_state->crtc_w;
 473	else
 474		pix_per_line = vc4_state->src_w[0];
 475
 476	if (!vc4_state->is_yuv) {
 477		if (vc4_state->y_scaling[0] == VC4_SCALING_TPZ)
 478			lbm = pix_per_line * 8;
 479		else {
 480			/* In special cases, this multiplier might be 12. */
 481			lbm = pix_per_line * 16;
 482		}
 483	} else {
 484		/* There are cases for this going down to a multiplier
 485		 * of 2, but according to the firmware source, the
 486		 * table in the docs is somewhat wrong.
 487		 */
 488		lbm = pix_per_line * 16;
 489	}
 490
 491	/* Align it to 64 or 128 (hvs5) bytes */
 492	lbm = roundup(lbm, vc4->is_vc5 ? 128 : 64);
 493
 494	/* Each "word" of the LBM memory contains 2 or 4 (hvs5) pixels */
 495	lbm /= vc4->is_vc5 ? 4 : 2;
 496
 497	return lbm;
 498}
 499
 500static void vc4_write_scaling_parameters(struct drm_plane_state *state,
 501					 int channel)
 502{
 503	struct vc4_plane_state *vc4_state = to_vc4_plane_state(state);
 504
 505	/* Ch0 H-PPF Word 0: Scaling Parameters */
 506	if (vc4_state->x_scaling[channel] == VC4_SCALING_PPF) {
 507		vc4_write_ppf(vc4_state,
 508			      vc4_state->src_w[channel], vc4_state->crtc_w);
 509	}
 510
 511	/* Ch0 V-PPF Words 0-1: Scaling Parameters, Context */
 512	if (vc4_state->y_scaling[channel] == VC4_SCALING_PPF) {
 513		vc4_write_ppf(vc4_state,
 514			      vc4_state->src_h[channel], vc4_state->crtc_h);
 515		vc4_dlist_write(vc4_state, 0xc0c0c0c0);
 516	}
 517
 518	/* Ch0 H-TPZ Words 0-1: Scaling Parameters, Recip */
 519	if (vc4_state->x_scaling[channel] == VC4_SCALING_TPZ) {
 520		vc4_write_tpz(vc4_state,
 521			      vc4_state->src_w[channel], vc4_state->crtc_w);
 522	}
 523
 524	/* Ch0 V-TPZ Words 0-2: Scaling Parameters, Recip, Context */
 525	if (vc4_state->y_scaling[channel] == VC4_SCALING_TPZ) {
 526		vc4_write_tpz(vc4_state,
 527			      vc4_state->src_h[channel], vc4_state->crtc_h);
 528		vc4_dlist_write(vc4_state, 0xc0c0c0c0);
 529	}
 530}
 531
 532static void vc4_plane_calc_load(struct drm_plane_state *state)
 533{
 534	unsigned int hvs_load_shift, vrefresh, i;
 535	struct drm_framebuffer *fb = state->fb;
 536	struct vc4_plane_state *vc4_state;
 537	struct drm_crtc_state *crtc_state;
 538	unsigned int vscale_factor;
 539
 540	vc4_state = to_vc4_plane_state(state);
 541	crtc_state = drm_atomic_get_existing_crtc_state(state->state,
 542							state->crtc);
 543	vrefresh = drm_mode_vrefresh(&crtc_state->adjusted_mode);
 544
 545	/* The HVS is able to process 2 pixels/cycle when scaling the source,
 546	 * 4 pixels/cycle otherwise.
 547	 * Alpha blending step seems to be pipelined and it's always operating
 548	 * at 4 pixels/cycle, so the limiting aspect here seems to be the
 549	 * scaler block.
 550	 * HVS load is expressed in clk-cycles/sec (AKA Hz).
 551	 */
 552	if (vc4_state->x_scaling[0] != VC4_SCALING_NONE ||
 553	    vc4_state->x_scaling[1] != VC4_SCALING_NONE ||
 554	    vc4_state->y_scaling[0] != VC4_SCALING_NONE ||
 555	    vc4_state->y_scaling[1] != VC4_SCALING_NONE)
 556		hvs_load_shift = 1;
 557	else
 558		hvs_load_shift = 2;
 559
 560	vc4_state->membus_load = 0;
 561	vc4_state->hvs_load = 0;
 562	for (i = 0; i < fb->format->num_planes; i++) {
 563		/* Even if the bandwidth/plane required for a single frame is
 564		 *
 565		 * vc4_state->src_w[i] * vc4_state->src_h[i] * cpp * vrefresh
 566		 *
 567		 * when downscaling, we have to read more pixels per line in
 568		 * the time frame reserved for a single line, so the bandwidth
 569		 * demand can be punctually higher. To account for that, we
 570		 * calculate the down-scaling factor and multiply the plane
 571		 * load by this number. We're likely over-estimating the read
 572		 * demand, but that's better than under-estimating it.
 573		 */
 574		vscale_factor = DIV_ROUND_UP(vc4_state->src_h[i],
 575					     vc4_state->crtc_h);
 576		vc4_state->membus_load += vc4_state->src_w[i] *
 577					  vc4_state->src_h[i] * vscale_factor *
 578					  fb->format->cpp[i];
 579		vc4_state->hvs_load += vc4_state->crtc_h * vc4_state->crtc_w;
 580	}
 581
 582	vc4_state->hvs_load *= vrefresh;
 583	vc4_state->hvs_load >>= hvs_load_shift;
 584	vc4_state->membus_load *= vrefresh;
 585}
 586
 587static int vc4_plane_allocate_lbm(struct drm_plane_state *state)
 588{
 589	struct vc4_dev *vc4 = to_vc4_dev(state->plane->dev);
 590	struct vc4_plane_state *vc4_state = to_vc4_plane_state(state);
 591	unsigned long irqflags;
 592	u32 lbm_size;
 593
 594	lbm_size = vc4_lbm_size(state);
 595	if (!lbm_size)
 596		return 0;
 597
 598	if (WARN_ON(!vc4_state->lbm_offset))
 599		return -EINVAL;
 600
 601	/* Allocate the LBM memory that the HVS will use for temporary
 602	 * storage due to our scaling/format conversion.
 603	 */
 604	if (!drm_mm_node_allocated(&vc4_state->lbm)) {
 605		int ret;
 606
 607		spin_lock_irqsave(&vc4->hvs->mm_lock, irqflags);
 608		ret = drm_mm_insert_node_generic(&vc4->hvs->lbm_mm,
 609						 &vc4_state->lbm,
 610						 lbm_size,
 611						 vc4->is_vc5 ? 64 : 32,
 612						 0, 0);
 613		spin_unlock_irqrestore(&vc4->hvs->mm_lock, irqflags);
 614
 615		if (ret)
 616			return ret;
 617	} else {
 618		WARN_ON_ONCE(lbm_size != vc4_state->lbm.size);
 619	}
 620
 621	vc4_state->dlist[vc4_state->lbm_offset] = vc4_state->lbm.start;
 622
 623	return 0;
 624}
 625
 626/*
 627 * The colorspace conversion matrices are held in 3 entries in the dlist.
 628 * Create an array of them, with entries for each full and limited mode, and
 629 * each supported colorspace.
 630 */
 631static const u32 colorspace_coeffs[2][DRM_COLOR_ENCODING_MAX][3] = {
 632	{
 633		/* Limited range */
 634		{
 635			/* BT601 */
 636			SCALER_CSC0_ITR_R_601_5,
 637			SCALER_CSC1_ITR_R_601_5,
 638			SCALER_CSC2_ITR_R_601_5,
 639		}, {
 640			/* BT709 */
 641			SCALER_CSC0_ITR_R_709_3,
 642			SCALER_CSC1_ITR_R_709_3,
 643			SCALER_CSC2_ITR_R_709_3,
 644		}, {
 645			/* BT2020 */
 646			SCALER_CSC0_ITR_R_2020,
 647			SCALER_CSC1_ITR_R_2020,
 648			SCALER_CSC2_ITR_R_2020,
 649		}
 650	}, {
 651		/* Full range */
 652		{
 653			/* JFIF */
 654			SCALER_CSC0_JPEG_JFIF,
 655			SCALER_CSC1_JPEG_JFIF,
 656			SCALER_CSC2_JPEG_JFIF,
 657		}, {
 658			/* BT709 */
 659			SCALER_CSC0_ITR_R_709_3_FR,
 660			SCALER_CSC1_ITR_R_709_3_FR,
 661			SCALER_CSC2_ITR_R_709_3_FR,
 662		}, {
 663			/* BT2020 */
 664			SCALER_CSC0_ITR_R_2020_FR,
 665			SCALER_CSC1_ITR_R_2020_FR,
 666			SCALER_CSC2_ITR_R_2020_FR,
 667		}
 668	}
 669};
 670
 671static u32 vc4_hvs4_get_alpha_blend_mode(struct drm_plane_state *state)
 672{
 673	if (!state->fb->format->has_alpha)
 674		return VC4_SET_FIELD(SCALER_POS2_ALPHA_MODE_FIXED,
 675				     SCALER_POS2_ALPHA_MODE);
 676
 677	switch (state->pixel_blend_mode) {
 678	case DRM_MODE_BLEND_PIXEL_NONE:
 679		return VC4_SET_FIELD(SCALER_POS2_ALPHA_MODE_FIXED,
 680				     SCALER_POS2_ALPHA_MODE);
 681	default:
 682	case DRM_MODE_BLEND_PREMULTI:
 683		return VC4_SET_FIELD(SCALER_POS2_ALPHA_MODE_PIPELINE,
 684				     SCALER_POS2_ALPHA_MODE) |
 685			SCALER_POS2_ALPHA_PREMULT;
 686	case DRM_MODE_BLEND_COVERAGE:
 687		return VC4_SET_FIELD(SCALER_POS2_ALPHA_MODE_PIPELINE,
 688				     SCALER_POS2_ALPHA_MODE);
 689	}
 690}
 691
 692static u32 vc4_hvs5_get_alpha_blend_mode(struct drm_plane_state *state)
 693{
 694	if (!state->fb->format->has_alpha)
 695		return VC4_SET_FIELD(SCALER5_CTL2_ALPHA_MODE_FIXED,
 696				     SCALER5_CTL2_ALPHA_MODE);
 697
 698	switch (state->pixel_blend_mode) {
 699	case DRM_MODE_BLEND_PIXEL_NONE:
 700		return VC4_SET_FIELD(SCALER5_CTL2_ALPHA_MODE_FIXED,
 701				     SCALER5_CTL2_ALPHA_MODE);
 702	default:
 703	case DRM_MODE_BLEND_PREMULTI:
 704		return VC4_SET_FIELD(SCALER5_CTL2_ALPHA_MODE_PIPELINE,
 705				     SCALER5_CTL2_ALPHA_MODE) |
 706			SCALER5_CTL2_ALPHA_PREMULT;
 707	case DRM_MODE_BLEND_COVERAGE:
 708		return VC4_SET_FIELD(SCALER5_CTL2_ALPHA_MODE_PIPELINE,
 709				     SCALER5_CTL2_ALPHA_MODE);
 710	}
 711}
 712
 713/* Writes out a full display list for an active plane to the plane's
 714 * private dlist state.
 715 */
 716static int vc4_plane_mode_set(struct drm_plane *plane,
 717			      struct drm_plane_state *state)
 718{
 719	struct vc4_dev *vc4 = to_vc4_dev(plane->dev);
 720	struct vc4_plane_state *vc4_state = to_vc4_plane_state(state);
 721	struct drm_framebuffer *fb = state->fb;
 722	u32 ctl0_offset = vc4_state->dlist_count;
 723	const struct hvs_format *format = vc4_get_hvs_format(fb->format->format);
 724	u64 base_format_mod = fourcc_mod_broadcom_mod(fb->modifier);
 725	int num_planes = fb->format->num_planes;
 726	u32 h_subsample = fb->format->hsub;
 727	u32 v_subsample = fb->format->vsub;
 728	bool mix_plane_alpha;
 729	bool covers_screen;
 730	u32 scl0, scl1, pitch0;
 731	u32 tiling, src_y;
 732	u32 hvs_format = format->hvs;
 733	unsigned int rotation;
 734	int ret, i;
 735
 736	if (vc4_state->dlist_initialized)
 737		return 0;
 738
 739	ret = vc4_plane_setup_clipping_and_scaling(state);
 740	if (ret)
 741		return ret;
 742
 743	/* SCL1 is used for Cb/Cr scaling of planar formats.  For RGB
 744	 * and 4:4:4, scl1 should be set to scl0 so both channels of
 745	 * the scaler do the same thing.  For YUV, the Y plane needs
 746	 * to be put in channel 1 and Cb/Cr in channel 0, so we swap
 747	 * the scl fields here.
 748	 */
 749	if (num_planes == 1) {
 750		scl0 = vc4_get_scl_field(state, 0);
 751		scl1 = scl0;
 752	} else {
 753		scl0 = vc4_get_scl_field(state, 1);
 754		scl1 = vc4_get_scl_field(state, 0);
 755	}
 756
 757	rotation = drm_rotation_simplify(state->rotation,
 758					 DRM_MODE_ROTATE_0 |
 759					 DRM_MODE_REFLECT_X |
 760					 DRM_MODE_REFLECT_Y);
 761
 762	/* We must point to the last line when Y reflection is enabled. */
 763	src_y = vc4_state->src_y;
 764	if (rotation & DRM_MODE_REFLECT_Y)
 765		src_y += vc4_state->src_h[0] - 1;
 766
 767	switch (base_format_mod) {
 768	case DRM_FORMAT_MOD_LINEAR:
 769		tiling = SCALER_CTL0_TILING_LINEAR;
 770		pitch0 = VC4_SET_FIELD(fb->pitches[0], SCALER_SRC_PITCH);
 771
 772		/* Adjust the base pointer to the first pixel to be scanned
 773		 * out.
 774		 */
 775		for (i = 0; i < num_planes; i++) {
 776			vc4_state->offsets[i] += src_y /
 777						 (i ? v_subsample : 1) *
 778						 fb->pitches[i];
 779
 780			vc4_state->offsets[i] += vc4_state->src_x /
 781						 (i ? h_subsample : 1) *
 782						 fb->format->cpp[i];
 783		}
 784
 785		break;
 786
 787	case DRM_FORMAT_MOD_BROADCOM_VC4_T_TILED: {
 788		u32 tile_size_shift = 12; /* T tiles are 4kb */
 789		/* Whole-tile offsets, mostly for setting the pitch. */
 790		u32 tile_w_shift = fb->format->cpp[0] == 2 ? 6 : 5;
 791		u32 tile_h_shift = 5; /* 16 and 32bpp are 32 pixels high */
 792		u32 tile_w_mask = (1 << tile_w_shift) - 1;
 793		/* The height mask on 32-bit-per-pixel tiles is 63, i.e. twice
 794		 * the height (in pixels) of a 4k tile.
 795		 */
 796		u32 tile_h_mask = (2 << tile_h_shift) - 1;
 797		/* For T-tiled, the FB pitch is "how many bytes from one row to
 798		 * the next, such that
 799		 *
 800		 *	pitch * tile_h == tile_size * tiles_per_row
 801		 */
 802		u32 tiles_w = fb->pitches[0] >> (tile_size_shift - tile_h_shift);
 803		u32 tiles_l = vc4_state->src_x >> tile_w_shift;
 804		u32 tiles_r = tiles_w - tiles_l;
 805		u32 tiles_t = src_y >> tile_h_shift;
 806		/* Intra-tile offsets, which modify the base address (the
 807		 * SCALER_PITCH0_TILE_Y_OFFSET tells HVS how to walk from that
 808		 * base address).
 809		 */
 810		u32 tile_y = (src_y >> 4) & 1;
 811		u32 subtile_y = (src_y >> 2) & 3;
 812		u32 utile_y = src_y & 3;
 813		u32 x_off = vc4_state->src_x & tile_w_mask;
 814		u32 y_off = src_y & tile_h_mask;
 815
 816		/* When Y reflection is requested we must set the
 817		 * SCALER_PITCH0_TILE_LINE_DIR flag to tell HVS that all lines
 818		 * after the initial one should be fetched in descending order,
 819		 * which makes sense since we start from the last line and go
 820		 * backward.
 821		 * Don't know why we need y_off = max_y_off - y_off, but it's
 822		 * definitely required (I guess it's also related to the "going
 823		 * backward" situation).
 824		 */
 825		if (rotation & DRM_MODE_REFLECT_Y) {
 826			y_off = tile_h_mask - y_off;
 827			pitch0 = SCALER_PITCH0_TILE_LINE_DIR;
 828		} else {
 829			pitch0 = 0;
 830		}
 831
 832		tiling = SCALER_CTL0_TILING_256B_OR_T;
 833		pitch0 |= (VC4_SET_FIELD(x_off, SCALER_PITCH0_SINK_PIX) |
 834			   VC4_SET_FIELD(y_off, SCALER_PITCH0_TILE_Y_OFFSET) |
 835			   VC4_SET_FIELD(tiles_l, SCALER_PITCH0_TILE_WIDTH_L) |
 836			   VC4_SET_FIELD(tiles_r, SCALER_PITCH0_TILE_WIDTH_R));
 837		vc4_state->offsets[0] += tiles_t * (tiles_w << tile_size_shift);
 838		vc4_state->offsets[0] += subtile_y << 8;
 839		vc4_state->offsets[0] += utile_y << 4;
 840
 841		/* Rows of tiles alternate left-to-right and right-to-left. */
 842		if (tiles_t & 1) {
 843			pitch0 |= SCALER_PITCH0_TILE_INITIAL_LINE_DIR;
 844			vc4_state->offsets[0] += (tiles_w - tiles_l) <<
 845						 tile_size_shift;
 846			vc4_state->offsets[0] -= (1 + !tile_y) << 10;
 847		} else {
 848			vc4_state->offsets[0] += tiles_l << tile_size_shift;
 849			vc4_state->offsets[0] += tile_y << 10;
 850		}
 851
 852		break;
 853	}
 854
 855	case DRM_FORMAT_MOD_BROADCOM_SAND64:
 856	case DRM_FORMAT_MOD_BROADCOM_SAND128:
 857	case DRM_FORMAT_MOD_BROADCOM_SAND256: {
 858		uint32_t param = fourcc_mod_broadcom_param(fb->modifier);
 859
 860		if (param > SCALER_TILE_HEIGHT_MASK) {
 861			DRM_DEBUG_KMS("SAND height too large (%d)\n",
 862				      param);
 863			return -EINVAL;
 864		}
 865
 866		if (fb->format->format == DRM_FORMAT_P030) {
 867			hvs_format = HVS_PIXEL_FORMAT_YCBCR_10BIT;
 868			tiling = SCALER_CTL0_TILING_128B;
 869		} else {
 870			hvs_format = HVS_PIXEL_FORMAT_H264;
 871
 872			switch (base_format_mod) {
 873			case DRM_FORMAT_MOD_BROADCOM_SAND64:
 874				tiling = SCALER_CTL0_TILING_64B;
 875				break;
 876			case DRM_FORMAT_MOD_BROADCOM_SAND128:
 877				tiling = SCALER_CTL0_TILING_128B;
 878				break;
 879			case DRM_FORMAT_MOD_BROADCOM_SAND256:
 880				tiling = SCALER_CTL0_TILING_256B_OR_T;
 881				break;
 882			default:
 883				return -EINVAL;
 884			}
 885		}
 886
 887		/* Adjust the base pointer to the first pixel to be scanned
 888		 * out.
 889		 *
 890		 * For P030, y_ptr [31:4] is the 128bit word for the start pixel
 891		 * y_ptr [3:0] is the pixel (0-11) contained within that 128bit
 892		 * word that should be taken as the first pixel.
 893		 * Ditto uv_ptr [31:4] vs [3:0], however [3:0] contains the
 894		 * element within the 128bit word, eg for pixel 3 the value
 895		 * should be 6.
 896		 */
 897		for (i = 0; i < num_planes; i++) {
 898			u32 tile_w, tile, x_off, pix_per_tile;
 899
 900			if (fb->format->format == DRM_FORMAT_P030) {
 901				/*
 902				 * Spec says: bits [31:4] of the given address
 903				 * should point to the 128-bit word containing
 904				 * the desired starting pixel, and bits[3:0]
 905				 * should be between 0 and 11, indicating which
 906				 * of the 12-pixels in that 128-bit word is the
 907				 * first pixel to be used
 908				 */
 909				u32 remaining_pixels = vc4_state->src_x % 96;
 910				u32 aligned = remaining_pixels / 12;
 911				u32 last_bits = remaining_pixels % 12;
 912
 913				x_off = aligned * 16 + last_bits;
 914				tile_w = 128;
 915				pix_per_tile = 96;
 916			} else {
 917				switch (base_format_mod) {
 918				case DRM_FORMAT_MOD_BROADCOM_SAND64:
 919					tile_w = 64;
 920					break;
 921				case DRM_FORMAT_MOD_BROADCOM_SAND128:
 922					tile_w = 128;
 923					break;
 924				case DRM_FORMAT_MOD_BROADCOM_SAND256:
 925					tile_w = 256;
 926					break;
 927				default:
 928					return -EINVAL;
 929				}
 930				pix_per_tile = tile_w / fb->format->cpp[0];
 931				x_off = (vc4_state->src_x % pix_per_tile) /
 932					(i ? h_subsample : 1) *
 933					fb->format->cpp[i];
 934			}
 935
 936			tile = vc4_state->src_x / pix_per_tile;
 937
 938			vc4_state->offsets[i] += param * tile_w * tile;
 939			vc4_state->offsets[i] += src_y /
 940						 (i ? v_subsample : 1) *
 941						 tile_w;
 942			vc4_state->offsets[i] += x_off & ~(i ? 1 : 0);
 943		}
 944
 945		pitch0 = VC4_SET_FIELD(param, SCALER_TILE_HEIGHT);
 946		break;
 947	}
 948
 949	default:
 950		DRM_DEBUG_KMS("Unsupported FB tiling flag 0x%16llx",
 951			      (long long)fb->modifier);
 952		return -EINVAL;
 953	}
 954
 955	/* Don't waste cycles mixing with plane alpha if the set alpha
 956	 * is opaque or there is no per-pixel alpha information.
 957	 * In any case we use the alpha property value as the fixed alpha.
 958	 */
 959	mix_plane_alpha = state->alpha != DRM_BLEND_ALPHA_OPAQUE &&
 960			  fb->format->has_alpha;
 961
 962	if (!vc4->is_vc5) {
 963	/* Control word */
 964		vc4_dlist_write(vc4_state,
 965				SCALER_CTL0_VALID |
 966				(rotation & DRM_MODE_REFLECT_X ? SCALER_CTL0_HFLIP : 0) |
 967				(rotation & DRM_MODE_REFLECT_Y ? SCALER_CTL0_VFLIP : 0) |
 968				VC4_SET_FIELD(SCALER_CTL0_RGBA_EXPAND_ROUND, SCALER_CTL0_RGBA_EXPAND) |
 969				(format->pixel_order << SCALER_CTL0_ORDER_SHIFT) |
 970				(hvs_format << SCALER_CTL0_PIXEL_FORMAT_SHIFT) |
 971				VC4_SET_FIELD(tiling, SCALER_CTL0_TILING) |
 972				(vc4_state->is_unity ? SCALER_CTL0_UNITY : 0) |
 973				VC4_SET_FIELD(scl0, SCALER_CTL0_SCL0) |
 974				VC4_SET_FIELD(scl1, SCALER_CTL0_SCL1));
 975
 976		/* Position Word 0: Image Positions and Alpha Value */
 977		vc4_state->pos0_offset = vc4_state->dlist_count;
 978		vc4_dlist_write(vc4_state,
 979				VC4_SET_FIELD(state->alpha >> 8, SCALER_POS0_FIXED_ALPHA) |
 980				VC4_SET_FIELD(vc4_state->crtc_x, SCALER_POS0_START_X) |
 981				VC4_SET_FIELD(vc4_state->crtc_y, SCALER_POS0_START_Y));
 982
 983		/* Position Word 1: Scaled Image Dimensions. */
 984		if (!vc4_state->is_unity) {
 985			vc4_dlist_write(vc4_state,
 986					VC4_SET_FIELD(vc4_state->crtc_w,
 987						      SCALER_POS1_SCL_WIDTH) |
 988					VC4_SET_FIELD(vc4_state->crtc_h,
 989						      SCALER_POS1_SCL_HEIGHT));
 990		}
 991
 992		/* Position Word 2: Source Image Size, Alpha */
 993		vc4_state->pos2_offset = vc4_state->dlist_count;
 994		vc4_dlist_write(vc4_state,
 995				(mix_plane_alpha ? SCALER_POS2_ALPHA_MIX : 0) |
 996				vc4_hvs4_get_alpha_blend_mode(state) |
 997				VC4_SET_FIELD(vc4_state->src_w[0],
 998					      SCALER_POS2_WIDTH) |
 999				VC4_SET_FIELD(vc4_state->src_h[0],
1000					      SCALER_POS2_HEIGHT));
1001
1002		/* Position Word 3: Context.  Written by the HVS. */
1003		vc4_dlist_write(vc4_state, 0xc0c0c0c0);
1004
1005	} else {
1006		u32 hvs_pixel_order = format->pixel_order;
1007
1008		if (format->pixel_order_hvs5)
1009			hvs_pixel_order = format->pixel_order_hvs5;
1010
1011		/* Control word */
1012		vc4_dlist_write(vc4_state,
1013				SCALER_CTL0_VALID |
1014				(hvs_pixel_order << SCALER_CTL0_ORDER_SHIFT) |
1015				(hvs_format << SCALER_CTL0_PIXEL_FORMAT_SHIFT) |
1016				VC4_SET_FIELD(tiling, SCALER_CTL0_TILING) |
1017				(vc4_state->is_unity ?
1018						SCALER5_CTL0_UNITY : 0) |
1019				VC4_SET_FIELD(scl0, SCALER_CTL0_SCL0) |
1020				VC4_SET_FIELD(scl1, SCALER_CTL0_SCL1) |
1021				SCALER5_CTL0_ALPHA_EXPAND |
1022				SCALER5_CTL0_RGB_EXPAND);
1023
1024		/* Position Word 0: Image Positions and Alpha Value */
1025		vc4_state->pos0_offset = vc4_state->dlist_count;
1026		vc4_dlist_write(vc4_state,
1027				(rotation & DRM_MODE_REFLECT_Y ?
1028						SCALER5_POS0_VFLIP : 0) |
1029				VC4_SET_FIELD(vc4_state->crtc_x,
1030					      SCALER_POS0_START_X) |
1031				(rotation & DRM_MODE_REFLECT_X ?
1032					      SCALER5_POS0_HFLIP : 0) |
1033				VC4_SET_FIELD(vc4_state->crtc_y,
1034					      SCALER5_POS0_START_Y)
1035			       );
1036
1037		/* Control Word 2 */
1038		vc4_dlist_write(vc4_state,
1039				VC4_SET_FIELD(state->alpha >> 4,
1040					      SCALER5_CTL2_ALPHA) |
1041				vc4_hvs5_get_alpha_blend_mode(state) |
1042				(mix_plane_alpha ?
1043					SCALER5_CTL2_ALPHA_MIX : 0)
1044			       );
1045
1046		/* Position Word 1: Scaled Image Dimensions. */
1047		if (!vc4_state->is_unity) {
1048			vc4_dlist_write(vc4_state,
1049					VC4_SET_FIELD(vc4_state->crtc_w,
1050						      SCALER5_POS1_SCL_WIDTH) |
1051					VC4_SET_FIELD(vc4_state->crtc_h,
1052						      SCALER5_POS1_SCL_HEIGHT));
1053		}
1054
1055		/* Position Word 2: Source Image Size */
1056		vc4_state->pos2_offset = vc4_state->dlist_count;
1057		vc4_dlist_write(vc4_state,
1058				VC4_SET_FIELD(vc4_state->src_w[0],
1059					      SCALER5_POS2_WIDTH) |
1060				VC4_SET_FIELD(vc4_state->src_h[0],
1061					      SCALER5_POS2_HEIGHT));
1062
1063		/* Position Word 3: Context.  Written by the HVS. */
1064		vc4_dlist_write(vc4_state, 0xc0c0c0c0);
1065	}
1066
1067
1068	/* Pointer Word 0/1/2: RGB / Y / Cb / Cr Pointers
1069	 *
1070	 * The pointers may be any byte address.
1071	 */
1072	vc4_state->ptr0_offset = vc4_state->dlist_count;
1073	for (i = 0; i < num_planes; i++)
1074		vc4_dlist_write(vc4_state, vc4_state->offsets[i]);
1075
1076	/* Pointer Context Word 0/1/2: Written by the HVS */
1077	for (i = 0; i < num_planes; i++)
1078		vc4_dlist_write(vc4_state, 0xc0c0c0c0);
1079
1080	/* Pitch word 0 */
1081	vc4_dlist_write(vc4_state, pitch0);
1082
1083	/* Pitch word 1/2 */
1084	for (i = 1; i < num_planes; i++) {
1085		if (hvs_format != HVS_PIXEL_FORMAT_H264 &&
1086		    hvs_format != HVS_PIXEL_FORMAT_YCBCR_10BIT) {
1087			vc4_dlist_write(vc4_state,
1088					VC4_SET_FIELD(fb->pitches[i],
1089						      SCALER_SRC_PITCH));
1090		} else {
1091			vc4_dlist_write(vc4_state, pitch0);
1092		}
1093	}
1094
1095	/* Colorspace conversion words */
1096	if (vc4_state->is_yuv) {
1097		enum drm_color_encoding color_encoding = state->color_encoding;
1098		enum drm_color_range color_range = state->color_range;
1099		const u32 *ccm;
1100
1101		if (color_encoding >= DRM_COLOR_ENCODING_MAX)
1102			color_encoding = DRM_COLOR_YCBCR_BT601;
1103		if (color_range >= DRM_COLOR_RANGE_MAX)
1104			color_range = DRM_COLOR_YCBCR_LIMITED_RANGE;
1105
1106		ccm = colorspace_coeffs[color_range][color_encoding];
1107
1108		vc4_dlist_write(vc4_state, ccm[0]);
1109		vc4_dlist_write(vc4_state, ccm[1]);
1110		vc4_dlist_write(vc4_state, ccm[2]);
1111	}
1112
1113	vc4_state->lbm_offset = 0;
1114
1115	if (vc4_state->x_scaling[0] != VC4_SCALING_NONE ||
1116	    vc4_state->x_scaling[1] != VC4_SCALING_NONE ||
1117	    vc4_state->y_scaling[0] != VC4_SCALING_NONE ||
1118	    vc4_state->y_scaling[1] != VC4_SCALING_NONE) {
1119		/* Reserve a slot for the LBM Base Address. The real value will
1120		 * be set when calling vc4_plane_allocate_lbm().
1121		 */
1122		if (vc4_state->y_scaling[0] != VC4_SCALING_NONE ||
1123		    vc4_state->y_scaling[1] != VC4_SCALING_NONE) {
1124			vc4_state->lbm_offset = vc4_state->dlist_count;
1125			vc4_dlist_counter_increment(vc4_state);
1126		}
1127
1128		if (num_planes > 1) {
1129			/* Emit Cb/Cr as channel 0 and Y as channel
1130			 * 1. This matches how we set up scl0/scl1
1131			 * above.
1132			 */
1133			vc4_write_scaling_parameters(state, 1);
1134		}
1135		vc4_write_scaling_parameters(state, 0);
1136
1137		/* If any PPF setup was done, then all the kernel
1138		 * pointers get uploaded.
1139		 */
1140		if (vc4_state->x_scaling[0] == VC4_SCALING_PPF ||
1141		    vc4_state->y_scaling[0] == VC4_SCALING_PPF ||
1142		    vc4_state->x_scaling[1] == VC4_SCALING_PPF ||
1143		    vc4_state->y_scaling[1] == VC4_SCALING_PPF) {
1144			u32 kernel = VC4_SET_FIELD(vc4->hvs->mitchell_netravali_filter.start,
1145						   SCALER_PPF_KERNEL_OFFSET);
1146
1147			/* HPPF plane 0 */
1148			vc4_dlist_write(vc4_state, kernel);
1149			/* VPPF plane 0 */
1150			vc4_dlist_write(vc4_state, kernel);
1151			/* HPPF plane 1 */
1152			vc4_dlist_write(vc4_state, kernel);
1153			/* VPPF plane 1 */
1154			vc4_dlist_write(vc4_state, kernel);
1155		}
1156	}
1157
1158	vc4_state->dlist[ctl0_offset] |=
1159		VC4_SET_FIELD(vc4_state->dlist_count, SCALER_CTL0_SIZE);
1160
1161	/* crtc_* are already clipped coordinates. */
1162	covers_screen = vc4_state->crtc_x == 0 && vc4_state->crtc_y == 0 &&
1163			vc4_state->crtc_w == state->crtc->mode.hdisplay &&
1164			vc4_state->crtc_h == state->crtc->mode.vdisplay;
1165	/* Background fill might be necessary when the plane has per-pixel
1166	 * alpha content or a non-opaque plane alpha and could blend from the
1167	 * background or does not cover the entire screen.
1168	 */
1169	vc4_state->needs_bg_fill = fb->format->has_alpha || !covers_screen ||
1170				   state->alpha != DRM_BLEND_ALPHA_OPAQUE;
1171
1172	/* Flag the dlist as initialized to avoid checking it twice in case
1173	 * the async update check already called vc4_plane_mode_set() and
1174	 * decided to fallback to sync update because async update was not
1175	 * possible.
1176	 */
1177	vc4_state->dlist_initialized = 1;
1178
1179	vc4_plane_calc_load(state);
1180
1181	return 0;
1182}
1183
1184/* If a modeset involves changing the setup of a plane, the atomic
1185 * infrastructure will call this to validate a proposed plane setup.
1186 * However, if a plane isn't getting updated, this (and the
1187 * corresponding vc4_plane_atomic_update) won't get called.  Thus, we
1188 * compute the dlist here and have all active plane dlists get updated
1189 * in the CRTC's flush.
1190 */
1191static int vc4_plane_atomic_check(struct drm_plane *plane,
1192				  struct drm_atomic_state *state)
1193{
1194	struct drm_plane_state *new_plane_state = drm_atomic_get_new_plane_state(state,
1195										 plane);
1196	struct vc4_plane_state *vc4_state = to_vc4_plane_state(new_plane_state);
1197	int ret;
1198
1199	vc4_state->dlist_count = 0;
1200
1201	if (!plane_enabled(new_plane_state))
1202		return 0;
1203
1204	ret = vc4_plane_mode_set(plane, new_plane_state);
1205	if (ret)
1206		return ret;
1207
1208	return vc4_plane_allocate_lbm(new_plane_state);
1209}
1210
1211static void vc4_plane_atomic_update(struct drm_plane *plane,
1212				    struct drm_atomic_state *state)
1213{
1214	/* No contents here.  Since we don't know where in the CRTC's
1215	 * dlist we should be stored, our dlist is uploaded to the
1216	 * hardware with vc4_plane_write_dlist() at CRTC atomic_flush
1217	 * time.
1218	 */
1219}
1220
1221u32 vc4_plane_write_dlist(struct drm_plane *plane, u32 __iomem *dlist)
1222{
1223	struct vc4_plane_state *vc4_state = to_vc4_plane_state(plane->state);
1224	int i;
1225	int idx;
1226
1227	if (!drm_dev_enter(plane->dev, &idx))
1228		goto out;
1229
1230	vc4_state->hw_dlist = dlist;
1231
1232	/* Can't memcpy_toio() because it needs to be 32-bit writes. */
1233	for (i = 0; i < vc4_state->dlist_count; i++)
1234		writel(vc4_state->dlist[i], &dlist[i]);
1235
1236	drm_dev_exit(idx);
1237
1238out:
1239	return vc4_state->dlist_count;
1240}
1241
1242u32 vc4_plane_dlist_size(const struct drm_plane_state *state)
1243{
1244	const struct vc4_plane_state *vc4_state =
1245		container_of(state, typeof(*vc4_state), base);
1246
1247	return vc4_state->dlist_count;
1248}
1249
1250/* Updates the plane to immediately (well, once the FIFO needs
1251 * refilling) scan out from at a new framebuffer.
1252 */
1253void vc4_plane_async_set_fb(struct drm_plane *plane, struct drm_framebuffer *fb)
1254{
1255	struct vc4_plane_state *vc4_state = to_vc4_plane_state(plane->state);
1256	struct drm_gem_dma_object *bo = drm_fb_dma_get_gem_obj(fb, 0);
1257	uint32_t addr;
1258	int idx;
1259
1260	if (!drm_dev_enter(plane->dev, &idx))
1261		return;
1262
1263	/* We're skipping the address adjustment for negative origin,
1264	 * because this is only called on the primary plane.
1265	 */
1266	WARN_ON_ONCE(plane->state->crtc_x < 0 || plane->state->crtc_y < 0);
1267	addr = bo->dma_addr + fb->offsets[0];
1268
1269	/* Write the new address into the hardware immediately.  The
1270	 * scanout will start from this address as soon as the FIFO
1271	 * needs to refill with pixels.
1272	 */
1273	writel(addr, &vc4_state->hw_dlist[vc4_state->ptr0_offset]);
1274
1275	/* Also update the CPU-side dlist copy, so that any later
1276	 * atomic updates that don't do a new modeset on our plane
1277	 * also use our updated address.
1278	 */
1279	vc4_state->dlist[vc4_state->ptr0_offset] = addr;
1280
1281	drm_dev_exit(idx);
1282}
1283
1284static void vc4_plane_atomic_async_update(struct drm_plane *plane,
1285					  struct drm_atomic_state *state)
1286{
1287	struct drm_plane_state *new_plane_state = drm_atomic_get_new_plane_state(state,
1288										 plane);
1289	struct vc4_plane_state *vc4_state, *new_vc4_state;
1290	int idx;
1291
1292	if (!drm_dev_enter(plane->dev, &idx))
1293		return;
1294
1295	swap(plane->state->fb, new_plane_state->fb);
1296	plane->state->crtc_x = new_plane_state->crtc_x;
1297	plane->state->crtc_y = new_plane_state->crtc_y;
1298	plane->state->crtc_w = new_plane_state->crtc_w;
1299	plane->state->crtc_h = new_plane_state->crtc_h;
1300	plane->state->src_x = new_plane_state->src_x;
1301	plane->state->src_y = new_plane_state->src_y;
1302	plane->state->src_w = new_plane_state->src_w;
1303	plane->state->src_h = new_plane_state->src_h;
1304	plane->state->alpha = new_plane_state->alpha;
1305	plane->state->pixel_blend_mode = new_plane_state->pixel_blend_mode;
1306	plane->state->rotation = new_plane_state->rotation;
1307	plane->state->zpos = new_plane_state->zpos;
1308	plane->state->normalized_zpos = new_plane_state->normalized_zpos;
1309	plane->state->color_encoding = new_plane_state->color_encoding;
1310	plane->state->color_range = new_plane_state->color_range;
1311	plane->state->src = new_plane_state->src;
1312	plane->state->dst = new_plane_state->dst;
1313	plane->state->visible = new_plane_state->visible;
1314
1315	new_vc4_state = to_vc4_plane_state(new_plane_state);
1316	vc4_state = to_vc4_plane_state(plane->state);
1317
1318	vc4_state->crtc_x = new_vc4_state->crtc_x;
1319	vc4_state->crtc_y = new_vc4_state->crtc_y;
1320	vc4_state->crtc_h = new_vc4_state->crtc_h;
1321	vc4_state->crtc_w = new_vc4_state->crtc_w;
1322	vc4_state->src_x = new_vc4_state->src_x;
1323	vc4_state->src_y = new_vc4_state->src_y;
1324	memcpy(vc4_state->src_w, new_vc4_state->src_w,
1325	       sizeof(vc4_state->src_w));
1326	memcpy(vc4_state->src_h, new_vc4_state->src_h,
1327	       sizeof(vc4_state->src_h));
1328	memcpy(vc4_state->x_scaling, new_vc4_state->x_scaling,
1329	       sizeof(vc4_state->x_scaling));
1330	memcpy(vc4_state->y_scaling, new_vc4_state->y_scaling,
1331	       sizeof(vc4_state->y_scaling));
1332	vc4_state->is_unity = new_vc4_state->is_unity;
1333	vc4_state->is_yuv = new_vc4_state->is_yuv;
1334	memcpy(vc4_state->offsets, new_vc4_state->offsets,
1335	       sizeof(vc4_state->offsets));
1336	vc4_state->needs_bg_fill = new_vc4_state->needs_bg_fill;
1337
1338	/* Update the current vc4_state pos0, pos2 and ptr0 dlist entries. */
1339	vc4_state->dlist[vc4_state->pos0_offset] =
1340		new_vc4_state->dlist[vc4_state->pos0_offset];
1341	vc4_state->dlist[vc4_state->pos2_offset] =
1342		new_vc4_state->dlist[vc4_state->pos2_offset];
1343	vc4_state->dlist[vc4_state->ptr0_offset] =
1344		new_vc4_state->dlist[vc4_state->ptr0_offset];
1345
1346	/* Note that we can't just call vc4_plane_write_dlist()
1347	 * because that would smash the context data that the HVS is
1348	 * currently using.
1349	 */
1350	writel(vc4_state->dlist[vc4_state->pos0_offset],
1351	       &vc4_state->hw_dlist[vc4_state->pos0_offset]);
1352	writel(vc4_state->dlist[vc4_state->pos2_offset],
1353	       &vc4_state->hw_dlist[vc4_state->pos2_offset]);
1354	writel(vc4_state->dlist[vc4_state->ptr0_offset],
1355	       &vc4_state->hw_dlist[vc4_state->ptr0_offset]);
1356
1357	drm_dev_exit(idx);
1358}
1359
1360static int vc4_plane_atomic_async_check(struct drm_plane *plane,
1361					struct drm_atomic_state *state)
1362{
1363	struct drm_plane_state *new_plane_state = drm_atomic_get_new_plane_state(state,
1364										 plane);
1365	struct vc4_plane_state *old_vc4_state, *new_vc4_state;
1366	int ret;
1367	u32 i;
1368
1369	ret = vc4_plane_mode_set(plane, new_plane_state);
1370	if (ret)
1371		return ret;
1372
1373	old_vc4_state = to_vc4_plane_state(plane->state);
1374	new_vc4_state = to_vc4_plane_state(new_plane_state);
1375
1376	if (!new_vc4_state->hw_dlist)
1377		return -EINVAL;
1378
1379	if (old_vc4_state->dlist_count != new_vc4_state->dlist_count ||
1380	    old_vc4_state->pos0_offset != new_vc4_state->pos0_offset ||
1381	    old_vc4_state->pos2_offset != new_vc4_state->pos2_offset ||
1382	    old_vc4_state->ptr0_offset != new_vc4_state->ptr0_offset ||
1383	    vc4_lbm_size(plane->state) != vc4_lbm_size(new_plane_state))
1384		return -EINVAL;
1385
1386	/* Only pos0, pos2 and ptr0 DWORDS can be updated in an async update
1387	 * if anything else has changed, fallback to a sync update.
1388	 */
1389	for (i = 0; i < new_vc4_state->dlist_count; i++) {
1390		if (i == new_vc4_state->pos0_offset ||
1391		    i == new_vc4_state->pos2_offset ||
1392		    i == new_vc4_state->ptr0_offset ||
1393		    (new_vc4_state->lbm_offset &&
1394		     i == new_vc4_state->lbm_offset))
1395			continue;
1396
1397		if (new_vc4_state->dlist[i] != old_vc4_state->dlist[i])
1398			return -EINVAL;
1399	}
1400
1401	return 0;
1402}
1403
1404static int vc4_prepare_fb(struct drm_plane *plane,
1405			  struct drm_plane_state *state)
1406{
1407	struct vc4_bo *bo;
1408
1409	if (!state->fb)
1410		return 0;
1411
1412	bo = to_vc4_bo(&drm_fb_dma_get_gem_obj(state->fb, 0)->base);
1413
1414	drm_gem_plane_helper_prepare_fb(plane, state);
1415
1416	if (plane->state->fb == state->fb)
1417		return 0;
1418
1419	return vc4_bo_inc_usecnt(bo);
1420}
1421
1422static void vc4_cleanup_fb(struct drm_plane *plane,
1423			   struct drm_plane_state *state)
1424{
1425	struct vc4_bo *bo;
1426
1427	if (plane->state->fb == state->fb || !state->fb)
1428		return;
1429
1430	bo = to_vc4_bo(&drm_fb_dma_get_gem_obj(state->fb, 0)->base);
1431	vc4_bo_dec_usecnt(bo);
1432}
1433
1434static const struct drm_plane_helper_funcs vc4_plane_helper_funcs = {
1435	.atomic_check = vc4_plane_atomic_check,
1436	.atomic_update = vc4_plane_atomic_update,
1437	.prepare_fb = vc4_prepare_fb,
1438	.cleanup_fb = vc4_cleanup_fb,
1439	.atomic_async_check = vc4_plane_atomic_async_check,
1440	.atomic_async_update = vc4_plane_atomic_async_update,
1441};
1442
1443static const struct drm_plane_helper_funcs vc5_plane_helper_funcs = {
1444	.atomic_check = vc4_plane_atomic_check,
1445	.atomic_update = vc4_plane_atomic_update,
1446	.atomic_async_check = vc4_plane_atomic_async_check,
1447	.atomic_async_update = vc4_plane_atomic_async_update,
1448};
1449
1450static bool vc4_format_mod_supported(struct drm_plane *plane,
1451				     uint32_t format,
1452				     uint64_t modifier)
1453{
1454	/* Support T_TILING for RGB formats only. */
1455	switch (format) {
1456	case DRM_FORMAT_XRGB8888:
1457	case DRM_FORMAT_ARGB8888:
1458	case DRM_FORMAT_ABGR8888:
1459	case DRM_FORMAT_XBGR8888:
1460	case DRM_FORMAT_RGB565:
1461	case DRM_FORMAT_BGR565:
1462	case DRM_FORMAT_ARGB1555:
1463	case DRM_FORMAT_XRGB1555:
1464		switch (fourcc_mod_broadcom_mod(modifier)) {
1465		case DRM_FORMAT_MOD_LINEAR:
1466		case DRM_FORMAT_MOD_BROADCOM_VC4_T_TILED:
1467			return true;
1468		default:
1469			return false;
1470		}
1471	case DRM_FORMAT_NV12:
1472	case DRM_FORMAT_NV21:
1473		switch (fourcc_mod_broadcom_mod(modifier)) {
1474		case DRM_FORMAT_MOD_LINEAR:
1475		case DRM_FORMAT_MOD_BROADCOM_SAND64:
1476		case DRM_FORMAT_MOD_BROADCOM_SAND128:
1477		case DRM_FORMAT_MOD_BROADCOM_SAND256:
1478			return true;
1479		default:
1480			return false;
1481		}
1482	case DRM_FORMAT_P030:
1483		switch (fourcc_mod_broadcom_mod(modifier)) {
1484		case DRM_FORMAT_MOD_BROADCOM_SAND128:
1485			return true;
1486		default:
1487			return false;
1488		}
1489	case DRM_FORMAT_RGBX1010102:
1490	case DRM_FORMAT_BGRX1010102:
1491	case DRM_FORMAT_RGBA1010102:
1492	case DRM_FORMAT_BGRA1010102:
1493	case DRM_FORMAT_YUV422:
1494	case DRM_FORMAT_YVU422:
1495	case DRM_FORMAT_YUV420:
1496	case DRM_FORMAT_YVU420:
1497	case DRM_FORMAT_NV16:
1498	case DRM_FORMAT_NV61:
1499	default:
1500		return (modifier == DRM_FORMAT_MOD_LINEAR);
1501	}
1502}
1503
1504static const struct drm_plane_funcs vc4_plane_funcs = {
1505	.update_plane = drm_atomic_helper_update_plane,
1506	.disable_plane = drm_atomic_helper_disable_plane,
1507	.reset = vc4_plane_reset,
1508	.atomic_duplicate_state = vc4_plane_duplicate_state,
1509	.atomic_destroy_state = vc4_plane_destroy_state,
1510	.format_mod_supported = vc4_format_mod_supported,
1511};
1512
1513struct drm_plane *vc4_plane_init(struct drm_device *dev,
1514				 enum drm_plane_type type,
1515				 uint32_t possible_crtcs)
1516{
1517	struct vc4_dev *vc4 = to_vc4_dev(dev);
1518	struct drm_plane *plane;
1519	struct vc4_plane *vc4_plane;
1520	u32 formats[ARRAY_SIZE(hvs_formats)];
1521	int num_formats = 0;
1522	unsigned i;
1523	static const uint64_t modifiers[] = {
1524		DRM_FORMAT_MOD_BROADCOM_VC4_T_TILED,
1525		DRM_FORMAT_MOD_BROADCOM_SAND128,
1526		DRM_FORMAT_MOD_BROADCOM_SAND64,
1527		DRM_FORMAT_MOD_BROADCOM_SAND256,
1528		DRM_FORMAT_MOD_LINEAR,
1529		DRM_FORMAT_MOD_INVALID
1530	};
1531
1532	for (i = 0; i < ARRAY_SIZE(hvs_formats); i++) {
1533		if (!hvs_formats[i].hvs5_only || vc4->is_vc5) {
1534			formats[num_formats] = hvs_formats[i].drm;
1535			num_formats++;
1536		}
1537	}
1538
1539	vc4_plane = drmm_universal_plane_alloc(dev, struct vc4_plane, base,
1540					       possible_crtcs,
1541					       &vc4_plane_funcs,
1542					       formats, num_formats,
1543					       modifiers, type, NULL);
1544	if (IS_ERR(vc4_plane))
1545		return ERR_CAST(vc4_plane);
1546	plane = &vc4_plane->base;
1547
1548	if (vc4->is_vc5)
1549		drm_plane_helper_add(plane, &vc5_plane_helper_funcs);
1550	else
1551		drm_plane_helper_add(plane, &vc4_plane_helper_funcs);
1552
1553	drm_plane_create_alpha_property(plane);
1554	drm_plane_create_blend_mode_property(plane,
1555					     BIT(DRM_MODE_BLEND_PIXEL_NONE) |
1556					     BIT(DRM_MODE_BLEND_PREMULTI) |
1557					     BIT(DRM_MODE_BLEND_COVERAGE));
1558	drm_plane_create_rotation_property(plane, DRM_MODE_ROTATE_0,
1559					   DRM_MODE_ROTATE_0 |
1560					   DRM_MODE_ROTATE_180 |
1561					   DRM_MODE_REFLECT_X |
1562					   DRM_MODE_REFLECT_Y);
1563
1564	drm_plane_create_color_properties(plane,
1565					  BIT(DRM_COLOR_YCBCR_BT601) |
1566					  BIT(DRM_COLOR_YCBCR_BT709) |
1567					  BIT(DRM_COLOR_YCBCR_BT2020),
1568					  BIT(DRM_COLOR_YCBCR_LIMITED_RANGE) |
1569					  BIT(DRM_COLOR_YCBCR_FULL_RANGE),
1570					  DRM_COLOR_YCBCR_BT709,
1571					  DRM_COLOR_YCBCR_LIMITED_RANGE);
1572
1573	return plane;
1574}
1575
1576int vc4_plane_create_additional_planes(struct drm_device *drm)
1577{
1578	struct drm_plane *cursor_plane;
1579	struct drm_crtc *crtc;
1580	unsigned int i;
1581
1582	/* Set up some arbitrary number of planes.  We're not limited
1583	 * by a set number of physical registers, just the space in
1584	 * the HVS (16k) and how small an plane can be (28 bytes).
1585	 * However, each plane we set up takes up some memory, and
1586	 * increases the cost of looping over planes, which atomic
1587	 * modesetting does quite a bit.  As a result, we pick a
1588	 * modest number of planes to expose, that should hopefully
1589	 * still cover any sane usecase.
1590	 */
1591	for (i = 0; i < 16; i++) {
1592		struct drm_plane *plane =
1593			vc4_plane_init(drm, DRM_PLANE_TYPE_OVERLAY,
1594				       GENMASK(drm->mode_config.num_crtc - 1, 0));
1595
1596		if (IS_ERR(plane))
1597			continue;
1598	}
1599
1600	drm_for_each_crtc(crtc, drm) {
1601		/* Set up the legacy cursor after overlay initialization,
1602		 * since we overlay planes on the CRTC in the order they were
1603		 * initialized.
1604		 */
1605		cursor_plane = vc4_plane_init(drm, DRM_PLANE_TYPE_CURSOR,
1606					      drm_crtc_mask(crtc));
1607		if (!IS_ERR(cursor_plane)) {
1608			crtc->cursor = cursor_plane;
1609		}
1610	}
1611
1612	return 0;
1613}