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