xref: /openbmc/linux/drivers/gpu/drm/vc4/vc4_plane.c (revision 94d964e5)
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 
31 static const struct hvs_format {
32 	u32 drm; /* DRM_FORMAT_* */
33 	u32 hvs; /* HVS_FORMAT_* */
34 	u32 pixel_order;
35 	u32 pixel_order_hvs5;
36 	bool hvs5_only;
37 } hvs_formats[] = {
38 	{
39 		.drm = DRM_FORMAT_XRGB8888,
40 		.hvs = HVS_PIXEL_FORMAT_RGBA8888,
41 		.pixel_order = HVS_PIXEL_ORDER_ABGR,
42 		.pixel_order_hvs5 = HVS_PIXEL_ORDER_ARGB,
43 	},
44 	{
45 		.drm = DRM_FORMAT_ARGB8888,
46 		.hvs = HVS_PIXEL_FORMAT_RGBA8888,
47 		.pixel_order = HVS_PIXEL_ORDER_ABGR,
48 		.pixel_order_hvs5 = HVS_PIXEL_ORDER_ARGB,
49 	},
50 	{
51 		.drm = DRM_FORMAT_ABGR8888,
52 		.hvs = HVS_PIXEL_FORMAT_RGBA8888,
53 		.pixel_order = HVS_PIXEL_ORDER_ARGB,
54 		.pixel_order_hvs5 = HVS_PIXEL_ORDER_ABGR,
55 	},
56 	{
57 		.drm = DRM_FORMAT_XBGR8888,
58 		.hvs = HVS_PIXEL_FORMAT_RGBA8888,
59 		.pixel_order = HVS_PIXEL_ORDER_ARGB,
60 		.pixel_order_hvs5 = HVS_PIXEL_ORDER_ABGR,
61 	},
62 	{
63 		.drm = DRM_FORMAT_RGB565,
64 		.hvs = HVS_PIXEL_FORMAT_RGB565,
65 		.pixel_order = HVS_PIXEL_ORDER_XRGB,
66 	},
67 	{
68 		.drm = DRM_FORMAT_BGR565,
69 		.hvs = HVS_PIXEL_FORMAT_RGB565,
70 		.pixel_order = HVS_PIXEL_ORDER_XBGR,
71 	},
72 	{
73 		.drm = DRM_FORMAT_ARGB1555,
74 		.hvs = HVS_PIXEL_FORMAT_RGBA5551,
75 		.pixel_order = HVS_PIXEL_ORDER_ABGR,
76 	},
77 	{
78 		.drm = DRM_FORMAT_XRGB1555,
79 		.hvs = HVS_PIXEL_FORMAT_RGBA5551,
80 		.pixel_order = HVS_PIXEL_ORDER_ABGR,
81 	},
82 	{
83 		.drm = DRM_FORMAT_RGB888,
84 		.hvs = HVS_PIXEL_FORMAT_RGB888,
85 		.pixel_order = HVS_PIXEL_ORDER_XRGB,
86 	},
87 	{
88 		.drm = DRM_FORMAT_BGR888,
89 		.hvs = HVS_PIXEL_FORMAT_RGB888,
90 		.pixel_order = HVS_PIXEL_ORDER_XBGR,
91 	},
92 	{
93 		.drm = DRM_FORMAT_YUV422,
94 		.hvs = HVS_PIXEL_FORMAT_YCBCR_YUV422_3PLANE,
95 		.pixel_order = HVS_PIXEL_ORDER_XYCBCR,
96 	},
97 	{
98 		.drm = DRM_FORMAT_YVU422,
99 		.hvs = HVS_PIXEL_FORMAT_YCBCR_YUV422_3PLANE,
100 		.pixel_order = HVS_PIXEL_ORDER_XYCRCB,
101 	},
102 	{
103 		.drm = DRM_FORMAT_YUV420,
104 		.hvs = HVS_PIXEL_FORMAT_YCBCR_YUV420_3PLANE,
105 		.pixel_order = HVS_PIXEL_ORDER_XYCBCR,
106 	},
107 	{
108 		.drm = DRM_FORMAT_YVU420,
109 		.hvs = HVS_PIXEL_FORMAT_YCBCR_YUV420_3PLANE,
110 		.pixel_order = HVS_PIXEL_ORDER_XYCRCB,
111 	},
112 	{
113 		.drm = DRM_FORMAT_NV12,
114 		.hvs = HVS_PIXEL_FORMAT_YCBCR_YUV420_2PLANE,
115 		.pixel_order = HVS_PIXEL_ORDER_XYCBCR,
116 	},
117 	{
118 		.drm = DRM_FORMAT_NV21,
119 		.hvs = HVS_PIXEL_FORMAT_YCBCR_YUV420_2PLANE,
120 		.pixel_order = HVS_PIXEL_ORDER_XYCRCB,
121 	},
122 	{
123 		.drm = DRM_FORMAT_NV16,
124 		.hvs = HVS_PIXEL_FORMAT_YCBCR_YUV422_2PLANE,
125 		.pixel_order = HVS_PIXEL_ORDER_XYCBCR,
126 	},
127 	{
128 		.drm = DRM_FORMAT_NV61,
129 		.hvs = HVS_PIXEL_FORMAT_YCBCR_YUV422_2PLANE,
130 		.pixel_order = HVS_PIXEL_ORDER_XYCRCB,
131 	},
132 	{
133 		.drm = DRM_FORMAT_P030,
134 		.hvs = HVS_PIXEL_FORMAT_YCBCR_10BIT,
135 		.pixel_order = HVS_PIXEL_ORDER_XYCBCR,
136 		.hvs5_only = true,
137 	},
138 };
139 
140 static const struct hvs_format *vc4_get_hvs_format(u32 drm_format)
141 {
142 	unsigned i;
143 
144 	for (i = 0; i < ARRAY_SIZE(hvs_formats); i++) {
145 		if (hvs_formats[i].drm == drm_format)
146 			return &hvs_formats[i];
147 	}
148 
149 	return NULL;
150 }
151 
152 static enum vc4_scaling_mode vc4_get_scaling_mode(u32 src, u32 dst)
153 {
154 	if (dst == src)
155 		return VC4_SCALING_NONE;
156 	if (3 * dst >= 2 * src)
157 		return VC4_SCALING_PPF;
158 	else
159 		return VC4_SCALING_TPZ;
160 }
161 
162 static bool plane_enabled(struct drm_plane_state *state)
163 {
164 	return state->fb && !WARN_ON(!state->crtc);
165 }
166 
167 static struct drm_plane_state *vc4_plane_duplicate_state(struct drm_plane *plane)
168 {
169 	struct vc4_plane_state *vc4_state;
170 
171 	if (WARN_ON(!plane->state))
172 		return NULL;
173 
174 	vc4_state = kmemdup(plane->state, sizeof(*vc4_state), GFP_KERNEL);
175 	if (!vc4_state)
176 		return NULL;
177 
178 	memset(&vc4_state->lbm, 0, sizeof(vc4_state->lbm));
179 	vc4_state->dlist_initialized = 0;
180 
181 	__drm_atomic_helper_plane_duplicate_state(plane, &vc4_state->base);
182 
183 	if (vc4_state->dlist) {
184 		vc4_state->dlist = kmemdup(vc4_state->dlist,
185 					   vc4_state->dlist_count * 4,
186 					   GFP_KERNEL);
187 		if (!vc4_state->dlist) {
188 			kfree(vc4_state);
189 			return NULL;
190 		}
191 		vc4_state->dlist_size = vc4_state->dlist_count;
192 	}
193 
194 	return &vc4_state->base;
195 }
196 
197 static void vc4_plane_destroy_state(struct drm_plane *plane,
198 				    struct drm_plane_state *state)
199 {
200 	struct vc4_dev *vc4 = to_vc4_dev(plane->dev);
201 	struct vc4_plane_state *vc4_state = to_vc4_plane_state(state);
202 
203 	if (drm_mm_node_allocated(&vc4_state->lbm)) {
204 		unsigned long irqflags;
205 
206 		spin_lock_irqsave(&vc4->hvs->mm_lock, irqflags);
207 		drm_mm_remove_node(&vc4_state->lbm);
208 		spin_unlock_irqrestore(&vc4->hvs->mm_lock, irqflags);
209 	}
210 
211 	kfree(vc4_state->dlist);
212 	__drm_atomic_helper_plane_destroy_state(&vc4_state->base);
213 	kfree(state);
214 }
215 
216 /* Called during init to allocate the plane's atomic state. */
217 static void vc4_plane_reset(struct drm_plane *plane)
218 {
219 	struct vc4_plane_state *vc4_state;
220 
221 	WARN_ON(plane->state);
222 
223 	vc4_state = kzalloc(sizeof(*vc4_state), GFP_KERNEL);
224 	if (!vc4_state)
225 		return;
226 
227 	__drm_atomic_helper_plane_reset(plane, &vc4_state->base);
228 }
229 
230 static void vc4_dlist_counter_increment(struct vc4_plane_state *vc4_state)
231 {
232 	if (vc4_state->dlist_count == vc4_state->dlist_size) {
233 		u32 new_size = max(4u, vc4_state->dlist_count * 2);
234 		u32 *new_dlist = kmalloc_array(new_size, 4, GFP_KERNEL);
235 
236 		if (!new_dlist)
237 			return;
238 		memcpy(new_dlist, vc4_state->dlist, vc4_state->dlist_count * 4);
239 
240 		kfree(vc4_state->dlist);
241 		vc4_state->dlist = new_dlist;
242 		vc4_state->dlist_size = new_size;
243 	}
244 
245 	vc4_state->dlist_count++;
246 }
247 
248 static void vc4_dlist_write(struct vc4_plane_state *vc4_state, u32 val)
249 {
250 	unsigned int idx = vc4_state->dlist_count;
251 
252 	vc4_dlist_counter_increment(vc4_state);
253 	vc4_state->dlist[idx] = val;
254 }
255 
256 /* Returns the scl0/scl1 field based on whether the dimensions need to
257  * be up/down/non-scaled.
258  *
259  * This is a replication of a table from the spec.
260  */
261 static u32 vc4_get_scl_field(struct drm_plane_state *state, int plane)
262 {
263 	struct vc4_plane_state *vc4_state = to_vc4_plane_state(state);
264 
265 	switch (vc4_state->x_scaling[plane] << 2 | vc4_state->y_scaling[plane]) {
266 	case VC4_SCALING_PPF << 2 | VC4_SCALING_PPF:
267 		return SCALER_CTL0_SCL_H_PPF_V_PPF;
268 	case VC4_SCALING_TPZ << 2 | VC4_SCALING_PPF:
269 		return SCALER_CTL0_SCL_H_TPZ_V_PPF;
270 	case VC4_SCALING_PPF << 2 | VC4_SCALING_TPZ:
271 		return SCALER_CTL0_SCL_H_PPF_V_TPZ;
272 	case VC4_SCALING_TPZ << 2 | VC4_SCALING_TPZ:
273 		return SCALER_CTL0_SCL_H_TPZ_V_TPZ;
274 	case VC4_SCALING_PPF << 2 | VC4_SCALING_NONE:
275 		return SCALER_CTL0_SCL_H_PPF_V_NONE;
276 	case VC4_SCALING_NONE << 2 | VC4_SCALING_PPF:
277 		return SCALER_CTL0_SCL_H_NONE_V_PPF;
278 	case VC4_SCALING_NONE << 2 | VC4_SCALING_TPZ:
279 		return SCALER_CTL0_SCL_H_NONE_V_TPZ;
280 	case VC4_SCALING_TPZ << 2 | VC4_SCALING_NONE:
281 		return SCALER_CTL0_SCL_H_TPZ_V_NONE;
282 	default:
283 	case VC4_SCALING_NONE << 2 | VC4_SCALING_NONE:
284 		/* The unity case is independently handled by
285 		 * SCALER_CTL0_UNITY.
286 		 */
287 		return 0;
288 	}
289 }
290 
291 static int vc4_plane_margins_adj(struct drm_plane_state *pstate)
292 {
293 	struct vc4_plane_state *vc4_pstate = to_vc4_plane_state(pstate);
294 	unsigned int left, right, top, bottom, adjhdisplay, adjvdisplay;
295 	struct drm_crtc_state *crtc_state;
296 
297 	crtc_state = drm_atomic_get_new_crtc_state(pstate->state,
298 						   pstate->crtc);
299 
300 	vc4_crtc_get_margins(crtc_state, &left, &right, &top, &bottom);
301 	if (!left && !right && !top && !bottom)
302 		return 0;
303 
304 	if (left + right >= crtc_state->mode.hdisplay ||
305 	    top + bottom >= crtc_state->mode.vdisplay)
306 		return -EINVAL;
307 
308 	adjhdisplay = crtc_state->mode.hdisplay - (left + right);
309 	vc4_pstate->crtc_x = DIV_ROUND_CLOSEST(vc4_pstate->crtc_x *
310 					       adjhdisplay,
311 					       crtc_state->mode.hdisplay);
312 	vc4_pstate->crtc_x += left;
313 	if (vc4_pstate->crtc_x > crtc_state->mode.hdisplay - left)
314 		vc4_pstate->crtc_x = crtc_state->mode.hdisplay - left;
315 
316 	adjvdisplay = crtc_state->mode.vdisplay - (top + bottom);
317 	vc4_pstate->crtc_y = DIV_ROUND_CLOSEST(vc4_pstate->crtc_y *
318 					       adjvdisplay,
319 					       crtc_state->mode.vdisplay);
320 	vc4_pstate->crtc_y += top;
321 	if (vc4_pstate->crtc_y > crtc_state->mode.vdisplay - top)
322 		vc4_pstate->crtc_y = crtc_state->mode.vdisplay - top;
323 
324 	vc4_pstate->crtc_w = DIV_ROUND_CLOSEST(vc4_pstate->crtc_w *
325 					       adjhdisplay,
326 					       crtc_state->mode.hdisplay);
327 	vc4_pstate->crtc_h = DIV_ROUND_CLOSEST(vc4_pstate->crtc_h *
328 					       adjvdisplay,
329 					       crtc_state->mode.vdisplay);
330 
331 	if (!vc4_pstate->crtc_w || !vc4_pstate->crtc_h)
332 		return -EINVAL;
333 
334 	return 0;
335 }
336 
337 static int vc4_plane_setup_clipping_and_scaling(struct drm_plane_state *state)
338 {
339 	struct vc4_plane_state *vc4_state = to_vc4_plane_state(state);
340 	struct drm_framebuffer *fb = state->fb;
341 	struct drm_gem_cma_object *bo = drm_fb_cma_get_gem_obj(fb, 0);
342 	u32 subpixel_src_mask = (1 << 16) - 1;
343 	int num_planes = fb->format->num_planes;
344 	struct drm_crtc_state *crtc_state;
345 	u32 h_subsample = fb->format->hsub;
346 	u32 v_subsample = fb->format->vsub;
347 	int i, ret;
348 
349 	crtc_state = drm_atomic_get_existing_crtc_state(state->state,
350 							state->crtc);
351 	if (!crtc_state) {
352 		DRM_DEBUG_KMS("Invalid crtc state\n");
353 		return -EINVAL;
354 	}
355 
356 	ret = drm_atomic_helper_check_plane_state(state, crtc_state, 1,
357 						  INT_MAX, true, true);
358 	if (ret)
359 		return ret;
360 
361 	for (i = 0; i < num_planes; i++)
362 		vc4_state->offsets[i] = bo->paddr + fb->offsets[i];
363 
364 	/* We don't support subpixel source positioning for scaling. */
365 	if ((state->src.x1 & subpixel_src_mask) ||
366 	    (state->src.x2 & subpixel_src_mask) ||
367 	    (state->src.y1 & subpixel_src_mask) ||
368 	    (state->src.y2 & subpixel_src_mask)) {
369 		return -EINVAL;
370 	}
371 
372 	vc4_state->src_x = state->src.x1 >> 16;
373 	vc4_state->src_y = state->src.y1 >> 16;
374 	vc4_state->src_w[0] = (state->src.x2 - state->src.x1) >> 16;
375 	vc4_state->src_h[0] = (state->src.y2 - state->src.y1) >> 16;
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 
424 static 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 
442 static 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 
452 static 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->hvs->hvs5 ? 128 : 64);
493 
494 	/* Each "word" of the LBM memory contains 2 or 4 (hvs5) pixels */
495 	lbm /= vc4->hvs->hvs5 ? 4 : 2;
496 
497 	return lbm;
498 }
499 
500 static 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 
532 static 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 
587 static 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->hvs->hvs5 ? 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  */
631 static 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 
671 /* Writes out a full display list for an active plane to the plane's
672  * private dlist state.
673  */
674 static int vc4_plane_mode_set(struct drm_plane *plane,
675 			      struct drm_plane_state *state)
676 {
677 	struct vc4_dev *vc4 = to_vc4_dev(plane->dev);
678 	struct vc4_plane_state *vc4_state = to_vc4_plane_state(state);
679 	struct drm_framebuffer *fb = state->fb;
680 	u32 ctl0_offset = vc4_state->dlist_count;
681 	const struct hvs_format *format = vc4_get_hvs_format(fb->format->format);
682 	u64 base_format_mod = fourcc_mod_broadcom_mod(fb->modifier);
683 	int num_planes = fb->format->num_planes;
684 	u32 h_subsample = fb->format->hsub;
685 	u32 v_subsample = fb->format->vsub;
686 	bool mix_plane_alpha;
687 	bool covers_screen;
688 	u32 scl0, scl1, pitch0;
689 	u32 tiling, src_y;
690 	u32 hvs_format = format->hvs;
691 	unsigned int rotation;
692 	int ret, i;
693 
694 	if (vc4_state->dlist_initialized)
695 		return 0;
696 
697 	ret = vc4_plane_setup_clipping_and_scaling(state);
698 	if (ret)
699 		return ret;
700 
701 	/* SCL1 is used for Cb/Cr scaling of planar formats.  For RGB
702 	 * and 4:4:4, scl1 should be set to scl0 so both channels of
703 	 * the scaler do the same thing.  For YUV, the Y plane needs
704 	 * to be put in channel 1 and Cb/Cr in channel 0, so we swap
705 	 * the scl fields here.
706 	 */
707 	if (num_planes == 1) {
708 		scl0 = vc4_get_scl_field(state, 0);
709 		scl1 = scl0;
710 	} else {
711 		scl0 = vc4_get_scl_field(state, 1);
712 		scl1 = vc4_get_scl_field(state, 0);
713 	}
714 
715 	rotation = drm_rotation_simplify(state->rotation,
716 					 DRM_MODE_ROTATE_0 |
717 					 DRM_MODE_REFLECT_X |
718 					 DRM_MODE_REFLECT_Y);
719 
720 	/* We must point to the last line when Y reflection is enabled. */
721 	src_y = vc4_state->src_y;
722 	if (rotation & DRM_MODE_REFLECT_Y)
723 		src_y += vc4_state->src_h[0] - 1;
724 
725 	switch (base_format_mod) {
726 	case DRM_FORMAT_MOD_LINEAR:
727 		tiling = SCALER_CTL0_TILING_LINEAR;
728 		pitch0 = VC4_SET_FIELD(fb->pitches[0], SCALER_SRC_PITCH);
729 
730 		/* Adjust the base pointer to the first pixel to be scanned
731 		 * out.
732 		 */
733 		for (i = 0; i < num_planes; i++) {
734 			vc4_state->offsets[i] += src_y /
735 						 (i ? v_subsample : 1) *
736 						 fb->pitches[i];
737 
738 			vc4_state->offsets[i] += vc4_state->src_x /
739 						 (i ? h_subsample : 1) *
740 						 fb->format->cpp[i];
741 		}
742 
743 		break;
744 
745 	case DRM_FORMAT_MOD_BROADCOM_VC4_T_TILED: {
746 		u32 tile_size_shift = 12; /* T tiles are 4kb */
747 		/* Whole-tile offsets, mostly for setting the pitch. */
748 		u32 tile_w_shift = fb->format->cpp[0] == 2 ? 6 : 5;
749 		u32 tile_h_shift = 5; /* 16 and 32bpp are 32 pixels high */
750 		u32 tile_w_mask = (1 << tile_w_shift) - 1;
751 		/* The height mask on 32-bit-per-pixel tiles is 63, i.e. twice
752 		 * the height (in pixels) of a 4k tile.
753 		 */
754 		u32 tile_h_mask = (2 << tile_h_shift) - 1;
755 		/* For T-tiled, the FB pitch is "how many bytes from one row to
756 		 * the next, such that
757 		 *
758 		 *	pitch * tile_h == tile_size * tiles_per_row
759 		 */
760 		u32 tiles_w = fb->pitches[0] >> (tile_size_shift - tile_h_shift);
761 		u32 tiles_l = vc4_state->src_x >> tile_w_shift;
762 		u32 tiles_r = tiles_w - tiles_l;
763 		u32 tiles_t = src_y >> tile_h_shift;
764 		/* Intra-tile offsets, which modify the base address (the
765 		 * SCALER_PITCH0_TILE_Y_OFFSET tells HVS how to walk from that
766 		 * base address).
767 		 */
768 		u32 tile_y = (src_y >> 4) & 1;
769 		u32 subtile_y = (src_y >> 2) & 3;
770 		u32 utile_y = src_y & 3;
771 		u32 x_off = vc4_state->src_x & tile_w_mask;
772 		u32 y_off = src_y & tile_h_mask;
773 
774 		/* When Y reflection is requested we must set the
775 		 * SCALER_PITCH0_TILE_LINE_DIR flag to tell HVS that all lines
776 		 * after the initial one should be fetched in descending order,
777 		 * which makes sense since we start from the last line and go
778 		 * backward.
779 		 * Don't know why we need y_off = max_y_off - y_off, but it's
780 		 * definitely required (I guess it's also related to the "going
781 		 * backward" situation).
782 		 */
783 		if (rotation & DRM_MODE_REFLECT_Y) {
784 			y_off = tile_h_mask - y_off;
785 			pitch0 = SCALER_PITCH0_TILE_LINE_DIR;
786 		} else {
787 			pitch0 = 0;
788 		}
789 
790 		tiling = SCALER_CTL0_TILING_256B_OR_T;
791 		pitch0 |= (VC4_SET_FIELD(x_off, SCALER_PITCH0_SINK_PIX) |
792 			   VC4_SET_FIELD(y_off, SCALER_PITCH0_TILE_Y_OFFSET) |
793 			   VC4_SET_FIELD(tiles_l, SCALER_PITCH0_TILE_WIDTH_L) |
794 			   VC4_SET_FIELD(tiles_r, SCALER_PITCH0_TILE_WIDTH_R));
795 		vc4_state->offsets[0] += tiles_t * (tiles_w << tile_size_shift);
796 		vc4_state->offsets[0] += subtile_y << 8;
797 		vc4_state->offsets[0] += utile_y << 4;
798 
799 		/* Rows of tiles alternate left-to-right and right-to-left. */
800 		if (tiles_t & 1) {
801 			pitch0 |= SCALER_PITCH0_TILE_INITIAL_LINE_DIR;
802 			vc4_state->offsets[0] += (tiles_w - tiles_l) <<
803 						 tile_size_shift;
804 			vc4_state->offsets[0] -= (1 + !tile_y) << 10;
805 		} else {
806 			vc4_state->offsets[0] += tiles_l << tile_size_shift;
807 			vc4_state->offsets[0] += tile_y << 10;
808 		}
809 
810 		break;
811 	}
812 
813 	case DRM_FORMAT_MOD_BROADCOM_SAND64:
814 	case DRM_FORMAT_MOD_BROADCOM_SAND128:
815 	case DRM_FORMAT_MOD_BROADCOM_SAND256: {
816 		uint32_t param = fourcc_mod_broadcom_param(fb->modifier);
817 
818 		if (param > SCALER_TILE_HEIGHT_MASK) {
819 			DRM_DEBUG_KMS("SAND height too large (%d)\n",
820 				      param);
821 			return -EINVAL;
822 		}
823 
824 		if (fb->format->format == DRM_FORMAT_P030) {
825 			hvs_format = HVS_PIXEL_FORMAT_YCBCR_10BIT;
826 			tiling = SCALER_CTL0_TILING_128B;
827 		} else {
828 			hvs_format = HVS_PIXEL_FORMAT_H264;
829 
830 			switch (base_format_mod) {
831 			case DRM_FORMAT_MOD_BROADCOM_SAND64:
832 				tiling = SCALER_CTL0_TILING_64B;
833 				break;
834 			case DRM_FORMAT_MOD_BROADCOM_SAND128:
835 				tiling = SCALER_CTL0_TILING_128B;
836 				break;
837 			case DRM_FORMAT_MOD_BROADCOM_SAND256:
838 				tiling = SCALER_CTL0_TILING_256B_OR_T;
839 				break;
840 			default:
841 				return -EINVAL;
842 			}
843 		}
844 
845 		/* Adjust the base pointer to the first pixel to be scanned
846 		 * out.
847 		 *
848 		 * For P030, y_ptr [31:4] is the 128bit word for the start pixel
849 		 * y_ptr [3:0] is the pixel (0-11) contained within that 128bit
850 		 * word that should be taken as the first pixel.
851 		 * Ditto uv_ptr [31:4] vs [3:0], however [3:0] contains the
852 		 * element within the 128bit word, eg for pixel 3 the value
853 		 * should be 6.
854 		 */
855 		for (i = 0; i < num_planes; i++) {
856 			u32 tile_w, tile, x_off, pix_per_tile;
857 
858 			if (fb->format->format == DRM_FORMAT_P030) {
859 				/*
860 				 * Spec says: bits [31:4] of the given address
861 				 * should point to the 128-bit word containing
862 				 * the desired starting pixel, and bits[3:0]
863 				 * should be between 0 and 11, indicating which
864 				 * of the 12-pixels in that 128-bit word is the
865 				 * first pixel to be used
866 				 */
867 				u32 remaining_pixels = vc4_state->src_x % 96;
868 				u32 aligned = remaining_pixels / 12;
869 				u32 last_bits = remaining_pixels % 12;
870 
871 				x_off = aligned * 16 + last_bits;
872 				tile_w = 128;
873 				pix_per_tile = 96;
874 			} else {
875 				switch (base_format_mod) {
876 				case DRM_FORMAT_MOD_BROADCOM_SAND64:
877 					tile_w = 64;
878 					break;
879 				case DRM_FORMAT_MOD_BROADCOM_SAND128:
880 					tile_w = 128;
881 					break;
882 				case DRM_FORMAT_MOD_BROADCOM_SAND256:
883 					tile_w = 256;
884 					break;
885 				default:
886 					return -EINVAL;
887 				}
888 				pix_per_tile = tile_w / fb->format->cpp[0];
889 				x_off = (vc4_state->src_x % pix_per_tile) /
890 					(i ? h_subsample : 1) *
891 					fb->format->cpp[i];
892 			}
893 
894 			tile = vc4_state->src_x / pix_per_tile;
895 
896 			vc4_state->offsets[i] += param * tile_w * tile;
897 			vc4_state->offsets[i] += src_y /
898 						 (i ? v_subsample : 1) *
899 						 tile_w;
900 			vc4_state->offsets[i] += x_off & ~(i ? 1 : 0);
901 		}
902 
903 		pitch0 = VC4_SET_FIELD(param, SCALER_TILE_HEIGHT);
904 		break;
905 	}
906 
907 	default:
908 		DRM_DEBUG_KMS("Unsupported FB tiling flag 0x%16llx",
909 			      (long long)fb->modifier);
910 		return -EINVAL;
911 	}
912 
913 	/* Don't waste cycles mixing with plane alpha if the set alpha
914 	 * is opaque or there is no per-pixel alpha information.
915 	 * In any case we use the alpha property value as the fixed alpha.
916 	 */
917 	mix_plane_alpha = state->alpha != DRM_BLEND_ALPHA_OPAQUE &&
918 			  fb->format->has_alpha;
919 
920 	if (!vc4->hvs->hvs5) {
921 	/* Control word */
922 		vc4_dlist_write(vc4_state,
923 				SCALER_CTL0_VALID |
924 				(rotation & DRM_MODE_REFLECT_X ? SCALER_CTL0_HFLIP : 0) |
925 				(rotation & DRM_MODE_REFLECT_Y ? SCALER_CTL0_VFLIP : 0) |
926 				VC4_SET_FIELD(SCALER_CTL0_RGBA_EXPAND_ROUND, SCALER_CTL0_RGBA_EXPAND) |
927 				(format->pixel_order << SCALER_CTL0_ORDER_SHIFT) |
928 				(hvs_format << SCALER_CTL0_PIXEL_FORMAT_SHIFT) |
929 				VC4_SET_FIELD(tiling, SCALER_CTL0_TILING) |
930 				(vc4_state->is_unity ? SCALER_CTL0_UNITY : 0) |
931 				VC4_SET_FIELD(scl0, SCALER_CTL0_SCL0) |
932 				VC4_SET_FIELD(scl1, SCALER_CTL0_SCL1));
933 
934 		/* Position Word 0: Image Positions and Alpha Value */
935 		vc4_state->pos0_offset = vc4_state->dlist_count;
936 		vc4_dlist_write(vc4_state,
937 				VC4_SET_FIELD(state->alpha >> 8, SCALER_POS0_FIXED_ALPHA) |
938 				VC4_SET_FIELD(vc4_state->crtc_x, SCALER_POS0_START_X) |
939 				VC4_SET_FIELD(vc4_state->crtc_y, SCALER_POS0_START_Y));
940 
941 		/* Position Word 1: Scaled Image Dimensions. */
942 		if (!vc4_state->is_unity) {
943 			vc4_dlist_write(vc4_state,
944 					VC4_SET_FIELD(vc4_state->crtc_w,
945 						      SCALER_POS1_SCL_WIDTH) |
946 					VC4_SET_FIELD(vc4_state->crtc_h,
947 						      SCALER_POS1_SCL_HEIGHT));
948 		}
949 
950 		/* Position Word 2: Source Image Size, Alpha */
951 		vc4_state->pos2_offset = vc4_state->dlist_count;
952 		vc4_dlist_write(vc4_state,
953 				VC4_SET_FIELD(fb->format->has_alpha ?
954 					      SCALER_POS2_ALPHA_MODE_PIPELINE :
955 					      SCALER_POS2_ALPHA_MODE_FIXED,
956 					      SCALER_POS2_ALPHA_MODE) |
957 				(mix_plane_alpha ? SCALER_POS2_ALPHA_MIX : 0) |
958 				(fb->format->has_alpha ?
959 						SCALER_POS2_ALPHA_PREMULT : 0) |
960 				VC4_SET_FIELD(vc4_state->src_w[0],
961 					      SCALER_POS2_WIDTH) |
962 				VC4_SET_FIELD(vc4_state->src_h[0],
963 					      SCALER_POS2_HEIGHT));
964 
965 		/* Position Word 3: Context.  Written by the HVS. */
966 		vc4_dlist_write(vc4_state, 0xc0c0c0c0);
967 
968 	} else {
969 		u32 hvs_pixel_order = format->pixel_order;
970 
971 		if (format->pixel_order_hvs5)
972 			hvs_pixel_order = format->pixel_order_hvs5;
973 
974 		/* Control word */
975 		vc4_dlist_write(vc4_state,
976 				SCALER_CTL0_VALID |
977 				(hvs_pixel_order << SCALER_CTL0_ORDER_SHIFT) |
978 				(hvs_format << SCALER_CTL0_PIXEL_FORMAT_SHIFT) |
979 				VC4_SET_FIELD(tiling, SCALER_CTL0_TILING) |
980 				(vc4_state->is_unity ?
981 						SCALER5_CTL0_UNITY : 0) |
982 				VC4_SET_FIELD(scl0, SCALER_CTL0_SCL0) |
983 				VC4_SET_FIELD(scl1, SCALER_CTL0_SCL1) |
984 				SCALER5_CTL0_ALPHA_EXPAND |
985 				SCALER5_CTL0_RGB_EXPAND);
986 
987 		/* Position Word 0: Image Positions and Alpha Value */
988 		vc4_state->pos0_offset = vc4_state->dlist_count;
989 		vc4_dlist_write(vc4_state,
990 				(rotation & DRM_MODE_REFLECT_Y ?
991 						SCALER5_POS0_VFLIP : 0) |
992 				VC4_SET_FIELD(vc4_state->crtc_x,
993 					      SCALER_POS0_START_X) |
994 				(rotation & DRM_MODE_REFLECT_X ?
995 					      SCALER5_POS0_HFLIP : 0) |
996 				VC4_SET_FIELD(vc4_state->crtc_y,
997 					      SCALER5_POS0_START_Y)
998 			       );
999 
1000 		/* Control Word 2 */
1001 		vc4_dlist_write(vc4_state,
1002 				VC4_SET_FIELD(state->alpha >> 4,
1003 					      SCALER5_CTL2_ALPHA) |
1004 				(fb->format->has_alpha ?
1005 					SCALER5_CTL2_ALPHA_PREMULT : 0) |
1006 				(mix_plane_alpha ?
1007 					SCALER5_CTL2_ALPHA_MIX : 0) |
1008 				VC4_SET_FIELD(fb->format->has_alpha ?
1009 				      SCALER5_CTL2_ALPHA_MODE_PIPELINE :
1010 				      SCALER5_CTL2_ALPHA_MODE_FIXED,
1011 				      SCALER5_CTL2_ALPHA_MODE)
1012 			       );
1013 
1014 		/* Position Word 1: Scaled Image Dimensions. */
1015 		if (!vc4_state->is_unity) {
1016 			vc4_dlist_write(vc4_state,
1017 					VC4_SET_FIELD(vc4_state->crtc_w,
1018 						      SCALER5_POS1_SCL_WIDTH) |
1019 					VC4_SET_FIELD(vc4_state->crtc_h,
1020 						      SCALER5_POS1_SCL_HEIGHT));
1021 		}
1022 
1023 		/* Position Word 2: Source Image Size */
1024 		vc4_state->pos2_offset = vc4_state->dlist_count;
1025 		vc4_dlist_write(vc4_state,
1026 				VC4_SET_FIELD(vc4_state->src_w[0],
1027 					      SCALER5_POS2_WIDTH) |
1028 				VC4_SET_FIELD(vc4_state->src_h[0],
1029 					      SCALER5_POS2_HEIGHT));
1030 
1031 		/* Position Word 3: Context.  Written by the HVS. */
1032 		vc4_dlist_write(vc4_state, 0xc0c0c0c0);
1033 	}
1034 
1035 
1036 	/* Pointer Word 0/1/2: RGB / Y / Cb / Cr Pointers
1037 	 *
1038 	 * The pointers may be any byte address.
1039 	 */
1040 	vc4_state->ptr0_offset = vc4_state->dlist_count;
1041 	for (i = 0; i < num_planes; i++)
1042 		vc4_dlist_write(vc4_state, vc4_state->offsets[i]);
1043 
1044 	/* Pointer Context Word 0/1/2: Written by the HVS */
1045 	for (i = 0; i < num_planes; i++)
1046 		vc4_dlist_write(vc4_state, 0xc0c0c0c0);
1047 
1048 	/* Pitch word 0 */
1049 	vc4_dlist_write(vc4_state, pitch0);
1050 
1051 	/* Pitch word 1/2 */
1052 	for (i = 1; i < num_planes; i++) {
1053 		if (hvs_format != HVS_PIXEL_FORMAT_H264 &&
1054 		    hvs_format != HVS_PIXEL_FORMAT_YCBCR_10BIT) {
1055 			vc4_dlist_write(vc4_state,
1056 					VC4_SET_FIELD(fb->pitches[i],
1057 						      SCALER_SRC_PITCH));
1058 		} else {
1059 			vc4_dlist_write(vc4_state, pitch0);
1060 		}
1061 	}
1062 
1063 	/* Colorspace conversion words */
1064 	if (vc4_state->is_yuv) {
1065 		enum drm_color_encoding color_encoding = state->color_encoding;
1066 		enum drm_color_range color_range = state->color_range;
1067 		const u32 *ccm;
1068 
1069 		if (color_encoding >= DRM_COLOR_ENCODING_MAX)
1070 			color_encoding = DRM_COLOR_YCBCR_BT601;
1071 		if (color_range >= DRM_COLOR_RANGE_MAX)
1072 			color_range = DRM_COLOR_YCBCR_LIMITED_RANGE;
1073 
1074 		ccm = colorspace_coeffs[color_range][color_encoding];
1075 
1076 		vc4_dlist_write(vc4_state, ccm[0]);
1077 		vc4_dlist_write(vc4_state, ccm[1]);
1078 		vc4_dlist_write(vc4_state, ccm[2]);
1079 	}
1080 
1081 	vc4_state->lbm_offset = 0;
1082 
1083 	if (vc4_state->x_scaling[0] != VC4_SCALING_NONE ||
1084 	    vc4_state->x_scaling[1] != VC4_SCALING_NONE ||
1085 	    vc4_state->y_scaling[0] != VC4_SCALING_NONE ||
1086 	    vc4_state->y_scaling[1] != VC4_SCALING_NONE) {
1087 		/* Reserve a slot for the LBM Base Address. The real value will
1088 		 * be set when calling vc4_plane_allocate_lbm().
1089 		 */
1090 		if (vc4_state->y_scaling[0] != VC4_SCALING_NONE ||
1091 		    vc4_state->y_scaling[1] != VC4_SCALING_NONE) {
1092 			vc4_state->lbm_offset = vc4_state->dlist_count;
1093 			vc4_dlist_counter_increment(vc4_state);
1094 		}
1095 
1096 		if (num_planes > 1) {
1097 			/* Emit Cb/Cr as channel 0 and Y as channel
1098 			 * 1. This matches how we set up scl0/scl1
1099 			 * above.
1100 			 */
1101 			vc4_write_scaling_parameters(state, 1);
1102 		}
1103 		vc4_write_scaling_parameters(state, 0);
1104 
1105 		/* If any PPF setup was done, then all the kernel
1106 		 * pointers get uploaded.
1107 		 */
1108 		if (vc4_state->x_scaling[0] == VC4_SCALING_PPF ||
1109 		    vc4_state->y_scaling[0] == VC4_SCALING_PPF ||
1110 		    vc4_state->x_scaling[1] == VC4_SCALING_PPF ||
1111 		    vc4_state->y_scaling[1] == VC4_SCALING_PPF) {
1112 			u32 kernel = VC4_SET_FIELD(vc4->hvs->mitchell_netravali_filter.start,
1113 						   SCALER_PPF_KERNEL_OFFSET);
1114 
1115 			/* HPPF plane 0 */
1116 			vc4_dlist_write(vc4_state, kernel);
1117 			/* VPPF plane 0 */
1118 			vc4_dlist_write(vc4_state, kernel);
1119 			/* HPPF plane 1 */
1120 			vc4_dlist_write(vc4_state, kernel);
1121 			/* VPPF plane 1 */
1122 			vc4_dlist_write(vc4_state, kernel);
1123 		}
1124 	}
1125 
1126 	vc4_state->dlist[ctl0_offset] |=
1127 		VC4_SET_FIELD(vc4_state->dlist_count, SCALER_CTL0_SIZE);
1128 
1129 	/* crtc_* are already clipped coordinates. */
1130 	covers_screen = vc4_state->crtc_x == 0 && vc4_state->crtc_y == 0 &&
1131 			vc4_state->crtc_w == state->crtc->mode.hdisplay &&
1132 			vc4_state->crtc_h == state->crtc->mode.vdisplay;
1133 	/* Background fill might be necessary when the plane has per-pixel
1134 	 * alpha content or a non-opaque plane alpha and could blend from the
1135 	 * background or does not cover the entire screen.
1136 	 */
1137 	vc4_state->needs_bg_fill = fb->format->has_alpha || !covers_screen ||
1138 				   state->alpha != DRM_BLEND_ALPHA_OPAQUE;
1139 
1140 	/* Flag the dlist as initialized to avoid checking it twice in case
1141 	 * the async update check already called vc4_plane_mode_set() and
1142 	 * decided to fallback to sync update because async update was not
1143 	 * possible.
1144 	 */
1145 	vc4_state->dlist_initialized = 1;
1146 
1147 	vc4_plane_calc_load(state);
1148 
1149 	return 0;
1150 }
1151 
1152 /* If a modeset involves changing the setup of a plane, the atomic
1153  * infrastructure will call this to validate a proposed plane setup.
1154  * However, if a plane isn't getting updated, this (and the
1155  * corresponding vc4_plane_atomic_update) won't get called.  Thus, we
1156  * compute the dlist here and have all active plane dlists get updated
1157  * in the CRTC's flush.
1158  */
1159 static int vc4_plane_atomic_check(struct drm_plane *plane,
1160 				  struct drm_atomic_state *state)
1161 {
1162 	struct drm_plane_state *new_plane_state = drm_atomic_get_new_plane_state(state,
1163 										 plane);
1164 	struct vc4_plane_state *vc4_state = to_vc4_plane_state(new_plane_state);
1165 	int ret;
1166 
1167 	vc4_state->dlist_count = 0;
1168 
1169 	if (!plane_enabled(new_plane_state))
1170 		return 0;
1171 
1172 	ret = vc4_plane_mode_set(plane, new_plane_state);
1173 	if (ret)
1174 		return ret;
1175 
1176 	return vc4_plane_allocate_lbm(new_plane_state);
1177 }
1178 
1179 static void vc4_plane_atomic_update(struct drm_plane *plane,
1180 				    struct drm_atomic_state *state)
1181 {
1182 	/* No contents here.  Since we don't know where in the CRTC's
1183 	 * dlist we should be stored, our dlist is uploaded to the
1184 	 * hardware with vc4_plane_write_dlist() at CRTC atomic_flush
1185 	 * time.
1186 	 */
1187 }
1188 
1189 u32 vc4_plane_write_dlist(struct drm_plane *plane, u32 __iomem *dlist)
1190 {
1191 	struct vc4_plane_state *vc4_state = to_vc4_plane_state(plane->state);
1192 	int i;
1193 
1194 	vc4_state->hw_dlist = dlist;
1195 
1196 	/* Can't memcpy_toio() because it needs to be 32-bit writes. */
1197 	for (i = 0; i < vc4_state->dlist_count; i++)
1198 		writel(vc4_state->dlist[i], &dlist[i]);
1199 
1200 	return vc4_state->dlist_count;
1201 }
1202 
1203 u32 vc4_plane_dlist_size(const struct drm_plane_state *state)
1204 {
1205 	const struct vc4_plane_state *vc4_state =
1206 		container_of(state, typeof(*vc4_state), base);
1207 
1208 	return vc4_state->dlist_count;
1209 }
1210 
1211 /* Updates the plane to immediately (well, once the FIFO needs
1212  * refilling) scan out from at a new framebuffer.
1213  */
1214 void vc4_plane_async_set_fb(struct drm_plane *plane, struct drm_framebuffer *fb)
1215 {
1216 	struct vc4_plane_state *vc4_state = to_vc4_plane_state(plane->state);
1217 	struct drm_gem_cma_object *bo = drm_fb_cma_get_gem_obj(fb, 0);
1218 	uint32_t addr;
1219 
1220 	/* We're skipping the address adjustment for negative origin,
1221 	 * because this is only called on the primary plane.
1222 	 */
1223 	WARN_ON_ONCE(plane->state->crtc_x < 0 || plane->state->crtc_y < 0);
1224 	addr = bo->paddr + fb->offsets[0];
1225 
1226 	/* Write the new address into the hardware immediately.  The
1227 	 * scanout will start from this address as soon as the FIFO
1228 	 * needs to refill with pixels.
1229 	 */
1230 	writel(addr, &vc4_state->hw_dlist[vc4_state->ptr0_offset]);
1231 
1232 	/* Also update the CPU-side dlist copy, so that any later
1233 	 * atomic updates that don't do a new modeset on our plane
1234 	 * also use our updated address.
1235 	 */
1236 	vc4_state->dlist[vc4_state->ptr0_offset] = addr;
1237 }
1238 
1239 static void vc4_plane_atomic_async_update(struct drm_plane *plane,
1240 					  struct drm_atomic_state *state)
1241 {
1242 	struct drm_plane_state *new_plane_state = drm_atomic_get_new_plane_state(state,
1243 										 plane);
1244 	struct vc4_plane_state *vc4_state, *new_vc4_state;
1245 
1246 	swap(plane->state->fb, new_plane_state->fb);
1247 	plane->state->crtc_x = new_plane_state->crtc_x;
1248 	plane->state->crtc_y = new_plane_state->crtc_y;
1249 	plane->state->crtc_w = new_plane_state->crtc_w;
1250 	plane->state->crtc_h = new_plane_state->crtc_h;
1251 	plane->state->src_x = new_plane_state->src_x;
1252 	plane->state->src_y = new_plane_state->src_y;
1253 	plane->state->src_w = new_plane_state->src_w;
1254 	plane->state->src_h = new_plane_state->src_h;
1255 	plane->state->alpha = new_plane_state->alpha;
1256 	plane->state->pixel_blend_mode = new_plane_state->pixel_blend_mode;
1257 	plane->state->rotation = new_plane_state->rotation;
1258 	plane->state->zpos = new_plane_state->zpos;
1259 	plane->state->normalized_zpos = new_plane_state->normalized_zpos;
1260 	plane->state->color_encoding = new_plane_state->color_encoding;
1261 	plane->state->color_range = new_plane_state->color_range;
1262 	plane->state->src = new_plane_state->src;
1263 	plane->state->dst = new_plane_state->dst;
1264 	plane->state->visible = new_plane_state->visible;
1265 
1266 	new_vc4_state = to_vc4_plane_state(new_plane_state);
1267 	vc4_state = to_vc4_plane_state(plane->state);
1268 
1269 	vc4_state->crtc_x = new_vc4_state->crtc_x;
1270 	vc4_state->crtc_y = new_vc4_state->crtc_y;
1271 	vc4_state->crtc_h = new_vc4_state->crtc_h;
1272 	vc4_state->crtc_w = new_vc4_state->crtc_w;
1273 	vc4_state->src_x = new_vc4_state->src_x;
1274 	vc4_state->src_y = new_vc4_state->src_y;
1275 	memcpy(vc4_state->src_w, new_vc4_state->src_w,
1276 	       sizeof(vc4_state->src_w));
1277 	memcpy(vc4_state->src_h, new_vc4_state->src_h,
1278 	       sizeof(vc4_state->src_h));
1279 	memcpy(vc4_state->x_scaling, new_vc4_state->x_scaling,
1280 	       sizeof(vc4_state->x_scaling));
1281 	memcpy(vc4_state->y_scaling, new_vc4_state->y_scaling,
1282 	       sizeof(vc4_state->y_scaling));
1283 	vc4_state->is_unity = new_vc4_state->is_unity;
1284 	vc4_state->is_yuv = new_vc4_state->is_yuv;
1285 	memcpy(vc4_state->offsets, new_vc4_state->offsets,
1286 	       sizeof(vc4_state->offsets));
1287 	vc4_state->needs_bg_fill = new_vc4_state->needs_bg_fill;
1288 
1289 	/* Update the current vc4_state pos0, pos2 and ptr0 dlist entries. */
1290 	vc4_state->dlist[vc4_state->pos0_offset] =
1291 		new_vc4_state->dlist[vc4_state->pos0_offset];
1292 	vc4_state->dlist[vc4_state->pos2_offset] =
1293 		new_vc4_state->dlist[vc4_state->pos2_offset];
1294 	vc4_state->dlist[vc4_state->ptr0_offset] =
1295 		new_vc4_state->dlist[vc4_state->ptr0_offset];
1296 
1297 	/* Note that we can't just call vc4_plane_write_dlist()
1298 	 * because that would smash the context data that the HVS is
1299 	 * currently using.
1300 	 */
1301 	writel(vc4_state->dlist[vc4_state->pos0_offset],
1302 	       &vc4_state->hw_dlist[vc4_state->pos0_offset]);
1303 	writel(vc4_state->dlist[vc4_state->pos2_offset],
1304 	       &vc4_state->hw_dlist[vc4_state->pos2_offset]);
1305 	writel(vc4_state->dlist[vc4_state->ptr0_offset],
1306 	       &vc4_state->hw_dlist[vc4_state->ptr0_offset]);
1307 }
1308 
1309 static int vc4_plane_atomic_async_check(struct drm_plane *plane,
1310 					struct drm_atomic_state *state)
1311 {
1312 	struct drm_plane_state *new_plane_state = drm_atomic_get_new_plane_state(state,
1313 										 plane);
1314 	struct vc4_plane_state *old_vc4_state, *new_vc4_state;
1315 	int ret;
1316 	u32 i;
1317 
1318 	ret = vc4_plane_mode_set(plane, new_plane_state);
1319 	if (ret)
1320 		return ret;
1321 
1322 	old_vc4_state = to_vc4_plane_state(plane->state);
1323 	new_vc4_state = to_vc4_plane_state(new_plane_state);
1324 	if (old_vc4_state->dlist_count != new_vc4_state->dlist_count ||
1325 	    old_vc4_state->pos0_offset != new_vc4_state->pos0_offset ||
1326 	    old_vc4_state->pos2_offset != new_vc4_state->pos2_offset ||
1327 	    old_vc4_state->ptr0_offset != new_vc4_state->ptr0_offset ||
1328 	    vc4_lbm_size(plane->state) != vc4_lbm_size(new_plane_state))
1329 		return -EINVAL;
1330 
1331 	/* Only pos0, pos2 and ptr0 DWORDS can be updated in an async update
1332 	 * if anything else has changed, fallback to a sync update.
1333 	 */
1334 	for (i = 0; i < new_vc4_state->dlist_count; i++) {
1335 		if (i == new_vc4_state->pos0_offset ||
1336 		    i == new_vc4_state->pos2_offset ||
1337 		    i == new_vc4_state->ptr0_offset ||
1338 		    (new_vc4_state->lbm_offset &&
1339 		     i == new_vc4_state->lbm_offset))
1340 			continue;
1341 
1342 		if (new_vc4_state->dlist[i] != old_vc4_state->dlist[i])
1343 			return -EINVAL;
1344 	}
1345 
1346 	return 0;
1347 }
1348 
1349 static int vc4_prepare_fb(struct drm_plane *plane,
1350 			  struct drm_plane_state *state)
1351 {
1352 	struct vc4_bo *bo;
1353 	int ret;
1354 
1355 	if (!state->fb)
1356 		return 0;
1357 
1358 	bo = to_vc4_bo(&drm_fb_cma_get_gem_obj(state->fb, 0)->base);
1359 
1360 	drm_gem_plane_helper_prepare_fb(plane, state);
1361 
1362 	if (plane->state->fb == state->fb)
1363 		return 0;
1364 
1365 	ret = vc4_bo_inc_usecnt(bo);
1366 	if (ret)
1367 		return ret;
1368 
1369 	return 0;
1370 }
1371 
1372 static void vc4_cleanup_fb(struct drm_plane *plane,
1373 			   struct drm_plane_state *state)
1374 {
1375 	struct vc4_bo *bo;
1376 
1377 	if (plane->state->fb == state->fb || !state->fb)
1378 		return;
1379 
1380 	bo = to_vc4_bo(&drm_fb_cma_get_gem_obj(state->fb, 0)->base);
1381 	vc4_bo_dec_usecnt(bo);
1382 }
1383 
1384 static const struct drm_plane_helper_funcs vc4_plane_helper_funcs = {
1385 	.atomic_check = vc4_plane_atomic_check,
1386 	.atomic_update = vc4_plane_atomic_update,
1387 	.prepare_fb = vc4_prepare_fb,
1388 	.cleanup_fb = vc4_cleanup_fb,
1389 	.atomic_async_check = vc4_plane_atomic_async_check,
1390 	.atomic_async_update = vc4_plane_atomic_async_update,
1391 };
1392 
1393 static bool vc4_format_mod_supported(struct drm_plane *plane,
1394 				     uint32_t format,
1395 				     uint64_t modifier)
1396 {
1397 	/* Support T_TILING for RGB formats only. */
1398 	switch (format) {
1399 	case DRM_FORMAT_XRGB8888:
1400 	case DRM_FORMAT_ARGB8888:
1401 	case DRM_FORMAT_ABGR8888:
1402 	case DRM_FORMAT_XBGR8888:
1403 	case DRM_FORMAT_RGB565:
1404 	case DRM_FORMAT_BGR565:
1405 	case DRM_FORMAT_ARGB1555:
1406 	case DRM_FORMAT_XRGB1555:
1407 		switch (fourcc_mod_broadcom_mod(modifier)) {
1408 		case DRM_FORMAT_MOD_LINEAR:
1409 		case DRM_FORMAT_MOD_BROADCOM_VC4_T_TILED:
1410 			return true;
1411 		default:
1412 			return false;
1413 		}
1414 	case DRM_FORMAT_NV12:
1415 	case DRM_FORMAT_NV21:
1416 		switch (fourcc_mod_broadcom_mod(modifier)) {
1417 		case DRM_FORMAT_MOD_LINEAR:
1418 		case DRM_FORMAT_MOD_BROADCOM_SAND64:
1419 		case DRM_FORMAT_MOD_BROADCOM_SAND128:
1420 		case DRM_FORMAT_MOD_BROADCOM_SAND256:
1421 			return true;
1422 		default:
1423 			return false;
1424 		}
1425 	case DRM_FORMAT_P030:
1426 		switch (fourcc_mod_broadcom_mod(modifier)) {
1427 		case DRM_FORMAT_MOD_BROADCOM_SAND128:
1428 			return true;
1429 		default:
1430 			return false;
1431 		}
1432 	case DRM_FORMAT_RGBX1010102:
1433 	case DRM_FORMAT_BGRX1010102:
1434 	case DRM_FORMAT_RGBA1010102:
1435 	case DRM_FORMAT_BGRA1010102:
1436 	case DRM_FORMAT_YUV422:
1437 	case DRM_FORMAT_YVU422:
1438 	case DRM_FORMAT_YUV420:
1439 	case DRM_FORMAT_YVU420:
1440 	case DRM_FORMAT_NV16:
1441 	case DRM_FORMAT_NV61:
1442 	default:
1443 		return (modifier == DRM_FORMAT_MOD_LINEAR);
1444 	}
1445 }
1446 
1447 static const struct drm_plane_funcs vc4_plane_funcs = {
1448 	.update_plane = drm_atomic_helper_update_plane,
1449 	.disable_plane = drm_atomic_helper_disable_plane,
1450 	.destroy = drm_plane_cleanup,
1451 	.set_property = NULL,
1452 	.reset = vc4_plane_reset,
1453 	.atomic_duplicate_state = vc4_plane_duplicate_state,
1454 	.atomic_destroy_state = vc4_plane_destroy_state,
1455 	.format_mod_supported = vc4_format_mod_supported,
1456 };
1457 
1458 struct drm_plane *vc4_plane_init(struct drm_device *dev,
1459 				 enum drm_plane_type type)
1460 {
1461 	struct drm_plane *plane = NULL;
1462 	struct vc4_plane *vc4_plane;
1463 	u32 formats[ARRAY_SIZE(hvs_formats)];
1464 	int num_formats = 0;
1465 	int ret = 0;
1466 	unsigned i;
1467 	bool hvs5 = of_device_is_compatible(dev->dev->of_node,
1468 					    "brcm,bcm2711-vc5");
1469 	static const uint64_t modifiers[] = {
1470 		DRM_FORMAT_MOD_BROADCOM_VC4_T_TILED,
1471 		DRM_FORMAT_MOD_BROADCOM_SAND128,
1472 		DRM_FORMAT_MOD_BROADCOM_SAND64,
1473 		DRM_FORMAT_MOD_BROADCOM_SAND256,
1474 		DRM_FORMAT_MOD_LINEAR,
1475 		DRM_FORMAT_MOD_INVALID
1476 	};
1477 
1478 	vc4_plane = devm_kzalloc(dev->dev, sizeof(*vc4_plane),
1479 				 GFP_KERNEL);
1480 	if (!vc4_plane)
1481 		return ERR_PTR(-ENOMEM);
1482 
1483 	for (i = 0; i < ARRAY_SIZE(hvs_formats); i++) {
1484 		if (!hvs_formats[i].hvs5_only || hvs5) {
1485 			formats[num_formats] = hvs_formats[i].drm;
1486 			num_formats++;
1487 		}
1488 	}
1489 
1490 	plane = &vc4_plane->base;
1491 	ret = drm_universal_plane_init(dev, plane, 0,
1492 				       &vc4_plane_funcs,
1493 				       formats, num_formats,
1494 				       modifiers, type, NULL);
1495 	if (ret)
1496 		return ERR_PTR(ret);
1497 
1498 	drm_plane_helper_add(plane, &vc4_plane_helper_funcs);
1499 
1500 	drm_plane_create_alpha_property(plane);
1501 	drm_plane_create_rotation_property(plane, DRM_MODE_ROTATE_0,
1502 					   DRM_MODE_ROTATE_0 |
1503 					   DRM_MODE_ROTATE_180 |
1504 					   DRM_MODE_REFLECT_X |
1505 					   DRM_MODE_REFLECT_Y);
1506 
1507 	drm_plane_create_color_properties(plane,
1508 					  BIT(DRM_COLOR_YCBCR_BT601) |
1509 					  BIT(DRM_COLOR_YCBCR_BT709) |
1510 					  BIT(DRM_COLOR_YCBCR_BT2020),
1511 					  BIT(DRM_COLOR_YCBCR_LIMITED_RANGE) |
1512 					  BIT(DRM_COLOR_YCBCR_FULL_RANGE),
1513 					  DRM_COLOR_YCBCR_BT709,
1514 					  DRM_COLOR_YCBCR_LIMITED_RANGE);
1515 
1516 	return plane;
1517 }
1518 
1519 int vc4_plane_create_additional_planes(struct drm_device *drm)
1520 {
1521 	struct drm_plane *cursor_plane;
1522 	struct drm_crtc *crtc;
1523 	unsigned int i;
1524 
1525 	/* Set up some arbitrary number of planes.  We're not limited
1526 	 * by a set number of physical registers, just the space in
1527 	 * the HVS (16k) and how small an plane can be (28 bytes).
1528 	 * However, each plane we set up takes up some memory, and
1529 	 * increases the cost of looping over planes, which atomic
1530 	 * modesetting does quite a bit.  As a result, we pick a
1531 	 * modest number of planes to expose, that should hopefully
1532 	 * still cover any sane usecase.
1533 	 */
1534 	for (i = 0; i < 16; i++) {
1535 		struct drm_plane *plane =
1536 			vc4_plane_init(drm, DRM_PLANE_TYPE_OVERLAY);
1537 
1538 		if (IS_ERR(plane))
1539 			continue;
1540 
1541 		plane->possible_crtcs =
1542 			GENMASK(drm->mode_config.num_crtc - 1, 0);
1543 	}
1544 
1545 	drm_for_each_crtc(crtc, drm) {
1546 		/* Set up the legacy cursor after overlay initialization,
1547 		 * since we overlay planes on the CRTC in the order they were
1548 		 * initialized.
1549 		 */
1550 		cursor_plane = vc4_plane_init(drm, DRM_PLANE_TYPE_CURSOR);
1551 		if (!IS_ERR(cursor_plane)) {
1552 			cursor_plane->possible_crtcs = drm_crtc_mask(crtc);
1553 			crtc->cursor = cursor_plane;
1554 		}
1555 	}
1556 
1557 	return 0;
1558 }
1559