xref: /linux/drivers/gpu/drm/vc4/vc4_plane.c (revision c32e64e852f3f5c0fd709f84bc94736840088375)
1 /*
2  * Copyright (C) 2015 Broadcom
3  *
4  * This program is free software; you can redistribute it and/or modify
5  * it under the terms of the GNU General Public License version 2 as
6  * published by the Free Software Foundation.
7  */
8 
9 /**
10  * DOC: VC4 plane module
11  *
12  * Each DRM plane is a layer of pixels being scanned out by the HVS.
13  *
14  * At atomic modeset check time, we compute the HVS display element
15  * state that would be necessary for displaying the plane (giving us a
16  * chance to figure out if a plane configuration is invalid), then at
17  * atomic flush time the CRTC will ask us to write our element state
18  * into the region of the HVS that it has allocated for us.
19  */
20 
21 #include <drm/drm_atomic.h>
22 #include <drm/drm_atomic_helper.h>
23 #include <drm/drm_fb_cma_helper.h>
24 #include <drm/drm_plane_helper.h>
25 
26 #include "uapi/drm/vc4_drm.h"
27 #include "vc4_drv.h"
28 #include "vc4_regs.h"
29 
30 static const struct hvs_format {
31 	u32 drm; /* DRM_FORMAT_* */
32 	u32 hvs; /* HVS_FORMAT_* */
33 	u32 pixel_order;
34 } hvs_formats[] = {
35 	{
36 		.drm = DRM_FORMAT_XRGB8888, .hvs = HVS_PIXEL_FORMAT_RGBA8888,
37 		.pixel_order = HVS_PIXEL_ORDER_ABGR,
38 	},
39 	{
40 		.drm = DRM_FORMAT_ARGB8888, .hvs = HVS_PIXEL_FORMAT_RGBA8888,
41 		.pixel_order = HVS_PIXEL_ORDER_ABGR,
42 	},
43 	{
44 		.drm = DRM_FORMAT_ABGR8888, .hvs = HVS_PIXEL_FORMAT_RGBA8888,
45 		.pixel_order = HVS_PIXEL_ORDER_ARGB,
46 	},
47 	{
48 		.drm = DRM_FORMAT_XBGR8888, .hvs = HVS_PIXEL_FORMAT_RGBA8888,
49 		.pixel_order = HVS_PIXEL_ORDER_ARGB,
50 	},
51 	{
52 		.drm = DRM_FORMAT_RGB565, .hvs = HVS_PIXEL_FORMAT_RGB565,
53 		.pixel_order = HVS_PIXEL_ORDER_XRGB,
54 	},
55 	{
56 		.drm = DRM_FORMAT_BGR565, .hvs = HVS_PIXEL_FORMAT_RGB565,
57 		.pixel_order = HVS_PIXEL_ORDER_XBGR,
58 	},
59 	{
60 		.drm = DRM_FORMAT_ARGB1555, .hvs = HVS_PIXEL_FORMAT_RGBA5551,
61 		.pixel_order = HVS_PIXEL_ORDER_ABGR,
62 	},
63 	{
64 		.drm = DRM_FORMAT_XRGB1555, .hvs = HVS_PIXEL_FORMAT_RGBA5551,
65 		.pixel_order = HVS_PIXEL_ORDER_ABGR,
66 	},
67 	{
68 		.drm = DRM_FORMAT_RGB888, .hvs = HVS_PIXEL_FORMAT_RGB888,
69 		.pixel_order = HVS_PIXEL_ORDER_XRGB,
70 	},
71 	{
72 		.drm = DRM_FORMAT_BGR888, .hvs = HVS_PIXEL_FORMAT_RGB888,
73 		.pixel_order = HVS_PIXEL_ORDER_XBGR,
74 	},
75 	{
76 		.drm = DRM_FORMAT_YUV422,
77 		.hvs = HVS_PIXEL_FORMAT_YCBCR_YUV422_3PLANE,
78 		.pixel_order = HVS_PIXEL_ORDER_XYCBCR,
79 	},
80 	{
81 		.drm = DRM_FORMAT_YVU422,
82 		.hvs = HVS_PIXEL_FORMAT_YCBCR_YUV422_3PLANE,
83 		.pixel_order = HVS_PIXEL_ORDER_XYCRCB,
84 	},
85 	{
86 		.drm = DRM_FORMAT_YUV420,
87 		.hvs = HVS_PIXEL_FORMAT_YCBCR_YUV420_3PLANE,
88 		.pixel_order = HVS_PIXEL_ORDER_XYCBCR,
89 	},
90 	{
91 		.drm = DRM_FORMAT_YVU420,
92 		.hvs = HVS_PIXEL_FORMAT_YCBCR_YUV420_3PLANE,
93 		.pixel_order = HVS_PIXEL_ORDER_XYCRCB,
94 	},
95 	{
96 		.drm = DRM_FORMAT_NV12,
97 		.hvs = HVS_PIXEL_FORMAT_YCBCR_YUV420_2PLANE,
98 		.pixel_order = HVS_PIXEL_ORDER_XYCBCR,
99 	},
100 	{
101 		.drm = DRM_FORMAT_NV21,
102 		.hvs = HVS_PIXEL_FORMAT_YCBCR_YUV420_2PLANE,
103 		.pixel_order = HVS_PIXEL_ORDER_XYCRCB,
104 	},
105 	{
106 		.drm = DRM_FORMAT_NV16,
107 		.hvs = HVS_PIXEL_FORMAT_YCBCR_YUV422_2PLANE,
108 		.pixel_order = HVS_PIXEL_ORDER_XYCBCR,
109 	},
110 	{
111 		.drm = DRM_FORMAT_NV61,
112 		.hvs = HVS_PIXEL_FORMAT_YCBCR_YUV422_2PLANE,
113 		.pixel_order = HVS_PIXEL_ORDER_XYCRCB,
114 	},
115 };
116 
117 static const struct hvs_format *vc4_get_hvs_format(u32 drm_format)
118 {
119 	unsigned i;
120 
121 	for (i = 0; i < ARRAY_SIZE(hvs_formats); i++) {
122 		if (hvs_formats[i].drm == drm_format)
123 			return &hvs_formats[i];
124 	}
125 
126 	return NULL;
127 }
128 
129 static enum vc4_scaling_mode vc4_get_scaling_mode(u32 src, u32 dst)
130 {
131 	if (dst > src)
132 		return VC4_SCALING_PPF;
133 	else if (dst < src)
134 		return VC4_SCALING_TPZ;
135 	else
136 		return VC4_SCALING_NONE;
137 }
138 
139 static bool plane_enabled(struct drm_plane_state *state)
140 {
141 	return state->fb && state->crtc;
142 }
143 
144 static struct drm_plane_state *vc4_plane_duplicate_state(struct drm_plane *plane)
145 {
146 	struct vc4_plane_state *vc4_state;
147 
148 	if (WARN_ON(!plane->state))
149 		return NULL;
150 
151 	vc4_state = kmemdup(plane->state, sizeof(*vc4_state), GFP_KERNEL);
152 	if (!vc4_state)
153 		return NULL;
154 
155 	memset(&vc4_state->lbm, 0, sizeof(vc4_state->lbm));
156 
157 	__drm_atomic_helper_plane_duplicate_state(plane, &vc4_state->base);
158 
159 	if (vc4_state->dlist) {
160 		vc4_state->dlist = kmemdup(vc4_state->dlist,
161 					   vc4_state->dlist_count * 4,
162 					   GFP_KERNEL);
163 		if (!vc4_state->dlist) {
164 			kfree(vc4_state);
165 			return NULL;
166 		}
167 		vc4_state->dlist_size = vc4_state->dlist_count;
168 	}
169 
170 	return &vc4_state->base;
171 }
172 
173 static void vc4_plane_destroy_state(struct drm_plane *plane,
174 				    struct drm_plane_state *state)
175 {
176 	struct vc4_dev *vc4 = to_vc4_dev(plane->dev);
177 	struct vc4_plane_state *vc4_state = to_vc4_plane_state(state);
178 
179 	if (vc4_state->lbm.allocated) {
180 		unsigned long irqflags;
181 
182 		spin_lock_irqsave(&vc4->hvs->mm_lock, irqflags);
183 		drm_mm_remove_node(&vc4_state->lbm);
184 		spin_unlock_irqrestore(&vc4->hvs->mm_lock, irqflags);
185 	}
186 
187 	kfree(vc4_state->dlist);
188 	__drm_atomic_helper_plane_destroy_state(&vc4_state->base);
189 	kfree(state);
190 }
191 
192 /* Called during init to allocate the plane's atomic state. */
193 static void vc4_plane_reset(struct drm_plane *plane)
194 {
195 	struct vc4_plane_state *vc4_state;
196 
197 	WARN_ON(plane->state);
198 
199 	vc4_state = kzalloc(sizeof(*vc4_state), GFP_KERNEL);
200 	if (!vc4_state)
201 		return;
202 
203 	plane->state = &vc4_state->base;
204 	plane->state->alpha = DRM_BLEND_ALPHA_OPAQUE;
205 	vc4_state->base.plane = plane;
206 }
207 
208 static void vc4_dlist_write(struct vc4_plane_state *vc4_state, u32 val)
209 {
210 	if (vc4_state->dlist_count == vc4_state->dlist_size) {
211 		u32 new_size = max(4u, vc4_state->dlist_count * 2);
212 		u32 *new_dlist = kmalloc_array(new_size, 4, GFP_KERNEL);
213 
214 		if (!new_dlist)
215 			return;
216 		memcpy(new_dlist, vc4_state->dlist, vc4_state->dlist_count * 4);
217 
218 		kfree(vc4_state->dlist);
219 		vc4_state->dlist = new_dlist;
220 		vc4_state->dlist_size = new_size;
221 	}
222 
223 	vc4_state->dlist[vc4_state->dlist_count++] = val;
224 }
225 
226 /* Returns the scl0/scl1 field based on whether the dimensions need to
227  * be up/down/non-scaled.
228  *
229  * This is a replication of a table from the spec.
230  */
231 static u32 vc4_get_scl_field(struct drm_plane_state *state, int plane)
232 {
233 	struct vc4_plane_state *vc4_state = to_vc4_plane_state(state);
234 
235 	switch (vc4_state->x_scaling[plane] << 2 | vc4_state->y_scaling[plane]) {
236 	case VC4_SCALING_PPF << 2 | VC4_SCALING_PPF:
237 		return SCALER_CTL0_SCL_H_PPF_V_PPF;
238 	case VC4_SCALING_TPZ << 2 | VC4_SCALING_PPF:
239 		return SCALER_CTL0_SCL_H_TPZ_V_PPF;
240 	case VC4_SCALING_PPF << 2 | VC4_SCALING_TPZ:
241 		return SCALER_CTL0_SCL_H_PPF_V_TPZ;
242 	case VC4_SCALING_TPZ << 2 | VC4_SCALING_TPZ:
243 		return SCALER_CTL0_SCL_H_TPZ_V_TPZ;
244 	case VC4_SCALING_PPF << 2 | VC4_SCALING_NONE:
245 		return SCALER_CTL0_SCL_H_PPF_V_NONE;
246 	case VC4_SCALING_NONE << 2 | VC4_SCALING_PPF:
247 		return SCALER_CTL0_SCL_H_NONE_V_PPF;
248 	case VC4_SCALING_NONE << 2 | VC4_SCALING_TPZ:
249 		return SCALER_CTL0_SCL_H_NONE_V_TPZ;
250 	case VC4_SCALING_TPZ << 2 | VC4_SCALING_NONE:
251 		return SCALER_CTL0_SCL_H_TPZ_V_NONE;
252 	default:
253 	case VC4_SCALING_NONE << 2 | VC4_SCALING_NONE:
254 		/* The unity case is independently handled by
255 		 * SCALER_CTL0_UNITY.
256 		 */
257 		return 0;
258 	}
259 }
260 
261 static int vc4_plane_setup_clipping_and_scaling(struct drm_plane_state *state)
262 {
263 	struct drm_plane *plane = state->plane;
264 	struct vc4_plane_state *vc4_state = to_vc4_plane_state(state);
265 	struct drm_framebuffer *fb = state->fb;
266 	struct drm_gem_cma_object *bo = drm_fb_cma_get_gem_obj(fb, 0);
267 	u32 subpixel_src_mask = (1 << 16) - 1;
268 	u32 format = fb->format->format;
269 	int num_planes = fb->format->num_planes;
270 	u32 h_subsample = 1;
271 	u32 v_subsample = 1;
272 	int i;
273 
274 	for (i = 0; i < num_planes; i++)
275 		vc4_state->offsets[i] = bo->paddr + fb->offsets[i];
276 
277 	/* We don't support subpixel source positioning for scaling. */
278 	if ((state->src_x & subpixel_src_mask) ||
279 	    (state->src_y & subpixel_src_mask) ||
280 	    (state->src_w & subpixel_src_mask) ||
281 	    (state->src_h & subpixel_src_mask)) {
282 		return -EINVAL;
283 	}
284 
285 	vc4_state->src_x = state->src_x >> 16;
286 	vc4_state->src_y = state->src_y >> 16;
287 	vc4_state->src_w[0] = state->src_w >> 16;
288 	vc4_state->src_h[0] = state->src_h >> 16;
289 
290 	vc4_state->crtc_x = state->crtc_x;
291 	vc4_state->crtc_y = state->crtc_y;
292 	vc4_state->crtc_w = state->crtc_w;
293 	vc4_state->crtc_h = state->crtc_h;
294 
295 	vc4_state->x_scaling[0] = vc4_get_scaling_mode(vc4_state->src_w[0],
296 						       vc4_state->crtc_w);
297 	vc4_state->y_scaling[0] = vc4_get_scaling_mode(vc4_state->src_h[0],
298 						       vc4_state->crtc_h);
299 
300 	vc4_state->is_unity = (vc4_state->x_scaling[0] == VC4_SCALING_NONE &&
301 			       vc4_state->y_scaling[0] == VC4_SCALING_NONE);
302 
303 	if (num_planes > 1) {
304 		vc4_state->is_yuv = true;
305 
306 		h_subsample = drm_format_horz_chroma_subsampling(format);
307 		v_subsample = drm_format_vert_chroma_subsampling(format);
308 		vc4_state->src_w[1] = vc4_state->src_w[0] / h_subsample;
309 		vc4_state->src_h[1] = vc4_state->src_h[0] / v_subsample;
310 
311 		vc4_state->x_scaling[1] =
312 			vc4_get_scaling_mode(vc4_state->src_w[1],
313 					     vc4_state->crtc_w);
314 		vc4_state->y_scaling[1] =
315 			vc4_get_scaling_mode(vc4_state->src_h[1],
316 					     vc4_state->crtc_h);
317 
318 		/* YUV conversion requires that horizontal scaling be enabled,
319 		 * even on a plane that's otherwise 1:1. Looks like only PPF
320 		 * works in that case, so let's pick that one.
321 		 */
322 		if (vc4_state->is_unity)
323 			vc4_state->x_scaling[0] = VC4_SCALING_PPF;
324 	} else {
325 		vc4_state->x_scaling[1] = VC4_SCALING_NONE;
326 		vc4_state->y_scaling[1] = VC4_SCALING_NONE;
327 	}
328 
329 	/* No configuring scaling on the cursor plane, since it gets
330 	   non-vblank-synced updates, and scaling requires requires
331 	   LBM changes which have to be vblank-synced.
332 	 */
333 	if (plane->type == DRM_PLANE_TYPE_CURSOR && !vc4_state->is_unity)
334 		return -EINVAL;
335 
336 	/* Clamp the on-screen start x/y to 0.  The hardware doesn't
337 	 * support negative y, and negative x wastes bandwidth.
338 	 */
339 	if (vc4_state->crtc_x < 0) {
340 		for (i = 0; i < num_planes; i++) {
341 			u32 cpp = fb->format->cpp[i];
342 			u32 subs = ((i == 0) ? 1 : h_subsample);
343 
344 			vc4_state->offsets[i] += (cpp *
345 						  (-vc4_state->crtc_x) / subs);
346 		}
347 		vc4_state->src_w[0] += vc4_state->crtc_x;
348 		vc4_state->src_w[1] += vc4_state->crtc_x / h_subsample;
349 		vc4_state->crtc_x = 0;
350 	}
351 
352 	if (vc4_state->crtc_y < 0) {
353 		for (i = 0; i < num_planes; i++) {
354 			u32 subs = ((i == 0) ? 1 : v_subsample);
355 
356 			vc4_state->offsets[i] += (fb->pitches[i] *
357 						  (-vc4_state->crtc_y) / subs);
358 		}
359 		vc4_state->src_h[0] += vc4_state->crtc_y;
360 		vc4_state->src_h[1] += vc4_state->crtc_y / v_subsample;
361 		vc4_state->crtc_y = 0;
362 	}
363 
364 	return 0;
365 }
366 
367 static void vc4_write_tpz(struct vc4_plane_state *vc4_state, u32 src, u32 dst)
368 {
369 	u32 scale, recip;
370 
371 	scale = (1 << 16) * src / dst;
372 
373 	/* The specs note that while the reciprocal would be defined
374 	 * as (1<<32)/scale, ~0 is close enough.
375 	 */
376 	recip = ~0 / scale;
377 
378 	vc4_dlist_write(vc4_state,
379 			VC4_SET_FIELD(scale, SCALER_TPZ0_SCALE) |
380 			VC4_SET_FIELD(0, SCALER_TPZ0_IPHASE));
381 	vc4_dlist_write(vc4_state,
382 			VC4_SET_FIELD(recip, SCALER_TPZ1_RECIP));
383 }
384 
385 static void vc4_write_ppf(struct vc4_plane_state *vc4_state, u32 src, u32 dst)
386 {
387 	u32 scale = (1 << 16) * src / dst;
388 
389 	vc4_dlist_write(vc4_state,
390 			SCALER_PPF_AGC |
391 			VC4_SET_FIELD(scale, SCALER_PPF_SCALE) |
392 			VC4_SET_FIELD(0, SCALER_PPF_IPHASE));
393 }
394 
395 static u32 vc4_lbm_size(struct drm_plane_state *state)
396 {
397 	struct vc4_plane_state *vc4_state = to_vc4_plane_state(state);
398 	/* This is the worst case number.  One of the two sizes will
399 	 * be used depending on the scaling configuration.
400 	 */
401 	u32 pix_per_line = max(vc4_state->src_w[0], (u32)vc4_state->crtc_w);
402 	u32 lbm;
403 
404 	if (!vc4_state->is_yuv) {
405 		if (vc4_state->is_unity)
406 			return 0;
407 		else if (vc4_state->y_scaling[0] == VC4_SCALING_TPZ)
408 			lbm = pix_per_line * 8;
409 		else {
410 			/* In special cases, this multiplier might be 12. */
411 			lbm = pix_per_line * 16;
412 		}
413 	} else {
414 		/* There are cases for this going down to a multiplier
415 		 * of 2, but according to the firmware source, the
416 		 * table in the docs is somewhat wrong.
417 		 */
418 		lbm = pix_per_line * 16;
419 	}
420 
421 	lbm = roundup(lbm, 32);
422 
423 	return lbm;
424 }
425 
426 static void vc4_write_scaling_parameters(struct drm_plane_state *state,
427 					 int channel)
428 {
429 	struct vc4_plane_state *vc4_state = to_vc4_plane_state(state);
430 
431 	/* Ch0 H-PPF Word 0: Scaling Parameters */
432 	if (vc4_state->x_scaling[channel] == VC4_SCALING_PPF) {
433 		vc4_write_ppf(vc4_state,
434 			      vc4_state->src_w[channel], vc4_state->crtc_w);
435 	}
436 
437 	/* Ch0 V-PPF Words 0-1: Scaling Parameters, Context */
438 	if (vc4_state->y_scaling[channel] == VC4_SCALING_PPF) {
439 		vc4_write_ppf(vc4_state,
440 			      vc4_state->src_h[channel], vc4_state->crtc_h);
441 		vc4_dlist_write(vc4_state, 0xc0c0c0c0);
442 	}
443 
444 	/* Ch0 H-TPZ Words 0-1: Scaling Parameters, Recip */
445 	if (vc4_state->x_scaling[channel] == VC4_SCALING_TPZ) {
446 		vc4_write_tpz(vc4_state,
447 			      vc4_state->src_w[channel], vc4_state->crtc_w);
448 	}
449 
450 	/* Ch0 V-TPZ Words 0-2: Scaling Parameters, Recip, Context */
451 	if (vc4_state->y_scaling[channel] == VC4_SCALING_TPZ) {
452 		vc4_write_tpz(vc4_state,
453 			      vc4_state->src_h[channel], vc4_state->crtc_h);
454 		vc4_dlist_write(vc4_state, 0xc0c0c0c0);
455 	}
456 }
457 
458 /* Writes out a full display list for an active plane to the plane's
459  * private dlist state.
460  */
461 static int vc4_plane_mode_set(struct drm_plane *plane,
462 			      struct drm_plane_state *state)
463 {
464 	struct vc4_dev *vc4 = to_vc4_dev(plane->dev);
465 	struct vc4_plane_state *vc4_state = to_vc4_plane_state(state);
466 	struct drm_framebuffer *fb = state->fb;
467 	u32 ctl0_offset = vc4_state->dlist_count;
468 	const struct hvs_format *format = vc4_get_hvs_format(fb->format->format);
469 	u64 base_format_mod = fourcc_mod_broadcom_mod(fb->modifier);
470 	int num_planes = drm_format_num_planes(format->drm);
471 	bool mix_plane_alpha;
472 	bool covers_screen;
473 	u32 scl0, scl1, pitch0;
474 	u32 lbm_size, tiling;
475 	unsigned long irqflags;
476 	u32 hvs_format = format->hvs;
477 	int ret, i;
478 
479 	ret = vc4_plane_setup_clipping_and_scaling(state);
480 	if (ret)
481 		return ret;
482 
483 	/* Allocate the LBM memory that the HVS will use for temporary
484 	 * storage due to our scaling/format conversion.
485 	 */
486 	lbm_size = vc4_lbm_size(state);
487 	if (lbm_size) {
488 		if (!vc4_state->lbm.allocated) {
489 			spin_lock_irqsave(&vc4->hvs->mm_lock, irqflags);
490 			ret = drm_mm_insert_node_generic(&vc4->hvs->lbm_mm,
491 							 &vc4_state->lbm,
492 							 lbm_size, 32, 0, 0);
493 			spin_unlock_irqrestore(&vc4->hvs->mm_lock, irqflags);
494 		} else {
495 			WARN_ON_ONCE(lbm_size != vc4_state->lbm.size);
496 		}
497 	}
498 
499 	if (ret)
500 		return ret;
501 
502 	/* SCL1 is used for Cb/Cr scaling of planar formats.  For RGB
503 	 * and 4:4:4, scl1 should be set to scl0 so both channels of
504 	 * the scaler do the same thing.  For YUV, the Y plane needs
505 	 * to be put in channel 1 and Cb/Cr in channel 0, so we swap
506 	 * the scl fields here.
507 	 */
508 	if (num_planes == 1) {
509 		scl0 = vc4_get_scl_field(state, 0);
510 		scl1 = scl0;
511 	} else {
512 		scl0 = vc4_get_scl_field(state, 1);
513 		scl1 = vc4_get_scl_field(state, 0);
514 	}
515 
516 	switch (base_format_mod) {
517 	case DRM_FORMAT_MOD_LINEAR:
518 		tiling = SCALER_CTL0_TILING_LINEAR;
519 		pitch0 = VC4_SET_FIELD(fb->pitches[0], SCALER_SRC_PITCH);
520 		break;
521 
522 	case DRM_FORMAT_MOD_BROADCOM_VC4_T_TILED: {
523 		/* For T-tiled, the FB pitch is "how many bytes from
524 		 * one row to the next, such that pitch * tile_h ==
525 		 * tile_size * tiles_per_row."
526 		 */
527 		u32 tile_size_shift = 12; /* T tiles are 4kb */
528 		u32 tile_h_shift = 5; /* 16 and 32bpp are 32 pixels high */
529 		u32 tiles_w = fb->pitches[0] >> (tile_size_shift - tile_h_shift);
530 
531 		tiling = SCALER_CTL0_TILING_256B_OR_T;
532 
533 		pitch0 = (VC4_SET_FIELD(0, SCALER_PITCH0_TILE_Y_OFFSET) |
534 			  VC4_SET_FIELD(0, SCALER_PITCH0_TILE_WIDTH_L) |
535 			  VC4_SET_FIELD(tiles_w, SCALER_PITCH0_TILE_WIDTH_R));
536 		break;
537 	}
538 
539 	case DRM_FORMAT_MOD_BROADCOM_SAND64:
540 	case DRM_FORMAT_MOD_BROADCOM_SAND128:
541 	case DRM_FORMAT_MOD_BROADCOM_SAND256: {
542 		uint32_t param = fourcc_mod_broadcom_param(fb->modifier);
543 
544 		/* Column-based NV12 or RGBA.
545 		 */
546 		if (fb->format->num_planes > 1) {
547 			if (hvs_format != HVS_PIXEL_FORMAT_YCBCR_YUV420_2PLANE) {
548 				DRM_DEBUG_KMS("SAND format only valid for NV12/21");
549 				return -EINVAL;
550 			}
551 			hvs_format = HVS_PIXEL_FORMAT_H264;
552 		} else {
553 			if (base_format_mod == DRM_FORMAT_MOD_BROADCOM_SAND256) {
554 				DRM_DEBUG_KMS("SAND256 format only valid for H.264");
555 				return -EINVAL;
556 			}
557 		}
558 
559 		switch (base_format_mod) {
560 		case DRM_FORMAT_MOD_BROADCOM_SAND64:
561 			tiling = SCALER_CTL0_TILING_64B;
562 			break;
563 		case DRM_FORMAT_MOD_BROADCOM_SAND128:
564 			tiling = SCALER_CTL0_TILING_128B;
565 			break;
566 		case DRM_FORMAT_MOD_BROADCOM_SAND256:
567 			tiling = SCALER_CTL0_TILING_256B_OR_T;
568 			break;
569 		default:
570 			break;
571 		}
572 
573 		if (param > SCALER_TILE_HEIGHT_MASK) {
574 			DRM_DEBUG_KMS("SAND height too large (%d)\n", param);
575 			return -EINVAL;
576 		}
577 
578 		pitch0 = VC4_SET_FIELD(param, SCALER_TILE_HEIGHT);
579 		break;
580 	}
581 
582 	default:
583 		DRM_DEBUG_KMS("Unsupported FB tiling flag 0x%16llx",
584 			      (long long)fb->modifier);
585 		return -EINVAL;
586 	}
587 
588 	/* Control word */
589 	vc4_dlist_write(vc4_state,
590 			SCALER_CTL0_VALID |
591 			VC4_SET_FIELD(SCALER_CTL0_RGBA_EXPAND_ROUND, SCALER_CTL0_RGBA_EXPAND) |
592 			(format->pixel_order << SCALER_CTL0_ORDER_SHIFT) |
593 			(hvs_format << SCALER_CTL0_PIXEL_FORMAT_SHIFT) |
594 			VC4_SET_FIELD(tiling, SCALER_CTL0_TILING) |
595 			(vc4_state->is_unity ? SCALER_CTL0_UNITY : 0) |
596 			VC4_SET_FIELD(scl0, SCALER_CTL0_SCL0) |
597 			VC4_SET_FIELD(scl1, SCALER_CTL0_SCL1));
598 
599 	/* Position Word 0: Image Positions and Alpha Value */
600 	vc4_state->pos0_offset = vc4_state->dlist_count;
601 	vc4_dlist_write(vc4_state,
602 			VC4_SET_FIELD(state->alpha >> 8, SCALER_POS0_FIXED_ALPHA) |
603 			VC4_SET_FIELD(vc4_state->crtc_x, SCALER_POS0_START_X) |
604 			VC4_SET_FIELD(vc4_state->crtc_y, SCALER_POS0_START_Y));
605 
606 	/* Position Word 1: Scaled Image Dimensions. */
607 	if (!vc4_state->is_unity) {
608 		vc4_dlist_write(vc4_state,
609 				VC4_SET_FIELD(vc4_state->crtc_w,
610 					      SCALER_POS1_SCL_WIDTH) |
611 				VC4_SET_FIELD(vc4_state->crtc_h,
612 					      SCALER_POS1_SCL_HEIGHT));
613 	}
614 
615 	/* Don't waste cycles mixing with plane alpha if the set alpha
616 	 * is opaque or there is no per-pixel alpha information.
617 	 * In any case we use the alpha property value as the fixed alpha.
618 	 */
619 	mix_plane_alpha = state->alpha != DRM_BLEND_ALPHA_OPAQUE &&
620 			  fb->format->has_alpha;
621 
622 	/* Position Word 2: Source Image Size, Alpha */
623 	vc4_state->pos2_offset = vc4_state->dlist_count;
624 	vc4_dlist_write(vc4_state,
625 			VC4_SET_FIELD(fb->format->has_alpha ?
626 				      SCALER_POS2_ALPHA_MODE_PIPELINE :
627 				      SCALER_POS2_ALPHA_MODE_FIXED,
628 				      SCALER_POS2_ALPHA_MODE) |
629 			(mix_plane_alpha ? SCALER_POS2_ALPHA_MIX : 0) |
630 			(fb->format->has_alpha ? SCALER_POS2_ALPHA_PREMULT : 0) |
631 			VC4_SET_FIELD(vc4_state->src_w[0], SCALER_POS2_WIDTH) |
632 			VC4_SET_FIELD(vc4_state->src_h[0], SCALER_POS2_HEIGHT));
633 
634 	/* Position Word 3: Context.  Written by the HVS. */
635 	vc4_dlist_write(vc4_state, 0xc0c0c0c0);
636 
637 
638 	/* Pointer Word 0/1/2: RGB / Y / Cb / Cr Pointers
639 	 *
640 	 * The pointers may be any byte address.
641 	 */
642 	vc4_state->ptr0_offset = vc4_state->dlist_count;
643 	for (i = 0; i < num_planes; i++)
644 		vc4_dlist_write(vc4_state, vc4_state->offsets[i]);
645 
646 	/* Pointer Context Word 0/1/2: Written by the HVS */
647 	for (i = 0; i < num_planes; i++)
648 		vc4_dlist_write(vc4_state, 0xc0c0c0c0);
649 
650 	/* Pitch word 0 */
651 	vc4_dlist_write(vc4_state, pitch0);
652 
653 	/* Pitch word 1/2 */
654 	for (i = 1; i < num_planes; i++) {
655 		if (hvs_format != HVS_PIXEL_FORMAT_H264) {
656 			vc4_dlist_write(vc4_state,
657 					VC4_SET_FIELD(fb->pitches[i],
658 						      SCALER_SRC_PITCH));
659 		} else {
660 			vc4_dlist_write(vc4_state, pitch0);
661 		}
662 	}
663 
664 	/* Colorspace conversion words */
665 	if (vc4_state->is_yuv) {
666 		vc4_dlist_write(vc4_state, SCALER_CSC0_ITR_R_601_5);
667 		vc4_dlist_write(vc4_state, SCALER_CSC1_ITR_R_601_5);
668 		vc4_dlist_write(vc4_state, SCALER_CSC2_ITR_R_601_5);
669 	}
670 
671 	if (vc4_state->x_scaling[0] != VC4_SCALING_NONE ||
672 	    vc4_state->x_scaling[1] != VC4_SCALING_NONE ||
673 	    vc4_state->y_scaling[0] != VC4_SCALING_NONE ||
674 	    vc4_state->y_scaling[1] != VC4_SCALING_NONE) {
675 		/* LBM Base Address. */
676 		if (vc4_state->y_scaling[0] != VC4_SCALING_NONE ||
677 		    vc4_state->y_scaling[1] != VC4_SCALING_NONE) {
678 			vc4_dlist_write(vc4_state, vc4_state->lbm.start);
679 		}
680 
681 		if (num_planes > 1) {
682 			/* Emit Cb/Cr as channel 0 and Y as channel
683 			 * 1. This matches how we set up scl0/scl1
684 			 * above.
685 			 */
686 			vc4_write_scaling_parameters(state, 1);
687 		}
688 		vc4_write_scaling_parameters(state, 0);
689 
690 		/* If any PPF setup was done, then all the kernel
691 		 * pointers get uploaded.
692 		 */
693 		if (vc4_state->x_scaling[0] == VC4_SCALING_PPF ||
694 		    vc4_state->y_scaling[0] == VC4_SCALING_PPF ||
695 		    vc4_state->x_scaling[1] == VC4_SCALING_PPF ||
696 		    vc4_state->y_scaling[1] == VC4_SCALING_PPF) {
697 			u32 kernel = VC4_SET_FIELD(vc4->hvs->mitchell_netravali_filter.start,
698 						   SCALER_PPF_KERNEL_OFFSET);
699 
700 			/* HPPF plane 0 */
701 			vc4_dlist_write(vc4_state, kernel);
702 			/* VPPF plane 0 */
703 			vc4_dlist_write(vc4_state, kernel);
704 			/* HPPF plane 1 */
705 			vc4_dlist_write(vc4_state, kernel);
706 			/* VPPF plane 1 */
707 			vc4_dlist_write(vc4_state, kernel);
708 		}
709 	}
710 
711 	vc4_state->dlist[ctl0_offset] |=
712 		VC4_SET_FIELD(vc4_state->dlist_count, SCALER_CTL0_SIZE);
713 
714 	/* crtc_* are already clipped coordinates. */
715 	covers_screen = vc4_state->crtc_x == 0 && vc4_state->crtc_y == 0 &&
716 			vc4_state->crtc_w == state->crtc->mode.hdisplay &&
717 			vc4_state->crtc_h == state->crtc->mode.vdisplay;
718 	/* Background fill might be necessary when the plane has per-pixel
719 	 * alpha content or a non-opaque plane alpha and could blend from the
720 	 * background or does not cover the entire screen.
721 	 */
722 	vc4_state->needs_bg_fill = fb->format->has_alpha || !covers_screen ||
723 				   state->alpha != DRM_BLEND_ALPHA_OPAQUE;
724 
725 	return 0;
726 }
727 
728 /* If a modeset involves changing the setup of a plane, the atomic
729  * infrastructure will call this to validate a proposed plane setup.
730  * However, if a plane isn't getting updated, this (and the
731  * corresponding vc4_plane_atomic_update) won't get called.  Thus, we
732  * compute the dlist here and have all active plane dlists get updated
733  * in the CRTC's flush.
734  */
735 static int vc4_plane_atomic_check(struct drm_plane *plane,
736 				  struct drm_plane_state *state)
737 {
738 	struct vc4_plane_state *vc4_state = to_vc4_plane_state(state);
739 
740 	vc4_state->dlist_count = 0;
741 
742 	if (plane_enabled(state))
743 		return vc4_plane_mode_set(plane, state);
744 	else
745 		return 0;
746 }
747 
748 static void vc4_plane_atomic_update(struct drm_plane *plane,
749 				    struct drm_plane_state *old_state)
750 {
751 	/* No contents here.  Since we don't know where in the CRTC's
752 	 * dlist we should be stored, our dlist is uploaded to the
753 	 * hardware with vc4_plane_write_dlist() at CRTC atomic_flush
754 	 * time.
755 	 */
756 }
757 
758 u32 vc4_plane_write_dlist(struct drm_plane *plane, u32 __iomem *dlist)
759 {
760 	struct vc4_plane_state *vc4_state = to_vc4_plane_state(plane->state);
761 	int i;
762 
763 	vc4_state->hw_dlist = dlist;
764 
765 	/* Can't memcpy_toio() because it needs to be 32-bit writes. */
766 	for (i = 0; i < vc4_state->dlist_count; i++)
767 		writel(vc4_state->dlist[i], &dlist[i]);
768 
769 	return vc4_state->dlist_count;
770 }
771 
772 u32 vc4_plane_dlist_size(const struct drm_plane_state *state)
773 {
774 	const struct vc4_plane_state *vc4_state =
775 		container_of(state, typeof(*vc4_state), base);
776 
777 	return vc4_state->dlist_count;
778 }
779 
780 /* Updates the plane to immediately (well, once the FIFO needs
781  * refilling) scan out from at a new framebuffer.
782  */
783 void vc4_plane_async_set_fb(struct drm_plane *plane, struct drm_framebuffer *fb)
784 {
785 	struct vc4_plane_state *vc4_state = to_vc4_plane_state(plane->state);
786 	struct drm_gem_cma_object *bo = drm_fb_cma_get_gem_obj(fb, 0);
787 	uint32_t addr;
788 
789 	/* We're skipping the address adjustment for negative origin,
790 	 * because this is only called on the primary plane.
791 	 */
792 	WARN_ON_ONCE(plane->state->crtc_x < 0 || plane->state->crtc_y < 0);
793 	addr = bo->paddr + fb->offsets[0];
794 
795 	/* Write the new address into the hardware immediately.  The
796 	 * scanout will start from this address as soon as the FIFO
797 	 * needs to refill with pixels.
798 	 */
799 	writel(addr, &vc4_state->hw_dlist[vc4_state->ptr0_offset]);
800 
801 	/* Also update the CPU-side dlist copy, so that any later
802 	 * atomic updates that don't do a new modeset on our plane
803 	 * also use our updated address.
804 	 */
805 	vc4_state->dlist[vc4_state->ptr0_offset] = addr;
806 }
807 
808 static void vc4_plane_atomic_async_update(struct drm_plane *plane,
809 					  struct drm_plane_state *state)
810 {
811 	struct vc4_plane_state *vc4_state = to_vc4_plane_state(plane->state);
812 
813 	if (plane->state->fb != state->fb) {
814 		vc4_plane_async_set_fb(plane, state->fb);
815 		drm_atomic_set_fb_for_plane(plane->state, state->fb);
816 	}
817 
818 	/* Set the cursor's position on the screen.  This is the
819 	 * expected change from the drm_mode_cursor_universal()
820 	 * helper.
821 	 */
822 	plane->state->crtc_x = state->crtc_x;
823 	plane->state->crtc_y = state->crtc_y;
824 
825 	/* Allow changing the start position within the cursor BO, if
826 	 * that matters.
827 	 */
828 	plane->state->src_x = state->src_x;
829 	plane->state->src_y = state->src_y;
830 
831 	/* Update the display list based on the new crtc_x/y. */
832 	vc4_plane_atomic_check(plane, plane->state);
833 
834 	/* Note that we can't just call vc4_plane_write_dlist()
835 	 * because that would smash the context data that the HVS is
836 	 * currently using.
837 	 */
838 	writel(vc4_state->dlist[vc4_state->pos0_offset],
839 	       &vc4_state->hw_dlist[vc4_state->pos0_offset]);
840 	writel(vc4_state->dlist[vc4_state->pos2_offset],
841 	       &vc4_state->hw_dlist[vc4_state->pos2_offset]);
842 	writel(vc4_state->dlist[vc4_state->ptr0_offset],
843 	       &vc4_state->hw_dlist[vc4_state->ptr0_offset]);
844 }
845 
846 static int vc4_plane_atomic_async_check(struct drm_plane *plane,
847 					struct drm_plane_state *state)
848 {
849 	/* No configuring new scaling in the fast path. */
850 	if (plane->state->crtc_w != state->crtc_w ||
851 	    plane->state->crtc_h != state->crtc_h ||
852 	    plane->state->src_w != state->src_w ||
853 	    plane->state->src_h != state->src_h)
854 		return -EINVAL;
855 
856 	return 0;
857 }
858 
859 static int vc4_prepare_fb(struct drm_plane *plane,
860 			  struct drm_plane_state *state)
861 {
862 	struct vc4_bo *bo;
863 	struct dma_fence *fence;
864 	int ret;
865 
866 	if (!state->fb)
867 		return 0;
868 
869 	bo = to_vc4_bo(&drm_fb_cma_get_gem_obj(state->fb, 0)->base);
870 
871 	fence = reservation_object_get_excl_rcu(bo->resv);
872 	drm_atomic_set_fence_for_plane(state, fence);
873 
874 	if (plane->state->fb == state->fb)
875 		return 0;
876 
877 	ret = vc4_bo_inc_usecnt(bo);
878 	if (ret)
879 		return ret;
880 
881 	return 0;
882 }
883 
884 static void vc4_cleanup_fb(struct drm_plane *plane,
885 			   struct drm_plane_state *state)
886 {
887 	struct vc4_bo *bo;
888 
889 	if (plane->state->fb == state->fb || !state->fb)
890 		return;
891 
892 	bo = to_vc4_bo(&drm_fb_cma_get_gem_obj(state->fb, 0)->base);
893 	vc4_bo_dec_usecnt(bo);
894 }
895 
896 static const struct drm_plane_helper_funcs vc4_plane_helper_funcs = {
897 	.atomic_check = vc4_plane_atomic_check,
898 	.atomic_update = vc4_plane_atomic_update,
899 	.prepare_fb = vc4_prepare_fb,
900 	.cleanup_fb = vc4_cleanup_fb,
901 	.atomic_async_check = vc4_plane_atomic_async_check,
902 	.atomic_async_update = vc4_plane_atomic_async_update,
903 };
904 
905 static void vc4_plane_destroy(struct drm_plane *plane)
906 {
907 	drm_plane_helper_disable(plane, NULL);
908 	drm_plane_cleanup(plane);
909 }
910 
911 static bool vc4_format_mod_supported(struct drm_plane *plane,
912 				     uint32_t format,
913 				     uint64_t modifier)
914 {
915 	/* Support T_TILING for RGB formats only. */
916 	switch (format) {
917 	case DRM_FORMAT_XRGB8888:
918 	case DRM_FORMAT_ARGB8888:
919 	case DRM_FORMAT_ABGR8888:
920 	case DRM_FORMAT_XBGR8888:
921 	case DRM_FORMAT_RGB565:
922 	case DRM_FORMAT_BGR565:
923 	case DRM_FORMAT_ARGB1555:
924 	case DRM_FORMAT_XRGB1555:
925 		switch (fourcc_mod_broadcom_mod(modifier)) {
926 		case DRM_FORMAT_MOD_LINEAR:
927 		case DRM_FORMAT_MOD_BROADCOM_VC4_T_TILED:
928 		case DRM_FORMAT_MOD_BROADCOM_SAND64:
929 		case DRM_FORMAT_MOD_BROADCOM_SAND128:
930 			return true;
931 		default:
932 			return false;
933 		}
934 	case DRM_FORMAT_NV12:
935 	case DRM_FORMAT_NV21:
936 		switch (fourcc_mod_broadcom_mod(modifier)) {
937 		case DRM_FORMAT_MOD_LINEAR:
938 		case DRM_FORMAT_MOD_BROADCOM_SAND64:
939 		case DRM_FORMAT_MOD_BROADCOM_SAND128:
940 		case DRM_FORMAT_MOD_BROADCOM_SAND256:
941 			return true;
942 		default:
943 			return false;
944 		}
945 	case DRM_FORMAT_YUV422:
946 	case DRM_FORMAT_YVU422:
947 	case DRM_FORMAT_YUV420:
948 	case DRM_FORMAT_YVU420:
949 	case DRM_FORMAT_NV16:
950 	case DRM_FORMAT_NV61:
951 	default:
952 		return (modifier == DRM_FORMAT_MOD_LINEAR);
953 	}
954 }
955 
956 static const struct drm_plane_funcs vc4_plane_funcs = {
957 	.update_plane = drm_atomic_helper_update_plane,
958 	.disable_plane = drm_atomic_helper_disable_plane,
959 	.destroy = vc4_plane_destroy,
960 	.set_property = NULL,
961 	.reset = vc4_plane_reset,
962 	.atomic_duplicate_state = vc4_plane_duplicate_state,
963 	.atomic_destroy_state = vc4_plane_destroy_state,
964 	.format_mod_supported = vc4_format_mod_supported,
965 };
966 
967 struct drm_plane *vc4_plane_init(struct drm_device *dev,
968 				 enum drm_plane_type type)
969 {
970 	struct drm_plane *plane = NULL;
971 	struct vc4_plane *vc4_plane;
972 	u32 formats[ARRAY_SIZE(hvs_formats)];
973 	u32 num_formats = 0;
974 	int ret = 0;
975 	unsigned i;
976 	static const uint64_t modifiers[] = {
977 		DRM_FORMAT_MOD_BROADCOM_VC4_T_TILED,
978 		DRM_FORMAT_MOD_BROADCOM_SAND128,
979 		DRM_FORMAT_MOD_BROADCOM_SAND64,
980 		DRM_FORMAT_MOD_BROADCOM_SAND256,
981 		DRM_FORMAT_MOD_LINEAR,
982 		DRM_FORMAT_MOD_INVALID
983 	};
984 
985 	vc4_plane = devm_kzalloc(dev->dev, sizeof(*vc4_plane),
986 				 GFP_KERNEL);
987 	if (!vc4_plane)
988 		return ERR_PTR(-ENOMEM);
989 
990 	for (i = 0; i < ARRAY_SIZE(hvs_formats); i++) {
991 		/* Don't allow YUV in cursor planes, since that means
992 		 * tuning on the scaler, which we don't allow for the
993 		 * cursor.
994 		 */
995 		if (type != DRM_PLANE_TYPE_CURSOR ||
996 		    hvs_formats[i].hvs < HVS_PIXEL_FORMAT_YCBCR_YUV420_3PLANE) {
997 			formats[num_formats++] = hvs_formats[i].drm;
998 		}
999 	}
1000 	plane = &vc4_plane->base;
1001 	ret = drm_universal_plane_init(dev, plane, 0,
1002 				       &vc4_plane_funcs,
1003 				       formats, num_formats,
1004 				       modifiers, type, NULL);
1005 
1006 	drm_plane_helper_add(plane, &vc4_plane_helper_funcs);
1007 
1008 	drm_plane_create_alpha_property(plane);
1009 
1010 	return plane;
1011 }
1012