xref: /openbmc/linux/drivers/gpu/drm/vc4/vc4_plane.c (revision 05cf4fe738242183f1237f1b3a28b4479348c0a1)
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 #include <drm/drm_atomic_uapi.h>
26 
27 #include "uapi/drm/vc4_drm.h"
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 } hvs_formats[] = {
36 	{
37 		.drm = DRM_FORMAT_XRGB8888, .hvs = HVS_PIXEL_FORMAT_RGBA8888,
38 		.pixel_order = HVS_PIXEL_ORDER_ABGR,
39 	},
40 	{
41 		.drm = DRM_FORMAT_ARGB8888, .hvs = HVS_PIXEL_FORMAT_RGBA8888,
42 		.pixel_order = HVS_PIXEL_ORDER_ABGR,
43 	},
44 	{
45 		.drm = DRM_FORMAT_ABGR8888, .hvs = HVS_PIXEL_FORMAT_RGBA8888,
46 		.pixel_order = HVS_PIXEL_ORDER_ARGB,
47 	},
48 	{
49 		.drm = DRM_FORMAT_XBGR8888, .hvs = HVS_PIXEL_FORMAT_RGBA8888,
50 		.pixel_order = HVS_PIXEL_ORDER_ARGB,
51 	},
52 	{
53 		.drm = DRM_FORMAT_RGB565, .hvs = HVS_PIXEL_FORMAT_RGB565,
54 		.pixel_order = HVS_PIXEL_ORDER_XRGB,
55 	},
56 	{
57 		.drm = DRM_FORMAT_BGR565, .hvs = HVS_PIXEL_FORMAT_RGB565,
58 		.pixel_order = HVS_PIXEL_ORDER_XBGR,
59 	},
60 	{
61 		.drm = DRM_FORMAT_ARGB1555, .hvs = HVS_PIXEL_FORMAT_RGBA5551,
62 		.pixel_order = HVS_PIXEL_ORDER_ABGR,
63 	},
64 	{
65 		.drm = DRM_FORMAT_XRGB1555, .hvs = HVS_PIXEL_FORMAT_RGBA5551,
66 		.pixel_order = HVS_PIXEL_ORDER_ABGR,
67 	},
68 	{
69 		.drm = DRM_FORMAT_RGB888, .hvs = HVS_PIXEL_FORMAT_RGB888,
70 		.pixel_order = HVS_PIXEL_ORDER_XRGB,
71 	},
72 	{
73 		.drm = DRM_FORMAT_BGR888, .hvs = HVS_PIXEL_FORMAT_RGB888,
74 		.pixel_order = HVS_PIXEL_ORDER_XBGR,
75 	},
76 	{
77 		.drm = DRM_FORMAT_YUV422,
78 		.hvs = HVS_PIXEL_FORMAT_YCBCR_YUV422_3PLANE,
79 		.pixel_order = HVS_PIXEL_ORDER_XYCBCR,
80 	},
81 	{
82 		.drm = DRM_FORMAT_YVU422,
83 		.hvs = HVS_PIXEL_FORMAT_YCBCR_YUV422_3PLANE,
84 		.pixel_order = HVS_PIXEL_ORDER_XYCRCB,
85 	},
86 	{
87 		.drm = DRM_FORMAT_YUV420,
88 		.hvs = HVS_PIXEL_FORMAT_YCBCR_YUV420_3PLANE,
89 		.pixel_order = HVS_PIXEL_ORDER_XYCBCR,
90 	},
91 	{
92 		.drm = DRM_FORMAT_YVU420,
93 		.hvs = HVS_PIXEL_FORMAT_YCBCR_YUV420_3PLANE,
94 		.pixel_order = HVS_PIXEL_ORDER_XYCRCB,
95 	},
96 	{
97 		.drm = DRM_FORMAT_NV12,
98 		.hvs = HVS_PIXEL_FORMAT_YCBCR_YUV420_2PLANE,
99 		.pixel_order = HVS_PIXEL_ORDER_XYCBCR,
100 	},
101 	{
102 		.drm = DRM_FORMAT_NV21,
103 		.hvs = HVS_PIXEL_FORMAT_YCBCR_YUV420_2PLANE,
104 		.pixel_order = HVS_PIXEL_ORDER_XYCRCB,
105 	},
106 	{
107 		.drm = DRM_FORMAT_NV16,
108 		.hvs = HVS_PIXEL_FORMAT_YCBCR_YUV422_2PLANE,
109 		.pixel_order = HVS_PIXEL_ORDER_XYCBCR,
110 	},
111 	{
112 		.drm = DRM_FORMAT_NV61,
113 		.hvs = HVS_PIXEL_FORMAT_YCBCR_YUV422_2PLANE,
114 		.pixel_order = HVS_PIXEL_ORDER_XYCRCB,
115 	},
116 };
117 
118 static const struct hvs_format *vc4_get_hvs_format(u32 drm_format)
119 {
120 	unsigned i;
121 
122 	for (i = 0; i < ARRAY_SIZE(hvs_formats); i++) {
123 		if (hvs_formats[i].drm == drm_format)
124 			return &hvs_formats[i];
125 	}
126 
127 	return NULL;
128 }
129 
130 static enum vc4_scaling_mode vc4_get_scaling_mode(u32 src, u32 dst)
131 {
132 	if (dst > src)
133 		return VC4_SCALING_PPF;
134 	else if (dst < src)
135 		return VC4_SCALING_TPZ;
136 	else
137 		return VC4_SCALING_NONE;
138 }
139 
140 static bool plane_enabled(struct drm_plane_state *state)
141 {
142 	return state->fb && state->crtc;
143 }
144 
145 static struct drm_plane_state *vc4_plane_duplicate_state(struct drm_plane *plane)
146 {
147 	struct vc4_plane_state *vc4_state;
148 
149 	if (WARN_ON(!plane->state))
150 		return NULL;
151 
152 	vc4_state = kmemdup(plane->state, sizeof(*vc4_state), GFP_KERNEL);
153 	if (!vc4_state)
154 		return NULL;
155 
156 	memset(&vc4_state->lbm, 0, sizeof(vc4_state->lbm));
157 
158 	__drm_atomic_helper_plane_duplicate_state(plane, &vc4_state->base);
159 
160 	if (vc4_state->dlist) {
161 		vc4_state->dlist = kmemdup(vc4_state->dlist,
162 					   vc4_state->dlist_count * 4,
163 					   GFP_KERNEL);
164 		if (!vc4_state->dlist) {
165 			kfree(vc4_state);
166 			return NULL;
167 		}
168 		vc4_state->dlist_size = vc4_state->dlist_count;
169 	}
170 
171 	return &vc4_state->base;
172 }
173 
174 static void vc4_plane_destroy_state(struct drm_plane *plane,
175 				    struct drm_plane_state *state)
176 {
177 	struct vc4_dev *vc4 = to_vc4_dev(plane->dev);
178 	struct vc4_plane_state *vc4_state = to_vc4_plane_state(state);
179 
180 	if (vc4_state->lbm.allocated) {
181 		unsigned long irqflags;
182 
183 		spin_lock_irqsave(&vc4->hvs->mm_lock, irqflags);
184 		drm_mm_remove_node(&vc4_state->lbm);
185 		spin_unlock_irqrestore(&vc4->hvs->mm_lock, irqflags);
186 	}
187 
188 	kfree(vc4_state->dlist);
189 	__drm_atomic_helper_plane_destroy_state(&vc4_state->base);
190 	kfree(state);
191 }
192 
193 /* Called during init to allocate the plane's atomic state. */
194 static void vc4_plane_reset(struct drm_plane *plane)
195 {
196 	struct vc4_plane_state *vc4_state;
197 
198 	WARN_ON(plane->state);
199 
200 	vc4_state = kzalloc(sizeof(*vc4_state), GFP_KERNEL);
201 	if (!vc4_state)
202 		return;
203 
204 	__drm_atomic_helper_plane_reset(plane, &vc4_state->base);
205 }
206 
207 static void vc4_dlist_write(struct vc4_plane_state *vc4_state, u32 val)
208 {
209 	if (vc4_state->dlist_count == vc4_state->dlist_size) {
210 		u32 new_size = max(4u, vc4_state->dlist_count * 2);
211 		u32 *new_dlist = kmalloc_array(new_size, 4, GFP_KERNEL);
212 
213 		if (!new_dlist)
214 			return;
215 		memcpy(new_dlist, vc4_state->dlist, vc4_state->dlist_count * 4);
216 
217 		kfree(vc4_state->dlist);
218 		vc4_state->dlist = new_dlist;
219 		vc4_state->dlist_size = new_size;
220 	}
221 
222 	vc4_state->dlist[vc4_state->dlist_count++] = val;
223 }
224 
225 /* Returns the scl0/scl1 field based on whether the dimensions need to
226  * be up/down/non-scaled.
227  *
228  * This is a replication of a table from the spec.
229  */
230 static u32 vc4_get_scl_field(struct drm_plane_state *state, int plane)
231 {
232 	struct vc4_plane_state *vc4_state = to_vc4_plane_state(state);
233 
234 	switch (vc4_state->x_scaling[plane] << 2 | vc4_state->y_scaling[plane]) {
235 	case VC4_SCALING_PPF << 2 | VC4_SCALING_PPF:
236 		return SCALER_CTL0_SCL_H_PPF_V_PPF;
237 	case VC4_SCALING_TPZ << 2 | VC4_SCALING_PPF:
238 		return SCALER_CTL0_SCL_H_TPZ_V_PPF;
239 	case VC4_SCALING_PPF << 2 | VC4_SCALING_TPZ:
240 		return SCALER_CTL0_SCL_H_PPF_V_TPZ;
241 	case VC4_SCALING_TPZ << 2 | VC4_SCALING_TPZ:
242 		return SCALER_CTL0_SCL_H_TPZ_V_TPZ;
243 	case VC4_SCALING_PPF << 2 | VC4_SCALING_NONE:
244 		return SCALER_CTL0_SCL_H_PPF_V_NONE;
245 	case VC4_SCALING_NONE << 2 | VC4_SCALING_PPF:
246 		return SCALER_CTL0_SCL_H_NONE_V_PPF;
247 	case VC4_SCALING_NONE << 2 | VC4_SCALING_TPZ:
248 		return SCALER_CTL0_SCL_H_NONE_V_TPZ;
249 	case VC4_SCALING_TPZ << 2 | VC4_SCALING_NONE:
250 		return SCALER_CTL0_SCL_H_TPZ_V_NONE;
251 	default:
252 	case VC4_SCALING_NONE << 2 | VC4_SCALING_NONE:
253 		/* The unity case is independently handled by
254 		 * SCALER_CTL0_UNITY.
255 		 */
256 		return 0;
257 	}
258 }
259 
260 static int vc4_plane_setup_clipping_and_scaling(struct drm_plane_state *state)
261 {
262 	struct drm_plane *plane = state->plane;
263 	struct vc4_plane_state *vc4_state = to_vc4_plane_state(state);
264 	struct drm_framebuffer *fb = state->fb;
265 	struct drm_gem_cma_object *bo = drm_fb_cma_get_gem_obj(fb, 0);
266 	u32 subpixel_src_mask = (1 << 16) - 1;
267 	u32 format = fb->format->format;
268 	int num_planes = fb->format->num_planes;
269 	u32 h_subsample = 1;
270 	u32 v_subsample = 1;
271 	int i;
272 
273 	for (i = 0; i < num_planes; i++)
274 		vc4_state->offsets[i] = bo->paddr + fb->offsets[i];
275 
276 	/* We don't support subpixel source positioning for scaling. */
277 	if ((state->src_x & subpixel_src_mask) ||
278 	    (state->src_y & subpixel_src_mask) ||
279 	    (state->src_w & subpixel_src_mask) ||
280 	    (state->src_h & subpixel_src_mask)) {
281 		return -EINVAL;
282 	}
283 
284 	vc4_state->src_x = state->src_x >> 16;
285 	vc4_state->src_y = state->src_y >> 16;
286 	vc4_state->src_w[0] = state->src_w >> 16;
287 	vc4_state->src_h[0] = state->src_h >> 16;
288 
289 	vc4_state->crtc_x = state->crtc_x;
290 	vc4_state->crtc_y = state->crtc_y;
291 	vc4_state->crtc_w = state->crtc_w;
292 	vc4_state->crtc_h = state->crtc_h;
293 
294 	vc4_state->x_scaling[0] = vc4_get_scaling_mode(vc4_state->src_w[0],
295 						       vc4_state->crtc_w);
296 	vc4_state->y_scaling[0] = vc4_get_scaling_mode(vc4_state->src_h[0],
297 						       vc4_state->crtc_h);
298 
299 	vc4_state->is_unity = (vc4_state->x_scaling[0] == VC4_SCALING_NONE &&
300 			       vc4_state->y_scaling[0] == VC4_SCALING_NONE);
301 
302 	if (num_planes > 1) {
303 		vc4_state->is_yuv = true;
304 
305 		h_subsample = drm_format_horz_chroma_subsampling(format);
306 		v_subsample = drm_format_vert_chroma_subsampling(format);
307 		vc4_state->src_w[1] = vc4_state->src_w[0] / h_subsample;
308 		vc4_state->src_h[1] = vc4_state->src_h[0] / v_subsample;
309 
310 		vc4_state->x_scaling[1] =
311 			vc4_get_scaling_mode(vc4_state->src_w[1],
312 					     vc4_state->crtc_w);
313 		vc4_state->y_scaling[1] =
314 			vc4_get_scaling_mode(vc4_state->src_h[1],
315 					     vc4_state->crtc_h);
316 
317 		/* YUV conversion requires that horizontal scaling be enabled,
318 		 * even on a plane that's otherwise 1:1. Looks like only PPF
319 		 * works in that case, so let's pick that one.
320 		 */
321 		if (vc4_state->is_unity)
322 			vc4_state->x_scaling[0] = VC4_SCALING_PPF;
323 	} else {
324 		vc4_state->x_scaling[1] = VC4_SCALING_NONE;
325 		vc4_state->y_scaling[1] = VC4_SCALING_NONE;
326 	}
327 
328 	/* No configuring scaling on the cursor plane, since it gets
329 	   non-vblank-synced updates, and scaling requires requires
330 	   LBM changes which have to be vblank-synced.
331 	 */
332 	if (plane->type == DRM_PLANE_TYPE_CURSOR && !vc4_state->is_unity)
333 		return -EINVAL;
334 
335 	/* Clamp the on-screen start x/y to 0.  The hardware doesn't
336 	 * support negative y, and negative x wastes bandwidth.
337 	 */
338 	if (vc4_state->crtc_x < 0) {
339 		for (i = 0; i < num_planes; i++) {
340 			u32 cpp = fb->format->cpp[i];
341 			u32 subs = ((i == 0) ? 1 : h_subsample);
342 
343 			vc4_state->offsets[i] += (cpp *
344 						  (-vc4_state->crtc_x) / subs);
345 		}
346 		vc4_state->src_w[0] += vc4_state->crtc_x;
347 		vc4_state->src_w[1] += vc4_state->crtc_x / h_subsample;
348 		vc4_state->crtc_x = 0;
349 	}
350 
351 	if (vc4_state->crtc_y < 0) {
352 		for (i = 0; i < num_planes; i++) {
353 			u32 subs = ((i == 0) ? 1 : v_subsample);
354 
355 			vc4_state->offsets[i] += (fb->pitches[i] *
356 						  (-vc4_state->crtc_y) / subs);
357 		}
358 		vc4_state->src_h[0] += vc4_state->crtc_y;
359 		vc4_state->src_h[1] += vc4_state->crtc_y / v_subsample;
360 		vc4_state->crtc_y = 0;
361 	}
362 
363 	return 0;
364 }
365 
366 static void vc4_write_tpz(struct vc4_plane_state *vc4_state, u32 src, u32 dst)
367 {
368 	u32 scale, recip;
369 
370 	scale = (1 << 16) * src / dst;
371 
372 	/* The specs note that while the reciprocal would be defined
373 	 * as (1<<32)/scale, ~0 is close enough.
374 	 */
375 	recip = ~0 / scale;
376 
377 	vc4_dlist_write(vc4_state,
378 			VC4_SET_FIELD(scale, SCALER_TPZ0_SCALE) |
379 			VC4_SET_FIELD(0, SCALER_TPZ0_IPHASE));
380 	vc4_dlist_write(vc4_state,
381 			VC4_SET_FIELD(recip, SCALER_TPZ1_RECIP));
382 }
383 
384 static void vc4_write_ppf(struct vc4_plane_state *vc4_state, u32 src, u32 dst)
385 {
386 	u32 scale = (1 << 16) * src / dst;
387 
388 	vc4_dlist_write(vc4_state,
389 			SCALER_PPF_AGC |
390 			VC4_SET_FIELD(scale, SCALER_PPF_SCALE) |
391 			VC4_SET_FIELD(0, SCALER_PPF_IPHASE));
392 }
393 
394 static u32 vc4_lbm_size(struct drm_plane_state *state)
395 {
396 	struct vc4_plane_state *vc4_state = to_vc4_plane_state(state);
397 	/* This is the worst case number.  One of the two sizes will
398 	 * be used depending on the scaling configuration.
399 	 */
400 	u32 pix_per_line = max(vc4_state->src_w[0], (u32)vc4_state->crtc_w);
401 	u32 lbm;
402 
403 	if (!vc4_state->is_yuv) {
404 		if (vc4_state->is_unity)
405 			return 0;
406 		else if (vc4_state->y_scaling[0] == VC4_SCALING_TPZ)
407 			lbm = pix_per_line * 8;
408 		else {
409 			/* In special cases, this multiplier might be 12. */
410 			lbm = pix_per_line * 16;
411 		}
412 	} else {
413 		/* There are cases for this going down to a multiplier
414 		 * of 2, but according to the firmware source, the
415 		 * table in the docs is somewhat wrong.
416 		 */
417 		lbm = pix_per_line * 16;
418 	}
419 
420 	lbm = roundup(lbm, 32);
421 
422 	return lbm;
423 }
424 
425 static void vc4_write_scaling_parameters(struct drm_plane_state *state,
426 					 int channel)
427 {
428 	struct vc4_plane_state *vc4_state = to_vc4_plane_state(state);
429 
430 	/* Ch0 H-PPF Word 0: Scaling Parameters */
431 	if (vc4_state->x_scaling[channel] == VC4_SCALING_PPF) {
432 		vc4_write_ppf(vc4_state,
433 			      vc4_state->src_w[channel], vc4_state->crtc_w);
434 	}
435 
436 	/* Ch0 V-PPF Words 0-1: Scaling Parameters, Context */
437 	if (vc4_state->y_scaling[channel] == VC4_SCALING_PPF) {
438 		vc4_write_ppf(vc4_state,
439 			      vc4_state->src_h[channel], vc4_state->crtc_h);
440 		vc4_dlist_write(vc4_state, 0xc0c0c0c0);
441 	}
442 
443 	/* Ch0 H-TPZ Words 0-1: Scaling Parameters, Recip */
444 	if (vc4_state->x_scaling[channel] == VC4_SCALING_TPZ) {
445 		vc4_write_tpz(vc4_state,
446 			      vc4_state->src_w[channel], vc4_state->crtc_w);
447 	}
448 
449 	/* Ch0 V-TPZ Words 0-2: Scaling Parameters, Recip, Context */
450 	if (vc4_state->y_scaling[channel] == VC4_SCALING_TPZ) {
451 		vc4_write_tpz(vc4_state,
452 			      vc4_state->src_h[channel], vc4_state->crtc_h);
453 		vc4_dlist_write(vc4_state, 0xc0c0c0c0);
454 	}
455 }
456 
457 /* Writes out a full display list for an active plane to the plane's
458  * private dlist state.
459  */
460 static int vc4_plane_mode_set(struct drm_plane *plane,
461 			      struct drm_plane_state *state)
462 {
463 	struct vc4_dev *vc4 = to_vc4_dev(plane->dev);
464 	struct vc4_plane_state *vc4_state = to_vc4_plane_state(state);
465 	struct drm_framebuffer *fb = state->fb;
466 	u32 ctl0_offset = vc4_state->dlist_count;
467 	const struct hvs_format *format = vc4_get_hvs_format(fb->format->format);
468 	u64 base_format_mod = fourcc_mod_broadcom_mod(fb->modifier);
469 	int num_planes = drm_format_num_planes(format->drm);
470 	bool mix_plane_alpha;
471 	bool covers_screen;
472 	u32 scl0, scl1, pitch0;
473 	u32 lbm_size, tiling;
474 	unsigned long irqflags;
475 	u32 hvs_format = format->hvs;
476 	int ret, i;
477 
478 	ret = vc4_plane_setup_clipping_and_scaling(state);
479 	if (ret)
480 		return ret;
481 
482 	/* Allocate the LBM memory that the HVS will use for temporary
483 	 * storage due to our scaling/format conversion.
484 	 */
485 	lbm_size = vc4_lbm_size(state);
486 	if (lbm_size) {
487 		if (!vc4_state->lbm.allocated) {
488 			spin_lock_irqsave(&vc4->hvs->mm_lock, irqflags);
489 			ret = drm_mm_insert_node_generic(&vc4->hvs->lbm_mm,
490 							 &vc4_state->lbm,
491 							 lbm_size, 32, 0, 0);
492 			spin_unlock_irqrestore(&vc4->hvs->mm_lock, irqflags);
493 		} else {
494 			WARN_ON_ONCE(lbm_size != vc4_state->lbm.size);
495 		}
496 	}
497 
498 	if (ret)
499 		return ret;
500 
501 	/* SCL1 is used for Cb/Cr scaling of planar formats.  For RGB
502 	 * and 4:4:4, scl1 should be set to scl0 so both channels of
503 	 * the scaler do the same thing.  For YUV, the Y plane needs
504 	 * to be put in channel 1 and Cb/Cr in channel 0, so we swap
505 	 * the scl fields here.
506 	 */
507 	if (num_planes == 1) {
508 		scl0 = vc4_get_scl_field(state, 0);
509 		scl1 = scl0;
510 	} else {
511 		scl0 = vc4_get_scl_field(state, 1);
512 		scl1 = vc4_get_scl_field(state, 0);
513 	}
514 
515 	switch (base_format_mod) {
516 	case DRM_FORMAT_MOD_LINEAR:
517 		tiling = SCALER_CTL0_TILING_LINEAR;
518 		pitch0 = VC4_SET_FIELD(fb->pitches[0], SCALER_SRC_PITCH);
519 		break;
520 
521 	case DRM_FORMAT_MOD_BROADCOM_VC4_T_TILED: {
522 		/* For T-tiled, the FB pitch is "how many bytes from
523 		 * one row to the next, such that pitch * tile_h ==
524 		 * tile_size * tiles_per_row."
525 		 */
526 		u32 tile_size_shift = 12; /* T tiles are 4kb */
527 		u32 tile_h_shift = 5; /* 16 and 32bpp are 32 pixels high */
528 		u32 tiles_w = fb->pitches[0] >> (tile_size_shift - tile_h_shift);
529 
530 		tiling = SCALER_CTL0_TILING_256B_OR_T;
531 
532 		pitch0 = (VC4_SET_FIELD(0, SCALER_PITCH0_TILE_Y_OFFSET) |
533 			  VC4_SET_FIELD(0, SCALER_PITCH0_TILE_WIDTH_L) |
534 			  VC4_SET_FIELD(tiles_w, SCALER_PITCH0_TILE_WIDTH_R));
535 		break;
536 	}
537 
538 	case DRM_FORMAT_MOD_BROADCOM_SAND64:
539 	case DRM_FORMAT_MOD_BROADCOM_SAND128:
540 	case DRM_FORMAT_MOD_BROADCOM_SAND256: {
541 		uint32_t param = fourcc_mod_broadcom_param(fb->modifier);
542 
543 		/* Column-based NV12 or RGBA.
544 		 */
545 		if (fb->format->num_planes > 1) {
546 			if (hvs_format != HVS_PIXEL_FORMAT_YCBCR_YUV420_2PLANE) {
547 				DRM_DEBUG_KMS("SAND format only valid for NV12/21");
548 				return -EINVAL;
549 			}
550 			hvs_format = HVS_PIXEL_FORMAT_H264;
551 		} else {
552 			if (base_format_mod == DRM_FORMAT_MOD_BROADCOM_SAND256) {
553 				DRM_DEBUG_KMS("SAND256 format only valid for H.264");
554 				return -EINVAL;
555 			}
556 		}
557 
558 		switch (base_format_mod) {
559 		case DRM_FORMAT_MOD_BROADCOM_SAND64:
560 			tiling = SCALER_CTL0_TILING_64B;
561 			break;
562 		case DRM_FORMAT_MOD_BROADCOM_SAND128:
563 			tiling = SCALER_CTL0_TILING_128B;
564 			break;
565 		case DRM_FORMAT_MOD_BROADCOM_SAND256:
566 			tiling = SCALER_CTL0_TILING_256B_OR_T;
567 			break;
568 		default:
569 			break;
570 		}
571 
572 		if (param > SCALER_TILE_HEIGHT_MASK) {
573 			DRM_DEBUG_KMS("SAND height too large (%d)\n", param);
574 			return -EINVAL;
575 		}
576 
577 		pitch0 = VC4_SET_FIELD(param, SCALER_TILE_HEIGHT);
578 		break;
579 	}
580 
581 	default:
582 		DRM_DEBUG_KMS("Unsupported FB tiling flag 0x%16llx",
583 			      (long long)fb->modifier);
584 		return -EINVAL;
585 	}
586 
587 	/* Control word */
588 	vc4_dlist_write(vc4_state,
589 			SCALER_CTL0_VALID |
590 			VC4_SET_FIELD(SCALER_CTL0_RGBA_EXPAND_ROUND, SCALER_CTL0_RGBA_EXPAND) |
591 			(format->pixel_order << SCALER_CTL0_ORDER_SHIFT) |
592 			(hvs_format << SCALER_CTL0_PIXEL_FORMAT_SHIFT) |
593 			VC4_SET_FIELD(tiling, SCALER_CTL0_TILING) |
594 			(vc4_state->is_unity ? SCALER_CTL0_UNITY : 0) |
595 			VC4_SET_FIELD(scl0, SCALER_CTL0_SCL0) |
596 			VC4_SET_FIELD(scl1, SCALER_CTL0_SCL1));
597 
598 	/* Position Word 0: Image Positions and Alpha Value */
599 	vc4_state->pos0_offset = vc4_state->dlist_count;
600 	vc4_dlist_write(vc4_state,
601 			VC4_SET_FIELD(state->alpha >> 8, SCALER_POS0_FIXED_ALPHA) |
602 			VC4_SET_FIELD(vc4_state->crtc_x, SCALER_POS0_START_X) |
603 			VC4_SET_FIELD(vc4_state->crtc_y, SCALER_POS0_START_Y));
604 
605 	/* Position Word 1: Scaled Image Dimensions. */
606 	if (!vc4_state->is_unity) {
607 		vc4_dlist_write(vc4_state,
608 				VC4_SET_FIELD(vc4_state->crtc_w,
609 					      SCALER_POS1_SCL_WIDTH) |
610 				VC4_SET_FIELD(vc4_state->crtc_h,
611 					      SCALER_POS1_SCL_HEIGHT));
612 	}
613 
614 	/* Don't waste cycles mixing with plane alpha if the set alpha
615 	 * is opaque or there is no per-pixel alpha information.
616 	 * In any case we use the alpha property value as the fixed alpha.
617 	 */
618 	mix_plane_alpha = state->alpha != DRM_BLEND_ALPHA_OPAQUE &&
619 			  fb->format->has_alpha;
620 
621 	/* Position Word 2: Source Image Size, Alpha */
622 	vc4_state->pos2_offset = vc4_state->dlist_count;
623 	vc4_dlist_write(vc4_state,
624 			VC4_SET_FIELD(fb->format->has_alpha ?
625 				      SCALER_POS2_ALPHA_MODE_PIPELINE :
626 				      SCALER_POS2_ALPHA_MODE_FIXED,
627 				      SCALER_POS2_ALPHA_MODE) |
628 			(mix_plane_alpha ? SCALER_POS2_ALPHA_MIX : 0) |
629 			(fb->format->has_alpha ? SCALER_POS2_ALPHA_PREMULT : 0) |
630 			VC4_SET_FIELD(vc4_state->src_w[0], SCALER_POS2_WIDTH) |
631 			VC4_SET_FIELD(vc4_state->src_h[0], SCALER_POS2_HEIGHT));
632 
633 	/* Position Word 3: Context.  Written by the HVS. */
634 	vc4_dlist_write(vc4_state, 0xc0c0c0c0);
635 
636 
637 	/* Pointer Word 0/1/2: RGB / Y / Cb / Cr Pointers
638 	 *
639 	 * The pointers may be any byte address.
640 	 */
641 	vc4_state->ptr0_offset = vc4_state->dlist_count;
642 	for (i = 0; i < num_planes; i++)
643 		vc4_dlist_write(vc4_state, vc4_state->offsets[i]);
644 
645 	/* Pointer Context Word 0/1/2: Written by the HVS */
646 	for (i = 0; i < num_planes; i++)
647 		vc4_dlist_write(vc4_state, 0xc0c0c0c0);
648 
649 	/* Pitch word 0 */
650 	vc4_dlist_write(vc4_state, pitch0);
651 
652 	/* Pitch word 1/2 */
653 	for (i = 1; i < num_planes; i++) {
654 		if (hvs_format != HVS_PIXEL_FORMAT_H264) {
655 			vc4_dlist_write(vc4_state,
656 					VC4_SET_FIELD(fb->pitches[i],
657 						      SCALER_SRC_PITCH));
658 		} else {
659 			vc4_dlist_write(vc4_state, pitch0);
660 		}
661 	}
662 
663 	/* Colorspace conversion words */
664 	if (vc4_state->is_yuv) {
665 		vc4_dlist_write(vc4_state, SCALER_CSC0_ITR_R_601_5);
666 		vc4_dlist_write(vc4_state, SCALER_CSC1_ITR_R_601_5);
667 		vc4_dlist_write(vc4_state, SCALER_CSC2_ITR_R_601_5);
668 	}
669 
670 	if (vc4_state->x_scaling[0] != VC4_SCALING_NONE ||
671 	    vc4_state->x_scaling[1] != VC4_SCALING_NONE ||
672 	    vc4_state->y_scaling[0] != VC4_SCALING_NONE ||
673 	    vc4_state->y_scaling[1] != VC4_SCALING_NONE) {
674 		/* LBM Base Address. */
675 		if (vc4_state->y_scaling[0] != VC4_SCALING_NONE ||
676 		    vc4_state->y_scaling[1] != VC4_SCALING_NONE) {
677 			vc4_dlist_write(vc4_state, vc4_state->lbm.start);
678 		}
679 
680 		if (num_planes > 1) {
681 			/* Emit Cb/Cr as channel 0 and Y as channel
682 			 * 1. This matches how we set up scl0/scl1
683 			 * above.
684 			 */
685 			vc4_write_scaling_parameters(state, 1);
686 		}
687 		vc4_write_scaling_parameters(state, 0);
688 
689 		/* If any PPF setup was done, then all the kernel
690 		 * pointers get uploaded.
691 		 */
692 		if (vc4_state->x_scaling[0] == VC4_SCALING_PPF ||
693 		    vc4_state->y_scaling[0] == VC4_SCALING_PPF ||
694 		    vc4_state->x_scaling[1] == VC4_SCALING_PPF ||
695 		    vc4_state->y_scaling[1] == VC4_SCALING_PPF) {
696 			u32 kernel = VC4_SET_FIELD(vc4->hvs->mitchell_netravali_filter.start,
697 						   SCALER_PPF_KERNEL_OFFSET);
698 
699 			/* HPPF plane 0 */
700 			vc4_dlist_write(vc4_state, kernel);
701 			/* VPPF plane 0 */
702 			vc4_dlist_write(vc4_state, kernel);
703 			/* HPPF plane 1 */
704 			vc4_dlist_write(vc4_state, kernel);
705 			/* VPPF plane 1 */
706 			vc4_dlist_write(vc4_state, kernel);
707 		}
708 	}
709 
710 	vc4_state->dlist[ctl0_offset] |=
711 		VC4_SET_FIELD(vc4_state->dlist_count, SCALER_CTL0_SIZE);
712 
713 	/* crtc_* are already clipped coordinates. */
714 	covers_screen = vc4_state->crtc_x == 0 && vc4_state->crtc_y == 0 &&
715 			vc4_state->crtc_w == state->crtc->mode.hdisplay &&
716 			vc4_state->crtc_h == state->crtc->mode.vdisplay;
717 	/* Background fill might be necessary when the plane has per-pixel
718 	 * alpha content or a non-opaque plane alpha and could blend from the
719 	 * background or does not cover the entire screen.
720 	 */
721 	vc4_state->needs_bg_fill = fb->format->has_alpha || !covers_screen ||
722 				   state->alpha != DRM_BLEND_ALPHA_OPAQUE;
723 
724 	return 0;
725 }
726 
727 /* If a modeset involves changing the setup of a plane, the atomic
728  * infrastructure will call this to validate a proposed plane setup.
729  * However, if a plane isn't getting updated, this (and the
730  * corresponding vc4_plane_atomic_update) won't get called.  Thus, we
731  * compute the dlist here and have all active plane dlists get updated
732  * in the CRTC's flush.
733  */
734 static int vc4_plane_atomic_check(struct drm_plane *plane,
735 				  struct drm_plane_state *state)
736 {
737 	struct vc4_plane_state *vc4_state = to_vc4_plane_state(state);
738 
739 	vc4_state->dlist_count = 0;
740 
741 	if (plane_enabled(state))
742 		return vc4_plane_mode_set(plane, state);
743 	else
744 		return 0;
745 }
746 
747 static void vc4_plane_atomic_update(struct drm_plane *plane,
748 				    struct drm_plane_state *old_state)
749 {
750 	/* No contents here.  Since we don't know where in the CRTC's
751 	 * dlist we should be stored, our dlist is uploaded to the
752 	 * hardware with vc4_plane_write_dlist() at CRTC atomic_flush
753 	 * time.
754 	 */
755 }
756 
757 u32 vc4_plane_write_dlist(struct drm_plane *plane, u32 __iomem *dlist)
758 {
759 	struct vc4_plane_state *vc4_state = to_vc4_plane_state(plane->state);
760 	int i;
761 
762 	vc4_state->hw_dlist = dlist;
763 
764 	/* Can't memcpy_toio() because it needs to be 32-bit writes. */
765 	for (i = 0; i < vc4_state->dlist_count; i++)
766 		writel(vc4_state->dlist[i], &dlist[i]);
767 
768 	return vc4_state->dlist_count;
769 }
770 
771 u32 vc4_plane_dlist_size(const struct drm_plane_state *state)
772 {
773 	const struct vc4_plane_state *vc4_state =
774 		container_of(state, typeof(*vc4_state), base);
775 
776 	return vc4_state->dlist_count;
777 }
778 
779 /* Updates the plane to immediately (well, once the FIFO needs
780  * refilling) scan out from at a new framebuffer.
781  */
782 void vc4_plane_async_set_fb(struct drm_plane *plane, struct drm_framebuffer *fb)
783 {
784 	struct vc4_plane_state *vc4_state = to_vc4_plane_state(plane->state);
785 	struct drm_gem_cma_object *bo = drm_fb_cma_get_gem_obj(fb, 0);
786 	uint32_t addr;
787 
788 	/* We're skipping the address adjustment for negative origin,
789 	 * because this is only called on the primary plane.
790 	 */
791 	WARN_ON_ONCE(plane->state->crtc_x < 0 || plane->state->crtc_y < 0);
792 	addr = bo->paddr + fb->offsets[0];
793 
794 	/* Write the new address into the hardware immediately.  The
795 	 * scanout will start from this address as soon as the FIFO
796 	 * needs to refill with pixels.
797 	 */
798 	writel(addr, &vc4_state->hw_dlist[vc4_state->ptr0_offset]);
799 
800 	/* Also update the CPU-side dlist copy, so that any later
801 	 * atomic updates that don't do a new modeset on our plane
802 	 * also use our updated address.
803 	 */
804 	vc4_state->dlist[vc4_state->ptr0_offset] = addr;
805 }
806 
807 static void vc4_plane_atomic_async_update(struct drm_plane *plane,
808 					  struct drm_plane_state *state)
809 {
810 	struct vc4_plane_state *vc4_state, *new_vc4_state;
811 
812 	if (plane->state->fb != state->fb) {
813 		vc4_plane_async_set_fb(plane, state->fb);
814 		drm_atomic_set_fb_for_plane(plane->state, state->fb);
815 	}
816 
817 	/* Set the cursor's position on the screen.  This is the
818 	 * expected change from the drm_mode_cursor_universal()
819 	 * helper.
820 	 */
821 	plane->state->crtc_x = state->crtc_x;
822 	plane->state->crtc_y = state->crtc_y;
823 
824 	/* Allow changing the start position within the cursor BO, if
825 	 * that matters.
826 	 */
827 	plane->state->src_x = state->src_x;
828 	plane->state->src_y = state->src_y;
829 
830 	/* Update the display list based on the new crtc_x/y. */
831 	vc4_plane_atomic_check(plane, state);
832 
833 	new_vc4_state = to_vc4_plane_state(state);
834 	vc4_state = to_vc4_plane_state(plane->state);
835 
836 	/* Update the current vc4_state pos0, pos2 and ptr0 dlist entries. */
837 	vc4_state->dlist[vc4_state->pos0_offset] =
838 		new_vc4_state->dlist[vc4_state->pos0_offset];
839 	vc4_state->dlist[vc4_state->pos2_offset] =
840 		new_vc4_state->dlist[vc4_state->pos2_offset];
841 	vc4_state->dlist[vc4_state->ptr0_offset] =
842 		new_vc4_state->dlist[vc4_state->ptr0_offset];
843 
844 	/* Note that we can't just call vc4_plane_write_dlist()
845 	 * because that would smash the context data that the HVS is
846 	 * currently using.
847 	 */
848 	writel(vc4_state->dlist[vc4_state->pos0_offset],
849 	       &vc4_state->hw_dlist[vc4_state->pos0_offset]);
850 	writel(vc4_state->dlist[vc4_state->pos2_offset],
851 	       &vc4_state->hw_dlist[vc4_state->pos2_offset]);
852 	writel(vc4_state->dlist[vc4_state->ptr0_offset],
853 	       &vc4_state->hw_dlist[vc4_state->ptr0_offset]);
854 }
855 
856 static int vc4_plane_atomic_async_check(struct drm_plane *plane,
857 					struct drm_plane_state *state)
858 {
859 	/* No configuring new scaling in the fast path. */
860 	if (plane->state->crtc_w != state->crtc_w ||
861 	    plane->state->crtc_h != state->crtc_h ||
862 	    plane->state->src_w != state->src_w ||
863 	    plane->state->src_h != state->src_h)
864 		return -EINVAL;
865 
866 	return 0;
867 }
868 
869 static int vc4_prepare_fb(struct drm_plane *plane,
870 			  struct drm_plane_state *state)
871 {
872 	struct vc4_bo *bo;
873 	struct dma_fence *fence;
874 	int ret;
875 
876 	if (!state->fb)
877 		return 0;
878 
879 	bo = to_vc4_bo(&drm_fb_cma_get_gem_obj(state->fb, 0)->base);
880 
881 	fence = reservation_object_get_excl_rcu(bo->resv);
882 	drm_atomic_set_fence_for_plane(state, fence);
883 
884 	if (plane->state->fb == state->fb)
885 		return 0;
886 
887 	ret = vc4_bo_inc_usecnt(bo);
888 	if (ret)
889 		return ret;
890 
891 	return 0;
892 }
893 
894 static void vc4_cleanup_fb(struct drm_plane *plane,
895 			   struct drm_plane_state *state)
896 {
897 	struct vc4_bo *bo;
898 
899 	if (plane->state->fb == state->fb || !state->fb)
900 		return;
901 
902 	bo = to_vc4_bo(&drm_fb_cma_get_gem_obj(state->fb, 0)->base);
903 	vc4_bo_dec_usecnt(bo);
904 }
905 
906 static const struct drm_plane_helper_funcs vc4_plane_helper_funcs = {
907 	.atomic_check = vc4_plane_atomic_check,
908 	.atomic_update = vc4_plane_atomic_update,
909 	.prepare_fb = vc4_prepare_fb,
910 	.cleanup_fb = vc4_cleanup_fb,
911 	.atomic_async_check = vc4_plane_atomic_async_check,
912 	.atomic_async_update = vc4_plane_atomic_async_update,
913 };
914 
915 static void vc4_plane_destroy(struct drm_plane *plane)
916 {
917 	drm_plane_helper_disable(plane, NULL);
918 	drm_plane_cleanup(plane);
919 }
920 
921 static bool vc4_format_mod_supported(struct drm_plane *plane,
922 				     uint32_t format,
923 				     uint64_t modifier)
924 {
925 	/* Support T_TILING for RGB formats only. */
926 	switch (format) {
927 	case DRM_FORMAT_XRGB8888:
928 	case DRM_FORMAT_ARGB8888:
929 	case DRM_FORMAT_ABGR8888:
930 	case DRM_FORMAT_XBGR8888:
931 	case DRM_FORMAT_RGB565:
932 	case DRM_FORMAT_BGR565:
933 	case DRM_FORMAT_ARGB1555:
934 	case DRM_FORMAT_XRGB1555:
935 		switch (fourcc_mod_broadcom_mod(modifier)) {
936 		case DRM_FORMAT_MOD_LINEAR:
937 		case DRM_FORMAT_MOD_BROADCOM_VC4_T_TILED:
938 		case DRM_FORMAT_MOD_BROADCOM_SAND64:
939 		case DRM_FORMAT_MOD_BROADCOM_SAND128:
940 			return true;
941 		default:
942 			return false;
943 		}
944 	case DRM_FORMAT_NV12:
945 	case DRM_FORMAT_NV21:
946 		switch (fourcc_mod_broadcom_mod(modifier)) {
947 		case DRM_FORMAT_MOD_LINEAR:
948 		case DRM_FORMAT_MOD_BROADCOM_SAND64:
949 		case DRM_FORMAT_MOD_BROADCOM_SAND128:
950 		case DRM_FORMAT_MOD_BROADCOM_SAND256:
951 			return true;
952 		default:
953 			return false;
954 		}
955 	case DRM_FORMAT_YUV422:
956 	case DRM_FORMAT_YVU422:
957 	case DRM_FORMAT_YUV420:
958 	case DRM_FORMAT_YVU420:
959 	case DRM_FORMAT_NV16:
960 	case DRM_FORMAT_NV61:
961 	default:
962 		return (modifier == DRM_FORMAT_MOD_LINEAR);
963 	}
964 }
965 
966 static const struct drm_plane_funcs vc4_plane_funcs = {
967 	.update_plane = drm_atomic_helper_update_plane,
968 	.disable_plane = drm_atomic_helper_disable_plane,
969 	.destroy = vc4_plane_destroy,
970 	.set_property = NULL,
971 	.reset = vc4_plane_reset,
972 	.atomic_duplicate_state = vc4_plane_duplicate_state,
973 	.atomic_destroy_state = vc4_plane_destroy_state,
974 	.format_mod_supported = vc4_format_mod_supported,
975 };
976 
977 struct drm_plane *vc4_plane_init(struct drm_device *dev,
978 				 enum drm_plane_type type)
979 {
980 	struct drm_plane *plane = NULL;
981 	struct vc4_plane *vc4_plane;
982 	u32 formats[ARRAY_SIZE(hvs_formats)];
983 	u32 num_formats = 0;
984 	int ret = 0;
985 	unsigned i;
986 	static const uint64_t modifiers[] = {
987 		DRM_FORMAT_MOD_BROADCOM_VC4_T_TILED,
988 		DRM_FORMAT_MOD_BROADCOM_SAND128,
989 		DRM_FORMAT_MOD_BROADCOM_SAND64,
990 		DRM_FORMAT_MOD_BROADCOM_SAND256,
991 		DRM_FORMAT_MOD_LINEAR,
992 		DRM_FORMAT_MOD_INVALID
993 	};
994 
995 	vc4_plane = devm_kzalloc(dev->dev, sizeof(*vc4_plane),
996 				 GFP_KERNEL);
997 	if (!vc4_plane)
998 		return ERR_PTR(-ENOMEM);
999 
1000 	for (i = 0; i < ARRAY_SIZE(hvs_formats); i++) {
1001 		/* Don't allow YUV in cursor planes, since that means
1002 		 * tuning on the scaler, which we don't allow for the
1003 		 * cursor.
1004 		 */
1005 		if (type != DRM_PLANE_TYPE_CURSOR ||
1006 		    hvs_formats[i].hvs < HVS_PIXEL_FORMAT_YCBCR_YUV420_3PLANE) {
1007 			formats[num_formats++] = hvs_formats[i].drm;
1008 		}
1009 	}
1010 	plane = &vc4_plane->base;
1011 	ret = drm_universal_plane_init(dev, plane, 0,
1012 				       &vc4_plane_funcs,
1013 				       formats, num_formats,
1014 				       modifiers, type, NULL);
1015 
1016 	drm_plane_helper_add(plane, &vc4_plane_helper_funcs);
1017 
1018 	drm_plane_create_alpha_property(plane);
1019 
1020 	return plane;
1021 }
1022