xref: /linux/drivers/gpu/drm/vc4/vc4_plane.c (revision 7f4f3b14e8079ecde096bd734af10e30d40c27b7)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Copyright (C) 2015 Broadcom
4  */
5 
6 /**
7  * DOC: VC4 plane module
8  *
9  * Each DRM plane is a layer of pixels being scanned out by the HVS.
10  *
11  * At atomic modeset check time, we compute the HVS display element
12  * state that would be necessary for displaying the plane (giving us a
13  * chance to figure out if a plane configuration is invalid), then at
14  * atomic flush time the CRTC will ask us to write our element state
15  * into the region of the HVS that it has allocated for us.
16  */
17 
18 #include <drm/drm_atomic.h>
19 #include <drm/drm_atomic_helper.h>
20 #include <drm/drm_atomic_uapi.h>
21 #include <drm/drm_blend.h>
22 #include <drm/drm_drv.h>
23 #include <drm/drm_fb_dma_helper.h>
24 #include <drm/drm_fourcc.h>
25 #include <drm/drm_framebuffer.h>
26 #include <drm/drm_gem_atomic_helper.h>
27 
28 #include "uapi/drm/vc4_drm.h"
29 
30 #include "vc4_drv.h"
31 #include "vc4_regs.h"
32 
33 static const struct hvs_format {
34 	u32 drm; /* DRM_FORMAT_* */
35 	u32 hvs; /* HVS_FORMAT_* */
36 	u32 pixel_order;
37 	u32 pixel_order_hvs5;
38 	bool hvs5_only;
39 } hvs_formats[] = {
40 	{
41 		.drm = DRM_FORMAT_XRGB8888,
42 		.hvs = HVS_PIXEL_FORMAT_RGBA8888,
43 		.pixel_order = HVS_PIXEL_ORDER_ABGR,
44 		.pixel_order_hvs5 = HVS_PIXEL_ORDER_ARGB,
45 	},
46 	{
47 		.drm = DRM_FORMAT_ARGB8888,
48 		.hvs = HVS_PIXEL_FORMAT_RGBA8888,
49 		.pixel_order = HVS_PIXEL_ORDER_ABGR,
50 		.pixel_order_hvs5 = HVS_PIXEL_ORDER_ARGB,
51 	},
52 	{
53 		.drm = DRM_FORMAT_ABGR8888,
54 		.hvs = HVS_PIXEL_FORMAT_RGBA8888,
55 		.pixel_order = HVS_PIXEL_ORDER_ARGB,
56 		.pixel_order_hvs5 = HVS_PIXEL_ORDER_ABGR,
57 	},
58 	{
59 		.drm = DRM_FORMAT_XBGR8888,
60 		.hvs = HVS_PIXEL_FORMAT_RGBA8888,
61 		.pixel_order = HVS_PIXEL_ORDER_ARGB,
62 		.pixel_order_hvs5 = HVS_PIXEL_ORDER_ABGR,
63 	},
64 	{
65 		.drm = DRM_FORMAT_RGB565,
66 		.hvs = HVS_PIXEL_FORMAT_RGB565,
67 		.pixel_order = HVS_PIXEL_ORDER_XRGB,
68 		.pixel_order_hvs5 = HVS_PIXEL_ORDER_XRGB,
69 	},
70 	{
71 		.drm = DRM_FORMAT_BGR565,
72 		.hvs = HVS_PIXEL_FORMAT_RGB565,
73 		.pixel_order = HVS_PIXEL_ORDER_XBGR,
74 		.pixel_order_hvs5 = HVS_PIXEL_ORDER_XBGR,
75 	},
76 	{
77 		.drm = DRM_FORMAT_ARGB1555,
78 		.hvs = HVS_PIXEL_FORMAT_RGBA5551,
79 		.pixel_order = HVS_PIXEL_ORDER_ABGR,
80 		.pixel_order_hvs5 = HVS_PIXEL_ORDER_ARGB,
81 	},
82 	{
83 		.drm = DRM_FORMAT_XRGB1555,
84 		.hvs = HVS_PIXEL_FORMAT_RGBA5551,
85 		.pixel_order = HVS_PIXEL_ORDER_ABGR,
86 		.pixel_order_hvs5 = HVS_PIXEL_ORDER_ARGB,
87 	},
88 	{
89 		.drm = DRM_FORMAT_RGB888,
90 		.hvs = HVS_PIXEL_FORMAT_RGB888,
91 		.pixel_order = HVS_PIXEL_ORDER_XRGB,
92 		.pixel_order_hvs5 = HVS_PIXEL_ORDER_XRGB,
93 	},
94 	{
95 		.drm = DRM_FORMAT_BGR888,
96 		.hvs = HVS_PIXEL_FORMAT_RGB888,
97 		.pixel_order = HVS_PIXEL_ORDER_XBGR,
98 		.pixel_order_hvs5 = HVS_PIXEL_ORDER_XBGR,
99 	},
100 	{
101 		.drm = DRM_FORMAT_YUV422,
102 		.hvs = HVS_PIXEL_FORMAT_YCBCR_YUV422_3PLANE,
103 		.pixel_order = HVS_PIXEL_ORDER_XYCBCR,
104 		.pixel_order_hvs5 = HVS_PIXEL_ORDER_XYCBCR,
105 	},
106 	{
107 		.drm = DRM_FORMAT_YVU422,
108 		.hvs = HVS_PIXEL_FORMAT_YCBCR_YUV422_3PLANE,
109 		.pixel_order = HVS_PIXEL_ORDER_XYCRCB,
110 		.pixel_order_hvs5 = HVS_PIXEL_ORDER_XYCRCB,
111 	},
112 	{
113 		.drm = DRM_FORMAT_YUV444,
114 		.hvs = HVS_PIXEL_FORMAT_YCBCR_YUV422_3PLANE,
115 		.pixel_order = HVS_PIXEL_ORDER_XYCBCR,
116 		.pixel_order_hvs5 = HVS_PIXEL_ORDER_XYCBCR,
117 	},
118 	{
119 		.drm = DRM_FORMAT_YVU444,
120 		.hvs = HVS_PIXEL_FORMAT_YCBCR_YUV422_3PLANE,
121 		.pixel_order = HVS_PIXEL_ORDER_XYCRCB,
122 		.pixel_order_hvs5 = HVS_PIXEL_ORDER_XYCRCB,
123 	},
124 	{
125 		.drm = DRM_FORMAT_YUV420,
126 		.hvs = HVS_PIXEL_FORMAT_YCBCR_YUV420_3PLANE,
127 		.pixel_order = HVS_PIXEL_ORDER_XYCBCR,
128 		.pixel_order_hvs5 = HVS_PIXEL_ORDER_XYCBCR,
129 	},
130 	{
131 		.drm = DRM_FORMAT_YVU420,
132 		.hvs = HVS_PIXEL_FORMAT_YCBCR_YUV420_3PLANE,
133 		.pixel_order = HVS_PIXEL_ORDER_XYCRCB,
134 		.pixel_order_hvs5 = HVS_PIXEL_ORDER_XYCRCB,
135 	},
136 	{
137 		.drm = DRM_FORMAT_NV12,
138 		.hvs = HVS_PIXEL_FORMAT_YCBCR_YUV420_2PLANE,
139 		.pixel_order = HVS_PIXEL_ORDER_XYCBCR,
140 		.pixel_order_hvs5 = HVS_PIXEL_ORDER_XYCBCR,
141 	},
142 	{
143 		.drm = DRM_FORMAT_NV21,
144 		.hvs = HVS_PIXEL_FORMAT_YCBCR_YUV420_2PLANE,
145 		.pixel_order = HVS_PIXEL_ORDER_XYCRCB,
146 		.pixel_order_hvs5 = HVS_PIXEL_ORDER_XYCRCB,
147 	},
148 	{
149 		.drm = DRM_FORMAT_NV16,
150 		.hvs = HVS_PIXEL_FORMAT_YCBCR_YUV422_2PLANE,
151 		.pixel_order = HVS_PIXEL_ORDER_XYCBCR,
152 		.pixel_order_hvs5 = HVS_PIXEL_ORDER_XYCBCR,
153 	},
154 	{
155 		.drm = DRM_FORMAT_NV61,
156 		.hvs = HVS_PIXEL_FORMAT_YCBCR_YUV422_2PLANE,
157 		.pixel_order = HVS_PIXEL_ORDER_XYCRCB,
158 		.pixel_order_hvs5 = HVS_PIXEL_ORDER_XYCRCB,
159 	},
160 	{
161 		.drm = DRM_FORMAT_P030,
162 		.hvs = HVS_PIXEL_FORMAT_YCBCR_10BIT,
163 		.pixel_order_hvs5 = HVS_PIXEL_ORDER_XYCBCR,
164 		.hvs5_only = true,
165 	},
166 	{
167 		.drm = DRM_FORMAT_XRGB2101010,
168 		.hvs = HVS_PIXEL_FORMAT_RGBA1010102,
169 		.pixel_order_hvs5 = HVS_PIXEL_ORDER_ARGB,
170 		.hvs5_only = true,
171 	},
172 	{
173 		.drm = DRM_FORMAT_ARGB2101010,
174 		.hvs = HVS_PIXEL_FORMAT_RGBA1010102,
175 		.pixel_order_hvs5 = HVS_PIXEL_ORDER_ARGB,
176 		.hvs5_only = true,
177 	},
178 	{
179 		.drm = DRM_FORMAT_ABGR2101010,
180 		.hvs = HVS_PIXEL_FORMAT_RGBA1010102,
181 		.pixel_order_hvs5 = HVS_PIXEL_ORDER_ABGR,
182 		.hvs5_only = true,
183 	},
184 	{
185 		.drm = DRM_FORMAT_XBGR2101010,
186 		.hvs = HVS_PIXEL_FORMAT_RGBA1010102,
187 		.pixel_order_hvs5 = HVS_PIXEL_ORDER_ABGR,
188 		.hvs5_only = true,
189 	},
190 	{
191 		.drm = DRM_FORMAT_RGB332,
192 		.hvs = HVS_PIXEL_FORMAT_RGB332,
193 		.pixel_order = HVS_PIXEL_ORDER_ARGB,
194 		.pixel_order_hvs5 = HVS_PIXEL_ORDER_ARGB,
195 	},
196 	{
197 		.drm = DRM_FORMAT_BGR233,
198 		.hvs = HVS_PIXEL_FORMAT_RGB332,
199 		.pixel_order = HVS_PIXEL_ORDER_ABGR,
200 		.pixel_order_hvs5 = HVS_PIXEL_ORDER_ABGR,
201 	},
202 	{
203 		.drm = DRM_FORMAT_XRGB4444,
204 		.hvs = HVS_PIXEL_FORMAT_RGBA4444,
205 		.pixel_order = HVS_PIXEL_ORDER_ABGR,
206 		.pixel_order_hvs5 = HVS_PIXEL_ORDER_ARGB,
207 	},
208 	{
209 		.drm = DRM_FORMAT_ARGB4444,
210 		.hvs = HVS_PIXEL_FORMAT_RGBA4444,
211 		.pixel_order = HVS_PIXEL_ORDER_ABGR,
212 		.pixel_order_hvs5 = HVS_PIXEL_ORDER_ARGB,
213 	},
214 	{
215 		.drm = DRM_FORMAT_XBGR4444,
216 		.hvs = HVS_PIXEL_FORMAT_RGBA4444,
217 		.pixel_order = HVS_PIXEL_ORDER_ARGB,
218 		.pixel_order_hvs5 = HVS_PIXEL_ORDER_ABGR,
219 	},
220 	{
221 		.drm = DRM_FORMAT_ABGR4444,
222 		.hvs = HVS_PIXEL_FORMAT_RGBA4444,
223 		.pixel_order = HVS_PIXEL_ORDER_ARGB,
224 		.pixel_order_hvs5 = HVS_PIXEL_ORDER_ABGR,
225 	},
226 	{
227 		.drm = DRM_FORMAT_BGRX4444,
228 		.hvs = HVS_PIXEL_FORMAT_RGBA4444,
229 		.pixel_order = HVS_PIXEL_ORDER_RGBA,
230 		.pixel_order_hvs5 = HVS_PIXEL_ORDER_BGRA,
231 	},
232 	{
233 		.drm = DRM_FORMAT_BGRA4444,
234 		.hvs = HVS_PIXEL_FORMAT_RGBA4444,
235 		.pixel_order = HVS_PIXEL_ORDER_RGBA,
236 		.pixel_order_hvs5 = HVS_PIXEL_ORDER_BGRA,
237 	},
238 	{
239 		.drm = DRM_FORMAT_RGBX4444,
240 		.hvs = HVS_PIXEL_FORMAT_RGBA4444,
241 		.pixel_order = HVS_PIXEL_ORDER_BGRA,
242 		.pixel_order_hvs5 = HVS_PIXEL_ORDER_RGBA,
243 	},
244 	{
245 		.drm = DRM_FORMAT_RGBA4444,
246 		.hvs = HVS_PIXEL_FORMAT_RGBA4444,
247 		.pixel_order = HVS_PIXEL_ORDER_BGRA,
248 		.pixel_order_hvs5 = HVS_PIXEL_ORDER_RGBA,
249 	},
250 };
251 
252 static const struct hvs_format *vc4_get_hvs_format(u32 drm_format)
253 {
254 	unsigned i;
255 
256 	for (i = 0; i < ARRAY_SIZE(hvs_formats); i++) {
257 		if (hvs_formats[i].drm == drm_format)
258 			return &hvs_formats[i];
259 	}
260 
261 	return NULL;
262 }
263 
264 static enum vc4_scaling_mode vc4_get_scaling_mode(u32 src, u32 dst)
265 {
266 	if (dst == src >> 16)
267 		return VC4_SCALING_NONE;
268 	if (3 * dst >= 2 * (src >> 16))
269 		return VC4_SCALING_PPF;
270 	else
271 		return VC4_SCALING_TPZ;
272 }
273 
274 static bool plane_enabled(struct drm_plane_state *state)
275 {
276 	return state->fb && !WARN_ON(!state->crtc);
277 }
278 
279 static struct drm_plane_state *vc4_plane_duplicate_state(struct drm_plane *plane)
280 {
281 	struct vc4_plane_state *vc4_state;
282 
283 	if (WARN_ON(!plane->state))
284 		return NULL;
285 
286 	vc4_state = kmemdup(plane->state, sizeof(*vc4_state), GFP_KERNEL);
287 	if (!vc4_state)
288 		return NULL;
289 
290 	memset(&vc4_state->lbm, 0, sizeof(vc4_state->lbm));
291 	vc4_state->dlist_initialized = 0;
292 
293 	__drm_atomic_helper_plane_duplicate_state(plane, &vc4_state->base);
294 
295 	if (vc4_state->dlist) {
296 		vc4_state->dlist = kmemdup(vc4_state->dlist,
297 					   vc4_state->dlist_count * 4,
298 					   GFP_KERNEL);
299 		if (!vc4_state->dlist) {
300 			kfree(vc4_state);
301 			return NULL;
302 		}
303 		vc4_state->dlist_size = vc4_state->dlist_count;
304 	}
305 
306 	return &vc4_state->base;
307 }
308 
309 static void vc4_plane_destroy_state(struct drm_plane *plane,
310 				    struct drm_plane_state *state)
311 {
312 	struct vc4_dev *vc4 = to_vc4_dev(plane->dev);
313 	struct vc4_plane_state *vc4_state = to_vc4_plane_state(state);
314 
315 	if (drm_mm_node_allocated(&vc4_state->lbm)) {
316 		unsigned long irqflags;
317 
318 		spin_lock_irqsave(&vc4->hvs->mm_lock, irqflags);
319 		drm_mm_remove_node(&vc4_state->lbm);
320 		spin_unlock_irqrestore(&vc4->hvs->mm_lock, irqflags);
321 	}
322 
323 	kfree(vc4_state->dlist);
324 	__drm_atomic_helper_plane_destroy_state(&vc4_state->base);
325 	kfree(state);
326 }
327 
328 /* Called during init to allocate the plane's atomic state. */
329 static void vc4_plane_reset(struct drm_plane *plane)
330 {
331 	struct vc4_plane_state *vc4_state;
332 
333 	WARN_ON(plane->state);
334 
335 	vc4_state = kzalloc(sizeof(*vc4_state), GFP_KERNEL);
336 	if (!vc4_state)
337 		return;
338 
339 	__drm_atomic_helper_plane_reset(plane, &vc4_state->base);
340 }
341 
342 static void vc4_dlist_counter_increment(struct vc4_plane_state *vc4_state)
343 {
344 	if (vc4_state->dlist_count == vc4_state->dlist_size) {
345 		u32 new_size = max(4u, vc4_state->dlist_count * 2);
346 		u32 *new_dlist = kmalloc_array(new_size, 4, GFP_KERNEL);
347 
348 		if (!new_dlist)
349 			return;
350 		memcpy(new_dlist, vc4_state->dlist, vc4_state->dlist_count * 4);
351 
352 		kfree(vc4_state->dlist);
353 		vc4_state->dlist = new_dlist;
354 		vc4_state->dlist_size = new_size;
355 	}
356 
357 	vc4_state->dlist_count++;
358 }
359 
360 static void vc4_dlist_write(struct vc4_plane_state *vc4_state, u32 val)
361 {
362 	unsigned int idx = vc4_state->dlist_count;
363 
364 	vc4_dlist_counter_increment(vc4_state);
365 	vc4_state->dlist[idx] = val;
366 }
367 
368 /* Returns the scl0/scl1 field based on whether the dimensions need to
369  * be up/down/non-scaled.
370  *
371  * This is a replication of a table from the spec.
372  */
373 static u32 vc4_get_scl_field(struct drm_plane_state *state, int plane)
374 {
375 	struct vc4_plane_state *vc4_state = to_vc4_plane_state(state);
376 
377 	switch (vc4_state->x_scaling[plane] << 2 | vc4_state->y_scaling[plane]) {
378 	case VC4_SCALING_PPF << 2 | VC4_SCALING_PPF:
379 		return SCALER_CTL0_SCL_H_PPF_V_PPF;
380 	case VC4_SCALING_TPZ << 2 | VC4_SCALING_PPF:
381 		return SCALER_CTL0_SCL_H_TPZ_V_PPF;
382 	case VC4_SCALING_PPF << 2 | VC4_SCALING_TPZ:
383 		return SCALER_CTL0_SCL_H_PPF_V_TPZ;
384 	case VC4_SCALING_TPZ << 2 | VC4_SCALING_TPZ:
385 		return SCALER_CTL0_SCL_H_TPZ_V_TPZ;
386 	case VC4_SCALING_PPF << 2 | VC4_SCALING_NONE:
387 		return SCALER_CTL0_SCL_H_PPF_V_NONE;
388 	case VC4_SCALING_NONE << 2 | VC4_SCALING_PPF:
389 		return SCALER_CTL0_SCL_H_NONE_V_PPF;
390 	case VC4_SCALING_NONE << 2 | VC4_SCALING_TPZ:
391 		return SCALER_CTL0_SCL_H_NONE_V_TPZ;
392 	case VC4_SCALING_TPZ << 2 | VC4_SCALING_NONE:
393 		return SCALER_CTL0_SCL_H_TPZ_V_NONE;
394 	default:
395 	case VC4_SCALING_NONE << 2 | VC4_SCALING_NONE:
396 		/* The unity case is independently handled by
397 		 * SCALER_CTL0_UNITY.
398 		 */
399 		return 0;
400 	}
401 }
402 
403 static int vc4_plane_margins_adj(struct drm_plane_state *pstate)
404 {
405 	struct vc4_plane_state *vc4_pstate = to_vc4_plane_state(pstate);
406 	unsigned int left, right, top, bottom, adjhdisplay, adjvdisplay;
407 	struct drm_crtc_state *crtc_state;
408 
409 	crtc_state = drm_atomic_get_new_crtc_state(pstate->state,
410 						   pstate->crtc);
411 
412 	vc4_crtc_get_margins(crtc_state, &left, &right, &top, &bottom);
413 	if (!left && !right && !top && !bottom)
414 		return 0;
415 
416 	if (left + right >= crtc_state->mode.hdisplay ||
417 	    top + bottom >= crtc_state->mode.vdisplay)
418 		return -EINVAL;
419 
420 	adjhdisplay = crtc_state->mode.hdisplay - (left + right);
421 	vc4_pstate->crtc_x = DIV_ROUND_CLOSEST(vc4_pstate->crtc_x *
422 					       adjhdisplay,
423 					       crtc_state->mode.hdisplay);
424 	vc4_pstate->crtc_x += left;
425 	if (vc4_pstate->crtc_x > crtc_state->mode.hdisplay - right)
426 		vc4_pstate->crtc_x = crtc_state->mode.hdisplay - right;
427 
428 	adjvdisplay = crtc_state->mode.vdisplay - (top + bottom);
429 	vc4_pstate->crtc_y = DIV_ROUND_CLOSEST(vc4_pstate->crtc_y *
430 					       adjvdisplay,
431 					       crtc_state->mode.vdisplay);
432 	vc4_pstate->crtc_y += top;
433 	if (vc4_pstate->crtc_y > crtc_state->mode.vdisplay - bottom)
434 		vc4_pstate->crtc_y = crtc_state->mode.vdisplay - bottom;
435 
436 	vc4_pstate->crtc_w = DIV_ROUND_CLOSEST(vc4_pstate->crtc_w *
437 					       adjhdisplay,
438 					       crtc_state->mode.hdisplay);
439 	vc4_pstate->crtc_h = DIV_ROUND_CLOSEST(vc4_pstate->crtc_h *
440 					       adjvdisplay,
441 					       crtc_state->mode.vdisplay);
442 
443 	if (!vc4_pstate->crtc_w || !vc4_pstate->crtc_h)
444 		return -EINVAL;
445 
446 	return 0;
447 }
448 
449 static int vc4_plane_setup_clipping_and_scaling(struct drm_plane_state *state)
450 {
451 	struct vc4_plane_state *vc4_state = to_vc4_plane_state(state);
452 	struct drm_framebuffer *fb = state->fb;
453 	int num_planes = fb->format->num_planes;
454 	struct drm_crtc_state *crtc_state;
455 	u32 h_subsample = fb->format->hsub;
456 	u32 v_subsample = fb->format->vsub;
457 	int ret;
458 
459 	crtc_state = drm_atomic_get_existing_crtc_state(state->state,
460 							state->crtc);
461 	if (!crtc_state) {
462 		DRM_DEBUG_KMS("Invalid crtc state\n");
463 		return -EINVAL;
464 	}
465 
466 	ret = drm_atomic_helper_check_plane_state(state, crtc_state, 1,
467 						  INT_MAX, true, true);
468 	if (ret)
469 		return ret;
470 
471 	vc4_state->src_x = state->src.x1;
472 	vc4_state->src_y = state->src.y1;
473 	vc4_state->src_w[0] = state->src.x2 - vc4_state->src_x;
474 	vc4_state->src_h[0] = state->src.y2 - vc4_state->src_y;
475 
476 	vc4_state->crtc_x = state->dst.x1;
477 	vc4_state->crtc_y = state->dst.y1;
478 	vc4_state->crtc_w = state->dst.x2 - state->dst.x1;
479 	vc4_state->crtc_h = state->dst.y2 - state->dst.y1;
480 
481 	ret = vc4_plane_margins_adj(state);
482 	if (ret)
483 		return ret;
484 
485 	vc4_state->x_scaling[0] = vc4_get_scaling_mode(vc4_state->src_w[0],
486 						       vc4_state->crtc_w);
487 	vc4_state->y_scaling[0] = vc4_get_scaling_mode(vc4_state->src_h[0],
488 						       vc4_state->crtc_h);
489 
490 	vc4_state->is_unity = (vc4_state->x_scaling[0] == VC4_SCALING_NONE &&
491 			       vc4_state->y_scaling[0] == VC4_SCALING_NONE);
492 
493 	if (num_planes > 1) {
494 		vc4_state->is_yuv = true;
495 
496 		vc4_state->src_w[1] = vc4_state->src_w[0] / h_subsample;
497 		vc4_state->src_h[1] = vc4_state->src_h[0] / v_subsample;
498 
499 		vc4_state->x_scaling[1] =
500 			vc4_get_scaling_mode(vc4_state->src_w[1],
501 					     vc4_state->crtc_w);
502 		vc4_state->y_scaling[1] =
503 			vc4_get_scaling_mode(vc4_state->src_h[1],
504 					     vc4_state->crtc_h);
505 
506 		/* YUV conversion requires that horizontal scaling be enabled
507 		 * on the UV plane even if vc4_get_scaling_mode() returned
508 		 * VC4_SCALING_NONE (which can happen when the down-scaling
509 		 * ratio is 0.5). Let's force it to VC4_SCALING_PPF in this
510 		 * case.
511 		 */
512 		if (vc4_state->x_scaling[1] == VC4_SCALING_NONE)
513 			vc4_state->x_scaling[1] = VC4_SCALING_PPF;
514 
515 		/* Similarly UV needs vertical scaling to be enabled.
516 		 * Without this a 1:1 scaled YUV422 plane isn't rendered.
517 		 */
518 		if (vc4_state->y_scaling[1] == VC4_SCALING_NONE)
519 			vc4_state->y_scaling[1] = VC4_SCALING_PPF;
520 	} else {
521 		vc4_state->is_yuv = false;
522 		vc4_state->x_scaling[1] = VC4_SCALING_NONE;
523 		vc4_state->y_scaling[1] = VC4_SCALING_NONE;
524 	}
525 
526 	return 0;
527 }
528 
529 static void vc4_write_tpz(struct vc4_plane_state *vc4_state, u32 src, u32 dst)
530 {
531 	u32 scale, recip;
532 
533 	scale = src / dst;
534 
535 	/* The specs note that while the reciprocal would be defined
536 	 * as (1<<32)/scale, ~0 is close enough.
537 	 */
538 	recip = ~0 / scale;
539 
540 	vc4_dlist_write(vc4_state,
541 			VC4_SET_FIELD(scale, SCALER_TPZ0_SCALE) |
542 			VC4_SET_FIELD(0, SCALER_TPZ0_IPHASE));
543 	vc4_dlist_write(vc4_state,
544 			VC4_SET_FIELD(recip, SCALER_TPZ1_RECIP));
545 }
546 
547 /* phase magnitude bits */
548 #define PHASE_BITS 6
549 
550 static void vc4_write_ppf(struct vc4_plane_state *vc4_state, u32 src, u32 dst,
551 			  u32 xy, int channel)
552 {
553 	u32 scale = src / dst;
554 	s32 offset, offset2;
555 	s32 phase;
556 
557 	/*
558 	 * Start the phase at 1/2 pixel from the 1st pixel at src_x.
559 	 * 1/4 pixel for YUV.
560 	 */
561 	if (channel) {
562 		/*
563 		 * The phase is relative to scale_src->x, so shift it for
564 		 * display list's x value
565 		 */
566 		offset = (xy & 0x1ffff) >> (16 - PHASE_BITS) >> 1;
567 		offset += -(1 << PHASE_BITS >> 2);
568 	} else {
569 		/*
570 		 * The phase is relative to scale_src->x, so shift it for
571 		 * display list's x value
572 		 */
573 		offset = (xy & 0xffff) >> (16 - PHASE_BITS);
574 		offset += -(1 << PHASE_BITS >> 1);
575 
576 		/*
577 		 * This is a kludge to make sure the scaling factors are
578 		 * consistent with YUV's luma scaling. We lose 1-bit precision
579 		 * because of this.
580 		 */
581 		scale &= ~1;
582 	}
583 
584 	/*
585 	 * There may be a also small error introduced by precision of scale.
586 	 * Add half of that as a compromise
587 	 */
588 	offset2 = src - dst * scale;
589 	offset2 >>= 16 - PHASE_BITS;
590 	phase = offset + (offset2 >> 1);
591 
592 	/* Ensure +ve values don't touch the sign bit, then truncate negative values */
593 	if (phase >= 1 << PHASE_BITS)
594 		phase = (1 << PHASE_BITS) - 1;
595 
596 	phase &= SCALER_PPF_IPHASE_MASK;
597 
598 	vc4_dlist_write(vc4_state,
599 			SCALER_PPF_AGC |
600 			VC4_SET_FIELD(scale, SCALER_PPF_SCALE) |
601 			VC4_SET_FIELD(phase, SCALER_PPF_IPHASE));
602 }
603 
604 static u32 vc4_lbm_size(struct drm_plane_state *state)
605 {
606 	struct vc4_plane_state *vc4_state = to_vc4_plane_state(state);
607 	struct vc4_dev *vc4 = to_vc4_dev(state->plane->dev);
608 	u32 pix_per_line;
609 	u32 lbm;
610 
611 	/* LBM is not needed when there's no vertical scaling. */
612 	if (vc4_state->y_scaling[0] == VC4_SCALING_NONE &&
613 	    vc4_state->y_scaling[1] == VC4_SCALING_NONE)
614 		return 0;
615 
616 	/*
617 	 * This can be further optimized in the RGB/YUV444 case if the PPF
618 	 * decimation factor is between 0.5 and 1.0 by using crtc_w.
619 	 *
620 	 * It's not an issue though, since in that case since src_w[0] is going
621 	 * to be greater than or equal to crtc_w.
622 	 */
623 	if (vc4_state->x_scaling[0] == VC4_SCALING_TPZ)
624 		pix_per_line = vc4_state->crtc_w;
625 	else
626 		pix_per_line = vc4_state->src_w[0] >> 16;
627 
628 	if (!vc4_state->is_yuv) {
629 		if (vc4_state->y_scaling[0] == VC4_SCALING_TPZ)
630 			lbm = pix_per_line * 8;
631 		else {
632 			/* In special cases, this multiplier might be 12. */
633 			lbm = pix_per_line * 16;
634 		}
635 	} else {
636 		/* There are cases for this going down to a multiplier
637 		 * of 2, but according to the firmware source, the
638 		 * table in the docs is somewhat wrong.
639 		 */
640 		lbm = pix_per_line * 16;
641 	}
642 
643 	/* Align it to 64 or 128 (hvs5) bytes */
644 	lbm = roundup(lbm, vc4->gen == VC4_GEN_5 ? 128 : 64);
645 
646 	/* Each "word" of the LBM memory contains 2 or 4 (hvs5) pixels */
647 	lbm /= vc4->gen == VC4_GEN_5 ? 4 : 2;
648 
649 	return lbm;
650 }
651 
652 static void vc4_write_scaling_parameters(struct drm_plane_state *state,
653 					 int channel)
654 {
655 	struct vc4_plane_state *vc4_state = to_vc4_plane_state(state);
656 
657 	/* Ch0 H-PPF Word 0: Scaling Parameters */
658 	if (vc4_state->x_scaling[channel] == VC4_SCALING_PPF) {
659 		vc4_write_ppf(vc4_state, vc4_state->src_w[channel],
660 			      vc4_state->crtc_w, vc4_state->src_x, channel);
661 	}
662 
663 	/* Ch0 V-PPF Words 0-1: Scaling Parameters, Context */
664 	if (vc4_state->y_scaling[channel] == VC4_SCALING_PPF) {
665 		vc4_write_ppf(vc4_state, vc4_state->src_h[channel],
666 			      vc4_state->crtc_h, vc4_state->src_y, channel);
667 		vc4_dlist_write(vc4_state, 0xc0c0c0c0);
668 	}
669 
670 	/* Ch0 H-TPZ Words 0-1: Scaling Parameters, Recip */
671 	if (vc4_state->x_scaling[channel] == VC4_SCALING_TPZ) {
672 		vc4_write_tpz(vc4_state, vc4_state->src_w[channel],
673 			      vc4_state->crtc_w);
674 	}
675 
676 	/* Ch0 V-TPZ Words 0-2: Scaling Parameters, Recip, Context */
677 	if (vc4_state->y_scaling[channel] == VC4_SCALING_TPZ) {
678 		vc4_write_tpz(vc4_state, vc4_state->src_h[channel],
679 			      vc4_state->crtc_h);
680 		vc4_dlist_write(vc4_state, 0xc0c0c0c0);
681 	}
682 }
683 
684 static void vc4_plane_calc_load(struct drm_plane_state *state)
685 {
686 	unsigned int hvs_load_shift, vrefresh, i;
687 	struct drm_framebuffer *fb = state->fb;
688 	struct vc4_plane_state *vc4_state;
689 	struct drm_crtc_state *crtc_state;
690 	unsigned int vscale_factor;
691 
692 	vc4_state = to_vc4_plane_state(state);
693 	crtc_state = drm_atomic_get_existing_crtc_state(state->state,
694 							state->crtc);
695 	vrefresh = drm_mode_vrefresh(&crtc_state->adjusted_mode);
696 
697 	/* The HVS is able to process 2 pixels/cycle when scaling the source,
698 	 * 4 pixels/cycle otherwise.
699 	 * Alpha blending step seems to be pipelined and it's always operating
700 	 * at 4 pixels/cycle, so the limiting aspect here seems to be the
701 	 * scaler block.
702 	 * HVS load is expressed in clk-cycles/sec (AKA Hz).
703 	 */
704 	if (vc4_state->x_scaling[0] != VC4_SCALING_NONE ||
705 	    vc4_state->x_scaling[1] != VC4_SCALING_NONE ||
706 	    vc4_state->y_scaling[0] != VC4_SCALING_NONE ||
707 	    vc4_state->y_scaling[1] != VC4_SCALING_NONE)
708 		hvs_load_shift = 1;
709 	else
710 		hvs_load_shift = 2;
711 
712 	vc4_state->membus_load = 0;
713 	vc4_state->hvs_load = 0;
714 	for (i = 0; i < fb->format->num_planes; i++) {
715 		/* Even if the bandwidth/plane required for a single frame is
716 		 *
717 		 * (vc4_state->src_w[i] >> 16) * (vc4_state->src_h[i] >> 16) *
718 		 *  cpp * vrefresh
719 		 *
720 		 * when downscaling, we have to read more pixels per line in
721 		 * the time frame reserved for a single line, so the bandwidth
722 		 * demand can be punctually higher. To account for that, we
723 		 * calculate the down-scaling factor and multiply the plane
724 		 * load by this number. We're likely over-estimating the read
725 		 * demand, but that's better than under-estimating it.
726 		 */
727 		vscale_factor = DIV_ROUND_UP(vc4_state->src_h[i] >> 16,
728 					     vc4_state->crtc_h);
729 		vc4_state->membus_load += (vc4_state->src_w[i] >> 16) *
730 					  (vc4_state->src_h[i] >> 16) *
731 					  vscale_factor * fb->format->cpp[i];
732 		vc4_state->hvs_load += vc4_state->crtc_h * vc4_state->crtc_w;
733 	}
734 
735 	vc4_state->hvs_load *= vrefresh;
736 	vc4_state->hvs_load >>= hvs_load_shift;
737 	vc4_state->membus_load *= vrefresh;
738 }
739 
740 static int vc4_plane_allocate_lbm(struct drm_plane_state *state)
741 {
742 	struct drm_device *drm = state->plane->dev;
743 	struct vc4_dev *vc4 = to_vc4_dev(drm);
744 	struct drm_plane *plane = state->plane;
745 	struct vc4_plane_state *vc4_state = to_vc4_plane_state(state);
746 	unsigned long irqflags;
747 	u32 lbm_size;
748 
749 	lbm_size = vc4_lbm_size(state);
750 	if (!lbm_size)
751 		return 0;
752 
753 	if (vc4->gen == VC4_GEN_5)
754 		lbm_size = ALIGN(lbm_size, 64);
755 	else if (vc4->gen == VC4_GEN_4)
756 		lbm_size = ALIGN(lbm_size, 32);
757 
758 	drm_dbg_driver(drm, "[PLANE:%d:%s] LBM Allocation Size: %u\n",
759 		       plane->base.id, plane->name, lbm_size);
760 
761 	if (WARN_ON(!vc4_state->lbm_offset))
762 		return -EINVAL;
763 
764 	/* Allocate the LBM memory that the HVS will use for temporary
765 	 * storage due to our scaling/format conversion.
766 	 */
767 	if (!drm_mm_node_allocated(&vc4_state->lbm)) {
768 		int ret;
769 
770 		spin_lock_irqsave(&vc4->hvs->mm_lock, irqflags);
771 		ret = drm_mm_insert_node_generic(&vc4->hvs->lbm_mm,
772 						 &vc4_state->lbm,
773 						 lbm_size, 1,
774 						 0, 0);
775 		spin_unlock_irqrestore(&vc4->hvs->mm_lock, irqflags);
776 
777 		if (ret) {
778 			drm_err(drm, "Failed to allocate LBM entry: %d\n", ret);
779 			return ret;
780 		}
781 	} else {
782 		WARN_ON_ONCE(lbm_size != vc4_state->lbm.size);
783 	}
784 
785 	vc4_state->dlist[vc4_state->lbm_offset] = vc4_state->lbm.start;
786 
787 	return 0;
788 }
789 
790 /*
791  * The colorspace conversion matrices are held in 3 entries in the dlist.
792  * Create an array of them, with entries for each full and limited mode, and
793  * each supported colorspace.
794  */
795 static const u32 colorspace_coeffs[2][DRM_COLOR_ENCODING_MAX][3] = {
796 	{
797 		/* Limited range */
798 		{
799 			/* BT601 */
800 			SCALER_CSC0_ITR_R_601_5,
801 			SCALER_CSC1_ITR_R_601_5,
802 			SCALER_CSC2_ITR_R_601_5,
803 		}, {
804 			/* BT709 */
805 			SCALER_CSC0_ITR_R_709_3,
806 			SCALER_CSC1_ITR_R_709_3,
807 			SCALER_CSC2_ITR_R_709_3,
808 		}, {
809 			/* BT2020 */
810 			SCALER_CSC0_ITR_R_2020,
811 			SCALER_CSC1_ITR_R_2020,
812 			SCALER_CSC2_ITR_R_2020,
813 		}
814 	}, {
815 		/* Full range */
816 		{
817 			/* JFIF */
818 			SCALER_CSC0_JPEG_JFIF,
819 			SCALER_CSC1_JPEG_JFIF,
820 			SCALER_CSC2_JPEG_JFIF,
821 		}, {
822 			/* BT709 */
823 			SCALER_CSC0_ITR_R_709_3_FR,
824 			SCALER_CSC1_ITR_R_709_3_FR,
825 			SCALER_CSC2_ITR_R_709_3_FR,
826 		}, {
827 			/* BT2020 */
828 			SCALER_CSC0_ITR_R_2020_FR,
829 			SCALER_CSC1_ITR_R_2020_FR,
830 			SCALER_CSC2_ITR_R_2020_FR,
831 		}
832 	}
833 };
834 
835 static u32 vc4_hvs4_get_alpha_blend_mode(struct drm_plane_state *state)
836 {
837 	if (!state->fb->format->has_alpha)
838 		return VC4_SET_FIELD(SCALER_POS2_ALPHA_MODE_FIXED,
839 				     SCALER_POS2_ALPHA_MODE);
840 
841 	switch (state->pixel_blend_mode) {
842 	case DRM_MODE_BLEND_PIXEL_NONE:
843 		return VC4_SET_FIELD(SCALER_POS2_ALPHA_MODE_FIXED,
844 				     SCALER_POS2_ALPHA_MODE);
845 	default:
846 	case DRM_MODE_BLEND_PREMULTI:
847 		return VC4_SET_FIELD(SCALER_POS2_ALPHA_MODE_PIPELINE,
848 				     SCALER_POS2_ALPHA_MODE) |
849 			SCALER_POS2_ALPHA_PREMULT;
850 	case DRM_MODE_BLEND_COVERAGE:
851 		return VC4_SET_FIELD(SCALER_POS2_ALPHA_MODE_PIPELINE,
852 				     SCALER_POS2_ALPHA_MODE);
853 	}
854 }
855 
856 static u32 vc4_hvs5_get_alpha_blend_mode(struct drm_plane_state *state)
857 {
858 	if (!state->fb->format->has_alpha)
859 		return VC4_SET_FIELD(SCALER5_CTL2_ALPHA_MODE_FIXED,
860 				     SCALER5_CTL2_ALPHA_MODE);
861 
862 	switch (state->pixel_blend_mode) {
863 	case DRM_MODE_BLEND_PIXEL_NONE:
864 		return VC4_SET_FIELD(SCALER5_CTL2_ALPHA_MODE_FIXED,
865 				     SCALER5_CTL2_ALPHA_MODE);
866 	default:
867 	case DRM_MODE_BLEND_PREMULTI:
868 		return VC4_SET_FIELD(SCALER5_CTL2_ALPHA_MODE_PIPELINE,
869 				     SCALER5_CTL2_ALPHA_MODE) |
870 			SCALER5_CTL2_ALPHA_PREMULT;
871 	case DRM_MODE_BLEND_COVERAGE:
872 		return VC4_SET_FIELD(SCALER5_CTL2_ALPHA_MODE_PIPELINE,
873 				     SCALER5_CTL2_ALPHA_MODE);
874 	}
875 }
876 
877 /* Writes out a full display list for an active plane to the plane's
878  * private dlist state.
879  */
880 static int vc4_plane_mode_set(struct drm_plane *plane,
881 			      struct drm_plane_state *state)
882 {
883 	struct vc4_dev *vc4 = to_vc4_dev(plane->dev);
884 	struct vc4_plane_state *vc4_state = to_vc4_plane_state(state);
885 	struct drm_framebuffer *fb = state->fb;
886 	u32 ctl0_offset = vc4_state->dlist_count;
887 	const struct hvs_format *format = vc4_get_hvs_format(fb->format->format);
888 	u64 base_format_mod = fourcc_mod_broadcom_mod(fb->modifier);
889 	int num_planes = fb->format->num_planes;
890 	u32 h_subsample = fb->format->hsub;
891 	u32 v_subsample = fb->format->vsub;
892 	bool mix_plane_alpha;
893 	bool covers_screen;
894 	u32 scl0, scl1, pitch0;
895 	u32 tiling, src_x, src_y;
896 	u32 width, height;
897 	u32 hvs_format = format->hvs;
898 	unsigned int rotation;
899 	u32 offsets[3] = { 0 };
900 	int ret, i;
901 
902 	if (vc4_state->dlist_initialized)
903 		return 0;
904 
905 	ret = vc4_plane_setup_clipping_and_scaling(state);
906 	if (ret)
907 		return ret;
908 
909 	width = vc4_state->src_w[0] >> 16;
910 	height = vc4_state->src_h[0] >> 16;
911 
912 	/* SCL1 is used for Cb/Cr scaling of planar formats.  For RGB
913 	 * and 4:4:4, scl1 should be set to scl0 so both channels of
914 	 * the scaler do the same thing.  For YUV, the Y plane needs
915 	 * to be put in channel 1 and Cb/Cr in channel 0, so we swap
916 	 * the scl fields here.
917 	 */
918 	if (num_planes == 1) {
919 		scl0 = vc4_get_scl_field(state, 0);
920 		scl1 = scl0;
921 	} else {
922 		scl0 = vc4_get_scl_field(state, 1);
923 		scl1 = vc4_get_scl_field(state, 0);
924 	}
925 
926 	rotation = drm_rotation_simplify(state->rotation,
927 					 DRM_MODE_ROTATE_0 |
928 					 DRM_MODE_REFLECT_X |
929 					 DRM_MODE_REFLECT_Y);
930 
931 	/* We must point to the last line when Y reflection is enabled. */
932 	src_y = vc4_state->src_y >> 16;
933 	if (rotation & DRM_MODE_REFLECT_Y)
934 		src_y += height - 1;
935 
936 	src_x = vc4_state->src_x >> 16;
937 
938 	switch (base_format_mod) {
939 	case DRM_FORMAT_MOD_LINEAR:
940 		tiling = SCALER_CTL0_TILING_LINEAR;
941 		pitch0 = VC4_SET_FIELD(fb->pitches[0], SCALER_SRC_PITCH);
942 
943 		/* Adjust the base pointer to the first pixel to be scanned
944 		 * out.
945 		 */
946 		for (i = 0; i < num_planes; i++) {
947 			offsets[i] += src_y / (i ? v_subsample : 1) * fb->pitches[i];
948 			offsets[i] += src_x / (i ? h_subsample : 1) * fb->format->cpp[i];
949 		}
950 
951 		break;
952 
953 	case DRM_FORMAT_MOD_BROADCOM_VC4_T_TILED: {
954 		u32 tile_size_shift = 12; /* T tiles are 4kb */
955 		/* Whole-tile offsets, mostly for setting the pitch. */
956 		u32 tile_w_shift = fb->format->cpp[0] == 2 ? 6 : 5;
957 		u32 tile_h_shift = 5; /* 16 and 32bpp are 32 pixels high */
958 		u32 tile_w_mask = (1 << tile_w_shift) - 1;
959 		/* The height mask on 32-bit-per-pixel tiles is 63, i.e. twice
960 		 * the height (in pixels) of a 4k tile.
961 		 */
962 		u32 tile_h_mask = (2 << tile_h_shift) - 1;
963 		/* For T-tiled, the FB pitch is "how many bytes from one row to
964 		 * the next, such that
965 		 *
966 		 *	pitch * tile_h == tile_size * tiles_per_row
967 		 */
968 		u32 tiles_w = fb->pitches[0] >> (tile_size_shift - tile_h_shift);
969 		u32 tiles_l = src_x >> tile_w_shift;
970 		u32 tiles_r = tiles_w - tiles_l;
971 		u32 tiles_t = src_y >> tile_h_shift;
972 		/* Intra-tile offsets, which modify the base address (the
973 		 * SCALER_PITCH0_TILE_Y_OFFSET tells HVS how to walk from that
974 		 * base address).
975 		 */
976 		u32 tile_y = (src_y >> 4) & 1;
977 		u32 subtile_y = (src_y >> 2) & 3;
978 		u32 utile_y = src_y & 3;
979 		u32 x_off = src_x & tile_w_mask;
980 		u32 y_off = src_y & tile_h_mask;
981 
982 		/* When Y reflection is requested we must set the
983 		 * SCALER_PITCH0_TILE_LINE_DIR flag to tell HVS that all lines
984 		 * after the initial one should be fetched in descending order,
985 		 * which makes sense since we start from the last line and go
986 		 * backward.
987 		 * Don't know why we need y_off = max_y_off - y_off, but it's
988 		 * definitely required (I guess it's also related to the "going
989 		 * backward" situation).
990 		 */
991 		if (rotation & DRM_MODE_REFLECT_Y) {
992 			y_off = tile_h_mask - y_off;
993 			pitch0 = SCALER_PITCH0_TILE_LINE_DIR;
994 		} else {
995 			pitch0 = 0;
996 		}
997 
998 		tiling = SCALER_CTL0_TILING_256B_OR_T;
999 		pitch0 |= (VC4_SET_FIELD(x_off, SCALER_PITCH0_SINK_PIX) |
1000 			   VC4_SET_FIELD(y_off, SCALER_PITCH0_TILE_Y_OFFSET) |
1001 			   VC4_SET_FIELD(tiles_l, SCALER_PITCH0_TILE_WIDTH_L) |
1002 			   VC4_SET_FIELD(tiles_r, SCALER_PITCH0_TILE_WIDTH_R));
1003 		offsets[0] += tiles_t * (tiles_w << tile_size_shift);
1004 		offsets[0] += subtile_y << 8;
1005 		offsets[0] += utile_y << 4;
1006 
1007 		/* Rows of tiles alternate left-to-right and right-to-left. */
1008 		if (tiles_t & 1) {
1009 			pitch0 |= SCALER_PITCH0_TILE_INITIAL_LINE_DIR;
1010 			offsets[0] += (tiles_w - tiles_l) << tile_size_shift;
1011 			offsets[0] -= (1 + !tile_y) << 10;
1012 		} else {
1013 			offsets[0] += tiles_l << tile_size_shift;
1014 			offsets[0] += tile_y << 10;
1015 		}
1016 
1017 		break;
1018 	}
1019 
1020 	case DRM_FORMAT_MOD_BROADCOM_SAND64:
1021 	case DRM_FORMAT_MOD_BROADCOM_SAND128:
1022 	case DRM_FORMAT_MOD_BROADCOM_SAND256: {
1023 		uint32_t param = fourcc_mod_broadcom_param(fb->modifier);
1024 
1025 		if (param > SCALER_TILE_HEIGHT_MASK) {
1026 			DRM_DEBUG_KMS("SAND height too large (%d)\n",
1027 				      param);
1028 			return -EINVAL;
1029 		}
1030 
1031 		if (fb->format->format == DRM_FORMAT_P030) {
1032 			hvs_format = HVS_PIXEL_FORMAT_YCBCR_10BIT;
1033 			tiling = SCALER_CTL0_TILING_128B;
1034 		} else {
1035 			hvs_format = HVS_PIXEL_FORMAT_H264;
1036 
1037 			switch (base_format_mod) {
1038 			case DRM_FORMAT_MOD_BROADCOM_SAND64:
1039 				tiling = SCALER_CTL0_TILING_64B;
1040 				break;
1041 			case DRM_FORMAT_MOD_BROADCOM_SAND128:
1042 				tiling = SCALER_CTL0_TILING_128B;
1043 				break;
1044 			case DRM_FORMAT_MOD_BROADCOM_SAND256:
1045 				tiling = SCALER_CTL0_TILING_256B_OR_T;
1046 				break;
1047 			default:
1048 				return -EINVAL;
1049 			}
1050 		}
1051 
1052 		/* Adjust the base pointer to the first pixel to be scanned
1053 		 * out.
1054 		 *
1055 		 * For P030, y_ptr [31:4] is the 128bit word for the start pixel
1056 		 * y_ptr [3:0] is the pixel (0-11) contained within that 128bit
1057 		 * word that should be taken as the first pixel.
1058 		 * Ditto uv_ptr [31:4] vs [3:0], however [3:0] contains the
1059 		 * element within the 128bit word, eg for pixel 3 the value
1060 		 * should be 6.
1061 		 */
1062 		for (i = 0; i < num_planes; i++) {
1063 			u32 tile_w, tile, x_off, pix_per_tile;
1064 
1065 			if (fb->format->format == DRM_FORMAT_P030) {
1066 				/*
1067 				 * Spec says: bits [31:4] of the given address
1068 				 * should point to the 128-bit word containing
1069 				 * the desired starting pixel, and bits[3:0]
1070 				 * should be between 0 and 11, indicating which
1071 				 * of the 12-pixels in that 128-bit word is the
1072 				 * first pixel to be used
1073 				 */
1074 				u32 remaining_pixels = src_x % 96;
1075 				u32 aligned = remaining_pixels / 12;
1076 				u32 last_bits = remaining_pixels % 12;
1077 
1078 				x_off = aligned * 16 + last_bits;
1079 				tile_w = 128;
1080 				pix_per_tile = 96;
1081 			} else {
1082 				switch (base_format_mod) {
1083 				case DRM_FORMAT_MOD_BROADCOM_SAND64:
1084 					tile_w = 64;
1085 					break;
1086 				case DRM_FORMAT_MOD_BROADCOM_SAND128:
1087 					tile_w = 128;
1088 					break;
1089 				case DRM_FORMAT_MOD_BROADCOM_SAND256:
1090 					tile_w = 256;
1091 					break;
1092 				default:
1093 					return -EINVAL;
1094 				}
1095 				pix_per_tile = tile_w / fb->format->cpp[0];
1096 				x_off = (src_x % pix_per_tile) /
1097 					(i ? h_subsample : 1) *
1098 					fb->format->cpp[i];
1099 			}
1100 
1101 			tile = src_x / pix_per_tile;
1102 
1103 			offsets[i] += param * tile_w * tile;
1104 			offsets[i] += src_y / (i ? v_subsample : 1) * tile_w;
1105 			offsets[i] += x_off & ~(i ? 1 : 0);
1106 		}
1107 
1108 		pitch0 = VC4_SET_FIELD(param, SCALER_TILE_HEIGHT);
1109 		break;
1110 	}
1111 
1112 	default:
1113 		DRM_DEBUG_KMS("Unsupported FB tiling flag 0x%16llx",
1114 			      (long long)fb->modifier);
1115 		return -EINVAL;
1116 	}
1117 
1118 	/* fetch an extra pixel if we don't actually line up with the left edge. */
1119 	if ((vc4_state->src_x & 0xffff) && vc4_state->src_x < (state->fb->width << 16))
1120 		width++;
1121 
1122 	/* same for the right side */
1123 	if (((vc4_state->src_x + vc4_state->src_w[0]) & 0xffff) &&
1124 	    vc4_state->src_x + vc4_state->src_w[0] < (state->fb->width << 16))
1125 		width++;
1126 
1127 	/* now for the top */
1128 	if ((vc4_state->src_y & 0xffff) && vc4_state->src_y < (state->fb->height << 16))
1129 		height++;
1130 
1131 	/* and the bottom */
1132 	if (((vc4_state->src_y + vc4_state->src_h[0]) & 0xffff) &&
1133 	    vc4_state->src_y + vc4_state->src_h[0] < (state->fb->height << 16))
1134 		height++;
1135 
1136 	/* For YUV444 the hardware wants double the width, otherwise it doesn't
1137 	 * fetch full width of chroma
1138 	 */
1139 	if (format->drm == DRM_FORMAT_YUV444 || format->drm == DRM_FORMAT_YVU444)
1140 		width <<= 1;
1141 
1142 	/* Don't waste cycles mixing with plane alpha if the set alpha
1143 	 * is opaque or there is no per-pixel alpha information.
1144 	 * In any case we use the alpha property value as the fixed alpha.
1145 	 */
1146 	mix_plane_alpha = state->alpha != DRM_BLEND_ALPHA_OPAQUE &&
1147 			  fb->format->has_alpha;
1148 
1149 	if (vc4->gen == VC4_GEN_4) {
1150 	/* Control word */
1151 		vc4_dlist_write(vc4_state,
1152 				SCALER_CTL0_VALID |
1153 				(rotation & DRM_MODE_REFLECT_X ? SCALER_CTL0_HFLIP : 0) |
1154 				(rotation & DRM_MODE_REFLECT_Y ? SCALER_CTL0_VFLIP : 0) |
1155 				VC4_SET_FIELD(SCALER_CTL0_RGBA_EXPAND_ROUND, SCALER_CTL0_RGBA_EXPAND) |
1156 				(format->pixel_order << SCALER_CTL0_ORDER_SHIFT) |
1157 				(hvs_format << SCALER_CTL0_PIXEL_FORMAT_SHIFT) |
1158 				VC4_SET_FIELD(tiling, SCALER_CTL0_TILING) |
1159 				(vc4_state->is_unity ? SCALER_CTL0_UNITY : 0) |
1160 				VC4_SET_FIELD(scl0, SCALER_CTL0_SCL0) |
1161 				VC4_SET_FIELD(scl1, SCALER_CTL0_SCL1));
1162 
1163 		/* Position Word 0: Image Positions and Alpha Value */
1164 		vc4_state->pos0_offset = vc4_state->dlist_count;
1165 		vc4_dlist_write(vc4_state,
1166 				VC4_SET_FIELD(state->alpha >> 8, SCALER_POS0_FIXED_ALPHA) |
1167 				VC4_SET_FIELD(vc4_state->crtc_x, SCALER_POS0_START_X) |
1168 				VC4_SET_FIELD(vc4_state->crtc_y, SCALER_POS0_START_Y));
1169 
1170 		/* Position Word 1: Scaled Image Dimensions. */
1171 		if (!vc4_state->is_unity) {
1172 			vc4_dlist_write(vc4_state,
1173 					VC4_SET_FIELD(vc4_state->crtc_w,
1174 						      SCALER_POS1_SCL_WIDTH) |
1175 					VC4_SET_FIELD(vc4_state->crtc_h,
1176 						      SCALER_POS1_SCL_HEIGHT));
1177 		}
1178 
1179 		/* Position Word 2: Source Image Size, Alpha */
1180 		vc4_state->pos2_offset = vc4_state->dlist_count;
1181 		vc4_dlist_write(vc4_state,
1182 				(mix_plane_alpha ? SCALER_POS2_ALPHA_MIX : 0) |
1183 				vc4_hvs4_get_alpha_blend_mode(state) |
1184 				VC4_SET_FIELD(width, SCALER_POS2_WIDTH) |
1185 				VC4_SET_FIELD(height, SCALER_POS2_HEIGHT));
1186 
1187 		/* Position Word 3: Context.  Written by the HVS. */
1188 		vc4_dlist_write(vc4_state, 0xc0c0c0c0);
1189 
1190 	} else {
1191 		/* Control word */
1192 		vc4_dlist_write(vc4_state,
1193 				SCALER_CTL0_VALID |
1194 				(format->pixel_order_hvs5 << SCALER_CTL0_ORDER_SHIFT) |
1195 				(hvs_format << SCALER_CTL0_PIXEL_FORMAT_SHIFT) |
1196 				VC4_SET_FIELD(tiling, SCALER_CTL0_TILING) |
1197 				(vc4_state->is_unity ?
1198 						SCALER5_CTL0_UNITY : 0) |
1199 				VC4_SET_FIELD(scl0, SCALER_CTL0_SCL0) |
1200 				VC4_SET_FIELD(scl1, SCALER_CTL0_SCL1) |
1201 				SCALER5_CTL0_ALPHA_EXPAND |
1202 				SCALER5_CTL0_RGB_EXPAND);
1203 
1204 		/* Position Word 0: Image Positions and Alpha Value */
1205 		vc4_state->pos0_offset = vc4_state->dlist_count;
1206 		vc4_dlist_write(vc4_state,
1207 				(rotation & DRM_MODE_REFLECT_Y ?
1208 						SCALER5_POS0_VFLIP : 0) |
1209 				VC4_SET_FIELD(vc4_state->crtc_x,
1210 					      SCALER_POS0_START_X) |
1211 				(rotation & DRM_MODE_REFLECT_X ?
1212 					      SCALER5_POS0_HFLIP : 0) |
1213 				VC4_SET_FIELD(vc4_state->crtc_y,
1214 					      SCALER5_POS0_START_Y)
1215 			       );
1216 
1217 		/* Control Word 2 */
1218 		vc4_dlist_write(vc4_state,
1219 				VC4_SET_FIELD(state->alpha >> 4,
1220 					      SCALER5_CTL2_ALPHA) |
1221 				vc4_hvs5_get_alpha_blend_mode(state) |
1222 				(mix_plane_alpha ?
1223 					SCALER5_CTL2_ALPHA_MIX : 0)
1224 			       );
1225 
1226 		/* Position Word 1: Scaled Image Dimensions. */
1227 		if (!vc4_state->is_unity) {
1228 			vc4_dlist_write(vc4_state,
1229 					VC4_SET_FIELD(vc4_state->crtc_w,
1230 						      SCALER5_POS1_SCL_WIDTH) |
1231 					VC4_SET_FIELD(vc4_state->crtc_h,
1232 						      SCALER5_POS1_SCL_HEIGHT));
1233 		}
1234 
1235 		/* Position Word 2: Source Image Size */
1236 		vc4_state->pos2_offset = vc4_state->dlist_count;
1237 		vc4_dlist_write(vc4_state,
1238 				VC4_SET_FIELD(width, SCALER5_POS2_WIDTH) |
1239 				VC4_SET_FIELD(height, SCALER5_POS2_HEIGHT));
1240 
1241 		/* Position Word 3: Context.  Written by the HVS. */
1242 		vc4_dlist_write(vc4_state, 0xc0c0c0c0);
1243 	}
1244 
1245 
1246 	/* Pointer Word 0/1/2: RGB / Y / Cb / Cr Pointers
1247 	 *
1248 	 * The pointers may be any byte address.
1249 	 */
1250 	vc4_state->ptr0_offset[0] = vc4_state->dlist_count;
1251 
1252 	for (i = 0; i < num_planes; i++) {
1253 		struct drm_gem_dma_object *bo = drm_fb_dma_get_gem_obj(fb, i);
1254 
1255 		vc4_dlist_write(vc4_state, bo->dma_addr + fb->offsets[i] + offsets[i]);
1256 	}
1257 
1258 	/* Pointer Context Word 0/1/2: Written by the HVS */
1259 	for (i = 0; i < num_planes; i++)
1260 		vc4_dlist_write(vc4_state, 0xc0c0c0c0);
1261 
1262 	/* Pitch word 0 */
1263 	vc4_dlist_write(vc4_state, pitch0);
1264 
1265 	/* Pitch word 1/2 */
1266 	for (i = 1; i < num_planes; i++) {
1267 		if (hvs_format != HVS_PIXEL_FORMAT_H264 &&
1268 		    hvs_format != HVS_PIXEL_FORMAT_YCBCR_10BIT) {
1269 			vc4_dlist_write(vc4_state,
1270 					VC4_SET_FIELD(fb->pitches[i],
1271 						      SCALER_SRC_PITCH));
1272 		} else {
1273 			vc4_dlist_write(vc4_state, pitch0);
1274 		}
1275 	}
1276 
1277 	/* Colorspace conversion words */
1278 	if (vc4_state->is_yuv) {
1279 		enum drm_color_encoding color_encoding = state->color_encoding;
1280 		enum drm_color_range color_range = state->color_range;
1281 		const u32 *ccm;
1282 
1283 		if (color_encoding >= DRM_COLOR_ENCODING_MAX)
1284 			color_encoding = DRM_COLOR_YCBCR_BT601;
1285 		if (color_range >= DRM_COLOR_RANGE_MAX)
1286 			color_range = DRM_COLOR_YCBCR_LIMITED_RANGE;
1287 
1288 		ccm = colorspace_coeffs[color_range][color_encoding];
1289 
1290 		vc4_dlist_write(vc4_state, ccm[0]);
1291 		vc4_dlist_write(vc4_state, ccm[1]);
1292 		vc4_dlist_write(vc4_state, ccm[2]);
1293 	}
1294 
1295 	vc4_state->lbm_offset = 0;
1296 
1297 	if (vc4_state->x_scaling[0] != VC4_SCALING_NONE ||
1298 	    vc4_state->x_scaling[1] != VC4_SCALING_NONE ||
1299 	    vc4_state->y_scaling[0] != VC4_SCALING_NONE ||
1300 	    vc4_state->y_scaling[1] != VC4_SCALING_NONE) {
1301 		/* Reserve a slot for the LBM Base Address. The real value will
1302 		 * be set when calling vc4_plane_allocate_lbm().
1303 		 */
1304 		if (vc4_state->y_scaling[0] != VC4_SCALING_NONE ||
1305 		    vc4_state->y_scaling[1] != VC4_SCALING_NONE) {
1306 			vc4_state->lbm_offset = vc4_state->dlist_count;
1307 			vc4_dlist_counter_increment(vc4_state);
1308 		}
1309 
1310 		if (num_planes > 1) {
1311 			/* Emit Cb/Cr as channel 0 and Y as channel
1312 			 * 1. This matches how we set up scl0/scl1
1313 			 * above.
1314 			 */
1315 			vc4_write_scaling_parameters(state, 1);
1316 		}
1317 		vc4_write_scaling_parameters(state, 0);
1318 
1319 		/* If any PPF setup was done, then all the kernel
1320 		 * pointers get uploaded.
1321 		 */
1322 		if (vc4_state->x_scaling[0] == VC4_SCALING_PPF ||
1323 		    vc4_state->y_scaling[0] == VC4_SCALING_PPF ||
1324 		    vc4_state->x_scaling[1] == VC4_SCALING_PPF ||
1325 		    vc4_state->y_scaling[1] == VC4_SCALING_PPF) {
1326 			u32 kernel = VC4_SET_FIELD(vc4->hvs->mitchell_netravali_filter.start,
1327 						   SCALER_PPF_KERNEL_OFFSET);
1328 
1329 			/* HPPF plane 0 */
1330 			vc4_dlist_write(vc4_state, kernel);
1331 			/* VPPF plane 0 */
1332 			vc4_dlist_write(vc4_state, kernel);
1333 			/* HPPF plane 1 */
1334 			vc4_dlist_write(vc4_state, kernel);
1335 			/* VPPF plane 1 */
1336 			vc4_dlist_write(vc4_state, kernel);
1337 		}
1338 	}
1339 
1340 	vc4_state->dlist[ctl0_offset] |=
1341 		VC4_SET_FIELD(vc4_state->dlist_count, SCALER_CTL0_SIZE);
1342 
1343 	/* crtc_* are already clipped coordinates. */
1344 	covers_screen = vc4_state->crtc_x == 0 && vc4_state->crtc_y == 0 &&
1345 			vc4_state->crtc_w == state->crtc->mode.hdisplay &&
1346 			vc4_state->crtc_h == state->crtc->mode.vdisplay;
1347 	/* Background fill might be necessary when the plane has per-pixel
1348 	 * alpha content or a non-opaque plane alpha and could blend from the
1349 	 * background or does not cover the entire screen.
1350 	 */
1351 	vc4_state->needs_bg_fill = fb->format->has_alpha || !covers_screen ||
1352 				   state->alpha != DRM_BLEND_ALPHA_OPAQUE;
1353 
1354 	/* Flag the dlist as initialized to avoid checking it twice in case
1355 	 * the async update check already called vc4_plane_mode_set() and
1356 	 * decided to fallback to sync update because async update was not
1357 	 * possible.
1358 	 */
1359 	vc4_state->dlist_initialized = 1;
1360 
1361 	vc4_plane_calc_load(state);
1362 
1363 	return 0;
1364 }
1365 
1366 /* If a modeset involves changing the setup of a plane, the atomic
1367  * infrastructure will call this to validate a proposed plane setup.
1368  * However, if a plane isn't getting updated, this (and the
1369  * corresponding vc4_plane_atomic_update) won't get called.  Thus, we
1370  * compute the dlist here and have all active plane dlists get updated
1371  * in the CRTC's flush.
1372  */
1373 static int vc4_plane_atomic_check(struct drm_plane *plane,
1374 				  struct drm_atomic_state *state)
1375 {
1376 	struct drm_plane_state *new_plane_state = drm_atomic_get_new_plane_state(state,
1377 										 plane);
1378 	struct vc4_plane_state *vc4_state = to_vc4_plane_state(new_plane_state);
1379 	int ret;
1380 
1381 	vc4_state->dlist_count = 0;
1382 
1383 	if (!plane_enabled(new_plane_state))
1384 		return 0;
1385 
1386 	ret = vc4_plane_mode_set(plane, new_plane_state);
1387 	if (ret)
1388 		return ret;
1389 
1390 	ret = vc4_plane_allocate_lbm(new_plane_state);
1391 	if (ret)
1392 		return ret;
1393 
1394 	return 0;
1395 }
1396 
1397 static void vc4_plane_atomic_update(struct drm_plane *plane,
1398 				    struct drm_atomic_state *state)
1399 {
1400 	/* No contents here.  Since we don't know where in the CRTC's
1401 	 * dlist we should be stored, our dlist is uploaded to the
1402 	 * hardware with vc4_plane_write_dlist() at CRTC atomic_flush
1403 	 * time.
1404 	 */
1405 }
1406 
1407 u32 vc4_plane_write_dlist(struct drm_plane *plane, u32 __iomem *dlist)
1408 {
1409 	struct vc4_plane_state *vc4_state = to_vc4_plane_state(plane->state);
1410 	int i;
1411 	int idx;
1412 
1413 	if (!drm_dev_enter(plane->dev, &idx))
1414 		goto out;
1415 
1416 	vc4_state->hw_dlist = dlist;
1417 
1418 	/* Can't memcpy_toio() because it needs to be 32-bit writes. */
1419 	for (i = 0; i < vc4_state->dlist_count; i++)
1420 		writel(vc4_state->dlist[i], &dlist[i]);
1421 
1422 	drm_dev_exit(idx);
1423 
1424 out:
1425 	return vc4_state->dlist_count;
1426 }
1427 
1428 u32 vc4_plane_dlist_size(const struct drm_plane_state *state)
1429 {
1430 	const struct vc4_plane_state *vc4_state = to_vc4_plane_state(state);
1431 
1432 	return vc4_state->dlist_count;
1433 }
1434 
1435 /* Updates the plane to immediately (well, once the FIFO needs
1436  * refilling) scan out from at a new framebuffer.
1437  */
1438 void vc4_plane_async_set_fb(struct drm_plane *plane, struct drm_framebuffer *fb)
1439 {
1440 	struct vc4_plane_state *vc4_state = to_vc4_plane_state(plane->state);
1441 	struct drm_gem_dma_object *bo = drm_fb_dma_get_gem_obj(fb, 0);
1442 	uint32_t addr;
1443 	int idx;
1444 
1445 	if (!drm_dev_enter(plane->dev, &idx))
1446 		return;
1447 
1448 	/* We're skipping the address adjustment for negative origin,
1449 	 * because this is only called on the primary plane.
1450 	 */
1451 	WARN_ON_ONCE(plane->state->crtc_x < 0 || plane->state->crtc_y < 0);
1452 	addr = bo->dma_addr + fb->offsets[0];
1453 
1454 	/* Write the new address into the hardware immediately.  The
1455 	 * scanout will start from this address as soon as the FIFO
1456 	 * needs to refill with pixels.
1457 	 */
1458 	writel(addr, &vc4_state->hw_dlist[vc4_state->ptr0_offset[0]]);
1459 
1460 	/* Also update the CPU-side dlist copy, so that any later
1461 	 * atomic updates that don't do a new modeset on our plane
1462 	 * also use our updated address.
1463 	 */
1464 	vc4_state->dlist[vc4_state->ptr0_offset[0]] = addr;
1465 
1466 	drm_dev_exit(idx);
1467 }
1468 
1469 static void vc4_plane_atomic_async_update(struct drm_plane *plane,
1470 					  struct drm_atomic_state *state)
1471 {
1472 	struct drm_plane_state *new_plane_state = drm_atomic_get_new_plane_state(state,
1473 										 plane);
1474 	struct vc4_plane_state *vc4_state, *new_vc4_state;
1475 	int idx;
1476 
1477 	if (!drm_dev_enter(plane->dev, &idx))
1478 		return;
1479 
1480 	swap(plane->state->fb, new_plane_state->fb);
1481 	plane->state->crtc_x = new_plane_state->crtc_x;
1482 	plane->state->crtc_y = new_plane_state->crtc_y;
1483 	plane->state->crtc_w = new_plane_state->crtc_w;
1484 	plane->state->crtc_h = new_plane_state->crtc_h;
1485 	plane->state->src_x = new_plane_state->src_x;
1486 	plane->state->src_y = new_plane_state->src_y;
1487 	plane->state->src_w = new_plane_state->src_w;
1488 	plane->state->src_h = new_plane_state->src_h;
1489 	plane->state->alpha = new_plane_state->alpha;
1490 	plane->state->pixel_blend_mode = new_plane_state->pixel_blend_mode;
1491 	plane->state->rotation = new_plane_state->rotation;
1492 	plane->state->zpos = new_plane_state->zpos;
1493 	plane->state->normalized_zpos = new_plane_state->normalized_zpos;
1494 	plane->state->color_encoding = new_plane_state->color_encoding;
1495 	plane->state->color_range = new_plane_state->color_range;
1496 	plane->state->src = new_plane_state->src;
1497 	plane->state->dst = new_plane_state->dst;
1498 	plane->state->visible = new_plane_state->visible;
1499 
1500 	new_vc4_state = to_vc4_plane_state(new_plane_state);
1501 	vc4_state = to_vc4_plane_state(plane->state);
1502 
1503 	vc4_state->crtc_x = new_vc4_state->crtc_x;
1504 	vc4_state->crtc_y = new_vc4_state->crtc_y;
1505 	vc4_state->crtc_h = new_vc4_state->crtc_h;
1506 	vc4_state->crtc_w = new_vc4_state->crtc_w;
1507 	vc4_state->src_x = new_vc4_state->src_x;
1508 	vc4_state->src_y = new_vc4_state->src_y;
1509 	memcpy(vc4_state->src_w, new_vc4_state->src_w,
1510 	       sizeof(vc4_state->src_w));
1511 	memcpy(vc4_state->src_h, new_vc4_state->src_h,
1512 	       sizeof(vc4_state->src_h));
1513 	memcpy(vc4_state->x_scaling, new_vc4_state->x_scaling,
1514 	       sizeof(vc4_state->x_scaling));
1515 	memcpy(vc4_state->y_scaling, new_vc4_state->y_scaling,
1516 	       sizeof(vc4_state->y_scaling));
1517 	vc4_state->is_unity = new_vc4_state->is_unity;
1518 	vc4_state->is_yuv = new_vc4_state->is_yuv;
1519 	vc4_state->needs_bg_fill = new_vc4_state->needs_bg_fill;
1520 
1521 	/* Update the current vc4_state pos0, pos2 and ptr0 dlist entries. */
1522 	vc4_state->dlist[vc4_state->pos0_offset] =
1523 		new_vc4_state->dlist[vc4_state->pos0_offset];
1524 	vc4_state->dlist[vc4_state->pos2_offset] =
1525 		new_vc4_state->dlist[vc4_state->pos2_offset];
1526 	vc4_state->dlist[vc4_state->ptr0_offset[0]] =
1527 		new_vc4_state->dlist[vc4_state->ptr0_offset[0]];
1528 
1529 	/* Note that we can't just call vc4_plane_write_dlist()
1530 	 * because that would smash the context data that the HVS is
1531 	 * currently using.
1532 	 */
1533 	writel(vc4_state->dlist[vc4_state->pos0_offset],
1534 	       &vc4_state->hw_dlist[vc4_state->pos0_offset]);
1535 	writel(vc4_state->dlist[vc4_state->pos2_offset],
1536 	       &vc4_state->hw_dlist[vc4_state->pos2_offset]);
1537 	writel(vc4_state->dlist[vc4_state->ptr0_offset[0]],
1538 	       &vc4_state->hw_dlist[vc4_state->ptr0_offset[0]]);
1539 
1540 	drm_dev_exit(idx);
1541 }
1542 
1543 static int vc4_plane_atomic_async_check(struct drm_plane *plane,
1544 					struct drm_atomic_state *state)
1545 {
1546 	struct drm_plane_state *new_plane_state = drm_atomic_get_new_plane_state(state,
1547 										 plane);
1548 	struct vc4_plane_state *old_vc4_state, *new_vc4_state;
1549 	int ret;
1550 	u32 i;
1551 
1552 	ret = vc4_plane_mode_set(plane, new_plane_state);
1553 	if (ret)
1554 		return ret;
1555 
1556 	old_vc4_state = to_vc4_plane_state(plane->state);
1557 	new_vc4_state = to_vc4_plane_state(new_plane_state);
1558 
1559 	if (!new_vc4_state->hw_dlist)
1560 		return -EINVAL;
1561 
1562 	if (old_vc4_state->dlist_count != new_vc4_state->dlist_count ||
1563 	    old_vc4_state->pos0_offset != new_vc4_state->pos0_offset ||
1564 	    old_vc4_state->pos2_offset != new_vc4_state->pos2_offset ||
1565 	    old_vc4_state->ptr0_offset[0] != new_vc4_state->ptr0_offset[0] ||
1566 	    vc4_lbm_size(plane->state) != vc4_lbm_size(new_plane_state))
1567 		return -EINVAL;
1568 
1569 	/* Only pos0, pos2 and ptr0 DWORDS can be updated in an async update
1570 	 * if anything else has changed, fallback to a sync update.
1571 	 */
1572 	for (i = 0; i < new_vc4_state->dlist_count; i++) {
1573 		if (i == new_vc4_state->pos0_offset ||
1574 		    i == new_vc4_state->pos2_offset ||
1575 		    i == new_vc4_state->ptr0_offset[0] ||
1576 		    (new_vc4_state->lbm_offset &&
1577 		     i == new_vc4_state->lbm_offset))
1578 			continue;
1579 
1580 		if (new_vc4_state->dlist[i] != old_vc4_state->dlist[i])
1581 			return -EINVAL;
1582 	}
1583 
1584 	return 0;
1585 }
1586 
1587 static int vc4_prepare_fb(struct drm_plane *plane,
1588 			  struct drm_plane_state *state)
1589 {
1590 	struct vc4_bo *bo;
1591 	int ret;
1592 
1593 	if (!state->fb)
1594 		return 0;
1595 
1596 	bo = to_vc4_bo(&drm_fb_dma_get_gem_obj(state->fb, 0)->base);
1597 
1598 	ret = drm_gem_plane_helper_prepare_fb(plane, state);
1599 	if (ret)
1600 		return ret;
1601 
1602 	return vc4_bo_inc_usecnt(bo);
1603 }
1604 
1605 static void vc4_cleanup_fb(struct drm_plane *plane,
1606 			   struct drm_plane_state *state)
1607 {
1608 	struct vc4_bo *bo;
1609 
1610 	if (!state->fb)
1611 		return;
1612 
1613 	bo = to_vc4_bo(&drm_fb_dma_get_gem_obj(state->fb, 0)->base);
1614 	vc4_bo_dec_usecnt(bo);
1615 }
1616 
1617 static const struct drm_plane_helper_funcs vc4_plane_helper_funcs = {
1618 	.atomic_check = vc4_plane_atomic_check,
1619 	.atomic_update = vc4_plane_atomic_update,
1620 	.prepare_fb = vc4_prepare_fb,
1621 	.cleanup_fb = vc4_cleanup_fb,
1622 	.atomic_async_check = vc4_plane_atomic_async_check,
1623 	.atomic_async_update = vc4_plane_atomic_async_update,
1624 };
1625 
1626 static const struct drm_plane_helper_funcs vc5_plane_helper_funcs = {
1627 	.atomic_check = vc4_plane_atomic_check,
1628 	.atomic_update = vc4_plane_atomic_update,
1629 	.atomic_async_check = vc4_plane_atomic_async_check,
1630 	.atomic_async_update = vc4_plane_atomic_async_update,
1631 };
1632 
1633 static bool vc4_format_mod_supported(struct drm_plane *plane,
1634 				     uint32_t format,
1635 				     uint64_t modifier)
1636 {
1637 	/* Support T_TILING for RGB formats only. */
1638 	switch (format) {
1639 	case DRM_FORMAT_XRGB8888:
1640 	case DRM_FORMAT_ARGB8888:
1641 	case DRM_FORMAT_ABGR8888:
1642 	case DRM_FORMAT_XBGR8888:
1643 	case DRM_FORMAT_RGB565:
1644 	case DRM_FORMAT_BGR565:
1645 	case DRM_FORMAT_ARGB1555:
1646 	case DRM_FORMAT_XRGB1555:
1647 		switch (fourcc_mod_broadcom_mod(modifier)) {
1648 		case DRM_FORMAT_MOD_LINEAR:
1649 		case DRM_FORMAT_MOD_BROADCOM_VC4_T_TILED:
1650 			return true;
1651 		default:
1652 			return false;
1653 		}
1654 	case DRM_FORMAT_NV12:
1655 	case DRM_FORMAT_NV21:
1656 		switch (fourcc_mod_broadcom_mod(modifier)) {
1657 		case DRM_FORMAT_MOD_LINEAR:
1658 		case DRM_FORMAT_MOD_BROADCOM_SAND64:
1659 		case DRM_FORMAT_MOD_BROADCOM_SAND128:
1660 		case DRM_FORMAT_MOD_BROADCOM_SAND256:
1661 			return true;
1662 		default:
1663 			return false;
1664 		}
1665 	case DRM_FORMAT_P030:
1666 		switch (fourcc_mod_broadcom_mod(modifier)) {
1667 		case DRM_FORMAT_MOD_BROADCOM_SAND128:
1668 			return true;
1669 		default:
1670 			return false;
1671 		}
1672 	case DRM_FORMAT_RGBX1010102:
1673 	case DRM_FORMAT_BGRX1010102:
1674 	case DRM_FORMAT_RGBA1010102:
1675 	case DRM_FORMAT_BGRA1010102:
1676 	case DRM_FORMAT_XRGB4444:
1677 	case DRM_FORMAT_ARGB4444:
1678 	case DRM_FORMAT_XBGR4444:
1679 	case DRM_FORMAT_ABGR4444:
1680 	case DRM_FORMAT_RGBX4444:
1681 	case DRM_FORMAT_RGBA4444:
1682 	case DRM_FORMAT_BGRX4444:
1683 	case DRM_FORMAT_BGRA4444:
1684 	case DRM_FORMAT_RGB332:
1685 	case DRM_FORMAT_BGR233:
1686 	case DRM_FORMAT_YUV422:
1687 	case DRM_FORMAT_YVU422:
1688 	case DRM_FORMAT_YUV420:
1689 	case DRM_FORMAT_YVU420:
1690 	case DRM_FORMAT_NV16:
1691 	case DRM_FORMAT_NV61:
1692 	default:
1693 		return (modifier == DRM_FORMAT_MOD_LINEAR);
1694 	}
1695 }
1696 
1697 static const struct drm_plane_funcs vc4_plane_funcs = {
1698 	.update_plane = drm_atomic_helper_update_plane,
1699 	.disable_plane = drm_atomic_helper_disable_plane,
1700 	.reset = vc4_plane_reset,
1701 	.atomic_duplicate_state = vc4_plane_duplicate_state,
1702 	.atomic_destroy_state = vc4_plane_destroy_state,
1703 	.format_mod_supported = vc4_format_mod_supported,
1704 };
1705 
1706 struct drm_plane *vc4_plane_init(struct drm_device *dev,
1707 				 enum drm_plane_type type,
1708 				 uint32_t possible_crtcs)
1709 {
1710 	struct vc4_dev *vc4 = to_vc4_dev(dev);
1711 	struct drm_plane *plane;
1712 	struct vc4_plane *vc4_plane;
1713 	u32 formats[ARRAY_SIZE(hvs_formats)];
1714 	int num_formats = 0;
1715 	unsigned i;
1716 	static const uint64_t modifiers[] = {
1717 		DRM_FORMAT_MOD_BROADCOM_VC4_T_TILED,
1718 		DRM_FORMAT_MOD_BROADCOM_SAND128,
1719 		DRM_FORMAT_MOD_BROADCOM_SAND64,
1720 		DRM_FORMAT_MOD_BROADCOM_SAND256,
1721 		DRM_FORMAT_MOD_LINEAR,
1722 		DRM_FORMAT_MOD_INVALID
1723 	};
1724 
1725 	for (i = 0; i < ARRAY_SIZE(hvs_formats); i++) {
1726 		if (!hvs_formats[i].hvs5_only || vc4->gen == VC4_GEN_5) {
1727 			formats[num_formats] = hvs_formats[i].drm;
1728 			num_formats++;
1729 		}
1730 	}
1731 
1732 	vc4_plane = drmm_universal_plane_alloc(dev, struct vc4_plane, base,
1733 					       possible_crtcs,
1734 					       &vc4_plane_funcs,
1735 					       formats, num_formats,
1736 					       modifiers, type, NULL);
1737 	if (IS_ERR(vc4_plane))
1738 		return ERR_CAST(vc4_plane);
1739 	plane = &vc4_plane->base;
1740 
1741 	if (vc4->gen == VC4_GEN_5)
1742 		drm_plane_helper_add(plane, &vc5_plane_helper_funcs);
1743 	else
1744 		drm_plane_helper_add(plane, &vc4_plane_helper_funcs);
1745 
1746 	drm_plane_create_alpha_property(plane);
1747 	drm_plane_create_blend_mode_property(plane,
1748 					     BIT(DRM_MODE_BLEND_PIXEL_NONE) |
1749 					     BIT(DRM_MODE_BLEND_PREMULTI) |
1750 					     BIT(DRM_MODE_BLEND_COVERAGE));
1751 	drm_plane_create_rotation_property(plane, DRM_MODE_ROTATE_0,
1752 					   DRM_MODE_ROTATE_0 |
1753 					   DRM_MODE_ROTATE_180 |
1754 					   DRM_MODE_REFLECT_X |
1755 					   DRM_MODE_REFLECT_Y);
1756 
1757 	drm_plane_create_color_properties(plane,
1758 					  BIT(DRM_COLOR_YCBCR_BT601) |
1759 					  BIT(DRM_COLOR_YCBCR_BT709) |
1760 					  BIT(DRM_COLOR_YCBCR_BT2020),
1761 					  BIT(DRM_COLOR_YCBCR_LIMITED_RANGE) |
1762 					  BIT(DRM_COLOR_YCBCR_FULL_RANGE),
1763 					  DRM_COLOR_YCBCR_BT709,
1764 					  DRM_COLOR_YCBCR_LIMITED_RANGE);
1765 
1766 	if (type == DRM_PLANE_TYPE_PRIMARY)
1767 		drm_plane_create_zpos_immutable_property(plane, 0);
1768 
1769 	return plane;
1770 }
1771 
1772 #define VC4_NUM_OVERLAY_PLANES	16
1773 
1774 int vc4_plane_create_additional_planes(struct drm_device *drm)
1775 {
1776 	struct drm_plane *cursor_plane;
1777 	struct drm_crtc *crtc;
1778 	unsigned int i;
1779 
1780 	/* Set up some arbitrary number of planes.  We're not limited
1781 	 * by a set number of physical registers, just the space in
1782 	 * the HVS (16k) and how small an plane can be (28 bytes).
1783 	 * However, each plane we set up takes up some memory, and
1784 	 * increases the cost of looping over planes, which atomic
1785 	 * modesetting does quite a bit.  As a result, we pick a
1786 	 * modest number of planes to expose, that should hopefully
1787 	 * still cover any sane usecase.
1788 	 */
1789 	for (i = 0; i < VC4_NUM_OVERLAY_PLANES; i++) {
1790 		struct drm_plane *plane =
1791 			vc4_plane_init(drm, DRM_PLANE_TYPE_OVERLAY,
1792 				       GENMASK(drm->mode_config.num_crtc - 1, 0));
1793 
1794 		if (IS_ERR(plane))
1795 			continue;
1796 
1797 		/* Create zpos property. Max of all the overlays + 1 primary +
1798 		 * 1 cursor plane on a crtc.
1799 		 */
1800 		drm_plane_create_zpos_property(plane, i + 1, 1,
1801 					       VC4_NUM_OVERLAY_PLANES + 1);
1802 	}
1803 
1804 	drm_for_each_crtc(crtc, drm) {
1805 		/* Set up the legacy cursor after overlay initialization,
1806 		 * since the zpos fallback is that planes are rendered by plane
1807 		 * ID order, and that then puts the cursor on top.
1808 		 */
1809 		cursor_plane = vc4_plane_init(drm, DRM_PLANE_TYPE_CURSOR,
1810 					      drm_crtc_mask(crtc));
1811 		if (!IS_ERR(cursor_plane)) {
1812 			crtc->cursor = cursor_plane;
1813 
1814 			drm_plane_create_zpos_property(cursor_plane,
1815 						       VC4_NUM_OVERLAY_PLANES + 1,
1816 						       1,
1817 						       VC4_NUM_OVERLAY_PLANES + 1);
1818 		}
1819 	}
1820 
1821 	return 0;
1822 }
1823