xref: /openbmc/linux/drivers/gpu/drm/vc4/vc4_plane.c (revision 0edbfea5)
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 "vc4_drv.h"
22 #include "vc4_regs.h"
23 #include "drm_atomic_helper.h"
24 #include "drm_fb_cma_helper.h"
25 #include "drm_plane_helper.h"
26 
27 enum vc4_scaling_mode {
28 	VC4_SCALING_NONE,
29 	VC4_SCALING_TPZ,
30 	VC4_SCALING_PPF,
31 };
32 
33 struct vc4_plane_state {
34 	struct drm_plane_state base;
35 	/* System memory copy of the display list for this element, computed
36 	 * at atomic_check time.
37 	 */
38 	u32 *dlist;
39 	u32 dlist_size; /* Number of dwords allocated for the display list */
40 	u32 dlist_count; /* Number of used dwords in the display list. */
41 
42 	/* Offset in the dlist to various words, for pageflip or
43 	 * cursor updates.
44 	 */
45 	u32 pos0_offset;
46 	u32 pos2_offset;
47 	u32 ptr0_offset;
48 
49 	/* Offset where the plane's dlist was last stored in the
50 	 * hardware at vc4_crtc_atomic_flush() time.
51 	 */
52 	u32 __iomem *hw_dlist;
53 
54 	/* Clipped coordinates of the plane on the display. */
55 	int crtc_x, crtc_y, crtc_w, crtc_h;
56 	/* Clipped area being scanned from in the FB. */
57 	u32 src_x, src_y;
58 
59 	u32 src_w[2], src_h[2];
60 
61 	/* Scaling selection for the RGB/Y plane and the Cb/Cr planes. */
62 	enum vc4_scaling_mode x_scaling[2], y_scaling[2];
63 	bool is_unity;
64 	bool is_yuv;
65 
66 	/* Offset to start scanning out from the start of the plane's
67 	 * BO.
68 	 */
69 	u32 offsets[3];
70 
71 	/* Our allocation in LBM for temporary storage during scaling. */
72 	struct drm_mm_node lbm;
73 };
74 
75 static inline struct vc4_plane_state *
76 to_vc4_plane_state(struct drm_plane_state *state)
77 {
78 	return (struct vc4_plane_state *)state;
79 }
80 
81 static const struct hvs_format {
82 	u32 drm; /* DRM_FORMAT_* */
83 	u32 hvs; /* HVS_FORMAT_* */
84 	u32 pixel_order;
85 	bool has_alpha;
86 	bool flip_cbcr;
87 } hvs_formats[] = {
88 	{
89 		.drm = DRM_FORMAT_XRGB8888, .hvs = HVS_PIXEL_FORMAT_RGBA8888,
90 		.pixel_order = HVS_PIXEL_ORDER_ABGR, .has_alpha = false,
91 	},
92 	{
93 		.drm = DRM_FORMAT_ARGB8888, .hvs = HVS_PIXEL_FORMAT_RGBA8888,
94 		.pixel_order = HVS_PIXEL_ORDER_ABGR, .has_alpha = true,
95 	},
96 	{
97 		.drm = DRM_FORMAT_RGB565, .hvs = HVS_PIXEL_FORMAT_RGB565,
98 		.pixel_order = HVS_PIXEL_ORDER_XRGB, .has_alpha = false,
99 	},
100 	{
101 		.drm = DRM_FORMAT_BGR565, .hvs = HVS_PIXEL_FORMAT_RGB565,
102 		.pixel_order = HVS_PIXEL_ORDER_XBGR, .has_alpha = false,
103 	},
104 	{
105 		.drm = DRM_FORMAT_ARGB1555, .hvs = HVS_PIXEL_FORMAT_RGBA5551,
106 		.pixel_order = HVS_PIXEL_ORDER_ABGR, .has_alpha = true,
107 	},
108 	{
109 		.drm = DRM_FORMAT_XRGB1555, .hvs = HVS_PIXEL_FORMAT_RGBA5551,
110 		.pixel_order = HVS_PIXEL_ORDER_ABGR, .has_alpha = false,
111 	},
112 	{
113 		.drm = DRM_FORMAT_YUV422,
114 		.hvs = HVS_PIXEL_FORMAT_YCBCR_YUV422_3PLANE,
115 	},
116 	{
117 		.drm = DRM_FORMAT_YVU422,
118 		.hvs = HVS_PIXEL_FORMAT_YCBCR_YUV422_3PLANE,
119 		.flip_cbcr = true,
120 	},
121 	{
122 		.drm = DRM_FORMAT_YUV420,
123 		.hvs = HVS_PIXEL_FORMAT_YCBCR_YUV420_3PLANE,
124 	},
125 	{
126 		.drm = DRM_FORMAT_YVU420,
127 		.hvs = HVS_PIXEL_FORMAT_YCBCR_YUV420_3PLANE,
128 		.flip_cbcr = true,
129 	},
130 	{
131 		.drm = DRM_FORMAT_NV12,
132 		.hvs = HVS_PIXEL_FORMAT_YCBCR_YUV420_2PLANE,
133 	},
134 	{
135 		.drm = DRM_FORMAT_NV16,
136 		.hvs = HVS_PIXEL_FORMAT_YCBCR_YUV422_2PLANE,
137 	},
138 };
139 
140 static const struct hvs_format *vc4_get_hvs_format(u32 drm_format)
141 {
142 	unsigned i;
143 
144 	for (i = 0; i < ARRAY_SIZE(hvs_formats); i++) {
145 		if (hvs_formats[i].drm == drm_format)
146 			return &hvs_formats[i];
147 	}
148 
149 	return NULL;
150 }
151 
152 static enum vc4_scaling_mode vc4_get_scaling_mode(u32 src, u32 dst)
153 {
154 	if (dst > src)
155 		return VC4_SCALING_PPF;
156 	else if (dst < src)
157 		return VC4_SCALING_TPZ;
158 	else
159 		return VC4_SCALING_NONE;
160 }
161 
162 static bool plane_enabled(struct drm_plane_state *state)
163 {
164 	return state->fb && state->crtc;
165 }
166 
167 static struct drm_plane_state *vc4_plane_duplicate_state(struct drm_plane *plane)
168 {
169 	struct vc4_plane_state *vc4_state;
170 
171 	if (WARN_ON(!plane->state))
172 		return NULL;
173 
174 	vc4_state = kmemdup(plane->state, sizeof(*vc4_state), GFP_KERNEL);
175 	if (!vc4_state)
176 		return NULL;
177 
178 	memset(&vc4_state->lbm, 0, sizeof(vc4_state->lbm));
179 
180 	__drm_atomic_helper_plane_duplicate_state(plane, &vc4_state->base);
181 
182 	if (vc4_state->dlist) {
183 		vc4_state->dlist = kmemdup(vc4_state->dlist,
184 					   vc4_state->dlist_count * 4,
185 					   GFP_KERNEL);
186 		if (!vc4_state->dlist) {
187 			kfree(vc4_state);
188 			return NULL;
189 		}
190 		vc4_state->dlist_size = vc4_state->dlist_count;
191 	}
192 
193 	return &vc4_state->base;
194 }
195 
196 static void vc4_plane_destroy_state(struct drm_plane *plane,
197 				    struct drm_plane_state *state)
198 {
199 	struct vc4_dev *vc4 = to_vc4_dev(plane->dev);
200 	struct vc4_plane_state *vc4_state = to_vc4_plane_state(state);
201 
202 	if (vc4_state->lbm.allocated) {
203 		unsigned long irqflags;
204 
205 		spin_lock_irqsave(&vc4->hvs->mm_lock, irqflags);
206 		drm_mm_remove_node(&vc4_state->lbm);
207 		spin_unlock_irqrestore(&vc4->hvs->mm_lock, irqflags);
208 	}
209 
210 	kfree(vc4_state->dlist);
211 	__drm_atomic_helper_plane_destroy_state(&vc4_state->base);
212 	kfree(state);
213 }
214 
215 /* Called during init to allocate the plane's atomic state. */
216 static void vc4_plane_reset(struct drm_plane *plane)
217 {
218 	struct vc4_plane_state *vc4_state;
219 
220 	WARN_ON(plane->state);
221 
222 	vc4_state = kzalloc(sizeof(*vc4_state), GFP_KERNEL);
223 	if (!vc4_state)
224 		return;
225 
226 	plane->state = &vc4_state->base;
227 	vc4_state->base.plane = plane;
228 }
229 
230 static void vc4_dlist_write(struct vc4_plane_state *vc4_state, u32 val)
231 {
232 	if (vc4_state->dlist_count == vc4_state->dlist_size) {
233 		u32 new_size = max(4u, vc4_state->dlist_count * 2);
234 		u32 *new_dlist = kmalloc(new_size * 4, GFP_KERNEL);
235 
236 		if (!new_dlist)
237 			return;
238 		memcpy(new_dlist, vc4_state->dlist, vc4_state->dlist_count * 4);
239 
240 		kfree(vc4_state->dlist);
241 		vc4_state->dlist = new_dlist;
242 		vc4_state->dlist_size = new_size;
243 	}
244 
245 	vc4_state->dlist[vc4_state->dlist_count++] = val;
246 }
247 
248 /* Returns the scl0/scl1 field based on whether the dimensions need to
249  * be up/down/non-scaled.
250  *
251  * This is a replication of a table from the spec.
252  */
253 static u32 vc4_get_scl_field(struct drm_plane_state *state, int plane)
254 {
255 	struct vc4_plane_state *vc4_state = to_vc4_plane_state(state);
256 
257 	switch (vc4_state->x_scaling[plane] << 2 | vc4_state->y_scaling[plane]) {
258 	case VC4_SCALING_PPF << 2 | VC4_SCALING_PPF:
259 		return SCALER_CTL0_SCL_H_PPF_V_PPF;
260 	case VC4_SCALING_TPZ << 2 | VC4_SCALING_PPF:
261 		return SCALER_CTL0_SCL_H_TPZ_V_PPF;
262 	case VC4_SCALING_PPF << 2 | VC4_SCALING_TPZ:
263 		return SCALER_CTL0_SCL_H_PPF_V_TPZ;
264 	case VC4_SCALING_TPZ << 2 | VC4_SCALING_TPZ:
265 		return SCALER_CTL0_SCL_H_TPZ_V_TPZ;
266 	case VC4_SCALING_PPF << 2 | VC4_SCALING_NONE:
267 		return SCALER_CTL0_SCL_H_PPF_V_NONE;
268 	case VC4_SCALING_NONE << 2 | VC4_SCALING_PPF:
269 		return SCALER_CTL0_SCL_H_NONE_V_PPF;
270 	case VC4_SCALING_NONE << 2 | VC4_SCALING_TPZ:
271 		return SCALER_CTL0_SCL_H_NONE_V_TPZ;
272 	case VC4_SCALING_TPZ << 2 | VC4_SCALING_NONE:
273 		return SCALER_CTL0_SCL_H_TPZ_V_NONE;
274 	default:
275 	case VC4_SCALING_NONE << 2 | VC4_SCALING_NONE:
276 		/* The unity case is independently handled by
277 		 * SCALER_CTL0_UNITY.
278 		 */
279 		return 0;
280 	}
281 }
282 
283 static int vc4_plane_setup_clipping_and_scaling(struct drm_plane_state *state)
284 {
285 	struct drm_plane *plane = state->plane;
286 	struct vc4_plane_state *vc4_state = to_vc4_plane_state(state);
287 	struct drm_framebuffer *fb = state->fb;
288 	struct drm_gem_cma_object *bo = drm_fb_cma_get_gem_obj(fb, 0);
289 	u32 subpixel_src_mask = (1 << 16) - 1;
290 	u32 format = fb->pixel_format;
291 	int num_planes = drm_format_num_planes(format);
292 	u32 h_subsample = 1;
293 	u32 v_subsample = 1;
294 	int i;
295 
296 	for (i = 0; i < num_planes; i++)
297 		vc4_state->offsets[i] = bo->paddr + fb->offsets[i];
298 
299 	/* We don't support subpixel source positioning for scaling. */
300 	if ((state->src_x & subpixel_src_mask) ||
301 	    (state->src_y & subpixel_src_mask) ||
302 	    (state->src_w & subpixel_src_mask) ||
303 	    (state->src_h & subpixel_src_mask)) {
304 		return -EINVAL;
305 	}
306 
307 	vc4_state->src_x = state->src_x >> 16;
308 	vc4_state->src_y = state->src_y >> 16;
309 	vc4_state->src_w[0] = state->src_w >> 16;
310 	vc4_state->src_h[0] = state->src_h >> 16;
311 
312 	vc4_state->crtc_x = state->crtc_x;
313 	vc4_state->crtc_y = state->crtc_y;
314 	vc4_state->crtc_w = state->crtc_w;
315 	vc4_state->crtc_h = state->crtc_h;
316 
317 	vc4_state->x_scaling[0] = vc4_get_scaling_mode(vc4_state->src_w[0],
318 						       vc4_state->crtc_w);
319 	vc4_state->y_scaling[0] = vc4_get_scaling_mode(vc4_state->src_h[0],
320 						       vc4_state->crtc_h);
321 
322 	if (num_planes > 1) {
323 		vc4_state->is_yuv = true;
324 
325 		h_subsample = drm_format_horz_chroma_subsampling(format);
326 		v_subsample = drm_format_vert_chroma_subsampling(format);
327 		vc4_state->src_w[1] = vc4_state->src_w[0] / h_subsample;
328 		vc4_state->src_h[1] = vc4_state->src_h[0] / v_subsample;
329 
330 		vc4_state->x_scaling[1] =
331 			vc4_get_scaling_mode(vc4_state->src_w[1],
332 					     vc4_state->crtc_w);
333 		vc4_state->y_scaling[1] =
334 			vc4_get_scaling_mode(vc4_state->src_h[1],
335 					     vc4_state->crtc_h);
336 
337 		/* YUV conversion requires that scaling be enabled,
338 		 * even on a plane that's otherwise 1:1.  Choose TPZ
339 		 * for simplicity.
340 		 */
341 		if (vc4_state->x_scaling[0] == VC4_SCALING_NONE)
342 			vc4_state->x_scaling[0] = VC4_SCALING_TPZ;
343 		if (vc4_state->y_scaling[0] == VC4_SCALING_NONE)
344 			vc4_state->y_scaling[0] = VC4_SCALING_TPZ;
345 	}
346 
347 	vc4_state->is_unity = (vc4_state->x_scaling[0] == VC4_SCALING_NONE &&
348 			       vc4_state->y_scaling[0] == VC4_SCALING_NONE &&
349 			       vc4_state->x_scaling[1] == VC4_SCALING_NONE &&
350 			       vc4_state->y_scaling[1] == VC4_SCALING_NONE);
351 
352 	/* No configuring scaling on the cursor plane, since it gets
353 	   non-vblank-synced updates, and scaling requires requires
354 	   LBM changes which have to be vblank-synced.
355 	 */
356 	if (plane->type == DRM_PLANE_TYPE_CURSOR && !vc4_state->is_unity)
357 		return -EINVAL;
358 
359 	/* Clamp the on-screen start x/y to 0.  The hardware doesn't
360 	 * support negative y, and negative x wastes bandwidth.
361 	 */
362 	if (vc4_state->crtc_x < 0) {
363 		for (i = 0; i < num_planes; i++) {
364 			u32 cpp = drm_format_plane_cpp(fb->pixel_format, i);
365 			u32 subs = ((i == 0) ? 1 : h_subsample);
366 
367 			vc4_state->offsets[i] += (cpp *
368 						  (-vc4_state->crtc_x) / subs);
369 		}
370 		vc4_state->src_w[0] += vc4_state->crtc_x;
371 		vc4_state->src_w[1] += vc4_state->crtc_x / h_subsample;
372 		vc4_state->crtc_x = 0;
373 	}
374 
375 	if (vc4_state->crtc_y < 0) {
376 		for (i = 0; i < num_planes; i++) {
377 			u32 subs = ((i == 0) ? 1 : v_subsample);
378 
379 			vc4_state->offsets[i] += (fb->pitches[i] *
380 						  (-vc4_state->crtc_y) / subs);
381 		}
382 		vc4_state->src_h[0] += vc4_state->crtc_y;
383 		vc4_state->src_h[1] += vc4_state->crtc_y / v_subsample;
384 		vc4_state->crtc_y = 0;
385 	}
386 
387 	return 0;
388 }
389 
390 static void vc4_write_tpz(struct vc4_plane_state *vc4_state, u32 src, u32 dst)
391 {
392 	u32 scale, recip;
393 
394 	scale = (1 << 16) * src / dst;
395 
396 	/* The specs note that while the reciprocal would be defined
397 	 * as (1<<32)/scale, ~0 is close enough.
398 	 */
399 	recip = ~0 / scale;
400 
401 	vc4_dlist_write(vc4_state,
402 			VC4_SET_FIELD(scale, SCALER_TPZ0_SCALE) |
403 			VC4_SET_FIELD(0, SCALER_TPZ0_IPHASE));
404 	vc4_dlist_write(vc4_state,
405 			VC4_SET_FIELD(recip, SCALER_TPZ1_RECIP));
406 }
407 
408 static void vc4_write_ppf(struct vc4_plane_state *vc4_state, u32 src, u32 dst)
409 {
410 	u32 scale = (1 << 16) * src / dst;
411 
412 	vc4_dlist_write(vc4_state,
413 			SCALER_PPF_AGC |
414 			VC4_SET_FIELD(scale, SCALER_PPF_SCALE) |
415 			VC4_SET_FIELD(0, SCALER_PPF_IPHASE));
416 }
417 
418 static u32 vc4_lbm_size(struct drm_plane_state *state)
419 {
420 	struct vc4_plane_state *vc4_state = to_vc4_plane_state(state);
421 	/* This is the worst case number.  One of the two sizes will
422 	 * be used depending on the scaling configuration.
423 	 */
424 	u32 pix_per_line = max(vc4_state->src_w[0], (u32)vc4_state->crtc_w);
425 	u32 lbm;
426 
427 	if (!vc4_state->is_yuv) {
428 		if (vc4_state->is_unity)
429 			return 0;
430 		else if (vc4_state->y_scaling[0] == VC4_SCALING_TPZ)
431 			lbm = pix_per_line * 8;
432 		else {
433 			/* In special cases, this multiplier might be 12. */
434 			lbm = pix_per_line * 16;
435 		}
436 	} else {
437 		/* There are cases for this going down to a multiplier
438 		 * of 2, but according to the firmware source, the
439 		 * table in the docs is somewhat wrong.
440 		 */
441 		lbm = pix_per_line * 16;
442 	}
443 
444 	lbm = roundup(lbm, 32);
445 
446 	return lbm;
447 }
448 
449 static void vc4_write_scaling_parameters(struct drm_plane_state *state,
450 					 int channel)
451 {
452 	struct vc4_plane_state *vc4_state = to_vc4_plane_state(state);
453 
454 	/* Ch0 H-PPF Word 0: Scaling Parameters */
455 	if (vc4_state->x_scaling[channel] == VC4_SCALING_PPF) {
456 		vc4_write_ppf(vc4_state,
457 			      vc4_state->src_w[channel], vc4_state->crtc_w);
458 	}
459 
460 	/* Ch0 V-PPF Words 0-1: Scaling Parameters, Context */
461 	if (vc4_state->y_scaling[channel] == VC4_SCALING_PPF) {
462 		vc4_write_ppf(vc4_state,
463 			      vc4_state->src_h[channel], vc4_state->crtc_h);
464 		vc4_dlist_write(vc4_state, 0xc0c0c0c0);
465 	}
466 
467 	/* Ch0 H-TPZ Words 0-1: Scaling Parameters, Recip */
468 	if (vc4_state->x_scaling[channel] == VC4_SCALING_TPZ) {
469 		vc4_write_tpz(vc4_state,
470 			      vc4_state->src_w[channel], vc4_state->crtc_w);
471 	}
472 
473 	/* Ch0 V-TPZ Words 0-2: Scaling Parameters, Recip, Context */
474 	if (vc4_state->y_scaling[channel] == VC4_SCALING_TPZ) {
475 		vc4_write_tpz(vc4_state,
476 			      vc4_state->src_h[channel], vc4_state->crtc_h);
477 		vc4_dlist_write(vc4_state, 0xc0c0c0c0);
478 	}
479 }
480 
481 /* Writes out a full display list for an active plane to the plane's
482  * private dlist state.
483  */
484 static int vc4_plane_mode_set(struct drm_plane *plane,
485 			      struct drm_plane_state *state)
486 {
487 	struct vc4_dev *vc4 = to_vc4_dev(plane->dev);
488 	struct vc4_plane_state *vc4_state = to_vc4_plane_state(state);
489 	struct drm_framebuffer *fb = state->fb;
490 	u32 ctl0_offset = vc4_state->dlist_count;
491 	const struct hvs_format *format = vc4_get_hvs_format(fb->pixel_format);
492 	int num_planes = drm_format_num_planes(format->drm);
493 	u32 scl0, scl1;
494 	u32 lbm_size;
495 	unsigned long irqflags;
496 	int ret, i;
497 
498 	ret = vc4_plane_setup_clipping_and_scaling(state);
499 	if (ret)
500 		return ret;
501 
502 	/* Allocate the LBM memory that the HVS will use for temporary
503 	 * storage due to our scaling/format conversion.
504 	 */
505 	lbm_size = vc4_lbm_size(state);
506 	if (lbm_size) {
507 		if (!vc4_state->lbm.allocated) {
508 			spin_lock_irqsave(&vc4->hvs->mm_lock, irqflags);
509 			ret = drm_mm_insert_node(&vc4->hvs->lbm_mm,
510 						 &vc4_state->lbm,
511 						 lbm_size, 32, 0);
512 			spin_unlock_irqrestore(&vc4->hvs->mm_lock, irqflags);
513 		} else {
514 			WARN_ON_ONCE(lbm_size != vc4_state->lbm.size);
515 		}
516 	}
517 
518 	if (ret)
519 		return ret;
520 
521 	/* SCL1 is used for Cb/Cr scaling of planar formats.  For RGB
522 	 * and 4:4:4, scl1 should be set to scl0 so both channels of
523 	 * the scaler do the same thing.  For YUV, the Y plane needs
524 	 * to be put in channel 1 and Cb/Cr in channel 0, so we swap
525 	 * the scl fields here.
526 	 */
527 	if (num_planes == 1) {
528 		scl0 = vc4_get_scl_field(state, 1);
529 		scl1 = scl0;
530 	} else {
531 		scl0 = vc4_get_scl_field(state, 1);
532 		scl1 = vc4_get_scl_field(state, 0);
533 	}
534 
535 	/* Control word */
536 	vc4_dlist_write(vc4_state,
537 			SCALER_CTL0_VALID |
538 			(format->pixel_order << SCALER_CTL0_ORDER_SHIFT) |
539 			(format->hvs << SCALER_CTL0_PIXEL_FORMAT_SHIFT) |
540 			(vc4_state->is_unity ? SCALER_CTL0_UNITY : 0) |
541 			VC4_SET_FIELD(scl0, SCALER_CTL0_SCL0) |
542 			VC4_SET_FIELD(scl1, SCALER_CTL0_SCL1));
543 
544 	/* Position Word 0: Image Positions and Alpha Value */
545 	vc4_state->pos0_offset = vc4_state->dlist_count;
546 	vc4_dlist_write(vc4_state,
547 			VC4_SET_FIELD(0xff, SCALER_POS0_FIXED_ALPHA) |
548 			VC4_SET_FIELD(vc4_state->crtc_x, SCALER_POS0_START_X) |
549 			VC4_SET_FIELD(vc4_state->crtc_y, SCALER_POS0_START_Y));
550 
551 	/* Position Word 1: Scaled Image Dimensions. */
552 	if (!vc4_state->is_unity) {
553 		vc4_dlist_write(vc4_state,
554 				VC4_SET_FIELD(vc4_state->crtc_w,
555 					      SCALER_POS1_SCL_WIDTH) |
556 				VC4_SET_FIELD(vc4_state->crtc_h,
557 					      SCALER_POS1_SCL_HEIGHT));
558 	}
559 
560 	/* Position Word 2: Source Image Size, Alpha Mode */
561 	vc4_state->pos2_offset = vc4_state->dlist_count;
562 	vc4_dlist_write(vc4_state,
563 			VC4_SET_FIELD(format->has_alpha ?
564 				      SCALER_POS2_ALPHA_MODE_PIPELINE :
565 				      SCALER_POS2_ALPHA_MODE_FIXED,
566 				      SCALER_POS2_ALPHA_MODE) |
567 			VC4_SET_FIELD(vc4_state->src_w[0], SCALER_POS2_WIDTH) |
568 			VC4_SET_FIELD(vc4_state->src_h[0], SCALER_POS2_HEIGHT));
569 
570 	/* Position Word 3: Context.  Written by the HVS. */
571 	vc4_dlist_write(vc4_state, 0xc0c0c0c0);
572 
573 
574 	/* Pointer Word 0/1/2: RGB / Y / Cb / Cr Pointers
575 	 *
576 	 * The pointers may be any byte address.
577 	 */
578 	vc4_state->ptr0_offset = vc4_state->dlist_count;
579 	if (!format->flip_cbcr) {
580 		for (i = 0; i < num_planes; i++)
581 			vc4_dlist_write(vc4_state, vc4_state->offsets[i]);
582 	} else {
583 		WARN_ON_ONCE(num_planes != 3);
584 		vc4_dlist_write(vc4_state, vc4_state->offsets[0]);
585 		vc4_dlist_write(vc4_state, vc4_state->offsets[2]);
586 		vc4_dlist_write(vc4_state, vc4_state->offsets[1]);
587 	}
588 
589 	/* Pointer Context Word 0/1/2: Written by the HVS */
590 	for (i = 0; i < num_planes; i++)
591 		vc4_dlist_write(vc4_state, 0xc0c0c0c0);
592 
593 	/* Pitch word 0/1/2 */
594 	for (i = 0; i < num_planes; i++) {
595 		vc4_dlist_write(vc4_state,
596 				VC4_SET_FIELD(fb->pitches[i], SCALER_SRC_PITCH));
597 	}
598 
599 	/* Colorspace conversion words */
600 	if (vc4_state->is_yuv) {
601 		vc4_dlist_write(vc4_state, SCALER_CSC0_ITR_R_601_5);
602 		vc4_dlist_write(vc4_state, SCALER_CSC1_ITR_R_601_5);
603 		vc4_dlist_write(vc4_state, SCALER_CSC2_ITR_R_601_5);
604 	}
605 
606 	if (!vc4_state->is_unity) {
607 		/* LBM Base Address. */
608 		if (vc4_state->y_scaling[0] != VC4_SCALING_NONE ||
609 		    vc4_state->y_scaling[1] != VC4_SCALING_NONE) {
610 			vc4_dlist_write(vc4_state, vc4_state->lbm.start);
611 		}
612 
613 		if (num_planes > 1) {
614 			/* Emit Cb/Cr as channel 0 and Y as channel
615 			 * 1. This matches how we set up scl0/scl1
616 			 * above.
617 			 */
618 			vc4_write_scaling_parameters(state, 1);
619 		}
620 		vc4_write_scaling_parameters(state, 0);
621 
622 		/* If any PPF setup was done, then all the kernel
623 		 * pointers get uploaded.
624 		 */
625 		if (vc4_state->x_scaling[0] == VC4_SCALING_PPF ||
626 		    vc4_state->y_scaling[0] == VC4_SCALING_PPF ||
627 		    vc4_state->x_scaling[1] == VC4_SCALING_PPF ||
628 		    vc4_state->y_scaling[1] == VC4_SCALING_PPF) {
629 			u32 kernel = VC4_SET_FIELD(vc4->hvs->mitchell_netravali_filter.start,
630 						   SCALER_PPF_KERNEL_OFFSET);
631 
632 			/* HPPF plane 0 */
633 			vc4_dlist_write(vc4_state, kernel);
634 			/* VPPF plane 0 */
635 			vc4_dlist_write(vc4_state, kernel);
636 			/* HPPF plane 1 */
637 			vc4_dlist_write(vc4_state, kernel);
638 			/* VPPF plane 1 */
639 			vc4_dlist_write(vc4_state, kernel);
640 		}
641 	}
642 
643 	vc4_state->dlist[ctl0_offset] |=
644 		VC4_SET_FIELD(vc4_state->dlist_count, SCALER_CTL0_SIZE);
645 
646 	return 0;
647 }
648 
649 /* If a modeset involves changing the setup of a plane, the atomic
650  * infrastructure will call this to validate a proposed plane setup.
651  * However, if a plane isn't getting updated, this (and the
652  * corresponding vc4_plane_atomic_update) won't get called.  Thus, we
653  * compute the dlist here and have all active plane dlists get updated
654  * in the CRTC's flush.
655  */
656 static int vc4_plane_atomic_check(struct drm_plane *plane,
657 				  struct drm_plane_state *state)
658 {
659 	struct vc4_plane_state *vc4_state = to_vc4_plane_state(state);
660 
661 	vc4_state->dlist_count = 0;
662 
663 	if (plane_enabled(state))
664 		return vc4_plane_mode_set(plane, state);
665 	else
666 		return 0;
667 }
668 
669 static void vc4_plane_atomic_update(struct drm_plane *plane,
670 				    struct drm_plane_state *old_state)
671 {
672 	/* No contents here.  Since we don't know where in the CRTC's
673 	 * dlist we should be stored, our dlist is uploaded to the
674 	 * hardware with vc4_plane_write_dlist() at CRTC atomic_flush
675 	 * time.
676 	 */
677 }
678 
679 u32 vc4_plane_write_dlist(struct drm_plane *plane, u32 __iomem *dlist)
680 {
681 	struct vc4_plane_state *vc4_state = to_vc4_plane_state(plane->state);
682 	int i;
683 
684 	vc4_state->hw_dlist = dlist;
685 
686 	/* Can't memcpy_toio() because it needs to be 32-bit writes. */
687 	for (i = 0; i < vc4_state->dlist_count; i++)
688 		writel(vc4_state->dlist[i], &dlist[i]);
689 
690 	return vc4_state->dlist_count;
691 }
692 
693 u32 vc4_plane_dlist_size(struct drm_plane_state *state)
694 {
695 	struct vc4_plane_state *vc4_state = to_vc4_plane_state(state);
696 
697 	return vc4_state->dlist_count;
698 }
699 
700 /* Updates the plane to immediately (well, once the FIFO needs
701  * refilling) scan out from at a new framebuffer.
702  */
703 void vc4_plane_async_set_fb(struct drm_plane *plane, struct drm_framebuffer *fb)
704 {
705 	struct vc4_plane_state *vc4_state = to_vc4_plane_state(plane->state);
706 	struct drm_gem_cma_object *bo = drm_fb_cma_get_gem_obj(fb, 0);
707 	uint32_t addr;
708 
709 	/* We're skipping the address adjustment for negative origin,
710 	 * because this is only called on the primary plane.
711 	 */
712 	WARN_ON_ONCE(plane->state->crtc_x < 0 || plane->state->crtc_y < 0);
713 	addr = bo->paddr + fb->offsets[0];
714 
715 	/* Write the new address into the hardware immediately.  The
716 	 * scanout will start from this address as soon as the FIFO
717 	 * needs to refill with pixels.
718 	 */
719 	writel(addr, &vc4_state->hw_dlist[vc4_state->ptr0_offset]);
720 
721 	/* Also update the CPU-side dlist copy, so that any later
722 	 * atomic updates that don't do a new modeset on our plane
723 	 * also use our updated address.
724 	 */
725 	vc4_state->dlist[vc4_state->ptr0_offset] = addr;
726 }
727 
728 static const struct drm_plane_helper_funcs vc4_plane_helper_funcs = {
729 	.prepare_fb = NULL,
730 	.cleanup_fb = NULL,
731 	.atomic_check = vc4_plane_atomic_check,
732 	.atomic_update = vc4_plane_atomic_update,
733 };
734 
735 static void vc4_plane_destroy(struct drm_plane *plane)
736 {
737 	drm_plane_helper_disable(plane);
738 	drm_plane_cleanup(plane);
739 }
740 
741 /* Implements immediate (non-vblank-synced) updates of the cursor
742  * position, or falls back to the atomic helper otherwise.
743  */
744 static int
745 vc4_update_plane(struct drm_plane *plane,
746 		 struct drm_crtc *crtc,
747 		 struct drm_framebuffer *fb,
748 		 int crtc_x, int crtc_y,
749 		 unsigned int crtc_w, unsigned int crtc_h,
750 		 uint32_t src_x, uint32_t src_y,
751 		 uint32_t src_w, uint32_t src_h)
752 {
753 	struct drm_plane_state *plane_state;
754 	struct vc4_plane_state *vc4_state;
755 
756 	if (plane != crtc->cursor)
757 		goto out;
758 
759 	plane_state = plane->state;
760 	vc4_state = to_vc4_plane_state(plane_state);
761 
762 	if (!plane_state)
763 		goto out;
764 
765 	/* If we're changing the cursor contents, do that in the
766 	 * normal vblank-synced atomic path.
767 	 */
768 	if (fb != plane_state->fb)
769 		goto out;
770 
771 	/* No configuring new scaling in the fast path. */
772 	if (crtc_w != plane_state->crtc_w ||
773 	    crtc_h != plane_state->crtc_h ||
774 	    src_w != plane_state->src_w ||
775 	    src_h != plane_state->src_h) {
776 		goto out;
777 	}
778 
779 	/* Set the cursor's position on the screen.  This is the
780 	 * expected change from the drm_mode_cursor_universal()
781 	 * helper.
782 	 */
783 	plane_state->crtc_x = crtc_x;
784 	plane_state->crtc_y = crtc_y;
785 
786 	/* Allow changing the start position within the cursor BO, if
787 	 * that matters.
788 	 */
789 	plane_state->src_x = src_x;
790 	plane_state->src_y = src_y;
791 
792 	/* Update the display list based on the new crtc_x/y. */
793 	vc4_plane_atomic_check(plane, plane_state);
794 
795 	/* Note that we can't just call vc4_plane_write_dlist()
796 	 * because that would smash the context data that the HVS is
797 	 * currently using.
798 	 */
799 	writel(vc4_state->dlist[vc4_state->pos0_offset],
800 	       &vc4_state->hw_dlist[vc4_state->pos0_offset]);
801 	writel(vc4_state->dlist[vc4_state->pos2_offset],
802 	       &vc4_state->hw_dlist[vc4_state->pos2_offset]);
803 	writel(vc4_state->dlist[vc4_state->ptr0_offset],
804 	       &vc4_state->hw_dlist[vc4_state->ptr0_offset]);
805 
806 	return 0;
807 
808 out:
809 	return drm_atomic_helper_update_plane(plane, crtc, fb,
810 					      crtc_x, crtc_y,
811 					      crtc_w, crtc_h,
812 					      src_x, src_y,
813 					      src_w, src_h);
814 }
815 
816 static const struct drm_plane_funcs vc4_plane_funcs = {
817 	.update_plane = vc4_update_plane,
818 	.disable_plane = drm_atomic_helper_disable_plane,
819 	.destroy = vc4_plane_destroy,
820 	.set_property = NULL,
821 	.reset = vc4_plane_reset,
822 	.atomic_duplicate_state = vc4_plane_duplicate_state,
823 	.atomic_destroy_state = vc4_plane_destroy_state,
824 };
825 
826 struct drm_plane *vc4_plane_init(struct drm_device *dev,
827 				 enum drm_plane_type type)
828 {
829 	struct drm_plane *plane = NULL;
830 	struct vc4_plane *vc4_plane;
831 	u32 formats[ARRAY_SIZE(hvs_formats)];
832 	u32 num_formats = 0;
833 	int ret = 0;
834 	unsigned i;
835 
836 	vc4_plane = devm_kzalloc(dev->dev, sizeof(*vc4_plane),
837 				 GFP_KERNEL);
838 	if (!vc4_plane) {
839 		ret = -ENOMEM;
840 		goto fail;
841 	}
842 
843 	for (i = 0; i < ARRAY_SIZE(hvs_formats); i++) {
844 		/* Don't allow YUV in cursor planes, since that means
845 		 * tuning on the scaler, which we don't allow for the
846 		 * cursor.
847 		 */
848 		if (type != DRM_PLANE_TYPE_CURSOR ||
849 		    hvs_formats[i].hvs < HVS_PIXEL_FORMAT_YCBCR_YUV420_3PLANE) {
850 			formats[num_formats++] = hvs_formats[i].drm;
851 		}
852 	}
853 	plane = &vc4_plane->base;
854 	ret = drm_universal_plane_init(dev, plane, 0xff,
855 				       &vc4_plane_funcs,
856 				       formats, num_formats,
857 				       type, NULL);
858 
859 	drm_plane_helper_add(plane, &vc4_plane_helper_funcs);
860 
861 	return plane;
862 fail:
863 	if (plane)
864 		vc4_plane_destroy(plane);
865 
866 	return ERR_PTR(ret);
867 }
868