xref: /openbmc/linux/drivers/gpu/drm/vc4/vc4_plane.c (revision 5a244f48)
1 /*
2  * Copyright (C) 2015 Broadcom
3  *
4  * This program is free software; you can redistribute it and/or modify
5  * it under the terms of the GNU General Public License version 2 as
6  * published by the Free Software Foundation.
7  */
8 
9 /**
10  * DOC: VC4 plane module
11  *
12  * Each DRM plane is a layer of pixels being scanned out by the HVS.
13  *
14  * At atomic modeset check time, we compute the HVS display element
15  * state that would be necessary for displaying the plane (giving us a
16  * chance to figure out if a plane configuration is invalid), then at
17  * atomic flush time the CRTC will ask us to write our element state
18  * into the region of the HVS that it has allocated for us.
19  */
20 
21 #include <drm/drm_atomic.h>
22 #include <drm/drm_atomic_helper.h>
23 #include <drm/drm_fb_cma_helper.h>
24 #include <drm/drm_plane_helper.h>
25 
26 #include "vc4_drv.h"
27 #include "vc4_regs.h"
28 
29 enum vc4_scaling_mode {
30 	VC4_SCALING_NONE,
31 	VC4_SCALING_TPZ,
32 	VC4_SCALING_PPF,
33 };
34 
35 struct vc4_plane_state {
36 	struct drm_plane_state base;
37 	/* System memory copy of the display list for this element, computed
38 	 * at atomic_check time.
39 	 */
40 	u32 *dlist;
41 	u32 dlist_size; /* Number of dwords allocated for the display list */
42 	u32 dlist_count; /* Number of used dwords in the display list. */
43 
44 	/* Offset in the dlist to various words, for pageflip or
45 	 * cursor updates.
46 	 */
47 	u32 pos0_offset;
48 	u32 pos2_offset;
49 	u32 ptr0_offset;
50 
51 	/* Offset where the plane's dlist was last stored in the
52 	 * hardware at vc4_crtc_atomic_flush() time.
53 	 */
54 	u32 __iomem *hw_dlist;
55 
56 	/* Clipped coordinates of the plane on the display. */
57 	int crtc_x, crtc_y, crtc_w, crtc_h;
58 	/* Clipped area being scanned from in the FB. */
59 	u32 src_x, src_y;
60 
61 	u32 src_w[2], src_h[2];
62 
63 	/* Scaling selection for the RGB/Y plane and the Cb/Cr planes. */
64 	enum vc4_scaling_mode x_scaling[2], y_scaling[2];
65 	bool is_unity;
66 	bool is_yuv;
67 
68 	/* Offset to start scanning out from the start of the plane's
69 	 * BO.
70 	 */
71 	u32 offsets[3];
72 
73 	/* Our allocation in LBM for temporary storage during scaling. */
74 	struct drm_mm_node lbm;
75 };
76 
77 static inline struct vc4_plane_state *
78 to_vc4_plane_state(struct drm_plane_state *state)
79 {
80 	return (struct vc4_plane_state *)state;
81 }
82 
83 static const struct hvs_format {
84 	u32 drm; /* DRM_FORMAT_* */
85 	u32 hvs; /* HVS_FORMAT_* */
86 	u32 pixel_order;
87 	bool has_alpha;
88 	bool flip_cbcr;
89 } hvs_formats[] = {
90 	{
91 		.drm = DRM_FORMAT_XRGB8888, .hvs = HVS_PIXEL_FORMAT_RGBA8888,
92 		.pixel_order = HVS_PIXEL_ORDER_ABGR, .has_alpha = false,
93 	},
94 	{
95 		.drm = DRM_FORMAT_ARGB8888, .hvs = HVS_PIXEL_FORMAT_RGBA8888,
96 		.pixel_order = HVS_PIXEL_ORDER_ABGR, .has_alpha = true,
97 	},
98 	{
99 		.drm = DRM_FORMAT_ABGR8888, .hvs = HVS_PIXEL_FORMAT_RGBA8888,
100 		.pixel_order = HVS_PIXEL_ORDER_ARGB, .has_alpha = true,
101 	},
102 	{
103 		.drm = DRM_FORMAT_XBGR8888, .hvs = HVS_PIXEL_FORMAT_RGBA8888,
104 		.pixel_order = HVS_PIXEL_ORDER_ARGB, .has_alpha = false,
105 	},
106 	{
107 		.drm = DRM_FORMAT_RGB565, .hvs = HVS_PIXEL_FORMAT_RGB565,
108 		.pixel_order = HVS_PIXEL_ORDER_XRGB, .has_alpha = false,
109 	},
110 	{
111 		.drm = DRM_FORMAT_BGR565, .hvs = HVS_PIXEL_FORMAT_RGB565,
112 		.pixel_order = HVS_PIXEL_ORDER_XBGR, .has_alpha = false,
113 	},
114 	{
115 		.drm = DRM_FORMAT_ARGB1555, .hvs = HVS_PIXEL_FORMAT_RGBA5551,
116 		.pixel_order = HVS_PIXEL_ORDER_ABGR, .has_alpha = true,
117 	},
118 	{
119 		.drm = DRM_FORMAT_XRGB1555, .hvs = HVS_PIXEL_FORMAT_RGBA5551,
120 		.pixel_order = HVS_PIXEL_ORDER_ABGR, .has_alpha = false,
121 	},
122 	{
123 		.drm = DRM_FORMAT_YUV422,
124 		.hvs = HVS_PIXEL_FORMAT_YCBCR_YUV422_3PLANE,
125 	},
126 	{
127 		.drm = DRM_FORMAT_YVU422,
128 		.hvs = HVS_PIXEL_FORMAT_YCBCR_YUV422_3PLANE,
129 		.flip_cbcr = true,
130 	},
131 	{
132 		.drm = DRM_FORMAT_YUV420,
133 		.hvs = HVS_PIXEL_FORMAT_YCBCR_YUV420_3PLANE,
134 	},
135 	{
136 		.drm = DRM_FORMAT_YVU420,
137 		.hvs = HVS_PIXEL_FORMAT_YCBCR_YUV420_3PLANE,
138 		.flip_cbcr = true,
139 	},
140 	{
141 		.drm = DRM_FORMAT_NV12,
142 		.hvs = HVS_PIXEL_FORMAT_YCBCR_YUV420_2PLANE,
143 	},
144 	{
145 		.drm = DRM_FORMAT_NV16,
146 		.hvs = HVS_PIXEL_FORMAT_YCBCR_YUV422_2PLANE,
147 	},
148 };
149 
150 static const struct hvs_format *vc4_get_hvs_format(u32 drm_format)
151 {
152 	unsigned i;
153 
154 	for (i = 0; i < ARRAY_SIZE(hvs_formats); i++) {
155 		if (hvs_formats[i].drm == drm_format)
156 			return &hvs_formats[i];
157 	}
158 
159 	return NULL;
160 }
161 
162 static enum vc4_scaling_mode vc4_get_scaling_mode(u32 src, u32 dst)
163 {
164 	if (dst > src)
165 		return VC4_SCALING_PPF;
166 	else if (dst < src)
167 		return VC4_SCALING_TPZ;
168 	else
169 		return VC4_SCALING_NONE;
170 }
171 
172 static bool plane_enabled(struct drm_plane_state *state)
173 {
174 	return state->fb && state->crtc;
175 }
176 
177 static struct drm_plane_state *vc4_plane_duplicate_state(struct drm_plane *plane)
178 {
179 	struct vc4_plane_state *vc4_state;
180 
181 	if (WARN_ON(!plane->state))
182 		return NULL;
183 
184 	vc4_state = kmemdup(plane->state, sizeof(*vc4_state), GFP_KERNEL);
185 	if (!vc4_state)
186 		return NULL;
187 
188 	memset(&vc4_state->lbm, 0, sizeof(vc4_state->lbm));
189 
190 	__drm_atomic_helper_plane_duplicate_state(plane, &vc4_state->base);
191 
192 	if (vc4_state->dlist) {
193 		vc4_state->dlist = kmemdup(vc4_state->dlist,
194 					   vc4_state->dlist_count * 4,
195 					   GFP_KERNEL);
196 		if (!vc4_state->dlist) {
197 			kfree(vc4_state);
198 			return NULL;
199 		}
200 		vc4_state->dlist_size = vc4_state->dlist_count;
201 	}
202 
203 	return &vc4_state->base;
204 }
205 
206 static void vc4_plane_destroy_state(struct drm_plane *plane,
207 				    struct drm_plane_state *state)
208 {
209 	struct vc4_dev *vc4 = to_vc4_dev(plane->dev);
210 	struct vc4_plane_state *vc4_state = to_vc4_plane_state(state);
211 
212 	if (vc4_state->lbm.allocated) {
213 		unsigned long irqflags;
214 
215 		spin_lock_irqsave(&vc4->hvs->mm_lock, irqflags);
216 		drm_mm_remove_node(&vc4_state->lbm);
217 		spin_unlock_irqrestore(&vc4->hvs->mm_lock, irqflags);
218 	}
219 
220 	kfree(vc4_state->dlist);
221 	__drm_atomic_helper_plane_destroy_state(&vc4_state->base);
222 	kfree(state);
223 }
224 
225 /* Called during init to allocate the plane's atomic state. */
226 static void vc4_plane_reset(struct drm_plane *plane)
227 {
228 	struct vc4_plane_state *vc4_state;
229 
230 	WARN_ON(plane->state);
231 
232 	vc4_state = kzalloc(sizeof(*vc4_state), GFP_KERNEL);
233 	if (!vc4_state)
234 		return;
235 
236 	plane->state = &vc4_state->base;
237 	vc4_state->base.plane = plane;
238 }
239 
240 static void vc4_dlist_write(struct vc4_plane_state *vc4_state, u32 val)
241 {
242 	if (vc4_state->dlist_count == vc4_state->dlist_size) {
243 		u32 new_size = max(4u, vc4_state->dlist_count * 2);
244 		u32 *new_dlist = kmalloc(new_size * 4, GFP_KERNEL);
245 
246 		if (!new_dlist)
247 			return;
248 		memcpy(new_dlist, vc4_state->dlist, vc4_state->dlist_count * 4);
249 
250 		kfree(vc4_state->dlist);
251 		vc4_state->dlist = new_dlist;
252 		vc4_state->dlist_size = new_size;
253 	}
254 
255 	vc4_state->dlist[vc4_state->dlist_count++] = val;
256 }
257 
258 /* Returns the scl0/scl1 field based on whether the dimensions need to
259  * be up/down/non-scaled.
260  *
261  * This is a replication of a table from the spec.
262  */
263 static u32 vc4_get_scl_field(struct drm_plane_state *state, int plane)
264 {
265 	struct vc4_plane_state *vc4_state = to_vc4_plane_state(state);
266 
267 	switch (vc4_state->x_scaling[plane] << 2 | vc4_state->y_scaling[plane]) {
268 	case VC4_SCALING_PPF << 2 | VC4_SCALING_PPF:
269 		return SCALER_CTL0_SCL_H_PPF_V_PPF;
270 	case VC4_SCALING_TPZ << 2 | VC4_SCALING_PPF:
271 		return SCALER_CTL0_SCL_H_TPZ_V_PPF;
272 	case VC4_SCALING_PPF << 2 | VC4_SCALING_TPZ:
273 		return SCALER_CTL0_SCL_H_PPF_V_TPZ;
274 	case VC4_SCALING_TPZ << 2 | VC4_SCALING_TPZ:
275 		return SCALER_CTL0_SCL_H_TPZ_V_TPZ;
276 	case VC4_SCALING_PPF << 2 | VC4_SCALING_NONE:
277 		return SCALER_CTL0_SCL_H_PPF_V_NONE;
278 	case VC4_SCALING_NONE << 2 | VC4_SCALING_PPF:
279 		return SCALER_CTL0_SCL_H_NONE_V_PPF;
280 	case VC4_SCALING_NONE << 2 | VC4_SCALING_TPZ:
281 		return SCALER_CTL0_SCL_H_NONE_V_TPZ;
282 	case VC4_SCALING_TPZ << 2 | VC4_SCALING_NONE:
283 		return SCALER_CTL0_SCL_H_TPZ_V_NONE;
284 	default:
285 	case VC4_SCALING_NONE << 2 | VC4_SCALING_NONE:
286 		/* The unity case is independently handled by
287 		 * SCALER_CTL0_UNITY.
288 		 */
289 		return 0;
290 	}
291 }
292 
293 static int vc4_plane_setup_clipping_and_scaling(struct drm_plane_state *state)
294 {
295 	struct drm_plane *plane = state->plane;
296 	struct vc4_plane_state *vc4_state = to_vc4_plane_state(state);
297 	struct drm_framebuffer *fb = state->fb;
298 	struct drm_gem_cma_object *bo = drm_fb_cma_get_gem_obj(fb, 0);
299 	u32 subpixel_src_mask = (1 << 16) - 1;
300 	u32 format = fb->format->format;
301 	int num_planes = fb->format->num_planes;
302 	u32 h_subsample = 1;
303 	u32 v_subsample = 1;
304 	int i;
305 
306 	for (i = 0; i < num_planes; i++)
307 		vc4_state->offsets[i] = bo->paddr + fb->offsets[i];
308 
309 	/* We don't support subpixel source positioning for scaling. */
310 	if ((state->src_x & subpixel_src_mask) ||
311 	    (state->src_y & subpixel_src_mask) ||
312 	    (state->src_w & subpixel_src_mask) ||
313 	    (state->src_h & subpixel_src_mask)) {
314 		return -EINVAL;
315 	}
316 
317 	vc4_state->src_x = state->src_x >> 16;
318 	vc4_state->src_y = state->src_y >> 16;
319 	vc4_state->src_w[0] = state->src_w >> 16;
320 	vc4_state->src_h[0] = state->src_h >> 16;
321 
322 	vc4_state->crtc_x = state->crtc_x;
323 	vc4_state->crtc_y = state->crtc_y;
324 	vc4_state->crtc_w = state->crtc_w;
325 	vc4_state->crtc_h = state->crtc_h;
326 
327 	vc4_state->x_scaling[0] = vc4_get_scaling_mode(vc4_state->src_w[0],
328 						       vc4_state->crtc_w);
329 	vc4_state->y_scaling[0] = vc4_get_scaling_mode(vc4_state->src_h[0],
330 						       vc4_state->crtc_h);
331 
332 	if (num_planes > 1) {
333 		vc4_state->is_yuv = true;
334 
335 		h_subsample = drm_format_horz_chroma_subsampling(format);
336 		v_subsample = drm_format_vert_chroma_subsampling(format);
337 		vc4_state->src_w[1] = vc4_state->src_w[0] / h_subsample;
338 		vc4_state->src_h[1] = vc4_state->src_h[0] / v_subsample;
339 
340 		vc4_state->x_scaling[1] =
341 			vc4_get_scaling_mode(vc4_state->src_w[1],
342 					     vc4_state->crtc_w);
343 		vc4_state->y_scaling[1] =
344 			vc4_get_scaling_mode(vc4_state->src_h[1],
345 					     vc4_state->crtc_h);
346 
347 		/* YUV conversion requires that scaling be enabled,
348 		 * even on a plane that's otherwise 1:1.  Choose TPZ
349 		 * for simplicity.
350 		 */
351 		if (vc4_state->x_scaling[0] == VC4_SCALING_NONE)
352 			vc4_state->x_scaling[0] = VC4_SCALING_TPZ;
353 		if (vc4_state->y_scaling[0] == VC4_SCALING_NONE)
354 			vc4_state->y_scaling[0] = VC4_SCALING_TPZ;
355 	}
356 
357 	vc4_state->is_unity = (vc4_state->x_scaling[0] == VC4_SCALING_NONE &&
358 			       vc4_state->y_scaling[0] == VC4_SCALING_NONE &&
359 			       vc4_state->x_scaling[1] == VC4_SCALING_NONE &&
360 			       vc4_state->y_scaling[1] == VC4_SCALING_NONE);
361 
362 	/* No configuring scaling on the cursor plane, since it gets
363 	   non-vblank-synced updates, and scaling requires requires
364 	   LBM changes which have to be vblank-synced.
365 	 */
366 	if (plane->type == DRM_PLANE_TYPE_CURSOR && !vc4_state->is_unity)
367 		return -EINVAL;
368 
369 	/* Clamp the on-screen start x/y to 0.  The hardware doesn't
370 	 * support negative y, and negative x wastes bandwidth.
371 	 */
372 	if (vc4_state->crtc_x < 0) {
373 		for (i = 0; i < num_planes; i++) {
374 			u32 cpp = fb->format->cpp[i];
375 			u32 subs = ((i == 0) ? 1 : h_subsample);
376 
377 			vc4_state->offsets[i] += (cpp *
378 						  (-vc4_state->crtc_x) / subs);
379 		}
380 		vc4_state->src_w[0] += vc4_state->crtc_x;
381 		vc4_state->src_w[1] += vc4_state->crtc_x / h_subsample;
382 		vc4_state->crtc_x = 0;
383 	}
384 
385 	if (vc4_state->crtc_y < 0) {
386 		for (i = 0; i < num_planes; i++) {
387 			u32 subs = ((i == 0) ? 1 : v_subsample);
388 
389 			vc4_state->offsets[i] += (fb->pitches[i] *
390 						  (-vc4_state->crtc_y) / subs);
391 		}
392 		vc4_state->src_h[0] += vc4_state->crtc_y;
393 		vc4_state->src_h[1] += vc4_state->crtc_y / v_subsample;
394 		vc4_state->crtc_y = 0;
395 	}
396 
397 	return 0;
398 }
399 
400 static void vc4_write_tpz(struct vc4_plane_state *vc4_state, u32 src, u32 dst)
401 {
402 	u32 scale, recip;
403 
404 	scale = (1 << 16) * src / dst;
405 
406 	/* The specs note that while the reciprocal would be defined
407 	 * as (1<<32)/scale, ~0 is close enough.
408 	 */
409 	recip = ~0 / scale;
410 
411 	vc4_dlist_write(vc4_state,
412 			VC4_SET_FIELD(scale, SCALER_TPZ0_SCALE) |
413 			VC4_SET_FIELD(0, SCALER_TPZ0_IPHASE));
414 	vc4_dlist_write(vc4_state,
415 			VC4_SET_FIELD(recip, SCALER_TPZ1_RECIP));
416 }
417 
418 static void vc4_write_ppf(struct vc4_plane_state *vc4_state, u32 src, u32 dst)
419 {
420 	u32 scale = (1 << 16) * src / dst;
421 
422 	vc4_dlist_write(vc4_state,
423 			SCALER_PPF_AGC |
424 			VC4_SET_FIELD(scale, SCALER_PPF_SCALE) |
425 			VC4_SET_FIELD(0, SCALER_PPF_IPHASE));
426 }
427 
428 static u32 vc4_lbm_size(struct drm_plane_state *state)
429 {
430 	struct vc4_plane_state *vc4_state = to_vc4_plane_state(state);
431 	/* This is the worst case number.  One of the two sizes will
432 	 * be used depending on the scaling configuration.
433 	 */
434 	u32 pix_per_line = max(vc4_state->src_w[0], (u32)vc4_state->crtc_w);
435 	u32 lbm;
436 
437 	if (!vc4_state->is_yuv) {
438 		if (vc4_state->is_unity)
439 			return 0;
440 		else if (vc4_state->y_scaling[0] == VC4_SCALING_TPZ)
441 			lbm = pix_per_line * 8;
442 		else {
443 			/* In special cases, this multiplier might be 12. */
444 			lbm = pix_per_line * 16;
445 		}
446 	} else {
447 		/* There are cases for this going down to a multiplier
448 		 * of 2, but according to the firmware source, the
449 		 * table in the docs is somewhat wrong.
450 		 */
451 		lbm = pix_per_line * 16;
452 	}
453 
454 	lbm = roundup(lbm, 32);
455 
456 	return lbm;
457 }
458 
459 static void vc4_write_scaling_parameters(struct drm_plane_state *state,
460 					 int channel)
461 {
462 	struct vc4_plane_state *vc4_state = to_vc4_plane_state(state);
463 
464 	/* Ch0 H-PPF Word 0: Scaling Parameters */
465 	if (vc4_state->x_scaling[channel] == VC4_SCALING_PPF) {
466 		vc4_write_ppf(vc4_state,
467 			      vc4_state->src_w[channel], vc4_state->crtc_w);
468 	}
469 
470 	/* Ch0 V-PPF Words 0-1: Scaling Parameters, Context */
471 	if (vc4_state->y_scaling[channel] == VC4_SCALING_PPF) {
472 		vc4_write_ppf(vc4_state,
473 			      vc4_state->src_h[channel], vc4_state->crtc_h);
474 		vc4_dlist_write(vc4_state, 0xc0c0c0c0);
475 	}
476 
477 	/* Ch0 H-TPZ Words 0-1: Scaling Parameters, Recip */
478 	if (vc4_state->x_scaling[channel] == VC4_SCALING_TPZ) {
479 		vc4_write_tpz(vc4_state,
480 			      vc4_state->src_w[channel], vc4_state->crtc_w);
481 	}
482 
483 	/* Ch0 V-TPZ Words 0-2: Scaling Parameters, Recip, Context */
484 	if (vc4_state->y_scaling[channel] == VC4_SCALING_TPZ) {
485 		vc4_write_tpz(vc4_state,
486 			      vc4_state->src_h[channel], vc4_state->crtc_h);
487 		vc4_dlist_write(vc4_state, 0xc0c0c0c0);
488 	}
489 }
490 
491 /* Writes out a full display list for an active plane to the plane's
492  * private dlist state.
493  */
494 static int vc4_plane_mode_set(struct drm_plane *plane,
495 			      struct drm_plane_state *state)
496 {
497 	struct vc4_dev *vc4 = to_vc4_dev(plane->dev);
498 	struct vc4_plane_state *vc4_state = to_vc4_plane_state(state);
499 	struct drm_framebuffer *fb = state->fb;
500 	u32 ctl0_offset = vc4_state->dlist_count;
501 	const struct hvs_format *format = vc4_get_hvs_format(fb->format->format);
502 	int num_planes = drm_format_num_planes(format->drm);
503 	u32 scl0, scl1, pitch0;
504 	u32 lbm_size, tiling;
505 	unsigned long irqflags;
506 	int ret, i;
507 
508 	ret = vc4_plane_setup_clipping_and_scaling(state);
509 	if (ret)
510 		return ret;
511 
512 	/* Allocate the LBM memory that the HVS will use for temporary
513 	 * storage due to our scaling/format conversion.
514 	 */
515 	lbm_size = vc4_lbm_size(state);
516 	if (lbm_size) {
517 		if (!vc4_state->lbm.allocated) {
518 			spin_lock_irqsave(&vc4->hvs->mm_lock, irqflags);
519 			ret = drm_mm_insert_node_generic(&vc4->hvs->lbm_mm,
520 							 &vc4_state->lbm,
521 							 lbm_size, 32, 0, 0);
522 			spin_unlock_irqrestore(&vc4->hvs->mm_lock, irqflags);
523 		} else {
524 			WARN_ON_ONCE(lbm_size != vc4_state->lbm.size);
525 		}
526 	}
527 
528 	if (ret)
529 		return ret;
530 
531 	/* SCL1 is used for Cb/Cr scaling of planar formats.  For RGB
532 	 * and 4:4:4, scl1 should be set to scl0 so both channels of
533 	 * the scaler do the same thing.  For YUV, the Y plane needs
534 	 * to be put in channel 1 and Cb/Cr in channel 0, so we swap
535 	 * the scl fields here.
536 	 */
537 	if (num_planes == 1) {
538 		scl0 = vc4_get_scl_field(state, 1);
539 		scl1 = scl0;
540 	} else {
541 		scl0 = vc4_get_scl_field(state, 1);
542 		scl1 = vc4_get_scl_field(state, 0);
543 	}
544 
545 	switch (fb->modifier) {
546 	case DRM_FORMAT_MOD_LINEAR:
547 		tiling = SCALER_CTL0_TILING_LINEAR;
548 		pitch0 = VC4_SET_FIELD(fb->pitches[0], SCALER_SRC_PITCH);
549 		break;
550 	case DRM_FORMAT_MOD_BROADCOM_VC4_T_TILED:
551 		tiling = SCALER_CTL0_TILING_256B_OR_T;
552 
553 		pitch0 = (VC4_SET_FIELD(0, SCALER_PITCH0_TILE_Y_OFFSET),
554 			  VC4_SET_FIELD(0, SCALER_PITCH0_TILE_WIDTH_L),
555 			  VC4_SET_FIELD((vc4_state->src_w[0] + 31) >> 5,
556 					SCALER_PITCH0_TILE_WIDTH_R));
557 		break;
558 	default:
559 		DRM_DEBUG_KMS("Unsupported FB tiling flag 0x%16llx",
560 			      (long long)fb->modifier);
561 		return -EINVAL;
562 	}
563 
564 	/* Control word */
565 	vc4_dlist_write(vc4_state,
566 			SCALER_CTL0_VALID |
567 			(format->pixel_order << SCALER_CTL0_ORDER_SHIFT) |
568 			(format->hvs << SCALER_CTL0_PIXEL_FORMAT_SHIFT) |
569 			VC4_SET_FIELD(tiling, SCALER_CTL0_TILING) |
570 			(vc4_state->is_unity ? SCALER_CTL0_UNITY : 0) |
571 			VC4_SET_FIELD(scl0, SCALER_CTL0_SCL0) |
572 			VC4_SET_FIELD(scl1, SCALER_CTL0_SCL1));
573 
574 	/* Position Word 0: Image Positions and Alpha Value */
575 	vc4_state->pos0_offset = vc4_state->dlist_count;
576 	vc4_dlist_write(vc4_state,
577 			VC4_SET_FIELD(0xff, SCALER_POS0_FIXED_ALPHA) |
578 			VC4_SET_FIELD(vc4_state->crtc_x, SCALER_POS0_START_X) |
579 			VC4_SET_FIELD(vc4_state->crtc_y, SCALER_POS0_START_Y));
580 
581 	/* Position Word 1: Scaled Image Dimensions. */
582 	if (!vc4_state->is_unity) {
583 		vc4_dlist_write(vc4_state,
584 				VC4_SET_FIELD(vc4_state->crtc_w,
585 					      SCALER_POS1_SCL_WIDTH) |
586 				VC4_SET_FIELD(vc4_state->crtc_h,
587 					      SCALER_POS1_SCL_HEIGHT));
588 	}
589 
590 	/* Position Word 2: Source Image Size, Alpha Mode */
591 	vc4_state->pos2_offset = vc4_state->dlist_count;
592 	vc4_dlist_write(vc4_state,
593 			VC4_SET_FIELD(format->has_alpha ?
594 				      SCALER_POS2_ALPHA_MODE_PIPELINE :
595 				      SCALER_POS2_ALPHA_MODE_FIXED,
596 				      SCALER_POS2_ALPHA_MODE) |
597 			VC4_SET_FIELD(vc4_state->src_w[0], SCALER_POS2_WIDTH) |
598 			VC4_SET_FIELD(vc4_state->src_h[0], SCALER_POS2_HEIGHT));
599 
600 	/* Position Word 3: Context.  Written by the HVS. */
601 	vc4_dlist_write(vc4_state, 0xc0c0c0c0);
602 
603 
604 	/* Pointer Word 0/1/2: RGB / Y / Cb / Cr Pointers
605 	 *
606 	 * The pointers may be any byte address.
607 	 */
608 	vc4_state->ptr0_offset = vc4_state->dlist_count;
609 	if (!format->flip_cbcr) {
610 		for (i = 0; i < num_planes; i++)
611 			vc4_dlist_write(vc4_state, vc4_state->offsets[i]);
612 	} else {
613 		WARN_ON_ONCE(num_planes != 3);
614 		vc4_dlist_write(vc4_state, vc4_state->offsets[0]);
615 		vc4_dlist_write(vc4_state, vc4_state->offsets[2]);
616 		vc4_dlist_write(vc4_state, vc4_state->offsets[1]);
617 	}
618 
619 	/* Pointer Context Word 0/1/2: Written by the HVS */
620 	for (i = 0; i < num_planes; i++)
621 		vc4_dlist_write(vc4_state, 0xc0c0c0c0);
622 
623 	/* Pitch word 0 */
624 	vc4_dlist_write(vc4_state, pitch0);
625 
626 	/* Pitch word 1/2 */
627 	for (i = 1; i < num_planes; i++) {
628 		vc4_dlist_write(vc4_state,
629 				VC4_SET_FIELD(fb->pitches[i], SCALER_SRC_PITCH));
630 	}
631 
632 	/* Colorspace conversion words */
633 	if (vc4_state->is_yuv) {
634 		vc4_dlist_write(vc4_state, SCALER_CSC0_ITR_R_601_5);
635 		vc4_dlist_write(vc4_state, SCALER_CSC1_ITR_R_601_5);
636 		vc4_dlist_write(vc4_state, SCALER_CSC2_ITR_R_601_5);
637 	}
638 
639 	if (!vc4_state->is_unity) {
640 		/* LBM Base Address. */
641 		if (vc4_state->y_scaling[0] != VC4_SCALING_NONE ||
642 		    vc4_state->y_scaling[1] != VC4_SCALING_NONE) {
643 			vc4_dlist_write(vc4_state, vc4_state->lbm.start);
644 		}
645 
646 		if (num_planes > 1) {
647 			/* Emit Cb/Cr as channel 0 and Y as channel
648 			 * 1. This matches how we set up scl0/scl1
649 			 * above.
650 			 */
651 			vc4_write_scaling_parameters(state, 1);
652 		}
653 		vc4_write_scaling_parameters(state, 0);
654 
655 		/* If any PPF setup was done, then all the kernel
656 		 * pointers get uploaded.
657 		 */
658 		if (vc4_state->x_scaling[0] == VC4_SCALING_PPF ||
659 		    vc4_state->y_scaling[0] == VC4_SCALING_PPF ||
660 		    vc4_state->x_scaling[1] == VC4_SCALING_PPF ||
661 		    vc4_state->y_scaling[1] == VC4_SCALING_PPF) {
662 			u32 kernel = VC4_SET_FIELD(vc4->hvs->mitchell_netravali_filter.start,
663 						   SCALER_PPF_KERNEL_OFFSET);
664 
665 			/* HPPF plane 0 */
666 			vc4_dlist_write(vc4_state, kernel);
667 			/* VPPF plane 0 */
668 			vc4_dlist_write(vc4_state, kernel);
669 			/* HPPF plane 1 */
670 			vc4_dlist_write(vc4_state, kernel);
671 			/* VPPF plane 1 */
672 			vc4_dlist_write(vc4_state, kernel);
673 		}
674 	}
675 
676 	vc4_state->dlist[ctl0_offset] |=
677 		VC4_SET_FIELD(vc4_state->dlist_count, SCALER_CTL0_SIZE);
678 
679 	return 0;
680 }
681 
682 /* If a modeset involves changing the setup of a plane, the atomic
683  * infrastructure will call this to validate a proposed plane setup.
684  * However, if a plane isn't getting updated, this (and the
685  * corresponding vc4_plane_atomic_update) won't get called.  Thus, we
686  * compute the dlist here and have all active plane dlists get updated
687  * in the CRTC's flush.
688  */
689 static int vc4_plane_atomic_check(struct drm_plane *plane,
690 				  struct drm_plane_state *state)
691 {
692 	struct vc4_plane_state *vc4_state = to_vc4_plane_state(state);
693 
694 	vc4_state->dlist_count = 0;
695 
696 	if (plane_enabled(state))
697 		return vc4_plane_mode_set(plane, state);
698 	else
699 		return 0;
700 }
701 
702 static void vc4_plane_atomic_update(struct drm_plane *plane,
703 				    struct drm_plane_state *old_state)
704 {
705 	/* No contents here.  Since we don't know where in the CRTC's
706 	 * dlist we should be stored, our dlist is uploaded to the
707 	 * hardware with vc4_plane_write_dlist() at CRTC atomic_flush
708 	 * time.
709 	 */
710 }
711 
712 u32 vc4_plane_write_dlist(struct drm_plane *plane, u32 __iomem *dlist)
713 {
714 	struct vc4_plane_state *vc4_state = to_vc4_plane_state(plane->state);
715 	int i;
716 
717 	vc4_state->hw_dlist = dlist;
718 
719 	/* Can't memcpy_toio() because it needs to be 32-bit writes. */
720 	for (i = 0; i < vc4_state->dlist_count; i++)
721 		writel(vc4_state->dlist[i], &dlist[i]);
722 
723 	return vc4_state->dlist_count;
724 }
725 
726 u32 vc4_plane_dlist_size(const struct drm_plane_state *state)
727 {
728 	const struct vc4_plane_state *vc4_state =
729 		container_of(state, typeof(*vc4_state), base);
730 
731 	return vc4_state->dlist_count;
732 }
733 
734 /* Updates the plane to immediately (well, once the FIFO needs
735  * refilling) scan out from at a new framebuffer.
736  */
737 void vc4_plane_async_set_fb(struct drm_plane *plane, struct drm_framebuffer *fb)
738 {
739 	struct vc4_plane_state *vc4_state = to_vc4_plane_state(plane->state);
740 	struct drm_gem_cma_object *bo = drm_fb_cma_get_gem_obj(fb, 0);
741 	uint32_t addr;
742 
743 	/* We're skipping the address adjustment for negative origin,
744 	 * because this is only called on the primary plane.
745 	 */
746 	WARN_ON_ONCE(plane->state->crtc_x < 0 || plane->state->crtc_y < 0);
747 	addr = bo->paddr + fb->offsets[0];
748 
749 	/* Write the new address into the hardware immediately.  The
750 	 * scanout will start from this address as soon as the FIFO
751 	 * needs to refill with pixels.
752 	 */
753 	writel(addr, &vc4_state->hw_dlist[vc4_state->ptr0_offset]);
754 
755 	/* Also update the CPU-side dlist copy, so that any later
756 	 * atomic updates that don't do a new modeset on our plane
757 	 * also use our updated address.
758 	 */
759 	vc4_state->dlist[vc4_state->ptr0_offset] = addr;
760 }
761 
762 static int vc4_prepare_fb(struct drm_plane *plane,
763 			  struct drm_plane_state *state)
764 {
765 	struct vc4_bo *bo;
766 	struct dma_fence *fence;
767 
768 	if ((plane->state->fb == state->fb) || !state->fb)
769 		return 0;
770 
771 	bo = to_vc4_bo(&drm_fb_cma_get_gem_obj(state->fb, 0)->base);
772 	fence = reservation_object_get_excl_rcu(bo->resv);
773 	drm_atomic_set_fence_for_plane(state, fence);
774 
775 	return 0;
776 }
777 
778 static const struct drm_plane_helper_funcs vc4_plane_helper_funcs = {
779 	.atomic_check = vc4_plane_atomic_check,
780 	.atomic_update = vc4_plane_atomic_update,
781 	.prepare_fb = vc4_prepare_fb,
782 };
783 
784 static void vc4_plane_destroy(struct drm_plane *plane)
785 {
786 	drm_plane_helper_disable(plane);
787 	drm_plane_cleanup(plane);
788 }
789 
790 /* Implements immediate (non-vblank-synced) updates of the cursor
791  * position, or falls back to the atomic helper otherwise.
792  */
793 static int
794 vc4_update_plane(struct drm_plane *plane,
795 		 struct drm_crtc *crtc,
796 		 struct drm_framebuffer *fb,
797 		 int crtc_x, int crtc_y,
798 		 unsigned int crtc_w, unsigned int crtc_h,
799 		 uint32_t src_x, uint32_t src_y,
800 		 uint32_t src_w, uint32_t src_h,
801 		 struct drm_modeset_acquire_ctx *ctx)
802 {
803 	struct drm_plane_state *plane_state;
804 	struct vc4_plane_state *vc4_state;
805 
806 	if (plane != crtc->cursor)
807 		goto out;
808 
809 	plane_state = plane->state;
810 	vc4_state = to_vc4_plane_state(plane_state);
811 
812 	if (!plane_state)
813 		goto out;
814 
815 	/* No configuring new scaling in the fast path. */
816 	if (crtc_w != plane_state->crtc_w ||
817 	    crtc_h != plane_state->crtc_h ||
818 	    src_w != plane_state->src_w ||
819 	    src_h != plane_state->src_h) {
820 		goto out;
821 	}
822 
823 	if (fb != plane_state->fb) {
824 		drm_atomic_set_fb_for_plane(plane->state, fb);
825 		vc4_plane_async_set_fb(plane, fb);
826 	}
827 
828 	/* Set the cursor's position on the screen.  This is the
829 	 * expected change from the drm_mode_cursor_universal()
830 	 * helper.
831 	 */
832 	plane_state->crtc_x = crtc_x;
833 	plane_state->crtc_y = crtc_y;
834 
835 	/* Allow changing the start position within the cursor BO, if
836 	 * that matters.
837 	 */
838 	plane_state->src_x = src_x;
839 	plane_state->src_y = src_y;
840 
841 	/* Update the display list based on the new crtc_x/y. */
842 	vc4_plane_atomic_check(plane, plane_state);
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 	return 0;
856 
857 out:
858 	return drm_atomic_helper_update_plane(plane, crtc, fb,
859 					      crtc_x, crtc_y,
860 					      crtc_w, crtc_h,
861 					      src_x, src_y,
862 					      src_w, src_h,
863 					      ctx);
864 }
865 
866 static const struct drm_plane_funcs vc4_plane_funcs = {
867 	.update_plane = vc4_update_plane,
868 	.disable_plane = drm_atomic_helper_disable_plane,
869 	.destroy = vc4_plane_destroy,
870 	.set_property = NULL,
871 	.reset = vc4_plane_reset,
872 	.atomic_duplicate_state = vc4_plane_duplicate_state,
873 	.atomic_destroy_state = vc4_plane_destroy_state,
874 };
875 
876 struct drm_plane *vc4_plane_init(struct drm_device *dev,
877 				 enum drm_plane_type type)
878 {
879 	struct drm_plane *plane = NULL;
880 	struct vc4_plane *vc4_plane;
881 	u32 formats[ARRAY_SIZE(hvs_formats)];
882 	u32 num_formats = 0;
883 	int ret = 0;
884 	unsigned i;
885 
886 	vc4_plane = devm_kzalloc(dev->dev, sizeof(*vc4_plane),
887 				 GFP_KERNEL);
888 	if (!vc4_plane)
889 		return ERR_PTR(-ENOMEM);
890 
891 	for (i = 0; i < ARRAY_SIZE(hvs_formats); i++) {
892 		/* Don't allow YUV in cursor planes, since that means
893 		 * tuning on the scaler, which we don't allow for the
894 		 * cursor.
895 		 */
896 		if (type != DRM_PLANE_TYPE_CURSOR ||
897 		    hvs_formats[i].hvs < HVS_PIXEL_FORMAT_YCBCR_YUV420_3PLANE) {
898 			formats[num_formats++] = hvs_formats[i].drm;
899 		}
900 	}
901 	plane = &vc4_plane->base;
902 	ret = drm_universal_plane_init(dev, plane, 0,
903 				       &vc4_plane_funcs,
904 				       formats, num_formats,
905 				       NULL, type, NULL);
906 
907 	drm_plane_helper_add(plane, &vc4_plane_helper_funcs);
908 
909 	return plane;
910 }
911