xref: /openbmc/linux/drivers/gpu/drm/vc4/vc4_plane.c (revision 98ddec80)
1 /*
2  * Copyright (C) 2015 Broadcom
3  *
4  * This program is free software; you can redistribute it and/or modify
5  * it under the terms of the GNU General Public License version 2 as
6  * published by the Free Software Foundation.
7  */
8 
9 /**
10  * DOC: VC4 plane module
11  *
12  * Each DRM plane is a layer of pixels being scanned out by the HVS.
13  *
14  * At atomic modeset check time, we compute the HVS display element
15  * state that would be necessary for displaying the plane (giving us a
16  * chance to figure out if a plane configuration is invalid), then at
17  * atomic flush time the CRTC will ask us to write our element state
18  * into the region of the HVS that it has allocated for us.
19  */
20 
21 #include <drm/drm_atomic.h>
22 #include <drm/drm_atomic_helper.h>
23 #include <drm/drm_fb_cma_helper.h>
24 #include <drm/drm_plane_helper.h>
25 
26 #include "uapi/drm/vc4_drm.h"
27 #include "vc4_drv.h"
28 #include "vc4_regs.h"
29 
30 static const struct hvs_format {
31 	u32 drm; /* DRM_FORMAT_* */
32 	u32 hvs; /* HVS_FORMAT_* */
33 	u32 pixel_order;
34 } hvs_formats[] = {
35 	{
36 		.drm = DRM_FORMAT_XRGB8888, .hvs = HVS_PIXEL_FORMAT_RGBA8888,
37 		.pixel_order = HVS_PIXEL_ORDER_ABGR,
38 	},
39 	{
40 		.drm = DRM_FORMAT_ARGB8888, .hvs = HVS_PIXEL_FORMAT_RGBA8888,
41 		.pixel_order = HVS_PIXEL_ORDER_ABGR,
42 	},
43 	{
44 		.drm = DRM_FORMAT_ABGR8888, .hvs = HVS_PIXEL_FORMAT_RGBA8888,
45 		.pixel_order = HVS_PIXEL_ORDER_ARGB,
46 	},
47 	{
48 		.drm = DRM_FORMAT_XBGR8888, .hvs = HVS_PIXEL_FORMAT_RGBA8888,
49 		.pixel_order = HVS_PIXEL_ORDER_ARGB,
50 	},
51 	{
52 		.drm = DRM_FORMAT_RGB565, .hvs = HVS_PIXEL_FORMAT_RGB565,
53 		.pixel_order = HVS_PIXEL_ORDER_XRGB,
54 	},
55 	{
56 		.drm = DRM_FORMAT_BGR565, .hvs = HVS_PIXEL_FORMAT_RGB565,
57 		.pixel_order = HVS_PIXEL_ORDER_XBGR,
58 	},
59 	{
60 		.drm = DRM_FORMAT_ARGB1555, .hvs = HVS_PIXEL_FORMAT_RGBA5551,
61 		.pixel_order = HVS_PIXEL_ORDER_ABGR,
62 	},
63 	{
64 		.drm = DRM_FORMAT_XRGB1555, .hvs = HVS_PIXEL_FORMAT_RGBA5551,
65 		.pixel_order = HVS_PIXEL_ORDER_ABGR,
66 	},
67 	{
68 		.drm = DRM_FORMAT_RGB888, .hvs = HVS_PIXEL_FORMAT_RGB888,
69 		.pixel_order = HVS_PIXEL_ORDER_XRGB,
70 	},
71 	{
72 		.drm = DRM_FORMAT_BGR888, .hvs = HVS_PIXEL_FORMAT_RGB888,
73 		.pixel_order = HVS_PIXEL_ORDER_XBGR,
74 	},
75 	{
76 		.drm = DRM_FORMAT_YUV422,
77 		.hvs = HVS_PIXEL_FORMAT_YCBCR_YUV422_3PLANE,
78 		.pixel_order = HVS_PIXEL_ORDER_XYCBCR,
79 	},
80 	{
81 		.drm = DRM_FORMAT_YVU422,
82 		.hvs = HVS_PIXEL_FORMAT_YCBCR_YUV422_3PLANE,
83 		.pixel_order = HVS_PIXEL_ORDER_XYCRCB,
84 	},
85 	{
86 		.drm = DRM_FORMAT_YUV420,
87 		.hvs = HVS_PIXEL_FORMAT_YCBCR_YUV420_3PLANE,
88 		.pixel_order = HVS_PIXEL_ORDER_XYCBCR,
89 	},
90 	{
91 		.drm = DRM_FORMAT_YVU420,
92 		.hvs = HVS_PIXEL_FORMAT_YCBCR_YUV420_3PLANE,
93 		.pixel_order = HVS_PIXEL_ORDER_XYCRCB,
94 	},
95 	{
96 		.drm = DRM_FORMAT_NV12,
97 		.hvs = HVS_PIXEL_FORMAT_YCBCR_YUV420_2PLANE,
98 		.pixel_order = HVS_PIXEL_ORDER_XYCBCR,
99 	},
100 	{
101 		.drm = DRM_FORMAT_NV21,
102 		.hvs = HVS_PIXEL_FORMAT_YCBCR_YUV420_2PLANE,
103 		.pixel_order = HVS_PIXEL_ORDER_XYCRCB,
104 	},
105 	{
106 		.drm = DRM_FORMAT_NV16,
107 		.hvs = HVS_PIXEL_FORMAT_YCBCR_YUV422_2PLANE,
108 		.pixel_order = HVS_PIXEL_ORDER_XYCBCR,
109 	},
110 	{
111 		.drm = DRM_FORMAT_NV61,
112 		.hvs = HVS_PIXEL_FORMAT_YCBCR_YUV422_2PLANE,
113 		.pixel_order = HVS_PIXEL_ORDER_XYCRCB,
114 	},
115 };
116 
117 static const struct hvs_format *vc4_get_hvs_format(u32 drm_format)
118 {
119 	unsigned i;
120 
121 	for (i = 0; i < ARRAY_SIZE(hvs_formats); i++) {
122 		if (hvs_formats[i].drm == drm_format)
123 			return &hvs_formats[i];
124 	}
125 
126 	return NULL;
127 }
128 
129 static enum vc4_scaling_mode vc4_get_scaling_mode(u32 src, u32 dst)
130 {
131 	if (dst > src)
132 		return VC4_SCALING_PPF;
133 	else if (dst < src)
134 		return VC4_SCALING_TPZ;
135 	else
136 		return VC4_SCALING_NONE;
137 }
138 
139 static bool plane_enabled(struct drm_plane_state *state)
140 {
141 	return state->fb && state->crtc;
142 }
143 
144 static struct drm_plane_state *vc4_plane_duplicate_state(struct drm_plane *plane)
145 {
146 	struct vc4_plane_state *vc4_state;
147 
148 	if (WARN_ON(!plane->state))
149 		return NULL;
150 
151 	vc4_state = kmemdup(plane->state, sizeof(*vc4_state), GFP_KERNEL);
152 	if (!vc4_state)
153 		return NULL;
154 
155 	memset(&vc4_state->lbm, 0, sizeof(vc4_state->lbm));
156 
157 	__drm_atomic_helper_plane_duplicate_state(plane, &vc4_state->base);
158 
159 	if (vc4_state->dlist) {
160 		vc4_state->dlist = kmemdup(vc4_state->dlist,
161 					   vc4_state->dlist_count * 4,
162 					   GFP_KERNEL);
163 		if (!vc4_state->dlist) {
164 			kfree(vc4_state);
165 			return NULL;
166 		}
167 		vc4_state->dlist_size = vc4_state->dlist_count;
168 	}
169 
170 	return &vc4_state->base;
171 }
172 
173 static void vc4_plane_destroy_state(struct drm_plane *plane,
174 				    struct drm_plane_state *state)
175 {
176 	struct vc4_dev *vc4 = to_vc4_dev(plane->dev);
177 	struct vc4_plane_state *vc4_state = to_vc4_plane_state(state);
178 
179 	if (vc4_state->lbm.allocated) {
180 		unsigned long irqflags;
181 
182 		spin_lock_irqsave(&vc4->hvs->mm_lock, irqflags);
183 		drm_mm_remove_node(&vc4_state->lbm);
184 		spin_unlock_irqrestore(&vc4->hvs->mm_lock, irqflags);
185 	}
186 
187 	kfree(vc4_state->dlist);
188 	__drm_atomic_helper_plane_destroy_state(&vc4_state->base);
189 	kfree(state);
190 }
191 
192 /* Called during init to allocate the plane's atomic state. */
193 static void vc4_plane_reset(struct drm_plane *plane)
194 {
195 	struct vc4_plane_state *vc4_state;
196 
197 	WARN_ON(plane->state);
198 
199 	vc4_state = kzalloc(sizeof(*vc4_state), GFP_KERNEL);
200 	if (!vc4_state)
201 		return;
202 
203 	plane->state = &vc4_state->base;
204 	plane->state->alpha = DRM_BLEND_ALPHA_OPAQUE;
205 	vc4_state->base.plane = plane;
206 }
207 
208 static void vc4_dlist_write(struct vc4_plane_state *vc4_state, u32 val)
209 {
210 	if (vc4_state->dlist_count == vc4_state->dlist_size) {
211 		u32 new_size = max(4u, vc4_state->dlist_count * 2);
212 		u32 *new_dlist = kmalloc_array(new_size, 4, GFP_KERNEL);
213 
214 		if (!new_dlist)
215 			return;
216 		memcpy(new_dlist, vc4_state->dlist, vc4_state->dlist_count * 4);
217 
218 		kfree(vc4_state->dlist);
219 		vc4_state->dlist = new_dlist;
220 		vc4_state->dlist_size = new_size;
221 	}
222 
223 	vc4_state->dlist[vc4_state->dlist_count++] = val;
224 }
225 
226 /* Returns the scl0/scl1 field based on whether the dimensions need to
227  * be up/down/non-scaled.
228  *
229  * This is a replication of a table from the spec.
230  */
231 static u32 vc4_get_scl_field(struct drm_plane_state *state, int plane)
232 {
233 	struct vc4_plane_state *vc4_state = to_vc4_plane_state(state);
234 
235 	switch (vc4_state->x_scaling[plane] << 2 | vc4_state->y_scaling[plane]) {
236 	case VC4_SCALING_PPF << 2 | VC4_SCALING_PPF:
237 		return SCALER_CTL0_SCL_H_PPF_V_PPF;
238 	case VC4_SCALING_TPZ << 2 | VC4_SCALING_PPF:
239 		return SCALER_CTL0_SCL_H_TPZ_V_PPF;
240 	case VC4_SCALING_PPF << 2 | VC4_SCALING_TPZ:
241 		return SCALER_CTL0_SCL_H_PPF_V_TPZ;
242 	case VC4_SCALING_TPZ << 2 | VC4_SCALING_TPZ:
243 		return SCALER_CTL0_SCL_H_TPZ_V_TPZ;
244 	case VC4_SCALING_PPF << 2 | VC4_SCALING_NONE:
245 		return SCALER_CTL0_SCL_H_PPF_V_NONE;
246 	case VC4_SCALING_NONE << 2 | VC4_SCALING_PPF:
247 		return SCALER_CTL0_SCL_H_NONE_V_PPF;
248 	case VC4_SCALING_NONE << 2 | VC4_SCALING_TPZ:
249 		return SCALER_CTL0_SCL_H_NONE_V_TPZ;
250 	case VC4_SCALING_TPZ << 2 | VC4_SCALING_NONE:
251 		return SCALER_CTL0_SCL_H_TPZ_V_NONE;
252 	default:
253 	case VC4_SCALING_NONE << 2 | VC4_SCALING_NONE:
254 		/* The unity case is independently handled by
255 		 * SCALER_CTL0_UNITY.
256 		 */
257 		return 0;
258 	}
259 }
260 
261 static int vc4_plane_setup_clipping_and_scaling(struct drm_plane_state *state)
262 {
263 	struct drm_plane *plane = state->plane;
264 	struct vc4_plane_state *vc4_state = to_vc4_plane_state(state);
265 	struct drm_framebuffer *fb = state->fb;
266 	struct drm_gem_cma_object *bo = drm_fb_cma_get_gem_obj(fb, 0);
267 	u32 subpixel_src_mask = (1 << 16) - 1;
268 	u32 format = fb->format->format;
269 	int num_planes = fb->format->num_planes;
270 	u32 h_subsample = 1;
271 	u32 v_subsample = 1;
272 	int i;
273 
274 	for (i = 0; i < num_planes; i++)
275 		vc4_state->offsets[i] = bo->paddr + fb->offsets[i];
276 
277 	/* We don't support subpixel source positioning for scaling. */
278 	if ((state->src_x & subpixel_src_mask) ||
279 	    (state->src_y & subpixel_src_mask) ||
280 	    (state->src_w & subpixel_src_mask) ||
281 	    (state->src_h & subpixel_src_mask)) {
282 		return -EINVAL;
283 	}
284 
285 	vc4_state->src_x = state->src_x >> 16;
286 	vc4_state->src_y = state->src_y >> 16;
287 	vc4_state->src_w[0] = state->src_w >> 16;
288 	vc4_state->src_h[0] = state->src_h >> 16;
289 
290 	vc4_state->crtc_x = state->crtc_x;
291 	vc4_state->crtc_y = state->crtc_y;
292 	vc4_state->crtc_w = state->crtc_w;
293 	vc4_state->crtc_h = state->crtc_h;
294 
295 	vc4_state->x_scaling[0] = vc4_get_scaling_mode(vc4_state->src_w[0],
296 						       vc4_state->crtc_w);
297 	vc4_state->y_scaling[0] = vc4_get_scaling_mode(vc4_state->src_h[0],
298 						       vc4_state->crtc_h);
299 
300 	if (num_planes > 1) {
301 		vc4_state->is_yuv = true;
302 
303 		h_subsample = drm_format_horz_chroma_subsampling(format);
304 		v_subsample = drm_format_vert_chroma_subsampling(format);
305 		vc4_state->src_w[1] = vc4_state->src_w[0] / h_subsample;
306 		vc4_state->src_h[1] = vc4_state->src_h[0] / v_subsample;
307 
308 		vc4_state->x_scaling[1] =
309 			vc4_get_scaling_mode(vc4_state->src_w[1],
310 					     vc4_state->crtc_w);
311 		vc4_state->y_scaling[1] =
312 			vc4_get_scaling_mode(vc4_state->src_h[1],
313 					     vc4_state->crtc_h);
314 
315 		/* YUV conversion requires that scaling be enabled,
316 		 * even on a plane that's otherwise 1:1.  Choose TPZ
317 		 * for simplicity.
318 		 */
319 		if (vc4_state->x_scaling[0] == VC4_SCALING_NONE)
320 			vc4_state->x_scaling[0] = VC4_SCALING_TPZ;
321 		if (vc4_state->y_scaling[0] == VC4_SCALING_NONE)
322 			vc4_state->y_scaling[0] = VC4_SCALING_TPZ;
323 	}
324 
325 	vc4_state->is_unity = (vc4_state->x_scaling[0] == VC4_SCALING_NONE &&
326 			       vc4_state->y_scaling[0] == VC4_SCALING_NONE &&
327 			       vc4_state->x_scaling[1] == VC4_SCALING_NONE &&
328 			       vc4_state->y_scaling[1] == VC4_SCALING_NONE);
329 
330 	/* No configuring scaling on the cursor plane, since it gets
331 	   non-vblank-synced updates, and scaling requires requires
332 	   LBM changes which have to be vblank-synced.
333 	 */
334 	if (plane->type == DRM_PLANE_TYPE_CURSOR && !vc4_state->is_unity)
335 		return -EINVAL;
336 
337 	/* Clamp the on-screen start x/y to 0.  The hardware doesn't
338 	 * support negative y, and negative x wastes bandwidth.
339 	 */
340 	if (vc4_state->crtc_x < 0) {
341 		for (i = 0; i < num_planes; i++) {
342 			u32 cpp = fb->format->cpp[i];
343 			u32 subs = ((i == 0) ? 1 : h_subsample);
344 
345 			vc4_state->offsets[i] += (cpp *
346 						  (-vc4_state->crtc_x) / subs);
347 		}
348 		vc4_state->src_w[0] += vc4_state->crtc_x;
349 		vc4_state->src_w[1] += vc4_state->crtc_x / h_subsample;
350 		vc4_state->crtc_x = 0;
351 	}
352 
353 	if (vc4_state->crtc_y < 0) {
354 		for (i = 0; i < num_planes; i++) {
355 			u32 subs = ((i == 0) ? 1 : v_subsample);
356 
357 			vc4_state->offsets[i] += (fb->pitches[i] *
358 						  (-vc4_state->crtc_y) / subs);
359 		}
360 		vc4_state->src_h[0] += vc4_state->crtc_y;
361 		vc4_state->src_h[1] += vc4_state->crtc_y / v_subsample;
362 		vc4_state->crtc_y = 0;
363 	}
364 
365 	return 0;
366 }
367 
368 static void vc4_write_tpz(struct vc4_plane_state *vc4_state, u32 src, u32 dst)
369 {
370 	u32 scale, recip;
371 
372 	scale = (1 << 16) * src / dst;
373 
374 	/* The specs note that while the reciprocal would be defined
375 	 * as (1<<32)/scale, ~0 is close enough.
376 	 */
377 	recip = ~0 / scale;
378 
379 	vc4_dlist_write(vc4_state,
380 			VC4_SET_FIELD(scale, SCALER_TPZ0_SCALE) |
381 			VC4_SET_FIELD(0, SCALER_TPZ0_IPHASE));
382 	vc4_dlist_write(vc4_state,
383 			VC4_SET_FIELD(recip, SCALER_TPZ1_RECIP));
384 }
385 
386 static void vc4_write_ppf(struct vc4_plane_state *vc4_state, u32 src, u32 dst)
387 {
388 	u32 scale = (1 << 16) * src / dst;
389 
390 	vc4_dlist_write(vc4_state,
391 			SCALER_PPF_AGC |
392 			VC4_SET_FIELD(scale, SCALER_PPF_SCALE) |
393 			VC4_SET_FIELD(0, SCALER_PPF_IPHASE));
394 }
395 
396 static u32 vc4_lbm_size(struct drm_plane_state *state)
397 {
398 	struct vc4_plane_state *vc4_state = to_vc4_plane_state(state);
399 	/* This is the worst case number.  One of the two sizes will
400 	 * be used depending on the scaling configuration.
401 	 */
402 	u32 pix_per_line = max(vc4_state->src_w[0], (u32)vc4_state->crtc_w);
403 	u32 lbm;
404 
405 	if (!vc4_state->is_yuv) {
406 		if (vc4_state->is_unity)
407 			return 0;
408 		else if (vc4_state->y_scaling[0] == VC4_SCALING_TPZ)
409 			lbm = pix_per_line * 8;
410 		else {
411 			/* In special cases, this multiplier might be 12. */
412 			lbm = pix_per_line * 16;
413 		}
414 	} else {
415 		/* There are cases for this going down to a multiplier
416 		 * of 2, but according to the firmware source, the
417 		 * table in the docs is somewhat wrong.
418 		 */
419 		lbm = pix_per_line * 16;
420 	}
421 
422 	lbm = roundup(lbm, 32);
423 
424 	return lbm;
425 }
426 
427 static void vc4_write_scaling_parameters(struct drm_plane_state *state,
428 					 int channel)
429 {
430 	struct vc4_plane_state *vc4_state = to_vc4_plane_state(state);
431 
432 	/* Ch0 H-PPF Word 0: Scaling Parameters */
433 	if (vc4_state->x_scaling[channel] == VC4_SCALING_PPF) {
434 		vc4_write_ppf(vc4_state,
435 			      vc4_state->src_w[channel], vc4_state->crtc_w);
436 	}
437 
438 	/* Ch0 V-PPF Words 0-1: Scaling Parameters, Context */
439 	if (vc4_state->y_scaling[channel] == VC4_SCALING_PPF) {
440 		vc4_write_ppf(vc4_state,
441 			      vc4_state->src_h[channel], vc4_state->crtc_h);
442 		vc4_dlist_write(vc4_state, 0xc0c0c0c0);
443 	}
444 
445 	/* Ch0 H-TPZ Words 0-1: Scaling Parameters, Recip */
446 	if (vc4_state->x_scaling[channel] == VC4_SCALING_TPZ) {
447 		vc4_write_tpz(vc4_state,
448 			      vc4_state->src_w[channel], vc4_state->crtc_w);
449 	}
450 
451 	/* Ch0 V-TPZ Words 0-2: Scaling Parameters, Recip, Context */
452 	if (vc4_state->y_scaling[channel] == VC4_SCALING_TPZ) {
453 		vc4_write_tpz(vc4_state,
454 			      vc4_state->src_h[channel], vc4_state->crtc_h);
455 		vc4_dlist_write(vc4_state, 0xc0c0c0c0);
456 	}
457 }
458 
459 /* Writes out a full display list for an active plane to the plane's
460  * private dlist state.
461  */
462 static int vc4_plane_mode_set(struct drm_plane *plane,
463 			      struct drm_plane_state *state)
464 {
465 	struct vc4_dev *vc4 = to_vc4_dev(plane->dev);
466 	struct vc4_plane_state *vc4_state = to_vc4_plane_state(state);
467 	struct drm_framebuffer *fb = state->fb;
468 	u32 ctl0_offset = vc4_state->dlist_count;
469 	const struct hvs_format *format = vc4_get_hvs_format(fb->format->format);
470 	int num_planes = drm_format_num_planes(format->drm);
471 	bool mix_plane_alpha;
472 	bool covers_screen;
473 	u32 scl0, scl1, pitch0;
474 	u32 lbm_size, tiling;
475 	unsigned long irqflags;
476 	int ret, i;
477 
478 	ret = vc4_plane_setup_clipping_and_scaling(state);
479 	if (ret)
480 		return ret;
481 
482 	/* Allocate the LBM memory that the HVS will use for temporary
483 	 * storage due to our scaling/format conversion.
484 	 */
485 	lbm_size = vc4_lbm_size(state);
486 	if (lbm_size) {
487 		if (!vc4_state->lbm.allocated) {
488 			spin_lock_irqsave(&vc4->hvs->mm_lock, irqflags);
489 			ret = drm_mm_insert_node_generic(&vc4->hvs->lbm_mm,
490 							 &vc4_state->lbm,
491 							 lbm_size, 32, 0, 0);
492 			spin_unlock_irqrestore(&vc4->hvs->mm_lock, irqflags);
493 		} else {
494 			WARN_ON_ONCE(lbm_size != vc4_state->lbm.size);
495 		}
496 	}
497 
498 	if (ret)
499 		return ret;
500 
501 	/* SCL1 is used for Cb/Cr scaling of planar formats.  For RGB
502 	 * and 4:4:4, scl1 should be set to scl0 so both channels of
503 	 * the scaler do the same thing.  For YUV, the Y plane needs
504 	 * to be put in channel 1 and Cb/Cr in channel 0, so we swap
505 	 * the scl fields here.
506 	 */
507 	if (num_planes == 1) {
508 		scl0 = vc4_get_scl_field(state, 0);
509 		scl1 = scl0;
510 	} else {
511 		scl0 = vc4_get_scl_field(state, 1);
512 		scl1 = vc4_get_scl_field(state, 0);
513 	}
514 
515 	switch (fb->modifier) {
516 	case DRM_FORMAT_MOD_LINEAR:
517 		tiling = SCALER_CTL0_TILING_LINEAR;
518 		pitch0 = VC4_SET_FIELD(fb->pitches[0], SCALER_SRC_PITCH);
519 		break;
520 
521 	case DRM_FORMAT_MOD_BROADCOM_VC4_T_TILED: {
522 		/* For T-tiled, the FB pitch is "how many bytes from
523 		 * one row to the next, such that pitch * tile_h ==
524 		 * tile_size * tiles_per_row."
525 		 */
526 		u32 tile_size_shift = 12; /* T tiles are 4kb */
527 		u32 tile_h_shift = 5; /* 16 and 32bpp are 32 pixels high */
528 		u32 tiles_w = fb->pitches[0] >> (tile_size_shift - tile_h_shift);
529 
530 		tiling = SCALER_CTL0_TILING_256B_OR_T;
531 
532 		pitch0 = (VC4_SET_FIELD(0, SCALER_PITCH0_TILE_Y_OFFSET) |
533 			  VC4_SET_FIELD(0, SCALER_PITCH0_TILE_WIDTH_L) |
534 			  VC4_SET_FIELD(tiles_w, SCALER_PITCH0_TILE_WIDTH_R));
535 		break;
536 	}
537 
538 	default:
539 		DRM_DEBUG_KMS("Unsupported FB tiling flag 0x%16llx",
540 			      (long long)fb->modifier);
541 		return -EINVAL;
542 	}
543 
544 	/* Control word */
545 	vc4_dlist_write(vc4_state,
546 			SCALER_CTL0_VALID |
547 			(format->pixel_order << SCALER_CTL0_ORDER_SHIFT) |
548 			(format->hvs << SCALER_CTL0_PIXEL_FORMAT_SHIFT) |
549 			VC4_SET_FIELD(tiling, SCALER_CTL0_TILING) |
550 			(vc4_state->is_unity ? SCALER_CTL0_UNITY : 0) |
551 			VC4_SET_FIELD(scl0, SCALER_CTL0_SCL0) |
552 			VC4_SET_FIELD(scl1, SCALER_CTL0_SCL1));
553 
554 	/* Position Word 0: Image Positions and Alpha Value */
555 	vc4_state->pos0_offset = vc4_state->dlist_count;
556 	vc4_dlist_write(vc4_state,
557 			VC4_SET_FIELD(state->alpha >> 8, SCALER_POS0_FIXED_ALPHA) |
558 			VC4_SET_FIELD(vc4_state->crtc_x, SCALER_POS0_START_X) |
559 			VC4_SET_FIELD(vc4_state->crtc_y, SCALER_POS0_START_Y));
560 
561 	/* Position Word 1: Scaled Image Dimensions. */
562 	if (!vc4_state->is_unity) {
563 		vc4_dlist_write(vc4_state,
564 				VC4_SET_FIELD(vc4_state->crtc_w,
565 					      SCALER_POS1_SCL_WIDTH) |
566 				VC4_SET_FIELD(vc4_state->crtc_h,
567 					      SCALER_POS1_SCL_HEIGHT));
568 	}
569 
570 	/* Don't waste cycles mixing with plane alpha if the set alpha
571 	 * is opaque or there is no per-pixel alpha information.
572 	 * In any case we use the alpha property value as the fixed alpha.
573 	 */
574 	mix_plane_alpha = state->alpha != DRM_BLEND_ALPHA_OPAQUE &&
575 			  fb->format->has_alpha;
576 
577 	/* Position Word 2: Source Image Size, Alpha */
578 	vc4_state->pos2_offset = vc4_state->dlist_count;
579 	vc4_dlist_write(vc4_state,
580 			VC4_SET_FIELD(fb->format->has_alpha ?
581 				      SCALER_POS2_ALPHA_MODE_PIPELINE :
582 				      SCALER_POS2_ALPHA_MODE_FIXED,
583 				      SCALER_POS2_ALPHA_MODE) |
584 			(mix_plane_alpha ? SCALER_POS2_ALPHA_MIX : 0) |
585 			(fb->format->has_alpha ? SCALER_POS2_ALPHA_PREMULT : 0) |
586 			VC4_SET_FIELD(vc4_state->src_w[0], SCALER_POS2_WIDTH) |
587 			VC4_SET_FIELD(vc4_state->src_h[0], SCALER_POS2_HEIGHT));
588 
589 	/* Position Word 3: Context.  Written by the HVS. */
590 	vc4_dlist_write(vc4_state, 0xc0c0c0c0);
591 
592 
593 	/* Pointer Word 0/1/2: RGB / Y / Cb / Cr Pointers
594 	 *
595 	 * The pointers may be any byte address.
596 	 */
597 	vc4_state->ptr0_offset = vc4_state->dlist_count;
598 	for (i = 0; i < num_planes; i++)
599 		vc4_dlist_write(vc4_state, vc4_state->offsets[i]);
600 
601 	/* Pointer Context Word 0/1/2: Written by the HVS */
602 	for (i = 0; i < num_planes; i++)
603 		vc4_dlist_write(vc4_state, 0xc0c0c0c0);
604 
605 	/* Pitch word 0 */
606 	vc4_dlist_write(vc4_state, pitch0);
607 
608 	/* Pitch word 1/2 */
609 	for (i = 1; i < num_planes; i++) {
610 		vc4_dlist_write(vc4_state,
611 				VC4_SET_FIELD(fb->pitches[i], SCALER_SRC_PITCH));
612 	}
613 
614 	/* Colorspace conversion words */
615 	if (vc4_state->is_yuv) {
616 		vc4_dlist_write(vc4_state, SCALER_CSC0_ITR_R_601_5);
617 		vc4_dlist_write(vc4_state, SCALER_CSC1_ITR_R_601_5);
618 		vc4_dlist_write(vc4_state, SCALER_CSC2_ITR_R_601_5);
619 	}
620 
621 	if (!vc4_state->is_unity) {
622 		/* LBM Base Address. */
623 		if (vc4_state->y_scaling[0] != VC4_SCALING_NONE ||
624 		    vc4_state->y_scaling[1] != VC4_SCALING_NONE) {
625 			vc4_dlist_write(vc4_state, vc4_state->lbm.start);
626 		}
627 
628 		if (num_planes > 1) {
629 			/* Emit Cb/Cr as channel 0 and Y as channel
630 			 * 1. This matches how we set up scl0/scl1
631 			 * above.
632 			 */
633 			vc4_write_scaling_parameters(state, 1);
634 		}
635 		vc4_write_scaling_parameters(state, 0);
636 
637 		/* If any PPF setup was done, then all the kernel
638 		 * pointers get uploaded.
639 		 */
640 		if (vc4_state->x_scaling[0] == VC4_SCALING_PPF ||
641 		    vc4_state->y_scaling[0] == VC4_SCALING_PPF ||
642 		    vc4_state->x_scaling[1] == VC4_SCALING_PPF ||
643 		    vc4_state->y_scaling[1] == VC4_SCALING_PPF) {
644 			u32 kernel = VC4_SET_FIELD(vc4->hvs->mitchell_netravali_filter.start,
645 						   SCALER_PPF_KERNEL_OFFSET);
646 
647 			/* HPPF plane 0 */
648 			vc4_dlist_write(vc4_state, kernel);
649 			/* VPPF plane 0 */
650 			vc4_dlist_write(vc4_state, kernel);
651 			/* HPPF plane 1 */
652 			vc4_dlist_write(vc4_state, kernel);
653 			/* VPPF plane 1 */
654 			vc4_dlist_write(vc4_state, kernel);
655 		}
656 	}
657 
658 	vc4_state->dlist[ctl0_offset] |=
659 		VC4_SET_FIELD(vc4_state->dlist_count, SCALER_CTL0_SIZE);
660 
661 	/* crtc_* are already clipped coordinates. */
662 	covers_screen = vc4_state->crtc_x == 0 && vc4_state->crtc_y == 0 &&
663 			vc4_state->crtc_w == state->crtc->mode.hdisplay &&
664 			vc4_state->crtc_h == state->crtc->mode.vdisplay;
665 	/* Background fill might be necessary when the plane has per-pixel
666 	 * alpha content or a non-opaque plane alpha and could blend from the
667 	 * background or does not cover the entire screen.
668 	 */
669 	vc4_state->needs_bg_fill = fb->format->has_alpha || !covers_screen ||
670 				   state->alpha != DRM_BLEND_ALPHA_OPAQUE;
671 
672 	return 0;
673 }
674 
675 /* If a modeset involves changing the setup of a plane, the atomic
676  * infrastructure will call this to validate a proposed plane setup.
677  * However, if a plane isn't getting updated, this (and the
678  * corresponding vc4_plane_atomic_update) won't get called.  Thus, we
679  * compute the dlist here and have all active plane dlists get updated
680  * in the CRTC's flush.
681  */
682 static int vc4_plane_atomic_check(struct drm_plane *plane,
683 				  struct drm_plane_state *state)
684 {
685 	struct vc4_plane_state *vc4_state = to_vc4_plane_state(state);
686 
687 	vc4_state->dlist_count = 0;
688 
689 	if (plane_enabled(state))
690 		return vc4_plane_mode_set(plane, state);
691 	else
692 		return 0;
693 }
694 
695 static void vc4_plane_atomic_update(struct drm_plane *plane,
696 				    struct drm_plane_state *old_state)
697 {
698 	/* No contents here.  Since we don't know where in the CRTC's
699 	 * dlist we should be stored, our dlist is uploaded to the
700 	 * hardware with vc4_plane_write_dlist() at CRTC atomic_flush
701 	 * time.
702 	 */
703 }
704 
705 u32 vc4_plane_write_dlist(struct drm_plane *plane, u32 __iomem *dlist)
706 {
707 	struct vc4_plane_state *vc4_state = to_vc4_plane_state(plane->state);
708 	int i;
709 
710 	vc4_state->hw_dlist = dlist;
711 
712 	/* Can't memcpy_toio() because it needs to be 32-bit writes. */
713 	for (i = 0; i < vc4_state->dlist_count; i++)
714 		writel(vc4_state->dlist[i], &dlist[i]);
715 
716 	return vc4_state->dlist_count;
717 }
718 
719 u32 vc4_plane_dlist_size(const struct drm_plane_state *state)
720 {
721 	const struct vc4_plane_state *vc4_state =
722 		container_of(state, typeof(*vc4_state), base);
723 
724 	return vc4_state->dlist_count;
725 }
726 
727 /* Updates the plane to immediately (well, once the FIFO needs
728  * refilling) scan out from at a new framebuffer.
729  */
730 void vc4_plane_async_set_fb(struct drm_plane *plane, struct drm_framebuffer *fb)
731 {
732 	struct vc4_plane_state *vc4_state = to_vc4_plane_state(plane->state);
733 	struct drm_gem_cma_object *bo = drm_fb_cma_get_gem_obj(fb, 0);
734 	uint32_t addr;
735 
736 	/* We're skipping the address adjustment for negative origin,
737 	 * because this is only called on the primary plane.
738 	 */
739 	WARN_ON_ONCE(plane->state->crtc_x < 0 || plane->state->crtc_y < 0);
740 	addr = bo->paddr + fb->offsets[0];
741 
742 	/* Write the new address into the hardware immediately.  The
743 	 * scanout will start from this address as soon as the FIFO
744 	 * needs to refill with pixels.
745 	 */
746 	writel(addr, &vc4_state->hw_dlist[vc4_state->ptr0_offset]);
747 
748 	/* Also update the CPU-side dlist copy, so that any later
749 	 * atomic updates that don't do a new modeset on our plane
750 	 * also use our updated address.
751 	 */
752 	vc4_state->dlist[vc4_state->ptr0_offset] = addr;
753 }
754 
755 static void vc4_plane_atomic_async_update(struct drm_plane *plane,
756 					  struct drm_plane_state *state)
757 {
758 	struct vc4_plane_state *vc4_state = to_vc4_plane_state(plane->state);
759 
760 	if (plane->state->fb != state->fb) {
761 		vc4_plane_async_set_fb(plane, state->fb);
762 		drm_atomic_set_fb_for_plane(plane->state, state->fb);
763 	}
764 
765 	/* Set the cursor's position on the screen.  This is the
766 	 * expected change from the drm_mode_cursor_universal()
767 	 * helper.
768 	 */
769 	plane->state->crtc_x = state->crtc_x;
770 	plane->state->crtc_y = state->crtc_y;
771 
772 	/* Allow changing the start position within the cursor BO, if
773 	 * that matters.
774 	 */
775 	plane->state->src_x = state->src_x;
776 	plane->state->src_y = state->src_y;
777 
778 	/* Update the display list based on the new crtc_x/y. */
779 	vc4_plane_atomic_check(plane, plane->state);
780 
781 	/* Note that we can't just call vc4_plane_write_dlist()
782 	 * because that would smash the context data that the HVS is
783 	 * currently using.
784 	 */
785 	writel(vc4_state->dlist[vc4_state->pos0_offset],
786 	       &vc4_state->hw_dlist[vc4_state->pos0_offset]);
787 	writel(vc4_state->dlist[vc4_state->pos2_offset],
788 	       &vc4_state->hw_dlist[vc4_state->pos2_offset]);
789 	writel(vc4_state->dlist[vc4_state->ptr0_offset],
790 	       &vc4_state->hw_dlist[vc4_state->ptr0_offset]);
791 }
792 
793 static int vc4_plane_atomic_async_check(struct drm_plane *plane,
794 					struct drm_plane_state *state)
795 {
796 	/* No configuring new scaling in the fast path. */
797 	if (plane->state->crtc_w != state->crtc_w ||
798 	    plane->state->crtc_h != state->crtc_h ||
799 	    plane->state->src_w != state->src_w ||
800 	    plane->state->src_h != state->src_h)
801 		return -EINVAL;
802 
803 	return 0;
804 }
805 
806 static int vc4_prepare_fb(struct drm_plane *plane,
807 			  struct drm_plane_state *state)
808 {
809 	struct vc4_bo *bo;
810 	struct dma_fence *fence;
811 	int ret;
812 
813 	if ((plane->state->fb == state->fb) || !state->fb)
814 		return 0;
815 
816 	bo = to_vc4_bo(&drm_fb_cma_get_gem_obj(state->fb, 0)->base);
817 
818 	ret = vc4_bo_inc_usecnt(bo);
819 	if (ret)
820 		return ret;
821 
822 	fence = reservation_object_get_excl_rcu(bo->resv);
823 	drm_atomic_set_fence_for_plane(state, fence);
824 
825 	return 0;
826 }
827 
828 static void vc4_cleanup_fb(struct drm_plane *plane,
829 			   struct drm_plane_state *state)
830 {
831 	struct vc4_bo *bo;
832 
833 	if (plane->state->fb == state->fb || !state->fb)
834 		return;
835 
836 	bo = to_vc4_bo(&drm_fb_cma_get_gem_obj(state->fb, 0)->base);
837 	vc4_bo_dec_usecnt(bo);
838 }
839 
840 static const struct drm_plane_helper_funcs vc4_plane_helper_funcs = {
841 	.atomic_check = vc4_plane_atomic_check,
842 	.atomic_update = vc4_plane_atomic_update,
843 	.prepare_fb = vc4_prepare_fb,
844 	.cleanup_fb = vc4_cleanup_fb,
845 	.atomic_async_check = vc4_plane_atomic_async_check,
846 	.atomic_async_update = vc4_plane_atomic_async_update,
847 };
848 
849 static void vc4_plane_destroy(struct drm_plane *plane)
850 {
851 	drm_plane_helper_disable(plane);
852 	drm_plane_cleanup(plane);
853 }
854 
855 static bool vc4_format_mod_supported(struct drm_plane *plane,
856 				     uint32_t format,
857 				     uint64_t modifier)
858 {
859 	/* Support T_TILING for RGB formats only. */
860 	switch (format) {
861 	case DRM_FORMAT_XRGB8888:
862 	case DRM_FORMAT_ARGB8888:
863 	case DRM_FORMAT_ABGR8888:
864 	case DRM_FORMAT_XBGR8888:
865 	case DRM_FORMAT_RGB565:
866 	case DRM_FORMAT_BGR565:
867 	case DRM_FORMAT_ARGB1555:
868 	case DRM_FORMAT_XRGB1555:
869 		return true;
870 	case DRM_FORMAT_YUV422:
871 	case DRM_FORMAT_YVU422:
872 	case DRM_FORMAT_YUV420:
873 	case DRM_FORMAT_YVU420:
874 	case DRM_FORMAT_NV12:
875 	case DRM_FORMAT_NV16:
876 	default:
877 		return (modifier == DRM_FORMAT_MOD_LINEAR);
878 	}
879 }
880 
881 static const struct drm_plane_funcs vc4_plane_funcs = {
882 	.update_plane = drm_atomic_helper_update_plane,
883 	.disable_plane = drm_atomic_helper_disable_plane,
884 	.destroy = vc4_plane_destroy,
885 	.set_property = NULL,
886 	.reset = vc4_plane_reset,
887 	.atomic_duplicate_state = vc4_plane_duplicate_state,
888 	.atomic_destroy_state = vc4_plane_destroy_state,
889 	.format_mod_supported = vc4_format_mod_supported,
890 };
891 
892 struct drm_plane *vc4_plane_init(struct drm_device *dev,
893 				 enum drm_plane_type type)
894 {
895 	struct drm_plane *plane = NULL;
896 	struct vc4_plane *vc4_plane;
897 	u32 formats[ARRAY_SIZE(hvs_formats)];
898 	u32 num_formats = 0;
899 	int ret = 0;
900 	unsigned i;
901 	static const uint64_t modifiers[] = {
902 		DRM_FORMAT_MOD_BROADCOM_VC4_T_TILED,
903 		DRM_FORMAT_MOD_LINEAR,
904 		DRM_FORMAT_MOD_INVALID
905 	};
906 
907 	vc4_plane = devm_kzalloc(dev->dev, sizeof(*vc4_plane),
908 				 GFP_KERNEL);
909 	if (!vc4_plane)
910 		return ERR_PTR(-ENOMEM);
911 
912 	for (i = 0; i < ARRAY_SIZE(hvs_formats); i++) {
913 		/* Don't allow YUV in cursor planes, since that means
914 		 * tuning on the scaler, which we don't allow for the
915 		 * cursor.
916 		 */
917 		if (type != DRM_PLANE_TYPE_CURSOR ||
918 		    hvs_formats[i].hvs < HVS_PIXEL_FORMAT_YCBCR_YUV420_3PLANE) {
919 			formats[num_formats++] = hvs_formats[i].drm;
920 		}
921 	}
922 	plane = &vc4_plane->base;
923 	ret = drm_universal_plane_init(dev, plane, 0,
924 				       &vc4_plane_funcs,
925 				       formats, num_formats,
926 				       modifiers, type, NULL);
927 
928 	drm_plane_helper_add(plane, &vc4_plane_helper_funcs);
929 
930 	drm_plane_create_alpha_property(plane);
931 
932 	return plane;
933 }
934