• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
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 *
to_vc4_plane_state(struct drm_plane_state * 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 
vc4_get_hvs_format(u32 drm_format)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 
vc4_get_scaling_mode(u32 src,u32 dst)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 
plane_enabled(struct drm_plane_state * state)172 static bool plane_enabled(struct drm_plane_state *state)
173 {
174 	return state->fb && state->crtc;
175 }
176 
vc4_plane_duplicate_state(struct drm_plane * plane)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 
vc4_plane_destroy_state(struct drm_plane * plane,struct drm_plane_state * state)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. */
vc4_plane_reset(struct drm_plane * plane)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 
vc4_dlist_write(struct vc4_plane_state * vc4_state,u32 val)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  */
vc4_get_scl_field(struct drm_plane_state * state,int plane)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 
vc4_plane_setup_clipping_and_scaling(struct drm_plane_state * state)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 	vc4_state->is_unity = (vc4_state->x_scaling[0] == VC4_SCALING_NONE &&
333 			       vc4_state->y_scaling[0] == VC4_SCALING_NONE);
334 
335 	if (num_planes > 1) {
336 		vc4_state->is_yuv = true;
337 
338 		h_subsample = drm_format_horz_chroma_subsampling(format);
339 		v_subsample = drm_format_vert_chroma_subsampling(format);
340 		vc4_state->src_w[1] = vc4_state->src_w[0] / h_subsample;
341 		vc4_state->src_h[1] = vc4_state->src_h[0] / v_subsample;
342 
343 		vc4_state->x_scaling[1] =
344 			vc4_get_scaling_mode(vc4_state->src_w[1],
345 					     vc4_state->crtc_w);
346 		vc4_state->y_scaling[1] =
347 			vc4_get_scaling_mode(vc4_state->src_h[1],
348 					     vc4_state->crtc_h);
349 
350 		/* YUV conversion requires that horizontal scaling be enabled
351 		 * on the UV plane even if vc4_get_scaling_mode() returned
352 		 * VC4_SCALING_NONE (which can happen when the down-scaling
353 		 * ratio is 0.5). Let's force it to VC4_SCALING_PPF in this
354 		 * case.
355 		 */
356 		if (vc4_state->x_scaling[1] == VC4_SCALING_NONE)
357 			vc4_state->x_scaling[1] = VC4_SCALING_PPF;
358 	} else {
359 		vc4_state->is_yuv = false;
360 		vc4_state->x_scaling[1] = VC4_SCALING_NONE;
361 		vc4_state->y_scaling[1] = VC4_SCALING_NONE;
362 	}
363 
364 	/* No configuring scaling on the cursor plane, since it gets
365 	   non-vblank-synced updates, and scaling requires requires
366 	   LBM changes which have to be vblank-synced.
367 	 */
368 	if (plane->type == DRM_PLANE_TYPE_CURSOR && !vc4_state->is_unity)
369 		return -EINVAL;
370 
371 	/* Clamp the on-screen start x/y to 0.  The hardware doesn't
372 	 * support negative y, and negative x wastes bandwidth.
373 	 */
374 	if (vc4_state->crtc_x < 0) {
375 		for (i = 0; i < num_planes; i++) {
376 			u32 cpp = fb->format->cpp[i];
377 			u32 subs = ((i == 0) ? 1 : h_subsample);
378 
379 			vc4_state->offsets[i] += (cpp *
380 						  (-vc4_state->crtc_x) / subs);
381 		}
382 		vc4_state->src_w[0] += vc4_state->crtc_x;
383 		vc4_state->src_w[1] += vc4_state->crtc_x / h_subsample;
384 		vc4_state->crtc_x = 0;
385 	}
386 
387 	if (vc4_state->crtc_y < 0) {
388 		for (i = 0; i < num_planes; i++) {
389 			u32 subs = ((i == 0) ? 1 : v_subsample);
390 
391 			vc4_state->offsets[i] += (fb->pitches[i] *
392 						  (-vc4_state->crtc_y) / subs);
393 		}
394 		vc4_state->src_h[0] += vc4_state->crtc_y;
395 		vc4_state->src_h[1] += vc4_state->crtc_y / v_subsample;
396 		vc4_state->crtc_y = 0;
397 	}
398 
399 	return 0;
400 }
401 
vc4_write_tpz(struct vc4_plane_state * vc4_state,u32 src,u32 dst)402 static void vc4_write_tpz(struct vc4_plane_state *vc4_state, u32 src, u32 dst)
403 {
404 	u32 scale, recip;
405 
406 	scale = (1 << 16) * src / dst;
407 
408 	/* The specs note that while the reciprocal would be defined
409 	 * as (1<<32)/scale, ~0 is close enough.
410 	 */
411 	recip = ~0 / scale;
412 
413 	vc4_dlist_write(vc4_state,
414 			VC4_SET_FIELD(scale, SCALER_TPZ0_SCALE) |
415 			VC4_SET_FIELD(0, SCALER_TPZ0_IPHASE));
416 	vc4_dlist_write(vc4_state,
417 			VC4_SET_FIELD(recip, SCALER_TPZ1_RECIP));
418 }
419 
vc4_write_ppf(struct vc4_plane_state * vc4_state,u32 src,u32 dst)420 static void vc4_write_ppf(struct vc4_plane_state *vc4_state, u32 src, u32 dst)
421 {
422 	u32 scale = (1 << 16) * src / dst;
423 
424 	vc4_dlist_write(vc4_state,
425 			SCALER_PPF_AGC |
426 			VC4_SET_FIELD(scale, SCALER_PPF_SCALE) |
427 			VC4_SET_FIELD(0, SCALER_PPF_IPHASE));
428 }
429 
vc4_lbm_size(struct drm_plane_state * state)430 static u32 vc4_lbm_size(struct drm_plane_state *state)
431 {
432 	struct vc4_plane_state *vc4_state = to_vc4_plane_state(state);
433 	/* This is the worst case number.  One of the two sizes will
434 	 * be used depending on the scaling configuration.
435 	 */
436 	u32 pix_per_line = max(vc4_state->src_w[0], (u32)vc4_state->crtc_w);
437 	u32 lbm;
438 
439 	if (!vc4_state->is_yuv) {
440 		if (vc4_state->is_unity)
441 			return 0;
442 		else if (vc4_state->y_scaling[0] == VC4_SCALING_TPZ)
443 			lbm = pix_per_line * 8;
444 		else {
445 			/* In special cases, this multiplier might be 12. */
446 			lbm = pix_per_line * 16;
447 		}
448 	} else {
449 		/* There are cases for this going down to a multiplier
450 		 * of 2, but according to the firmware source, the
451 		 * table in the docs is somewhat wrong.
452 		 */
453 		lbm = pix_per_line * 16;
454 	}
455 
456 	lbm = roundup(lbm, 32);
457 
458 	return lbm;
459 }
460 
vc4_write_scaling_parameters(struct drm_plane_state * state,int channel)461 static void vc4_write_scaling_parameters(struct drm_plane_state *state,
462 					 int channel)
463 {
464 	struct vc4_plane_state *vc4_state = to_vc4_plane_state(state);
465 
466 	/* Ch0 H-PPF Word 0: Scaling Parameters */
467 	if (vc4_state->x_scaling[channel] == VC4_SCALING_PPF) {
468 		vc4_write_ppf(vc4_state,
469 			      vc4_state->src_w[channel], vc4_state->crtc_w);
470 	}
471 
472 	/* Ch0 V-PPF Words 0-1: Scaling Parameters, Context */
473 	if (vc4_state->y_scaling[channel] == VC4_SCALING_PPF) {
474 		vc4_write_ppf(vc4_state,
475 			      vc4_state->src_h[channel], vc4_state->crtc_h);
476 		vc4_dlist_write(vc4_state, 0xc0c0c0c0);
477 	}
478 
479 	/* Ch0 H-TPZ Words 0-1: Scaling Parameters, Recip */
480 	if (vc4_state->x_scaling[channel] == VC4_SCALING_TPZ) {
481 		vc4_write_tpz(vc4_state,
482 			      vc4_state->src_w[channel], vc4_state->crtc_w);
483 	}
484 
485 	/* Ch0 V-TPZ Words 0-2: Scaling Parameters, Recip, Context */
486 	if (vc4_state->y_scaling[channel] == VC4_SCALING_TPZ) {
487 		vc4_write_tpz(vc4_state,
488 			      vc4_state->src_h[channel], vc4_state->crtc_h);
489 		vc4_dlist_write(vc4_state, 0xc0c0c0c0);
490 	}
491 }
492 
493 /* Writes out a full display list for an active plane to the plane's
494  * private dlist state.
495  */
vc4_plane_mode_set(struct drm_plane * plane,struct drm_plane_state * state)496 static int vc4_plane_mode_set(struct drm_plane *plane,
497 			      struct drm_plane_state *state)
498 {
499 	struct vc4_dev *vc4 = to_vc4_dev(plane->dev);
500 	struct vc4_plane_state *vc4_state = to_vc4_plane_state(state);
501 	struct drm_framebuffer *fb = state->fb;
502 	u32 ctl0_offset = vc4_state->dlist_count;
503 	const struct hvs_format *format = vc4_get_hvs_format(fb->format->format);
504 	int num_planes = drm_format_num_planes(format->drm);
505 	u32 scl0, scl1, pitch0;
506 	u32 lbm_size, tiling;
507 	unsigned long irqflags;
508 	int ret, i;
509 
510 	ret = vc4_plane_setup_clipping_and_scaling(state);
511 	if (ret)
512 		return ret;
513 
514 	/* Allocate the LBM memory that the HVS will use for temporary
515 	 * storage due to our scaling/format conversion.
516 	 */
517 	lbm_size = vc4_lbm_size(state);
518 	if (lbm_size) {
519 		if (!vc4_state->lbm.allocated) {
520 			spin_lock_irqsave(&vc4->hvs->mm_lock, irqflags);
521 			ret = drm_mm_insert_node_generic(&vc4->hvs->lbm_mm,
522 							 &vc4_state->lbm,
523 							 lbm_size, 32, 0, 0);
524 			spin_unlock_irqrestore(&vc4->hvs->mm_lock, irqflags);
525 		} else {
526 			WARN_ON_ONCE(lbm_size != vc4_state->lbm.size);
527 		}
528 	}
529 
530 	if (ret)
531 		return ret;
532 
533 	/* SCL1 is used for Cb/Cr scaling of planar formats.  For RGB
534 	 * and 4:4:4, scl1 should be set to scl0 so both channels of
535 	 * the scaler do the same thing.  For YUV, the Y plane needs
536 	 * to be put in channel 1 and Cb/Cr in channel 0, so we swap
537 	 * the scl fields here.
538 	 */
539 	if (num_planes == 1) {
540 		scl0 = vc4_get_scl_field(state, 0);
541 		scl1 = scl0;
542 	} else {
543 		scl0 = vc4_get_scl_field(state, 1);
544 		scl1 = vc4_get_scl_field(state, 0);
545 	}
546 
547 	switch (fb->modifier) {
548 	case DRM_FORMAT_MOD_LINEAR:
549 		tiling = SCALER_CTL0_TILING_LINEAR;
550 		pitch0 = VC4_SET_FIELD(fb->pitches[0], SCALER_SRC_PITCH);
551 		break;
552 	case DRM_FORMAT_MOD_BROADCOM_VC4_T_TILED:
553 		tiling = SCALER_CTL0_TILING_256B_OR_T;
554 
555 		pitch0 = (VC4_SET_FIELD(0, SCALER_PITCH0_TILE_Y_OFFSET),
556 			  VC4_SET_FIELD(0, SCALER_PITCH0_TILE_WIDTH_L),
557 			  VC4_SET_FIELD((vc4_state->src_w[0] + 31) >> 5,
558 					SCALER_PITCH0_TILE_WIDTH_R));
559 		break;
560 	default:
561 		DRM_DEBUG_KMS("Unsupported FB tiling flag 0x%16llx",
562 			      (long long)fb->modifier);
563 		return -EINVAL;
564 	}
565 
566 	/* Control word */
567 	vc4_dlist_write(vc4_state,
568 			SCALER_CTL0_VALID |
569 			(format->pixel_order << SCALER_CTL0_ORDER_SHIFT) |
570 			(format->hvs << SCALER_CTL0_PIXEL_FORMAT_SHIFT) |
571 			VC4_SET_FIELD(tiling, SCALER_CTL0_TILING) |
572 			(vc4_state->is_unity ? SCALER_CTL0_UNITY : 0) |
573 			VC4_SET_FIELD(scl0, SCALER_CTL0_SCL0) |
574 			VC4_SET_FIELD(scl1, SCALER_CTL0_SCL1));
575 
576 	/* Position Word 0: Image Positions and Alpha Value */
577 	vc4_state->pos0_offset = vc4_state->dlist_count;
578 	vc4_dlist_write(vc4_state,
579 			VC4_SET_FIELD(0xff, SCALER_POS0_FIXED_ALPHA) |
580 			VC4_SET_FIELD(vc4_state->crtc_x, SCALER_POS0_START_X) |
581 			VC4_SET_FIELD(vc4_state->crtc_y, SCALER_POS0_START_Y));
582 
583 	/* Position Word 1: Scaled Image Dimensions. */
584 	if (!vc4_state->is_unity) {
585 		vc4_dlist_write(vc4_state,
586 				VC4_SET_FIELD(vc4_state->crtc_w,
587 					      SCALER_POS1_SCL_WIDTH) |
588 				VC4_SET_FIELD(vc4_state->crtc_h,
589 					      SCALER_POS1_SCL_HEIGHT));
590 	}
591 
592 	/* Position Word 2: Source Image Size, Alpha Mode */
593 	vc4_state->pos2_offset = vc4_state->dlist_count;
594 	vc4_dlist_write(vc4_state,
595 			VC4_SET_FIELD(format->has_alpha ?
596 				      SCALER_POS2_ALPHA_MODE_PIPELINE :
597 				      SCALER_POS2_ALPHA_MODE_FIXED,
598 				      SCALER_POS2_ALPHA_MODE) |
599 			VC4_SET_FIELD(vc4_state->src_w[0], SCALER_POS2_WIDTH) |
600 			VC4_SET_FIELD(vc4_state->src_h[0], SCALER_POS2_HEIGHT));
601 
602 	/* Position Word 3: Context.  Written by the HVS. */
603 	vc4_dlist_write(vc4_state, 0xc0c0c0c0);
604 
605 
606 	/* Pointer Word 0/1/2: RGB / Y / Cb / Cr Pointers
607 	 *
608 	 * The pointers may be any byte address.
609 	 */
610 	vc4_state->ptr0_offset = vc4_state->dlist_count;
611 	if (!format->flip_cbcr) {
612 		for (i = 0; i < num_planes; i++)
613 			vc4_dlist_write(vc4_state, vc4_state->offsets[i]);
614 	} else {
615 		WARN_ON_ONCE(num_planes != 3);
616 		vc4_dlist_write(vc4_state, vc4_state->offsets[0]);
617 		vc4_dlist_write(vc4_state, vc4_state->offsets[2]);
618 		vc4_dlist_write(vc4_state, vc4_state->offsets[1]);
619 	}
620 
621 	/* Pointer Context Word 0/1/2: Written by the HVS */
622 	for (i = 0; i < num_planes; i++)
623 		vc4_dlist_write(vc4_state, 0xc0c0c0c0);
624 
625 	/* Pitch word 0 */
626 	vc4_dlist_write(vc4_state, pitch0);
627 
628 	/* Pitch word 1/2 */
629 	for (i = 1; i < num_planes; i++) {
630 		vc4_dlist_write(vc4_state,
631 				VC4_SET_FIELD(fb->pitches[i], SCALER_SRC_PITCH));
632 	}
633 
634 	/* Colorspace conversion words */
635 	if (vc4_state->is_yuv) {
636 		vc4_dlist_write(vc4_state, SCALER_CSC0_ITR_R_601_5);
637 		vc4_dlist_write(vc4_state, SCALER_CSC1_ITR_R_601_5);
638 		vc4_dlist_write(vc4_state, SCALER_CSC2_ITR_R_601_5);
639 	}
640 
641 	if (vc4_state->x_scaling[0] != VC4_SCALING_NONE ||
642 	    vc4_state->x_scaling[1] != VC4_SCALING_NONE ||
643 	    vc4_state->y_scaling[0] != VC4_SCALING_NONE ||
644 	    vc4_state->y_scaling[1] != VC4_SCALING_NONE) {
645 		/* LBM Base Address. */
646 		if (vc4_state->y_scaling[0] != VC4_SCALING_NONE ||
647 		    vc4_state->y_scaling[1] != VC4_SCALING_NONE) {
648 			vc4_dlist_write(vc4_state, vc4_state->lbm.start);
649 		}
650 
651 		if (num_planes > 1) {
652 			/* Emit Cb/Cr as channel 0 and Y as channel
653 			 * 1. This matches how we set up scl0/scl1
654 			 * above.
655 			 */
656 			vc4_write_scaling_parameters(state, 1);
657 		}
658 		vc4_write_scaling_parameters(state, 0);
659 
660 		/* If any PPF setup was done, then all the kernel
661 		 * pointers get uploaded.
662 		 */
663 		if (vc4_state->x_scaling[0] == VC4_SCALING_PPF ||
664 		    vc4_state->y_scaling[0] == VC4_SCALING_PPF ||
665 		    vc4_state->x_scaling[1] == VC4_SCALING_PPF ||
666 		    vc4_state->y_scaling[1] == VC4_SCALING_PPF) {
667 			u32 kernel = VC4_SET_FIELD(vc4->hvs->mitchell_netravali_filter.start,
668 						   SCALER_PPF_KERNEL_OFFSET);
669 
670 			/* HPPF plane 0 */
671 			vc4_dlist_write(vc4_state, kernel);
672 			/* VPPF plane 0 */
673 			vc4_dlist_write(vc4_state, kernel);
674 			/* HPPF plane 1 */
675 			vc4_dlist_write(vc4_state, kernel);
676 			/* VPPF plane 1 */
677 			vc4_dlist_write(vc4_state, kernel);
678 		}
679 	}
680 
681 	vc4_state->dlist[ctl0_offset] |=
682 		VC4_SET_FIELD(vc4_state->dlist_count, SCALER_CTL0_SIZE);
683 
684 	return 0;
685 }
686 
687 /* If a modeset involves changing the setup of a plane, the atomic
688  * infrastructure will call this to validate a proposed plane setup.
689  * However, if a plane isn't getting updated, this (and the
690  * corresponding vc4_plane_atomic_update) won't get called.  Thus, we
691  * compute the dlist here and have all active plane dlists get updated
692  * in the CRTC's flush.
693  */
vc4_plane_atomic_check(struct drm_plane * plane,struct drm_plane_state * state)694 static int vc4_plane_atomic_check(struct drm_plane *plane,
695 				  struct drm_plane_state *state)
696 {
697 	struct vc4_plane_state *vc4_state = to_vc4_plane_state(state);
698 
699 	vc4_state->dlist_count = 0;
700 
701 	if (plane_enabled(state))
702 		return vc4_plane_mode_set(plane, state);
703 	else
704 		return 0;
705 }
706 
vc4_plane_atomic_update(struct drm_plane * plane,struct drm_plane_state * old_state)707 static void vc4_plane_atomic_update(struct drm_plane *plane,
708 				    struct drm_plane_state *old_state)
709 {
710 	/* No contents here.  Since we don't know where in the CRTC's
711 	 * dlist we should be stored, our dlist is uploaded to the
712 	 * hardware with vc4_plane_write_dlist() at CRTC atomic_flush
713 	 * time.
714 	 */
715 }
716 
vc4_plane_write_dlist(struct drm_plane * plane,u32 __iomem * dlist)717 u32 vc4_plane_write_dlist(struct drm_plane *plane, u32 __iomem *dlist)
718 {
719 	struct vc4_plane_state *vc4_state = to_vc4_plane_state(plane->state);
720 	int i;
721 
722 	vc4_state->hw_dlist = dlist;
723 
724 	/* Can't memcpy_toio() because it needs to be 32-bit writes. */
725 	for (i = 0; i < vc4_state->dlist_count; i++)
726 		writel(vc4_state->dlist[i], &dlist[i]);
727 
728 	return vc4_state->dlist_count;
729 }
730 
vc4_plane_dlist_size(const struct drm_plane_state * state)731 u32 vc4_plane_dlist_size(const struct drm_plane_state *state)
732 {
733 	const struct vc4_plane_state *vc4_state =
734 		container_of(state, typeof(*vc4_state), base);
735 
736 	return vc4_state->dlist_count;
737 }
738 
739 /* Updates the plane to immediately (well, once the FIFO needs
740  * refilling) scan out from at a new framebuffer.
741  */
vc4_plane_async_set_fb(struct drm_plane * plane,struct drm_framebuffer * fb)742 void vc4_plane_async_set_fb(struct drm_plane *plane, struct drm_framebuffer *fb)
743 {
744 	struct vc4_plane_state *vc4_state = to_vc4_plane_state(plane->state);
745 	struct drm_gem_cma_object *bo = drm_fb_cma_get_gem_obj(fb, 0);
746 	uint32_t addr;
747 
748 	/* We're skipping the address adjustment for negative origin,
749 	 * because this is only called on the primary plane.
750 	 */
751 	WARN_ON_ONCE(plane->state->crtc_x < 0 || plane->state->crtc_y < 0);
752 	addr = bo->paddr + fb->offsets[0];
753 
754 	/* Write the new address into the hardware immediately.  The
755 	 * scanout will start from this address as soon as the FIFO
756 	 * needs to refill with pixels.
757 	 */
758 	writel(addr, &vc4_state->hw_dlist[vc4_state->ptr0_offset]);
759 
760 	/* Also update the CPU-side dlist copy, so that any later
761 	 * atomic updates that don't do a new modeset on our plane
762 	 * also use our updated address.
763 	 */
764 	vc4_state->dlist[vc4_state->ptr0_offset] = addr;
765 }
766 
vc4_prepare_fb(struct drm_plane * plane,struct drm_plane_state * state)767 static int vc4_prepare_fb(struct drm_plane *plane,
768 			  struct drm_plane_state *state)
769 {
770 	struct vc4_bo *bo;
771 	struct dma_fence *fence;
772 
773 	if ((plane->state->fb == state->fb) || !state->fb)
774 		return 0;
775 
776 	bo = to_vc4_bo(&drm_fb_cma_get_gem_obj(state->fb, 0)->base);
777 	fence = reservation_object_get_excl_rcu(bo->resv);
778 	drm_atomic_set_fence_for_plane(state, fence);
779 
780 	return 0;
781 }
782 
783 static const struct drm_plane_helper_funcs vc4_plane_helper_funcs = {
784 	.atomic_check = vc4_plane_atomic_check,
785 	.atomic_update = vc4_plane_atomic_update,
786 	.prepare_fb = vc4_prepare_fb,
787 };
788 
vc4_plane_destroy(struct drm_plane * plane)789 static void vc4_plane_destroy(struct drm_plane *plane)
790 {
791 	drm_plane_helper_disable(plane);
792 	drm_plane_cleanup(plane);
793 }
794 
795 /* Implements immediate (non-vblank-synced) updates of the cursor
796  * position, or falls back to the atomic helper otherwise.
797  */
798 static int
vc4_update_plane(struct drm_plane * plane,struct drm_crtc * crtc,struct drm_framebuffer * fb,int crtc_x,int crtc_y,unsigned int crtc_w,unsigned int crtc_h,uint32_t src_x,uint32_t src_y,uint32_t src_w,uint32_t src_h,struct drm_modeset_acquire_ctx * ctx)799 vc4_update_plane(struct drm_plane *plane,
800 		 struct drm_crtc *crtc,
801 		 struct drm_framebuffer *fb,
802 		 int crtc_x, int crtc_y,
803 		 unsigned int crtc_w, unsigned int crtc_h,
804 		 uint32_t src_x, uint32_t src_y,
805 		 uint32_t src_w, uint32_t src_h,
806 		 struct drm_modeset_acquire_ctx *ctx)
807 {
808 	struct drm_plane_state *plane_state;
809 	struct vc4_plane_state *vc4_state;
810 
811 	if (plane != crtc->cursor)
812 		goto out;
813 
814 	plane_state = plane->state;
815 	vc4_state = to_vc4_plane_state(plane_state);
816 
817 	if (!plane_state)
818 		goto out;
819 
820 	/* No configuring new scaling in the fast path. */
821 	if (crtc_w != plane_state->crtc_w ||
822 	    crtc_h != plane_state->crtc_h ||
823 	    src_w != plane_state->src_w ||
824 	    src_h != plane_state->src_h) {
825 		goto out;
826 	}
827 
828 	if (fb != plane_state->fb) {
829 		drm_atomic_set_fb_for_plane(plane->state, fb);
830 		vc4_plane_async_set_fb(plane, fb);
831 	}
832 
833 	/* Set the cursor's position on the screen.  This is the
834 	 * expected change from the drm_mode_cursor_universal()
835 	 * helper.
836 	 */
837 	plane_state->crtc_x = crtc_x;
838 	plane_state->crtc_y = crtc_y;
839 
840 	/* Allow changing the start position within the cursor BO, if
841 	 * that matters.
842 	 */
843 	plane_state->src_x = src_x;
844 	plane_state->src_y = src_y;
845 
846 	/* Update the display list based on the new crtc_x/y. */
847 	vc4_plane_atomic_check(plane, plane_state);
848 
849 	/* Note that we can't just call vc4_plane_write_dlist()
850 	 * because that would smash the context data that the HVS is
851 	 * currently using.
852 	 */
853 	writel(vc4_state->dlist[vc4_state->pos0_offset],
854 	       &vc4_state->hw_dlist[vc4_state->pos0_offset]);
855 	writel(vc4_state->dlist[vc4_state->pos2_offset],
856 	       &vc4_state->hw_dlist[vc4_state->pos2_offset]);
857 	writel(vc4_state->dlist[vc4_state->ptr0_offset],
858 	       &vc4_state->hw_dlist[vc4_state->ptr0_offset]);
859 
860 	return 0;
861 
862 out:
863 	return drm_atomic_helper_update_plane(plane, crtc, fb,
864 					      crtc_x, crtc_y,
865 					      crtc_w, crtc_h,
866 					      src_x, src_y,
867 					      src_w, src_h,
868 					      ctx);
869 }
870 
871 static const struct drm_plane_funcs vc4_plane_funcs = {
872 	.update_plane = vc4_update_plane,
873 	.disable_plane = drm_atomic_helper_disable_plane,
874 	.destroy = vc4_plane_destroy,
875 	.set_property = NULL,
876 	.reset = vc4_plane_reset,
877 	.atomic_duplicate_state = vc4_plane_duplicate_state,
878 	.atomic_destroy_state = vc4_plane_destroy_state,
879 };
880 
vc4_plane_init(struct drm_device * dev,enum drm_plane_type type)881 struct drm_plane *vc4_plane_init(struct drm_device *dev,
882 				 enum drm_plane_type type)
883 {
884 	struct drm_plane *plane = NULL;
885 	struct vc4_plane *vc4_plane;
886 	u32 formats[ARRAY_SIZE(hvs_formats)];
887 	u32 num_formats = 0;
888 	int ret = 0;
889 	unsigned i;
890 
891 	vc4_plane = devm_kzalloc(dev->dev, sizeof(*vc4_plane),
892 				 GFP_KERNEL);
893 	if (!vc4_plane)
894 		return ERR_PTR(-ENOMEM);
895 
896 	for (i = 0; i < ARRAY_SIZE(hvs_formats); i++) {
897 		/* Don't allow YUV in cursor planes, since that means
898 		 * tuning on the scaler, which we don't allow for the
899 		 * cursor.
900 		 */
901 		if (type != DRM_PLANE_TYPE_CURSOR ||
902 		    hvs_formats[i].hvs < HVS_PIXEL_FORMAT_YCBCR_YUV420_3PLANE) {
903 			formats[num_formats++] = hvs_formats[i].drm;
904 		}
905 	}
906 	plane = &vc4_plane->base;
907 	ret = drm_universal_plane_init(dev, plane, 0,
908 				       &vc4_plane_funcs,
909 				       formats, num_formats,
910 				       NULL, type, NULL);
911 
912 	drm_plane_helper_add(plane, &vc4_plane_helper_funcs);
913 
914 	return plane;
915 }
916