xref: /linux/drivers/media/platform/renesas/vsp1/vsp1_rpf.c (revision 2b0cfa6e49566c8fa6759734cf821aa6e8271a9e)
1 // SPDX-License-Identifier: GPL-2.0+
2 /*
3  * vsp1_rpf.c  --  R-Car VSP1 Read Pixel Formatter
4  *
5  * Copyright (C) 2013-2014 Renesas Electronics Corporation
6  *
7  * Contact: Laurent Pinchart (laurent.pinchart@ideasonboard.com)
8  */
9 
10 #include <linux/device.h>
11 
12 #include <media/v4l2-subdev.h>
13 
14 #include "vsp1.h"
15 #include "vsp1_dl.h"
16 #include "vsp1_pipe.h"
17 #include "vsp1_rwpf.h"
18 #include "vsp1_video.h"
19 
20 #define RPF_MAX_WIDTH				8190
21 #define RPF_MAX_HEIGHT				8190
22 
23 /* Pre extended display list command data structure. */
24 struct vsp1_extcmd_auto_fld_body {
25 	u32 top_y0;
26 	u32 bottom_y0;
27 	u32 top_c0;
28 	u32 bottom_c0;
29 	u32 top_c1;
30 	u32 bottom_c1;
31 	u32 reserved0;
32 	u32 reserved1;
33 } __packed;
34 
35 /* -----------------------------------------------------------------------------
36  * Device Access
37  */
38 
39 static inline void vsp1_rpf_write(struct vsp1_rwpf *rpf,
40 				  struct vsp1_dl_body *dlb, u32 reg, u32 data)
41 {
42 	vsp1_dl_body_write(dlb, reg + rpf->entity.index * VI6_RPF_OFFSET,
43 			       data);
44 }
45 
46 /* -----------------------------------------------------------------------------
47  * VSP1 Entity Operations
48  */
49 
50 static void rpf_configure_stream(struct vsp1_entity *entity,
51 				 struct vsp1_pipeline *pipe,
52 				 struct vsp1_dl_list *dl,
53 				 struct vsp1_dl_body *dlb)
54 {
55 	struct vsp1_rwpf *rpf = to_rwpf(&entity->subdev);
56 	const struct vsp1_format_info *fmtinfo = rpf->fmtinfo;
57 	const struct v4l2_pix_format_mplane *format = &rpf->format;
58 	const struct v4l2_mbus_framefmt *source_format;
59 	const struct v4l2_mbus_framefmt *sink_format;
60 	unsigned int left = 0;
61 	unsigned int top = 0;
62 	u32 pstride;
63 	u32 infmt;
64 
65 	/* Stride */
66 	pstride = format->plane_fmt[0].bytesperline
67 		<< VI6_RPF_SRCM_PSTRIDE_Y_SHIFT;
68 	if (format->num_planes > 1)
69 		pstride |= format->plane_fmt[1].bytesperline
70 			<< VI6_RPF_SRCM_PSTRIDE_C_SHIFT;
71 
72 	/*
73 	 * pstride has both STRIDE_Y and STRIDE_C, but multiplying the whole
74 	 * of pstride by 2 is conveniently OK here as we are multiplying both
75 	 * values.
76 	 */
77 	if (pipe->interlaced)
78 		pstride *= 2;
79 
80 	vsp1_rpf_write(rpf, dlb, VI6_RPF_SRCM_PSTRIDE, pstride);
81 
82 	/* Format */
83 	sink_format = vsp1_entity_get_pad_format(&rpf->entity,
84 						 rpf->entity.state,
85 						 RWPF_PAD_SINK);
86 	source_format = vsp1_entity_get_pad_format(&rpf->entity,
87 						   rpf->entity.state,
88 						   RWPF_PAD_SOURCE);
89 
90 	infmt = VI6_RPF_INFMT_CIPM
91 	      | (fmtinfo->hwfmt << VI6_RPF_INFMT_RDFMT_SHIFT);
92 
93 	if (fmtinfo->swap_yc)
94 		infmt |= VI6_RPF_INFMT_SPYCS;
95 	if (fmtinfo->swap_uv)
96 		infmt |= VI6_RPF_INFMT_SPUVS;
97 
98 	if (sink_format->code != source_format->code)
99 		infmt |= VI6_RPF_INFMT_CSC;
100 
101 	vsp1_rpf_write(rpf, dlb, VI6_RPF_INFMT, infmt);
102 	vsp1_rpf_write(rpf, dlb, VI6_RPF_DSWAP, fmtinfo->swap);
103 
104 	if (entity->vsp1->info->gen == 4) {
105 		u32 ext_infmt0;
106 		u32 ext_infmt1;
107 		u32 ext_infmt2;
108 
109 		switch (fmtinfo->fourcc) {
110 		case V4L2_PIX_FMT_RGBX1010102:
111 			ext_infmt0 = VI6_RPF_EXT_INFMT0_BYPP_M1_RGB10;
112 			ext_infmt1 = VI6_RPF_EXT_INFMT1_PACK_CPOS(0, 10, 20, 0);
113 			ext_infmt2 = VI6_RPF_EXT_INFMT2_PACK_CLEN(10, 10, 10, 0);
114 			break;
115 
116 		case V4L2_PIX_FMT_RGBA1010102:
117 			ext_infmt0 = VI6_RPF_EXT_INFMT0_BYPP_M1_RGB10;
118 			ext_infmt1 = VI6_RPF_EXT_INFMT1_PACK_CPOS(0, 10, 20, 30);
119 			ext_infmt2 = VI6_RPF_EXT_INFMT2_PACK_CLEN(10, 10, 10, 2);
120 			break;
121 
122 		case V4L2_PIX_FMT_ARGB2101010:
123 			ext_infmt0 = VI6_RPF_EXT_INFMT0_BYPP_M1_RGB10;
124 			ext_infmt1 = VI6_RPF_EXT_INFMT1_PACK_CPOS(2, 12, 22, 0);
125 			ext_infmt2 = VI6_RPF_EXT_INFMT2_PACK_CLEN(10, 10, 10, 2);
126 			break;
127 
128 		case V4L2_PIX_FMT_Y210:
129 			ext_infmt0 = VI6_RPF_EXT_INFMT0_F2B |
130 				     VI6_RPF_EXT_INFMT0_IPBD_Y_10 |
131 				     VI6_RPF_EXT_INFMT0_IPBD_C_10;
132 			ext_infmt1 = 0x0;
133 			ext_infmt2 = 0x0;
134 			break;
135 
136 		case V4L2_PIX_FMT_Y212:
137 			ext_infmt0 = VI6_RPF_EXT_INFMT0_F2B |
138 				     VI6_RPF_EXT_INFMT0_IPBD_Y_12 |
139 				     VI6_RPF_EXT_INFMT0_IPBD_C_12;
140 			ext_infmt1 = 0x0;
141 			ext_infmt2 = 0x0;
142 			break;
143 
144 		default:
145 			ext_infmt0 = 0;
146 			ext_infmt1 = 0;
147 			ext_infmt2 = 0;
148 			break;
149 		}
150 
151 		vsp1_rpf_write(rpf, dlb, VI6_RPF_EXT_INFMT0, ext_infmt0);
152 		vsp1_rpf_write(rpf, dlb, VI6_RPF_EXT_INFMT1, ext_infmt1);
153 		vsp1_rpf_write(rpf, dlb, VI6_RPF_EXT_INFMT2, ext_infmt2);
154 	}
155 
156 	/* Output location. */
157 	if (pipe->brx) {
158 		const struct v4l2_rect *compose;
159 
160 		compose = vsp1_entity_get_pad_selection(pipe->brx,
161 							pipe->brx->state,
162 							rpf->brx_input,
163 							V4L2_SEL_TGT_COMPOSE);
164 		left = compose->left;
165 		top = compose->top;
166 	}
167 
168 	if (pipe->interlaced)
169 		top /= 2;
170 
171 	vsp1_rpf_write(rpf, dlb, VI6_RPF_LOC,
172 		       (left << VI6_RPF_LOC_HCOORD_SHIFT) |
173 		       (top << VI6_RPF_LOC_VCOORD_SHIFT));
174 
175 	/*
176 	 * On Gen2 use the alpha channel (extended to 8 bits) when available or
177 	 * a fixed alpha value set through the V4L2_CID_ALPHA_COMPONENT control
178 	 * otherwise.
179 	 *
180 	 * The Gen3+ RPF has extended alpha capability and can both multiply the
181 	 * alpha channel by a fixed global alpha value, and multiply the pixel
182 	 * components to convert the input to premultiplied alpha.
183 	 *
184 	 * As alpha premultiplication is available in the BRx for both Gen2 and
185 	 * Gen3+ we handle it there and use the Gen3 alpha multiplier for global
186 	 * alpha multiplication only. This however prevents conversion to
187 	 * premultiplied alpha if no BRx is present in the pipeline. If that use
188 	 * case turns out to be useful we will revisit the implementation (for
189 	 * Gen3 only).
190 	 *
191 	 * We enable alpha multiplication on Gen3+ using the fixed alpha value
192 	 * set through the V4L2_CID_ALPHA_COMPONENT control when the input
193 	 * contains an alpha channel. On Gen2 the global alpha is ignored in
194 	 * that case.
195 	 *
196 	 * In all cases, disable color keying.
197 	 */
198 	vsp1_rpf_write(rpf, dlb, VI6_RPF_ALPH_SEL, VI6_RPF_ALPH_SEL_AEXT_EXT |
199 		       (fmtinfo->alpha ? VI6_RPF_ALPH_SEL_ASEL_PACKED
200 				       : VI6_RPF_ALPH_SEL_ASEL_FIXED));
201 
202 	if (entity->vsp1->info->gen >= 3) {
203 		u32 mult;
204 
205 		if (fmtinfo->alpha) {
206 			/*
207 			 * When the input contains an alpha channel enable the
208 			 * alpha multiplier. If the input is premultiplied we
209 			 * need to multiply both the alpha channel and the pixel
210 			 * components by the global alpha value to keep them
211 			 * premultiplied. Otherwise multiply the alpha channel
212 			 * only.
213 			 */
214 			bool premultiplied = format->flags
215 					   & V4L2_PIX_FMT_FLAG_PREMUL_ALPHA;
216 
217 			mult = VI6_RPF_MULT_ALPHA_A_MMD_RATIO
218 			     | (premultiplied ?
219 				VI6_RPF_MULT_ALPHA_P_MMD_RATIO :
220 				VI6_RPF_MULT_ALPHA_P_MMD_NONE);
221 		} else {
222 			/*
223 			 * When the input doesn't contain an alpha channel the
224 			 * global alpha value is applied in the unpacking unit,
225 			 * the alpha multiplier isn't needed and must be
226 			 * disabled.
227 			 */
228 			mult = VI6_RPF_MULT_ALPHA_A_MMD_NONE
229 			     | VI6_RPF_MULT_ALPHA_P_MMD_NONE;
230 		}
231 
232 		rpf->mult_alpha = mult;
233 	}
234 
235 	vsp1_rpf_write(rpf, dlb, VI6_RPF_MSK_CTRL, 0);
236 	vsp1_rpf_write(rpf, dlb, VI6_RPF_CKEY_CTRL, 0);
237 
238 }
239 
240 static void vsp1_rpf_configure_autofld(struct vsp1_rwpf *rpf,
241 				       struct vsp1_dl_list *dl)
242 {
243 	const struct v4l2_pix_format_mplane *format = &rpf->format;
244 	struct vsp1_dl_ext_cmd *cmd;
245 	struct vsp1_extcmd_auto_fld_body *auto_fld;
246 	u32 offset_y, offset_c;
247 
248 	cmd = vsp1_dl_get_pre_cmd(dl);
249 	if (WARN_ONCE(!cmd, "Failed to obtain an autofld cmd"))
250 		return;
251 
252 	/* Re-index our auto_fld to match the current RPF. */
253 	auto_fld = cmd->data;
254 	auto_fld = &auto_fld[rpf->entity.index];
255 
256 	auto_fld->top_y0 = rpf->mem.addr[0];
257 	auto_fld->top_c0 = rpf->mem.addr[1];
258 	auto_fld->top_c1 = rpf->mem.addr[2];
259 
260 	offset_y = format->plane_fmt[0].bytesperline;
261 	offset_c = format->plane_fmt[1].bytesperline;
262 
263 	auto_fld->bottom_y0 = rpf->mem.addr[0] + offset_y;
264 	auto_fld->bottom_c0 = rpf->mem.addr[1] + offset_c;
265 	auto_fld->bottom_c1 = rpf->mem.addr[2] + offset_c;
266 
267 	cmd->flags |= VI6_DL_EXT_AUTOFLD_INT | BIT(16 + rpf->entity.index);
268 }
269 
270 static void rpf_configure_frame(struct vsp1_entity *entity,
271 				struct vsp1_pipeline *pipe,
272 				struct vsp1_dl_list *dl,
273 				struct vsp1_dl_body *dlb)
274 {
275 	struct vsp1_rwpf *rpf = to_rwpf(&entity->subdev);
276 
277 	vsp1_rpf_write(rpf, dlb, VI6_RPF_VRTCOL_SET,
278 		       rpf->alpha << VI6_RPF_VRTCOL_SET_LAYA_SHIFT);
279 	vsp1_rpf_write(rpf, dlb, VI6_RPF_MULT_ALPHA, rpf->mult_alpha |
280 		       (rpf->alpha << VI6_RPF_MULT_ALPHA_RATIO_SHIFT));
281 
282 	vsp1_pipeline_propagate_alpha(pipe, dlb, rpf->alpha);
283 }
284 
285 static void rpf_configure_partition(struct vsp1_entity *entity,
286 				    struct vsp1_pipeline *pipe,
287 				    struct vsp1_dl_list *dl,
288 				    struct vsp1_dl_body *dlb)
289 {
290 	struct vsp1_rwpf *rpf = to_rwpf(&entity->subdev);
291 	struct vsp1_rwpf_memory mem = rpf->mem;
292 	struct vsp1_device *vsp1 = rpf->entity.vsp1;
293 	const struct vsp1_format_info *fmtinfo = rpf->fmtinfo;
294 	const struct v4l2_pix_format_mplane *format = &rpf->format;
295 	struct v4l2_rect crop;
296 
297 	/*
298 	 * Source size and crop offsets.
299 	 *
300 	 * The crop offsets correspond to the location of the crop
301 	 * rectangle top left corner in the plane buffer. Only two
302 	 * offsets are needed, as planes 2 and 3 always have identical
303 	 * strides.
304 	 */
305 	crop = *vsp1_rwpf_get_crop(rpf, rpf->entity.state);
306 
307 	/*
308 	 * Partition Algorithm Control
309 	 *
310 	 * The partition algorithm can split this frame into multiple
311 	 * slices. We must scale our partition window based on the pipe
312 	 * configuration to match the destination partition window.
313 	 * To achieve this, we adjust our crop to provide a 'sub-crop'
314 	 * matching the expected partition window. Only 'left' and
315 	 * 'width' need to be adjusted.
316 	 */
317 	if (pipe->partitions > 1) {
318 		crop.width = pipe->partition->rpf.width;
319 		crop.left += pipe->partition->rpf.left;
320 	}
321 
322 	if (pipe->interlaced) {
323 		crop.height = round_down(crop.height / 2, fmtinfo->vsub);
324 		crop.top = round_down(crop.top / 2, fmtinfo->vsub);
325 	}
326 
327 	vsp1_rpf_write(rpf, dlb, VI6_RPF_SRC_BSIZE,
328 		       (crop.width << VI6_RPF_SRC_BSIZE_BHSIZE_SHIFT) |
329 		       (crop.height << VI6_RPF_SRC_BSIZE_BVSIZE_SHIFT));
330 	vsp1_rpf_write(rpf, dlb, VI6_RPF_SRC_ESIZE,
331 		       (crop.width << VI6_RPF_SRC_ESIZE_EHSIZE_SHIFT) |
332 		       (crop.height << VI6_RPF_SRC_ESIZE_EVSIZE_SHIFT));
333 
334 	mem.addr[0] += crop.top * format->plane_fmt[0].bytesperline
335 		     + crop.left * fmtinfo->bpp[0] / 8;
336 
337 	if (format->num_planes > 1) {
338 		unsigned int bpl = format->plane_fmt[1].bytesperline;
339 		unsigned int offset;
340 
341 		offset = crop.top / fmtinfo->vsub * bpl
342 		       + crop.left / fmtinfo->hsub * fmtinfo->bpp[1] / 8;
343 		mem.addr[1] += offset;
344 		mem.addr[2] += offset;
345 	}
346 
347 	/*
348 	 * On Gen3+ hardware the SPUVS bit has no effect on 3-planar
349 	 * formats. Swap the U and V planes manually in that case.
350 	 */
351 	if (vsp1->info->gen >= 3 && format->num_planes == 3 &&
352 	    fmtinfo->swap_uv)
353 		swap(mem.addr[1], mem.addr[2]);
354 
355 	/*
356 	 * Interlaced pipelines will use the extended pre-cmd to process
357 	 * SRCM_ADDR_{Y,C0,C1}.
358 	 */
359 	if (pipe->interlaced) {
360 		vsp1_rpf_configure_autofld(rpf, dl);
361 	} else {
362 		vsp1_rpf_write(rpf, dlb, VI6_RPF_SRCM_ADDR_Y, mem.addr[0]);
363 		vsp1_rpf_write(rpf, dlb, VI6_RPF_SRCM_ADDR_C0, mem.addr[1]);
364 		vsp1_rpf_write(rpf, dlb, VI6_RPF_SRCM_ADDR_C1, mem.addr[2]);
365 	}
366 }
367 
368 static void rpf_partition(struct vsp1_entity *entity,
369 			  struct vsp1_pipeline *pipe,
370 			  struct vsp1_partition *partition,
371 			  unsigned int partition_idx,
372 			  struct vsp1_partition_window *window)
373 {
374 	partition->rpf = *window;
375 }
376 
377 static const struct vsp1_entity_operations rpf_entity_ops = {
378 	.configure_stream = rpf_configure_stream,
379 	.configure_frame = rpf_configure_frame,
380 	.configure_partition = rpf_configure_partition,
381 	.partition = rpf_partition,
382 };
383 
384 /* -----------------------------------------------------------------------------
385  * Initialization and Cleanup
386  */
387 
388 struct vsp1_rwpf *vsp1_rpf_create(struct vsp1_device *vsp1, unsigned int index)
389 {
390 	struct vsp1_rwpf *rpf;
391 	char name[6];
392 	int ret;
393 
394 	rpf = devm_kzalloc(vsp1->dev, sizeof(*rpf), GFP_KERNEL);
395 	if (rpf == NULL)
396 		return ERR_PTR(-ENOMEM);
397 
398 	rpf->max_width = RPF_MAX_WIDTH;
399 	rpf->max_height = RPF_MAX_HEIGHT;
400 
401 	rpf->entity.ops = &rpf_entity_ops;
402 	rpf->entity.type = VSP1_ENTITY_RPF;
403 	rpf->entity.index = index;
404 
405 	sprintf(name, "rpf.%u", index);
406 	ret = vsp1_entity_init(vsp1, &rpf->entity, name, 2, &vsp1_rwpf_subdev_ops,
407 			       MEDIA_ENT_F_PROC_VIDEO_PIXEL_FORMATTER);
408 	if (ret < 0)
409 		return ERR_PTR(ret);
410 
411 	/* Initialize the control handler. */
412 	ret = vsp1_rwpf_init_ctrls(rpf, 0);
413 	if (ret < 0) {
414 		dev_err(vsp1->dev, "rpf%u: failed to initialize controls\n",
415 			index);
416 		goto error;
417 	}
418 
419 	v4l2_ctrl_handler_setup(&rpf->ctrls);
420 
421 	return rpf;
422 
423 error:
424 	vsp1_entity_destroy(&rpf->entity);
425 	return ERR_PTR(ret);
426 }
427