vkms_composer.c

  1// SPDX-License-Identifier: GPL-2.0+
  2
  3#include <linux/crc32.h>
  4
  5#include <drm/drm_atomic.h>
  6#include <drm/drm_atomic_helper.h>
  7#include <drm/drm_blend.h>
  8#include <drm/drm_colorop.h>
  9#include <drm/drm_fourcc.h>
 10#include <drm/drm_fixed.h>
 11#include <drm/drm_gem_framebuffer_helper.h>
 12#include <drm/drm_print.h>
 13#include <drm/drm_vblank.h>
 14#include <linux/minmax.h>
 15#include <kunit/visibility.h>
 16
 17#include "vkms_composer.h"
 18#include "vkms_luts.h"
 19
 20static u16 pre_mul_blend_channel(u16 src, u16 dst, u16 alpha)
 21{
 22	u32 new_color;
 23
 24	new_color = (src * 0xffff + dst * (0xffff - alpha));
 25
 26	return DIV_ROUND_CLOSEST(new_color, 0xffff);
 27}
 28
 29/**
 30 * pre_mul_alpha_blend - alpha blending equation
 31 * @stage_buffer: The line with the pixels from src_plane
 32 * @output_buffer: A line buffer that receives all the blends output
 33 * @x_start: The start offset
 34 * @pixel_count: The number of pixels to blend
 35 *
 36 * The pixels [@x_start;@x_start+@pixel_count) in stage_buffer are blended at
 37 * [@x_start;@x_start+@pixel_count) in output_buffer.
 38 *
 39 * The current DRM assumption is that pixel color values have been already
 40 * pre-multiplied with the alpha channel values. See more
 41 * drm_plane_create_blend_mode_property(). Also, this formula assumes a
 42 * completely opaque background.
 43 */
 44static void pre_mul_alpha_blend(const struct line_buffer *stage_buffer,
 45				struct line_buffer *output_buffer, int x_start, int pixel_count)
 46{
 47	struct pixel_argb_u16 *out = &output_buffer->pixels[x_start];
 48	const struct pixel_argb_u16 *in = &stage_buffer->pixels[x_start];
 49
 50	for (int i = 0; i < pixel_count; i++) {
 51		out[i].a = (u16)0xffff;
 52		out[i].r = pre_mul_blend_channel(in[i].r, out[i].r, in[i].a);
 53		out[i].g = pre_mul_blend_channel(in[i].g, out[i].g, in[i].a);
 54		out[i].b = pre_mul_blend_channel(in[i].b, out[i].b, in[i].a);
 55	}
 56}
 57
 58
 59static void fill_background(const struct pixel_argb_u16 *background_color,
 60			    struct line_buffer *output_buffer)
 61{
 62	for (size_t i = 0; i < output_buffer->n_pixels; i++)
 63		output_buffer->pixels[i] = *background_color;
 64}
 65
 66// lerp(a, b, t) = a + (b - a) * t
 67VISIBLE_IF_KUNIT u16 lerp_u16(u16 a, u16 b, s64 t)
 68{
 69	s64 a_fp = drm_int2fixp(a);
 70	s64 b_fp = drm_int2fixp(b);
 71
 72	s64 delta = drm_fixp_mul(b_fp - a_fp, t);
 73
 74	return drm_fixp2int_round(a_fp + delta);
 75}
 76EXPORT_SYMBOL_IF_KUNIT(lerp_u16);
 77
 78VISIBLE_IF_KUNIT s64 get_lut_index(const struct vkms_color_lut *lut, u16 channel_value)
 79{
 80	s64 color_channel_fp = drm_int2fixp(channel_value);
 81
 82	return drm_fixp_mul(color_channel_fp, lut->channel_value2index_ratio);
 83}
 84EXPORT_SYMBOL_IF_KUNIT(get_lut_index);
 85
 86VISIBLE_IF_KUNIT u16 apply_lut_to_channel_value(const struct vkms_color_lut *lut, u16 channel_value,
 87						enum lut_channel channel)
 88{
 89	s64 lut_index = get_lut_index(lut, channel_value);
 90	u16 *floor_lut_value, *ceil_lut_value;
 91	u16 floor_channel_value, ceil_channel_value;
 92
 93	/*
 94	 * This checks if `struct drm_color_lut` has any gap added by the compiler
 95	 * between the struct fields.
 96	 */
 97	static_assert(sizeof(struct drm_color_lut) == sizeof(__u16) * 4);
 98
 99	floor_lut_value = (__u16 *)&lut->base[drm_fixp2int(lut_index)];
100	if (drm_fixp2int(lut_index) == (lut->lut_length - 1))
101		/* We're at the end of the LUT array, use same value for ceil and floor */
102		ceil_lut_value = floor_lut_value;
103	else
104		ceil_lut_value = (__u16 *)&lut->base[drm_fixp2int_ceil(lut_index)];
105
106	floor_channel_value = floor_lut_value[channel];
107	ceil_channel_value = ceil_lut_value[channel];
108
109	return lerp_u16(floor_channel_value, ceil_channel_value,
110			lut_index & DRM_FIXED_DECIMAL_MASK);
111}
112EXPORT_SYMBOL_IF_KUNIT(apply_lut_to_channel_value);
113
114
115static void apply_lut(const struct vkms_crtc_state *crtc_state, struct line_buffer *output_buffer)
116{
117	if (!crtc_state->gamma_lut.base)
118		return;
119
120	if (!crtc_state->gamma_lut.lut_length)
121		return;
122
123	for (size_t x = 0; x < output_buffer->n_pixels; x++) {
124		struct pixel_argb_u16 *pixel = &output_buffer->pixels[x];
125
126		pixel->r = apply_lut_to_channel_value(&crtc_state->gamma_lut, pixel->r, LUT_RED);
127		pixel->g = apply_lut_to_channel_value(&crtc_state->gamma_lut, pixel->g, LUT_GREEN);
128		pixel->b = apply_lut_to_channel_value(&crtc_state->gamma_lut, pixel->b, LUT_BLUE);
129	}
130}
131
132VISIBLE_IF_KUNIT void apply_3x4_matrix(struct pixel_argb_s32 *pixel,
133				       const struct drm_color_ctm_3x4 *matrix)
134{
135	s64 rf, gf, bf;
136	s64 r, g, b;
137
138	r = drm_int2fixp(pixel->r);
139	g = drm_int2fixp(pixel->g);
140	b = drm_int2fixp(pixel->b);
141
142	rf = drm_fixp_mul(drm_sm2fixp(matrix->matrix[0]), r) +
143	     drm_fixp_mul(drm_sm2fixp(matrix->matrix[1]), g) +
144	     drm_fixp_mul(drm_sm2fixp(matrix->matrix[2]), b) +
145	     drm_sm2fixp(matrix->matrix[3]);
146
147	gf = drm_fixp_mul(drm_sm2fixp(matrix->matrix[4]), r) +
148	     drm_fixp_mul(drm_sm2fixp(matrix->matrix[5]), g) +
149	     drm_fixp_mul(drm_sm2fixp(matrix->matrix[6]), b) +
150	     drm_sm2fixp(matrix->matrix[7]);
151
152	bf = drm_fixp_mul(drm_sm2fixp(matrix->matrix[8]), r) +
153	     drm_fixp_mul(drm_sm2fixp(matrix->matrix[9]), g) +
154	     drm_fixp_mul(drm_sm2fixp(matrix->matrix[10]), b) +
155	     drm_sm2fixp(matrix->matrix[11]);
156
157	pixel->r = drm_fixp2int_round(rf);
158	pixel->g = drm_fixp2int_round(gf);
159	pixel->b = drm_fixp2int_round(bf);
160}
161EXPORT_SYMBOL_IF_KUNIT(apply_3x4_matrix);
162
163static void apply_colorop(struct pixel_argb_s32 *pixel, struct drm_colorop *colorop)
164{
165	struct drm_colorop_state *colorop_state = colorop->state;
166	struct drm_device *dev = colorop->dev;
167
168	if (colorop->type == DRM_COLOROP_1D_CURVE) {
169		switch (colorop_state->curve_1d_type) {
170		case DRM_COLOROP_1D_CURVE_SRGB_INV_EOTF:
171			pixel->r = apply_lut_to_channel_value(&srgb_inv_eotf, pixel->r, LUT_RED);
172			pixel->g = apply_lut_to_channel_value(&srgb_inv_eotf, pixel->g, LUT_GREEN);
173			pixel->b = apply_lut_to_channel_value(&srgb_inv_eotf, pixel->b, LUT_BLUE);
174			break;
175		case DRM_COLOROP_1D_CURVE_SRGB_EOTF:
176			pixel->r = apply_lut_to_channel_value(&srgb_eotf, pixel->r, LUT_RED);
177			pixel->g = apply_lut_to_channel_value(&srgb_eotf, pixel->g, LUT_GREEN);
178			pixel->b = apply_lut_to_channel_value(&srgb_eotf, pixel->b, LUT_BLUE);
179			break;
180		default:
181			drm_WARN_ONCE(dev, true,
182				      "unknown colorop 1D curve type %d\n",
183				      colorop_state->curve_1d_type);
184			break;
185		}
186	} else if (colorop->type == DRM_COLOROP_CTM_3X4) {
187		if (colorop_state->data)
188			apply_3x4_matrix(pixel,
189					 (struct drm_color_ctm_3x4 *)colorop_state->data->data);
190	}
191}
192
193static void pre_blend_color_transform(const struct vkms_plane_state *plane_state,
194				      struct line_buffer *output_buffer)
195{
196	struct pixel_argb_s32 pixel;
197
198	for (size_t x = 0; x < output_buffer->n_pixels; x++) {
199		struct drm_colorop *colorop = plane_state->base.base.color_pipeline;
200
201		/*
202		 * Some operations, such as applying a BT709 encoding matrix,
203		 * followed by a decoding matrix, require that we preserve
204		 * values above 1.0 and below 0.0 until the end of the pipeline.
205		 *
206		 * Pack the 16-bit UNORM values into s32 to give us head-room to
207		 * avoid clipping until we're at the end of the pipeline. Clip
208		 * intentionally at the end of the pipeline before packing
209		 * UNORM values back into u16.
210		 */
211		pixel.a = output_buffer->pixels[x].a;
212		pixel.r = output_buffer->pixels[x].r;
213		pixel.g = output_buffer->pixels[x].g;
214		pixel.b = output_buffer->pixels[x].b;
215
216		while (colorop) {
217			struct drm_colorop_state *colorop_state;
218
219			colorop_state = colorop->state;
220
221			if (!colorop_state)
222				return;
223
224			if (!colorop_state->bypass)
225				apply_colorop(&pixel, colorop);
226
227			colorop = colorop->next;
228		}
229
230		/* clamp values */
231		output_buffer->pixels[x].a = clamp_val(pixel.a, 0, 0xffff);
232		output_buffer->pixels[x].r = clamp_val(pixel.r, 0, 0xffff);
233		output_buffer->pixels[x].g = clamp_val(pixel.g, 0, 0xffff);
234		output_buffer->pixels[x].b = clamp_val(pixel.b, 0, 0xffff);
235	}
236}
237
238/**
239 * direction_for_rotation() - Get the correct reading direction for a given rotation
240 *
241 * @rotation: Rotation to analyze. It correspond the field @frame_info.rotation.
242 *
243 * This function will use the @rotation setting of a source plane to compute the reading
244 * direction in this plane which correspond to a "left to right writing" in the CRTC.
245 * For example, if the buffer is reflected on X axis, the pixel must be read from right to left
246 * to be written from left to right on the CRTC.
247 */
248static enum pixel_read_direction direction_for_rotation(unsigned int rotation)
249{
250	struct drm_rect tmp_a, tmp_b;
251	int x, y;
252
253	/*
254	 * Points A and B are depicted as zero-size rectangles on the CRTC.
255	 * The CRTC writing direction is from A to B. The plane reading direction
256	 * is discovered by inverse-transforming A and B.
257	 * The reading direction is computed by rotating the vector AB (top-left to top-right) in a
258	 * 1x1 square.
259	 */
260
261	tmp_a = DRM_RECT_INIT(0, 0, 0, 0);
262	tmp_b = DRM_RECT_INIT(1, 0, 0, 0);
263	drm_rect_rotate_inv(&tmp_a, 1, 1, rotation);
264	drm_rect_rotate_inv(&tmp_b, 1, 1, rotation);
265
266	x = tmp_b.x1 - tmp_a.x1;
267	y = tmp_b.y1 - tmp_a.y1;
268
269	if (x == 1 && y == 0)
270		return READ_LEFT_TO_RIGHT;
271	else if (x == -1 && y == 0)
272		return READ_RIGHT_TO_LEFT;
273	else if (y == 1 && x == 0)
274		return READ_TOP_TO_BOTTOM;
275	else if (y == -1 && x == 0)
276		return READ_BOTTOM_TO_TOP;
277
278	WARN_ONCE(true, "The inverse of the rotation gives an incorrect direction.");
279	return READ_LEFT_TO_RIGHT;
280}
281
282/**
283 * clamp_line_coordinates() - Compute and clamp the coordinate to read and write during the blend
284 * process.
285 *
286 * @direction: direction of the reading
287 * @current_plane: current plane blended
288 * @src_line: source line of the reading. Only the top-left coordinate is used. This rectangle
289 * must be rotated and have a shape of 1*pixel_count if @direction is vertical and a shape of
290 * pixel_count*1 if @direction is horizontal.
291 * @src_x_start: x start coordinate for the line reading
292 * @src_y_start: y start coordinate for the line reading
293 * @dst_x_start: x coordinate to blend the read line
294 * @pixel_count: number of pixels to blend
295 *
296 * This function is mainly a safety net to avoid reading outside the source buffer. As the
297 * userspace should never ask to read outside the source plane, all the cases covered here should
298 * be dead code.
299 */
300static void clamp_line_coordinates(enum pixel_read_direction direction,
301				   const struct vkms_plane_state *current_plane,
302				   const struct drm_rect *src_line, int *src_x_start,
303				   int *src_y_start, int *dst_x_start, int *pixel_count)
304{
305	/* By default the start points are correct */
306	*src_x_start = src_line->x1;
307	*src_y_start = src_line->y1;
308	*dst_x_start = current_plane->frame_info->dst.x1;
309
310	/* Get the correct number of pixel to blend, it depends of the direction */
311	switch (direction) {
312	case READ_LEFT_TO_RIGHT:
313	case READ_RIGHT_TO_LEFT:
314		*pixel_count = drm_rect_width(src_line);
315		break;
316	case READ_BOTTOM_TO_TOP:
317	case READ_TOP_TO_BOTTOM:
318		*pixel_count = drm_rect_height(src_line);
319		break;
320	}
321
322	/*
323	 * Clamp the coordinates to avoid reading outside the buffer
324	 *
325	 * This is mainly a security check to avoid reading outside the buffer, the userspace
326	 * should never request to read outside the source buffer.
327	 */
328	switch (direction) {
329	case READ_LEFT_TO_RIGHT:
330	case READ_RIGHT_TO_LEFT:
331		if (*src_x_start < 0) {
332			*pixel_count += *src_x_start;
333			*dst_x_start -= *src_x_start;
334			*src_x_start = 0;
335		}
336		if (*src_x_start + *pixel_count > current_plane->frame_info->fb->width)
337			*pixel_count = max(0, (int)current_plane->frame_info->fb->width -
338				*src_x_start);
339		break;
340	case READ_BOTTOM_TO_TOP:
341	case READ_TOP_TO_BOTTOM:
342		if (*src_y_start < 0) {
343			*pixel_count += *src_y_start;
344			*dst_x_start -= *src_y_start;
345			*src_y_start = 0;
346		}
347		if (*src_y_start + *pixel_count > current_plane->frame_info->fb->height)
348			*pixel_count = max(0, (int)current_plane->frame_info->fb->height -
349				*src_y_start);
350		break;
351	}
352}
353
354/**
355 * blend_line() - Blend a line from a plane to the output buffer
356 *
357 * @current_plane: current plane to work on
358 * @y: line to write in the output buffer
359 * @crtc_x_limit: width of the output buffer
360 * @stage_buffer: temporary buffer to convert the pixel line from the source buffer
361 * @output_buffer: buffer to blend the read line into.
362 */
363static void blend_line(struct vkms_plane_state *current_plane, int y,
364		       int crtc_x_limit, struct line_buffer *stage_buffer,
365		       struct line_buffer *output_buffer)
366{
367	int src_x_start, src_y_start, dst_x_start, pixel_count;
368	struct drm_rect dst_line, tmp_src, src_line;
369
370	/* Avoid rendering useless lines */
371	if (y < current_plane->frame_info->dst.y1 ||
372	    y >= current_plane->frame_info->dst.y2)
373		return;
374
375	/*
376	 * dst_line is the line to copy. The initial coordinates are inside the
377	 * destination framebuffer, and then drm_rect_* helpers are used to
378	 * compute the correct position into the source framebuffer.
379	 */
380	dst_line = DRM_RECT_INIT(current_plane->frame_info->dst.x1, y,
381				 drm_rect_width(&current_plane->frame_info->dst),
382				 1);
383
384	drm_rect_fp_to_int(&tmp_src, &current_plane->frame_info->src);
385
386	/*
387	 * [1]: Clamping src_line to the crtc_x_limit to avoid writing outside of
388	 * the destination buffer
389	 */
390	dst_line.x1 = max_t(int, dst_line.x1, 0);
391	dst_line.x2 = min_t(int, dst_line.x2, crtc_x_limit);
392	/* The destination is completely outside of the crtc. */
393	if (dst_line.x2 <= dst_line.x1)
394		return;
395
396	src_line = dst_line;
397
398	/*
399	 * Transform the coordinate x/y from the crtc to coordinates into
400	 * coordinates for the src buffer.
401	 *
402	 * - Cancel the offset of the dst buffer.
403	 * - Invert the rotation. This assumes that
404	 *   dst = drm_rect_rotate(src, rotation) (dst and src have the
405	 *   same size, but can be rotated).
406	 * - Apply the offset of the source rectangle to the coordinate.
407	 */
408	drm_rect_translate(&src_line, -current_plane->frame_info->dst.x1,
409			   -current_plane->frame_info->dst.y1);
410	drm_rect_rotate_inv(&src_line, drm_rect_width(&tmp_src),
411			    drm_rect_height(&tmp_src),
412			    current_plane->frame_info->rotation);
413	drm_rect_translate(&src_line, tmp_src.x1, tmp_src.y1);
414
415	/* Get the correct reading direction in the source buffer. */
416
417	enum pixel_read_direction direction =
418		direction_for_rotation(current_plane->frame_info->rotation);
419
420	/* [2]: Compute and clamp the number of pixel to read */
421	clamp_line_coordinates(direction, current_plane, &src_line, &src_x_start, &src_y_start,
422			       &dst_x_start, &pixel_count);
423
424	if (pixel_count <= 0) {
425		/* Nothing to read, so avoid multiple function calls */
426		return;
427	}
428
429	/*
430	 * Modify the starting point to take in account the rotation
431	 *
432	 * src_line is the top-left corner, so when reading READ_RIGHT_TO_LEFT or
433	 * READ_BOTTOM_TO_TOP, it must be changed to the top-right/bottom-left
434	 * corner.
435	 */
436	if (direction == READ_RIGHT_TO_LEFT) {
437		// src_x_start is now the right point
438		src_x_start += pixel_count - 1;
439	} else if (direction == READ_BOTTOM_TO_TOP) {
440		// src_y_start is now the bottom point
441		src_y_start += pixel_count - 1;
442	}
443
444	/*
445	 * Perform the conversion and the blending
446	 *
447	 * Here we know that the read line (x_start, y_start, pixel_count) is
448	 * inside the source buffer [2] and we don't write outside the stage
449	 * buffer [1].
450	 */
451	current_plane->pixel_read_line(current_plane, src_x_start, src_y_start, direction,
452				       pixel_count, &stage_buffer->pixels[dst_x_start]);
453	pre_blend_color_transform(current_plane, stage_buffer);
454	pre_mul_alpha_blend(stage_buffer, output_buffer,
455			    dst_x_start, pixel_count);
456}
457
458/**
459 * blend - blend the pixels from all planes and compute crc
460 * @wb: The writeback frame buffer metadata
461 * @crtc_state: The crtc state
462 * @crc32: The crc output of the final frame
463 * @output_buffer: A buffer of a row that will receive the result of the blend(s)
464 * @stage_buffer: The line with the pixels from plane being blend to the output
465 * @row_size: The size, in bytes, of a single row
466 *
467 * This function blends the pixels (Using the `pre_mul_alpha_blend`)
468 * from all planes, calculates the crc32 of the output from the former step,
469 * and, if necessary, convert and store the output to the writeback buffer.
470 */
471static void blend(struct vkms_writeback_job *wb,
472		  struct vkms_crtc_state *crtc_state,
473		  u32 *crc32, struct line_buffer *stage_buffer,
474		  struct line_buffer *output_buffer, size_t row_size)
475{
476	struct vkms_plane_state **plane = crtc_state->active_planes;
477	u32 n_active_planes = crtc_state->num_active_planes;
478
479	const struct pixel_argb_u16 background_color = { .a = 0xffff };
480
481	int crtc_y_limit = crtc_state->base.mode.vdisplay;
482	int crtc_x_limit = crtc_state->base.mode.hdisplay;
483
484	/*
485	 * The planes are composed line-by-line to avoid heavy memory usage. It is a necessary
486	 * complexity to avoid poor blending performance.
487	 *
488	 * The function pixel_read_line callback is used to read a line, using an efficient
489	 * algorithm for a specific format, into the staging buffer.
490	 */
491	for (int y = 0; y < crtc_y_limit; y++) {
492		fill_background(&background_color, output_buffer);
493
494		/* The active planes are composed associatively in z-order. */
495		for (size_t i = 0; i < n_active_planes; i++) {
496			blend_line(plane[i], y, crtc_x_limit, stage_buffer, output_buffer);
497		}
498
499		apply_lut(crtc_state, output_buffer);
500
501		*crc32 = crc32_le(*crc32, (void *)output_buffer->pixels, row_size);
502
503		if (wb)
504			vkms_writeback_row(wb, output_buffer, y);
505	}
506}
507
508static int check_format_funcs(struct vkms_crtc_state *crtc_state,
509			      struct vkms_writeback_job *active_wb)
510{
511	struct vkms_plane_state **planes = crtc_state->active_planes;
512	u32 n_active_planes = crtc_state->num_active_planes;
513
514	for (size_t i = 0; i < n_active_planes; i++)
515		if (!planes[i]->pixel_read_line)
516			return -1;
517
518	if (active_wb && !active_wb->pixel_write)
519		return -1;
520
521	return 0;
522}
523
524static int check_iosys_map(struct vkms_crtc_state *crtc_state)
525{
526	struct vkms_plane_state **plane_state = crtc_state->active_planes;
527	u32 n_active_planes = crtc_state->num_active_planes;
528
529	for (size_t i = 0; i < n_active_planes; i++)
530		if (iosys_map_is_null(&plane_state[i]->frame_info->map[0]))
531			return -1;
532
533	return 0;
534}
535
536static int compose_active_planes(struct vkms_writeback_job *active_wb,
537				 struct vkms_crtc_state *crtc_state,
538				 u32 *crc32)
539{
540	size_t line_width, pixel_size = sizeof(struct pixel_argb_u16);
541	struct line_buffer output_buffer, stage_buffer;
542	int ret = 0;
543
544	/*
545	 * This check exists so we can call `crc32_le` for the entire line
546	 * instead doing it for each channel of each pixel in case
547	 * `struct `pixel_argb_u16` had any gap added by the compiler
548	 * between the struct fields.
549	 */
550	static_assert(sizeof(struct pixel_argb_u16) == 8);
551
552	if (WARN_ON(check_iosys_map(crtc_state)))
553		return -EINVAL;
554
555	if (WARN_ON(check_format_funcs(crtc_state, active_wb)))
556		return -EINVAL;
557
558	line_width = crtc_state->base.mode.hdisplay;
559	stage_buffer.n_pixels = line_width;
560	output_buffer.n_pixels = line_width;
561
562	stage_buffer.pixels = kvmalloc(line_width * pixel_size, GFP_KERNEL);
563	if (!stage_buffer.pixels) {
564		DRM_ERROR("Cannot allocate memory for the output line buffer");
565		return -ENOMEM;
566	}
567
568	output_buffer.pixels = kvmalloc(line_width * pixel_size, GFP_KERNEL);
569	if (!output_buffer.pixels) {
570		DRM_ERROR("Cannot allocate memory for intermediate line buffer");
571		ret = -ENOMEM;
572		goto free_stage_buffer;
573	}
574
575	blend(active_wb, crtc_state, crc32, &stage_buffer,
576	      &output_buffer, line_width * pixel_size);
577
578	kvfree(output_buffer.pixels);
579free_stage_buffer:
580	kvfree(stage_buffer.pixels);
581
582	return ret;
583}
584
585/**
586 * vkms_composer_worker - ordered work_struct to compute CRC
587 *
588 * @work: work_struct
589 *
590 * Work handler for composing and computing CRCs. work_struct scheduled in
591 * an ordered workqueue that's periodically scheduled to run by
592 * vkms_vblank_simulate() and flushed at vkms_atomic_commit_tail().
593 */
594void vkms_composer_worker(struct work_struct *work)
595{
596	struct vkms_crtc_state *crtc_state = container_of(work,
597							  struct vkms_crtc_state,
598							  composer_work);
599	struct drm_crtc *crtc = crtc_state->base.crtc;
600	struct vkms_writeback_job *active_wb = crtc_state->active_writeback;
601	struct vkms_output *out = drm_crtc_to_vkms_output(crtc);
602	bool crc_pending, wb_pending;
603	u64 frame_start, frame_end;
604	u32 crc32 = 0;
605	int ret;
606
607	spin_lock_irq(&out->composer_lock);
608	frame_start = crtc_state->frame_start;
609	frame_end = crtc_state->frame_end;
610	crc_pending = crtc_state->crc_pending;
611	wb_pending = crtc_state->wb_pending;
612	crtc_state->frame_start = 0;
613	crtc_state->frame_end = 0;
614	crtc_state->crc_pending = false;
615
616	if (crtc->state->gamma_lut) {
617		s64 max_lut_index_fp;
618		s64 u16_max_fp = drm_int2fixp(0xffff);
619
620		crtc_state->gamma_lut.base = (struct drm_color_lut *)crtc->state->gamma_lut->data;
621		crtc_state->gamma_lut.lut_length =
622			crtc->state->gamma_lut->length / sizeof(struct drm_color_lut);
623		max_lut_index_fp = drm_int2fixp(crtc_state->gamma_lut.lut_length - 1);
624		crtc_state->gamma_lut.channel_value2index_ratio = drm_fixp_div(max_lut_index_fp,
625									       u16_max_fp);
626
627	} else {
628		crtc_state->gamma_lut.base = NULL;
629	}
630
631	spin_unlock_irq(&out->composer_lock);
632
633	/*
634	 * We raced with the vblank hrtimer and previous work already computed
635	 * the crc, nothing to do.
636	 */
637	if (!crc_pending)
638		return;
639
640	if (wb_pending)
641		ret = compose_active_planes(active_wb, crtc_state, &crc32);
642	else
643		ret = compose_active_planes(NULL, crtc_state, &crc32);
644
645	if (ret)
646		return;
647
648	if (wb_pending) {
649		drm_writeback_signal_completion(&out->wb_connector, 0);
650		spin_lock_irq(&out->composer_lock);
651		crtc_state->wb_pending = false;
652		spin_unlock_irq(&out->composer_lock);
653	}
654
655	/*
656	 * The worker can fall behind the vblank hrtimer, make sure we catch up.
657	 */
658	while (frame_start <= frame_end)
659		drm_crtc_add_crc_entry(crtc, true, frame_start++, &crc32);
660}
661
662static const char *const pipe_crc_sources[] = { "auto" };
663
664const char *const *vkms_get_crc_sources(struct drm_crtc *crtc,
665					size_t *count)
666{
667	*count = ARRAY_SIZE(pipe_crc_sources);
668	return pipe_crc_sources;
669}
670
671static int vkms_crc_parse_source(const char *src_name, bool *enabled)
672{
673	int ret = 0;
674
675	if (!src_name) {
676		*enabled = false;
677	} else if (strcmp(src_name, "auto") == 0) {
678		*enabled = true;
679	} else {
680		*enabled = false;
681		ret = -EINVAL;
682	}
683
684	return ret;
685}
686
687int vkms_verify_crc_source(struct drm_crtc *crtc, const char *src_name,
688			   size_t *values_cnt)
689{
690	bool enabled;
691
692	if (vkms_crc_parse_source(src_name, &enabled) < 0) {
693		DRM_DEBUG_DRIVER("unknown source %s\n", src_name);
694		return -EINVAL;
695	}
696
697	*values_cnt = 1;
698
699	return 0;
700}
701
702void vkms_set_composer(struct vkms_output *out, bool enabled)
703{
704	bool old_enabled;
705
706	if (enabled)
707		drm_crtc_vblank_get(&out->crtc);
708
709	spin_lock_irq(&out->lock);
710	old_enabled = out->composer_enabled;
711	out->composer_enabled = enabled;
712	spin_unlock_irq(&out->lock);
713
714	if (old_enabled)
715		drm_crtc_vblank_put(&out->crtc);
716}
717
718int vkms_set_crc_source(struct drm_crtc *crtc, const char *src_name)
719{
720	struct vkms_output *out = drm_crtc_to_vkms_output(crtc);
721	bool enabled = false;
722	int ret = 0;
723
724	ret = vkms_crc_parse_source(src_name, &enabled);
725
726	vkms_set_composer(out, enabled);
727
728	return ret;
729}