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_fourcc.h>
9 #include <drm/drm_fixed.h>
10 #include <drm/drm_gem_framebuffer_helper.h>
11 #include <drm/drm_vblank.h>
12 #include <linux/minmax.h>
13 
14 #include "vkms_drv.h"
15 
16 static u16 pre_mul_blend_channel(u16 src, u16 dst, u16 alpha)
17 {
18 	u32 new_color;
19 
20 	new_color = (src * 0xffff + dst * (0xffff - alpha));
21 
22 	return DIV_ROUND_CLOSEST(new_color, 0xffff);
23 }
24 
25 /**
26  * pre_mul_alpha_blend - alpha blending equation
27  * @frame_info: Source framebuffer's metadata
28  * @stage_buffer: The line with the pixels from src_plane
29  * @output_buffer: A line buffer that receives all the blends output
30  *
31  * Using the information from the `frame_info`, this blends only the
32  * necessary pixels from the `stage_buffer` to the `output_buffer`
33  * using premultiplied blend formula.
34  *
35  * The current DRM assumption is that pixel color values have been already
36  * pre-multiplied with the alpha channel values. See more
37  * drm_plane_create_blend_mode_property(). Also, this formula assumes a
38  * completely opaque background.
39  */
40 static void pre_mul_alpha_blend(struct vkms_frame_info *frame_info,
41 				struct line_buffer *stage_buffer,
42 				struct line_buffer *output_buffer)
43 {
44 	int x_dst = frame_info->dst.x1;
45 	struct pixel_argb_u16 *out = output_buffer->pixels + x_dst;
46 	struct pixel_argb_u16 *in = stage_buffer->pixels;
47 	int x_limit = min_t(size_t, drm_rect_width(&frame_info->dst),
48 			    stage_buffer->n_pixels);
49 
50 	for (int x = 0; x < x_limit; x++) {
51 		out[x].a = (u16)0xffff;
52 		out[x].r = pre_mul_blend_channel(in[x].r, out[x].r, in[x].a);
53 		out[x].g = pre_mul_blend_channel(in[x].g, out[x].g, in[x].a);
54 		out[x].b = pre_mul_blend_channel(in[x].b, out[x].b, in[x].a);
55 	}
56 }
57 
58 static int get_y_pos(struct vkms_frame_info *frame_info, int y)
59 {
60 	if (frame_info->rotation & DRM_MODE_REFLECT_Y)
61 		return drm_rect_height(&frame_info->rotated) - y - 1;
62 
63 	switch (frame_info->rotation & DRM_MODE_ROTATE_MASK) {
64 	case DRM_MODE_ROTATE_90:
65 		return frame_info->rotated.x2 - y - 1;
66 	case DRM_MODE_ROTATE_270:
67 		return y + frame_info->rotated.x1;
68 	default:
69 		return y;
70 	}
71 }
72 
73 static bool check_limit(struct vkms_frame_info *frame_info, int pos)
74 {
75 	if (drm_rotation_90_or_270(frame_info->rotation)) {
76 		if (pos >= 0 && pos < drm_rect_width(&frame_info->rotated))
77 			return true;
78 	} else {
79 		if (pos >= frame_info->rotated.y1 && pos < frame_info->rotated.y2)
80 			return true;
81 	}
82 
83 	return false;
84 }
85 
86 static void fill_background(const struct pixel_argb_u16 *background_color,
87 			    struct line_buffer *output_buffer)
88 {
89 	for (size_t i = 0; i < output_buffer->n_pixels; i++)
90 		output_buffer->pixels[i] = *background_color;
91 }
92 
93 // lerp(a, b, t) = a + (b - a) * t
94 static u16 lerp_u16(u16 a, u16 b, s64 t)
95 {
96 	s64 a_fp = drm_int2fixp(a);
97 	s64 b_fp = drm_int2fixp(b);
98 
99 	s64 delta = drm_fixp_mul(b_fp - a_fp,  t);
100 
101 	return drm_fixp2int(a_fp + delta);
102 }
103 
104 static s64 get_lut_index(const struct vkms_color_lut *lut, u16 channel_value)
105 {
106 	s64 color_channel_fp = drm_int2fixp(channel_value);
107 
108 	return drm_fixp_mul(color_channel_fp, lut->channel_value2index_ratio);
109 }
110 
111 /*
112  * This enum is related to the positions of the variables inside
113  * `struct drm_color_lut`, so the order of both needs to be the same.
114  */
115 enum lut_channel {
116 	LUT_RED = 0,
117 	LUT_GREEN,
118 	LUT_BLUE,
119 	LUT_RESERVED
120 };
121 
122 static u16 apply_lut_to_channel_value(const struct vkms_color_lut *lut, u16 channel_value,
123 				      enum lut_channel channel)
124 {
125 	s64 lut_index = get_lut_index(lut, channel_value);
126 
127 	/*
128 	 * This checks if `struct drm_color_lut` has any gap added by the compiler
129 	 * between the struct fields.
130 	 */
131 	static_assert(sizeof(struct drm_color_lut) == sizeof(__u16) * 4);
132 
133 	u16 *floor_lut_value = (__u16 *)&lut->base[drm_fixp2int(lut_index)];
134 	u16 *ceil_lut_value = (__u16 *)&lut->base[drm_fixp2int_ceil(lut_index)];
135 
136 	u16 floor_channel_value = floor_lut_value[channel];
137 	u16 ceil_channel_value = ceil_lut_value[channel];
138 
139 	return lerp_u16(floor_channel_value, ceil_channel_value,
140 			lut_index & DRM_FIXED_DECIMAL_MASK);
141 }
142 
143 static void apply_lut(const struct vkms_crtc_state *crtc_state, struct line_buffer *output_buffer)
144 {
145 	if (!crtc_state->gamma_lut.base)
146 		return;
147 
148 	if (!crtc_state->gamma_lut.lut_length)
149 		return;
150 
151 	for (size_t x = 0; x < output_buffer->n_pixels; x++) {
152 		struct pixel_argb_u16 *pixel = &output_buffer->pixels[x];
153 
154 		pixel->r = apply_lut_to_channel_value(&crtc_state->gamma_lut, pixel->r, LUT_RED);
155 		pixel->g = apply_lut_to_channel_value(&crtc_state->gamma_lut, pixel->g, LUT_GREEN);
156 		pixel->b = apply_lut_to_channel_value(&crtc_state->gamma_lut, pixel->b, LUT_BLUE);
157 	}
158 }
159 
160 /**
161  * blend - blend the pixels from all planes and compute crc
162  * @wb: The writeback frame buffer metadata
163  * @crtc_state: The crtc state
164  * @crc32: The crc output of the final frame
165  * @output_buffer: A buffer of a row that will receive the result of the blend(s)
166  * @stage_buffer: The line with the pixels from plane being blend to the output
167  * @row_size: The size, in bytes, of a single row
168  *
169  * This function blends the pixels (Using the `pre_mul_alpha_blend`)
170  * from all planes, calculates the crc32 of the output from the former step,
171  * and, if necessary, convert and store the output to the writeback buffer.
172  */
173 static void blend(struct vkms_writeback_job *wb,
174 		  struct vkms_crtc_state *crtc_state,
175 		  u32 *crc32, struct line_buffer *stage_buffer,
176 		  struct line_buffer *output_buffer, size_t row_size)
177 {
178 	struct vkms_plane_state **plane = crtc_state->active_planes;
179 	u32 n_active_planes = crtc_state->num_active_planes;
180 	int y_pos;
181 
182 	const struct pixel_argb_u16 background_color = { .a = 0xffff };
183 
184 	size_t crtc_y_limit = crtc_state->base.crtc->mode.vdisplay;
185 
186 	for (size_t y = 0; y < crtc_y_limit; y++) {
187 		fill_background(&background_color, output_buffer);
188 
189 		/* The active planes are composed associatively in z-order. */
190 		for (size_t i = 0; i < n_active_planes; i++) {
191 			y_pos = get_y_pos(plane[i]->frame_info, y);
192 
193 			if (!check_limit(plane[i]->frame_info, y_pos))
194 				continue;
195 
196 			vkms_compose_row(stage_buffer, plane[i], y_pos);
197 			pre_mul_alpha_blend(plane[i]->frame_info, stage_buffer,
198 					    output_buffer);
199 		}
200 
201 		apply_lut(crtc_state, output_buffer);
202 
203 		*crc32 = crc32_le(*crc32, (void *)output_buffer->pixels, row_size);
204 
205 		if (wb)
206 			vkms_writeback_row(wb, output_buffer, y_pos);
207 	}
208 }
209 
210 static int check_format_funcs(struct vkms_crtc_state *crtc_state,
211 			      struct vkms_writeback_job *active_wb)
212 {
213 	struct vkms_plane_state **planes = crtc_state->active_planes;
214 	u32 n_active_planes = crtc_state->num_active_planes;
215 
216 	for (size_t i = 0; i < n_active_planes; i++)
217 		if (!planes[i]->pixel_read)
218 			return -1;
219 
220 	if (active_wb && !active_wb->pixel_write)
221 		return -1;
222 
223 	return 0;
224 }
225 
226 static int check_iosys_map(struct vkms_crtc_state *crtc_state)
227 {
228 	struct vkms_plane_state **plane_state = crtc_state->active_planes;
229 	u32 n_active_planes = crtc_state->num_active_planes;
230 
231 	for (size_t i = 0; i < n_active_planes; i++)
232 		if (iosys_map_is_null(&plane_state[i]->frame_info->map[0]))
233 			return -1;
234 
235 	return 0;
236 }
237 
238 static int compose_active_planes(struct vkms_writeback_job *active_wb,
239 				 struct vkms_crtc_state *crtc_state,
240 				 u32 *crc32)
241 {
242 	size_t line_width, pixel_size = sizeof(struct pixel_argb_u16);
243 	struct line_buffer output_buffer, stage_buffer;
244 	int ret = 0;
245 
246 	/*
247 	 * This check exists so we can call `crc32_le` for the entire line
248 	 * instead doing it for each channel of each pixel in case
249 	 * `struct `pixel_argb_u16` had any gap added by the compiler
250 	 * between the struct fields.
251 	 */
252 	static_assert(sizeof(struct pixel_argb_u16) == 8);
253 
254 	if (WARN_ON(check_iosys_map(crtc_state)))
255 		return -EINVAL;
256 
257 	if (WARN_ON(check_format_funcs(crtc_state, active_wb)))
258 		return -EINVAL;
259 
260 	line_width = crtc_state->base.crtc->mode.hdisplay;
261 	stage_buffer.n_pixels = line_width;
262 	output_buffer.n_pixels = line_width;
263 
264 	stage_buffer.pixels = kvmalloc(line_width * pixel_size, GFP_KERNEL);
265 	if (!stage_buffer.pixels) {
266 		DRM_ERROR("Cannot allocate memory for the output line buffer");
267 		return -ENOMEM;
268 	}
269 
270 	output_buffer.pixels = kvmalloc(line_width * pixel_size, GFP_KERNEL);
271 	if (!output_buffer.pixels) {
272 		DRM_ERROR("Cannot allocate memory for intermediate line buffer");
273 		ret = -ENOMEM;
274 		goto free_stage_buffer;
275 	}
276 
277 	blend(active_wb, crtc_state, crc32, &stage_buffer,
278 	      &output_buffer, line_width * pixel_size);
279 
280 	kvfree(output_buffer.pixels);
281 free_stage_buffer:
282 	kvfree(stage_buffer.pixels);
283 
284 	return ret;
285 }
286 
287 /**
288  * vkms_composer_worker - ordered work_struct to compute CRC
289  *
290  * @work: work_struct
291  *
292  * Work handler for composing and computing CRCs. work_struct scheduled in
293  * an ordered workqueue that's periodically scheduled to run by
294  * vkms_vblank_simulate() and flushed at vkms_atomic_commit_tail().
295  */
296 void vkms_composer_worker(struct work_struct *work)
297 {
298 	struct vkms_crtc_state *crtc_state = container_of(work,
299 						struct vkms_crtc_state,
300 						composer_work);
301 	struct drm_crtc *crtc = crtc_state->base.crtc;
302 	struct vkms_writeback_job *active_wb = crtc_state->active_writeback;
303 	struct vkms_output *out = drm_crtc_to_vkms_output(crtc);
304 	bool crc_pending, wb_pending;
305 	u64 frame_start, frame_end;
306 	u32 crc32 = 0;
307 	int ret;
308 
309 	spin_lock_irq(&out->composer_lock);
310 	frame_start = crtc_state->frame_start;
311 	frame_end = crtc_state->frame_end;
312 	crc_pending = crtc_state->crc_pending;
313 	wb_pending = crtc_state->wb_pending;
314 	crtc_state->frame_start = 0;
315 	crtc_state->frame_end = 0;
316 	crtc_state->crc_pending = false;
317 
318 	if (crtc->state->gamma_lut) {
319 		s64 max_lut_index_fp;
320 		s64 u16_max_fp = drm_int2fixp(0xffff);
321 
322 		crtc_state->gamma_lut.base = (struct drm_color_lut *)crtc->state->gamma_lut->data;
323 		crtc_state->gamma_lut.lut_length =
324 			crtc->state->gamma_lut->length / sizeof(struct drm_color_lut);
325 		max_lut_index_fp = drm_int2fixp(crtc_state->gamma_lut.lut_length  - 1);
326 		crtc_state->gamma_lut.channel_value2index_ratio = drm_fixp_div(max_lut_index_fp,
327 									       u16_max_fp);
328 
329 	} else {
330 		crtc_state->gamma_lut.base = NULL;
331 	}
332 
333 	spin_unlock_irq(&out->composer_lock);
334 
335 	/*
336 	 * We raced with the vblank hrtimer and previous work already computed
337 	 * the crc, nothing to do.
338 	 */
339 	if (!crc_pending)
340 		return;
341 
342 	if (wb_pending)
343 		ret = compose_active_planes(active_wb, crtc_state, &crc32);
344 	else
345 		ret = compose_active_planes(NULL, crtc_state, &crc32);
346 
347 	if (ret)
348 		return;
349 
350 	if (wb_pending) {
351 		drm_writeback_signal_completion(&out->wb_connector, 0);
352 		spin_lock_irq(&out->composer_lock);
353 		crtc_state->wb_pending = false;
354 		spin_unlock_irq(&out->composer_lock);
355 	}
356 
357 	/*
358 	 * The worker can fall behind the vblank hrtimer, make sure we catch up.
359 	 */
360 	while (frame_start <= frame_end)
361 		drm_crtc_add_crc_entry(crtc, true, frame_start++, &crc32);
362 }
363 
364 static const char * const pipe_crc_sources[] = {"auto"};
365 
366 const char *const *vkms_get_crc_sources(struct drm_crtc *crtc,
367 					size_t *count)
368 {
369 	*count = ARRAY_SIZE(pipe_crc_sources);
370 	return pipe_crc_sources;
371 }
372 
373 static int vkms_crc_parse_source(const char *src_name, bool *enabled)
374 {
375 	int ret = 0;
376 
377 	if (!src_name) {
378 		*enabled = false;
379 	} else if (strcmp(src_name, "auto") == 0) {
380 		*enabled = true;
381 	} else {
382 		*enabled = false;
383 		ret = -EINVAL;
384 	}
385 
386 	return ret;
387 }
388 
389 int vkms_verify_crc_source(struct drm_crtc *crtc, const char *src_name,
390 			   size_t *values_cnt)
391 {
392 	bool enabled;
393 
394 	if (vkms_crc_parse_source(src_name, &enabled) < 0) {
395 		DRM_DEBUG_DRIVER("unknown source %s\n", src_name);
396 		return -EINVAL;
397 	}
398 
399 	*values_cnt = 1;
400 
401 	return 0;
402 }
403 
404 void vkms_set_composer(struct vkms_output *out, bool enabled)
405 {
406 	bool old_enabled;
407 
408 	if (enabled)
409 		drm_crtc_vblank_get(&out->crtc);
410 
411 	mutex_lock(&out->enabled_lock);
412 	old_enabled = out->composer_enabled;
413 	out->composer_enabled = enabled;
414 
415 	/* the composition wasn't enabled, so unlock the lock to make sure the lock
416 	 * will be balanced even if we have a failed commit
417 	 */
418 	if (!out->composer_enabled)
419 		mutex_unlock(&out->enabled_lock);
420 
421 	if (old_enabled)
422 		drm_crtc_vblank_put(&out->crtc);
423 }
424 
425 int vkms_set_crc_source(struct drm_crtc *crtc, const char *src_name)
426 {
427 	struct vkms_output *out = drm_crtc_to_vkms_output(crtc);
428 	bool enabled = false;
429 	int ret = 0;
430 
431 	ret = vkms_crc_parse_source(src_name, &enabled);
432 
433 	vkms_set_composer(out, enabled);
434 
435 	return ret;
436 }
437