summaryrefslogtreecommitdiff
path: root/src/fb.c
blob: 250bd378e33403a6ea95ed2c76e9f70cb5c7b3d5 (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
#include <sys/types.h>
#include <sys/stat.h>
#include <fcntl.h>
#include <pthread.h>
#include <stdlib.h>
#include <string.h>
#include <stdint.h>
#include <inttypes.h>
#include <xf86drm.h>
#include <xf86drmMode.h>
#include <stddef.h>
#include <stdint.h>
#include <sys/ioctl.h>
#include <sys/mman.h>

#include "fb.h"
#include "util.h"

/* Copyright (C) 2016 Vito Caputo <vcaputo@pengaru.com> */


/* I've used a separate thread for page-flipping duties because the libdrm api
 * (and related kernel ioctl) for page flips doesn't appear to support queueing
 * multiple flip requests.  In this use case we aren't interactive and wish to
 * just accumulate rendered pages until we run out of spare pages, allowing the
 * renderer to get as far ahead of vsync as possible, and certainly never
 * blocked waiting for vsync unless there's no spare page available for drawing
 * into.
 *
 * In lieu of a queueing mechanism on the drm fd, we must submit the next page
 * once the currently submitted page is flipped to - it's at that moment we
 * won't get EBUSY from the ioctl any longer.  Without a dedicated thread
 * submitting flip requests and synchronously consuming their flip events,
 * we're liable to introduce latency in the page flip submission if implemented
 * in a more opportunistic manner whenever the fb api is entered from the
 * render loop.
 *
 * If the kernel simply let us queue multiple flip requests we could maintain
 * our submission queue entirely in the drm fd, and get available pages from
 * the drm event handler once our pool of pages is depleted.  The kernel on
 * vsync could check the fd to see if another flip is queued and there would be
 * the least latency possible in submitting the flips - the least likely to
 * miss a vsync.  This would also elide the need for synchronization in
 * userspace between the renderer and the flipper thread, since there would no
 * longer be a flipper thread.
 *
 * Let me know if you're aware of a better way with existing mainline drm!
 */


/* Most of fb_page_t is kept private, the public part is
 * just an fb_fragment_t describing the whole page.
 */
typedef struct _fb_page_t _fb_page_t;
struct _fb_page_t {
	_fb_page_t	*next;
	uint32_t	*mmap;
	size_t		mmap_size;
	uint32_t	drm_dumb_handle;
	uint32_t	drm_fb_id;
	fb_page_t	public_page;
};

typedef struct fb_t {
	pthread_t	thread;
	int		drm_fd;
	uint32_t	drm_crtc_id;

	_fb_page_t	*active_page;		/* page currently displayed */

	pthread_mutex_t	ready_mutex;
	pthread_cond_t	ready_cond;
	_fb_page_t	*ready_pages_head;	/* next pages to flip to */
	_fb_page_t	*ready_pages_tail;

	pthread_mutex_t	inactive_mutex;
	pthread_cond_t	inactive_cond;
	_fb_page_t	*inactive_pages;	/* finished pages available for (re)use */

	unsigned	put_pages_count;
} fb_t;

#ifndef container_of
#define container_of(_ptr, _type, _member) \
	(_type *)((void *)(_ptr) - offsetof(_type, _member))
#endif


/* Synchronize with the page flip by waiting for its event. */
static inline void fb_drm_flip_wait(fb_t *fb)
{
	drmEventContext	drm_ev_ctx = {
				.version = DRM_EVENT_CONTEXT_VERSION,
				.vblank_handler = NULL,
				.page_flip_handler = NULL
			};
	drmHandleEvent(fb->drm_fd, &drm_ev_ctx);
}


/* Consumes ready pages queued via fb_page_put(), submits them to drm to flip
 * on vsync.  Produces inactive pages from those replaced, making them
 * available to fb_page_get(). */
static void * fb_flipper_thread(void *_fb)
{
	fb_t	*fb = _fb;

	for (;;) {
		_fb_page_t	*next_active_page;
		/* wait for a flip req, submit the req page for flip on vsync, wait for it to flip before making the
		 * active page inactive/available, repeat.
		 */
		pthread_mutex_lock(&fb->ready_mutex);
		while (!fb->ready_pages_head)
			pthread_cond_wait(&fb->ready_cond, &fb->ready_mutex);

		next_active_page = fb->ready_pages_head;
		fb->ready_pages_head = next_active_page->next;
		if (!fb->ready_pages_head)
			fb->ready_pages_tail = NULL;
		pthread_mutex_unlock(&fb->ready_mutex);

		/* submit the next active page for page flip on vsync, and wait for it. */
		pexit_if(drmModePageFlip(fb->drm_fd, fb->drm_crtc_id, next_active_page->drm_fb_id, DRM_MODE_PAGE_FLIP_EVENT, NULL) < 0,
			"unable to flip page");
		fb_drm_flip_wait(fb);

		/* now that we're displaying a new page, make the previously active one inactive so rendering can reuse it */
		pthread_mutex_lock(&fb->inactive_mutex);
		fb->active_page->next = fb->inactive_pages;
		fb->inactive_pages = fb->active_page;
		pthread_cond_signal(&fb->inactive_cond);
		pthread_mutex_unlock(&fb->inactive_mutex);

		fb->active_page = next_active_page;
	}
}


/* creates a framebuffer page, which is a coupled drm_fb object and mmap region of memory */
static void fb_page_new(fb_t *fb, drmModeModeInfoPtr mode)
{
	_fb_page_t			*page;
	struct drm_mode_create_dumb	create_dumb = {
						.width = mode->hdisplay,
						.height = mode->vdisplay,
						.bpp = 32,
						.flags = 0, // unused,
					};
	struct drm_mode_map_dumb	map_dumb = {
						.pad = 0, // unused
					};
	uint32_t			*map, fb_id;

	page = calloc(1, sizeof(_fb_page_t));

	pexit_if(ioctl(fb->drm_fd, DRM_IOCTL_MODE_CREATE_DUMB, &create_dumb) < 0,
		"unable to create dumb buffer");

	map_dumb.handle = create_dumb.handle;
	pexit_if(ioctl(fb->drm_fd, DRM_IOCTL_MODE_MAP_DUMB, &map_dumb) < 0,
		"unable to prepare dumb buffer for mmap");
	pexit_if(!(map = mmap(NULL, create_dumb.size, PROT_READ|PROT_WRITE, MAP_SHARED, fb->drm_fd, map_dumb.offset)),
		"unable to mmap dumb buffer");
	pexit_if(drmModeAddFB(fb->drm_fd, mode->hdisplay, mode->vdisplay, 24, 32, create_dumb.pitch, create_dumb.handle, &fb_id) < 0,
		"unable to add dumb buffer");

	page->mmap = map;
	page->mmap_size = create_dumb.size;
	page->drm_dumb_handle = map_dumb.handle;
	page->drm_fb_id = fb_id;

	page->public_page.fragment.buf = map;
	page->public_page.fragment.width = mode->hdisplay;
	page->public_page.fragment.frame_width = mode->hdisplay;
	page->public_page.fragment.height = mode->vdisplay;
	page->public_page.fragment.frame_height = mode->vdisplay;
	page->public_page.fragment.stride = create_dumb.pitch - (mode->hdisplay * 4);

	page->next = fb->inactive_pages;
	fb->inactive_pages = page;
}


static void _fb_page_free(fb_t *fb, _fb_page_t *page)
{
	struct drm_mode_destroy_dumb	destroy_dumb = {
						.handle = page->drm_dumb_handle,
					};

	drmModeRmFB(fb->drm_fd, page->drm_fb_id);
	munmap(page->mmap, page->mmap_size);
	ioctl(fb->drm_fd, DRM_IOCTL_MODE_DESTROY_DUMB, &destroy_dumb); // XXX: errors?
	free(page);
}


/* get the next inactive page from the fb, waiting if necessary. */
static inline _fb_page_t * _fb_page_get(fb_t *fb)
{
	_fb_page_t	*page;

	/* As long as n_pages is >= 3 this won't block unless we're submitting
	 * pages faster than vhz.
	 */
	pthread_mutex_lock(&fb->inactive_mutex);
	while (!(page = fb->inactive_pages))
		pthread_cond_wait(&fb->inactive_cond, &fb->inactive_mutex);
	fb->inactive_pages = page->next;
	pthread_mutex_unlock(&fb->inactive_mutex);

	page->next = NULL;

	return page;
}


/* public interface */
fb_page_t * fb_page_get(fb_t *fb)
{
	return &(_fb_page_get(fb)->public_page);
}


/* put a page into the fb, queueing for display */
static inline void _fb_page_put(fb_t *fb, _fb_page_t *page)
{
	pthread_mutex_lock(&fb->ready_mutex);
	if (fb->ready_pages_tail)
		fb->ready_pages_tail->next = page;
	else
		fb->ready_pages_head = page;

	fb->ready_pages_tail = page;
	pthread_cond_signal(&fb->ready_cond);
	pthread_mutex_unlock(&fb->ready_mutex);
}


/* public interface */

/* put a page into the fb, queueing for display */
void fb_page_put(fb_t *fb, fb_page_t *page)
{
	fb->put_pages_count++;

	_fb_page_put(fb, container_of(page, _fb_page_t, public_page));
}


/* get (and reset) the current count of put pages */
void fb_get_put_pages_count(fb_t *fb, unsigned *count)
{
	*count = fb->put_pages_count;
	fb->put_pages_count = 0;
}


/* free the fb and associated resources */
void fb_free(fb_t *fb)
{
	if (fb->active_page) {
		pthread_cancel(fb->thread);
		pthread_join(fb->thread, NULL);
	}

	/* TODO: free all the pages */

	pthread_mutex_destroy(&fb->ready_mutex);
	pthread_cond_destroy(&fb->ready_cond);
	pthread_mutex_destroy(&fb->inactive_mutex);
	pthread_cond_destroy(&fb->inactive_cond);

	free(fb);
}


/* create a new fb instance */
fb_t * fb_new(int drm_fd, uint32_t crtc_id, uint32_t *connectors, int n_connectors, drmModeModeInfoPtr mode, int n_pages)
{
	int		i;
	_fb_page_t	*page;
	fb_t		*fb;

	/* XXX: page-flipping is the only supported rendering model, requiring 2+ pages. */
	if (n_pages < 2)
		return NULL;

	fb = calloc(1, sizeof(fb_t));
	if (!fb)
		return NULL;

	fb->drm_fd = drm_fd;
	fb->drm_crtc_id = crtc_id;

	for (i = 0; i < n_pages; i++)
		fb_page_new(fb, mode);

	pthread_mutex_init(&fb->ready_mutex, NULL);
	pthread_cond_init(&fb->ready_cond, NULL);
	pthread_mutex_init(&fb->inactive_mutex, NULL);
	pthread_cond_init(&fb->inactive_cond, NULL);

	page = _fb_page_get(fb);

	/* set the video mode, pinning this page, set it as the active page. */
	if (drmModeSetCrtc(drm_fd, crtc_id, page->drm_fb_id, 0, 0, connectors, n_connectors, mode) < 0) {
		_fb_page_free(fb, page);
		fb_free(fb);
		return NULL;
	}

	fb->active_page = page;

	/* start up the page flipper thread */
	pthread_create(&fb->thread, NULL, fb_flipper_thread, fb);

	return fb;
}


/* helpers for fragmenting incrementally */
int fb_fragment_divide_single(const fb_fragment_t *fragment, unsigned n_fragments, unsigned num, fb_fragment_t *res_fragment)
{
	unsigned	slice = fragment->height / n_fragments;
	unsigned	yoff = slice * num;
	unsigned	pitch;

	if (yoff >= fragment->height)
		return 0;

	pitch = (fragment->width * 4) + fragment->stride;

	res_fragment->buf = ((void *)fragment->buf) + yoff * pitch;
	res_fragment->x = fragment->x;
	res_fragment->y = yoff;
	res_fragment->width = fragment->width;
	res_fragment->height = MIN(fragment->height - yoff, slice);
	res_fragment->frame_width = fragment->frame_width;
	res_fragment->frame_height = fragment->frame_height;
	res_fragment->stride = fragment->stride;

	return 1;
}


int fb_fragment_tile_single(const fb_fragment_t *fragment, unsigned tile_size, unsigned num, fb_fragment_t *res_fragment)
{
	unsigned	w = fragment->width / tile_size, h = fragment->height / tile_size;
	unsigned	pitch = (fragment->width * 4) + fragment->stride;
	unsigned	x, y, xoff, yoff;

	if (w * tile_size < fragment->width)
		w++;

	if (h * tile_size < fragment->height)
		h++;

	y = num / w;
	if (y >= h)
		return 0;

	x = num - (y * w);

	xoff = x * tile_size;
	yoff = y * tile_size;

	res_fragment->buf = (void *)fragment->buf + (yoff * pitch) + (xoff * 4);
	res_fragment->x = fragment->x + xoff;
	res_fragment->y = fragment->y + yoff;
	res_fragment->width = MIN(fragment->width - xoff, tile_size);
	res_fragment->height = MIN(fragment->height - yoff, tile_size);
	res_fragment->frame_width = fragment->frame_width;
	res_fragment->frame_height = fragment->frame_height;
	res_fragment->stride = fragment->stride + ((fragment->width - res_fragment->width) * 4);

	return 1;
}
© All Rights Reserved