Bump version: 5.44 → 5.45

Clang reported "warning: variable 'slc' is incremented both in the loop
header and in the loop body".

.bumpversion.cfg

@@ -1,7 +1,7 @@
 [bumpversion]
 commit = True
 tag = True
-current_version = 5.40
+current_version = 5.45
 parse = (?P<major>\d+)\.(?P<minor>\d+)
 serialize = 
 	{major}.{minor}

.gitignore

@@ -4,6 +4,7 @@
 /pkg/arch/src/
 /src/build/
 /python/build/
+/python/ustreamer.egg-info/
 /janus/build/
 /ustreamer
 /ustreamer-dump

README.md

@@ -88,7 +88,7 @@ Without arguments, ```ustreamer``` will try to open ```/dev/video0``` with 640x4
 :exclamation: Please note that since µStreamer v2.0 cross-domain requests were disabled by default for [security reasons](https://developer.mozilla.org/en-US/docs/Web/HTTP/CORS). To enable the old behavior, use the option `--allow-origin=\*`.
 
 The recommended way of running µStreamer with [Auvidea B101](https://www.raspberrypi.org/forums/viewtopic.php?f=38&t=120702&start=400#p1339178) on Raspberry Pi:
-```bash
+```
 $ ./ustreamer \
     --format=uyvy \ # Device input format
     --encoder=m2m-image \ # Hardware encoding on V4L2 M2M driver
@@ -115,7 +115,7 @@ dtoverlay=tc358743
 
 Check size of CMA:
 
-```bash
+```
 $ dmesg | grep cma-reserved
 [    0.000000] Memory: 7700524K/8244224K available (11772K kernel code, 1278K rwdata, 4320K rodata, 4096K init, 1077K bss, 281556K reserved, 262144K cma-reserved)
 ```
@@ -131,14 +131,14 @@ Save changes and reboot.
 ## Launch
 Start container:
 
-```bash
+```
 $ docker run --device /dev/video0:/dev/video0 -e EDID=1 -p 8080:8080 pikvm/ustreamer:latest
 ```
 
 Then access the web interface at port 8080 (e.g. http://raspberrypi.local:8080).
 
 ## Custom config
-```bash
+```
 $ docker run --rm pikvm/ustreamer:latest \
     --format=uyvy \
     --workers=3 \
@@ -153,8 +153,15 @@ Add `-e EDID=1` to set HDMI EDID before starting ustreamer. Use together with `-
 -----
 # Raspberry Pi Camera Example
 
+Example usage for the Raspberry Pi v3 camera (required `libcamerify` which is located in `libcamera-tools` on Raspbian):
+```
+$ sudo modprobe bcm2835-v4l2
+$ libcamerify ./ustreamer --host :: -e m2m-image
+```
+
+
 Example usage for the Raspberry Pi v1 camera:
-```bash
+```
 $ sudo modprobe bcm2835-v4l2
 $ ./ustreamer --host :: -m jpeg --device-timeout=5 --buffers=3 -r 2592x1944
 ```
@@ -163,7 +170,7 @@ $ ./ustreamer --host :: -m jpeg --device-timeout=5 --buffers=3 -r 2592x1944
 
 :exclamation: If you get a poor framerate, it could be that the camera is switched to photo mode, which produces a low framerate (but a higher quality picture). This is because `bcm2835-v4l2` switches to photo mode at resolutions higher than `1280x720`. To work around this, pass the `max_video_width` and `max_video_height` module parameters like so:
 
-```bash
+```
 $ modprobe bcm2835-v4l2 max_video_width=2592 max_video_height=1944
 ```
 

docs/ssl/README.md

@@ -0,0 +1,50 @@
+# Adding SSL
+These days, browsers are not happy if you have HTTP content on an HTTPS page.
+The browser will not show an HTTP stream on a page if the parent page is from a site which is using HTTPS.
+
+The files in this folder configure an Nginx proxy in front of the µStreamer stream.
+Using certbot, an SSL cert is created from Let's Encrypt and installed.
+These scripts can be modified to add SSL to just about any HTTP server.
+
+The scripts are not fire and forget.
+They will require some pre-configuration and are interactive (you'll be asked questions while they're running).
+They have been tested using the following setup.
+1. A Raspberry Pi 4
+1. µStreamer set up and running as a service
+    1. Internally on port 8080
+    1. Public port will be 5101
+1. Verizon home Wi-Fi router
+1. Domain registration from GoDaddy
+
+## The Script
+Below is an overview of the steps performed by `ssl-config.sh` (for Raspberry OS):
+1. Install snapd - certbot uses this for installation
+1. Install certbot
+1. Get a free cert from Let's Encrypt using certbot
+1. Install nginx
+1. Configures nginx to proxy for µStreamer
+
+## Steps
+1. Create a public DNS entry.
+    1. Pointing to the Pi itself or the public IP of the router behind which the Pi sits.
+    1. This would be managed in the domain registrar, such as GoDaddy.
+    1. Use a subdomain, such as `webcam.domain.com`
+1. Port Forwarding
+    1. If using a Wi-Fi router, create a port forwarding rule which passes traffic from port 80 to the Pi. This is needed for certbot to ensure your DNS entry reaches the Pi, even if your final port will be something else.
+    1. Create a second rule for your final setup. For example, forward traffic from the router on port 5101 to the Pi's IP port 8080.
+1. Update the ustreamer-proxy file in this folder
+    1. Replace `your.domain.com` with a fully qualified domain, it's three places in the proxy file.
+    1. Modify the line `listen 5101 ssl` port if needed. This is the public port, not the port on which the µStreamer service is running
+    1. Modify `proxy_pass http://127.0.0.1:8080;` with the working address of the internal µStreamer service.
+1. Run the script
+    1. Stand buy, certbot asks some basic questions, such as email, domain, agree to terms, etc.
+    1. `bash ssl-config.sh`
+1. Test your URL!
+
+## Down the Road
+Two important points to keep in mind for the future:
+1. Dynamic IP - Most routers do not have a static IP address on the WAN side. So, if you reboot your router or if your internet provider gives you a new IP, you'll have to update the DNS entry.
+    1. Many routers have some sort of dynamic DNS feature. This would automatically update the DNS entry for you. That functionality is outside the scope of this document.
+1. SSL Renewals - certbot automatically creates a task to renew the SSL cert before it expires. Assuming the Pi is running all the time, this shouldn't be an issue.
+
+## Enjoy!

docs/ssl/ssl-config.sh

@@ -0,0 +1,20 @@
+#!/bin/sh
+
+echo -e "\e[32mInstalling snapd...\e[0m"
+sudo apt install snapd -y
+sudo snap install core
+
+
+echo -e "\e[32mInstalling certbot, don't leave, it's going to ask questions...\e[0m"
+sudo snap install --classic certbot
+sudo ln -s /snap/bin/certbot /usr/bin/certbot
+sudo certbot certonly --standalone
+sudo certbot renew --dry-run
+
+
+echo -e "\e[32mInstalling nginx...\e[0m"
+sudo apt-get install nginx -y
+sudo cp ustreamer-proxy /etc/nginx/sites-available/ustreamer-proxy
+sudo ln -s /etc/nginx/sites-available/ustreamer-proxy /etc/nginx/sites-enabled/
+sudo nginx -t
+sudo systemctl reload nginx

docs/ssl/ustreamer-proxy

@@ -0,0 +1,13 @@
+server {
+    listen 5101 ssl;
+    server_name your.domain.com;
+
+    ssl_certificate /etc/letsencrypt/live/your.domain.com/fullchain.pem;
+    ssl_certificate_key /etc/letsencrypt/live/your.domain.com/privkey.pem;
+
+    location / {
+        proxy_pass http://127.0.0.1:8080;  # Change this to the uStreamer server address
+        proxy_set_header Host $host;
+        proxy_set_header X-Real-IP $remote_addr;
+    }
+}

janus/src/client.c

@@ -116,7 +116,9 @@ static void *_common_thread(void *v_client, bool video) {
 					packet.extensions.max_delay = 0;
 				} else {
 					packet.extensions.min_delay = 0;
-					packet.extensions.max_delay = 1000;
+					// 10s - Chromium/WebRTC default
+					// 3s - Firefox default
+					packet.extensions.max_delay = 300; // == 3s, i.e. 10ms granularity
 				}*/
 
 				client->gw->relay_rtp(client->session, &packet);

janus/src/plugin.c

@@ -56,8 +56,8 @@
 #include "config.h"
 
 
-static us_config_s	*_g_config = NULL;
-const useconds_t	_g_watchers_polling = 100000;
+static us_config_s		*_g_config = NULL;
+static const useconds_t	_g_watchers_polling = 100000;
 
 static us_janus_client_s	*_g_clients = NULL;
 static janus_callbacks		*_g_gw = NULL;
@@ -107,7 +107,8 @@ static void *_video_rtp_thread(UNUSED void *arg) {
 		us_frame_s *frame;
 		if (us_queue_get(_g_video_queue, (void **)&frame, 0.1) == 0) {
 			_LOCK_VIDEO;
-			us_rtpv_wrap(_g_rtpv, frame);
+			const bool zero_playout_delay = (frame->gop == 0);
+			us_rtpv_wrap(_g_rtpv, frame, zero_playout_delay);
 			_UNLOCK_VIDEO;
 			us_frame_destroy(frame);
 		}

janus/src/rtpv.c

@@ -71,13 +71,13 @@ char *us_rtpv_make_sdp(us_rtpv_s *rtpv) {
 
 #define _PRE 3 // Annex B prefix length
 
-void us_rtpv_wrap(us_rtpv_s *rtpv, const us_frame_s *frame) {
+void us_rtpv_wrap(us_rtpv_s *rtpv, const us_frame_s *frame, bool zero_playout_delay) {
 	// There is a complicated logic here but everything works as it should:
 	//   - https://github.com/pikvm/ustreamer/issues/115#issuecomment-893071775
 
 	assert(frame->format == V4L2_PIX_FMT_H264);
 
-	rtpv->rtp->zero_playout_delay = (frame->gop == 0);
+	rtpv->rtp->zero_playout_delay = zero_playout_delay;
 
 	const uint32_t pts = us_get_now_monotonic_u64() * 9 / 100; // PTS units are in 90 kHz
 	ssize_t last_offset = -_PRE;

janus/src/rtpv.h

@@ -47,4 +47,4 @@ us_rtpv_s *us_rtpv_init(us_rtp_callback_f callback);
 void us_rtpv_destroy(us_rtpv_s *rtpv);
 
 char *us_rtpv_make_sdp(us_rtpv_s *rtpv);
-void us_rtpv_wrap(us_rtpv_s *rtpv, const us_frame_s *frame);
+void us_rtpv_wrap(us_rtpv_s *rtpv, const us_frame_s *frame, bool zero_playout_delay);

linters/cppcheck.h

@@ -2,5 +2,3 @@
 #define WITH_GPIO
 #define JANUS_PLUGIN_INIT(...) { __VA_ARGS__ }
 #define EVTHREAD_USE_PTHREADS_IMPLEMENTED 1
-#define CLOCK_MONOTONIC_RAW 1
-#define CLOCK_MONOTONIC_FAST 1

linters/tox.ini

@@ -9,7 +9,6 @@ changedir = /src
 [testenv:cppcheck]
 allowlist_externals = cppcheck
 commands = cppcheck \
-			-j4 \
 			--force \
 			--std=c17 \
 			--error-exitcode=1 \

man/ustreamer-dump.1

@@ -1,6 +1,6 @@
 .\" Manpage for ustreamer-dump.
 .\" Open an issue or pull request to https://github.com/pikvm/ustreamer to correct errors or typos
-.TH USTREAMER-DUMP 1 "version 5.40" "January 2021"
+.TH USTREAMER-DUMP 1 "version 5.45" "January 2021"
 
 .SH NAME
 ustreamer-dump \- Dump uStreamer's memory sink to file

man/ustreamer.1

@@ -1,6 +1,6 @@
 .\" Manpage for ustreamer.
 .\" Open an issue or pull request to https://github.com/pikvm/ustreamer to correct errors or typos
-.TH USTREAMER 1 "version 5.40" "November 2020"
+.TH USTREAMER 1 "version 5.45" "November 2020"
 
 .SH NAME
 ustreamer \- stream MJPEG video from any V4L2 device to the network
@@ -248,7 +248,7 @@ Timeout for lock. Default: 1.
 H264 bitrate in Kbps. Default: 5000.
 .TP
 .BR \-\-h264\-gop\ \fIN
-Intarval between keyframes. Default: 30.
+Interval between keyframes. Default: 30.
 .TP
 .BR \-\-h264\-m2m\-device\ \fI/dev/path
 Path to V4L2 mem-to-mem encoder device. Default: auto-select.

pkg/arch/PKGBUILD

@@ -3,7 +3,7 @@
 
 
 pkgname=ustreamer
-pkgver=5.40
+pkgver=5.45
 pkgrel=1
 pkgdesc="Lightweight and fast MJPEG-HTTP streamer"
 url="https://github.com/pikvm/ustreamer"

pkg/openwrt/Makefile

@@ -6,7 +6,7 @@
 include $(TOPDIR)/rules.mk
 
 PKG_NAME:=ustreamer
-PKG_VERSION:=5.40
+PKG_VERSION:=5.45
 PKG_RELEASE:=1
 PKG_MAINTAINER:=Maxim Devaev <mdevaev@gmail.com>
 

python/Makefile

@@ -17,4 +17,4 @@ install:
 
 
 clean:
-	rm -rf build
+	rm -rf build ustreamer.egg-info

python/setup.py

@@ -17,7 +17,7 @@ def _find_sources(suffix: str) -> list[str]:
 if __name__ == "__main__":
     setup(
         name="ustreamer",
-        version="5.40",
+        version="5.45",
         description="uStreamer tools",
         author="Maxim Devaev",
         author_email="mdevaev@gmail.com",

src/libs/const.h

@@ -23,7 +23,7 @@
 #pragma once
 
 #define US_VERSION_MAJOR 5
-#define US_VERSION_MINOR 40
+#define US_VERSION_MINOR 45
 
 #define US_MAKE_VERSION2(_major, _minor) #_major "." #_minor
 #define US_MAKE_VERSION1(_major, _minor) US_MAKE_VERSION2(_major, _minor)

src/libs/frame.c

@@ -75,6 +75,7 @@ unsigned us_frame_get_padding(const us_frame_s *frame) {
 		case V4L2_PIX_FMT_YUYV:
 		case V4L2_PIX_FMT_UYVY:
 		case V4L2_PIX_FMT_RGB565: bytes_per_pixel = 2; break;
+		case V4L2_PIX_FMT_BGR24:
 		case V4L2_PIX_FMT_RGB24: bytes_per_pixel = 3; break;
 		// case V4L2_PIX_FMT_H264:
 		case V4L2_PIX_FMT_MJPEG:

src/libs/process.h

@@ -98,7 +98,7 @@ INLINE int us_process_track_parent_death(void) {
 #ifdef WITH_SETPROCTITLE
 #	pragma GCC diagnostic ignored "-Wunused-parameter"
 #	pragma GCC diagnostic push
-INLINE void us_process_set_name_prefix(int argc, char *argv[], const char *prefix) {
+INLINE void us_process_set_name_prefix(int argc, char *argv[], const char *prefix) { // cppcheck-suppress constParameter
 #	pragma GCC diagnostic pop
 
 	char *cmdline = NULL;

src/libs/tools.h

@@ -117,29 +117,19 @@ INLINE void us_get_now(clockid_t clk_id, time_t *sec, long *msec) {
 	}
 }
 
-#if defined(CLOCK_MONOTONIC_RAW)
-#	define _X_CLOCK_MONOTONIC CLOCK_MONOTONIC_RAW
-#elif defined(CLOCK_MONOTONIC_FAST)
-#	define _X_CLOCK_MONOTONIC CLOCK_MONOTONIC_FAST
-#else
-#	define _X_CLOCK_MONOTONIC CLOCK_MONOTONIC
-#endif
-
 INLINE long double us_get_now_monotonic(void) {
 	time_t sec;
 	long msec;
-	us_get_now(_X_CLOCK_MONOTONIC, &sec, &msec);
+	us_get_now(CLOCK_MONOTONIC, &sec, &msec);
 	return (long double)sec + ((long double)msec) / 1000;
 }
 
 INLINE uint64_t us_get_now_monotonic_u64(void) {
 	struct timespec ts;
-	assert(!clock_gettime(_X_CLOCK_MONOTONIC, &ts));
+	assert(!clock_gettime(CLOCK_MONOTONIC, &ts));
 	return (uint64_t)(ts.tv_nsec / 1000) + (uint64_t)ts.tv_sec * 1000000;
 }
 
-#undef _X_CLOCK_MONOTONIC
-
 INLINE uint64_t us_get_now_id(void) {
 	const uint64_t now = us_get_now_monotonic_u64();
 	return (uint64_t)us_triple_u32(now) | ((uint64_t)us_triple_u32(now + 12345) << 32);

src/ustreamer/device.c

@@ -41,6 +41,7 @@ static const struct {
 	{"UYVY",	V4L2_PIX_FMT_UYVY},
 	{"RGB565",	V4L2_PIX_FMT_RGB565},
 	{"RGB24",	V4L2_PIX_FMT_RGB24},
+	{"BGR24",	V4L2_PIX_FMT_BGR24},
 	{"MJPEG",	V4L2_PIX_FMT_MJPEG},
 	{"JPEG",	V4L2_PIX_FMT_JPEG},
 };
@@ -54,6 +55,8 @@ static const struct {
 };
 
 
+static void _v4l2_buffer_copy(const struct v4l2_buffer *src, struct v4l2_buffer *dest);
+static bool _device_is_buffer_valid(us_device_s *dev, const struct v4l2_buffer *buf, const uint8_t *data);
 static int _device_open_check_cap(us_device_s *dev);
 static int _device_open_dv_timings(us_device_s *dev);
 static int _device_apply_dv_timings(us_device_s *dev);
@@ -71,7 +74,7 @@ static int _device_query_control(
 	us_device_s *dev, struct v4l2_queryctrl *query,
 	const char *name, unsigned cid, bool quiet);
 static void _device_set_control(
-	us_device_s *dev, struct v4l2_queryctrl *query,
+	us_device_s *dev, const struct v4l2_queryctrl *query,
 	const char *name, unsigned cid, int value, bool quiet);
 
 static const char *_format_to_string_nullable(unsigned format);
@@ -82,6 +85,7 @@ static const char *_io_method_to_string_supported(enum v4l2_memory io_method);
 
 #define _RUN(x_next)	dev->run->x_next
 #define _D_XIOCTL(...)	us_xioctl(_RUN(fd), __VA_ARGS__)
+#define _D_IS_MPLANE	(_RUN(capture_type) == V4L2_BUF_TYPE_VIDEO_CAPTURE_MPLANE)
 
 
 us_device_s *us_device_init(void) {
@@ -194,6 +198,10 @@ void us_device_close(us_device_s *dev) {
 				US_DELETE(HW(raw.data), free);
 			}
 
+			if (_D_IS_MPLANE) {
+				free(HW(buf.m.planes));
+			}
+
 #			undef HW
 		}
 		_RUN(n_bufs) = 0;
@@ -217,7 +225,7 @@ int us_device_export_to_dma(us_device_s *dev) {
 
 	for (unsigned index = 0; index < _RUN(n_bufs); ++index) {
 		struct v4l2_exportbuffer exp = {0};
-		exp.type = V4L2_BUF_TYPE_VIDEO_CAPTURE;
+		exp.type = _RUN(capture_type);
 		exp.index = index;
 
 		US_LOG_DEBUG("Exporting device buffer=%u to DMA ...", index);
@@ -244,7 +252,7 @@ int us_device_export_to_dma(us_device_s *dev) {
 
 int us_device_switch_capturing(us_device_s *dev, bool enable) {
 	if (enable != _RUN(capturing)) {
-		enum v4l2_buf_type type = V4L2_BUF_TYPE_VIDEO_CAPTURE;
+		enum v4l2_buf_type type = _RUN(capture_type);
 
 		US_LOG_DEBUG("%s device capturing ...", (enable ? "Starting" : "Stopping"));
 		if (_D_XIOCTL((enable ? VIDIOC_STREAMON : VIDIOC_STREAMOFF), &type) < 0) {
@@ -310,46 +318,90 @@ int us_device_grab_buffer(us_device_s *dev, us_hw_buffer_s **hw) {
 	*hw = NULL;
 
 	struct v4l2_buffer buf = {0};
-	buf.type = V4L2_BUF_TYPE_VIDEO_CAPTURE;
-	buf.memory = dev->io_method;
+	struct v4l2_plane buf_planes[VIDEO_MAX_PLANES] = {0};
+	if (_D_IS_MPLANE) {
+		// Just for _v4l2_buffer_copy(), buf.length is not needed here
+		buf.m.planes = buf_planes;
+	}
+
+	bool buf_got = false;
+	unsigned skipped = 0;
+	bool broken = false;
 
 	US_LOG_DEBUG("Grabbing device buffer ...");
-	if (_D_XIOCTL(VIDIOC_DQBUF, &buf) < 0) {
-		US_LOG_PERROR("Can't grab device buffer");
-		return -1;
-	}
 
-	US_LOG_DEBUG("Grabbed new frame: buffer=%u, bytesused=%u", buf.index, buf.bytesused);
+	do {
+		struct v4l2_buffer new = {0};
+		struct v4l2_plane new_planes[VIDEO_MAX_PLANES] = {0};
+		new.type = _RUN(capture_type);
+		new.memory = dev->io_method;
+		if (_D_IS_MPLANE) {
+			new.length = VIDEO_MAX_PLANES;
+			new.m.planes = new_planes;
+		}
+		
+		const bool new_got = (_D_XIOCTL(VIDIOC_DQBUF, &new) >= 0);
 
-	if (buf.index >= _RUN(n_bufs)) {
-		US_LOG_ERROR("V4L2 error: grabbed invalid device buffer=%u, n_bufs=%u", buf.index, _RUN(n_bufs));
-		return -1;
-	}
+		if (new_got) {
+			if (new.index >= _RUN(n_bufs)) {
+				US_LOG_ERROR("V4L2 error: grabbed invalid device buffer=%u, n_bufs=%u", new.index, _RUN(n_bufs));
+				return -1;
+			}
 
-	// Workaround for broken, corrupted frames:
-	// Under low light conditions corrupted frames may get captured.
-	// The good thing is such frames are quite small compared to the regular frames.
-	// For example a VGA (640x480) webcam frame is normally >= 8kByte large,
-	// corrupted frames are smaller.
-	if (buf.bytesused < dev->min_frame_size) {
-		US_LOG_DEBUG("Dropped too small frame, assuming it was broken: buffer=%u, bytesused=%u",
-			buf.index, buf.bytesused);
-		US_LOG_DEBUG("Releasing device buffer=%u (broken frame) ...", buf.index);
-		if (_D_XIOCTL(VIDIOC_QBUF, &buf) < 0) {
-			US_LOG_PERROR("Can't release device buffer=%u (broken frame)", buf.index);
+#			define GRABBED(x_buf) _RUN(hw_bufs)[x_buf.index].grabbed
+#			define FRAME_DATA(x_buf) _RUN(hw_bufs)[x_buf.index].raw.data
+
+			if (GRABBED(new)) {
+				US_LOG_ERROR("V4L2 error: grabbed device buffer=%u is already used", new.index);
+				return -1;
+			}
+			GRABBED(new) = true;
+
+			if (_D_IS_MPLANE) {
+				new.bytesused = new.m.planes[0].bytesused;
+			}
+
+			broken = !_device_is_buffer_valid(dev, &new, FRAME_DATA(new));
+			if (broken) {
+				US_LOG_DEBUG("Releasing device buffer=%u (broken frame) ...", new.index);
+				if (_D_XIOCTL(VIDIOC_QBUF, &new) < 0) {
+					US_LOG_PERROR("Can't release device buffer=%u (broken frame)", new.index);
+					return -1;
+				}
+				GRABBED(new) = false;
+				continue;
+			}
+
+			if (buf_got) {
+				if (_D_XIOCTL(VIDIOC_QBUF, &buf) < 0) {
+					US_LOG_PERROR("Can't release device buffer=%u (skipped frame)", buf.index);
+					return -1;
+				}
+				GRABBED(buf) = false;
+				++skipped;
+				// buf_got = false;
+			}
+
+#			undef GRABBED
+#			undef FRAME_DATA
+
+			_v4l2_buffer_copy(&new, &buf);
+			buf_got = true;
+
+		} else {
+			if (errno == EAGAIN) {
+				if (buf_got) {
+					break; // Process any latest valid frame
+				} else if (broken) {
+					return -2; // If we have only broken frames on this capture session
+				}
+			}
+			US_LOG_PERROR("Can't grab device buffer");
 			return -1;
 		}
-		return -2;
-	}
+	} while (true);
 
 #	define HW(x_next) _RUN(hw_bufs)[buf.index].x_next
-
-	if (HW(grabbed)) {
-		US_LOG_ERROR("V4L2 error: grabbed device buffer=%u is already used", buf.index);
-		return -1;
-	}
-	HW(grabbed) = true;
-
 	HW(raw.dma_fd) = HW(dma_fd);
 	HW(raw.used) = buf.bytesused;
 	HW(raw.width) = _RUN(width);
@@ -357,10 +409,12 @@ int us_device_grab_buffer(us_device_s *dev, us_hw_buffer_s **hw) {
 	HW(raw.format) = _RUN(format);
 	HW(raw.stride) = _RUN(stride);
 	HW(raw.online) = true;
-	memcpy(&HW(buf), &buf, sizeof(struct v4l2_buffer));
-	HW(raw.grab_ts) = us_get_now_monotonic();
-
+	_v4l2_buffer_copy(&buf, &HW(buf));
+	HW(raw.grab_ts) = (long double)((buf.timestamp.tv_sec * (uint64_t)1000) + (buf.timestamp.tv_usec / 1000)) / 1000;
+	US_LOG_DEBUG("Grabbed new frame: buffer=%u, bytesused=%u, grab_ts=%.3Lf, latency=%.3Lf, skipped=%u",
+		buf.index, buf.bytesused, HW(raw.grab_ts), us_get_now_monotonic() - HW(raw.grab_ts), skipped);
 #	undef HW
+
 	*hw = &_RUN(hw_bufs[buf.index]);
 	return buf.index;
 }
@@ -396,6 +450,55 @@ int us_device_consume_event(us_device_s *dev) {
 	return 0;
 }
 
+static void _v4l2_buffer_copy(const struct v4l2_buffer *src, struct v4l2_buffer *dest) {
+	struct v4l2_plane *dest_planes = dest->m.planes;
+	memcpy(dest, src, sizeof(struct v4l2_buffer));
+	if (src->type == V4L2_BUF_TYPE_VIDEO_CAPTURE_MPLANE) {
+		assert(dest_planes);
+		dest->m.planes = dest_planes;
+		memcpy(dest->m.planes, src->m.planes, sizeof(struct v4l2_plane) * VIDEO_MAX_PLANES);
+	}
+}
+
+bool _device_is_buffer_valid(us_device_s *dev, const struct v4l2_buffer *buf, const uint8_t *data) {
+	// Workaround for broken, corrupted frames:
+	// Under low light conditions corrupted frames may get captured.
+	// The good thing is such frames are quite small compared to the regular frames.
+	// For example a VGA (640x480) webcam frame is normally >= 8kByte large,
+	// corrupted frames are smaller.
+	if (buf->bytesused < dev->min_frame_size) {
+		US_LOG_DEBUG("Dropped too small frame, assuming it was broken: buffer=%u, bytesused=%u",
+			buf->index, buf->bytesused);
+		return false;
+	}
+
+	// Workaround for truncated JPEG frames:
+	// Some inexpensive CCTV-style USB webcams such as the ELP-USB100W03M send
+	// large amounts of these frames when using MJPEG streams. Checks that the
+	// buffer ends with either the JPEG end of image marker (0xFFD9), the last
+	// marker byte plus a padding byte (0xD900), or just padding bytes (0x0000)
+	// A more sophisticated method would scan for the end of image marker, but
+	// that takes precious CPU cycles and this should be good enough for most
+	// cases.
+	if (us_is_jpeg(dev->run->format)) {
+		if (buf->bytesused < 125) {
+			// https://stackoverflow.com/questions/2253404/what-is-the-smallest-valid-jpeg-file-size-in-bytes
+			US_LOG_DEBUG("Discarding invalid frame, too small to be a valid JPEG: bytesused=%u", buf->bytesused);
+			return false;
+		}
+
+		const uint8_t *const end_ptr = data + buf->bytesused;
+		const uint8_t *const eoi_ptr = end_ptr - 2;
+		const uint16_t eoi_marker = (((uint16_t)(eoi_ptr[0]) << 8) | eoi_ptr[1]);
+		if (eoi_marker != 0xFFD9 && eoi_marker != 0xD900 && eoi_marker != 0x0000) {
+			US_LOG_DEBUG("Discarding truncated JPEG frame: eoi_marker=0x%04x, bytesused=%u", eoi_marker, buf->bytesused);
+			return false;
+		}
+	}
+
+	return true;
+}
+
 static int _device_open_check_cap(us_device_s *dev) {
 	struct v4l2_capability cap = {0};
 
@@ -405,7 +508,13 @@ static int _device_open_check_cap(us_device_s *dev) {
 		return -1;
 	}
 
-	if (!(cap.capabilities & V4L2_CAP_VIDEO_CAPTURE)) {
+	if (cap.capabilities & V4L2_CAP_VIDEO_CAPTURE) {
+		_RUN(capture_type) = V4L2_BUF_TYPE_VIDEO_CAPTURE;
+		US_LOG_INFO("Using capture type: single-planar");
+	} else if (cap.capabilities & V4L2_CAP_VIDEO_CAPTURE_MPLANE) {
+		_RUN(capture_type) = V4L2_BUF_TYPE_VIDEO_CAPTURE_MPLANE;
+		US_LOG_INFO("Using capture type: multi-planar");
+	} else {
 		US_LOG_ERROR("Video capture is not supported by device");
 		return -1;
 	}
@@ -415,11 +524,13 @@ static int _device_open_check_cap(us_device_s *dev) {
 		return -1;
 	}
 
-	int input = dev->input; // Needs a pointer to int for ioctl()
-	US_LOG_INFO("Using input channel: %d", input);
-	if (_D_XIOCTL(VIDIOC_S_INPUT, &input) < 0) {
-		US_LOG_ERROR("Can't set input channel");
-		return -1;
+	if (!_D_IS_MPLANE) {
+		int input = dev->input; // Needs a pointer to int for ioctl()
+		US_LOG_INFO("Using input channel: %d", input);
+		if (_D_XIOCTL(VIDIOC_S_INPUT, &input) < 0) {
+			US_LOG_ERROR("Can't set input channel");
+			return -1;
+		}
 	}
 
 	if (dev->standard != V4L2_STD_UNKNOWN) {
@@ -497,16 +608,25 @@ static int _device_apply_dv_timings(us_device_s *dev) {
 	return 0;
 }
 
-static int _device_open_format(us_device_s *dev, bool first) {
+static int _device_open_format(us_device_s *dev, bool first) { // FIXME
 	const unsigned stride = us_align_size(_RUN(width), 32) << 1;
 
 	struct v4l2_format fmt = {0};
-	fmt.type = V4L2_BUF_TYPE_VIDEO_CAPTURE;
-	fmt.fmt.pix.width = _RUN(width);
-	fmt.fmt.pix.height = _RUN(height);
-	fmt.fmt.pix.pixelformat = dev->format;
-	fmt.fmt.pix.field = V4L2_FIELD_ANY;
-	fmt.fmt.pix.bytesperline = stride;
+	fmt.type = _RUN(capture_type);
+	if (_D_IS_MPLANE) {
+		fmt.fmt.pix_mp.width = _RUN(width);
+		fmt.fmt.pix_mp.height = _RUN(height);
+		fmt.fmt.pix_mp.pixelformat = dev->format;
+		fmt.fmt.pix_mp.field = V4L2_FIELD_ANY;
+		fmt.fmt.pix_mp.flags = 0;
+		fmt.fmt.pix_mp.num_planes = 1;
+	} else {
+		fmt.fmt.pix.width = _RUN(width);
+		fmt.fmt.pix.height = _RUN(height);
+		fmt.fmt.pix.pixelformat = dev->format;
+		fmt.fmt.pix.field = V4L2_FIELD_ANY;
+		fmt.fmt.pix.bytesperline = stride;
+	}
 
 	// Set format
 	US_LOG_DEBUG("Probing device format=%s, stride=%u, resolution=%ux%u ...",
@@ -516,13 +636,21 @@ static int _device_open_format(us_device_s *dev, bool first) {
 		return -1;
 	}
 
+	if (fmt.type != _RUN(capture_type)) {
+		US_LOG_ERROR("Capture format mismatch, please report to the developer");
+		return -1;
+	}
+
+#	define FMT(x_next)	(_D_IS_MPLANE ? fmt.fmt.pix_mp.x_next : fmt.fmt.pix.x_next)
+#	define FMTS(x_next)	(_D_IS_MPLANE ? fmt.fmt.pix_mp.plane_fmt[0].x_next : fmt.fmt.pix.x_next)
+
 	// Check resolution
 	bool retry = false;
-	if (fmt.fmt.pix.width != _RUN(width) || fmt.fmt.pix.height != _RUN(height)) {
+	if (FMT(width) != _RUN(width) || FMT(height) != _RUN(height)) {
 		US_LOG_ERROR("Requested resolution=%ux%u is unavailable", _RUN(width), _RUN(height));
 		retry = true;
 	}
-	if (_device_apply_resolution(dev, fmt.fmt.pix.width, fmt.fmt.pix.height) < 0) {
+	if (_device_apply_resolution(dev, FMT(width), FMT(height)) < 0) {
 		return -1;
 	}
 	if (first && retry) {
@@ -531,27 +659,32 @@ static int _device_open_format(us_device_s *dev, bool first) {
 	US_LOG_INFO("Using resolution: %ux%u", _RUN(width), _RUN(height));
 
 	// Check format
-	if (fmt.fmt.pix.pixelformat != dev->format) {
+	if (FMT(pixelformat) != dev->format) {
 		US_LOG_ERROR("Could not obtain the requested format=%s; driver gave us %s",
 			_format_to_string_supported(dev->format),
-			_format_to_string_supported(fmt.fmt.pix.pixelformat));
+			_format_to_string_supported(FMT(pixelformat)));
 
 		char *format_str;
-		if ((format_str = (char *)_format_to_string_nullable(fmt.fmt.pix.pixelformat)) != NULL) {
+		if ((format_str = (char *)_format_to_string_nullable(FMT(pixelformat))) != NULL) {
 			US_LOG_INFO("Falling back to format=%s", format_str);
 		} else {
 			char fourcc_str[8];
 			US_LOG_ERROR("Unsupported format=%s (fourcc)",
-				us_fourcc_to_string(fmt.fmt.pix.pixelformat, fourcc_str, 8));
+				us_fourcc_to_string(FMT(pixelformat), fourcc_str, 8));
 			return -1;
 		}
 	}
 
-	_RUN(format) = fmt.fmt.pix.pixelformat;
+	_RUN(format) = FMT(pixelformat);
 	US_LOG_INFO("Using format: %s", _format_to_string_supported(_RUN(format)));
 
-	_RUN(stride) = fmt.fmt.pix.bytesperline;
-	_RUN(raw_size) = fmt.fmt.pix.sizeimage; // Only for userptr
+
+	_RUN(stride) = FMTS(bytesperline);
+	_RUN(raw_size) = FMTS(sizeimage); // Only for userptr
+
+#	undef FMTS
+#	undef FMT
+
 	return 0;
 }
 
@@ -559,7 +692,7 @@ static void _device_open_hw_fps(us_device_s *dev) {
 	_RUN(hw_fps) = 0;
 
 	struct v4l2_streamparm setfps = {0};
-	setfps.type = V4L2_BUF_TYPE_VIDEO_CAPTURE;
+	setfps.type = _RUN(capture_type);
 
 	US_LOG_DEBUG("Querying HW FPS ...");
 	if (_D_XIOCTL(VIDIOC_G_PARM, &setfps) < 0) {
@@ -579,7 +712,7 @@ static void _device_open_hw_fps(us_device_s *dev) {
 #	define SETFPS_TPF(x_next) setfps.parm.capture.timeperframe.x_next
 
 	US_MEMSET_ZERO(setfps);
-	setfps.type = V4L2_BUF_TYPE_VIDEO_CAPTURE;
+	setfps.type = _RUN(capture_type);
 	SETFPS_TPF(numerator) = 1;
 	SETFPS_TPF(denominator) = (dev->desired_fps == 0 ? 255 : dev->desired_fps);
 
@@ -642,7 +775,7 @@ static int _device_open_io_method(us_device_s *dev) {
 static int _device_open_io_method_mmap(us_device_s *dev) {
 	struct v4l2_requestbuffers req = {0};
 	req.count = dev->n_bufs;
-	req.type = V4L2_BUF_TYPE_VIDEO_CAPTURE;
+	req.type = _RUN(capture_type);
 	req.memory = V4L2_MEMORY_MMAP;
 
 	US_LOG_DEBUG("Requesting %u device buffers for MMAP ...", req.count);
@@ -663,9 +796,14 @@ static int _device_open_io_method_mmap(us_device_s *dev) {
 	US_CALLOC(_RUN(hw_bufs), req.count);
 	for (_RUN(n_bufs) = 0; _RUN(n_bufs) < req.count; ++_RUN(n_bufs)) {
 		struct v4l2_buffer buf = {0};
-		buf.type = V4L2_BUF_TYPE_VIDEO_CAPTURE;
+		struct v4l2_plane planes[VIDEO_MAX_PLANES] = {0};
+		buf.type = _RUN(capture_type);
 		buf.memory = V4L2_MEMORY_MMAP;
 		buf.index = _RUN(n_bufs);
+		if (_D_IS_MPLANE) {
+			buf.m.planes = planes;
+			buf.length = VIDEO_MAX_PLANES;
+		}
 
 		US_LOG_DEBUG("Calling us_xioctl(VIDIOC_QUERYBUF) for device buffer=%u ...", _RUN(n_bufs));
 		if (_D_XIOCTL(VIDIOC_QUERYBUF, &buf) < 0) {
@@ -677,20 +815,28 @@ static int _device_open_io_method_mmap(us_device_s *dev) {
 
 		HW(dma_fd) = -1;
 
+		const size_t buf_size = (_D_IS_MPLANE ? buf.m.planes[0].length : buf.length);
+		const off_t buf_offset = (_D_IS_MPLANE ? buf.m.planes[0].m.mem_offset : buf.m.offset);
+
 		US_LOG_DEBUG("Mapping device buffer=%u ...", _RUN(n_bufs));
 		if ((HW(raw.data) = mmap(
 			NULL,
-			buf.length,
+			buf_size,
 			PROT_READ | PROT_WRITE,
 			MAP_SHARED,
 			_RUN(fd),
-			buf.m.offset
+			buf_offset
 		)) == MAP_FAILED) {
 			US_LOG_PERROR("Can't map device buffer=%u", _RUN(n_bufs));
 			return -1;
 		}
 		assert(HW(raw.data) != NULL);
-		HW(raw.allocated) = buf.length;
+
+		HW(raw.allocated) = buf_size;
+
+		if (_D_IS_MPLANE) {
+			US_CALLOC(HW(buf.m.planes), VIDEO_MAX_PLANES);
+		}
 
 #		undef HW
 	}
@@ -700,7 +846,7 @@ static int _device_open_io_method_mmap(us_device_s *dev) {
 static int _device_open_io_method_userptr(us_device_s *dev) {
 	struct v4l2_requestbuffers req = {0};
 	req.count = dev->n_bufs;
-	req.type = V4L2_BUF_TYPE_VIDEO_CAPTURE;
+	req.type = _RUN(capture_type);
 	req.memory = V4L2_MEMORY_USERPTR;
 
 	US_LOG_DEBUG("Requesting %u device buffers for USERPTR ...", req.count);
@@ -728,6 +874,9 @@ static int _device_open_io_method_userptr(us_device_s *dev) {
 		assert((HW(raw.data) = aligned_alloc(page_size, buf_size)) != NULL);
 		memset(HW(raw.data), 0, buf_size);
 		HW(raw.allocated) = buf_size;
+		if (_D_IS_MPLANE) {
+			US_CALLOC(HW(buf.m.planes), VIDEO_MAX_PLANES);
+		}
 #		undef HW
 	}
 	return 0;
@@ -736,10 +885,18 @@ static int _device_open_io_method_userptr(us_device_s *dev) {
 static int _device_open_queue_buffers(us_device_s *dev) {
 	for (unsigned index = 0; index < _RUN(n_bufs); ++index) {
 		struct v4l2_buffer buf = {0};
-		buf.type = V4L2_BUF_TYPE_VIDEO_CAPTURE;
+		struct v4l2_plane planes[VIDEO_MAX_PLANES] = {0};
+		buf.type = _RUN(capture_type);
 		buf.memory = dev->io_method;
 		buf.index = index;
+		if (_D_IS_MPLANE) {
+			buf.m.planes = planes;
+			buf.length = 1;
+		}
+		
 		if (dev->io_method == V4L2_MEMORY_USERPTR) {
+			// I am not sure, may be this is incorrect for mplane device, 
+			// but i don't have one which supports V4L2_MEMORY_USERPTR
 			buf.m.userptr = (unsigned long)_RUN(hw_bufs)[index].raw.data;
 			buf.length = _RUN(hw_bufs)[index].raw.allocated;
 		}
@@ -843,7 +1000,7 @@ static int _device_query_control(
 }
 
 static void _device_set_control(
-	us_device_s *dev, struct v4l2_queryctrl *query,
+	us_device_s *dev, const struct v4l2_queryctrl *query,
 	const char *name, unsigned cid, int value, bool quiet) {
 
 	if (value < query->minimum || value > query->maximum || value % query->step != 0) {

src/ustreamer/device.h

@@ -50,10 +50,10 @@
 
 
 #define US_VIDEO_MIN_WIDTH		((unsigned)160)
-#define US_VIDEO_MAX_WIDTH		((unsigned)10240)
+#define US_VIDEO_MAX_WIDTH		((unsigned)15360)
 
 #define US_VIDEO_MIN_HEIGHT		((unsigned)120)
-#define US_VIDEO_MAX_HEIGHT		((unsigned)4320)
+#define US_VIDEO_MAX_HEIGHT		((unsigned)8640)
 
 #define US_VIDEO_MAX_FPS		((unsigned)120)
 
@@ -61,7 +61,7 @@
 #define US_STANDARDS_STR		"PAL, NTSC, SECAM"
 
 #define US_FORMAT_UNKNOWN		-1
-#define US_FORMATS_STR			"YUYV, UYVY, RGB565, RGB24, MJPEG, JPEG"
+#define US_FORMATS_STR			"YUYV, UYVY, RGB565, RGB24, BGR24, MJPEG, JPEG"
 
 #define US_IO_METHOD_UNKNOWN	-1
 #define US_IO_METHODS_STR		"MMAP, USERPTR"
@@ -75,18 +75,19 @@ typedef struct {
 } us_hw_buffer_s;
 
 typedef struct {
-	int				fd;
-	unsigned		width;
-	unsigned		height;
-	unsigned		format;
-	unsigned		stride;
-	unsigned		hw_fps;
-	unsigned		jpeg_quality;
-	size_t			raw_size;
-	unsigned		n_bufs;
-	us_hw_buffer_s	*hw_bufs;
-	bool			capturing;
-	bool			persistent_timeout_reported;
+	int					fd;
+	unsigned			width;
+	unsigned			height;
+	unsigned			format;
+	unsigned			stride;
+	unsigned			hw_fps;
+	unsigned			jpeg_quality;
+	size_t				raw_size;
+	unsigned			n_bufs;
+	us_hw_buffer_s		*hw_bufs;
+	enum v4l2_buf_type	capture_type;
+	bool				capturing;
+	bool				persistent_timeout_reported;
 } us_device_runtime_s;
 
 typedef enum {
@@ -132,9 +133,7 @@ typedef struct {
 	size_t				min_frame_size;
 	bool				persistent;
 	unsigned			timeout;
-
 	us_controls_s 		ctl;
-
 	us_device_runtime_s *run;
 } us_device_s;
 

src/ustreamer/encoder.c

@@ -195,7 +195,7 @@ static void _worker_job_destroy(void *v_job) {
 static bool _worker_run_job(us_worker_s *wr) {
 	us_encoder_job_s *job = (us_encoder_job_s *)wr->job;
 	us_encoder_s *enc = job->enc; // Just for _ER()
-	us_frame_s *src = &job->hw->raw;
+	const us_frame_s *src = &job->hw->raw;
 	us_frame_s *dest = job->dest;
 
 	assert(_ER(type) != US_ENCODER_TYPE_UNKNOWN);

src/ustreamer/encoders/cpu/encoder.c

@@ -41,6 +41,7 @@ static void _jpeg_write_scanlines_yuyv(struct jpeg_compress_struct *jpeg, const
 static void _jpeg_write_scanlines_uyvy(struct jpeg_compress_struct *jpeg, const us_frame_s *frame);
 static void _jpeg_write_scanlines_rgb565(struct jpeg_compress_struct *jpeg, const us_frame_s *frame);
 static void _jpeg_write_scanlines_rgb24(struct jpeg_compress_struct *jpeg, const us_frame_s *frame);
+static void _jpeg_write_scanlines_bgr24(struct jpeg_compress_struct *jpeg, const us_frame_s *frame);
 
 static void _jpeg_init_destination(j_compress_ptr jpeg);
 static boolean _jpeg_empty_output_buffer(j_compress_ptr jpeg);
@@ -63,7 +64,7 @@ void us_cpu_encoder_compress(const us_frame_s *src, us_frame_s *dest, unsigned q
 	jpeg.image_width = src->width;
 	jpeg.image_height = src->height;
 	jpeg.input_components = 3;
-	jpeg.in_color_space = JCS_RGB;
+	jpeg.in_color_space = ((src->format == V4L2_PIX_FMT_YUYV || src->format == V4L2_PIX_FMT_UYVY) ? JCS_YCbCr : JCS_RGB);
 
 	jpeg_set_defaults(&jpeg);
 	jpeg_set_quality(&jpeg, quality, TRUE);
@@ -79,6 +80,7 @@ void us_cpu_encoder_compress(const us_frame_s *src, us_frame_s *dest, unsigned q
 		WRITE_SCANLINES(V4L2_PIX_FMT_UYVY, _jpeg_write_scanlines_uyvy);
 		WRITE_SCANLINES(V4L2_PIX_FMT_RGB565, _jpeg_write_scanlines_rgb565);
 		WRITE_SCANLINES(V4L2_PIX_FMT_RGB24, _jpeg_write_scanlines_rgb24);
+		WRITE_SCANLINES(V4L2_PIX_FMT_BGR24, _jpeg_write_scanlines_bgr24);
 		default: assert(0 && "Unsupported input format for CPU encoder");
 	}
 
@@ -106,39 +108,29 @@ static void _jpeg_set_dest_frame(j_compress_ptr jpeg, us_frame_s *frame) {
 	frame->used = 0;
 }
 
-#define YUV_R(_y, _, _v)	(((_y) + (359 * (_v))) >> 8)
-#define YUV_G(_y, _u, _v)	(((_y) - (88 * (_u)) - (183 * (_v))) >> 8)
-#define YUV_B(_y, _u, _)	(((_y) + (454 * (_u))) >> 8)
-#define NORM_COMPONENT(_x)	(((_x) > 255) ? 255 : (((_x) < 0) ? 0 : (_x)))
-
 static void _jpeg_write_scanlines_yuyv(struct jpeg_compress_struct *jpeg, const us_frame_s *frame) {
 	uint8_t *line_buf;
 	US_CALLOC(line_buf, frame->width * 3);
 
 	const unsigned padding = us_frame_get_padding(frame);
 	const uint8_t *data = frame->data;
-	unsigned z = 0;
 
 	while (jpeg->next_scanline < frame->height) {
 		uint8_t *ptr = line_buf;
 
 		for (unsigned x = 0; x < frame->width; ++x) {
-			const int y = (!z ? data[0] << 8 : data[2] << 8);
-			const int u = data[1] - 128;
-			const int v = data[3] - 128;
+			// See also: https://www.kernel.org/doc/html/v4.8/media/uapi/v4l/pixfmt-yuyv.html
+			const bool is_odd_pixel = x & 1;
+			const uint8_t y = data[is_odd_pixel ? 2 : 0];
+			const uint8_t u = data[1];
+			const uint8_t v = data[3];
 
-			const int r = YUV_R(y, u, v);
-			const int g = YUV_G(y, u, v);
-			const int b = YUV_B(y, u, v);
+			ptr[0] = y;
+			ptr[1] = u;
+			ptr[2] = v;
+			ptr += 3;
 
-			*(ptr++) = NORM_COMPONENT(r);
-			*(ptr++) = NORM_COMPONENT(g);
-			*(ptr++) = NORM_COMPONENT(b);
-
-			if (z++) {
-				z = 0;
-				data += 4;
-			}
+			data += (is_odd_pixel ? 4: 0);
 		}
 		data += padding;
 
@@ -155,28 +147,23 @@ static void _jpeg_write_scanlines_uyvy(struct jpeg_compress_struct *jpeg, const
 
 	const unsigned padding = us_frame_get_padding(frame);
 	const uint8_t *data = frame->data;
-	unsigned z = 0;
 
 	while (jpeg->next_scanline < frame->height) {
 		uint8_t *ptr = line_buf;
 
 		for (unsigned x = 0; x < frame->width; ++x) {
-			const int y = (!z ? data[1] << 8 : data[3] << 8);
-			const int u = data[0] - 128;
-			const int v = data[2] - 128;
+			// See also: https://www.kernel.org/doc/html/v4.8/media/uapi/v4l/pixfmt-uyvy.html
+			const bool is_odd_pixel = x & 1;
+			const uint8_t y = data[is_odd_pixel ? 3 : 1];
+			const uint8_t u = data[0];
+			const uint8_t v = data[2];
 
-			const int r = YUV_R(y, u, v);
-			const int g = YUV_G(y, u, v);
-			const int b = YUV_B(y, u, v);
+			ptr[0] = y;
+			ptr[1] = u;
+			ptr[2] = v;
+			ptr += 3;
 
-			*(ptr++) = NORM_COMPONENT(r);
-			*(ptr++) = NORM_COMPONENT(g);
-			*(ptr++) = NORM_COMPONENT(b);
-
-			if (z++) {
-				z = 0;
-				data += 4;
-			}
+			data += (is_odd_pixel ? 4 : 0);
 		}
 		data += padding;
 
@@ -187,11 +174,6 @@ static void _jpeg_write_scanlines_uyvy(struct jpeg_compress_struct *jpeg, const
 	free(line_buf);
 }
 
-#undef NORM_COMPONENT
-#undef YUV_B
-#undef YUV_G
-#undef YUV_R
-
 static void _jpeg_write_scanlines_rgb565(struct jpeg_compress_struct *jpeg, const us_frame_s *frame) {
 	uint8_t *line_buf;
 	US_CALLOC(line_buf, frame->width * 3);
@@ -205,9 +187,10 @@ static void _jpeg_write_scanlines_rgb565(struct jpeg_compress_struct *jpeg, cons
 		for (unsigned x = 0; x < frame->width; ++x) {
 			const unsigned int two_byte = (data[1] << 8) + data[0];
 
-			*(ptr++) = data[1] & 248; // Red
-			*(ptr++) = (uint8_t)((two_byte & 2016) >> 3); // Green
-			*(ptr++) = (data[0] & 31) * 8; // Blue
+			ptr[0] = data[1] & 248; // Red
+			ptr[1] = (uint8_t)((two_byte & 2016) >> 3); // Green
+			ptr[2] = (data[0] & 31) * 8; // Blue
+			ptr += 3;
 
 			data += 2;
 		}
@@ -232,6 +215,33 @@ static void _jpeg_write_scanlines_rgb24(struct jpeg_compress_struct *jpeg, const
 	}
 }
 
+static void _jpeg_write_scanlines_bgr24(struct jpeg_compress_struct *jpeg, const us_frame_s *frame) {
+	uint8_t *line_buf;
+	US_CALLOC(line_buf, frame->width * 3);
+
+	const unsigned padding = us_frame_get_padding(frame);
+	uint8_t *data = frame->data;
+
+	while (jpeg->next_scanline < frame->height) {
+		uint8_t *ptr = line_buf;
+
+		// swap B and R values
+		for (unsigned x = 0; x < frame->width * 3; x += 3) {
+			ptr[0] = data[x + 2];
+			ptr[1] = data[x + 1];
+			ptr[2] = data[x];
+			ptr += 3;
+		}
+		
+		JSAMPROW scanlines[1] = {line_buf};
+		jpeg_write_scanlines(jpeg, scanlines, 1);
+
+		data += (frame->width * 3) + padding;
+	}
+
+	free(line_buf);
+}
+
 #define JPEG_OUTPUT_BUFFER_SIZE ((size_t)4096)
 
 static void _jpeg_init_destination(j_compress_ptr jpeg) {

src/ustreamer/options.c

@@ -233,7 +233,7 @@ static int _parse_resolution(const char *str, unsigned *width, unsigned *height,
 static int _check_instance_id(const char *str);
 
 static void _features(void);
-static void _help(FILE *fp, us_device_s *dev, us_encoder_s *enc, us_stream_s *stream, us_server_s *server);
+static void _help(FILE *fp, const us_device_s *dev, const us_encoder_s *enc, const us_stream_s *stream, const us_server_s *server);
 
 
 us_options_s *us_options_init(unsigned argc, char *argv[]) {
@@ -586,7 +586,7 @@ static void _features(void) {
 #	endif
 }
 
-static void _help(FILE *fp, us_device_s *dev, us_encoder_s *enc, us_stream_s *stream, us_server_s *server) {
+static void _help(FILE *fp, const us_device_s *dev, const us_encoder_s *enc, const us_stream_s *stream, const us_server_s *server) {
 #	define SAY(x_msg, ...) fprintf(fp, x_msg "\n", ##__VA_ARGS__)
 	SAY("\nuStreamer - Lightweight and fast MJPEG-HTTP streamer");
 	SAY("═══════════════════════════════════════════════════");
@@ -696,7 +696,7 @@ static void _help(FILE *fp, us_device_s *dev, us_encoder_s *enc, us_stream_s *st
 	ADD_SINK("RAW", "raw-")
 	ADD_SINK("H264", "h264-")
 	SAY("    --h264-bitrate <kbps>  ───────── H264 bitrate in Kbps. Default: %u.\n", stream->h264_bitrate);
-	SAY("    --h264-gop <N>  ──────────────── Intarval between keyframes. Default: %u.\n", stream->h264_gop);
+	SAY("    --h264-gop <N>  ──────────────── Interval between keyframes. Default: %u.\n", stream->h264_gop);
 	SAY("    --h264-m2m-device </dev/path>  ─ Path to V4L2 M2M encoder device. Default: auto select.\n");
 #	undef ADD_SINK
 #	ifdef WITH_GPIO

src/ustreamer/stream.c

@@ -112,9 +112,10 @@ void us_stream_loop(us_stream_s *stream) {
 				if (!ready_wr->job_failed) {
 					if (ready_wr->job_timely) {
 						_stream_expose_frame(stream, ready_job->dest, captured_fps);
-						US_LOG_PERF("##### Encoded frame exposed; worker=%s", ready_wr->name);
+						US_LOG_PERF("##### Encoded JPEG exposed; worker=%s, latency=%.3Lf",
+							ready_wr->name, us_get_now_monotonic() - ready_job->dest->grab_ts);
 					} else {
-						US_LOG_PERF("----- Encoded frame dropped; worker=%s", ready_wr->name);
+						US_LOG_PERF("----- Encoded JPEG dropped; worker=%s", ready_wr->name);
 					}
 				} else {
 					break;
@@ -126,7 +127,6 @@ void us_stream_loop(us_stream_s *stream) {
 				unsigned slc = 0;
 				for (; slc < 10 && !atomic_load(&_RUN(stop)) && !us_stream_has_clients(stream); ++slc) {
 					usleep(100000);
-					++slc;
 				}
 				h264_force_key = (slc == 10);
 			}

src/ustreamer/workers.c

@@ -157,7 +157,7 @@ void us_workers_pool_assign(us_workers_pool_s *pool, us_worker_s *ready_wr/*, vo
 	US_MUTEX_UNLOCK(pool->free_workers_mutex);
 }
 
-long double us_workers_pool_get_fluency_delay(us_workers_pool_s *pool, us_worker_s *ready_wr) {
+long double us_workers_pool_get_fluency_delay(us_workers_pool_s *pool, const us_worker_s *ready_wr) {
 	const long double approx_job_time = pool->approx_job_time * 0.9 + ready_wr->last_job_time * 0.1;
 
 	US_LOG_VERBOSE("Correcting pool's %s approx_job_time: %.3Lf -> %.3Lf (last_job_time=%.3Lf)",

src/ustreamer/workers.h

@@ -92,4 +92,4 @@ void us_workers_pool_destroy(us_workers_pool_s *pool);
 us_worker_s *us_workers_pool_wait(us_workers_pool_s *pool);
 void us_workers_pool_assign(us_workers_pool_s *pool, us_worker_s *ready_wr/*, void *job*/);
 
-long double us_workers_pool_get_fluency_delay(us_workers_pool_s *pool, us_worker_s *ready_wr);
+long double us_workers_pool_get_fluency_delay(us_workers_pool_s *pool, const us_worker_s *ready_wr);