vdr December 2022

vdr@linuxtv.org

4 participants
5 discussions

Running VDR from systemd on Raspberry Pi OS
by Marko Mäkelä 29 Dec '22

29 Dec '22

Yesterday, I finally bought external storage for my Raspberry Pi based VDR setup, a Samsung Portable SSD T7. It supports USB 3, but it also works on the Raspberry Pi 2's USB 2.0 and does not consume too much power. My old tower PC case based system that I had set up in 2004 has now been replaced with something that is better in every thinkable respect: power consumption, noise (passive cooling, no HDDs), speed, and size (not much larger than the remote control unit). Hardware: * Raspberry Pi 2 * Pi TV hat * TV hat case * an IR receiver attached via soldered wires to the TV hat, at GPIO pin 18 * a remote control unit (from an old Hauppauge Nova-T PCI card) * Samsung Portable SSD T7 (1 TB) Software: * Raspberry OS Legacy installed on a MicroSD card, with no GUI * sudo apt install ir-keytable * VDR 2.6.3 (or 2.6.2) compiled from source * https://github.com/reufer/rpihddevice/ compiled from source * "make install" to /usr/local My /boot/config.txt includes the following lines: dtoverlay=gpio-ir,gpio_pin=18 dtparam=audio=on gpu_mem=256 In /etc/rc_maps.cfg (the configuration file of ir-keytable), ensure that there is a line like the following that will match the remote control unit that you are using: * * hauppauge.toml The above works for several RC5 based Hauppauge remote control units. To prevent the Power button on the remote control unit from shutting down the entire system, add the following to the [Login] section of /etc/systemd/logind.conf: HandlePowerKey=ignore The default is HandlePowerKey=poweroff. Use mkfs.ext4 to replace the FAT file system of the only partition of the T7. There is no need to change the partitioning or specify a block size or alignment, because the physical block size is reported as 512 bytes. Optionally, you may set a label by executing something like this: tune2fs -L VDR /dev/sda1 You may create a mount point: sudo mkdir -m 000 /video Then, add a line like this to /etc/fstab to have the storage mounted automatically: LABEL=VDR /video ext4 defaults,noatime,nofail 0 1 You may replace the LABEL=VDR with whatever symbolic link you have in /dev/disk/by-label (see also tune2fs above). On my system, I actually wrote PARTUUID=33d32895-01 because there is a symbolic link /dev/disk/by-partuuid/33d32895-01 that identifies the partition. Once the storage is mounted, execute the following: sudo mkdir /video/video sudo chown pi:pi /video/video The next step is to configure VDR to start up correctly. I have some configuration files in /var/lib/vdr. For testing, I used to start VDR manually from the command line, and shut it down by choosing "restart" from the OSD menu. Now I want it to restart automatically, but only if suitable USB storage has been plugged in: sudo tee /etc/systemd/system/vdr.service << EOF [Unit] After=systemd-user-sessions.service plymouth-quit-wait.service After=rc-local.service After=getty(a)tty1.service After=video.mount Conflicts=getty(a)tty1.service ConditionPathExists=/video/video [Service] User=pi ExecStart=/usr/local/bin/vdr --no-kbd --lirc=/dev/lirc0 -Prpihddevice -v /video/video -s /var/lib/vdr/vdr-shutdown.sh TimeoutStartSec=infinity Type=idle Restart=on-failure RestartSec=1s TTYVTDisallocate=yes [Install] Alias=display-manager.service EOF This will replace the getty process on virtual terminal 1 (tty1). If the storage is not plugged within 90 seconds from startup (I do not know how to configure that timeout), then an error message will appear on the console. No getty will be started on tty1 in any case; you can always log in from tty2 by pressing Alt+F2. The shutdown script /var/lib/vdr/vdr-shutdown.sh does not work as intended yet: #!/bin/sh if [ "$5" = 1 ] then sudo service vdr stop sudo umount /video sudo udisksctl power-off -b /dev/sda fi The first step appears to terminate the shell script, because the shell is a subprocess of VDR. So, the storage will remain mounted and powered on. I guess that we need to launch a separate "vdr-shutdown" service that would take care of the remaining steps. Has someone already implemented something like this? After the "umount" and "udisksctl" commands are executed, it is safe to unplug the storage. The LED of the SSD will shortly change color and then turn off during the execution of the "udisksctl" command. What I am also missing is a udev rule that would automatically mount the storage and attempt to start up VDR as soon as the storage is plugged in. Currently, I have to manually execute the following if I plug in the drive to an already running system: sudo mount /video sudo service vdr start This configuration provides a rather simple user interface for VDR. No keyboard or mouse is needed, just the remote control unit, a display, and optionally the USB cable, if the system has other uses that are independent of VDR. For timed recordings, I think that on the Raspberry Pi, it is easiest to let the VDR process run all the time. Starting up the Pi based on timer would require additional hardware. One thing that I'd like to improve in that regard is to let VDR shut down all tuners when the system is idle. This could be based on an inactivity timer or an explicit user action, such as pressing a button on the remote control. The video output would shut off, and the tuner would be powered off, except when needed for something (EPG scan, recording). As soon as a button is pressed on the remote control unit, the user interface would spring back to life. Happy holidays, Marko

3 8

[ANNOUNCE] VDR version 2.6.3 released
by Klaus Schmidinger 14 Dec '22

14 Dec '22

VDR version 2.6.3 is now available at the official VDR GIT archive git://git.tvdr.de You can also get the latest stable version with git clone --branch stable/2.6 git://git.tvdr.de/vdr.git or as a tar archive with http://git.tvdr.de/?p=vdr.git;a=snapshot;h=refs/tags/2.6.3;sf=tbz2 This version fixes a few bugs that came up after the release of version 2.6.2. The changes since version 2.6.2: - Fixed a compiler warning. - Fixed generating the index file in the cutter (reported by Christoph Haubrich). - Fixed a faulty 'Timer still recording' query when canceling an editing job. - Added code for the 'qks' audio track (thanks to Johann Friedrichs). - Fixed a possible heap-use-after-free in cDvbTuner::Action() (reported by Marko Mäkelä). - Fixed initializing cDvbPlayerControl and cTransferControl (reported by Marko Mäkelä). - Now avoiding calling poll() in cSectionHandler::Action() if there are no filters (reported by Marko Mäkelä). - Now avoiding the memcpy() call in cGlyph::cGlyph() if the bitmap is empty (thanks to Marko Mäkelä). - Now avoiding unnecessary processing in cDvbSubtitleConverter::FinishPage() if there are no areas (thanks to Marko Mäkelä). - Avoiding a zero sized array in cDevice::GetDevice() (thanks to Marko Mäkelä). - Now checking the video directory after setting the user id. Homepage: http://www.tvdr.de Facebook: https://www.facebook.com/VideoDiskRecorder Have fun! Klaus

1 0

[PATCH] ThreadSanitizer warnings for cThread
by Marko Mäkelä 14 Dec '22

14 Dec '22

Because of the heap-use-after-free race condition that was rather easily reproducible with AddressSanitizer (-fsanitize=address), I thought that I should finally try to learn to use ThreadSanitizer (TSAN, -fsanitize=thread in GCC and clang). https://clang.llvm.org/docs/ThreadSanitizer.html Because VDR makes use of POSIX thread synchronization primitives, no additional instrumentation via <sanitizer/tsan_interface.h> should be necessary. Before C++11 defined a memory model for multi-threaded applications, semantics around shared data structures were rather unclear, and I would guess that most multi-threaded pre-C++11 code bases would trip ThreadSanitizer. Also, multi-threaded CPUs were rare in the early days, and the Total Store Order of the x86 is very forgiving, compared to the weak memory model of ARM (see https://www.cl.cam.ac.uk/~pes20/cpp/cpp0xmappings.html for some examples). To silence one prominent source of TSAN warnings in the cThread destructor, I applied the attached patch. It is not ideal, because std::atomic defaults to std::memory_order_seq_cst while std::memory_order_relaxed would likely suffice here. Even after applying the first patch, a simple test with no DVB receiver device and no valid data directory would still produce a couple of data race reports (see the end of this message). I recorded a trace of such a run with "rr record" (https://rr-project.org) and debugged it in "rr replay". Unsurprisingly, I did not find actual evidence of a race condition. Finally, I figured out what is causing the first report: cThread::description is not protected by cThread::mutex. Possibly, most cThread data members (including cThread::active) should be protected by cThread::mutex? With both attached patches applied, the report of the first race will disappear. The second race is apparently about some memory that is allocated inside opendir(). I did not figure it out yet. Related to this, cThread::Cancel() especially when invoked with WaitSeconds=-1 looks problematic to me, and I see that VDR is invoking pthread_detach() and never invoking pthread_join(). The second patch includes an attempt to clean this up as well. Both patches are just a proof of concept; I did not test them beyond the simple failing-to-start VDR run under TSAN. Unfortunately, TSAN is not available for my primary VDR hardware, running on 32-bit ARM. With best regards, Marko vdr: error while reading '/var/lib/vdr/sources.conf' vdr: error while reading '/var/lib/vdr/channels.conf' vdr: no primary device found - using first device! ================== WARNING: ThreadSanitizer: data race on vptr (ctor/dtor vs virtual call) (pid=96847) Write of size 8 at 0x7ffc7f773e60 by main thread: #0 cThread::~cThread() /dev/shm/vdr/thread.c:249 (vdr+0x1d72b8) #1 cEpgDataReader::~cEpgDataReader() /dev/shm/vdr/epg.h:236 (vdr+0xa956b) #2 main /dev/shm/vdr/vdr.c:731 (vdr+0xa956b) Previous read of size 8 at 0x7ffc7f773e60 by thread T2: #0 cThread::StartThread(cThread*) /dev/shm/vdr/thread.c:293 (vdr+0x1d76d9) Location is stack of main thread. Location is global '<null>' at 0x000000000000 ([stack]+0x1fe60) Thread T2 'epg data reader' (tid=96855, finished) created by main thread at: #0 pthread_create ../../../../src/libsanitizer/tsan/tsan_interceptors_posix.cpp:1001 (libtsan.so.2+0x5e686) #1 cThread::Start() /dev/shm/vdr/thread.c:316 (vdr+0x1d6fe0) #2 main /dev/shm/vdr/vdr.c:804 (vdr+0xaa477) SUMMARY: ThreadSanitizer: data race on vptr (ctor/dtor vs virtual call) /dev/shm/vdr/thread.c:249 in cThread::~cThread() ================== ================== WARNING: ThreadSanitizer: data race (pid=96847) Write of size 8 at 0x7b04000005a0 by main thread: #0 free ../../../../src/libsanitizer/tsan/tsan_interceptors_posix.cpp:706 (libtsan.so.2+0x47e82) #1 cString::~cString() /dev/shm/vdr/tools.c:1097 (vdr+0x1e52df) #2 cxa_at_exit_wrapper ../../../../src/libsanitizer/tsan/tsan_interceptors_posix.cpp:389 (libtsan.so.2+0x2dee3) Previous read of size 8 at 0x7b04000005a0 by thread T1: #0 opendir ../../../../src/libsanitizer/sanitizer_common/sanitizer_common_interceptors.inc:3271 (libtsan.so.2+0x4c641) #1 cReadDir::cReadDir(char const*) /dev/shm/vdr/tools.c:1553 (vdr+0x1ea8bd) #2 cVideoDirectoryScannerThread::ScanVideoDir(char const*, int, int) /dev/shm/vdr/recording.c:1439 (vdr+0x180620) #3 cVideoDirectoryScannerThread::Action() /dev/shm/vdr/recording.c:1433 (vdr+0x180bfc) #4 cThread::StartThread(cThread*) /dev/shm/vdr/thread.c:293 (vdr+0x1d76ea) Thread T1 'video directory' (tid=96854, running) created by main thread at: #0 pthread_create ../../../../src/libsanitizer/tsan/tsan_interceptors_posix.cpp:1001 (libtsan.so.2+0x5e686) #1 cThread::Start() /dev/shm/vdr/thread.c:316 (vdr+0x1d6fe0) #2 cRecordings::Update(bool) /dev/shm/vdr/recording.c:1554 (vdr+0x175387) #3 main /dev/shm/vdr/vdr.c:788 (vdr+0xaa3f8) SUMMARY: ThreadSanitizer: data race /dev/shm/vdr/tools.c:1097 in cString::~cString() ================== ThreadSanitizer: reported 2 warnings

2 4

[PATCH] Fix undefined behaviour
by Marko Mäkelä 06 Dec '22

06 Dec '22

Another day, another sanitizer. After fixing issues reported by -fsanitize=address yesterday, I gave -fsanitize=undefined a try. The GCC documentation points to the clang documentation: https://clang.llvm.org/docs/UndefinedBehaviorSanitizer.html The issues related to cControl::player were tricky. In the end, I figured it out after setting UBSAN_OPTIONS=print_stacktrace=1 and setting a breakpoint on _Unwind_Backtrace(). The name of the reporting function in my system was __ubsan_handle_dynamic_type_cache_miss(), nothing about "vptr". Also, the diagnostics was misleadingly pointing to the body of the constructor, and not the initializer list where a data member was being assigned to before the base class had been initialized. The -fsanitize=undefined in clang might report more things. Next I may give -fsanitize=thread a try. GCC does not implement -fsanitize=memory (checking for the use of uninitialized memory) at all. It will require clang and libc++ (not libstdc++) and that all libraries except libc are built with -fsanitize=memory. If you are familiar with Valgrind's default memcheck tool, it is roughly comparable to the combination of -fsanitize=address and -fsanitize=memory. Marko

3 9

[ANNOUNCE] VDR version 2.6.2 released
by Klaus Schmidinger 03 Dec '22

03 Dec '22

VDR version 2.6.2 is now available at the official VDR GIT archive git://git.tvdr.de You can also get the latest stable version with git clone --branch stable/2.6 git://git.tvdr.de/vdr.git or as a tar archive with http://git.tvdr.de/?p=vdr.git;a=snapshot;h=refs/tags/2.6.2;sf=tbz2 This version fixes a few bugs that came up after the release of version 2.6.1. The changes since version 2.6.1: - Added UPDATE-2.6.0, which was missing in the official 2.6.0 release. - Fixed unexpected calls of the '-r' script when a recording is interrupted and the timer has not yet finished. - Now dropping capabilities after opening terminal. - Now assuming the lock when removing deleted recordings even if the disk is full (reported by Claus Muus). - When checking whether a recording is still active, VDR no longer checks whether the index file is being written, but rather checks for the presence of a '.timer' file. The cutter now writes a dummy '.timer' file with timer ID '0' to make this work for recordings that are currently being edited. - Fixed a possible crash if an editing process is canceled while the edited recording is being replayed. - Added a warning if an attempt is made to obtain a write lock twice from the same thread. - Fixed default values for DVB-T (thanks to Winfried Köhler and Jose Angel). - Removed some unnecessary locks from SVDRPClientHandler. - Fixed a possible deadlock in case two SVDRP clients send each other POLL commands at the same time. - Added a missing 'const' to cTimers::GetTimerForEvent() (reported by Markus Ehrnsperger). - Added a chapter about locking to PLUGINS.html (suggested by Markus Ehrnsperger). - Implemented parsing frame rate and image size for MPEG2, H.264 and H.265 (thanks to Christoph Haubrich). - Using the frame rate parsed from the stream, with fall back to determining it from PTS values. - Fixed printing/scanning values for systems where %ld doesn't work for time_t. - Added support for kernel based LIRC driver (thanks to Marko Mäkelä). Use the command line option '--lirc=/dev/lirc0' to use this. Requires kernel version 5.10 or above. - Added periodic calls to malloc_trim(0) to reduce memory consumption (thanks to Onur Sentürk). - Fixed regenerating the index file of a recording in case it is present, but empty (reported by Stefan Herdler). - Added missing rounding when dividing frequencies in processing the NIT (thanks to Winfried Köhler). Homepage: http://www.tvdr.de Facebook: https://www.facebook.com/VideoDiskRecorder Have fun!

2 1

2024

2023

2022

2021

2020

2019

2018

2017

2016

2015

2014

2013

2012

2011

2010

2009

2008

2007

2006

2005

vdr December 2022