KWin/Debugging: Difference between revisions

From KDE Community Wiki
Line 179: Line 179:
= Automatically fetch core dumps with coredumpctl =
= Automatically fetch core dumps with coredumpctl =


TODO edit /etc/systemd/coredump.conf
After installing systemd-coredumpctl, you should now have a file located in /etc/systemd/coredump.conf that can be used to configure coredumpctl.


However, kwin_wayland crashes might be a bit trickier to debug. Because KWin plays the traditional role of what is known as Display Server (in X11 lingo) in a Wayland session, whenever it crashes, you get thrown off to the login screen. Therefore, it's advisable to use a different TTY or SSH and attach GDB directly instead of using coredumpctl (as systemd-coredump might not be able to fetch the stacktrace and it might get truncated).
The <tt>Storage=</tt> setting defaults to external. This means that core dumps, in addition to being available through coredumpctl, are stored as physical files under /var/lib/systemd/coredump/. If you wish to be able to rotate core dumps together with the journal, you might want to set Storage=journal.
 
{{Warning| Crashes in kwin_wayland might be a bit trickier to debug. Because KWin plays the traditional role of what is known as Display Server (in X11 lingo) in a Wayland session, whenever it is killed, you get thrown off to the login screen. On rare occasions, coredumpctl cannot fetch a full backtrace, rendering a truncated core dump. If you encounter these issues with coredumpctl, it's advisable to use a different TTY or SSH and attach GDB directly instead.}}
 
{{Note| Older versions of kwin_wayland used to throw the user back to the login screen if it it got killed (SIGKILL) or suffered a segmentation fault (SIGSEGV). As per <nowiki>https://invent.kde.org/plasma/kwin/-/merge_requests/483</nowiki>, SIGSEGV no longer causes this issue.}}
 
The main instructions on using coredumpctl to fetch are available on [[Guidelines_and_HOWTOs/Debugging/How_to_create_useful_crash_reports#Retrieving_a_backtrace_using_coredumpctl| the wiki page on crash reporting]]. Basically, it amounts to this:
 
<tt>coredumpctl</tt> &rarr; check the PID of the crashed process &rarr; coredumpctl dbg <KWin PID here> &rarr; bt &rarr; copy into a log file
 
Or, alternatively:
 
{{Input|1=<nowiki>
coredumpctl dump <KWin PID>
</nowiki>}}
 
It's possible to create a log file containing the backtrace by setting the --output flag to cat and redirecting it to a file:
 
{{Input|1=<nowiki>
coredumpctl dump <KWin PID> --output=cat > kwincrash.log
</nowiki>}}


= Getting debug log output =
= Getting debug log output =

Revision as of 01:36, 10 January 2021

 
Under Construction
This is a new page, currently under construction!


Getting diagnosis information for bug reports

If you want to fetch relevant information to report a bug with KWin, the following command will provide a general list of data that should help the KWin developers diagnose your problem.

qdbus org.kde.KWin /KWin supportInformation

Depending on your distro (e.g. openSUSE), by default the command might be named a bit differently:

qdbus-qt5 org.kde.KWin /KWin supportInformation

On occasion, xwininfo might be asked for by the developers if the issue concerns screens and windows, and xprop if the issue concerns window information. Both commands can be run standalone, but depending on the issue it might be useful to run them on specific windows.

Report issues via DrKonqi

Generally speaking, Plasmashell handles all widgets (including the menu), KWin handles windows and compositing (such as window decorations and desktop effects), and KGlobalAccel5 handles keyboard shortcuts.

In a situation where the Plasmashell and KGlobalAccel5 processes are still running but KWin has crashed, you will probably see a sad face in your notification tray, that's DrKonqi, the KDE Crash Handler, getting sad that you experienced a crash. Click its icon and you should be able to follow through with the crash reporting process in a straightforward manner.

In a situation where KGlobalAccel5 is still running but both Plasmashell and KWin are running, you can still invoke any keyboard shortcut to run a program that allows you to run a command, such as KRunner, Konsole or Yakuake. With that, you can restart the plasmashell process with:

plasmashell --replace

Or if you have manually enabled the new systemd initialization:

systemctl --user restart plasma-plasmashell

And you might get a DrKonqi icon on your panel mentioning the KWin crash.

In case you were unable to create a backtrace using the KDE Crash Handler, proceed to the section KWin/Debugging#Debug KWin with GDB.

Install debug symbols

Depending on your distribution, you might need additional steps before you're able to install debug symbols, as detailed on the instructions on how to install debugging packages.

Debug KWin with GDB

TL;DR for bug reporters

For ease of reference, users wanting to report a KWin crash can just copy-paste the following two commands, wait for the crash to happen and ignore the rest of this page. You'll get a file named kwin.gdb in your home folder (or wherever folder you run this command).

echo 0 | sudo tee /proc/sys/kernel/yama/ptrace_scope
gdb -pid $(pidof kwin_x11) -batch -ex "set logging file kwin.gdb" -ex "set logging on" -ex "continue" -ex "thread apply all backtrace" -ex "quit"

If you want to know what these commands do (recommended), keep reading.

General information

While interacting with GDB, the debugged process is stopped - that's of course nasty if the debugged process is what paints what you see.

KWin used to have issues with attaching GDB, but it is no longer the case. Nevertheless, if you ever encounter issues doing this, it might be preferable to debug KWin from a side-channel, eg. another TTY or via SSH depending on what's available on your system. This is especially true if you are unable to both fetch a backtrace and a core dump from kwin_wayland.

Note

Debugging from an SSH connection is generally preferable over a TTY, since it doesn't impact the framebuffer/scanout buffer state.


GDB says "ptrace: Operation not permitted."

This is probably the first thing you'll need to circumvent. By default, you're not allowed to attach GDB to KWin.

This is a security feature in "newer" Linux kernels, you'll need to explicitly allow GDB to attach to a non-inferior process:

echo 0 | sudo tee /proc/sys/kernel/yama/ptrace_scope

Do I have to write down the debug output by hand ????

No ;-)

You just copy the gdb output into a file.

gdb --pid `pidof kwin_x11` 2>&1 | tee kwin.gdb

Alternatively, you can use the lengthier method made available by GDB.

gdb --pid `pidof kwin_x11`
set logging file kwin.gdb
set logging on

I know nothing about GDB, how do I obtain a stacktrace?

A stacktrace (or backtrace) is a set of data containing information of the state the program was when it crashed. It is the primary means for the KWin developers to find out what happened on your machine.

Nowadays it's common for programs to take advantage of multiple threads to run more efficiently and make better use of the CPU cores available on your machine. Different threads might run different parts of a program, and each thread can render a different stacktrace.

So after attaching GDB to the application process, you'll see the GDB shell. It's where you'll run instructions so GDB can fetch the information you want.

Immediately after attaching GDB, it will stop the process, but you'll likely want to cause a certain condition (halt or crash) while KWin is running - in that case you first need to

continue

the process before it can be crashed.

If KWin does not crash, but you want to inspect the stack at some other time, you'll first need to interrupt the process by pressing

Ctrl+C

To dump a stacktrace, issue

bt

With this command you will get the stacktrace for the main thread where KWin crashed.

Then, hit the

Enter

key until you reach the end of the stack.

Sometimes, you may want to see what's in another thread

thread 2
bt
thread 3
bt

Alternatively, to dump a stacktrace the same way DrKonqi does (i.e. showing all available threads), use

thread apply all backtrace

This is probably the best method if you want to provide some awesome stacktraces for the KWin devs!

Finally, to leave GDB:

detach
quit

Automating the creation of backtraces

GDB provides an easy way to automate debugging: the -batch flag. After enabling it, you should be able to run commands sequentially with -ex or --eval-command. A more detailed explanation of the procedure can be seen in Plasma/Debugging.

gdb -pid $(pidof kwin_x11) -batch -ex "set logging file kwin.gdb" -ex "set logging on" -ex "continue" -ex "thread apply all backtrace" -ex "quit"

Debug KWin with Valgrind

It is not possible to attach Valgrind to running processes for fetching backtraces, but it is possible for Valgrind to run a nested KWin Wayland session with an XWayland server running rootless:

valgrind --log-file=kwinxwayland.log dbus-run-session kwin_wayland --xwayland

And a pure KWin Wayland session:

valgrind --log-file=kwinwayland.log dbus-run-session kwin_wayland

KWin X11 is limited to using the currently running dbus session, and therefore it cannot produce nested sessions. The --replace flag is used for substituting your current kwin_x11 process.

valgrind --log-file=kwinx11.log kwin_x11 --replace

Note

It's not possible to run a KWin X11 session from a KWin Wayland session.

In addition to creating backtraces, Valgrind is also able to log memory leaks.

valgrind --leak-check=full --show-reachable=yes --track-origins=yes --trace-children=yes --log-file=kwinxwaylandmem.log dbus-run-session kwin_wayland --xwayland"

Automatically fetch core dumps with coredumpctl

After installing systemd-coredumpctl, you should now have a file located in /etc/systemd/coredump.conf that can be used to configure coredumpctl.

The Storage= setting defaults to external. This means that core dumps, in addition to being available through coredumpctl, are stored as physical files under /var/lib/systemd/coredump/. If you wish to be able to rotate core dumps together with the journal, you might want to set Storage=journal.

Warning

Crashes in kwin_wayland might be a bit trickier to debug. Because KWin plays the traditional role of what is known as Display Server (in X11 lingo) in a Wayland session, whenever it is killed, you get thrown off to the login screen. On rare occasions, coredumpctl cannot fetch a full backtrace, rendering a truncated core dump. If you encounter these issues with coredumpctl, it's advisable to use a different TTY or SSH and attach GDB directly instead.


Note

Older versions of kwin_wayland used to throw the user back to the login screen if it it got killed (SIGKILL) or suffered a segmentation fault (SIGSEGV). As per https://invent.kde.org/plasma/kwin/-/merge_requests/483, SIGSEGV no longer causes this issue.


The main instructions on using coredumpctl to fetch are available on the wiki page on crash reporting. Basically, it amounts to this:

coredumpctl → check the PID of the crashed process → coredumpctl dbg <KWin PID here> → bt → copy into a log file

Or, alternatively:

coredumpctl dump <KWin PID>

It's possible to create a log file containing the backtrace by setting the --output flag to cat and redirecting it to a file:

coredumpctl dump <KWin PID> --output=cat > kwincrash.log

Getting debug log output

The environment variable QT_LOGGING_RULES can be used to turn on full debug output from KWin:

export QT_LOGGING_RULES="kwin_*.debug=true"

The logs in X are at

~/.xsession-errors

The logs in Wayland are in

~/.local/share/sddm/wayland-session.log

TODO KDE_DEBUG=1 to disable DrKonqi