On 11.10.2026 01:42, Kamil Jońca wrote:
Fetched gitlab source and make bisect:
This is what I'd do myself if I had a reproducer...
# first 'bad' commit: [d616781f2d63fa602b12020cedbb274d15797e41]
hw/char/serial-isa.c: declare IRQ as shared in ACPI IRQ descriptor
for the reference:
after fetching I make
./configure
and then
make
sudo make install repeatedly
If you will ever need to do that again, there's much easier way:
./configure --target-list=x86_64-softmmu --disable-werror
(this will request only one qemu binary and omit irrelevant parts)
ninja qemu-system-x86_64
(this will build just this binary and nothing else)
This is much, much faster than having to build everything each time.
Now, once you have this binary, you can point libvirt to it, by using
custom qemu binary in its settings, pointing to just built
/qemu/build/dir/qemu-system-x86_64 (whatever the full path is).
I don't recommend to install it just for this kind of testing.
Also, you can add --enable-debug to the ./configure line, this will
help with gdb and friends.
During compiling I have to make some minor changes due to compiler warnings
like:
+#ifndef FUTEX_CMD_MASK
#define FUTEX_CMD_MASK ~(FUTEX_PRIVATE_FLAG | FUTEX_CLOCK_REALTIME)
-
+#endif
But I think these does not affect my tests.
Yes, this is okay and I think with --disable-werror it wont even be
needed (but I'm not sure).
As it is "serial" in name file I tried to remove "serial" device from
machines:
<serial type="pty">
<source path="/dev/pts/7"/>
<target type="isa-serial" port="0">
<model name="isa-serial"/>
</target>
<alias name="serial0"/>
</serial>
So this is just one serial port, not 3 or 4. Interesting. I think this
is this part in your command line:
-chardev pty,id=charserial0
-device
{"driver":"isa-serial","chardev":"charserial0","id":"serial0","index":0}
Ok, I don't have vista image here, lemme try to find one and see if
I can reproduce it myself.
Thank you for doing all this! You took much harder route but you ended
in the right place.
/mjt
