Re: [Bug 1882817] [NEW] Segfault in audio_pcm_sw_write with audio over VNC

[Philippe Mathieu-Daudé]

On 6/9/20 7:10 PM, Artyom wrote:
> Public bug reported:
> 
> QEMU 5.0.0, built with ./configure --target-list=x86_64-softmmu
> --enable-debug --disable-strip --disable-docs --disable-sdl
> 
> Running on a headless host (Ryzen 3600), Arch Linux, 64bit latest.
> Guest is also Arch Linux, 64bit.
> 
> Started with qemu-system-x86_64 -vnc 0.0.0.0:0 -enable-kvm -m 4096 -cpu host 
> -smp cores=2,threads=1,sockets=1 -machine q35 -vga std -device
>  ich9-ahci,id=ahci -drive file=vm0.qcow2,format=qcow2,if=none,id=dsk0 -device 
> ide-hd,drive=dsk0,bus=ahci.0 -soundhw hda
> 
> So, a headless VM is running on a server and is being connected to over
> VNC. The virtual sound card is detected and speaker test is running
> inside the VM. So far so good.
> 
> Then, i tell the VNC client to enable audio (QEMU Audio Client Message,
> 255,1,0). QEMU responds with a "stream is about to start" message (QEMU
> Audio Server Message, 255,1,1) and then promptly crashes without sending
> anything else.
> 
> Running it in GDB produces a crash at audio/audio.c:739
> 
> Thread 1 "qemu-system-x86" received signal SIGSEGV, Segmentation fault.
> audio_pcm_sw_write (sw=0x5555575bbf30, buf=0x0, size=1628) at 
> audio/audio.c:739
> 739             if (!sw->hw->pcm_ops->volume_out) {

Isn't it the one fixed by
https://www.mail-archive.com/qemu-devel@nongnu.org/msg705896.html?

> 
> The exact sequence of events does not matter - i can enable sound before
> playing anything, and then it would say nothing and keep working, but
> crash with the same message once anything sound-playing is launched in
> the VM.
> 
> Using different soundhw or adding various audiodev options does not seem
> to affect anything.
> 
> I can't quite figure out if the QEMU Audio VNC extension is supposed to
> work at all or not, but it would be handy to me if it is.
> 
> ** Affects: qemu
>      Importance: Undecided
>          Status: New
>