[Spice-devel] Fix red_worker threads consuming 100% cpu when a client is connected

[Hans de Goede]

Hi,

On 03/10/2012 08:05 PM, Hans de Goede wrote:
> Hi All,
>
> This patch series fixes an issue where red_worker thread consume as much CPU
> as they can get when a client is connected (the interesting part explaining
> what is going on is in the 4th patch.
>
> Unfortunately with this series I can reproducably crash the vm (qemu segfaults)
> when using multiple heads. I've been debugging the crash, it is deep inside
> the glz_encoding code, and I believe that the cause is a glz encoding bug,
> and this patch set simply changes some timing in such a way that the bug
> triggers. More on this in my next mail.

Let me correct myself on this, while typing my next mail I've found the cause
of this multihead crash and the problem is not inside the glz encoder,
but with how red_worker creates the encoder (both display channels get the
same encoder id, and then bad things start happening). To be continued ...

Regards,

Hans