[Top][All Lists]
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
[Qemu-devel] qxl screendump: how and why it is broken
From: |
Alon Levy |
Subject: |
[Qemu-devel] qxl screendump: how and why it is broken |
Date: |
Tue, 6 Mar 2012 19:21:12 +0200 |
User-agent: |
Mutt/1.5.21 (2010-09-15) |
..and why we got to this situation.
(per Anthony's request, and if you'll excuse the puny attempt to write
biblically, a new beginning)
In the begining there was screendump. It was a synchronous monitor
command. Libvirt said: I want a screendump. And qemu waited for the vga
device to produce it. When the vga device said it was done, libvirt was
done.
Then there came qxl. Qxl used a spice library to render the contents. It
would say to libvirt: wait, while I read on this thread, which will tell
me I am done. And so it was possible to hang the libvirt process:
tl;dr if spice client hangs libvirt screenshot hangs and qemu's vcpus starve.
http://lists.gnu.org/archive/html/qemu-devel/2012-03/msg00864.html
Then a new patchset landed, which stopped the read, by waiting for a
callback from the spice server. It is here:
http://patchwork.ozlabs.org/patch/142985/
commited as 81fb6f1504fb9ef71f2382f44af34756668296e8
But now screendump for qxl is broken: It produces a file that is old,
because it calls ppm_save before spice server is done rendering.
Now we come to now. My understanding from the previous thread is:
* monitor async is not good because Luiz said so.
(yes, I know I'm putting my head in the sand here)
If the above is enough to convince you of $SUBJECT, then the
screendump-async patchset is still waiting review :)
http://patchwork.ozlabs.org/patch/144704/
http://patchwork.ozlabs.org/patch/144705/
http://patchwork.ozlabs.org/patch/144706/
Alon
[Prev in Thread] |
Current Thread |
[Next in Thread] |
- [Qemu-devel] qxl screendump: how and why it is broken,
Alon Levy <=