On 12/13/2012 08:56 AM, Hans de Goede wrote:
Hi,
On 12/08/2012 09:47 AM, Cole Robinson wrote:
> Hi all,
>
> I've pushed qemu 1.3 to the fedora git repo, but haven't pushed a build to
> rawhide yet
Cool to see 1.3 coming to rawhide. And I see that you've already forward
ported the chardev flowcontrol patches needed for reliable spice agent /
usbredir operation it will need the usual chardev flowcontrol patches,
thanks!
While you mention it, anyone on virt-devel care to chime in with long term
plans here? Has there been any movement on Anthony's 'proper' solution?
At the least, it would help out quite a bit if the first two patches of the
series were merged:
http://pkgs.fedoraproject.org/cgit/qemu.git/tree/0100-char-Split-out-tcp-...
http://pkgs.fedoraproject.org/cgit/qemu.git/tree/0101-char-Add-a-QemuChrH...
Note that I usually have a branch with them forward ported long
before
the Fedora packages get rebased, so you can save yourself this tedious
job by taking them from a branch here:
http://cgit.freedesktop.org/~jwrdegoede/qemu/log/?h=qemu-1.3-usbredir-wip
You will want the patches starting at:
"char: Split out tcp socket close code in a separate function"
and ending at:
"hw/virtio-serial-bus: replay guest open on destination"
Thanks Hans, I'll use that going forward.
Note that I once more add to plan some usb-core + usb-redir patches
to the Fedora packages for some USB goodness. Currently pending
there is:
-Some more performance tweaks for ehci giving another 20% performance
boost reading from usb mass storage devices
-Throughput improvements for uhci (1000 urbs / sec instead if 500 in
cases where the guest uses only 1 urb).
-Buffered bulk input support. This decouples qemu and the real usb-dev
adding fifo in between for bulk in endpoints on certain devices (like
we already do for isoc endpoints). This makes USB <-> serial adapters
work reliable even when we hit some latency spikes / issues in qemu
(which we regularly do). Without this received serial data may get lost
due to the small fifo these devices have internally overrunning.
Note the intends of this is a heads-up about me planning to bring these
upstream improvements to the Fedora qemu-1.3 packages is to allow people
to object. If I receive no objections I will move forward with this when
I find some time for it :)
It's one thing to backport feature/enhancement bits as part of an accepted
fedora 'feature', but IMO it should not be a regular occurrence. Packaging
churn, update churn, and risk for not a lot of gain.
Also in this particular instance, qemu 1.3 is only going to exist in
rawhide/virt-preview, so the audience is pretty small. Nowdays there's only 2
months between qemu GA and qemu+1 RC0, so the window of usefulness is small
anyways.
That last bit about usb serial sounds like it results in a bug fix, but if
it's sufficiently non trivial that qemu-stable wouldn't take it, no one is
actively complaining about the bug, and it's not required for a fedora
feature, then skip the backport.
Feel free to change my mind about any of that :)
Backporting isolated bug fixes is always fine (though if you do, please also
send a pull request to qemu-stable).
Thanks,
Cole