Hello,
Fedora has carried flow control patches for qemu chardevs since a very long time now. They weren't upstream, and they were required for spice, usb-redir to work properly and not freeze the guest.
Upstream qemu has now merged an alternative implementation of the same concept, and most of the bugs have been shaken out upstream (we don't know of any known bugs now). So I'm of the view we should replace the older patches in F19's qemu with the backport of the newer, upstream ones.
Is this fine? I can work with Cole / Hans who have been keeping the older patches uptodate on Fedora to replace the patches.
Thanks,
Amit
Hi,
On 04/23/2013 07:57 AM, Amit Shah wrote:
Hello,
Fedora has carried flow control patches for qemu chardevs since a very long time now. They weren't upstream, and they were required for spice, usb-redir to work properly and not freeze the guest.
Upstream qemu has now merged an alternative implementation of the same concept, and most of the bugs have been shaken out upstream (we don't know of any known bugs now). So I'm of the view we should replace the older patches in F19's qemu with the backport of the newer, upstream ones.
Is this fine?
This is fine by me.
Regards,
Hans
On 04/23/2013 01:57 AM, Amit Shah wrote:
Hello,
Fedora has carried flow control patches for qemu chardevs since a very long time now. They weren't upstream, and they were required for spice, usb-redir to work properly and not freeze the guest.
Upstream qemu has now merged an alternative implementation of the same concept, and most of the bugs have been shaken out upstream (we don't know of any known bugs now). So I'm of the view we should replace the older patches in F19's qemu with the backport of the newer, upstream ones.
Is this fine? I can work with Cole / Hans who have been keeping the older patches uptodate on Fedora to replace the patches.
If the backport is fairly clean and self contained it should be okay, but can't say I'm excited about it. What's the benefit besides more testing of the new code? Yes we get to drop the old patches but at this point they are a long tested known quantity. If it was a month ago I'd be less resistant but this will be past the alpha.
I'll leave it to your discretion but if the backport is pulling in unrelated patches so things apply cleanly, or needs non-trivial alteration to work on 1.4, we should re-evaluate.
Thanks, Cole
On (Tue) 23 Apr 2013 [10:32:56], Cole Robinson wrote:
On 04/23/2013 01:57 AM, Amit Shah wrote:
Hello,
Fedora has carried flow control patches for qemu chardevs since a very long time now. They weren't upstream, and they were required for spice, usb-redir to work properly and not freeze the guest.
Upstream qemu has now merged an alternative implementation of the same concept, and most of the bugs have been shaken out upstream (we don't know of any known bugs now). So I'm of the view we should replace the older patches in F19's qemu with the backport of the newer, upstream ones.
Is this fine? I can work with Cole / Hans who have been keeping the older patches uptodate on Fedora to replace the patches.
If the backport is fairly clean and self contained it should be okay, but can't say I'm excited about it. What's the benefit besides more testing of the new code? Yes we get to drop the old patches but at this point they are a long tested known quantity. If it was a month ago I'd be less resistant but this will be past the alpha.
Yes, I was concerned about the timing too.
I'll leave it to your discretion but if the backport is pulling in unrelated patches so things apply cleanly, or needs non-trivial alteration to work on 1.4, we should re-evaluate.
I haven't tried the backport yet.
Since people haven't been finding bugs in the older code, it's obviously safer to just keep it and drop all the patches in rawhide once we rebase to qemu-1.5 (1.5 will contain all the new upstream work). OTOH getting testing for the new code will be a big positive.
From all indications, we've ironed out the bugs in the most-used
areas. There might be some lurking in corner cases, and we do have some time till the final release.
Looking at the current state, I am inclined to push the upstream changes since we've been rigorously testing for regressions (and everything looks settled now).
Amit
On (Tue) 23 Apr 2013 [20:41:03], Amit Shah wrote:
On (Tue) 23 Apr 2013 [10:32:56], Cole Robinson wrote:
On 04/23/2013 01:57 AM, Amit Shah wrote:
Hello,
Fedora has carried flow control patches for qemu chardevs since a very long time now. They weren't upstream, and they were required for spice, usb-redir to work properly and not freeze the guest.
Upstream qemu has now merged an alternative implementation of the same concept, and most of the bugs have been shaken out upstream (we don't know of any known bugs now). So I'm of the view we should replace the older patches in F19's qemu with the backport of the newer, upstream ones.
Is this fine? I can work with Cole / Hans who have been keeping the older patches uptodate on Fedora to replace the patches.
If the backport is fairly clean and self contained it should be okay, but can't say I'm excited about it. What's the benefit besides more testing of the new code? Yes we get to drop the old patches but at this point they are a long tested known quantity. If it was a month ago I'd be less resistant but this will be past the alpha.
Yes, I was concerned about the timing too.
Cole mentioned on IRC about rawhide virt-preview, which I agree will be a good idea to get this code into users' hands. So let's drop the backport idea for now -- I'm quite relieved :)
Amit
On Tue, Apr 23, 2013 at 11:27:09AM +0530, Amit Shah wrote:
Hello,
Fedora has carried flow control patches for qemu chardevs since a very long time now. They weren't upstream, and they were required for spice, usb-redir to work properly and not freeze the guest.
Upstream qemu has now merged an alternative implementation of the same concept, and most of the bugs have been shaken out upstream (we don't know of any known bugs now). So I'm of the view we should replace the older patches in F19's qemu with the backport of the newer, upstream ones.
Is this fine? I can work with Cole / Hans who have been keeping the older patches uptodate on Fedora to replace the patches.
Would this affect virtio-serial (ie. libguestfs)?
Rich.
On (Sat) 27 Apr 2013 [21:15:54], Richard W.M. Jones wrote:
On Tue, Apr 23, 2013 at 11:27:09AM +0530, Amit Shah wrote:
Hello,
Fedora has carried flow control patches for qemu chardevs since a very long time now. They weren't upstream, and they were required for spice, usb-redir to work properly and not freeze the guest.
Upstream qemu has now merged an alternative implementation of the same concept, and most of the bugs have been shaken out upstream (we don't know of any known bugs now). So I'm of the view we should replace the older patches in F19's qemu with the backport of the newer, upstream ones.
Is this fine? I can work with Cole / Hans who have been keeping the older patches uptodate on Fedora to replace the patches.
Would this affect virtio-serial (ie. libguestfs)?
Yes, it affects everything that uses chardevs, including virtio-serial. Without the flow control patches, guest writes to virtio-serial ports could freeze the guest. We've carried non-upstream patches in Fedora for a long time now.
Amit
-
Amit Shah -
Cole Robinson -
Hans de Goede -
Richard W.M. Jones