Re: [Xen-devel] [PATCH] evtchn: make support for different ABIs tunable

[Eslam Elnikety <elnikety@xxxxxxxxxx>]

On 14.08.19 15:02, Andrew Cooper wrote:
> On 14/08/2019 13:51, George Dunlap wrote:
> > On 8/7/19 5:03 PM, Jan Beulich wrote:
> > > Whatever we do in Xen, it'll only allow to work around that issue.
> > > An actual fix belongs in the kernel(s). For this reason I suppose
> > > what we're talking about here is a feature (from Xen's pov), not a
> > > bug fix. And it being a feature, it should preferably be coded in
> > > a way that's usable also going forward.
> > FWIW, I agree with what I understand Jan to be saying:
> >
> > 1. It would be good to be able to disable FIFO event channels, but
> >
> > 2. Disabling them in Xen isn't really the right way to fix Amazon's
> > issue. Rather, probably the soft reboot should reset the event channel
> > state.
> Depends what you mean about "fix the issue".
>
> Amazon have real customer VMs in this situation which will break on a
> Xen old => new upgrade.  Modifying Xen is the only rational option.
>
> They are also doing this in an upstream compatible manner (which is
> great).  One way or another, Xen needs to gain this ability to work
> around broken-linux which is already in the field.
>
> As for the ideal way to fix this, this bug has existed in Linux longer
> than soft-reset has been a thing, and frankly, its still a gruesome
> hack.  We need some interfaces which don't suck so terribly.
>
> ~Andrew
Thanks, Andrew. I second those points.

I have just sent v3 of this patch, mostly addressing comments from Jan. 
Have a look, and let me know if there are further tweaks you would 
rather see. Thanks.
Cheers,
Eslam

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxxx
https://lists.xenproject.org/mailman/listinfo/xen-devel