Re: [Xen-devel] Early ACPI events prevent subsequent ACPI functionality on xen 4.3 + HVM domU

[Dave Scott <Dave.Scott@xxxxxxxxxxxxx>]

Hi,

> On Tue, 2013-05-21 at 15:34 +0100, George Dunlap wrote:
> > On 05/21/2013 03:20 PM, Ian Campbell wrote:
> > > On Tue, 2013-05-21 at 15:16 +0100, George Dunlap wrote:
> > >> And will the Citrix PV drivers check for the existence of shutdown
> > >> commands before they start a watch?
> > >
> > > Watches fire immediately when you register them, to allow this race
> > > to be easily avoided without really having to think about it.
> >
> > Well then that watch mechanism is broken,
> 
> The firing happens in xenstored, and that certainly works everywhere else,
> or all sorts of things would be broken.

IIRC the watch event should appear in the xenstore (access?) log. I agree that 
it's very unlikely that xenstore is broken.
 
> That's not to say there isn't a problem with e.g. the kernels watch event
> handling causing the watch which is fired to not get as far as the appropriate
> handler.

For a long time our test cases have had a 'sleep 10' after a 'xe vm-start' and 
before a 'xe vm-shutdown'. IIRC when I last looked into this (years ago) it was 
claimed that the PV kernel was trying to signal init to change runlevel, but it 
was too soon and the event/signal/whatever  was dropped. Recently to speed up 
the tests we've started switching to a custom miniOS kernel (based on Mirage) 
which doesn't have this problem. However it would be great to finally fix this 
race...

Cheers,
Dave
_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
http://lists.xen.org/xen-devel