Re: [Xen-devel] regression from c/s 22071:c5aed2e049bc (ept: Put locks around ept_get_entry) ?

[George Dunlap <George.Dunlap@xxxxxxxxxxxxx>]

Sorry for the delay in responding; I caught a bad cold, and was in no
state to comment on making lockless races benign. :-)

On Fri, Dec 17, 2010 at 11:15 AM, Tim Deegan <Tim.Deegan@xxxxxxxxxx> wrote:
> George, I think the underlying logic is still racy - the
> check-and-populate function is checking a pointer that was found outside
> the lock.  It needs to start again from the beginning to be safe, which
> probably means just dropping the "check" part and letting the
> p2m_pod_demand_populate handle lost races.  Also, why doesn't
> ept_get_entry use a single read at the lowest level?

My main goal when I wrote this patch was to fix a bug our testing
found that was going to slip a release, and then move on to other
pressing issues.  So I'm sure there's more raciness to clean up in
here that just isn't triggered generally.  I think you're right,
dropping the check and doing it in p2m_pod_demand_populate() is
probably the Right Thing to do.  I've got that stack of patches to
deal with in January, I'll work on it then.

 -George Dunlap

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-devel