Re: [Xen-devel] [PATCH 3 of 6] x86/mm/shadow: fix potential p2m/paging deadlock when emulating page table writes

[Tim Deegan <tim@xxxxxxx>]

At 09:25 -0700 on 18 Apr (1334741158), Andres Lagar-Cavilla wrote:
> > Nack, at least for now; we spent a fair amount of effort taking all this
> > emulation code out from under the shadow lock; serializing it under the
> > p2m lock would be unfortunate.  The other patches are less worrying,
> > since they wrap a shadow_lock() in a p2m_lock() but I hope they can all
> > be avoided.
> >
> > The existing interlock between the shadow code and the p2m code prevents
> > any p2m modifications from happening when the shadow lock is held, so I
> > hope I can solve this by switching to unlocked lookups instead.  I'm
> > building a test kernel now to tell me exactly which lookps are to
> > blame.
> 
> FWIW, get_gfn_query for the target gfn of the PTE entry that is being
> checked in validate_gl?e entry, is the call that causes the panic this
> patch tries to fix.
> 
> As for the other two patches, sh_resync_l1 is the trigger (again,
> get_gfn_query on the gfn that is contained by the PTE being verified)

Thanks.  I've taken a sweep through mm/shadow, making the lookups into
unlocked ones wherever the paging_lock is already held.  With the
attached patch and your final patch to enable the locking, I can start
HVM guests without crashing Xen.  I haven't given it any more serious
testing yet.

Cheers,

Tim.

Attachment: shadow-vs-p2m
Description: Text document

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