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

[George Dunlap <George.Dunlap@xxxxxxxxxxxxx>]

Yes, I have to apologize: I have a queue of PoD, p2m, and ept-related
fixes that I haven't pushed to the list because:
* they require non-negligible reworking
* it's been really difficult for me to set up an OSS-based system to test them

It actually turns out that doing locking in ept_get_entry() is the
wrong thing to do anyway; it can cause the following deadlock:

p2m_change_type [grabs p2m lock] -> set_p2m_entry -> ept_set_entry ->
ept_set_middle_level -> p2m_alloc [grabs hap lock]

write cr4 -> hap_update_paging_modes [grabes hap lock] ->
hap_update_cr3 -> gfn_to_mfn -> ept_get_entry -> [grabs p2m lock]

Attached is a ported patch that removes locking in ept_get_entry(),
and implements access-once semantics for reading and writing.  This
solves the original problem (a race between reading and writing the
table) without causing deadlocks.  I haven't had a chance to test it
-- can you give it a spin?

Thanks,
 -George

On Tue, Dec 14, 2010 at 8:39 AM, Jan Beulich <JBeulich@xxxxxxxxxx> wrote:
> George,
>
> with ept_get_entry() calling ept_pod_check_and_populate(), which in
> turn unconditionally calls p2m_lock(), we got a report of the BUG() in
> p2m_lock() triggering (in a backport of the patch on top of 4.0.1 - the
> logic seems unchanged in -unstable, and hence the issue ought to
> similarly exist there). Wouldn't therefore ept_pod_check_and_populate(),
> only ever called from ept_get_entry(), not need to do any locking on
> its own at all?
>
> Thanks, Jan
>
>
> _______________________________________________
> Xen-devel mailing list
> Xen-devel@xxxxxxxxxxxxxxxxxxx
> http://lists.xensource.com/xen-devel
>

ept-access-once.diff
Description: Text Data

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