[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Xen-devel] [PATCH for-4.12 V4] x86/altp2m: fix HVMOP_altp2m_set_domain_state race

On 2/15/19 3:37 PM, George Dunlap wrote:

On Feb 15, 2019, at 1:24 PM, Jan Beulich <JBeulich@xxxxxxxx> wrote:

On 15.02.19 at 13:52, <George.Dunlap@xxxxxxxxxx> wrote:
On Feb 12, 2019, at 11:42 AM, Razvan Cojocaru <rcojocaru@xxxxxxxxxxxxxxx> wrote:

HVMOP_altp2m_set_domain_state does not domain_pause(), presumably
on purpose (as it was originally supposed to cater to a in-guest
agent, and a domain pausing itself is not a good idea).

Sorry to come in here on v4 and suggest changing everything, but I don’t
really like the solution you have here.  Not setting altp2m to ‘active’ until
after the vcpus are set up makes sense; but passing this true/false value in
seems ugly, and still seems a bit racy (i.e., what if p2m_active() is
disabled between the check in HVMOP_altp2m_switch_p2m and the time we
actually call altp2m_vcpu_update_p2m()?)

I certainly don’t think domain_pause() should be our go-to solution for race
avoidance, but in this case it really seems like switching the global p2m for
every vcpu at once makes the most sense; and trying to safely change this on
an unpaused domain is not only overly complicated, but probably not what we
wanted anyway.

p2m_altp2m_destroy_by_id() and p2m_switch_domain_altp2m_by_id() already use
domain_pause_except_self(); so it seems like not using it for
altp2m_set_domain_state was probably more of an oversight than an intentional
decision.  Using that here seems like a more robust solution.

Ah, I didn't even recall there was such a function. As this now
also allows covering a domain requesting the operation for itself,
I don't mind the pausing approach anymore.

Yeah, I forgot too until I was grepping for “domain_pause” to figure out what 
everyone else was doing. :-)

The one issue is that domain_pause_except_self() currently is actually a
deadlock risk if two different vcpus start it at the same time.  I think the
attached patch (compile-tested only) should fix this issue; after this patch
you should be able to use domain_pause_except_self() in
altp2m_set_domain_state instead.

There's one thing I don't really like here, which is a result of the
(necessary) re-use of the hypercall deadlock mutex: This
certainly poses the risk of getting called from a context where
the lock was already acquired. Therefore I'd like to suggest to
use this lock in a recursive way (here and elsewhere).

And two cosmetic remarks - there's no need to re-specify
__must_check on the function definition, as the function
declaration ought to be in scope anyway. And there's a stray
blank inside the likely() you add.

I don’t see that I added a ‘likely’; there’s one in context, but I don’t see 
any stray blanks there.

The other two points make sense — Razvan, would you be willing to make those changes 
(and test the result, as I haven’t done more than compile-test it)?

Of course, happy to. Just to make sure I understand where we stand: I'll try to leave the mutex alone for now and only switch to a recursive one if anything blows up.


Xen-devel mailing list



Lists.xenproject.org is hosted with RackSpace, monitoring our
servers 24x7x365 and backed by RackSpace's Fanatical Support®.