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

Re: [Xen-devel] [PATCH] mm: fix LLVM code-generation issue

On Fri, Nov 23, 2018 at 11:39:32AM +0000, Julien Grall wrote:
> Hi Jan,
> On 23/11/2018 11:33, Jan Beulich wrote:
> > > > > On 23.11.18 at 12:23, <roger.pau@xxxxxxxxxx> wrote:
> > > Image an AP reads system_state < SYS_STATE_active, then BSP sets
> > > system_state = SYS_STATE_active and clears the init mappings, then
> > > when the AP tries to read an init variable it would get a page fault.
> > 
> > As said before - APs are supposed to be quiescent during this
> > particular transition of system_state, and the idle scrubbing added
> > an exception to that rule. I don't see why such an exception may
> > not involve exceptional treatment on the particular reading path.
> > Furthermore APs, which may also be brought up at run time, may
> > never ever touch init mappings of any kind.
> The "supposed" is quite worrying because we have no way to verify that
> assumption until we get a random crash.
> Also, the cost of the barrier should be quite cheap over the rest of the 
> function.

Right, but unless I'm missing something that won't completely fix the
issue either?

The only way I see to be completely safe is to either use a mutex or
to make opt_bootscrub non-init.


Xen-devel mailing list



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