On Thu, Dec 01, 2016 at 07:15:01PM +0100, Borislav Petkov wrote:
> On Thu, Dec 01, 2016 at 11:02:04AM -0500, Yazen Ghannam wrote:
> > A deferred error is an uncorrectable error whose handling can be
> > deferred, i.e. it's not urgent. This affects the system behavior, but
> > I'm now thinking that this shouldn't affect users' behavior. I think it
> > would be simpler to just classify deferred errors as uncorrectable
> > errors so that users treat them as such.
> Why would we want to lie about deferred errors being uncorrectable?
They are uncorrectable errors that can be handled differently. If you
can't handle them then there's not much difference.
> And I believe deferred errors can be handled differently like freeze the
> process using the page instead of killing it. And so on...
If deferred errors can be handled differently in userspace, then you're
right we should maintain the distinction. I was thinking we'd only
handle them in the kernel.
> Why aren't you simply adding the documentation about
> HW_EVENT_ERR_DEFERRED and be done with it? The downstream path like
> tracepoint and all can handle all that just fine.
Okay, will do.
> > Boris,
> > Can we drop or revert commit d12a969ebbfc?
> No can do. It is a public branch and there's no touching it.
Okay, got it.