Re: [Intel-gfx] [PATCH 2/5] kernel.h: Add non_block_start/end()

[Jason Gunthorpe]

On Fri, Aug 16, 2019 at 06:36:52PM +0200, Daniel Vetter wrote:
> On Fri, Aug 16, 2019 at 4:38 PM Jason Gunthorpe <jgg@xxxxxxxx> wrote:
> >
> > On Fri, Aug 16, 2019 at 04:11:34PM +0200, Daniel Vetter wrote:
> > > Also, aside from this patch (which is prep for the next) and some
> > > simple reordering conflicts they're all independent. So if there's no
> > > way to paint this bikeshed here (technicolor perhaps?) then I'd like
> > > to get at least the others considered.
> >
> > Sure, I think for conflict avoidance reasons I'm probably taking
> > mmu_notifier stuff via hmm.git, so:
> >
> > - Andrew had a minor remark on #1, I am ambivalent and would take it
> >   as-is. Your decision if you want to respin.
> 
> I like mine better, see also the reply from Ralph Campbell.

Sure

> > - #2/#3 is this issue, I would stand by the preempt_disable/etc path
> >   Our situation matches yours, debug tests run lockdep/etc.
>
> Since Michal requested the current flavour I think we need spin a bit
> more on these here. I guess I'll just rebase them to the end so
> they're not holding up the others.
> 
> > - #4 I like a lot, except the map should enclose range_end too,
> >   this can be done after the mm_has_notifiers inside the
> >   __mmu_notifier function
> 
> To make sure I get this right: The same lockdep context, but also
> wrapped around invalidate_range_end? 

Yes, the locking context of _range_start and _range_end should be
identical, last time I checked callers this was the case.

So, just add it to __mmu_notifier_invalidate_range_end() outside the
SRCU as there is no reason to burden debug kernel callers twice when
mmu notifiers are not enabled

Jason