Re: [PATCH 3/3] mm/mmu_gather: send tlb_remove_table_smp_sync IPI only to CPUs in kernel mode

[Peter Zijlstra]

On Thu, Apr 06, 2023 at 03:11:52PM +0100, Valentin Schneider wrote:
> On 06/04/23 15:38, Peter Zijlstra wrote:
> > On Wed, Apr 05, 2023 at 01:45:02PM +0100, Valentin Schneider wrote:
> >>
> >> I've been hacking on something like this (CSD deferral for NOHZ-full),
> >> and unfortunately this uses the CPU-local cfd_data storage thing, which
> >> means any further smp_call_function() from the same CPU to the same
> >> destination will spin on csd_lock_wait(), waiting for the target CPU to
> >> come out of userspace and flush the queue - and we've just spent extra
> >> effort into *not* disturbing it, so that'll take a while :(
> >
> > I'm not sure I buy into deferring stuff.. a NOHZ_FULL cpu might 'never'
> > come back. Queueing data just in case it does seems wasteful.
> 
> Putting those callbacks straight into the bin would make my life much
> easier!

Well, it's either they get inhibited at the source like the parent patch
does, or they go through. I really don't see a sane middle way here.

> Unfortunately, even if they really should, I don't believe all of the
> things being crammed onto NOHZ_FULL CPUs have the same definition of
> 'never' as we do :/

That's not entirely the point, the point is that there are proper
NOHZ_FULL users that won't return to the kernel until the machine shuts
down. Buffering stuff for them is more or less a direct memory leak.