Re: [PATCH 1/4] mm: Check if mmu notifier callbacks are allowed to fail

From: Jason Gunthorpe
Date: Wed Jun 19 2019 - 16:47:55 EST


On Wed, Jun 19, 2019 at 10:18:43PM +0200, Daniel Vetter wrote:
> On Wed, Jun 19, 2019 at 10:13 PM Jason Gunthorpe <jgg@xxxxxxxx> wrote:
> > On Wed, Jun 19, 2019 at 09:57:15PM +0200, Daniel Vetter wrote:
> > > On Wed, Jun 19, 2019 at 6:50 PM Jason Gunthorpe <jgg@xxxxxxxx> wrote:
> > > > On Tue, Jun 18, 2019 at 05:22:15PM +0200, Daniel Vetter wrote:
> > > > > On Tue, May 21, 2019 at 11:44:11AM -0400, Jerome Glisse wrote:
> > > > > > On Mon, May 20, 2019 at 11:39:42PM +0200, Daniel Vetter wrote:
> > > > > > > Just a bit of paranoia, since if we start pushing this deep into
> > > > > > > callchains it's hard to spot all places where an mmu notifier
> > > > > > > implementation might fail when it's not allowed to.
> > > > > > >
> > > > > > > Inspired by some confusion we had discussing i915 mmu notifiers and
> > > > > > > whether we could use the newly-introduced return value to handle some
> > > > > > > corner cases. Until we realized that these are only for when a task
> > > > > > > has been killed by the oom reaper.
> > > > > > >
> > > > > > > An alternative approach would be to split the callback into two
> > > > > > > versions, one with the int return value, and the other with void
> > > > > > > return value like in older kernels. But that's a lot more churn for
> > > > > > > fairly little gain I think.
> > > > > > >
> > > > > > > Summary from the m-l discussion on why we want something at warning
> > > > > > > level: This allows automated tooling in CI to catch bugs without
> > > > > > > humans having to look at everything. If we just upgrade the existing
> > > > > > > pr_info to a pr_warn, then we'll have false positives. And as-is, no
> > > > > > > one will ever spot the problem since it's lost in the massive amounts
> > > > > > > of overall dmesg noise.
> > > > > > >
> > > > > > > v2: Drop the full WARN_ON backtrace in favour of just a pr_warn for
> > > > > > > the problematic case (Michal Hocko).
> > > >
> > > > I disagree with this v2 note, the WARN_ON/WARN will trigger checkers
> > > > like syzkaller to report a bug, while a random pr_warn probably will
> > > > not.
> > > >
> > > > I do agree the backtrace is not useful here, but we don't have a
> > > > warn-no-backtrace version..
> > > >
> > > > IMHO, kernel/driver bugs should always be reported by WARN &
> > > > friends. We never expect to see the print, so why do we care how big
> > > > it is?
> > > >
> > > > Also note that WARN integrates an unlikely() into it so the codegen is
> > > > automatically a bit more optimal that the if & pr_warn combination.
> > >
> > > Where do you make a difference between a WARN without backtrace and a
> > > pr_warn? They're both dumped at the same log-level ...
> >
> > WARN panics the kernel when you set
> >
> > /proc/sys/kernel/panic_on_warn
> >
> > So auto testing tools can set that and get a clean detection that the
> > kernel has failed the test in some way.
> >
> > Otherwise you are left with frail/ugly grepping of dmesg.
>
> Hm right.
>
> Anyway, I'm happy to repaint the bikeshed in any color that's desired,
> if that helps with landing it. WARN_WITHOUT_BACKTRACE might take a bit
> longer (need to find a bit of time, plus it'll definitely attract more
> comments).

I was actually just writing something very similar when looking at the
hmm things..

Also, is the test backwards?

mmu_notifier_range_blockable() == true means the callback must return
zero

mmu_notififer_range_blockable() == false means the callback can return
0 or -EAGAIN.

Suggest this:

pr_info("%pS callback failed with %d in %sblockable context.\n",
mn->ops->invalidate_range_start, _ret,
!mmu_notifier_range_blockable(range) ? "non-" : "");
+ WARN_ON(mmu_notifier_range_blockable(range) ||
+ _ret != -EAGAIN);
ret = _ret;
}
}

To express the API invariant.

Jason