Re: [PATCH] mm: use in_atomic() in print_vma_addr()

[Michal Hocko]

On Mon 06-11-17 11:43:54, Michal Hocko wrote:
> On Mon 06-11-17 11:05:58, Peter Zijlstra wrote:
> > On Sun, Nov 05, 2017 at 09:19:46AM +0100, Michal Hocko wrote:
> > > [CC Peter]
> > > 
> > > On Fri 03-11-17 20:09:49, Bart Van Assche wrote:
> > > > On Fri, 2017-11-03 at 11:02 -0700, Andrew Morton wrote:
> > > > > Also, checkpatch says
> > > > > 
> > > > > WARNING: use of in_atomic() is incorrect outside core kernel code
> > > > > #43: FILE: mm/memory.c:4491:
> > > > > +       if (in_atomic())
> > > > > 
> > > > > I don't recall why we did that, but perhaps this should be revisited?
> > > > 
> > > > Is the comment above in_atomic() still up-to-date? From <linux/preempt.h>:
> > > > 
> > > > /*
> > > >  * Are we running in atomic context?  WARNING: this macro cannot
> > > >  * always detect atomic context; in particular, it cannot know about
> > > >  * held spinlocks in non-preemptible kernels.  Thus it should not be
> > > >  * used in the general case to determine whether sleeping is possible.
> > > >  * Do not use in_atomic() in driver code.
> > > >  */
> > > > #define in_atomic()	(preempt_count() != 0)
> > > 
> > > I can still see preempt_disable NOOP for !CONFIG_PREEMPT_COUNT kernels
> > > which makes me think this is still a valid comment.
> > 
> > Yes the comment is very much accurate.
> 
> Which suggests that print_vma_addr might be problematic, right?
> Shouldn't we do trylock on mmap_sem instead?

I might be missing something but the check seems to be broken. The
original commit by Ingo e8bff74afbdb ("x86: fix "BUG: sleeping function
called from invalid context" in print_vma_addr()") relied on elevated
preempt count by preempt_conditional_sti which is gone for quite some
time. First replaced by explicit preempt_disable in d99e1bd175f4
("x86/entry/traps: Refactor preemption and interrupt flag handling").

So unless I am missing something this check doesn't work and we should
rather do the trylock thing.
-- 
Michal Hocko
SUSE Labs