Re: [PATCH v4] kmemleak: survive in a low-memory situation

[Michal Hocko]

On Wed 27-03-19 07:34:32, Qian Cai wrote:
> On 3/27/19 4:44 AM, Michal Hocko wrote:
> >> diff --git a/mm/kmemleak.c b/mm/kmemleak.c
> >> index a2d894d3de07..7f4545ab1f84 100644
> >> --- a/mm/kmemleak.c
> >> +++ b/mm/kmemleak.c
> >> @@ -580,7 +580,16 @@ static struct kmemleak_object *create_object(unsigned long ptr, size_t size,
> >>  	struct rb_node **link, *rb_parent;
> >>  	unsigned long untagged_ptr;
> >>  
> >> -	object = kmem_cache_alloc(object_cache, gfp_kmemleak_mask(gfp));
> >> +	/*
> >> +	 * The tracked memory was allocated successful, if the kmemleak object
> >> +	 * failed to allocate for some reasons, it ends up with the whole
> >> +	 * kmemleak disabled, so try it harder.
> >> +	 */
> >> +	gfp = (in_atomic() || irqs_disabled()) ?
> >> +	       gfp_kmemleak_mask(gfp) | GFP_ATOMIC :
> >> +	       gfp_kmemleak_mask(gfp) | __GFP_DIRECT_RECLAIM;
> > 
> > 
> > The comment for in_atomic says:
> >  * 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.
> 
> That is why it needs both in_atomic() and irqs_disabled(), so irqs_disabled()
> can detect kernel functions held spinlocks even in non-preemptible kernels.
> 
> According to [1],
> 
> "This [2] is useful if you know that the data in question is only ever
> manipulated from a "process context", ie no interrupts involved."
> 
> Since kmemleak only deal with kernel context, if a spinlock was held, it always
> has local interrupt disabled.

What? Normal spin lock implementation doesn't disable interrupts. So
either I misunderstand what you are saying or you seem to be confused.
the thing is that in_atomic relies on preempt_count to work properly and
if you have CONFIG_PREEMPT_COUNT=n then you simply never know whether
preemption is disabled so you do not know that a spin_lock is held.
irqs_disabled on the other hand checks whether arch specific flag for
IRQs handling is set (or cleared). So you would only catch irq safe spin
locks with the above check.

> ftrace is in the same boat where this commit was merged a while back that has
> the same check.
> 
> ef99b88b16be
> tracing: Handle ftrace_dump() atomic context in graph_trace_open()
> 
> [1] https://www.kernel.org/doc/Documentation/locking/spinlocks.txt
> [2]
> 	spin_lock(&lock);
> 	...
> 	spin_unlock(&lock);

-- 
Michal Hocko
SUSE Labs