On Wed, 2012-05-02 at 12:46 -0700, Andrew Morton wrote:
> On Wed,  2 May 2012 13:28:09 +0900
> Minchan Kim <minchan@kernel.org> wrote:
> 
> > Now there are several places to use __vmalloc with GFP_ATOMIC,
> > GFP_NOIO, GFP_NOFS but unfortunately __vmalloc calls map_vm_area
> > which calls alloc_pages with GFP_KERNEL to allocate page tables.
> > It means it's possible to happen deadlock.
> > I don't know why it doesn't have reported until now.
> > 
> > Firstly, I tried passing gfp_t to lower functions to support __vmalloc
> > with such flags but other mm guys don't want and decided that
> > all of caller should be fixed.
> > 
> > http://marc.info/?l=linux-kernel&m=133517143616544&w=2
> > 
> > To begin with, let's listen other's opinion whether they can fix it
> > by other approach without calling __vmalloc with such flags.
> > 
> > So this patch adds warning in __vmalloc_node_range to detect it and
> > to be fixed hopely. __vmalloc_node_range isn't random chocie because
> > all caller which has gfp_mask of map_vm_area use it through __vmalloc_area_node.
> > And __vmalloc_area_node is current static function and is called by only
> > __vmalloc_node_range. So warning in __vmalloc_node_range would cover all
> > vmalloc functions which have gfp_t argument.
> >
> > I Cced related maintainers.
> > If I miss someone, please Cced them.
> > 
> > --- a/mm/vmalloc.c
> > +++ b/mm/vmalloc.c
> > @@ -1648,6 +1648,10 @@ void *__vmalloc_node_range(unsigned long size, unsigned long align,
> >  	void *addr;
> >  	unsigned long real_size = size;
> >  
> > +	WARN_ON_ONCE(!(gfp_mask & __GFP_WAIT) ||
> > +			!(gfp_mask & __GFP_IO) ||
> > +			!(gfp_mask & __GFP_FS));
> > +
> >  	size = PAGE_ALIGN(size);
> >  	if (!size || (size >> PAGE_SHIFT) > totalram_pages)
> >  		goto fail;
> 
> Well.  What are we actually doing here?  Causing the kernel to spew a
> warning due to known-buggy callsites, so that users will report the
> warnings, eventually goading maintainers into fixing their stuff.
> 
> This isn't very efficient :(
> 
> It would be better to fix that stuff first, then add the warning to
> prevent reoccurrences.  Yes, maintainers are very naughty and probably
> do need cattle prods^W^W warnings to motivate them to fix stuff, but we
> should first make an effort to get these things fixed without
> irritating and alarming our users.  
> 
> Where are these offending callsites?

OK, I checked my part - both UBI and UBIFS call __vmalloc() with
GFP_NOFS in several places of the _debugging_ code, and this is why we
do not see any issues - the debugging code is used very rarely for
validating purposes. All the places look fixable, I'll fix them a bit
later.

WARN_ON_ONCE() looks like a good first step. An I think it is better if
maintainers fix their areas rather than if someone who does not know how
the subsystem works starts trying to do that.

-- 
Best Regards,
Artem Bityutskiy