Re: [syzbot] [mm?] WARNING in mas_nomem

[Liam R. Howlett]

On 26/08/07 04:26PM, Jason Gunthorpe wrote:
> On Fri, Aug 07, 2026 at 03:24:24PM -0400, Liam R. Howlett wrote:
> > On 26/08/07 09:19AM, Liam R. Howlett wrote:
> > > On 26/08/06 12:48PM, Jason Gunthorpe wrote:
> > > > On Thu, Aug 06, 2026 at 10:29:27AM -0400, Liam R. Howlett wrote:
> > > > 
> > > > > @@ -5609,8 +5612,7 @@ void *mas_store(struct ma_state *mas, void *entry)
> > > > >  		goto store;
> > > > >  
> > > > >  	mas_alloc_nodes(mas, GFP_NOWAIT);
> > > > > -	if (mas_is_err(mas))
> > > > > -		return NULL;
> > > > > +	mas_nomem(mas, GFP_KERNEL | __GFP_NOFAIL);
> > > > 
> > > > It still feels really weird to ignore the return code from mas_nomem..
> > > > 
> > > >  void mas_nomem_nofail() 
> > > 
> > > Sure, that's fair.
> > > 
> > > > 
> > > > ? and maybe it can have a WARN_ON
> > > 
> > > 
> > > The warning would never trigger.  If the allocation fails with nofail,
> > > then you'd never reach the warning.
> > 
> > Upon rereading my reply, I don't think it was clear what's going to
> > happen here.
> > 
> > If there's really no memory and no reserves, then we're going to panic
> > on trying to allocate.  So, even if I rewrite this to have a void
> > function to print a warning, it will never trigger, at that point why
> > have another function?  Are you looking to document what's happening in
> > the code with this?  If so, maybe a comment would be better.
> 
> Yeah, it is clearer to document it with a function that can't be
> called wrong, IMHO

Okay, I'll figure out how to do that later.  I'd like to get this in for
the next weeks -next testing as we're getting late in the cycle.

Cheers,
Liam