Re: [PATCH] padata: Fix possible divide-by-0 panic in padata_mt_helper()

[Waiman Long]

On 8/10/24 21:45, Herbert Xu wrote:
> On Sat, Aug 10, 2024 at 09:30:25PM -0400, Waiman Long wrote:
> > On 8/10/24 00:05, Herbert Xu wrote:
> > > Waiman Long <longman@xxxxxxxxxx> wrote:
> > > > diff --git a/kernel/padata.c b/kernel/padata.c
> > > > index 53f4bc912712..0fa6c2895460 100644
> > > > --- a/kernel/padata.c
> > > > +++ b/kernel/padata.c
> > > > @@ -517,6 +517,13 @@ void __init padata_do_multithreaded(struct padata_mt_job *job)
> > > >          ps.chunk_size = max(ps.chunk_size, job->min_chunk);
> > > >          ps.chunk_size = roundup(ps.chunk_size, job->align);
> > > >
> > > > +       /*
> > > > +        * chunk_size can be 0 if the caller sets min_chunk to 0. So force it
> > > > +        * to at least 1 to prevent divide-by-0 panic in padata_mt_helper().`
> > > > +        */
> > > > +       if (!ps.chunk_size)
> > > > +               ps.chunk_size = 1U;
> > > Perhaps change the first ps.chunk_size assignment to use DIV_ROUND_UP
> > > instead?
> > I think DIV_ROUND_UP() will exactly the same problem that if chunk_size is
> > 0, you still got a 0 result. round_up() only if the 2nd argument is a power
> > of 2 while with DIV_ROUND_UP(), the second argument can be any number except
> > 0.
> Unless I'm missing something chunk_size cannot be zero before the
> division because that's the first thing we check upon entry into
> this function.

chunk_size is initialized as

ps.chunk_size = job->size / (ps.nworks * load_balance_factor);

chunk_size will be 0 if job->size < (ps.nworks * load_balance_factor). 
If min_chunk is 0, chunk_size will remain 0.

After looking at the dump file when the crash happen at 
padata_mt_helper(). I had determined that ps->chunk_size was indeed 0 
which caused the divide-by-0 panic. I actually got 2 different bug 
reports of this div-by-0 panic, one with a debug and another one with a 
non-debug kernel.

Cheers,
Longman