On Sat, 02 Jun 2018, Herbert Xu wrote:

On Fri, Jun 01, 2018 at 09:53:47AM -0700, Davidlohr Bueso wrote:

quoted

Curious, are these users setting up the param structure dynamically
or something that they can pass along bogus values?

If that's the case then yes, I definitely agree.

It's just a quality of implementation issue.  This is a generic API.
Sure for early-boot users like yours it makes sense to just WARN_ON
rather than deal with the messy hash table allocation failure.

But for a driver author writing some kernel module it isn't nice
to WARN_ON and then crash on a NULL-pointer dereference when we
can cleanly fail the table init.

Fine, at least patch 2 applies without this one. So Andrew, if you
consider taking this series please drop patch 1 and 5 (which no
longer makes sense as rhashtable_init() won't be returning void
in the future).

If you want me to resend (assuming other issues are not pointed out),
I can do but I wanted to avoid spamming more the necessary.

Thanks,
Davidlohr