On Sun, Apr 29, 2018 at 09:59:27AM -0700, Kees Cook wrote:
> Did this ever happen?

Not yet.  I brought it up at LSFMM, and I'll repost the patches soon.

> I'd also like to see kmalloc_array_3d() or
> something that takes three size arguments. We have a lot of this
> pattern too:
> 
> kmalloc(sizeof(foo) * A * B, gfp...)
> 
> And we could turn that into:
> 
> kmalloc_array_3d(sizeof(foo), A, B, gfp...)

Are either of A or B constant?  Because if so, we could just use
kmalloc_array.  If not, then kmalloc_array_3d becomes a little more
expensive than kmalloc_array because we have to do a divide at runtime
instead of compile-time.  that's still better than allocating too few
bytes, of course.

I'm wondering how far down the abc + ab + ac + bc + d rabbit-hole we're
going to end up going.  As far as we have to, I guess.

Reply via email to