Stefano Brivio <sbrivio@xxxxxxxxxx> wrote:
> > - fill_map = kcalloc(m->bsize_max, sizeof(*res_map), GFP_ATOMIC);
> > + fill_map = kcalloc(m->bsize_max, sizeof(*res_map), GFP_KERNEL);
>
> I haven't re-checked the whole logic, but can't nft_pipapo_deactivate()
> (hence pipapo_deactivate() and pipapo_get()) be called from the data
> path for some reason?
Not that I know. Deactivate turns off an element in the next generation
and that concept is tied to the transaction, and that needs the nft
mutex.
> If I recall correctly that's why I used GFP_ATOMIC here, but I'm not
> sure anymore and I guess you know better.
I'll have a look at original version to see if there was a reason
for this, if so I'll update commit message or move this to its own
change.
> > + if (src->rules > 0) {
> > + dst->mt = kvmalloc_array(src->rules, sizeof(*src->mt), GFP_KERNEL);
>
> Nit: equally readable within 80 columns:
>
> dst->mt = kvmalloc_array(src->rules, sizeof(*src->mt),
> GFP_KERNEL);
OK. I'll reformat.
