On 21 July 2017 at 14:44, Herbert Xu <herbert@xxxxxxxxxxxxxxxxxxx> wrote: > On Fri, Jul 21, 2017 at 02:42:20PM +0100, Ard Biesheuvel wrote: >> >> >> - Would you mind a patch that makes the code only use the per-CPU >> >> buffers if we are running atomically to begin with? >> > >> > That would mean dropping the first packet so no. >> > >> >> I think you misunderstood me: the idea is not to allocate the per-CPU >> buffers at that time, but avoiding them and use the heap directly (as >> my patch does now). This way, we don't unnecessarily disable >> preemption for calls executing in process context. > > When we actually have a user other than IPcomp then we could > try this optimisation. Otherwise I think it'd be a waste of > time. > Fair enough.
