Re: [PATCH bpf-next v3 2/6] mm, bpf: Introduce free_pages_nolock()

[Alexei Starovoitov <alexei.starovoitov@gmail.com>]

On Wed, Dec 18, 2024 at 3:32 AM Michal Hocko <mhocko@suse.com> wrote:
>
> On Tue 17-12-24 19:07:15, alexei.starovoitov@gmail.com wrote:
> > From: Alexei Starovoitov <ast@kernel.org>
> >
> > Introduce free_pages_nolock() that can free pages without taking locks.
> > It relies on trylock and can be called from any context.
> > Since spin_trylock() cannot be used in RT from hard IRQ or NMI
> > it uses lockless link list to stash the pages which will be freed
> > by subsequent free_pages() from good context.
>
> Yes, this makes sense. Have you tried a simpler implementation that
> would just queue on the lockless link list unconditionally? That would
> certainly reduce the complexity. Essentially something similar that we
> do in vfree_atomic (well, except the queue_work which is likely too
> heavy for the usecase and potentialy not reentrant).

We cannot use llist approach unconditionally.
One of the ways bpf maps are used is non-stop alloc/free.
We cannot delay the free part. When memory is free it's better to
be available for kernel and bpf uses right away.
llist is the last resort.