On Fri, Oct 11, 2019 at 12:54 PM Wei Wang <wei...@google.com> wrote:
> Hmm... Yes... I would think a per-CPU input cache should work for the
> case above.
> Another idea is: instead of calling dst_dev_put() in rt_cache_route()
> to switch out the dev, we call, rt_add_uncached_list() to add this
> obsolete dst cache to the uncached list. And if the device gets
> unregistered, rt_flush_dev() takes care of all dst entries in the
> uncached list. I think that would work too.
>
> diff --git a/net/ipv4/route.c b/net/ipv4/route.c
> index dc1f510a7c81..ee618d4234ce 100644
> --- a/net/ipv4/route.c
> +++ b/net/ipv4/route.c
> @@ -1482,7 +1482,7 @@ static bool rt_cache_route(struct fib_nh_common
> *nhc, struct rtable *rt)
> prev = cmpxchg(p, orig, rt);
> if (prev == orig) {
> if (orig) {
> - dst_dev_put(&orig->dst);
> + rt_add_uncached_list(orig);
> dst_release(&orig->dst);
> }
> } else {
>
Thanks Wei for your work on this issue,
Any chance this patch will make it into 5.4?
