Thanks for reporting this!
I think we should allow the illusion that ovsrcu_init is a function, and it’s
parameters are evaluated before the body. Otherwise problems like this are
bound to be reintroduced. I’ll post a patch soon.
Jarno
On Jun 2, 2014, at 3:34 AM, YAMAMOTO Takashi <yamam...@valinux.co.jp> wrote:
> In the case of ovs-atomic-pthreads, the previous coding is expanded
> into successive atomic_lock__ calls which ends up with deadlock.
>
> Signed-off-by: YAMAMOTO Takashi <yamam...@valinux.co.jp>
> ---
> lib/cmap.c | 4 +++-
> 1 file changed, 3 insertions(+), 1 deletion(-)
>
> diff --git a/lib/cmap.c b/lib/cmap.c
> index a760235..7892d50 100644
> --- a/lib/cmap.c
> +++ b/lib/cmap.c
> @@ -690,7 +690,9 @@ cmap_replace__(struct cmap_impl *impl, struct cmap_node
> *node,
> replacement = cmap_node_next_protected(node);
> } else {
> /* 'replacement' takes the position of 'node' in the list. */
> - ovsrcu_init(&replacement->next, cmap_node_next_protected(node));
> + struct cmap_node *next = cmap_node_next_protected(node);
> +
> + ovsrcu_init(&replacement->next, next);
> }
>
> struct cmap_node *iter = &b->nodes[slot];
> --
> 1.8.3.1
>
> _______________________________________________
> dev mailing list
> dev@openvswitch.org
> http://openvswitch.org/mailman/listinfo/dev
_______________________________________________
dev mailing list
dev@openvswitch.org
http://openvswitch.org/mailman/listinfo/dev
