On 6/18/2020 3:29 PM, Eric Dumazet wrote:
On Thu, Jun 18, 2020 at 2:21 PM Amritha Nambiar
<amritha.namb...@intel.com> wrote:

This will be useful to allow busy poll for tunneled traffic. In case of
busy poll for sessions over tunnels, the underlying physical device's
queues need to be polled.

Tunnels schedule NAPI either via netif_rx() for backlog queue or
schedule the gro_cell_poll(). netif_rx() propagates the valid skb->napi_id
to the socket. OTOH, gro_cell_poll() stamps the skb->napi_id again by
calling skb_mark_napi_id() with the tunnel NAPI which is not a busy poll
candidate.


Yes the tunnel NAPI id should be 0 really (look for NAPI_STATE_NO_BUSY_POLL)

This was preventing tunneled traffic to use busy poll. A valid
NAPI ID in the skb indicates it was already marked for busy poll by a
NAPI driver and hence needs to be copied into the socket.

Signed-off-by: Amritha Nambiar <amritha.namb...@intel.com>
---
  include/net/busy_poll.h |    6 +++++-
  1 file changed, 5 insertions(+), 1 deletion(-)

diff --git a/include/net/busy_poll.h b/include/net/busy_poll.h
index 86e028388bad..b001fa91c14e 100644
--- a/include/net/busy_poll.h
+++ b/include/net/busy_poll.h
@@ -114,7 +114,11 @@ static inline void skb_mark_napi_id(struct sk_buff *skb,
                                     struct napi_struct *napi)
  {
  #ifdef CONFIG_NET_RX_BUSY_POLL
-       skb->napi_id = napi->napi_id;
+       /* If the skb was already marked with a valid NAPI ID, avoid overwriting
+        * it.
+        */
+       if (skb->napi_id < MIN_NAPI_ID)
+               skb->napi_id = napi->napi_id;


It should be faster to not depend on MIN_NAPI_ID (aka NR_CPUS + 1)

     if (napi->napi_id)
        skb->napi_id = napi->napi_id;

Probably not a big deal.

Reviewed-by: Eric Dumazet <eduma...@google.com>


Thanks for the review. Should I send a v2 with this change?







  #endif
  }


Reply via email to