Thanks Bhanu and Jan,

For this proposed simple fix, should the exact match flow key->len not equal to 
megaflow length from cr.flow.len?  Since one megaflow could match many exact 
match flows...

Thanks
Yipeng

>-----Original Message-----
>From: Jan Scheurich [mailto:[email protected]]
>Sent: Wednesday, February 21, 2018 12:09 AM
>To: Bodireddy, Bhanuprakash <[email protected]>; Wang, Yipeng1 
><[email protected]>;
>[email protected]
>Cc: Tai, Charlie <[email protected]>
>Subject: RE: [ovs-dev] [RFC 2/4] dpif-netdev: Fix EMC key length
>
>I think there is an even simpler fix for the problem:
>
>+        /* Found a match in DFC. Insert into EMC for subsequent lookups.
>+         * We use probabilistic insertion here so that mainly elephant
>+         * flows enter EMC.
>+         * Initialize the key's length from the matched flow. */
>+        key->len = flow->cr.flow.len;
>+        emc_probabilistic_insert(&cache->emc_cache, key, flow);
>+        *exact_match = false;
>+        return flow;
>+    } else {
>
>Please use that in the next version of 1/4.
>
>BR, Jan
>
>> -----Original Message-----
>> From: Bodireddy, Bhanuprakash [mailto:[email protected]]
>> Sent: Tuesday, 20 February, 2018 22:14
>> To: Wang, Yipeng1 <[email protected]>; [email protected]; Jan 
>> Scheurich <[email protected]>
>> Cc: Tai, Charlie <[email protected]>
>> Subject: RE: [ovs-dev] [RFC 2/4] dpif-netdev: Fix EMC key length
>>
>> This fix is needed and can  be included in 1/4 in next revision.
>>
>> - Bhanuprakash.
>>
>> >-----Original Message-----
>> >From: [email protected] [mailto:ovs-dev-
>> >[email protected]] On Behalf Of Yipeng Wang
>> >Sent: Thursday, January 18, 2018 6:20 PM
>> >To: [email protected]; [email protected]
>> >Cc: Tai, Charlie <[email protected]>
>> >Subject: [ovs-dev] [RFC 2/4] dpif-netdev: Fix EMC key length
>> >
>> >EMC's key length is not initialized when insertion. Initialize the key 
>> >length
>> >before insertion.
>> >
>> >The code might be put in another place, for now I just put it in dfc_lookup.
>> >
>> >Signed-off-by: Yipeng Wang <[email protected]>
>> >---
>> > lib/dpif-netdev.c | 3 ++-
>> > 1 file changed, 2 insertions(+), 1 deletion(-)
>> >
>> >diff --git a/lib/dpif-netdev.c b/lib/dpif-netdev.c index b9f4b6d..3e87992
>> >100644
>> >--- a/lib/dpif-netdev.c
>> >+++ b/lib/dpif-netdev.c
>> >@@ -2295,7 +2295,7 @@ dfc_insert(struct dp_netdev_pmd_thread *pmd,  }
>> >
>> > static inline struct dp_netdev_flow *
>> >-dfc_lookup(struct dfc_cache *cache, const struct netdev_flow_key *key,
>> >+dfc_lookup(struct dfc_cache *cache, struct netdev_flow_key *key,
>> >            bool *exact_match)
>> > {
>> >     struct dp_netdev_flow *flow;
>> >@@ -2317,6 +2317,7 @@ dfc_lookup(struct dfc_cache *cache, const struct
>> >netdev_flow_key *key,
>> >         /* Found a match in DFC. Insert into EMC for subsequent lookups.
>> >          * We use probabilistic insertion here so that mainly elephant
>> >          * flows enter EMC. */
>> >+        key->len = netdev_flow_key_size(miniflow_n_values(&key->mf));
>> >         emc_probabilistic_insert(&cache->emc_cache, key, flow);
>> >         *exact_match = false;
>> >         return flow;
>> >--
>> >2.7.4
>> >
>> >_______________________________________________
>> >dev mailing list
>> >[email protected]
>> >https://mail.openvswitch.org/mailman/listinfo/ovs-dev
_______________________________________________
dev mailing list
[email protected]
https://mail.openvswitch.org/mailman/listinfo/ovs-dev

Reply via email to