[dpdk-dev] Safe practices for adding to rte_table_hash_lru

[dpdk-dev] Safe practices for adding to rte_table_hash_lru

[Saha, Avik (AWS)]

Hey guys
   Is it safe to add an entry to the rte_table_hash while the pipeline is being run - for instance if I were to try and add an entry on a port reader action when the packet enters the pipeline?

Thanks
Avik

Re: [dpdk-dev] Safe practices for adding to rte_table_hash_lru

[Neil Horman]

On Wed, Sep 24, 2014 at 07:38:49PM +0000, Saha, Avik (AWS) wrote:
> Hey guys
>    Is it safe to add an entry to the rte_table_hash while the pipeline is being run - for instance if I were to try and add an entry on a port reader action when the packet enters the pipeline?
> 
> Thanks
> Avik
> 

No, none of the table implementations are currently thread safe.  They all
assume that you will do your own locking between threads and processes.  There
were a few patches to support thread safety recently, but they are being
reworked currently.

Neil

Re: [dpdk-dev] Safe practices for adding to rte_table_hash_lru

[Gray, Mark D]

> 
> Hey guys
>    Is it safe to add an entry to the rte_table_hash while the pipeline is being
> run - for instance if I were to try and add an entry on a port reader action
> when the packet enters the pipeline?

If the table is part of the pipeline, it is not safe. i.e. the table is not threadsafe.

You can use a control message ring to add entries to the table in an 
asynchronous manner. Your main dpdk loop would then periodically handle
these messages. There are examples of this in Intel DPDK vSwitch (https://github.com/01org/dpdk-ovs)
and also, I think, in the DPDK ip_pipeline sample application.

> 
> Thanks
> Avik