[PATCH v2 2/9] bpf: document ethdev callback threading limitations

[Stephen Hemminger]

On Sat, 24 Oct 2026 15:10:13 +0100
Marat Khalili <qm2k21 at gmail.com> wrote:

> On 19/10/2026 22:10, Konstantin Ananyev wrote:
> >> Document the data plane limitation in the file docstring of
> >> rte_bpf_ethdev.h: packet callback execution expects only one
> >> thread per port/queue pair at a time, making concurrent bursts
> >> (such as with RTE_ETH_TX_OFFLOAD_MT_LOCKFREE, shared RX queues,
> >> or multi-threaded software PMDs) unsupported.  
> > Not sure what do you mean by:
> > 'shared RX queues' and 'multi-threaded software PMDs'. Can you 
> > probably clarify?  
> 
> I mean any hypothetical solutions that call rx/tx burst concurrently 
> from multiple threads or processes. To avoid confusion, will just leave 
> only an RTE_ETH_TX_OFFLOAD_MT_LOCKFREE example in the next version.

Other than TX lockfree DPDK is not safe to use rx/tx burst from
multiple threads on the same queue. The queue either needs to be
bound to a thread, or locking in the application. Many drivers
would break if Rx was called concurrently on multiple threads.
This design is intentional and one of the reasons DPDK is fast.