[dpdk-users] Is rte_ring_dequeue_burst/rte_ring_enqueue_burst just memory address copy from enqueue side to dequeue side?

Huai-En Tseng the at csie.io
Thu Jun 6 05:31:17 CEST 2019


I am curious about rte_ring_dequeue and rte_ring_enqueue.

I notice in rte_ring.h, the dequeue and enqueue mechanism, DEQUEUE_PTRS and ENQUEUE_PTRS marcos, both are just memory address assignment.

At the same time, I get a strange bug in my project.

Thread 1 continually enqueue 1 packet every time. 
Thread 2 also continually dequeue dequeue packets whatever it can and does some logic operation.

However, when thread 2 does some operation taking too long time such as printf and file io or thread 1 enqueue enqueue packet twice in very short time interval.

The packet thread 2 dequeue will be the next packet content which thread 1 enqueue.

I assume that despite thread 2 dequeue 1 packet, but it actually dequeue the memory address.

Then the cons and prod pointer do dequeue operation and thread 1 is ready to enqueue next packet to the memory address which thread 2 just dequeue.

However, at the same time, thread 2 is using this memory address and it is the root cause.

Is my assumption correct or something I miss?

If my assumption is correct, is there any method can avoid this except using memcpy?

Thanks all.

More information about the users mailing list