On 2/26/19 9:14 PM, Stephen Hemminger wrote:
I noticed an anomaly in how receive interrupt control is done.
In l3fwd-power there is a per-port lock around calls to enable interrupts
but no locking around the call to disable interrupts. This looks broken
since intr_disable requires multiple operations on some hardware.
There are three better options:
1) do locking in rte_eth_dev (not application)
2) push any necessary locking down into the driver
3) lock on both places in l3fwd-power (and other usages)
I am thinking #1 is easiest solution.
Which ever solution is done, the documentation should be as well.
I think #2 is a way to go - some drivers require locking, some not.
I think it is a bad idea to enforce locking for drivers which do not
require it.
Andrew.
