[ofa-general] Re: [Bug, PATCH and another Bug] Was: Fix refcounting problem with netif_rx_reschedule()

[Krishna Kumar2]

Hi Dave,

David Miller <davem at davemloft.net> wrote on 09/19/2007 09:35:57 PM:

> The NAPI_STATE_SCHED flag bit should provide all of the necessary
> synchornization.
>
> Only the setter of that bit should add the NAPI instance to the
> polling list.
>
> The polling loop runs atomically on the cpu where the NAPI instance
> got added to the per-cpu polling list.  And therefore decisions to
> complete NAPI are serialized too.
>
> That serialized completion decision is also when the list deletion
> occurs.

About the "list deletion occurs", isn't the race I mentioned still present?
If done < budget, the driver does netif_rx_complete (at which time some
other
cpu can add this NAPI to their list). But the first cpu might do some more
actions on the napi, like ipoib_poll() calls request_notify_cq(priv->cq),
when other cpu might have started using this napi.

(net_rx_action's 'list_move' however will not execute since work != weight)

Thanks,

- KK