Stephen Hemminger <shemminger@xxxxxxxx> :
> The rx_frags statistic is a software statistic and not a hardware statistic
> and it is pasted on to the end. Since the statistic is really a warning about
> overlength frames, it shouldn't be in the ethtool stats.
>
> Francois, could you make sure 8169 and 8139cp report fragments the same way.
(not sure I understood the question)
Do you mean that the r8169 driver should provide something like
cp_rx_err_acct() ?
[...]
> diff -Nru a/drivers/net/8139cp.c b/drivers/net/8139cp.c
> --- a/drivers/net/8139cp.c 2005-03-16 14:55:11 -08:00
> +++ b/drivers/net/8139cp.c 2005-03-16 14:55:11 -08:00
[...]
> + /* we don't support incoming fragmented frames.
> + * instead, we attempt to ensure that the
> + * pre-allocated RX skbs are properly sized such
> + * that RX fragments are never encountered
> + */
> + if ((status & (FirstFrag | LastFrag)) != (FirstFrag | LastFrag)
> + || (status & (RxError | RxErrFIFO))) {
> cp_rx_err_acct(cp, rx_tail, status, len);
> goto rx_next;
> }
I'll verify how the 8169 behaves wrt the Rx fifo error indication. It can
apparently be set a bit too much (i.e. on every packet).
--
Ueimor
|