net driver error accounting

net driver error accounting

[Andrew Morton]

Looking at http://bugzilla.kernel.org/show_bug.cgi?id=8106

Guys, could we please have a ruling here?

When a net driver encounters a tx_fifo_error, should this also contribute
to the tx_error count, or should it not?

More generally, should netdev drivers accumulate all the detailed
rx_errors into net_device_stats.rx_errors in real time, or should they not?

Thanks.

Re: net driver error accounting

[Jeff Garzik]

Andrew Morton wrote:
> Looking at http://bugzilla.kernel.org/show_bug.cgi?id=8106
> 
> Guys, could we please have a ruling here?
> 
> When a net driver encounters a tx_fifo_error, should this also contribute
> to the tx_error count, or should it not?

For each TX error, (a) tx_error is incremented and (b) a more-specific 
TX error stat is also potentially incremented.  So, yes, tx_error 
accumulates.

See cp_tx() in 8139cp.


> More generally, should netdev drivers accumulate all the detailed
> rx_errors into net_device_stats.rx_errors in real time, or should they not?

For each RX error, (a) rx_error is incremented and (b) a more-specific 
RX error stat is also potentially incremented.  So, yes, rx_error 
accumulates.

See cp_rx_err_acct() in 8139cp.

	Jeff

Re: net driver error accounting

[Andrew Morton]

On Thu, 02 Aug 2007 16:38:09 -0400
Jeff Garzik <jeff@garzik.org> wrote:

> Andrew Morton wrote:
> > Looking at http://bugzilla.kernel.org/show_bug.cgi?id=8106
> > 
> > Guys, could we please have a ruling here?
> > 
> > When a net driver encounters a tx_fifo_error, should this also contribute
> > to the tx_error count, or should it not?
> 
> For each TX error, (a) tx_error is incremented and (b) a more-specific 
> TX error stat is also potentially incremented.  So, yes, tx_error 
> accumulates.
> 
> See cp_tx() in 8139cp.
> 
> 
> > More generally, should netdev drivers accumulate all the detailed
> > rx_errors into net_device_stats.rx_errors in real time, or should they not?
> 
> For each RX error, (a) rx_error is incremented and (b) a more-specific 
> RX error stat is also potentially incremented.  So, yes, rx_error 
> accumulates.
> 
> See cp_rx_err_acct() in 8139cp.
> 

OK, thanks.

One does wonder why the overall rx_error exists all all, but whatever.  The
main thing is to get all the net drivers doing the same thing.


So I guess bug 8106 wants something like this?


diff -puN drivers/net/natsemi.c~a drivers/net/natsemi.c
--- a/drivers/net/natsemi.c~a
+++ a/drivers/net/natsemi.c
@@ -2438,13 +2438,16 @@ static void netdev_error(struct net_devi
 				dev->name);
 		}
 		np->stats.rx_fifo_errors++;
+		np->stats.rx_errors++;
 	}
 	/* Hmmmmm, it's not clear how to recover from PCI faults. */
 	if (intr_status & IntrPCIErr) {
 		printk(KERN_NOTICE "%s: PCI error %#08x\n", dev->name,
 			intr_status & IntrPCIErr);
 		np->stats.tx_fifo_errors++;
+		np->stats.tx_errors++;
 		np->stats.rx_fifo_errors++;
+		np->stats.rx_errors++;
 	}
 	spin_unlock(&np->lock);
 }
_