netdev
[Top] [All Lists]

Re: zerocopy changes in 3c59x.c

To: <kuznet@xxxxxxxxxxxxx>
Subject: Re: zerocopy changes in 3c59x.c
From: Bogdan Costescu <bogdan.costescu@xxxxxxxxxxxxxxxxxxxxx>
Date: Fri, 26 Jan 2001 19:04:45 +0100 (CET)
Cc: Andrew Morton <andrewm@xxxxxxxxxx>, <netdev@xxxxxxxxxxx>
In-reply-to: <200101261746.UAA27262@xxxxxxxxxxxxx>
Sender: owner-netdev@xxxxxxxxxxx
On Fri, 26 Jan 2001 kuznet@xxxxxxxxxxxxx wrote:

> > Five second transmit timeout.  Why is this?
> I would ask: why 400msec? 8)

That was the original value in Don's older drivers. I think that it was
computed based on packet transmission time, inter-packet gap and so on.

But there is a difference: Don didn't have Tx mitigation enabled by
default, so a packet going through would trigger an interrupt which would
clear tbusy and tx_full. With Tx mitigation, for a filled-up ring and
collisions, the interrupt might be delayed past Tx timeout value.

> BTW it really helped to avoid watchdog, when card had some spurious stalls
> due to excessive collisions or those link problems.

Andrew has enabled Tx mitigation by default to reduce interrupt rate
under load. See above.

Sincerely,

Bogdan Costescu

IWR - Interdisziplinaeres Zentrum fuer Wissenschaftliches Rechnen
Universitaet Heidelberg, INF 368, D-69120 Heidelberg, GERMANY
Telephone: +49 6221 54 8869, Telefax: +49 6221 54 8868
E-mail: Bogdan.Costescu@xxxxxxxxxxxxxxxxxxxxx


<Prev in Thread] Current Thread [Next in Thread>