[vortex] Re: vortex 3c982, 3C90x etc

[Bogdan Costescu]

On Tue, 13 Aug 2002, Thomas Mierau wrote:

> This is one part of a hole coversation that you had
> http://www.geocrawler.com/archives/3/433/2002/1/0/7568423/

That thread had dealt with 2 different problems: missed interrupts and
probably a race in the Tx part of the driver. However, your computer seems
to have another problem, generating too many interrupts. So I think that 
they cannot compare.

> In fact the multiple serial card would never show up in the
> proc/interrupts.

That's something strange...

> It is a PCI card thogh and of course uses an irq. But there is no
> difference in this card beeing installed or not, or sharing the IRQ or
> not.

In the listing that you attached, it does share an interrupt with dpti0 
and eth0. Do you observe any strange behaviour with eth0 as well ?

> Very simple, when the raid systems starts with some intensive work like
> backup of the data base, the errors on the IRQ for the eth0 increases.
> Some how the etho is loosing data packages.

Probably dpti0 processes interrupts for too long. That's a know cause for
getting "Too much work in interrupt" messages and loosing packets. You
could try to dissociate dpti0 and eth0 from the same interrupt; check the
BIOS if it allows association between PCI slot and IRQ line or move the
card to another PCI slot.

>   5: 2603035427 2603245593   IO-APIC-level  eth1

That's certainly too much...
>           RX packets:126196 errors:0 dropped:0 overruns:0 frame:0
>           TX packets:63016 errors:0 dropped:0 overruns:0 carrier:0
>           RX bytes:12366600 (11.7 Mb)  TX bytes:6164576 (5.8 Mb)

... for such low data volumes.

> The blocks of 5 return at the same time , as the time differnce is
> always 1sec off look pretty nasty

Do you run by any chance eth3 with default 5 seconds comit interval ?

I don't believe that the NICs or the driver are responsible for such 
behaviour. Most likely there's an APIC misbehaviour or something strange 
with the PCI bus.

-- 
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@IWR.Uni-Heidelberg.De