[vortex] Problems with 3c59x

Bogdan Costescu bogdan.costescu@iwr.uni-heidelberg.de
Sat Dec 14 06:02:01 2002


On Sat, 14 Dec 2002, Lists (lst) wrote:

> I don't understand this. Be more explicit and tell me what to do when the 
> card hangs again.

Just before that line there should have been one with a "status" value. 
This value presumably has the lowest significant bit set, of a probable 
form of 0xeXX1.

> Why the card loosing the interrupt? Is this a kernel bug, a driver bug or 
> a mainboard bug?

It's not the card, it's the whole system. The card generates the signal, 
but the interrupt routine from the driver is not called. The generation of 
signal from the card is a state kept until explicitly cleared, so it can 
be checked whenever other driver code is executed.
So there is no driver bug; but between a kernel and mainboard bug, I can't 
tell... There have been problems with some older kernel versions, that 
were supposed to be fixed some time ago. Also being on the linux-smp list 
I frequently got messages about problems with the SMP VIA chipset that 
were solved by upgrading BIOS or changing the BIOS MPS level.

> This option will afect my SMP box?

The difference is in handling of interrupts: noapic disables the IO-APIC, 
which makes interrupts function in XT-PIC mode which is the mode used 
on uniprocessor machines. This will make interrupt requests only be 
delivered to one CPU which will also be the only CPU servicing the 
interrupt routines from drivers. In some cases this is a good thing as it 
provides better cache locality, while for a system which receives very 
many interrupts, it might degrade performance.

-- 
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