Re: Avanti network card config

["Brian E.W. Wood" <beww intac com>]

On 11-May-98 Maciej W. Rozycki wrote:
> 
>  I had similar symptoms one day.  Their reason was the CPU losing
> interrupts.  In fact, the PCI-ISA bridge (if that tells you anything...)
> had the PCI to ISA IRQ routing broken.
> 
>  As of modules -- they work flawlessly in 2.1.x without any extra steps.
> AFAIK, there is a way to make them work perfectly with 2.0.x, but I never
> bothered to find out how.
> 
> 
>  I guess you are using a 2.0.x and a CONFIG_ALPHA_SRM_SETUP configuration. 
> The problem is currently fixed in the 2.1.x series.  You may choose one of
> the following solutions: 
> 
> 1. Disable CONFIG_ALPHA_SRM_SETUP.  You won't be able to halt/reboot
> cleanly.  It will be OK from Linux's side: you will just have to type
> 'init' within SRM after halt.  And you will see somewhat annoying
> behaviour of reboot.  (Well, I suppose you observe this behaviour anyway,
> as there is a small omission present in 2.0.33, which prevents
> CONFIG_ALPHA_SRM_SETUP from having a positive impact on Avanti machines.) 
> ;-) 
> 
> 2. Wait for 2.0.34 be released by Alan.  Hopefully Jay will survive an
> extra overload caused by the need of preparing patches for Alpha/2.0.34
> and they will finally appear on 'gatekeeper'.
> 
> 3. Use the current 2.1.x release -- 2.1.101 works for me but YMMV.

Right now I'd settle for just finding out how avanti-s was compiled (what
patches etc.). It works fine but needs to have a few more things compiled in.

I don't know how long 2.0.34 will be and I don't know if it is a good idea to
give my boss a developmental kernel, although I will probably try out 2.1.x at
home. Does 2.1.101 work on an SX machine? I heard that the
2.1.x series didn't work on the SX, but that was some time ago. Does 2.1.101
need patching for the Alpha? I think I heard it did not.

OTOH, if the vendors have some really great deals at LinuxExpo who knows what I
may come home with :-)

> 
>> Clearly the address is the problem. Again I wish I knew what RedHat did
>> to get their kernel to work. I really hate it when they do something and
>> do not document it or make the source they used available, Isn't this
>> what the GPL is supposed to prevent? 
> 
>  Apparently, you did not use the RH's kernel but the one from
> 'gatekeeper'.  Or did RH apply that patch finally?

I'm pretty sure I got it from RH. I also recall a mention on this list that the
"-s" was from the "clever gnomes at RH" and meant it was for use with SRM.

The kernels I compile that do not work properly are from 2.0.33 sources with
alpha-patches-2.0.32 applied. Are there other patches I should have applied?
 
> 
>> I'm also booting from the SRM console and the SRM PCI support is
>> labelled as "new". I may try blowing ARC into the flash and booting from
>> that although I don't see why that should matter since my other PCI
>> stuff like scsi works fine. 
> 
>  I wouldn't do that.  ARC is not so nice as SRM and also it prints that
> nasty "Copyright Microsoft...".  SCSI works by accident -- the IRQ route
> happens to be correct for this slot.
> 
>> Thanks for the note, I think the 0x10100 address you reported is
>> significant to me. 
> 
>  It just hints that SRM PCI configuration is in use -- Linux does not set
> up devices in such a way because, the I/O space on Intels is limited to 64
> kB.
> 
>  BTW, cc to axp-list as others may be interested, too.

I usually do, but I've put a lot of traffic about this problem on the list and
I figured I would wait 'till I get it figured out and then post a summary -
more or less the way the Sun-Managers and the Alpha-OSF lists work. I probably
wouldn't have made that crack about the GPL in a list message :-)

Much thanks for your input.

---
Brian E.W. Wood  beww@intac.com