[IPCop-user] Re: IRQs Suck

[Bryan J. S. <b.j...@ie...>]

John Edwards wrote:
> PCI cards should be able to share the same IRQ if they both play nice 
> and use interrupt sharing. There may be some performance loss on PCI 
> throughput but you probably will not notice that unless they are 
> Gigabit cards.

When you're talking Gigabit cards, I/O contention with other PCI cards 
on the same PCI bus is your biggest issue.  I.e., never put a Gigabit 
NIC and a disk controller on the same PCI bus.  Intel thanks got for 
ServerWorks everyday (i.e., Intel didn't have a multi-PCI bus chipset 
until it started licensing the designs from ServerWorks).

> Nope, they are talking about different things. Donald Becker is 
> discussing how PCI cards in great detail. The FAQ is noting that 
> some cards (usually ISA ones) are not automatically configured 
> or detected and so you need a setup utility.
> Just use a DOS boot disk, aka "Windows Emergency Repair Disk" from 
> Windows 95 or 98.

In case everyone doesn't know, Windows 95, 98 and ME _are_ MS-DOS 7.x. 
7.0 in 95 through 95A, 7.1 from 95B on-ward.  It's still good'ole 16-bit 
running _everything_ underneath -- usually with the 386+ CPU in 
Virtual386 mode using the DOS Protected Mode Interface (DPMI) APIs 
mapped into a subset of the Win32 API.


> So they are cheap, no name, no support. Is there anything to differentiate 
> them from a piece of junk ?
> First you'll need to find out chipset they use, which can sometimes be 
> written on the chips. Otherwise try querying the PCI info (Device Manager 
> in Windows or "cat /proc/pci" in Linux/IPCop).
> Secondly you'll need to check that there is a driver for that card in 
> the Linux kernel. If you still don't know you the chipset and autodetection 
> doesn't work then you could take a guess for the common cheapos - NE2000 
> or Realtek 8139.
> I've never had to fiddle with INT settings in the BIOS to get a network 
> card to work, and I've worked with some really wierd ones. 

I've never had to with Linux unless the system used AMI BIOS.  Then you 
have to make sure the IRQs are not shared.  AMI BIOS makes both NT-based 
Windows and Linux a serious PITA, and very DOS-based Windows-like when 
it comes to such things.

> My best recommendation is to return the cheap pieces of junk that are 
> being sold as PCI cards on grounds that they do not seem to following 
> the PCI specs. Then get a good brand name card such as 3com 905 or 
> Intel EtherExpress Pro. It will cost you about 20 more US dollars and 
> save you weeks of grief.

Bewary of some i82558 chips (early Intel EtherExpress Pro/100).  They 
had serious transmitter issues, although the e100 (Intel) as well as 
newer eepro100 (GPL) drivers have addressed it.

It's not that Intel products have bugs -- _all_ ICs do, you just can't 
test everything (e.g., modern CPUs typcially have 50-100 known bugs that 
are easily accomodated in software).  No, Intel has a nasty habit of not 
being "forthcoming" with its "errata."  It's large enough that it 
doesn't need to, from a "bean counter" standpoint.  ;-ppp

-- 
Bryan J. Smith, E.I.   b.j...@ie...   http://thebs.org
If you want the stupid letters:   http://thebs.org/certs.pdf
------------------------------------------------------------
Running Windows applications under Linux does not reduce any
political/legal "costs" but does increase the overall tech-
nical "costs."  Linux is not a better Windows than Windows.