Subject: RE: de0/pciide0 performance conflict?
To: 'tls@rek.tjls.com' <tls@rek.tjls.com>
From: Adam Glass (Exchange) <adamg@Exchange.Microsoft.com>
List: port-i386
Date: 10/05/1999 09:20:51
thus the "theoretically".  For hot plug pci i've seen proposals to just set
it to a fixed number and shrug.

my general concern with screwing with pci parameters is that perhaps the
bios may know something we don't.  This particular parameter is pretty safe
but some of the others may be turned on/off based on specific knowledge of
buggy hardware, interoperability concerns, etc.

-----Original Message-----
From: Thor Lancelot Simon [mailto:tls@rek.tjls.com]
Sent: Monday, October 04, 1999 11:05 PM
To: Adam Glass (Exchange)
Cc: port-i386@netbsd.org
Subject: Re: de0/pciide0 performance conflict?


On Mon, Oct 04, 1999 at 10:29:52PM -0700, Adam Glass (Exchange) wrote:
> 
> Generally this is not a parameter that an OS should screw with.
> Theoretically these parameters are balanced by the bios to reflect a
latency
> appropriate to the number/type devices in the system.  No MS OS modifies
> this parameter.  You'll get some performance this way but you might starve
> something else.

It'd be nice if the BIOS actually "balanced" the requirements of the
system's devices to do this, but I have yet to see that happen.  After
we ran into the problem with the AlphaPC 164SX machines that didn't set
this value *at all* I took a look at what all the x86 boxes I could
find did.

They all seemed to just pick either 32 or 64 for all devices.  Some of
them were nice enough to let *me* set the value they'd use for every
device, some weren't.  Considering the simplistic approach the BIOS
vendors have taken here, and the existence of broken firmware that
doesn't get it right at all, I don't think it would be unreasonable
for the OS to try to do a better job, or at least provide a user
override of what the firmware did.