Linux/AMD Bug - solutions?

Adcadet

Storage Freak
Joined
Jan 14, 2002
Messages
1,861
Location
44.8, -91.5
Does anybody know a good source that explains how to work around the Linux2.4/AMD Athlon/AGP bug?

Seems Mandrake 8.2 won't install on my Tyan Tiger MP system (crashes right after configuring X), and I'm wondering if it's because of that. Any thoughts?
 

Mercutio

Fatwah on Western Digital
Joined
Jan 17, 2002
Messages
21,562
Location
I am omnipresent
Slashdot.org had a couple of 500-reply articles on that very subject. I haven't gone back and read them yet, but I'll bet if you take the time, you'll see something useful.

kuro5shin is another place to look. Usually there are more technical comments there.
 

The JoJo

Wannabe Storage Freak
Joined
Jan 25, 2002
Messages
1,490
Location
Finland, Turku
Website
www.thejojo.com
I put the following line into lilo.conf :
append="mem=nopentium"

And then in my bios lowered agp4x to 2x.

I haven't read too much about this, so If anyone comes upon any more useful/better ideas, please post here.
 

Mercutio

Fatwah on Western Digital
Joined
Jan 17, 2002
Messages
21,562
Location
I am omnipresent
Is AGP @ 2x really a problem that needs solving? Most people don't play games or run modeling apps on Linux machines, as AFAIK those are the things that actually make use of AGP's enhanced speed.

Anyway, I just started reading what I could on slashdot. Maybe I'll see something else.

AD, you don't have an SB live on that Tyan board, do you? I recall that being a source of problems with AMD dualies under linux.
 

The JoJo

Wannabe Storage Freak
Joined
Jan 25, 2002
Messages
1,490
Location
Finland, Turku
Website
www.thejojo.com
I've been using my system now for a while and it seems that the few freezes I experienced before are gone...
Not sure if it's "the" bug or not that caused them though.
 

LiamC

Storage Is My Life
Joined
Feb 7, 2002
Messages
2,016
Location
Canberra
The JoJo,

you don't need to lower your AGP speed because the bug has no relevance to the AGP speed setting.

The bug only impacts when your GART is using 4MB tables instead of 4KB tables, *and* if bit 21 is set high, *and* the software does an invalid page table buffer flush.

Only since Pentium class CPU's has there been an option to use 4MB pages instead of 4KB - which is what the "nopentium" switch is doing - forcing the use of 4KB pages exclusively. Performance hit is *slight*
 
Top