Re: 2.1.111: IDE DMA disabled...BLAH...BLAH...

Wed, 29 Jul 1998 21:16:31 +0200 (CEST)

On Wed, 29 Jul 1998, Alan Cox wrote:

> Two people has 2.0 DMA problems, many people have 2.1 DMA problems.
> Many people who have SCSI have 2.1 problems but do not have 2.0 problems.

not necessarily. Eg., i have another (old) SMP box, which is heavily
overclocked. Back when 2.1 got better and better, errors started popping
up. Mysterious file corruptions (which went away after reboot), stuff like
that, typical RAM corruption problems. Didnt happen with 2.0, happens with
2.1, still 2.1's only 'fault' is that it's much more effective. So many
things have been reimplemented to be as effective as it can go. The last
thing that made this particular box fail finally was the %esp 'current'
improvement on SMP, which removed that crazy 30 cycles delay caused by
reading the CPU ID from the local APIC. Another thing when i had to make
BIOS timings somewhat more conservative and add yet another fan was when
the dentry cache started being so fast around 2.1.50.

> BTW: Have you checked Ingo's bug report on a possible race introduced in
> the IDE driver in 2.1 ???

i do not see any corruption, 'only' lockups, so my problem seems to be
different. Also (as i've said in my report), this particular sti() has
been in 2.0 as well. So it's just a demonstration, maybe a small clue,
maybe maybe a solution.

-- mingo

ps. if it wasnt for this flamewar i wouldnt have gone back and wouldnt
have produced this smallish patch, so the 2.1.111 'sledgehammer'
change did indeed have some use ;)

ps2. <flame> in 2.0, there was one particular very subtle and hard to
reproduce IDE DMA Triton bug observed by a friend of mine. The only
developer who cared about this bug was Gadi Oxman (and Linus too).
finally, after ~2 months of debugging, it turned out to be a chipset
bug, fixed by a flash BIOS update. </flame>

To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to
Please read the FAQ at