On 23-Nov-97 Jochen Heuer wrote:
>Hello everybody,
>
>I'm experiencing some problems with filesystem corruption under
>2.0.32 because I could not compile anything anymore. Having a
>little look I found out that some files got *strange* permissions.
>ide: i82371 PIIX (Triton) on PCI bus 0 function 57
> ide0: BM-DMA at 0xf000-0xf007
>hda: WDC AC21600H, 1549MB w/128kB Cache, CHS=787/64/63, DMA
>ide0 at 0x1f0-0x1f7,0x3f6 on irq 14
Disable the Triton Bus Mastering DMA support in the kernel. We are
getting sporadic reports of problems with this option enabled. The
problems tend to go away with this option disabled.
Grrr...
I use this on some system here (hdparm -u1 -c1 -d1 -X34). In general DMA is
a win but at times, usually of high disk activity, BUT NOT ALWAYS, I get
cruft like:
Nov 13 16:02:12 caffeine kernel: hda: irq timeout: status=0xd0 { Busy }
Nov 13 16:02:12 caffeine kernel: hda: DMA disabled
which is a real pain.... there's a chance its the drives and not the
controller though.
I'm not sure whether or not other OS have this problem, but it looks as if
NT and 95 probably use bus mastering when its available (a guess based upon
looking at interrupts numbers before and after a certain about of disk
activity) and they don't seem to have this problem.
I keep hearing about Ultra SCSI, ATA-3, etc. and the narrowing gap between
SCSI and IDE - but I just don't see it.
For the mean time I'm probably going to start get 4GB Wide-SCSI Seagate
Hawk's because they're cheap and I can stripe 4 of them to get reasonable
performance. Better suggestions are welcome.
-Chris