> >> I'd be willing to bet a dime that the PIO problem is a write-behind cache
> >> problem that is being masked by turning off interrupts during PIO
> >> operation, I'll bet that changing write-back to write-through "fixes"
> >> the problem.
>
> >On a 386 with no cache. Man that'd be clever.
>
> You're pretty quick to dismiss the possibility that there's something
> wrong with the linux kernel and/or IDE code.
>
> FreeBSD does not have the Linux weakness of needing to disable
> interrupts during IDE disk I/O to avoid filesystem corruption.
This is likely just because FreeBSD has *far* fewer users than Linux,
and probably even fewer "vintage" machines. The problem machines
number very few, and nowadays we'd be fairly safe just plain enabling
interrupts during IDE I/O on all PCI-based boards.
Two years ago, the problem was much more widespread, but just about
all problem systems were ISA or VLB, not PCI.
-ml
the (old) Linux IDE guy