I have seen the same problem with a 486DX100 with VLB bus and a VLB IDE
controller. I reverted to 1.3.3.

A file cat of /netbsd to /dev/null took 8 seconds on 1.4.1 and under a
second on 1.3.3.

dmesg output is the same for both systems:

        wdc0 at isa0 port 0x1f0-0x1f7 irq 14
        wd0 at wdc0 drive 0: <ST1144AT>
        wd0 using 1-sector 16-bit pio transfers, chs mode
        wd0: 124MB, 1001 cyl, 15 head, 17 sec, 512 bytes/sec
        (wd1 is the same type of disk)

My guess is that the interrupt per sector is taking longer than the
inter-sector gap time and the transfer is paying a disk rotation time
for each sector transferred.

Cheers,

George D Detlefsen

Frank van der Linden wrote:
> 
> On Thu, Nov 25, 1999 at 11:39:37AM -0500, static wrote:
> >       I have an old 486DX25 with two HDs - a 80mb Western Digital and a
> > Seagate 100mb (both IDE). The WD drive works fine in normal speed, but the
> > Seagate is barely usable because it is so slow. Formatting takes forever
> > on that drive, and using it causes the system to go slow (it is my /usr).
> > I'm currently running 1.3.2 on that machine because of that, and recently
> > tried a 1.4N kernel and got the same problems.
> 
> I've seen this problem too, with old IDE disks. Some other people have
> as well. I've not been able to track this down, unfortunately, and
> Manuel doesn't have any disk that it shows up on, so it's hard for
> him to fix :-(
> 
> - Frank