I built a 2.6.0 kernel and have been testing it with my installed slackware 9.1 which normally uses a 2.4.22 kernel. I noticed that it seemed slow, especially disk-intensive things. Testing with hdparm shows it is MUCH slower than 2.4.22 --

hdparm -t on 2.4.22:
buffered disk read: 64 MB in 1.61 seconds

hdparm -t on 2.6.0:
buffered disk read: 64 MB in 34.01 seconds

Naturally both runs are using the same disk/partition, nothing else running, tried mutliple times, etc. I can't find anything left to enable in the kernel config that seems like it might help. Any suggestions as to what might explain this?

did you enable dma? or did you compile dma support?
do "hdparm <drive>" as root to see the difference.

Yes dma is enabled, at least in the kernel config ("use by default"). But I did notice the following difference from doing hdparm -I:

2.4.22:
R/W multiple sector transfer: Max = 16 Current = 16

2.6.0:
R/W multiple sector transfer: Max = 16 Current = 0

That's the only difference (from hdparm). Where does this get configured?

try "hdparm -m16 <device>"
and use "hdparm -tT <device>" to test how fast it is?

Are you sure you compiled in chipset-specific support in 2.6? The specific chipset support allows the kernel to manipulate the PIO, DMA, UDMA and IRQ settings of the IDE core, which allows you to increase performance without having to run hdparm every time the system boots.

Thanks for that tip, helped me to finally solve this... I had the chipset (VIA) support configured as a module instead of built-in, and had most of the other chipsets also enabled as modules, plus the "generic" ide was enabled. Changed to VIA as built-in, and removed all other chipset support including "generic", and now it uses the correct driver. Before it was not enabling dma (couldn't even get hdparm to enable it), now it is. However it's still not enabled multiple-sector transfers, but that can be enabled with hdparm. (Dma accounts for most of the speed-up anyway.)

Next problem is getting the sound to work...