On 2007/12/08 13:37, Antti Harri wrote: > On Sat, 8 Dec 2007, Stuart Henderson wrote: > >> On 2007/12/08 10:59, Antti Harri wrote: >>> anyone figured out where the problem is with OpenBSD & CF? >> >> Naddy noticed that DMA is only used if the drive supports >> multi-sector transfers. >> >>> wd1 at wdc2 channel 0 drive 0: <TOSHIBA THNCF512MQG> >>> wd1: 1-sector PIO, LBA, 488MB, 1000944 sectors >> >> My slow cards are 1-sector, my fast cards are >1. > > Ok.. But no ideas why other systems perform better > with the same hardware?
If you try accessing the card on some other OS and see low CPU use while it takes place, it's probably using DMA. But there might well be a reason _why_ we don't do that (other than "until recently most IDE/CF weren't even wired for DMA because the lines weren't in older CF spec"). > PS. I can test diffs (and probably a friend of mine too) > if someone is working on it. Due to the type of computer where IDE flash tends to be used, this needs to be done really conservatively, with plenty of testing (lots of machines and cards).