Re: 2.1.111: IDE DMA disabled?

Andrew Derrick Balsa (
Mon, 27 Jul 1998 12:03:17 +0200

Hello Linus, Alan,

I think the original point was that Linus wanted to disable IDE DMA by default
in kernel 2.2.x, and Alan was arguing that this may not be an optimal decision,
because IDE DMA performs _much_ better than IDE PIO.

I tend to agree with Alan. The IDE DMA driver by Mark Lord is a good piece of
code, as reliable (or more) as any other piece of code in the kernel.

Disabling IDE DMA because of the odd (rare) cases of flawed hardware that have
been reported on the list, is like saying that the 100BaseTx Ethernet drivers
in the kernel should be set by default to work at 10BaseT, because there may be
some bad cat5 cables out there.

That doesn't make sense, IMHO.

I have mentionned various (controlable) parameters that affect IDE DMA
reliability, among others the BIOS revision (mentionned earlier on by Alan). One
can follow these simple rules and get reliable DMA mode2 and/or UDMA operation
quite easily, _if_ the hardware is OK.

One last note: before you decide on enabling/disabling IDE DMA by default in
kernels 2.2.x, may I suggest you check with Mark Lord first? He is the author of
the code and is entitled to a decision on this issue, or at least to voicing
his opinion, IMHO.


Andrew D. Balsa

PS: like Alan, I have noticed many posts mentionning silent fs corruption on i386 SMP systems, whether using SCSI DMA or IDE DMA. So, there may be a stray spinlock or similar problem in the SMP code, related to DMA accesses. Or even a still unknown, obscure problem with the Intel chipsets related to SMP?

- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to Please read the FAQ at