Re: Underruns on Audio CD Burns
On Mon, Dec 06, 2004 at 10:28:06PM -0500, Bill Davidsen wrote:
> I've had mixed results using non-ide-scsi with 2.4 kernels. There has
> been a lot of discussion on the kernel list on this (and I'm guilty of
> starting part of it), and I've had no problems using ATAPI: with a 2.6
> kernel up to 48x but I'd rather not on 2.4.
Uh, you'd rather not what on 2.4? Use ATAPI? Smart, as there's no DMA
then at all, with ide-scsi you at least have DMA when writing data CD's.
> My puzzlement is not how to get around the issue, but why it happens. I
> do appreciate the suggestion,
As fas as I know, this is caused by a performance problem in your
southbridge when doing PIO. Since there's no way to use DMA with Linux
kernel 2.4 and audio CD writing, you're stuck with slow writes with
them and any other kind of CD that has a sector size not multiple of
512 bytes, like VCD.
I had this problem with VIA's notorious 686b southbridge. 12x audio
burns were OK, 16x burns went actually at 14x or so and my machine was
somewhat bogged down, but not totally. The number of IRQs generated was
unbelievably high with 16x burns. Apparently the 16x LG writer I had back
then was smart enough to limit write speed by itself.
After I asked about this on lkml sometime in 2002, I was contacted by a
couple of people with Intel hardware and the same problem. I was very
surprised to hear that apparently Intel's ICH4 southbridge (or maybe
ICH5, I've lost those mails) has this problem as well, even though
Intel's much older PIIX3 does not...
If you're interested, I got around this problem by switching to FreeBSD,
then going back to Linux and getting a motherboard with SiS745 chipset,
which had no problems with PIO and 16x burns. Current solution is
Linux kernel 2.6.
Reply to: