Bug#273106: cmd64x driver doesn't load properly on ia64
On Saturday 16 October 2004 11:45, Joey Hess wrote:
> Jim, I noticed your patch to prebaseconfig for this bug. I wonder if
> closing the bug report there is really the right thing to do, since the
> real bug seems to be in the kernel, doesn't it?
yes, the "real" bug is in the ide subsystem but I looked at it and
a backport of the probe code would hit a lot of code. The other
distros solve this by not compiling ide as modules and I don't see
Marcelo or others being interested in accepting a backport now when
2.6 is the stable kernel. Probe order is the least of ide's problems.
By the same token, it doesn't make sense to carry along a
debian specific backport either.
If the real bug is in the kernel one might say that the
solution to 2.4 bugs is to trade them in for 2.6 bugs... In the
end, I recommend closing the kernel bug if for no other reason
than this is not any more egregeous a workaround than any other
distro. with < 2.6 kernels.
>
> I can reproduce this on my ia64.
>
> The comment says that "50prebaseconfig" (really it's 50register-module)
> adds modules in file name sort order. But really it adds them in
> whatever order they were registered. hw-detect is careful about the
> order, but makes an i386 centric assumption (which may not even be
> correct there) -- that ide-detect and ide-generic will be enough to make
> most CD drives work. The cmd64x module is not registered in /etc/modules
> at all.
hw-detect is careful about order it creates files but register-module
creates a file by module name and they are then taken out via a
"for f in `ls ...`" or somesuch. ide-cd before ide-detect
works as a result... Otherwise, I would have stuffed it in there.
Yes, cmd64x would also work but most of the others wouldn't.
If you want true fifo order, the register-module should be using
an append to a single file, which would also be more efficient.
>
> We rarely see this case on i386 because it only happens if the system
> has a scsi hard drive and ide cd; if the hard drive is ide then the
> kernel's initrd takes care of loading all the appropriate ide modules in
> the right order.
This is correct. The same applies to the installers h/w discovery.
>
> I think that it might be good to hunt down the used ide driver(s) using
> the method you found on i386 too, and register-module those drives
> before ide-detect and ide-generic, so they get loaded first. This could
> be done in hw-detect. It's probably too late to make this change for
> sarge and so I see your prebaseconfig hack as a suitable workaround for
> now.
One can always add other archs to the initial test. I was restrictive
because I only have an ia64 for this. The problem applies to any
system which only uses the ide controller for cdrom/dvd, in my case,
an rx2600. I'm sure there is a sata or scsi system out there that uses
an ide for the cdrom... As for moving this to hw-detect.sh, that would
be a good idea if the storage method was also fixed. I went for the
minimal change.
Jim
--
*************
Jim Lieb Wild Open Source Inc.
lieb@wildopensource.com Cell: 831.295.9317
Office: 831.421.0883 Fax: 831.421.0885
Reply to: