ata_piix installation/device detection problems on Dell server

Package: debian-installer
Version: RC3
Thanks

Hi.

I Cc this message to the ata_piix maintainer, the debian bug tracking system for debian-installer (since it affects installation) as well as the linux-ide mailinglist (as requested in the ata_piix source) and the debian-boot mailinglist (though it should get a report from the BTS). If you think any of these are inappropriate, please let me know and remove them from replies please.

I just experienced a problem with a Dell server which is using combined mode to provide a PATA port for its CD-ROM and an SATA port for its HD.

The problem I experienced is as follows:

the ata_piix driver detects PATA and SATA in combined mode (which does seem correct as far as I can tell). However, only the HD is detected/registered in the kernel while the CD-ROM remains undetected/unregistered. I therefore can't install using the default settings of debian-installer rc3 (netinst CD tested, other still untested, but since it's a kernel issue, I doubt they would make a difference.

I was able to work around this issue by _not_ loading the ata_piix module at first, but load the generic ide driver module instead. Of course, that module doesn't detect the SATA part, so I would be able to access the CD-ROM but not the HD. However, loading ata_piix _after_ the generic IDE driver notes that the first port is busy (PATA, used by ide_generic) and only initializes the SATA part. This makes it possible to access both SATA (HD) and PATA (CD).

If anyone wants access to the machine in question, I guess I could arrange both root and serial console access to it for a limited time. Will have to check with my boss though. Tell me if you would need that.

regards,
Sven

PS: Please Cc me in answers unless you also send a copy to the debian-boot mailinglist (which I read).

Hi.

I just found out that libata by default doesn't support PATA devices. This can be changed in include/linux/libata.h (or so) at compile time of the kernel by defining ATA_ENABLE_PATA (it's #undef'ed there by default). Since ata_piix blocks the PATA part of the chip(set), it would be good if either that default is changed to enable PATA in libata drivers or that ata_piix doesn't touch the PATA part when ATA_ENABLE_PATA isn't defined (or, preferably, both).

Regarding debian-installer, I guess there are currently three ways to solve this:
1) Re-compile the kernel with ATA_ENABLE_PATA defined (which is currently only used in ata_piix.c anyway, so it shouldn't have ill side effects)
2) Document the problem and provide a description of the possible workaround.
3) Implement the workaround in debian-installer. (Loading the generic IDE driver (ide_generic?) before ata_piix)

I would prefer to see (1), but I assume this will not happen pre-Sarge.

PS: I created a bug report for debian-installer (#303135).

The offer still stands.

Regards,
Sven