Debian "forgets" ide controller after install

[ljones0]

Hello all!

I'm just trying to install a newer version of debian on an older machine here - and although the install appears to go ok after install I am unable to boot. I'm trying to install debian stretch (though just a command line), kernel 4.9.0-7-686. I can get partly into the system by using rescue mode from the installer.

FIrst of all the machine's specs -- they are nothing amazing as this is quite an old PC. The CPU is an old AMD athlon 1100 with 1GB ram with an IDE dvd rom drive and 20GB IDE hard drive. The board uses a SiS chipset.

Install seems to go ok (and the installer detects the IDE controllers and hard drive with no problem). But after boot I get this.

I then get dropped out into a initramfs command line. After looking around online it appears sometimes this problem can happen if you use a swap partition but I didn't use any swap partition when installing and don't have any swap partitions on the hard drive. I also tried some of the ideas on [url=
https://askubuntu.com/questions/676707/ ... -in-initrd]this page[/url] to try to load a module but it did not work.

In the initramfs debian did not find my hard drive or any ide device at all. I can confirm these devices do work however - for example, I can install freedos on the hard drive in question and it works fine; also I can boot a debian live (command line) CD and it will find the hard drive/ide controller with no problem.

In the initramfs there don't seem to be very many modules loaded! I counted 4 - evdev, serio_raw, fan and thermal. Was thinking that something like the module "all_generic_ide" would be present, but it isn't.

I did try the solutions on this site but even after trying to force-load all_generic_ide it dosen't seem to work (even after modifying and running update-initramfs -u) . Listing the modules in the initramfs still dosen't show all_generic_ide.

Yet that module seems to be in the initramfs! It just never seems to get loaded or started.

I have read that sometimes the problem can be in a file called /etc/initramfs-tools/conf.d/resume but there is no file on my system called that (the directory is there, but empty) .

BTW I did try both bootloaders, grub and lilo. They install without problem but they don't fix the problem above.

Does anyone know what could be going wrong? Thanks

ljones

[day]

set

Code: Select all

MODULES=most

in file /etc/initramfs-tools/initramfs.conf, then regenerate initrd.
If you know the exact name of module(s), you can write it to /etc/initramfs-tools/modules.

[ljones0]

Ok I gave that a go and updated the initramfs but it still didn't seem to want to work. /proc/modules while in the initramfs still just shows 4 modules listed and no sign of the module all_generic_ide . Not sure why it isn't working -- I wonder what the installer's kernel and initramfs do differently? Still got this!

ljones

[Segfault]

All IDE drivers are depreciated, PATA drivers are used instead. Boot with some Linux CD and check with lspci -k which driver is loaded. Your drive will not be /dev/hda as it would with IDE drivers, it will be /dev/sda.

[ljones0]

Understood and I think I had the wrong module in any case, "all_generic_ide" - !

I rebooted with the debian installer cd and went into rescue mode and executed "lspci -k". That said that the module 'pata_sis' was being used.

So I rebooted normally (and dropped again back into initramfs) and tried the following:

Code: Select all

$ cd /lib/modules/4.9.0.7-686-pae/
$ cd kernel/drivers/ata
$ insmod pata-sis.ko

But I got a whole bunch of [url=http://slothpuck.no-ip.org/data/oldpc/9.jpg]unknown symbol error messages[/url] :-(!

ljones

[Segfault]

This error usually comes up when the modules are not built for running kernel, in other words, version mismatch.

[ljones0]

That's really weird then as these files are what debian installed. The initramfs prompt I'm getting from there is what happens if I try to boot normally (directly from the hard drive, not the cd).

Really confused as from within the initramfs (which in turn is running busybox) uname -r returns "4.9.0-7-686-pae", and looking at /lib/modules that too is 4.9.0-7-686-pae . Wonder why it'd be a different version?

Even tried modinfo pata_sis.ko to try to list information about the module. It says it depends on "libata" and is version "4.9.0-7-686-pae".

Did some further checking within the initramfs btw;

- pata_sis.ko is version 4.9.0.7-686-pae, depends on libata.
- libata.ko is version 4.9.0-7-686-pae, depends on scsi_mod
- scsi_mod is version 4.9.0-7-686-pae, depends on nothing.

I found if I manually insmod'd scsi_mod.ko, then libata.ko and then pata_sis.ko I didn't get any of the unknown symbol error mesages. But it still dosen't see any ide devices either x.x !

Going back to the debian installer cd in rescue mode I can see with lsmod uname -r

Going back to the debian installer cd in rescue mode for a moment I can see that its kernel is apparently "4.9.0-7-686" reported by uname -r. Would the "pae" part matter?

I used lsmod from the cd in rescue mode and found the following modules being used:

- ata_generic
- pata_sis
- libata
- scsi_mod

There's also a sis900 and mii which I'm not sure about. I did try to modprobe/insmod these in the initramfs as well, but no joy no ide devices detected.

ljones

[p.H]

This error usually comes up when the modules are not built for running kernel, in other words, version mismatch.

No, it comes up because insmod does not take care of modules dependencies.
Use modprobe instead.

[Segfault]

This error usually comes up when the modules are not built for running kernel, in other words, version mismatch.

No, it comes up because insmod does not take care of modules dependencies.
Use modprobe instead.

So you are saying it never comes up in case of version mismatch? (I agree, this time it may be insmod problem.)

[ljones0]

Random thought: With regards to the kernel that debian installed and the install cd the only difference is that the cd's kernel isn't pae, whereas the kernel debian installed is pae. Maybe my PC dosen't like this?

(debian install cd == 4.9.0.7-686, installed kernel == 4.9.0.7-686-pae) .

Maybe there's a certian order to modprobe in but I have no idea what that would be - modprobe scsi_mod, then libata and pata_sis dosen't seem to detect any devices.

ljones

[ljones0]

Ok, this odd! I did another debian install on a different hard drive - only this time the only difference was that I used 'expert mode' and did not install a PAE kernel. And the system seems to have booted ok from the hard drive. The kenel reported by uname -r is "4.9.0-7-686", and not "4.9.0-7-686-pae".

Which is strange because cat /proc/cpuinfo | grep pae displays output; pae is listed in the flags section of /proc/cpuinfo. So this older cpu (which is an amd athlon 1100) does support pae! Why it dosen't like using it though is anyone's guess......

/proc/cpuinfo reports this btw:
processor : 0
vendor: authenticamd
cpu_family: 6
model: 4
model name: amd athlon(tm) processor
stepping: 2
cpu mhz:1102.483
cache size: 256kb
physical id: 0
siblings: 1
core id: 0
cpu cores: 1
apicid : 0

Maybe the newer modules with the pae kernel in the initramfs (again running a pae kernel) are just too new? Does debian/linux have any sort of special 'pae' module?

ljones

[p.H]

Forgot to comment on this :

There's also a sis900 and mii which I'm not sure about.

Irrelevant. These are ethernet drivers.

I found if I manually insmod'd scsi_mod.ko, then libata.ko and then pata_sis.ko I didn't get any of the unknown symbol error mesages. But it still dosen't see any ide devices either

The module sd_mod.ko must also be loaded to detect PATA/SATA/SCSI/USB /dev/sd* disks.

So you are saying it never comes up in case of version mismatch?

It should not. Version mismatch should be detected earlier and produce a different message from insmod :

Code: Select all

insmod: ERROR: could not insert module <path>/<module>.ko: Invalid module format

and in the kernel log :

Code: Select all

<module>: disagrees about version of symbol module_layout

Otherwise, missing symbols suggest an ABI mismatch without a version mismatch. This should not happen with stock Debian kernels. In case of an ABI change brought by an update, the ABI number in the kernel version (and package name) should be incremented, for example from 4.9.0-6 to 4.9.0-7, in order to avoid such ABI mismatch.

However, this is the theory. In practice, I must admit that I have observed the missing symbol error before rebooting after a kernel update. This has happened very rarely, once or twice, and last time was long ago.

pae is listed in the flags section of /proc/cpuinfo. So this older cpu (which is an amd athlon 1100) does support pae!

Of course it does. PAE has been supported by almost any x86 CPU since Pentium Pro (686) with rare exceptions such as early Pentium M. If the CPU did not support PAE, the installer would have installed the non-PAE kernel and the PAE kernel would not even boot.

However, I do not think that such an old computer would benefit much from the PAE kernel. PAE is required to enable the desirable NX/XD bit (prevent execution of data pages), but this feature was not implemented in 32 bit Athlon CPU yet, it first appeared in Athlon 64.

Anyway, I do not see how PAE could be the cause of the breakage.

Maybe there's a certian order to modprobe in

No. Just modprobe the module you need and it will automatically load other modules it depends on.
However, udev should also automatically load modules for detected hardware, so you should not need to manually load it. It seems that this is the missing feature in the initramfs. Maybe something went wrong during the first installation and the initramfs creation.

[ljones0]

Switched back to 'old' hard drive and gave modprobing sd_mod.ko a try but still nothing even after modprobing pata-sis x.x . Will wipe this install and re-install and see if that fixes anything although I don't remember the installer coming back with any missing files or errors during installation. Prehaps udev got scrambled up somewhere as you say?

ljones

[ljones0]

Reinstalled, but still have the same problem. However I did notice 'something' possibly maybe going wrong during install somewhere around 5% on base install. But I was unable to see what it said (the installer carried on installing btw) it was quite literally a "blink-and-you'll-miss it" moment - if you had your head turned at that moment you'd never have seen it.

I did manage to save the syslog from the installer however.

Looks like around Aug 18 19:19:51 according to the log either the kernel or busybox crashed. x.x !

ljones

[debiman]

it is amazing how you all try to solve this the technical way, together (no sarcasm).

but...

The board uses a SiS chipset.

SiS has always been bad news for Linux; i think you should concentrate your searches here.
maybe you should also try a different distro - one that is designed to run on old hardware - namely antiX.

[p.H]

I know about SiS chips bad reputation and I have always carefully avoided them. However PAE is a CPU thing, and I fail to see how it would affect the PATA controller operation.