CDROM drive failure under Linux seems odd

[dunbar]

Ok, here is why I think a certain CDROM drive is still good: Linux boot images (vmlinuz, used on install CDs and with LILO until mid way through the boot process) access it perfectly first try, every try. Rescue disks are the same because they tend to use vmlinuz.

Here is why I think the drive is bad: First off, I replaced it with a different CDROM drive, all seems fine as of now. Second, back when the 'bad' CDROM was still installed, under certain conditions, I had to suffer 3 ATAPI resets before it would allow me to use it. Those conditions were: always need to perform 3x ATAPI reset on the very first access [after the machine was cold booted] or [warm rebooted]. This was not specific to one distro: I tried Slackware 8.0, Debian something and several RPM based distros (RH, MDK, and some others). All suffered the problem. DEVFS was used in some installs, not used in others. Third, sometimes during installations, the install would fail "package error , package foo not installed", but the very next install attempt fails elswhere, or possibly not even failing at all!

The puzzler to me is the fact that bootimages ever got flawless access first try....

Something is assumed differently between the 2 cases of IDE access (vmlinuz vs installed kernel), and the vmlinuz instance is the better choice, IMO. The asumption of a difference makes sense to my limited Linux knowledge, because the 2 kernels themselves are different (amongst other diferences): the modular kernel seems to be the post install / post boot choice; whereas I suspect that the vmlinuz kernel is using more elementary (read that as 'less assumptions').

Any ideas?

[KraM]

Well, I think when you boot from the CD, the bootloader (NOT Linux) loads an initrd image into memory which the kernel uses as the root filesystem. Whatever install programs that are on there, probably are inside the image and are run (by the kernel) from memory, not from CD.

When the kernel is accessing the CD drive (eg. copying packages) it has problems....

I think this is what you are asking...not real sure...

HTH

[dunbar]

Originally posted by KraM
Well, I think when you boot from the CD, the bootloader (NOT Linux) loads an initrd image into memory which the kernel uses as the root filesystem. Whatever install programs that are on there, probably are inside the image and are run (by the kernel) from memory, not from CD.

It is a really tricky issue: I booted from CD without the errors, on the first try. Then, after the installer routine loads (flawlessly), that program then reads the first data from the CD - flawlessly... remember the initrd you mentioned? It came from the CD, as I am not using a floppy to boot with. Really odd...

[KraM]

Yes, but the bootloader loaded the image. It just sounds like linux is having trouble with your particular combination of hardware. Or, there is something wrong with you hardware.

I'm taking an educated guess here: I think the bootloader uses the BIOS to access the IDE devices. But, Linux, being the advanced piece of software that it is , accessing these devices directly, thus bypassing potentially buggy BIOSes.

You can force Linux to access the PCI bus through the BIOS, and the IDE controllers are on the PCI bus (??). Anyway, to do this, while configuring a kernel:

make menuconfig -> General Setup -> PCI Access Mode -> BIOS

I use menuconfig when configuring the kernel, it may be somewhat different if you use xconfig.

Worth a try I think May open up a whole new can of worms if your BIOS is buggy etc...

If it doesn't work, no big deal, just go back to your old kernel. It may also be worth trying just selecting direct instead of any, which is the default AFAIK.

[caddmannq]

Whoa! This is an eye-opener!

I've got two distros that have failed to install from CD's I burned on another PC. I've been blaming the CDs themselves, even though they seemed to check out clean when I made them.

I've installed Linux from CD on this box before, but it had a diffferent CD drive...

[dunbar]

Originally posted by KraM
Yes, but the bootloader loaded the image.

Exactly where did the image come from???? It came from the CDROM disc called 'distro install disc one'. Flawless performance on that first try when accessing the CDROM drive. After that point, the loaded data contained something that eventually modified how the current process would access the CDROM. Buggy BIOS or no, the same BIOS worked correctly before the software loaded in order to get the bootloader (did not need ATAPI resets at that point... I'd have heard it, very distinct pattern of head positioning attempts), then everything still worked correctly while loading the first booted code and even worked while loading of a few subsequent screens. After this, the system then began working incorrectly. And the same BIOS works fine on a different CDROM.

It just sounds like linux is having trouble with your particular combination of hardware. Or, there is something wrong with you hardware.

Granted. I use a different CDROM now because I have already decided the first CDROM drive might be bad... I'm now using a diffferent drive on the same system and things now act correctly. We'll see how the different drive functions over the next few weeks, but no ATAPI resets thus far means I like it more than the first drive.

I'm taking an educated guess here: I think the bootloader uses the BIOS to access the IDE devices. But, Linux, being the advanced piece of software that it is , accessing these devices directly, thus bypassing potentially buggy BIOSes.

Might the device be designed exactly the other way around - assuming the BIOS will be the only method? Might the Linux assumption that 'all CDROM drives are standardized' be a bit less than realistic?

You can force Linux to access the PCI bus through the BIOS, and the IDE controllers are on the PCI bus (??). Anyway, to do this, while configuring a kernel:

Not while booting from CD for the first install....
Thanks for trying, but I remain uninformed.

[Peter M]

Point here probably is that the board's BIOS runs the CDROM drive in a slower transfer mode than Linux is trying to do. Typically, CDROM boot is performed using some kind of PIO mode, while modern Linux distros try to auto-setup a suitable DMA mode. There aren't exactly few IDE CDROMs that screw up in DMA modes. You need to use the "idedma=" kernel parameter on bootup for those.

regards, Peter

[caddmannq]

Hey there, PeterM...regarding "idedma=" parameter....

..equals what? Am I pointing to a certain DMA channel?
...or a certain CDROM drive number?
...or ???

Cadd.

[Peter M]

I don't use IDE here Please refer to the Linux kernel documentation for the exact syntax. It should be in the usual FAQs too since this is quite common a problem.

regards, Peter

[caddmannq]

AHA! To enable for all IDE drives, which is what I wanna do:

"idedma=yes" (from the SUSE FAQ)

Thanks once again...
Cadd.