Prev: Xine can't play DVDs or Xvids?
Next: slackware 12 on i810, monitor sony bravia KVL-S40A10 (40inch 1280x768)
From: Franck Barbenoire on 20 Jul 2007 16:51
Hi,

1) Purpose

My purpose is to run the generic kernel (GK) on a raid 1 array of 2 sata
discs. I use standard unmodified packages from the installation Slack 12
dvd. All mirrored partitions are formatted using reiserfs file system.

Description of the array :

md0 is composed of /dev/sda1 and /dev/sdb1 mounted on /boot
md1 is composed of /dev/sda5 and /dev/sdb5 mounted on /

GK stands for Generic Kernel
HK stands for Huge Kernel

2) Tries

I first made a successfull install of the huge kernel (HK).

Then, I made an initrd.gz image including only the reiserfs module :

# mkinitrd -c -k 2.6.21.5-smp -m reiserfs -f reiserfs -r /dev/md1

In lilo.conf, I copied the entry made for the HK to a new entry that I
renamed "Test", I changed the kernel image filename and I added initrd.gz
line.

At the end og GK rebooting, I got a kernel panic :

....
initrd.gz: Loading kernel modules from initrd image
Using : /lib/modules/2.6.21.5-smp/kernel/fs/reiserfs/reiserfs.ko
mount : mounting /dev/md0 on /mnt failed
....
kernel panic

3) Investigations

First, I make a diff between the configurations of the HK and GK. There is
no difference concerning sata and raid drivers ; the HK boots without an
initrd.gz. Reiserfs is built as a module in GH.

Second, I break my raid array. On one disc, I turn my partitions to standard
Linux (type=83 instead of fd) and reinstall HK then GK on this disk. For
GK, I make an initrd the same way as in 2), changing the device names of
course.

Both HK and GK work one one sata disc with reiserfs formatted partitions !

So, the problem seems to come from raid detection in GK. It is not related
to sata nor reiserfs.

Looking in /boot/initrd-tree/dev directory, I notice that all device nodes
(sda* and sdb*) corresponding to all partitions exist but md* are symbolic
links to inexisting files !

In order to satisfy these symbolic links, I make a md directory
in /boot/initrd-tree/dev and created 0, 1,... nodes in it with the same
permission, owner, group, type, major, minor as in the regular /dev/md.

I run mkinitrd without an argument in order to rebuild initrd.gz, rerun
Lilo, reboot. Things have changed, the kernel log is now :

....
initrd.gz: Loading kernel modules from initrd image
Using : /lib/modules/2.6.21.5-smp/kernel/fs/reiserfs/reiserfs.ko
ReiserFS : md0 : warning: sh-2006: read_super_block: bread failed (dev, md0,
block 2, size 4096)
ReiserFS : md0 : warning: sh-2006: read_super_block: bread failed (dev, md0,
block 16, size 4096)
mount : mounting /dev/md0 on /mnt failed
....
kernel panic

I reboot HK, reiserfs fs are clean.

4) Conclusion

I feel like running an inird.gz at boot time affects badly raid arrays
detection and mounting.

For the moment I will stick on HK but I'd like to know whether I went to an
untested area or I did bad things.

Many thanks for your answers !

Franck Barbenoire




From: cbxbiker61 on 21 Jul 2007 08:44
Franck Barbenoire wrote:
> Hi,
>
> 1) Purpose
>
> My purpose is to run the generic kernel (GK) on a raid 1 array of 2 sata
> discs. I use standard unmodified packages from the installation Slack 12
> dvd. All mirrored partitions are formatted using reiserfs file system.
>
> Description of the array :
>
> md0 is composed of /dev/sda1 and /dev/sdb1 mounted on /boot
> md1 is composed of /dev/sda5 and /dev/sdb5 mounted on /
>
> GK stands for Generic Kernel
> HK stands for Huge Kernel
>
> 2) Tries
>
> I first made a successfull install of the huge kernel (HK).
>
> Then, I made an initrd.gz image including only the reiserfs module :
>
> # mkinitrd -c -k 2.6.21.5-smp -m reiserfs -f reiserfs -r /dev/md1
>
> In lilo.conf, I copied the entry made for the HK to a new entry that I
> renamed "Test", I changed the kernel image filename and I added initrd.gz
> line.
>
> At the end og GK rebooting, I got a kernel panic :
>
> ...
> initrd.gz: Loading kernel modules from initrd image
> Using : /lib/modules/2.6.21.5-smp/kernel/fs/reiserfs/reiserfs.ko
> mount : mounting /dev/md0 on /mnt failed
> ...
> kernel panic
>
> 3) Investigations
>
> First, I make a diff between the configurations of the HK and GK. There is
> no difference concerning sata and raid drivers ; the HK boots without an
> initrd.gz. Reiserfs is built as a module in GH.
>
> Second, I break my raid array. On one disc, I turn my partitions to standard
> Linux (type=83 instead of fd) and reinstall HK then GK on this disk. For
> GK, I make an initrd the same way as in 2), changing the device names of
> course.
>
> Both HK and GK work one one sata disc with reiserfs formatted partitions !
>
> So, the problem seems to come from raid detection in GK. It is not related
> to sata nor reiserfs.
>
> Looking in /boot/initrd-tree/dev directory, I notice that all device nodes
> (sda* and sdb*) corresponding to all partitions exist but md* are symbolic
> links to inexisting files !
>
> In order to satisfy these symbolic links, I make a md directory
> in /boot/initrd-tree/dev and created 0, 1,... nodes in it with the same
> permission, owner, group, type, major, minor as in the regular /dev/md.
>
> I run mkinitrd without an argument in order to rebuild initrd.gz, rerun
> Lilo, reboot. Things have changed, the kernel log is now :
>
> ...
> initrd.gz: Loading kernel modules from initrd image
> Using : /lib/modules/2.6.21.5-smp/kernel/fs/reiserfs/reiserfs.ko
> ReiserFS : md0 : warning: sh-2006: read_super_block: bread failed (dev, md0,
> block 2, size 4096)
> ReiserFS : md0 : warning: sh-2006: read_super_block: bread failed (dev, md0,
> block 16, size 4096)
> mount : mounting /dev/md0 on /mnt failed
> ...
> kernel panic
>
> I reboot HK, reiserfs fs are clean.
>
> 4) Conclusion
>
> I feel like running an inird.gz at boot time affects badly raid arrays
> detection and mounting.
>
> For the moment I will stick on HK but I'd like to know whether I went to an
> untested area or I did bad things.
>
> Many thanks for your answers !
>
> Franck Barbenoire
>
>
>
>
I had been thinking of trying an initrd on my software raid. I thought
it may be tricky and I guess you proved that. No real need to use
initrd when you build custom kernels. My kernel compile times are down
to half what they were. I was using an Athlon XP 3200. Now I'm using a
Core Duo T2250.
From: Franck Barbenoire on 21 Jul 2007 18:31
cbxbiker61 wrote:

> Franck Barbenoire wrote:
>> Hi,
>>
>> 1) Purpose
>>
>> My purpose is to run the generic kernel (GK) on a raid 1 array of 2 sata
>> discs. I use standard unmodified packages from the installation Slack 12
>> dvd. All mirrored partitions are formatted using reiserfs file system.
>>
>> Description of the array :
>>
>> md0 is composed of /dev/sda1 and /dev/sdb1 mounted on /boot
>> md1 is composed of /dev/sda5 and /dev/sdb5 mounted on /
>>
>> GK stands for Generic Kernel
>> HK stands for Huge Kernel
>>
>> 2) Tries
>>
>> I first made a successfull install of the huge kernel (HK).
>>
>> Then, I made an initrd.gz image including only the reiserfs module :
>>
>> # mkinitrd -c -k 2.6.21.5-smp -m reiserfs -f reiserfs -r /dev/md1
>>
>> In lilo.conf, I copied the entry made for the HK to a new entry that I
>> renamed "Test", I changed the kernel image filename and I added initrd.gz
>> line.
>>
>> At the end og GK rebooting, I got a kernel panic :
>>
>> ...
>> initrd.gz: Loading kernel modules from initrd image
>> Using : /lib/modules/2.6.21.5-smp/kernel/fs/reiserfs/reiserfs.ko
>> mount : mounting /dev/md0 on /mnt failed
>> ...
>> kernel panic
>>
>> 3) Investigations
>>
>> First, I make a diff between the configurations of the HK and GK. There
>> is no difference concerning sata and raid drivers ; the HK boots without
>> an initrd.gz. Reiserfs is built as a module in GH.
>>
>> Second, I break my raid array. On one disc, I turn my partitions to
>> standard Linux (type=83 instead of fd) and reinstall HK then GK on this
>> disk. For GK, I make an initrd the same way as in 2), changing the device
>> names of course.
>>
>> Both HK and GK work one one sata disc with reiserfs formatted partitions
>> !
>>
>> So, the problem seems to come from raid detection in GK. It is not
>> related to sata nor reiserfs.
>>
>> Looking in /boot/initrd-tree/dev directory, I notice that all device
>> nodes (sda* and sdb*) corresponding to all partitions exist but md* are
>> symbolic links to inexisting files !
>>
>> In order to satisfy these symbolic links, I make a md directory
>> in /boot/initrd-tree/dev and created 0, 1,... nodes in it with the same
>> permission, owner, group, type, major, minor as in the regular /dev/md.
>>
>> I run mkinitrd without an argument in order to rebuild initrd.gz, rerun
>> Lilo, reboot. Things have changed, the kernel log is now :
>>
>> ...
>> initrd.gz: Loading kernel modules from initrd image
>> Using : /lib/modules/2.6.21.5-smp/kernel/fs/reiserfs/reiserfs.ko
>> ReiserFS : md0 : warning: sh-2006: read_super_block: bread failed (dev,
>> md0, block 2, size 4096)
>> ReiserFS : md0 : warning: sh-2006: read_super_block: bread failed (dev,
>> md0, block 16, size 4096)
>> mount : mounting /dev/md0 on /mnt failed
>> ...
>> kernel panic
>>
>> I reboot HK, reiserfs fs are clean.
>>
>> 4) Conclusion
>>
>> I feel like running an inird.gz at boot time affects badly raid arrays
>> detection and mounting.
>>
>> For the moment I will stick on HK but I'd like to know whether I went to
>> an untested area or I did bad things.
>>
>> Many thanks for your answers !
>>
>> Franck Barbenoire
>>
>>
>>
>>
> I had been thinking of trying an initrd on my software raid. I thought
> it may be tricky and I guess you proved that. No real need to use
> initrd when you build custom kernels. My kernel compile times are down
> to half what they were. I was using an Athlon XP 3200. Now I'm using a
> Core Duo T2250.

I agree with you : you do not need an initrd when compiling the kernel
yourself, you just put anything you need in it. I just not wanted to do
that because I wanted to use the generic kernel from the dvd.

In order to test and see the difference, I will add reiserfs support to the
generic kernel configuration, and rebuild it, thus avoiding the use of
inird.

I will report the result as a post in this thread.

Franck
From: Guy Macon on 21 Jul 2007 20:08



cbxbiker61 wrote:

>I had been thinking of trying an initrd on my software raid. I thought
>it may be tricky and I guess you proved that. No real need to use
>initrd when you build custom kernels.

If you do decide to go with initrd, take a close look at the
possiblility of using initfs instead.

--
Guy Macon
<http://www.guymacon.com/>


From: Franck Barbenoire on 22 Jul 2007 15:09
Franck Barbenoire wrote:

> cbxbiker61 wrote:
>
>> Franck Barbenoire wrote:
>>> Hi,
>>>
>>> 1) Purpose
>>>
>>> My purpose is to run the generic kernel (GK) on a raid 1 array of 2 sata
>>> discs. I use standard unmodified packages from the installation Slack 12
>>> dvd. All mirrored partitions are formatted using reiserfs file system.
>>>
>>> Description of the array :
>>>
>>> md0 is composed of /dev/sda1 and /dev/sdb1 mounted on /boot
>>> md1 is composed of /dev/sda5 and /dev/sdb5 mounted on /
>>>
>>> GK stands for Generic Kernel
>>> HK stands for Huge Kernel
>>>
>>> 2) Tries
>>>
>>> I first made a successfull install of the huge kernel (HK).
>>>
>>> Then, I made an initrd.gz image including only the reiserfs module :
>>>
>>> # mkinitrd -c -k 2.6.21.5-smp -m reiserfs -f reiserfs -r /dev/md1
>>>
>>> In lilo.conf, I copied the entry made for the HK to a new entry that I
>>> renamed "Test", I changed the kernel image filename and I added
>>> initrd.gz line.
>>>
>>> At the end og GK rebooting, I got a kernel panic :
>>>
>>> ...
>>> initrd.gz: Loading kernel modules from initrd image
>>> Using : /lib/modules/2.6.21.5-smp/kernel/fs/reiserfs/reiserfs.ko
>>> mount : mounting /dev/md0 on /mnt failed
>>> ...
>>> kernel panic
>>>
>>> 3) Investigations
>>>
>>> First, I make a diff between the configurations of the HK and GK. There
>>> is no difference concerning sata and raid drivers ; the HK boots without
>>> an initrd.gz. Reiserfs is built as a module in GH.
>>>
>>> Second, I break my raid array. On one disc, I turn my partitions to
>>> standard Linux (type=83 instead of fd) and reinstall HK then GK on this
>>> disk. For GK, I make an initrd the same way as in 2), changing the
>>> device names of course.
>>>
>>> Both HK and GK work one one sata disc with reiserfs formatted partitions
>>> !
>>>
>>> So, the problem seems to come from raid detection in GK. It is not
>>> related to sata nor reiserfs.
>>>
>>> Looking in /boot/initrd-tree/dev directory, I notice that all device
>>> nodes (sda* and sdb*) corresponding to all partitions exist but md* are
>>> symbolic links to inexisting files !
>>>
>>> In order to satisfy these symbolic links, I make a md directory
>>> in /boot/initrd-tree/dev and created 0, 1,... nodes in it with the same
>>> permission, owner, group, type, major, minor as in the regular /dev/md.
>>>
>>> I run mkinitrd without an argument in order to rebuild initrd.gz, rerun
>>> Lilo, reboot. Things have changed, the kernel log is now :
>>>
>>> ...
>>> initrd.gz: Loading kernel modules from initrd image
>>> Using : /lib/modules/2.6.21.5-smp/kernel/fs/reiserfs/reiserfs.ko
>>> ReiserFS : md0 : warning: sh-2006: read_super_block: bread failed (dev,
>>> md0, block 2, size 4096)
>>> ReiserFS : md0 : warning: sh-2006: read_super_block: bread failed (dev,
>>> md0, block 16, size 4096)
>>> mount : mounting /dev/md0 on /mnt failed
>>> ...
>>> kernel panic
>>>
>>> I reboot HK, reiserfs fs are clean.
>>>
>>> 4) Conclusion
>>>
>>> I feel like running an inird.gz at boot time affects badly raid arrays
>>> detection and mounting.
>>>
>>> For the moment I will stick on HK but I'd like to know whether I went to
>>> an untested area or I did bad things.
>>>
>>> Many thanks for your answers !
>>>
>>> Franck Barbenoire
>>>
>>>
>>>
>>>
>> I had been thinking of trying an initrd on my software raid. I thought
>> it may be tricky and I guess you proved that. No real need to use
>> initrd when you build custom kernels. My kernel compile times are down
>> to half what they were. I was using an Athlon XP 3200. Now I'm using a
>> Core Duo T2250.
>
> I agree with you : you do not need an initrd when compiling the kernel
> yourself, you just put anything you need in it. I just not wanted to do
> that because I wanted to use the generic kernel from the dvd.
>
> In order to test and see the difference, I will add reiserfs support to
> the generic kernel configuration, and rebuild it, thus avoiding the use of
> inird.
>
> I will report the result as a post in this thread.
>
> Franck

I recompiled the kernel, starting from the generic kernel configuration, I
added support for reiserfs. Rebooting this kernel without use of inird.gz
works fine : raid 1 array is correctly detected and mounted.

So the problem is demonstrated : initrd + raid 1 = problem !

Franck
 |  Next  |  Last
Pages: 1 2
Prev: Xine can't play DVDs or Xvids?
Next: slackware 12 on i810, monitor sony bravia KVL-S40A10 (40inch 1280x768)