Posts Tagged ‘kernel panic’

How to disable ACPI on productive Linux servers to decrease kernel panics and increase CPU fan lifespan

Tuesday, May 15th, 2012

Linux TUX ACPI logo / Tux Hates ACPI logohttps://www.pc-freak.net/images/linux_tux_acpi_logo-tux-hates-acpi.png

Why would anyone disable ACPI support on a server machine??
Well  ACPI support kernel loaded code is just another piece of code constantly being present in the memory,  that makes the probability for a fatal memory mess up leading to  a fatal bug resulting in system crash (kernel panic) more likely.

Many computers ship with buggy or out of specifications ACPI firmware which can cause a severe oddities on a brand new bought piece of comp equipment.

One such oddity related to ACPI motherboard support problems is if you notice your machine randomly powering off or failing to boot with a brand new Linux installed on it.

Another reason to switch off ACPI code will would to be prevent the CPU FAN rotation from being kernel controlled.

If the kernel controls the CPU fan on  high CPU heat up it will instruct the fan to rotate quickly and on low system loads it will bring back the fan to loose speed.
 This frequent switch of FAN from high speed to low speed  increases the probability for a short fan damage due to frequent changes of fan speed. Such a fan damage leads often to  system outage due to fan failure to rotate properly.

Therefore in my view it is better ACPI support is switched off completely on  servers. On some servers ACPI is useful as it can be used to track CPU temperature with embedded motherboard sensors with lm_sensors or any piece of hardwre vendor specific software provided. On many machines, however lm_sensors will not properly recognize the integrated CPU temperature sensors and hence ACPI is mostly useless.

There are 3 ways to disable fully or partially ACPI support.

- One is to disable it straight for BIOS (best way IMHO)
- Disable via GRUB or LILO passing a kernel parameter
- Partial ACPI off-ing - /disabling the software that controls the CPU fan/

1. Disable ACPI in BIOS level

Press DEL, F1, F2, F10 or whatever the enter bios key combination is go through all the different menus (depending on the vios BENDOR) and make sure every occurance of ACPI is set to off / disable whatever it is called.

Below is a screenshot of menus with ACPI stuff on a motherboard equipped with Phoenix AwardBIOS:

BIOS ACPI Disable power Off Phoenix BIOS

This is the in my opinon best and safest way to disable ACPI power saving, Unfortunately some newer PCs lack the functionality to disable ACPI; (probably due to the crazy "green" policy the whole world is nowdays mad of).

If that's the case with you, thanksfully there is a "software way" to disable ACPI via passing kernel options via GRUB and LILO boot loaders.

2. Disabling ACPI support on kernel boot level through GRUB boot loader config

There is a tiny difference in command to pass in order to disable  ACPI depending on the Linux installed  GRUB ver. 1.x or GRUB 2.x.

a) In GRUB 0.99 (GRUB version 1)

Edit file /etc/grub/menu.lst or /etc/grub/grub.conf (location differs across Linux distribution). Therein append:

acpi=off

to the end of kernel command line.

Here is an example of a kernel command line with ACPI not disabled (example taken from CentOS server grub.conf):

[root@centos ~]# grep -i title -A 4 /etc/grub/grub.conf
title Red Hat Enterprise Linux Server (2.6.18-36.el5)
root (hd0,0)
kernel /vmlinuz-2.6.18-36.el5 ro root=/dev/VolGroup00/LogVol00 console=ttyS0,115200n8
initrd /initrd-2.6.18-36.el5.img

The edited version of the file with acpi=off included should look like so:

title Red Hat Enterprise Linux Server (2.6.18-36.el5)
root (hd0,0)
kernel /vmlinuz-2.6.18-36.el5 ro root=/dev/VolGroup00/LogVol00 console=ttyS0,115200n8 acpi=off
initrd /initrd-2.6.18-36.el5.img

The kernel option root=/dev/VolGroup00/LogVol00 means the the server is configured to use LVM (Logical Volume Manager).

b) Disabling ACPI on GRUB version 1.99 +

This version is by default installed on newer Ubuntu and Debian Linux-es.

In grub 1.99 on latest Debian Squeeze, the file to edit is located in /boot/grub/grub.cfg. The file is more messy than with its predecessor menu.lst (grub 0.99).
Thanks God there is no need to directly edit the file (though this is possible), but on newer Linuces (as of time of writting the post), there is another simplied grub config file /etc/grub/config

Hence to add the acpi=off to 1.99 open /etc/grub/config find the line reading:

GRUB_CMDLINE_LINUX_DEFAULT="quiet"

and append the "acpi=off" option, e.g. the line has to change to:

GRUB_CMDLINE_LINUX_DEFAULT="quiet acpi=off"

On some servers it might be better to also disable APIC along with ACPI:

Just in case you don't know what is the difference between ACPI and APIC, here is a short explanation:

ACPI = Advanced Configuration and Power Interface

APIC = Advanced Programmable Interrupt Controllers

ACPI is the system that controls your dynamic speed fans, the power button behavior, sleep states, etc.

APIC is the replacement for the old PIC chip that used to come imbedded on motherboards that allowed you to setup interrupts for your soundcard, ide controllers, etc.

Hence on some machines experiencing still problems with even ACPI switched off, it is helpful  to disable the APIC support too, by using:

acpi=off noapic noacpi

Anyways, while doing the changes, be very very cautious or you might end up with un-boot-able server. Don't blame me if this happens :); be sure you have a backup option if server doesn't boot.

To assure faultless kernel boot, GRUB has ability to be configured to automatically load up a second kernel if 1st one fails to boot, if you need that read the grub documentation on that.

To load up the kernel with the new setting, give it a restart:

[root@centos ~]# shutdown -r now
....

3. Disable ACPI support on kernel boot time on Slackware or other Linuxes still booting kernel with LILO

Still, some Linux distros like Slackware, decided to keep the old way and use LILO (LInux LOader) as a default boot loader.

Disabling ACPI support in LILO is done through /etc/lilo.conf

By default in /etc/lilo.conf, there is a line:

append= acpi=on

it should be changed to:

append= acpi=off

Next to load up the new acpi disabled setting, lilo has to be reloaded:

slackware:~# /sbin/lilo -c /etc/lilo.conf
....

Finally a reboot is required:

slackware:~# reboot
....

(If you don't have a physical access or someone near the server you better not 🙂 )

4. Disable ACPI fan control support on a running Linux server without restart

This is the most secure work-around, to disabling the ACPI control over the machine CPU fan, however it has a downside that still the ACPI code will be loaded in the kernel and could cause kernel issues possibly in the long run – lets say the machine has uptime of more than 2 years…

The acpi support on a user level  is controlled by acpid or haldaemon (depending on the Linux distro), hence to disable the fan control on servers this services has to be switched off:

a) disabling ACPI on Debian and deb based Linux-es

As of time of writting on Debian Linux servers acpid (Advanced Configuration and Power Interface event daemon) is there to control how power management will be handled. To disable it stop it as a service (if running):

debian:~# /etc/init.d/acpid stop

To permanently remove acpid from boot up on system boot disable it with update-rc.d:

debian:~# update-rc.d acpid disable 2 3 4 5
update-rc.d: using dependency based boot sequencing
insserv: Script iptables is broken: incomplete LSB comment.
insserv: missing `Required-Start:' entry: please add even if empty.
insserv: warning: current start runlevel(s) (empty) of script `acpid' overwrites defaults (2 3 4 5).
insserv: warning: current stop runlevel(s) (2 3 4 5) of script `acpid' overwrites defaults (empty).
insserv: missing `Required-Start:' entry: please add even if empty.

b) disabling ACPI on RHEL, Fedora and other Redhat-s (also known as RedHacks 🙂 )

I'm not sure if this is safe,as many newer rpm based server system services,  might not work properly with haldaemon disabled.

Anyways you can give it a try if when it is stopped there are issues just bring it up again.

[root@rhel ~]# /etc/init.d/haldaemon stop

If all is fine with the haldaemon switched off (hope so), you can completely disable it to load on start up with:

[root@centos ~]# /sbin/chkconfig --level 2 3 4 5 haldaemon off

Disabling ACPI could increase a bit your server bills, but same time decrease losses from downtimes, so I guess it worths its costs 🙂

 

How to automatically reboot (restart) Debian GNU Lenny / Squeeze Linux on kernel panic, some general CPU overload or system crash

Monday, June 21st, 2010

If you are a system administrator, you have probably wondered at least once ohw to configure your Linux server to automatically reboot itself if it crashes, is going through a mass CPU overload, e.g. the server load average “hits the sky”.
I just learned from a nice article found here that there is a kernel variable which when enabled takes care to automatically restart a crashed server with the terrible Kernel Panic message we all know.

The variable I’m taking about is kernel.panic for instance kernel.panic = 20 would instruct your GNU Linux kernel to automatically reboot if it experiences a kernel panic system crash within a time limit of 20 seconds.

To start using the auto-reboot linux capabilities on a kernel panic occurance just set the variable to /etc/sysctl.conf

debian-server:~# echo 'kernel.panic = 20' >> /etc/sysctl.conf

Now we will also have to enable the variable to start being use on the system, so execute:

debian-server:~# sysctl -p There you go automatic system reboots on kernel panics is now on.
Now to further assure yourself the linux server you’re responsible of will automatically restart itself on a emergency situation like a system overload I suggest you check Watchdog

You might consider checking out this auto reboot tutorial which explains in simple words how watchdog is installed and configured.
On Debian installing and maintaining watchdog is really simple and comes to installing and enabling the watchdog system service, right afteryou made two changes in it’s configuration file /etc/watchdog.conf

To do so execute:

debian-server:~# apt-get install watchdog
debian-server:~# echo "file = /var/log/messages" >> /etc/watchdog.conf
debian-server:~# echo "watchdog-device = /dev/watchdog" >> /etc/watchdog.conf

Well that should be it, you might also need to load some kernel module to monitor your watchdog.
On my system the kernel modules related to watchdog are located in:

/lib/modules/2.6.26-2-amd64/kernel/drivers/watchdog/
If not then you should certainly try the software watchdog linux kernel module called softdog , to do so issue:
debian-server:~# /sbin/modprobe softdog

It’s best if you load the module while the softdog daemon is disabled.
If you consider auto loadig the softdog software watchdog kernel driver you should exec:

debian-server:~# echo 'softdog' >> /etc/modules

Finally a start of the watchdog is necessery:

 


debian-server:~# /etc/init.d/watchdog start
Stopping watchdog keepalive daemon....
Starting watchdog daemon....

That should be all your automatic system reboots should be now on! 🙂

How to auto restart CentOS Linux server with software watchdog (softdog) to reduce server downtime

Wednesday, August 10th, 2011

How to auto restart centos with software watchdog daemon to mitigate server downtimes, watchdog linux artistic logo

I’m in charge of dozen of Linux servers these days and therefore am required to restart many of the servers with a support ticket (because many of the Data Centers where the servers are co-located does not have a web interface or IPKVM connected to the server for that purpose). Therefore the server restart requests in case of crash sometimes gets processed in few hours or in best case in at least half an hour.

I’m aware of the existence of Hardware Watchdog devices, which are capable to detect if a server is hanged and auto-restart it, however the servers I administrate does not have Hardware support for Watchdog timer.

Thanksfully there is a free software project called Watchdog which is easily configured and mitigates the terrible downtimes caused every now and then by a server crash and respective delays by tech support in Data Centers.

I’ve recently blogged on the topic of Debian Linux auto-restart in case of kernel panic , however now i had to conifgure watchdog on some dozen of CentOS Linux servers.

It appeared installation & configuration of Watchdog on CentOS is a piece of cake and comes to simply following few easy steps, which I’ll explain quickly in this post:

1. Install with yum watchdog to CentOS

[root@centos:/etc/init.d ]# yum install watchdog
...

2. Add to configuration a log file to log watchdog activities and location of the watchdog device

The quickest way to add this two is to use echo to append it in /etc/watchdog.conf:

[root@centos:/etc/init.d ]# echo 'file = /var/log/messages' >> /etc/watchdog.conf
echo 'watchdog-device = /dev/watchdog' >> /etc/watchdog.conf

3. Load the softdog kernel module to initialize the software watchdog via /dev/watchdog

[root@centos:/etc/init.d ]# /sbin/modprobe softdog

Initialization of softdog should be indicated by a line in dmesg kernel log like the one above:

[root@centos:/etc/init.d ]# dmesg |grep -i watchdog
Software Watchdog Timer: 0.07 initialized. soft_noboot=0 soft_margin=60 sec (nowayout= 0)

4. Include the softdog kernel module to load on CentOS boot up

This is necessery, because otherwise after reboot the softdog would not be auto initialized and without it being initialized, the watchdog daemon service could not function as it does automatically auto reboots the server if the /dev/watchdog disappears.

It’s better that the softdog module is not loaded via /etc/rc.local but the default CentOS methodology to load module from /etc/rc.module is used:

[root@centos:/etc/init.d ]# echo modprobe softdog >> /etc/rc.modules
[root@centos:/etc/init.d ]# chmod +x /etc/rc.modules

5. Start the watchdog daemon service

The succesful intialization of softdog in step 4, should have provided the system with /dev/watchdog, before proceeding with starting up the watchdog daemon it’s wise to first check if /dev/watchdog is existent on the system. Here is how:

[root@centos:/etc/init.d ]# ls -al /dev/watchdogcrw------- 1 root root 10, 130 Aug 10 14:03 /dev/watchdog

Being sure, that /dev/watchdog is there, I’ll start the watchdog service.

[root@centos:/etc/init.d ]# service watchdog restart
...

Very important note to make here is that you should never ever configure watchdog service to run on boot time with chkconfig. In other words the status from chkconfig for watchdog boot on all levels should be off like so:

[root@centos:/etc/init.d ]# chkconfig --list |grep -i watchdog
watchdog 0:off 1:off 2:off 3:off 4:off 5:off 6:off

Enabling the watchdog from the chkconfig will cause watchdog to automatically restart the system as it will probably start the watchdog daemon before the softdog module is initialized. As watchdog will be unable to read the /dev/watchdog it will though the system has hanged even though the system might be in a boot process. Therefore it will end up in an endless loops of reboots which can only be fixed in a linux single user mode!!! Once again BEWARE, never ever activate watchdog via chkconfig!

Next step to be absolutely sure that watchdog device is running it can be checked with normal ps command:

[root@centos:/etc/init.d ]# ps aux|grep -i watchdog
root@hosting1-fr [~]# ps axu|grep -i watch|grep -v greproot 18692 0.0 0.0 1816 1812 ? SNLs 14:03 0:00 /usr/sbin/watchdog
root 25225 0.0 0.0 0 0 ? ZN 17:25 0:00 [watchdog] <defunct>

You have probably noticed the defunct state of watchdog, consider that as absolutely normal, above output indicates that now watchdog is properly running on the host and waiting to auto reboot in case of sudden /dev/watchdog disappearance.

As a last step before, after being sure its initialized properly, it’s necessery to add watchdog to run on boot time via /etc/rc.local post init script, like so:

[root@centos:/etc/init.d ]# echo 'echo /sbin/service watchdog start' >> /etc/rc.local

Now enjoy, watchdog is up and running and will automatically restart the CentOS host 😉