ブートデバッグ

提供: ArchWiki
2015年2月14日 (土) 18:28時点におけるKusakata (トーク | 投稿記録)による版 (ページの作成:「Category:ブートプロセス Category:システムリカバリ en:Boot debugging it:Boot Debugging {{Related articles start}} {{Related|Arch ブート...」)
(差分) ← 古い版 | 最新版 (差分) | 新しい版 → (差分)
ナビゲーションに移動 検索に移動

関連記事

様々なデバッグ出力のレベルで既存のシステムを簡単に出来るように、カーネルはあらゆる高度な設定を行うための便利な手段を用意しています [1]。ブートローダーの設定にエントリを追加するだけで、とても簡単に複数のレベルのデバッグを作成することが可能です。電源やハードウェアの問題によって、後で問題が起こった時に、手数を減らすことができ、また、システムについて習熟したいと思った時にデバッグ出力に勝るものはありません。

まず、カーネルパラメータの記事を読んでブートローダーにカーネルパラメータを追加する方法を見て下さい。

デバッグレベル

Light Debug

A quick way to see more verbose messages on your console is to boot up your bootloader entry after appending verbose to the kernel line.

Medium Debug

Adding the debug kernel parameter to your kernel line is recognized by a lot of Linux internals and enables quite a bit of debugging compared to the default.

Heavy Debug

An even more impressive kernel parameter is the ignore_loglevel, which causes the system to ignore any loglevel and keeps the internal loglevel at the maximum debugging level, basically rendering dmesg unable to lower the debug level.

Extreme Debug

If the "Heavy Debug" seemed like a lot of output, that is about half of the logging that occurs with kernel parameters: 

debug ignore_loglevel log_buf_len=10M print_fatal_signals=1 LOGLEVEL=8 earlyprintk=vga,keep sched_debug

Or for EFI systems 

debug ignore_loglevel log_buf_len=10M print_fatal_signals=1 LOGLEVEL=8 earlyprintk=efi,keep sched_debug

This does a couple things:

  • It uses the earlyprintk parameter to setup your kernel for early printing of messages to your vga screen or EFI framebuffer.
  • The keep parameter just lets it stay on the screen longer. This will let you see logs that normally are hidden due to the boot-up process.
  • log_buf_len=10M changes the log buffer length to 10MB
  • Instructs that any fatal signals be printed with print_fatal_signals
  • With the last one, sched_debug, you can look up in the very excellent kernel documentation on kernel parameters.

Insane Debug

The first few debugging kernel parameters turn on really verbose debugging. This kind of debugging is absolutely critical if you want to max out your system or just learn more about what is going on behind the scenes. But there is a final trick that is my favorite, which is the ability to set both environment variables and, more importantly, module parameters at boot.

As an example, here is some truly insane debugging kernel parameters. Note that the actual GRUB entry is all on one line. This turns on many devices to be at the absolute max debug level. The machine may be slower than a TI-89 calculator (See Improve boot performance). 

rootwait ignore_loglevel debug debug_locks_verbose=1 sched_debug initcall_debug 
mminit_loglevel=4 udev.log_priority=8 loglevel=8 earlyprintk=vga,keep log_buf_len=10M
print_fatal_signals=1 apm.debug=Y i8042.debug=Y drm.debug=1 scsi_logging_level=1
usbserial.debug=Y option.debug=Y pl2303.debug=Y firewire_ohci.debug=1 hid.debug=1
pci_hotplug.debug=Y pci_hotplug.debug_acpi=Y shpchp.shpchp_debug=Y apic=debug
show_lapic=all hpet=verbose lmb=debug pause_on_oops=5 panic=10 sysrq_always_enabled

A couple key items are sysrq_always_enabled which forces on the SysRq magic, which really is a lifesaver when debugging at this level, as your machine will freeze/stop-responding sometimes, and it is nice to use SysRq to kill all tasks, change the log level, unmount all file systems, or do a hard reboot. Another key parameter is initcall_debug, which debugs the init process in excruciating detail, which can be very useful at times. The last parameter I find very useful is udev.log_priority=8 to turn on udev logging.

デバッグの方法

モジュールパラメータ

カーネルモジュールを見て下さい。

Break into Init

For instance, if you add break=y to your kernel command line, init will pause early in the boot process (after loading modules) and launch an interactive sh shell which can be used for troubleshooting purposes. Normal boot continues after logging out. This is very similar to the shell that shows up if your computer gets turned off before it is able to shutdown properly. But using this parameter lets you enter into this mode differently at will.

init のデバッグ

The awesome parameter udev.log_priority=8 does the same thing as editing the file /etc/udev/udev.conf except it executes earlier, turning on debugging output for udev. If you want to know your hardware, that is the key parameter right there. Another trick is if you change the /etc/udev/udev.conf to be verbose, then you can make your initramfs image include that file to turn on verbose udev debugging by adding it to your /etc/mkinitcpio.conf like: 

FILES="/etc/modprobe.d/modprobe.conf /etc/udev/udev.conf"

which, on Arch, is as easy as 

# mkinitcpio -p linux

Debugging udev is important because the initramfs performs a root change at the end of its run to usually launch a program like /sbin/init as part of a chroot, and unless the new file system has a valid /dev directory, udev must be initialized before invoking chroot in order to provide /dev/console. 

exec chroot . /sbin/init <dev/console >dev/console 2>&1

So basically, you are not able to view the logs that are generated before /dev/console is initialized by udev or by a special initramfs you compiled yourself. One method the kernel developers use to be able to still get the log messages generated before /dev/console is available, is to provide an alternative console that you can enable or disable from GRUB.

netconsole

If you read through the kernel documentation regarding debugging, you will read about netconsole, which can be compiled into your kernel or loaded at runtime as a module. Having a netconsole entry in your kernel parameters is most excellent for debugging slower computers like old laptops or thin-clients. It is easy to use:

  1. Set up a 2nd computer (running Arch) to accept syslog requests on a remote port, which is one line in syslog.conf.
  2. Then you could use a log-color-parser like ccze to view all syslog logs, or just tail your /var/log/everything.log file.
  3. On your laptop, boot up and add netconsole=514@10.0.0.2/12:34:56:78:9a:bc debug ignore_loglevel into kernel parameters.
  4. You will start seeing as much logging as you want on your syslog system. This logging lets you view even earlier log output than is available with the earlyprintk=vga kernel parameter, as netconsole is used by kernel hackers and developers, so it is very powerful.

cmdline をハイジャック

If you do not have access to GRUB or the kernel boot-time command line, like on a server or virtual machine, as long as you have root permissions, you can still enable this kind of simplistic verbose logging using a neat hack. While you cannot modify the /proc/cmdline even as root, you can place your own cmdline file on top of /proc/cmdline, so that accessing /proc/cmdline actually accesses your file instead.

For example, if I cat /proc/cmdline, I have the following: 

root=/dev/disk/by-label/ROOT ro console=tty1 logo.nologo quiet

So I use a simple sed command to replace quiet with verbose like: 

sed '/root=/s/ quiet/ verbose/g' /proc/cmdline > /root/cmdline

Then I bind mount /root/cmdline so that it becomes /proc/cmdline, using the -n option to mount, so that this mount will not be recorded in the system's mtab. 

mount -n --bind -o ro /root/cmdline /proc/cmdline

Now if I cat /proc/cmdline, I have the following: 

root=/dev/disk/by-label/ROOT ro console=tty1 logo.nologo verbose

トラブルシューティング

Output cut off

Cut off when switching resolution

If the first part of your boot output disappears when the resolution changes you can try to disable KMS for debugging.

Scrolling limited

If your boot output is limited to 4 or 5 pages and you need more so you can use the debug parameters above, add the following item to your kernel parameters 

fbcon=scrollback:512k

You can read more about this in Scrollback buffer.

Repairing with Arch live CD

In case grub is unable to boot your kernel, or if your initramfs is broken, you can boot into a safe system using an Arch live CD. Once finished with repairs, unmount the broken system, and reboot.

Mounting and chrooting a broken system

Once booted and at a console prompt, use the following to mount and repair your broken system (where /dev/sda3 is / and /dev/sda1 is /boot):

First, create the mount point, and mount your root / file system to it; then, cd into it. 

# mkdir /mnt/arch
# mount /dev/sda3 /mnt/arch
# cd /mnt/arch

Now, create the proc, sysfs, and dev file systems 

# mount -t proc proc proc/
# mount -t sysfs sys sys/
# mount -o bind /dev dev/

Next, mount the boot partition if you use one. 

# mount /dev/sda1 boot/

Finally, chroot into /mnt/arch which will become /. 

# chroot .

Turn on networking.

Reinstalling with pacman

The author uses pacman in a chrooted broken system to reinstall the kernel, GRUB, initramfs, udev, and any other packages that may be broken and/or needed to get the system up and running.

This will reinstall the kernel and initramfs so check that /etc/mkinitcpio.conf is correct or remove the file entirely and re-install mkinitcpio. 

# pacman -Syyu mkinitcpio linux udev

Afterwards, unmount and reboot.

参照

「https://wiki.archlinux.jp/index.php?title=ブートデバッグ&oldid=2226」から取得