summaryrefslogtreecommitdiffstats
path: root/Documentation/kdump/kdump.txt
blob: 212cf3c21abf8849e39b5964607263e23da8e77f (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
Documentation for kdump - the kexec-based crash dumping solution
================================================================

DESIGN
======

Kdump uses kexec to reboot to a second kernel whenever a dump needs to be
taken. This second kernel is booted with very little memory. The first kernel
reserves the section of memory that the second kernel uses. This ensures that
on-going DMA from the first kernel does not corrupt the second kernel.

All the necessary information about Core image is encoded in ELF format and
stored in reserved area of memory before crash. Physical address of start of
ELF header is passed to new kernel through command line parameter elfcorehdr=.

On i386, the first 640 KB of physical memory is needed to boot, irrespective
of where the kernel loads. Hence, this region is backed up by kexec just before
rebooting into the new kernel.

In the second kernel, "old memory" can be accessed in two ways.

- The first one is through a /dev/oldmem device interface. A capture utility
  can read the device file and write out the memory in raw format. This is raw
  dump of memory and analysis/capture tool should be intelligent enough to
  determine where to look for the right information. ELF headers (elfcorehdr=)
  can become handy here.

- The second interface is through /proc/vmcore. This exports the dump as an ELF
  format file which can be written out using any file copy command
  (cp, scp, etc). Further, gdb can be used to perform limited debugging on
  the dump file. This method ensures methods ensure that there is correct
  ordering of the dump pages (corresponding to the first 640 KB that has been
  relocated).

SETUP
=====

1) Download the upstream kexec-tools userspace package from
   http://www.xmission.com/~ebiederm/files/kexec/kexec-tools-1.101.tar.gz.

   Apply the latest consolidated kdump patch on top of kexec-tools-1.101
   from http://lse.sourceforge.net/kdump/. This arrangment has been made
   till all the userspace patches supporting kdump are integrated with
   upstream kexec-tools userspace.

2) Download and build the appropriate (2.6.13-rc1 onwards) vanilla kernels.
   Two kernels need to be built in order to get this feature working.
   Following are the steps to properly configure the two kernels specific
   to kexec and kdump features:

  A) First kernel or regular kernel:
  ----------------------------------
   a) Enable "kexec system call" feature (in Processor type and features).
      CONFIG_KEXEC=y
   b) Enable "sysfs file system support" (in Pseudo filesystems).
      CONFIG_SYSFS=y
   c) make
   d) Boot into first kernel with the command line parameter "crashkernel=Y@X".
      Use appropriate values for X and Y. Y denotes how much memory to reserve
      for the second kernel, and X denotes at what physical address the
      reserved memory section starts. For example: "crashkernel=64M@16M".


  B) Second kernel or dump capture kernel:
  ---------------------------------------
   a) For i386 architecture enable Highmem support
      CONFIG_HIGHMEM=y
   b) Enable "kernel crash dumps" feature (under "Processor type and features")
      CONFIG_CRASH_DUMP=y
   c) Make sure a suitable value for "Physical address where the kernel is
      loaded" (under "Processor type and features"). By default this value
      is 0x1000000 (16MB) and it should be same as X (See option d above),
      e.g., 16 MB or 0x1000000.
      CONFIG_PHYSICAL_START=0x1000000
   d) Enable "/proc/vmcore support" (Optional, under "Pseudo filesystems").
      CONFIG_PROC_VMCORE=y

3) After booting to regular kernel or first kernel, load the second kernel
   using the following command:

   kexec -p <second-kernel> --args-linux --elf32-core-headers
   --append="root=<root-dev> init 1 irqpoll maxcpus=1"

   Notes:
   ======
     i) <second-kernel> has to be a vmlinux image ie uncompressed elf image.
        bzImage will not work, as of now.
    ii) --args-linux has to be speicfied as if kexec it loading an elf image,
        it needs to know that the arguments supplied are of linux type.
   iii) By default ELF headers are stored in ELF64 format to support systems
        with more than 4GB memory. Option --elf32-core-headers forces generation
        of ELF32 headers. The reason for this option being, as of now gdb can
        not open vmcore file with ELF64 headers on a 32 bit systems. So ELF32
        headers can be used if one has non-PAE systems and hence memory less
        than 4GB.
    iv) Specify "irqpoll" as command line parameter. This reduces driver
         initialization failures in second kernel due to shared interrupts.
     v) <root-dev> needs to be specified in a format corresponding to the root
        device name in the output of mount command.
    vi) If you have built the drivers required to mount root file system as
        modules in <second-kernel>, then, specify
        --initrd=<initrd-for-second-kernel>.
   vii) Specify maxcpus=1 as, if during first kernel run, if panic happens on
        non-boot cpus, second kernel doesn't seem to be boot up all the cpus.
        The other option is to always built the second kernel without SMP
        support ie CONFIG_SMP=n

4) After successfully loading the second kernel as above, if a panic occurs
   system reboots into the second kernel. A module can be written to force
   the panic or "ALT-SysRq-c" can be used initiate a crash dump for testing
   purposes.

5) Once the second kernel has booted, write out the dump file using

   cp /proc/vmcore <dump-file>

   Dump memory can also be accessed as a /dev/oldmem device for a linear/raw
   view.  To create the device, type:

   mknod /dev/oldmem c 1 12

   Use "dd" with suitable options for count, bs and skip to access specific
   portions of the dump.

   Entire memory:  dd if=/dev/oldmem of=oldmem.001


ANALYSIS
========
Limited analysis can be done using gdb on the dump file copied out of
/proc/vmcore. Use vmlinux built with -g and run

  gdb vmlinux <dump-file>

Stack trace for the task on processor 0, register display, memory display
work fine.

Note: gdb cannot analyse core files generated in ELF64 format for i386.

Latest "crash" (crash-4.0-2.18) as available on Dave Anderson's site
http://people.redhat.com/~anderson/ works well with kdump format.


TODO
====
1) Provide a kernel pages filtering mechanism so that core file size is not
   insane on systems having huge memory banks.
2) Relocatable kernel can help in maintaining multiple kernels for crashdump
   and same kernel as the first kernel can be used to capture the dump.


CONTACT
=======
Vivek Goyal (vgoyal@in.ibm.com)
Maneesh Soni (maneesh@in.ibm.com)