xref: /linux/Documentation/admin-guide/kdump/kdump.rst (revision f4fee216df7d28b87d1c9cc60bcebfecb51c1a05)
1================================================================
2Documentation for Kdump - The kexec-based Crash Dumping Solution
3================================================================
4
5This document includes overview, setup, installation, and analysis
6information.
7
8Overview
9========
10
11Kdump uses kexec to quickly boot to a dump-capture kernel whenever a
12dump of the system kernel's memory needs to be taken (for example, when
13the system panics). The system kernel's memory image is preserved across
14the reboot and is accessible to the dump-capture kernel.
15
16You can use common commands, such as cp, scp or makedumpfile to copy
17the memory image to a dump file on the local disk, or across the network
18to a remote system.
19
20Kdump and kexec are currently supported on the x86, x86_64, ppc64,
21s390x, arm and arm64 architectures.
22
23When the system kernel boots, it reserves a small section of memory for
24the dump-capture kernel. This ensures that ongoing Direct Memory Access
25(DMA) from the system kernel does not corrupt the dump-capture kernel.
26The kexec -p command loads the dump-capture kernel into this reserved
27memory.
28
29On x86 machines, the first 640 KB of physical memory is needed for boot,
30regardless of where the kernel loads. For simpler handling, the whole
31low 1M is reserved to avoid any later kernel or device driver writing
32data into this area. Like this, the low 1M can be reused as system RAM
33by kdump kernel without extra handling.
34
35On PPC64 machines first 32KB of physical memory is needed for booting
36regardless of where the kernel is loaded and to support 64K page size
37kexec backs up the first 64KB memory.
38
39For s390x, when kdump is triggered, the crashkernel region is exchanged
40with the region [0, crashkernel region size] and then the kdump kernel
41runs in [0, crashkernel region size]. Therefore no relocatable kernel is
42needed for s390x.
43
44All of the necessary information about the system kernel's core image is
45encoded in the ELF format, and stored in a reserved area of memory
46before a crash. The physical address of the start of the ELF header is
47passed to the dump-capture kernel through the elfcorehdr= boot
48parameter. Optionally the size of the ELF header can also be passed
49when using the elfcorehdr=[size[KMG]@]offset[KMG] syntax.
50
51With the dump-capture kernel, you can access the memory image through
52/proc/vmcore. This exports the dump as an ELF-format file that you can
53write out using file copy commands such as cp or scp. You can also use
54makedumpfile utility to analyze and write out filtered contents with
55options, e.g with '-d 31' it will only write out kernel data. Further,
56you can use analysis tools such as the GNU Debugger (GDB) and the Crash
57tool to debug the dump file. This method ensures that the dump pages are
58correctly ordered.
59
60Setup and Installation
61======================
62
63Install kexec-tools
64-------------------
65
661) Login as the root user.
67
682) Download the kexec-tools user-space package from the following URL:
69
70http://kernel.org/pub/linux/utils/kernel/kexec/kexec-tools.tar.gz
71
72This is a symlink to the latest version.
73
74The latest kexec-tools git tree is available at:
75
76- git://git.kernel.org/pub/scm/utils/kernel/kexec/kexec-tools.git
77- http://www.kernel.org/pub/scm/utils/kernel/kexec/kexec-tools.git
78
79There is also a gitweb interface available at
80http://www.kernel.org/git/?p=utils/kernel/kexec/kexec-tools.git
81
82More information about kexec-tools can be found at
83http://horms.net/projects/kexec/
84
853) Unpack the tarball with the tar command, as follows::
86
87	tar xvpzf kexec-tools.tar.gz
88
894) Change to the kexec-tools directory, as follows::
90
91	cd kexec-tools-VERSION
92
935) Configure the package, as follows::
94
95	./configure
96
976) Compile the package, as follows::
98
99	make
100
1017) Install the package, as follows::
102
103	make install
104
105
106Build the system and dump-capture kernels
107-----------------------------------------
108There are two possible methods of using Kdump.
109
1101) Build a separate custom dump-capture kernel for capturing the
111   kernel core dump.
112
1132) Or use the system kernel binary itself as dump-capture kernel and there is
114   no need to build a separate dump-capture kernel. This is possible
115   only with the architectures which support a relocatable kernel. As
116   of today, i386, x86_64, ppc64, arm and arm64 architectures support
117   relocatable kernel.
118
119Building a relocatable kernel is advantageous from the point of view that
120one does not have to build a second kernel for capturing the dump. But
121at the same time one might want to build a custom dump capture kernel
122suitable to his needs.
123
124Following are the configuration setting required for system and
125dump-capture kernels for enabling kdump support.
126
127System kernel config options
128----------------------------
129
1301) Enable "kexec system call" or "kexec file based system call" in
131   "Processor type and features."::
132
133	CONFIG_KEXEC=y or CONFIG_KEXEC_FILE=y
134
135   And both of them will select KEXEC_CORE::
136
137	CONFIG_KEXEC_CORE=y
138
139   Subsequently, CRASH_CORE is selected by KEXEC_CORE::
140
141	CONFIG_CRASH_CORE=y
142
1432) Enable "sysfs file system support" in "Filesystem" -> "Pseudo
144   filesystems." This is usually enabled by default::
145
146	CONFIG_SYSFS=y
147
148   Note that "sysfs file system support" might not appear in the "Pseudo
149   filesystems" menu if "Configure standard kernel features (expert users)"
150   is not enabled in "General Setup." In this case, check the .config file
151   itself to ensure that sysfs is turned on, as follows::
152
153	grep 'CONFIG_SYSFS' .config
154
1553) Enable "Compile the kernel with debug info" in "Kernel hacking."::
156
157	CONFIG_DEBUG_INFO=Y
158
159   This causes the kernel to be built with debug symbols. The dump
160   analysis tools require a vmlinux with debug symbols in order to read
161   and analyze a dump file.
162
163Dump-capture kernel config options (Arch Independent)
164-----------------------------------------------------
165
1661) Enable "kernel crash dumps" support under "Processor type and
167   features"::
168
169	CONFIG_CRASH_DUMP=y
170
1712) Enable "/proc/vmcore support" under "Filesystems" -> "Pseudo filesystems"::
172
173	CONFIG_PROC_VMCORE=y
174
175   (CONFIG_PROC_VMCORE is set by default when CONFIG_CRASH_DUMP is selected.)
176
177Dump-capture kernel config options (Arch Dependent, i386 and x86_64)
178--------------------------------------------------------------------
179
1801) On i386, enable high memory support under "Processor type and
181   features"::
182
183	CONFIG_HIGHMEM64G=y
184
185   or::
186
187	CONFIG_HIGHMEM4G
188
1892) With CONFIG_SMP=y, usually nr_cpus=1 need specified on the kernel
190   command line when loading the dump-capture kernel because one
191   CPU is enough for kdump kernel to dump vmcore on most of systems.
192
193   However, you can also specify nr_cpus=X to enable multiple processors
194   in kdump kernel.
195
196   With CONFIG_SMP=n, the above things are not related.
197
1983) A relocatable kernel is suggested to be built by default. If not yet,
199   enable "Build a relocatable kernel" support under "Processor type and
200   features"::
201
202	CONFIG_RELOCATABLE=y
203
2044) Use a suitable value for "Physical address where the kernel is
205   loaded" (under "Processor type and features"). This only appears when
206   "kernel crash dumps" is enabled. A suitable value depends upon
207   whether kernel is relocatable or not.
208
209   If you are using a relocatable kernel use CONFIG_PHYSICAL_START=0x100000
210   This will compile the kernel for physical address 1MB, but given the fact
211   kernel is relocatable, it can be run from any physical address hence
212   kexec boot loader will load it in memory region reserved for dump-capture
213   kernel.
214
215   Otherwise it should be the start of memory region reserved for
216   second kernel using boot parameter "crashkernel=Y@X". Here X is
217   start of memory region reserved for dump-capture kernel.
218   Generally X is 16MB (0x1000000). So you can set
219   CONFIG_PHYSICAL_START=0x1000000
220
2215) Make and install the kernel and its modules. DO NOT add this kernel
222   to the boot loader configuration files.
223
224Dump-capture kernel config options (Arch Dependent, ppc64)
225----------------------------------------------------------
226
2271) Enable "Build a kdump crash kernel" support under "Kernel" options::
228
229	CONFIG_CRASH_DUMP=y
230
2312)   Enable "Build a relocatable kernel" support::
232
233	CONFIG_RELOCATABLE=y
234
235   Make and install the kernel and its modules.
236
237Dump-capture kernel config options (Arch Dependent, arm)
238----------------------------------------------------------
239
240-   To use a relocatable kernel,
241    Enable "AUTO_ZRELADDR" support under "Boot" options::
242
243	AUTO_ZRELADDR=y
244
245Dump-capture kernel config options (Arch Dependent, arm64)
246----------------------------------------------------------
247
248- Please note that kvm of the dump-capture kernel will not be enabled
249  on non-VHE systems even if it is configured. This is because the CPU
250  will not be reset to EL2 on panic.
251
252crashkernel syntax
253===========================
2541) crashkernel=size@offset
255
256   Here 'size' specifies how much memory to reserve for the dump-capture kernel
257   and 'offset' specifies the beginning of this reserved memory. For example,
258   "crashkernel=64M@16M" tells the system kernel to reserve 64 MB of memory
259   starting at physical address 0x01000000 (16MB) for the dump-capture kernel.
260
261   The crashkernel region can be automatically placed by the system
262   kernel at run time. This is done by specifying the base address as 0,
263   or omitting it all together::
264
265         crashkernel=256M@0
266
267   or::
268
269         crashkernel=256M
270
271   If the start address is specified, note that the start address of the
272   kernel will be aligned to a value (which is Arch dependent), so if the
273   start address is not then any space below the alignment point will be
274   wasted.
275
2762) range1:size1[,range2:size2,...][@offset]
277
278   While the "crashkernel=size[@offset]" syntax is sufficient for most
279   configurations, sometimes it's handy to have the reserved memory dependent
280   on the value of System RAM -- that's mostly for distributors that pre-setup
281   the kernel command line to avoid a unbootable system after some memory has
282   been removed from the machine.
283
284   The syntax is::
285
286       crashkernel=<range1>:<size1>[,<range2>:<size2>,...][@offset]
287       range=start-[end]
288
289   For example::
290
291       crashkernel=512M-2G:64M,2G-:128M
292
293   This would mean:
294
295       1) if the RAM is smaller than 512M, then don't reserve anything
296          (this is the "rescue" case)
297       2) if the RAM size is between 512M and 2G (exclusive), then reserve 64M
298       3) if the RAM size is larger than 2G, then reserve 128M
299
3003) crashkernel=size,high and crashkernel=size,low
301
302   If memory above 4G is preferred, crashkernel=size,high can be used to
303   fulfill that. With it, physical memory is allowed to be allocated from top,
304   so could be above 4G if system has more than 4G RAM installed. Otherwise,
305   memory region will be allocated below 4G if available.
306
307   When crashkernel=X,high is passed, kernel could allocate physical memory
308   region above 4G, low memory under 4G is needed in this case. There are
309   three ways to get low memory:
310
311      1) Kernel will allocate at least 256M memory below 4G automatically
312         if crashkernel=Y,low is not specified.
313      2) Let user specify low memory size instead.
314      3) Specified value 0 will disable low memory allocation::
315
316            crashkernel=0,low
317
318Boot into System Kernel
319-----------------------
3201) Update the boot loader (such as grub, yaboot, or lilo) configuration
321   files as necessary.
322
3232) Boot the system kernel with the boot parameter "crashkernel=Y@X".
324
325   On x86 and x86_64, use "crashkernel=Y[@X]". Most of the time, the
326   start address 'X' is not necessary, kernel will search a suitable
327   area. Unless an explicit start address is expected.
328
329   On ppc64, use "crashkernel=128M@32M".
330
331   On s390x, typically use "crashkernel=xxM". The value of xx is dependent
332   on the memory consumption of the kdump system. In general this is not
333   dependent on the memory size of the production system.
334
335   On arm, the use of "crashkernel=Y@X" is no longer necessary; the
336   kernel will automatically locate the crash kernel image within the
337   first 512MB of RAM if X is not given.
338
339   On arm64, use "crashkernel=Y[@X]".  Note that the start address of
340   the kernel, X if explicitly specified, must be aligned to 2MiB (0x200000).
341
342Load the Dump-capture Kernel
343============================
344
345After booting to the system kernel, dump-capture kernel needs to be
346loaded.
347
348Based on the architecture and type of image (relocatable or not), one
349can choose to load the uncompressed vmlinux or compressed bzImage/vmlinuz
350of dump-capture kernel. Following is the summary.
351
352For i386 and x86_64:
353
354	- Use bzImage/vmlinuz if kernel is relocatable.
355	- Use vmlinux if kernel is not relocatable.
356
357For ppc64:
358
359	- Use vmlinux
360
361For s390x:
362
363	- Use image or bzImage
364
365For arm:
366
367	- Use zImage
368
369For arm64:
370
371	- Use vmlinux or Image
372
373If you are using an uncompressed vmlinux image then use following command
374to load dump-capture kernel::
375
376   kexec -p <dump-capture-kernel-vmlinux-image> \
377   --initrd=<initrd-for-dump-capture-kernel> --args-linux \
378   --append="root=<root-dev> <arch-specific-options>"
379
380If you are using a compressed bzImage/vmlinuz, then use following command
381to load dump-capture kernel::
382
383   kexec -p <dump-capture-kernel-bzImage> \
384   --initrd=<initrd-for-dump-capture-kernel> \
385   --append="root=<root-dev> <arch-specific-options>"
386
387If you are using a compressed zImage, then use following command
388to load dump-capture kernel::
389
390   kexec --type zImage -p <dump-capture-kernel-bzImage> \
391   --initrd=<initrd-for-dump-capture-kernel> \
392   --dtb=<dtb-for-dump-capture-kernel> \
393   --append="root=<root-dev> <arch-specific-options>"
394
395If you are using an uncompressed Image, then use following command
396to load dump-capture kernel::
397
398   kexec -p <dump-capture-kernel-Image> \
399   --initrd=<initrd-for-dump-capture-kernel> \
400   --append="root=<root-dev> <arch-specific-options>"
401
402Following are the arch specific command line options to be used while
403loading dump-capture kernel.
404
405For i386 and x86_64:
406
407	"1 irqpoll nr_cpus=1 reset_devices"
408
409For ppc64:
410
411	"1 maxcpus=1 noirqdistrib reset_devices"
412
413For s390x:
414
415	"1 nr_cpus=1 cgroup_disable=memory"
416
417For arm:
418
419	"1 maxcpus=1 reset_devices"
420
421For arm64:
422
423	"1 nr_cpus=1 reset_devices"
424
425Notes on loading the dump-capture kernel:
426
427* By default, the ELF headers are stored in ELF64 format to support
428  systems with more than 4GB memory. On i386, kexec automatically checks if
429  the physical RAM size exceeds the 4 GB limit and if not, uses ELF32.
430  So, on non-PAE systems, ELF32 is always used.
431
432  The --elf32-core-headers option can be used to force the generation of ELF32
433  headers. This is necessary because GDB currently cannot open vmcore files
434  with ELF64 headers on 32-bit systems.
435
436* The "irqpoll" boot parameter reduces driver initialization failures
437  due to shared interrupts in the dump-capture kernel.
438
439* You must specify <root-dev> in the format corresponding to the root
440  device name in the output of mount command.
441
442* Boot parameter "1" boots the dump-capture kernel into single-user
443  mode without networking. If you want networking, use "3".
444
445* We generally don't have to bring up a SMP kernel just to capture the
446  dump. Hence generally it is useful either to build a UP dump-capture
447  kernel or specify maxcpus=1 option while loading dump-capture kernel.
448  Note, though maxcpus always works, you had better replace it with
449  nr_cpus to save memory if supported by the current ARCH, such as x86.
450
451* You should enable multi-cpu support in dump-capture kernel if you intend
452  to use multi-thread programs with it, such as parallel dump feature of
453  makedumpfile. Otherwise, the multi-thread program may have a great
454  performance degradation. To enable multi-cpu support, you should bring up an
455  SMP dump-capture kernel and specify maxcpus/nr_cpus options while loading it.
456
457* For s390x there are two kdump modes: If a ELF header is specified with
458  the elfcorehdr= kernel parameter, it is used by the kdump kernel as it
459  is done on all other architectures. If no elfcorehdr= kernel parameter is
460  specified, the s390x kdump kernel dynamically creates the header. The
461  second mode has the advantage that for CPU and memory hotplug, kdump has
462  not to be reloaded with kexec_load().
463
464* For s390x systems with many attached devices the "cio_ignore" kernel
465  parameter should be used for the kdump kernel in order to prevent allocation
466  of kernel memory for devices that are not relevant for kdump. The same
467  applies to systems that use SCSI/FCP devices. In that case the
468  "allow_lun_scan" zfcp module parameter should be set to zero before
469  setting FCP devices online.
470
471Kernel Panic
472============
473
474After successfully loading the dump-capture kernel as previously
475described, the system will reboot into the dump-capture kernel if a
476system crash is triggered.  Trigger points are located in panic(),
477die(), die_nmi() and in the sysrq handler (ALT-SysRq-c).
478
479The following conditions will execute a crash trigger point:
480
481If a hard lockup is detected and "NMI watchdog" is configured, the system
482will boot into the dump-capture kernel ( die_nmi() ).
483
484If die() is called, and it happens to be a thread with pid 0 or 1, or die()
485is called inside interrupt context or die() is called and panic_on_oops is set,
486the system will boot into the dump-capture kernel.
487
488On powerpc systems when a soft-reset is generated, die() is called by all cpus
489and the system will boot into the dump-capture kernel.
490
491For testing purposes, you can trigger a crash by using "ALT-SysRq-c",
492"echo c > /proc/sysrq-trigger" or write a module to force the panic.
493
494Write Out the Dump File
495=======================
496
497After the dump-capture kernel is booted, write out the dump file with
498the following command::
499
500   cp /proc/vmcore <dump-file>
501
502or use scp to write out the dump file between hosts on a network, e.g::
503
504   scp /proc/vmcore remote_username@remote_ip:<dump-file>
505
506You can also use makedumpfile utility to write out the dump file
507with specified options to filter out unwanted contents, e.g::
508
509   makedumpfile -l --message-level 1 -d 31 /proc/vmcore <dump-file>
510
511Analysis
512========
513
514Before analyzing the dump image, you should reboot into a stable kernel.
515
516You can do limited analysis using GDB on the dump file copied out of
517/proc/vmcore. Use the debug vmlinux built with -g and run the following
518command::
519
520   gdb vmlinux <dump-file>
521
522Stack trace for the task on processor 0, register display, and memory
523display work fine.
524
525Note: GDB cannot analyze core files generated in ELF64 format for x86.
526On systems with a maximum of 4GB of memory, you can generate
527ELF32-format headers using the --elf32-core-headers kernel option on the
528dump kernel.
529
530You can also use the Crash utility to analyze dump files in Kdump
531format. Crash is available at the following URL:
532
533   https://github.com/crash-utility/crash
534
535Crash document can be found at:
536   https://crash-utility.github.io/
537
538Trigger Kdump on WARN()
539=======================
540
541The kernel parameter, panic_on_warn, calls panic() in all WARN() paths.  This
542will cause a kdump to occur at the panic() call.  In cases where a user wants
543to specify this during runtime, /proc/sys/kernel/panic_on_warn can be set to 1
544to achieve the same behaviour.
545
546Trigger Kdump on add_taint()
547============================
548
549The kernel parameter panic_on_taint facilitates a conditional call to panic()
550from within add_taint() whenever the value set in this bitmask matches with the
551bit flag being set by add_taint().
552This will cause a kdump to occur at the add_taint()->panic() call.
553
554Contact
555=======
556
557- kexec@lists.infradead.org
558
559GDB macros
560==========
561
562.. include:: gdbmacros.txt
563   :literal:
564