2                   Hypervisor-Assisted Dump
3                   ------------------------
4                       November 2007
6The goal of hypervisor-assisted dump is to enable the dump of
7a crashed system, and to do so from a fully-reset system, and
8to minimize the total elapsed time until the system is back
9in production use.
11As compared to kdump or other strategies, hypervisor-assisted
12dump offers several strong, practical advantages:
14-- Unlike kdump, the system has been reset, and loaded
15   with a fresh copy of the kernel. In particular,
16   PCI and I/O devices have been reinitialized and are
17   in a clean, consistent state.
18-- As the dump is performed, the dumped memory becomes
19   immediately available to the system for normal use.
20-- After the dump is completed, no further reboots are
21   required; the system will be fully usable, and running
22   in its normal, production mode on its normal kernel.
24The above can only be accomplished by coordination with,
25and assistance from the hypervisor. The procedure is
26as follows:
28-- When a system crashes, the hypervisor will save
29   the low 256MB of RAM to a previously registered
30   save region. It will also save system state, system
31   registers, and hardware PTE's.
33-- After the low 256MB area has been saved, the
34   hypervisor will reset PCI and other hardware state.
35   It will *not* clear RAM. It will then launch the
36   bootloader, as normal.
38-- The freshly booted kernel will notice that there
39   is a new node (ibm,dump-kernel) in the device tree,
40   indicating that there is crash data available from
41   a previous boot. It will boot into only 256MB of RAM,
42   reserving the rest of system memory.
44-- Userspace tools will parse /sys/kernel/release_region
45   and read /proc/vmcore to obtain the contents of memory,
46   which holds the previous crashed kernel. The userspace
47   tools may copy this info to disk, or network, nas, san,
48   iscsi, etc. as desired.
50   For Example: the values in /sys/kernel/release-region
51   would look something like this (address-range pairs).
52   CPU:0x177fee000-0x10000: HPTE:0x177ffe020-0x1000: /
53   DUMP:0x177fff020-0x10000000, 0x10000000-0x16F1D370A
55-- As the userspace tools complete saving a portion of
56   dump, they echo an offset and size to
57   /sys/kernel/release_region to release the reserved
58   memory back to general use.
60   An example of this is:
61     "echo 0x40000000 0x10000000 > /sys/kernel/release_region"
62   which will release 256MB at the 1GB boundary.
64Please note that the hypervisor-assisted dump feature
65is only available on Power6-based systems with recent
66firmware versions.
68Implementation details:
71During boot, a check is made to see if firmware supports
72this feature on this particular machine. If it does, then
73we check to see if a active dump is waiting for us. If yes
74then everything but 256 MB of RAM is reserved during early
75boot. This area is released once we collect a dump from user
76land scripts that are run. If there is dump data, then
77the /sys/kernel/release_region file is created, and
78the reserved memory is held.
80If there is no waiting dump data, then only the highest
81256MB of the ram is reserved as a scratch area. This area
82is *not* released: this region will be kept permanently
83reserved, so that it can act as a receptacle for a copy
84of the low 256MB in the case a crash does occur. See,
85however, "open issues" below, as to whether
86such a reserved region is really needed.
88Currently the dump will be copied from /proc/vmcore to a
89a new file upon user intervention. The starting address
90to be read and the range for each data point in provided
91in /sys/kernel/release_region.
93The tools to examine the dump will be same as the ones
94used for kdump.
96General notes:
98Security: please note that there are potential security issues
99with any sort of dump mechanism. In particular, plaintext
100(unencrypted) data, and possibly passwords, may be present in
101the dump data. Userspace tools must take adequate precautions to
102preserve security.
104Open issues/ToDo:
106 o The various code paths that tell the hypervisor that a crash
107   occurred, vs. it simply being a normal reboot, should be
108   reviewed, and possibly clarified/fixed.
110 o Instead of using /sys/kernel, should there be a /sys/dump
111   instead? There is a dump_subsys being created by the s390 code,
112   perhaps the pseries code should use a similar layout as well.
114 o Is reserving a 256MB region really required? The goal of
115   reserving a 256MB scratch area is to make sure that no
116   important crash data is clobbered when the hypervisor
117   save low mem to the scratch area. But, if one could assure
118   that nothing important is located in some 256MB area, then
119   it would not need to be reserved. Something that can be
120   improved in subsequent versions.
122 o Still working the kdump team to integrate this with kdump,
123   some work remains but this would not affect the current
124   patches.
126 o Still need to write a shell script, to copy the dump away.
127   Currently I am parsing it manually.

Archive Download this file