[Ksummit-2012-discuss] [ATTEND] kernel core dump and "dying breath"
tony.luck at intel.com
Fri Jun 22 20:28:28 UTC 2012
> I have to admit I am curios on both fronts. For pstore, it got me
> wondering about what kind of atomic interface exists, and if this is
> used in any distros as of yet. With an atomic interface to pstore
> that doesn't touch pieces of the FS handlers that require locks, I
> could see this being really useful. The documentation is a bit scant
> so I'll probably have to play with this.
pstore has a number of platform specific back-ends to stash data:
1) ERST - this is part of ACPI spec - an interface to some platform level
non-volatile storage. Tends to only be implemented on server systems,
and the amount of storage available is quite limited (tens of kilobytes).
But this is usually enough so store enough lines from the kernel log from
a handful of OOPS and a panic.
2) EFI variables - available on more platforms every day as EFI starts to
take over the world (once Windows 8 is launched EFI will be the default
on just about everything since it is a Win8 sticker requirement). Even
more limited in non-volatile space - just one panic log can be stored.
3) RAMoops - very recent addition to mainline. Needs some area of memory
that the BIOS will not scribble on. Can provide much more space, and because
it is RAM can be used for more exciting flight recorder modes (e.g. there are
patches to hook into ftrace to keep a running stack dump present ).
More information about the Ksummit-2012-discuss