|| ||"Luck, Tony" <firstname.lastname@example.org> |
|| ||"Ingo Molnar" <email@example.com>, "Borislav Petkov" <firstname.lastname@example.org> |
|| ||[RFC] reworked machine check recovery patches |
|| ||Thu, 09 Jun 2011 14:25:38 -0700|
|| ||email@example.com, "Huang, Ying" <firstname.lastname@example.org>,
"Hidetoshi Seto" <email@example.com>,
"Avi Kivity" <firstname.lastname@example.org>|
|| ||Article, Thread
This is the in-kernel recovery path for memory errors. Still
no change to the eventual reporting mechanism (mcelog vs. perf).
Lots of changes based on feedback from the previous set of
discussions. Big stuff:
0) Scaled back on ambitions - just worry about recovery from
memory errors while executing in user mode. Adding recovery
for some kernel cases is deferred to some future date.
1) Took Seto-san's "mce_gather_info()" patch in place of the
one to just move mce_get_rip() around. Added an expanded
comment to explain that we need to collect all this information
to make decisions about severity for errors we find when scanning
the machine check banks. The rest of Seto-san's patches to
make the severity table easier to read look interesting.
2) Ingo pointed me the the user-return-notifier code as an alternative
to TIF_MCE_NOTIFY. Part 7 uses return notifiers to cover the
existing use of this TIF bit (Avi: Thanks for the review)
3) Take the newly freed TIF_MCE_NOTIFY bit and use it to implement
an extension in part 8 to make a "per-task" user notifier.
3) Re-implement the recovery path using the task_return_notifier ...
this gets rid of the "mce_error_pfn" field that was used in the
previous implementation that was so obnoxious.
N.B. both the existing parts and my extensions to the user return
notifiers are not NMI safe because they use ordinary Linux lists.
If this seems to be a worthwhile direction, this deficiency can
be fixed using Ying's NMI safe single linked lists. Ditto the
hacky "allocator" of task_notify structures in mce_action_required().
Just a proof of concept ... should be replaced with a more
generic NMI safe allocator.
arch/x86/Kconfig | 1 +
arch/x86/include/asm/thread_info.h | 8 +-
arch/x86/kernel/cpu/mcheck/mce-severity.c | 37 +++-
arch/x86/kernel/cpu/mcheck/mce.c | 339 ++++++++++++++++++++++-------
arch/x86/kernel/signal.c | 7 +-
include/linux/sched.h | 3 +
include/linux/user-return-notifier.h | 14 ++
kernel/fork.c | 1 +
kernel/user-return-notifier.c | 36 +++
mm/memory-failure.c | 28 ++-
10 files changed, 369 insertions(+), 105 deletions(-)
Andi Kleen (3):
MCE: Move ADDR/MISC reading code into common function
MCE: Mask out address mask bits below address granuality
HWPOISON: Handle hwpoison in current process
Hidetoshi Seto (1):
MCE: introduce mce_gather_info()
Tony Luck (6):
MCE: fixes for mce severity table
MCE: save most severe error information
MCE: replace mce.c use of TIF_MCE_NOTIFY with user_return_notifier
NOTIFIER: Take over TIF_MCE_NOTIFY and implement task return notifier
MCE: run through processors with more severe problems first
MCE: Add Action-Required support
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to email@example.com
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/