x86/mce fix (ready for 3.6 merge window)

The following changes since commit 6887a4131da3adaab011613776d865f4bcfb5678:

Linux 3.5-rc5 (2012-06-30 16:08:57 -0700)

are available in the git repository at:

git://git.kernel.org/pub/scm/linux/kernel/git/ras/ras.git tags/please-pull-mce-ripvfix

for you to fetch changes up to b99c2fc9366d4e32b8d087cd28f6dbf5f7932dae:

x86/mce: Need to let kill_proc() send signal to doomed process (2012-07-10 10:18:29 -0700)

----------------------------------------------------------------
Fix application level machine check recovery

----------------------------------------------------------------
Tony Luck (1):
x86/mce: Need to let kill_proc() send signal to doomed process

 arch/x86/kernel/cpu/mcheck/mce.c |  6 ++++--
 include/linux/mm.h               |  1 +
 mm/memory-failure.c              | 10 ++++++----
 3 files changed, 11 insertions(+), 6 deletions(-)

A couple of commit log details:

- If it's for v3.6 then the Cc: stable backport is not justified. Either it's for tip:x86/urgent and then we'll merge it straight away, or for tip:x86/mce for v3.6 and then there's no Cc: stable tag.

- This reference to a commit is a bit unusual:

In commit dad1743e5993f19b3d7e7bd0fb35dc45b5326626
x86/mce: Only restart instruction after machine check recovery if it is safe

the canonical format is something like:

In commit dad1743e5993f1 ("x86/mce: Only restart instruction after machine check recovery if it is safe") ...

- We tend to use such an ordering of tags:

I.e. Tested-by and Reported-by tags first (if any), then author SOB, then SOB chain (if any), then Reviewed-by and Acked-by, then stable tags, then Cc:s.

(I fix this up silently for email space patches, for Git pulls
I cannot do that.)

- The title is suboptimal:

x86/mce: Need to let kill_proc() send signal to doomed process

In titles we prefer proper sentences, starting with a verb, so something like this would do:

x86/mce: Allow kill_proc() to send signals to doomed process

Or rather, use a good title that talks about what the change is really about:

x86/mce: Fix siginfo_t->si_addr value for non-recoverable memory faults

- (Spelling nitpick: it's not "et. al." but "et al.".)

- Style nit, this:

       if (mi->restartable == 0)

is better written as:

       if (!mi->restartable)

because mi->restartable's role here is not really an integer value, but a boolean in essence.

- The 'doit' flag was significantly misnamed when kill_procs() was written and now it spreads further, it's a totally opaque name that tells nothing about the role of the flag.

How about 'force'?

Thanks,

Ingo

This could be part of checkpatch - whenever a stable tag is added to a patch commit msg, it should at least warn the patch author to check with <Documentation/stable_kernel_rules.txt> first.

Commit referencing in commit messages doesn't come up for the first time so can we get this as a rule into checkpatch so that we can have unified commit reference format?

The regex would be probably hairy and generate a couple of false positives but sure it will help in a lot of other situations.

Also, how many chars of the commit id we keep? The first 12, 14, 15? I'm thinking of commit id uniqueness sometime far in the future.

patch tags order could be checked for in checkpatch too?

[ … ]

Yes, we talked about this but having a bool as a u8 there would add padding to the struct so it's the same thing, space-wise. It could be converted to a bitfield if more flags are added/needed.

Even better, make it even more descriptive: 'force_kill' or 'do_kill' or 'really_kill' - this way one knows exactly what one is looking at.

Thanks.

Regards/Gruss,
    Boris.

The commit that this fixes was included in 3.4 - so I was thinking that this should be backported to stable 3.4 ... but 3.4 isn't a long term release, so it will be abandoned when 3.5 is released.

Since we are at -rc6 and this isn't a regression from the current merge, I figured that it wasn't a candidate for immediate merge ... hence the 3.6 request. So it would appear that there isn't a way that this can make it back to 3.4-stable.

I'll fix up all the other stuff and send a new pull (still for 3.6, and still with
a Cc: stable for backport to 3.5)

-Tony

The following changes since commit 6887a4131da3adaab011613776d865f4bcfb5678:

Linux 3.5-rc5 (2012-06-30 16:08:57 -0700)

are available in the git repository at:

git://git.kernel.org/pub/scm/linux/kernel/git/ras/ras.git mce-ripvfix

for you to fetch changes up to 6751ed65dc6642af64f7b8a440a75563c8aab7ae:

x86/mce: Fix siginfo_t->si_addr value for non-recoverable memory faults (2012-07-11 10:20:47 -0700)

----------------------------------------------------------------
Fix application level machine check recovery

----------------------------------------------------------------
Tony Luck (1):

      x86/mce: Fix siginfo_t->si_addr value for non-recoverable memory faults

 arch/x86/kernel/cpu/mcheck/mce.c |  6 ++++--
 include/linux/mm.h               |  1 +
 mm/memory-failure.c              | 14 ++++++++------
 3 files changed, 13 insertions(+), 8 deletions(-)

Pulled, thanks Tony!

Ingo