lkml.org 
[lkml]   [2017]   [Apr]   [12]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [RFC PATCH] x86, mce: change the mce notifier to 'blocking' from 'atomic'
On 04/12, Borislav Petkov wrote:
> On Tue, Apr 11, 2017 at 04:44:57PM -0600, Vishal Verma wrote:
> > The NFIT MCE handler callback (for handling media errors on NVDIMMs)
> > takes a mutex to add the location of a memory error to a list. But since
> > the notifier call chain for machine checks (x86_mce_decoder_chain) is
> > atomic, we get a lockdep splat like:
> >
> > BUG: sleeping function called from invalid context at kernel/locking/mutex.c:620
> > in_atomic(): 1, irqs_disabled(): 0, pid: 4, name: kworker/0:0
> > [..]
> > Call Trace:
> > dump_stack+0x86/0xc3
> > ___might_sleep+0x178/0x240
> > __might_sleep+0x4a/0x80
> > mutex_lock_nested+0x43/0x3f0
> > ? __lock_acquire+0xcbc/0x1290
> > nfit_handle_mce+0x33/0x180 [nfit]
> > notifier_call_chain+0x4a/0x70
> > atomic_notifier_call_chain+0x6e/0x110
> > ? atomic_notifier_call_chain+0x5/0x110
> > mce_gen_pool_process+0x41/0x70
> >
> > Commit 648ed94038c030245a06e4be59744fd5cdc18c40
> > x86/mce: Provide a lockless memory pool to save error records
> > Changes the mce notifier callbacks to be run in a process context, and
> > this can allow us to use the 'blocking' type notifier, where we can take
> > mutexes etc. in the call chain functions.
> >
> > Reported-by: Ross Zwisler <ross.zwisler@linux.intel.com>
> > Cc: Borislav Petkov <bp@suse.de>
> > Cc: Tony Luck <tony.luck@intel.com>
> > Cc: Dan Williams <dan.j.williams@intel.com>
> > Signed-off-by: Vishal Verma <vishal.l.verma@intel.com>
> > ---
> > arch/x86/kernel/cpu/mcheck/mce-genpool.c | 2 +-
> > arch/x86/kernel/cpu/mcheck/mce-internal.h | 2 +-
> > arch/x86/kernel/cpu/mcheck/mce.c | 8 ++++----
> > 3 files changed, 6 insertions(+), 6 deletions(-)
> >
> > While this patch almost solves the problem, I think it is not quite right.
> > The x86_mce_decoder_chain is also called from print_mce for fatal machine
> > checks, and that is, afaict, still from an atomic context. One thing Tony
> > suggested was splitting the notifier chain into two distinct chains, one
> > for regular logging and recoverable actions that allows blocking, the
> > other from the panic path.
>
> Well, if Mohammad won't come to the mountain...
>
> So the NFIT handler has:
>
> /* We only care about memory errors */
> if (!(mce->status & MCACOD))
> return NOTIFY_DONE;
>
> what severity are we talking here? Errors which can be reported on the
> panic path, i.e., in atomic context or only AO/AR ones which don't raise
> an #MC exception?

I don't think we can do anything about the panic path errors. The NFIT
handler takes the recoverable machine checks, and essentially, adds the
location to a list.

>
> --
> Regards/Gruss,
> Boris.
>
> SUSE Linux GmbH, GF: Felix Imendörffer, Jane Smithard, Graham Norton, HRB 21284 (AG Nürnberg)
> --

\
 
 \ /
  Last update: 2017-04-12 22:00    [W:0.067 / U:2.908 seconds]
©2003-2018 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site