Re: [PATCH] mce: fix RCU lockdep from mce_log()

[Davidlohr Bueso <dave@gnu.org>]

On Tue, 2011-03-29 at 11:45 +0200, Zdenek Kabelac wrote:
> 2010/11/8 Paul E. McKenney <paulmck@linux.vnet.ibm.com>:
> > On Mon, Nov 08, 2010 at 08:30:19AM -0300, Davidlohr Bueso wrote:
> >> On Sun, 2010-11-07 at 05:39 -0800, Paul E. McKenney wrote:
> >> > On Sat, Nov 06, 2010 at 07:53:50PM +0100, Andi Kleen wrote:
> >> > > On Fri, Nov 05, 2010 at 06:44:59PM -0300, Davidlohr Bueso wrote:
> >> > > > Hi,
> >> > > >
> >> > > > Please review this patch, I am not very familiar with MCE/RCU so I'm not sure that this is the correct fix (otherwise consider it a bug report :)).
> >> > > > This does "fix" the message though and I can use MCE normally.
> >> > >
> >> > > The patch is certainly not correct. The variable needs to be read
> >> > > independently of the mutex.
> >> >
> >> > This code is simply checking the value of the pointer, and therefore
> >> > need not protect any actual dereferences.  So why not replace the
> >> > rcu_dereference_check_mce() with rcu_access_pointer()?  If this is
> >> > OK, please see the patch below.
> >> >
> >> > BTW, assigning the value returned by rcu_access_pointer() into a
> >> > variable often indicates a bug.  ;-)
> >> >
> >> >                                                     Thanx, Paul
> >> >
> >> > Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
> >> >
> >> > diff --git a/arch/x86/kernel/cpu/mcheck/mce.c b/arch/x86/kernel/cpu/mcheck/mce.c
> >> > index 7a35b72..4d29d50 100644
> >> > --- a/arch/x86/kernel/cpu/mcheck/mce.c
> >> > +++ b/arch/x86/kernel/cpu/mcheck/mce.c
> >> > @@ -1625,7 +1625,7 @@ out:
> >> >  static unsigned int mce_poll(struct file *file, poll_table *wait)
> >> >  {
> >> >     poll_wait(file, &mce_wait, wait);
> >> > -   if (rcu_dereference_check_mce(mcelog.next))
> >> > +   if (rcu_access_pointer(mcelog.next))
> >>
> >> this doesn't compile (mcelog.next is an index):
> >>
> >> arch/x86/kernel/cpu/mcheck/mce.c: In function ‘mce_poll’:
> >> arch/x86/kernel/cpu/mcheck/mce.c:1628: error: invalid type argument of
> >> ‘unary *’ (have ‘unsigned int’)
> >> arch/x86/kernel/cpu/mcheck/mce.c:1628: warning: type defaults to ‘int’
> >> in declaration of ‘_________p1’
> >> arch/x86/kernel/cpu/mcheck/mce.c:1628: error: invalid type argument of
> >> ‘unary *’ (have ‘unsigned int’)
> >> arch/x86/kernel/cpu/mcheck/mce.c:1628: warning: type defaults to ‘int’
> >> in declaration of ‘type name’
> >> arch/x86/kernel/cpu/mcheck/mce.c:1628: warning: cast to pointer from
> >> integer of different size
> >> arch/x86/kernel/cpu/mcheck/mce.c:1628: error: invalid type argument of
> >> ‘unary *’ (have ‘unsigned int’)
> >> arch/x86/kernel/cpu/mcheck/mce.c:1628: warning: type defaults to ‘int’
> >> in declaration of ‘type name’
> >> make[4]: *** [arch/x86/kernel/cpu/mcheck/mce.o] Error 1
> >>
> >>
> >> Since the mutex is independent, what about this patch?
> >
> > Looks good to me!
> >
> > Acked-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
> >
> >>  Signed-off-by: Davidlohr Bueso <dave@gnu.org>
> >>
> >> ---
> >>  arch/x86/kernel/cpu/mcheck/mce.c |    2 +-
> >>  1 files changed, 1 insertions(+), 1 deletions(-)
> >>
> >> diff --git a/arch/x86/kernel/cpu/mcheck/mce.c
> >> b/arch/x86/kernel/cpu/mcheck/mce.c
> >> index 7a35b72..cc1c673 100644
> >> --- a/arch/x86/kernel/cpu/mcheck/mce.c
> >> +++ b/arch/x86/kernel/cpu/mcheck/mce.c
> >> @@ -1625,7 +1625,7 @@ out:
> >>  static unsigned int mce_poll(struct file *file, poll_table *wait)
> >>  {
> >>       poll_wait(file, &mce_wait, wait);
> >> -     if (rcu_dereference_check_mce(mcelog.next))
> >> +     if (rcu_dereference_index_check(mcelog.next,
> >> rcu_read_lock_sched_held()))
> >>               return POLLIN | POLLRDNORM;
> >>       if (!mce_apei_read_done && apei_check_mce())
> >>               return POLLIN | POLLRDNORM;
> 
> 
> 
> Any chance to have this ever fixed upstream ?
> (still happens with today's vanialla build)

I'm still quite interested in getting this fixed, I run into it several
times a day.