[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <20170907070404.5j2wrblttsxd2qyb@hirez.programming.kicks-ass.net>
Date: Thu, 7 Sep 2017 09:04:04 +0200
From: Peter Zijlstra <peterz@...radead.org>
To: Josh Poimboeuf <jpoimboe@...hat.com>
Cc: Miguel Bernal Marin <miguel.bernal.marin@...ux.intel.com>,
Ingo Molnar <mingo@...hat.com>,
Thomas Gleixner <tglx@...utronix.de>,
"H . Peter Anvin" <hpa@...or.com>, linux-kernel@...r.kernel.org,
x86@...nel.org
Subject: Re: [PATCH 3/3] locking/rwsem/x86: Add stack frame dependency for
__downgrade_write()
On Wed, Sep 06, 2017 at 04:33:02PM -0500, Josh Poimboeuf wrote:
> On Wed, Sep 06, 2017 at 12:26:13AM -0500, Miguel Bernal Marin wrote:
> > kernel/locking/rwsem.o: warning: objtool: downgrade_write()+0x22: call without frame pointer save/setup
> >
> > The warning means gcc 7.2.0 placed the __downgrade_write() inline asm (and
> > its call instruction) before the frame pointer setup in downgrade_write(),
> > which breaks frame pointer convention and can result in incorrect
> > stack traces.
> >
> > Force a stack frame to be created before the call instruction by listing
> > the stack pointer as an output operand in the inline asm statement.
> >
> > Signed-off-by: Miguel Bernal Marin <miguel.bernal.marin@...ux.intel.com>
> > ---
> > arch/x86/include/asm/rwsem.h | 6 ++++--
> > 1 file changed, 4 insertions(+), 2 deletions(-)
> >
> > diff --git a/arch/x86/include/asm/rwsem.h b/arch/x86/include/asm/rwsem.h
> > index d26b6916b935..a749dc6a3103 100644
> > --- a/arch/x86/include/asm/rwsem.h
> > +++ b/arch/x86/include/asm/rwsem.h
> > @@ -205,8 +205,10 @@ static inline void __up_write(struct rw_semaphore *sem)
> > */
> > static inline void __downgrade_write(struct rw_semaphore *sem)
> > {
> > + register void *__sp asm(_ASM_SP);
> > +
> > asm volatile("# beginning __downgrade_write\n\t"
> > - LOCK_PREFIX _ASM_ADD "%2,(%1)\n\t"
> > + LOCK_PREFIX _ASM_ADD "%2,(%2)\n\t"
>
> The '%2' should be changed to '%3'
>
> LOCK_PREFIX _ASM_ADD "%3,(%2)\n\t"
>
> because both inputs' indices are shifted by the new output constraint.
Even better would be to used named operands.
Powered by blists - more mailing lists