Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
Date:   Fri, 26 Jun 2020 20:45:22 +0900
From:   Stafford Horne <shorne@gmail.com>
To:     Peter Zijlstra <peterz@infradead.org>
Cc:     LKML <linux-kernel@vger.kernel.org>,
        "Paul E. McKenney" <paulmck@kernel.org>,
        Marco Elver <elver@google.com>,
        "Eric W. Biederman" <ebiederm@xmission.com>,
        Ingo Molnar <mingo@kernel.org>,
        Thomas Gleixner <tglx@linutronix.de>,
        Mike Rapoport <rppt@linux.ibm.com>,
        Sami Tolvanen <samitolvanen@google.com>
Subject: Re: [PATCH] init: Align init_task to avoid conflict with MUTEX_FLAGS
Message-ID: <20200626114522.GI1401039@lianli.shorne-pla.net>
References: <20200625204410.3168791-1-shorne@gmail.com>
 <20200626111208.GD4817@hirez.programming.kicks-ass.net>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20200626111208.GD4817@hirez.programming.kicks-ass.net>
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On Fri, Jun 26, 2020 at 01:12:08PM +0200, Peter Zijlstra wrote:
> On Fri, Jun 26, 2020 at 05:44:09AM +0900, Stafford Horne wrote:
> > When booting on 32-bit machines (seen on OpenRISC) I saw this warning
> > with CONFIG_DEBUG_MUTEXES turned on.
> 
> > I traced this to kernel/locking/mutex.c storing 3 bits of MUTEX_FLAGS in
> > the task_struct pointer (mutex.owner).  There is a comment saying that
> > task_structs are always aligned to L1_CACHE_BYTES.  This is not true for
> > the init_task.
> > 
> > On 64-bit machines this is not a problem because symbol addresses are
> > naturally aligned to 64-bits providing 3 bits for MUTEX_FLAGS.  Howerver,
> > for 32-bit machines the symbol address only has 2 bits available.
> > 
> > Fix this by setting init_task alignment to at least L1_CACHE_BYTES.
> 
> Whoopsie, sorry about that.
> 
> > Cc: Peter Zijlstra <peterz@infradead.org>
> > Signed-off-by: Stafford Horne <shorne@gmail.com>
> > ---
> >  init/init_task.c | 2 ++
> >  1 file changed, 2 insertions(+)
> > 
> > diff --git a/init/init_task.c b/init/init_task.c
> > index 15089d15010a..d2d2af018d0d 100644
> > --- a/init/init_task.c
> > +++ b/init/init_task.c
> > @@ -64,6 +64,8 @@ unsigned long init_shadow_call_stack[SCS_SIZE / sizeof(long)]
> >  struct task_struct init_task
> >  #ifdef CONFIG_ARCH_TASK_STRUCT_ON_STACK
> >  	__init_task_data
> > +#else
> > +	__aligned(L1_CACHE_BYTES)
> >  #endif
> 
> Why make this conditional? task_struct_cachep (in kernel/fork.c) has
> max_t(int, L1_CACHE_BYTES, ARCH_MIN_TASKALIGN) alignment, so this really
> should be aligned on L1_CACHE_BYTES at least.

I think we can make this unconditional.  The only reason I used the condition is
because the only architecture that sets CONFIG_ARCH_TASK_STRUCT_ON_STACK is ia64
which already has 64-bit alignment.

I can change it to (not incorporating max_t(ARCH_MIN_TASKALIGN) as not sure how):

diff --git a/init/init_task.c b/init/init_task.c
index 15089d15010a..ab6173f8e6a8 100644
--- a/init/init_task.c
+++ b/init/init_task.c
@@ -65,6 +65,7 @@ struct task_struct init_task
 #ifdef CONFIG_ARCH_TASK_STRUCT_ON_STACK
        __init_task_data
 #endif
+       __aligned(L1_CACHE_BYTES)
 = {
 #ifdef CONFIG_THREAD_INFO_IN_TASK
        .thread_info    = INIT_THREAD_INFO(init_task),

-Stafford