Re: [PATCH 2/2] spinlock_debug: Print kallsyms name for lock

From: Stephen Boyd <sboyd@codeaurora.org>
To: Andrew Morton <akpm@linux-foundation.org>
Cc: linux-kernel@vger.kernel.org
Subject: Re: [PATCH 2/2] spinlock_debug: Print kallsyms name for lock
Date: Thu, 10 May 2012 11:53:53 -0700	[thread overview]
Message-ID: <4FAC0EC1.7050803@codeaurora.org> (raw)
In-Reply-To: <20120424145402.4f9be035.akpm@linux-foundation.org>

On 04/24/12 14:54, Andrew Morton wrote:
> On Mon, 23 Apr 2012 20:17:03 -0700
> Stephen Boyd <sboyd@codeaurora.org> wrote:
>
>> On 04/23/12 14:54, Andrew Morton wrote:
>>>> --- a/lib/spinlock_debug.c
>>>> +++ b/lib/spinlock_debug.c
>>>> @@ -58,7 +58,7 @@ static void spin_dump(raw_spinlock_t *lock, const char *msg)
>>>>  	printk(KERN_EMERG "BUG: spinlock %s on CPU#%d, %s/%d\n",
>>>>  		msg, raw_smp_processor_id(),
>>>>  		current->comm, task_pid_nr(current));
>>>> -	printk(KERN_EMERG " lock: %p, .magic: %08x, .owner: %s/%d, "
>>>> +	printk(KERN_EMERG " lock: %ps, .magic: %08x, .owner: %s/%d, "
>>>>  			".owner_cpu: %d\n",
>>>>  		lock, lock->magic,
>>>>  		owner ? owner->comm : "<none>",
>>> Maybe.  It will only do useful things for statically-allocated locks
>>> which are rare and which we are unlikely to screw up as easily as locks
>>> which lie in dynamically allocated memory.
>> Agreed. It catches the really stupid stuff and that's about it. I was
>> thinking we could get more information about dynamic allocated locks by
>> adding some code to slab to find the slab that a pointer is allocated
>> in. Does that sound possible?
> Well lockdep knows lots of things about the lock, including numerous
> stack backtraces which can be used to identify the lock.  Making 
> spinlock_debug use lockdep infrastructure seems a good fit.
>

I was thinking about this more. In the case of spinlock bad magic I want
to find out who freed this region of memory last because that code most
likely stomped all over my lock and corrupted the magic. With lockdep I
can't find that. I can only find out that the allocator of a lock did a
lock/unlock after a kfree() and I believe lockdep already checks to make
sure live locks are not being freed.

Enabling slub debugging might help, but I would have to get lucky and
allocate the lock after the previous user corrupted it. So I really want
a way to query slab/slub about who last allocated and free this memory
so I can find them when I detect magic corruption.

-- 
Sent by an employee of the Qualcomm Innovation Center, Inc.
The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum.