[PATCH] make lockdep happy with r/o bind mounts

[PATCH] make lockdep happy with r/o bind mounts

[Dave Hansen]

With the r/o bind mount patches, we can have as many
spinlocks nested as there are CPUs on the system.
Lockdep freaks out after 8.

So, create a new lockdep class of locks for the
mnt_writer spinlocks, and initialize each of the
cpu locks to be in a different class.

It should shut up warnings like this, while still
allowing some of the lockdep goodness to remain:

=============================================
[ INFO: possible recursive locking detected ]
2.6.23-rc6 #6
---------------------------------------------
swapper/1 is trying to acquire lock:
 (&writer->lock){--..}, at: [<c016c094>] lock_and_coalesce_cpu_mnt_writer_counts+0x29/0x5c

but task is already holding lock:
 (&writer->lock){--..}, at: [<c016c094>] lock_and_coalesce_cpu_mnt_writer_counts+0x29/0x5c

other info that might help us debug this:
1 lock held by swapper/1:
 #0:  (&writer->lock){--..}, at: [<c016c094>] lock_and_coalesce_cpu_mnt_writer_counts+0x29/0x5c

stack backtrace:
 [<c01034ab>] show_trace_log_lvl+0x19/0x2e
 [<c01034d2>] show_trace+0x12/0x14
 [<c01035be>] dump_stack+0x14/0x16
 [<c0130ab2>] print_deadlock_bug+0xb7/0xc1
 [<c0130afc>] check_deadlock+0x40/0x54
 [<c013100a>] validate_chain+0x26e/0x319
 [<c0132cca>] __lock_acquire+0x6c4/0x731
 [<c01332da>] lock_acquire+0x7a/0x94
 [<c031bb01>] _spin_lock+0x38/0x62
 [<c016c094>] lock_and_coalesce_cpu_mnt_writer_counts+0x29/0x5c
 [<c016c5f8>] mntput_no_expire+0x38/0xad
 [<c015fbed>] path_release_on_umount+0x15/0x18
 [<c016cc76>] sys_umount+0x67/0x6f
 [<c010058a>] name_to_dev_t+0x22a/0x24c
 [<c0427207>] prepare_namespace+0x89/0x187
 [<c0426cdd>] kernel_init+0x9a/0xa3
 [<c010336f>] kernel_thread_helper+0x7/0x10
 =======================

Signed-off-by: Dave Hansen <haveblue@us.ibm.com>
---

 lxc-dave/fs/namespace.c |    2 ++
 1 file changed, 2 insertions(+)

diff -puN fs/namespace.c~make-lockdep-happy-with-ro-bind-mounts fs/namespace.c
--- lxc/fs/namespace.c~make-lockdep-happy-with-ro-bind-mounts	2007-09-24 10:32:23.000000000 -0700
+++ lxc-dave/fs/namespace.c	2007-09-24 10:41:37.000000000 -0700
@@ -116,6 +116,7 @@ struct mnt_writer {
 	 * must be ordered by cpu number.
 	 */
 	spinlock_t lock;
+	struct lock_class_key lock_class; /* compiles out with !lockdep */
 	unsigned long count;
 	struct vfsmount *mnt;
 } ____cacheline_aligned_in_smp;
@@ -127,6 +128,7 @@ static int __init init_mnt_writers(void)
 	for_each_possible_cpu(cpu) {
 		struct mnt_writer *writer = &per_cpu(mnt_writers, cpu);
 		spin_lock_init(&writer->lock);
+		lockdep_set_class(&writer->lock, &writer->lock_class);
 		writer->count = 0;
 	}
 	return 0;
diff -puN include/linux/lockdep.h~make-lockdep-happy-with-ro-bind-mounts include/linux/lockdep.h
_

Re: [PATCH] make lockdep happy with r/o bind mounts

[Peter Zijlstra]

On Mon, 24 Sep 2007 10:46:58 -0700 Dave Hansen <haveblue@us.ibm.com>
wrote:

> 
> With the r/o bind mount patches, we can have as many
> spinlocks nested as there are CPUs on the system.
> Lockdep freaks out after 8.
> 
> So, create a new lockdep class of locks for the
> mnt_writer spinlocks, and initialize each of the
> cpu locks to be in a different class.
> 
> It should shut up warnings like this, while still
> allowing some of the lockdep goodness to remain:
> 
> =============================================
> [ INFO: possible recursive locking detected ]
> 2.6.23-rc6 #6
> ---------------------------------------------
> swapper/1 is trying to acquire lock:
>  (&writer->lock){--..}, at: [<c016c094>] lock_and_coalesce_cpu_mnt_writer_counts+0x29/0x5c
> 
> but task is already holding lock:
>  (&writer->lock){--..}, at: [<c016c094>] lock_and_coalesce_cpu_mnt_writer_counts+0x29/0x5c
> 
> other info that might help us debug this:
> 1 lock held by swapper/1:
>  #0:  (&writer->lock){--..}, at: [<c016c094>] lock_and_coalesce_cpu_mnt_writer_counts+0x29/0x5c
> 
> stack backtrace:
>  [<c01034ab>] show_trace_log_lvl+0x19/0x2e
>  [<c01034d2>] show_trace+0x12/0x14
>  [<c01035be>] dump_stack+0x14/0x16
>  [<c0130ab2>] print_deadlock_bug+0xb7/0xc1
>  [<c0130afc>] check_deadlock+0x40/0x54
>  [<c013100a>] validate_chain+0x26e/0x319
>  [<c0132cca>] __lock_acquire+0x6c4/0x731
>  [<c01332da>] lock_acquire+0x7a/0x94
>  [<c031bb01>] _spin_lock+0x38/0x62
>  [<c016c094>] lock_and_coalesce_cpu_mnt_writer_counts+0x29/0x5c
>  [<c016c5f8>] mntput_no_expire+0x38/0xad
>  [<c015fbed>] path_release_on_umount+0x15/0x18
>  [<c016cc76>] sys_umount+0x67/0x6f
>  [<c010058a>] name_to_dev_t+0x22a/0x24c
>  [<c0427207>] prepare_namespace+0x89/0x187
>  [<c0426cdd>] kernel_init+0x9a/0xa3
>  [<c010336f>] kernel_thread_helper+0x7/0x10
>  =======================
> 
> Signed-off-by: Dave Hansen <haveblue@us.ibm.com>

Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl>

> ---
> 
>  lxc-dave/fs/namespace.c |    2 ++
>  1 file changed, 2 insertions(+)
> 
> diff -puN fs/namespace.c~make-lockdep-happy-with-ro-bind-mounts fs/namespace.c
> --- lxc/fs/namespace.c~make-lockdep-happy-with-ro-bind-mounts	2007-09-24 10:32:23.000000000 -0700
> +++ lxc-dave/fs/namespace.c	2007-09-24 10:41:37.000000000 -0700
> @@ -116,6 +116,7 @@ struct mnt_writer {
>  	 * must be ordered by cpu number.
>  	 */
>  	spinlock_t lock;
> +	struct lock_class_key lock_class; /* compiles out with !lockdep */
>  	unsigned long count;
>  	struct vfsmount *mnt;
>  } ____cacheline_aligned_in_smp;
> @@ -127,6 +128,7 @@ static int __init init_mnt_writers(void)
>  	for_each_possible_cpu(cpu) {
>  		struct mnt_writer *writer = &per_cpu(mnt_writers, cpu);
>  		spin_lock_init(&writer->lock);
> +		lockdep_set_class(&writer->lock, &writer->lock_class);
>  		writer->count = 0;
>  	}
>  	return 0;
> diff -puN include/linux/lockdep.h~make-lockdep-happy-with-ro-bind-mounts include/linux/lockdep.h
> _

Re: [PATCH] make lockdep happy with r/o bind mounts

[Arjan van de Ven]

On Mon, 24 Sep 2007 10:46:58 -0700
Dave Hansen <haveblue@us.ibm.com> wrote:

> 
> With the r/o bind mount patches, we can have as many
> spinlocks nested as there are CPUs on the system.
> Lockdep freaks out after 8.

The patch implements the right solution for this "invent your own lock"
situation... 

Acked-by: Arjan van de Ven <arjan@linux.intel.com>