Re: [PATCH 1/1 linux-next] fs/ufs/super.c: remove CONFIG_SMP/CONFIG_PREEMPT testing

[Andrew Morton]

On Sun,  1 Feb 2015 12:36:57 +0100 Fabian Frederick <fabf@xxxxxxxxx> wrote:

> Let locking subsystem decide on mutex management.
> 
> ...
>
> --- a/fs/ufs/super.c
> +++ b/fs/ufs/super.c
> @@ -95,22 +95,18 @@
>  
>  void lock_ufs(struct super_block *sb)
>  {
> -#if defined(CONFIG_SMP) || defined (CONFIG_PREEMPT)
>  	struct ufs_sb_info *sbi = UFS_SB(sb);
>  
>  	mutex_lock(&sbi->mutex);
>  	sbi->mutex_owner = current;
> -#endif
>  }

Good heavens.

This patch is a bugfix.  lock_ufs() is assuming that on non-preempt
uniprocessor, the calling code will run atomically up to the matching
unlock_ufs().

But that isn't true.  The very first site I looked at (ufs_frag_map)
does sb_bread() under lock_ufs(.  And sb_bread() will call schedule(),
very commonly.

The ->mutex_owner stuff is a bit hacky but should work OK.

Care to resend the patch with a more alarming changelog?
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/