lock context warnings

lock context warnings

[Christoph Hellwig]

Since a few month sparse keeps spewing tons of wrong lock context
warnings like:

/home/hch/work/linux-2.6/fs/xfs/quota/xfs_dquot.c:1341:24: warning:
context problem in 'xfs_qm_dqflush_done': 'xfs_trans_ail_delete'
expected different context
/home/hch/work/linux-2.6/fs/xfs/quota/xfs_dquot.c:1341:24:    context
'xa_lock': wanted >= 1, got 0
/home/hch/work/linux-2.6/fs/xfs/quota/xfs_dquot_item.c:565:22: warning:
context problem in 'xfs_qm_qoffend_logitem_committed':
'xfs_trans_ail_delete' expected different context
/home/hch/work/linux-2.6/fs/xfs/quota/xfs_dquot_item.c:565:22:
context 'xa_lock': wanted >= 1, got 0

which mean lock context annotations just don't seem to work, e.g. the
fist one is:

		/* xfs_trans_ail_delete() drops the AIL lock. */
		spin_lock(&ailp->xa_lock);
		if (qip->qli_item.li_lsn == qip->qli_flush_lsn)	
			xfs_trans_ail_delete(ailp, (xfs_log_item_t*)qip);
		else
			spin_unlock(&ailp->xa_lock);


where xfs_trans_ail_delete prototyped as:

void                    xfs_trans_ail_delete(struct xfs_ail *ailp,
                                        struct xfs_log_item *lip)
					__releases(ailp->xa_lock);


These messages are highly annoying in a kernel build, so if you can't
fix them easily, please at least remove the lock context stuff from
the default check flags.

Re: lock context warnings

[Alexey Zaytsev]

On Wed, Jan 7, 2009 at 11:45, Christoph Hellwig <hch@lst.de> wrote:
> Since a few month sparse keeps spewing tons of wrong lock context
> warnings like:

Yes, the warnings are wrong, and the patches that were supposeed to
fix them breoke other things, so they were never merged.
Christopher reverted the patches that introduced the new warnings
in his tree: git://git.kernel.org/pub/scm/devel/sparse/chrisl/sparse.git
until someone fixes or rewrites the context checking.

It looks like Christopher is serious about maintaining this tree, so
people should probably switch to using it.

Re: lock context warnings

[Christoph Hellwig]

On Wed, Jan 07, 2009 at 12:20:18PM +0300, Alexey Zaytsev wrote:
> On Wed, Jan 7, 2009 at 11:45, Christoph Hellwig <hch@lst.de> wrote:
> > Since a few month sparse keeps spewing tons of wrong lock context
> > warnings like:
> 
> Yes, the warnings are wrong, and the patches that were supposeed to
> fix them breoke other things, so they were never merged.
> Christopher reverted the patches that introduced the new warnings
> in his tree: git://git.kernel.org/pub/scm/devel/sparse/chrisl/sparse.git
> until someone fixes or rewrites the context checking.
> 
> It looks like Christopher is serious about maintaining this tree, so
> people should probably switch to using it.

That tree looks much better, only a few lock context warnings left which
look legit from a first scan.  I'll keep using that tree for now.