Re: [patch 01/12] xfs: split tail_lsn assignments from log space wakeups

[Christoph Hellwig <hch@xxxxxxxxxxxxx>]

> >     if (!list_empty(&tmp))
> >             xfs_ail_splice(ailp, cur, &tmp, lsn);
> > +   spin_unlock(&ailp->xa_lock);
> 
> Right.  I am uncomfortable with the idea of dropping the ail lock here
> and then retaking it below in xlog_assign_tail_lsn.  Your suggestion
> that a variant of xlog_assign_tail_lsn which expects the lock to be held
> seems reasonable.

There is no risk in dropping it in terms of correctness, the only
downside is doupling the amount of lock roundtrips.  The reason why
I didn't do the version that is called with the lock held is that it
would be fairly intrusive and ugly.