Re: [tip:sched/core 16/27] fs/jbd/commit.c:105:12: sparse: context imbalance in 'inverted_lock' - wrong count at exit

[Josh Triplett <josh@joshtriplett.org>]

On Mon, Sep 30, 2013 at 03:44:34PM +0200, Peter Zijlstra wrote:
> On Wed, Sep 25, 2013 at 02:47:20PM -0700, Josh Triplett wrote:
> > That's expressible in Sparse; look at how spin_trylock and _cond_lock,
> > and write a _cond_unlock.
> 
> Yeah, I know about __cond_lock() its an abomination that should die.

Can't argue with that.

> I
> did take a stab at teach sparse something saner but got stuck.. was
> years ago, can't remember more.

I only see two obvious ways to extend Sparse to remove the need for
__cond_lock, and only one makes sense.

First, you could add an attribute for conditional context changes, which
takes an expression; however, that would require an expression
evaluator, which internally would construct code a lot like __cond_lock,
and it would require some syntax to reference the return value.  That
seems excessively painful, and not significantly better than
__cond_lock.

Second, the real solution: teach Sparse to do whole-program
analysis, similar to GCC LTO.

Alternatively, someone could write a GCC plugin that understands the
context attribute and __context__ statement, and then does whole-program
context analysis using GCC; that seems easiest, relatively speaking.

- Josh Triplett