Re: [RFC] seqlock,lockdep: Add lock primitives to read_seqbegin().

[Peter Zijlstra <peterz@infradead.org>]

On Tue, 2011-03-29 at 13:30 +0900, Tetsuo Handa wrote:

> static int locktest_open1(struct inode *inode, struct file *file)
> {
> 	write_seqlock(&seqlock1);
> 	msleep(1000); /* Open /proc/locktest2 while sleeping here. */
> 	br_read_lock(brlock1);
> 	br_read_unlock(brlock1);
> 	write_sequnlock(&seqlock1);
> 	return -EINVAL;
> }
> 
> static int locktest_open2(struct inode *inode, struct file *file)
> {
> 	br_write_lock(brlock1);
> 	read_seqbegin2(&seqlock1);
> 	br_write_unlock(brlock1);
> 	return -EINVAL;
> }
> 
> static int locktest_open3(struct inode *inode, struct file *file)
> {
> 	static DEFINE_MUTEX(mutex1);
> 	mutex_lock(&mutex1);
> 	locktest_open1(inode, file);
> 	mutex_unlock(&mutex1);
> 	return -EINVAL;
> }


That's quite horrid as far as test-cases go, why bother with the
userspace part at all?

In order to hit your inversion you need to do something like:

cat /proc/locktest1 & cat /proc/locktest2

if you do them serialized you'll never hit that inversion.


That said, there are some out-standing issues with rw_locks and lockdep,
Gautham and I worked on that for a while but we never persevered and
finished it..

  http://lkml.org/lkml/2009/5/11/203

And I think you're hitting that case.