Re: A question about sparse: how to use __acquires() and __releases() correctly ?

[Johannes Berg <johannes@sipsolutions.net>]

On Thu, 4 Dec 2008 13:12:58 +0100, "Bart Van Assche"
<bart.vanassche@gmail.com> wrote:
> [ping]
> 
> Is there anyone who can help me with the question below ?


>> void dev_seq_stop(struct seq_file *seq, void *v)
>>        __releases(dev_base_lock)
>> {
>>        read_unlock(&dev_base_lock);
>> }
>>
>> The command "make C=2 M=net/core" produces the following output for
>> the above function (using a sparse binary built from the sparse git
>> repository, last updated on August 26, 2008):
>>
>> net/core/dev.c:2579:2: warning: context problem in 'dev_seq_stop':
>> '_read_unlock' expected different context
>> net/core/dev.c:2579:2:    context 'lock': wanted >= 1, got 0

I don't think sparse can properly handle this yet, at least not in a way
you'd expect it to. I've extended sparse to handle it, but the current git
tree has only a partial set of my changes applied, and the remaining ones
have been contested. (I still think my initial changes should be reverted
in the meantime)

>> My questions are as follows:
>> * Which argument type should be passed to __releases() -- a pointer to
>> a lock structure or the lock strucure itself ? In the header file
>> include/linux/spinlock_api_smp.h a pointer is passed to __acquires()
>> and __releases(), while other code (like the above) passes the lock
>> structure itself to the __acquires() and __releases() annotations.

sparse prett much ignores the first argument anyway, this isn't defined
yet.

>> * If the __releases() annotation is used correctly in net/core/dev.c,
>> why does sparse complain about a context problem ?

Maybe it is? I don't know off-hand.

johannes