Re: [PATCH 03/10] rwsem: lighter active count checks when waking up readers

[David Howells <dhowells@redhat.com>]

Michel Lespinasse <walken@google.com> wrote:

> ... When there are waiter threads on a rwsem and the spinlock is held, other
> threads can only increment the active count by trying to grab the rwsem in
> up_xxxx().

That's not true.  A thread attempting to get an rwsem by issuing a down_read()
or down_write() will also unconditionally increment the active count before it
considers calling out to the slow path.

Maybe what you mean is that other threads wanting to do a wake up can only
increase the active count for the processes being woken up whilst holding the
rwsem's spinlock.

> +	/* If we come here from up_xxxx(), another thread might have reached
> +	 * rwsem_down_failed_common() before we acquired the spinlock and
> +	 * woken up an active locker.

Do you mean a waiter rather than an active locker?  If a process is still
registering activity on the rwsem, then it can't be woken yet.
Michel Lespinasse <walken@google.com> wrote:

> +	 * Note that we do not need to update the rwsem count: any writer
> +	 * trying to acquire rwsem will run rwsem_down_write_failed() due
> +	 * to the waiting threads, and block trying to acquire the spinlock.

That comma shouldn't be there.

>  	/* Grant an infinite number of read locks to the readers at the front
>  	 * of the queue.  Note we increment the 'active part' of the count by

I wonder if I should've called it the 'activity part' of the count rather than
the 'active part'.

Apart from that, the patch looks fine.  That's all comment/description fixes.

David