Re: [PATCH v3 1/2] locking/qrwlock: Better optimization for interrupt context readers

[Will Deacon <will.deacon@arm.com>]

Hi Waiman,

On Mon, Jun 15, 2015 at 11:24:02PM +0100, Waiman Long wrote:
> The qrwlock is fair in the process context, but becoming unfair when
> in the interrupt context to support use cases like the tasklist_lock.
> 
> The current code isn't that well-documented on what happens when
> in the interrupt context. The rspin_until_writer_unlock() will only
> spin if the writer has gotten the lock. If the writer is still in the
> waiting state, the increment in the reader count will cause the writer
> to remain in the waiting state and the new interrupt context reader
> will get the lock and return immediately. The current code, however,
> do an additional read of the lock value which is not necessary as the
> information have already been there in the fast path. This may sometime
> cause an additional cacheline load when the lock is highly contended.
> 
> This patch passes the lock value information gotten in the fast path
> to the slow path to eliminate the additional read. It also clarify the
> action for the interrupt context readers more explicitly.
> 
> Signed-off-by: Waiman Long <Waiman.Long@hp.com>
> ---
>  include/asm-generic/qrwlock.h |    4 ++--
>  kernel/locking/qrwlock.c      |   14 ++++++++------
>  2 files changed, 10 insertions(+), 8 deletions(-)

[...]

> diff --git a/kernel/locking/qrwlock.c b/kernel/locking/qrwlock.c
> index 00c12bb..d7d7557 100644
> --- a/kernel/locking/qrwlock.c
> +++ b/kernel/locking/qrwlock.c
> @@ -43,22 +43,24 @@ rspin_until_writer_unlock(struct qrwlock *lock, u32 cnts)
>   * queue_read_lock_slowpath - acquire read lock of a queue rwlock
>   * @lock: Pointer to queue rwlock structure
>   */
> -void queue_read_lock_slowpath(struct qrwlock *lock)
> +void queue_read_lock_slowpath(struct qrwlock *lock, u32 cnts)
>  {
> -	u32 cnts;
> -
>  	/*
>  	 * Readers come here when they cannot get the lock without waiting
>  	 */
>  	if (unlikely(in_interrupt())) {
>  		/*
> -		 * Readers in interrupt context will spin until the lock is
> -		 * available without waiting in the queue.
> +		 * Readers in interrupt context will get the lock immediately
> +		 * if the writer is just waiting (not holding the lock yet)
> +		 * or they will spin until the lock is available without
> +		 * waiting in the queue.
>  		 */
> -		cnts = smp_load_acquire((u32 *)&lock->cnts);
> +		if ((cnts & _QW_WMASK) != _QW_LOCKED)
> +			return;

I really doubt the check here is gaining you any performance, given
rspin_until_write_unlock does the same check immediately and should be
inlined. Just dropping the acquire and passing cnts through should be
sufficient.

Will