Re: [RFC PATCH] QRCU fastpath optimization

2007-02-12 Thread Paul E. McKenney
On Mon, Feb 12, 2007 at 07:22:09AM +0100, Jens Axboe wrote: > On Sun, Feb 11 2007, Paul E. McKenney wrote: > > This patch optimizes the "quick" RCU update-side fastpath, so that in the > > absence of readers, synchronize_qrcu() does four non-atomic comparisons > > and three memory barriers,

Re: [RFC PATCH] QRCU fastpath optimization

2007-02-12 Thread Paul E. McKenney
On Mon, Feb 12, 2007 at 07:22:09AM +0100, Jens Axboe wrote: On Sun, Feb 11 2007, Paul E. McKenney wrote: This patch optimizes the quick RCU update-side fastpath, so that in the absence of readers, synchronize_qrcu() does four non-atomic comparisons and three memory barriers, eliminating the

Re: [RFC PATCH] QRCU fastpath optimization

2007-02-11 Thread Jens Axboe
On Sun, Feb 11 2007, Paul E. McKenney wrote: > This patch optimizes the "quick" RCU update-side fastpath, so that in the > absence of readers, synchronize_qrcu() does four non-atomic comparisons > and three memory barriers, eliminating the need to acquire the global > lock in this case. Lightly

[RFC PATCH] QRCU fastpath optimization

2007-02-11 Thread Paul E. McKenney
This patch optimizes the "quick" RCU update-side fastpath, so that in the absence of readers, synchronize_qrcu() does four non-atomic comparisons and three memory barriers, eliminating the need to acquire the global lock in this case. Lightly tested. Algorithm has been validated for the

[RFC PATCH] QRCU fastpath optimization

2007-02-11 Thread Paul E. McKenney
This patch optimizes the quick RCU update-side fastpath, so that in the absence of readers, synchronize_qrcu() does four non-atomic comparisons and three memory barriers, eliminating the need to acquire the global lock in this case. Lightly tested. Algorithm has been validated for the

Re: [RFC PATCH] QRCU fastpath optimization

2007-02-11 Thread Jens Axboe
On Sun, Feb 11 2007, Paul E. McKenney wrote: This patch optimizes the quick RCU update-side fastpath, so that in the absence of readers, synchronize_qrcu() does four non-atomic comparisons and three memory barriers, eliminating the need to acquire the global lock in this case. Lightly tested.