Re: [PATCH 04/13] locking/qspinlock: move pv lock word helpers into qspinlock.c
From: Nicholas Piggin
Date: Mon Jul 11 2022 - 20:11:37 EST
Excerpts from Waiman Long's message of July 6, 2022 5:34 am:
> On 7/4/22 10:38, Nicholas Piggin wrote:
>> There is no real reason not to keep all the bit manipulation together.
>> Signed-off-by: Nicholas Piggin <npiggin@xxxxxxxxx>
>> kernel/locking/qspinlock.c | 107 ++++++++++++++++------------
>> kernel/locking/qspinlock_paravirt.h | 51 -------------
>> 2 files changed, 63 insertions(+), 95 deletions(-)
>> diff --git a/kernel/locking/qspinlock.c b/kernel/locking/qspinlock.c
>> index 7360d643de29..8f2173e22479 100644
>> --- a/kernel/locking/qspinlock.c
>> +++ b/kernel/locking/qspinlock.c
>> @@ -141,7 +141,24 @@ struct qnode *grab_qnode(struct qnode *base, int idx)
>> #define _Q_LOCKED_PENDING_MASK (_Q_LOCKED_MASK | _Q_PENDING_MASK)
>> + * set_pending - set the pending bit.
>> + * @lock: Pointer to queued spinlock structure
>> + *
>> + * *,0,* -> *,1,*
>> + *
>> + * The pending bit is used by the queue head vCPU to indicate that it
>> + * is actively spinning on the lock and no lock stealing is allowed.
> The pending bit has different usage in pv and non-pv cases. The
> description here refers to the pv case. For non-pv, it is used to avoid
> loading the the extra node cacheline in likely contended case.
Ah thank you, I can expand on that.