Re: [PATCH v2 04/13] locking/qspinlock: Remove unbounded cmpxchg loop from locking slowpath

From: Will Deacon
Date: Thu Apr 12 2018 - 10:18:17 EST


On Thu, Apr 12, 2018 at 10:16:55AM -0400, Waiman Long wrote:
> On 04/12/2018 10:06 AM, Will Deacon wrote:
> > On Wed, Apr 11, 2018 at 03:53:16PM -0400, Waiman Long wrote:
> >>> diff --git a/kernel/locking/qspinlock.c b/kernel/locking/qspinlock.c
> >>> index 396701e8c62d..a8fc402b3f3a 100644
> >>> --- a/kernel/locking/qspinlock.c
> >>> +++ b/kernel/locking/qspinlock.c
> >>> @@ -162,6 +162,17 @@ struct __qspinlock {
> >>>
> >>> #if _Q_PENDING_BITS == 8
> >>> /**
> >>> + * clear_pending - clear the pending bit.
> >>> + * @lock: Pointer to queued spinlock structure
> >>> + *
> >>> + * *,1,* -> *,0,*
> >>> + */
> >>> +static __always_inline void clear_pending(struct qspinlock *lock)
> >>> +{
> >>> + WRITE_ONCE(lock->pending, 0);
> >>> +}
> >>> +
> >>> +/**
> >>> * clear_pending_set_locked - take ownership and clear the pending bit.
> >>> * @lock: Pointer to queued spinlock structure
> >>> *
> >>> @@ -201,6 +212,17 @@ static __always_inline u32 xchg_tail(struct qspinlock *lock, u32 tail)
> >>> #else /* _Q_PENDING_BITS == 8 */
> >>>
> >>> /**
> >>> + * clear_pending - clear the pending bit.
> >>> + * @lock: Pointer to queued spinlock structure
> >>> + *
> >>> + * *,1,* -> *,0,*
> >>> + */
> >>> +static __always_inline void clear_pending(struct qspinlock *lock)
> >>> +{
> >>> + atomic_andnot(_Q_PENDING_VAL, &lock->val);
> >>> +}
> >>> +
> >>> +/**
> >>> * clear_pending_set_locked - take ownership and clear the pending bit.
> >>> * @lock: Pointer to queued spinlock structure
> >>> *
> >> BTW, there is a similar clear_pending() function in
> >> qspinlock_paravirt.c. I think you need to remove that with this patch.
> > Thanks, I'll do that. I did build and bisect this series... for arm64, which
> > is completely useless as it doesn't get compiled until the final patch which
> > I haven't posted here.
> >
> > Will
>
> You certainly need to get a x86 system and compiles with the patches
> applied one-by-one with CONFIG_PARAVIRT_SPINLOCKS enabled.

Yup, doing that now. I just botched my test scripts before, which default to
arm64.

Will