Re: [PATCH] locking/pvqspinlock: Add lock holder CPU argument to pv_wait()

[Waiman Long <waiman.long@hpe.com>]

On 04/14/2016 10:34 AM, Pan Xinhui wrote:
> hello, Waiman
> I try your patch, thanks!
>
> also I do some improvement.
> below code diff has been tested, it works for me. :)
>
> diff --git a/kernel/locking/qspinlock.c b/kernel/locking/qspinlock.c
> index ce2f75e..99f31e4 100644
> --- a/kernel/locking/qspinlock.c
> +++ b/kernel/locking/qspinlock.c
> @@ -248,7 +248,8 @@ static __always_inline void set_locked(struct qspinlock *lock)
>    */
>
>   static __always_inline void __pv_init_node(struct mcs_spinlock *node) { }
> -static __always_inline void __pv_wait_node(struct mcs_spinlock *node,
> +static __always_inline void __pv_wait_node(struct qspinlock *lock,
> +                                          struct mcs_spinlock *node,
>                                             struct mcs_spinlock *prev) { }
>   static __always_inline void __pv_kick_node(struct qspinlock *lock,
>                                             struct mcs_spinlock *node) { }
> @@ -407,7 +408,7 @@ queue:
>                  prev = decode_tail(old);
>                  WRITE_ONCE(prev->next, node);
>
> -               pv_wait_node(node, prev);
> +               pv_wait_node(lock, node, prev);
>                  arch_mcs_spin_lock_contended(&node->locked);
>
>                  /*
> diff --git a/kernel/locking/qspinlock_paravirt.h b/kernel/locking/qspinlock_paravirt.h
> index 01a6d16..75ccfd3 100644
> --- a/kernel/locking/qspinlock_paravirt.h
> +++ b/kernel/locking/qspinlock_paravirt.h
> @@ -255,6 +257,19 @@ static struct pv_node *pv_unhash(struct qspinlock *lock)
>          BUG();
>   }
>
> +static struct pv_node *pv_hash_lookup(struct qspinlock *lock)
> +{
> +       unsigned long offset, hash = hash_ptr(lock, pv_lock_hash_bits);
> +       struct pv_hash_entry *he;
> +
> +       for_each_hash_entry(he, offset, hash) {
> +               if (READ_ONCE(he->lock) == lock) {
> +                       return he->node;
> +               }
> +       }
> +       return NULL;
> +}
> +
>   /*
>    * Return true if when it is time to check the previous node which is not
>    * in a running state.
> @@ -287,14 +303,17 @@ static void pv_init_node(struct mcs_spinlock *node)
>    * pv_kick_node() is used to set _Q_SLOW_VAL and fill in hash table on its
>    * behalf.
>    */
> -static void pv_wait_node(struct mcs_spinlock *node, struct mcs_spinlock *prev)
> +static void pv_wait_node(struct qspinlock *lock, struct mcs_spinlock *node,
> +                       struct mcs_spinlock *prev)
>   {
>          struct pv_node *pn = (struct pv_node *)node;
>          struct pv_node *pp = (struct pv_node *)prev;
> +       struct pv_node *ph;
>          int waitcnt = 0;
>          int loop;
>          bool wait_early;
>
> +       pn->prev_cpu = pp->cpu;
>          /* waitcnt processing will be compiled out if !QUEUED_LOCK_STAT */
>          for (;; waitcnt++) {
>                  for (wait_early = false, loop = SPIN_THRESHOLD; loop; loop--) {
> @@ -322,7 +341,10 @@ static void pv_wait_node(struct mcs_spinlock *node, struct mcs_spinlock *prev)
>                          qstat_inc(qstat_pv_wait_node, true);
>                          qstat_inc(qstat_pv_wait_again, waitcnt);
>                          qstat_inc(qstat_pv_wait_early, wait_early);
> -                       pv_wait(&pn->state, vcpu_halted);
> +                       ph = pv_hash_lookup(lock);
> +                       if (!ph)
> +                               ph = pp;
> +                       pv_wait(&pn->state, vcpu_halted, ph->prev_cpu);
>                  }
>
>                  /*
>
> any comments are welcome. I put my patch here just for simplicity, or need I send it out in a new thread?
>
> thanks
> xinhui
>
>

I have sent out a v2 patch that incorporate your change with some minor 
twists.

Cheers,
Longman