Message-ID: <563CE93E.6090000@hpe.com>
Date: Fri, 06 Nov 2015 12:54:06 -0500
From: Waiman Long <waiman.long@hpe.com>
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:10.0.12) Gecko/20130109 Thunderbird/10.0.12
MIME-Version: 1.0
To: Peter Zijlstra <peterz@infradead.org>
CC: Ingo Molnar <mingo@redhat.com>, Thomas Gleixner <tglx@linutronix.de>,
        "H. Peter Anvin" <hpa@zytor.com>, x86@kernel.org,
        linux-kernel@vger.kernel.org, Scott J Norton <scott.norton@hpe.com>,
        Douglas Hatch <doug.hatch@hpe.com>,
        Davidlohr Bueso <dave@stgolabs.net>
Subject: Re: [PATCH tip/locking/core v9 6/6] locking/pvqspinlock: Queue node
 adaptive spinning
References: <1446247597-61863-1-git-send-email-Waiman.Long@hpe.com> <1446247597-61863-7-git-send-email-Waiman.Long@hpe.com> <20151106150149.GV17308@twins.programming.kicks-ass.net>
In-Reply-To: <20151106150149.GV17308@twins.programming.kicks-ass.net>
Content-Type: text/plain; charset=ISO-8859-1; format=flowed
Content-Transfer-Encoding: 7bit
Sender: linux-kernel-owner@vger.kernel.org
Content-Length: 3486
Lines: 104

On 11/06/2015 10:01 AM, Peter Zijlstra wrote:
> On Fri, Oct 30, 2015 at 07:26:37PM -0400, Waiman Long wrote:
>> +++ b/kernel/locking/qspinlock_paravirt.h
>> @@ -23,6 +23,19 @@
>>   #define _Q_SLOW_VAL	(3U<<  _Q_LOCKED_OFFSET)
>>
>>   /*
>> + * Queue Node Adaptive Spinning
>> + *
>> + * A queue node vCPU will stop spinning if the vCPU in the previous node is
>> + * not running. The one lock stealing attempt allowed at slowpath entry
>> + * mitigates the slight slowdown for non-overcommitted guest with this
>> + * aggressive wait-early mechanism.
>> + *
>> + * The status of the previous node will be checked at fixed interval
>> + * controlled by PV_PREV_CHECK_MASK.
>> + */
>> +#define PV_PREV_CHECK_MASK	0xff
>> +
>> +/*
>>    * Queue node uses: vcpu_running&  vcpu_halted.
>>    * Queue head uses: vcpu_running&  vcpu_hashed.
>>    */
>> @@ -202,6 +215,20 @@ static struct pv_node *pv_unhash(struct qspinlock *lock)
>>   }
>>
>>   /*
>> + * Return true if when it is time to check the previous node which is not
>> + * in a running state.
>> + */
>> +static inline bool
>> +pv_wait_early(struct pv_node *prev, int loop)
>> +{
>> +
>> +	if ((loop&  PV_PREV_CHECK_MASK) != 0)
>> +		return false;
>> +
>> +	return READ_ONCE(prev->state) != vcpu_running;
>> +}
> So it appears to me the sole purpose of PV_PREV_CHECK_MASK it to avoid
> touching the prev->state cacheline too hard. Yet that is not mentioned
> anywhere above.

Yes, that is true. I will add a comment to that effect.

>
>> +static void pv_wait_node(struct mcs_spinlock *node, struct mcs_spinlock *prev)
>>   {
>>   	struct pv_node *pn = (struct pv_node *)node;
>> +	struct pv_node *pp = (struct pv_node *)prev;
>>   	int waitcnt = 0;
>>   	int loop;
>> +	bool wait_early;
>>
>>   	/* waitcnt processing will be compiled out if !QUEUED_LOCK_STAT */
>>   	for (;; waitcnt++) {
>> -		for (loop = SPIN_THRESHOLD; loop; loop--) {
>> +		for (wait_early = false, loop = SPIN_THRESHOLD; loop; loop--) {
>>   			if (READ_ONCE(node->locked))
>>   				return;
>> +			if (pv_wait_early(pp, loop)) {
>> +				wait_early = true;
>> +				break;
>> +			}
>>   			cpu_relax();
>>   		}
>>
> So if prev points to another node, it will never see vcpu_running. Was
> that fully intended?

I had added code in pv_wait_head_or_lock to set the state appropriately 
for the queue head vCPU.

         for (;; waitcnt++) {
                 /*
+                * Set correct vCPU state to be used by queue node 
wait-early
+                * mechanism.
+                */
+               WRITE_ONCE(pn->state, vcpu_running);
+
+               /*
                  * Set the pending bit in the active lock spinning loop to
                  * disable lock stealing. However, the pending bit check in
                  * pv_queued_spin_trylock_unfair() and the setting/clearing
@@ -374,6 +414,7 @@ static u32 pv_wait_head_lock(struct qspinlock *lock, 
struct mcs_spinlock *node)
                                 goto gotlock;
                         }
                 }
+               WRITE_ONCE(pn->state, vcpu_halted);

> FYI, I think I've now seen all patches ;-)

Thanks for the review. I will work on fixing the issues you identified 
and issue a new patch series next week.

Cheers,
Longman
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/