[PATCH] lib/cpumask.c: Optimize __any_online_cpu() calculation

[PATCH] lib/cpumask.c: Optimize __any_online_cpu() calculation

[Srivatsa S. Bhat]

__any_online_cpu() uses a for loop at the moment.
Instead, use cpumask_* operations to speed it up.

Signed-off-by: Srivatsa S. Bhat <srivatsa.bhat@linux.vnet.ibm.com>
---
Actually, the patch posted at https://lkml.org/lkml/2012/2/15/101 removed
the last user of any_online_cpu() (and hence __any_online_cpu()).
However, since this is an exported symbol, I refrained from removing this
function altogether.

 lib/cpumask.c |    8 +-------
 1 files changed, 1 insertions(+), 7 deletions(-)

diff --git a/lib/cpumask.c b/lib/cpumask.c
index af3e5817..b25f53d 100644
--- a/lib/cpumask.c
+++ b/lib/cpumask.c
@@ -28,13 +28,7 @@ EXPORT_SYMBOL(__next_cpu_nr);
 
 int __any_online_cpu(const cpumask_t *mask)
 {
-	int cpu;
-
-	for_each_cpu(cpu, mask) {
-		if (cpu_online(cpu))
-			break;
-	}
-	return cpu;
+	return cpumask_any_and(mask, cpu_online_mask);
 }
 EXPORT_SYMBOL(__any_online_cpu);
 

Re: [PATCH] lib/cpumask.c: Optimize __any_online_cpu() calculation

[Eric Dumazet]

Le mercredi 15 février 2012 à 18:05 +0530, Srivatsa S. Bhat a écrit :
> __any_online_cpu() uses a for loop at the moment.
> Instead, use cpumask_* operations to speed it up.
> 
> Signed-off-by: Srivatsa S. Bhat <srivatsa.bhat@linux.vnet.ibm.com>
> ---
> Actually, the patch posted at https://lkml.org/lkml/2012/2/15/101 removed
> the last user of any_online_cpu() (and hence __any_online_cpu()).
> However, since this is an exported symbol, I refrained from removing this
> function altogether.

Just remove the export/function and eventually leave the helper ?

#define any_online_cpu(mask) cpumask_any_and(mask, cpu_online_mask)

Re: [PATCH] lib/cpumask.c: Optimize __any_online_cpu() calculation

[Srivatsa S. Bhat]

On 02/15/2012 06:27 PM, Eric Dumazet wrote:

> Le mercredi 15 février 2012 à 18:05 +0530, Srivatsa S. Bhat a écrit :
>> __any_online_cpu() uses a for loop at the moment.
>> Instead, use cpumask_* operations to speed it up.
>>
>> Signed-off-by: Srivatsa S. Bhat <srivatsa.bhat@linux.vnet.ibm.com>
>> ---
>> Actually, the patch posted at https://lkml.org/lkml/2012/2/15/101 removed
>> the last user of any_online_cpu() (and hence __any_online_cpu()).
>> However, since this is an exported symbol, I refrained from removing this
>> function altogether.
> 
> Just remove the export/function and eventually leave the helper ?
> 
> #define any_online_cpu(mask) cpumask_any_and(mask, cpu_online_mask)
> 


That sounds good. Here is v2 of the patch.
(Since this v2 touches include/linux/cpumask.h, I rebased it on top of the
patch posted at: https://lkml.org/lkml/2012/2/15/2. Even without that
patch it should apply with minor line number changes..)

---
From: Srivatsa S. Bhat <srivatsa.bhat@linux.vnet.ibm.com>
Subject: [PATCH v2] lib/cpumask.c: Remove __any_online_cpu()

__any_online_cpu() is not optimal and also unnecessary.
So, replace its use by faster cpumask_* operations.

Signed-off-by: Srivatsa S. Bhat <srivatsa.bhat@linux.vnet.ibm.com>
---

 include/linux/cpumask.h |    3 +--
 lib/cpumask.c           |   12 ------------
 2 files changed, 1 insertions(+), 14 deletions(-)

diff --git a/include/linux/cpumask.h b/include/linux/cpumask.h
index 42fec19..d3ccd2d 100644
--- a/include/linux/cpumask.h
+++ b/include/linux/cpumask.h
@@ -808,11 +808,10 @@ static inline const struct cpumask *get_cpu_mask(unsigned int cpu)
 #else /* NR_CPUS > 1 */
 int __first_cpu(const cpumask_t *srcp);
 int __next_cpu(int n, const cpumask_t *srcp);
-int __any_online_cpu(const cpumask_t *mask);
 
 #define first_cpu(src)		__first_cpu(&(src))
 #define next_cpu(n, src)	__next_cpu((n), &(src))
-#define any_online_cpu(mask) __any_online_cpu(&(mask))
+#define any_online_cpu(mask) cpumask_any_and(&mask, cpu_online_mask)
 #define for_each_cpu_mask(cpu, mask)			\
 	for ((cpu) = -1;				\
 		(cpu) = next_cpu((cpu), (mask)),	\
diff --git a/lib/cpumask.c b/lib/cpumask.c
index af3e5817..0470742 100644
--- a/lib/cpumask.c
+++ b/lib/cpumask.c
@@ -26,18 +26,6 @@ int __next_cpu_nr(int n, const cpumask_t *srcp)
 EXPORT_SYMBOL(__next_cpu_nr);
 #endif
 
-int __any_online_cpu(const cpumask_t *mask)
-{
-	int cpu;
-
-	for_each_cpu(cpu, mask) {
-		if (cpu_online(cpu))
-			break;
-	}
-	return cpu;
-}
-EXPORT_SYMBOL(__any_online_cpu);
-
 /**
  * cpumask_next_and - get the next cpu in *src1p & *src2p
  * @n: the cpu prior to the place to search (ie. return will be > @n)

Re: [PATCH] lib/cpumask.c: Optimize __any_online_cpu() calculation

[Rusty Russell]

On Wed, 15 Feb 2012 19:49:00 +0530, "Srivatsa S. Bhat" <srivatsa.bhat@linux.vnet.ibm.com> wrote:
> On 02/15/2012 06:27 PM, Eric Dumazet wrote:
> 
> > Le mercredi 15 février 2012 à 18:05 +0530, Srivatsa S. Bhat a écrit :
> >> __any_online_cpu() uses a for loop at the moment.
> >> Instead, use cpumask_* operations to speed it up.
> >>
> >> Signed-off-by: Srivatsa S. Bhat <srivatsa.bhat@linux.vnet.ibm.com>
> >> ---
> >> Actually, the patch posted at https://lkml.org/lkml/2012/2/15/101 removed
> >> the last user of any_online_cpu() (and hence __any_online_cpu()).
> >> However, since this is an exported symbol, I refrained from removing this
> >> function altogether.
> > 
> > Just remove the export/function and eventually leave the helper ?
> > 
> > #define any_online_cpu(mask) cpumask_any_and(mask, cpu_online_mask)
> > 
> 
> 
> That sounds good. Here is v2 of the patch.
> (Since this v2 touches include/linux/cpumask.h, I rebased it on top of the
> patch posted at: https://lkml.org/lkml/2012/2/15/2. Even without that
> patch it should apply with minor line number changes..)

No.  Just git rid of the only caller of any_online_cpu():

arch/ia64/kernel/smpboot.c:693:			new_cpei_cpu = any_online_cpu(cpu_online_map);

Thanks,
Rusty.

Re: [PATCH] lib/cpumask.c: Optimize __any_online_cpu() calculation

[Srivatsa S. Bhat]

On 02/16/2012 04:21 AM, Rusty Russell wrote:

> On Wed, 15 Feb 2012 19:49:00 +0530, "Srivatsa S. Bhat" <srivatsa.bhat@linux.vnet.ibm.com> wrote:
>> On 02/15/2012 06:27 PM, Eric Dumazet wrote:
>>
>>> Le mercredi 15 février 2012 à 18:05 +0530, Srivatsa S. Bhat a écrit :
>>>> __any_online_cpu() uses a for loop at the moment.
>>>> Instead, use cpumask_* operations to speed it up.
>>>>
>>>> Signed-off-by: Srivatsa S. Bhat <srivatsa.bhat@linux.vnet.ibm.com>
>>>> ---
>>>> Actually, the patch posted at https://lkml.org/lkml/2012/2/15/101 removed
>>>> the last user of any_online_cpu() (and hence __any_online_cpu()).
>>>> However, since this is an exported symbol, I refrained from removing this
>>>> function altogether.
>>>
>>> Just remove the export/function and eventually leave the helper ?
>>>
>>> #define any_online_cpu(mask) cpumask_any_and(mask, cpu_online_mask)
>>>
>>
>>
>> That sounds good. Here is v2 of the patch.
>> (Since this v2 touches include/linux/cpumask.h, I rebased it on top of the
>> patch posted at: https://lkml.org/lkml/2012/2/15/2. Even without that
>> patch it should apply with minor line number changes..)
> 
> No.  Just git rid of the only caller of any_online_cpu():
> 
> arch/ia64/kernel/smpboot.c:693:			new_cpei_cpu = any_online_cpu(cpu_online_map);
> 


My ia64 patch already got rid of that caller! So this patch should be fine as
is. And it looks like Andrew Morton has taken this patch into -mm. So nothing
more needs to be done about this :-)

Regards,
Srivatsa S. Bhat