* Re: [PATCH 43/54] drivers/hv: replace cpumask_weight with cpumask_weight_eq
2022-01-23 18:39 ` [PATCH 43/54] drivers/hv: replace cpumask_weight with cpumask_weight_eq Yury Norov
@ 2022-01-23 22:00 ` Wei Liu
2022-01-23 22:02 ` Haiyang Zhang
2022-01-24 9:20 ` Vitaly Kuznetsov
2 siblings, 0 replies; 6+ messages in thread
From: Wei Liu @ 2022-01-23 22:00 UTC (permalink / raw)
To: Yury Norov
Cc: Andy Shevchenko, Rasmus Villemoes, Andrew Morton,
Michał Mirosław, Greg Kroah-Hartman, Peter Zijlstra,
David Laight, Joe Perches, Dennis Zhou, Emil Renner Berthing,
Nicholas Piggin, Matti Vaittinen, Alexey Klimov, linux-kernel,
K. Y. Srinivasan, Haiyang Zhang, Stephen Hemminger, Wei Liu,
Dexuan Cui, linux-hyperv
On Sun, Jan 23, 2022 at 10:39:14AM -0800, Yury Norov wrote:
> init_vp_index() calls cpumask_weight() to compare the weights of cpumasks
> We can do it more efficiently with cpumask_weight_eq because conditional
> cpumask_weight may stop traversing the cpumask earlier (at least one), as
> soon as condition is met.
>
> Signed-off-by: Yury Norov <yury.norov@gmail.com>
Acked-by: Wei Liu <wei.liu@kernel.org>
^ permalink raw reply [flat|nested] 6+ messages in thread
* RE: [PATCH 43/54] drivers/hv: replace cpumask_weight with cpumask_weight_eq
2022-01-23 18:39 ` [PATCH 43/54] drivers/hv: replace cpumask_weight with cpumask_weight_eq Yury Norov
2022-01-23 22:00 ` Wei Liu
@ 2022-01-23 22:02 ` Haiyang Zhang
2022-01-24 9:20 ` Vitaly Kuznetsov
2 siblings, 0 replies; 6+ messages in thread
From: Haiyang Zhang @ 2022-01-23 22:02 UTC (permalink / raw)
To: Yury Norov, Andy Shevchenko, Rasmus Villemoes, Andrew Morton,
Michał Mirosław, Greg Kroah-Hartman, Peter Zijlstra,
David Laight, Joe Perches, Dennis Zhou, Emil Renner Berthing,
Nicholas Piggin, Matti Vaittinen, Alexey Klimov,
linux-kernel@vger.kernel.org, KY Srinivasan, Stephen Hemminger,
Wei Liu, Dexuan Cui, linux-hyperv@vger.kernel.org
> -----Original Message-----
> From: Yury Norov <yury.norov@gmail.com>
> Sent: Sunday, January 23, 2022 1:39 PM
> To: Yury Norov <yury.norov@gmail.com>; Andy Shevchenko <andriy.shevchenko@linux.intel.com>;
> Rasmus Villemoes <linux@rasmusvillemoes.dk>; Andrew Morton <akpm@linux-foundation.org>;
> Michał Mirosław <mirq-linux@rere.qmqm.pl>; Greg Kroah-Hartman <gregkh@linuxfoundation.org>;
> Peter Zijlstra <peterz@infradead.org>; David Laight <David.Laight@aculab.com>; Joe Perches
> <joe@perches.com>; Dennis Zhou <dennis@kernel.org>; Emil Renner Berthing <kernel@esmil.dk>;
> Nicholas Piggin <npiggin@gmail.com>; Matti Vaittinen <matti.vaittinen@fi.rohmeurope.com>;
> Alexey Klimov <aklimov@redhat.com>; linux-kernel@vger.kernel.org; KY Srinivasan
> <kys@microsoft.com>; Haiyang Zhang <haiyangz@microsoft.com>; Stephen Hemminger
> <sthemmin@microsoft.com>; Wei Liu <wei.liu@kernel.org>; Dexuan Cui <decui@microsoft.com>;
> linux-hyperv@vger.kernel.org
> Subject: [PATCH 43/54] drivers/hv: replace cpumask_weight with cpumask_weight_eq
>
> init_vp_index() calls cpumask_weight() to compare the weights of cpumasks
> We can do it more efficiently with cpumask_weight_eq because conditional
> cpumask_weight may stop traversing the cpumask earlier (at least one), as
> soon as condition is met.
>
> Signed-off-by: Yury Norov <yury.norov@gmail.com>
> ---
> drivers/hv/channel_mgmt.c | 4 ++--
> 1 file changed, 2 insertions(+), 2 deletions(-)
>
> diff --git a/drivers/hv/channel_mgmt.c b/drivers/hv/channel_mgmt.c
> index 60375879612f..7420a5fd47b5 100644
> --- a/drivers/hv/channel_mgmt.c
> +++ b/drivers/hv/channel_mgmt.c
> @@ -762,8 +762,8 @@ static void init_vp_index(struct vmbus_channel *channel)
> }
> alloced_mask = &hv_context.hv_numa_map[numa_node];
>
> - if (cpumask_weight(alloced_mask) ==
> - cpumask_weight(cpumask_of_node(numa_node))) {
> + if (cpumask_weight_eq(alloced_mask,
> + cpumask_weight(cpumask_of_node(numa_node)))) {
> /*
> * We have cycled through all the CPUs in the node;
> * reset the alloced map.
Thanks.
Reviewed-by: Haiyang Zhang <haiyangz@microsoft.com>
^ permalink raw reply [flat|nested] 6+ messages in thread* Re: [PATCH 43/54] drivers/hv: replace cpumask_weight with cpumask_weight_eq
2022-01-23 18:39 ` [PATCH 43/54] drivers/hv: replace cpumask_weight with cpumask_weight_eq Yury Norov
2022-01-23 22:00 ` Wei Liu
2022-01-23 22:02 ` Haiyang Zhang
@ 2022-01-24 9:20 ` Vitaly Kuznetsov
2022-01-27 15:02 ` Michael Kelley (LINUX)
2 siblings, 1 reply; 6+ messages in thread
From: Vitaly Kuznetsov @ 2022-01-24 9:20 UTC (permalink / raw)
To: Yury Norov
Cc: Andy Shevchenko, Rasmus Villemoes, Andrew Morton,
Michał Mirosław, Greg Kroah-Hartman, Peter Zijlstra,
David Laight, Joe Perches, Dennis Zhou, Emil Renner Berthing,
Nicholas Piggin, Matti Vaittinen, Alexey Klimov, linux-kernel,
K. Y. Srinivasan, Haiyang Zhang, Stephen Hemminger, Wei Liu,
Dexuan Cui, linux-hyperv
Yury Norov <yury.norov@gmail.com> writes:
> init_vp_index() calls cpumask_weight() to compare the weights of cpumasks
> We can do it more efficiently with cpumask_weight_eq because conditional
> cpumask_weight may stop traversing the cpumask earlier (at least one), as
> soon as condition is met.
Same comment as for "PATCH 41/54": cpumask_weight_eq() can only stop
earlier if the condition is not met, to prove the equality all bits need
always have to be examined.
>
> Signed-off-by: Yury Norov <yury.norov@gmail.com>
> ---
> drivers/hv/channel_mgmt.c | 4 ++--
> 1 file changed, 2 insertions(+), 2 deletions(-)
>
> diff --git a/drivers/hv/channel_mgmt.c b/drivers/hv/channel_mgmt.c
> index 60375879612f..7420a5fd47b5 100644
> --- a/drivers/hv/channel_mgmt.c
> +++ b/drivers/hv/channel_mgmt.c
> @@ -762,8 +762,8 @@ static void init_vp_index(struct vmbus_channel *channel)
> }
> alloced_mask = &hv_context.hv_numa_map[numa_node];
>
> - if (cpumask_weight(alloced_mask) ==
> - cpumask_weight(cpumask_of_node(numa_node))) {
> + if (cpumask_weight_eq(alloced_mask,
> + cpumask_weight(cpumask_of_node(numa_node)))) {
This code is not performace critical and I prefer the old version:
cpumask_weight() == cpumask_weight()
looks better than
cpumask_weight_eq(..., cpumask_weight())
(let alone the inner cpumask_of_node()) to me.
> /*
> * We have cycled through all the CPUs in the node;
> * reset the alloced map.
--
Vitaly
^ permalink raw reply [flat|nested] 6+ messages in thread* RE: [PATCH 43/54] drivers/hv: replace cpumask_weight with cpumask_weight_eq
2022-01-24 9:20 ` Vitaly Kuznetsov
@ 2022-01-27 15:02 ` Michael Kelley (LINUX)
2022-01-28 9:31 ` Vitaly Kuznetsov
0 siblings, 1 reply; 6+ messages in thread
From: Michael Kelley (LINUX) @ 2022-01-27 15:02 UTC (permalink / raw)
To: vkuznets, Yury Norov
Cc: Andy Shevchenko, Rasmus Villemoes, Andrew Morton,
Michał Mirosław, Greg Kroah-Hartman, Peter Zijlstra,
David Laight, Joe Perches, Dennis Zhou, Emil Renner Berthing,
Nicholas Piggin, Matti Vaittinen, Alexey Klimov,
linux-kernel@vger.kernel.org, KY Srinivasan, Haiyang Zhang,
Stephen Hemminger, Wei Liu, Dexuan Cui,
linux-hyperv@vger.kernel.org
From: Vitaly Kuznetsov <vkuznets@redhat.com> Sent: Monday, January 24, 2022 1:20 AM
>
> Yury Norov <yury.norov@gmail.com> writes:
>
> > init_vp_index() calls cpumask_weight() to compare the weights of cpumasks
> > We can do it more efficiently with cpumask_weight_eq because conditional
> > cpumask_weight may stop traversing the cpumask earlier (at least one), as
> > soon as condition is met.
>
> Same comment as for "PATCH 41/54": cpumask_weight_eq() can only stop
> earlier if the condition is not met, to prove the equality all bits need
> always have to be examined.
>
> >
> > Signed-off-by: Yury Norov <yury.norov@gmail.com>
> > ---
> > drivers/hv/channel_mgmt.c | 4 ++--
> > 1 file changed, 2 insertions(+), 2 deletions(-)
> >
> > diff --git a/drivers/hv/channel_mgmt.c b/drivers/hv/channel_mgmt.c
> > index 60375879612f..7420a5fd47b5 100644
> > --- a/drivers/hv/channel_mgmt.c
> > +++ b/drivers/hv/channel_mgmt.c
> > @@ -762,8 +762,8 @@ static void init_vp_index(struct vmbus_channel *channel)
> > }
> > alloced_mask = &hv_context.hv_numa_map[numa_node];
> >
> > - if (cpumask_weight(alloced_mask) ==
> > - cpumask_weight(cpumask_of_node(numa_node))) {
> > + if (cpumask_weight_eq(alloced_mask,
> > + cpumask_weight(cpumask_of_node(numa_node)))) {
>
> This code is not performace critical and I prefer the old version:
>
> cpumask_weight() == cpumask_weight()
>
> looks better than
>
> cpumask_weight_eq(..., cpumask_weight())
>
> (let alone the inner cpumask_of_node()) to me.
>
> > /*
> > * We have cycled through all the CPUs in the node;
> > * reset the alloced map.
>
> --
> Vitaly
I agree with Vitaly in preferring the old version, and indeed performance
here is a shrug. But actually, I think the old version is a poorly coded way
to determine if the two cpumasks are equal. The following would correctly
capture the intent:
if (cpumask_equal(alloced_mask, cpumask_of_node(numa_node))
Michael
^ permalink raw reply [flat|nested] 6+ messages in thread* RE: [PATCH 43/54] drivers/hv: replace cpumask_weight with cpumask_weight_eq
2022-01-27 15:02 ` Michael Kelley (LINUX)
@ 2022-01-28 9:31 ` Vitaly Kuznetsov
0 siblings, 0 replies; 6+ messages in thread
From: Vitaly Kuznetsov @ 2022-01-28 9:31 UTC (permalink / raw)
To: Michael Kelley (LINUX)
Cc: Andy Shevchenko, Rasmus Villemoes, Andrew Morton,
Michał Mirosław, Greg Kroah-Hartman, Peter Zijlstra,
David Laight, Joe Perches, Dennis Zhou, Emil Renner Berthing,
Nicholas Piggin, Matti Vaittinen, Alexey Klimov,
linux-kernel@vger.kernel.org, KY Srinivasan, Haiyang Zhang,
Stephen Hemminger, Wei Liu, Dexuan Cui,
linux-hyperv@vger.kernel.org, Yury Norov
"Michael Kelley (LINUX)" <mikelley@microsoft.com> writes:
> From: Vitaly Kuznetsov <vkuznets@redhat.com> Sent: Monday, January 24, 2022 1:20 AM
>>
>> Yury Norov <yury.norov@gmail.com> writes:
>>
...
>> >
>> > diff --git a/drivers/hv/channel_mgmt.c b/drivers/hv/channel_mgmt.c
>> > index 60375879612f..7420a5fd47b5 100644
>> > --- a/drivers/hv/channel_mgmt.c
>> > +++ b/drivers/hv/channel_mgmt.c
>> > @@ -762,8 +762,8 @@ static void init_vp_index(struct vmbus_channel *channel)
>> > }
>> > alloced_mask = &hv_context.hv_numa_map[numa_node];
>> >
>> > - if (cpumask_weight(alloced_mask) ==
>> > - cpumask_weight(cpumask_of_node(numa_node))) {
>> > + if (cpumask_weight_eq(alloced_mask,
>> > + cpumask_weight(cpumask_of_node(numa_node)))) {
>>
>> This code is not performace critical and I prefer the old version:
>>
>> cpumask_weight() == cpumask_weight()
>>
>> looks better than
>>
>> cpumask_weight_eq(..., cpumask_weight())
>>
>> (let alone the inner cpumask_of_node()) to me.
>>
>> > /*
>> > * We have cycled through all the CPUs in the node;
>> > * reset the alloced map.
>>
> I agree with Vitaly in preferring the old version, and indeed performance
> here is a shrug. But actually, I think the old version is a poorly coded way
> to determine if the two cpumasks are equal. The following would correctly
> capture the intent:
>
> if (cpumask_equal(alloced_mask, cpumask_of_node(numa_node))
>
Indeed. While it seems that only CPUs from 'cpumask_of_node(numa_node)'
can be set in 'alloced_mask' (and thus the comparison is valid), there's
no real need to weigh anything. I'll send a patch.
--
Vitaly
^ permalink raw reply [flat|nested] 6+ messages in thread