Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]

Groups > linux.kernel > #1539197 > unrolled thread

[RFC 0/5] rcu: Introduce leaf_node_for_each_mask_possible_cpu() and its friend

Back to article view | Back to linux.kernel

Contents

  [RFC 0/5] rcu: Introduce leaf_node_for_each_mask_possible_cpu() and its friend Boqun Feng <boqun.feng@gmail.com> - 2016-12-09 09:50 +0100
    [RFC 5/5] rcu: Use leaf_node_for_each_mask_*() for leaf node online CPU iteration Boqun Feng <boqun.feng@gmail.com> - 2016-12-09 10:00 +0100
    Re: [RFC 0/5] rcu: Introduce leaf_node_for_each_mask_possible_cpu()  and its friend "Paul E. McKenney" <paulmck@linux.vnet.ibm.com> - 2016-12-10 00:50 +0100
      Re: [RFC 0/5] rcu: Introduce leaf_node_for_each_mask_possible_cpu()  and its friend Boqun Feng <boqun.feng@gmail.com> - 2016-12-10 01:50 +0100
        Re: [RFC 0/5] rcu: Introduce leaf_node_for_each_mask_possible_cpu()  and its friend "Paul E. McKenney" <paulmck@linux.vnet.ibm.com> - 2016-12-10 05:30 +0100
          Re: [RFC 0/5] rcu: Introduce leaf_node_for_each_mask_possible_cpu()  and its friend Boqun Feng <boqun.feng@gmail.com> - 2016-12-10 14:40 +0100
            Re: [RFC 0/5] rcu: Introduce leaf_node_for_each_mask_possible_cpu()  and its friend "Paul E. McKenney" <paulmck@linux.vnet.ibm.com> - 2016-12-10 18:40 +0100
              Re: [RFC 0/5] rcu: Introduce leaf_node_for_each_mask_possible_cpu()  and its friend Boqun Feng <boqun.feng@gmail.com> - 2016-12-11 01:10 +0100

#1539197 — [RFC 0/5] rcu: Introduce leaf_node_for_each_mask_possible_cpu() and its friend

Hi Paul,

While reading the discussion at:

https://marc.info/?l=linux-kernel&m=148044253400769

I figured we might use this fact to save some extra checks in RCU core code,
currently we iterate over all the possible CPUs on a leaf node, check whether
they were masked in a certain mask and do something. However, given the fact
that the masks on a leaf node should always be sparse than the corresponding
part of cpu_possible_mask, we'd better iterate over all bits in a mask and
check whether the corresponding CPU is possible or not.

So I made this RFC, I did a simple build/boot/rcutorture test on my box with
SMP=4, nothing bad happens. Currently I'm waiting for the 0day and trying to
test this one a bigger system, in the meanwhile, looking forwards to any
comment and suggestion.

So thoughts?

Regards,
Boqun

[toc] | [next] | [standalone]

#1539204 — [RFC 5/5] rcu: Use leaf_node_for_each_mask_*() for leaf node online CPU iteration

Though mostly identical, ->qsmaskinit(A.K.A rcu_rnp_online_cpus) is
sometimes more sparse than the corresponding part of cpu_possible_mask
for an RCU leaf node. So we use leaf_node_for_each_mask_possible_cpu()
in rcu_boost_kthread_setaffinity() instead to save some extra checks.

Signed-off-by: Boqun Feng <boqun.feng@gmail.com>
---
 kernel/rcu/tree_plugin.h | 8 +++++---
 1 file changed, 5 insertions(+), 3 deletions(-)

diff --git a/kernel/rcu/tree_plugin.h b/kernel/rcu/tree_plugin.h
index 69c6eb27c37f..c954c2a7a9ba 100644
--- a/kernel/rcu/tree_plugin.h
+++ b/kernel/rcu/tree_plugin.h
@@ -1170,15 +1170,17 @@ static void rcu_boost_kthread_setaffinity(struct rcu_node *rnp, int outgoingcpu)
 	unsigned long mask = rcu_rnp_online_cpus(rnp);
 	cpumask_var_t cm;
 	int cpu;
+	unsigned long bit;
 
 	if (!t)
 		return;
 	if (!zalloc_cpumask_var(&cm, GFP_KERNEL))
 		return;
-	for_each_leaf_node_possible_cpu(rnp, cpu)
-		if ((mask & leaf_node_cpu_bit(rnp, cpu)) &&
-		    cpu != outgoingcpu)
+
+	leaf_node_for_each_mask_possible_cpu(rnp, mask, bit, cpu)
+		if (cpu != outgoingcpu)
 			cpumask_set_cpu(cpu, cm);
+
 	if (cpumask_weight(cm) == 0)
 		cpumask_setall(cm);
 	set_cpus_allowed_ptr(t, cm);
-- 
2.10.2

[toc] | [prev] | [next] | [standalone]

#1539737 — Re: [RFC 0/5] rcu: Introduce leaf_node_for_each_mask_possible_cpu() and its friend

On Fri, Dec 09, 2016 at 04:48:22PM +0800, Boqun Feng wrote:
> Hi Paul,
> 
> While reading the discussion at:
> 
> https://marc.info/?l=linux-kernel&m=148044253400769

This discussion was for stalls specifically, rather than for routine
scans of the bitmasks.

But it does look to save some code, so worth looking into.

> I figured we might use this fact to save some extra checks in RCU core code,
> currently we iterate over all the possible CPUs on a leaf node, check whether
> they were masked in a certain mask and do something. However, given the fact
> that the masks on a leaf node should always be sparse than the corresponding
> part of cpu_possible_mask, we'd better iterate over all bits in a mask and
> check whether the corresponding CPU is possible or not.
> 
> So I made this RFC, I did a simple build/boot/rcutorture test on my box with
> SMP=4, nothing bad happens. Currently I'm waiting for the 0day and trying to
> test this one a bigger system, in the meanwhile, looking forwards to any
> comment and suggestion.
> 
> So thoughts?

By analogy with for_each_cpu() and for_each_possible_cpu(), the name
should instead be for_each_leaf_node_cpu(), the tradition of excessively
long names in RCU notwithstanding.  ;-)

							Thanx, Paul

[toc] | [prev] | [next] | [standalone]

#1539749 — Re: [RFC 0/5] rcu: Introduce leaf_node_for_each_mask_possible_cpu() and its friend

[Multipart message — attachments visible in raw view] — view raw

On Fri, Dec 09, 2016 at 03:49:45PM -0800, Paul E. McKenney wrote:
> On Fri, Dec 09, 2016 at 04:48:22PM +0800, Boqun Feng wrote:
> > Hi Paul,
> > 
> > While reading the discussion at:
> > 
> > https://marc.info/?l=linux-kernel&m=148044253400769
> 
> This discussion was for stalls specifically, rather than for routine
> scans of the bitmasks.
> 
> But it does look to save some code, so worth looking into.
> 
> > I figured we might use this fact to save some extra checks in RCU core code,
> > currently we iterate over all the possible CPUs on a leaf node, check whether
> > they were masked in a certain mask and do something. However, given the fact
> > that the masks on a leaf node should always be sparse than the corresponding
> > part of cpu_possible_mask, we'd better iterate over all bits in a mask and
> > check whether the corresponding CPU is possible or not.
> > 
> > So I made this RFC, I did a simple build/boot/rcutorture test on my box with
> > SMP=4, nothing bad happens. Currently I'm waiting for the 0day and trying to
> > test this one a bigger system, in the meanwhile, looking forwards to any
> > comment and suggestion.
> > 
> > So thoughts?
> 
> By analogy with for_each_cpu() and for_each_possible_cpu(), the name
> should instead be for_each_leaf_node_cpu(), the tradition of excessively
> long names in RCU notwithstanding.  ;-)
> 

Make sense ;-)

I think it's more appropriate to call it for_each_leaf_node_mask_cpu(),
because we don't iterate all cpus of a leaf node. The word "possible"
could be dropped because obviously we won't iterate over "impossible"
cpus in a leaf node ;-)

Will modify that in next version.

Regards,
Boqun

> 							Thanx, Paul
> 

[toc] | [prev] | [next] | [standalone]

#1539769 — Re: [RFC 0/5] rcu: Introduce leaf_node_for_each_mask_possible_cpu() and its friend

On Sat, Dec 10, 2016 at 08:45:38AM +0800, Boqun Feng wrote:
> On Fri, Dec 09, 2016 at 03:49:45PM -0800, Paul E. McKenney wrote:
> > On Fri, Dec 09, 2016 at 04:48:22PM +0800, Boqun Feng wrote:
> > > Hi Paul,
> > > 
> > > While reading the discussion at:
> > > 
> > > https://marc.info/?l=linux-kernel&m=148044253400769
> > 
> > This discussion was for stalls specifically, rather than for routine
> > scans of the bitmasks.
> > 
> > But it does look to save some code, so worth looking into.
> > 
> > > I figured we might use this fact to save some extra checks in RCU core code,
> > > currently we iterate over all the possible CPUs on a leaf node, check whether
> > > they were masked in a certain mask and do something. However, given the fact
> > > that the masks on a leaf node should always be sparse than the corresponding
> > > part of cpu_possible_mask, we'd better iterate over all bits in a mask and
> > > check whether the corresponding CPU is possible or not.
> > > 
> > > So I made this RFC, I did a simple build/boot/rcutorture test on my box with
> > > SMP=4, nothing bad happens. Currently I'm waiting for the 0day and trying to
> > > test this one a bigger system, in the meanwhile, looking forwards to any
> > > comment and suggestion.
> > > 
> > > So thoughts?
> > 
> > By analogy with for_each_cpu() and for_each_possible_cpu(), the name
> > should instead be for_each_leaf_node_cpu(), the tradition of excessively
> > long names in RCU notwithstanding.  ;-)
> > 
> 
> Make sense ;-)
> 
> I think it's more appropriate to call it for_each_leaf_node_mask_cpu(),
> because we don't iterate all cpus of a leaf node. The word "possible"
> could be dropped because obviously we won't iterate over "impossible"
> cpus in a leaf node ;-)

C'mon, Boqun!  The for_each_leaf_node_cpu() is not only consistent
with the for_each_cpu() family, it is shorter!  ;-)

							Thanx, Paul

> Will modify that in next version.
> 
> Regards,
> Boqun
> 
> > 							Thanx, Paul
> > 

[toc] | [prev] | [next] | [standalone]

#1539833 — Re: [RFC 0/5] rcu: Introduce leaf_node_for_each_mask_possible_cpu() and its friend

[Multipart message — attachments visible in raw view] — view raw

On Fri, Dec 09, 2016 at 08:28:05PM -0800, Paul E. McKenney wrote:
> On Sat, Dec 10, 2016 at 08:45:38AM +0800, Boqun Feng wrote:
> > On Fri, Dec 09, 2016 at 03:49:45PM -0800, Paul E. McKenney wrote:
> > > On Fri, Dec 09, 2016 at 04:48:22PM +0800, Boqun Feng wrote:
> > > > Hi Paul,
> > > > 
> > > > While reading the discussion at:
> > > > 
> > > > https://marc.info/?l=linux-kernel&m=148044253400769
> > > 
> > > This discussion was for stalls specifically, rather than for routine
> > > scans of the bitmasks.
> > > 
> > > But it does look to save some code, so worth looking into.
> > > 
> > > > I figured we might use this fact to save some extra checks in RCU core code,
> > > > currently we iterate over all the possible CPUs on a leaf node, check whether
> > > > they were masked in a certain mask and do something. However, given the fact
> > > > that the masks on a leaf node should always be sparse than the corresponding
> > > > part of cpu_possible_mask, we'd better iterate over all bits in a mask and
> > > > check whether the corresponding CPU is possible or not.
> > > > 
> > > > So I made this RFC, I did a simple build/boot/rcutorture test on my box with
> > > > SMP=4, nothing bad happens. Currently I'm waiting for the 0day and trying to
> > > > test this one a bigger system, in the meanwhile, looking forwards to any
> > > > comment and suggestion.
> > > > 
> > > > So thoughts?
> > > 
> > > By analogy with for_each_cpu() and for_each_possible_cpu(), the name
> > > should instead be for_each_leaf_node_cpu(), the tradition of excessively
> > > long names in RCU notwithstanding.  ;-)
> > > 
> > 
> > Make sense ;-)
> > 
> > I think it's more appropriate to call it for_each_leaf_node_mask_cpu(),
> > because we don't iterate all cpus of a leaf node. The word "possible"
> > could be dropped because obviously we won't iterate over "impossible"
> > cpus in a leaf node ;-)
> 
> C'mon, Boqun!  The for_each_leaf_node_cpu() is not only consistent
> with the for_each_cpu() family, it is shorter!  ;-)
> 

Sure ;-) But for_each_leaf_node_cpu() seems like an operation that
iterates over _all_ cpus in a leaf node, but I actually implement it as
an operation that iterates only the _masked_ cpus. So I feel like word
"mask" better be added in the name.

If we call it for_each_leaf_node_cpu(rnp, mask,...), we will rely on the
hope that readers could figure it out what the primitive actually does
by the indication of the parameter @mask.

I like shorter names too, but not sure whether putting "mask" in the
name is better. After all, naming is one of the most difficult
challenges in programming ;-)

Regards,
Boqun

> 							Thanx, Paul
> 
> > Will modify that in next version.
> > 
> > Regards,
> > Boqun
> > 
> > > 							Thanx, Paul
> > > 
> 
> 

[toc] | [prev] | [next] | [standalone]

#1539867 — Re: [RFC 0/5] rcu: Introduce leaf_node_for_each_mask_possible_cpu() and its friend

On Sat, Dec 10, 2016 at 09:36:29PM +0800, Boqun Feng wrote:
> On Fri, Dec 09, 2016 at 08:28:05PM -0800, Paul E. McKenney wrote:
> > On Sat, Dec 10, 2016 at 08:45:38AM +0800, Boqun Feng wrote:
> > > On Fri, Dec 09, 2016 at 03:49:45PM -0800, Paul E. McKenney wrote:
> > > > On Fri, Dec 09, 2016 at 04:48:22PM +0800, Boqun Feng wrote:
> > > > > Hi Paul,
> > > > > 
> > > > > While reading the discussion at:
> > > > > 
> > > > > https://marc.info/?l=linux-kernel&m=148044253400769
> > > > 
> > > > This discussion was for stalls specifically, rather than for routine
> > > > scans of the bitmasks.
> > > > 
> > > > But it does look to save some code, so worth looking into.
> > > > 
> > > > > I figured we might use this fact to save some extra checks in RCU core code,
> > > > > currently we iterate over all the possible CPUs on a leaf node, check whether
> > > > > they were masked in a certain mask and do something. However, given the fact
> > > > > that the masks on a leaf node should always be sparse than the corresponding
> > > > > part of cpu_possible_mask, we'd better iterate over all bits in a mask and
> > > > > check whether the corresponding CPU is possible or not.
> > > > > 
> > > > > So I made this RFC, I did a simple build/boot/rcutorture test on my box with
> > > > > SMP=4, nothing bad happens. Currently I'm waiting for the 0day and trying to
> > > > > test this one a bigger system, in the meanwhile, looking forwards to any
> > > > > comment and suggestion.
> > > > > 
> > > > > So thoughts?
> > > > 
> > > > By analogy with for_each_cpu() and for_each_possible_cpu(), the name
> > > > should instead be for_each_leaf_node_cpu(), the tradition of excessively
> > > > long names in RCU notwithstanding.  ;-)
> > > > 
> > > 
> > > Make sense ;-)
> > > 
> > > I think it's more appropriate to call it for_each_leaf_node_mask_cpu(),
> > > because we don't iterate all cpus of a leaf node. The word "possible"
> > > could be dropped because obviously we won't iterate over "impossible"
> > > cpus in a leaf node ;-)
> > 
> > C'mon, Boqun!  The for_each_leaf_node_cpu() is not only consistent
> > with the for_each_cpu() family, it is shorter!  ;-)
> 
> Sure ;-) But for_each_leaf_node_cpu() seems like an operation that
> iterates over _all_ cpus in a leaf node, but I actually implement it as
> an operation that iterates only the _masked_ cpus. So I feel like word
> "mask" better be added in the name.

Although that is a fair point, the same can be said of for_each_cpu().
Which people seem to be able to use without undue pain.

> If we call it for_each_leaf_node_cpu(rnp, mask,...), we will rely on the
> hope that readers could figure it out what the primitive actually does
> by the indication of the parameter @mask.
> 
> I like shorter names too, but not sure whether putting "mask" in the
> name is better. After all, naming is one of the most difficult
> challenges in programming ;-)

The two most difficult challenges in programming are the last two hard
things that the person speaking worked on.  ;-)

Consistency is more important than the stand-alone understanding of
this particular name.  You can always add a comment pointing out that
it follows for_each_cpu().

							Thanx, Paul

> Regards,
> Boqun
> 
> > 							Thanx, Paul
> > 
> > > Will modify that in next version.
> > > 
> > > Regards,
> > > Boqun
> > > 
> > > > 							Thanx, Paul
> > > > 
> > 
> > 

[toc] | [prev] | [next] | [standalone]

#1539931 — Re: [RFC 0/5] rcu: Introduce leaf_node_for_each_mask_possible_cpu() and its friend

[Multipart message — attachments visible in raw view] — view raw

On Sat, Dec 10, 2016 at 09:38:54AM -0800, Paul E. McKenney wrote:
> On Sat, Dec 10, 2016 at 09:36:29PM +0800, Boqun Feng wrote:
> > On Fri, Dec 09, 2016 at 08:28:05PM -0800, Paul E. McKenney wrote:
> > > On Sat, Dec 10, 2016 at 08:45:38AM +0800, Boqun Feng wrote:
> > > > On Fri, Dec 09, 2016 at 03:49:45PM -0800, Paul E. McKenney wrote:
> > > > > On Fri, Dec 09, 2016 at 04:48:22PM +0800, Boqun Feng wrote:
> > > > > > Hi Paul,
> > > > > > 
> > > > > > While reading the discussion at:
> > > > > > 
> > > > > > https://marc.info/?l=linux-kernel&m=148044253400769
> > > > > 
> > > > > This discussion was for stalls specifically, rather than for routine
> > > > > scans of the bitmasks.
> > > > > 
> > > > > But it does look to save some code, so worth looking into.
> > > > > 
> > > > > > I figured we might use this fact to save some extra checks in RCU core code,
> > > > > > currently we iterate over all the possible CPUs on a leaf node, check whether
> > > > > > they were masked in a certain mask and do something. However, given the fact
> > > > > > that the masks on a leaf node should always be sparse than the corresponding
> > > > > > part of cpu_possible_mask, we'd better iterate over all bits in a mask and
> > > > > > check whether the corresponding CPU is possible or not.
> > > > > > 
> > > > > > So I made this RFC, I did a simple build/boot/rcutorture test on my box with
> > > > > > SMP=4, nothing bad happens. Currently I'm waiting for the 0day and trying to
> > > > > > test this one a bigger system, in the meanwhile, looking forwards to any
> > > > > > comment and suggestion.
> > > > > > 
> > > > > > So thoughts?
> > > > > 
> > > > > By analogy with for_each_cpu() and for_each_possible_cpu(), the name
> > > > > should instead be for_each_leaf_node_cpu(), the tradition of excessively
> > > > > long names in RCU notwithstanding.  ;-)
> > > > > 
> > > > 
> > > > Make sense ;-)
> > > > 
> > > > I think it's more appropriate to call it for_each_leaf_node_mask_cpu(),
> > > > because we don't iterate all cpus of a leaf node. The word "possible"
> > > > could be dropped because obviously we won't iterate over "impossible"
> > > > cpus in a leaf node ;-)
> > > 
> > > C'mon, Boqun!  The for_each_leaf_node_cpu() is not only consistent
> > > with the for_each_cpu() family, it is shorter!  ;-)
> > 
> > Sure ;-) But for_each_leaf_node_cpu() seems like an operation that
> > iterates over _all_ cpus in a leaf node, but I actually implement it as
> > an operation that iterates only the _masked_ cpus. So I feel like word
> > "mask" better be added in the name.
> 
> Although that is a fair point, the same can be said of for_each_cpu().
> Which people seem to be able to use without undue pain.
> 
> > If we call it for_each_leaf_node_cpu(rnp, mask,...), we will rely on the
> > hope that readers could figure it out what the primitive actually does
> > by the indication of the parameter @mask.
> > 
> > I like shorter names too, but not sure whether putting "mask" in the
> > name is better. After all, naming is one of the most difficult
> > challenges in programming ;-)
> 
> The two most difficult challenges in programming are the last two hard
> things that the person speaking worked on.  ;-)
> 

;-)

> Consistency is more important than the stand-alone understanding of
> this particular name.  You can always add a comment pointing out that
> it follows for_each_cpu().
> 

Fair enough. Let us name it for_each_leaf_node_cpu() ;-)

Regards,
Boqun

> 							Thanx, Paul
> 
> > Regards,
> > Boqun
> > 
> > > 							Thanx, Paul
> > > 
> > > > Will modify that in next version.
> > > > 
> > > > Regards,
> > > > Boqun
> > > > 
> > > > > 							Thanx, Paul
> > > > > 
> > > 
> > > 
> 
> 

[toc] | [prev] | [standalone]

Back to top | Article view | linux.kernel

csiph-web