1. Patchew
  2. linux
  3. View series

[PATCH v2] srcu/tiny: Remove preempt_disable/enable() in srcu_gp_start_if_needed()

Zqiang posted 1 patch 5 months ago
There is a newer version of this series
kernel/rcu/srcutiny.c | 4 +---
1 file changed, 1 insertion(+), 3 deletions(-)
[PATCH v2] srcu/tiny: Remove preempt_disable/enable() in srcu_gp_start_if_needed()
Posted by Zqiang 5 months ago 
Currently, the srcu_gp_start_if_needed() is always be invoked in
preempt disable's critical section, this commit therefore remove
redundant preempt_disable/enable() in srcu_gp_start_if_needed()
and adds a call to lockdep_assert_preemption_disabled() in order
to enable lockdep to diagnose mistaken invocations of this function
from preempts-enabled code.

Fixes: 65b4a59557f6 ("srcu: Make Tiny SRCU explicitly disable preemption")
Signed-off-by: Zqiang <qiang.zhang@linux.dev>
---
 kernel/rcu/srcutiny.c | 4 +---
 1 file changed, 1 insertion(+), 3 deletions(-)

diff --git a/kernel/rcu/srcutiny.c b/kernel/rcu/srcutiny.c
index b52ec45698e8..b2da188133fc 100644
--- a/kernel/rcu/srcutiny.c
+++ b/kernel/rcu/srcutiny.c
@@ -181,10 +181,9 @@ static void srcu_gp_start_if_needed(struct srcu_struct *ssp)
 {
 	unsigned long cookie;
 
-	preempt_disable();  // Needed for PREEMPT_LAZY
+	lockdep_assert_preemption_disabled();
 	cookie = get_state_synchronize_srcu(ssp);
 	if (ULONG_CMP_GE(READ_ONCE(ssp->srcu_idx_max), cookie)) {
-		preempt_enable();
 		return;
 	}
 	WRITE_ONCE(ssp->srcu_idx_max, cookie);
@@ -194,7 +193,6 @@ static void srcu_gp_start_if_needed(struct srcu_struct *ssp)
 		else if (list_empty(&ssp->srcu_work.entry))
 			list_add(&ssp->srcu_work.entry, &srcu_boot_list);
 	}
-	preempt_enable();
 }
 
 /*
-- 
2.48.1
Re: [PATCH v2] srcu/tiny: Remove preempt_disable/enable() in srcu_gp_start_if_needed()
Posted by Paul E. McKenney 5 months ago 
On Tue, Sep 09, 2025 at 09:39:28PM +0800, Zqiang wrote:
> Currently, the srcu_gp_start_if_needed() is always be invoked in
> preempt disable's critical section, this commit therefore remove
> redundant preempt_disable/enable() in srcu_gp_start_if_needed()
> and adds a call to lockdep_assert_preemption_disabled() in order
> to enable lockdep to diagnose mistaken invocations of this function
> from preempts-enabled code.
> 
> Fixes: 65b4a59557f6 ("srcu: Make Tiny SRCU explicitly disable preemption")
> Signed-off-by: Zqiang <qiang.zhang@linux.dev>

Very good, applied for testing and further review, thank you!

							Thanx, Paul

> ---
>  kernel/rcu/srcutiny.c | 4 +---
>  1 file changed, 1 insertion(+), 3 deletions(-)
> 
> diff --git a/kernel/rcu/srcutiny.c b/kernel/rcu/srcutiny.c
> index b52ec45698e8..b2da188133fc 100644
> --- a/kernel/rcu/srcutiny.c
> +++ b/kernel/rcu/srcutiny.c
> @@ -181,10 +181,9 @@ static void srcu_gp_start_if_needed(struct srcu_struct *ssp)
>  {
>  	unsigned long cookie;
>  
> -	preempt_disable();  // Needed for PREEMPT_LAZY
> +	lockdep_assert_preemption_disabled();
>  	cookie = get_state_synchronize_srcu(ssp);
>  	if (ULONG_CMP_GE(READ_ONCE(ssp->srcu_idx_max), cookie)) {
> -		preempt_enable();
>  		return;
>  	}
>  	WRITE_ONCE(ssp->srcu_idx_max, cookie);
> @@ -194,7 +193,6 @@ static void srcu_gp_start_if_needed(struct srcu_struct *ssp)
>  		else if (list_empty(&ssp->srcu_work.entry))
>  			list_add(&ssp->srcu_work.entry, &srcu_boot_list);
>  	}
> -	preempt_enable();
>  }
>  
>  /*
> -- 
> 2.48.1
>
Re: [PATCH v2] srcu/tiny: Remove preempt_disable/enable() in srcu_gp_start_if_needed()
Posted by Joel Fernandes 5 months ago 
[..]
> >  kernel/rcu/srcutiny.c | 4 +---
> >  1 file changed, 1 insertion(+), 3 deletions(-)
> > 
> > diff --git a/kernel/rcu/srcutiny.c b/kernel/rcu/srcutiny.c
> > index b52ec45698e8..b2da188133fc 100644
> > --- a/kernel/rcu/srcutiny.c
> > +++ b/kernel/rcu/srcutiny.c
> > @@ -181,10 +181,9 @@ static void srcu_gp_start_if_needed(struct srcu_struct *ssp)
> >  {
> >  	unsigned long cookie;
> >  
> > -	preempt_disable();  // Needed for PREEMPT_LAZY
> > +	lockdep_assert_preemption_disabled();

nit: Do we still want to keep the comment that the expectation of preemption
being disabled is for the LAZY case?

thanks,

 - Joel


> >  	cookie = get_state_synchronize_srcu(ssp);
> >  	if (ULONG_CMP_GE(READ_ONCE(ssp->srcu_idx_max), cookie)) {
> > -		preempt_enable();
> >  		return;
> >  	}
> >  	WRITE_ONCE(ssp->srcu_idx_max, cookie);
> > @@ -194,7 +193,6 @@ static void srcu_gp_start_if_needed(struct srcu_struct *ssp)
> >  		else if (list_empty(&ssp->srcu_work.entry))
> >  			list_add(&ssp->srcu_work.entry, &srcu_boot_list);
> >  	}
> > -	preempt_enable();
> >  }
> >  
> >  /*
> > -- 
> > 2.48.1
> >
Re: [PATCH v2] srcu/tiny: Remove preempt_disable/enable() in srcu_gp_start_if_needed()
Posted by Paul E. McKenney 5 months ago 
On Wed, Sep 10, 2025 at 10:36:20AM -0400, Joel Fernandes wrote:
> [..]
> > >  kernel/rcu/srcutiny.c | 4 +---
> > >  1 file changed, 1 insertion(+), 3 deletions(-)
> > > 
> > > diff --git a/kernel/rcu/srcutiny.c b/kernel/rcu/srcutiny.c
> > > index b52ec45698e8..b2da188133fc 100644
> > > --- a/kernel/rcu/srcutiny.c
> > > +++ b/kernel/rcu/srcutiny.c
> > > @@ -181,10 +181,9 @@ static void srcu_gp_start_if_needed(struct srcu_struct *ssp)
> > >  {
> > >  	unsigned long cookie;
> > >  
> > > -	preempt_disable();  // Needed for PREEMPT_LAZY
> > > +	lockdep_assert_preemption_disabled();
> 
> nit: Do we still want to keep the comment that the expectation of preemption
> being disabled is for the LAZY case?

Good point, and I do believe that we do.  Zqiang, any reason not to
add this comment back in?

							Thanx, Paul

> thanks,
> 
>  - Joel
> 
> 
> > >  	cookie = get_state_synchronize_srcu(ssp);
> > >  	if (ULONG_CMP_GE(READ_ONCE(ssp->srcu_idx_max), cookie)) {
> > > -		preempt_enable();
> > >  		return;
> > >  	}
> > >  	WRITE_ONCE(ssp->srcu_idx_max, cookie);
> > > @@ -194,7 +193,6 @@ static void srcu_gp_start_if_needed(struct srcu_struct *ssp)
> > >  		else if (list_empty(&ssp->srcu_work.entry))
> > >  			list_add(&ssp->srcu_work.entry, &srcu_boot_list);
> > >  	}
> > > -	preempt_enable();
> > >  }
> > >  
> > >  /*
> > > -- 
> > > 2.48.1
> > >
Re: [PATCH v2] srcu/tiny: Remove preempt_disable/enable() in srcu_gp_start_if_needed()
Posted by Zqiang 4 months, 4 weeks ago 
> 
> On Wed, Sep 10, 2025 at 10:36:20AM -0400, Joel Fernandes wrote:
> 
> > 
> > [..]
> >  > kernel/rcu/srcutiny.c | 4 +---
> >  > 1 file changed, 1 insertion(+), 3 deletions(-)
> >  > 
> >  > diff --git a/kernel/rcu/srcutiny.c b/kernel/rcu/srcutiny.c
> >  > index b52ec45698e8..b2da188133fc 100644
> >  > --- a/kernel/rcu/srcutiny.c
> >  > +++ b/kernel/rcu/srcutiny.c
> >  > @@ -181,10 +181,9 @@ static void srcu_gp_start_if_needed(struct srcu_struct *ssp)
> >  > {
> >  > unsigned long cookie;
> >  > 
> >  > - preempt_disable(); // Needed for PREEMPT_LAZY
> >  > + lockdep_assert_preemption_disabled();
> >  
> >  nit: Do we still want to keep the comment that the expectation of preemption
> >  being disabled is for the LAZY case?
> > 
> Good point, and I do believe that we do. Zqiang, any reason not to
> add this comment back in?

in rcu-tree, this commit:

(935147775c977 "EXP srcu: Enable Tiny SRCU On all CONFIG_SMP=n kernels")

make preempt disable needed for CONFIG_PREEMPT=y or CONFIG_PREEMPT_LAZY=y
when the CONFIG_SMP=n. do we need to replace "Needed for PREEMPT_LAZY"
comments with "Needed for PREEMPT or PREEMPT_LAZY"?

Thanks
Zqiang


> 
>  Thanx, Paul
> 
> > 
> > thanks,
> >  
> >  - Joel
> >  
> >  
> >  > cookie = get_state_synchronize_srcu(ssp);
> >  > if (ULONG_CMP_GE(READ_ONCE(ssp->srcu_idx_max), cookie)) {
> >  > - preempt_enable();
> >  > return;
> >  > }
> >  > WRITE_ONCE(ssp->srcu_idx_max, cookie);
> >  > @@ -194,7 +193,6 @@ static void srcu_gp_start_if_needed(struct srcu_struct *ssp)
> >  > else if (list_empty(&ssp->srcu_work.entry))
> >  > list_add(&ssp->srcu_work.entry, &srcu_boot_list);
> >  > }
> >  > - preempt_enable();
> >  > }
> >  > 
> >  > /*
> >  > -- 
> >  > 2.48.1
> >  >
> >
>
Re: [PATCH v2] srcu/tiny: Remove preempt_disable/enable() in srcu_gp_start_if_needed()
Posted by Paul E. McKenney 4 months, 4 weeks ago 
On Thu, Sep 11, 2025 at 12:36:45AM +0000, Zqiang wrote:
> > 
> > On Wed, Sep 10, 2025 at 10:36:20AM -0400, Joel Fernandes wrote:
> > 
> > > 
> > > [..]
> > >  > kernel/rcu/srcutiny.c | 4 +---
> > >  > 1 file changed, 1 insertion(+), 3 deletions(-)
> > >  > 
> > >  > diff --git a/kernel/rcu/srcutiny.c b/kernel/rcu/srcutiny.c
> > >  > index b52ec45698e8..b2da188133fc 100644
> > >  > --- a/kernel/rcu/srcutiny.c
> > >  > +++ b/kernel/rcu/srcutiny.c
> > >  > @@ -181,10 +181,9 @@ static void srcu_gp_start_if_needed(struct srcu_struct *ssp)
> > >  > {
> > >  > unsigned long cookie;
> > >  > 
> > >  > - preempt_disable(); // Needed for PREEMPT_LAZY
> > >  > + lockdep_assert_preemption_disabled();
> > >  
> > >  nit: Do we still want to keep the comment that the expectation of preemption
> > >  being disabled is for the LAZY case?
> > > 
> > Good point, and I do believe that we do. Zqiang, any reason not to
> > add this comment back in?
> 
> in rcu-tree, this commit:
> 
> (935147775c977 "EXP srcu: Enable Tiny SRCU On all CONFIG_SMP=n kernels")
> 
> make preempt disable needed for CONFIG_PREEMPT=y or CONFIG_PREEMPT_LAZY=y
> when the CONFIG_SMP=n. do we need to replace "Needed for PREEMPT_LAZY"
> comments with "Needed for PREEMPT or PREEMPT_LAZY"?

Good point as well, thank you!  And I need to decide whether I should
send that patch upstream.  Its original purpose was to test PREEMPT_LAZY=y
better than could be tested with PREEMPT_LAZY.

Thoughts?

							Thanx, Paul

> Thanks
> Zqiang
> 
> 
> > 
> >  Thanx, Paul
> > 
> > > 
> > > thanks,
> > >  
> > >  - Joel
> > >  
> > >  
> > >  > cookie = get_state_synchronize_srcu(ssp);
> > >  > if (ULONG_CMP_GE(READ_ONCE(ssp->srcu_idx_max), cookie)) {
> > >  > - preempt_enable();
> > >  > return;
> > >  > }
> > >  > WRITE_ONCE(ssp->srcu_idx_max, cookie);
> > >  > @@ -194,7 +193,6 @@ static void srcu_gp_start_if_needed(struct srcu_struct *ssp)
> > >  > else if (list_empty(&ssp->srcu_work.entry))
> > >  > list_add(&ssp->srcu_work.entry, &srcu_boot_list);
> > >  > }
> > >  > - preempt_enable();
> > >  > }
> > >  > 
> > >  > /*
> > >  > -- 
> > >  > 2.48.1
> > >  >
> > >
> >
Re: [PATCH v2] srcu/tiny: Remove preempt_disable/enable() in srcu_gp_start_if_needed()
Posted by Zqiang 4 months, 4 weeks ago 
> 
> On Thu, Sep 11, 2025 at 12:36:45AM +0000, Zqiang wrote:
> 
> > 
> > On Wed, Sep 10, 2025 at 10:36:20AM -0400, Joel Fernandes wrote:
> >  
> >  > 
> >  > [..]
> >  > > kernel/rcu/srcutiny.c | 4 +---
> >  > > 1 file changed, 1 insertion(+), 3 deletions(-)
> >  > > 
> >  > > diff --git a/kernel/rcu/srcutiny.c b/kernel/rcu/srcutiny.c
> >  > > index b52ec45698e8..b2da188133fc 100644
> >  > > --- a/kernel/rcu/srcutiny.c
> >  > > +++ b/kernel/rcu/srcutiny.c
> >  > > @@ -181,10 +181,9 @@ static void srcu_gp_start_if_needed(struct srcu_struct *ssp)
> >  > > {
> >  > > unsigned long cookie;
> >  > > 
> >  > > - preempt_disable(); // Needed for PREEMPT_LAZY
> >  > > + lockdep_assert_preemption_disabled();
> >  > 
> >  > nit: Do we still want to keep the comment that the expectation of preemption
> >  > being disabled is for the LAZY case?
> >  > 
> >  Good point, and I do believe that we do. Zqiang, any reason not to
> >  add this comment back in?
> >  
> >  in rcu-tree, this commit:
> >  
> >  (935147775c977 "EXP srcu: Enable Tiny SRCU On all CONFIG_SMP=n kernels")
> >  
> >  make preempt disable needed for CONFIG_PREEMPT=y or CONFIG_PREEMPT_LAZY=y
> >  when the CONFIG_SMP=n. do we need to replace "Needed for PREEMPT_LAZY"
> >  comments with "Needed for PREEMPT or PREEMPT_LAZY"?
> > 
> Good point as well, thank you! And I need to decide whether I should
> send that patch upstream. Its original purpose was to test PREEMPT_LAZY=y
> better than could be tested with PREEMPT_LAZY.
> 
> Thoughts?

I will add "Needed for PREEMPT_LAZY" comments, if this commit (935147775c977) is
send to upstream, will update comments again in the future.

Thanks
Zqiang

> 
>  Thanx, Paul
> 
> > 
> > Thanks
> >  Zqiang
> >  
> >  
> >  
> >  Thanx, Paul
> >  
> >  > 
> >  > thanks,
> >  > 
> >  > - Joel
> >  > 
> >  > 
> >  > > cookie = get_state_synchronize_srcu(ssp);
> >  > > if (ULONG_CMP_GE(READ_ONCE(ssp->srcu_idx_max), cookie)) {
> >  > > - preempt_enable();
> >  > > return;
> >  > > }
> >  > > WRITE_ONCE(ssp->srcu_idx_max, cookie);
> >  > > @@ -194,7 +193,6 @@ static void srcu_gp_start_if_needed(struct srcu_struct *ssp)
> >  > > else if (list_empty(&ssp->srcu_work.entry))
> >  > > list_add(&ssp->srcu_work.entry, &srcu_boot_list);
> >  > > }
> >  > > - preempt_enable();
> >  > > }
> >  > > 
> >  > > /*
> >  > > -- 
> >  > > 2.48.1
> >  > >
> >  >
> >
>
Re: [PATCH v2] srcu/tiny: Remove preempt_disable/enable() in srcu_gp_start_if_needed()
Posted by Paul E. McKenney 4 months, 4 weeks ago 
On Thu, Sep 11, 2025 at 11:46:36AM +0000, Zqiang wrote:
> > 
> > On Thu, Sep 11, 2025 at 12:36:45AM +0000, Zqiang wrote:
> > 
> > > 
> > > On Wed, Sep 10, 2025 at 10:36:20AM -0400, Joel Fernandes wrote:
> > >  
> > >  > 
> > >  > [..]
> > >  > > kernel/rcu/srcutiny.c | 4 +---
> > >  > > 1 file changed, 1 insertion(+), 3 deletions(-)
> > >  > > 
> > >  > > diff --git a/kernel/rcu/srcutiny.c b/kernel/rcu/srcutiny.c
> > >  > > index b52ec45698e8..b2da188133fc 100644
> > >  > > --- a/kernel/rcu/srcutiny.c
> > >  > > +++ b/kernel/rcu/srcutiny.c
> > >  > > @@ -181,10 +181,9 @@ static void srcu_gp_start_if_needed(struct srcu_struct *ssp)
> > >  > > {
> > >  > > unsigned long cookie;
> > >  > > 
> > >  > > - preempt_disable(); // Needed for PREEMPT_LAZY
> > >  > > + lockdep_assert_preemption_disabled();
> > >  > 
> > >  > nit: Do we still want to keep the comment that the expectation of preemption
> > >  > being disabled is for the LAZY case?
> > >  > 
> > >  Good point, and I do believe that we do. Zqiang, any reason not to
> > >  add this comment back in?
> > >  
> > >  in rcu-tree, this commit:
> > >  
> > >  (935147775c977 "EXP srcu: Enable Tiny SRCU On all CONFIG_SMP=n kernels")
> > >  
> > >  make preempt disable needed for CONFIG_PREEMPT=y or CONFIG_PREEMPT_LAZY=y
> > >  when the CONFIG_SMP=n. do we need to replace "Needed for PREEMPT_LAZY"
> > >  comments with "Needed for PREEMPT or PREEMPT_LAZY"?
> > > 
> > Good point as well, thank you! And I need to decide whether I should
> > send that patch upstream. Its original purpose was to test PREEMPT_LAZY=y
> > better than could be tested with PREEMPT_LAZY.
> > 
> > Thoughts?
> 
> I will add "Needed for PREEMPT_LAZY" comments, if this commit (935147775c977) is
> send to upstream, will update comments again in the future.

That sounds good to me, thank you!

							Thanx, Paul

> Thanks
> Zqiang
> 
> > 
> >  Thanx, Paul
> > 
> > > 
> > > Thanks
> > >  Zqiang
> > >  
> > >  
> > >  
> > >  Thanx, Paul
> > >  
> > >  > 
> > >  > thanks,
> > >  > 
> > >  > - Joel
> > >  > 
> > >  > 
> > >  > > cookie = get_state_synchronize_srcu(ssp);
> > >  > > if (ULONG_CMP_GE(READ_ONCE(ssp->srcu_idx_max), cookie)) {
> > >  > > - preempt_enable();
> > >  > > return;
> > >  > > }
> > >  > > WRITE_ONCE(ssp->srcu_idx_max, cookie);
> > >  > > @@ -194,7 +193,6 @@ static void srcu_gp_start_if_needed(struct srcu_struct *ssp)
> > >  > > else if (list_empty(&ssp->srcu_work.entry))
> > >  > > list_add(&ssp->srcu_work.entry, &srcu_boot_list);
> > >  > > }
> > >  > > - preempt_enable();
> > >  > > }
> > >  > > 
> > >  > > /*
> > >  > > -- 
> > >  > > 2.48.1
> > >  > >
> > >  >
> > >
> >

Patchew 2.1-devel-b67e4c2

© 2016 - 2026 Red Hat, Inc.