[PATCH] s390/pkey: Remove synchronize_rcu from pkey_handler_register

[PATCH] s390/pkey: Remove synchronize_rcu from pkey_handler_register

[lirongqing]

From: Li RongQing <lirongqing@baidu.com>

The synchronize_rcu() call after adding a handler to the handler_list
is redundant because RCU readers will either see the old list or the
new list. Removing this synchronization point reduces the blocking
time during handler registration.

Signed-off-by: Li RongQing <lirongqing@baidu.com>
---
 drivers/s390/crypto/pkey_base.c | 1 -
 1 file changed, 1 deletion(-)

diff --git a/drivers/s390/crypto/pkey_base.c b/drivers/s390/crypto/pkey_base.c
index d60cd98..c2e29d6 100644
--- a/drivers/s390/crypto/pkey_base.c
+++ b/drivers/s390/crypto/pkey_base.c
@@ -60,7 +60,6 @@ int pkey_handler_register(struct pkey_handler *handler)
 
 	list_add_rcu(&handler->list, &handler_list);
 	spin_unlock(&handler_list_write_lock);
-	synchronize_rcu();
 
 	module_put(handler->module);
 
-- 
2.9.4

Re: [PATCH] s390/pkey: Remove synchronize_rcu from pkey_handler_register

[Harald Freudenberger]

On 2026-03-13 06:23, lirongqing wrote:
> From: Li RongQing <lirongqing@baidu.com>
> 
> The synchronize_rcu() call after adding a handler to the handler_list
> is redundant because RCU readers will either see the old list or the
> new list. Removing this synchronization point reduces the blocking
> time during handler registration.
> 
> Signed-off-by: Li RongQing <lirongqing@baidu.com>
> ---
>  drivers/s390/crypto/pkey_base.c | 1 -
>  1 file changed, 1 deletion(-)
> 
> diff --git a/drivers/s390/crypto/pkey_base.c 
> b/drivers/s390/crypto/pkey_base.c
> index d60cd98..c2e29d6 100644
> --- a/drivers/s390/crypto/pkey_base.c
> +++ b/drivers/s390/crypto/pkey_base.c
> @@ -60,7 +60,6 @@ int pkey_handler_register(struct pkey_handler 
> *handler)
> 
>  	list_add_rcu(&handler->list, &handler_list);
>  	spin_unlock(&handler_list_write_lock);
> -	synchronize_rcu();
> 
>  	module_put(handler->module);

Thanks Li RongQing

Reviewed-by: Harald Freudenberger <freude@linux.ibm.com>

@Heiko/@Vasily/@Alexander will you pick this?

Re: [PATCH] s390/pkey: Remove synchronize_rcu from pkey_handler_register

[Heiko Carstens]

On Fri, Mar 13, 2026 at 09:32:39AM +0100, Harald Freudenberger wrote:
> On 2026-03-13 06:23, lirongqing wrote:
> > From: Li RongQing <lirongqing@baidu.com>
> > 
> > The synchronize_rcu() call after adding a handler to the handler_list
> > is redundant because RCU readers will either see the old list or the
> > new list. Removing this synchronization point reduces the blocking
> > time during handler registration.
> > 
> > Signed-off-by: Li RongQing <lirongqing@baidu.com>
> > ---
> >  drivers/s390/crypto/pkey_base.c | 1 -
> >  1 file changed, 1 deletion(-)
> > 
> > diff --git a/drivers/s390/crypto/pkey_base.c
> > b/drivers/s390/crypto/pkey_base.c
> > index d60cd98..c2e29d6 100644
> > --- a/drivers/s390/crypto/pkey_base.c
> > +++ b/drivers/s390/crypto/pkey_base.c
> > @@ -60,7 +60,6 @@ int pkey_handler_register(struct pkey_handler
> > *handler)
> > 
> >  	list_add_rcu(&handler->list, &handler_list);
> >  	spin_unlock(&handler_list_write_lock);
> > -	synchronize_rcu();
> > 
> >  	module_put(handler->module);
> 
> Thanks Li RongQing
> 
> Reviewed-by: Harald Freudenberger <freude@linux.ibm.com>

With synchronize_rcu() you do know that other CPUs will see the new
list after this function returned, without not. Is that ok?

Re: [PATCH] s390/pkey: Remove synchronize_rcu from pkey_handler_register

[Harald Freudenberger]

On 2026-03-13 10:06, Heiko Carstens wrote:
> On Fri, Mar 13, 2026 at 09:32:39AM +0100, Harald Freudenberger wrote:
>> On 2026-03-13 06:23, lirongqing wrote:
>> > From: Li RongQing <lirongqing@baidu.com>
>> >
>> > The synchronize_rcu() call after adding a handler to the handler_list
>> > is redundant because RCU readers will either see the old list or the
>> > new list. Removing this synchronization point reduces the blocking
>> > time during handler registration.
>> >
>> > Signed-off-by: Li RongQing <lirongqing@baidu.com>
>> > ---
>> >  drivers/s390/crypto/pkey_base.c | 1 -
>> >  1 file changed, 1 deletion(-)
>> >
>> > diff --git a/drivers/s390/crypto/pkey_base.c
>> > b/drivers/s390/crypto/pkey_base.c
>> > index d60cd98..c2e29d6 100644
>> > --- a/drivers/s390/crypto/pkey_base.c
>> > +++ b/drivers/s390/crypto/pkey_base.c
>> > @@ -60,7 +60,6 @@ int pkey_handler_register(struct pkey_handler
>> > *handler)
>> >
>> >  	list_add_rcu(&handler->list, &handler_list);
>> >  	spin_unlock(&handler_list_write_lock);
>> > -	synchronize_rcu();
>> >
>> >  	module_put(handler->module);
>> 
>> Thanks Li RongQing
>> 
>> Reviewed-by: Harald Freudenberger <freude@linux.ibm.com>
> 
> With synchronize_rcu() you do know that other CPUs will see the new
> list after this function returned, without not. Is that ok?

No - If this has side effects then I'll discuss this with you and
then may also run some tests - so please hold back until I'll give
a green light on this. Thanks for this hint.

Re: [PATCH] s390/pkey: Remove synchronize_rcu from pkey_handler_register

[Harald Freudenberger]

On 2026-03-13 06:23, lirongqing wrote:
> From: Li RongQing <lirongqing@baidu.com>
> 
> The synchronize_rcu() call after adding a handler to the handler_list
> is redundant because RCU readers will either see the old list or the
> new list. Removing this synchronization point reduces the blocking
> time during handler registration.
> 
> Signed-off-by: Li RongQing <lirongqing@baidu.com>
> ---
>  drivers/s390/crypto/pkey_base.c | 1 -
>  1 file changed, 1 deletion(-)
> 
> diff --git a/drivers/s390/crypto/pkey_base.c 
> b/drivers/s390/crypto/pkey_base.c
> index d60cd98..c2e29d6 100644
> --- a/drivers/s390/crypto/pkey_base.c
> +++ b/drivers/s390/crypto/pkey_base.c
> @@ -60,7 +60,6 @@ int pkey_handler_register(struct pkey_handler 
> *handler)
> 
>  	list_add_rcu(&handler->list, &handler_list);
>  	spin_unlock(&handler_list_write_lock);
> -	synchronize_rcu();
> 
>  	module_put(handler->module);

After some deep dive into RCU, and some code review with Holger we
decided to leave this statement in the code: This is the fast path to
push the info about the updated list to the other cpus. If removed, the
other cpus may get the updated list when the RCU context is synched. As
this code is in general not performance critical and the list update
more or less only occurs at the early time in system startup the focus
is on concurrency versus performance. Maybe I'll add a comment onto
this code line...

Thanks Li RongQing for your patch and suggestions.

Harald Freudenberger

答复: [????] Re: [PATCH] s390/pkey: Remove synchronize_rcu from pkey_handler_register

[Li,Rongqing(ACG CCN)]

> After some deep dive into RCU, and some code review with Holger we decided to
> leave this statement in the code: This is the fast path to push the info about the
> updated list to the other cpus. If removed, the other cpus may get the updated
> list when the RCU context is synched. As this code is in general not performance
> critical and the list update more or less only occurs at the early time in system
> startup the focus is on concurrency versus performance. Maybe I'll add a
> comment onto this code line...
> 

Thanks for explanation, I think a comment is needed
sorry for noise.

[Li,Rongqing]