On January 30, 2020 12:33:41 PM GMT+01:00, Oleg Nesterov <oleg@xxxxxxxxxx> wrote:
>On 01/30, Christian Brauner wrote:
>>
>> On Thu, Jan 30, 2020 at 11:50:28AM +0530,
>madhuparnabhowmik10@xxxxxxxxx wrote:
>> > From: Madhuparna Bhowmik <madhuparnabhowmik10@xxxxxxxxx>
>> >
>> > This patch fixes the following sparse error:
>> > kernel/exit.c:627:25: error: incompatible types in comparison
>expression
>> >
>> > And the following warning:
>> > kernel/exit.c:626:40: warning: incorrect type in assignment
>> >
>> > Signed-off-by: Madhuparna Bhowmik <madhuparnabhowmik10@xxxxxxxxx>
>>
>> I think the previous version was already fine but hopefully
>> RCU_INIT_POINTER() really saves some overhead. In any case:
>
>It is not about overhead, RCU_INIT_POINTER() documents the fact that we
>didn't make any changes to the new parent, we only need to change the
>pointer.
Right, I wasn't complaining. RCU_INIT_POINTER() claims that it has less overhead than rcu_assign_pointer().
So that is an additional argument for it.
>
>And btw, I don't really understand the __rcu annotations. Say,
>according
>to sparse this code is wrong:
>
> int __rcu *P;
>
> void func(int *p)
> {
> P = p;
> }
>
>OK, although quite possibly it is fine.
>
>However, this code
>
> int __rcu *P;
>
> void func(int __rcu *p)
> {
> *p = 10;
> P = p;
> }
>
>is almost certainly wrong but sparse is happy, asn is the same.
That's more an argument to fix sparse I guess?
The annotations themselves are rather useful I think.
They at least help me when reading the code.
It's not that rcu lifetimes are trivial and anything that helps remind me that an object wants rcu semantics I'm happy to take it. :)
>
>
>> Acked-by: Christian Brauner <christian.brauner@xxxxxxxxxx>
>
>Acked-by: Oleg Nesterov <oleg@xxxxxxxxxx>
