Re: [RFC 2/4] bpf, security: Add Checmate

On Thu, Aug 04, 2016 at 05:34:32PM +0800, zhuyj wrote:

 Sure.
Is it better to add
#ifndef CONFIG_PREEMPT_RCU ?

On Thu, Aug 4, 2016 at 4:28 PM, Eric Dumazet [off-list ref] wrote:

quoted

Please do not top post

On Thu, 2016-08-04 at 16:08 +0800, zhuyj wrote:

quoted

 +void register_checmate_prog_ops(void);
maybe it is extern void register_checmate_prog_ops(void);?

+       preempt_disable();
+       rcu_read_lock();
IMHO, it is not necessary to use the above 2 since rcu_read_lock will
call preempt_disable.

You might double check if this claim is true if CONFIG_PREEMPT_RCU=y

Thanks for your feedback zhuyj, Looking at kernel documentation itself, it looks 
like this is the preferred mechanism[1]. Their example:

 1 preempt_disable();
 2 rcu_read_lock();
 3 do_something();
 4 rcu_read_unlock();
 5 preempt_enable();

But, I think you're right. Do you know if there's a great benefit of doing this? 
Or does it make sense to implement a new macro, a la 
rcu_read_lock_and_preent_disable()?

[1] https://www.kernel.org/doc/Documentation/RCU/Design/Requirements/Requirements.html#Disabling Preemption Does Not Block Grace Periods