question on pktgen shutdown logic.

question on pktgen shutdown logic.

[Ben Greear]

While looking at pktgen.c in 3.9.3, I cannot explain how
it does not race on shutdown.

The thread worker checks for pktgen_exiting, and will bail out
of it's main loop and initiate cleanup if this is true.

The pg_net_exit sets the pktgen_exiting to true, and then
proceedes to call

		kthread_stop(t->tsk);
		kfree(t);

What I don't understand is what keeps the kfree(t) from happening
before the thread has noticed pktgen_exiting is set and properly
bailed out of it's loop and finished cleanup?

Thanks,
Ben

-- 
Ben Greear <greearb@candelatech.com>
Candela Technologies Inc  http://www.candelatech.com

Re: question on pktgen shutdown logic.

[Eric Dumazet]

On Wed, 2013-05-22 at 16:15 -0700, Ben Greear wrote:
> While looking at pktgen.c in 3.9.3, I cannot explain how
> it does not race on shutdown.
> 
> The thread worker checks for pktgen_exiting, and will bail out
> of it's main loop and initiate cleanup if this is true.
> 
> The pg_net_exit sets the pktgen_exiting to true, and then
> proceedes to call
> 
> 		kthread_stop(t->tsk);
> 		kfree(t);
> 
> What I don't understand is what keeps the kfree(t) from happening
> before the thread has noticed pktgen_exiting is set and properly
> bailed out of it's loop and finished cleanup?

kthread_stop() is synchronous.

Re: question on pktgen shutdown logic.

[Ben Greear]

On 05/22/2013 04:22 PM, Eric Dumazet wrote:
> On Wed, 2013-05-22 at 16:15 -0700, Ben Greear wrote:
>> While looking at pktgen.c in 3.9.3, I cannot explain how
>> it does not race on shutdown.
>>
>> The thread worker checks for pktgen_exiting, and will bail out
>> of it's main loop and initiate cleanup if this is true.
>>
>> The pg_net_exit sets the pktgen_exiting to true, and then
>> proceedes to call
>>
>> 		kthread_stop(t->tsk);
>> 		kfree(t);
>>
>> What I don't understand is what keeps the kfree(t) from happening
>> before the thread has noticed pktgen_exiting is set and properly
>> bailed out of it's loop and finished cleanup?
>
> kthread_stop() is synchronous.

Ahh, so it waits for the pktgen_thread_worker method to return?

Why do we bother with the pktgen_exiting flag at all...it
would seem that just checking kthread_should_stop() would
do the trick?

Thanks,
Ben

-- 
Ben Greear <greearb@candelatech.com>
Candela Technologies Inc  http://www.candelatech.com

Re: question on pktgen shutdown logic.

[Eric Dumazet]

On Wed, 2013-05-22 at 16:28 -0700, Ben Greear wrote:

> Ahh, so it waits for the pktgen_thread_worker method to return?
> 
> Why do we bother with the pktgen_exiting flag at all...it
> would seem that just checking kthread_should_stop() would
> do the trick?

If you have 4096 cpus, setting pktgen_exiting permits all pktgen exiting
faster.

Otherwise you wait about 400 seconds (4096 * HZ/10)

Re: question on pktgen shutdown logic.

[Ben Greear]

On 05/22/2013 04:35 PM, Eric Dumazet wrote:
> On Wed, 2013-05-22 at 16:28 -0700, Ben Greear wrote:
>
>> Ahh, so it waits for the pktgen_thread_worker method to return?
>>
>> Why do we bother with the pktgen_exiting flag at all...it
>> would seem that just checking kthread_should_stop() would
>> do the trick?
>
> If you have 4096 cpus, setting pktgen_exiting permits all pktgen exiting
> faster.
>
> Otherwise you wait about 400 seconds (4096 * HZ/10)

Ok, I think I understand now.

Thanks,
Ben

>
>


-- 
Ben Greear <greearb@candelatech.com>
Candela Technologies Inc  http://www.candelatech.com