2005-12-05 10:57:26

by Srivatsa Vaddagiri

Subject: [PATCH] Fix bug in RCU torture test

While doing some test of RCU torture module, I hit a OOPS in rcu_do_batch,
which was trying to processes callback of a module that was just removed.
This is because we weren't waiting long enough for all callbacks to fire.

Patch below fixes that.

Signed-off-by : Srivatsa Vaddagiri <[email protected]>


---

linux-2.6.15-rc5-mm1-root/kernel/rcutorture.c | 3 +--
1 files changed, 1 insertion(+), 2 deletions(-)

diff -puN kernel/rcutorture.c~rcutorture_fix kernel/rcutorture.c
--- linux-2.6.15-rc5-mm1/kernel/rcutorture.c~rcutorture_fix 2005-12-05 15:33:06.000000000 +0530
+++ linux-2.6.15-rc5-mm1-root/kernel/rcutorture.c 2005-12-05 15:33:17.000000000 +0530
@@ -408,9 +408,8 @@ rcu_torture_cleanup(void)
stats_task = NULL;

/* Wait for all RCU callbacks to fire. */
+ rcu_barrier();

- for (i = 0; i < RCU_TORTURE_PIPE_LEN; i++)
- synchronize_rcu();
rcu_torture_stats_print(); /* -After- the stats thread is stopped! */
printk(KERN_ALERT TORTURE_FLAG
"--- End of test: %s\n",

_


--


Thanks and Regards,
Srivatsa Vaddagiri,
Linux Technology Center,
IBM Software Labs,
Bangalore, INDIA - 560017

2005-12-05 12:32:42

by Dipankar Sarma

Subject: Re: [PATCH] Fix bug in RCU torture test

On Mon, Dec 05, 2005 at 04:28:49PM +0530, Srivatsa Vaddagiri wrote:
> While doing some test of RCU torture module, I hit a OOPS in rcu_do_batch,
> which was trying to processes callback of a module that was just removed.
> This is because we weren't waiting long enough for all callbacks to fire.
>
> diff -puN kernel/rcutorture.c~rcutorture_fix kernel/rcutorture.c
> --- linux-2.6.15-rc5-mm1/kernel/rcutorture.c~rcutorture_fix 2005-12-05 15:33:06.000000000 +0530
> +++ linux-2.6.15-rc5-mm1-root/kernel/rcutorture.c 2005-12-05 15:33:17.000000000 +0530
> @@ -408,9 +408,8 @@ rcu_torture_cleanup(void)
> stats_task = NULL;
>
> /* Wait for all RCU callbacks to fire. */
> + rcu_barrier();

Andrew,

This patch is dependent on the rcu_barrier patch buried in reiser4
patchset. It makes sense to merge that patch without waiting
for reiser4 since rcutorture is already in mainline.
I will send an updated documentation patch describing why
and how modules using RCU should use rcu_barrier().

Thanks
Dipankar

2005-12-05 13:56:32

by Paul E. McKenney

Subject: Re: [PATCH] Fix bug in RCU torture test

On Mon, Dec 05, 2005 at 04:28:49PM +0530, Srivatsa Vaddagiri wrote:
> While doing some test of RCU torture module, I hit a OOPS in rcu_do_batch,
> which was trying to processes callback of a module that was just removed.
> This is because we weren't waiting long enough for all callbacks to fire.
>
> Patch below fixes that.

Good catch -- I had forgotten about rcu_barrier().

Thanx, Paul

Acked-by: <[email protected]>

> Signed-off-by : Srivatsa Vaddagiri <[email protected]>
>
>
> ---
>
> linux-2.6.15-rc5-mm1-root/kernel/rcutorture.c | 3 +--
> 1 files changed, 1 insertion(+), 2 deletions(-)
>
> diff -puN kernel/rcutorture.c~rcutorture_fix kernel/rcutorture.c
> --- linux-2.6.15-rc5-mm1/kernel/rcutorture.c~rcutorture_fix 2005-12-05 15:33:06.000000000 +0530
> +++ linux-2.6.15-rc5-mm1-root/kernel/rcutorture.c 2005-12-05 15:33:17.000000000 +0530
> @@ -408,9 +408,8 @@ rcu_torture_cleanup(void)
> stats_task = NULL;
>
> /* Wait for all RCU callbacks to fire. */
> + rcu_barrier();
>
> - for (i = 0; i < RCU_TORTURE_PIPE_LEN; i++)
> - synchronize_rcu();
> rcu_torture_stats_print(); /* -After- the stats thread is stopped! */
> printk(KERN_ALERT TORTURE_FLAG
> "--- End of test: %s\n",
>
> _
>
>
> --
>
>
> Thanks and Regards,
> Srivatsa Vaddagiri,
> Linux Technology Center,
> IBM Software Labs,
> Bangalore, INDIA - 560017
>