[PATCH] cgroup: Properly init nr_tasks in cgroup_taskset

[PATCH] cgroup: Properly init nr_tasks in cgroup_taskset

[Waiman Long]

Commit 610467270fb3 ("cgroup: don't call migration methods if there
are no tasks to migrate") introduces a new field nr_tasks to the
cgroup_taskset structure for keeping track of the number of tasks
contained in the structure.  The initial value of this field, however,
is not guaranteed to be 0 as all the cgroup_taskset structures are
allocated from stack.  Therefore, we need to explicitly initilized
it in the CGROUP_TASKSET_INIT() macro for the new code to behave
correctly.

Signed-off-by: Waiman Long <longman@redhat.com>
---
 kernel/cgroup/cgroup-internal.h | 1 +
 1 file changed, 1 insertion(+)

diff --git a/kernel/cgroup/cgroup-internal.h b/kernel/cgroup/cgroup-internal.h
index 5151ff2..6b4c04e 100644
--- a/kernel/cgroup/cgroup-internal.h
+++ b/kernel/cgroup/cgroup-internal.h
@@ -76,6 +76,7 @@ struct cgroup_mgctx {
 	.src_csets		= LIST_HEAD_INIT(tset.src_csets),		\
 	.dst_csets		= LIST_HEAD_INIT(tset.dst_csets),		\
 	.csets			= &tset.src_csets,				\
+	.nr_tasks		= 0,						\
 }
 
 #define CGROUP_MGCTX_INIT(name)							\
-- 
1.8.3.1

Re: [PATCH] cgroup: Properly init nr_tasks in cgroup_taskset

[Tejun Heo]

Hello, Waiman.

On Thu, Sep 14, 2017 at 01:45:13PM -0700, Waiman Long wrote:
> Commit 610467270fb3 ("cgroup: don't call migration methods if there
> are no tasks to migrate") introduces a new field nr_tasks to the
> cgroup_taskset structure for keeping track of the number of tasks
> contained in the structure.  The initial value of this field, however,
> is not guaranteed to be 0 as all the cgroup_taskset structures are
> allocated from stack.  Therefore, we need to explicitly initilized
> it in the CGROUP_TASKSET_INIT() macro for the new code to behave
> correctly.
> 
> Signed-off-by: Waiman Long <longman@redhat.com>
> ---
>  kernel/cgroup/cgroup-internal.h | 1 +
>  1 file changed, 1 insertion(+)
> 
> diff --git a/kernel/cgroup/cgroup-internal.h b/kernel/cgroup/cgroup-internal.h
> index 5151ff2..6b4c04e 100644
> --- a/kernel/cgroup/cgroup-internal.h
> +++ b/kernel/cgroup/cgroup-internal.h
> @@ -76,6 +76,7 @@ struct cgroup_mgctx {
>  	.src_csets		= LIST_HEAD_INIT(tset.src_csets),		\
>  	.dst_csets		= LIST_HEAD_INIT(tset.dst_csets),		\
>  	.csets			= &tset.src_csets,				\
> +	.nr_tasks		= 0,						\

But it's an initialization macro, which means that the only way these
are used is as a part of whole struct initialization and the compiler
would always set the unspecified fields to zero.

Thanks.

-- 
tejun

Re: [PATCH] cgroup: Properly init nr_tasks in cgroup_taskset

[Waiman Long]

On 09/15/2017 08:47 AM, Tejun Heo wrote:
> Hello, Waiman.
>
> On Thu, Sep 14, 2017 at 01:45:13PM -0700, Waiman Long wrote:
>> Commit 610467270fb3 ("cgroup: don't call migration methods if there
>> are no tasks to migrate") introduces a new field nr_tasks to the
>> cgroup_taskset structure for keeping track of the number of tasks
>> contained in the structure.  The initial value of this field, however,
>> is not guaranteed to be 0 as all the cgroup_taskset structures are
>> allocated from stack.  Therefore, we need to explicitly initilized
>> it in the CGROUP_TASKSET_INIT() macro for the new code to behave
>> correctly.
>>
>> Signed-off-by: Waiman Long <longman@redhat.com>
>> ---
>>  kernel/cgroup/cgroup-internal.h | 1 +
>>  1 file changed, 1 insertion(+)
>>
>> diff --git a/kernel/cgroup/cgroup-internal.h b/kernel/cgroup/cgroup-internal.h
>> index 5151ff2..6b4c04e 100644
>> --- a/kernel/cgroup/cgroup-internal.h
>> +++ b/kernel/cgroup/cgroup-internal.h
>> @@ -76,6 +76,7 @@ struct cgroup_mgctx {
>>  	.src_csets		= LIST_HEAD_INIT(tset.src_csets),		\
>>  	.dst_csets		= LIST_HEAD_INIT(tset.dst_csets),		\
>>  	.csets			= &tset.src_csets,	Y			\
>> +	.nr_tasks		= 0,						\
> But it's an initialization macro, which means that the only way thhese
> are used is as a part of whole struct initialization and the compiler
> would always set the unspecified fields to zero.
>
> Thanks.
>
You are right. I checked the assembly code and it did initialize the
other fields of the structure to 0. So my fix is bogus.

I sent this patch out because our QE people were still seeing the same
panic in PPC machines even with your patch applied. So I mistakenly
thought maybe the structure initialization wasn't complete. I will have
to further into why this was still happening.

Sorry for bothering you.

Cheers,
Longman