Login | Register For Free | Help
Search for: (Advanced)

Mailing List Archive: Linux: Kernel

[PATCH v2 19/29] skip memcg kmem allocations in specified code regions

 

 

Linux kernel RSS feed   Index | Next | Previous | View Threaded


glommer at parallels

May 11, 2012, 10:44 AM

Post #1 of 4 (108 views)
Permalink
[PATCH v2 19/29] skip memcg kmem allocations in specified code regions

This patch creates a mechanism that skip memcg allocations during
certain pieces of our core code. It basically works in the same way
as preempt_disable()/preempt_enable(): By marking a region under
which all allocations will be accounted to the root memcg.

We need this to prevent races in early cache creation, when we
allocate data using caches that are not necessarily created already.

Signed-off-by: Glauber Costa <glommer [at] parallels>
CC: Christoph Lameter <cl [at] linux>
CC: Pekka Enberg <penberg [at] cs>
CC: Michal Hocko <mhocko [at] suse>
CC: Kamezawa Hiroyuki <kamezawa.hiroyu [at] jp>
CC: Johannes Weiner <hannes [at] cmpxchg>
CC: Suleiman Souhlal <suleiman [at] google>
---
include/linux/sched.h | 1 +
mm/memcontrol.c | 25 +++++++++++++++++++++++++
2 files changed, 26 insertions(+), 0 deletions(-)

diff --git a/include/linux/sched.h b/include/linux/sched.h
index 81a173c..0501114 100644
--- a/include/linux/sched.h
+++ b/include/linux/sched.h
@@ -1613,6 +1613,7 @@ struct task_struct {
unsigned long nr_pages; /* uncharged usage */
unsigned long memsw_nr_pages; /* uncharged mem+swap usage */
} memcg_batch;
+ atomic_t memcg_kmem_skip_account;
#endif
#ifdef CONFIG_HAVE_HW_BREAKPOINT
atomic_t ptrace_bp_refcnt;
diff --git a/mm/memcontrol.c b/mm/memcontrol.c
index 5a7416b..c4ecf9c 100644
--- a/mm/memcontrol.c
+++ b/mm/memcontrol.c
@@ -479,6 +479,21 @@ struct cg_proto *tcp_proto_cgroup(struct mem_cgroup *memcg)
EXPORT_SYMBOL(tcp_proto_cgroup);
#endif /* CONFIG_INET */

+static void memcg_stop_kmem_account(void)
+{
+ if (!current->mm)
+ return;
+
+ atomic_inc(&current->memcg_kmem_skip_account);
+}
+
+static void memcg_resume_kmem_account(void)
+{
+ if (!current->mm)
+ return;
+
+ atomic_dec(&current->memcg_kmem_skip_account);
+}
char *mem_cgroup_cache_name(struct mem_cgroup *memcg, struct kmem_cache *cachep)
{
char *name;
@@ -540,7 +555,9 @@ static struct kmem_cache *memcg_create_kmem_cache(struct mem_cgroup *memcg,
if (new_cachep)
goto out;

+ memcg_stop_kmem_account();
new_cachep = kmem_cache_dup(memcg, cachep);
+ memcg_resume_kmem_account();

if (new_cachep == NULL) {
new_cachep = cachep;
@@ -631,7 +648,9 @@ static void memcg_create_cache_enqueue(struct mem_cgroup *memcg,
if (!css_tryget(&memcg->css))
return;

+ memcg_stop_kmem_account();
cw = kmalloc(sizeof(struct create_work), GFP_NOWAIT);
+ memcg_resume_kmem_account();
if (cw == NULL) {
css_put(&memcg->css);
return;
@@ -666,6 +685,9 @@ struct kmem_cache *__mem_cgroup_get_kmem_cache(struct kmem_cache *cachep,
int idx;
struct task_struct *p;

+ if (!current->mm || atomic_read(&current->memcg_kmem_skip_account))
+ return cachep;
+
gfp |= cachep->allocflags;

if (cachep->memcg_params.memcg)
@@ -700,6 +722,9 @@ bool __mem_cgroup_new_kmem_page(struct page *page, gfp_t gfp)
if (!current->mm || in_interrupt())
return true;

+ if (!current->mm || atomic_read(&current->memcg_kmem_skip_account))
+ return true;
+
rcu_read_lock();
p = rcu_dereference(current->mm->owner);
memcg = mem_cgroup_from_task(p);
--
1.7.7.6

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo [at] vger
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/


kamezawa.hiroyu at jp

May 14, 2012, 7:46 PM

Post #2 of 4 (97 views)
Permalink
Re: [PATCH v2 19/29] skip memcg kmem allocations in specified code regions [In reply to]

(2012/05/12 2:44), Glauber Costa wrote:

> This patch creates a mechanism that skip memcg allocations during
> certain pieces of our core code. It basically works in the same way
> as preempt_disable()/preempt_enable(): By marking a region under
> which all allocations will be accounted to the root memcg.
>
> We need this to prevent races in early cache creation, when we
> allocate data using caches that are not necessarily created already.
>
> Signed-off-by: Glauber Costa <glommer [at] parallels>
> CC: Christoph Lameter <cl [at] linux>
> CC: Pekka Enberg <penberg [at] cs>
> CC: Michal Hocko <mhocko [at] suse>
> CC: Kamezawa Hiroyuki <kamezawa.hiroyu [at] jp>
> CC: Johannes Weiner <hannes [at] cmpxchg>
> CC: Suleiman Souhlal <suleiman [at] google>


The concept seems okay to me but...

> ---
> include/linux/sched.h | 1 +
> mm/memcontrol.c | 25 +++++++++++++++++++++++++
> 2 files changed, 26 insertions(+), 0 deletions(-)
>
> diff --git a/include/linux/sched.h b/include/linux/sched.h
> index 81a173c..0501114 100644
> --- a/include/linux/sched.h
> +++ b/include/linux/sched.h
> @@ -1613,6 +1613,7 @@ struct task_struct {
> unsigned long nr_pages; /* uncharged usage */
> unsigned long memsw_nr_pages; /* uncharged mem+swap usage */
> } memcg_batch;
> + atomic_t memcg_kmem_skip_account;


If only 'current' thread touch this, you don't need to make this atomic counter.
you can use 'long'.

Thanks,
-Kame

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo [at] vger
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/


glommer at parallels

May 15, 2012, 11:19 PM

Post #3 of 4 (93 views)
Permalink
Re: [PATCH v2 19/29] skip memcg kmem allocations in specified code regions [In reply to]

On 05/15/2012 06:46 AM, KAMEZAWA Hiroyuki wrote:
> (2012/05/12 2:44), Glauber Costa wrote:
>
>> This patch creates a mechanism that skip memcg allocations during
>> certain pieces of our core code. It basically works in the same way
>> as preempt_disable()/preempt_enable(): By marking a region under
>> which all allocations will be accounted to the root memcg.
>>
>> We need this to prevent races in early cache creation, when we
>> allocate data using caches that are not necessarily created already.
>>
>> Signed-off-by: Glauber Costa<glommer [at] parallels>
>> CC: Christoph Lameter<cl [at] linux>
>> CC: Pekka Enberg<penberg [at] cs>
>> CC: Michal Hocko<mhocko [at] suse>
>> CC: Kamezawa Hiroyuki<kamezawa.hiroyu [at] jp>
>> CC: Johannes Weiner<hannes [at] cmpxchg>
>> CC: Suleiman Souhlal<suleiman [at] google>
>
>
> The concept seems okay to me but...
>
>> ---
>> include/linux/sched.h | 1 +
>> mm/memcontrol.c | 25 +++++++++++++++++++++++++
>> 2 files changed, 26 insertions(+), 0 deletions(-)
>>
>> diff --git a/include/linux/sched.h b/include/linux/sched.h
>> index 81a173c..0501114 100644
>> --- a/include/linux/sched.h
>> +++ b/include/linux/sched.h
>> @@ -1613,6 +1613,7 @@ struct task_struct {
>> unsigned long nr_pages; /* uncharged usage */
>> unsigned long memsw_nr_pages; /* uncharged mem+swap usage */
>> } memcg_batch;
>> + atomic_t memcg_kmem_skip_account;
>
>
> If only 'current' thread touch this, you don't need to make this atomic counter.
> you can use 'long'.
>
You're absolutely right, Kame, thanks.
I first used atomic_t because I had it tested against current->mm->owner.

Do you, btw, agree to use current instead of owner here?
You can find the rationale in earlier mails between me and Suleiman.


--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo [at] vger
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/


kamezawa.hiroyu at jp

May 16, 2012, 12:55 AM

Post #4 of 4 (93 views)
Permalink
Re: [PATCH v2 19/29] skip memcg kmem allocations in specified code regions [In reply to]

(2012/05/16 15:19), Glauber Costa wrote:

> On 05/15/2012 06:46 AM, KAMEZAWA Hiroyuki wrote:
>> (2012/05/12 2:44), Glauber Costa wrote:
>>
>>> This patch creates a mechanism that skip memcg allocations during
>>> certain pieces of our core code. It basically works in the same way
>>> as preempt_disable()/preempt_enable(): By marking a region under
>>> which all allocations will be accounted to the root memcg.
>>>
>>> We need this to prevent races in early cache creation, when we
>>> allocate data using caches that are not necessarily created already.
>>>
>>> Signed-off-by: Glauber Costa<glommer [at] parallels>
>>> CC: Christoph Lameter<cl [at] linux>
>>> CC: Pekka Enberg<penberg [at] cs>
>>> CC: Michal Hocko<mhocko [at] suse>
>>> CC: Kamezawa Hiroyuki<kamezawa.hiroyu [at] jp>
>>> CC: Johannes Weiner<hannes [at] cmpxchg>
>>> CC: Suleiman Souhlal<suleiman [at] google>
>>
>>
>> The concept seems okay to me but...
>>
>>> ---
>>> include/linux/sched.h | 1 +
>>> mm/memcontrol.c | 25 +++++++++++++++++++++++++
>>> 2 files changed, 26 insertions(+), 0 deletions(-)
>>>
>>> diff --git a/include/linux/sched.h b/include/linux/sched.h
>>> index 81a173c..0501114 100644
>>> --- a/include/linux/sched.h
>>> +++ b/include/linux/sched.h
>>> @@ -1613,6 +1613,7 @@ struct task_struct {
>>> unsigned long nr_pages; /* uncharged usage */
>>> unsigned long memsw_nr_pages; /* uncharged mem+swap usage */
>>> } memcg_batch;
>>> + atomic_t memcg_kmem_skip_account;
>>
>>
>> If only 'current' thread touch this, you don't need to make this atomic counter.
>> you can use 'long'.
>>
> You're absolutely right, Kame, thanks.
> I first used atomic_t because I had it tested against current->mm->owner.
>
> Do you, btw, agree to use current instead of owner here?
> You can find the rationale in earlier mails between me and Suleiman.

I agree to use current. This information depends on the context of callers.

Thanks,
-Kame


--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo [at] vger
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/

Linux kernel RSS feed   Index | Next | Previous | View Threaded
 
 


Interested in having your list archived? Contact Gossamer Threads
 
  Web Applications & Managed Hosting Powered by Gossamer Threads Inc.