| lists.openwall.net | lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC | |
|
Open Source and information security mailing list archives
| ||
|
Message-Id: <20120207091906.1fd6eb40.kamezawa.hiroyu@jp.fujitsu.com>
Date: Tue, 7 Feb 2012 09:19:06 +0900
From: KAMEZAWA Hiroyuki <kamezawa.hiroyu@...fujitsu.com>
To: Andrew Morton <akpm@...ux-foundation.org>
Cc: "linux-mm@...ck.org" <linux-mm@...ck.org>,
"linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>,
"hannes@...xchg.org" <hannes@...xchg.org>,
Michal Hocko <mhocko@...e.cz>, Ying Han <yinghan@...gle.com>,
Hugh Dickins <hughd@...gle.com>
Subject: Re: [PATCH 1/6] memcg: simplify move_account() check.
On Mon, 6 Feb 2012 14:38:53 -0800
Andrew Morton <akpm@...ux-foundation.org> wrote:
> On Mon, 6 Feb 2012 19:07:59 +0900
> KAMEZAWA Hiroyuki <kamezawa.hiroyu@...fujitsu.com> wrote:
>
> > >From c75cc843ca0cb36de97ab814e59fb4ab7b1ffbd1 Mon Sep 17 00:00:00 2001
> > From: KAMEZAWA Hiroyuki <kamezawa.hiroyu@...fujitsu.com>
> > Date: Thu, 2 Feb 2012 10:02:39 +0900
> > Subject: [PATCH 1/6] memcg: simplify move_account() check.
> >
> > In memcg, for avoiding take-lock-irq-off at accessing page_cgroup,
> > a logic, flag + rcu_read_lock(), is used. This works as following
> >
> > CPU-A CPU-B
> > rcu_read_lock()
> > set flag
> > if(flag is set)
> > take heavy lock
> > do job.
> > synchronize_rcu() rcu_read_unlock()
> >
> > In recent discussion, it's argued that using per-cpu value for this
> > flag just complicates the code because 'set flag' is very rare.
> >
> > This patch changes 'flag' implementation from percpu to atomic_t.
> > This will be much simpler.
> >
>
> To me, "RFC" says "might not be ready for merging yet". You're up to
> v3 - why is it still RFC? You're still expecting to make significant
> changes?
>
Yes, I made changes discussed in v2. and need to show how it looks.
I'm sorry that changelog wasn't enough.
> >
> > }
> > +/*
> > + * memcg->moving_account is used for checking possibility that some thread is
> > + * calling move_account(). When a thread on CPU-A starts moving pages under
> > + * a memcg, other threads sholud check memcg->moving_account under
>
> "should"
>
Sure..
> > + * rcu_read_lock(), like this:
> > + *
> > + * CPU-A CPU-B
> > + * rcu_read_lock()
> > + * memcg->moving_account+1 if (memcg->mocing_account)
> > + * take havier locks.
> > + * syncronize_rcu() update something.
> > + * rcu_read_unlock()
> > + * start move here.
> > + */
> >
> > static void mem_cgroup_start_move(struct mem_cgroup *memcg)
> > {
> > - int cpu;
> > -
> > - get_online_cpus();
> > - spin_lock(&memcg->pcp_counter_lock);
> > - for_each_online_cpu(cpu)
> > - per_cpu(memcg->stat->count[MEM_CGROUP_ON_MOVE], cpu) += 1;
> > - memcg->nocpu_base.count[MEM_CGROUP_ON_MOVE] += 1;
> > - spin_unlock(&memcg->pcp_counter_lock);
> > - put_online_cpus();
> > -
> > + atomic_inc(&memcg->moving_account);
> > synchronize_rcu();
> > }
> >
> > static void mem_cgroup_end_move(struct mem_cgroup *memcg)
> > {
> > - int cpu;
> > -
> > - if (!memcg)
> > - return;
> > - get_online_cpus();
> > - spin_lock(&memcg->pcp_counter_lock);
> > - for_each_online_cpu(cpu)
> > - per_cpu(memcg->stat->count[MEM_CGROUP_ON_MOVE], cpu) -= 1;
> > - memcg->nocpu_base.count[MEM_CGROUP_ON_MOVE] -= 1;
> > - spin_unlock(&memcg->pcp_counter_lock);
> > - put_online_cpus();
> > + if (memcg)
> > + atomic_dec(&memcg->moving_account);
> > }
>
> It's strange that end_move handles a NULL memcg but start_move does not.
>
Ah, the reason was that mem_cgroup_end_move() can called in mem_cgroup_clear_mc().
This mem_cgroup_clear_mc() can call mem_cgroup_end_move(NULL)...
Then, this function has NULL check in callee side.
I'll add comments.
> > /*
> > * 2 routines for checking "mem" is under move_account() or not.
> > @@ -1298,7 +1297,7 @@ static void mem_cgroup_end_move(struct mem_cgroup *memcg)
> > static bool mem_cgroup_stealed(struct mem_cgroup *memcg)
> > {
> > VM_BUG_ON(!rcu_read_lock_held());
> > - return this_cpu_read(memcg->stat->count[MEM_CGROUP_ON_MOVE]) > 0;
> > + return atomic_read(&memcg->moving_account);
> > }
>
> So a bool-returning function can return something > 1?
>
> I don't know what the compiler would make of that. Presumably "if (b)"
> will work OK, but will "if (b1 == b2)"?
>
if (!mem_cgroup_stealed(memcg))
ffffffff8116e278: 85 c0 test %eax,%eax
ffffffff8116e27a: 74 1f je ffffffff8116e29b <__mem_cgroup_begin_update_page_stat+0x7b>
return;
ffffffff8116e29b: 5b pop %rbx
ffffffff8116e29c: 41 5c pop %r12
ffffffff8116e29e: 41 5d pop %r13
ffffffff8116e2a0: 41 5e pop %r14
ffffffff8116e2a2: c9 leaveq
ffffffff8116e2a3: c3 retq
Maybe works as expected but... I'll rewrite..how about this ?.
>From 3cefc03a4da41843ea2439f1c0ca630c64e8cf87 Mon Sep 17 00:00:00 2001
From: KAMEZAWA Hiroyuki <kamezawa.hiroyu@...fujitsu.com>
Date: Thu, 2 Feb 2012 10:02:39 +0900
Subject: [PATCH] memcg: simplify move_account() check.
In memcg, for avoiding take-lock-irq-off at accessing page_cgroup,
a logic, flag + rcu_read_lock(), is used. This works as following
CPU-A CPU-B
rcu_read_lock()
set flag
if(flag is set)
take heavy lock
do job.
synchronize_rcu() rcu_read_unlock()
In recent discussion, it's argued that using per-cpu value for this
flag just complicates the code because 'set flag' is very rare.
This patch changes 'flag' implementation from percpu to atomic_t.
This will be much simpler.
Changelog: v4
- fixed many typos.
- check return variables should be bool
- add comments.
Changelog: v3
- this is a new patch since v3.
memcontrol.c | 65 ++++++++++++++++++++++-------------------------------------
1 file changed, 25 insertions(+), 40 deletions(-)
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@...fujitsu.com>
---
mm/memcontrol.c | 70 +++++++++++++++++++++++-------------------------------
1 files changed, 30 insertions(+), 40 deletions(-)
diff --git a/mm/memcontrol.c b/mm/memcontrol.c
index ab315ab..0359175 100644
--- a/mm/memcontrol.c
+++ b/mm/memcontrol.c
@@ -89,7 +89,6 @@ enum mem_cgroup_stat_index {
MEM_CGROUP_STAT_FILE_MAPPED, /* # of pages charged as file rss */
MEM_CGROUP_STAT_SWAPOUT, /* # of pages, swapped out */
MEM_CGROUP_STAT_DATA, /* end of data requires synchronization */
- MEM_CGROUP_ON_MOVE, /* someone is moving account between groups */
MEM_CGROUP_STAT_NSTATS,
};
@@ -278,6 +277,10 @@ struct mem_cgroup {
*/
unsigned long move_charge_at_immigrate;
/*
+ * set > 0 if pages under this cgroup are moving to other cgroup.
+ */
+ atomic_t moving_account;
+ /*
* percpu counter.
*/
struct mem_cgroup_stat_cpu *stat;
@@ -1253,36 +1256,37 @@ int mem_cgroup_swappiness(struct mem_cgroup *memcg)
return memcg->swappiness;
}
+/*
+ * memcg->moving_account is used for checking possibility that some thread is
+ * calling move_account(). When a thread on CPU-A starts moving pages under
+ * a memcg, other threads should check memcg->moving_account under
+ * rcu_read_lock(), like this:
+ *
+ * CPU-A CPU-B
+ * rcu_read_lock()
+ * memcg->moving_account+1 if (memcg->mocing_account)
+ * take heavy locks.
+ * synchronize_rcu() update something.
+ * rcu_read_unlock()
+ * start move here.
+ */
static void mem_cgroup_start_move(struct mem_cgroup *memcg)
{
- int cpu;
-
- get_online_cpus();
- spin_lock(&memcg->pcp_counter_lock);
- for_each_online_cpu(cpu)
- per_cpu(memcg->stat->count[MEM_CGROUP_ON_MOVE], cpu) += 1;
- memcg->nocpu_base.count[MEM_CGROUP_ON_MOVE] += 1;
- spin_unlock(&memcg->pcp_counter_lock);
- put_online_cpus();
-
+ atomic_inc(&memcg->moving_account);
synchronize_rcu();
}
static void mem_cgroup_end_move(struct mem_cgroup *memcg)
{
- int cpu;
-
- if (!memcg)
- return;
- get_online_cpus();
- spin_lock(&memcg->pcp_counter_lock);
- for_each_online_cpu(cpu)
- per_cpu(memcg->stat->count[MEM_CGROUP_ON_MOVE], cpu) -= 1;
- memcg->nocpu_base.count[MEM_CGROUP_ON_MOVE] -= 1;
- spin_unlock(&memcg->pcp_counter_lock);
- put_online_cpus();
+ /*
+ * Now, mem_cgroup_clear_mc() may call this function with NULL.
+ * We check NULL in callee rather than caller.
+ */
+ if (memcg)
+ atomic_dec(&memcg->moving_account);
}
+
/*
* 2 routines for checking "mem" is under move_account() or not.
*
@@ -1298,7 +1302,7 @@ static void mem_cgroup_end_move(struct mem_cgroup *memcg)
static bool mem_cgroup_stealed(struct mem_cgroup *memcg)
{
VM_BUG_ON(!rcu_read_lock_held());
- return this_cpu_read(memcg->stat->count[MEM_CGROUP_ON_MOVE]) > 0;
+ return atomic_read(&memcg->moving_account) > 0;
}
static bool mem_cgroup_under_move(struct mem_cgroup *memcg)
@@ -1849,8 +1853,8 @@ bool mem_cgroup_handle_oom(struct mem_cgroup *memcg, gfp_t mask)
* by flags.
*
* Considering "move", this is an only case we see a race. To make the race
- * small, we check MEM_CGROUP_ON_MOVE percpu value and detect there are
- * possibility of race condition. If there is, we take a lock.
+ * small, we check mm->moving_account and detect there are possibility of race
+ * If there is, we take a lock.
*/
void mem_cgroup_update_page_stat(struct page *page,
@@ -2068,17 +2072,6 @@ static void mem_cgroup_drain_pcp_counter(struct mem_cgroup *memcg, int cpu)
per_cpu(memcg->stat->events[i], cpu) = 0;
memcg->nocpu_base.events[i] += x;
}
- /* need to clear ON_MOVE value, works as a kind of lock. */
- per_cpu(memcg->stat->count[MEM_CGROUP_ON_MOVE], cpu) = 0;
- spin_unlock(&memcg->pcp_counter_lock);
-}
-
-static void synchronize_mem_cgroup_on_move(struct mem_cgroup *memcg, int cpu)
-{
- int idx = MEM_CGROUP_ON_MOVE;
-
- spin_lock(&memcg->pcp_counter_lock);
- per_cpu(memcg->stat->count[idx], cpu) = memcg->nocpu_base.count[idx];
spin_unlock(&memcg->pcp_counter_lock);
}
@@ -2090,11 +2083,8 @@ static int __cpuinit memcg_cpu_hotplug_callback(struct notifier_block *nb,
struct memcg_stock_pcp *stock;
struct mem_cgroup *iter;
- if ((action == CPU_ONLINE)) {
- for_each_mem_cgroup(iter)
- synchronize_mem_cgroup_on_move(iter, cpu);
+ if ((action == CPU_ONLINE))
return NOTIFY_OK;
- }
if ((action != CPU_DEAD) || action != CPU_DEAD_FROZEN)
return NOTIFY_OK;
--
1.7.4.1
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
Powered by blists - more mailing lists