On Thu, May 28, 2020 at 11:21:36PM +0800, Kleen, Andi wrote:
>
>
> >If it's true, then there could be 2 solutions, one is to skip the WARN_ONCE as it has no practical value, as the real >check is the following code, the other is to rectify the percpu counter when the policy is changing to >OVERCOMMIT_NEVER.
>
> I think it's better to fix it up when the policy changes. That's the right place. The WARN_ON might be useful to catch other bugs.
If we keep the WARN_ON, then the draft fix patch I can think of looks like:
diff --git a/lib/percpu_counter.c b/lib/percpu_counter.c
index a66595b..02d87fc 100644
--- a/lib/percpu_counter.c
+++ b/lib/percpu_counter.c
@@ -98,6 +98,20 @@ void percpu_counter_add_batch(struct percpu_counter *fbc, s64 amount, s32 batch)
}
EXPORT_SYMBOL(percpu_counter_add_batch);
+void percpu_counter_sync(struct percpu_counter *fbc)
+{
+ unsigned long flags;
+ s64 count;
+
+ raw_spin_lock_irqsave(&fbc->lock, flags);
+ count = __this_cpu_read(*fbc->counters);
+ fbc->count += count;
+ __this_cpu_sub(*fbc->counters, count);
+ raw_spin_unlock_irqrestore(&fbc->lock, flags);
+}
+EXPORT_SYMBOL(percpu_counter_sync);
+
+
/*
* Add up all the per-cpu counts, return the result. This is a more accurate
* but much slower version of percpu_counter_read_positive()
diff --git a/mm/util.c b/mm/util.c
index 580d268..24322da 100644
--- a/mm/util.c
+++ b/mm/util.c
@@ -746,14 +746,24 @@ int overcommit_ratio_handler(struct ctl_table *table, int write, void *buffer,
return ret;
}
+static void sync_overcommit_as(struct work_struct *dummy)
+{
+ percpu_counter_sync(&vm_committed_as);
+}
+
int overcommit_policy_handler(struct ctl_table *table, int write, void *buffer,
size_t *lenp, loff_t *ppos)
{
int ret;
ret = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
- if (ret == 0 && write)
+ if (ret == 0 && write) {
+ if (sysctl_overcommit_memory == OVERCOMMIT_NEVER)
+ schedule_on_each_cpu(sync_overcommit_as);
+
mm_compute_batch();
+ }
return ret;
}
Any comments?
Thanks,
Feng
> -Andi
>
