Yasuaki Ishimatsu hit a bug when the numa mapping between CPU and node
is changed. And the previous path fixup wq_numa_possible_cpumask.
(See more information form the changelog of that patch)
After wq_numa_possible_cpumask was updated, the new pool->node will be
correct, but the existing pools (and workers) are still running, some of
them are running with the wrong pool->node, or even worse, with pool->node
which is quitted node, they create_worker() on wrong pool->node.
These create_worker() may create workers on wrong node or failed without
any progress (when with pool->node which is quitted node).
So we need to update the pool->node when the numa mapping is changed.
We simply re-calc the pool->node when the numa mapping changed. It reuses
the code from get_unbound_pool() for unbound pool.
Reported-by: Yasuaki Ishimatsu <isimatu.yasu...@jp.fujitsu.com>
Cc: Tejun Heo <t...@kernel.org>
Cc: Yasuaki Ishimatsu <isimatu.yasu...@jp.fujitsu.com>
Cc: "Gu, Zheng" <guz.f...@cn.fujitsu.com>
Cc: tangchen <tangc...@cn.fujitsu.com>
Cc: Hiroyuki KAMEZAWA <kamezawa.hir...@jp.fujitsu.com>
Signed-off-by: Lai Jiangshan <la...@cn.fujitsu.com>
---
kernel/workqueue.c | 53 ++++++++++++++++++++++++++++++++++++++-------------
1 files changed, 39 insertions(+), 14 deletions(-)
diff --git a/kernel/workqueue.c b/kernel/workqueue.c
index 4c88b61..7fbabf6 100644
--- a/kernel/workqueue.c
+++ b/kernel/workqueue.c
@@ -3439,6 +3439,26 @@ static void put_unbound_pool(struct worker_pool *pool)
call_rcu_sched(&pool->rcu, rcu_free_pool);
}
+static int calc_pool_node(struct worker_pool *pool)
+{
+ int node;
+
+ if (pool->cpu >= 0)
+ return cpu_to_node(pool->cpu);
+
+ /* if cpumask is contained inside a NUMA node, we belong to that node */
+ if (wq_numa_enabled) {
+ for_each_node(node) {
+ if (cpumask_subset(pool->attrs->cpumask,
+ wq_numa_possible_cpumask[node])) {
+ return node;
+ }
+ }
+ }
+
+ return NUMA_NO_NODE;
+}
+
/**
* get_unbound_pool - get a worker_pool with the specified attributes
* @attrs: the attributes of the worker_pool to get
@@ -3457,7 +3477,6 @@ static struct worker_pool *get_unbound_pool(const struct
workqueue_attrs *attrs)
{
u32 hash = wqattrs_hash(attrs);
struct worker_pool *pool;
- int node;
lockdep_assert_held(&wq_pool_mutex);
@@ -3482,17 +3501,7 @@ static struct worker_pool *get_unbound_pool(const struct
workqueue_attrs *attrs)
* 'struct workqueue_attrs' comments for detail.
*/
pool->attrs->no_numa = false;
-
- /* if cpumask is contained inside a NUMA node, we belong to that node */
- if (wq_numa_enabled) {
- for_each_node(node) {
- if (cpumask_subset(pool->attrs->cpumask,
- wq_numa_possible_cpumask[node])) {
- pool->node = node;
- break;
- }
- }
- }
+ pool->node = calc_pool_node(pool);
if (worker_pool_assign_id(pool) < 0)
goto fail;
@@ -3952,6 +3961,8 @@ out_unlock:
static void wq_update_numa_mapping(int cpu)
{
int node, orig_node = NUMA_NO_NODE, new_node = cpu_to_node(cpu);
+ struct worker_pool *pool;
+ int pi;
lockdep_assert_held(&wq_pool_mutex);
@@ -3985,6 +3996,17 @@ static void wq_update_numa_mapping(int cpu)
else if (orig_node != NUMA_NO_NODE && node == orig_node)
cpumask_set_cpu(cpu,
wq_numa_possible_cpumask[orig_node]);
}
+
+ /*
+ * Fixup pool->node, it needs to test for all pools, pool->node might be
+ * changed from orig_node to new_node or from new_node to NUMA_NO_NODE
+ * or NUMA_NO_NODE to orig_node or some other possibilities.
+ */
+ for_each_pool(pool, pi) {
+ node = calc_pool_node(pool);
+ if (pool->node != node)
+ pool->node = node;
+ }
}
static int alloc_and_link_pwqs(struct workqueue_struct *wq)
@@ -4614,10 +4636,13 @@ static int workqueue_cpu_up_callback(struct
notifier_block *nfb,
for_each_pool(pool, pi) {
mutex_lock(&pool->attach_mutex);
- if (pool->cpu == cpu)
+ if (pool->cpu == cpu) {
+ if (unlikely(pool->node != cpu_to_node(cpu)))
+ pool->node = cpu_to_node(cpu);
rebind_workers(pool);
- else if (pool->cpu < 0)
+ } else if (pool->cpu < 0) {
restore_unbound_workers_cpumask(pool, cpu);
+ }
mutex_unlock(&pool->attach_mutex);
}
--
1.7.4.4
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majord...@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
