Add rcu_sync_struct->exclusive boolean set by rcu_sync_init(),
it obviously controls the exclusiveness of rcu_sync_enter().
This is what percpu_down_write() actually wants.
We turn ->gp_wait into "struct completion gp_comp", it is used
as a resource counter in "exclusive" mode. Otherwise we only use
its completion->wait member for wait_event/wake_up_all. We never
mix the completion/wait_queue_head_t operations.
Note: it would be more clean to do __complete_locked() under
->rss_lock in rcu_sync_exit() in the "else" branch, but we don't
have this trivial helper.
Signed-off-by: Oleg Nesterov <o...@redhat.com>
---
include/linux/rcusync.h | 29 ++++++++++++++++-------------
kernel/cpu.c | 2 +-
kernel/rcusync.c | 22 +++++++++++++++++-----
3 files changed, 34 insertions(+), 19 deletions(-)
diff --git a/include/linux/rcusync.h b/include/linux/rcusync.h
index 33864a0..5689f24 100644
--- a/include/linux/rcusync.h
+++ b/include/linux/rcusync.h
@@ -1,8 +1,8 @@
#ifndef _LINUX_RCUSYNC_H_
#define _LINUX_RCUSYNC_H_
-#include <linux/wait.h>
#include <linux/rcupdate.h>
+#include <linux/completion.h>
struct rcu_sync_ops {
void (*sync)(void);
@@ -16,11 +16,12 @@ struct rcu_sync_ops {
struct rcu_sync_struct {
int gp_state;
int gp_count;
- wait_queue_head_t gp_wait;
+ struct completion gp_comp;
int cb_state;
struct rcu_head cb_head;
+ bool exclusive;
struct rcu_sync_ops *ops;
};
@@ -34,32 +35,34 @@ static inline bool rcu_sync_is_idle(struct rcu_sync_struct
*rss)
enum rcu_sync_type { RCU_SYNC, RCU_SCHED_SYNC, RCU_BH_SYNC };
-extern void rcu_sync_init(struct rcu_sync_struct *, enum rcu_sync_type);
+extern void rcu_sync_init(struct rcu_sync_struct *,
+ enum rcu_sync_type, bool excl);
extern void rcu_sync_enter(struct rcu_sync_struct *);
extern void rcu_sync_exit(struct rcu_sync_struct *);
extern void rcu_sync_dtor(struct rcu_sync_struct *);
extern struct rcu_sync_ops rcu_sync_ops_array[];
-#define __RCU_SYNC_INITIALIZER(name, type) { \
+#define __RCU_SYNC_INITIALIZER(name, type, excl) { \
.gp_state = 0, \
.gp_count = 0, \
- .gp_wait = __WAIT_QUEUE_HEAD_INITIALIZER(name.gp_wait), \
+ .gp_comp = COMPLETION_INITIALIZER(name.gp_comp), \
.cb_state = 0, \
+ .exclusive = excl, \
.ops = rcu_sync_ops_array + (type), \
}
-#define __DEFINE_RCU_SYNC(name, type) \
- struct rcu_sync_struct name = __RCU_SYNC_INITIALIZER(name, type)
+#define __DEFINE_RCU_SYNC(name, type, excl) \
+ struct rcu_sync_struct name = __RCU_SYNC_INITIALIZER(name, type, excl)
-#define DEFINE_RCU_SYNC(name) \
- __DEFINE_RCU_SYNC(name, RCU_SYNC)
+#define DEFINE_RCU_SYNC(name, excl) \
+ __DEFINE_RCU_SYNC(name, RCU_SYNC, excl)
-#define DEFINE_RCU_SCHED_SYNC(name) \
- __DEFINE_RCU_SYNC(name, RCU_SCHED_SYNC)
+#define DEFINE_RCU_SCHED_SYNC(name, excl) \
+ __DEFINE_RCU_SYNC(name, RCU_SCHED_SYNC, excl)
-#define DEFINE_RCU_BH_SYNC(name) \
- __DEFINE_RCU_SYNC(name, RCU_BH_SYNC)
+#define DEFINE_RCU_BH_SYNC(name, excl) \
+ __DEFINE_RCU_SYNC(name, RCU_BH_SYNC, excl)
#endif /* _LINUX_RCUSYNC_H_ */
diff --git a/kernel/cpu.c b/kernel/cpu.c
index d5f475a..fb1bdf0 100644
--- a/kernel/cpu.c
+++ b/kernel/cpu.c
@@ -51,7 +51,7 @@ static int cpu_hotplug_disabled;
enum { readers_slow, readers_block };
-DEFINE_RCU_SCHED_SYNC(__cpuhp_rss);
+DEFINE_RCU_SCHED_SYNC(__cpuhp_rss, false);
EXPORT_SYMBOL_GPL(__cpuhp_rss);
DEFINE_PER_CPU(unsigned int, __cpuhp_refcount);
diff --git a/kernel/rcusync.c b/kernel/rcusync.c
index bb311eb..667eb7d 100644
--- a/kernel/rcusync.c
+++ b/kernel/rcusync.c
@@ -4,7 +4,7 @@
enum { GP_IDLE = 0, GP_PENDING, GP_PASSED };
enum { CB_IDLE = 0, CB_PENDING, CB_REPLAY };
-#define rss_lock gp_wait.lock
+#define rss_lock gp_comp.wait.lock
#ifdef CONFIG_PROVE_RCU
#define __INIT_HELD(func) .held = func,
@@ -33,11 +33,13 @@ struct rcu_sync_ops rcu_sync_ops_array[] = {
},
};
-void rcu_sync_init(struct rcu_sync_struct *rss, enum rcu_sync_type type)
+void rcu_sync_init(struct rcu_sync_struct *rss,
+ enum rcu_sync_type type, bool excl)
{
memset(rss, 0, sizeof(*rss));
- init_waitqueue_head(&rss->gp_wait);
+ init_completion(&rss->gp_comp);
rss->ops = rcu_sync_ops_array + type;
+ rss->exclusive = excl;
}
void rcu_sync_enter(struct rcu_sync_struct *rss)
@@ -56,9 +58,13 @@ void rcu_sync_enter(struct rcu_sync_struct *rss)
if (need_sync) {
rss->ops->sync();
rss->gp_state = GP_PASSED;
- wake_up_all(&rss->gp_wait);
+ if (!rss->exclusive)
+ wake_up_all(&rss->gp_comp.wait);
} else if (need_wait) {
- wait_event(rss->gp_wait, rss->gp_state == GP_PASSED);
+ if (!rss->exclusive)
+ wait_event(rss->gp_comp.wait, rss->gp_state ==
GP_PASSED);
+ else
+ wait_for_completion(&rss->gp_comp);
} else {
/*
* Possible when there's a pending CB from a rcu_sync_exit().
@@ -105,6 +111,8 @@ static void rcu_sync_func(struct rcu_head *rcu)
void rcu_sync_exit(struct rcu_sync_struct *rss)
{
+ bool wakeup_excl = rss->exclusive;
+
spin_lock_irq(&rss->rss_lock);
if (!--rss->gp_count) {
if (rss->cb_state == CB_IDLE) {
@@ -113,8 +121,12 @@ void rcu_sync_exit(struct rcu_sync_struct *rss)
} else if (rss->cb_state == CB_PENDING) {
rss->cb_state = CB_REPLAY;
}
+ wakeup_excl = false;
}
spin_unlock_irq(&rss->rss_lock);
+
+ if (wakeup_excl)
+ complete(&rss->gp_comp);
}
void rcu_sync_dtor(struct rcu_sync_struct *rss)
--
1.5.5.1
--
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/
