[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [Qemu-devel] [PATCH] rcu: actually register threads that have RCU re
From: |
Wen Congyang |
Subject: |
Re: [Qemu-devel] [PATCH] rcu: actually register threads that have RCU read-side critical sections |
Date: |
Fri, 24 Jul 2015 13:56:16 +0800 |
User-agent: |
Mozilla/5.0 (X11; Linux x86_64; rv:38.0) Gecko/20100101 Thunderbird/38.1.0 |
On 07/24/2015 12:58 AM, Paolo Bonzini wrote:
>
>
> On 23/07/2015 14:59, Wen Congyang wrote:
>>>>
>>>> If the thread doesn't use RCU, rcu_register_thread() is harmless, is
>>>> it right?
>>>
>>> Every rcu_register_thread() makes synchronize_rcu() a little slower.
>>
>> Yes, but synchronize_rcu() is very slow...
>
> Hmm, worse, rcu_register_thread() if called together with
> synchronize_rcu() it waits for the synchronize_rcu() to finish. :/
What about this modification:
diff --git a/util/rcu.c b/util/rcu.c
index 7270151..ccf8cfa 100644
--- a/util/rcu.c
+++ b/util/rcu.c
@@ -48,6 +48,7 @@ unsigned long rcu_gp_ctr = RCU_GP_LOCKED;
QemuEvent rcu_gp_event;
static QemuMutex rcu_gp_lock;
+static QemuMutex rcu_sync_lock;
/*
* Check whether a quiescent state was crossed between the beginning of
@@ -115,9 +116,12 @@ static void wait_for_readers(void)
}
/* Wait for one thread to report a quiescent state and
- * try again.
+ * try again. Release rcu_gp_lock, so rcu_(un)register_thread()
+ * doesn't wait too much time.
*/
+ qemu_mutex_unlock(&rcu_gp_lock);
qemu_event_wait(&rcu_gp_event);
+ qemu_mutex_lock(&rcu_gp_lock);
}
/* put back the reader list in the registry */
@@ -126,6 +130,7 @@ static void wait_for_readers(void)
void synchronize_rcu(void)
{
+ qemu_mutex_lock(&rcu_sync_lock);
qemu_mutex_lock(&rcu_gp_lock);
if (!QLIST_EMPTY(®istry)) {
@@ -150,6 +155,7 @@ void synchronize_rcu(void)
}
qemu_mutex_unlock(&rcu_gp_lock);
+ qemu_mutex_unlock(&rcu_sync_lock);
}
@@ -288,6 +294,7 @@ static void rcu_init_complete(void)
QemuThread thread;
qemu_mutex_init(&rcu_gp_lock);
+ qemu_mutex_init(&rcu_sync_lock);
qemu_event_init(&rcu_gp_event, true);
qemu_event_init(&rcu_call_ready_event, false);
@@ -304,12 +311,14 @@ static void rcu_init_complete(void)
#ifdef CONFIG_POSIX
static void rcu_init_lock(void)
{
+ qemu_mutex_lock(&rcu_sync_lock);
qemu_mutex_lock(&rcu_gp_lock);
}
static void rcu_init_unlock(void)
{
qemu_mutex_unlock(&rcu_gp_lock);
+ qemu_mutex_unlock(&rcu_sync_lock);
}
#endif
rcu_register_thread() will be a littl slower when it is
called together with synchronize_rcu().
Thanks
Wen Congyang
>
> Paolo
>
>>>
>>>>>> be simpler to add an assertion in rcu_register_thread. I'm just a bit
>>>>>> wary of doing little more than the bare minimum in 2.4, because of the
>>>>>> OS X failure that I didn't quite understand.
>>>> Which problem? I don't find it in the maillist.
>>>
>>> http://article.gmane.org/gmane.comp.emulators.qemu/351548
>>
>> Hmm, I guess rcu_reader is invalid when pthread key is destroyed.
>> pthread key and __thread
>> variable, which is destroyed first? I don't find any document to
>> describe it.
>
- [Qemu-devel] [PATCH] rcu: actually register threads that have RCU read-side critical sections, Paolo Bonzini, 2015/07/22
- Re: [Qemu-devel] [PATCH] rcu: actually register threads that have RCU read-side critical sections, Wen Congyang, 2015/07/22
- Re: [Qemu-devel] [PATCH] rcu: actually register threads that have RCU read-side critical sections, Christian Borntraeger, 2015/07/23
- Re: [Qemu-devel] [PATCH] rcu: actually register threads that have RCU read-side critical sections, Paolo Bonzini, 2015/07/23
- Re: [Qemu-devel] [PATCH] rcu: actually register threads that have RCU read-side critical sections, Wen Congyang, 2015/07/23
- Re: [Qemu-devel] [PATCH] rcu: actually register threads that have RCU read-side critical sections, Paolo Bonzini, 2015/07/23
- Re: [Qemu-devel] [PATCH] rcu: actually register threads that have RCU read-side critical sections, Wen Congyang, 2015/07/23
- Re: [Qemu-devel] [PATCH] rcu: actually register threads that have RCU read-side critical sections, Paolo Bonzini, 2015/07/23
- Re: [Qemu-devel] [PATCH] rcu: actually register threads that have RCU read-side critical sections, Wen Congyang, 2015/07/23
- Re: [Qemu-devel] [PATCH] rcu: actually register threads that have RCU read-side critical sections,
Wen Congyang <=
- Re: [Qemu-devel] [PATCH] rcu: actually register threads that have RCU read-side critical sections, Paolo Bonzini, 2015/07/24
- Re: [Qemu-devel] [PATCH] rcu: actually register threads that have RCU read-side critical sections, Wen Congyang, 2015/07/24