linux-kernel - Re: [RFC PATCH 1/1] rcu: Use rcu_gp_kthread

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

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Date:	Fri, 25 Jul 2014 19:29:37 -0400
From:	Pranith Kumar <bobby.prani@...il.com>
To:	paulmck@...ux.vnet.ibm.com
CC:	Josh Triplett <josh@...htriplett.org>,
	Steven Rostedt <rostedt@...dmis.org>,
	Mathieu Desnoyers <mathieu.desnoyers@...icios.com>,
	Lai Jiangshan <laijs@...fujitsu.com>,
	"open list:READ-COPY UPDATE..." <linux-kernel@...r.kernel.org>
Subject: Re: [RFC PATCH 1/1] rcu: Use rcu_gp_kthread_wake() to wake up kthreads


On 07/25/2014 07:15 PM, Paul E. McKenney wrote:
> On Fri, Jul 25, 2014 at 06:23:41PM -0400, Pranith Kumar wrote:
>> Here total is the total number of times we enter th function rcu_report_qs_rsp()
>> and unnecessary is the times we call wake_up() unnecessarily.
>> case1, 2, 3 are the cases I listed above.
>>
>> Note that the frequency has gone way up than before, I am not sure why that is.
>>
>> *ALL* the wakeups seem to be unnecessary from that location. And the
>> main reason is that gp_flags is 0.
>>
>> My rcugp file has the following:
>>
>> completed=257515  gpnum=257516  age=1  max=1684
>>
>> Thoughts?
> Hard to believe in the rcutorture case.  My guess was that rcutorture was
> doing about 9000 wakeups, 2000 of which were unnecessary.  Which would
> of course still tilt things very much in favor of your patch.
>
> I am not surprised in the mostly-idle case, as the RCU grace-period
> kthread would most likely be the one ending the grace period, which
> would therefore almost always be a self-wakeup.
>
> Any chance of a peek at your debugging code?
>
> 							Thanx, Paul
>

Sure, I am also attaching my dmesg output. Hope it helps!

--
Pranith

---
 kernel/rcu/tree.c | 19 +++++++++++++++++++
 1 file changed, 19 insertions(+)

diff --git a/kernel/rcu/tree.c b/kernel/rcu/tree.c
index 946d47b..10ac44e 100644
--- a/kernel/rcu/tree.c
+++ b/kernel/rcu/tree.c
@@ -1936,8 +1936,27 @@ static bool rcu_start_gp(struct rcu_state *rsp)
 static void rcu_report_qs_rsp(struct rcu_state *rsp, unsigned long flags)
     __releases(rcu_get_root(rsp)->lock)
 {
+    static unsigned long total_wakeups = 0, unnecessary_wakeups = 0;
+    static unsigned long case1 = 0, case2 = 0, case3 = 0;
+
     WARN_ON_ONCE(!rcu_gp_in_progress(rsp));
     raw_spin_unlock_irqrestore(&rcu_get_root(rsp)->lock, flags);
+    total_wakeups++;
+
+    if (current == rsp->gp_kthread ||
+        !ACCESS_ONCE(rsp->gp_flags) ||
+        !rsp->gp_kthread) {
+
+        unnecessary_wakeups++;
+        if (current == rsp->gp_kthread) case1++;
+        if (!ACCESS_ONCE(rsp->gp_flags)) case2++;
+        if (!rsp->gp_kthread) case3++;
+
+        if (unnecessary_wakeups % 2000 == 0)
+            pr_info("Total:%lu, unnecessary:%lu, case1:%lu, case2:%lu, case3:%lu\n",
+                    total_wakeups, unnecessary_wakeups, case1, case2, case3);
+
+    }
     wake_up(&rsp->gp_wq);  /* Memory barrier implied by wake_up() path. */
 }
 
-- 
2.0.1

View attachment "dmesg.txt" of type "text/plain" (103790 bytes)