linux-kernel - Re: [RFC PATCH 0/2] tiny_rcu: simplify tiny

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]

Message-ID: <5486C42F.7010407@cn.fujitsu.com>
Date:	Tue, 9 Dec 2014 17:43:11 +0800
From:	Lai Jiangshan <laijs@...fujitsu.com>
To:	<paulmck@...ux.vnet.ibm.com>
CC:	<linux-kernel@...r.kernel.org>,
	Josh Triplett <josh@...htriplett.org>,
	Steven Rostedt <rostedt@...dmis.org>,
	Mathieu Desnoyers <mathieu.desnoyers@...icios.com>
Subject: Re: [RFC PATCH 0/2] tiny_rcu: simplify tiny_rcu

On 11/28/2014 11:42 PM, Paul E. McKenney wrote:
> On Fri, Nov 28, 2014 at 05:43:31PM +0800, Lai Jiangshan wrote:
>> Hi, Paul
>>
>> These two patches use the special feture of the UP system:
>> 	In UP, quiescent state == grace period.
>>
>> For rcu_bh, rcu_process_callbacks() == a bh == QS == GP
>> so we can pass rcu_bh-QS and advance GP(and callbacks) in
>> rcu_process_callbacks(). After doing so, rcu_bh_qs() is useless
>> since its work is handled by rcu_process_callbacks().
>>
>> For rcu_sched, context-switch = QS = GP, thus we can force a
>> context-switch when call_rcu_sched() is happened on idle_task.
>> After doing so, rcu_idle/irq_enter/exit() are useless.
>>
>> These patches remove the useless code to simplify the tiny_rcu.
>>
>> We can change rcu_bh_qs() rcu_idle/irq_enter/exit() to static-inline-functions
>> to reduce the binary size after these two patches accepted.
>>
>> Thanks,
>> Lai
> 
> These look interesting and promising!
> 
> Could you please add the change in the tiny.o size to the commit logs?
> 
> Also, one potential performance problem with this patchset is that it is
> moving the RAISE_SOFTIRQ() from the scheduling-clock interrupt handler
> (where it simply sets a bit in a per-CPU variable) to mainline code
> (where it is often a much more expensive full wakeup of ksoftirqd).
> My guess (and hope) is that this will be in the noise for most workloads,
> but given that there is some chance of performance regressions, could
> you please measure the performance difference on some reasonable
> benchmark?  The kernbench benchmark should be just fine for this purpose.


Hi, Paul

The kernbench tests showed no performance changed.  It is expected due
to kernbench doesn't hit any network code which use RCU_BH.

But I still drop the patch1 since I only benchmarked it by kernbench.

Thanks,
Lai

> 
> 							Thanx, Paul
> 
>> Cc: Josh Triplett <josh@...htriplett.org>
>> Cc: Steven Rostedt <rostedt@...dmis.org>
>> Cc: Mathieu Desnoyers <mathieu.desnoyers@...icios.com>
>>
>> Lai Jiangshan (2):
>>   record rcu_bh quiescent state in RCU_SOFTIRQ
>>   tiny_rcu: resched when call_rcu() on idle_task
>>
>>  kernel/rcu/rcu.h         |    6 ++
>>  kernel/rcu/tiny.c        |  134 ++++++++--------------------------------------
>>  kernel/rcu/tiny_plugin.h |    2 +-
>>  kernel/rcu/tree.c        |    6 --
>>  4 files changed, 29 insertions(+), 119 deletions(-)
>>
>> -- 
>> 1.7.4.4
>>
> 
> .
> 

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/