linux-kernel - Re: [PATCH] rcu: mark debug_lockdep_rcu

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: <f26ebaed-6868-5ff2-e24c-fc73676256a0@yandex-team.ru>
Date:   Sun, 21 May 2017 11:51:39 +0300
From:   Konstantin Khlebnikov <khlebnikov@...dex-team.ru>
To:     paulmck@...ux.vnet.ibm.com
Cc:     linux-kernel@...r.kernel.org,
        Peter Zijlstra <peterz@...radead.org>,
        Lai Jiangshan <jiangshanlai@...il.com>,
        Josh Triplett <josh@...htriplett.org>,
        Steven Rostedt <rostedt@...dmis.org>,
        Ingo Molnar <mingo@...hat.com>,
        Mathieu Desnoyers <mathieu.desnoyers@...icios.com>
Subject: Re: [PATCH] rcu: mark debug_lockdep_rcu_enabled() as pure

On 20.05.2017 19:42, Paul E. McKenney wrote:
> On Fri, May 19, 2017 at 10:03:59AM +0300, Konstantin Khlebnikov wrote:
>> This allows to get rid of unneeded invocations.
>>
>> Function debug_lockdep_rcu_enabled() becomes really hot if several
>> debug options are enabled together with CONFIG_PROVE_RCU.
>>
>> Hottest path ends with:
>>    debug_lockdep_rcu_enabled
>>    is_ftrace_trampoline
>>    __kernel_text_address
>>
>> Here debug_lockdep_rcu_enabled() is called from condition
>> (debug_lockdep_rcu_enabled() && !__warned && (c)) inside macro
>> do_for_each_ftrace_op(), where "c" is false.
>>
>> With this patch "netperf -H localhost" shows boost from 2400 to 2500.
>>
>> Signed-off-by: Konstantin Khlebnikov <khlebnikov@...dex-team.ru>
> 
> Nice performance increase!
> 
> The gcc documentation says that __attribute__((pure)) functions are
> supposed to have return values that depend only at the function's
> arguments and the values of global variables.  However, it also says:
> 
> 	Interesting non-pure functions are functions with infinite loops
> 	or those depending on volatile memory or other system resource,
> 	that may change between two consecutive calls (such as feof in
> 	a multithreading environment).
> 
> This is OK for current->lockdep_recursion because this variable is changed
> only by the current task (I think so, anyway).
> 
> It is sort of OK for debug_locks.  This could be set to zero at any time
> by any other task, but if we have a race condition that very rarely causes
> two lockdep splats instead of just one, so what?  (But I am sure that
> some of the people on CC will correct me if I am wrong here.)
> 
> It should be OK for rcu_scheduler_active because the transition from
> RCU_SCHEDULER_INACTIVE to RCU_SCHEDULER_INIT happens before the first
> context switch, and the various barrier() calls, implied as well as
> explicit, should keep things straight.
> 
> But I don't totally trust my analysis.  Could you please get someone
> more gcc-savvy to review this and give their ack/review?  Given that,
> I will queue it.
> 
> 							Thanx, Paul

Thank you for analisys.

__attribute__((pure)) allows compiler to deduplicate / eliminate calls.
This indeed might expand race windows when global switches like debug_locks
changes their state. But strict synchronization here isn't required.

> 
>> ---
>>   include/linux/rcupdate.h |    2 +-
>>   kernel/rcu/update.c      |    2 +-
>>   2 files changed, 2 insertions(+), 2 deletions(-)
>>
>> diff --git a/include/linux/rcupdate.h b/include/linux/rcupdate.h
>> index e1e5d002fdb9..9ecb3cb715bd 100644
>> --- a/include/linux/rcupdate.h
>> +++ b/include/linux/rcupdate.h
>> @@ -472,7 +472,7 @@ extern struct lockdep_map rcu_lock_map;
>>   extern struct lockdep_map rcu_bh_lock_map;
>>   extern struct lockdep_map rcu_sched_lock_map;
>>   extern struct lockdep_map rcu_callback_map;
>> -int debug_lockdep_rcu_enabled(void);
>> +int __pure debug_lockdep_rcu_enabled(void);
>>
>>   int rcu_read_lock_held(void);
>>   int rcu_read_lock_bh_held(void);
>> diff --git a/kernel/rcu/update.c b/kernel/rcu/update.c
>> index 273e869ca21d..a0c30abefdcd 100644
>> --- a/kernel/rcu/update.c
>> +++ b/kernel/rcu/update.c
>> @@ -292,7 +292,7 @@ struct lockdep_map rcu_callback_map =
>>   	STATIC_LOCKDEP_MAP_INIT("rcu_callback", &rcu_callback_key);
>>   EXPORT_SYMBOL_GPL(rcu_callback_map);
>>
>> -int notrace debug_lockdep_rcu_enabled(void)
>> +int __pure notrace debug_lockdep_rcu_enabled(void)
>>   {
>>   	return rcu_scheduler_active != RCU_SCHEDULER_INACTIVE && debug_locks &&
>>   	       current->lockdep_recursion == 0;
>>
>