Re: [PATCH] locking/rtmutex: remove unused cmpxchg_relaxed

[Alex Shi <alex.shi@linux.alibaba.com>]

在 2020/2/1 上午4:23, Thomas Gleixner 写道:
> Davidlohr Bueso <dave@stgolabs.net> writes:
>> On Tue, 21 Jan 2020, Alex Shi wrote:
> 
>   Subject: locking/rtmutex: remove unused cmpxchg_relaxed
> 
> should be
> 
>   Subject: locking/rtmutex: Remove unused rt_mutex_cmpxchg_relaxed()
> 
> You're not removing cmpxchg_relaxed, right?
> 
>>> No one use this macro after it was introduced. Better to remove it?
> 
> Please make that factual.
> 
>  The macro was never used at all. Remove it.
> 
>> You also need to remove it for the CONFIG_DEBUG_RT_MUTEXES=y case.
> 
> Yes.
> 
>> Hmm unrelated, but do we want CCAS for rtmutex fastpath? Ie:
>>
>>      (l->owner == c && cmpxchg_acquire(&l->owner, c, n) == c)
>>
>> That would optimize for the contended case and avoid the cmpxchg - it would
>> also help if we ever do the top-waiter spin thing.
> 
> Not sure if it buys much, but it kinda makes sense.
> 
> Thanks,
> 
>         tglx
> 
Thanks Thomas and David!
Is this following patch ok?

Thanks
Alex
---
From 4cf9e38a73c67c6894f3addb2ddca26bb51b1a28 Mon Sep 17 00:00:00 2001
From: Alex Shi <alex.shi@linux.alibaba.com>
Date: Tue, 21 Jan 2020 15:03:33 +0800
Subject: [PATCH v2] locking/rtmutex: optimize rt_mutex_cmpxchg_xxx series func

rt_mutex_cmpxchg_relexed isn't interested by anyone, so remove it.
And Davidlohr Bueso suggests check l->owner before cmpxchg to reduce
lock contention.

Signed-off-by: Alex Shi <alex.shi@linux.alibaba.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Davidlohr Bueso <dave@stgolabs.net>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Will Deacon <will@kernel.org>
Cc: linux-kernel@vger.kernel.org
---
 kernel/locking/rtmutex.c | 8 ++++----
 1 file changed, 4 insertions(+), 4 deletions(-)

diff --git a/kernel/locking/rtmutex.c b/kernel/locking/rtmutex.c
index 851bbb10819d..eb26f4e57ce4 100644
--- a/kernel/locking/rtmutex.c
+++ b/kernel/locking/rtmutex.c
@@ -141,9 +141,10 @@ static void fixup_rt_mutex_waiters(struct rt_mutex *lock)
  * set up.
  */
 #ifndef CONFIG_DEBUG_RT_MUTEXES
-# define rt_mutex_cmpxchg_relaxed(l,c,n) (cmpxchg_relaxed(&l->owner, c, n) == c)
-# define rt_mutex_cmpxchg_acquire(l,c,n) (cmpxchg_acquire(&l->owner, c, n) == c)
-# define rt_mutex_cmpxchg_release(l,c,n) (cmpxchg_release(&l->owner, c, n) == c)
+# define rt_mutex_cmpxchg_acquire(l,c,n)	\
+		(l->owner == c && cmpxchg_acquire(&l->owner, c, n) == c)
+# define rt_mutex_cmpxchg_release(l,c,n)	\
+		(l->owner == c && cmpxchg_release(&l->owner, c, n) == c)
 
 /*
  * Callers must hold the ->wait_lock -- which is the whole purpose as we force
@@ -202,7 +203,6 @@ static inline bool unlock_rt_mutex_safe(struct rt_mutex *lock,
 }
 
 #else
-# define rt_mutex_cmpxchg_relaxed(l,c,n)	(0)
 # define rt_mutex_cmpxchg_acquire(l,c,n)	(0)
 # define rt_mutex_cmpxchg_release(l,c,n)	(0)
 
-- 
1.8.3.1