Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Message-ID: <2437ac29-29f0-34f9-b7cb-f0e294db7dc6@loongson.cn>
Date:   Tue, 1 Aug 2023 09:37:33 +0800
MIME-Version: 1.0
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:102.0) Gecko/20100101
 Thunderbird/102.13.0
Subject: Re: [PATCH V2] asm-generic: ticket-lock: Optimize
 arch_spin_value_unlocked
Content-Language: en-US
To:     Waiman Long <longman@redhat.com>, guoren@kernel.org,
        David.Laight@ACULAB.COM, will@kernel.org, peterz@infradead.org,
        mingo@redhat.com
Cc:     linux-arch@vger.kernel.org, linux-kernel@vger.kernel.org,
        linux-riscv@lists.infradead.org, Guo Ren <guoren@linux.alibaba.com>
References: <20230731023308.3748432-1-guoren@kernel.org>
 <c603e7f1-a562-6826-1c86-995c8127abee@redhat.com>
From:   bibo mao <maobibo@loongson.cn>
In-Reply-To: <c603e7f1-a562-6826-1c86-995c8127abee@redhat.com>
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit
Precedence: bulk


在 2023/7/31 23:16, Waiman Long 写道:
> On 7/30/23 22:33, guoren@kernel.org wrote:
>> From: Guo Ren <guoren@linux.alibaba.com>
>>
>> The arch_spin_value_unlocked would cause an unnecessary memory
>> access to the contended value. Although it won't cause a significant
>> performance gap in most architectures, the arch_spin_value_unlocked
>> argument contains enough information. Thus, remove unnecessary
>> atomic_read in arch_spin_value_unlocked().
>>
>> The caller of arch_spin_value_unlocked() could benefit from this
>> change. Currently, the only caller is lockref.
>>
>> Signed-off-by: Guo Ren <guoren@kernel.org>
>> Cc: Waiman Long <longman@redhat.com>
>> Cc: David Laight <David.Laight@ACULAB.COM>
>> Cc: Peter Zijlstra <peterz@infradead.org>
>> Signed-off-by: Guo Ren <guoren@linux.alibaba.com>
>> ---
>> Changelog
>> V2:
>>   - Fixup commit log with Waiman advice.
>>   - Add Waiman comment in the commit msg.
>> ---
>>   include/asm-generic/spinlock.h | 16 +++++++++-------
>>   1 file changed, 9 insertions(+), 7 deletions(-)
>>
>> diff --git a/include/asm-generic/spinlock.h b/include/asm-generic/spinlock.h
>> index fdfebcb050f4..90803a826ba0 100644
>> --- a/include/asm-generic/spinlock.h
>> +++ b/include/asm-generic/spinlock.h
>> @@ -68,11 +68,18 @@ static __always_inline void arch_spin_unlock(arch_spinlock_t *lock)
>>       smp_store_release(ptr, (u16)val + 1);
>>   }
>>   +static __always_inline int arch_spin_value_unlocked(arch_spinlock_t lock)
>> +{
>> +    u32 val = lock.counter;
>> +
>> +    return ((val >> 16) == (val & 0xffff));
>> +}
>> +
>>   static __always_inline int arch_spin_is_locked(arch_spinlock_t *lock)
>>   {
>> -    u32 val = atomic_read(lock);
>> +    arch_spinlock_t val = READ_ONCE(*lock);
>>   -    return ((val >> 16) != (val & 0xffff));
>> +    return !arch_spin_value_unlocked(val);
>>   }
>>     static __always_inline int arch_spin_is_contended(arch_spinlock_t *lock)
>> @@ -82,11 +89,6 @@ static __always_inline int arch_spin_is_contended(arch_spinlock_t *lock)
>>       return (s16)((val >> 16) - (val & 0xffff)) > 1;
>>   }
>>   -static __always_inline int arch_spin_value_unlocked(arch_spinlock_t lock)
>> -{
>> -    return !arch_spin_is_locked(&lock);
>> -}
>> -
>>   #include <asm/qrwlock.h>
>>     #endif /* __ASM_GENERIC_SPINLOCK_H */
> 
> I am fine with the current change. However, modern optimizing compiler should be able to avoid the redundant memory read anyway. So this patch may not have an impact from the performance point of view.

arch_spin_value_unlocked is called with lockref like this:

#define CMPXCHG_LOOP(CODE, SUCCESS) do {                                        \
        int retry = 100;                                                        \
        struct lockref old;                                                     \
        BUILD_BUG_ON(sizeof(old) != 8);                                         \
        old.lock_count = READ_ONCE(lockref->lock_count);                        \
        while (likely(arch_spin_value_unlocked(old.lock.rlock.raw_lock))) {     \

With modern optimizing compiler, Is it possible that old value of 
old.lock.rlock.raw_lock is cached in register, despite that try_cmpxchg64_relaxed
modifies the memory of old.lock_count with new value? 

Regards
Bibo Mao

                struct lockref new = old;                                       \
                CODE                                                            \
                if (likely(try_cmpxchg64_relaxed(&lockref->lock_count,          \
                                                 &old.lock_count,               \
                                                 new.lock_count))) {            \
                        SUCCESS;                                                \
                }                                                               \
                if (!--retry)                                                   \
                        break;                                                  \
        }                                                                       \
} while (0)

> 
> Acked-by: Waiman Long <longman@redhat.com>