Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::3:7 as permitted sender) client-ip=2620:137:e000::3:7;
Feedback-ID: id8a24192:Fastmail
Message-ID: <73e673d6-ecb8-dec9-bdc0-6dde9c4e76cb@fastmail.fm>
Date:   Tue, 19 Sep 2023 15:12:13 +0200
MIME-Version: 1.0
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:102.0) Gecko/20100101
 Thunderbird/102.15.1
Subject: Re: [PATCH] fuse: remove unneeded lock which protecting update of
 congestion_threshold
Content-Language: en-US, de-DE
To:     Kemeng Shi <shikemeng@huaweicloud.com>, miklos@szeredi.hu,
        linux-fsdevel@vger.kernel.org, linux-kernel@vger.kernel.org
References: <20230914154553.71939-1-shikemeng@huaweicloud.com>
 <9a5d4c82-1ab3-e96d-98bb-369acc8404d1@fastmail.fm>
 <177d891e-9258-68bb-72aa-4d4126403b7e@huaweicloud.com>
From:   Bernd Schubert <bernd.schubert@fastmail.fm>
In-Reply-To: <177d891e-9258-68bb-72aa-4d4126403b7e@huaweicloud.com>
Content-Type: text/plain; charset=UTF-8; format=flowed
Content-Transfer-Encoding: 8bit
Precedence: bulk


On 9/19/23 08:11, Kemeng Shi wrote:
> 
> 
> on 9/16/2023 7:06 PM, Bernd Schubert wrote:
>>
>>
>> On 9/14/23 17:45, Kemeng Shi wrote:
>>> Commit 670d21c6e17f6 ("fuse: remove reliance on bdi congestion") change how
>>> congestion_threshold is used and lock in
>>> fuse_conn_congestion_threshold_write is not needed anymore.
>>> 1. Access to supe_block is removed along with removing of bdi congestion.
>>> Then down_read(&fc->killsb) which protecting access to super_block is no
>>> needed.
>>> 2. Compare num_background and congestion_threshold without holding
>>> bg_lock. Then there is no need to hold bg_lock to update
>>> congestion_threshold.
>>>
>>> Signed-off-by: Kemeng Shi <shikemeng@huaweicloud.com>
>>> ---
>>>    fs/fuse/control.c | 4 ----
>>>    1 file changed, 4 deletions(-)
>>>
>>> diff --git a/fs/fuse/control.c b/fs/fuse/control.c
>>> index 247ef4f76761..c5d7bf80efed 100644
>>> --- a/fs/fuse/control.c
>>> +++ b/fs/fuse/control.c
>>> @@ -174,11 +174,7 @@ static ssize_t fuse_conn_congestion_threshold_write(struct file *file,
>>>        if (!fc)
>>>            goto out;
>>>    -    down_read(&fc->killsb);
>>> -    spin_lock(&fc->bg_lock);
>>>        fc->congestion_threshold = val;
>>> -    spin_unlock(&fc->bg_lock);
>>> -    up_read(&fc->killsb);
>>>        fuse_conn_put(fc);
>>>    out:
>>>        return ret;
>>
>> Yeah, I don't see readers holding any of these locks.
>> I just wonder if it wouldn't be better to use WRITE_ONCE to ensure a single atomic operation to store the value.
> Sure, WRITE_ONCE looks better. I wonder if we should use READ_ONCE from reader.
> Would like to get any advice. Thanks!

I'm not entirely sure either, but I _think_ the compiler is free to 
store a 32 bit value  with multiple operations (like 2 x 16 bit). In 
that case a competing reading thread might read garbage...
Although I don't see this documented here
https://www.kernel.org/doc/Documentation/memory-barriers.txt
Though documented there is that the compile is free to optimize out the 
storage at all, see
"(*) Similarly, the compiler is within its rights to omit a store entirely"


Regarding READ_ONCE, I don't have a strong opinion, if the compiler 
makes some optimizations and the value would be wrong for a few cycles, 
would that matter for that variable? Unless the compiler would be really 
creative and the variable would get never updated... For sure READ_ONCE 
would be safer, but I don't know if it is needed
SSee section
"The compiler is within its rights to omit a load entirely if it know"
in the document above.

Thanks,
Bernd