Message-ID: <f00bff72-c923-2552-f181-f1bcc034bff9@redhat.com>
Date:   Mon, 23 Jan 2023 16:10:02 -0500
MIME-Version: 1.0
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:102.0) Gecko/20100101
 Thunderbird/102.6.0
Subject: Re: [PATCH v6 0/6] lockinig/rwsem: Fix rwsem bugs & enable true lock
 handoff
Content-Language: en-US
From:   Waiman Long <longman@redhat.com>
To:     Peter Zijlstra <peterz@infradead.org>
Cc:     Ingo Molnar <mingo@redhat.com>, Will Deacon <will@kernel.org>,
        Boqun Feng <boqun.feng@gmail.com>,
        linux-kernel@vger.kernel.org, john.p.donnelly@oracle.com,
        Hillf Danton <hdanton@sina.com>,
        Mukesh Ojha <quic_mojha@quicinc.com>,
        =?UTF-8?B?VGluZzExIFdhbmcg546L5am3?= <wangting11@xiaomi.com>
References: <20221118022016.462070-1-longman@redhat.com>
 <Y80+SLxrdEVyJHMo@hirez.programming.kicks-ass.net>
 <8eea871c-eed0-cb4b-1b81-3a6b502c3c34@redhat.com>
In-Reply-To: <8eea871c-eed0-cb4b-1b81-3a6b502c3c34@redhat.com>
Content-Type: text/plain; charset=UTF-8; format=flowed
Content-Transfer-Encoding: 8bit
Precedence: bulk

On 1/22/23 22:40, Waiman Long wrote:
>
> On 1/22/23 08:46, Peter Zijlstra wrote:
>> On Thu, Nov 17, 2022 at 09:20:10PM -0500, Waiman Long wrote:
>>
>>> Waiman Long (6):
>>>    locking/rwsem: Prevent non-first waiter from spinning in 
>>> down_write() slowpath
>>>    locking/rwsem: Disable preemption at all down_read*() and 
>>> up_read() code paths
>>>    locking/rwsem: Disable preemption at all down_write*() and 
>>> up_write() code paths
>>>    locking/rwsem: Change waiter->hanodff_set to a handoff_state enum
>> After all these we still have down_read_non_owner() using
>> __rwsem_set_reader_owner() outside of the preempt_disable() region.
>>
>> Now, let me go stare at this one:
>
> Thanks for spotting that. I kind of overlook the fact we have a 
> down_read_non_owner(). Will update that as well.

After looking further into this, we will need to either pass one more 
parameter to __down_read() for this special case or to do 
preempt_disable/enable in each of the higher level down_read functions. 
As down_read_non_owner() is a rarely called function, I doubt it is 
worth the extra effort to do that since the owner value (other than the 
RWSEM_READER_OWNED bit) is for debugging purpose only and is not 
critical to the correct functioning of a rwsem. I will add a comment to 
note that.

Cheers,
Longman