Re: [PATCH] locking/rwsem: Prevent non-first waiter from spinning in down_write() slowpath

[Waiman Long <longman@redhat.com>]

On 9/29/22 17:15, John Donnelly wrote:
> On 9/29/22 13:04, Waiman Long wrote:
>> A non-first waiter can potentially spin in the for loop of
>> rwsem_down_write_slowpath() without sleeping but fail to acquire the
>> lock even if the rwsem is free if the following sequence happens:
>>
>>    Non-first waiter       First waiter      Lock holder
>>    ----------------       ------------      -----------
>>    Acquire wait_lock
>>    rwsem_try_write_lock():
>>      Set handoff bit if RT or
>>        wait too long
>>      Set waiter->handoff_set
>>    Release wait_lock
>>                           Acquire wait_lock
>>                           Inherit waiter->handoff_set
>>                           Release wait_lock
>>                        Clear owner
>>                                             Release lock
>>    if (waiter.handoff_set) {
>>      rwsem_spin_on_owner(();
>>      if (OWNER_NULL)
>>        goto trylock_again;
>>    }
>>    trylock_again:
>>    Acquire wait_lock
>>    rwsem_try_write_lock():
>>       if (first->handoff_set && (waiter != first))
>>           return false;
>>    Release wait_lock
>>
>> It is especially problematic if the non-first waiter is an RT task and
>> it is running on the same CPU as the first waiter as this can lead to
>> live lock.
>>
>> Fixes: d257cc8cb8d5 ("locking/rwsem: Make handoff bit handling more 
>> consistent")
>> Signed-off-by: Waiman Long <longman@redhat.com>
>> ---
>>   kernel/locking/rwsem.c | 13 ++++++++++---
>>   1 file changed, 10 insertions(+), 3 deletions(-)
>>
>> diff --git a/kernel/locking/rwsem.c b/kernel/locking/rwsem.c
>> index 65f0262f635e..ad676e99e0b3 100644
>> --- a/kernel/locking/rwsem.c
>> +++ b/kernel/locking/rwsem.c
>> @@ -628,6 +628,11 @@ static inline bool rwsem_try_write_lock(struct 
>> rw_semaphore *sem,
>>           new = count;
>>             if (count & RWSEM_LOCK_MASK) {
>> +            /*
>> +             * A waiter (first or not) can set the handoff bit
>> +             * if it is an RT task or wait in the wait queue
>> +             * for too long.
>> +             */
>>               if (has_handoff || (!rt_task(waiter->task) &&
>>                           !time_after(jiffies, waiter->timeout)))
>>                   return false;
>> @@ -643,11 +648,13 @@ static inline bool rwsem_try_write_lock(struct 
>> rw_semaphore *sem,
>>       } while (!atomic_long_try_cmpxchg_acquire(&sem->count, &count, 
>> new));
>>         /*
>> -     * We have either acquired the lock with handoff bit cleared or
>> -     * set the handoff bit.
>> +     * We have either acquired the lock with handoff bit cleared or set
>> +     * the handoff bit. Only the first waiter can have its handoff_set
>> +     * set here to enable optimistic spinning in slowpath loop.
>>        */
>>       if (new & RWSEM_FLAG_HANDOFF) {
>> -        waiter->handoff_set = true;
>> +        if (waiter == first)
>> +            waiter->handoff_set = true;
>>           lockevent_inc(rwsem_wlock_handoff);
>>           return false;
>>       }
> Hi,.
>
> Are you missing
>
>
> [PATCH 5.18 87/88] locking/rwsem: Allow slowpath writer to ignore 
> handoff bit if not set by first waiter
>
>
>
> Fixes: d257cc8cb8d5 ("locking/rwsem: Make handoff bit handling more 
> consistent")
>
> Or is this another regression ?
>
It is another regression which.

Cheers,
Longman