Re: [PATCH-tip v7 00/20] locking/rwsem: Rwsem rearchitecture part 2

[Waiman Long]

On 4/28/19 6:46 PM, Linus Torvalds wrote:
> This doesn't seem to be the full diff - looking at that patch 1 you
> seem to have taken my suggested list_cut_before() change too.
>
> I'm not against it (it does seem to be simpler and better), I just
> hope you double-checked it, since I kind of hand-waved it.
>
>                     Linus

I implemented your suggestion in patch 1 as it will produce simpler and
faster code. However, one of the changes in my patchset is to wake up
all the readers in the wait list. This means I have to jump over the
writers and wake up the readers behind them as well. See patch 11 for
details. As a result, I have to revert back to use list_add_tail() and
list_for_each_entry_safe() for the first pass. That is why the diff for
the whole patchset is just the below change. It is done on purpose, not
an omission.

Cheers,
Longman

>
> On Sun, Apr 28, 2019 at 2:26 PM Waiman Long <longman@xxxxxxxxxx> wrote:
>>  v6=>v7 diff
>>  -----------
>> diff --git a/kernel/locking/rwsem.c b/kernel/locking/rwsem.c
>> index 97a2334d9cd3..60783267b50d 100644
>> --- a/kernel/locking/rwsem.c
>> +++ b/kernel/locking/rwsem.c
>> @@ -693,7 +693,7 @@ static void __rwsem_mark_wake(struct rw_semaphore *sem,
>>                 atomic_long_add(adjustment, &sem->count);
>>
>>         /* 2nd pass */
>> -       list_for_each_entry(waiter, &wlist, list) {
>> +       list_for_each_entry_safe(waiter, tmp, &wlist, list) {
>>                 struct task_struct *tsk;
>>
>>                 tsk = waiter->task;
>>