[Devel] [RFC PATCH vz9 v6 07/62] dm-ploop: convert wait_list and wb_batch_llist to use lockless lists

[Pavel Tikhomirov]

On 1/14/25 22:20, Alexander Atanasov wrote:
> On 9.01.25 7:00, Pavel Tikhomirov wrote:
>>
>>
>> On 12/6/24 05:55, Alexander Atanasov wrote:
>>> diff --git a/drivers/md/dm-ploop-map.c b/drivers/md/dm-ploop-map.c
>>> index a0fe92a592a1..fb49dc7bd7b0 100644
>>> --- a/drivers/md/dm-ploop-map.c
>>> +++ b/drivers/md/dm-ploop-map.c
>>> @@ -382,13 +382,14 @@ static bool ploop_delay_if_md_busy(struct ploop 
>>> *ploop, struct md_page *md,
>>>       WARN_ON_ONCE(!list_empty(&pio->list));
>>> -    write_lock_irqsave(&ploop->bat_rwlock, flags);
>>> +    /* lock protects piwb */
>>> +    read_lock_irqsave(&ploop->bat_rwlock, flags);
>>>       piwb = md->piwb;
>>>       if (piwb && (piwb->type != type || test_bit(MD_WRITEBACK, &md- 
>>> >status))) {
>>> -        list_add_tail(&pio->list, &md->wait_list);
>>> +        llist_add((struct llist_node *)(&pio->list), &md->wait_llist);
>>
>> Probably we can have:
>>
>> struct pio {
>>    union {
>>      struct list_head list;
>>      struct llist_node llist;
>>    };
>> }
>>
>> And use llist member where we need lock-less lists and use list where 
>> we want doubly linked list.
> 
> 
> tried that - did not work.
> compiler errors from list macros, they do type checking - there is a 
> third struct too - struct llist_head.

That is really strange, if you use right name from the union in right 
places all type checkings should be satisfied, but maybe I miss something.

> 
> 

-- 
Best regards, Tikhomirov Pavel
Senior Software Developer, Virtuozzo.