[Devel] [RFC PATCH vz9 v6 14/62] dm-ploop: move preparations of pios into the caller from worker

[Alexander Atanasov]

On 13.12.24 11:21, Alexey Kuznetsov wrote:
> [Some people who received this message don't often get email from alexey.n.kuznetsov at gmail.com. Learn why this is important at https://aka.ms/LearnAboutSenderIdentification ]
> 
> Hello!
> 
> Frankly, I would seriosly consider dropping off llist. It is just an opinion.
> 
> This braindead construct is a lie, it is _not_ lockless at all, it
> uses the slowest possible
> way of synchronization (cmpxchg) and still it has damn complicated set
> of usage rules.
> Personally I would never use it. Either crystal clear and not less
> efficient dumb construct:
> 
> struct lockedlist  {
>          spinlock_t      lock;
>          struct hlist_head       list;
> }
> 
> or when I am ready to go berserk, something really lockless, without cmpxchg and
> with full *mb() nightmare.


the *mb() Andrey pointed out is a leftover from a different thing not 
related to llist.

There might be better options but llist was the way to do small steps 
forward as requested. There are a lot of things that can be improved 
besides lists. I will keep this in mind when doing qcow2.


-- 
Regards,
Alexander Atanasov