MySQL Bugs: #97228: rwlock: refine lock->lock

Bug #97228	rwlock: refine lock->lock_word with C11 atomics
Submitted:	15 Oct 2019 10:40	Modified:	25 Oct 2019 10:41
Reporter:	Cai Yibo (OCA)	Email Updates:
Status:	Verified	Impact on me:	None
Category:	MySQL Server: InnoDB storage engine	Severity:	S5 (Performance)
Version:	8.0	OS:	Any
Assigned to:		CPU Architecture:	ARM
Tags:	Contribution

Description:
This is followup patch of Bug #96504(Refine atomics and barriers for weak memory order platform). It depends on the patch for Bug #97150.

lock_word is core of rwlock implementation. It's a signed integer
reflecting current locking status(S, SX, X).

The most important changes of this patch are locking and unlocking
code. Refine locking(rw_lock_lock_word_decr) with acquire order,
and unlocking(rw_lock_lock_word_incr) with release order.

Unlocking is followed by lock->waiters checking to signal waiting
threads. Incrementing lock->word must happen before checking
lock->waiters. After changing unlocking to release order, this
sequence is not guaranteed. So adding an explicit acquire fence
before loading lock->waiters to fix the problem.

Another improvement comes from removing of "volatile" definition,
which eliminates some unnecessary memory accesses in code.

How to repeat:
NA

patch (*) I confirm the code being submitted is offered under the terms of the OCA, and that I am authorized to contribute it.

Contribution: 0002-rwlock-refine-lock-lock_word-with-C-11-atomics.patch (text/x-patch), 17.74 KiB.

Better to review together with Bug #97230.

Hi Mr. Yibo,

Thank you for your bug report and contribution for improving our performance on the ARM CPU family.

Verified as reported.

Hi Sinisa,

Thanks for reviewing. Is there anything I need to improve for this patch?

Yibo

Hi Mr. Yibo,

We shall contact you as soon as we get some feedback from our Engineering.