News for hackers. Stuff that matters.
Re:The Fastest Mutexes by xxs on Thursday October 03, 2024 @02:59AM (#41727979)
Lots of code is 'lock free' and uses the same pattern (more or less). Attempt the change optimistically, if it fails spin it, if it keeps failing (Massive Contention), back off.
Re:The Fastest Mutexes by adastra22 on Thursday October 03, 2024 @05:39AM (#41728971)
A properly lock-free architecture wouldn’t have spin-locks, no?
Re:The Fastest Mutexes by pjdesno on Thursday October 03, 2024 @11:30AM (#41731684)
A properly lock-free architecture is a spin-lock :-)
Re:The Fastest Mutexes by xxs on Friday October 04, 2024 @05:28AM (#41739515)
spin locks prevent forward grantees, e.g. when the thread is scheduled out (or serves an interrupt), no other thread can make progress. Lock free allows progress - there is no exclusivity. Of course if working on the kernel and controlling the scheduler, etc. the restriction does not apply immediately.

Realistically though, lots of lock free employs copy-on-write or rcu.

Re:The Fastest Mutexes by xxs on Friday October 04, 2024 @05:31AM (#41739539)
of course, no lock.

I have written quite the fair share of lock free code (back then when it was new), and generally enjoy writing it (to this day). Yet, for most folks it's just too hard to reason/understand/maintain it.

Just a note, lock free is a lesser requirement than wait-free.