Re: [RFC/PoC] malloc: use wfcqueue to speed up remote frees

[Mathieu Desnoyers via Libc-alpha <libc-alpha@sourceware.org>]

On 2023-01-17 14:05, Mathieu Desnoyers wrote:
> On 2023-01-17 01:42, Eric Wong wrote:
>> Carlos O'Donell <carlos@redhat.com> wrote:
>>>>> - Adding urcu as a build-time dependency is not acceptable for
>>>>> bootstrap, instead we would bundle a copy of urcu and keep it
>>>>> in sync with upstream. Would that make your work easier?
>>>
>>> Eric Wong wrote:
>>>> Yes, bundling that sounds great.  I assume it's something for
>>>> you or one of the regular contributors to work on (build systems
>>>> scare me :x)
>>>
>>> Yes, that is something we'd have to do.
>>
>> Hi, bringing this topic from 2018 up again (+Cc Mathieu):
>> https://inbox.sourceware.org/libc-alpha/c061de55-cc2a-88fe-564b-2ea9c4a7e632@redhat.com/T/
>>
>> I'm wondering if URCU-in-glibc is still on the table.  I'm also
>> considering an learning C11 atomics and deriving a standalone
>> wfcqueue w/o URCU atomics.
> 
> Hi Eric,
> 
> I'm very much interested to contribute portions of liburcu to glibc. I 
> think what we would need at this point is to document where we see that 
> liburcu infrastructure can improve glibc, and create a gradual 
> integration roadmap based on priorities. We should identify the 
> stake-holders interested in seeing this done, and then we can discuss 
> the time-frame and resources available for this project.
> 
> That being said, I suspect we'd want to cover a few pieces of technology 
> in this list, namely:
> 
> - Restartable Sequences (rseq system call),
> - membarrier system call,
> - liburcu and libside [1,2] RCU implementations,
> - liburcu data structures (e.g. wfcqueue).

Here is a PoC implementing liburcu wfcqueue with C11 atomics:

https://review.lttng.org/c/userspace-rcu/+/9271 PoC: wfcqueue: remove dependencies on liburcu headers

Feedback is welcome!

Thanks,

Mathieu

> 
> I would be tempted to go for an approach closer to the RCU 
> implementation I have done in the libside project for glibc, because it 
> supports having multiple RCU grace period domains per process, which 
> allows to nicely split the locking dependencies into sub-domains, and 
> therefore lessen the chances of deadlocks due to nesting of a 
> global-domain RCU read-side/synchronize_rcu, and locking (e.g. mutexes).
> 
> Thanks,
> 
> Mathieu
> 
> [1] https://github.com/efficios/libside/blob/master/src/rcu.c
> [2] https://github.com/efficios/libside/blob/master/src/rcu.h
> 

-- 
Mathieu Desnoyers
EfficiOS Inc.
https://www.efficios.com