[PATCH] malloc: Improve malloc initialization

[DJ Delorie]

Wilco Dijkstra <Wilco.Dijkstra@arm.com> writes:
>>> Move malloc initialization out of hot paths - it is only required after checking
>>> fastbins in _int_malloc or before tcache initialization. Bench-malloc-thread
>>> improves by 1.8% for 1 thread and 1.3% for 32 threads on Neoverse V2.
>>
>> ptmalloc_init sets up the tunables needed to properly manage tcache and
>> fastbins...
>
> And that's fine because one can't enter tcache code when it has not been
> initialized yet. The same is true for the fastbins - global_max_fast is zero
> intialized, so we can't enter the fastbin case until initialization has been run.
>
> Or would you prefer calling ptmalloc_init explicitly during startup?

I think I brought that up in the distant past but don't recall the
reason.  My concern is that ptmalloc_init does a *lot* and malloc is
complicated; if we're going to defer calling it we need to be really
sure that anything we do access is in some stable state without
ptmalloc_init being called.  I mean, do we even know if there will be
any arenas to have an uninitialized fastbin in?