Re: Increase PTHREAD_KEYS_MAX

[Paul Goyette <paul%vps1.whooppee.com@localhost>]

On Wed, 13 May 2015, Emmanuel Dreyfus wrote:

> On Wed, May 13, 2015 at 02:02:26PM +0200, Joerg Sonnenberger wrote:
> > > - fix _POSIX_THREAD_KEYS_MAX to be 128 instad of 256
> > This doesn't make sense. It should reflect the actual limit.
>
> The actual limit is PTHREAD_KEYS_MAX. _POSIX_THREAD_KEYS_MAX
> is the constant defined in the standard.
>
> > > - increase PTHREAD_KEYS_MAX to 2048 so that we lead again for some time
> > >   this would increase memory footprint of programs linked with -lpthread
> > >   of 14 kB on ILP32 systems, and 28 kB on LP64 systems.
> >
> > I am against raising it that much. The overhead is *per thread*, so it
> > sums up easily.
>
> This is not my understanding of both the code and the documentation.
> pthread_key_create(3) saus:
>     The pthread_key_create() function creates a thread-specific data key
>     visible to all threads in the process.
>
> The data attached to a pthread_key_t is thread-specific. The array is
> not.
>
> > A queue doesn't make sense, O(1) is really required here. On-demand
> > growing of the array shouldn't be too hard to implement though.
>
> I am not sure if we can use realloc() there. Even if we can, it means
> we have to call malloc() on first usage.

I would guess that you wouldn't have to reallocate very often.  Could 
you maybe do something like what the SYSV* code does?  Allocate one big 
chunk of memory, and if you have to grow it, you allocate a bigger 
chunk, migrate all the old stuff to the new chunk (one entry at a time), 
and then free the old chunk?

Yup, it's not elegant, but it would not be a frequent occurrence.  You 
simply need to use an "appropriate" increment.



-------------------------------------------------------------------------
| Paul Goyette     | PGP Key fingerprint:     | E-mail addresses:       |
| (Retired)        | FA29 0E3B 35AF E8AE 6651 | paul at whooppee.com    |
| Kernel Developer | 0786 F758 55DE 53BA 7731 | pgoyette at netbsd.org  |
-------------------------------------------------------------------------