Re: [PATCH 0/3] 64-bit futexes: Intro

[Ulrich Drepper <drepper@redhat.com>]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Linus Torvalds wrote:
> Is it possible? I dunno.

Once again not without limitations which are too low.  The main problem
is that there is far more information needed on the reader side.  There
are two counters and a number of bits for state information.  Which
means the counters can be at most 14 bits in size.

A counter this low also means I cannot unconditionally increment them
(since there can be more than 2^14 threads sharing one page).  This
means you get in the cache line race I described in the last mail.  You
first have to load a word and then perform a the write operation while
at the same time other threads competing with you for the same cache line.


> I personally suspect it is. I also suspect you 
> didn't even try.

I have to disappoint you.  I thought of that and had to reject it.

- --
➧ Ulrich Drepper ➧ Red Hat, Inc. ➧ 444 Castro St ➧ Mountain View, CA ❖
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.9 (GNU/Linux)

iEYEARECAAYFAkhA1j8ACgkQ2ijCOnn/RHTUXQCgnkUy6PNSuUI4O6BD4abxqqKr
PzMAn2qcvrXGM4qrhOytk1Yie0BsMeVa
=zvxZ
-----END PGP SIGNATURE-----