Memory barriers

Poul-Henning Kamp phk at
Wed May 19 21:30:17 CEST 2010

In message <15639.1274170643 at>, "Poul-Henning Kamp" writes:
>In message <4BF24C0A.40508 at>, Nils Goroll writes:
>>Once you're at it, it would appear to me that adding differentiated macros for
>>load/store/both (AMD64: lfence/sfence/mfence) or even more specific macros was >a good idea.
>Only if I need them.

Ok, I just found out that Linux does not expose the various membar
macros to userland (there is a proposal for a syscall to do it ??)
So I bit the bullet and add the VMB VTLA and put a pthread-compat
version in it, since that will be cheaper than a syscall on any OS
with a decent pthread implementation.


PS: Please hand me the ability to edit POSIX standards unhindered for
just one day, and the world will be a better place, I promise...

Poul-Henning Kamp       | UNIX since Zilog Zeus 3.20
phk at FreeBSD.ORG         | TCP/IP since RFC 956
FreeBSD committer       | BSD since 4.3-tahoe    
Never attribute to malice what can adequately be explained by incompetence.

More information about the varnish-dev mailing list