Re: How could HP help Debian
On Sun, 2003-09-14 at 22:45, Matthew Wilcox wrote:
> [moving to debian-hppa, since that seems to be the best place. doing my
> best to remove all traces of the original mail since that's decidedly
> On Sun, Sep 14, 2003 at 10:30:18PM -0400, Grzegorz B. Prokopski wrote:
> > port "libffi" to HPPA if they want to see LGPL-ed SableVM JVM (and
> > probably also GCJ) working on these arches in near future [*].
> Randolph Chung has libffi working quite well. He's got copyright papers
> sorted recently, and I believe he's just waiting for a spare couple of
> hours to get it all submitted.
Oh - that's great news, really!
If we don't make a VMS port of SableVM this week - then we could at
least get it working on HPPA ;-)
> > PS: [*] And somebody tell me how to do atomic cmp&swap on HPPA - then
> > you have a working SableVM JVM on HPPA in 10 minutes. Seriously.
> You can't. hppa's only got one atomic instruction -- load-and-zero word.
> If you can use that as a locking primitive, ie lock read and write
> sides, then you're fine. If you can live with atomic assignment, you're
> also fine. But if you rely on something like cmpxchg() as a locking
> primitive, you're in trouble.
Hmm... I must say that I saw that *one* instruction long time ago, but
hoped that someone more knowledgable might know sth. better.
If there's nothing better - then I think I'll simply have to actively
loop [*] until I get the lock, then lock, do non-atomic cmp&swap, then
unlock and return whether it succeeded. Hmm... Should do, methinks.
So - when can I expect libffi w/ hppa support in Debian?
Apparently it's currently built from gcc sources and I don't know
whether gcc people are happy to quickly put up a new version w/
Randolph's improvements of libffi?
I should be able to prepare needed cmp&swap code meanwhile.
It would be ideally if someone could drop me a note when HPPA's libffi
becomes available in Debian.
Grzegorz B. Prokopski
[*] It may be better to give up processor to another thread though
so that it could release the lock we need.
/me noting that we don't do that for other arches...
Grzegorz B. Prokopski <email@example.com>
Debian GNU/Linux http://www.debian.org