From: | Bruce Momjian <maillist(at)candle(dot)pha(dot)pa(dot)us> |
---|---|
To: | teunis(at)mauve(dot)computersupportcentre(dot)com (teunis) |
Cc: | scrappy(at)hub(dot)org, pgsql-hackers(at)postgreSQL(dot)org |
Subject: | Re: [HACKERS] S_LOCK() change produces error... |
Date: | 1998-01-19 23:59:32 |
Message-ID: | 199801192359.SAA05291@candle.pha.pa.us |
Views: | Raw Message | Whole Thread | Download mbox | Resend email |
Thread: | |
Lists: | pgsql-hackers |
>
> On Sat, 17 Jan 1998, The Hermit Hacker wrote:
>
> > On Sat, 17 Jan 1998, Bruce Momjian wrote:
> > > > I installed some patches today for the univel port, and one of the changes
> > > > did the following to include/storage/s_lock.h:
> > > >
> > > > 302c318
> > > > < __asm__("xchgb %0,%1": "=q"(_res), "=m"(*lock):"0"(0x1)); \
> > > > ---
> > > > > __asm__("lock xchgb %0,%1": "=q"(_res), "=m"(*lock):"0"(0x1)); \
> > >
> > > I guess this is a multiple cpu modifier for asm, and most people don't
> > > run multiple cpus. I guess our gcc's call it an error, rather than
> > > ignore it. I think we need an OS-specific ifdef there. We can't have
> > > Univel changing the normal i386 stuff that works so well now.
> >
> > Actually, I think that the patch was meant to improve...if you look at the
> > code, he put all the Univel stuff inside of its own #ifdef...see around
> > line 297 in include/storage/s_lock.h and you'll see what I mean.
> >
> > He seems to have only added a 'lock' to the beginning of the __asm__,
> > which is what is breaking things under FreeBSD, but unless it affects every
> > other port, I'm loath to remove it without just throwing in a FreeBSD #ifdef
> > in there...
>
> (clip from SMP support in linux' asm/spinlocks.h)
> #define spin_unlock(lock) \
> __asm__ __volatile__( \
> "lock ; btrl $0,%0" \
> :"=m" (__dummy_lock(lock)))
>
> in linux the lock has ";" following.
> Yep - it's for multiCPU systems (SMP). Handy for shared-memory systems
> too if you're really into multithreading-speed.
>
> It locks that particular byte (word?) of memory against access by other
> CPU's accessing it IIRC...
>
> Perhaps your GAS is too old? (GNU binutils)
> (does BSD support multiple CPU's under intel?)
>
> multiprocessor really isn't that rare under linux - even Linus Torvalds
> uses a SMP system *grin*...
>
> Maybe he encountered a locking problem with a multicpu host and needed a
> semaphore (or equiv) to lock things? Just trying to figure this out...
> (sometimes necessary if you're doing shared memory across processes)
Marc, I will try 'lock;' and if it works, will submit a patch.
--
Bruce Momjian
maillist(at)candle(dot)pha(dot)pa(dot)us
From | Date | Subject | |
---|---|---|---|
Next Message | Bruce Momjian | 1998-01-20 05:03:09 | parser cleanup |
Previous Message | teunis | 1998-01-19 23:28:46 | Re: [HACKERS] S_LOCK() change produces error... |