Date: Tue, 17 Oct 2000 12:21:50 +1100 From: Peter Jeremy <peter.jeremy@alcatel.com.au> To: Terry Lambert <tlambert@primenet.com> Cc: freebsd-arch@freebsd.org Subject: Re: we need atomic_t Message-ID: <00Oct17.122151est.115206@border.alcanet.com.au>
next in thread | raw e-mail | index | archive | help
[Apologies if this is a duplicate, I haven't seen the first copy come
back via the mailing list]
On Fri, Oct 13, 2000 at 02:19:20AM +0000, Terry Lambert wrote:
>[ ... atomic_t ... ]
>
>> My unspoken minimum precision was going to be 24 bits, for situations
>> where that wasn't enough the idea was to provide a atomic64_t, but
>> only if the demand was reasonable.
>
>How would you handle this type on 386, 486, and Pentium machines,
>if somone used it in code?
The Pentium and later CPUs have a compare-double-and-swap. This could
be used inside a loop to produce atomic operations on 64-bit objects
(much the same as the Alpha atomic routines). The basic code sequence
is:
atomic_op64:
movl mem,%eax
movl mem+4,%edx
loop:
op_q {%ecx:%ebx} = {%edx:%eax} op {operand}
lock cmpxchg8b mem
jne loop
Sticking the whole thing inside di/ei reduces the load/store window,
but may or may not be necessary. Adding a spin count might also be
worthwhile.
It isn't possible to have atomic 64-bit operations on 386 or 486, but
there was a previous agreement that FreeBSD wouldn't support SMP on
either of these CPU's either. Since (AFAIK) no bus-master peripherals
would need atomic 64-bit operations, protecting it with a
disable/enable interrupts block should be adequate.
Peter
To Unsubscribe: send mail to majordomo@FreeBSD.org
with "unsubscribe freebsd-arch" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?00Oct17.122151est.115206>