Messages in this thread | | | Date | Fri, 8 Feb 2002 13:16:28 -0800 | From | Mike Fedyk <> | Subject | Re: [RFC] New locking primitive for 2.5 |
| |
On Fri, Feb 08, 2002 at 03:04:34PM -0500, Jeff Garzik wrote: > Linus Torvalds wrote: > > > > On Fri, 8 Feb 2002, Ingo Molnar wrote: > > > > > > and regarding the reintroduction of BKL, *please* do not just use a global > > > locks around such pieces of code, lock bouncing sucks on SMP, even if > > > there is no overhead. > > > > I'd suggest not having a lock at all, but instead add two functions: one > > to read a 64-bit value atomically, the other to write it atomically (and > > they'd be atomic only wrt each other, no memory barriers etc implied). > > > > On 64-bit architectures that's just a direct dereference, and even on x86 > > it's just a "cmpxchg8b". > > Are there architectures out there that absolutely must implement this > with a spinlock? Your suggested API of functions to read/write 64-bit > values atomically would work for such a case, but still I am just > curious. >
SMP 486s would need that (if there is such a beast). What point does x86 get the 64 bit instructions? If after 586, then it would definately need a spinlock or somesuch in those functions.
Mike - To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
| |