Messages in this thread | | | Date | Tue, 30 Nov 1999 22:50:21 +0100 (CET) | From | Ingo Molnar <> | Subject | Re: [patch] new spinlock variant, spinlock-2.3.30-A4 |
| |
On Tue, 30 Nov 1999, Davide Libenzi wrote:
> > movb $1, 3(%%esi) # spin_lock > > movl 4(%%esi), %%edx > > addl (%%esi), %%edx > > cmpl $0x01000000, %%edx > > je slow_path > > This can lead to some dead loops, due to the fact that an atomic ( locked ) > test and set is not used. > The lock acquire is based on the assumption that only one CPU execute the > lock code.
The spinlock is correct with an arbitrary number of CPUs.
> In theory if there are more than one CPUs executing the code, they can > spend a lot of loops without getting the N bytes at zero. This > probability is higher greater is the number of CPUs. Using a test and > set instruction, even if N CPUs execute the same code at the same > time, only one gets the lock value.
yes, true. Keep in mind that this only affects the slow path. The above spinlock would be implemented as a function anyway (we do not want to inline it), and in that case the slow path can eg. use queued spinlocks (spinlock chains through on-stack variables) or exponential backoff, or whatever technique.
-- mingo
- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.rutgers.edu Please read the FAQ at http://www.tux.org/lkml/
| |