Re: [PATCH] one of $BIGNUM devfs races

[Richard Gooch]

Anton Altaparmakov writes:
> If you are introducing a spinlock for the sole purpose of protecting
> a counter, I would suggest to drop the spinlock, make the counter an
> atomic_t and just use atomic operations on it. That should be both
> faster and generate shorter code.

Firstly, I don't see an atomic_get_and_inc(). Sure, I can atomically
increment, and I can atomically read. But I can't read and increment
atomically.

Secondly, the range is 24 bits. While 24 bits is "probably enough",
I'd prefer not to waste bits.

Finally, a spinlock is less code (0) and faster on UP.

				Regards,

					Richard....
Permanent: [e-mail]
Current:   [e-mail]