From: Guo Ren

Sent: 30 March 2021 04:14

...

quoted

Step 1 would be to get your architecute fixed such that it can provide
fwd progress guarantees for LL/SC. Otherwise there's absolutely no point
in building complex systems with it.

Quote Waiman's comment [1] on xchg16 optimization:

"This optimization is needed to make the qspinlock achieve performance
parity with ticket spinlock at light load."

[1] https://lore.kernel.org/kvm/1429901803-29771-6-git-send-email-Waiman.Long@hp.com/ (local)

So for a non-xhg16 machine:
 - ticket-lock for small numbers of CPUs
 - qspinlock for large numbers of CPUs

Okay, I'll put all of them into the next patch :P

Doesn't that also imply that you need ticket-locks for
lightly contended locks even with a lot of CPUs?

If you actually get a lot of CPUs waiting on the same spinlock
you probably ought to change the code to reduce lock contention.

	David

-
Registered Address Lakeside, Bramley Road, Mount Farm, Milton Keynes, MK1 1PT, UK
Registration No: 1397386 (Wales)