Re: [parisc-linux] [patch 15/23] Add cmpxchg_local to parisc
David Miller <[email protected]> Mon, 27 Aug 2007 14:20:17 -0700 (PDT)
| Newsgroups | gmane.linux.kernel,gmane.linux.ports.hppa |
|---|---|
| Message-ID | <[email protected]> |
From: Grant Grundler <[email protected]> Date: Mon, 27 Aug 2007 15:04:32 -0600 > Doesn't explain the difference between "local" and "non-local" either. > Per CPU data should only need memory barriers (in some cases) and > protection against interrupts (in probably more cases). So I'm not > understanding why a new set of APIs is needed. > > Can you add a description to Documentation/atomic_ops.txt ? My understanding is that the local versions can run without memory barriers. Only local cpu execution and local interrupt threads of control can access the variable, and interrupts force all stores to become visible to the local cpu so an interrupt can't see the datum from before the atomic update if one was in progress when the interrupt triggered, and stuff like that.