Messages in this thread | | | From | Arnd Bergmann <> | Subject | Re: [PATCH 1/2] lkdtm: use atomic_t to replace count_lock | Date | Thu, 2 Feb 2012 14:55:38 +0000 |
| |
On Thursday 02 February 2012, Cong Wang wrote: > On 02/02/2012 09:44 PM, Arnd Bergmann wrote: > > On Thursday 02 February 2012, Cong Wang wrote: > >>> In order to have an atomic here, you have to use a loop around > >>> atomic_cmpxchg, like > >>> > >>> > >>> int old, new; > >>> old = atomic_read(&count); > >>> do { > >>> new = old ? old - 1 : cpoint_count; > >>> old = cmpxchg(&count, old, new);
^^^^^^^ I guess I meant "new = cmpxchg(...)" here, sorry.
> >>> } while (old != new); > >>> > >>> I suppose you could also just keep the spinlock and move lkdtm_do_action() > >>> outside of it? > >> > >> If we still need spinlock, I think we don't need to bother atomic_t at all. > > > > Yes, it's one or the other: If you use the cmpxchg loop, you don't need a > > spinlock and vice versa. > > > > The cmpxchg loop is for comparing and assigning to 'count', but still > there is a printk() above that needs to read 'count'. Combining these > two operations means we have to use a spinlock, correct? Because there > is a chance that another process could change 'count' in between.
No, you can just print the value of "old" in the above example, which was atomically read.
Arnd
| |