Messages in this thread | | | Date | Mon, 01 Jul 2013 09:40:29 -0400 | From | Waiman Long <> | Subject | Re: [PATCH v2 1/2] spinlock: New spinlock_refcount.h for lockless update of refcount |
| |
On 06/29/2013 06:47 PM, Linus Torvalds wrote: > On Sat, Jun 29, 2013 at 2:58 PM, Linus Torvalds > <torvalds@linux-foundation.org> wrote: >> STEP 1: >> >> - create a new set of config option (say >> "CONFIG_[ARCH_]SPINLOCK_REFCOUNT") that defaults to 'y' and isn't >> actually asked about anywhere. > That "defauls to 'y'" was incomplete/misleading. The > ARCH_SPINLOCK_REFCOUNT config option should default to 'n' (the > example code snippet did that correctly: a bool config option defaults > to 'n' unless stated otherwise). > > The SPINLOCK_REFCOUNT in turn _should_ have a "default y" (the sample > snippet I posted didn't have that), but that 'y' is then obviously > normally disabled by the "depends on" logic, so it only takes effect > when ARCH_SPINLOCK_REFCOUNT is set, and none of the spinlock debug > things are enabled. > > I hope that was all fairly obvious from the concept (and the real > examples of us already doing things like this with the whole > CONFIG_GENERIC_STRNLEN_USER and CONFIG_DCACHE_WORD_ACCESS examples), > but I thought I'd follow up with an explicit correction, since both > the explanation and sample snippets were somewhat misleading. > > Linus
Thank for the clarification, but I had already understood the implication of your suggestion. Each architecture has to explicitly opt-in by defining an ARCH_SPINLOCK_REFCOUNT in its Kconfig file to have this lockref feature enabled. In this way, each architecture can take its time to verify it works before enabling it. My patch will enable enable the x86 as I don't have test machines for the other architectures.
Regards, Longman
| |