Re: [PATCH v2 3/5] locking/qspinlock: Introduce CNA into the slow path of qspinlock
From: Waiman Long
Date: Wed Apr 03 2019 - 13:40:25 EST
On 04/03/2019 01:16 PM, Peter Zijlstra wrote:
> On Wed, Apr 03, 2019 at 12:33:20PM -0400, Waiman Long wrote:
>> static inline void queued_spin_lock_slowpath(struct qspinlock *lock, u32
>> ÂÂÂÂÂÂÂ if (static_branch_unlikely(&use_numa_spinlock))
>> ÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂ numa_queued_spin_lock_slowpath(lock, val);
>> ÂÂÂÂÂÂÂ elseÂÂÂ
>> ÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂ native_queued_spin_lock_slowpath(lock, val);
> That's horrible for the exact reason you state.
>> Alternatively, we can also call numa_queued_spin_lock_slowpath() in
>> native_queued_spin_lock_slowpath() if we don't want to increase the code
>> size of spinlock call sites.
> Yeah, still don't much like that though, we're littering the fast path
> of that slow path with all sorts of crap.
Yes, I know it is less than ideal, but that is probably the only option
if we don't have static_call or paravirt. On the other hand, I am
perfectly fine with making CNA depends on PARAVIRT_SPINLOCKS for now
until static_call is available.