Re: [PATCH -mm -V7] mm, swap: fix race between swapoff and some swap operations

["Huang\, Ying" <ying.huang@intel.com>]

Andrea Arcangeli <aarcange@redhat.com> writes:

> On Thu, Feb 14, 2019 at 04:07:37PM +0800, Huang, Ying wrote:
>> Before, we choose to use stop_machine() to reduce the overhead of hot
>> path (page fault handler) as much as possible.  But now, I found
>> rcu_read_lock_sched() is just a wrapper of preempt_disable().  So maybe
>> we can switch to RCU version now.
>
> rcu_read_lock looks more efficient than rcu_read_lock_sched. So for
> this purpose in the fast path rcu_read_lock()/unlock() should be the
> preferred methods, no need to force preempt_disable() (except for
> debug purposes if sleep debug is enabled). Server builds are done with
> voluntary preempt (no preempt shouldn't even exist as config option)
> and there rcu_read_lock might be just a noop.

If

CONFIG_PREEMPT_VOLUNTARY=y
CONFIG_PREEMPT=n
CONFIG_PREEMPT_COUNT=n

which is common for servers,

rcu_read_lock() will be a noop, rcu_read_lock_sched() and
preempt_disable() will be barrier().  So rcu_read_lock() is a little
better.

> Against a fast path rcu_read_lock/unlock before the consolidation
> synchronize_rcu would have been enough, now after the consolidation
> even more certain that it's enough because it's equivalent to _mult.

Yes.  Will change to rcu_read_lock/unlock based method.

Best Regards,
Huang, Ying