Re: Raw spinlocks and memory allocation

[Matthew Wilcox]

On Fri, Jul 31, 2020 at 03:30:16PM -0700, Paul E. McKenney wrote:
> On Fri, Jul 31, 2020 at 02:29:19PM -0700, Andrew Morton wrote:
> > On Fri, 31 Jul 2020 14:24:57 -0700 "Paul E. McKenney" <paulmck@xxxxxxxxxx> wrote:
> > 
> > > The reason for this restriction is that in -rt, the spin_lock(&zone->lock)
> > > in rmqueue_bulk() can sleep.
> > 
> > So if there is runtime overhead, this overhead could be restricted to
> > -rt kernels with suitable ifdefs?
> 
> In theory, yes.  In practice, with CONFIG_PROVE_RAW_LOCK_NESTING=y,
> lockdep will complain regardless of -rt or not.

On non-RT, we could make that lock a raw spinlock.  On RT, we could
decline to take the lock.  We'd need to abstract the spin_lock() away
behind zone_lock(zone), but that should be OK.

But let's see if we need to do that.