Re: [PATCH] mm: might_sleep warning

[Andrew Morton]

On Tue, 6 Mar 2018 15:48:26 -0500 Pavel Tatashin <pasha.tatashin@xxxxxxxxxx> wrote:

> On Tue, Mar 6, 2018 at 3:36 PM, Andrew Morton <akpm@xxxxxxxxxxxxxxxxxxxx>
> wrote:
> 
> > On Tue,  6 Mar 2018 14:20:22 -0500 Pavel Tatashin <
> > pasha.tatashin@xxxxxxxxxx> wrote:
> >
> > >       spin_lock(&deferred_zone_grow_lock);
> > > -     static_branch_disable(&deferred_pages);
> > > +     deferred_zone_grow = false;
> > >       spin_unlock(&deferred_zone_grow_lock);
> > > +     static_branch_disable(&deferred_pages);
> > >
> > >       /* There will be num_node_state(N_MEMORY) threads */
> > >       atomic_set(&pgdat_init_n_undone, num_node_state(N_MEMORY));
> >
> > Kinda ugly, but I can see the logic behind the decisions.
> >
> > Can we instead turn deferred_zone_grow_lock into a mutex?

(top-posting repaired.  Please don't top-post).

> [CCed everyone]
> 
> Hi Andrew,
> 
> I afraid we cannot change this spinlock to mutex
> because deferred_grow_zone() might be called from an interrupt context if
> interrupt thread needs to allocate memory.
> 

OK.  But if deferred_grow_zone() can be called from interrupt then
page_alloc_init_late() should be using spin_lock_irq(), shouldn't it? 
I'm surprised that lockdep didn't detect that.


--- a/mm/page_alloc.c~mm-initialize-pages-on-demand-during-boot-fix-4-fix
+++ a/mm/page_alloc.c
@@ -1689,9 +1689,9 @@ void __init page_alloc_init_late(void)
 	 * context. Since, spin_lock() disables preemption, we must use an
 	 * extra boolean deferred_zone_grow.
 	 */
-	spin_lock(&deferred_zone_grow_lock);
+	spin_lock_irq(&deferred_zone_grow_lock);
 	deferred_zone_grow = false;
-	spin_unlock(&deferred_zone_grow_lock);
+	spin_unlock_irq(&deferred_zone_grow_lock);
 	static_branch_disable(&deferred_pages);
 
 	/* There will be num_node_state(N_MEMORY) threads */
_