On Fri, 29 May 2015 14:26:14 -0700
Andrew Morton [off-list ref] wrote:

On Fri, 29 May 2015 10:48:15 -0500 Clark Williams [off-list ref] wrote:

quoted

The irqs_disabled() check in mem_cgroup_swapout() fails on the latest
RT kernel because RT mutexes do not disable interrupts when held. Change
the test for the lock being held to use spin_is_locked.

...

--- a/mm/memcontrol.c
+++ b/mm/memcontrol.c

@@ -5845,7 +5845,7 @@ void mem_cgroup_swapout(struct page *page,

swp_entry_t entry) page_counter_uncharge(&memcg->memory, 1);
 
 	/* XXX: caller holds IRQ-safe mapping->tree_lock */
-	VM_BUG_ON(!irqs_disabled());
+	VM_BUG_ON(!spin_is_locked(&page_mapping(page)->tree_lock));
 
 	mem_cgroup_charge_statistics(memcg, page, -1);
 	memcg_check_events(memcg, page);

spin_is_locked() returns zero on uniprocessor builds.  The results will
be unhappy.  

I suggest just deleting the check.

Guess this is Johannes call. We can just #ifdef it out and that would
remain the same when we finally merge PREEMPT_RT in mainline. 

If Johannes wants to keep the check on non-RT, here's a patch:

From: Clark Williams <redacted>
Date: Mon, 1 Jun 2015 13:10:39 -0500
Subject: [PATCH] mm: ifdef out VM_BUG_ON check on PREEMPT_RT_FULL

The irqs_disabled() check in mem_cgroup_swapout() fails on the latest
RT kernel because RT mutexes do not disable interrupts when held. Ifdef
this check out for PREEMPT_RT_FULL.

Signed-off-by: Clark Williams <redacted>
---
 mm/memcontrol.c | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/mm/memcontrol.c b/mm/memcontrol.c
index 9da0f3e9c1f3..f3fcef7713f6 100644
--- a/mm/memcontrol.c
+++ b/mm/memcontrol.c

@@ -5844,8 +5844,10 @@ void mem_cgroup_swapout(struct page *page, swp_entry_t entry)
 	if (!mem_cgroup_is_root(memcg))
 		page_counter_uncharge(&memcg->memory, 1);
 
+#ifndef CONFIG_PREEMPT_RT_FULL
 	/* XXX: caller holds IRQ-safe mapping->tree_lock */
 	VM_BUG_ON(!irqs_disabled());
+#endif
 
 	mem_cgroup_charge_statistics(memcg, page, -1);
 	memcg_check_events(memcg, page);

-- 
2.1.0