s390/mm: fix race on mm->context.flush_mm

The order in __tlb_flush_mm_lazy is to flush TLB first and then clear
the mm->context.flush_mm bit. This can lead to missed flushes as the
bit can be set anytime, the order needs to be the other way aronud.

But this leads to a different race, __tlb_flush_mm_lazy may be called
on two CPUs concurrently. If mm->context.flush_mm is cleared first then
another CPU can bypass __tlb_flush_mm_lazy although the first CPU has
not done the flush yet. In a virtualized environment the time until the
flush is finally completed can be arbitrarily long.

Add a spinlock to serialize __tlb_flush_mm_lazy and use the function
in finish_arch_post_lock_switch as well.

Cc: <stable@vger.kernel.org>
Reviewed-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>

1 files changed, 2 insertions, 2 deletions

diff --git a/arch/s390/include/asm/mmu_context.h b/arch/s390/include/asm/mmu_context.h
index 8823e35f69a9..484efe8f4234 100644
--- a/arch/s390/include/asm/mmu_context.h
+++ b/arch/s390/include/asm/mmu_context.h
@@ -17,6 +17,7 @@
 static inline int init_new_context(struct task_struct *tsk,
 				   struct mm_struct *mm)
 {
+	spin_lock_init(&mm->context.lock);
 	spin_lock_init(&mm->context.pgtable_lock);
 	INIT_LIST_HEAD(&mm->context.pgtable_list);
 	spin_lock_init(&mm->context.gmap_lock);
@@ -121,8 +122,7 @@ static inline void finish_arch_post_lock_switch(void)
 		while (atomic_read(&mm->context.flush_count))
 			cpu_relax();
 		cpumask_set_cpu(smp_processor_id(), mm_cpumask(mm));
-		if (mm->context.flush_mm)
-			__tlb_flush_mm(mm);
+		__tlb_flush_mm_lazy(mm);
 		preempt_enable();
 	}
 	set_fs(current->thread.mm_segment);