diff --git a/arch/x86/mm/tlb.c b/arch/x86/mm/tlb.c
index e491c75b2a68..3c81edd54c5c 100644
--- a/arch/x86/mm/tlb.c
+++ b/arch/x86/mm/tlb.c
@@ -621,7 +621,11 @@ void switch_mm_irqs_off(struct mm_struct *unused, struct mm_struct *next,
 
 		choose_new_asid(next, next_tlb_gen, &new_asid, &need_flush);
 
-		/* Let nmi_uaccess_okay() know that we're changing CR3. */
+		/*
+		 * Indicate that CR3 is about to change. nmi_uaccess_okay()
+		 * and others are sensitive to the window where mm_cpumask(),
+		 * CR3 and cpu_tlbstate.loaded_mm are not all in sync.
+ 		 */
 		this_cpu_write(cpu_tlbstate.loaded_mm, LOADED_MM_SWITCHING);
 		barrier();
 	}
@@ -895,8 +899,16 @@ static void flush_tlb_func(void *info)
 
 static bool should_flush_tlb(int cpu, void *data)
 {
+	struct mm_struct *loaded_mm = per_cpu(cpu_tlbstate.loaded_mm, cpu);
 	struct flush_tlb_info *info = data;
 
+	/*
+	 * Order the 'loaded_mm' and 'is_lazy' against their
+	 * write ordering in switch_mm_irqs_off(). Ensure
+	 * 'is_lazy' is at least as new as 'loaded_mm'.
+	 */
+	smp_rmb();
+
 	/* Lazy TLB will get flushed at the next context switch. */
 	if (per_cpu(cpu_tlbstate_shared.is_lazy, cpu))
 		return false;
@@ -905,8 +917,15 @@ static bool should_flush_tlb(int cpu, void *data)
 	if (!info->mm)
 		return true;
 
+	/*
+	 * While switching, the remote CPU could have state from
+	 * either the prev or next mm. Assume the worst and flush.
+	 */
+	if (loaded_mm == LOADED_MM_SWITCHING)
+		return true;
+
 	/* The target mm is loaded, and the CPU is not lazy. */
-	if (per_cpu(cpu_tlbstate.loaded_mm, cpu) == info->mm)
+	if (loaded_mm == info->mm)
 		return true;
 
 	/* In cpumask, but not the loaded mm? Periodically remove by flushing. */