s390/mm,tlb: race of lazy TLB flush vs. recreation of TLB entries

Git commit 050eef364ad70059 "[S390] fix tlb flushing vs. concurrent
/proc accesses" introduced the attach counter to avoid using the
mm_users value to decide between IPTE for every PTE and lazy TLB
flushing with IDTE. That fixed the problem with mm_users but it
introduced another subtle race, fortunately one that is very hard
to hit.
The background is the requirement of the architecture that a valid
PTE may not be changed while it can be used concurrently by another
cpu. The decision between IPTE and lazy TLB flushing needs to be
done while the PTE is still valid. Now if the virtual cpu is
temporarily stopped after the decision to use lazy TLB flushing but
before the invalid bit of the PTE has been set, another cpu can attach
the mm, find that flush_mm is set, do the IDTE, return to userspace,
and recreate a TLB that uses the PTE in question. When the first,
stopped cpu continues it will change the PTE while it is attached on
another cpu. The first cpu will do another IDTE shortly after the
modification of the PTE which makes the race window quite short.

To fix this race the CPU that wants to attach the address space of a
user space thread needs to wait for the end of the PTE modification.
The number of concurrent TLB flushers for an mm is tracked in the
upper 16 bits of the attach_count and finish_arch_post_lock_switch
is used to wait for the end of the flush operation if required.

Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>

1 files changed, 34 insertions, 5 deletions

diff --git a/arch/s390/include/asm/mmu_context.h b/arch/s390/include/asm/mmu_context.h
index 5d1f950704dc..38149b63dc44 100644
--- a/arch/s390/include/asm/mmu_context.h
+++ b/arch/s390/include/asm/mmu_context.h
@@ -48,13 +48,42 @@ static inline void update_mm(struct mm_struct *mm, struct task_struct *tsk)
 static inline void switch_mm(struct mm_struct *prev, struct mm_struct *next,
 			     struct task_struct *tsk)
 {
-	cpumask_set_cpu(smp_processor_id(), mm_cpumask(next));
-	update_mm(next, tsk);
+	int cpu = smp_processor_id();
+
+	if (prev == next)
+		return;
+	if (atomic_inc_return(&next->context.attach_count) >> 16) {
+		/* Delay update_mm until all TLB flushes are done. */
+		set_tsk_thread_flag(tsk, TIF_TLB_WAIT);
+	} else {
+		cpumask_set_cpu(cpu, mm_cpumask(next));
+		update_mm(next, tsk);
+		if (next->context.flush_mm)
+			/* Flush pending TLBs */
+			__tlb_flush_mm(next);
+	}
 	atomic_dec(&prev->context.attach_count);
 	WARN_ON(atomic_read(&prev->context.attach_count) < 0);
-	atomic_inc(&next->context.attach_count);
-	/* Check for TLBs not flushed yet */
-	__tlb_flush_mm_lazy(next);
+}
+
+#define finish_arch_post_lock_switch finish_arch_post_lock_switch
+static inline void finish_arch_post_lock_switch(void)
+{
+	struct task_struct *tsk = current;
+	struct mm_struct *mm = tsk->mm;
+
+	if (!test_tsk_thread_flag(tsk, TIF_TLB_WAIT))
+		return;
+	preempt_disable();
+	clear_tsk_thread_flag(tsk, TIF_TLB_WAIT);
+	while (atomic_read(&mm->context.attach_count) >> 16)
+		cpu_relax();
+
+	cpumask_set_cpu(smp_processor_id(), mm_cpumask(mm));
+	update_mm(mm, tsk);
+	if (mm->context.flush_mm)
+		__tlb_flush_mm(mm);
+	preempt_enable();
 }
 
 #define enter_lazy_tlb(mm,tsk)	do { } while (0)