[PATCH 4.4 01/64] ring-buffer: Fix recursion protection transitions between interrupt context

[Greg Kroah-Hartman <gregkh@linuxfoundation.org>]

From: Steven Rostedt (VMware) <rostedt@goodmis.org>

[ Upstream commit b02414c8f045ab3b9afc816c3735bc98c5c3d262 ]

The recursion protection of the ring buffer depends on preempt_count() to be
correct. But it is possible that the ring buffer gets called after an
interrupt comes in but before it updates the preempt_count(). This will
trigger a false positive in the recursion code.

Use the same trick from the ftrace function callback recursion code which
uses a "transition" bit that gets set, to allow for a single recursion for
to handle transitions between contexts.

Cc: stable@vger.kernel.org
Fixes: 567cd4da54ff4 ("ring-buffer: User context bit recursion checking")
Signed-off-by: Steven Rostedt (VMware) <rostedt@goodmis.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
---
 kernel/trace/ring_buffer.c | 54 +++++++++++++++++++++++++++++++-------
 1 file changed, 44 insertions(+), 10 deletions(-)

diff --git a/kernel/trace/ring_buffer.c b/kernel/trace/ring_buffer.c
index 1c1ecc1d49ad2..547a3a5ac57b5 100644
--- a/kernel/trace/ring_buffer.c
+++ b/kernel/trace/ring_buffer.c
@@ -416,14 +416,16 @@ struct rb_event_info {
 
 /*
  * Used for which event context the event is in.
- *  NMI     = 0
- *  IRQ     = 1
- *  SOFTIRQ = 2
- *  NORMAL  = 3
+ *  TRANSITION = 0
+ *  NMI     = 1
+ *  IRQ     = 2
+ *  SOFTIRQ = 3
+ *  NORMAL  = 4
  *
  * See trace_recursive_lock() comment below for more details.
  */
 enum {
+	RB_CTX_TRANSITION,
 	RB_CTX_NMI,
 	RB_CTX_IRQ,
 	RB_CTX_SOFTIRQ,
@@ -2585,10 +2587,10 @@ rb_wakeups(struct ring_buffer *buffer, struct ring_buffer_per_cpu *cpu_buffer)
  * a bit of overhead in something as critical as function tracing,
  * we use a bitmask trick.
  *
- *  bit 0 =  NMI context
- *  bit 1 =  IRQ context
- *  bit 2 =  SoftIRQ context
- *  bit 3 =  normal context.
+ *  bit 1 =  NMI context
+ *  bit 2 =  IRQ context
+ *  bit 3 =  SoftIRQ context
+ *  bit 4 =  normal context.
  *
  * This works because this is the order of contexts that can
  * preempt other contexts. A SoftIRQ never preempts an IRQ
@@ -2611,6 +2613,30 @@ rb_wakeups(struct ring_buffer *buffer, struct ring_buffer_per_cpu *cpu_buffer)
  * The least significant bit can be cleared this way, and it
  * just so happens that it is the same bit corresponding to
  * the current context.
+ *
+ * Now the TRANSITION bit breaks the above slightly. The TRANSITION bit
+ * is set when a recursion is detected at the current context, and if
+ * the TRANSITION bit is already set, it will fail the recursion.
+ * This is needed because there's a lag between the changing of
+ * interrupt context and updating the preempt count. In this case,
+ * a false positive will be found. To handle this, one extra recursion
+ * is allowed, and this is done by the TRANSITION bit. If the TRANSITION
+ * bit is already set, then it is considered a recursion and the function
+ * ends. Otherwise, the TRANSITION bit is set, and that bit is returned.
+ *
+ * On the trace_recursive_unlock(), the TRANSITION bit will be the first
+ * to be cleared. Even if it wasn't the context that set it. That is,
+ * if an interrupt comes in while NORMAL bit is set and the ring buffer
+ * is called before preempt_count() is updated, since the check will
+ * be on the NORMAL bit, the TRANSITION bit will then be set. If an
+ * NMI then comes in, it will set the NMI bit, but when the NMI code
+ * does the trace_recursive_unlock() it will clear the TRANSTION bit
+ * and leave the NMI bit set. But this is fine, because the interrupt
+ * code that set the TRANSITION bit will then clear the NMI bit when it
+ * calls trace_recursive_unlock(). If another NMI comes in, it will
+ * set the TRANSITION bit and continue.
+ *
+ * Note: The TRANSITION bit only handles a single transition between context.
  */
 
 static __always_inline int
@@ -2629,8 +2655,16 @@ trace_recursive_lock(struct ring_buffer_per_cpu *cpu_buffer)
 	} else
 		bit = RB_CTX_NORMAL;
 
-	if (unlikely(val & (1 << bit)))
-		return 1;
+	if (unlikely(val & (1 << bit))) {
+		/*
+		 * It is possible that this was called by transitioning
+		 * between interrupt context, and preempt_count() has not
+		 * been updated yet. In this case, use the TRANSITION bit.
+		 */
+		bit = RB_CTX_TRANSITION;
+		if (val & (1 << bit))
+			return 1;
+	}
 
 	val |= (1 << bit);
 	cpu_buffer->current_context = val;
-- 
2.27.0