Date:   Mon, 29 Jul 2019 10:07:34 +0900
From:   Eiichi Tsukata <devel@...ukata.com>
To:     joel@...lfernandes.org, paulmck@...ux.vnet.ibm.com,
        tglx@...utronix.de, peterz@...radead.org, rostedt@...dmis.org,
        mingo@...hat.com, fweisbec@...il.com, luto@...capital.net,
        linux-kernel@...r.kernel.org
Cc:     Eiichi Tsukata <devel@...ukata.com>
Subject: [PATCH] tracing: Prevent RCU EQS breakage in preemptirq events

If context tracking is enabled, causing page fault in preemptirq
irq_enable or irq_disable events triggers the following RCU EQS warning.

Reproducer:

  // CONFIG_PREEMPTIRQ_EVENTS=y
  // CONFIG_CONTEXT_TRACKING=y
  // CONFIG_RCU_EQS_DEBUG=y
  # echo 1 > events/preemptirq/irq_disable/enable
  # echo 1 > options/userstacktrace

  WARNING: CPU: 0 PID: 2574 at kernel/rcu/tree.c:262 rcu_dynticks_eqs_exit+0x48/0x50
  CPU: 0 PID: 2574 Comm: sh Not tainted 5.3.0-rc1+ #105
  RIP: 0010:rcu_dynticks_eqs_exit+0x48/0x50
  Call Trace:
   rcu_eqs_exit+0x4e/0xd0
   rcu_user_exit+0x13/0x20
   __context_tracking_exit.part.0+0x74/0x120
   context_tracking_exit.part.0+0x28/0x50
   context_tracking_exit+0x1d/0x20
   do_page_fault+0xab/0x1b0
   do_async_page_fault+0x35/0xb0
   async_page_fault+0x3e/0x50
  RIP: 0010:arch_stack_walk_user+0x8e/0x100
   stack_trace_save_user+0x7d/0xa9
   trace_buffer_unlock_commit_regs+0x178/0x220
   trace_event_buffer_commit+0x6b/0x200
   trace_event_raw_event_preemptirq_template+0x7b/0xc0
   trace_hardirqs_off_caller+0xb3/0xf0
   trace_hardirqs_off_thunk+0x1a/0x20
   entry_SYSCALL_64_after_hwframe+0x3e/0xbe

Details of call trace and RCU EQS/Context:

  entry_SYSCALL_64_after_hwframe() EQS: IN, CTX: USER
    trace_irq_disable_rcuidle()
      rcu_irq_enter_irqson()
        rcu_dynticks_eqs_exit() EQS: IN => OUT
      stack_trace_save_user() EQS: OUT, CTX: USER
        page_fault()
          do_page_fault()
            exception_enter() EQS: OUT, CTX: USER
              context_tracking_exit()
                rcu_eqs_exit()
		  rcu_dynticks_eqs_exit() EQS: OUT => OUT? (warning)

trace_irq_disable/enable_rcuidle() are called from user mode in entry
code, and calls rcu_irq_enter_irqson() in __DO_TRACE(). This can cause
the state "RCU ESQ: OUT but CTX: USER", then stack_trace_save_user() can
cause page fault which calls rcu_dynticks_eqs_exit() again leads to hit
the EQS validation warning if CONFIG_RCU_EQS_DEBUG is enabled.

Fix it by calling exception_enter/exit() around
trace_irq_disable/enable_rcuidle() to enter CONTEXT_KERNEL before
tracing code causes page fault. Also makes the timing of state change to
CONTEXT_KERNL earlier to prevent tracing codes from calling
context_tracking_exit() recursively.

Ideally, the problem can be fixed by calling enter_from_user_mode() before
TRACE_IRQS_OFF in entry codes (then we need to tell lockdep that IRQs are
off eariler) and calling prepare_exit_to_usermode() after TRACE_IRQS_ON.
But this patch will be much simpler and limit the most change to tracing
codes.

Fixes: 865e63b04e9b ("tracing: Add back in rcu_irq_enter/exit_irqson() for rcuidle tracepoints")
Signed-off-by: Eiichi Tsukata <devel@...ukata.com>
---
 kernel/context_tracking.c       |  6 +++++-
 kernel/trace/trace_preemptirq.c | 15 +++++++++++++--
 2 files changed, 18 insertions(+), 3 deletions(-)

diff --git a/kernel/context_tracking.c b/kernel/context_tracking.c
index be01a4d627c9..860eaf9780e5 100644
--- a/kernel/context_tracking.c
+++ b/kernel/context_tracking.c
@@ -148,6 +148,11 @@ void __context_tracking_exit(enum ctx_state state)
 		return;
 
 	if (__this_cpu_read(context_tracking.state) == state) {
+		/*
+		 * Change state before executing codes which can trigger
+		 * page fault leading unnecessary re-entrance.
+		 */
+		__this_cpu_write(context_tracking.state, CONTEXT_KERNEL);
 		if (__this_cpu_read(context_tracking.active)) {
 			/*
 			 * We are going to run code that may use RCU. Inform
@@ -159,7 +164,6 @@ void __context_tracking_exit(enum ctx_state state)
 				trace_user_exit(0);
 			}
 		}
-		__this_cpu_write(context_tracking.state, CONTEXT_KERNEL);
 	}
 	context_tracking_recursion_exit();
 }
diff --git a/kernel/trace/trace_preemptirq.c b/kernel/trace/trace_preemptirq.c
index 4d8e99fdbbbe..031b51cb94d0 100644
--- a/kernel/trace/trace_preemptirq.c
+++ b/kernel/trace/trace_preemptirq.c
@@ -10,6 +10,7 @@
 #include <linux/module.h>
 #include <linux/ftrace.h>
 #include <linux/kprobes.h>
+#include <linux/context_tracking.h>
 #include "trace.h"
 
 #define CREATE_TRACE_POINTS
@@ -49,9 +50,14 @@ NOKPROBE_SYMBOL(trace_hardirqs_off);
 
 __visible void trace_hardirqs_on_caller(unsigned long caller_addr)
 {
+	enum ctx_state prev_state;
+
 	if (this_cpu_read(tracing_irq_cpu)) {
-		if (!in_nmi())
+		if (!in_nmi()) {
+			prev_state = exception_enter();
 			trace_irq_enable_rcuidle(CALLER_ADDR0, caller_addr);
+			exception_exit(prev_state);
+		}
 		tracer_hardirqs_on(CALLER_ADDR0, caller_addr);
 		this_cpu_write(tracing_irq_cpu, 0);
 	}
@@ -63,11 +69,16 @@ NOKPROBE_SYMBOL(trace_hardirqs_on_caller);
 
 __visible void trace_hardirqs_off_caller(unsigned long caller_addr)
 {
+	enum ctx_state prev_state;
+
 	if (!this_cpu_read(tracing_irq_cpu)) {
 		this_cpu_write(tracing_irq_cpu, 1);
 		tracer_hardirqs_off(CALLER_ADDR0, caller_addr);
-		if (!in_nmi())
+		if (!in_nmi()) {
+			prev_state = exception_enter();
 			trace_irq_disable_rcuidle(CALLER_ADDR0, caller_addr);
+			exception_exit(prev_state);
+		}
 	}
 
 	lockdep_hardirqs_off(CALLER_ADDR0);
-- 
2.21.0

Powered by blists - more mailing lists