[PATCH] rcu: Fix unrecovered RCU user mode in syscall_trace_leave()
From: Frederic Weisbecker
Date: Fri Oct 26 2012 - 05:40:27 EST
On x86-64 syscall exit, 3 non exclusive events may happen
looping in the following order:
1) Check if we need resched for user preemption, if so call
schedule_user()
2) Check if we have pending signals, if so call do_notify_resume()
3) Check if we do syscall tracing, if so call syscall_trace_leave()
However syscall_trace_leave() has been written assuming it directly
follows the syscall and forget about the above possible 1st and 2nd
steps.
Now schedule_user() and do_notify_resume() exit in RCU user mode
because they have most chances to resume userspace immediately and
this avoids an rcu_user_enter() call in the syscall fast path.
So by the time we call syscall_trace_leave(), we may well be in RCU
user mode. To fix this up, simply call rcu_user_exit() in the beginning
of this function.
This fixes some reported RCU uses in extended quiescent state.
Reported-by: Dave Jones <davej@xxxxxxxxxx>
Reported-by: Sergey Senozhatsky <sergey.senozhatsky@xxxxxxxxx>
Signed-off-by: Frederic Weisbecker <fweisbec@xxxxxxxxx>
---
arch/x86/kernel/ptrace.c | 7 +++++++
1 files changed, 7 insertions(+), 0 deletions(-)
diff --git a/arch/x86/kernel/ptrace.c b/arch/x86/kernel/ptrace.c
index b00b33a..eff5b8c 100644
--- a/arch/x86/kernel/ptrace.c
+++ b/arch/x86/kernel/ptrace.c
@@ -1511,6 +1511,13 @@ void syscall_trace_leave(struct pt_regs *regs)
{
bool step;
+ /*
+ * We may come here right after calling schedule_user()
+ * or do_notify_resume(), in which case we can be in RCU
+ * user mode.
+ */
+ rcu_user_exit();
+
audit_syscall_exit(regs);
if (unlikely(test_thread_flag(TIF_SYSCALL_TRACEPOINT)))
--
1.7.5.4
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/