x86, mm, perf: Allow recursive faults from interrupts
authorPeter Zijlstra <[email protected]>
Fri, 10 Jan 2014 20:06:03 +0000 (21:06 +0100)
committerIngo Molnar <[email protected]>
Thu, 16 Jan 2014 08:19:48 +0000 (09:19 +0100)
Waiman managed to trigger a PMI while in a emulate_vsyscall() fault,
the PMI in turn managed to trigger a fault while obtaining a stack
trace. This triggered the sig_on_uaccess_error recursive fault logic
and killed the process dead.

Fix this by explicitly excluding interrupts from the recursive fault
logic.

Reported-and-Tested-by: Waiman Long <[email protected]>
Fixes: e00b12e64be9 ("perf/x86: Further optimize copy_from_user_nmi()")
Cc: Aswin Chandramouleeswaran <[email protected]>
Cc: Scott J Norton <[email protected]>
Cc: Linus Torvalds <[email protected]>
Cc: Andy Lutomirski <[email protected]>
Cc: Arnaldo Carvalho de Melo <[email protected]>
Cc: Andrew Morton <[email protected]>
Signed-off-by: Peter Zijlstra <[email protected]>
Link: http://lkml.kernel.org/r/[email protected]
Signed-off-by: Ingo Molnar <[email protected]>
arch/x86/mm/fault.c

index 9ff85bb8dd698182fa6605ee565a00f7843e6303..9d591c895803101e2decbc85a0ce9f23a0b4eaeb 100644 (file)
@@ -641,6 +641,20 @@ no_context(struct pt_regs *regs, unsigned long error_code,
 
        /* Are we prepared to handle this kernel fault? */
        if (fixup_exception(regs)) {
+               /*
+                * Any interrupt that takes a fault gets the fixup. This makes
+                * the below recursive fault logic only apply to a faults from
+                * task context.
+                */
+               if (in_interrupt())
+                       return;
+
+               /*
+                * Per the above we're !in_interrupt(), aka. task context.
+                *
+                * In this case we need to make sure we're not recursively
+                * faulting through the emulate_vsyscall() logic.
+                */
                if (current_thread_info()->sig_on_uaccess_error && signal) {
                        tsk->thread.trap_nr = X86_TRAP_PF;
                        tsk->thread.error_code = error_code | PF_USER;
@@ -649,6 +663,10 @@ no_context(struct pt_regs *regs, unsigned long error_code,
                        /* XXX: hwpoison faults will set the wrong code. */
                        force_sig_info_fault(signal, si_code, address, tsk, 0);
                }
+
+               /*
+                * Barring that, we can do the fixup and be happy.
+                */
                return;
        }