panic: decrease oops_in_progress only after having done the panic

Impact: eliminate secondary warnings during panic()

We can panic() in a number of difficult, atomic contexts, hence
we use bust_spinlocks(1) in panic() to increase oops_in_progress,
which prevents various debug checks we have in place.

But in practice this protection only covers the first few printk's
done by panic() - it does not cover the later attempt to stop all
other CPUs and kexec(). If a secondary warning triggers in one of
those facilities that can make the panic message scroll off.

So do bust_spinlocks(0) only much later in panic(). (which code
is only reached if panic policy is relaxed that it can return
after a warning message)

Reported-by: Jan Beulich <[email protected]>
LKML-Reference: <49B91A7E[email protected]>
Signed-off-by: Ingo Molnar <[email protected]>

diff --git a/kernel/panic.c b/kernel/panic.c
index 32fe4eff1b89ecf75af7c229fc8db2e46179c924..57fb005de546e46d454590c1246edb8921bcef44 100644 (file)
--- a/kernel/panic.c
+++ b/kernel/panic.c
@@ -77,7 +77,6 @@ NORET_TYPE void panic(const char * fmt, ...)
 #ifdef CONFIG_DEBUG_BUGVERBOSE
        dump_stack();
 #endif
-       bust_spinlocks(0);
 
        /*
         * If we have crashed and we have a crash kernel loaded let it handle
@@ -136,6 +135,7 @@ NORET_TYPE void panic(const char * fmt, ...)
                mdelay(1);
                i++;
        }
+       bust_spinlocks(0);
 }
 
 EXPORT_SYMBOL(panic);