Re: [regression] boot failure on alpha, bisected

[Oleg Nesterov <oleg@redhat.com>]

On 10/07, Oleg Nesterov wrote:
>
> Hmm. I know nothing about arch/alpha and I can't understand its entry.S.
> But _it seems_ to me that do_notify_resume() is called with irqs disabled.
> If this is true, then imho arch/alpha should be fixed.
>
> Before this commit task_work_run() enabled irqs, but this was the "side
> effect" of spin_lock_irq/spin_unlock_irq, we should not rely on this.

Could you please test the debugging patch below?

Oleg.


--- x/arch/alpha/kernel/signal.c
+++ x/arch/alpha/kernel/signal.c
@@ -567,11 +567,19 @@ do_signal(struct pt_regs * regs, struct 
 		ptrace_set_bpt(current);	/* re-set breakpoint */
 }
 
+#include <linux/ratelimit.h>
+
 void
 do_notify_resume(struct pt_regs *regs, struct switch_stack *sw,
 		 unsigned long thread_info_flags,
 		 unsigned long r0, unsigned long r19)
 {
+	if (irqs_disabled()) {
+		printk_ratelimited(KERN_WARNING
+			"NOTIFY with irqs_disabled:%lx\n", thread_info_flags);
+		local_irq_enable();
+	}
+
 	if (thread_info_flags & _TIF_SIGPENDING)
 		do_signal(regs, sw, r0, r19);