xref: /openbmc/linux/lib/nmi_backtrace.c (revision 5d0e4d78)
1 /*
2  *  NMI backtrace support
3  *
4  * Gratuitously copied from arch/x86/kernel/apic/hw_nmi.c by Russell King,
5  * with the following header:
6  *
7  *  HW NMI watchdog support
8  *
9  *  started by Don Zickus, Copyright (C) 2010 Red Hat, Inc.
10  *
11  *  Arch specific calls to support NMI watchdog
12  *
13  *  Bits copied from original nmi.c file
14  */
15 #include <linux/cpumask.h>
16 #include <linux/delay.h>
17 #include <linux/kprobes.h>
18 #include <linux/nmi.h>
19 #include <linux/cpu.h>
20 #include <linux/sched/debug.h>
21 
22 #ifdef arch_trigger_cpumask_backtrace
23 /* For reliability, we're prepared to waste bits here. */
24 static DECLARE_BITMAP(backtrace_mask, NR_CPUS) __read_mostly;
25 
26 /* "in progress" flag of arch_trigger_cpumask_backtrace */
27 static unsigned long backtrace_flag;
28 
29 /*
30  * When raise() is called it will be passed a pointer to the
31  * backtrace_mask. Architectures that call nmi_cpu_backtrace()
32  * directly from their raise() functions may rely on the mask
33  * they are passed being updated as a side effect of this call.
34  */
35 void nmi_trigger_cpumask_backtrace(const cpumask_t *mask,
36 				   bool exclude_self,
37 				   void (*raise)(cpumask_t *mask))
38 {
39 	int i, this_cpu = get_cpu();
40 
41 	if (test_and_set_bit(0, &backtrace_flag)) {
42 		/*
43 		 * If there is already a trigger_all_cpu_backtrace() in progress
44 		 * (backtrace_flag == 1), don't output double cpu dump infos.
45 		 */
46 		put_cpu();
47 		return;
48 	}
49 
50 	cpumask_copy(to_cpumask(backtrace_mask), mask);
51 	if (exclude_self)
52 		cpumask_clear_cpu(this_cpu, to_cpumask(backtrace_mask));
53 
54 	/*
55 	 * Don't try to send an NMI to this cpu; it may work on some
56 	 * architectures, but on others it may not, and we'll get
57 	 * information at least as useful just by doing a dump_stack() here.
58 	 * Note that nmi_cpu_backtrace(NULL) will clear the cpu bit.
59 	 */
60 	if (cpumask_test_cpu(this_cpu, to_cpumask(backtrace_mask)))
61 		nmi_cpu_backtrace(NULL);
62 
63 	if (!cpumask_empty(to_cpumask(backtrace_mask))) {
64 		pr_info("Sending NMI from CPU %d to CPUs %*pbl:\n",
65 			this_cpu, nr_cpumask_bits, to_cpumask(backtrace_mask));
66 		raise(to_cpumask(backtrace_mask));
67 	}
68 
69 	/* Wait for up to 10 seconds for all CPUs to do the backtrace */
70 	for (i = 0; i < 10 * 1000; i++) {
71 		if (cpumask_empty(to_cpumask(backtrace_mask)))
72 			break;
73 		mdelay(1);
74 		touch_softlockup_watchdog();
75 	}
76 
77 	/*
78 	 * Force flush any remote buffers that might be stuck in IRQ context
79 	 * and therefore could not run their irq_work.
80 	 */
81 	printk_safe_flush();
82 
83 	clear_bit_unlock(0, &backtrace_flag);
84 	put_cpu();
85 }
86 
87 bool nmi_cpu_backtrace(struct pt_regs *regs)
88 {
89 	static arch_spinlock_t lock = __ARCH_SPIN_LOCK_UNLOCKED;
90 	int cpu = smp_processor_id();
91 
92 	if (cpumask_test_cpu(cpu, to_cpumask(backtrace_mask))) {
93 		arch_spin_lock(&lock);
94 		if (regs && cpu_in_idle(instruction_pointer(regs))) {
95 			pr_warn("NMI backtrace for cpu %d skipped: idling at pc %#lx\n",
96 				cpu, instruction_pointer(regs));
97 		} else {
98 			pr_warn("NMI backtrace for cpu %d\n", cpu);
99 			if (regs)
100 				show_regs(regs);
101 			else
102 				dump_stack();
103 		}
104 		arch_spin_unlock(&lock);
105 		cpumask_clear_cpu(cpu, to_cpumask(backtrace_mask));
106 		return true;
107 	}
108 
109 	return false;
110 }
111 NOKPROBE_SYMBOL(nmi_cpu_backtrace);
112 #endif
113