xref: /openbmc/linux/lib/nmi_backtrace.c (revision 2d99a7ec)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  *  NMI backtrace support
4  *
5  * Gratuitously copied from arch/x86/kernel/apic/hw_nmi.c by Russell King,
6  * with the following header:
7  *
8  *  HW NMI watchdog support
9  *
10  *  started by Don Zickus, Copyright (C) 2010 Red Hat, Inc.
11  *
12  *  Arch specific calls to support NMI watchdog
13  *
14  *  Bits copied from original nmi.c file
15  */
16 #include <linux/cpumask.h>
17 #include <linux/delay.h>
18 #include <linux/kprobes.h>
19 #include <linux/nmi.h>
20 #include <linux/cpu.h>
21 #include <linux/sched/debug.h>
22 
23 #ifdef arch_trigger_cpumask_backtrace
24 /* For reliability, we're prepared to waste bits here. */
25 static DECLARE_BITMAP(backtrace_mask, NR_CPUS) __read_mostly;
26 
27 /* "in progress" flag of arch_trigger_cpumask_backtrace */
28 static unsigned long backtrace_flag;
29 
30 /*
31  * When raise() is called it will be passed a pointer to the
32  * backtrace_mask. Architectures that call nmi_cpu_backtrace()
33  * directly from their raise() functions may rely on the mask
34  * they are passed being updated as a side effect of this call.
35  */
36 void nmi_trigger_cpumask_backtrace(const cpumask_t *mask,
37 				   bool exclude_self,
38 				   void (*raise)(cpumask_t *mask))
39 {
40 	int i, this_cpu = get_cpu();
41 
42 	if (test_and_set_bit(0, &backtrace_flag)) {
43 		/*
44 		 * If there is already a trigger_all_cpu_backtrace() in progress
45 		 * (backtrace_flag == 1), don't output double cpu dump infos.
46 		 */
47 		put_cpu();
48 		return;
49 	}
50 
51 	cpumask_copy(to_cpumask(backtrace_mask), mask);
52 	if (exclude_self)
53 		cpumask_clear_cpu(this_cpu, to_cpumask(backtrace_mask));
54 
55 	/*
56 	 * Don't try to send an NMI to this cpu; it may work on some
57 	 * architectures, but on others it may not, and we'll get
58 	 * information at least as useful just by doing a dump_stack() here.
59 	 * Note that nmi_cpu_backtrace(NULL) will clear the cpu bit.
60 	 */
61 	if (cpumask_test_cpu(this_cpu, to_cpumask(backtrace_mask)))
62 		nmi_cpu_backtrace(NULL);
63 
64 	if (!cpumask_empty(to_cpumask(backtrace_mask))) {
65 		pr_info("Sending NMI from CPU %d to CPUs %*pbl:\n",
66 			this_cpu, nr_cpumask_bits, to_cpumask(backtrace_mask));
67 		nmi_backtrace_stall_snap(to_cpumask(backtrace_mask));
68 		raise(to_cpumask(backtrace_mask));
69 	}
70 
71 	/* Wait for up to 10 seconds for all CPUs to do the backtrace */
72 	for (i = 0; i < 10 * 1000; i++) {
73 		if (cpumask_empty(to_cpumask(backtrace_mask)))
74 			break;
75 		mdelay(1);
76 		touch_softlockup_watchdog();
77 	}
78 	nmi_backtrace_stall_check(to_cpumask(backtrace_mask));
79 
80 	/*
81 	 * Force flush any remote buffers that might be stuck in IRQ context
82 	 * and therefore could not run their irq_work.
83 	 */
84 	printk_trigger_flush();
85 
86 	clear_bit_unlock(0, &backtrace_flag);
87 	put_cpu();
88 }
89 
90 // Dump stacks even for idle CPUs.
91 static bool backtrace_idle;
92 module_param(backtrace_idle, bool, 0644);
93 
94 bool nmi_cpu_backtrace(struct pt_regs *regs)
95 {
96 	int cpu = smp_processor_id();
97 	unsigned long flags;
98 
99 	if (cpumask_test_cpu(cpu, to_cpumask(backtrace_mask))) {
100 		/*
101 		 * Allow nested NMI backtraces while serializing
102 		 * against other CPUs.
103 		 */
104 		printk_cpu_sync_get_irqsave(flags);
105 		if (!READ_ONCE(backtrace_idle) && regs && cpu_in_idle(instruction_pointer(regs))) {
106 			pr_warn("NMI backtrace for cpu %d skipped: idling at %pS\n",
107 				cpu, (void *)instruction_pointer(regs));
108 		} else {
109 			pr_warn("NMI backtrace for cpu %d\n", cpu);
110 			if (regs)
111 				show_regs(regs);
112 			else
113 				dump_stack();
114 		}
115 		printk_cpu_sync_put_irqrestore(flags);
116 		cpumask_clear_cpu(cpu, to_cpumask(backtrace_mask));
117 		return true;
118 	}
119 
120 	return false;
121 }
122 NOKPROBE_SYMBOL(nmi_cpu_backtrace);
123 #endif
124