Don Zickus | 1d48922 | 2011-09-30 15:06:19 -0400 | [diff] [blame] | 1 | /* |
| 2 | * Copyright (C) 1991, 1992 Linus Torvalds |
| 3 | * Copyright (C) 2000, 2001, 2002 Andi Kleen, SuSE Labs |
Don Zickus | 9c48f1c | 2011-09-30 15:06:21 -0400 | [diff] [blame] | 4 | * Copyright (C) 2011 Don Zickus Red Hat, Inc. |
Don Zickus | 1d48922 | 2011-09-30 15:06:19 -0400 | [diff] [blame] | 5 | * |
| 6 | * Pentium III FXSR, SSE support |
| 7 | * Gareth Hughes <gareth@valinux.com>, May 2000 |
| 8 | */ |
| 9 | |
| 10 | /* |
| 11 | * Handle hardware traps and faults. |
| 12 | */ |
| 13 | #include <linux/spinlock.h> |
| 14 | #include <linux/kprobes.h> |
| 15 | #include <linux/kdebug.h> |
Ingo Molnar | b17b015 | 2017-02-08 18:51:35 +0100 | [diff] [blame] | 16 | #include <linux/sched/debug.h> |
Don Zickus | 1d48922 | 2011-09-30 15:06:19 -0400 | [diff] [blame] | 17 | #include <linux/nmi.h> |
Dave Hansen | 2ab00456 | 2013-06-21 08:51:35 -0700 | [diff] [blame] | 18 | #include <linux/debugfs.h> |
Don Zickus | c9126b2 | 2011-09-30 15:06:20 -0400 | [diff] [blame] | 19 | #include <linux/delay.h> |
| 20 | #include <linux/hardirq.h> |
Arnd Bergmann | c361db5 | 2016-06-06 16:09:52 +0200 | [diff] [blame] | 21 | #include <linux/ratelimit.h> |
Don Zickus | c9126b2 | 2011-09-30 15:06:20 -0400 | [diff] [blame] | 22 | #include <linux/slab.h> |
Paul Gortmaker | 69c60c8 | 2011-05-26 12:22:53 -0400 | [diff] [blame] | 23 | #include <linux/export.h> |
Ingo Molnar | e601757 | 2017-02-01 16:36:40 +0100 | [diff] [blame] | 24 | #include <linux/sched/clock.h> |
Don Zickus | 1d48922 | 2011-09-30 15:06:19 -0400 | [diff] [blame] | 25 | |
| 26 | #if defined(CONFIG_EDAC) |
| 27 | #include <linux/edac.h> |
| 28 | #endif |
| 29 | |
| 30 | #include <linux/atomic.h> |
| 31 | #include <asm/traps.h> |
| 32 | #include <asm/mach_traps.h> |
Don Zickus | c9126b2 | 2011-09-30 15:06:20 -0400 | [diff] [blame] | 33 | #include <asm/nmi.h> |
Mathias Nyman | 6fd36ba | 2011-11-10 13:45:24 +0000 | [diff] [blame] | 34 | #include <asm/x86_init.h> |
Hidehiro Kawai | b279d67 | 2015-12-14 11:19:13 +0100 | [diff] [blame] | 35 | #include <asm/reboot.h> |
Kostenzer Felix | 8e2a7f5 | 2016-03-06 23:20:06 +0100 | [diff] [blame] | 36 | #include <asm/cache.h> |
Thomas Gleixner | 04dcbdb | 2019-02-18 23:42:51 +0100 | [diff] [blame] | 37 | #include <asm/nospec-branch.h> |
Don Zickus | c9126b2 | 2011-09-30 15:06:20 -0400 | [diff] [blame] | 38 | |
Dave Hansen | 0c4df02 | 2013-06-21 08:51:38 -0700 | [diff] [blame] | 39 | #define CREATE_TRACE_POINTS |
| 40 | #include <trace/events/nmi.h> |
| 41 | |
Don Zickus | c9126b2 | 2011-09-30 15:06:20 -0400 | [diff] [blame] | 42 | struct nmi_desc { |
Scott Wood | c455fd9 | 2017-07-24 16:32:42 -0500 | [diff] [blame] | 43 | raw_spinlock_t lock; |
Don Zickus | c9126b2 | 2011-09-30 15:06:20 -0400 | [diff] [blame] | 44 | struct list_head head; |
| 45 | }; |
| 46 | |
| 47 | static struct nmi_desc nmi_desc[NMI_MAX] = |
| 48 | { |
| 49 | { |
Scott Wood | c455fd9 | 2017-07-24 16:32:42 -0500 | [diff] [blame] | 50 | .lock = __RAW_SPIN_LOCK_UNLOCKED(&nmi_desc[0].lock), |
Don Zickus | c9126b2 | 2011-09-30 15:06:20 -0400 | [diff] [blame] | 51 | .head = LIST_HEAD_INIT(nmi_desc[0].head), |
| 52 | }, |
| 53 | { |
Scott Wood | c455fd9 | 2017-07-24 16:32:42 -0500 | [diff] [blame] | 54 | .lock = __RAW_SPIN_LOCK_UNLOCKED(&nmi_desc[1].lock), |
Don Zickus | c9126b2 | 2011-09-30 15:06:20 -0400 | [diff] [blame] | 55 | .head = LIST_HEAD_INIT(nmi_desc[1].head), |
| 56 | }, |
Don Zickus | 553222f | 2012-03-29 16:11:16 -0400 | [diff] [blame] | 57 | { |
Scott Wood | c455fd9 | 2017-07-24 16:32:42 -0500 | [diff] [blame] | 58 | .lock = __RAW_SPIN_LOCK_UNLOCKED(&nmi_desc[2].lock), |
Don Zickus | 553222f | 2012-03-29 16:11:16 -0400 | [diff] [blame] | 59 | .head = LIST_HEAD_INIT(nmi_desc[2].head), |
| 60 | }, |
| 61 | { |
Scott Wood | c455fd9 | 2017-07-24 16:32:42 -0500 | [diff] [blame] | 62 | .lock = __RAW_SPIN_LOCK_UNLOCKED(&nmi_desc[3].lock), |
Don Zickus | 553222f | 2012-03-29 16:11:16 -0400 | [diff] [blame] | 63 | .head = LIST_HEAD_INIT(nmi_desc[3].head), |
| 64 | }, |
Don Zickus | c9126b2 | 2011-09-30 15:06:20 -0400 | [diff] [blame] | 65 | |
| 66 | }; |
Don Zickus | 1d48922 | 2011-09-30 15:06:19 -0400 | [diff] [blame] | 67 | |
Don Zickus | efc3aac | 2011-09-30 15:06:23 -0400 | [diff] [blame] | 68 | struct nmi_stats { |
| 69 | unsigned int normal; |
| 70 | unsigned int unknown; |
| 71 | unsigned int external; |
| 72 | unsigned int swallow; |
| 73 | }; |
| 74 | |
| 75 | static DEFINE_PER_CPU(struct nmi_stats, nmi_stats); |
| 76 | |
Kostenzer Felix | 8e2a7f5 | 2016-03-06 23:20:06 +0100 | [diff] [blame] | 77 | static int ignore_nmis __read_mostly; |
Don Zickus | 1d48922 | 2011-09-30 15:06:19 -0400 | [diff] [blame] | 78 | |
| 79 | int unknown_nmi_panic; |
| 80 | /* |
| 81 | * Prevent NMI reason port (0x61) being accessed simultaneously, can |
| 82 | * only be used in NMI handler. |
| 83 | */ |
| 84 | static DEFINE_RAW_SPINLOCK(nmi_reason_lock); |
| 85 | |
| 86 | static int __init setup_unknown_nmi_panic(char *str) |
| 87 | { |
| 88 | unknown_nmi_panic = 1; |
| 89 | return 1; |
| 90 | } |
| 91 | __setup("unknown_nmi_panic", setup_unknown_nmi_panic); |
| 92 | |
Don Zickus | c9126b2 | 2011-09-30 15:06:20 -0400 | [diff] [blame] | 93 | #define nmi_to_desc(type) (&nmi_desc[type]) |
| 94 | |
Dave Hansen | 2ab00456 | 2013-06-21 08:51:35 -0700 | [diff] [blame] | 95 | static u64 nmi_longest_ns = 1 * NSEC_PER_MSEC; |
Peter Zijlstra | e90c785 | 2014-02-03 18:02:09 +0100 | [diff] [blame] | 96 | |
Dave Hansen | 2ab00456 | 2013-06-21 08:51:35 -0700 | [diff] [blame] | 97 | static int __init nmi_warning_debugfs(void) |
| 98 | { |
| 99 | debugfs_create_u64("nmi_longest_ns", 0644, |
| 100 | arch_debugfs_dir, &nmi_longest_ns); |
| 101 | return 0; |
| 102 | } |
| 103 | fs_initcall(nmi_warning_debugfs); |
| 104 | |
Peter Zijlstra | e90c785 | 2014-02-03 18:02:09 +0100 | [diff] [blame] | 105 | static void nmi_max_handler(struct irq_work *w) |
| 106 | { |
| 107 | struct nmiaction *a = container_of(w, struct nmiaction, irq_work); |
| 108 | int remainder_ns, decimal_msecs; |
Mark Rutland | 6aa7de0 | 2017-10-23 14:07:29 -0700 | [diff] [blame] | 109 | u64 whole_msecs = READ_ONCE(a->max_duration); |
Peter Zijlstra | e90c785 | 2014-02-03 18:02:09 +0100 | [diff] [blame] | 110 | |
| 111 | remainder_ns = do_div(whole_msecs, (1000 * 1000)); |
| 112 | decimal_msecs = remainder_ns / 1000; |
| 113 | |
| 114 | printk_ratelimited(KERN_INFO |
| 115 | "INFO: NMI handler (%ps) took too long to run: %lld.%03d msecs\n", |
| 116 | a->handler, whole_msecs, decimal_msecs); |
| 117 | } |
| 118 | |
Andy Lutomirski | bf9f2ee | 2015-07-20 11:52:23 -0700 | [diff] [blame] | 119 | static int nmi_handle(unsigned int type, struct pt_regs *regs) |
Don Zickus | c9126b2 | 2011-09-30 15:06:20 -0400 | [diff] [blame] | 120 | { |
| 121 | struct nmi_desc *desc = nmi_to_desc(type); |
| 122 | struct nmiaction *a; |
| 123 | int handled=0; |
| 124 | |
| 125 | rcu_read_lock(); |
| 126 | |
| 127 | /* |
| 128 | * NMIs are edge-triggered, which means if you have enough |
| 129 | * of them concurrently, you can lose some because only one |
| 130 | * can be latched at any given time. Walk the whole list |
| 131 | * to handle those situations. |
| 132 | */ |
Dave Hansen | 2ab00456 | 2013-06-21 08:51:35 -0700 | [diff] [blame] | 133 | list_for_each_entry_rcu(a, &desc->head, list) { |
Peter Zijlstra | e90c785 | 2014-02-03 18:02:09 +0100 | [diff] [blame] | 134 | int thishandled; |
| 135 | u64 delta; |
Dave Hansen | 2ab00456 | 2013-06-21 08:51:35 -0700 | [diff] [blame] | 136 | |
Peter Zijlstra | e90c785 | 2014-02-03 18:02:09 +0100 | [diff] [blame] | 137 | delta = sched_clock(); |
Dave Hansen | 0c4df02 | 2013-06-21 08:51:38 -0700 | [diff] [blame] | 138 | thishandled = a->handler(type, regs); |
| 139 | handled += thishandled; |
Peter Zijlstra | e90c785 | 2014-02-03 18:02:09 +0100 | [diff] [blame] | 140 | delta = sched_clock() - delta; |
Dave Hansen | 0c4df02 | 2013-06-21 08:51:38 -0700 | [diff] [blame] | 141 | trace_nmi_handler(a->handler, (int)delta, thishandled); |
Dave Hansen | 2ab00456 | 2013-06-21 08:51:35 -0700 | [diff] [blame] | 142 | |
Peter Zijlstra | e90c785 | 2014-02-03 18:02:09 +0100 | [diff] [blame] | 143 | if (delta < nmi_longest_ns || delta < a->max_duration) |
Dave Hansen | 2ab00456 | 2013-06-21 08:51:35 -0700 | [diff] [blame] | 144 | continue; |
| 145 | |
Peter Zijlstra | e90c785 | 2014-02-03 18:02:09 +0100 | [diff] [blame] | 146 | a->max_duration = delta; |
| 147 | irq_work_queue(&a->irq_work); |
Dave Hansen | 2ab00456 | 2013-06-21 08:51:35 -0700 | [diff] [blame] | 148 | } |
Don Zickus | c9126b2 | 2011-09-30 15:06:20 -0400 | [diff] [blame] | 149 | |
Don Zickus | c9126b2 | 2011-09-30 15:06:20 -0400 | [diff] [blame] | 150 | rcu_read_unlock(); |
| 151 | |
| 152 | /* return total number of NMI events handled */ |
| 153 | return handled; |
| 154 | } |
Masami Hiramatsu | 9326638 | 2014-04-17 17:18:14 +0900 | [diff] [blame] | 155 | NOKPROBE_SYMBOL(nmi_handle); |
Don Zickus | c9126b2 | 2011-09-30 15:06:20 -0400 | [diff] [blame] | 156 | |
Li Zhong | 72b3fb2 | 2012-03-29 16:11:17 -0400 | [diff] [blame] | 157 | int __register_nmi_handler(unsigned int type, struct nmiaction *action) |
Don Zickus | c9126b2 | 2011-09-30 15:06:20 -0400 | [diff] [blame] | 158 | { |
| 159 | struct nmi_desc *desc = nmi_to_desc(type); |
| 160 | unsigned long flags; |
| 161 | |
Li Zhong | 72b3fb2 | 2012-03-29 16:11:17 -0400 | [diff] [blame] | 162 | if (!action->handler) |
| 163 | return -EINVAL; |
| 164 | |
Peter Zijlstra | e90c785 | 2014-02-03 18:02:09 +0100 | [diff] [blame] | 165 | init_irq_work(&action->irq_work, nmi_max_handler); |
| 166 | |
Scott Wood | c455fd9 | 2017-07-24 16:32:42 -0500 | [diff] [blame] | 167 | raw_spin_lock_irqsave(&desc->lock, flags); |
Don Zickus | c9126b2 | 2011-09-30 15:06:20 -0400 | [diff] [blame] | 168 | |
| 169 | /* |
Mike Travis | 0d443b7 | 2017-03-07 15:08:42 -0600 | [diff] [blame] | 170 | * Indicate if there are multiple registrations on the |
| 171 | * internal NMI handler call chains (SERR and IO_CHECK). |
Don Zickus | b227e23 | 2011-09-30 15:06:22 -0400 | [diff] [blame] | 172 | */ |
Don Zickus | 553222f | 2012-03-29 16:11:16 -0400 | [diff] [blame] | 173 | WARN_ON_ONCE(type == NMI_SERR && !list_empty(&desc->head)); |
| 174 | WARN_ON_ONCE(type == NMI_IO_CHECK && !list_empty(&desc->head)); |
Don Zickus | b227e23 | 2011-09-30 15:06:22 -0400 | [diff] [blame] | 175 | |
| 176 | /* |
Don Zickus | c9126b2 | 2011-09-30 15:06:20 -0400 | [diff] [blame] | 177 | * some handlers need to be executed first otherwise a fake |
| 178 | * event confuses some handlers (kdump uses this flag) |
| 179 | */ |
| 180 | if (action->flags & NMI_FLAG_FIRST) |
| 181 | list_add_rcu(&action->list, &desc->head); |
| 182 | else |
| 183 | list_add_tail_rcu(&action->list, &desc->head); |
| 184 | |
Scott Wood | c455fd9 | 2017-07-24 16:32:42 -0500 | [diff] [blame] | 185 | raw_spin_unlock_irqrestore(&desc->lock, flags); |
Don Zickus | c9126b2 | 2011-09-30 15:06:20 -0400 | [diff] [blame] | 186 | return 0; |
| 187 | } |
Li Zhong | 72b3fb2 | 2012-03-29 16:11:17 -0400 | [diff] [blame] | 188 | EXPORT_SYMBOL(__register_nmi_handler); |
Don Zickus | c9126b2 | 2011-09-30 15:06:20 -0400 | [diff] [blame] | 189 | |
Li Zhong | 72b3fb2 | 2012-03-29 16:11:17 -0400 | [diff] [blame] | 190 | void unregister_nmi_handler(unsigned int type, const char *name) |
Don Zickus | c9126b2 | 2011-09-30 15:06:20 -0400 | [diff] [blame] | 191 | { |
| 192 | struct nmi_desc *desc = nmi_to_desc(type); |
| 193 | struct nmiaction *n; |
| 194 | unsigned long flags; |
| 195 | |
Scott Wood | c455fd9 | 2017-07-24 16:32:42 -0500 | [diff] [blame] | 196 | raw_spin_lock_irqsave(&desc->lock, flags); |
Don Zickus | c9126b2 | 2011-09-30 15:06:20 -0400 | [diff] [blame] | 197 | |
| 198 | list_for_each_entry_rcu(n, &desc->head, list) { |
| 199 | /* |
| 200 | * the name passed in to describe the nmi handler |
| 201 | * is used as the lookup key |
| 202 | */ |
| 203 | if (!strcmp(n->name, name)) { |
| 204 | WARN(in_nmi(), |
| 205 | "Trying to free NMI (%s) from NMI context!\n", n->name); |
| 206 | list_del_rcu(&n->list); |
| 207 | break; |
| 208 | } |
| 209 | } |
| 210 | |
Scott Wood | c455fd9 | 2017-07-24 16:32:42 -0500 | [diff] [blame] | 211 | raw_spin_unlock_irqrestore(&desc->lock, flags); |
Don Zickus | c9126b2 | 2011-09-30 15:06:20 -0400 | [diff] [blame] | 212 | synchronize_rcu(); |
Don Zickus | c9126b2 | 2011-09-30 15:06:20 -0400 | [diff] [blame] | 213 | } |
Don Zickus | c9126b2 | 2011-09-30 15:06:20 -0400 | [diff] [blame] | 214 | EXPORT_SYMBOL_GPL(unregister_nmi_handler); |
| 215 | |
Masami Hiramatsu | 9326638 | 2014-04-17 17:18:14 +0900 | [diff] [blame] | 216 | static void |
Don Zickus | 1d48922 | 2011-09-30 15:06:19 -0400 | [diff] [blame] | 217 | pci_serr_error(unsigned char reason, struct pt_regs *regs) |
| 218 | { |
Don Zickus | 553222f | 2012-03-29 16:11:16 -0400 | [diff] [blame] | 219 | /* check to see if anyone registered against these types of errors */ |
Andy Lutomirski | bf9f2ee | 2015-07-20 11:52:23 -0700 | [diff] [blame] | 220 | if (nmi_handle(NMI_SERR, regs)) |
Don Zickus | 553222f | 2012-03-29 16:11:16 -0400 | [diff] [blame] | 221 | return; |
| 222 | |
Don Zickus | 1d48922 | 2011-09-30 15:06:19 -0400 | [diff] [blame] | 223 | pr_emerg("NMI: PCI system error (SERR) for reason %02x on CPU %d.\n", |
| 224 | reason, smp_processor_id()); |
| 225 | |
Don Zickus | 1d48922 | 2011-09-30 15:06:19 -0400 | [diff] [blame] | 226 | if (panic_on_unrecovered_nmi) |
Hidehiro Kawai | 58c5661 | 2015-12-14 11:19:10 +0100 | [diff] [blame] | 227 | nmi_panic(regs, "NMI: Not continuing"); |
Don Zickus | 1d48922 | 2011-09-30 15:06:19 -0400 | [diff] [blame] | 228 | |
| 229 | pr_emerg("Dazed and confused, but trying to continue\n"); |
| 230 | |
| 231 | /* Clear and disable the PCI SERR error line. */ |
| 232 | reason = (reason & NMI_REASON_CLEAR_MASK) | NMI_REASON_CLEAR_SERR; |
| 233 | outb(reason, NMI_REASON_PORT); |
| 234 | } |
Masami Hiramatsu | 9326638 | 2014-04-17 17:18:14 +0900 | [diff] [blame] | 235 | NOKPROBE_SYMBOL(pci_serr_error); |
Don Zickus | 1d48922 | 2011-09-30 15:06:19 -0400 | [diff] [blame] | 236 | |
Masami Hiramatsu | 9326638 | 2014-04-17 17:18:14 +0900 | [diff] [blame] | 237 | static void |
Don Zickus | 1d48922 | 2011-09-30 15:06:19 -0400 | [diff] [blame] | 238 | io_check_error(unsigned char reason, struct pt_regs *regs) |
| 239 | { |
| 240 | unsigned long i; |
| 241 | |
Don Zickus | 553222f | 2012-03-29 16:11:16 -0400 | [diff] [blame] | 242 | /* check to see if anyone registered against these types of errors */ |
Andy Lutomirski | bf9f2ee | 2015-07-20 11:52:23 -0700 | [diff] [blame] | 243 | if (nmi_handle(NMI_IO_CHECK, regs)) |
Don Zickus | 553222f | 2012-03-29 16:11:16 -0400 | [diff] [blame] | 244 | return; |
| 245 | |
Don Zickus | 1d48922 | 2011-09-30 15:06:19 -0400 | [diff] [blame] | 246 | pr_emerg( |
| 247 | "NMI: IOCK error (debug interrupt?) for reason %02x on CPU %d.\n", |
| 248 | reason, smp_processor_id()); |
Jan Beulich | 57da8b9 | 2012-05-09 08:47:37 +0100 | [diff] [blame] | 249 | show_regs(regs); |
Don Zickus | 1d48922 | 2011-09-30 15:06:19 -0400 | [diff] [blame] | 250 | |
Hidehiro Kawai | 1717f20 | 2015-12-14 11:19:09 +0100 | [diff] [blame] | 251 | if (panic_on_io_nmi) { |
Hidehiro Kawai | 58c5661 | 2015-12-14 11:19:10 +0100 | [diff] [blame] | 252 | nmi_panic(regs, "NMI IOCK error: Not continuing"); |
Hidehiro Kawai | 1717f20 | 2015-12-14 11:19:09 +0100 | [diff] [blame] | 253 | |
| 254 | /* |
| 255 | * If we end up here, it means we have received an NMI while |
| 256 | * processing panic(). Simply return without delaying and |
| 257 | * re-enabling NMIs. |
| 258 | */ |
| 259 | return; |
| 260 | } |
Don Zickus | 1d48922 | 2011-09-30 15:06:19 -0400 | [diff] [blame] | 261 | |
| 262 | /* Re-enable the IOCK line, wait for a few seconds */ |
| 263 | reason = (reason & NMI_REASON_CLEAR_MASK) | NMI_REASON_CLEAR_IOCHK; |
| 264 | outb(reason, NMI_REASON_PORT); |
| 265 | |
| 266 | i = 20000; |
| 267 | while (--i) { |
| 268 | touch_nmi_watchdog(); |
| 269 | udelay(100); |
| 270 | } |
| 271 | |
| 272 | reason &= ~NMI_REASON_CLEAR_IOCHK; |
| 273 | outb(reason, NMI_REASON_PORT); |
| 274 | } |
Masami Hiramatsu | 9326638 | 2014-04-17 17:18:14 +0900 | [diff] [blame] | 275 | NOKPROBE_SYMBOL(io_check_error); |
Don Zickus | 1d48922 | 2011-09-30 15:06:19 -0400 | [diff] [blame] | 276 | |
Masami Hiramatsu | 9326638 | 2014-04-17 17:18:14 +0900 | [diff] [blame] | 277 | static void |
Don Zickus | 1d48922 | 2011-09-30 15:06:19 -0400 | [diff] [blame] | 278 | unknown_nmi_error(unsigned char reason, struct pt_regs *regs) |
| 279 | { |
Don Zickus | 9c48f1c | 2011-09-30 15:06:21 -0400 | [diff] [blame] | 280 | int handled; |
| 281 | |
Don Zickus | b227e23 | 2011-09-30 15:06:22 -0400 | [diff] [blame] | 282 | /* |
| 283 | * Use 'false' as back-to-back NMIs are dealt with one level up. |
| 284 | * Of course this makes having multiple 'unknown' handlers useless |
| 285 | * as only the first one is ever run (unless it can actually determine |
| 286 | * if it caused the NMI) |
| 287 | */ |
Andy Lutomirski | bf9f2ee | 2015-07-20 11:52:23 -0700 | [diff] [blame] | 288 | handled = nmi_handle(NMI_UNKNOWN, regs); |
Don Zickus | efc3aac | 2011-09-30 15:06:23 -0400 | [diff] [blame] | 289 | if (handled) { |
| 290 | __this_cpu_add(nmi_stats.unknown, handled); |
Don Zickus | 1d48922 | 2011-09-30 15:06:19 -0400 | [diff] [blame] | 291 | return; |
Don Zickus | efc3aac | 2011-09-30 15:06:23 -0400 | [diff] [blame] | 292 | } |
| 293 | |
| 294 | __this_cpu_add(nmi_stats.unknown, 1); |
| 295 | |
Don Zickus | 1d48922 | 2011-09-30 15:06:19 -0400 | [diff] [blame] | 296 | pr_emerg("Uhhuh. NMI received for unknown reason %02x on CPU %d.\n", |
| 297 | reason, smp_processor_id()); |
| 298 | |
| 299 | pr_emerg("Do you have a strange power saving mode enabled?\n"); |
| 300 | if (unknown_nmi_panic || panic_on_unrecovered_nmi) |
Hidehiro Kawai | 58c5661 | 2015-12-14 11:19:10 +0100 | [diff] [blame] | 301 | nmi_panic(regs, "NMI: Not continuing"); |
Don Zickus | 1d48922 | 2011-09-30 15:06:19 -0400 | [diff] [blame] | 302 | |
| 303 | pr_emerg("Dazed and confused, but trying to continue\n"); |
| 304 | } |
Masami Hiramatsu | 9326638 | 2014-04-17 17:18:14 +0900 | [diff] [blame] | 305 | NOKPROBE_SYMBOL(unknown_nmi_error); |
Don Zickus | 1d48922 | 2011-09-30 15:06:19 -0400 | [diff] [blame] | 306 | |
Don Zickus | b227e23 | 2011-09-30 15:06:22 -0400 | [diff] [blame] | 307 | static DEFINE_PER_CPU(bool, swallow_nmi); |
| 308 | static DEFINE_PER_CPU(unsigned long, last_nmi_rip); |
| 309 | |
Masami Hiramatsu | 9326638 | 2014-04-17 17:18:14 +0900 | [diff] [blame] | 310 | static void default_do_nmi(struct pt_regs *regs) |
Don Zickus | 1d48922 | 2011-09-30 15:06:19 -0400 | [diff] [blame] | 311 | { |
| 312 | unsigned char reason = 0; |
Don Zickus | 9c48f1c | 2011-09-30 15:06:21 -0400 | [diff] [blame] | 313 | int handled; |
Don Zickus | b227e23 | 2011-09-30 15:06:22 -0400 | [diff] [blame] | 314 | bool b2b = false; |
Don Zickus | 1d48922 | 2011-09-30 15:06:19 -0400 | [diff] [blame] | 315 | |
| 316 | /* |
| 317 | * CPU-specific NMI must be processed before non-CPU-specific |
| 318 | * NMI, otherwise we may lose it, because the CPU-specific |
| 319 | * NMI can not be detected/processed on other CPUs. |
| 320 | */ |
Don Zickus | b227e23 | 2011-09-30 15:06:22 -0400 | [diff] [blame] | 321 | |
| 322 | /* |
| 323 | * Back-to-back NMIs are interesting because they can either |
| 324 | * be two NMI or more than two NMIs (any thing over two is dropped |
| 325 | * due to NMI being edge-triggered). If this is the second half |
| 326 | * of the back-to-back NMI, assume we dropped things and process |
| 327 | * more handlers. Otherwise reset the 'swallow' NMI behaviour |
| 328 | */ |
| 329 | if (regs->ip == __this_cpu_read(last_nmi_rip)) |
| 330 | b2b = true; |
| 331 | else |
| 332 | __this_cpu_write(swallow_nmi, false); |
| 333 | |
| 334 | __this_cpu_write(last_nmi_rip, regs->ip); |
| 335 | |
Andy Lutomirski | bf9f2ee | 2015-07-20 11:52:23 -0700 | [diff] [blame] | 336 | handled = nmi_handle(NMI_LOCAL, regs); |
Don Zickus | efc3aac | 2011-09-30 15:06:23 -0400 | [diff] [blame] | 337 | __this_cpu_add(nmi_stats.normal, handled); |
Don Zickus | b227e23 | 2011-09-30 15:06:22 -0400 | [diff] [blame] | 338 | if (handled) { |
| 339 | /* |
| 340 | * There are cases when a NMI handler handles multiple |
| 341 | * events in the current NMI. One of these events may |
| 342 | * be queued for in the next NMI. Because the event is |
| 343 | * already handled, the next NMI will result in an unknown |
| 344 | * NMI. Instead lets flag this for a potential NMI to |
| 345 | * swallow. |
| 346 | */ |
| 347 | if (handled > 1) |
| 348 | __this_cpu_write(swallow_nmi, true); |
Don Zickus | 1d48922 | 2011-09-30 15:06:19 -0400 | [diff] [blame] | 349 | return; |
Don Zickus | b227e23 | 2011-09-30 15:06:22 -0400 | [diff] [blame] | 350 | } |
Don Zickus | 1d48922 | 2011-09-30 15:06:19 -0400 | [diff] [blame] | 351 | |
Hidehiro Kawai | b279d67 | 2015-12-14 11:19:13 +0100 | [diff] [blame] | 352 | /* |
| 353 | * Non-CPU-specific NMI: NMI sources can be processed on any CPU. |
| 354 | * |
| 355 | * Another CPU may be processing panic routines while holding |
| 356 | * nmi_reason_lock. Check if the CPU issued the IPI for crash dumping, |
| 357 | * and if so, call its callback directly. If there is no CPU preparing |
| 358 | * crash dump, we simply loop here. |
| 359 | */ |
| 360 | while (!raw_spin_trylock(&nmi_reason_lock)) { |
| 361 | run_crash_ipi_callback(regs); |
| 362 | cpu_relax(); |
| 363 | } |
| 364 | |
Jacob Pan | 064a59b | 2011-11-10 13:43:05 +0000 | [diff] [blame] | 365 | reason = x86_platform.get_nmi_reason(); |
Don Zickus | 1d48922 | 2011-09-30 15:06:19 -0400 | [diff] [blame] | 366 | |
| 367 | if (reason & NMI_REASON_MASK) { |
| 368 | if (reason & NMI_REASON_SERR) |
| 369 | pci_serr_error(reason, regs); |
| 370 | else if (reason & NMI_REASON_IOCHK) |
| 371 | io_check_error(reason, regs); |
| 372 | #ifdef CONFIG_X86_32 |
| 373 | /* |
| 374 | * Reassert NMI in case it became active |
| 375 | * meanwhile as it's edge-triggered: |
| 376 | */ |
| 377 | reassert_nmi(); |
| 378 | #endif |
Don Zickus | efc3aac | 2011-09-30 15:06:23 -0400 | [diff] [blame] | 379 | __this_cpu_add(nmi_stats.external, 1); |
Don Zickus | 1d48922 | 2011-09-30 15:06:19 -0400 | [diff] [blame] | 380 | raw_spin_unlock(&nmi_reason_lock); |
| 381 | return; |
| 382 | } |
| 383 | raw_spin_unlock(&nmi_reason_lock); |
| 384 | |
Don Zickus | b227e23 | 2011-09-30 15:06:22 -0400 | [diff] [blame] | 385 | /* |
| 386 | * Only one NMI can be latched at a time. To handle |
| 387 | * this we may process multiple nmi handlers at once to |
| 388 | * cover the case where an NMI is dropped. The downside |
| 389 | * to this approach is we may process an NMI prematurely, |
| 390 | * while its real NMI is sitting latched. This will cause |
| 391 | * an unknown NMI on the next run of the NMI processing. |
| 392 | * |
| 393 | * We tried to flag that condition above, by setting the |
| 394 | * swallow_nmi flag when we process more than one event. |
| 395 | * This condition is also only present on the second half |
| 396 | * of a back-to-back NMI, so we flag that condition too. |
| 397 | * |
| 398 | * If both are true, we assume we already processed this |
| 399 | * NMI previously and we swallow it. Otherwise we reset |
| 400 | * the logic. |
| 401 | * |
| 402 | * There are scenarios where we may accidentally swallow |
| 403 | * a 'real' unknown NMI. For example, while processing |
| 404 | * a perf NMI another perf NMI comes in along with a |
| 405 | * 'real' unknown NMI. These two NMIs get combined into |
| 406 | * one (as descibed above). When the next NMI gets |
| 407 | * processed, it will be flagged by perf as handled, but |
| 408 | * noone will know that there was a 'real' unknown NMI sent |
| 409 | * also. As a result it gets swallowed. Or if the first |
| 410 | * perf NMI returns two events handled then the second |
| 411 | * NMI will get eaten by the logic below, again losing a |
| 412 | * 'real' unknown NMI. But this is the best we can do |
| 413 | * for now. |
| 414 | */ |
| 415 | if (b2b && __this_cpu_read(swallow_nmi)) |
Don Zickus | efc3aac | 2011-09-30 15:06:23 -0400 | [diff] [blame] | 416 | __this_cpu_add(nmi_stats.swallow, 1); |
Don Zickus | b227e23 | 2011-09-30 15:06:22 -0400 | [diff] [blame] | 417 | else |
| 418 | unknown_nmi_error(reason, regs); |
Don Zickus | 1d48922 | 2011-09-30 15:06:19 -0400 | [diff] [blame] | 419 | } |
Masami Hiramatsu | 9326638 | 2014-04-17 17:18:14 +0900 | [diff] [blame] | 420 | NOKPROBE_SYMBOL(default_do_nmi); |
Don Zickus | 1d48922 | 2011-09-30 15:06:19 -0400 | [diff] [blame] | 421 | |
Steven Rostedt | ccd49c2 | 2011-12-13 16:44:16 -0500 | [diff] [blame] | 422 | /* |
Andy Lutomirski | 0b22930 | 2015-07-15 10:29:36 -0700 | [diff] [blame] | 423 | * NMIs can page fault or hit breakpoints which will cause it to lose |
| 424 | * its NMI context with the CPU when the breakpoint or page fault does an IRET. |
Andy Lutomirski | 9d05041 | 2015-07-15 10:29:33 -0700 | [diff] [blame] | 425 | * |
| 426 | * As a result, NMIs can nest if NMIs get unmasked due an IRET during |
| 427 | * NMI processing. On x86_64, the asm glue protects us from nested NMIs |
| 428 | * if the outer NMI came from kernel mode, but we can still nest if the |
| 429 | * outer NMI came from user mode. |
| 430 | * |
| 431 | * To handle these nested NMIs, we have three states: |
Steven Rostedt | ccd49c2 | 2011-12-13 16:44:16 -0500 | [diff] [blame] | 432 | * |
| 433 | * 1) not running |
| 434 | * 2) executing |
| 435 | * 3) latched |
| 436 | * |
| 437 | * When no NMI is in progress, it is in the "not running" state. |
| 438 | * When an NMI comes in, it goes into the "executing" state. |
| 439 | * Normally, if another NMI is triggered, it does not interrupt |
| 440 | * the running NMI and the HW will simply latch it so that when |
| 441 | * the first NMI finishes, it will restart the second NMI. |
| 442 | * (Note, the latch is binary, thus multiple NMIs triggering, |
| 443 | * when one is running, are ignored. Only one NMI is restarted.) |
| 444 | * |
Andy Lutomirski | 9d05041 | 2015-07-15 10:29:33 -0700 | [diff] [blame] | 445 | * If an NMI executes an iret, another NMI can preempt it. We do not |
| 446 | * want to allow this new NMI to run, but we want to execute it when the |
| 447 | * first one finishes. We set the state to "latched", and the exit of |
| 448 | * the first NMI will perform a dec_return, if the result is zero |
| 449 | * (NOT_RUNNING), then it will simply exit the NMI handler. If not, the |
| 450 | * dec_return would have set the state to NMI_EXECUTING (what we want it |
| 451 | * to be when we are running). In this case, we simply jump back to |
| 452 | * rerun the NMI handler again, and restart the 'latched' NMI. |
Steven Rostedt | c7d65a7 | 2012-06-07 11:03:00 -0400 | [diff] [blame] | 453 | * |
| 454 | * No trap (breakpoint or page fault) should be hit before nmi_restart, |
| 455 | * thus there is no race between the first check of state for NOT_RUNNING |
| 456 | * and setting it to NMI_EXECUTING. The HW will prevent nested NMIs |
| 457 | * at this point. |
Steven Rostedt | 70fb74a | 2012-06-07 11:54:37 -0400 | [diff] [blame] | 458 | * |
| 459 | * In case the NMI takes a page fault, we need to save off the CR2 |
| 460 | * because the NMI could have preempted another page fault and corrupt |
| 461 | * the CR2 that is about to be read. As nested NMIs must be restarted |
| 462 | * and they can not take breakpoints or page faults, the update of the |
| 463 | * CR2 must be done before converting the nmi state back to NOT_RUNNING. |
| 464 | * Otherwise, there would be a race of another nested NMI coming in |
| 465 | * after setting state to NOT_RUNNING but before updating the nmi_cr2. |
Steven Rostedt | ccd49c2 | 2011-12-13 16:44:16 -0500 | [diff] [blame] | 466 | */ |
| 467 | enum nmi_states { |
Steven Rostedt | c7d65a7 | 2012-06-07 11:03:00 -0400 | [diff] [blame] | 468 | NMI_NOT_RUNNING = 0, |
Steven Rostedt | ccd49c2 | 2011-12-13 16:44:16 -0500 | [diff] [blame] | 469 | NMI_EXECUTING, |
| 470 | NMI_LATCHED, |
| 471 | }; |
| 472 | static DEFINE_PER_CPU(enum nmi_states, nmi_state); |
Steven Rostedt | 70fb74a | 2012-06-07 11:54:37 -0400 | [diff] [blame] | 473 | static DEFINE_PER_CPU(unsigned long, nmi_cr2); |
Steven Rostedt | 228bdaa | 2011-12-09 03:02:19 -0500 | [diff] [blame] | 474 | |
Andy Lutomirski | 9d05041 | 2015-07-15 10:29:33 -0700 | [diff] [blame] | 475 | #ifdef CONFIG_X86_64 |
Steven Rostedt | ccd49c2 | 2011-12-13 16:44:16 -0500 | [diff] [blame] | 476 | /* |
Andy Lutomirski | 9d05041 | 2015-07-15 10:29:33 -0700 | [diff] [blame] | 477 | * In x86_64, we need to handle breakpoint -> NMI -> breakpoint. Without |
| 478 | * some care, the inner breakpoint will clobber the outer breakpoint's |
| 479 | * stack. |
Steven Rostedt | ccd49c2 | 2011-12-13 16:44:16 -0500 | [diff] [blame] | 480 | * |
Andy Lutomirski | 9d05041 | 2015-07-15 10:29:33 -0700 | [diff] [blame] | 481 | * If a breakpoint is being processed, and the debug stack is being |
| 482 | * used, if an NMI comes in and also hits a breakpoint, the stack |
| 483 | * pointer will be set to the same fixed address as the breakpoint that |
| 484 | * was interrupted, causing that stack to be corrupted. To handle this |
| 485 | * case, check if the stack that was interrupted is the debug stack, and |
| 486 | * if so, change the IDT so that new breakpoints will use the current |
| 487 | * stack and not switch to the fixed address. On return of the NMI, |
| 488 | * switch back to the original IDT. |
Steven Rostedt | ccd49c2 | 2011-12-13 16:44:16 -0500 | [diff] [blame] | 489 | */ |
| 490 | static DEFINE_PER_CPU(int, update_debug_stack); |
Andy Lutomirski | 9d05041 | 2015-07-15 10:29:33 -0700 | [diff] [blame] | 491 | #endif |
Steven Rostedt | ccd49c2 | 2011-12-13 16:44:16 -0500 | [diff] [blame] | 492 | |
Andy Lutomirski | 9d05041 | 2015-07-15 10:29:33 -0700 | [diff] [blame] | 493 | dotraplinkage notrace void |
| 494 | do_nmi(struct pt_regs *regs, long error_code) |
Steven Rostedt | ccd49c2 | 2011-12-13 16:44:16 -0500 | [diff] [blame] | 495 | { |
Andy Lutomirski | 9d05041 | 2015-07-15 10:29:33 -0700 | [diff] [blame] | 496 | if (this_cpu_read(nmi_state) != NMI_NOT_RUNNING) { |
| 497 | this_cpu_write(nmi_state, NMI_LATCHED); |
| 498 | return; |
| 499 | } |
| 500 | this_cpu_write(nmi_state, NMI_EXECUTING); |
| 501 | this_cpu_write(nmi_cr2, read_cr2()); |
| 502 | nmi_restart: |
| 503 | |
| 504 | #ifdef CONFIG_X86_64 |
Steven Rostedt | 228bdaa | 2011-12-09 03:02:19 -0500 | [diff] [blame] | 505 | /* |
| 506 | * If we interrupted a breakpoint, it is possible that |
| 507 | * the nmi handler will have breakpoints too. We need to |
| 508 | * change the IDT such that breakpoints that happen here |
| 509 | * continue to use the NMI stack. |
| 510 | */ |
| 511 | if (unlikely(is_debug_stack(regs->sp))) { |
| 512 | debug_stack_set_zero(); |
Steven Rostedt | c0525a6 | 2012-05-30 11:43:19 -0400 | [diff] [blame] | 513 | this_cpu_write(update_debug_stack, 1); |
Steven Rostedt | 228bdaa | 2011-12-09 03:02:19 -0500 | [diff] [blame] | 514 | } |
Steven Rostedt | ccd49c2 | 2011-12-13 16:44:16 -0500 | [diff] [blame] | 515 | #endif |
| 516 | |
Don Zickus | 1d48922 | 2011-09-30 15:06:19 -0400 | [diff] [blame] | 517 | nmi_enter(); |
| 518 | |
| 519 | inc_irq_stat(__nmi_count); |
| 520 | |
| 521 | if (!ignore_nmis) |
| 522 | default_do_nmi(regs); |
| 523 | |
| 524 | nmi_exit(); |
Steven Rostedt | 228bdaa | 2011-12-09 03:02:19 -0500 | [diff] [blame] | 525 | |
Andy Lutomirski | 9d05041 | 2015-07-15 10:29:33 -0700 | [diff] [blame] | 526 | #ifdef CONFIG_X86_64 |
| 527 | if (unlikely(this_cpu_read(update_debug_stack))) { |
| 528 | debug_stack_reset(); |
| 529 | this_cpu_write(update_debug_stack, 0); |
| 530 | } |
| 531 | #endif |
| 532 | |
| 533 | if (unlikely(this_cpu_read(nmi_cr2) != read_cr2())) |
| 534 | write_cr2(this_cpu_read(nmi_cr2)); |
| 535 | if (this_cpu_dec_return(nmi_state)) |
| 536 | goto nmi_restart; |
Thomas Gleixner | 04dcbdb | 2019-02-18 23:42:51 +0100 | [diff] [blame] | 537 | |
| 538 | if (user_mode(regs)) |
| 539 | mds_user_clear_cpu_buffers(); |
Don Zickus | 1d48922 | 2011-09-30 15:06:19 -0400 | [diff] [blame] | 540 | } |
Masami Hiramatsu | 9326638 | 2014-04-17 17:18:14 +0900 | [diff] [blame] | 541 | NOKPROBE_SYMBOL(do_nmi); |
Don Zickus | 1d48922 | 2011-09-30 15:06:19 -0400 | [diff] [blame] | 542 | |
| 543 | void stop_nmi(void) |
| 544 | { |
| 545 | ignore_nmis++; |
| 546 | } |
| 547 | |
| 548 | void restart_nmi(void) |
| 549 | { |
| 550 | ignore_nmis--; |
| 551 | } |
Don Zickus | b227e23 | 2011-09-30 15:06:22 -0400 | [diff] [blame] | 552 | |
| 553 | /* reset the back-to-back NMI logic */ |
| 554 | void local_touch_nmi(void) |
| 555 | { |
| 556 | __this_cpu_write(last_nmi_rip, 0); |
| 557 | } |
Jacob Pan | 29c6fb7 | 2013-01-04 11:12:44 +0000 | [diff] [blame] | 558 | EXPORT_SYMBOL_GPL(local_touch_nmi); |