Commit | Line | Data |
---|---|---|
1d48922c DZ |
1 | /* |
2 | * Copyright (C) 1991, 1992 Linus Torvalds | |
3 | * Copyright (C) 2000, 2001, 2002 Andi Kleen, SuSE Labs | |
9c48f1c6 | 4 | * Copyright (C) 2011 Don Zickus Red Hat, Inc. |
1d48922c DZ |
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> | |
16 | #include <linux/nmi.h> | |
2ab00456 | 17 | #include <linux/debugfs.h> |
c9126b2e DZ |
18 | #include <linux/delay.h> |
19 | #include <linux/hardirq.h> | |
20 | #include <linux/slab.h> | |
69c60c88 | 21 | #include <linux/export.h> |
1d48922c DZ |
22 | |
23 | #if defined(CONFIG_EDAC) | |
24 | #include <linux/edac.h> | |
25 | #endif | |
26 | ||
27 | #include <linux/atomic.h> | |
28 | #include <asm/traps.h> | |
29 | #include <asm/mach_traps.h> | |
c9126b2e | 30 | #include <asm/nmi.h> |
6fd36ba0 | 31 | #include <asm/x86_init.h> |
c9126b2e | 32 | |
0c4df02d DH |
33 | #define CREATE_TRACE_POINTS |
34 | #include <trace/events/nmi.h> | |
35 | ||
c9126b2e DZ |
36 | struct nmi_desc { |
37 | spinlock_t lock; | |
38 | struct list_head head; | |
39 | }; | |
40 | ||
41 | static struct nmi_desc nmi_desc[NMI_MAX] = | |
42 | { | |
43 | { | |
44 | .lock = __SPIN_LOCK_UNLOCKED(&nmi_desc[0].lock), | |
45 | .head = LIST_HEAD_INIT(nmi_desc[0].head), | |
46 | }, | |
47 | { | |
48 | .lock = __SPIN_LOCK_UNLOCKED(&nmi_desc[1].lock), | |
49 | .head = LIST_HEAD_INIT(nmi_desc[1].head), | |
50 | }, | |
553222f3 DZ |
51 | { |
52 | .lock = __SPIN_LOCK_UNLOCKED(&nmi_desc[2].lock), | |
53 | .head = LIST_HEAD_INIT(nmi_desc[2].head), | |
54 | }, | |
55 | { | |
56 | .lock = __SPIN_LOCK_UNLOCKED(&nmi_desc[3].lock), | |
57 | .head = LIST_HEAD_INIT(nmi_desc[3].head), | |
58 | }, | |
c9126b2e DZ |
59 | |
60 | }; | |
1d48922c | 61 | |
efc3aac5 DZ |
62 | struct nmi_stats { |
63 | unsigned int normal; | |
64 | unsigned int unknown; | |
65 | unsigned int external; | |
66 | unsigned int swallow; | |
67 | }; | |
68 | ||
69 | static DEFINE_PER_CPU(struct nmi_stats, nmi_stats); | |
70 | ||
1d48922c DZ |
71 | static int ignore_nmis; |
72 | ||
73 | int unknown_nmi_panic; | |
74 | /* | |
75 | * Prevent NMI reason port (0x61) being accessed simultaneously, can | |
76 | * only be used in NMI handler. | |
77 | */ | |
78 | static DEFINE_RAW_SPINLOCK(nmi_reason_lock); | |
79 | ||
80 | static int __init setup_unknown_nmi_panic(char *str) | |
81 | { | |
82 | unknown_nmi_panic = 1; | |
83 | return 1; | |
84 | } | |
85 | __setup("unknown_nmi_panic", setup_unknown_nmi_panic); | |
86 | ||
c9126b2e DZ |
87 | #define nmi_to_desc(type) (&nmi_desc[type]) |
88 | ||
2ab00456 | 89 | static u64 nmi_longest_ns = 1 * NSEC_PER_MSEC; |
e90c7853 | 90 | |
2ab00456 DH |
91 | static int __init nmi_warning_debugfs(void) |
92 | { | |
93 | debugfs_create_u64("nmi_longest_ns", 0644, | |
94 | arch_debugfs_dir, &nmi_longest_ns); | |
95 | return 0; | |
96 | } | |
97 | fs_initcall(nmi_warning_debugfs); | |
98 | ||
e90c7853 PZ |
99 | static void nmi_max_handler(struct irq_work *w) |
100 | { | |
101 | struct nmiaction *a = container_of(w, struct nmiaction, irq_work); | |
102 | int remainder_ns, decimal_msecs; | |
103 | u64 whole_msecs = ACCESS_ONCE(a->max_duration); | |
104 | ||
105 | remainder_ns = do_div(whole_msecs, (1000 * 1000)); | |
106 | decimal_msecs = remainder_ns / 1000; | |
107 | ||
108 | printk_ratelimited(KERN_INFO | |
109 | "INFO: NMI handler (%ps) took too long to run: %lld.%03d msecs\n", | |
110 | a->handler, whole_msecs, decimal_msecs); | |
111 | } | |
112 | ||
9326638c | 113 | static int nmi_handle(unsigned int type, struct pt_regs *regs, bool b2b) |
c9126b2e DZ |
114 | { |
115 | struct nmi_desc *desc = nmi_to_desc(type); | |
116 | struct nmiaction *a; | |
117 | int handled=0; | |
118 | ||
119 | rcu_read_lock(); | |
120 | ||
121 | /* | |
122 | * NMIs are edge-triggered, which means if you have enough | |
123 | * of them concurrently, you can lose some because only one | |
124 | * can be latched at any given time. Walk the whole list | |
125 | * to handle those situations. | |
126 | */ | |
2ab00456 | 127 | list_for_each_entry_rcu(a, &desc->head, list) { |
e90c7853 PZ |
128 | int thishandled; |
129 | u64 delta; | |
2ab00456 | 130 | |
e90c7853 | 131 | delta = sched_clock(); |
0c4df02d DH |
132 | thishandled = a->handler(type, regs); |
133 | handled += thishandled; | |
e90c7853 | 134 | delta = sched_clock() - delta; |
0c4df02d | 135 | trace_nmi_handler(a->handler, (int)delta, thishandled); |
2ab00456 | 136 | |
e90c7853 | 137 | if (delta < nmi_longest_ns || delta < a->max_duration) |
2ab00456 DH |
138 | continue; |
139 | ||
e90c7853 PZ |
140 | a->max_duration = delta; |
141 | irq_work_queue(&a->irq_work); | |
2ab00456 | 142 | } |
c9126b2e | 143 | |
c9126b2e DZ |
144 | rcu_read_unlock(); |
145 | ||
146 | /* return total number of NMI events handled */ | |
147 | return handled; | |
148 | } | |
9326638c | 149 | NOKPROBE_SYMBOL(nmi_handle); |
c9126b2e | 150 | |
72b3fb24 | 151 | int __register_nmi_handler(unsigned int type, struct nmiaction *action) |
c9126b2e DZ |
152 | { |
153 | struct nmi_desc *desc = nmi_to_desc(type); | |
154 | unsigned long flags; | |
155 | ||
72b3fb24 LZ |
156 | if (!action->handler) |
157 | return -EINVAL; | |
158 | ||
e90c7853 PZ |
159 | init_irq_work(&action->irq_work, nmi_max_handler); |
160 | ||
c9126b2e DZ |
161 | spin_lock_irqsave(&desc->lock, flags); |
162 | ||
b227e233 DZ |
163 | /* |
164 | * most handlers of type NMI_UNKNOWN never return because | |
165 | * they just assume the NMI is theirs. Just a sanity check | |
166 | * to manage expectations | |
167 | */ | |
168 | WARN_ON_ONCE(type == NMI_UNKNOWN && !list_empty(&desc->head)); | |
553222f3 DZ |
169 | WARN_ON_ONCE(type == NMI_SERR && !list_empty(&desc->head)); |
170 | WARN_ON_ONCE(type == NMI_IO_CHECK && !list_empty(&desc->head)); | |
b227e233 | 171 | |
c9126b2e DZ |
172 | /* |
173 | * some handlers need to be executed first otherwise a fake | |
174 | * event confuses some handlers (kdump uses this flag) | |
175 | */ | |
176 | if (action->flags & NMI_FLAG_FIRST) | |
177 | list_add_rcu(&action->list, &desc->head); | |
178 | else | |
179 | list_add_tail_rcu(&action->list, &desc->head); | |
180 | ||
181 | spin_unlock_irqrestore(&desc->lock, flags); | |
182 | return 0; | |
183 | } | |
72b3fb24 | 184 | EXPORT_SYMBOL(__register_nmi_handler); |
c9126b2e | 185 | |
72b3fb24 | 186 | void unregister_nmi_handler(unsigned int type, const char *name) |
c9126b2e DZ |
187 | { |
188 | struct nmi_desc *desc = nmi_to_desc(type); | |
189 | struct nmiaction *n; | |
190 | unsigned long flags; | |
191 | ||
192 | spin_lock_irqsave(&desc->lock, flags); | |
193 | ||
194 | list_for_each_entry_rcu(n, &desc->head, list) { | |
195 | /* | |
196 | * the name passed in to describe the nmi handler | |
197 | * is used as the lookup key | |
198 | */ | |
199 | if (!strcmp(n->name, name)) { | |
200 | WARN(in_nmi(), | |
201 | "Trying to free NMI (%s) from NMI context!\n", n->name); | |
202 | list_del_rcu(&n->list); | |
203 | break; | |
204 | } | |
205 | } | |
206 | ||
207 | spin_unlock_irqrestore(&desc->lock, flags); | |
208 | synchronize_rcu(); | |
c9126b2e | 209 | } |
c9126b2e DZ |
210 | EXPORT_SYMBOL_GPL(unregister_nmi_handler); |
211 | ||
9326638c | 212 | static void |
1d48922c DZ |
213 | pci_serr_error(unsigned char reason, struct pt_regs *regs) |
214 | { | |
553222f3 DZ |
215 | /* check to see if anyone registered against these types of errors */ |
216 | if (nmi_handle(NMI_SERR, regs, false)) | |
217 | return; | |
218 | ||
1d48922c DZ |
219 | pr_emerg("NMI: PCI system error (SERR) for reason %02x on CPU %d.\n", |
220 | reason, smp_processor_id()); | |
221 | ||
222 | /* | |
223 | * On some machines, PCI SERR line is used to report memory | |
224 | * errors. EDAC makes use of it. | |
225 | */ | |
226 | #if defined(CONFIG_EDAC) | |
227 | if (edac_handler_set()) { | |
228 | edac_atomic_assert_error(); | |
229 | return; | |
230 | } | |
231 | #endif | |
232 | ||
233 | if (panic_on_unrecovered_nmi) | |
234 | panic("NMI: Not continuing"); | |
235 | ||
236 | pr_emerg("Dazed and confused, but trying to continue\n"); | |
237 | ||
238 | /* Clear and disable the PCI SERR error line. */ | |
239 | reason = (reason & NMI_REASON_CLEAR_MASK) | NMI_REASON_CLEAR_SERR; | |
240 | outb(reason, NMI_REASON_PORT); | |
241 | } | |
9326638c | 242 | NOKPROBE_SYMBOL(pci_serr_error); |
1d48922c | 243 | |
9326638c | 244 | static void |
1d48922c DZ |
245 | io_check_error(unsigned char reason, struct pt_regs *regs) |
246 | { | |
247 | unsigned long i; | |
248 | ||
553222f3 DZ |
249 | /* check to see if anyone registered against these types of errors */ |
250 | if (nmi_handle(NMI_IO_CHECK, regs, false)) | |
251 | return; | |
252 | ||
1d48922c DZ |
253 | pr_emerg( |
254 | "NMI: IOCK error (debug interrupt?) for reason %02x on CPU %d.\n", | |
255 | reason, smp_processor_id()); | |
57da8b96 | 256 | show_regs(regs); |
1d48922c DZ |
257 | |
258 | if (panic_on_io_nmi) | |
259 | panic("NMI IOCK error: Not continuing"); | |
260 | ||
261 | /* Re-enable the IOCK line, wait for a few seconds */ | |
262 | reason = (reason & NMI_REASON_CLEAR_MASK) | NMI_REASON_CLEAR_IOCHK; | |
263 | outb(reason, NMI_REASON_PORT); | |
264 | ||
265 | i = 20000; | |
266 | while (--i) { | |
267 | touch_nmi_watchdog(); | |
268 | udelay(100); | |
269 | } | |
270 | ||
271 | reason &= ~NMI_REASON_CLEAR_IOCHK; | |
272 | outb(reason, NMI_REASON_PORT); | |
273 | } | |
9326638c | 274 | NOKPROBE_SYMBOL(io_check_error); |
1d48922c | 275 | |
9326638c | 276 | static void |
1d48922c DZ |
277 | unknown_nmi_error(unsigned char reason, struct pt_regs *regs) |
278 | { | |
9c48f1c6 DZ |
279 | int handled; |
280 | ||
b227e233 DZ |
281 | /* |
282 | * Use 'false' as back-to-back NMIs are dealt with one level up. | |
283 | * Of course this makes having multiple 'unknown' handlers useless | |
284 | * as only the first one is ever run (unless it can actually determine | |
285 | * if it caused the NMI) | |
286 | */ | |
287 | handled = nmi_handle(NMI_UNKNOWN, regs, false); | |
efc3aac5 DZ |
288 | if (handled) { |
289 | __this_cpu_add(nmi_stats.unknown, handled); | |
1d48922c | 290 | return; |
efc3aac5 DZ |
291 | } |
292 | ||
293 | __this_cpu_add(nmi_stats.unknown, 1); | |
294 | ||
1d48922c DZ |
295 | pr_emerg("Uhhuh. NMI received for unknown reason %02x on CPU %d.\n", |
296 | reason, smp_processor_id()); | |
297 | ||
298 | pr_emerg("Do you have a strange power saving mode enabled?\n"); | |
299 | if (unknown_nmi_panic || panic_on_unrecovered_nmi) | |
300 | panic("NMI: Not continuing"); | |
301 | ||
302 | pr_emerg("Dazed and confused, but trying to continue\n"); | |
303 | } | |
9326638c | 304 | NOKPROBE_SYMBOL(unknown_nmi_error); |
1d48922c | 305 | |
b227e233 DZ |
306 | static DEFINE_PER_CPU(bool, swallow_nmi); |
307 | static DEFINE_PER_CPU(unsigned long, last_nmi_rip); | |
308 | ||
9326638c | 309 | static void default_do_nmi(struct pt_regs *regs) |
1d48922c DZ |
310 | { |
311 | unsigned char reason = 0; | |
9c48f1c6 | 312 | int handled; |
b227e233 | 313 | bool b2b = false; |
1d48922c DZ |
314 | |
315 | /* | |
316 | * CPU-specific NMI must be processed before non-CPU-specific | |
317 | * NMI, otherwise we may lose it, because the CPU-specific | |
318 | * NMI can not be detected/processed on other CPUs. | |
319 | */ | |
b227e233 DZ |
320 | |
321 | /* | |
322 | * Back-to-back NMIs are interesting because they can either | |
323 | * be two NMI or more than two NMIs (any thing over two is dropped | |
324 | * due to NMI being edge-triggered). If this is the second half | |
325 | * of the back-to-back NMI, assume we dropped things and process | |
326 | * more handlers. Otherwise reset the 'swallow' NMI behaviour | |
327 | */ | |
328 | if (regs->ip == __this_cpu_read(last_nmi_rip)) | |
329 | b2b = true; | |
330 | else | |
331 | __this_cpu_write(swallow_nmi, false); | |
332 | ||
333 | __this_cpu_write(last_nmi_rip, regs->ip); | |
334 | ||
335 | handled = nmi_handle(NMI_LOCAL, regs, b2b); | |
efc3aac5 | 336 | __this_cpu_add(nmi_stats.normal, handled); |
b227e233 DZ |
337 | if (handled) { |
338 | /* | |
339 | * There are cases when a NMI handler handles multiple | |
340 | * events in the current NMI. One of these events may | |
341 | * be queued for in the next NMI. Because the event is | |
342 | * already handled, the next NMI will result in an unknown | |
343 | * NMI. Instead lets flag this for a potential NMI to | |
344 | * swallow. | |
345 | */ | |
346 | if (handled > 1) | |
347 | __this_cpu_write(swallow_nmi, true); | |
1d48922c | 348 | return; |
b227e233 | 349 | } |
1d48922c DZ |
350 | |
351 | /* Non-CPU-specific NMI: NMI sources can be processed on any CPU */ | |
352 | raw_spin_lock(&nmi_reason_lock); | |
064a59b6 | 353 | reason = x86_platform.get_nmi_reason(); |
1d48922c DZ |
354 | |
355 | if (reason & NMI_REASON_MASK) { | |
356 | if (reason & NMI_REASON_SERR) | |
357 | pci_serr_error(reason, regs); | |
358 | else if (reason & NMI_REASON_IOCHK) | |
359 | io_check_error(reason, regs); | |
360 | #ifdef CONFIG_X86_32 | |
361 | /* | |
362 | * Reassert NMI in case it became active | |
363 | * meanwhile as it's edge-triggered: | |
364 | */ | |
365 | reassert_nmi(); | |
366 | #endif | |
efc3aac5 | 367 | __this_cpu_add(nmi_stats.external, 1); |
1d48922c DZ |
368 | raw_spin_unlock(&nmi_reason_lock); |
369 | return; | |
370 | } | |
371 | raw_spin_unlock(&nmi_reason_lock); | |
372 | ||
b227e233 DZ |
373 | /* |
374 | * Only one NMI can be latched at a time. To handle | |
375 | * this we may process multiple nmi handlers at once to | |
376 | * cover the case where an NMI is dropped. The downside | |
377 | * to this approach is we may process an NMI prematurely, | |
378 | * while its real NMI is sitting latched. This will cause | |
379 | * an unknown NMI on the next run of the NMI processing. | |
380 | * | |
381 | * We tried to flag that condition above, by setting the | |
382 | * swallow_nmi flag when we process more than one event. | |
383 | * This condition is also only present on the second half | |
384 | * of a back-to-back NMI, so we flag that condition too. | |
385 | * | |
386 | * If both are true, we assume we already processed this | |
387 | * NMI previously and we swallow it. Otherwise we reset | |
388 | * the logic. | |
389 | * | |
390 | * There are scenarios where we may accidentally swallow | |
391 | * a 'real' unknown NMI. For example, while processing | |
392 | * a perf NMI another perf NMI comes in along with a | |
393 | * 'real' unknown NMI. These two NMIs get combined into | |
394 | * one (as descibed above). When the next NMI gets | |
395 | * processed, it will be flagged by perf as handled, but | |
396 | * noone will know that there was a 'real' unknown NMI sent | |
397 | * also. As a result it gets swallowed. Or if the first | |
398 | * perf NMI returns two events handled then the second | |
399 | * NMI will get eaten by the logic below, again losing a | |
400 | * 'real' unknown NMI. But this is the best we can do | |
401 | * for now. | |
402 | */ | |
403 | if (b2b && __this_cpu_read(swallow_nmi)) | |
efc3aac5 | 404 | __this_cpu_add(nmi_stats.swallow, 1); |
b227e233 DZ |
405 | else |
406 | unknown_nmi_error(reason, regs); | |
1d48922c | 407 | } |
9326638c | 408 | NOKPROBE_SYMBOL(default_do_nmi); |
1d48922c | 409 | |
ccd49c23 SR |
410 | /* |
411 | * NMIs can hit breakpoints which will cause it to lose its | |
412 | * NMI context with the CPU when the breakpoint does an iret. | |
413 | */ | |
414 | #ifdef CONFIG_X86_32 | |
415 | /* | |
416 | * For i386, NMIs use the same stack as the kernel, and we can | |
c7d65a78 SR |
417 | * add a workaround to the iret problem in C (preventing nested |
418 | * NMIs if an NMI takes a trap). Simply have 3 states the NMI | |
419 | * can be in: | |
ccd49c23 SR |
420 | * |
421 | * 1) not running | |
422 | * 2) executing | |
423 | * 3) latched | |
424 | * | |
425 | * When no NMI is in progress, it is in the "not running" state. | |
426 | * When an NMI comes in, it goes into the "executing" state. | |
427 | * Normally, if another NMI is triggered, it does not interrupt | |
428 | * the running NMI and the HW will simply latch it so that when | |
429 | * the first NMI finishes, it will restart the second NMI. | |
430 | * (Note, the latch is binary, thus multiple NMIs triggering, | |
431 | * when one is running, are ignored. Only one NMI is restarted.) | |
432 | * | |
433 | * If an NMI hits a breakpoint that executes an iret, another | |
434 | * NMI can preempt it. We do not want to allow this new NMI | |
435 | * to run, but we want to execute it when the first one finishes. | |
c7d65a78 SR |
436 | * We set the state to "latched", and the exit of the first NMI will |
437 | * perform a dec_return, if the result is zero (NOT_RUNNING), then | |
438 | * it will simply exit the NMI handler. If not, the dec_return | |
439 | * would have set the state to NMI_EXECUTING (what we want it to | |
440 | * be when we are running). In this case, we simply jump back | |
441 | * to rerun the NMI handler again, and restart the 'latched' NMI. | |
442 | * | |
443 | * No trap (breakpoint or page fault) should be hit before nmi_restart, | |
444 | * thus there is no race between the first check of state for NOT_RUNNING | |
445 | * and setting it to NMI_EXECUTING. The HW will prevent nested NMIs | |
446 | * at this point. | |
70fb74a5 SR |
447 | * |
448 | * In case the NMI takes a page fault, we need to save off the CR2 | |
449 | * because the NMI could have preempted another page fault and corrupt | |
450 | * the CR2 that is about to be read. As nested NMIs must be restarted | |
451 | * and they can not take breakpoints or page faults, the update of the | |
452 | * CR2 must be done before converting the nmi state back to NOT_RUNNING. | |
453 | * Otherwise, there would be a race of another nested NMI coming in | |
454 | * after setting state to NOT_RUNNING but before updating the nmi_cr2. | |
ccd49c23 SR |
455 | */ |
456 | enum nmi_states { | |
c7d65a78 | 457 | NMI_NOT_RUNNING = 0, |
ccd49c23 SR |
458 | NMI_EXECUTING, |
459 | NMI_LATCHED, | |
460 | }; | |
461 | static DEFINE_PER_CPU(enum nmi_states, nmi_state); | |
70fb74a5 | 462 | static DEFINE_PER_CPU(unsigned long, nmi_cr2); |
ccd49c23 SR |
463 | |
464 | #define nmi_nesting_preprocess(regs) \ | |
465 | do { \ | |
c7d65a78 SR |
466 | if (this_cpu_read(nmi_state) != NMI_NOT_RUNNING) { \ |
467 | this_cpu_write(nmi_state, NMI_LATCHED); \ | |
ccd49c23 SR |
468 | return; \ |
469 | } \ | |
c7d65a78 | 470 | this_cpu_write(nmi_state, NMI_EXECUTING); \ |
70fb74a5 | 471 | this_cpu_write(nmi_cr2, read_cr2()); \ |
c7d65a78 SR |
472 | } while (0); \ |
473 | nmi_restart: | |
ccd49c23 SR |
474 | |
475 | #define nmi_nesting_postprocess() \ | |
476 | do { \ | |
70fb74a5 SR |
477 | if (unlikely(this_cpu_read(nmi_cr2) != read_cr2())) \ |
478 | write_cr2(this_cpu_read(nmi_cr2)); \ | |
c7d65a78 | 479 | if (this_cpu_dec_return(nmi_state)) \ |
ccd49c23 SR |
480 | goto nmi_restart; \ |
481 | } while (0) | |
482 | #else /* x86_64 */ | |
483 | /* | |
484 | * In x86_64 things are a bit more difficult. This has the same problem | |
485 | * where an NMI hitting a breakpoint that calls iret will remove the | |
486 | * NMI context, allowing a nested NMI to enter. What makes this more | |
487 | * difficult is that both NMIs and breakpoints have their own stack. | |
488 | * When a new NMI or breakpoint is executed, the stack is set to a fixed | |
489 | * point. If an NMI is nested, it will have its stack set at that same | |
490 | * fixed address that the first NMI had, and will start corrupting the | |
491 | * stack. This is handled in entry_64.S, but the same problem exists with | |
492 | * the breakpoint stack. | |
493 | * | |
494 | * If a breakpoint is being processed, and the debug stack is being used, | |
495 | * if an NMI comes in and also hits a breakpoint, the stack pointer | |
496 | * will be set to the same fixed address as the breakpoint that was | |
497 | * interrupted, causing that stack to be corrupted. To handle this case, | |
498 | * check if the stack that was interrupted is the debug stack, and if | |
499 | * so, change the IDT so that new breakpoints will use the current stack | |
500 | * and not switch to the fixed address. On return of the NMI, switch back | |
501 | * to the original IDT. | |
502 | */ | |
503 | static DEFINE_PER_CPU(int, update_debug_stack); | |
228bdaa9 | 504 | |
ccd49c23 SR |
505 | static inline void nmi_nesting_preprocess(struct pt_regs *regs) |
506 | { | |
228bdaa9 SR |
507 | /* |
508 | * If we interrupted a breakpoint, it is possible that | |
509 | * the nmi handler will have breakpoints too. We need to | |
510 | * change the IDT such that breakpoints that happen here | |
511 | * continue to use the NMI stack. | |
512 | */ | |
513 | if (unlikely(is_debug_stack(regs->sp))) { | |
514 | debug_stack_set_zero(); | |
c0525a69 | 515 | this_cpu_write(update_debug_stack, 1); |
228bdaa9 | 516 | } |
ccd49c23 SR |
517 | } |
518 | ||
519 | static inline void nmi_nesting_postprocess(void) | |
520 | { | |
c0525a69 | 521 | if (unlikely(this_cpu_read(update_debug_stack))) { |
ccd49c23 | 522 | debug_stack_reset(); |
c0525a69 SR |
523 | this_cpu_write(update_debug_stack, 0); |
524 | } | |
ccd49c23 SR |
525 | } |
526 | #endif | |
527 | ||
9326638c | 528 | dotraplinkage notrace void |
ccd49c23 SR |
529 | do_nmi(struct pt_regs *regs, long error_code) |
530 | { | |
531 | nmi_nesting_preprocess(regs); | |
532 | ||
1d48922c DZ |
533 | nmi_enter(); |
534 | ||
535 | inc_irq_stat(__nmi_count); | |
536 | ||
537 | if (!ignore_nmis) | |
538 | default_do_nmi(regs); | |
539 | ||
540 | nmi_exit(); | |
228bdaa9 | 541 | |
ccd49c23 SR |
542 | /* On i386, may loop back to preprocess */ |
543 | nmi_nesting_postprocess(); | |
1d48922c | 544 | } |
9326638c | 545 | NOKPROBE_SYMBOL(do_nmi); |
1d48922c DZ |
546 | |
547 | void stop_nmi(void) | |
548 | { | |
549 | ignore_nmis++; | |
550 | } | |
551 | ||
552 | void restart_nmi(void) | |
553 | { | |
554 | ignore_nmis--; | |
555 | } | |
b227e233 DZ |
556 | |
557 | /* reset the back-to-back NMI logic */ | |
558 | void local_touch_nmi(void) | |
559 | { | |
560 | __this_cpu_write(last_nmi_rip, 0); | |
561 | } | |
29c6fb7b | 562 | EXPORT_SYMBOL_GPL(local_touch_nmi); |