summaryrefslogtreecommitdiffstats
path: root/arch/x86/platform/uv/uv_nmi.c
blob: 8dd80050d705eea6099212724f953de8364355ff (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
/*
 * SGI NMI support routines
 *
 *  This program is free software; you can redistribute it and/or modify
 *  it under the terms of the GNU General Public License as published by
 *  the Free Software Foundation; either version 2 of the License, or
 *  (at your option) any later version.
 *
 *  This program is distributed in the hope that it will be useful,
 *  but WITHOUT ANY WARRANTY; without even the implied warranty of
 *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 *  GNU General Public License for more details.
 *
 *  You should have received a copy of the GNU General Public License
 *  along with this program; if not, write to the Free Software
 *  Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307 USA
 *
 *  Copyright (c) 2009-2013 Silicon Graphics, Inc.  All Rights Reserved.
 *  Copyright (c) Mike Travis
 */

#include <linux/cpu.h>
#include <linux/delay.h>
#include <linux/kdb.h>
#include <linux/kexec.h>
#include <linux/kgdb.h>
#include <linux/module.h>
#include <linux/nmi.h>
#include <linux/sched.h>
#include <linux/slab.h>
#include <linux/clocksource.h>

#include <asm/apic.h>
#include <asm/current.h>
#include <asm/kdebug.h>
#include <asm/local64.h>
#include <asm/nmi.h>
#include <asm/traps.h>
#include <asm/uv/uv.h>
#include <asm/uv/uv_hub.h>
#include <asm/uv/uv_mmrs.h>

/*
 * UV handler for NMI
 *
 * Handle system-wide NMI events generated by the global 'power nmi' command.
 *
 * Basic operation is to field the NMI interrupt on each cpu and wait
 * until all cpus have arrived into the nmi handler.  If some cpus do not
 * make it into the handler, try and force them in with the IPI(NMI) signal.
 *
 * We also have to lessen UV Hub MMR accesses as much as possible as this
 * disrupts the UV Hub's primary mission of directing NumaLink traffic and
 * can cause system problems to occur.
 *
 * To do this we register our primary NMI notifier on the NMI_UNKNOWN
 * chain.  This reduces the number of false NMI calls when the perf
 * tools are running which generate an enormous number of NMIs per
 * second (~4M/s for 1024 cpu threads).  Our secondary NMI handler is
 * very short as it only checks that if it has been "pinged" with the
 * IPI(NMI) signal as mentioned above, and does not read the UV Hub's MMR.
 *
 */

static struct uv_hub_nmi_s **uv_hub_nmi_list;

DEFINE_PER_CPU(struct uv_cpu_nmi_s, uv_cpu_nmi);
EXPORT_PER_CPU_SYMBOL_GPL(uv_cpu_nmi);

static unsigned long nmi_mmr;
static unsigned long nmi_mmr_clear;
static unsigned long nmi_mmr_pending;

static atomic_t	uv_in_nmi;
static atomic_t uv_nmi_cpu = ATOMIC_INIT(-1);
static atomic_t uv_nmi_cpus_in_nmi = ATOMIC_INIT(-1);
static atomic_t uv_nmi_slave_continue;
static cpumask_var_t uv_nmi_cpu_mask;

/* Values for uv_nmi_slave_continue */
#define SLAVE_CLEAR	0
#define SLAVE_CONTINUE	1
#define SLAVE_EXIT	2

/*
 * Default is all stack dumps go to the console and buffer.
 * Lower level to send to log buffer only.
 */
static int uv_nmi_loglevel = CONSOLE_LOGLEVEL_DEFAULT;
module_param_named(dump_loglevel, uv_nmi_loglevel, int, 0644);

/*
 * The following values show statistics on how perf events are affecting
 * this system.
 */
static int param_get_local64(char *buffer, const struct kernel_param *kp)
{
	return sprintf(buffer, "%lu\n", local64_read((local64_t *)kp->arg));
}

static int param_set_local64(const char *val, const struct kernel_param *kp)
{
	/* clear on any write */
	local64_set((local64_t *)kp->arg, 0);
	return 0;
}

static const struct kernel_param_ops param_ops_local64 = {
	.get = param_get_local64,
	.set = param_set_local64,
};
#define param_check_local64(name, p) __param_check(name, p, local64_t)

static local64_t uv_nmi_count;
module_param_named(nmi_count, uv_nmi_count, local64, 0644);

static local64_t uv_nmi_misses;
module_param_named(nmi_misses, uv_nmi_misses, local64, 0644);

static local64_t uv_nmi_ping_count;
module_param_named(ping_count, uv_nmi_ping_count, local64, 0644);

static local64_t uv_nmi_ping_misses;
module_param_named(ping_misses, uv_nmi_ping_misses, local64, 0644);

/*
 * Following values allow tuning for large systems under heavy loading
 */
static int uv_nmi_initial_delay = 100;
module_param_named(initial_delay, uv_nmi_initial_delay, int, 0644);

static int uv_nmi_slave_delay = 100;
module_param_named(slave_delay, uv_nmi_slave_delay, int, 0644);

static int uv_nmi_loop_delay = 100;
module_param_named(loop_delay, uv_nmi_loop_delay, int, 0644);

static int uv_nmi_trigger_delay = 10000;
module_param_named(trigger_delay, uv_nmi_trigger_delay, int, 0644);

static int uv_nmi_wait_count = 100;
module_param_named(wait_count, uv_nmi_wait_count, int, 0644);

static int uv_nmi_retry_count = 500;
module_param_named(retry_count, uv_nmi_retry_count, int, 0644);

/*
 * Valid NMI Actions:
 *  "dump"	- dump process stack for each cpu
 *  "ips"	- dump IP info for each cpu
 *  "kdump"	- do crash dump
 *  "kdb"	- enter KDB (default)
 *  "kgdb"	- enter KGDB
 */
static char uv_nmi_action[8] = "kdb";
module_param_string(action, uv_nmi_action, sizeof(uv_nmi_action), 0644);

static inline bool uv_nmi_action_is(const char *action)
{
	return (strncmp(uv_nmi_action, action, strlen(action)) == 0);
}

/* Setup which NMI support is present in system */
static void uv_nmi_setup_mmrs(void)
{
	if (uv_read_local_mmr(UVH_NMI_MMRX_SUPPORTED)) {
		uv_write_local_mmr(UVH_NMI_MMRX_REQ,
					1UL << UVH_NMI_MMRX_REQ_SHIFT);
		nmi_mmr = UVH_NMI_MMRX;
		nmi_mmr_clear = UVH_NMI_MMRX_CLEAR;
		nmi_mmr_pending = 1UL << UVH_NMI_MMRX_SHIFT;
		pr_info("UV: SMI NMI support: %s\n", UVH_NMI_MMRX_TYPE);
	} else {
		nmi_mmr = UVH_NMI_MMR;
		nmi_mmr_clear = UVH_NMI_MMR_CLEAR;
		nmi_mmr_pending = 1UL << UVH_NMI_MMR_SHIFT;
		pr_info("UV: SMI NMI support: %s\n", UVH_NMI_MMR_TYPE);
	}
}

/* Read NMI MMR and check if NMI flag was set by BMC. */
static inline int uv_nmi_test_mmr(struct uv_hub_nmi_s *hub_nmi)
{
	hub_nmi->nmi_value = uv_read_local_mmr(nmi_mmr);
	atomic_inc(&hub_nmi->read_mmr_count);
	return !!(hub_nmi->nmi_value & nmi_mmr_pending);
}

static inline void uv_local_mmr_clear_nmi(void)
{
	uv_write_local_mmr(nmi_mmr_clear, nmi_mmr_pending);
}

/*
 * If first cpu in on this hub, set hub_nmi "in_nmi" and "owner" values and
 * return true.  If first cpu in on the system, set global "in_nmi" flag.
 */
static int uv_set_in_nmi(int cpu, struct uv_hub_nmi_s *hub_nmi)
{
	int first = atomic_add_unless(&hub_nmi->in_nmi, 1, 1);

	if (first) {
		atomic_set(&hub_nmi->cpu_owner, cpu);
		if (atomic_add_unless(&uv_in_nmi, 1, 1))
			atomic_set(&uv_nmi_cpu, cpu);

		atomic_inc(&hub_nmi->nmi_count);
	}
	return first;
}

/* Check if this is a system NMI event */
static int uv_check_nmi(struct uv_hub_nmi_s *hub_nmi)
{
	int cpu = smp_processor_id();
	int nmi = 0;

	local64_inc(&uv_nmi_count);
	this_cpu_inc(uv_cpu_nmi.queries);

	do {
		nmi = atomic_read(&hub_nmi->in_nmi);
		if (nmi)
			break;

		if (raw_spin_trylock(&hub_nmi->nmi_lock)) {

			/* check hub MMR NMI flag */
			if (uv_nmi_test_mmr(hub_nmi)) {
				uv_set_in_nmi(cpu, hub_nmi);
				nmi = 1;
				break;
			}

			/* MMR NMI flag is clear */
			raw_spin_unlock(&hub_nmi->nmi_lock);

		} else {
			/* wait a moment for the hub nmi locker to set flag */
			cpu_relax();
			udelay(uv_nmi_slave_delay);

			/* re-check hub in_nmi flag */
			nmi = atomic_read(&hub_nmi->in_nmi);
			if (nmi)
				break;
		}

		/* check if this BMC missed setting the MMR NMI flag */
		if (!nmi) {
			nmi = atomic_read(&uv_in_nmi);
			if (nmi)
				uv_set_in_nmi(cpu, hub_nmi);
		}

	} while (0);

	if (!nmi)
		local64_inc(&uv_nmi_misses);

	return nmi;
}

/* Need to reset the NMI MMR register, but only once per hub. */
static inline void uv_clear_nmi(int cpu)
{
	struct uv_hub_nmi_s *hub_nmi = uv_hub_nmi;

	if (cpu == atomic_read(&hub_nmi->cpu_owner)) {
		atomic_set(&hub_nmi->cpu_owner, -1);
		atomic_set(&hub_nmi->in_nmi, 0);
		uv_local_mmr_clear_nmi();
		raw_spin_unlock(&hub_nmi->nmi_lock);
	}
}

/* Ping non-responding cpus attemping to force them into the NMI handler */
static void uv_nmi_nr_cpus_ping(void)
{
	int cpu;

	for_each_cpu(cpu, uv_nmi_cpu_mask)
		uv_cpu_nmi_per(cpu).pinging = 1;

	apic->send_IPI_mask(uv_nmi_cpu_mask, APIC_DM_NMI);
}

/* Clean up flags for cpus that ignored both NMI and ping */
static void uv_nmi_cleanup_mask(void)
{
	int cpu;

	for_each_cpu(cpu, uv_nmi_cpu_mask) {
		uv_cpu_nmi_per(cpu).pinging =  0;
		uv_cpu_nmi_per(cpu).state = UV_NMI_STATE_OUT;
		cpumask_clear_cpu(cpu, uv_nmi_cpu_mask);
	}
}

/* Loop waiting as cpus enter nmi handler */
static int uv_nmi_wait_cpus(int first)
{
	int i, j, k, n = num_online_cpus();
	int last_k = 0, waiting = 0;

	if (first) {
		cpumask_copy(uv_nmi_cpu_mask, cpu_online_mask);
		k = 0;
	} else {
		k = n - cpumask_weight(uv_nmi_cpu_mask);
	}

	udelay(uv_nmi_initial_delay);
	for (i = 0; i < uv_nmi_retry_count; i++) {
		int loop_delay = uv_nmi_loop_delay;

		for_each_cpu(j, uv_nmi_cpu_mask) {
			if (uv_cpu_nmi_per(j).state) {
				cpumask_clear_cpu(j, uv_nmi_cpu_mask);
				if (++k >= n)
					break;
			}
		}
		if (k >= n) {		/* all in? */
			k = n;
			break;
		}
		if (last_k != k) {	/* abort if no new cpus coming in */
			last_k = k;
			waiting = 0;
		} else if (++waiting > uv_nmi_wait_count)
			break;

		/* extend delay if waiting only for cpu 0 */
		if (waiting && (n - k) == 1 &&
		    cpumask_test_cpu(0, uv_nmi_cpu_mask))
			loop_delay *= 100;

		udelay(loop_delay);
	}
	atomic_set(&uv_nmi_cpus_in_nmi, k);
	return n - k;
}

/* Wait until all slave cpus have entered UV NMI handler */
static void uv_nmi_wait(int master)
{
	/* indicate this cpu is in */
	this_cpu_write(uv_cpu_nmi.state, UV_NMI_STATE_IN);

	/* if not the first cpu in (the master), then we are a slave cpu */
	if (!master)
		return;

	do {
		/* wait for all other cpus to gather here */
		if (!uv_nmi_wait_cpus(1))
			break;

		/* if not all made it in, send IPI NMI to them */
		pr_alert("UV: Sending NMI IPI to %d non-responding CPUs: %*pbl\n",
			 cpumask_weight(uv_nmi_cpu_mask),
			 cpumask_pr_args(uv_nmi_cpu_mask));

		uv_nmi_nr_cpus_ping();

		/* if all cpus are in, then done */
		if (!uv_nmi_wait_cpus(0))
			break;

		pr_alert("UV: %d CPUs not in NMI loop: %*pbl\n",
			 cpumask_weight(uv_nmi_cpu_mask),
			 cpumask_pr_args(uv_nmi_cpu_mask));
	} while (0);

	pr_alert("UV: %d of %d CPUs in NMI\n",
		atomic_read(&uv_nmi_cpus_in_nmi), num_online_cpus());
}

/* Dump Instruction Pointer header */
static void uv_nmi_dump_cpu_ip_hdr(void)
{
	pr_info("\nUV: %4s %6s %-32s %s   (Note: PID 0 not listed)\n",
		"CPU", "PID", "COMMAND", "IP");
}

/* Dump Instruction Pointer info */
static void uv_nmi_dump_cpu_ip(int cpu, struct pt_regs *regs)
{
	pr_info("UV: %4d %6d %-32.32s ", cpu, current->pid, current->comm);
	printk_address(regs->ip);
}

/*
 * Dump this CPU's state.  If action was set to "kdump" and the crash_kexec
 * failed, then we provide "dump" as an alternate action.  Action "dump" now
 * also includes the show "ips" (instruction pointers) action whereas the
 * action "ips" only displays instruction pointers for the non-idle CPU's.
 * This is an abbreviated form of the "ps" command.
 */
static void uv_nmi_dump_state_cpu(int cpu, struct pt_regs *regs)
{
	const char *dots = " ................................. ";

	if (cpu == 0)
		uv_nmi_dump_cpu_ip_hdr();

	if (current->pid != 0 || !uv_nmi_action_is("ips"))
		uv_nmi_dump_cpu_ip(cpu, regs);

	if (uv_nmi_action_is("dump")) {
		pr_info("UV:%sNMI process trace for CPU %d\n", dots, cpu);
		show_regs(regs);
	}

	this_cpu_write(uv_cpu_nmi.state, UV_NMI_STATE_DUMP_DONE);
}

/* Trigger a slave cpu to dump it's state */
static void uv_nmi_trigger_dump(int cpu)
{
	int retry = uv_nmi_trigger_delay;

	if (uv_cpu_nmi_per(cpu).state != UV_NMI_STATE_IN)
		return;

	uv_cpu_nmi_per(cpu).state = UV_NMI_STATE_DUMP;
	do {
		cpu_relax();
		udelay(10);
		if (uv_cpu_nmi_per(cpu).state
				!= UV_NMI_STATE_DUMP)
			return;
	} while (--retry > 0);

	pr_crit("UV: CPU %d stuck in process dump function\n", cpu);
	uv_cpu_nmi_per(cpu).state = UV_NMI_STATE_DUMP_DONE;
}

/* Wait until all cpus ready to exit */
static void uv_nmi_sync_exit(int master)
{
	atomic_dec(&uv_nmi_cpus_in_nmi);
	if (master) {
		while (atomic_read(&uv_nmi_cpus_in_nmi) > 0)
			cpu_relax();
		atomic_set(&uv_nmi_slave_continue, SLAVE_CLEAR);
	} else {
		while (atomic_read(&uv_nmi_slave_continue))
			cpu_relax();
	}
}

/* Walk through cpu list and dump state of each */
static void uv_nmi_dump_state(int cpu, struct pt_regs *regs, int master)
{
	if (master) {
		int tcpu;
		int ignored = 0;
		int saved_console_loglevel = console_loglevel;

		pr_alert("UV: tracing %s for %d CPUs from CPU %d\n",
			uv_nmi_action_is("ips") ? "IPs" : "processes",
			atomic_read(&uv_nmi_cpus_in_nmi), cpu);

		console_loglevel = uv_nmi_loglevel;
		atomic_set(&uv_nmi_slave_continue, SLAVE_EXIT);
		for_each_online_cpu(tcpu) {
			if (cpumask_test_cpu(tcpu, uv_nmi_cpu_mask))
				ignored++;
			else if (tcpu == cpu)
				uv_nmi_dump_state_cpu(tcpu, regs);
			else
				uv_nmi_trigger_dump(tcpu);
		}
		if (ignored)
			pr_alert("UV: %d CPUs ignored NMI\n", ignored);

		console_loglevel = saved_console_loglevel;
		pr_alert("UV: process trace complete\n");
	} else {
		while (!atomic_read(&uv_nmi_slave_continue))
			cpu_relax();
		while (this_cpu_read(uv_cpu_nmi.state) != UV_NMI_STATE_DUMP)
			cpu_relax();
		uv_nmi_dump_state_cpu(cpu, regs);
	}
	uv_nmi_sync_exit(master);
}

static void uv_nmi_touch_watchdogs(void)
{
	touch_softlockup_watchdog_sync();
	clocksource_touch_watchdog();
	rcu_cpu_stall_reset();
	touch_nmi_watchdog();
}

static atomic_t uv_nmi_kexec_failed;

#if defined(CONFIG_KEXEC_CORE)
static void uv_nmi_kdump(int cpu, int master, struct pt_regs *regs)
{
	/* Call crash to dump system state */
	if (master) {
		pr_emerg("UV: NMI executing crash_kexec on CPU%d\n", cpu);
		crash_kexec(regs);

		pr_emerg("UV: crash_kexec unexpectedly returned, ");
		atomic_set(&uv_nmi_kexec_failed, 1);
		if (!kexec_crash_image) {
			pr_cont("crash kernel not loaded\n");
			return;
		}
		pr_cont("kexec busy, stalling cpus while waiting\n");
	}

	/* If crash exec fails the slaves should return, otherwise stall */
	while (atomic_read(&uv_nmi_kexec_failed) == 0)
		mdelay(10);
}

#else /* !CONFIG_KEXEC_CORE */
static inline void uv_nmi_kdump(int cpu, int master, struct pt_regs *regs)
{
	if (master)
		pr_err("UV: NMI kdump: KEXEC not supported in this kernel\n");
	atomic_set(&uv_nmi_kexec_failed, 1);
}
#endif /* !CONFIG_KEXEC_CORE */

#ifdef CONFIG_KGDB
#ifdef CONFIG_KGDB_KDB
static inline int uv_nmi_kdb_reason(void)
{
	return KDB_REASON_SYSTEM_NMI;
}
#else /* !CONFIG_KGDB_KDB */
static inline int uv_nmi_kdb_reason(void)
{
	/* Insure user is expecting to attach gdb remote */
	if (uv_nmi_action_is("kgdb"))
		return 0;

	pr_err("UV: NMI error: KDB is not enabled in this kernel\n");
	return -1;
}
#endif /* CONFIG_KGDB_KDB */

/*
 * Call KGDB/KDB from NMI handler
 *
 * Note that if both KGDB and KDB are configured, then the action of 'kgdb' or
 * 'kdb' has no affect on which is used.  See the KGDB documention for further
 * information.
 */
static void uv_call_kgdb_kdb(int cpu, struct pt_regs *regs, int master)
{
	if (master) {
		int reason = uv_nmi_kdb_reason();
		int ret;

		if (reason < 0)
			return;

		/* call KGDB NMI handler as MASTER */
		ret = kgdb_nmicallin(cpu, X86_TRAP_NMI, regs, reason,
				&uv_nmi_slave_continue);
		if (ret) {
			pr_alert("KGDB returned error, is kgdboc set?\n");
			atomic_set(&uv_nmi_slave_continue, SLAVE_EXIT);
		}
	} else {
		/* wait for KGDB signal that it's ready for slaves to enter */
		int sig;

		do {
			cpu_relax();
			sig = atomic_read(&uv_nmi_slave_continue);
		} while (!sig);

		/* call KGDB as slave */
		if (sig == SLAVE_CONTINUE)
			kgdb_nmicallback(cpu, regs);
	}
	uv_nmi_sync_exit(master);
}

#else /* !CONFIG_KGDB */
static inline void uv_call_kgdb_kdb(int cpu, struct pt_regs *regs, int master)
{
	pr_err("UV: NMI error: KGDB is not enabled in this kernel\n");
}
#endif /* !CONFIG_KGDB */

/*
 * UV NMI handler
 */
int uv_handle_nmi(unsigned int reason, struct pt_regs *regs)
{
	struct uv_hub_nmi_s *hub_nmi = uv_hub_nmi;
	int cpu = smp_processor_id();
	int master = 0;
	unsigned long flags;

	local_irq_save(flags);

	/* If not a UV System NMI, ignore */
	if (!this_cpu_read(uv_cpu_nmi.pinging) && !uv_check_nmi(hub_nmi)) {
		local_irq_restore(flags);
		return NMI_DONE;
	}

	/* Indicate we are the first CPU into the NMI handler */
	master = (atomic_read(&uv_nmi_cpu) == cpu);

	/* If NMI action is "kdump", then attempt to do it */
	if (uv_nmi_action_is("kdump")) {
		uv_nmi_kdump(cpu, master, regs);

		/* Unexpected return, revert action to "dump" */
		if (master)
			strncpy(uv_nmi_action, "dump", strlen(uv_nmi_action));
	}

	/* Pause as all cpus enter the NMI handler */
	uv_nmi_wait(master);

	/* Dump state of each cpu */
	if (uv_nmi_action_is("ips") || uv_nmi_action_is("dump"))
		uv_nmi_dump_state(cpu, regs, master);

	/* Call KGDB/KDB if enabled */
	else if (uv_nmi_action_is("kdb") || uv_nmi_action_is("kgdb"))
		uv_call_kgdb_kdb(cpu, regs, master);

	/* Clear per_cpu "in nmi" flag */
	this_cpu_write(uv_cpu_nmi.state, UV_NMI_STATE_OUT);

	/* Clear MMR NMI flag on each hub */
	uv_clear_nmi(cpu);

	/* Clear global flags */
	if (master) {
		if (cpumask_weight(uv_nmi_cpu_mask))
			uv_nmi_cleanup_mask();
		atomic_set(&uv_nmi_cpus_in_nmi, -1);
		atomic_set(&uv_nmi_cpu, -1);
		atomic_set(&uv_in_nmi, 0);
		atomic_set(&uv_nmi_kexec_failed, 0);
	}

	uv_nmi_touch_watchdogs();
	local_irq_restore(flags);

	return NMI_HANDLED;
}

/*
 * NMI handler for pulling in CPUs when perf events are grabbing our NMI
 */
static int uv_handle_nmi_ping(unsigned int reason, struct pt_regs *regs)
{
	int ret;

	this_cpu_inc(uv_cpu_nmi.queries);
	if (!this_cpu_read(uv_cpu_nmi.pinging)) {
		local64_inc(&uv_nmi_ping_misses);
		return NMI_DONE;
	}

	this_cpu_inc(uv_cpu_nmi.pings);
	local64_inc(&uv_nmi_ping_count);
	ret = uv_handle_nmi(reason, regs);
	this_cpu_write(uv_cpu_nmi.pinging, 0);
	return ret;
}

static void uv_register_nmi_notifier(void)
{
	if (register_nmi_handler(NMI_UNKNOWN, uv_handle_nmi, 0, "uv"))
		pr_warn("UV: NMI handler failed to register\n");

	if (register_nmi_handler(NMI_LOCAL, uv_handle_nmi_ping, 0, "uvping"))
		pr_warn("UV: PING NMI handler failed to register\n");
}

void uv_nmi_init(void)
{
	unsigned int value;

	/*
	 * Unmask NMI on all cpus
	 */
	value = apic_read(APIC_LVT1) | APIC_DM_NMI;
	value &= ~APIC_LVT_MASKED;
	apic_write(APIC_LVT1, value);
}

void uv_nmi_setup(void)
{
	int size = sizeof(void *) * (1 << NODES_SHIFT);
	int cpu, nid;

	/* Setup hub nmi info */
	uv_nmi_setup_mmrs();
	uv_hub_nmi_list = kzalloc(size, GFP_KERNEL);
	pr_info("UV: NMI hub list @ 0x%p (%d)\n", uv_hub_nmi_list, size);
	BUG_ON(!uv_hub_nmi_list);
	size = sizeof(struct uv_hub_nmi_s);
	for_each_present_cpu(cpu) {
		nid = cpu_to_node(cpu);
		if (uv_hub_nmi_list[nid] == NULL) {
			uv_hub_nmi_list[nid] = kzalloc_node(size,
							    GFP_KERNEL, nid);
			BUG_ON(!uv_hub_nmi_list[nid]);
			raw_spin_lock_init(&(uv_hub_nmi_list[nid]->nmi_lock));
			atomic_set(&uv_hub_nmi_list[nid]->cpu_owner, -1);
		}
		uv_hub_nmi_per(cpu) = uv_hub_nmi_list[nid];
	}
	BUG_ON(!alloc_cpumask_var(&uv_nmi_cpu_mask, GFP_KERNEL));
	uv_register_nmi_notifier();
}