xref: /openbmc/linux/arch/x86/kernel/cpu/mce/inject.c (revision 236a9bf2)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Machine check injection support.
4  * Copyright 2008 Intel Corporation.
5  *
6  * Authors:
7  * Andi Kleen
8  * Ying Huang
9  *
10  * The AMD part (from mce_amd_inj.c): a simple MCE injection facility
11  * for testing different aspects of the RAS code. This driver should be
12  * built as module so that it can be loaded on production kernels for
13  * testing purposes.
14  *
15  * Copyright (c) 2010-17:  Borislav Petkov <bp@alien8.de>
16  *			   Advanced Micro Devices Inc.
17  */
18 
19 #include <linux/cpu.h>
20 #include <linux/debugfs.h>
21 #include <linux/kernel.h>
22 #include <linux/module.h>
23 #include <linux/notifier.h>
24 #include <linux/pci.h>
25 #include <linux/uaccess.h>
26 
27 #include <asm/amd_nb.h>
28 #include <asm/apic.h>
29 #include <asm/irq_vectors.h>
30 #include <asm/mce.h>
31 #include <asm/nmi.h>
32 #include <asm/smp.h>
33 
34 #include "internal.h"
35 
36 static bool hw_injection_possible = true;
37 
38 /*
39  * Collect all the MCi_XXX settings
40  */
41 static struct mce i_mce;
42 static struct dentry *dfs_inj;
43 
44 #define MAX_FLAG_OPT_SIZE	4
45 #define NBCFG			0x44
46 
47 enum injection_type {
48 	SW_INJ = 0,	/* SW injection, simply decode the error */
49 	HW_INJ,		/* Trigger a #MC */
50 	DFR_INT_INJ,    /* Trigger Deferred error interrupt */
51 	THR_INT_INJ,    /* Trigger threshold interrupt */
52 	N_INJ_TYPES,
53 };
54 
55 static const char * const flags_options[] = {
56 	[SW_INJ] = "sw",
57 	[HW_INJ] = "hw",
58 	[DFR_INT_INJ] = "df",
59 	[THR_INT_INJ] = "th",
60 	NULL
61 };
62 
63 /* Set default injection to SW_INJ */
64 static enum injection_type inj_type = SW_INJ;
65 
66 #define MCE_INJECT_SET(reg)						\
67 static int inj_##reg##_set(void *data, u64 val)				\
68 {									\
69 	struct mce *m = (struct mce *)data;				\
70 									\
71 	m->reg = val;							\
72 	return 0;							\
73 }
74 
75 MCE_INJECT_SET(status);
76 MCE_INJECT_SET(misc);
77 MCE_INJECT_SET(addr);
78 MCE_INJECT_SET(synd);
79 
80 #define MCE_INJECT_GET(reg)						\
81 static int inj_##reg##_get(void *data, u64 *val)			\
82 {									\
83 	struct mce *m = (struct mce *)data;				\
84 									\
85 	*val = m->reg;							\
86 	return 0;							\
87 }
88 
89 MCE_INJECT_GET(status);
90 MCE_INJECT_GET(misc);
91 MCE_INJECT_GET(addr);
92 MCE_INJECT_GET(synd);
93 MCE_INJECT_GET(ipid);
94 
95 DEFINE_SIMPLE_ATTRIBUTE(status_fops, inj_status_get, inj_status_set, "%llx\n");
96 DEFINE_SIMPLE_ATTRIBUTE(misc_fops, inj_misc_get, inj_misc_set, "%llx\n");
97 DEFINE_SIMPLE_ATTRIBUTE(addr_fops, inj_addr_get, inj_addr_set, "%llx\n");
98 DEFINE_SIMPLE_ATTRIBUTE(synd_fops, inj_synd_get, inj_synd_set, "%llx\n");
99 
100 /* Use the user provided IPID value on a sw injection. */
101 static int inj_ipid_set(void *data, u64 val)
102 {
103 	struct mce *m = (struct mce *)data;
104 
105 	if (cpu_feature_enabled(X86_FEATURE_SMCA)) {
106 		if (inj_type == SW_INJ)
107 			m->ipid = val;
108 	}
109 
110 	return 0;
111 }
112 
113 DEFINE_SIMPLE_ATTRIBUTE(ipid_fops, inj_ipid_get, inj_ipid_set, "%llx\n");
114 
115 static void setup_inj_struct(struct mce *m)
116 {
117 	memset(m, 0, sizeof(struct mce));
118 
119 	m->cpuvendor = boot_cpu_data.x86_vendor;
120 	m->time	     = ktime_get_real_seconds();
121 	m->cpuid     = cpuid_eax(1);
122 	m->microcode = boot_cpu_data.microcode;
123 }
124 
125 /* Update fake mce registers on current CPU. */
126 static void inject_mce(struct mce *m)
127 {
128 	struct mce *i = &per_cpu(injectm, m->extcpu);
129 
130 	/* Make sure no one reads partially written injectm */
131 	i->finished = 0;
132 	mb();
133 	m->finished = 0;
134 	/* First set the fields after finished */
135 	i->extcpu = m->extcpu;
136 	mb();
137 	/* Now write record in order, finished last (except above) */
138 	memcpy(i, m, sizeof(struct mce));
139 	/* Finally activate it */
140 	mb();
141 	i->finished = 1;
142 }
143 
144 static void raise_poll(struct mce *m)
145 {
146 	unsigned long flags;
147 	mce_banks_t b;
148 
149 	memset(&b, 0xff, sizeof(mce_banks_t));
150 	local_irq_save(flags);
151 	machine_check_poll(0, &b);
152 	local_irq_restore(flags);
153 	m->finished = 0;
154 }
155 
156 static void raise_exception(struct mce *m, struct pt_regs *pregs)
157 {
158 	struct pt_regs regs;
159 	unsigned long flags;
160 
161 	if (!pregs) {
162 		memset(&regs, 0, sizeof(struct pt_regs));
163 		regs.ip = m->ip;
164 		regs.cs = m->cs;
165 		pregs = &regs;
166 	}
167 	/* do_machine_check() expects interrupts disabled -- at least */
168 	local_irq_save(flags);
169 	do_machine_check(pregs);
170 	local_irq_restore(flags);
171 	m->finished = 0;
172 }
173 
174 static cpumask_var_t mce_inject_cpumask;
175 static DEFINE_MUTEX(mce_inject_mutex);
176 
177 static int mce_raise_notify(unsigned int cmd, struct pt_regs *regs)
178 {
179 	int cpu = smp_processor_id();
180 	struct mce *m = this_cpu_ptr(&injectm);
181 	if (!cpumask_test_cpu(cpu, mce_inject_cpumask))
182 		return NMI_DONE;
183 	cpumask_clear_cpu(cpu, mce_inject_cpumask);
184 	if (m->inject_flags & MCJ_EXCEPTION)
185 		raise_exception(m, regs);
186 	else if (m->status)
187 		raise_poll(m);
188 	return NMI_HANDLED;
189 }
190 
191 static void mce_irq_ipi(void *info)
192 {
193 	int cpu = smp_processor_id();
194 	struct mce *m = this_cpu_ptr(&injectm);
195 
196 	if (cpumask_test_cpu(cpu, mce_inject_cpumask) &&
197 			m->inject_flags & MCJ_EXCEPTION) {
198 		cpumask_clear_cpu(cpu, mce_inject_cpumask);
199 		raise_exception(m, NULL);
200 	}
201 }
202 
203 /* Inject mce on current CPU */
204 static int raise_local(void)
205 {
206 	struct mce *m = this_cpu_ptr(&injectm);
207 	int context = MCJ_CTX(m->inject_flags);
208 	int ret = 0;
209 	int cpu = m->extcpu;
210 
211 	if (m->inject_flags & MCJ_EXCEPTION) {
212 		pr_info("Triggering MCE exception on CPU %d\n", cpu);
213 		switch (context) {
214 		case MCJ_CTX_IRQ:
215 			/*
216 			 * Could do more to fake interrupts like
217 			 * calling irq_enter, but the necessary
218 			 * machinery isn't exported currently.
219 			 */
220 			fallthrough;
221 		case MCJ_CTX_PROCESS:
222 			raise_exception(m, NULL);
223 			break;
224 		default:
225 			pr_info("Invalid MCE context\n");
226 			ret = -EINVAL;
227 		}
228 		pr_info("MCE exception done on CPU %d\n", cpu);
229 	} else if (m->status) {
230 		pr_info("Starting machine check poll CPU %d\n", cpu);
231 		raise_poll(m);
232 		mce_notify_irq();
233 		pr_info("Machine check poll done on CPU %d\n", cpu);
234 	} else
235 		m->finished = 0;
236 
237 	return ret;
238 }
239 
240 static void __maybe_unused raise_mce(struct mce *m)
241 {
242 	int context = MCJ_CTX(m->inject_flags);
243 
244 	inject_mce(m);
245 
246 	if (context == MCJ_CTX_RANDOM)
247 		return;
248 
249 	if (m->inject_flags & (MCJ_IRQ_BROADCAST | MCJ_NMI_BROADCAST)) {
250 		unsigned long start;
251 		int cpu;
252 
253 		cpus_read_lock();
254 		cpumask_copy(mce_inject_cpumask, cpu_online_mask);
255 		cpumask_clear_cpu(get_cpu(), mce_inject_cpumask);
256 		for_each_online_cpu(cpu) {
257 			struct mce *mcpu = &per_cpu(injectm, cpu);
258 			if (!mcpu->finished ||
259 			    MCJ_CTX(mcpu->inject_flags) != MCJ_CTX_RANDOM)
260 				cpumask_clear_cpu(cpu, mce_inject_cpumask);
261 		}
262 		if (!cpumask_empty(mce_inject_cpumask)) {
263 			if (m->inject_flags & MCJ_IRQ_BROADCAST) {
264 				/*
265 				 * don't wait because mce_irq_ipi is necessary
266 				 * to be sync with following raise_local
267 				 */
268 				preempt_disable();
269 				smp_call_function_many(mce_inject_cpumask,
270 					mce_irq_ipi, NULL, 0);
271 				preempt_enable();
272 			} else if (m->inject_flags & MCJ_NMI_BROADCAST)
273 				__apic_send_IPI_mask(mce_inject_cpumask, NMI_VECTOR);
274 		}
275 		start = jiffies;
276 		while (!cpumask_empty(mce_inject_cpumask)) {
277 			if (!time_before(jiffies, start + 2*HZ)) {
278 				pr_err("Timeout waiting for mce inject %lx\n",
279 					*cpumask_bits(mce_inject_cpumask));
280 				break;
281 			}
282 			cpu_relax();
283 		}
284 		raise_local();
285 		put_cpu();
286 		cpus_read_unlock();
287 	} else {
288 		preempt_disable();
289 		raise_local();
290 		preempt_enable();
291 	}
292 }
293 
294 static int mce_inject_raise(struct notifier_block *nb, unsigned long val,
295 			    void *data)
296 {
297 	struct mce *m = (struct mce *)data;
298 
299 	if (!m)
300 		return NOTIFY_DONE;
301 
302 	mutex_lock(&mce_inject_mutex);
303 	raise_mce(m);
304 	mutex_unlock(&mce_inject_mutex);
305 
306 	return NOTIFY_DONE;
307 }
308 
309 static struct notifier_block inject_nb = {
310 	.notifier_call  = mce_inject_raise,
311 };
312 
313 /*
314  * Caller needs to be make sure this cpu doesn't disappear
315  * from under us, i.e.: get_cpu/put_cpu.
316  */
317 static int toggle_hw_mce_inject(unsigned int cpu, bool enable)
318 {
319 	u32 l, h;
320 	int err;
321 
322 	err = rdmsr_on_cpu(cpu, MSR_K7_HWCR, &l, &h);
323 	if (err) {
324 		pr_err("%s: error reading HWCR\n", __func__);
325 		return err;
326 	}
327 
328 	enable ? (l |= BIT(18)) : (l &= ~BIT(18));
329 
330 	err = wrmsr_on_cpu(cpu, MSR_K7_HWCR, l, h);
331 	if (err)
332 		pr_err("%s: error writing HWCR\n", __func__);
333 
334 	return err;
335 }
336 
337 static int __set_inj(const char *buf)
338 {
339 	int i;
340 
341 	for (i = 0; i < N_INJ_TYPES; i++) {
342 		if (!strncmp(flags_options[i], buf, strlen(flags_options[i]))) {
343 			if (i > SW_INJ && !hw_injection_possible)
344 				continue;
345 			inj_type = i;
346 			return 0;
347 		}
348 	}
349 	return -EINVAL;
350 }
351 
352 static ssize_t flags_read(struct file *filp, char __user *ubuf,
353 			  size_t cnt, loff_t *ppos)
354 {
355 	char buf[MAX_FLAG_OPT_SIZE];
356 	int n;
357 
358 	n = sprintf(buf, "%s\n", flags_options[inj_type]);
359 
360 	return simple_read_from_buffer(ubuf, cnt, ppos, buf, n);
361 }
362 
363 static ssize_t flags_write(struct file *filp, const char __user *ubuf,
364 			   size_t cnt, loff_t *ppos)
365 {
366 	char buf[MAX_FLAG_OPT_SIZE], *__buf;
367 	int err;
368 
369 	if (!cnt || cnt > MAX_FLAG_OPT_SIZE)
370 		return -EINVAL;
371 
372 	if (copy_from_user(&buf, ubuf, cnt))
373 		return -EFAULT;
374 
375 	buf[cnt - 1] = 0;
376 
377 	/* strip whitespace */
378 	__buf = strstrip(buf);
379 
380 	err = __set_inj(__buf);
381 	if (err) {
382 		pr_err("%s: Invalid flags value: %s\n", __func__, __buf);
383 		return err;
384 	}
385 
386 	*ppos += cnt;
387 
388 	return cnt;
389 }
390 
391 static const struct file_operations flags_fops = {
392 	.read           = flags_read,
393 	.write          = flags_write,
394 	.llseek         = generic_file_llseek,
395 };
396 
397 /*
398  * On which CPU to inject?
399  */
400 MCE_INJECT_GET(extcpu);
401 
402 static int inj_extcpu_set(void *data, u64 val)
403 {
404 	struct mce *m = (struct mce *)data;
405 
406 	if (val >= nr_cpu_ids || !cpu_online(val)) {
407 		pr_err("%s: Invalid CPU: %llu\n", __func__, val);
408 		return -EINVAL;
409 	}
410 	m->extcpu = val;
411 	return 0;
412 }
413 
414 DEFINE_SIMPLE_ATTRIBUTE(extcpu_fops, inj_extcpu_get, inj_extcpu_set, "%llu\n");
415 
416 static void trigger_mce(void *info)
417 {
418 	asm volatile("int $18");
419 }
420 
421 static void trigger_dfr_int(void *info)
422 {
423 	asm volatile("int %0" :: "i" (DEFERRED_ERROR_VECTOR));
424 }
425 
426 static void trigger_thr_int(void *info)
427 {
428 	asm volatile("int %0" :: "i" (THRESHOLD_APIC_VECTOR));
429 }
430 
431 static u32 get_nbc_for_node(int node_id)
432 {
433 	struct cpuinfo_x86 *c = &boot_cpu_data;
434 	u32 cores_per_node;
435 
436 	cores_per_node = (c->x86_max_cores * smp_num_siblings) / amd_get_nodes_per_socket();
437 
438 	return cores_per_node * node_id;
439 }
440 
441 static void toggle_nb_mca_mst_cpu(u16 nid)
442 {
443 	struct amd_northbridge *nb;
444 	struct pci_dev *F3;
445 	u32 val;
446 	int err;
447 
448 	nb = node_to_amd_nb(nid);
449 	if (!nb)
450 		return;
451 
452 	F3 = nb->misc;
453 	if (!F3)
454 		return;
455 
456 	err = pci_read_config_dword(F3, NBCFG, &val);
457 	if (err) {
458 		pr_err("%s: Error reading F%dx%03x.\n",
459 		       __func__, PCI_FUNC(F3->devfn), NBCFG);
460 		return;
461 	}
462 
463 	if (val & BIT(27))
464 		return;
465 
466 	pr_err("%s: Set D18F3x44[NbMcaToMstCpuEn] which BIOS hasn't done.\n",
467 	       __func__);
468 
469 	val |= BIT(27);
470 	err = pci_write_config_dword(F3, NBCFG, val);
471 	if (err)
472 		pr_err("%s: Error writing F%dx%03x.\n",
473 		       __func__, PCI_FUNC(F3->devfn), NBCFG);
474 }
475 
476 static void prepare_msrs(void *info)
477 {
478 	struct mce m = *(struct mce *)info;
479 	u8 b = m.bank;
480 
481 	wrmsrl(MSR_IA32_MCG_STATUS, m.mcgstatus);
482 
483 	if (boot_cpu_has(X86_FEATURE_SMCA)) {
484 		if (m.inject_flags == DFR_INT_INJ) {
485 			wrmsrl(MSR_AMD64_SMCA_MCx_DESTAT(b), m.status);
486 			wrmsrl(MSR_AMD64_SMCA_MCx_DEADDR(b), m.addr);
487 		} else {
488 			wrmsrl(MSR_AMD64_SMCA_MCx_STATUS(b), m.status);
489 			wrmsrl(MSR_AMD64_SMCA_MCx_ADDR(b), m.addr);
490 		}
491 
492 		wrmsrl(MSR_AMD64_SMCA_MCx_MISC(b), m.misc);
493 		wrmsrl(MSR_AMD64_SMCA_MCx_SYND(b), m.synd);
494 	} else {
495 		wrmsrl(MSR_IA32_MCx_STATUS(b), m.status);
496 		wrmsrl(MSR_IA32_MCx_ADDR(b), m.addr);
497 		wrmsrl(MSR_IA32_MCx_MISC(b), m.misc);
498 	}
499 }
500 
501 static void do_inject(void)
502 {
503 	u64 mcg_status = 0;
504 	unsigned int cpu = i_mce.extcpu;
505 	u8 b = i_mce.bank;
506 
507 	i_mce.tsc = rdtsc_ordered();
508 
509 	i_mce.status |= MCI_STATUS_VAL;
510 
511 	if (i_mce.misc)
512 		i_mce.status |= MCI_STATUS_MISCV;
513 
514 	if (i_mce.synd)
515 		i_mce.status |= MCI_STATUS_SYNDV;
516 
517 	if (inj_type == SW_INJ) {
518 		mce_log(&i_mce);
519 		return;
520 	}
521 
522 	/* prep MCE global settings for the injection */
523 	mcg_status = MCG_STATUS_MCIP | MCG_STATUS_EIPV;
524 
525 	if (!(i_mce.status & MCI_STATUS_PCC))
526 		mcg_status |= MCG_STATUS_RIPV;
527 
528 	/*
529 	 * Ensure necessary status bits for deferred errors:
530 	 * - MCx_STATUS[Deferred]: make sure it is a deferred error
531 	 * - MCx_STATUS[UC] cleared: deferred errors are _not_ UC
532 	 */
533 	if (inj_type == DFR_INT_INJ) {
534 		i_mce.status |= MCI_STATUS_DEFERRED;
535 		i_mce.status &= ~MCI_STATUS_UC;
536 	}
537 
538 	/*
539 	 * For multi node CPUs, logging and reporting of bank 4 errors happens
540 	 * only on the node base core. Refer to D18F3x44[NbMcaToMstCpuEn] for
541 	 * Fam10h and later BKDGs.
542 	 */
543 	if (boot_cpu_has(X86_FEATURE_AMD_DCM) &&
544 	    b == 4 &&
545 	    boot_cpu_data.x86 < 0x17) {
546 		toggle_nb_mca_mst_cpu(topology_die_id(cpu));
547 		cpu = get_nbc_for_node(topology_die_id(cpu));
548 	}
549 
550 	cpus_read_lock();
551 	if (!cpu_online(cpu))
552 		goto err;
553 
554 	toggle_hw_mce_inject(cpu, true);
555 
556 	i_mce.mcgstatus = mcg_status;
557 	i_mce.inject_flags = inj_type;
558 	smp_call_function_single(cpu, prepare_msrs, &i_mce, 0);
559 
560 	toggle_hw_mce_inject(cpu, false);
561 
562 	switch (inj_type) {
563 	case DFR_INT_INJ:
564 		smp_call_function_single(cpu, trigger_dfr_int, NULL, 0);
565 		break;
566 	case THR_INT_INJ:
567 		smp_call_function_single(cpu, trigger_thr_int, NULL, 0);
568 		break;
569 	default:
570 		smp_call_function_single(cpu, trigger_mce, NULL, 0);
571 	}
572 
573 err:
574 	cpus_read_unlock();
575 
576 }
577 
578 /*
579  * This denotes into which bank we're injecting and triggers
580  * the injection, at the same time.
581  */
582 static int inj_bank_set(void *data, u64 val)
583 {
584 	struct mce *m = (struct mce *)data;
585 	u8 n_banks;
586 	u64 cap;
587 
588 	/* Get bank count on target CPU so we can handle non-uniform values. */
589 	rdmsrl_on_cpu(m->extcpu, MSR_IA32_MCG_CAP, &cap);
590 	n_banks = cap & MCG_BANKCNT_MASK;
591 
592 	if (val >= n_banks) {
593 		pr_err("MCA bank %llu non-existent on CPU%d\n", val, m->extcpu);
594 		return -EINVAL;
595 	}
596 
597 	m->bank = val;
598 
599 	/*
600 	 * sw-only injection allows to write arbitrary values into the MCA
601 	 * registers because it tests only the decoding paths.
602 	 */
603 	if (inj_type == SW_INJ)
604 		goto inject;
605 
606 	/*
607 	 * Read IPID value to determine if a bank is populated on the target
608 	 * CPU.
609 	 */
610 	if (cpu_feature_enabled(X86_FEATURE_SMCA)) {
611 		u64 ipid;
612 
613 		if (rdmsrl_on_cpu(m->extcpu, MSR_AMD64_SMCA_MCx_IPID(val), &ipid)) {
614 			pr_err("Error reading IPID on CPU%d\n", m->extcpu);
615 			return -EINVAL;
616 		}
617 
618 		if (!ipid) {
619 			pr_err("Cannot inject into unpopulated bank %llu\n", val);
620 			return -ENODEV;
621 		}
622 	}
623 
624 inject:
625 	do_inject();
626 
627 	/* Reset injection struct */
628 	setup_inj_struct(&i_mce);
629 
630 	return 0;
631 }
632 
633 MCE_INJECT_GET(bank);
634 
635 DEFINE_SIMPLE_ATTRIBUTE(bank_fops, inj_bank_get, inj_bank_set, "%llu\n");
636 
637 static const char readme_msg[] =
638 "Description of the files and their usages:\n"
639 "\n"
640 "Note1: i refers to the bank number below.\n"
641 "Note2: See respective BKDGs for the exact bit definitions of the files below\n"
642 "as they mirror the hardware registers.\n"
643 "\n"
644 "status:\t Set MCi_STATUS: the bits in that MSR control the error type and\n"
645 "\t attributes of the error which caused the MCE.\n"
646 "\n"
647 "misc:\t Set MCi_MISC: provide auxiliary info about the error. It is mostly\n"
648 "\t used for error thresholding purposes and its validity is indicated by\n"
649 "\t MCi_STATUS[MiscV].\n"
650 "\n"
651 "synd:\t Set MCi_SYND: provide syndrome info about the error. Only valid on\n"
652 "\t Scalable MCA systems, and its validity is indicated by MCi_STATUS[SyndV].\n"
653 "\n"
654 "addr:\t Error address value to be written to MCi_ADDR. Log address information\n"
655 "\t associated with the error.\n"
656 "\n"
657 "cpu:\t The CPU to inject the error on.\n"
658 "\n"
659 "bank:\t Specify the bank you want to inject the error into: the number of\n"
660 "\t banks in a processor varies and is family/model-specific, therefore, the\n"
661 "\t supplied value is sanity-checked. Setting the bank value also triggers the\n"
662 "\t injection.\n"
663 "\n"
664 "flags:\t Injection type to be performed. Writing to this file will trigger a\n"
665 "\t real machine check, an APIC interrupt or invoke the error decoder routines\n"
666 "\t for AMD processors.\n"
667 "\n"
668 "\t Allowed error injection types:\n"
669 "\t  - \"sw\": Software error injection. Decode error to a human-readable \n"
670 "\t    format only. Safe to use.\n"
671 "\t  - \"hw\": Hardware error injection. Causes the #MC exception handler to \n"
672 "\t    handle the error. Be warned: might cause system panic if MCi_STATUS[PCC] \n"
673 "\t    is set. Therefore, consider setting (debugfs_mountpoint)/mce/fake_panic \n"
674 "\t    before injecting.\n"
675 "\t  - \"df\": Trigger APIC interrupt for Deferred error. Causes deferred \n"
676 "\t    error APIC interrupt handler to handle the error if the feature is \n"
677 "\t    is present in hardware. \n"
678 "\t  - \"th\": Trigger APIC interrupt for Threshold errors. Causes threshold \n"
679 "\t    APIC interrupt handler to handle the error. \n"
680 "\n"
681 "ipid:\t IPID (AMD-specific)\n"
682 "\n";
683 
684 static ssize_t
685 inj_readme_read(struct file *filp, char __user *ubuf,
686 		       size_t cnt, loff_t *ppos)
687 {
688 	return simple_read_from_buffer(ubuf, cnt, ppos,
689 					readme_msg, strlen(readme_msg));
690 }
691 
692 static const struct file_operations readme_fops = {
693 	.read		= inj_readme_read,
694 };
695 
696 static struct dfs_node {
697 	char *name;
698 	const struct file_operations *fops;
699 	umode_t perm;
700 } dfs_fls[] = {
701 	{ .name = "status",	.fops = &status_fops, .perm = S_IRUSR | S_IWUSR },
702 	{ .name = "misc",	.fops = &misc_fops,   .perm = S_IRUSR | S_IWUSR },
703 	{ .name = "addr",	.fops = &addr_fops,   .perm = S_IRUSR | S_IWUSR },
704 	{ .name = "synd",	.fops = &synd_fops,   .perm = S_IRUSR | S_IWUSR },
705 	{ .name = "ipid",	.fops = &ipid_fops,   .perm = S_IRUSR | S_IWUSR },
706 	{ .name = "bank",	.fops = &bank_fops,   .perm = S_IRUSR | S_IWUSR },
707 	{ .name = "flags",	.fops = &flags_fops,  .perm = S_IRUSR | S_IWUSR },
708 	{ .name = "cpu",	.fops = &extcpu_fops, .perm = S_IRUSR | S_IWUSR },
709 	{ .name = "README",	.fops = &readme_fops, .perm = S_IRUSR | S_IRGRP | S_IROTH },
710 };
711 
712 static void __init debugfs_init(void)
713 {
714 	unsigned int i;
715 
716 	dfs_inj = debugfs_create_dir("mce-inject", NULL);
717 
718 	for (i = 0; i < ARRAY_SIZE(dfs_fls); i++)
719 		debugfs_create_file(dfs_fls[i].name, dfs_fls[i].perm, dfs_inj,
720 				    &i_mce, dfs_fls[i].fops);
721 }
722 
723 static void check_hw_inj_possible(void)
724 {
725 	int cpu;
726 	u8 bank;
727 
728 	/*
729 	 * This behavior exists only on SMCA systems though its not directly
730 	 * related to SMCA.
731 	 */
732 	if (!cpu_feature_enabled(X86_FEATURE_SMCA))
733 		return;
734 
735 	cpu = get_cpu();
736 
737 	for (bank = 0; bank < MAX_NR_BANKS; ++bank) {
738 		u64 status = MCI_STATUS_VAL, ipid;
739 
740 		/* Check whether bank is populated */
741 		rdmsrl(MSR_AMD64_SMCA_MCx_IPID(bank), ipid);
742 		if (!ipid)
743 			continue;
744 
745 		toggle_hw_mce_inject(cpu, true);
746 
747 		wrmsrl_safe(mca_msr_reg(bank, MCA_STATUS), status);
748 		rdmsrl_safe(mca_msr_reg(bank, MCA_STATUS), &status);
749 		wrmsrl_safe(mca_msr_reg(bank, MCA_STATUS), 0);
750 
751 		if (!status) {
752 			hw_injection_possible = false;
753 			pr_warn("Platform does not allow *hardware* error injection."
754 				"Try using APEI EINJ instead.\n");
755 		}
756 
757 		toggle_hw_mce_inject(cpu, false);
758 
759 		break;
760 	}
761 
762 	put_cpu();
763 }
764 
765 static int __init inject_init(void)
766 {
767 	if (!alloc_cpumask_var(&mce_inject_cpumask, GFP_KERNEL))
768 		return -ENOMEM;
769 
770 	check_hw_inj_possible();
771 
772 	debugfs_init();
773 
774 	register_nmi_handler(NMI_LOCAL, mce_raise_notify, 0, "mce_notify");
775 	mce_register_injector_chain(&inject_nb);
776 
777 	setup_inj_struct(&i_mce);
778 
779 	pr_info("Machine check injector initialized\n");
780 
781 	return 0;
782 }
783 
784 static void __exit inject_exit(void)
785 {
786 
787 	mce_unregister_injector_chain(&inject_nb);
788 	unregister_nmi_handler(NMI_LOCAL, "mce_notify");
789 
790 	debugfs_remove_recursive(dfs_inj);
791 	dfs_inj = NULL;
792 
793 	memset(&dfs_fls, 0, sizeof(dfs_fls));
794 
795 	free_cpumask_var(mce_inject_cpumask);
796 }
797 
798 module_init(inject_init);
799 module_exit(inject_exit);
800 MODULE_LICENSE("GPL");
801