Linux Audio

Check our new training course

Loading...
Note: File does not exist in v4.17.
   1// SPDX-License-Identifier: GPL-2.0-only
   2/*
   3 * Machine check handler.
   4 *
   5 * K8 parts Copyright 2002,2003 Andi Kleen, SuSE Labs.
   6 * Rest from unknown author(s).
   7 * 2004 Andi Kleen. Rewrote most of it.
   8 * Copyright 2008 Intel Corporation
   9 * Author: Andi Kleen
  10 */
  11
  12#include <linux/thread_info.h>
  13#include <linux/capability.h>
  14#include <linux/miscdevice.h>
  15#include <linux/ratelimit.h>
  16#include <linux/rcupdate.h>
  17#include <linux/kobject.h>
  18#include <linux/uaccess.h>
  19#include <linux/kdebug.h>
  20#include <linux/kernel.h>
  21#include <linux/percpu.h>
  22#include <linux/string.h>
  23#include <linux/device.h>
  24#include <linux/syscore_ops.h>
  25#include <linux/delay.h>
  26#include <linux/ctype.h>
  27#include <linux/sched.h>
  28#include <linux/sysfs.h>
  29#include <linux/types.h>
  30#include <linux/slab.h>
  31#include <linux/init.h>
  32#include <linux/kmod.h>
  33#include <linux/poll.h>
  34#include <linux/nmi.h>
  35#include <linux/cpu.h>
  36#include <linux/ras.h>
  37#include <linux/smp.h>
  38#include <linux/fs.h>
  39#include <linux/mm.h>
  40#include <linux/debugfs.h>
  41#include <linux/irq_work.h>
  42#include <linux/export.h>
  43#include <linux/jump_label.h>
  44#include <linux/set_memory.h>
  45
  46#include <asm/intel-family.h>
  47#include <asm/processor.h>
  48#include <asm/traps.h>
  49#include <asm/tlbflush.h>
  50#include <asm/mce.h>
  51#include <asm/msr.h>
  52#include <asm/reboot.h>
  53
  54#include "internal.h"
  55
  56static DEFINE_MUTEX(mce_log_mutex);
  57
  58/* sysfs synchronization */
  59static DEFINE_MUTEX(mce_sysfs_mutex);
  60
  61#define CREATE_TRACE_POINTS
  62#include <trace/events/mce.h>
  63
  64#define SPINUNIT		100	/* 100ns */
  65
  66DEFINE_PER_CPU(unsigned, mce_exception_count);
  67
  68DEFINE_PER_CPU_READ_MOSTLY(unsigned int, mce_num_banks);
  69
  70struct mce_bank {
  71	u64			ctl;			/* subevents to enable */
  72	bool			init;			/* initialise bank? */
  73};
  74static DEFINE_PER_CPU_READ_MOSTLY(struct mce_bank[MAX_NR_BANKS], mce_banks_array);
  75
  76#define ATTR_LEN               16
  77/* One object for each MCE bank, shared by all CPUs */
  78struct mce_bank_dev {
  79	struct device_attribute	attr;			/* device attribute */
  80	char			attrname[ATTR_LEN];	/* attribute name */
  81	u8			bank;			/* bank number */
  82};
  83static struct mce_bank_dev mce_bank_devs[MAX_NR_BANKS];
  84
  85struct mce_vendor_flags mce_flags __read_mostly;
  86
  87struct mca_config mca_cfg __read_mostly = {
  88	.bootlog  = -1,
  89	/*
  90	 * Tolerant levels:
  91	 * 0: always panic on uncorrected errors, log corrected errors
  92	 * 1: panic or SIGBUS on uncorrected errors, log corrected errors
  93	 * 2: SIGBUS or log uncorrected errors (if possible), log corr. errors
  94	 * 3: never panic or SIGBUS, log all errors (for testing only)
  95	 */
  96	.tolerant = 1,
  97	.monarch_timeout = -1
  98};
  99
 100static DEFINE_PER_CPU(struct mce, mces_seen);
 101static unsigned long mce_need_notify;
 102static int cpu_missing;
 103
 104/*
 105 * MCA banks polled by the period polling timer for corrected events.
 106 * With Intel CMCI, this only has MCA banks which do not support CMCI (if any).
 107 */
 108DEFINE_PER_CPU(mce_banks_t, mce_poll_banks) = {
 109	[0 ... BITS_TO_LONGS(MAX_NR_BANKS)-1] = ~0UL
 110};
 111
 112/*
 113 * MCA banks controlled through firmware first for corrected errors.
 114 * This is a global list of banks for which we won't enable CMCI and we
 115 * won't poll. Firmware controls these banks and is responsible for
 116 * reporting corrected errors through GHES. Uncorrected/recoverable
 117 * errors are still notified through a machine check.
 118 */
 119mce_banks_t mce_banks_ce_disabled;
 120
 121static struct work_struct mce_work;
 122static struct irq_work mce_irq_work;
 123
 124static void (*quirk_no_way_out)(int bank, struct mce *m, struct pt_regs *regs);
 125
 126/*
 127 * CPU/chipset specific EDAC code can register a notifier call here to print
 128 * MCE errors in a human-readable form.
 129 */
 130BLOCKING_NOTIFIER_HEAD(x86_mce_decoder_chain);
 131
 132/* Do initial initialization of a struct mce */
 133void mce_setup(struct mce *m)
 134{
 135	memset(m, 0, sizeof(struct mce));
 136	m->cpu = m->extcpu = smp_processor_id();
 137	/* need the internal __ version to avoid deadlocks */
 138	m->time = __ktime_get_real_seconds();
 139	m->cpuvendor = boot_cpu_data.x86_vendor;
 140	m->cpuid = cpuid_eax(1);
 141	m->socketid = cpu_data(m->extcpu).phys_proc_id;
 142	m->apicid = cpu_data(m->extcpu).initial_apicid;
 143	rdmsrl(MSR_IA32_MCG_CAP, m->mcgcap);
 144
 145	if (this_cpu_has(X86_FEATURE_INTEL_PPIN))
 146		rdmsrl(MSR_PPIN, m->ppin);
 147
 148	m->microcode = boot_cpu_data.microcode;
 149}
 150
 151DEFINE_PER_CPU(struct mce, injectm);
 152EXPORT_PER_CPU_SYMBOL_GPL(injectm);
 153
 154void mce_log(struct mce *m)
 155{
 156	if (!mce_gen_pool_add(m))
 157		irq_work_queue(&mce_irq_work);
 158}
 159
 160void mce_inject_log(struct mce *m)
 161{
 162	mutex_lock(&mce_log_mutex);
 163	mce_log(m);
 164	mutex_unlock(&mce_log_mutex);
 165}
 166EXPORT_SYMBOL_GPL(mce_inject_log);
 167
 168static struct notifier_block mce_srao_nb;
 169
 170/*
 171 * We run the default notifier if we have only the SRAO, the first and the
 172 * default notifier registered. I.e., the mandatory NUM_DEFAULT_NOTIFIERS
 173 * notifiers registered on the chain.
 174 */
 175#define NUM_DEFAULT_NOTIFIERS	3
 176static atomic_t num_notifiers;
 177
 178void mce_register_decode_chain(struct notifier_block *nb)
 179{
 180	if (WARN_ON(nb->priority > MCE_PRIO_MCELOG && nb->priority < MCE_PRIO_EDAC))
 181		return;
 182
 183	atomic_inc(&num_notifiers);
 184
 185	blocking_notifier_chain_register(&x86_mce_decoder_chain, nb);
 186}
 187EXPORT_SYMBOL_GPL(mce_register_decode_chain);
 188
 189void mce_unregister_decode_chain(struct notifier_block *nb)
 190{
 191	atomic_dec(&num_notifiers);
 192
 193	blocking_notifier_chain_unregister(&x86_mce_decoder_chain, nb);
 194}
 195EXPORT_SYMBOL_GPL(mce_unregister_decode_chain);
 196
 197static inline u32 ctl_reg(int bank)
 198{
 199	return MSR_IA32_MCx_CTL(bank);
 200}
 201
 202static inline u32 status_reg(int bank)
 203{
 204	return MSR_IA32_MCx_STATUS(bank);
 205}
 206
 207static inline u32 addr_reg(int bank)
 208{
 209	return MSR_IA32_MCx_ADDR(bank);
 210}
 211
 212static inline u32 misc_reg(int bank)
 213{
 214	return MSR_IA32_MCx_MISC(bank);
 215}
 216
 217static inline u32 smca_ctl_reg(int bank)
 218{
 219	return MSR_AMD64_SMCA_MCx_CTL(bank);
 220}
 221
 222static inline u32 smca_status_reg(int bank)
 223{
 224	return MSR_AMD64_SMCA_MCx_STATUS(bank);
 225}
 226
 227static inline u32 smca_addr_reg(int bank)
 228{
 229	return MSR_AMD64_SMCA_MCx_ADDR(bank);
 230}
 231
 232static inline u32 smca_misc_reg(int bank)
 233{
 234	return MSR_AMD64_SMCA_MCx_MISC(bank);
 235}
 236
 237struct mca_msr_regs msr_ops = {
 238	.ctl	= ctl_reg,
 239	.status	= status_reg,
 240	.addr	= addr_reg,
 241	.misc	= misc_reg
 242};
 243
 244static void __print_mce(struct mce *m)
 245{
 246	pr_emerg(HW_ERR "CPU %d: Machine Check%s: %Lx Bank %d: %016Lx\n",
 247		 m->extcpu,
 248		 (m->mcgstatus & MCG_STATUS_MCIP ? " Exception" : ""),
 249		 m->mcgstatus, m->bank, m->status);
 250
 251	if (m->ip) {
 252		pr_emerg(HW_ERR "RIP%s %02x:<%016Lx> ",
 253			!(m->mcgstatus & MCG_STATUS_EIPV) ? " !INEXACT!" : "",
 254			m->cs, m->ip);
 255
 256		if (m->cs == __KERNEL_CS)
 257			pr_cont("{%pS}", (void *)(unsigned long)m->ip);
 258		pr_cont("\n");
 259	}
 260
 261	pr_emerg(HW_ERR "TSC %llx ", m->tsc);
 262	if (m->addr)
 263		pr_cont("ADDR %llx ", m->addr);
 264	if (m->misc)
 265		pr_cont("MISC %llx ", m->misc);
 266
 267	if (mce_flags.smca) {
 268		if (m->synd)
 269			pr_cont("SYND %llx ", m->synd);
 270		if (m->ipid)
 271			pr_cont("IPID %llx ", m->ipid);
 272	}
 273
 274	pr_cont("\n");
 275	/*
 276	 * Note this output is parsed by external tools and old fields
 277	 * should not be changed.
 278	 */
 279	pr_emerg(HW_ERR "PROCESSOR %u:%x TIME %llu SOCKET %u APIC %x microcode %x\n",
 280		m->cpuvendor, m->cpuid, m->time, m->socketid, m->apicid,
 281		m->microcode);
 282}
 283
 284static void print_mce(struct mce *m)
 285{
 286	__print_mce(m);
 287
 288	if (m->cpuvendor != X86_VENDOR_AMD && m->cpuvendor != X86_VENDOR_HYGON)
 289		pr_emerg_ratelimited(HW_ERR "Run the above through 'mcelog --ascii'\n");
 290}
 291
 292#define PANIC_TIMEOUT 5 /* 5 seconds */
 293
 294static atomic_t mce_panicked;
 295
 296static int fake_panic;
 297static atomic_t mce_fake_panicked;
 298
 299/* Panic in progress. Enable interrupts and wait for final IPI */
 300static void wait_for_panic(void)
 301{
 302	long timeout = PANIC_TIMEOUT*USEC_PER_SEC;
 303
 304	preempt_disable();
 305	local_irq_enable();
 306	while (timeout-- > 0)
 307		udelay(1);
 308	if (panic_timeout == 0)
 309		panic_timeout = mca_cfg.panic_timeout;
 310	panic("Panicing machine check CPU died");
 311}
 312
 313static void mce_panic(const char *msg, struct mce *final, char *exp)
 314{
 315	int apei_err = 0;
 316	struct llist_node *pending;
 317	struct mce_evt_llist *l;
 318
 319	if (!fake_panic) {
 320		/*
 321		 * Make sure only one CPU runs in machine check panic
 322		 */
 323		if (atomic_inc_return(&mce_panicked) > 1)
 324			wait_for_panic();
 325		barrier();
 326
 327		bust_spinlocks(1);
 328		console_verbose();
 329	} else {
 330		/* Don't log too much for fake panic */
 331		if (atomic_inc_return(&mce_fake_panicked) > 1)
 332			return;
 333	}
 334	pending = mce_gen_pool_prepare_records();
 335	/* First print corrected ones that are still unlogged */
 336	llist_for_each_entry(l, pending, llnode) {
 337		struct mce *m = &l->mce;
 338		if (!(m->status & MCI_STATUS_UC)) {
 339			print_mce(m);
 340			if (!apei_err)
 341				apei_err = apei_write_mce(m);
 342		}
 343	}
 344	/* Now print uncorrected but with the final one last */
 345	llist_for_each_entry(l, pending, llnode) {
 346		struct mce *m = &l->mce;
 347		if (!(m->status & MCI_STATUS_UC))
 348			continue;
 349		if (!final || mce_cmp(m, final)) {
 350			print_mce(m);
 351			if (!apei_err)
 352				apei_err = apei_write_mce(m);
 353		}
 354	}
 355	if (final) {
 356		print_mce(final);
 357		if (!apei_err)
 358			apei_err = apei_write_mce(final);
 359	}
 360	if (cpu_missing)
 361		pr_emerg(HW_ERR "Some CPUs didn't answer in synchronization\n");
 362	if (exp)
 363		pr_emerg(HW_ERR "Machine check: %s\n", exp);
 364	if (!fake_panic) {
 365		if (panic_timeout == 0)
 366			panic_timeout = mca_cfg.panic_timeout;
 367		panic(msg);
 368	} else
 369		pr_emerg(HW_ERR "Fake kernel panic: %s\n", msg);
 370}
 371
 372/* Support code for software error injection */
 373
 374static int msr_to_offset(u32 msr)
 375{
 376	unsigned bank = __this_cpu_read(injectm.bank);
 377
 378	if (msr == mca_cfg.rip_msr)
 379		return offsetof(struct mce, ip);
 380	if (msr == msr_ops.status(bank))
 381		return offsetof(struct mce, status);
 382	if (msr == msr_ops.addr(bank))
 383		return offsetof(struct mce, addr);
 384	if (msr == msr_ops.misc(bank))
 385		return offsetof(struct mce, misc);
 386	if (msr == MSR_IA32_MCG_STATUS)
 387		return offsetof(struct mce, mcgstatus);
 388	return -1;
 389}
 390
 391/* MSR access wrappers used for error injection */
 392static u64 mce_rdmsrl(u32 msr)
 393{
 394	u64 v;
 395
 396	if (__this_cpu_read(injectm.finished)) {
 397		int offset = msr_to_offset(msr);
 398
 399		if (offset < 0)
 400			return 0;
 401		return *(u64 *)((char *)this_cpu_ptr(&injectm) + offset);
 402	}
 403
 404	if (rdmsrl_safe(msr, &v)) {
 405		WARN_ONCE(1, "mce: Unable to read MSR 0x%x!\n", msr);
 406		/*
 407		 * Return zero in case the access faulted. This should
 408		 * not happen normally but can happen if the CPU does
 409		 * something weird, or if the code is buggy.
 410		 */
 411		v = 0;
 412	}
 413
 414	return v;
 415}
 416
 417static void mce_wrmsrl(u32 msr, u64 v)
 418{
 419	if (__this_cpu_read(injectm.finished)) {
 420		int offset = msr_to_offset(msr);
 421
 422		if (offset >= 0)
 423			*(u64 *)((char *)this_cpu_ptr(&injectm) + offset) = v;
 424		return;
 425	}
 426	wrmsrl(msr, v);
 427}
 428
 429/*
 430 * Collect all global (w.r.t. this processor) status about this machine
 431 * check into our "mce" struct so that we can use it later to assess
 432 * the severity of the problem as we read per-bank specific details.
 433 */
 434static inline void mce_gather_info(struct mce *m, struct pt_regs *regs)
 435{
 436	mce_setup(m);
 437
 438	m->mcgstatus = mce_rdmsrl(MSR_IA32_MCG_STATUS);
 439	if (regs) {
 440		/*
 441		 * Get the address of the instruction at the time of
 442		 * the machine check error.
 443		 */
 444		if (m->mcgstatus & (MCG_STATUS_RIPV|MCG_STATUS_EIPV)) {
 445			m->ip = regs->ip;
 446			m->cs = regs->cs;
 447
 448			/*
 449			 * When in VM86 mode make the cs look like ring 3
 450			 * always. This is a lie, but it's better than passing
 451			 * the additional vm86 bit around everywhere.
 452			 */
 453			if (v8086_mode(regs))
 454				m->cs |= 3;
 455		}
 456		/* Use accurate RIP reporting if available. */
 457		if (mca_cfg.rip_msr)
 458			m->ip = mce_rdmsrl(mca_cfg.rip_msr);
 459	}
 460}
 461
 462int mce_available(struct cpuinfo_x86 *c)
 463{
 464	if (mca_cfg.disabled)
 465		return 0;
 466	return cpu_has(c, X86_FEATURE_MCE) && cpu_has(c, X86_FEATURE_MCA);
 467}
 468
 469static void mce_schedule_work(void)
 470{
 471	if (!mce_gen_pool_empty())
 472		schedule_work(&mce_work);
 473}
 474
 475static void mce_irq_work_cb(struct irq_work *entry)
 476{
 477	mce_schedule_work();
 478}
 479
 480/*
 481 * Check if the address reported by the CPU is in a format we can parse.
 482 * It would be possible to add code for most other cases, but all would
 483 * be somewhat complicated (e.g. segment offset would require an instruction
 484 * parser). So only support physical addresses up to page granuality for now.
 485 */
 486int mce_usable_address(struct mce *m)
 487{
 488	if (!(m->status & MCI_STATUS_ADDRV))
 489		return 0;
 490
 491	/* Checks after this one are Intel-specific: */
 492	if (boot_cpu_data.x86_vendor != X86_VENDOR_INTEL)
 493		return 1;
 494
 495	if (!(m->status & MCI_STATUS_MISCV))
 496		return 0;
 497
 498	if (MCI_MISC_ADDR_LSB(m->misc) > PAGE_SHIFT)
 499		return 0;
 500
 501	if (MCI_MISC_ADDR_MODE(m->misc) != MCI_MISC_ADDR_PHYS)
 502		return 0;
 503
 504	return 1;
 505}
 506EXPORT_SYMBOL_GPL(mce_usable_address);
 507
 508bool mce_is_memory_error(struct mce *m)
 509{
 510	if (m->cpuvendor == X86_VENDOR_AMD ||
 511	    m->cpuvendor == X86_VENDOR_HYGON) {
 512		return amd_mce_is_memory_error(m);
 513	} else if (m->cpuvendor == X86_VENDOR_INTEL) {
 514		/*
 515		 * Intel SDM Volume 3B - 15.9.2 Compound Error Codes
 516		 *
 517		 * Bit 7 of the MCACOD field of IA32_MCi_STATUS is used for
 518		 * indicating a memory error. Bit 8 is used for indicating a
 519		 * cache hierarchy error. The combination of bit 2 and bit 3
 520		 * is used for indicating a `generic' cache hierarchy error
 521		 * But we can't just blindly check the above bits, because if
 522		 * bit 11 is set, then it is a bus/interconnect error - and
 523		 * either way the above bits just gives more detail on what
 524		 * bus/interconnect error happened. Note that bit 12 can be
 525		 * ignored, as it's the "filter" bit.
 526		 */
 527		return (m->status & 0xef80) == BIT(7) ||
 528		       (m->status & 0xef00) == BIT(8) ||
 529		       (m->status & 0xeffc) == 0xc;
 530	}
 531
 532	return false;
 533}
 534EXPORT_SYMBOL_GPL(mce_is_memory_error);
 535
 536bool mce_is_correctable(struct mce *m)
 537{
 538	if (m->cpuvendor == X86_VENDOR_AMD && m->status & MCI_STATUS_DEFERRED)
 539		return false;
 540
 541	if (m->cpuvendor == X86_VENDOR_HYGON && m->status & MCI_STATUS_DEFERRED)
 542		return false;
 543
 544	if (m->status & MCI_STATUS_UC)
 545		return false;
 546
 547	return true;
 548}
 549EXPORT_SYMBOL_GPL(mce_is_correctable);
 550
 551static bool cec_add_mce(struct mce *m)
 552{
 553	if (!m)
 554		return false;
 555
 556	/* We eat only correctable DRAM errors with usable addresses. */
 557	if (mce_is_memory_error(m) &&
 558	    mce_is_correctable(m)  &&
 559	    mce_usable_address(m))
 560		if (!cec_add_elem(m->addr >> PAGE_SHIFT))
 561			return true;
 562
 563	return false;
 564}
 565
 566static int mce_first_notifier(struct notifier_block *nb, unsigned long val,
 567			      void *data)
 568{
 569	struct mce *m = (struct mce *)data;
 570
 571	if (!m)
 572		return NOTIFY_DONE;
 573
 574	if (cec_add_mce(m))
 575		return NOTIFY_STOP;
 576
 577	/* Emit the trace record: */
 578	trace_mce_record(m);
 579
 580	set_bit(0, &mce_need_notify);
 581
 582	mce_notify_irq();
 583
 584	return NOTIFY_DONE;
 585}
 586
 587static struct notifier_block first_nb = {
 588	.notifier_call	= mce_first_notifier,
 589	.priority	= MCE_PRIO_FIRST,
 590};
 591
 592static int srao_decode_notifier(struct notifier_block *nb, unsigned long val,
 593				void *data)
 594{
 595	struct mce *mce = (struct mce *)data;
 596	unsigned long pfn;
 597
 598	if (!mce)
 599		return NOTIFY_DONE;
 600
 601	if (mce_usable_address(mce) && (mce->severity == MCE_AO_SEVERITY)) {
 602		pfn = mce->addr >> PAGE_SHIFT;
 603		if (!memory_failure(pfn, 0))
 604			set_mce_nospec(pfn);
 605	}
 606
 607	return NOTIFY_OK;
 608}
 609static struct notifier_block mce_srao_nb = {
 610	.notifier_call	= srao_decode_notifier,
 611	.priority	= MCE_PRIO_SRAO,
 612};
 613
 614static int mce_default_notifier(struct notifier_block *nb, unsigned long val,
 615				void *data)
 616{
 617	struct mce *m = (struct mce *)data;
 618
 619	if (!m)
 620		return NOTIFY_DONE;
 621
 622	if (atomic_read(&num_notifiers) > NUM_DEFAULT_NOTIFIERS)
 623		return NOTIFY_DONE;
 624
 625	__print_mce(m);
 626
 627	return NOTIFY_DONE;
 628}
 629
 630static struct notifier_block mce_default_nb = {
 631	.notifier_call	= mce_default_notifier,
 632	/* lowest prio, we want it to run last. */
 633	.priority	= MCE_PRIO_LOWEST,
 634};
 635
 636/*
 637 * Read ADDR and MISC registers.
 638 */
 639static void mce_read_aux(struct mce *m, int i)
 640{
 641	if (m->status & MCI_STATUS_MISCV)
 642		m->misc = mce_rdmsrl(msr_ops.misc(i));
 643
 644	if (m->status & MCI_STATUS_ADDRV) {
 645		m->addr = mce_rdmsrl(msr_ops.addr(i));
 646
 647		/*
 648		 * Mask the reported address by the reported granularity.
 649		 */
 650		if (mca_cfg.ser && (m->status & MCI_STATUS_MISCV)) {
 651			u8 shift = MCI_MISC_ADDR_LSB(m->misc);
 652			m->addr >>= shift;
 653			m->addr <<= shift;
 654		}
 655
 656		/*
 657		 * Extract [55:<lsb>] where lsb is the least significant
 658		 * *valid* bit of the address bits.
 659		 */
 660		if (mce_flags.smca) {
 661			u8 lsb = (m->addr >> 56) & 0x3f;
 662
 663			m->addr &= GENMASK_ULL(55, lsb);
 664		}
 665	}
 666
 667	if (mce_flags.smca) {
 668		m->ipid = mce_rdmsrl(MSR_AMD64_SMCA_MCx_IPID(i));
 669
 670		if (m->status & MCI_STATUS_SYNDV)
 671			m->synd = mce_rdmsrl(MSR_AMD64_SMCA_MCx_SYND(i));
 672	}
 673}
 674
 675DEFINE_PER_CPU(unsigned, mce_poll_count);
 676
 677/*
 678 * Poll for corrected events or events that happened before reset.
 679 * Those are just logged through /dev/mcelog.
 680 *
 681 * This is executed in standard interrupt context.
 682 *
 683 * Note: spec recommends to panic for fatal unsignalled
 684 * errors here. However this would be quite problematic --
 685 * we would need to reimplement the Monarch handling and
 686 * it would mess up the exclusion between exception handler
 687 * and poll handler -- * so we skip this for now.
 688 * These cases should not happen anyways, or only when the CPU
 689 * is already totally * confused. In this case it's likely it will
 690 * not fully execute the machine check handler either.
 691 */
 692bool machine_check_poll(enum mcp_flags flags, mce_banks_t *b)
 693{
 694	struct mce_bank *mce_banks = this_cpu_ptr(mce_banks_array);
 695	bool error_seen = false;
 696	struct mce m;
 697	int i;
 698
 699	this_cpu_inc(mce_poll_count);
 700
 701	mce_gather_info(&m, NULL);
 702
 703	if (flags & MCP_TIMESTAMP)
 704		m.tsc = rdtsc();
 705
 706	for (i = 0; i < this_cpu_read(mce_num_banks); i++) {
 707		if (!mce_banks[i].ctl || !test_bit(i, *b))
 708			continue;
 709
 710		m.misc = 0;
 711		m.addr = 0;
 712		m.bank = i;
 713
 714		barrier();
 715		m.status = mce_rdmsrl(msr_ops.status(i));
 716
 717		/* If this entry is not valid, ignore it */
 718		if (!(m.status & MCI_STATUS_VAL))
 719			continue;
 720
 721		/*
 722		 * If we are logging everything (at CPU online) or this
 723		 * is a corrected error, then we must log it.
 724		 */
 725		if ((flags & MCP_UC) || !(m.status & MCI_STATUS_UC))
 726			goto log_it;
 727
 728		/*
 729		 * Newer Intel systems that support software error
 730		 * recovery need to make additional checks. Other
 731		 * CPUs should skip over uncorrected errors, but log
 732		 * everything else.
 733		 */
 734		if (!mca_cfg.ser) {
 735			if (m.status & MCI_STATUS_UC)
 736				continue;
 737			goto log_it;
 738		}
 739
 740		/* Log "not enabled" (speculative) errors */
 741		if (!(m.status & MCI_STATUS_EN))
 742			goto log_it;
 743
 744		/*
 745		 * Log UCNA (SDM: 15.6.3 "UCR Error Classification")
 746		 * UC == 1 && PCC == 0 && S == 0
 747		 */
 748		if (!(m.status & MCI_STATUS_PCC) && !(m.status & MCI_STATUS_S))
 749			goto log_it;
 750
 751		/*
 752		 * Skip anything else. Presumption is that our read of this
 753		 * bank is racing with a machine check. Leave the log alone
 754		 * for do_machine_check() to deal with it.
 755		 */
 756		continue;
 757
 758log_it:
 759		error_seen = true;
 760
 761		mce_read_aux(&m, i);
 762
 763		m.severity = mce_severity(&m, mca_cfg.tolerant, NULL, false);
 764
 765		/*
 766		 * Don't get the IP here because it's unlikely to
 767		 * have anything to do with the actual error location.
 768		 */
 769		if (!(flags & MCP_DONTLOG) && !mca_cfg.dont_log_ce)
 770			mce_log(&m);
 771		else if (mce_usable_address(&m)) {
 772			/*
 773			 * Although we skipped logging this, we still want
 774			 * to take action. Add to the pool so the registered
 775			 * notifiers will see it.
 776			 */
 777			if (!mce_gen_pool_add(&m))
 778				mce_schedule_work();
 779		}
 780
 781		/*
 782		 * Clear state for this bank.
 783		 */
 784		mce_wrmsrl(msr_ops.status(i), 0);
 785	}
 786
 787	/*
 788	 * Don't clear MCG_STATUS here because it's only defined for
 789	 * exceptions.
 790	 */
 791
 792	sync_core();
 793
 794	return error_seen;
 795}
 796EXPORT_SYMBOL_GPL(machine_check_poll);
 797
 798/*
 799 * Do a quick check if any of the events requires a panic.
 800 * This decides if we keep the events around or clear them.
 801 */
 802static int mce_no_way_out(struct mce *m, char **msg, unsigned long *validp,
 803			  struct pt_regs *regs)
 804{
 805	char *tmp;
 806	int i;
 807
 808	for (i = 0; i < this_cpu_read(mce_num_banks); i++) {
 809		m->status = mce_rdmsrl(msr_ops.status(i));
 810		if (!(m->status & MCI_STATUS_VAL))
 811			continue;
 812
 813		__set_bit(i, validp);
 814		if (quirk_no_way_out)
 815			quirk_no_way_out(i, m, regs);
 816
 817		if (mce_severity(m, mca_cfg.tolerant, &tmp, true) >= MCE_PANIC_SEVERITY) {
 818			m->bank = i;
 819			mce_read_aux(m, i);
 820			*msg = tmp;
 821			return 1;
 822		}
 823	}
 824	return 0;
 825}
 826
 827/*
 828 * Variable to establish order between CPUs while scanning.
 829 * Each CPU spins initially until executing is equal its number.
 830 */
 831static atomic_t mce_executing;
 832
 833/*
 834 * Defines order of CPUs on entry. First CPU becomes Monarch.
 835 */
 836static atomic_t mce_callin;
 837
 838/*
 839 * Check if a timeout waiting for other CPUs happened.
 840 */
 841static int mce_timed_out(u64 *t, const char *msg)
 842{
 843	/*
 844	 * The others already did panic for some reason.
 845	 * Bail out like in a timeout.
 846	 * rmb() to tell the compiler that system_state
 847	 * might have been modified by someone else.
 848	 */
 849	rmb();
 850	if (atomic_read(&mce_panicked))
 851		wait_for_panic();
 852	if (!mca_cfg.monarch_timeout)
 853		goto out;
 854	if ((s64)*t < SPINUNIT) {
 855		if (mca_cfg.tolerant <= 1)
 856			mce_panic(msg, NULL, NULL);
 857		cpu_missing = 1;
 858		return 1;
 859	}
 860	*t -= SPINUNIT;
 861out:
 862	touch_nmi_watchdog();
 863	return 0;
 864}
 865
 866/*
 867 * The Monarch's reign.  The Monarch is the CPU who entered
 868 * the machine check handler first. It waits for the others to
 869 * raise the exception too and then grades them. When any
 870 * error is fatal panic. Only then let the others continue.
 871 *
 872 * The other CPUs entering the MCE handler will be controlled by the
 873 * Monarch. They are called Subjects.
 874 *
 875 * This way we prevent any potential data corruption in a unrecoverable case
 876 * and also makes sure always all CPU's errors are examined.
 877 *
 878 * Also this detects the case of a machine check event coming from outer
 879 * space (not detected by any CPUs) In this case some external agent wants
 880 * us to shut down, so panic too.
 881 *
 882 * The other CPUs might still decide to panic if the handler happens
 883 * in a unrecoverable place, but in this case the system is in a semi-stable
 884 * state and won't corrupt anything by itself. It's ok to let the others
 885 * continue for a bit first.
 886 *
 887 * All the spin loops have timeouts; when a timeout happens a CPU
 888 * typically elects itself to be Monarch.
 889 */
 890static void mce_reign(void)
 891{
 892	int cpu;
 893	struct mce *m = NULL;
 894	int global_worst = 0;
 895	char *msg = NULL;
 896	char *nmsg = NULL;
 897
 898	/*
 899	 * This CPU is the Monarch and the other CPUs have run
 900	 * through their handlers.
 901	 * Grade the severity of the errors of all the CPUs.
 902	 */
 903	for_each_possible_cpu(cpu) {
 904		int severity = mce_severity(&per_cpu(mces_seen, cpu),
 905					    mca_cfg.tolerant,
 906					    &nmsg, true);
 907		if (severity > global_worst) {
 908			msg = nmsg;
 909			global_worst = severity;
 910			m = &per_cpu(mces_seen, cpu);
 911		}
 912	}
 913
 914	/*
 915	 * Cannot recover? Panic here then.
 916	 * This dumps all the mces in the log buffer and stops the
 917	 * other CPUs.
 918	 */
 919	if (m && global_worst >= MCE_PANIC_SEVERITY && mca_cfg.tolerant < 3)
 920		mce_panic("Fatal machine check", m, msg);
 921
 922	/*
 923	 * For UC somewhere we let the CPU who detects it handle it.
 924	 * Also must let continue the others, otherwise the handling
 925	 * CPU could deadlock on a lock.
 926	 */
 927
 928	/*
 929	 * No machine check event found. Must be some external
 930	 * source or one CPU is hung. Panic.
 931	 */
 932	if (global_worst <= MCE_KEEP_SEVERITY && mca_cfg.tolerant < 3)
 933		mce_panic("Fatal machine check from unknown source", NULL, NULL);
 934
 935	/*
 936	 * Now clear all the mces_seen so that they don't reappear on
 937	 * the next mce.
 938	 */
 939	for_each_possible_cpu(cpu)
 940		memset(&per_cpu(mces_seen, cpu), 0, sizeof(struct mce));
 941}
 942
 943static atomic_t global_nwo;
 944
 945/*
 946 * Start of Monarch synchronization. This waits until all CPUs have
 947 * entered the exception handler and then determines if any of them
 948 * saw a fatal event that requires panic. Then it executes them
 949 * in the entry order.
 950 * TBD double check parallel CPU hotunplug
 951 */
 952static int mce_start(int *no_way_out)
 953{
 954	int order;
 955	int cpus = num_online_cpus();
 956	u64 timeout = (u64)mca_cfg.monarch_timeout * NSEC_PER_USEC;
 957
 958	if (!timeout)
 959		return -1;
 960
 961	atomic_add(*no_way_out, &global_nwo);
 962	/*
 963	 * Rely on the implied barrier below, such that global_nwo
 964	 * is updated before mce_callin.
 965	 */
 966	order = atomic_inc_return(&mce_callin);
 967
 968	/*
 969	 * Wait for everyone.
 970	 */
 971	while (atomic_read(&mce_callin) != cpus) {
 972		if (mce_timed_out(&timeout,
 973				  "Timeout: Not all CPUs entered broadcast exception handler")) {
 974			atomic_set(&global_nwo, 0);
 975			return -1;
 976		}
 977		ndelay(SPINUNIT);
 978	}
 979
 980	/*
 981	 * mce_callin should be read before global_nwo
 982	 */
 983	smp_rmb();
 984
 985	if (order == 1) {
 986		/*
 987		 * Monarch: Starts executing now, the others wait.
 988		 */
 989		atomic_set(&mce_executing, 1);
 990	} else {
 991		/*
 992		 * Subject: Now start the scanning loop one by one in
 993		 * the original callin order.
 994		 * This way when there are any shared banks it will be
 995		 * only seen by one CPU before cleared, avoiding duplicates.
 996		 */
 997		while (atomic_read(&mce_executing) < order) {
 998			if (mce_timed_out(&timeout,
 999					  "Timeout: Subject CPUs unable to finish machine check processing")) {
1000				atomic_set(&global_nwo, 0);
1001				return -1;
1002			}
1003			ndelay(SPINUNIT);
1004		}
1005	}
1006
1007	/*
1008	 * Cache the global no_way_out state.
1009	 */
1010	*no_way_out = atomic_read(&global_nwo);
1011
1012	return order;
1013}
1014
1015/*
1016 * Synchronize between CPUs after main scanning loop.
1017 * This invokes the bulk of the Monarch processing.
1018 */
1019static int mce_end(int order)
1020{
1021	int ret = -1;
1022	u64 timeout = (u64)mca_cfg.monarch_timeout * NSEC_PER_USEC;
1023
1024	if (!timeout)
1025		goto reset;
1026	if (order < 0)
1027		goto reset;
1028
1029	/*
1030	 * Allow others to run.
1031	 */
1032	atomic_inc(&mce_executing);
1033
1034	if (order == 1) {
1035		/* CHECKME: Can this race with a parallel hotplug? */
1036		int cpus = num_online_cpus();
1037
1038		/*
1039		 * Monarch: Wait for everyone to go through their scanning
1040		 * loops.
1041		 */
1042		while (atomic_read(&mce_executing) <= cpus) {
1043			if (mce_timed_out(&timeout,
1044					  "Timeout: Monarch CPU unable to finish machine check processing"))
1045				goto reset;
1046			ndelay(SPINUNIT);
1047		}
1048
1049		mce_reign();
1050		barrier();
1051		ret = 0;
1052	} else {
1053		/*
1054		 * Subject: Wait for Monarch to finish.
1055		 */
1056		while (atomic_read(&mce_executing) != 0) {
1057			if (mce_timed_out(&timeout,
1058					  "Timeout: Monarch CPU did not finish machine check processing"))
1059				goto reset;
1060			ndelay(SPINUNIT);
1061		}
1062
1063		/*
1064		 * Don't reset anything. That's done by the Monarch.
1065		 */
1066		return 0;
1067	}
1068
1069	/*
1070	 * Reset all global state.
1071	 */
1072reset:
1073	atomic_set(&global_nwo, 0);
1074	atomic_set(&mce_callin, 0);
1075	barrier();
1076
1077	/*
1078	 * Let others run again.
1079	 */
1080	atomic_set(&mce_executing, 0);
1081	return ret;
1082}
1083
1084static void mce_clear_state(unsigned long *toclear)
1085{
1086	int i;
1087
1088	for (i = 0; i < this_cpu_read(mce_num_banks); i++) {
1089		if (test_bit(i, toclear))
1090			mce_wrmsrl(msr_ops.status(i), 0);
1091	}
1092}
1093
1094static int do_memory_failure(struct mce *m)
1095{
1096	int flags = MF_ACTION_REQUIRED;
1097	int ret;
1098
1099	pr_err("Uncorrected hardware memory error in user-access at %llx", m->addr);
1100	if (!(m->mcgstatus & MCG_STATUS_RIPV))
1101		flags |= MF_MUST_KILL;
1102	ret = memory_failure(m->addr >> PAGE_SHIFT, flags);
1103	if (ret)
1104		pr_err("Memory error not recovered");
1105	else
1106		set_mce_nospec(m->addr >> PAGE_SHIFT);
1107	return ret;
1108}
1109
1110
1111/*
1112 * Cases where we avoid rendezvous handler timeout:
1113 * 1) If this CPU is offline.
1114 *
1115 * 2) If crashing_cpu was set, e.g. we're entering kdump and we need to
1116 *  skip those CPUs which remain looping in the 1st kernel - see
1117 *  crash_nmi_callback().
1118 *
1119 * Note: there still is a small window between kexec-ing and the new,
1120 * kdump kernel establishing a new #MC handler where a broadcasted MCE
1121 * might not get handled properly.
1122 */
1123static bool __mc_check_crashing_cpu(int cpu)
1124{
1125	if (cpu_is_offline(cpu) ||
1126	    (crashing_cpu != -1 && crashing_cpu != cpu)) {
1127		u64 mcgstatus;
1128
1129		mcgstatus = mce_rdmsrl(MSR_IA32_MCG_STATUS);
1130		if (mcgstatus & MCG_STATUS_RIPV) {
1131			mce_wrmsrl(MSR_IA32_MCG_STATUS, 0);
1132			return true;
1133		}
1134	}
1135	return false;
1136}
1137
1138static void __mc_scan_banks(struct mce *m, struct mce *final,
1139			    unsigned long *toclear, unsigned long *valid_banks,
1140			    int no_way_out, int *worst)
1141{
1142	struct mce_bank *mce_banks = this_cpu_ptr(mce_banks_array);
1143	struct mca_config *cfg = &mca_cfg;
1144	int severity, i;
1145
1146	for (i = 0; i < this_cpu_read(mce_num_banks); i++) {
1147		__clear_bit(i, toclear);
1148		if (!test_bit(i, valid_banks))
1149			continue;
1150
1151		if (!mce_banks[i].ctl)
1152			continue;
1153
1154		m->misc = 0;
1155		m->addr = 0;
1156		m->bank = i;
1157
1158		m->status = mce_rdmsrl(msr_ops.status(i));
1159		if (!(m->status & MCI_STATUS_VAL))
1160			continue;
1161
1162		/*
1163		 * Corrected or non-signaled errors are handled by
1164		 * machine_check_poll(). Leave them alone, unless this panics.
1165		 */
1166		if (!(m->status & (cfg->ser ? MCI_STATUS_S : MCI_STATUS_UC)) &&
1167			!no_way_out)
1168			continue;
1169
1170		/* Set taint even when machine check was not enabled. */
1171		add_taint(TAINT_MACHINE_CHECK, LOCKDEP_NOW_UNRELIABLE);
1172
1173		severity = mce_severity(m, cfg->tolerant, NULL, true);
1174
1175		/*
1176		 * When machine check was for corrected/deferred handler don't
1177		 * touch, unless we're panicking.
1178		 */
1179		if ((severity == MCE_KEEP_SEVERITY ||
1180		     severity == MCE_UCNA_SEVERITY) && !no_way_out)
1181			continue;
1182
1183		__set_bit(i, toclear);
1184
1185		/* Machine check event was not enabled. Clear, but ignore. */
1186		if (severity == MCE_NO_SEVERITY)
1187			continue;
1188
1189		mce_read_aux(m, i);
1190
1191		/* assuming valid severity level != 0 */
1192		m->severity = severity;
1193
1194		mce_log(m);
1195
1196		if (severity > *worst) {
1197			*final = *m;
1198			*worst = severity;
1199		}
1200	}
1201
1202	/* mce_clear_state will clear *final, save locally for use later */
1203	*m = *final;
1204}
1205
1206/*
1207 * The actual machine check handler. This only handles real
1208 * exceptions when something got corrupted coming in through int 18.
1209 *
1210 * This is executed in NMI context not subject to normal locking rules. This
1211 * implies that most kernel services cannot be safely used. Don't even
1212 * think about putting a printk in there!
1213 *
1214 * On Intel systems this is entered on all CPUs in parallel through
1215 * MCE broadcast. However some CPUs might be broken beyond repair,
1216 * so be always careful when synchronizing with others.
1217 */
1218void do_machine_check(struct pt_regs *regs, long error_code)
1219{
1220	DECLARE_BITMAP(valid_banks, MAX_NR_BANKS);
1221	DECLARE_BITMAP(toclear, MAX_NR_BANKS);
1222	struct mca_config *cfg = &mca_cfg;
1223	int cpu = smp_processor_id();
1224	char *msg = "Unknown";
1225	struct mce m, *final;
1226	int worst = 0;
1227
1228	/*
1229	 * Establish sequential order between the CPUs entering the machine
1230	 * check handler.
1231	 */
1232	int order = -1;
1233
1234	/*
1235	 * If no_way_out gets set, there is no safe way to recover from this
1236	 * MCE.  If mca_cfg.tolerant is cranked up, we'll try anyway.
1237	 */
1238	int no_way_out = 0;
1239
1240	/*
1241	 * If kill_it gets set, there might be a way to recover from this
1242	 * error.
1243	 */
1244	int kill_it = 0;
1245
1246	/*
1247	 * MCEs are always local on AMD. Same is determined by MCG_STATUS_LMCES
1248	 * on Intel.
1249	 */
1250	int lmce = 1;
1251
1252	if (__mc_check_crashing_cpu(cpu))
1253		return;
1254
1255	ist_enter(regs);
1256
1257	this_cpu_inc(mce_exception_count);
1258
1259	mce_gather_info(&m, regs);
1260	m.tsc = rdtsc();
1261
1262	final = this_cpu_ptr(&mces_seen);
1263	*final = m;
1264
1265	memset(valid_banks, 0, sizeof(valid_banks));
1266	no_way_out = mce_no_way_out(&m, &msg, valid_banks, regs);
1267
1268	barrier();
1269
1270	/*
1271	 * When no restart IP might need to kill or panic.
1272	 * Assume the worst for now, but if we find the
1273	 * severity is MCE_AR_SEVERITY we have other options.
1274	 */
1275	if (!(m.mcgstatus & MCG_STATUS_RIPV))
1276		kill_it = 1;
1277
1278	/*
1279	 * Check if this MCE is signaled to only this logical processor,
1280	 * on Intel only.
1281	 */
1282	if (m.cpuvendor == X86_VENDOR_INTEL)
1283		lmce = m.mcgstatus & MCG_STATUS_LMCES;
1284
1285	/*
1286	 * Local machine check may already know that we have to panic.
1287	 * Broadcast machine check begins rendezvous in mce_start()
1288	 * Go through all banks in exclusion of the other CPUs. This way we
1289	 * don't report duplicated events on shared banks because the first one
1290	 * to see it will clear it.
1291	 */
1292	if (lmce) {
1293		if (no_way_out)
1294			mce_panic("Fatal local machine check", &m, msg);
1295	} else {
1296		order = mce_start(&no_way_out);
1297	}
1298
1299	__mc_scan_banks(&m, final, toclear, valid_banks, no_way_out, &worst);
1300
1301	if (!no_way_out)
1302		mce_clear_state(toclear);
1303
1304	/*
1305	 * Do most of the synchronization with other CPUs.
1306	 * When there's any problem use only local no_way_out state.
1307	 */
1308	if (!lmce) {
1309		if (mce_end(order) < 0)
1310			no_way_out = worst >= MCE_PANIC_SEVERITY;
1311	} else {
1312		/*
1313		 * If there was a fatal machine check we should have
1314		 * already called mce_panic earlier in this function.
1315		 * Since we re-read the banks, we might have found
1316		 * something new. Check again to see if we found a
1317		 * fatal error. We call "mce_severity()" again to
1318		 * make sure we have the right "msg".
1319		 */
1320		if (worst >= MCE_PANIC_SEVERITY && mca_cfg.tolerant < 3) {
1321			mce_severity(&m, cfg->tolerant, &msg, true);
1322			mce_panic("Local fatal machine check!", &m, msg);
1323		}
1324	}
1325
1326	/*
1327	 * If tolerant is at an insane level we drop requests to kill
1328	 * processes and continue even when there is no way out.
1329	 */
1330	if (cfg->tolerant == 3)
1331		kill_it = 0;
1332	else if (no_way_out)
1333		mce_panic("Fatal machine check on current CPU", &m, msg);
1334
1335	if (worst > 0)
1336		irq_work_queue(&mce_irq_work);
1337
1338	mce_wrmsrl(MSR_IA32_MCG_STATUS, 0);
1339
1340	sync_core();
1341
1342	if (worst != MCE_AR_SEVERITY && !kill_it)
1343		goto out_ist;
1344
1345	/* Fault was in user mode and we need to take some action */
1346	if ((m.cs & 3) == 3) {
1347		ist_begin_non_atomic(regs);
1348		local_irq_enable();
1349
1350		if (kill_it || do_memory_failure(&m))
1351			force_sig(SIGBUS);
1352		local_irq_disable();
1353		ist_end_non_atomic();
1354	} else {
1355		if (!fixup_exception(regs, X86_TRAP_MC, error_code, 0))
1356			mce_panic("Failed kernel mode recovery", &m, NULL);
1357	}
1358
1359out_ist:
1360	ist_exit(regs);
1361}
1362EXPORT_SYMBOL_GPL(do_machine_check);
1363
1364#ifndef CONFIG_MEMORY_FAILURE
1365int memory_failure(unsigned long pfn, int flags)
1366{
1367	/* mce_severity() should not hand us an ACTION_REQUIRED error */
1368	BUG_ON(flags & MF_ACTION_REQUIRED);
1369	pr_err("Uncorrected memory error in page 0x%lx ignored\n"
1370	       "Rebuild kernel with CONFIG_MEMORY_FAILURE=y for smarter handling\n",
1371	       pfn);
1372
1373	return 0;
1374}
1375#endif
1376
1377/*
1378 * Periodic polling timer for "silent" machine check errors.  If the
1379 * poller finds an MCE, poll 2x faster.  When the poller finds no more
1380 * errors, poll 2x slower (up to check_interval seconds).
1381 */
1382static unsigned long check_interval = INITIAL_CHECK_INTERVAL;
1383
1384static DEFINE_PER_CPU(unsigned long, mce_next_interval); /* in jiffies */
1385static DEFINE_PER_CPU(struct timer_list, mce_timer);
1386
1387static unsigned long mce_adjust_timer_default(unsigned long interval)
1388{
1389	return interval;
1390}
1391
1392static unsigned long (*mce_adjust_timer)(unsigned long interval) = mce_adjust_timer_default;
1393
1394static void __start_timer(struct timer_list *t, unsigned long interval)
1395{
1396	unsigned long when = jiffies + interval;
1397	unsigned long flags;
1398
1399	local_irq_save(flags);
1400
1401	if (!timer_pending(t) || time_before(when, t->expires))
1402		mod_timer(t, round_jiffies(when));
1403
1404	local_irq_restore(flags);
1405}
1406
1407static void mce_timer_fn(struct timer_list *t)
1408{
1409	struct timer_list *cpu_t = this_cpu_ptr(&mce_timer);
1410	unsigned long iv;
1411
1412	WARN_ON(cpu_t != t);
1413
1414	iv = __this_cpu_read(mce_next_interval);
1415
1416	if (mce_available(this_cpu_ptr(&cpu_info))) {
1417		machine_check_poll(0, this_cpu_ptr(&mce_poll_banks));
1418
1419		if (mce_intel_cmci_poll()) {
1420			iv = mce_adjust_timer(iv);
1421			goto done;
1422		}
1423	}
1424
1425	/*
1426	 * Alert userspace if needed. If we logged an MCE, reduce the polling
1427	 * interval, otherwise increase the polling interval.
1428	 */
1429	if (mce_notify_irq())
1430		iv = max(iv / 2, (unsigned long) HZ/100);
1431	else
1432		iv = min(iv * 2, round_jiffies_relative(check_interval * HZ));
1433
1434done:
1435	__this_cpu_write(mce_next_interval, iv);
1436	__start_timer(t, iv);
1437}
1438
1439/*
1440 * Ensure that the timer is firing in @interval from now.
1441 */
1442void mce_timer_kick(unsigned long interval)
1443{
1444	struct timer_list *t = this_cpu_ptr(&mce_timer);
1445	unsigned long iv = __this_cpu_read(mce_next_interval);
1446
1447	__start_timer(t, interval);
1448
1449	if (interval < iv)
1450		__this_cpu_write(mce_next_interval, interval);
1451}
1452
1453/* Must not be called in IRQ context where del_timer_sync() can deadlock */
1454static void mce_timer_delete_all(void)
1455{
1456	int cpu;
1457
1458	for_each_online_cpu(cpu)
1459		del_timer_sync(&per_cpu(mce_timer, cpu));
1460}
1461
1462/*
1463 * Notify the user(s) about new machine check events.
1464 * Can be called from interrupt context, but not from machine check/NMI
1465 * context.
1466 */
1467int mce_notify_irq(void)
1468{
1469	/* Not more than two messages every minute */
1470	static DEFINE_RATELIMIT_STATE(ratelimit, 60*HZ, 2);
1471
1472	if (test_and_clear_bit(0, &mce_need_notify)) {
1473		mce_work_trigger();
1474
1475		if (__ratelimit(&ratelimit))
1476			pr_info(HW_ERR "Machine check events logged\n");
1477
1478		return 1;
1479	}
1480	return 0;
1481}
1482EXPORT_SYMBOL_GPL(mce_notify_irq);
1483
1484static void __mcheck_cpu_mce_banks_init(void)
1485{
1486	struct mce_bank *mce_banks = this_cpu_ptr(mce_banks_array);
1487	u8 n_banks = this_cpu_read(mce_num_banks);
1488	int i;
1489
1490	for (i = 0; i < n_banks; i++) {
1491		struct mce_bank *b = &mce_banks[i];
1492
1493		/*
1494		 * Init them all, __mcheck_cpu_apply_quirks() is going to apply
1495		 * the required vendor quirks before
1496		 * __mcheck_cpu_init_clear_banks() does the final bank setup.
1497		 */
1498		b->ctl = -1ULL;
1499		b->init = 1;
1500	}
1501}
1502
1503/*
1504 * Initialize Machine Checks for a CPU.
1505 */
1506static void __mcheck_cpu_cap_init(void)
1507{
1508	u64 cap;
1509	u8 b;
1510
1511	rdmsrl(MSR_IA32_MCG_CAP, cap);
1512
1513	b = cap & MCG_BANKCNT_MASK;
1514
1515	if (b > MAX_NR_BANKS) {
1516		pr_warn("CPU%d: Using only %u machine check banks out of %u\n",
1517			smp_processor_id(), MAX_NR_BANKS, b);
1518		b = MAX_NR_BANKS;
1519	}
1520
1521	this_cpu_write(mce_num_banks, b);
1522
1523	__mcheck_cpu_mce_banks_init();
1524
1525	/* Use accurate RIP reporting if available. */
1526	if ((cap & MCG_EXT_P) && MCG_EXT_CNT(cap) >= 9)
1527		mca_cfg.rip_msr = MSR_IA32_MCG_EIP;
1528
1529	if (cap & MCG_SER_P)
1530		mca_cfg.ser = 1;
1531}
1532
1533static void __mcheck_cpu_init_generic(void)
1534{
1535	enum mcp_flags m_fl = 0;
1536	mce_banks_t all_banks;
1537	u64 cap;
1538
1539	if (!mca_cfg.bootlog)
1540		m_fl = MCP_DONTLOG;
1541
1542	/*
1543	 * Log the machine checks left over from the previous reset.
1544	 */
1545	bitmap_fill(all_banks, MAX_NR_BANKS);
1546	machine_check_poll(MCP_UC | m_fl, &all_banks);
1547
1548	cr4_set_bits(X86_CR4_MCE);
1549
1550	rdmsrl(MSR_IA32_MCG_CAP, cap);
1551	if (cap & MCG_CTL_P)
1552		wrmsr(MSR_IA32_MCG_CTL, 0xffffffff, 0xffffffff);
1553}
1554
1555static void __mcheck_cpu_init_clear_banks(void)
1556{
1557	struct mce_bank *mce_banks = this_cpu_ptr(mce_banks_array);
1558	int i;
1559
1560	for (i = 0; i < this_cpu_read(mce_num_banks); i++) {
1561		struct mce_bank *b = &mce_banks[i];
1562
1563		if (!b->init)
1564			continue;
1565		wrmsrl(msr_ops.ctl(i), b->ctl);
1566		wrmsrl(msr_ops.status(i), 0);
1567	}
1568}
1569
1570/*
1571 * Do a final check to see if there are any unused/RAZ banks.
1572 *
1573 * This must be done after the banks have been initialized and any quirks have
1574 * been applied.
1575 *
1576 * Do not call this from any user-initiated flows, e.g. CPU hotplug or sysfs.
1577 * Otherwise, a user who disables a bank will not be able to re-enable it
1578 * without a system reboot.
1579 */
1580static void __mcheck_cpu_check_banks(void)
1581{
1582	struct mce_bank *mce_banks = this_cpu_ptr(mce_banks_array);
1583	u64 msrval;
1584	int i;
1585
1586	for (i = 0; i < this_cpu_read(mce_num_banks); i++) {
1587		struct mce_bank *b = &mce_banks[i];
1588
1589		if (!b->init)
1590			continue;
1591
1592		rdmsrl(msr_ops.ctl(i), msrval);
1593		b->init = !!msrval;
1594	}
1595}
1596
1597/*
1598 * During IFU recovery Sandy Bridge -EP4S processors set the RIPV and
1599 * EIPV bits in MCG_STATUS to zero on the affected logical processor (SDM
1600 * Vol 3B Table 15-20). But this confuses both the code that determines
1601 * whether the machine check occurred in kernel or user mode, and also
1602 * the severity assessment code. Pretend that EIPV was set, and take the
1603 * ip/cs values from the pt_regs that mce_gather_info() ignored earlier.
1604 */
1605static void quirk_sandybridge_ifu(int bank, struct mce *m, struct pt_regs *regs)
1606{
1607	if (bank != 0)
1608		return;
1609	if ((m->mcgstatus & (MCG_STATUS_EIPV|MCG_STATUS_RIPV)) != 0)
1610		return;
1611	if ((m->status & (MCI_STATUS_OVER|MCI_STATUS_UC|
1612		          MCI_STATUS_EN|MCI_STATUS_MISCV|MCI_STATUS_ADDRV|
1613			  MCI_STATUS_PCC|MCI_STATUS_S|MCI_STATUS_AR|
1614			  MCACOD)) !=
1615			 (MCI_STATUS_UC|MCI_STATUS_EN|
1616			  MCI_STATUS_MISCV|MCI_STATUS_ADDRV|MCI_STATUS_S|
1617			  MCI_STATUS_AR|MCACOD_INSTR))
1618		return;
1619
1620	m->mcgstatus |= MCG_STATUS_EIPV;
1621	m->ip = regs->ip;
1622	m->cs = regs->cs;
1623}
1624
1625/* Add per CPU specific workarounds here */
1626static int __mcheck_cpu_apply_quirks(struct cpuinfo_x86 *c)
1627{
1628	struct mce_bank *mce_banks = this_cpu_ptr(mce_banks_array);
1629	struct mca_config *cfg = &mca_cfg;
1630
1631	if (c->x86_vendor == X86_VENDOR_UNKNOWN) {
1632		pr_info("unknown CPU type - not enabling MCE support\n");
1633		return -EOPNOTSUPP;
1634	}
1635
1636	/* This should be disabled by the BIOS, but isn't always */
1637	if (c->x86_vendor == X86_VENDOR_AMD) {
1638		if (c->x86 == 15 && this_cpu_read(mce_num_banks) > 4) {
1639			/*
1640			 * disable GART TBL walk error reporting, which
1641			 * trips off incorrectly with the IOMMU & 3ware
1642			 * & Cerberus:
1643			 */
1644			clear_bit(10, (unsigned long *)&mce_banks[4].ctl);
1645		}
1646		if (c->x86 < 0x11 && cfg->bootlog < 0) {
1647			/*
1648			 * Lots of broken BIOS around that don't clear them
1649			 * by default and leave crap in there. Don't log:
1650			 */
1651			cfg->bootlog = 0;
1652		}
1653		/*
1654		 * Various K7s with broken bank 0 around. Always disable
1655		 * by default.
1656		 */
1657		if (c->x86 == 6 && this_cpu_read(mce_num_banks) > 0)
1658			mce_banks[0].ctl = 0;
1659
1660		/*
1661		 * overflow_recov is supported for F15h Models 00h-0fh
1662		 * even though we don't have a CPUID bit for it.
1663		 */
1664		if (c->x86 == 0x15 && c->x86_model <= 0xf)
1665			mce_flags.overflow_recov = 1;
1666
1667	}
1668
1669	if (c->x86_vendor == X86_VENDOR_INTEL) {
1670		/*
1671		 * SDM documents that on family 6 bank 0 should not be written
1672		 * because it aliases to another special BIOS controlled
1673		 * register.
1674		 * But it's not aliased anymore on model 0x1a+
1675		 * Don't ignore bank 0 completely because there could be a
1676		 * valid event later, merely don't write CTL0.
1677		 */
1678
1679		if (c->x86 == 6 && c->x86_model < 0x1A && this_cpu_read(mce_num_banks) > 0)
1680			mce_banks[0].init = 0;
1681
1682		/*
1683		 * All newer Intel systems support MCE broadcasting. Enable
1684		 * synchronization with a one second timeout.
1685		 */
1686		if ((c->x86 > 6 || (c->x86 == 6 && c->x86_model >= 0xe)) &&
1687			cfg->monarch_timeout < 0)
1688			cfg->monarch_timeout = USEC_PER_SEC;
1689
1690		/*
1691		 * There are also broken BIOSes on some Pentium M and
1692		 * earlier systems:
1693		 */
1694		if (c->x86 == 6 && c->x86_model <= 13 && cfg->bootlog < 0)
1695			cfg->bootlog = 0;
1696
1697		if (c->x86 == 6 && c->x86_model == 45)
1698			quirk_no_way_out = quirk_sandybridge_ifu;
1699	}
1700	if (cfg->monarch_timeout < 0)
1701		cfg->monarch_timeout = 0;
1702	if (cfg->bootlog != 0)
1703		cfg->panic_timeout = 30;
1704
1705	return 0;
1706}
1707
1708static int __mcheck_cpu_ancient_init(struct cpuinfo_x86 *c)
1709{
1710	if (c->x86 != 5)
1711		return 0;
1712
1713	switch (c->x86_vendor) {
1714	case X86_VENDOR_INTEL:
1715		intel_p5_mcheck_init(c);
1716		return 1;
1717		break;
1718	case X86_VENDOR_CENTAUR:
1719		winchip_mcheck_init(c);
1720		return 1;
1721		break;
1722	default:
1723		return 0;
1724	}
1725
1726	return 0;
1727}
1728
1729/*
1730 * Init basic CPU features needed for early decoding of MCEs.
1731 */
1732static void __mcheck_cpu_init_early(struct cpuinfo_x86 *c)
1733{
1734	if (c->x86_vendor == X86_VENDOR_AMD || c->x86_vendor == X86_VENDOR_HYGON) {
1735		mce_flags.overflow_recov = !!cpu_has(c, X86_FEATURE_OVERFLOW_RECOV);
1736		mce_flags.succor	 = !!cpu_has(c, X86_FEATURE_SUCCOR);
1737		mce_flags.smca		 = !!cpu_has(c, X86_FEATURE_SMCA);
1738
1739		if (mce_flags.smca) {
1740			msr_ops.ctl	= smca_ctl_reg;
1741			msr_ops.status	= smca_status_reg;
1742			msr_ops.addr	= smca_addr_reg;
1743			msr_ops.misc	= smca_misc_reg;
1744		}
1745	}
1746}
1747
1748static void mce_centaur_feature_init(struct cpuinfo_x86 *c)
1749{
1750	struct mca_config *cfg = &mca_cfg;
1751
1752	 /*
1753	  * All newer Centaur CPUs support MCE broadcasting. Enable
1754	  * synchronization with a one second timeout.
1755	  */
1756	if ((c->x86 == 6 && c->x86_model == 0xf && c->x86_stepping >= 0xe) ||
1757	     c->x86 > 6) {
1758		if (cfg->monarch_timeout < 0)
1759			cfg->monarch_timeout = USEC_PER_SEC;
1760	}
1761}
1762
1763static void __mcheck_cpu_init_vendor(struct cpuinfo_x86 *c)
1764{
1765	switch (c->x86_vendor) {
1766	case X86_VENDOR_INTEL:
1767		mce_intel_feature_init(c);
1768		mce_adjust_timer = cmci_intel_adjust_timer;
1769		break;
1770
1771	case X86_VENDOR_AMD: {
1772		mce_amd_feature_init(c);
1773		break;
1774		}
1775
1776	case X86_VENDOR_HYGON:
1777		mce_hygon_feature_init(c);
1778		break;
1779
1780	case X86_VENDOR_CENTAUR:
1781		mce_centaur_feature_init(c);
1782		break;
1783
1784	default:
1785		break;
1786	}
1787}
1788
1789static void __mcheck_cpu_clear_vendor(struct cpuinfo_x86 *c)
1790{
1791	switch (c->x86_vendor) {
1792	case X86_VENDOR_INTEL:
1793		mce_intel_feature_clear(c);
1794		break;
1795	default:
1796		break;
1797	}
1798}
1799
1800static void mce_start_timer(struct timer_list *t)
1801{
1802	unsigned long iv = check_interval * HZ;
1803
1804	if (mca_cfg.ignore_ce || !iv)
1805		return;
1806
1807	this_cpu_write(mce_next_interval, iv);
1808	__start_timer(t, iv);
1809}
1810
1811static void __mcheck_cpu_setup_timer(void)
1812{
1813	struct timer_list *t = this_cpu_ptr(&mce_timer);
1814
1815	timer_setup(t, mce_timer_fn, TIMER_PINNED);
1816}
1817
1818static void __mcheck_cpu_init_timer(void)
1819{
1820	struct timer_list *t = this_cpu_ptr(&mce_timer);
1821
1822	timer_setup(t, mce_timer_fn, TIMER_PINNED);
1823	mce_start_timer(t);
1824}
1825
1826bool filter_mce(struct mce *m)
1827{
1828	if (boot_cpu_data.x86_vendor == X86_VENDOR_AMD)
1829		return amd_filter_mce(m);
1830
1831	return false;
1832}
1833
1834/* Handle unconfigured int18 (should never happen) */
1835static void unexpected_machine_check(struct pt_regs *regs, long error_code)
1836{
1837	pr_err("CPU#%d: Unexpected int18 (Machine Check)\n",
1838	       smp_processor_id());
1839}
1840
1841/* Call the installed machine check handler for this CPU setup. */
1842void (*machine_check_vector)(struct pt_regs *, long error_code) =
1843						unexpected_machine_check;
1844
1845dotraplinkage void do_mce(struct pt_regs *regs, long error_code)
1846{
1847	machine_check_vector(regs, error_code);
1848}
1849
1850/*
1851 * Called for each booted CPU to set up machine checks.
1852 * Must be called with preempt off:
1853 */
1854void mcheck_cpu_init(struct cpuinfo_x86 *c)
1855{
1856	if (mca_cfg.disabled)
1857		return;
1858
1859	if (__mcheck_cpu_ancient_init(c))
1860		return;
1861
1862	if (!mce_available(c))
1863		return;
1864
1865	__mcheck_cpu_cap_init();
1866
1867	if (__mcheck_cpu_apply_quirks(c) < 0) {
1868		mca_cfg.disabled = 1;
1869		return;
1870	}
1871
1872	if (mce_gen_pool_init()) {
1873		mca_cfg.disabled = 1;
1874		pr_emerg("Couldn't allocate MCE records pool!\n");
1875		return;
1876	}
1877
1878	machine_check_vector = do_machine_check;
1879
1880	__mcheck_cpu_init_early(c);
1881	__mcheck_cpu_init_generic();
1882	__mcheck_cpu_init_vendor(c);
1883	__mcheck_cpu_init_clear_banks();
1884	__mcheck_cpu_check_banks();
1885	__mcheck_cpu_setup_timer();
1886}
1887
1888/*
1889 * Called for each booted CPU to clear some machine checks opt-ins
1890 */
1891void mcheck_cpu_clear(struct cpuinfo_x86 *c)
1892{
1893	if (mca_cfg.disabled)
1894		return;
1895
1896	if (!mce_available(c))
1897		return;
1898
1899	/*
1900	 * Possibly to clear general settings generic to x86
1901	 * __mcheck_cpu_clear_generic(c);
1902	 */
1903	__mcheck_cpu_clear_vendor(c);
1904
1905}
1906
1907static void __mce_disable_bank(void *arg)
1908{
1909	int bank = *((int *)arg);
1910	__clear_bit(bank, this_cpu_ptr(mce_poll_banks));
1911	cmci_disable_bank(bank);
1912}
1913
1914void mce_disable_bank(int bank)
1915{
1916	if (bank >= this_cpu_read(mce_num_banks)) {
1917		pr_warn(FW_BUG
1918			"Ignoring request to disable invalid MCA bank %d.\n",
1919			bank);
1920		return;
1921	}
1922	set_bit(bank, mce_banks_ce_disabled);
1923	on_each_cpu(__mce_disable_bank, &bank, 1);
1924}
1925
1926/*
1927 * mce=off Disables machine check
1928 * mce=no_cmci Disables CMCI
1929 * mce=no_lmce Disables LMCE
1930 * mce=dont_log_ce Clears corrected events silently, no log created for CEs.
1931 * mce=ignore_ce Disables polling and CMCI, corrected events are not cleared.
1932 * mce=TOLERANCELEVEL[,monarchtimeout] (number, see above)
1933 *	monarchtimeout is how long to wait for other CPUs on machine
1934 *	check, or 0 to not wait
1935 * mce=bootlog Log MCEs from before booting. Disabled by default on AMD Fam10h
1936	and older.
1937 * mce=nobootlog Don't log MCEs from before booting.
1938 * mce=bios_cmci_threshold Don't program the CMCI threshold
1939 * mce=recovery force enable memcpy_mcsafe()
1940 */
1941static int __init mcheck_enable(char *str)
1942{
1943	struct mca_config *cfg = &mca_cfg;
1944
1945	if (*str == 0) {
1946		enable_p5_mce();
1947		return 1;
1948	}
1949	if (*str == '=')
1950		str++;
1951	if (!strcmp(str, "off"))
1952		cfg->disabled = 1;
1953	else if (!strcmp(str, "no_cmci"))
1954		cfg->cmci_disabled = true;
1955	else if (!strcmp(str, "no_lmce"))
1956		cfg->lmce_disabled = 1;
1957	else if (!strcmp(str, "dont_log_ce"))
1958		cfg->dont_log_ce = true;
1959	else if (!strcmp(str, "ignore_ce"))
1960		cfg->ignore_ce = true;
1961	else if (!strcmp(str, "bootlog") || !strcmp(str, "nobootlog"))
1962		cfg->bootlog = (str[0] == 'b');
1963	else if (!strcmp(str, "bios_cmci_threshold"))
1964		cfg->bios_cmci_threshold = 1;
1965	else if (!strcmp(str, "recovery"))
1966		cfg->recovery = 1;
1967	else if (isdigit(str[0])) {
1968		if (get_option(&str, &cfg->tolerant) == 2)
1969			get_option(&str, &(cfg->monarch_timeout));
1970	} else {
1971		pr_info("mce argument %s ignored. Please use /sys\n", str);
1972		return 0;
1973	}
1974	return 1;
1975}
1976__setup("mce", mcheck_enable);
1977
1978int __init mcheck_init(void)
1979{
1980	mcheck_intel_therm_init();
1981	mce_register_decode_chain(&first_nb);
1982	mce_register_decode_chain(&mce_srao_nb);
1983	mce_register_decode_chain(&mce_default_nb);
1984	mcheck_vendor_init_severity();
1985
1986	INIT_WORK(&mce_work, mce_gen_pool_process);
1987	init_irq_work(&mce_irq_work, mce_irq_work_cb);
1988
1989	return 0;
1990}
1991
1992/*
1993 * mce_syscore: PM support
1994 */
1995
1996/*
1997 * Disable machine checks on suspend and shutdown. We can't really handle
1998 * them later.
1999 */
2000static void mce_disable_error_reporting(void)
2001{
2002	struct mce_bank *mce_banks = this_cpu_ptr(mce_banks_array);
2003	int i;
2004
2005	for (i = 0; i < this_cpu_read(mce_num_banks); i++) {
2006		struct mce_bank *b = &mce_banks[i];
2007
2008		if (b->init)
2009			wrmsrl(msr_ops.ctl(i), 0);
2010	}
2011	return;
2012}
2013
2014static void vendor_disable_error_reporting(void)
2015{
2016	/*
2017	 * Don't clear on Intel or AMD or Hygon CPUs. Some of these MSRs
2018	 * are socket-wide.
2019	 * Disabling them for just a single offlined CPU is bad, since it will
2020	 * inhibit reporting for all shared resources on the socket like the
2021	 * last level cache (LLC), the integrated memory controller (iMC), etc.
2022	 */
2023	if (boot_cpu_data.x86_vendor == X86_VENDOR_INTEL ||
2024	    boot_cpu_data.x86_vendor == X86_VENDOR_HYGON ||
2025	    boot_cpu_data.x86_vendor == X86_VENDOR_AMD)
2026		return;
2027
2028	mce_disable_error_reporting();
2029}
2030
2031static int mce_syscore_suspend(void)
2032{
2033	vendor_disable_error_reporting();
2034	return 0;
2035}
2036
2037static void mce_syscore_shutdown(void)
2038{
2039	vendor_disable_error_reporting();
2040}
2041
2042/*
2043 * On resume clear all MCE state. Don't want to see leftovers from the BIOS.
2044 * Only one CPU is active at this time, the others get re-added later using
2045 * CPU hotplug:
2046 */
2047static void mce_syscore_resume(void)
2048{
2049	__mcheck_cpu_init_generic();
2050	__mcheck_cpu_init_vendor(raw_cpu_ptr(&cpu_info));
2051	__mcheck_cpu_init_clear_banks();
2052}
2053
2054static struct syscore_ops mce_syscore_ops = {
2055	.suspend	= mce_syscore_suspend,
2056	.shutdown	= mce_syscore_shutdown,
2057	.resume		= mce_syscore_resume,
2058};
2059
2060/*
2061 * mce_device: Sysfs support
2062 */
2063
2064static void mce_cpu_restart(void *data)
2065{
2066	if (!mce_available(raw_cpu_ptr(&cpu_info)))
2067		return;
2068	__mcheck_cpu_init_generic();
2069	__mcheck_cpu_init_clear_banks();
2070	__mcheck_cpu_init_timer();
2071}
2072
2073/* Reinit MCEs after user configuration changes */
2074static void mce_restart(void)
2075{
2076	mce_timer_delete_all();
2077	on_each_cpu(mce_cpu_restart, NULL, 1);
2078}
2079
2080/* Toggle features for corrected errors */
2081static void mce_disable_cmci(void *data)
2082{
2083	if (!mce_available(raw_cpu_ptr(&cpu_info)))
2084		return;
2085	cmci_clear();
2086}
2087
2088static void mce_enable_ce(void *all)
2089{
2090	if (!mce_available(raw_cpu_ptr(&cpu_info)))
2091		return;
2092	cmci_reenable();
2093	cmci_recheck();
2094	if (all)
2095		__mcheck_cpu_init_timer();
2096}
2097
2098static struct bus_type mce_subsys = {
2099	.name		= "machinecheck",
2100	.dev_name	= "machinecheck",
2101};
2102
2103DEFINE_PER_CPU(struct device *, mce_device);
2104
2105static inline struct mce_bank_dev *attr_to_bank(struct device_attribute *attr)
2106{
2107	return container_of(attr, struct mce_bank_dev, attr);
2108}
2109
2110static ssize_t show_bank(struct device *s, struct device_attribute *attr,
2111			 char *buf)
2112{
2113	u8 bank = attr_to_bank(attr)->bank;
2114	struct mce_bank *b;
2115
2116	if (bank >= per_cpu(mce_num_banks, s->id))
2117		return -EINVAL;
2118
2119	b = &per_cpu(mce_banks_array, s->id)[bank];
2120
2121	if (!b->init)
2122		return -ENODEV;
2123
2124	return sprintf(buf, "%llx\n", b->ctl);
2125}
2126
2127static ssize_t set_bank(struct device *s, struct device_attribute *attr,
2128			const char *buf, size_t size)
2129{
2130	u8 bank = attr_to_bank(attr)->bank;
2131	struct mce_bank *b;
2132	u64 new;
2133
2134	if (kstrtou64(buf, 0, &new) < 0)
2135		return -EINVAL;
2136
2137	if (bank >= per_cpu(mce_num_banks, s->id))
2138		return -EINVAL;
2139
2140	b = &per_cpu(mce_banks_array, s->id)[bank];
2141
2142	if (!b->init)
2143		return -ENODEV;
2144
2145	b->ctl = new;
2146	mce_restart();
2147
2148	return size;
2149}
2150
2151static ssize_t set_ignore_ce(struct device *s,
2152			     struct device_attribute *attr,
2153			     const char *buf, size_t size)
2154{
2155	u64 new;
2156
2157	if (kstrtou64(buf, 0, &new) < 0)
2158		return -EINVAL;
2159
2160	mutex_lock(&mce_sysfs_mutex);
2161	if (mca_cfg.ignore_ce ^ !!new) {
2162		if (new) {
2163			/* disable ce features */
2164			mce_timer_delete_all();
2165			on_each_cpu(mce_disable_cmci, NULL, 1);
2166			mca_cfg.ignore_ce = true;
2167		} else {
2168			/* enable ce features */
2169			mca_cfg.ignore_ce = false;
2170			on_each_cpu(mce_enable_ce, (void *)1, 1);
2171		}
2172	}
2173	mutex_unlock(&mce_sysfs_mutex);
2174
2175	return size;
2176}
2177
2178static ssize_t set_cmci_disabled(struct device *s,
2179				 struct device_attribute *attr,
2180				 const char *buf, size_t size)
2181{
2182	u64 new;
2183
2184	if (kstrtou64(buf, 0, &new) < 0)
2185		return -EINVAL;
2186
2187	mutex_lock(&mce_sysfs_mutex);
2188	if (mca_cfg.cmci_disabled ^ !!new) {
2189		if (new) {
2190			/* disable cmci */
2191			on_each_cpu(mce_disable_cmci, NULL, 1);
2192			mca_cfg.cmci_disabled = true;
2193		} else {
2194			/* enable cmci */
2195			mca_cfg.cmci_disabled = false;
2196			on_each_cpu(mce_enable_ce, NULL, 1);
2197		}
2198	}
2199	mutex_unlock(&mce_sysfs_mutex);
2200
2201	return size;
2202}
2203
2204static ssize_t store_int_with_restart(struct device *s,
2205				      struct device_attribute *attr,
2206				      const char *buf, size_t size)
2207{
2208	unsigned long old_check_interval = check_interval;
2209	ssize_t ret = device_store_ulong(s, attr, buf, size);
2210
2211	if (check_interval == old_check_interval)
2212		return ret;
2213
2214	mutex_lock(&mce_sysfs_mutex);
2215	mce_restart();
2216	mutex_unlock(&mce_sysfs_mutex);
2217
2218	return ret;
2219}
2220
2221static DEVICE_INT_ATTR(tolerant, 0644, mca_cfg.tolerant);
2222static DEVICE_INT_ATTR(monarch_timeout, 0644, mca_cfg.monarch_timeout);
2223static DEVICE_BOOL_ATTR(dont_log_ce, 0644, mca_cfg.dont_log_ce);
2224
2225static struct dev_ext_attribute dev_attr_check_interval = {
2226	__ATTR(check_interval, 0644, device_show_int, store_int_with_restart),
2227	&check_interval
2228};
2229
2230static struct dev_ext_attribute dev_attr_ignore_ce = {
2231	__ATTR(ignore_ce, 0644, device_show_bool, set_ignore_ce),
2232	&mca_cfg.ignore_ce
2233};
2234
2235static struct dev_ext_attribute dev_attr_cmci_disabled = {
2236	__ATTR(cmci_disabled, 0644, device_show_bool, set_cmci_disabled),
2237	&mca_cfg.cmci_disabled
2238};
2239
2240static struct device_attribute *mce_device_attrs[] = {
2241	&dev_attr_tolerant.attr,
2242	&dev_attr_check_interval.attr,
2243#ifdef CONFIG_X86_MCELOG_LEGACY
2244	&dev_attr_trigger,
2245#endif
2246	&dev_attr_monarch_timeout.attr,
2247	&dev_attr_dont_log_ce.attr,
2248	&dev_attr_ignore_ce.attr,
2249	&dev_attr_cmci_disabled.attr,
2250	NULL
2251};
2252
2253static cpumask_var_t mce_device_initialized;
2254
2255static void mce_device_release(struct device *dev)
2256{
2257	kfree(dev);
2258}
2259
2260/* Per CPU device init. All of the CPUs still share the same bank device: */
2261static int mce_device_create(unsigned int cpu)
2262{
2263	struct device *dev;
2264	int err;
2265	int i, j;
2266
2267	if (!mce_available(&boot_cpu_data))
2268		return -EIO;
2269
2270	dev = per_cpu(mce_device, cpu);
2271	if (dev)
2272		return 0;
2273
2274	dev = kzalloc(sizeof(*dev), GFP_KERNEL);
2275	if (!dev)
2276		return -ENOMEM;
2277	dev->id  = cpu;
2278	dev->bus = &mce_subsys;
2279	dev->release = &mce_device_release;
2280
2281	err = device_register(dev);
2282	if (err) {
2283		put_device(dev);
2284		return err;
2285	}
2286
2287	for (i = 0; mce_device_attrs[i]; i++) {
2288		err = device_create_file(dev, mce_device_attrs[i]);
2289		if (err)
2290			goto error;
2291	}
2292	for (j = 0; j < per_cpu(mce_num_banks, cpu); j++) {
2293		err = device_create_file(dev, &mce_bank_devs[j].attr);
2294		if (err)
2295			goto error2;
2296	}
2297	cpumask_set_cpu(cpu, mce_device_initialized);
2298	per_cpu(mce_device, cpu) = dev;
2299
2300	return 0;
2301error2:
2302	while (--j >= 0)
2303		device_remove_file(dev, &mce_bank_devs[j].attr);
2304error:
2305	while (--i >= 0)
2306		device_remove_file(dev, mce_device_attrs[i]);
2307
2308	device_unregister(dev);
2309
2310	return err;
2311}
2312
2313static void mce_device_remove(unsigned int cpu)
2314{
2315	struct device *dev = per_cpu(mce_device, cpu);
2316	int i;
2317
2318	if (!cpumask_test_cpu(cpu, mce_device_initialized))
2319		return;
2320
2321	for (i = 0; mce_device_attrs[i]; i++)
2322		device_remove_file(dev, mce_device_attrs[i]);
2323
2324	for (i = 0; i < per_cpu(mce_num_banks, cpu); i++)
2325		device_remove_file(dev, &mce_bank_devs[i].attr);
2326
2327	device_unregister(dev);
2328	cpumask_clear_cpu(cpu, mce_device_initialized);
2329	per_cpu(mce_device, cpu) = NULL;
2330}
2331
2332/* Make sure there are no machine checks on offlined CPUs. */
2333static void mce_disable_cpu(void)
2334{
2335	if (!mce_available(raw_cpu_ptr(&cpu_info)))
2336		return;
2337
2338	if (!cpuhp_tasks_frozen)
2339		cmci_clear();
2340
2341	vendor_disable_error_reporting();
2342}
2343
2344static void mce_reenable_cpu(void)
2345{
2346	struct mce_bank *mce_banks = this_cpu_ptr(mce_banks_array);
2347	int i;
2348
2349	if (!mce_available(raw_cpu_ptr(&cpu_info)))
2350		return;
2351
2352	if (!cpuhp_tasks_frozen)
2353		cmci_reenable();
2354	for (i = 0; i < this_cpu_read(mce_num_banks); i++) {
2355		struct mce_bank *b = &mce_banks[i];
2356
2357		if (b->init)
2358			wrmsrl(msr_ops.ctl(i), b->ctl);
2359	}
2360}
2361
2362static int mce_cpu_dead(unsigned int cpu)
2363{
2364	mce_intel_hcpu_update(cpu);
2365
2366	/* intentionally ignoring frozen here */
2367	if (!cpuhp_tasks_frozen)
2368		cmci_rediscover();
2369	return 0;
2370}
2371
2372static int mce_cpu_online(unsigned int cpu)
2373{
2374	struct timer_list *t = this_cpu_ptr(&mce_timer);
2375	int ret;
2376
2377	mce_device_create(cpu);
2378
2379	ret = mce_threshold_create_device(cpu);
2380	if (ret) {
2381		mce_device_remove(cpu);
2382		return ret;
2383	}
2384	mce_reenable_cpu();
2385	mce_start_timer(t);
2386	return 0;
2387}
2388
2389static int mce_cpu_pre_down(unsigned int cpu)
2390{
2391	struct timer_list *t = this_cpu_ptr(&mce_timer);
2392
2393	mce_disable_cpu();
2394	del_timer_sync(t);
2395	mce_threshold_remove_device(cpu);
2396	mce_device_remove(cpu);
2397	return 0;
2398}
2399
2400static __init void mce_init_banks(void)
2401{
2402	int i;
2403
2404	for (i = 0; i < MAX_NR_BANKS; i++) {
2405		struct mce_bank_dev *b = &mce_bank_devs[i];
2406		struct device_attribute *a = &b->attr;
2407
2408		b->bank = i;
2409
2410		sysfs_attr_init(&a->attr);
2411		a->attr.name	= b->attrname;
2412		snprintf(b->attrname, ATTR_LEN, "bank%d", i);
2413
2414		a->attr.mode	= 0644;
2415		a->show		= show_bank;
2416		a->store	= set_bank;
2417	}
2418}
2419
2420static __init int mcheck_init_device(void)
2421{
2422	int err;
2423
2424	/*
2425	 * Check if we have a spare virtual bit. This will only become
2426	 * a problem if/when we move beyond 5-level page tables.
2427	 */
2428	MAYBE_BUILD_BUG_ON(__VIRTUAL_MASK_SHIFT >= 63);
2429
2430	if (!mce_available(&boot_cpu_data)) {
2431		err = -EIO;
2432		goto err_out;
2433	}
2434
2435	if (!zalloc_cpumask_var(&mce_device_initialized, GFP_KERNEL)) {
2436		err = -ENOMEM;
2437		goto err_out;
2438	}
2439
2440	mce_init_banks();
2441
2442	err = subsys_system_register(&mce_subsys, NULL);
2443	if (err)
2444		goto err_out_mem;
2445
2446	err = cpuhp_setup_state(CPUHP_X86_MCE_DEAD, "x86/mce:dead", NULL,
2447				mce_cpu_dead);
2448	if (err)
2449		goto err_out_mem;
2450
2451	err = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "x86/mce:online",
2452				mce_cpu_online, mce_cpu_pre_down);
2453	if (err < 0)
2454		goto err_out_online;
2455
2456	register_syscore_ops(&mce_syscore_ops);
2457
2458	return 0;
2459
2460err_out_online:
2461	cpuhp_remove_state(CPUHP_X86_MCE_DEAD);
2462
2463err_out_mem:
2464	free_cpumask_var(mce_device_initialized);
2465
2466err_out:
2467	pr_err("Unable to init MCE device (rc: %d)\n", err);
2468
2469	return err;
2470}
2471device_initcall_sync(mcheck_init_device);
2472
2473/*
2474 * Old style boot options parsing. Only for compatibility.
2475 */
2476static int __init mcheck_disable(char *str)
2477{
2478	mca_cfg.disabled = 1;
2479	return 1;
2480}
2481__setup("nomce", mcheck_disable);
2482
2483#ifdef CONFIG_DEBUG_FS
2484struct dentry *mce_get_debugfs_dir(void)
2485{
2486	static struct dentry *dmce;
2487
2488	if (!dmce)
2489		dmce = debugfs_create_dir("mce", NULL);
2490
2491	return dmce;
2492}
2493
2494static void mce_reset(void)
2495{
2496	cpu_missing = 0;
2497	atomic_set(&mce_fake_panicked, 0);
2498	atomic_set(&mce_executing, 0);
2499	atomic_set(&mce_callin, 0);
2500	atomic_set(&global_nwo, 0);
2501}
2502
2503static int fake_panic_get(void *data, u64 *val)
2504{
2505	*val = fake_panic;
2506	return 0;
2507}
2508
2509static int fake_panic_set(void *data, u64 val)
2510{
2511	mce_reset();
2512	fake_panic = val;
2513	return 0;
2514}
2515
2516DEFINE_DEBUGFS_ATTRIBUTE(fake_panic_fops, fake_panic_get, fake_panic_set,
2517			 "%llu\n");
2518
2519static void __init mcheck_debugfs_init(void)
2520{
2521	struct dentry *dmce;
2522
2523	dmce = mce_get_debugfs_dir();
2524	debugfs_create_file_unsafe("fake_panic", 0444, dmce, NULL,
2525				   &fake_panic_fops);
2526}
2527#else
2528static void __init mcheck_debugfs_init(void) { }
2529#endif
2530
2531DEFINE_STATIC_KEY_FALSE(mcsafe_key);
2532EXPORT_SYMBOL_GPL(mcsafe_key);
2533
2534static int __init mcheck_late_init(void)
2535{
2536	if (mca_cfg.recovery)
2537		static_branch_inc(&mcsafe_key);
2538
2539	mcheck_debugfs_init();
2540	cec_init();
2541
2542	/*
2543	 * Flush out everything that has been logged during early boot, now that
2544	 * everything has been initialized (workqueues, decoders, ...).
2545	 */
2546	mce_schedule_work();
2547
2548	return 0;
2549}
2550late_initcall(mcheck_late_init);