nmi.c

  1// SPDX-License-Identifier: GPL-2.0
  2/*
  3 *   Machine check handler
  4 *
  5 *    Copyright IBM Corp. 2000, 2009
  6 *    Author(s): Ingo Adlung <adlung@de.ibm.com>,
  7 *		 Martin Schwidefsky <schwidefsky@de.ibm.com>,
  8 *		 Cornelia Huck <cornelia.huck@de.ibm.com>,
  9 */
 10
 11#include <linux/kernel_stat.h>
 12#include <linux/utsname.h>
 13#include <linux/cpufeature.h>
 14#include <linux/init.h>
 15#include <linux/errno.h>
 16#include <linux/entry-common.h>
 17#include <linux/hardirq.h>
 18#include <linux/log2.h>
 19#include <linux/kprobes.h>
 20#include <linux/kmemleak.h>
 21#include <linux/time.h>
 22#include <linux/module.h>
 23#include <linux/sched/signal.h>
 24#include <linux/kvm_host.h>
 25#include <asm/lowcore.h>
 26#include <asm/ctlreg.h>
 27#include <asm/fpu.h>
 28#include <asm/smp.h>
 29#include <asm/stp.h>
 30#include <asm/cputime.h>
 31#include <asm/nmi.h>
 32#include <asm/crw.h>
 33#include <asm/asm-offsets.h>
 34#include <asm/pai.h>
 35#include <asm/vtime.h>
 36
 37struct mcck_struct {
 38	unsigned int kill_task : 1;
 39	unsigned int channel_report : 1;
 40	unsigned int warning : 1;
 41	unsigned int stp_queue : 1;
 42	unsigned long mcck_code;
 43};
 44
 45static DEFINE_PER_CPU(struct mcck_struct, cpu_mcck);
 46
 47static inline int nmi_needs_mcesa(void)
 48{
 49	return cpu_has_vx() || cpu_has_gs();
 50}
 51
 52/*
 53 * The initial machine check extended save area for the boot CPU.
 54 * It will be replaced on the boot CPU reinit with an allocated
 55 * structure. The structure is required for machine check happening
 56 * early in the boot process.
 57 */
 58static struct mcesa boot_mcesa __aligned(MCESA_MAX_SIZE);
 59
 60void __init nmi_alloc_mcesa_early(u64 *mcesad)
 61{
 62	if (!nmi_needs_mcesa())
 63		return;
 64	*mcesad = __pa(&boot_mcesa);
 65	if (cpu_has_gs())
 66		*mcesad |= ilog2(MCESA_MAX_SIZE);
 67}
 68
 69int nmi_alloc_mcesa(u64 *mcesad)
 70{
 71	unsigned long size;
 72	void *origin;
 73
 74	*mcesad = 0;
 75	if (!nmi_needs_mcesa())
 76		return 0;
 77	size = cpu_has_gs() ? MCESA_MAX_SIZE : MCESA_MIN_SIZE;
 78	origin = kmalloc(size, GFP_KERNEL);
 79	if (!origin)
 80		return -ENOMEM;
 81	/* The pointer is stored with mcesa_bits ORed in */
 82	kmemleak_not_leak(origin);
 83	*mcesad = __pa(origin);
 84	if (cpu_has_gs())
 85		*mcesad |= ilog2(MCESA_MAX_SIZE);
 86	return 0;
 87}
 88
 89void nmi_free_mcesa(u64 *mcesad)
 90{
 91	if (!nmi_needs_mcesa())
 92		return;
 93	kfree(__va(*mcesad & MCESA_ORIGIN_MASK));
 94}
 95
 96static __always_inline char *nmi_puts(char *dest, const char *src)
 97{
 98	while (*src)
 99		*dest++ = *src++;
100	*dest = 0;
101	return dest;
102}
103
104static __always_inline char *u64_to_hex(char *dest, u64 val)
105{
106	int i, num;
107
108	for (i = 1; i <= 16; i++) {
109		num = (val >> (64 - 4 * i)) & 0xf;
110		if (num >= 10)
111			*dest++ = 'A' + num - 10;
112		else
113			*dest++ = '0' + num;
114	}
115	*dest = 0;
116	return dest;
117}
118
119static notrace void nmi_print_info(void)
120{
121	struct lowcore *lc = get_lowcore();
122	char message[100];
123	char *ptr;
124	int i;
125
126	ptr = nmi_puts(message, "Unrecoverable machine check, code: ");
127	ptr = u64_to_hex(ptr, lc->mcck_interruption_code);
128	ptr = nmi_puts(ptr, "\n");
129	sclp_emergency_printk(message);
130
131	ptr = nmi_puts(message, init_utsname()->release);
132	ptr = nmi_puts(ptr, "\n");
133	sclp_emergency_printk(message);
134
135	ptr = nmi_puts(message, arch_hw_string);
136	ptr = nmi_puts(ptr, "\n");
137	sclp_emergency_printk(message);
138
139	ptr = nmi_puts(message, "PSW: ");
140	ptr = u64_to_hex(ptr, lc->mcck_old_psw.mask);
141	ptr = nmi_puts(ptr, " ");
142	ptr = u64_to_hex(ptr, lc->mcck_old_psw.addr);
143	ptr = nmi_puts(ptr, " PFX: ");
144	ptr = u64_to_hex(ptr, (u64)get_lowcore());
145	ptr = nmi_puts(ptr, "\n");
146	sclp_emergency_printk(message);
147
148	ptr = nmi_puts(message, "LBA: ");
149	ptr = u64_to_hex(ptr, lc->last_break_save_area);
150	ptr = nmi_puts(ptr, " EDC: ");
151	ptr = u64_to_hex(ptr, lc->external_damage_code);
152	ptr = nmi_puts(ptr, " FSA: ");
153	ptr = u64_to_hex(ptr, lc->failing_storage_address);
154	ptr = nmi_puts(ptr, "\n");
155	sclp_emergency_printk(message);
156
157	ptr = nmi_puts(message, "CRS:\n");
158	sclp_emergency_printk(message);
159	ptr = message;
160	for (i = 0; i < 16; i++) {
161		ptr = u64_to_hex(ptr, lc->cregs_save_area[i].val);
162		ptr = nmi_puts(ptr, " ");
163		if ((i + 1) % 4 == 0) {
164			ptr = nmi_puts(ptr, "\n");
165			sclp_emergency_printk(message);
166			ptr = message;
167		}
168	}
169
170	ptr = nmi_puts(message, "GPRS:\n");
171	sclp_emergency_printk(message);
172	ptr = message;
173	for (i = 0; i < 16; i++) {
174		ptr = u64_to_hex(ptr, lc->gpregs_save_area[i]);
175		ptr = nmi_puts(ptr, " ");
176		if ((i + 1) % 4 == 0) {
177			ptr = nmi_puts(ptr, "\n");
178			sclp_emergency_printk(message);
179			ptr = message;
180		}
181	}
182
183	ptr = nmi_puts(message, "System stopped\n");
184	sclp_emergency_printk(message);
185}
186
187static notrace void __noreturn s390_handle_damage(void)
188{
189	struct lowcore *lc = get_lowcore();
190	union ctlreg0 cr0, cr0_new;
191	psw_t psw_save;
192
193	smp_emergency_stop();
194	diag_amode31_ops.diag308_reset();
195
196	/*
197	 * Disable low address protection and make machine check new PSW a
198	 * disabled wait PSW. Any additional machine check cannot be handled.
199	 */
200	local_ctl_store(0, &cr0.reg);
201	cr0_new = cr0;
202	cr0_new.lap = 0;
203	local_ctl_load(0, &cr0_new.reg);
204	psw_save = lc->mcck_new_psw;
205	psw_bits(lc->mcck_new_psw).io = 0;
206	psw_bits(lc->mcck_new_psw).ext = 0;
207	psw_bits(lc->mcck_new_psw).wait = 1;
208	nmi_print_info();
209
210	/*
211	 * Restore machine check new PSW and control register 0 to original
212	 * values. This makes possible system dump analysis easier.
213	 */
214	lc->mcck_new_psw = psw_save;
215	local_ctl_load(0, &cr0.reg);
216	disabled_wait();
217}
218NOKPROBE_SYMBOL(s390_handle_damage);
219
220/*
221 * Main machine check handler function. Will be called with interrupts disabled
222 * and machine checks enabled.
223 */
224void s390_handle_mcck(void)
225{
226	struct mcck_struct mcck;
227	unsigned long mflags;
228
229	/*
230	 * Disable machine checks and get the current state of accumulated
231	 * machine checks. Afterwards delete the old state and enable machine
232	 * checks again.
233	 */
234	local_mcck_save(mflags);
235	mcck = *this_cpu_ptr(&cpu_mcck);
236	memset(this_cpu_ptr(&cpu_mcck), 0, sizeof(mcck));
237	local_mcck_restore(mflags);
238
239	if (mcck.channel_report)
240		crw_handle_channel_report();
241	/*
242	 * A warning may remain for a prolonged period on the bare iron.
243	 * (actually until the machine is powered off, or the problem is gone)
244	 * So we just stop listening for the WARNING MCH and avoid continuously
245	 * being interrupted.  One caveat is however, that we must do this per
246	 * processor and cannot use the smp version of ctl_clear_bit().
247	 * On VM we only get one interrupt per virtally presented machinecheck.
248	 * Though one suffices, we may get one interrupt per (virtual) cpu.
249	 */
250	if (mcck.warning) {	/* WARNING pending ? */
251		static int mchchk_wng_posted = 0;
252
253		/* Use single cpu clear, as we cannot handle smp here. */
254		local_ctl_clear_bit(14, CR14_WARNING_SUBMASK_BIT);
255		if (xchg(&mchchk_wng_posted, 1) == 0)
256			kill_cad_pid(SIGPWR, 1);
257	}
258	if (mcck.stp_queue)
259		stp_queue_work();
260	if (mcck.kill_task) {
261		printk(KERN_EMERG "mcck: Terminating task because of machine "
262		       "malfunction (code 0x%016lx).\n", mcck.mcck_code);
263		printk(KERN_EMERG "mcck: task: %s, pid: %d.\n",
264		       current->comm, current->pid);
265		if (is_global_init(current))
266			panic("mcck: Attempting to kill init!\n");
267		do_send_sig_info(SIGKILL, SEND_SIG_PRIV, current, PIDTYPE_PID);
268	}
269}
270
271/**
272 * nmi_registers_valid - verify if registers are valid
273 * @mci: machine check interruption code
274 *
275 * Inspect a machine check interruption code and verify if all required
276 * registers are valid. For some registers the corresponding validity bit is
277 * ignored and the registers are set to the expected value.
278 * Returns true if all registers are valid, otherwise false.
279 */
280static bool notrace nmi_registers_valid(union mci mci)
281{
282	union ctlreg2 cr2;
283
284	/*
285	 * The getcpu vdso syscall reads the CPU number from the programmable
286	 * field of the TOD clock. Disregard the TOD programmable register
287	 * validity bit and load the CPU number into the TOD programmable field
288	 * unconditionally.
289	 */
290	set_tod_programmable_field(raw_smp_processor_id());
291	/*
292	 * Set the clock comparator register to the next expected value.
293	 */
294	set_clock_comparator(get_lowcore()->clock_comparator);
295	if (!mci.gr || !mci.fp || !mci.fc)
296		return false;
297	/*
298	 * The vector validity must only be checked if not running a
299	 * KVM guest. For KVM guests the machine check is forwarded by
300	 * KVM and it is the responsibility of the guest to take
301	 * appropriate actions. The host vector or FPU values have been
302	 * saved by KVM and will be restored by KVM.
303	 */
304	if (!mci.vr && !test_cpu_flag(CIF_MCCK_GUEST))
305		return false;
306	if (!mci.ar)
307		return false;
308	/*
309	 * Two cases for guarded storage registers:
310	 * - machine check in kernel or userspace
311	 * - machine check while running SIE (KVM guest)
312	 * For kernel or userspace the userspace values of guarded storage
313	 * control can not be recreated, the process must be terminated.
314	 * For SIE the guest values of guarded storage can not be recreated.
315	 * This is either due to a bug or due to GS being disabled in the
316	 * guest. The guest will be notified by KVM code and the guests machine
317	 * check handling must take care of this. The host values are saved by
318	 * KVM and are not affected.
319	 */
320	cr2.reg = get_lowcore()->cregs_save_area[2];
321	if (cr2.gse && !mci.gs && !test_cpu_flag(CIF_MCCK_GUEST))
322		return false;
323	if (!mci.ms || !mci.pm || !mci.ia)
324		return false;
325	return true;
326}
327NOKPROBE_SYMBOL(nmi_registers_valid);
328
329/*
330 * Backup the guest's machine check info to its description block
331 */
332static void notrace s390_backup_mcck_info(struct pt_regs *regs)
333{
334	struct mcck_volatile_info *mcck_backup;
335	struct sie_page *sie_page;
336
337	/* r14 contains the sie block, which was set in sie64a */
338	struct kvm_s390_sie_block *sie_block = phys_to_virt(regs->gprs[14]);
339
340	if (sie_block == NULL)
341		/* Something's seriously wrong, stop system. */
342		s390_handle_damage();
343
344	sie_page = container_of(sie_block, struct sie_page, sie_block);
345	mcck_backup = &sie_page->mcck_info;
346	mcck_backup->mcic = get_lowcore()->mcck_interruption_code &
347				~(MCCK_CODE_CP | MCCK_CODE_EXT_DAMAGE);
348	mcck_backup->ext_damage_code = get_lowcore()->external_damage_code;
349	mcck_backup->failing_storage_address = get_lowcore()->failing_storage_address;
350}
351NOKPROBE_SYMBOL(s390_backup_mcck_info);
352
353#define MAX_IPD_COUNT	29
354#define MAX_IPD_TIME	(5 * 60 * USEC_PER_SEC) /* 5 minutes */
355
356#define ED_STP_ISLAND	6	/* External damage STP island check */
357#define ED_STP_SYNC	7	/* External damage STP sync check */
358
359#define MCCK_CODE_NO_GUEST	(MCCK_CODE_CP | MCCK_CODE_EXT_DAMAGE)
360
361/*
362 * machine check handler.
363 */
364void notrace s390_do_machine_check(struct pt_regs *regs)
365{
366	static int ipd_count;
367	static DEFINE_SPINLOCK(ipd_lock);
368	static unsigned long long last_ipd;
369	struct lowcore *lc = get_lowcore();
370	struct mcck_struct *mcck;
371	unsigned long long tmp;
372	irqentry_state_t irq_state;
373	union mci mci;
374	unsigned long mcck_dam_code;
375	int mcck_pending = 0;
376
377	irq_state = irqentry_nmi_enter(regs);
378
379	if (user_mode(regs))
380		update_timer_mcck();
381	inc_irq_stat(NMI_NMI);
382	mci.val = lc->mcck_interruption_code;
383	mcck = this_cpu_ptr(&cpu_mcck);
384
385	/*
386	 * Reinject the instruction processing damages' machine checks
387	 * including Delayed Access Exception into the guest
388	 * instead of damaging the host if they happen in the guest.
389	 */
390	if (mci.pd && !test_cpu_flag(CIF_MCCK_GUEST)) {
391		if (mci.b) {
392			/* Processing backup -> verify if we can survive this */
393			u64 z_mcic, o_mcic, t_mcic;
394			z_mcic = (1ULL<<63 | 1ULL<<59 | 1ULL<<29);
395			o_mcic = (1ULL<<43 | 1ULL<<42 | 1ULL<<41 | 1ULL<<40 |
396				  1ULL<<36 | 1ULL<<35 | 1ULL<<34 | 1ULL<<32 |
397				  1ULL<<30 | 1ULL<<21 | 1ULL<<20 | 1ULL<<17 |
398				  1ULL<<16);
399			t_mcic = mci.val;
400
401			if (((t_mcic & z_mcic) != 0) ||
402			    ((t_mcic & o_mcic) != o_mcic)) {
403				s390_handle_damage();
404			}
405
406			/*
407			 * Nullifying exigent condition, therefore we might
408			 * retry this instruction.
409			 */
410			spin_lock(&ipd_lock);
411			tmp = get_tod_clock();
412			if (((tmp - last_ipd) >> 12) < MAX_IPD_TIME)
413				ipd_count++;
414			else
415				ipd_count = 1;
416			last_ipd = tmp;
417			if (ipd_count == MAX_IPD_COUNT)
418				s390_handle_damage();
419			spin_unlock(&ipd_lock);
420		} else {
421			/* Processing damage -> stopping machine */
422			s390_handle_damage();
423		}
424	}
425	if (!nmi_registers_valid(mci)) {
426		if (!user_mode(regs))
427			s390_handle_damage();
428		/*
429		 * Couldn't restore all register contents for the
430		 * user space process -> mark task for termination.
431		 */
432		mcck->kill_task = 1;
433		mcck->mcck_code = mci.val;
434		mcck_pending = 1;
435	}
436
437	/*
438	 * Backup the machine check's info if it happens when the guest
439	 * is running.
440	 */
441	if (test_cpu_flag(CIF_MCCK_GUEST))
442		s390_backup_mcck_info(regs);
443
444	if (mci.cd) {
445		/* Timing facility damage */
446		s390_handle_damage();
447	}
448	if (mci.ed && mci.ec) {
449		/* External damage */
450		if (lc->external_damage_code & (1U << ED_STP_SYNC))
451			mcck->stp_queue |= stp_sync_check();
452		if (lc->external_damage_code & (1U << ED_STP_ISLAND))
453			mcck->stp_queue |= stp_island_check();
454		mcck_pending = 1;
455	}
456	/*
457	 * Reinject storage related machine checks into the guest if they
458	 * happen when the guest is running.
459	 */
460	if (!test_cpu_flag(CIF_MCCK_GUEST)) {
461		/* Storage error uncorrected */
462		if (mci.se)
463			s390_handle_damage();
464		/* Storage key-error uncorrected */
465		if (mci.ke)
466			s390_handle_damage();
467		/* Storage degradation */
468		if (mci.ds && mci.fa)
469			s390_handle_damage();
470	}
471	if (mci.cp) {
472		/* Channel report word pending */
473		mcck->channel_report = 1;
474		mcck_pending = 1;
475	}
476	if (mci.w) {
477		/* Warning pending */
478		mcck->warning = 1;
479		mcck_pending = 1;
480	}
481
482	/*
483	 * If there are only Channel Report Pending and External Damage
484	 * machine checks, they will not be reinjected into the guest
485	 * because they refer to host conditions only.
486	 */
487	mcck_dam_code = (mci.val & MCIC_SUBCLASS_MASK);
488	if (test_cpu_flag(CIF_MCCK_GUEST) &&
489	(mcck_dam_code & MCCK_CODE_NO_GUEST) != mcck_dam_code) {
490		/* Set exit reason code for host's later handling */
491		*((long *)(regs->gprs[15] + __SF_SIE_REASON)) = -EINTR;
492	}
493	clear_cpu_flag(CIF_MCCK_GUEST);
494
495	if (mcck_pending)
496		schedule_mcck_handler();
497
498	irqentry_nmi_exit(regs, irq_state);
499}
500NOKPROBE_SYMBOL(s390_do_machine_check);
501
502static int __init machine_check_init(void)
503{
504	system_ctl_set_bit(14, CR14_EXTERNAL_DAMAGE_SUBMASK_BIT);
505	system_ctl_set_bit(14, CR14_RECOVERY_SUBMASK_BIT);
506	system_ctl_set_bit(14, CR14_WARNING_SUBMASK_BIT);
507	return 0;
508}
509early_initcall(machine_check_init);