1 // SPDX-License-Identifier: GPL-2.0
3 * Machine check handler
5 * Copyright IBM Corp. 2000, 2009
11 #include <linux/kernel_stat.h>
12 #include <linux/init.h>
13 #include <linux/errno.h>
14 #include <linux/entry-common.h>
15 #include <linux/hardirq.h>
16 #include <linux/log2.h>
17 #include <linux/kprobes.h>
18 #include <linux/kmemleak.h>
19 #include <linux/time.h>
20 #include <linux/module.h>
21 #include <linux/sched/signal.h>
22 #include <linux/kvm_host.h>
23 #include <linux/export.h>
24 #include <asm/lowcore.h>
27 #include <asm/cputime.h>
30 #include <asm/switch_to.h>
31 #include <asm/ctl_reg.h>
32 #include <asm/asm-offsets.h>
34 #include <asm/vx-insn.h>
37 unsigned int kill_task : 1;
38 unsigned int channel_report : 1;
39 unsigned int warning : 1;
40 unsigned int stp_queue : 1;
41 unsigned long mcck_code;
44 static DEFINE_PER_CPU(struct mcck_struct, cpu_mcck);
46 static inline int nmi_needs_mcesa(void)
48 return MACHINE_HAS_VX || MACHINE_HAS_GS;
52 * The initial machine check extended save area for the boot CPU.
53 * It will be replaced on the boot CPU reinit with an allocated
54 * structure. The structure is required for machine check happening
55 * early in the boot process.
57 static struct mcesa boot_mcesa __aligned(MCESA_MAX_SIZE);
59 void __init nmi_alloc_mcesa_early(u64 *mcesad)
61 if (!nmi_needs_mcesa())
63 *mcesad = __pa(&boot_mcesa);
65 *mcesad |= ilog2(MCESA_MAX_SIZE);
68 int nmi_alloc_mcesa(u64 *mcesad)
74 if (!nmi_needs_mcesa())
76 size = MACHINE_HAS_GS ? MCESA_MAX_SIZE : MCESA_MIN_SIZE;
77 origin = kmalloc(size, GFP_KERNEL);
80 /* The pointer is stored with mcesa_bits ORed in */
81 kmemleak_not_leak(origin);
82 *mcesad = __pa(origin);
84 *mcesad |= ilog2(MCESA_MAX_SIZE);
88 void nmi_free_mcesa(u64 *mcesad)
90 if (!nmi_needs_mcesa())
92 kfree(__va(*mcesad & MCESA_ORIGIN_MASK));
95 static __always_inline char *nmi_puts(char *dest, const char *src)
103 static __always_inline char *u64_to_hex(char *dest, u64 val)
107 for (i = 1; i <= 16; i++) {
108 num = (val >> (64 - 4 * i)) & 0xf;
110 *dest++ = 'A' + num - 10;
118 static notrace void s390_handle_damage(void)
120 union ctlreg0 cr0, cr0_new;
125 smp_emergency_stop();
126 diag_amode31_ops.diag308_reset();
127 ptr = nmi_puts(message, "System stopped due to unrecoverable machine check, code: 0x");
128 u64_to_hex(ptr, S390_lowcore.mcck_interruption_code);
131 * Disable low address protection and make machine check new PSW a
132 * disabled wait PSW. Any additional machine check cannot be handled.
134 __ctl_store(cr0.val, 0, 0);
137 __ctl_load(cr0_new.val, 0, 0);
138 psw_save = S390_lowcore.mcck_new_psw;
139 psw_bits(S390_lowcore.mcck_new_psw).io = 0;
140 psw_bits(S390_lowcore.mcck_new_psw).ext = 0;
141 psw_bits(S390_lowcore.mcck_new_psw).wait = 1;
142 sclp_emergency_printk(message);
145 * Restore machine check new PSW and control register 0 to original
146 * values. This makes possible system dump analysis easier.
148 S390_lowcore.mcck_new_psw = psw_save;
149 __ctl_load(cr0.val, 0, 0);
153 NOKPROBE_SYMBOL(s390_handle_damage);
156 * Main machine check handler function. Will be called with interrupts disabled
157 * and machine checks enabled.
159 void __s390_handle_mcck(void)
161 struct mcck_struct mcck;
164 * Disable machine checks and get the current state of accumulated
165 * machine checks. Afterwards delete the old state and enable machine
168 local_mcck_disable();
169 mcck = *this_cpu_ptr(&cpu_mcck);
170 memset(this_cpu_ptr(&cpu_mcck), 0, sizeof(mcck));
173 if (mcck.channel_report)
174 crw_handle_channel_report();
176 * A warning may remain for a prolonged period on the bare iron.
177 * (actually until the machine is powered off, or the problem is gone)
178 * So we just stop listening for the WARNING MCH and avoid continuously
179 * being interrupted. One caveat is however, that we must do this per
180 * processor and cannot use the smp version of ctl_clear_bit().
181 * On VM we only get one interrupt per virtally presented machinecheck.
182 * Though one suffices, we may get one interrupt per (virtual) cpu.
184 if (mcck.warning) { /* WARNING pending ? */
185 static int mchchk_wng_posted = 0;
187 /* Use single cpu clear, as we cannot handle smp here. */
188 __ctl_clear_bit(14, 24); /* Disable WARNING MCH */
189 if (xchg(&mchchk_wng_posted, 1) == 0)
190 kill_cad_pid(SIGPWR, 1);
194 if (mcck.kill_task) {
196 printk(KERN_EMERG "mcck: Terminating task because of machine "
197 "malfunction (code 0x%016lx).\n", mcck.mcck_code);
198 printk(KERN_EMERG "mcck: task: %s, pid: %d.\n",
199 current->comm, current->pid);
200 make_task_dead(SIGSEGV);
204 void noinstr s390_handle_mcck(struct pt_regs *regs)
206 trace_hardirqs_off();
207 pai_kernel_enter(regs);
208 __s390_handle_mcck();
209 pai_kernel_exit(regs);
213 * returns 0 if register contents could be validated
214 * returns 1 otherwise
216 static int notrace s390_validate_registers(union mci mci)
227 if (!mci.gr || !mci.fp)
229 fpt_save_area = &S390_lowcore.floating_pt_save_area;
240 : "Q" (S390_lowcore.fpt_creg_save_area));
243 mcesa = __va(S390_lowcore.mcesad & MCESA_ORIGIN_MASK);
244 if (!MACHINE_HAS_VX) {
245 /* Validate floating point registers */
264 : "a" (fpt_save_area)
267 /* Validate vector registers */
271 * The vector validity must only be checked if not running a
272 * KVM guest. For KVM guests the machine check is forwarded by
273 * KVM and it is the responsibility of the guest to take
274 * appropriate actions. The host vector or FPU values have been
275 * saved by KVM and will be restored by KVM.
277 if (!mci.vr && !test_cpu_flag(CIF_MCCK_GUEST))
279 cr0.val = S390_lowcore.cregs_save_area[0];
280 cr0.afp = cr0.vx = 1;
281 __ctl_load(cr0.val, 0, 0);
287 : "Q" (*(struct vx_array *)mcesa->vector_save_area)
289 __ctl_load(S390_lowcore.cregs_save_area[0], 0, 0);
291 /* Validate access registers */
295 : "a" (&S390_lowcore.access_regs_save_area)
299 /* Validate guarded storage registers */
300 cr2.val = S390_lowcore.cregs_save_area[2];
305 * - machine check in kernel or userspace
306 * - machine check while running SIE (KVM guest)
307 * For kernel or userspace the userspace values of
308 * guarded storage control can not be recreated, the
309 * process must be terminated.
310 * For SIE the guest values of guarded storage can not
311 * be recreated. This is either due to a bug or due to
312 * GS being disabled in the guest. The guest will be
313 * notified by KVM code and the guests machine check
314 * handling must take care of this. The host values
315 * are saved by KVM and are not affected.
317 if (!test_cpu_flag(CIF_MCCK_GUEST))
320 load_gs_cb((struct gs_cb *)mcesa->guarded_storage_save_area);
324 * The getcpu vdso syscall reads CPU number from the programmable
325 * field of the TOD clock. Disregard the TOD programmable register
326 * validity bit and load the CPU number into the TOD programmable
327 * field unconditionally.
329 set_tod_programmable_field(raw_smp_processor_id());
330 /* Validate clock comparator register */
331 set_clock_comparator(S390_lowcore.clock_comparator);
333 if (!mci.ms || !mci.pm || !mci.ia)
338 NOKPROBE_SYMBOL(s390_validate_registers);
341 * Backup the guest's machine check info to its description block
343 static void notrace s390_backup_mcck_info(struct pt_regs *regs)
345 struct mcck_volatile_info *mcck_backup;
346 struct sie_page *sie_page;
348 /* r14 contains the sie block, which was set in sie64a */
349 struct kvm_s390_sie_block *sie_block =
350 (struct kvm_s390_sie_block *) regs->gprs[14];
352 if (sie_block == NULL)
353 /* Something's seriously wrong, stop system. */
354 s390_handle_damage();
356 sie_page = container_of(sie_block, struct sie_page, sie_block);
357 mcck_backup = &sie_page->mcck_info;
358 mcck_backup->mcic = S390_lowcore.mcck_interruption_code &
359 ~(MCCK_CODE_CP | MCCK_CODE_EXT_DAMAGE);
360 mcck_backup->ext_damage_code = S390_lowcore.external_damage_code;
361 mcck_backup->failing_storage_address
362 = S390_lowcore.failing_storage_address;
364 NOKPROBE_SYMBOL(s390_backup_mcck_info);
366 #define MAX_IPD_COUNT 29
367 #define MAX_IPD_TIME (5 * 60 * USEC_PER_SEC) /* 5 minutes */
369 #define ED_STP_ISLAND 6 /* External damage STP island check */
370 #define ED_STP_SYNC 7 /* External damage STP sync check */
372 #define MCCK_CODE_NO_GUEST (MCCK_CODE_CP | MCCK_CODE_EXT_DAMAGE)
375 * machine check handler.
377 int notrace s390_do_machine_check(struct pt_regs *regs)
379 static int ipd_count;
380 static DEFINE_SPINLOCK(ipd_lock);
381 static unsigned long long last_ipd;
382 struct mcck_struct *mcck;
383 unsigned long long tmp;
384 irqentry_state_t irq_state;
386 unsigned long mcck_dam_code;
387 int mcck_pending = 0;
389 irq_state = irqentry_nmi_enter(regs);
393 inc_irq_stat(NMI_NMI);
394 mci.val = S390_lowcore.mcck_interruption_code;
395 mcck = this_cpu_ptr(&cpu_mcck);
398 * Reinject the instruction processing damages' machine checks
399 * including Delayed Access Exception into the guest
400 * instead of damaging the host if they happen in the guest.
402 if (mci.pd && !test_cpu_flag(CIF_MCCK_GUEST)) {
404 /* Processing backup -> verify if we can survive this */
405 u64 z_mcic, o_mcic, t_mcic;
406 z_mcic = (1ULL<<63 | 1ULL<<59 | 1ULL<<29);
407 o_mcic = (1ULL<<43 | 1ULL<<42 | 1ULL<<41 | 1ULL<<40 |
408 1ULL<<36 | 1ULL<<35 | 1ULL<<34 | 1ULL<<32 |
409 1ULL<<30 | 1ULL<<21 | 1ULL<<20 | 1ULL<<17 |
413 if (((t_mcic & z_mcic) != 0) ||
414 ((t_mcic & o_mcic) != o_mcic)) {
415 s390_handle_damage();
419 * Nullifying exigent condition, therefore we might
420 * retry this instruction.
422 spin_lock(&ipd_lock);
423 tmp = get_tod_clock();
424 if (((tmp - last_ipd) >> 12) < MAX_IPD_TIME)
429 if (ipd_count == MAX_IPD_COUNT)
430 s390_handle_damage();
431 spin_unlock(&ipd_lock);
433 /* Processing damage -> stopping machine */
434 s390_handle_damage();
437 if (s390_validate_registers(mci)) {
438 if (!user_mode(regs))
439 s390_handle_damage();
441 * Couldn't restore all register contents for the
442 * user space process -> mark task for termination.
445 mcck->mcck_code = mci.val;
450 * Backup the machine check's info if it happens when the guest
453 if (test_cpu_flag(CIF_MCCK_GUEST))
454 s390_backup_mcck_info(regs);
457 /* Timing facility damage */
458 s390_handle_damage();
460 if (mci.ed && mci.ec) {
461 /* External damage */
462 if (S390_lowcore.external_damage_code & (1U << ED_STP_SYNC))
463 mcck->stp_queue |= stp_sync_check();
464 if (S390_lowcore.external_damage_code & (1U << ED_STP_ISLAND))
465 mcck->stp_queue |= stp_island_check();
469 * Reinject storage related machine checks into the guest if they
470 * happen when the guest is running.
472 if (!test_cpu_flag(CIF_MCCK_GUEST)) {
473 /* Storage error uncorrected */
475 s390_handle_damage();
476 /* Storage key-error uncorrected */
478 s390_handle_damage();
479 /* Storage degradation */
480 if (mci.ds && mci.fa)
481 s390_handle_damage();
484 /* Channel report word pending */
485 mcck->channel_report = 1;
489 /* Warning pending */
495 * If there are only Channel Report Pending and External Damage
496 * machine checks, they will not be reinjected into the guest
497 * because they refer to host conditions only.
499 mcck_dam_code = (mci.val & MCIC_SUBCLASS_MASK);
500 if (test_cpu_flag(CIF_MCCK_GUEST) &&
501 (mcck_dam_code & MCCK_CODE_NO_GUEST) != mcck_dam_code) {
502 /* Set exit reason code for host's later handling */
503 *((long *)(regs->gprs[15] + __SF_SIE_REASON)) = -EINTR;
505 clear_cpu_flag(CIF_MCCK_GUEST);
507 if (user_mode(regs) && mcck_pending) {
508 irqentry_nmi_exit(regs, irq_state);
513 schedule_mcck_handler();
515 irqentry_nmi_exit(regs, irq_state);
518 NOKPROBE_SYMBOL(s390_do_machine_check);
520 static int __init machine_check_init(void)
522 ctl_set_bit(14, 25); /* enable external damage MCH */
523 ctl_set_bit(14, 27); /* enable system recovery MCH */
524 ctl_set_bit(14, 24); /* enable warning MCH */
527 early_initcall(machine_check_init);