1 // SPDX-License-Identifier: GPL-2.0
3 * Machine check handler
5 * Copyright IBM Corp. 2000, 2009
12 #include <linux/kernel_stat.h>
13 #include <linux/init.h>
14 #include <linux/errno.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>
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>
33 #include <linux/kvm_host.h>
36 unsigned int kill_task : 1;
37 unsigned int channel_report : 1;
38 unsigned int warning : 1;
39 unsigned int stp_queue : 1;
40 unsigned long mcck_code;
43 static DEFINE_PER_CPU(struct mcck_struct, cpu_mcck);
44 static struct kmem_cache *mcesa_cache;
45 static unsigned long mcesa_origin_lc;
47 static inline int nmi_needs_mcesa(void)
49 return MACHINE_HAS_VX || MACHINE_HAS_GS;
52 static inline unsigned long nmi_get_mcesa_size(void)
55 return MCESA_MAX_SIZE;
56 return MCESA_MIN_SIZE;
60 * The initial machine check extended save area for the boot CPU.
61 * It will be replaced by nmi_init() with an allocated structure.
62 * The structure is required for machine check happening early in
65 static struct mcesa boot_mcesa __initdata __aligned(MCESA_MAX_SIZE);
67 void __init nmi_alloc_boot_cpu(struct lowcore *lc)
69 if (!nmi_needs_mcesa())
71 lc->mcesad = (unsigned long) &boot_mcesa;
73 lc->mcesad |= ilog2(MCESA_MAX_SIZE);
76 static int __init nmi_init(void)
78 unsigned long origin, cr0, size;
80 if (!nmi_needs_mcesa())
82 size = nmi_get_mcesa_size();
83 if (size > MCESA_MIN_SIZE)
84 mcesa_origin_lc = ilog2(size);
85 /* create slab cache for the machine-check-extended-save-areas */
86 mcesa_cache = kmem_cache_create("nmi_save_areas", size, size, 0, NULL);
88 panic("Couldn't create nmi save area cache");
89 origin = (unsigned long) kmem_cache_alloc(mcesa_cache, GFP_KERNEL);
91 panic("Couldn't allocate nmi save area");
92 /* The pointer is stored with mcesa_bits ORed in */
93 kmemleak_not_leak((void *) origin);
94 __ctl_store(cr0, 0, 0);
95 __ctl_clear_bit(0, 28); /* disable lowcore protection */
96 /* Replace boot_mcesa on the boot CPU */
97 S390_lowcore.mcesad = origin | mcesa_origin_lc;
98 __ctl_load(cr0, 0, 0);
101 early_initcall(nmi_init);
103 int nmi_alloc_per_cpu(struct lowcore *lc)
105 unsigned long origin;
107 if (!nmi_needs_mcesa())
109 origin = (unsigned long) kmem_cache_alloc(mcesa_cache, GFP_KERNEL);
112 /* The pointer is stored with mcesa_bits ORed in */
113 kmemleak_not_leak((void *) origin);
114 lc->mcesad = origin | mcesa_origin_lc;
118 void nmi_free_per_cpu(struct lowcore *lc)
120 if (!nmi_needs_mcesa())
122 kmem_cache_free(mcesa_cache, (void *)(lc->mcesad & MCESA_ORIGIN_MASK));
125 static notrace void s390_handle_damage(void)
127 smp_emergency_stop();
131 NOKPROBE_SYMBOL(s390_handle_damage);
134 * Main machine check handler function. Will be called with interrupts disabled
135 * and machine checks enabled.
137 void __s390_handle_mcck(void)
139 struct mcck_struct mcck;
142 * Disable machine checks and get the current state of accumulated
143 * machine checks. Afterwards delete the old state and enable machine
146 local_mcck_disable();
147 mcck = *this_cpu_ptr(&cpu_mcck);
148 memset(this_cpu_ptr(&cpu_mcck), 0, sizeof(mcck));
151 if (mcck.channel_report)
152 crw_handle_channel_report();
154 * A warning may remain for a prolonged period on the bare iron.
155 * (actually until the machine is powered off, or the problem is gone)
156 * So we just stop listening for the WARNING MCH and avoid continuously
157 * being interrupted. One caveat is however, that we must do this per
158 * processor and cannot use the smp version of ctl_clear_bit().
159 * On VM we only get one interrupt per virtally presented machinecheck.
160 * Though one suffices, we may get one interrupt per (virtual) cpu.
162 if (mcck.warning) { /* WARNING pending ? */
163 static int mchchk_wng_posted = 0;
165 /* Use single cpu clear, as we cannot handle smp here. */
166 __ctl_clear_bit(14, 24); /* Disable WARNING MCH */
167 if (xchg(&mchchk_wng_posted, 1) == 0)
168 kill_cad_pid(SIGPWR, 1);
172 if (mcck.kill_task) {
174 printk(KERN_EMERG "mcck: Terminating task because of machine "
175 "malfunction (code 0x%016lx).\n", mcck.mcck_code);
176 printk(KERN_EMERG "mcck: task: %s, pid: %d.\n",
177 current->comm, current->pid);
182 void noinstr s390_handle_mcck(void)
184 trace_hardirqs_off();
185 __s390_handle_mcck();
189 * returns 0 if all required registers are available
190 * returns 1 otherwise
192 static int notrace s390_validate_registers(union mci mci, int umode)
205 * General purpose registers couldn't be restored and have
206 * unknown contents. Stop system or terminate process.
209 s390_handle_damage();
214 * Floating point registers can't be restored. If the
215 * kernel currently uses floating point registers the
216 * system is stopped. If the process has its floating
217 * pointer registers loaded it is terminated.
219 if (S390_lowcore.fpu_flags & KERNEL_VXR_V0V7)
220 s390_handle_damage();
221 if (!test_cpu_flag(CIF_FPU))
224 fpt_save_area = &S390_lowcore.floating_pt_save_area;
227 * Floating point control register can't be restored.
228 * If the kernel currently uses the floating pointer
229 * registers and needs the FPC register the system is
230 * stopped. If the process has its floating pointer
231 * registers loaded it is terminated. Otherwise the
232 * FPC is just validated.
234 if (S390_lowcore.fpu_flags & KERNEL_FPC)
235 s390_handle_damage();
240 if (!test_cpu_flag(CIF_FPU))
246 : "Q" (S390_lowcore.fpt_creg_save_area));
249 mcesa = (struct mcesa *)(S390_lowcore.mcesad & MCESA_ORIGIN_MASK);
250 if (!MACHINE_HAS_VX) {
251 /* Validate floating point registers */
270 : "a" (fpt_save_area)
273 /* Validate vector registers */
278 * Vector registers can't be restored. If the kernel
279 * currently uses vector registers the system is
280 * stopped. If the process has its vector registers
281 * loaded it is terminated. Otherwise just validate
284 if (S390_lowcore.fpu_flags & KERNEL_VXR)
285 s390_handle_damage();
286 if (!test_cpu_flag(CIF_FPU))
289 cr0.val = S390_lowcore.cregs_save_area[0];
290 cr0.afp = cr0.vx = 1;
291 __ctl_load(cr0.val, 0, 0);
294 " .word 0xe70f,0x1000,0x0036\n" /* vlm 0,15,0(1) */
295 " .word 0xe70f,0x1100,0x0c36\n" /* vlm 16,31,256(1) */
297 : "Q" (*(struct vx_array *)mcesa->vector_save_area)
299 __ctl_load(S390_lowcore.cregs_save_area[0], 0, 0);
301 /* Validate access registers */
305 : "a" (&S390_lowcore.access_regs_save_area)
309 * Access registers have unknown contents.
314 /* Validate guarded storage registers */
315 cr2.val = S390_lowcore.cregs_save_area[2];
319 * Guarded storage register can't be restored and
320 * the current processes uses guarded storage.
321 * It has to be terminated.
325 load_gs_cb((struct gs_cb *)mcesa->guarded_storage_save_area);
329 * The getcpu vdso syscall reads CPU number from the programmable
330 * field of the TOD clock. Disregard the TOD programmable register
331 * validity bit and load the CPU number into the TOD programmable
332 * field unconditionally.
334 set_tod_programmable_field(raw_smp_processor_id());
335 /* Validate clock comparator register */
336 set_clock_comparator(S390_lowcore.clock_comparator);
338 if (!mci.ms || !mci.pm || !mci.ia)
343 NOKPROBE_SYMBOL(s390_validate_registers);
346 * Backup the guest's machine check info to its description block
348 static void notrace s390_backup_mcck_info(struct pt_regs *regs)
350 struct mcck_volatile_info *mcck_backup;
351 struct sie_page *sie_page;
353 /* r14 contains the sie block, which was set in sie64a */
354 struct kvm_s390_sie_block *sie_block =
355 (struct kvm_s390_sie_block *) regs->gprs[14];
357 if (sie_block == NULL)
358 /* Something's seriously wrong, stop system. */
359 s390_handle_damage();
361 sie_page = container_of(sie_block, struct sie_page, sie_block);
362 mcck_backup = &sie_page->mcck_info;
363 mcck_backup->mcic = S390_lowcore.mcck_interruption_code &
364 ~(MCCK_CODE_CP | MCCK_CODE_EXT_DAMAGE);
365 mcck_backup->ext_damage_code = S390_lowcore.external_damage_code;
366 mcck_backup->failing_storage_address
367 = S390_lowcore.failing_storage_address;
369 NOKPROBE_SYMBOL(s390_backup_mcck_info);
371 #define MAX_IPD_COUNT 29
372 #define MAX_IPD_TIME (5 * 60 * USEC_PER_SEC) /* 5 minutes */
374 #define ED_STP_ISLAND 6 /* External damage STP island check */
375 #define ED_STP_SYNC 7 /* External damage STP sync check */
377 #define MCCK_CODE_NO_GUEST (MCCK_CODE_CP | MCCK_CODE_EXT_DAMAGE)
380 * machine check handler.
382 int notrace s390_do_machine_check(struct pt_regs *regs)
384 static int ipd_count;
385 static DEFINE_SPINLOCK(ipd_lock);
386 static unsigned long long last_ipd;
387 struct mcck_struct *mcck;
388 unsigned long long tmp;
390 unsigned long mcck_dam_code;
391 int mcck_pending = 0;
397 inc_irq_stat(NMI_NMI);
398 mci.val = S390_lowcore.mcck_interruption_code;
399 mcck = this_cpu_ptr(&cpu_mcck);
402 * Reinject the instruction processing damages' machine checks
403 * including Delayed Access Exception into the guest
404 * instead of damaging the host if they happen in the guest.
406 if (mci.pd && !test_cpu_flag(CIF_MCCK_GUEST)) {
408 /* Processing backup -> verify if we can survive this */
409 u64 z_mcic, o_mcic, t_mcic;
410 z_mcic = (1ULL<<63 | 1ULL<<59 | 1ULL<<29);
411 o_mcic = (1ULL<<43 | 1ULL<<42 | 1ULL<<41 | 1ULL<<40 |
412 1ULL<<36 | 1ULL<<35 | 1ULL<<34 | 1ULL<<32 |
413 1ULL<<30 | 1ULL<<21 | 1ULL<<20 | 1ULL<<17 |
417 if (((t_mcic & z_mcic) != 0) ||
418 ((t_mcic & o_mcic) != o_mcic)) {
419 s390_handle_damage();
423 * Nullifying exigent condition, therefore we might
424 * retry this instruction.
426 spin_lock(&ipd_lock);
427 tmp = get_tod_clock();
428 if (((tmp - last_ipd) >> 12) < MAX_IPD_TIME)
433 if (ipd_count == MAX_IPD_COUNT)
434 s390_handle_damage();
435 spin_unlock(&ipd_lock);
437 /* Processing damage -> stopping machine */
438 s390_handle_damage();
441 if (s390_validate_registers(mci, user_mode(regs))) {
443 * Couldn't restore all register contents for the
444 * user space process -> mark task for termination.
447 mcck->mcck_code = mci.val;
452 * Backup the machine check's info if it happens when the guest
455 if (test_cpu_flag(CIF_MCCK_GUEST))
456 s390_backup_mcck_info(regs);
459 /* Timing facility damage */
460 s390_handle_damage();
462 if (mci.ed && mci.ec) {
463 /* External damage */
464 if (S390_lowcore.external_damage_code & (1U << ED_STP_SYNC))
465 mcck->stp_queue |= stp_sync_check();
466 if (S390_lowcore.external_damage_code & (1U << ED_STP_ISLAND))
467 mcck->stp_queue |= stp_island_check();
472 /* Channel report word pending */
473 mcck->channel_report = 1;
477 /* Warning pending */
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.
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;
493 clear_cpu_flag(CIF_MCCK_GUEST);
495 if (user_mode(regs) && mcck_pending) {
501 schedule_mcck_handler();
506 NOKPROBE_SYMBOL(s390_do_machine_check);
508 static int __init machine_check_init(void)
510 ctl_set_bit(14, 25); /* enable external damage MCH */
511 ctl_set_bit(14, 27); /* enable system recovery MCH */
512 ctl_set_bit(14, 24); /* enable warning MCH */
515 early_initcall(machine_check_init);