Loading...
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/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>
25#include <asm/ctlreg.h>
26#include <asm/smp.h>
27#include <asm/stp.h>
28#include <asm/cputime.h>
29#include <asm/nmi.h>
30#include <asm/crw.h>
31#include <asm/switch_to.h>
32#include <asm/asm-offsets.h>
33#include <asm/pai.h>
34#include <asm/vx-insn.h>
35#include <asm/fpu/api.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() || MACHINE_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 (MACHINE_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 = MACHINE_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 (MACHINE_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 s390_handle_damage(void)
120{
121 union ctlreg0 cr0, cr0_new;
122 char message[100];
123 psw_t psw_save;
124 char *ptr;
125
126 smp_emergency_stop();
127 diag_amode31_ops.diag308_reset();
128 ptr = nmi_puts(message, "System stopped due to unrecoverable machine check, code: 0x");
129 u64_to_hex(ptr, S390_lowcore.mcck_interruption_code);
130
131 /*
132 * Disable low address protection and make machine check new PSW a
133 * disabled wait PSW. Any additional machine check cannot be handled.
134 */
135 local_ctl_store(0, &cr0.reg);
136 cr0_new = cr0;
137 cr0_new.lap = 0;
138 local_ctl_load(0, &cr0_new.reg);
139 psw_save = S390_lowcore.mcck_new_psw;
140 psw_bits(S390_lowcore.mcck_new_psw).io = 0;
141 psw_bits(S390_lowcore.mcck_new_psw).ext = 0;
142 psw_bits(S390_lowcore.mcck_new_psw).wait = 1;
143 sclp_emergency_printk(message);
144
145 /*
146 * Restore machine check new PSW and control register 0 to original
147 * values. This makes possible system dump analysis easier.
148 */
149 S390_lowcore.mcck_new_psw = psw_save;
150 local_ctl_load(0, &cr0.reg);
151 disabled_wait();
152 while (1);
153}
154NOKPROBE_SYMBOL(s390_handle_damage);
155
156/*
157 * Main machine check handler function. Will be called with interrupts disabled
158 * and machine checks enabled.
159 */
160void s390_handle_mcck(void)
161{
162 struct mcck_struct mcck;
163 unsigned long mflags;
164
165 /*
166 * Disable machine checks and get the current state of accumulated
167 * machine checks. Afterwards delete the old state and enable machine
168 * checks again.
169 */
170 local_mcck_save(mflags);
171 mcck = *this_cpu_ptr(&cpu_mcck);
172 memset(this_cpu_ptr(&cpu_mcck), 0, sizeof(mcck));
173 local_mcck_restore(mflags);
174
175 if (mcck.channel_report)
176 crw_handle_channel_report();
177 /*
178 * A warning may remain for a prolonged period on the bare iron.
179 * (actually until the machine is powered off, or the problem is gone)
180 * So we just stop listening for the WARNING MCH and avoid continuously
181 * being interrupted. One caveat is however, that we must do this per
182 * processor and cannot use the smp version of ctl_clear_bit().
183 * On VM we only get one interrupt per virtally presented machinecheck.
184 * Though one suffices, we may get one interrupt per (virtual) cpu.
185 */
186 if (mcck.warning) { /* WARNING pending ? */
187 static int mchchk_wng_posted = 0;
188
189 /* Use single cpu clear, as we cannot handle smp here. */
190 local_ctl_clear_bit(14, CR14_WARNING_SUBMASK_BIT);
191 if (xchg(&mchchk_wng_posted, 1) == 0)
192 kill_cad_pid(SIGPWR, 1);
193 }
194 if (mcck.stp_queue)
195 stp_queue_work();
196 if (mcck.kill_task) {
197 printk(KERN_EMERG "mcck: Terminating task because of machine "
198 "malfunction (code 0x%016lx).\n", mcck.mcck_code);
199 printk(KERN_EMERG "mcck: task: %s, pid: %d.\n",
200 current->comm, current->pid);
201 if (is_global_init(current))
202 panic("mcck: Attempting to kill init!\n");
203 do_send_sig_info(SIGKILL, SEND_SIG_PRIV, current, PIDTYPE_PID);
204 }
205}
206
207/*
208 * returns 0 if register contents could be validated
209 * returns 1 otherwise
210 */
211static int notrace s390_validate_registers(union mci mci)
212{
213 struct mcesa *mcesa;
214 void *fpt_save_area;
215 union ctlreg2 cr2;
216 int kill_task;
217 u64 zero;
218
219 kill_task = 0;
220 zero = 0;
221
222 if (!mci.gr || !mci.fp)
223 kill_task = 1;
224 fpt_save_area = &S390_lowcore.floating_pt_save_area;
225 if (!mci.fc) {
226 kill_task = 1;
227 asm volatile(
228 " lfpc %0\n"
229 :
230 : "Q" (zero));
231 } else {
232 asm volatile(
233 " lfpc %0\n"
234 :
235 : "Q" (S390_lowcore.fpt_creg_save_area));
236 }
237
238 mcesa = __va(S390_lowcore.mcesad & MCESA_ORIGIN_MASK);
239 if (!cpu_has_vx()) {
240 /* Validate floating point registers */
241 asm volatile(
242 " ld 0,0(%0)\n"
243 " ld 1,8(%0)\n"
244 " ld 2,16(%0)\n"
245 " ld 3,24(%0)\n"
246 " ld 4,32(%0)\n"
247 " ld 5,40(%0)\n"
248 " ld 6,48(%0)\n"
249 " ld 7,56(%0)\n"
250 " ld 8,64(%0)\n"
251 " ld 9,72(%0)\n"
252 " ld 10,80(%0)\n"
253 " ld 11,88(%0)\n"
254 " ld 12,96(%0)\n"
255 " ld 13,104(%0)\n"
256 " ld 14,112(%0)\n"
257 " ld 15,120(%0)\n"
258 :
259 : "a" (fpt_save_area)
260 : "memory");
261 } else {
262 /* Validate vector registers */
263 union ctlreg0 cr0;
264
265 /*
266 * The vector validity must only be checked if not running a
267 * KVM guest. For KVM guests the machine check is forwarded by
268 * KVM and it is the responsibility of the guest to take
269 * appropriate actions. The host vector or FPU values have been
270 * saved by KVM and will be restored by KVM.
271 */
272 if (!mci.vr && !test_cpu_flag(CIF_MCCK_GUEST))
273 kill_task = 1;
274 cr0.reg = S390_lowcore.cregs_save_area[0];
275 cr0.afp = cr0.vx = 1;
276 local_ctl_load(0, &cr0.reg);
277 asm volatile(
278 " la 1,%0\n"
279 " VLM 0,15,0,1\n"
280 " VLM 16,31,256,1\n"
281 :
282 : "Q" (*(struct vx_array *)mcesa->vector_save_area)
283 : "1");
284 local_ctl_load(0, &S390_lowcore.cregs_save_area[0]);
285 }
286 /* Validate access registers */
287 asm volatile(
288 " lam 0,15,0(%0)\n"
289 :
290 : "a" (&S390_lowcore.access_regs_save_area)
291 : "memory");
292 if (!mci.ar)
293 kill_task = 1;
294 /* Validate guarded storage registers */
295 cr2.reg = S390_lowcore.cregs_save_area[2];
296 if (cr2.gse) {
297 if (!mci.gs) {
298 /*
299 * 2 cases:
300 * - machine check in kernel or userspace
301 * - machine check while running SIE (KVM guest)
302 * For kernel or userspace the userspace values of
303 * guarded storage control can not be recreated, the
304 * process must be terminated.
305 * For SIE the guest values of guarded storage can not
306 * be recreated. This is either due to a bug or due to
307 * GS being disabled in the guest. The guest will be
308 * notified by KVM code and the guests machine check
309 * handling must take care of this. The host values
310 * are saved by KVM and are not affected.
311 */
312 if (!test_cpu_flag(CIF_MCCK_GUEST))
313 kill_task = 1;
314 } else {
315 load_gs_cb((struct gs_cb *)mcesa->guarded_storage_save_area);
316 }
317 }
318 /*
319 * The getcpu vdso syscall reads CPU number from the programmable
320 * field of the TOD clock. Disregard the TOD programmable register
321 * validity bit and load the CPU number into the TOD programmable
322 * field unconditionally.
323 */
324 set_tod_programmable_field(raw_smp_processor_id());
325 /* Validate clock comparator register */
326 set_clock_comparator(S390_lowcore.clock_comparator);
327
328 if (!mci.ms || !mci.pm || !mci.ia)
329 kill_task = 1;
330
331 return kill_task;
332}
333NOKPROBE_SYMBOL(s390_validate_registers);
334
335/*
336 * Backup the guest's machine check info to its description block
337 */
338static void notrace s390_backup_mcck_info(struct pt_regs *regs)
339{
340 struct mcck_volatile_info *mcck_backup;
341 struct sie_page *sie_page;
342
343 /* r14 contains the sie block, which was set in sie64a */
344 struct kvm_s390_sie_block *sie_block = phys_to_virt(regs->gprs[14]);
345
346 if (sie_block == NULL)
347 /* Something's seriously wrong, stop system. */
348 s390_handle_damage();
349
350 sie_page = container_of(sie_block, struct sie_page, sie_block);
351 mcck_backup = &sie_page->mcck_info;
352 mcck_backup->mcic = S390_lowcore.mcck_interruption_code &
353 ~(MCCK_CODE_CP | MCCK_CODE_EXT_DAMAGE);
354 mcck_backup->ext_damage_code = S390_lowcore.external_damage_code;
355 mcck_backup->failing_storage_address
356 = S390_lowcore.failing_storage_address;
357}
358NOKPROBE_SYMBOL(s390_backup_mcck_info);
359
360#define MAX_IPD_COUNT 29
361#define MAX_IPD_TIME (5 * 60 * USEC_PER_SEC) /* 5 minutes */
362
363#define ED_STP_ISLAND 6 /* External damage STP island check */
364#define ED_STP_SYNC 7 /* External damage STP sync check */
365
366#define MCCK_CODE_NO_GUEST (MCCK_CODE_CP | MCCK_CODE_EXT_DAMAGE)
367
368/*
369 * machine check handler.
370 */
371void notrace s390_do_machine_check(struct pt_regs *regs)
372{
373 static int ipd_count;
374 static DEFINE_SPINLOCK(ipd_lock);
375 static unsigned long long last_ipd;
376 struct mcck_struct *mcck;
377 unsigned long long tmp;
378 irqentry_state_t irq_state;
379 union mci mci;
380 unsigned long mcck_dam_code;
381 int mcck_pending = 0;
382
383 irq_state = irqentry_nmi_enter(regs);
384
385 if (user_mode(regs))
386 update_timer_mcck();
387 inc_irq_stat(NMI_NMI);
388 mci.val = S390_lowcore.mcck_interruption_code;
389 mcck = this_cpu_ptr(&cpu_mcck);
390
391 /*
392 * Reinject the instruction processing damages' machine checks
393 * including Delayed Access Exception into the guest
394 * instead of damaging the host if they happen in the guest.
395 */
396 if (mci.pd && !test_cpu_flag(CIF_MCCK_GUEST)) {
397 if (mci.b) {
398 /* Processing backup -> verify if we can survive this */
399 u64 z_mcic, o_mcic, t_mcic;
400 z_mcic = (1ULL<<63 | 1ULL<<59 | 1ULL<<29);
401 o_mcic = (1ULL<<43 | 1ULL<<42 | 1ULL<<41 | 1ULL<<40 |
402 1ULL<<36 | 1ULL<<35 | 1ULL<<34 | 1ULL<<32 |
403 1ULL<<30 | 1ULL<<21 | 1ULL<<20 | 1ULL<<17 |
404 1ULL<<16);
405 t_mcic = mci.val;
406
407 if (((t_mcic & z_mcic) != 0) ||
408 ((t_mcic & o_mcic) != o_mcic)) {
409 s390_handle_damage();
410 }
411
412 /*
413 * Nullifying exigent condition, therefore we might
414 * retry this instruction.
415 */
416 spin_lock(&ipd_lock);
417 tmp = get_tod_clock();
418 if (((tmp - last_ipd) >> 12) < MAX_IPD_TIME)
419 ipd_count++;
420 else
421 ipd_count = 1;
422 last_ipd = tmp;
423 if (ipd_count == MAX_IPD_COUNT)
424 s390_handle_damage();
425 spin_unlock(&ipd_lock);
426 } else {
427 /* Processing damage -> stopping machine */
428 s390_handle_damage();
429 }
430 }
431 if (s390_validate_registers(mci)) {
432 if (!user_mode(regs))
433 s390_handle_damage();
434 /*
435 * Couldn't restore all register contents for the
436 * user space process -> mark task for termination.
437 */
438 mcck->kill_task = 1;
439 mcck->mcck_code = mci.val;
440 mcck_pending = 1;
441 }
442
443 /*
444 * Backup the machine check's info if it happens when the guest
445 * is running.
446 */
447 if (test_cpu_flag(CIF_MCCK_GUEST))
448 s390_backup_mcck_info(regs);
449
450 if (mci.cd) {
451 /* Timing facility damage */
452 s390_handle_damage();
453 }
454 if (mci.ed && mci.ec) {
455 /* External damage */
456 if (S390_lowcore.external_damage_code & (1U << ED_STP_SYNC))
457 mcck->stp_queue |= stp_sync_check();
458 if (S390_lowcore.external_damage_code & (1U << ED_STP_ISLAND))
459 mcck->stp_queue |= stp_island_check();
460 mcck_pending = 1;
461 }
462 /*
463 * Reinject storage related machine checks into the guest if they
464 * happen when the guest is running.
465 */
466 if (!test_cpu_flag(CIF_MCCK_GUEST)) {
467 /* Storage error uncorrected */
468 if (mci.se)
469 s390_handle_damage();
470 /* Storage key-error uncorrected */
471 if (mci.ke)
472 s390_handle_damage();
473 /* Storage degradation */
474 if (mci.ds && mci.fa)
475 s390_handle_damage();
476 }
477 if (mci.cp) {
478 /* Channel report word pending */
479 mcck->channel_report = 1;
480 mcck_pending = 1;
481 }
482 if (mci.w) {
483 /* Warning pending */
484 mcck->warning = 1;
485 mcck_pending = 1;
486 }
487
488 /*
489 * If there are only Channel Report Pending and External Damage
490 * machine checks, they will not be reinjected into the guest
491 * because they refer to host conditions only.
492 */
493 mcck_dam_code = (mci.val & MCIC_SUBCLASS_MASK);
494 if (test_cpu_flag(CIF_MCCK_GUEST) &&
495 (mcck_dam_code & MCCK_CODE_NO_GUEST) != mcck_dam_code) {
496 /* Set exit reason code for host's later handling */
497 *((long *)(regs->gprs[15] + __SF_SIE_REASON)) = -EINTR;
498 }
499 clear_cpu_flag(CIF_MCCK_GUEST);
500
501 if (mcck_pending)
502 schedule_mcck_handler();
503
504 irqentry_nmi_exit(regs, irq_state);
505}
506NOKPROBE_SYMBOL(s390_do_machine_check);
507
508static int __init machine_check_init(void)
509{
510 system_ctl_set_bit(14, CR14_EXTERNAL_DAMAGE_SUBMASK_BIT);
511 system_ctl_set_bit(14, CR14_RECOVERY_SUBMASK_BIT);
512 system_ctl_set_bit(14, CR14_WARNING_SUBMASK_BIT);
513 return 0;
514}
515early_initcall(machine_check_init);
1/*
2 * Machine check handler
3 *
4 * Copyright IBM Corp. 2000,2009
5 * Author(s): Ingo Adlung <adlung@de.ibm.com>,
6 * Martin Schwidefsky <schwidefsky@de.ibm.com>,
7 * Cornelia Huck <cornelia.huck@de.ibm.com>,
8 * Heiko Carstens <heiko.carstens@de.ibm.com>,
9 */
10
11#include <linux/kernel_stat.h>
12#include <linux/init.h>
13#include <linux/errno.h>
14#include <linux/hardirq.h>
15#include <linux/time.h>
16#include <linux/module.h>
17#include <asm/lowcore.h>
18#include <asm/smp.h>
19#include <asm/etr.h>
20#include <asm/cputime.h>
21#include <asm/nmi.h>
22#include <asm/crw.h>
23
24struct mcck_struct {
25 int kill_task;
26 int channel_report;
27 int warning;
28 unsigned long long mcck_code;
29};
30
31static DEFINE_PER_CPU(struct mcck_struct, cpu_mcck);
32
33static NORET_TYPE void s390_handle_damage(char *msg)
34{
35 smp_send_stop();
36 disabled_wait((unsigned long) __builtin_return_address(0));
37 while (1);
38}
39
40/*
41 * Main machine check handler function. Will be called with interrupts enabled
42 * or disabled and machine checks enabled or disabled.
43 */
44void s390_handle_mcck(void)
45{
46 unsigned long flags;
47 struct mcck_struct mcck;
48
49 /*
50 * Disable machine checks and get the current state of accumulated
51 * machine checks. Afterwards delete the old state and enable machine
52 * checks again.
53 */
54 local_irq_save(flags);
55 local_mcck_disable();
56 mcck = __get_cpu_var(cpu_mcck);
57 memset(&__get_cpu_var(cpu_mcck), 0, sizeof(struct mcck_struct));
58 clear_thread_flag(TIF_MCCK_PENDING);
59 local_mcck_enable();
60 local_irq_restore(flags);
61
62 if (mcck.channel_report)
63 crw_handle_channel_report();
64 /*
65 * A warning may remain for a prolonged period on the bare iron.
66 * (actually until the machine is powered off, or the problem is gone)
67 * So we just stop listening for the WARNING MCH and avoid continuously
68 * being interrupted. One caveat is however, that we must do this per
69 * processor and cannot use the smp version of ctl_clear_bit().
70 * On VM we only get one interrupt per virtally presented machinecheck.
71 * Though one suffices, we may get one interrupt per (virtual) cpu.
72 */
73 if (mcck.warning) { /* WARNING pending ? */
74 static int mchchk_wng_posted = 0;
75
76 /* Use single cpu clear, as we cannot handle smp here. */
77 __ctl_clear_bit(14, 24); /* Disable WARNING MCH */
78 if (xchg(&mchchk_wng_posted, 1) == 0)
79 kill_cad_pid(SIGPWR, 1);
80 }
81 if (mcck.kill_task) {
82 local_irq_enable();
83 printk(KERN_EMERG "mcck: Terminating task because of machine "
84 "malfunction (code 0x%016llx).\n", mcck.mcck_code);
85 printk(KERN_EMERG "mcck: task: %s, pid: %d.\n",
86 current->comm, current->pid);
87 do_exit(SIGSEGV);
88 }
89}
90EXPORT_SYMBOL_GPL(s390_handle_mcck);
91
92/*
93 * returns 0 if all registers could be validated
94 * returns 1 otherwise
95 */
96static int notrace s390_revalidate_registers(struct mci *mci)
97{
98 int kill_task;
99 u64 zero;
100 void *fpt_save_area, *fpt_creg_save_area;
101
102 kill_task = 0;
103 zero = 0;
104
105 if (!mci->gr) {
106 /*
107 * General purpose registers couldn't be restored and have
108 * unknown contents. Process needs to be terminated.
109 */
110 kill_task = 1;
111 }
112 if (!mci->fp) {
113 /*
114 * Floating point registers can't be restored and
115 * therefore the process needs to be terminated.
116 */
117 kill_task = 1;
118 }
119#ifndef CONFIG_64BIT
120 asm volatile(
121 " ld 0,0(%0)\n"
122 " ld 2,8(%0)\n"
123 " ld 4,16(%0)\n"
124 " ld 6,24(%0)"
125 : : "a" (&S390_lowcore.floating_pt_save_area));
126#endif
127
128 if (MACHINE_HAS_IEEE) {
129#ifdef CONFIG_64BIT
130 fpt_save_area = &S390_lowcore.floating_pt_save_area;
131 fpt_creg_save_area = &S390_lowcore.fpt_creg_save_area;
132#else
133 fpt_save_area = (void *) S390_lowcore.extended_save_area_addr;
134 fpt_creg_save_area = fpt_save_area + 128;
135#endif
136 if (!mci->fc) {
137 /*
138 * Floating point control register can't be restored.
139 * Task will be terminated.
140 */
141 asm volatile("lfpc 0(%0)" : : "a" (&zero), "m" (zero));
142 kill_task = 1;
143
144 } else
145 asm volatile("lfpc 0(%0)" : : "a" (fpt_creg_save_area));
146
147 asm volatile(
148 " ld 0,0(%0)\n"
149 " ld 1,8(%0)\n"
150 " ld 2,16(%0)\n"
151 " ld 3,24(%0)\n"
152 " ld 4,32(%0)\n"
153 " ld 5,40(%0)\n"
154 " ld 6,48(%0)\n"
155 " ld 7,56(%0)\n"
156 " ld 8,64(%0)\n"
157 " ld 9,72(%0)\n"
158 " ld 10,80(%0)\n"
159 " ld 11,88(%0)\n"
160 " ld 12,96(%0)\n"
161 " ld 13,104(%0)\n"
162 " ld 14,112(%0)\n"
163 " ld 15,120(%0)\n"
164 : : "a" (fpt_save_area));
165 }
166 /* Revalidate access registers */
167 asm volatile(
168 " lam 0,15,0(%0)"
169 : : "a" (&S390_lowcore.access_regs_save_area));
170 if (!mci->ar) {
171 /*
172 * Access registers have unknown contents.
173 * Terminating task.
174 */
175 kill_task = 1;
176 }
177 /* Revalidate control registers */
178 if (!mci->cr) {
179 /*
180 * Control registers have unknown contents.
181 * Can't recover and therefore stopping machine.
182 */
183 s390_handle_damage("invalid control registers.");
184 } else {
185#ifdef CONFIG_64BIT
186 asm volatile(
187 " lctlg 0,15,0(%0)"
188 : : "a" (&S390_lowcore.cregs_save_area));
189#else
190 asm volatile(
191 " lctl 0,15,0(%0)"
192 : : "a" (&S390_lowcore.cregs_save_area));
193#endif
194 }
195 /*
196 * We don't even try to revalidate the TOD register, since we simply
197 * can't write something sensible into that register.
198 */
199#ifdef CONFIG_64BIT
200 /*
201 * See if we can revalidate the TOD programmable register with its
202 * old contents (should be zero) otherwise set it to zero.
203 */
204 if (!mci->pr)
205 asm volatile(
206 " sr 0,0\n"
207 " sckpf"
208 : : : "0", "cc");
209 else
210 asm volatile(
211 " l 0,0(%0)\n"
212 " sckpf"
213 : : "a" (&S390_lowcore.tod_progreg_save_area)
214 : "0", "cc");
215#endif
216 /* Revalidate clock comparator register */
217 if (S390_lowcore.clock_comparator == -1)
218 set_clock_comparator(S390_lowcore.mcck_clock);
219 else
220 set_clock_comparator(S390_lowcore.clock_comparator);
221 /* Check if old PSW is valid */
222 if (!mci->wp)
223 /*
224 * Can't tell if we come from user or kernel mode
225 * -> stopping machine.
226 */
227 s390_handle_damage("old psw invalid.");
228
229 if (!mci->ms || !mci->pm || !mci->ia)
230 kill_task = 1;
231
232 return kill_task;
233}
234
235#define MAX_IPD_COUNT 29
236#define MAX_IPD_TIME (5 * 60 * USEC_PER_SEC) /* 5 minutes */
237
238#define ED_STP_ISLAND 6 /* External damage STP island check */
239#define ED_STP_SYNC 7 /* External damage STP sync check */
240#define ED_ETR_SYNC 12 /* External damage ETR sync check */
241#define ED_ETR_SWITCH 13 /* External damage ETR switch to local */
242
243/*
244 * machine check handler.
245 */
246void notrace s390_do_machine_check(struct pt_regs *regs)
247{
248 static int ipd_count;
249 static DEFINE_SPINLOCK(ipd_lock);
250 static unsigned long long last_ipd;
251 struct mcck_struct *mcck;
252 unsigned long long tmp;
253 struct mci *mci;
254 int umode;
255
256 nmi_enter();
257 s390_idle_check(regs, S390_lowcore.mcck_clock,
258 S390_lowcore.mcck_enter_timer);
259 kstat_cpu(smp_processor_id()).irqs[NMI_NMI]++;
260 mci = (struct mci *) &S390_lowcore.mcck_interruption_code;
261 mcck = &__get_cpu_var(cpu_mcck);
262 umode = user_mode(regs);
263
264 if (mci->sd) {
265 /* System damage -> stopping machine */
266 s390_handle_damage("received system damage machine check.");
267 }
268 if (mci->pd) {
269 if (mci->b) {
270 /* Processing backup -> verify if we can survive this */
271 u64 z_mcic, o_mcic, t_mcic;
272#ifdef CONFIG_64BIT
273 z_mcic = (1ULL<<63 | 1ULL<<59 | 1ULL<<29);
274 o_mcic = (1ULL<<43 | 1ULL<<42 | 1ULL<<41 | 1ULL<<40 |
275 1ULL<<36 | 1ULL<<35 | 1ULL<<34 | 1ULL<<32 |
276 1ULL<<30 | 1ULL<<21 | 1ULL<<20 | 1ULL<<17 |
277 1ULL<<16);
278#else
279 z_mcic = (1ULL<<63 | 1ULL<<59 | 1ULL<<57 | 1ULL<<50 |
280 1ULL<<29);
281 o_mcic = (1ULL<<43 | 1ULL<<42 | 1ULL<<41 | 1ULL<<40 |
282 1ULL<<36 | 1ULL<<35 | 1ULL<<34 | 1ULL<<32 |
283 1ULL<<30 | 1ULL<<20 | 1ULL<<17 | 1ULL<<16);
284#endif
285 t_mcic = *(u64 *)mci;
286
287 if (((t_mcic & z_mcic) != 0) ||
288 ((t_mcic & o_mcic) != o_mcic)) {
289 s390_handle_damage("processing backup machine "
290 "check with damage.");
291 }
292
293 /*
294 * Nullifying exigent condition, therefore we might
295 * retry this instruction.
296 */
297 spin_lock(&ipd_lock);
298 tmp = get_clock();
299 if (((tmp - last_ipd) >> 12) < MAX_IPD_TIME)
300 ipd_count++;
301 else
302 ipd_count = 1;
303 last_ipd = tmp;
304 if (ipd_count == MAX_IPD_COUNT)
305 s390_handle_damage("too many ipd retries.");
306 spin_unlock(&ipd_lock);
307 } else {
308 /* Processing damage -> stopping machine */
309 s390_handle_damage("received instruction processing "
310 "damage machine check.");
311 }
312 }
313 if (s390_revalidate_registers(mci)) {
314 if (umode) {
315 /*
316 * Couldn't restore all register contents while in
317 * user mode -> mark task for termination.
318 */
319 mcck->kill_task = 1;
320 mcck->mcck_code = *(unsigned long long *) mci;
321 set_thread_flag(TIF_MCCK_PENDING);
322 } else {
323 /*
324 * Couldn't restore all register contents while in
325 * kernel mode -> stopping machine.
326 */
327 s390_handle_damage("unable to revalidate registers.");
328 }
329 }
330 if (mci->cd) {
331 /* Timing facility damage */
332 s390_handle_damage("TOD clock damaged");
333 }
334 if (mci->ed && mci->ec) {
335 /* External damage */
336 if (S390_lowcore.external_damage_code & (1U << ED_ETR_SYNC))
337 etr_sync_check();
338 if (S390_lowcore.external_damage_code & (1U << ED_ETR_SWITCH))
339 etr_switch_to_local();
340 if (S390_lowcore.external_damage_code & (1U << ED_STP_SYNC))
341 stp_sync_check();
342 if (S390_lowcore.external_damage_code & (1U << ED_STP_ISLAND))
343 stp_island_check();
344 }
345 if (mci->se)
346 /* Storage error uncorrected */
347 s390_handle_damage("received storage error uncorrected "
348 "machine check.");
349 if (mci->ke)
350 /* Storage key-error uncorrected */
351 s390_handle_damage("received storage key-error uncorrected "
352 "machine check.");
353 if (mci->ds && mci->fa)
354 /* Storage degradation */
355 s390_handle_damage("received storage degradation machine "
356 "check.");
357 if (mci->cp) {
358 /* Channel report word pending */
359 mcck->channel_report = 1;
360 set_thread_flag(TIF_MCCK_PENDING);
361 }
362 if (mci->w) {
363 /* Warning pending */
364 mcck->warning = 1;
365 set_thread_flag(TIF_MCCK_PENDING);
366 }
367 nmi_exit();
368}
369
370static int __init machine_check_init(void)
371{
372 ctl_set_bit(14, 25); /* enable external damage MCH */
373 ctl_set_bit(14, 27); /* enable system recovery MCH */
374 ctl_set_bit(14, 24); /* enable warning MCH */
375 return 0;
376}
377arch_initcall(machine_check_init);