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/smp.h>
26 #include <asm/stp.h>
27 #include <asm/cputime.h>
28 #include <asm/nmi.h>
29 #include <asm/crw.h>
30 #include <asm/switch_to.h>
31 #include <asm/ctl_reg.h>
32 #include <asm/asm-offsets.h>
33 #include <asm/pai.h>
34 #include <asm/vx-insn.h>
35
36 struct mcck_struct {
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;
42 };
43
44 static DEFINE_PER_CPU(struct mcck_struct, cpu_mcck);
45
nmi_needs_mcesa(void)46 static inline int nmi_needs_mcesa(void)
47 {
48 return MACHINE_HAS_VX || MACHINE_HAS_GS;
49 }
50
51 /*
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.
56 */
57 static struct mcesa boot_mcesa __aligned(MCESA_MAX_SIZE);
58
nmi_alloc_mcesa_early(u64 * mcesad)59 void __init nmi_alloc_mcesa_early(u64 *mcesad)
60 {
61 if (!nmi_needs_mcesa())
62 return;
63 *mcesad = __pa(&boot_mcesa);
64 if (MACHINE_HAS_GS)
65 *mcesad |= ilog2(MCESA_MAX_SIZE);
66 }
67
nmi_alloc_mcesa(u64 * mcesad)68 int nmi_alloc_mcesa(u64 *mcesad)
69 {
70 unsigned long size;
71 void *origin;
72
73 *mcesad = 0;
74 if (!nmi_needs_mcesa())
75 return 0;
76 size = MACHINE_HAS_GS ? MCESA_MAX_SIZE : MCESA_MIN_SIZE;
77 origin = kmalloc(size, GFP_KERNEL);
78 if (!origin)
79 return -ENOMEM;
80 /* The pointer is stored with mcesa_bits ORed in */
81 kmemleak_not_leak(origin);
82 *mcesad = __pa(origin);
83 if (MACHINE_HAS_GS)
84 *mcesad |= ilog2(MCESA_MAX_SIZE);
85 return 0;
86 }
87
nmi_free_mcesa(u64 * mcesad)88 void nmi_free_mcesa(u64 *mcesad)
89 {
90 if (!nmi_needs_mcesa())
91 return;
92 kfree(__va(*mcesad & MCESA_ORIGIN_MASK));
93 }
94
nmi_puts(char * dest,const char * src)95 static __always_inline char *nmi_puts(char *dest, const char *src)
96 {
97 while (*src)
98 *dest++ = *src++;
99 *dest = 0;
100 return dest;
101 }
102
u64_to_hex(char * dest,u64 val)103 static __always_inline char *u64_to_hex(char *dest, u64 val)
104 {
105 int i, num;
106
107 for (i = 1; i <= 16; i++) {
108 num = (val >> (64 - 4 * i)) & 0xf;
109 if (num >= 10)
110 *dest++ = 'A' + num - 10;
111 else
112 *dest++ = '0' + num;
113 }
114 *dest = 0;
115 return dest;
116 }
117
s390_handle_damage(void)118 static notrace void s390_handle_damage(void)
119 {
120 union ctlreg0 cr0, cr0_new;
121 char message[100];
122 psw_t psw_save;
123 char *ptr;
124
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);
129
130 /*
131 * Disable low address protection and make machine check new PSW a
132 * disabled wait PSW. Any additional machine check cannot be handled.
133 */
134 __ctl_store(cr0.val, 0, 0);
135 cr0_new = cr0;
136 cr0_new.lap = 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);
143
144 /*
145 * Restore machine check new PSW and control register 0 to original
146 * values. This makes possible system dump analysis easier.
147 */
148 S390_lowcore.mcck_new_psw = psw_save;
149 __ctl_load(cr0.val, 0, 0);
150 disabled_wait();
151 while (1);
152 }
153 NOKPROBE_SYMBOL(s390_handle_damage);
154
155 /*
156 * Main machine check handler function. Will be called with interrupts disabled
157 * and machine checks enabled.
158 */
s390_handle_mcck(void)159 void s390_handle_mcck(void)
160 {
161 struct mcck_struct mcck;
162
163 /*
164 * Disable machine checks and get the current state of accumulated
165 * machine checks. Afterwards delete the old state and enable machine
166 * checks again.
167 */
168 local_mcck_disable();
169 mcck = *this_cpu_ptr(&cpu_mcck);
170 memset(this_cpu_ptr(&cpu_mcck), 0, sizeof(mcck));
171 local_mcck_enable();
172
173 if (mcck.channel_report)
174 crw_handle_channel_report();
175 /*
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.
183 */
184 if (mcck.warning) { /* WARNING pending ? */
185 static int mchchk_wng_posted = 0;
186
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);
191 }
192 if (mcck.stp_queue)
193 stp_queue_work();
194 if (mcck.kill_task) {
195 printk(KERN_EMERG "mcck: Terminating task because of machine "
196 "malfunction (code 0x%016lx).\n", mcck.mcck_code);
197 printk(KERN_EMERG "mcck: task: %s, pid: %d.\n",
198 current->comm, current->pid);
199 if (is_global_init(current))
200 panic("mcck: Attempting to kill init!\n");
201 do_send_sig_info(SIGKILL, SEND_SIG_PRIV, current, PIDTYPE_PID);
202 }
203 }
204
205 /*
206 * returns 0 if register contents could be validated
207 * returns 1 otherwise
208 */
s390_validate_registers(union mci mci)209 static int notrace s390_validate_registers(union mci mci)
210 {
211 struct mcesa *mcesa;
212 void *fpt_save_area;
213 union ctlreg2 cr2;
214 int kill_task;
215 u64 zero;
216
217 kill_task = 0;
218 zero = 0;
219
220 if (!mci.gr || !mci.fp)
221 kill_task = 1;
222 fpt_save_area = &S390_lowcore.floating_pt_save_area;
223 if (!mci.fc) {
224 kill_task = 1;
225 asm volatile(
226 " lfpc %0\n"
227 :
228 : "Q" (zero));
229 } else {
230 asm volatile(
231 " lfpc %0\n"
232 :
233 : "Q" (S390_lowcore.fpt_creg_save_area));
234 }
235
236 mcesa = __va(S390_lowcore.mcesad & MCESA_ORIGIN_MASK);
237 if (!MACHINE_HAS_VX) {
238 /* Validate floating point registers */
239 asm volatile(
240 " ld 0,0(%0)\n"
241 " ld 1,8(%0)\n"
242 " ld 2,16(%0)\n"
243 " ld 3,24(%0)\n"
244 " ld 4,32(%0)\n"
245 " ld 5,40(%0)\n"
246 " ld 6,48(%0)\n"
247 " ld 7,56(%0)\n"
248 " ld 8,64(%0)\n"
249 " ld 9,72(%0)\n"
250 " ld 10,80(%0)\n"
251 " ld 11,88(%0)\n"
252 " ld 12,96(%0)\n"
253 " ld 13,104(%0)\n"
254 " ld 14,112(%0)\n"
255 " ld 15,120(%0)\n"
256 :
257 : "a" (fpt_save_area)
258 : "memory");
259 } else {
260 /* Validate vector registers */
261 union ctlreg0 cr0;
262
263 /*
264 * The vector validity must only be checked if not running a
265 * KVM guest. For KVM guests the machine check is forwarded by
266 * KVM and it is the responsibility of the guest to take
267 * appropriate actions. The host vector or FPU values have been
268 * saved by KVM and will be restored by KVM.
269 */
270 if (!mci.vr && !test_cpu_flag(CIF_MCCK_GUEST))
271 kill_task = 1;
272 cr0.val = S390_lowcore.cregs_save_area[0];
273 cr0.afp = cr0.vx = 1;
274 __ctl_load(cr0.val, 0, 0);
275 asm volatile(
276 " la 1,%0\n"
277 " VLM 0,15,0,1\n"
278 " VLM 16,31,256,1\n"
279 :
280 : "Q" (*(struct vx_array *)mcesa->vector_save_area)
281 : "1");
282 __ctl_load(S390_lowcore.cregs_save_area[0], 0, 0);
283 }
284 /* Validate access registers */
285 asm volatile(
286 " lam 0,15,0(%0)\n"
287 :
288 : "a" (&S390_lowcore.access_regs_save_area)
289 : "memory");
290 if (!mci.ar)
291 kill_task = 1;
292 /* Validate guarded storage registers */
293 cr2.val = S390_lowcore.cregs_save_area[2];
294 if (cr2.gse) {
295 if (!mci.gs) {
296 /*
297 * 2 cases:
298 * - machine check in kernel or userspace
299 * - machine check while running SIE (KVM guest)
300 * For kernel or userspace the userspace values of
301 * guarded storage control can not be recreated, the
302 * process must be terminated.
303 * For SIE the guest values of guarded storage can not
304 * be recreated. This is either due to a bug or due to
305 * GS being disabled in the guest. The guest will be
306 * notified by KVM code and the guests machine check
307 * handling must take care of this. The host values
308 * are saved by KVM and are not affected.
309 */
310 if (!test_cpu_flag(CIF_MCCK_GUEST))
311 kill_task = 1;
312 } else {
313 load_gs_cb((struct gs_cb *)mcesa->guarded_storage_save_area);
314 }
315 }
316 /*
317 * The getcpu vdso syscall reads CPU number from the programmable
318 * field of the TOD clock. Disregard the TOD programmable register
319 * validity bit and load the CPU number into the TOD programmable
320 * field unconditionally.
321 */
322 set_tod_programmable_field(raw_smp_processor_id());
323 /* Validate clock comparator register */
324 set_clock_comparator(S390_lowcore.clock_comparator);
325
326 if (!mci.ms || !mci.pm || !mci.ia)
327 kill_task = 1;
328
329 return kill_task;
330 }
331 NOKPROBE_SYMBOL(s390_validate_registers);
332
333 /*
334 * Backup the guest's machine check info to its description block
335 */
s390_backup_mcck_info(struct pt_regs * regs)336 static void notrace s390_backup_mcck_info(struct pt_regs *regs)
337 {
338 struct mcck_volatile_info *mcck_backup;
339 struct sie_page *sie_page;
340
341 /* r14 contains the sie block, which was set in sie64a */
342 struct kvm_s390_sie_block *sie_block = phys_to_virt(regs->gprs[14]);
343
344 if (sie_block == NULL)
345 /* Something's seriously wrong, stop system. */
346 s390_handle_damage();
347
348 sie_page = container_of(sie_block, struct sie_page, sie_block);
349 mcck_backup = &sie_page->mcck_info;
350 mcck_backup->mcic = S390_lowcore.mcck_interruption_code &
351 ~(MCCK_CODE_CP | MCCK_CODE_EXT_DAMAGE);
352 mcck_backup->ext_damage_code = S390_lowcore.external_damage_code;
353 mcck_backup->failing_storage_address
354 = S390_lowcore.failing_storage_address;
355 }
356 NOKPROBE_SYMBOL(s390_backup_mcck_info);
357
358 #define MAX_IPD_COUNT 29
359 #define MAX_IPD_TIME (5 * 60 * USEC_PER_SEC) /* 5 minutes */
360
361 #define ED_STP_ISLAND 6 /* External damage STP island check */
362 #define ED_STP_SYNC 7 /* External damage STP sync check */
363
364 #define MCCK_CODE_NO_GUEST (MCCK_CODE_CP | MCCK_CODE_EXT_DAMAGE)
365
366 /*
367 * machine check handler.
368 */
s390_do_machine_check(struct pt_regs * regs)369 void notrace s390_do_machine_check(struct pt_regs *regs)
370 {
371 static int ipd_count;
372 static DEFINE_SPINLOCK(ipd_lock);
373 static unsigned long long last_ipd;
374 struct mcck_struct *mcck;
375 unsigned long long tmp;
376 irqentry_state_t irq_state;
377 union mci mci;
378 unsigned long mcck_dam_code;
379 int mcck_pending = 0;
380
381 irq_state = irqentry_nmi_enter(regs);
382
383 if (user_mode(regs))
384 update_timer_mcck();
385 inc_irq_stat(NMI_NMI);
386 mci.val = S390_lowcore.mcck_interruption_code;
387 mcck = this_cpu_ptr(&cpu_mcck);
388
389 /*
390 * Reinject the instruction processing damages' machine checks
391 * including Delayed Access Exception into the guest
392 * instead of damaging the host if they happen in the guest.
393 */
394 if (mci.pd && !test_cpu_flag(CIF_MCCK_GUEST)) {
395 if (mci.b) {
396 /* Processing backup -> verify if we can survive this */
397 u64 z_mcic, o_mcic, t_mcic;
398 z_mcic = (1ULL<<63 | 1ULL<<59 | 1ULL<<29);
399 o_mcic = (1ULL<<43 | 1ULL<<42 | 1ULL<<41 | 1ULL<<40 |
400 1ULL<<36 | 1ULL<<35 | 1ULL<<34 | 1ULL<<32 |
401 1ULL<<30 | 1ULL<<21 | 1ULL<<20 | 1ULL<<17 |
402 1ULL<<16);
403 t_mcic = mci.val;
404
405 if (((t_mcic & z_mcic) != 0) ||
406 ((t_mcic & o_mcic) != o_mcic)) {
407 s390_handle_damage();
408 }
409
410 /*
411 * Nullifying exigent condition, therefore we might
412 * retry this instruction.
413 */
414 spin_lock(&ipd_lock);
415 tmp = get_tod_clock();
416 if (((tmp - last_ipd) >> 12) < MAX_IPD_TIME)
417 ipd_count++;
418 else
419 ipd_count = 1;
420 last_ipd = tmp;
421 if (ipd_count == MAX_IPD_COUNT)
422 s390_handle_damage();
423 spin_unlock(&ipd_lock);
424 } else {
425 /* Processing damage -> stopping machine */
426 s390_handle_damage();
427 }
428 }
429 if (s390_validate_registers(mci)) {
430 if (!user_mode(regs))
431 s390_handle_damage();
432 /*
433 * Couldn't restore all register contents for the
434 * user space process -> mark task for termination.
435 */
436 mcck->kill_task = 1;
437 mcck->mcck_code = mci.val;
438 mcck_pending = 1;
439 }
440
441 /*
442 * Backup the machine check's info if it happens when the guest
443 * is running.
444 */
445 if (test_cpu_flag(CIF_MCCK_GUEST))
446 s390_backup_mcck_info(regs);
447
448 if (mci.cd) {
449 /* Timing facility damage */
450 s390_handle_damage();
451 }
452 if (mci.ed && mci.ec) {
453 /* External damage */
454 if (S390_lowcore.external_damage_code & (1U << ED_STP_SYNC))
455 mcck->stp_queue |= stp_sync_check();
456 if (S390_lowcore.external_damage_code & (1U << ED_STP_ISLAND))
457 mcck->stp_queue |= stp_island_check();
458 mcck_pending = 1;
459 }
460 /*
461 * Reinject storage related machine checks into the guest if they
462 * happen when the guest is running.
463 */
464 if (!test_cpu_flag(CIF_MCCK_GUEST)) {
465 /* Storage error uncorrected */
466 if (mci.se)
467 s390_handle_damage();
468 /* Storage key-error uncorrected */
469 if (mci.ke)
470 s390_handle_damage();
471 /* Storage degradation */
472 if (mci.ds && mci.fa)
473 s390_handle_damage();
474 }
475 if (mci.cp) {
476 /* Channel report word pending */
477 mcck->channel_report = 1;
478 mcck_pending = 1;
479 }
480 if (mci.w) {
481 /* Warning pending */
482 mcck->warning = 1;
483 mcck_pending = 1;
484 }
485
486 /*
487 * If there are only Channel Report Pending and External Damage
488 * machine checks, they will not be reinjected into the guest
489 * because they refer to host conditions only.
490 */
491 mcck_dam_code = (mci.val & MCIC_SUBCLASS_MASK);
492 if (test_cpu_flag(CIF_MCCK_GUEST) &&
493 (mcck_dam_code & MCCK_CODE_NO_GUEST) != mcck_dam_code) {
494 /* Set exit reason code for host's later handling */
495 *((long *)(regs->gprs[15] + __SF_SIE_REASON)) = -EINTR;
496 }
497 clear_cpu_flag(CIF_MCCK_GUEST);
498
499 if (mcck_pending)
500 schedule_mcck_handler();
501
502 irqentry_nmi_exit(regs, irq_state);
503 }
504 NOKPROBE_SYMBOL(s390_do_machine_check);
505
machine_check_init(void)506 static int __init machine_check_init(void)
507 {
508 ctl_set_bit(14, 25); /* enable external damage MCH */
509 ctl_set_bit(14, 27); /* enable system recovery MCH */
510 ctl_set_bit(14, 24); /* enable warning MCH */
511 return 0;
512 }
513 early_initcall(machine_check_init);
514