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 * Heiko Carstens <heiko.carstens@de.ibm.com>,
10 */
11
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>
22
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 <linux/kvm_host.h>
34
35 struct mcck_struct {
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;
41 };
42
43 static DEFINE_PER_CPU(struct mcck_struct, cpu_mcck);
44 static struct kmem_cache *mcesa_cache;
45 static unsigned long mcesa_origin_lc;
46
nmi_needs_mcesa(void)47 static inline int nmi_needs_mcesa(void)
48 {
49 return MACHINE_HAS_VX || MACHINE_HAS_GS;
50 }
51
nmi_get_mcesa_size(void)52 static inline unsigned long nmi_get_mcesa_size(void)
53 {
54 if (MACHINE_HAS_GS)
55 return MCESA_MAX_SIZE;
56 return MCESA_MIN_SIZE;
57 }
58
59 /*
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
63 * the boot process.
64 */
65 static struct mcesa boot_mcesa __aligned(MCESA_MAX_SIZE);
66
nmi_alloc_boot_cpu(struct lowcore * lc)67 void __init nmi_alloc_boot_cpu(struct lowcore *lc)
68 {
69 if (!nmi_needs_mcesa())
70 return;
71 lc->mcesad = (unsigned long) &boot_mcesa;
72 if (MACHINE_HAS_GS)
73 lc->mcesad |= ilog2(MCESA_MAX_SIZE);
74 }
75
nmi_init(void)76 static int __init nmi_init(void)
77 {
78 unsigned long origin, cr0, size;
79
80 if (!nmi_needs_mcesa())
81 return 0;
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);
87 if (!mcesa_cache)
88 panic("Couldn't create nmi save area cache");
89 origin = (unsigned long) kmem_cache_alloc(mcesa_cache, GFP_KERNEL);
90 if (!origin)
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);
99 return 0;
100 }
101 early_initcall(nmi_init);
102
nmi_alloc_per_cpu(struct lowcore * lc)103 int nmi_alloc_per_cpu(struct lowcore *lc)
104 {
105 unsigned long origin;
106
107 if (!nmi_needs_mcesa())
108 return 0;
109 origin = (unsigned long) kmem_cache_alloc(mcesa_cache, GFP_KERNEL);
110 if (!origin)
111 return -ENOMEM;
112 /* The pointer is stored with mcesa_bits ORed in */
113 kmemleak_not_leak((void *) origin);
114 lc->mcesad = origin | mcesa_origin_lc;
115 return 0;
116 }
117
nmi_free_per_cpu(struct lowcore * lc)118 void nmi_free_per_cpu(struct lowcore *lc)
119 {
120 if (!nmi_needs_mcesa())
121 return;
122 kmem_cache_free(mcesa_cache, (void *)(lc->mcesad & MCESA_ORIGIN_MASK));
123 }
124
s390_handle_damage(void)125 static notrace void s390_handle_damage(void)
126 {
127 smp_emergency_stop();
128 disabled_wait();
129 while (1);
130 }
131 NOKPROBE_SYMBOL(s390_handle_damage);
132
133 /*
134 * Main machine check handler function. Will be called with interrupts disabled
135 * and machine checks enabled.
136 */
__s390_handle_mcck(void)137 void __s390_handle_mcck(void)
138 {
139 struct mcck_struct mcck;
140
141 /*
142 * Disable machine checks and get the current state of accumulated
143 * machine checks. Afterwards delete the old state and enable machine
144 * checks again.
145 */
146 local_mcck_disable();
147 mcck = *this_cpu_ptr(&cpu_mcck);
148 memset(this_cpu_ptr(&cpu_mcck), 0, sizeof(mcck));
149 local_mcck_enable();
150
151 if (mcck.channel_report)
152 crw_handle_channel_report();
153 /*
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.
161 */
162 if (mcck.warning) { /* WARNING pending ? */
163 static int mchchk_wng_posted = 0;
164
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);
169 }
170 if (mcck.stp_queue)
171 stp_queue_work();
172 if (mcck.kill_task) {
173 local_irq_enable();
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);
178 make_task_dead(SIGSEGV);
179 }
180 }
181
s390_handle_mcck(void)182 void noinstr s390_handle_mcck(void)
183 {
184 trace_hardirqs_off();
185 __s390_handle_mcck();
186 trace_hardirqs_on();
187 }
188 /*
189 * returns 0 if all required registers are available
190 * returns 1 otherwise
191 */
s390_validate_registers(union mci mci,int umode)192 static int notrace s390_validate_registers(union mci mci, int umode)
193 {
194 struct mcesa *mcesa;
195 void *fpt_save_area;
196 union ctlreg2 cr2;
197 int kill_task;
198 u64 zero;
199
200 kill_task = 0;
201 zero = 0;
202
203 if (!mci.gr) {
204 /*
205 * General purpose registers couldn't be restored and have
206 * unknown contents. Stop system or terminate process.
207 */
208 if (!umode)
209 s390_handle_damage();
210 kill_task = 1;
211 }
212 if (!mci.fp) {
213 /*
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.
218 */
219 if (S390_lowcore.fpu_flags & KERNEL_VXR_V0V7)
220 s390_handle_damage();
221 if (!test_cpu_flag(CIF_FPU))
222 kill_task = 1;
223 }
224 fpt_save_area = &S390_lowcore.floating_pt_save_area;
225 if (!mci.fc) {
226 /*
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.
233 */
234 if (S390_lowcore.fpu_flags & KERNEL_FPC)
235 s390_handle_damage();
236 asm volatile(
237 " lfpc %0\n"
238 :
239 : "Q" (zero));
240 if (!test_cpu_flag(CIF_FPU))
241 kill_task = 1;
242 } else {
243 asm volatile(
244 " lfpc %0\n"
245 :
246 : "Q" (S390_lowcore.fpt_creg_save_area));
247 }
248
249 mcesa = (struct mcesa *)(S390_lowcore.mcesad & MCESA_ORIGIN_MASK);
250 if (!MACHINE_HAS_VX) {
251 /* Validate floating point registers */
252 asm volatile(
253 " ld 0,0(%0)\n"
254 " ld 1,8(%0)\n"
255 " ld 2,16(%0)\n"
256 " ld 3,24(%0)\n"
257 " ld 4,32(%0)\n"
258 " ld 5,40(%0)\n"
259 " ld 6,48(%0)\n"
260 " ld 7,56(%0)\n"
261 " ld 8,64(%0)\n"
262 " ld 9,72(%0)\n"
263 " ld 10,80(%0)\n"
264 " ld 11,88(%0)\n"
265 " ld 12,96(%0)\n"
266 " ld 13,104(%0)\n"
267 " ld 14,112(%0)\n"
268 " ld 15,120(%0)\n"
269 :
270 : "a" (fpt_save_area)
271 : "memory");
272 } else {
273 /* Validate vector registers */
274 union ctlreg0 cr0;
275
276 /*
277 * The vector validity must only be checked if not running a
278 * KVM guest. For KVM guests the machine check is forwarded by
279 * KVM and it is the responsibility of the guest to take
280 * appropriate actions. The host vector or FPU values have been
281 * saved by KVM and will be restored by KVM.
282 */
283 if (!mci.vr && !test_cpu_flag(CIF_MCCK_GUEST)) {
284 /*
285 * Vector registers can't be restored. If the kernel
286 * currently uses vector registers the system is
287 * stopped. If the process has its vector registers
288 * loaded it is terminated. Otherwise just validate
289 * the registers.
290 */
291 if (S390_lowcore.fpu_flags & KERNEL_VXR)
292 s390_handle_damage();
293 if (!test_cpu_flag(CIF_FPU))
294 kill_task = 1;
295 }
296 cr0.val = S390_lowcore.cregs_save_area[0];
297 cr0.afp = cr0.vx = 1;
298 __ctl_load(cr0.val, 0, 0);
299 asm volatile(
300 " la 1,%0\n"
301 " .word 0xe70f,0x1000,0x0036\n" /* vlm 0,15,0(1) */
302 " .word 0xe70f,0x1100,0x0c36\n" /* vlm 16,31,256(1) */
303 :
304 : "Q" (*(struct vx_array *)mcesa->vector_save_area)
305 : "1");
306 __ctl_load(S390_lowcore.cregs_save_area[0], 0, 0);
307 }
308 /* Validate access registers */
309 asm volatile(
310 " lam 0,15,0(%0)\n"
311 :
312 : "a" (&S390_lowcore.access_regs_save_area)
313 : "memory");
314 if (!mci.ar) {
315 /*
316 * Access registers have unknown contents.
317 * Terminating task.
318 */
319 kill_task = 1;
320 }
321 /* Validate guarded storage registers */
322 cr2.val = S390_lowcore.cregs_save_area[2];
323 if (cr2.gse) {
324 if (!mci.gs) {
325 /*
326 * 2 cases:
327 * - machine check in kernel or userspace
328 * - machine check while running SIE (KVM guest)
329 * For kernel or userspace the userspace values of
330 * guarded storage control can not be recreated, the
331 * process must be terminated.
332 * For SIE the guest values of guarded storage can not
333 * be recreated. This is either due to a bug or due to
334 * GS being disabled in the guest. The guest will be
335 * notified by KVM code and the guests machine check
336 * handling must take care of this. The host values
337 * are saved by KVM and are not affected.
338 */
339 if (!test_cpu_flag(CIF_MCCK_GUEST))
340 kill_task = 1;
341 } else {
342 load_gs_cb((struct gs_cb *)mcesa->guarded_storage_save_area);
343 }
344 }
345 /*
346 * The getcpu vdso syscall reads CPU number from the programmable
347 * field of the TOD clock. Disregard the TOD programmable register
348 * validity bit and load the CPU number into the TOD programmable
349 * field unconditionally.
350 */
351 set_tod_programmable_field(raw_smp_processor_id());
352 /* Validate clock comparator register */
353 set_clock_comparator(S390_lowcore.clock_comparator);
354
355 if (!mci.ms || !mci.pm || !mci.ia)
356 kill_task = 1;
357
358 return kill_task;
359 }
360 NOKPROBE_SYMBOL(s390_validate_registers);
361
362 /*
363 * Backup the guest's machine check info to its description block
364 */
s390_backup_mcck_info(struct pt_regs * regs)365 static void notrace s390_backup_mcck_info(struct pt_regs *regs)
366 {
367 struct mcck_volatile_info *mcck_backup;
368 struct sie_page *sie_page;
369
370 /* r14 contains the sie block, which was set in sie64a */
371 struct kvm_s390_sie_block *sie_block =
372 (struct kvm_s390_sie_block *) regs->gprs[14];
373
374 if (sie_block == NULL)
375 /* Something's seriously wrong, stop system. */
376 s390_handle_damage();
377
378 sie_page = container_of(sie_block, struct sie_page, sie_block);
379 mcck_backup = &sie_page->mcck_info;
380 mcck_backup->mcic = S390_lowcore.mcck_interruption_code &
381 ~(MCCK_CODE_CP | MCCK_CODE_EXT_DAMAGE);
382 mcck_backup->ext_damage_code = S390_lowcore.external_damage_code;
383 mcck_backup->failing_storage_address
384 = S390_lowcore.failing_storage_address;
385 }
386 NOKPROBE_SYMBOL(s390_backup_mcck_info);
387
388 #define MAX_IPD_COUNT 29
389 #define MAX_IPD_TIME (5 * 60 * USEC_PER_SEC) /* 5 minutes */
390
391 #define ED_STP_ISLAND 6 /* External damage STP island check */
392 #define ED_STP_SYNC 7 /* External damage STP sync check */
393
394 #define MCCK_CODE_NO_GUEST (MCCK_CODE_CP | MCCK_CODE_EXT_DAMAGE)
395
396 /*
397 * machine check handler.
398 */
s390_do_machine_check(struct pt_regs * regs)399 int notrace s390_do_machine_check(struct pt_regs *regs)
400 {
401 static int ipd_count;
402 static DEFINE_SPINLOCK(ipd_lock);
403 static unsigned long long last_ipd;
404 struct mcck_struct *mcck;
405 unsigned long long tmp;
406 union mci mci;
407 unsigned long mcck_dam_code;
408 int mcck_pending = 0;
409
410 nmi_enter();
411
412 if (user_mode(regs))
413 update_timer_mcck();
414 inc_irq_stat(NMI_NMI);
415 mci.val = S390_lowcore.mcck_interruption_code;
416 mcck = this_cpu_ptr(&cpu_mcck);
417
418 /*
419 * Reinject the instruction processing damages' machine checks
420 * including Delayed Access Exception into the guest
421 * instead of damaging the host if they happen in the guest.
422 */
423 if (mci.pd && !test_cpu_flag(CIF_MCCK_GUEST)) {
424 if (mci.b) {
425 /* Processing backup -> verify if we can survive this */
426 u64 z_mcic, o_mcic, t_mcic;
427 z_mcic = (1ULL<<63 | 1ULL<<59 | 1ULL<<29);
428 o_mcic = (1ULL<<43 | 1ULL<<42 | 1ULL<<41 | 1ULL<<40 |
429 1ULL<<36 | 1ULL<<35 | 1ULL<<34 | 1ULL<<32 |
430 1ULL<<30 | 1ULL<<21 | 1ULL<<20 | 1ULL<<17 |
431 1ULL<<16);
432 t_mcic = mci.val;
433
434 if (((t_mcic & z_mcic) != 0) ||
435 ((t_mcic & o_mcic) != o_mcic)) {
436 s390_handle_damage();
437 }
438
439 /*
440 * Nullifying exigent condition, therefore we might
441 * retry this instruction.
442 */
443 spin_lock(&ipd_lock);
444 tmp = get_tod_clock();
445 if (((tmp - last_ipd) >> 12) < MAX_IPD_TIME)
446 ipd_count++;
447 else
448 ipd_count = 1;
449 last_ipd = tmp;
450 if (ipd_count == MAX_IPD_COUNT)
451 s390_handle_damage();
452 spin_unlock(&ipd_lock);
453 } else {
454 /* Processing damage -> stopping machine */
455 s390_handle_damage();
456 }
457 }
458 if (s390_validate_registers(mci, user_mode(regs))) {
459 /*
460 * Couldn't restore all register contents for the
461 * user space process -> mark task for termination.
462 */
463 mcck->kill_task = 1;
464 mcck->mcck_code = mci.val;
465 mcck_pending = 1;
466 }
467
468 /*
469 * Backup the machine check's info if it happens when the guest
470 * is running.
471 */
472 if (test_cpu_flag(CIF_MCCK_GUEST))
473 s390_backup_mcck_info(regs);
474
475 if (mci.cd) {
476 /* Timing facility damage */
477 s390_handle_damage();
478 }
479 if (mci.ed && mci.ec) {
480 /* External damage */
481 if (S390_lowcore.external_damage_code & (1U << ED_STP_SYNC))
482 mcck->stp_queue |= stp_sync_check();
483 if (S390_lowcore.external_damage_code & (1U << ED_STP_ISLAND))
484 mcck->stp_queue |= stp_island_check();
485 mcck_pending = 1;
486 }
487
488 if (mci.cp) {
489 /* Channel report word pending */
490 mcck->channel_report = 1;
491 mcck_pending = 1;
492 }
493 if (mci.w) {
494 /* Warning pending */
495 mcck->warning = 1;
496 mcck_pending = 1;
497 }
498
499 /*
500 * If there are only Channel Report Pending and External Damage
501 * machine checks, they will not be reinjected into the guest
502 * because they refer to host conditions only.
503 */
504 mcck_dam_code = (mci.val & MCIC_SUBCLASS_MASK);
505 if (test_cpu_flag(CIF_MCCK_GUEST) &&
506 (mcck_dam_code & MCCK_CODE_NO_GUEST) != mcck_dam_code) {
507 /* Set exit reason code for host's later handling */
508 *((long *)(regs->gprs[15] + __SF_SIE_REASON)) = -EINTR;
509 }
510 clear_cpu_flag(CIF_MCCK_GUEST);
511
512 if (user_mode(regs) && mcck_pending) {
513 nmi_exit();
514 return 1;
515 }
516
517 if (mcck_pending)
518 schedule_mcck_handler();
519
520 nmi_exit();
521 return 0;
522 }
523 NOKPROBE_SYMBOL(s390_do_machine_check);
524
machine_check_init(void)525 static int __init machine_check_init(void)
526 {
527 ctl_set_bit(14, 25); /* enable external damage MCH */
528 ctl_set_bit(14, 27); /* enable system recovery MCH */
529 ctl_set_bit(14, 24); /* enable warning MCH */
530 return 0;
531 }
532 early_initcall(machine_check_init);
533