1 // SPDX-License-Identifier: GPL-2.0
3 * Machine check handler
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>,
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 enabled
135 * or disabled and machine checks enabled or disabled.
137 void s390_handle_mcck(void)
140 struct mcck_struct mcck
;
143 * Disable machine checks and get the current state of accumulated
144 * machine checks. Afterwards delete the old state and enable machine
147 local_irq_save(flags
);
148 local_mcck_disable();
149 mcck
= *this_cpu_ptr(&cpu_mcck
);
150 memset(this_cpu_ptr(&cpu_mcck
), 0, sizeof(mcck
));
152 local_irq_restore(flags
);
154 if (mcck
.channel_report
)
155 crw_handle_channel_report();
157 * A warning may remain for a prolonged period on the bare iron.
158 * (actually until the machine is powered off, or the problem is gone)
159 * So we just stop listening for the WARNING MCH and avoid continuously
160 * being interrupted. One caveat is however, that we must do this per
161 * processor and cannot use the smp version of ctl_clear_bit().
162 * On VM we only get one interrupt per virtally presented machinecheck.
163 * Though one suffices, we may get one interrupt per (virtual) cpu.
165 if (mcck
.warning
) { /* WARNING pending ? */
166 static int mchchk_wng_posted
= 0;
168 /* Use single cpu clear, as we cannot handle smp here. */
169 __ctl_clear_bit(14, 24); /* Disable WARNING MCH */
170 if (xchg(&mchchk_wng_posted
, 1) == 0)
171 kill_cad_pid(SIGPWR
, 1);
175 if (mcck
.kill_task
) {
177 printk(KERN_EMERG
"mcck: Terminating task because of machine "
178 "malfunction (code 0x%016lx).\n", mcck
.mcck_code
);
179 printk(KERN_EMERG
"mcck: task: %s, pid: %d.\n",
180 current
->comm
, current
->pid
);
184 EXPORT_SYMBOL_GPL(s390_handle_mcck
);
187 * returns 0 if all required registers are available
188 * returns 1 otherwise
190 static int notrace
s390_check_registers(union mci mci
, int umode
)
199 * General purpose registers couldn't be restored and have
200 * unknown contents. Stop system or terminate process.
203 s390_handle_damage();
206 /* Check control registers */
209 * Control registers have unknown contents.
210 * Can't recover and therefore stopping machine.
212 s390_handle_damage();
216 * Floating point registers can't be restored. If the
217 * kernel currently uses floating point registers the
218 * system is stopped. If the process has its floating
219 * pointer registers loaded it is terminated.
221 if (S390_lowcore
.fpu_flags
& KERNEL_VXR_V0V7
)
222 s390_handle_damage();
223 if (!test_cpu_flag(CIF_FPU
))
228 * Floating point control register can't be restored.
229 * If the kernel currently uses the floating pointer
230 * registers and needs the FPC register the system is
231 * stopped. If the process has its floating pointer
232 * registers loaded it is terminated.
234 if (S390_lowcore
.fpu_flags
& KERNEL_FPC
)
235 s390_handle_damage();
236 if (!test_cpu_flag(CIF_FPU
))
240 if (MACHINE_HAS_VX
) {
243 * Vector registers can't be restored. If the kernel
244 * currently uses vector registers the system is
245 * stopped. If the process has its vector registers
246 * loaded it is terminated.
248 if (S390_lowcore
.fpu_flags
& KERNEL_VXR
)
249 s390_handle_damage();
250 if (!test_cpu_flag(CIF_FPU
))
254 /* Check if access registers are valid */
257 * Access registers have unknown contents.
262 /* Check guarded storage registers */
263 cr2
.val
= S390_lowcore
.cregs_save_area
[2];
267 * Guarded storage register can't be restored and
268 * the current processes uses guarded storage.
269 * It has to be terminated.
274 /* Check if old PSW is valid */
277 * Can't tell if we come from user or kernel mode
278 * -> stopping machine.
280 s390_handle_damage();
282 /* Check for invalid kernel instruction address */
283 if (!mci
.ia
&& !umode
) {
285 * The instruction address got lost while running
286 * in the kernel -> stopping machine.
288 s390_handle_damage();
291 if (!mci
.ms
|| !mci
.pm
|| !mci
.ia
)
296 NOKPROBE_SYMBOL(s390_check_registers
);
299 * Backup the guest's machine check info to its description block
301 static void notrace
s390_backup_mcck_info(struct pt_regs
*regs
)
303 struct mcck_volatile_info
*mcck_backup
;
304 struct sie_page
*sie_page
;
306 /* r14 contains the sie block, which was set in sie64a */
307 struct kvm_s390_sie_block
*sie_block
=
308 (struct kvm_s390_sie_block
*) regs
->gprs
[14];
310 if (sie_block
== NULL
)
311 /* Something's seriously wrong, stop system. */
312 s390_handle_damage();
314 sie_page
= container_of(sie_block
, struct sie_page
, sie_block
);
315 mcck_backup
= &sie_page
->mcck_info
;
316 mcck_backup
->mcic
= S390_lowcore
.mcck_interruption_code
&
317 ~(MCCK_CODE_CP
| MCCK_CODE_EXT_DAMAGE
);
318 mcck_backup
->ext_damage_code
= S390_lowcore
.external_damage_code
;
319 mcck_backup
->failing_storage_address
320 = S390_lowcore
.failing_storage_address
;
322 NOKPROBE_SYMBOL(s390_backup_mcck_info
);
324 #define MAX_IPD_COUNT 29
325 #define MAX_IPD_TIME (5 * 60 * USEC_PER_SEC) /* 5 minutes */
327 #define ED_STP_ISLAND 6 /* External damage STP island check */
328 #define ED_STP_SYNC 7 /* External damage STP sync check */
330 #define MCCK_CODE_NO_GUEST (MCCK_CODE_CP | MCCK_CODE_EXT_DAMAGE)
333 * machine check handler.
335 int notrace
s390_do_machine_check(struct pt_regs
*regs
)
337 static int ipd_count
;
338 static DEFINE_SPINLOCK(ipd_lock
);
339 static unsigned long long last_ipd
;
340 struct mcck_struct
*mcck
;
341 unsigned long long tmp
;
343 unsigned long mcck_dam_code
;
344 int mcck_pending
= 0;
347 inc_irq_stat(NMI_NMI
);
348 mci
.val
= S390_lowcore
.mcck_interruption_code
;
349 mcck
= this_cpu_ptr(&cpu_mcck
);
352 /* System damage -> stopping machine */
353 s390_handle_damage();
357 * Reinject the instruction processing damages' machine checks
358 * including Delayed Access Exception into the guest
359 * instead of damaging the host if they happen in the guest.
361 if (mci
.pd
&& !test_cpu_flag(CIF_MCCK_GUEST
)) {
363 /* Processing backup -> verify if we can survive this */
364 u64 z_mcic
, o_mcic
, t_mcic
;
365 z_mcic
= (1ULL<<63 | 1ULL<<59 | 1ULL<<29);
366 o_mcic
= (1ULL<<43 | 1ULL<<42 | 1ULL<<41 | 1ULL<<40 |
367 1ULL<<36 | 1ULL<<35 | 1ULL<<34 | 1ULL<<32 |
368 1ULL<<30 | 1ULL<<21 | 1ULL<<20 | 1ULL<<17 |
372 if (((t_mcic
& z_mcic
) != 0) ||
373 ((t_mcic
& o_mcic
) != o_mcic
)) {
374 s390_handle_damage();
378 * Nullifying exigent condition, therefore we might
379 * retry this instruction.
381 spin_lock(&ipd_lock
);
382 tmp
= get_tod_clock();
383 if (((tmp
- last_ipd
) >> 12) < MAX_IPD_TIME
)
388 if (ipd_count
== MAX_IPD_COUNT
)
389 s390_handle_damage();
390 spin_unlock(&ipd_lock
);
392 /* Processing damage -> stopping machine */
393 s390_handle_damage();
396 if (s390_check_registers(mci
, user_mode(regs
))) {
398 * Couldn't restore all register contents for the
399 * user space process -> mark task for termination.
402 mcck
->mcck_code
= mci
.val
;
407 * Backup the machine check's info if it happens when the guest
410 if (test_cpu_flag(CIF_MCCK_GUEST
))
411 s390_backup_mcck_info(regs
);
414 /* Timing facility damage */
415 s390_handle_damage();
417 if (mci
.ed
&& mci
.ec
) {
418 /* External damage */
419 if (S390_lowcore
.external_damage_code
& (1U << ED_STP_SYNC
))
420 mcck
->stp_queue
|= stp_sync_check();
421 if (S390_lowcore
.external_damage_code
& (1U << ED_STP_ISLAND
))
422 mcck
->stp_queue
|= stp_island_check();
427 * Reinject storage related machine checks into the guest if they
428 * happen when the guest is running.
430 if (!test_cpu_flag(CIF_MCCK_GUEST
)) {
432 /* Storage error uncorrected */
433 s390_handle_damage();
435 /* Storage key-error uncorrected */
436 s390_handle_damage();
437 if (mci
.ds
&& mci
.fa
)
438 /* Storage degradation */
439 s390_handle_damage();
442 /* Channel report word pending */
443 mcck
->channel_report
= 1;
447 /* Warning pending */
453 * If there are only Channel Report Pending and External Damage
454 * machine checks, they will not be reinjected into the guest
455 * because they refer to host conditions only.
457 mcck_dam_code
= (mci
.val
& MCIC_SUBCLASS_MASK
);
458 if (test_cpu_flag(CIF_MCCK_GUEST
) &&
459 (mcck_dam_code
& MCCK_CODE_NO_GUEST
) != mcck_dam_code
) {
460 /* Set exit reason code for host's later handling */
461 *((long *)(regs
->gprs
[15] + __SF_SIE_REASON
)) = -EINTR
;
463 clear_cpu_flag(CIF_MCCK_GUEST
);
465 if (user_mode(regs
) && mcck_pending
) {
471 schedule_mcck_handler();
476 NOKPROBE_SYMBOL(s390_do_machine_check
);
478 static int __init
machine_check_init(void)
480 ctl_set_bit(14, 25); /* enable external damage MCH */
481 ctl_set_bit(14, 27); /* enable system recovery MCH */
482 ctl_set_bit(14, 24); /* enable warning MCH */
485 early_initcall(machine_check_init
);