2 * drivers/s390/s390mach.c
3 * S/390 machine check handler
6 * Copyright (C) 2000 IBM Deutschland Entwicklung GmbH, IBM Corporation
7 * Author(s): Ingo Adlung (adlung@de.ibm.com)
8 * Martin Schwidefsky (schwidefsky@de.ibm.com)
11 #include <linux/init.h>
12 #include <linux/sched.h>
13 #include <linux/errno.h>
14 #include <linux/workqueue.h>
15 #include <linux/time.h>
16 #include <linux/kthread.h>
18 #include <asm/lowcore.h>
22 static struct semaphore m_sem
;
24 extern int css_process_crw(int, int);
25 extern int chsc_process_crw(void);
26 extern int chp_process_crw(int, int);
27 extern void css_reiterate_subchannels(void);
29 extern struct workqueue_struct
*slow_path_wq
;
30 extern struct work_struct slow_path_work
;
32 static NORET_TYPE
void
33 s390_handle_damage(char *msg
)
38 disabled_wait((unsigned long) __builtin_return_address(0));
43 * Retrieve CRWs and call function to handle event.
45 * Note : we currently process CRWs for io and chsc subchannels only
48 s390_collect_crw_info(void *param
)
52 struct semaphore
*sem
;
55 sem
= (struct semaphore
*)param
;
57 down_interruptible(sem
);
61 if (unlikely(chain
> 1)) {
64 printk(KERN_WARNING
"%s: Code does not support more "
65 "than two chained crws; please report to "
66 "linux390@de.ibm.com!\n", __FUNCTION__
);
67 ccode
= stcrw(&tmp_crw
);
68 printk(KERN_WARNING
"%s: crw reports slct=%d, oflw=%d, "
69 "chn=%d, rsc=%X, anc=%d, erc=%X, rsid=%X\n",
70 __FUNCTION__
, tmp_crw
.slct
, tmp_crw
.oflw
,
71 tmp_crw
.chn
, tmp_crw
.rsc
, tmp_crw
.anc
,
72 tmp_crw
.erc
, tmp_crw
.rsid
);
73 printk(KERN_WARNING
"%s: This was crw number %x in the "
74 "chain\n", __FUNCTION__
, chain
);
77 chain
= tmp_crw
.chn
? chain
+ 1 : 0;
80 ccode
= stcrw(&crw
[chain
]);
83 printk(KERN_DEBUG
"crw_info : CRW reports slct=%d, oflw=%d, "
84 "chn=%d, rsc=%X, anc=%d, erc=%X, rsid=%X\n",
85 crw
[chain
].slct
, crw
[chain
].oflw
, crw
[chain
].chn
,
86 crw
[chain
].rsc
, crw
[chain
].anc
, crw
[chain
].erc
,
88 /* Check for overflows. */
89 if (crw
[chain
].oflw
) {
90 pr_debug("%s: crw overflow detected!\n", __FUNCTION__
);
91 css_reiterate_subchannels();
96 switch (crw
[chain
].rsc
) {
98 if (crw
[0].chn
&& !chain
)
100 pr_debug("source is subchannel %04X\n", crw
[0].rsid
);
101 ret
= css_process_crw (crw
[0].rsid
,
102 chain
? crw
[1].rsid
: 0);
106 case CRW_RSC_MONITOR
:
107 pr_debug("source is monitoring facility\n");
110 pr_debug("source is channel path %02X\n", crw
[0].rsid
);
112 * Check for solicited machine checks. These are
113 * created by reset channel path and need not be
114 * reported to the common I/O layer.
116 if (crw
[chain
].slct
) {
117 pr_debug("solicited machine check for "
118 "channel path %02X\n", crw
[0].rsid
);
121 switch (crw
[0].erc
) {
122 case CRW_ERC_IPARM
: /* Path has come. */
123 ret
= chp_process_crw(crw
[0].rsid
, 1);
125 case CRW_ERC_PERRI
: /* Path has gone. */
127 ret
= chp_process_crw(crw
[0].rsid
, 0);
130 pr_debug("Don't know how to handle erc=%x\n",
138 pr_debug("source is configuration-alert facility\n");
141 pr_debug("source is channel subsystem\n");
142 ret
= chsc_process_crw();
147 pr_debug("unknown source\n");
150 /* chain is always 0 or 1 here. */
151 chain
= crw
[chain
].chn
? chain
+ 1 : 0;
154 queue_work(slow_path_wq
, &slow_path_work
);
163 unsigned long long mcck_code
;
166 static DEFINE_PER_CPU(struct mcck_struct
, cpu_mcck
);
169 * Main machine check handler function. Will be called with interrupts enabled
170 * or disabled and machine checks enabled or disabled.
173 s390_handle_mcck(void)
176 struct mcck_struct mcck
;
179 * Disable machine checks and get the current state of accumulated
180 * machine checks. Afterwards delete the old state and enable machine
183 local_irq_save(flags
);
184 local_mcck_disable();
185 mcck
= __get_cpu_var(cpu_mcck
);
186 memset(&__get_cpu_var(cpu_mcck
), 0, sizeof(struct mcck_struct
));
187 clear_thread_flag(TIF_MCCK_PENDING
);
189 local_irq_restore(flags
);
191 if (mcck
.channel_report
)
194 #ifdef CONFIG_MACHCHK_WARNING
196 * The warning may remain for a prolonged period on the bare iron.
197 * (actually till the machine is powered off, or until the problem is gone)
198 * So we just stop listening for the WARNING MCH and prevent continuously
199 * being interrupted. One caveat is however, that we must do this per
200 * processor and cannot use the smp version of ctl_clear_bit().
201 * On VM we only get one interrupt per virtally presented machinecheck.
202 * Though one suffices, we may get one interrupt per (virtual) processor.
204 if (mcck
.warning
) { /* WARNING pending ? */
205 static int mchchk_wng_posted
= 0;
207 * Use single machine clear, as we cannot handle smp right now
209 __ctl_clear_bit(14, 24); /* Disable WARNING MCH */
210 if (xchg(&mchchk_wng_posted
, 1) == 0)
211 kill_proc(1, SIGPWR
, 1);
215 if (mcck
.kill_task
) {
217 printk(KERN_EMERG
"mcck: Terminating task because of machine "
218 "malfunction (code 0x%016llx).\n", mcck
.mcck_code
);
219 printk(KERN_EMERG
"mcck: task: %s, pid: %d.\n",
220 current
->comm
, current
->pid
);
226 * returns 0 if all registers could be validated
227 * returns 1 otherwise
230 s390_revalidate_registers(struct mci
*mci
)
235 void *fpt_save_area
, *fpt_creg_save_area
;
239 /* General purpose registers */
242 * General purpose registers couldn't be restored and have
243 * unknown contents. Process needs to be terminated.
247 /* Revalidate floating point registers */
250 * Floating point registers can't be restored and
251 * therefore the process needs to be terminated.
256 asm volatile("ld 0,0(%0)\n"
260 : : "a" (&S390_lowcore
.floating_pt_save_area
));
263 if (MACHINE_HAS_IEEE
) {
265 fpt_save_area
= &S390_lowcore
.floating_pt_save_area
;
266 fpt_creg_save_area
= &S390_lowcore
.fpt_creg_save_area
;
268 fpt_save_area
= (void *) S390_lowcore
.extended_save_area_addr
;
269 fpt_creg_save_area
= fpt_save_area
+128;
271 /* Floating point control register */
274 * Floating point control register can't be restored.
275 * Task will be terminated.
277 asm volatile ("lfpc 0(%0)" : : "a" (&zero
), "m" (zero
));
284 : : "a" (fpt_creg_save_area
));
286 asm volatile("ld 0,0(%0)\n"
302 : : "a" (fpt_save_area
));
305 /* Revalidate access registers */
306 asm volatile("lam 0,15,0(%0)"
307 : : "a" (&S390_lowcore
.access_regs_save_area
));
310 * Access registers have unknown contents.
315 /* Revalidate control registers */
318 * Control registers have unknown contents.
319 * Can't recover and therefore stopping machine.
321 s390_handle_damage("invalid control registers.");
324 asm volatile("lctlg 0,15,0(%0)"
325 : : "a" (&S390_lowcore
.cregs_save_area
));
327 asm volatile("lctl 0,15,0(%0)"
328 : : "a" (&S390_lowcore
.cregs_save_area
));
332 * We don't even try to revalidate the TOD register, since we simply
333 * can't write something sensible into that register.
338 * See if we can revalidate the TOD programmable register with its
339 * old contents (should be zero) otherwise set it to zero.
342 asm volatile("sr 0,0\n"
349 : : "a" (&S390_lowcore
.tod_progreg_save_area
) : "0", "cc");
352 /* Revalidate clock comparator register */
353 asm volatile ("stck 0(%1)\n"
355 : "=m" (tmpclock
) : "a" (&(tmpclock
)) : "cc", "memory");
357 /* Check if old PSW is valid */
360 * Can't tell if we come from user or kernel mode
361 * -> stopping machine.
363 s390_handle_damage("old psw invalid.");
365 if (!mci
->ms
|| !mci
->pm
|| !mci
->ia
)
371 #define MAX_IPD_COUNT 29
372 #define MAX_IPD_TIME (5 * 60 * USEC_PER_SEC) /* 5 minutes */
375 * machine check handler.
378 s390_do_machine_check(struct pt_regs
*regs
)
380 static DEFINE_SPINLOCK(ipd_lock
);
381 static unsigned long long last_ipd
;
382 static int ipd_count
;
383 unsigned long long tmp
;
385 struct mcck_struct
*mcck
;
390 mci
= (struct mci
*) &S390_lowcore
.mcck_interruption_code
;
391 mcck
= &__get_cpu_var(cpu_mcck
);
392 umode
= user_mode(regs
);
395 /* System damage -> stopping machine */
396 s390_handle_damage("received system damage machine check.");
400 /* Processing backup -> verify if we can survive this */
401 u64 z_mcic
, o_mcic
, t_mcic
;
403 z_mcic
= (1ULL<<63 | 1ULL<<59 | 1ULL<<29);
404 o_mcic
= (1ULL<<43 | 1ULL<<42 | 1ULL<<41 | 1ULL<<40 |
405 1ULL<<36 | 1ULL<<35 | 1ULL<<34 | 1ULL<<32 |
406 1ULL<<30 | 1ULL<<21 | 1ULL<<20 | 1ULL<<17 |
409 z_mcic
= (1ULL<<63 | 1ULL<<59 | 1ULL<<57 | 1ULL<<50 |
411 o_mcic
= (1ULL<<43 | 1ULL<<42 | 1ULL<<41 | 1ULL<<40 |
412 1ULL<<36 | 1ULL<<35 | 1ULL<<34 | 1ULL<<32 |
413 1ULL<<30 | 1ULL<<20 | 1ULL<<17 | 1ULL<<16);
415 t_mcic
= *(u64
*)mci
;
417 if (((t_mcic
& z_mcic
) != 0) ||
418 ((t_mcic
& o_mcic
) != o_mcic
)) {
419 s390_handle_damage("processing backup machine "
420 "check with damage.");
424 * Nullifying exigent condition, therefore we might
425 * retry this instruction.
428 spin_lock(&ipd_lock
);
432 if (((tmp
- last_ipd
) >> 12) < MAX_IPD_TIME
)
439 if (ipd_count
== MAX_IPD_COUNT
)
440 s390_handle_damage("too many ipd retries.");
442 spin_unlock(&ipd_lock
);
445 /* Processing damage -> stopping machine */
446 s390_handle_damage("received instruction processing "
447 "damage machine check.");
450 if (s390_revalidate_registers(mci
)) {
453 * Couldn't restore all register contents while in
454 * user mode -> mark task for termination.
457 mcck
->mcck_code
= *(unsigned long long *) mci
;
458 set_thread_flag(TIF_MCCK_PENDING
);
462 * Couldn't restore all register contents while in
463 * kernel mode -> stopping machine.
465 s390_handle_damage("unable to revalidate registers.");
469 /* Storage error uncorrected */
470 s390_handle_damage("received storage error uncorrected "
474 /* Storage key-error uncorrected */
475 s390_handle_damage("received storage key-error uncorrected "
478 if (mci
->ds
&& mci
->fa
)
479 /* Storage degradation */
480 s390_handle_damage("received storage degradation machine "
484 /* Channel report word pending */
485 mcck
->channel_report
= 1;
486 set_thread_flag(TIF_MCCK_PENDING
);
490 /* Warning pending */
492 set_thread_flag(TIF_MCCK_PENDING
);
498 * s390_init_machine_check
500 * initialize machine check handling
503 machine_check_init(void)
505 init_MUTEX_LOCKED(&m_sem
);
506 ctl_clear_bit(14, 25); /* disable external damage MCH */
507 ctl_set_bit(14, 27); /* enable system recovery MCH */
508 #ifdef CONFIG_MACHCHK_WARNING
509 ctl_set_bit(14, 24); /* enable warning MCH */
515 * Initialize the machine check handler really early to be able to
516 * catch all machine checks that happen during boot
518 arch_initcall(machine_check_init
);
521 * Machine checks for the channel subsystem must be enabled
522 * after the channel subsystem is initialized
525 machine_check_crw_init (void)
527 kthread_run(s390_collect_crw_info
, &m_sem
, "kmcheck");
528 ctl_set_bit(14, 28); /* enable channel report MCH */
532 device_initcall (machine_check_crw_init
);