2 * drivers/s390/s390mach.c
3 * S/390 machine check handler
5 * Copyright IBM Corp. 2000,2008
6 * Author(s): Ingo Adlung (adlung@de.ibm.com)
7 * Martin Schwidefsky (schwidefsky@de.ibm.com)
8 * Cornelia Huck <cornelia.huck@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/device.h>
17 #include <linux/kthread.h>
19 #include <asm/lowcore.h>
24 static struct semaphore m_sem;
26 static NORET_TYPE void
27 s390_handle_damage(char *msg)
32 disabled_wait((unsigned long) __builtin_return_address(0));
36 static crw_handler_t crw_handlers[NR_RSCS];
39 * s390_register_crw_handler() - register a channel report word handler
40 * @rsc: reporting source code to handle
41 * @handler: handler to be registered
43 * Returns %0 on success and a negative error value otherwise.
45 int s390_register_crw_handler(int rsc, crw_handler_t handler)
47 if ((rsc < 0) || (rsc >= NR_RSCS))
49 if (!cmpxchg(&crw_handlers[rsc], NULL, handler))
55 * s390_unregister_crw_handler() - unregister a channel report word handler
56 * @rsc: reporting source code to handle
58 void s390_unregister_crw_handler(int rsc)
60 if ((rsc < 0) || (rsc >= NR_RSCS))
62 xchg(&crw_handlers[rsc], NULL);
67 * Retrieve CRWs and call function to handle event.
69 static int s390_collect_crw_info(void *param)
73 struct semaphore *sem;
77 sem = (struct semaphore *)param;
79 ignore = down_interruptible(sem);
82 if (unlikely(chain > 1)) {
85 printk(KERN_WARNING"%s: Code does not support more "
86 "than two chained crws; please report to "
87 "linux390@de.ibm.com!\n", __func__);
88 ccode = stcrw(&tmp_crw);
89 printk(KERN_WARNING"%s: crw reports slct=%d, oflw=%d, "
90 "chn=%d, rsc=%X, anc=%d, erc=%X, rsid=%X\n",
91 __func__, tmp_crw.slct, tmp_crw.oflw,
92 tmp_crw.chn, tmp_crw.rsc, tmp_crw.anc,
93 tmp_crw.erc, tmp_crw.rsid);
94 printk(KERN_WARNING"%s: This was crw number %x in the "
95 "chain\n", __func__, chain);
98 chain = tmp_crw.chn ? chain + 1 : 0;
101 ccode = stcrw(&crw[chain]);
104 printk(KERN_DEBUG "crw_info : CRW reports slct=%d, oflw=%d, "
105 "chn=%d, rsc=%X, anc=%d, erc=%X, rsid=%X\n",
106 crw[chain].slct, crw[chain].oflw, crw[chain].chn,
107 crw[chain].rsc, crw[chain].anc, crw[chain].erc,
109 /* Check for overflows. */
110 if (crw[chain].oflw) {
113 pr_debug("%s: crw overflow detected!\n", __func__);
114 for (i = 0; i < NR_RSCS; i++) {
116 crw_handlers[i](NULL, NULL, 1);
121 if (crw[0].chn && !chain) {
125 if (crw_handlers[crw[chain].rsc])
126 crw_handlers[crw[chain].rsc](&crw[0],
127 chain ? &crw[1] : NULL,
129 /* chain is always 0 or 1 here. */
130 chain = crw[chain].chn ? chain + 1 : 0;
140 unsigned long long mcck_code;
143 static DEFINE_PER_CPU(struct mcck_struct, cpu_mcck);
146 * Main machine check handler function. Will be called with interrupts enabled
147 * or disabled and machine checks enabled or disabled.
150 s390_handle_mcck(void)
153 struct mcck_struct mcck;
156 * Disable machine checks and get the current state of accumulated
157 * machine checks. Afterwards delete the old state and enable machine
160 local_irq_save(flags);
161 local_mcck_disable();
162 mcck = __get_cpu_var(cpu_mcck);
163 memset(&__get_cpu_var(cpu_mcck), 0, sizeof(struct mcck_struct));
164 clear_thread_flag(TIF_MCCK_PENDING);
166 local_irq_restore(flags);
168 if (mcck.channel_report)
171 #ifdef CONFIG_MACHCHK_WARNING
173 * The warning may remain for a prolonged period on the bare iron.
174 * (actually till the machine is powered off, or until the problem is gone)
175 * So we just stop listening for the WARNING MCH and prevent continuously
176 * being interrupted. One caveat is however, that we must do this per
177 * processor and cannot use the smp version of ctl_clear_bit().
178 * On VM we only get one interrupt per virtally presented machinecheck.
179 * Though one suffices, we may get one interrupt per (virtual) processor.
181 if (mcck.warning) { /* WARNING pending ? */
182 static int mchchk_wng_posted = 0;
184 * Use single machine clear, as we cannot handle smp right now
186 __ctl_clear_bit(14, 24); /* Disable WARNING MCH */
187 if (xchg(&mchchk_wng_posted, 1) == 0)
188 kill_cad_pid(SIGPWR, 1);
192 if (mcck.kill_task) {
194 printk(KERN_EMERG "mcck: Terminating task because of machine "
195 "malfunction (code 0x%016llx).\n", mcck.mcck_code);
196 printk(KERN_EMERG "mcck: task: %s, pid: %d.\n",
197 current->comm, current->pid);
201 EXPORT_SYMBOL_GPL(s390_handle_mcck);
204 * returns 0 if all registers could be validated
205 * returns 1 otherwise
208 s390_revalidate_registers(struct mci *mci)
213 void *fpt_save_area, *fpt_creg_save_area;
217 /* General purpose registers */
220 * General purpose registers couldn't be restored and have
221 * unknown contents. Process needs to be terminated.
225 /* Revalidate floating point registers */
228 * Floating point registers can't be restored and
229 * therefore the process needs to be terminated.
239 : : "a" (&S390_lowcore.floating_pt_save_area));
242 if (MACHINE_HAS_IEEE) {
244 fpt_save_area = &S390_lowcore.floating_pt_save_area;
245 fpt_creg_save_area = &S390_lowcore.fpt_creg_save_area;
247 fpt_save_area = (void *) S390_lowcore.extended_save_area_addr;
248 fpt_creg_save_area = fpt_save_area+128;
250 /* Floating point control register */
253 * Floating point control register can't be restored.
254 * Task will be terminated.
256 asm volatile("lfpc 0(%0)" : : "a" (&zero), "m" (zero));
260 asm volatile("lfpc 0(%0)" : : "a" (fpt_creg_save_area));
279 : : "a" (fpt_save_area));
282 /* Revalidate access registers */
285 : : "a" (&S390_lowcore.access_regs_save_area));
288 * Access registers have unknown contents.
293 /* Revalidate control registers */
296 * Control registers have unknown contents.
297 * Can't recover and therefore stopping machine.
299 s390_handle_damage("invalid control registers.");
304 : : "a" (&S390_lowcore.cregs_save_area));
308 : : "a" (&S390_lowcore.cregs_save_area));
312 * We don't even try to revalidate the TOD register, since we simply
313 * can't write something sensible into that register.
318 * See if we can revalidate the TOD programmable register with its
319 * old contents (should be zero) otherwise set it to zero.
330 : : "a" (&S390_lowcore.tod_progreg_save_area)
334 /* Revalidate clock comparator register */
338 : "=m" (tmpclock) : "a" (&(tmpclock)) : "cc", "memory");
340 /* Check if old PSW is valid */
343 * Can't tell if we come from user or kernel mode
344 * -> stopping machine.
346 s390_handle_damage("old psw invalid.");
348 if (!mci->ms || !mci->pm || !mci->ia)
354 #define MAX_IPD_COUNT 29
355 #define MAX_IPD_TIME (5 * 60 * USEC_PER_SEC) /* 5 minutes */
358 * machine check handler.
360 void notrace s390_do_machine_check(struct pt_regs *regs)
362 static DEFINE_SPINLOCK(ipd_lock);
363 static unsigned long long last_ipd;
364 static int ipd_count;
365 unsigned long long tmp;
367 struct mcck_struct *mcck;
374 mci = (struct mci *) &S390_lowcore.mcck_interruption_code;
375 mcck = &__get_cpu_var(cpu_mcck);
376 umode = user_mode(regs);
379 /* System damage -> stopping machine */
380 s390_handle_damage("received system damage machine check.");
384 /* Processing backup -> verify if we can survive this */
385 u64 z_mcic, o_mcic, t_mcic;
387 z_mcic = (1ULL<<63 | 1ULL<<59 | 1ULL<<29);
388 o_mcic = (1ULL<<43 | 1ULL<<42 | 1ULL<<41 | 1ULL<<40 |
389 1ULL<<36 | 1ULL<<35 | 1ULL<<34 | 1ULL<<32 |
390 1ULL<<30 | 1ULL<<21 | 1ULL<<20 | 1ULL<<17 |
393 z_mcic = (1ULL<<63 | 1ULL<<59 | 1ULL<<57 | 1ULL<<50 |
395 o_mcic = (1ULL<<43 | 1ULL<<42 | 1ULL<<41 | 1ULL<<40 |
396 1ULL<<36 | 1ULL<<35 | 1ULL<<34 | 1ULL<<32 |
397 1ULL<<30 | 1ULL<<20 | 1ULL<<17 | 1ULL<<16);
399 t_mcic = *(u64 *)mci;
401 if (((t_mcic & z_mcic) != 0) ||
402 ((t_mcic & o_mcic) != o_mcic)) {
403 s390_handle_damage("processing backup machine "
404 "check with damage.");
408 * Nullifying exigent condition, therefore we might
409 * retry this instruction.
412 spin_lock(&ipd_lock);
416 if (((tmp - last_ipd) >> 12) < MAX_IPD_TIME)
423 if (ipd_count == MAX_IPD_COUNT)
424 s390_handle_damage("too many ipd retries.");
426 spin_unlock(&ipd_lock);
429 /* Processing damage -> stopping machine */
430 s390_handle_damage("received instruction processing "
431 "damage machine check.");
434 if (s390_revalidate_registers(mci)) {
437 * Couldn't restore all register contents while in
438 * user mode -> mark task for termination.
441 mcck->mcck_code = *(unsigned long long *) mci;
442 set_thread_flag(TIF_MCCK_PENDING);
446 * Couldn't restore all register contents while in
447 * kernel mode -> stopping machine.
449 s390_handle_damage("unable to revalidate registers.");
453 /* Timing facility damage */
454 s390_handle_damage("TOD clock damaged");
457 if (mci->ed && mci->ec) {
458 /* External damage */
459 if (S390_lowcore.external_damage_code & (1U << ED_ETR_SYNC))
461 if (S390_lowcore.external_damage_code & (1U << ED_ETR_SWITCH))
462 etr_switch_to_local();
463 if (S390_lowcore.external_damage_code & (1U << ED_STP_SYNC))
465 if (S390_lowcore.external_damage_code & (1U << ED_STP_ISLAND))
470 /* Storage error uncorrected */
471 s390_handle_damage("received storage error uncorrected "
475 /* Storage key-error uncorrected */
476 s390_handle_damage("received storage key-error uncorrected "
479 if (mci->ds && mci->fa)
480 /* Storage degradation */
481 s390_handle_damage("received storage degradation machine "
485 /* Channel report word pending */
486 mcck->channel_report = 1;
487 set_thread_flag(TIF_MCCK_PENDING);
491 /* Warning pending */
493 set_thread_flag(TIF_MCCK_PENDING);
499 * s390_init_machine_check
501 * initialize machine check handling
504 machine_check_init(void)
506 init_MUTEX_LOCKED(&m_sem);
507 ctl_set_bit(14, 25); /* enable external damage MCH */
508 ctl_set_bit(14, 27); /* enable system recovery MCH */
509 #ifdef CONFIG_MACHCHK_WARNING
510 ctl_set_bit(14, 24); /* enable warning MCH */
516 * Initialize the machine check handler really early to be able to
517 * catch all machine checks that happen during boot
519 arch_initcall(machine_check_init);
522 * Machine checks for the channel subsystem must be enabled
523 * after the channel subsystem is initialized
526 machine_check_crw_init (void)
528 struct task_struct *task;
530 task = kthread_run(s390_collect_crw_info, &m_sem, "kmcheck");
532 return PTR_ERR(task);
533 ctl_set_bit(14, 28); /* enable channel report MCH */
537 device_initcall (machine_check_crw_init);