2 * Machine check handler
4 * Copyright IBM Corp. 2000,2009
5 * Author(s): Ingo Adlung <adlung@de.ibm.com>,
6 * Martin Schwidefsky <schwidefsky@de.ibm.com>,
7 * Cornelia Huck <cornelia.huck@de.ibm.com>,
8 * Heiko Carstens <heiko.carstens@de.ibm.com>,
11 #include <linux/kernel_stat.h>
12 #include <linux/init.h>
13 #include <linux/errno.h>
14 #include <linux/hardirq.h>
15 #include <linux/time.h>
16 #include <linux/module.h>
17 #include <asm/lowcore.h>
20 #include <asm/cputime.h>
28 unsigned long long mcck_code
;
31 static DEFINE_PER_CPU(struct mcck_struct
, cpu_mcck
);
33 static NORET_TYPE
void s390_handle_damage(char *msg
)
36 disabled_wait((unsigned long) __builtin_return_address(0));
41 * Main machine check handler function. Will be called with interrupts enabled
42 * or disabled and machine checks enabled or disabled.
44 void s390_handle_mcck(void)
47 struct mcck_struct mcck
;
50 * Disable machine checks and get the current state of accumulated
51 * machine checks. Afterwards delete the old state and enable machine
54 local_irq_save(flags
);
56 mcck
= __get_cpu_var(cpu_mcck
);
57 memset(&__get_cpu_var(cpu_mcck
), 0, sizeof(struct mcck_struct
));
58 clear_thread_flag(TIF_MCCK_PENDING
);
60 local_irq_restore(flags
);
62 if (mcck
.channel_report
)
63 crw_handle_channel_report();
65 * A warning may remain for a prolonged period on the bare iron.
66 * (actually until the machine is powered off, or the problem is gone)
67 * So we just stop listening for the WARNING MCH and avoid continuously
68 * being interrupted. One caveat is however, that we must do this per
69 * processor and cannot use the smp version of ctl_clear_bit().
70 * On VM we only get one interrupt per virtally presented machinecheck.
71 * Though one suffices, we may get one interrupt per (virtual) cpu.
73 if (mcck
.warning
) { /* WARNING pending ? */
74 static int mchchk_wng_posted
= 0;
76 /* Use single cpu clear, as we cannot handle smp here. */
77 __ctl_clear_bit(14, 24); /* Disable WARNING MCH */
78 if (xchg(&mchchk_wng_posted
, 1) == 0)
79 kill_cad_pid(SIGPWR
, 1);
83 printk(KERN_EMERG
"mcck: Terminating task because of machine "
84 "malfunction (code 0x%016llx).\n", mcck
.mcck_code
);
85 printk(KERN_EMERG
"mcck: task: %s, pid: %d.\n",
86 current
->comm
, current
->pid
);
90 EXPORT_SYMBOL_GPL(s390_handle_mcck
);
93 * returns 0 if all registers could be validated
96 static int notrace
s390_revalidate_registers(struct mci
*mci
)
100 void *fpt_save_area
, *fpt_creg_save_area
;
107 * General purpose registers couldn't be restored and have
108 * unknown contents. Process needs to be terminated.
114 * Floating point registers can't be restored and
115 * therefore the process needs to be terminated.
125 : : "a" (&S390_lowcore
.floating_pt_save_area
));
128 if (MACHINE_HAS_IEEE
) {
130 fpt_save_area
= &S390_lowcore
.floating_pt_save_area
;
131 fpt_creg_save_area
= &S390_lowcore
.fpt_creg_save_area
;
133 fpt_save_area
= (void *) S390_lowcore
.extended_save_area_addr
;
134 fpt_creg_save_area
= fpt_save_area
+ 128;
138 * Floating point control register can't be restored.
139 * Task will be terminated.
141 asm volatile("lfpc 0(%0)" : : "a" (&zero
), "m" (zero
));
145 asm volatile("lfpc 0(%0)" : : "a" (fpt_creg_save_area
));
164 : : "a" (fpt_save_area
));
166 /* Revalidate access registers */
169 : : "a" (&S390_lowcore
.access_regs_save_area
));
172 * Access registers have unknown contents.
177 /* Revalidate control registers */
180 * Control registers have unknown contents.
181 * Can't recover and therefore stopping machine.
183 s390_handle_damage("invalid control registers.");
188 : : "a" (&S390_lowcore
.cregs_save_area
));
192 : : "a" (&S390_lowcore
.cregs_save_area
));
196 * We don't even try to revalidate the TOD register, since we simply
197 * can't write something sensible into that register.
201 * See if we can revalidate the TOD programmable register with its
202 * old contents (should be zero) otherwise set it to zero.
213 : : "a" (&S390_lowcore
.tod_progreg_save_area
)
216 /* Revalidate clock comparator register */
217 if (S390_lowcore
.clock_comparator
== -1)
218 set_clock_comparator(S390_lowcore
.mcck_clock
);
220 set_clock_comparator(S390_lowcore
.clock_comparator
);
221 /* Check if old PSW is valid */
224 * Can't tell if we come from user or kernel mode
225 * -> stopping machine.
227 s390_handle_damage("old psw invalid.");
229 if (!mci
->ms
|| !mci
->pm
|| !mci
->ia
)
235 #define MAX_IPD_COUNT 29
236 #define MAX_IPD_TIME (5 * 60 * USEC_PER_SEC) /* 5 minutes */
238 #define ED_STP_ISLAND 6 /* External damage STP island check */
239 #define ED_STP_SYNC 7 /* External damage STP sync check */
240 #define ED_ETR_SYNC 12 /* External damage ETR sync check */
241 #define ED_ETR_SWITCH 13 /* External damage ETR switch to local */
244 * machine check handler.
246 void notrace
s390_do_machine_check(struct pt_regs
*regs
)
248 static int ipd_count
;
249 static DEFINE_SPINLOCK(ipd_lock
);
250 static unsigned long long last_ipd
;
251 struct mcck_struct
*mcck
;
252 unsigned long long tmp
;
257 s390_idle_check(regs
, S390_lowcore
.mcck_clock
,
258 S390_lowcore
.mcck_enter_timer
);
259 kstat_cpu(smp_processor_id()).irqs
[NMI_NMI
]++;
260 mci
= (struct mci
*) &S390_lowcore
.mcck_interruption_code
;
261 mcck
= &__get_cpu_var(cpu_mcck
);
262 umode
= user_mode(regs
);
265 /* System damage -> stopping machine */
266 s390_handle_damage("received system damage machine check.");
270 /* Processing backup -> verify if we can survive this */
271 u64 z_mcic
, o_mcic
, t_mcic
;
273 z_mcic
= (1ULL<<63 | 1ULL<<59 | 1ULL<<29);
274 o_mcic
= (1ULL<<43 | 1ULL<<42 | 1ULL<<41 | 1ULL<<40 |
275 1ULL<<36 | 1ULL<<35 | 1ULL<<34 | 1ULL<<32 |
276 1ULL<<30 | 1ULL<<21 | 1ULL<<20 | 1ULL<<17 |
279 z_mcic
= (1ULL<<63 | 1ULL<<59 | 1ULL<<57 | 1ULL<<50 |
281 o_mcic
= (1ULL<<43 | 1ULL<<42 | 1ULL<<41 | 1ULL<<40 |
282 1ULL<<36 | 1ULL<<35 | 1ULL<<34 | 1ULL<<32 |
283 1ULL<<30 | 1ULL<<20 | 1ULL<<17 | 1ULL<<16);
285 t_mcic
= *(u64
*)mci
;
287 if (((t_mcic
& z_mcic
) != 0) ||
288 ((t_mcic
& o_mcic
) != o_mcic
)) {
289 s390_handle_damage("processing backup machine "
290 "check with damage.");
294 * Nullifying exigent condition, therefore we might
295 * retry this instruction.
297 spin_lock(&ipd_lock
);
299 if (((tmp
- last_ipd
) >> 12) < MAX_IPD_TIME
)
304 if (ipd_count
== MAX_IPD_COUNT
)
305 s390_handle_damage("too many ipd retries.");
306 spin_unlock(&ipd_lock
);
308 /* Processing damage -> stopping machine */
309 s390_handle_damage("received instruction processing "
310 "damage machine check.");
313 if (s390_revalidate_registers(mci
)) {
316 * Couldn't restore all register contents while in
317 * user mode -> mark task for termination.
320 mcck
->mcck_code
= *(unsigned long long *) mci
;
321 set_thread_flag(TIF_MCCK_PENDING
);
324 * Couldn't restore all register contents while in
325 * kernel mode -> stopping machine.
327 s390_handle_damage("unable to revalidate registers.");
331 /* Timing facility damage */
332 s390_handle_damage("TOD clock damaged");
334 if (mci
->ed
&& mci
->ec
) {
335 /* External damage */
336 if (S390_lowcore
.external_damage_code
& (1U << ED_ETR_SYNC
))
338 if (S390_lowcore
.external_damage_code
& (1U << ED_ETR_SWITCH
))
339 etr_switch_to_local();
340 if (S390_lowcore
.external_damage_code
& (1U << ED_STP_SYNC
))
342 if (S390_lowcore
.external_damage_code
& (1U << ED_STP_ISLAND
))
346 /* Storage error uncorrected */
347 s390_handle_damage("received storage error uncorrected "
350 /* Storage key-error uncorrected */
351 s390_handle_damage("received storage key-error uncorrected "
353 if (mci
->ds
&& mci
->fa
)
354 /* Storage degradation */
355 s390_handle_damage("received storage degradation machine "
358 /* Channel report word pending */
359 mcck
->channel_report
= 1;
360 set_thread_flag(TIF_MCCK_PENDING
);
363 /* Warning pending */
365 set_thread_flag(TIF_MCCK_PENDING
);
370 static int __init
machine_check_init(void)
372 ctl_set_bit(14, 25); /* enable external damage MCH */
373 ctl_set_bit(14, 27); /* enable system recovery MCH */
374 ctl_set_bit(14, 24); /* enable warning MCH */
377 arch_initcall(machine_check_init
);