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/init.h>
12 #include <linux/errno.h>
13 #include <linux/hardirq.h>
14 #include <linux/time.h>
15 #include <linux/module.h>
16 #include <asm/lowcore.h>
19 #include <asm/cputime.h>
27 unsigned long long mcck_code
;
30 static DEFINE_PER_CPU(struct mcck_struct
, cpu_mcck
);
32 static NORET_TYPE
void s390_handle_damage(char *msg
)
35 disabled_wait((unsigned long) __builtin_return_address(0));
40 * Main machine check handler function. Will be called with interrupts enabled
41 * or disabled and machine checks enabled or disabled.
43 void s390_handle_mcck(void)
46 struct mcck_struct mcck
;
49 * Disable machine checks and get the current state of accumulated
50 * machine checks. Afterwards delete the old state and enable machine
53 local_irq_save(flags
);
55 mcck
= __get_cpu_var(cpu_mcck
);
56 memset(&__get_cpu_var(cpu_mcck
), 0, sizeof(struct mcck_struct
));
57 clear_thread_flag(TIF_MCCK_PENDING
);
59 local_irq_restore(flags
);
61 if (mcck
.channel_report
)
62 crw_handle_channel_report();
64 * A warning may remain for a prolonged period on the bare iron.
65 * (actually until the machine is powered off, or the problem is gone)
66 * So we just stop listening for the WARNING MCH and avoid continuously
67 * being interrupted. One caveat is however, that we must do this per
68 * processor and cannot use the smp version of ctl_clear_bit().
69 * On VM we only get one interrupt per virtally presented machinecheck.
70 * Though one suffices, we may get one interrupt per (virtual) cpu.
72 if (mcck
.warning
) { /* WARNING pending ? */
73 static int mchchk_wng_posted
= 0;
75 /* Use single cpu clear, as we cannot handle smp here. */
76 __ctl_clear_bit(14, 24); /* Disable WARNING MCH */
77 if (xchg(&mchchk_wng_posted
, 1) == 0)
78 kill_cad_pid(SIGPWR
, 1);
82 printk(KERN_EMERG
"mcck: Terminating task because of machine "
83 "malfunction (code 0x%016llx).\n", mcck
.mcck_code
);
84 printk(KERN_EMERG
"mcck: task: %s, pid: %d.\n",
85 current
->comm
, current
->pid
);
89 EXPORT_SYMBOL_GPL(s390_handle_mcck
);
92 * returns 0 if all registers could be validated
95 static int notrace
s390_revalidate_registers(struct mci
*mci
)
99 void *fpt_save_area
, *fpt_creg_save_area
;
106 * General purpose registers couldn't be restored and have
107 * unknown contents. Process needs to be terminated.
113 * Floating point registers can't be restored and
114 * therefore the process needs to be terminated.
124 : : "a" (&S390_lowcore
.floating_pt_save_area
));
127 if (MACHINE_HAS_IEEE
) {
129 fpt_save_area
= &S390_lowcore
.floating_pt_save_area
;
130 fpt_creg_save_area
= &S390_lowcore
.fpt_creg_save_area
;
132 fpt_save_area
= (void *) S390_lowcore
.extended_save_area_addr
;
133 fpt_creg_save_area
= fpt_save_area
+ 128;
137 * Floating point control register can't be restored.
138 * Task will be terminated.
140 asm volatile("lfpc 0(%0)" : : "a" (&zero
), "m" (zero
));
144 asm volatile("lfpc 0(%0)" : : "a" (fpt_creg_save_area
));
163 : : "a" (fpt_save_area
));
165 /* Revalidate access registers */
168 : : "a" (&S390_lowcore
.access_regs_save_area
));
171 * Access registers have unknown contents.
176 /* Revalidate control registers */
179 * Control registers have unknown contents.
180 * Can't recover and therefore stopping machine.
182 s390_handle_damage("invalid control registers.");
187 : : "a" (&S390_lowcore
.cregs_save_area
));
191 : : "a" (&S390_lowcore
.cregs_save_area
));
195 * We don't even try to revalidate the TOD register, since we simply
196 * can't write something sensible into that register.
200 * See if we can revalidate the TOD programmable register with its
201 * old contents (should be zero) otherwise set it to zero.
212 : : "a" (&S390_lowcore
.tod_progreg_save_area
)
215 /* Revalidate clock comparator register */
216 if (S390_lowcore
.clock_comparator
== -1)
217 set_clock_comparator(get_clock());
219 set_clock_comparator(S390_lowcore
.clock_comparator
);
220 /* Check if old PSW is valid */
223 * Can't tell if we come from user or kernel mode
224 * -> stopping machine.
226 s390_handle_damage("old psw invalid.");
228 if (!mci
->ms
|| !mci
->pm
|| !mci
->ia
)
234 #define MAX_IPD_COUNT 29
235 #define MAX_IPD_TIME (5 * 60 * USEC_PER_SEC) /* 5 minutes */
237 #define ED_STP_ISLAND 6 /* External damage STP island check */
238 #define ED_STP_SYNC 7 /* External damage STP sync check */
239 #define ED_ETR_SYNC 12 /* External damage ETR sync check */
240 #define ED_ETR_SWITCH 13 /* External damage ETR switch to local */
243 * machine check handler.
245 void notrace
s390_do_machine_check(struct pt_regs
*regs
)
247 static int ipd_count
;
248 static DEFINE_SPINLOCK(ipd_lock
);
249 static unsigned long long last_ipd
;
250 struct mcck_struct
*mcck
;
251 unsigned long long tmp
;
258 mci
= (struct mci
*) &S390_lowcore
.mcck_interruption_code
;
259 mcck
= &__get_cpu_var(cpu_mcck
);
260 umode
= user_mode(regs
);
263 /* System damage -> stopping machine */
264 s390_handle_damage("received system damage machine check.");
268 /* Processing backup -> verify if we can survive this */
269 u64 z_mcic
, o_mcic
, t_mcic
;
271 z_mcic
= (1ULL<<63 | 1ULL<<59 | 1ULL<<29);
272 o_mcic
= (1ULL<<43 | 1ULL<<42 | 1ULL<<41 | 1ULL<<40 |
273 1ULL<<36 | 1ULL<<35 | 1ULL<<34 | 1ULL<<32 |
274 1ULL<<30 | 1ULL<<21 | 1ULL<<20 | 1ULL<<17 |
277 z_mcic
= (1ULL<<63 | 1ULL<<59 | 1ULL<<57 | 1ULL<<50 |
279 o_mcic
= (1ULL<<43 | 1ULL<<42 | 1ULL<<41 | 1ULL<<40 |
280 1ULL<<36 | 1ULL<<35 | 1ULL<<34 | 1ULL<<32 |
281 1ULL<<30 | 1ULL<<20 | 1ULL<<17 | 1ULL<<16);
283 t_mcic
= *(u64
*)mci
;
285 if (((t_mcic
& z_mcic
) != 0) ||
286 ((t_mcic
& o_mcic
) != o_mcic
)) {
287 s390_handle_damage("processing backup machine "
288 "check with damage.");
292 * Nullifying exigent condition, therefore we might
293 * retry this instruction.
295 spin_lock(&ipd_lock
);
297 if (((tmp
- last_ipd
) >> 12) < MAX_IPD_TIME
)
302 if (ipd_count
== MAX_IPD_COUNT
)
303 s390_handle_damage("too many ipd retries.");
304 spin_unlock(&ipd_lock
);
306 /* Processing damage -> stopping machine */
307 s390_handle_damage("received instruction processing "
308 "damage machine check.");
311 if (s390_revalidate_registers(mci
)) {
314 * Couldn't restore all register contents while in
315 * user mode -> mark task for termination.
318 mcck
->mcck_code
= *(unsigned long long *) mci
;
319 set_thread_flag(TIF_MCCK_PENDING
);
322 * Couldn't restore all register contents while in
323 * kernel mode -> stopping machine.
325 s390_handle_damage("unable to revalidate registers.");
329 /* Timing facility damage */
330 s390_handle_damage("TOD clock damaged");
332 if (mci
->ed
&& mci
->ec
) {
333 /* External damage */
334 if (S390_lowcore
.external_damage_code
& (1U << ED_ETR_SYNC
))
336 if (S390_lowcore
.external_damage_code
& (1U << ED_ETR_SWITCH
))
337 etr_switch_to_local();
338 if (S390_lowcore
.external_damage_code
& (1U << ED_STP_SYNC
))
340 if (S390_lowcore
.external_damage_code
& (1U << ED_STP_ISLAND
))
344 /* Storage error uncorrected */
345 s390_handle_damage("received storage error uncorrected "
348 /* Storage key-error uncorrected */
349 s390_handle_damage("received storage key-error uncorrected "
351 if (mci
->ds
&& mci
->fa
)
352 /* Storage degradation */
353 s390_handle_damage("received storage degradation machine "
356 /* Channel report word pending */
357 mcck
->channel_report
= 1;
358 set_thread_flag(TIF_MCCK_PENDING
);
361 /* Warning pending */
363 set_thread_flag(TIF_MCCK_PENDING
);
368 static int __init
machine_check_init(void)
370 ctl_set_bit(14, 25); /* enable external damage MCH */
371 ctl_set_bit(14, 27); /* enable system recovery MCH */
372 ctl_set_bit(14, 24); /* enable warning MCH */
375 arch_initcall(machine_check_init
);