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>
23 #include <asm/switch_to.h>
24 #include <asm/fpu-internal.h>
25 #include <asm/ctl_reg.h>
31 unsigned long long mcck_code
;
34 static DEFINE_PER_CPU(struct mcck_struct
, cpu_mcck
);
36 static void s390_handle_damage(char *msg
)
39 disabled_wait((unsigned long) __builtin_return_address(0));
44 * Main machine check handler function. Will be called with interrupts enabled
45 * or disabled and machine checks enabled or disabled.
47 void s390_handle_mcck(void)
50 struct mcck_struct mcck
;
53 * Disable machine checks and get the current state of accumulated
54 * machine checks. Afterwards delete the old state and enable machine
57 local_irq_save(flags
);
59 mcck
= *this_cpu_ptr(&cpu_mcck
);
60 memset(this_cpu_ptr(&cpu_mcck
), 0, sizeof(mcck
));
61 clear_cpu_flag(CIF_MCCK_PENDING
);
63 local_irq_restore(flags
);
65 if (mcck
.channel_report
)
66 crw_handle_channel_report();
68 * A warning may remain for a prolonged period on the bare iron.
69 * (actually until the machine is powered off, or the problem is gone)
70 * So we just stop listening for the WARNING MCH and avoid continuously
71 * being interrupted. One caveat is however, that we must do this per
72 * processor and cannot use the smp version of ctl_clear_bit().
73 * On VM we only get one interrupt per virtally presented machinecheck.
74 * Though one suffices, we may get one interrupt per (virtual) cpu.
76 if (mcck
.warning
) { /* WARNING pending ? */
77 static int mchchk_wng_posted
= 0;
79 /* Use single cpu clear, as we cannot handle smp here. */
80 __ctl_clear_bit(14, 24); /* Disable WARNING MCH */
81 if (xchg(&mchchk_wng_posted
, 1) == 0)
82 kill_cad_pid(SIGPWR
, 1);
86 printk(KERN_EMERG
"mcck: Terminating task because of machine "
87 "malfunction (code 0x%016llx).\n", mcck
.mcck_code
);
88 printk(KERN_EMERG
"mcck: task: %s, pid: %d.\n",
89 current
->comm
, current
->pid
);
93 EXPORT_SYMBOL_GPL(s390_handle_mcck
);
96 * returns 0 if all registers could be validated
99 static int notrace
s390_revalidate_registers(struct mci
*mci
)
103 void *fpt_save_area
, *fpt_creg_save_area
;
110 * General purpose registers couldn't be restored and have
111 * unknown contents. Process needs to be terminated.
117 * Floating point registers can't be restored and
118 * therefore the process needs to be terminated.
122 fpt_save_area
= &S390_lowcore
.floating_pt_save_area
;
123 fpt_creg_save_area
= &S390_lowcore
.fpt_creg_save_area
;
126 * Floating point control register can't be restored.
127 * Task will be terminated.
129 asm volatile("lfpc 0(%0)" : : "a" (&zero
), "m" (zero
));
132 asm volatile("lfpc 0(%0)" : : "a" (fpt_creg_save_area
));
134 if (!MACHINE_HAS_VX
) {
135 /* Revalidate floating point registers */
153 : : "a" (fpt_save_area
));
155 /* Revalidate vector registers */
160 * Vector registers can't be restored and therefore
161 * the process needs to be terminated.
165 cr0
.val
= S390_lowcore
.cregs_save_area
[0];
166 cr0
.afp
= cr0
.vx
= 1;
167 __ctl_load(cr0
.val
, 0, 0);
170 " .word 0xe70f,0x1000,0x0036\n" /* vlm 0,15,0(1) */
171 " .word 0xe70f,0x1100,0x0c36\n" /* vlm 16,31,256(1) */
172 : : "Q" (*(struct vx_array
*)
173 &S390_lowcore
.vector_save_area
) : "1");
174 __ctl_load(S390_lowcore
.cregs_save_area
[0], 0, 0);
176 /* Revalidate access registers */
179 : : "a" (&S390_lowcore
.access_regs_save_area
));
182 * Access registers have unknown contents.
187 /* Revalidate control registers */
190 * Control registers have unknown contents.
191 * Can't recover and therefore stopping machine.
193 s390_handle_damage("invalid control registers.");
197 : : "a" (&S390_lowcore
.cregs_save_area
));
200 * We don't even try to revalidate the TOD register, since we simply
201 * can't write something sensible into that register.
204 * See if we can revalidate the TOD programmable register with its
205 * old contents (should be zero) otherwise set it to zero.
216 : : "a" (&S390_lowcore
.tod_progreg_save_area
)
218 /* Revalidate clock comparator register */
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
;
256 inc_irq_stat(NMI_NMI
);
257 mci
= (struct mci
*) &S390_lowcore
.mcck_interruption_code
;
258 mcck
= this_cpu_ptr(&cpu_mcck
);
259 umode
= user_mode(regs
);
262 /* System damage -> stopping machine */
263 s390_handle_damage("received system damage machine check.");
267 /* Processing backup -> verify if we can survive this */
268 u64 z_mcic
, o_mcic
, t_mcic
;
269 z_mcic
= (1ULL<<63 | 1ULL<<59 | 1ULL<<29);
270 o_mcic
= (1ULL<<43 | 1ULL<<42 | 1ULL<<41 | 1ULL<<40 |
271 1ULL<<36 | 1ULL<<35 | 1ULL<<34 | 1ULL<<32 |
272 1ULL<<30 | 1ULL<<21 | 1ULL<<20 | 1ULL<<17 |
274 t_mcic
= *(u64
*)mci
;
276 if (((t_mcic
& z_mcic
) != 0) ||
277 ((t_mcic
& o_mcic
) != o_mcic
)) {
278 s390_handle_damage("processing backup machine "
279 "check with damage.");
283 * Nullifying exigent condition, therefore we might
284 * retry this instruction.
286 spin_lock(&ipd_lock
);
287 tmp
= get_tod_clock();
288 if (((tmp
- last_ipd
) >> 12) < MAX_IPD_TIME
)
293 if (ipd_count
== MAX_IPD_COUNT
)
294 s390_handle_damage("too many ipd retries.");
295 spin_unlock(&ipd_lock
);
297 /* Processing damage -> stopping machine */
298 s390_handle_damage("received instruction processing "
299 "damage machine check.");
302 if (s390_revalidate_registers(mci
)) {
305 * Couldn't restore all register contents while in
306 * user mode -> mark task for termination.
309 mcck
->mcck_code
= *(unsigned long long *) mci
;
310 set_cpu_flag(CIF_MCCK_PENDING
);
313 * Couldn't restore all register contents while in
314 * kernel mode -> stopping machine.
316 s390_handle_damage("unable to revalidate registers.");
320 /* Timing facility damage */
321 s390_handle_damage("TOD clock damaged");
323 if (mci
->ed
&& mci
->ec
) {
324 /* External damage */
325 if (S390_lowcore
.external_damage_code
& (1U << ED_ETR_SYNC
))
327 if (S390_lowcore
.external_damage_code
& (1U << ED_ETR_SWITCH
))
328 etr_switch_to_local();
329 if (S390_lowcore
.external_damage_code
& (1U << ED_STP_SYNC
))
331 if (S390_lowcore
.external_damage_code
& (1U << ED_STP_ISLAND
))
335 /* Storage error uncorrected */
336 s390_handle_damage("received storage error uncorrected "
339 /* Storage key-error uncorrected */
340 s390_handle_damage("received storage key-error uncorrected "
342 if (mci
->ds
&& mci
->fa
)
343 /* Storage degradation */
344 s390_handle_damage("received storage degradation machine "
347 /* Channel report word pending */
348 mcck
->channel_report
= 1;
349 set_cpu_flag(CIF_MCCK_PENDING
);
352 /* Warning pending */
354 set_cpu_flag(CIF_MCCK_PENDING
);
359 static int __init
machine_check_init(void)
361 ctl_set_bit(14, 25); /* enable external damage MCH */
362 ctl_set_bit(14, 27); /* enable system recovery MCH */
363 ctl_set_bit(14, 24); /* enable warning MCH */
366 early_initcall(machine_check_init
);