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/time.h>
14 #include <linux/module.h>
15 #include <asm/lowcore.h>
26 unsigned long long mcck_code;
29 static DEFINE_PER_CPU(struct mcck_struct, cpu_mcck);
31 static NORET_TYPE void s390_handle_damage(char *msg)
34 disabled_wait((unsigned long) __builtin_return_address(0));
39 * Main machine check handler function. Will be called with interrupts enabled
40 * or disabled and machine checks enabled or disabled.
42 void s390_handle_mcck(void)
45 struct mcck_struct mcck;
48 * Disable machine checks and get the current state of accumulated
49 * machine checks. Afterwards delete the old state and enable machine
52 local_irq_save(flags);
54 mcck = __get_cpu_var(cpu_mcck);
55 memset(&__get_cpu_var(cpu_mcck), 0, sizeof(struct mcck_struct));
56 clear_thread_flag(TIF_MCCK_PENDING);
58 local_irq_restore(flags);
60 if (mcck.channel_report)
61 crw_handle_channel_report();
63 #ifdef CONFIG_MACHCHK_WARNING
65 * The warning may remain for a prolonged period on the bare iron.
66 * (actually till the machine is powered off, or until the problem is gone)
67 * So we just stop listening for the WARNING MCH and prevent 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) processor.
73 if (mcck.warning) { /* WARNING pending ? */
74 static int mchchk_wng_posted = 0;
76 * Use single machine clear, as we cannot handle smp right now
78 __ctl_clear_bit(14, 24); /* Disable WARNING MCH */
79 if (xchg(&mchchk_wng_posted, 1) == 0)
80 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)
104 void *fpt_save_area, *fpt_creg_save_area;
111 * General purpose registers couldn't be restored and have
112 * unknown contents. Process needs to be terminated.
118 * Floating point registers can't be restored and
119 * therefore the process needs to be terminated.
129 : : "a" (&S390_lowcore.floating_pt_save_area));
132 if (MACHINE_HAS_IEEE) {
134 fpt_save_area = &S390_lowcore.floating_pt_save_area;
135 fpt_creg_save_area = &S390_lowcore.fpt_creg_save_area;
137 fpt_save_area = (void *) S390_lowcore.extended_save_area_addr;
138 fpt_creg_save_area = fpt_save_area + 128;
142 * Floating point control register can't be restored.
143 * Task will be terminated.
145 asm volatile("lfpc 0(%0)" : : "a" (&zero), "m" (zero));
149 asm volatile("lfpc 0(%0)" : : "a" (fpt_creg_save_area));
168 : : "a" (fpt_save_area));
170 /* Revalidate access registers */
173 : : "a" (&S390_lowcore.access_regs_save_area));
176 * Access registers have unknown contents.
181 /* Revalidate control registers */
184 * Control registers have unknown contents.
185 * Can't recover and therefore stopping machine.
187 s390_handle_damage("invalid control registers.");
192 : : "a" (&S390_lowcore.cregs_save_area));
196 : : "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.
205 * See if we can revalidate the TOD programmable register with its
206 * old contents (should be zero) otherwise set it to zero.
217 : : "a" (&S390_lowcore.tod_progreg_save_area)
220 /* Revalidate clock comparator register */
224 : "=m" (tmpclock) : "a" (&(tmpclock)) : "cc", "memory");
226 /* Check if old PSW is valid */
229 * Can't tell if we come from user or kernel mode
230 * -> stopping machine.
232 s390_handle_damage("old psw invalid.");
234 if (!mci->ms || !mci->pm || !mci->ia)
240 #define MAX_IPD_COUNT 29
241 #define MAX_IPD_TIME (5 * 60 * USEC_PER_SEC) /* 5 minutes */
243 #define ED_STP_ISLAND 6 /* External damage STP island check */
244 #define ED_STP_SYNC 7 /* External damage STP sync check */
245 #define ED_ETR_SYNC 12 /* External damage ETR sync check */
246 #define ED_ETR_SWITCH 13 /* External damage ETR switch to local */
249 * machine check handler.
251 void notrace s390_do_machine_check(struct pt_regs *regs)
253 static int ipd_count;
254 static DEFINE_SPINLOCK(ipd_lock);
255 static unsigned long long last_ipd;
256 struct mcck_struct *mcck;
257 unsigned long long tmp;
264 mci = (struct mci *) &S390_lowcore.mcck_interruption_code;
265 mcck = &__get_cpu_var(cpu_mcck);
266 umode = user_mode(regs);
269 /* System damage -> stopping machine */
270 s390_handle_damage("received system damage machine check.");
274 /* Processing backup -> verify if we can survive this */
275 u64 z_mcic, o_mcic, t_mcic;
277 z_mcic = (1ULL<<63 | 1ULL<<59 | 1ULL<<29);
278 o_mcic = (1ULL<<43 | 1ULL<<42 | 1ULL<<41 | 1ULL<<40 |
279 1ULL<<36 | 1ULL<<35 | 1ULL<<34 | 1ULL<<32 |
280 1ULL<<30 | 1ULL<<21 | 1ULL<<20 | 1ULL<<17 |
283 z_mcic = (1ULL<<63 | 1ULL<<59 | 1ULL<<57 | 1ULL<<50 |
285 o_mcic = (1ULL<<43 | 1ULL<<42 | 1ULL<<41 | 1ULL<<40 |
286 1ULL<<36 | 1ULL<<35 | 1ULL<<34 | 1ULL<<32 |
287 1ULL<<30 | 1ULL<<20 | 1ULL<<17 | 1ULL<<16);
289 t_mcic = *(u64 *)mci;
291 if (((t_mcic & z_mcic) != 0) ||
292 ((t_mcic & o_mcic) != o_mcic)) {
293 s390_handle_damage("processing backup machine "
294 "check with damage.");
298 * Nullifying exigent condition, therefore we might
299 * retry this instruction.
301 spin_lock(&ipd_lock);
303 if (((tmp - last_ipd) >> 12) < MAX_IPD_TIME)
308 if (ipd_count == MAX_IPD_COUNT)
309 s390_handle_damage("too many ipd retries.");
310 spin_unlock(&ipd_lock);
312 /* Processing damage -> stopping machine */
313 s390_handle_damage("received instruction processing "
314 "damage machine check.");
317 if (s390_revalidate_registers(mci)) {
320 * Couldn't restore all register contents while in
321 * user mode -> mark task for termination.
324 mcck->mcck_code = *(unsigned long long *) mci;
325 set_thread_flag(TIF_MCCK_PENDING);
328 * Couldn't restore all register contents while in
329 * kernel mode -> stopping machine.
331 s390_handle_damage("unable to revalidate registers.");
335 /* Timing facility damage */
336 s390_handle_damage("TOD clock damaged");
338 if (mci->ed && mci->ec) {
339 /* External damage */
340 if (S390_lowcore.external_damage_code & (1U << ED_ETR_SYNC))
342 if (S390_lowcore.external_damage_code & (1U << ED_ETR_SWITCH))
343 etr_switch_to_local();
344 if (S390_lowcore.external_damage_code & (1U << ED_STP_SYNC))
346 if (S390_lowcore.external_damage_code & (1U << ED_STP_ISLAND))
350 /* Storage error uncorrected */
351 s390_handle_damage("received storage error uncorrected "
354 /* Storage key-error uncorrected */
355 s390_handle_damage("received storage key-error uncorrected "
357 if (mci->ds && mci->fa)
358 /* Storage degradation */
359 s390_handle_damage("received storage degradation machine "
362 /* Channel report word pending */
363 mcck->channel_report = 1;
364 set_thread_flag(TIF_MCCK_PENDING);
367 /* Warning pending */
369 set_thread_flag(TIF_MCCK_PENDING);
374 static int __init machine_check_init(void)
376 ctl_set_bit(14, 25); /* enable external damage MCH */
377 ctl_set_bit(14, 27); /* enable system recovery MCH */
378 #ifdef CONFIG_MACHCHK_WARNING
379 ctl_set_bit(14, 24); /* enable warning MCH */
383 arch_initcall(machine_check_init);