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 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);
57 * Ummm... Does this make sense at all? Copying the percpu struct
58 * and then zapping it one statement later?
60 memcpy(&mcck, this_cpu_ptr(&cpu_mcck), sizeof(mcck));
61 memset(&mcck, 0, sizeof(struct mcck_struct));
62 clear_cpu_flag(CIF_MCCK_PENDING);
64 local_irq_restore(flags);
66 if (mcck.channel_report)
67 crw_handle_channel_report();
69 * A warning may remain for a prolonged period on the bare iron.
70 * (actually until the machine is powered off, or the problem is gone)
71 * So we just stop listening for the WARNING MCH and avoid continuously
72 * being interrupted. One caveat is however, that we must do this per
73 * processor and cannot use the smp version of ctl_clear_bit().
74 * On VM we only get one interrupt per virtally presented machinecheck.
75 * Though one suffices, we may get one interrupt per (virtual) cpu.
77 if (mcck.warning) { /* WARNING pending ? */
78 static int mchchk_wng_posted = 0;
80 /* Use single cpu clear, as we cannot handle smp here. */
81 __ctl_clear_bit(14, 24); /* Disable WARNING MCH */
82 if (xchg(&mchchk_wng_posted, 1) == 0)
83 kill_cad_pid(SIGPWR, 1);
87 printk(KERN_EMERG "mcck: Terminating task because of machine "
88 "malfunction (code 0x%016llx).\n", mcck.mcck_code);
89 printk(KERN_EMERG "mcck: task: %s, pid: %d.\n",
90 current->comm, current->pid);
94 EXPORT_SYMBOL_GPL(s390_handle_mcck);
97 * returns 0 if all registers could be validated
100 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 */
221 set_clock_comparator(S390_lowcore.clock_comparator);
222 /* Check if old PSW is valid */
225 * Can't tell if we come from user or kernel mode
226 * -> stopping machine.
228 s390_handle_damage("old psw invalid.");
230 if (!mci->ms || !mci->pm || !mci->ia)
236 #define MAX_IPD_COUNT 29
237 #define MAX_IPD_TIME (5 * 60 * USEC_PER_SEC) /* 5 minutes */
239 #define ED_STP_ISLAND 6 /* External damage STP island check */
240 #define ED_STP_SYNC 7 /* External damage STP sync check */
241 #define ED_ETR_SYNC 12 /* External damage ETR sync check */
242 #define ED_ETR_SWITCH 13 /* External damage ETR switch to local */
245 * machine check handler.
247 void notrace s390_do_machine_check(struct pt_regs *regs)
249 static int ipd_count;
250 static DEFINE_SPINLOCK(ipd_lock);
251 static unsigned long long last_ipd;
252 struct mcck_struct *mcck;
253 unsigned long long tmp;
258 inc_irq_stat(NMI_NMI);
259 mci = (struct mci *) &S390_lowcore.mcck_interruption_code;
260 mcck = this_cpu_ptr(&cpu_mcck);
261 umode = user_mode(regs);
264 /* System damage -> stopping machine */
265 s390_handle_damage("received system damage machine check.");
269 /* Processing backup -> verify if we can survive this */
270 u64 z_mcic, o_mcic, t_mcic;
272 z_mcic = (1ULL<<63 | 1ULL<<59 | 1ULL<<29);
273 o_mcic = (1ULL<<43 | 1ULL<<42 | 1ULL<<41 | 1ULL<<40 |
274 1ULL<<36 | 1ULL<<35 | 1ULL<<34 | 1ULL<<32 |
275 1ULL<<30 | 1ULL<<21 | 1ULL<<20 | 1ULL<<17 |
278 z_mcic = (1ULL<<63 | 1ULL<<59 | 1ULL<<57 | 1ULL<<50 |
280 o_mcic = (1ULL<<43 | 1ULL<<42 | 1ULL<<41 | 1ULL<<40 |
281 1ULL<<36 | 1ULL<<35 | 1ULL<<34 | 1ULL<<32 |
282 1ULL<<30 | 1ULL<<20 | 1ULL<<17 | 1ULL<<16);
284 t_mcic = *(u64 *)mci;
286 if (((t_mcic & z_mcic) != 0) ||
287 ((t_mcic & o_mcic) != o_mcic)) {
288 s390_handle_damage("processing backup machine "
289 "check with damage.");
293 * Nullifying exigent condition, therefore we might
294 * retry this instruction.
296 spin_lock(&ipd_lock);
297 tmp = get_tod_clock();
298 if (((tmp - last_ipd) >> 12) < MAX_IPD_TIME)
303 if (ipd_count == MAX_IPD_COUNT)
304 s390_handle_damage("too many ipd retries.");
305 spin_unlock(&ipd_lock);
307 /* Processing damage -> stopping machine */
308 s390_handle_damage("received instruction processing "
309 "damage machine check.");
312 if (s390_revalidate_registers(mci)) {
315 * Couldn't restore all register contents while in
316 * user mode -> mark task for termination.
319 mcck->mcck_code = *(unsigned long long *) mci;
320 set_cpu_flag(CIF_MCCK_PENDING);
323 * Couldn't restore all register contents while in
324 * kernel mode -> stopping machine.
326 s390_handle_damage("unable to revalidate registers.");
330 /* Timing facility damage */
331 s390_handle_damage("TOD clock damaged");
333 if (mci->ed && mci->ec) {
334 /* External damage */
335 if (S390_lowcore.external_damage_code & (1U << ED_ETR_SYNC))
337 if (S390_lowcore.external_damage_code & (1U << ED_ETR_SWITCH))
338 etr_switch_to_local();
339 if (S390_lowcore.external_damage_code & (1U << ED_STP_SYNC))
341 if (S390_lowcore.external_damage_code & (1U << ED_STP_ISLAND))
345 /* Storage error uncorrected */
346 s390_handle_damage("received storage error uncorrected "
349 /* Storage key-error uncorrected */
350 s390_handle_damage("received storage key-error uncorrected "
352 if (mci->ds && mci->fa)
353 /* Storage degradation */
354 s390_handle_damage("received storage degradation machine "
357 /* Channel report word pending */
358 mcck->channel_report = 1;
359 set_cpu_flag(CIF_MCCK_PENDING);
362 /* Warning pending */
364 set_cpu_flag(CIF_MCCK_PENDING);
369 static int __init machine_check_init(void)
371 ctl_set_bit(14, 25); /* enable external damage MCH */
372 ctl_set_bit(14, 27); /* enable system recovery MCH */
373 ctl_set_bit(14, 24); /* enable warning MCH */
376 arch_initcall(machine_check_init);