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/ctl_reg.h>
30 unsigned long long mcck_code;
33 static DEFINE_PER_CPU(struct mcck_struct, cpu_mcck);
35 static void s390_handle_damage(char *msg)
38 disabled_wait((unsigned long) __builtin_return_address(0));
43 * Main machine check handler function. Will be called with interrupts enabled
44 * or disabled and machine checks enabled or disabled.
46 void s390_handle_mcck(void)
49 struct mcck_struct mcck;
52 * Disable machine checks and get the current state of accumulated
53 * machine checks. Afterwards delete the old state and enable machine
56 local_irq_save(flags);
58 mcck = *this_cpu_ptr(&cpu_mcck);
59 memset(this_cpu_ptr(&cpu_mcck), 0, sizeof(mcck));
60 clear_cpu_flag(CIF_MCCK_PENDING);
62 local_irq_restore(flags);
64 if (mcck.channel_report)
65 crw_handle_channel_report();
67 * A warning may remain for a prolonged period on the bare iron.
68 * (actually until the machine is powered off, or the problem is gone)
69 * So we just stop listening for the WARNING MCH and avoid continuously
70 * being interrupted. One caveat is however, that we must do this per
71 * processor and cannot use the smp version of ctl_clear_bit().
72 * On VM we only get one interrupt per virtally presented machinecheck.
73 * Though one suffices, we may get one interrupt per (virtual) cpu.
75 if (mcck.warning) { /* WARNING pending ? */
76 static int mchchk_wng_posted = 0;
78 /* Use single cpu clear, as we cannot handle smp here. */
79 __ctl_clear_bit(14, 24); /* Disable WARNING MCH */
80 if (xchg(&mchchk_wng_posted, 1) == 0)
81 kill_cad_pid(SIGPWR, 1);
85 printk(KERN_EMERG "mcck: Terminating task because of machine "
86 "malfunction (code 0x%016llx).\n", mcck.mcck_code);
87 printk(KERN_EMERG "mcck: task: %s, pid: %d.\n",
88 current->comm, current->pid);
92 EXPORT_SYMBOL_GPL(s390_handle_mcck);
95 * returns 0 if all registers could be validated
98 static int notrace s390_revalidate_registers(struct mci *mci)
102 void *fpt_save_area, *fpt_creg_save_area;
109 * General purpose registers couldn't be restored and have
110 * unknown contents. Process needs to be terminated.
116 * Floating point registers can't be restored and
117 * therefore the process needs to be terminated.
121 fpt_save_area = &S390_lowcore.floating_pt_save_area;
122 fpt_creg_save_area = &S390_lowcore.fpt_creg_save_area;
125 * Floating point control register can't be restored.
126 * Task will be terminated.
128 asm volatile("lfpc 0(%0)" : : "a" (&zero), "m" (zero));
131 asm volatile("lfpc 0(%0)" : : "a" (fpt_creg_save_area));
133 if (!MACHINE_HAS_VX) {
134 /* Revalidate floating point registers */
152 : : "a" (fpt_save_area));
154 /* Revalidate vector registers */
159 * Vector registers can't be restored and therefore
160 * the process needs to be terminated.
164 cr0.val = S390_lowcore.cregs_save_area[0];
165 cr0.afp = cr0.vx = 1;
166 __ctl_load(cr0.val, 0, 0);
167 restore_vx_regs((__vector128 *)
168 &S390_lowcore.vector_save_area);
169 __ctl_load(S390_lowcore.cregs_save_area[0], 0, 0);
171 /* Revalidate access registers */
174 : : "a" (&S390_lowcore.access_regs_save_area));
177 * Access registers have unknown contents.
182 /* Revalidate control registers */
185 * Control registers have unknown contents.
186 * Can't recover and therefore stopping machine.
188 s390_handle_damage("invalid control registers.");
192 : : "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.
199 * See if we can revalidate the TOD programmable register with its
200 * old contents (should be zero) otherwise set it to zero.
211 : : "a" (&S390_lowcore.tod_progreg_save_area)
213 /* Revalidate clock comparator register */
214 set_clock_comparator(S390_lowcore.clock_comparator);
215 /* Check if old PSW is valid */
218 * Can't tell if we come from user or kernel mode
219 * -> stopping machine.
221 s390_handle_damage("old psw invalid.");
223 if (!mci->ms || !mci->pm || !mci->ia)
229 #define MAX_IPD_COUNT 29
230 #define MAX_IPD_TIME (5 * 60 * USEC_PER_SEC) /* 5 minutes */
232 #define ED_STP_ISLAND 6 /* External damage STP island check */
233 #define ED_STP_SYNC 7 /* External damage STP sync check */
234 #define ED_ETR_SYNC 12 /* External damage ETR sync check */
235 #define ED_ETR_SWITCH 13 /* External damage ETR switch to local */
238 * machine check handler.
240 void notrace s390_do_machine_check(struct pt_regs *regs)
242 static int ipd_count;
243 static DEFINE_SPINLOCK(ipd_lock);
244 static unsigned long long last_ipd;
245 struct mcck_struct *mcck;
246 unsigned long long tmp;
251 inc_irq_stat(NMI_NMI);
252 mci = (struct mci *) &S390_lowcore.mcck_interruption_code;
253 mcck = this_cpu_ptr(&cpu_mcck);
254 umode = user_mode(regs);
257 /* System damage -> stopping machine */
258 s390_handle_damage("received system damage machine check.");
262 /* Processing backup -> verify if we can survive this */
263 u64 z_mcic, o_mcic, t_mcic;
264 z_mcic = (1ULL<<63 | 1ULL<<59 | 1ULL<<29);
265 o_mcic = (1ULL<<43 | 1ULL<<42 | 1ULL<<41 | 1ULL<<40 |
266 1ULL<<36 | 1ULL<<35 | 1ULL<<34 | 1ULL<<32 |
267 1ULL<<30 | 1ULL<<21 | 1ULL<<20 | 1ULL<<17 |
269 t_mcic = *(u64 *)mci;
271 if (((t_mcic & z_mcic) != 0) ||
272 ((t_mcic & o_mcic) != o_mcic)) {
273 s390_handle_damage("processing backup machine "
274 "check with damage.");
278 * Nullifying exigent condition, therefore we might
279 * retry this instruction.
281 spin_lock(&ipd_lock);
282 tmp = get_tod_clock();
283 if (((tmp - last_ipd) >> 12) < MAX_IPD_TIME)
288 if (ipd_count == MAX_IPD_COUNT)
289 s390_handle_damage("too many ipd retries.");
290 spin_unlock(&ipd_lock);
292 /* Processing damage -> stopping machine */
293 s390_handle_damage("received instruction processing "
294 "damage machine check.");
297 if (s390_revalidate_registers(mci)) {
300 * Couldn't restore all register contents while in
301 * user mode -> mark task for termination.
304 mcck->mcck_code = *(unsigned long long *) mci;
305 set_cpu_flag(CIF_MCCK_PENDING);
308 * Couldn't restore all register contents while in
309 * kernel mode -> stopping machine.
311 s390_handle_damage("unable to revalidate registers.");
315 /* Timing facility damage */
316 s390_handle_damage("TOD clock damaged");
318 if (mci->ed && mci->ec) {
319 /* External damage */
320 if (S390_lowcore.external_damage_code & (1U << ED_ETR_SYNC))
322 if (S390_lowcore.external_damage_code & (1U << ED_ETR_SWITCH))
323 etr_switch_to_local();
324 if (S390_lowcore.external_damage_code & (1U << ED_STP_SYNC))
326 if (S390_lowcore.external_damage_code & (1U << ED_STP_ISLAND))
330 /* Storage error uncorrected */
331 s390_handle_damage("received storage error uncorrected "
334 /* Storage key-error uncorrected */
335 s390_handle_damage("received storage key-error uncorrected "
337 if (mci->ds && mci->fa)
338 /* Storage degradation */
339 s390_handle_damage("received storage degradation machine "
342 /* Channel report word pending */
343 mcck->channel_report = 1;
344 set_cpu_flag(CIF_MCCK_PENDING);
347 /* Warning pending */
349 set_cpu_flag(CIF_MCCK_PENDING);
354 static int __init machine_check_init(void)
356 ctl_set_bit(14, 25); /* enable external damage MCH */
357 ctl_set_bit(14, 27); /* enable system recovery MCH */
358 ctl_set_bit(14, 24); /* enable warning MCH */
361 arch_initcall(machine_check_init);