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>
29 unsigned long long mcck_code;
32 static DEFINE_PER_CPU(struct mcck_struct, cpu_mcck);
34 static void s390_handle_damage(char *msg)
37 disabled_wait((unsigned long) __builtin_return_address(0));
42 * Main machine check handler function. Will be called with interrupts enabled
43 * or disabled and machine checks enabled or disabled.
45 void s390_handle_mcck(void)
48 struct mcck_struct mcck;
51 * Disable machine checks and get the current state of accumulated
52 * machine checks. Afterwards delete the old state and enable machine
55 local_irq_save(flags);
57 mcck = *this_cpu_ptr(&cpu_mcck);
58 memset(this_cpu_ptr(&cpu_mcck), 0, sizeof(mcck));
59 clear_cpu_flag(CIF_MCCK_PENDING);
61 local_irq_restore(flags);
63 if (mcck.channel_report)
64 crw_handle_channel_report();
66 * A warning may remain for a prolonged period on the bare iron.
67 * (actually until the machine is powered off, or the problem is gone)
68 * So we just stop listening for the WARNING MCH and avoid continuously
69 * being interrupted. One caveat is however, that we must do this per
70 * processor and cannot use the smp version of ctl_clear_bit().
71 * On VM we only get one interrupt per virtally presented machinecheck.
72 * Though one suffices, we may get one interrupt per (virtual) cpu.
74 if (mcck.warning) { /* WARNING pending ? */
75 static int mchchk_wng_posted = 0;
77 /* Use single cpu clear, as we cannot handle smp here. */
78 __ctl_clear_bit(14, 24); /* Disable WARNING MCH */
79 if (xchg(&mchchk_wng_posted, 1) == 0)
80 kill_cad_pid(SIGPWR, 1);
84 printk(KERN_EMERG "mcck: Terminating task because of machine "
85 "malfunction (code 0x%016llx).\n", mcck.mcck_code);
86 printk(KERN_EMERG "mcck: task: %s, pid: %d.\n",
87 current->comm, current->pid);
91 EXPORT_SYMBOL_GPL(s390_handle_mcck);
94 * returns 0 if all registers could be validated
97 static int notrace s390_revalidate_registers(struct mci *mci)
101 void *fpt_save_area, *fpt_creg_save_area;
108 * General purpose registers couldn't be restored and have
109 * unknown contents. Process needs to be terminated.
115 * Floating point registers can't be restored and
116 * therefore the process needs to be terminated.
120 fpt_save_area = &S390_lowcore.floating_pt_save_area;
121 fpt_creg_save_area = &S390_lowcore.fpt_creg_save_area;
124 * Floating point control register can't be restored.
125 * Task will be terminated.
127 asm volatile("lfpc 0(%0)" : : "a" (&zero), "m" (zero));
130 asm volatile("lfpc 0(%0)" : : "a" (fpt_creg_save_area));
149 : : "a" (fpt_save_area));
150 /* Revalidate vector registers */
151 if (MACHINE_HAS_VX && current->thread.vxrs) {
154 * Vector registers can't be restored and therefore
155 * the process needs to be terminated.
159 restore_vx_regs((__vector128 *)
160 S390_lowcore.vector_save_area_addr);
162 /* Revalidate access registers */
165 : : "a" (&S390_lowcore.access_regs_save_area));
168 * Access registers have unknown contents.
173 /* Revalidate control registers */
176 * Control registers have unknown contents.
177 * Can't recover and therefore stopping machine.
179 s390_handle_damage("invalid control registers.");
183 : : "a" (&S390_lowcore.cregs_save_area));
186 * We don't even try to revalidate the TOD register, since we simply
187 * can't write something sensible into that register.
190 * See if we can revalidate the TOD programmable register with its
191 * old contents (should be zero) otherwise set it to zero.
202 : : "a" (&S390_lowcore.tod_progreg_save_area)
204 /* Revalidate clock comparator register */
205 set_clock_comparator(S390_lowcore.clock_comparator);
206 /* Check if old PSW is valid */
209 * Can't tell if we come from user or kernel mode
210 * -> stopping machine.
212 s390_handle_damage("old psw invalid.");
214 if (!mci->ms || !mci->pm || !mci->ia)
220 #define MAX_IPD_COUNT 29
221 #define MAX_IPD_TIME (5 * 60 * USEC_PER_SEC) /* 5 minutes */
223 #define ED_STP_ISLAND 6 /* External damage STP island check */
224 #define ED_STP_SYNC 7 /* External damage STP sync check */
225 #define ED_ETR_SYNC 12 /* External damage ETR sync check */
226 #define ED_ETR_SWITCH 13 /* External damage ETR switch to local */
229 * machine check handler.
231 void notrace s390_do_machine_check(struct pt_regs *regs)
233 static int ipd_count;
234 static DEFINE_SPINLOCK(ipd_lock);
235 static unsigned long long last_ipd;
236 struct mcck_struct *mcck;
237 unsigned long long tmp;
242 inc_irq_stat(NMI_NMI);
243 mci = (struct mci *) &S390_lowcore.mcck_interruption_code;
244 mcck = this_cpu_ptr(&cpu_mcck);
245 umode = user_mode(regs);
248 /* System damage -> stopping machine */
249 s390_handle_damage("received system damage machine check.");
253 /* Processing backup -> verify if we can survive this */
254 u64 z_mcic, o_mcic, t_mcic;
255 z_mcic = (1ULL<<63 | 1ULL<<59 | 1ULL<<29);
256 o_mcic = (1ULL<<43 | 1ULL<<42 | 1ULL<<41 | 1ULL<<40 |
257 1ULL<<36 | 1ULL<<35 | 1ULL<<34 | 1ULL<<32 |
258 1ULL<<30 | 1ULL<<21 | 1ULL<<20 | 1ULL<<17 |
260 t_mcic = *(u64 *)mci;
262 if (((t_mcic & z_mcic) != 0) ||
263 ((t_mcic & o_mcic) != o_mcic)) {
264 s390_handle_damage("processing backup machine "
265 "check with damage.");
269 * Nullifying exigent condition, therefore we might
270 * retry this instruction.
272 spin_lock(&ipd_lock);
273 tmp = get_tod_clock();
274 if (((tmp - last_ipd) >> 12) < MAX_IPD_TIME)
279 if (ipd_count == MAX_IPD_COUNT)
280 s390_handle_damage("too many ipd retries.");
281 spin_unlock(&ipd_lock);
283 /* Processing damage -> stopping machine */
284 s390_handle_damage("received instruction processing "
285 "damage machine check.");
288 if (s390_revalidate_registers(mci)) {
291 * Couldn't restore all register contents while in
292 * user mode -> mark task for termination.
295 mcck->mcck_code = *(unsigned long long *) mci;
296 set_cpu_flag(CIF_MCCK_PENDING);
299 * Couldn't restore all register contents while in
300 * kernel mode -> stopping machine.
302 s390_handle_damage("unable to revalidate registers.");
306 /* Timing facility damage */
307 s390_handle_damage("TOD clock damaged");
309 if (mci->ed && mci->ec) {
310 /* External damage */
311 if (S390_lowcore.external_damage_code & (1U << ED_ETR_SYNC))
313 if (S390_lowcore.external_damage_code & (1U << ED_ETR_SWITCH))
314 etr_switch_to_local();
315 if (S390_lowcore.external_damage_code & (1U << ED_STP_SYNC))
317 if (S390_lowcore.external_damage_code & (1U << ED_STP_ISLAND))
321 /* Storage error uncorrected */
322 s390_handle_damage("received storage error uncorrected "
325 /* Storage key-error uncorrected */
326 s390_handle_damage("received storage key-error uncorrected "
328 if (mci->ds && mci->fa)
329 /* Storage degradation */
330 s390_handle_damage("received storage degradation machine "
333 /* Channel report word pending */
334 mcck->channel_report = 1;
335 set_cpu_flag(CIF_MCCK_PENDING);
338 /* Warning pending */
340 set_cpu_flag(CIF_MCCK_PENDING);
345 static int __init machine_check_init(void)
347 ctl_set_bit(14, 25); /* enable external damage MCH */
348 ctl_set_bit(14, 27); /* enable system recovery MCH */
349 ctl_set_bit(14, 24); /* enable warning MCH */
352 arch_initcall(machine_check_init);