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 int etr_queue : 1;
32 unsigned int stp_queue : 1;
33 unsigned long long mcck_code;
36 static DEFINE_PER_CPU(struct mcck_struct, cpu_mcck);
38 static void s390_handle_damage(char *msg)
41 disabled_wait((unsigned long) __builtin_return_address(0));
46 * Main machine check handler function. Will be called with interrupts enabled
47 * or disabled and machine checks enabled or disabled.
49 void s390_handle_mcck(void)
52 struct mcck_struct mcck;
55 * Disable machine checks and get the current state of accumulated
56 * machine checks. Afterwards delete the old state and enable machine
59 local_irq_save(flags);
61 mcck = *this_cpu_ptr(&cpu_mcck);
62 memset(this_cpu_ptr(&cpu_mcck), 0, sizeof(mcck));
63 clear_cpu_flag(CIF_MCCK_PENDING);
65 local_irq_restore(flags);
67 if (mcck.channel_report)
68 crw_handle_channel_report();
70 * A warning may remain for a prolonged period on the bare iron.
71 * (actually until the machine is powered off, or the problem is gone)
72 * So we just stop listening for the WARNING MCH and avoid continuously
73 * being interrupted. One caveat is however, that we must do this per
74 * processor and cannot use the smp version of ctl_clear_bit().
75 * On VM we only get one interrupt per virtally presented machinecheck.
76 * Though one suffices, we may get one interrupt per (virtual) cpu.
78 if (mcck.warning) { /* WARNING pending ? */
79 static int mchchk_wng_posted = 0;
81 /* Use single cpu clear, as we cannot handle smp here. */
82 __ctl_clear_bit(14, 24); /* Disable WARNING MCH */
83 if (xchg(&mchchk_wng_posted, 1) == 0)
84 kill_cad_pid(SIGPWR, 1);
92 printk(KERN_EMERG "mcck: Terminating task because of machine "
93 "malfunction (code 0x%016llx).\n", mcck.mcck_code);
94 printk(KERN_EMERG "mcck: task: %s, pid: %d.\n",
95 current->comm, current->pid);
99 EXPORT_SYMBOL_GPL(s390_handle_mcck);
102 * returns 0 if all registers could be validated
103 * returns 1 otherwise
105 static int notrace s390_revalidate_registers(struct mci *mci)
109 void *fpt_save_area, *fpt_creg_save_area;
116 * General purpose registers couldn't be restored and have
117 * unknown contents. Process needs to be terminated.
123 * Floating point registers can't be restored and
124 * therefore the process needs to be terminated.
128 fpt_save_area = &S390_lowcore.floating_pt_save_area;
129 fpt_creg_save_area = &S390_lowcore.fpt_creg_save_area;
132 * Floating point control register can't be restored.
133 * Task will be terminated.
135 asm volatile("lfpc 0(%0)" : : "a" (&zero), "m" (zero));
138 asm volatile("lfpc 0(%0)" : : "a" (fpt_creg_save_area));
140 if (!MACHINE_HAS_VX) {
141 /* Revalidate floating point registers */
159 : : "a" (fpt_save_area));
161 /* Revalidate vector registers */
166 * Vector registers can't be restored and therefore
167 * the process needs to be terminated.
171 cr0.val = S390_lowcore.cregs_save_area[0];
172 cr0.afp = cr0.vx = 1;
173 __ctl_load(cr0.val, 0, 0);
176 " .word 0xe70f,0x1000,0x0036\n" /* vlm 0,15,0(1) */
177 " .word 0xe70f,0x1100,0x0c36\n" /* vlm 16,31,256(1) */
178 : : "Q" (*(struct vx_array *)
179 &S390_lowcore.vector_save_area) : "1");
180 __ctl_load(S390_lowcore.cregs_save_area[0], 0, 0);
182 /* Revalidate access registers */
185 : : "a" (&S390_lowcore.access_regs_save_area));
188 * Access registers have unknown contents.
193 /* Revalidate control registers */
196 * Control registers have unknown contents.
197 * Can't recover and therefore stopping machine.
199 s390_handle_damage("invalid control registers.");
203 : : "a" (&S390_lowcore.cregs_save_area));
206 * We don't even try to revalidate the TOD register, since we simply
207 * can't write something sensible into that register.
210 * See if we can revalidate the TOD programmable register with its
211 * old contents (should be zero) otherwise set it to zero.
222 : : "a" (&S390_lowcore.tod_progreg_save_area)
224 /* Revalidate clock comparator register */
225 set_clock_comparator(S390_lowcore.clock_comparator);
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;
262 inc_irq_stat(NMI_NMI);
263 mci = (struct mci *) &S390_lowcore.mcck_interruption_code;
264 mcck = this_cpu_ptr(&cpu_mcck);
265 umode = user_mode(regs);
268 /* System damage -> stopping machine */
269 s390_handle_damage("received system damage machine check.");
273 /* Processing backup -> verify if we can survive this */
274 u64 z_mcic, o_mcic, t_mcic;
275 z_mcic = (1ULL<<63 | 1ULL<<59 | 1ULL<<29);
276 o_mcic = (1ULL<<43 | 1ULL<<42 | 1ULL<<41 | 1ULL<<40 |
277 1ULL<<36 | 1ULL<<35 | 1ULL<<34 | 1ULL<<32 |
278 1ULL<<30 | 1ULL<<21 | 1ULL<<20 | 1ULL<<17 |
280 t_mcic = *(u64 *)mci;
282 if (((t_mcic & z_mcic) != 0) ||
283 ((t_mcic & o_mcic) != o_mcic)) {
284 s390_handle_damage("processing backup machine "
285 "check with damage.");
289 * Nullifying exigent condition, therefore we might
290 * retry this instruction.
292 spin_lock(&ipd_lock);
293 tmp = get_tod_clock();
294 if (((tmp - last_ipd) >> 12) < MAX_IPD_TIME)
299 if (ipd_count == MAX_IPD_COUNT)
300 s390_handle_damage("too many ipd retries.");
301 spin_unlock(&ipd_lock);
303 /* Processing damage -> stopping machine */
304 s390_handle_damage("received instruction processing "
305 "damage machine check.");
308 if (s390_revalidate_registers(mci)) {
311 * Couldn't restore all register contents while in
312 * user mode -> mark task for termination.
315 mcck->mcck_code = *(unsigned long long *) mci;
316 set_cpu_flag(CIF_MCCK_PENDING);
319 * Couldn't restore all register contents while in
320 * kernel mode -> stopping machine.
322 s390_handle_damage("unable to revalidate registers.");
326 /* Timing facility damage */
327 s390_handle_damage("TOD clock damaged");
329 if (mci->ed && mci->ec) {
330 /* External damage */
331 if (S390_lowcore.external_damage_code & (1U << ED_ETR_SYNC))
332 mcck->etr_queue |= etr_sync_check();
333 if (S390_lowcore.external_damage_code & (1U << ED_ETR_SWITCH))
334 mcck->etr_queue |= etr_switch_to_local();
335 if (S390_lowcore.external_damage_code & (1U << ED_STP_SYNC))
336 mcck->stp_queue |= stp_sync_check();
337 if (S390_lowcore.external_damage_code & (1U << ED_STP_ISLAND))
338 mcck->stp_queue |= stp_island_check();
339 if (mcck->etr_queue || mcck->stp_queue)
340 set_cpu_flag(CIF_MCCK_PENDING);
343 /* Storage error uncorrected */
344 s390_handle_damage("received storage error uncorrected "
347 /* Storage key-error uncorrected */
348 s390_handle_damage("received storage key-error uncorrected "
350 if (mci->ds && mci->fa)
351 /* Storage degradation */
352 s390_handle_damage("received storage degradation machine "
355 /* Channel report word pending */
356 mcck->channel_report = 1;
357 set_cpu_flag(CIF_MCCK_PENDING);
360 /* Warning pending */
362 set_cpu_flag(CIF_MCCK_PENDING);
367 static int __init machine_check_init(void)
369 ctl_set_bit(14, 25); /* enable external damage MCH */
370 ctl_set_bit(14, 27); /* enable system recovery MCH */
371 ctl_set_bit(14, 24); /* enable warning MCH */
374 early_initcall(machine_check_init);