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>
27 unsigned int kill_task : 1;
28 unsigned int channel_report : 1;
29 unsigned int warning : 1;
30 unsigned int stp_queue : 1;
31 unsigned long mcck_code;
34 static DEFINE_PER_CPU(struct mcck_struct, cpu_mcck);
36 static void s390_handle_damage(void)
39 disabled_wait((unsigned long) __builtin_return_address(0));
44 * Main machine check handler function. Will be called with interrupts enabled
45 * or disabled and machine checks enabled or disabled.
47 void s390_handle_mcck(void)
50 struct mcck_struct mcck;
53 * Disable machine checks and get the current state of accumulated
54 * machine checks. Afterwards delete the old state and enable machine
57 local_irq_save(flags);
59 mcck = *this_cpu_ptr(&cpu_mcck);
60 memset(this_cpu_ptr(&cpu_mcck), 0, sizeof(mcck));
61 clear_cpu_flag(CIF_MCCK_PENDING);
63 local_irq_restore(flags);
65 if (mcck.channel_report)
66 crw_handle_channel_report();
68 * A warning may remain for a prolonged period on the bare iron.
69 * (actually until the machine is powered off, or the problem is gone)
70 * So we just stop listening for the WARNING MCH and avoid continuously
71 * being interrupted. One caveat is however, that we must do this per
72 * processor and cannot use the smp version of ctl_clear_bit().
73 * On VM we only get one interrupt per virtally presented machinecheck.
74 * Though one suffices, we may get one interrupt per (virtual) cpu.
76 if (mcck.warning) { /* WARNING pending ? */
77 static int mchchk_wng_posted = 0;
79 /* Use single cpu clear, as we cannot handle smp here. */
80 __ctl_clear_bit(14, 24); /* Disable WARNING MCH */
81 if (xchg(&mchchk_wng_posted, 1) == 0)
82 kill_cad_pid(SIGPWR, 1);
88 printk(KERN_EMERG "mcck: Terminating task because of machine "
89 "malfunction (code 0x%016lx).\n", mcck.mcck_code);
90 printk(KERN_EMERG "mcck: task: %s, pid: %d.\n",
91 current->comm, current->pid);
95 EXPORT_SYMBOL_GPL(s390_handle_mcck);
98 * returns 0 if all registers could be validated
101 static int notrace s390_validate_registers(union mci mci)
105 void *fpt_save_area, *fpt_creg_save_area;
112 * General purpose registers couldn't be restored and have
113 * unknown contents. Process needs to be terminated.
119 * Floating point registers can't be restored and
120 * therefore the process needs to be terminated.
124 fpt_save_area = &S390_lowcore.floating_pt_save_area;
125 fpt_creg_save_area = &S390_lowcore.fpt_creg_save_area;
128 * Floating point control register can't be restored.
129 * Task will be terminated.
131 asm volatile("lfpc 0(%0)" : : "a" (&zero), "m" (zero));
134 asm volatile("lfpc 0(%0)" : : "a" (fpt_creg_save_area));
136 if (!MACHINE_HAS_VX) {
137 /* Validate floating point registers */
155 : : "a" (fpt_save_area));
157 /* Validate vector registers */
162 * Vector registers can't be restored and therefore
163 * the process needs to be terminated.
167 cr0.val = S390_lowcore.cregs_save_area[0];
168 cr0.afp = cr0.vx = 1;
169 __ctl_load(cr0.val, 0, 0);
172 " .word 0xe70f,0x1000,0x0036\n" /* vlm 0,15,0(1) */
173 " .word 0xe70f,0x1100,0x0c36\n" /* vlm 16,31,256(1) */
174 : : "Q" (*(struct vx_array *)
175 &S390_lowcore.vector_save_area) : "1");
176 __ctl_load(S390_lowcore.cregs_save_area[0], 0, 0);
178 /* Validate access registers */
181 : : "a" (&S390_lowcore.access_regs_save_area));
184 * Access registers have unknown contents.
189 /* Validate control registers */
192 * Control registers have unknown contents.
193 * Can't recover and therefore stopping machine.
195 s390_handle_damage();
199 : : "a" (&S390_lowcore.cregs_save_area));
202 * We don't even try to validate the TOD register, since we simply
203 * can't write something sensible into that register.
206 * See if we can validate the TOD programmable register with its
207 * old contents (should be zero) otherwise set it to zero.
218 : : "a" (&S390_lowcore.tod_progreg_save_area)
220 /* Validate 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();
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 */
243 * machine check handler.
245 void notrace s390_do_machine_check(struct pt_regs *regs)
247 static int ipd_count;
248 static DEFINE_SPINLOCK(ipd_lock);
249 static unsigned long long last_ipd;
250 struct mcck_struct *mcck;
251 unsigned long long tmp;
256 inc_irq_stat(NMI_NMI);
257 mci.val = S390_lowcore.mcck_interruption_code;
258 mcck = this_cpu_ptr(&cpu_mcck);
259 umode = user_mode(regs);
262 /* System damage -> stopping machine */
263 s390_handle_damage();
267 /* Processing backup -> verify if we can survive this */
268 u64 z_mcic, o_mcic, t_mcic;
269 z_mcic = (1ULL<<63 | 1ULL<<59 | 1ULL<<29);
270 o_mcic = (1ULL<<43 | 1ULL<<42 | 1ULL<<41 | 1ULL<<40 |
271 1ULL<<36 | 1ULL<<35 | 1ULL<<34 | 1ULL<<32 |
272 1ULL<<30 | 1ULL<<21 | 1ULL<<20 | 1ULL<<17 |
276 if (((t_mcic & z_mcic) != 0) ||
277 ((t_mcic & o_mcic) != o_mcic)) {
278 s390_handle_damage();
282 * Nullifying exigent condition, therefore we might
283 * retry this instruction.
285 spin_lock(&ipd_lock);
286 tmp = get_tod_clock();
287 if (((tmp - last_ipd) >> 12) < MAX_IPD_TIME)
292 if (ipd_count == MAX_IPD_COUNT)
293 s390_handle_damage();
294 spin_unlock(&ipd_lock);
296 /* Processing damage -> stopping machine */
297 s390_handle_damage();
300 if (s390_validate_registers(mci)) {
303 * Couldn't restore all register contents while in
304 * user mode -> mark task for termination.
307 mcck->mcck_code = mci.val;
308 set_cpu_flag(CIF_MCCK_PENDING);
311 * Couldn't restore all register contents while in
312 * kernel mode -> stopping machine.
314 s390_handle_damage();
318 /* Timing facility damage */
319 s390_handle_damage();
321 if (mci.ed && mci.ec) {
322 /* External damage */
323 if (S390_lowcore.external_damage_code & (1U << ED_STP_SYNC))
324 mcck->stp_queue |= stp_sync_check();
325 if (S390_lowcore.external_damage_code & (1U << ED_STP_ISLAND))
326 mcck->stp_queue |= stp_island_check();
328 set_cpu_flag(CIF_MCCK_PENDING);
331 /* Storage error uncorrected */
332 s390_handle_damage();
334 /* Storage key-error uncorrected */
335 s390_handle_damage();
336 if (mci.ds && mci.fa)
337 /* Storage degradation */
338 s390_handle_damage();
340 /* Channel report word pending */
341 mcck->channel_report = 1;
342 set_cpu_flag(CIF_MCCK_PENDING);
345 /* Warning pending */
347 set_cpu_flag(CIF_MCCK_PENDING);
352 static int __init machine_check_init(void)
354 ctl_set_bit(14, 25); /* enable external damage MCH */
355 ctl_set_bit(14, 27); /* enable system recovery MCH */
356 ctl_set_bit(14, 24); /* enable warning MCH */
359 early_initcall(machine_check_init);