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, int umode)
105 void *fpt_save_area, *fpt_creg_save_area;
112 * General purpose registers couldn't be restored and have
113 * unknown contents. Stop system or terminate process.
116 s390_handle_damage();
121 * Floating point registers can't be restored. If the
122 * kernel currently uses floating point registers the
123 * system is stopped. If the process has its floating
124 * pointer registers loaded it is terminated.
125 * Otherwise just revalidate the registers.
127 if (S390_lowcore.fpu_flags & KERNEL_VXR_V0V7)
128 s390_handle_damage();
129 if (!test_cpu_flag(CIF_FPU))
132 fpt_save_area = &S390_lowcore.floating_pt_save_area;
133 fpt_creg_save_area = &S390_lowcore.fpt_creg_save_area;
136 * Floating point control register can't be restored.
137 * If the kernel currently uses the floating pointer
138 * registers and needs the FPC register the system is
139 * stopped. If the process has its floating pointer
140 * registers loaded it is terminated. Otherwiese the
141 * FPC is just revalidated.
143 if (S390_lowcore.fpu_flags & KERNEL_FPC)
144 s390_handle_damage();
145 asm volatile("lfpc 0(%0)" : : "a" (&zero), "m" (zero));
146 if (!test_cpu_flag(CIF_FPU))
149 asm volatile("lfpc 0(%0)" : : "a" (fpt_creg_save_area));
151 if (!MACHINE_HAS_VX) {
152 /* Validate floating point registers */
170 : : "a" (fpt_save_area));
172 /* Validate vector registers */
177 * Vector registers can't be restored. If the kernel
178 * currently uses vector registers the system is
179 * stopped. If the process has its vector registers
180 * loaded it is terminated. Otherwise just revalidate
183 if (S390_lowcore.fpu_flags & KERNEL_VXR)
184 s390_handle_damage();
185 if (!test_cpu_flag(CIF_FPU))
188 cr0.val = S390_lowcore.cregs_save_area[0];
189 cr0.afp = cr0.vx = 1;
190 __ctl_load(cr0.val, 0, 0);
193 " .word 0xe70f,0x1000,0x0036\n" /* vlm 0,15,0(1) */
194 " .word 0xe70f,0x1100,0x0c36\n" /* vlm 16,31,256(1) */
195 : : "Q" (*(struct vx_array *)
196 &S390_lowcore.vector_save_area) : "1");
197 __ctl_load(S390_lowcore.cregs_save_area[0], 0, 0);
199 /* Validate access registers */
202 : : "a" (&S390_lowcore.access_regs_save_area));
205 * Access registers have unknown contents.
210 /* Validate control registers */
213 * Control registers have unknown contents.
214 * Can't recover and therefore stopping machine.
216 s390_handle_damage();
220 : : "a" (&S390_lowcore.cregs_save_area));
223 * We don't even try to validate the TOD register, since we simply
224 * can't write something sensible into that register.
227 * See if we can validate the TOD programmable register with its
228 * old contents (should be zero) otherwise set it to zero.
239 : : "a" (&S390_lowcore.tod_progreg_save_area)
241 /* Validate clock comparator register */
242 set_clock_comparator(S390_lowcore.clock_comparator);
243 /* Check if old PSW is valid */
246 * Can't tell if we come from user or kernel mode
247 * -> stopping machine.
249 s390_handle_damage();
251 if (!mci.ms || !mci.pm || !mci.ia)
257 #define MAX_IPD_COUNT 29
258 #define MAX_IPD_TIME (5 * 60 * USEC_PER_SEC) /* 5 minutes */
260 #define ED_STP_ISLAND 6 /* External damage STP island check */
261 #define ED_STP_SYNC 7 /* External damage STP sync check */
264 * machine check handler.
266 void notrace s390_do_machine_check(struct pt_regs *regs)
268 static int ipd_count;
269 static DEFINE_SPINLOCK(ipd_lock);
270 static unsigned long long last_ipd;
271 struct mcck_struct *mcck;
272 unsigned long long tmp;
276 inc_irq_stat(NMI_NMI);
277 mci.val = S390_lowcore.mcck_interruption_code;
278 mcck = this_cpu_ptr(&cpu_mcck);
281 /* System damage -> stopping machine */
282 s390_handle_damage();
286 /* Processing backup -> verify if we can survive this */
287 u64 z_mcic, o_mcic, t_mcic;
288 z_mcic = (1ULL<<63 | 1ULL<<59 | 1ULL<<29);
289 o_mcic = (1ULL<<43 | 1ULL<<42 | 1ULL<<41 | 1ULL<<40 |
290 1ULL<<36 | 1ULL<<35 | 1ULL<<34 | 1ULL<<32 |
291 1ULL<<30 | 1ULL<<21 | 1ULL<<20 | 1ULL<<17 |
295 if (((t_mcic & z_mcic) != 0) ||
296 ((t_mcic & o_mcic) != o_mcic)) {
297 s390_handle_damage();
301 * Nullifying exigent condition, therefore we might
302 * retry this instruction.
304 spin_lock(&ipd_lock);
305 tmp = get_tod_clock();
306 if (((tmp - last_ipd) >> 12) < MAX_IPD_TIME)
311 if (ipd_count == MAX_IPD_COUNT)
312 s390_handle_damage();
313 spin_unlock(&ipd_lock);
315 /* Processing damage -> stopping machine */
316 s390_handle_damage();
319 if (s390_validate_registers(mci, user_mode(regs))) {
321 * Couldn't restore all register contents for the
322 * user space process -> mark task for termination.
325 mcck->mcck_code = mci.val;
326 set_cpu_flag(CIF_MCCK_PENDING);
329 /* Timing facility damage */
330 s390_handle_damage();
332 if (mci.ed && mci.ec) {
333 /* External damage */
334 if (S390_lowcore.external_damage_code & (1U << ED_STP_SYNC))
335 mcck->stp_queue |= stp_sync_check();
336 if (S390_lowcore.external_damage_code & (1U << ED_STP_ISLAND))
337 mcck->stp_queue |= stp_island_check();
339 set_cpu_flag(CIF_MCCK_PENDING);
342 /* Storage error uncorrected */
343 s390_handle_damage();
345 /* Storage key-error uncorrected */
346 s390_handle_damage();
347 if (mci.ds && mci.fa)
348 /* Storage degradation */
349 s390_handle_damage();
351 /* Channel report word pending */
352 mcck->channel_report = 1;
353 set_cpu_flag(CIF_MCCK_PENDING);
356 /* Warning pending */
358 set_cpu_flag(CIF_MCCK_PENDING);
363 static int __init machine_check_init(void)
365 ctl_set_bit(14, 25); /* enable external damage MCH */
366 ctl_set_bit(14, 27); /* enable system recovery MCH */
367 ctl_set_bit(14, 24); /* enable warning MCH */
370 early_initcall(machine_check_init);