aboutsummaryrefslogtreecommitdiffstats
path: root/arch/s390/kernel/nmi.c
diff options
context:
space:
mode:
Diffstat (limited to 'arch/s390/kernel/nmi.c')
-rw-r--r--arch/s390/kernel/nmi.c376
1 files changed, 376 insertions, 0 deletions
diff --git a/arch/s390/kernel/nmi.c b/arch/s390/kernel/nmi.c
new file mode 100644
index 000000000000..4bfdc421d7e9
--- /dev/null
+++ b/arch/s390/kernel/nmi.c
@@ -0,0 +1,376 @@
1/*
2 * Machine check handler
3 *
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>,
9 */
10
11#include <linux/init.h>
12#include <linux/errno.h>
13#include <linux/time.h>
14#include <linux/module.h>
15#include <asm/lowcore.h>
16#include <asm/smp.h>
17#include <asm/etr.h>
18#include <asm/cpu.h>
19#include <asm/nmi.h>
20#include <asm/crw.h>
21
22struct mcck_struct {
23 int kill_task;
24 int channel_report;
25 int warning;
26 unsigned long long mcck_code;
27};
28
29static DEFINE_PER_CPU(struct mcck_struct, cpu_mcck);
30
31static NORET_TYPE void s390_handle_damage(char *msg)
32{
33 smp_send_stop();
34 disabled_wait((unsigned long) __builtin_return_address(0));
35 while (1);
36}
37
38/*
39 * Main machine check handler function. Will be called with interrupts enabled
40 * or disabled and machine checks enabled or disabled.
41 */
42void s390_handle_mcck(void)
43{
44 unsigned long flags;
45 struct mcck_struct mcck;
46
47 /*
48 * Disable machine checks and get the current state of accumulated
49 * machine checks. Afterwards delete the old state and enable machine
50 * checks again.
51 */
52 local_irq_save(flags);
53 local_mcck_disable();
54 mcck = __get_cpu_var(cpu_mcck);
55 memset(&__get_cpu_var(cpu_mcck), 0, sizeof(struct mcck_struct));
56 clear_thread_flag(TIF_MCCK_PENDING);
57 local_mcck_enable();
58 local_irq_restore(flags);
59
60 if (mcck.channel_report)
61 crw_handle_channel_report();
62 /*
63 * A warning may remain for a prolonged period on the bare iron.
64 * (actually until the machine is powered off, or the problem is gone)
65 * So we just stop listening for the WARNING MCH and avoid continuously
66 * being interrupted. One caveat is however, that we must do this per
67 * processor and cannot use the smp version of ctl_clear_bit().
68 * On VM we only get one interrupt per virtally presented machinecheck.
69 * Though one suffices, we may get one interrupt per (virtual) cpu.
70 */
71 if (mcck.warning) { /* WARNING pending ? */
72 static int mchchk_wng_posted = 0;
73
74 /* Use single cpu clear, as we cannot handle smp here. */
75 __ctl_clear_bit(14, 24); /* Disable WARNING MCH */
76 if (xchg(&mchchk_wng_posted, 1) == 0)
77 kill_cad_pid(SIGPWR, 1);
78 }
79 if (mcck.kill_task) {
80 local_irq_enable();
81 printk(KERN_EMERG "mcck: Terminating task because of machine "
82 "malfunction (code 0x%016llx).\n", mcck.mcck_code);
83 printk(KERN_EMERG "mcck: task: %s, pid: %d.\n",
84 current->comm, current->pid);
85 do_exit(SIGSEGV);
86 }
87}
88EXPORT_SYMBOL_GPL(s390_handle_mcck);
89
90/*
91 * returns 0 if all registers could be validated
92 * returns 1 otherwise
93 */
94static int notrace s390_revalidate_registers(struct mci *mci)
95{
96 int kill_task;
97 u64 tmpclock;
98 u64 zero;
99 void *fpt_save_area, *fpt_creg_save_area;
100
101 kill_task = 0;
102 zero = 0;
103
104 if (!mci->gr) {
105 /*
106 * General purpose registers couldn't be restored and have
107 * unknown contents. Process needs to be terminated.
108 */
109 kill_task = 1;
110 }
111 if (!mci->fp) {
112 /*
113 * Floating point registers can't be restored and
114 * therefore the process needs to be terminated.
115 */
116 kill_task = 1;
117 }
118#ifndef CONFIG_64BIT
119 asm volatile(
120 " ld 0,0(%0)\n"
121 " ld 2,8(%0)\n"
122 " ld 4,16(%0)\n"
123 " ld 6,24(%0)"
124 : : "a" (&S390_lowcore.floating_pt_save_area));
125#endif
126
127 if (MACHINE_HAS_IEEE) {
128#ifdef CONFIG_64BIT
129 fpt_save_area = &S390_lowcore.floating_pt_save_area;
130 fpt_creg_save_area = &S390_lowcore.fpt_creg_save_area;
131#else
132 fpt_save_area = (void *) S390_lowcore.extended_save_area_addr;
133 fpt_creg_save_area = fpt_save_area + 128;
134#endif
135 if (!mci->fc) {
136 /*
137 * Floating point control register can't be restored.
138 * Task will be terminated.
139 */
140 asm volatile("lfpc 0(%0)" : : "a" (&zero), "m" (zero));
141 kill_task = 1;
142
143 } else
144 asm volatile("lfpc 0(%0)" : : "a" (fpt_creg_save_area));
145
146 asm volatile(
147 " ld 0,0(%0)\n"
148 " ld 1,8(%0)\n"
149 " ld 2,16(%0)\n"
150 " ld 3,24(%0)\n"
151 " ld 4,32(%0)\n"
152 " ld 5,40(%0)\n"
153 " ld 6,48(%0)\n"
154 " ld 7,56(%0)\n"
155 " ld 8,64(%0)\n"
156 " ld 9,72(%0)\n"
157 " ld 10,80(%0)\n"
158 " ld 11,88(%0)\n"
159 " ld 12,96(%0)\n"
160 " ld 13,104(%0)\n"
161 " ld 14,112(%0)\n"
162 " ld 15,120(%0)\n"
163 : : "a" (fpt_save_area));
164 }
165 /* Revalidate access registers */
166 asm volatile(
167 " lam 0,15,0(%0)"
168 : : "a" (&S390_lowcore.access_regs_save_area));
169 if (!mci->ar) {
170 /*
171 * Access registers have unknown contents.
172 * Terminating task.
173 */
174 kill_task = 1;
175 }
176 /* Revalidate control registers */
177 if (!mci->cr) {
178 /*
179 * Control registers have unknown contents.
180 * Can't recover and therefore stopping machine.
181 */
182 s390_handle_damage("invalid control registers.");
183 } else {
184#ifdef CONFIG_64BIT
185 asm volatile(
186 " lctlg 0,15,0(%0)"
187 : : "a" (&S390_lowcore.cregs_save_area));
188#else
189 asm volatile(
190 " lctl 0,15,0(%0)"
191 : : "a" (&S390_lowcore.cregs_save_area));
192#endif
193 }
194 /*
195 * We don't even try to revalidate the TOD register, since we simply
196 * can't write something sensible into that register.
197 */
198#ifdef CONFIG_64BIT
199 /*
200 * See if we can revalidate the TOD programmable register with its
201 * old contents (should be zero) otherwise set it to zero.
202 */
203 if (!mci->pr)
204 asm volatile(
205 " sr 0,0\n"
206 " sckpf"
207 : : : "0", "cc");
208 else
209 asm volatile(
210 " l 0,0(%0)\n"
211 " sckpf"
212 : : "a" (&S390_lowcore.tod_progreg_save_area)
213 : "0", "cc");
214#endif
215 /* Revalidate clock comparator register */
216 asm volatile(
217 " stck 0(%1)\n"
218 " sckc 0(%1)"
219 : "=m" (tmpclock) : "a" (&(tmpclock)) : "cc", "memory");
220
221 /* Check if old PSW is valid */
222 if (!mci->wp)
223 /*
224 * Can't tell if we come from user or kernel mode
225 * -> stopping machine.
226 */
227 s390_handle_damage("old psw invalid.");
228
229 if (!mci->ms || !mci->pm || !mci->ia)
230 kill_task = 1;
231
232 return kill_task;
233}
234
235#define MAX_IPD_COUNT 29
236#define MAX_IPD_TIME (5 * 60 * USEC_PER_SEC) /* 5 minutes */
237
238#define ED_STP_ISLAND 6 /* External damage STP island check */
239#define ED_STP_SYNC 7 /* External damage STP sync check */
240#define ED_ETR_SYNC 12 /* External damage ETR sync check */
241#define ED_ETR_SWITCH 13 /* External damage ETR switch to local */
242
243/*
244 * machine check handler.
245 */
246void notrace s390_do_machine_check(struct pt_regs *regs)
247{
248 static int ipd_count;
249 static DEFINE_SPINLOCK(ipd_lock);
250 static unsigned long long last_ipd;
251 struct mcck_struct *mcck;
252 unsigned long long tmp;
253 struct mci *mci;
254 int umode;
255
256 lockdep_off();
257 s390_idle_check();
258
259 mci = (struct mci *) &S390_lowcore.mcck_interruption_code;
260 mcck = &__get_cpu_var(cpu_mcck);
261 umode = user_mode(regs);
262
263 if (mci->sd) {
264 /* System damage -> stopping machine */
265 s390_handle_damage("received system damage machine check.");
266 }
267 if (mci->pd) {
268 if (mci->b) {
269 /* Processing backup -> verify if we can survive this */
270 u64 z_mcic, o_mcic, t_mcic;
271#ifdef CONFIG_64BIT
272 z_mcic = (1ULL<<63 | 1ULL<<59 | 1ULL<<29);
273 o_mcic = (1ULL<<43 | 1ULL<<42 | 1ULL<<41 | 1ULL<<40 |
274 1ULL<<36 | 1ULL<<35 | 1ULL<<34 | 1ULL<<32 |
275 1ULL<<30 | 1ULL<<21 | 1ULL<<20 | 1ULL<<17 |
276 1ULL<<16);
277#else
278 z_mcic = (1ULL<<63 | 1ULL<<59 | 1ULL<<57 | 1ULL<<50 |
279 1ULL<<29);
280 o_mcic = (1ULL<<43 | 1ULL<<42 | 1ULL<<41 | 1ULL<<40 |
281 1ULL<<36 | 1ULL<<35 | 1ULL<<34 | 1ULL<<32 |
282 1ULL<<30 | 1ULL<<20 | 1ULL<<17 | 1ULL<<16);
283#endif
284 t_mcic = *(u64 *)mci;
285
286 if (((t_mcic & z_mcic) != 0) ||
287 ((t_mcic & o_mcic) != o_mcic)) {
288 s390_handle_damage("processing backup machine "
289 "check with damage.");
290 }
291
292 /*
293 * Nullifying exigent condition, therefore we might
294 * retry this instruction.
295 */
296 spin_lock(&ipd_lock);
297 tmp = get_clock();
298 if (((tmp - last_ipd) >> 12) < MAX_IPD_TIME)
299 ipd_count++;
300 else
301 ipd_count = 1;
302 last_ipd = tmp;
303 if (ipd_count == MAX_IPD_COUNT)
304 s390_handle_damage("too many ipd retries.");
305 spin_unlock(&ipd_lock);
306 } else {
307 /* Processing damage -> stopping machine */
308 s390_handle_damage("received instruction processing "
309 "damage machine check.");
310 }
311 }
312 if (s390_revalidate_registers(mci)) {
313 if (umode) {
314 /*
315 * Couldn't restore all register contents while in
316 * user mode -> mark task for termination.
317 */
318 mcck->kill_task = 1;
319 mcck->mcck_code = *(unsigned long long *) mci;
320 set_thread_flag(TIF_MCCK_PENDING);
321 } else {
322 /*
323 * Couldn't restore all register contents while in
324 * kernel mode -> stopping machine.
325 */
326 s390_handle_damage("unable to revalidate registers.");
327 }
328 }
329 if (mci->cd) {
330 /* Timing facility damage */
331 s390_handle_damage("TOD clock damaged");
332 }
333 if (mci->ed && mci->ec) {
334 /* External damage */
335 if (S390_lowcore.external_damage_code & (1U << ED_ETR_SYNC))
336 etr_sync_check();
337 if (S390_lowcore.external_damage_code & (1U << ED_ETR_SWITCH))
338 etr_switch_to_local();
339 if (S390_lowcore.external_damage_code & (1U << ED_STP_SYNC))
340 stp_sync_check();
341 if (S390_lowcore.external_damage_code & (1U << ED_STP_ISLAND))
342 stp_island_check();
343 }
344 if (mci->se)
345 /* Storage error uncorrected */
346 s390_handle_damage("received storage error uncorrected "
347 "machine check.");
348 if (mci->ke)
349 /* Storage key-error uncorrected */
350 s390_handle_damage("received storage key-error uncorrected "
351 "machine check.");
352 if (mci->ds && mci->fa)
353 /* Storage degradation */
354 s390_handle_damage("received storage degradation machine "
355 "check.");
356 if (mci->cp) {
357 /* Channel report word pending */
358 mcck->channel_report = 1;
359 set_thread_flag(TIF_MCCK_PENDING);
360 }
361 if (mci->w) {
362 /* Warning pending */
363 mcck->warning = 1;
364 set_thread_flag(TIF_MCCK_PENDING);
365 }
366 lockdep_on();
367}
368
369static int __init machine_check_init(void)
370{
371 ctl_set_bit(14, 25); /* enable external damage MCH */
372 ctl_set_bit(14, 27); /* enable system recovery MCH */
373 ctl_set_bit(14, 24); /* enable warning MCH */
374 return 0;
375}
376arch_initcall(machine_check_init);
generated by cgit v1.2.2 (git 2.25.0) at 2025-01-06 07:28:43 -0500