projects / deliverable / linux.git / blob
? search:


834e9ee7e934f0e3222d17d0e7e929db565a615d
[deliverable/linux.git] / drivers / s390 / s390mach.c
1 /*
2 * drivers/s390/s390mach.c
3 * S/390 machine check handler
4 *
5 * Copyright IBM Corp. 2000,2008
6 * Author(s): Ingo Adlung (adlung@de.ibm.com)
7 * Martin Schwidefsky (schwidefsky@de.ibm.com)
8 * Cornelia Huck <cornelia.huck@de.ibm.com>
9 */
10
11 #include <linux/init.h>
12 #include <linux/sched.h>
13 #include <linux/errno.h>
14 #include <linux/workqueue.h>
15 #include <linux/time.h>
16 #include <linux/device.h>
17 #include <linux/kthread.h>
18 #include <asm/etr.h>
19 #include <asm/lowcore.h>
20 #include <asm/cio.h>
21 #include "s390mach.h"
22
23 static struct semaphore m_sem;
24
25 static NORET_TYPE void
26 s390_handle_damage(char *msg)
27 {
28 #ifdef CONFIG_SMP
29 smp_send_stop();
30 #endif
31 disabled_wait((unsigned long) __builtin_return_address(0));
32 for(;;);
33 }
34
35 static crw_handler_t crw_handlers[NR_RSCS];
36
37 /**
38 * s390_register_crw_handler() - register a channel report word handler
39 * @rsc: reporting source code to handle
40 * @handler: handler to be registered
41 *
42 * Returns %0 on success and a negative error value otherwise.
43 */
44 int s390_register_crw_handler(int rsc, crw_handler_t handler)
45 {
46 if ((rsc < 0) || (rsc >= NR_RSCS))
47 return -EINVAL;
48 if (!cmpxchg(&crw_handlers[rsc], NULL, handler))
49 return 0;
50 return -EBUSY;
51 }
52
53 /**
54 * s390_unregister_crw_handler() - unregister a channel report word handler
55 * @rsc: reporting source code to handle
56 */
57 void s390_unregister_crw_handler(int rsc)
58 {
59 if ((rsc < 0) || (rsc >= NR_RSCS))
60 return;
61 xchg(&crw_handlers[rsc], NULL);
62 synchronize_sched();
63 }
64
65 /*
66 * Retrieve CRWs and call function to handle event.
67 */
68 static int s390_collect_crw_info(void *param)
69 {
70 struct crw crw[2];
71 int ccode;
72 struct semaphore *sem;
73 unsigned int chain;
74 int ignore;
75
76 sem = (struct semaphore *)param;
77 repeat:
78 ignore = down_interruptible(sem);
79 chain = 0;
80 while (1) {
81 if (unlikely(chain > 1)) {
82 struct crw tmp_crw;
83
84 printk(KERN_WARNING"%s: Code does not support more "
85 "than two chained crws; please report to "
86 "linux390@de.ibm.com!\n", __func__);
87 ccode = stcrw(&tmp_crw);
88 printk(KERN_WARNING"%s: crw reports slct=%d, oflw=%d, "
89 "chn=%d, rsc=%X, anc=%d, erc=%X, rsid=%X\n",
90 __func__, tmp_crw.slct, tmp_crw.oflw,
91 tmp_crw.chn, tmp_crw.rsc, tmp_crw.anc,
92 tmp_crw.erc, tmp_crw.rsid);
93 printk(KERN_WARNING"%s: This was crw number %x in the "
94 "chain\n", __func__, chain);
95 if (ccode != 0)
96 break;
97 chain = tmp_crw.chn ? chain + 1 : 0;
98 continue;
99 }
100 ccode = stcrw(&crw[chain]);
101 if (ccode != 0)
102 break;
103 printk(KERN_DEBUG "crw_info : CRW reports slct=%d, oflw=%d, "
104 "chn=%d, rsc=%X, anc=%d, erc=%X, rsid=%X\n",
105 crw[chain].slct, crw[chain].oflw, crw[chain].chn,
106 crw[chain].rsc, crw[chain].anc, crw[chain].erc,
107 crw[chain].rsid);
108 /* Check for overflows. */
109 if (crw[chain].oflw) {
110 int i;
111
112 pr_debug("%s: crw overflow detected!\n", __func__);
113 for (i = 0; i < NR_RSCS; i++) {
114 if (crw_handlers[i])
115 crw_handlers[i](NULL, NULL, 1);
116 }
117 chain = 0;
118 continue;
119 }
120 if (crw[0].chn && !chain) {
121 chain++;
122 continue;
123 }
124 if (crw_handlers[crw[chain].rsc])
125 crw_handlers[crw[chain].rsc](&crw[0],
126 chain ? &crw[1] : NULL,
127 0);
128 /* chain is always 0 or 1 here. */
129 chain = crw[chain].chn ? chain + 1 : 0;
130 }
131 goto repeat;
132 return 0;
133 }
134
135 struct mcck_struct {
136 int kill_task;
137 int channel_report;
138 int warning;
139 unsigned long long mcck_code;
140 };
141
142 static DEFINE_PER_CPU(struct mcck_struct, cpu_mcck);
143
144 /*
145 * Main machine check handler function. Will be called with interrupts enabled
146 * or disabled and machine checks enabled or disabled.
147 */
148 void
149 s390_handle_mcck(void)
150 {
151 unsigned long flags;
152 struct mcck_struct mcck;
153
154 /*
155 * Disable machine checks and get the current state of accumulated
156 * machine checks. Afterwards delete the old state and enable machine
157 * checks again.
158 */
159 local_irq_save(flags);
160 local_mcck_disable();
161 mcck = __get_cpu_var(cpu_mcck);
162 memset(&__get_cpu_var(cpu_mcck), 0, sizeof(struct mcck_struct));
163 clear_thread_flag(TIF_MCCK_PENDING);
164 local_mcck_enable();
165 local_irq_restore(flags);
166
167 if (mcck.channel_report)
168 up(&m_sem);
169
170 #ifdef CONFIG_MACHCHK_WARNING
171 /*
172 * The warning may remain for a prolonged period on the bare iron.
173 * (actually till the machine is powered off, or until the problem is gone)
174 * So we just stop listening for the WARNING MCH and prevent continuously
175 * being interrupted. One caveat is however, that we must do this per
176 * processor and cannot use the smp version of ctl_clear_bit().
177 * On VM we only get one interrupt per virtally presented machinecheck.
178 * Though one suffices, we may get one interrupt per (virtual) processor.
179 */
180 if (mcck.warning) { /* WARNING pending ? */
181 static int mchchk_wng_posted = 0;
182 /*
183 * Use single machine clear, as we cannot handle smp right now
184 */
185 __ctl_clear_bit(14, 24); /* Disable WARNING MCH */
186 if (xchg(&mchchk_wng_posted, 1) == 0)
187 kill_cad_pid(SIGPWR, 1);
188 }
189 #endif
190
191 if (mcck.kill_task) {
192 local_irq_enable();
193 printk(KERN_EMERG "mcck: Terminating task because of machine "
194 "malfunction (code 0x%016llx).\n", mcck.mcck_code);
195 printk(KERN_EMERG "mcck: task: %s, pid: %d.\n",
196 current->comm, current->pid);
197 do_exit(SIGSEGV);
198 }
199 }
200 EXPORT_SYMBOL_GPL(s390_handle_mcck);
201
202 /*
203 * returns 0 if all registers could be validated
204 * returns 1 otherwise
205 */
206 static int
207 s390_revalidate_registers(struct mci *mci)
208 {
209 int kill_task;
210 u64 tmpclock;
211 u64 zero;
212 void *fpt_save_area, *fpt_creg_save_area;
213
214 kill_task = 0;
215 zero = 0;
216 /* General purpose registers */
217 if (!mci->gr)
218 /*
219 * General purpose registers couldn't be restored and have
220 * unknown contents. Process needs to be terminated.
221 */
222 kill_task = 1;
223
224 /* Revalidate floating point registers */
225 if (!mci->fp)
226 /*
227 * Floating point registers can't be restored and
228 * therefore the process needs to be terminated.
229 */
230 kill_task = 1;
231
232 #ifndef CONFIG_64BIT
233 asm volatile(
234 " ld 0,0(%0)\n"
235 " ld 2,8(%0)\n"
236 " ld 4,16(%0)\n"
237 " ld 6,24(%0)"
238 : : "a" (&S390_lowcore.floating_pt_save_area));
239 #endif
240
241 if (MACHINE_HAS_IEEE) {
242 #ifdef CONFIG_64BIT
243 fpt_save_area = &S390_lowcore.floating_pt_save_area;
244 fpt_creg_save_area = &S390_lowcore.fpt_creg_save_area;
245 #else
246 fpt_save_area = (void *) S390_lowcore.extended_save_area_addr;
247 fpt_creg_save_area = fpt_save_area+128;
248 #endif
249 /* Floating point control register */
250 if (!mci->fc) {
251 /*
252 * Floating point control register can't be restored.
253 * Task will be terminated.
254 */
255 asm volatile("lfpc 0(%0)" : : "a" (&zero), "m" (zero));
256 kill_task = 1;
257
258 } else
259 asm volatile("lfpc 0(%0)" : : "a" (fpt_creg_save_area));
260
261 asm volatile(
262 " ld 0,0(%0)\n"
263 " ld 1,8(%0)\n"
264 " ld 2,16(%0)\n"
265 " ld 3,24(%0)\n"
266 " ld 4,32(%0)\n"
267 " ld 5,40(%0)\n"
268 " ld 6,48(%0)\n"
269 " ld 7,56(%0)\n"
270 " ld 8,64(%0)\n"
271 " ld 9,72(%0)\n"
272 " ld 10,80(%0)\n"
273 " ld 11,88(%0)\n"
274 " ld 12,96(%0)\n"
275 " ld 13,104(%0)\n"
276 " ld 14,112(%0)\n"
277 " ld 15,120(%0)\n"
278 : : "a" (fpt_save_area));
279 }
280
281 /* Revalidate access registers */
282 asm volatile(
283 " lam 0,15,0(%0)"
284 : : "a" (&S390_lowcore.access_regs_save_area));
285 if (!mci->ar)
286 /*
287 * Access registers have unknown contents.
288 * Terminating task.
289 */
290 kill_task = 1;
291
292 /* Revalidate control registers */
293 if (!mci->cr)
294 /*
295 * Control registers have unknown contents.
296 * Can't recover and therefore stopping machine.
297 */
298 s390_handle_damage("invalid control registers.");
299 else
300 #ifdef CONFIG_64BIT
301 asm volatile(
302 " lctlg 0,15,0(%0)"
303 : : "a" (&S390_lowcore.cregs_save_area));
304 #else
305 asm volatile(
306 " lctl 0,15,0(%0)"
307 : : "a" (&S390_lowcore.cregs_save_area));
308 #endif
309
310 /*
311 * We don't even try to revalidate the TOD register, since we simply
312 * can't write something sensible into that register.
313 */
314
315 #ifdef CONFIG_64BIT
316 /*
317 * See if we can revalidate the TOD programmable register with its
318 * old contents (should be zero) otherwise set it to zero.
319 */
320 if (!mci->pr)
321 asm volatile(
322 " sr 0,0\n"
323 " sckpf"
324 : : : "0", "cc");
325 else
326 asm volatile(
327 " l 0,0(%0)\n"
328 " sckpf"
329 : : "a" (&S390_lowcore.tod_progreg_save_area)
330 : "0", "cc");
331 #endif
332
333 /* Revalidate clock comparator register */
334 asm volatile(
335 " stck 0(%1)\n"
336 " sckc 0(%1)"
337 : "=m" (tmpclock) : "a" (&(tmpclock)) : "cc", "memory");
338
339 /* Check if old PSW is valid */
340 if (!mci->wp)
341 /*
342 * Can't tell if we come from user or kernel mode
343 * -> stopping machine.
344 */
345 s390_handle_damage("old psw invalid.");
346
347 if (!mci->ms || !mci->pm || !mci->ia)
348 kill_task = 1;
349
350 return kill_task;
351 }
352
353 #define MAX_IPD_COUNT 29
354 #define MAX_IPD_TIME (5 * 60 * USEC_PER_SEC) /* 5 minutes */
355
356 /*
357 * machine check handler.
358 */
359 void
360 s390_do_machine_check(struct pt_regs *regs)
361 {
362 static DEFINE_SPINLOCK(ipd_lock);
363 static unsigned long long last_ipd;
364 static int ipd_count;
365 unsigned long long tmp;
366 struct mci *mci;
367 struct mcck_struct *mcck;
368 int umode;
369
370 lockdep_off();
371
372 mci = (struct mci *) &S390_lowcore.mcck_interruption_code;
373 mcck = &__get_cpu_var(cpu_mcck);
374 umode = user_mode(regs);
375
376 if (mci->sd)
377 /* System damage -> stopping machine */
378 s390_handle_damage("received system damage machine check.");
379
380 if (mci->pd) {
381 if (mci->b) {
382 /* Processing backup -> verify if we can survive this */
383 u64 z_mcic, o_mcic, t_mcic;
384 #ifdef CONFIG_64BIT
385 z_mcic = (1ULL<<63 | 1ULL<<59 | 1ULL<<29);
386 o_mcic = (1ULL<<43 | 1ULL<<42 | 1ULL<<41 | 1ULL<<40 |
387 1ULL<<36 | 1ULL<<35 | 1ULL<<34 | 1ULL<<32 |
388 1ULL<<30 | 1ULL<<21 | 1ULL<<20 | 1ULL<<17 |
389 1ULL<<16);
390 #else
391 z_mcic = (1ULL<<63 | 1ULL<<59 | 1ULL<<57 | 1ULL<<50 |
392 1ULL<<29);
393 o_mcic = (1ULL<<43 | 1ULL<<42 | 1ULL<<41 | 1ULL<<40 |
394 1ULL<<36 | 1ULL<<35 | 1ULL<<34 | 1ULL<<32 |
395 1ULL<<30 | 1ULL<<20 | 1ULL<<17 | 1ULL<<16);
396 #endif
397 t_mcic = *(u64 *)mci;
398
399 if (((t_mcic & z_mcic) != 0) ||
400 ((t_mcic & o_mcic) != o_mcic)) {
401 s390_handle_damage("processing backup machine "
402 "check with damage.");
403 }
404
405 /*
406 * Nullifying exigent condition, therefore we might
407 * retry this instruction.
408 */
409
410 spin_lock(&ipd_lock);
411
412 tmp = get_clock();
413
414 if (((tmp - last_ipd) >> 12) < MAX_IPD_TIME)
415 ipd_count++;
416 else
417 ipd_count = 1;
418
419 last_ipd = tmp;
420
421 if (ipd_count == MAX_IPD_COUNT)
422 s390_handle_damage("too many ipd retries.");
423
424 spin_unlock(&ipd_lock);
425 }
426 else {
427 /* Processing damage -> stopping machine */
428 s390_handle_damage("received instruction processing "
429 "damage machine check.");
430 }
431 }
432 if (s390_revalidate_registers(mci)) {
433 if (umode) {
434 /*
435 * Couldn't restore all register contents while in
436 * user mode -> mark task for termination.
437 */
438 mcck->kill_task = 1;
439 mcck->mcck_code = *(unsigned long long *) mci;
440 set_thread_flag(TIF_MCCK_PENDING);
441 }
442 else
443 /*
444 * Couldn't restore all register contents while in
445 * kernel mode -> stopping machine.
446 */
447 s390_handle_damage("unable to revalidate registers.");
448 }
449
450 if (mci->cd) {
451 /* Timing facility damage */
452 s390_handle_damage("TOD clock damaged");
453 }
454
455 if (mci->ed && mci->ec) {
456 /* External damage */
457 if (S390_lowcore.external_damage_code & (1U << ED_ETR_SYNC))
458 etr_sync_check();
459 if (S390_lowcore.external_damage_code & (1U << ED_ETR_SWITCH))
460 etr_switch_to_local();
461 if (S390_lowcore.external_damage_code & (1U << ED_STP_SYNC))
462 stp_sync_check();
463 if (S390_lowcore.external_damage_code & (1U << ED_STP_ISLAND))
464 stp_island_check();
465 }
466
467 if (mci->se)
468 /* Storage error uncorrected */
469 s390_handle_damage("received storage error uncorrected "
470 "machine check.");
471
472 if (mci->ke)
473 /* Storage key-error uncorrected */
474 s390_handle_damage("received storage key-error uncorrected "
475 "machine check.");
476
477 if (mci->ds && mci->fa)
478 /* Storage degradation */
479 s390_handle_damage("received storage degradation machine "
480 "check.");
481
482 if (mci->cp) {
483 /* Channel report word pending */
484 mcck->channel_report = 1;
485 set_thread_flag(TIF_MCCK_PENDING);
486 }
487
488 if (mci->w) {
489 /* Warning pending */
490 mcck->warning = 1;
491 set_thread_flag(TIF_MCCK_PENDING);
492 }
493 lockdep_on();
494 }
495
496 /*
497 * s390_init_machine_check
498 *
499 * initialize machine check handling
500 */
501 static int
502 machine_check_init(void)
503 {
504 init_MUTEX_LOCKED(&m_sem);
505 ctl_set_bit(14, 25); /* enable external damage MCH */
506 ctl_set_bit(14, 27); /* enable system recovery MCH */
507 #ifdef CONFIG_MACHCHK_WARNING
508 ctl_set_bit(14, 24); /* enable warning MCH */
509 #endif
510 return 0;
511 }
512
513 /*
514 * Initialize the machine check handler really early to be able to
515 * catch all machine checks that happen during boot
516 */
517 arch_initcall(machine_check_init);
518
519 /*
520 * Machine checks for the channel subsystem must be enabled
521 * after the channel subsystem is initialized
522 */
523 static int __init
524 machine_check_crw_init (void)
525 {
526 struct task_struct *task;
527
528 task = kthread_run(s390_collect_crw_info, &m_sem, "kmcheck");
529 if (IS_ERR(task))
530 return PTR_ERR(task);
531 ctl_set_bit(14, 28); /* enable channel report MCH */
532 return 0;
533 }
534
535 device_initcall (machine_check_crw_init);
Linux kernel - Deliverables
This page took 0.044117 seconds and 5 git commands to generate.