?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
pull master
[hh.org.git] / drivers / s390 / s390mach.c
blobe088b5e287111b7268e322ad62dc838663d4815c
1 /*
2 * drivers/s390/s390mach.c
3 * S/390 machine check handler
4 *
5 * S390 version
6 * Copyright (C) 2000 IBM Deutschland Entwicklung GmbH, IBM Corporation
7 * Author(s): Ingo Adlung (adlung@de.ibm.com)
8 * Martin Schwidefsky (schwidefsky@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/kthread.h>
17
18 #include <asm/lowcore.h>
19
20 #include "s390mach.h"
21
22 static struct semaphore m_sem;
23
24 extern int css_process_crw(int, int);
25 extern int chsc_process_crw(void);
26 extern int chp_process_crw(int, int);
27 extern void css_reiterate_subchannels(void);
28
29 extern struct workqueue_struct *slow_path_wq;
30 extern struct work_struct slow_path_work;
31
32 static NORET_TYPE void
33 s390_handle_damage(char *msg)
34 {
35 #ifdef CONFIG_SMP
36 smp_send_stop();
37 #endif
38 disabled_wait((unsigned long) __builtin_return_address(0));
39 for(;;);
40 }
41
42 /*
43 * Retrieve CRWs and call function to handle event.
44 *
45 * Note : we currently process CRWs for io and chsc subchannels only
46 */
47 static int
48 s390_collect_crw_info(void *param)
49 {
50 struct crw crw[2];
51 int ccode, ret, slow;
52 struct semaphore *sem;
53 unsigned int chain;
54
55 sem = (struct semaphore *)param;
56 repeat:
57 down_interruptible(sem);
58 slow = 0;
59 chain = 0;
60 while (1) {
61 if (unlikely(chain > 1)) {
62 struct crw tmp_crw;
63
64 printk(KERN_WARNING"%s: Code does not support more "
65 "than two chained crws; please report to "
66 "linux390@de.ibm.com!\n", __FUNCTION__);
67 ccode = stcrw(&tmp_crw);
68 printk(KERN_WARNING"%s: crw reports slct=%d, oflw=%d, "
69 "chn=%d, rsc=%X, anc=%d, erc=%X, rsid=%X\n",
70 __FUNCTION__, tmp_crw.slct, tmp_crw.oflw,
71 tmp_crw.chn, tmp_crw.rsc, tmp_crw.anc,
72 tmp_crw.erc, tmp_crw.rsid);
73 printk(KERN_WARNING"%s: This was crw number %x in the "
74 "chain\n", __FUNCTION__, chain);
75 if (ccode != 0)
76 break;
77 chain = tmp_crw.chn ? chain + 1 : 0;
78 continue;
79 }
80 ccode = stcrw(&crw[chain]);
81 if (ccode != 0)
82 break;
83 printk(KERN_DEBUG "crw_info : CRW reports slct=%d, oflw=%d, "
84 "chn=%d, rsc=%X, anc=%d, erc=%X, rsid=%X\n",
85 crw[chain].slct, crw[chain].oflw, crw[chain].chn,
86 crw[chain].rsc, crw[chain].anc, crw[chain].erc,
87 crw[chain].rsid);
88 /* Check for overflows. */
89 if (crw[chain].oflw) {
90 pr_debug("%s: crw overflow detected!\n", __FUNCTION__);
91 css_reiterate_subchannels();
92 chain = 0;
93 slow = 1;
94 continue;
95 }
96 switch (crw[chain].rsc) {
97 case CRW_RSC_SCH:
98 if (crw[0].chn && !chain)
99 break;
100 pr_debug("source is subchannel %04X\n", crw[0].rsid);
101 ret = css_process_crw (crw[0].rsid,
102 chain ? crw[1].rsid : 0);
103 if (ret == -EAGAIN)
104 slow = 1;
105 break;
106 case CRW_RSC_MONITOR:
107 pr_debug("source is monitoring facility\n");
108 break;
109 case CRW_RSC_CPATH:
110 pr_debug("source is channel path %02X\n", crw[0].rsid);
111 /*
112 * Check for solicited machine checks. These are
113 * created by reset channel path and need not be
114 * reported to the common I/O layer.
115 */
116 if (crw[chain].slct) {
117 pr_debug("solicited machine check for "
118 "channel path %02X\n", crw[0].rsid);
119 break;
120 }
121 switch (crw[0].erc) {
122 case CRW_ERC_IPARM: /* Path has come. */
123 ret = chp_process_crw(crw[0].rsid, 1);
124 break;
125 case CRW_ERC_PERRI: /* Path has gone. */
126 case CRW_ERC_PERRN:
127 ret = chp_process_crw(crw[0].rsid, 0);
128 break;
129 default:
130 pr_debug("Don't know how to handle erc=%x\n",
131 crw[0].erc);
132 ret = 0;
133 }
134 if (ret == -EAGAIN)
135 slow = 1;
136 break;
137 case CRW_RSC_CONFIG:
138 pr_debug("source is configuration-alert facility\n");
139 break;
140 case CRW_RSC_CSS:
141 pr_debug("source is channel subsystem\n");
142 ret = chsc_process_crw();
143 if (ret == -EAGAIN)
144 slow = 1;
145 break;
146 default:
147 pr_debug("unknown source\n");
148 break;
149 }
150 /* chain is always 0 or 1 here. */
151 chain = crw[chain].chn ? chain + 1 : 0;
152 }
153 if (slow)
154 queue_work(slow_path_wq, &slow_path_work);
155 goto repeat;
156 return 0;
157 }
158
159 struct mcck_struct {
160 int kill_task;
161 int channel_report;
162 int warning;
163 unsigned long long mcck_code;
164 };
165
166 static DEFINE_PER_CPU(struct mcck_struct, cpu_mcck);
167
168 /*
169 * Main machine check handler function. Will be called with interrupts enabled
170 * or disabled and machine checks enabled or disabled.
171 */
172 void
173 s390_handle_mcck(void)
174 {
175 unsigned long flags;
176 struct mcck_struct mcck;
177
178 /*
179 * Disable machine checks and get the current state of accumulated
180 * machine checks. Afterwards delete the old state and enable machine
181 * checks again.
182 */
183 local_irq_save(flags);
184 local_mcck_disable();
185 mcck = __get_cpu_var(cpu_mcck);
186 memset(&__get_cpu_var(cpu_mcck), 0, sizeof(struct mcck_struct));
187 clear_thread_flag(TIF_MCCK_PENDING);
188 local_mcck_enable();
189 local_irq_restore(flags);
190
191 if (mcck.channel_report)
192 up(&m_sem);
193
194 #ifdef CONFIG_MACHCHK_WARNING
195 /*
196 * The warning may remain for a prolonged period on the bare iron.
197 * (actually till the machine is powered off, or until the problem is gone)
198 * So we just stop listening for the WARNING MCH and prevent continuously
199 * being interrupted. One caveat is however, that we must do this per
200 * processor and cannot use the smp version of ctl_clear_bit().
201 * On VM we only get one interrupt per virtally presented machinecheck.
202 * Though one suffices, we may get one interrupt per (virtual) processor.
203 */
204 if (mcck.warning) { /* WARNING pending ? */
205 static int mchchk_wng_posted = 0;
206 /*
207 * Use single machine clear, as we cannot handle smp right now
208 */
209 __ctl_clear_bit(14, 24); /* Disable WARNING MCH */
210 if (xchg(&mchchk_wng_posted, 1) == 0)
211 kill_cad_pid(SIGPWR, 1);
212 }
213 #endif
214
215 if (mcck.kill_task) {
216 local_irq_enable();
217 printk(KERN_EMERG "mcck: Terminating task because of machine "
218 "malfunction (code 0x%016llx).\n", mcck.mcck_code);
219 printk(KERN_EMERG "mcck: task: %s, pid: %d.\n",
220 current->comm, current->pid);
221 do_exit(SIGSEGV);
222 }
223 }
224
225 /*
226 * returns 0 if all registers could be validated
227 * returns 1 otherwise
228 */
229 static int
230 s390_revalidate_registers(struct mci *mci)
231 {
232 int kill_task;
233 u64 tmpclock;
234 u64 zero;
235 void *fpt_save_area, *fpt_creg_save_area;
236
237 kill_task = 0;
238 zero = 0;
239 /* General purpose registers */
240 if (!mci->gr)
241 /*
242 * General purpose registers couldn't be restored and have
243 * unknown contents. Process needs to be terminated.
244 */
245 kill_task = 1;
246
247 /* Revalidate floating point registers */
248 if (!mci->fp)
249 /*
250 * Floating point registers can't be restored and
251 * therefore the process needs to be terminated.
252 */
253 kill_task = 1;
254
255 #ifndef CONFIG_64BIT
256 asm volatile(
257 " ld 0,0(%0)\n"
258 " ld 2,8(%0)\n"
259 " ld 4,16(%0)\n"
260 " ld 6,24(%0)"
261 : : "a" (&S390_lowcore.floating_pt_save_area));
262 #endif
263
264 if (MACHINE_HAS_IEEE) {
265 #ifdef CONFIG_64BIT
266 fpt_save_area = &S390_lowcore.floating_pt_save_area;
267 fpt_creg_save_area = &S390_lowcore.fpt_creg_save_area;
268 #else
269 fpt_save_area = (void *) S390_lowcore.extended_save_area_addr;
270 fpt_creg_save_area = fpt_save_area+128;
271 #endif
272 /* Floating point control register */
273 if (!mci->fc) {
274 /*
275 * Floating point control register can't be restored.
276 * Task will be terminated.
277 */
278 asm volatile("lfpc 0(%0)" : : "a" (&zero), "m" (zero));
279 kill_task = 1;
280
281 } else
282 asm volatile("lfpc 0(%0)" : : "a" (fpt_creg_save_area));
283
284 asm volatile(
285 " ld 0,0(%0)\n"
286 " ld 1,8(%0)\n"
287 " ld 2,16(%0)\n"
288 " ld 3,24(%0)\n"
289 " ld 4,32(%0)\n"
290 " ld 5,40(%0)\n"
291 " ld 6,48(%0)\n"
292 " ld 7,56(%0)\n"
293 " ld 8,64(%0)\n"
294 " ld 9,72(%0)\n"
295 " ld 10,80(%0)\n"
296 " ld 11,88(%0)\n"
297 " ld 12,96(%0)\n"
298 " ld 13,104(%0)\n"
299 " ld 14,112(%0)\n"
300 " ld 15,120(%0)\n"
301 : : "a" (fpt_save_area));
302 }
303
304 /* Revalidate access registers */
305 asm volatile(
306 " lam 0,15,0(%0)"
307 : : "a" (&S390_lowcore.access_regs_save_area));
308 if (!mci->ar)
309 /*
310 * Access registers have unknown contents.
311 * Terminating task.
312 */
313 kill_task = 1;
314
315 /* Revalidate control registers */
316 if (!mci->cr)
317 /*
318 * Control registers have unknown contents.
319 * Can't recover and therefore stopping machine.
320 */
321 s390_handle_damage("invalid control registers.");
322 else
323 #ifdef CONFIG_64BIT
324 asm volatile(
325 " lctlg 0,15,0(%0)"
326 : : "a" (&S390_lowcore.cregs_save_area));
327 #else
328 asm volatile(
329 " lctl 0,15,0(%0)"
330 : : "a" (&S390_lowcore.cregs_save_area));
331 #endif
332
333 /*
334 * We don't even try to revalidate the TOD register, since we simply
335 * can't write something sensible into that register.
336 */
337
338 #ifdef CONFIG_64BIT
339 /*
340 * See if we can revalidate the TOD programmable register with its
341 * old contents (should be zero) otherwise set it to zero.
342 */
343 if (!mci->pr)
344 asm volatile(
345 " sr 0,0\n"
346 " sckpf"
347 : : : "0", "cc");
348 else
349 asm volatile(
350 " l 0,0(%0)\n"
351 " sckpf"
352 : : "a" (&S390_lowcore.tod_progreg_save_area)
353 : "0", "cc");
354 #endif
355
356 /* Revalidate clock comparator register */
357 asm volatile(
358 " stck 0(%1)\n"
359 " sckc 0(%1)"
360 : "=m" (tmpclock) : "a" (&(tmpclock)) : "cc", "memory");
361
362 /* Check if old PSW is valid */
363 if (!mci->wp)
364 /*
365 * Can't tell if we come from user or kernel mode
366 * -> stopping machine.
367 */
368 s390_handle_damage("old psw invalid.");
369
370 if (!mci->ms || !mci->pm || !mci->ia)
371 kill_task = 1;
372
373 return kill_task;
374 }
375
376 #define MAX_IPD_COUNT 29
377 #define MAX_IPD_TIME (5 * 60 * USEC_PER_SEC) /* 5 minutes */
378
379 /*
380 * machine check handler.
381 */
382 void
383 s390_do_machine_check(struct pt_regs *regs)
384 {
385 static DEFINE_SPINLOCK(ipd_lock);
386 static unsigned long long last_ipd;
387 static int ipd_count;
388 unsigned long long tmp;
389 struct mci *mci;
390 struct mcck_struct *mcck;
391 int umode;
392
393 lockdep_off();
394
395 mci = (struct mci *) &S390_lowcore.mcck_interruption_code;
396 mcck = &__get_cpu_var(cpu_mcck);
397 umode = user_mode(regs);
398
399 if (mci->sd)
400 /* System damage -> stopping machine */
401 s390_handle_damage("received system damage machine check.");
402
403 if (mci->pd) {
404 if (mci->b) {
405 /* Processing backup -> verify if we can survive this */
406 u64 z_mcic, o_mcic, t_mcic;
407 #ifdef CONFIG_64BIT
408 z_mcic = (1ULL<<63 | 1ULL<<59 | 1ULL<<29);
409 o_mcic = (1ULL<<43 | 1ULL<<42 | 1ULL<<41 | 1ULL<<40 |
410 1ULL<<36 | 1ULL<<35 | 1ULL<<34 | 1ULL<<32 |
411 1ULL<<30 | 1ULL<<21 | 1ULL<<20 | 1ULL<<17 |
412 1ULL<<16);
413 #else
414 z_mcic = (1ULL<<63 | 1ULL<<59 | 1ULL<<57 | 1ULL<<50 |
415 1ULL<<29);
416 o_mcic = (1ULL<<43 | 1ULL<<42 | 1ULL<<41 | 1ULL<<40 |
417 1ULL<<36 | 1ULL<<35 | 1ULL<<34 | 1ULL<<32 |
418 1ULL<<30 | 1ULL<<20 | 1ULL<<17 | 1ULL<<16);
419 #endif
420 t_mcic = *(u64 *)mci;
421
422 if (((t_mcic & z_mcic) != 0) ||
423 ((t_mcic & o_mcic) != o_mcic)) {
424 s390_handle_damage("processing backup machine "
425 "check with damage.");
426 }
427
428 /*
429 * Nullifying exigent condition, therefore we might
430 * retry this instruction.
431 */
432
433 spin_lock(&ipd_lock);
434
435 tmp = get_clock();
436
437 if (((tmp - last_ipd) >> 12) < MAX_IPD_TIME)
438 ipd_count++;
439 else
440 ipd_count = 1;
441
442 last_ipd = tmp;
443
444 if (ipd_count == MAX_IPD_COUNT)
445 s390_handle_damage("too many ipd retries.");
446
447 spin_unlock(&ipd_lock);
448 }
449 else {
450 /* Processing damage -> stopping machine */
451 s390_handle_damage("received instruction processing "
452 "damage machine check.");
453 }
454 }
455 if (s390_revalidate_registers(mci)) {
456 if (umode) {
457 /*
458 * Couldn't restore all register contents while in
459 * user mode -> mark task for termination.
460 */
461 mcck->kill_task = 1;
462 mcck->mcck_code = *(unsigned long long *) mci;
463 set_thread_flag(TIF_MCCK_PENDING);
464 }
465 else
466 /*
467 * Couldn't restore all register contents while in
468 * kernel mode -> stopping machine.
469 */
470 s390_handle_damage("unable to revalidate registers.");
471 }
472
473 if (mci->se)
474 /* Storage error uncorrected */
475 s390_handle_damage("received storage error uncorrected "
476 "machine check.");
477
478 if (mci->ke)
479 /* Storage key-error uncorrected */
480 s390_handle_damage("received storage key-error uncorrected "
481 "machine check.");
482
483 if (mci->ds && mci->fa)
484 /* Storage degradation */
485 s390_handle_damage("received storage degradation machine "
486 "check.");
487
488 if (mci->cp) {
489 /* Channel report word pending */
490 mcck->channel_report = 1;
491 set_thread_flag(TIF_MCCK_PENDING);
492 }
493
494 if (mci->w) {
495 /* Warning pending */
496 mcck->warning = 1;
497 set_thread_flag(TIF_MCCK_PENDING);
498 }
499 lockdep_on();
500 }
501
502 /*
503 * s390_init_machine_check
504 *
505 * initialize machine check handling
506 */
507 static int
508 machine_check_init(void)
509 {
510 init_MUTEX_LOCKED(&m_sem);
511 ctl_clear_bit(14, 25); /* disable external damage MCH */
512 ctl_set_bit(14, 27); /* enable system recovery MCH */
513 #ifdef CONFIG_MACHCHK_WARNING
514 ctl_set_bit(14, 24); /* enable warning MCH */
515 #endif
516 return 0;
517 }
518
519 /*
520 * Initialize the machine check handler really early to be able to
521 * catch all machine checks that happen during boot
522 */
523 arch_initcall(machine_check_init);
524
525 /*
526 * Machine checks for the channel subsystem must be enabled
527 * after the channel subsystem is initialized
528 */
529 static int __init
530 machine_check_crw_init (void)
531 {
532 kthread_run(s390_collect_crw_info, &m_sem, "kmcheck");
533 ctl_set_bit(14, 28); /* enable channel report MCH */
534 return 0;
535 }
536
537 device_initcall (machine_check_crw_init);
hh.org GIT migration