?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Linux 2.6.21-rc3
[linux/fpc-iii.git] / drivers / s390 / s390mach.c
blob806bb1a921eb99e83822a9b2668ba150b26eecf3
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/device.h>
17 #include <linux/kthread.h>
18 #include <asm/etr.h>
19 #include <asm/lowcore.h>
20 #include <asm/cio.h>
21 #include "cio/cio.h"
22 #include "cio/chsc.h"
23 #include "cio/css.h"
24 #include "s390mach.h"
25
26 static struct semaphore m_sem;
27
28 static NORET_TYPE void
29 s390_handle_damage(char *msg)
30 {
31 #ifdef CONFIG_SMP
32 smp_send_stop();
33 #endif
34 disabled_wait((unsigned long) __builtin_return_address(0));
35 for(;;);
36 }
37
38 /*
39 * Retrieve CRWs and call function to handle event.
40 *
41 * Note : we currently process CRWs for io and chsc subchannels only
42 */
43 static int
44 s390_collect_crw_info(void *param)
45 {
46 struct crw crw[2];
47 int ccode, ret, slow;
48 struct semaphore *sem;
49 unsigned int chain;
50
51 sem = (struct semaphore *)param;
52 repeat:
53 down_interruptible(sem);
54 slow = 0;
55 chain = 0;
56 while (1) {
57 if (unlikely(chain > 1)) {
58 struct crw tmp_crw;
59
60 printk(KERN_WARNING"%s: Code does not support more "
61 "than two chained crws; please report to "
62 "linux390@de.ibm.com!\n", __FUNCTION__);
63 ccode = stcrw(&tmp_crw);
64 printk(KERN_WARNING"%s: crw reports slct=%d, oflw=%d, "
65 "chn=%d, rsc=%X, anc=%d, erc=%X, rsid=%X\n",
66 __FUNCTION__, tmp_crw.slct, tmp_crw.oflw,
67 tmp_crw.chn, tmp_crw.rsc, tmp_crw.anc,
68 tmp_crw.erc, tmp_crw.rsid);
69 printk(KERN_WARNING"%s: This was crw number %x in the "
70 "chain\n", __FUNCTION__, chain);
71 if (ccode != 0)
72 break;
73 chain = tmp_crw.chn ? chain + 1 : 0;
74 continue;
75 }
76 ccode = stcrw(&crw[chain]);
77 if (ccode != 0)
78 break;
79 printk(KERN_DEBUG "crw_info : CRW reports slct=%d, oflw=%d, "
80 "chn=%d, rsc=%X, anc=%d, erc=%X, rsid=%X\n",
81 crw[chain].slct, crw[chain].oflw, crw[chain].chn,
82 crw[chain].rsc, crw[chain].anc, crw[chain].erc,
83 crw[chain].rsid);
84 /* Check for overflows. */
85 if (crw[chain].oflw) {
86 pr_debug("%s: crw overflow detected!\n", __FUNCTION__);
87 css_reiterate_subchannels();
88 chain = 0;
89 slow = 1;
90 continue;
91 }
92 switch (crw[chain].rsc) {
93 case CRW_RSC_SCH:
94 if (crw[0].chn && !chain)
95 break;
96 pr_debug("source is subchannel %04X\n", crw[0].rsid);
97 ret = css_process_crw (crw[0].rsid,
98 chain ? crw[1].rsid : 0);
99 if (ret == -EAGAIN)
100 slow = 1;
101 break;
102 case CRW_RSC_MONITOR:
103 pr_debug("source is monitoring facility\n");
104 break;
105 case CRW_RSC_CPATH:
106 pr_debug("source is channel path %02X\n", crw[0].rsid);
107 /*
108 * Check for solicited machine checks. These are
109 * created by reset channel path and need not be
110 * reported to the common I/O layer.
111 */
112 if (crw[chain].slct) {
113 pr_debug("solicited machine check for "
114 "channel path %02X\n", crw[0].rsid);
115 break;
116 }
117 switch (crw[0].erc) {
118 case CRW_ERC_IPARM: /* Path has come. */
119 ret = chp_process_crw(crw[0].rsid, 1);
120 break;
121 case CRW_ERC_PERRI: /* Path has gone. */
122 case CRW_ERC_PERRN:
123 ret = chp_process_crw(crw[0].rsid, 0);
124 break;
125 default:
126 pr_debug("Don't know how to handle erc=%x\n",
127 crw[0].erc);
128 ret = 0;
129 }
130 if (ret == -EAGAIN)
131 slow = 1;
132 break;
133 case CRW_RSC_CONFIG:
134 pr_debug("source is configuration-alert facility\n");
135 break;
136 case CRW_RSC_CSS:
137 pr_debug("source is channel subsystem\n");
138 ret = chsc_process_crw();
139 if (ret == -EAGAIN)
140 slow = 1;
141 break;
142 default:
143 pr_debug("unknown source\n");
144 break;
145 }
146 /* chain is always 0 or 1 here. */
147 chain = crw[chain].chn ? chain + 1 : 0;
148 }
149 if (slow)
150 queue_work(slow_path_wq, &slow_path_work);
151 goto repeat;
152 return 0;
153 }
154
155 struct mcck_struct {
156 int kill_task;
157 int channel_report;
158 int warning;
159 unsigned long long mcck_code;
160 };
161
162 static DEFINE_PER_CPU(struct mcck_struct, cpu_mcck);
163
164 /*
165 * Main machine check handler function. Will be called with interrupts enabled
166 * or disabled and machine checks enabled or disabled.
167 */
168 void
169 s390_handle_mcck(void)
170 {
171 unsigned long flags;
172 struct mcck_struct mcck;
173
174 /*
175 * Disable machine checks and get the current state of accumulated
176 * machine checks. Afterwards delete the old state and enable machine
177 * checks again.
178 */
179 local_irq_save(flags);
180 local_mcck_disable();
181 mcck = __get_cpu_var(cpu_mcck);
182 memset(&__get_cpu_var(cpu_mcck), 0, sizeof(struct mcck_struct));
183 clear_thread_flag(TIF_MCCK_PENDING);
184 local_mcck_enable();
185 local_irq_restore(flags);
186
187 if (mcck.channel_report)
188 up(&m_sem);
189
190 #ifdef CONFIG_MACHCHK_WARNING
191 /*
192 * The warning may remain for a prolonged period on the bare iron.
193 * (actually till the machine is powered off, or until the problem is gone)
194 * So we just stop listening for the WARNING MCH and prevent continuously
195 * being interrupted. One caveat is however, that we must do this per
196 * processor and cannot use the smp version of ctl_clear_bit().
197 * On VM we only get one interrupt per virtally presented machinecheck.
198 * Though one suffices, we may get one interrupt per (virtual) processor.
199 */
200 if (mcck.warning) { /* WARNING pending ? */
201 static int mchchk_wng_posted = 0;
202 /*
203 * Use single machine clear, as we cannot handle smp right now
204 */
205 __ctl_clear_bit(14, 24); /* Disable WARNING MCH */
206 if (xchg(&mchchk_wng_posted, 1) == 0)
207 kill_cad_pid(SIGPWR, 1);
208 }
209 #endif
210
211 if (mcck.kill_task) {
212 local_irq_enable();
213 printk(KERN_EMERG "mcck: Terminating task because of machine "
214 "malfunction (code 0x%016llx).\n", mcck.mcck_code);
215 printk(KERN_EMERG "mcck: task: %s, pid: %d.\n",
216 current->comm, current->pid);
217 do_exit(SIGSEGV);
218 }
219 }
220
221 /*
222 * returns 0 if all registers could be validated
223 * returns 1 otherwise
224 */
225 static int
226 s390_revalidate_registers(struct mci *mci)
227 {
228 int kill_task;
229 u64 tmpclock;
230 u64 zero;
231 void *fpt_save_area, *fpt_creg_save_area;
232
233 kill_task = 0;
234 zero = 0;
235 /* General purpose registers */
236 if (!mci->gr)
237 /*
238 * General purpose registers couldn't be restored and have
239 * unknown contents. Process needs to be terminated.
240 */
241 kill_task = 1;
242
243 /* Revalidate floating point registers */
244 if (!mci->fp)
245 /*
246 * Floating point registers can't be restored and
247 * therefore the process needs to be terminated.
248 */
249 kill_task = 1;
250
251 #ifndef CONFIG_64BIT
252 asm volatile(
253 " ld 0,0(%0)\n"
254 " ld 2,8(%0)\n"
255 " ld 4,16(%0)\n"
256 " ld 6,24(%0)"
257 : : "a" (&S390_lowcore.floating_pt_save_area));
258 #endif
259
260 if (MACHINE_HAS_IEEE) {
261 #ifdef CONFIG_64BIT
262 fpt_save_area = &S390_lowcore.floating_pt_save_area;
263 fpt_creg_save_area = &S390_lowcore.fpt_creg_save_area;
264 #else
265 fpt_save_area = (void *) S390_lowcore.extended_save_area_addr;
266 fpt_creg_save_area = fpt_save_area+128;
267 #endif
268 /* Floating point control register */
269 if (!mci->fc) {
270 /*
271 * Floating point control register can't be restored.
272 * Task will be terminated.
273 */
274 asm volatile("lfpc 0(%0)" : : "a" (&zero), "m" (zero));
275 kill_task = 1;
276
277 } else
278 asm volatile("lfpc 0(%0)" : : "a" (fpt_creg_save_area));
279
280 asm volatile(
281 " ld 0,0(%0)\n"
282 " ld 1,8(%0)\n"
283 " ld 2,16(%0)\n"
284 " ld 3,24(%0)\n"
285 " ld 4,32(%0)\n"
286 " ld 5,40(%0)\n"
287 " ld 6,48(%0)\n"
288 " ld 7,56(%0)\n"
289 " ld 8,64(%0)\n"
290 " ld 9,72(%0)\n"
291 " ld 10,80(%0)\n"
292 " ld 11,88(%0)\n"
293 " ld 12,96(%0)\n"
294 " ld 13,104(%0)\n"
295 " ld 14,112(%0)\n"
296 " ld 15,120(%0)\n"
297 : : "a" (fpt_save_area));
298 }
299
300 /* Revalidate access registers */
301 asm volatile(
302 " lam 0,15,0(%0)"
303 : : "a" (&S390_lowcore.access_regs_save_area));
304 if (!mci->ar)
305 /*
306 * Access registers have unknown contents.
307 * Terminating task.
308 */
309 kill_task = 1;
310
311 /* Revalidate control registers */
312 if (!mci->cr)
313 /*
314 * Control registers have unknown contents.
315 * Can't recover and therefore stopping machine.
316 */
317 s390_handle_damage("invalid control registers.");
318 else
319 #ifdef CONFIG_64BIT
320 asm volatile(
321 " lctlg 0,15,0(%0)"
322 : : "a" (&S390_lowcore.cregs_save_area));
323 #else
324 asm volatile(
325 " lctl 0,15,0(%0)"
326 : : "a" (&S390_lowcore.cregs_save_area));
327 #endif
328
329 /*
330 * We don't even try to revalidate the TOD register, since we simply
331 * can't write something sensible into that register.
332 */
333
334 #ifdef CONFIG_64BIT
335 /*
336 * See if we can revalidate the TOD programmable register with its
337 * old contents (should be zero) otherwise set it to zero.
338 */
339 if (!mci->pr)
340 asm volatile(
341 " sr 0,0\n"
342 " sckpf"
343 : : : "0", "cc");
344 else
345 asm volatile(
346 " l 0,0(%0)\n"
347 " sckpf"
348 : : "a" (&S390_lowcore.tod_progreg_save_area)
349 : "0", "cc");
350 #endif
351
352 /* Revalidate clock comparator register */
353 asm volatile(
354 " stck 0(%1)\n"
355 " sckc 0(%1)"
356 : "=m" (tmpclock) : "a" (&(tmpclock)) : "cc", "memory");
357
358 /* Check if old PSW is valid */
359 if (!mci->wp)
360 /*
361 * Can't tell if we come from user or kernel mode
362 * -> stopping machine.
363 */
364 s390_handle_damage("old psw invalid.");
365
366 if (!mci->ms || !mci->pm || !mci->ia)
367 kill_task = 1;
368
369 return kill_task;
370 }
371
372 #define MAX_IPD_COUNT 29
373 #define MAX_IPD_TIME (5 * 60 * USEC_PER_SEC) /* 5 minutes */
374
375 /*
376 * machine check handler.
377 */
378 void
379 s390_do_machine_check(struct pt_regs *regs)
380 {
381 static DEFINE_SPINLOCK(ipd_lock);
382 static unsigned long long last_ipd;
383 static int ipd_count;
384 unsigned long long tmp;
385 struct mci *mci;
386 struct mcck_struct *mcck;
387 int umode;
388
389 lockdep_off();
390
391 mci = (struct mci *) &S390_lowcore.mcck_interruption_code;
392 mcck = &__get_cpu_var(cpu_mcck);
393 umode = user_mode(regs);
394
395 if (mci->sd)
396 /* System damage -> stopping machine */
397 s390_handle_damage("received system damage machine check.");
398
399 if (mci->pd) {
400 if (mci->b) {
401 /* Processing backup -> verify if we can survive this */
402 u64 z_mcic, o_mcic, t_mcic;
403 #ifdef CONFIG_64BIT
404 z_mcic = (1ULL<<63 | 1ULL<<59 | 1ULL<<29);
405 o_mcic = (1ULL<<43 | 1ULL<<42 | 1ULL<<41 | 1ULL<<40 |
406 1ULL<<36 | 1ULL<<35 | 1ULL<<34 | 1ULL<<32 |
407 1ULL<<30 | 1ULL<<21 | 1ULL<<20 | 1ULL<<17 |
408 1ULL<<16);
409 #else
410 z_mcic = (1ULL<<63 | 1ULL<<59 | 1ULL<<57 | 1ULL<<50 |
411 1ULL<<29);
412 o_mcic = (1ULL<<43 | 1ULL<<42 | 1ULL<<41 | 1ULL<<40 |
413 1ULL<<36 | 1ULL<<35 | 1ULL<<34 | 1ULL<<32 |
414 1ULL<<30 | 1ULL<<20 | 1ULL<<17 | 1ULL<<16);
415 #endif
416 t_mcic = *(u64 *)mci;
417
418 if (((t_mcic & z_mcic) != 0) ||
419 ((t_mcic & o_mcic) != o_mcic)) {
420 s390_handle_damage("processing backup machine "
421 "check with damage.");
422 }
423
424 /*
425 * Nullifying exigent condition, therefore we might
426 * retry this instruction.
427 */
428
429 spin_lock(&ipd_lock);
430
431 tmp = get_clock();
432
433 if (((tmp - last_ipd) >> 12) < MAX_IPD_TIME)
434 ipd_count++;
435 else
436 ipd_count = 1;
437
438 last_ipd = tmp;
439
440 if (ipd_count == MAX_IPD_COUNT)
441 s390_handle_damage("too many ipd retries.");
442
443 spin_unlock(&ipd_lock);
444 }
445 else {
446 /* Processing damage -> stopping machine */
447 s390_handle_damage("received instruction processing "
448 "damage machine check.");
449 }
450 }
451 if (s390_revalidate_registers(mci)) {
452 if (umode) {
453 /*
454 * Couldn't restore all register contents while in
455 * user mode -> mark task for termination.
456 */
457 mcck->kill_task = 1;
458 mcck->mcck_code = *(unsigned long long *) mci;
459 set_thread_flag(TIF_MCCK_PENDING);
460 }
461 else
462 /*
463 * Couldn't restore all register contents while in
464 * kernel mode -> stopping machine.
465 */
466 s390_handle_damage("unable to revalidate registers.");
467 }
468
469 if (mci->cd) {
470 /* Timing facility damage */
471 s390_handle_damage("TOD clock damaged");
472 }
473
474 if (mci->ed && mci->ec) {
475 /* External damage */
476 if (S390_lowcore.external_damage_code & (1U << ED_ETR_SYNC))
477 etr_sync_check();
478 if (S390_lowcore.external_damage_code & (1U << ED_ETR_SWITCH))
479 etr_switch_to_local();
480 }
481
482 if (mci->se)
483 /* Storage error uncorrected */
484 s390_handle_damage("received storage error uncorrected "
485 "machine check.");
486
487 if (mci->ke)
488 /* Storage key-error uncorrected */
489 s390_handle_damage("received storage key-error uncorrected "
490 "machine check.");
491
492 if (mci->ds && mci->fa)
493 /* Storage degradation */
494 s390_handle_damage("received storage degradation machine "
495 "check.");
496
497 if (mci->cp) {
498 /* Channel report word pending */
499 mcck->channel_report = 1;
500 set_thread_flag(TIF_MCCK_PENDING);
501 }
502
503 if (mci->w) {
504 /* Warning pending */
505 mcck->warning = 1;
506 set_thread_flag(TIF_MCCK_PENDING);
507 }
508 lockdep_on();
509 }
510
511 /*
512 * s390_init_machine_check
513 *
514 * initialize machine check handling
515 */
516 static int
517 machine_check_init(void)
518 {
519 init_MUTEX_LOCKED(&m_sem);
520 ctl_set_bit(14, 25); /* enable external damage MCH */
521 ctl_set_bit(14, 27); /* enable system recovery MCH */
522 #ifdef CONFIG_MACHCHK_WARNING
523 ctl_set_bit(14, 24); /* enable warning MCH */
524 #endif
525 return 0;
526 }
527
528 /*
529 * Initialize the machine check handler really early to be able to
530 * catch all machine checks that happen during boot
531 */
532 arch_initcall(machine_check_init);
533
534 /*
535 * Machine checks for the channel subsystem must be enabled
536 * after the channel subsystem is initialized
537 */
538 static int __init
539 machine_check_crw_init (void)
540 {
541 struct task_struct *task;
542
543 task = kthread_run(s390_collect_crw_info, &m_sem, "kmcheck");
544 if (IS_ERR(task))
545 return PTR_ERR(task);
546 ctl_set_bit(14, 28); /* enable channel report MCH */
547 return 0;
548 }
549
550 device_initcall (machine_check_crw_init);
Linux with added FPC-III support