blob: 3bf466603512a32f2c6a933525492eed58963f33 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
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/config.h>
12#include <linux/init.h>
13#include <linux/sched.h>
14#include <linux/errno.h>
15#include <linux/workqueue.h>
16
17#include <asm/lowcore.h>
18
19#include "s390mach.h"
20
21#define DBG printk
22// #define DBG(args,...) do {} while (0);
23
24static struct semaphore m_sem;
25
Cornelia Huckfb6958a2006-01-06 00:19:25 -080026extern int css_process_crw(int, int);
Linus Torvalds1da177e2005-04-16 15:20:36 -070027extern int chsc_process_crw(void);
28extern int chp_process_crw(int, int);
29extern void css_reiterate_subchannels(void);
30
31extern struct workqueue_struct *slow_path_wq;
32extern struct work_struct slow_path_work;
33
Heiko Carstens77fa2242005-06-25 14:55:30 -070034static NORET_TYPE void
Linus Torvalds1da177e2005-04-16 15:20:36 -070035s390_handle_damage(char *msg)
36{
Linus Torvalds1da177e2005-04-16 15:20:36 -070037#ifdef CONFIG_SMP
38 smp_send_stop();
39#endif
40 disabled_wait((unsigned long) __builtin_return_address(0));
Heiko Carstens77fa2242005-06-25 14:55:30 -070041 for(;;);
Linus Torvalds1da177e2005-04-16 15:20:36 -070042}
43
44/*
45 * Retrieve CRWs and call function to handle event.
46 *
47 * Note : we currently process CRWs for io and chsc subchannels only
48 */
49static int
50s390_collect_crw_info(void *param)
51{
Cornelia Huckfb6958a2006-01-06 00:19:25 -080052 struct crw crw[2];
Linus Torvalds1da177e2005-04-16 15:20:36 -070053 int ccode, ret, slow;
54 struct semaphore *sem;
Cornelia Huckfb6958a2006-01-06 00:19:25 -080055 unsigned int chain;
Linus Torvalds1da177e2005-04-16 15:20:36 -070056
57 sem = (struct semaphore *)param;
58 /* Set a nice name. */
59 daemonize("kmcheck");
60repeat:
61 down_interruptible(sem);
62 slow = 0;
Cornelia Huckfb6958a2006-01-06 00:19:25 -080063 chain = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -070064 while (1) {
Cornelia Huckfb6958a2006-01-06 00:19:25 -080065 if (unlikely(chain > 1)) {
66 struct crw tmp_crw;
67
68 printk(KERN_WARNING"%s: Code does not support more "
69 "than two chained crws; please report to "
70 "linux390@de.ibm.com!\n", __FUNCTION__);
71 ccode = stcrw(&tmp_crw);
72 printk(KERN_WARNING"%s: crw reports slct=%d, oflw=%d, "
73 "chn=%d, rsc=%X, anc=%d, erc=%X, rsid=%X\n",
74 __FUNCTION__, tmp_crw.slct, tmp_crw.oflw,
75 tmp_crw.chn, tmp_crw.rsc, tmp_crw.anc,
76 tmp_crw.erc, tmp_crw.rsid);
77 printk(KERN_WARNING"%s: This was crw number %x in the "
78 "chain\n", __FUNCTION__, chain);
79 if (ccode != 0)
80 break;
81 chain = tmp_crw.chn ? chain + 1 : 0;
82 continue;
83 }
84 ccode = stcrw(&crw[chain]);
Linus Torvalds1da177e2005-04-16 15:20:36 -070085 if (ccode != 0)
86 break;
87 DBG(KERN_DEBUG "crw_info : CRW reports slct=%d, oflw=%d, "
88 "chn=%d, rsc=%X, anc=%d, erc=%X, rsid=%X\n",
Cornelia Huckfb6958a2006-01-06 00:19:25 -080089 crw[chain].slct, crw[chain].oflw, crw[chain].chn,
90 crw[chain].rsc, crw[chain].anc, crw[chain].erc,
91 crw[chain].rsid);
Linus Torvalds1da177e2005-04-16 15:20:36 -070092 /* Check for overflows. */
Cornelia Huckfb6958a2006-01-06 00:19:25 -080093 if (crw[chain].oflw) {
Linus Torvalds1da177e2005-04-16 15:20:36 -070094 pr_debug("%s: crw overflow detected!\n", __FUNCTION__);
95 css_reiterate_subchannels();
Cornelia Huckfb6958a2006-01-06 00:19:25 -080096 chain = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -070097 slow = 1;
98 continue;
99 }
Cornelia Huckfb6958a2006-01-06 00:19:25 -0800100 switch (crw[chain].rsc) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700101 case CRW_RSC_SCH:
Cornelia Huckfb6958a2006-01-06 00:19:25 -0800102 if (crw[0].chn && !chain)
103 break;
104 pr_debug("source is subchannel %04X\n", crw[0].rsid);
105 ret = css_process_crw (crw[0].rsid,
106 chain ? crw[1].rsid : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700107 if (ret == -EAGAIN)
108 slow = 1;
109 break;
110 case CRW_RSC_MONITOR:
111 pr_debug("source is monitoring facility\n");
112 break;
113 case CRW_RSC_CPATH:
Cornelia Huckfb6958a2006-01-06 00:19:25 -0800114 pr_debug("source is channel path %02X\n", crw[0].rsid);
115 switch (crw[0].erc) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700116 case CRW_ERC_IPARM: /* Path has come. */
Cornelia Huckfb6958a2006-01-06 00:19:25 -0800117 ret = chp_process_crw(crw[0].rsid, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700118 break;
119 case CRW_ERC_PERRI: /* Path has gone. */
120 case CRW_ERC_PERRN:
Cornelia Huckfb6958a2006-01-06 00:19:25 -0800121 ret = chp_process_crw(crw[0].rsid, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700122 break;
123 default:
124 pr_debug("Don't know how to handle erc=%x\n",
Cornelia Huckfb6958a2006-01-06 00:19:25 -0800125 crw[0].erc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700126 ret = 0;
127 }
128 if (ret == -EAGAIN)
129 slow = 1;
130 break;
131 case CRW_RSC_CONFIG:
132 pr_debug("source is configuration-alert facility\n");
133 break;
134 case CRW_RSC_CSS:
135 pr_debug("source is channel subsystem\n");
136 ret = chsc_process_crw();
137 if (ret == -EAGAIN)
138 slow = 1;
139 break;
140 default:
141 pr_debug("unknown source\n");
142 break;
143 }
Cornelia Huckfb6958a2006-01-06 00:19:25 -0800144 /* chain is always 0 or 1 here. */
145 chain = crw[chain].chn ? chain + 1 : 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700146 }
147 if (slow)
148 queue_work(slow_path_wq, &slow_path_work);
149 goto repeat;
150 return 0;
151}
152
Heiko Carstens77fa2242005-06-25 14:55:30 -0700153struct mcck_struct {
154 int kill_task;
155 int channel_report;
156 int warning;
157 unsigned long long mcck_code;
158};
159
160static DEFINE_PER_CPU(struct mcck_struct, cpu_mcck);
161
Linus Torvalds1da177e2005-04-16 15:20:36 -0700162/*
Heiko Carstens77fa2242005-06-25 14:55:30 -0700163 * Main machine check handler function. Will be called with interrupts enabled
164 * or disabled and machine checks enabled or disabled.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700165 */
166void
Heiko Carstens77fa2242005-06-25 14:55:30 -0700167s390_handle_mcck(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700168{
Heiko Carstens77fa2242005-06-25 14:55:30 -0700169 unsigned long flags;
170 struct mcck_struct mcck;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700171
Heiko Carstens77fa2242005-06-25 14:55:30 -0700172 /*
173 * Disable machine checks and get the current state of accumulated
174 * machine checks. Afterwards delete the old state and enable machine
175 * checks again.
176 */
177 local_irq_save(flags);
178 local_mcck_disable();
179 mcck = __get_cpu_var(cpu_mcck);
180 memset(&__get_cpu_var(cpu_mcck), 0, sizeof(struct mcck_struct));
181 clear_thread_flag(TIF_MCCK_PENDING);
182 local_mcck_enable();
183 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700184
Heiko Carstens77fa2242005-06-25 14:55:30 -0700185 if (mcck.channel_report)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700186 up(&m_sem);
187
188#ifdef CONFIG_MACHCHK_WARNING
189/*
190 * The warning may remain for a prolonged period on the bare iron.
191 * (actually till the machine is powered off, or until the problem is gone)
192 * So we just stop listening for the WARNING MCH and prevent continuously
193 * being interrupted. One caveat is however, that we must do this per
194 * processor and cannot use the smp version of ctl_clear_bit().
195 * On VM we only get one interrupt per virtally presented machinecheck.
196 * Though one suffices, we may get one interrupt per (virtual) processor.
197 */
Heiko Carstens77fa2242005-06-25 14:55:30 -0700198 if (mcck.warning) { /* WARNING pending ? */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700199 static int mchchk_wng_posted = 0;
200 /*
201 * Use single machine clear, as we cannot handle smp right now
202 */
203 __ctl_clear_bit(14, 24); /* Disable WARNING MCH */
204 if (xchg(&mchchk_wng_posted, 1) == 0)
205 kill_proc(1, SIGPWR, 1);
206 }
207#endif
Heiko Carstens77fa2242005-06-25 14:55:30 -0700208
209 if (mcck.kill_task) {
210 local_irq_enable();
211 printk(KERN_EMERG "mcck: Terminating task because of machine "
212 "malfunction (code 0x%016llx).\n", mcck.mcck_code);
213 printk(KERN_EMERG "mcck: task: %s, pid: %d.\n",
214 current->comm, current->pid);
215 do_exit(SIGSEGV);
216 }
217}
218
219/*
220 * returns 0 if all registers could be validated
221 * returns 1 otherwise
222 */
223static int
224s390_revalidate_registers(struct mci *mci)
225{
226 int kill_task;
227 u64 tmpclock;
228 u64 zero;
229 void *fpt_save_area, *fpt_creg_save_area;
230
231 kill_task = 0;
232 zero = 0;
233 /* General purpose registers */
234 if (!mci->gr)
235 /*
236 * General purpose registers couldn't be restored and have
237 * unknown contents. Process needs to be terminated.
238 */
239 kill_task = 1;
240
241 /* Revalidate floating point registers */
242 if (!mci->fp)
243 /*
244 * Floating point registers can't be restored and
245 * therefore the process needs to be terminated.
246 */
247 kill_task = 1;
248
Martin Schwidefsky347a8dc2006-01-06 00:19:28 -0800249#ifndef CONFIG_64BIT
Heiko Carstens77fa2242005-06-25 14:55:30 -0700250 asm volatile("ld 0,0(%0)\n"
251 "ld 2,8(%0)\n"
252 "ld 4,16(%0)\n"
253 "ld 6,24(%0)"
254 : : "a" (&S390_lowcore.floating_pt_save_area));
255#endif
256
257 if (MACHINE_HAS_IEEE) {
Martin Schwidefsky347a8dc2006-01-06 00:19:28 -0800258#ifdef CONFIG_64BIT
Heiko Carstens77fa2242005-06-25 14:55:30 -0700259 fpt_save_area = &S390_lowcore.floating_pt_save_area;
260 fpt_creg_save_area = &S390_lowcore.fpt_creg_save_area;
261#else
262 fpt_save_area = (void *) S390_lowcore.extended_save_area_addr;
263 fpt_creg_save_area = fpt_save_area+128;
264#endif
265 /* Floating point control register */
266 if (!mci->fc) {
267 /*
268 * Floating point control register can't be restored.
269 * Task will be terminated.
270 */
Martin Schwidefskyae6aa2e2005-09-03 15:57:56 -0700271 asm volatile ("lfpc 0(%0)" : : "a" (&zero), "m" (zero));
Heiko Carstens77fa2242005-06-25 14:55:30 -0700272 kill_task = 1;
273
274 }
275 else
276 asm volatile (
277 "lfpc 0(%0)"
278 : : "a" (fpt_creg_save_area));
279
280 asm volatile("ld 0,0(%0)\n"
281 "ld 1,8(%0)\n"
282 "ld 2,16(%0)\n"
283 "ld 3,24(%0)\n"
284 "ld 4,32(%0)\n"
285 "ld 5,40(%0)\n"
286 "ld 6,48(%0)\n"
287 "ld 7,56(%0)\n"
288 "ld 8,64(%0)\n"
289 "ld 9,72(%0)\n"
290 "ld 10,80(%0)\n"
291 "ld 11,88(%0)\n"
292 "ld 12,96(%0)\n"
293 "ld 13,104(%0)\n"
294 "ld 14,112(%0)\n"
295 "ld 15,120(%0)\n"
296 : : "a" (fpt_save_area));
297 }
298
299 /* Revalidate access registers */
300 asm volatile("lam 0,15,0(%0)"
301 : : "a" (&S390_lowcore.access_regs_save_area));
302 if (!mci->ar)
303 /*
304 * Access registers have unknown contents.
305 * Terminating task.
306 */
307 kill_task = 1;
308
309 /* Revalidate control registers */
310 if (!mci->cr)
311 /*
312 * Control registers have unknown contents.
313 * Can't recover and therefore stopping machine.
314 */
315 s390_handle_damage("invalid control registers.");
316 else
Martin Schwidefsky347a8dc2006-01-06 00:19:28 -0800317#ifdef CONFIG_64BIT
Heiko Carstens77fa2242005-06-25 14:55:30 -0700318 asm volatile("lctlg 0,15,0(%0)"
319 : : "a" (&S390_lowcore.cregs_save_area));
320#else
321 asm volatile("lctl 0,15,0(%0)"
322 : : "a" (&S390_lowcore.cregs_save_area));
323#endif
324
325 /*
326 * We don't even try to revalidate the TOD register, since we simply
327 * can't write something sensible into that register.
328 */
329
Martin Schwidefsky347a8dc2006-01-06 00:19:28 -0800330#ifdef CONFIG_64BIT
Heiko Carstens77fa2242005-06-25 14:55:30 -0700331 /*
332 * See if we can revalidate the TOD programmable register with its
333 * old contents (should be zero) otherwise set it to zero.
334 */
335 if (!mci->pr)
336 asm volatile("sr 0,0\n"
337 "sckpf"
338 : : : "0", "cc");
339 else
340 asm volatile(
341 "l 0,0(%0)\n"
342 "sckpf"
343 : : "a" (&S390_lowcore.tod_progreg_save_area) : "0", "cc");
344#endif
345
346 /* Revalidate clock comparator register */
347 asm volatile ("stck 0(%1)\n"
348 "sckc 0(%1)"
349 : "=m" (tmpclock) : "a" (&(tmpclock)) : "cc", "memory");
350
351 /* Check if old PSW is valid */
352 if (!mci->wp)
353 /*
354 * Can't tell if we come from user or kernel mode
355 * -> stopping machine.
356 */
357 s390_handle_damage("old psw invalid.");
358
359 if (!mci->ms || !mci->pm || !mci->ia)
360 kill_task = 1;
361
362 return kill_task;
363}
364
365/*
366 * machine check handler.
367 */
368void
369s390_do_machine_check(struct pt_regs *regs)
370{
371 struct mci *mci;
372 struct mcck_struct *mcck;
373 int umode;
374
375 mci = (struct mci *) &S390_lowcore.mcck_interruption_code;
376 mcck = &__get_cpu_var(cpu_mcck);
377 umode = user_mode(regs);
378
379 if (mci->sd)
380 /* System damage -> stopping machine */
381 s390_handle_damage("received system damage machine check.");
382
383 if (mci->pd) {
384 if (mci->b) {
385 /* Processing backup -> verify if we can survive this */
386 u64 z_mcic, o_mcic, t_mcic;
Martin Schwidefsky347a8dc2006-01-06 00:19:28 -0800387#ifdef CONFIG_64BIT
Heiko Carstens77fa2242005-06-25 14:55:30 -0700388 z_mcic = (1ULL<<63 | 1ULL<<59 | 1ULL<<29);
389 o_mcic = (1ULL<<43 | 1ULL<<42 | 1ULL<<41 | 1ULL<<40 |
390 1ULL<<36 | 1ULL<<35 | 1ULL<<34 | 1ULL<<32 |
391 1ULL<<30 | 1ULL<<21 | 1ULL<<20 | 1ULL<<17 |
392 1ULL<<16);
393#else
394 z_mcic = (1ULL<<63 | 1ULL<<59 | 1ULL<<57 | 1ULL<<50 |
395 1ULL<<29);
396 o_mcic = (1ULL<<43 | 1ULL<<42 | 1ULL<<41 | 1ULL<<40 |
397 1ULL<<36 | 1ULL<<35 | 1ULL<<34 | 1ULL<<32 |
398 1ULL<<30 | 1ULL<<20 | 1ULL<<17 | 1ULL<<16);
399#endif
400 t_mcic = *(u64 *)mci;
401
402 if (((t_mcic & z_mcic) != 0) ||
403 ((t_mcic & o_mcic) != o_mcic)) {
404 s390_handle_damage("processing backup machine "
405 "check with damage.");
406 }
407 if (!umode)
408 s390_handle_damage("processing backup machine "
409 "check in kernel mode.");
410 mcck->kill_task = 1;
411 mcck->mcck_code = *(unsigned long long *) mci;
412 }
413 else {
414 /* Processing damage -> stopping machine */
415 s390_handle_damage("received instruction processing "
416 "damage machine check.");
417 }
418 }
419 if (s390_revalidate_registers(mci)) {
420 if (umode) {
421 /*
422 * Couldn't restore all register contents while in
423 * user mode -> mark task for termination.
424 */
425 mcck->kill_task = 1;
426 mcck->mcck_code = *(unsigned long long *) mci;
427 set_thread_flag(TIF_MCCK_PENDING);
428 }
429 else
430 /*
431 * Couldn't restore all register contents while in
432 * kernel mode -> stopping machine.
433 */
434 s390_handle_damage("unable to revalidate registers.");
435 }
436
437 if (mci->se)
438 /* Storage error uncorrected */
439 s390_handle_damage("received storage error uncorrected "
440 "machine check.");
441
442 if (mci->ke)
443 /* Storage key-error uncorrected */
444 s390_handle_damage("received storage key-error uncorrected "
445 "machine check.");
446
447 if (mci->ds && mci->fa)
448 /* Storage degradation */
449 s390_handle_damage("received storage degradation machine "
450 "check.");
451
452 if (mci->cp) {
453 /* Channel report word pending */
454 mcck->channel_report = 1;
455 set_thread_flag(TIF_MCCK_PENDING);
456 }
457
458 if (mci->w) {
459 /* Warning pending */
460 mcck->warning = 1;
461 set_thread_flag(TIF_MCCK_PENDING);
462 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700463}
464
465/*
466 * s390_init_machine_check
467 *
468 * initialize machine check handling
469 */
470static int
471machine_check_init(void)
472{
473 init_MUTEX_LOCKED(&m_sem);
Heiko Carstens77fa2242005-06-25 14:55:30 -0700474 ctl_clear_bit(14, 25); /* disable external damage MCH */
475 ctl_set_bit(14, 27); /* enable system recovery MCH */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700476#ifdef CONFIG_MACHCHK_WARNING
477 ctl_set_bit(14, 24); /* enable warning MCH */
478#endif
479 return 0;
480}
481
482/*
483 * Initialize the machine check handler really early to be able to
484 * catch all machine checks that happen during boot
485 */
486arch_initcall(machine_check_init);
487
488/*
489 * Machine checks for the channel subsystem must be enabled
490 * after the channel subsystem is initialized
491 */
492static int __init
493machine_check_crw_init (void)
494{
495 kernel_thread(s390_collect_crw_info, &m_sem, CLONE_FS|CLONE_FILES);
496 ctl_set_bit(14, 28); /* enable channel report MCH */
497 return 0;
498}
499
500device_initcall (machine_check_crw_init);