Merge branch 'for-linus' of master.kernel.org:/pub/scm/linux/kernel/git/roland/infiniband
[linux-2.6] / drivers / s390 / s390mach.c
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 "cio/chp.h"
25 #include "s390mach.h"
26
27 static struct semaphore m_sem;
28
29 static NORET_TYPE void
30 s390_handle_damage(char *msg)
31 {
32 #ifdef CONFIG_SMP
33         smp_send_stop();
34 #endif
35         disabled_wait((unsigned long) __builtin_return_address(0));
36         for(;;);
37 }
38
39 /*
40  * Retrieve CRWs and call function to handle event.
41  *
42  * Note : we currently process CRWs for io and chsc subchannels only
43  */
44 static int
45 s390_collect_crw_info(void *param)
46 {
47         struct crw crw[2];
48         int ccode;
49         struct semaphore *sem;
50         unsigned int chain;
51
52         sem = (struct semaphore *)param;
53 repeat:
54         down_interruptible(sem);
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_schedule_eval_all();
88                         chain = 0;
89                         continue;
90                 }
91                 switch (crw[chain].rsc) {
92                 case CRW_RSC_SCH:
93                         if (crw[0].chn && !chain)
94                                 break;
95                         pr_debug("source is subchannel %04X\n", crw[0].rsid);
96                         css_process_crw(crw[0].rsid, chain ? crw[1].rsid : 0);
97                         break;
98                 case CRW_RSC_MONITOR:
99                         pr_debug("source is monitoring facility\n");
100                         break;
101                 case CRW_RSC_CPATH:
102                         pr_debug("source is channel path %02X\n", crw[0].rsid);
103                         /*
104                          * Check for solicited machine checks. These are
105                          * created by reset channel path and need not be
106                          * reported to the common I/O layer.
107                          */
108                         if (crw[chain].slct) {
109                                 pr_debug("solicited machine check for "
110                                          "channel path %02X\n", crw[0].rsid);
111                                 break;
112                         }
113                         switch (crw[0].erc) {
114                         case CRW_ERC_IPARM: /* Path has come. */
115                                 chp_process_crw(crw[0].rsid, 1);
116                                 break;
117                         case CRW_ERC_PERRI: /* Path has gone. */
118                         case CRW_ERC_PERRN:
119                                 chp_process_crw(crw[0].rsid, 0);
120                                 break;
121                         default:
122                                 pr_debug("Don't know how to handle erc=%x\n",
123                                          crw[0].erc);
124                         }
125                         break;
126                 case CRW_RSC_CONFIG:
127                         pr_debug("source is configuration-alert facility\n");
128                         break;
129                 case CRW_RSC_CSS:
130                         pr_debug("source is channel subsystem\n");
131                         chsc_process_crw();
132                         break;
133                 default:
134                         pr_debug("unknown source\n");
135                         break;
136                 }
137                 /* chain is always 0 or 1 here. */
138                 chain = crw[chain].chn ? chain + 1 : 0;
139         }
140         goto repeat;
141         return 0;
142 }
143
144 struct mcck_struct {
145         int kill_task;
146         int channel_report;
147         int warning;
148         unsigned long long mcck_code;
149 };
150
151 static DEFINE_PER_CPU(struct mcck_struct, cpu_mcck);
152
153 /*
154  * Main machine check handler function. Will be called with interrupts enabled
155  * or disabled and machine checks enabled or disabled.
156  */
157 void
158 s390_handle_mcck(void)
159 {
160         unsigned long flags;
161         struct mcck_struct mcck;
162
163         /*
164          * Disable machine checks and get the current state of accumulated
165          * machine checks. Afterwards delete the old state and enable machine
166          * checks again.
167          */
168         local_irq_save(flags);
169         local_mcck_disable();
170         mcck = __get_cpu_var(cpu_mcck);
171         memset(&__get_cpu_var(cpu_mcck), 0, sizeof(struct mcck_struct));
172         clear_thread_flag(TIF_MCCK_PENDING);
173         local_mcck_enable();
174         local_irq_restore(flags);
175
176         if (mcck.channel_report)
177                 up(&m_sem);
178
179 #ifdef CONFIG_MACHCHK_WARNING
180 /*
181  * The warning may remain for a prolonged period on the bare iron.
182  * (actually till the machine is powered off, or until the problem is gone)
183  * So we just stop listening for the WARNING MCH and prevent continuously
184  * being interrupted.  One caveat is however, that we must do this per
185  * processor and cannot use the smp version of ctl_clear_bit().
186  * On VM we only get one interrupt per virtally presented machinecheck.
187  * Though one suffices, we may get one interrupt per (virtual) processor.
188  */
189         if (mcck.warning) {     /* WARNING pending ? */
190                 static int mchchk_wng_posted = 0;
191                 /*
192                  * Use single machine clear, as we cannot handle smp right now
193                  */
194                 __ctl_clear_bit(14, 24);        /* Disable WARNING MCH */
195                 if (xchg(&mchchk_wng_posted, 1) == 0)
196                         kill_cad_pid(SIGPWR, 1);
197         }
198 #endif
199
200         if (mcck.kill_task) {
201                 local_irq_enable();
202                 printk(KERN_EMERG "mcck: Terminating task because of machine "
203                        "malfunction (code 0x%016llx).\n", mcck.mcck_code);
204                 printk(KERN_EMERG "mcck: task: %s, pid: %d.\n",
205                        current->comm, current->pid);
206                 do_exit(SIGSEGV);
207         }
208 }
209
210 /*
211  * returns 0 if all registers could be validated
212  * returns 1 otherwise
213  */
214 static int
215 s390_revalidate_registers(struct mci *mci)
216 {
217         int kill_task;
218         u64 tmpclock;
219         u64 zero;
220         void *fpt_save_area, *fpt_creg_save_area;
221
222         kill_task = 0;
223         zero = 0;
224         /* General purpose registers */
225         if (!mci->gr)
226                 /*
227                  * General purpose registers couldn't be restored and have
228                  * unknown contents. Process needs to be terminated.
229                  */
230                 kill_task = 1;
231
232         /* Revalidate floating point registers */
233         if (!mci->fp)
234                 /*
235                  * Floating point registers can't be restored and
236                  * therefore the process needs to be terminated.
237                  */
238                 kill_task = 1;
239
240 #ifndef CONFIG_64BIT
241         asm volatile(
242                 "       ld      0,0(%0)\n"
243                 "       ld      2,8(%0)\n"
244                 "       ld      4,16(%0)\n"
245                 "       ld      6,24(%0)"
246                 : : "a" (&S390_lowcore.floating_pt_save_area));
247 #endif
248
249         if (MACHINE_HAS_IEEE) {
250 #ifdef CONFIG_64BIT
251                 fpt_save_area = &S390_lowcore.floating_pt_save_area;
252                 fpt_creg_save_area = &S390_lowcore.fpt_creg_save_area;
253 #else
254                 fpt_save_area = (void *) S390_lowcore.extended_save_area_addr;
255                 fpt_creg_save_area = fpt_save_area+128;
256 #endif
257                 /* Floating point control register */
258                 if (!mci->fc) {
259                         /*
260                          * Floating point control register can't be restored.
261                          * Task will be terminated.
262                          */
263                         asm volatile("lfpc 0(%0)" : : "a" (&zero), "m" (zero));
264                         kill_task = 1;
265
266                 } else
267                         asm volatile("lfpc 0(%0)" : : "a" (fpt_creg_save_area));
268
269                 asm volatile(
270                         "       ld      0,0(%0)\n"
271                         "       ld      1,8(%0)\n"
272                         "       ld      2,16(%0)\n"
273                         "       ld      3,24(%0)\n"
274                         "       ld      4,32(%0)\n"
275                         "       ld      5,40(%0)\n"
276                         "       ld      6,48(%0)\n"
277                         "       ld      7,56(%0)\n"
278                         "       ld      8,64(%0)\n"
279                         "       ld      9,72(%0)\n"
280                         "       ld      10,80(%0)\n"
281                         "       ld      11,88(%0)\n"
282                         "       ld      12,96(%0)\n"
283                         "       ld      13,104(%0)\n"
284                         "       ld      14,112(%0)\n"
285                         "       ld      15,120(%0)\n"
286                         : : "a" (fpt_save_area));
287         }
288
289         /* Revalidate access registers */
290         asm volatile(
291                 "       lam     0,15,0(%0)"
292                 : : "a" (&S390_lowcore.access_regs_save_area));
293         if (!mci->ar)
294                 /*
295                  * Access registers have unknown contents.
296                  * Terminating task.
297                  */
298                 kill_task = 1;
299
300         /* Revalidate control registers */
301         if (!mci->cr)
302                 /*
303                  * Control registers have unknown contents.
304                  * Can't recover and therefore stopping machine.
305                  */
306                 s390_handle_damage("invalid control registers.");
307         else
308 #ifdef CONFIG_64BIT
309                 asm volatile(
310                         "       lctlg   0,15,0(%0)"
311                         : : "a" (&S390_lowcore.cregs_save_area));
312 #else
313                 asm volatile(
314                         "       lctl    0,15,0(%0)"
315                         : : "a" (&S390_lowcore.cregs_save_area));
316 #endif
317
318         /*
319          * We don't even try to revalidate the TOD register, since we simply
320          * can't write something sensible into that register.
321          */
322
323 #ifdef CONFIG_64BIT
324         /*
325          * See if we can revalidate the TOD programmable register with its
326          * old contents (should be zero) otherwise set it to zero.
327          */
328         if (!mci->pr)
329                 asm volatile(
330                         "       sr      0,0\n"
331                         "       sckpf"
332                         : : : "0", "cc");
333         else
334                 asm volatile(
335                         "       l       0,0(%0)\n"
336                         "       sckpf"
337                         : : "a" (&S390_lowcore.tod_progreg_save_area)
338                         : "0", "cc");
339 #endif
340
341         /* Revalidate clock comparator register */
342         asm volatile(
343                 "       stck    0(%1)\n"
344                 "       sckc    0(%1)"
345                 : "=m" (tmpclock) : "a" (&(tmpclock)) : "cc", "memory");
346
347         /* Check if old PSW is valid */
348         if (!mci->wp)
349                 /*
350                  * Can't tell if we come from user or kernel mode
351                  * -> stopping machine.
352                  */
353                 s390_handle_damage("old psw invalid.");
354
355         if (!mci->ms || !mci->pm || !mci->ia)
356                 kill_task = 1;
357
358         return kill_task;
359 }
360
361 #define MAX_IPD_COUNT   29
362 #define MAX_IPD_TIME    (5 * 60 * USEC_PER_SEC) /* 5 minutes */
363
364 /*
365  * machine check handler.
366  */
367 void
368 s390_do_machine_check(struct pt_regs *regs)
369 {
370         static DEFINE_SPINLOCK(ipd_lock);
371         static unsigned long long last_ipd;
372         static int ipd_count;
373         unsigned long long tmp;
374         struct mci *mci;
375         struct mcck_struct *mcck;
376         int umode;
377
378         lockdep_off();
379
380         mci = (struct mci *) &S390_lowcore.mcck_interruption_code;
381         mcck = &__get_cpu_var(cpu_mcck);
382         umode = user_mode(regs);
383
384         if (mci->sd)
385                 /* System damage -> stopping machine */
386                 s390_handle_damage("received system damage machine check.");
387
388         if (mci->pd) {
389                 if (mci->b) {
390                         /* Processing backup -> verify if we can survive this */
391                         u64 z_mcic, o_mcic, t_mcic;
392 #ifdef CONFIG_64BIT
393                         z_mcic = (1ULL<<63 | 1ULL<<59 | 1ULL<<29);
394                         o_mcic = (1ULL<<43 | 1ULL<<42 | 1ULL<<41 | 1ULL<<40 |
395                                   1ULL<<36 | 1ULL<<35 | 1ULL<<34 | 1ULL<<32 |
396                                   1ULL<<30 | 1ULL<<21 | 1ULL<<20 | 1ULL<<17 |
397                                   1ULL<<16);
398 #else
399                         z_mcic = (1ULL<<63 | 1ULL<<59 | 1ULL<<57 | 1ULL<<50 |
400                                   1ULL<<29);
401                         o_mcic = (1ULL<<43 | 1ULL<<42 | 1ULL<<41 | 1ULL<<40 |
402                                   1ULL<<36 | 1ULL<<35 | 1ULL<<34 | 1ULL<<32 |
403                                   1ULL<<30 | 1ULL<<20 | 1ULL<<17 | 1ULL<<16);
404 #endif
405                         t_mcic = *(u64 *)mci;
406
407                         if (((t_mcic & z_mcic) != 0) ||
408                             ((t_mcic & o_mcic) != o_mcic)) {
409                                 s390_handle_damage("processing backup machine "
410                                                    "check with damage.");
411                         }
412
413                         /*
414                          * Nullifying exigent condition, therefore we might
415                          * retry this instruction.
416                          */
417
418                         spin_lock(&ipd_lock);
419
420                         tmp = get_clock();
421
422                         if (((tmp - last_ipd) >> 12) < MAX_IPD_TIME)
423                                 ipd_count++;
424                         else
425                                 ipd_count = 1;
426
427                         last_ipd = tmp;
428
429                         if (ipd_count == MAX_IPD_COUNT)
430                                 s390_handle_damage("too many ipd retries.");
431
432                         spin_unlock(&ipd_lock);
433                 }
434                 else {
435                         /* Processing damage -> stopping machine */
436                         s390_handle_damage("received instruction processing "
437                                            "damage machine check.");
438                 }
439         }
440         if (s390_revalidate_registers(mci)) {
441                 if (umode) {
442                         /*
443                          * Couldn't restore all register contents while in
444                          * user mode -> mark task for termination.
445                          */
446                         mcck->kill_task = 1;
447                         mcck->mcck_code = *(unsigned long long *) mci;
448                         set_thread_flag(TIF_MCCK_PENDING);
449                 }
450                 else
451                         /*
452                          * Couldn't restore all register contents while in
453                          * kernel mode -> stopping machine.
454                          */
455                         s390_handle_damage("unable to revalidate registers.");
456         }
457
458         if (mci->cd) {
459                 /* Timing facility damage */
460                 s390_handle_damage("TOD clock damaged");
461         }
462
463         if (mci->ed && mci->ec) {
464                 /* External damage */
465                 if (S390_lowcore.external_damage_code & (1U << ED_ETR_SYNC))
466                         etr_sync_check();
467                 if (S390_lowcore.external_damage_code & (1U << ED_ETR_SWITCH))
468                         etr_switch_to_local();
469         }
470
471         if (mci->se)
472                 /* Storage error uncorrected */
473                 s390_handle_damage("received storage error uncorrected "
474                                    "machine check.");
475
476         if (mci->ke)
477                 /* Storage key-error uncorrected */
478                 s390_handle_damage("received storage key-error uncorrected "
479                                    "machine check.");
480
481         if (mci->ds && mci->fa)
482                 /* Storage degradation */
483                 s390_handle_damage("received storage degradation machine "
484                                    "check.");
485
486         if (mci->cp) {
487                 /* Channel report word pending */
488                 mcck->channel_report = 1;
489                 set_thread_flag(TIF_MCCK_PENDING);
490         }
491
492         if (mci->w) {
493                 /* Warning pending */
494                 mcck->warning = 1;
495                 set_thread_flag(TIF_MCCK_PENDING);
496         }
497         lockdep_on();
498 }
499
500 /*
501  * s390_init_machine_check
502  *
503  * initialize machine check handling
504  */
505 static int
506 machine_check_init(void)
507 {
508         init_MUTEX_LOCKED(&m_sem);
509         ctl_set_bit(14, 25);    /* enable external damage MCH */
510         ctl_set_bit(14, 27);    /* enable system recovery MCH */
511 #ifdef CONFIG_MACHCHK_WARNING
512         ctl_set_bit(14, 24);    /* enable warning MCH */
513 #endif
514         return 0;
515 }
516
517 /*
518  * Initialize the machine check handler really early to be able to
519  * catch all machine checks that happen during boot
520  */
521 arch_initcall(machine_check_init);
522
523 /*
524  * Machine checks for the channel subsystem must be enabled
525  * after the channel subsystem is initialized
526  */
527 static int __init
528 machine_check_crw_init (void)
529 {
530         struct task_struct *task;
531
532         task = kthread_run(s390_collect_crw_info, &m_sem, "kmcheck");
533         if (IS_ERR(task))
534                 return PTR_ERR(task);
535         ctl_set_bit(14, 28);    /* enable channel report MCH */
536         return 0;
537 }
538
539 device_initcall (machine_check_crw_init);