Merge rsync://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux-2.6
[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/kthread.h>
17
18 #include <asm/lowcore.h>
19
20 #include "s390mach.h"
21
22 #define DBG printk
23 // #define DBG(args,...) do {} while (0);
24
25 static struct semaphore m_sem;
26
27 extern int css_process_crw(int, int);
28 extern int chsc_process_crw(void);
29 extern int chp_process_crw(int, int);
30 extern void css_reiterate_subchannels(void);
31
32 extern struct workqueue_struct *slow_path_wq;
33 extern struct work_struct slow_path_work;
34
35 static NORET_TYPE void
36 s390_handle_damage(char *msg)
37 {
38 #ifdef CONFIG_SMP
39         smp_send_stop();
40 #endif
41         disabled_wait((unsigned long) __builtin_return_address(0));
42         for(;;);
43 }
44
45 /*
46  * Retrieve CRWs and call function to handle event.
47  *
48  * Note : we currently process CRWs for io and chsc subchannels only
49  */
50 static int
51 s390_collect_crw_info(void *param)
52 {
53         struct crw crw[2];
54         int ccode, ret, slow;
55         struct semaphore *sem;
56         unsigned int chain;
57
58         sem = (struct semaphore *)param;
59 repeat:
60         down_interruptible(sem);
61         slow = 0;
62         chain = 0;
63         while (1) {
64                 if (unlikely(chain > 1)) {
65                         struct crw tmp_crw;
66
67                         printk(KERN_WARNING"%s: Code does not support more "
68                                "than two chained crws; please report to "
69                                "linux390@de.ibm.com!\n", __FUNCTION__);
70                         ccode = stcrw(&tmp_crw);
71                         printk(KERN_WARNING"%s: crw reports slct=%d, oflw=%d, "
72                                "chn=%d, rsc=%X, anc=%d, erc=%X, rsid=%X\n",
73                                __FUNCTION__, tmp_crw.slct, tmp_crw.oflw,
74                                tmp_crw.chn, tmp_crw.rsc, tmp_crw.anc,
75                                tmp_crw.erc, tmp_crw.rsid);
76                         printk(KERN_WARNING"%s: This was crw number %x in the "
77                                "chain\n", __FUNCTION__, chain);
78                         if (ccode != 0)
79                                 break;
80                         chain = tmp_crw.chn ? chain + 1 : 0;
81                         continue;
82                 }
83                 ccode = stcrw(&crw[chain]);
84                 if (ccode != 0)
85                         break;
86                 DBG(KERN_DEBUG "crw_info : CRW reports slct=%d, oflw=%d, "
87                     "chn=%d, rsc=%X, anc=%d, erc=%X, rsid=%X\n",
88                     crw[chain].slct, crw[chain].oflw, crw[chain].chn,
89                     crw[chain].rsc, crw[chain].anc, crw[chain].erc,
90                     crw[chain].rsid);
91                 /* Check for overflows. */
92                 if (crw[chain].oflw) {
93                         pr_debug("%s: crw overflow detected!\n", __FUNCTION__);
94                         css_reiterate_subchannels();
95                         chain = 0;
96                         slow = 1;
97                         continue;
98                 }
99                 switch (crw[chain].rsc) {
100                 case CRW_RSC_SCH:
101                         if (crw[0].chn && !chain)
102                                 break;
103                         pr_debug("source is subchannel %04X\n", crw[0].rsid);
104                         ret = css_process_crw (crw[0].rsid,
105                                                chain ? crw[1].rsid : 0);
106                         if (ret == -EAGAIN)
107                                 slow = 1;
108                         break;
109                 case CRW_RSC_MONITOR:
110                         pr_debug("source is monitoring facility\n");
111                         break;
112                 case CRW_RSC_CPATH:
113                         pr_debug("source is channel path %02X\n", crw[0].rsid);
114                         switch (crw[0].erc) {
115                         case CRW_ERC_IPARM: /* Path has come. */
116                                 ret = chp_process_crw(crw[0].rsid, 1);
117                                 break;
118                         case CRW_ERC_PERRI: /* Path has gone. */
119                         case CRW_ERC_PERRN:
120                                 ret = chp_process_crw(crw[0].rsid, 0);
121                                 break;
122                         default:
123                                 pr_debug("Don't know how to handle erc=%x\n",
124                                          crw[0].erc);
125                                 ret = 0;
126                         }
127                         if (ret == -EAGAIN)
128                                 slow = 1;
129                         break;
130                 case CRW_RSC_CONFIG:
131                         pr_debug("source is configuration-alert facility\n");
132                         break;
133                 case CRW_RSC_CSS:
134                         pr_debug("source is channel subsystem\n");
135                         ret = chsc_process_crw();
136                         if (ret == -EAGAIN)
137                                 slow = 1;
138                         break;
139                 default:
140                         pr_debug("unknown source\n");
141                         break;
142                 }
143                 /* chain is always 0 or 1 here. */
144                 chain = crw[chain].chn ? chain + 1 : 0;
145         }
146         if (slow)
147                 queue_work(slow_path_wq, &slow_path_work);
148         goto repeat;
149         return 0;
150 }
151
152 struct mcck_struct {
153         int kill_task;
154         int channel_report;
155         int warning;
156         unsigned long long mcck_code;
157 };
158
159 static DEFINE_PER_CPU(struct mcck_struct, cpu_mcck);
160
161 /*
162  * Main machine check handler function. Will be called with interrupts enabled
163  * or disabled and machine checks enabled or disabled.
164  */
165 void
166 s390_handle_mcck(void)
167 {
168         unsigned long flags;
169         struct mcck_struct mcck;
170
171         /*
172          * Disable machine checks and get the current state of accumulated
173          * machine checks. Afterwards delete the old state and enable machine
174          * checks again.
175          */
176         local_irq_save(flags);
177         local_mcck_disable();
178         mcck = __get_cpu_var(cpu_mcck);
179         memset(&__get_cpu_var(cpu_mcck), 0, sizeof(struct mcck_struct));
180         clear_thread_flag(TIF_MCCK_PENDING);
181         local_mcck_enable();
182         local_irq_restore(flags);
183
184         if (mcck.channel_report)
185                 up(&m_sem);
186
187 #ifdef CONFIG_MACHCHK_WARNING
188 /*
189  * The warning may remain for a prolonged period on the bare iron.
190  * (actually till the machine is powered off, or until the problem is gone)
191  * So we just stop listening for the WARNING MCH and prevent continuously
192  * being interrupted.  One caveat is however, that we must do this per
193  * processor and cannot use the smp version of ctl_clear_bit().
194  * On VM we only get one interrupt per virtally presented machinecheck.
195  * Though one suffices, we may get one interrupt per (virtual) processor.
196  */
197         if (mcck.warning) {     /* WARNING pending ? */
198                 static int mchchk_wng_posted = 0;
199                 /*
200                  * Use single machine clear, as we cannot handle smp right now
201                  */
202                 __ctl_clear_bit(14, 24);        /* Disable WARNING MCH */
203                 if (xchg(&mchchk_wng_posted, 1) == 0)
204                         kill_proc(1, SIGPWR, 1);
205         }
206 #endif
207
208         if (mcck.kill_task) {
209                 local_irq_enable();
210                 printk(KERN_EMERG "mcck: Terminating task because of machine "
211                        "malfunction (code 0x%016llx).\n", mcck.mcck_code);
212                 printk(KERN_EMERG "mcck: task: %s, pid: %d.\n",
213                        current->comm, current->pid);
214                 do_exit(SIGSEGV);
215         }
216 }
217
218 /*
219  * returns 0 if all registers could be validated
220  * returns 1 otherwise
221  */
222 static int
223 s390_revalidate_registers(struct mci *mci)
224 {
225         int kill_task;
226         u64 tmpclock;
227         u64 zero;
228         void *fpt_save_area, *fpt_creg_save_area;
229
230         kill_task = 0;
231         zero = 0;
232         /* General purpose registers */
233         if (!mci->gr)
234                 /*
235                  * General purpose registers couldn't be restored and have
236                  * unknown contents. Process needs to be terminated.
237                  */
238                 kill_task = 1;
239
240         /* Revalidate floating point registers */
241         if (!mci->fp)
242                 /*
243                  * Floating point registers can't be restored and
244                  * therefore the process needs to be terminated.
245                  */
246                 kill_task = 1;
247
248 #ifndef CONFIG_64BIT
249         asm volatile("ld 0,0(%0)\n"
250                      "ld 2,8(%0)\n"
251                      "ld 4,16(%0)\n"
252                      "ld 6,24(%0)"
253                      : : "a" (&S390_lowcore.floating_pt_save_area));
254 #endif
255
256         if (MACHINE_HAS_IEEE) {
257 #ifdef CONFIG_64BIT
258                 fpt_save_area = &S390_lowcore.floating_pt_save_area;
259                 fpt_creg_save_area = &S390_lowcore.fpt_creg_save_area;
260 #else
261                 fpt_save_area = (void *) S390_lowcore.extended_save_area_addr;
262                 fpt_creg_save_area = fpt_save_area+128;
263 #endif
264                 /* Floating point control register */
265                 if (!mci->fc) {
266                         /*
267                          * Floating point control register can't be restored.
268                          * Task will be terminated.
269                          */
270                         asm volatile ("lfpc 0(%0)" : : "a" (&zero), "m" (zero));
271                         kill_task = 1;
272
273                 }
274                 else
275                         asm volatile (
276                                 "lfpc 0(%0)"
277                                 : : "a" (fpt_creg_save_area));
278
279                 asm volatile("ld  0,0(%0)\n"
280                              "ld  1,8(%0)\n"
281                              "ld  2,16(%0)\n"
282                              "ld  3,24(%0)\n"
283                              "ld  4,32(%0)\n"
284                              "ld  5,40(%0)\n"
285                              "ld  6,48(%0)\n"
286                              "ld  7,56(%0)\n"
287                              "ld  8,64(%0)\n"
288                              "ld  9,72(%0)\n"
289                              "ld 10,80(%0)\n"
290                              "ld 11,88(%0)\n"
291                              "ld 12,96(%0)\n"
292                              "ld 13,104(%0)\n"
293                              "ld 14,112(%0)\n"
294                              "ld 15,120(%0)\n"
295                              : : "a" (fpt_save_area));
296         }
297
298         /* Revalidate access registers */
299         asm volatile("lam 0,15,0(%0)"
300                      : : "a" (&S390_lowcore.access_regs_save_area));
301         if (!mci->ar)
302                 /*
303                  * Access registers have unknown contents.
304                  * Terminating task.
305                  */
306                 kill_task = 1;
307
308         /* Revalidate control registers */
309         if (!mci->cr)
310                 /*
311                  * Control registers have unknown contents.
312                  * Can't recover and therefore stopping machine.
313                  */
314                 s390_handle_damage("invalid control registers.");
315         else
316 #ifdef CONFIG_64BIT
317                 asm volatile("lctlg 0,15,0(%0)"
318                              : : "a" (&S390_lowcore.cregs_save_area));
319 #else
320                 asm volatile("lctl 0,15,0(%0)"
321                              : : "a" (&S390_lowcore.cregs_save_area));
322 #endif
323
324         /*
325          * We don't even try to revalidate the TOD register, since we simply
326          * can't write something sensible into that register.
327          */
328
329 #ifdef CONFIG_64BIT
330         /*
331          * See if we can revalidate the TOD programmable register with its
332          * old contents (should be zero) otherwise set it to zero.
333          */
334         if (!mci->pr)
335                 asm volatile("sr 0,0\n"
336                              "sckpf"
337                              : : : "0", "cc");
338         else
339                 asm volatile(
340                         "l 0,0(%0)\n"
341                         "sckpf"
342                         : : "a" (&S390_lowcore.tod_progreg_save_area) : "0", "cc");
343 #endif
344
345         /* Revalidate clock comparator register */
346         asm volatile ("stck 0(%1)\n"
347                       "sckc 0(%1)"
348                       : "=m" (tmpclock) : "a" (&(tmpclock)) : "cc", "memory");
349
350         /* Check if old PSW is valid */
351         if (!mci->wp)
352                 /*
353                  * Can't tell if we come from user or kernel mode
354                  * -> stopping machine.
355                  */
356                 s390_handle_damage("old psw invalid.");
357
358         if (!mci->ms || !mci->pm || !mci->ia)
359                 kill_task = 1;
360
361         return kill_task;
362 }
363
364 #define MAX_IPD_COUNT   29
365 #define MAX_IPD_TIME    (5 * 60 * USEC_PER_SEC) /* 5 minutes */
366
367 /*
368  * machine check handler.
369  */
370 void
371 s390_do_machine_check(struct pt_regs *regs)
372 {
373         static DEFINE_SPINLOCK(ipd_lock);
374         static unsigned long long last_ipd;
375         static int ipd_count;
376         unsigned long long tmp;
377         struct mci *mci;
378         struct mcck_struct *mcck;
379         int umode;
380
381         mci = (struct mci *) &S390_lowcore.mcck_interruption_code;
382         mcck = &__get_cpu_var(cpu_mcck);
383         umode = user_mode(regs);
384
385         if (mci->sd)
386                 /* System damage -> stopping machine */
387                 s390_handle_damage("received system damage machine check.");
388
389         if (mci->pd) {
390                 if (mci->b) {
391                         /* Processing backup -> verify if we can survive this */
392                         u64 z_mcic, o_mcic, t_mcic;
393 #ifdef CONFIG_64BIT
394                         z_mcic = (1ULL<<63 | 1ULL<<59 | 1ULL<<29);
395                         o_mcic = (1ULL<<43 | 1ULL<<42 | 1ULL<<41 | 1ULL<<40 |
396                                   1ULL<<36 | 1ULL<<35 | 1ULL<<34 | 1ULL<<32 |
397                                   1ULL<<30 | 1ULL<<21 | 1ULL<<20 | 1ULL<<17 |
398                                   1ULL<<16);
399 #else
400                         z_mcic = (1ULL<<63 | 1ULL<<59 | 1ULL<<57 | 1ULL<<50 |
401                                   1ULL<<29);
402                         o_mcic = (1ULL<<43 | 1ULL<<42 | 1ULL<<41 | 1ULL<<40 |
403                                   1ULL<<36 | 1ULL<<35 | 1ULL<<34 | 1ULL<<32 |
404                                   1ULL<<30 | 1ULL<<20 | 1ULL<<17 | 1ULL<<16);
405 #endif
406                         t_mcic = *(u64 *)mci;
407
408                         if (((t_mcic & z_mcic) != 0) ||
409                             ((t_mcic & o_mcic) != o_mcic)) {
410                                 s390_handle_damage("processing backup machine "
411                                                    "check with damage.");
412                         }
413
414                         /*
415                          * Nullifying exigent condition, therefore we might
416                          * retry this instruction.
417                          */
418
419                         spin_lock(&ipd_lock);
420
421                         tmp = get_clock();
422
423                         if (((tmp - last_ipd) >> 12) < MAX_IPD_TIME)
424                                 ipd_count++;
425                         else
426                                 ipd_count = 1;
427
428                         last_ipd = tmp;
429
430                         if (ipd_count == MAX_IPD_COUNT)
431                                 s390_handle_damage("too many ipd retries.");
432
433                         spin_unlock(&ipd_lock);
434                 }
435                 else {
436                         /* Processing damage -> stopping machine */
437                         s390_handle_damage("received instruction processing "
438                                            "damage machine check.");
439                 }
440         }
441         if (s390_revalidate_registers(mci)) {
442                 if (umode) {
443                         /*
444                          * Couldn't restore all register contents while in
445                          * user mode -> mark task for termination.
446                          */
447                         mcck->kill_task = 1;
448                         mcck->mcck_code = *(unsigned long long *) mci;
449                         set_thread_flag(TIF_MCCK_PENDING);
450                 }
451                 else
452                         /*
453                          * Couldn't restore all register contents while in
454                          * kernel mode -> stopping machine.
455                          */
456                         s390_handle_damage("unable to revalidate registers.");
457         }
458
459         if (mci->se)
460                 /* Storage error uncorrected */
461                 s390_handle_damage("received storage error uncorrected "
462                                    "machine check.");
463
464         if (mci->ke)
465                 /* Storage key-error uncorrected */
466                 s390_handle_damage("received storage key-error uncorrected "
467                                    "machine check.");
468
469         if (mci->ds && mci->fa)
470                 /* Storage degradation */
471                 s390_handle_damage("received storage degradation machine "
472                                    "check.");
473
474         if (mci->cp) {
475                 /* Channel report word pending */
476                 mcck->channel_report = 1;
477                 set_thread_flag(TIF_MCCK_PENDING);
478         }
479
480         if (mci->w) {
481                 /* Warning pending */
482                 mcck->warning = 1;
483                 set_thread_flag(TIF_MCCK_PENDING);
484         }
485 }
486
487 /*
488  * s390_init_machine_check
489  *
490  * initialize machine check handling
491  */
492 static int
493 machine_check_init(void)
494 {
495         init_MUTEX_LOCKED(&m_sem);
496         ctl_clear_bit(14, 25);  /* disable external damage MCH */
497         ctl_set_bit(14, 27);    /* enable system recovery MCH */
498 #ifdef CONFIG_MACHCHK_WARNING
499         ctl_set_bit(14, 24);    /* enable warning MCH */
500 #endif
501         return 0;
502 }
503
504 /*
505  * Initialize the machine check handler really early to be able to
506  * catch all machine checks that happen during boot
507  */
508 arch_initcall(machine_check_init);
509
510 /*
511  * Machine checks for the channel subsystem must be enabled
512  * after the channel subsystem is initialized
513  */
514 static int __init
515 machine_check_crw_init (void)
516 {
517         kthread_run(s390_collect_crw_info, &m_sem, "kmcheck");
518         ctl_set_bit(14, 28);    /* enable channel report MCH */
519         return 0;
520 }
521
522 device_initcall (machine_check_crw_init);