Merge git://git.infradead.org/~dwmw2/rbtree-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/config.h>
12 #include <linux/init.h>
13 #include <linux/sched.h>
14 #include <linux/errno.h>
15 #include <linux/workqueue.h>
16 #include <linux/time.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         /* Set a nice name. */
60         daemonize("kmcheck");
61 repeat:
62         down_interruptible(sem);
63         slow = 0;
64         chain = 0;
65         while (1) {
66                 if (unlikely(chain > 1)) {
67                         struct crw tmp_crw;
68
69                         printk(KERN_WARNING"%s: Code does not support more "
70                                "than two chained crws; please report to "
71                                "linux390@de.ibm.com!\n", __FUNCTION__);
72                         ccode = stcrw(&tmp_crw);
73                         printk(KERN_WARNING"%s: crw reports slct=%d, oflw=%d, "
74                                "chn=%d, rsc=%X, anc=%d, erc=%X, rsid=%X\n",
75                                __FUNCTION__, tmp_crw.slct, tmp_crw.oflw,
76                                tmp_crw.chn, tmp_crw.rsc, tmp_crw.anc,
77                                tmp_crw.erc, tmp_crw.rsid);
78                         printk(KERN_WARNING"%s: This was crw number %x in the "
79                                "chain\n", __FUNCTION__, chain);
80                         if (ccode != 0)
81                                 break;
82                         chain = tmp_crw.chn ? chain + 1 : 0;
83                         continue;
84                 }
85                 ccode = stcrw(&crw[chain]);
86                 if (ccode != 0)
87                         break;
88                 DBG(KERN_DEBUG "crw_info : CRW reports slct=%d, oflw=%d, "
89                     "chn=%d, rsc=%X, anc=%d, erc=%X, rsid=%X\n",
90                     crw[chain].slct, crw[chain].oflw, crw[chain].chn,
91                     crw[chain].rsc, crw[chain].anc, crw[chain].erc,
92                     crw[chain].rsid);
93                 /* Check for overflows. */
94                 if (crw[chain].oflw) {
95                         pr_debug("%s: crw overflow detected!\n", __FUNCTION__);
96                         css_reiterate_subchannels();
97                         chain = 0;
98                         slow = 1;
99                         continue;
100                 }
101                 switch (crw[chain].rsc) {
102                 case CRW_RSC_SCH:
103                         if (crw[0].chn && !chain)
104                                 break;
105                         pr_debug("source is subchannel %04X\n", crw[0].rsid);
106                         ret = css_process_crw (crw[0].rsid,
107                                                chain ? crw[1].rsid : 0);
108                         if (ret == -EAGAIN)
109                                 slow = 1;
110                         break;
111                 case CRW_RSC_MONITOR:
112                         pr_debug("source is monitoring facility\n");
113                         break;
114                 case CRW_RSC_CPATH:
115                         pr_debug("source is channel path %02X\n", crw[0].rsid);
116                         switch (crw[0].erc) {
117                         case CRW_ERC_IPARM: /* Path has come. */
118                                 ret = chp_process_crw(crw[0].rsid, 1);
119                                 break;
120                         case CRW_ERC_PERRI: /* Path has gone. */
121                         case CRW_ERC_PERRN:
122                                 ret = chp_process_crw(crw[0].rsid, 0);
123                                 break;
124                         default:
125                                 pr_debug("Don't know how to handle erc=%x\n",
126                                          crw[0].erc);
127                                 ret = 0;
128                         }
129                         if (ret == -EAGAIN)
130                                 slow = 1;
131                         break;
132                 case CRW_RSC_CONFIG:
133                         pr_debug("source is configuration-alert facility\n");
134                         break;
135                 case CRW_RSC_CSS:
136                         pr_debug("source is channel subsystem\n");
137                         ret = chsc_process_crw();
138                         if (ret == -EAGAIN)
139                                 slow = 1;
140                         break;
141                 default:
142                         pr_debug("unknown source\n");
143                         break;
144                 }
145                 /* chain is always 0 or 1 here. */
146                 chain = crw[chain].chn ? chain + 1 : 0;
147         }
148         if (slow)
149                 queue_work(slow_path_wq, &slow_path_work);
150         goto repeat;
151         return 0;
152 }
153
154 struct mcck_struct {
155         int kill_task;
156         int channel_report;
157         int warning;
158         unsigned long long mcck_code;
159 };
160
161 static DEFINE_PER_CPU(struct mcck_struct, cpu_mcck);
162
163 /*
164  * Main machine check handler function. Will be called with interrupts enabled
165  * or disabled and machine checks enabled or disabled.
166  */
167 void
168 s390_handle_mcck(void)
169 {
170         unsigned long flags;
171         struct mcck_struct mcck;
172
173         /*
174          * Disable machine checks and get the current state of accumulated
175          * machine checks. Afterwards delete the old state and enable machine
176          * checks again.
177          */
178         local_irq_save(flags);
179         local_mcck_disable();
180         mcck = __get_cpu_var(cpu_mcck);
181         memset(&__get_cpu_var(cpu_mcck), 0, sizeof(struct mcck_struct));
182         clear_thread_flag(TIF_MCCK_PENDING);
183         local_mcck_enable();
184         local_irq_restore(flags);
185
186         if (mcck.channel_report)
187                 up(&m_sem);
188
189 #ifdef CONFIG_MACHCHK_WARNING
190 /*
191  * The warning may remain for a prolonged period on the bare iron.
192  * (actually till the machine is powered off, or until the problem is gone)
193  * So we just stop listening for the WARNING MCH and prevent continuously
194  * being interrupted.  One caveat is however, that we must do this per
195  * processor and cannot use the smp version of ctl_clear_bit().
196  * On VM we only get one interrupt per virtally presented machinecheck.
197  * Though one suffices, we may get one interrupt per (virtual) processor.
198  */
199         if (mcck.warning) {     /* WARNING pending ? */
200                 static int mchchk_wng_posted = 0;
201                 /*
202                  * Use single machine clear, as we cannot handle smp right now
203                  */
204                 __ctl_clear_bit(14, 24);        /* Disable WARNING MCH */
205                 if (xchg(&mchchk_wng_posted, 1) == 0)
206                         kill_proc(1, SIGPWR, 1);
207         }
208 #endif
209
210         if (mcck.kill_task) {
211                 local_irq_enable();
212                 printk(KERN_EMERG "mcck: Terminating task because of machine "
213                        "malfunction (code 0x%016llx).\n", mcck.mcck_code);
214                 printk(KERN_EMERG "mcck: task: %s, pid: %d.\n",
215                        current->comm, current->pid);
216                 do_exit(SIGSEGV);
217         }
218 }
219
220 /*
221  * returns 0 if all registers could be validated
222  * returns 1 otherwise
223  */
224 static int
225 s390_revalidate_registers(struct mci *mci)
226 {
227         int kill_task;
228         u64 tmpclock;
229         u64 zero;
230         void *fpt_save_area, *fpt_creg_save_area;
231
232         kill_task = 0;
233         zero = 0;
234         /* General purpose registers */
235         if (!mci->gr)
236                 /*
237                  * General purpose registers couldn't be restored and have
238                  * unknown contents. Process needs to be terminated.
239                  */
240                 kill_task = 1;
241
242         /* Revalidate floating point registers */
243         if (!mci->fp)
244                 /*
245                  * Floating point registers can't be restored and
246                  * therefore the process needs to be terminated.
247                  */
248                 kill_task = 1;
249
250 #ifndef CONFIG_64BIT
251         asm volatile("ld 0,0(%0)\n"
252                      "ld 2,8(%0)\n"
253                      "ld 4,16(%0)\n"
254                      "ld 6,24(%0)"
255                      : : "a" (&S390_lowcore.floating_pt_save_area));
256 #endif
257
258         if (MACHINE_HAS_IEEE) {
259 #ifdef CONFIG_64BIT
260                 fpt_save_area = &S390_lowcore.floating_pt_save_area;
261                 fpt_creg_save_area = &S390_lowcore.fpt_creg_save_area;
262 #else
263                 fpt_save_area = (void *) S390_lowcore.extended_save_area_addr;
264                 fpt_creg_save_area = fpt_save_area+128;
265 #endif
266                 /* Floating point control register */
267                 if (!mci->fc) {
268                         /*
269                          * Floating point control register can't be restored.
270                          * Task will be terminated.
271                          */
272                         asm volatile ("lfpc 0(%0)" : : "a" (&zero), "m" (zero));
273                         kill_task = 1;
274
275                 }
276                 else
277                         asm volatile (
278                                 "lfpc 0(%0)"
279                                 : : "a" (fpt_creg_save_area));
280
281                 asm volatile("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("lam 0,15,0(%0)"
302                      : : "a" (&S390_lowcore.access_regs_save_area));
303         if (!mci->ar)
304                 /*
305                  * Access registers have unknown contents.
306                  * Terminating task.
307                  */
308                 kill_task = 1;
309
310         /* Revalidate control registers */
311         if (!mci->cr)
312                 /*
313                  * Control registers have unknown contents.
314                  * Can't recover and therefore stopping machine.
315                  */
316                 s390_handle_damage("invalid control registers.");
317         else
318 #ifdef CONFIG_64BIT
319                 asm volatile("lctlg 0,15,0(%0)"
320                              : : "a" (&S390_lowcore.cregs_save_area));
321 #else
322                 asm volatile("lctl 0,15,0(%0)"
323                              : : "a" (&S390_lowcore.cregs_save_area));
324 #endif
325
326         /*
327          * We don't even try to revalidate the TOD register, since we simply
328          * can't write something sensible into that register.
329          */
330
331 #ifdef CONFIG_64BIT
332         /*
333          * See if we can revalidate the TOD programmable register with its
334          * old contents (should be zero) otherwise set it to zero.
335          */
336         if (!mci->pr)
337                 asm volatile("sr 0,0\n"
338                              "sckpf"
339                              : : : "0", "cc");
340         else
341                 asm volatile(
342                         "l 0,0(%0)\n"
343                         "sckpf"
344                         : : "a" (&S390_lowcore.tod_progreg_save_area) : "0", "cc");
345 #endif
346
347         /* Revalidate clock comparator register */
348         asm volatile ("stck 0(%1)\n"
349                       "sckc 0(%1)"
350                       : "=m" (tmpclock) : "a" (&(tmpclock)) : "cc", "memory");
351
352         /* Check if old PSW is valid */
353         if (!mci->wp)
354                 /*
355                  * Can't tell if we come from user or kernel mode
356                  * -> stopping machine.
357                  */
358                 s390_handle_damage("old psw invalid.");
359
360         if (!mci->ms || !mci->pm || !mci->ia)
361                 kill_task = 1;
362
363         return kill_task;
364 }
365
366 #define MAX_IPD_COUNT   29
367 #define MAX_IPD_TIME    (5 * 60 * USEC_PER_SEC) /* 5 minutes */
368
369 /*
370  * machine check handler.
371  */
372 void
373 s390_do_machine_check(struct pt_regs *regs)
374 {
375         static DEFINE_SPINLOCK(ipd_lock);
376         static unsigned long long last_ipd;
377         static int ipd_count;
378         unsigned long long tmp;
379         struct mci *mci;
380         struct mcck_struct *mcck;
381         int umode;
382
383         mci = (struct mci *) &S390_lowcore.mcck_interruption_code;
384         mcck = &__get_cpu_var(cpu_mcck);
385         umode = user_mode(regs);
386
387         if (mci->sd)
388                 /* System damage -> stopping machine */
389                 s390_handle_damage("received system damage machine check.");
390
391         if (mci->pd) {
392                 if (mci->b) {
393                         /* Processing backup -> verify if we can survive this */
394                         u64 z_mcic, o_mcic, t_mcic;
395 #ifdef CONFIG_64BIT
396                         z_mcic = (1ULL<<63 | 1ULL<<59 | 1ULL<<29);
397                         o_mcic = (1ULL<<43 | 1ULL<<42 | 1ULL<<41 | 1ULL<<40 |
398                                   1ULL<<36 | 1ULL<<35 | 1ULL<<34 | 1ULL<<32 |
399                                   1ULL<<30 | 1ULL<<21 | 1ULL<<20 | 1ULL<<17 |
400                                   1ULL<<16);
401 #else
402                         z_mcic = (1ULL<<63 | 1ULL<<59 | 1ULL<<57 | 1ULL<<50 |
403                                   1ULL<<29);
404                         o_mcic = (1ULL<<43 | 1ULL<<42 | 1ULL<<41 | 1ULL<<40 |
405                                   1ULL<<36 | 1ULL<<35 | 1ULL<<34 | 1ULL<<32 |
406                                   1ULL<<30 | 1ULL<<20 | 1ULL<<17 | 1ULL<<16);
407 #endif
408                         t_mcic = *(u64 *)mci;
409
410                         if (((t_mcic & z_mcic) != 0) ||
411                             ((t_mcic & o_mcic) != o_mcic)) {
412                                 s390_handle_damage("processing backup machine "
413                                                    "check with damage.");
414                         }
415
416                         /*
417                          * Nullifying exigent condition, therefore we might
418                          * retry this instruction.
419                          */
420
421                         spin_lock(&ipd_lock);
422
423                         tmp = get_clock();
424
425                         if (((tmp - last_ipd) >> 12) < MAX_IPD_TIME)
426                                 ipd_count++;
427                         else
428                                 ipd_count = 1;
429
430                         last_ipd = tmp;
431
432                         if (ipd_count == MAX_IPD_COUNT)
433                                 s390_handle_damage("too many ipd retries.");
434
435                         spin_unlock(&ipd_lock);
436                 }
437                 else {
438                         /* Processing damage -> stopping machine */
439                         s390_handle_damage("received instruction processing "
440                                            "damage machine check.");
441                 }
442         }
443         if (s390_revalidate_registers(mci)) {
444                 if (umode) {
445                         /*
446                          * Couldn't restore all register contents while in
447                          * user mode -> mark task for termination.
448                          */
449                         mcck->kill_task = 1;
450                         mcck->mcck_code = *(unsigned long long *) mci;
451                         set_thread_flag(TIF_MCCK_PENDING);
452                 }
453                 else
454                         /*
455                          * Couldn't restore all register contents while in
456                          * kernel mode -> stopping machine.
457                          */
458                         s390_handle_damage("unable to revalidate registers.");
459         }
460
461         if (mci->se)
462                 /* Storage error uncorrected */
463                 s390_handle_damage("received storage error uncorrected "
464                                    "machine check.");
465
466         if (mci->ke)
467                 /* Storage key-error uncorrected */
468                 s390_handle_damage("received storage key-error uncorrected "
469                                    "machine check.");
470
471         if (mci->ds && mci->fa)
472                 /* Storage degradation */
473                 s390_handle_damage("received storage degradation machine "
474                                    "check.");
475
476         if (mci->cp) {
477                 /* Channel report word pending */
478                 mcck->channel_report = 1;
479                 set_thread_flag(TIF_MCCK_PENDING);
480         }
481
482         if (mci->w) {
483                 /* Warning pending */
484                 mcck->warning = 1;
485                 set_thread_flag(TIF_MCCK_PENDING);
486         }
487 }
488
489 /*
490  * s390_init_machine_check
491  *
492  * initialize machine check handling
493  */
494 static int
495 machine_check_init(void)
496 {
497         init_MUTEX_LOCKED(&m_sem);
498         ctl_clear_bit(14, 25);  /* disable external damage MCH */
499         ctl_set_bit(14, 27);    /* enable system recovery MCH */
500 #ifdef CONFIG_MACHCHK_WARNING
501         ctl_set_bit(14, 24);    /* enable warning MCH */
502 #endif
503         return 0;
504 }
505
506 /*
507  * Initialize the machine check handler really early to be able to
508  * catch all machine checks that happen during boot
509  */
510 arch_initcall(machine_check_init);
511
512 /*
513  * Machine checks for the channel subsystem must be enabled
514  * after the channel subsystem is initialized
515  */
516 static int __init
517 machine_check_crw_init (void)
518 {
519         kernel_thread(s390_collect_crw_info, &m_sem, CLONE_FS|CLONE_FILES);
520         ctl_set_bit(14, 28);    /* enable channel report MCH */
521         return 0;
522 }
523
524 device_initcall (machine_check_crw_init);