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