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, ret, slow;
  49         struct semaphore *sem;
  50         unsigned int chain;
  51
  52         sem = (struct semaphore *)param;
  53 repeat:
  54         down_interruptible(sem);
  55         slow = 0;
  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", __FUNCTION__);
  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                                __FUNCTION__, 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", __FUNCTION__, 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", __FUNCTION__);
  88                         css_reiterate_subchannels();
  89                         chain = 0;
  90                         slow = 1;
  91                         continue;
  92                 }
  93                 switch (crw[chain].rsc) {
  94                 case CRW_RSC_SCH:
  95                         if (crw[0].chn && !chain)
  96                                 break;
  97                         pr_debug("source is subchannel %04X\n", crw[0].rsid);
  98                         ret = css_process_crw (crw[0].rsid,
  99                                                chain ? crw[1].rsid : 0);
 100                         if (ret == -EAGAIN)
 101                                 slow = 1;
 102                         break;
 103                 case CRW_RSC_MONITOR:
 104                         pr_debug("source is monitoring facility\n");
 105                         break;
 106                 case CRW_RSC_CPATH:
 107                         pr_debug("source is channel path %02X\n", crw[0].rsid);
 108                         /*
 109                          * Check for solicited machine checks. These are
 110                          * created by reset channel path and need not be
 111                          * reported to the common I/O layer.
 112                          */
 113                         if (crw[chain].slct) {
 114                                 pr_debug("solicited machine check for "
 115                                          "channel path %02X\n", crw[0].rsid);
 116                                 break;
 117                         }
 118                         switch (crw[0].erc) {
 119                         case CRW_ERC_IPARM: /* Path has come. */
 120                                 ret = chp_process_crw(crw[0].rsid, 1);
 121                                 break;
 122                         case CRW_ERC_PERRI: /* Path has gone. */
 123                         case CRW_ERC_PERRN:
 124                                 ret = chp_process_crw(crw[0].rsid, 0);
 125                                 break;
 126                         default:
 127                                 pr_debug("Don't know how to handle erc=%x\n",
 128                                          crw[0].erc);
 129                                 ret = 0;
 130                         }
 131                         if (ret == -EAGAIN)
 132                                 slow = 1;
 133                         break;
 134                 case CRW_RSC_CONFIG:
 135                         pr_debug("source is configuration-alert facility\n");
 136                         break;
 137                 case CRW_RSC_CSS:
 138                         pr_debug("source is channel subsystem\n");
 139                         ret = chsc_process_crw();
 140                         if (ret == -EAGAIN)
 141                                 slow = 1;
 142                         break;
 143                 default:
 144                         pr_debug("unknown source\n");
 145                         break;
 146                 }
 147                 /* chain is always 0 or 1 here. */
 148                 chain = crw[chain].chn ? chain + 1 : 0;
 149         }
 150         if (slow)
 151                 queue_work(slow_path_wq, &slow_path_work);
 152         goto repeat;
 153         return 0;
 154 }
 155
 156 struct mcck_struct {
 157         int kill_task;
 158         int channel_report;
 159         int warning;
 160         unsigned long long mcck_code;
 161 };
 162
 163 static DEFINE_PER_CPU(struct mcck_struct, cpu_mcck);
 164
 165 /*
 166  * Main machine check handler function. Will be called with interrupts enabled
 167  * or disabled and machine checks enabled or disabled.
 168  */
 169 void
 170 s390_handle_mcck(void)
 171 {
 172         unsigned long flags;
 173         struct mcck_struct mcck;
 174
 175         /*
 176          * Disable machine checks and get the current state of accumulated
 177          * machine checks. Afterwards delete the old state and enable machine
 178          * checks again.
 179          */
 180         local_irq_save(flags);
 181         local_mcck_disable();
 182         mcck = __get_cpu_var(cpu_mcck);
 183         memset(&__get_cpu_var(cpu_mcck), 0, sizeof(struct mcck_struct));
 184         clear_thread_flag(TIF_MCCK_PENDING);
 185         local_mcck_enable();
 186         local_irq_restore(flags);
 187
 188         if (mcck.channel_report)
 189                 up(&m_sem);
 190
 191 #ifdef CONFIG_MACHCHK_WARNING
 192 /*
 193  * The warning may remain for a prolonged period on the bare iron.
 194  * (actually till the machine is powered off, or until the problem is gone)
 195  * So we just stop listening for the WARNING MCH and prevent continuously
 196  * being interrupted.  One caveat is however, that we must do this per
 197  * processor and cannot use the smp version of ctl_clear_bit().
 198  * On VM we only get one interrupt per virtally presented machinecheck.
 199  * Though one suffices, we may get one interrupt per (virtual) processor.
 200  */
 201         if (mcck.warning) {     /* WARNING pending ? */
 202                 static int mchchk_wng_posted = 0;
 203                 /*
 204                  * Use single machine clear, as we cannot handle smp right now
 205                  */
 206                 __ctl_clear_bit(14, 24);        /* Disable WARNING MCH */
 207                 if (xchg(&mchchk_wng_posted, 1) == 0)
 208                         kill_cad_pid(SIGPWR, 1);
 209         }
 210 #endif
 211
 212         if (mcck.kill_task) {
 213                 local_irq_enable();
 214                 printk(KERN_EMERG "mcck: Terminating task because of machine "
 215                        "malfunction (code 0x%016llx).\n", mcck.mcck_code);
 216                 printk(KERN_EMERG "mcck: task: %s, pid: %d.\n",
 217                        current->comm, current->pid);
 218                 do_exit(SIGSEGV);
 219         }
 220 }
 221
 222 /*
 223  * returns 0 if all registers could be validated
 224  * returns 1 otherwise
 225  */
 226 static int
 227 s390_revalidate_registers(struct mci *mci)
 228 {
 229         int kill_task;
 230         u64 tmpclock;
 231         u64 zero;
 232         void *fpt_save_area, *fpt_creg_save_area;
 233
 234         kill_task = 0;
 235         zero = 0;
 236         /* General purpose registers */
 237         if (!mci->gr)
 238                 /*
 239                  * General purpose registers couldn't be restored and have
 240                  * unknown contents. Process needs to be terminated.
 241                  */
 242                 kill_task = 1;
 243
 244         /* Revalidate floating point registers */
 245         if (!mci->fp)
 246                 /*
 247                  * Floating point registers can't be restored and
 248                  * therefore the process needs to be terminated.
 249                  */
 250                 kill_task = 1;
 251
 252 #ifndef CONFIG_64BIT
 253         asm volatile(
 254                 "       ld      0,0(%0)\n"
 255                 "       ld      2,8(%0)\n"
 256                 "       ld      4,16(%0)\n"
 257                 "       ld      6,24(%0)"
 258                 : : "a" (&S390_lowcore.floating_pt_save_area));
 259 #endif
 260
 261         if (MACHINE_HAS_IEEE) {
 262 #ifdef CONFIG_64BIT
 263                 fpt_save_area = &S390_lowcore.floating_pt_save_area;
 264                 fpt_creg_save_area = &S390_lowcore.fpt_creg_save_area;
 265 #else
 266                 fpt_save_area = (void *) S390_lowcore.extended_save_area_addr;
 267                 fpt_creg_save_area = fpt_save_area+128;
 268 #endif
 269                 /* Floating point control register */
 270                 if (!mci->fc) {
 271                         /*
 272                          * Floating point control register can't be restored.
 273                          * Task will be terminated.
 274                          */
 275                         asm volatile("lfpc 0(%0)" : : "a" (&zero), "m" (zero));
 276                         kill_task = 1;
 277
 278                 } else
 279                         asm volatile("lfpc 0(%0)" : : "a" (fpt_creg_save_area));
 280
 281                 asm volatile(
 282                         "       ld      0,0(%0)\n"
 283                         "       ld      1,8(%0)\n"
 284                         "       ld      2,16(%0)\n"
 285                         "       ld      3,24(%0)\n"
 286                         "       ld      4,32(%0)\n"
 287                         "       ld      5,40(%0)\n"
 288                         "       ld      6,48(%0)\n"
 289                         "       ld      7,56(%0)\n"
 290                         "       ld      8,64(%0)\n"
 291                         "       ld      9,72(%0)\n"
 292                         "       ld      10,80(%0)\n"
 293                         "       ld      11,88(%0)\n"
 294                         "       ld      12,96(%0)\n"
 295                         "       ld      13,104(%0)\n"
 296                         "       ld      14,112(%0)\n"
 297                         "       ld      15,120(%0)\n"
 298                         : : "a" (fpt_save_area));
 299         }
 300
 301         /* Revalidate access registers */
 302         asm volatile(
 303                 "       lam     0,15,0(%0)"
 304                 : : "a" (&S390_lowcore.access_regs_save_area));
 305         if (!mci->ar)
 306                 /*
 307                  * Access registers have unknown contents.
 308                  * Terminating task.
 309                  */
 310                 kill_task = 1;
 311
 312         /* Revalidate control registers */
 313         if (!mci->cr)
 314                 /*
 315                  * Control registers have unknown contents.
 316                  * Can't recover and therefore stopping machine.
 317                  */
 318                 s390_handle_damage("invalid control registers.");
 319         else
 320 #ifdef CONFIG_64BIT
 321                 asm volatile(
 322                         "       lctlg   0,15,0(%0)"
 323                         : : "a" (&S390_lowcore.cregs_save_area));
 324 #else
 325                 asm volatile(
 326                         "       lctl    0,15,0(%0)"
 327                         : : "a" (&S390_lowcore.cregs_save_area));
 328 #endif
 329
 330         /*
 331          * We don't even try to revalidate the TOD register, since we simply
 332          * can't write something sensible into that register.
 333          */
 334
 335 #ifdef CONFIG_64BIT
 336         /*
 337          * See if we can revalidate the TOD programmable register with its
 338          * old contents (should be zero) otherwise set it to zero.
 339          */
 340         if (!mci->pr)
 341                 asm volatile(
 342                         "       sr      0,0\n"
 343                         "       sckpf"
 344                         : : : "0", "cc");
 345         else
 346                 asm volatile(
 347                         "       l       0,0(%0)\n"
 348                         "       sckpf"
 349                         : : "a" (&S390_lowcore.tod_progreg_save_area)
 350                         : "0", "cc");
 351 #endif
 352
 353         /* Revalidate clock comparator register */
 354         asm volatile(
 355                 "       stck    0(%1)\n"
 356                 "       sckc    0(%1)"
 357                 : "=m" (tmpclock) : "a" (&(tmpclock)) : "cc", "memory");
 358
 359         /* Check if old PSW is valid */
 360         if (!mci->wp)
 361                 /*
 362                  * Can't tell if we come from user or kernel mode
 363                  * -> stopping machine.
 364                  */
 365                 s390_handle_damage("old psw invalid.");
 366
 367         if (!mci->ms || !mci->pm || !mci->ia)
 368                 kill_task = 1;
 369
 370         return kill_task;
 371 }
 372
 373 #define MAX_IPD_COUNT   29
 374 #define MAX_IPD_TIME    (5 * 60 * USEC_PER_SEC) /* 5 minutes */
 375
 376 /*
 377  * machine check handler.
 378  */
 379 void
 380 s390_do_machine_check(struct pt_regs *regs)
 381 {
 382         static DEFINE_SPINLOCK(ipd_lock);
 383         static unsigned long long last_ipd;
 384         static int ipd_count;
 385         unsigned long long tmp;
 386         struct mci *mci;
 387         struct mcck_struct *mcck;
 388         int umode;
 389
 390         lockdep_off();
 391
 392         mci = (struct mci *) &S390_lowcore.mcck_interruption_code;
 393         mcck = &__get_cpu_var(cpu_mcck);
 394         umode = user_mode(regs);
 395
 396         if (mci->sd)
 397                 /* System damage -> stopping machine */
 398                 s390_handle_damage("received system damage machine check.");
 399
 400         if (mci->pd) {
 401                 if (mci->b) {
 402                         /* Processing backup -> verify if we can survive this */
 403                         u64 z_mcic, o_mcic, t_mcic;
 404 #ifdef CONFIG_64BIT
 405                         z_mcic = (1ULL<<63 | 1ULL<<59 | 1ULL<<29);
 406                         o_mcic = (1ULL<<43 | 1ULL<<42 | 1ULL<<41 | 1ULL<<40 |
 407                                   1ULL<<36 | 1ULL<<35 | 1ULL<<34 | 1ULL<<32 |
 408                                   1ULL<<30 | 1ULL<<21 | 1ULL<<20 | 1ULL<<17 |
 409                                   1ULL<<16);
 410 #else
 411                         z_mcic = (1ULL<<63 | 1ULL<<59 | 1ULL<<57 | 1ULL<<50 |
 412                                   1ULL<<29);
 413                         o_mcic = (1ULL<<43 | 1ULL<<42 | 1ULL<<41 | 1ULL<<40 |
 414                                   1ULL<<36 | 1ULL<<35 | 1ULL<<34 | 1ULL<<32 |
 415                                   1ULL<<30 | 1ULL<<20 | 1ULL<<17 | 1ULL<<16);
 416 #endif
 417                         t_mcic = *(u64 *)mci;
 418
 419                         if (((t_mcic & z_mcic) != 0) ||
 420                             ((t_mcic & o_mcic) != o_mcic)) {
 421                                 s390_handle_damage("processing backup machine "
 422                                                    "check with damage.");
 423                         }
 424
 425                         /*
 426                          * Nullifying exigent condition, therefore we might
 427                          * retry this instruction.
 428                          */
 429
 430                         spin_lock(&ipd_lock);
 431
 432                         tmp = get_clock();
 433
 434                         if (((tmp - last_ipd) >> 12) < MAX_IPD_TIME)
 435                                 ipd_count++;
 436                         else
 437                                 ipd_count = 1;
 438
 439                         last_ipd = tmp;
 440
 441                         if (ipd_count == MAX_IPD_COUNT)
 442                                 s390_handle_damage("too many ipd retries.");
 443
 444                         spin_unlock(&ipd_lock);
 445                 }
 446                 else {
 447                         /* Processing damage -> stopping machine */
 448                         s390_handle_damage("received instruction processing "
 449                                            "damage machine check.");
 450                 }
 451         }
 452         if (s390_revalidate_registers(mci)) {
 453                 if (umode) {
 454                         /*
 455                          * Couldn't restore all register contents while in
 456                          * user mode -> mark task for termination.
 457                          */
 458                         mcck->kill_task = 1;
 459                         mcck->mcck_code = *(unsigned long long *) mci;
 460                         set_thread_flag(TIF_MCCK_PENDING);
 461                 }
 462                 else
 463                         /*
 464                          * Couldn't restore all register contents while in
 465                          * kernel mode -> stopping machine.
 466                          */
 467                         s390_handle_damage("unable to revalidate registers.");
 468         }
 469
 470         if (mci->cd) {
 471                 /* Timing facility damage */
 472                 s390_handle_damage("TOD clock damaged");
 473         }
 474
 475         if (mci->ed && mci->ec) {
 476                 /* External damage */
 477                 if (S390_lowcore.external_damage_code & (1U << ED_ETR_SYNC))
 478                         etr_sync_check();
 479                 if (S390_lowcore.external_damage_code & (1U << ED_ETR_SWITCH))
 480                         etr_switch_to_local();
 481         }
 482
 483         if (mci->se)
 484                 /* Storage error uncorrected */
 485                 s390_handle_damage("received storage error uncorrected "
 486                                    "machine check.");
 487
 488         if (mci->ke)
 489                 /* Storage key-error uncorrected */
 490                 s390_handle_damage("received storage key-error uncorrected "
 491                                    "machine check.");
 492
 493         if (mci->ds && mci->fa)
 494                 /* Storage degradation */
 495                 s390_handle_damage("received storage degradation machine "
 496                                    "check.");
 497
 498         if (mci->cp) {
 499                 /* Channel report word pending */
 500                 mcck->channel_report = 1;
 501                 set_thread_flag(TIF_MCCK_PENDING);
 502         }
 503
 504         if (mci->w) {
 505                 /* Warning pending */
 506                 mcck->warning = 1;
 507                 set_thread_flag(TIF_MCCK_PENDING);
 508         }
 509         lockdep_on();
 510 }
 511
 512 /*
 513  * s390_init_machine_check
 514  *
 515  * initialize machine check handling
 516  */
 517 static int
 518 machine_check_init(void)
 519 {
 520         init_MUTEX_LOCKED(&m_sem);
 521         ctl_set_bit(14, 25);    /* enable external damage MCH */
 522         ctl_set_bit(14, 27);    /* enable system recovery MCH */
 523 #ifdef CONFIG_MACHCHK_WARNING
 524         ctl_set_bit(14, 24);    /* enable warning MCH */
 525 #endif
 526         return 0;
 527 }
 528
 529 /*
 530  * Initialize the machine check handler really early to be able to
 531  * catch all machine checks that happen during boot
 532  */
 533 arch_initcall(machine_check_init);
 534
 535 /*
 536  * Machine checks for the channel subsystem must be enabled
 537  * after the channel subsystem is initialized
 538  */
 539 static int __init
 540 machine_check_crw_init (void)
 541 {
 542         struct task_struct *task;
 543
 544         task = kthread_run(s390_collect_crw_info, &m_sem, "kmcheck");
 545         if (IS_ERR(task))
 546                 return PTR_ERR(task);
 547         ctl_set_bit(14, 28);    /* enable channel report MCH */
 548         return 0;
 549 }
 550
 551 device_initcall (machine_check_crw_init);