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
  17 #include <asm/lowcore.h>
  18
  19 #include "s390mach.h"
  20
  21 #define DBG printk
  22 // #define DBG(args,...) do {} while (0);
  23
  24 static struct semaphore m_sem;
  25
  26 extern int css_process_crw(int, int);
  27 extern int chsc_process_crw(void);
  28 extern int chp_process_crw(int, int);
  29 extern void css_reiterate_subchannels(void);
  30
  31 extern struct workqueue_struct *slow_path_wq;
  32 extern struct work_struct slow_path_work;
  33
  34 static NORET_TYPE void
  35 s390_handle_damage(char *msg)
  36 {
  37 #ifdef CONFIG_SMP
  38         smp_send_stop();
  39 #endif
  40         disabled_wait((unsigned long) __builtin_return_address(0));
  41         for(;;);
  42 }
  43
  44 /*
  45  * Retrieve CRWs and call function to handle event.
  46  *
  47  * Note : we currently process CRWs for io and chsc subchannels only
  48  */
  49 static int
  50 s390_collect_crw_info(void *param)
  51 {
  52         struct crw crw[2];
  53         int ccode, ret, slow;
  54         struct semaphore *sem;
  55         unsigned int chain;
  56
  57         sem = (struct semaphore *)param;
  58         /* Set a nice name. */
  59         daemonize("kmcheck");
  60 repeat:
  61         down_interruptible(sem);
  62         slow = 0;
  63         chain = 0;
  64         while (1) {
  65                 if (unlikely(chain > 1)) {
  66                         struct crw tmp_crw;
  67
  68                         printk(KERN_WARNING"%s: Code does not support more "
  69                                "than two chained crws; please report to "
  70                                "linux390@de.ibm.com!\n", __FUNCTION__);
  71                         ccode = stcrw(&tmp_crw);
  72                         printk(KERN_WARNING"%s: crw reports slct=%d, oflw=%d, "
  73                                "chn=%d, rsc=%X, anc=%d, erc=%X, rsid=%X\n",
  74                                __FUNCTION__, tmp_crw.slct, tmp_crw.oflw,
  75                                tmp_crw.chn, tmp_crw.rsc, tmp_crw.anc,
  76                                tmp_crw.erc, tmp_crw.rsid);
  77                         printk(KERN_WARNING"%s: This was crw number %x in the "
  78                                "chain\n", __FUNCTION__, chain);
  79                         if (ccode != 0)
  80                                 break;
  81                         chain = tmp_crw.chn ? chain + 1 : 0;
  82                         continue;
  83                 }
  84                 ccode = stcrw(&crw[chain]);
  85                 if (ccode != 0)
  86                         break;
  87                 DBG(KERN_DEBUG "crw_info : CRW reports slct=%d, oflw=%d, "
  88                     "chn=%d, rsc=%X, anc=%d, erc=%X, rsid=%X\n",
  89                     crw[chain].slct, crw[chain].oflw, crw[chain].chn,
  90                     crw[chain].rsc, crw[chain].anc, crw[chain].erc,
  91                     crw[chain].rsid);
  92                 /* Check for overflows. */
  93                 if (crw[chain].oflw) {
  94                         pr_debug("%s: crw overflow detected!\n", __FUNCTION__);
  95                         css_reiterate_subchannels();
  96                         chain = 0;
  97                         slow = 1;
  98                         continue;
  99                 }
 100                 switch (crw[chain].rsc) {
 101                 case CRW_RSC_SCH:
 102                         if (crw[0].chn && !chain)
 103                                 break;
 104                         pr_debug("source is subchannel %04X\n", crw[0].rsid);
 105                         ret = css_process_crw (crw[0].rsid,
 106                                                chain ? crw[1].rsid : 0);
 107                         if (ret == -EAGAIN)
 108                                 slow = 1;
 109                         break;
 110                 case CRW_RSC_MONITOR:
 111                         pr_debug("source is monitoring facility\n");
 112                         break;
 113                 case CRW_RSC_CPATH:
 114                         pr_debug("source is channel path %02X\n", crw[0].rsid);
 115                         switch (crw[0].erc) {
 116                         case CRW_ERC_IPARM: /* Path has come. */
 117                                 ret = chp_process_crw(crw[0].rsid, 1);
 118                                 break;
 119                         case CRW_ERC_PERRI: /* Path has gone. */
 120                         case CRW_ERC_PERRN:
 121                                 ret = chp_process_crw(crw[0].rsid, 0);
 122                                 break;
 123                         default:
 124                                 pr_debug("Don't know how to handle erc=%x\n",
 125                                          crw[0].erc);
 126                                 ret = 0;
 127                         }
 128                         if (ret == -EAGAIN)
 129                                 slow = 1;
 130                         break;
 131                 case CRW_RSC_CONFIG:
 132                         pr_debug("source is configuration-alert facility\n");
 133                         break;
 134                 case CRW_RSC_CSS:
 135                         pr_debug("source is channel subsystem\n");
 136                         ret = chsc_process_crw();
 137                         if (ret == -EAGAIN)
 138                                 slow = 1;
 139                         break;
 140                 default:
 141                         pr_debug("unknown source\n");
 142                         break;
 143                 }
 144                 /* chain is always 0 or 1 here. */
 145                 chain = crw[chain].chn ? chain + 1 : 0;
 146         }
 147         if (slow)
 148                 queue_work(slow_path_wq, &slow_path_work);
 149         goto repeat;
 150         return 0;
 151 }
 152
 153 struct mcck_struct {
 154         int kill_task;
 155         int channel_report;
 156         int warning;
 157         unsigned long long mcck_code;
 158 };
 159
 160 static DEFINE_PER_CPU(struct mcck_struct, cpu_mcck);
 161
 162 /*
 163  * Main machine check handler function. Will be called with interrupts enabled
 164  * or disabled and machine checks enabled or disabled.
 165  */
 166 void
 167 s390_handle_mcck(void)
 168 {
 169         unsigned long flags;
 170         struct mcck_struct mcck;
 171
 172         /*
 173          * Disable machine checks and get the current state of accumulated
 174          * machine checks. Afterwards delete the old state and enable machine
 175          * checks again.
 176          */
 177         local_irq_save(flags);
 178         local_mcck_disable();
 179         mcck = __get_cpu_var(cpu_mcck);
 180         memset(&__get_cpu_var(cpu_mcck), 0, sizeof(struct mcck_struct));
 181         clear_thread_flag(TIF_MCCK_PENDING);
 182         local_mcck_enable();
 183         local_irq_restore(flags);
 184
 185         if (mcck.channel_report)
 186                 up(&m_sem);
 187
 188 #ifdef CONFIG_MACHCHK_WARNING
 189 /*
 190  * The warning may remain for a prolonged period on the bare iron.
 191  * (actually till the machine is powered off, or until the problem is gone)
 192  * So we just stop listening for the WARNING MCH and prevent continuously
 193  * being interrupted.  One caveat is however, that we must do this per
 194  * processor and cannot use the smp version of ctl_clear_bit().
 195  * On VM we only get one interrupt per virtally presented machinecheck.
 196  * Though one suffices, we may get one interrupt per (virtual) processor.
 197  */
 198         if (mcck.warning) {     /* WARNING pending ? */
 199                 static int mchchk_wng_posted = 0;
 200                 /*
 201                  * Use single machine clear, as we cannot handle smp right now
 202                  */
 203                 __ctl_clear_bit(14, 24);        /* Disable WARNING MCH */
 204                 if (xchg(&mchchk_wng_posted, 1) == 0)
 205                         kill_proc(1, SIGPWR, 1);
 206         }
 207 #endif
 208
 209         if (mcck.kill_task) {
 210                 local_irq_enable();
 211                 printk(KERN_EMERG "mcck: Terminating task because of machine "
 212                        "malfunction (code 0x%016llx).\n", mcck.mcck_code);
 213                 printk(KERN_EMERG "mcck: task: %s, pid: %d.\n",
 214                        current->comm, current->pid);
 215                 do_exit(SIGSEGV);
 216         }
 217 }
 218
 219 /*
 220  * returns 0 if all registers could be validated
 221  * returns 1 otherwise
 222  */
 223 static int
 224 s390_revalidate_registers(struct mci *mci)
 225 {
 226         int kill_task;
 227         u64 tmpclock;
 228         u64 zero;
 229         void *fpt_save_area, *fpt_creg_save_area;
 230
 231         kill_task = 0;
 232         zero = 0;
 233         /* General purpose registers */
 234         if (!mci->gr)
 235                 /*
 236                  * General purpose registers couldn't be restored and have
 237                  * unknown contents. Process needs to be terminated.
 238                  */
 239                 kill_task = 1;
 240
 241         /* Revalidate floating point registers */
 242         if (!mci->fp)
 243                 /*
 244                  * Floating point registers can't be restored and
 245                  * therefore the process needs to be terminated.
 246                  */
 247                 kill_task = 1;
 248
 249 #ifndef CONFIG_64BIT
 250         asm volatile("ld 0,0(%0)\n"
 251                      "ld 2,8(%0)\n"
 252                      "ld 4,16(%0)\n"
 253                      "ld 6,24(%0)"
 254                      : : "a" (&S390_lowcore.floating_pt_save_area));
 255 #endif
 256
 257         if (MACHINE_HAS_IEEE) {
 258 #ifdef CONFIG_64BIT
 259                 fpt_save_area = &S390_lowcore.floating_pt_save_area;
 260                 fpt_creg_save_area = &S390_lowcore.fpt_creg_save_area;
 261 #else
 262                 fpt_save_area = (void *) S390_lowcore.extended_save_area_addr;
 263                 fpt_creg_save_area = fpt_save_area+128;
 264 #endif
 265                 /* Floating point control register */
 266                 if (!mci->fc) {
 267                         /*
 268                          * Floating point control register can't be restored.
 269                          * Task will be terminated.
 270                          */
 271                         asm volatile ("lfpc 0(%0)" : : "a" (&zero), "m" (zero));
 272                         kill_task = 1;
 273
 274                 }
 275                 else
 276                         asm volatile (
 277                                 "lfpc 0(%0)"
 278                                 : : "a" (fpt_creg_save_area));
 279
 280                 asm volatile("ld  0,0(%0)\n"
 281                              "ld  1,8(%0)\n"
 282                              "ld  2,16(%0)\n"
 283                              "ld  3,24(%0)\n"
 284                              "ld  4,32(%0)\n"
 285                              "ld  5,40(%0)\n"
 286                              "ld  6,48(%0)\n"
 287                              "ld  7,56(%0)\n"
 288                              "ld  8,64(%0)\n"
 289                              "ld  9,72(%0)\n"
 290                              "ld 10,80(%0)\n"
 291                              "ld 11,88(%0)\n"
 292                              "ld 12,96(%0)\n"
 293                              "ld 13,104(%0)\n"
 294                              "ld 14,112(%0)\n"
 295                              "ld 15,120(%0)\n"
 296                              : : "a" (fpt_save_area));
 297         }
 298
 299         /* Revalidate access registers */
 300         asm volatile("lam 0,15,0(%0)"
 301                      : : "a" (&S390_lowcore.access_regs_save_area));
 302         if (!mci->ar)
 303                 /*
 304                  * Access registers have unknown contents.
 305                  * Terminating task.
 306                  */
 307                 kill_task = 1;
 308
 309         /* Revalidate control registers */
 310         if (!mci->cr)
 311                 /*
 312                  * Control registers have unknown contents.
 313                  * Can't recover and therefore stopping machine.
 314                  */
 315                 s390_handle_damage("invalid control registers.");
 316         else
 317 #ifdef CONFIG_64BIT
 318                 asm volatile("lctlg 0,15,0(%0)"
 319                              : : "a" (&S390_lowcore.cregs_save_area));
 320 #else
 321                 asm volatile("lctl 0,15,0(%0)"
 322                              : : "a" (&S390_lowcore.cregs_save_area));
 323 #endif
 324
 325         /*
 326          * We don't even try to revalidate the TOD register, since we simply
 327          * can't write something sensible into that register.
 328          */
 329
 330 #ifdef CONFIG_64BIT
 331         /*
 332          * See if we can revalidate the TOD programmable register with its
 333          * old contents (should be zero) otherwise set it to zero.
 334          */
 335         if (!mci->pr)
 336                 asm volatile("sr 0,0\n"
 337                              "sckpf"
 338                              : : : "0", "cc");
 339         else
 340                 asm volatile(
 341                         "l 0,0(%0)\n"
 342                         "sckpf"
 343                         : : "a" (&S390_lowcore.tod_progreg_save_area) : "0", "cc");
 344 #endif
 345
 346         /* Revalidate clock comparator register */
 347         asm volatile ("stck 0(%1)\n"
 348                       "sckc 0(%1)"
 349                       : "=m" (tmpclock) : "a" (&(tmpclock)) : "cc", "memory");
 350
 351         /* Check if old PSW is valid */
 352         if (!mci->wp)
 353                 /*
 354                  * Can't tell if we come from user or kernel mode
 355                  * -> stopping machine.
 356                  */
 357                 s390_handle_damage("old psw invalid.");
 358
 359         if (!mci->ms || !mci->pm || !mci->ia)
 360                 kill_task = 1;
 361
 362         return kill_task;
 363 }
 364
 365 #define MAX_IPD_COUNT   29
 366 #define MAX_IPD_TIME    (5 * 60 * 100 * 1000) /* 5 minutes */
 367
 368 /*
 369  * machine check handler.
 370  */
 371 void
 372 s390_do_machine_check(struct pt_regs *regs)
 373 {
 374         static DEFINE_SPINLOCK(ipd_lock);
 375         static unsigned long long last_ipd;
 376         static int ipd_count;
 377         unsigned long long tmp;
 378         struct mci *mci;
 379         struct mcck_struct *mcck;
 380         int umode;
 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->se)
 461                 /* Storage error uncorrected */
 462                 s390_handle_damage("received storage error uncorrected "
 463                                    "machine check.");
 464
 465         if (mci->ke)
 466                 /* Storage key-error uncorrected */
 467                 s390_handle_damage("received storage key-error uncorrected "
 468                                    "machine check.");
 469
 470         if (mci->ds && mci->fa)
 471                 /* Storage degradation */
 472                 s390_handle_damage("received storage degradation machine "
 473                                    "check.");
 474
 475         if (mci->cp) {
 476                 /* Channel report word pending */
 477                 mcck->channel_report = 1;
 478                 set_thread_flag(TIF_MCCK_PENDING);
 479         }
 480
 481         if (mci->w) {
 482                 /* Warning pending */
 483                 mcck->warning = 1;
 484                 set_thread_flag(TIF_MCCK_PENDING);
 485         }
 486 }
 487
 488 /*
 489  * s390_init_machine_check
 490  *
 491  * initialize machine check handling
 492  */
 493 static int
 494 machine_check_init(void)
 495 {
 496         init_MUTEX_LOCKED(&m_sem);
 497         ctl_clear_bit(14, 25);  /* disable external damage MCH */
 498         ctl_set_bit(14, 27);    /* enable system recovery MCH */
 499 #ifdef CONFIG_MACHCHK_WARNING
 500         ctl_set_bit(14, 24);    /* enable warning MCH */
 501 #endif
 502         return 0;
 503 }
 504
 505 /*
 506  * Initialize the machine check handler really early to be able to
 507  * catch all machine checks that happen during boot
 508  */
 509 arch_initcall(machine_check_init);
 510
 511 /*
 512  * Machine checks for the channel subsystem must be enabled
 513  * after the channel subsystem is initialized
 514  */
 515 static int __init
 516 machine_check_crw_init (void)
 517 {
 518         kernel_thread(s390_collect_crw_info, &m_sem, CLONE_FS|CLONE_FILES);
 519         ctl_set_bit(14, 28);    /* enable channel report MCH */
 520         return 0;
 521 }
 522
 523 device_initcall (machine_check_crw_init);