arch/s390/kernel/nmi.c

   1 /*
   2  *   Machine check handler
   3  *
   4  *    Copyright IBM Corp. 2000, 2009
   5  *    Author(s): Ingo Adlung <adlung@de.ibm.com>,
   6  *               Martin Schwidefsky <schwidefsky@de.ibm.com>,
   7  *               Cornelia Huck <cornelia.huck@de.ibm.com>,
   8  *               Heiko Carstens <heiko.carstens@de.ibm.com>,
   9  */
  10
  11 #include <linux/kernel_stat.h>
  12 #include <linux/init.h>
  13 #include <linux/errno.h>
  14 #include <linux/hardirq.h>
  15 #include <linux/time.h>
  16 #include <linux/module.h>
  17 #include <asm/lowcore.h>
  18 #include <asm/smp.h>
  19 #include <asm/etr.h>
  20 #include <asm/cputime.h>
  21 #include <asm/nmi.h>
  22 #include <asm/crw.h>
  23 #include <asm/switch_to.h>
  24
  25 struct mcck_struct {
  26         int kill_task;
  27         int channel_report;
  28         int warning;
  29         unsigned long long mcck_code;
  30 };
  31
  32 static DEFINE_PER_CPU(struct mcck_struct, cpu_mcck);
  33
  34 static void s390_handle_damage(char *msg)
  35 {
  36         smp_send_stop();
  37         disabled_wait((unsigned long) __builtin_return_address(0));
  38         while (1);
  39 }
  40
  41 /*
  42  * Main machine check handler function. Will be called with interrupts enabled
  43  * or disabled and machine checks enabled or disabled.
  44  */
  45 void s390_handle_mcck(void)
  46 {
  47         unsigned long flags;
  48         struct mcck_struct mcck;
  49
  50         /*
  51          * Disable machine checks and get the current state of accumulated
  52          * machine checks. Afterwards delete the old state and enable machine
  53          * checks again.
  54          */
  55         local_irq_save(flags);
  56         local_mcck_disable();
  57         mcck = *this_cpu_ptr(&cpu_mcck);
  58         memset(this_cpu_ptr(&cpu_mcck), 0, sizeof(mcck));
  59         clear_cpu_flag(CIF_MCCK_PENDING);
  60         local_mcck_enable();
  61         local_irq_restore(flags);
  62
  63         if (mcck.channel_report)
  64                 crw_handle_channel_report();
  65         /*
  66          * A warning may remain for a prolonged period on the bare iron.
  67          * (actually until the machine is powered off, or the problem is gone)
  68          * So we just stop listening for the WARNING MCH and avoid continuously
  69          * being interrupted.  One caveat is however, that we must do this per
  70          * processor and cannot use the smp version of ctl_clear_bit().
  71          * On VM we only get one interrupt per virtally presented machinecheck.
  72          * Though one suffices, we may get one interrupt per (virtual) cpu.
  73          */
  74         if (mcck.warning) {     /* WARNING pending ? */
  75                 static int mchchk_wng_posted = 0;
  76
  77                 /* Use single cpu clear, as we cannot handle smp here. */
  78                 __ctl_clear_bit(14, 24);        /* Disable WARNING MCH */
  79                 if (xchg(&mchchk_wng_posted, 1) == 0)
  80                         kill_cad_pid(SIGPWR, 1);
  81         }
  82         if (mcck.kill_task) {
  83                 local_irq_enable();
  84                 printk(KERN_EMERG "mcck: Terminating task because of machine "
  85                        "malfunction (code 0x%016llx).\n", mcck.mcck_code);
  86                 printk(KERN_EMERG "mcck: task: %s, pid: %d.\n",
  87                        current->comm, current->pid);
  88                 do_exit(SIGSEGV);
  89         }
  90 }
  91 EXPORT_SYMBOL_GPL(s390_handle_mcck);
  92
  93 /*
  94  * returns 0 if all registers could be validated
  95  * returns 1 otherwise
  96  */
  97 static int notrace s390_revalidate_registers(struct mci *mci)
  98 {
  99         int kill_task;
 100         u64 zero;
 101         void *fpt_save_area, *fpt_creg_save_area;
 102
 103         kill_task = 0;
 104         zero = 0;
 105
 106         if (!mci->gr) {
 107                 /*
 108                  * General purpose registers couldn't be restored and have
 109                  * unknown contents. Process needs to be terminated.
 110                  */
 111                 kill_task = 1;
 112         }
 113         if (!mci->fp) {
 114                 /*
 115                  * Floating point registers can't be restored and
 116                  * therefore the process needs to be terminated.
 117                  */
 118                 kill_task = 1;
 119         }
 120 #ifndef CONFIG_64BIT
 121         asm volatile(
 122                 "       ld      0,0(%0)\n"
 123                 "       ld      2,8(%0)\n"
 124                 "       ld      4,16(%0)\n"
 125                 "       ld      6,24(%0)"
 126                 : : "a" (&S390_lowcore.floating_pt_save_area));
 127 #endif
 128
 129         if (MACHINE_HAS_IEEE) {
 130 #ifdef CONFIG_64BIT
 131                 fpt_save_area = &S390_lowcore.floating_pt_save_area;
 132                 fpt_creg_save_area = &S390_lowcore.fpt_creg_save_area;
 133 #else
 134                 fpt_save_area = (void *) S390_lowcore.extended_save_area_addr;
 135                 fpt_creg_save_area = fpt_save_area + 128;
 136 #endif
 137                 if (!mci->fc) {
 138                         /*
 139                          * Floating point control register can't be restored.
 140                          * Task will be terminated.
 141                          */
 142                         asm volatile("lfpc 0(%0)" : : "a" (&zero), "m" (zero));
 143                         kill_task = 1;
 144
 145                 } else
 146                         asm volatile("lfpc 0(%0)" : : "a" (fpt_creg_save_area));
 147
 148                 asm volatile(
 149                         "       ld      0,0(%0)\n"
 150                         "       ld      1,8(%0)\n"
 151                         "       ld      2,16(%0)\n"
 152                         "       ld      3,24(%0)\n"
 153                         "       ld      4,32(%0)\n"
 154                         "       ld      5,40(%0)\n"
 155                         "       ld      6,48(%0)\n"
 156                         "       ld      7,56(%0)\n"
 157                         "       ld      8,64(%0)\n"
 158                         "       ld      9,72(%0)\n"
 159                         "       ld      10,80(%0)\n"
 160                         "       ld      11,88(%0)\n"
 161                         "       ld      12,96(%0)\n"
 162                         "       ld      13,104(%0)\n"
 163                         "       ld      14,112(%0)\n"
 164                         "       ld      15,120(%0)\n"
 165                         : : "a" (fpt_save_area));
 166         }
 167
 168 #ifdef CONFIG_64BIT
 169         /* Revalidate vector registers */
 170         if (MACHINE_HAS_VX && current->thread.vxrs) {
 171                 if (!mci->vr) {
 172                         /*
 173                          * Vector registers can't be restored and therefore
 174                          * the process needs to be terminated.
 175                          */
 176                         kill_task = 1;
 177                 }
 178                 restore_vx_regs((__vector128 *)
 179                                 S390_lowcore.vector_save_area_addr);
 180         }
 181 #endif
 182         /* Revalidate access registers */
 183         asm volatile(
 184                 "       lam     0,15,0(%0)"
 185                 : : "a" (&S390_lowcore.access_regs_save_area));
 186         if (!mci->ar) {
 187                 /*
 188                  * Access registers have unknown contents.
 189                  * Terminating task.
 190                  */
 191                 kill_task = 1;
 192         }
 193         /* Revalidate control registers */
 194         if (!mci->cr) {
 195                 /*
 196                  * Control registers have unknown contents.
 197                  * Can't recover and therefore stopping machine.
 198                  */
 199                 s390_handle_damage("invalid control registers.");
 200         } else {
 201 #ifdef CONFIG_64BIT
 202                 asm volatile(
 203                         "       lctlg   0,15,0(%0)"
 204                         : : "a" (&S390_lowcore.cregs_save_area));
 205 #else
 206                 asm volatile(
 207                         "       lctl    0,15,0(%0)"
 208                         : : "a" (&S390_lowcore.cregs_save_area));
 209 #endif
 210         }
 211         /*
 212          * We don't even try to revalidate the TOD register, since we simply
 213          * can't write something sensible into that register.
 214          */
 215 #ifdef CONFIG_64BIT
 216         /*
 217          * See if we can revalidate the TOD programmable register with its
 218          * old contents (should be zero) otherwise set it to zero.
 219          */
 220         if (!mci->pr)
 221                 asm volatile(
 222                         "       sr      0,0\n"
 223                         "       sckpf"
 224                         : : : "0", "cc");
 225         else
 226                 asm volatile(
 227                         "       l       0,0(%0)\n"
 228                         "       sckpf"
 229                         : : "a" (&S390_lowcore.tod_progreg_save_area)
 230                         : "0", "cc");
 231 #endif
 232         /* Revalidate clock comparator register */
 233         set_clock_comparator(S390_lowcore.clock_comparator);
 234         /* Check if old PSW is valid */
 235         if (!mci->wp)
 236                 /*
 237                  * Can't tell if we come from user or kernel mode
 238                  * -> stopping machine.
 239                  */
 240                 s390_handle_damage("old psw invalid.");
 241
 242         if (!mci->ms || !mci->pm || !mci->ia)
 243                 kill_task = 1;
 244
 245         return kill_task;
 246 }
 247
 248 #define MAX_IPD_COUNT   29
 249 #define MAX_IPD_TIME    (5 * 60 * USEC_PER_SEC) /* 5 minutes */
 250
 251 #define ED_STP_ISLAND   6       /* External damage STP island check */
 252 #define ED_STP_SYNC     7       /* External damage STP sync check */
 253 #define ED_ETR_SYNC     12      /* External damage ETR sync check */
 254 #define ED_ETR_SWITCH   13      /* External damage ETR switch to local */
 255
 256 /*
 257  * machine check handler.
 258  */
 259 void notrace s390_do_machine_check(struct pt_regs *regs)
 260 {
 261         static int ipd_count;
 262         static DEFINE_SPINLOCK(ipd_lock);
 263         static unsigned long long last_ipd;
 264         struct mcck_struct *mcck;
 265         unsigned long long tmp;
 266         struct mci *mci;
 267         int umode;
 268
 269         nmi_enter();
 270         inc_irq_stat(NMI_NMI);
 271         mci = (struct mci *) &S390_lowcore.mcck_interruption_code;
 272         mcck = this_cpu_ptr(&cpu_mcck);
 273         umode = user_mode(regs);
 274
 275         if (mci->sd) {
 276                 /* System damage -> stopping machine */
 277                 s390_handle_damage("received system damage machine check.");
 278         }
 279         if (mci->pd) {
 280                 if (mci->b) {
 281                         /* Processing backup -> verify if we can survive this */
 282                         u64 z_mcic, o_mcic, t_mcic;
 283 #ifdef CONFIG_64BIT
 284                         z_mcic = (1ULL<<63 | 1ULL<<59 | 1ULL<<29);
 285                         o_mcic = (1ULL<<43 | 1ULL<<42 | 1ULL<<41 | 1ULL<<40 |
 286                                   1ULL<<36 | 1ULL<<35 | 1ULL<<34 | 1ULL<<32 |
 287                                   1ULL<<30 | 1ULL<<21 | 1ULL<<20 | 1ULL<<17 |
 288                                   1ULL<<16);
 289 #else
 290                         z_mcic = (1ULL<<63 | 1ULL<<59 | 1ULL<<57 | 1ULL<<50 |
 291                                   1ULL<<29);
 292                         o_mcic = (1ULL<<43 | 1ULL<<42 | 1ULL<<41 | 1ULL<<40 |
 293                                   1ULL<<36 | 1ULL<<35 | 1ULL<<34 | 1ULL<<32 |
 294                                   1ULL<<30 | 1ULL<<20 | 1ULL<<17 | 1ULL<<16);
 295 #endif
 296                         t_mcic = *(u64 *)mci;
 297
 298                         if (((t_mcic & z_mcic) != 0) ||
 299                             ((t_mcic & o_mcic) != o_mcic)) {
 300                                 s390_handle_damage("processing backup machine "
 301                                                    "check with damage.");
 302                         }
 303
 304                         /*
 305                          * Nullifying exigent condition, therefore we might
 306                          * retry this instruction.
 307                          */
 308                         spin_lock(&ipd_lock);
 309                         tmp = get_tod_clock();
 310                         if (((tmp - last_ipd) >> 12) < MAX_IPD_TIME)
 311                                 ipd_count++;
 312                         else
 313                                 ipd_count = 1;
 314                         last_ipd = tmp;
 315                         if (ipd_count == MAX_IPD_COUNT)
 316                                 s390_handle_damage("too many ipd retries.");
 317                         spin_unlock(&ipd_lock);
 318                 } else {
 319                         /* Processing damage -> stopping machine */
 320                         s390_handle_damage("received instruction processing "
 321                                            "damage machine check.");
 322                 }
 323         }
 324         if (s390_revalidate_registers(mci)) {
 325                 if (umode) {
 326                         /*
 327                          * Couldn't restore all register contents while in
 328                          * user mode -> mark task for termination.
 329                          */
 330                         mcck->kill_task = 1;
 331                         mcck->mcck_code = *(unsigned long long *) mci;
 332                         set_cpu_flag(CIF_MCCK_PENDING);
 333                 } else {
 334                         /*
 335                          * Couldn't restore all register contents while in
 336                          * kernel mode -> stopping machine.
 337                          */
 338                         s390_handle_damage("unable to revalidate registers.");
 339                 }
 340         }
 341         if (mci->cd) {
 342                 /* Timing facility damage */
 343                 s390_handle_damage("TOD clock damaged");
 344         }
 345         if (mci->ed && mci->ec) {
 346                 /* External damage */
 347                 if (S390_lowcore.external_damage_code & (1U << ED_ETR_SYNC))
 348                         etr_sync_check();
 349                 if (S390_lowcore.external_damage_code & (1U << ED_ETR_SWITCH))
 350                         etr_switch_to_local();
 351                 if (S390_lowcore.external_damage_code & (1U << ED_STP_SYNC))
 352                         stp_sync_check();
 353                 if (S390_lowcore.external_damage_code & (1U << ED_STP_ISLAND))
 354                         stp_island_check();
 355         }
 356         if (mci->se)
 357                 /* Storage error uncorrected */
 358                 s390_handle_damage("received storage error uncorrected "
 359                                    "machine check.");
 360         if (mci->ke)
 361                 /* Storage key-error uncorrected */
 362                 s390_handle_damage("received storage key-error uncorrected "
 363                                    "machine check.");
 364         if (mci->ds && mci->fa)
 365                 /* Storage degradation */
 366                 s390_handle_damage("received storage degradation machine "
 367                                    "check.");
 368         if (mci->cp) {
 369                 /* Channel report word pending */
 370                 mcck->channel_report = 1;
 371                 set_cpu_flag(CIF_MCCK_PENDING);
 372         }
 373         if (mci->w) {
 374                 /* Warning pending */
 375                 mcck->warning = 1;
 376                 set_cpu_flag(CIF_MCCK_PENDING);
 377         }
 378         nmi_exit();
 379 }
 380
 381 static int __init machine_check_init(void)
 382 {
 383         ctl_set_bit(14, 25);    /* enable external damage MCH */
 384         ctl_set_bit(14, 27);    /* enable system recovery MCH */
 385         ctl_set_bit(14, 24);    /* enable warning MCH */
 386         return 0;
 387 }
 388 arch_initcall(machine_check_init);