arch/ia64/kernel/mca.c

   1 /*
   2  * File:        mca.c
   3  * Purpose:     Generic MCA handling layer
   4  *
   5  * Updated for latest kernel
   6  * Copyright (C) 2003 Hewlett-Packard Co
   7  *      David Mosberger-Tang <davidm@hpl.hp.com>
   8  *
   9  * Copyright (C) 2002 Dell Inc.
  10  * Copyright (C) Matt Domsch (Matt_Domsch@dell.com)
  11  *
  12  * Copyright (C) 2002 Intel
  13  * Copyright (C) Jenna Hall (jenna.s.hall@intel.com)
  14  *
  15  * Copyright (C) 2001 Intel
  16  * Copyright (C) Fred Lewis (frederick.v.lewis@intel.com)
  17  *
  18  * Copyright (C) 2000 Intel
  19  * Copyright (C) Chuck Fleckenstein (cfleck@co.intel.com)
  20  *
  21  * Copyright (C) 1999, 2004 Silicon Graphics, Inc.
  22  * Copyright (C) Vijay Chander(vijay@engr.sgi.com)
  23  *
  24  * 03/04/15 D. Mosberger Added INIT backtrace support.
  25  * 02/03/25 M. Domsch   GUID cleanups
  26  *
  27  * 02/01/04 J. Hall     Aligned MCA stack to 16 bytes, added platform vs. CPU
  28  *                      error flag, set SAL default return values, changed
  29  *                      error record structure to linked list, added init call
  30  *                      to sal_get_state_info_size().
  31  *
  32  * 01/01/03 F. Lewis    Added setup of CMCI and CPEI IRQs, logging of corrected
  33  *                      platform errors, completed code for logging of
  34  *                      corrected & uncorrected machine check errors, and
  35  *                      updated for conformance with Nov. 2000 revision of the
  36  *                      SAL 3.0 spec.
  37  * 00/03/29 C. Fleckenstein  Fixed PAL/SAL update issues, began MCA bug fixes, logging issues,
  38  *                           added min save state dump, added INIT handler.
  39  *
  40  * 2003-12-08 Keith Owens <kaos@sgi.com>
  41  *            smp_call_function() must not be called from interrupt context (can
  42  *            deadlock on tasklist_lock).  Use keventd to call smp_call_function().
  43  *
  44  * 2004-02-01 Keith Owens <kaos@sgi.com>
  45  *            Avoid deadlock when using printk() for MCA and INIT records.
  46  *            Delete all record printing code, moved to salinfo_decode in user space.
  47  *            Mark variables and functions static where possible.
  48  *            Delete dead variables and functions.
  49  *            Reorder to remove the need for forward declarations and to consolidate
  50  *            related code.
  51  */
  52 #include <linux/config.h>
  53 #include <linux/types.h>
  54 #include <linux/init.h>
  55 #include <linux/sched.h>
  56 #include <linux/interrupt.h>
  57 #include <linux/irq.h>
  58 #include <linux/kallsyms.h>
  59 #include <linux/smp_lock.h>
  60 #include <linux/bootmem.h>
  61 #include <linux/acpi.h>
  62 #include <linux/timer.h>
  63 #include <linux/module.h>
  64 #include <linux/kernel.h>
  65 #include <linux/smp.h>
  66 #include <linux/workqueue.h>
  67
  68 #include <asm/delay.h>
  69 #include <asm/machvec.h>
  70 #include <asm/meminit.h>
  71 #include <asm/page.h>
  72 #include <asm/ptrace.h>
  73 #include <asm/system.h>
  74 #include <asm/sal.h>
  75 #include <asm/mca.h>
  76
  77 #include <asm/irq.h>
  78 #include <asm/hw_irq.h>
  79
  80 #if defined(IA64_MCA_DEBUG_INFO)
  81 # define IA64_MCA_DEBUG(fmt...) printk(fmt)
  82 #else
  83 # define IA64_MCA_DEBUG(fmt...)
  84 #endif
  85
  86 /* Used by mca_asm.S */
  87 ia64_mca_sal_to_os_state_t      ia64_sal_to_os_handoff_state;
  88 ia64_mca_os_to_sal_state_t      ia64_os_to_sal_handoff_state;
  89 u64                             ia64_mca_serialize;
  90 DEFINE_PER_CPU(u64, ia64_mca_data); /* == __per_cpu_mca[smp_processor_id()] */
  91 DEFINE_PER_CPU(u64, ia64_mca_per_cpu_pte); /* PTE to map per-CPU area */
  92 DEFINE_PER_CPU(u64, ia64_mca_pal_pte);      /* PTE to map PAL code */
  93 DEFINE_PER_CPU(u64, ia64_mca_pal_base);    /* vaddr PAL code granule */
  94
  95 unsigned long __per_cpu_mca[NR_CPUS];
  96
  97 /* In mca_asm.S */
  98 extern void                     ia64_monarch_init_handler (void);
  99 extern void                     ia64_slave_init_handler (void);
 100
 101 static ia64_mc_info_t           ia64_mc_info;
 102
 103 #define MAX_CPE_POLL_INTERVAL (15*60*HZ) /* 15 minutes */
 104 #define MIN_CPE_POLL_INTERVAL (2*60*HZ)  /* 2 minutes */
 105 #define CMC_POLL_INTERVAL     (1*60*HZ)  /* 1 minute */
 106 #define CPE_HISTORY_LENGTH    5
 107 #define CMC_HISTORY_LENGTH    5
 108
 109 static struct timer_list cpe_poll_timer;
 110 static struct timer_list cmc_poll_timer;
 111 /*
 112  * This variable tells whether we are currently in polling mode.
 113  * Start with this in the wrong state so we won't play w/ timers
 114  * before the system is ready.
 115  */
 116 static int cmc_polling_enabled = 1;
 117
 118 /*
 119  * Clearing this variable prevents CPE polling from getting activated
 120  * in mca_late_init.  Use it if your system doesn't provide a CPEI,
 121  * but encounters problems retrieving CPE logs.  This should only be
 122  * necessary for debugging.
 123  */
 124 static int cpe_poll_enabled = 1;
 125
 126 extern void salinfo_log_wakeup(int type, u8 *buffer, u64 size, int irqsafe);
 127
 128 static int mca_init;
 129
 130 /*
 131  * IA64_MCA log support
 132  */
 133 #define IA64_MAX_LOGS           2       /* Double-buffering for nested MCAs */
 134 #define IA64_MAX_LOG_TYPES      4   /* MCA, INIT, CMC, CPE */
 135
 136 typedef struct ia64_state_log_s
 137 {
 138         spinlock_t      isl_lock;
 139         int             isl_index;
 140         unsigned long   isl_count;
 141         ia64_err_rec_t  *isl_log[IA64_MAX_LOGS]; /* need space to store header + error log */
 142 } ia64_state_log_t;
 143
 144 static ia64_state_log_t ia64_state_log[IA64_MAX_LOG_TYPES];
 145
 146 #define IA64_LOG_ALLOCATE(it, size) \
 147         {ia64_state_log[it].isl_log[IA64_LOG_CURR_INDEX(it)] = \
 148                 (ia64_err_rec_t *)alloc_bootmem(size); \
 149         ia64_state_log[it].isl_log[IA64_LOG_NEXT_INDEX(it)] = \
 150                 (ia64_err_rec_t *)alloc_bootmem(size);}
 151 #define IA64_LOG_LOCK_INIT(it) spin_lock_init(&ia64_state_log[it].isl_lock)
 152 #define IA64_LOG_LOCK(it)      spin_lock_irqsave(&ia64_state_log[it].isl_lock, s)
 153 #define IA64_LOG_UNLOCK(it)    spin_unlock_irqrestore(&ia64_state_log[it].isl_lock,s)
 154 #define IA64_LOG_NEXT_INDEX(it)    ia64_state_log[it].isl_index
 155 #define IA64_LOG_CURR_INDEX(it)    1 - ia64_state_log[it].isl_index
 156 #define IA64_LOG_INDEX_INC(it) \
 157     {ia64_state_log[it].isl_index = 1 - ia64_state_log[it].isl_index; \
 158     ia64_state_log[it].isl_count++;}
 159 #define IA64_LOG_INDEX_DEC(it) \
 160     ia64_state_log[it].isl_index = 1 - ia64_state_log[it].isl_index
 161 #define IA64_LOG_NEXT_BUFFER(it)   (void *)((ia64_state_log[it].isl_log[IA64_LOG_NEXT_INDEX(it)]))
 162 #define IA64_LOG_CURR_BUFFER(it)   (void *)((ia64_state_log[it].isl_log[IA64_LOG_CURR_INDEX(it)]))
 163 #define IA64_LOG_COUNT(it)         ia64_state_log[it].isl_count
 164
 165 /*
 166  * ia64_log_init
 167  *      Reset the OS ia64 log buffer
 168  * Inputs   :   info_type   (SAL_INFO_TYPE_{MCA,INIT,CMC,CPE})
 169  * Outputs      :       None
 170  */
 171 static void
 172 ia64_log_init(int sal_info_type)
 173 {
 174         u64     max_size = 0;
 175
 176         IA64_LOG_NEXT_INDEX(sal_info_type) = 0;
 177         IA64_LOG_LOCK_INIT(sal_info_type);
 178
 179         // SAL will tell us the maximum size of any error record of this type
 180         max_size = ia64_sal_get_state_info_size(sal_info_type);
 181         if (!max_size)
 182                 /* alloc_bootmem() doesn't like zero-sized allocations! */
 183                 return;
 184
 185         // set up OS data structures to hold error info
 186         IA64_LOG_ALLOCATE(sal_info_type, max_size);
 187         memset(IA64_LOG_CURR_BUFFER(sal_info_type), 0, max_size);
 188         memset(IA64_LOG_NEXT_BUFFER(sal_info_type), 0, max_size);
 189 }
 190
 191 /*
 192  * ia64_log_get
 193  *
 194  *      Get the current MCA log from SAL and copy it into the OS log buffer.
 195  *
 196  *  Inputs  :   info_type   (SAL_INFO_TYPE_{MCA,INIT,CMC,CPE})
 197  *              irq_safe    whether you can use printk at this point
 198  *  Outputs :   size        (total record length)
 199  *              *buffer     (ptr to error record)
 200  *
 201  */
 202 static u64
 203 ia64_log_get(int sal_info_type, u8 **buffer, int irq_safe)
 204 {
 205         sal_log_record_header_t     *log_buffer;
 206         u64                         total_len = 0;
 207         int                         s;
 208
 209         IA64_LOG_LOCK(sal_info_type);
 210
 211         /* Get the process state information */
 212         log_buffer = IA64_LOG_NEXT_BUFFER(sal_info_type);
 213
 214         total_len = ia64_sal_get_state_info(sal_info_type, (u64 *)log_buffer);
 215
 216         if (total_len) {
 217                 IA64_LOG_INDEX_INC(sal_info_type);
 218                 IA64_LOG_UNLOCK(sal_info_type);
 219                 if (irq_safe) {
 220                         IA64_MCA_DEBUG("%s: SAL error record type %d retrieved. "
 221                                        "Record length = %ld\n", __FUNCTION__, sal_info_type, total_len);
 222                 }
 223                 *buffer = (u8 *) log_buffer;
 224                 return total_len;
 225         } else {
 226                 IA64_LOG_UNLOCK(sal_info_type);
 227                 return 0;
 228         }
 229 }
 230
 231 /*
 232  *  ia64_mca_log_sal_error_record
 233  *
 234  *  This function retrieves a specified error record type from SAL
 235  *  and wakes up any processes waiting for error records.
 236  *
 237  *  Inputs  :   sal_info_type   (Type of error record MCA/CMC/CPE/INIT)
 238  */
 239 static void
 240 ia64_mca_log_sal_error_record(int sal_info_type)
 241 {
 242         u8 *buffer;
 243         sal_log_record_header_t *rh;
 244         u64 size;
 245         int irq_safe = sal_info_type != SAL_INFO_TYPE_MCA && sal_info_type != SAL_INFO_TYPE_INIT;
 246 #ifdef IA64_MCA_DEBUG_INFO
 247         static const char * const rec_name[] = { "MCA", "INIT", "CMC", "CPE" };
 248 #endif
 249
 250         size = ia64_log_get(sal_info_type, &buffer, irq_safe);
 251         if (!size)
 252                 return;
 253
 254         salinfo_log_wakeup(sal_info_type, buffer, size, irq_safe);
 255
 256         if (irq_safe)
 257                 IA64_MCA_DEBUG("CPU %d: SAL log contains %s error record\n",
 258                         smp_processor_id(),
 259                         sal_info_type < ARRAY_SIZE(rec_name) ? rec_name[sal_info_type] : "UNKNOWN");
 260
 261         /* Clear logs from corrected errors in case there's no user-level logger */
 262         rh = (sal_log_record_header_t *)buffer;
 263         if (rh->severity == sal_log_severity_corrected)
 264                 ia64_sal_clear_state_info(sal_info_type);
 265 }
 266
 267 /*
 268  * platform dependent error handling
 269  */
 270 #ifndef PLATFORM_MCA_HANDLERS
 271
 272 #ifdef CONFIG_ACPI
 273
 274 int cpe_vector = -1;
 275
 276 static irqreturn_t
 277 ia64_mca_cpe_int_handler (int cpe_irq, void *arg, struct pt_regs *ptregs)
 278 {
 279         static unsigned long    cpe_history[CPE_HISTORY_LENGTH];
 280         static int              index;
 281         static DEFINE_SPINLOCK(cpe_history_lock);
 282
 283         IA64_MCA_DEBUG("%s: received interrupt vector = %#x on CPU %d\n",
 284                        __FUNCTION__, cpe_irq, smp_processor_id());
 285
 286         /* SAL spec states this should run w/ interrupts enabled */
 287         local_irq_enable();
 288
 289         /* Get the CPE error record and log it */
 290         ia64_mca_log_sal_error_record(SAL_INFO_TYPE_CPE);
 291
 292         spin_lock(&cpe_history_lock);
 293         if (!cpe_poll_enabled && cpe_vector >= 0) {
 294
 295                 int i, count = 1; /* we know 1 happened now */
 296                 unsigned long now = jiffies;
 297
 298                 for (i = 0; i < CPE_HISTORY_LENGTH; i++) {
 299                         if (now - cpe_history[i] <= HZ)
 300                                 count++;
 301                 }
 302
 303                 IA64_MCA_DEBUG(KERN_INFO "CPE threshold %d/%d\n", count, CPE_HISTORY_LENGTH);
 304                 if (count >= CPE_HISTORY_LENGTH) {
 305
 306                         cpe_poll_enabled = 1;
 307                         spin_unlock(&cpe_history_lock);
 308                         disable_irq_nosync(local_vector_to_irq(IA64_CPE_VECTOR));
 309
 310                         /*
 311                          * Corrected errors will still be corrected, but
 312                          * make sure there's a log somewhere that indicates
 313                          * something is generating more than we can handle.
 314                          */
 315                         printk(KERN_WARNING "WARNING: Switching to polling CPE handler; error records may be lost\n");
 316
 317                         mod_timer(&cpe_poll_timer, jiffies + MIN_CPE_POLL_INTERVAL);
 318
 319                         /* lock already released, get out now */
 320                         return IRQ_HANDLED;
 321                 } else {
 322                         cpe_history[index++] = now;
 323                         if (index == CPE_HISTORY_LENGTH)
 324                                 index = 0;
 325                 }
 326         }
 327         spin_unlock(&cpe_history_lock);
 328         return IRQ_HANDLED;
 329 }
 330
 331 #endif /* CONFIG_ACPI */
 332
 333 static void
 334 show_min_state (pal_min_state_area_t *minstate)
 335 {
 336         u64 iip = minstate->pmsa_iip + ((struct ia64_psr *)(&minstate->pmsa_ipsr))->ri;
 337         u64 xip = minstate->pmsa_xip + ((struct ia64_psr *)(&minstate->pmsa_xpsr))->ri;
 338
 339         printk("NaT bits\t%016lx\n", minstate->pmsa_nat_bits);
 340         printk("pr\t\t%016lx\n", minstate->pmsa_pr);
 341         printk("b0\t\t%016lx ", minstate->pmsa_br0); print_symbol("%s\n", minstate->pmsa_br0);
 342         printk("ar.rsc\t\t%016lx\n", minstate->pmsa_rsc);
 343         printk("cr.iip\t\t%016lx ", iip); print_symbol("%s\n", iip);
 344         printk("cr.ipsr\t\t%016lx\n", minstate->pmsa_ipsr);
 345         printk("cr.ifs\t\t%016lx\n", minstate->pmsa_ifs);
 346         printk("xip\t\t%016lx ", xip); print_symbol("%s\n", xip);
 347         printk("xpsr\t\t%016lx\n", minstate->pmsa_xpsr);
 348         printk("xfs\t\t%016lx\n", minstate->pmsa_xfs);
 349         printk("b1\t\t%016lx ", minstate->pmsa_br1);
 350         print_symbol("%s\n", minstate->pmsa_br1);
 351
 352         printk("\nstatic registers r0-r15:\n");
 353         printk(" r0- 3 %016lx %016lx %016lx %016lx\n",
 354                0UL, minstate->pmsa_gr[0], minstate->pmsa_gr[1], minstate->pmsa_gr[2]);
 355         printk(" r4- 7 %016lx %016lx %016lx %016lx\n",
 356                minstate->pmsa_gr[3], minstate->pmsa_gr[4],
 357                minstate->pmsa_gr[5], minstate->pmsa_gr[6]);
 358         printk(" r8-11 %016lx %016lx %016lx %016lx\n",
 359                minstate->pmsa_gr[7], minstate->pmsa_gr[8],
 360                minstate->pmsa_gr[9], minstate->pmsa_gr[10]);
 361         printk("r12-15 %016lx %016lx %016lx %016lx\n",
 362                minstate->pmsa_gr[11], minstate->pmsa_gr[12],
 363                minstate->pmsa_gr[13], minstate->pmsa_gr[14]);
 364
 365         printk("\nbank 0:\n");
 366         printk("r16-19 %016lx %016lx %016lx %016lx\n",
 367                minstate->pmsa_bank0_gr[0], minstate->pmsa_bank0_gr[1],
 368                minstate->pmsa_bank0_gr[2], minstate->pmsa_bank0_gr[3]);
 369         printk("r20-23 %016lx %016lx %016lx %016lx\n",
 370                minstate->pmsa_bank0_gr[4], minstate->pmsa_bank0_gr[5],
 371                minstate->pmsa_bank0_gr[6], minstate->pmsa_bank0_gr[7]);
 372         printk("r24-27 %016lx %016lx %016lx %016lx\n",
 373                minstate->pmsa_bank0_gr[8], minstate->pmsa_bank0_gr[9],
 374                minstate->pmsa_bank0_gr[10], minstate->pmsa_bank0_gr[11]);
 375         printk("r28-31 %016lx %016lx %016lx %016lx\n",
 376                minstate->pmsa_bank0_gr[12], minstate->pmsa_bank0_gr[13],
 377                minstate->pmsa_bank0_gr[14], minstate->pmsa_bank0_gr[15]);
 378
 379         printk("\nbank 1:\n");
 380         printk("r16-19 %016lx %016lx %016lx %016lx\n",
 381                minstate->pmsa_bank1_gr[0], minstate->pmsa_bank1_gr[1],
 382                minstate->pmsa_bank1_gr[2], minstate->pmsa_bank1_gr[3]);
 383         printk("r20-23 %016lx %016lx %016lx %016lx\n",
 384                minstate->pmsa_bank1_gr[4], minstate->pmsa_bank1_gr[5],
 385                minstate->pmsa_bank1_gr[6], minstate->pmsa_bank1_gr[7]);
 386         printk("r24-27 %016lx %016lx %016lx %016lx\n",
 387                minstate->pmsa_bank1_gr[8], minstate->pmsa_bank1_gr[9],
 388                minstate->pmsa_bank1_gr[10], minstate->pmsa_bank1_gr[11]);
 389         printk("r28-31 %016lx %016lx %016lx %016lx\n",
 390                minstate->pmsa_bank1_gr[12], minstate->pmsa_bank1_gr[13],
 391                minstate->pmsa_bank1_gr[14], minstate->pmsa_bank1_gr[15]);
 392 }
 393
 394 static void
 395 fetch_min_state (pal_min_state_area_t *ms, struct pt_regs *pt, struct switch_stack *sw)
 396 {
 397         u64 *dst_banked, *src_banked, bit, shift, nat_bits;
 398         int i;
 399
 400         /*
 401          * First, update the pt-regs and switch-stack structures with the contents stored
 402          * in the min-state area:
 403          */
 404         if (((struct ia64_psr *) &ms->pmsa_ipsr)->ic == 0) {
 405                 pt->cr_ipsr = ms->pmsa_xpsr;
 406                 pt->cr_iip = ms->pmsa_xip;
 407                 pt->cr_ifs = ms->pmsa_xfs;
 408         } else {
 409                 pt->cr_ipsr = ms->pmsa_ipsr;
 410                 pt->cr_iip = ms->pmsa_iip;
 411                 pt->cr_ifs = ms->pmsa_ifs;
 412         }
 413         pt->ar_rsc = ms->pmsa_rsc;
 414         pt->pr = ms->pmsa_pr;
 415         pt->r1 = ms->pmsa_gr[0];
 416         pt->r2 = ms->pmsa_gr[1];
 417         pt->r3 = ms->pmsa_gr[2];
 418         sw->r4 = ms->pmsa_gr[3];
 419         sw->r5 = ms->pmsa_gr[4];
 420         sw->r6 = ms->pmsa_gr[5];
 421         sw->r7 = ms->pmsa_gr[6];
 422         pt->r8 = ms->pmsa_gr[7];
 423         pt->r9 = ms->pmsa_gr[8];
 424         pt->r10 = ms->pmsa_gr[9];
 425         pt->r11 = ms->pmsa_gr[10];
 426         pt->r12 = ms->pmsa_gr[11];
 427         pt->r13 = ms->pmsa_gr[12];
 428         pt->r14 = ms->pmsa_gr[13];
 429         pt->r15 = ms->pmsa_gr[14];
 430         dst_banked = &pt->r16;          /* r16-r31 are contiguous in struct pt_regs */
 431         src_banked = ms->pmsa_bank1_gr;
 432         for (i = 0; i < 16; ++i)
 433                 dst_banked[i] = src_banked[i];
 434         pt->b0 = ms->pmsa_br0;
 435         sw->b1 = ms->pmsa_br1;
 436
 437         /* construct the NaT bits for the pt-regs structure: */
 438 #       define PUT_NAT_BIT(dst, addr)                                   \
 439         do {                                                            \
 440                 bit = nat_bits & 1; nat_bits >>= 1;                     \
 441                 shift = ((unsigned long) addr >> 3) & 0x3f;             \
 442                 dst = ((dst) & ~(1UL << shift)) | (bit << shift);       \
 443         } while (0)
 444
 445         /* Rotate the saved NaT bits such that bit 0 corresponds to pmsa_gr[0]: */
 446         shift = ((unsigned long) &ms->pmsa_gr[0] >> 3) & 0x3f;
 447         nat_bits = (ms->pmsa_nat_bits >> shift) | (ms->pmsa_nat_bits << (64 - shift));
 448
 449         PUT_NAT_BIT(sw->caller_unat, &pt->r1);
 450         PUT_NAT_BIT(sw->caller_unat, &pt->r2);
 451         PUT_NAT_BIT(sw->caller_unat, &pt->r3);
 452         PUT_NAT_BIT(sw->ar_unat, &sw->r4);
 453         PUT_NAT_BIT(sw->ar_unat, &sw->r5);
 454         PUT_NAT_BIT(sw->ar_unat, &sw->r6);
 455         PUT_NAT_BIT(sw->ar_unat, &sw->r7);
 456         PUT_NAT_BIT(sw->caller_unat, &pt->r8);  PUT_NAT_BIT(sw->caller_unat, &pt->r9);
 457         PUT_NAT_BIT(sw->caller_unat, &pt->r10); PUT_NAT_BIT(sw->caller_unat, &pt->r11);
 458         PUT_NAT_BIT(sw->caller_unat, &pt->r12); PUT_NAT_BIT(sw->caller_unat, &pt->r13);
 459         PUT_NAT_BIT(sw->caller_unat, &pt->r14); PUT_NAT_BIT(sw->caller_unat, &pt->r15);
 460         nat_bits >>= 16;        /* skip over bank0 NaT bits */
 461         PUT_NAT_BIT(sw->caller_unat, &pt->r16); PUT_NAT_BIT(sw->caller_unat, &pt->r17);
 462         PUT_NAT_BIT(sw->caller_unat, &pt->r18); PUT_NAT_BIT(sw->caller_unat, &pt->r19);
 463         PUT_NAT_BIT(sw->caller_unat, &pt->r20); PUT_NAT_BIT(sw->caller_unat, &pt->r21);
 464         PUT_NAT_BIT(sw->caller_unat, &pt->r22); PUT_NAT_BIT(sw->caller_unat, &pt->r23);
 465         PUT_NAT_BIT(sw->caller_unat, &pt->r24); PUT_NAT_BIT(sw->caller_unat, &pt->r25);
 466         PUT_NAT_BIT(sw->caller_unat, &pt->r26); PUT_NAT_BIT(sw->caller_unat, &pt->r27);
 467         PUT_NAT_BIT(sw->caller_unat, &pt->r28); PUT_NAT_BIT(sw->caller_unat, &pt->r29);
 468         PUT_NAT_BIT(sw->caller_unat, &pt->r30); PUT_NAT_BIT(sw->caller_unat, &pt->r31);
 469 }
 470
 471 static void
 472 init_handler_platform (pal_min_state_area_t *ms,
 473                        struct pt_regs *pt, struct switch_stack *sw)
 474 {
 475         struct unw_frame_info info;
 476
 477         /* if a kernel debugger is available call it here else just dump the registers */
 478
 479         /*
 480          * Wait for a bit.  On some machines (e.g., HP's zx2000 and zx6000, INIT can be
 481          * generated via the BMC's command-line interface, but since the console is on the
 482          * same serial line, the user will need some time to switch out of the BMC before
 483          * the dump begins.
 484          */
 485         printk("Delaying for 5 seconds...\n");
 486         udelay(5*1000000);
 487         show_min_state(ms);
 488
 489         printk("Backtrace of current task (pid %d, %s)\n", current->pid, current->comm);
 490         fetch_min_state(ms, pt, sw);
 491         unw_init_from_interruption(&info, current, pt, sw);
 492         ia64_do_show_stack(&info, NULL);
 493
 494 #ifdef CONFIG_SMP
 495         /* read_trylock() would be handy... */
 496         if (!tasklist_lock.write_lock)
 497                 read_lock(&tasklist_lock);
 498 #endif
 499         {
 500                 struct task_struct *g, *t;
 501                 do_each_thread (g, t) {
 502                         if (t == current)
 503                                 continue;
 504
 505                         printk("\nBacktrace of pid %d (%s)\n", t->pid, t->comm);
 506                         show_stack(t, NULL);
 507                 } while_each_thread (g, t);
 508         }
 509 #ifdef CONFIG_SMP
 510         if (!tasklist_lock.write_lock)
 511                 read_unlock(&tasklist_lock);
 512 #endif
 513
 514         printk("\nINIT dump complete.  Please reboot now.\n");
 515         while (1);                      /* hang city if no debugger */
 516 }
 517
 518 #ifdef CONFIG_ACPI
 519 /*
 520  * ia64_mca_register_cpev
 521  *
 522  *  Register the corrected platform error vector with SAL.
 523  *
 524  *  Inputs
 525  *      cpev        Corrected Platform Error Vector number
 526  *
 527  *  Outputs
 528  *      None
 529  */
 530 static void
 531 ia64_mca_register_cpev (int cpev)
 532 {
 533         /* Register the CPE interrupt vector with SAL */
 534         struct ia64_sal_retval isrv;
 535
 536         isrv = ia64_sal_mc_set_params(SAL_MC_PARAM_CPE_INT, SAL_MC_PARAM_MECHANISM_INT, cpev, 0, 0);
 537         if (isrv.status) {
 538                 printk(KERN_ERR "Failed to register Corrected Platform "
 539                        "Error interrupt vector with SAL (status %ld)\n", isrv.status);
 540                 return;
 541         }
 542
 543         IA64_MCA_DEBUG("%s: corrected platform error "
 544                        "vector %#x registered\n", __FUNCTION__, cpev);
 545 }
 546 #endif /* CONFIG_ACPI */
 547
 548 #endif /* PLATFORM_MCA_HANDLERS */
 549
 550 /*
 551  * ia64_mca_cmc_vector_setup
 552  *
 553  *  Setup the corrected machine check vector register in the processor.
 554  *  (The interrupt is masked on boot. ia64_mca_late_init unmask this.)
 555  *  This function is invoked on a per-processor basis.
 556  *
 557  * Inputs
 558  *      None
 559  *
 560  * Outputs
 561  *      None
 562  */
 563 void
 564 ia64_mca_cmc_vector_setup (void)
 565 {
 566         cmcv_reg_t      cmcv;
 567
 568         cmcv.cmcv_regval        = 0;
 569         cmcv.cmcv_mask          = 1;        /* Mask/disable interrupt at first */
 570         cmcv.cmcv_vector        = IA64_CMC_VECTOR;
 571         ia64_setreg(_IA64_REG_CR_CMCV, cmcv.cmcv_regval);
 572
 573         IA64_MCA_DEBUG("%s: CPU %d corrected "
 574                        "machine check vector %#x registered.\n",
 575                        __FUNCTION__, smp_processor_id(), IA64_CMC_VECTOR);
 576
 577         IA64_MCA_DEBUG("%s: CPU %d CMCV = %#016lx\n",
 578                        __FUNCTION__, smp_processor_id(), ia64_getreg(_IA64_REG_CR_CMCV));
 579 }
 580
 581 /*
 582  * ia64_mca_cmc_vector_disable
 583  *
 584  *  Mask the corrected machine check vector register in the processor.
 585  *  This function is invoked on a per-processor basis.
 586  *
 587  * Inputs
 588  *      dummy(unused)
 589  *
 590  * Outputs
 591  *      None
 592  */
 593 static void
 594 ia64_mca_cmc_vector_disable (void *dummy)
 595 {
 596         cmcv_reg_t      cmcv;
 597
 598         cmcv.cmcv_regval = ia64_getreg(_IA64_REG_CR_CMCV);
 599
 600         cmcv.cmcv_mask = 1; /* Mask/disable interrupt */
 601         ia64_setreg(_IA64_REG_CR_CMCV, cmcv.cmcv_regval);
 602
 603         IA64_MCA_DEBUG("%s: CPU %d corrected "
 604                        "machine check vector %#x disabled.\n",
 605                        __FUNCTION__, smp_processor_id(), cmcv.cmcv_vector);
 606 }
 607
 608 /*
 609  * ia64_mca_cmc_vector_enable
 610  *
 611  *  Unmask the corrected machine check vector register in the processor.
 612  *  This function is invoked on a per-processor basis.
 613  *
 614  * Inputs
 615  *      dummy(unused)
 616  *
 617  * Outputs
 618  *      None
 619  */
 620 static void
 621 ia64_mca_cmc_vector_enable (void *dummy)
 622 {
 623         cmcv_reg_t      cmcv;
 624
 625         cmcv.cmcv_regval = ia64_getreg(_IA64_REG_CR_CMCV);
 626
 627         cmcv.cmcv_mask = 0; /* Unmask/enable interrupt */
 628         ia64_setreg(_IA64_REG_CR_CMCV, cmcv.cmcv_regval);
 629
 630         IA64_MCA_DEBUG("%s: CPU %d corrected "
 631                        "machine check vector %#x enabled.\n",
 632                        __FUNCTION__, smp_processor_id(), cmcv.cmcv_vector);
 633 }
 634
 635 /*
 636  * ia64_mca_cmc_vector_disable_keventd
 637  *
 638  * Called via keventd (smp_call_function() is not safe in interrupt context) to
 639  * disable the cmc interrupt vector.
 640  */
 641 static void
 642 ia64_mca_cmc_vector_disable_keventd(void *unused)
 643 {
 644         on_each_cpu(ia64_mca_cmc_vector_disable, NULL, 1, 0);
 645 }
 646
 647 /*
 648  * ia64_mca_cmc_vector_enable_keventd
 649  *
 650  * Called via keventd (smp_call_function() is not safe in interrupt context) to
 651  * enable the cmc interrupt vector.
 652  */
 653 static void
 654 ia64_mca_cmc_vector_enable_keventd(void *unused)
 655 {
 656         on_each_cpu(ia64_mca_cmc_vector_enable, NULL, 1, 0);
 657 }
 658
 659 /*
 660  * ia64_mca_wakeup_ipi_wait
 661  *
 662  *      Wait for the inter-cpu interrupt to be sent by the
 663  *      monarch processor once it is done with handling the
 664  *      MCA.
 665  *
 666  *  Inputs  :   None
 667  *  Outputs :   None
 668  */
 669 static void
 670 ia64_mca_wakeup_ipi_wait(void)
 671 {
 672         int     irr_num = (IA64_MCA_WAKEUP_VECTOR >> 6);
 673         int     irr_bit = (IA64_MCA_WAKEUP_VECTOR & 0x3f);
 674         u64     irr = 0;
 675
 676         do {
 677                 switch(irr_num) {
 678                       case 0:
 679                         irr = ia64_getreg(_IA64_REG_CR_IRR0);
 680                         break;
 681                       case 1:
 682                         irr = ia64_getreg(_IA64_REG_CR_IRR1);
 683                         break;
 684                       case 2:
 685                         irr = ia64_getreg(_IA64_REG_CR_IRR2);
 686                         break;
 687                       case 3:
 688                         irr = ia64_getreg(_IA64_REG_CR_IRR3);
 689                         break;
 690                 }
 691                 cpu_relax();
 692         } while (!(irr & (1UL << irr_bit))) ;
 693 }
 694
 695 /*
 696  * ia64_mca_wakeup
 697  *
 698  *      Send an inter-cpu interrupt to wake-up a particular cpu
 699  *      and mark that cpu to be out of rendez.
 700  *
 701  *  Inputs  :   cpuid
 702  *  Outputs :   None
 703  */
 704 static void
 705 ia64_mca_wakeup(int cpu)
 706 {
 707         platform_send_ipi(cpu, IA64_MCA_WAKEUP_VECTOR, IA64_IPI_DM_INT, 0);
 708         ia64_mc_info.imi_rendez_checkin[cpu] = IA64_MCA_RENDEZ_CHECKIN_NOTDONE;
 709
 710 }
 711
 712 /*
 713  * ia64_mca_wakeup_all
 714  *
 715  *      Wakeup all the cpus which have rendez'ed previously.
 716  *
 717  *  Inputs  :   None
 718  *  Outputs :   None
 719  */
 720 static void
 721 ia64_mca_wakeup_all(void)
 722 {
 723         int cpu;
 724
 725         /* Clear the Rendez checkin flag for all cpus */
 726         for(cpu = 0; cpu < NR_CPUS; cpu++) {
 727                 if (!cpu_online(cpu))
 728                         continue;
 729                 if (ia64_mc_info.imi_rendez_checkin[cpu] == IA64_MCA_RENDEZ_CHECKIN_DONE)
 730                         ia64_mca_wakeup(cpu);
 731         }
 732
 733 }
 734
 735 /*
 736  * ia64_mca_rendez_interrupt_handler
 737  *
 738  *      This is handler used to put slave processors into spinloop
 739  *      while the monarch processor does the mca handling and later
 740  *      wake each slave up once the monarch is done.
 741  *
 742  *  Inputs  :   None
 743  *  Outputs :   None
 744  */
 745 static irqreturn_t
 746 ia64_mca_rendez_int_handler(int rendez_irq, void *arg, struct pt_regs *ptregs)
 747 {
 748         unsigned long flags;
 749         int cpu = smp_processor_id();
 750
 751         /* Mask all interrupts */
 752         local_irq_save(flags);
 753
 754         ia64_mc_info.imi_rendez_checkin[cpu] = IA64_MCA_RENDEZ_CHECKIN_DONE;
 755         /* Register with the SAL monarch that the slave has
 756          * reached SAL
 757          */
 758         ia64_sal_mc_rendez();
 759
 760         /* Wait for the wakeup IPI from the monarch
 761          * This waiting is done by polling on the wakeup-interrupt
 762          * vector bit in the processor's IRRs
 763          */
 764         ia64_mca_wakeup_ipi_wait();
 765
 766         /* Enable all interrupts */
 767         local_irq_restore(flags);
 768         return IRQ_HANDLED;
 769 }
 770
 771 /*
 772  * ia64_mca_wakeup_int_handler
 773  *
 774  *      The interrupt handler for processing the inter-cpu interrupt to the
 775  *      slave cpu which was spinning in the rendez loop.
 776  *      Since this spinning is done by turning off the interrupts and
 777  *      polling on the wakeup-interrupt bit in the IRR, there is
 778  *      nothing useful to be done in the handler.
 779  *
 780  *  Inputs  :   wakeup_irq  (Wakeup-interrupt bit)
 781  *      arg             (Interrupt handler specific argument)
 782  *      ptregs          (Exception frame at the time of the interrupt)
 783  *  Outputs :   None
 784  *
 785  */
 786 static irqreturn_t
 787 ia64_mca_wakeup_int_handler(int wakeup_irq, void *arg, struct pt_regs *ptregs)
 788 {
 789         return IRQ_HANDLED;
 790 }
 791
 792 /*
 793  * ia64_return_to_sal_check
 794  *
 795  *      This is function called before going back from the OS_MCA handler
 796  *      to the OS_MCA dispatch code which finally takes the control back
 797  *      to the SAL.
 798  *      The main purpose of this routine is to setup the OS_MCA to SAL
 799  *      return state which can be used by the OS_MCA dispatch code
 800  *      just before going back to SAL.
 801  *
 802  *  Inputs  :   None
 803  *  Outputs :   None
 804  */
 805
 806 static void
 807 ia64_return_to_sal_check(int recover)
 808 {
 809
 810         /* Copy over some relevant stuff from the sal_to_os_mca_handoff
 811          * so that it can be used at the time of os_mca_to_sal_handoff
 812          */
 813         ia64_os_to_sal_handoff_state.imots_sal_gp =
 814                 ia64_sal_to_os_handoff_state.imsto_sal_gp;
 815
 816         ia64_os_to_sal_handoff_state.imots_sal_check_ra =
 817                 ia64_sal_to_os_handoff_state.imsto_sal_check_ra;
 818
 819         if (recover)
 820                 ia64_os_to_sal_handoff_state.imots_os_status = IA64_MCA_CORRECTED;
 821         else
 822                 ia64_os_to_sal_handoff_state.imots_os_status = IA64_MCA_COLD_BOOT;
 823
 824         /* Default = tell SAL to return to same context */
 825         ia64_os_to_sal_handoff_state.imots_context = IA64_MCA_SAME_CONTEXT;
 826
 827         ia64_os_to_sal_handoff_state.imots_new_min_state =
 828                 (u64 *)ia64_sal_to_os_handoff_state.pal_min_state;
 829
 830 }
 831
 832 /* Function pointer for extra MCA recovery */
 833 int (*ia64_mca_ucmc_extension)
 834         (void*,ia64_mca_sal_to_os_state_t*,ia64_mca_os_to_sal_state_t*)
 835         = NULL;
 836
 837 int
 838 ia64_reg_MCA_extension(void *fn)
 839 {
 840         if (ia64_mca_ucmc_extension)
 841                 return 1;
 842
 843         ia64_mca_ucmc_extension = fn;
 844         return 0;
 845 }
 846
 847 void
 848 ia64_unreg_MCA_extension(void)
 849 {
 850         if (ia64_mca_ucmc_extension)
 851                 ia64_mca_ucmc_extension = NULL;
 852 }
 853
 854 EXPORT_SYMBOL(ia64_reg_MCA_extension);
 855 EXPORT_SYMBOL(ia64_unreg_MCA_extension);
 856
 857 /*
 858  * ia64_mca_ucmc_handler
 859  *
 860  *      This is uncorrectable machine check handler called from OS_MCA
 861  *      dispatch code which is in turn called from SAL_CHECK().
 862  *      This is the place where the core of OS MCA handling is done.
 863  *      Right now the logs are extracted and displayed in a well-defined
 864  *      format. This handler code is supposed to be run only on the
 865  *      monarch processor. Once the monarch is done with MCA handling
 866  *      further MCA logging is enabled by clearing logs.
 867  *      Monarch also has the duty of sending wakeup-IPIs to pull the
 868  *      slave processors out of rendezvous spinloop.
 869  *
 870  *  Inputs  :   None
 871  *  Outputs :   None
 872  */
 873 void
 874 ia64_mca_ucmc_handler(void)
 875 {
 876         pal_processor_state_info_t *psp = (pal_processor_state_info_t *)
 877                 &ia64_sal_to_os_handoff_state.proc_state_param;
 878         int recover;
 879
 880         /* Get the MCA error record and log it */
 881         ia64_mca_log_sal_error_record(SAL_INFO_TYPE_MCA);
 882
 883         /* TLB error is only exist in this SAL error record */
 884         recover = (psp->tc && !(psp->cc || psp->bc || psp->rc || psp->uc))
 885         /* other error recovery */
 886            || (ia64_mca_ucmc_extension
 887                 && ia64_mca_ucmc_extension(
 888                         IA64_LOG_CURR_BUFFER(SAL_INFO_TYPE_MCA),
 889                         &ia64_sal_to_os_handoff_state,
 890                         &ia64_os_to_sal_handoff_state));
 891
 892         if (recover) {
 893                 sal_log_record_header_t *rh = IA64_LOG_CURR_BUFFER(SAL_INFO_TYPE_MCA);
 894                 rh->severity = sal_log_severity_corrected;
 895                 ia64_sal_clear_state_info(SAL_INFO_TYPE_MCA);
 896         }
 897         /*
 898          *  Wakeup all the processors which are spinning in the rendezvous
 899          *  loop.
 900          */
 901         ia64_mca_wakeup_all();
 902
 903         /* Return to SAL */
 904         ia64_return_to_sal_check(recover);
 905 }
 906
 907 static DECLARE_WORK(cmc_disable_work, ia64_mca_cmc_vector_disable_keventd, NULL);
 908 static DECLARE_WORK(cmc_enable_work, ia64_mca_cmc_vector_enable_keventd, NULL);
 909
 910 /*
 911  * ia64_mca_cmc_int_handler
 912  *
 913  *  This is corrected machine check interrupt handler.
 914  *      Right now the logs are extracted and displayed in a well-defined
 915  *      format.
 916  *
 917  * Inputs
 918  *      interrupt number
 919  *      client data arg ptr
 920  *      saved registers ptr
 921  *
 922  * Outputs
 923  *      None
 924  */
 925 static irqreturn_t
 926 ia64_mca_cmc_int_handler(int cmc_irq, void *arg, struct pt_regs *ptregs)
 927 {
 928         static unsigned long    cmc_history[CMC_HISTORY_LENGTH];
 929         static int              index;
 930         static DEFINE_SPINLOCK(cmc_history_lock);
 931
 932         IA64_MCA_DEBUG("%s: received interrupt vector = %#x on CPU %d\n",
 933                        __FUNCTION__, cmc_irq, smp_processor_id());
 934
 935         /* SAL spec states this should run w/ interrupts enabled */
 936         local_irq_enable();
 937
 938         /* Get the CMC error record and log it */
 939         ia64_mca_log_sal_error_record(SAL_INFO_TYPE_CMC);
 940
 941         spin_lock(&cmc_history_lock);
 942         if (!cmc_polling_enabled) {
 943                 int i, count = 1; /* we know 1 happened now */
 944                 unsigned long now = jiffies;
 945
 946                 for (i = 0; i < CMC_HISTORY_LENGTH; i++) {
 947                         if (now - cmc_history[i] <= HZ)
 948                                 count++;
 949                 }
 950
 951                 IA64_MCA_DEBUG(KERN_INFO "CMC threshold %d/%d\n", count, CMC_HISTORY_LENGTH);
 952                 if (count >= CMC_HISTORY_LENGTH) {
 953
 954                         cmc_polling_enabled = 1;
 955                         spin_unlock(&cmc_history_lock);
 956                         schedule_work(&cmc_disable_work);
 957
 958                         /*
 959                          * Corrected errors will still be corrected, but
 960                          * make sure there's a log somewhere that indicates
 961                          * something is generating more than we can handle.
 962                          */
 963                         printk(KERN_WARNING "WARNING: Switching to polling CMC handler; error records may be lost\n");
 964
 965                         mod_timer(&cmc_poll_timer, jiffies + CMC_POLL_INTERVAL);
 966
 967                         /* lock already released, get out now */
 968                         return IRQ_HANDLED;
 969                 } else {
 970                         cmc_history[index++] = now;
 971                         if (index == CMC_HISTORY_LENGTH)
 972                                 index = 0;
 973                 }
 974         }
 975         spin_unlock(&cmc_history_lock);
 976         return IRQ_HANDLED;
 977 }
 978
 979 /*
 980  *  ia64_mca_cmc_int_caller
 981  *
 982  *      Triggered by sw interrupt from CMC polling routine.  Calls
 983  *      real interrupt handler and either triggers a sw interrupt
 984  *      on the next cpu or does cleanup at the end.
 985  *
 986  * Inputs
 987  *      interrupt number
 988  *      client data arg ptr
 989  *      saved registers ptr
 990  * Outputs
 991  *      handled
 992  */
 993 static irqreturn_t
 994 ia64_mca_cmc_int_caller(int cmc_irq, void *arg, struct pt_regs *ptregs)
 995 {
 996         static int start_count = -1;
 997         unsigned int cpuid;
 998
 999         cpuid = smp_processor_id();
1000
1001         /* If first cpu, update count */
1002         if (start_count == -1)
1003                 start_count = IA64_LOG_COUNT(SAL_INFO_TYPE_CMC);
1004
1005         ia64_mca_cmc_int_handler(cmc_irq, arg, ptregs);
1006
1007         for (++cpuid ; cpuid < NR_CPUS && !cpu_online(cpuid) ; cpuid++);
1008
1009         if (cpuid < NR_CPUS) {
1010                 platform_send_ipi(cpuid, IA64_CMCP_VECTOR, IA64_IPI_DM_INT, 0);
1011         } else {
1012                 /* If no log record, switch out of polling mode */
1013                 if (start_count == IA64_LOG_COUNT(SAL_INFO_TYPE_CMC)) {
1014
1015                         printk(KERN_WARNING "Returning to interrupt driven CMC handler\n");
1016                         schedule_work(&cmc_enable_work);
1017                         cmc_polling_enabled = 0;
1018
1019                 } else {
1020
1021                         mod_timer(&cmc_poll_timer, jiffies + CMC_POLL_INTERVAL);
1022                 }
1023
1024                 start_count = -1;
1025         }
1026
1027         return IRQ_HANDLED;
1028 }
1029
1030 /*
1031  *  ia64_mca_cmc_poll
1032  *
1033  *      Poll for Corrected Machine Checks (CMCs)
1034  *
1035  * Inputs   :   dummy(unused)
1036  * Outputs  :   None
1037  *
1038  */
1039 static void
1040 ia64_mca_cmc_poll (unsigned long dummy)
1041 {
1042         /* Trigger a CMC interrupt cascade  */
1043         platform_send_ipi(first_cpu(cpu_online_map), IA64_CMCP_VECTOR, IA64_IPI_DM_INT, 0);
1044 }
1045
1046 /*
1047  *  ia64_mca_cpe_int_caller
1048  *
1049  *      Triggered by sw interrupt from CPE polling routine.  Calls
1050  *      real interrupt handler and either triggers a sw interrupt
1051  *      on the next cpu or does cleanup at the end.
1052  *
1053  * Inputs
1054  *      interrupt number
1055  *      client data arg ptr
1056  *      saved registers ptr
1057  * Outputs
1058  *      handled
1059  */
1060 #ifdef CONFIG_ACPI
1061
1062 static irqreturn_t
1063 ia64_mca_cpe_int_caller(int cpe_irq, void *arg, struct pt_regs *ptregs)
1064 {
1065         static int start_count = -1;
1066         static int poll_time = MIN_CPE_POLL_INTERVAL;
1067         unsigned int cpuid;
1068
1069         cpuid = smp_processor_id();
1070
1071         /* If first cpu, update count */
1072         if (start_count == -1)
1073                 start_count = IA64_LOG_COUNT(SAL_INFO_TYPE_CPE);
1074
1075         ia64_mca_cpe_int_handler(cpe_irq, arg, ptregs);
1076
1077         for (++cpuid ; cpuid < NR_CPUS && !cpu_online(cpuid) ; cpuid++);
1078
1079         if (cpuid < NR_CPUS) {
1080                 platform_send_ipi(cpuid, IA64_CPEP_VECTOR, IA64_IPI_DM_INT, 0);
1081         } else {
1082                 /*
1083                  * If a log was recorded, increase our polling frequency,
1084                  * otherwise, backoff or return to interrupt mode.
1085                  */
1086                 if (start_count != IA64_LOG_COUNT(SAL_INFO_TYPE_CPE)) {
1087                         poll_time = max(MIN_CPE_POLL_INTERVAL, poll_time / 2);
1088                 } else if (cpe_vector < 0) {
1089                         poll_time = min(MAX_CPE_POLL_INTERVAL, poll_time * 2);
1090                 } else {
1091                         poll_time = MIN_CPE_POLL_INTERVAL;
1092
1093                         printk(KERN_WARNING "Returning to interrupt driven CPE handler\n");
1094                         enable_irq(local_vector_to_irq(IA64_CPE_VECTOR));
1095                         cpe_poll_enabled = 0;
1096                 }
1097
1098                 if (cpe_poll_enabled)
1099                         mod_timer(&cpe_poll_timer, jiffies + poll_time);
1100                 start_count = -1;
1101         }
1102
1103         return IRQ_HANDLED;
1104 }
1105
1106 /*
1107  *  ia64_mca_cpe_poll
1108  *
1109  *      Poll for Corrected Platform Errors (CPEs), trigger interrupt
1110  *      on first cpu, from there it will trickle through all the cpus.
1111  *
1112  * Inputs   :   dummy(unused)
1113  * Outputs  :   None
1114  *
1115  */
1116 static void
1117 ia64_mca_cpe_poll (unsigned long dummy)
1118 {
1119         /* Trigger a CPE interrupt cascade  */
1120         platform_send_ipi(first_cpu(cpu_online_map), IA64_CPEP_VECTOR, IA64_IPI_DM_INT, 0);
1121 }
1122
1123 #endif /* CONFIG_ACPI */
1124
1125 /*
1126  * C portion of the OS INIT handler
1127  *
1128  * Called from ia64_monarch_init_handler
1129  *
1130  * Inputs: pointer to pt_regs where processor info was saved.
1131  *
1132  * Returns:
1133  *   0 if SAL must warm boot the System
1134  *   1 if SAL must return to interrupted context using PAL_MC_RESUME
1135  *
1136  */
1137 void
1138 ia64_init_handler (struct pt_regs *pt, struct switch_stack *sw)
1139 {
1140         pal_min_state_area_t *ms;
1141
1142         oops_in_progress = 1;   /* avoid deadlock in printk, but it makes recovery dodgy */
1143         console_loglevel = 15;  /* make sure printks make it to console */
1144
1145         printk(KERN_INFO "Entered OS INIT handler. PSP=%lx\n",
1146                 ia64_sal_to_os_handoff_state.proc_state_param);
1147
1148         /*
1149          * Address of minstate area provided by PAL is physical,
1150          * uncacheable (bit 63 set). Convert to Linux virtual
1151          * address in region 6.
1152          */
1153         ms = (pal_min_state_area_t *)(ia64_sal_to_os_handoff_state.pal_min_state | (6ul<<61));
1154
1155         init_handler_platform(ms, pt, sw);      /* call platform specific routines */
1156 }
1157
1158 static int __init
1159 ia64_mca_disable_cpe_polling(char *str)
1160 {
1161         cpe_poll_enabled = 0;
1162         return 1;
1163 }
1164
1165 __setup("disable_cpe_poll", ia64_mca_disable_cpe_polling);
1166
1167 static struct irqaction cmci_irqaction = {
1168         .handler =      ia64_mca_cmc_int_handler,
1169         .flags =        SA_INTERRUPT,
1170         .name =         "cmc_hndlr"
1171 };
1172
1173 static struct irqaction cmcp_irqaction = {
1174         .handler =      ia64_mca_cmc_int_caller,
1175         .flags =        SA_INTERRUPT,
1176         .name =         "cmc_poll"
1177 };
1178
1179 static struct irqaction mca_rdzv_irqaction = {
1180         .handler =      ia64_mca_rendez_int_handler,
1181         .flags =        SA_INTERRUPT,
1182         .name =         "mca_rdzv"
1183 };
1184
1185 static struct irqaction mca_wkup_irqaction = {
1186         .handler =      ia64_mca_wakeup_int_handler,
1187         .flags =        SA_INTERRUPT,
1188         .name =         "mca_wkup"
1189 };
1190
1191 #ifdef CONFIG_ACPI
1192 static struct irqaction mca_cpe_irqaction = {
1193         .handler =      ia64_mca_cpe_int_handler,
1194         .flags =        SA_INTERRUPT,
1195         .name =         "cpe_hndlr"
1196 };
1197
1198 static struct irqaction mca_cpep_irqaction = {
1199         .handler =      ia64_mca_cpe_int_caller,
1200         .flags =        SA_INTERRUPT,
1201         .name =         "cpe_poll"
1202 };
1203 #endif /* CONFIG_ACPI */
1204
1205 /* Do per-CPU MCA-related initialization.  */
1206
1207 void __devinit
1208 ia64_mca_cpu_init(void *cpu_data)
1209 {
1210         void *pal_vaddr;
1211
1212         if (smp_processor_id() == 0) {
1213                 void *mca_data;
1214                 int cpu;
1215
1216                 mca_data = alloc_bootmem(sizeof(struct ia64_mca_cpu)
1217                                          * NR_CPUS);
1218                 for (cpu = 0; cpu < NR_CPUS; cpu++) {
1219                         __per_cpu_mca[cpu] = __pa(mca_data);
1220                         mca_data += sizeof(struct ia64_mca_cpu);
1221                 }
1222         }
1223
1224         /*
1225          * The MCA info structure was allocated earlier and its
1226          * physical address saved in __per_cpu_mca[cpu].  Copy that
1227          * address * to ia64_mca_data so we can access it as a per-CPU
1228          * variable.
1229          */
1230         __get_cpu_var(ia64_mca_data) = __per_cpu_mca[smp_processor_id()];
1231
1232         /*
1233          * Stash away a copy of the PTE needed to map the per-CPU page.
1234          * We may need it during MCA recovery.
1235          */
1236         __get_cpu_var(ia64_mca_per_cpu_pte) =
1237                 pte_val(mk_pte_phys(__pa(cpu_data), PAGE_KERNEL));
1238
1239         /*
1240          * Also, stash away a copy of the PAL address and the PTE
1241          * needed to map it.
1242          */
1243         pal_vaddr = efi_get_pal_addr();
1244         if (!pal_vaddr)
1245                 return;
1246         __get_cpu_var(ia64_mca_pal_base) =
1247                 GRANULEROUNDDOWN((unsigned long) pal_vaddr);
1248         __get_cpu_var(ia64_mca_pal_pte) = pte_val(mk_pte_phys(__pa(pal_vaddr),
1249                                                               PAGE_KERNEL));
1250 }
1251
1252 /*
1253  * ia64_mca_init
1254  *
1255  *  Do all the system level mca specific initialization.
1256  *
1257  *      1. Register spinloop and wakeup request interrupt vectors
1258  *
1259  *      2. Register OS_MCA handler entry point
1260  *
1261  *      3. Register OS_INIT handler entry point
1262  *
1263  *  4. Initialize MCA/CMC/INIT related log buffers maintained by the OS.
1264  *
1265  *  Note that this initialization is done very early before some kernel
1266  *  services are available.
1267  *
1268  *  Inputs  :   None
1269  *
1270  *  Outputs :   None
1271  */
1272 void __init
1273 ia64_mca_init(void)
1274 {
1275         ia64_fptr_t *mon_init_ptr = (ia64_fptr_t *)ia64_monarch_init_handler;
1276         ia64_fptr_t *slave_init_ptr = (ia64_fptr_t *)ia64_slave_init_handler;
1277         ia64_fptr_t *mca_hldlr_ptr = (ia64_fptr_t *)ia64_os_mca_dispatch;
1278         int i;
1279         s64 rc;
1280         struct ia64_sal_retval isrv;
1281         u64 timeout = IA64_MCA_RENDEZ_TIMEOUT;  /* platform specific */
1282
1283         IA64_MCA_DEBUG("%s: begin\n", __FUNCTION__);
1284
1285         /* Clear the Rendez checkin flag for all cpus */
1286         for(i = 0 ; i < NR_CPUS; i++)
1287                 ia64_mc_info.imi_rendez_checkin[i] = IA64_MCA_RENDEZ_CHECKIN_NOTDONE;
1288
1289         /*
1290          * Register the rendezvous spinloop and wakeup mechanism with SAL
1291          */
1292
1293         /* Register the rendezvous interrupt vector with SAL */
1294         while (1) {
1295                 isrv = ia64_sal_mc_set_params(SAL_MC_PARAM_RENDEZ_INT,
1296                                               SAL_MC_PARAM_MECHANISM_INT,
1297                                               IA64_MCA_RENDEZ_VECTOR,
1298                                               timeout,
1299                                               SAL_MC_PARAM_RZ_ALWAYS);
1300                 rc = isrv.status;
1301                 if (rc == 0)
1302                         break;
1303                 if (rc == -2) {
1304                         printk(KERN_INFO "Increasing MCA rendezvous timeout from "
1305                                 "%ld to %ld milliseconds\n", timeout, isrv.v0);
1306                         timeout = isrv.v0;
1307                         continue;
1308                 }
1309                 printk(KERN_ERR "Failed to register rendezvous interrupt "
1310                        "with SAL (status %ld)\n", rc);
1311                 return;
1312         }
1313
1314         /* Register the wakeup interrupt vector with SAL */
1315         isrv = ia64_sal_mc_set_params(SAL_MC_PARAM_RENDEZ_WAKEUP,
1316                                       SAL_MC_PARAM_MECHANISM_INT,
1317                                       IA64_MCA_WAKEUP_VECTOR,
1318                                       0, 0);
1319         rc = isrv.status;
1320         if (rc) {
1321                 printk(KERN_ERR "Failed to register wakeup interrupt with SAL "
1322                        "(status %ld)\n", rc);
1323                 return;
1324         }
1325
1326         IA64_MCA_DEBUG("%s: registered MCA rendezvous spinloop and wakeup mech.\n", __FUNCTION__);
1327
1328         ia64_mc_info.imi_mca_handler        = ia64_tpa(mca_hldlr_ptr->fp);
1329         /*
1330          * XXX - disable SAL checksum by setting size to 0; should be
1331          *      ia64_tpa(ia64_os_mca_dispatch_end) - ia64_tpa(ia64_os_mca_dispatch);
1332          */
1333         ia64_mc_info.imi_mca_handler_size       = 0;
1334
1335         /* Register the os mca handler with SAL */
1336         if ((rc = ia64_sal_set_vectors(SAL_VECTOR_OS_MCA,
1337                                        ia64_mc_info.imi_mca_handler,
1338                                        ia64_tpa(mca_hldlr_ptr->gp),
1339                                        ia64_mc_info.imi_mca_handler_size,
1340                                        0, 0, 0)))
1341         {
1342                 printk(KERN_ERR "Failed to register OS MCA handler with SAL "
1343                        "(status %ld)\n", rc);
1344                 return;
1345         }
1346
1347         IA64_MCA_DEBUG("%s: registered OS MCA handler with SAL at 0x%lx, gp = 0x%lx\n", __FUNCTION__,
1348                        ia64_mc_info.imi_mca_handler, ia64_tpa(mca_hldlr_ptr->gp));
1349
1350         /*
1351          * XXX - disable SAL checksum by setting size to 0, should be
1352          * size of the actual init handler in mca_asm.S.
1353          */
1354         ia64_mc_info.imi_monarch_init_handler           = ia64_tpa(mon_init_ptr->fp);
1355         ia64_mc_info.imi_monarch_init_handler_size      = 0;
1356         ia64_mc_info.imi_slave_init_handler             = ia64_tpa(slave_init_ptr->fp);
1357         ia64_mc_info.imi_slave_init_handler_size        = 0;
1358
1359         IA64_MCA_DEBUG("%s: OS INIT handler at %lx\n", __FUNCTION__,
1360                        ia64_mc_info.imi_monarch_init_handler);
1361
1362         /* Register the os init handler with SAL */
1363         if ((rc = ia64_sal_set_vectors(SAL_VECTOR_OS_INIT,
1364                                        ia64_mc_info.imi_monarch_init_handler,
1365                                        ia64_tpa(ia64_getreg(_IA64_REG_GP)),
1366                                        ia64_mc_info.imi_monarch_init_handler_size,
1367                                        ia64_mc_info.imi_slave_init_handler,
1368                                        ia64_tpa(ia64_getreg(_IA64_REG_GP)),
1369                                        ia64_mc_info.imi_slave_init_handler_size)))
1370         {
1371                 printk(KERN_ERR "Failed to register m/s INIT handlers with SAL "
1372                        "(status %ld)\n", rc);
1373                 return;
1374         }
1375
1376         IA64_MCA_DEBUG("%s: registered OS INIT handler with SAL\n", __FUNCTION__);
1377
1378         /*
1379          *  Configure the CMCI/P vector and handler. Interrupts for CMC are
1380          *  per-processor, so AP CMC interrupts are setup in smp_callin() (smpboot.c).
1381          */
1382         register_percpu_irq(IA64_CMC_VECTOR, &cmci_irqaction);
1383         register_percpu_irq(IA64_CMCP_VECTOR, &cmcp_irqaction);
1384         ia64_mca_cmc_vector_setup();       /* Setup vector on BSP */
1385
1386         /* Setup the MCA rendezvous interrupt vector */
1387         register_percpu_irq(IA64_MCA_RENDEZ_VECTOR, &mca_rdzv_irqaction);
1388
1389         /* Setup the MCA wakeup interrupt vector */
1390         register_percpu_irq(IA64_MCA_WAKEUP_VECTOR, &mca_wkup_irqaction);
1391
1392 #ifdef CONFIG_ACPI
1393         /* Setup the CPEI/P handler */
1394         register_percpu_irq(IA64_CPEP_VECTOR, &mca_cpep_irqaction);
1395 #endif
1396
1397         /* Initialize the areas set aside by the OS to buffer the
1398          * platform/processor error states for MCA/INIT/CMC
1399          * handling.
1400          */
1401         ia64_log_init(SAL_INFO_TYPE_MCA);
1402         ia64_log_init(SAL_INFO_TYPE_INIT);
1403         ia64_log_init(SAL_INFO_TYPE_CMC);
1404         ia64_log_init(SAL_INFO_TYPE_CPE);
1405
1406         mca_init = 1;
1407         printk(KERN_INFO "MCA related initialization done\n");
1408 }
1409
1410 /*
1411  * ia64_mca_late_init
1412  *
1413  *      Opportunity to setup things that require initialization later
1414  *      than ia64_mca_init.  Setup a timer to poll for CPEs if the
1415  *      platform doesn't support an interrupt driven mechanism.
1416  *
1417  *  Inputs  :   None
1418  *  Outputs :   Status
1419  */
1420 static int __init
1421 ia64_mca_late_init(void)
1422 {
1423         if (!mca_init)
1424                 return 0;
1425
1426         /* Setup the CMCI/P vector and handler */
1427         init_timer(&cmc_poll_timer);
1428         cmc_poll_timer.function = ia64_mca_cmc_poll;
1429
1430         /* Unmask/enable the vector */
1431         cmc_polling_enabled = 0;
1432         schedule_work(&cmc_enable_work);
1433
1434         IA64_MCA_DEBUG("%s: CMCI/P setup and enabled.\n", __FUNCTION__);
1435
1436 #ifdef CONFIG_ACPI
1437         /* Setup the CPEI/P vector and handler */
1438         cpe_vector = acpi_request_vector(ACPI_INTERRUPT_CPEI);
1439         init_timer(&cpe_poll_timer);
1440         cpe_poll_timer.function = ia64_mca_cpe_poll;
1441
1442         {
1443                 irq_desc_t *desc;
1444                 unsigned int irq;
1445
1446                 if (cpe_vector >= 0) {
1447                         /* If platform supports CPEI, enable the irq. */
1448                         cpe_poll_enabled = 0;
1449                         for (irq = 0; irq < NR_IRQS; ++irq)
1450                                 if (irq_to_vector(irq) == cpe_vector) {
1451                                         desc = irq_descp(irq);
1452                                         desc->status |= IRQ_PER_CPU;
1453                                         setup_irq(irq, &mca_cpe_irqaction);
1454                                 }
1455                         ia64_mca_register_cpev(cpe_vector);
1456                         IA64_MCA_DEBUG("%s: CPEI/P setup and enabled.\n", __FUNCTION__);
1457                 } else {
1458                         /* If platform doesn't support CPEI, get the timer going. */
1459                         if (cpe_poll_enabled) {
1460                                 ia64_mca_cpe_poll(0UL);
1461                                 IA64_MCA_DEBUG("%s: CPEP setup and enabled.\n", __FUNCTION__);
1462                         }
1463                 }
1464         }
1465 #endif
1466
1467         return 0;
1468 }
1469
1470 device_initcall(ia64_mca_late_init);