1 /*
   2  * CDDL HEADER START
   3  *
   4  * The contents of this file are subject to the terms of the
   5  * Common Development and Distribution License (the "License").
   6  * You may not use this file except in compliance with the License.
   7  *
   8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
   9  * or http://www.opensolaris.org/os/licensing.
  10  * See the License for the specific language governing permissions
  11  * and limitations under the License.
  12  *
  13  * When distributing Covered Code, include this CDDL HEADER in each
  14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
  15  * If applicable, add the following below this CDDL HEADER, with the
  16  * fields enclosed by brackets "[]" replaced with your own identifying
  17  * information: Portions Copyright [yyyy] [name of copyright owner]
  18  *
  19  * CDDL HEADER END
  20  */
  21 
  22 /*
  23  * Copyright (c) 1993, 2010, Oracle and/or its affiliates. All rights reserved.
  24  */
  25 /*
  26  * Copyright (c) 2010, Intel Corporation.
  27  * All rights reserved.
  28  */
  29 /*
  30  * Copyright (c) 2013, Joyent, Inc.  All rights reserved.
  31  */
  32 
  33 /*
  34  * To understand how the pcplusmp module interacts with the interrupt subsystem
  35  * read the theory statement in uts/i86pc/os/intr.c.
  36  */
  37 
  38 /*
  39  * PSMI 1.1 extensions are supported only in 2.6 and later versions.
  40  * PSMI 1.2 extensions are supported only in 2.7 and later versions.
  41  * PSMI 1.3 and 1.4 extensions are supported in Solaris 10.
  42  * PSMI 1.5 extensions are supported in Solaris Nevada.
  43  * PSMI 1.6 extensions are supported in Solaris Nevada.
  44  * PSMI 1.7 extensions are supported in Solaris Nevada.
  45  */
  46 #define PSMI_1_7
  47 
  48 #include <sys/processor.h>
  49 #include <sys/time.h>
  50 #include <sys/psm.h>
  51 #include <sys/smp_impldefs.h>
  52 #include <sys/cram.h>
  53 #include <sys/acpi/acpi.h>
  54 #include <sys/acpica.h>
  55 #include <sys/psm_common.h>
  56 #include <sys/apic.h>
  57 #include <sys/pit.h>
  58 #include <sys/ddi.h>
  59 #include <sys/sunddi.h>
  60 #include <sys/ddi_impldefs.h>
  61 #include <sys/pci.h>
  62 #include <sys/promif.h>
  63 #include <sys/x86_archext.h>
  64 #include <sys/cpc_impl.h>
  65 #include <sys/uadmin.h>
  66 #include <sys/panic.h>
  67 #include <sys/debug.h>
  68 #include <sys/archsystm.h>
  69 #include <sys/trap.h>
  70 #include <sys/machsystm.h>
  71 #include <sys/sysmacros.h>
  72 #include <sys/cpuvar.h>
  73 #include <sys/rm_platter.h>
  74 #include <sys/privregs.h>
  75 #include <sys/note.h>
  76 #include <sys/pci_intr_lib.h>
  77 #include <sys/spl.h>
  78 #include <sys/clock.h>
  79 #include <sys/cyclic.h>
  80 #include <sys/dditypes.h>
  81 #include <sys/sunddi.h>
  82 #include <sys/x_call.h>
  83 #include <sys/reboot.h>
  84 #include <sys/hpet.h>
  85 #include <sys/apic_common.h>
  86 #include <sys/apic_timer.h>
  87 
  88 /*
  89  *      Local Function Prototypes
  90  */
  91 static void apic_init_intr(void);
  92 
  93 /*
  94  *      standard MP entries
  95  */
  96 static int      apic_probe(void);
  97 static int      apic_getclkirq(int ipl);
  98 static void     apic_init(void);
  99 static void     apic_picinit(void);
 100 static int      apic_post_cpu_start(void);
 101 static int      apic_intr_enter(int ipl, int *vect);
 102 static void     apic_setspl(int ipl);
 103 static void     x2apic_setspl(int ipl);
 104 static int      apic_addspl(int ipl, int vector, int min_ipl, int max_ipl);
 105 static int      apic_delspl(int ipl, int vector, int min_ipl, int max_ipl);
 106 static int      apic_disable_intr(processorid_t cpun);
 107 static void     apic_enable_intr(processorid_t cpun);
 108 static int              apic_get_ipivect(int ipl, int type);
 109 static void     apic_post_cyclic_setup(void *arg);
 110 
 111 /*
 112  * The following vector assignments influence the value of ipltopri and
 113  * vectortoipl. Note that vectors 0 - 0x1f are not used. We can program
 114  * idle to 0 and IPL 0 to 0xf to differentiate idle in case
 115  * we care to do so in future. Note some IPLs which are rarely used
 116  * will share the vector ranges and heavily used IPLs (5 and 6) have
 117  * a wide range.
 118  *
 119  * This array is used to initialize apic_ipls[] (in apic_init()).
 120  *
 121  *      IPL             Vector range.           as passed to intr_enter
 122  *      0               none.
 123  *      1,2,3           0x20-0x2f               0x0-0xf
 124  *      4               0x30-0x3f               0x10-0x1f
 125  *      5               0x40-0x5f               0x20-0x3f
 126  *      6               0x60-0x7f               0x40-0x5f
 127  *      7,8,9           0x80-0x8f               0x60-0x6f
 128  *      10              0x90-0x9f               0x70-0x7f
 129  *      11              0xa0-0xaf               0x80-0x8f
 130  *      ...             ...
 131  *      15              0xe0-0xef               0xc0-0xcf
 132  *      15              0xf0-0xff               0xd0-0xdf
 133  */
 134 uchar_t apic_vectortoipl[APIC_AVAIL_VECTOR / APIC_VECTOR_PER_IPL] = {
 135         3, 4, 5, 5, 6, 6, 9, 10, 11, 12, 13, 14, 15, 15
 136 };
 137         /*
 138          * The ipl of an ISR at vector X is apic_vectortoipl[X>>4]
 139          * NOTE that this is vector as passed into intr_enter which is
 140          * programmed vector - 0x20 (APIC_BASE_VECT)
 141          */
 142 
 143 uchar_t apic_ipltopri[MAXIPL + 1];      /* unix ipl to apic pri */
 144         /* The taskpri to be programmed into apic to mask given ipl */
 145 
 146 #if defined(__amd64)
 147 static unsigned char dummy_cpu_pri[MAXIPL + 1];
 148 #endif
 149 
 150 /*
 151  * Correlation of the hardware vector to the IPL in use, initialized
 152  * from apic_vectortoipl[] in apic_init().  The final IPLs may not correlate
 153  * to the IPLs in apic_vectortoipl on some systems that share interrupt lines
 154  * connected to errata-stricken IOAPICs
 155  */
 156 uchar_t apic_ipls[APIC_AVAIL_VECTOR];
 157 
 158 /*
 159  * Patchable global variables.
 160  */
 161 int     apic_enable_hwsoftint = 0;      /* 0 - disable, 1 - enable      */
 162 int     apic_enable_bind_log = 1;       /* 1 - display interrupt binding log */
 163 
 164 /*
 165  *      Local static data
 166  */
 167 static struct   psm_ops apic_ops = {
 168         apic_probe,
 169 
 170         apic_init,
 171         apic_picinit,
 172         apic_intr_enter,
 173         apic_intr_exit,
 174         apic_setspl,
 175         apic_addspl,
 176         apic_delspl,
 177         apic_disable_intr,
 178         apic_enable_intr,
 179         (int (*)(int))NULL,             /* psm_softlvl_to_irq */
 180         (void (*)(int))NULL,            /* psm_set_softintr */
 181 
 182         apic_set_idlecpu,
 183         apic_unset_idlecpu,
 184 
 185         apic_clkinit,
 186         apic_getclkirq,
 187         (void (*)(void))NULL,           /* psm_hrtimeinit */
 188         apic_gethrtime,
 189 
 190         apic_get_next_processorid,
 191         apic_cpu_start,
 192         apic_post_cpu_start,
 193         apic_shutdown,
 194         apic_get_ipivect,
 195         apic_send_ipi,
 196 
 197         (int (*)(dev_info_t *, int))NULL,       /* psm_translate_irq */
 198         (void (*)(int, char *))NULL,    /* psm_notify_error */
 199         (void (*)(int))NULL,            /* psm_notify_func */
 200         apic_timer_reprogram,
 201         apic_timer_enable,
 202         apic_timer_disable,
 203         apic_post_cyclic_setup,
 204         apic_preshutdown,
 205         apic_intr_ops,                  /* Advanced DDI Interrupt framework */
 206         apic_state,                     /* save, restore apic state for S3 */
 207         apic_cpu_ops,                   /* CPU control interface. */
 208 };
 209 
 210 struct psm_ops *psmops = &apic_ops;
 211 
 212 static struct   psm_info apic_psm_info = {
 213         PSM_INFO_VER01_7,                       /* version */
 214         PSM_OWN_EXCLUSIVE,                      /* ownership */
 215         (struct psm_ops *)&apic_ops,                /* operation */
 216         APIC_PCPLUSMP_NAME,                     /* machine name */
 217         "pcplusmp v1.4 compatible",
 218 };
 219 
 220 static void *apic_hdlp;
 221 
 222 /*
 223  * apic_let_idle_redistribute can have the following values:
 224  * 0 - If clock decremented it from 1 to 0, clock has to call redistribute.
 225  * apic_redistribute_lock prevents multiple idle cpus from redistributing
 226  */
 227 int     apic_num_idle_redistributions = 0;
 228 static  int apic_let_idle_redistribute = 0;
 229 
 230 /* to gather intr data and redistribute */
 231 static void apic_redistribute_compute(void);
 232 
 233 /*
 234  *      This is the loadable module wrapper
 235  */
 236 
 237 int
 238 _init(void)
 239 {
 240         if (apic_coarse_hrtime)
 241                 apic_ops.psm_gethrtime = &apic_gettime;
 242         return (psm_mod_init(&apic_hdlp, &apic_psm_info));
 243 }
 244 
 245 int
 246 _fini(void)
 247 {
 248         return (psm_mod_fini(&apic_hdlp, &apic_psm_info));
 249 }
 250 
 251 int
 252 _info(struct modinfo *modinfop)
 253 {
 254         return (psm_mod_info(&apic_hdlp, &apic_psm_info, modinfop));
 255 }
 256 
 257 static int
 258 apic_probe(void)
 259 {
 260         /* check if apix is initialized */
 261         if (apix_enable && apix_loaded())
 262                 return (PSM_FAILURE);
 263         else
 264                 apix_enable = 0; /* continue using pcplusmp PSM */
 265 
 266         return (apic_probe_common(apic_psm_info.p_mach_idstring));
 267 }
 268 
 269 static uchar_t
 270 apic_xlate_vector_by_irq(uchar_t irq)
 271 {
 272         if (apic_irq_table[irq] == NULL)
 273                 return (0);
 274 
 275         return (apic_irq_table[irq]->airq_vector);
 276 }
 277 
 278 void
 279 apic_init(void)
 280 {
 281         int i;
 282         int     j = 1;
 283 
 284         psm_get_ioapicid = apic_get_ioapicid;
 285         psm_get_localapicid = apic_get_localapicid;
 286         psm_xlate_vector_by_irq = apic_xlate_vector_by_irq;
 287 
 288         apic_ipltopri[0] = APIC_VECTOR_PER_IPL; /* leave 0 for idle */
 289         for (i = 0; i < (APIC_AVAIL_VECTOR / APIC_VECTOR_PER_IPL); i++) {
 290                 if ((i < ((APIC_AVAIL_VECTOR / APIC_VECTOR_PER_IPL) - 1)) &&
 291                     (apic_vectortoipl[i + 1] == apic_vectortoipl[i]))
 292                         /* get to highest vector at the same ipl */
 293                         continue;
 294                 for (; j <= apic_vectortoipl[i]; j++) {
 295                         apic_ipltopri[j] = (i << APIC_IPL_SHIFT) +
 296                             APIC_BASE_VECT;
 297                 }
 298         }
 299         for (; j < MAXIPL + 1; j++)
 300                 /* fill up any empty ipltopri slots */
 301                 apic_ipltopri[j] = (i << APIC_IPL_SHIFT) + APIC_BASE_VECT;
 302         apic_init_common();
 303 #if defined(__amd64)
 304         CPU->cpu_pri_data = dummy_cpu_pri;
 305 #else
 306         if (cpuid_have_cr8access(CPU))
 307                 apic_have_32bit_cr8 = 1;
 308 #endif  /* __amd64 */
 309 }
 310 
 311 static void
 312 apic_init_intr(void)
 313 {
 314         processorid_t   cpun = psm_get_cpu_id();
 315         uint_t nlvt;
 316         uint32_t svr = AV_UNIT_ENABLE | APIC_SPUR_INTR;
 317 
 318         apic_reg_ops->apic_write_task_reg(APIC_MASK_ALL);
 319 
 320         if (apic_mode == LOCAL_APIC) {
 321                 /*
 322                  * We are running APIC in MMIO mode.
 323                  */
 324                 if (apic_flat_model) {
 325                         apic_reg_ops->apic_write(APIC_FORMAT_REG,
 326                             APIC_FLAT_MODEL);
 327                 } else {
 328                         apic_reg_ops->apic_write(APIC_FORMAT_REG,
 329                             APIC_CLUSTER_MODEL);
 330                 }
 331 
 332                 apic_reg_ops->apic_write(APIC_DEST_REG,
 333                     AV_HIGH_ORDER >> cpun);
 334         }
 335 
 336         if (apic_directed_EOI_supported()) {
 337                 /*
 338                  * Setting the 12th bit in the Spurious Interrupt Vector
 339                  * Register suppresses broadcast EOIs generated by the local
 340                  * APIC. The suppression of broadcast EOIs happens only when
 341                  * interrupts are level-triggered.
 342                  */
 343                 svr |= APIC_SVR_SUPPRESS_BROADCAST_EOI;
 344         }
 345 
 346         /* need to enable APIC before unmasking NMI */
 347         apic_reg_ops->apic_write(APIC_SPUR_INT_REG, svr);
 348 
 349         /*
 350          * Presence of an invalid vector with delivery mode AV_FIXED can
 351          * cause an error interrupt, even if the entry is masked...so
 352          * write a valid vector to LVT entries along with the mask bit
 353          */
 354 
 355         /* All APICs have timer and LINT0/1 */
 356         apic_reg_ops->apic_write(APIC_LOCAL_TIMER, AV_MASK|APIC_RESV_IRQ);
 357         apic_reg_ops->apic_write(APIC_INT_VECT0, AV_MASK|APIC_RESV_IRQ);
 358         apic_reg_ops->apic_write(APIC_INT_VECT1, AV_NMI);    /* enable NMI */
 359 
 360         /*
 361          * On integrated APICs, the number of LVT entries is
 362          * 'Max LVT entry' + 1; on 82489DX's (non-integrated
 363          * APICs), nlvt is "3" (LINT0, LINT1, and timer)
 364          */
 365 
 366         if (apic_cpus[cpun].aci_local_ver < APIC_INTEGRATED_VERS) {
 367                 nlvt = 3;
 368         } else {
 369                 nlvt = ((apic_reg_ops->apic_read(APIC_VERS_REG) >> 16) &
 370                     0xFF) + 1;
 371         }
 372 
 373         if (nlvt >= 5) {
 374                 /* Enable performance counter overflow interrupt */
 375 
 376                 if (!is_x86_feature(x86_featureset, X86FSET_MSR))
 377                         apic_enable_cpcovf_intr = 0;
 378                 if (apic_enable_cpcovf_intr) {
 379                         if (apic_cpcovf_vect == 0) {
 380                                 int ipl = APIC_PCINT_IPL;
 381                                 int irq = apic_get_ipivect(ipl, -1);
 382 
 383                                 ASSERT(irq != -1);
 384                                 apic_cpcovf_vect =
 385                                     apic_irq_table[irq]->airq_vector;
 386                                 ASSERT(apic_cpcovf_vect);
 387                                 (void) add_avintr(NULL, ipl,
 388                                     (avfunc)kcpc_hw_overflow_intr,
 389                                     "apic pcint", irq, NULL, NULL, NULL, NULL);
 390                                 kcpc_hw_overflow_intr_installed = 1;
 391                                 kcpc_hw_enable_cpc_intr =
 392                                     apic_cpcovf_mask_clear;
 393                         }
 394                         apic_reg_ops->apic_write(APIC_PCINT_VECT,
 395                             apic_cpcovf_vect);
 396                 }
 397         }
 398 
 399         if (nlvt >= 6) {
 400                 /* Only mask TM intr if the BIOS apparently doesn't use it */
 401 
 402                 uint32_t lvtval;
 403 
 404                 lvtval = apic_reg_ops->apic_read(APIC_THERM_VECT);
 405                 if (((lvtval & AV_MASK) == AV_MASK) ||
 406                     ((lvtval & AV_DELIV_MODE) != AV_SMI)) {
 407                         apic_reg_ops->apic_write(APIC_THERM_VECT,
 408                             AV_MASK|APIC_RESV_IRQ);
 409                 }
 410         }
 411 
 412         /* Enable error interrupt */
 413 
 414         if (nlvt >= 4 && apic_enable_error_intr) {
 415                 if (apic_errvect == 0) {
 416                         int ipl = 0xf;  /* get highest priority intr */
 417                         int irq = apic_get_ipivect(ipl, -1);
 418 
 419                         ASSERT(irq != -1);
 420                         apic_errvect = apic_irq_table[irq]->airq_vector;
 421                         ASSERT(apic_errvect);
 422                         /*
 423                          * Not PSMI compliant, but we are going to merge
 424                          * with ON anyway
 425                          */
 426                         (void) add_avintr((void *)NULL, ipl,
 427                             (avfunc)apic_error_intr, "apic error intr",
 428                             irq, NULL, NULL, NULL, NULL);
 429                 }
 430                 apic_reg_ops->apic_write(APIC_ERR_VECT, apic_errvect);
 431                 apic_reg_ops->apic_write(APIC_ERROR_STATUS, 0);
 432                 apic_reg_ops->apic_write(APIC_ERROR_STATUS, 0);
 433         }
 434 
 435         /* Enable CMCI interrupt */
 436         if (cmi_enable_cmci) {
 437 
 438                 mutex_enter(&cmci_cpu_setup_lock);
 439                 if (cmci_cpu_setup_registered == 0) {
 440                         mutex_enter(&cpu_lock);
 441                         register_cpu_setup_func(cmci_cpu_setup, NULL);
 442                         mutex_exit(&cpu_lock);
 443                         cmci_cpu_setup_registered = 1;
 444                 }
 445                 mutex_exit(&cmci_cpu_setup_lock);
 446 
 447                 if (apic_cmci_vect == 0) {
 448                         int ipl = 0x2;
 449                         int irq = apic_get_ipivect(ipl, -1);
 450 
 451                         ASSERT(irq != -1);
 452                         apic_cmci_vect = apic_irq_table[irq]->airq_vector;
 453                         ASSERT(apic_cmci_vect);
 454 
 455                         (void) add_avintr(NULL, ipl,
 456                             (avfunc)cmi_cmci_trap,
 457                             "apic cmci intr", irq, NULL, NULL, NULL, NULL);
 458                 }
 459                 apic_reg_ops->apic_write(APIC_CMCI_VECT, apic_cmci_vect);
 460         }
 461 }
 462 
 463 static void
 464 apic_picinit(void)
 465 {
 466         int i, j;
 467         uint_t isr;
 468 
 469         /*
 470          * Initialize and enable interrupt remapping before apic
 471          * hardware initialization
 472          */
 473         apic_intrmap_init(apic_mode);
 474 
 475         /*
 476          * On UniSys Model 6520, the BIOS leaves vector 0x20 isr
 477          * bit on without clearing it with EOI.  Since softint
 478          * uses vector 0x20 to interrupt itself, so softint will
 479          * not work on this machine.  In order to fix this problem
 480          * a check is made to verify all the isr bits are clear.
 481          * If not, EOIs are issued to clear the bits.
 482          */
 483         for (i = 7; i >= 1; i--) {
 484                 isr = apic_reg_ops->apic_read(APIC_ISR_REG + (i * 4));
 485                 if (isr != 0)
 486                         for (j = 0; ((j < 32) && (isr != 0)); j++)
 487                                 if (isr & (1 << j)) {
 488                                         apic_reg_ops->apic_write(
 489                                             APIC_EOI_REG, 0);
 490                                         isr &= ~(1 << j);
 491                                         apic_error |= APIC_ERR_BOOT_EOI;
 492                                 }
 493         }
 494 
 495         /* set a flag so we know we have run apic_picinit() */
 496         apic_picinit_called = 1;
 497         LOCK_INIT_CLEAR(&apic_gethrtime_lock);
 498         LOCK_INIT_CLEAR(&apic_ioapic_lock);
 499         LOCK_INIT_CLEAR(&apic_error_lock);
 500         LOCK_INIT_CLEAR(&apic_mode_switch_lock);
 501 
 502         picsetup();      /* initialise the 8259 */
 503 
 504         /* add nmi handler - least priority nmi handler */
 505         LOCK_INIT_CLEAR(&apic_nmi_lock);
 506 
 507         if (!psm_add_nmintr(0, (avfunc) apic_nmi_intr,
 508             "pcplusmp NMI handler", (caddr_t)NULL))
 509                 cmn_err(CE_WARN, "pcplusmp: Unable to add nmi handler");
 510 
 511         /*
 512          * Check for directed-EOI capability in the local APIC.
 513          */
 514         if (apic_directed_EOI_supported() == 1) {
 515                 apic_set_directed_EOI_handler();
 516         }
 517 
 518         apic_init_intr();
 519 
 520         /* enable apic mode if imcr present */
 521         if (apic_imcrp) {
 522                 outb(APIC_IMCR_P1, (uchar_t)APIC_IMCR_SELECT);
 523                 outb(APIC_IMCR_P2, (uchar_t)APIC_IMCR_APIC);
 524         }
 525 
 526         ioapic_init_intr(IOAPIC_MASK);
 527 }
 528 
 529 #ifdef  DEBUG
 530 void
 531 apic_break(void)
 532 {
 533 }
 534 #endif /* DEBUG */
 535 
 536 /*
 537  * platform_intr_enter
 538  *
 539  *      Called at the beginning of the interrupt service routine to
 540  *      mask all level equal to and below the interrupt priority
 541  *      of the interrupting vector.  An EOI should be given to
 542  *      the interrupt controller to enable other HW interrupts.
 543  *
 544  *      Return -1 for spurious interrupts
 545  *
 546  */
 547 /*ARGSUSED*/
 548 static int
 549 apic_intr_enter(int ipl, int *vectorp)
 550 {
 551         uchar_t vector;
 552         int nipl;
 553         int irq;
 554         ulong_t iflag;
 555         apic_cpus_info_t *cpu_infop;
 556 
 557         /*
 558          * The real vector delivered is (*vectorp + 0x20), but our caller
 559          * subtracts 0x20 from the vector before passing it to us.
 560          * (That's why APIC_BASE_VECT is 0x20.)
 561          */
 562         vector = (uchar_t)*vectorp;
 563 
 564         /* if interrupted by the clock, increment apic_nsec_since_boot */
 565         if (vector == apic_clkvect) {
 566                 if (!apic_oneshot) {
 567                         /* NOTE: this is not MT aware */
 568                         apic_hrtime_stamp++;
 569                         apic_nsec_since_boot += apic_nsec_per_intr;
 570                         apic_hrtime_stamp++;
 571                         last_count_read = apic_hertz_count;
 572                         apic_redistribute_compute();
 573                 }
 574 
 575                 /* We will avoid all the book keeping overhead for clock */
 576                 nipl = apic_ipls[vector];
 577 
 578                 *vectorp = apic_vector_to_irq[vector + APIC_BASE_VECT];
 579                 if (apic_mode == LOCAL_APIC) {
 580 #if defined(__amd64)
 581                         setcr8((ulong_t)(apic_ipltopri[nipl] >>
 582                             APIC_IPL_SHIFT));
 583 #else
 584                         if (apic_have_32bit_cr8)
 585                                 setcr8((ulong_t)(apic_ipltopri[nipl] >>
 586                                     APIC_IPL_SHIFT));
 587                         else
 588                                 LOCAL_APIC_WRITE_REG(APIC_TASK_REG,
 589                                     (uint32_t)apic_ipltopri[nipl]);
 590 #endif
 591                         LOCAL_APIC_WRITE_REG(APIC_EOI_REG, 0);
 592                 } else {
 593                         X2APIC_WRITE(APIC_TASK_REG, apic_ipltopri[nipl]);
 594                         X2APIC_WRITE(APIC_EOI_REG, 0);
 595                 }
 596 
 597                 return (nipl);
 598         }
 599 
 600         cpu_infop = &apic_cpus[psm_get_cpu_id()];
 601 
 602         if (vector == (APIC_SPUR_INTR - APIC_BASE_VECT)) {
 603                 cpu_infop->aci_spur_cnt++;
 604                 return (APIC_INT_SPURIOUS);
 605         }
 606 
 607         /* Check if the vector we got is really what we need */
 608         if (apic_revector_pending) {
 609                 /*
 610                  * Disable interrupts for the duration of
 611                  * the vector translation to prevent a self-race for
 612                  * the apic_revector_lock.  This cannot be done
 613                  * in apic_xlate_vector because it is recursive and
 614                  * we want the vector translation to be atomic with
 615                  * respect to other (higher-priority) interrupts.
 616                  */
 617                 iflag = intr_clear();
 618                 vector = apic_xlate_vector(vector + APIC_BASE_VECT) -
 619                     APIC_BASE_VECT;
 620                 intr_restore(iflag);
 621         }
 622 
 623         nipl = apic_ipls[vector];
 624         *vectorp = irq = apic_vector_to_irq[vector + APIC_BASE_VECT];
 625 
 626         if (apic_mode == LOCAL_APIC) {
 627 #if defined(__amd64)
 628                 setcr8((ulong_t)(apic_ipltopri[nipl] >> APIC_IPL_SHIFT));
 629 #else
 630                 if (apic_have_32bit_cr8)
 631                         setcr8((ulong_t)(apic_ipltopri[nipl] >>
 632                             APIC_IPL_SHIFT));
 633                 else
 634                         LOCAL_APIC_WRITE_REG(APIC_TASK_REG,
 635                             (uint32_t)apic_ipltopri[nipl]);
 636 #endif
 637         } else {
 638                 X2APIC_WRITE(APIC_TASK_REG, apic_ipltopri[nipl]);
 639         }
 640 
 641         cpu_infop->aci_current[nipl] = (uchar_t)irq;
 642         cpu_infop->aci_curipl = (uchar_t)nipl;
 643         cpu_infop->aci_ISR_in_progress |= 1 << nipl;
 644 
 645         /*
 646          * apic_level_intr could have been assimilated into the irq struct.
 647          * but, having it as a character array is more efficient in terms of
 648          * cache usage. So, we leave it as is.
 649          */
 650         if (!apic_level_intr[irq]) {
 651                 if (apic_mode == LOCAL_APIC) {
 652                         LOCAL_APIC_WRITE_REG(APIC_EOI_REG, 0);
 653                 } else {
 654                         X2APIC_WRITE(APIC_EOI_REG, 0);
 655                 }
 656         }
 657 
 658 #ifdef  DEBUG
 659         APIC_DEBUG_BUF_PUT(vector);
 660         APIC_DEBUG_BUF_PUT(irq);
 661         APIC_DEBUG_BUF_PUT(nipl);
 662         APIC_DEBUG_BUF_PUT(psm_get_cpu_id());
 663         if ((apic_stretch_interrupts) && (apic_stretch_ISR & (1 << nipl)))
 664                 drv_usecwait(apic_stretch_interrupts);
 665 
 666         if (apic_break_on_cpu == psm_get_cpu_id())
 667                 apic_break();
 668 #endif /* DEBUG */
 669         return (nipl);
 670 }
 671 
 672 /*
 673  * This macro is a common code used by MMIO local apic and X2APIC
 674  * local apic.
 675  */
 676 #define APIC_INTR_EXIT() \
 677 { \
 678         cpu_infop = &apic_cpus[psm_get_cpu_id()]; \
 679         if (apic_level_intr[irq]) \
 680                 apic_reg_ops->apic_send_eoi(irq); \
 681         cpu_infop->aci_curipl = (uchar_t)prev_ipl; \
 682         /* ISR above current pri could not be in progress */ \
 683         cpu_infop->aci_ISR_in_progress &= (2 << prev_ipl) - 1; \
 684 }
 685 
 686 /*
 687  * Any changes made to this function must also change X2APIC
 688  * version of intr_exit.
 689  */
 690 void
 691 apic_intr_exit(int prev_ipl, int irq)
 692 {
 693         apic_cpus_info_t *cpu_infop;
 694 
 695 #if defined(__amd64)
 696         setcr8((ulong_t)(apic_ipltopri[prev_ipl] >> APIC_IPL_SHIFT));
 697 #else
 698         if (apic_have_32bit_cr8)
 699                 setcr8((ulong_t)(apic_ipltopri[prev_ipl] >> APIC_IPL_SHIFT));
 700         else
 701                 apicadr[APIC_TASK_REG] = apic_ipltopri[prev_ipl];
 702 #endif
 703 
 704         APIC_INTR_EXIT();
 705 }
 706 
 707 /*
 708  * Same as apic_intr_exit() except it uses MSR rather than MMIO
 709  * to access local apic registers.
 710  */
 711 void
 712 x2apic_intr_exit(int prev_ipl, int irq)
 713 {
 714         apic_cpus_info_t *cpu_infop;
 715 
 716         X2APIC_WRITE(APIC_TASK_REG, apic_ipltopri[prev_ipl]);
 717         APIC_INTR_EXIT();
 718 }
 719 
 720 intr_exit_fn_t
 721 psm_intr_exit_fn(void)
 722 {
 723         if (apic_mode == LOCAL_X2APIC)
 724                 return (x2apic_intr_exit);
 725 
 726         return (apic_intr_exit);
 727 }
 728 
 729 /*
 730  * Mask all interrupts below or equal to the given IPL.
 731  * Any changes made to this function must also change X2APIC
 732  * version of setspl.
 733  */
 734 static void
 735 apic_setspl(int ipl)
 736 {
 737 #if defined(__amd64)
 738         setcr8((ulong_t)(apic_ipltopri[ipl] >> APIC_IPL_SHIFT));
 739 #else
 740         if (apic_have_32bit_cr8)
 741                 setcr8((ulong_t)(apic_ipltopri[ipl] >> APIC_IPL_SHIFT));
 742         else
 743                 apicadr[APIC_TASK_REG] = apic_ipltopri[ipl];
 744 #endif
 745 
 746         /* interrupts at ipl above this cannot be in progress */
 747         apic_cpus[psm_get_cpu_id()].aci_ISR_in_progress &= (2 << ipl) - 1;
 748         /*
 749          * this is a patch fix for the ALR QSMP P5 machine, so that interrupts
 750          * have enough time to come in before the priority is raised again
 751          * during the idle() loop.
 752          */
 753         if (apic_setspl_delay)
 754                 (void) apic_reg_ops->apic_get_pri();
 755 }
 756 
 757 /*
 758  * X2APIC version of setspl.
 759  * Mask all interrupts below or equal to the given IPL
 760  */
 761 static void
 762 x2apic_setspl(int ipl)
 763 {
 764         X2APIC_WRITE(APIC_TASK_REG, apic_ipltopri[ipl]);
 765 
 766         /* interrupts at ipl above this cannot be in progress */
 767         apic_cpus[psm_get_cpu_id()].aci_ISR_in_progress &= (2 << ipl) - 1;
 768 }
 769 
 770 /*ARGSUSED*/
 771 static int
 772 apic_addspl(int irqno, int ipl, int min_ipl, int max_ipl)
 773 {
 774         return (apic_addspl_common(irqno, ipl, min_ipl, max_ipl));
 775 }
 776 
 777 static int
 778 apic_delspl(int irqno, int ipl, int min_ipl, int max_ipl)
 779 {
 780         return (apic_delspl_common(irqno, ipl, min_ipl,  max_ipl));
 781 }
 782 
 783 static int
 784 apic_post_cpu_start(void)
 785 {
 786         int cpun;
 787         static int cpus_started = 1;
 788 
 789         /* We know this CPU + BSP  started successfully. */
 790         cpus_started++;
 791 
 792         /*
 793          * On BSP we would have enabled X2APIC, if supported by processor,
 794          * in acpi_probe(), but on AP we do it here.
 795          *
 796          * We enable X2APIC mode only if BSP is running in X2APIC & the
 797          * local APIC mode of the current CPU is MMIO (xAPIC).
 798          */
 799         if (apic_mode == LOCAL_X2APIC && apic_detect_x2apic() &&
 800             apic_local_mode() == LOCAL_APIC) {
 801                 apic_enable_x2apic();
 802         }
 803 
 804         /*
 805          * Switch back to x2apic IPI sending method for performance when target
 806          * CPU has entered x2apic mode.
 807          */
 808         if (apic_mode == LOCAL_X2APIC) {
 809                 apic_switch_ipi_callback(B_FALSE);
 810         }
 811 
 812         splx(ipltospl(LOCK_LEVEL));
 813         apic_init_intr();
 814 
 815         /*
 816          * since some systems don't enable the internal cache on the non-boot
 817          * cpus, so we have to enable them here
 818          */
 819         setcr0(getcr0() & ~(CR0_CD | CR0_NW));
 820 
 821 #ifdef  DEBUG
 822         APIC_AV_PENDING_SET();
 823 #else
 824         if (apic_mode == LOCAL_APIC)
 825                 APIC_AV_PENDING_SET();
 826 #endif  /* DEBUG */
 827 
 828         /*
 829          * We may be booting, or resuming from suspend; aci_status will
 830          * be APIC_CPU_INTR_ENABLE if coming from suspend, so we add the
 831          * APIC_CPU_ONLINE flag here rather than setting aci_status completely.
 832          */
 833         cpun = psm_get_cpu_id();
 834         apic_cpus[cpun].aci_status |= APIC_CPU_ONLINE;
 835 
 836         apic_reg_ops->apic_write(APIC_DIVIDE_REG, apic_divide_reg_init);
 837         return (PSM_SUCCESS);
 838 }
 839 
 840 /*
 841  * type == -1 indicates it is an internal request. Do not change
 842  * resv_vector for these requests
 843  */
 844 static int
 845 apic_get_ipivect(int ipl, int type)
 846 {
 847         uchar_t vector;
 848         int irq;
 849 
 850         if ((irq = apic_allocate_irq(APIC_VECTOR(ipl))) != -1) {
 851                 if (vector = apic_allocate_vector(ipl, irq, 1)) {
 852                         apic_irq_table[irq]->airq_mps_intr_index =
 853                             RESERVE_INDEX;
 854                         apic_irq_table[irq]->airq_vector = vector;
 855                         if (type != -1) {
 856                                 apic_resv_vector[ipl] = vector;
 857                         }
 858                         return (irq);
 859                 }
 860         }
 861         apic_error |= APIC_ERR_GET_IPIVECT_FAIL;
 862         return (-1);    /* shouldn't happen */
 863 }
 864 
 865 static int
 866 apic_getclkirq(int ipl)
 867 {
 868         int     irq;
 869 
 870         if ((irq = apic_get_ipivect(ipl, -1)) == -1)
 871                 return (-1);
 872         /*
 873          * Note the vector in apic_clkvect for per clock handling.
 874          */
 875         apic_clkvect = apic_irq_table[irq]->airq_vector - APIC_BASE_VECT;
 876         APIC_VERBOSE_IOAPIC((CE_NOTE, "get_clkirq: vector = %x\n",
 877             apic_clkvect));
 878         return (irq);
 879 }
 880 
 881 /*
 882  * Try and disable all interrupts. We just assign interrupts to other
 883  * processors based on policy. If any were bound by user request, we
 884  * let them continue and return failure. We do not bother to check
 885  * for cache affinity while rebinding.
 886  */
 887 
 888 static int
 889 apic_disable_intr(processorid_t cpun)
 890 {
 891         int bind_cpu = 0, i, hardbound = 0;
 892         apic_irq_t *irq_ptr;
 893         ulong_t iflag;
 894 
 895         iflag = intr_clear();
 896         lock_set(&apic_ioapic_lock);
 897 
 898         for (i = 0; i <= APIC_MAX_VECTOR; i++) {
 899                 if (apic_reprogram_info[i].done == B_FALSE) {
 900                         if (apic_reprogram_info[i].bindcpu == cpun) {
 901                                 /*
 902                                  * CPU is busy -- it's the target of
 903                                  * a pending reprogramming attempt
 904                                  */
 905                                 lock_clear(&apic_ioapic_lock);
 906                                 intr_restore(iflag);
 907                                 return (PSM_FAILURE);
 908                         }
 909                 }
 910         }
 911 
 912         apic_cpus[cpun].aci_status &= ~APIC_CPU_INTR_ENABLE;
 913 
 914         apic_cpus[cpun].aci_curipl = 0;
 915 
 916         i = apic_min_device_irq;
 917         for (; i <= apic_max_device_irq; i++) {
 918                 /*
 919                  * If there are bound interrupts on this cpu, then
 920                  * rebind them to other processors.
 921                  */
 922                 if ((irq_ptr = apic_irq_table[i]) != NULL) {
 923                         ASSERT((irq_ptr->airq_temp_cpu == IRQ_UNBOUND) ||
 924                             (irq_ptr->airq_temp_cpu == IRQ_UNINIT) ||
 925                             (apic_cpu_in_range(irq_ptr->airq_temp_cpu)));
 926 
 927                         if (irq_ptr->airq_temp_cpu == (cpun | IRQ_USER_BOUND)) {
 928                                 hardbound = 1;
 929                                 continue;
 930                         }
 931 
 932                         if (irq_ptr->airq_temp_cpu == cpun) {
 933                                 do {
 934                                         bind_cpu =
 935                                             apic_find_cpu(APIC_CPU_INTR_ENABLE);
 936                                 } while (apic_rebind_all(irq_ptr, bind_cpu));
 937                         }
 938                 }
 939         }
 940 
 941         lock_clear(&apic_ioapic_lock);
 942         intr_restore(iflag);
 943 
 944         if (hardbound) {
 945                 cmn_err(CE_WARN, "Could not disable interrupts on %d"
 946                     "due to user bound interrupts", cpun);
 947                 return (PSM_FAILURE);
 948         }
 949         else
 950                 return (PSM_SUCCESS);
 951 }
 952 
 953 /*
 954  * Bind interrupts to the CPU's local APIC.
 955  * Interrupts should not be bound to a CPU's local APIC until the CPU
 956  * is ready to receive interrupts.
 957  */
 958 static void
 959 apic_enable_intr(processorid_t cpun)
 960 {
 961         int     i;
 962         apic_irq_t *irq_ptr;
 963         ulong_t iflag;
 964 
 965         iflag = intr_clear();
 966         lock_set(&apic_ioapic_lock);
 967 
 968         apic_cpus[cpun].aci_status |= APIC_CPU_INTR_ENABLE;
 969 
 970         i = apic_min_device_irq;
 971         for (i = apic_min_device_irq; i <= apic_max_device_irq; i++) {
 972                 if ((irq_ptr = apic_irq_table[i]) != NULL) {
 973                         if ((irq_ptr->airq_cpu & ~IRQ_USER_BOUND) == cpun) {
 974                                 (void) apic_rebind_all(irq_ptr,
 975                                     irq_ptr->airq_cpu);
 976                         }
 977                 }
 978         }
 979 
 980         if (apic_cpus[cpun].aci_status & APIC_CPU_SUSPEND)
 981                 apic_cpus[cpun].aci_status &= ~APIC_CPU_SUSPEND;
 982 
 983         lock_clear(&apic_ioapic_lock);
 984         intr_restore(iflag);
 985 }
 986 
 987 /*
 988  * If this module needs a periodic handler for the interrupt distribution, it
 989  * can be added here. The argument to the periodic handler is not currently
 990  * used, but is reserved for future.
 991  */
 992 static void
 993 apic_post_cyclic_setup(void *arg)
 994 {
 995 _NOTE(ARGUNUSED(arg))
 996 
 997         cyc_handler_t cyh;
 998         cyc_time_t cyt;
 999 
1000         /* cpu_lock is held */
1001         /* set up a periodic handler for intr redistribution */
1002 
1003         /*
1004          * In peridoc mode intr redistribution processing is done in
1005          * apic_intr_enter during clk intr processing
1006          */
1007         if (!apic_oneshot)
1008                 return;
1009 
1010         /*
1011          * Register a periodical handler for the redistribution processing.
1012          * Though we would generally prefer to use the DDI interface for
1013          * periodic handler invocation, ddi_periodic_add(9F), we are
1014          * unfortunately already holding cpu_lock, which ddi_periodic_add will
1015          * attempt to take for us.  Thus, we add our own cyclic directly:
1016          */
1017         cyh.cyh_func = (void (*)(void *))apic_redistribute_compute;
1018         cyh.cyh_arg = NULL;
1019         cyh.cyh_level = CY_LOW_LEVEL;
1020 
1021         cyt.cyt_when = 0;
1022         cyt.cyt_interval = apic_redistribute_sample_interval;
1023 
1024         apic_cyclic_id = cyclic_add(&cyh, &cyt);
1025 }
1026 
1027 static void
1028 apic_redistribute_compute(void)
1029 {
1030         int     i, j, max_busy;
1031 
1032         if (apic_enable_dynamic_migration) {
1033                 if (++apic_nticks == apic_sample_factor_redistribution) {
1034                         /*
1035                          * Time to call apic_intr_redistribute().
1036                          * reset apic_nticks. This will cause max_busy
1037                          * to be calculated below and if it is more than
1038                          * apic_int_busy, we will do the whole thing
1039                          */
1040                         apic_nticks = 0;
1041                 }
1042                 max_busy = 0;
1043                 for (i = 0; i < apic_nproc; i++) {
1044                         if (!apic_cpu_in_range(i))
1045                                 continue;
1046 
1047                         /*
1048                          * Check if curipl is non zero & if ISR is in
1049                          * progress
1050                          */
1051                         if (((j = apic_cpus[i].aci_curipl) != 0) &&
1052                             (apic_cpus[i].aci_ISR_in_progress & (1 << j))) {
1053 
1054                                 int     irq;
1055                                 apic_cpus[i].aci_busy++;
1056                                 irq = apic_cpus[i].aci_current[j];
1057                                 apic_irq_table[irq]->airq_busy++;
1058                         }
1059 
1060                         if (!apic_nticks &&
1061                             (apic_cpus[i].aci_busy > max_busy))
1062                                 max_busy = apic_cpus[i].aci_busy;
1063                 }
1064                 if (!apic_nticks) {
1065                         if (max_busy > apic_int_busy_mark) {
1066                         /*
1067                          * We could make the following check be
1068                          * skipped > 1 in which case, we get a
1069                          * redistribution at half the busy mark (due to
1070                          * double interval). Need to be able to collect
1071                          * more empirical data to decide if that is a
1072                          * good strategy. Punt for now.
1073                          */
1074                                 if (apic_skipped_redistribute) {
1075                                         apic_cleanup_busy();
1076                                         apic_skipped_redistribute = 0;
1077                                 } else {
1078                                         apic_intr_redistribute();
1079                                 }
1080                         } else
1081                                 apic_skipped_redistribute++;
1082                 }
1083         }
1084 }
1085 
1086 
1087 /*
1088  * The following functions are in the platform specific file so that they
1089  * can be different functions depending on whether we are running on
1090  * bare metal or a hypervisor.
1091  */
1092 
1093 /*
1094  * Check to make sure there are enough irq slots
1095  */
1096 int
1097 apic_check_free_irqs(int count)
1098 {
1099         int i, avail;
1100 
1101         avail = 0;
1102         for (i = APIC_FIRST_FREE_IRQ; i < APIC_RESV_IRQ; i++) {
1103                 if ((apic_irq_table[i] == NULL) ||
1104                     apic_irq_table[i]->airq_mps_intr_index == FREE_INDEX) {
1105                         if (++avail >= count)
1106                                 return (PSM_SUCCESS);
1107                 }
1108         }
1109         return (PSM_FAILURE);
1110 }
1111 
1112 /*
1113  * This function allocates "count" MSI vector(s) for the given "dip/pri/type"
1114  */
1115 int
1116 apic_alloc_msi_vectors(dev_info_t *dip, int inum, int count, int pri,
1117     int behavior)
1118 {
1119         int     rcount, i;
1120         uchar_t start, irqno;
1121         uint32_t cpu;
1122         major_t major;
1123         apic_irq_t      *irqptr;
1124 
1125         DDI_INTR_IMPLDBG((CE_CONT, "apic_alloc_msi_vectors: dip=0x%p "
1126             "inum=0x%x  pri=0x%x count=0x%x behavior=%d\n",
1127             (void *)dip, inum, pri, count, behavior));
1128 
1129         if (count > 1) {
1130                 if (behavior == DDI_INTR_ALLOC_STRICT &&
1131                     apic_multi_msi_enable == 0)
1132                         return (0);
1133                 if (apic_multi_msi_enable == 0)
1134                         count = 1;
1135         }
1136 
1137         if ((rcount = apic_navail_vector(dip, pri)) > count)
1138                 rcount = count;
1139         else if (rcount == 0 || (rcount < count &&
1140             behavior == DDI_INTR_ALLOC_STRICT))
1141                 return (0);
1142 
1143         /* if not ISP2, then round it down */
1144         if (!ISP2(rcount))
1145                 rcount = 1 << (highbit(rcount) - 1);
1146 
1147         mutex_enter(&airq_mutex);
1148 
1149         for (start = 0; rcount > 0; rcount >>= 1) {
1150                 if ((start = apic_find_multi_vectors(pri, rcount)) != 0 ||
1151                     behavior == DDI_INTR_ALLOC_STRICT)
1152                         break;
1153         }
1154 
1155         if (start == 0) {
1156                 /* no vector available */
1157                 mutex_exit(&airq_mutex);
1158                 return (0);
1159         }
1160 
1161         if (apic_check_free_irqs(rcount) == PSM_FAILURE) {
1162                 /* not enough free irq slots available */
1163                 mutex_exit(&airq_mutex);
1164                 return (0);
1165         }
1166 
1167         major = (dip != NULL) ? ddi_driver_major(dip) : 0;
1168         for (i = 0; i < rcount; i++) {
1169                 if ((irqno = apic_allocate_irq(apic_first_avail_irq)) ==
1170                     (uchar_t)-1) {
1171                         /*
1172                          * shouldn't happen because of the
1173                          * apic_check_free_irqs() check earlier
1174                          */
1175                         mutex_exit(&airq_mutex);
1176                         DDI_INTR_IMPLDBG((CE_CONT, "apic_alloc_msi_vectors: "
1177                             "apic_allocate_irq failed\n"));
1178                         return (i);
1179                 }
1180                 apic_max_device_irq = max(irqno, apic_max_device_irq);
1181                 apic_min_device_irq = min(irqno, apic_min_device_irq);
1182                 irqptr = apic_irq_table[irqno];
1183 #ifdef  DEBUG
1184                 if (apic_vector_to_irq[start + i] != APIC_RESV_IRQ)
1185                         DDI_INTR_IMPLDBG((CE_CONT, "apic_alloc_msi_vectors: "
1186                             "apic_vector_to_irq is not APIC_RESV_IRQ\n"));
1187 #endif
1188                 apic_vector_to_irq[start + i] = (uchar_t)irqno;
1189 
1190                 irqptr->airq_vector = (uchar_t)(start + i);
1191                 irqptr->airq_ioapicindex = (uchar_t)inum;    /* start */
1192                 irqptr->airq_intin_no = (uchar_t)rcount;
1193                 irqptr->airq_ipl = pri;
1194                 irqptr->airq_vector = start + i;
1195                 irqptr->airq_origirq = (uchar_t)(inum + i);
1196                 irqptr->airq_share_id = 0;
1197                 irqptr->airq_mps_intr_index = MSI_INDEX;
1198                 irqptr->airq_dip = dip;
1199                 irqptr->airq_major = major;
1200                 if (i == 0) /* they all bound to the same cpu */
1201                         cpu = irqptr->airq_cpu = apic_bind_intr(dip, irqno,
1202                             0xff, 0xff);
1203                 else
1204                         irqptr->airq_cpu = cpu;
1205                 DDI_INTR_IMPLDBG((CE_CONT, "apic_alloc_msi_vectors: irq=0x%x "
1206                     "dip=0x%p vector=0x%x origirq=0x%x pri=0x%x\n", irqno,
1207                     (void *)irqptr->airq_dip, irqptr->airq_vector,
1208                     irqptr->airq_origirq, pri));
1209         }
1210         mutex_exit(&airq_mutex);
1211         return (rcount);
1212 }
1213 
1214 /*
1215  * This function allocates "count" MSI-X vector(s) for the given "dip/pri/type"
1216  */
1217 int
1218 apic_alloc_msix_vectors(dev_info_t *dip, int inum, int count, int pri,
1219     int behavior)
1220 {
1221         int     rcount, i;
1222         major_t major;
1223 
1224         mutex_enter(&airq_mutex);
1225 
1226         if ((rcount = apic_navail_vector(dip, pri)) > count)
1227                 rcount = count;
1228         else if (rcount == 0 || (rcount < count &&
1229             behavior == DDI_INTR_ALLOC_STRICT)) {
1230                 rcount = 0;
1231                 goto out;
1232         }
1233 
1234         if (apic_check_free_irqs(rcount) == PSM_FAILURE) {
1235                 /* not enough free irq slots available */
1236                 rcount = 0;
1237                 goto out;
1238         }
1239 
1240         major = (dip != NULL) ? ddi_driver_major(dip) : 0;
1241         for (i = 0; i < rcount; i++) {
1242                 uchar_t vector, irqno;
1243                 apic_irq_t      *irqptr;
1244 
1245                 if ((irqno = apic_allocate_irq(apic_first_avail_irq)) ==
1246                     (uchar_t)-1) {
1247                         /*
1248                          * shouldn't happen because of the
1249                          * apic_check_free_irqs() check earlier
1250                          */
1251                         DDI_INTR_IMPLDBG((CE_CONT, "apic_alloc_msix_vectors: "
1252                             "apic_allocate_irq failed\n"));
1253                         rcount = i;
1254                         goto out;
1255                 }
1256                 if ((vector = apic_allocate_vector(pri, irqno, 1)) == 0) {
1257                         /*
1258                          * shouldn't happen because of the
1259                          * apic_navail_vector() call earlier
1260                          */
1261                         DDI_INTR_IMPLDBG((CE_CONT, "apic_alloc_msix_vectors: "
1262                             "apic_allocate_vector failed\n"));
1263                         rcount = i;
1264                         goto out;
1265                 }
1266                 apic_max_device_irq = max(irqno, apic_max_device_irq);
1267                 apic_min_device_irq = min(irqno, apic_min_device_irq);
1268                 irqptr = apic_irq_table[irqno];
1269                 irqptr->airq_vector = (uchar_t)vector;
1270                 irqptr->airq_ipl = pri;
1271                 irqptr->airq_origirq = (uchar_t)(inum + i);
1272                 irqptr->airq_share_id = 0;
1273                 irqptr->airq_mps_intr_index = MSIX_INDEX;
1274                 irqptr->airq_dip = dip;
1275                 irqptr->airq_major = major;
1276                 irqptr->airq_cpu = apic_bind_intr(dip, irqno, 0xff, 0xff);
1277         }
1278 out:
1279         mutex_exit(&airq_mutex);
1280         return (rcount);
1281 }
1282 
1283 /*
1284  * Allocate a free vector for irq at ipl. Takes care of merging of multiple
1285  * IPLs into a single APIC level as well as stretching some IPLs onto multiple
1286  * levels. APIC_HI_PRI_VECTS interrupts are reserved for high priority
1287  * requests and allocated only when pri is set.
1288  */
1289 uchar_t
1290 apic_allocate_vector(int ipl, int irq, int pri)
1291 {
1292         int     lowest, highest, i;
1293 
1294         highest = apic_ipltopri[ipl] + APIC_VECTOR_MASK;
1295         lowest = apic_ipltopri[ipl - 1] + APIC_VECTOR_PER_IPL;
1296 
1297         if (highest < lowest) /* Both ipl and ipl - 1 map to same pri */
1298                 lowest -= APIC_VECTOR_PER_IPL;
1299 
1300 #ifdef  DEBUG
1301         if (apic_restrict_vector)       /* for testing shared interrupt logic */
1302                 highest = lowest + apic_restrict_vector + APIC_HI_PRI_VECTS;
1303 #endif /* DEBUG */
1304         if (pri == 0)
1305                 highest -= APIC_HI_PRI_VECTS;
1306 
1307         for (i = lowest; i <= highest; i++) {
1308                 if (APIC_CHECK_RESERVE_VECTORS(i))
1309                         continue;
1310                 if (apic_vector_to_irq[i] == APIC_RESV_IRQ) {
1311                         apic_vector_to_irq[i] = (uchar_t)irq;
1312                         return (i);
1313                 }
1314         }
1315 
1316         return (0);
1317 }
1318 
1319 /* Mark vector as not being used by any irq */
1320 void
1321 apic_free_vector(uchar_t vector)
1322 {
1323         apic_vector_to_irq[vector] = APIC_RESV_IRQ;
1324 }
1325 
1326 /*
1327  * Call rebind to do the actual programming.
1328  * Must be called with interrupts disabled and apic_ioapic_lock held
1329  * 'p' is polymorphic -- if this function is called to process a deferred
1330  * reprogramming, p is of type 'struct ioapic_reprogram_data *', from which
1331  * the irq pointer is retrieved.  If not doing deferred reprogramming,
1332  * p is of the type 'apic_irq_t *'.
1333  *
1334  * apic_ioapic_lock must be held across this call, as it protects apic_rebind
1335  * and it protects apic_get_next_bind_cpu() from a race in which a CPU can be
1336  * taken offline after a cpu is selected, but before apic_rebind is called to
1337  * bind interrupts to it.
1338  */
1339 int
1340 apic_setup_io_intr(void *p, int irq, boolean_t deferred)
1341 {
1342         apic_irq_t *irqptr;
1343         struct ioapic_reprogram_data *drep = NULL;
1344         int rv;
1345 
1346         if (deferred) {
1347                 drep = (struct ioapic_reprogram_data *)p;
1348                 ASSERT(drep != NULL);
1349                 irqptr = drep->irqp;
1350         } else
1351                 irqptr = (apic_irq_t *)p;
1352 
1353         ASSERT(irqptr != NULL);
1354 
1355         rv = apic_rebind(irqptr, apic_irq_table[irq]->airq_cpu, drep);
1356         if (rv) {
1357                 /*
1358                  * CPU is not up or interrupts are disabled. Fall back to
1359                  * the first available CPU
1360                  */
1361                 rv = apic_rebind(irqptr, apic_find_cpu(APIC_CPU_INTR_ENABLE),
1362                     drep);
1363         }
1364 
1365         return (rv);
1366 }
1367 
1368 
1369 uchar_t
1370 apic_modify_vector(uchar_t vector, int irq)
1371 {
1372         apic_vector_to_irq[vector] = (uchar_t)irq;
1373         return (vector);
1374 }
1375 
1376 char *
1377 apic_get_apic_type(void)
1378 {
1379         return (apic_psm_info.p_mach_idstring);
1380 }
1381 
1382 void
1383 x2apic_update_psm(void)
1384 {
1385         struct psm_ops *pops = &apic_ops;
1386 
1387         ASSERT(pops != NULL);
1388 
1389         pops->psm_intr_exit = x2apic_intr_exit;
1390         pops->psm_setspl = x2apic_setspl;
1391 
1392         pops->psm_send_ipi =  x2apic_send_ipi;
1393         send_dirintf = pops->psm_send_ipi;
1394 
1395         apic_mode = LOCAL_X2APIC;
1396         apic_change_ops();
1397 }