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