The Design and Implementation of the FreeBSD Operating System, Second Edition
Now available: The Design and Implementation of the FreeBSD Operating System (Second Edition)


[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]

FreeBSD/Linux Kernel Cross Reference
sys/i386/i386/local_apic.c

Version: -  FREEBSD  -  FREEBSD-13-STABLE  -  FREEBSD-13-0  -  FREEBSD-12-STABLE  -  FREEBSD-12-0  -  FREEBSD-11-STABLE  -  FREEBSD-11-0  -  FREEBSD-10-STABLE  -  FREEBSD-10-0  -  FREEBSD-9-STABLE  -  FREEBSD-9-0  -  FREEBSD-8-STABLE  -  FREEBSD-8-0  -  FREEBSD-7-STABLE  -  FREEBSD-7-0  -  FREEBSD-6-STABLE  -  FREEBSD-6-0  -  FREEBSD-5-STABLE  -  FREEBSD-5-0  -  FREEBSD-4-STABLE  -  FREEBSD-3-STABLE  -  FREEBSD22  -  l41  -  OPENBSD  -  linux-2.6  -  MK84  -  PLAN9  -  xnu-8792 
SearchContext: -  none  -  3  -  10 

    1 /*-
    2  * Copyright (c) 2003 John Baldwin <jhb@FreeBSD.org>
    3  * Copyright (c) 1996, by Steve Passe
    4  * All rights reserved.
    5  *
    6  * Redistribution and use in source and binary forms, with or without
    7  * modification, are permitted provided that the following conditions
    8  * are met:
    9  * 1. Redistributions of source code must retain the above copyright
   10  *    notice, this list of conditions and the following disclaimer.
   11  * 2. The name of the developer may NOT be used to endorse or promote products
   12  *    derived from this software without specific prior written permission.
   13  * 3. Neither the name of the author nor the names of any co-contributors
   14  *    may be used to endorse or promote products derived from this software
   15  *    without specific prior written permission.
   16  *
   17  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
   18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
   19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
   20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
   21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
   22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
   23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
   24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
   25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
   26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
   27  * SUCH DAMAGE.
   28  */
   29 
   30 /*
   31  * Local APIC support on Pentium and later processors.
   32  */
   33 
   34 #include <sys/cdefs.h>
   35 __FBSDID("$FreeBSD: releng/7.4/sys/i386/i386/local_apic.c 215528 2010-11-19 16:26:51Z jhb $");
   36 
   37 #include "opt_hwpmc_hooks.h"
   38 #include "opt_kdtrace.h"
   39 
   40 #include "opt_ddb.h"
   41 
   42 #include <sys/param.h>
   43 #include <sys/systm.h>
   44 #include <sys/bus.h>
   45 #include <sys/kernel.h>
   46 #include <sys/lock.h>
   47 #include <sys/mutex.h>
   48 #include <sys/pcpu.h>
   49 #include <sys/smp.h>
   50 
   51 #include <vm/vm.h>
   52 #include <vm/pmap.h>
   53 
   54 #include <machine/apicreg.h>
   55 #include <machine/cpu.h>
   56 #include <machine/cputypes.h>
   57 #include <machine/frame.h>
   58 #include <machine/intr_machdep.h>
   59 #include <machine/apicvar.h>
   60 #include <machine/mca.h>
   61 #include <machine/md_var.h>
   62 #include <machine/smp.h>
   63 #include <machine/specialreg.h>
   64 
   65 #ifdef DDB
   66 #include <sys/interrupt.h>
   67 #include <ddb/ddb.h>
   68 #endif
   69 
   70 #ifdef KDTRACE_HOOKS
   71 #include <sys/dtrace_bsd.h>
   72 cyclic_clock_func_t     lapic_cyclic_clock_func[MAXCPU];
   73 #endif
   74 
   75 /* Sanity checks on IDT vectors. */
   76 CTASSERT(APIC_IO_INTS + APIC_NUM_IOINTS == APIC_TIMER_INT);
   77 CTASSERT(APIC_TIMER_INT < APIC_LOCAL_INTS);
   78 CTASSERT(APIC_LOCAL_INTS == 240);
   79 CTASSERT(IPI_STOP < APIC_SPURIOUS_INT);
   80 
   81 /* Magic IRQ values for the timer and syscalls. */
   82 #define IRQ_TIMER       (NUM_IO_INTS + 1)
   83 #define IRQ_SYSCALL     (NUM_IO_INTS + 2)
   84 
   85 /*
   86  * Support for local APICs.  Local APICs manage interrupts on each
   87  * individual processor as opposed to I/O APICs which receive interrupts
   88  * from I/O devices and then forward them on to the local APICs.
   89  *
   90  * Local APICs can also send interrupts to each other thus providing the
   91  * mechanism for IPIs.
   92  */
   93 
   94 struct lvt {
   95         u_int lvt_edgetrigger:1;
   96         u_int lvt_activehi:1;
   97         u_int lvt_masked:1;
   98         u_int lvt_active:1;
   99         u_int lvt_mode:16;
  100         u_int lvt_vector:8;
  101 };
  102 
  103 struct lapic {
  104         struct lvt la_lvts[LVT_MAX + 1];
  105         u_int la_id:8;
  106         u_int la_cluster:4;
  107         u_int la_cluster_id:2;
  108         u_int la_present:1;
  109         u_long *la_timer_count;
  110         u_long la_hard_ticks;
  111         u_long la_stat_ticks;
  112         u_long la_prof_ticks;
  113 } static lapics[MAX_APIC_ID + 1];
  114 
  115 /* Global defaults for local APIC LVT entries. */
  116 static struct lvt lvts[LVT_MAX + 1] = {
  117         { 1, 1, 1, 1, APIC_LVT_DM_EXTINT, 0 },  /* LINT0: masked ExtINT */
  118         { 1, 1, 0, 1, APIC_LVT_DM_NMI, 0 },     /* LINT1: NMI */
  119         { 1, 1, 1, 1, APIC_LVT_DM_FIXED, APIC_TIMER_INT },      /* Timer */
  120         { 1, 1, 0, 1, APIC_LVT_DM_FIXED, APIC_ERROR_INT },      /* Error */
  121         { 1, 1, 1, 1, APIC_LVT_DM_NMI, 0 },     /* PMC */
  122         { 1, 1, 1, 1, APIC_LVT_DM_FIXED, APIC_THERMAL_INT },    /* Thermal */
  123         { 1, 1, 1, 1, APIC_LVT_DM_FIXED, APIC_CMC_INT },        /* CMCI */
  124 };
  125 
  126 static inthand_t *ioint_handlers[] = {
  127         NULL,                   /* 0 - 31 */
  128         IDTVEC(apic_isr1),      /* 32 - 63 */
  129         IDTVEC(apic_isr2),      /* 64 - 95 */
  130         IDTVEC(apic_isr3),      /* 96 - 127 */
  131         IDTVEC(apic_isr4),      /* 128 - 159 */
  132         IDTVEC(apic_isr5),      /* 160 - 191 */
  133         IDTVEC(apic_isr6),      /* 192 - 223 */
  134         IDTVEC(apic_isr7),      /* 224 - 255 */
  135 };
  136 
  137 /* Include IDT_SYSCALL to make indexing easier. */
  138 static u_int ioint_irqs[APIC_NUM_IOINTS + 1];
  139 
  140 static u_int32_t lapic_timer_divisors[] = { 
  141         APIC_TDCR_1, APIC_TDCR_2, APIC_TDCR_4, APIC_TDCR_8, APIC_TDCR_16,
  142         APIC_TDCR_32, APIC_TDCR_64, APIC_TDCR_128
  143 };
  144 
  145 extern inthand_t IDTVEC(rsvd);
  146 
  147 volatile lapic_t *lapic;
  148 vm_paddr_t lapic_paddr;
  149 static u_long lapic_timer_divisor, lapic_timer_period, lapic_timer_hz;
  150 
  151 static void     lapic_enable(void);
  152 static void     lapic_resume(struct pic *pic);
  153 static void     lapic_timer_enable_intr(void);
  154 static void     lapic_timer_oneshot(u_int count);
  155 static void     lapic_timer_periodic(u_int count);
  156 static void     lapic_timer_set_divisor(u_int divisor);
  157 static uint32_t lvt_mode(struct lapic *la, u_int pin, uint32_t value);
  158 
  159 struct pic lapic_pic = { .pic_resume = lapic_resume };
  160 
  161 static uint32_t
  162 lvt_mode(struct lapic *la, u_int pin, uint32_t value)
  163 {
  164         struct lvt *lvt;
  165 
  166         KASSERT(pin <= LVT_MAX, ("%s: pin %u out of range", __func__, pin));
  167         if (la->la_lvts[pin].lvt_active)
  168                 lvt = &la->la_lvts[pin];
  169         else
  170                 lvt = &lvts[pin];
  171 
  172         value &= ~(APIC_LVT_M | APIC_LVT_TM | APIC_LVT_IIPP | APIC_LVT_DM |
  173             APIC_LVT_VECTOR);
  174         if (lvt->lvt_edgetrigger == 0)
  175                 value |= APIC_LVT_TM;
  176         if (lvt->lvt_activehi == 0)
  177                 value |= APIC_LVT_IIPP_INTALO;
  178         if (lvt->lvt_masked)
  179                 value |= APIC_LVT_M;
  180         value |= lvt->lvt_mode;
  181         switch (lvt->lvt_mode) {
  182         case APIC_LVT_DM_NMI:
  183         case APIC_LVT_DM_SMI:
  184         case APIC_LVT_DM_INIT:
  185         case APIC_LVT_DM_EXTINT:
  186                 if (!lvt->lvt_edgetrigger) {
  187                         printf("lapic%u: Forcing LINT%u to edge trigger\n",
  188                             la->la_id, pin);
  189                         value |= APIC_LVT_TM;
  190                 }
  191                 /* Use a vector of 0. */
  192                 break;
  193         case APIC_LVT_DM_FIXED:
  194                 value |= lvt->lvt_vector;
  195                 break;
  196         default:
  197                 panic("bad APIC LVT delivery mode: %#x\n", value);
  198         }
  199         return (value);
  200 }
  201 
  202 /*
  203  * Map the local APIC and setup necessary interrupt vectors.
  204  */
  205 void
  206 lapic_init(vm_paddr_t addr)
  207 {
  208 
  209         /* Map the local APIC and setup the spurious interrupt handler. */
  210         KASSERT(trunc_page(addr) == addr,
  211             ("local APIC not aligned on a page boundary"));
  212         lapic = pmap_mapdev(addr, sizeof(lapic_t));
  213         lapic_paddr = addr;
  214         setidt(APIC_SPURIOUS_INT, IDTVEC(spuriousint), SDT_SYS386IGT, SEL_KPL,
  215             GSEL(GCODE_SEL, SEL_KPL));
  216 
  217         /* Perform basic initialization of the BSP's local APIC. */
  218         lapic_enable();
  219         ioint_irqs[IDT_SYSCALL - APIC_IO_INTS] = IRQ_SYSCALL;
  220 
  221         /* Set BSP's per-CPU local APIC ID. */
  222         PCPU_SET(apic_id, lapic_id());
  223 
  224         /* Local APIC timer interrupt. */
  225         setidt(APIC_TIMER_INT, IDTVEC(timerint), SDT_SYS386IGT, SEL_KPL,
  226             GSEL(GCODE_SEL, SEL_KPL));
  227         ioint_irqs[APIC_TIMER_INT - APIC_IO_INTS] = IRQ_TIMER;
  228 
  229         /* Local APIC error interrupt. */
  230         setidt(APIC_ERROR_INT, IDTVEC(errorint), SDT_SYS386IGT, SEL_KPL,
  231             GSEL(GCODE_SEL, SEL_KPL));
  232 
  233         /* XXX: Thermal interrupt */
  234 
  235         /* Local APIC CMCI. */
  236         setidt(APIC_CMC_INT, IDTVEC(cmcint), SDT_SYS386TGT, SEL_KPL,
  237             GSEL(GCODE_SEL, SEL_KPL));
  238 }
  239 
  240 /*
  241  * Create a local APIC instance.
  242  */
  243 void
  244 lapic_create(u_int apic_id, int boot_cpu)
  245 {
  246         int i;
  247 
  248         if (apic_id > MAX_APIC_ID) {
  249                 printf("APIC: Ignoring local APIC with ID %d\n", apic_id);
  250                 if (boot_cpu)
  251                         panic("Can't ignore BSP");
  252                 return;
  253         }
  254         KASSERT(!lapics[apic_id].la_present, ("duplicate local APIC %u",
  255             apic_id));
  256 
  257         /*
  258          * Assume no local LVT overrides and a cluster of 0 and
  259          * intra-cluster ID of 0.
  260          */
  261         lapics[apic_id].la_present = 1;
  262         lapics[apic_id].la_id = apic_id;
  263         for (i = 0; i <= LVT_MAX; i++) {
  264                 lapics[apic_id].la_lvts[i] = lvts[i];
  265                 lapics[apic_id].la_lvts[i].lvt_active = 0;
  266         }
  267 
  268 #ifdef SMP
  269         cpu_add(apic_id, boot_cpu);
  270 #endif
  271 }
  272 
  273 /*
  274  * Dump contents of local APIC registers
  275  */
  276 void
  277 lapic_dump(const char* str)
  278 {
  279         uint32_t maxlvt;
  280 
  281         maxlvt = (lapic->version & APIC_VER_MAXLVT) >> MAXLVTSHIFT;
  282         printf("cpu%d %s:\n", PCPU_GET(cpuid), str);
  283         printf("     ID: 0x%08x   VER: 0x%08x LDR: 0x%08x DFR: 0x%08x\n",
  284             lapic->id, lapic->version, lapic->ldr, lapic->dfr);
  285         printf("  lint0: 0x%08x lint1: 0x%08x TPR: 0x%08x SVR: 0x%08x\n",
  286             lapic->lvt_lint0, lapic->lvt_lint1, lapic->tpr, lapic->svr);
  287         printf("  timer: 0x%08x therm: 0x%08x err: 0x%08x",
  288             lapic->lvt_timer, lapic->lvt_thermal, lapic->lvt_error);
  289         if (maxlvt >= LVT_PMC)
  290                 printf(" pmc: 0x%08x", lapic->lvt_pcint);
  291         printf("\n");
  292         if (maxlvt >= LVT_CMCI)
  293                 printf("   cmci: 0x%08x\n", lapic->lvt_cmci);
  294 }
  295 
  296 void
  297 lapic_setup(int boot)
  298 {
  299         struct lapic *la;
  300         u_int32_t maxlvt;
  301         register_t eflags;
  302         char buf[MAXCOMLEN + 1];
  303 
  304         la = &lapics[lapic_id()];
  305         KASSERT(la->la_present, ("missing APIC structure"));
  306         eflags = intr_disable();
  307         maxlvt = (lapic->version & APIC_VER_MAXLVT) >> MAXLVTSHIFT;
  308 
  309         /* Initialize the TPR to allow all interrupts. */
  310         lapic_set_tpr(0);
  311 
  312         /* Setup spurious vector and enable the local APIC. */
  313         lapic_enable();
  314 
  315         /* Program LINT[01] LVT entries. */
  316         lapic->lvt_lint0 = lvt_mode(la, LVT_LINT0, lapic->lvt_lint0);
  317         lapic->lvt_lint1 = lvt_mode(la, LVT_LINT1, lapic->lvt_lint1);
  318 
  319         /* Program the PMC LVT entry if present. */
  320         if (maxlvt >= LVT_PMC)
  321                 lapic->lvt_pcint = lvt_mode(la, LVT_PMC, lapic->lvt_pcint);
  322 
  323         /* Program timer LVT and setup handler. */
  324         lapic->lvt_timer = lvt_mode(la, LVT_TIMER, lapic->lvt_timer);
  325         if (boot) {
  326                 snprintf(buf, sizeof(buf), "cpu%d: timer", PCPU_GET(cpuid));
  327                 intrcnt_add(buf, &la->la_timer_count);
  328         }
  329 
  330         /* We don't setup the timer during boot on the BSP until later. */
  331         if (!(boot && PCPU_GET(cpuid) == 0)) {
  332                 KASSERT(lapic_timer_period != 0, ("lapic%u: zero divisor",
  333                     lapic_id()));
  334                 lapic_timer_set_divisor(lapic_timer_divisor);
  335                 lapic_timer_periodic(lapic_timer_period);
  336                 lapic_timer_enable_intr();
  337         }
  338 
  339         /* Program error LVT and clear any existing errors. */
  340         lapic->lvt_error = lvt_mode(la, LVT_ERROR, lapic->lvt_error);
  341         lapic->esr = 0;
  342 
  343         /* XXX: Thermal LVT */
  344 
  345         if (cpu_vendor_id == CPU_VENDOR_AMD) {
  346                 /*
  347                  * Detect the presence of C1E capability mostly on latest
  348                  * dual-cores (or future) k8 family.  This feature renders
  349                  * the local APIC timer dead, so we disable it by reading
  350                  * the Interrupt Pending Message register and clearing both
  351                  * C1eOnCmpHalt (bit 28) and SmiOnCmpHalt (bit 27).
  352                  * 
  353                  * Reference:
  354                  *   "BIOS and Kernel Developer's Guide for AMD NPT
  355                  *    Family 0Fh Processors"
  356                  *   #32559 revision 3.00
  357                  */
  358                 if ((cpu_id & 0x00000f00) == 0x00000f00 &&
  359                     (cpu_id & 0x0fff0000) >=  0x00040000) {
  360                         uint64_t msr;
  361 
  362                         msr = rdmsr(0xc0010055);
  363                         if (msr & 0x18000000)
  364                                 wrmsr(0xc0010055, msr & ~0x18000000ULL);
  365                 }
  366         }
  367 
  368         /* Program the CMCI LVT entry if present. */
  369         if (maxlvt >= LVT_CMCI)
  370                 lapic->lvt_cmci = lvt_mode(la, LVT_CMCI, lapic->lvt_cmci);
  371             
  372         intr_restore(eflags);
  373 }
  374 
  375 void
  376 lapic_reenable_pmc(void)
  377 {
  378 #ifdef HWPMC_HOOKS
  379         uint32_t value;
  380 
  381         value =  lapic->lvt_pcint;
  382         value &= ~APIC_LVT_M;
  383         lapic->lvt_pcint = value;
  384 #endif
  385 }
  386 
  387 #ifdef HWPMC_HOOKS
  388 static void
  389 lapic_update_pmc(void *dummy)
  390 {
  391         struct lapic *la;
  392 
  393         la = &lapics[lapic_id()];
  394         lapic->lvt_pcint = lvt_mode(la, LVT_PMC, lapic->lvt_pcint);
  395 }
  396 #endif
  397 
  398 int
  399 lapic_enable_pmc(void)
  400 {
  401 #ifdef HWPMC_HOOKS
  402         u_int32_t maxlvt;
  403 
  404         /* Fail if the local APIC is not present. */
  405         if (lapic == NULL)
  406                 return (0);
  407 
  408         /* Fail if the PMC LVT is not present. */
  409         maxlvt = (lapic->version & APIC_VER_MAXLVT) >> MAXLVTSHIFT;
  410         if (maxlvt < LVT_PMC)
  411                 return (0);
  412 
  413         lvts[LVT_PMC].lvt_masked = 0;
  414 
  415 #ifdef SMP
  416         /*
  417          * If hwpmc was loaded at boot time then the APs may not be
  418          * started yet.  In that case, don't forward the request to
  419          * them as they will program the lvt when they start.
  420          */
  421         if (smp_started)
  422                 smp_rendezvous(NULL, lapic_update_pmc, NULL, NULL);
  423         else
  424 #endif
  425                 lapic_update_pmc(NULL);
  426         return (1);
  427 #else
  428         return (0);
  429 #endif
  430 }
  431 
  432 void
  433 lapic_disable_pmc(void)
  434 {
  435 #ifdef HWPMC_HOOKS
  436         u_int32_t maxlvt;
  437 
  438         /* Fail if the local APIC is not present. */
  439         if (lapic == NULL)
  440                 return;
  441 
  442         /* Fail if the PMC LVT is not present. */
  443         maxlvt = (lapic->version & APIC_VER_MAXLVT) >> MAXLVTSHIFT;
  444         if (maxlvt < LVT_PMC)
  445                 return;
  446 
  447         lvts[LVT_PMC].lvt_masked = 1;
  448 
  449 #ifdef SMP
  450         /* The APs should always be started when hwpmc is unloaded. */
  451         KASSERT(mp_ncpus == 1 || smp_started, ("hwpmc unloaded too early"));
  452 #endif
  453         smp_rendezvous(NULL, lapic_update_pmc, NULL, NULL);
  454 #endif
  455 }
  456 
  457 /*
  458  * Called by cpu_initclocks() on the BSP to setup the local APIC timer so
  459  * that it can drive hardclock, statclock, and profclock.  This function
  460  * returns true if it is able to use the local APIC timer to drive the
  461  * clocks and false if it is not able.
  462  */
  463 int
  464 lapic_setup_clock(void)
  465 {
  466         u_long value;
  467 
  468         /* Can't drive the timer without a local APIC. */
  469         if (lapic == NULL)
  470                 return (0);
  471 
  472         /* Start off with a divisor of 2 (power on reset default). */
  473         lapic_timer_divisor = 2;
  474 
  475         /* Try to calibrate the local APIC timer. */
  476         do {
  477                 lapic_timer_set_divisor(lapic_timer_divisor);
  478                 lapic_timer_oneshot(APIC_TIMER_MAX_COUNT);
  479                 DELAY(2000000);
  480                 value = APIC_TIMER_MAX_COUNT - lapic->ccr_timer;
  481                 if (value != APIC_TIMER_MAX_COUNT)
  482                         break;
  483                 lapic_timer_divisor <<= 1;
  484         } while (lapic_timer_divisor <= 128);
  485         if (lapic_timer_divisor > 128)
  486                 panic("lapic: Divisor too big");
  487         value /= 2;
  488         if (bootverbose)
  489                 printf("lapic: Divisor %lu, Frequency %lu Hz\n",
  490                     lapic_timer_divisor, value);
  491 
  492         /*
  493          * We want to run stathz in the neighborhood of 128hz.  We would
  494          * like profhz to run as often as possible, so we let it run on
  495          * each clock tick.  We try to honor the requested 'hz' value as
  496          * much as possible.
  497          *
  498          * If 'hz' is above 1500, then we just let the lapic timer
  499          * (and profhz) run at hz.  If 'hz' is below 1500 but above
  500          * 750, then we let the lapic timer run at 2 * 'hz'.  If 'hz'
  501          * is below 750 then we let the lapic timer run at 4 * 'hz'.
  502          */
  503         if (hz >= 1500)
  504                 lapic_timer_hz = hz;
  505         else if (hz >= 750)
  506                 lapic_timer_hz = hz * 2;
  507         else
  508                 lapic_timer_hz = hz * 4;
  509         if (lapic_timer_hz < 128)
  510                 stathz = lapic_timer_hz;
  511         else
  512                 stathz = lapic_timer_hz / (lapic_timer_hz / 128);
  513         profhz = lapic_timer_hz;
  514         lapic_timer_period = value / lapic_timer_hz;
  515 
  516         /*
  517          * Start up the timer on the BSP.  The APs will kick off their
  518          * timer during lapic_setup().
  519          */
  520         lapic_timer_periodic(lapic_timer_period);
  521         lapic_timer_enable_intr();
  522         return (1);
  523 }
  524 
  525 void
  526 lapic_disable(void)
  527 {
  528         uint32_t value;
  529 
  530         /* Software disable the local APIC. */
  531         value = lapic->svr;
  532         value &= ~APIC_SVR_SWEN;
  533         lapic->svr = value;
  534 }
  535 
  536 static void
  537 lapic_enable(void)
  538 {
  539         u_int32_t value;
  540 
  541         /* Program the spurious vector to enable the local APIC. */
  542         value = lapic->svr;
  543         value &= ~(APIC_SVR_VECTOR | APIC_SVR_FOCUS);
  544         value |= (APIC_SVR_FEN | APIC_SVR_SWEN | APIC_SPURIOUS_INT);
  545         lapic->svr = value;
  546 }
  547 
  548 /* Reset the local APIC on the BSP during resume. */
  549 static void
  550 lapic_resume(struct pic *pic)
  551 {
  552 
  553         lapic_setup(0);
  554 }
  555 
  556 int
  557 lapic_id(void)
  558 {
  559 
  560         KASSERT(lapic != NULL, ("local APIC is not mapped"));
  561         return (lapic->id >> APIC_ID_SHIFT);
  562 }
  563 
  564 int
  565 lapic_intr_pending(u_int vector)
  566 {
  567         volatile u_int32_t *irr;
  568 
  569         /*
  570          * The IRR registers are an array of 128-bit registers each of
  571          * which only describes 32 interrupts in the low 32 bits..  Thus,
  572          * we divide the vector by 32 to get the 128-bit index.  We then
  573          * multiply that index by 4 to get the equivalent index from
  574          * treating the IRR as an array of 32-bit registers.  Finally, we
  575          * modulus the vector by 32 to determine the individual bit to
  576          * test.
  577          */
  578         irr = &lapic->irr0;
  579         return (irr[(vector / 32) * 4] & 1 << (vector % 32));
  580 }
  581 
  582 void
  583 lapic_set_logical_id(u_int apic_id, u_int cluster, u_int cluster_id)
  584 {
  585         struct lapic *la;
  586 
  587         KASSERT(lapics[apic_id].la_present, ("%s: APIC %u doesn't exist",
  588             __func__, apic_id));
  589         KASSERT(cluster <= APIC_MAX_CLUSTER, ("%s: cluster %u too big",
  590             __func__, cluster));
  591         KASSERT(cluster_id <= APIC_MAX_INTRACLUSTER_ID,
  592             ("%s: intra cluster id %u too big", __func__, cluster_id));
  593         la = &lapics[apic_id];
  594         la->la_cluster = cluster;
  595         la->la_cluster_id = cluster_id;
  596 }
  597 
  598 int
  599 lapic_set_lvt_mask(u_int apic_id, u_int pin, u_char masked)
  600 {
  601 
  602         if (pin > LVT_MAX)
  603                 return (EINVAL);
  604         if (apic_id == APIC_ID_ALL) {
  605                 lvts[pin].lvt_masked = masked;
  606                 if (bootverbose)
  607                         printf("lapic:");
  608         } else {
  609                 KASSERT(lapics[apic_id].la_present,
  610                     ("%s: missing APIC %u", __func__, apic_id));
  611                 lapics[apic_id].la_lvts[pin].lvt_masked = masked;
  612                 lapics[apic_id].la_lvts[pin].lvt_active = 1;
  613                 if (bootverbose)
  614                         printf("lapic%u:", apic_id);
  615         }
  616         if (bootverbose)
  617                 printf(" LINT%u %s\n", pin, masked ? "masked" : "unmasked");
  618         return (0);
  619 }
  620 
  621 int
  622 lapic_set_lvt_mode(u_int apic_id, u_int pin, u_int32_t mode)
  623 {
  624         struct lvt *lvt;
  625 
  626         if (pin > LVT_MAX)
  627                 return (EINVAL);
  628         if (apic_id == APIC_ID_ALL) {
  629                 lvt = &lvts[pin];
  630                 if (bootverbose)
  631                         printf("lapic:");
  632         } else {
  633                 KASSERT(lapics[apic_id].la_present,
  634                     ("%s: missing APIC %u", __func__, apic_id));
  635                 lvt = &lapics[apic_id].la_lvts[pin];
  636                 lvt->lvt_active = 1;
  637                 if (bootverbose)
  638                         printf("lapic%u:", apic_id);
  639         }
  640         lvt->lvt_mode = mode;
  641         switch (mode) {
  642         case APIC_LVT_DM_NMI:
  643         case APIC_LVT_DM_SMI:
  644         case APIC_LVT_DM_INIT:
  645         case APIC_LVT_DM_EXTINT:
  646                 lvt->lvt_edgetrigger = 1;
  647                 lvt->lvt_activehi = 1;
  648                 if (mode == APIC_LVT_DM_EXTINT)
  649                         lvt->lvt_masked = 1;
  650                 else
  651                         lvt->lvt_masked = 0;
  652                 break;
  653         default:
  654                 panic("Unsupported delivery mode: 0x%x\n", mode);
  655         }
  656         if (bootverbose) {
  657                 printf(" Routing ");
  658                 switch (mode) {
  659                 case APIC_LVT_DM_NMI:
  660                         printf("NMI");
  661                         break;
  662                 case APIC_LVT_DM_SMI:
  663                         printf("SMI");
  664                         break;
  665                 case APIC_LVT_DM_INIT:
  666                         printf("INIT");
  667                         break;
  668                 case APIC_LVT_DM_EXTINT:
  669                         printf("ExtINT");
  670                         break;
  671                 }
  672                 printf(" -> LINT%u\n", pin);
  673         }
  674         return (0);
  675 }
  676 
  677 int
  678 lapic_set_lvt_polarity(u_int apic_id, u_int pin, enum intr_polarity pol)
  679 {
  680 
  681         if (pin > LVT_MAX || pol == INTR_POLARITY_CONFORM)
  682                 return (EINVAL);
  683         if (apic_id == APIC_ID_ALL) {
  684                 lvts[pin].lvt_activehi = (pol == INTR_POLARITY_HIGH);
  685                 if (bootverbose)
  686                         printf("lapic:");
  687         } else {
  688                 KASSERT(lapics[apic_id].la_present,
  689                     ("%s: missing APIC %u", __func__, apic_id));
  690                 lapics[apic_id].la_lvts[pin].lvt_active = 1;
  691                 lapics[apic_id].la_lvts[pin].lvt_activehi =
  692                     (pol == INTR_POLARITY_HIGH);
  693                 if (bootverbose)
  694                         printf("lapic%u:", apic_id);
  695         }
  696         if (bootverbose)
  697                 printf(" LINT%u polarity: %s\n", pin,
  698                     pol == INTR_POLARITY_HIGH ? "high" : "low");
  699         return (0);
  700 }
  701 
  702 int
  703 lapic_set_lvt_triggermode(u_int apic_id, u_int pin, enum intr_trigger trigger)
  704 {
  705 
  706         if (pin > LVT_MAX || trigger == INTR_TRIGGER_CONFORM)
  707                 return (EINVAL);
  708         if (apic_id == APIC_ID_ALL) {
  709                 lvts[pin].lvt_edgetrigger = (trigger == INTR_TRIGGER_EDGE);
  710                 if (bootverbose)
  711                         printf("lapic:");
  712         } else {
  713                 KASSERT(lapics[apic_id].la_present,
  714                     ("%s: missing APIC %u", __func__, apic_id));
  715                 lapics[apic_id].la_lvts[pin].lvt_edgetrigger =
  716                     (trigger == INTR_TRIGGER_EDGE);
  717                 lapics[apic_id].la_lvts[pin].lvt_active = 1;
  718                 if (bootverbose)
  719                         printf("lapic%u:", apic_id);
  720         }
  721         if (bootverbose)
  722                 printf(" LINT%u trigger: %s\n", pin,
  723                     trigger == INTR_TRIGGER_EDGE ? "edge" : "level");
  724         return (0);
  725 }
  726 
  727 /*
  728  * Adjust the TPR of the current CPU so that it blocks all interrupts below
  729  * the passed in vector.
  730  */
  731 void
  732 lapic_set_tpr(u_int vector)
  733 {
  734 #ifdef CHEAP_TPR
  735         lapic->tpr = vector;
  736 #else
  737         u_int32_t tpr;
  738 
  739         tpr = lapic->tpr & ~APIC_TPR_PRIO;
  740         tpr |= vector;
  741         lapic->tpr = tpr;
  742 #endif
  743 }
  744 
  745 void
  746 lapic_eoi(void)
  747 {
  748 
  749         lapic->eoi = 0;
  750 }
  751 
  752 void
  753 lapic_handle_intr(int vector, struct trapframe *frame)
  754 {
  755         struct intsrc *isrc;
  756 
  757         if (vector == -1)
  758                 panic("Couldn't get vector from ISR!");
  759         isrc = intr_lookup_source(apic_idt_to_irq(vector));
  760         intr_execute_handlers(isrc, frame);
  761 }
  762 
  763 void
  764 lapic_handle_timer(struct trapframe *frame)
  765 {
  766         struct lapic *la;
  767 
  768         /* Send EOI first thing. */
  769         lapic_eoi();
  770 
  771 #if defined(SMP) && !defined(SCHED_ULE)
  772         /*
  773          * Don't do any accounting for the disabled HTT cores, since it
  774          * will provide misleading numbers for the userland.
  775          *
  776          * No locking is necessary here, since even if we loose the race
  777          * when hlt_cpus_mask changes it is not a big deal, really.
  778          *
  779          * Don't do that for ULE, since ULE doesn't consider hlt_cpus_mask
  780          * and unlike other schedulers it actually schedules threads to
  781          * those CPUs.
  782          */
  783         if ((hlt_cpus_mask & (1 << PCPU_GET(cpuid))) != 0)
  784                 return;
  785 #endif
  786 
  787         /* Look up our local APIC structure for the tick counters. */
  788         la = &lapics[PCPU_GET(apic_id)];
  789         (*la->la_timer_count)++;
  790         critical_enter();
  791 
  792 #ifdef KDTRACE_HOOKS
  793         /*
  794          * If the DTrace hooks are configured and a callback function
  795          * has been registered, then call it to process the high speed
  796          * timers.
  797          */
  798         int cpu = PCPU_GET(cpuid);
  799         if (lapic_cyclic_clock_func[cpu] != NULL)
  800                 (*lapic_cyclic_clock_func[cpu])(frame);
  801 #endif
  802 
  803         /* Fire hardclock at hz. */
  804         la->la_hard_ticks += hz;
  805         if (la->la_hard_ticks >= lapic_timer_hz) {
  806                 la->la_hard_ticks -= lapic_timer_hz;
  807                 if (PCPU_GET(cpuid) == 0)
  808                         hardclock(TRAPF_USERMODE(frame), TRAPF_PC(frame));
  809                 else
  810                         hardclock_cpu(TRAPF_USERMODE(frame));
  811         }
  812 
  813         /* Fire statclock at stathz. */
  814         la->la_stat_ticks += stathz;
  815         if (la->la_stat_ticks >= lapic_timer_hz) {
  816                 la->la_stat_ticks -= lapic_timer_hz;
  817                 statclock(TRAPF_USERMODE(frame));
  818         }
  819 
  820         /* Fire profclock at profhz, but only when needed. */
  821         la->la_prof_ticks += profhz;
  822         if (la->la_prof_ticks >= lapic_timer_hz) {
  823                 la->la_prof_ticks -= lapic_timer_hz;
  824                 if (profprocs != 0)
  825                         profclock(TRAPF_USERMODE(frame), TRAPF_PC(frame));
  826         }
  827         critical_exit();
  828 }
  829 
  830 static void
  831 lapic_timer_set_divisor(u_int divisor)
  832 {
  833 
  834         KASSERT(powerof2(divisor), ("lapic: invalid divisor %u", divisor));
  835         KASSERT(ffs(divisor) <= sizeof(lapic_timer_divisors) /
  836             sizeof(u_int32_t), ("lapic: invalid divisor %u", divisor));
  837         lapic->dcr_timer = lapic_timer_divisors[ffs(divisor) - 1];
  838 }
  839 
  840 static void
  841 lapic_timer_oneshot(u_int count)
  842 {
  843         u_int32_t value;
  844 
  845         value = lapic->lvt_timer;
  846         value &= ~APIC_LVTT_TM;
  847         value |= APIC_LVTT_TM_ONE_SHOT;
  848         lapic->lvt_timer = value;
  849         lapic->icr_timer = count;
  850 }
  851 
  852 static void
  853 lapic_timer_periodic(u_int count)
  854 {
  855         u_int32_t value;
  856 
  857         value = lapic->lvt_timer;
  858         value &= ~APIC_LVTT_TM;
  859         value |= APIC_LVTT_TM_PERIODIC;
  860         lapic->lvt_timer = value;
  861         lapic->icr_timer = count;
  862 }
  863 
  864 static void
  865 lapic_timer_enable_intr(void)
  866 {
  867         u_int32_t value;
  868 
  869         value = lapic->lvt_timer;
  870         value &= ~APIC_LVT_M;
  871         lapic->lvt_timer = value;
  872 }
  873 
  874 void
  875 lapic_handle_cmc(void)
  876 {
  877 
  878         lapic_eoi();
  879         cmc_intr();
  880 }
  881 
  882 /*
  883  * Called from the mca_init() to activate the CMC interrupt if this CPU is
  884  * responsible for monitoring any MC banks for CMC events.  Since mca_init()
  885  * is called prior to lapic_setup() during boot, this just needs to unmask
  886  * this CPU's LVT_CMCI entry.
  887  */
  888 void
  889 lapic_enable_cmc(void)
  890 {
  891         u_int apic_id;
  892 
  893         apic_id = PCPU_GET(apic_id);
  894         KASSERT(lapics[apic_id].la_present,
  895             ("%s: missing APIC %u", __func__, apic_id));
  896         lapics[apic_id].la_lvts[LVT_CMCI].lvt_masked = 0;
  897         lapics[apic_id].la_lvts[LVT_CMCI].lvt_active = 1;
  898         if (bootverbose)
  899                 printf("lapic%u: CMCI unmasked\n", apic_id);
  900 }
  901 
  902 void
  903 lapic_handle_error(void)
  904 {
  905         u_int32_t esr;
  906 
  907         /*
  908          * Read the contents of the error status register.  Write to
  909          * the register first before reading from it to force the APIC
  910          * to update its value to indicate any errors that have
  911          * occurred since the previous write to the register.
  912          */
  913         lapic->esr = 0;
  914         esr = lapic->esr;
  915 
  916         printf("CPU%d: local APIC error 0x%x\n", PCPU_GET(cpuid), esr);
  917         lapic_eoi();
  918 }
  919 
  920 /* Request a free IDT vector to be used by the specified IRQ. */
  921 u_int
  922 apic_alloc_vector(u_int irq)
  923 {
  924         u_int vector;
  925 
  926         KASSERT(irq < NUM_IO_INTS, ("Invalid IRQ %u", irq));
  927 
  928         /*
  929          * Search for a free vector.  Currently we just use a very simple
  930          * algorithm to find the first free vector.
  931          */
  932         mtx_lock_spin(&icu_lock);
  933         for (vector = 0; vector < APIC_NUM_IOINTS; vector++) {
  934                 if (ioint_irqs[vector] != 0)
  935                         continue;
  936                 ioint_irqs[vector] = irq;
  937                 mtx_unlock_spin(&icu_lock);
  938                 return (vector + APIC_IO_INTS);
  939         }
  940         mtx_unlock_spin(&icu_lock);
  941         panic("Couldn't find an APIC vector for IRQ %u", irq);
  942 }
  943 
  944 /*
  945  * Request 'count' free contiguous IDT vectors to be used by 'count'
  946  * IRQs.  'count' must be a power of two and the vectors will be
  947  * aligned on a boundary of 'align'.  If the request cannot be
  948  * satisfied, 0 is returned.
  949  */
  950 u_int
  951 apic_alloc_vectors(u_int *irqs, u_int count, u_int align)
  952 {
  953         u_int first, run, vector;
  954 
  955         KASSERT(powerof2(count), ("bad count"));
  956         KASSERT(powerof2(align), ("bad align"));
  957         KASSERT(align >= count, ("align < count"));
  958 #ifdef INVARIANTS
  959         for (run = 0; run < count; run++)
  960                 KASSERT(irqs[run] < NUM_IO_INTS, ("Invalid IRQ %u at index %u",
  961                     irqs[run], run));
  962 #endif
  963 
  964         /*
  965          * Search for 'count' free vectors.  As with apic_alloc_vector(),
  966          * this just uses a simple first fit algorithm.
  967          */
  968         run = 0;
  969         first = 0;
  970         mtx_lock_spin(&icu_lock);
  971         for (vector = 0; vector < APIC_NUM_IOINTS; vector++) {
  972 
  973                 /* Vector is in use, end run. */
  974                 if (ioint_irqs[vector] != 0) {
  975                         run = 0;
  976                         first = 0;
  977                         continue;
  978                 }
  979 
  980                 /* Start a new run if run == 0 and vector is aligned. */
  981                 if (run == 0) {
  982                         if ((vector & (align - 1)) != 0)
  983                                 continue;
  984                         first = vector;
  985                 }
  986                 run++;
  987 
  988                 /* Keep looping if the run isn't long enough yet. */
  989                 if (run < count)
  990                         continue;
  991 
  992                 /* Found a run, assign IRQs and return the first vector. */
  993                 for (vector = 0; vector < count; vector++)
  994                         ioint_irqs[first + vector] = irqs[vector];
  995                 mtx_unlock_spin(&icu_lock);
  996                 return (first + APIC_IO_INTS);
  997         }
  998         mtx_unlock_spin(&icu_lock);
  999         printf("APIC: Couldn't find APIC vectors for %u IRQs\n", count);
 1000         return (0);
 1001 }
 1002 
 1003 void
 1004 apic_enable_vector(u_int vector)
 1005 {
 1006 
 1007         KASSERT(vector != IDT_SYSCALL, ("Attempt to overwrite syscall entry"));
 1008         KASSERT(ioint_handlers[vector / 32] != NULL,
 1009             ("No ISR handler for vector %u", vector));
 1010         setidt(vector, ioint_handlers[vector / 32], SDT_SYS386IGT, SEL_KPL,
 1011             GSEL(GCODE_SEL, SEL_KPL));
 1012 }
 1013 
 1014 void
 1015 apic_disable_vector(u_int vector)
 1016 {
 1017 
 1018         KASSERT(vector != IDT_SYSCALL, ("Attempt to overwrite syscall entry"));
 1019         KASSERT(ioint_handlers[vector / 32] != NULL,
 1020             ("No ISR handler for vector %u", vector));
 1021         setidt(vector, &IDTVEC(rsvd), SDT_SYS386TGT, SEL_KPL,
 1022             GSEL(GCODE_SEL, SEL_KPL));
 1023 }
 1024 
 1025 /* Release an APIC vector when it's no longer in use. */
 1026 void
 1027 apic_free_vector(u_int vector, u_int irq)
 1028 {
 1029         KASSERT(vector >= APIC_IO_INTS && vector != IDT_SYSCALL &&
 1030             vector <= APIC_IO_INTS + APIC_NUM_IOINTS,
 1031             ("Vector %u does not map to an IRQ line", vector));
 1032         KASSERT(irq < NUM_IO_INTS, ("Invalid IRQ %u", irq));
 1033         KASSERT(ioint_irqs[vector - APIC_IO_INTS] == irq, ("IRQ mismatch"));
 1034         mtx_lock_spin(&icu_lock);
 1035         ioint_irqs[vector - APIC_IO_INTS] = 0;
 1036         mtx_unlock_spin(&icu_lock);
 1037 }
 1038 
 1039 /* Map an IDT vector (APIC) to an IRQ (interrupt source). */
 1040 u_int
 1041 apic_idt_to_irq(u_int vector)
 1042 {
 1043 
 1044         KASSERT(vector >= APIC_IO_INTS && vector != IDT_SYSCALL &&
 1045             vector <= APIC_IO_INTS + APIC_NUM_IOINTS,
 1046             ("Vector %u does not map to an IRQ line", vector));
 1047         return (ioint_irqs[vector - APIC_IO_INTS]);
 1048 }
 1049 
 1050 #ifdef DDB
 1051 /*
 1052  * Dump data about APIC IDT vector mappings.
 1053  */
 1054 DB_SHOW_COMMAND(apic, db_show_apic)
 1055 {
 1056         struct intsrc *isrc;
 1057         int i, verbose;
 1058         u_int irq;
 1059 
 1060         if (strcmp(modif, "vv") == 0)
 1061                 verbose = 2;
 1062         else if (strcmp(modif, "v") == 0)
 1063                 verbose = 1;
 1064         else
 1065                 verbose = 0;
 1066         for (i = 0; i < APIC_NUM_IOINTS + 1 && !db_pager_quit; i++) {
 1067                 irq = ioint_irqs[i];
 1068                 if (irq != 0 && irq != IRQ_SYSCALL) {
 1069                         db_printf("vec 0x%2x -> ", i + APIC_IO_INTS);
 1070                         if (irq == IRQ_TIMER)
 1071                                 db_printf("lapic timer\n");
 1072                         else if (irq < NUM_IO_INTS) {
 1073                                 isrc = intr_lookup_source(irq);
 1074                                 if (isrc == NULL || verbose == 0)
 1075                                         db_printf("IRQ %u\n", irq);
 1076                                 else
 1077                                         db_dump_intr_event(isrc->is_event,
 1078                                             verbose == 2);
 1079                         } else
 1080                                 db_printf("IRQ %u ???\n", irq);
 1081                 }
 1082         }
 1083 }
 1084 
 1085 static void
 1086 dump_mask(const char *prefix, uint32_t v, int base)
 1087 {
 1088         int i, first;
 1089 
 1090         first = 1;
 1091         for (i = 0; i < 32; i++)
 1092                 if (v & (1 << i)) {
 1093                         if (first) {
 1094                                 db_printf("%s:", prefix);
 1095                                 first = 0;
 1096                         }
 1097                         db_printf(" %02x", base + i);
 1098                 }
 1099         if (!first)
 1100                 db_printf("\n");
 1101 }
 1102 
 1103 /* Show info from the lapic regs for this CPU. */
 1104 DB_SHOW_COMMAND(lapic, db_show_lapic)
 1105 {
 1106         uint32_t v;
 1107 
 1108         db_printf("lapic ID = %d\n", lapic_id());
 1109         v = lapic->version;
 1110         db_printf("version  = %d.%d\n", (v & APIC_VER_VERSION) >> 4,
 1111             v & 0xf);
 1112         db_printf("max LVT  = %d\n", (v & APIC_VER_MAXLVT) >> MAXLVTSHIFT);
 1113         v = lapic->svr;
 1114         db_printf("SVR      = %02x (%s)\n", v & APIC_SVR_VECTOR,
 1115             v & APIC_SVR_ENABLE ? "enabled" : "disabled");
 1116         db_printf("TPR      = %02x\n", lapic->tpr);
 1117 
 1118 #define dump_field(prefix, index)                                       \
 1119         dump_mask(__XSTRING(prefix ## index), lapic->prefix ## index,   \
 1120             index * 32)
 1121 
 1122         db_printf("In-service Interrupts:\n");
 1123         dump_field(isr, 0);
 1124         dump_field(isr, 1);
 1125         dump_field(isr, 2);
 1126         dump_field(isr, 3);
 1127         dump_field(isr, 4);
 1128         dump_field(isr, 5);
 1129         dump_field(isr, 6);
 1130         dump_field(isr, 7);
 1131 
 1132         db_printf("TMR Interrupts:\n");
 1133         dump_field(tmr, 0);
 1134         dump_field(tmr, 1);
 1135         dump_field(tmr, 2);
 1136         dump_field(tmr, 3);
 1137         dump_field(tmr, 4);
 1138         dump_field(tmr, 5);
 1139         dump_field(tmr, 6);
 1140         dump_field(tmr, 7);
 1141 
 1142         db_printf("IRR Interrupts:\n");
 1143         dump_field(irr, 0);
 1144         dump_field(irr, 1);
 1145         dump_field(irr, 2);
 1146         dump_field(irr, 3);
 1147         dump_field(irr, 4);
 1148         dump_field(irr, 5);
 1149         dump_field(irr, 6);
 1150         dump_field(irr, 7);
 1151 
 1152 #undef dump_field
 1153 }
 1154 #endif
 1155 
 1156 /*
 1157  * APIC probing support code.  This includes code to manage enumerators.
 1158  */
 1159 
 1160 static SLIST_HEAD(, apic_enumerator) enumerators =
 1161         SLIST_HEAD_INITIALIZER(enumerators);
 1162 static struct apic_enumerator *best_enum;
 1163         
 1164 void
 1165 apic_register_enumerator(struct apic_enumerator *enumerator)
 1166 {
 1167 #ifdef INVARIANTS
 1168         struct apic_enumerator *apic_enum;
 1169 
 1170         SLIST_FOREACH(apic_enum, &enumerators, apic_next) {
 1171                 if (apic_enum == enumerator)
 1172                         panic("%s: Duplicate register of %s", __func__,
 1173                             enumerator->apic_name);
 1174         }
 1175 #endif
 1176         SLIST_INSERT_HEAD(&enumerators, enumerator, apic_next);
 1177 }
 1178 
 1179 /*
 1180  * Probe the APIC enumerators, enumerate CPUs, and initialize the
 1181  * local APIC.
 1182  */
 1183 static void
 1184 apic_init(void *dummy __unused)
 1185 {
 1186         struct apic_enumerator *enumerator;
 1187         uint64_t apic_base;
 1188         int retval, best;
 1189 
 1190         /* We only support built in local APICs. */
 1191         if (!(cpu_feature & CPUID_APIC))
 1192                 return;
 1193 
 1194         /* Don't probe if APIC mode is disabled. */
 1195         if (resource_disabled("apic", 0))
 1196                 return;
 1197 
 1198         /* First, probe all the enumerators to find the best match. */
 1199         best_enum = NULL;
 1200         best = 0;
 1201         SLIST_FOREACH(enumerator, &enumerators, apic_next) {
 1202                 retval = enumerator->apic_probe();
 1203                 if (retval > 0)
 1204                         continue;
 1205                 if (best_enum == NULL || best < retval) {
 1206                         best_enum = enumerator;
 1207                         best = retval;
 1208                 }
 1209         }
 1210         if (best_enum == NULL) {
 1211                 if (bootverbose)
 1212                         printf("APIC: Could not find any APICs.\n");
 1213                 return;
 1214         }
 1215 
 1216         if (bootverbose)
 1217                 printf("APIC: Using the %s enumerator.\n",
 1218                     best_enum->apic_name);
 1219 
 1220         /*
 1221          * To work around an errata, we disable the local APIC on some
 1222          * CPUs during early startup.  We need to turn the local APIC back
 1223          * on on such CPUs now.
 1224          */
 1225         if (cpu == CPU_686 && cpu_vendor_id == CPU_VENDOR_INTEL &&
 1226             (cpu_id & 0xff0) == 0x610) {
 1227                 apic_base = rdmsr(MSR_APICBASE);
 1228                 apic_base |= APICBASE_ENABLED;
 1229                 wrmsr(MSR_APICBASE, apic_base);
 1230         }
 1231 
 1232         /* Second, probe the CPU's in the system. */
 1233         retval = best_enum->apic_probe_cpus();
 1234         if (retval != 0)
 1235                 printf("%s: Failed to probe CPUs: returned %d\n",
 1236                     best_enum->apic_name, retval);
 1237 
 1238         /* Third, initialize the local APIC. */
 1239         retval = best_enum->apic_setup_local();
 1240         if (retval != 0)
 1241                 printf("%s: Failed to setup the local APIC: returned %d\n",
 1242                     best_enum->apic_name, retval);
 1243 }
 1244 SYSINIT(apic_init, SI_SUB_CPU, SI_ORDER_SECOND, apic_init, NULL);
 1245 
 1246 /*
 1247  * Setup the I/O APICs.
 1248  */
 1249 static void
 1250 apic_setup_io(void *dummy __unused)
 1251 {
 1252         int retval;
 1253 
 1254         if (best_enum == NULL)
 1255                 return;
 1256         retval = best_enum->apic_setup_io();
 1257         if (retval != 0)
 1258                 printf("%s: Failed to setup I/O APICs: returned %d\n",
 1259                     best_enum->apic_name, retval);
 1260 
 1261         /*
 1262          * Finish setting up the local APIC on the BSP once we know how to
 1263          * properly program the LINT pins.
 1264          */
 1265         lapic_setup(1);
 1266         intr_register_pic(&lapic_pic);
 1267         if (bootverbose)
 1268                 lapic_dump("BSP");
 1269 
 1270         /* Enable the MSI "pic". */
 1271         msi_init();
 1272 }
 1273 SYSINIT(apic_setup_io, SI_SUB_INTR, SI_ORDER_SECOND, apic_setup_io, NULL);
 1274 
 1275 #ifdef SMP
 1276 /*
 1277  * Inter Processor Interrupt functions.  The lapic_ipi_*() functions are
 1278  * private to the sys/i386 code.  The public interface for the rest of the
 1279  * kernel is defined in mp_machdep.c.
 1280  */
 1281 int
 1282 lapic_ipi_wait(int delay)
 1283 {
 1284         int x, incr;
 1285 
 1286         /*
 1287          * Wait delay loops for IPI to be sent.  This is highly bogus
 1288          * since this is sensitive to CPU clock speed.  If delay is
 1289          * -1, we wait forever.
 1290          */
 1291         if (delay == -1) {
 1292                 incr = 0;
 1293                 delay = 1;
 1294         } else
 1295                 incr = 1;
 1296         for (x = 0; x < delay; x += incr) {
 1297                 if ((lapic->icr_lo & APIC_DELSTAT_MASK) == APIC_DELSTAT_IDLE)
 1298                         return (1);
 1299                 ia32_pause();
 1300         }
 1301         return (0);
 1302 }
 1303 
 1304 void
 1305 lapic_ipi_raw(register_t icrlo, u_int dest)
 1306 {
 1307         register_t value, eflags;
 1308 
 1309         /* XXX: Need more sanity checking of icrlo? */
 1310         KASSERT(lapic != NULL, ("%s called too early", __func__));
 1311         KASSERT((dest & ~(APIC_ID_MASK >> APIC_ID_SHIFT)) == 0,
 1312             ("%s: invalid dest field", __func__));
 1313         KASSERT((icrlo & APIC_ICRLO_RESV_MASK) == 0,
 1314             ("%s: reserved bits set in ICR LO register", __func__));
 1315 
 1316         /* Set destination in ICR HI register if it is being used. */
 1317         eflags = intr_disable();
 1318         if ((icrlo & APIC_DEST_MASK) == APIC_DEST_DESTFLD) {
 1319                 value = lapic->icr_hi;
 1320                 value &= ~APIC_ID_MASK;
 1321                 value |= dest << APIC_ID_SHIFT;
 1322                 lapic->icr_hi = value;
 1323         }
 1324 
 1325         /* Program the contents of the IPI and dispatch it. */
 1326         value = lapic->icr_lo;
 1327         value &= APIC_ICRLO_RESV_MASK;
 1328         value |= icrlo;
 1329         lapic->icr_lo = value;
 1330         intr_restore(eflags);
 1331 }
 1332 
 1333 #define BEFORE_SPIN     1000000
 1334 #ifdef DETECT_DEADLOCK
 1335 #define AFTER_SPIN      1000
 1336 #endif
 1337 
 1338 void
 1339 lapic_ipi_vectored(u_int vector, int dest)
 1340 {
 1341         register_t icrlo, destfield;
 1342 
 1343         KASSERT((vector & ~APIC_VECTOR_MASK) == 0,
 1344             ("%s: invalid vector %d", __func__, vector));
 1345 
 1346         icrlo = vector | APIC_DELMODE_FIXED | APIC_DESTMODE_PHY |
 1347             APIC_LEVEL_DEASSERT | APIC_TRIGMOD_EDGE;
 1348         destfield = 0;
 1349         switch (dest) {
 1350         case APIC_IPI_DEST_SELF:
 1351                 icrlo |= APIC_DEST_SELF;
 1352                 break;
 1353         case APIC_IPI_DEST_ALL:
 1354                 icrlo |= APIC_DEST_ALLISELF;
 1355                 break;
 1356         case APIC_IPI_DEST_OTHERS:
 1357                 icrlo |= APIC_DEST_ALLESELF;
 1358                 break;
 1359         default:
 1360                 KASSERT((dest & ~(APIC_ID_MASK >> APIC_ID_SHIFT)) == 0,
 1361                     ("%s: invalid destination 0x%x", __func__, dest));
 1362                 destfield = dest;
 1363         }
 1364 
 1365         /* Wait for an earlier IPI to finish. */
 1366         if (!lapic_ipi_wait(BEFORE_SPIN)) {
 1367                 if (panicstr != NULL)
 1368                         return;
 1369                 else
 1370                         panic("APIC: Previous IPI is stuck");
 1371         }
 1372 
 1373         lapic_ipi_raw(icrlo, destfield);
 1374 
 1375 #ifdef DETECT_DEADLOCK
 1376         /* Wait for IPI to be delivered. */
 1377         if (!lapic_ipi_wait(AFTER_SPIN)) {
 1378 #ifdef needsattention
 1379                 /*
 1380                  * XXX FIXME:
 1381                  *
 1382                  * The above function waits for the message to actually be
 1383                  * delivered.  It breaks out after an arbitrary timeout
 1384                  * since the message should eventually be delivered (at
 1385                  * least in theory) and that if it wasn't we would catch
 1386                  * the failure with the check above when the next IPI is
 1387                  * sent.
 1388                  *
 1389                  * We could skip this wait entirely, EXCEPT it probably
 1390                  * protects us from other routines that assume that the
 1391                  * message was delivered and acted upon when this function
 1392                  * returns.
 1393                  */
 1394                 printf("APIC: IPI might be stuck\n");
 1395 #else /* !needsattention */
 1396                 /* Wait until mesage is sent without a timeout. */
 1397                 while (lapic->icr_lo & APIC_DELSTAT_PEND)
 1398                         ia32_pause();
 1399 #endif /* needsattention */
 1400         }
 1401 #endif /* DETECT_DEADLOCK */
 1402 }
 1403 #endif /* SMP */

Cache object: 4f4ae9e61c77b4f39cf73e6238be21f7


[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]


This page is part of the FreeBSD/Linux Linux Kernel Cross-Reference, and was automatically generated using a modified version of the LXR engine.