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/amd64/amd64/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/8.0/sys/amd64/amd64/local_apic.c 196996 2009-09-08 21:50:34Z 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/proc.h>
   50 #include <sys/sched.h>
   51 #include <sys/smp.h>
   52 
   53 #include <vm/vm.h>
   54 #include <vm/pmap.h>
   55 
   56 #include <machine/apicreg.h>
   57 #include <machine/cpu.h>
   58 #include <machine/cputypes.h>
   59 #include <machine/frame.h>
   60 #include <machine/intr_machdep.h>
   61 #include <machine/apicvar.h>
   62 #include <machine/md_var.h>
   63 #include <machine/smp.h>
   64 #include <machine/specialreg.h>
   65 
   66 #ifdef DDB
   67 #include <sys/interrupt.h>
   68 #include <ddb/ddb.h>
   69 #endif
   70 
   71 #ifdef KDTRACE_HOOKS
   72 #include <sys/dtrace_bsd.h>
   73 cyclic_clock_func_t     lapic_cyclic_clock_func[MAXCPU];
   74 #endif
   75 
   76 /* Sanity checks on IDT vectors. */
   77 CTASSERT(APIC_IO_INTS + APIC_NUM_IOINTS == APIC_TIMER_INT);
   78 CTASSERT(APIC_TIMER_INT < APIC_LOCAL_INTS);
   79 CTASSERT(APIC_LOCAL_INTS == 240);
   80 CTASSERT(IPI_STOP < APIC_SPURIOUS_INT);
   81 
   82 /* Magic IRQ values for the timer and syscalls. */
   83 #define IRQ_TIMER       (NUM_IO_INTS + 1)
   84 #define IRQ_SYSCALL     (NUM_IO_INTS + 2)
   85 
   86 /*
   87  * Support for local APICs.  Local APICs manage interrupts on each
   88  * individual processor as opposed to I/O APICs which receive interrupts
   89  * from I/O devices and then forward them on to the local APICs.
   90  *
   91  * Local APICs can also send interrupts to each other thus providing the
   92  * mechanism for IPIs.
   93  */
   94 
   95 struct lvt {
   96         u_int lvt_edgetrigger:1;
   97         u_int lvt_activehi:1;
   98         u_int lvt_masked:1;
   99         u_int lvt_active:1;
  100         u_int lvt_mode:16;
  101         u_int lvt_vector:8;
  102 };
  103 
  104 struct lapic {
  105         struct lvt la_lvts[LVT_MAX + 1];
  106         u_int la_id:8;
  107         u_int la_cluster:4;
  108         u_int la_cluster_id:2;
  109         u_int la_present:1;
  110         u_long *la_timer_count;
  111         u_long la_hard_ticks;
  112         u_long la_stat_ticks;
  113         u_long la_prof_ticks;
  114         /* Include IDT_SYSCALL to make indexing easier. */
  115         int la_ioint_irqs[APIC_NUM_IOINTS + 1];
  116 } static lapics[MAX_APIC_ID + 1];
  117 
  118 /* XXX: should thermal be an NMI? */
  119 
  120 /* Global defaults for local APIC LVT entries. */
  121 static struct lvt lvts[LVT_MAX + 1] = {
  122         { 1, 1, 1, 1, APIC_LVT_DM_EXTINT, 0 },  /* LINT0: masked ExtINT */
  123         { 1, 1, 0, 1, APIC_LVT_DM_NMI, 0 },     /* LINT1: NMI */
  124         { 1, 1, 1, 1, APIC_LVT_DM_FIXED, APIC_TIMER_INT },      /* Timer */
  125         { 1, 1, 1, 1, APIC_LVT_DM_FIXED, APIC_ERROR_INT },      /* Error */
  126         { 1, 1, 1, 1, APIC_LVT_DM_NMI, 0 },     /* PMC */
  127         { 1, 1, 1, 1, APIC_LVT_DM_FIXED, APIC_THERMAL_INT },    /* Thermal */
  128 };
  129 
  130 static inthand_t *ioint_handlers[] = {
  131         NULL,                   /* 0 - 31 */
  132         IDTVEC(apic_isr1),      /* 32 - 63 */
  133         IDTVEC(apic_isr2),      /* 64 - 95 */
  134         IDTVEC(apic_isr3),      /* 96 - 127 */
  135         IDTVEC(apic_isr4),      /* 128 - 159 */
  136         IDTVEC(apic_isr5),      /* 160 - 191 */
  137         IDTVEC(apic_isr6),      /* 192 - 223 */
  138         IDTVEC(apic_isr7),      /* 224 - 255 */
  139 };
  140 
  141 
  142 static u_int32_t lapic_timer_divisors[] = {
  143         APIC_TDCR_1, APIC_TDCR_2, APIC_TDCR_4, APIC_TDCR_8, APIC_TDCR_16,
  144         APIC_TDCR_32, APIC_TDCR_64, APIC_TDCR_128
  145 };
  146 
  147 extern inthand_t IDTVEC(rsvd);
  148 
  149 volatile lapic_t *lapic;
  150 vm_paddr_t lapic_paddr;
  151 static u_long lapic_timer_divisor, lapic_timer_period, lapic_timer_hz;
  152 
  153 static void     lapic_enable(void);
  154 static void     lapic_resume(struct pic *pic);
  155 static void     lapic_timer_enable_intr(void);
  156 static void     lapic_timer_oneshot(u_int count);
  157 static void     lapic_timer_periodic(u_int count);
  158 static void     lapic_timer_set_divisor(u_int divisor);
  159 static uint32_t lvt_mode(struct lapic *la, u_int pin, uint32_t value);
  160 
  161 struct pic lapic_pic = { .pic_resume = lapic_resume };
  162 
  163 static uint32_t
  164 lvt_mode(struct lapic *la, u_int pin, uint32_t value)
  165 {
  166         struct lvt *lvt;
  167 
  168         KASSERT(pin <= LVT_MAX, ("%s: pin %u out of range", __func__, pin));
  169         if (la->la_lvts[pin].lvt_active)
  170                 lvt = &la->la_lvts[pin];
  171         else
  172                 lvt = &lvts[pin];
  173 
  174         value &= ~(APIC_LVT_M | APIC_LVT_TM | APIC_LVT_IIPP | APIC_LVT_DM |
  175             APIC_LVT_VECTOR);
  176         if (lvt->lvt_edgetrigger == 0)
  177                 value |= APIC_LVT_TM;
  178         if (lvt->lvt_activehi == 0)
  179                 value |= APIC_LVT_IIPP_INTALO;
  180         if (lvt->lvt_masked)
  181                 value |= APIC_LVT_M;
  182         value |= lvt->lvt_mode;
  183         switch (lvt->lvt_mode) {
  184         case APIC_LVT_DM_NMI:
  185         case APIC_LVT_DM_SMI:
  186         case APIC_LVT_DM_INIT:
  187         case APIC_LVT_DM_EXTINT:
  188                 if (!lvt->lvt_edgetrigger) {
  189                         printf("lapic%u: Forcing LINT%u to edge trigger\n",
  190                             la->la_id, pin);
  191                         value |= APIC_LVT_TM;
  192                 }
  193                 /* Use a vector of 0. */
  194                 break;
  195         case APIC_LVT_DM_FIXED:
  196                 value |= lvt->lvt_vector;
  197                 break;
  198         default:
  199                 panic("bad APIC LVT delivery mode: %#x\n", value);
  200         }
  201         return (value);
  202 }
  203 
  204 /*
  205  * Map the local APIC and setup necessary interrupt vectors.
  206  */
  207 void
  208 lapic_init(vm_paddr_t addr)
  209 {
  210 
  211         /* Map the local APIC and setup the spurious interrupt handler. */
  212         KASSERT(trunc_page(addr) == addr,
  213             ("local APIC not aligned on a page boundary"));
  214         lapic = pmap_mapdev(addr, sizeof(lapic_t));
  215         lapic_paddr = addr;
  216         setidt(APIC_SPURIOUS_INT, IDTVEC(spuriousint), SDT_SYSIGT, SEL_KPL, 0);
  217 
  218         /* Perform basic initialization of the BSP's local APIC. */
  219         lapic_enable();
  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_SYSIGT, SEL_KPL, 0);
  226 
  227         /* XXX: error/thermal interrupts */
  228 }
  229 
  230 /*
  231  * Create a local APIC instance.
  232  */
  233 void
  234 lapic_create(u_int apic_id, int boot_cpu)
  235 {
  236         int i;
  237 
  238         if (apic_id > MAX_APIC_ID) {
  239                 printf("APIC: Ignoring local APIC with ID %d\n", apic_id);
  240                 if (boot_cpu)
  241                         panic("Can't ignore BSP");
  242                 return;
  243         }
  244         KASSERT(!lapics[apic_id].la_present, ("duplicate local APIC %u",
  245             apic_id));
  246 
  247         /*
  248          * Assume no local LVT overrides and a cluster of 0 and
  249          * intra-cluster ID of 0.
  250          */
  251         lapics[apic_id].la_present = 1;
  252         lapics[apic_id].la_id = apic_id;
  253         for (i = 0; i < LVT_MAX; i++) {
  254                 lapics[apic_id].la_lvts[i] = lvts[i];
  255                 lapics[apic_id].la_lvts[i].lvt_active = 0;
  256         }
  257         for (i = 0; i <= APIC_NUM_IOINTS; i++)
  258             lapics[apic_id].la_ioint_irqs[i] = -1;
  259         lapics[apic_id].la_ioint_irqs[IDT_SYSCALL - APIC_IO_INTS] = IRQ_SYSCALL;
  260         lapics[apic_id].la_ioint_irqs[APIC_TIMER_INT - APIC_IO_INTS] =
  261             IRQ_TIMER;
  262 
  263 #ifdef SMP
  264         cpu_add(apic_id, boot_cpu);
  265 #endif
  266 }
  267 
  268 /*
  269  * Dump contents of local APIC registers
  270  */
  271 void
  272 lapic_dump(const char* str)
  273 {
  274 
  275         printf("cpu%d %s:\n", PCPU_GET(cpuid), str);
  276         printf("     ID: 0x%08x   VER: 0x%08x LDR: 0x%08x DFR: 0x%08x\n",
  277             lapic->id, lapic->version, lapic->ldr, lapic->dfr);
  278         printf("  lint0: 0x%08x lint1: 0x%08x TPR: 0x%08x SVR: 0x%08x\n",
  279             lapic->lvt_lint0, lapic->lvt_lint1, lapic->tpr, lapic->svr);
  280         printf("  timer: 0x%08x therm: 0x%08x err: 0x%08x pcm: 0x%08x\n",
  281             lapic->lvt_timer, lapic->lvt_thermal, lapic->lvt_error,
  282             lapic->lvt_pcint);
  283 }
  284 
  285 void
  286 lapic_setup(int boot)
  287 {
  288         struct lapic *la;
  289         u_int32_t maxlvt;
  290         register_t eflags;
  291         char buf[MAXCOMLEN + 1];
  292 
  293         la = &lapics[lapic_id()];
  294         KASSERT(la->la_present, ("missing APIC structure"));
  295         eflags = intr_disable();
  296         maxlvt = (lapic->version & APIC_VER_MAXLVT) >> MAXLVTSHIFT;
  297 
  298         /* Initialize the TPR to allow all interrupts. */
  299         lapic_set_tpr(0);
  300 
  301         /* Setup spurious vector and enable the local APIC. */
  302         lapic_enable();
  303 
  304         /* Program LINT[01] LVT entries. */
  305         lapic->lvt_lint0 = lvt_mode(la, LVT_LINT0, lapic->lvt_lint0);
  306         lapic->lvt_lint1 = lvt_mode(la, LVT_LINT1, lapic->lvt_lint1);
  307 
  308         /* Program the PMC LVT entry if present. */
  309         if (maxlvt >= LVT_PMC)
  310                 lapic->lvt_pcint = lvt_mode(la, LVT_PMC, lapic->lvt_pcint);
  311 
  312         /* Program timer LVT and setup handler. */
  313         lapic->lvt_timer = lvt_mode(la, LVT_TIMER, lapic->lvt_timer);
  314         if (boot) {
  315                 snprintf(buf, sizeof(buf), "cpu%d: timer", PCPU_GET(cpuid));
  316                 intrcnt_add(buf, &la->la_timer_count);
  317         }
  318 
  319         /* We don't setup the timer during boot on the BSP until later. */
  320         if (!(boot && PCPU_GET(cpuid) == 0) && lapic_timer_hz != 0) {
  321                 KASSERT(lapic_timer_period != 0, ("lapic%u: zero divisor",
  322                     lapic_id()));
  323                 lapic_timer_set_divisor(lapic_timer_divisor);
  324                 lapic_timer_periodic(lapic_timer_period);
  325                 lapic_timer_enable_intr();
  326         }
  327 
  328         /* XXX: Error and thermal LVTs */
  329 
  330         intr_restore(eflags);
  331 }
  332 
  333 void
  334 lapic_reenable_pmc(void)
  335 {
  336 #ifdef HWPMC_HOOKS
  337         uint32_t value;
  338 
  339         value =  lapic->lvt_pcint;
  340         value &= ~APIC_LVT_M;
  341         lapic->lvt_pcint = value;
  342 #endif
  343 }
  344 
  345 #ifdef HWPMC_HOOKS
  346 static void
  347 lapic_update_pmc(void *dummy)
  348 {
  349         struct lapic *la;
  350 
  351         la = &lapics[lapic_id()];
  352         lapic->lvt_pcint = lvt_mode(la, LVT_PMC, lapic->lvt_pcint);
  353 }
  354 #endif
  355 
  356 int
  357 lapic_enable_pmc(void)
  358 {
  359 #ifdef HWPMC_HOOKS
  360         u_int32_t maxlvt;
  361 
  362         /* Fail if the local APIC is not present. */
  363         if (lapic == NULL)
  364                 return (0);
  365 
  366         /* Fail if the PMC LVT is not present. */
  367         maxlvt = (lapic->version & APIC_VER_MAXLVT) >> MAXLVTSHIFT;
  368         if (maxlvt < LVT_PMC)
  369                 return (0);
  370 
  371         lvts[LVT_PMC].lvt_masked = 0;
  372 
  373 #ifdef SMP
  374         /*
  375          * If hwpmc was loaded at boot time then the APs may not be
  376          * started yet.  In that case, don't forward the request to
  377          * them as they will program the lvt when they start.
  378          */
  379         if (smp_started)
  380                 smp_rendezvous(NULL, lapic_update_pmc, NULL, NULL);
  381         else
  382 #endif
  383                 lapic_update_pmc(NULL);
  384         return (1);
  385 #else
  386         return (0);
  387 #endif
  388 }
  389 
  390 void
  391 lapic_disable_pmc(void)
  392 {
  393 #ifdef HWPMC_HOOKS
  394         u_int32_t maxlvt;
  395 
  396         /* Fail if the local APIC is not present. */
  397         if (lapic == NULL)
  398                 return;
  399 
  400         /* Fail if the PMC LVT is not present. */
  401         maxlvt = (lapic->version & APIC_VER_MAXLVT) >> MAXLVTSHIFT;
  402         if (maxlvt < LVT_PMC)
  403                 return;
  404 
  405         lvts[LVT_PMC].lvt_masked = 1;
  406 
  407 #ifdef SMP
  408         /* The APs should always be started when hwpmc is unloaded. */
  409         KASSERT(mp_ncpus == 1 || smp_started, ("hwpmc unloaded too early"));
  410 #endif
  411         smp_rendezvous(NULL, lapic_update_pmc, NULL, NULL);
  412 #endif
  413 }
  414 
  415 /*
  416  * Called by cpu_initclocks() on the BSP to setup the local APIC timer so
  417  * that it can drive hardclock, statclock, and profclock.  This function
  418  * returns true if it is able to use the local APIC timer to drive the
  419  * clocks and false if it is not able.
  420  */
  421 int
  422 lapic_setup_clock(void)
  423 {
  424         u_long value;
  425         int i;
  426 
  427         /* Can't drive the timer without a local APIC. */
  428         if (lapic == NULL)
  429                 return (0);
  430 
  431         if (resource_int_value("apic", 0, "clock", &i) == 0 && i == 0)
  432                 return (0);
  433 
  434         /* Start off with a divisor of 2 (power on reset default). */
  435         lapic_timer_divisor = 2;
  436 
  437         /* Try to calibrate the local APIC timer. */
  438         do {
  439                 lapic_timer_set_divisor(lapic_timer_divisor);
  440                 lapic_timer_oneshot(APIC_TIMER_MAX_COUNT);
  441                 DELAY(2000000);
  442                 value = APIC_TIMER_MAX_COUNT - lapic->ccr_timer;
  443                 if (value != APIC_TIMER_MAX_COUNT)
  444                         break;
  445                 lapic_timer_divisor <<= 1;
  446         } while (lapic_timer_divisor <= 128);
  447         if (lapic_timer_divisor > 128)
  448                 panic("lapic: Divisor too big");
  449         value /= 2;
  450         if (bootverbose)
  451                 printf("lapic: Divisor %lu, Frequency %lu hz\n",
  452                     lapic_timer_divisor, value);
  453 
  454         /*
  455          * We want to run stathz in the neighborhood of 128hz.  We would
  456          * like profhz to run as often as possible, so we let it run on
  457          * each clock tick.  We try to honor the requested 'hz' value as
  458          * much as possible.
  459          *
  460          * If 'hz' is above 1500, then we just let the lapic timer
  461          * (and profhz) run at hz.  If 'hz' is below 1500 but above
  462          * 750, then we let the lapic timer run at 2 * 'hz'.  If 'hz'
  463          * is below 750 then we let the lapic timer run at 4 * 'hz'.
  464          */
  465         if (hz >= 1500)
  466                 lapic_timer_hz = hz;
  467         else if (hz >= 750)
  468                 lapic_timer_hz = hz * 2;
  469         else
  470                 lapic_timer_hz = hz * 4;
  471         if (lapic_timer_hz < 128)
  472                 stathz = lapic_timer_hz;
  473         else
  474                 stathz = lapic_timer_hz / (lapic_timer_hz / 128);
  475         profhz = lapic_timer_hz;
  476         lapic_timer_period = value / lapic_timer_hz;
  477 
  478         /*
  479          * Start up the timer on the BSP.  The APs will kick off their
  480          * timer during lapic_setup().
  481          */
  482         lapic_timer_periodic(lapic_timer_period);
  483         lapic_timer_enable_intr();
  484         return (1);
  485 }
  486 
  487 void
  488 lapic_disable(void)
  489 {
  490         uint32_t value;
  491 
  492         /* Software disable the local APIC. */
  493         value = lapic->svr;
  494         value &= ~APIC_SVR_SWEN;
  495         lapic->svr = value;
  496 }
  497 
  498 static void
  499 lapic_enable(void)
  500 {
  501         u_int32_t value;
  502 
  503         /* Program the spurious vector to enable the local APIC. */
  504         value = lapic->svr;
  505         value &= ~(APIC_SVR_VECTOR | APIC_SVR_FOCUS);
  506         value |= (APIC_SVR_FEN | APIC_SVR_SWEN | APIC_SPURIOUS_INT);
  507         lapic->svr = value;
  508 }
  509 
  510 /* Reset the local APIC on the BSP during resume. */
  511 static void
  512 lapic_resume(struct pic *pic)
  513 {
  514 
  515         lapic_setup(0);
  516 }
  517 
  518 int
  519 lapic_id(void)
  520 {
  521 
  522         KASSERT(lapic != NULL, ("local APIC is not mapped"));
  523         return (lapic->id >> APIC_ID_SHIFT);
  524 }
  525 
  526 int
  527 lapic_intr_pending(u_int vector)
  528 {
  529         volatile u_int32_t *irr;
  530 
  531         /*
  532          * The IRR registers are an array of 128-bit registers each of
  533          * which only describes 32 interrupts in the low 32 bits..  Thus,
  534          * we divide the vector by 32 to get the 128-bit index.  We then
  535          * multiply that index by 4 to get the equivalent index from
  536          * treating the IRR as an array of 32-bit registers.  Finally, we
  537          * modulus the vector by 32 to determine the individual bit to
  538          * test.
  539          */
  540         irr = &lapic->irr0;
  541         return (irr[(vector / 32) * 4] & 1 << (vector % 32));
  542 }
  543 
  544 void
  545 lapic_set_logical_id(u_int apic_id, u_int cluster, u_int cluster_id)
  546 {
  547         struct lapic *la;
  548 
  549         KASSERT(lapics[apic_id].la_present, ("%s: APIC %u doesn't exist",
  550             __func__, apic_id));
  551         KASSERT(cluster <= APIC_MAX_CLUSTER, ("%s: cluster %u too big",
  552             __func__, cluster));
  553         KASSERT(cluster_id <= APIC_MAX_INTRACLUSTER_ID,
  554             ("%s: intra cluster id %u too big", __func__, cluster_id));
  555         la = &lapics[apic_id];
  556         la->la_cluster = cluster;
  557         la->la_cluster_id = cluster_id;
  558 }
  559 
  560 int
  561 lapic_set_lvt_mask(u_int apic_id, u_int pin, u_char masked)
  562 {
  563 
  564         if (pin > LVT_MAX)
  565                 return (EINVAL);
  566         if (apic_id == APIC_ID_ALL) {
  567                 lvts[pin].lvt_masked = masked;
  568                 if (bootverbose)
  569                         printf("lapic:");
  570         } else {
  571                 KASSERT(lapics[apic_id].la_present,
  572                     ("%s: missing APIC %u", __func__, apic_id));
  573                 lapics[apic_id].la_lvts[pin].lvt_masked = masked;
  574                 lapics[apic_id].la_lvts[pin].lvt_active = 1;
  575                 if (bootverbose)
  576                         printf("lapic%u:", apic_id);
  577         }
  578         if (bootverbose)
  579                 printf(" LINT%u %s\n", pin, masked ? "masked" : "unmasked");
  580         return (0);
  581 }
  582 
  583 int
  584 lapic_set_lvt_mode(u_int apic_id, u_int pin, u_int32_t mode)
  585 {
  586         struct lvt *lvt;
  587 
  588         if (pin > LVT_MAX)
  589                 return (EINVAL);
  590         if (apic_id == APIC_ID_ALL) {
  591                 lvt = &lvts[pin];
  592                 if (bootverbose)
  593                         printf("lapic:");
  594         } else {
  595                 KASSERT(lapics[apic_id].la_present,
  596                     ("%s: missing APIC %u", __func__, apic_id));
  597                 lvt = &lapics[apic_id].la_lvts[pin];
  598                 lvt->lvt_active = 1;
  599                 if (bootverbose)
  600                         printf("lapic%u:", apic_id);
  601         }
  602         lvt->lvt_mode = mode;
  603         switch (mode) {
  604         case APIC_LVT_DM_NMI:
  605         case APIC_LVT_DM_SMI:
  606         case APIC_LVT_DM_INIT:
  607         case APIC_LVT_DM_EXTINT:
  608                 lvt->lvt_edgetrigger = 1;
  609                 lvt->lvt_activehi = 1;
  610                 if (mode == APIC_LVT_DM_EXTINT)
  611                         lvt->lvt_masked = 1;
  612                 else
  613                         lvt->lvt_masked = 0;
  614                 break;
  615         default:
  616                 panic("Unsupported delivery mode: 0x%x\n", mode);
  617         }
  618         if (bootverbose) {
  619                 printf(" Routing ");
  620                 switch (mode) {
  621                 case APIC_LVT_DM_NMI:
  622                         printf("NMI");
  623                         break;
  624                 case APIC_LVT_DM_SMI:
  625                         printf("SMI");
  626                         break;
  627                 case APIC_LVT_DM_INIT:
  628                         printf("INIT");
  629                         break;
  630                 case APIC_LVT_DM_EXTINT:
  631                         printf("ExtINT");
  632                         break;
  633                 }
  634                 printf(" -> LINT%u\n", pin);
  635         }
  636         return (0);
  637 }
  638 
  639 int
  640 lapic_set_lvt_polarity(u_int apic_id, u_int pin, enum intr_polarity pol)
  641 {
  642 
  643         if (pin > LVT_MAX || pol == INTR_POLARITY_CONFORM)
  644                 return (EINVAL);
  645         if (apic_id == APIC_ID_ALL) {
  646                 lvts[pin].lvt_activehi = (pol == INTR_POLARITY_HIGH);
  647                 if (bootverbose)
  648                         printf("lapic:");
  649         } else {
  650                 KASSERT(lapics[apic_id].la_present,
  651                     ("%s: missing APIC %u", __func__, apic_id));
  652                 lapics[apic_id].la_lvts[pin].lvt_active = 1;
  653                 lapics[apic_id].la_lvts[pin].lvt_activehi =
  654                     (pol == INTR_POLARITY_HIGH);
  655                 if (bootverbose)
  656                         printf("lapic%u:", apic_id);
  657         }
  658         if (bootverbose)
  659                 printf(" LINT%u polarity: %s\n", pin,
  660                     pol == INTR_POLARITY_HIGH ? "high" : "low");
  661         return (0);
  662 }
  663 
  664 int
  665 lapic_set_lvt_triggermode(u_int apic_id, u_int pin, enum intr_trigger trigger)
  666 {
  667 
  668         if (pin > LVT_MAX || trigger == INTR_TRIGGER_CONFORM)
  669                 return (EINVAL);
  670         if (apic_id == APIC_ID_ALL) {
  671                 lvts[pin].lvt_edgetrigger = (trigger == INTR_TRIGGER_EDGE);
  672                 if (bootverbose)
  673                         printf("lapic:");
  674         } else {
  675                 KASSERT(lapics[apic_id].la_present,
  676                     ("%s: missing APIC %u", __func__, apic_id));
  677                 lapics[apic_id].la_lvts[pin].lvt_edgetrigger =
  678                     (trigger == INTR_TRIGGER_EDGE);
  679                 lapics[apic_id].la_lvts[pin].lvt_active = 1;
  680                 if (bootverbose)
  681                         printf("lapic%u:", apic_id);
  682         }
  683         if (bootverbose)
  684                 printf(" LINT%u trigger: %s\n", pin,
  685                     trigger == INTR_TRIGGER_EDGE ? "edge" : "level");
  686         return (0);
  687 }
  688 
  689 /*
  690  * Adjust the TPR of the current CPU so that it blocks all interrupts below
  691  * the passed in vector.
  692  */
  693 void
  694 lapic_set_tpr(u_int vector)
  695 {
  696 #ifdef CHEAP_TPR
  697         lapic->tpr = vector;
  698 #else
  699         u_int32_t tpr;
  700 
  701         tpr = lapic->tpr & ~APIC_TPR_PRIO;
  702         tpr |= vector;
  703         lapic->tpr = tpr;
  704 #endif
  705 }
  706 
  707 void
  708 lapic_eoi(void)
  709 {
  710 
  711         lapic->eoi = 0;
  712 }
  713 
  714 /*
  715  * Read the contents of the error status register.  We have to write
  716  * to the register first before reading from it.
  717  */
  718 u_int
  719 lapic_error(void)
  720 {
  721 
  722         lapic->esr = 0;
  723         return (lapic->esr);
  724 }
  725 
  726 void
  727 lapic_handle_intr(int vector, struct trapframe *frame)
  728 {
  729         struct intsrc *isrc;
  730 
  731         if (vector == -1)
  732                 panic("Couldn't get vector from ISR!");
  733         isrc = intr_lookup_source(apic_idt_to_irq(PCPU_GET(apic_id),
  734             vector));
  735         intr_execute_handlers(isrc, frame);
  736 }
  737 
  738 void
  739 lapic_handle_timer(struct trapframe *frame)
  740 {
  741         struct lapic *la;
  742 
  743         /* Send EOI first thing. */
  744         lapic_eoi();
  745 
  746 #if defined(SMP) && !defined(SCHED_ULE)
  747         /*
  748          * Don't do any accounting for the disabled HTT cores, since it
  749          * will provide misleading numbers for the userland.
  750          *
  751          * No locking is necessary here, since even if we loose the race
  752          * when hlt_cpus_mask changes it is not a big deal, really.
  753          *
  754          * Don't do that for ULE, since ULE doesn't consider hlt_cpus_mask
  755          * and unlike other schedulers it actually schedules threads to
  756          * those CPUs.
  757          */
  758         if ((hlt_cpus_mask & (1 << PCPU_GET(cpuid))) != 0)
  759                 return;
  760 #endif
  761 
  762         /* Look up our local APIC structure for the tick counters. */
  763         la = &lapics[PCPU_GET(apic_id)];
  764         (*la->la_timer_count)++;
  765         critical_enter();
  766 
  767 #ifdef KDTRACE_HOOKS
  768         /*
  769          * If the DTrace hooks are configured and a callback function
  770          * has been registered, then call it to process the high speed
  771          * timers.
  772          */
  773         int cpu = PCPU_GET(cpuid);
  774         if (lapic_cyclic_clock_func[cpu] != NULL)
  775                 (*lapic_cyclic_clock_func[cpu])(frame);
  776 #endif
  777 
  778         /* Fire hardclock at hz. */
  779         la->la_hard_ticks += hz;
  780         if (la->la_hard_ticks >= lapic_timer_hz) {
  781                 la->la_hard_ticks -= lapic_timer_hz;
  782                 if (PCPU_GET(cpuid) == 0)
  783                         hardclock(TRAPF_USERMODE(frame), TRAPF_PC(frame));
  784                 else
  785                         hardclock_cpu(TRAPF_USERMODE(frame));
  786         }
  787 
  788         /* Fire statclock at stathz. */
  789         la->la_stat_ticks += stathz;
  790         if (la->la_stat_ticks >= lapic_timer_hz) {
  791                 la->la_stat_ticks -= lapic_timer_hz;
  792                 statclock(TRAPF_USERMODE(frame));
  793         }
  794 
  795         /* Fire profclock at profhz, but only when needed. */
  796         la->la_prof_ticks += profhz;
  797         if (la->la_prof_ticks >= lapic_timer_hz) {
  798                 la->la_prof_ticks -= lapic_timer_hz;
  799                 if (profprocs != 0)
  800                         profclock(TRAPF_USERMODE(frame), TRAPF_PC(frame));
  801         }
  802         critical_exit();
  803 }
  804 
  805 static void
  806 lapic_timer_set_divisor(u_int divisor)
  807 {
  808 
  809         KASSERT(powerof2(divisor), ("lapic: invalid divisor %u", divisor));
  810         KASSERT(ffs(divisor) <= sizeof(lapic_timer_divisors) /
  811             sizeof(u_int32_t), ("lapic: invalid divisor %u", divisor));
  812         lapic->dcr_timer = lapic_timer_divisors[ffs(divisor) - 1];
  813 }
  814 
  815 static void
  816 lapic_timer_oneshot(u_int count)
  817 {
  818         u_int32_t value;
  819 
  820         value = lapic->lvt_timer;
  821         value &= ~APIC_LVTT_TM;
  822         value |= APIC_LVTT_TM_ONE_SHOT;
  823         lapic->lvt_timer = value;
  824         lapic->icr_timer = count;
  825 }
  826 
  827 static void
  828 lapic_timer_periodic(u_int count)
  829 {
  830         u_int32_t value;
  831 
  832         value = lapic->lvt_timer;
  833         value &= ~APIC_LVTT_TM;
  834         value |= APIC_LVTT_TM_PERIODIC;
  835         lapic->lvt_timer = value;
  836         lapic->icr_timer = count;
  837 }
  838 
  839 static void
  840 lapic_timer_enable_intr(void)
  841 {
  842         u_int32_t value;
  843 
  844         value = lapic->lvt_timer;
  845         value &= ~APIC_LVT_M;
  846         lapic->lvt_timer = value;
  847 }
  848 
  849 u_int
  850 apic_cpuid(u_int apic_id)
  851 {
  852 #ifdef SMP
  853         return apic_cpuids[apic_id];
  854 #else
  855         return 0;
  856 #endif
  857 }
  858 
  859 /* Request a free IDT vector to be used by the specified IRQ. */
  860 u_int
  861 apic_alloc_vector(u_int apic_id, u_int irq)
  862 {
  863         u_int vector;
  864 
  865         KASSERT(irq < NUM_IO_INTS, ("Invalid IRQ %u", irq));
  866 
  867         /*
  868          * Search for a free vector.  Currently we just use a very simple
  869          * algorithm to find the first free vector.
  870          */
  871         mtx_lock_spin(&icu_lock);
  872         for (vector = 0; vector < APIC_NUM_IOINTS; vector++) {
  873                 if (lapics[apic_id].la_ioint_irqs[vector] != -1)
  874                         continue;
  875                 lapics[apic_id].la_ioint_irqs[vector] = irq;
  876                 mtx_unlock_spin(&icu_lock);
  877                 return (vector + APIC_IO_INTS);
  878         }
  879         mtx_unlock_spin(&icu_lock);
  880         return (0);
  881 }
  882 
  883 /*
  884  * Request 'count' free contiguous IDT vectors to be used by 'count'
  885  * IRQs.  'count' must be a power of two and the vectors will be
  886  * aligned on a boundary of 'align'.  If the request cannot be
  887  * satisfied, 0 is returned.
  888  */
  889 u_int
  890 apic_alloc_vectors(u_int apic_id, u_int *irqs, u_int count, u_int align)
  891 {
  892         u_int first, run, vector;
  893 
  894         KASSERT(powerof2(count), ("bad count"));
  895         KASSERT(powerof2(align), ("bad align"));
  896         KASSERT(align >= count, ("align < count"));
  897 #ifdef INVARIANTS
  898         for (run = 0; run < count; run++)
  899                 KASSERT(irqs[run] < NUM_IO_INTS, ("Invalid IRQ %u at index %u",
  900                     irqs[run], run));
  901 #endif
  902 
  903         /*
  904          * Search for 'count' free vectors.  As with apic_alloc_vector(),
  905          * this just uses a simple first fit algorithm.
  906          */
  907         run = 0;
  908         first = 0;
  909         mtx_lock_spin(&icu_lock);
  910         for (vector = 0; vector < APIC_NUM_IOINTS; vector++) {
  911 
  912                 /* Vector is in use, end run. */
  913                 if (lapics[apic_id].la_ioint_irqs[vector] != -1) {
  914                         run = 0;
  915                         first = 0;
  916                         continue;
  917                 }
  918 
  919                 /* Start a new run if run == 0 and vector is aligned. */
  920                 if (run == 0) {
  921                         if ((vector & (align - 1)) != 0)
  922                                 continue;
  923                         first = vector;
  924                 }
  925                 run++;
  926 
  927                 /* Keep looping if the run isn't long enough yet. */
  928                 if (run < count)
  929                         continue;
  930 
  931                 /* Found a run, assign IRQs and return the first vector. */
  932                 for (vector = 0; vector < count; vector++)
  933                         lapics[apic_id].la_ioint_irqs[first + vector] =
  934                             irqs[vector];
  935                 mtx_unlock_spin(&icu_lock);
  936                 return (first + APIC_IO_INTS);
  937         }
  938         mtx_unlock_spin(&icu_lock);
  939         printf("APIC: Couldn't find APIC vectors for %u IRQs\n", count);
  940         return (0);
  941 }
  942 
  943 /*
  944  * Enable a vector for a particular apic_id.  Since all lapics share idt
  945  * entries and ioint_handlers this enables the vector on all lapics.  lapics
  946  * which do not have the vector configured would report spurious interrupts
  947  * should it fire.
  948  */
  949 void
  950 apic_enable_vector(u_int apic_id, u_int vector)
  951 {
  952 
  953         KASSERT(vector != IDT_SYSCALL, ("Attempt to overwrite syscall entry"));
  954         KASSERT(ioint_handlers[vector / 32] != NULL,
  955             ("No ISR handler for vector %u", vector));
  956         setidt(vector, ioint_handlers[vector / 32], SDT_SYSIGT, SEL_KPL, 0);
  957 }
  958 
  959 void
  960 apic_disable_vector(u_int apic_id, u_int vector)
  961 {
  962 
  963         KASSERT(vector != IDT_SYSCALL, ("Attempt to overwrite syscall entry"));
  964         KASSERT(ioint_handlers[vector / 32] != NULL,
  965             ("No ISR handler for vector %u", vector));
  966 #ifdef notyet
  967         /*
  968          * We can not currently clear the idt entry because other cpus
  969          * may have a valid vector at this offset.
  970          */
  971         setidt(vector, &IDTVEC(rsvd), SDT_SYSIGT, SEL_KPL, 0);
  972 #endif
  973 }
  974 
  975 /* Release an APIC vector when it's no longer in use. */
  976 void
  977 apic_free_vector(u_int apic_id, u_int vector, u_int irq)
  978 {
  979         struct thread *td;
  980 
  981         KASSERT(vector >= APIC_IO_INTS && vector != IDT_SYSCALL &&
  982             vector <= APIC_IO_INTS + APIC_NUM_IOINTS,
  983             ("Vector %u does not map to an IRQ line", vector));
  984         KASSERT(irq < NUM_IO_INTS, ("Invalid IRQ %u", irq));
  985         KASSERT(lapics[apic_id].la_ioint_irqs[vector - APIC_IO_INTS] ==
  986             irq, ("IRQ mismatch"));
  987 
  988         /*
  989          * Bind us to the cpu that owned the vector before freeing it so
  990          * we don't lose an interrupt delivery race.
  991          */
  992         td = curthread;
  993         if (!rebooting) {
  994                 thread_lock(td);
  995                 if (sched_is_bound(td))
  996                         panic("apic_free_vector: Thread already bound.\n");
  997                 sched_bind(td, apic_cpuid(apic_id));
  998                 thread_unlock(td);
  999         }
 1000         mtx_lock_spin(&icu_lock);
 1001         lapics[apic_id].la_ioint_irqs[vector - APIC_IO_INTS] = -1;
 1002         mtx_unlock_spin(&icu_lock);
 1003         if (!rebooting) {
 1004                 thread_lock(td);
 1005                 sched_unbind(td);
 1006                 thread_unlock(td);
 1007         }
 1008 }
 1009 
 1010 /* Map an IDT vector (APIC) to an IRQ (interrupt source). */
 1011 u_int
 1012 apic_idt_to_irq(u_int apic_id, u_int vector)
 1013 {
 1014         int irq;
 1015 
 1016         KASSERT(vector >= APIC_IO_INTS && vector != IDT_SYSCALL &&
 1017             vector <= APIC_IO_INTS + APIC_NUM_IOINTS,
 1018             ("Vector %u does not map to an IRQ line", vector));
 1019         irq = lapics[apic_id].la_ioint_irqs[vector - APIC_IO_INTS];
 1020         if (irq < 0)
 1021                 irq = 0;
 1022         return (irq);
 1023 }
 1024 
 1025 #ifdef DDB
 1026 /*
 1027  * Dump data about APIC IDT vector mappings.
 1028  */
 1029 DB_SHOW_COMMAND(apic, db_show_apic)
 1030 {
 1031         struct intsrc *isrc;
 1032         int i, verbose;
 1033         u_int apic_id;
 1034         u_int irq;
 1035 
 1036         if (strcmp(modif, "vv") == 0)
 1037                 verbose = 2;
 1038         else if (strcmp(modif, "v") == 0)
 1039                 verbose = 1;
 1040         else
 1041                 verbose = 0;
 1042         for (apic_id = 0; apic_id <= MAX_APIC_ID; apic_id++) {
 1043                 if (lapics[apic_id].la_present == 0)
 1044                         continue;
 1045                 db_printf("Interrupts bound to lapic %u\n", apic_id);
 1046                 for (i = 0; i < APIC_NUM_IOINTS + 1 && !db_pager_quit; i++) {
 1047                         irq = lapics[apic_id].la_ioint_irqs[i];
 1048                         if (irq == -1 || irq == IRQ_SYSCALL)
 1049                                 continue;
 1050                         db_printf("vec 0x%2x -> ", i + APIC_IO_INTS);
 1051                         if (irq == IRQ_TIMER)
 1052                                 db_printf("lapic timer\n");
 1053                         else if (irq < NUM_IO_INTS) {
 1054                                 isrc = intr_lookup_source(irq);
 1055                                 if (isrc == NULL || verbose == 0)
 1056                                         db_printf("IRQ %u\n", irq);
 1057                                 else
 1058                                         db_dump_intr_event(isrc->is_event,
 1059                                             verbose == 2);
 1060                         } else
 1061                                 db_printf("IRQ %u ???\n", irq);
 1062                 }
 1063         }
 1064 }
 1065 
 1066 static void
 1067 dump_mask(const char *prefix, uint32_t v, int base)
 1068 {
 1069         int i, first;
 1070 
 1071         first = 1;
 1072         for (i = 0; i < 32; i++)
 1073                 if (v & (1 << i)) {
 1074                         if (first) {
 1075                                 db_printf("%s:", prefix);
 1076                                 first = 0;
 1077                         }
 1078                         db_printf(" %02x", base + i);
 1079                 }
 1080         if (!first)
 1081                 db_printf("\n");
 1082 }
 1083 
 1084 /* Show info from the lapic regs for this CPU. */
 1085 DB_SHOW_COMMAND(lapic, db_show_lapic)
 1086 {
 1087         uint32_t v;
 1088 
 1089         db_printf("lapic ID = %d\n", lapic_id());
 1090         v = lapic->version;
 1091         db_printf("version  = %d.%d\n", (v & APIC_VER_VERSION) >> 4,
 1092             v & 0xf);
 1093         db_printf("max LVT  = %d\n", (v & APIC_VER_MAXLVT) >> MAXLVTSHIFT);
 1094         v = lapic->svr;
 1095         db_printf("SVR      = %02x (%s)\n", v & APIC_SVR_VECTOR,
 1096             v & APIC_SVR_ENABLE ? "enabled" : "disabled");
 1097         db_printf("TPR      = %02x\n", lapic->tpr);
 1098 
 1099 #define dump_field(prefix, index)                                       \
 1100         dump_mask(__XSTRING(prefix ## index), lapic->prefix ## index,   \
 1101             index * 32)
 1102 
 1103         db_printf("In-service Interrupts:\n");
 1104         dump_field(isr, 0);
 1105         dump_field(isr, 1);
 1106         dump_field(isr, 2);
 1107         dump_field(isr, 3);
 1108         dump_field(isr, 4);
 1109         dump_field(isr, 5);
 1110         dump_field(isr, 6);
 1111         dump_field(isr, 7);
 1112 
 1113         db_printf("TMR Interrupts:\n");
 1114         dump_field(tmr, 0);
 1115         dump_field(tmr, 1);
 1116         dump_field(tmr, 2);
 1117         dump_field(tmr, 3);
 1118         dump_field(tmr, 4);
 1119         dump_field(tmr, 5);
 1120         dump_field(tmr, 6);
 1121         dump_field(tmr, 7);
 1122 
 1123         db_printf("IRR Interrupts:\n");
 1124         dump_field(irr, 0);
 1125         dump_field(irr, 1);
 1126         dump_field(irr, 2);
 1127         dump_field(irr, 3);
 1128         dump_field(irr, 4);
 1129         dump_field(irr, 5);
 1130         dump_field(irr, 6);
 1131         dump_field(irr, 7);
 1132 
 1133 #undef dump_field
 1134 }
 1135 #endif
 1136 
 1137 /*
 1138  * APIC probing support code.  This includes code to manage enumerators.
 1139  */
 1140 
 1141 static SLIST_HEAD(, apic_enumerator) enumerators =
 1142         SLIST_HEAD_INITIALIZER(enumerators);
 1143 static struct apic_enumerator *best_enum;
 1144 
 1145 void
 1146 apic_register_enumerator(struct apic_enumerator *enumerator)
 1147 {
 1148 #ifdef INVARIANTS
 1149         struct apic_enumerator *apic_enum;
 1150 
 1151         SLIST_FOREACH(apic_enum, &enumerators, apic_next) {
 1152                 if (apic_enum == enumerator)
 1153                         panic("%s: Duplicate register of %s", __func__,
 1154                             enumerator->apic_name);
 1155         }
 1156 #endif
 1157         SLIST_INSERT_HEAD(&enumerators, enumerator, apic_next);
 1158 }
 1159 
 1160 /*
 1161  * We have to look for CPU's very, very early because certain subsystems
 1162  * want to know how many CPU's we have extremely early on in the boot
 1163  * process.
 1164  */
 1165 static void
 1166 apic_init(void *dummy __unused)
 1167 {
 1168         struct apic_enumerator *enumerator;
 1169         int retval, best;
 1170 
 1171         /* Don't probe if APIC mode is disabled. */
 1172         if (resource_disabled("apic", 0))
 1173                 return;
 1174 
 1175         /* First, probe all the enumerators to find the best match. */
 1176         best_enum = NULL;
 1177         best = 0;
 1178         SLIST_FOREACH(enumerator, &enumerators, apic_next) {
 1179                 retval = enumerator->apic_probe();
 1180                 if (retval > 0)
 1181                         continue;
 1182                 if (best_enum == NULL || best < retval) {
 1183                         best_enum = enumerator;
 1184                         best = retval;
 1185                 }
 1186         }
 1187         if (best_enum == NULL) {
 1188                 if (bootverbose)
 1189                         printf("APIC: Could not find any APICs.\n");
 1190                 return;
 1191         }
 1192 
 1193         if (bootverbose)
 1194                 printf("APIC: Using the %s enumerator.\n",
 1195                     best_enum->apic_name);
 1196 
 1197         /* Second, probe the CPU's in the system. */
 1198         retval = best_enum->apic_probe_cpus();
 1199         if (retval != 0)
 1200                 printf("%s: Failed to probe CPUs: returned %d\n",
 1201                     best_enum->apic_name, retval);
 1202 }
 1203 SYSINIT(apic_init, SI_SUB_TUNABLES - 1, SI_ORDER_SECOND, apic_init, NULL);
 1204 
 1205 /*
 1206  * Setup the local APIC.  We have to do this prior to starting up the APs
 1207  * in the SMP case.
 1208  */
 1209 static void
 1210 apic_setup_local(void *dummy __unused)
 1211 {
 1212         int retval;
 1213 
 1214         if (best_enum == NULL)
 1215                 return;
 1216         retval = best_enum->apic_setup_local();
 1217         if (retval != 0)
 1218                 printf("%s: Failed to setup the local APIC: returned %d\n",
 1219                     best_enum->apic_name, retval);
 1220 }
 1221 SYSINIT(apic_setup_local, SI_SUB_CPU, SI_ORDER_SECOND, apic_setup_local,
 1222     NULL);
 1223 
 1224 /*
 1225  * Setup the I/O APICs.
 1226  */
 1227 static void
 1228 apic_setup_io(void *dummy __unused)
 1229 {
 1230         int retval;
 1231 
 1232         if (best_enum == NULL)
 1233                 return;
 1234         retval = best_enum->apic_setup_io();
 1235         if (retval != 0)
 1236                 printf("%s: Failed to setup I/O APICs: returned %d\n",
 1237                     best_enum->apic_name, retval);
 1238 
 1239         /*
 1240          * Finish setting up the local APIC on the BSP once we know how to
 1241          * properly program the LINT pins.
 1242          */
 1243         lapic_setup(1);
 1244         intr_register_pic(&lapic_pic);
 1245         if (bootverbose)
 1246                 lapic_dump("BSP");
 1247 
 1248         /* Enable the MSI "pic". */
 1249         msi_init();
 1250 }
 1251 SYSINIT(apic_setup_io, SI_SUB_INTR, SI_ORDER_SECOND, apic_setup_io, NULL);
 1252 
 1253 #ifdef SMP
 1254 /*
 1255  * Inter Processor Interrupt functions.  The lapic_ipi_*() functions are
 1256  * private to the sys/amd64 code.  The public interface for the rest of the
 1257  * kernel is defined in mp_machdep.c.
 1258  */
 1259 int
 1260 lapic_ipi_wait(int delay)
 1261 {
 1262         int x, incr;
 1263 
 1264         /*
 1265          * Wait delay loops for IPI to be sent.  This is highly bogus
 1266          * since this is sensitive to CPU clock speed.  If delay is
 1267          * -1, we wait forever.
 1268          */
 1269         if (delay == -1) {
 1270                 incr = 0;
 1271                 delay = 1;
 1272         } else
 1273                 incr = 1;
 1274         for (x = 0; x < delay; x += incr) {
 1275                 if ((lapic->icr_lo & APIC_DELSTAT_MASK) == APIC_DELSTAT_IDLE)
 1276                         return (1);
 1277                 ia32_pause();
 1278         }
 1279         return (0);
 1280 }
 1281 
 1282 void
 1283 lapic_ipi_raw(register_t icrlo, u_int dest)
 1284 {
 1285         register_t value, eflags;
 1286 
 1287         /* XXX: Need more sanity checking of icrlo? */
 1288         KASSERT(lapic != NULL, ("%s called too early", __func__));
 1289         KASSERT((dest & ~(APIC_ID_MASK >> APIC_ID_SHIFT)) == 0,
 1290             ("%s: invalid dest field", __func__));
 1291         KASSERT((icrlo & APIC_ICRLO_RESV_MASK) == 0,
 1292             ("%s: reserved bits set in ICR LO register", __func__));
 1293 
 1294         /* Set destination in ICR HI register if it is being used. */
 1295         eflags = intr_disable();
 1296         if ((icrlo & APIC_DEST_MASK) == APIC_DEST_DESTFLD) {
 1297                 value = lapic->icr_hi;
 1298                 value &= ~APIC_ID_MASK;
 1299                 value |= dest << APIC_ID_SHIFT;
 1300                 lapic->icr_hi = value;
 1301         }
 1302 
 1303         /* Program the contents of the IPI and dispatch it. */
 1304         value = lapic->icr_lo;
 1305         value &= APIC_ICRLO_RESV_MASK;
 1306         value |= icrlo;
 1307         lapic->icr_lo = value;
 1308         intr_restore(eflags);
 1309 }
 1310 
 1311 #define BEFORE_SPIN     1000000
 1312 #ifdef DETECT_DEADLOCK
 1313 #define AFTER_SPIN      1000
 1314 #endif
 1315 
 1316 void
 1317 lapic_ipi_vectored(u_int vector, int dest)
 1318 {
 1319         register_t icrlo, destfield;
 1320 
 1321         KASSERT((vector & ~APIC_VECTOR_MASK) == 0,
 1322             ("%s: invalid vector %d", __func__, vector));
 1323 
 1324         icrlo = APIC_DESTMODE_PHY | APIC_TRIGMOD_EDGE;
 1325 
 1326         /*
 1327          * IPI_STOP_HARD is just a "fake" vector used to send a NMI.
 1328          * Use special rules regard NMI if passed, otherwise specify
 1329          * the vector.
 1330          */
 1331         if (vector == IPI_STOP_HARD)
 1332                 icrlo |= APIC_DELMODE_NMI | APIC_LEVEL_ASSERT;
 1333         else
 1334                 icrlo |= vector | APIC_DELMODE_FIXED | APIC_LEVEL_DEASSERT;
 1335         destfield = 0;
 1336         switch (dest) {
 1337         case APIC_IPI_DEST_SELF:
 1338                 icrlo |= APIC_DEST_SELF;
 1339                 break;
 1340         case APIC_IPI_DEST_ALL:
 1341                 icrlo |= APIC_DEST_ALLISELF;
 1342                 break;
 1343         case APIC_IPI_DEST_OTHERS:
 1344                 icrlo |= APIC_DEST_ALLESELF;
 1345                 break;
 1346         default:
 1347                 KASSERT((dest & ~(APIC_ID_MASK >> APIC_ID_SHIFT)) == 0,
 1348                     ("%s: invalid destination 0x%x", __func__, dest));
 1349                 destfield = dest;
 1350         }
 1351 
 1352         /* Wait for an earlier IPI to finish. */
 1353         if (!lapic_ipi_wait(BEFORE_SPIN)) {
 1354                 if (panicstr != NULL)
 1355                         return;
 1356                 else
 1357                         panic("APIC: Previous IPI is stuck");
 1358         }
 1359 
 1360         lapic_ipi_raw(icrlo, destfield);
 1361 
 1362 #ifdef DETECT_DEADLOCK
 1363         /* Wait for IPI to be delivered. */
 1364         if (!lapic_ipi_wait(AFTER_SPIN)) {
 1365 #ifdef needsattention
 1366                 /*
 1367                  * XXX FIXME:
 1368                  *
 1369                  * The above function waits for the message to actually be
 1370                  * delivered.  It breaks out after an arbitrary timeout
 1371                  * since the message should eventually be delivered (at
 1372                  * least in theory) and that if it wasn't we would catch
 1373                  * the failure with the check above when the next IPI is
 1374                  * sent.
 1375                  *
 1376                  * We could skip this wait entirely, EXCEPT it probably
 1377                  * protects us from other routines that assume that the
 1378                  * message was delivered and acted upon when this function
 1379                  * returns.
 1380                  */
 1381                 printf("APIC: IPI might be stuck\n");
 1382 #else /* !needsattention */
 1383                 /* Wait until mesage is sent without a timeout. */
 1384                 while (lapic->icr_lo & APIC_DELSTAT_PEND)
 1385                         ia32_pause();
 1386 #endif /* needsattention */
 1387         }
 1388 #endif /* DETECT_DEADLOCK */
 1389 }
 1390 #endif /* SMP */

Cache object: d521c2a1cad26e9685c8d43181a71122


[ 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.