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/kern/kern_intr.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  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
    3  *
    4  * Copyright (c) 1997, Stefan Esser <se@freebsd.org>
    5  * All rights reserved.
    6  *
    7  * Redistribution and use in source and binary forms, with or without
    8  * modification, are permitted provided that the following conditions
    9  * are met:
   10  * 1. Redistributions of source code must retain the above copyright
   11  *    notice unmodified, this list of conditions, and the following
   12  *    disclaimer.
   13  * 2. Redistributions in binary form must reproduce the above copyright
   14  *    notice, this list of conditions and the following disclaimer in the
   15  *    documentation and/or other materials provided with the distribution.
   16  *
   17  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
   18  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
   19  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
   20  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
   21  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
   22  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
   23  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
   24  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
   25  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
   26  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
   27  */
   28 
   29 #include <sys/cdefs.h>
   30 __FBSDID("$FreeBSD$");
   31 
   32 #include "opt_ddb.h"
   33 #include "opt_kstack_usage_prof.h"
   34 
   35 #include <sys/param.h>
   36 #include <sys/bus.h>
   37 #include <sys/conf.h>
   38 #include <sys/cpuset.h>
   39 #include <sys/rtprio.h>
   40 #include <sys/systm.h>
   41 #include <sys/interrupt.h>
   42 #include <sys/kernel.h>
   43 #include <sys/kthread.h>
   44 #include <sys/ktr.h>
   45 #include <sys/limits.h>
   46 #include <sys/lock.h>
   47 #include <sys/malloc.h>
   48 #include <sys/mutex.h>
   49 #include <sys/priv.h>
   50 #include <sys/proc.h>
   51 #include <sys/epoch.h>
   52 #include <sys/random.h>
   53 #include <sys/resourcevar.h>
   54 #include <sys/sched.h>
   55 #include <sys/smp.h>
   56 #include <sys/sysctl.h>
   57 #include <sys/syslog.h>
   58 #include <sys/unistd.h>
   59 #include <sys/vmmeter.h>
   60 #include <machine/atomic.h>
   61 #include <machine/cpu.h>
   62 #include <machine/md_var.h>
   63 #include <machine/smp.h>
   64 #include <machine/stdarg.h>
   65 #ifdef DDB
   66 #include <ddb/ddb.h>
   67 #include <ddb/db_sym.h>
   68 #endif
   69 
   70 /*
   71  * Describe an interrupt thread.  There is one of these per interrupt event.
   72  */
   73 struct intr_thread {
   74         struct intr_event *it_event;
   75         struct thread *it_thread;       /* Kernel thread. */
   76         int     it_flags;               /* (j) IT_* flags. */
   77         int     it_need;                /* Needs service. */
   78 };
   79 
   80 /* Interrupt thread flags kept in it_flags */
   81 #define IT_DEAD         0x000001        /* Thread is waiting to exit. */
   82 #define IT_WAIT         0x000002        /* Thread is waiting for completion. */
   83 
   84 struct  intr_entropy {
   85         struct  thread *td;
   86         uintptr_t event;
   87 };
   88 
   89 struct  intr_event *clk_intr_event;
   90 struct proc *intrproc;
   91 
   92 static MALLOC_DEFINE(M_ITHREAD, "ithread", "Interrupt Threads");
   93 
   94 static int intr_storm_threshold = 0;
   95 SYSCTL_INT(_hw, OID_AUTO, intr_storm_threshold, CTLFLAG_RWTUN,
   96     &intr_storm_threshold, 0,
   97     "Number of consecutive interrupts before storm protection is enabled");
   98 static int intr_epoch_batch = 1000;
   99 SYSCTL_INT(_hw, OID_AUTO, intr_epoch_batch, CTLFLAG_RWTUN, &intr_epoch_batch,
  100     0, "Maximum interrupt handler executions without re-entering epoch(9)");
  101 static TAILQ_HEAD(, intr_event) event_list =
  102     TAILQ_HEAD_INITIALIZER(event_list);
  103 static struct mtx event_lock;
  104 MTX_SYSINIT(intr_event_list, &event_lock, "intr event list", MTX_DEF);
  105 
  106 static void     intr_event_update(struct intr_event *ie);
  107 static int      intr_event_schedule_thread(struct intr_event *ie);
  108 static struct intr_thread *ithread_create(const char *name);
  109 static void     ithread_destroy(struct intr_thread *ithread);
  110 static void     ithread_execute_handlers(struct proc *p, 
  111                     struct intr_event *ie);
  112 static void     ithread_loop(void *);
  113 static void     ithread_update(struct intr_thread *ithd);
  114 static void     start_softintr(void *);
  115 
  116 /* Map an interrupt type to an ithread priority. */
  117 u_char
  118 intr_priority(enum intr_type flags)
  119 {
  120         u_char pri;
  121 
  122         flags &= (INTR_TYPE_TTY | INTR_TYPE_BIO | INTR_TYPE_NET |
  123             INTR_TYPE_CAM | INTR_TYPE_MISC | INTR_TYPE_CLK | INTR_TYPE_AV);
  124         switch (flags) {
  125         case INTR_TYPE_TTY:
  126                 pri = PI_TTY;
  127                 break;
  128         case INTR_TYPE_BIO:
  129                 pri = PI_DISK;
  130                 break;
  131         case INTR_TYPE_NET:
  132                 pri = PI_NET;
  133                 break;
  134         case INTR_TYPE_CAM:
  135                 pri = PI_DISK;
  136                 break;
  137         case INTR_TYPE_AV:
  138                 pri = PI_AV;
  139                 break;
  140         case INTR_TYPE_CLK:
  141                 pri = PI_REALTIME;
  142                 break;
  143         case INTR_TYPE_MISC:
  144                 pri = PI_DULL;          /* don't care */
  145                 break;
  146         default:
  147                 /* We didn't specify an interrupt level. */
  148                 panic("intr_priority: no interrupt type in flags");
  149         }
  150 
  151         return pri;
  152 }
  153 
  154 /*
  155  * Update an ithread based on the associated intr_event.
  156  */
  157 static void
  158 ithread_update(struct intr_thread *ithd)
  159 {
  160         struct intr_event *ie;
  161         struct thread *td;
  162         u_char pri;
  163 
  164         ie = ithd->it_event;
  165         td = ithd->it_thread;
  166         mtx_assert(&ie->ie_lock, MA_OWNED);
  167 
  168         /* Determine the overall priority of this event. */
  169         if (CK_SLIST_EMPTY(&ie->ie_handlers))
  170                 pri = PRI_MAX_ITHD;
  171         else
  172                 pri = CK_SLIST_FIRST(&ie->ie_handlers)->ih_pri;
  173 
  174         /* Update name and priority. */
  175         strlcpy(td->td_name, ie->ie_fullname, sizeof(td->td_name));
  176 #ifdef KTR
  177         sched_clear_tdname(td);
  178 #endif
  179         thread_lock(td);
  180         sched_prio(td, pri);
  181         thread_unlock(td);
  182 }
  183 
  184 /*
  185  * Regenerate the full name of an interrupt event and update its priority.
  186  */
  187 static void
  188 intr_event_update(struct intr_event *ie)
  189 {
  190         struct intr_handler *ih;
  191         char *last;
  192         int missed, space, flags;
  193 
  194         /* Start off with no entropy and just the name of the event. */
  195         mtx_assert(&ie->ie_lock, MA_OWNED);
  196         strlcpy(ie->ie_fullname, ie->ie_name, sizeof(ie->ie_fullname));
  197         flags = 0;
  198         missed = 0;
  199         space = 1;
  200 
  201         /* Run through all the handlers updating values. */
  202         CK_SLIST_FOREACH(ih, &ie->ie_handlers, ih_next) {
  203                 if (strlen(ie->ie_fullname) + strlen(ih->ih_name) + 1 <
  204                     sizeof(ie->ie_fullname)) {
  205                         strcat(ie->ie_fullname, " ");
  206                         strcat(ie->ie_fullname, ih->ih_name);
  207                         space = 0;
  208                 } else
  209                         missed++;
  210                 flags |= ih->ih_flags;
  211         }
  212         ie->ie_hflags = flags;
  213 
  214         /*
  215          * If there is only one handler and its name is too long, just copy in
  216          * as much of the end of the name (includes the unit number) as will
  217          * fit.  Otherwise, we have multiple handlers and not all of the names
  218          * will fit.  Add +'s to indicate missing names.  If we run out of room
  219          * and still have +'s to add, change the last character from a + to a *.
  220          */
  221         if (missed == 1 && space == 1) {
  222                 ih = CK_SLIST_FIRST(&ie->ie_handlers);
  223                 missed = strlen(ie->ie_fullname) + strlen(ih->ih_name) + 2 -
  224                     sizeof(ie->ie_fullname);
  225                 strcat(ie->ie_fullname, (missed == 0) ? " " : "-");
  226                 strcat(ie->ie_fullname, &ih->ih_name[missed]);
  227                 missed = 0;
  228         }
  229         last = &ie->ie_fullname[sizeof(ie->ie_fullname) - 2];
  230         while (missed-- > 0) {
  231                 if (strlen(ie->ie_fullname) + 1 == sizeof(ie->ie_fullname)) {
  232                         if (*last == '+') {
  233                                 *last = '*';
  234                                 break;
  235                         } else
  236                                 *last = '+';
  237                 } else if (space) {
  238                         strcat(ie->ie_fullname, " +");
  239                         space = 0;
  240                 } else
  241                         strcat(ie->ie_fullname, "+");
  242         }
  243 
  244         /*
  245          * If this event has an ithread, update it's priority and
  246          * name.
  247          */
  248         if (ie->ie_thread != NULL)
  249                 ithread_update(ie->ie_thread);
  250         CTR2(KTR_INTR, "%s: updated %s", __func__, ie->ie_fullname);
  251 }
  252 
  253 int
  254 intr_event_create(struct intr_event **event, void *source, int flags, int irq,
  255     void (*pre_ithread)(void *), void (*post_ithread)(void *),
  256     void (*post_filter)(void *), int (*assign_cpu)(void *, int),
  257     const char *fmt, ...)
  258 {
  259         struct intr_event *ie;
  260         va_list ap;
  261 
  262         /* The only valid flag during creation is IE_SOFT. */
  263         if ((flags & ~IE_SOFT) != 0)
  264                 return (EINVAL);
  265         ie = malloc(sizeof(struct intr_event), M_ITHREAD, M_WAITOK | M_ZERO);
  266         ie->ie_source = source;
  267         ie->ie_pre_ithread = pre_ithread;
  268         ie->ie_post_ithread = post_ithread;
  269         ie->ie_post_filter = post_filter;
  270         ie->ie_assign_cpu = assign_cpu;
  271         ie->ie_flags = flags;
  272         ie->ie_irq = irq;
  273         ie->ie_cpu = NOCPU;
  274         CK_SLIST_INIT(&ie->ie_handlers);
  275         mtx_init(&ie->ie_lock, "intr event", NULL, MTX_DEF);
  276 
  277         va_start(ap, fmt);
  278         vsnprintf(ie->ie_name, sizeof(ie->ie_name), fmt, ap);
  279         va_end(ap);
  280         strlcpy(ie->ie_fullname, ie->ie_name, sizeof(ie->ie_fullname));
  281         mtx_lock(&event_lock);
  282         TAILQ_INSERT_TAIL(&event_list, ie, ie_list);
  283         mtx_unlock(&event_lock);
  284         if (event != NULL)
  285                 *event = ie;
  286         CTR2(KTR_INTR, "%s: created %s", __func__, ie->ie_name);
  287         return (0);
  288 }
  289 
  290 /*
  291  * Bind an interrupt event to the specified CPU.  Note that not all
  292  * platforms support binding an interrupt to a CPU.  For those
  293  * platforms this request will fail.  Using a cpu id of NOCPU unbinds
  294  * the interrupt event.
  295  */
  296 static int
  297 _intr_event_bind(struct intr_event *ie, int cpu, bool bindirq, bool bindithread)
  298 {
  299         lwpid_t id;
  300         int error;
  301 
  302         /* Need a CPU to bind to. */
  303         if (cpu != NOCPU && CPU_ABSENT(cpu))
  304                 return (EINVAL);
  305 
  306         if (ie->ie_assign_cpu == NULL)
  307                 return (EOPNOTSUPP);
  308 
  309         error = priv_check(curthread, PRIV_SCHED_CPUSET_INTR);
  310         if (error)
  311                 return (error);
  312 
  313         /*
  314          * If we have any ithreads try to set their mask first to verify
  315          * permissions, etc.
  316          */
  317         if (bindithread) {
  318                 mtx_lock(&ie->ie_lock);
  319                 if (ie->ie_thread != NULL) {
  320                         id = ie->ie_thread->it_thread->td_tid;
  321                         mtx_unlock(&ie->ie_lock);
  322                         error = cpuset_setithread(id, cpu);
  323                         if (error)
  324                                 return (error);
  325                 } else
  326                         mtx_unlock(&ie->ie_lock);
  327         }
  328         if (bindirq)
  329                 error = ie->ie_assign_cpu(ie->ie_source, cpu);
  330         if (error) {
  331                 if (bindithread) {
  332                         mtx_lock(&ie->ie_lock);
  333                         if (ie->ie_thread != NULL) {
  334                                 cpu = ie->ie_cpu;
  335                                 id = ie->ie_thread->it_thread->td_tid;
  336                                 mtx_unlock(&ie->ie_lock);
  337                                 (void)cpuset_setithread(id, cpu);
  338                         } else
  339                                 mtx_unlock(&ie->ie_lock);
  340                 }
  341                 return (error);
  342         }
  343 
  344         if (bindirq) {
  345                 mtx_lock(&ie->ie_lock);
  346                 ie->ie_cpu = cpu;
  347                 mtx_unlock(&ie->ie_lock);
  348         }
  349 
  350         return (error);
  351 }
  352 
  353 /*
  354  * Bind an interrupt event to the specified CPU.  For supported platforms, any
  355  * associated ithreads as well as the primary interrupt context will be bound
  356  * to the specificed CPU.
  357  */
  358 int
  359 intr_event_bind(struct intr_event *ie, int cpu)
  360 {
  361 
  362         return (_intr_event_bind(ie, cpu, true, true));
  363 }
  364 
  365 /*
  366  * Bind an interrupt event to the specified CPU, but do not bind associated
  367  * ithreads.
  368  */
  369 int
  370 intr_event_bind_irqonly(struct intr_event *ie, int cpu)
  371 {
  372 
  373         return (_intr_event_bind(ie, cpu, true, false));
  374 }
  375 
  376 /*
  377  * Bind an interrupt event's ithread to the specified CPU.
  378  */
  379 int
  380 intr_event_bind_ithread(struct intr_event *ie, int cpu)
  381 {
  382 
  383         return (_intr_event_bind(ie, cpu, false, true));
  384 }
  385 
  386 /*
  387  * Bind an interrupt event's ithread to the specified cpuset.
  388  */
  389 int
  390 intr_event_bind_ithread_cpuset(struct intr_event *ie, cpuset_t *cs)
  391 {
  392         lwpid_t id;
  393 
  394         mtx_lock(&ie->ie_lock);
  395         if (ie->ie_thread != NULL) {
  396                 id = ie->ie_thread->it_thread->td_tid;
  397                 mtx_unlock(&ie->ie_lock);
  398                 return (cpuset_setthread(id, cs));
  399         } else {
  400                 mtx_unlock(&ie->ie_lock);
  401         }
  402         return (ENODEV);
  403 }
  404 
  405 static struct intr_event *
  406 intr_lookup(int irq)
  407 {
  408         struct intr_event *ie;
  409 
  410         mtx_lock(&event_lock);
  411         TAILQ_FOREACH(ie, &event_list, ie_list)
  412                 if (ie->ie_irq == irq &&
  413                     (ie->ie_flags & IE_SOFT) == 0 &&
  414                     CK_SLIST_FIRST(&ie->ie_handlers) != NULL)
  415                         break;
  416         mtx_unlock(&event_lock);
  417         return (ie);
  418 }
  419 
  420 int
  421 intr_setaffinity(int irq, int mode, void *m)
  422 {
  423         struct intr_event *ie;
  424         cpuset_t *mask;
  425         int cpu, n;
  426 
  427         mask = m;
  428         cpu = NOCPU;
  429         /*
  430          * If we're setting all cpus we can unbind.  Otherwise make sure
  431          * only one cpu is in the set.
  432          */
  433         if (CPU_CMP(cpuset_root, mask)) {
  434                 for (n = 0; n < CPU_SETSIZE; n++) {
  435                         if (!CPU_ISSET(n, mask))
  436                                 continue;
  437                         if (cpu != NOCPU)
  438                                 return (EINVAL);
  439                         cpu = n;
  440                 }
  441         }
  442         ie = intr_lookup(irq);
  443         if (ie == NULL)
  444                 return (ESRCH);
  445         switch (mode) {
  446         case CPU_WHICH_IRQ:
  447                 return (intr_event_bind(ie, cpu));
  448         case CPU_WHICH_INTRHANDLER:
  449                 return (intr_event_bind_irqonly(ie, cpu));
  450         case CPU_WHICH_ITHREAD:
  451                 return (intr_event_bind_ithread(ie, cpu));
  452         default:
  453                 return (EINVAL);
  454         }
  455 }
  456 
  457 int
  458 intr_getaffinity(int irq, int mode, void *m)
  459 {
  460         struct intr_event *ie;
  461         struct thread *td;
  462         struct proc *p;
  463         cpuset_t *mask;
  464         lwpid_t id;
  465         int error;
  466 
  467         mask = m;
  468         ie = intr_lookup(irq);
  469         if (ie == NULL)
  470                 return (ESRCH);
  471 
  472         error = 0;
  473         CPU_ZERO(mask);
  474         switch (mode) {
  475         case CPU_WHICH_IRQ:
  476         case CPU_WHICH_INTRHANDLER:
  477                 mtx_lock(&ie->ie_lock);
  478                 if (ie->ie_cpu == NOCPU)
  479                         CPU_COPY(cpuset_root, mask);
  480                 else
  481                         CPU_SET(ie->ie_cpu, mask);
  482                 mtx_unlock(&ie->ie_lock);
  483                 break;
  484         case CPU_WHICH_ITHREAD:
  485                 mtx_lock(&ie->ie_lock);
  486                 if (ie->ie_thread == NULL) {
  487                         mtx_unlock(&ie->ie_lock);
  488                         CPU_COPY(cpuset_root, mask);
  489                 } else {
  490                         id = ie->ie_thread->it_thread->td_tid;
  491                         mtx_unlock(&ie->ie_lock);
  492                         error = cpuset_which(CPU_WHICH_TID, id, &p, &td, NULL);
  493                         if (error != 0)
  494                                 return (error);
  495                         CPU_COPY(&td->td_cpuset->cs_mask, mask);
  496                         PROC_UNLOCK(p);
  497                 }
  498         default:
  499                 return (EINVAL);
  500         }
  501         return (0);
  502 }
  503 
  504 int
  505 intr_event_destroy(struct intr_event *ie)
  506 {
  507 
  508         if (ie == NULL)
  509                 return (EINVAL);
  510 
  511         mtx_lock(&event_lock);
  512         mtx_lock(&ie->ie_lock);
  513         if (!CK_SLIST_EMPTY(&ie->ie_handlers)) {
  514                 mtx_unlock(&ie->ie_lock);
  515                 mtx_unlock(&event_lock);
  516                 return (EBUSY);
  517         }
  518         TAILQ_REMOVE(&event_list, ie, ie_list);
  519 #ifndef notyet
  520         if (ie->ie_thread != NULL) {
  521                 ithread_destroy(ie->ie_thread);
  522                 ie->ie_thread = NULL;
  523         }
  524 #endif
  525         mtx_unlock(&ie->ie_lock);
  526         mtx_unlock(&event_lock);
  527         mtx_destroy(&ie->ie_lock);
  528         free(ie, M_ITHREAD);
  529         return (0);
  530 }
  531 
  532 static struct intr_thread *
  533 ithread_create(const char *name)
  534 {
  535         struct intr_thread *ithd;
  536         struct thread *td;
  537         int error;
  538 
  539         ithd = malloc(sizeof(struct intr_thread), M_ITHREAD, M_WAITOK | M_ZERO);
  540 
  541         error = kproc_kthread_add(ithread_loop, ithd, &intrproc,
  542                     &td, RFSTOPPED | RFHIGHPID,
  543                     0, "intr", "%s", name);
  544         if (error)
  545                 panic("kproc_create() failed with %d", error);
  546         thread_lock(td);
  547         sched_class(td, PRI_ITHD);
  548         TD_SET_IWAIT(td);
  549         thread_unlock(td);
  550         td->td_pflags |= TDP_ITHREAD;
  551         ithd->it_thread = td;
  552         CTR2(KTR_INTR, "%s: created %s", __func__, name);
  553         return (ithd);
  554 }
  555 
  556 static void
  557 ithread_destroy(struct intr_thread *ithread)
  558 {
  559         struct thread *td;
  560 
  561         CTR2(KTR_INTR, "%s: killing %s", __func__, ithread->it_event->ie_name);
  562         td = ithread->it_thread;
  563         thread_lock(td);
  564         ithread->it_flags |= IT_DEAD;
  565         if (TD_AWAITING_INTR(td)) {
  566                 TD_CLR_IWAIT(td);
  567                 sched_add(td, SRQ_INTR);
  568         } else
  569                 thread_unlock(td);
  570 }
  571 
  572 int
  573 intr_event_add_handler(struct intr_event *ie, const char *name,
  574     driver_filter_t filter, driver_intr_t handler, void *arg, u_char pri,
  575     enum intr_type flags, void **cookiep)
  576 {
  577         struct intr_handler *ih, *temp_ih;
  578         struct intr_handler **prevptr;
  579         struct intr_thread *it;
  580 
  581         if (ie == NULL || name == NULL || (handler == NULL && filter == NULL))
  582                 return (EINVAL);
  583 
  584         /* Allocate and populate an interrupt handler structure. */
  585         ih = malloc(sizeof(struct intr_handler), M_ITHREAD, M_WAITOK | M_ZERO);
  586         ih->ih_filter = filter;
  587         ih->ih_handler = handler;
  588         ih->ih_argument = arg;
  589         strlcpy(ih->ih_name, name, sizeof(ih->ih_name));
  590         ih->ih_event = ie;
  591         ih->ih_pri = pri;
  592         if (flags & INTR_EXCL)
  593                 ih->ih_flags = IH_EXCLUSIVE;
  594         if (flags & INTR_MPSAFE)
  595                 ih->ih_flags |= IH_MPSAFE;
  596         if (flags & INTR_ENTROPY)
  597                 ih->ih_flags |= IH_ENTROPY;
  598         if (flags & INTR_TYPE_NET)
  599                 ih->ih_flags |= IH_NET;
  600 
  601         /* We can only have one exclusive handler in a event. */
  602         mtx_lock(&ie->ie_lock);
  603         if (!CK_SLIST_EMPTY(&ie->ie_handlers)) {
  604                 if ((flags & INTR_EXCL) ||
  605                     (CK_SLIST_FIRST(&ie->ie_handlers)->ih_flags & IH_EXCLUSIVE)) {
  606                         mtx_unlock(&ie->ie_lock);
  607                         free(ih, M_ITHREAD);
  608                         return (EINVAL);
  609                 }
  610         }
  611 
  612         /* Create a thread if we need one. */
  613         while (ie->ie_thread == NULL && handler != NULL) {
  614                 if (ie->ie_flags & IE_ADDING_THREAD)
  615                         msleep(ie, &ie->ie_lock, 0, "ithread", 0);
  616                 else {
  617                         ie->ie_flags |= IE_ADDING_THREAD;
  618                         mtx_unlock(&ie->ie_lock);
  619                         it = ithread_create("intr: newborn");
  620                         mtx_lock(&ie->ie_lock);
  621                         ie->ie_flags &= ~IE_ADDING_THREAD;
  622                         ie->ie_thread = it;
  623                         it->it_event = ie;
  624                         ithread_update(it);
  625                         wakeup(ie);
  626                 }
  627         }
  628 
  629         /* Add the new handler to the event in priority order. */
  630         CK_SLIST_FOREACH_PREVPTR(temp_ih, prevptr, &ie->ie_handlers, ih_next) {
  631                 if (temp_ih->ih_pri > ih->ih_pri)
  632                         break;
  633         }
  634         CK_SLIST_INSERT_PREVPTR(prevptr, temp_ih, ih, ih_next);
  635 
  636         intr_event_update(ie);
  637 
  638         CTR3(KTR_INTR, "%s: added %s to %s", __func__, ih->ih_name,
  639             ie->ie_name);
  640         mtx_unlock(&ie->ie_lock);
  641 
  642         if (cookiep != NULL)
  643                 *cookiep = ih;
  644         return (0);
  645 }
  646 
  647 /*
  648  * Append a description preceded by a ':' to the name of the specified
  649  * interrupt handler.
  650  */
  651 int
  652 intr_event_describe_handler(struct intr_event *ie, void *cookie,
  653     const char *descr)
  654 {
  655         struct intr_handler *ih;
  656         size_t space;
  657         char *start;
  658 
  659         mtx_lock(&ie->ie_lock);
  660 #ifdef INVARIANTS
  661         CK_SLIST_FOREACH(ih, &ie->ie_handlers, ih_next) {
  662                 if (ih == cookie)
  663                         break;
  664         }
  665         if (ih == NULL) {
  666                 mtx_unlock(&ie->ie_lock);
  667                 panic("handler %p not found in interrupt event %p", cookie, ie);
  668         }
  669 #endif
  670         ih = cookie;
  671 
  672         /*
  673          * Look for an existing description by checking for an
  674          * existing ":".  This assumes device names do not include
  675          * colons.  If one is found, prepare to insert the new
  676          * description at that point.  If one is not found, find the
  677          * end of the name to use as the insertion point.
  678          */
  679         start = strchr(ih->ih_name, ':');
  680         if (start == NULL)
  681                 start = strchr(ih->ih_name, 0);
  682 
  683         /*
  684          * See if there is enough remaining room in the string for the
  685          * description + ":".  The "- 1" leaves room for the trailing
  686          * '\0'.  The "+ 1" accounts for the colon.
  687          */
  688         space = sizeof(ih->ih_name) - (start - ih->ih_name) - 1;
  689         if (strlen(descr) + 1 > space) {
  690                 mtx_unlock(&ie->ie_lock);
  691                 return (ENOSPC);
  692         }
  693 
  694         /* Append a colon followed by the description. */
  695         *start = ':';
  696         strcpy(start + 1, descr);
  697         intr_event_update(ie);
  698         mtx_unlock(&ie->ie_lock);
  699         return (0);
  700 }
  701 
  702 /*
  703  * Return the ie_source field from the intr_event an intr_handler is
  704  * associated with.
  705  */
  706 void *
  707 intr_handler_source(void *cookie)
  708 {
  709         struct intr_handler *ih;
  710         struct intr_event *ie;
  711 
  712         ih = (struct intr_handler *)cookie;
  713         if (ih == NULL)
  714                 return (NULL);
  715         ie = ih->ih_event;
  716         KASSERT(ie != NULL,
  717             ("interrupt handler \"%s\" has a NULL interrupt event",
  718             ih->ih_name));
  719         return (ie->ie_source);
  720 }
  721 
  722 /*
  723  * If intr_event_handle() is running in the ISR context at the time of the call,
  724  * then wait for it to complete.
  725  */
  726 static void
  727 intr_event_barrier(struct intr_event *ie)
  728 {
  729         int phase;
  730 
  731         mtx_assert(&ie->ie_lock, MA_OWNED);
  732         phase = ie->ie_phase;
  733 
  734         /*
  735          * Switch phase to direct future interrupts to the other active counter.
  736          * Make sure that any preceding stores are visible before the switch.
  737          */
  738         KASSERT(ie->ie_active[!phase] == 0, ("idle phase has activity"));
  739         atomic_store_rel_int(&ie->ie_phase, !phase);
  740 
  741         /*
  742          * This code cooperates with wait-free iteration of ie_handlers
  743          * in intr_event_handle.
  744          * Make sure that the removal and the phase update are not reordered
  745          * with the active count check.
  746          * Note that no combination of acquire and release fences can provide
  747          * that guarantee as Store->Load sequences can always be reordered.
  748          */
  749         atomic_thread_fence_seq_cst();
  750 
  751         /*
  752          * Now wait on the inactive phase.
  753          * The acquire fence is needed so that all post-barrier accesses
  754          * are after the check.
  755          */
  756         while (ie->ie_active[phase] > 0)
  757                 cpu_spinwait();
  758         atomic_thread_fence_acq();
  759 }
  760 
  761 static void
  762 intr_handler_barrier(struct intr_handler *handler)
  763 {
  764         struct intr_event *ie;
  765 
  766         ie = handler->ih_event;
  767         mtx_assert(&ie->ie_lock, MA_OWNED);
  768         KASSERT((handler->ih_flags & IH_DEAD) == 0,
  769             ("update for a removed handler"));
  770 
  771         if (ie->ie_thread == NULL) {
  772                 intr_event_barrier(ie);
  773                 return;
  774         }
  775         if ((handler->ih_flags & IH_CHANGED) == 0) {
  776                 handler->ih_flags |= IH_CHANGED;
  777                 intr_event_schedule_thread(ie);
  778         }
  779         while ((handler->ih_flags & IH_CHANGED) != 0)
  780                 msleep(handler, &ie->ie_lock, 0, "ih_barr", 0);
  781 }
  782 
  783 /*
  784  * Sleep until an ithread finishes executing an interrupt handler.
  785  *
  786  * XXX Doesn't currently handle interrupt filters or fast interrupt
  787  * handlers. This is intended for LinuxKPI drivers only.
  788  * Do not use in BSD code.
  789  */
  790 void
  791 _intr_drain(int irq)
  792 {
  793         struct intr_event *ie;
  794         struct intr_thread *ithd;
  795         struct thread *td;
  796 
  797         ie = intr_lookup(irq);
  798         if (ie == NULL)
  799                 return;
  800         if (ie->ie_thread == NULL)
  801                 return;
  802         ithd = ie->ie_thread;
  803         td = ithd->it_thread;
  804         /*
  805          * We set the flag and wait for it to be cleared to avoid
  806          * long delays with potentially busy interrupt handlers
  807          * were we to only sample TD_AWAITING_INTR() every tick.
  808          */
  809         thread_lock(td);
  810         if (!TD_AWAITING_INTR(td)) {
  811                 ithd->it_flags |= IT_WAIT;
  812                 while (ithd->it_flags & IT_WAIT) {
  813                         thread_unlock(td);
  814                         pause("idrain", 1);
  815                         thread_lock(td);
  816                 }
  817         }
  818         thread_unlock(td);
  819         return;
  820 }
  821 
  822 int
  823 intr_event_remove_handler(void *cookie)
  824 {
  825         struct intr_handler *handler = (struct intr_handler *)cookie;
  826         struct intr_event *ie;
  827         struct intr_handler *ih;
  828         struct intr_handler **prevptr;
  829 #ifdef notyet
  830         int dead;
  831 #endif
  832 
  833         if (handler == NULL)
  834                 return (EINVAL);
  835         ie = handler->ih_event;
  836         KASSERT(ie != NULL,
  837             ("interrupt handler \"%s\" has a NULL interrupt event",
  838             handler->ih_name));
  839 
  840         mtx_lock(&ie->ie_lock);
  841         CTR3(KTR_INTR, "%s: removing %s from %s", __func__, handler->ih_name,
  842             ie->ie_name);
  843         CK_SLIST_FOREACH_PREVPTR(ih, prevptr, &ie->ie_handlers, ih_next) {
  844                 if (ih == handler)
  845                         break;
  846         }
  847         if (ih == NULL) {
  848                 panic("interrupt handler \"%s\" not found in "
  849                     "interrupt event \"%s\"", handler->ih_name, ie->ie_name);
  850         }
  851 
  852         /*
  853          * If there is no ithread, then directly remove the handler.  Note that
  854          * intr_event_handle() iterates ie_handlers in a lock-less fashion, so
  855          * care needs to be taken to keep ie_handlers consistent and to free
  856          * the removed handler only when ie_handlers is quiescent.
  857          */
  858         if (ie->ie_thread == NULL) {
  859                 CK_SLIST_REMOVE_PREVPTR(prevptr, ih, ih_next);
  860                 intr_event_barrier(ie);
  861                 intr_event_update(ie);
  862                 mtx_unlock(&ie->ie_lock);
  863                 free(handler, M_ITHREAD);
  864                 return (0);
  865         }
  866 
  867         /*
  868          * Let the interrupt thread do the job.
  869          * The interrupt source is disabled when the interrupt thread is
  870          * running, so it does not have to worry about interaction with
  871          * intr_event_handle().
  872          */
  873         KASSERT((handler->ih_flags & IH_DEAD) == 0,
  874             ("duplicate handle remove"));
  875         handler->ih_flags |= IH_DEAD;
  876         intr_event_schedule_thread(ie);
  877         while (handler->ih_flags & IH_DEAD)
  878                 msleep(handler, &ie->ie_lock, 0, "iev_rmh", 0);
  879         intr_event_update(ie);
  880 
  881 #ifdef notyet
  882         /*
  883          * XXX: This could be bad in the case of ppbus(8).  Also, I think
  884          * this could lead to races of stale data when servicing an
  885          * interrupt.
  886          */
  887         dead = 1;
  888         CK_SLIST_FOREACH(ih, &ie->ie_handlers, ih_next) {
  889                 if (ih->ih_handler != NULL) {
  890                         dead = 0;
  891                         break;
  892                 }
  893         }
  894         if (dead) {
  895                 ithread_destroy(ie->ie_thread);
  896                 ie->ie_thread = NULL;
  897         }
  898 #endif
  899         mtx_unlock(&ie->ie_lock);
  900         free(handler, M_ITHREAD);
  901         return (0);
  902 }
  903 
  904 int
  905 intr_event_suspend_handler(void *cookie)
  906 {
  907         struct intr_handler *handler = (struct intr_handler *)cookie;
  908         struct intr_event *ie;
  909 
  910         if (handler == NULL)
  911                 return (EINVAL);
  912         ie = handler->ih_event;
  913         KASSERT(ie != NULL,
  914             ("interrupt handler \"%s\" has a NULL interrupt event",
  915             handler->ih_name));
  916         mtx_lock(&ie->ie_lock);
  917         handler->ih_flags |= IH_SUSP;
  918         intr_handler_barrier(handler);
  919         mtx_unlock(&ie->ie_lock);
  920         return (0);
  921 }
  922 
  923 int
  924 intr_event_resume_handler(void *cookie)
  925 {
  926         struct intr_handler *handler = (struct intr_handler *)cookie;
  927         struct intr_event *ie;
  928 
  929         if (handler == NULL)
  930                 return (EINVAL);
  931         ie = handler->ih_event;
  932         KASSERT(ie != NULL,
  933             ("interrupt handler \"%s\" has a NULL interrupt event",
  934             handler->ih_name));
  935 
  936         /*
  937          * intr_handler_barrier() acts not only as a barrier,
  938          * it also allows to check for any pending interrupts.
  939          */
  940         mtx_lock(&ie->ie_lock);
  941         handler->ih_flags &= ~IH_SUSP;
  942         intr_handler_barrier(handler);
  943         mtx_unlock(&ie->ie_lock);
  944         return (0);
  945 }
  946 
  947 static int
  948 intr_event_schedule_thread(struct intr_event *ie)
  949 {
  950         struct intr_entropy entropy;
  951         struct intr_thread *it;
  952         struct thread *td;
  953         struct thread *ctd;
  954 
  955         /*
  956          * If no ithread or no handlers, then we have a stray interrupt.
  957          */
  958         if (ie == NULL || CK_SLIST_EMPTY(&ie->ie_handlers) ||
  959             ie->ie_thread == NULL)
  960                 return (EINVAL);
  961 
  962         ctd = curthread;
  963         it = ie->ie_thread;
  964         td = it->it_thread;
  965 
  966         /*
  967          * If any of the handlers for this ithread claim to be good
  968          * sources of entropy, then gather some.
  969          */
  970         if (ie->ie_hflags & IH_ENTROPY) {
  971                 entropy.event = (uintptr_t)ie;
  972                 entropy.td = ctd;
  973                 random_harvest_queue(&entropy, sizeof(entropy), RANDOM_INTERRUPT);
  974         }
  975 
  976         KASSERT(td->td_proc != NULL, ("ithread %s has no process", ie->ie_name));
  977 
  978         /*
  979          * Set it_need to tell the thread to keep running if it is already
  980          * running.  Then, lock the thread and see if we actually need to
  981          * put it on the runqueue.
  982          *
  983          * Use store_rel to arrange that the store to ih_need in
  984          * swi_sched() is before the store to it_need and prepare for
  985          * transfer of this order to loads in the ithread.
  986          */
  987         atomic_store_rel_int(&it->it_need, 1);
  988         thread_lock(td);
  989         if (TD_AWAITING_INTR(td)) {
  990                 CTR3(KTR_INTR, "%s: schedule pid %d (%s)", __func__, td->td_proc->p_pid,
  991                     td->td_name);
  992                 TD_CLR_IWAIT(td);
  993                 sched_add(td, SRQ_INTR);
  994         } else {
  995                 CTR5(KTR_INTR, "%s: pid %d (%s): it_need %d, state %d",
  996                     __func__, td->td_proc->p_pid, td->td_name, it->it_need, td->td_state);
  997                 thread_unlock(td);
  998         }
  999 
 1000         return (0);
 1001 }
 1002 
 1003 /*
 1004  * Allow interrupt event binding for software interrupt handlers -- a no-op,
 1005  * since interrupts are generated in software rather than being directed by
 1006  * a PIC.
 1007  */
 1008 static int
 1009 swi_assign_cpu(void *arg, int cpu)
 1010 {
 1011 
 1012         return (0);
 1013 }
 1014 
 1015 /*
 1016  * Add a software interrupt handler to a specified event.  If a given event
 1017  * is not specified, then a new event is created.
 1018  */
 1019 int
 1020 swi_add(struct intr_event **eventp, const char *name, driver_intr_t handler,
 1021             void *arg, int pri, enum intr_type flags, void **cookiep)
 1022 {
 1023         struct intr_event *ie;
 1024         int error = 0;
 1025 
 1026         if (flags & INTR_ENTROPY)
 1027                 return (EINVAL);
 1028 
 1029         ie = (eventp != NULL) ? *eventp : NULL;
 1030 
 1031         if (ie != NULL) {
 1032                 if (!(ie->ie_flags & IE_SOFT))
 1033                         return (EINVAL);
 1034         } else {
 1035                 error = intr_event_create(&ie, NULL, IE_SOFT, 0,
 1036                     NULL, NULL, NULL, swi_assign_cpu, "swi%d:", pri);
 1037                 if (error)
 1038                         return (error);
 1039                 if (eventp != NULL)
 1040                         *eventp = ie;
 1041         }
 1042         if (handler != NULL) {
 1043                 error = intr_event_add_handler(ie, name, NULL, handler, arg,
 1044                     PI_SWI(pri), flags, cookiep);
 1045         }
 1046         return (error);
 1047 }
 1048 
 1049 /*
 1050  * Schedule a software interrupt thread.
 1051  */
 1052 void
 1053 swi_sched(void *cookie, int flags)
 1054 {
 1055         struct intr_handler *ih = (struct intr_handler *)cookie;
 1056         struct intr_event *ie = ih->ih_event;
 1057         struct intr_entropy entropy;
 1058         int error __unused;
 1059 
 1060         CTR3(KTR_INTR, "swi_sched: %s %s need=%d", ie->ie_name, ih->ih_name,
 1061             ih->ih_need);
 1062 
 1063         if ((flags & SWI_FROMNMI) == 0) {
 1064                 entropy.event = (uintptr_t)ih;
 1065                 entropy.td = curthread;
 1066                 random_harvest_queue(&entropy, sizeof(entropy), RANDOM_SWI);
 1067         }
 1068 
 1069         /*
 1070          * Set ih_need for this handler so that if the ithread is already
 1071          * running it will execute this handler on the next pass.  Otherwise,
 1072          * it will execute it the next time it runs.
 1073          */
 1074         ih->ih_need = 1;
 1075 
 1076         if (flags & SWI_DELAY)
 1077                 return;
 1078 
 1079         if (flags & SWI_FROMNMI) {
 1080 #if defined(SMP) && (defined(__i386__) || defined(__amd64__))
 1081                 KASSERT(ie == clk_intr_event,
 1082                     ("SWI_FROMNMI used not with clk_intr_event"));
 1083                 ipi_self_from_nmi(IPI_SWI);
 1084 #endif
 1085         } else {
 1086                 VM_CNT_INC(v_soft);
 1087                 error = intr_event_schedule_thread(ie);
 1088                 KASSERT(error == 0, ("stray software interrupt"));
 1089         }
 1090 }
 1091 
 1092 /*
 1093  * Remove a software interrupt handler.  Currently this code does not
 1094  * remove the associated interrupt event if it becomes empty.  Calling code
 1095  * may do so manually via intr_event_destroy(), but that's not really
 1096  * an optimal interface.
 1097  */
 1098 int
 1099 swi_remove(void *cookie)
 1100 {
 1101 
 1102         return (intr_event_remove_handler(cookie));
 1103 }
 1104 
 1105 static void
 1106 intr_event_execute_handlers(struct proc *p, struct intr_event *ie)
 1107 {
 1108         struct intr_handler *ih, *ihn, *ihp;
 1109 
 1110         ihp = NULL;
 1111         CK_SLIST_FOREACH_SAFE(ih, &ie->ie_handlers, ih_next, ihn) {
 1112                 /*
 1113                  * If this handler is marked for death, remove it from
 1114                  * the list of handlers and wake up the sleeper.
 1115                  */
 1116                 if (ih->ih_flags & IH_DEAD) {
 1117                         mtx_lock(&ie->ie_lock);
 1118                         if (ihp == NULL)
 1119                                 CK_SLIST_REMOVE_HEAD(&ie->ie_handlers, ih_next);
 1120                         else
 1121                                 CK_SLIST_REMOVE_AFTER(ihp, ih_next);
 1122                         ih->ih_flags &= ~IH_DEAD;
 1123                         wakeup(ih);
 1124                         mtx_unlock(&ie->ie_lock);
 1125                         continue;
 1126                 }
 1127 
 1128                 /*
 1129                  * Now that we know that the current element won't be removed
 1130                  * update the previous element.
 1131                  */
 1132                 ihp = ih;
 1133 
 1134                 if ((ih->ih_flags & IH_CHANGED) != 0) {
 1135                         mtx_lock(&ie->ie_lock);
 1136                         ih->ih_flags &= ~IH_CHANGED;
 1137                         wakeup(ih);
 1138                         mtx_unlock(&ie->ie_lock);
 1139                 }
 1140 
 1141                 /* Skip filter only handlers */
 1142                 if (ih->ih_handler == NULL)
 1143                         continue;
 1144 
 1145                 /* Skip suspended handlers */
 1146                 if ((ih->ih_flags & IH_SUSP) != 0)
 1147                         continue;
 1148 
 1149                 /*
 1150                  * For software interrupt threads, we only execute
 1151                  * handlers that have their need flag set.  Hardware
 1152                  * interrupt threads always invoke all of their handlers.
 1153                  *
 1154                  * ih_need can only be 0 or 1.  Failed cmpset below
 1155                  * means that there is no request to execute handlers,
 1156                  * so a retry of the cmpset is not needed.
 1157                  */
 1158                 if ((ie->ie_flags & IE_SOFT) != 0 &&
 1159                     atomic_cmpset_int(&ih->ih_need, 1, 0) == 0)
 1160                         continue;
 1161 
 1162                 /* Execute this handler. */
 1163                 CTR6(KTR_INTR, "%s: pid %d exec %p(%p) for %s flg=%x",
 1164                     __func__, p->p_pid, (void *)ih->ih_handler, 
 1165                     ih->ih_argument, ih->ih_name, ih->ih_flags);
 1166 
 1167                 if (!(ih->ih_flags & IH_MPSAFE))
 1168                         mtx_lock(&Giant);
 1169                 ih->ih_handler(ih->ih_argument);
 1170                 if (!(ih->ih_flags & IH_MPSAFE))
 1171                         mtx_unlock(&Giant);
 1172         }
 1173 }
 1174 
 1175 static void
 1176 ithread_execute_handlers(struct proc *p, struct intr_event *ie)
 1177 {
 1178 
 1179         /* Interrupt handlers should not sleep. */
 1180         if (!(ie->ie_flags & IE_SOFT))
 1181                 THREAD_NO_SLEEPING();
 1182         intr_event_execute_handlers(p, ie);
 1183         if (!(ie->ie_flags & IE_SOFT))
 1184                 THREAD_SLEEPING_OK();
 1185 
 1186         /*
 1187          * Interrupt storm handling:
 1188          *
 1189          * If this interrupt source is currently storming, then throttle
 1190          * it to only fire the handler once  per clock tick.
 1191          *
 1192          * If this interrupt source is not currently storming, but the
 1193          * number of back to back interrupts exceeds the storm threshold,
 1194          * then enter storming mode.
 1195          */
 1196         if (intr_storm_threshold != 0 && ie->ie_count >= intr_storm_threshold &&
 1197             !(ie->ie_flags & IE_SOFT)) {
 1198                 /* Report the message only once every second. */
 1199                 if (ppsratecheck(&ie->ie_warntm, &ie->ie_warncnt, 1)) {
 1200                         printf(
 1201         "interrupt storm detected on \"%s\"; throttling interrupt source\n",
 1202                             ie->ie_name);
 1203                 }
 1204                 pause("istorm", 1);
 1205         } else
 1206                 ie->ie_count++;
 1207 
 1208         /*
 1209          * Now that all the handlers have had a chance to run, reenable
 1210          * the interrupt source.
 1211          */
 1212         if (ie->ie_post_ithread != NULL)
 1213                 ie->ie_post_ithread(ie->ie_source);
 1214 }
 1215 
 1216 /*
 1217  * This is the main code for interrupt threads.
 1218  */
 1219 static void
 1220 ithread_loop(void *arg)
 1221 {
 1222         struct epoch_tracker et;
 1223         struct intr_thread *ithd;
 1224         struct intr_event *ie;
 1225         struct thread *td;
 1226         struct proc *p;
 1227         int wake, epoch_count;
 1228         bool needs_epoch;
 1229 
 1230         td = curthread;
 1231         p = td->td_proc;
 1232         ithd = (struct intr_thread *)arg;
 1233         KASSERT(ithd->it_thread == td,
 1234             ("%s: ithread and proc linkage out of sync", __func__));
 1235         ie = ithd->it_event;
 1236         ie->ie_count = 0;
 1237         wake = 0;
 1238 
 1239         /*
 1240          * As long as we have interrupts outstanding, go through the
 1241          * list of handlers, giving each one a go at it.
 1242          */
 1243         for (;;) {
 1244                 /*
 1245                  * If we are an orphaned thread, then just die.
 1246                  */
 1247                 if (ithd->it_flags & IT_DEAD) {
 1248                         CTR3(KTR_INTR, "%s: pid %d (%s) exiting", __func__,
 1249                             p->p_pid, td->td_name);
 1250                         free(ithd, M_ITHREAD);
 1251                         kthread_exit();
 1252                 }
 1253 
 1254                 /*
 1255                  * Service interrupts.  If another interrupt arrives while
 1256                  * we are running, it will set it_need to note that we
 1257                  * should make another pass.
 1258                  *
 1259                  * The load_acq part of the following cmpset ensures
 1260                  * that the load of ih_need in ithread_execute_handlers()
 1261                  * is ordered after the load of it_need here.
 1262                  */
 1263                 needs_epoch =
 1264                     (atomic_load_int(&ie->ie_hflags) & IH_NET) != 0;
 1265                 if (needs_epoch) {
 1266                         epoch_count = 0;
 1267                         NET_EPOCH_ENTER(et);
 1268                 }
 1269                 while (atomic_cmpset_acq_int(&ithd->it_need, 1, 0) != 0) {
 1270                         ithread_execute_handlers(p, ie);
 1271                         if (needs_epoch &&
 1272                             ++epoch_count >= intr_epoch_batch) {
 1273                                 NET_EPOCH_EXIT(et);
 1274                                 epoch_count = 0;
 1275                                 NET_EPOCH_ENTER(et);
 1276                         }
 1277                 }
 1278                 if (needs_epoch)
 1279                         NET_EPOCH_EXIT(et);
 1280                 WITNESS_WARN(WARN_PANIC, NULL, "suspending ithread");
 1281                 mtx_assert(&Giant, MA_NOTOWNED);
 1282 
 1283                 /*
 1284                  * Processed all our interrupts.  Now get the sched
 1285                  * lock.  This may take a while and it_need may get
 1286                  * set again, so we have to check it again.
 1287                  */
 1288                 thread_lock(td);
 1289                 if (atomic_load_acq_int(&ithd->it_need) == 0 &&
 1290                     (ithd->it_flags & (IT_DEAD | IT_WAIT)) == 0) {
 1291                         TD_SET_IWAIT(td);
 1292                         ie->ie_count = 0;
 1293                         mi_switch(SW_VOL | SWT_IWAIT);
 1294                 } else {
 1295                         if (ithd->it_flags & IT_WAIT) {
 1296                                 wake = 1;
 1297                                 ithd->it_flags &= ~IT_WAIT;
 1298                         }
 1299                         thread_unlock(td);
 1300                 }
 1301                 if (wake) {
 1302                         wakeup(ithd);
 1303                         wake = 0;
 1304                 }
 1305         }
 1306 }
 1307 
 1308 /*
 1309  * Main interrupt handling body.
 1310  *
 1311  * Input:
 1312  * o ie:                        the event connected to this interrupt.
 1313  * o frame:                     some archs (i.e. i386) pass a frame to some.
 1314  *                              handlers as their main argument.
 1315  * Return value:
 1316  * o 0:                         everything ok.
 1317  * o EINVAL:                    stray interrupt.
 1318  */
 1319 int
 1320 intr_event_handle(struct intr_event *ie, struct trapframe *frame)
 1321 {
 1322         struct intr_handler *ih;
 1323         struct trapframe *oldframe;
 1324         struct thread *td;
 1325         int phase;
 1326         int ret;
 1327         bool filter, thread;
 1328 
 1329         td = curthread;
 1330 
 1331 #ifdef KSTACK_USAGE_PROF
 1332         intr_prof_stack_use(td, frame);
 1333 #endif
 1334 
 1335         /* An interrupt with no event or handlers is a stray interrupt. */
 1336         if (ie == NULL || CK_SLIST_EMPTY(&ie->ie_handlers))
 1337                 return (EINVAL);
 1338 
 1339         /*
 1340          * Execute fast interrupt handlers directly.
 1341          * To support clock handlers, if a handler registers
 1342          * with a NULL argument, then we pass it a pointer to
 1343          * a trapframe as its argument.
 1344          */
 1345         td->td_intr_nesting_level++;
 1346         filter = false;
 1347         thread = false;
 1348         ret = 0;
 1349         critical_enter();
 1350         oldframe = td->td_intr_frame;
 1351         td->td_intr_frame = frame;
 1352 
 1353         phase = ie->ie_phase;
 1354         atomic_add_int(&ie->ie_active[phase], 1);
 1355 
 1356         /*
 1357          * This fence is required to ensure that no later loads are
 1358          * re-ordered before the ie_active store.
 1359          */
 1360         atomic_thread_fence_seq_cst();
 1361 
 1362         CK_SLIST_FOREACH(ih, &ie->ie_handlers, ih_next) {
 1363                 if ((ih->ih_flags & IH_SUSP) != 0)
 1364                         continue;
 1365                 if ((ie->ie_flags & IE_SOFT) != 0 && ih->ih_need == 0)
 1366                         continue;
 1367                 if (ih->ih_filter == NULL) {
 1368                         thread = true;
 1369                         continue;
 1370                 }
 1371                 CTR4(KTR_INTR, "%s: exec %p(%p) for %s", __func__,
 1372                     ih->ih_filter, ih->ih_argument == NULL ? frame :
 1373                     ih->ih_argument, ih->ih_name);
 1374                 if (ih->ih_argument == NULL)
 1375                         ret = ih->ih_filter(frame);
 1376                 else
 1377                         ret = ih->ih_filter(ih->ih_argument);
 1378                 KASSERT(ret == FILTER_STRAY ||
 1379                     ((ret & (FILTER_SCHEDULE_THREAD | FILTER_HANDLED)) != 0 &&
 1380                     (ret & ~(FILTER_SCHEDULE_THREAD | FILTER_HANDLED)) == 0),
 1381                     ("%s: incorrect return value %#x from %s", __func__, ret,
 1382                     ih->ih_name));
 1383                 filter = filter || ret == FILTER_HANDLED;
 1384 
 1385                 /*
 1386                  * Wrapper handler special handling:
 1387                  *
 1388                  * in some particular cases (like pccard and pccbb),
 1389                  * the _real_ device handler is wrapped in a couple of
 1390                  * functions - a filter wrapper and an ithread wrapper.
 1391                  * In this case (and just in this case), the filter wrapper
 1392                  * could ask the system to schedule the ithread and mask
 1393                  * the interrupt source if the wrapped handler is composed
 1394                  * of just an ithread handler.
 1395                  *
 1396                  * TODO: write a generic wrapper to avoid people rolling
 1397                  * their own.
 1398                  */
 1399                 if (!thread) {
 1400                         if (ret == FILTER_SCHEDULE_THREAD)
 1401                                 thread = true;
 1402                 }
 1403         }
 1404         atomic_add_rel_int(&ie->ie_active[phase], -1);
 1405 
 1406         td->td_intr_frame = oldframe;
 1407 
 1408         if (thread) {
 1409                 if (ie->ie_pre_ithread != NULL)
 1410                         ie->ie_pre_ithread(ie->ie_source);
 1411         } else {
 1412                 if (ie->ie_post_filter != NULL)
 1413                         ie->ie_post_filter(ie->ie_source);
 1414         }
 1415 
 1416         /* Schedule the ithread if needed. */
 1417         if (thread) {
 1418                 int error __unused;
 1419 
 1420                 error =  intr_event_schedule_thread(ie);
 1421                 KASSERT(error == 0, ("bad stray interrupt"));
 1422         }
 1423         critical_exit();
 1424         td->td_intr_nesting_level--;
 1425 #ifdef notyet
 1426         /* The interrupt is not aknowledged by any filter and has no ithread. */
 1427         if (!thread && !filter)
 1428                 return (EINVAL);
 1429 #endif
 1430         return (0);
 1431 }
 1432 
 1433 #ifdef DDB
 1434 /*
 1435  * Dump details about an interrupt handler
 1436  */
 1437 static void
 1438 db_dump_intrhand(struct intr_handler *ih)
 1439 {
 1440         int comma;
 1441 
 1442         db_printf("\t%-10s ", ih->ih_name);
 1443         switch (ih->ih_pri) {
 1444         case PI_REALTIME:
 1445                 db_printf("CLK ");
 1446                 break;
 1447         case PI_AV:
 1448                 db_printf("AV  ");
 1449                 break;
 1450         case PI_TTY:
 1451                 db_printf("TTY ");
 1452                 break;
 1453         case PI_NET:
 1454                 db_printf("NET ");
 1455                 break;
 1456         case PI_DISK:
 1457                 db_printf("DISK");
 1458                 break;
 1459         case PI_DULL:
 1460                 db_printf("DULL");
 1461                 break;
 1462         default:
 1463                 if (ih->ih_pri >= PI_SOFT)
 1464                         db_printf("SWI ");
 1465                 else
 1466                         db_printf("%4u", ih->ih_pri);
 1467                 break;
 1468         }
 1469         db_printf(" ");
 1470         if (ih->ih_filter != NULL) {
 1471                 db_printf("[F]");
 1472                 db_printsym((uintptr_t)ih->ih_filter, DB_STGY_PROC);
 1473         }
 1474         if (ih->ih_handler != NULL) {
 1475                 if (ih->ih_filter != NULL)
 1476                         db_printf(",");
 1477                 db_printf("[H]");
 1478                 db_printsym((uintptr_t)ih->ih_handler, DB_STGY_PROC);
 1479         }
 1480         db_printf("(%p)", ih->ih_argument);
 1481         if (ih->ih_need ||
 1482             (ih->ih_flags & (IH_EXCLUSIVE | IH_ENTROPY | IH_DEAD |
 1483             IH_MPSAFE)) != 0) {
 1484                 db_printf(" {");
 1485                 comma = 0;
 1486                 if (ih->ih_flags & IH_EXCLUSIVE) {
 1487                         if (comma)
 1488                                 db_printf(", ");
 1489                         db_printf("EXCL");
 1490                         comma = 1;
 1491                 }
 1492                 if (ih->ih_flags & IH_ENTROPY) {
 1493                         if (comma)
 1494                                 db_printf(", ");
 1495                         db_printf("ENTROPY");
 1496                         comma = 1;
 1497                 }
 1498                 if (ih->ih_flags & IH_DEAD) {
 1499                         if (comma)
 1500                                 db_printf(", ");
 1501                         db_printf("DEAD");
 1502                         comma = 1;
 1503                 }
 1504                 if (ih->ih_flags & IH_MPSAFE) {
 1505                         if (comma)
 1506                                 db_printf(", ");
 1507                         db_printf("MPSAFE");
 1508                         comma = 1;
 1509                 }
 1510                 if (ih->ih_need) {
 1511                         if (comma)
 1512                                 db_printf(", ");
 1513                         db_printf("NEED");
 1514                 }
 1515                 db_printf("}");
 1516         }
 1517         db_printf("\n");
 1518 }
 1519 
 1520 /*
 1521  * Dump details about a event.
 1522  */
 1523 void
 1524 db_dump_intr_event(struct intr_event *ie, int handlers)
 1525 {
 1526         struct intr_handler *ih;
 1527         struct intr_thread *it;
 1528         int comma;
 1529 
 1530         db_printf("%s ", ie->ie_fullname);
 1531         it = ie->ie_thread;
 1532         if (it != NULL)
 1533                 db_printf("(pid %d)", it->it_thread->td_proc->p_pid);
 1534         else
 1535                 db_printf("(no thread)");
 1536         if ((ie->ie_flags & (IE_SOFT | IE_ADDING_THREAD)) != 0 ||
 1537             (it != NULL && it->it_need)) {
 1538                 db_printf(" {");
 1539                 comma = 0;
 1540                 if (ie->ie_flags & IE_SOFT) {
 1541                         db_printf("SOFT");
 1542                         comma = 1;
 1543                 }
 1544                 if (ie->ie_flags & IE_ADDING_THREAD) {
 1545                         if (comma)
 1546                                 db_printf(", ");
 1547                         db_printf("ADDING_THREAD");
 1548                         comma = 1;
 1549                 }
 1550                 if (it != NULL && it->it_need) {
 1551                         if (comma)
 1552                                 db_printf(", ");
 1553                         db_printf("NEED");
 1554                 }
 1555                 db_printf("}");
 1556         }
 1557         db_printf("\n");
 1558 
 1559         if (handlers)
 1560                 CK_SLIST_FOREACH(ih, &ie->ie_handlers, ih_next)
 1561                     db_dump_intrhand(ih);
 1562 }
 1563 
 1564 /*
 1565  * Dump data about interrupt handlers
 1566  */
 1567 DB_SHOW_COMMAND(intr, db_show_intr)
 1568 {
 1569         struct intr_event *ie;
 1570         int all, verbose;
 1571 
 1572         verbose = strchr(modif, 'v') != NULL;
 1573         all = strchr(modif, 'a') != NULL;
 1574         TAILQ_FOREACH(ie, &event_list, ie_list) {
 1575                 if (!all && CK_SLIST_EMPTY(&ie->ie_handlers))
 1576                         continue;
 1577                 db_dump_intr_event(ie, verbose);
 1578                 if (db_pager_quit)
 1579                         break;
 1580         }
 1581 }
 1582 #endif /* DDB */
 1583 
 1584 /*
 1585  * Start standard software interrupt threads
 1586  */
 1587 static void
 1588 start_softintr(void *dummy)
 1589 {
 1590 
 1591         if (swi_add(&clk_intr_event, "clk", NULL, NULL, SWI_CLOCK,
 1592             INTR_MPSAFE, NULL))
 1593                 panic("died while creating clk swi ithread");
 1594 }
 1595 SYSINIT(start_softintr, SI_SUB_SOFTINTR, SI_ORDER_FIRST, start_softintr,
 1596     NULL);
 1597 
 1598 /*
 1599  * Sysctls used by systat and others: hw.intrnames and hw.intrcnt.
 1600  * The data for this machine dependent, and the declarations are in machine
 1601  * dependent code.  The layout of intrnames and intrcnt however is machine
 1602  * independent.
 1603  *
 1604  * We do not know the length of intrcnt and intrnames at compile time, so
 1605  * calculate things at run time.
 1606  */
 1607 static int
 1608 sysctl_intrnames(SYSCTL_HANDLER_ARGS)
 1609 {
 1610         return (sysctl_handle_opaque(oidp, intrnames, sintrnames, req));
 1611 }
 1612 
 1613 SYSCTL_PROC(_hw, OID_AUTO, intrnames,
 1614     CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
 1615     sysctl_intrnames, "",
 1616     "Interrupt Names");
 1617 
 1618 static int
 1619 sysctl_intrcnt(SYSCTL_HANDLER_ARGS)
 1620 {
 1621 #ifdef SCTL_MASK32
 1622         uint32_t *intrcnt32;
 1623         unsigned i;
 1624         int error;
 1625 
 1626         if (req->flags & SCTL_MASK32) {
 1627                 if (!req->oldptr)
 1628                         return (sysctl_handle_opaque(oidp, NULL, sintrcnt / 2, req));
 1629                 intrcnt32 = malloc(sintrcnt / 2, M_TEMP, M_NOWAIT);
 1630                 if (intrcnt32 == NULL)
 1631                         return (ENOMEM);
 1632                 for (i = 0; i < sintrcnt / sizeof (u_long); i++)
 1633                         intrcnt32[i] = intrcnt[i];
 1634                 error = sysctl_handle_opaque(oidp, intrcnt32, sintrcnt / 2, req);
 1635                 free(intrcnt32, M_TEMP);
 1636                 return (error);
 1637         }
 1638 #endif
 1639         return (sysctl_handle_opaque(oidp, intrcnt, sintrcnt, req));
 1640 }
 1641 
 1642 SYSCTL_PROC(_hw, OID_AUTO, intrcnt,
 1643     CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
 1644     sysctl_intrcnt, "",
 1645     "Interrupt Counts");
 1646 
 1647 #ifdef DDB
 1648 /*
 1649  * DDB command to dump the interrupt statistics.
 1650  */
 1651 DB_SHOW_COMMAND(intrcnt, db_show_intrcnt)
 1652 {
 1653         u_long *i;
 1654         char *cp;
 1655         u_int j;
 1656 
 1657         cp = intrnames;
 1658         j = 0;
 1659         for (i = intrcnt; j < (sintrcnt / sizeof(u_long)) && !db_pager_quit;
 1660             i++, j++) {
 1661                 if (*cp == '\0')
 1662                         break;
 1663                 if (*i != 0)
 1664                         db_printf("%s\t%lu\n", cp, *i);
 1665                 cp += strlen(cp) + 1;
 1666         }
 1667 }
 1668 #endif

Cache object: a914bbe9b990e4f310c731ded7698091


[ 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.