2 * Copyright (c) 2003 Matthew Dillon <dillon@backplane.com> All rights reserved.
3 * Copyright (c) 1997, Stefan Esser <se@freebsd.org> All rights reserved.
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
8 * 1. Redistributions of source code must retain the above copyright
9 * notice unmodified, this list of conditions, and the following
11 * 2. Redistributions in binary form must reproduce the above copyright
12 * notice, this list of conditions and the following disclaimer in the
13 * documentation and/or other materials provided with the distribution.
15 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
16 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
17 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
18 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
19 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
20 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
21 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
22 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
23 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
24 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
26 * $FreeBSD: src/sys/kern/kern_intr.c,v 1.24.2.1 2001/10/14 20:05:50 luigi Exp $
27 * $DragonFly: src/sys/kern/kern_intr.c,v 1.55 2008/09/01 12:49:00 sephe Exp $
31 #include <sys/param.h>
32 #include <sys/systm.h>
33 #include <sys/malloc.h>
34 #include <sys/kernel.h>
35 #include <sys/sysctl.h>
36 #include <sys/thread.h>
38 #include <sys/thread2.h>
39 #include <sys/random.h>
40 #include <sys/serialize.h>
41 #include <sys/interrupt.h>
43 #include <sys/machintr.h>
45 #include <machine/frame.h>
47 #include <sys/interrupt.h>
51 typedef struct intrec {
53 struct intr_info *info;
59 struct lwkt_serialize *serializer;
64 struct thread i_thread;
65 struct random_softc i_random;
67 long i_count; /* interrupts dispatched */
68 int i_mplock_required;
73 unsigned long i_straycount;
74 } intr_info_ary[MAX_INTS];
76 int max_installed_hard_intr;
77 int max_installed_soft_intr;
79 #define EMERGENCY_INTR_POLLING_FREQ_MAX 20000
81 static int sysctl_emergency_freq(SYSCTL_HANDLER_ARGS);
82 static int sysctl_emergency_enable(SYSCTL_HANDLER_ARGS);
83 static void emergency_intr_timer_callback(systimer_t, struct intrframe *);
84 static void ithread_handler(void *arg);
85 static void ithread_emergency(void *arg);
86 static void report_stray_interrupt(int intr, struct intr_info *info);
87 static void int_moveto_destcpu(int *, int *, int);
88 static void int_moveto_origcpu(int, int);
90 static void intr_get_mplock(void);
93 int intr_info_size = sizeof(intr_info_ary) / sizeof(intr_info_ary[0]);
95 static struct systimer emergency_intr_timer;
96 static struct thread emergency_intr_thread;
98 #define ISTATE_NOTHREAD 0
99 #define ISTATE_NORMAL 1
100 #define ISTATE_LIVELOCKED 2
103 static int intr_mpsafe = 1;
104 static int intr_migrate = 0;
105 static int intr_migrate_count;
106 TUNABLE_INT("kern.intr_mpsafe", &intr_mpsafe);
107 SYSCTL_INT(_kern, OID_AUTO, intr_mpsafe,
108 CTLFLAG_RW, &intr_mpsafe, 0, "Run INTR_MPSAFE handlers without the BGL");
109 SYSCTL_INT(_kern, OID_AUTO, intr_migrate,
110 CTLFLAG_RW, &intr_migrate, 0, "Migrate to cpu holding BGL");
111 SYSCTL_INT(_kern, OID_AUTO, intr_migrate_count,
112 CTLFLAG_RW, &intr_migrate_count, 0, "");
114 static int livelock_limit = 40000;
115 static int livelock_lowater = 20000;
116 static int livelock_debug = -1;
117 SYSCTL_INT(_kern, OID_AUTO, livelock_limit,
118 CTLFLAG_RW, &livelock_limit, 0, "Livelock interrupt rate limit");
119 SYSCTL_INT(_kern, OID_AUTO, livelock_lowater,
120 CTLFLAG_RW, &livelock_lowater, 0, "Livelock low-water mark restore");
121 SYSCTL_INT(_kern, OID_AUTO, livelock_debug,
122 CTLFLAG_RW, &livelock_debug, 0, "Livelock debug intr#");
124 static int emergency_intr_enable = 0; /* emergency interrupt polling */
125 TUNABLE_INT("kern.emergency_intr_enable", &emergency_intr_enable);
126 SYSCTL_PROC(_kern, OID_AUTO, emergency_intr_enable, CTLTYPE_INT | CTLFLAG_RW,
127 0, 0, sysctl_emergency_enable, "I", "Emergency Interrupt Poll Enable");
129 static int emergency_intr_freq = 10; /* emergency polling frequency */
130 TUNABLE_INT("kern.emergency_intr_freq", &emergency_intr_freq);
131 SYSCTL_PROC(_kern, OID_AUTO, emergency_intr_freq, CTLTYPE_INT | CTLFLAG_RW,
132 0, 0, sysctl_emergency_freq, "I", "Emergency Interrupt Poll Frequency");
135 * Sysctl support routines
138 sysctl_emergency_enable(SYSCTL_HANDLER_ARGS)
142 enabled = emergency_intr_enable;
143 error = sysctl_handle_int(oidp, &enabled, 0, req);
144 if (error || req->newptr == NULL)
146 emergency_intr_enable = enabled;
147 if (emergency_intr_enable) {
148 systimer_adjust_periodic(&emergency_intr_timer,
149 emergency_intr_freq);
151 systimer_adjust_periodic(&emergency_intr_timer, 1);
157 sysctl_emergency_freq(SYSCTL_HANDLER_ARGS)
161 phz = emergency_intr_freq;
162 error = sysctl_handle_int(oidp, &phz, 0, req);
163 if (error || req->newptr == NULL)
167 else if (phz > EMERGENCY_INTR_POLLING_FREQ_MAX)
168 phz = EMERGENCY_INTR_POLLING_FREQ_MAX;
170 emergency_intr_freq = phz;
171 if (emergency_intr_enable) {
172 systimer_adjust_periodic(&emergency_intr_timer,
173 emergency_intr_freq);
175 systimer_adjust_periodic(&emergency_intr_timer, 1);
181 * Register an SWI or INTerrupt handler.
184 register_swi(int intr, inthand2_t *handler, void *arg, const char *name,
185 struct lwkt_serialize *serializer)
187 if (intr < FIRST_SOFTINT || intr >= MAX_INTS)
188 panic("register_swi: bad intr %d", intr);
189 return(register_int(intr, handler, arg, name, serializer, 0));
193 register_swi_mp(int intr, inthand2_t *handler, void *arg, const char *name,
194 struct lwkt_serialize *serializer)
196 if (intr < FIRST_SOFTINT || intr >= MAX_INTS)
197 panic("register_swi: bad intr %d", intr);
198 return(register_int(intr, handler, arg, name, serializer, INTR_MPSAFE));
202 register_int(int intr, inthand2_t *handler, void *arg, const char *name,
203 struct lwkt_serialize *serializer, int intr_flags)
205 struct intr_info *info;
206 struct intrec **list;
208 int orig_cpuid, cpuid;
210 if (intr < 0 || intr >= MAX_INTS)
211 panic("register_int: bad intr %d", intr);
214 info = &intr_info_ary[intr];
217 * Construct an interrupt handler record
219 rec = kmalloc(sizeof(struct intrec), M_DEVBUF, M_INTWAIT);
220 rec->name = kmalloc(strlen(name) + 1, M_DEVBUF, M_INTWAIT);
221 strcpy(rec->name, name);
224 rec->handler = handler;
227 rec->intr_flags = intr_flags;
229 rec->serializer = serializer;
232 * Create an emergency polling thread and set up a systimer to wake
235 if (emergency_intr_thread.td_kstack == NULL) {
236 lwkt_create(ithread_emergency, NULL, NULL,
237 &emergency_intr_thread, TDF_STOPREQ|TDF_INTTHREAD, -1,
239 systimer_init_periodic_nq(&emergency_intr_timer,
240 emergency_intr_timer_callback, &emergency_intr_thread,
241 (emergency_intr_enable ? emergency_intr_freq : 1));
244 int_moveto_destcpu(&orig_cpuid, &cpuid, intr);
247 * Create an interrupt thread if necessary, leave it in an unscheduled
250 if (info->i_state == ISTATE_NOTHREAD) {
251 info->i_state = ISTATE_NORMAL;
252 lwkt_create((void *)ithread_handler, (void *)(intptr_t)intr, NULL,
253 &info->i_thread, TDF_STOPREQ|TDF_INTTHREAD|TDF_MPSAFE, -1,
255 if (intr >= FIRST_SOFTINT)
256 lwkt_setpri(&info->i_thread, TDPRI_SOFT_NORM);
258 lwkt_setpri(&info->i_thread, TDPRI_INT_MED);
259 info->i_thread.td_preemptable = lwkt_preempt;
262 list = &info->i_reclist;
265 * Keep track of how many fast and slow interrupts we have.
266 * Set i_mplock_required if any handler in the chain requires
267 * the MP lock to operate.
269 if ((intr_flags & INTR_MPSAFE) == 0)
270 info->i_mplock_required = 1;
271 if (intr_flags & INTR_FAST)
277 * Enable random number generation keying off of this interrupt.
279 if ((intr_flags & INTR_NOENTROPY) == 0 && info->i_random.sc_enabled == 0) {
280 info->i_random.sc_enabled = 1;
281 info->i_random.sc_intr = intr;
285 * Add the record to the interrupt list.
288 while (*list != NULL)
289 list = &(*list)->next;
294 * Update max_installed_hard_intr to make the emergency intr poll
295 * a bit more efficient.
297 if (intr < FIRST_SOFTINT) {
298 if (max_installed_hard_intr <= intr)
299 max_installed_hard_intr = intr + 1;
301 if (max_installed_soft_intr <= intr)
302 max_installed_soft_intr = intr + 1;
306 * Setup the machine level interrupt vector
308 * XXX temporary workaround for some ACPI brokedness. ACPI installs
309 * its interrupt too early, before the IOAPICs have been configured,
310 * which means the IOAPIC is not enabled by the registration of the
311 * ACPI interrupt. Anything else sharing that IRQ will wind up not
312 * being enabled. Temporarily work around the problem by always
313 * installing and enabling on every new interrupt handler, even
314 * if one has already been setup on that irq.
316 if (intr < FIRST_SOFTINT /* && info->i_slow + info->i_fast == 1*/) {
317 if (machintr_vector_setup(intr, intr_flags))
318 kprintf("machintr_vector_setup: failed on irq %d\n", intr);
321 int_moveto_origcpu(orig_cpuid, cpuid);
327 unregister_swi(void *id)
333 unregister_int(void *id)
335 struct intr_info *info;
336 struct intrec **list;
338 int intr, orig_cpuid, cpuid;
340 intr = ((intrec_t)id)->intr;
342 if (intr < 0 || intr >= MAX_INTS)
343 panic("register_int: bad intr %d", intr);
345 info = &intr_info_ary[intr];
347 int_moveto_destcpu(&orig_cpuid, &cpuid, intr);
350 * Remove the interrupt descriptor, adjust the descriptor count,
351 * and teardown the machine level vector if this was the last interrupt.
354 list = &info->i_reclist;
355 while ((rec = *list) != NULL) {
364 if (rec->intr_flags & INTR_FAST)
368 if (intr < FIRST_SOFTINT && info->i_fast + info->i_slow == 0)
369 machintr_vector_teardown(intr);
372 * Clear i_mplock_required if no handlers in the chain require the
375 for (rec0 = info->i_reclist; rec0; rec0 = rec0->next) {
376 if ((rec0->intr_flags & INTR_MPSAFE) == 0)
380 info->i_mplock_required = 0;
385 int_moveto_origcpu(orig_cpuid, cpuid);
391 kfree(rec->name, M_DEVBUF);
392 kfree(rec, M_DEVBUF);
394 kprintf("warning: unregister_int: int %d handler for %s not found\n",
395 intr, ((intrec_t)id)->name);
400 get_registered_name(int intr)
404 if (intr < 0 || intr >= MAX_INTS)
405 panic("register_int: bad intr %d", intr);
407 if ((rec = intr_info_ary[intr].i_reclist) == NULL)
416 count_registered_ints(int intr)
418 struct intr_info *info;
420 if (intr < 0 || intr >= MAX_INTS)
421 panic("register_int: bad intr %d", intr);
422 info = &intr_info_ary[intr];
423 return(info->i_fast + info->i_slow);
427 get_interrupt_counter(int intr)
429 struct intr_info *info;
431 if (intr < 0 || intr >= MAX_INTS)
432 panic("register_int: bad intr %d", intr);
433 info = &intr_info_ary[intr];
434 return(info->i_count);
439 swi_setpriority(int intr, int pri)
441 struct intr_info *info;
443 if (intr < FIRST_SOFTINT || intr >= MAX_INTS)
444 panic("register_swi: bad intr %d", intr);
445 info = &intr_info_ary[intr];
446 if (info->i_state != ISTATE_NOTHREAD)
447 lwkt_setpri(&info->i_thread, pri);
451 register_randintr(int intr)
453 struct intr_info *info;
455 if (intr < 0 || intr >= MAX_INTS)
456 panic("register_randintr: bad intr %d", intr);
457 info = &intr_info_ary[intr];
458 info->i_random.sc_intr = intr;
459 info->i_random.sc_enabled = 1;
463 unregister_randintr(int intr)
465 struct intr_info *info;
467 if (intr < 0 || intr >= MAX_INTS)
468 panic("register_swi: bad intr %d", intr);
469 info = &intr_info_ary[intr];
470 info->i_random.sc_enabled = -1;
474 next_registered_randintr(int intr)
476 struct intr_info *info;
478 if (intr < 0 || intr >= MAX_INTS)
479 panic("register_swi: bad intr %d", intr);
480 while (intr < MAX_INTS) {
481 info = &intr_info_ary[intr];
482 if (info->i_random.sc_enabled > 0)
490 * Dispatch an interrupt. If there's nothing to do we have a stray
491 * interrupt and can just return, leaving the interrupt masked.
493 * We need to schedule the interrupt and set its i_running bit. If
494 * we are not on the interrupt thread's cpu we have to send a message
495 * to the correct cpu that will issue the desired action (interlocking
496 * with the interrupt thread's critical section). We do NOT attempt to
497 * reschedule interrupts whos i_running bit is already set because
498 * this would prematurely wakeup a livelock-limited interrupt thread.
500 * i_running is only tested/set on the same cpu as the interrupt thread.
502 * We are NOT in a critical section, which will allow the scheduled
503 * interrupt to preempt us. The MP lock might *NOT* be held here.
508 sched_ithd_remote(void *arg)
510 sched_ithd((int)(intptr_t)arg);
518 struct intr_info *info;
520 info = &intr_info_ary[intr];
523 if (info->i_state != ISTATE_NOTHREAD) {
524 if (info->i_reclist == NULL) {
525 report_stray_interrupt(intr, info);
528 if (info->i_thread.td_gd == mycpu) {
529 if (info->i_running == 0) {
531 if (info->i_state != ISTATE_LIVELOCKED)
532 lwkt_schedule(&info->i_thread); /* MIGHT PREEMPT */
535 lwkt_send_ipiq(info->i_thread.td_gd,
536 sched_ithd_remote, (void *)(intptr_t)intr);
539 if (info->i_running == 0) {
541 if (info->i_state != ISTATE_LIVELOCKED)
542 lwkt_schedule(&info->i_thread); /* MIGHT PREEMPT */
547 report_stray_interrupt(intr, info);
552 report_stray_interrupt(int intr, struct intr_info *info)
554 ++info->i_straycount;
555 if (info->i_straycount < 10) {
556 if (info->i_errorticks == ticks)
558 info->i_errorticks = ticks;
559 kprintf("sched_ithd: stray interrupt %d on cpu %d\n",
561 } else if (info->i_straycount == 10) {
562 kprintf("sched_ithd: %ld stray interrupts %d on cpu %d - "
563 "there will be no further reports\n",
564 info->i_straycount, intr, mycpuid);
569 * This is run from a periodic SYSTIMER (and thus must be MP safe, the BGL
570 * might not be held).
573 ithread_livelock_wakeup(systimer_t st)
575 struct intr_info *info;
577 info = &intr_info_ary[(int)(intptr_t)st->data];
578 if (info->i_state != ISTATE_NOTHREAD)
579 lwkt_schedule(&info->i_thread);
583 * Schedule ithread within fast intr handler
585 * XXX Protect sched_ithd() call with gd_intr_nesting_level?
586 * Interrupts aren't enabled, but still...
589 ithread_fast_sched(int intr, thread_t td)
594 * We are already in critical section, exit it now to
599 crit_enter_quick(td);
605 * This function is called directly from the ICU or APIC vector code assembly
606 * to process an interrupt. The critical section and interrupt deferral
607 * checks have already been done but the function is entered WITHOUT
608 * a critical section held. The BGL may or may not be held.
610 * Must return non-zero if we do not want the vector code to re-enable
611 * the interrupt (which we don't if we have to schedule the interrupt)
613 int ithread_fast_handler(struct intrframe *frame);
616 ithread_fast_handler(struct intrframe *frame)
619 struct intr_info *info;
620 struct intrec **list;
625 intrec_t rec, next_rec;
629 intr = frame->if_vec;
633 /* We must be in critical section. */
634 KKASSERT(td->td_pri >= TDPRI_CRIT);
636 info = &intr_info_ary[intr];
639 * If we are not processing any FAST interrupts, just schedule the thing.
641 if (info->i_fast == 0) {
643 ithread_fast_sched(intr, td);
648 * This should not normally occur since interrupts ought to be
649 * masked if the ithread has been scheduled or is running.
655 * Bump the interrupt nesting level to process any FAST interrupts.
656 * Obtain the MP lock as necessary. If the MP lock cannot be obtained,
657 * schedule the interrupt thread to deal with the issue instead.
659 * To reduce overhead, just leave the MP lock held once it has been
662 ++gd->gd_intr_nesting_level;
664 must_schedule = info->i_slow;
669 list = &info->i_reclist;
670 for (rec = *list; rec; rec = next_rec) {
671 next_rec = rec->next; /* rec may be invalid after call */
673 if (rec->intr_flags & INTR_FAST) {
675 if ((rec->intr_flags & INTR_MPSAFE) == 0 && got_mplock == 0) {
676 if (try_mplock() == 0) {
677 /* Couldn't get the MP lock; just schedule it. */
684 if (rec->serializer) {
685 must_schedule += lwkt_serialize_handler_try(
686 rec->serializer, rec->handler,
687 rec->argument, frame);
689 rec->handler(rec->argument, frame);
697 --gd->gd_intr_nesting_level;
704 * If we had a problem, or mixed fast and slow interrupt handlers are
705 * registered, schedule the ithread to catch the missed records (it
706 * will just re-run all of them). A return value of 0 indicates that
707 * all handlers have been run and the interrupt can be re-enabled, and
708 * a non-zero return indicates that the interrupt thread controls
711 if (must_schedule > 0)
712 ithread_fast_sched(intr, td);
713 else if (must_schedule == 0)
715 return(must_schedule);
719 * Interrupt threads run this as their main loop.
721 * The handler begins execution outside a critical section and with the BGL
724 * The i_running state starts at 0. When an interrupt occurs, the hardware
725 * interrupt is disabled and sched_ithd() The HW interrupt remains disabled
726 * until all routines have run. We then call ithread_done() to reenable
727 * the HW interrupt and deschedule us until the next interrupt.
729 * We are responsible for atomically checking i_running and ithread_done()
730 * is responsible for atomically checking for platform-specific delayed
731 * interrupts. i_running for our irq is only set in the context of our cpu,
732 * so a critical section is a sufficient interlock.
734 #define LIVELOCK_TIMEFRAME(freq) ((freq) >> 2) /* 1/4 second */
737 ithread_handler(void *arg)
739 struct intr_info *info;
744 struct intrec **list;
747 struct systimer ill_timer; /* enforced freq. timer */
748 u_int ill_count; /* interrupt livelock counter */
751 intr = (int)(intptr_t)arg;
752 info = &intr_info_ary[intr];
753 list = &info->i_reclist;
756 * The loop must be entered with one critical section held. The thread
757 * is created with TDF_MPSAFE so the MP lock is not held on start.
760 lseconds = gd->gd_time_seconds;
766 * The chain is only considered MPSAFE if all its interrupt handlers
767 * are MPSAFE. However, if intr_mpsafe has been turned off we
768 * always operate with the BGL.
771 if (intr_mpsafe == 0) {
776 } else if (info->i_mplock_required != mpheld) {
777 if (info->i_mplock_required) {
778 KKASSERT(mpheld == 0);
782 KKASSERT(mpheld != 0);
789 * scheduled cpu may have changed, see intr_get_mplock()
795 * If an interrupt is pending, clear i_running and execute the
796 * handlers. Note that certain types of interrupts can re-trigger
797 * and set i_running again.
799 * Each handler is run in a critical section. Note that we run both
800 * FAST and SLOW designated service routines.
802 if (info->i_running) {
807 report_stray_interrupt(intr, info);
809 for (rec = *list; rec; rec = nrec) {
811 if (rec->serializer) {
812 lwkt_serialize_handler_call(rec->serializer, rec->handler,
813 rec->argument, NULL);
815 rec->handler(rec->argument, NULL);
821 * This is our interrupt hook to add rate randomness to the random
824 if (info->i_random.sc_enabled > 0)
825 add_interrupt_randomness(intr);
828 * Unmask the interrupt to allow it to trigger again. This only
829 * applies to certain types of interrupts (typ level interrupts).
830 * This can result in the interrupt retriggering, but the retrigger
831 * will not be processed until we cycle our critical section.
833 * Only unmask interrupts while handlers are installed. It is
834 * possible to hit a situation where no handlers are installed
835 * due to a device driver livelocking and then tearing down its
836 * interrupt on close (the parallel bus being a good example).
839 machintr_intren(intr);
842 * Do a quick exit/enter to catch any higher-priority interrupt
843 * sources, such as the statclock, so thread time accounting
844 * will still work. This may also cause an interrupt to re-trigger.
850 * LIVELOCK STATE MACHINE
852 switch(info->i_state) {
855 * Reset the count each second.
857 if (lseconds != gd->gd_time_seconds) {
858 lseconds = gd->gd_time_seconds;
863 * If we did not exceed the frequency limit, we are done.
864 * If the interrupt has not retriggered we deschedule ourselves.
866 if (ill_count <= livelock_limit) {
867 if (info->i_running == 0) {
869 if (mpheld && intr_migrate) {
874 lwkt_deschedule_self(gd->gd_curthread);
881 * Otherwise we are livelocked. Set up a periodic systimer
882 * to wake the thread up at the limit frequency.
884 kprintf("intr %d at %d/%d hz, livelocked limit engaged!\n",
885 intr, ill_count, livelock_limit);
886 info->i_state = ISTATE_LIVELOCKED;
887 if ((use_limit = livelock_limit) < 100)
889 else if (use_limit > 500000)
891 systimer_init_periodic_nq(&ill_timer, ithread_livelock_wakeup,
892 (void *)(intptr_t)intr, use_limit);
894 case ISTATE_LIVELOCKED:
896 * Wait for our periodic timer to go off. Since the interrupt
897 * has re-armed it can still set i_running, but it will not
898 * reschedule us while we are in a livelocked state.
900 lwkt_deschedule_self(gd->gd_curthread);
904 * Check once a second to see if the livelock condition no
907 if (lseconds != gd->gd_time_seconds) {
908 lseconds = gd->gd_time_seconds;
909 if (ill_count < livelock_lowater) {
910 info->i_state = ISTATE_NORMAL;
911 systimer_del(&ill_timer);
912 kprintf("intr %d at %d/%d hz, livelock removed\n",
913 intr, ill_count, livelock_lowater);
914 } else if (livelock_debug == intr ||
915 (bootverbose && cold)) {
916 kprintf("intr %d at %d/%d hz, in livelock\n",
917 intr, ill_count, livelock_lowater);
930 * An interrupt thread is trying to get the MP lock. To avoid cpu-bound
931 * code in the kernel on cpu X from interfering we chase the MP lock.
934 intr_get_mplock(void)
938 if (intr_migrate == 0) {
942 while (try_mplock() == 0) {
943 owner = owner_mplock();
944 if (owner >= 0 && owner != mycpu->gd_cpuid) {
945 lwkt_migratecpu(owner);
946 ++intr_migrate_count;
956 * Emergency interrupt polling thread. The thread begins execution
957 * outside a critical section with the BGL held.
959 * If emergency interrupt polling is enabled, this thread will
960 * execute all system interrupts not marked INTR_NOPOLL at the
961 * specified polling frequency.
963 * WARNING! This thread runs *ALL* interrupt service routines that
964 * are not marked INTR_NOPOLL, which basically means everything except
965 * the 8254 clock interrupt and the ATA interrupt. It has very high
966 * overhead and should only be used in situations where the machine
967 * cannot otherwise be made to work. Due to the severe performance
968 * degredation, it should not be enabled on production machines.
971 ithread_emergency(void *arg __unused)
973 struct intr_info *info;
978 for (intr = 0; intr < max_installed_hard_intr; ++intr) {
979 info = &intr_info_ary[intr];
980 for (rec = info->i_reclist; rec; rec = nrec) {
981 if ((rec->intr_flags & INTR_NOPOLL) == 0) {
982 if (rec->serializer) {
983 lwkt_serialize_handler_call(rec->serializer,
984 rec->handler, rec->argument, NULL);
986 rec->handler(rec->argument, NULL);
992 lwkt_deschedule_self(curthread);
998 * Systimer callback - schedule the emergency interrupt poll thread
999 * if emergency polling is enabled.
1003 emergency_intr_timer_callback(systimer_t info, struct intrframe *frame __unused)
1005 if (emergency_intr_enable)
1006 lwkt_schedule(info->data);
1010 ithread_cpuid(int intr)
1012 const struct intr_info *info;
1014 KKASSERT(intr >= 0 && intr < MAX_INTS);
1015 info = &intr_info_ary[intr];
1017 if (info->i_state == ISTATE_NOTHREAD)
1019 return info->i_thread.td_gd->gd_cpuid;
1023 * Sysctls used by systat and others: hw.intrnames and hw.intrcnt.
1024 * The data for this machine dependent, and the declarations are in machine
1025 * dependent code. The layout of intrnames and intrcnt however is machine
1028 * We do not know the length of intrcnt and intrnames at compile time, so
1029 * calculate things at run time.
1033 sysctl_intrnames(SYSCTL_HANDLER_ARGS)
1035 struct intr_info *info;
1042 for (intr = 0; error == 0 && intr < MAX_INTS; ++intr) {
1043 info = &intr_info_ary[intr];
1047 for (rec = info->i_reclist; rec; rec = rec->next) {
1048 ksnprintf(buf + len, sizeof(buf) - len, "%s%s",
1049 (len ? "/" : ""), rec->name);
1050 len += strlen(buf + len);
1053 ksnprintf(buf, sizeof(buf), "irq%d", intr);
1056 error = SYSCTL_OUT(req, buf, len + 1);
1062 SYSCTL_PROC(_hw, OID_AUTO, intrnames, CTLTYPE_OPAQUE | CTLFLAG_RD,
1063 NULL, 0, sysctl_intrnames, "", "Interrupt Names");
1066 sysctl_intrcnt(SYSCTL_HANDLER_ARGS)
1068 struct intr_info *info;
1072 for (intr = 0; intr < max_installed_hard_intr; ++intr) {
1073 info = &intr_info_ary[intr];
1075 error = SYSCTL_OUT(req, &info->i_count, sizeof(info->i_count));
1079 for (intr = FIRST_SOFTINT; intr < max_installed_soft_intr; ++intr) {
1080 info = &intr_info_ary[intr];
1082 error = SYSCTL_OUT(req, &info->i_count, sizeof(info->i_count));
1090 SYSCTL_PROC(_hw, OID_AUTO, intrcnt, CTLTYPE_OPAQUE | CTLFLAG_RD,
1091 NULL, 0, sysctl_intrcnt, "", "Interrupt Counts");
1094 int_moveto_destcpu(int *orig_cpuid0, int *cpuid0, int intr)
1096 int orig_cpuid = mycpuid, cpuid;
1100 ksnprintf(envpath, sizeof(envpath), "hw.irq.%d.dest", intr);
1101 kgetenv_int(envpath, &cpuid);
1105 if (cpuid != orig_cpuid)
1106 lwkt_migratecpu(cpuid);
1108 *orig_cpuid0 = orig_cpuid;
1113 int_moveto_origcpu(int orig_cpuid, int cpuid)
1115 if (cpuid != orig_cpuid)
1116 lwkt_migratecpu(orig_cpuid);