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 $
30 #include <sys/param.h>
31 #include <sys/systm.h>
32 #include <sys/malloc.h>
33 #include <sys/kernel.h>
34 #include <sys/sysctl.h>
35 #include <sys/thread.h>
37 #include <sys/random.h>
38 #include <sys/serialize.h>
39 #include <sys/interrupt.h>
41 #include <sys/machintr.h>
43 #include <machine/frame.h>
45 #include <sys/thread2.h>
46 #include <sys/mplock2.h>
50 typedef struct intrec {
52 struct intr_info *info;
58 struct lwkt_serialize *serializer;
63 struct thread *i_thread; /* don't embed struct thread */
64 struct random_softc i_random;
66 long i_count; /* interrupts dispatched */
67 int i_mplock_required;
72 unsigned long i_straycount;
77 static struct intr_info intr_info_ary[MAXCPU][MAX_INTS];
78 static struct intr_info *swi_info_ary[MAX_SOFTINTS];
80 static int max_installed_hard_intr[MAXCPU];
82 #define EMERGENCY_INTR_POLLING_FREQ_MAX 20000
85 * Assert that callers into interrupt handlers don't return with
86 * dangling tokens, spinlocks, or mp locks.
90 #define TD_INVARIANTS_DECLARE \
94 #define TD_INVARIANTS_GET(td) \
96 spincount = (td)->td_gd->gd_spinlocks; \
97 curstop = (td)->td_toks_stop; \
100 #define TD_INVARIANTS_TEST(td, name) \
102 KASSERT(spincount == (td)->td_gd->gd_spinlocks, \
103 ("spincount mismatch after interrupt handler %s", \
105 KASSERT(curstop == (td)->td_toks_stop, \
106 ("token count mismatch after interrupt handler %s", \
114 #define TD_INVARIANTS_DECLARE
115 #define TD_INVARIANTS_GET(td)
116 #define TD_INVARIANTS_TEST(td, name)
118 #endif /* ndef INVARIANTS */
120 static int sysctl_emergency_freq(SYSCTL_HANDLER_ARGS);
121 static int sysctl_emergency_enable(SYSCTL_HANDLER_ARGS);
122 static void emergency_intr_timer_callback(systimer_t, int, struct intrframe *);
123 static void ithread_handler(void *arg);
124 static void ithread_emergency(void *arg);
125 static void report_stray_interrupt(struct intr_info *info, const char *func);
126 static void int_moveto_destcpu(int *, int);
127 static void int_moveto_origcpu(int, int);
128 static void sched_ithd_intern(struct intr_info *info);
130 static struct systimer emergency_intr_timer[MAXCPU];
131 static struct thread emergency_intr_thread[MAXCPU];
133 #define ISTATE_NOTHREAD 0
134 #define ISTATE_NORMAL 1
135 #define ISTATE_LIVELOCKED 2
137 static int livelock_limit = 40000;
138 static int livelock_lowater = 20000;
139 static int livelock_debug = -1;
140 SYSCTL_INT(_kern, OID_AUTO, livelock_limit,
141 CTLFLAG_RW, &livelock_limit, 0, "Livelock interrupt rate limit");
142 SYSCTL_INT(_kern, OID_AUTO, livelock_lowater,
143 CTLFLAG_RW, &livelock_lowater, 0, "Livelock low-water mark restore");
144 SYSCTL_INT(_kern, OID_AUTO, livelock_debug,
145 CTLFLAG_RW, &livelock_debug, 0, "Livelock debug intr#");
147 static int emergency_intr_enable = 0; /* emergency interrupt polling */
148 TUNABLE_INT("kern.emergency_intr_enable", &emergency_intr_enable);
149 SYSCTL_PROC(_kern, OID_AUTO, emergency_intr_enable, CTLTYPE_INT | CTLFLAG_RW,
150 0, 0, sysctl_emergency_enable, "I", "Emergency Interrupt Poll Enable");
152 static int emergency_intr_freq = 10; /* emergency polling frequency */
153 TUNABLE_INT("kern.emergency_intr_freq", &emergency_intr_freq);
154 SYSCTL_PROC(_kern, OID_AUTO, emergency_intr_freq, CTLTYPE_INT | CTLFLAG_RW,
155 0, 0, sysctl_emergency_freq, "I", "Emergency Interrupt Poll Frequency");
158 * Sysctl support routines
161 sysctl_emergency_enable(SYSCTL_HANDLER_ARGS)
163 int error, enabled, cpuid, freq;
165 enabled = emergency_intr_enable;
166 error = sysctl_handle_int(oidp, &enabled, 0, req);
167 if (error || req->newptr == NULL)
169 emergency_intr_enable = enabled;
170 if (emergency_intr_enable)
171 freq = emergency_intr_freq;
175 for (cpuid = 0; cpuid < ncpus; ++cpuid)
176 systimer_adjust_periodic(&emergency_intr_timer[cpuid], freq);
181 sysctl_emergency_freq(SYSCTL_HANDLER_ARGS)
183 int error, phz, cpuid, freq;
185 phz = emergency_intr_freq;
186 error = sysctl_handle_int(oidp, &phz, 0, req);
187 if (error || req->newptr == NULL)
191 else if (phz > EMERGENCY_INTR_POLLING_FREQ_MAX)
192 phz = EMERGENCY_INTR_POLLING_FREQ_MAX;
194 emergency_intr_freq = phz;
195 if (emergency_intr_enable)
196 freq = emergency_intr_freq;
200 for (cpuid = 0; cpuid < ncpus; ++cpuid)
201 systimer_adjust_periodic(&emergency_intr_timer[cpuid], freq);
206 * Register an SWI or INTerrupt handler.
209 register_swi(int intr, inthand2_t *handler, void *arg, const char *name,
210 struct lwkt_serialize *serializer, int cpuid)
212 if (intr < FIRST_SOFTINT || intr >= MAX_INTS)
213 panic("register_swi: bad intr %d", intr);
216 cpuid = intr % ncpus;
217 return(register_int(intr, handler, arg, name, serializer, 0, cpuid));
221 register_swi_mp(int intr, inthand2_t *handler, void *arg, const char *name,
222 struct lwkt_serialize *serializer, int cpuid)
224 if (intr < FIRST_SOFTINT || intr >= MAX_INTS)
225 panic("register_swi: bad intr %d", intr);
228 cpuid = intr % ncpus;
229 return(register_int(intr, handler, arg, name, serializer,
230 INTR_MPSAFE, cpuid));
234 register_int(int intr, inthand2_t *handler, void *arg, const char *name,
235 struct lwkt_serialize *serializer, int intr_flags, int cpuid)
237 struct intr_info *info;
238 struct intrec **list;
242 KKASSERT(cpuid >= 0 && cpuid < ncpus);
244 if (intr < 0 || intr >= MAX_INTS)
245 panic("register_int: bad intr %d", intr);
248 info = &intr_info_ary[cpuid][intr];
251 * Construct an interrupt handler record
253 rec = kmalloc(sizeof(struct intrec), M_DEVBUF, M_INTWAIT);
254 rec->name = kmalloc(strlen(name) + 1, M_DEVBUF, M_INTWAIT);
255 strcpy(rec->name, name);
258 rec->handler = handler;
261 rec->intr_flags = intr_flags;
263 rec->serializer = serializer;
265 int_moveto_destcpu(&orig_cpuid, cpuid);
268 * Create an emergency polling thread and set up a systimer to wake
271 if (emergency_intr_thread[cpuid].td_kstack == NULL) {
272 lwkt_create(ithread_emergency, NULL, NULL,
273 &emergency_intr_thread[cpuid],
274 TDF_NOSTART | TDF_INTTHREAD, cpuid, "ithreadE %d",
276 systimer_init_periodic_nq(&emergency_intr_timer[cpuid],
277 emergency_intr_timer_callback,
278 &emergency_intr_thread[cpuid],
279 (emergency_intr_enable ? emergency_intr_freq : 1));
283 * Create an interrupt thread if necessary, leave it in an unscheduled
286 if (info->i_state == ISTATE_NOTHREAD) {
287 info->i_state = ISTATE_NORMAL;
288 info->i_thread = kmalloc(sizeof(struct thread), M_DEVBUF,
290 lwkt_create(ithread_handler, (void *)(intptr_t)intr, NULL,
291 info->i_thread, TDF_NOSTART | TDF_INTTHREAD, cpuid,
292 "ithread%d %d", intr, cpuid);
293 if (intr >= FIRST_SOFTINT)
294 lwkt_setpri(info->i_thread, TDPRI_SOFT_NORM);
296 lwkt_setpri(info->i_thread, TDPRI_INT_MED);
297 info->i_thread->td_preemptable = lwkt_preempt;
300 list = &info->i_reclist;
303 * Keep track of how many fast and slow interrupts we have.
304 * Set i_mplock_required if any handler in the chain requires
305 * the MP lock to operate.
307 if ((intr_flags & INTR_MPSAFE) == 0)
308 info->i_mplock_required = 1;
309 if (intr_flags & INTR_CLOCK)
315 * Enable random number generation keying off of this interrupt.
317 if ((intr_flags & INTR_NOENTROPY) == 0 && info->i_random.sc_enabled == 0) {
318 info->i_random.sc_enabled = 1;
319 info->i_random.sc_intr = intr;
323 * Add the record to the interrupt list.
326 while (*list != NULL)
327 list = &(*list)->next;
332 * Update max_installed_hard_intr to make the emergency intr poll
333 * a bit more efficient.
335 if (intr < FIRST_SOFTINT) {
336 if (max_installed_hard_intr[cpuid] <= intr)
337 max_installed_hard_intr[cpuid] = intr + 1;
340 if (intr >= FIRST_SOFTINT)
341 swi_info_ary[intr - FIRST_SOFTINT] = info;
344 * Setup the machine level interrupt vector
346 if (intr < FIRST_SOFTINT && info->i_slow + info->i_fast == 1)
347 machintr_intr_setup(intr, intr_flags);
349 int_moveto_origcpu(orig_cpuid, cpuid);
355 unregister_swi(void *id, int intr, int cpuid)
358 cpuid = intr % ncpus;
360 unregister_int(id, cpuid);
364 unregister_int(void *id, int cpuid)
366 struct intr_info *info;
367 struct intrec **list;
369 int intr, orig_cpuid;
371 KKASSERT(cpuid >= 0 && cpuid < ncpus);
373 intr = ((intrec_t)id)->intr;
375 if (intr < 0 || intr >= MAX_INTS)
376 panic("register_int: bad intr %d", intr);
378 info = &intr_info_ary[cpuid][intr];
380 int_moveto_destcpu(&orig_cpuid, cpuid);
383 * Remove the interrupt descriptor, adjust the descriptor count,
384 * and teardown the machine level vector if this was the last interrupt.
387 list = &info->i_reclist;
388 while ((rec = *list) != NULL) {
397 if (rec->intr_flags & INTR_CLOCK)
401 if (intr < FIRST_SOFTINT && info->i_fast + info->i_slow == 0)
402 machintr_intr_teardown(intr);
405 * Clear i_mplock_required if no handlers in the chain require the
408 for (rec0 = info->i_reclist; rec0; rec0 = rec0->next) {
409 if ((rec0->intr_flags & INTR_MPSAFE) == 0)
413 info->i_mplock_required = 0;
416 if (intr >= FIRST_SOFTINT && info->i_reclist == NULL)
417 swi_info_ary[intr - FIRST_SOFTINT] = NULL;
421 int_moveto_origcpu(orig_cpuid, cpuid);
427 kfree(rec->name, M_DEVBUF);
428 kfree(rec, M_DEVBUF);
430 kprintf("warning: unregister_int: int %d handler for %s not found\n",
431 intr, ((intrec_t)id)->name);
436 get_interrupt_counter(int intr, int cpuid)
438 struct intr_info *info;
440 KKASSERT(cpuid >= 0 && cpuid < ncpus);
442 if (intr < 0 || intr >= MAX_INTS)
443 panic("register_int: bad intr %d", intr);
444 info = &intr_info_ary[cpuid][intr];
445 return(info->i_count);
449 register_randintr(int intr)
451 struct intr_info *info;
454 if (intr < 0 || intr >= MAX_INTS)
455 panic("register_randintr: bad intr %d", intr);
457 for (cpuid = 0; cpuid < ncpus; ++cpuid) {
458 info = &intr_info_ary[cpuid][intr];
459 info->i_random.sc_intr = intr;
460 info->i_random.sc_enabled = 1;
465 unregister_randintr(int intr)
467 struct intr_info *info;
470 if (intr < 0 || intr >= MAX_INTS)
471 panic("register_swi: bad intr %d", intr);
473 for (cpuid = 0; cpuid < ncpus; ++cpuid) {
474 info = &intr_info_ary[cpuid][intr];
475 info->i_random.sc_enabled = -1;
480 next_registered_randintr(int intr)
482 struct intr_info *info;
484 if (intr < 0 || intr >= MAX_INTS)
485 panic("register_swi: bad intr %d", intr);
487 while (intr < MAX_INTS) {
490 for (cpuid = 0; cpuid < ncpus; ++cpuid) {
491 info = &intr_info_ary[cpuid][intr];
492 if (info->i_random.sc_enabled > 0)
501 * Dispatch an interrupt. If there's nothing to do we have a stray
502 * interrupt and can just return, leaving the interrupt masked.
504 * We need to schedule the interrupt and set its i_running bit. If
505 * we are not on the interrupt thread's cpu we have to send a message
506 * to the correct cpu that will issue the desired action (interlocking
507 * with the interrupt thread's critical section). We do NOT attempt to
508 * reschedule interrupts whos i_running bit is already set because
509 * this would prematurely wakeup a livelock-limited interrupt thread.
511 * i_running is only tested/set on the same cpu as the interrupt thread.
513 * We are NOT in a critical section, which will allow the scheduled
514 * interrupt to preempt us. The MP lock might *NOT* be held here.
517 sched_ithd_remote(void *arg)
519 sched_ithd_intern(arg);
523 sched_ithd_intern(struct intr_info *info)
526 if (info->i_state != ISTATE_NOTHREAD) {
527 if (info->i_reclist == NULL) {
528 report_stray_interrupt(info, "sched_ithd");
530 if (info->i_thread->td_gd == mycpu) {
531 if (info->i_running == 0) {
533 if (info->i_state != ISTATE_LIVELOCKED)
534 lwkt_schedule(info->i_thread); /* MIGHT PREEMPT */
537 lwkt_send_ipiq(info->i_thread->td_gd, sched_ithd_remote, info);
541 report_stray_interrupt(info, "sched_ithd");
546 sched_ithd_soft(int intr)
548 struct intr_info *info;
550 KKASSERT(intr >= FIRST_SOFTINT && intr < MAX_INTS);
552 info = swi_info_ary[intr - FIRST_SOFTINT];
554 sched_ithd_intern(info);
556 kprintf("unregistered softint %d got scheduled on cpu%d\n",
562 sched_ithd_hard(int intr)
564 KKASSERT(intr >= 0 && intr < MAX_HARDINTS);
565 sched_ithd_intern(&intr_info_ary[mycpuid][intr]);
568 #ifdef _KERNEL_VIRTUAL
571 sched_ithd_hard_virtual(int intr)
573 KKASSERT(intr >= 0 && intr < MAX_HARDINTS);
574 sched_ithd_intern(&intr_info_ary[0][intr]);
578 register_int_virtual(int intr, inthand2_t *handler, void *arg, const char *name,
579 struct lwkt_serialize *serializer, int intr_flags)
581 return register_int(intr, handler, arg, name, serializer, intr_flags, 0);
585 unregister_int_virtual(void *id)
587 unregister_int(id, 0);
590 #endif /* _KERN_VIRTUAL */
593 report_stray_interrupt(struct intr_info *info, const char *func)
595 ++info->i_straycount;
596 if (info->i_straycount < 10) {
597 if (info->i_errorticks == ticks)
599 info->i_errorticks = ticks;
600 kprintf("%s: stray interrupt %d on cpu%d\n",
601 func, info->i_intr, mycpuid);
602 } else if (info->i_straycount == 10) {
603 kprintf("%s: %ld stray interrupts %d on cpu%d - "
604 "there will be no further reports\n", func,
605 info->i_straycount, info->i_intr, mycpuid);
610 * This is run from a periodic SYSTIMER (and thus must be MP safe, the BGL
611 * might not be held).
614 ithread_livelock_wakeup(systimer_t st, int in_ipi __unused,
615 struct intrframe *frame __unused)
617 struct intr_info *info;
619 info = &intr_info_ary[mycpuid][(int)(intptr_t)st->data];
620 if (info->i_state != ISTATE_NOTHREAD)
621 lwkt_schedule(info->i_thread);
625 * Schedule ithread within fast intr handler
627 * XXX Protect sched_ithd_hard() call with gd_intr_nesting_level?
628 * Interrupts aren't enabled, but still...
631 ithread_fast_sched(int intr, thread_t td)
636 * We are already in critical section, exit it now to
640 sched_ithd_hard(intr);
641 crit_enter_quick(td);
647 * This function is called directly from the ICU or APIC vector code assembly
648 * to process an interrupt. The critical section and interrupt deferral
649 * checks have already been done but the function is entered WITHOUT
650 * a critical section held. The BGL may or may not be held.
652 * Must return non-zero if we do not want the vector code to re-enable
653 * the interrupt (which we don't if we have to schedule the interrupt)
655 int ithread_fast_handler(struct intrframe *frame);
658 ithread_fast_handler(struct intrframe *frame)
661 struct intr_info *info;
662 struct intrec **list;
665 TD_INVARIANTS_DECLARE;
670 intr = frame->if_vec;
674 /* We must be in critical section. */
675 KKASSERT(td->td_critcount);
677 info = &intr_info_ary[mycpuid][intr];
680 * If we are not processing any FAST interrupts, just schedule the thing.
682 if (info->i_fast == 0) {
684 ithread_fast_sched(intr, td);
689 * This should not normally occur since interrupts ought to be
690 * masked if the ithread has been scheduled or is running.
696 * Bump the interrupt nesting level to process any FAST interrupts.
697 * Obtain the MP lock as necessary. If the MP lock cannot be obtained,
698 * schedule the interrupt thread to deal with the issue instead.
700 * To reduce overhead, just leave the MP lock held once it has been
703 ++gd->gd_intr_nesting_level;
705 must_schedule = info->i_slow;
708 TD_INVARIANTS_GET(td);
709 list = &info->i_reclist;
711 for (rec = *list; rec; rec = nrec) {
712 /* rec may be invalid after call */
715 if (rec->intr_flags & INTR_CLOCK) {
716 if ((rec->intr_flags & INTR_MPSAFE) == 0 && got_mplock == 0) {
717 if (try_mplock() == 0) {
718 /* Couldn't get the MP lock; just schedule it. */
724 if (rec->serializer) {
725 must_schedule += lwkt_serialize_handler_try(
726 rec->serializer, rec->handler,
727 rec->argument, frame);
729 rec->handler(rec->argument, frame);
731 TD_INVARIANTS_TEST(td, rec->name);
738 --gd->gd_intr_nesting_level;
743 * If we had a problem, or mixed fast and slow interrupt handlers are
744 * registered, schedule the ithread to catch the missed records (it
745 * will just re-run all of them). A return value of 0 indicates that
746 * all handlers have been run and the interrupt can be re-enabled, and
747 * a non-zero return indicates that the interrupt thread controls
750 if (must_schedule > 0)
751 ithread_fast_sched(intr, td);
752 else if (must_schedule == 0)
754 return(must_schedule);
758 * Interrupt threads run this as their main loop.
760 * The handler begins execution outside a critical section and no MP lock.
762 * The i_running state starts at 0. When an interrupt occurs, the hardware
763 * interrupt is disabled and sched_ithd_hard() The HW interrupt remains
764 * disabled until all routines have run. We then call ithread_done() to
765 * reenable the HW interrupt and deschedule us until the next interrupt.
767 * We are responsible for atomically checking i_running and ithread_done()
768 * is responsible for atomically checking for platform-specific delayed
769 * interrupts. i_running for our irq is only set in the context of our cpu,
770 * so a critical section is a sufficient interlock.
772 #define LIVELOCK_TIMEFRAME(freq) ((freq) >> 2) /* 1/4 second */
775 ithread_handler(void *arg)
777 struct intr_info *info;
780 int intr, cpuid = mycpuid;
782 struct intrec **list;
785 struct systimer ill_timer; /* enforced freq. timer */
786 u_int ill_count; /* interrupt livelock counter */
787 TD_INVARIANTS_DECLARE;
790 intr = (int)(intptr_t)arg;
791 info = &intr_info_ary[cpuid][intr];
792 list = &info->i_reclist;
795 * The loop must be entered with one critical section held. The thread
796 * does not hold the mplock on startup.
799 lseconds = gd->gd_time_seconds;
805 * The chain is only considered MPSAFE if all its interrupt handlers
806 * are MPSAFE. However, if intr_mpsafe has been turned off we
807 * always operate with the BGL.
809 if (info->i_mplock_required != mpheld) {
810 if (info->i_mplock_required) {
811 KKASSERT(mpheld == 0);
815 KKASSERT(mpheld != 0);
821 TD_INVARIANTS_GET(gd->gd_curthread);
824 * If an interrupt is pending, clear i_running and execute the
825 * handlers. Note that certain types of interrupts can re-trigger
826 * and set i_running again.
828 * Each handler is run in a critical section. Note that we run both
829 * FAST and SLOW designated service routines.
831 if (info->i_running) {
836 report_stray_interrupt(info, "ithread_handler");
838 for (rec = *list; rec; rec = nrec) {
839 /* rec may be invalid after call */
841 if (rec->serializer) {
842 lwkt_serialize_handler_call(rec->serializer, rec->handler,
843 rec->argument, NULL);
845 rec->handler(rec->argument, NULL);
847 TD_INVARIANTS_TEST(gd->gd_curthread, rec->name);
852 * This is our interrupt hook to add rate randomness to the random
855 if (info->i_random.sc_enabled > 0)
856 add_interrupt_randomness(intr);
859 * Unmask the interrupt to allow it to trigger again. This only
860 * applies to certain types of interrupts (typ level interrupts).
861 * This can result in the interrupt retriggering, but the retrigger
862 * will not be processed until we cycle our critical section.
864 * Only unmask interrupts while handlers are installed. It is
865 * possible to hit a situation where no handlers are installed
866 * due to a device driver livelocking and then tearing down its
867 * interrupt on close (the parallel bus being a good example).
869 if (intr < FIRST_SOFTINT && *list)
870 machintr_intr_enable(intr);
873 * Do a quick exit/enter to catch any higher-priority interrupt
874 * sources, such as the statclock, so thread time accounting
875 * will still work. This may also cause an interrupt to re-trigger.
881 * LIVELOCK STATE MACHINE
883 switch(info->i_state) {
886 * Reset the count each second.
888 if (lseconds != gd->gd_time_seconds) {
889 lseconds = gd->gd_time_seconds;
894 * If we did not exceed the frequency limit, we are done.
895 * If the interrupt has not retriggered we deschedule ourselves.
897 if (ill_count <= livelock_limit) {
898 if (info->i_running == 0) {
899 lwkt_deschedule_self(gd->gd_curthread);
906 * Otherwise we are livelocked. Set up a periodic systimer
907 * to wake the thread up at the limit frequency.
909 kprintf("intr %d on cpu%d at %d/%d hz, livelocked limit engaged!\n",
910 intr, cpuid, ill_count, livelock_limit);
911 info->i_state = ISTATE_LIVELOCKED;
912 if ((use_limit = livelock_limit) < 100)
914 else if (use_limit > 500000)
916 systimer_init_periodic_nq(&ill_timer, ithread_livelock_wakeup,
917 (void *)(intptr_t)intr, use_limit);
919 case ISTATE_LIVELOCKED:
921 * Wait for our periodic timer to go off. Since the interrupt
922 * has re-armed it can still set i_running, but it will not
923 * reschedule us while we are in a livelocked state.
925 lwkt_deschedule_self(gd->gd_curthread);
929 * Check once a second to see if the livelock condition no
932 if (lseconds != gd->gd_time_seconds) {
933 lseconds = gd->gd_time_seconds;
934 if (ill_count < livelock_lowater) {
935 info->i_state = ISTATE_NORMAL;
936 systimer_del(&ill_timer);
937 kprintf("intr %d on cpu%d at %d/%d hz, livelock removed\n",
938 intr, cpuid, ill_count, livelock_lowater);
939 } else if (livelock_debug == intr ||
940 (bootverbose && cold)) {
941 kprintf("intr %d on cpu%d at %d/%d hz, in livelock\n",
942 intr, cpuid, ill_count, livelock_lowater);
953 * Emergency interrupt polling thread. The thread begins execution
954 * outside a critical section with the BGL held.
956 * If emergency interrupt polling is enabled, this thread will
957 * execute all system interrupts not marked INTR_NOPOLL at the
958 * specified polling frequency.
960 * WARNING! This thread runs *ALL* interrupt service routines that
961 * are not marked INTR_NOPOLL, which basically means everything except
962 * the 8254 clock interrupt and the ATA interrupt. It has very high
963 * overhead and should only be used in situations where the machine
964 * cannot otherwise be made to work. Due to the severe performance
965 * degredation, it should not be enabled on production machines.
968 ithread_emergency(void *arg __unused)
970 globaldata_t gd = mycpu;
971 struct intr_info *info;
973 int intr, cpuid = mycpuid;
974 TD_INVARIANTS_DECLARE;
978 TD_INVARIANTS_GET(gd->gd_curthread);
981 for (intr = 0; intr < max_installed_hard_intr[cpuid]; ++intr) {
982 info = &intr_info_ary[cpuid][intr];
983 for (rec = info->i_reclist; rec; rec = nrec) {
984 /* rec may be invalid after call */
986 if ((rec->intr_flags & INTR_NOPOLL) == 0) {
987 if (rec->serializer) {
988 lwkt_serialize_handler_try(rec->serializer,
989 rec->handler, rec->argument, NULL);
991 rec->handler(rec->argument, NULL);
993 TD_INVARIANTS_TEST(gd->gd_curthread, rec->name);
997 lwkt_deschedule_self(gd->gd_curthread);
1004 * Systimer callback - schedule the emergency interrupt poll thread
1005 * if emergency polling is enabled.
1009 emergency_intr_timer_callback(systimer_t info, int in_ipi __unused,
1010 struct intrframe *frame __unused)
1012 if (emergency_intr_enable)
1013 lwkt_schedule(info->data);
1017 * Sysctls used by systat and others: hw.intrnames and hw.intrcnt.
1018 * The data for this machine dependent, and the declarations are in machine
1019 * dependent code. The layout of intrnames and intrcnt however is machine
1022 * We do not know the length of intrcnt and intrnames at compile time, so
1023 * calculate things at run time.
1027 sysctl_intrnames(SYSCTL_HANDLER_ARGS)
1029 struct intr_info *info;
1036 for (cpuid = 0; cpuid < ncpus; ++cpuid) {
1037 for (intr = 0; error == 0 && intr < MAX_INTS; ++intr) {
1038 info = &intr_info_ary[cpuid][intr];
1042 for (rec = info->i_reclist; rec; rec = rec->next) {
1043 ksnprintf(buf + len, sizeof(buf) - len, "%s%s",
1044 (len ? "/" : ""), rec->name);
1045 len += strlen(buf + len);
1048 ksnprintf(buf, sizeof(buf), "irq%d", intr);
1051 error = SYSCTL_OUT(req, buf, len + 1);
1057 SYSCTL_PROC(_hw, OID_AUTO, intrnames, CTLTYPE_OPAQUE | CTLFLAG_RD,
1058 NULL, 0, sysctl_intrnames, "", "Interrupt Names");
1061 sysctl_intrcnt_all(SYSCTL_HANDLER_ARGS)
1063 struct intr_info *info;
1067 for (cpuid = 0; cpuid < ncpus; ++cpuid) {
1068 for (intr = 0; intr < MAX_INTS; ++intr) {
1069 info = &intr_info_ary[cpuid][intr];
1071 error = SYSCTL_OUT(req, &info->i_count, sizeof(info->i_count));
1080 SYSCTL_PROC(_hw, OID_AUTO, intrcnt_all, CTLTYPE_OPAQUE | CTLFLAG_RD,
1081 NULL, 0, sysctl_intrcnt_all, "", "Interrupt Counts");
1083 SYSCTL_PROC(_hw, OID_AUTO, intrcnt, CTLTYPE_OPAQUE | CTLFLAG_RD,
1084 NULL, 0, sysctl_intrcnt_all, "", "Interrupt Counts");
1087 int_moveto_destcpu(int *orig_cpuid0, int cpuid)
1089 int orig_cpuid = mycpuid;
1091 if (cpuid != orig_cpuid)
1092 lwkt_migratecpu(cpuid);
1094 *orig_cpuid0 = orig_cpuid;
1098 int_moveto_origcpu(int orig_cpuid, int cpuid)
1100 if (cpuid != orig_cpuid)
1101 lwkt_migratecpu(orig_cpuid);
1105 intr_init(void *dummy __unused)
1109 kprintf("Initialize MI interrupts\n");
1111 for (cpuid = 0; cpuid < ncpus; ++cpuid) {
1114 for (intr = 0; intr < MAX_INTS; ++intr) {
1115 struct intr_info *info = &intr_info_ary[cpuid][intr];
1117 info->i_cpuid = cpuid;
1118 info->i_intr = intr;
1122 SYSINIT(intr_init, SI_BOOT2_FINISH_PIC, SI_ORDER_ANY, intr_init, NULL);