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/random.h>
39 #include <sys/serialize.h>
40 #include <sys/interrupt.h>
42 #include <sys/machintr.h>
44 #include <machine/frame.h>
46 #include <sys/interrupt.h>
48 #include <sys/thread2.h>
49 #include <sys/mplock2.h>
53 typedef struct intrec {
55 struct intr_info *info;
61 struct lwkt_serialize *serializer;
66 struct thread i_thread;
67 struct random_softc i_random;
69 long i_count; /* interrupts dispatched */
70 int i_mplock_required;
75 unsigned long i_straycount;
76 } intr_info_ary[MAX_INTS];
78 int max_installed_hard_intr;
79 int max_installed_soft_intr;
81 #define EMERGENCY_INTR_POLLING_FREQ_MAX 20000
84 * Assert that callers into interrupt handlers don't return with
85 * dangling tokens, spinlocks, or mp locks.
91 #define SMP_INVARIANTS_DECLARE \
94 #define SMP_INVARIANTS_GET(td) \
95 mpcount = (td)->td_mpcount
97 #define SMP_INVARIANTS_TEST(td, name) \
98 KASSERT(mpcount == (td)->td_mpcount, \
99 ("mpcount mismatch after interrupt handler %s", \
102 #define SMP_INVARIANTS_ADJMP(count) mpcount += (count)
106 #define SMP_INVARIANTS_DECLARE
107 #define SMP_INVARIANTS_GET(td)
108 #define SMP_INVARIANTS_TEST(td, name)
112 #define TD_INVARIANTS_DECLARE \
113 SMP_INVARIANTS_DECLARE \
115 lwkt_tokref_t curstop
117 #define TD_INVARIANTS_GET(td) \
119 SMP_INVARIANTS_GET(td); \
120 spincount = (td)->td_gd->gd_spinlocks_wr; \
121 curstop = (td)->td_toks_stop; \
124 #define TD_INVARIANTS_TEST(td, name) \
126 KASSERT(spincount == (td)->td_gd->gd_spinlocks_wr, \
127 ("spincount mismatch after interrupt handler %s", \
129 KASSERT(curstop == (td)->td_toks_stop, \
130 ("token count mismatch after interrupt handler %s", \
132 SMP_INVARIANTS_TEST(td, name); \
137 #define TD_INVARIANTS_DECLARE
138 #define TD_INVARIANTS_GET(td)
139 #define TD_INVARIANTS_TEST(td)
140 #define TD_INVARIANTS_ADJMP(count)
144 static int sysctl_emergency_freq(SYSCTL_HANDLER_ARGS);
145 static int sysctl_emergency_enable(SYSCTL_HANDLER_ARGS);
146 static void emergency_intr_timer_callback(systimer_t, struct intrframe *);
147 static void ithread_handler(void *arg);
148 static void ithread_emergency(void *arg);
149 static void report_stray_interrupt(int intr, struct intr_info *info);
150 static void int_moveto_destcpu(int *, int *, int);
151 static void int_moveto_origcpu(int, int);
153 int intr_info_size = sizeof(intr_info_ary) / sizeof(intr_info_ary[0]);
155 static struct systimer emergency_intr_timer;
156 static struct thread emergency_intr_thread;
158 #define ISTATE_NOTHREAD 0
159 #define ISTATE_NORMAL 1
160 #define ISTATE_LIVELOCKED 2
162 static int livelock_limit = 40000;
163 static int livelock_lowater = 20000;
164 static int livelock_debug = -1;
165 SYSCTL_INT(_kern, OID_AUTO, livelock_limit,
166 CTLFLAG_RW, &livelock_limit, 0, "Livelock interrupt rate limit");
167 SYSCTL_INT(_kern, OID_AUTO, livelock_lowater,
168 CTLFLAG_RW, &livelock_lowater, 0, "Livelock low-water mark restore");
169 SYSCTL_INT(_kern, OID_AUTO, livelock_debug,
170 CTLFLAG_RW, &livelock_debug, 0, "Livelock debug intr#");
172 static int emergency_intr_enable = 0; /* emergency interrupt polling */
173 TUNABLE_INT("kern.emergency_intr_enable", &emergency_intr_enable);
174 SYSCTL_PROC(_kern, OID_AUTO, emergency_intr_enable, CTLTYPE_INT | CTLFLAG_RW,
175 0, 0, sysctl_emergency_enable, "I", "Emergency Interrupt Poll Enable");
177 static int emergency_intr_freq = 10; /* emergency polling frequency */
178 TUNABLE_INT("kern.emergency_intr_freq", &emergency_intr_freq);
179 SYSCTL_PROC(_kern, OID_AUTO, emergency_intr_freq, CTLTYPE_INT | CTLFLAG_RW,
180 0, 0, sysctl_emergency_freq, "I", "Emergency Interrupt Poll Frequency");
183 * Sysctl support routines
186 sysctl_emergency_enable(SYSCTL_HANDLER_ARGS)
190 enabled = emergency_intr_enable;
191 error = sysctl_handle_int(oidp, &enabled, 0, req);
192 if (error || req->newptr == NULL)
194 emergency_intr_enable = enabled;
195 if (emergency_intr_enable) {
196 systimer_adjust_periodic(&emergency_intr_timer,
197 emergency_intr_freq);
199 systimer_adjust_periodic(&emergency_intr_timer, 1);
205 sysctl_emergency_freq(SYSCTL_HANDLER_ARGS)
209 phz = emergency_intr_freq;
210 error = sysctl_handle_int(oidp, &phz, 0, req);
211 if (error || req->newptr == NULL)
215 else if (phz > EMERGENCY_INTR_POLLING_FREQ_MAX)
216 phz = EMERGENCY_INTR_POLLING_FREQ_MAX;
218 emergency_intr_freq = phz;
219 if (emergency_intr_enable) {
220 systimer_adjust_periodic(&emergency_intr_timer,
221 emergency_intr_freq);
223 systimer_adjust_periodic(&emergency_intr_timer, 1);
229 * Register an SWI or INTerrupt handler.
232 register_swi(int intr, inthand2_t *handler, void *arg, const char *name,
233 struct lwkt_serialize *serializer)
235 if (intr < FIRST_SOFTINT || intr >= MAX_INTS)
236 panic("register_swi: bad intr %d", intr);
237 return(register_int(intr, handler, arg, name, serializer, 0));
241 register_swi_mp(int intr, inthand2_t *handler, void *arg, const char *name,
242 struct lwkt_serialize *serializer)
244 if (intr < FIRST_SOFTINT || intr >= MAX_INTS)
245 panic("register_swi: bad intr %d", intr);
246 return(register_int(intr, handler, arg, name, serializer, INTR_MPSAFE));
250 register_int(int intr, inthand2_t *handler, void *arg, const char *name,
251 struct lwkt_serialize *serializer, int intr_flags)
253 struct intr_info *info;
254 struct intrec **list;
256 int orig_cpuid, cpuid;
258 if (intr < 0 || intr >= MAX_INTS)
259 panic("register_int: bad intr %d", intr);
262 info = &intr_info_ary[intr];
265 * Construct an interrupt handler record
267 rec = kmalloc(sizeof(struct intrec), M_DEVBUF, M_INTWAIT);
268 rec->name = kmalloc(strlen(name) + 1, M_DEVBUF, M_INTWAIT);
269 strcpy(rec->name, name);
272 rec->handler = handler;
275 rec->intr_flags = intr_flags;
277 rec->serializer = serializer;
280 * Create an emergency polling thread and set up a systimer to wake
283 if (emergency_intr_thread.td_kstack == NULL) {
284 lwkt_create(ithread_emergency, NULL, NULL, &emergency_intr_thread,
285 TDF_STOPREQ | TDF_INTTHREAD, -1, "ithread emerg");
286 systimer_init_periodic_nq(&emergency_intr_timer,
287 emergency_intr_timer_callback, &emergency_intr_thread,
288 (emergency_intr_enable ? emergency_intr_freq : 1));
291 int_moveto_destcpu(&orig_cpuid, &cpuid, intr);
294 * Create an interrupt thread if necessary, leave it in an unscheduled
297 if (info->i_state == ISTATE_NOTHREAD) {
298 info->i_state = ISTATE_NORMAL;
299 lwkt_create(ithread_handler, (void *)(intptr_t)intr, NULL,
300 &info->i_thread, TDF_STOPREQ | TDF_INTTHREAD, -1,
302 if (intr >= FIRST_SOFTINT)
303 lwkt_setpri(&info->i_thread, TDPRI_SOFT_NORM);
305 lwkt_setpri(&info->i_thread, TDPRI_INT_MED);
306 info->i_thread.td_preemptable = lwkt_preempt;
309 list = &info->i_reclist;
312 * Keep track of how many fast and slow interrupts we have.
313 * Set i_mplock_required if any handler in the chain requires
314 * the MP lock to operate.
316 if ((intr_flags & INTR_MPSAFE) == 0)
317 info->i_mplock_required = 1;
318 if (intr_flags & INTR_CLOCK)
324 * Enable random number generation keying off of this interrupt.
326 if ((intr_flags & INTR_NOENTROPY) == 0 && info->i_random.sc_enabled == 0) {
327 info->i_random.sc_enabled = 1;
328 info->i_random.sc_intr = intr;
332 * Add the record to the interrupt list.
335 while (*list != NULL)
336 list = &(*list)->next;
341 * Update max_installed_hard_intr to make the emergency intr poll
342 * a bit more efficient.
344 if (intr < FIRST_SOFTINT) {
345 if (max_installed_hard_intr <= intr)
346 max_installed_hard_intr = intr + 1;
348 if (max_installed_soft_intr <= intr)
349 max_installed_soft_intr = intr + 1;
353 * Setup the machine level interrupt vector
355 * XXX temporary workaround for some ACPI brokedness. ACPI installs
356 * its interrupt too early, before the IOAPICs have been configured,
357 * which means the IOAPIC is not enabled by the registration of the
358 * ACPI interrupt. Anything else sharing that IRQ will wind up not
359 * being enabled. Temporarily work around the problem by always
360 * installing and enabling on every new interrupt handler, even
361 * if one has already been setup on that irq.
363 if (intr < FIRST_SOFTINT /* && info->i_slow + info->i_fast == 1*/) {
364 if (machintr_vector_setup(intr, intr_flags))
365 kprintf("machintr_vector_setup: failed on irq %d\n", intr);
368 int_moveto_origcpu(orig_cpuid, cpuid);
374 unregister_swi(void *id)
380 unregister_int(void *id)
382 struct intr_info *info;
383 struct intrec **list;
385 int intr, orig_cpuid, cpuid;
387 intr = ((intrec_t)id)->intr;
389 if (intr < 0 || intr >= MAX_INTS)
390 panic("register_int: bad intr %d", intr);
392 info = &intr_info_ary[intr];
394 int_moveto_destcpu(&orig_cpuid, &cpuid, intr);
397 * Remove the interrupt descriptor, adjust the descriptor count,
398 * and teardown the machine level vector if this was the last interrupt.
401 list = &info->i_reclist;
402 while ((rec = *list) != NULL) {
411 if (rec->intr_flags & INTR_CLOCK)
415 if (intr < FIRST_SOFTINT && info->i_fast + info->i_slow == 0)
416 machintr_vector_teardown(intr);
419 * Clear i_mplock_required if no handlers in the chain require the
422 for (rec0 = info->i_reclist; rec0; rec0 = rec0->next) {
423 if ((rec0->intr_flags & INTR_MPSAFE) == 0)
427 info->i_mplock_required = 0;
432 int_moveto_origcpu(orig_cpuid, cpuid);
438 kfree(rec->name, M_DEVBUF);
439 kfree(rec, M_DEVBUF);
441 kprintf("warning: unregister_int: int %d handler for %s not found\n",
442 intr, ((intrec_t)id)->name);
447 get_registered_name(int intr)
451 if (intr < 0 || intr >= MAX_INTS)
452 panic("register_int: bad intr %d", intr);
454 if ((rec = intr_info_ary[intr].i_reclist) == NULL)
463 count_registered_ints(int intr)
465 struct intr_info *info;
467 if (intr < 0 || intr >= MAX_INTS)
468 panic("register_int: bad intr %d", intr);
469 info = &intr_info_ary[intr];
470 return(info->i_fast + info->i_slow);
474 get_interrupt_counter(int intr)
476 struct intr_info *info;
478 if (intr < 0 || intr >= MAX_INTS)
479 panic("register_int: bad intr %d", intr);
480 info = &intr_info_ary[intr];
481 return(info->i_count);
486 swi_setpriority(int intr, int pri)
488 struct intr_info *info;
490 if (intr < FIRST_SOFTINT || intr >= MAX_INTS)
491 panic("register_swi: bad intr %d", intr);
492 info = &intr_info_ary[intr];
493 if (info->i_state != ISTATE_NOTHREAD)
494 lwkt_setpri(&info->i_thread, pri);
498 register_randintr(int intr)
500 struct intr_info *info;
502 if (intr < 0 || intr >= MAX_INTS)
503 panic("register_randintr: bad intr %d", intr);
504 info = &intr_info_ary[intr];
505 info->i_random.sc_intr = intr;
506 info->i_random.sc_enabled = 1;
510 unregister_randintr(int intr)
512 struct intr_info *info;
514 if (intr < 0 || intr >= MAX_INTS)
515 panic("register_swi: bad intr %d", intr);
516 info = &intr_info_ary[intr];
517 info->i_random.sc_enabled = -1;
521 next_registered_randintr(int intr)
523 struct intr_info *info;
525 if (intr < 0 || intr >= MAX_INTS)
526 panic("register_swi: bad intr %d", intr);
527 while (intr < MAX_INTS) {
528 info = &intr_info_ary[intr];
529 if (info->i_random.sc_enabled > 0)
537 * Dispatch an interrupt. If there's nothing to do we have a stray
538 * interrupt and can just return, leaving the interrupt masked.
540 * We need to schedule the interrupt and set its i_running bit. If
541 * we are not on the interrupt thread's cpu we have to send a message
542 * to the correct cpu that will issue the desired action (interlocking
543 * with the interrupt thread's critical section). We do NOT attempt to
544 * reschedule interrupts whos i_running bit is already set because
545 * this would prematurely wakeup a livelock-limited interrupt thread.
547 * i_running is only tested/set on the same cpu as the interrupt thread.
549 * We are NOT in a critical section, which will allow the scheduled
550 * interrupt to preempt us. The MP lock might *NOT* be held here.
555 sched_ithd_remote(void *arg)
557 sched_ithd((int)(intptr_t)arg);
565 struct intr_info *info;
567 info = &intr_info_ary[intr];
570 if (info->i_state != ISTATE_NOTHREAD) {
571 if (info->i_reclist == NULL) {
572 report_stray_interrupt(intr, info);
575 if (info->i_thread.td_gd == mycpu) {
576 if (info->i_running == 0) {
578 if (info->i_state != ISTATE_LIVELOCKED)
579 lwkt_schedule(&info->i_thread); /* MIGHT PREEMPT */
582 lwkt_send_ipiq(info->i_thread.td_gd,
583 sched_ithd_remote, (void *)(intptr_t)intr);
586 if (info->i_running == 0) {
588 if (info->i_state != ISTATE_LIVELOCKED)
589 lwkt_schedule(&info->i_thread); /* MIGHT PREEMPT */
594 report_stray_interrupt(intr, info);
599 report_stray_interrupt(int intr, struct intr_info *info)
601 ++info->i_straycount;
602 if (info->i_straycount < 10) {
603 if (info->i_errorticks == ticks)
605 info->i_errorticks = ticks;
606 kprintf("sched_ithd: stray interrupt %d on cpu %d\n",
608 } else if (info->i_straycount == 10) {
609 kprintf("sched_ithd: %ld stray interrupts %d on cpu %d - "
610 "there will be no further reports\n",
611 info->i_straycount, intr, mycpuid);
616 * This is run from a periodic SYSTIMER (and thus must be MP safe, the BGL
617 * might not be held).
620 ithread_livelock_wakeup(systimer_t st)
622 struct intr_info *info;
624 info = &intr_info_ary[(int)(intptr_t)st->data];
625 if (info->i_state != ISTATE_NOTHREAD)
626 lwkt_schedule(&info->i_thread);
630 * Schedule ithread within fast intr handler
632 * XXX Protect sched_ithd() call with gd_intr_nesting_level?
633 * Interrupts aren't enabled, but still...
636 ithread_fast_sched(int intr, thread_t td)
641 * We are already in critical section, exit it now to
646 crit_enter_quick(td);
652 * This function is called directly from the ICU or APIC vector code assembly
653 * to process an interrupt. The critical section and interrupt deferral
654 * checks have already been done but the function is entered WITHOUT
655 * a critical section held. The BGL may or may not be held.
657 * Must return non-zero if we do not want the vector code to re-enable
658 * the interrupt (which we don't if we have to schedule the interrupt)
660 int ithread_fast_handler(struct intrframe *frame);
663 ithread_fast_handler(struct intrframe *frame)
666 struct intr_info *info;
667 struct intrec **list;
672 TD_INVARIANTS_DECLARE;
673 intrec_t rec, next_rec;
677 intr = frame->if_vec;
681 /* We must be in critical section. */
682 KKASSERT(td->td_critcount);
684 info = &intr_info_ary[intr];
687 * If we are not processing any FAST interrupts, just schedule the thing.
689 if (info->i_fast == 0) {
691 ithread_fast_sched(intr, td);
696 * This should not normally occur since interrupts ought to be
697 * masked if the ithread has been scheduled or is running.
703 * Bump the interrupt nesting level to process any FAST interrupts.
704 * Obtain the MP lock as necessary. If the MP lock cannot be obtained,
705 * schedule the interrupt thread to deal with the issue instead.
707 * To reduce overhead, just leave the MP lock held once it has been
710 ++gd->gd_intr_nesting_level;
712 must_schedule = info->i_slow;
717 TD_INVARIANTS_GET(td);
718 list = &info->i_reclist;
720 for (rec = *list; rec; rec = next_rec) {
721 next_rec = rec->next; /* rec may be invalid after call */
723 if (rec->intr_flags & INTR_CLOCK) {
725 if ((rec->intr_flags & INTR_MPSAFE) == 0 && got_mplock == 0) {
726 if (try_mplock() == 0) {
727 /* Couldn't get the MP lock; just schedule it. */
732 SMP_INVARIANTS_ADJMP(1);
735 if (rec->serializer) {
736 must_schedule += lwkt_serialize_handler_try(
737 rec->serializer, rec->handler,
738 rec->argument, frame);
740 rec->handler(rec->argument, frame);
742 TD_INVARIANTS_TEST(td, rec->name);
749 --gd->gd_intr_nesting_level;
756 * If we had a problem, or mixed fast and slow interrupt handlers are
757 * registered, schedule the ithread to catch the missed records (it
758 * will just re-run all of them). A return value of 0 indicates that
759 * all handlers have been run and the interrupt can be re-enabled, and
760 * a non-zero return indicates that the interrupt thread controls
763 if (must_schedule > 0)
764 ithread_fast_sched(intr, td);
765 else if (must_schedule == 0)
767 return(must_schedule);
771 * Interrupt threads run this as their main loop.
773 * The handler begins execution outside a critical section and no MP lock.
775 * The i_running state starts at 0. When an interrupt occurs, the hardware
776 * interrupt is disabled and sched_ithd() The HW interrupt remains disabled
777 * until all routines have run. We then call ithread_done() to reenable
778 * the HW interrupt and deschedule us until the next interrupt.
780 * We are responsible for atomically checking i_running and ithread_done()
781 * is responsible for atomically checking for platform-specific delayed
782 * interrupts. i_running for our irq is only set in the context of our cpu,
783 * so a critical section is a sufficient interlock.
785 #define LIVELOCK_TIMEFRAME(freq) ((freq) >> 2) /* 1/4 second */
788 ithread_handler(void *arg)
790 struct intr_info *info;
795 struct intrec **list;
798 struct systimer ill_timer; /* enforced freq. timer */
799 u_int ill_count; /* interrupt livelock counter */
800 TD_INVARIANTS_DECLARE;
803 intr = (int)(intptr_t)arg;
804 info = &intr_info_ary[intr];
805 list = &info->i_reclist;
808 * The loop must be entered with one critical section held. The thread
809 * does not hold the mplock on startup.
812 lseconds = gd->gd_time_seconds;
818 * The chain is only considered MPSAFE if all its interrupt handlers
819 * are MPSAFE. However, if intr_mpsafe has been turned off we
820 * always operate with the BGL.
823 if (info->i_mplock_required != mpheld) {
824 if (info->i_mplock_required) {
825 KKASSERT(mpheld == 0);
829 KKASSERT(mpheld != 0);
836 TD_INVARIANTS_GET(gd->gd_curthread);
839 * If an interrupt is pending, clear i_running and execute the
840 * handlers. Note that certain types of interrupts can re-trigger
841 * and set i_running again.
843 * Each handler is run in a critical section. Note that we run both
844 * FAST and SLOW designated service routines.
846 if (info->i_running) {
851 report_stray_interrupt(intr, info);
853 for (rec = *list; rec; rec = nrec) {
855 if (rec->serializer) {
856 lwkt_serialize_handler_call(rec->serializer, rec->handler,
857 rec->argument, NULL);
859 rec->handler(rec->argument, NULL);
861 TD_INVARIANTS_TEST(gd->gd_curthread, rec->name);
866 * This is our interrupt hook to add rate randomness to the random
869 if (info->i_random.sc_enabled > 0)
870 add_interrupt_randomness(intr);
873 * Unmask the interrupt to allow it to trigger again. This only
874 * applies to certain types of interrupts (typ level interrupts).
875 * This can result in the interrupt retriggering, but the retrigger
876 * will not be processed until we cycle our critical section.
878 * Only unmask interrupts while handlers are installed. It is
879 * possible to hit a situation where no handlers are installed
880 * due to a device driver livelocking and then tearing down its
881 * interrupt on close (the parallel bus being a good example).
884 machintr_intren(intr);
887 * Do a quick exit/enter to catch any higher-priority interrupt
888 * sources, such as the statclock, so thread time accounting
889 * will still work. This may also cause an interrupt to re-trigger.
895 * LIVELOCK STATE MACHINE
897 switch(info->i_state) {
900 * Reset the count each second.
902 if (lseconds != gd->gd_time_seconds) {
903 lseconds = gd->gd_time_seconds;
908 * If we did not exceed the frequency limit, we are done.
909 * If the interrupt has not retriggered we deschedule ourselves.
911 if (ill_count <= livelock_limit) {
912 if (info->i_running == 0) {
913 lwkt_deschedule_self(gd->gd_curthread);
920 * Otherwise we are livelocked. Set up a periodic systimer
921 * to wake the thread up at the limit frequency.
923 kprintf("intr %d at %d/%d hz, livelocked limit engaged!\n",
924 intr, ill_count, livelock_limit);
925 info->i_state = ISTATE_LIVELOCKED;
926 if ((use_limit = livelock_limit) < 100)
928 else if (use_limit > 500000)
930 systimer_init_periodic_nq(&ill_timer, ithread_livelock_wakeup,
931 (void *)(intptr_t)intr, use_limit);
933 case ISTATE_LIVELOCKED:
935 * Wait for our periodic timer to go off. Since the interrupt
936 * has re-armed it can still set i_running, but it will not
937 * reschedule us while we are in a livelocked state.
939 lwkt_deschedule_self(gd->gd_curthread);
943 * Check once a second to see if the livelock condition no
946 if (lseconds != gd->gd_time_seconds) {
947 lseconds = gd->gd_time_seconds;
948 if (ill_count < livelock_lowater) {
949 info->i_state = ISTATE_NORMAL;
950 systimer_del(&ill_timer);
951 kprintf("intr %d at %d/%d hz, livelock removed\n",
952 intr, ill_count, livelock_lowater);
953 } else if (livelock_debug == intr ||
954 (bootverbose && cold)) {
955 kprintf("intr %d at %d/%d hz, in livelock\n",
956 intr, ill_count, livelock_lowater);
967 * Emergency interrupt polling thread. The thread begins execution
968 * outside a critical section with the BGL held.
970 * If emergency interrupt polling is enabled, this thread will
971 * execute all system interrupts not marked INTR_NOPOLL at the
972 * specified polling frequency.
974 * WARNING! This thread runs *ALL* interrupt service routines that
975 * are not marked INTR_NOPOLL, which basically means everything except
976 * the 8254 clock interrupt and the ATA interrupt. It has very high
977 * overhead and should only be used in situations where the machine
978 * cannot otherwise be made to work. Due to the severe performance
979 * degredation, it should not be enabled on production machines.
982 ithread_emergency(void *arg __unused)
984 struct intr_info *info;
987 thread_t td __debugvar = curthread;
988 TD_INVARIANTS_DECLARE;
991 TD_INVARIANTS_GET(td);
994 for (intr = 0; intr < max_installed_hard_intr; ++intr) {
995 info = &intr_info_ary[intr];
996 for (rec = info->i_reclist; rec; rec = nrec) {
997 if ((rec->intr_flags & INTR_NOPOLL) == 0) {
998 if (rec->serializer) {
999 lwkt_serialize_handler_call(rec->serializer,
1000 rec->handler, rec->argument, NULL);
1002 rec->handler(rec->argument, NULL);
1004 TD_INVARIANTS_TEST(td, rec->name);
1009 lwkt_deschedule_self(curthread);
1015 * Systimer callback - schedule the emergency interrupt poll thread
1016 * if emergency polling is enabled.
1020 emergency_intr_timer_callback(systimer_t info, struct intrframe *frame __unused)
1022 if (emergency_intr_enable)
1023 lwkt_schedule(info->data);
1027 ithread_cpuid(int intr)
1029 const struct intr_info *info;
1031 KKASSERT(intr >= 0 && intr < MAX_INTS);
1032 info = &intr_info_ary[intr];
1034 if (info->i_state == ISTATE_NOTHREAD)
1036 return info->i_thread.td_gd->gd_cpuid;
1040 * Sysctls used by systat and others: hw.intrnames and hw.intrcnt.
1041 * The data for this machine dependent, and the declarations are in machine
1042 * dependent code. The layout of intrnames and intrcnt however is machine
1045 * We do not know the length of intrcnt and intrnames at compile time, so
1046 * calculate things at run time.
1050 sysctl_intrnames(SYSCTL_HANDLER_ARGS)
1052 struct intr_info *info;
1059 for (intr = 0; error == 0 && intr < MAX_INTS; ++intr) {
1060 info = &intr_info_ary[intr];
1064 for (rec = info->i_reclist; rec; rec = rec->next) {
1065 ksnprintf(buf + len, sizeof(buf) - len, "%s%s",
1066 (len ? "/" : ""), rec->name);
1067 len += strlen(buf + len);
1070 ksnprintf(buf, sizeof(buf), "irq%d", intr);
1073 error = SYSCTL_OUT(req, buf, len + 1);
1079 SYSCTL_PROC(_hw, OID_AUTO, intrnames, CTLTYPE_OPAQUE | CTLFLAG_RD,
1080 NULL, 0, sysctl_intrnames, "", "Interrupt Names");
1083 sysctl_intrcnt(SYSCTL_HANDLER_ARGS)
1085 struct intr_info *info;
1089 for (intr = 0; intr < max_installed_hard_intr; ++intr) {
1090 info = &intr_info_ary[intr];
1092 error = SYSCTL_OUT(req, &info->i_count, sizeof(info->i_count));
1096 for (intr = FIRST_SOFTINT; intr < max_installed_soft_intr; ++intr) {
1097 info = &intr_info_ary[intr];
1099 error = SYSCTL_OUT(req, &info->i_count, sizeof(info->i_count));
1107 SYSCTL_PROC(_hw, OID_AUTO, intrcnt, CTLTYPE_OPAQUE | CTLFLAG_RD,
1108 NULL, 0, sysctl_intrcnt, "", "Interrupt Counts");
1111 int_moveto_destcpu(int *orig_cpuid0, int *cpuid0, int intr)
1113 int orig_cpuid = mycpuid, cpuid;
1117 ksnprintf(envpath, sizeof(envpath), "hw.irq.%d.dest", intr);
1118 kgetenv_int(envpath, &cpuid);
1122 if (cpuid != orig_cpuid)
1123 lwkt_migratecpu(cpuid);
1125 *orig_cpuid0 = orig_cpuid;
1130 int_moveto_origcpu(int orig_cpuid, int cpuid)
1132 if (cpuid != orig_cpuid)
1133 lwkt_migratecpu(orig_cpuid);