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 if (intr
< FIRST_SOFTINT
&& info
->i_slow
+ info
->i_fast
== 1) {
309 if (machintr_vector_setup(intr
, intr_flags
))
310 kprintf("machintr_vector_setup: failed on irq %d\n", intr
);
313 int_moveto_origcpu(orig_cpuid
, cpuid
);
319 unregister_swi(void *id
)
325 unregister_int(void *id
)
327 struct intr_info
*info
;
328 struct intrec
**list
;
330 int intr
, orig_cpuid
, cpuid
;
332 intr
= ((intrec_t
)id
)->intr
;
334 if (intr
< 0 || intr
>= MAX_INTS
)
335 panic("register_int: bad intr %d", intr
);
337 info
= &intr_info_ary
[intr
];
339 int_moveto_destcpu(&orig_cpuid
, &cpuid
, intr
);
342 * Remove the interrupt descriptor, adjust the descriptor count,
343 * and teardown the machine level vector if this was the last interrupt.
346 list
= &info
->i_reclist
;
347 while ((rec
= *list
) != NULL
) {
356 if (rec
->intr_flags
& INTR_FAST
)
360 if (intr
< FIRST_SOFTINT
&& info
->i_fast
+ info
->i_slow
== 0)
361 machintr_vector_teardown(intr
);
364 * Clear i_mplock_required if no handlers in the chain require the
367 for (rec0
= info
->i_reclist
; rec0
; rec0
= rec0
->next
) {
368 if ((rec0
->intr_flags
& INTR_MPSAFE
) == 0)
372 info
->i_mplock_required
= 0;
377 int_moveto_origcpu(orig_cpuid
, cpuid
);
383 kfree(rec
->name
, M_DEVBUF
);
384 kfree(rec
, M_DEVBUF
);
386 kprintf("warning: unregister_int: int %d handler for %s not found\n",
387 intr
, ((intrec_t
)id
)->name
);
392 get_registered_name(int intr
)
396 if (intr
< 0 || intr
>= MAX_INTS
)
397 panic("register_int: bad intr %d", intr
);
399 if ((rec
= intr_info_ary
[intr
].i_reclist
) == NULL
)
408 count_registered_ints(int intr
)
410 struct intr_info
*info
;
412 if (intr
< 0 || intr
>= MAX_INTS
)
413 panic("register_int: bad intr %d", intr
);
414 info
= &intr_info_ary
[intr
];
415 return(info
->i_fast
+ info
->i_slow
);
419 get_interrupt_counter(int intr
)
421 struct intr_info
*info
;
423 if (intr
< 0 || intr
>= MAX_INTS
)
424 panic("register_int: bad intr %d", intr
);
425 info
= &intr_info_ary
[intr
];
426 return(info
->i_count
);
431 swi_setpriority(int intr
, int pri
)
433 struct intr_info
*info
;
435 if (intr
< FIRST_SOFTINT
|| intr
>= MAX_INTS
)
436 panic("register_swi: bad intr %d", intr
);
437 info
= &intr_info_ary
[intr
];
438 if (info
->i_state
!= ISTATE_NOTHREAD
)
439 lwkt_setpri(&info
->i_thread
, pri
);
443 register_randintr(int intr
)
445 struct intr_info
*info
;
447 if (intr
< 0 || intr
>= MAX_INTS
)
448 panic("register_randintr: bad intr %d", intr
);
449 info
= &intr_info_ary
[intr
];
450 info
->i_random
.sc_intr
= intr
;
451 info
->i_random
.sc_enabled
= 1;
455 unregister_randintr(int intr
)
457 struct intr_info
*info
;
459 if (intr
< 0 || intr
>= MAX_INTS
)
460 panic("register_swi: bad intr %d", intr
);
461 info
= &intr_info_ary
[intr
];
462 info
->i_random
.sc_enabled
= -1;
466 next_registered_randintr(int intr
)
468 struct intr_info
*info
;
470 if (intr
< 0 || intr
>= MAX_INTS
)
471 panic("register_swi: bad intr %d", intr
);
472 while (intr
< MAX_INTS
) {
473 info
= &intr_info_ary
[intr
];
474 if (info
->i_random
.sc_enabled
> 0)
482 * Dispatch an interrupt. If there's nothing to do we have a stray
483 * interrupt and can just return, leaving the interrupt masked.
485 * We need to schedule the interrupt and set its i_running bit. If
486 * we are not on the interrupt thread's cpu we have to send a message
487 * to the correct cpu that will issue the desired action (interlocking
488 * with the interrupt thread's critical section). We do NOT attempt to
489 * reschedule interrupts whos i_running bit is already set because
490 * this would prematurely wakeup a livelock-limited interrupt thread.
492 * i_running is only tested/set on the same cpu as the interrupt thread.
494 * We are NOT in a critical section, which will allow the scheduled
495 * interrupt to preempt us. The MP lock might *NOT* be held here.
500 sched_ithd_remote(void *arg
)
502 sched_ithd((int)(intptr_t)arg
);
510 struct intr_info
*info
;
512 info
= &intr_info_ary
[intr
];
515 if (info
->i_state
!= ISTATE_NOTHREAD
) {
516 if (info
->i_reclist
== NULL
) {
517 report_stray_interrupt(intr
, info
);
520 if (info
->i_thread
.td_gd
== mycpu
) {
521 if (info
->i_running
== 0) {
523 if (info
->i_state
!= ISTATE_LIVELOCKED
)
524 lwkt_schedule(&info
->i_thread
); /* MIGHT PREEMPT */
527 lwkt_send_ipiq(info
->i_thread
.td_gd
,
528 sched_ithd_remote
, (void *)(intptr_t)intr
);
531 if (info
->i_running
== 0) {
533 if (info
->i_state
!= ISTATE_LIVELOCKED
)
534 lwkt_schedule(&info
->i_thread
); /* MIGHT PREEMPT */
539 report_stray_interrupt(intr
, info
);
544 report_stray_interrupt(int intr
, struct intr_info
*info
)
546 ++info
->i_straycount
;
547 if (info
->i_straycount
< 10) {
548 if (info
->i_errorticks
== ticks
)
550 info
->i_errorticks
= ticks
;
551 kprintf("sched_ithd: stray interrupt %d on cpu %d\n",
553 } else if (info
->i_straycount
== 10) {
554 kprintf("sched_ithd: %ld stray interrupts %d on cpu %d - "
555 "there will be no further reports\n",
556 info
->i_straycount
, intr
, mycpuid
);
561 * This is run from a periodic SYSTIMER (and thus must be MP safe, the BGL
562 * might not be held).
565 ithread_livelock_wakeup(systimer_t st
)
567 struct intr_info
*info
;
569 info
= &intr_info_ary
[(int)(intptr_t)st
->data
];
570 if (info
->i_state
!= ISTATE_NOTHREAD
)
571 lwkt_schedule(&info
->i_thread
);
575 * Schedule ithread within fast intr handler
577 * XXX Protect sched_ithd() call with gd_intr_nesting_level?
578 * Interrupts aren't enabled, but still...
581 ithread_fast_sched(int intr
, thread_t td
)
586 * We are already in critical section, exit it now to
591 crit_enter_quick(td
);
597 * This function is called directly from the ICU or APIC vector code assembly
598 * to process an interrupt. The critical section and interrupt deferral
599 * checks have already been done but the function is entered WITHOUT
600 * a critical section held. The BGL may or may not be held.
602 * Must return non-zero if we do not want the vector code to re-enable
603 * the interrupt (which we don't if we have to schedule the interrupt)
605 int ithread_fast_handler(struct intrframe
*frame
);
608 ithread_fast_handler(struct intrframe
*frame
)
611 struct intr_info
*info
;
612 struct intrec
**list
;
617 intrec_t rec
, next_rec
;
621 intr
= frame
->if_vec
;
625 /* We must be in critical section. */
626 KKASSERT(td
->td_pri
>= TDPRI_CRIT
);
628 info
= &intr_info_ary
[intr
];
631 * If we are not processing any FAST interrupts, just schedule the thing.
633 if (info
->i_fast
== 0) {
635 ithread_fast_sched(intr
, td
);
640 * This should not normally occur since interrupts ought to be
641 * masked if the ithread has been scheduled or is running.
647 * Bump the interrupt nesting level to process any FAST interrupts.
648 * Obtain the MP lock as necessary. If the MP lock cannot be obtained,
649 * schedule the interrupt thread to deal with the issue instead.
651 * To reduce overhead, just leave the MP lock held once it has been
654 ++gd
->gd_intr_nesting_level
;
656 must_schedule
= info
->i_slow
;
661 list
= &info
->i_reclist
;
662 for (rec
= *list
; rec
; rec
= next_rec
) {
663 next_rec
= rec
->next
; /* rec may be invalid after call */
665 if (rec
->intr_flags
& INTR_FAST
) {
667 if ((rec
->intr_flags
& INTR_MPSAFE
) == 0 && got_mplock
== 0) {
668 if (try_mplock() == 0) {
669 /* Couldn't get the MP lock; just schedule it. */
676 if (rec
->serializer
) {
677 must_schedule
+= lwkt_serialize_handler_try(
678 rec
->serializer
, rec
->handler
,
679 rec
->argument
, frame
);
681 rec
->handler(rec
->argument
, frame
);
689 --gd
->gd_intr_nesting_level
;
696 * If we had a problem, or mixed fast and slow interrupt handlers are
697 * registered, schedule the ithread to catch the missed records (it
698 * will just re-run all of them). A return value of 0 indicates that
699 * all handlers have been run and the interrupt can be re-enabled, and
700 * a non-zero return indicates that the interrupt thread controls
703 if (must_schedule
> 0)
704 ithread_fast_sched(intr
, td
);
705 else if (must_schedule
== 0)
707 return(must_schedule
);
711 * Interrupt threads run this as their main loop.
713 * The handler begins execution outside a critical section and with the BGL
716 * The i_running state starts at 0. When an interrupt occurs, the hardware
717 * interrupt is disabled and sched_ithd() The HW interrupt remains disabled
718 * until all routines have run. We then call ithread_done() to reenable
719 * the HW interrupt and deschedule us until the next interrupt.
721 * We are responsible for atomically checking i_running and ithread_done()
722 * is responsible for atomically checking for platform-specific delayed
723 * interrupts. i_running for our irq is only set in the context of our cpu,
724 * so a critical section is a sufficient interlock.
726 #define LIVELOCK_TIMEFRAME(freq) ((freq) >> 2) /* 1/4 second */
729 ithread_handler(void *arg
)
731 struct intr_info
*info
;
736 struct intrec
**list
;
739 struct systimer ill_timer
; /* enforced freq. timer */
740 u_int ill_count
; /* interrupt livelock counter */
743 intr
= (int)(intptr_t)arg
;
744 info
= &intr_info_ary
[intr
];
745 list
= &info
->i_reclist
;
748 * The loop must be entered with one critical section held. The thread
749 * is created with TDF_MPSAFE so the MP lock is not held on start.
752 lseconds
= gd
->gd_time_seconds
;
758 * The chain is only considered MPSAFE if all its interrupt handlers
759 * are MPSAFE. However, if intr_mpsafe has been turned off we
760 * always operate with the BGL.
763 if (intr_mpsafe
== 0) {
768 } else if (info
->i_mplock_required
!= mpheld
) {
769 if (info
->i_mplock_required
) {
770 KKASSERT(mpheld
== 0);
774 KKASSERT(mpheld
!= 0);
781 * scheduled cpu may have changed, see intr_get_mplock()
787 * If an interrupt is pending, clear i_running and execute the
788 * handlers. Note that certain types of interrupts can re-trigger
789 * and set i_running again.
791 * Each handler is run in a critical section. Note that we run both
792 * FAST and SLOW designated service routines.
794 if (info
->i_running
) {
799 report_stray_interrupt(intr
, info
);
801 for (rec
= *list
; rec
; rec
= nrec
) {
803 if (rec
->serializer
) {
804 lwkt_serialize_handler_call(rec
->serializer
, rec
->handler
,
805 rec
->argument
, NULL
);
807 rec
->handler(rec
->argument
, NULL
);
813 * This is our interrupt hook to add rate randomness to the random
816 if (info
->i_random
.sc_enabled
> 0)
817 add_interrupt_randomness(intr
);
820 * Unmask the interrupt to allow it to trigger again. This only
821 * applies to certain types of interrupts (typ level interrupts).
822 * This can result in the interrupt retriggering, but the retrigger
823 * will not be processed until we cycle our critical section.
825 * Only unmask interrupts while handlers are installed. It is
826 * possible to hit a situation where no handlers are installed
827 * due to a device driver livelocking and then tearing down its
828 * interrupt on close (the parallel bus being a good example).
831 machintr_intren(intr
);
834 * Do a quick exit/enter to catch any higher-priority interrupt
835 * sources, such as the statclock, so thread time accounting
836 * will still work. This may also cause an interrupt to re-trigger.
842 * LIVELOCK STATE MACHINE
844 switch(info
->i_state
) {
847 * Reset the count each second.
849 if (lseconds
!= gd
->gd_time_seconds
) {
850 lseconds
= gd
->gd_time_seconds
;
855 * If we did not exceed the frequency limit, we are done.
856 * If the interrupt has not retriggered we deschedule ourselves.
858 if (ill_count
<= livelock_limit
) {
859 if (info
->i_running
== 0) {
861 if (mpheld
&& intr_migrate
) {
866 lwkt_deschedule_self(gd
->gd_curthread
);
873 * Otherwise we are livelocked. Set up a periodic systimer
874 * to wake the thread up at the limit frequency.
876 kprintf("intr %d at %d/%d hz, livelocked limit engaged!\n",
877 intr
, ill_count
, livelock_limit
);
878 info
->i_state
= ISTATE_LIVELOCKED
;
879 if ((use_limit
= livelock_limit
) < 100)
881 else if (use_limit
> 500000)
883 systimer_init_periodic_nq(&ill_timer
, ithread_livelock_wakeup
,
884 (void *)(intptr_t)intr
, use_limit
);
886 case ISTATE_LIVELOCKED
:
888 * Wait for our periodic timer to go off. Since the interrupt
889 * has re-armed it can still set i_running, but it will not
890 * reschedule us while we are in a livelocked state.
892 lwkt_deschedule_self(gd
->gd_curthread
);
896 * Check once a second to see if the livelock condition no
899 if (lseconds
!= gd
->gd_time_seconds
) {
900 lseconds
= gd
->gd_time_seconds
;
901 if (ill_count
< livelock_lowater
) {
902 info
->i_state
= ISTATE_NORMAL
;
903 systimer_del(&ill_timer
);
904 kprintf("intr %d at %d/%d hz, livelock removed\n",
905 intr
, ill_count
, livelock_lowater
);
906 } else if (livelock_debug
== intr
||
907 (bootverbose
&& cold
)) {
908 kprintf("intr %d at %d/%d hz, in livelock\n",
909 intr
, ill_count
, livelock_lowater
);
922 * An interrupt thread is trying to get the MP lock. To avoid cpu-bound
923 * code in the kernel on cpu X from interfering we chase the MP lock.
926 intr_get_mplock(void)
930 if (intr_migrate
== 0) {
934 while (try_mplock() == 0) {
935 owner
= owner_mplock();
936 if (owner
>= 0 && owner
!= mycpu
->gd_cpuid
) {
937 lwkt_migratecpu(owner
);
938 ++intr_migrate_count
;
948 * Emergency interrupt polling thread. The thread begins execution
949 * outside a critical section with the BGL held.
951 * If emergency interrupt polling is enabled, this thread will
952 * execute all system interrupts not marked INTR_NOPOLL at the
953 * specified polling frequency.
955 * WARNING! This thread runs *ALL* interrupt service routines that
956 * are not marked INTR_NOPOLL, which basically means everything except
957 * the 8254 clock interrupt and the ATA interrupt. It has very high
958 * overhead and should only be used in situations where the machine
959 * cannot otherwise be made to work. Due to the severe performance
960 * degredation, it should not be enabled on production machines.
963 ithread_emergency(void *arg __unused
)
965 struct intr_info
*info
;
970 for (intr
= 0; intr
< max_installed_hard_intr
; ++intr
) {
971 info
= &intr_info_ary
[intr
];
972 for (rec
= info
->i_reclist
; rec
; rec
= nrec
) {
973 if ((rec
->intr_flags
& INTR_NOPOLL
) == 0) {
974 if (rec
->serializer
) {
975 lwkt_serialize_handler_call(rec
->serializer
,
976 rec
->handler
, rec
->argument
, NULL
);
978 rec
->handler(rec
->argument
, NULL
);
984 lwkt_deschedule_self(curthread
);
990 * Systimer callback - schedule the emergency interrupt poll thread
991 * if emergency polling is enabled.
995 emergency_intr_timer_callback(systimer_t info
, struct intrframe
*frame __unused
)
997 if (emergency_intr_enable
)
998 lwkt_schedule(info
->data
);
1002 ithread_cpuid(int intr
)
1004 const struct intr_info
*info
;
1006 KKASSERT(intr
>= 0 && intr
< MAX_INTS
);
1007 info
= &intr_info_ary
[intr
];
1009 if (info
->i_state
== ISTATE_NOTHREAD
)
1011 return info
->i_thread
.td_gd
->gd_cpuid
;
1015 * Sysctls used by systat and others: hw.intrnames and hw.intrcnt.
1016 * The data for this machine dependent, and the declarations are in machine
1017 * dependent code. The layout of intrnames and intrcnt however is machine
1020 * We do not know the length of intrcnt and intrnames at compile time, so
1021 * calculate things at run time.
1025 sysctl_intrnames(SYSCTL_HANDLER_ARGS
)
1027 struct intr_info
*info
;
1034 for (intr
= 0; error
== 0 && intr
< MAX_INTS
; ++intr
) {
1035 info
= &intr_info_ary
[intr
];
1039 for (rec
= info
->i_reclist
; rec
; rec
= rec
->next
) {
1040 ksnprintf(buf
+ len
, sizeof(buf
) - len
, "%s%s",
1041 (len
? "/" : ""), rec
->name
);
1042 len
+= strlen(buf
+ len
);
1045 ksnprintf(buf
, sizeof(buf
), "irq%d", intr
);
1048 error
= SYSCTL_OUT(req
, buf
, len
+ 1);
1054 SYSCTL_PROC(_hw
, OID_AUTO
, intrnames
, CTLTYPE_OPAQUE
| CTLFLAG_RD
,
1055 NULL
, 0, sysctl_intrnames
, "", "Interrupt Names");
1058 sysctl_intrcnt(SYSCTL_HANDLER_ARGS
)
1060 struct intr_info
*info
;
1064 for (intr
= 0; intr
< max_installed_hard_intr
; ++intr
) {
1065 info
= &intr_info_ary
[intr
];
1067 error
= SYSCTL_OUT(req
, &info
->i_count
, sizeof(info
->i_count
));
1071 for (intr
= FIRST_SOFTINT
; intr
< max_installed_soft_intr
; ++intr
) {
1072 info
= &intr_info_ary
[intr
];
1074 error
= SYSCTL_OUT(req
, &info
->i_count
, sizeof(info
->i_count
));
1082 SYSCTL_PROC(_hw
, OID_AUTO
, intrcnt
, CTLTYPE_OPAQUE
| CTLFLAG_RD
,
1083 NULL
, 0, sysctl_intrcnt
, "", "Interrupt Counts");
1086 int_moveto_destcpu(int *orig_cpuid0
, int *cpuid0
, int intr
)
1088 int orig_cpuid
= mycpuid
, cpuid
;
1092 ksnprintf(envpath
, sizeof(envpath
), "hw.irq.%d.dest", intr
);
1093 kgetenv_int(envpath
, &cpuid
);
1097 if (cpuid
!= orig_cpuid
)
1098 lwkt_migratecpu(cpuid
);
1100 *orig_cpuid0
= orig_cpuid
;
1105 int_moveto_origcpu(int orig_cpuid
, int cpuid
)
1107 if (cpuid
!= orig_cpuid
)
1108 lwkt_migratecpu(orig_cpuid
);