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 1.21 2005/06/06 15:02:27 dillon 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>
41 #include <machine/ipl.h>
43 #include <sys/interrupt.h>
45 typedef struct intrec {
48 intrmask_t *maskptr; /* LEGACY */
54 static intrec_t *intlists[NHWI+NSWI];
55 static thread_t ithreads[NHWI+NSWI];
56 static struct thread ithread_ary[NHWI+NSWI];
57 static struct random_softc irandom_ary[NHWI+NSWI];
58 static int irunning[NHWI+NSWI];
59 static u_int ill_count[NHWI+NSWI]; /* interrupt livelock counter */
60 static u_int ill_ticks[NHWI+NSWI]; /* track elapsed to calculate freq */
61 static u_int ill_delta[NHWI+NSWI]; /* track elapsed to calculate freq */
62 static int ill_state[NHWI+NSWI]; /* current state */
63 static struct systimer ill_timer[NHWI+NSWI]; /* enforced freq. timer */
64 static struct systimer ill_rtimer[NHWI+NSWI]; /* recovery timer */
65 static intrmask_t dummy_intr_mask;
67 #define LIVELOCK_NONE 0
68 #define LIVELOCK_LIMITED 1
70 static int livelock_limit = 50000;
71 static int livelock_fallback = 20000;
72 SYSCTL_INT(_kern, OID_AUTO, livelock_limit,
73 CTLFLAG_RW, &livelock_limit, 0, "Livelock interrupt rate limit");
74 SYSCTL_INT(_kern, OID_AUTO, livelock_fallback,
75 CTLFLAG_RW, &livelock_fallback, 0, "Livelock interrupt fallback rate");
77 static void ithread_handler(void *arg);
80 * Register an SWI or INTerrupt handler.
82 * Note that maskptr exists to support legacy spl handling and is not intended
83 * to be permanent (because spls are not compatible with BGL removal).
86 register_swi(int intr, inthand2_t *handler, void *arg, const char *name,
89 if (intr < NHWI || intr >= NHWI + NSWI)
90 panic("register_swi: bad intr %d", intr);
91 return(register_int(intr, handler, arg, name, maskptr));
95 register_int(int intr, inthand2_t *handler, void *arg, const char *name,
102 if (intr < 0 || intr >= NHWI + NSWI)
103 panic("register_int: bad intr %d", intr);
105 maskptr = &dummy_intr_mask;
107 rec = malloc(sizeof(intrec_t), M_DEVBUF, M_NOWAIT);
109 panic("register_swi: malloc failed");
110 rec->handler = handler;
111 rec->maskptr = maskptr;
117 list = &intlists[intr];
120 * Create an interrupt thread if necessary, leave it in an unscheduled
123 if ((td = ithreads[intr]) == NULL) {
124 lwkt_create((void *)ithread_handler, (void *)intr, &ithreads[intr],
125 &ithread_ary[intr], TDF_STOPREQ|TDF_INTTHREAD, -1,
128 if (intr >= NHWI && intr < NHWI + NSWI)
129 lwkt_setpri(td, TDPRI_SOFT_NORM);
131 lwkt_setpri(td, TDPRI_INT_MED);
135 * Add the record to the interrupt list
137 crit_enter(); /* token */
138 while (*list != NULL)
139 list = &(*list)->next;
146 unregister_swi(int intr, inthand2_t *handler)
148 if (intr < NHWI || intr >= NHWI + NSWI)
149 panic("register_swi: bad intr %d", intr);
150 unregister_int(intr, handler);
154 unregister_int(int intr, inthand2_t handler)
159 if (intr < 0 || intr > NHWI + NSWI)
160 panic("register_int: bad intr %d", intr);
161 list = &intlists[intr];
163 while ((rec = *list) != NULL) {
164 if (rec->handler == (void *)handler) {
174 printf("warning: unregister_int: int %d handler %p not found\n",
180 swi_setpriority(int intr, int pri)
184 if (intr < NHWI || intr >= NHWI + NSWI)
185 panic("register_swi: bad intr %d", intr);
186 if ((td = ithreads[intr]) != NULL)
187 lwkt_setpri(td, pri);
191 register_randintr(int intr)
193 struct random_softc *sc = &irandom_ary[intr];
199 unregister_randintr(int intr)
201 struct random_softc *sc = &irandom_ary[intr];
206 * Dispatch an interrupt. If there's nothing to do we have a stray
207 * interrupt and can just return, leaving the interrupt masked.
209 * We need to schedule the interrupt and set its irunning[] bit. If
210 * we are not on the interrupt thread's cpu we have to send a message
211 * to the correct cpu that will issue the desired action (interlocking
212 * with the interrupt thread's critical section).
214 * We are NOT in a critical section, which will allow the scheduled
215 * interrupt to preempt us. The MP lock might *NOT* be held here.
218 sched_ithd_remote(void *arg)
220 sched_ithd((int)arg);
228 if ((td = ithreads[intr]) != NULL) {
229 if (intlists[intr] == NULL) {
230 printf("sched_ithd: stray interrupt %d\n", intr);
232 if (td->td_gd == mycpu) {
234 lwkt_schedule(td); /* preemption handled internally */
236 lwkt_send_ipiq(td->td_gd, sched_ithd_remote, (void *)intr);
240 printf("sched_ithd: stray interrupt %d\n", intr);
245 * This is run from a periodic SYSTIMER (and thus must be MP safe, the BGL
246 * might not be held).
249 ithread_livelock_wakeup(systimer_t info)
251 int intr = (int)info->data;
254 if ((td = ithreads[intr]) != NULL)
260 * Interrupt threads run this as their main loop.
262 * The handler begins execution outside a critical section and with the BGL
265 * The irunning state starts at 0. When an interrupt occurs, the hardware
266 * interrupt is disabled and sched_ithd() The HW interrupt remains disabled
267 * until all routines have run. We then call ithread_done() to reenable
268 * the HW interrupt and deschedule us until the next interrupt.
270 * We are responsible for atomically checking irunning[] and ithread_done()
271 * is responsible for atomically checking for platform-specific delayed
272 * interrupts. irunning[] for our irq is only set in the context of our cpu,
273 * so a critical section is a sufficient interlock.
275 #define LIVELOCK_TIMEFRAME(freq) ((freq) >> 2) /* 1/4 second */
278 ithread_handler(void *arg)
284 intrec_t **list = &intlists[intr];
287 struct random_softc *sc = &irandom_ary[intr];
288 globaldata_t gd = mycpu;
291 * The loop must be entered with one critical section held.
297 * We can get woken up by the livelock periodic code too, run the
298 * handlers only if there is a real interrupt pending. XXX
300 * Clear irunning[] prior to running the handlers to interlock
301 * again new events occuring during processing of existing events.
303 * For now run each handler in a critical section.
306 for (rec = *list; rec; rec = nrec) {
308 rec->handler(rec->argument);
312 * Do a quick exit/enter to catch any higher-priority
313 * interrupt sources and so user/system/interrupt statistics
314 * work for interrupt threads.
320 * This is our interrupt hook to add rate randomness to the random
324 add_interrupt_randomness(intr);
327 * This is our livelock test. If we hit the rate limit we
328 * limit ourselves to X interrupts/sec until the rate
329 * falls below 50% of that value, then we unlimit again.
331 * XXX calling cputimer_count() is expensive but a livelock may
332 * prevent other interrupts from occuring so we cannot use ticks.
334 cputicks = sys_cputimer->count();
336 bticks = cputicks - ill_ticks[intr];
337 ill_ticks[intr] = cputicks;
338 if (bticks > sys_cputimer->freq)
339 bticks = sys_cputimer->freq;
341 switch(ill_state[intr]) {
343 ill_delta[intr] += bticks;
344 if (ill_delta[intr] < LIVELOCK_TIMEFRAME(sys_cputimer->freq))
346 freq = (int64_t)ill_count[intr] * sys_cputimer->freq /
350 if (freq < livelock_limit)
352 printf("intr %d at %d hz, livelocked! limiting at %d hz\n",
353 intr, freq, livelock_fallback);
354 ill_state[intr] = LIVELOCK_LIMITED;
356 /* force periodic check to avoid stale removal (if ints stop) */
357 systimer_init_periodic(&ill_rtimer[intr], ithread_livelock_wakeup,
360 case LIVELOCK_LIMITED:
362 * Delay (us) before rearming the interrupt
364 systimer_init_oneshot(&ill_timer[intr], ithread_livelock_wakeup,
365 (void *)intr, 1 + 1000000 / livelock_fallback);
366 lwkt_deschedule_self(curthread);
369 /* in case we were woken up by something else */
370 systimer_del(&ill_timer[intr]);
373 * Calculate interrupt rate (note that due to our delay it
374 * will not exceed livelock_fallback).
376 ill_delta[intr] += bticks;
377 if (ill_delta[intr] < LIVELOCK_TIMEFRAME(sys_cputimer->freq))
379 freq = (int64_t)ill_count[intr] * sys_cputimer->freq /
383 if (freq < (livelock_fallback >> 1)) {
384 printf("intr %d at %d hz, removing livelock limit\n",
386 ill_state[intr] = LIVELOCK_NONE;
387 systimer_del(&ill_rtimer[intr]);
393 * There are two races here. irunning[] is set by sched_ithd()
394 * in the context of our cpu and is critical-section safe. We
395 * are responsible for checking it. ipending is not critical
396 * section safe and must be handled by the platform specific
397 * ithread_done() routine.
399 if (irunning[intr] == 0)
401 /* must be in critical section on loop */
407 * Sysctls used by systat and others: hw.intrnames and hw.intrcnt.
408 * The data for this machine dependent, and the declarations are in machine
409 * dependent code. The layout of intrnames and intrcnt however is machine
412 * We do not know the length of intrcnt and intrnames at compile time, so
413 * calculate things at run time.
416 sysctl_intrnames(SYSCTL_HANDLER_ARGS)
418 return (sysctl_handle_opaque(oidp, intrnames, eintrnames - intrnames,
422 SYSCTL_PROC(_hw, OID_AUTO, intrnames, CTLTYPE_OPAQUE | CTLFLAG_RD,
423 NULL, 0, sysctl_intrnames, "", "Interrupt Names");
426 sysctl_intrcnt(SYSCTL_HANDLER_ARGS)
428 return (sysctl_handle_opaque(oidp, intrcnt,
429 (char *)eintrcnt - (char *)intrcnt, req));
432 SYSCTL_PROC(_hw, OID_AUTO, intrcnt, CTLTYPE_OPAQUE | CTLFLAG_RD,
433 NULL, 0, sysctl_intrcnt, "", "Interrupt Counts");