xref: /dflybsd-src/sys/kern/kern_intr.c (revision 10f84ad9e1cb939878f6c1ec7e73c30dd133a0f2)
1 /*
2  * Copyright (c) 2003 Matthew Dillon <dillon@backplane.com> All rights reserved.
3  * Copyright (c) 1997, Stefan Esser <se@freebsd.org> All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice unmodified, this list of conditions, and the following
10  *    disclaimer.
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.
14  *
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.
25  *
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 1.18 2004/06/28 05:02:56 dillon Exp $
28  *
29  */
30 
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>
37 #include <sys/proc.h>
38 #include <sys/thread2.h>
39 #include <sys/random.h>
40 
41 #include <machine/ipl.h>
42 
43 #include <sys/interrupt.h>
44 
45 typedef struct intrec {
46     struct intrec *next;
47     inthand2_t	*handler;
48     void	*argument;
49     const char	*name;
50     int		intr;
51 } intrec_t;
52 
53 static intrec_t	*intlists[NHWI+NSWI];
54 static thread_t ithreads[NHWI+NSWI];
55 static struct thread ithread_ary[NHWI+NSWI];
56 static struct random_softc irandom_ary[NHWI+NSWI];
57 static int irunning[NHWI+NSWI];
58 static u_int ill_count[NHWI+NSWI];	/* interrupt livelock counter */
59 static u_int ill_ticks[NHWI+NSWI];	/* track elapsed to calculate freq */
60 static u_int ill_delta[NHWI+NSWI];	/* track elapsed to calculate freq */
61 static int ill_state[NHWI+NSWI];	/* current state */
62 static struct systimer ill_timer[NHWI+NSWI];	/* enforced freq. timer */
63 static struct systimer ill_rtimer[NHWI+NSWI];	/* recovery timer */
64 
65 #define LIVELOCK_NONE		0
66 #define LIVELOCK_LIMITED	1
67 
68 static int livelock_limit = 50000;
69 static int livelock_fallback = 20000;
70 SYSCTL_INT(_kern, OID_AUTO, livelock_limit,
71         CTLFLAG_RW, &livelock_limit, 0, "Livelock interrupt rate limit");
72 SYSCTL_INT(_kern, OID_AUTO, livelock_fallback,
73         CTLFLAG_RW, &livelock_fallback, 0, "Livelock interrupt fallback rate");
74 
75 static void ithread_handler(void *arg);
76 
77 thread_t
78 register_swi(int intr, inthand2_t *handler, void *arg, const char *name)
79 {
80     if (intr < NHWI || intr >= NHWI + NSWI)
81 	panic("register_swi: bad intr %d", intr);
82     return(register_int(intr, handler, arg, name));
83 }
84 
85 thread_t
86 register_int(int intr, inthand2_t *handler, void *arg, const char *name)
87 {
88     intrec_t **list;
89     intrec_t *rec;
90     thread_t td;
91 
92     if (intr < 0 || intr >= NHWI + NSWI)
93 	panic("register_int: bad intr %d", intr);
94 
95     rec = malloc(sizeof(intrec_t), M_DEVBUF, M_NOWAIT);
96     if (rec == NULL)
97 	panic("register_swi: malloc failed");
98     rec->handler = handler;
99     rec->argument = arg;
100     rec->name = name;
101     rec->intr = intr;
102     rec->next = NULL;
103 
104     list = &intlists[intr];
105 
106     /*
107      * Create an interrupt thread if necessary, leave it in an unscheduled
108      * state.  The kthread restore function exits a critical section before
109      * starting the function so we need *TWO* critical sections in order
110      * for the handler to begin running in one.
111      */
112     if ((td = ithreads[intr]) == NULL) {
113 	lwkt_create((void *)ithread_handler, (void *)intr, &ithreads[intr],
114 	    &ithread_ary[intr], TDF_STOPREQ|TDF_INTTHREAD, -1,
115 	    "ithread %d", intr);
116 	td = ithreads[intr];
117 	if (intr >= NHWI && intr < NHWI + NSWI)
118 	    lwkt_setpri(td, TDPRI_SOFT_NORM + TDPRI_CRIT * 2);
119 	else
120 	    lwkt_setpri(td, TDPRI_INT_MED + TDPRI_CRIT * 2);
121     }
122 
123     /*
124      * Add the record to the interrupt list
125      */
126     crit_enter();	/* token */
127     while (*list != NULL)
128 	list = &(*list)->next;
129     *list = rec;
130     crit_exit();
131     return(td);
132 }
133 
134 void
135 unregister_swi(int intr, inthand2_t *handler)
136 {
137     if (intr < NHWI || intr >= NHWI + NSWI)
138 	panic("register_swi: bad intr %d", intr);
139     unregister_int(intr, handler);
140 }
141 
142 void
143 unregister_int(int intr, inthand2_t handler)
144 {
145     intrec_t **list;
146     intrec_t *rec;
147 
148     if (intr < 0 || intr > NHWI + NSWI)
149 	panic("register_int: bad intr %d", intr);
150     list = &intlists[intr];
151     crit_enter();
152     while ((rec = *list) != NULL) {
153 	if (rec->handler == (void *)handler) {
154 	    *list = rec->next;
155 	    break;
156 	}
157 	list = &rec->next;
158     }
159     crit_exit();
160     if (rec != NULL) {
161 	free(rec, M_DEVBUF);
162     } else {
163 	printf("warning: unregister_int: int %d handler %p not found\n",
164 	    intr, handler);
165     }
166 }
167 
168 void
169 swi_setpriority(int intr, int pri)
170 {
171     struct thread *td;
172 
173     if (intr < NHWI || intr >= NHWI + NSWI)
174 	panic("register_swi: bad intr %d", intr);
175     if ((td = ithreads[intr]) != NULL)
176 	lwkt_setpri(td, pri);
177 }
178 
179 void
180 register_randintr(int intr)
181 {
182     struct random_softc *sc = &irandom_ary[intr];
183     sc->sc_intr = intr;
184     sc->sc_enabled = 1;
185 }
186 
187 void
188 unregister_randintr(int intr)
189 {
190     struct random_softc *sc = &irandom_ary[intr];
191     sc->sc_enabled = 0;
192 }
193 
194 /*
195  * Dispatch an interrupt.  If there's nothing to do we have a stray
196  * interrupt and can just return, leaving the interrupt masked.
197  *
198  * We need to schedule the interrupt and set its irunning[] bit.  If
199  * we are not on the interrupt thread's cpu we have to send a message
200  * to the correct cpu that will issue the desired action (interlocking
201  * with the interrupt thread's critical section).
202  *
203  * We are NOT in a critical section, which will allow the scheduled
204  * interrupt to preempt us.  The MP lock might *NOT* be held here.
205  */
206 static void
207 sched_ithd_remote(void *arg)
208 {
209     sched_ithd((int)arg);
210 }
211 
212 void
213 sched_ithd(int intr)
214 {
215     thread_t td;
216 
217     if ((td = ithreads[intr]) != NULL) {
218 	if (intlists[intr] == NULL) {
219 	    printf("sched_ithd: stray interrupt %d\n", intr);
220 	} else {
221 	    if (td->td_gd == mycpu) {
222 		irunning[intr] = 1;
223 		lwkt_schedule(td);	/* preemption handled internally */
224 	    } else {
225 		lwkt_send_ipiq(td->td_gd, sched_ithd_remote, (void *)intr);
226 	    }
227 	}
228     } else {
229 	printf("sched_ithd: stray interrupt %d\n", intr);
230     }
231 }
232 
233 /*
234  * This is run from a periodic SYSTIMER (and thus must be MP safe, the BGL
235  * might not be held).
236  */
237 static void
238 ithread_livelock_wakeup(systimer_t info)
239 {
240     int intr = (int)info->data;
241     thread_t td;
242 
243     if ((td = ithreads[intr]) != NULL)
244 	lwkt_schedule(td);
245 }
246 
247 
248 /*
249  * Interrupt threads run this as their main loop.  The handler should be
250  * in a critical section on entry and the BGL is usually left held (for now).
251  *
252  * The irunning state starts at 0.  When an interrupt occurs, the hardware
253  * interrupt is disabled and sched_ithd() The HW interrupt remains disabled
254  * until all routines have run.  We then call ithread_done() to reenable
255  * the HW interrupt and deschedule us until the next interrupt.
256  */
257 
258 #define LIVELOCK_TIMEFRAME(freq)	((freq) >> 2)	/* 1/4 second */
259 
260 static void
261 ithread_handler(void *arg)
262 {
263     int intr = (int)arg;
264     int freq;
265     u_int bticks;
266     u_int cputicks;
267     intrec_t **list = &intlists[intr];
268     intrec_t *rec;
269     intrec_t *nrec;
270     struct random_softc *sc = &irandom_ary[intr];
271 
272     KKASSERT(curthread->td_pri >= TDPRI_CRIT);
273     for (;;) {
274 	/*
275 	 * We can get woken up by the livelock periodic code too, run the
276 	 * handlers only if there is a real interrupt pending.  Clear
277 	 * irunning[] prior to running the handlers to interlock new
278 	 * events.
279 	 */
280 	irunning[intr] = 0;
281 	for (rec = *list; rec; rec = nrec) {
282 	    nrec = rec->next;
283 	    rec->handler(rec->argument);
284 	}
285 
286 	/*
287 	 * This is our interrupt hook to add rate randomness to the random
288 	 * number generator.
289 	 */
290 	if (sc->sc_enabled)
291 	    add_interrupt_randomness(intr);
292 
293 	/*
294 	 * This is our livelock test.  If we hit the rate limit we
295 	 * limit ourselves to 10000 interrupts/sec until the rate
296 	 * falls below 50% of that value, then we unlimit again.
297 	 */
298 	cputicks = cputimer_count();
299 	++ill_count[intr];
300 	bticks = cputicks - ill_ticks[intr];
301 	ill_ticks[intr] = cputicks;
302 	if (bticks > cputimer_freq)
303 	    bticks = cputimer_freq;
304 
305 	switch(ill_state[intr]) {
306 	case LIVELOCK_NONE:
307 	    ill_delta[intr] += bticks;
308 	    if (ill_delta[intr] < LIVELOCK_TIMEFRAME(cputimer_freq))
309 		break;
310 	    freq = (int64_t)ill_count[intr] * cputimer_freq / ill_delta[intr];
311 	    ill_delta[intr] = 0;
312 	    ill_count[intr] = 0;
313 	    if (freq < livelock_limit)
314 		break;
315 	    printf("intr %d at %d hz, livelocked! limiting at %d hz\n",
316 		intr, freq, livelock_fallback);
317 	    ill_state[intr] = LIVELOCK_LIMITED;
318 	    bticks = 0;
319 	    /* force periodic check to avoid stale removal (if ints stop) */
320 	    systimer_init_periodic(&ill_rtimer[intr], ithread_livelock_wakeup,
321 				(void *)intr, 1);
322 	    /* fall through */
323 	case LIVELOCK_LIMITED:
324 	    /*
325 	     * Delay (us) before rearming the interrupt
326 	     */
327 	    systimer_init_oneshot(&ill_timer[intr], ithread_livelock_wakeup,
328 				(void *)intr, 1 + 1000000 / livelock_fallback);
329 	    lwkt_deschedule_self(curthread);
330 	    lwkt_switch();
331 
332 	    /* in case we were woken up by something else */
333 	    systimer_del(&ill_timer[intr]);
334 
335 	    /*
336 	     * Calculate interrupt rate (note that due to our delay it
337 	     * will not exceed livelock_fallback).
338 	     */
339 	    ill_delta[intr] += bticks;
340 	    if (ill_delta[intr] < LIVELOCK_TIMEFRAME(cputimer_freq))
341 		break;
342 	    freq = (int64_t)ill_count[intr] * cputimer_freq / ill_delta[intr];
343 	    ill_delta[intr] = 0;
344 	    ill_count[intr] = 0;
345 	    if (freq < (livelock_fallback >> 1)) {
346 		printf("intr %d at %d hz, removing livelock limit\n",
347 			intr, freq);
348 		ill_state[intr] = LIVELOCK_NONE;
349 		systimer_del(&ill_rtimer[intr]);
350 	    }
351 	    break;
352 	}
353 
354 	/*
355 	 * If another interrupt has not been queued we can reenable the
356 	 * hardware interrupt and go to sleep.
357 	 */
358 	if (irunning[intr] == 0)
359 	    ithread_done(intr);
360     }
361 }
362 
363 /*
364  * Sysctls used by systat and others: hw.intrnames and hw.intrcnt.
365  * The data for this machine dependent, and the declarations are in machine
366  * dependent code.  The layout of intrnames and intrcnt however is machine
367  * independent.
368  *
369  * We do not know the length of intrcnt and intrnames at compile time, so
370  * calculate things at run time.
371  */
372 static int
373 sysctl_intrnames(SYSCTL_HANDLER_ARGS)
374 {
375 	return (sysctl_handle_opaque(oidp, intrnames, eintrnames - intrnames,
376 	    req));
377 }
378 
379 SYSCTL_PROC(_hw, OID_AUTO, intrnames, CTLTYPE_OPAQUE | CTLFLAG_RD,
380 	NULL, 0, sysctl_intrnames, "", "Interrupt Names");
381 
382 static int
383 sysctl_intrcnt(SYSCTL_HANDLER_ARGS)
384 {
385 	return (sysctl_handle_opaque(oidp, intrcnt,
386 	    (char *)eintrcnt - (char *)intrcnt, req));
387 }
388 
389 SYSCTL_PROC(_hw, OID_AUTO, intrcnt, CTLTYPE_OPAQUE | CTLFLAG_RD,
390 	NULL, 0, sysctl_intrcnt, "", "Interrupt Counts");
391