subr_xcall.c revision 1.13.16.2 1 1.13.16.2 snj /* $NetBSD: subr_xcall.c,v 1.13.16.2 2017/07/06 15:18:23 snj Exp $ */
2 1.2 ad
3 1.2 ad /*-
4 1.12 rmind * Copyright (c) 2007-2010 The NetBSD Foundation, Inc.
5 1.2 ad * All rights reserved.
6 1.2 ad *
7 1.2 ad * This code is derived from software contributed to The NetBSD Foundation
8 1.12 rmind * by Andrew Doran and Mindaugas Rasiukevicius.
9 1.2 ad *
10 1.2 ad * Redistribution and use in source and binary forms, with or without
11 1.2 ad * modification, are permitted provided that the following conditions
12 1.2 ad * are met:
13 1.2 ad * 1. Redistributions of source code must retain the above copyright
14 1.2 ad * notice, this list of conditions and the following disclaimer.
15 1.2 ad * 2. Redistributions in binary form must reproduce the above copyright
16 1.2 ad * notice, this list of conditions and the following disclaimer in the
17 1.2 ad * documentation and/or other materials provided with the distribution.
18 1.2 ad *
19 1.2 ad * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
20 1.2 ad * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
21 1.2 ad * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
22 1.2 ad * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
23 1.2 ad * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
24 1.2 ad * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
25 1.2 ad * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
26 1.2 ad * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
27 1.2 ad * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
28 1.2 ad * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
29 1.2 ad * POSSIBILITY OF SUCH DAMAGE.
30 1.2 ad */
31 1.2 ad
32 1.2 ad /*
33 1.2 ad * Cross call support
34 1.2 ad *
35 1.2 ad * Background
36 1.2 ad *
37 1.2 ad * Sometimes it is necessary to modify hardware state that is tied
38 1.2 ad * directly to individual CPUs (such as a CPU's local timer), and
39 1.2 ad * these updates can not be done remotely by another CPU. The LWP
40 1.2 ad * requesting the update may be unable to guarantee that it will be
41 1.2 ad * running on the CPU where the update must occur, when the update
42 1.2 ad * occurs.
43 1.2 ad *
44 1.2 ad * Additionally, it's sometimes necessary to modify per-CPU software
45 1.2 ad * state from a remote CPU. Where these update operations are so
46 1.2 ad * rare or the access to the per-CPU data so frequent that the cost
47 1.2 ad * of using locking or atomic operations to provide coherency is
48 1.4 ad * prohibitive, another way must be found.
49 1.2 ad *
50 1.2 ad * Cross calls help to solve these types of problem by allowing
51 1.2 ad * any CPU in the system to request that an arbitrary function be
52 1.2 ad * executed on any other CPU.
53 1.2 ad *
54 1.2 ad * Implementation
55 1.2 ad *
56 1.2 ad * A slow mechanism for making 'low priority' cross calls is
57 1.2 ad * provided. The function to be executed runs on the remote CPU
58 1.2 ad * within a bound kthread. No queueing is provided, and the
59 1.2 ad * implementation uses global state. The function being called may
60 1.2 ad * block briefly on locks, but in doing so must be careful to not
61 1.2 ad * interfere with other cross calls in the system. The function is
62 1.2 ad * called with thread context and not from a soft interrupt, so it
63 1.2 ad * can ensure that it is not interrupting other code running on the
64 1.2 ad * CPU, and so has exclusive access to the CPU. Since this facility
65 1.2 ad * is heavyweight, it's expected that it will not be used often.
66 1.2 ad *
67 1.4 ad * Cross calls must not allocate memory, as the pagedaemon uses
68 1.4 ad * them (and memory allocation may need to wait on the pagedaemon).
69 1.4 ad *
70 1.12 rmind * A low-overhead mechanism for high priority calls (XC_HIGHPRI) is
71 1.12 rmind * also provided. The function to be executed runs on a software
72 1.12 rmind * interrupt context, at SOFTINT_CLOCK level, and is expected to be
73 1.12 rmind * very lightweight, e.g. avoid blocking.
74 1.2 ad */
75 1.2 ad
76 1.2 ad #include <sys/cdefs.h>
77 1.13.16.2 snj __KERNEL_RCSID(0, "$NetBSD: subr_xcall.c,v 1.13.16.2 2017/07/06 15:18:23 snj Exp $");
78 1.2 ad
79 1.2 ad #include <sys/types.h>
80 1.2 ad #include <sys/param.h>
81 1.2 ad #include <sys/xcall.h>
82 1.2 ad #include <sys/mutex.h>
83 1.2 ad #include <sys/condvar.h>
84 1.2 ad #include <sys/evcnt.h>
85 1.2 ad #include <sys/kthread.h>
86 1.3 ad #include <sys/cpu.h>
87 1.2 ad
88 1.12 rmind /* Cross-call state box. */
89 1.12 rmind typedef struct {
90 1.12 rmind kmutex_t xc_lock;
91 1.12 rmind kcondvar_t xc_busy;
92 1.12 rmind xcfunc_t xc_func;
93 1.12 rmind void * xc_arg1;
94 1.12 rmind void * xc_arg2;
95 1.12 rmind uint64_t xc_headp;
96 1.12 rmind uint64_t xc_donep;
97 1.12 rmind } xc_state_t;
98 1.12 rmind
99 1.12 rmind /* Bit indicating high (1) or low (0) priority. */
100 1.12 rmind #define XC_PRI_BIT (1ULL << 63)
101 1.12 rmind
102 1.12 rmind /* Low priority xcall structures. */
103 1.13 rmind static xc_state_t xc_low_pri __cacheline_aligned;
104 1.12 rmind
105 1.12 rmind /* High priority xcall structures. */
106 1.13 rmind static xc_state_t xc_high_pri __cacheline_aligned;
107 1.13 rmind static void * xc_sih __cacheline_aligned;
108 1.2 ad
109 1.12 rmind /* Event counters. */
110 1.13 rmind static struct evcnt xc_unicast_ev __cacheline_aligned;
111 1.13 rmind static struct evcnt xc_broadcast_ev __cacheline_aligned;
112 1.12 rmind
113 1.12 rmind static void xc_init(void);
114 1.12 rmind static void xc_thread(void *);
115 1.12 rmind static void xc_highpri_intr(void *);
116 1.12 rmind
117 1.12 rmind static inline uint64_t xc_highpri(xcfunc_t, void *, void *, struct cpu_info *);
118 1.12 rmind static inline uint64_t xc_lowpri(xcfunc_t, void *, void *, struct cpu_info *);
119 1.12 rmind
120 1.12 rmind /*
121 1.12 rmind * xc_init:
122 1.12 rmind *
123 1.12 rmind * Initialize low and high priority cross-call structures.
124 1.12 rmind */
125 1.12 rmind static void
126 1.12 rmind xc_init(void)
127 1.12 rmind {
128 1.12 rmind xc_state_t *xclo = &xc_low_pri, *xchi = &xc_high_pri;
129 1.12 rmind
130 1.12 rmind memset(xclo, 0, sizeof(xc_state_t));
131 1.12 rmind mutex_init(&xclo->xc_lock, MUTEX_DEFAULT, IPL_NONE);
132 1.12 rmind cv_init(&xclo->xc_busy, "xclocv");
133 1.12 rmind
134 1.12 rmind memset(xchi, 0, sizeof(xc_state_t));
135 1.12 rmind mutex_init(&xchi->xc_lock, MUTEX_DEFAULT, IPL_SOFTCLOCK);
136 1.12 rmind cv_init(&xchi->xc_busy, "xchicv");
137 1.12 rmind xc_sih = softint_establish(SOFTINT_CLOCK | SOFTINT_MPSAFE,
138 1.12 rmind xc_highpri_intr, NULL);
139 1.12 rmind KASSERT(xc_sih != NULL);
140 1.12 rmind
141 1.12 rmind evcnt_attach_dynamic(&xc_unicast_ev, EVCNT_TYPE_MISC, NULL,
142 1.12 rmind "crosscall", "unicast");
143 1.12 rmind evcnt_attach_dynamic(&xc_broadcast_ev, EVCNT_TYPE_MISC, NULL,
144 1.12 rmind "crosscall", "broadcast");
145 1.12 rmind }
146 1.2 ad
147 1.2 ad /*
148 1.2 ad * xc_init_cpu:
149 1.2 ad *
150 1.2 ad * Initialize the cross-call subsystem. Called once for each CPU
151 1.2 ad * in the system as they are attached.
152 1.2 ad */
153 1.2 ad void
154 1.2 ad xc_init_cpu(struct cpu_info *ci)
155 1.2 ad {
156 1.11 pooka static bool again = false;
157 1.2 ad int error;
158 1.2 ad
159 1.2 ad if (!again) {
160 1.2 ad /* Autoconfiguration will prevent re-entry. */
161 1.12 rmind xc_init();
162 1.2 ad again = true;
163 1.2 ad }
164 1.2 ad cv_init(&ci->ci_data.cpu_xcall, "xcall");
165 1.2 ad error = kthread_create(PRI_XCALL, KTHREAD_MPSAFE, ci, xc_thread,
166 1.6 martin NULL, NULL, "xcall/%u", ci->ci_index);
167 1.12 rmind KASSERT(error == 0);
168 1.2 ad }
169 1.2 ad
170 1.2 ad /*
171 1.7 ad * xc_broadcast:
172 1.2 ad *
173 1.2 ad * Trigger a call on all CPUs in the system.
174 1.2 ad */
175 1.2 ad uint64_t
176 1.2 ad xc_broadcast(u_int flags, xcfunc_t func, void *arg1, void *arg2)
177 1.2 ad {
178 1.2 ad
179 1.12 rmind KASSERT(!cpu_intr_p() && !cpu_softintr_p());
180 1.12 rmind
181 1.2 ad if ((flags & XC_HIGHPRI) != 0) {
182 1.12 rmind return xc_highpri(func, arg1, arg2, NULL);
183 1.2 ad } else {
184 1.12 rmind return xc_lowpri(func, arg1, arg2, NULL);
185 1.2 ad }
186 1.2 ad }
187 1.2 ad
188 1.2 ad /*
189 1.2 ad * xc_unicast:
190 1.2 ad *
191 1.2 ad * Trigger a call on one CPU.
192 1.2 ad */
193 1.2 ad uint64_t
194 1.2 ad xc_unicast(u_int flags, xcfunc_t func, void *arg1, void *arg2,
195 1.2 ad struct cpu_info *ci)
196 1.2 ad {
197 1.2 ad
198 1.12 rmind KASSERT(ci != NULL);
199 1.12 rmind KASSERT(!cpu_intr_p() && !cpu_softintr_p());
200 1.12 rmind
201 1.2 ad if ((flags & XC_HIGHPRI) != 0) {
202 1.12 rmind return xc_highpri(func, arg1, arg2, ci);
203 1.12 rmind } else {
204 1.12 rmind return xc_lowpri(func, arg1, arg2, ci);
205 1.12 rmind }
206 1.12 rmind }
207 1.12 rmind
208 1.12 rmind /*
209 1.12 rmind * xc_wait:
210 1.12 rmind *
211 1.12 rmind * Wait for a cross call to complete.
212 1.12 rmind */
213 1.12 rmind void
214 1.12 rmind xc_wait(uint64_t where)
215 1.12 rmind {
216 1.12 rmind xc_state_t *xc;
217 1.12 rmind
218 1.12 rmind KASSERT(!cpu_intr_p() && !cpu_softintr_p());
219 1.12 rmind
220 1.12 rmind /* Determine whether it is high or low priority cross-call. */
221 1.12 rmind if ((where & XC_PRI_BIT) != 0) {
222 1.12 rmind xc = &xc_high_pri;
223 1.12 rmind where &= ~XC_PRI_BIT;
224 1.2 ad } else {
225 1.12 rmind xc = &xc_low_pri;
226 1.12 rmind }
227 1.12 rmind
228 1.12 rmind /* Fast path, if already done. */
229 1.12 rmind if (xc->xc_donep >= where) {
230 1.12 rmind return;
231 1.12 rmind }
232 1.12 rmind
233 1.12 rmind /* Slow path: block until awoken. */
234 1.12 rmind mutex_enter(&xc->xc_lock);
235 1.12 rmind while (xc->xc_donep < where) {
236 1.12 rmind cv_wait(&xc->xc_busy, &xc->xc_lock);
237 1.2 ad }
238 1.12 rmind mutex_exit(&xc->xc_lock);
239 1.2 ad }
240 1.2 ad
241 1.2 ad /*
242 1.2 ad * xc_lowpri:
243 1.2 ad *
244 1.2 ad * Trigger a low priority call on one or more CPUs.
245 1.2 ad */
246 1.12 rmind static inline uint64_t
247 1.12 rmind xc_lowpri(xcfunc_t func, void *arg1, void *arg2, struct cpu_info *ci)
248 1.2 ad {
249 1.12 rmind xc_state_t *xc = &xc_low_pri;
250 1.2 ad CPU_INFO_ITERATOR cii;
251 1.10 uebayasi uint64_t where;
252 1.2 ad
253 1.12 rmind mutex_enter(&xc->xc_lock);
254 1.13.16.2 snj while (xc->xc_headp != xc->xc_donep) {
255 1.12 rmind cv_wait(&xc->xc_busy, &xc->xc_lock);
256 1.12 rmind }
257 1.12 rmind xc->xc_arg1 = arg1;
258 1.12 rmind xc->xc_arg2 = arg2;
259 1.12 rmind xc->xc_func = func;
260 1.2 ad if (ci == NULL) {
261 1.2 ad xc_broadcast_ev.ev_count++;
262 1.2 ad for (CPU_INFO_FOREACH(cii, ci)) {
263 1.8 ad if ((ci->ci_schedstate.spc_flags & SPCF_RUNNING) == 0)
264 1.8 ad continue;
265 1.12 rmind xc->xc_headp += 1;
266 1.2 ad ci->ci_data.cpu_xcall_pending = true;
267 1.2 ad cv_signal(&ci->ci_data.cpu_xcall);
268 1.2 ad }
269 1.2 ad } else {
270 1.2 ad xc_unicast_ev.ev_count++;
271 1.12 rmind xc->xc_headp += 1;
272 1.2 ad ci->ci_data.cpu_xcall_pending = true;
273 1.2 ad cv_signal(&ci->ci_data.cpu_xcall);
274 1.2 ad }
275 1.13.16.2 snj KASSERT(xc->xc_donep < xc->xc_headp);
276 1.12 rmind where = xc->xc_headp;
277 1.12 rmind mutex_exit(&xc->xc_lock);
278 1.2 ad
279 1.12 rmind /* Return a low priority ticket. */
280 1.12 rmind KASSERT((where & XC_PRI_BIT) == 0);
281 1.2 ad return where;
282 1.2 ad }
283 1.2 ad
284 1.2 ad /*
285 1.2 ad * xc_thread:
286 1.2 ad *
287 1.2 ad * One thread per-CPU to dispatch low priority calls.
288 1.2 ad */
289 1.2 ad static void
290 1.2 ad xc_thread(void *cookie)
291 1.2 ad {
292 1.12 rmind struct cpu_info *ci = curcpu();
293 1.12 rmind xc_state_t *xc = &xc_low_pri;
294 1.2 ad void *arg1, *arg2;
295 1.2 ad xcfunc_t func;
296 1.2 ad
297 1.12 rmind mutex_enter(&xc->xc_lock);
298 1.2 ad for (;;) {
299 1.2 ad while (!ci->ci_data.cpu_xcall_pending) {
300 1.13.16.2 snj if (xc->xc_headp == xc->xc_donep) {
301 1.12 rmind cv_broadcast(&xc->xc_busy);
302 1.12 rmind }
303 1.12 rmind cv_wait(&ci->ci_data.cpu_xcall, &xc->xc_lock);
304 1.2 ad KASSERT(ci == curcpu());
305 1.2 ad }
306 1.2 ad ci->ci_data.cpu_xcall_pending = false;
307 1.12 rmind func = xc->xc_func;
308 1.12 rmind arg1 = xc->xc_arg1;
309 1.12 rmind arg2 = xc->xc_arg2;
310 1.12 rmind mutex_exit(&xc->xc_lock);
311 1.2 ad
312 1.12 rmind KASSERT(func != NULL);
313 1.2 ad (*func)(arg1, arg2);
314 1.2 ad
315 1.12 rmind mutex_enter(&xc->xc_lock);
316 1.12 rmind xc->xc_donep++;
317 1.2 ad }
318 1.2 ad /* NOTREACHED */
319 1.2 ad }
320 1.12 rmind
321 1.12 rmind /*
322 1.12 rmind * xc_ipi_handler:
323 1.12 rmind *
324 1.12 rmind * Handler of cross-call IPI.
325 1.12 rmind */
326 1.12 rmind void
327 1.12 rmind xc_ipi_handler(void)
328 1.12 rmind {
329 1.12 rmind /* Executes xc_highpri_intr() via software interrupt. */
330 1.12 rmind softint_schedule(xc_sih);
331 1.12 rmind }
332 1.12 rmind
333 1.12 rmind /*
334 1.12 rmind * xc_highpri_intr:
335 1.12 rmind *
336 1.12 rmind * A software interrupt handler for high priority calls.
337 1.12 rmind */
338 1.12 rmind static void
339 1.12 rmind xc_highpri_intr(void *dummy)
340 1.12 rmind {
341 1.12 rmind xc_state_t *xc = &xc_high_pri;
342 1.12 rmind void *arg1, *arg2;
343 1.12 rmind xcfunc_t func;
344 1.12 rmind
345 1.12 rmind KASSERT(cpu_softintr_p());
346 1.12 rmind /*
347 1.12 rmind * Lock-less fetch of function and its arguments.
348 1.12 rmind * Safe since it cannot change at this point.
349 1.12 rmind */
350 1.12 rmind KASSERT(xc->xc_donep < xc->xc_headp);
351 1.12 rmind func = xc->xc_func;
352 1.12 rmind arg1 = xc->xc_arg1;
353 1.12 rmind arg2 = xc->xc_arg2;
354 1.12 rmind
355 1.12 rmind KASSERT(func != NULL);
356 1.12 rmind (*func)(arg1, arg2);
357 1.12 rmind
358 1.12 rmind /*
359 1.12 rmind * Note the request as done, and if we have reached the head,
360 1.12 rmind * cross-call has been processed - notify waiters, if any.
361 1.12 rmind */
362 1.12 rmind mutex_enter(&xc->xc_lock);
363 1.12 rmind if (++xc->xc_donep == xc->xc_headp) {
364 1.12 rmind cv_broadcast(&xc->xc_busy);
365 1.12 rmind }
366 1.12 rmind mutex_exit(&xc->xc_lock);
367 1.12 rmind }
368 1.12 rmind
369 1.12 rmind /*
370 1.12 rmind * xc_highpri:
371 1.12 rmind *
372 1.12 rmind * Trigger a high priority call on one or more CPUs.
373 1.12 rmind */
374 1.12 rmind static inline uint64_t
375 1.12 rmind xc_highpri(xcfunc_t func, void *arg1, void *arg2, struct cpu_info *ci)
376 1.12 rmind {
377 1.12 rmind xc_state_t *xc = &xc_high_pri;
378 1.12 rmind uint64_t where;
379 1.12 rmind
380 1.12 rmind mutex_enter(&xc->xc_lock);
381 1.12 rmind while (xc->xc_headp != xc->xc_donep) {
382 1.12 rmind cv_wait(&xc->xc_busy, &xc->xc_lock);
383 1.12 rmind }
384 1.12 rmind xc->xc_func = func;
385 1.12 rmind xc->xc_arg1 = arg1;
386 1.12 rmind xc->xc_arg2 = arg2;
387 1.12 rmind xc->xc_headp += (ci ? 1 : ncpu);
388 1.12 rmind where = xc->xc_headp;
389 1.12 rmind mutex_exit(&xc->xc_lock);
390 1.12 rmind
391 1.12 rmind /*
392 1.12 rmind * Send the IPI once lock is released.
393 1.12 rmind * Note: it will handle the local CPU case.
394 1.12 rmind */
395 1.12 rmind
396 1.12 rmind #ifdef MULTIPROCESSOR
397 1.12 rmind kpreempt_disable();
398 1.12 rmind if (curcpu() == ci) {
399 1.12 rmind /* Unicast: local CPU. */
400 1.12 rmind xc_ipi_handler();
401 1.12 rmind } else if (ci) {
402 1.12 rmind /* Unicast: remote CPU. */
403 1.12 rmind xc_send_ipi(ci);
404 1.12 rmind } else {
405 1.12 rmind /* Broadcast: all, including local. */
406 1.12 rmind xc_send_ipi(NULL);
407 1.12 rmind xc_ipi_handler();
408 1.12 rmind }
409 1.12 rmind kpreempt_enable();
410 1.12 rmind #else
411 1.13.16.1 bouyer KASSERT(ci == NULL || curcpu() == ci);
412 1.12 rmind xc_ipi_handler();
413 1.12 rmind #endif
414 1.12 rmind
415 1.12 rmind /* Indicate a high priority ticket. */
416 1.12 rmind return (where | XC_PRI_BIT);
417 1.12 rmind }
418