sys_pset.c revision 1.9 1 /* $NetBSD: sys_pset.c,v 1.9 2008/09/30 16:28:45 rmind Exp $ */
2
3 /*
4 * Copyright (c) 2008, Mindaugas Rasiukevicius <rmind at NetBSD org>
5 * All rights reserved.
6 *
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
9 * are met:
10 * 1. Redistributions of source code must retain the above copyright
11 * notice, this list of conditions and the following disclaimer.
12 * 2. Redistributions in binary form must reproduce the above copyright
13 * notice, this list of conditions and the following disclaimer in the
14 * documentation and/or other materials provided with the distribution.
15 *
16 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
17 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
20 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
26 * SUCH DAMAGE.
27 */
28
29 /*
30 * Implementation of the Processor Sets.
31 *
32 * Locking
33 * The array of the processor-set structures and its members are protected
34 * by the global cpu_lock. Note that in scheduler, the very l_psid value
35 * might be used without lock held.
36 */
37
38 #include <sys/cdefs.h>
39 __KERNEL_RCSID(0, "$NetBSD: sys_pset.c,v 1.9 2008/09/30 16:28:45 rmind Exp $");
40
41 #include <sys/param.h>
42
43 #include <sys/cpu.h>
44 #include <sys/kauth.h>
45 #include <sys/kmem.h>
46 #include <sys/lwp.h>
47 #include <sys/mutex.h>
48 #include <sys/proc.h>
49 #include <sys/pset.h>
50 #include <sys/sched.h>
51 #include <sys/syscallargs.h>
52 #include <sys/sysctl.h>
53 #include <sys/systm.h>
54 #include <sys/types.h>
55
56 static pset_info_t ** psets;
57 static u_int psets_max;
58 static u_int psets_count;
59
60 static int psets_realloc(int);
61 static int psid_validate(psetid_t, bool);
62 static int kern_pset_create(psetid_t *);
63 static int kern_pset_destroy(psetid_t);
64
65 /*
66 * Initialization of the processor-sets.
67 */
68 void
69 psets_init(void)
70 {
71
72 psets_max = max(MAXCPUS, 32);
73 psets = kmem_zalloc(psets_max * sizeof(void *), KM_SLEEP);
74 psets_count = 0;
75 }
76
77 /*
78 * Reallocate the array of the processor-set structures.
79 */
80 static int
81 psets_realloc(int new_psets_max)
82 {
83 pset_info_t **new_psets, **old_psets;
84 const u_int newsize = new_psets_max * sizeof(void *);
85 u_int i, oldsize;
86
87 if (new_psets_max < 1)
88 return EINVAL;
89
90 new_psets = kmem_zalloc(newsize, KM_SLEEP);
91 mutex_enter(&cpu_lock);
92 old_psets = psets;
93 oldsize = psets_max * sizeof(void *);
94
95 /* Check if we can lower the size of the array */
96 if (new_psets_max < psets_max) {
97 for (i = new_psets_max; i < psets_max; i++) {
98 if (psets[i] == NULL)
99 continue;
100 mutex_exit(&cpu_lock);
101 kmem_free(new_psets, newsize);
102 return EBUSY;
103 }
104 }
105
106 /* Copy all pointers to the new array */
107 memcpy(new_psets, psets, newsize);
108 psets_max = new_psets_max;
109 psets = new_psets;
110 mutex_exit(&cpu_lock);
111
112 kmem_free(old_psets, oldsize);
113 return 0;
114 }
115
116 /*
117 * Validate processor-set ID.
118 */
119 static int
120 psid_validate(psetid_t psid, bool chkps)
121 {
122
123 KASSERT(mutex_owned(&cpu_lock));
124
125 if (chkps && (psid == PS_NONE || psid == PS_QUERY || psid == PS_MYID))
126 return 0;
127 if (psid <= 0 || psid > psets_max)
128 return EINVAL;
129 if (psets[psid - 1] == NULL)
130 return EINVAL;
131 if (psets[psid - 1]->ps_flags & PSET_BUSY)
132 return EBUSY;
133
134 return 0;
135 }
136
137 /*
138 * Create a processor-set.
139 */
140 static int
141 kern_pset_create(psetid_t *psid)
142 {
143 pset_info_t *pi;
144 u_int i;
145
146 if (psets_count == psets_max)
147 return ENOMEM;
148
149 pi = kmem_zalloc(sizeof(pset_info_t), KM_SLEEP);
150
151 mutex_enter(&cpu_lock);
152 if (psets_count == psets_max) {
153 mutex_exit(&cpu_lock);
154 kmem_free(pi, sizeof(pset_info_t));
155 return ENOMEM;
156 }
157
158 /* Find a free entry in the array */
159 for (i = 0; i < psets_max; i++)
160 if (psets[i] == NULL)
161 break;
162 KASSERT(i != psets_max);
163
164 psets[i] = pi;
165 psets_count++;
166 mutex_exit(&cpu_lock);
167
168 *psid = i + 1;
169 return 0;
170 }
171
172 /*
173 * Destroy a processor-set.
174 */
175 static int
176 kern_pset_destroy(psetid_t psid)
177 {
178 struct cpu_info *ci;
179 pset_info_t *pi;
180 struct lwp *l;
181 CPU_INFO_ITERATOR cii;
182 int error;
183
184 mutex_enter(&cpu_lock);
185 if (psid == PS_MYID) {
186 /* Use caller's processor-set ID */
187 psid = curlwp->l_psid;
188 }
189 error = psid_validate(psid, false);
190 if (error) {
191 mutex_exit(&cpu_lock);
192 return error;
193 }
194
195 /* Release the processor-set from all CPUs */
196 for (CPU_INFO_FOREACH(cii, ci)) {
197 struct schedstate_percpu *spc;
198
199 spc = &ci->ci_schedstate;
200 if (spc->spc_psid != psid)
201 continue;
202 spc->spc_psid = PS_NONE;
203 }
204 /* Mark that processor-set is going to be destroyed */
205 pi = psets[psid - 1];
206 pi->ps_flags |= PSET_BUSY;
207 mutex_exit(&cpu_lock);
208
209 /* Unmark the processor-set ID from each thread */
210 mutex_enter(proc_lock);
211 LIST_FOREACH(l, &alllwp, l_list) {
212 /* Safe to check and set without lock held */
213 if (l->l_psid != psid)
214 continue;
215 l->l_psid = PS_NONE;
216 }
217 mutex_exit(proc_lock);
218
219 /* Destroy the processor-set */
220 mutex_enter(&cpu_lock);
221 psets[psid - 1] = NULL;
222 psets_count--;
223 mutex_exit(&cpu_lock);
224
225 kmem_free(pi, sizeof(pset_info_t));
226 return 0;
227 }
228
229 /*
230 * General system calls for the processor-sets.
231 */
232
233 int
234 sys_pset_create(struct lwp *l, const struct sys_pset_create_args *uap,
235 register_t *retval)
236 {
237 /* {
238 syscallarg(psetid_t) *psid;
239 } */
240 psetid_t psid;
241 int error;
242
243 /* Available only for super-user */
244 if (kauth_authorize_system(l->l_cred, KAUTH_SYSTEM_PSET,
245 KAUTH_REQ_SYSTEM_PSET_CREATE, NULL, NULL, NULL))
246 return EPERM;
247
248 error = kern_pset_create(&psid);
249 if (error)
250 return error;
251
252 error = copyout(&psid, SCARG(uap, psid), sizeof(psetid_t));
253 if (error)
254 (void)kern_pset_destroy(psid);
255
256 return error;
257 }
258
259 int
260 sys_pset_destroy(struct lwp *l, const struct sys_pset_destroy_args *uap,
261 register_t *retval)
262 {
263 /* {
264 syscallarg(psetid_t) psid;
265 } */
266
267 /* Available only for super-user */
268 if (kauth_authorize_system(l->l_cred, KAUTH_SYSTEM_PSET,
269 KAUTH_REQ_SYSTEM_PSET_DESTROY,
270 KAUTH_ARG(SCARG(uap, psid)), NULL, NULL))
271 return EPERM;
272
273 return kern_pset_destroy(SCARG(uap, psid));
274 }
275
276 int
277 sys_pset_assign(struct lwp *l, const struct sys_pset_assign_args *uap,
278 register_t *retval)
279 {
280 /* {
281 syscallarg(psetid_t) psid;
282 syscallarg(cpuid_t) cpuid;
283 syscallarg(psetid_t) *opsid;
284 } */
285 struct cpu_info *ci;
286 struct schedstate_percpu *spc = NULL;
287 psetid_t psid = SCARG(uap, psid), opsid = 0;
288 CPU_INFO_ITERATOR cii;
289 int error = 0, nnone = 0;
290
291 /* Available only for super-user, except the case of PS_QUERY */
292 if (kauth_authorize_system(l->l_cred, KAUTH_SYSTEM_PSET,
293 KAUTH_REQ_SYSTEM_PSET_ASSIGN, KAUTH_ARG(SCARG(uap, psid)), NULL,
294 NULL))
295 return EPERM;
296
297 /* Find the target CPU */
298 mutex_enter(&cpu_lock);
299 for (CPU_INFO_FOREACH(cii, ci)) {
300 if (cpu_index(ci) == SCARG(uap, cpuid))
301 spc = &ci->ci_schedstate;
302 nnone += (ci->ci_schedstate.spc_psid == PS_NONE);
303 }
304 if (spc == NULL) {
305 mutex_exit(&cpu_lock);
306 return EINVAL;
307 }
308 error = psid_validate(psid, true);
309 if (error) {
310 mutex_exit(&cpu_lock);
311 return error;
312 }
313 opsid = spc->spc_psid;
314 switch (psid) {
315 case PS_QUERY:
316 break;
317 case PS_MYID:
318 psid = curlwp->l_psid;
319 /* FALLTHROUGH */
320 default:
321 /*
322 * Ensure at least one CPU stays in the default set,
323 * and that specified CPU is not offline.
324 */
325 if (psid != PS_NONE && ((spc->spc_flags & SPCF_OFFLINE) ||
326 (nnone == 1 && spc->spc_psid == PS_NONE))) {
327 mutex_exit(&cpu_lock);
328 return EBUSY;
329 }
330 spc->spc_psid = psid;
331 break;
332 }
333 mutex_exit(&cpu_lock);
334
335 if (SCARG(uap, opsid) != NULL)
336 error = copyout(&opsid, SCARG(uap, opsid), sizeof(psetid_t));
337
338 return error;
339 }
340
341 int
342 sys__pset_bind(struct lwp *l, const struct sys__pset_bind_args *uap,
343 register_t *retval)
344 {
345 /* {
346 syscallarg(idtype_t) idtype;
347 syscallarg(id_t) first_id;
348 syscallarg(id_t) second_id;
349 syscallarg(psetid_t) psid;
350 syscallarg(psetid_t) *opsid;
351 } */
352 struct cpu_info *ci;
353 struct proc *p;
354 struct lwp *t;
355 id_t id1, id2;
356 pid_t pid = 0;
357 lwpid_t lid = 0;
358 psetid_t psid, opsid;
359 int error = 0, lcnt;
360
361 psid = SCARG(uap, psid);
362
363 /* Available only for super-user, except the case of PS_QUERY */
364 if (kauth_authorize_system(l->l_cred, KAUTH_SYSTEM_PSET,
365 KAUTH_REQ_SYSTEM_PSET_BIND, KAUTH_ARG(SCARG(uap, psid)), NULL,
366 NULL))
367 return EPERM;
368
369 mutex_enter(&cpu_lock);
370 error = psid_validate(psid, true);
371 if (error) {
372 mutex_exit(&cpu_lock);
373 return error;
374 }
375 if (psid == PS_MYID)
376 psid = curlwp->l_psid;
377 if (psid != PS_QUERY && psid != PS_NONE)
378 psets[psid - 1]->ps_flags |= PSET_BUSY;
379 mutex_exit(&cpu_lock);
380
381 /*
382 * Get PID and LID from the ID.
383 */
384 p = l->l_proc;
385 id1 = SCARG(uap, first_id);
386 id2 = SCARG(uap, second_id);
387
388 switch (SCARG(uap, idtype)) {
389 case P_PID:
390 /*
391 * Process:
392 * First ID - PID;
393 * Second ID - ignored;
394 */
395 pid = (id1 == P_MYID) ? p->p_pid : id1;
396 lid = 0;
397 break;
398 case P_LWPID:
399 /*
400 * Thread (LWP):
401 * First ID - LID;
402 * Second ID - PID;
403 */
404 if (id1 == P_MYID) {
405 pid = p->p_pid;
406 lid = l->l_lid;
407 break;
408 }
409 lid = id1;
410 pid = (id2 == P_MYID) ? p->p_pid : id2;
411 break;
412 default:
413 error = EINVAL;
414 goto error;
415 }
416
417 /* Find the process */
418 mutex_enter(proc_lock);
419 p = p_find(pid, PFIND_LOCKED);
420 if (p == NULL) {
421 mutex_exit(proc_lock);
422 error = ESRCH;
423 goto error;
424 }
425 mutex_enter(p->p_lock);
426 mutex_exit(proc_lock);
427
428 /* Disallow modification of the system processes */
429 if (p->p_flag & PK_SYSTEM) {
430 mutex_exit(p->p_lock);
431 error = EPERM;
432 goto error;
433 }
434
435 /* Find the LWP(s) */
436 lcnt = 0;
437 ci = NULL;
438 LIST_FOREACH(t, &p->p_lwps, l_sibling) {
439 if (lid && lid != t->l_lid)
440 continue;
441 /*
442 * Bind the thread to the processor-set,
443 * take some CPU and migrate.
444 */
445 lwp_lock(t);
446 opsid = t->l_psid;
447 t->l_psid = psid;
448 ci = sched_takecpu(l);
449 /* Unlocks LWP */
450 lwp_migrate(t, ci);
451 lcnt++;
452 }
453 mutex_exit(p->p_lock);
454 if (lcnt == 0) {
455 error = ESRCH;
456 goto error;
457 }
458 if (SCARG(uap, opsid))
459 error = copyout(&opsid, SCARG(uap, opsid), sizeof(psetid_t));
460 error:
461 if (psid != PS_QUERY && psid != PS_NONE) {
462 mutex_enter(&cpu_lock);
463 psets[psid - 1]->ps_flags &= ~PSET_BUSY;
464 mutex_exit(&cpu_lock);
465 }
466 return error;
467 }
468
469 /*
470 * Sysctl nodes and initialization.
471 */
472
473 static int
474 sysctl_psets_max(SYSCTLFN_ARGS)
475 {
476 struct sysctlnode node;
477 int error, newsize;
478
479 node = *rnode;
480 node.sysctl_data = &newsize;
481
482 newsize = psets_max;
483 error = sysctl_lookup(SYSCTLFN_CALL(&node));
484 if (error || newp == NULL)
485 return error;
486
487 if (newsize <= 0)
488 return EINVAL;
489
490 sysctl_unlock();
491 error = psets_realloc(newsize);
492 sysctl_relock();
493 return error;
494 }
495
496 static int
497 sysctl_psets_list(SYSCTLFN_ARGS)
498 {
499 const size_t bufsz = 1024;
500 char *buf, tbuf[16];
501 int i, error;
502 size_t len;
503
504 sysctl_unlock();
505 buf = kmem_alloc(bufsz, KM_SLEEP);
506 snprintf(buf, bufsz, "%d:1", PS_NONE); /* XXX */
507
508 mutex_enter(&cpu_lock);
509 for (i = 0; i < psets_max; i++) {
510 if (psets[i] == NULL)
511 continue;
512 snprintf(tbuf, sizeof(tbuf), ",%d:2", i + 1); /* XXX */
513 strlcat(buf, tbuf, bufsz);
514 }
515 mutex_exit(&cpu_lock);
516 len = strlen(buf) + 1;
517 error = 0;
518 if (oldp != NULL)
519 error = copyout(buf, oldp, min(len, *oldlenp));
520 *oldlenp = len;
521 kmem_free(buf, bufsz);
522 sysctl_relock();
523 return error;
524 }
525
526 SYSCTL_SETUP(sysctl_pset_setup, "sysctl kern.pset subtree setup")
527 {
528 const struct sysctlnode *node = NULL;
529
530 sysctl_createv(clog, 0, NULL, NULL,
531 CTLFLAG_PERMANENT,
532 CTLTYPE_NODE, "kern", NULL,
533 NULL, 0, NULL, 0,
534 CTL_KERN, CTL_EOL);
535 sysctl_createv(clog, 0, NULL, &node,
536 CTLFLAG_PERMANENT,
537 CTLTYPE_NODE, "pset",
538 SYSCTL_DESCR("Processor-set options"),
539 NULL, 0, NULL, 0,
540 CTL_KERN, CTL_CREATE, CTL_EOL);
541
542 if (node == NULL)
543 return;
544
545 sysctl_createv(clog, 0, &node, NULL,
546 CTLFLAG_PERMANENT | CTLFLAG_READWRITE,
547 CTLTYPE_INT, "psets_max",
548 SYSCTL_DESCR("Maximal count of the processor-sets"),
549 sysctl_psets_max, 0, &psets_max, 0,
550 CTL_CREATE, CTL_EOL);
551 sysctl_createv(clog, 0, &node, NULL,
552 CTLFLAG_PERMANENT,
553 CTLTYPE_STRING, "list",
554 SYSCTL_DESCR("List of active sets"),
555 sysctl_psets_list, 0, NULL, 0,
556 CTL_CREATE, CTL_EOL);
557 }
558