lwp: Add two syscalls to set/get lwp's CPU affinity mask.
[dragonfly.git] / sys / kern / kern_usched.c
blobfb0b9e628a6865cd801150177933f2365d7306c6
1 /*
2 * Copyright (c) 2005 The DragonFly Project. All rights reserved.
3 *
4 * This code is derived from software contributed to The DragonFly Project
5 * by Sergey Glushchenko <deen@smz.com.ua>
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:
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in
15 * the documentation and/or other materials provided with the
16 * distribution.
17 * 3. Neither the name of The DragonFly Project nor the names of its
18 * contributors may be used to endorse or promote products derived
19 * from this software without specific, prior written permission.
21 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
24 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
25 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
26 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
27 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
28 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
29 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
30 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
31 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32 * SUCH DAMAGE.
36 #include <sys/errno.h>
37 #include <sys/globaldata.h> /* curthread */
38 #include <sys/proc.h>
39 #include <sys/priv.h>
40 #include <sys/sysproto.h> /* struct usched_set_args */
41 #include <sys/systm.h> /* strcmp() */
42 #include <sys/usched.h>
44 #include <sys/mplock2.h>
46 #include <machine/cpumask.h>
47 #include <machine/smp.h>
49 static TAILQ_HEAD(, usched) usched_list = TAILQ_HEAD_INITIALIZER(usched_list);
51 cpumask_t usched_mastermask = CPUMASK_INITIALIZER_ALLONES;
54 * Called from very low level boot code, i386/i386/machdep.c/init386().
55 * We cannot do anything fancy. no malloc's, no nothing other then
56 * static initialization.
58 struct usched *
59 usched_init(void)
61 const char *defsched;
63 defsched = kgetenv("kern.user_scheduler");
66 * Add various userland schedulers to the system.
68 usched_ctl(&usched_bsd4, USCH_ADD);
69 usched_ctl(&usched_dfly, USCH_ADD);
70 usched_ctl(&usched_dummy, USCH_ADD);
71 if (defsched == NULL )
72 return(&usched_dfly);
73 if (strcmp(defsched, "bsd4") == 0)
74 return(&usched_bsd4);
75 if (strcmp(defsched, "dfly") == 0)
76 return(&usched_dfly);
77 kprintf("WARNING: Running dummy userland scheduler\n");
78 return(&usched_dummy);
82 * USCHED_CTL
84 * SYNOPSIS:
85 * Add/remove usched to/from list.
87 * ARGUMENTS:
88 * usched - pointer to target scheduler
89 * action - addition or removal ?
91 * RETURN VALUES:
92 * 0 - success
93 * EINVAL - error
95 int
96 usched_ctl(struct usched *usched, int action)
98 struct usched *item; /* temporaly for TAILQ processing */
99 int error = 0;
101 switch(action) {
102 case USCH_ADD:
104 * Make sure it isn't already on the list
106 #ifdef INVARIANTS
107 TAILQ_FOREACH(item, &usched_list, entry) {
108 KKASSERT(item != usched);
110 #endif
112 * Optional callback to the scheduler before we officially
113 * add it to the list.
115 if (usched->usched_register)
116 usched->usched_register();
117 TAILQ_INSERT_TAIL(&usched_list, usched, entry);
118 break;
119 case USCH_REM:
121 * Do not allow the default scheduler to be removed
123 if (strcmp(usched->name, "bsd4") == 0) {
124 error = EINVAL;
125 break;
127 TAILQ_FOREACH(item, &usched_list, entry) {
128 if (item == usched)
129 break;
131 if (item) {
132 if (item->usched_unregister)
133 item->usched_unregister();
134 TAILQ_REMOVE(&usched_list, item, entry);
135 } else {
136 error = EINVAL;
138 break;
139 default:
140 error = EINVAL;
141 break;
143 return (error);
147 * Called from the scheduler clock on each cpu independently at the
148 * common scheduling rate. If th scheduler clock interrupted a running
149 * lwp the lp will be non-NULL.
151 void
152 usched_schedulerclock(struct lwp *lp, sysclock_t periodic, sysclock_t time)
154 struct usched *item;
156 TAILQ_FOREACH(item, &usched_list, entry) {
157 if (lp && lp->lwp_proc->p_usched == item)
158 item->schedulerclock(lp, periodic, time);
159 else
160 item->schedulerclock(NULL, periodic, time);
165 * USCHED_SET(syscall)
167 * SYNOPSIS:
168 * Setting up a proc's usched.
170 * ARGUMENTS:
171 * pid -
172 * cmd -
173 * data -
174 * bytes -
175 * RETURN VALUES:
176 * 0 - success
177 * EFBIG - error (invalid cpu#)
178 * EPERM - error (failed to delete cpu#)
179 * EINVAL - error (other reasons)
181 * MPALMOSTSAFE
184 sys_usched_set(struct usched_set_args *uap)
186 struct proc *p = curthread->td_proc;
187 struct usched *item; /* temporaly for TAILQ processing */
188 int error;
189 char buffer[NAME_LENGTH];
190 cpumask_t mask;
191 struct lwp *lp;
192 int cpuid;
194 if (uap->pid != 0 && uap->pid != curthread->td_proc->p_pid)
195 return (EINVAL);
197 lp = curthread->td_lwp;
198 get_mplock();
200 switch (uap->cmd) {
201 case USCHED_SET_SCHEDULER:
202 if ((error = priv_check(curthread, PRIV_SCHED_SET)) != 0)
203 break;
204 error = copyinstr(uap->data, buffer, sizeof(buffer), NULL);
205 if (error)
206 break;
207 TAILQ_FOREACH(item, &usched_list, entry) {
208 if ((strcmp(item->name, buffer) == 0))
209 break;
213 * If the scheduler for a process is being changed, disassociate
214 * the old scheduler before switching to the new one.
216 * XXX we might have to add an additional ABI call to do a 'full
217 * disassociation' and another ABI call to do a 'full
218 * reassociation'
220 /* XXX lwp have to deal with multiple lwps here */
221 if (p->p_nthreads != 1) {
222 error = EINVAL;
223 break;
225 if (item && item != p->p_usched) {
226 /* XXX lwp */
227 p->p_usched->release_curproc(ONLY_LWP_IN_PROC(p));
228 p->p_usched->heuristic_exiting(ONLY_LWP_IN_PROC(p), p);
229 p->p_usched = item;
230 } else if (item == NULL) {
231 error = EINVAL;
233 break;
234 case USCHED_SET_CPU:
235 if ((error = priv_check(curthread, PRIV_SCHED_CPUSET)) != 0)
236 break;
237 if (uap->bytes != sizeof(int)) {
238 error = EINVAL;
239 break;
241 error = copyin(uap->data, &cpuid, sizeof(int));
242 if (error)
243 break;
244 if (cpuid < 0 || cpuid >= ncpus) {
245 error = EFBIG;
246 break;
248 if (CPUMASK_TESTBIT(smp_active_mask, cpuid) == 0) {
249 error = EINVAL;
250 break;
252 CPUMASK_ASSBIT(lp->lwp_cpumask, cpuid);
253 if (cpuid != mycpu->gd_cpuid) {
254 lwkt_migratecpu(cpuid);
255 p->p_usched->changedcpu(lp);
257 break;
258 case USCHED_GET_CPU:
259 /* USCHED_GET_CPU doesn't require special privileges. */
260 if (uap->bytes != sizeof(int)) {
261 error = EINVAL;
262 break;
264 error = copyout(&(mycpu->gd_cpuid), uap->data, sizeof(int));
265 break;
266 case USCHED_GET_CPUMASK:
267 /* USCHED_GET_CPUMASK doesn't require special privileges. */
268 if (uap->bytes != sizeof(cpumask_t)) {
269 error = EINVAL;
270 break;
272 error = copyout(&lp->lwp_cpumask, uap->data, sizeof(cpumask_t));
273 break;
274 case USCHED_ADD_CPU:
275 if ((error = priv_check(curthread, PRIV_SCHED_CPUSET)) != 0)
276 break;
277 if (uap->bytes != sizeof(int)) {
278 error = EINVAL;
279 break;
281 error = copyin(uap->data, &cpuid, sizeof(int));
282 if (error)
283 break;
284 if (cpuid < 0 || cpuid >= ncpus) {
285 error = EFBIG;
286 break;
288 if (CPUMASK_TESTBIT(smp_active_mask, cpuid) == 0) {
289 error = EINVAL;
290 break;
292 CPUMASK_ORBIT(lp->lwp_cpumask, cpuid);
293 break;
294 case USCHED_DEL_CPU:
295 /* USCHED_DEL_CPU doesn't require special privileges. */
296 if (uap->bytes != sizeof(int)) {
297 error = EINVAL;
298 break;
300 error = copyin(uap->data, &cpuid, sizeof(int));
301 if (error)
302 break;
303 if (cpuid < 0 || cpuid >= ncpus) {
304 error = EFBIG;
305 break;
307 lp = curthread->td_lwp;
308 mask = lp->lwp_cpumask;
309 CPUMASK_ANDMASK(mask, smp_active_mask);
310 CPUMASK_NANDBIT(mask, cpuid);
311 if (CPUMASK_TESTZERO(mask)) {
312 error = EPERM;
313 } else {
314 CPUMASK_NANDBIT(lp->lwp_cpumask, cpuid);
315 if (CPUMASK_TESTMASK(lp->lwp_cpumask,
316 mycpu->gd_cpumask) == 0) {
317 mask = lp->lwp_cpumask;
318 CPUMASK_ANDMASK(mask, smp_active_mask);
319 cpuid = BSFCPUMASK(mask);
320 lwkt_migratecpu(cpuid);
321 p->p_usched->changedcpu(lp);
324 break;
325 case USCHED_SET_CPUMASK:
326 if ((error = priv_check(curthread, PRIV_SCHED_CPUSET)) != 0)
327 break;
328 if (uap->bytes != sizeof(mask)) {
329 error = EINVAL;
330 break;
332 error = copyin(uap->data, &mask, sizeof(mask));
333 if (error)
334 break;
336 CPUMASK_ANDMASK(mask, smp_active_mask);
337 if (CPUMASK_TESTZERO(mask)) {
338 error = EPERM;
339 break;
341 /* Commit the new cpumask. */
342 lp->lwp_cpumask = mask;
344 /* Migrate if necessary. */
345 if (CPUMASK_TESTMASK(lp->lwp_cpumask, mycpu->gd_cpumask) == 0) {
346 cpuid = BSFCPUMASK(lp->lwp_cpumask);
347 lwkt_migratecpu(cpuid);
348 p->p_usched->changedcpu(lp);
350 break;
351 default:
352 error = EINVAL;
353 break;
355 rel_mplock();
356 return (error);
360 sys_lwp_getaffinity(struct lwp_getaffinity_args *uap)
362 struct proc *p;
363 cpumask_t mask;
364 struct lwp *lp;
365 int error = 0;
367 if (uap->pid < 0) {
368 return (EINVAL);
369 } else if (uap->pid == 0) {
370 p = curproc;
371 PHOLD(p);
372 } else {
373 p = pfind(uap->pid);
374 if (p == NULL)
375 return (ESRCH);
376 /* pfind PHOLDs 'p'. */
379 get_mplock();
380 lwkt_gettoken(&p->p_token);
382 if (uap->tid < 0)
383 lp = curthread->td_lwp;
384 else
385 lp = lwp_rb_tree_RB_LOOKUP(&p->p_lwp_tree, uap->tid);
386 if (lp == NULL) {
387 error = ESRCH;
388 } else {
389 /* Take a snapshot for copyout, which may block. */
390 mask = lp->lwp_cpumask;
391 CPUMASK_ANDMASK(mask, smp_active_mask);
394 lwkt_reltoken(&p->p_token);
395 rel_mplock();
396 PRELE(p);
398 if (!error)
399 error = copyout(&mask, uap->mask, sizeof(cpumask_t));
400 return (error);
404 sys_lwp_setaffinity(struct lwp_setaffinity_args *uap)
406 struct proc *p;
407 cpumask_t mask;
408 struct lwp *lp;
409 int error;
411 if ((error = priv_check(curthread, PRIV_SCHED_CPUSET)) != 0)
412 return (error);
414 error = copyin(uap->mask, &mask, sizeof(mask));
415 if (error)
416 return (error);
418 CPUMASK_ANDMASK(mask, smp_active_mask);
419 if (CPUMASK_TESTZERO(mask))
420 return (EPERM);
422 if (uap->pid < 0) {
423 return (EINVAL);
424 } else if (uap->pid == 0) {
425 p = curproc;
426 PHOLD(p);
427 } else {
428 p = pfind(uap->pid);
429 if (p == NULL)
430 return (ESRCH);
431 /* pfind PHOLDs 'p'. */
434 get_mplock();
435 lwkt_gettoken(&p->p_token);
437 if (uap->tid < 0)
438 lp = curthread->td_lwp;
439 else
440 lp = lwp_rb_tree_RB_LOOKUP(&p->p_lwp_tree, uap->tid);
441 if (lp == NULL) {
442 error = ESRCH;
443 } else {
444 /* Commit the new cpumask. */
445 lp->lwp_cpumask = mask;
447 if (lp == curthread->td_lwp) {
449 * Self migration can be done immediately,
450 * if necessary.
452 if (CPUMASK_TESTBIT(lp->lwp_cpumask,
453 mycpu->gd_cpuid) == 0) {
454 lwkt_migratecpu(BSFCPUMASK(lp->lwp_cpumask));
455 p->p_usched->changedcpu(lp);
457 } else {
459 * NOTE:
460 * Real migration happens upon next rescheduling.
465 lwkt_reltoken(&p->p_token);
466 rel_mplock();
467 PRELE(p);
469 return (error);