sys/kern/kern_cpuset.c

74  * Each process belongs to an identified set, by default this is set 1.  Each
76  * named set.  This creates an anonymous set which other threads and processes
79  * The named set is referred to herein as the 'base' set to avoid ambiguity.
80  * This set is usually a child of a 'root' set while the anonymous set may
84  * Threads inherit their set from their creator whether it be anonymous or
86  * shared.  To modify an anonymous set a new set is created with the desired
87  * mask and the same parent as the existing anonymous set.  This gives the
91  * or mask that is discovered via a pid, tid, or setid.  Modifying a set
94  * exist within the assigned parent set.
127 cpuset_ref(struct cpuset *set)
130 	refcount_acquire(&set->cs_ref);
131 	return (set);
135  * Walks up the tree from 'set' to find the root.  Returns the root
139 cpuset_refroot(struct cpuset *set)
142 	for (; set->cs_parent != NULL; set = set->cs_parent)
143 		if (set->cs_flags & CPU_SET_ROOT)
145 	cpuset_ref(set);
147 	return (set);
151  * Find the first non-anonymous set starting from 'set'.  Returns this set
152  * referenced.  May return the passed in set with an extra ref if it is
156 cpuset_refbase(struct cpuset *set)
159 	if (set->cs_id == CPUSET_INVALID)
160 		set = set->cs_parent;
161 	cpuset_ref(set);
163 	return (set);
170 cpuset_rel(struct cpuset *set)
174 	if (refcount_release(&set->cs_ref) == 0)
177 	LIST_REMOVE(set, cs_siblings);
178 	id = set->cs_id;
180 		LIST_REMOVE(set, cs_link);
182 	cpuset_rel(set->cs_parent);
183 	uma_zfree(cpuset_zone, set);
193 cpuset_rel_defer(struct setlist *head, struct cpuset *set)
196 	if (refcount_release(&set->cs_ref) == 0)
199 	LIST_REMOVE(set, cs_siblings);
200 	if (set->cs_id != CPUSET_INVALID)
201 		LIST_REMOVE(set, cs_link);
202 	LIST_INSERT_HEAD(head, set, cs_link);
207  * Complete a deferred release.  Removes the set from the list provided to
211 cpuset_rel_complete(struct cpuset *set)
213 	LIST_REMOVE(set, cs_link);
214 	cpuset_rel(set->cs_parent);
215 	uma_zfree(cpuset_zone, set);
219  * Find a set based on an id.  Returns it with a ref.
224 	struct cpuset *set;
229 	LIST_FOREACH(set, &cpuset_ids, cs_link)
230 		if (set->cs_id == setid)
232 	if (set)
233 		cpuset_ref(set);
237 	if (set != NULL && jailed(td->td_ucred)) {
241 		for (tset = set; tset != NULL; tset = tset->cs_parent)
245 			cpuset_rel(set);
246 			set = NULL;
250 	return (set);
254  * Create a set in the space provided in 'set' with the provided parameters.
255  * The set is returned with a single ref.  May return EDEADLK if the set
259 _cpuset_create(struct cpuset *set, struct cpuset *parent, const cpuset_t *mask,
265 	CPU_COPY(mask, &set->cs_mask);
266 	LIST_INIT(&set->cs_children);
267 	refcount_init(&set->cs_ref, 1);
268 	set->cs_flags = 0;
270 	CPU_AND(&set->cs_mask, &parent->cs_mask);
271 	set->cs_id = id;
272 	set->cs_parent = cpuset_ref(parent);
273 	LIST_INSERT_HEAD(&parent->cs_children, set, cs_siblings);
274 	if (set->cs_id != CPUSET_INVALID)
275 		LIST_INSERT_HEAD(&cpuset_ids, set, cs_link);
282  * Create a new non-anonymous set with the requested parent and mask.  May
289 	struct cpuset *set;
296 	*setp = set = uma_zalloc(cpuset_zone, M_WAITOK);
297 	error = _cpuset_create(set, parent, mask, id);
301 	uma_zfree(cpuset_zone, set);
308  * the tree of sets starting at 'set'.  Checks for sets that would become
312 cpuset_testupdate(struct cpuset *set, cpuset_t *mask, int check_mask)
319 	if (set->cs_flags & CPU_SET_RDONLY)
322 		if (!CPU_OVERLAP(&set->cs_mask, mask))
324 		CPU_COPY(&set->cs_mask, &newmask);
329 	LIST_FOREACH(nset, &set->cs_children, cs_siblings)
339 cpuset_update(struct cpuset *set, cpuset_t *mask)
344 	CPU_AND(&set->cs_mask, mask);
345 	LIST_FOREACH(nset, &set->cs_children, cs_siblings)
346 		cpuset_update(nset, &set->cs_mask);
352  * Modify the set 'set' to use a copy of the mask provided.  Apply this new
357 cpuset_modify(struct cpuset *set, cpuset_t *mask)
372 	    set->cs_flags & CPU_SET_ROOT)
375 	 * Verify that we have access to this set of
378 	root = set->cs_parent;
382 	error = cpuset_testupdate(set, mask, 0);
385 	CPU_COPY(mask, &set->cs_mask);
386 	cpuset_update(set, mask);
399  * For WHICH_SET returns a valid set with a new reference.
402  * the base set of the current thread.  May fail with ESRCH/EPERM.
408 	struct cpuset *set;
415 	*setp = set = NULL;
441 			set = cpuset_refbase(curthread->td_cpuset);
444 			set = cpuset_lookup(id, curthread);
445 		if (set) {
446 			*setp = set;
452 		/* Find `set' for prison with given id. */
484  * Create an anonymous set with the provided mask in the space provided by
485  * 'fset'.  If the passed in set is anonymous we use its parent otherwise
486  * the new set is a child of 'set'.
489 cpuset_shadow(struct cpuset *set, struct cpuset *fset, const cpuset_t *mask)
493 	if (set->cs_id == CPUSET_INVALID)
494 		parent = set->cs_parent;
496 		parent = set;
503  * Handle two cases for replacing the base set or mask of an entire process.
506  *    to the provided set and replaces all non-anonymous td_cpusets with the
507  *    provided set.
508  * 2) Mask is non-null and set is null.  This replaces or creates anonymous
516 cpuset_setproc(pid_t pid, struct cpuset *set, cpuset_t *mask)
563 		 * the set the thread is a member of.
571 		 * Verify that a new set won't leave an existing thread
573 		 * the set.
576 			if (!CPU_OVERLAP(&set->cs_mask, &tdset->cs_mask))
591 		 * If we presently have an anonymous set or are applying a
592 		 * mask we must create an anonymous shadow set.  That is
593 		 * either parented to our existing base or the supplied set.
595 		 * If we have a base set with no anonymous shadow we simply
605 				error = _cpuset_create(nset, set,
613 			nset = cpuset_ref(set);
636 cpusetobj_strprint(char *buf, const cpuset_t *set)
646 		bytesp = snprintf(tbuf, bufsiz, "%lx,", set->__bits[i]);
650 	snprintf(tbuf, bufsiz, "%lx", set->__bits[_NCPUWORDS - 1]);
659 cpusetobj_strscan(cpuset_t *set, const char *buf)
675 	CPU_ZERO(set);
677 		ret = sscanf(buf, "%lx,", &set->__bits[i]);
685 	ret = sscanf(buf, "%lx", &set->__bits[nwords - 1]);
698 	struct cpuset *set;
704 	error = cpuset_which(CPU_WHICH_TID, id, &p, &td, &set);
707 	set = NULL;
711 		set = td->td_cpuset;
718 	if (set)
719 		cpuset_rel(set);
760 		 * roll back to default set. We're not using cpuset_shadow()
762 		 * if default set does not contain all CPUs.
774 		 * Current set is either default (1) or
775 		 * shadowed version of default set.
777 		 * Allocate new root set to be able to shadow it
791 		/* Assume existing set was already allocated by previous call */
824  * 0 - The root set which should represent all valid processors in the
827  *     runs.  This set is immutable.
828  * 1 - The default set which all processes are a member of until changed.
835 	struct cpuset *set;
843 	 * Create the root system set for the whole machine.  Doesn't use
846 	set = uma_zalloc(cpuset_zone, M_WAITOK | M_ZERO);
847 	CPU_FILL(&set->cs_mask);
848 	LIST_INIT(&set->cs_children);
849 	LIST_INSERT_HEAD(&cpuset_ids, set, cs_link);
850 	set->cs_ref = 1;
851 	set->cs_flags = CPU_SET_ROOT;
852 	cpuset_zero = set;
853 	cpuset_root = &set->cs_mask;
856 	 * Now derive a default, modifiable set from that to give out.
858 	set = uma_zalloc(cpuset_zone, M_WAITOK);
859 	error = _cpuset_create(set, cpuset_zero, &cpuset_zero->cs_mask, 1);
860 	KASSERT(error == 0, ("Error creating default set: %d\n", error));
861 	cpuset_default = set;
878 	return (set);
883  * mark the new 'set' as root.
888  * In case of no error, returns the set in *setp locked with a reference.
893 	struct cpuset *set;
906 	/* Mark the set as root. */
907 	set = *setp;
908 	set->cs_flags |= CPU_SET_ROOT;
914 cpuset_setproc_update_set(struct proc *p, struct cpuset *set)
919 	KASSERT(set != NULL, ("[%s:%d] invalid set", __func__, __LINE__));
921 	cpuset_ref(set);
922 	error = cpuset_setproc(p->p_pid, set, NULL);
925 	cpuset_rel(set);
930  * This is called once the final set of system cpus is known.  Modifies
931  * the root set and all children and mark the root read-only.
940 		panic("Can't set initial cpuset mask.\n");
954 	struct cpuset *set;
960 	error = cpuset_create(&set, root, &root->cs_mask);
964 	error = copyout(&set->cs_id, uap->setid, sizeof(set->cs_id));
966 		error = cpuset_setproc(-1, set, NULL);
967 	cpuset_rel(set);
989 	struct cpuset *set;
997 	set = cpuset_lookup(setid, td);
998 	if (set == NULL)
1000 	error = cpuset_setproc(id, set, NULL);
1001 	cpuset_rel(set);
1026 	struct cpuset *set;
1034 	error = cpuset_which(which, id, &p, &ttd, &set);
1041 		set = cpuset_refbase(ttd->td_cpuset);
1054 		nset = cpuset_refroot(set);
1055 		cpuset_rel(set);
1056 		set = nset;
1063 	tmpid = set->cs_id;
1064 	cpuset_rel(set);
1094 	struct cpuset *set;
1102 	/* In Capability mode, you can only get your own CPU set. */
1113 	error = cpuset_which(which, id, &p, &ttd, &set);
1123 			set = cpuset_ref(ttd->td_cpuset);
1137 			nset = cpuset_refroot(set);
1139 			nset = cpuset_refbase(set);
1159 			CPU_COPY(&set->cs_mask, mask);
1178 	if (set)
1179 		cpuset_rel(set);
1211 	struct cpuset *set;
1219 	/* In Capability mode, you can only set your own CPU set. */
1233 	 * Verify that no high bits are set.
1252 		error = cpuset_which(which, id, &p, &ttd, &set);
1259 			set = cpuset_ref(ttd->td_cpuset);
1274 			nset = cpuset_refroot(set);
1276 			nset = cpuset_refbase(set);
1279 		cpuset_rel(set);
1291 			error = cpuset_which(which, id, &p, &ttd, &set);
1293 				error = cpuset_modify(set, mask);
1294 				cpuset_rel(set);
1318 ddb_display_cpuset(const cpuset_t *set)
1323 		if (CPU_ISSET(cpu, set)) {
1337 	struct cpuset *set;
1339 	LIST_FOREACH(set, &cpuset_ids, cs_link) {
1340 		db_printf("set=%p id=%-6u ref=%-6d flags=0x%04x parent id=%d\n",
1341 		    set, set->cs_id, set->cs_ref, set->cs_flags,
1342 		    (set->cs_parent != NULL) ? set->cs_parent->cs_id : 0);
1344 		ddb_display_cpuset(&set->cs_mask);