]>
Commit | Line | Data |
---|---|---|
029632fb PZ |
1 | #include "sched.h" |
2 | ||
5091faa4 MG |
3 | #include <linux/proc_fs.h> |
4 | #include <linux/seq_file.h> | |
5 | #include <linux/kallsyms.h> | |
6 | #include <linux/utsname.h> | |
029632fb PZ |
7 | #include <linux/security.h> |
8 | #include <linux/export.h> | |
5091faa4 MG |
9 | |
10 | unsigned int __read_mostly sysctl_sched_autogroup_enabled = 1; | |
11 | static struct autogroup autogroup_default; | |
12 | static atomic_t autogroup_seq_nr; | |
13 | ||
029632fb | 14 | void __init autogroup_init(struct task_struct *init_task) |
5091faa4 | 15 | { |
07e06b01 | 16 | autogroup_default.tg = &root_task_group; |
5091faa4 MG |
17 | kref_init(&autogroup_default.kref); |
18 | init_rwsem(&autogroup_default.lock); | |
19 | init_task->signal->autogroup = &autogroup_default; | |
20 | } | |
21 | ||
029632fb | 22 | void autogroup_free(struct task_group *tg) |
5091faa4 MG |
23 | { |
24 | kfree(tg->autogroup); | |
25 | } | |
26 | ||
27 | static inline void autogroup_destroy(struct kref *kref) | |
28 | { | |
29 | struct autogroup *ag = container_of(kref, struct autogroup, kref); | |
30 | ||
f4493771 MG |
31 | #ifdef CONFIG_RT_GROUP_SCHED |
32 | /* We've redirected RT tasks to the root task group... */ | |
33 | ag->tg->rt_se = NULL; | |
34 | ag->tg->rt_rq = NULL; | |
35 | #endif | |
ace783b9 | 36 | sched_offline_group(ag->tg); |
5091faa4 MG |
37 | sched_destroy_group(ag->tg); |
38 | } | |
39 | ||
40 | static inline void autogroup_kref_put(struct autogroup *ag) | |
41 | { | |
42 | kref_put(&ag->kref, autogroup_destroy); | |
43 | } | |
44 | ||
45 | static inline struct autogroup *autogroup_kref_get(struct autogroup *ag) | |
46 | { | |
47 | kref_get(&ag->kref); | |
48 | return ag; | |
49 | } | |
50 | ||
4f821987 MG |
51 | static inline struct autogroup *autogroup_task_get(struct task_struct *p) |
52 | { | |
53 | struct autogroup *ag; | |
54 | unsigned long flags; | |
55 | ||
56 | if (!lock_task_sighand(p, &flags)) | |
57 | return autogroup_kref_get(&autogroup_default); | |
58 | ||
59 | ag = autogroup_kref_get(p->signal->autogroup); | |
60 | unlock_task_sighand(p, &flags); | |
61 | ||
62 | return ag; | |
63 | } | |
64 | ||
5091faa4 MG |
65 | static inline struct autogroup *autogroup_create(void) |
66 | { | |
67 | struct autogroup *ag = kzalloc(sizeof(*ag), GFP_KERNEL); | |
68 | struct task_group *tg; | |
69 | ||
70 | if (!ag) | |
71 | goto out_fail; | |
72 | ||
07e06b01 | 73 | tg = sched_create_group(&root_task_group); |
5091faa4 MG |
74 | if (IS_ERR(tg)) |
75 | goto out_free; | |
76 | ||
77 | kref_init(&ag->kref); | |
78 | init_rwsem(&ag->lock); | |
79 | ag->id = atomic_inc_return(&autogroup_seq_nr); | |
80 | ag->tg = tg; | |
f4493771 MG |
81 | #ifdef CONFIG_RT_GROUP_SCHED |
82 | /* | |
83 | * Autogroup RT tasks are redirected to the root task group | |
84 | * so we don't have to move tasks around upon policy change, | |
85 | * or flail around trying to allocate bandwidth on the fly. | |
86 | * A bandwidth exception in __sched_setscheduler() allows | |
1fe89e1b | 87 | * the policy change to proceed. |
f4493771 MG |
88 | */ |
89 | free_rt_sched_group(tg); | |
90 | tg->rt_se = root_task_group.rt_se; | |
91 | tg->rt_rq = root_task_group.rt_rq; | |
92 | #endif | |
5091faa4 MG |
93 | tg->autogroup = ag; |
94 | ||
41261b6a | 95 | sched_online_group(tg, &root_task_group); |
5091faa4 MG |
96 | return ag; |
97 | ||
98 | out_free: | |
99 | kfree(ag); | |
100 | out_fail: | |
101 | if (printk_ratelimit()) { | |
102 | printk(KERN_WARNING "autogroup_create: %s failure.\n", | |
1e58565e | 103 | ag ? "sched_create_group()" : "kzalloc()"); |
5091faa4 MG |
104 | } |
105 | ||
106 | return autogroup_kref_get(&autogroup_default); | |
107 | } | |
108 | ||
029632fb | 109 | bool task_wants_autogroup(struct task_struct *p, struct task_group *tg) |
5091faa4 MG |
110 | { |
111 | if (tg != &root_task_group) | |
112 | return false; | |
5091faa4 | 113 | /* |
18f649ef ON |
114 | * If we race with autogroup_move_group() the caller can use the old |
115 | * value of signal->autogroup but in this case sched_move_task() will | |
116 | * be called again before autogroup_kref_put(). | |
8e5bfa8c ON |
117 | * |
118 | * However, there is no way sched_autogroup_exit_task() could tell us | |
119 | * to avoid autogroup->tg, so we abuse PF_EXITING flag for this case. | |
5091faa4 | 120 | */ |
8e5bfa8c ON |
121 | if (p->flags & PF_EXITING) |
122 | return false; | |
123 | ||
5091faa4 MG |
124 | return true; |
125 | } | |
126 | ||
8e5bfa8c ON |
127 | void sched_autogroup_exit_task(struct task_struct *p) |
128 | { | |
129 | /* | |
130 | * We are going to call exit_notify() and autogroup_move_group() can't | |
131 | * see this thread after that: we can no longer use signal->autogroup. | |
132 | * See the PF_EXITING check in task_wants_autogroup(). | |
133 | */ | |
134 | sched_move_task(p); | |
135 | } | |
136 | ||
5091faa4 MG |
137 | static void |
138 | autogroup_move_group(struct task_struct *p, struct autogroup *ag) | |
139 | { | |
140 | struct autogroup *prev; | |
141 | struct task_struct *t; | |
142 | unsigned long flags; | |
143 | ||
144 | BUG_ON(!lock_task_sighand(p, &flags)); | |
145 | ||
146 | prev = p->signal->autogroup; | |
147 | if (prev == ag) { | |
148 | unlock_task_sighand(p, &flags); | |
149 | return; | |
150 | } | |
151 | ||
152 | p->signal->autogroup = autogroup_kref_get(ag); | |
18f649ef ON |
153 | /* |
154 | * We can't avoid sched_move_task() after we changed signal->autogroup, | |
155 | * this process can already run with task_group() == prev->tg or we can | |
156 | * race with cgroup code which can read autogroup = prev under rq->lock. | |
157 | * In the latter case for_each_thread() can not miss a migrating thread, | |
158 | * cpu_cgroup_attach() must not be possible after cgroup_exit() and it | |
159 | * can't be removed from thread list, we hold ->siglock. | |
8e5bfa8c ON |
160 | * |
161 | * If an exiting thread was already removed from thread list we rely on | |
162 | * sched_autogroup_exit_task(). | |
18f649ef | 163 | */ |
5aface53 | 164 | for_each_thread(p, t) |
5091faa4 | 165 | sched_move_task(t); |
18f649ef | 166 | |
5091faa4 MG |
167 | unlock_task_sighand(p, &flags); |
168 | autogroup_kref_put(prev); | |
169 | } | |
170 | ||
171 | /* Allocates GFP_KERNEL, cannot be called under any spinlock */ | |
172 | void sched_autogroup_create_attach(struct task_struct *p) | |
173 | { | |
c1ad41f1 | 174 | struct autogroup *ag = autogroup_create(); |
5091faa4 MG |
175 | |
176 | autogroup_move_group(p, ag); | |
25985edc | 177 | /* drop extra reference added by autogroup_create() */ |
5091faa4 MG |
178 | autogroup_kref_put(ag); |
179 | } | |
180 | EXPORT_SYMBOL(sched_autogroup_create_attach); | |
181 | ||
182 | /* Cannot be called under siglock. Currently has no users */ | |
183 | void sched_autogroup_detach(struct task_struct *p) | |
184 | { | |
185 | autogroup_move_group(p, &autogroup_default); | |
186 | } | |
187 | EXPORT_SYMBOL(sched_autogroup_detach); | |
188 | ||
189 | void sched_autogroup_fork(struct signal_struct *sig) | |
190 | { | |
4f821987 | 191 | sig->autogroup = autogroup_task_get(current); |
5091faa4 MG |
192 | } |
193 | ||
194 | void sched_autogroup_exit(struct signal_struct *sig) | |
195 | { | |
196 | autogroup_kref_put(sig->autogroup); | |
197 | } | |
198 | ||
199 | static int __init setup_autogroup(char *str) | |
200 | { | |
201 | sysctl_sched_autogroup_enabled = 0; | |
202 | ||
203 | return 1; | |
204 | } | |
205 | ||
206 | __setup("noautogroup", setup_autogroup); | |
207 | ||
c1ad41f1 IM |
208 | #ifdef CONFIG_PROC_FS |
209 | ||
210 | int proc_sched_autogroup_set_nice(struct task_struct *p, int nice) | |
211 | { | |
212 | static unsigned long next = INITIAL_JIFFIES; | |
213 | struct autogroup *ag; | |
83929cce | 214 | unsigned long shares; |
c1ad41f1 IM |
215 | int err; |
216 | ||
75e45d51 | 217 | if (nice < MIN_NICE || nice > MAX_NICE) |
c1ad41f1 IM |
218 | return -EINVAL; |
219 | ||
220 | err = security_task_setnice(current, nice); | |
221 | if (err) | |
222 | return err; | |
223 | ||
224 | if (nice < 0 && !can_nice(current, nice)) | |
225 | return -EPERM; | |
226 | ||
227 | /* this is a heavy operation taking global locks.. */ | |
228 | if (!capable(CAP_SYS_ADMIN) && time_before(jiffies, next)) | |
229 | return -EAGAIN; | |
230 | ||
231 | next = HZ / 10 + jiffies; | |
232 | ag = autogroup_task_get(p); | |
83929cce | 233 | shares = scale_load(sched_prio_to_weight[nice + 20]); |
c1ad41f1 IM |
234 | |
235 | down_write(&ag->lock); | |
83929cce | 236 | err = sched_group_set_shares(ag->tg, shares); |
c1ad41f1 IM |
237 | if (!err) |
238 | ag->nice = nice; | |
239 | up_write(&ag->lock); | |
240 | ||
241 | autogroup_kref_put(ag); | |
242 | ||
243 | return err; | |
244 | } | |
245 | ||
246 | void proc_sched_autogroup_show_task(struct task_struct *p, struct seq_file *m) | |
247 | { | |
248 | struct autogroup *ag = autogroup_task_get(p); | |
249 | ||
250 | if (!task_group_is_autogroup(ag->tg)) | |
251 | goto out; | |
252 | ||
253 | down_read(&ag->lock); | |
254 | seq_printf(m, "/autogroup-%ld nice %d\n", ag->id, ag->nice); | |
255 | up_read(&ag->lock); | |
256 | ||
257 | out: | |
258 | autogroup_kref_put(ag); | |
259 | } | |
260 | #endif /* CONFIG_PROC_FS */ | |
261 | ||
5091faa4 | 262 | #ifdef CONFIG_SCHED_DEBUG |
029632fb | 263 | int autogroup_path(struct task_group *tg, char *buf, int buflen) |
5091faa4 | 264 | { |
511f67a5 | 265 | if (!task_group_is_autogroup(tg)) |
8ecedd7a BR |
266 | return 0; |
267 | ||
5091faa4 MG |
268 | return snprintf(buf, buflen, "%s-%ld", "/autogroup", tg->autogroup->id); |
269 | } | |
270 | #endif /* CONFIG_SCHED_DEBUG */ |