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