diff options
Diffstat (limited to 'kernel/stop_machine.c')
-rw-r--r-- | kernel/stop_machine.c | 288 |
1 files changed, 131 insertions, 157 deletions
diff --git a/kernel/stop_machine.c b/kernel/stop_machine.c index 738b411ff2d3..e446c7c7d6a9 100644 --- a/kernel/stop_machine.c +++ b/kernel/stop_machine.c | |||
@@ -1,4 +1,4 @@ | |||
1 | /* Copyright 2005 Rusty Russell rusty@rustcorp.com.au IBM Corporation. | 1 | /* Copyright 2008, 2005 Rusty Russell rusty@rustcorp.com.au IBM Corporation. |
2 | * GPL v2 and any later version. | 2 | * GPL v2 and any later version. |
3 | */ | 3 | */ |
4 | #include <linux/cpu.h> | 4 | #include <linux/cpu.h> |
@@ -13,204 +13,178 @@ | |||
13 | #include <asm/atomic.h> | 13 | #include <asm/atomic.h> |
14 | #include <asm/uaccess.h> | 14 | #include <asm/uaccess.h> |
15 | 15 | ||
16 | /* Since we effect priority and affinity (both of which are visible | 16 | /* This controls the threads on each CPU. */ |
17 | * to, and settable by outside processes) we do indirection via a | ||
18 | * kthread. */ | ||
19 | |||
20 | /* Thread to stop each CPU in user context. */ | ||
21 | enum stopmachine_state { | 17 | enum stopmachine_state { |
22 | STOPMACHINE_WAIT, | 18 | /* Dummy starting state for thread. */ |
19 | STOPMACHINE_NONE, | ||
20 | /* Awaiting everyone to be scheduled. */ | ||
23 | STOPMACHINE_PREPARE, | 21 | STOPMACHINE_PREPARE, |
22 | /* Disable interrupts. */ | ||
24 | STOPMACHINE_DISABLE_IRQ, | 23 | STOPMACHINE_DISABLE_IRQ, |
24 | /* Run the function */ | ||
25 | STOPMACHINE_RUN, | ||
26 | /* Exit */ | ||
25 | STOPMACHINE_EXIT, | 27 | STOPMACHINE_EXIT, |
26 | }; | 28 | }; |
29 | static enum stopmachine_state state; | ||
27 | 30 | ||
28 | static enum stopmachine_state stopmachine_state; | 31 | struct stop_machine_data { |
29 | static unsigned int stopmachine_num_threads; | 32 | int (*fn)(void *); |
30 | static atomic_t stopmachine_thread_ack; | 33 | void *data; |
31 | 34 | int fnret; | |
32 | static int stopmachine(void *cpu) | 35 | }; |
33 | { | ||
34 | int irqs_disabled = 0; | ||
35 | int prepared = 0; | ||
36 | cpumask_of_cpu_ptr(cpumask, (int)(long)cpu); | ||
37 | |||
38 | set_cpus_allowed_ptr(current, cpumask); | ||
39 | |||
40 | /* Ack: we are alive */ | ||
41 | smp_mb(); /* Theoretically the ack = 0 might not be on this CPU yet. */ | ||
42 | atomic_inc(&stopmachine_thread_ack); | ||
43 | |||
44 | /* Simple state machine */ | ||
45 | while (stopmachine_state != STOPMACHINE_EXIT) { | ||
46 | if (stopmachine_state == STOPMACHINE_DISABLE_IRQ | ||
47 | && !irqs_disabled) { | ||
48 | local_irq_disable(); | ||
49 | hard_irq_disable(); | ||
50 | irqs_disabled = 1; | ||
51 | /* Ack: irqs disabled. */ | ||
52 | smp_mb(); /* Must read state first. */ | ||
53 | atomic_inc(&stopmachine_thread_ack); | ||
54 | } else if (stopmachine_state == STOPMACHINE_PREPARE | ||
55 | && !prepared) { | ||
56 | /* Everyone is in place, hold CPU. */ | ||
57 | preempt_disable(); | ||
58 | prepared = 1; | ||
59 | smp_mb(); /* Must read state first. */ | ||
60 | atomic_inc(&stopmachine_thread_ack); | ||
61 | } | ||
62 | /* Yield in first stage: migration threads need to | ||
63 | * help our sisters onto their CPUs. */ | ||
64 | if (!prepared && !irqs_disabled) | ||
65 | yield(); | ||
66 | cpu_relax(); | ||
67 | } | ||
68 | |||
69 | /* Ack: we are exiting. */ | ||
70 | smp_mb(); /* Must read state first. */ | ||
71 | atomic_inc(&stopmachine_thread_ack); | ||
72 | |||
73 | if (irqs_disabled) | ||
74 | local_irq_enable(); | ||
75 | if (prepared) | ||
76 | preempt_enable(); | ||
77 | 36 | ||
78 | return 0; | 37 | /* Like num_online_cpus(), but hotplug cpu uses us, so we need this. */ |
79 | } | 38 | static unsigned int num_threads; |
39 | static atomic_t thread_ack; | ||
40 | static struct completion finished; | ||
41 | static DEFINE_MUTEX(lock); | ||
80 | 42 | ||
81 | /* Change the thread state */ | 43 | static void set_state(enum stopmachine_state newstate) |
82 | static void stopmachine_set_state(enum stopmachine_state state) | ||
83 | { | 44 | { |
84 | atomic_set(&stopmachine_thread_ack, 0); | 45 | /* Reset ack counter. */ |
46 | atomic_set(&thread_ack, num_threads); | ||
85 | smp_wmb(); | 47 | smp_wmb(); |
86 | stopmachine_state = state; | 48 | state = newstate; |
87 | while (atomic_read(&stopmachine_thread_ack) != stopmachine_num_threads) | ||
88 | cpu_relax(); | ||
89 | } | 49 | } |
90 | 50 | ||
91 | static int stop_machine(void) | 51 | /* Last one to ack a state moves to the next state. */ |
52 | static void ack_state(void) | ||
92 | { | 53 | { |
93 | int i, ret = 0; | 54 | if (atomic_dec_and_test(&thread_ack)) { |
94 | 55 | /* If we're the last one to ack the EXIT, we're finished. */ | |
95 | atomic_set(&stopmachine_thread_ack, 0); | 56 | if (state == STOPMACHINE_EXIT) |
96 | stopmachine_num_threads = 0; | 57 | complete(&finished); |
97 | stopmachine_state = STOPMACHINE_WAIT; | 58 | else |
98 | 59 | set_state(state + 1); | |
99 | for_each_online_cpu(i) { | ||
100 | if (i == raw_smp_processor_id()) | ||
101 | continue; | ||
102 | ret = kernel_thread(stopmachine, (void *)(long)i,CLONE_KERNEL); | ||
103 | if (ret < 0) | ||
104 | break; | ||
105 | stopmachine_num_threads++; | ||
106 | } | ||
107 | |||
108 | /* Wait for them all to come to life. */ | ||
109 | while (atomic_read(&stopmachine_thread_ack) != stopmachine_num_threads) { | ||
110 | yield(); | ||
111 | cpu_relax(); | ||
112 | } | 60 | } |
61 | } | ||
113 | 62 | ||
114 | /* If some failed, kill them all. */ | 63 | /* This is the actual thread which stops the CPU. It exits by itself rather |
115 | if (ret < 0) { | 64 | * than waiting for kthread_stop(), because it's easier for hotplug CPU. */ |
116 | stopmachine_set_state(STOPMACHINE_EXIT); | 65 | static int stop_cpu(struct stop_machine_data *smdata) |
117 | return ret; | 66 | { |
118 | } | 67 | enum stopmachine_state curstate = STOPMACHINE_NONE; |
68 | int uninitialized_var(ret); | ||
119 | 69 | ||
120 | /* Now they are all started, make them hold the CPUs, ready. */ | 70 | /* Simple state machine */ |
121 | preempt_disable(); | 71 | do { |
122 | stopmachine_set_state(STOPMACHINE_PREPARE); | 72 | /* Chill out and ensure we re-read stopmachine_state. */ |
73 | cpu_relax(); | ||
74 | if (state != curstate) { | ||
75 | curstate = state; | ||
76 | switch (curstate) { | ||
77 | case STOPMACHINE_DISABLE_IRQ: | ||
78 | local_irq_disable(); | ||
79 | hard_irq_disable(); | ||
80 | break; | ||
81 | case STOPMACHINE_RUN: | ||
82 | /* |= allows error detection if functions on | ||
83 | * multiple CPUs. */ | ||
84 | smdata->fnret |= smdata->fn(smdata->data); | ||
85 | break; | ||
86 | default: | ||
87 | break; | ||
88 | } | ||
89 | ack_state(); | ||
90 | } | ||
91 | } while (curstate != STOPMACHINE_EXIT); | ||
123 | 92 | ||
124 | /* Make them disable irqs. */ | 93 | local_irq_enable(); |
125 | local_irq_disable(); | 94 | do_exit(0); |
126 | hard_irq_disable(); | 95 | } |
127 | stopmachine_set_state(STOPMACHINE_DISABLE_IRQ); | ||
128 | 96 | ||
97 | /* Callback for CPUs which aren't supposed to do anything. */ | ||
98 | static int chill(void *unused) | ||
99 | { | ||
129 | return 0; | 100 | return 0; |
130 | } | 101 | } |
131 | 102 | ||
132 | static void restart_machine(void) | 103 | int __stop_machine(int (*fn)(void *), void *data, const cpumask_t *cpus) |
133 | { | 104 | { |
134 | stopmachine_set_state(STOPMACHINE_EXIT); | 105 | int i, err; |
135 | local_irq_enable(); | 106 | struct stop_machine_data active, idle; |
136 | preempt_enable_no_resched(); | 107 | struct task_struct **threads; |
137 | } | 108 | |
109 | active.fn = fn; | ||
110 | active.data = data; | ||
111 | active.fnret = 0; | ||
112 | idle.fn = chill; | ||
113 | idle.data = NULL; | ||
114 | |||
115 | /* This could be too big for stack on large machines. */ | ||
116 | threads = kcalloc(NR_CPUS, sizeof(threads[0]), GFP_KERNEL); | ||
117 | if (!threads) | ||
118 | return -ENOMEM; | ||
119 | |||
120 | /* Set up initial state. */ | ||
121 | mutex_lock(&lock); | ||
122 | init_completion(&finished); | ||
123 | num_threads = num_online_cpus(); | ||
124 | set_state(STOPMACHINE_PREPARE); | ||
138 | 125 | ||
139 | struct stop_machine_data { | 126 | for_each_online_cpu(i) { |
140 | int (*fn)(void *); | 127 | struct stop_machine_data *smdata = &idle; |
141 | void *data; | 128 | struct sched_param param = { .sched_priority = MAX_RT_PRIO-1 }; |
142 | struct completion done; | ||
143 | }; | ||
144 | 129 | ||
145 | static int do_stop(void *_smdata) | 130 | if (!cpus) { |
146 | { | 131 | if (i == first_cpu(cpu_online_map)) |
147 | struct stop_machine_data *smdata = _smdata; | 132 | smdata = &active; |
148 | int ret; | 133 | } else { |
134 | if (cpu_isset(i, *cpus)) | ||
135 | smdata = &active; | ||
136 | } | ||
149 | 137 | ||
150 | ret = stop_machine(); | 138 | threads[i] = kthread_create((void *)stop_cpu, smdata, "kstop%u", |
151 | if (ret == 0) { | 139 | i); |
152 | ret = smdata->fn(smdata->data); | 140 | if (IS_ERR(threads[i])) { |
153 | restart_machine(); | 141 | err = PTR_ERR(threads[i]); |
154 | } | 142 | threads[i] = NULL; |
143 | goto kill_threads; | ||
144 | } | ||
155 | 145 | ||
156 | /* We're done: you can kthread_stop us now */ | 146 | /* Place it onto correct cpu. */ |
157 | complete(&smdata->done); | 147 | kthread_bind(threads[i], i); |
158 | 148 | ||
159 | /* Wait for kthread_stop */ | 149 | /* Make it highest prio. */ |
160 | set_current_state(TASK_INTERRUPTIBLE); | 150 | if (sched_setscheduler_nocheck(threads[i], SCHED_FIFO, ¶m)) |
161 | while (!kthread_should_stop()) { | 151 | BUG(); |
162 | schedule(); | ||
163 | set_current_state(TASK_INTERRUPTIBLE); | ||
164 | } | 152 | } |
165 | __set_current_state(TASK_RUNNING); | ||
166 | return ret; | ||
167 | } | ||
168 | 153 | ||
169 | struct task_struct *__stop_machine_run(int (*fn)(void *), void *data, | 154 | /* We've created all the threads. Wake them all: hold this CPU so one |
170 | unsigned int cpu) | 155 | * doesn't hit this CPU until we're ready. */ |
171 | { | 156 | get_cpu(); |
172 | static DEFINE_MUTEX(stopmachine_mutex); | 157 | for_each_online_cpu(i) |
173 | struct stop_machine_data smdata; | 158 | wake_up_process(threads[i]); |
174 | struct task_struct *p; | ||
175 | 159 | ||
176 | smdata.fn = fn; | 160 | /* This will release the thread on our CPU. */ |
177 | smdata.data = data; | 161 | put_cpu(); |
178 | init_completion(&smdata.done); | 162 | wait_for_completion(&finished); |
163 | mutex_unlock(&lock); | ||
179 | 164 | ||
180 | mutex_lock(&stopmachine_mutex); | 165 | kfree(threads); |
181 | 166 | ||
182 | /* If they don't care which CPU fn runs on, bind to any online one. */ | 167 | return active.fnret; |
183 | if (cpu == NR_CPUS) | ||
184 | cpu = raw_smp_processor_id(); | ||
185 | 168 | ||
186 | p = kthread_create(do_stop, &smdata, "kstopmachine"); | 169 | kill_threads: |
187 | if (!IS_ERR(p)) { | 170 | for_each_online_cpu(i) |
188 | struct sched_param param = { .sched_priority = MAX_RT_PRIO-1 }; | 171 | if (threads[i]) |
172 | kthread_stop(threads[i]); | ||
173 | mutex_unlock(&lock); | ||
189 | 174 | ||
190 | /* One high-prio thread per cpu. We'll do this one. */ | 175 | kfree(threads); |
191 | sched_setscheduler_nocheck(p, SCHED_FIFO, ¶m); | 176 | return err; |
192 | kthread_bind(p, cpu); | ||
193 | wake_up_process(p); | ||
194 | wait_for_completion(&smdata.done); | ||
195 | } | ||
196 | mutex_unlock(&stopmachine_mutex); | ||
197 | return p; | ||
198 | } | 177 | } |
199 | 178 | ||
200 | int stop_machine_run(int (*fn)(void *), void *data, unsigned int cpu) | 179 | int stop_machine(int (*fn)(void *), void *data, const cpumask_t *cpus) |
201 | { | 180 | { |
202 | struct task_struct *p; | ||
203 | int ret; | 181 | int ret; |
204 | 182 | ||
205 | /* No CPUs can come up or down during this. */ | 183 | /* No CPUs can come up or down during this. */ |
206 | get_online_cpus(); | 184 | get_online_cpus(); |
207 | p = __stop_machine_run(fn, data, cpu); | 185 | ret = __stop_machine(fn, data, cpus); |
208 | if (!IS_ERR(p)) | ||
209 | ret = kthread_stop(p); | ||
210 | else | ||
211 | ret = PTR_ERR(p); | ||
212 | put_online_cpus(); | 186 | put_online_cpus(); |
213 | 187 | ||
214 | return ret; | 188 | return ret; |
215 | } | 189 | } |
216 | EXPORT_SYMBOL_GPL(stop_machine_run); | 190 | EXPORT_SYMBOL_GPL(stop_machine); |