Commit | Line | Data |
---|---|---|
1c33be57 NP |
1 | /* |
2 | * arch/arm/common/bL_switcher.c -- big.LITTLE cluster switcher core driver | |
3 | * | |
4 | * Created by: Nicolas Pitre, March 2012 | |
5 | * Copyright: (C) 2012-2013 Linaro Limited | |
6 | * | |
7 | * This program is free software; you can redistribute it and/or modify | |
8 | * it under the terms of the GNU General Public License version 2 as | |
9 | * published by the Free Software Foundation. | |
10 | */ | |
11 | ||
0577fee2 | 12 | #include <linux/atomic.h> |
1c33be57 NP |
13 | #include <linux/init.h> |
14 | #include <linux/kernel.h> | |
15 | #include <linux/module.h> | |
16 | #include <linux/sched.h> | |
17 | #include <linux/interrupt.h> | |
18 | #include <linux/cpu_pm.h> | |
71ce1dee | 19 | #include <linux/cpu.h> |
3f09d479 | 20 | #include <linux/cpumask.h> |
71ce1dee NP |
21 | #include <linux/kthread.h> |
22 | #include <linux/wait.h> | |
1bfbddb6 | 23 | #include <linux/time.h> |
3f09d479 LP |
24 | #include <linux/clockchips.h> |
25 | #include <linux/hrtimer.h> | |
26 | #include <linux/tick.h> | |
491990e2 | 27 | #include <linux/notifier.h> |
1c33be57 | 28 | #include <linux/mm.h> |
c0f43751 | 29 | #include <linux/mutex.h> |
b09bbe5b | 30 | #include <linux/smp.h> |
0577fee2 | 31 | #include <linux/spinlock.h> |
1c33be57 | 32 | #include <linux/string.h> |
6b7437ae | 33 | #include <linux/sysfs.h> |
1c33be57 | 34 | #include <linux/irqchip/arm-gic.h> |
c4821c05 | 35 | #include <linux/moduleparam.h> |
1c33be57 NP |
36 | |
37 | #include <asm/smp_plat.h> | |
1bfbddb6 | 38 | #include <asm/cputype.h> |
1c33be57 NP |
39 | #include <asm/suspend.h> |
40 | #include <asm/mcpm.h> | |
41 | #include <asm/bL_switcher.h> | |
42 | ||
1bfbddb6 DM |
43 | #define CREATE_TRACE_POINTS |
44 | #include <trace/events/power_cpu_migrate.h> | |
45 | ||
1c33be57 NP |
46 | |
47 | /* | |
48 | * Use our own MPIDR accessors as the generic ones in asm/cputype.h have | |
49 | * __attribute_const__ and we don't want the compiler to assume any | |
50 | * constness here as the value _does_ change along some code paths. | |
51 | */ | |
52 | ||
53 | static int read_mpidr(void) | |
54 | { | |
55 | unsigned int id; | |
56 | asm volatile ("mrc p15, 0, %0, c0, c0, 5" : "=r" (id)); | |
57 | return id & MPIDR_HWID_BITMASK; | |
58 | } | |
59 | ||
1bfbddb6 DM |
60 | /* |
61 | * Get a global nanosecond time stamp for tracing. | |
62 | */ | |
63 | static s64 get_ns(void) | |
64 | { | |
65 | struct timespec ts; | |
66 | getnstimeofday(&ts); | |
67 | return timespec_to_ns(&ts); | |
68 | } | |
69 | ||
1c33be57 NP |
70 | /* |
71 | * bL switcher core code. | |
72 | */ | |
73 | ||
108a9640 | 74 | static void bL_do_switch(void *_arg) |
1c33be57 | 75 | { |
38c35d4f | 76 | unsigned ib_mpidr, ib_cpu, ib_cluster; |
108a9640 | 77 | long volatile handshake, **handshake_ptr = _arg; |
1c33be57 | 78 | |
1c33be57 NP |
79 | pr_debug("%s\n", __func__); |
80 | ||
38c35d4f NP |
81 | ib_mpidr = cpu_logical_map(smp_processor_id()); |
82 | ib_cpu = MPIDR_AFFINITY_LEVEL(ib_mpidr, 0); | |
83 | ib_cluster = MPIDR_AFFINITY_LEVEL(ib_mpidr, 1); | |
1c33be57 | 84 | |
108a9640 NP |
85 | /* Advertise our handshake location */ |
86 | if (handshake_ptr) { | |
87 | handshake = 0; | |
88 | *handshake_ptr = &handshake; | |
89 | } else | |
90 | handshake = -1; | |
91 | ||
1c33be57 NP |
92 | /* |
93 | * Our state has been saved at this point. Let's release our | |
94 | * inbound CPU. | |
95 | */ | |
38c35d4f | 96 | mcpm_set_entry_vector(ib_cpu, ib_cluster, cpu_resume); |
1c33be57 NP |
97 | sev(); |
98 | ||
99 | /* | |
100 | * From this point, we must assume that our counterpart CPU might | |
101 | * have taken over in its parallel world already, as if execution | |
102 | * just returned from cpu_suspend(). It is therefore important to | |
103 | * be very careful not to make any change the other guy is not | |
104 | * expecting. This is why we need stack isolation. | |
105 | * | |
106 | * Fancy under cover tasks could be performed here. For now | |
107 | * we have none. | |
108 | */ | |
109 | ||
108a9640 NP |
110 | /* |
111 | * Let's wait until our inbound is alive. | |
112 | */ | |
113 | while (!handshake) { | |
114 | wfe(); | |
115 | smp_mb(); | |
116 | } | |
117 | ||
1c33be57 NP |
118 | /* Let's put ourself down. */ |
119 | mcpm_cpu_power_down(); | |
120 | ||
121 | /* should never get here */ | |
122 | BUG(); | |
123 | } | |
124 | ||
125 | /* | |
c052de26 NP |
126 | * Stack isolation. To ensure 'current' remains valid, we just use another |
127 | * piece of our thread's stack space which should be fairly lightly used. | |
128 | * The selected area starts just above the thread_info structure located | |
129 | * at the very bottom of the stack, aligned to a cache line, and indexed | |
130 | * with the cluster number. | |
1c33be57 | 131 | */ |
c052de26 | 132 | #define STACK_SIZE 512 |
1c33be57 NP |
133 | extern void call_with_stack(void (*fn)(void *), void *arg, void *sp); |
134 | static int bL_switchpoint(unsigned long _arg) | |
135 | { | |
136 | unsigned int mpidr = read_mpidr(); | |
1c33be57 | 137 | unsigned int clusterid = MPIDR_AFFINITY_LEVEL(mpidr, 1); |
c052de26 | 138 | void *stack = current_thread_info() + 1; |
1c33be57 | 139 | stack = PTR_ALIGN(stack, L1_CACHE_BYTES); |
c052de26 | 140 | stack += clusterid * STACK_SIZE + STACK_SIZE; |
1c33be57 NP |
141 | call_with_stack(bL_do_switch, (void *)_arg, stack); |
142 | BUG(); | |
143 | } | |
144 | ||
145 | /* | |
146 | * Generic switcher interface | |
147 | */ | |
148 | ||
ed96762e | 149 | static unsigned int bL_gic_id[MAX_CPUS_PER_CLUSTER][MAX_NR_CLUSTERS]; |
38c35d4f | 150 | static int bL_switcher_cpu_pairing[NR_CPUS]; |
ed96762e | 151 | |
1c33be57 NP |
152 | /* |
153 | * bL_switch_to - Switch to a specific cluster for the current CPU | |
154 | * @new_cluster_id: the ID of the cluster to switch to. | |
155 | * | |
156 | * This function must be called on the CPU to be switched. | |
157 | * Returns 0 on success, else a negative status code. | |
158 | */ | |
159 | static int bL_switch_to(unsigned int new_cluster_id) | |
160 | { | |
38c35d4f NP |
161 | unsigned int mpidr, this_cpu, that_cpu; |
162 | unsigned int ob_mpidr, ob_cpu, ob_cluster, ib_mpidr, ib_cpu, ib_cluster; | |
6137eba6 | 163 | struct completion inbound_alive; |
3f09d479 LP |
164 | struct tick_device *tdev; |
165 | enum clock_event_mode tdev_mode; | |
108a9640 | 166 | long volatile *handshake_ptr; |
6137eba6 | 167 | int ipi_nr, ret; |
1c33be57 | 168 | |
38c35d4f NP |
169 | this_cpu = smp_processor_id(); |
170 | ob_mpidr = read_mpidr(); | |
171 | ob_cpu = MPIDR_AFFINITY_LEVEL(ob_mpidr, 0); | |
172 | ob_cluster = MPIDR_AFFINITY_LEVEL(ob_mpidr, 1); | |
173 | BUG_ON(cpu_logical_map(this_cpu) != ob_mpidr); | |
1c33be57 | 174 | |
38c35d4f | 175 | if (new_cluster_id == ob_cluster) |
1c33be57 NP |
176 | return 0; |
177 | ||
38c35d4f NP |
178 | that_cpu = bL_switcher_cpu_pairing[this_cpu]; |
179 | ib_mpidr = cpu_logical_map(that_cpu); | |
180 | ib_cpu = MPIDR_AFFINITY_LEVEL(ib_mpidr, 0); | |
181 | ib_cluster = MPIDR_AFFINITY_LEVEL(ib_mpidr, 1); | |
182 | ||
183 | pr_debug("before switch: CPU %d MPIDR %#x -> %#x\n", | |
184 | this_cpu, ob_mpidr, ib_mpidr); | |
1c33be57 | 185 | |
6137eba6 NP |
186 | this_cpu = smp_processor_id(); |
187 | ||
1c33be57 | 188 | /* Close the gate for our entry vectors */ |
38c35d4f NP |
189 | mcpm_set_entry_vector(ob_cpu, ob_cluster, NULL); |
190 | mcpm_set_entry_vector(ib_cpu, ib_cluster, NULL); | |
1c33be57 | 191 | |
6137eba6 NP |
192 | /* Install our "inbound alive" notifier. */ |
193 | init_completion(&inbound_alive); | |
194 | ipi_nr = register_ipi_completion(&inbound_alive, this_cpu); | |
195 | ipi_nr |= ((1 << 16) << bL_gic_id[ob_cpu][ob_cluster]); | |
196 | mcpm_set_early_poke(ib_cpu, ib_cluster, gic_get_sgir_physaddr(), ipi_nr); | |
197 | ||
1c33be57 NP |
198 | /* |
199 | * Let's wake up the inbound CPU now in case it requires some delay | |
200 | * to come online, but leave it gated in our entry vector code. | |
201 | */ | |
38c35d4f | 202 | ret = mcpm_cpu_power_up(ib_cpu, ib_cluster); |
1c33be57 NP |
203 | if (ret) { |
204 | pr_err("%s: mcpm_cpu_power_up() returned %d\n", __func__, ret); | |
205 | return ret; | |
206 | } | |
207 | ||
6137eba6 NP |
208 | /* |
209 | * Raise a SGI on the inbound CPU to make sure it doesn't stall | |
210 | * in a possible WFI, such as in bL_power_down(). | |
211 | */ | |
212 | gic_send_sgi(bL_gic_id[ib_cpu][ib_cluster], 0); | |
213 | ||
214 | /* | |
215 | * Wait for the inbound to come up. This allows for other | |
216 | * tasks to be scheduled in the mean time. | |
217 | */ | |
218 | wait_for_completion(&inbound_alive); | |
219 | mcpm_set_early_poke(ib_cpu, ib_cluster, 0, 0); | |
220 | ||
1c33be57 NP |
221 | /* |
222 | * From this point we are entering the switch critical zone | |
223 | * and can't take any interrupts anymore. | |
224 | */ | |
225 | local_irq_disable(); | |
226 | local_fiq_disable(); | |
1bfbddb6 | 227 | trace_cpu_migrate_begin(get_ns(), ob_mpidr); |
1c33be57 | 228 | |
1c33be57 | 229 | /* redirect GIC's SGIs to our counterpart */ |
38c35d4f | 230 | gic_migrate_target(bL_gic_id[ib_cpu][ib_cluster]); |
1c33be57 | 231 | |
3f09d479 LP |
232 | tdev = tick_get_device(this_cpu); |
233 | if (tdev && !cpumask_equal(tdev->evtdev->cpumask, cpumask_of(this_cpu))) | |
234 | tdev = NULL; | |
235 | if (tdev) { | |
236 | tdev_mode = tdev->evtdev->mode; | |
237 | clockevents_set_mode(tdev->evtdev, CLOCK_EVT_MODE_SHUTDOWN); | |
238 | } | |
239 | ||
1c33be57 NP |
240 | ret = cpu_pm_enter(); |
241 | ||
242 | /* we can not tolerate errors at this point */ | |
243 | if (ret) | |
244 | panic("%s: cpu_pm_enter() returned %d\n", __func__, ret); | |
245 | ||
38c35d4f NP |
246 | /* Swap the physical CPUs in the logical map for this logical CPU. */ |
247 | cpu_logical_map(this_cpu) = ib_mpidr; | |
248 | cpu_logical_map(that_cpu) = ob_mpidr; | |
1c33be57 NP |
249 | |
250 | /* Let's do the actual CPU switch. */ | |
108a9640 | 251 | ret = cpu_suspend((unsigned long)&handshake_ptr, bL_switchpoint); |
1c33be57 NP |
252 | if (ret > 0) |
253 | panic("%s: cpu_suspend() returned %d\n", __func__, ret); | |
254 | ||
255 | /* We are executing on the inbound CPU at this point */ | |
256 | mpidr = read_mpidr(); | |
38c35d4f NP |
257 | pr_debug("after switch: CPU %d MPIDR %#x\n", this_cpu, mpidr); |
258 | BUG_ON(mpidr != ib_mpidr); | |
1c33be57 NP |
259 | |
260 | mcpm_cpu_powered_up(); | |
261 | ||
262 | ret = cpu_pm_exit(); | |
263 | ||
3f09d479 LP |
264 | if (tdev) { |
265 | clockevents_set_mode(tdev->evtdev, tdev_mode); | |
266 | clockevents_program_event(tdev->evtdev, | |
267 | tdev->evtdev->next_event, 1); | |
268 | } | |
269 | ||
1bfbddb6 | 270 | trace_cpu_migrate_finish(get_ns(), ib_mpidr); |
1c33be57 NP |
271 | local_fiq_enable(); |
272 | local_irq_enable(); | |
273 | ||
108a9640 NP |
274 | *handshake_ptr = 1; |
275 | dsb_sev(); | |
276 | ||
1c33be57 NP |
277 | if (ret) |
278 | pr_err("%s exiting with error %d\n", __func__, ret); | |
279 | return ret; | |
280 | } | |
281 | ||
71ce1dee | 282 | struct bL_thread { |
0577fee2 | 283 | spinlock_t lock; |
71ce1dee NP |
284 | struct task_struct *task; |
285 | wait_queue_head_t wq; | |
286 | int wanted_cluster; | |
6b7437ae | 287 | struct completion started; |
0577fee2 DM |
288 | bL_switch_completion_handler completer; |
289 | void *completer_cookie; | |
1c33be57 NP |
290 | }; |
291 | ||
71ce1dee NP |
292 | static struct bL_thread bL_threads[NR_CPUS]; |
293 | ||
294 | static int bL_switcher_thread(void *arg) | |
295 | { | |
296 | struct bL_thread *t = arg; | |
297 | struct sched_param param = { .sched_priority = 1 }; | |
298 | int cluster; | |
0577fee2 DM |
299 | bL_switch_completion_handler completer; |
300 | void *completer_cookie; | |
71ce1dee NP |
301 | |
302 | sched_setscheduler_nocheck(current, SCHED_FIFO, ¶m); | |
6b7437ae | 303 | complete(&t->started); |
71ce1dee NP |
304 | |
305 | do { | |
306 | if (signal_pending(current)) | |
307 | flush_signals(current); | |
308 | wait_event_interruptible(t->wq, | |
309 | t->wanted_cluster != -1 || | |
310 | kthread_should_stop()); | |
0577fee2 DM |
311 | |
312 | spin_lock(&t->lock); | |
313 | cluster = t->wanted_cluster; | |
314 | completer = t->completer; | |
315 | completer_cookie = t->completer_cookie; | |
316 | t->wanted_cluster = -1; | |
317 | t->completer = NULL; | |
318 | spin_unlock(&t->lock); | |
319 | ||
320 | if (cluster != -1) { | |
71ce1dee | 321 | bL_switch_to(cluster); |
0577fee2 DM |
322 | |
323 | if (completer) | |
324 | completer(completer_cookie); | |
325 | } | |
71ce1dee NP |
326 | } while (!kthread_should_stop()); |
327 | ||
328 | return 0; | |
329 | } | |
330 | ||
6b7437ae | 331 | static struct task_struct *bL_switcher_thread_create(int cpu, void *arg) |
1c33be57 | 332 | { |
71ce1dee NP |
333 | struct task_struct *task; |
334 | ||
335 | task = kthread_create_on_node(bL_switcher_thread, arg, | |
336 | cpu_to_node(cpu), "kswitcher_%d", cpu); | |
337 | if (!IS_ERR(task)) { | |
338 | kthread_bind(task, cpu); | |
339 | wake_up_process(task); | |
340 | } else | |
341 | pr_err("%s failed for CPU %d\n", __func__, cpu); | |
342 | return task; | |
1c33be57 NP |
343 | } |
344 | ||
345 | /* | |
0577fee2 DM |
346 | * bL_switch_request_cb - Switch to a specific cluster for the given CPU, |
347 | * with completion notification via a callback | |
1c33be57 NP |
348 | * |
349 | * @cpu: the CPU to switch | |
350 | * @new_cluster_id: the ID of the cluster to switch to. | |
0577fee2 DM |
351 | * @completer: switch completion callback. if non-NULL, |
352 | * @completer(@completer_cookie) will be called on completion of | |
353 | * the switch, in non-atomic context. | |
354 | * @completer_cookie: opaque context argument for @completer. | |
1c33be57 | 355 | * |
71ce1dee NP |
356 | * This function causes a cluster switch on the given CPU by waking up |
357 | * the appropriate switcher thread. This function may or may not return | |
358 | * before the switch has occurred. | |
0577fee2 DM |
359 | * |
360 | * If a @completer callback function is supplied, it will be called when | |
361 | * the switch is complete. This can be used to determine asynchronously | |
362 | * when the switch is complete, regardless of when bL_switch_request() | |
363 | * returns. When @completer is supplied, no new switch request is permitted | |
364 | * for the affected CPU until after the switch is complete, and @completer | |
365 | * has returned. | |
1c33be57 | 366 | */ |
0577fee2 DM |
367 | int bL_switch_request_cb(unsigned int cpu, unsigned int new_cluster_id, |
368 | bL_switch_completion_handler completer, | |
369 | void *completer_cookie) | |
1c33be57 | 370 | { |
71ce1dee | 371 | struct bL_thread *t; |
1c33be57 | 372 | |
71ce1dee NP |
373 | if (cpu >= ARRAY_SIZE(bL_threads)) { |
374 | pr_err("%s: cpu %d out of bounds\n", __func__, cpu); | |
375 | return -EINVAL; | |
1c33be57 | 376 | } |
1c33be57 | 377 | |
71ce1dee | 378 | t = &bL_threads[cpu]; |
0577fee2 | 379 | |
71ce1dee NP |
380 | if (IS_ERR(t->task)) |
381 | return PTR_ERR(t->task); | |
382 | if (!t->task) | |
383 | return -ESRCH; | |
384 | ||
0577fee2 DM |
385 | spin_lock(&t->lock); |
386 | if (t->completer) { | |
387 | spin_unlock(&t->lock); | |
388 | return -EBUSY; | |
389 | } | |
390 | t->completer = completer; | |
391 | t->completer_cookie = completer_cookie; | |
71ce1dee | 392 | t->wanted_cluster = new_cluster_id; |
0577fee2 | 393 | spin_unlock(&t->lock); |
71ce1dee NP |
394 | wake_up(&t->wq); |
395 | return 0; | |
1c33be57 | 396 | } |
0577fee2 | 397 | EXPORT_SYMBOL_GPL(bL_switch_request_cb); |
71ce1dee | 398 | |
9797a0e9 NP |
399 | /* |
400 | * Activation and configuration code. | |
401 | */ | |
402 | ||
c0f43751 | 403 | static DEFINE_MUTEX(bL_switcher_activation_lock); |
491990e2 | 404 | static BLOCKING_NOTIFIER_HEAD(bL_activation_notifier); |
6b7437ae | 405 | static unsigned int bL_switcher_active; |
38c35d4f | 406 | static unsigned int bL_switcher_cpu_original_cluster[NR_CPUS]; |
9797a0e9 NP |
407 | static cpumask_t bL_switcher_removed_logical_cpus; |
408 | ||
491990e2 DM |
409 | int bL_switcher_register_notifier(struct notifier_block *nb) |
410 | { | |
411 | return blocking_notifier_chain_register(&bL_activation_notifier, nb); | |
412 | } | |
413 | EXPORT_SYMBOL_GPL(bL_switcher_register_notifier); | |
414 | ||
415 | int bL_switcher_unregister_notifier(struct notifier_block *nb) | |
416 | { | |
417 | return blocking_notifier_chain_unregister(&bL_activation_notifier, nb); | |
418 | } | |
419 | EXPORT_SYMBOL_GPL(bL_switcher_unregister_notifier); | |
420 | ||
421 | static int bL_activation_notify(unsigned long val) | |
422 | { | |
423 | int ret; | |
424 | ||
425 | ret = blocking_notifier_call_chain(&bL_activation_notifier, val, NULL); | |
426 | if (ret & NOTIFY_STOP_MASK) | |
427 | pr_err("%s: notifier chain failed with status 0x%x\n", | |
428 | __func__, ret); | |
429 | return notifier_to_errno(ret); | |
430 | } | |
431 | ||
6b7437ae | 432 | static void bL_switcher_restore_cpus(void) |
9797a0e9 NP |
433 | { |
434 | int i; | |
435 | ||
3f8517e7 NP |
436 | for_each_cpu(i, &bL_switcher_removed_logical_cpus) { |
437 | struct device *cpu_dev = get_cpu_device(i); | |
438 | int ret = device_online(cpu_dev); | |
439 | if (ret) | |
440 | dev_err(cpu_dev, "switcher: unable to restore CPU\n"); | |
441 | } | |
9797a0e9 NP |
442 | } |
443 | ||
6b7437ae | 444 | static int bL_switcher_halve_cpus(void) |
9797a0e9 | 445 | { |
38c35d4f NP |
446 | int i, j, cluster_0, gic_id, ret; |
447 | unsigned int cpu, cluster, mask; | |
448 | cpumask_t available_cpus; | |
9797a0e9 | 449 | |
38c35d4f NP |
450 | /* First pass to validate what we have */ |
451 | mask = 0; | |
9797a0e9 | 452 | for_each_online_cpu(i) { |
38c35d4f NP |
453 | cpu = MPIDR_AFFINITY_LEVEL(cpu_logical_map(i), 0); |
454 | cluster = MPIDR_AFFINITY_LEVEL(cpu_logical_map(i), 1); | |
9797a0e9 NP |
455 | if (cluster >= 2) { |
456 | pr_err("%s: only dual cluster systems are supported\n", __func__); | |
457 | return -EINVAL; | |
458 | } | |
38c35d4f NP |
459 | if (WARN_ON(cpu >= MAX_CPUS_PER_CLUSTER)) |
460 | return -EINVAL; | |
461 | mask |= (1 << cluster); | |
9797a0e9 | 462 | } |
38c35d4f NP |
463 | if (mask != 3) { |
464 | pr_err("%s: no CPU pairing possible\n", __func__); | |
9797a0e9 NP |
465 | return -EINVAL; |
466 | } | |
467 | ||
38c35d4f NP |
468 | /* |
469 | * Now let's do the pairing. We match each CPU with another CPU | |
470 | * from a different cluster. To get a uniform scheduling behavior | |
471 | * without fiddling with CPU topology and compute capacity data, | |
472 | * we'll use logical CPUs initially belonging to the same cluster. | |
473 | */ | |
474 | memset(bL_switcher_cpu_pairing, -1, sizeof(bL_switcher_cpu_pairing)); | |
475 | cpumask_copy(&available_cpus, cpu_online_mask); | |
476 | cluster_0 = -1; | |
477 | for_each_cpu(i, &available_cpus) { | |
478 | int match = -1; | |
479 | cluster = MPIDR_AFFINITY_LEVEL(cpu_logical_map(i), 1); | |
480 | if (cluster_0 == -1) | |
481 | cluster_0 = cluster; | |
482 | if (cluster != cluster_0) | |
483 | continue; | |
484 | cpumask_clear_cpu(i, &available_cpus); | |
485 | for_each_cpu(j, &available_cpus) { | |
486 | cluster = MPIDR_AFFINITY_LEVEL(cpu_logical_map(j), 1); | |
9797a0e9 | 487 | /* |
38c35d4f NP |
488 | * Let's remember the last match to create "odd" |
489 | * pairings on purpose in order for other code not | |
490 | * to assume any relation between physical and | |
491 | * logical CPU numbers. | |
9797a0e9 | 492 | */ |
38c35d4f NP |
493 | if (cluster != cluster_0) |
494 | match = j; | |
495 | } | |
496 | if (match != -1) { | |
497 | bL_switcher_cpu_pairing[i] = match; | |
498 | cpumask_clear_cpu(match, &available_cpus); | |
499 | pr_info("CPU%d paired with CPU%d\n", i, match); | |
500 | } | |
501 | } | |
502 | ||
503 | /* | |
504 | * Now we disable the unwanted CPUs i.e. everything that has no | |
505 | * pairing information (that includes the pairing counterparts). | |
506 | */ | |
507 | cpumask_clear(&bL_switcher_removed_logical_cpus); | |
508 | for_each_online_cpu(i) { | |
509 | cpu = MPIDR_AFFINITY_LEVEL(cpu_logical_map(i), 0); | |
510 | cluster = MPIDR_AFFINITY_LEVEL(cpu_logical_map(i), 1); | |
511 | ||
512 | /* Let's take note of the GIC ID for this CPU */ | |
513 | gic_id = gic_get_cpu_id(i); | |
514 | if (gic_id < 0) { | |
515 | pr_err("%s: bad GIC ID for CPU %d\n", __func__, i); | |
516 | bL_switcher_restore_cpus(); | |
517 | return -EINVAL; | |
518 | } | |
519 | bL_gic_id[cpu][cluster] = gic_id; | |
520 | pr_info("GIC ID for CPU %u cluster %u is %u\n", | |
521 | cpu, cluster, gic_id); | |
522 | ||
523 | if (bL_switcher_cpu_pairing[i] != -1) { | |
524 | bL_switcher_cpu_original_cluster[i] = cluster; | |
525 | continue; | |
9797a0e9 NP |
526 | } |
527 | ||
3f8517e7 | 528 | ret = device_offline(get_cpu_device(i)); |
9797a0e9 NP |
529 | if (ret) { |
530 | bL_switcher_restore_cpus(); | |
531 | return ret; | |
532 | } | |
533 | cpumask_set_cpu(i, &bL_switcher_removed_logical_cpus); | |
534 | } | |
535 | ||
536 | return 0; | |
537 | } | |
538 | ||
d08e2e09 DM |
539 | /* Determine the logical CPU a given physical CPU is grouped on. */ |
540 | int bL_switcher_get_logical_index(u32 mpidr) | |
541 | { | |
542 | int cpu; | |
543 | ||
544 | if (!bL_switcher_active) | |
545 | return -EUNATCH; | |
546 | ||
547 | mpidr &= MPIDR_HWID_BITMASK; | |
548 | for_each_online_cpu(cpu) { | |
549 | int pairing = bL_switcher_cpu_pairing[cpu]; | |
550 | if (pairing == -1) | |
551 | continue; | |
552 | if ((mpidr == cpu_logical_map(cpu)) || | |
553 | (mpidr == cpu_logical_map(pairing))) | |
554 | return cpu; | |
555 | } | |
556 | return -EINVAL; | |
557 | } | |
558 | ||
b09bbe5b DM |
559 | static void bL_switcher_trace_trigger_cpu(void *__always_unused info) |
560 | { | |
561 | trace_cpu_migrate_current(get_ns(), read_mpidr()); | |
562 | } | |
563 | ||
29064b88 | 564 | int bL_switcher_trace_trigger(void) |
b09bbe5b DM |
565 | { |
566 | int ret; | |
567 | ||
568 | preempt_disable(); | |
569 | ||
570 | bL_switcher_trace_trigger_cpu(NULL); | |
571 | ret = smp_call_function(bL_switcher_trace_trigger_cpu, NULL, true); | |
572 | ||
573 | preempt_enable(); | |
574 | ||
575 | return ret; | |
576 | } | |
29064b88 | 577 | EXPORT_SYMBOL_GPL(bL_switcher_trace_trigger); |
b09bbe5b | 578 | |
6b7437ae | 579 | static int bL_switcher_enable(void) |
71ce1dee | 580 | { |
9797a0e9 | 581 | int cpu, ret; |
71ce1dee | 582 | |
c0f43751 | 583 | mutex_lock(&bL_switcher_activation_lock); |
b0ced9d2 | 584 | lock_device_hotplug(); |
6b7437ae | 585 | if (bL_switcher_active) { |
b0ced9d2 | 586 | unlock_device_hotplug(); |
c0f43751 | 587 | mutex_unlock(&bL_switcher_activation_lock); |
6b7437ae | 588 | return 0; |
9797a0e9 NP |
589 | } |
590 | ||
6b7437ae NP |
591 | pr_info("big.LITTLE switcher initializing\n"); |
592 | ||
491990e2 DM |
593 | ret = bL_activation_notify(BL_NOTIFY_PRE_ENABLE); |
594 | if (ret) | |
595 | goto error; | |
596 | ||
9797a0e9 | 597 | ret = bL_switcher_halve_cpus(); |
491990e2 DM |
598 | if (ret) |
599 | goto error; | |
9797a0e9 | 600 | |
b09bbe5b DM |
601 | bL_switcher_trace_trigger(); |
602 | ||
71ce1dee NP |
603 | for_each_online_cpu(cpu) { |
604 | struct bL_thread *t = &bL_threads[cpu]; | |
0577fee2 | 605 | spin_lock_init(&t->lock); |
71ce1dee | 606 | init_waitqueue_head(&t->wq); |
6b7437ae | 607 | init_completion(&t->started); |
71ce1dee NP |
608 | t->wanted_cluster = -1; |
609 | t->task = bL_switcher_thread_create(cpu, t); | |
610 | } | |
6b7437ae NP |
611 | |
612 | bL_switcher_active = 1; | |
491990e2 | 613 | bL_activation_notify(BL_NOTIFY_POST_ENABLE); |
71ce1dee | 614 | pr_info("big.LITTLE switcher initialized\n"); |
491990e2 DM |
615 | goto out; |
616 | ||
617 | error: | |
618 | pr_warn("big.LITTLE switcher initialization failed\n"); | |
619 | bL_activation_notify(BL_NOTIFY_POST_DISABLE); | |
c0f43751 | 620 | |
491990e2 | 621 | out: |
b0ced9d2 | 622 | unlock_device_hotplug(); |
c0f43751 | 623 | mutex_unlock(&bL_switcher_activation_lock); |
491990e2 | 624 | return ret; |
71ce1dee NP |
625 | } |
626 | ||
6b7437ae NP |
627 | #ifdef CONFIG_SYSFS |
628 | ||
629 | static void bL_switcher_disable(void) | |
630 | { | |
38c35d4f | 631 | unsigned int cpu, cluster; |
6b7437ae NP |
632 | struct bL_thread *t; |
633 | struct task_struct *task; | |
634 | ||
c0f43751 | 635 | mutex_lock(&bL_switcher_activation_lock); |
b0ced9d2 | 636 | lock_device_hotplug(); |
491990e2 DM |
637 | |
638 | if (!bL_switcher_active) | |
639 | goto out; | |
640 | ||
641 | if (bL_activation_notify(BL_NOTIFY_PRE_DISABLE) != 0) { | |
642 | bL_activation_notify(BL_NOTIFY_POST_ENABLE); | |
643 | goto out; | |
6b7437ae | 644 | } |
491990e2 | 645 | |
6b7437ae NP |
646 | bL_switcher_active = 0; |
647 | ||
648 | /* | |
649 | * To deactivate the switcher, we must shut down the switcher | |
650 | * threads to prevent any other requests from being accepted. | |
651 | * Then, if the final cluster for given logical CPU is not the | |
652 | * same as the original one, we'll recreate a switcher thread | |
653 | * just for the purpose of switching the CPU back without any | |
654 | * possibility for interference from external requests. | |
655 | */ | |
656 | for_each_online_cpu(cpu) { | |
6b7437ae NP |
657 | t = &bL_threads[cpu]; |
658 | task = t->task; | |
659 | t->task = NULL; | |
660 | if (!task || IS_ERR(task)) | |
661 | continue; | |
662 | kthread_stop(task); | |
663 | /* no more switch may happen on this CPU at this point */ | |
664 | cluster = MPIDR_AFFINITY_LEVEL(cpu_logical_map(cpu), 1); | |
665 | if (cluster == bL_switcher_cpu_original_cluster[cpu]) | |
666 | continue; | |
667 | init_completion(&t->started); | |
668 | t->wanted_cluster = bL_switcher_cpu_original_cluster[cpu]; | |
669 | task = bL_switcher_thread_create(cpu, t); | |
670 | if (!IS_ERR(task)) { | |
671 | wait_for_completion(&t->started); | |
672 | kthread_stop(task); | |
673 | cluster = MPIDR_AFFINITY_LEVEL(cpu_logical_map(cpu), 1); | |
674 | if (cluster == bL_switcher_cpu_original_cluster[cpu]) | |
675 | continue; | |
676 | } | |
677 | /* If execution gets here, we're in trouble. */ | |
678 | pr_crit("%s: unable to restore original cluster for CPU %d\n", | |
679 | __func__, cpu); | |
38c35d4f NP |
680 | pr_crit("%s: CPU %d can't be restored\n", |
681 | __func__, bL_switcher_cpu_pairing[cpu]); | |
682 | cpumask_clear_cpu(bL_switcher_cpu_pairing[cpu], | |
683 | &bL_switcher_removed_logical_cpus); | |
6b7437ae NP |
684 | } |
685 | ||
686 | bL_switcher_restore_cpus(); | |
b09bbe5b DM |
687 | bL_switcher_trace_trigger(); |
688 | ||
491990e2 DM |
689 | bL_activation_notify(BL_NOTIFY_POST_DISABLE); |
690 | ||
691 | out: | |
b0ced9d2 | 692 | unlock_device_hotplug(); |
c0f43751 | 693 | mutex_unlock(&bL_switcher_activation_lock); |
6b7437ae NP |
694 | } |
695 | ||
696 | static ssize_t bL_switcher_active_show(struct kobject *kobj, | |
697 | struct kobj_attribute *attr, char *buf) | |
698 | { | |
699 | return sprintf(buf, "%u\n", bL_switcher_active); | |
700 | } | |
701 | ||
702 | static ssize_t bL_switcher_active_store(struct kobject *kobj, | |
703 | struct kobj_attribute *attr, const char *buf, size_t count) | |
704 | { | |
705 | int ret; | |
706 | ||
707 | switch (buf[0]) { | |
708 | case '0': | |
709 | bL_switcher_disable(); | |
710 | ret = 0; | |
711 | break; | |
712 | case '1': | |
713 | ret = bL_switcher_enable(); | |
714 | break; | |
715 | default: | |
716 | ret = -EINVAL; | |
717 | } | |
718 | ||
719 | return (ret >= 0) ? count : ret; | |
720 | } | |
721 | ||
b09bbe5b DM |
722 | static ssize_t bL_switcher_trace_trigger_store(struct kobject *kobj, |
723 | struct kobj_attribute *attr, const char *buf, size_t count) | |
724 | { | |
725 | int ret = bL_switcher_trace_trigger(); | |
726 | ||
727 | return ret ? ret : count; | |
728 | } | |
729 | ||
6b7437ae NP |
730 | static struct kobj_attribute bL_switcher_active_attr = |
731 | __ATTR(active, 0644, bL_switcher_active_show, bL_switcher_active_store); | |
732 | ||
b09bbe5b DM |
733 | static struct kobj_attribute bL_switcher_trace_trigger_attr = |
734 | __ATTR(trace_trigger, 0200, NULL, bL_switcher_trace_trigger_store); | |
735 | ||
6b7437ae NP |
736 | static struct attribute *bL_switcher_attrs[] = { |
737 | &bL_switcher_active_attr.attr, | |
b09bbe5b | 738 | &bL_switcher_trace_trigger_attr.attr, |
6b7437ae NP |
739 | NULL, |
740 | }; | |
741 | ||
742 | static struct attribute_group bL_switcher_attr_group = { | |
743 | .attrs = bL_switcher_attrs, | |
744 | }; | |
745 | ||
746 | static struct kobject *bL_switcher_kobj; | |
747 | ||
748 | static int __init bL_switcher_sysfs_init(void) | |
749 | { | |
750 | int ret; | |
751 | ||
752 | bL_switcher_kobj = kobject_create_and_add("bL_switcher", kernel_kobj); | |
753 | if (!bL_switcher_kobj) | |
754 | return -ENOMEM; | |
755 | ret = sysfs_create_group(bL_switcher_kobj, &bL_switcher_attr_group); | |
756 | if (ret) | |
757 | kobject_put(bL_switcher_kobj); | |
758 | return ret; | |
759 | } | |
760 | ||
761 | #endif /* CONFIG_SYSFS */ | |
762 | ||
c0f43751 DM |
763 | bool bL_switcher_get_enabled(void) |
764 | { | |
765 | mutex_lock(&bL_switcher_activation_lock); | |
766 | ||
767 | return bL_switcher_active; | |
768 | } | |
769 | EXPORT_SYMBOL_GPL(bL_switcher_get_enabled); | |
770 | ||
771 | void bL_switcher_put_enabled(void) | |
772 | { | |
773 | mutex_unlock(&bL_switcher_activation_lock); | |
774 | } | |
775 | EXPORT_SYMBOL_GPL(bL_switcher_put_enabled); | |
776 | ||
27261435 NP |
777 | /* |
778 | * Veto any CPU hotplug operation on those CPUs we've removed | |
779 | * while the switcher is active. | |
780 | * We're just not ready to deal with that given the trickery involved. | |
781 | */ | |
782 | static int bL_switcher_hotplug_callback(struct notifier_block *nfb, | |
783 | unsigned long action, void *hcpu) | |
784 | { | |
785 | if (bL_switcher_active) { | |
786 | int pairing = bL_switcher_cpu_pairing[(unsigned long)hcpu]; | |
787 | switch (action & 0xf) { | |
788 | case CPU_UP_PREPARE: | |
789 | case CPU_DOWN_PREPARE: | |
790 | if (pairing == -1) | |
791 | return NOTIFY_BAD; | |
792 | } | |
793 | } | |
794 | return NOTIFY_DONE; | |
795 | } | |
796 | ||
c4821c05 NP |
797 | static bool no_bL_switcher; |
798 | core_param(no_bL_switcher, no_bL_switcher, bool, 0644); | |
799 | ||
6b7437ae NP |
800 | static int __init bL_switcher_init(void) |
801 | { | |
802 | int ret; | |
803 | ||
4530e4b6 NP |
804 | if (!mcpm_is_available()) |
805 | return -ENODEV; | |
6b7437ae | 806 | |
27261435 NP |
807 | cpu_notifier(bL_switcher_hotplug_callback, 0); |
808 | ||
c4821c05 NP |
809 | if (!no_bL_switcher) { |
810 | ret = bL_switcher_enable(); | |
811 | if (ret) | |
812 | return ret; | |
813 | } | |
6b7437ae NP |
814 | |
815 | #ifdef CONFIG_SYSFS | |
816 | ret = bL_switcher_sysfs_init(); | |
817 | if (ret) | |
818 | pr_err("%s: unable to create sysfs entry\n", __func__); | |
819 | #endif | |
820 | ||
821 | return 0; | |
822 | } | |
823 | ||
71ce1dee | 824 | late_initcall(bL_switcher_init); |