1 // SPDX-License-Identifier: GPL-2.0-only
4 * Copyright (C) 2016 ARM Limited
7 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
9 #include <linux/atomic.h>
10 #include <linux/completion.h>
11 #include <linux/cpu.h>
12 #include <linux/cpuidle.h>
13 #include <linux/cpu_pm.h>
14 #include <linux/kernel.h>
15 #include <linux/kthread.h>
16 #include <uapi/linux/sched/types.h>
17 #include <linux/module.h>
18 #include <linux/preempt.h>
19 #include <linux/psci.h>
20 #include <linux/slab.h>
21 #include <linux/tick.h>
22 #include <linux/topology.h>
24 #include <asm/cpuidle.h>
26 #include <uapi/linux/psci.h>
28 #define NUM_SUSPEND_CYCLE (10)
30 static unsigned int nb_available_cpus
;
31 static int tos_resident_cpu
= -1;
33 static atomic_t nb_active_threads
;
34 static struct completion suspend_threads_started
=
35 COMPLETION_INITIALIZER(suspend_threads_started
);
36 static struct completion suspend_threads_done
=
37 COMPLETION_INITIALIZER(suspend_threads_done
);
40 * We assume that PSCI operations are used if they are available. This is not
41 * necessarily true on arm64, since the decision is based on the
42 * "enable-method" property of each CPU in the DT, but given that there is no
43 * arch-specific way to check this, we assume that the DT is sensible.
45 static int psci_ops_check(void)
47 int migrate_type
= -1;
50 if (!(psci_ops
.cpu_off
&& psci_ops
.cpu_on
&& psci_ops
.cpu_suspend
)) {
51 pr_warn("Missing PSCI operations, aborting tests\n");
55 if (psci_ops
.migrate_info_type
)
56 migrate_type
= psci_ops
.migrate_info_type();
58 if (migrate_type
== PSCI_0_2_TOS_UP_MIGRATE
||
59 migrate_type
== PSCI_0_2_TOS_UP_NO_MIGRATE
) {
60 /* There is a UP Trusted OS, find on which core it resides. */
61 for_each_online_cpu(cpu
)
62 if (psci_tos_resident_on(cpu
)) {
63 tos_resident_cpu
= cpu
;
66 if (tos_resident_cpu
== -1)
67 pr_warn("UP Trusted OS resides on no online CPU\n");
74 * offlined_cpus is a temporary array but passing it as an argument avoids
75 * multiple allocations.
77 static unsigned int down_and_up_cpus(const struct cpumask
*cpus
,
78 struct cpumask
*offlined_cpus
)
83 cpumask_clear(offlined_cpus
);
85 /* Try to power down all CPUs in the mask. */
86 for_each_cpu(cpu
, cpus
) {
87 int ret
= cpu_down(cpu
);
90 * cpu_down() checks the number of online CPUs before the TOS
93 if (cpumask_weight(offlined_cpus
) + 1 == nb_available_cpus
) {
95 pr_err("Unexpected return code %d while trying "
96 "to power down last online CPU %d\n",
100 } else if (cpu
== tos_resident_cpu
) {
102 pr_err("Unexpected return code %d while trying "
103 "to power down TOS resident CPU %d\n",
107 } else if (ret
!= 0) {
108 pr_err("Error occurred (%d) while trying "
109 "to power down CPU %d\n", ret
, cpu
);
114 cpumask_set_cpu(cpu
, offlined_cpus
);
117 /* Try to power up all the CPUs that have been offlined. */
118 for_each_cpu(cpu
, offlined_cpus
) {
119 int ret
= cpu_up(cpu
);
122 pr_err("Error occurred (%d) while trying "
123 "to power up CPU %d\n", ret
, cpu
);
126 cpumask_clear_cpu(cpu
, offlined_cpus
);
131 * Something went bad at some point and some CPUs could not be turned
134 WARN_ON(!cpumask_empty(offlined_cpus
) ||
135 num_online_cpus() != nb_available_cpus
);
140 static void free_cpu_groups(int num
, cpumask_var_t
**pcpu_groups
)
143 cpumask_var_t
*cpu_groups
= *pcpu_groups
;
145 for (i
= 0; i
< num
; ++i
)
146 free_cpumask_var(cpu_groups
[i
]);
150 static int alloc_init_cpu_groups(cpumask_var_t
**pcpu_groups
)
153 cpumask_var_t tmp
, *cpu_groups
;
155 if (!alloc_cpumask_var(&tmp
, GFP_KERNEL
))
158 cpu_groups
= kcalloc(nb_available_cpus
, sizeof(cpu_groups
),
163 cpumask_copy(tmp
, cpu_online_mask
);
165 while (!cpumask_empty(tmp
)) {
166 const struct cpumask
*cpu_group
=
167 topology_core_cpumask(cpumask_any(tmp
));
169 if (!alloc_cpumask_var(&cpu_groups
[num_groups
], GFP_KERNEL
)) {
170 free_cpu_groups(num_groups
, &cpu_groups
);
173 cpumask_copy(cpu_groups
[num_groups
++], cpu_group
);
174 cpumask_andnot(tmp
, tmp
, cpu_group
);
177 free_cpumask_var(tmp
);
178 *pcpu_groups
= cpu_groups
;
183 static int hotplug_tests(void)
185 int i
, nb_cpu_group
, err
= -ENOMEM
;
186 cpumask_var_t offlined_cpus
, *cpu_groups
;
189 if (!alloc_cpumask_var(&offlined_cpus
, GFP_KERNEL
))
192 nb_cpu_group
= alloc_init_cpu_groups(&cpu_groups
);
193 if (nb_cpu_group
< 0)
195 page_buf
= (char *)__get_free_page(GFP_KERNEL
);
197 goto out_free_cpu_groups
;
201 * Of course the last CPU cannot be powered down and cpu_down() should
204 pr_info("Trying to turn off and on again all CPUs\n");
205 err
+= down_and_up_cpus(cpu_online_mask
, offlined_cpus
);
208 * Take down CPUs by cpu group this time. When the last CPU is turned
209 * off, the cpu group itself should shut down.
211 for (i
= 0; i
< nb_cpu_group
; ++i
) {
212 ssize_t len
= cpumap_print_to_pagebuf(true, page_buf
,
214 /* Remove trailing newline. */
215 page_buf
[len
- 1] = '\0';
216 pr_info("Trying to turn off and on again group %d (CPUs %s)\n",
218 err
+= down_and_up_cpus(cpu_groups
[i
], offlined_cpus
);
221 free_page((unsigned long)page_buf
);
223 free_cpu_groups(nb_cpu_group
, &cpu_groups
);
225 free_cpumask_var(offlined_cpus
);
229 static void dummy_callback(struct timer_list
*unused
) {}
231 static int suspend_cpu(struct cpuidle_device
*dev
,
232 struct cpuidle_driver
*drv
, int index
)
234 struct cpuidle_state
*state
= &drv
->states
[index
];
235 bool broadcast
= state
->flags
& CPUIDLE_FLAG_TIMER_STOP
;
238 arch_cpu_idle_enter();
242 * The local timer will be shut down, we need to enter tick
245 ret
= tick_broadcast_enter();
248 * In the absence of hardware broadcast mechanism,
249 * this CPU might be used to broadcast wakeups, which
250 * may be why entering tick broadcast has failed.
251 * There is little the kernel can do to work around
252 * that, so enter WFI instead (idle state 0).
260 ret
= state
->enter(dev
, drv
, index
);
263 tick_broadcast_exit();
266 arch_cpu_idle_exit();
271 static int suspend_test_thread(void *arg
)
274 int i
, nb_suspend
= 0, nb_shallow_sleep
= 0, nb_err
= 0;
275 struct sched_param sched_priority
= { .sched_priority
= MAX_RT_PRIO
-1 };
276 struct cpuidle_device
*dev
;
277 struct cpuidle_driver
*drv
;
278 /* No need for an actual callback, we just want to wake up the CPU. */
279 struct timer_list wakeup_timer
;
281 /* Wait for the main thread to give the start signal. */
282 wait_for_completion(&suspend_threads_started
);
284 /* Set maximum priority to preempt all other threads on this CPU. */
285 if (sched_setscheduler_nocheck(current
, SCHED_FIFO
, &sched_priority
))
286 pr_warn("Failed to set suspend thread scheduler on CPU %d\n",
289 dev
= this_cpu_read(cpuidle_devices
);
290 drv
= cpuidle_get_cpu_driver(dev
);
292 pr_info("CPU %d entering suspend cycles, states 1 through %d\n",
293 cpu
, drv
->state_count
- 1);
295 timer_setup_on_stack(&wakeup_timer
, dummy_callback
, 0);
296 for (i
= 0; i
< NUM_SUSPEND_CYCLE
; ++i
) {
299 * Test all possible states, except 0 (which is usually WFI and
302 for (index
= 1; index
< drv
->state_count
; ++index
) {
304 struct cpuidle_state
*state
= &drv
->states
[index
];
307 * Set the timer to wake this CPU up in some time (which
308 * should be largely sufficient for entering suspend).
309 * If the local tick is disabled when entering suspend,
310 * suspend_cpu() takes care of switching to a broadcast
311 * tick, so the timer will still wake us up.
313 mod_timer(&wakeup_timer
, jiffies
+
314 usecs_to_jiffies(state
->target_residency
));
316 /* IRQs must be disabled during suspend operations. */
319 ret
= suspend_cpu(dev
, drv
, index
);
322 * We have woken up. Re-enable IRQs to handle any
323 * pending interrupt, do not wait until the end of the
330 } else if (ret
>= 0) {
331 /* We did not enter the expected state. */
334 pr_err("Failed to suspend CPU %d: error %d "
335 "(requested state %d, cycle %d)\n",
343 * Disable the timer to make sure that the timer will not trigger
346 del_timer(&wakeup_timer
);
347 destroy_timer_on_stack(&wakeup_timer
);
349 if (atomic_dec_return_relaxed(&nb_active_threads
) == 0)
350 complete(&suspend_threads_done
);
352 /* Give up on RT scheduling and wait for termination. */
353 sched_priority
.sched_priority
= 0;
354 if (sched_setscheduler_nocheck(current
, SCHED_NORMAL
, &sched_priority
))
355 pr_warn("Failed to set suspend thread scheduler on CPU %d\n",
358 /* Needs to be set first to avoid missing a wakeup. */
359 set_current_state(TASK_INTERRUPTIBLE
);
360 if (kthread_should_park())
365 pr_info("CPU %d suspend test results: success %d, shallow states %d, errors %d\n",
366 cpu
, nb_suspend
, nb_shallow_sleep
, nb_err
);
373 static int suspend_tests(void)
376 struct task_struct
**threads
;
379 threads
= kmalloc_array(nb_available_cpus
, sizeof(*threads
),
385 * Stop cpuidle to prevent the idle tasks from entering a deep sleep
386 * mode, as it might interfere with the suspend threads on other CPUs.
387 * This does not prevent the suspend threads from using cpuidle (only
388 * the idle tasks check this status). Take the idle lock so that
389 * the cpuidle driver and device look-up can be carried out safely.
391 cpuidle_pause_and_lock();
393 for_each_online_cpu(cpu
) {
394 struct task_struct
*thread
;
395 /* Check that cpuidle is available on that CPU. */
396 struct cpuidle_device
*dev
= per_cpu(cpuidle_devices
, cpu
);
397 struct cpuidle_driver
*drv
= cpuidle_get_cpu_driver(dev
);
400 pr_warn("cpuidle not available on CPU %d, ignoring\n",
405 thread
= kthread_create_on_cpu(suspend_test_thread
,
406 (void *)(long)cpu
, cpu
,
407 "psci_suspend_test");
409 pr_err("Failed to create kthread on CPU %d\n", cpu
);
411 threads
[nb_threads
++] = thread
;
414 if (nb_threads
< 1) {
419 atomic_set(&nb_active_threads
, nb_threads
);
422 * Wake up the suspend threads. To avoid the main thread being preempted
423 * before all the threads have been unparked, the suspend threads will
424 * wait for the completion of suspend_threads_started.
426 for (i
= 0; i
< nb_threads
; ++i
)
427 wake_up_process(threads
[i
]);
428 complete_all(&suspend_threads_started
);
430 wait_for_completion(&suspend_threads_done
);
433 /* Stop and destroy all threads, get return status. */
434 for (i
= 0; i
< nb_threads
; ++i
) {
435 err
+= kthread_park(threads
[i
]);
436 err
+= kthread_stop(threads
[i
]);
439 cpuidle_resume_and_unlock();
444 static int __init
psci_checker(void)
449 * Since we're in an initcall, we assume that all the CPUs that all
450 * CPUs that can be onlined have been onlined.
452 * The tests assume that hotplug is enabled but nobody else is using it,
453 * otherwise the results will be unpredictable. However, since there
454 * is no userspace yet in initcalls, that should be fine, as long as
455 * no torture test is running at the same time (see Kconfig).
457 nb_available_cpus
= num_online_cpus();
459 /* Check PSCI operations are set up and working. */
460 ret
= psci_ops_check();
464 pr_info("PSCI checker started using %u CPUs\n", nb_available_cpus
);
466 pr_info("Starting hotplug tests\n");
467 ret
= hotplug_tests();
469 pr_info("Hotplug tests passed OK\n");
471 pr_err("%d error(s) encountered in hotplug tests\n", ret
);
473 pr_err("Out of memory\n");
477 pr_info("Starting suspend tests (%d cycles per state)\n",
479 ret
= suspend_tests();
481 pr_info("Suspend tests passed OK\n");
483 pr_err("%d error(s) encountered in suspend tests\n", ret
);
487 pr_err("Out of memory\n");
490 pr_warn("Could not start suspend tests on any CPU\n");
495 pr_info("PSCI checker completed\n");
496 return ret
< 0 ? ret
: 0;
498 late_initcall(psci_checker
);