| 1 |
// SPDX-License-Identifier: GPL-2.0-only |
| 2 |
/* |
| 3 |
* |
| 4 |
* Copyright (C) 2016 ARM Limited |
| 5 |
*/ |
| 6 |
|
| 7 |
#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt |
| 8 |
|
| 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> |
| 23 |
|
| 24 |
#include <asm/cpuidle.h> |
| 25 |
|
| 26 |
#include <uapi/linux/psci.h> |
| 27 |
|
| 28 |
#define NUM_SUSPEND_CYCLE (10) |
| 29 |
|
| 30 |
static unsigned int nb_available_cpus; |
| 31 |
static int tos_resident_cpu = -1; |
| 32 |
|
| 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); |
| 38 |
|
| 39 |
/* |
| 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. |
| 44 |
*/ |
| 45 |
static int psci_ops_check(void) |
| 46 |
{ |
| 47 |
int migrate_type = -1; |
| 48 |
int cpu; |
| 49 |
|
| 50 |
if (!(psci_ops.cpu_off && psci_ops.cpu_on && psci_ops.cpu_suspend)) { |
| 51 |
pr_warn("Missing PSCI operations, aborting tests\n"); |
| 52 |
return -EOPNOTSUPP; |
| 53 |
} |
| 54 |
|
| 55 |
if (psci_ops.migrate_info_type) |
| 56 |
migrate_type = psci_ops.migrate_info_type(); |
| 57 |
|
| 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; |
| 64 |
break; |
| 65 |
} |
| 66 |
if (tos_resident_cpu == -1) |
| 67 |
pr_warn("UP Trusted OS resides on no online CPU\n"); |
| 68 |
} |
| 69 |
|
| 70 |
return 0; |
| 71 |
} |
| 72 |
|
| 73 |
/* |
| 74 |
* offlined_cpus is a temporary array but passing it as an argument avoids |
| 75 |
* multiple allocations. |
| 76 |
*/ |
| 77 |
static unsigned int down_and_up_cpus(const struct cpumask *cpus, |
| 78 |
struct cpumask *offlined_cpus) |
| 79 |
{ |
| 80 |
int cpu; |
| 81 |
int err = 0; |
| 82 |
|
| 83 |
cpumask_clear(offlined_cpus); |
| 84 |
|
| 85 |
/* Try to power down all CPUs in the mask. */ |
| 86 |
for_each_cpu(cpu, cpus) { |
| 87 |
int ret = remove_cpu(cpu); |
| 88 |
|
| 89 |
/* |
| 90 |
* cpu_down() checks the number of online CPUs before the TOS |
| 91 |
* resident CPU. |
| 92 |
*/ |
| 93 |
if (cpumask_weight(offlined_cpus) + 1 == nb_available_cpus) { |
| 94 |
if (ret != -EBUSY) { |
| 95 |
pr_err("Unexpected return code %d while trying " |
| 96 |
"to power down last online CPU %d\n", |
| 97 |
ret, cpu); |
| 98 |
++err; |
| 99 |
} |
| 100 |
} else if (cpu == tos_resident_cpu) { |
| 101 |
if (ret != -EPERM) { |
| 102 |
pr_err("Unexpected return code %d while trying " |
| 103 |
"to power down TOS resident CPU %d\n", |
| 104 |
ret, cpu); |
| 105 |
++err; |
| 106 |
} |
| 107 |
} else if (ret != 0) { |
| 108 |
pr_err("Error occurred (%d) while trying " |
| 109 |
"to power down CPU %d\n", ret, cpu); |
| 110 |
++err; |
| 111 |
} |
| 112 |
|
| 113 |
if (ret == 0) |
| 114 |
cpumask_set_cpu(cpu, offlined_cpus); |
| 115 |
} |
| 116 |
|
| 117 |
/* Try to power up all the CPUs that have been offlined. */ |
| 118 |
for_each_cpu(cpu, offlined_cpus) { |
| 119 |
int ret = add_cpu(cpu); |
| 120 |
|
| 121 |
if (ret != 0) { |
| 122 |
pr_err("Error occurred (%d) while trying " |
| 123 |
"to power up CPU %d\n", ret, cpu); |
| 124 |
++err; |
| 125 |
} else { |
| 126 |
cpumask_clear_cpu(cpu, offlined_cpus); |
| 127 |
} |
| 128 |
} |
| 129 |
|
| 130 |
/* |
| 131 |
* Something went bad at some point and some CPUs could not be turned |
| 132 |
* back on. |
| 133 |
*/ |
| 134 |
WARN_ON(!cpumask_empty(offlined_cpus) || |
| 135 |
num_online_cpus() != nb_available_cpus); |
| 136 |
|
| 137 |
return err; |
| 138 |
} |
| 139 |
|
| 140 |
static void free_cpu_groups(int num, cpumask_var_t **pcpu_groups) |
| 141 |
{ |
| 142 |
int i; |
| 143 |
cpumask_var_t *cpu_groups = *pcpu_groups; |
| 144 |
|
| 145 |
for (i = 0; i < num; ++i) |
| 146 |
free_cpumask_var(cpu_groups[i]); |
| 147 |
kfree(cpu_groups); |
| 148 |
} |
| 149 |
|
| 150 |
static int alloc_init_cpu_groups(cpumask_var_t **pcpu_groups) |
| 151 |
{ |
| 152 |
int num_groups = 0; |
| 153 |
cpumask_var_t tmp, *cpu_groups; |
| 154 |
|
| 155 |
if (!alloc_cpumask_var(&tmp, GFP_KERNEL)) |
| 156 |
return -ENOMEM; |
| 157 |
|
| 158 |
cpu_groups = kzalloc_objs(*cpu_groups, nb_available_cpus); |
| 159 |
if (!cpu_groups) { |
| 160 |
free_cpumask_var(tmp); |
| 161 |
return -ENOMEM; |
| 162 |
} |
| 163 |
|
| 164 |
cpumask_copy(tmp, cpu_online_mask); |
| 165 |
|
| 166 |
while (!cpumask_empty(tmp)) { |
| 167 |
const struct cpumask *cpu_group = |
| 168 |
topology_core_cpumask(cpumask_any(tmp)); |
| 169 |
|
| 170 |
if (!alloc_cpumask_var(&cpu_groups[num_groups], GFP_KERNEL)) { |
| 171 |
free_cpumask_var(tmp); |
| 172 |
free_cpu_groups(num_groups, &cpu_groups); |
| 173 |
return -ENOMEM; |
| 174 |
} |
| 175 |
cpumask_copy(cpu_groups[num_groups++], cpu_group); |
| 176 |
cpumask_andnot(tmp, tmp, cpu_group); |
| 177 |
} |
| 178 |
|
| 179 |
free_cpumask_var(tmp); |
| 180 |
*pcpu_groups = cpu_groups; |
| 181 |
|
| 182 |
return num_groups; |
| 183 |
} |
| 184 |
|
| 185 |
static int hotplug_tests(void) |
| 186 |
{ |
| 187 |
int i, nb_cpu_group, err = -ENOMEM; |
| 188 |
cpumask_var_t offlined_cpus, *cpu_groups; |
| 189 |
char *page_buf; |
| 190 |
|
| 191 |
if (!alloc_cpumask_var(&offlined_cpus, GFP_KERNEL)) |
| 192 |
return err; |
| 193 |
|
| 194 |
nb_cpu_group = alloc_init_cpu_groups(&cpu_groups); |
| 195 |
if (nb_cpu_group < 0) |
| 196 |
goto out_free_cpus; |
| 197 |
page_buf = (char *)__get_free_page(GFP_KERNEL); |
| 198 |
if (!page_buf) |
| 199 |
goto out_free_cpu_groups; |
| 200 |
|
| 201 |
/* |
| 202 |
* Of course the last CPU cannot be powered down and cpu_down() should |
| 203 |
* refuse doing that. |
| 204 |
*/ |
| 205 |
pr_info("Trying to turn off and on again all CPUs\n"); |
| 206 |
err = down_and_up_cpus(cpu_online_mask, offlined_cpus); |
| 207 |
|
| 208 |
/* |
| 209 |
* Take down CPUs by cpu group this time. When the last CPU is turned |
| 210 |
* off, the cpu group itself should shut down. |
| 211 |
*/ |
| 212 |
for (i = 0; i < nb_cpu_group; ++i) { |
| 213 |
ssize_t len = cpumap_print_to_pagebuf(true, page_buf, |
| 214 |
cpu_groups[i]); |
| 215 |
/* Remove trailing newline. */ |
| 216 |
page_buf[len - 1] = '\0'; |
| 217 |
pr_info("Trying to turn off and on again group %d (CPUs %s)\n", |
| 218 |
i, page_buf); |
| 219 |
err += down_and_up_cpus(cpu_groups[i], offlined_cpus); |
| 220 |
} |
| 221 |
|
| 222 |
free_page((unsigned long)page_buf); |
| 223 |
out_free_cpu_groups: |
| 224 |
free_cpu_groups(nb_cpu_group, &cpu_groups); |
| 225 |
out_free_cpus: |
| 226 |
free_cpumask_var(offlined_cpus); |
| 227 |
return err; |
| 228 |
} |
| 229 |
|
| 230 |
static void dummy_callback(struct timer_list *unused) {} |
| 231 |
|
| 232 |
static int suspend_cpu(struct cpuidle_device *dev, |
| 233 |
struct cpuidle_driver *drv, int index) |
| 234 |
{ |
| 235 |
struct cpuidle_state *state = &drv->states[index]; |
| 236 |
bool broadcast = state->flags & CPUIDLE_FLAG_TIMER_STOP; |
| 237 |
int ret; |
| 238 |
|
| 239 |
arch_cpu_idle_enter(); |
| 240 |
|
| 241 |
if (broadcast) { |
| 242 |
/* |
| 243 |
* The local timer will be shut down, we need to enter tick |
| 244 |
* broadcast. |
| 245 |
*/ |
| 246 |
ret = tick_broadcast_enter(); |
| 247 |
if (ret) { |
| 248 |
/* |
| 249 |
* In the absence of hardware broadcast mechanism, |
| 250 |
* this CPU might be used to broadcast wakeups, which |
| 251 |
* may be why entering tick broadcast has failed. |
| 252 |
* There is little the kernel can do to work around |
| 253 |
* that, so enter WFI instead (idle state 0). |
| 254 |
*/ |
| 255 |
cpu_do_idle(); |
| 256 |
ret = 0; |
| 257 |
goto out_arch_exit; |
| 258 |
} |
| 259 |
} |
| 260 |
|
| 261 |
ret = state->enter(dev, drv, index); |
| 262 |
|
| 263 |
if (broadcast) |
| 264 |
tick_broadcast_exit(); |
| 265 |
|
| 266 |
out_arch_exit: |
| 267 |
arch_cpu_idle_exit(); |
| 268 |
|
| 269 |
return ret; |
| 270 |
} |
| 271 |
|
| 272 |
static int suspend_test_thread(void *arg) |
| 273 |
{ |
| 274 |
int cpu = (long)arg; |
| 275 |
int i, nb_suspend = 0, nb_shallow_sleep = 0, nb_err = 0; |
| 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; |
| 280 |
|
| 281 |
/* Wait for the main thread to give the start signal. */ |
| 282 |
wait_for_completion(&suspend_threads_started); |
| 283 |
|
| 284 |
/* Set maximum priority to preempt all other threads on this CPU. */ |
| 285 |
sched_set_fifo(current); |
| 286 |
|
| 287 |
dev = this_cpu_read(cpuidle_devices); |
| 288 |
drv = cpuidle_get_cpu_driver(dev); |
| 289 |
|
| 290 |
pr_info("CPU %d entering suspend cycles, states 1 through %d\n", |
| 291 |
cpu, drv->state_count - 1); |
| 292 |
|
| 293 |
timer_setup_on_stack(&wakeup_timer, dummy_callback, 0); |
| 294 |
for (i = 0; i < NUM_SUSPEND_CYCLE; ++i) { |
| 295 |
int index; |
| 296 |
/* |
| 297 |
* Test all possible states, except 0 (which is usually WFI and |
| 298 |
* doesn't use PSCI). |
| 299 |
*/ |
| 300 |
for (index = 1; index < drv->state_count; ++index) { |
| 301 |
int ret; |
| 302 |
struct cpuidle_state *state = &drv->states[index]; |
| 303 |
|
| 304 |
/* |
| 305 |
* Set the timer to wake this CPU up in some time (which |
| 306 |
* should be largely sufficient for entering suspend). |
| 307 |
* If the local tick is disabled when entering suspend, |
| 308 |
* suspend_cpu() takes care of switching to a broadcast |
| 309 |
* tick, so the timer will still wake us up. |
| 310 |
*/ |
| 311 |
mod_timer(&wakeup_timer, jiffies + |
| 312 |
usecs_to_jiffies(state->target_residency)); |
| 313 |
|
| 314 |
/* IRQs must be disabled during suspend operations. */ |
| 315 |
local_irq_disable(); |
| 316 |
|
| 317 |
ret = suspend_cpu(dev, drv, index); |
| 318 |
|
| 319 |
/* |
| 320 |
* We have woken up. Re-enable IRQs to handle any |
| 321 |
* pending interrupt, do not wait until the end of the |
| 322 |
* loop. |
| 323 |
*/ |
| 324 |
local_irq_enable(); |
| 325 |
|
| 326 |
if (ret == index) { |
| 327 |
++nb_suspend; |
| 328 |
} else if (ret >= 0) { |
| 329 |
/* We did not enter the expected state. */ |
| 330 |
++nb_shallow_sleep; |
| 331 |
} else { |
| 332 |
pr_err("Failed to suspend CPU %d: error %d " |
| 333 |
"(requested state %d, cycle %d)\n", |
| 334 |
cpu, ret, index, i); |
| 335 |
++nb_err; |
| 336 |
} |
| 337 |
} |
| 338 |
} |
| 339 |
|
| 340 |
/* |
| 341 |
* Disable the timer to make sure that the timer will not trigger |
| 342 |
* later. |
| 343 |
*/ |
| 344 |
timer_delete(&wakeup_timer); |
| 345 |
timer_destroy_on_stack(&wakeup_timer); |
| 346 |
|
| 347 |
if (atomic_dec_return_relaxed(&nb_active_threads) == 0) |
| 348 |
complete(&suspend_threads_done); |
| 349 |
|
| 350 |
for (;;) { |
| 351 |
/* Needs to be set first to avoid missing a wakeup. */ |
| 352 |
set_current_state(TASK_INTERRUPTIBLE); |
| 353 |
if (kthread_should_park()) |
| 354 |
break; |
| 355 |
schedule(); |
| 356 |
} |
| 357 |
|
| 358 |
pr_info("CPU %d suspend test results: success %d, shallow states %d, errors %d\n", |
| 359 |
cpu, nb_suspend, nb_shallow_sleep, nb_err); |
| 360 |
|
| 361 |
kthread_parkme(); |
| 362 |
|
| 363 |
return nb_err; |
| 364 |
} |
| 365 |
|
| 366 |
static int suspend_tests(void) |
| 367 |
{ |
| 368 |
int i, cpu, err = 0; |
| 369 |
struct task_struct **threads; |
| 370 |
int nb_threads = 0; |
| 371 |
|
| 372 |
threads = kmalloc_objs(*threads, nb_available_cpus); |
| 373 |
if (!threads) |
| 374 |
return -ENOMEM; |
| 375 |
|
| 376 |
/* |
| 377 |
* Stop cpuidle to prevent the idle tasks from entering a deep sleep |
| 378 |
* mode, as it might interfere with the suspend threads on other CPUs. |
| 379 |
* This does not prevent the suspend threads from using cpuidle (only |
| 380 |
* the idle tasks check this status). Take the idle lock so that |
| 381 |
* the cpuidle driver and device look-up can be carried out safely. |
| 382 |
*/ |
| 383 |
cpuidle_pause_and_lock(); |
| 384 |
|
| 385 |
for_each_online_cpu(cpu) { |
| 386 |
struct task_struct *thread; |
| 387 |
/* Check that cpuidle is available on that CPU. */ |
| 388 |
struct cpuidle_device *dev = per_cpu(cpuidle_devices, cpu); |
| 389 |
struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev); |
| 390 |
|
| 391 |
if (!dev || !drv) { |
| 392 |
pr_warn("cpuidle not available on CPU %d, ignoring\n", |
| 393 |
cpu); |
| 394 |
continue; |
| 395 |
} |
| 396 |
|
| 397 |
thread = kthread_create_on_cpu(suspend_test_thread, |
| 398 |
(void *)(long)cpu, cpu, |
| 399 |
"psci_suspend_test"); |
| 400 |
if (IS_ERR(thread)) |
| 401 |
pr_err("Failed to create kthread on CPU %d\n", cpu); |
| 402 |
else |
| 403 |
threads[nb_threads++] = thread; |
| 404 |
} |
| 405 |
|
| 406 |
if (nb_threads < 1) { |
| 407 |
err = -ENODEV; |
| 408 |
goto out; |
| 409 |
} |
| 410 |
|
| 411 |
atomic_set(&nb_active_threads, nb_threads); |
| 412 |
|
| 413 |
/* |
| 414 |
* Wake up the suspend threads. To avoid the main thread being preempted |
| 415 |
* before all the threads have been unparked, the suspend threads will |
| 416 |
* wait for the completion of suspend_threads_started. |
| 417 |
*/ |
| 418 |
for (i = 0; i < nb_threads; ++i) |
| 419 |
wake_up_process(threads[i]); |
| 420 |
complete_all(&suspend_threads_started); |
| 421 |
|
| 422 |
wait_for_completion(&suspend_threads_done); |
| 423 |
|
| 424 |
|
| 425 |
/* Stop and destroy all threads, get return status. */ |
| 426 |
for (i = 0; i < nb_threads; ++i) { |
| 427 |
err += kthread_park(threads[i]); |
| 428 |
err += kthread_stop(threads[i]); |
| 429 |
} |
| 430 |
out: |
| 431 |
cpuidle_resume_and_unlock(); |
| 432 |
kfree(threads); |
| 433 |
return err; |
| 434 |
} |
| 435 |
|
| 436 |
static int __init psci_checker(void) |
| 437 |
{ |
| 438 |
int ret; |
| 439 |
|
| 440 |
/* |
| 441 |
* Since we're in an initcall, we assume that all the CPUs that all |
| 442 |
* CPUs that can be onlined have been onlined. |
| 443 |
* |
| 444 |
* The tests assume that hotplug is enabled but nobody else is using it, |
| 445 |
* otherwise the results will be unpredictable. However, since there |
| 446 |
* is no userspace yet in initcalls, that should be fine, as long as |
| 447 |
* no torture test is running at the same time (see Kconfig). |
| 448 |
*/ |
| 449 |
nb_available_cpus = num_online_cpus(); |
| 450 |
|
| 451 |
/* Check PSCI operations are set up and working. */ |
| 452 |
ret = psci_ops_check(); |
| 453 |
if (ret) |
| 454 |
return ret; |
| 455 |
|
| 456 |
pr_info("PSCI checker started using %u CPUs\n", nb_available_cpus); |
| 457 |
|
| 458 |
pr_info("Starting hotplug tests\n"); |
| 459 |
ret = hotplug_tests(); |
| 460 |
if (ret == 0) |
| 461 |
pr_info("Hotplug tests passed OK\n"); |
| 462 |
else if (ret > 0) |
| 463 |
pr_err("%d error(s) encountered in hotplug tests\n", ret); |
| 464 |
else { |
| 465 |
pr_err("Out of memory\n"); |
| 466 |
return ret; |
| 467 |
} |
| 468 |
|
| 469 |
pr_info("Starting suspend tests (%d cycles per state)\n", |
| 470 |
NUM_SUSPEND_CYCLE); |
| 471 |
ret = suspend_tests(); |
| 472 |
if (ret == 0) |
| 473 |
pr_info("Suspend tests passed OK\n"); |
| 474 |
else if (ret > 0) |
| 475 |
pr_err("%d error(s) encountered in suspend tests\n", ret); |
| 476 |
else { |
| 477 |
switch (ret) { |
| 478 |
case -ENOMEM: |
| 479 |
pr_err("Out of memory\n"); |
| 480 |
break; |
| 481 |
case -ENODEV: |
| 482 |
pr_warn("Could not start suspend tests on any CPU\n"); |
| 483 |
break; |
| 484 |
} |
| 485 |
} |
| 486 |
|
| 487 |
pr_info("PSCI checker completed\n"); |
| 488 |
return ret < 0 ? ret : 0; |
| 489 |
} |
| 490 |
late_initcall(psci_checker); |