Commit | Line | Data |
---|---|---|
2025cf9e | 1 | // SPDX-License-Identifier: GPL-2.0-only |
8e0af514 SL |
2 | /* |
3 | * acpi_pad.c ACPI Processor Aggregator Driver | |
4 | * | |
5 | * Copyright (c) 2009, Intel Corporation. | |
8e0af514 SL |
6 | */ |
7 | ||
8 | #include <linux/kernel.h> | |
9 | #include <linux/cpumask.h> | |
10 | #include <linux/module.h> | |
11 | #include <linux/init.h> | |
12 | #include <linux/types.h> | |
13 | #include <linux/kthread.h> | |
ae7e81c0 | 14 | #include <uapi/linux/sched/types.h> |
8e0af514 SL |
15 | #include <linux/freezer.h> |
16 | #include <linux/cpu.h> | |
979081e7 | 17 | #include <linux/tick.h> |
5a0e3ad6 | 18 | #include <linux/slab.h> |
8b48463f | 19 | #include <linux/acpi.h> |
2a606a18 | 20 | #include <linux/perf_event.h> |
dd0261bb | 21 | #include <linux/platform_device.h> |
bc83cccc | 22 | #include <asm/mwait.h> |
e311404f | 23 | #include <xen/xen.h> |
8e0af514 | 24 | |
a40770a9 | 25 | #define ACPI_PROCESSOR_AGGREGATOR_CLASS "acpi_pad" |
8e0af514 SL |
26 | #define ACPI_PROCESSOR_AGGREGATOR_DEVICE_NAME "Processor Aggregator" |
27 | #define ACPI_PROCESSOR_AGGREGATOR_NOTIFY 0x80 | |
28 | static DEFINE_MUTEX(isolated_cpus_lock); | |
5f160126 | 29 | static DEFINE_MUTEX(round_robin_lock); |
8e0af514 | 30 | |
8e0af514 | 31 | static unsigned long power_saving_mwait_eax; |
0dc698b9 VP |
32 | |
33 | static unsigned char tsc_detected_unstable; | |
34 | static unsigned char tsc_marked_unstable; | |
35 | ||
8e0af514 SL |
36 | static void power_saving_mwait_init(void) |
37 | { | |
38 | unsigned int eax, ebx, ecx, edx; | |
39 | unsigned int highest_cstate = 0; | |
40 | unsigned int highest_subcstate = 0; | |
41 | int i; | |
42 | ||
43 | if (!boot_cpu_has(X86_FEATURE_MWAIT)) | |
44 | return; | |
45 | if (boot_cpu_data.cpuid_level < CPUID_MWAIT_LEAF) | |
46 | return; | |
47 | ||
48 | cpuid(CPUID_MWAIT_LEAF, &eax, &ebx, &ecx, &edx); | |
49 | ||
50 | if (!(ecx & CPUID5_ECX_EXTENSIONS_SUPPORTED) || | |
51 | !(ecx & CPUID5_ECX_INTERRUPT_BREAK)) | |
52 | return; | |
53 | ||
54 | edx >>= MWAIT_SUBSTATE_SIZE; | |
55 | for (i = 0; i < 7 && edx; i++, edx >>= MWAIT_SUBSTATE_SIZE) { | |
56 | if (edx & MWAIT_SUBSTATE_MASK) { | |
57 | highest_cstate = i; | |
58 | highest_subcstate = edx & MWAIT_SUBSTATE_MASK; | |
59 | } | |
60 | } | |
61 | power_saving_mwait_eax = (highest_cstate << MWAIT_SUBSTATE_SIZE) | | |
62 | (highest_subcstate - 1); | |
63 | ||
592913ec | 64 | #if defined(CONFIG_X86) |
8e0af514 | 65 | switch (boot_cpu_data.x86_vendor) { |
7377ed4b | 66 | case X86_VENDOR_HYGON: |
8e0af514 SL |
67 | case X86_VENDOR_AMD: |
68 | case X86_VENDOR_INTEL: | |
773b2f30 | 69 | case X86_VENDOR_ZHAOXIN: |
b72f301c | 70 | case X86_VENDOR_CENTAUR: |
8e0af514 SL |
71 | /* |
72 | * AMD Fam10h TSC will tick in all | |
73 | * C/P/S0/S1 states when this bit is set. | |
74 | */ | |
8aa4b14e CG |
75 | if (!boot_cpu_has(X86_FEATURE_NONSTOP_TSC)) |
76 | tsc_detected_unstable = 1; | |
8aa4b14e | 77 | break; |
8e0af514 | 78 | default: |
3ff70551 | 79 | /* TSC could halt in idle */ |
0dc698b9 | 80 | tsc_detected_unstable = 1; |
8e0af514 SL |
81 | } |
82 | #endif | |
83 | } | |
84 | ||
85 | static unsigned long cpu_weight[NR_CPUS]; | |
86 | static int tsk_in_cpu[NR_CPUS] = {[0 ... NR_CPUS-1] = -1}; | |
87 | static DECLARE_BITMAP(pad_busy_cpus_bits, NR_CPUS); | |
88 | static void round_robin_cpu(unsigned int tsk_index) | |
89 | { | |
90 | struct cpumask *pad_busy_cpus = to_cpumask(pad_busy_cpus_bits); | |
91 | cpumask_var_t tmp; | |
92 | int cpu; | |
f67538f8 | 93 | unsigned long min_weight = -1; |
134043cd | 94 | unsigned long preferred_cpu; |
8e0af514 SL |
95 | |
96 | if (!alloc_cpumask_var(&tmp, GFP_KERNEL)) | |
97 | return; | |
98 | ||
5f160126 | 99 | mutex_lock(&round_robin_lock); |
8e0af514 SL |
100 | cpumask_clear(tmp); |
101 | for_each_cpu(cpu, pad_busy_cpus) | |
06931e62 | 102 | cpumask_or(tmp, tmp, topology_sibling_cpumask(cpu)); |
8e0af514 | 103 | cpumask_andnot(tmp, cpu_online_mask, tmp); |
51a23b1b | 104 | /* avoid HT siblings if possible */ |
8e0af514 SL |
105 | if (cpumask_empty(tmp)) |
106 | cpumask_andnot(tmp, cpu_online_mask, pad_busy_cpus); | |
107 | if (cpumask_empty(tmp)) { | |
5f160126 | 108 | mutex_unlock(&round_robin_lock); |
8b29d29a | 109 | free_cpumask_var(tmp); |
8e0af514 SL |
110 | return; |
111 | } | |
112 | for_each_cpu(cpu, tmp) { | |
113 | if (cpu_weight[cpu] < min_weight) { | |
114 | min_weight = cpu_weight[cpu]; | |
115 | preferred_cpu = cpu; | |
116 | } | |
117 | } | |
118 | ||
119 | if (tsk_in_cpu[tsk_index] != -1) | |
120 | cpumask_clear_cpu(tsk_in_cpu[tsk_index], pad_busy_cpus); | |
121 | tsk_in_cpu[tsk_index] = preferred_cpu; | |
122 | cpumask_set_cpu(preferred_cpu, pad_busy_cpus); | |
123 | cpu_weight[preferred_cpu]++; | |
5f160126 | 124 | mutex_unlock(&round_robin_lock); |
8e0af514 SL |
125 | |
126 | set_cpus_allowed_ptr(current, cpumask_of(preferred_cpu)); | |
8b29d29a LS |
127 | |
128 | free_cpumask_var(tmp); | |
8e0af514 SL |
129 | } |
130 | ||
131 | static void exit_round_robin(unsigned int tsk_index) | |
132 | { | |
133 | struct cpumask *pad_busy_cpus = to_cpumask(pad_busy_cpus_bits); | |
c8eb628c | 134 | |
8e0af514 SL |
135 | cpumask_clear_cpu(tsk_in_cpu[tsk_index], pad_busy_cpus); |
136 | tsk_in_cpu[tsk_index] = -1; | |
137 | } | |
138 | ||
139 | static unsigned int idle_pct = 5; /* percentage */ | |
fa7584e1 | 140 | static unsigned int round_robin_time = 1; /* second */ |
8e0af514 SL |
141 | static int power_saving_thread(void *data) |
142 | { | |
8e0af514 SL |
143 | int do_sleep; |
144 | unsigned int tsk_index = (unsigned long)data; | |
145 | u64 last_jiffies = 0; | |
146 | ||
4ca6c1a0 | 147 | sched_set_fifo_low(current); |
8e0af514 SL |
148 | |
149 | while (!kthread_should_stop()) { | |
4ff248f3 | 150 | unsigned long expire_time; |
8e0af514 | 151 | |
8e0af514 | 152 | /* round robin to cpus */ |
4ff248f3 MS |
153 | expire_time = last_jiffies + round_robin_time * HZ; |
154 | if (time_before(expire_time, jiffies)) { | |
8e0af514 SL |
155 | last_jiffies = jiffies; |
156 | round_robin_cpu(tsk_index); | |
157 | } | |
158 | ||
159 | do_sleep = 0; | |
160 | ||
8e0af514 SL |
161 | expire_time = jiffies + HZ * (100 - idle_pct) / 100; |
162 | ||
163 | while (!need_resched()) { | |
0dc698b9 VP |
164 | if (tsc_detected_unstable && !tsc_marked_unstable) { |
165 | /* TSC could halt in idle, so notify users */ | |
166 | mark_tsc_unstable("TSC halts in idle"); | |
167 | tsc_marked_unstable = 1; | |
168 | } | |
8e0af514 | 169 | local_irq_disable(); |
2a606a18 SE |
170 | |
171 | perf_lopwr_cb(true); | |
172 | ||
979081e7 | 173 | tick_broadcast_enable(); |
c7952135 | 174 | tick_broadcast_enter(); |
8e0af514 SL |
175 | stop_critical_timings(); |
176 | ||
16824255 | 177 | mwait_idle_with_hints(power_saving_mwait_eax, 1); |
8e0af514 SL |
178 | |
179 | start_critical_timings(); | |
c7952135 | 180 | tick_broadcast_exit(); |
2a606a18 SE |
181 | |
182 | perf_lopwr_cb(false); | |
183 | ||
8e0af514 SL |
184 | local_irq_enable(); |
185 | ||
4ff248f3 | 186 | if (time_before(expire_time, jiffies)) { |
8e0af514 SL |
187 | do_sleep = 1; |
188 | break; | |
189 | } | |
190 | } | |
191 | ||
8e0af514 SL |
192 | /* |
193 | * current sched_rt has threshold for rt task running time. | |
194 | * When a rt task uses 95% CPU time, the rt thread will be | |
195 | * scheduled out for 5% CPU time to not starve other tasks. But | |
196 | * the mechanism only works when all CPUs have RT task running, | |
197 | * as if one CPU hasn't RT task, RT task from other CPUs will | |
198 | * borrow CPU time from this CPU and cause RT task use > 95% | |
3b8cb427 | 199 | * CPU time. To make 'avoid starvation' work, takes a nap here. |
8e0af514 | 200 | */ |
5b59c69e | 201 | if (unlikely(do_sleep)) |
8e0af514 | 202 | schedule_timeout_killable(HZ * idle_pct / 100); |
5b59c69e TC |
203 | |
204 | /* If an external event has set the need_resched flag, then | |
205 | * we need to deal with it, or this loop will continue to | |
206 | * spin without calling __mwait(). | |
207 | */ | |
208 | if (unlikely(need_resched())) | |
209 | schedule(); | |
8e0af514 SL |
210 | } |
211 | ||
212 | exit_round_robin(tsk_index); | |
213 | return 0; | |
214 | } | |
215 | ||
216 | static struct task_struct *ps_tsks[NR_CPUS]; | |
217 | static unsigned int ps_tsk_num; | |
218 | static int create_power_saving_task(void) | |
219 | { | |
5d7e4386 | 220 | int rc; |
3b8cb427 | 221 | |
8e0af514 SL |
222 | ps_tsks[ps_tsk_num] = kthread_run(power_saving_thread, |
223 | (void *)(unsigned long)ps_tsk_num, | |
150ed86f | 224 | "acpi_pad/%d", ps_tsk_num); |
5d7e4386 RR |
225 | |
226 | if (IS_ERR(ps_tsks[ps_tsk_num])) { | |
227 | rc = PTR_ERR(ps_tsks[ps_tsk_num]); | |
3b8cb427 | 228 | ps_tsks[ps_tsk_num] = NULL; |
5d7e4386 RR |
229 | } else { |
230 | rc = 0; | |
231 | ps_tsk_num++; | |
232 | } | |
3b8cb427 CG |
233 | |
234 | return rc; | |
8e0af514 SL |
235 | } |
236 | ||
237 | static void destroy_power_saving_task(void) | |
238 | { | |
239 | if (ps_tsk_num > 0) { | |
240 | ps_tsk_num--; | |
241 | kthread_stop(ps_tsks[ps_tsk_num]); | |
3b8cb427 | 242 | ps_tsks[ps_tsk_num] = NULL; |
8e0af514 SL |
243 | } |
244 | } | |
245 | ||
246 | static void set_power_saving_task_num(unsigned int num) | |
247 | { | |
248 | if (num > ps_tsk_num) { | |
249 | while (ps_tsk_num < num) { | |
250 | if (create_power_saving_task()) | |
251 | return; | |
252 | } | |
253 | } else if (num < ps_tsk_num) { | |
254 | while (ps_tsk_num > num) | |
255 | destroy_power_saving_task(); | |
256 | } | |
257 | } | |
258 | ||
3b8cb427 | 259 | static void acpi_pad_idle_cpus(unsigned int num_cpus) |
8e0af514 | 260 | { |
95ac7067 | 261 | cpus_read_lock(); |
8e0af514 SL |
262 | |
263 | num_cpus = min_t(unsigned int, num_cpus, num_online_cpus()); | |
264 | set_power_saving_task_num(num_cpus); | |
265 | ||
95ac7067 | 266 | cpus_read_unlock(); |
8e0af514 SL |
267 | } |
268 | ||
269 | static uint32_t acpi_pad_idle_cpus_num(void) | |
270 | { | |
271 | return ps_tsk_num; | |
272 | } | |
273 | ||
0f39ee83 | 274 | static ssize_t rrtime_store(struct device *dev, |
8e0af514 SL |
275 | struct device_attribute *attr, const char *buf, size_t count) |
276 | { | |
277 | unsigned long num; | |
c8eb628c | 278 | |
73d4511a | 279 | if (kstrtoul(buf, 0, &num)) |
8e0af514 SL |
280 | return -EINVAL; |
281 | if (num < 1 || num >= 100) | |
282 | return -EINVAL; | |
283 | mutex_lock(&isolated_cpus_lock); | |
284 | round_robin_time = num; | |
285 | mutex_unlock(&isolated_cpus_lock); | |
286 | return count; | |
287 | } | |
288 | ||
0f39ee83 | 289 | static ssize_t rrtime_show(struct device *dev, |
8e0af514 SL |
290 | struct device_attribute *attr, char *buf) |
291 | { | |
92266c65 | 292 | return sysfs_emit(buf, "%d\n", round_robin_time); |
8e0af514 | 293 | } |
0f39ee83 | 294 | static DEVICE_ATTR_RW(rrtime); |
8e0af514 | 295 | |
0f39ee83 | 296 | static ssize_t idlepct_store(struct device *dev, |
8e0af514 SL |
297 | struct device_attribute *attr, const char *buf, size_t count) |
298 | { | |
299 | unsigned long num; | |
c8eb628c | 300 | |
73d4511a | 301 | if (kstrtoul(buf, 0, &num)) |
8e0af514 SL |
302 | return -EINVAL; |
303 | if (num < 1 || num >= 100) | |
304 | return -EINVAL; | |
305 | mutex_lock(&isolated_cpus_lock); | |
306 | idle_pct = num; | |
307 | mutex_unlock(&isolated_cpus_lock); | |
308 | return count; | |
309 | } | |
310 | ||
0f39ee83 | 311 | static ssize_t idlepct_show(struct device *dev, |
8e0af514 SL |
312 | struct device_attribute *attr, char *buf) |
313 | { | |
92266c65 | 314 | return sysfs_emit(buf, "%d\n", idle_pct); |
8e0af514 | 315 | } |
0f39ee83 | 316 | static DEVICE_ATTR_RW(idlepct); |
8e0af514 | 317 | |
0f39ee83 | 318 | static ssize_t idlecpus_store(struct device *dev, |
8e0af514 SL |
319 | struct device_attribute *attr, const char *buf, size_t count) |
320 | { | |
321 | unsigned long num; | |
c8eb628c | 322 | |
73d4511a | 323 | if (kstrtoul(buf, 0, &num)) |
8e0af514 SL |
324 | return -EINVAL; |
325 | mutex_lock(&isolated_cpus_lock); | |
326 | acpi_pad_idle_cpus(num); | |
327 | mutex_unlock(&isolated_cpus_lock); | |
328 | return count; | |
329 | } | |
330 | ||
0f39ee83 | 331 | static ssize_t idlecpus_show(struct device *dev, |
8e0af514 SL |
332 | struct device_attribute *attr, char *buf) |
333 | { | |
5aaba363 SH |
334 | return cpumap_print_to_pagebuf(false, buf, |
335 | to_cpumask(pad_busy_cpus_bits)); | |
8e0af514 | 336 | } |
5aaba363 | 337 | |
0f39ee83 | 338 | static DEVICE_ATTR_RW(idlecpus); |
8e0af514 | 339 | |
d7228c71 MW |
340 | static struct attribute *acpi_pad_attrs[] = { |
341 | &dev_attr_idlecpus.attr, | |
342 | &dev_attr_idlepct.attr, | |
343 | &dev_attr_rrtime.attr, | |
344 | NULL | |
345 | }; | |
8e0af514 | 346 | |
d7228c71 | 347 | ATTRIBUTE_GROUPS(acpi_pad); |
8e0af514 | 348 | |
c9ad8e06 LB |
349 | /* |
350 | * Query firmware how many CPUs should be idle | |
351 | * return -1 on failure | |
352 | */ | |
353 | static int acpi_pad_pur(acpi_handle handle) | |
8e0af514 SL |
354 | { |
355 | struct acpi_buffer buffer = {ACPI_ALLOCATE_BUFFER, NULL}; | |
8e0af514 | 356 | union acpi_object *package; |
c9ad8e06 | 357 | int num = -1; |
8e0af514 | 358 | |
3b8cb427 | 359 | if (ACPI_FAILURE(acpi_evaluate_object(handle, "_PUR", NULL, &buffer))) |
c9ad8e06 | 360 | return num; |
3b8cb427 CG |
361 | |
362 | if (!buffer.length || !buffer.pointer) | |
c9ad8e06 | 363 | return num; |
3b8cb427 | 364 | |
8e0af514 | 365 | package = buffer.pointer; |
c9ad8e06 LB |
366 | |
367 | if (package->type == ACPI_TYPE_PACKAGE && | |
368 | package->package.count == 2 && | |
369 | package->package.elements[0].integer.value == 1) /* rev 1 */ | |
370 | ||
371 | num = package->package.elements[1].integer.value; | |
372 | ||
8e0af514 | 373 | kfree(buffer.pointer); |
c9ad8e06 | 374 | return num; |
8e0af514 SL |
375 | } |
376 | ||
8e0af514 SL |
377 | static void acpi_pad_handle_notify(acpi_handle handle) |
378 | { | |
3b8cb427 | 379 | int num_cpus; |
8e0af514 | 380 | uint32_t idle_cpus; |
8b296d94 JL |
381 | struct acpi_buffer param = { |
382 | .length = 4, | |
383 | .pointer = (void *)&idle_cpus, | |
384 | }; | |
8e0af514 SL |
385 | |
386 | mutex_lock(&isolated_cpus_lock); | |
c9ad8e06 LB |
387 | num_cpus = acpi_pad_pur(handle); |
388 | if (num_cpus < 0) { | |
8e0af514 SL |
389 | mutex_unlock(&isolated_cpus_lock); |
390 | return; | |
391 | } | |
3b8cb427 | 392 | acpi_pad_idle_cpus(num_cpus); |
8e0af514 | 393 | idle_cpus = acpi_pad_idle_cpus_num(); |
8b296d94 | 394 | acpi_evaluate_ost(handle, ACPI_PROCESSOR_AGGREGATOR_NOTIFY, 0, ¶m); |
8e0af514 SL |
395 | mutex_unlock(&isolated_cpus_lock); |
396 | } | |
397 | ||
398 | static void acpi_pad_notify(acpi_handle handle, u32 event, | |
399 | void *data) | |
400 | { | |
5ccd40c5 | 401 | struct acpi_device *adev = data; |
8e0af514 SL |
402 | |
403 | switch (event) { | |
404 | case ACPI_PROCESSOR_AGGREGATOR_NOTIFY: | |
405 | acpi_pad_handle_notify(handle); | |
5ccd40c5 MW |
406 | acpi_bus_generate_netlink_event(adev->pnp.device_class, |
407 | dev_name(&adev->dev), event, 0); | |
8e0af514 SL |
408 | break; |
409 | default: | |
73d4511a | 410 | pr_warn("Unsupported event [0x%x]\n", event); |
8e0af514 SL |
411 | break; |
412 | } | |
413 | } | |
414 | ||
dd0261bb | 415 | static int acpi_pad_probe(struct platform_device *pdev) |
8e0af514 | 416 | { |
5ccd40c5 | 417 | struct acpi_device *adev = ACPI_COMPANION(&pdev->dev); |
8e0af514 SL |
418 | acpi_status status; |
419 | ||
5ccd40c5 MW |
420 | strcpy(acpi_device_name(adev), ACPI_PROCESSOR_AGGREGATOR_DEVICE_NAME); |
421 | strcpy(acpi_device_class(adev), ACPI_PROCESSOR_AGGREGATOR_CLASS); | |
422 | ||
423 | status = acpi_install_notify_handler(adev->handle, | |
424 | ACPI_DEVICE_NOTIFY, acpi_pad_notify, adev); | |
8e0af514 | 425 | |
d7228c71 | 426 | if (ACPI_FAILURE(status)) |
8e0af514 | 427 | return -ENODEV; |
8e0af514 SL |
428 | |
429 | return 0; | |
430 | } | |
431 | ||
dd0261bb | 432 | static void acpi_pad_remove(struct platform_device *pdev) |
8e0af514 | 433 | { |
5ccd40c5 | 434 | struct acpi_device *adev = ACPI_COMPANION(&pdev->dev); |
dd0261bb | 435 | |
8e0af514 SL |
436 | mutex_lock(&isolated_cpus_lock); |
437 | acpi_pad_idle_cpus(0); | |
438 | mutex_unlock(&isolated_cpus_lock); | |
439 | ||
5ccd40c5 | 440 | acpi_remove_notify_handler(adev->handle, |
8e0af514 | 441 | ACPI_DEVICE_NOTIFY, acpi_pad_notify); |
8e0af514 SL |
442 | } |
443 | ||
444 | static const struct acpi_device_id pad_device_ids[] = { | |
445 | {"ACPI000C", 0}, | |
446 | {"", 0}, | |
447 | }; | |
448 | MODULE_DEVICE_TABLE(acpi, pad_device_ids); | |
449 | ||
dd0261bb MW |
450 | static struct platform_driver acpi_pad_driver = { |
451 | .probe = acpi_pad_probe, | |
452 | .remove_new = acpi_pad_remove, | |
453 | .driver = { | |
d7228c71 | 454 | .dev_groups = acpi_pad_groups, |
dd0261bb MW |
455 | .name = "processor_aggregator", |
456 | .acpi_match_table = pad_device_ids, | |
8e0af514 SL |
457 | }, |
458 | }; | |
459 | ||
460 | static int __init acpi_pad_init(void) | |
461 | { | |
e311404f JG |
462 | /* Xen ACPI PAD is used when running as Xen Dom0. */ |
463 | if (xen_initial_domain()) | |
464 | return -ENODEV; | |
465 | ||
8e0af514 SL |
466 | power_saving_mwait_init(); |
467 | if (power_saving_mwait_eax == 0) | |
468 | return -EINVAL; | |
469 | ||
dd0261bb | 470 | return platform_driver_register(&acpi_pad_driver); |
8e0af514 SL |
471 | } |
472 | ||
473 | static void __exit acpi_pad_exit(void) | |
474 | { | |
dd0261bb | 475 | platform_driver_unregister(&acpi_pad_driver); |
8e0af514 SL |
476 | } |
477 | ||
478 | module_init(acpi_pad_init); | |
479 | module_exit(acpi_pad_exit); | |
480 | MODULE_AUTHOR("Shaohua Li<shaohua.li@intel.com>"); | |
481 | MODULE_DESCRIPTION("ACPI Processor Aggregator Driver"); | |
482 | MODULE_LICENSE("GPL"); |