1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright IBM Corp. 2007, 2011
4 * Author(s): Heiko Carstens <heiko.carstens@de.ibm.com>
7 #define KMSG_COMPONENT "cpu"
8 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
10 #include <linux/workqueue.h>
11 #include <linux/memblock.h>
12 #include <linux/uaccess.h>
13 #include <linux/sysctl.h>
14 #include <linux/cpuset.h>
15 #include <linux/device.h>
16 #include <linux/export.h>
17 #include <linux/kernel.h>
18 #include <linux/sched.h>
19 #include <linux/sched/topology.h>
20 #include <linux/delay.h>
21 #include <linux/init.h>
22 #include <linux/slab.h>
23 #include <linux/cpu.h>
24 #include <linux/smp.h>
26 #include <linux/nodemask.h>
27 #include <linux/node.h>
28 #include <asm/sysinfo.h>
31 #define PTF_HORIZONTAL (0UL)
32 #define PTF_VERTICAL (1UL)
33 #define PTF_CHECK (2UL)
38 TOPOLOGY_MODE_PACKAGE,
39 TOPOLOGY_MODE_UNINITIALIZED
43 struct mask_info *next;
48 static int topology_mode = TOPOLOGY_MODE_UNINITIALIZED;
49 static void set_topology_timer(void);
50 static void topology_work_fn(struct work_struct *work);
51 static struct sysinfo_15_1_x *tl_info;
53 static DECLARE_WORK(topology_work, topology_work_fn);
56 * Socket/Book linked lists and cpu_topology updates are
57 * protected by "sched_domains_mutex".
59 static struct mask_info socket_info;
60 static struct mask_info book_info;
61 static struct mask_info drawer_info;
63 struct cpu_topology_s390 cpu_topology[NR_CPUS];
64 EXPORT_SYMBOL_GPL(cpu_topology);
66 cpumask_t cpus_with_topology;
68 static cpumask_t cpu_group_map(struct mask_info *info, unsigned int cpu)
72 cpumask_copy(&mask, cpumask_of(cpu));
73 switch (topology_mode) {
74 case TOPOLOGY_MODE_HW:
76 if (cpumask_test_cpu(cpu, &info->mask)) {
82 if (cpumask_empty(&mask))
83 cpumask_copy(&mask, cpumask_of(cpu));
85 case TOPOLOGY_MODE_PACKAGE:
86 cpumask_copy(&mask, cpu_present_mask);
90 case TOPOLOGY_MODE_SINGLE:
91 cpumask_copy(&mask, cpumask_of(cpu));
97 static cpumask_t cpu_thread_map(unsigned int cpu)
102 cpumask_copy(&mask, cpumask_of(cpu));
103 if (topology_mode != TOPOLOGY_MODE_HW)
105 cpu -= cpu % (smp_cpu_mtid + 1);
106 for (i = 0; i <= smp_cpu_mtid; i++)
107 if (cpu_present(cpu + i))
108 cpumask_set_cpu(cpu + i, &mask);
112 #define TOPOLOGY_CORE_BITS 64
114 static void add_cpus_to_mask(struct topology_core *tl_core,
115 struct mask_info *drawer,
116 struct mask_info *book,
117 struct mask_info *socket)
119 struct cpu_topology_s390 *topo;
122 for_each_set_bit(core, &tl_core->mask, TOPOLOGY_CORE_BITS) {
126 rcore = TOPOLOGY_CORE_BITS - 1 - core + tl_core->origin;
127 lcpu = smp_find_processor_id(rcore << smp_cpu_mt_shift);
130 for (i = 0; i <= smp_cpu_mtid; i++) {
131 topo = &cpu_topology[lcpu + i];
132 topo->drawer_id = drawer->id;
133 topo->book_id = book->id;
134 topo->socket_id = socket->id;
135 topo->core_id = rcore;
136 topo->thread_id = lcpu + i;
137 topo->dedicated = tl_core->d;
138 cpumask_set_cpu(lcpu + i, &drawer->mask);
139 cpumask_set_cpu(lcpu + i, &book->mask);
140 cpumask_set_cpu(lcpu + i, &socket->mask);
141 cpumask_set_cpu(lcpu + i, &cpus_with_topology);
142 smp_cpu_set_polarization(lcpu + i, tl_core->pp);
147 static void clear_masks(void)
149 struct mask_info *info;
153 cpumask_clear(&info->mask);
158 cpumask_clear(&info->mask);
163 cpumask_clear(&info->mask);
168 static union topology_entry *next_tle(union topology_entry *tle)
171 return (union topology_entry *)((struct topology_core *)tle + 1);
172 return (union topology_entry *)((struct topology_container *)tle + 1);
175 static void tl_to_masks(struct sysinfo_15_1_x *info)
177 struct mask_info *socket = &socket_info;
178 struct mask_info *book = &book_info;
179 struct mask_info *drawer = &drawer_info;
180 union topology_entry *tle, *end;
184 end = (union topology_entry *)((unsigned long)info + info->length);
188 drawer = drawer->next;
189 drawer->id = tle->container.id;
193 book->id = tle->container.id;
196 socket = socket->next;
197 socket->id = tle->container.id;
200 add_cpus_to_mask(&tle->cpu, drawer, book, socket);
210 static void topology_update_polarization_simple(void)
214 for_each_possible_cpu(cpu)
215 smp_cpu_set_polarization(cpu, POLARIZATION_HRZ);
218 static int ptf(unsigned long fc)
223 " .insn rre,0xb9a20000,%1,%1\n"
231 int topology_set_cpu_management(int fc)
235 if (!MACHINE_HAS_TOPOLOGY)
238 rc = ptf(PTF_VERTICAL);
240 rc = ptf(PTF_HORIZONTAL);
243 for_each_possible_cpu(cpu)
244 smp_cpu_set_polarization(cpu, POLARIZATION_UNKNOWN);
248 static void update_cpu_masks(void)
250 struct cpu_topology_s390 *topo;
253 for_each_possible_cpu(cpu) {
254 topo = &cpu_topology[cpu];
255 topo->thread_mask = cpu_thread_map(cpu);
256 topo->core_mask = cpu_group_map(&socket_info, cpu);
257 topo->book_mask = cpu_group_map(&book_info, cpu);
258 topo->drawer_mask = cpu_group_map(&drawer_info, cpu);
259 if (topology_mode != TOPOLOGY_MODE_HW) {
260 id = topology_mode == TOPOLOGY_MODE_PACKAGE ? 0 : cpu;
261 topo->thread_id = cpu;
263 topo->socket_id = id;
265 topo->drawer_id = id;
266 if (cpu_present(cpu))
267 cpumask_set_cpu(cpu, &cpus_with_topology);
270 numa_update_cpu_topology();
273 void store_topology(struct sysinfo_15_1_x *info)
275 stsi(info, 15, 1, topology_mnest_limit());
278 static void __arch_update_dedicated_flag(void *arg)
280 if (topology_cpu_dedicated(smp_processor_id()))
281 set_cpu_flag(CIF_DEDICATED_CPU);
283 clear_cpu_flag(CIF_DEDICATED_CPU);
286 static int __arch_update_cpu_topology(void)
288 struct sysinfo_15_1_x *info = tl_info;
291 mutex_lock(&smp_cpu_state_mutex);
292 cpumask_clear(&cpus_with_topology);
293 if (MACHINE_HAS_TOPOLOGY) {
295 store_topology(info);
299 if (!MACHINE_HAS_TOPOLOGY)
300 topology_update_polarization_simple();
301 mutex_unlock(&smp_cpu_state_mutex);
305 int arch_update_cpu_topology(void)
310 rc = __arch_update_cpu_topology();
311 on_each_cpu(__arch_update_dedicated_flag, NULL, 0);
312 for_each_online_cpu(cpu) {
313 dev = get_cpu_device(cpu);
314 kobject_uevent(&dev->kobj, KOBJ_CHANGE);
319 static void topology_work_fn(struct work_struct *work)
321 rebuild_sched_domains();
324 void topology_schedule_update(void)
326 schedule_work(&topology_work);
329 static void topology_flush_work(void)
331 flush_work(&topology_work);
334 static void topology_timer_fn(struct timer_list *unused)
337 topology_schedule_update();
338 set_topology_timer();
341 static struct timer_list topology_timer;
343 static atomic_t topology_poll = ATOMIC_INIT(0);
345 static void set_topology_timer(void)
347 if (atomic_add_unless(&topology_poll, -1, 0))
348 mod_timer(&topology_timer, jiffies + HZ / 10);
350 mod_timer(&topology_timer, jiffies + HZ * 60);
353 void topology_expect_change(void)
355 if (!MACHINE_HAS_TOPOLOGY)
357 /* This is racy, but it doesn't matter since it is just a heuristic.
358 * Worst case is that we poll in a higher frequency for a bit longer.
360 if (atomic_read(&topology_poll) > 60)
362 atomic_add(60, &topology_poll);
363 set_topology_timer();
366 static int cpu_management;
368 static ssize_t dispatching_show(struct device *dev,
369 struct device_attribute *attr,
374 mutex_lock(&smp_cpu_state_mutex);
375 count = sprintf(buf, "%d\n", cpu_management);
376 mutex_unlock(&smp_cpu_state_mutex);
380 static ssize_t dispatching_store(struct device *dev,
381 struct device_attribute *attr,
388 if (sscanf(buf, "%d %c", &val, &delim) != 1)
390 if (val != 0 && val != 1)
394 mutex_lock(&smp_cpu_state_mutex);
395 if (cpu_management == val)
397 rc = topology_set_cpu_management(val);
400 cpu_management = val;
401 topology_expect_change();
403 mutex_unlock(&smp_cpu_state_mutex);
405 return rc ? rc : count;
407 static DEVICE_ATTR_RW(dispatching);
409 static ssize_t cpu_polarization_show(struct device *dev,
410 struct device_attribute *attr, char *buf)
415 mutex_lock(&smp_cpu_state_mutex);
416 switch (smp_cpu_get_polarization(cpu)) {
417 case POLARIZATION_HRZ:
418 count = sprintf(buf, "horizontal\n");
420 case POLARIZATION_VL:
421 count = sprintf(buf, "vertical:low\n");
423 case POLARIZATION_VM:
424 count = sprintf(buf, "vertical:medium\n");
426 case POLARIZATION_VH:
427 count = sprintf(buf, "vertical:high\n");
430 count = sprintf(buf, "unknown\n");
433 mutex_unlock(&smp_cpu_state_mutex);
436 static DEVICE_ATTR(polarization, 0444, cpu_polarization_show, NULL);
438 static struct attribute *topology_cpu_attrs[] = {
439 &dev_attr_polarization.attr,
443 static struct attribute_group topology_cpu_attr_group = {
444 .attrs = topology_cpu_attrs,
447 static ssize_t cpu_dedicated_show(struct device *dev,
448 struct device_attribute *attr, char *buf)
453 mutex_lock(&smp_cpu_state_mutex);
454 count = sprintf(buf, "%d\n", topology_cpu_dedicated(cpu));
455 mutex_unlock(&smp_cpu_state_mutex);
458 static DEVICE_ATTR(dedicated, 0444, cpu_dedicated_show, NULL);
460 static struct attribute *topology_extra_cpu_attrs[] = {
461 &dev_attr_dedicated.attr,
465 static struct attribute_group topology_extra_cpu_attr_group = {
466 .attrs = topology_extra_cpu_attrs,
469 int topology_cpu_init(struct cpu *cpu)
473 rc = sysfs_create_group(&cpu->dev.kobj, &topology_cpu_attr_group);
474 if (rc || !MACHINE_HAS_TOPOLOGY)
476 rc = sysfs_create_group(&cpu->dev.kobj, &topology_extra_cpu_attr_group);
478 sysfs_remove_group(&cpu->dev.kobj, &topology_cpu_attr_group);
482 static const struct cpumask *cpu_thread_mask(int cpu)
484 return &cpu_topology[cpu].thread_mask;
488 const struct cpumask *cpu_coregroup_mask(int cpu)
490 return &cpu_topology[cpu].core_mask;
493 static const struct cpumask *cpu_book_mask(int cpu)
495 return &cpu_topology[cpu].book_mask;
498 static const struct cpumask *cpu_drawer_mask(int cpu)
500 return &cpu_topology[cpu].drawer_mask;
503 static struct sched_domain_topology_level s390_topology[] = {
504 { cpu_thread_mask, cpu_smt_flags, SD_INIT_NAME(SMT) },
505 { cpu_coregroup_mask, cpu_core_flags, SD_INIT_NAME(MC) },
506 { cpu_book_mask, SD_INIT_NAME(BOOK) },
507 { cpu_drawer_mask, SD_INIT_NAME(DRAWER) },
508 { cpu_cpu_mask, SD_INIT_NAME(DIE) },
512 static void __init alloc_masks(struct sysinfo_15_1_x *info,
513 struct mask_info *mask, int offset)
517 nr_masks = info->mag[TOPOLOGY_NR_MAG - offset];
518 for (i = 0; i < info->mnest - offset; i++)
519 nr_masks *= info->mag[TOPOLOGY_NR_MAG - offset - 1 - i];
520 nr_masks = max(nr_masks, 1);
521 for (i = 0; i < nr_masks; i++) {
522 mask->next = memblock_alloc(sizeof(*mask->next), 8);
524 panic("%s: Failed to allocate %zu bytes align=0x%x\n",
525 __func__, sizeof(*mask->next), 8);
530 void __init topology_init_early(void)
532 struct sysinfo_15_1_x *info;
534 set_sched_topology(s390_topology);
535 if (topology_mode == TOPOLOGY_MODE_UNINITIALIZED) {
536 if (MACHINE_HAS_TOPOLOGY)
537 topology_mode = TOPOLOGY_MODE_HW;
539 topology_mode = TOPOLOGY_MODE_SINGLE;
541 if (!MACHINE_HAS_TOPOLOGY)
543 tl_info = memblock_alloc(PAGE_SIZE, PAGE_SIZE);
545 panic("%s: Failed to allocate %lu bytes align=0x%lx\n",
546 __func__, PAGE_SIZE, PAGE_SIZE);
548 store_topology(info);
549 pr_info("The CPU configuration topology of the machine is: %d %d %d %d %d %d / %d\n",
550 info->mag[0], info->mag[1], info->mag[2], info->mag[3],
551 info->mag[4], info->mag[5], info->mnest);
552 alloc_masks(info, &socket_info, 1);
553 alloc_masks(info, &book_info, 2);
554 alloc_masks(info, &drawer_info, 3);
556 __arch_update_cpu_topology();
557 __arch_update_dedicated_flag(NULL);
560 static inline int topology_get_mode(int enabled)
563 return TOPOLOGY_MODE_SINGLE;
564 return MACHINE_HAS_TOPOLOGY ? TOPOLOGY_MODE_HW : TOPOLOGY_MODE_PACKAGE;
567 static inline int topology_is_enabled(void)
569 return topology_mode != TOPOLOGY_MODE_SINGLE;
572 static int __init topology_setup(char *str)
577 rc = kstrtobool(str, &enabled);
580 topology_mode = topology_get_mode(enabled);
583 early_param("topology", topology_setup);
585 static int topology_ctl_handler(struct ctl_table *ctl, int write,
586 void __user *buffer, size_t *lenp, loff_t *ppos)
588 int enabled = topology_is_enabled();
591 struct ctl_table ctl_entry = {
592 .procname = ctl->procname,
594 .maxlen = sizeof(int),
595 .extra1 = SYSCTL_ZERO,
596 .extra2 = SYSCTL_ONE,
599 rc = proc_douintvec_minmax(&ctl_entry, write, buffer, lenp, ppos);
600 if (rc < 0 || !write)
603 mutex_lock(&smp_cpu_state_mutex);
604 new_mode = topology_get_mode(enabled);
605 if (topology_mode != new_mode) {
606 topology_mode = new_mode;
607 topology_schedule_update();
609 mutex_unlock(&smp_cpu_state_mutex);
610 topology_flush_work();
615 static struct ctl_table topology_ctl_table[] = {
617 .procname = "topology",
619 .proc_handler = topology_ctl_handler,
624 static struct ctl_table topology_dir_table[] = {
629 .child = topology_ctl_table,
634 static int __init topology_init(void)
636 timer_setup(&topology_timer, topology_timer_fn, TIMER_DEFERRABLE);
637 if (MACHINE_HAS_TOPOLOGY)
638 set_topology_timer();
640 topology_update_polarization_simple();
641 register_sysctl_table(topology_dir_table);
642 return device_create_file(cpu_subsys.dev_root, &dev_attr_dispatching);
644 device_initcall(topology_init);