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>
30 #define PTF_HORIZONTAL (0UL)
31 #define PTF_VERTICAL (1UL)
32 #define PTF_CHECK (2UL)
37 TOPOLOGY_MODE_PACKAGE,
38 TOPOLOGY_MODE_UNINITIALIZED
42 struct mask_info *next;
47 static int topology_mode = TOPOLOGY_MODE_UNINITIALIZED;
48 static void set_topology_timer(void);
49 static void topology_work_fn(struct work_struct *work);
50 static struct sysinfo_15_1_x *tl_info;
52 static DECLARE_WORK(topology_work, topology_work_fn);
55 * Socket/Book linked lists and cpu_topology updates are
56 * protected by "sched_domains_mutex".
58 static struct mask_info socket_info;
59 static struct mask_info book_info;
60 static struct mask_info drawer_info;
62 struct cpu_topology_s390 cpu_topology[NR_CPUS];
63 EXPORT_SYMBOL_GPL(cpu_topology);
65 static void cpu_group_map(cpumask_t *dst, struct mask_info *info, unsigned int cpu)
67 static cpumask_t mask;
69 cpumask_copy(&mask, cpumask_of(cpu));
70 switch (topology_mode) {
71 case TOPOLOGY_MODE_HW:
73 if (cpumask_test_cpu(cpu, &info->mask)) {
74 cpumask_copy(&mask, &info->mask);
80 case TOPOLOGY_MODE_PACKAGE:
81 cpumask_copy(&mask, cpu_present_mask);
85 case TOPOLOGY_MODE_SINGLE:
86 cpumask_copy(&mask, cpumask_of(cpu));
89 cpumask_and(&mask, &mask, cpu_online_mask);
90 cpumask_copy(dst, &mask);
93 static void cpu_thread_map(cpumask_t *dst, unsigned int cpu)
95 static cpumask_t mask;
98 cpumask_copy(&mask, cpumask_of(cpu));
99 if (topology_mode != TOPOLOGY_MODE_HW)
101 cpu -= cpu % (smp_cpu_mtid + 1);
102 for (i = 0; i <= smp_cpu_mtid; i++)
103 if (cpu_present(cpu + i))
104 cpumask_set_cpu(cpu + i, &mask);
105 cpumask_and(&mask, &mask, cpu_online_mask);
107 cpumask_copy(dst, &mask);
110 #define TOPOLOGY_CORE_BITS 64
112 static void add_cpus_to_mask(struct topology_core *tl_core,
113 struct mask_info *drawer,
114 struct mask_info *book,
115 struct mask_info *socket)
117 struct cpu_topology_s390 *topo;
120 for_each_set_bit(core, &tl_core->mask, TOPOLOGY_CORE_BITS) {
124 rcore = TOPOLOGY_CORE_BITS - 1 - core + tl_core->origin;
125 lcpu = smp_find_processor_id(rcore << smp_cpu_mt_shift);
128 for (i = 0; i <= smp_cpu_mtid; i++) {
129 topo = &cpu_topology[lcpu + i];
130 topo->drawer_id = drawer->id;
131 topo->book_id = book->id;
132 topo->socket_id = socket->id;
133 topo->core_id = rcore;
134 topo->thread_id = lcpu + i;
135 topo->dedicated = tl_core->d;
136 cpumask_set_cpu(lcpu + i, &drawer->mask);
137 cpumask_set_cpu(lcpu + i, &book->mask);
138 cpumask_set_cpu(lcpu + i, &socket->mask);
139 smp_cpu_set_polarization(lcpu + i, tl_core->pp);
144 static void clear_masks(void)
146 struct mask_info *info;
150 cpumask_clear(&info->mask);
155 cpumask_clear(&info->mask);
160 cpumask_clear(&info->mask);
165 static union topology_entry *next_tle(union topology_entry *tle)
168 return (union topology_entry *)((struct topology_core *)tle + 1);
169 return (union topology_entry *)((struct topology_container *)tle + 1);
172 static void tl_to_masks(struct sysinfo_15_1_x *info)
174 struct mask_info *socket = &socket_info;
175 struct mask_info *book = &book_info;
176 struct mask_info *drawer = &drawer_info;
177 union topology_entry *tle, *end;
181 end = (union topology_entry *)((unsigned long)info + info->length);
185 drawer = drawer->next;
186 drawer->id = tle->container.id;
190 book->id = tle->container.id;
193 socket = socket->next;
194 socket->id = tle->container.id;
197 add_cpus_to_mask(&tle->cpu, drawer, book, socket);
207 static void topology_update_polarization_simple(void)
211 for_each_possible_cpu(cpu)
212 smp_cpu_set_polarization(cpu, POLARIZATION_HRZ);
215 static int ptf(unsigned long fc)
220 " .insn rre,0xb9a20000,%1,%1\n"
228 int topology_set_cpu_management(int fc)
232 if (!MACHINE_HAS_TOPOLOGY)
235 rc = ptf(PTF_VERTICAL);
237 rc = ptf(PTF_HORIZONTAL);
240 for_each_possible_cpu(cpu)
241 smp_cpu_set_polarization(cpu, POLARIZATION_UNKNOWN);
245 void update_cpu_masks(void)
247 struct cpu_topology_s390 *topo, *topo_package, *topo_sibling;
248 int cpu, sibling, pkg_first, smt_first, id;
250 for_each_possible_cpu(cpu) {
251 topo = &cpu_topology[cpu];
252 cpu_thread_map(&topo->thread_mask, cpu);
253 cpu_group_map(&topo->core_mask, &socket_info, cpu);
254 cpu_group_map(&topo->book_mask, &book_info, cpu);
255 cpu_group_map(&topo->drawer_mask, &drawer_info, cpu);
256 topo->booted_cores = 0;
257 if (topology_mode != TOPOLOGY_MODE_HW) {
258 id = topology_mode == TOPOLOGY_MODE_PACKAGE ? 0 : cpu;
259 topo->thread_id = cpu;
261 topo->socket_id = id;
263 topo->drawer_id = id;
266 for_each_online_cpu(cpu) {
267 topo = &cpu_topology[cpu];
268 pkg_first = cpumask_first(&topo->core_mask);
269 topo_package = &cpu_topology[pkg_first];
270 if (cpu == pkg_first) {
271 for_each_cpu(sibling, &topo->core_mask) {
272 topo_sibling = &cpu_topology[sibling];
273 smt_first = cpumask_first(&topo_sibling->thread_mask);
274 if (sibling == smt_first)
275 topo_package->booted_cores++;
278 topo->booted_cores = topo_package->booted_cores;
283 void store_topology(struct sysinfo_15_1_x *info)
285 stsi(info, 15, 1, topology_mnest_limit());
288 static void __arch_update_dedicated_flag(void *arg)
290 if (topology_cpu_dedicated(smp_processor_id()))
291 set_cpu_flag(CIF_DEDICATED_CPU);
293 clear_cpu_flag(CIF_DEDICATED_CPU);
296 static int __arch_update_cpu_topology(void)
298 struct sysinfo_15_1_x *info = tl_info;
301 mutex_lock(&smp_cpu_state_mutex);
302 if (MACHINE_HAS_TOPOLOGY) {
304 store_topology(info);
308 if (!MACHINE_HAS_TOPOLOGY)
309 topology_update_polarization_simple();
310 mutex_unlock(&smp_cpu_state_mutex);
314 int arch_update_cpu_topology(void)
319 rc = __arch_update_cpu_topology();
320 on_each_cpu(__arch_update_dedicated_flag, NULL, 0);
321 for_each_online_cpu(cpu) {
322 dev = get_cpu_device(cpu);
324 kobject_uevent(&dev->kobj, KOBJ_CHANGE);
329 static void topology_work_fn(struct work_struct *work)
331 rebuild_sched_domains();
334 void topology_schedule_update(void)
336 schedule_work(&topology_work);
339 static void topology_flush_work(void)
341 flush_work(&topology_work);
344 static void topology_timer_fn(struct timer_list *unused)
347 topology_schedule_update();
348 set_topology_timer();
351 static struct timer_list topology_timer;
353 static atomic_t topology_poll = ATOMIC_INIT(0);
355 static void set_topology_timer(void)
357 if (atomic_add_unless(&topology_poll, -1, 0))
358 mod_timer(&topology_timer, jiffies + msecs_to_jiffies(100));
360 mod_timer(&topology_timer, jiffies + msecs_to_jiffies(60 * MSEC_PER_SEC));
363 void topology_expect_change(void)
365 if (!MACHINE_HAS_TOPOLOGY)
367 /* This is racy, but it doesn't matter since it is just a heuristic.
368 * Worst case is that we poll in a higher frequency for a bit longer.
370 if (atomic_read(&topology_poll) > 60)
372 atomic_add(60, &topology_poll);
373 set_topology_timer();
376 static int cpu_management;
378 static ssize_t dispatching_show(struct device *dev,
379 struct device_attribute *attr,
384 mutex_lock(&smp_cpu_state_mutex);
385 count = sprintf(buf, "%d\n", cpu_management);
386 mutex_unlock(&smp_cpu_state_mutex);
390 static ssize_t dispatching_store(struct device *dev,
391 struct device_attribute *attr,
398 if (sscanf(buf, "%d %c", &val, &delim) != 1)
400 if (val != 0 && val != 1)
404 mutex_lock(&smp_cpu_state_mutex);
405 if (cpu_management == val)
407 rc = topology_set_cpu_management(val);
410 cpu_management = val;
411 topology_expect_change();
413 mutex_unlock(&smp_cpu_state_mutex);
415 return rc ? rc : count;
417 static DEVICE_ATTR_RW(dispatching);
419 static ssize_t cpu_polarization_show(struct device *dev,
420 struct device_attribute *attr, char *buf)
425 mutex_lock(&smp_cpu_state_mutex);
426 switch (smp_cpu_get_polarization(cpu)) {
427 case POLARIZATION_HRZ:
428 count = sprintf(buf, "horizontal\n");
430 case POLARIZATION_VL:
431 count = sprintf(buf, "vertical:low\n");
433 case POLARIZATION_VM:
434 count = sprintf(buf, "vertical:medium\n");
436 case POLARIZATION_VH:
437 count = sprintf(buf, "vertical:high\n");
440 count = sprintf(buf, "unknown\n");
443 mutex_unlock(&smp_cpu_state_mutex);
446 static DEVICE_ATTR(polarization, 0444, cpu_polarization_show, NULL);
448 static struct attribute *topology_cpu_attrs[] = {
449 &dev_attr_polarization.attr,
453 static struct attribute_group topology_cpu_attr_group = {
454 .attrs = topology_cpu_attrs,
457 static ssize_t cpu_dedicated_show(struct device *dev,
458 struct device_attribute *attr, char *buf)
463 mutex_lock(&smp_cpu_state_mutex);
464 count = sprintf(buf, "%d\n", topology_cpu_dedicated(cpu));
465 mutex_unlock(&smp_cpu_state_mutex);
468 static DEVICE_ATTR(dedicated, 0444, cpu_dedicated_show, NULL);
470 static struct attribute *topology_extra_cpu_attrs[] = {
471 &dev_attr_dedicated.attr,
475 static struct attribute_group topology_extra_cpu_attr_group = {
476 .attrs = topology_extra_cpu_attrs,
479 int topology_cpu_init(struct cpu *cpu)
483 rc = sysfs_create_group(&cpu->dev.kobj, &topology_cpu_attr_group);
484 if (rc || !MACHINE_HAS_TOPOLOGY)
486 rc = sysfs_create_group(&cpu->dev.kobj, &topology_extra_cpu_attr_group);
488 sysfs_remove_group(&cpu->dev.kobj, &topology_cpu_attr_group);
492 static const struct cpumask *cpu_thread_mask(int cpu)
494 return &cpu_topology[cpu].thread_mask;
498 const struct cpumask *cpu_coregroup_mask(int cpu)
500 return &cpu_topology[cpu].core_mask;
503 static const struct cpumask *cpu_book_mask(int cpu)
505 return &cpu_topology[cpu].book_mask;
508 static const struct cpumask *cpu_drawer_mask(int cpu)
510 return &cpu_topology[cpu].drawer_mask;
513 static struct sched_domain_topology_level s390_topology[] = {
514 { cpu_thread_mask, cpu_smt_flags, SD_INIT_NAME(SMT) },
515 { cpu_coregroup_mask, cpu_core_flags, SD_INIT_NAME(MC) },
516 { cpu_book_mask, SD_INIT_NAME(BOOK) },
517 { cpu_drawer_mask, SD_INIT_NAME(DRAWER) },
518 { cpu_cpu_mask, SD_INIT_NAME(DIE) },
522 static void __init alloc_masks(struct sysinfo_15_1_x *info,
523 struct mask_info *mask, int offset)
527 nr_masks = info->mag[TOPOLOGY_NR_MAG - offset];
528 for (i = 0; i < info->mnest - offset; i++)
529 nr_masks *= info->mag[TOPOLOGY_NR_MAG - offset - 1 - i];
530 nr_masks = max(nr_masks, 1);
531 for (i = 0; i < nr_masks; i++) {
532 mask->next = memblock_alloc(sizeof(*mask->next), 8);
534 panic("%s: Failed to allocate %zu bytes align=0x%x\n",
535 __func__, sizeof(*mask->next), 8);
540 void __init topology_init_early(void)
542 struct sysinfo_15_1_x *info;
544 set_sched_topology(s390_topology);
545 if (topology_mode == TOPOLOGY_MODE_UNINITIALIZED) {
546 if (MACHINE_HAS_TOPOLOGY)
547 topology_mode = TOPOLOGY_MODE_HW;
549 topology_mode = TOPOLOGY_MODE_SINGLE;
551 if (!MACHINE_HAS_TOPOLOGY)
553 tl_info = memblock_alloc(PAGE_SIZE, PAGE_SIZE);
555 panic("%s: Failed to allocate %lu bytes align=0x%lx\n",
556 __func__, PAGE_SIZE, PAGE_SIZE);
558 store_topology(info);
559 pr_info("The CPU configuration topology of the machine is: %d %d %d %d %d %d / %d\n",
560 info->mag[0], info->mag[1], info->mag[2], info->mag[3],
561 info->mag[4], info->mag[5], info->mnest);
562 alloc_masks(info, &socket_info, 1);
563 alloc_masks(info, &book_info, 2);
564 alloc_masks(info, &drawer_info, 3);
566 __arch_update_cpu_topology();
567 __arch_update_dedicated_flag(NULL);
570 static inline int topology_get_mode(int enabled)
573 return TOPOLOGY_MODE_SINGLE;
574 return MACHINE_HAS_TOPOLOGY ? TOPOLOGY_MODE_HW : TOPOLOGY_MODE_PACKAGE;
577 static inline int topology_is_enabled(void)
579 return topology_mode != TOPOLOGY_MODE_SINGLE;
582 static int __init topology_setup(char *str)
587 rc = kstrtobool(str, &enabled);
590 topology_mode = topology_get_mode(enabled);
593 early_param("topology", topology_setup);
595 static int topology_ctl_handler(struct ctl_table *ctl, int write,
596 void *buffer, size_t *lenp, loff_t *ppos)
598 int enabled = topology_is_enabled();
601 struct ctl_table ctl_entry = {
602 .procname = ctl->procname,
604 .maxlen = sizeof(int),
605 .extra1 = SYSCTL_ZERO,
606 .extra2 = SYSCTL_ONE,
609 rc = proc_douintvec_minmax(&ctl_entry, write, buffer, lenp, ppos);
610 if (rc < 0 || !write)
613 mutex_lock(&smp_cpu_state_mutex);
614 new_mode = topology_get_mode(enabled);
615 if (topology_mode != new_mode) {
616 topology_mode = new_mode;
617 topology_schedule_update();
619 mutex_unlock(&smp_cpu_state_mutex);
620 topology_flush_work();
625 static struct ctl_table topology_ctl_table[] = {
627 .procname = "topology",
629 .proc_handler = topology_ctl_handler,
634 static struct ctl_table topology_dir_table[] = {
639 .child = topology_ctl_table,
644 static int __init topology_init(void)
646 timer_setup(&topology_timer, topology_timer_fn, TIMER_DEFERRABLE);
647 if (MACHINE_HAS_TOPOLOGY)
648 set_topology_timer();
650 topology_update_polarization_simple();
651 register_sysctl_table(topology_dir_table);
652 return device_create_file(cpu_subsys.dev_root, &dev_attr_dispatching);
654 device_initcall(topology_init);