1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright IBM Corp. 2007, 2011
6 #define KMSG_COMPONENT "cpu"
7 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
9 #include <linux/workqueue.h>
10 #include <linux/memblock.h>
11 #include <linux/uaccess.h>
12 #include <linux/sysctl.h>
13 #include <linux/cpuset.h>
14 #include <linux/device.h>
15 #include <linux/export.h>
16 #include <linux/kernel.h>
17 #include <linux/sched.h>
18 #include <linux/sched/topology.h>
19 #include <linux/delay.h>
20 #include <linux/init.h>
21 #include <linux/slab.h>
22 #include <linux/cpu.h>
23 #include <linux/smp.h>
25 #include <linux/nodemask.h>
26 #include <linux/node.h>
27 #include <asm/sysinfo.h>
29 #define PTF_HORIZONTAL (0UL)
30 #define PTF_VERTICAL (1UL)
31 #define PTF_CHECK (2UL)
36 TOPOLOGY_MODE_PACKAGE,
37 TOPOLOGY_MODE_UNINITIALIZED
41 struct mask_info *next;
46 static int topology_mode = TOPOLOGY_MODE_UNINITIALIZED;
47 static void set_topology_timer(void);
48 static void topology_work_fn(struct work_struct *work);
49 static struct sysinfo_15_1_x *tl_info;
51 static DECLARE_WORK(topology_work, topology_work_fn);
54 * Socket/Book linked lists and cpu_topology updates are
55 * protected by "sched_domains_mutex".
57 static struct mask_info socket_info;
58 static struct mask_info book_info;
59 static struct mask_info drawer_info;
61 struct cpu_topology_s390 cpu_topology[NR_CPUS];
62 EXPORT_SYMBOL_GPL(cpu_topology);
64 static void cpu_group_map(cpumask_t *dst, struct mask_info *info, unsigned int cpu)
66 static cpumask_t mask;
69 if (!cpumask_test_cpu(cpu, &cpu_setup_mask))
71 cpumask_set_cpu(cpu, &mask);
72 switch (topology_mode) {
73 case TOPOLOGY_MODE_HW:
75 if (cpumask_test_cpu(cpu, &info->mask)) {
76 cpumask_copy(&mask, &info->mask);
82 case TOPOLOGY_MODE_PACKAGE:
83 cpumask_copy(&mask, cpu_present_mask);
87 case TOPOLOGY_MODE_SINGLE:
90 cpumask_and(&mask, &mask, &cpu_setup_mask);
92 cpumask_copy(dst, &mask);
95 static void cpu_thread_map(cpumask_t *dst, unsigned int cpu)
97 static cpumask_t mask;
100 cpumask_clear(&mask);
101 if (!cpumask_test_cpu(cpu, &cpu_setup_mask))
103 cpumask_set_cpu(cpu, &mask);
104 if (topology_mode != TOPOLOGY_MODE_HW)
106 cpu -= cpu % (smp_cpu_mtid + 1);
107 for (i = 0; i <= smp_cpu_mtid; i++) {
108 if (cpumask_test_cpu(cpu + i, &cpu_setup_mask))
109 cpumask_set_cpu(cpu + i, &mask);
112 cpumask_copy(dst, &mask);
115 #define TOPOLOGY_CORE_BITS 64
117 static void add_cpus_to_mask(struct topology_core *tl_core,
118 struct mask_info *drawer,
119 struct mask_info *book,
120 struct mask_info *socket)
122 struct cpu_topology_s390 *topo;
125 for_each_set_bit(core, &tl_core->mask, TOPOLOGY_CORE_BITS) {
129 rcore = TOPOLOGY_CORE_BITS - 1 - core + tl_core->origin;
130 lcpu = smp_find_processor_id(rcore << smp_cpu_mt_shift);
133 for (i = 0; i <= smp_cpu_mtid; i++) {
134 topo = &cpu_topology[lcpu + i];
135 topo->drawer_id = drawer->id;
136 topo->book_id = book->id;
137 topo->socket_id = socket->id;
138 topo->core_id = rcore;
139 topo->thread_id = lcpu + i;
140 topo->dedicated = tl_core->d;
141 cpumask_set_cpu(lcpu + i, &drawer->mask);
142 cpumask_set_cpu(lcpu + i, &book->mask);
143 cpumask_set_cpu(lcpu + i, &socket->mask);
144 smp_cpu_set_polarization(lcpu + i, tl_core->pp);
149 static void clear_masks(void)
151 struct mask_info *info;
155 cpumask_clear(&info->mask);
160 cpumask_clear(&info->mask);
165 cpumask_clear(&info->mask);
170 static union topology_entry *next_tle(union topology_entry *tle)
173 return (union topology_entry *)((struct topology_core *)tle + 1);
174 return (union topology_entry *)((struct topology_container *)tle + 1);
177 static void tl_to_masks(struct sysinfo_15_1_x *info)
179 struct mask_info *socket = &socket_info;
180 struct mask_info *book = &book_info;
181 struct mask_info *drawer = &drawer_info;
182 union topology_entry *tle, *end;
186 end = (union topology_entry *)((unsigned long)info + info->length);
190 drawer = drawer->next;
191 drawer->id = tle->container.id;
195 book->id = tle->container.id;
198 socket = socket->next;
199 socket->id = tle->container.id;
202 add_cpus_to_mask(&tle->cpu, drawer, book, socket);
212 static void topology_update_polarization_simple(void)
216 for_each_possible_cpu(cpu)
217 smp_cpu_set_polarization(cpu, POLARIZATION_HRZ);
220 static int ptf(unsigned long fc)
225 " .insn rre,0xb9a20000,%1,%1\n"
233 int topology_set_cpu_management(int fc)
237 if (!MACHINE_HAS_TOPOLOGY)
240 rc = ptf(PTF_VERTICAL);
242 rc = ptf(PTF_HORIZONTAL);
245 for_each_possible_cpu(cpu)
246 smp_cpu_set_polarization(cpu, POLARIZATION_UNKNOWN);
250 void update_cpu_masks(void)
252 struct cpu_topology_s390 *topo, *topo_package, *topo_sibling;
253 int cpu, sibling, pkg_first, smt_first, id;
255 for_each_possible_cpu(cpu) {
256 topo = &cpu_topology[cpu];
257 cpu_thread_map(&topo->thread_mask, cpu);
258 cpu_group_map(&topo->core_mask, &socket_info, cpu);
259 cpu_group_map(&topo->book_mask, &book_info, cpu);
260 cpu_group_map(&topo->drawer_mask, &drawer_info, cpu);
261 topo->booted_cores = 0;
262 if (topology_mode != TOPOLOGY_MODE_HW) {
263 id = topology_mode == TOPOLOGY_MODE_PACKAGE ? 0 : cpu;
264 topo->thread_id = cpu;
266 topo->socket_id = id;
268 topo->drawer_id = id;
271 for_each_online_cpu(cpu) {
272 topo = &cpu_topology[cpu];
273 pkg_first = cpumask_first(&topo->core_mask);
274 topo_package = &cpu_topology[pkg_first];
275 if (cpu == pkg_first) {
276 for_each_cpu(sibling, &topo->core_mask) {
277 topo_sibling = &cpu_topology[sibling];
278 smt_first = cpumask_first(&topo_sibling->thread_mask);
279 if (sibling == smt_first)
280 topo_package->booted_cores++;
283 topo->booted_cores = topo_package->booted_cores;
288 void store_topology(struct sysinfo_15_1_x *info)
290 stsi(info, 15, 1, topology_mnest_limit());
293 static void __arch_update_dedicated_flag(void *arg)
295 if (topology_cpu_dedicated(smp_processor_id()))
296 set_cpu_flag(CIF_DEDICATED_CPU);
298 clear_cpu_flag(CIF_DEDICATED_CPU);
301 static int __arch_update_cpu_topology(void)
303 struct sysinfo_15_1_x *info = tl_info;
306 mutex_lock(&smp_cpu_state_mutex);
307 if (MACHINE_HAS_TOPOLOGY) {
309 store_topology(info);
313 if (!MACHINE_HAS_TOPOLOGY)
314 topology_update_polarization_simple();
315 mutex_unlock(&smp_cpu_state_mutex);
319 int arch_update_cpu_topology(void)
324 rc = __arch_update_cpu_topology();
325 on_each_cpu(__arch_update_dedicated_flag, NULL, 0);
326 for_each_online_cpu(cpu) {
327 dev = get_cpu_device(cpu);
329 kobject_uevent(&dev->kobj, KOBJ_CHANGE);
334 static void topology_work_fn(struct work_struct *work)
336 rebuild_sched_domains();
339 void topology_schedule_update(void)
341 schedule_work(&topology_work);
344 static void topology_flush_work(void)
346 flush_work(&topology_work);
349 static void topology_timer_fn(struct timer_list *unused)
352 topology_schedule_update();
353 set_topology_timer();
356 static struct timer_list topology_timer;
358 static atomic_t topology_poll = ATOMIC_INIT(0);
360 static void set_topology_timer(void)
362 if (atomic_add_unless(&topology_poll, -1, 0))
363 mod_timer(&topology_timer, jiffies + msecs_to_jiffies(100));
365 mod_timer(&topology_timer, jiffies + msecs_to_jiffies(60 * MSEC_PER_SEC));
368 void topology_expect_change(void)
370 if (!MACHINE_HAS_TOPOLOGY)
372 /* This is racy, but it doesn't matter since it is just a heuristic.
373 * Worst case is that we poll in a higher frequency for a bit longer.
375 if (atomic_read(&topology_poll) > 60)
377 atomic_add(60, &topology_poll);
378 set_topology_timer();
381 static int cpu_management;
383 static ssize_t dispatching_show(struct device *dev,
384 struct device_attribute *attr,
389 mutex_lock(&smp_cpu_state_mutex);
390 count = sprintf(buf, "%d\n", cpu_management);
391 mutex_unlock(&smp_cpu_state_mutex);
395 static ssize_t dispatching_store(struct device *dev,
396 struct device_attribute *attr,
403 if (sscanf(buf, "%d %c", &val, &delim) != 1)
405 if (val != 0 && val != 1)
409 mutex_lock(&smp_cpu_state_mutex);
410 if (cpu_management == val)
412 rc = topology_set_cpu_management(val);
415 cpu_management = val;
416 topology_expect_change();
418 mutex_unlock(&smp_cpu_state_mutex);
420 return rc ? rc : count;
422 static DEVICE_ATTR_RW(dispatching);
424 static ssize_t cpu_polarization_show(struct device *dev,
425 struct device_attribute *attr, char *buf)
430 mutex_lock(&smp_cpu_state_mutex);
431 switch (smp_cpu_get_polarization(cpu)) {
432 case POLARIZATION_HRZ:
433 count = sprintf(buf, "horizontal\n");
435 case POLARIZATION_VL:
436 count = sprintf(buf, "vertical:low\n");
438 case POLARIZATION_VM:
439 count = sprintf(buf, "vertical:medium\n");
441 case POLARIZATION_VH:
442 count = sprintf(buf, "vertical:high\n");
445 count = sprintf(buf, "unknown\n");
448 mutex_unlock(&smp_cpu_state_mutex);
451 static DEVICE_ATTR(polarization, 0444, cpu_polarization_show, NULL);
453 static struct attribute *topology_cpu_attrs[] = {
454 &dev_attr_polarization.attr,
458 static struct attribute_group topology_cpu_attr_group = {
459 .attrs = topology_cpu_attrs,
462 static ssize_t cpu_dedicated_show(struct device *dev,
463 struct device_attribute *attr, char *buf)
468 mutex_lock(&smp_cpu_state_mutex);
469 count = sprintf(buf, "%d\n", topology_cpu_dedicated(cpu));
470 mutex_unlock(&smp_cpu_state_mutex);
473 static DEVICE_ATTR(dedicated, 0444, cpu_dedicated_show, NULL);
475 static struct attribute *topology_extra_cpu_attrs[] = {
476 &dev_attr_dedicated.attr,
480 static struct attribute_group topology_extra_cpu_attr_group = {
481 .attrs = topology_extra_cpu_attrs,
484 int topology_cpu_init(struct cpu *cpu)
488 rc = sysfs_create_group(&cpu->dev.kobj, &topology_cpu_attr_group);
489 if (rc || !MACHINE_HAS_TOPOLOGY)
491 rc = sysfs_create_group(&cpu->dev.kobj, &topology_extra_cpu_attr_group);
493 sysfs_remove_group(&cpu->dev.kobj, &topology_cpu_attr_group);
497 static const struct cpumask *cpu_thread_mask(int cpu)
499 return &cpu_topology[cpu].thread_mask;
503 const struct cpumask *cpu_coregroup_mask(int cpu)
505 return &cpu_topology[cpu].core_mask;
508 static const struct cpumask *cpu_book_mask(int cpu)
510 return &cpu_topology[cpu].book_mask;
513 static const struct cpumask *cpu_drawer_mask(int cpu)
515 return &cpu_topology[cpu].drawer_mask;
518 static struct sched_domain_topology_level s390_topology[] = {
519 { cpu_thread_mask, cpu_smt_flags, SD_INIT_NAME(SMT) },
520 { cpu_coregroup_mask, cpu_core_flags, SD_INIT_NAME(MC) },
521 { cpu_book_mask, SD_INIT_NAME(BOOK) },
522 { cpu_drawer_mask, SD_INIT_NAME(DRAWER) },
523 { cpu_cpu_mask, SD_INIT_NAME(DIE) },
527 static void __init alloc_masks(struct sysinfo_15_1_x *info,
528 struct mask_info *mask, int offset)
532 nr_masks = info->mag[TOPOLOGY_NR_MAG - offset];
533 for (i = 0; i < info->mnest - offset; i++)
534 nr_masks *= info->mag[TOPOLOGY_NR_MAG - offset - 1 - i];
535 nr_masks = max(nr_masks, 1);
536 for (i = 0; i < nr_masks; i++) {
537 mask->next = memblock_alloc(sizeof(*mask->next), 8);
539 panic("%s: Failed to allocate %zu bytes align=0x%x\n",
540 __func__, sizeof(*mask->next), 8);
545 void __init topology_init_early(void)
547 struct sysinfo_15_1_x *info;
549 set_sched_topology(s390_topology);
550 if (topology_mode == TOPOLOGY_MODE_UNINITIALIZED) {
551 if (MACHINE_HAS_TOPOLOGY)
552 topology_mode = TOPOLOGY_MODE_HW;
554 topology_mode = TOPOLOGY_MODE_SINGLE;
556 if (!MACHINE_HAS_TOPOLOGY)
558 tl_info = memblock_alloc(PAGE_SIZE, PAGE_SIZE);
560 panic("%s: Failed to allocate %lu bytes align=0x%lx\n",
561 __func__, PAGE_SIZE, PAGE_SIZE);
563 store_topology(info);
564 pr_info("The CPU configuration topology of the machine is: %d %d %d %d %d %d / %d\n",
565 info->mag[0], info->mag[1], info->mag[2], info->mag[3],
566 info->mag[4], info->mag[5], info->mnest);
567 alloc_masks(info, &socket_info, 1);
568 alloc_masks(info, &book_info, 2);
569 alloc_masks(info, &drawer_info, 3);
571 cpumask_set_cpu(0, &cpu_setup_mask);
572 __arch_update_cpu_topology();
573 __arch_update_dedicated_flag(NULL);
576 static inline int topology_get_mode(int enabled)
579 return TOPOLOGY_MODE_SINGLE;
580 return MACHINE_HAS_TOPOLOGY ? TOPOLOGY_MODE_HW : TOPOLOGY_MODE_PACKAGE;
583 static inline int topology_is_enabled(void)
585 return topology_mode != TOPOLOGY_MODE_SINGLE;
588 static int __init topology_setup(char *str)
593 rc = kstrtobool(str, &enabled);
596 topology_mode = topology_get_mode(enabled);
599 early_param("topology", topology_setup);
601 static int topology_ctl_handler(struct ctl_table *ctl, int write,
602 void *buffer, size_t *lenp, loff_t *ppos)
604 int enabled = topology_is_enabled();
607 struct ctl_table ctl_entry = {
608 .procname = ctl->procname,
610 .maxlen = sizeof(int),
611 .extra1 = SYSCTL_ZERO,
612 .extra2 = SYSCTL_ONE,
615 rc = proc_douintvec_minmax(&ctl_entry, write, buffer, lenp, ppos);
616 if (rc < 0 || !write)
619 mutex_lock(&smp_cpu_state_mutex);
620 new_mode = topology_get_mode(enabled);
621 if (topology_mode != new_mode) {
622 topology_mode = new_mode;
623 topology_schedule_update();
625 mutex_unlock(&smp_cpu_state_mutex);
626 topology_flush_work();
631 static struct ctl_table topology_ctl_table[] = {
633 .procname = "topology",
635 .proc_handler = topology_ctl_handler,
640 static struct ctl_table topology_dir_table[] = {
645 .child = topology_ctl_table,
650 static int __init topology_init(void)
652 timer_setup(&topology_timer, topology_timer_fn, TIMER_DEFERRABLE);
653 if (MACHINE_HAS_TOPOLOGY)
654 set_topology_timer();
656 topology_update_polarization_simple();
657 register_sysctl_table(topology_dir_table);
658 return device_create_file(cpu_subsys.dev_root, &dev_attr_dispatching);
660 device_initcall(topology_init);