2 * 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/bootmem.h>
11 #include <linux/cpuset.h>
12 #include <linux/device.h>
13 #include <linux/export.h>
14 #include <linux/kernel.h>
15 #include <linux/sched.h>
16 #include <linux/sched/topology.h>
17 #include <linux/delay.h>
18 #include <linux/init.h>
19 #include <linux/slab.h>
20 #include <linux/cpu.h>
21 #include <linux/smp.h>
23 #include <linux/nodemask.h>
24 #include <linux/node.h>
25 #include <asm/sysinfo.h>
28 #define PTF_HORIZONTAL (0UL)
29 #define PTF_VERTICAL (1UL)
30 #define PTF_CHECK (2UL)
33 struct mask_info *next;
38 static void set_topology_timer(void);
39 static void topology_work_fn(struct work_struct *work);
40 static struct sysinfo_15_1_x *tl_info;
42 static DECLARE_WORK(topology_work, topology_work_fn);
45 * Socket/Book linked lists and cpu_topology updates are
46 * protected by "sched_domains_mutex".
48 static struct mask_info socket_info;
49 static struct mask_info book_info;
50 static struct mask_info drawer_info;
52 struct cpu_topology_s390 cpu_topology[NR_CPUS];
53 EXPORT_SYMBOL_GPL(cpu_topology);
55 cpumask_t cpus_with_topology;
57 static cpumask_t cpu_group_map(struct mask_info *info, unsigned int cpu)
61 cpumask_copy(&mask, cpumask_of(cpu));
62 if (!MACHINE_HAS_TOPOLOGY)
64 for (; info; info = info->next) {
65 if (cpumask_test_cpu(cpu, &info->mask))
71 static cpumask_t cpu_thread_map(unsigned int cpu)
76 cpumask_copy(&mask, cpumask_of(cpu));
77 if (!MACHINE_HAS_TOPOLOGY)
79 cpu -= cpu % (smp_cpu_mtid + 1);
80 for (i = 0; i <= smp_cpu_mtid; i++)
81 if (cpu_present(cpu + i))
82 cpumask_set_cpu(cpu + i, &mask);
86 #define TOPOLOGY_CORE_BITS 64
88 static void add_cpus_to_mask(struct topology_core *tl_core,
89 struct mask_info *drawer,
90 struct mask_info *book,
91 struct mask_info *socket)
93 struct cpu_topology_s390 *topo;
96 for_each_set_bit(core, &tl_core->mask, TOPOLOGY_CORE_BITS) {
100 rcore = TOPOLOGY_CORE_BITS - 1 - core + tl_core->origin;
101 lcpu = smp_find_processor_id(rcore << smp_cpu_mt_shift);
104 for (i = 0; i <= smp_cpu_mtid; i++) {
105 topo = &cpu_topology[lcpu + i];
106 topo->drawer_id = drawer->id;
107 topo->book_id = book->id;
108 topo->socket_id = socket->id;
109 topo->core_id = rcore;
110 topo->thread_id = lcpu + i;
111 cpumask_set_cpu(lcpu + i, &drawer->mask);
112 cpumask_set_cpu(lcpu + i, &book->mask);
113 cpumask_set_cpu(lcpu + i, &socket->mask);
114 cpumask_set_cpu(lcpu + i, &cpus_with_topology);
115 smp_cpu_set_polarization(lcpu + i, tl_core->pp);
120 static void clear_masks(void)
122 struct mask_info *info;
126 cpumask_clear(&info->mask);
131 cpumask_clear(&info->mask);
136 cpumask_clear(&info->mask);
141 static union topology_entry *next_tle(union topology_entry *tle)
144 return (union topology_entry *)((struct topology_core *)tle + 1);
145 return (union topology_entry *)((struct topology_container *)tle + 1);
148 static void tl_to_masks(struct sysinfo_15_1_x *info)
150 struct mask_info *socket = &socket_info;
151 struct mask_info *book = &book_info;
152 struct mask_info *drawer = &drawer_info;
153 union topology_entry *tle, *end;
157 end = (union topology_entry *)((unsigned long)info + info->length);
161 drawer = drawer->next;
162 drawer->id = tle->container.id;
166 book->id = tle->container.id;
169 socket = socket->next;
170 socket->id = tle->container.id;
173 add_cpus_to_mask(&tle->cpu, drawer, book, socket);
183 static void topology_update_polarization_simple(void)
187 mutex_lock(&smp_cpu_state_mutex);
188 for_each_possible_cpu(cpu)
189 smp_cpu_set_polarization(cpu, POLARIZATION_HRZ);
190 mutex_unlock(&smp_cpu_state_mutex);
193 static int ptf(unsigned long fc)
198 " .insn rre,0xb9a20000,%1,%1\n"
206 int topology_set_cpu_management(int fc)
210 if (!MACHINE_HAS_TOPOLOGY)
213 rc = ptf(PTF_VERTICAL);
215 rc = ptf(PTF_HORIZONTAL);
218 for_each_possible_cpu(cpu)
219 smp_cpu_set_polarization(cpu, POLARIZATION_UNKNOWN);
223 static void update_cpu_masks(void)
225 struct cpu_topology_s390 *topo;
228 for_each_possible_cpu(cpu) {
229 topo = &cpu_topology[cpu];
230 topo->thread_mask = cpu_thread_map(cpu);
231 topo->core_mask = cpu_group_map(&socket_info, cpu);
232 topo->book_mask = cpu_group_map(&book_info, cpu);
233 topo->drawer_mask = cpu_group_map(&drawer_info, cpu);
234 if (!MACHINE_HAS_TOPOLOGY) {
235 topo->thread_id = cpu;
237 topo->socket_id = cpu;
239 topo->drawer_id = cpu;
240 if (cpu_present(cpu))
241 cpumask_set_cpu(cpu, &cpus_with_topology);
244 numa_update_cpu_topology();
247 void store_topology(struct sysinfo_15_1_x *info)
249 stsi(info, 15, 1, topology_mnest_limit());
252 static int __arch_update_cpu_topology(void)
254 struct sysinfo_15_1_x *info = tl_info;
257 cpumask_clear(&cpus_with_topology);
258 if (MACHINE_HAS_TOPOLOGY) {
260 store_topology(info);
264 if (!MACHINE_HAS_TOPOLOGY)
265 topology_update_polarization_simple();
269 int arch_update_cpu_topology(void)
274 rc = __arch_update_cpu_topology();
275 for_each_online_cpu(cpu) {
276 dev = get_cpu_device(cpu);
277 kobject_uevent(&dev->kobj, KOBJ_CHANGE);
282 static void topology_work_fn(struct work_struct *work)
284 rebuild_sched_domains();
287 void topology_schedule_update(void)
289 schedule_work(&topology_work);
292 static void topology_timer_fn(unsigned long ignored)
295 topology_schedule_update();
296 set_topology_timer();
299 static struct timer_list topology_timer =
300 TIMER_DEFERRED_INITIALIZER(topology_timer_fn, 0, 0);
302 static atomic_t topology_poll = ATOMIC_INIT(0);
304 static void set_topology_timer(void)
306 if (atomic_add_unless(&topology_poll, -1, 0))
307 mod_timer(&topology_timer, jiffies + HZ / 10);
309 mod_timer(&topology_timer, jiffies + HZ * 60);
312 void topology_expect_change(void)
314 if (!MACHINE_HAS_TOPOLOGY)
316 /* This is racy, but it doesn't matter since it is just a heuristic.
317 * Worst case is that we poll in a higher frequency for a bit longer.
319 if (atomic_read(&topology_poll) > 60)
321 atomic_add(60, &topology_poll);
322 set_topology_timer();
325 static int cpu_management;
327 static ssize_t dispatching_show(struct device *dev,
328 struct device_attribute *attr,
333 mutex_lock(&smp_cpu_state_mutex);
334 count = sprintf(buf, "%d\n", cpu_management);
335 mutex_unlock(&smp_cpu_state_mutex);
339 static ssize_t dispatching_store(struct device *dev,
340 struct device_attribute *attr,
347 if (sscanf(buf, "%d %c", &val, &delim) != 1)
349 if (val != 0 && val != 1)
353 mutex_lock(&smp_cpu_state_mutex);
354 if (cpu_management == val)
356 rc = topology_set_cpu_management(val);
359 cpu_management = val;
360 topology_expect_change();
362 mutex_unlock(&smp_cpu_state_mutex);
364 return rc ? rc : count;
366 static DEVICE_ATTR(dispatching, 0644, dispatching_show,
369 static ssize_t cpu_polarization_show(struct device *dev,
370 struct device_attribute *attr, char *buf)
375 mutex_lock(&smp_cpu_state_mutex);
376 switch (smp_cpu_get_polarization(cpu)) {
377 case POLARIZATION_HRZ:
378 count = sprintf(buf, "horizontal\n");
380 case POLARIZATION_VL:
381 count = sprintf(buf, "vertical:low\n");
383 case POLARIZATION_VM:
384 count = sprintf(buf, "vertical:medium\n");
386 case POLARIZATION_VH:
387 count = sprintf(buf, "vertical:high\n");
390 count = sprintf(buf, "unknown\n");
393 mutex_unlock(&smp_cpu_state_mutex);
396 static DEVICE_ATTR(polarization, 0444, cpu_polarization_show, NULL);
398 static struct attribute *topology_cpu_attrs[] = {
399 &dev_attr_polarization.attr,
403 static struct attribute_group topology_cpu_attr_group = {
404 .attrs = topology_cpu_attrs,
407 int topology_cpu_init(struct cpu *cpu)
409 return sysfs_create_group(&cpu->dev.kobj, &topology_cpu_attr_group);
412 static const struct cpumask *cpu_thread_mask(int cpu)
414 return &cpu_topology[cpu].thread_mask;
418 const struct cpumask *cpu_coregroup_mask(int cpu)
420 return &cpu_topology[cpu].core_mask;
423 static const struct cpumask *cpu_book_mask(int cpu)
425 return &cpu_topology[cpu].book_mask;
428 static const struct cpumask *cpu_drawer_mask(int cpu)
430 return &cpu_topology[cpu].drawer_mask;
433 static struct sched_domain_topology_level s390_topology[] = {
434 { cpu_thread_mask, cpu_smt_flags, SD_INIT_NAME(SMT) },
435 { cpu_coregroup_mask, cpu_core_flags, SD_INIT_NAME(MC) },
436 { cpu_book_mask, SD_INIT_NAME(BOOK) },
437 { cpu_drawer_mask, SD_INIT_NAME(DRAWER) },
438 { cpu_cpu_mask, SD_INIT_NAME(DIE) },
442 static void __init alloc_masks(struct sysinfo_15_1_x *info,
443 struct mask_info *mask, int offset)
447 nr_masks = info->mag[TOPOLOGY_NR_MAG - offset];
448 for (i = 0; i < info->mnest - offset; i++)
449 nr_masks *= info->mag[TOPOLOGY_NR_MAG - offset - 1 - i];
450 nr_masks = max(nr_masks, 1);
451 for (i = 0; i < nr_masks; i++) {
452 mask->next = memblock_virt_alloc(sizeof(*mask->next), 8);
457 void __init topology_init_early(void)
459 struct sysinfo_15_1_x *info;
461 set_sched_topology(s390_topology);
462 if (!MACHINE_HAS_TOPOLOGY)
464 tl_info = memblock_virt_alloc(PAGE_SIZE, PAGE_SIZE);
466 store_topology(info);
467 pr_info("The CPU configuration topology of the machine is: %d %d %d %d %d %d / %d\n",
468 info->mag[0], info->mag[1], info->mag[2], info->mag[3],
469 info->mag[4], info->mag[5], info->mnest);
470 alloc_masks(info, &socket_info, 1);
471 alloc_masks(info, &book_info, 2);
472 alloc_masks(info, &drawer_info, 3);
474 __arch_update_cpu_topology();
477 static int __init topology_init(void)
479 if (MACHINE_HAS_TOPOLOGY)
480 set_topology_timer();
482 topology_update_polarization_simple();
483 return device_create_file(cpu_subsys.dev_root, &dev_attr_dispatching);
485 device_initcall(topology_init);