Loading...
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Copyright IBM Corp. 2007, 2011
4 */
5
6#define KMSG_COMPONENT "cpu"
7#define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
8
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>
24#include <linux/mm.h>
25#include <linux/nodemask.h>
26#include <linux/node.h>
27#include <asm/hiperdispatch.h>
28#include <asm/sysinfo.h>
29#include <asm/asm.h>
30
31#define PTF_HORIZONTAL (0UL)
32#define PTF_VERTICAL (1UL)
33#define PTF_CHECK (2UL)
34
35enum {
36 TOPOLOGY_MODE_HW,
37 TOPOLOGY_MODE_SINGLE,
38 TOPOLOGY_MODE_PACKAGE,
39 TOPOLOGY_MODE_UNINITIALIZED
40};
41
42struct mask_info {
43 struct mask_info *next;
44 unsigned char id;
45 cpumask_t mask;
46};
47
48static int topology_mode = TOPOLOGY_MODE_UNINITIALIZED;
49static void set_topology_timer(void);
50static void topology_work_fn(struct work_struct *work);
51static struct sysinfo_15_1_x *tl_info;
52static int cpu_management;
53
54static DECLARE_WORK(topology_work, topology_work_fn);
55
56/*
57 * Socket/Book linked lists and cpu_topology updates are
58 * protected by "sched_domains_mutex".
59 */
60static struct mask_info socket_info;
61static struct mask_info book_info;
62static struct mask_info drawer_info;
63
64struct cpu_topology_s390 cpu_topology[NR_CPUS];
65EXPORT_SYMBOL_GPL(cpu_topology);
66
67static void cpu_group_map(cpumask_t *dst, struct mask_info *info, unsigned int cpu)
68{
69 static cpumask_t mask;
70
71 cpumask_clear(&mask);
72 if (!cpumask_test_cpu(cpu, &cpu_setup_mask))
73 goto out;
74 cpumask_set_cpu(cpu, &mask);
75 switch (topology_mode) {
76 case TOPOLOGY_MODE_HW:
77 while (info) {
78 if (cpumask_test_cpu(cpu, &info->mask)) {
79 cpumask_copy(&mask, &info->mask);
80 break;
81 }
82 info = info->next;
83 }
84 break;
85 case TOPOLOGY_MODE_PACKAGE:
86 cpumask_copy(&mask, cpu_present_mask);
87 break;
88 default:
89 fallthrough;
90 case TOPOLOGY_MODE_SINGLE:
91 break;
92 }
93 cpumask_and(&mask, &mask, &cpu_setup_mask);
94out:
95 cpumask_copy(dst, &mask);
96}
97
98static void cpu_thread_map(cpumask_t *dst, unsigned int cpu)
99{
100 static cpumask_t mask;
101 unsigned int max_cpu;
102
103 cpumask_clear(&mask);
104 if (!cpumask_test_cpu(cpu, &cpu_setup_mask))
105 goto out;
106 cpumask_set_cpu(cpu, &mask);
107 if (topology_mode != TOPOLOGY_MODE_HW)
108 goto out;
109 cpu -= cpu % (smp_cpu_mtid + 1);
110 max_cpu = min(cpu + smp_cpu_mtid, nr_cpu_ids - 1);
111 for (; cpu <= max_cpu; cpu++) {
112 if (cpumask_test_cpu(cpu, &cpu_setup_mask))
113 cpumask_set_cpu(cpu, &mask);
114 }
115out:
116 cpumask_copy(dst, &mask);
117}
118
119#define TOPOLOGY_CORE_BITS 64
120
121static void add_cpus_to_mask(struct topology_core *tl_core,
122 struct mask_info *drawer,
123 struct mask_info *book,
124 struct mask_info *socket)
125{
126 struct cpu_topology_s390 *topo;
127 unsigned int core;
128
129 for_each_set_bit(core, &tl_core->mask, TOPOLOGY_CORE_BITS) {
130 unsigned int max_cpu, rcore;
131 int cpu;
132
133 rcore = TOPOLOGY_CORE_BITS - 1 - core + tl_core->origin;
134 cpu = smp_find_processor_id(rcore << smp_cpu_mt_shift);
135 if (cpu < 0)
136 continue;
137 max_cpu = min(cpu + smp_cpu_mtid, nr_cpu_ids - 1);
138 for (; cpu <= max_cpu; cpu++) {
139 topo = &cpu_topology[cpu];
140 topo->drawer_id = drawer->id;
141 topo->book_id = book->id;
142 topo->socket_id = socket->id;
143 topo->core_id = rcore;
144 topo->thread_id = cpu;
145 topo->dedicated = tl_core->d;
146 cpumask_set_cpu(cpu, &drawer->mask);
147 cpumask_set_cpu(cpu, &book->mask);
148 cpumask_set_cpu(cpu, &socket->mask);
149 smp_cpu_set_polarization(cpu, tl_core->pp);
150 smp_cpu_set_capacity(cpu, CPU_CAPACITY_HIGH);
151 }
152 }
153}
154
155static void clear_masks(void)
156{
157 struct mask_info *info;
158
159 info = &socket_info;
160 while (info) {
161 cpumask_clear(&info->mask);
162 info = info->next;
163 }
164 info = &book_info;
165 while (info) {
166 cpumask_clear(&info->mask);
167 info = info->next;
168 }
169 info = &drawer_info;
170 while (info) {
171 cpumask_clear(&info->mask);
172 info = info->next;
173 }
174}
175
176static union topology_entry *next_tle(union topology_entry *tle)
177{
178 if (!tle->nl)
179 return (union topology_entry *)((struct topology_core *)tle + 1);
180 return (union topology_entry *)((struct topology_container *)tle + 1);
181}
182
183static void tl_to_masks(struct sysinfo_15_1_x *info)
184{
185 struct mask_info *socket = &socket_info;
186 struct mask_info *book = &book_info;
187 struct mask_info *drawer = &drawer_info;
188 union topology_entry *tle, *end;
189
190 clear_masks();
191 tle = info->tle;
192 end = (union topology_entry *)((unsigned long)info + info->length);
193 while (tle < end) {
194 switch (tle->nl) {
195 case 3:
196 drawer = drawer->next;
197 drawer->id = tle->container.id;
198 break;
199 case 2:
200 book = book->next;
201 book->id = tle->container.id;
202 break;
203 case 1:
204 socket = socket->next;
205 socket->id = tle->container.id;
206 break;
207 case 0:
208 add_cpus_to_mask(&tle->cpu, drawer, book, socket);
209 break;
210 default:
211 clear_masks();
212 return;
213 }
214 tle = next_tle(tle);
215 }
216}
217
218static void topology_update_polarization_simple(void)
219{
220 int cpu;
221
222 for_each_possible_cpu(cpu)
223 smp_cpu_set_polarization(cpu, POLARIZATION_HRZ);
224}
225
226static int ptf(unsigned long fc)
227{
228 int cc;
229
230 asm volatile(
231 " .insn rre,0xb9a20000,%[fc],%[fc]\n"
232 CC_IPM(cc)
233 : CC_OUT(cc, cc)
234 : [fc] "d" (fc)
235 : CC_CLOBBER);
236 return CC_TRANSFORM(cc);
237}
238
239int topology_set_cpu_management(int fc)
240{
241 int cpu, rc;
242
243 if (!MACHINE_HAS_TOPOLOGY)
244 return -EOPNOTSUPP;
245 if (fc)
246 rc = ptf(PTF_VERTICAL);
247 else
248 rc = ptf(PTF_HORIZONTAL);
249 if (rc)
250 return -EBUSY;
251 for_each_possible_cpu(cpu)
252 smp_cpu_set_polarization(cpu, POLARIZATION_UNKNOWN);
253 return rc;
254}
255
256void update_cpu_masks(void)
257{
258 struct cpu_topology_s390 *topo, *topo_package, *topo_sibling;
259 int cpu, sibling, pkg_first, smt_first, id;
260
261 for_each_possible_cpu(cpu) {
262 topo = &cpu_topology[cpu];
263 cpu_thread_map(&topo->thread_mask, cpu);
264 cpu_group_map(&topo->core_mask, &socket_info, cpu);
265 cpu_group_map(&topo->book_mask, &book_info, cpu);
266 cpu_group_map(&topo->drawer_mask, &drawer_info, cpu);
267 topo->booted_cores = 0;
268 if (topology_mode != TOPOLOGY_MODE_HW) {
269 id = topology_mode == TOPOLOGY_MODE_PACKAGE ? 0 : cpu;
270 topo->thread_id = cpu;
271 topo->core_id = cpu;
272 topo->socket_id = id;
273 topo->book_id = id;
274 topo->drawer_id = id;
275 }
276 }
277 hd_reset_state();
278 for_each_online_cpu(cpu) {
279 topo = &cpu_topology[cpu];
280 pkg_first = cpumask_first(&topo->core_mask);
281 topo_package = &cpu_topology[pkg_first];
282 if (cpu == pkg_first) {
283 for_each_cpu(sibling, &topo->core_mask) {
284 topo_sibling = &cpu_topology[sibling];
285 smt_first = cpumask_first(&topo_sibling->thread_mask);
286 if (sibling == smt_first) {
287 topo_package->booted_cores++;
288 hd_add_core(sibling);
289 }
290 }
291 } else {
292 topo->booted_cores = topo_package->booted_cores;
293 }
294 }
295}
296
297void store_topology(struct sysinfo_15_1_x *info)
298{
299 stsi(info, 15, 1, topology_mnest_limit());
300}
301
302static void __arch_update_dedicated_flag(void *arg)
303{
304 if (topology_cpu_dedicated(smp_processor_id()))
305 set_cpu_flag(CIF_DEDICATED_CPU);
306 else
307 clear_cpu_flag(CIF_DEDICATED_CPU);
308}
309
310static int __arch_update_cpu_topology(void)
311{
312 struct sysinfo_15_1_x *info = tl_info;
313 int rc, hd_status;
314
315 hd_status = 0;
316 rc = 0;
317 mutex_lock(&smp_cpu_state_mutex);
318 if (MACHINE_HAS_TOPOLOGY) {
319 rc = 1;
320 store_topology(info);
321 tl_to_masks(info);
322 }
323 update_cpu_masks();
324 if (!MACHINE_HAS_TOPOLOGY)
325 topology_update_polarization_simple();
326 if (cpu_management == 1)
327 hd_status = hd_enable_hiperdispatch();
328 mutex_unlock(&smp_cpu_state_mutex);
329 if (hd_status == 0)
330 hd_disable_hiperdispatch();
331 return rc;
332}
333
334int arch_update_cpu_topology(void)
335{
336 int rc;
337
338 rc = __arch_update_cpu_topology();
339 on_each_cpu(__arch_update_dedicated_flag, NULL, 0);
340 return rc;
341}
342
343static void topology_work_fn(struct work_struct *work)
344{
345 rebuild_sched_domains();
346}
347
348void topology_schedule_update(void)
349{
350 schedule_work(&topology_work);
351}
352
353static void topology_flush_work(void)
354{
355 flush_work(&topology_work);
356}
357
358static void topology_timer_fn(struct timer_list *unused)
359{
360 if (ptf(PTF_CHECK))
361 topology_schedule_update();
362 set_topology_timer();
363}
364
365static struct timer_list topology_timer;
366
367static atomic_t topology_poll = ATOMIC_INIT(0);
368
369static void set_topology_timer(void)
370{
371 if (atomic_add_unless(&topology_poll, -1, 0))
372 mod_timer(&topology_timer, jiffies + msecs_to_jiffies(100));
373 else
374 mod_timer(&topology_timer, jiffies + msecs_to_jiffies(60 * MSEC_PER_SEC));
375}
376
377void topology_expect_change(void)
378{
379 if (!MACHINE_HAS_TOPOLOGY)
380 return;
381 /* This is racy, but it doesn't matter since it is just a heuristic.
382 * Worst case is that we poll in a higher frequency for a bit longer.
383 */
384 if (atomic_read(&topology_poll) > 60)
385 return;
386 atomic_add(60, &topology_poll);
387 set_topology_timer();
388}
389
390static int set_polarization(int polarization)
391{
392 int rc = 0;
393
394 cpus_read_lock();
395 mutex_lock(&smp_cpu_state_mutex);
396 if (cpu_management == polarization)
397 goto out;
398 rc = topology_set_cpu_management(polarization);
399 if (rc)
400 goto out;
401 cpu_management = polarization;
402 topology_expect_change();
403out:
404 mutex_unlock(&smp_cpu_state_mutex);
405 cpus_read_unlock();
406 return rc;
407}
408
409static ssize_t dispatching_show(struct device *dev,
410 struct device_attribute *attr,
411 char *buf)
412{
413 ssize_t count;
414
415 mutex_lock(&smp_cpu_state_mutex);
416 count = sysfs_emit(buf, "%d\n", cpu_management);
417 mutex_unlock(&smp_cpu_state_mutex);
418 return count;
419}
420
421static ssize_t dispatching_store(struct device *dev,
422 struct device_attribute *attr,
423 const char *buf,
424 size_t count)
425{
426 int val, rc;
427 char delim;
428
429 if (sscanf(buf, "%d %c", &val, &delim) != 1)
430 return -EINVAL;
431 if (val != 0 && val != 1)
432 return -EINVAL;
433 rc = set_polarization(val);
434 return rc ? rc : count;
435}
436static DEVICE_ATTR_RW(dispatching);
437
438static ssize_t cpu_polarization_show(struct device *dev,
439 struct device_attribute *attr, char *buf)
440{
441 int cpu = dev->id;
442 ssize_t count;
443
444 mutex_lock(&smp_cpu_state_mutex);
445 switch (smp_cpu_get_polarization(cpu)) {
446 case POLARIZATION_HRZ:
447 count = sysfs_emit(buf, "horizontal\n");
448 break;
449 case POLARIZATION_VL:
450 count = sysfs_emit(buf, "vertical:low\n");
451 break;
452 case POLARIZATION_VM:
453 count = sysfs_emit(buf, "vertical:medium\n");
454 break;
455 case POLARIZATION_VH:
456 count = sysfs_emit(buf, "vertical:high\n");
457 break;
458 default:
459 count = sysfs_emit(buf, "unknown\n");
460 break;
461 }
462 mutex_unlock(&smp_cpu_state_mutex);
463 return count;
464}
465static DEVICE_ATTR(polarization, 0444, cpu_polarization_show, NULL);
466
467static struct attribute *topology_cpu_attrs[] = {
468 &dev_attr_polarization.attr,
469 NULL,
470};
471
472static struct attribute_group topology_cpu_attr_group = {
473 .attrs = topology_cpu_attrs,
474};
475
476static ssize_t cpu_dedicated_show(struct device *dev,
477 struct device_attribute *attr, char *buf)
478{
479 int cpu = dev->id;
480 ssize_t count;
481
482 mutex_lock(&smp_cpu_state_mutex);
483 count = sysfs_emit(buf, "%d\n", topology_cpu_dedicated(cpu));
484 mutex_unlock(&smp_cpu_state_mutex);
485 return count;
486}
487static DEVICE_ATTR(dedicated, 0444, cpu_dedicated_show, NULL);
488
489static struct attribute *topology_extra_cpu_attrs[] = {
490 &dev_attr_dedicated.attr,
491 NULL,
492};
493
494static struct attribute_group topology_extra_cpu_attr_group = {
495 .attrs = topology_extra_cpu_attrs,
496};
497
498int topology_cpu_init(struct cpu *cpu)
499{
500 int rc;
501
502 rc = sysfs_create_group(&cpu->dev.kobj, &topology_cpu_attr_group);
503 if (rc || !MACHINE_HAS_TOPOLOGY)
504 return rc;
505 rc = sysfs_create_group(&cpu->dev.kobj, &topology_extra_cpu_attr_group);
506 if (rc)
507 sysfs_remove_group(&cpu->dev.kobj, &topology_cpu_attr_group);
508 return rc;
509}
510
511static const struct cpumask *cpu_thread_mask(int cpu)
512{
513 return &cpu_topology[cpu].thread_mask;
514}
515
516
517const struct cpumask *cpu_coregroup_mask(int cpu)
518{
519 return &cpu_topology[cpu].core_mask;
520}
521
522static const struct cpumask *cpu_book_mask(int cpu)
523{
524 return &cpu_topology[cpu].book_mask;
525}
526
527static const struct cpumask *cpu_drawer_mask(int cpu)
528{
529 return &cpu_topology[cpu].drawer_mask;
530}
531
532static struct sched_domain_topology_level s390_topology[] = {
533 { cpu_thread_mask, cpu_smt_flags, SD_INIT_NAME(SMT) },
534 { cpu_coregroup_mask, cpu_core_flags, SD_INIT_NAME(MC) },
535 { cpu_book_mask, SD_INIT_NAME(BOOK) },
536 { cpu_drawer_mask, SD_INIT_NAME(DRAWER) },
537 { cpu_cpu_mask, SD_INIT_NAME(PKG) },
538 { NULL, },
539};
540
541static void __init alloc_masks(struct sysinfo_15_1_x *info,
542 struct mask_info *mask, int offset)
543{
544 int i, nr_masks;
545
546 nr_masks = info->mag[TOPOLOGY_NR_MAG - offset];
547 for (i = 0; i < info->mnest - offset; i++)
548 nr_masks *= info->mag[TOPOLOGY_NR_MAG - offset - 1 - i];
549 nr_masks = max(nr_masks, 1);
550 for (i = 0; i < nr_masks; i++) {
551 mask->next = memblock_alloc(sizeof(*mask->next), 8);
552 if (!mask->next)
553 panic("%s: Failed to allocate %zu bytes align=0x%x\n",
554 __func__, sizeof(*mask->next), 8);
555 mask = mask->next;
556 }
557}
558
559void __init topology_init_early(void)
560{
561 struct sysinfo_15_1_x *info;
562
563 set_sched_topology(s390_topology);
564 if (topology_mode == TOPOLOGY_MODE_UNINITIALIZED) {
565 if (MACHINE_HAS_TOPOLOGY)
566 topology_mode = TOPOLOGY_MODE_HW;
567 else
568 topology_mode = TOPOLOGY_MODE_SINGLE;
569 }
570 if (!MACHINE_HAS_TOPOLOGY)
571 goto out;
572 tl_info = memblock_alloc(PAGE_SIZE, PAGE_SIZE);
573 if (!tl_info)
574 panic("%s: Failed to allocate %lu bytes align=0x%lx\n",
575 __func__, PAGE_SIZE, PAGE_SIZE);
576 info = tl_info;
577 store_topology(info);
578 pr_info("The CPU configuration topology of the machine is: %d %d %d %d %d %d / %d\n",
579 info->mag[0], info->mag[1], info->mag[2], info->mag[3],
580 info->mag[4], info->mag[5], info->mnest);
581 alloc_masks(info, &socket_info, 1);
582 alloc_masks(info, &book_info, 2);
583 alloc_masks(info, &drawer_info, 3);
584out:
585 cpumask_set_cpu(0, &cpu_setup_mask);
586 __arch_update_cpu_topology();
587 __arch_update_dedicated_flag(NULL);
588}
589
590static inline int topology_get_mode(int enabled)
591{
592 if (!enabled)
593 return TOPOLOGY_MODE_SINGLE;
594 return MACHINE_HAS_TOPOLOGY ? TOPOLOGY_MODE_HW : TOPOLOGY_MODE_PACKAGE;
595}
596
597static inline int topology_is_enabled(void)
598{
599 return topology_mode != TOPOLOGY_MODE_SINGLE;
600}
601
602static int __init topology_setup(char *str)
603{
604 bool enabled;
605 int rc;
606
607 rc = kstrtobool(str, &enabled);
608 if (rc)
609 return rc;
610 topology_mode = topology_get_mode(enabled);
611 return 0;
612}
613early_param("topology", topology_setup);
614
615static int topology_ctl_handler(const struct ctl_table *ctl, int write,
616 void *buffer, size_t *lenp, loff_t *ppos)
617{
618 int enabled = topology_is_enabled();
619 int new_mode;
620 int rc;
621 struct ctl_table ctl_entry = {
622 .procname = ctl->procname,
623 .data = &enabled,
624 .maxlen = sizeof(int),
625 .extra1 = SYSCTL_ZERO,
626 .extra2 = SYSCTL_ONE,
627 };
628
629 rc = proc_douintvec_minmax(&ctl_entry, write, buffer, lenp, ppos);
630 if (rc < 0 || !write)
631 return rc;
632
633 mutex_lock(&smp_cpu_state_mutex);
634 new_mode = topology_get_mode(enabled);
635 if (topology_mode != new_mode) {
636 topology_mode = new_mode;
637 topology_schedule_update();
638 }
639 mutex_unlock(&smp_cpu_state_mutex);
640 topology_flush_work();
641
642 return rc;
643}
644
645static int polarization_ctl_handler(const struct ctl_table *ctl, int write,
646 void *buffer, size_t *lenp, loff_t *ppos)
647{
648 int polarization;
649 int rc;
650 struct ctl_table ctl_entry = {
651 .procname = ctl->procname,
652 .data = &polarization,
653 .maxlen = sizeof(int),
654 .extra1 = SYSCTL_ZERO,
655 .extra2 = SYSCTL_ONE,
656 };
657
658 polarization = cpu_management;
659 rc = proc_douintvec_minmax(&ctl_entry, write, buffer, lenp, ppos);
660 if (rc < 0 || !write)
661 return rc;
662 return set_polarization(polarization);
663}
664
665static struct ctl_table topology_ctl_table[] = {
666 {
667 .procname = "topology",
668 .mode = 0644,
669 .proc_handler = topology_ctl_handler,
670 },
671 {
672 .procname = "polarization",
673 .mode = 0644,
674 .proc_handler = polarization_ctl_handler,
675 },
676};
677
678static int __init topology_init(void)
679{
680 struct device *dev_root;
681 int rc = 0;
682
683 timer_setup(&topology_timer, topology_timer_fn, TIMER_DEFERRABLE);
684 if (MACHINE_HAS_TOPOLOGY)
685 set_topology_timer();
686 else
687 topology_update_polarization_simple();
688 if (IS_ENABLED(CONFIG_SCHED_TOPOLOGY_VERTICAL))
689 set_polarization(1);
690 register_sysctl("s390", topology_ctl_table);
691
692 dev_root = bus_get_dev_root(&cpu_subsys);
693 if (dev_root) {
694 rc = device_create_file(dev_root, &dev_attr_dispatching);
695 put_device(dev_root);
696 }
697 return rc;
698}
699device_initcall(topology_init);
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Copyright IBM Corp. 2007, 2011
4 * Author(s): Heiko Carstens <heiko.carstens@de.ibm.com>
5 */
6
7#define KMSG_COMPONENT "cpu"
8#define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
9
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>
25#include <linux/mm.h>
26#include <linux/nodemask.h>
27#include <linux/node.h>
28#include <asm/sysinfo.h>
29#include <asm/numa.h>
30
31#define PTF_HORIZONTAL (0UL)
32#define PTF_VERTICAL (1UL)
33#define PTF_CHECK (2UL)
34
35enum {
36 TOPOLOGY_MODE_HW,
37 TOPOLOGY_MODE_SINGLE,
38 TOPOLOGY_MODE_PACKAGE,
39 TOPOLOGY_MODE_UNINITIALIZED
40};
41
42struct mask_info {
43 struct mask_info *next;
44 unsigned char id;
45 cpumask_t mask;
46};
47
48static int topology_mode = TOPOLOGY_MODE_UNINITIALIZED;
49static void set_topology_timer(void);
50static void topology_work_fn(struct work_struct *work);
51static struct sysinfo_15_1_x *tl_info;
52
53static DECLARE_WORK(topology_work, topology_work_fn);
54
55/*
56 * Socket/Book linked lists and cpu_topology updates are
57 * protected by "sched_domains_mutex".
58 */
59static struct mask_info socket_info;
60static struct mask_info book_info;
61static struct mask_info drawer_info;
62
63struct cpu_topology_s390 cpu_topology[NR_CPUS];
64EXPORT_SYMBOL_GPL(cpu_topology);
65
66cpumask_t cpus_with_topology;
67
68static cpumask_t cpu_group_map(struct mask_info *info, unsigned int cpu)
69{
70 cpumask_t mask;
71
72 cpumask_copy(&mask, cpumask_of(cpu));
73 switch (topology_mode) {
74 case TOPOLOGY_MODE_HW:
75 while (info) {
76 if (cpumask_test_cpu(cpu, &info->mask)) {
77 mask = info->mask;
78 break;
79 }
80 info = info->next;
81 }
82 if (cpumask_empty(&mask))
83 cpumask_copy(&mask, cpumask_of(cpu));
84 break;
85 case TOPOLOGY_MODE_PACKAGE:
86 cpumask_copy(&mask, cpu_present_mask);
87 break;
88 default:
89 /* fallthrough */
90 case TOPOLOGY_MODE_SINGLE:
91 cpumask_copy(&mask, cpumask_of(cpu));
92 break;
93 }
94 return mask;
95}
96
97static cpumask_t cpu_thread_map(unsigned int cpu)
98{
99 cpumask_t mask;
100 int i;
101
102 cpumask_copy(&mask, cpumask_of(cpu));
103 if (topology_mode != TOPOLOGY_MODE_HW)
104 return mask;
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);
109 return mask;
110}
111
112#define TOPOLOGY_CORE_BITS 64
113
114static 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)
118{
119 struct cpu_topology_s390 *topo;
120 unsigned int core;
121
122 for_each_set_bit(core, &tl_core->mask, TOPOLOGY_CORE_BITS) {
123 unsigned int rcore;
124 int lcpu, i;
125
126 rcore = TOPOLOGY_CORE_BITS - 1 - core + tl_core->origin;
127 lcpu = smp_find_processor_id(rcore << smp_cpu_mt_shift);
128 if (lcpu < 0)
129 continue;
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);
143 }
144 }
145}
146
147static void clear_masks(void)
148{
149 struct mask_info *info;
150
151 info = &socket_info;
152 while (info) {
153 cpumask_clear(&info->mask);
154 info = info->next;
155 }
156 info = &book_info;
157 while (info) {
158 cpumask_clear(&info->mask);
159 info = info->next;
160 }
161 info = &drawer_info;
162 while (info) {
163 cpumask_clear(&info->mask);
164 info = info->next;
165 }
166}
167
168static union topology_entry *next_tle(union topology_entry *tle)
169{
170 if (!tle->nl)
171 return (union topology_entry *)((struct topology_core *)tle + 1);
172 return (union topology_entry *)((struct topology_container *)tle + 1);
173}
174
175static void tl_to_masks(struct sysinfo_15_1_x *info)
176{
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;
181
182 clear_masks();
183 tle = info->tle;
184 end = (union topology_entry *)((unsigned long)info + info->length);
185 while (tle < end) {
186 switch (tle->nl) {
187 case 3:
188 drawer = drawer->next;
189 drawer->id = tle->container.id;
190 break;
191 case 2:
192 book = book->next;
193 book->id = tle->container.id;
194 break;
195 case 1:
196 socket = socket->next;
197 socket->id = tle->container.id;
198 break;
199 case 0:
200 add_cpus_to_mask(&tle->cpu, drawer, book, socket);
201 break;
202 default:
203 clear_masks();
204 return;
205 }
206 tle = next_tle(tle);
207 }
208}
209
210static void topology_update_polarization_simple(void)
211{
212 int cpu;
213
214 for_each_possible_cpu(cpu)
215 smp_cpu_set_polarization(cpu, POLARIZATION_HRZ);
216}
217
218static int ptf(unsigned long fc)
219{
220 int rc;
221
222 asm volatile(
223 " .insn rre,0xb9a20000,%1,%1\n"
224 " ipm %0\n"
225 " srl %0,28\n"
226 : "=d" (rc)
227 : "d" (fc) : "cc");
228 return rc;
229}
230
231int topology_set_cpu_management(int fc)
232{
233 int cpu, rc;
234
235 if (!MACHINE_HAS_TOPOLOGY)
236 return -EOPNOTSUPP;
237 if (fc)
238 rc = ptf(PTF_VERTICAL);
239 else
240 rc = ptf(PTF_HORIZONTAL);
241 if (rc)
242 return -EBUSY;
243 for_each_possible_cpu(cpu)
244 smp_cpu_set_polarization(cpu, POLARIZATION_UNKNOWN);
245 return rc;
246}
247
248static void update_cpu_masks(void)
249{
250 struct cpu_topology_s390 *topo;
251 int cpu, id;
252
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;
262 topo->core_id = cpu;
263 topo->socket_id = id;
264 topo->book_id = id;
265 topo->drawer_id = id;
266 if (cpu_present(cpu))
267 cpumask_set_cpu(cpu, &cpus_with_topology);
268 }
269 }
270 numa_update_cpu_topology();
271}
272
273void store_topology(struct sysinfo_15_1_x *info)
274{
275 stsi(info, 15, 1, topology_mnest_limit());
276}
277
278static void __arch_update_dedicated_flag(void *arg)
279{
280 if (topology_cpu_dedicated(smp_processor_id()))
281 set_cpu_flag(CIF_DEDICATED_CPU);
282 else
283 clear_cpu_flag(CIF_DEDICATED_CPU);
284}
285
286static int __arch_update_cpu_topology(void)
287{
288 struct sysinfo_15_1_x *info = tl_info;
289 int rc = 0;
290
291 mutex_lock(&smp_cpu_state_mutex);
292 cpumask_clear(&cpus_with_topology);
293 if (MACHINE_HAS_TOPOLOGY) {
294 rc = 1;
295 store_topology(info);
296 tl_to_masks(info);
297 }
298 update_cpu_masks();
299 if (!MACHINE_HAS_TOPOLOGY)
300 topology_update_polarization_simple();
301 mutex_unlock(&smp_cpu_state_mutex);
302 return rc;
303}
304
305int arch_update_cpu_topology(void)
306{
307 struct device *dev;
308 int cpu, rc;
309
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 if (dev)
315 kobject_uevent(&dev->kobj, KOBJ_CHANGE);
316 }
317 return rc;
318}
319
320static void topology_work_fn(struct work_struct *work)
321{
322 rebuild_sched_domains();
323}
324
325void topology_schedule_update(void)
326{
327 schedule_work(&topology_work);
328}
329
330static void topology_flush_work(void)
331{
332 flush_work(&topology_work);
333}
334
335static void topology_timer_fn(struct timer_list *unused)
336{
337 if (ptf(PTF_CHECK))
338 topology_schedule_update();
339 set_topology_timer();
340}
341
342static struct timer_list topology_timer;
343
344static atomic_t topology_poll = ATOMIC_INIT(0);
345
346static void set_topology_timer(void)
347{
348 if (atomic_add_unless(&topology_poll, -1, 0))
349 mod_timer(&topology_timer, jiffies + HZ / 10);
350 else
351 mod_timer(&topology_timer, jiffies + HZ * 60);
352}
353
354void topology_expect_change(void)
355{
356 if (!MACHINE_HAS_TOPOLOGY)
357 return;
358 /* This is racy, but it doesn't matter since it is just a heuristic.
359 * Worst case is that we poll in a higher frequency for a bit longer.
360 */
361 if (atomic_read(&topology_poll) > 60)
362 return;
363 atomic_add(60, &topology_poll);
364 set_topology_timer();
365}
366
367static int cpu_management;
368
369static ssize_t dispatching_show(struct device *dev,
370 struct device_attribute *attr,
371 char *buf)
372{
373 ssize_t count;
374
375 mutex_lock(&smp_cpu_state_mutex);
376 count = sprintf(buf, "%d\n", cpu_management);
377 mutex_unlock(&smp_cpu_state_mutex);
378 return count;
379}
380
381static ssize_t dispatching_store(struct device *dev,
382 struct device_attribute *attr,
383 const char *buf,
384 size_t count)
385{
386 int val, rc;
387 char delim;
388
389 if (sscanf(buf, "%d %c", &val, &delim) != 1)
390 return -EINVAL;
391 if (val != 0 && val != 1)
392 return -EINVAL;
393 rc = 0;
394 get_online_cpus();
395 mutex_lock(&smp_cpu_state_mutex);
396 if (cpu_management == val)
397 goto out;
398 rc = topology_set_cpu_management(val);
399 if (rc)
400 goto out;
401 cpu_management = val;
402 topology_expect_change();
403out:
404 mutex_unlock(&smp_cpu_state_mutex);
405 put_online_cpus();
406 return rc ? rc : count;
407}
408static DEVICE_ATTR_RW(dispatching);
409
410static ssize_t cpu_polarization_show(struct device *dev,
411 struct device_attribute *attr, char *buf)
412{
413 int cpu = dev->id;
414 ssize_t count;
415
416 mutex_lock(&smp_cpu_state_mutex);
417 switch (smp_cpu_get_polarization(cpu)) {
418 case POLARIZATION_HRZ:
419 count = sprintf(buf, "horizontal\n");
420 break;
421 case POLARIZATION_VL:
422 count = sprintf(buf, "vertical:low\n");
423 break;
424 case POLARIZATION_VM:
425 count = sprintf(buf, "vertical:medium\n");
426 break;
427 case POLARIZATION_VH:
428 count = sprintf(buf, "vertical:high\n");
429 break;
430 default:
431 count = sprintf(buf, "unknown\n");
432 break;
433 }
434 mutex_unlock(&smp_cpu_state_mutex);
435 return count;
436}
437static DEVICE_ATTR(polarization, 0444, cpu_polarization_show, NULL);
438
439static struct attribute *topology_cpu_attrs[] = {
440 &dev_attr_polarization.attr,
441 NULL,
442};
443
444static struct attribute_group topology_cpu_attr_group = {
445 .attrs = topology_cpu_attrs,
446};
447
448static ssize_t cpu_dedicated_show(struct device *dev,
449 struct device_attribute *attr, char *buf)
450{
451 int cpu = dev->id;
452 ssize_t count;
453
454 mutex_lock(&smp_cpu_state_mutex);
455 count = sprintf(buf, "%d\n", topology_cpu_dedicated(cpu));
456 mutex_unlock(&smp_cpu_state_mutex);
457 return count;
458}
459static DEVICE_ATTR(dedicated, 0444, cpu_dedicated_show, NULL);
460
461static struct attribute *topology_extra_cpu_attrs[] = {
462 &dev_attr_dedicated.attr,
463 NULL,
464};
465
466static struct attribute_group topology_extra_cpu_attr_group = {
467 .attrs = topology_extra_cpu_attrs,
468};
469
470int topology_cpu_init(struct cpu *cpu)
471{
472 int rc;
473
474 rc = sysfs_create_group(&cpu->dev.kobj, &topology_cpu_attr_group);
475 if (rc || !MACHINE_HAS_TOPOLOGY)
476 return rc;
477 rc = sysfs_create_group(&cpu->dev.kobj, &topology_extra_cpu_attr_group);
478 if (rc)
479 sysfs_remove_group(&cpu->dev.kobj, &topology_cpu_attr_group);
480 return rc;
481}
482
483static const struct cpumask *cpu_thread_mask(int cpu)
484{
485 return &cpu_topology[cpu].thread_mask;
486}
487
488
489const struct cpumask *cpu_coregroup_mask(int cpu)
490{
491 return &cpu_topology[cpu].core_mask;
492}
493
494static const struct cpumask *cpu_book_mask(int cpu)
495{
496 return &cpu_topology[cpu].book_mask;
497}
498
499static const struct cpumask *cpu_drawer_mask(int cpu)
500{
501 return &cpu_topology[cpu].drawer_mask;
502}
503
504static struct sched_domain_topology_level s390_topology[] = {
505 { cpu_thread_mask, cpu_smt_flags, SD_INIT_NAME(SMT) },
506 { cpu_coregroup_mask, cpu_core_flags, SD_INIT_NAME(MC) },
507 { cpu_book_mask, SD_INIT_NAME(BOOK) },
508 { cpu_drawer_mask, SD_INIT_NAME(DRAWER) },
509 { cpu_cpu_mask, SD_INIT_NAME(DIE) },
510 { NULL, },
511};
512
513static void __init alloc_masks(struct sysinfo_15_1_x *info,
514 struct mask_info *mask, int offset)
515{
516 int i, nr_masks;
517
518 nr_masks = info->mag[TOPOLOGY_NR_MAG - offset];
519 for (i = 0; i < info->mnest - offset; i++)
520 nr_masks *= info->mag[TOPOLOGY_NR_MAG - offset - 1 - i];
521 nr_masks = max(nr_masks, 1);
522 for (i = 0; i < nr_masks; i++) {
523 mask->next = memblock_alloc(sizeof(*mask->next), 8);
524 if (!mask->next)
525 panic("%s: Failed to allocate %zu bytes align=0x%x\n",
526 __func__, sizeof(*mask->next), 8);
527 mask = mask->next;
528 }
529}
530
531void __init topology_init_early(void)
532{
533 struct sysinfo_15_1_x *info;
534
535 set_sched_topology(s390_topology);
536 if (topology_mode == TOPOLOGY_MODE_UNINITIALIZED) {
537 if (MACHINE_HAS_TOPOLOGY)
538 topology_mode = TOPOLOGY_MODE_HW;
539 else
540 topology_mode = TOPOLOGY_MODE_SINGLE;
541 }
542 if (!MACHINE_HAS_TOPOLOGY)
543 goto out;
544 tl_info = memblock_alloc(PAGE_SIZE, PAGE_SIZE);
545 if (!tl_info)
546 panic("%s: Failed to allocate %lu bytes align=0x%lx\n",
547 __func__, PAGE_SIZE, PAGE_SIZE);
548 info = tl_info;
549 store_topology(info);
550 pr_info("The CPU configuration topology of the machine is: %d %d %d %d %d %d / %d\n",
551 info->mag[0], info->mag[1], info->mag[2], info->mag[3],
552 info->mag[4], info->mag[5], info->mnest);
553 alloc_masks(info, &socket_info, 1);
554 alloc_masks(info, &book_info, 2);
555 alloc_masks(info, &drawer_info, 3);
556out:
557 __arch_update_cpu_topology();
558 __arch_update_dedicated_flag(NULL);
559}
560
561static inline int topology_get_mode(int enabled)
562{
563 if (!enabled)
564 return TOPOLOGY_MODE_SINGLE;
565 return MACHINE_HAS_TOPOLOGY ? TOPOLOGY_MODE_HW : TOPOLOGY_MODE_PACKAGE;
566}
567
568static inline int topology_is_enabled(void)
569{
570 return topology_mode != TOPOLOGY_MODE_SINGLE;
571}
572
573static int __init topology_setup(char *str)
574{
575 bool enabled;
576 int rc;
577
578 rc = kstrtobool(str, &enabled);
579 if (rc)
580 return rc;
581 topology_mode = topology_get_mode(enabled);
582 return 0;
583}
584early_param("topology", topology_setup);
585
586static int topology_ctl_handler(struct ctl_table *ctl, int write,
587 void __user *buffer, size_t *lenp, loff_t *ppos)
588{
589 int enabled = topology_is_enabled();
590 int new_mode;
591 int rc;
592 struct ctl_table ctl_entry = {
593 .procname = ctl->procname,
594 .data = &enabled,
595 .maxlen = sizeof(int),
596 .extra1 = SYSCTL_ZERO,
597 .extra2 = SYSCTL_ONE,
598 };
599
600 rc = proc_douintvec_minmax(&ctl_entry, write, buffer, lenp, ppos);
601 if (rc < 0 || !write)
602 return rc;
603
604 mutex_lock(&smp_cpu_state_mutex);
605 new_mode = topology_get_mode(enabled);
606 if (topology_mode != new_mode) {
607 topology_mode = new_mode;
608 topology_schedule_update();
609 }
610 mutex_unlock(&smp_cpu_state_mutex);
611 topology_flush_work();
612
613 return rc;
614}
615
616static struct ctl_table topology_ctl_table[] = {
617 {
618 .procname = "topology",
619 .mode = 0644,
620 .proc_handler = topology_ctl_handler,
621 },
622 { },
623};
624
625static struct ctl_table topology_dir_table[] = {
626 {
627 .procname = "s390",
628 .maxlen = 0,
629 .mode = 0555,
630 .child = topology_ctl_table,
631 },
632 { },
633};
634
635static int __init topology_init(void)
636{
637 timer_setup(&topology_timer, topology_timer_fn, TIMER_DEFERRABLE);
638 if (MACHINE_HAS_TOPOLOGY)
639 set_topology_timer();
640 else
641 topology_update_polarization_simple();
642 register_sysctl_table(topology_dir_table);
643 return device_create_file(cpu_subsys.dev_root, &dev_attr_dispatching);
644}
645device_initcall(topology_init);