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