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