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