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