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/cpufeature.h>
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/hiperdispatch.h>
29 #include <asm/sysinfo.h>
30 #include <asm/asm.h>
31 
32 #define PTF_HORIZONTAL	(0UL)
33 #define PTF_VERTICAL	(1UL)
34 #define PTF_CHECK	(2UL)
35 
36 enum {
37 	TOPOLOGY_MODE_HW,
38 	TOPOLOGY_MODE_SINGLE,
39 	TOPOLOGY_MODE_PACKAGE,
40 	TOPOLOGY_MODE_UNINITIALIZED
41 };
42 
43 struct mask_info {
44 	struct mask_info *next;
45 	unsigned char id;
46 	cpumask_t mask;
47 };
48 
49 static int topology_mode = TOPOLOGY_MODE_UNINITIALIZED;
50 static void set_topology_timer(void);
51 static void topology_work_fn(struct work_struct *work);
52 static struct sysinfo_15_1_x *tl_info;
53 static int cpu_management;
54 
55 static DECLARE_WORK(topology_work, topology_work_fn);
56 
57 /*
58  * Socket/Book linked lists and cpu_topology updates are
59  * protected by "sched_domains_mutex".
60  */
61 static struct mask_info socket_info;
62 static struct mask_info book_info;
63 static struct mask_info drawer_info;
64 
65 struct cpu_topology_s390 cpu_topology[NR_CPUS];
66 EXPORT_SYMBOL_GPL(cpu_topology);
67 
68 static void cpu_group_map(cpumask_t *dst, struct mask_info *info, unsigned int cpu)
69 {
70 	static cpumask_t mask;
71 
72 	cpumask_clear(&mask);
73 	if (!cpumask_test_cpu(cpu, &cpu_setup_mask))
74 		goto out;
75 	cpumask_set_cpu(cpu, &mask);
76 	switch (topology_mode) {
77 	case TOPOLOGY_MODE_HW:
78 		while (info) {
79 			if (cpumask_test_cpu(cpu, &info->mask)) {
80 				cpumask_copy(&mask, &info->mask);
81 				break;
82 			}
83 			info = info->next;
84 		}
85 		break;
86 	case TOPOLOGY_MODE_PACKAGE:
87 		cpumask_copy(&mask, cpu_present_mask);
88 		break;
89 	default:
90 		fallthrough;
91 	case TOPOLOGY_MODE_SINGLE:
92 		break;
93 	}
94 	cpumask_and(&mask, &mask, &cpu_setup_mask);
95 out:
96 	cpumask_copy(dst, &mask);
97 }
98 
99 static void cpu_thread_map(cpumask_t *dst, unsigned int cpu)
100 {
101 	static cpumask_t mask;
102 	unsigned int max_cpu;
103 
104 	cpumask_clear(&mask);
105 	if (!cpumask_test_cpu(cpu, &cpu_setup_mask))
106 		goto out;
107 	cpumask_set_cpu(cpu, &mask);
108 	if (topology_mode != TOPOLOGY_MODE_HW)
109 		goto out;
110 	cpu -= cpu % (smp_cpu_mtid + 1);
111 	max_cpu = min(cpu + smp_cpu_mtid, nr_cpu_ids - 1);
112 	for (; cpu <= max_cpu; cpu++) {
113 		if (cpumask_test_cpu(cpu, &cpu_setup_mask))
114 			cpumask_set_cpu(cpu, &mask);
115 	}
116 out:
117 	cpumask_copy(dst, &mask);
118 }
119 
120 #define TOPOLOGY_CORE_BITS	64
121 
122 static void add_cpus_to_mask(struct topology_core *tl_core,
123 			     struct mask_info *drawer,
124 			     struct mask_info *book,
125 			     struct mask_info *socket)
126 {
127 	struct cpu_topology_s390 *topo;
128 	unsigned int core;
129 
130 	for_each_set_bit(core, &tl_core->mask, TOPOLOGY_CORE_BITS) {
131 		unsigned int max_cpu, rcore;
132 		int cpu;
133 
134 		rcore = TOPOLOGY_CORE_BITS - 1 - core + tl_core->origin;
135 		cpu = smp_find_processor_id(rcore << smp_cpu_mt_shift);
136 		if (cpu < 0)
137 			continue;
138 		max_cpu = min(cpu + smp_cpu_mtid, nr_cpu_ids - 1);
139 		for (; cpu <= max_cpu; cpu++) {
140 			topo = &cpu_topology[cpu];
141 			topo->drawer_id = drawer->id;
142 			topo->book_id = book->id;
143 			topo->socket_id = socket->id;
144 			topo->core_id = rcore;
145 			topo->thread_id = cpu;
146 			topo->dedicated = tl_core->d;
147 			cpumask_set_cpu(cpu, &drawer->mask);
148 			cpumask_set_cpu(cpu, &book->mask);
149 			cpumask_set_cpu(cpu, &socket->mask);
150 			smp_cpu_set_polarization(cpu, tl_core->pp);
151 			smp_cpu_set_capacity(cpu, CPU_CAPACITY_HIGH);
152 		}
153 	}
154 }
155 
156 static void clear_masks(void)
157 {
158 	struct mask_info *info;
159 
160 	info = &socket_info;
161 	while (info) {
162 		cpumask_clear(&info->mask);
163 		info = info->next;
164 	}
165 	info = &book_info;
166 	while (info) {
167 		cpumask_clear(&info->mask);
168 		info = info->next;
169 	}
170 	info = &drawer_info;
171 	while (info) {
172 		cpumask_clear(&info->mask);
173 		info = info->next;
174 	}
175 }
176 
177 static union topology_entry *next_tle(union topology_entry *tle)
178 {
179 	if (!tle->nl)
180 		return (union topology_entry *)((struct topology_core *)tle + 1);
181 	return (union topology_entry *)((struct topology_container *)tle + 1);
182 }
183 
184 static void tl_to_masks(struct sysinfo_15_1_x *info)
185 {
186 	struct mask_info *socket = &socket_info;
187 	struct mask_info *book = &book_info;
188 	struct mask_info *drawer = &drawer_info;
189 	union topology_entry *tle, *end;
190 
191 	clear_masks();
192 	tle = info->tle;
193 	end = (union topology_entry *)((unsigned long)info + info->length);
194 	while (tle < end) {
195 		switch (tle->nl) {
196 		case 3:
197 			drawer = drawer->next;
198 			drawer->id = tle->container.id;
199 			break;
200 		case 2:
201 			book = book->next;
202 			book->id = tle->container.id;
203 			break;
204 		case 1:
205 			socket = socket->next;
206 			socket->id = tle->container.id;
207 			break;
208 		case 0:
209 			add_cpus_to_mask(&tle->cpu, drawer, book, socket);
210 			break;
211 		default:
212 			clear_masks();
213 			return;
214 		}
215 		tle = next_tle(tle);
216 	}
217 }
218 
219 static void topology_update_polarization_simple(void)
220 {
221 	int cpu;
222 
223 	for_each_possible_cpu(cpu)
224 		smp_cpu_set_polarization(cpu, POLARIZATION_HRZ);
225 }
226 
227 static int ptf(unsigned long fc)
228 {
229 	int cc;
230 
231 	asm volatile(
232 		"	.insn	rre,0xb9a20000,%[fc],%[fc]\n"
233 		CC_IPM(cc)
234 		: CC_OUT(cc, cc)
235 		: [fc] "d" (fc)
236 		: CC_CLOBBER);
237 	return CC_TRANSFORM(cc);
238 }
239 
240 int topology_set_cpu_management(int fc)
241 {
242 	int cpu, rc;
243 
244 	if (!cpu_has_topology())
245 		return -EOPNOTSUPP;
246 	if (fc)
247 		rc = ptf(PTF_VERTICAL);
248 	else
249 		rc = ptf(PTF_HORIZONTAL);
250 	if (rc)
251 		return -EBUSY;
252 	for_each_possible_cpu(cpu)
253 		smp_cpu_set_polarization(cpu, POLARIZATION_UNKNOWN);
254 	return rc;
255 }
256 
257 void update_cpu_masks(void)
258 {
259 	struct cpu_topology_s390 *topo, *topo_package, *topo_sibling;
260 	int cpu, sibling, pkg_first, smt_first, id;
261 
262 	for_each_possible_cpu(cpu) {
263 		topo = &cpu_topology[cpu];
264 		cpu_thread_map(&topo->thread_mask, cpu);
265 		cpu_group_map(&topo->core_mask, &socket_info, cpu);
266 		cpu_group_map(&topo->book_mask, &book_info, cpu);
267 		cpu_group_map(&topo->drawer_mask, &drawer_info, cpu);
268 		topo->booted_cores = 0;
269 		if (topology_mode != TOPOLOGY_MODE_HW) {
270 			id = topology_mode == TOPOLOGY_MODE_PACKAGE ? 0 : cpu;
271 			topo->thread_id = cpu;
272 			topo->core_id = cpu;
273 			topo->socket_id = id;
274 			topo->book_id = id;
275 			topo->drawer_id = id;
276 		}
277 	}
278 	hd_reset_state();
279 	for_each_online_cpu(cpu) {
280 		topo = &cpu_topology[cpu];
281 		pkg_first = cpumask_first(&topo->core_mask);
282 		topo_package = &cpu_topology[pkg_first];
283 		if (cpu == pkg_first) {
284 			for_each_cpu(sibling, &topo->core_mask) {
285 				topo_sibling = &cpu_topology[sibling];
286 				smt_first = cpumask_first(&topo_sibling->thread_mask);
287 				if (sibling == smt_first) {
288 					topo_package->booted_cores++;
289 					hd_add_core(sibling);
290 				}
291 			}
292 		} else {
293 			topo->booted_cores = topo_package->booted_cores;
294 		}
295 	}
296 }
297 
298 void store_topology(struct sysinfo_15_1_x *info)
299 {
300 	stsi(info, 15, 1, topology_mnest_limit());
301 }
302 
303 static void __arch_update_dedicated_flag(void *arg)
304 {
305 	if (topology_cpu_dedicated(smp_processor_id()))
306 		set_cpu_flag(CIF_DEDICATED_CPU);
307 	else
308 		clear_cpu_flag(CIF_DEDICATED_CPU);
309 }
310 
311 static int __arch_update_cpu_topology(void)
312 {
313 	struct sysinfo_15_1_x *info = tl_info;
314 	int rc, hd_status;
315 
316 	hd_status = 0;
317 	rc = 0;
318 	mutex_lock(&smp_cpu_state_mutex);
319 	if (cpu_has_topology()) {
320 		rc = 1;
321 		store_topology(info);
322 		tl_to_masks(info);
323 	}
324 	update_cpu_masks();
325 	if (!cpu_has_topology())
326 		topology_update_polarization_simple();
327 	if (cpu_management == 1)
328 		hd_status = hd_enable_hiperdispatch();
329 	mutex_unlock(&smp_cpu_state_mutex);
330 	if (hd_status == 0)
331 		hd_disable_hiperdispatch();
332 	return rc;
333 }
334 
335 int arch_update_cpu_topology(void)
336 {
337 	int rc;
338 
339 	rc = __arch_update_cpu_topology();
340 	on_each_cpu(__arch_update_dedicated_flag, NULL, 0);
341 	return rc;
342 }
343 
344 static void topology_work_fn(struct work_struct *work)
345 {
346 	rebuild_sched_domains();
347 }
348 
349 void topology_schedule_update(void)
350 {
351 	schedule_work(&topology_work);
352 }
353 
354 static void topology_flush_work(void)
355 {
356 	flush_work(&topology_work);
357 }
358 
359 static void topology_timer_fn(struct timer_list *unused)
360 {
361 	if (ptf(PTF_CHECK))
362 		topology_schedule_update();
363 	set_topology_timer();
364 }
365 
366 static struct timer_list topology_timer;
367 
368 static atomic_t topology_poll = ATOMIC_INIT(0);
369 
370 static void set_topology_timer(void)
371 {
372 	if (atomic_add_unless(&topology_poll, -1, 0))
373 		mod_timer(&topology_timer, jiffies + msecs_to_jiffies(100));
374 	else
375 		mod_timer(&topology_timer, jiffies + secs_to_jiffies(60));
376 }
377 
378 void topology_expect_change(void)
379 {
380 	if (!cpu_has_topology())
381 		return;
382 	/* This is racy, but it doesn't matter since it is just a heuristic.
383 	 * Worst case is that we poll in a higher frequency for a bit longer.
384 	 */
385 	if (atomic_read(&topology_poll) > 60)
386 		return;
387 	atomic_add(60, &topology_poll);
388 	set_topology_timer();
389 }
390 
391 static int set_polarization(int polarization)
392 {
393 	int rc = 0;
394 
395 	cpus_read_lock();
396 	mutex_lock(&smp_cpu_state_mutex);
397 	if (cpu_management == polarization)
398 		goto out;
399 	rc = topology_set_cpu_management(polarization);
400 	if (rc)
401 		goto out;
402 	cpu_management = polarization;
403 	topology_expect_change();
404 out:
405 	mutex_unlock(&smp_cpu_state_mutex);
406 	cpus_read_unlock();
407 	return rc;
408 }
409 
410 static ssize_t dispatching_show(struct device *dev,
411 				struct device_attribute *attr,
412 				char *buf)
413 {
414 	ssize_t count;
415 
416 	mutex_lock(&smp_cpu_state_mutex);
417 	count = sysfs_emit(buf, "%d\n", cpu_management);
418 	mutex_unlock(&smp_cpu_state_mutex);
419 	return count;
420 }
421 
422 static ssize_t dispatching_store(struct device *dev,
423 				 struct device_attribute *attr,
424 				 const char *buf,
425 				 size_t count)
426 {
427 	int val, rc;
428 	char delim;
429 
430 	if (sscanf(buf, "%d %c", &val, &delim) != 1)
431 		return -EINVAL;
432 	if (val != 0 && val != 1)
433 		return -EINVAL;
434 	rc = set_polarization(val);
435 	return rc ? rc : count;
436 }
437 static DEVICE_ATTR_RW(dispatching);
438 
439 static ssize_t cpu_polarization_show(struct device *dev,
440 				     struct device_attribute *attr, char *buf)
441 {
442 	int cpu = dev->id;
443 	ssize_t count;
444 
445 	mutex_lock(&smp_cpu_state_mutex);
446 	switch (smp_cpu_get_polarization(cpu)) {
447 	case POLARIZATION_HRZ:
448 		count = sysfs_emit(buf, "horizontal\n");
449 		break;
450 	case POLARIZATION_VL:
451 		count = sysfs_emit(buf, "vertical:low\n");
452 		break;
453 	case POLARIZATION_VM:
454 		count = sysfs_emit(buf, "vertical:medium\n");
455 		break;
456 	case POLARIZATION_VH:
457 		count = sysfs_emit(buf, "vertical:high\n");
458 		break;
459 	default:
460 		count = sysfs_emit(buf, "unknown\n");
461 		break;
462 	}
463 	mutex_unlock(&smp_cpu_state_mutex);
464 	return count;
465 }
466 static DEVICE_ATTR(polarization, 0444, cpu_polarization_show, NULL);
467 
468 static struct attribute *topology_cpu_attrs[] = {
469 	&dev_attr_polarization.attr,
470 	NULL,
471 };
472 
473 static struct attribute_group topology_cpu_attr_group = {
474 	.attrs = topology_cpu_attrs,
475 };
476 
477 static ssize_t cpu_dedicated_show(struct device *dev,
478 				  struct device_attribute *attr, char *buf)
479 {
480 	int cpu = dev->id;
481 	ssize_t count;
482 
483 	mutex_lock(&smp_cpu_state_mutex);
484 	count = sysfs_emit(buf, "%d\n", topology_cpu_dedicated(cpu));
485 	mutex_unlock(&smp_cpu_state_mutex);
486 	return count;
487 }
488 static DEVICE_ATTR(dedicated, 0444, cpu_dedicated_show, NULL);
489 
490 static struct attribute *topology_extra_cpu_attrs[] = {
491 	&dev_attr_dedicated.attr,
492 	NULL,
493 };
494 
495 static struct attribute_group topology_extra_cpu_attr_group = {
496 	.attrs = topology_extra_cpu_attrs,
497 };
498 
499 int topology_cpu_init(struct cpu *cpu)
500 {
501 	int rc;
502 
503 	rc = sysfs_create_group(&cpu->dev.kobj, &topology_cpu_attr_group);
504 	if (rc || !cpu_has_topology())
505 		return rc;
506 	rc = sysfs_create_group(&cpu->dev.kobj, &topology_extra_cpu_attr_group);
507 	if (rc)
508 		sysfs_remove_group(&cpu->dev.kobj, &topology_cpu_attr_group);
509 	return rc;
510 }
511 
512 static const struct cpumask *cpu_thread_mask(int cpu)
513 {
514 	return &cpu_topology[cpu].thread_mask;
515 }
516 
517 
518 const struct cpumask *cpu_coregroup_mask(int cpu)
519 {
520 	return &cpu_topology[cpu].core_mask;
521 }
522 
523 static const struct cpumask *cpu_book_mask(int cpu)
524 {
525 	return &cpu_topology[cpu].book_mask;
526 }
527 
528 static const struct cpumask *cpu_drawer_mask(int cpu)
529 {
530 	return &cpu_topology[cpu].drawer_mask;
531 }
532 
533 static struct sched_domain_topology_level s390_topology[] = {
534 	{ cpu_thread_mask, cpu_smt_flags, SD_INIT_NAME(SMT) },
535 	{ cpu_coregroup_mask, cpu_core_flags, SD_INIT_NAME(MC) },
536 	{ cpu_book_mask, SD_INIT_NAME(BOOK) },
537 	{ cpu_drawer_mask, SD_INIT_NAME(DRAWER) },
538 	{ cpu_cpu_mask, SD_INIT_NAME(PKG) },
539 	{ NULL, },
540 };
541 
542 static void __init alloc_masks(struct sysinfo_15_1_x *info,
543 			       struct mask_info *mask, int offset)
544 {
545 	int i, nr_masks;
546 
547 	nr_masks = info->mag[TOPOLOGY_NR_MAG - offset];
548 	for (i = 0; i < info->mnest - offset; i++)
549 		nr_masks *= info->mag[TOPOLOGY_NR_MAG - offset - 1 - i];
550 	nr_masks = max(nr_masks, 1);
551 	for (i = 0; i < nr_masks; i++) {
552 		mask->next = memblock_alloc_or_panic(sizeof(*mask->next), 8);
553 		mask = mask->next;
554 	}
555 }
556 
557 static int __init detect_polarization(union topology_entry *tle)
558 {
559 	struct topology_core *tl_core;
560 
561 	while (tle->nl)
562 		tle = next_tle(tle);
563 	tl_core = (struct topology_core *)tle;
564 	return tl_core->pp != POLARIZATION_HRZ;
565 }
566 
567 void __init topology_init_early(void)
568 {
569 	struct sysinfo_15_1_x *info;
570 
571 	set_sched_topology(s390_topology);
572 	if (topology_mode == TOPOLOGY_MODE_UNINITIALIZED) {
573 		if (cpu_has_topology())
574 			topology_mode = TOPOLOGY_MODE_HW;
575 		else
576 			topology_mode = TOPOLOGY_MODE_SINGLE;
577 	}
578 	if (!cpu_has_topology())
579 		goto out;
580 	tl_info = memblock_alloc_or_panic(PAGE_SIZE, PAGE_SIZE);
581 	info = tl_info;
582 	store_topology(info);
583 	cpu_management = detect_polarization(info->tle);
584 	pr_info("The CPU configuration topology of the machine is: %d %d %d %d %d %d / %d\n",
585 		info->mag[0], info->mag[1], info->mag[2], info->mag[3],
586 		info->mag[4], info->mag[5], info->mnest);
587 	alloc_masks(info, &socket_info, 1);
588 	alloc_masks(info, &book_info, 2);
589 	alloc_masks(info, &drawer_info, 3);
590 out:
591 	cpumask_set_cpu(0, &cpu_setup_mask);
592 	__arch_update_cpu_topology();
593 	__arch_update_dedicated_flag(NULL);
594 }
595 
596 static inline int topology_get_mode(int enabled)
597 {
598 	if (!enabled)
599 		return TOPOLOGY_MODE_SINGLE;
600 	return cpu_has_topology() ? TOPOLOGY_MODE_HW : TOPOLOGY_MODE_PACKAGE;
601 }
602 
603 static inline int topology_is_enabled(void)
604 {
605 	return topology_mode != TOPOLOGY_MODE_SINGLE;
606 }
607 
608 static int __init topology_setup(char *str)
609 {
610 	bool enabled;
611 	int rc;
612 
613 	rc = kstrtobool(str, &enabled);
614 	if (rc)
615 		return rc;
616 	topology_mode = topology_get_mode(enabled);
617 	return 0;
618 }
619 early_param("topology", topology_setup);
620 
621 static int topology_ctl_handler(const struct ctl_table *ctl, int write,
622 				void *buffer, size_t *lenp, loff_t *ppos)
623 {
624 	int enabled = topology_is_enabled();
625 	int new_mode;
626 	int rc;
627 	struct ctl_table ctl_entry = {
628 		.procname	= ctl->procname,
629 		.data		= &enabled,
630 		.maxlen		= sizeof(int),
631 		.extra1		= SYSCTL_ZERO,
632 		.extra2		= SYSCTL_ONE,
633 	};
634 
635 	rc = proc_douintvec_minmax(&ctl_entry, write, buffer, lenp, ppos);
636 	if (rc < 0 || !write)
637 		return rc;
638 
639 	mutex_lock(&smp_cpu_state_mutex);
640 	new_mode = topology_get_mode(enabled);
641 	if (topology_mode != new_mode) {
642 		topology_mode = new_mode;
643 		topology_schedule_update();
644 	}
645 	mutex_unlock(&smp_cpu_state_mutex);
646 	topology_flush_work();
647 
648 	return rc;
649 }
650 
651 static int polarization_ctl_handler(const struct ctl_table *ctl, int write,
652 				    void *buffer, size_t *lenp, loff_t *ppos)
653 {
654 	int polarization;
655 	int rc;
656 	struct ctl_table ctl_entry = {
657 		.procname	= ctl->procname,
658 		.data		= &polarization,
659 		.maxlen		= sizeof(int),
660 		.extra1		= SYSCTL_ZERO,
661 		.extra2		= SYSCTL_ONE,
662 	};
663 
664 	polarization = cpu_management;
665 	rc = proc_douintvec_minmax(&ctl_entry, write, buffer, lenp, ppos);
666 	if (rc < 0 || !write)
667 		return rc;
668 	return set_polarization(polarization);
669 }
670 
671 static const struct ctl_table topology_ctl_table[] = {
672 	{
673 		.procname	= "topology",
674 		.mode		= 0644,
675 		.proc_handler	= topology_ctl_handler,
676 	},
677 	{
678 		.procname	= "polarization",
679 		.mode		= 0644,
680 		.proc_handler	= polarization_ctl_handler,
681 	},
682 };
683 
684 static int __init topology_init(void)
685 {
686 	struct device *dev_root;
687 	int rc = 0;
688 
689 	timer_setup(&topology_timer, topology_timer_fn, TIMER_DEFERRABLE);
690 	if (cpu_has_topology())
691 		set_topology_timer();
692 	else
693 		topology_update_polarization_simple();
694 	if (IS_ENABLED(CONFIG_SCHED_TOPOLOGY_VERTICAL))
695 		set_polarization(1);
696 	register_sysctl("s390", topology_ctl_table);
697 
698 	dev_root = bus_get_dev_root(&cpu_subsys);
699 	if (dev_root) {
700 		rc = device_create_file(dev_root, &dev_attr_dispatching);
701 		put_device(dev_root);
702 	}
703 	return rc;
704 }
705 device_initcall(topology_init);
706