xref: /linux/arch/s390/kernel/topology.c (revision b77e0ce62d63a761ffb7f7245a215a49f5921c2f)
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 
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_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 				cpumask_copy(&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 	cpumask_copy(dst, &mask);
93 }
94 
95 static void cpu_thread_map(cpumask_t *dst, unsigned int cpu)
96 {
97 	static cpumask_t mask;
98 	int i;
99 
100 	cpumask_copy(&mask, cpumask_of(cpu));
101 	if (topology_mode != TOPOLOGY_MODE_HW)
102 		goto out;
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 out:
109 	cpumask_copy(dst, &mask);
110 }
111 
112 #define TOPOLOGY_CORE_BITS	64
113 
114 static void add_cpus_to_mask(struct topology_core *tl_core,
115 			     struct mask_info *drawer,
116 			     struct mask_info *book,
117 			     struct mask_info *socket)
118 {
119 	struct cpu_topology_s390 *topo;
120 	unsigned int core;
121 
122 	for_each_set_bit(core, &tl_core->mask, TOPOLOGY_CORE_BITS) {
123 		unsigned int rcore;
124 		int lcpu, i;
125 
126 		rcore = TOPOLOGY_CORE_BITS - 1 - core + tl_core->origin;
127 		lcpu = smp_find_processor_id(rcore << smp_cpu_mt_shift);
128 		if (lcpu < 0)
129 			continue;
130 		for (i = 0; i <= smp_cpu_mtid; i++) {
131 			topo = &cpu_topology[lcpu + i];
132 			topo->drawer_id = drawer->id;
133 			topo->book_id = book->id;
134 			topo->socket_id = socket->id;
135 			topo->core_id = rcore;
136 			topo->thread_id = lcpu + i;
137 			topo->dedicated = tl_core->d;
138 			cpumask_set_cpu(lcpu + i, &drawer->mask);
139 			cpumask_set_cpu(lcpu + i, &book->mask);
140 			cpumask_set_cpu(lcpu + i, &socket->mask);
141 			smp_cpu_set_polarization(lcpu + i, tl_core->pp);
142 		}
143 	}
144 }
145 
146 static void clear_masks(void)
147 {
148 	struct mask_info *info;
149 
150 	info = &socket_info;
151 	while (info) {
152 		cpumask_clear(&info->mask);
153 		info = info->next;
154 	}
155 	info = &book_info;
156 	while (info) {
157 		cpumask_clear(&info->mask);
158 		info = info->next;
159 	}
160 	info = &drawer_info;
161 	while (info) {
162 		cpumask_clear(&info->mask);
163 		info = info->next;
164 	}
165 }
166 
167 static union topology_entry *next_tle(union topology_entry *tle)
168 {
169 	if (!tle->nl)
170 		return (union topology_entry *)((struct topology_core *)tle + 1);
171 	return (union topology_entry *)((struct topology_container *)tle + 1);
172 }
173 
174 static void tl_to_masks(struct sysinfo_15_1_x *info)
175 {
176 	struct mask_info *socket = &socket_info;
177 	struct mask_info *book = &book_info;
178 	struct mask_info *drawer = &drawer_info;
179 	union topology_entry *tle, *end;
180 
181 	clear_masks();
182 	tle = info->tle;
183 	end = (union topology_entry *)((unsigned long)info + info->length);
184 	while (tle < end) {
185 		switch (tle->nl) {
186 		case 3:
187 			drawer = drawer->next;
188 			drawer->id = tle->container.id;
189 			break;
190 		case 2:
191 			book = book->next;
192 			book->id = tle->container.id;
193 			break;
194 		case 1:
195 			socket = socket->next;
196 			socket->id = tle->container.id;
197 			break;
198 		case 0:
199 			add_cpus_to_mask(&tle->cpu, drawer, book, socket);
200 			break;
201 		default:
202 			clear_masks();
203 			return;
204 		}
205 		tle = next_tle(tle);
206 	}
207 }
208 
209 static void topology_update_polarization_simple(void)
210 {
211 	int cpu;
212 
213 	for_each_possible_cpu(cpu)
214 		smp_cpu_set_polarization(cpu, POLARIZATION_HRZ);
215 }
216 
217 static int ptf(unsigned long fc)
218 {
219 	int rc;
220 
221 	asm volatile(
222 		"	.insn	rre,0xb9a20000,%1,%1\n"
223 		"	ipm	%0\n"
224 		"	srl	%0,28\n"
225 		: "=d" (rc)
226 		: "d" (fc)  : "cc");
227 	return rc;
228 }
229 
230 int topology_set_cpu_management(int fc)
231 {
232 	int cpu, rc;
233 
234 	if (!MACHINE_HAS_TOPOLOGY)
235 		return -EOPNOTSUPP;
236 	if (fc)
237 		rc = ptf(PTF_VERTICAL);
238 	else
239 		rc = ptf(PTF_HORIZONTAL);
240 	if (rc)
241 		return -EBUSY;
242 	for_each_possible_cpu(cpu)
243 		smp_cpu_set_polarization(cpu, POLARIZATION_UNKNOWN);
244 	return rc;
245 }
246 
247 void update_cpu_masks(void)
248 {
249 	struct cpu_topology_s390 *topo, *topo_package, *topo_sibling;
250 	int cpu, sibling, pkg_first, smt_first, id;
251 
252 	for_each_possible_cpu(cpu) {
253 		topo = &cpu_topology[cpu];
254 		cpu_thread_map(&topo->thread_mask, cpu);
255 		cpu_group_map(&topo->core_mask, &socket_info, cpu);
256 		cpu_group_map(&topo->book_mask, &book_info, cpu);
257 		cpu_group_map(&topo->drawer_mask, &drawer_info, cpu);
258 		topo->booted_cores = 0;
259 		if (topology_mode != TOPOLOGY_MODE_HW) {
260 			id = topology_mode == TOPOLOGY_MODE_PACKAGE ? 0 : cpu;
261 			topo->thread_id = cpu;
262 			topo->core_id = cpu;
263 			topo->socket_id = id;
264 			topo->book_id = id;
265 			topo->drawer_id = id;
266 		}
267 	}
268 	for_each_online_cpu(cpu) {
269 		topo = &cpu_topology[cpu];
270 		pkg_first = cpumask_first(&topo->core_mask);
271 		topo_package = &cpu_topology[pkg_first];
272 		if (cpu == pkg_first) {
273 			for_each_cpu(sibling, &topo->core_mask) {
274 				topo_sibling = &cpu_topology[sibling];
275 				smt_first = cpumask_first(&topo_sibling->thread_mask);
276 				if (sibling == smt_first)
277 					topo_package->booted_cores++;
278 			}
279 		} else {
280 			topo->booted_cores = topo_package->booted_cores;
281 		}
282 	}
283 }
284 
285 void store_topology(struct sysinfo_15_1_x *info)
286 {
287 	stsi(info, 15, 1, topology_mnest_limit());
288 }
289 
290 static void __arch_update_dedicated_flag(void *arg)
291 {
292 	if (topology_cpu_dedicated(smp_processor_id()))
293 		set_cpu_flag(CIF_DEDICATED_CPU);
294 	else
295 		clear_cpu_flag(CIF_DEDICATED_CPU);
296 }
297 
298 static int __arch_update_cpu_topology(void)
299 {
300 	struct sysinfo_15_1_x *info = tl_info;
301 	int rc = 0;
302 
303 	mutex_lock(&smp_cpu_state_mutex);
304 	if (MACHINE_HAS_TOPOLOGY) {
305 		rc = 1;
306 		store_topology(info);
307 		tl_to_masks(info);
308 	}
309 	update_cpu_masks();
310 	if (!MACHINE_HAS_TOPOLOGY)
311 		topology_update_polarization_simple();
312 	mutex_unlock(&smp_cpu_state_mutex);
313 	return rc;
314 }
315 
316 int arch_update_cpu_topology(void)
317 {
318 	struct device *dev;
319 	int cpu, rc;
320 
321 	rc = __arch_update_cpu_topology();
322 	on_each_cpu(__arch_update_dedicated_flag, NULL, 0);
323 	for_each_online_cpu(cpu) {
324 		dev = get_cpu_device(cpu);
325 		if (dev)
326 			kobject_uevent(&dev->kobj, KOBJ_CHANGE);
327 	}
328 	return rc;
329 }
330 
331 static void topology_work_fn(struct work_struct *work)
332 {
333 	rebuild_sched_domains();
334 }
335 
336 void topology_schedule_update(void)
337 {
338 	schedule_work(&topology_work);
339 }
340 
341 static void topology_flush_work(void)
342 {
343 	flush_work(&topology_work);
344 }
345 
346 static void topology_timer_fn(struct timer_list *unused)
347 {
348 	if (ptf(PTF_CHECK))
349 		topology_schedule_update();
350 	set_topology_timer();
351 }
352 
353 static struct timer_list topology_timer;
354 
355 static atomic_t topology_poll = ATOMIC_INIT(0);
356 
357 static void set_topology_timer(void)
358 {
359 	if (atomic_add_unless(&topology_poll, -1, 0))
360 		mod_timer(&topology_timer, jiffies + msecs_to_jiffies(100));
361 	else
362 		mod_timer(&topology_timer, jiffies + msecs_to_jiffies(60 * MSEC_PER_SEC));
363 }
364 
365 void topology_expect_change(void)
366 {
367 	if (!MACHINE_HAS_TOPOLOGY)
368 		return;
369 	/* This is racy, but it doesn't matter since it is just a heuristic.
370 	 * Worst case is that we poll in a higher frequency for a bit longer.
371 	 */
372 	if (atomic_read(&topology_poll) > 60)
373 		return;
374 	atomic_add(60, &topology_poll);
375 	set_topology_timer();
376 }
377 
378 static int cpu_management;
379 
380 static ssize_t dispatching_show(struct device *dev,
381 				struct device_attribute *attr,
382 				char *buf)
383 {
384 	ssize_t count;
385 
386 	mutex_lock(&smp_cpu_state_mutex);
387 	count = sprintf(buf, "%d\n", cpu_management);
388 	mutex_unlock(&smp_cpu_state_mutex);
389 	return count;
390 }
391 
392 static ssize_t dispatching_store(struct device *dev,
393 				 struct device_attribute *attr,
394 				 const char *buf,
395 				 size_t count)
396 {
397 	int val, rc;
398 	char delim;
399 
400 	if (sscanf(buf, "%d %c", &val, &delim) != 1)
401 		return -EINVAL;
402 	if (val != 0 && val != 1)
403 		return -EINVAL;
404 	rc = 0;
405 	get_online_cpus();
406 	mutex_lock(&smp_cpu_state_mutex);
407 	if (cpu_management == val)
408 		goto out;
409 	rc = topology_set_cpu_management(val);
410 	if (rc)
411 		goto out;
412 	cpu_management = val;
413 	topology_expect_change();
414 out:
415 	mutex_unlock(&smp_cpu_state_mutex);
416 	put_online_cpus();
417 	return rc ? rc : count;
418 }
419 static DEVICE_ATTR_RW(dispatching);
420 
421 static ssize_t cpu_polarization_show(struct device *dev,
422 				     struct device_attribute *attr, char *buf)
423 {
424 	int cpu = dev->id;
425 	ssize_t count;
426 
427 	mutex_lock(&smp_cpu_state_mutex);
428 	switch (smp_cpu_get_polarization(cpu)) {
429 	case POLARIZATION_HRZ:
430 		count = sprintf(buf, "horizontal\n");
431 		break;
432 	case POLARIZATION_VL:
433 		count = sprintf(buf, "vertical:low\n");
434 		break;
435 	case POLARIZATION_VM:
436 		count = sprintf(buf, "vertical:medium\n");
437 		break;
438 	case POLARIZATION_VH:
439 		count = sprintf(buf, "vertical:high\n");
440 		break;
441 	default:
442 		count = sprintf(buf, "unknown\n");
443 		break;
444 	}
445 	mutex_unlock(&smp_cpu_state_mutex);
446 	return count;
447 }
448 static DEVICE_ATTR(polarization, 0444, cpu_polarization_show, NULL);
449 
450 static struct attribute *topology_cpu_attrs[] = {
451 	&dev_attr_polarization.attr,
452 	NULL,
453 };
454 
455 static struct attribute_group topology_cpu_attr_group = {
456 	.attrs = topology_cpu_attrs,
457 };
458 
459 static ssize_t cpu_dedicated_show(struct device *dev,
460 				  struct device_attribute *attr, char *buf)
461 {
462 	int cpu = dev->id;
463 	ssize_t count;
464 
465 	mutex_lock(&smp_cpu_state_mutex);
466 	count = sprintf(buf, "%d\n", topology_cpu_dedicated(cpu));
467 	mutex_unlock(&smp_cpu_state_mutex);
468 	return count;
469 }
470 static DEVICE_ATTR(dedicated, 0444, cpu_dedicated_show, NULL);
471 
472 static struct attribute *topology_extra_cpu_attrs[] = {
473 	&dev_attr_dedicated.attr,
474 	NULL,
475 };
476 
477 static struct attribute_group topology_extra_cpu_attr_group = {
478 	.attrs = topology_extra_cpu_attrs,
479 };
480 
481 int topology_cpu_init(struct cpu *cpu)
482 {
483 	int rc;
484 
485 	rc = sysfs_create_group(&cpu->dev.kobj, &topology_cpu_attr_group);
486 	if (rc || !MACHINE_HAS_TOPOLOGY)
487 		return rc;
488 	rc = sysfs_create_group(&cpu->dev.kobj, &topology_extra_cpu_attr_group);
489 	if (rc)
490 		sysfs_remove_group(&cpu->dev.kobj, &topology_cpu_attr_group);
491 	return rc;
492 }
493 
494 static const struct cpumask *cpu_thread_mask(int cpu)
495 {
496 	return &cpu_topology[cpu].thread_mask;
497 }
498 
499 
500 const struct cpumask *cpu_coregroup_mask(int cpu)
501 {
502 	return &cpu_topology[cpu].core_mask;
503 }
504 
505 static const struct cpumask *cpu_book_mask(int cpu)
506 {
507 	return &cpu_topology[cpu].book_mask;
508 }
509 
510 static const struct cpumask *cpu_drawer_mask(int cpu)
511 {
512 	return &cpu_topology[cpu].drawer_mask;
513 }
514 
515 static struct sched_domain_topology_level s390_topology[] = {
516 	{ cpu_thread_mask, cpu_smt_flags, SD_INIT_NAME(SMT) },
517 	{ cpu_coregroup_mask, cpu_core_flags, SD_INIT_NAME(MC) },
518 	{ cpu_book_mask, SD_INIT_NAME(BOOK) },
519 	{ cpu_drawer_mask, SD_INIT_NAME(DRAWER) },
520 	{ cpu_cpu_mask, SD_INIT_NAME(DIE) },
521 	{ NULL, },
522 };
523 
524 static void __init alloc_masks(struct sysinfo_15_1_x *info,
525 			       struct mask_info *mask, int offset)
526 {
527 	int i, nr_masks;
528 
529 	nr_masks = info->mag[TOPOLOGY_NR_MAG - offset];
530 	for (i = 0; i < info->mnest - offset; i++)
531 		nr_masks *= info->mag[TOPOLOGY_NR_MAG - offset - 1 - i];
532 	nr_masks = max(nr_masks, 1);
533 	for (i = 0; i < nr_masks; i++) {
534 		mask->next = memblock_alloc(sizeof(*mask->next), 8);
535 		if (!mask->next)
536 			panic("%s: Failed to allocate %zu bytes align=0x%x\n",
537 			      __func__, sizeof(*mask->next), 8);
538 		mask = mask->next;
539 	}
540 }
541 
542 void __init topology_init_early(void)
543 {
544 	struct sysinfo_15_1_x *info;
545 
546 	set_sched_topology(s390_topology);
547 	if (topology_mode == TOPOLOGY_MODE_UNINITIALIZED) {
548 		if (MACHINE_HAS_TOPOLOGY)
549 			topology_mode = TOPOLOGY_MODE_HW;
550 		else
551 			topology_mode = TOPOLOGY_MODE_SINGLE;
552 	}
553 	if (!MACHINE_HAS_TOPOLOGY)
554 		goto out;
555 	tl_info = memblock_alloc(PAGE_SIZE, PAGE_SIZE);
556 	if (!tl_info)
557 		panic("%s: Failed to allocate %lu bytes align=0x%lx\n",
558 		      __func__, PAGE_SIZE, PAGE_SIZE);
559 	info = tl_info;
560 	store_topology(info);
561 	pr_info("The CPU configuration topology of the machine is: %d %d %d %d %d %d / %d\n",
562 		info->mag[0], info->mag[1], info->mag[2], info->mag[3],
563 		info->mag[4], info->mag[5], info->mnest);
564 	alloc_masks(info, &socket_info, 1);
565 	alloc_masks(info, &book_info, 2);
566 	alloc_masks(info, &drawer_info, 3);
567 out:
568 	__arch_update_cpu_topology();
569 	__arch_update_dedicated_flag(NULL);
570 }
571 
572 static inline int topology_get_mode(int enabled)
573 {
574 	if (!enabled)
575 		return TOPOLOGY_MODE_SINGLE;
576 	return MACHINE_HAS_TOPOLOGY ? TOPOLOGY_MODE_HW : TOPOLOGY_MODE_PACKAGE;
577 }
578 
579 static inline int topology_is_enabled(void)
580 {
581 	return topology_mode != TOPOLOGY_MODE_SINGLE;
582 }
583 
584 static int __init topology_setup(char *str)
585 {
586 	bool enabled;
587 	int rc;
588 
589 	rc = kstrtobool(str, &enabled);
590 	if (rc)
591 		return rc;
592 	topology_mode = topology_get_mode(enabled);
593 	return 0;
594 }
595 early_param("topology", topology_setup);
596 
597 static int topology_ctl_handler(struct ctl_table *ctl, int write,
598 				void *buffer, size_t *lenp, loff_t *ppos)
599 {
600 	int enabled = topology_is_enabled();
601 	int new_mode;
602 	int rc;
603 	struct ctl_table ctl_entry = {
604 		.procname	= ctl->procname,
605 		.data		= &enabled,
606 		.maxlen		= sizeof(int),
607 		.extra1		= SYSCTL_ZERO,
608 		.extra2		= SYSCTL_ONE,
609 	};
610 
611 	rc = proc_douintvec_minmax(&ctl_entry, write, buffer, lenp, ppos);
612 	if (rc < 0 || !write)
613 		return rc;
614 
615 	mutex_lock(&smp_cpu_state_mutex);
616 	new_mode = topology_get_mode(enabled);
617 	if (topology_mode != new_mode) {
618 		topology_mode = new_mode;
619 		topology_schedule_update();
620 	}
621 	mutex_unlock(&smp_cpu_state_mutex);
622 	topology_flush_work();
623 
624 	return rc;
625 }
626 
627 static struct ctl_table topology_ctl_table[] = {
628 	{
629 		.procname	= "topology",
630 		.mode		= 0644,
631 		.proc_handler	= topology_ctl_handler,
632 	},
633 	{ },
634 };
635 
636 static struct ctl_table topology_dir_table[] = {
637 	{
638 		.procname	= "s390",
639 		.maxlen		= 0,
640 		.mode		= 0555,
641 		.child		= topology_ctl_table,
642 	},
643 	{ },
644 };
645 
646 static int __init topology_init(void)
647 {
648 	timer_setup(&topology_timer, topology_timer_fn, TIMER_DEFERRABLE);
649 	if (MACHINE_HAS_TOPOLOGY)
650 		set_topology_timer();
651 	else
652 		topology_update_polarization_simple();
653 	register_sysctl_table(topology_dir_table);
654 	return device_create_file(cpu_subsys.dev_root, &dev_attr_dispatching);
655 }
656 device_initcall(topology_init);
657