xref: /linux/arch/s390/kernel/topology.c (revision 96ac6d435100450f0565708d9b885ea2a7400e0a)
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 #include <asm/numa.h>
30 
31 #define PTF_HORIZONTAL	(0UL)
32 #define PTF_VERTICAL	(1UL)
33 #define PTF_CHECK	(2UL)
34 
35 enum {
36 	TOPOLOGY_MODE_HW,
37 	TOPOLOGY_MODE_SINGLE,
38 	TOPOLOGY_MODE_PACKAGE,
39 	TOPOLOGY_MODE_UNINITIALIZED
40 };
41 
42 struct mask_info {
43 	struct mask_info *next;
44 	unsigned char id;
45 	cpumask_t mask;
46 };
47 
48 static int topology_mode = TOPOLOGY_MODE_UNINITIALIZED;
49 static void set_topology_timer(void);
50 static void topology_work_fn(struct work_struct *work);
51 static struct sysinfo_15_1_x *tl_info;
52 
53 static DECLARE_WORK(topology_work, topology_work_fn);
54 
55 /*
56  * Socket/Book linked lists and cpu_topology updates are
57  * protected by "sched_domains_mutex".
58  */
59 static struct mask_info socket_info;
60 static struct mask_info book_info;
61 static struct mask_info drawer_info;
62 
63 struct cpu_topology_s390 cpu_topology[NR_CPUS];
64 EXPORT_SYMBOL_GPL(cpu_topology);
65 
66 cpumask_t cpus_with_topology;
67 
68 static cpumask_t cpu_group_map(struct mask_info *info, unsigned int cpu)
69 {
70 	cpumask_t mask;
71 
72 	cpumask_copy(&mask, cpumask_of(cpu));
73 	switch (topology_mode) {
74 	case TOPOLOGY_MODE_HW:
75 		while (info) {
76 			if (cpumask_test_cpu(cpu, &info->mask)) {
77 				mask = info->mask;
78 				break;
79 			}
80 			info = info->next;
81 		}
82 		if (cpumask_empty(&mask))
83 			cpumask_copy(&mask, cpumask_of(cpu));
84 		break;
85 	case TOPOLOGY_MODE_PACKAGE:
86 		cpumask_copy(&mask, cpu_present_mask);
87 		break;
88 	default:
89 		/* fallthrough */
90 	case TOPOLOGY_MODE_SINGLE:
91 		cpumask_copy(&mask, cpumask_of(cpu));
92 		break;
93 	}
94 	return mask;
95 }
96 
97 static cpumask_t cpu_thread_map(unsigned int cpu)
98 {
99 	cpumask_t mask;
100 	int i;
101 
102 	cpumask_copy(&mask, cpumask_of(cpu));
103 	if (topology_mode != TOPOLOGY_MODE_HW)
104 		return mask;
105 	cpu -= cpu % (smp_cpu_mtid + 1);
106 	for (i = 0; i <= smp_cpu_mtid; i++)
107 		if (cpu_present(cpu + i))
108 			cpumask_set_cpu(cpu + i, &mask);
109 	return 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 			cpumask_set_cpu(lcpu + i, &cpus_with_topology);
142 			smp_cpu_set_polarization(lcpu + i, tl_core->pp);
143 		}
144 	}
145 }
146 
147 static void clear_masks(void)
148 {
149 	struct mask_info *info;
150 
151 	info = &socket_info;
152 	while (info) {
153 		cpumask_clear(&info->mask);
154 		info = info->next;
155 	}
156 	info = &book_info;
157 	while (info) {
158 		cpumask_clear(&info->mask);
159 		info = info->next;
160 	}
161 	info = &drawer_info;
162 	while (info) {
163 		cpumask_clear(&info->mask);
164 		info = info->next;
165 	}
166 }
167 
168 static union topology_entry *next_tle(union topology_entry *tle)
169 {
170 	if (!tle->nl)
171 		return (union topology_entry *)((struct topology_core *)tle + 1);
172 	return (union topology_entry *)((struct topology_container *)tle + 1);
173 }
174 
175 static void tl_to_masks(struct sysinfo_15_1_x *info)
176 {
177 	struct mask_info *socket = &socket_info;
178 	struct mask_info *book = &book_info;
179 	struct mask_info *drawer = &drawer_info;
180 	union topology_entry *tle, *end;
181 
182 	clear_masks();
183 	tle = info->tle;
184 	end = (union topology_entry *)((unsigned long)info + info->length);
185 	while (tle < end) {
186 		switch (tle->nl) {
187 		case 3:
188 			drawer = drawer->next;
189 			drawer->id = tle->container.id;
190 			break;
191 		case 2:
192 			book = book->next;
193 			book->id = tle->container.id;
194 			break;
195 		case 1:
196 			socket = socket->next;
197 			socket->id = tle->container.id;
198 			break;
199 		case 0:
200 			add_cpus_to_mask(&tle->cpu, drawer, book, socket);
201 			break;
202 		default:
203 			clear_masks();
204 			return;
205 		}
206 		tle = next_tle(tle);
207 	}
208 }
209 
210 static void topology_update_polarization_simple(void)
211 {
212 	int cpu;
213 
214 	for_each_possible_cpu(cpu)
215 		smp_cpu_set_polarization(cpu, POLARIZATION_HRZ);
216 }
217 
218 static int ptf(unsigned long fc)
219 {
220 	int rc;
221 
222 	asm volatile(
223 		"	.insn	rre,0xb9a20000,%1,%1\n"
224 		"	ipm	%0\n"
225 		"	srl	%0,28\n"
226 		: "=d" (rc)
227 		: "d" (fc)  : "cc");
228 	return rc;
229 }
230 
231 int topology_set_cpu_management(int fc)
232 {
233 	int cpu, rc;
234 
235 	if (!MACHINE_HAS_TOPOLOGY)
236 		return -EOPNOTSUPP;
237 	if (fc)
238 		rc = ptf(PTF_VERTICAL);
239 	else
240 		rc = ptf(PTF_HORIZONTAL);
241 	if (rc)
242 		return -EBUSY;
243 	for_each_possible_cpu(cpu)
244 		smp_cpu_set_polarization(cpu, POLARIZATION_UNKNOWN);
245 	return rc;
246 }
247 
248 static void update_cpu_masks(void)
249 {
250 	struct cpu_topology_s390 *topo;
251 	int cpu, id;
252 
253 	for_each_possible_cpu(cpu) {
254 		topo = &cpu_topology[cpu];
255 		topo->thread_mask = cpu_thread_map(cpu);
256 		topo->core_mask = cpu_group_map(&socket_info, cpu);
257 		topo->book_mask = cpu_group_map(&book_info, cpu);
258 		topo->drawer_mask = cpu_group_map(&drawer_info, cpu);
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 			if (cpu_present(cpu))
267 				cpumask_set_cpu(cpu, &cpus_with_topology);
268 		}
269 	}
270 	numa_update_cpu_topology();
271 }
272 
273 void store_topology(struct sysinfo_15_1_x *info)
274 {
275 	stsi(info, 15, 1, topology_mnest_limit());
276 }
277 
278 static void __arch_update_dedicated_flag(void *arg)
279 {
280 	if (topology_cpu_dedicated(smp_processor_id()))
281 		set_cpu_flag(CIF_DEDICATED_CPU);
282 	else
283 		clear_cpu_flag(CIF_DEDICATED_CPU);
284 }
285 
286 static int __arch_update_cpu_topology(void)
287 {
288 	struct sysinfo_15_1_x *info = tl_info;
289 	int rc = 0;
290 
291 	mutex_lock(&smp_cpu_state_mutex);
292 	cpumask_clear(&cpus_with_topology);
293 	if (MACHINE_HAS_TOPOLOGY) {
294 		rc = 1;
295 		store_topology(info);
296 		tl_to_masks(info);
297 	}
298 	update_cpu_masks();
299 	if (!MACHINE_HAS_TOPOLOGY)
300 		topology_update_polarization_simple();
301 	mutex_unlock(&smp_cpu_state_mutex);
302 	return rc;
303 }
304 
305 int arch_update_cpu_topology(void)
306 {
307 	struct device *dev;
308 	int cpu, rc;
309 
310 	rc = __arch_update_cpu_topology();
311 	on_each_cpu(__arch_update_dedicated_flag, NULL, 0);
312 	for_each_online_cpu(cpu) {
313 		dev = get_cpu_device(cpu);
314 		kobject_uevent(&dev->kobj, KOBJ_CHANGE);
315 	}
316 	return rc;
317 }
318 
319 static void topology_work_fn(struct work_struct *work)
320 {
321 	rebuild_sched_domains();
322 }
323 
324 void topology_schedule_update(void)
325 {
326 	schedule_work(&topology_work);
327 }
328 
329 static void topology_flush_work(void)
330 {
331 	flush_work(&topology_work);
332 }
333 
334 static void topology_timer_fn(struct timer_list *unused)
335 {
336 	if (ptf(PTF_CHECK))
337 		topology_schedule_update();
338 	set_topology_timer();
339 }
340 
341 static struct timer_list topology_timer;
342 
343 static atomic_t topology_poll = ATOMIC_INIT(0);
344 
345 static void set_topology_timer(void)
346 {
347 	if (atomic_add_unless(&topology_poll, -1, 0))
348 		mod_timer(&topology_timer, jiffies + HZ / 10);
349 	else
350 		mod_timer(&topology_timer, jiffies + HZ * 60);
351 }
352 
353 void topology_expect_change(void)
354 {
355 	if (!MACHINE_HAS_TOPOLOGY)
356 		return;
357 	/* This is racy, but it doesn't matter since it is just a heuristic.
358 	 * Worst case is that we poll in a higher frequency for a bit longer.
359 	 */
360 	if (atomic_read(&topology_poll) > 60)
361 		return;
362 	atomic_add(60, &topology_poll);
363 	set_topology_timer();
364 }
365 
366 static int cpu_management;
367 
368 static ssize_t dispatching_show(struct device *dev,
369 				struct device_attribute *attr,
370 				char *buf)
371 {
372 	ssize_t count;
373 
374 	mutex_lock(&smp_cpu_state_mutex);
375 	count = sprintf(buf, "%d\n", cpu_management);
376 	mutex_unlock(&smp_cpu_state_mutex);
377 	return count;
378 }
379 
380 static ssize_t dispatching_store(struct device *dev,
381 				 struct device_attribute *attr,
382 				 const char *buf,
383 				 size_t count)
384 {
385 	int val, rc;
386 	char delim;
387 
388 	if (sscanf(buf, "%d %c", &val, &delim) != 1)
389 		return -EINVAL;
390 	if (val != 0 && val != 1)
391 		return -EINVAL;
392 	rc = 0;
393 	get_online_cpus();
394 	mutex_lock(&smp_cpu_state_mutex);
395 	if (cpu_management == val)
396 		goto out;
397 	rc = topology_set_cpu_management(val);
398 	if (rc)
399 		goto out;
400 	cpu_management = val;
401 	topology_expect_change();
402 out:
403 	mutex_unlock(&smp_cpu_state_mutex);
404 	put_online_cpus();
405 	return rc ? rc : count;
406 }
407 static DEVICE_ATTR_RW(dispatching);
408 
409 static ssize_t cpu_polarization_show(struct device *dev,
410 				     struct device_attribute *attr, char *buf)
411 {
412 	int cpu = dev->id;
413 	ssize_t count;
414 
415 	mutex_lock(&smp_cpu_state_mutex);
416 	switch (smp_cpu_get_polarization(cpu)) {
417 	case POLARIZATION_HRZ:
418 		count = sprintf(buf, "horizontal\n");
419 		break;
420 	case POLARIZATION_VL:
421 		count = sprintf(buf, "vertical:low\n");
422 		break;
423 	case POLARIZATION_VM:
424 		count = sprintf(buf, "vertical:medium\n");
425 		break;
426 	case POLARIZATION_VH:
427 		count = sprintf(buf, "vertical:high\n");
428 		break;
429 	default:
430 		count = sprintf(buf, "unknown\n");
431 		break;
432 	}
433 	mutex_unlock(&smp_cpu_state_mutex);
434 	return count;
435 }
436 static DEVICE_ATTR(polarization, 0444, cpu_polarization_show, NULL);
437 
438 static struct attribute *topology_cpu_attrs[] = {
439 	&dev_attr_polarization.attr,
440 	NULL,
441 };
442 
443 static struct attribute_group topology_cpu_attr_group = {
444 	.attrs = topology_cpu_attrs,
445 };
446 
447 static ssize_t cpu_dedicated_show(struct device *dev,
448 				  struct device_attribute *attr, char *buf)
449 {
450 	int cpu = dev->id;
451 	ssize_t count;
452 
453 	mutex_lock(&smp_cpu_state_mutex);
454 	count = sprintf(buf, "%d\n", topology_cpu_dedicated(cpu));
455 	mutex_unlock(&smp_cpu_state_mutex);
456 	return count;
457 }
458 static DEVICE_ATTR(dedicated, 0444, cpu_dedicated_show, NULL);
459 
460 static struct attribute *topology_extra_cpu_attrs[] = {
461 	&dev_attr_dedicated.attr,
462 	NULL,
463 };
464 
465 static struct attribute_group topology_extra_cpu_attr_group = {
466 	.attrs = topology_extra_cpu_attrs,
467 };
468 
469 int topology_cpu_init(struct cpu *cpu)
470 {
471 	int rc;
472 
473 	rc = sysfs_create_group(&cpu->dev.kobj, &topology_cpu_attr_group);
474 	if (rc || !MACHINE_HAS_TOPOLOGY)
475 		return rc;
476 	rc = sysfs_create_group(&cpu->dev.kobj, &topology_extra_cpu_attr_group);
477 	if (rc)
478 		sysfs_remove_group(&cpu->dev.kobj, &topology_cpu_attr_group);
479 	return rc;
480 }
481 
482 static const struct cpumask *cpu_thread_mask(int cpu)
483 {
484 	return &cpu_topology[cpu].thread_mask;
485 }
486 
487 
488 const struct cpumask *cpu_coregroup_mask(int cpu)
489 {
490 	return &cpu_topology[cpu].core_mask;
491 }
492 
493 static const struct cpumask *cpu_book_mask(int cpu)
494 {
495 	return &cpu_topology[cpu].book_mask;
496 }
497 
498 static const struct cpumask *cpu_drawer_mask(int cpu)
499 {
500 	return &cpu_topology[cpu].drawer_mask;
501 }
502 
503 static struct sched_domain_topology_level s390_topology[] = {
504 	{ cpu_thread_mask, cpu_smt_flags, SD_INIT_NAME(SMT) },
505 	{ cpu_coregroup_mask, cpu_core_flags, SD_INIT_NAME(MC) },
506 	{ cpu_book_mask, SD_INIT_NAME(BOOK) },
507 	{ cpu_drawer_mask, SD_INIT_NAME(DRAWER) },
508 	{ cpu_cpu_mask, SD_INIT_NAME(DIE) },
509 	{ NULL, },
510 };
511 
512 static void __init alloc_masks(struct sysinfo_15_1_x *info,
513 			       struct mask_info *mask, int offset)
514 {
515 	int i, nr_masks;
516 
517 	nr_masks = info->mag[TOPOLOGY_NR_MAG - offset];
518 	for (i = 0; i < info->mnest - offset; i++)
519 		nr_masks *= info->mag[TOPOLOGY_NR_MAG - offset - 1 - i];
520 	nr_masks = max(nr_masks, 1);
521 	for (i = 0; i < nr_masks; i++) {
522 		mask->next = memblock_alloc(sizeof(*mask->next), 8);
523 		if (!mask->next)
524 			panic("%s: Failed to allocate %zu bytes align=0x%x\n",
525 			      __func__, sizeof(*mask->next), 8);
526 		mask = mask->next;
527 	}
528 }
529 
530 void __init topology_init_early(void)
531 {
532 	struct sysinfo_15_1_x *info;
533 
534 	set_sched_topology(s390_topology);
535 	if (topology_mode == TOPOLOGY_MODE_UNINITIALIZED) {
536 		if (MACHINE_HAS_TOPOLOGY)
537 			topology_mode = TOPOLOGY_MODE_HW;
538 		else
539 			topology_mode = TOPOLOGY_MODE_SINGLE;
540 	}
541 	if (!MACHINE_HAS_TOPOLOGY)
542 		goto out;
543 	tl_info = memblock_alloc(PAGE_SIZE, PAGE_SIZE);
544 	if (!tl_info)
545 		panic("%s: Failed to allocate %lu bytes align=0x%lx\n",
546 		      __func__, PAGE_SIZE, PAGE_SIZE);
547 	info = tl_info;
548 	store_topology(info);
549 	pr_info("The CPU configuration topology of the machine is: %d %d %d %d %d %d / %d\n",
550 		info->mag[0], info->mag[1], info->mag[2], info->mag[3],
551 		info->mag[4], info->mag[5], info->mnest);
552 	alloc_masks(info, &socket_info, 1);
553 	alloc_masks(info, &book_info, 2);
554 	alloc_masks(info, &drawer_info, 3);
555 out:
556 	__arch_update_cpu_topology();
557 	__arch_update_dedicated_flag(NULL);
558 }
559 
560 static inline int topology_get_mode(int enabled)
561 {
562 	if (!enabled)
563 		return TOPOLOGY_MODE_SINGLE;
564 	return MACHINE_HAS_TOPOLOGY ? TOPOLOGY_MODE_HW : TOPOLOGY_MODE_PACKAGE;
565 }
566 
567 static inline int topology_is_enabled(void)
568 {
569 	return topology_mode != TOPOLOGY_MODE_SINGLE;
570 }
571 
572 static int __init topology_setup(char *str)
573 {
574 	bool enabled;
575 	int rc;
576 
577 	rc = kstrtobool(str, &enabled);
578 	if (rc)
579 		return rc;
580 	topology_mode = topology_get_mode(enabled);
581 	return 0;
582 }
583 early_param("topology", topology_setup);
584 
585 static int topology_ctl_handler(struct ctl_table *ctl, int write,
586 				void __user *buffer, size_t *lenp, loff_t *ppos)
587 {
588 	int enabled = topology_is_enabled();
589 	int new_mode;
590 	int zero = 0;
591 	int one = 1;
592 	int rc;
593 	struct ctl_table ctl_entry = {
594 		.procname	= ctl->procname,
595 		.data		= &enabled,
596 		.maxlen		= sizeof(int),
597 		.extra1		= &zero,
598 		.extra2		= &one,
599 	};
600 
601 	rc = proc_douintvec_minmax(&ctl_entry, write, buffer, lenp, ppos);
602 	if (rc < 0 || !write)
603 		return rc;
604 
605 	mutex_lock(&smp_cpu_state_mutex);
606 	new_mode = topology_get_mode(enabled);
607 	if (topology_mode != new_mode) {
608 		topology_mode = new_mode;
609 		topology_schedule_update();
610 	}
611 	mutex_unlock(&smp_cpu_state_mutex);
612 	topology_flush_work();
613 
614 	return rc;
615 }
616 
617 static struct ctl_table topology_ctl_table[] = {
618 	{
619 		.procname	= "topology",
620 		.mode		= 0644,
621 		.proc_handler	= topology_ctl_handler,
622 	},
623 	{ },
624 };
625 
626 static struct ctl_table topology_dir_table[] = {
627 	{
628 		.procname	= "s390",
629 		.maxlen		= 0,
630 		.mode		= 0555,
631 		.child		= topology_ctl_table,
632 	},
633 	{ },
634 };
635 
636 static int __init topology_init(void)
637 {
638 	timer_setup(&topology_timer, topology_timer_fn, TIMER_DEFERRABLE);
639 	if (MACHINE_HAS_TOPOLOGY)
640 		set_topology_timer();
641 	else
642 		topology_update_polarization_simple();
643 	register_sysctl_table(topology_dir_table);
644 	return device_create_file(cpu_subsys.dev_root, &dev_attr_dispatching);
645 }
646 device_initcall(topology_init);
647