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