1#include <linux/sysdev.h>
2#include <linux/cpu.h>
3#include <linux/smp.h>
4#include <linux/percpu.h>
5#include <linux/init.h>
6#include <linux/sched.h>
7#include <linux/module.h>
8#include <linux/nodemask.h>
9#include <linux/cpumask.h>
10#include <linux/notifier.h>
11
12#include <asm/current.h>
13#include <asm/processor.h>
14#include <asm/cputable.h>
15#include <asm/firmware.h>
16#include <asm/hvcall.h>
17#include <asm/prom.h>
18#include <asm/paca.h>
19#include <asm/lppaca.h>
20#include <asm/machdep.h>
21#include <asm/smp.h>
22
23static DEFINE_PER_CPU(struct cpu, cpu_devices);
24
25/* SMT stuff */
26
27#ifdef CONFIG_PPC_MULTIPLATFORM
28/* Time in microseconds we delay before sleeping in the idle loop */
29DEFINE_PER_CPU(unsigned long, smt_snooze_delay) = { 100 };
30
31static ssize_t store_smt_snooze_delay(struct sys_device *dev, const char *buf,
32				      size_t count)
33{
34	struct cpu *cpu = container_of(dev, struct cpu, sysdev);
35	ssize_t ret;
36	unsigned long snooze;
37
38	ret = sscanf(buf, "%lu", &snooze);
39	if (ret != 1)
40		return -EINVAL;
41
42	per_cpu(smt_snooze_delay, cpu->sysdev.id) = snooze;
43
44	return count;
45}
46
47static ssize_t show_smt_snooze_delay(struct sys_device *dev, char *buf)
48{
49	struct cpu *cpu = container_of(dev, struct cpu, sysdev);
50
51	return sprintf(buf, "%lu\n", per_cpu(smt_snooze_delay, cpu->sysdev.id));
52}
53
54static SYSDEV_ATTR(smt_snooze_delay, 0644, show_smt_snooze_delay,
55		   store_smt_snooze_delay);
56
57/* Only parse OF options if the matching cmdline option was not specified */
58static int smt_snooze_cmdline;
59
60static int __init smt_setup(void)
61{
62	struct device_node *options;
63	const unsigned int *val;
64	unsigned int cpu;
65
66	if (!cpu_has_feature(CPU_FTR_SMT))
67		return -ENODEV;
68
69	options = of_find_node_by_path("/options");
70	if (!options)
71		return -ENODEV;
72
73	val = of_get_property(options, "ibm,smt-snooze-delay", NULL);
74	if (!smt_snooze_cmdline && val) {
75		for_each_possible_cpu(cpu)
76			per_cpu(smt_snooze_delay, cpu) = *val;
77	}
78
79	of_node_put(options);
80	return 0;
81}
82__initcall(smt_setup);
83
84static int __init setup_smt_snooze_delay(char *str)
85{
86	unsigned int cpu;
87	int snooze;
88
89	if (!cpu_has_feature(CPU_FTR_SMT))
90		return 1;
91
92	smt_snooze_cmdline = 1;
93
94	if (get_option(&str, &snooze)) {
95		for_each_possible_cpu(cpu)
96			per_cpu(smt_snooze_delay, cpu) = snooze;
97	}
98
99	return 1;
100}
101__setup("smt-snooze-delay=", setup_smt_snooze_delay);
102
103#endif /* CONFIG_PPC_MULTIPLATFORM */
104
105/*
106 * Enabling PMCs will slow partition context switch times so we only do
107 * it the first time we write to the PMCs.
108 */
109
110static DEFINE_PER_CPU(char, pmcs_enabled);
111
112void ppc64_enable_pmcs(void)
113{
114	/* Only need to enable them once */
115	if (__get_cpu_var(pmcs_enabled))
116		return;
117
118	__get_cpu_var(pmcs_enabled) = 1;
119
120	if (ppc_md.enable_pmcs)
121		ppc_md.enable_pmcs();
122}
123EXPORT_SYMBOL(ppc64_enable_pmcs);
124
125static unsigned long run_on_cpu(unsigned long cpu,
126			        unsigned long (*func)(unsigned long),
127				unsigned long arg)
128{
129	cpumask_t old_affinity = current->cpus_allowed;
130	unsigned long ret;
131
132	/* should return -EINVAL to userspace */
133	if (set_cpus_allowed(current, cpumask_of_cpu(cpu)))
134		return 0;
135
136	ret = func(arg);
137
138	set_cpus_allowed(current, old_affinity);
139
140	return ret;
141}
142
143#define SYSFS_PMCSETUP(NAME, ADDRESS) \
144static unsigned long read_##NAME(unsigned long junk) \
145{ \
146	return mfspr(ADDRESS); \
147} \
148static unsigned long write_##NAME(unsigned long val) \
149{ \
150	ppc64_enable_pmcs(); \
151	mtspr(ADDRESS, val); \
152	return 0; \
153} \
154static ssize_t show_##NAME(struct sys_device *dev, char *buf) \
155{ \
156	struct cpu *cpu = container_of(dev, struct cpu, sysdev); \
157	unsigned long val = run_on_cpu(cpu->sysdev.id, read_##NAME, 0); \
158	return sprintf(buf, "%lx\n", val); \
159} \
160static ssize_t __attribute_used__ \
161	store_##NAME(struct sys_device *dev, const char *buf, size_t count) \
162{ \
163	struct cpu *cpu = container_of(dev, struct cpu, sysdev); \
164	unsigned long val; \
165	int ret = sscanf(buf, "%lx", &val); \
166	if (ret != 1) \
167		return -EINVAL; \
168	run_on_cpu(cpu->sysdev.id, write_##NAME, val); \
169	return count; \
170}
171
172
173/* Let's define all possible registers, we'll only hook up the ones
174 * that are implemented on the current processor
175 */
176
177SYSFS_PMCSETUP(mmcr0, SPRN_MMCR0);
178SYSFS_PMCSETUP(mmcr1, SPRN_MMCR1);
179SYSFS_PMCSETUP(mmcra, SPRN_MMCRA);
180SYSFS_PMCSETUP(pmc1, SPRN_PMC1);
181SYSFS_PMCSETUP(pmc2, SPRN_PMC2);
182SYSFS_PMCSETUP(pmc3, SPRN_PMC3);
183SYSFS_PMCSETUP(pmc4, SPRN_PMC4);
184SYSFS_PMCSETUP(pmc5, SPRN_PMC5);
185SYSFS_PMCSETUP(pmc6, SPRN_PMC6);
186SYSFS_PMCSETUP(pmc7, SPRN_PMC7);
187SYSFS_PMCSETUP(pmc8, SPRN_PMC8);
188SYSFS_PMCSETUP(purr, SPRN_PURR);
189SYSFS_PMCSETUP(spurr, SPRN_SPURR);
190SYSFS_PMCSETUP(dscr, SPRN_DSCR);
191
192SYSFS_PMCSETUP(pa6t_pmc0, SPRN_PA6T_PMC0);
193SYSFS_PMCSETUP(pa6t_pmc1, SPRN_PA6T_PMC1);
194SYSFS_PMCSETUP(pa6t_pmc2, SPRN_PA6T_PMC2);
195SYSFS_PMCSETUP(pa6t_pmc3, SPRN_PA6T_PMC3);
196SYSFS_PMCSETUP(pa6t_pmc4, SPRN_PA6T_PMC4);
197SYSFS_PMCSETUP(pa6t_pmc5, SPRN_PA6T_PMC5);
198
199
200static SYSDEV_ATTR(mmcra, 0600, show_mmcra, store_mmcra);
201static SYSDEV_ATTR(spurr, 0600, show_spurr, NULL);
202static SYSDEV_ATTR(dscr, 0600, show_dscr, store_dscr);
203static SYSDEV_ATTR(purr, 0600, show_purr, store_purr);
204
205static struct sysdev_attribute ibm_common_attrs[] = {
206	_SYSDEV_ATTR(mmcr0, 0600, show_mmcr0, store_mmcr0),
207	_SYSDEV_ATTR(mmcr1, 0600, show_mmcr1, store_mmcr1),
208};
209
210static struct sysdev_attribute ibm_pmc_attrs[] = {
211	_SYSDEV_ATTR(pmc1, 0600, show_pmc1, store_pmc1),
212	_SYSDEV_ATTR(pmc2, 0600, show_pmc2, store_pmc2),
213	_SYSDEV_ATTR(pmc3, 0600, show_pmc3, store_pmc3),
214	_SYSDEV_ATTR(pmc4, 0600, show_pmc4, store_pmc4),
215	_SYSDEV_ATTR(pmc5, 0600, show_pmc5, store_pmc5),
216	_SYSDEV_ATTR(pmc6, 0600, show_pmc6, store_pmc6),
217	_SYSDEV_ATTR(pmc7, 0600, show_pmc7, store_pmc7),
218	_SYSDEV_ATTR(pmc8, 0600, show_pmc8, store_pmc8),
219};
220
221static struct sysdev_attribute pa6t_attrs[] = {
222	_SYSDEV_ATTR(mmcr0, 0600, show_mmcr0, store_mmcr0),
223	_SYSDEV_ATTR(mmcr1, 0600, show_mmcr1, store_mmcr1),
224	_SYSDEV_ATTR(pmc0, 0600, show_pa6t_pmc0, store_pa6t_pmc0),
225	_SYSDEV_ATTR(pmc1, 0600, show_pa6t_pmc1, store_pa6t_pmc1),
226	_SYSDEV_ATTR(pmc2, 0600, show_pa6t_pmc2, store_pa6t_pmc2),
227	_SYSDEV_ATTR(pmc3, 0600, show_pa6t_pmc3, store_pa6t_pmc3),
228	_SYSDEV_ATTR(pmc4, 0600, show_pa6t_pmc4, store_pa6t_pmc4),
229	_SYSDEV_ATTR(pmc5, 0600, show_pa6t_pmc5, store_pa6t_pmc5),
230};
231
232
233static void register_cpu_online(unsigned int cpu)
234{
235	struct cpu *c = &per_cpu(cpu_devices, cpu);
236	struct sys_device *s = &c->sysdev;
237	struct sysdev_attribute *attrs, *pmc_attrs;
238	int i, nattrs;
239
240	if (!firmware_has_feature(FW_FEATURE_ISERIES) &&
241			cpu_has_feature(CPU_FTR_SMT))
242		sysdev_create_file(s, &attr_smt_snooze_delay);
243
244	/* PMC stuff */
245	switch (cur_cpu_spec->pmc_type) {
246	case PPC_PMC_IBM:
247		attrs = ibm_common_attrs;
248		nattrs = sizeof(ibm_common_attrs) / sizeof(struct sysdev_attribute);
249		pmc_attrs = ibm_pmc_attrs;
250		break;
251	case PPC_PMC_PA6T:
252		/* PA Semi starts counting at PMC0 */
253		attrs = pa6t_attrs;
254		nattrs = sizeof(pa6t_attrs) / sizeof(struct sysdev_attribute);
255		pmc_attrs = NULL;
256		break;
257	default:
258		attrs = NULL;
259		nattrs = 0;
260		pmc_attrs = NULL;
261	}
262
263	for (i = 0; i < nattrs; i++)
264		sysdev_create_file(s, &attrs[i]);
265
266	if (pmc_attrs)
267		for (i = 0; i < cur_cpu_spec->num_pmcs; i++)
268			sysdev_create_file(s, &pmc_attrs[i]);
269
270	if (cpu_has_feature(CPU_FTR_MMCRA))
271		sysdev_create_file(s, &attr_mmcra);
272
273	if (cpu_has_feature(CPU_FTR_PURR))
274		sysdev_create_file(s, &attr_purr);
275
276	if (cpu_has_feature(CPU_FTR_SPURR))
277		sysdev_create_file(s, &attr_spurr);
278
279	if (cpu_has_feature(CPU_FTR_DSCR))
280		sysdev_create_file(s, &attr_dscr);
281}
282
283#ifdef CONFIG_HOTPLUG_CPU
284static void unregister_cpu_online(unsigned int cpu)
285{
286	struct cpu *c = &per_cpu(cpu_devices, cpu);
287	struct sys_device *s = &c->sysdev;
288	struct sysdev_attribute *attrs, *pmc_attrs;
289	int i, nattrs;
290
291	BUG_ON(!c->hotpluggable);
292
293	if (!firmware_has_feature(FW_FEATURE_ISERIES) &&
294			cpu_has_feature(CPU_FTR_SMT))
295		sysdev_remove_file(s, &attr_smt_snooze_delay);
296
297	/* PMC stuff */
298	switch (cur_cpu_spec->pmc_type) {
299	case PPC_PMC_IBM:
300		attrs = ibm_common_attrs;
301		nattrs = sizeof(ibm_common_attrs) / sizeof(struct sysdev_attribute);
302		pmc_attrs = ibm_pmc_attrs;
303		break;
304	case PPC_PMC_PA6T:
305		/* PA Semi starts counting at PMC0 */
306		attrs = pa6t_attrs;
307		nattrs = sizeof(pa6t_attrs) / sizeof(struct sysdev_attribute);
308		pmc_attrs = NULL;
309		break;
310	default:
311		attrs = NULL;
312		nattrs = 0;
313		pmc_attrs = NULL;
314	}
315
316	for (i = 0; i < nattrs; i++)
317		sysdev_remove_file(s, &attrs[i]);
318
319	if (pmc_attrs)
320		for (i = 0; i < cur_cpu_spec->num_pmcs; i++)
321			sysdev_remove_file(s, &pmc_attrs[i]);
322
323	if (cpu_has_feature(CPU_FTR_MMCRA))
324		sysdev_remove_file(s, &attr_mmcra);
325
326	if (cpu_has_feature(CPU_FTR_PURR))
327		sysdev_remove_file(s, &attr_purr);
328
329	if (cpu_has_feature(CPU_FTR_SPURR))
330		sysdev_remove_file(s, &attr_spurr);
331
332	if (cpu_has_feature(CPU_FTR_DSCR))
333		sysdev_remove_file(s, &attr_dscr);
334}
335#endif /* CONFIG_HOTPLUG_CPU */
336
337static int __cpuinit sysfs_cpu_notify(struct notifier_block *self,
338				      unsigned long action, void *hcpu)
339{
340	unsigned int cpu = (unsigned int)(long)hcpu;
341
342	switch (action) {
343	case CPU_ONLINE:
344	case CPU_ONLINE_FROZEN:
345		register_cpu_online(cpu);
346		break;
347#ifdef CONFIG_HOTPLUG_CPU
348	case CPU_DEAD:
349	case CPU_DEAD_FROZEN:
350		unregister_cpu_online(cpu);
351		break;
352#endif
353	}
354	return NOTIFY_OK;
355}
356
357static struct notifier_block __cpuinitdata sysfs_cpu_nb = {
358	.notifier_call	= sysfs_cpu_notify,
359};
360
361static DEFINE_MUTEX(cpu_mutex);
362
363int cpu_add_sysdev_attr(struct sysdev_attribute *attr)
364{
365	int cpu;
366
367	mutex_lock(&cpu_mutex);
368
369	for_each_possible_cpu(cpu) {
370		sysdev_create_file(get_cpu_sysdev(cpu), attr);
371	}
372
373	mutex_unlock(&cpu_mutex);
374	return 0;
375}
376EXPORT_SYMBOL_GPL(cpu_add_sysdev_attr);
377
378int cpu_add_sysdev_attr_group(struct attribute_group *attrs)
379{
380	int cpu;
381	struct sys_device *sysdev;
382
383	mutex_lock(&cpu_mutex);
384
385	for_each_possible_cpu(cpu) {
386		sysdev = get_cpu_sysdev(cpu);
387		sysfs_create_group(&sysdev->kobj, attrs);
388	}
389
390	mutex_unlock(&cpu_mutex);
391	return 0;
392}
393EXPORT_SYMBOL_GPL(cpu_add_sysdev_attr_group);
394
395
396void cpu_remove_sysdev_attr(struct sysdev_attribute *attr)
397{
398	int cpu;
399
400	mutex_lock(&cpu_mutex);
401
402	for_each_possible_cpu(cpu) {
403		sysdev_remove_file(get_cpu_sysdev(cpu), attr);
404	}
405
406	mutex_unlock(&cpu_mutex);
407}
408EXPORT_SYMBOL_GPL(cpu_remove_sysdev_attr);
409
410void cpu_remove_sysdev_attr_group(struct attribute_group *attrs)
411{
412	int cpu;
413	struct sys_device *sysdev;
414
415	mutex_lock(&cpu_mutex);
416
417	for_each_possible_cpu(cpu) {
418		sysdev = get_cpu_sysdev(cpu);
419		sysfs_remove_group(&sysdev->kobj, attrs);
420	}
421
422	mutex_unlock(&cpu_mutex);
423}
424EXPORT_SYMBOL_GPL(cpu_remove_sysdev_attr_group);
425
426
427/* NUMA stuff */
428
429#ifdef CONFIG_NUMA
430static void register_nodes(void)
431{
432	int i;
433
434	for (i = 0; i < MAX_NUMNODES; i++)
435		register_one_node(i);
436}
437
438int sysfs_add_device_to_node(struct sys_device *dev, int nid)
439{
440	struct node *node = &node_devices[nid];
441	return sysfs_create_link(&node->sysdev.kobj, &dev->kobj,
442			kobject_name(&dev->kobj));
443}
444
445void sysfs_remove_device_from_node(struct sys_device *dev, int nid)
446{
447	struct node *node = &node_devices[nid];
448	sysfs_remove_link(&node->sysdev.kobj, kobject_name(&dev->kobj));
449}
450
451#else
452static void register_nodes(void)
453{
454	return;
455}
456
457#endif
458
459EXPORT_SYMBOL_GPL(sysfs_add_device_to_node);
460EXPORT_SYMBOL_GPL(sysfs_remove_device_from_node);
461
462/* Only valid if CPU is present. */
463static ssize_t show_physical_id(struct sys_device *dev, char *buf)
464{
465	struct cpu *cpu = container_of(dev, struct cpu, sysdev);
466
467	return sprintf(buf, "%d\n", get_hard_smp_processor_id(cpu->sysdev.id));
468}
469static SYSDEV_ATTR(physical_id, 0444, show_physical_id, NULL);
470
471static int __init topology_init(void)
472{
473	int cpu;
474
475	register_nodes();
476	register_cpu_notifier(&sysfs_cpu_nb);
477
478	for_each_possible_cpu(cpu) {
479		struct cpu *c = &per_cpu(cpu_devices, cpu);
480
481		/*
482		 * For now, we just see if the system supports making
483		 * the RTAS calls for CPU hotplug.  But, there may be a
484		 * more comprehensive way to do this for an individual
485		 * CPU.  For instance, the boot cpu might never be valid
486		 * for hotplugging.
487		 */
488		if (ppc_md.cpu_die)
489			c->hotpluggable = 1;
490
491		if (cpu_online(cpu) || c->hotpluggable) {
492			register_cpu(c, cpu);
493
494			sysdev_create_file(&c->sysdev, &attr_physical_id);
495		}
496
497		if (cpu_online(cpu))
498			register_cpu_online(cpu);
499	}
500
501	return 0;
502}
503subsys_initcall(topology_init);
504