xref: /linux/arch/s390/kernel/smp.c (revision d524dac9279b6a41ffdf7ff7958c577f2e387db6)
1 /*
2  *  arch/s390/kernel/smp.c
3  *
4  *    Copyright IBM Corp. 1999, 2009
5  *    Author(s): Denis Joseph Barrow (djbarrow@de.ibm.com,barrow_dj@yahoo.com),
6  *		 Martin Schwidefsky (schwidefsky@de.ibm.com)
7  *		 Heiko Carstens (heiko.carstens@de.ibm.com)
8  *
9  *  based on other smp stuff by
10  *    (c) 1995 Alan Cox, CymruNET Ltd  <alan@cymru.net>
11  *    (c) 1998 Ingo Molnar
12  *
13  * We work with logical cpu numbering everywhere we can. The only
14  * functions using the real cpu address (got from STAP) are the sigp
15  * functions. For all other functions we use the identity mapping.
16  * That means that cpu_number_map[i] == i for every cpu. cpu_number_map is
17  * used e.g. to find the idle task belonging to a logical cpu. Every array
18  * in the kernel is sorted by the logical cpu number and not by the physical
19  * one which is causing all the confusion with __cpu_logical_map and
20  * cpu_number_map in other architectures.
21  */
22 
23 #define KMSG_COMPONENT "cpu"
24 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
25 
26 #include <linux/workqueue.h>
27 #include <linux/module.h>
28 #include <linux/init.h>
29 #include <linux/mm.h>
30 #include <linux/err.h>
31 #include <linux/spinlock.h>
32 #include <linux/kernel_stat.h>
33 #include <linux/delay.h>
34 #include <linux/cache.h>
35 #include <linux/interrupt.h>
36 #include <linux/irqflags.h>
37 #include <linux/cpu.h>
38 #include <linux/timex.h>
39 #include <linux/bootmem.h>
40 #include <linux/slab.h>
41 #include <asm/asm-offsets.h>
42 #include <asm/ipl.h>
43 #include <asm/setup.h>
44 #include <asm/sigp.h>
45 #include <asm/pgalloc.h>
46 #include <asm/irq.h>
47 #include <asm/s390_ext.h>
48 #include <asm/cpcmd.h>
49 #include <asm/tlbflush.h>
50 #include <asm/timer.h>
51 #include <asm/lowcore.h>
52 #include <asm/sclp.h>
53 #include <asm/cputime.h>
54 #include <asm/vdso.h>
55 #include <asm/cpu.h>
56 #include "entry.h"
57 
58 /* logical cpu to cpu address */
59 unsigned short __cpu_logical_map[NR_CPUS];
60 
61 static struct task_struct *current_set[NR_CPUS];
62 
63 static u8 smp_cpu_type;
64 static int smp_use_sigp_detection;
65 
66 enum s390_cpu_state {
67 	CPU_STATE_STANDBY,
68 	CPU_STATE_CONFIGURED,
69 };
70 
71 DEFINE_MUTEX(smp_cpu_state_mutex);
72 int smp_cpu_polarization[NR_CPUS];
73 static int smp_cpu_state[NR_CPUS];
74 static int cpu_management;
75 
76 static DEFINE_PER_CPU(struct cpu, cpu_devices);
77 
78 static void smp_ext_bitcall(int, int);
79 
80 static int raw_cpu_stopped(int cpu)
81 {
82 	u32 status;
83 
84 	switch (raw_sigp_ps(&status, 0, cpu, sigp_sense)) {
85 	case sigp_status_stored:
86 		/* Check for stopped and check stop state */
87 		if (status & 0x50)
88 			return 1;
89 		break;
90 	default:
91 		break;
92 	}
93 	return 0;
94 }
95 
96 static inline int cpu_stopped(int cpu)
97 {
98 	return raw_cpu_stopped(cpu_logical_map(cpu));
99 }
100 
101 void smp_switch_to_ipl_cpu(void (*func)(void *), void *data)
102 {
103 	struct _lowcore *lc, *current_lc;
104 	struct stack_frame *sf;
105 	struct pt_regs *regs;
106 	unsigned long sp;
107 
108 	if (smp_processor_id() == 0)
109 		func(data);
110 	__load_psw_mask(PSW_BASE_BITS | PSW_DEFAULT_KEY);
111 	/* Disable lowcore protection */
112 	__ctl_clear_bit(0, 28);
113 	current_lc = lowcore_ptr[smp_processor_id()];
114 	lc = lowcore_ptr[0];
115 	if (!lc)
116 		lc = current_lc;
117 	lc->restart_psw.mask = PSW_BASE_BITS | PSW_DEFAULT_KEY;
118 	lc->restart_psw.addr = PSW_ADDR_AMODE | (unsigned long) smp_restart_cpu;
119 	if (!cpu_online(0))
120 		smp_switch_to_cpu(func, data, 0, stap(), __cpu_logical_map[0]);
121 	while (sigp(0, sigp_stop_and_store_status) == sigp_busy)
122 		cpu_relax();
123 	sp = lc->panic_stack;
124 	sp -= sizeof(struct pt_regs);
125 	regs = (struct pt_regs *) sp;
126 	memcpy(&regs->gprs, &current_lc->gpregs_save_area, sizeof(regs->gprs));
127 	regs->psw = lc->psw_save_area;
128 	sp -= STACK_FRAME_OVERHEAD;
129 	sf = (struct stack_frame *) sp;
130 	sf->back_chain = regs->gprs[15];
131 	smp_switch_to_cpu(func, data, sp, stap(), __cpu_logical_map[0]);
132 }
133 
134 void smp_send_stop(void)
135 {
136 	int cpu, rc;
137 
138 	/* Disable all interrupts/machine checks */
139 	__load_psw_mask(psw_kernel_bits & ~PSW_MASK_MCHECK);
140 	trace_hardirqs_off();
141 
142 	/* stop all processors */
143 	for_each_online_cpu(cpu) {
144 		if (cpu == smp_processor_id())
145 			continue;
146 		do {
147 			rc = sigp(cpu, sigp_stop);
148 		} while (rc == sigp_busy);
149 
150 		while (!cpu_stopped(cpu))
151 			cpu_relax();
152 	}
153 }
154 
155 /*
156  * This is the main routine where commands issued by other
157  * cpus are handled.
158  */
159 
160 static void do_ext_call_interrupt(unsigned int ext_int_code,
161 				  unsigned int param32, unsigned long param64)
162 {
163 	unsigned long bits;
164 
165 	kstat_cpu(smp_processor_id()).irqs[EXTINT_IPI]++;
166 	/*
167 	 * handle bit signal external calls
168 	 *
169 	 * For the ec_schedule signal we have to do nothing. All the work
170 	 * is done automatically when we return from the interrupt.
171 	 */
172 	bits = xchg(&S390_lowcore.ext_call_fast, 0);
173 
174 	if (test_bit(ec_call_function, &bits))
175 		generic_smp_call_function_interrupt();
176 
177 	if (test_bit(ec_call_function_single, &bits))
178 		generic_smp_call_function_single_interrupt();
179 }
180 
181 /*
182  * Send an external call sigp to another cpu and return without waiting
183  * for its completion.
184  */
185 static void smp_ext_bitcall(int cpu, int sig)
186 {
187 	/*
188 	 * Set signaling bit in lowcore of target cpu and kick it
189 	 */
190 	set_bit(sig, (unsigned long *) &lowcore_ptr[cpu]->ext_call_fast);
191 	while (sigp(cpu, sigp_emergency_signal) == sigp_busy)
192 		udelay(10);
193 }
194 
195 void arch_send_call_function_ipi_mask(const struct cpumask *mask)
196 {
197 	int cpu;
198 
199 	for_each_cpu(cpu, mask)
200 		smp_ext_bitcall(cpu, ec_call_function);
201 }
202 
203 void arch_send_call_function_single_ipi(int cpu)
204 {
205 	smp_ext_bitcall(cpu, ec_call_function_single);
206 }
207 
208 #ifndef CONFIG_64BIT
209 /*
210  * this function sends a 'purge tlb' signal to another CPU.
211  */
212 static void smp_ptlb_callback(void *info)
213 {
214 	__tlb_flush_local();
215 }
216 
217 void smp_ptlb_all(void)
218 {
219 	on_each_cpu(smp_ptlb_callback, NULL, 1);
220 }
221 EXPORT_SYMBOL(smp_ptlb_all);
222 #endif /* ! CONFIG_64BIT */
223 
224 /*
225  * this function sends a 'reschedule' IPI to another CPU.
226  * it goes straight through and wastes no time serializing
227  * anything. Worst case is that we lose a reschedule ...
228  */
229 void smp_send_reschedule(int cpu)
230 {
231 	smp_ext_bitcall(cpu, ec_schedule);
232 }
233 
234 /*
235  * parameter area for the set/clear control bit callbacks
236  */
237 struct ec_creg_mask_parms {
238 	unsigned long orvals[16];
239 	unsigned long andvals[16];
240 };
241 
242 /*
243  * callback for setting/clearing control bits
244  */
245 static void smp_ctl_bit_callback(void *info)
246 {
247 	struct ec_creg_mask_parms *pp = info;
248 	unsigned long cregs[16];
249 	int i;
250 
251 	__ctl_store(cregs, 0, 15);
252 	for (i = 0; i <= 15; i++)
253 		cregs[i] = (cregs[i] & pp->andvals[i]) | pp->orvals[i];
254 	__ctl_load(cregs, 0, 15);
255 }
256 
257 /*
258  * Set a bit in a control register of all cpus
259  */
260 void smp_ctl_set_bit(int cr, int bit)
261 {
262 	struct ec_creg_mask_parms parms;
263 
264 	memset(&parms.orvals, 0, sizeof(parms.orvals));
265 	memset(&parms.andvals, 0xff, sizeof(parms.andvals));
266 	parms.orvals[cr] = 1 << bit;
267 	on_each_cpu(smp_ctl_bit_callback, &parms, 1);
268 }
269 EXPORT_SYMBOL(smp_ctl_set_bit);
270 
271 /*
272  * Clear a bit in a control register of all cpus
273  */
274 void smp_ctl_clear_bit(int cr, int bit)
275 {
276 	struct ec_creg_mask_parms parms;
277 
278 	memset(&parms.orvals, 0, sizeof(parms.orvals));
279 	memset(&parms.andvals, 0xff, sizeof(parms.andvals));
280 	parms.andvals[cr] = ~(1L << bit);
281 	on_each_cpu(smp_ctl_bit_callback, &parms, 1);
282 }
283 EXPORT_SYMBOL(smp_ctl_clear_bit);
284 
285 #ifdef CONFIG_ZFCPDUMP
286 
287 static void __init smp_get_save_area(unsigned int cpu, unsigned int phy_cpu)
288 {
289 	if (ipl_info.type != IPL_TYPE_FCP_DUMP)
290 		return;
291 	if (cpu >= NR_CPUS) {
292 		pr_warning("CPU %i exceeds the maximum %i and is excluded from "
293 			   "the dump\n", cpu, NR_CPUS - 1);
294 		return;
295 	}
296 	zfcpdump_save_areas[cpu] = kmalloc(sizeof(struct save_area), GFP_KERNEL);
297 	while (raw_sigp(phy_cpu, sigp_stop_and_store_status) == sigp_busy)
298 		cpu_relax();
299 	memcpy_real(zfcpdump_save_areas[cpu],
300 		    (void *)(unsigned long) store_prefix() + SAVE_AREA_BASE,
301 		    sizeof(struct save_area));
302 }
303 
304 struct save_area *zfcpdump_save_areas[NR_CPUS + 1];
305 EXPORT_SYMBOL_GPL(zfcpdump_save_areas);
306 
307 #else
308 
309 static inline void smp_get_save_area(unsigned int cpu, unsigned int phy_cpu) { }
310 
311 #endif /* CONFIG_ZFCPDUMP */
312 
313 static int cpu_known(int cpu_id)
314 {
315 	int cpu;
316 
317 	for_each_present_cpu(cpu) {
318 		if (__cpu_logical_map[cpu] == cpu_id)
319 			return 1;
320 	}
321 	return 0;
322 }
323 
324 static int smp_rescan_cpus_sigp(cpumask_t avail)
325 {
326 	int cpu_id, logical_cpu;
327 
328 	logical_cpu = cpumask_first(&avail);
329 	if (logical_cpu >= nr_cpu_ids)
330 		return 0;
331 	for (cpu_id = 0; cpu_id <= MAX_CPU_ADDRESS; cpu_id++) {
332 		if (cpu_known(cpu_id))
333 			continue;
334 		__cpu_logical_map[logical_cpu] = cpu_id;
335 		smp_cpu_polarization[logical_cpu] = POLARIZATION_UNKNWN;
336 		if (!cpu_stopped(logical_cpu))
337 			continue;
338 		cpu_set(logical_cpu, cpu_present_map);
339 		smp_cpu_state[logical_cpu] = CPU_STATE_CONFIGURED;
340 		logical_cpu = cpumask_next(logical_cpu, &avail);
341 		if (logical_cpu >= nr_cpu_ids)
342 			break;
343 	}
344 	return 0;
345 }
346 
347 static int smp_rescan_cpus_sclp(cpumask_t avail)
348 {
349 	struct sclp_cpu_info *info;
350 	int cpu_id, logical_cpu, cpu;
351 	int rc;
352 
353 	logical_cpu = cpumask_first(&avail);
354 	if (logical_cpu >= nr_cpu_ids)
355 		return 0;
356 	info = kmalloc(sizeof(*info), GFP_KERNEL);
357 	if (!info)
358 		return -ENOMEM;
359 	rc = sclp_get_cpu_info(info);
360 	if (rc)
361 		goto out;
362 	for (cpu = 0; cpu < info->combined; cpu++) {
363 		if (info->has_cpu_type && info->cpu[cpu].type != smp_cpu_type)
364 			continue;
365 		cpu_id = info->cpu[cpu].address;
366 		if (cpu_known(cpu_id))
367 			continue;
368 		__cpu_logical_map[logical_cpu] = cpu_id;
369 		smp_cpu_polarization[logical_cpu] = POLARIZATION_UNKNWN;
370 		cpu_set(logical_cpu, cpu_present_map);
371 		if (cpu >= info->configured)
372 			smp_cpu_state[logical_cpu] = CPU_STATE_STANDBY;
373 		else
374 			smp_cpu_state[logical_cpu] = CPU_STATE_CONFIGURED;
375 		logical_cpu = cpumask_next(logical_cpu, &avail);
376 		if (logical_cpu >= nr_cpu_ids)
377 			break;
378 	}
379 out:
380 	kfree(info);
381 	return rc;
382 }
383 
384 static int __smp_rescan_cpus(void)
385 {
386 	cpumask_t avail;
387 
388 	cpus_xor(avail, cpu_possible_map, cpu_present_map);
389 	if (smp_use_sigp_detection)
390 		return smp_rescan_cpus_sigp(avail);
391 	else
392 		return smp_rescan_cpus_sclp(avail);
393 }
394 
395 static void __init smp_detect_cpus(void)
396 {
397 	unsigned int cpu, c_cpus, s_cpus;
398 	struct sclp_cpu_info *info;
399 	u16 boot_cpu_addr, cpu_addr;
400 
401 	c_cpus = 1;
402 	s_cpus = 0;
403 	boot_cpu_addr = __cpu_logical_map[0];
404 	info = kmalloc(sizeof(*info), GFP_KERNEL);
405 	if (!info)
406 		panic("smp_detect_cpus failed to allocate memory\n");
407 	/* Use sigp detection algorithm if sclp doesn't work. */
408 	if (sclp_get_cpu_info(info)) {
409 		smp_use_sigp_detection = 1;
410 		for (cpu = 0; cpu <= MAX_CPU_ADDRESS; cpu++) {
411 			if (cpu == boot_cpu_addr)
412 				continue;
413 			if (!raw_cpu_stopped(cpu))
414 				continue;
415 			smp_get_save_area(c_cpus, cpu);
416 			c_cpus++;
417 		}
418 		goto out;
419 	}
420 
421 	if (info->has_cpu_type) {
422 		for (cpu = 0; cpu < info->combined; cpu++) {
423 			if (info->cpu[cpu].address == boot_cpu_addr) {
424 				smp_cpu_type = info->cpu[cpu].type;
425 				break;
426 			}
427 		}
428 	}
429 
430 	for (cpu = 0; cpu < info->combined; cpu++) {
431 		if (info->has_cpu_type && info->cpu[cpu].type != smp_cpu_type)
432 			continue;
433 		cpu_addr = info->cpu[cpu].address;
434 		if (cpu_addr == boot_cpu_addr)
435 			continue;
436 		if (!raw_cpu_stopped(cpu_addr)) {
437 			s_cpus++;
438 			continue;
439 		}
440 		smp_get_save_area(c_cpus, cpu_addr);
441 		c_cpus++;
442 	}
443 out:
444 	kfree(info);
445 	pr_info("%d configured CPUs, %d standby CPUs\n", c_cpus, s_cpus);
446 	get_online_cpus();
447 	__smp_rescan_cpus();
448 	put_online_cpus();
449 }
450 
451 /*
452  *	Activate a secondary processor.
453  */
454 int __cpuinit start_secondary(void *cpuvoid)
455 {
456 	/* Setup the cpu */
457 	cpu_init();
458 	preempt_disable();
459 	/* Enable TOD clock interrupts on the secondary cpu. */
460 	init_cpu_timer();
461 	/* Enable cpu timer interrupts on the secondary cpu. */
462 	init_cpu_vtimer();
463 	/* Enable pfault pseudo page faults on this cpu. */
464 	pfault_init();
465 
466 	/* call cpu notifiers */
467 	notify_cpu_starting(smp_processor_id());
468 	/* Mark this cpu as online */
469 	ipi_call_lock();
470 	cpu_set(smp_processor_id(), cpu_online_map);
471 	ipi_call_unlock();
472 	/* Switch on interrupts */
473 	local_irq_enable();
474 	/* cpu_idle will call schedule for us */
475 	cpu_idle();
476 	return 0;
477 }
478 
479 struct create_idle {
480 	struct work_struct work;
481 	struct task_struct *idle;
482 	struct completion done;
483 	int cpu;
484 };
485 
486 static void __cpuinit smp_fork_idle(struct work_struct *work)
487 {
488 	struct create_idle *c_idle;
489 
490 	c_idle = container_of(work, struct create_idle, work);
491 	c_idle->idle = fork_idle(c_idle->cpu);
492 	complete(&c_idle->done);
493 }
494 
495 static int __cpuinit smp_alloc_lowcore(int cpu)
496 {
497 	unsigned long async_stack, panic_stack;
498 	struct _lowcore *lowcore;
499 
500 	lowcore = (void *) __get_free_pages(GFP_KERNEL | GFP_DMA, LC_ORDER);
501 	if (!lowcore)
502 		return -ENOMEM;
503 	async_stack = __get_free_pages(GFP_KERNEL, ASYNC_ORDER);
504 	panic_stack = __get_free_page(GFP_KERNEL);
505 	if (!panic_stack || !async_stack)
506 		goto out;
507 	memcpy(lowcore, &S390_lowcore, 512);
508 	memset((char *)lowcore + 512, 0, sizeof(*lowcore) - 512);
509 	lowcore->async_stack = async_stack + ASYNC_SIZE;
510 	lowcore->panic_stack = panic_stack + PAGE_SIZE;
511 
512 #ifndef CONFIG_64BIT
513 	if (MACHINE_HAS_IEEE) {
514 		unsigned long save_area;
515 
516 		save_area = get_zeroed_page(GFP_KERNEL);
517 		if (!save_area)
518 			goto out;
519 		lowcore->extended_save_area_addr = (u32) save_area;
520 	}
521 #else
522 	if (vdso_alloc_per_cpu(cpu, lowcore))
523 		goto out;
524 #endif
525 	lowcore_ptr[cpu] = lowcore;
526 	return 0;
527 
528 out:
529 	free_page(panic_stack);
530 	free_pages(async_stack, ASYNC_ORDER);
531 	free_pages((unsigned long) lowcore, LC_ORDER);
532 	return -ENOMEM;
533 }
534 
535 static void smp_free_lowcore(int cpu)
536 {
537 	struct _lowcore *lowcore;
538 
539 	lowcore = lowcore_ptr[cpu];
540 #ifndef CONFIG_64BIT
541 	if (MACHINE_HAS_IEEE)
542 		free_page((unsigned long) lowcore->extended_save_area_addr);
543 #else
544 	vdso_free_per_cpu(cpu, lowcore);
545 #endif
546 	free_page(lowcore->panic_stack - PAGE_SIZE);
547 	free_pages(lowcore->async_stack - ASYNC_SIZE, ASYNC_ORDER);
548 	free_pages((unsigned long) lowcore, LC_ORDER);
549 	lowcore_ptr[cpu] = NULL;
550 }
551 
552 /* Upping and downing of CPUs */
553 int __cpuinit __cpu_up(unsigned int cpu)
554 {
555 	struct _lowcore *cpu_lowcore;
556 	struct create_idle c_idle;
557 	struct task_struct *idle;
558 	struct stack_frame *sf;
559 	u32 lowcore;
560 	int ccode;
561 
562 	if (smp_cpu_state[cpu] != CPU_STATE_CONFIGURED)
563 		return -EIO;
564 	idle = current_set[cpu];
565 	if (!idle) {
566 		c_idle.done = COMPLETION_INITIALIZER_ONSTACK(c_idle.done);
567 		INIT_WORK_ONSTACK(&c_idle.work, smp_fork_idle);
568 		c_idle.cpu = cpu;
569 		schedule_work(&c_idle.work);
570 		wait_for_completion(&c_idle.done);
571 		if (IS_ERR(c_idle.idle))
572 			return PTR_ERR(c_idle.idle);
573 		idle = c_idle.idle;
574 		current_set[cpu] = c_idle.idle;
575 	}
576 	init_idle(idle, cpu);
577 	if (smp_alloc_lowcore(cpu))
578 		return -ENOMEM;
579 	do {
580 		ccode = sigp(cpu, sigp_initial_cpu_reset);
581 		if (ccode == sigp_busy)
582 			udelay(10);
583 		if (ccode == sigp_not_operational)
584 			goto err_out;
585 	} while (ccode == sigp_busy);
586 
587 	lowcore = (u32)(unsigned long)lowcore_ptr[cpu];
588 	while (sigp_p(lowcore, cpu, sigp_set_prefix) == sigp_busy)
589 		udelay(10);
590 
591 	cpu_lowcore = lowcore_ptr[cpu];
592 	cpu_lowcore->kernel_stack = (unsigned long)
593 		task_stack_page(idle) + THREAD_SIZE;
594 	cpu_lowcore->thread_info = (unsigned long) task_thread_info(idle);
595 	sf = (struct stack_frame *) (cpu_lowcore->kernel_stack
596 				     - sizeof(struct pt_regs)
597 				     - sizeof(struct stack_frame));
598 	memset(sf, 0, sizeof(struct stack_frame));
599 	sf->gprs[9] = (unsigned long) sf;
600 	cpu_lowcore->save_area[15] = (unsigned long) sf;
601 	__ctl_store(cpu_lowcore->cregs_save_area, 0, 15);
602 	atomic_inc(&init_mm.context.attach_count);
603 	asm volatile(
604 		"	stam	0,15,0(%0)"
605 		: : "a" (&cpu_lowcore->access_regs_save_area) : "memory");
606 	cpu_lowcore->percpu_offset = __per_cpu_offset[cpu];
607 	cpu_lowcore->current_task = (unsigned long) idle;
608 	cpu_lowcore->cpu_nr = cpu;
609 	cpu_lowcore->kernel_asce = S390_lowcore.kernel_asce;
610 	cpu_lowcore->machine_flags = S390_lowcore.machine_flags;
611 	cpu_lowcore->ftrace_func = S390_lowcore.ftrace_func;
612 	memcpy(cpu_lowcore->stfle_fac_list, S390_lowcore.stfle_fac_list,
613 	       MAX_FACILITY_BIT/8);
614 	eieio();
615 
616 	while (sigp(cpu, sigp_restart) == sigp_busy)
617 		udelay(10);
618 
619 	while (!cpu_online(cpu))
620 		cpu_relax();
621 	return 0;
622 
623 err_out:
624 	smp_free_lowcore(cpu);
625 	return -EIO;
626 }
627 
628 static int __init setup_possible_cpus(char *s)
629 {
630 	int pcpus, cpu;
631 
632 	pcpus = simple_strtoul(s, NULL, 0);
633 	init_cpu_possible(cpumask_of(0));
634 	for (cpu = 1; cpu < pcpus && cpu < nr_cpu_ids; cpu++)
635 		set_cpu_possible(cpu, true);
636 	return 0;
637 }
638 early_param("possible_cpus", setup_possible_cpus);
639 
640 #ifdef CONFIG_HOTPLUG_CPU
641 
642 int __cpu_disable(void)
643 {
644 	struct ec_creg_mask_parms cr_parms;
645 	int cpu = smp_processor_id();
646 
647 	cpu_clear(cpu, cpu_online_map);
648 
649 	/* Disable pfault pseudo page faults on this cpu. */
650 	pfault_fini();
651 
652 	memset(&cr_parms.orvals, 0, sizeof(cr_parms.orvals));
653 	memset(&cr_parms.andvals, 0xff, sizeof(cr_parms.andvals));
654 
655 	/* disable all external interrupts */
656 	cr_parms.orvals[0] = 0;
657 	cr_parms.andvals[0] = ~(1 << 15 | 1 << 14 | 1 << 13 | 1 << 12 |
658 				1 << 11 | 1 << 10 | 1 <<  6 | 1 <<  4);
659 	/* disable all I/O interrupts */
660 	cr_parms.orvals[6] = 0;
661 	cr_parms.andvals[6] = ~(1 << 31 | 1 << 30 | 1 << 29 | 1 << 28 |
662 				1 << 27 | 1 << 26 | 1 << 25 | 1 << 24);
663 	/* disable most machine checks */
664 	cr_parms.orvals[14] = 0;
665 	cr_parms.andvals[14] = ~(1 << 28 | 1 << 27 | 1 << 26 |
666 				 1 << 25 | 1 << 24);
667 
668 	smp_ctl_bit_callback(&cr_parms);
669 
670 	return 0;
671 }
672 
673 void __cpu_die(unsigned int cpu)
674 {
675 	/* Wait until target cpu is down */
676 	while (!cpu_stopped(cpu))
677 		cpu_relax();
678 	while (sigp_p(0, cpu, sigp_set_prefix) == sigp_busy)
679 		udelay(10);
680 	smp_free_lowcore(cpu);
681 	atomic_dec(&init_mm.context.attach_count);
682 }
683 
684 void cpu_die(void)
685 {
686 	idle_task_exit();
687 	while (sigp(smp_processor_id(), sigp_stop) == sigp_busy)
688 		cpu_relax();
689 	for (;;);
690 }
691 
692 #endif /* CONFIG_HOTPLUG_CPU */
693 
694 void __init smp_prepare_cpus(unsigned int max_cpus)
695 {
696 #ifndef CONFIG_64BIT
697 	unsigned long save_area = 0;
698 #endif
699 	unsigned long async_stack, panic_stack;
700 	struct _lowcore *lowcore;
701 
702 	smp_detect_cpus();
703 
704 	/* request the 0x1201 emergency signal external interrupt */
705 	if (register_external_interrupt(0x1201, do_ext_call_interrupt) != 0)
706 		panic("Couldn't request external interrupt 0x1201");
707 
708 	/* Reallocate current lowcore, but keep its contents. */
709 	lowcore = (void *) __get_free_pages(GFP_KERNEL | GFP_DMA, LC_ORDER);
710 	panic_stack = __get_free_page(GFP_KERNEL);
711 	async_stack = __get_free_pages(GFP_KERNEL, ASYNC_ORDER);
712 	BUG_ON(!lowcore || !panic_stack || !async_stack);
713 #ifndef CONFIG_64BIT
714 	if (MACHINE_HAS_IEEE)
715 		save_area = get_zeroed_page(GFP_KERNEL);
716 #endif
717 	local_irq_disable();
718 	local_mcck_disable();
719 	lowcore_ptr[smp_processor_id()] = lowcore;
720 	*lowcore = S390_lowcore;
721 	lowcore->panic_stack = panic_stack + PAGE_SIZE;
722 	lowcore->async_stack = async_stack + ASYNC_SIZE;
723 #ifndef CONFIG_64BIT
724 	if (MACHINE_HAS_IEEE)
725 		lowcore->extended_save_area_addr = (u32) save_area;
726 #endif
727 	set_prefix((u32)(unsigned long) lowcore);
728 	local_mcck_enable();
729 	local_irq_enable();
730 #ifdef CONFIG_64BIT
731 	if (vdso_alloc_per_cpu(smp_processor_id(), &S390_lowcore))
732 		BUG();
733 #endif
734 }
735 
736 void __init smp_prepare_boot_cpu(void)
737 {
738 	BUG_ON(smp_processor_id() != 0);
739 
740 	current_thread_info()->cpu = 0;
741 	cpu_set(0, cpu_present_map);
742 	cpu_set(0, cpu_online_map);
743 	S390_lowcore.percpu_offset = __per_cpu_offset[0];
744 	current_set[0] = current;
745 	smp_cpu_state[0] = CPU_STATE_CONFIGURED;
746 	smp_cpu_polarization[0] = POLARIZATION_UNKNWN;
747 }
748 
749 void __init smp_cpus_done(unsigned int max_cpus)
750 {
751 }
752 
753 void __init smp_setup_processor_id(void)
754 {
755 	S390_lowcore.cpu_nr = 0;
756 	__cpu_logical_map[0] = stap();
757 }
758 
759 /*
760  * the frequency of the profiling timer can be changed
761  * by writing a multiplier value into /proc/profile.
762  *
763  * usually you want to run this on all CPUs ;)
764  */
765 int setup_profiling_timer(unsigned int multiplier)
766 {
767 	return 0;
768 }
769 
770 #ifdef CONFIG_HOTPLUG_CPU
771 static ssize_t cpu_configure_show(struct sys_device *dev,
772 				struct sysdev_attribute *attr, char *buf)
773 {
774 	ssize_t count;
775 
776 	mutex_lock(&smp_cpu_state_mutex);
777 	count = sprintf(buf, "%d\n", smp_cpu_state[dev->id]);
778 	mutex_unlock(&smp_cpu_state_mutex);
779 	return count;
780 }
781 
782 static ssize_t cpu_configure_store(struct sys_device *dev,
783 				  struct sysdev_attribute *attr,
784 				  const char *buf, size_t count)
785 {
786 	int cpu = dev->id;
787 	int val, rc;
788 	char delim;
789 
790 	if (sscanf(buf, "%d %c", &val, &delim) != 1)
791 		return -EINVAL;
792 	if (val != 0 && val != 1)
793 		return -EINVAL;
794 
795 	get_online_cpus();
796 	mutex_lock(&smp_cpu_state_mutex);
797 	rc = -EBUSY;
798 	/* disallow configuration changes of online cpus and cpu 0 */
799 	if (cpu_online(cpu) || cpu == 0)
800 		goto out;
801 	rc = 0;
802 	switch (val) {
803 	case 0:
804 		if (smp_cpu_state[cpu] == CPU_STATE_CONFIGURED) {
805 			rc = sclp_cpu_deconfigure(__cpu_logical_map[cpu]);
806 			if (!rc) {
807 				smp_cpu_state[cpu] = CPU_STATE_STANDBY;
808 				smp_cpu_polarization[cpu] = POLARIZATION_UNKNWN;
809 			}
810 		}
811 		break;
812 	case 1:
813 		if (smp_cpu_state[cpu] == CPU_STATE_STANDBY) {
814 			rc = sclp_cpu_configure(__cpu_logical_map[cpu]);
815 			if (!rc) {
816 				smp_cpu_state[cpu] = CPU_STATE_CONFIGURED;
817 				smp_cpu_polarization[cpu] = POLARIZATION_UNKNWN;
818 			}
819 		}
820 		break;
821 	default:
822 		break;
823 	}
824 out:
825 	mutex_unlock(&smp_cpu_state_mutex);
826 	put_online_cpus();
827 	return rc ? rc : count;
828 }
829 static SYSDEV_ATTR(configure, 0644, cpu_configure_show, cpu_configure_store);
830 #endif /* CONFIG_HOTPLUG_CPU */
831 
832 static ssize_t cpu_polarization_show(struct sys_device *dev,
833 				     struct sysdev_attribute *attr, char *buf)
834 {
835 	int cpu = dev->id;
836 	ssize_t count;
837 
838 	mutex_lock(&smp_cpu_state_mutex);
839 	switch (smp_cpu_polarization[cpu]) {
840 	case POLARIZATION_HRZ:
841 		count = sprintf(buf, "horizontal\n");
842 		break;
843 	case POLARIZATION_VL:
844 		count = sprintf(buf, "vertical:low\n");
845 		break;
846 	case POLARIZATION_VM:
847 		count = sprintf(buf, "vertical:medium\n");
848 		break;
849 	case POLARIZATION_VH:
850 		count = sprintf(buf, "vertical:high\n");
851 		break;
852 	default:
853 		count = sprintf(buf, "unknown\n");
854 		break;
855 	}
856 	mutex_unlock(&smp_cpu_state_mutex);
857 	return count;
858 }
859 static SYSDEV_ATTR(polarization, 0444, cpu_polarization_show, NULL);
860 
861 static ssize_t show_cpu_address(struct sys_device *dev,
862 				struct sysdev_attribute *attr, char *buf)
863 {
864 	return sprintf(buf, "%d\n", __cpu_logical_map[dev->id]);
865 }
866 static SYSDEV_ATTR(address, 0444, show_cpu_address, NULL);
867 
868 
869 static struct attribute *cpu_common_attrs[] = {
870 #ifdef CONFIG_HOTPLUG_CPU
871 	&attr_configure.attr,
872 #endif
873 	&attr_address.attr,
874 	&attr_polarization.attr,
875 	NULL,
876 };
877 
878 static struct attribute_group cpu_common_attr_group = {
879 	.attrs = cpu_common_attrs,
880 };
881 
882 static ssize_t show_capability(struct sys_device *dev,
883 				struct sysdev_attribute *attr, char *buf)
884 {
885 	unsigned int capability;
886 	int rc;
887 
888 	rc = get_cpu_capability(&capability);
889 	if (rc)
890 		return rc;
891 	return sprintf(buf, "%u\n", capability);
892 }
893 static SYSDEV_ATTR(capability, 0444, show_capability, NULL);
894 
895 static ssize_t show_idle_count(struct sys_device *dev,
896 				struct sysdev_attribute *attr, char *buf)
897 {
898 	struct s390_idle_data *idle;
899 	unsigned long long idle_count;
900 	unsigned int sequence;
901 
902 	idle = &per_cpu(s390_idle, dev->id);
903 repeat:
904 	sequence = idle->sequence;
905 	smp_rmb();
906 	if (sequence & 1)
907 		goto repeat;
908 	idle_count = idle->idle_count;
909 	if (idle->idle_enter)
910 		idle_count++;
911 	smp_rmb();
912 	if (idle->sequence != sequence)
913 		goto repeat;
914 	return sprintf(buf, "%llu\n", idle_count);
915 }
916 static SYSDEV_ATTR(idle_count, 0444, show_idle_count, NULL);
917 
918 static ssize_t show_idle_time(struct sys_device *dev,
919 				struct sysdev_attribute *attr, char *buf)
920 {
921 	struct s390_idle_data *idle;
922 	unsigned long long now, idle_time, idle_enter;
923 	unsigned int sequence;
924 
925 	idle = &per_cpu(s390_idle, dev->id);
926 	now = get_clock();
927 repeat:
928 	sequence = idle->sequence;
929 	smp_rmb();
930 	if (sequence & 1)
931 		goto repeat;
932 	idle_time = idle->idle_time;
933 	idle_enter = idle->idle_enter;
934 	if (idle_enter != 0ULL && idle_enter < now)
935 		idle_time += now - idle_enter;
936 	smp_rmb();
937 	if (idle->sequence != sequence)
938 		goto repeat;
939 	return sprintf(buf, "%llu\n", idle_time >> 12);
940 }
941 static SYSDEV_ATTR(idle_time_us, 0444, show_idle_time, NULL);
942 
943 static struct attribute *cpu_online_attrs[] = {
944 	&attr_capability.attr,
945 	&attr_idle_count.attr,
946 	&attr_idle_time_us.attr,
947 	NULL,
948 };
949 
950 static struct attribute_group cpu_online_attr_group = {
951 	.attrs = cpu_online_attrs,
952 };
953 
954 static int __cpuinit smp_cpu_notify(struct notifier_block *self,
955 				    unsigned long action, void *hcpu)
956 {
957 	unsigned int cpu = (unsigned int)(long)hcpu;
958 	struct cpu *c = &per_cpu(cpu_devices, cpu);
959 	struct sys_device *s = &c->sysdev;
960 	struct s390_idle_data *idle;
961 	int err = 0;
962 
963 	switch (action) {
964 	case CPU_ONLINE:
965 	case CPU_ONLINE_FROZEN:
966 		idle = &per_cpu(s390_idle, cpu);
967 		memset(idle, 0, sizeof(struct s390_idle_data));
968 		err = sysfs_create_group(&s->kobj, &cpu_online_attr_group);
969 		break;
970 	case CPU_DEAD:
971 	case CPU_DEAD_FROZEN:
972 		sysfs_remove_group(&s->kobj, &cpu_online_attr_group);
973 		break;
974 	}
975 	return notifier_from_errno(err);
976 }
977 
978 static struct notifier_block __cpuinitdata smp_cpu_nb = {
979 	.notifier_call = smp_cpu_notify,
980 };
981 
982 static int __devinit smp_add_present_cpu(int cpu)
983 {
984 	struct cpu *c = &per_cpu(cpu_devices, cpu);
985 	struct sys_device *s = &c->sysdev;
986 	int rc;
987 
988 	c->hotpluggable = 1;
989 	rc = register_cpu(c, cpu);
990 	if (rc)
991 		goto out;
992 	rc = sysfs_create_group(&s->kobj, &cpu_common_attr_group);
993 	if (rc)
994 		goto out_cpu;
995 	if (!cpu_online(cpu))
996 		goto out;
997 	rc = sysfs_create_group(&s->kobj, &cpu_online_attr_group);
998 	if (!rc)
999 		return 0;
1000 	sysfs_remove_group(&s->kobj, &cpu_common_attr_group);
1001 out_cpu:
1002 #ifdef CONFIG_HOTPLUG_CPU
1003 	unregister_cpu(c);
1004 #endif
1005 out:
1006 	return rc;
1007 }
1008 
1009 #ifdef CONFIG_HOTPLUG_CPU
1010 
1011 int __ref smp_rescan_cpus(void)
1012 {
1013 	cpumask_t newcpus;
1014 	int cpu;
1015 	int rc;
1016 
1017 	get_online_cpus();
1018 	mutex_lock(&smp_cpu_state_mutex);
1019 	newcpus = cpu_present_map;
1020 	rc = __smp_rescan_cpus();
1021 	if (rc)
1022 		goto out;
1023 	cpus_andnot(newcpus, cpu_present_map, newcpus);
1024 	for_each_cpu_mask(cpu, newcpus) {
1025 		rc = smp_add_present_cpu(cpu);
1026 		if (rc)
1027 			cpu_clear(cpu, cpu_present_map);
1028 	}
1029 	rc = 0;
1030 out:
1031 	mutex_unlock(&smp_cpu_state_mutex);
1032 	put_online_cpus();
1033 	if (!cpus_empty(newcpus))
1034 		topology_schedule_update();
1035 	return rc;
1036 }
1037 
1038 static ssize_t __ref rescan_store(struct sysdev_class *class,
1039 				  struct sysdev_class_attribute *attr,
1040 				  const char *buf,
1041 				  size_t count)
1042 {
1043 	int rc;
1044 
1045 	rc = smp_rescan_cpus();
1046 	return rc ? rc : count;
1047 }
1048 static SYSDEV_CLASS_ATTR(rescan, 0200, NULL, rescan_store);
1049 #endif /* CONFIG_HOTPLUG_CPU */
1050 
1051 static ssize_t dispatching_show(struct sysdev_class *class,
1052 				struct sysdev_class_attribute *attr,
1053 				char *buf)
1054 {
1055 	ssize_t count;
1056 
1057 	mutex_lock(&smp_cpu_state_mutex);
1058 	count = sprintf(buf, "%d\n", cpu_management);
1059 	mutex_unlock(&smp_cpu_state_mutex);
1060 	return count;
1061 }
1062 
1063 static ssize_t dispatching_store(struct sysdev_class *dev,
1064 				 struct sysdev_class_attribute *attr,
1065 				 const char *buf,
1066 				 size_t count)
1067 {
1068 	int val, rc;
1069 	char delim;
1070 
1071 	if (sscanf(buf, "%d %c", &val, &delim) != 1)
1072 		return -EINVAL;
1073 	if (val != 0 && val != 1)
1074 		return -EINVAL;
1075 	rc = 0;
1076 	get_online_cpus();
1077 	mutex_lock(&smp_cpu_state_mutex);
1078 	if (cpu_management == val)
1079 		goto out;
1080 	rc = topology_set_cpu_management(val);
1081 	if (!rc)
1082 		cpu_management = val;
1083 out:
1084 	mutex_unlock(&smp_cpu_state_mutex);
1085 	put_online_cpus();
1086 	return rc ? rc : count;
1087 }
1088 static SYSDEV_CLASS_ATTR(dispatching, 0644, dispatching_show,
1089 			 dispatching_store);
1090 
1091 static int __init topology_init(void)
1092 {
1093 	int cpu;
1094 	int rc;
1095 
1096 	register_cpu_notifier(&smp_cpu_nb);
1097 
1098 #ifdef CONFIG_HOTPLUG_CPU
1099 	rc = sysdev_class_create_file(&cpu_sysdev_class, &attr_rescan);
1100 	if (rc)
1101 		return rc;
1102 #endif
1103 	rc = sysdev_class_create_file(&cpu_sysdev_class, &attr_dispatching);
1104 	if (rc)
1105 		return rc;
1106 	for_each_present_cpu(cpu) {
1107 		rc = smp_add_present_cpu(cpu);
1108 		if (rc)
1109 			return rc;
1110 	}
1111 	return 0;
1112 }
1113 subsys_initcall(topology_init);
1114