xref: /linux/arch/s390/kernel/smp.c (revision f2ee442115c9b6219083c019939a9cc0c9abb2f8)
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 <linux/crash_dump.h>
42 #include <asm/asm-offsets.h>
43 #include <asm/ipl.h>
44 #include <asm/setup.h>
45 #include <asm/sigp.h>
46 #include <asm/pgalloc.h>
47 #include <asm/irq.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 /*
102  * Ensure that PSW restart is done on an online CPU
103  */
104 void smp_restart_with_online_cpu(void)
105 {
106 	int cpu;
107 
108 	for_each_online_cpu(cpu) {
109 		if (stap() == __cpu_logical_map[cpu]) {
110 			/* We are online: Enable DAT again and return */
111 			__load_psw_mask(psw_kernel_bits | PSW_MASK_DAT);
112 			return;
113 		}
114 	}
115 	/* We are not online: Do PSW restart on an online CPU */
116 	while (sigp(cpu, sigp_restart) == sigp_busy)
117 		cpu_relax();
118 	/* And stop ourself */
119 	while (raw_sigp(stap(), sigp_stop) == sigp_busy)
120 		cpu_relax();
121 	for (;;);
122 }
123 
124 void smp_switch_to_ipl_cpu(void (*func)(void *), void *data)
125 {
126 	struct _lowcore *lc, *current_lc;
127 	struct stack_frame *sf;
128 	struct pt_regs *regs;
129 	unsigned long sp;
130 
131 	if (smp_processor_id() == 0)
132 		func(data);
133 	__load_psw_mask(PSW_DEFAULT_KEY | PSW_MASK_BASE |
134 			PSW_MASK_EA | PSW_MASK_BA);
135 	/* Disable lowcore protection */
136 	__ctl_clear_bit(0, 28);
137 	current_lc = lowcore_ptr[smp_processor_id()];
138 	lc = lowcore_ptr[0];
139 	if (!lc)
140 		lc = current_lc;
141 	lc->restart_psw.mask =
142 		PSW_DEFAULT_KEY | PSW_MASK_BASE | PSW_MASK_EA | PSW_MASK_BA;
143 	lc->restart_psw.addr = PSW_ADDR_AMODE | (unsigned long) smp_restart_cpu;
144 	if (!cpu_online(0))
145 		smp_switch_to_cpu(func, data, 0, stap(), __cpu_logical_map[0]);
146 	while (sigp(0, sigp_stop_and_store_status) == sigp_busy)
147 		cpu_relax();
148 	sp = lc->panic_stack;
149 	sp -= sizeof(struct pt_regs);
150 	regs = (struct pt_regs *) sp;
151 	memcpy(&regs->gprs, &current_lc->gpregs_save_area, sizeof(regs->gprs));
152 	regs->psw = lc->psw_save_area;
153 	sp -= STACK_FRAME_OVERHEAD;
154 	sf = (struct stack_frame *) sp;
155 	sf->back_chain = regs->gprs[15];
156 	smp_switch_to_cpu(func, data, sp, stap(), __cpu_logical_map[0]);
157 }
158 
159 void smp_send_stop(void)
160 {
161 	int cpu, rc;
162 
163 	/* Disable all interrupts/machine checks */
164 	__load_psw_mask(psw_kernel_bits | PSW_MASK_DAT);
165 	trace_hardirqs_off();
166 
167 	/* stop all processors */
168 	for_each_online_cpu(cpu) {
169 		if (cpu == smp_processor_id())
170 			continue;
171 		do {
172 			rc = sigp(cpu, sigp_stop);
173 		} while (rc == sigp_busy);
174 
175 		while (!cpu_stopped(cpu))
176 			cpu_relax();
177 	}
178 }
179 
180 /*
181  * This is the main routine where commands issued by other
182  * cpus are handled.
183  */
184 
185 static void do_ext_call_interrupt(unsigned int ext_int_code,
186 				  unsigned int param32, unsigned long param64)
187 {
188 	unsigned long bits;
189 
190 	if (ext_int_code == 0x1202)
191 		kstat_cpu(smp_processor_id()).irqs[EXTINT_EXC]++;
192 	else
193 		kstat_cpu(smp_processor_id()).irqs[EXTINT_EMS]++;
194 	/*
195 	 * handle bit signal external calls
196 	 */
197 	bits = xchg(&S390_lowcore.ext_call_fast, 0);
198 
199 	if (test_bit(ec_schedule, &bits))
200 		scheduler_ipi();
201 
202 	if (test_bit(ec_call_function, &bits))
203 		generic_smp_call_function_interrupt();
204 
205 	if (test_bit(ec_call_function_single, &bits))
206 		generic_smp_call_function_single_interrupt();
207 }
208 
209 /*
210  * Send an external call sigp to another cpu and return without waiting
211  * for its completion.
212  */
213 static void smp_ext_bitcall(int cpu, int sig)
214 {
215 	int order;
216 
217 	/*
218 	 * Set signaling bit in lowcore of target cpu and kick it
219 	 */
220 	set_bit(sig, (unsigned long *) &lowcore_ptr[cpu]->ext_call_fast);
221 	while (1) {
222 		order = smp_vcpu_scheduled(cpu) ?
223 			sigp_external_call : sigp_emergency_signal;
224 		if (sigp(cpu, order) != sigp_busy)
225 			break;
226 		udelay(10);
227 	}
228 }
229 
230 void arch_send_call_function_ipi_mask(const struct cpumask *mask)
231 {
232 	int cpu;
233 
234 	for_each_cpu(cpu, mask)
235 		smp_ext_bitcall(cpu, ec_call_function);
236 }
237 
238 void arch_send_call_function_single_ipi(int cpu)
239 {
240 	smp_ext_bitcall(cpu, ec_call_function_single);
241 }
242 
243 #ifndef CONFIG_64BIT
244 /*
245  * this function sends a 'purge tlb' signal to another CPU.
246  */
247 static void smp_ptlb_callback(void *info)
248 {
249 	__tlb_flush_local();
250 }
251 
252 void smp_ptlb_all(void)
253 {
254 	on_each_cpu(smp_ptlb_callback, NULL, 1);
255 }
256 EXPORT_SYMBOL(smp_ptlb_all);
257 #endif /* ! CONFIG_64BIT */
258 
259 /*
260  * this function sends a 'reschedule' IPI to another CPU.
261  * it goes straight through and wastes no time serializing
262  * anything. Worst case is that we lose a reschedule ...
263  */
264 void smp_send_reschedule(int cpu)
265 {
266 	smp_ext_bitcall(cpu, ec_schedule);
267 }
268 
269 /*
270  * parameter area for the set/clear control bit callbacks
271  */
272 struct ec_creg_mask_parms {
273 	unsigned long orvals[16];
274 	unsigned long andvals[16];
275 };
276 
277 /*
278  * callback for setting/clearing control bits
279  */
280 static void smp_ctl_bit_callback(void *info)
281 {
282 	struct ec_creg_mask_parms *pp = info;
283 	unsigned long cregs[16];
284 	int i;
285 
286 	__ctl_store(cregs, 0, 15);
287 	for (i = 0; i <= 15; i++)
288 		cregs[i] = (cregs[i] & pp->andvals[i]) | pp->orvals[i];
289 	__ctl_load(cregs, 0, 15);
290 }
291 
292 /*
293  * Set a bit in a control register of all cpus
294  */
295 void smp_ctl_set_bit(int cr, int bit)
296 {
297 	struct ec_creg_mask_parms parms;
298 
299 	memset(&parms.orvals, 0, sizeof(parms.orvals));
300 	memset(&parms.andvals, 0xff, sizeof(parms.andvals));
301 	parms.orvals[cr] = 1UL << bit;
302 	on_each_cpu(smp_ctl_bit_callback, &parms, 1);
303 }
304 EXPORT_SYMBOL(smp_ctl_set_bit);
305 
306 /*
307  * Clear a bit in a control register of all cpus
308  */
309 void smp_ctl_clear_bit(int cr, int bit)
310 {
311 	struct ec_creg_mask_parms parms;
312 
313 	memset(&parms.orvals, 0, sizeof(parms.orvals));
314 	memset(&parms.andvals, 0xff, sizeof(parms.andvals));
315 	parms.andvals[cr] = ~(1UL << bit);
316 	on_each_cpu(smp_ctl_bit_callback, &parms, 1);
317 }
318 EXPORT_SYMBOL(smp_ctl_clear_bit);
319 
320 #if defined(CONFIG_ZFCPDUMP) || defined(CONFIG_CRASH_DUMP)
321 
322 static void __init smp_get_save_area(unsigned int cpu, unsigned int phy_cpu)
323 {
324 	if (ipl_info.type != IPL_TYPE_FCP_DUMP && !OLDMEM_BASE)
325 		return;
326 	if (is_kdump_kernel())
327 		return;
328 	if (cpu >= NR_CPUS) {
329 		pr_warning("CPU %i exceeds the maximum %i and is excluded from "
330 			   "the dump\n", cpu, NR_CPUS - 1);
331 		return;
332 	}
333 	zfcpdump_save_areas[cpu] = kmalloc(sizeof(struct save_area), GFP_KERNEL);
334 	while (raw_sigp(phy_cpu, sigp_stop_and_store_status) == sigp_busy)
335 		cpu_relax();
336 	memcpy_real(zfcpdump_save_areas[cpu],
337 		    (void *)(unsigned long) store_prefix() + SAVE_AREA_BASE,
338 		    sizeof(struct save_area));
339 }
340 
341 struct save_area *zfcpdump_save_areas[NR_CPUS + 1];
342 EXPORT_SYMBOL_GPL(zfcpdump_save_areas);
343 
344 #else
345 
346 static inline void smp_get_save_area(unsigned int cpu, unsigned int phy_cpu) { }
347 
348 #endif /* CONFIG_ZFCPDUMP */
349 
350 static int cpu_known(int cpu_id)
351 {
352 	int cpu;
353 
354 	for_each_present_cpu(cpu) {
355 		if (__cpu_logical_map[cpu] == cpu_id)
356 			return 1;
357 	}
358 	return 0;
359 }
360 
361 static int smp_rescan_cpus_sigp(cpumask_t avail)
362 {
363 	int cpu_id, logical_cpu;
364 
365 	logical_cpu = cpumask_first(&avail);
366 	if (logical_cpu >= nr_cpu_ids)
367 		return 0;
368 	for (cpu_id = 0; cpu_id <= MAX_CPU_ADDRESS; cpu_id++) {
369 		if (cpu_known(cpu_id))
370 			continue;
371 		__cpu_logical_map[logical_cpu] = cpu_id;
372 		smp_cpu_polarization[logical_cpu] = POLARIZATION_UNKNWN;
373 		if (!cpu_stopped(logical_cpu))
374 			continue;
375 		set_cpu_present(logical_cpu, true);
376 		smp_cpu_state[logical_cpu] = CPU_STATE_CONFIGURED;
377 		logical_cpu = cpumask_next(logical_cpu, &avail);
378 		if (logical_cpu >= nr_cpu_ids)
379 			break;
380 	}
381 	return 0;
382 }
383 
384 static int smp_rescan_cpus_sclp(cpumask_t avail)
385 {
386 	struct sclp_cpu_info *info;
387 	int cpu_id, logical_cpu, cpu;
388 	int rc;
389 
390 	logical_cpu = cpumask_first(&avail);
391 	if (logical_cpu >= nr_cpu_ids)
392 		return 0;
393 	info = kmalloc(sizeof(*info), GFP_KERNEL);
394 	if (!info)
395 		return -ENOMEM;
396 	rc = sclp_get_cpu_info(info);
397 	if (rc)
398 		goto out;
399 	for (cpu = 0; cpu < info->combined; cpu++) {
400 		if (info->has_cpu_type && info->cpu[cpu].type != smp_cpu_type)
401 			continue;
402 		cpu_id = info->cpu[cpu].address;
403 		if (cpu_known(cpu_id))
404 			continue;
405 		__cpu_logical_map[logical_cpu] = cpu_id;
406 		smp_cpu_polarization[logical_cpu] = POLARIZATION_UNKNWN;
407 		set_cpu_present(logical_cpu, true);
408 		if (cpu >= info->configured)
409 			smp_cpu_state[logical_cpu] = CPU_STATE_STANDBY;
410 		else
411 			smp_cpu_state[logical_cpu] = CPU_STATE_CONFIGURED;
412 		logical_cpu = cpumask_next(logical_cpu, &avail);
413 		if (logical_cpu >= nr_cpu_ids)
414 			break;
415 	}
416 out:
417 	kfree(info);
418 	return rc;
419 }
420 
421 static int __smp_rescan_cpus(void)
422 {
423 	cpumask_t avail;
424 
425 	cpumask_xor(&avail, cpu_possible_mask, cpu_present_mask);
426 	if (smp_use_sigp_detection)
427 		return smp_rescan_cpus_sigp(avail);
428 	else
429 		return smp_rescan_cpus_sclp(avail);
430 }
431 
432 static void __init smp_detect_cpus(void)
433 {
434 	unsigned int cpu, c_cpus, s_cpus;
435 	struct sclp_cpu_info *info;
436 	u16 boot_cpu_addr, cpu_addr;
437 
438 	c_cpus = 1;
439 	s_cpus = 0;
440 	boot_cpu_addr = __cpu_logical_map[0];
441 	info = kmalloc(sizeof(*info), GFP_KERNEL);
442 	if (!info)
443 		panic("smp_detect_cpus failed to allocate memory\n");
444 #ifdef CONFIG_CRASH_DUMP
445 	if (OLDMEM_BASE && !is_kdump_kernel()) {
446 		struct save_area *save_area;
447 
448 		save_area = kmalloc(sizeof(*save_area), GFP_KERNEL);
449 		if (!save_area)
450 			panic("could not allocate memory for save area\n");
451 		copy_oldmem_page(1, (void *) save_area, sizeof(*save_area),
452 				 0x200, 0);
453 		zfcpdump_save_areas[0] = save_area;
454 	}
455 #endif
456 	/* Use sigp detection algorithm if sclp doesn't work. */
457 	if (sclp_get_cpu_info(info)) {
458 		smp_use_sigp_detection = 1;
459 		for (cpu = 0; cpu <= MAX_CPU_ADDRESS; cpu++) {
460 			if (cpu == boot_cpu_addr)
461 				continue;
462 			if (!raw_cpu_stopped(cpu))
463 				continue;
464 			smp_get_save_area(c_cpus, cpu);
465 			c_cpus++;
466 		}
467 		goto out;
468 	}
469 
470 	if (info->has_cpu_type) {
471 		for (cpu = 0; cpu < info->combined; cpu++) {
472 			if (info->cpu[cpu].address == boot_cpu_addr) {
473 				smp_cpu_type = info->cpu[cpu].type;
474 				break;
475 			}
476 		}
477 	}
478 
479 	for (cpu = 0; cpu < info->combined; cpu++) {
480 		if (info->has_cpu_type && info->cpu[cpu].type != smp_cpu_type)
481 			continue;
482 		cpu_addr = info->cpu[cpu].address;
483 		if (cpu_addr == boot_cpu_addr)
484 			continue;
485 		if (!raw_cpu_stopped(cpu_addr)) {
486 			s_cpus++;
487 			continue;
488 		}
489 		smp_get_save_area(c_cpus, cpu_addr);
490 		c_cpus++;
491 	}
492 out:
493 	kfree(info);
494 	pr_info("%d configured CPUs, %d standby CPUs\n", c_cpus, s_cpus);
495 	get_online_cpus();
496 	__smp_rescan_cpus();
497 	put_online_cpus();
498 }
499 
500 /*
501  *	Activate a secondary processor.
502  */
503 int __cpuinit start_secondary(void *cpuvoid)
504 {
505 	cpu_init();
506 	preempt_disable();
507 	init_cpu_timer();
508 	init_cpu_vtimer();
509 	pfault_init();
510 
511 	notify_cpu_starting(smp_processor_id());
512 	ipi_call_lock();
513 	set_cpu_online(smp_processor_id(), true);
514 	ipi_call_unlock();
515 	__ctl_clear_bit(0, 28); /* Disable lowcore protection */
516 	S390_lowcore.restart_psw.mask =
517 		PSW_DEFAULT_KEY | PSW_MASK_BASE | PSW_MASK_EA | PSW_MASK_BA;
518 	S390_lowcore.restart_psw.addr =
519 		PSW_ADDR_AMODE | (unsigned long) psw_restart_int_handler;
520 	__ctl_set_bit(0, 28); /* Enable lowcore protection */
521 	/*
522 	 * Wait until the cpu which brought this one up marked it
523 	 * active before enabling interrupts.
524 	 */
525 	while (!cpumask_test_cpu(smp_processor_id(), cpu_active_mask))
526 		cpu_relax();
527 	local_irq_enable();
528 	/* cpu_idle will call schedule for us */
529 	cpu_idle();
530 	return 0;
531 }
532 
533 struct create_idle {
534 	struct work_struct work;
535 	struct task_struct *idle;
536 	struct completion done;
537 	int cpu;
538 };
539 
540 static void __cpuinit smp_fork_idle(struct work_struct *work)
541 {
542 	struct create_idle *c_idle;
543 
544 	c_idle = container_of(work, struct create_idle, work);
545 	c_idle->idle = fork_idle(c_idle->cpu);
546 	complete(&c_idle->done);
547 }
548 
549 static int __cpuinit smp_alloc_lowcore(int cpu)
550 {
551 	unsigned long async_stack, panic_stack;
552 	struct _lowcore *lowcore;
553 
554 	lowcore = (void *) __get_free_pages(GFP_KERNEL | GFP_DMA, LC_ORDER);
555 	if (!lowcore)
556 		return -ENOMEM;
557 	async_stack = __get_free_pages(GFP_KERNEL, ASYNC_ORDER);
558 	panic_stack = __get_free_page(GFP_KERNEL);
559 	if (!panic_stack || !async_stack)
560 		goto out;
561 	memcpy(lowcore, &S390_lowcore, 512);
562 	memset((char *)lowcore + 512, 0, sizeof(*lowcore) - 512);
563 	lowcore->async_stack = async_stack + ASYNC_SIZE;
564 	lowcore->panic_stack = panic_stack + PAGE_SIZE;
565 	lowcore->restart_psw.mask =
566 		PSW_DEFAULT_KEY | PSW_MASK_BASE | PSW_MASK_EA | PSW_MASK_BA;
567 	lowcore->restart_psw.addr =
568 		PSW_ADDR_AMODE | (unsigned long) restart_int_handler;
569 	if (user_mode != HOME_SPACE_MODE)
570 		lowcore->restart_psw.mask |= PSW_ASC_HOME;
571 #ifndef CONFIG_64BIT
572 	if (MACHINE_HAS_IEEE) {
573 		unsigned long save_area;
574 
575 		save_area = get_zeroed_page(GFP_KERNEL);
576 		if (!save_area)
577 			goto out;
578 		lowcore->extended_save_area_addr = (u32) save_area;
579 	}
580 #else
581 	if (vdso_alloc_per_cpu(cpu, lowcore))
582 		goto out;
583 #endif
584 	lowcore_ptr[cpu] = lowcore;
585 	return 0;
586 
587 out:
588 	free_page(panic_stack);
589 	free_pages(async_stack, ASYNC_ORDER);
590 	free_pages((unsigned long) lowcore, LC_ORDER);
591 	return -ENOMEM;
592 }
593 
594 static void smp_free_lowcore(int cpu)
595 {
596 	struct _lowcore *lowcore;
597 
598 	lowcore = lowcore_ptr[cpu];
599 #ifndef CONFIG_64BIT
600 	if (MACHINE_HAS_IEEE)
601 		free_page((unsigned long) lowcore->extended_save_area_addr);
602 #else
603 	vdso_free_per_cpu(cpu, lowcore);
604 #endif
605 	free_page(lowcore->panic_stack - PAGE_SIZE);
606 	free_pages(lowcore->async_stack - ASYNC_SIZE, ASYNC_ORDER);
607 	free_pages((unsigned long) lowcore, LC_ORDER);
608 	lowcore_ptr[cpu] = NULL;
609 }
610 
611 /* Upping and downing of CPUs */
612 int __cpuinit __cpu_up(unsigned int cpu)
613 {
614 	struct _lowcore *cpu_lowcore;
615 	struct create_idle c_idle;
616 	struct task_struct *idle;
617 	struct stack_frame *sf;
618 	u32 lowcore;
619 	int ccode;
620 
621 	if (smp_cpu_state[cpu] != CPU_STATE_CONFIGURED)
622 		return -EIO;
623 	idle = current_set[cpu];
624 	if (!idle) {
625 		c_idle.done = COMPLETION_INITIALIZER_ONSTACK(c_idle.done);
626 		INIT_WORK_ONSTACK(&c_idle.work, smp_fork_idle);
627 		c_idle.cpu = cpu;
628 		schedule_work(&c_idle.work);
629 		wait_for_completion(&c_idle.done);
630 		if (IS_ERR(c_idle.idle))
631 			return PTR_ERR(c_idle.idle);
632 		idle = c_idle.idle;
633 		current_set[cpu] = c_idle.idle;
634 	}
635 	init_idle(idle, cpu);
636 	if (smp_alloc_lowcore(cpu))
637 		return -ENOMEM;
638 	do {
639 		ccode = sigp(cpu, sigp_initial_cpu_reset);
640 		if (ccode == sigp_busy)
641 			udelay(10);
642 		if (ccode == sigp_not_operational)
643 			goto err_out;
644 	} while (ccode == sigp_busy);
645 
646 	lowcore = (u32)(unsigned long)lowcore_ptr[cpu];
647 	while (sigp_p(lowcore, cpu, sigp_set_prefix) == sigp_busy)
648 		udelay(10);
649 
650 	cpu_lowcore = lowcore_ptr[cpu];
651 	cpu_lowcore->kernel_stack = (unsigned long)
652 		task_stack_page(idle) + THREAD_SIZE;
653 	cpu_lowcore->thread_info = (unsigned long) task_thread_info(idle);
654 	sf = (struct stack_frame *) (cpu_lowcore->kernel_stack
655 				     - sizeof(struct pt_regs)
656 				     - sizeof(struct stack_frame));
657 	memset(sf, 0, sizeof(struct stack_frame));
658 	sf->gprs[9] = (unsigned long) sf;
659 	cpu_lowcore->save_area[15] = (unsigned long) sf;
660 	__ctl_store(cpu_lowcore->cregs_save_area, 0, 15);
661 	atomic_inc(&init_mm.context.attach_count);
662 	asm volatile(
663 		"	stam	0,15,0(%0)"
664 		: : "a" (&cpu_lowcore->access_regs_save_area) : "memory");
665 	cpu_lowcore->percpu_offset = __per_cpu_offset[cpu];
666 	cpu_lowcore->current_task = (unsigned long) idle;
667 	cpu_lowcore->cpu_nr = cpu;
668 	cpu_lowcore->kernel_asce = S390_lowcore.kernel_asce;
669 	cpu_lowcore->machine_flags = S390_lowcore.machine_flags;
670 	cpu_lowcore->ftrace_func = S390_lowcore.ftrace_func;
671 	memcpy(cpu_lowcore->stfle_fac_list, S390_lowcore.stfle_fac_list,
672 	       MAX_FACILITY_BIT/8);
673 	eieio();
674 
675 	while (sigp(cpu, sigp_restart) == sigp_busy)
676 		udelay(10);
677 
678 	while (!cpu_online(cpu))
679 		cpu_relax();
680 	return 0;
681 
682 err_out:
683 	smp_free_lowcore(cpu);
684 	return -EIO;
685 }
686 
687 static int __init setup_possible_cpus(char *s)
688 {
689 	int pcpus, cpu;
690 
691 	pcpus = simple_strtoul(s, NULL, 0);
692 	init_cpu_possible(cpumask_of(0));
693 	for (cpu = 1; cpu < pcpus && cpu < nr_cpu_ids; cpu++)
694 		set_cpu_possible(cpu, true);
695 	return 0;
696 }
697 early_param("possible_cpus", setup_possible_cpus);
698 
699 #ifdef CONFIG_HOTPLUG_CPU
700 
701 int __cpu_disable(void)
702 {
703 	struct ec_creg_mask_parms cr_parms;
704 	int cpu = smp_processor_id();
705 
706 	set_cpu_online(cpu, false);
707 
708 	/* Disable pfault pseudo page faults on this cpu. */
709 	pfault_fini();
710 
711 	memset(&cr_parms.orvals, 0, sizeof(cr_parms.orvals));
712 	memset(&cr_parms.andvals, 0xff, sizeof(cr_parms.andvals));
713 
714 	/* disable all external interrupts */
715 	cr_parms.orvals[0] = 0;
716 	cr_parms.andvals[0] = ~(1 << 15 | 1 << 14 | 1 << 13 | 1 << 11 |
717 				1 << 10 | 1 <<	9 | 1 <<  6 | 1 <<  5 |
718 				1 <<  4);
719 	/* disable all I/O interrupts */
720 	cr_parms.orvals[6] = 0;
721 	cr_parms.andvals[6] = ~(1 << 31 | 1 << 30 | 1 << 29 | 1 << 28 |
722 				1 << 27 | 1 << 26 | 1 << 25 | 1 << 24);
723 	/* disable most machine checks */
724 	cr_parms.orvals[14] = 0;
725 	cr_parms.andvals[14] = ~(1 << 28 | 1 << 27 | 1 << 26 |
726 				 1 << 25 | 1 << 24);
727 
728 	smp_ctl_bit_callback(&cr_parms);
729 
730 	return 0;
731 }
732 
733 void __cpu_die(unsigned int cpu)
734 {
735 	/* Wait until target cpu is down */
736 	while (!cpu_stopped(cpu))
737 		cpu_relax();
738 	while (sigp_p(0, cpu, sigp_set_prefix) == sigp_busy)
739 		udelay(10);
740 	smp_free_lowcore(cpu);
741 	atomic_dec(&init_mm.context.attach_count);
742 }
743 
744 void __noreturn cpu_die(void)
745 {
746 	idle_task_exit();
747 	while (sigp(smp_processor_id(), sigp_stop) == sigp_busy)
748 		cpu_relax();
749 	for (;;);
750 }
751 
752 #endif /* CONFIG_HOTPLUG_CPU */
753 
754 void __init smp_prepare_cpus(unsigned int max_cpus)
755 {
756 #ifndef CONFIG_64BIT
757 	unsigned long save_area = 0;
758 #endif
759 	unsigned long async_stack, panic_stack;
760 	struct _lowcore *lowcore;
761 
762 	smp_detect_cpus();
763 
764 	/* request the 0x1201 emergency signal external interrupt */
765 	if (register_external_interrupt(0x1201, do_ext_call_interrupt) != 0)
766 		panic("Couldn't request external interrupt 0x1201");
767 	/* request the 0x1202 external call external interrupt */
768 	if (register_external_interrupt(0x1202, do_ext_call_interrupt) != 0)
769 		panic("Couldn't request external interrupt 0x1202");
770 
771 	/* Reallocate current lowcore, but keep its contents. */
772 	lowcore = (void *) __get_free_pages(GFP_KERNEL | GFP_DMA, LC_ORDER);
773 	panic_stack = __get_free_page(GFP_KERNEL);
774 	async_stack = __get_free_pages(GFP_KERNEL, ASYNC_ORDER);
775 	BUG_ON(!lowcore || !panic_stack || !async_stack);
776 #ifndef CONFIG_64BIT
777 	if (MACHINE_HAS_IEEE)
778 		save_area = get_zeroed_page(GFP_KERNEL);
779 #endif
780 	local_irq_disable();
781 	local_mcck_disable();
782 	lowcore_ptr[smp_processor_id()] = lowcore;
783 	*lowcore = S390_lowcore;
784 	lowcore->panic_stack = panic_stack + PAGE_SIZE;
785 	lowcore->async_stack = async_stack + ASYNC_SIZE;
786 #ifndef CONFIG_64BIT
787 	if (MACHINE_HAS_IEEE)
788 		lowcore->extended_save_area_addr = (u32) save_area;
789 #endif
790 	set_prefix((u32)(unsigned long) lowcore);
791 	local_mcck_enable();
792 	local_irq_enable();
793 #ifdef CONFIG_64BIT
794 	if (vdso_alloc_per_cpu(smp_processor_id(), &S390_lowcore))
795 		BUG();
796 #endif
797 }
798 
799 void __init smp_prepare_boot_cpu(void)
800 {
801 	BUG_ON(smp_processor_id() != 0);
802 
803 	current_thread_info()->cpu = 0;
804 	set_cpu_present(0, true);
805 	set_cpu_online(0, true);
806 	S390_lowcore.percpu_offset = __per_cpu_offset[0];
807 	current_set[0] = current;
808 	smp_cpu_state[0] = CPU_STATE_CONFIGURED;
809 	smp_cpu_polarization[0] = POLARIZATION_UNKNWN;
810 }
811 
812 void __init smp_cpus_done(unsigned int max_cpus)
813 {
814 }
815 
816 void __init smp_setup_processor_id(void)
817 {
818 	S390_lowcore.cpu_nr = 0;
819 	__cpu_logical_map[0] = stap();
820 }
821 
822 /*
823  * the frequency of the profiling timer can be changed
824  * by writing a multiplier value into /proc/profile.
825  *
826  * usually you want to run this on all CPUs ;)
827  */
828 int setup_profiling_timer(unsigned int multiplier)
829 {
830 	return 0;
831 }
832 
833 #ifdef CONFIG_HOTPLUG_CPU
834 static ssize_t cpu_configure_show(struct sys_device *dev,
835 				struct sysdev_attribute *attr, char *buf)
836 {
837 	ssize_t count;
838 
839 	mutex_lock(&smp_cpu_state_mutex);
840 	count = sprintf(buf, "%d\n", smp_cpu_state[dev->id]);
841 	mutex_unlock(&smp_cpu_state_mutex);
842 	return count;
843 }
844 
845 static ssize_t cpu_configure_store(struct sys_device *dev,
846 				  struct sysdev_attribute *attr,
847 				  const char *buf, size_t count)
848 {
849 	int cpu = dev->id;
850 	int val, rc;
851 	char delim;
852 
853 	if (sscanf(buf, "%d %c", &val, &delim) != 1)
854 		return -EINVAL;
855 	if (val != 0 && val != 1)
856 		return -EINVAL;
857 
858 	get_online_cpus();
859 	mutex_lock(&smp_cpu_state_mutex);
860 	rc = -EBUSY;
861 	/* disallow configuration changes of online cpus and cpu 0 */
862 	if (cpu_online(cpu) || cpu == 0)
863 		goto out;
864 	rc = 0;
865 	switch (val) {
866 	case 0:
867 		if (smp_cpu_state[cpu] == CPU_STATE_CONFIGURED) {
868 			rc = sclp_cpu_deconfigure(__cpu_logical_map[cpu]);
869 			if (!rc) {
870 				smp_cpu_state[cpu] = CPU_STATE_STANDBY;
871 				smp_cpu_polarization[cpu] = POLARIZATION_UNKNWN;
872 			}
873 		}
874 		break;
875 	case 1:
876 		if (smp_cpu_state[cpu] == CPU_STATE_STANDBY) {
877 			rc = sclp_cpu_configure(__cpu_logical_map[cpu]);
878 			if (!rc) {
879 				smp_cpu_state[cpu] = CPU_STATE_CONFIGURED;
880 				smp_cpu_polarization[cpu] = POLARIZATION_UNKNWN;
881 			}
882 		}
883 		break;
884 	default:
885 		break;
886 	}
887 out:
888 	mutex_unlock(&smp_cpu_state_mutex);
889 	put_online_cpus();
890 	return rc ? rc : count;
891 }
892 static SYSDEV_ATTR(configure, 0644, cpu_configure_show, cpu_configure_store);
893 #endif /* CONFIG_HOTPLUG_CPU */
894 
895 static ssize_t cpu_polarization_show(struct sys_device *dev,
896 				     struct sysdev_attribute *attr, char *buf)
897 {
898 	int cpu = dev->id;
899 	ssize_t count;
900 
901 	mutex_lock(&smp_cpu_state_mutex);
902 	switch (smp_cpu_polarization[cpu]) {
903 	case POLARIZATION_HRZ:
904 		count = sprintf(buf, "horizontal\n");
905 		break;
906 	case POLARIZATION_VL:
907 		count = sprintf(buf, "vertical:low\n");
908 		break;
909 	case POLARIZATION_VM:
910 		count = sprintf(buf, "vertical:medium\n");
911 		break;
912 	case POLARIZATION_VH:
913 		count = sprintf(buf, "vertical:high\n");
914 		break;
915 	default:
916 		count = sprintf(buf, "unknown\n");
917 		break;
918 	}
919 	mutex_unlock(&smp_cpu_state_mutex);
920 	return count;
921 }
922 static SYSDEV_ATTR(polarization, 0444, cpu_polarization_show, NULL);
923 
924 static ssize_t show_cpu_address(struct sys_device *dev,
925 				struct sysdev_attribute *attr, char *buf)
926 {
927 	return sprintf(buf, "%d\n", __cpu_logical_map[dev->id]);
928 }
929 static SYSDEV_ATTR(address, 0444, show_cpu_address, NULL);
930 
931 
932 static struct attribute *cpu_common_attrs[] = {
933 #ifdef CONFIG_HOTPLUG_CPU
934 	&attr_configure.attr,
935 #endif
936 	&attr_address.attr,
937 	&attr_polarization.attr,
938 	NULL,
939 };
940 
941 static struct attribute_group cpu_common_attr_group = {
942 	.attrs = cpu_common_attrs,
943 };
944 
945 static ssize_t show_capability(struct sys_device *dev,
946 				struct sysdev_attribute *attr, char *buf)
947 {
948 	unsigned int capability;
949 	int rc;
950 
951 	rc = get_cpu_capability(&capability);
952 	if (rc)
953 		return rc;
954 	return sprintf(buf, "%u\n", capability);
955 }
956 static SYSDEV_ATTR(capability, 0444, show_capability, NULL);
957 
958 static ssize_t show_idle_count(struct sys_device *dev,
959 				struct sysdev_attribute *attr, char *buf)
960 {
961 	struct s390_idle_data *idle;
962 	unsigned long long idle_count;
963 	unsigned int sequence;
964 
965 	idle = &per_cpu(s390_idle, dev->id);
966 repeat:
967 	sequence = idle->sequence;
968 	smp_rmb();
969 	if (sequence & 1)
970 		goto repeat;
971 	idle_count = idle->idle_count;
972 	if (idle->idle_enter)
973 		idle_count++;
974 	smp_rmb();
975 	if (idle->sequence != sequence)
976 		goto repeat;
977 	return sprintf(buf, "%llu\n", idle_count);
978 }
979 static SYSDEV_ATTR(idle_count, 0444, show_idle_count, NULL);
980 
981 static ssize_t show_idle_time(struct sys_device *dev,
982 				struct sysdev_attribute *attr, char *buf)
983 {
984 	struct s390_idle_data *idle;
985 	unsigned long long now, idle_time, idle_enter;
986 	unsigned int sequence;
987 
988 	idle = &per_cpu(s390_idle, dev->id);
989 	now = get_clock();
990 repeat:
991 	sequence = idle->sequence;
992 	smp_rmb();
993 	if (sequence & 1)
994 		goto repeat;
995 	idle_time = idle->idle_time;
996 	idle_enter = idle->idle_enter;
997 	if (idle_enter != 0ULL && idle_enter < now)
998 		idle_time += now - idle_enter;
999 	smp_rmb();
1000 	if (idle->sequence != sequence)
1001 		goto repeat;
1002 	return sprintf(buf, "%llu\n", idle_time >> 12);
1003 }
1004 static SYSDEV_ATTR(idle_time_us, 0444, show_idle_time, NULL);
1005 
1006 static struct attribute *cpu_online_attrs[] = {
1007 	&attr_capability.attr,
1008 	&attr_idle_count.attr,
1009 	&attr_idle_time_us.attr,
1010 	NULL,
1011 };
1012 
1013 static struct attribute_group cpu_online_attr_group = {
1014 	.attrs = cpu_online_attrs,
1015 };
1016 
1017 static int __cpuinit smp_cpu_notify(struct notifier_block *self,
1018 				    unsigned long action, void *hcpu)
1019 {
1020 	unsigned int cpu = (unsigned int)(long)hcpu;
1021 	struct cpu *c = &per_cpu(cpu_devices, cpu);
1022 	struct sys_device *s = &c->sysdev;
1023 	struct s390_idle_data *idle;
1024 	int err = 0;
1025 
1026 	switch (action) {
1027 	case CPU_ONLINE:
1028 	case CPU_ONLINE_FROZEN:
1029 		idle = &per_cpu(s390_idle, cpu);
1030 		memset(idle, 0, sizeof(struct s390_idle_data));
1031 		err = sysfs_create_group(&s->kobj, &cpu_online_attr_group);
1032 		break;
1033 	case CPU_DEAD:
1034 	case CPU_DEAD_FROZEN:
1035 		sysfs_remove_group(&s->kobj, &cpu_online_attr_group);
1036 		break;
1037 	}
1038 	return notifier_from_errno(err);
1039 }
1040 
1041 static struct notifier_block __cpuinitdata smp_cpu_nb = {
1042 	.notifier_call = smp_cpu_notify,
1043 };
1044 
1045 static int __devinit smp_add_present_cpu(int cpu)
1046 {
1047 	struct cpu *c = &per_cpu(cpu_devices, cpu);
1048 	struct sys_device *s = &c->sysdev;
1049 	int rc;
1050 
1051 	c->hotpluggable = 1;
1052 	rc = register_cpu(c, cpu);
1053 	if (rc)
1054 		goto out;
1055 	rc = sysfs_create_group(&s->kobj, &cpu_common_attr_group);
1056 	if (rc)
1057 		goto out_cpu;
1058 	if (!cpu_online(cpu))
1059 		goto out;
1060 	rc = sysfs_create_group(&s->kobj, &cpu_online_attr_group);
1061 	if (!rc)
1062 		return 0;
1063 	sysfs_remove_group(&s->kobj, &cpu_common_attr_group);
1064 out_cpu:
1065 #ifdef CONFIG_HOTPLUG_CPU
1066 	unregister_cpu(c);
1067 #endif
1068 out:
1069 	return rc;
1070 }
1071 
1072 #ifdef CONFIG_HOTPLUG_CPU
1073 
1074 int __ref smp_rescan_cpus(void)
1075 {
1076 	cpumask_t newcpus;
1077 	int cpu;
1078 	int rc;
1079 
1080 	get_online_cpus();
1081 	mutex_lock(&smp_cpu_state_mutex);
1082 	cpumask_copy(&newcpus, cpu_present_mask);
1083 	rc = __smp_rescan_cpus();
1084 	if (rc)
1085 		goto out;
1086 	cpumask_andnot(&newcpus, cpu_present_mask, &newcpus);
1087 	for_each_cpu(cpu, &newcpus) {
1088 		rc = smp_add_present_cpu(cpu);
1089 		if (rc)
1090 			set_cpu_present(cpu, false);
1091 	}
1092 	rc = 0;
1093 out:
1094 	mutex_unlock(&smp_cpu_state_mutex);
1095 	put_online_cpus();
1096 	if (!cpumask_empty(&newcpus))
1097 		topology_schedule_update();
1098 	return rc;
1099 }
1100 
1101 static ssize_t __ref rescan_store(struct sysdev_class *class,
1102 				  struct sysdev_class_attribute *attr,
1103 				  const char *buf,
1104 				  size_t count)
1105 {
1106 	int rc;
1107 
1108 	rc = smp_rescan_cpus();
1109 	return rc ? rc : count;
1110 }
1111 static SYSDEV_CLASS_ATTR(rescan, 0200, NULL, rescan_store);
1112 #endif /* CONFIG_HOTPLUG_CPU */
1113 
1114 static ssize_t dispatching_show(struct sysdev_class *class,
1115 				struct sysdev_class_attribute *attr,
1116 				char *buf)
1117 {
1118 	ssize_t count;
1119 
1120 	mutex_lock(&smp_cpu_state_mutex);
1121 	count = sprintf(buf, "%d\n", cpu_management);
1122 	mutex_unlock(&smp_cpu_state_mutex);
1123 	return count;
1124 }
1125 
1126 static ssize_t dispatching_store(struct sysdev_class *dev,
1127 				 struct sysdev_class_attribute *attr,
1128 				 const char *buf,
1129 				 size_t count)
1130 {
1131 	int val, rc;
1132 	char delim;
1133 
1134 	if (sscanf(buf, "%d %c", &val, &delim) != 1)
1135 		return -EINVAL;
1136 	if (val != 0 && val != 1)
1137 		return -EINVAL;
1138 	rc = 0;
1139 	get_online_cpus();
1140 	mutex_lock(&smp_cpu_state_mutex);
1141 	if (cpu_management == val)
1142 		goto out;
1143 	rc = topology_set_cpu_management(val);
1144 	if (!rc)
1145 		cpu_management = val;
1146 out:
1147 	mutex_unlock(&smp_cpu_state_mutex);
1148 	put_online_cpus();
1149 	return rc ? rc : count;
1150 }
1151 static SYSDEV_CLASS_ATTR(dispatching, 0644, dispatching_show,
1152 			 dispatching_store);
1153 
1154 static int __init topology_init(void)
1155 {
1156 	int cpu;
1157 	int rc;
1158 
1159 	register_cpu_notifier(&smp_cpu_nb);
1160 
1161 #ifdef CONFIG_HOTPLUG_CPU
1162 	rc = sysdev_class_create_file(&cpu_sysdev_class, &attr_rescan);
1163 	if (rc)
1164 		return rc;
1165 #endif
1166 	rc = sysdev_class_create_file(&cpu_sysdev_class, &attr_dispatching);
1167 	if (rc)
1168 		return rc;
1169 	for_each_present_cpu(cpu) {
1170 		rc = smp_add_present_cpu(cpu);
1171 		if (rc)
1172 			return rc;
1173 	}
1174 	return 0;
1175 }
1176 subsys_initcall(topology_init);
1177