xref: /linux/arch/s390/kernel/smp.c (revision b43ab901d671e3e3cad425ea5e9a3c74e266dcdd)
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 static int smp_cpu_state[NR_CPUS];
73 
74 static DEFINE_PER_CPU(struct cpu, cpu_devices);
75 
76 static void smp_ext_bitcall(int, int);
77 
78 static int raw_cpu_stopped(int cpu)
79 {
80 	u32 status;
81 
82 	switch (raw_sigp_ps(&status, 0, cpu, sigp_sense)) {
83 	case sigp_status_stored:
84 		/* Check for stopped and check stop state */
85 		if (status & 0x50)
86 			return 1;
87 		break;
88 	default:
89 		break;
90 	}
91 	return 0;
92 }
93 
94 static inline int cpu_stopped(int cpu)
95 {
96 	return raw_cpu_stopped(cpu_logical_map(cpu));
97 }
98 
99 /*
100  * Ensure that PSW restart is done on an online CPU
101  */
102 void smp_restart_with_online_cpu(void)
103 {
104 	int cpu;
105 
106 	for_each_online_cpu(cpu) {
107 		if (stap() == __cpu_logical_map[cpu]) {
108 			/* We are online: Enable DAT again and return */
109 			__load_psw_mask(psw_kernel_bits | PSW_MASK_DAT);
110 			return;
111 		}
112 	}
113 	/* We are not online: Do PSW restart on an online CPU */
114 	while (sigp(cpu, sigp_restart) == sigp_busy)
115 		cpu_relax();
116 	/* And stop ourself */
117 	while (raw_sigp(stap(), sigp_stop) == sigp_busy)
118 		cpu_relax();
119 	for (;;);
120 }
121 
122 void smp_switch_to_ipl_cpu(void (*func)(void *), void *data)
123 {
124 	struct _lowcore *lc, *current_lc;
125 	struct stack_frame *sf;
126 	struct pt_regs *regs;
127 	unsigned long sp;
128 
129 	if (smp_processor_id() == 0)
130 		func(data);
131 	__load_psw_mask(PSW_DEFAULT_KEY | PSW_MASK_BASE |
132 			PSW_MASK_EA | PSW_MASK_BA);
133 	/* Disable lowcore protection */
134 	__ctl_clear_bit(0, 28);
135 	current_lc = lowcore_ptr[smp_processor_id()];
136 	lc = lowcore_ptr[0];
137 	if (!lc)
138 		lc = current_lc;
139 	lc->restart_psw.mask =
140 		PSW_DEFAULT_KEY | PSW_MASK_BASE | PSW_MASK_EA | PSW_MASK_BA;
141 	lc->restart_psw.addr = PSW_ADDR_AMODE | (unsigned long) smp_restart_cpu;
142 	if (!cpu_online(0))
143 		smp_switch_to_cpu(func, data, 0, stap(), __cpu_logical_map[0]);
144 	while (sigp(0, sigp_stop_and_store_status) == sigp_busy)
145 		cpu_relax();
146 	sp = lc->panic_stack;
147 	sp -= sizeof(struct pt_regs);
148 	regs = (struct pt_regs *) sp;
149 	memcpy(&regs->gprs, &current_lc->gpregs_save_area, sizeof(regs->gprs));
150 	regs->psw = current_lc->psw_save_area;
151 	sp -= STACK_FRAME_OVERHEAD;
152 	sf = (struct stack_frame *) sp;
153 	sf->back_chain = 0;
154 	smp_switch_to_cpu(func, data, sp, stap(), __cpu_logical_map[0]);
155 }
156 
157 static void smp_stop_cpu(void)
158 {
159 	while (sigp(smp_processor_id(), sigp_stop) == sigp_busy)
160 		cpu_relax();
161 }
162 
163 void smp_send_stop(void)
164 {
165 	cpumask_t cpumask;
166 	int cpu;
167 	u64 end;
168 
169 	/* Disable all interrupts/machine checks */
170 	__load_psw_mask(psw_kernel_bits | PSW_MASK_DAT);
171 	trace_hardirqs_off();
172 
173 	cpumask_copy(&cpumask, cpu_online_mask);
174 	cpumask_clear_cpu(smp_processor_id(), &cpumask);
175 
176 	if (oops_in_progress) {
177 		/*
178 		 * Give the other cpus the opportunity to complete
179 		 * outstanding interrupts before stopping them.
180 		 */
181 		end = get_clock() + (1000000UL << 12);
182 		for_each_cpu(cpu, &cpumask) {
183 			set_bit(ec_stop_cpu, (unsigned long *)
184 				&lowcore_ptr[cpu]->ext_call_fast);
185 			while (sigp(cpu, sigp_emergency_signal) == sigp_busy &&
186 			       get_clock() < end)
187 				cpu_relax();
188 		}
189 		while (get_clock() < end) {
190 			for_each_cpu(cpu, &cpumask)
191 				if (cpu_stopped(cpu))
192 					cpumask_clear_cpu(cpu, &cpumask);
193 			if (cpumask_empty(&cpumask))
194 				break;
195 			cpu_relax();
196 		}
197 	}
198 
199 	/* stop all processors */
200 	for_each_cpu(cpu, &cpumask) {
201 		while (sigp(cpu, sigp_stop) == sigp_busy)
202 			cpu_relax();
203 		while (!cpu_stopped(cpu))
204 			cpu_relax();
205 	}
206 }
207 
208 /*
209  * This is the main routine where commands issued by other
210  * cpus are handled.
211  */
212 
213 static void do_ext_call_interrupt(unsigned int ext_int_code,
214 				  unsigned int param32, unsigned long param64)
215 {
216 	unsigned long bits;
217 
218 	if ((ext_int_code & 0xffff) == 0x1202)
219 		kstat_cpu(smp_processor_id()).irqs[EXTINT_EXC]++;
220 	else
221 		kstat_cpu(smp_processor_id()).irqs[EXTINT_EMS]++;
222 	/*
223 	 * handle bit signal external calls
224 	 */
225 	bits = xchg(&S390_lowcore.ext_call_fast, 0);
226 
227 	if (test_bit(ec_stop_cpu, &bits))
228 		smp_stop_cpu();
229 
230 	if (test_bit(ec_schedule, &bits))
231 		scheduler_ipi();
232 
233 	if (test_bit(ec_call_function, &bits))
234 		generic_smp_call_function_interrupt();
235 
236 	if (test_bit(ec_call_function_single, &bits))
237 		generic_smp_call_function_single_interrupt();
238 
239 }
240 
241 /*
242  * Send an external call sigp to another cpu and return without waiting
243  * for its completion.
244  */
245 static void smp_ext_bitcall(int cpu, int sig)
246 {
247 	int order;
248 
249 	/*
250 	 * Set signaling bit in lowcore of target cpu and kick it
251 	 */
252 	set_bit(sig, (unsigned long *) &lowcore_ptr[cpu]->ext_call_fast);
253 	while (1) {
254 		order = smp_vcpu_scheduled(cpu) ?
255 			sigp_external_call : sigp_emergency_signal;
256 		if (sigp(cpu, order) != sigp_busy)
257 			break;
258 		udelay(10);
259 	}
260 }
261 
262 void arch_send_call_function_ipi_mask(const struct cpumask *mask)
263 {
264 	int cpu;
265 
266 	for_each_cpu(cpu, mask)
267 		smp_ext_bitcall(cpu, ec_call_function);
268 }
269 
270 void arch_send_call_function_single_ipi(int cpu)
271 {
272 	smp_ext_bitcall(cpu, ec_call_function_single);
273 }
274 
275 #ifndef CONFIG_64BIT
276 /*
277  * this function sends a 'purge tlb' signal to another CPU.
278  */
279 static void smp_ptlb_callback(void *info)
280 {
281 	__tlb_flush_local();
282 }
283 
284 void smp_ptlb_all(void)
285 {
286 	on_each_cpu(smp_ptlb_callback, NULL, 1);
287 }
288 EXPORT_SYMBOL(smp_ptlb_all);
289 #endif /* ! CONFIG_64BIT */
290 
291 /*
292  * this function sends a 'reschedule' IPI to another CPU.
293  * it goes straight through and wastes no time serializing
294  * anything. Worst case is that we lose a reschedule ...
295  */
296 void smp_send_reschedule(int cpu)
297 {
298 	smp_ext_bitcall(cpu, ec_schedule);
299 }
300 
301 /*
302  * parameter area for the set/clear control bit callbacks
303  */
304 struct ec_creg_mask_parms {
305 	unsigned long orvals[16];
306 	unsigned long andvals[16];
307 };
308 
309 /*
310  * callback for setting/clearing control bits
311  */
312 static void smp_ctl_bit_callback(void *info)
313 {
314 	struct ec_creg_mask_parms *pp = info;
315 	unsigned long cregs[16];
316 	int i;
317 
318 	__ctl_store(cregs, 0, 15);
319 	for (i = 0; i <= 15; i++)
320 		cregs[i] = (cregs[i] & pp->andvals[i]) | pp->orvals[i];
321 	__ctl_load(cregs, 0, 15);
322 }
323 
324 /*
325  * Set a bit in a control register of all cpus
326  */
327 void smp_ctl_set_bit(int cr, int bit)
328 {
329 	struct ec_creg_mask_parms parms;
330 
331 	memset(&parms.orvals, 0, sizeof(parms.orvals));
332 	memset(&parms.andvals, 0xff, sizeof(parms.andvals));
333 	parms.orvals[cr] = 1UL << bit;
334 	on_each_cpu(smp_ctl_bit_callback, &parms, 1);
335 }
336 EXPORT_SYMBOL(smp_ctl_set_bit);
337 
338 /*
339  * Clear a bit in a control register of all cpus
340  */
341 void smp_ctl_clear_bit(int cr, int bit)
342 {
343 	struct ec_creg_mask_parms parms;
344 
345 	memset(&parms.orvals, 0, sizeof(parms.orvals));
346 	memset(&parms.andvals, 0xff, sizeof(parms.andvals));
347 	parms.andvals[cr] = ~(1UL << bit);
348 	on_each_cpu(smp_ctl_bit_callback, &parms, 1);
349 }
350 EXPORT_SYMBOL(smp_ctl_clear_bit);
351 
352 #if defined(CONFIG_ZFCPDUMP) || defined(CONFIG_CRASH_DUMP)
353 
354 static void __init smp_get_save_area(unsigned int cpu, unsigned int phy_cpu)
355 {
356 	if (ipl_info.type != IPL_TYPE_FCP_DUMP && !OLDMEM_BASE)
357 		return;
358 	if (is_kdump_kernel())
359 		return;
360 	if (cpu >= NR_CPUS) {
361 		pr_warning("CPU %i exceeds the maximum %i and is excluded from "
362 			   "the dump\n", cpu, NR_CPUS - 1);
363 		return;
364 	}
365 	zfcpdump_save_areas[cpu] = kmalloc(sizeof(struct save_area), GFP_KERNEL);
366 	while (raw_sigp(phy_cpu, sigp_stop_and_store_status) == sigp_busy)
367 		cpu_relax();
368 	memcpy_real(zfcpdump_save_areas[cpu],
369 		    (void *)(unsigned long) store_prefix() + SAVE_AREA_BASE,
370 		    sizeof(struct save_area));
371 }
372 
373 struct save_area *zfcpdump_save_areas[NR_CPUS + 1];
374 EXPORT_SYMBOL_GPL(zfcpdump_save_areas);
375 
376 #else
377 
378 static inline void smp_get_save_area(unsigned int cpu, unsigned int phy_cpu) { }
379 
380 #endif /* CONFIG_ZFCPDUMP */
381 
382 static int cpu_known(int cpu_id)
383 {
384 	int cpu;
385 
386 	for_each_present_cpu(cpu) {
387 		if (__cpu_logical_map[cpu] == cpu_id)
388 			return 1;
389 	}
390 	return 0;
391 }
392 
393 static int smp_rescan_cpus_sigp(cpumask_t avail)
394 {
395 	int cpu_id, logical_cpu;
396 
397 	logical_cpu = cpumask_first(&avail);
398 	if (logical_cpu >= nr_cpu_ids)
399 		return 0;
400 	for (cpu_id = 0; cpu_id <= MAX_CPU_ADDRESS; cpu_id++) {
401 		if (cpu_known(cpu_id))
402 			continue;
403 		__cpu_logical_map[logical_cpu] = cpu_id;
404 		cpu_set_polarization(logical_cpu, POLARIZATION_UNKNOWN);
405 		if (!cpu_stopped(logical_cpu))
406 			continue;
407 		set_cpu_present(logical_cpu, true);
408 		smp_cpu_state[logical_cpu] = CPU_STATE_CONFIGURED;
409 		logical_cpu = cpumask_next(logical_cpu, &avail);
410 		if (logical_cpu >= nr_cpu_ids)
411 			break;
412 	}
413 	return 0;
414 }
415 
416 static int smp_rescan_cpus_sclp(cpumask_t avail)
417 {
418 	struct sclp_cpu_info *info;
419 	int cpu_id, logical_cpu, cpu;
420 	int rc;
421 
422 	logical_cpu = cpumask_first(&avail);
423 	if (logical_cpu >= nr_cpu_ids)
424 		return 0;
425 	info = kmalloc(sizeof(*info), GFP_KERNEL);
426 	if (!info)
427 		return -ENOMEM;
428 	rc = sclp_get_cpu_info(info);
429 	if (rc)
430 		goto out;
431 	for (cpu = 0; cpu < info->combined; cpu++) {
432 		if (info->has_cpu_type && info->cpu[cpu].type != smp_cpu_type)
433 			continue;
434 		cpu_id = info->cpu[cpu].address;
435 		if (cpu_known(cpu_id))
436 			continue;
437 		__cpu_logical_map[logical_cpu] = cpu_id;
438 		cpu_set_polarization(logical_cpu, POLARIZATION_UNKNOWN);
439 		set_cpu_present(logical_cpu, true);
440 		if (cpu >= info->configured)
441 			smp_cpu_state[logical_cpu] = CPU_STATE_STANDBY;
442 		else
443 			smp_cpu_state[logical_cpu] = CPU_STATE_CONFIGURED;
444 		logical_cpu = cpumask_next(logical_cpu, &avail);
445 		if (logical_cpu >= nr_cpu_ids)
446 			break;
447 	}
448 out:
449 	kfree(info);
450 	return rc;
451 }
452 
453 static int __smp_rescan_cpus(void)
454 {
455 	cpumask_t avail;
456 
457 	cpumask_xor(&avail, cpu_possible_mask, cpu_present_mask);
458 	if (smp_use_sigp_detection)
459 		return smp_rescan_cpus_sigp(avail);
460 	else
461 		return smp_rescan_cpus_sclp(avail);
462 }
463 
464 static void __init smp_detect_cpus(void)
465 {
466 	unsigned int cpu, c_cpus, s_cpus;
467 	struct sclp_cpu_info *info;
468 	u16 boot_cpu_addr, cpu_addr;
469 
470 	c_cpus = 1;
471 	s_cpus = 0;
472 	boot_cpu_addr = __cpu_logical_map[0];
473 	info = kmalloc(sizeof(*info), GFP_KERNEL);
474 	if (!info)
475 		panic("smp_detect_cpus failed to allocate memory\n");
476 #ifdef CONFIG_CRASH_DUMP
477 	if (OLDMEM_BASE && !is_kdump_kernel()) {
478 		struct save_area *save_area;
479 
480 		save_area = kmalloc(sizeof(*save_area), GFP_KERNEL);
481 		if (!save_area)
482 			panic("could not allocate memory for save area\n");
483 		copy_oldmem_page(1, (void *) save_area, sizeof(*save_area),
484 				 0x200, 0);
485 		zfcpdump_save_areas[0] = save_area;
486 	}
487 #endif
488 	/* Use sigp detection algorithm if sclp doesn't work. */
489 	if (sclp_get_cpu_info(info)) {
490 		smp_use_sigp_detection = 1;
491 		for (cpu = 0; cpu <= MAX_CPU_ADDRESS; cpu++) {
492 			if (cpu == boot_cpu_addr)
493 				continue;
494 			if (!raw_cpu_stopped(cpu))
495 				continue;
496 			smp_get_save_area(c_cpus, cpu);
497 			c_cpus++;
498 		}
499 		goto out;
500 	}
501 
502 	if (info->has_cpu_type) {
503 		for (cpu = 0; cpu < info->combined; cpu++) {
504 			if (info->cpu[cpu].address == boot_cpu_addr) {
505 				smp_cpu_type = info->cpu[cpu].type;
506 				break;
507 			}
508 		}
509 	}
510 
511 	for (cpu = 0; cpu < info->combined; cpu++) {
512 		if (info->has_cpu_type && info->cpu[cpu].type != smp_cpu_type)
513 			continue;
514 		cpu_addr = info->cpu[cpu].address;
515 		if (cpu_addr == boot_cpu_addr)
516 			continue;
517 		if (!raw_cpu_stopped(cpu_addr)) {
518 			s_cpus++;
519 			continue;
520 		}
521 		smp_get_save_area(c_cpus, cpu_addr);
522 		c_cpus++;
523 	}
524 out:
525 	kfree(info);
526 	pr_info("%d configured CPUs, %d standby CPUs\n", c_cpus, s_cpus);
527 	get_online_cpus();
528 	__smp_rescan_cpus();
529 	put_online_cpus();
530 }
531 
532 /*
533  *	Activate a secondary processor.
534  */
535 int __cpuinit start_secondary(void *cpuvoid)
536 {
537 	cpu_init();
538 	preempt_disable();
539 	init_cpu_timer();
540 	init_cpu_vtimer();
541 	pfault_init();
542 
543 	notify_cpu_starting(smp_processor_id());
544 	ipi_call_lock();
545 	set_cpu_online(smp_processor_id(), true);
546 	ipi_call_unlock();
547 	__ctl_clear_bit(0, 28); /* Disable lowcore protection */
548 	S390_lowcore.restart_psw.mask =
549 		PSW_DEFAULT_KEY | PSW_MASK_BASE | PSW_MASK_EA | PSW_MASK_BA;
550 	S390_lowcore.restart_psw.addr =
551 		PSW_ADDR_AMODE | (unsigned long) psw_restart_int_handler;
552 	__ctl_set_bit(0, 28); /* Enable lowcore protection */
553 	/*
554 	 * Wait until the cpu which brought this one up marked it
555 	 * active before enabling interrupts.
556 	 */
557 	while (!cpumask_test_cpu(smp_processor_id(), cpu_active_mask))
558 		cpu_relax();
559 	local_irq_enable();
560 	/* cpu_idle will call schedule for us */
561 	cpu_idle();
562 	return 0;
563 }
564 
565 struct create_idle {
566 	struct work_struct work;
567 	struct task_struct *idle;
568 	struct completion done;
569 	int cpu;
570 };
571 
572 static void __cpuinit smp_fork_idle(struct work_struct *work)
573 {
574 	struct create_idle *c_idle;
575 
576 	c_idle = container_of(work, struct create_idle, work);
577 	c_idle->idle = fork_idle(c_idle->cpu);
578 	complete(&c_idle->done);
579 }
580 
581 static int __cpuinit smp_alloc_lowcore(int cpu)
582 {
583 	unsigned long async_stack, panic_stack;
584 	struct _lowcore *lowcore;
585 
586 	lowcore = (void *) __get_free_pages(GFP_KERNEL | GFP_DMA, LC_ORDER);
587 	if (!lowcore)
588 		return -ENOMEM;
589 	async_stack = __get_free_pages(GFP_KERNEL, ASYNC_ORDER);
590 	panic_stack = __get_free_page(GFP_KERNEL);
591 	if (!panic_stack || !async_stack)
592 		goto out;
593 	memcpy(lowcore, &S390_lowcore, 512);
594 	memset((char *)lowcore + 512, 0, sizeof(*lowcore) - 512);
595 	lowcore->async_stack = async_stack + ASYNC_SIZE;
596 	lowcore->panic_stack = panic_stack + PAGE_SIZE;
597 	lowcore->restart_psw.mask =
598 		PSW_DEFAULT_KEY | PSW_MASK_BASE | PSW_MASK_EA | PSW_MASK_BA;
599 	lowcore->restart_psw.addr =
600 		PSW_ADDR_AMODE | (unsigned long) restart_int_handler;
601 	if (user_mode != HOME_SPACE_MODE)
602 		lowcore->restart_psw.mask |= PSW_ASC_HOME;
603 #ifndef CONFIG_64BIT
604 	if (MACHINE_HAS_IEEE) {
605 		unsigned long save_area;
606 
607 		save_area = get_zeroed_page(GFP_KERNEL);
608 		if (!save_area)
609 			goto out;
610 		lowcore->extended_save_area_addr = (u32) save_area;
611 	}
612 #else
613 	if (vdso_alloc_per_cpu(cpu, lowcore))
614 		goto out;
615 #endif
616 	lowcore_ptr[cpu] = lowcore;
617 	return 0;
618 
619 out:
620 	free_page(panic_stack);
621 	free_pages(async_stack, ASYNC_ORDER);
622 	free_pages((unsigned long) lowcore, LC_ORDER);
623 	return -ENOMEM;
624 }
625 
626 static void smp_free_lowcore(int cpu)
627 {
628 	struct _lowcore *lowcore;
629 
630 	lowcore = lowcore_ptr[cpu];
631 #ifndef CONFIG_64BIT
632 	if (MACHINE_HAS_IEEE)
633 		free_page((unsigned long) lowcore->extended_save_area_addr);
634 #else
635 	vdso_free_per_cpu(cpu, lowcore);
636 #endif
637 	free_page(lowcore->panic_stack - PAGE_SIZE);
638 	free_pages(lowcore->async_stack - ASYNC_SIZE, ASYNC_ORDER);
639 	free_pages((unsigned long) lowcore, LC_ORDER);
640 	lowcore_ptr[cpu] = NULL;
641 }
642 
643 /* Upping and downing of CPUs */
644 int __cpuinit __cpu_up(unsigned int cpu)
645 {
646 	struct _lowcore *cpu_lowcore;
647 	struct create_idle c_idle;
648 	struct task_struct *idle;
649 	struct stack_frame *sf;
650 	u32 lowcore;
651 	int ccode;
652 
653 	if (smp_cpu_state[cpu] != CPU_STATE_CONFIGURED)
654 		return -EIO;
655 	idle = current_set[cpu];
656 	if (!idle) {
657 		c_idle.done = COMPLETION_INITIALIZER_ONSTACK(c_idle.done);
658 		INIT_WORK_ONSTACK(&c_idle.work, smp_fork_idle);
659 		c_idle.cpu = cpu;
660 		schedule_work(&c_idle.work);
661 		wait_for_completion(&c_idle.done);
662 		if (IS_ERR(c_idle.idle))
663 			return PTR_ERR(c_idle.idle);
664 		idle = c_idle.idle;
665 		current_set[cpu] = c_idle.idle;
666 	}
667 	init_idle(idle, cpu);
668 	if (smp_alloc_lowcore(cpu))
669 		return -ENOMEM;
670 	do {
671 		ccode = sigp(cpu, sigp_initial_cpu_reset);
672 		if (ccode == sigp_busy)
673 			udelay(10);
674 		if (ccode == sigp_not_operational)
675 			goto err_out;
676 	} while (ccode == sigp_busy);
677 
678 	lowcore = (u32)(unsigned long)lowcore_ptr[cpu];
679 	while (sigp_p(lowcore, cpu, sigp_set_prefix) == sigp_busy)
680 		udelay(10);
681 
682 	cpu_lowcore = lowcore_ptr[cpu];
683 	cpu_lowcore->kernel_stack = (unsigned long)
684 		task_stack_page(idle) + THREAD_SIZE;
685 	cpu_lowcore->thread_info = (unsigned long) task_thread_info(idle);
686 	sf = (struct stack_frame *) (cpu_lowcore->kernel_stack
687 				     - sizeof(struct pt_regs)
688 				     - sizeof(struct stack_frame));
689 	memset(sf, 0, sizeof(struct stack_frame));
690 	sf->gprs[9] = (unsigned long) sf;
691 	cpu_lowcore->gpregs_save_area[15] = (unsigned long) sf;
692 	__ctl_store(cpu_lowcore->cregs_save_area, 0, 15);
693 	atomic_inc(&init_mm.context.attach_count);
694 	asm volatile(
695 		"	stam	0,15,0(%0)"
696 		: : "a" (&cpu_lowcore->access_regs_save_area) : "memory");
697 	cpu_lowcore->percpu_offset = __per_cpu_offset[cpu];
698 	cpu_lowcore->current_task = (unsigned long) idle;
699 	cpu_lowcore->cpu_nr = cpu;
700 	cpu_lowcore->kernel_asce = S390_lowcore.kernel_asce;
701 	cpu_lowcore->machine_flags = S390_lowcore.machine_flags;
702 	cpu_lowcore->ftrace_func = S390_lowcore.ftrace_func;
703 	memcpy(cpu_lowcore->stfle_fac_list, S390_lowcore.stfle_fac_list,
704 	       MAX_FACILITY_BIT/8);
705 	eieio();
706 
707 	while (sigp(cpu, sigp_restart) == sigp_busy)
708 		udelay(10);
709 
710 	while (!cpu_online(cpu))
711 		cpu_relax();
712 	return 0;
713 
714 err_out:
715 	smp_free_lowcore(cpu);
716 	return -EIO;
717 }
718 
719 static int __init setup_possible_cpus(char *s)
720 {
721 	int pcpus, cpu;
722 
723 	pcpus = simple_strtoul(s, NULL, 0);
724 	init_cpu_possible(cpumask_of(0));
725 	for (cpu = 1; cpu < pcpus && cpu < nr_cpu_ids; cpu++)
726 		set_cpu_possible(cpu, true);
727 	return 0;
728 }
729 early_param("possible_cpus", setup_possible_cpus);
730 
731 #ifdef CONFIG_HOTPLUG_CPU
732 
733 int __cpu_disable(void)
734 {
735 	struct ec_creg_mask_parms cr_parms;
736 	int cpu = smp_processor_id();
737 
738 	set_cpu_online(cpu, false);
739 
740 	/* Disable pfault pseudo page faults on this cpu. */
741 	pfault_fini();
742 
743 	memset(&cr_parms.orvals, 0, sizeof(cr_parms.orvals));
744 	memset(&cr_parms.andvals, 0xff, sizeof(cr_parms.andvals));
745 
746 	/* disable all external interrupts */
747 	cr_parms.orvals[0] = 0;
748 	cr_parms.andvals[0] = ~(1 << 15 | 1 << 14 | 1 << 13 | 1 << 11 |
749 				1 << 10 | 1 <<	9 | 1 <<  6 | 1 <<  5 |
750 				1 <<  4);
751 	/* disable all I/O interrupts */
752 	cr_parms.orvals[6] = 0;
753 	cr_parms.andvals[6] = ~(1 << 31 | 1 << 30 | 1 << 29 | 1 << 28 |
754 				1 << 27 | 1 << 26 | 1 << 25 | 1 << 24);
755 	/* disable most machine checks */
756 	cr_parms.orvals[14] = 0;
757 	cr_parms.andvals[14] = ~(1 << 28 | 1 << 27 | 1 << 26 |
758 				 1 << 25 | 1 << 24);
759 
760 	smp_ctl_bit_callback(&cr_parms);
761 
762 	return 0;
763 }
764 
765 void __cpu_die(unsigned int cpu)
766 {
767 	/* Wait until target cpu is down */
768 	while (!cpu_stopped(cpu))
769 		cpu_relax();
770 	while (sigp_p(0, cpu, sigp_set_prefix) == sigp_busy)
771 		udelay(10);
772 	smp_free_lowcore(cpu);
773 	atomic_dec(&init_mm.context.attach_count);
774 }
775 
776 void __noreturn cpu_die(void)
777 {
778 	idle_task_exit();
779 	while (sigp(smp_processor_id(), sigp_stop) == sigp_busy)
780 		cpu_relax();
781 	for (;;);
782 }
783 
784 #endif /* CONFIG_HOTPLUG_CPU */
785 
786 void __init smp_prepare_cpus(unsigned int max_cpus)
787 {
788 #ifndef CONFIG_64BIT
789 	unsigned long save_area = 0;
790 #endif
791 	unsigned long async_stack, panic_stack;
792 	struct _lowcore *lowcore;
793 
794 	smp_detect_cpus();
795 
796 	/* request the 0x1201 emergency signal external interrupt */
797 	if (register_external_interrupt(0x1201, do_ext_call_interrupt) != 0)
798 		panic("Couldn't request external interrupt 0x1201");
799 	/* request the 0x1202 external call external interrupt */
800 	if (register_external_interrupt(0x1202, do_ext_call_interrupt) != 0)
801 		panic("Couldn't request external interrupt 0x1202");
802 
803 	/* Reallocate current lowcore, but keep its contents. */
804 	lowcore = (void *) __get_free_pages(GFP_KERNEL | GFP_DMA, LC_ORDER);
805 	panic_stack = __get_free_page(GFP_KERNEL);
806 	async_stack = __get_free_pages(GFP_KERNEL, ASYNC_ORDER);
807 	BUG_ON(!lowcore || !panic_stack || !async_stack);
808 #ifndef CONFIG_64BIT
809 	if (MACHINE_HAS_IEEE)
810 		save_area = get_zeroed_page(GFP_KERNEL);
811 #endif
812 	local_irq_disable();
813 	local_mcck_disable();
814 	lowcore_ptr[smp_processor_id()] = lowcore;
815 	*lowcore = S390_lowcore;
816 	lowcore->panic_stack = panic_stack + PAGE_SIZE;
817 	lowcore->async_stack = async_stack + ASYNC_SIZE;
818 #ifndef CONFIG_64BIT
819 	if (MACHINE_HAS_IEEE)
820 		lowcore->extended_save_area_addr = (u32) save_area;
821 #endif
822 	set_prefix((u32)(unsigned long) lowcore);
823 	local_mcck_enable();
824 	local_irq_enable();
825 #ifdef CONFIG_64BIT
826 	if (vdso_alloc_per_cpu(smp_processor_id(), &S390_lowcore))
827 		BUG();
828 #endif
829 }
830 
831 void __init smp_prepare_boot_cpu(void)
832 {
833 	BUG_ON(smp_processor_id() != 0);
834 
835 	current_thread_info()->cpu = 0;
836 	set_cpu_present(0, true);
837 	set_cpu_online(0, true);
838 	S390_lowcore.percpu_offset = __per_cpu_offset[0];
839 	current_set[0] = current;
840 	smp_cpu_state[0] = CPU_STATE_CONFIGURED;
841 	cpu_set_polarization(0, POLARIZATION_UNKNOWN);
842 }
843 
844 void __init smp_cpus_done(unsigned int max_cpus)
845 {
846 }
847 
848 void __init smp_setup_processor_id(void)
849 {
850 	S390_lowcore.cpu_nr = 0;
851 	__cpu_logical_map[0] = stap();
852 }
853 
854 /*
855  * the frequency of the profiling timer can be changed
856  * by writing a multiplier value into /proc/profile.
857  *
858  * usually you want to run this on all CPUs ;)
859  */
860 int setup_profiling_timer(unsigned int multiplier)
861 {
862 	return 0;
863 }
864 
865 #ifdef CONFIG_HOTPLUG_CPU
866 static ssize_t cpu_configure_show(struct device *dev,
867 				struct device_attribute *attr, char *buf)
868 {
869 	ssize_t count;
870 
871 	mutex_lock(&smp_cpu_state_mutex);
872 	count = sprintf(buf, "%d\n", smp_cpu_state[dev->id]);
873 	mutex_unlock(&smp_cpu_state_mutex);
874 	return count;
875 }
876 
877 static ssize_t cpu_configure_store(struct device *dev,
878 				  struct device_attribute *attr,
879 				  const char *buf, size_t count)
880 {
881 	int cpu = dev->id;
882 	int val, rc;
883 	char delim;
884 
885 	if (sscanf(buf, "%d %c", &val, &delim) != 1)
886 		return -EINVAL;
887 	if (val != 0 && val != 1)
888 		return -EINVAL;
889 
890 	get_online_cpus();
891 	mutex_lock(&smp_cpu_state_mutex);
892 	rc = -EBUSY;
893 	/* disallow configuration changes of online cpus and cpu 0 */
894 	if (cpu_online(cpu) || cpu == 0)
895 		goto out;
896 	rc = 0;
897 	switch (val) {
898 	case 0:
899 		if (smp_cpu_state[cpu] == CPU_STATE_CONFIGURED) {
900 			rc = sclp_cpu_deconfigure(__cpu_logical_map[cpu]);
901 			if (!rc) {
902 				smp_cpu_state[cpu] = CPU_STATE_STANDBY;
903 				cpu_set_polarization(cpu, POLARIZATION_UNKNOWN);
904 				topology_expect_change();
905 			}
906 		}
907 		break;
908 	case 1:
909 		if (smp_cpu_state[cpu] == CPU_STATE_STANDBY) {
910 			rc = sclp_cpu_configure(__cpu_logical_map[cpu]);
911 			if (!rc) {
912 				smp_cpu_state[cpu] = CPU_STATE_CONFIGURED;
913 				cpu_set_polarization(cpu, POLARIZATION_UNKNOWN);
914 				topology_expect_change();
915 			}
916 		}
917 		break;
918 	default:
919 		break;
920 	}
921 out:
922 	mutex_unlock(&smp_cpu_state_mutex);
923 	put_online_cpus();
924 	return rc ? rc : count;
925 }
926 static DEVICE_ATTR(configure, 0644, cpu_configure_show, cpu_configure_store);
927 #endif /* CONFIG_HOTPLUG_CPU */
928 
929 static ssize_t show_cpu_address(struct device *dev,
930 				struct device_attribute *attr, char *buf)
931 {
932 	return sprintf(buf, "%d\n", __cpu_logical_map[dev->id]);
933 }
934 static DEVICE_ATTR(address, 0444, show_cpu_address, NULL);
935 
936 static struct attribute *cpu_common_attrs[] = {
937 #ifdef CONFIG_HOTPLUG_CPU
938 	&dev_attr_configure.attr,
939 #endif
940 	&dev_attr_address.attr,
941 	NULL,
942 };
943 
944 static struct attribute_group cpu_common_attr_group = {
945 	.attrs = cpu_common_attrs,
946 };
947 
948 static ssize_t show_capability(struct device *dev,
949 				struct device_attribute *attr, char *buf)
950 {
951 	unsigned int capability;
952 	int rc;
953 
954 	rc = get_cpu_capability(&capability);
955 	if (rc)
956 		return rc;
957 	return sprintf(buf, "%u\n", capability);
958 }
959 static DEVICE_ATTR(capability, 0444, show_capability, NULL);
960 
961 static ssize_t show_idle_count(struct device *dev,
962 				struct device_attribute *attr, char *buf)
963 {
964 	struct s390_idle_data *idle;
965 	unsigned long long idle_count;
966 	unsigned int sequence;
967 
968 	idle = &per_cpu(s390_idle, dev->id);
969 repeat:
970 	sequence = idle->sequence;
971 	smp_rmb();
972 	if (sequence & 1)
973 		goto repeat;
974 	idle_count = idle->idle_count;
975 	if (idle->idle_enter)
976 		idle_count++;
977 	smp_rmb();
978 	if (idle->sequence != sequence)
979 		goto repeat;
980 	return sprintf(buf, "%llu\n", idle_count);
981 }
982 static DEVICE_ATTR(idle_count, 0444, show_idle_count, NULL);
983 
984 static ssize_t show_idle_time(struct device *dev,
985 				struct device_attribute *attr, char *buf)
986 {
987 	struct s390_idle_data *idle;
988 	unsigned long long now, idle_time, idle_enter;
989 	unsigned int sequence;
990 
991 	idle = &per_cpu(s390_idle, dev->id);
992 	now = get_clock();
993 repeat:
994 	sequence = idle->sequence;
995 	smp_rmb();
996 	if (sequence & 1)
997 		goto repeat;
998 	idle_time = idle->idle_time;
999 	idle_enter = idle->idle_enter;
1000 	if (idle_enter != 0ULL && idle_enter < now)
1001 		idle_time += now - idle_enter;
1002 	smp_rmb();
1003 	if (idle->sequence != sequence)
1004 		goto repeat;
1005 	return sprintf(buf, "%llu\n", idle_time >> 12);
1006 }
1007 static DEVICE_ATTR(idle_time_us, 0444, show_idle_time, NULL);
1008 
1009 static struct attribute *cpu_online_attrs[] = {
1010 	&dev_attr_capability.attr,
1011 	&dev_attr_idle_count.attr,
1012 	&dev_attr_idle_time_us.attr,
1013 	NULL,
1014 };
1015 
1016 static struct attribute_group cpu_online_attr_group = {
1017 	.attrs = cpu_online_attrs,
1018 };
1019 
1020 static int __cpuinit smp_cpu_notify(struct notifier_block *self,
1021 				    unsigned long action, void *hcpu)
1022 {
1023 	unsigned int cpu = (unsigned int)(long)hcpu;
1024 	struct cpu *c = &per_cpu(cpu_devices, cpu);
1025 	struct device *s = &c->dev;
1026 	struct s390_idle_data *idle;
1027 	int err = 0;
1028 
1029 	switch (action) {
1030 	case CPU_ONLINE:
1031 	case CPU_ONLINE_FROZEN:
1032 		idle = &per_cpu(s390_idle, cpu);
1033 		memset(idle, 0, sizeof(struct s390_idle_data));
1034 		err = sysfs_create_group(&s->kobj, &cpu_online_attr_group);
1035 		break;
1036 	case CPU_DEAD:
1037 	case CPU_DEAD_FROZEN:
1038 		sysfs_remove_group(&s->kobj, &cpu_online_attr_group);
1039 		break;
1040 	}
1041 	return notifier_from_errno(err);
1042 }
1043 
1044 static struct notifier_block __cpuinitdata smp_cpu_nb = {
1045 	.notifier_call = smp_cpu_notify,
1046 };
1047 
1048 static int __devinit smp_add_present_cpu(int cpu)
1049 {
1050 	struct cpu *c = &per_cpu(cpu_devices, cpu);
1051 	struct device *s = &c->dev;
1052 	int rc;
1053 
1054 	c->hotpluggable = 1;
1055 	rc = register_cpu(c, cpu);
1056 	if (rc)
1057 		goto out;
1058 	rc = sysfs_create_group(&s->kobj, &cpu_common_attr_group);
1059 	if (rc)
1060 		goto out_cpu;
1061 	if (cpu_online(cpu)) {
1062 		rc = sysfs_create_group(&s->kobj, &cpu_online_attr_group);
1063 		if (rc)
1064 			goto out_online;
1065 	}
1066 	rc = topology_cpu_init(c);
1067 	if (rc)
1068 		goto out_topology;
1069 	return 0;
1070 
1071 out_topology:
1072 	if (cpu_online(cpu))
1073 		sysfs_remove_group(&s->kobj, &cpu_online_attr_group);
1074 out_online:
1075 	sysfs_remove_group(&s->kobj, &cpu_common_attr_group);
1076 out_cpu:
1077 #ifdef CONFIG_HOTPLUG_CPU
1078 	unregister_cpu(c);
1079 #endif
1080 out:
1081 	return rc;
1082 }
1083 
1084 #ifdef CONFIG_HOTPLUG_CPU
1085 
1086 int __ref smp_rescan_cpus(void)
1087 {
1088 	cpumask_t newcpus;
1089 	int cpu;
1090 	int rc;
1091 
1092 	get_online_cpus();
1093 	mutex_lock(&smp_cpu_state_mutex);
1094 	cpumask_copy(&newcpus, cpu_present_mask);
1095 	rc = __smp_rescan_cpus();
1096 	if (rc)
1097 		goto out;
1098 	cpumask_andnot(&newcpus, cpu_present_mask, &newcpus);
1099 	for_each_cpu(cpu, &newcpus) {
1100 		rc = smp_add_present_cpu(cpu);
1101 		if (rc)
1102 			set_cpu_present(cpu, false);
1103 	}
1104 	rc = 0;
1105 out:
1106 	mutex_unlock(&smp_cpu_state_mutex);
1107 	put_online_cpus();
1108 	if (!cpumask_empty(&newcpus))
1109 		topology_schedule_update();
1110 	return rc;
1111 }
1112 
1113 static ssize_t __ref rescan_store(struct device *dev,
1114 				  struct device_attribute *attr,
1115 				  const char *buf,
1116 				  size_t count)
1117 {
1118 	int rc;
1119 
1120 	rc = smp_rescan_cpus();
1121 	return rc ? rc : count;
1122 }
1123 static DEVICE_ATTR(rescan, 0200, NULL, rescan_store);
1124 #endif /* CONFIG_HOTPLUG_CPU */
1125 
1126 static int __init s390_smp_init(void)
1127 {
1128 	int cpu, rc;
1129 
1130 	register_cpu_notifier(&smp_cpu_nb);
1131 #ifdef CONFIG_HOTPLUG_CPU
1132 	rc = device_create_file(cpu_subsys.dev_root, &dev_attr_rescan);
1133 	if (rc)
1134 		return rc;
1135 #endif
1136 	for_each_present_cpu(cpu) {
1137 		rc = smp_add_present_cpu(cpu);
1138 		if (rc)
1139 			return rc;
1140 	}
1141 	return 0;
1142 }
1143 subsys_initcall(s390_smp_init);
1144