xref: /linux/arch/s390/kernel/smp.c (revision 5499b45190237ca90dd2ac86395cf464fe1f4cc7)
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/module.h>
27 #include <linux/init.h>
28 #include <linux/mm.h>
29 #include <linux/err.h>
30 #include <linux/spinlock.h>
31 #include <linux/kernel_stat.h>
32 #include <linux/delay.h>
33 #include <linux/cache.h>
34 #include <linux/interrupt.h>
35 #include <linux/irqflags.h>
36 #include <linux/cpu.h>
37 #include <linux/timex.h>
38 #include <linux/bootmem.h>
39 #include <asm/asm-offsets.h>
40 #include <asm/ipl.h>
41 #include <asm/setup.h>
42 #include <asm/sigp.h>
43 #include <asm/pgalloc.h>
44 #include <asm/irq.h>
45 #include <asm/s390_ext.h>
46 #include <asm/cpcmd.h>
47 #include <asm/tlbflush.h>
48 #include <asm/timer.h>
49 #include <asm/lowcore.h>
50 #include <asm/sclp.h>
51 #include <asm/cputime.h>
52 #include <asm/vdso.h>
53 #include <asm/cpu.h>
54 #include "entry.h"
55 
56 /* logical cpu to cpu address */
57 unsigned short __cpu_logical_map[NR_CPUS];
58 
59 static struct task_struct *current_set[NR_CPUS];
60 
61 static u8 smp_cpu_type;
62 static int smp_use_sigp_detection;
63 
64 enum s390_cpu_state {
65 	CPU_STATE_STANDBY,
66 	CPU_STATE_CONFIGURED,
67 };
68 
69 DEFINE_MUTEX(smp_cpu_state_mutex);
70 int smp_cpu_polarization[NR_CPUS];
71 static int smp_cpu_state[NR_CPUS];
72 static int cpu_management;
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 void smp_switch_to_ipl_cpu(void (*func)(void *), void *data)
100 {
101 	struct _lowcore *lc, *current_lc;
102 	struct stack_frame *sf;
103 	struct pt_regs *regs;
104 	unsigned long sp;
105 
106 	if (smp_processor_id() == 0)
107 		func(data);
108 	__load_psw_mask(PSW_BASE_BITS | PSW_DEFAULT_KEY);
109 	/* Disable lowcore protection */
110 	__ctl_clear_bit(0, 28);
111 	current_lc = lowcore_ptr[smp_processor_id()];
112 	lc = lowcore_ptr[0];
113 	if (!lc)
114 		lc = current_lc;
115 	lc->restart_psw.mask = PSW_BASE_BITS | PSW_DEFAULT_KEY;
116 	lc->restart_psw.addr = PSW_ADDR_AMODE | (unsigned long) smp_restart_cpu;
117 	if (!cpu_online(0))
118 		smp_switch_to_cpu(func, data, 0, stap(), __cpu_logical_map[0]);
119 	while (sigp(0, sigp_stop_and_store_status) == sigp_busy)
120 		cpu_relax();
121 	sp = lc->panic_stack;
122 	sp -= sizeof(struct pt_regs);
123 	regs = (struct pt_regs *) sp;
124 	memcpy(&regs->gprs, &current_lc->gpregs_save_area, sizeof(regs->gprs));
125 	regs->psw = lc->psw_save_area;
126 	sp -= STACK_FRAME_OVERHEAD;
127 	sf = (struct stack_frame *) sp;
128 	sf->back_chain = regs->gprs[15];
129 	smp_switch_to_cpu(func, data, sp, stap(), __cpu_logical_map[0]);
130 }
131 
132 void smp_send_stop(void)
133 {
134 	int cpu, rc;
135 
136 	/* Disable all interrupts/machine checks */
137 	__load_psw_mask(psw_kernel_bits & ~PSW_MASK_MCHECK);
138 	trace_hardirqs_off();
139 
140 	/* stop all processors */
141 	for_each_online_cpu(cpu) {
142 		if (cpu == smp_processor_id())
143 			continue;
144 		do {
145 			rc = sigp(cpu, sigp_stop);
146 		} while (rc == sigp_busy);
147 
148 		while (!cpu_stopped(cpu))
149 			cpu_relax();
150 	}
151 }
152 
153 /*
154  * This is the main routine where commands issued by other
155  * cpus are handled.
156  */
157 
158 static void do_ext_call_interrupt(__u16 code)
159 {
160 	unsigned long bits;
161 
162 	/*
163 	 * handle bit signal external calls
164 	 *
165 	 * For the ec_schedule signal we have to do nothing. All the work
166 	 * is done automatically when we return from the interrupt.
167 	 */
168 	bits = xchg(&S390_lowcore.ext_call_fast, 0);
169 
170 	if (test_bit(ec_call_function, &bits))
171 		generic_smp_call_function_interrupt();
172 
173 	if (test_bit(ec_call_function_single, &bits))
174 		generic_smp_call_function_single_interrupt();
175 }
176 
177 /*
178  * Send an external call sigp to another cpu and return without waiting
179  * for its completion.
180  */
181 static void smp_ext_bitcall(int cpu, int sig)
182 {
183 	/*
184 	 * Set signaling bit in lowcore of target cpu and kick it
185 	 */
186 	set_bit(sig, (unsigned long *) &lowcore_ptr[cpu]->ext_call_fast);
187 	while (sigp(cpu, sigp_emergency_signal) == sigp_busy)
188 		udelay(10);
189 }
190 
191 void arch_send_call_function_ipi_mask(const struct cpumask *mask)
192 {
193 	int cpu;
194 
195 	for_each_cpu(cpu, mask)
196 		smp_ext_bitcall(cpu, ec_call_function);
197 }
198 
199 void arch_send_call_function_single_ipi(int cpu)
200 {
201 	smp_ext_bitcall(cpu, ec_call_function_single);
202 }
203 
204 #ifndef CONFIG_64BIT
205 /*
206  * this function sends a 'purge tlb' signal to another CPU.
207  */
208 static void smp_ptlb_callback(void *info)
209 {
210 	__tlb_flush_local();
211 }
212 
213 void smp_ptlb_all(void)
214 {
215 	on_each_cpu(smp_ptlb_callback, NULL, 1);
216 }
217 EXPORT_SYMBOL(smp_ptlb_all);
218 #endif /* ! CONFIG_64BIT */
219 
220 /*
221  * this function sends a 'reschedule' IPI to another CPU.
222  * it goes straight through and wastes no time serializing
223  * anything. Worst case is that we lose a reschedule ...
224  */
225 void smp_send_reschedule(int cpu)
226 {
227 	smp_ext_bitcall(cpu, ec_schedule);
228 }
229 
230 /*
231  * parameter area for the set/clear control bit callbacks
232  */
233 struct ec_creg_mask_parms {
234 	unsigned long orvals[16];
235 	unsigned long andvals[16];
236 };
237 
238 /*
239  * callback for setting/clearing control bits
240  */
241 static void smp_ctl_bit_callback(void *info)
242 {
243 	struct ec_creg_mask_parms *pp = info;
244 	unsigned long cregs[16];
245 	int i;
246 
247 	__ctl_store(cregs, 0, 15);
248 	for (i = 0; i <= 15; i++)
249 		cregs[i] = (cregs[i] & pp->andvals[i]) | pp->orvals[i];
250 	__ctl_load(cregs, 0, 15);
251 }
252 
253 /*
254  * Set a bit in a control register of all cpus
255  */
256 void smp_ctl_set_bit(int cr, int bit)
257 {
258 	struct ec_creg_mask_parms parms;
259 
260 	memset(&parms.orvals, 0, sizeof(parms.orvals));
261 	memset(&parms.andvals, 0xff, sizeof(parms.andvals));
262 	parms.orvals[cr] = 1 << bit;
263 	on_each_cpu(smp_ctl_bit_callback, &parms, 1);
264 }
265 EXPORT_SYMBOL(smp_ctl_set_bit);
266 
267 /*
268  * Clear a bit in a control register of all cpus
269  */
270 void smp_ctl_clear_bit(int cr, int bit)
271 {
272 	struct ec_creg_mask_parms parms;
273 
274 	memset(&parms.orvals, 0, sizeof(parms.orvals));
275 	memset(&parms.andvals, 0xff, sizeof(parms.andvals));
276 	parms.andvals[cr] = ~(1L << bit);
277 	on_each_cpu(smp_ctl_bit_callback, &parms, 1);
278 }
279 EXPORT_SYMBOL(smp_ctl_clear_bit);
280 
281 #ifdef CONFIG_ZFCPDUMP
282 
283 static void __init smp_get_save_area(unsigned int cpu, unsigned int phy_cpu)
284 {
285 	if (ipl_info.type != IPL_TYPE_FCP_DUMP)
286 		return;
287 	if (cpu >= NR_CPUS) {
288 		pr_warning("CPU %i exceeds the maximum %i and is excluded from "
289 			   "the dump\n", cpu, NR_CPUS - 1);
290 		return;
291 	}
292 	zfcpdump_save_areas[cpu] = kmalloc(sizeof(struct save_area), GFP_KERNEL);
293 	while (raw_sigp(phy_cpu, sigp_stop_and_store_status) == sigp_busy)
294 		cpu_relax();
295 	memcpy(zfcpdump_save_areas[cpu],
296 	       (void *)(unsigned long) store_prefix() + SAVE_AREA_BASE,
297 	       sizeof(struct save_area));
298 }
299 
300 struct save_area *zfcpdump_save_areas[NR_CPUS + 1];
301 EXPORT_SYMBOL_GPL(zfcpdump_save_areas);
302 
303 #else
304 
305 static inline void smp_get_save_area(unsigned int cpu, unsigned int phy_cpu) { }
306 
307 #endif /* CONFIG_ZFCPDUMP */
308 
309 static int cpu_known(int cpu_id)
310 {
311 	int cpu;
312 
313 	for_each_present_cpu(cpu) {
314 		if (__cpu_logical_map[cpu] == cpu_id)
315 			return 1;
316 	}
317 	return 0;
318 }
319 
320 static int smp_rescan_cpus_sigp(cpumask_t avail)
321 {
322 	int cpu_id, logical_cpu;
323 
324 	logical_cpu = cpumask_first(&avail);
325 	if (logical_cpu >= nr_cpu_ids)
326 		return 0;
327 	for (cpu_id = 0; cpu_id <= MAX_CPU_ADDRESS; cpu_id++) {
328 		if (cpu_known(cpu_id))
329 			continue;
330 		__cpu_logical_map[logical_cpu] = cpu_id;
331 		smp_cpu_polarization[logical_cpu] = POLARIZATION_UNKNWN;
332 		if (!cpu_stopped(logical_cpu))
333 			continue;
334 		cpu_set(logical_cpu, cpu_present_map);
335 		smp_cpu_state[logical_cpu] = CPU_STATE_CONFIGURED;
336 		logical_cpu = cpumask_next(logical_cpu, &avail);
337 		if (logical_cpu >= nr_cpu_ids)
338 			break;
339 	}
340 	return 0;
341 }
342 
343 static int smp_rescan_cpus_sclp(cpumask_t avail)
344 {
345 	struct sclp_cpu_info *info;
346 	int cpu_id, logical_cpu, cpu;
347 	int rc;
348 
349 	logical_cpu = cpumask_first(&avail);
350 	if (logical_cpu >= nr_cpu_ids)
351 		return 0;
352 	info = kmalloc(sizeof(*info), GFP_KERNEL);
353 	if (!info)
354 		return -ENOMEM;
355 	rc = sclp_get_cpu_info(info);
356 	if (rc)
357 		goto out;
358 	for (cpu = 0; cpu < info->combined; cpu++) {
359 		if (info->has_cpu_type && info->cpu[cpu].type != smp_cpu_type)
360 			continue;
361 		cpu_id = info->cpu[cpu].address;
362 		if (cpu_known(cpu_id))
363 			continue;
364 		__cpu_logical_map[logical_cpu] = cpu_id;
365 		smp_cpu_polarization[logical_cpu] = POLARIZATION_UNKNWN;
366 		cpu_set(logical_cpu, cpu_present_map);
367 		if (cpu >= info->configured)
368 			smp_cpu_state[logical_cpu] = CPU_STATE_STANDBY;
369 		else
370 			smp_cpu_state[logical_cpu] = CPU_STATE_CONFIGURED;
371 		logical_cpu = cpumask_next(logical_cpu, &avail);
372 		if (logical_cpu >= nr_cpu_ids)
373 			break;
374 	}
375 out:
376 	kfree(info);
377 	return rc;
378 }
379 
380 static int __smp_rescan_cpus(void)
381 {
382 	cpumask_t avail;
383 
384 	cpus_xor(avail, cpu_possible_map, cpu_present_map);
385 	if (smp_use_sigp_detection)
386 		return smp_rescan_cpus_sigp(avail);
387 	else
388 		return smp_rescan_cpus_sclp(avail);
389 }
390 
391 static void __init smp_detect_cpus(void)
392 {
393 	unsigned int cpu, c_cpus, s_cpus;
394 	struct sclp_cpu_info *info;
395 	u16 boot_cpu_addr, cpu_addr;
396 
397 	c_cpus = 1;
398 	s_cpus = 0;
399 	boot_cpu_addr = __cpu_logical_map[0];
400 	info = kmalloc(sizeof(*info), GFP_KERNEL);
401 	if (!info)
402 		panic("smp_detect_cpus failed to allocate memory\n");
403 	/* Use sigp detection algorithm if sclp doesn't work. */
404 	if (sclp_get_cpu_info(info)) {
405 		smp_use_sigp_detection = 1;
406 		for (cpu = 0; cpu <= MAX_CPU_ADDRESS; cpu++) {
407 			if (cpu == boot_cpu_addr)
408 				continue;
409 			if (!raw_cpu_stopped(cpu))
410 				continue;
411 			smp_get_save_area(c_cpus, cpu);
412 			c_cpus++;
413 		}
414 		goto out;
415 	}
416 
417 	if (info->has_cpu_type) {
418 		for (cpu = 0; cpu < info->combined; cpu++) {
419 			if (info->cpu[cpu].address == boot_cpu_addr) {
420 				smp_cpu_type = info->cpu[cpu].type;
421 				break;
422 			}
423 		}
424 	}
425 
426 	for (cpu = 0; cpu < info->combined; cpu++) {
427 		if (info->has_cpu_type && info->cpu[cpu].type != smp_cpu_type)
428 			continue;
429 		cpu_addr = info->cpu[cpu].address;
430 		if (cpu_addr == boot_cpu_addr)
431 			continue;
432 		if (!raw_cpu_stopped(cpu_addr)) {
433 			s_cpus++;
434 			continue;
435 		}
436 		smp_get_save_area(c_cpus, cpu_addr);
437 		c_cpus++;
438 	}
439 out:
440 	kfree(info);
441 	pr_info("%d configured CPUs, %d standby CPUs\n", c_cpus, s_cpus);
442 	get_online_cpus();
443 	__smp_rescan_cpus();
444 	put_online_cpus();
445 }
446 
447 /*
448  *	Activate a secondary processor.
449  */
450 int __cpuinit start_secondary(void *cpuvoid)
451 {
452 	/* Setup the cpu */
453 	cpu_init();
454 	preempt_disable();
455 	/* Enable TOD clock interrupts on the secondary cpu. */
456 	init_cpu_timer();
457 	/* Enable cpu timer interrupts on the secondary cpu. */
458 	init_cpu_vtimer();
459 	/* Enable pfault pseudo page faults on this cpu. */
460 	pfault_init();
461 
462 	/* call cpu notifiers */
463 	notify_cpu_starting(smp_processor_id());
464 	/* Mark this cpu as online */
465 	ipi_call_lock();
466 	cpu_set(smp_processor_id(), cpu_online_map);
467 	ipi_call_unlock();
468 	/* Switch on interrupts */
469 	local_irq_enable();
470 	/* Print info about this processor */
471 	print_cpu_info();
472 	/* cpu_idle will call schedule for us */
473 	cpu_idle();
474 	return 0;
475 }
476 
477 static void __init smp_create_idle(unsigned int cpu)
478 {
479 	struct task_struct *p;
480 
481 	/*
482 	 *  don't care about the psw and regs settings since we'll never
483 	 *  reschedule the forked task.
484 	 */
485 	p = fork_idle(cpu);
486 	if (IS_ERR(p))
487 		panic("failed fork for CPU %u: %li", cpu, PTR_ERR(p));
488 	current_set[cpu] = p;
489 }
490 
491 static int __cpuinit smp_alloc_lowcore(int cpu)
492 {
493 	unsigned long async_stack, panic_stack;
494 	struct _lowcore *lowcore;
495 
496 	lowcore = (void *) __get_free_pages(GFP_KERNEL | GFP_DMA, LC_ORDER);
497 	if (!lowcore)
498 		return -ENOMEM;
499 	async_stack = __get_free_pages(GFP_KERNEL, ASYNC_ORDER);
500 	panic_stack = __get_free_page(GFP_KERNEL);
501 	if (!panic_stack || !async_stack)
502 		goto out;
503 	memcpy(lowcore, &S390_lowcore, 512);
504 	memset((char *)lowcore + 512, 0, sizeof(*lowcore) - 512);
505 	lowcore->async_stack = async_stack + ASYNC_SIZE;
506 	lowcore->panic_stack = panic_stack + PAGE_SIZE;
507 
508 #ifndef CONFIG_64BIT
509 	if (MACHINE_HAS_IEEE) {
510 		unsigned long save_area;
511 
512 		save_area = get_zeroed_page(GFP_KERNEL);
513 		if (!save_area)
514 			goto out;
515 		lowcore->extended_save_area_addr = (u32) save_area;
516 	}
517 #else
518 	if (vdso_alloc_per_cpu(cpu, lowcore))
519 		goto out;
520 #endif
521 	lowcore_ptr[cpu] = lowcore;
522 	return 0;
523 
524 out:
525 	free_page(panic_stack);
526 	free_pages(async_stack, ASYNC_ORDER);
527 	free_pages((unsigned long) lowcore, LC_ORDER);
528 	return -ENOMEM;
529 }
530 
531 static void smp_free_lowcore(int cpu)
532 {
533 	struct _lowcore *lowcore;
534 
535 	lowcore = lowcore_ptr[cpu];
536 #ifndef CONFIG_64BIT
537 	if (MACHINE_HAS_IEEE)
538 		free_page((unsigned long) lowcore->extended_save_area_addr);
539 #else
540 	vdso_free_per_cpu(cpu, lowcore);
541 #endif
542 	free_page(lowcore->panic_stack - PAGE_SIZE);
543 	free_pages(lowcore->async_stack - ASYNC_SIZE, ASYNC_ORDER);
544 	free_pages((unsigned long) lowcore, LC_ORDER);
545 	lowcore_ptr[cpu] = NULL;
546 }
547 
548 /* Upping and downing of CPUs */
549 int __cpuinit __cpu_up(unsigned int cpu)
550 {
551 	struct _lowcore *cpu_lowcore;
552 	struct task_struct *idle;
553 	struct stack_frame *sf;
554 	u32 lowcore;
555 	int ccode;
556 
557 	if (smp_cpu_state[cpu] != CPU_STATE_CONFIGURED)
558 		return -EIO;
559 	if (smp_alloc_lowcore(cpu))
560 		return -ENOMEM;
561 	do {
562 		ccode = sigp(cpu, sigp_initial_cpu_reset);
563 		if (ccode == sigp_busy)
564 			udelay(10);
565 		if (ccode == sigp_not_operational)
566 			goto err_out;
567 	} while (ccode == sigp_busy);
568 
569 	lowcore = (u32)(unsigned long)lowcore_ptr[cpu];
570 	while (sigp_p(lowcore, cpu, sigp_set_prefix) == sigp_busy)
571 		udelay(10);
572 
573 	idle = current_set[cpu];
574 	cpu_lowcore = lowcore_ptr[cpu];
575 	cpu_lowcore->kernel_stack = (unsigned long)
576 		task_stack_page(idle) + THREAD_SIZE;
577 	cpu_lowcore->thread_info = (unsigned long) task_thread_info(idle);
578 	sf = (struct stack_frame *) (cpu_lowcore->kernel_stack
579 				     - sizeof(struct pt_regs)
580 				     - sizeof(struct stack_frame));
581 	memset(sf, 0, sizeof(struct stack_frame));
582 	sf->gprs[9] = (unsigned long) sf;
583 	cpu_lowcore->save_area[15] = (unsigned long) sf;
584 	__ctl_store(cpu_lowcore->cregs_save_area, 0, 15);
585 	asm volatile(
586 		"	stam	0,15,0(%0)"
587 		: : "a" (&cpu_lowcore->access_regs_save_area) : "memory");
588 	cpu_lowcore->percpu_offset = __per_cpu_offset[cpu];
589 	cpu_lowcore->current_task = (unsigned long) idle;
590 	cpu_lowcore->cpu_nr = cpu;
591 	cpu_lowcore->kernel_asce = S390_lowcore.kernel_asce;
592 	cpu_lowcore->machine_flags = S390_lowcore.machine_flags;
593 	cpu_lowcore->ftrace_func = S390_lowcore.ftrace_func;
594 	eieio();
595 
596 	while (sigp(cpu, sigp_restart) == sigp_busy)
597 		udelay(10);
598 
599 	while (!cpu_online(cpu))
600 		cpu_relax();
601 	return 0;
602 
603 err_out:
604 	smp_free_lowcore(cpu);
605 	return -EIO;
606 }
607 
608 static int __init setup_possible_cpus(char *s)
609 {
610 	int pcpus, cpu;
611 
612 	pcpus = simple_strtoul(s, NULL, 0);
613 	init_cpu_possible(cpumask_of(0));
614 	for (cpu = 1; cpu < pcpus && cpu < nr_cpu_ids; cpu++)
615 		set_cpu_possible(cpu, true);
616 	return 0;
617 }
618 early_param("possible_cpus", setup_possible_cpus);
619 
620 #ifdef CONFIG_HOTPLUG_CPU
621 
622 int __cpu_disable(void)
623 {
624 	struct ec_creg_mask_parms cr_parms;
625 	int cpu = smp_processor_id();
626 
627 	cpu_clear(cpu, cpu_online_map);
628 
629 	/* Disable pfault pseudo page faults on this cpu. */
630 	pfault_fini();
631 
632 	memset(&cr_parms.orvals, 0, sizeof(cr_parms.orvals));
633 	memset(&cr_parms.andvals, 0xff, sizeof(cr_parms.andvals));
634 
635 	/* disable all external interrupts */
636 	cr_parms.orvals[0] = 0;
637 	cr_parms.andvals[0] = ~(1 << 15 | 1 << 14 | 1 << 13 | 1 << 12 |
638 				1 << 11 | 1 << 10 | 1 <<  6 | 1 <<  4);
639 	/* disable all I/O interrupts */
640 	cr_parms.orvals[6] = 0;
641 	cr_parms.andvals[6] = ~(1 << 31 | 1 << 30 | 1 << 29 | 1 << 28 |
642 				1 << 27 | 1 << 26 | 1 << 25 | 1 << 24);
643 	/* disable most machine checks */
644 	cr_parms.orvals[14] = 0;
645 	cr_parms.andvals[14] = ~(1 << 28 | 1 << 27 | 1 << 26 |
646 				 1 << 25 | 1 << 24);
647 
648 	smp_ctl_bit_callback(&cr_parms);
649 
650 	return 0;
651 }
652 
653 void __cpu_die(unsigned int cpu)
654 {
655 	/* Wait until target cpu is down */
656 	while (!cpu_stopped(cpu))
657 		cpu_relax();
658 	while (sigp_p(0, cpu, sigp_set_prefix) == sigp_busy)
659 		udelay(10);
660 	smp_free_lowcore(cpu);
661 	pr_info("Processor %d stopped\n", cpu);
662 }
663 
664 void cpu_die(void)
665 {
666 	idle_task_exit();
667 	while (sigp(smp_processor_id(), sigp_stop) == sigp_busy)
668 		cpu_relax();
669 	for (;;);
670 }
671 
672 #endif /* CONFIG_HOTPLUG_CPU */
673 
674 void __init smp_prepare_cpus(unsigned int max_cpus)
675 {
676 #ifndef CONFIG_64BIT
677 	unsigned long save_area = 0;
678 #endif
679 	unsigned long async_stack, panic_stack;
680 	struct _lowcore *lowcore;
681 	unsigned int cpu;
682 
683 	smp_detect_cpus();
684 
685 	/* request the 0x1201 emergency signal external interrupt */
686 	if (register_external_interrupt(0x1201, do_ext_call_interrupt) != 0)
687 		panic("Couldn't request external interrupt 0x1201");
688 	print_cpu_info();
689 
690 	/* Reallocate current lowcore, but keep its contents. */
691 	lowcore = (void *) __get_free_pages(GFP_KERNEL | GFP_DMA, LC_ORDER);
692 	panic_stack = __get_free_page(GFP_KERNEL);
693 	async_stack = __get_free_pages(GFP_KERNEL, ASYNC_ORDER);
694 	BUG_ON(!lowcore || !panic_stack || !async_stack);
695 #ifndef CONFIG_64BIT
696 	if (MACHINE_HAS_IEEE)
697 		save_area = get_zeroed_page(GFP_KERNEL);
698 #endif
699 	local_irq_disable();
700 	local_mcck_disable();
701 	lowcore_ptr[smp_processor_id()] = lowcore;
702 	*lowcore = S390_lowcore;
703 	lowcore->panic_stack = panic_stack + PAGE_SIZE;
704 	lowcore->async_stack = async_stack + ASYNC_SIZE;
705 #ifndef CONFIG_64BIT
706 	if (MACHINE_HAS_IEEE)
707 		lowcore->extended_save_area_addr = (u32) save_area;
708 #endif
709 	set_prefix((u32)(unsigned long) lowcore);
710 	local_mcck_enable();
711 	local_irq_enable();
712 #ifdef CONFIG_64BIT
713 	if (vdso_alloc_per_cpu(smp_processor_id(), &S390_lowcore))
714 		BUG();
715 #endif
716 	for_each_possible_cpu(cpu)
717 		if (cpu != smp_processor_id())
718 			smp_create_idle(cpu);
719 }
720 
721 void __init smp_prepare_boot_cpu(void)
722 {
723 	BUG_ON(smp_processor_id() != 0);
724 
725 	current_thread_info()->cpu = 0;
726 	cpu_set(0, cpu_present_map);
727 	cpu_set(0, cpu_online_map);
728 	S390_lowcore.percpu_offset = __per_cpu_offset[0];
729 	current_set[0] = current;
730 	smp_cpu_state[0] = CPU_STATE_CONFIGURED;
731 	smp_cpu_polarization[0] = POLARIZATION_UNKNWN;
732 }
733 
734 void __init smp_cpus_done(unsigned int max_cpus)
735 {
736 }
737 
738 void __init smp_setup_processor_id(void)
739 {
740 	S390_lowcore.cpu_nr = 0;
741 	__cpu_logical_map[0] = stap();
742 }
743 
744 /*
745  * the frequency of the profiling timer can be changed
746  * by writing a multiplier value into /proc/profile.
747  *
748  * usually you want to run this on all CPUs ;)
749  */
750 int setup_profiling_timer(unsigned int multiplier)
751 {
752 	return 0;
753 }
754 
755 #ifdef CONFIG_HOTPLUG_CPU
756 static ssize_t cpu_configure_show(struct sys_device *dev,
757 				struct sysdev_attribute *attr, char *buf)
758 {
759 	ssize_t count;
760 
761 	mutex_lock(&smp_cpu_state_mutex);
762 	count = sprintf(buf, "%d\n", smp_cpu_state[dev->id]);
763 	mutex_unlock(&smp_cpu_state_mutex);
764 	return count;
765 }
766 
767 static ssize_t cpu_configure_store(struct sys_device *dev,
768 				  struct sysdev_attribute *attr,
769 				  const char *buf, size_t count)
770 {
771 	int cpu = dev->id;
772 	int val, rc;
773 	char delim;
774 
775 	if (sscanf(buf, "%d %c", &val, &delim) != 1)
776 		return -EINVAL;
777 	if (val != 0 && val != 1)
778 		return -EINVAL;
779 
780 	get_online_cpus();
781 	mutex_lock(&smp_cpu_state_mutex);
782 	rc = -EBUSY;
783 	/* disallow configuration changes of online cpus and cpu 0 */
784 	if (cpu_online(cpu) || cpu == 0)
785 		goto out;
786 	rc = 0;
787 	switch (val) {
788 	case 0:
789 		if (smp_cpu_state[cpu] == CPU_STATE_CONFIGURED) {
790 			rc = sclp_cpu_deconfigure(__cpu_logical_map[cpu]);
791 			if (!rc) {
792 				smp_cpu_state[cpu] = CPU_STATE_STANDBY;
793 				smp_cpu_polarization[cpu] = POLARIZATION_UNKNWN;
794 			}
795 		}
796 		break;
797 	case 1:
798 		if (smp_cpu_state[cpu] == CPU_STATE_STANDBY) {
799 			rc = sclp_cpu_configure(__cpu_logical_map[cpu]);
800 			if (!rc) {
801 				smp_cpu_state[cpu] = CPU_STATE_CONFIGURED;
802 				smp_cpu_polarization[cpu] = POLARIZATION_UNKNWN;
803 			}
804 		}
805 		break;
806 	default:
807 		break;
808 	}
809 out:
810 	mutex_unlock(&smp_cpu_state_mutex);
811 	put_online_cpus();
812 	return rc ? rc : count;
813 }
814 static SYSDEV_ATTR(configure, 0644, cpu_configure_show, cpu_configure_store);
815 #endif /* CONFIG_HOTPLUG_CPU */
816 
817 static ssize_t cpu_polarization_show(struct sys_device *dev,
818 				     struct sysdev_attribute *attr, char *buf)
819 {
820 	int cpu = dev->id;
821 	ssize_t count;
822 
823 	mutex_lock(&smp_cpu_state_mutex);
824 	switch (smp_cpu_polarization[cpu]) {
825 	case POLARIZATION_HRZ:
826 		count = sprintf(buf, "horizontal\n");
827 		break;
828 	case POLARIZATION_VL:
829 		count = sprintf(buf, "vertical:low\n");
830 		break;
831 	case POLARIZATION_VM:
832 		count = sprintf(buf, "vertical:medium\n");
833 		break;
834 	case POLARIZATION_VH:
835 		count = sprintf(buf, "vertical:high\n");
836 		break;
837 	default:
838 		count = sprintf(buf, "unknown\n");
839 		break;
840 	}
841 	mutex_unlock(&smp_cpu_state_mutex);
842 	return count;
843 }
844 static SYSDEV_ATTR(polarization, 0444, cpu_polarization_show, NULL);
845 
846 static ssize_t show_cpu_address(struct sys_device *dev,
847 				struct sysdev_attribute *attr, char *buf)
848 {
849 	return sprintf(buf, "%d\n", __cpu_logical_map[dev->id]);
850 }
851 static SYSDEV_ATTR(address, 0444, show_cpu_address, NULL);
852 
853 
854 static struct attribute *cpu_common_attrs[] = {
855 #ifdef CONFIG_HOTPLUG_CPU
856 	&attr_configure.attr,
857 #endif
858 	&attr_address.attr,
859 	&attr_polarization.attr,
860 	NULL,
861 };
862 
863 static struct attribute_group cpu_common_attr_group = {
864 	.attrs = cpu_common_attrs,
865 };
866 
867 static ssize_t show_capability(struct sys_device *dev,
868 				struct sysdev_attribute *attr, char *buf)
869 {
870 	unsigned int capability;
871 	int rc;
872 
873 	rc = get_cpu_capability(&capability);
874 	if (rc)
875 		return rc;
876 	return sprintf(buf, "%u\n", capability);
877 }
878 static SYSDEV_ATTR(capability, 0444, show_capability, NULL);
879 
880 static ssize_t show_idle_count(struct sys_device *dev,
881 				struct sysdev_attribute *attr, char *buf)
882 {
883 	struct s390_idle_data *idle;
884 	unsigned long long idle_count;
885 	unsigned int sequence;
886 
887 	idle = &per_cpu(s390_idle, dev->id);
888 repeat:
889 	sequence = idle->sequence;
890 	smp_rmb();
891 	if (sequence & 1)
892 		goto repeat;
893 	idle_count = idle->idle_count;
894 	if (idle->idle_enter)
895 		idle_count++;
896 	smp_rmb();
897 	if (idle->sequence != sequence)
898 		goto repeat;
899 	return sprintf(buf, "%llu\n", idle_count);
900 }
901 static SYSDEV_ATTR(idle_count, 0444, show_idle_count, NULL);
902 
903 static ssize_t show_idle_time(struct sys_device *dev,
904 				struct sysdev_attribute *attr, char *buf)
905 {
906 	struct s390_idle_data *idle;
907 	unsigned long long now, idle_time, idle_enter;
908 	unsigned int sequence;
909 
910 	idle = &per_cpu(s390_idle, dev->id);
911 	now = get_clock();
912 repeat:
913 	sequence = idle->sequence;
914 	smp_rmb();
915 	if (sequence & 1)
916 		goto repeat;
917 	idle_time = idle->idle_time;
918 	idle_enter = idle->idle_enter;
919 	if (idle_enter != 0ULL && idle_enter < now)
920 		idle_time += now - idle_enter;
921 	smp_rmb();
922 	if (idle->sequence != sequence)
923 		goto repeat;
924 	return sprintf(buf, "%llu\n", idle_time >> 12);
925 }
926 static SYSDEV_ATTR(idle_time_us, 0444, show_idle_time, NULL);
927 
928 static struct attribute *cpu_online_attrs[] = {
929 	&attr_capability.attr,
930 	&attr_idle_count.attr,
931 	&attr_idle_time_us.attr,
932 	NULL,
933 };
934 
935 static struct attribute_group cpu_online_attr_group = {
936 	.attrs = cpu_online_attrs,
937 };
938 
939 static int __cpuinit smp_cpu_notify(struct notifier_block *self,
940 				    unsigned long action, void *hcpu)
941 {
942 	unsigned int cpu = (unsigned int)(long)hcpu;
943 	struct cpu *c = &per_cpu(cpu_devices, cpu);
944 	struct sys_device *s = &c->sysdev;
945 	struct s390_idle_data *idle;
946 
947 	switch (action) {
948 	case CPU_ONLINE:
949 	case CPU_ONLINE_FROZEN:
950 		idle = &per_cpu(s390_idle, cpu);
951 		memset(idle, 0, sizeof(struct s390_idle_data));
952 		if (sysfs_create_group(&s->kobj, &cpu_online_attr_group))
953 			return NOTIFY_BAD;
954 		break;
955 	case CPU_DEAD:
956 	case CPU_DEAD_FROZEN:
957 		sysfs_remove_group(&s->kobj, &cpu_online_attr_group);
958 		break;
959 	}
960 	return NOTIFY_OK;
961 }
962 
963 static struct notifier_block __cpuinitdata smp_cpu_nb = {
964 	.notifier_call = smp_cpu_notify,
965 };
966 
967 static int __devinit smp_add_present_cpu(int cpu)
968 {
969 	struct cpu *c = &per_cpu(cpu_devices, cpu);
970 	struct sys_device *s = &c->sysdev;
971 	int rc;
972 
973 	c->hotpluggable = 1;
974 	rc = register_cpu(c, cpu);
975 	if (rc)
976 		goto out;
977 	rc = sysfs_create_group(&s->kobj, &cpu_common_attr_group);
978 	if (rc)
979 		goto out_cpu;
980 	if (!cpu_online(cpu))
981 		goto out;
982 	rc = sysfs_create_group(&s->kobj, &cpu_online_attr_group);
983 	if (!rc)
984 		return 0;
985 	sysfs_remove_group(&s->kobj, &cpu_common_attr_group);
986 out_cpu:
987 #ifdef CONFIG_HOTPLUG_CPU
988 	unregister_cpu(c);
989 #endif
990 out:
991 	return rc;
992 }
993 
994 #ifdef CONFIG_HOTPLUG_CPU
995 
996 int __ref smp_rescan_cpus(void)
997 {
998 	cpumask_t newcpus;
999 	int cpu;
1000 	int rc;
1001 
1002 	get_online_cpus();
1003 	mutex_lock(&smp_cpu_state_mutex);
1004 	newcpus = cpu_present_map;
1005 	rc = __smp_rescan_cpus();
1006 	if (rc)
1007 		goto out;
1008 	cpus_andnot(newcpus, cpu_present_map, newcpus);
1009 	for_each_cpu_mask(cpu, newcpus) {
1010 		rc = smp_add_present_cpu(cpu);
1011 		if (rc)
1012 			cpu_clear(cpu, cpu_present_map);
1013 	}
1014 	rc = 0;
1015 out:
1016 	mutex_unlock(&smp_cpu_state_mutex);
1017 	put_online_cpus();
1018 	if (!cpus_empty(newcpus))
1019 		topology_schedule_update();
1020 	return rc;
1021 }
1022 
1023 static ssize_t __ref rescan_store(struct sysdev_class *class, const char *buf,
1024 				  size_t count)
1025 {
1026 	int rc;
1027 
1028 	rc = smp_rescan_cpus();
1029 	return rc ? rc : count;
1030 }
1031 static SYSDEV_CLASS_ATTR(rescan, 0200, NULL, rescan_store);
1032 #endif /* CONFIG_HOTPLUG_CPU */
1033 
1034 static ssize_t dispatching_show(struct sysdev_class *class, char *buf)
1035 {
1036 	ssize_t count;
1037 
1038 	mutex_lock(&smp_cpu_state_mutex);
1039 	count = sprintf(buf, "%d\n", cpu_management);
1040 	mutex_unlock(&smp_cpu_state_mutex);
1041 	return count;
1042 }
1043 
1044 static ssize_t dispatching_store(struct sysdev_class *dev, const char *buf,
1045 				 size_t count)
1046 {
1047 	int val, rc;
1048 	char delim;
1049 
1050 	if (sscanf(buf, "%d %c", &val, &delim) != 1)
1051 		return -EINVAL;
1052 	if (val != 0 && val != 1)
1053 		return -EINVAL;
1054 	rc = 0;
1055 	get_online_cpus();
1056 	mutex_lock(&smp_cpu_state_mutex);
1057 	if (cpu_management == val)
1058 		goto out;
1059 	rc = topology_set_cpu_management(val);
1060 	if (!rc)
1061 		cpu_management = val;
1062 out:
1063 	mutex_unlock(&smp_cpu_state_mutex);
1064 	put_online_cpus();
1065 	return rc ? rc : count;
1066 }
1067 static SYSDEV_CLASS_ATTR(dispatching, 0644, dispatching_show,
1068 			 dispatching_store);
1069 
1070 static int __init topology_init(void)
1071 {
1072 	int cpu;
1073 	int rc;
1074 
1075 	register_cpu_notifier(&smp_cpu_nb);
1076 
1077 #ifdef CONFIG_HOTPLUG_CPU
1078 	rc = sysdev_class_create_file(&cpu_sysdev_class, &attr_rescan);
1079 	if (rc)
1080 		return rc;
1081 #endif
1082 	rc = sysdev_class_create_file(&cpu_sysdev_class, &attr_dispatching);
1083 	if (rc)
1084 		return rc;
1085 	for_each_present_cpu(cpu) {
1086 		rc = smp_add_present_cpu(cpu);
1087 		if (rc)
1088 			return rc;
1089 	}
1090 	return 0;
1091 }
1092 subsys_initcall(topology_init);
1093