xref: /openbmc/linux/arch/s390/kernel/smp.c (revision baa7eb025ab14f3cba2e35c0a8648f9c9f01d24f)
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 <linux/slab.h>
40 #include <asm/asm-offsets.h>
41 #include <asm/ipl.h>
42 #include <asm/setup.h>
43 #include <asm/sigp.h>
44 #include <asm/pgalloc.h>
45 #include <asm/irq.h>
46 #include <asm/s390_ext.h>
47 #include <asm/cpcmd.h>
48 #include <asm/tlbflush.h>
49 #include <asm/timer.h>
50 #include <asm/lowcore.h>
51 #include <asm/sclp.h>
52 #include <asm/cputime.h>
53 #include <asm/vdso.h>
54 #include <asm/cpu.h>
55 #include "entry.h"
56 
57 /* logical cpu to cpu address */
58 unsigned short __cpu_logical_map[NR_CPUS];
59 
60 static struct task_struct *current_set[NR_CPUS];
61 
62 static u8 smp_cpu_type;
63 static int smp_use_sigp_detection;
64 
65 enum s390_cpu_state {
66 	CPU_STATE_STANDBY,
67 	CPU_STATE_CONFIGURED,
68 };
69 
70 DEFINE_MUTEX(smp_cpu_state_mutex);
71 int smp_cpu_polarization[NR_CPUS];
72 static int smp_cpu_state[NR_CPUS];
73 static int cpu_management;
74 
75 static DEFINE_PER_CPU(struct cpu, cpu_devices);
76 
77 static void smp_ext_bitcall(int, int);
78 
79 static int raw_cpu_stopped(int cpu)
80 {
81 	u32 status;
82 
83 	switch (raw_sigp_ps(&status, 0, cpu, sigp_sense)) {
84 	case sigp_status_stored:
85 		/* Check for stopped and check stop state */
86 		if (status & 0x50)
87 			return 1;
88 		break;
89 	default:
90 		break;
91 	}
92 	return 0;
93 }
94 
95 static inline int cpu_stopped(int cpu)
96 {
97 	return raw_cpu_stopped(cpu_logical_map(cpu));
98 }
99 
100 void smp_switch_to_ipl_cpu(void (*func)(void *), void *data)
101 {
102 	struct _lowcore *lc, *current_lc;
103 	struct stack_frame *sf;
104 	struct pt_regs *regs;
105 	unsigned long sp;
106 
107 	if (smp_processor_id() == 0)
108 		func(data);
109 	__load_psw_mask(PSW_BASE_BITS | PSW_DEFAULT_KEY);
110 	/* Disable lowcore protection */
111 	__ctl_clear_bit(0, 28);
112 	current_lc = lowcore_ptr[smp_processor_id()];
113 	lc = lowcore_ptr[0];
114 	if (!lc)
115 		lc = current_lc;
116 	lc->restart_psw.mask = PSW_BASE_BITS | PSW_DEFAULT_KEY;
117 	lc->restart_psw.addr = PSW_ADDR_AMODE | (unsigned long) smp_restart_cpu;
118 	if (!cpu_online(0))
119 		smp_switch_to_cpu(func, data, 0, stap(), __cpu_logical_map[0]);
120 	while (sigp(0, sigp_stop_and_store_status) == sigp_busy)
121 		cpu_relax();
122 	sp = lc->panic_stack;
123 	sp -= sizeof(struct pt_regs);
124 	regs = (struct pt_regs *) sp;
125 	memcpy(&regs->gprs, &current_lc->gpregs_save_area, sizeof(regs->gprs));
126 	regs->psw = lc->psw_save_area;
127 	sp -= STACK_FRAME_OVERHEAD;
128 	sf = (struct stack_frame *) sp;
129 	sf->back_chain = regs->gprs[15];
130 	smp_switch_to_cpu(func, data, sp, stap(), __cpu_logical_map[0]);
131 }
132 
133 void smp_send_stop(void)
134 {
135 	int cpu, rc;
136 
137 	/* Disable all interrupts/machine checks */
138 	__load_psw_mask(psw_kernel_bits & ~PSW_MASK_MCHECK);
139 	trace_hardirqs_off();
140 
141 	/* stop all processors */
142 	for_each_online_cpu(cpu) {
143 		if (cpu == smp_processor_id())
144 			continue;
145 		do {
146 			rc = sigp(cpu, sigp_stop);
147 		} while (rc == sigp_busy);
148 
149 		while (!cpu_stopped(cpu))
150 			cpu_relax();
151 	}
152 }
153 
154 /*
155  * This is the main routine where commands issued by other
156  * cpus are handled.
157  */
158 
159 static void do_ext_call_interrupt(unsigned int ext_int_code,
160 				  unsigned int param32, unsigned long param64)
161 {
162 	unsigned long bits;
163 
164 	/*
165 	 * handle bit signal external calls
166 	 *
167 	 * For the ec_schedule signal we have to do nothing. All the work
168 	 * is done automatically when we return from the interrupt.
169 	 */
170 	bits = xchg(&S390_lowcore.ext_call_fast, 0);
171 
172 	if (test_bit(ec_call_function, &bits))
173 		generic_smp_call_function_interrupt();
174 
175 	if (test_bit(ec_call_function_single, &bits))
176 		generic_smp_call_function_single_interrupt();
177 }
178 
179 /*
180  * Send an external call sigp to another cpu and return without waiting
181  * for its completion.
182  */
183 static void smp_ext_bitcall(int cpu, int sig)
184 {
185 	/*
186 	 * Set signaling bit in lowcore of target cpu and kick it
187 	 */
188 	set_bit(sig, (unsigned long *) &lowcore_ptr[cpu]->ext_call_fast);
189 	while (sigp(cpu, sigp_emergency_signal) == sigp_busy)
190 		udelay(10);
191 }
192 
193 void arch_send_call_function_ipi_mask(const struct cpumask *mask)
194 {
195 	int cpu;
196 
197 	for_each_cpu(cpu, mask)
198 		smp_ext_bitcall(cpu, ec_call_function);
199 }
200 
201 void arch_send_call_function_single_ipi(int cpu)
202 {
203 	smp_ext_bitcall(cpu, ec_call_function_single);
204 }
205 
206 #ifndef CONFIG_64BIT
207 /*
208  * this function sends a 'purge tlb' signal to another CPU.
209  */
210 static void smp_ptlb_callback(void *info)
211 {
212 	__tlb_flush_local();
213 }
214 
215 void smp_ptlb_all(void)
216 {
217 	on_each_cpu(smp_ptlb_callback, NULL, 1);
218 }
219 EXPORT_SYMBOL(smp_ptlb_all);
220 #endif /* ! CONFIG_64BIT */
221 
222 /*
223  * this function sends a 'reschedule' IPI to another CPU.
224  * it goes straight through and wastes no time serializing
225  * anything. Worst case is that we lose a reschedule ...
226  */
227 void smp_send_reschedule(int cpu)
228 {
229 	smp_ext_bitcall(cpu, ec_schedule);
230 }
231 
232 /*
233  * parameter area for the set/clear control bit callbacks
234  */
235 struct ec_creg_mask_parms {
236 	unsigned long orvals[16];
237 	unsigned long andvals[16];
238 };
239 
240 /*
241  * callback for setting/clearing control bits
242  */
243 static void smp_ctl_bit_callback(void *info)
244 {
245 	struct ec_creg_mask_parms *pp = info;
246 	unsigned long cregs[16];
247 	int i;
248 
249 	__ctl_store(cregs, 0, 15);
250 	for (i = 0; i <= 15; i++)
251 		cregs[i] = (cregs[i] & pp->andvals[i]) | pp->orvals[i];
252 	__ctl_load(cregs, 0, 15);
253 }
254 
255 /*
256  * Set a bit in a control register of all cpus
257  */
258 void smp_ctl_set_bit(int cr, int bit)
259 {
260 	struct ec_creg_mask_parms parms;
261 
262 	memset(&parms.orvals, 0, sizeof(parms.orvals));
263 	memset(&parms.andvals, 0xff, sizeof(parms.andvals));
264 	parms.orvals[cr] = 1 << bit;
265 	on_each_cpu(smp_ctl_bit_callback, &parms, 1);
266 }
267 EXPORT_SYMBOL(smp_ctl_set_bit);
268 
269 /*
270  * Clear a bit in a control register of all cpus
271  */
272 void smp_ctl_clear_bit(int cr, int bit)
273 {
274 	struct ec_creg_mask_parms parms;
275 
276 	memset(&parms.orvals, 0, sizeof(parms.orvals));
277 	memset(&parms.andvals, 0xff, sizeof(parms.andvals));
278 	parms.andvals[cr] = ~(1L << bit);
279 	on_each_cpu(smp_ctl_bit_callback, &parms, 1);
280 }
281 EXPORT_SYMBOL(smp_ctl_clear_bit);
282 
283 #ifdef CONFIG_ZFCPDUMP
284 
285 static void __init smp_get_save_area(unsigned int cpu, unsigned int phy_cpu)
286 {
287 	if (ipl_info.type != IPL_TYPE_FCP_DUMP)
288 		return;
289 	if (cpu >= NR_CPUS) {
290 		pr_warning("CPU %i exceeds the maximum %i and is excluded from "
291 			   "the dump\n", cpu, NR_CPUS - 1);
292 		return;
293 	}
294 	zfcpdump_save_areas[cpu] = kmalloc(sizeof(struct save_area), GFP_KERNEL);
295 	while (raw_sigp(phy_cpu, sigp_stop_and_store_status) == sigp_busy)
296 		cpu_relax();
297 	memcpy_real(zfcpdump_save_areas[cpu],
298 		    (void *)(unsigned long) store_prefix() + SAVE_AREA_BASE,
299 		    sizeof(struct save_area));
300 }
301 
302 struct save_area *zfcpdump_save_areas[NR_CPUS + 1];
303 EXPORT_SYMBOL_GPL(zfcpdump_save_areas);
304 
305 #else
306 
307 static inline void smp_get_save_area(unsigned int cpu, unsigned int phy_cpu) { }
308 
309 #endif /* CONFIG_ZFCPDUMP */
310 
311 static int cpu_known(int cpu_id)
312 {
313 	int cpu;
314 
315 	for_each_present_cpu(cpu) {
316 		if (__cpu_logical_map[cpu] == cpu_id)
317 			return 1;
318 	}
319 	return 0;
320 }
321 
322 static int smp_rescan_cpus_sigp(cpumask_t avail)
323 {
324 	int cpu_id, logical_cpu;
325 
326 	logical_cpu = cpumask_first(&avail);
327 	if (logical_cpu >= nr_cpu_ids)
328 		return 0;
329 	for (cpu_id = 0; cpu_id <= MAX_CPU_ADDRESS; cpu_id++) {
330 		if (cpu_known(cpu_id))
331 			continue;
332 		__cpu_logical_map[logical_cpu] = cpu_id;
333 		smp_cpu_polarization[logical_cpu] = POLARIZATION_UNKNWN;
334 		if (!cpu_stopped(logical_cpu))
335 			continue;
336 		cpu_set(logical_cpu, cpu_present_map);
337 		smp_cpu_state[logical_cpu] = CPU_STATE_CONFIGURED;
338 		logical_cpu = cpumask_next(logical_cpu, &avail);
339 		if (logical_cpu >= nr_cpu_ids)
340 			break;
341 	}
342 	return 0;
343 }
344 
345 static int smp_rescan_cpus_sclp(cpumask_t avail)
346 {
347 	struct sclp_cpu_info *info;
348 	int cpu_id, logical_cpu, cpu;
349 	int rc;
350 
351 	logical_cpu = cpumask_first(&avail);
352 	if (logical_cpu >= nr_cpu_ids)
353 		return 0;
354 	info = kmalloc(sizeof(*info), GFP_KERNEL);
355 	if (!info)
356 		return -ENOMEM;
357 	rc = sclp_get_cpu_info(info);
358 	if (rc)
359 		goto out;
360 	for (cpu = 0; cpu < info->combined; cpu++) {
361 		if (info->has_cpu_type && info->cpu[cpu].type != smp_cpu_type)
362 			continue;
363 		cpu_id = info->cpu[cpu].address;
364 		if (cpu_known(cpu_id))
365 			continue;
366 		__cpu_logical_map[logical_cpu] = cpu_id;
367 		smp_cpu_polarization[logical_cpu] = POLARIZATION_UNKNWN;
368 		cpu_set(logical_cpu, cpu_present_map);
369 		if (cpu >= info->configured)
370 			smp_cpu_state[logical_cpu] = CPU_STATE_STANDBY;
371 		else
372 			smp_cpu_state[logical_cpu] = CPU_STATE_CONFIGURED;
373 		logical_cpu = cpumask_next(logical_cpu, &avail);
374 		if (logical_cpu >= nr_cpu_ids)
375 			break;
376 	}
377 out:
378 	kfree(info);
379 	return rc;
380 }
381 
382 static int __smp_rescan_cpus(void)
383 {
384 	cpumask_t avail;
385 
386 	cpus_xor(avail, cpu_possible_map, cpu_present_map);
387 	if (smp_use_sigp_detection)
388 		return smp_rescan_cpus_sigp(avail);
389 	else
390 		return smp_rescan_cpus_sclp(avail);
391 }
392 
393 static void __init smp_detect_cpus(void)
394 {
395 	unsigned int cpu, c_cpus, s_cpus;
396 	struct sclp_cpu_info *info;
397 	u16 boot_cpu_addr, cpu_addr;
398 
399 	c_cpus = 1;
400 	s_cpus = 0;
401 	boot_cpu_addr = __cpu_logical_map[0];
402 	info = kmalloc(sizeof(*info), GFP_KERNEL);
403 	if (!info)
404 		panic("smp_detect_cpus failed to allocate memory\n");
405 	/* Use sigp detection algorithm if sclp doesn't work. */
406 	if (sclp_get_cpu_info(info)) {
407 		smp_use_sigp_detection = 1;
408 		for (cpu = 0; cpu <= MAX_CPU_ADDRESS; cpu++) {
409 			if (cpu == boot_cpu_addr)
410 				continue;
411 			if (!raw_cpu_stopped(cpu))
412 				continue;
413 			smp_get_save_area(c_cpus, cpu);
414 			c_cpus++;
415 		}
416 		goto out;
417 	}
418 
419 	if (info->has_cpu_type) {
420 		for (cpu = 0; cpu < info->combined; cpu++) {
421 			if (info->cpu[cpu].address == boot_cpu_addr) {
422 				smp_cpu_type = info->cpu[cpu].type;
423 				break;
424 			}
425 		}
426 	}
427 
428 	for (cpu = 0; cpu < info->combined; cpu++) {
429 		if (info->has_cpu_type && info->cpu[cpu].type != smp_cpu_type)
430 			continue;
431 		cpu_addr = info->cpu[cpu].address;
432 		if (cpu_addr == boot_cpu_addr)
433 			continue;
434 		if (!raw_cpu_stopped(cpu_addr)) {
435 			s_cpus++;
436 			continue;
437 		}
438 		smp_get_save_area(c_cpus, cpu_addr);
439 		c_cpus++;
440 	}
441 out:
442 	kfree(info);
443 	pr_info("%d configured CPUs, %d standby CPUs\n", c_cpus, s_cpus);
444 	get_online_cpus();
445 	__smp_rescan_cpus();
446 	put_online_cpus();
447 }
448 
449 /*
450  *	Activate a secondary processor.
451  */
452 int __cpuinit start_secondary(void *cpuvoid)
453 {
454 	/* Setup the cpu */
455 	cpu_init();
456 	preempt_disable();
457 	/* Enable TOD clock interrupts on the secondary cpu. */
458 	init_cpu_timer();
459 	/* Enable cpu timer interrupts on the secondary cpu. */
460 	init_cpu_vtimer();
461 	/* Enable pfault pseudo page faults on this cpu. */
462 	pfault_init();
463 
464 	/* call cpu notifiers */
465 	notify_cpu_starting(smp_processor_id());
466 	/* Mark this cpu as online */
467 	ipi_call_lock();
468 	cpu_set(smp_processor_id(), cpu_online_map);
469 	ipi_call_unlock();
470 	/* Switch on interrupts */
471 	local_irq_enable();
472 	/* Print info about this processor */
473 	print_cpu_info();
474 	/* cpu_idle will call schedule for us */
475 	cpu_idle();
476 	return 0;
477 }
478 
479 static void __init smp_create_idle(unsigned int cpu)
480 {
481 	struct task_struct *p;
482 
483 	/*
484 	 *  don't care about the psw and regs settings since we'll never
485 	 *  reschedule the forked task.
486 	 */
487 	p = fork_idle(cpu);
488 	if (IS_ERR(p))
489 		panic("failed fork for CPU %u: %li", cpu, PTR_ERR(p));
490 	current_set[cpu] = p;
491 }
492 
493 static int __cpuinit smp_alloc_lowcore(int cpu)
494 {
495 	unsigned long async_stack, panic_stack;
496 	struct _lowcore *lowcore;
497 
498 	lowcore = (void *) __get_free_pages(GFP_KERNEL | GFP_DMA, LC_ORDER);
499 	if (!lowcore)
500 		return -ENOMEM;
501 	async_stack = __get_free_pages(GFP_KERNEL, ASYNC_ORDER);
502 	panic_stack = __get_free_page(GFP_KERNEL);
503 	if (!panic_stack || !async_stack)
504 		goto out;
505 	memcpy(lowcore, &S390_lowcore, 512);
506 	memset((char *)lowcore + 512, 0, sizeof(*lowcore) - 512);
507 	lowcore->async_stack = async_stack + ASYNC_SIZE;
508 	lowcore->panic_stack = panic_stack + PAGE_SIZE;
509 
510 #ifndef CONFIG_64BIT
511 	if (MACHINE_HAS_IEEE) {
512 		unsigned long save_area;
513 
514 		save_area = get_zeroed_page(GFP_KERNEL);
515 		if (!save_area)
516 			goto out;
517 		lowcore->extended_save_area_addr = (u32) save_area;
518 	}
519 #else
520 	if (vdso_alloc_per_cpu(cpu, lowcore))
521 		goto out;
522 #endif
523 	lowcore_ptr[cpu] = lowcore;
524 	return 0;
525 
526 out:
527 	free_page(panic_stack);
528 	free_pages(async_stack, ASYNC_ORDER);
529 	free_pages((unsigned long) lowcore, LC_ORDER);
530 	return -ENOMEM;
531 }
532 
533 static void smp_free_lowcore(int cpu)
534 {
535 	struct _lowcore *lowcore;
536 
537 	lowcore = lowcore_ptr[cpu];
538 #ifndef CONFIG_64BIT
539 	if (MACHINE_HAS_IEEE)
540 		free_page((unsigned long) lowcore->extended_save_area_addr);
541 #else
542 	vdso_free_per_cpu(cpu, lowcore);
543 #endif
544 	free_page(lowcore->panic_stack - PAGE_SIZE);
545 	free_pages(lowcore->async_stack - ASYNC_SIZE, ASYNC_ORDER);
546 	free_pages((unsigned long) lowcore, LC_ORDER);
547 	lowcore_ptr[cpu] = NULL;
548 }
549 
550 /* Upping and downing of CPUs */
551 int __cpuinit __cpu_up(unsigned int cpu)
552 {
553 	struct _lowcore *cpu_lowcore;
554 	struct task_struct *idle;
555 	struct stack_frame *sf;
556 	u32 lowcore;
557 	int ccode;
558 
559 	if (smp_cpu_state[cpu] != CPU_STATE_CONFIGURED)
560 		return -EIO;
561 	if (smp_alloc_lowcore(cpu))
562 		return -ENOMEM;
563 	do {
564 		ccode = sigp(cpu, sigp_initial_cpu_reset);
565 		if (ccode == sigp_busy)
566 			udelay(10);
567 		if (ccode == sigp_not_operational)
568 			goto err_out;
569 	} while (ccode == sigp_busy);
570 
571 	lowcore = (u32)(unsigned long)lowcore_ptr[cpu];
572 	while (sigp_p(lowcore, cpu, sigp_set_prefix) == sigp_busy)
573 		udelay(10);
574 
575 	idle = current_set[cpu];
576 	cpu_lowcore = lowcore_ptr[cpu];
577 	cpu_lowcore->kernel_stack = (unsigned long)
578 		task_stack_page(idle) + THREAD_SIZE;
579 	cpu_lowcore->thread_info = (unsigned long) task_thread_info(idle);
580 	sf = (struct stack_frame *) (cpu_lowcore->kernel_stack
581 				     - sizeof(struct pt_regs)
582 				     - sizeof(struct stack_frame));
583 	memset(sf, 0, sizeof(struct stack_frame));
584 	sf->gprs[9] = (unsigned long) sf;
585 	cpu_lowcore->save_area[15] = (unsigned long) sf;
586 	__ctl_store(cpu_lowcore->cregs_save_area, 0, 15);
587 	atomic_inc(&init_mm.context.attach_count);
588 	asm volatile(
589 		"	stam	0,15,0(%0)"
590 		: : "a" (&cpu_lowcore->access_regs_save_area) : "memory");
591 	cpu_lowcore->percpu_offset = __per_cpu_offset[cpu];
592 	cpu_lowcore->current_task = (unsigned long) idle;
593 	cpu_lowcore->cpu_nr = cpu;
594 	cpu_lowcore->kernel_asce = S390_lowcore.kernel_asce;
595 	cpu_lowcore->machine_flags = S390_lowcore.machine_flags;
596 	cpu_lowcore->ftrace_func = S390_lowcore.ftrace_func;
597 	memcpy(cpu_lowcore->stfle_fac_list, S390_lowcore.stfle_fac_list,
598 	       MAX_FACILITY_BIT/8);
599 	eieio();
600 
601 	while (sigp(cpu, sigp_restart) == sigp_busy)
602 		udelay(10);
603 
604 	while (!cpu_online(cpu))
605 		cpu_relax();
606 	return 0;
607 
608 err_out:
609 	smp_free_lowcore(cpu);
610 	return -EIO;
611 }
612 
613 static int __init setup_possible_cpus(char *s)
614 {
615 	int pcpus, cpu;
616 
617 	pcpus = simple_strtoul(s, NULL, 0);
618 	init_cpu_possible(cpumask_of(0));
619 	for (cpu = 1; cpu < pcpus && cpu < nr_cpu_ids; cpu++)
620 		set_cpu_possible(cpu, true);
621 	return 0;
622 }
623 early_param("possible_cpus", setup_possible_cpus);
624 
625 #ifdef CONFIG_HOTPLUG_CPU
626 
627 int __cpu_disable(void)
628 {
629 	struct ec_creg_mask_parms cr_parms;
630 	int cpu = smp_processor_id();
631 
632 	cpu_clear(cpu, cpu_online_map);
633 
634 	/* Disable pfault pseudo page faults on this cpu. */
635 	pfault_fini();
636 
637 	memset(&cr_parms.orvals, 0, sizeof(cr_parms.orvals));
638 	memset(&cr_parms.andvals, 0xff, sizeof(cr_parms.andvals));
639 
640 	/* disable all external interrupts */
641 	cr_parms.orvals[0] = 0;
642 	cr_parms.andvals[0] = ~(1 << 15 | 1 << 14 | 1 << 13 | 1 << 12 |
643 				1 << 11 | 1 << 10 | 1 <<  6 | 1 <<  4);
644 	/* disable all I/O interrupts */
645 	cr_parms.orvals[6] = 0;
646 	cr_parms.andvals[6] = ~(1 << 31 | 1 << 30 | 1 << 29 | 1 << 28 |
647 				1 << 27 | 1 << 26 | 1 << 25 | 1 << 24);
648 	/* disable most machine checks */
649 	cr_parms.orvals[14] = 0;
650 	cr_parms.andvals[14] = ~(1 << 28 | 1 << 27 | 1 << 26 |
651 				 1 << 25 | 1 << 24);
652 
653 	smp_ctl_bit_callback(&cr_parms);
654 
655 	return 0;
656 }
657 
658 void __cpu_die(unsigned int cpu)
659 {
660 	/* Wait until target cpu is down */
661 	while (!cpu_stopped(cpu))
662 		cpu_relax();
663 	while (sigp_p(0, cpu, sigp_set_prefix) == sigp_busy)
664 		udelay(10);
665 	smp_free_lowcore(cpu);
666 	atomic_dec(&init_mm.context.attach_count);
667 	pr_info("Processor %d stopped\n", cpu);
668 }
669 
670 void cpu_die(void)
671 {
672 	idle_task_exit();
673 	while (sigp(smp_processor_id(), sigp_stop) == sigp_busy)
674 		cpu_relax();
675 	for (;;);
676 }
677 
678 #endif /* CONFIG_HOTPLUG_CPU */
679 
680 void __init smp_prepare_cpus(unsigned int max_cpus)
681 {
682 #ifndef CONFIG_64BIT
683 	unsigned long save_area = 0;
684 #endif
685 	unsigned long async_stack, panic_stack;
686 	struct _lowcore *lowcore;
687 	unsigned int cpu;
688 
689 	smp_detect_cpus();
690 
691 	/* request the 0x1201 emergency signal external interrupt */
692 	if (register_external_interrupt(0x1201, do_ext_call_interrupt) != 0)
693 		panic("Couldn't request external interrupt 0x1201");
694 	print_cpu_info();
695 
696 	/* Reallocate current lowcore, but keep its contents. */
697 	lowcore = (void *) __get_free_pages(GFP_KERNEL | GFP_DMA, LC_ORDER);
698 	panic_stack = __get_free_page(GFP_KERNEL);
699 	async_stack = __get_free_pages(GFP_KERNEL, ASYNC_ORDER);
700 	BUG_ON(!lowcore || !panic_stack || !async_stack);
701 #ifndef CONFIG_64BIT
702 	if (MACHINE_HAS_IEEE)
703 		save_area = get_zeroed_page(GFP_KERNEL);
704 #endif
705 	local_irq_disable();
706 	local_mcck_disable();
707 	lowcore_ptr[smp_processor_id()] = lowcore;
708 	*lowcore = S390_lowcore;
709 	lowcore->panic_stack = panic_stack + PAGE_SIZE;
710 	lowcore->async_stack = async_stack + ASYNC_SIZE;
711 #ifndef CONFIG_64BIT
712 	if (MACHINE_HAS_IEEE)
713 		lowcore->extended_save_area_addr = (u32) save_area;
714 #endif
715 	set_prefix((u32)(unsigned long) lowcore);
716 	local_mcck_enable();
717 	local_irq_enable();
718 #ifdef CONFIG_64BIT
719 	if (vdso_alloc_per_cpu(smp_processor_id(), &S390_lowcore))
720 		BUG();
721 #endif
722 	for_each_possible_cpu(cpu)
723 		if (cpu != smp_processor_id())
724 			smp_create_idle(cpu);
725 }
726 
727 void __init smp_prepare_boot_cpu(void)
728 {
729 	BUG_ON(smp_processor_id() != 0);
730 
731 	current_thread_info()->cpu = 0;
732 	cpu_set(0, cpu_present_map);
733 	cpu_set(0, cpu_online_map);
734 	S390_lowcore.percpu_offset = __per_cpu_offset[0];
735 	current_set[0] = current;
736 	smp_cpu_state[0] = CPU_STATE_CONFIGURED;
737 	smp_cpu_polarization[0] = POLARIZATION_UNKNWN;
738 }
739 
740 void __init smp_cpus_done(unsigned int max_cpus)
741 {
742 }
743 
744 void __init smp_setup_processor_id(void)
745 {
746 	S390_lowcore.cpu_nr = 0;
747 	__cpu_logical_map[0] = stap();
748 }
749 
750 /*
751  * the frequency of the profiling timer can be changed
752  * by writing a multiplier value into /proc/profile.
753  *
754  * usually you want to run this on all CPUs ;)
755  */
756 int setup_profiling_timer(unsigned int multiplier)
757 {
758 	return 0;
759 }
760 
761 #ifdef CONFIG_HOTPLUG_CPU
762 static ssize_t cpu_configure_show(struct sys_device *dev,
763 				struct sysdev_attribute *attr, char *buf)
764 {
765 	ssize_t count;
766 
767 	mutex_lock(&smp_cpu_state_mutex);
768 	count = sprintf(buf, "%d\n", smp_cpu_state[dev->id]);
769 	mutex_unlock(&smp_cpu_state_mutex);
770 	return count;
771 }
772 
773 static ssize_t cpu_configure_store(struct sys_device *dev,
774 				  struct sysdev_attribute *attr,
775 				  const char *buf, size_t count)
776 {
777 	int cpu = dev->id;
778 	int val, rc;
779 	char delim;
780 
781 	if (sscanf(buf, "%d %c", &val, &delim) != 1)
782 		return -EINVAL;
783 	if (val != 0 && val != 1)
784 		return -EINVAL;
785 
786 	get_online_cpus();
787 	mutex_lock(&smp_cpu_state_mutex);
788 	rc = -EBUSY;
789 	/* disallow configuration changes of online cpus and cpu 0 */
790 	if (cpu_online(cpu) || cpu == 0)
791 		goto out;
792 	rc = 0;
793 	switch (val) {
794 	case 0:
795 		if (smp_cpu_state[cpu] == CPU_STATE_CONFIGURED) {
796 			rc = sclp_cpu_deconfigure(__cpu_logical_map[cpu]);
797 			if (!rc) {
798 				smp_cpu_state[cpu] = CPU_STATE_STANDBY;
799 				smp_cpu_polarization[cpu] = POLARIZATION_UNKNWN;
800 			}
801 		}
802 		break;
803 	case 1:
804 		if (smp_cpu_state[cpu] == CPU_STATE_STANDBY) {
805 			rc = sclp_cpu_configure(__cpu_logical_map[cpu]);
806 			if (!rc) {
807 				smp_cpu_state[cpu] = CPU_STATE_CONFIGURED;
808 				smp_cpu_polarization[cpu] = POLARIZATION_UNKNWN;
809 			}
810 		}
811 		break;
812 	default:
813 		break;
814 	}
815 out:
816 	mutex_unlock(&smp_cpu_state_mutex);
817 	put_online_cpus();
818 	return rc ? rc : count;
819 }
820 static SYSDEV_ATTR(configure, 0644, cpu_configure_show, cpu_configure_store);
821 #endif /* CONFIG_HOTPLUG_CPU */
822 
823 static ssize_t cpu_polarization_show(struct sys_device *dev,
824 				     struct sysdev_attribute *attr, char *buf)
825 {
826 	int cpu = dev->id;
827 	ssize_t count;
828 
829 	mutex_lock(&smp_cpu_state_mutex);
830 	switch (smp_cpu_polarization[cpu]) {
831 	case POLARIZATION_HRZ:
832 		count = sprintf(buf, "horizontal\n");
833 		break;
834 	case POLARIZATION_VL:
835 		count = sprintf(buf, "vertical:low\n");
836 		break;
837 	case POLARIZATION_VM:
838 		count = sprintf(buf, "vertical:medium\n");
839 		break;
840 	case POLARIZATION_VH:
841 		count = sprintf(buf, "vertical:high\n");
842 		break;
843 	default:
844 		count = sprintf(buf, "unknown\n");
845 		break;
846 	}
847 	mutex_unlock(&smp_cpu_state_mutex);
848 	return count;
849 }
850 static SYSDEV_ATTR(polarization, 0444, cpu_polarization_show, NULL);
851 
852 static ssize_t show_cpu_address(struct sys_device *dev,
853 				struct sysdev_attribute *attr, char *buf)
854 {
855 	return sprintf(buf, "%d\n", __cpu_logical_map[dev->id]);
856 }
857 static SYSDEV_ATTR(address, 0444, show_cpu_address, NULL);
858 
859 
860 static struct attribute *cpu_common_attrs[] = {
861 #ifdef CONFIG_HOTPLUG_CPU
862 	&attr_configure.attr,
863 #endif
864 	&attr_address.attr,
865 	&attr_polarization.attr,
866 	NULL,
867 };
868 
869 static struct attribute_group cpu_common_attr_group = {
870 	.attrs = cpu_common_attrs,
871 };
872 
873 static ssize_t show_capability(struct sys_device *dev,
874 				struct sysdev_attribute *attr, char *buf)
875 {
876 	unsigned int capability;
877 	int rc;
878 
879 	rc = get_cpu_capability(&capability);
880 	if (rc)
881 		return rc;
882 	return sprintf(buf, "%u\n", capability);
883 }
884 static SYSDEV_ATTR(capability, 0444, show_capability, NULL);
885 
886 static ssize_t show_idle_count(struct sys_device *dev,
887 				struct sysdev_attribute *attr, char *buf)
888 {
889 	struct s390_idle_data *idle;
890 	unsigned long long idle_count;
891 	unsigned int sequence;
892 
893 	idle = &per_cpu(s390_idle, dev->id);
894 repeat:
895 	sequence = idle->sequence;
896 	smp_rmb();
897 	if (sequence & 1)
898 		goto repeat;
899 	idle_count = idle->idle_count;
900 	if (idle->idle_enter)
901 		idle_count++;
902 	smp_rmb();
903 	if (idle->sequence != sequence)
904 		goto repeat;
905 	return sprintf(buf, "%llu\n", idle_count);
906 }
907 static SYSDEV_ATTR(idle_count, 0444, show_idle_count, NULL);
908 
909 static ssize_t show_idle_time(struct sys_device *dev,
910 				struct sysdev_attribute *attr, char *buf)
911 {
912 	struct s390_idle_data *idle;
913 	unsigned long long now, idle_time, idle_enter;
914 	unsigned int sequence;
915 
916 	idle = &per_cpu(s390_idle, dev->id);
917 	now = get_clock();
918 repeat:
919 	sequence = idle->sequence;
920 	smp_rmb();
921 	if (sequence & 1)
922 		goto repeat;
923 	idle_time = idle->idle_time;
924 	idle_enter = idle->idle_enter;
925 	if (idle_enter != 0ULL && idle_enter < now)
926 		idle_time += now - idle_enter;
927 	smp_rmb();
928 	if (idle->sequence != sequence)
929 		goto repeat;
930 	return sprintf(buf, "%llu\n", idle_time >> 12);
931 }
932 static SYSDEV_ATTR(idle_time_us, 0444, show_idle_time, NULL);
933 
934 static struct attribute *cpu_online_attrs[] = {
935 	&attr_capability.attr,
936 	&attr_idle_count.attr,
937 	&attr_idle_time_us.attr,
938 	NULL,
939 };
940 
941 static struct attribute_group cpu_online_attr_group = {
942 	.attrs = cpu_online_attrs,
943 };
944 
945 static int __cpuinit smp_cpu_notify(struct notifier_block *self,
946 				    unsigned long action, void *hcpu)
947 {
948 	unsigned int cpu = (unsigned int)(long)hcpu;
949 	struct cpu *c = &per_cpu(cpu_devices, cpu);
950 	struct sys_device *s = &c->sysdev;
951 	struct s390_idle_data *idle;
952 	int err = 0;
953 
954 	switch (action) {
955 	case CPU_ONLINE:
956 	case CPU_ONLINE_FROZEN:
957 		idle = &per_cpu(s390_idle, cpu);
958 		memset(idle, 0, sizeof(struct s390_idle_data));
959 		err = sysfs_create_group(&s->kobj, &cpu_online_attr_group);
960 		break;
961 	case CPU_DEAD:
962 	case CPU_DEAD_FROZEN:
963 		sysfs_remove_group(&s->kobj, &cpu_online_attr_group);
964 		break;
965 	}
966 	return notifier_from_errno(err);
967 }
968 
969 static struct notifier_block __cpuinitdata smp_cpu_nb = {
970 	.notifier_call = smp_cpu_notify,
971 };
972 
973 static int __devinit smp_add_present_cpu(int cpu)
974 {
975 	struct cpu *c = &per_cpu(cpu_devices, cpu);
976 	struct sys_device *s = &c->sysdev;
977 	int rc;
978 
979 	c->hotpluggable = 1;
980 	rc = register_cpu(c, cpu);
981 	if (rc)
982 		goto out;
983 	rc = sysfs_create_group(&s->kobj, &cpu_common_attr_group);
984 	if (rc)
985 		goto out_cpu;
986 	if (!cpu_online(cpu))
987 		goto out;
988 	rc = sysfs_create_group(&s->kobj, &cpu_online_attr_group);
989 	if (!rc)
990 		return 0;
991 	sysfs_remove_group(&s->kobj, &cpu_common_attr_group);
992 out_cpu:
993 #ifdef CONFIG_HOTPLUG_CPU
994 	unregister_cpu(c);
995 #endif
996 out:
997 	return rc;
998 }
999 
1000 #ifdef CONFIG_HOTPLUG_CPU
1001 
1002 int __ref smp_rescan_cpus(void)
1003 {
1004 	cpumask_t newcpus;
1005 	int cpu;
1006 	int rc;
1007 
1008 	get_online_cpus();
1009 	mutex_lock(&smp_cpu_state_mutex);
1010 	newcpus = cpu_present_map;
1011 	rc = __smp_rescan_cpus();
1012 	if (rc)
1013 		goto out;
1014 	cpus_andnot(newcpus, cpu_present_map, newcpus);
1015 	for_each_cpu_mask(cpu, newcpus) {
1016 		rc = smp_add_present_cpu(cpu);
1017 		if (rc)
1018 			cpu_clear(cpu, cpu_present_map);
1019 	}
1020 	rc = 0;
1021 out:
1022 	mutex_unlock(&smp_cpu_state_mutex);
1023 	put_online_cpus();
1024 	if (!cpus_empty(newcpus))
1025 		topology_schedule_update();
1026 	return rc;
1027 }
1028 
1029 static ssize_t __ref rescan_store(struct sysdev_class *class,
1030 				  struct sysdev_class_attribute *attr,
1031 				  const char *buf,
1032 				  size_t count)
1033 {
1034 	int rc;
1035 
1036 	rc = smp_rescan_cpus();
1037 	return rc ? rc : count;
1038 }
1039 static SYSDEV_CLASS_ATTR(rescan, 0200, NULL, rescan_store);
1040 #endif /* CONFIG_HOTPLUG_CPU */
1041 
1042 static ssize_t dispatching_show(struct sysdev_class *class,
1043 				struct sysdev_class_attribute *attr,
1044 				char *buf)
1045 {
1046 	ssize_t count;
1047 
1048 	mutex_lock(&smp_cpu_state_mutex);
1049 	count = sprintf(buf, "%d\n", cpu_management);
1050 	mutex_unlock(&smp_cpu_state_mutex);
1051 	return count;
1052 }
1053 
1054 static ssize_t dispatching_store(struct sysdev_class *dev,
1055 				 struct sysdev_class_attribute *attr,
1056 				 const char *buf,
1057 				 size_t count)
1058 {
1059 	int val, rc;
1060 	char delim;
1061 
1062 	if (sscanf(buf, "%d %c", &val, &delim) != 1)
1063 		return -EINVAL;
1064 	if (val != 0 && val != 1)
1065 		return -EINVAL;
1066 	rc = 0;
1067 	get_online_cpus();
1068 	mutex_lock(&smp_cpu_state_mutex);
1069 	if (cpu_management == val)
1070 		goto out;
1071 	rc = topology_set_cpu_management(val);
1072 	if (!rc)
1073 		cpu_management = val;
1074 out:
1075 	mutex_unlock(&smp_cpu_state_mutex);
1076 	put_online_cpus();
1077 	return rc ? rc : count;
1078 }
1079 static SYSDEV_CLASS_ATTR(dispatching, 0644, dispatching_show,
1080 			 dispatching_store);
1081 
1082 static int __init topology_init(void)
1083 {
1084 	int cpu;
1085 	int rc;
1086 
1087 	register_cpu_notifier(&smp_cpu_nb);
1088 
1089 #ifdef CONFIG_HOTPLUG_CPU
1090 	rc = sysdev_class_create_file(&cpu_sysdev_class, &attr_rescan);
1091 	if (rc)
1092 		return rc;
1093 #endif
1094 	rc = sysdev_class_create_file(&cpu_sysdev_class, &attr_dispatching);
1095 	if (rc)
1096 		return rc;
1097 	for_each_present_cpu(cpu) {
1098 		rc = smp_add_present_cpu(cpu);
1099 		if (rc)
1100 			return rc;
1101 	}
1102 	return 0;
1103 }
1104 subsys_initcall(topology_init);
1105