xref: /linux/arch/s390/kernel/smp.c (revision 95e9fd10f06cb5642028b6b851e32b8c8afb4571)
1 /*
2  *  SMP related functions
3  *
4  *    Copyright IBM Corp. 1999, 2012
5  *    Author(s): Denis Joseph Barrow,
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  * The code outside of smp.c uses logical cpu numbers, only smp.c does
14  * the translation of logical to physical cpu ids. All new code that
15  * operates on physical cpu numbers needs to go into smp.c.
16  */
17 
18 #define KMSG_COMPONENT "cpu"
19 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
20 
21 #include <linux/workqueue.h>
22 #include <linux/module.h>
23 #include <linux/init.h>
24 #include <linux/mm.h>
25 #include <linux/err.h>
26 #include <linux/spinlock.h>
27 #include <linux/kernel_stat.h>
28 #include <linux/delay.h>
29 #include <linux/interrupt.h>
30 #include <linux/irqflags.h>
31 #include <linux/cpu.h>
32 #include <linux/slab.h>
33 #include <linux/crash_dump.h>
34 #include <asm/asm-offsets.h>
35 #include <asm/switch_to.h>
36 #include <asm/facility.h>
37 #include <asm/ipl.h>
38 #include <asm/setup.h>
39 #include <asm/irq.h>
40 #include <asm/tlbflush.h>
41 #include <asm/vtimer.h>
42 #include <asm/lowcore.h>
43 #include <asm/sclp.h>
44 #include <asm/vdso.h>
45 #include <asm/debug.h>
46 #include <asm/os_info.h>
47 #include <asm/sigp.h>
48 #include "entry.h"
49 
50 enum {
51 	ec_schedule = 0,
52 	ec_call_function,
53 	ec_call_function_single,
54 	ec_stop_cpu,
55 };
56 
57 enum {
58 	CPU_STATE_STANDBY,
59 	CPU_STATE_CONFIGURED,
60 };
61 
62 struct pcpu {
63 	struct cpu cpu;
64 	struct _lowcore *lowcore;	/* lowcore page(s) for the cpu */
65 	unsigned long async_stack;	/* async stack for the cpu */
66 	unsigned long panic_stack;	/* panic stack for the cpu */
67 	unsigned long ec_mask;		/* bit mask for ec_xxx functions */
68 	int state;			/* physical cpu state */
69 	u32 status;			/* last status received via sigp */
70 	u16 address;			/* physical cpu address */
71 };
72 
73 static u8 boot_cpu_type;
74 static u16 boot_cpu_address;
75 static struct pcpu pcpu_devices[NR_CPUS];
76 
77 DEFINE_MUTEX(smp_cpu_state_mutex);
78 
79 /*
80  * Signal processor helper functions.
81  */
82 static inline int __pcpu_sigp(u16 addr, u8 order, u32 parm, u32 *status)
83 {
84 	register unsigned int reg1 asm ("1") = parm;
85 	int cc;
86 
87 	asm volatile(
88 		"	sigp	%1,%2,0(%3)\n"
89 		"	ipm	%0\n"
90 		"	srl	%0,28\n"
91 		: "=d" (cc), "+d" (reg1) : "d" (addr), "a" (order) : "cc");
92 	if (status && cc == 1)
93 		*status = reg1;
94 	return cc;
95 }
96 
97 static inline int __pcpu_sigp_relax(u16 addr, u8 order, u32 parm, u32 *status)
98 {
99 	int cc;
100 
101 	while (1) {
102 		cc = __pcpu_sigp(addr, order, parm, status);
103 		if (cc != SIGP_CC_BUSY)
104 			return cc;
105 		cpu_relax();
106 	}
107 }
108 
109 static int pcpu_sigp_retry(struct pcpu *pcpu, u8 order, u32 parm)
110 {
111 	int cc, retry;
112 
113 	for (retry = 0; ; retry++) {
114 		cc = __pcpu_sigp(pcpu->address, order, parm, &pcpu->status);
115 		if (cc != SIGP_CC_BUSY)
116 			break;
117 		if (retry >= 3)
118 			udelay(10);
119 	}
120 	return cc;
121 }
122 
123 static inline int pcpu_stopped(struct pcpu *pcpu)
124 {
125 	if (__pcpu_sigp(pcpu->address, SIGP_SENSE,
126 			0, &pcpu->status) != SIGP_CC_STATUS_STORED)
127 		return 0;
128 	return !!(pcpu->status & (SIGP_STATUS_CHECK_STOP|SIGP_STATUS_STOPPED));
129 }
130 
131 static inline int pcpu_running(struct pcpu *pcpu)
132 {
133 	if (__pcpu_sigp(pcpu->address, SIGP_SENSE_RUNNING,
134 			0, &pcpu->status) != SIGP_CC_STATUS_STORED)
135 		return 1;
136 	/* Status stored condition code is equivalent to cpu not running. */
137 	return 0;
138 }
139 
140 /*
141  * Find struct pcpu by cpu address.
142  */
143 static struct pcpu *pcpu_find_address(const struct cpumask *mask, int address)
144 {
145 	int cpu;
146 
147 	for_each_cpu(cpu, mask)
148 		if (pcpu_devices[cpu].address == address)
149 			return pcpu_devices + cpu;
150 	return NULL;
151 }
152 
153 static void pcpu_ec_call(struct pcpu *pcpu, int ec_bit)
154 {
155 	int order;
156 
157 	set_bit(ec_bit, &pcpu->ec_mask);
158 	order = pcpu_running(pcpu) ?
159 		SIGP_EXTERNAL_CALL : SIGP_EMERGENCY_SIGNAL;
160 	pcpu_sigp_retry(pcpu, order, 0);
161 }
162 
163 static int __cpuinit pcpu_alloc_lowcore(struct pcpu *pcpu, int cpu)
164 {
165 	struct _lowcore *lc;
166 
167 	if (pcpu != &pcpu_devices[0]) {
168 		pcpu->lowcore =	(struct _lowcore *)
169 			__get_free_pages(GFP_KERNEL | GFP_DMA, LC_ORDER);
170 		pcpu->async_stack = __get_free_pages(GFP_KERNEL, ASYNC_ORDER);
171 		pcpu->panic_stack = __get_free_page(GFP_KERNEL);
172 		if (!pcpu->lowcore || !pcpu->panic_stack || !pcpu->async_stack)
173 			goto out;
174 	}
175 	lc = pcpu->lowcore;
176 	memcpy(lc, &S390_lowcore, 512);
177 	memset((char *) lc + 512, 0, sizeof(*lc) - 512);
178 	lc->async_stack = pcpu->async_stack + ASYNC_SIZE;
179 	lc->panic_stack = pcpu->panic_stack + PAGE_SIZE;
180 	lc->cpu_nr = cpu;
181 #ifndef CONFIG_64BIT
182 	if (MACHINE_HAS_IEEE) {
183 		lc->extended_save_area_addr = get_zeroed_page(GFP_KERNEL);
184 		if (!lc->extended_save_area_addr)
185 			goto out;
186 	}
187 #else
188 	if (vdso_alloc_per_cpu(lc))
189 		goto out;
190 #endif
191 	lowcore_ptr[cpu] = lc;
192 	pcpu_sigp_retry(pcpu, SIGP_SET_PREFIX, (u32)(unsigned long) lc);
193 	return 0;
194 out:
195 	if (pcpu != &pcpu_devices[0]) {
196 		free_page(pcpu->panic_stack);
197 		free_pages(pcpu->async_stack, ASYNC_ORDER);
198 		free_pages((unsigned long) pcpu->lowcore, LC_ORDER);
199 	}
200 	return -ENOMEM;
201 }
202 
203 #ifdef CONFIG_HOTPLUG_CPU
204 
205 static void pcpu_free_lowcore(struct pcpu *pcpu)
206 {
207 	pcpu_sigp_retry(pcpu, SIGP_SET_PREFIX, 0);
208 	lowcore_ptr[pcpu - pcpu_devices] = NULL;
209 #ifndef CONFIG_64BIT
210 	if (MACHINE_HAS_IEEE) {
211 		struct _lowcore *lc = pcpu->lowcore;
212 
213 		free_page((unsigned long) lc->extended_save_area_addr);
214 		lc->extended_save_area_addr = 0;
215 	}
216 #else
217 	vdso_free_per_cpu(pcpu->lowcore);
218 #endif
219 	if (pcpu != &pcpu_devices[0]) {
220 		free_page(pcpu->panic_stack);
221 		free_pages(pcpu->async_stack, ASYNC_ORDER);
222 		free_pages((unsigned long) pcpu->lowcore, LC_ORDER);
223 	}
224 }
225 
226 #endif /* CONFIG_HOTPLUG_CPU */
227 
228 static void pcpu_prepare_secondary(struct pcpu *pcpu, int cpu)
229 {
230 	struct _lowcore *lc = pcpu->lowcore;
231 
232 	atomic_inc(&init_mm.context.attach_count);
233 	lc->cpu_nr = cpu;
234 	lc->percpu_offset = __per_cpu_offset[cpu];
235 	lc->kernel_asce = S390_lowcore.kernel_asce;
236 	lc->machine_flags = S390_lowcore.machine_flags;
237 	lc->ftrace_func = S390_lowcore.ftrace_func;
238 	lc->user_timer = lc->system_timer = lc->steal_timer = 0;
239 	__ctl_store(lc->cregs_save_area, 0, 15);
240 	save_access_regs((unsigned int *) lc->access_regs_save_area);
241 	memcpy(lc->stfle_fac_list, S390_lowcore.stfle_fac_list,
242 	       MAX_FACILITY_BIT/8);
243 }
244 
245 static void pcpu_attach_task(struct pcpu *pcpu, struct task_struct *tsk)
246 {
247 	struct _lowcore *lc = pcpu->lowcore;
248 	struct thread_info *ti = task_thread_info(tsk);
249 
250 	lc->kernel_stack = (unsigned long) task_stack_page(tsk) + THREAD_SIZE;
251 	lc->thread_info = (unsigned long) task_thread_info(tsk);
252 	lc->current_task = (unsigned long) tsk;
253 	lc->user_timer = ti->user_timer;
254 	lc->system_timer = ti->system_timer;
255 	lc->steal_timer = 0;
256 }
257 
258 static void pcpu_start_fn(struct pcpu *pcpu, void (*func)(void *), void *data)
259 {
260 	struct _lowcore *lc = pcpu->lowcore;
261 
262 	lc->restart_stack = lc->kernel_stack;
263 	lc->restart_fn = (unsigned long) func;
264 	lc->restart_data = (unsigned long) data;
265 	lc->restart_source = -1UL;
266 	pcpu_sigp_retry(pcpu, SIGP_RESTART, 0);
267 }
268 
269 /*
270  * Call function via PSW restart on pcpu and stop the current cpu.
271  */
272 static void pcpu_delegate(struct pcpu *pcpu, void (*func)(void *),
273 			  void *data, unsigned long stack)
274 {
275 	struct _lowcore *lc = lowcore_ptr[pcpu - pcpu_devices];
276 	unsigned long source_cpu = stap();
277 
278 	__load_psw_mask(psw_kernel_bits);
279 	if (pcpu->address == source_cpu)
280 		func(data);	/* should not return */
281 	/* Stop target cpu (if func returns this stops the current cpu). */
282 	pcpu_sigp_retry(pcpu, SIGP_STOP, 0);
283 	/* Restart func on the target cpu and stop the current cpu. */
284 	mem_assign_absolute(lc->restart_stack, stack);
285 	mem_assign_absolute(lc->restart_fn, (unsigned long) func);
286 	mem_assign_absolute(lc->restart_data, (unsigned long) data);
287 	mem_assign_absolute(lc->restart_source, source_cpu);
288 	asm volatile(
289 		"0:	sigp	0,%0,%2	# sigp restart to target cpu\n"
290 		"	brc	2,0b	# busy, try again\n"
291 		"1:	sigp	0,%1,%3	# sigp stop to current cpu\n"
292 		"	brc	2,1b	# busy, try again\n"
293 		: : "d" (pcpu->address), "d" (source_cpu),
294 		    "K" (SIGP_RESTART), "K" (SIGP_STOP)
295 		: "0", "1", "cc");
296 	for (;;) ;
297 }
298 
299 /*
300  * Call function on an online CPU.
301  */
302 void smp_call_online_cpu(void (*func)(void *), void *data)
303 {
304 	struct pcpu *pcpu;
305 
306 	/* Use the current cpu if it is online. */
307 	pcpu = pcpu_find_address(cpu_online_mask, stap());
308 	if (!pcpu)
309 		/* Use the first online cpu. */
310 		pcpu = pcpu_devices + cpumask_first(cpu_online_mask);
311 	pcpu_delegate(pcpu, func, data, (unsigned long) restart_stack);
312 }
313 
314 /*
315  * Call function on the ipl CPU.
316  */
317 void smp_call_ipl_cpu(void (*func)(void *), void *data)
318 {
319 	pcpu_delegate(&pcpu_devices[0], func, data,
320 		      pcpu_devices->panic_stack + PAGE_SIZE);
321 }
322 
323 int smp_find_processor_id(u16 address)
324 {
325 	int cpu;
326 
327 	for_each_present_cpu(cpu)
328 		if (pcpu_devices[cpu].address == address)
329 			return cpu;
330 	return -1;
331 }
332 
333 int smp_vcpu_scheduled(int cpu)
334 {
335 	return pcpu_running(pcpu_devices + cpu);
336 }
337 
338 void smp_yield(void)
339 {
340 	if (MACHINE_HAS_DIAG44)
341 		asm volatile("diag 0,0,0x44");
342 }
343 
344 void smp_yield_cpu(int cpu)
345 {
346 	if (MACHINE_HAS_DIAG9C)
347 		asm volatile("diag %0,0,0x9c"
348 			     : : "d" (pcpu_devices[cpu].address));
349 	else if (MACHINE_HAS_DIAG44)
350 		asm volatile("diag 0,0,0x44");
351 }
352 
353 /*
354  * Send cpus emergency shutdown signal. This gives the cpus the
355  * opportunity to complete outstanding interrupts.
356  */
357 void smp_emergency_stop(cpumask_t *cpumask)
358 {
359 	u64 end;
360 	int cpu;
361 
362 	end = get_clock() + (1000000UL << 12);
363 	for_each_cpu(cpu, cpumask) {
364 		struct pcpu *pcpu = pcpu_devices + cpu;
365 		set_bit(ec_stop_cpu, &pcpu->ec_mask);
366 		while (__pcpu_sigp(pcpu->address, SIGP_EMERGENCY_SIGNAL,
367 				   0, NULL) == SIGP_CC_BUSY &&
368 		       get_clock() < end)
369 			cpu_relax();
370 	}
371 	while (get_clock() < end) {
372 		for_each_cpu(cpu, cpumask)
373 			if (pcpu_stopped(pcpu_devices + cpu))
374 				cpumask_clear_cpu(cpu, cpumask);
375 		if (cpumask_empty(cpumask))
376 			break;
377 		cpu_relax();
378 	}
379 }
380 
381 /*
382  * Stop all cpus but the current one.
383  */
384 void smp_send_stop(void)
385 {
386 	cpumask_t cpumask;
387 	int cpu;
388 
389 	/* Disable all interrupts/machine checks */
390 	__load_psw_mask(psw_kernel_bits | PSW_MASK_DAT);
391 	trace_hardirqs_off();
392 
393 	debug_set_critical();
394 	cpumask_copy(&cpumask, cpu_online_mask);
395 	cpumask_clear_cpu(smp_processor_id(), &cpumask);
396 
397 	if (oops_in_progress)
398 		smp_emergency_stop(&cpumask);
399 
400 	/* stop all processors */
401 	for_each_cpu(cpu, &cpumask) {
402 		struct pcpu *pcpu = pcpu_devices + cpu;
403 		pcpu_sigp_retry(pcpu, SIGP_STOP, 0);
404 		while (!pcpu_stopped(pcpu))
405 			cpu_relax();
406 	}
407 }
408 
409 /*
410  * Stop the current cpu.
411  */
412 void smp_stop_cpu(void)
413 {
414 	pcpu_sigp_retry(pcpu_devices + smp_processor_id(), SIGP_STOP, 0);
415 	for (;;) ;
416 }
417 
418 /*
419  * This is the main routine where commands issued by other
420  * cpus are handled.
421  */
422 static void do_ext_call_interrupt(struct ext_code ext_code,
423 				  unsigned int param32, unsigned long param64)
424 {
425 	unsigned long bits;
426 	int cpu;
427 
428 	cpu = smp_processor_id();
429 	if (ext_code.code == 0x1202)
430 		kstat_cpu(cpu).irqs[EXTINT_EXC]++;
431 	else
432 		kstat_cpu(cpu).irqs[EXTINT_EMS]++;
433 	/*
434 	 * handle bit signal external calls
435 	 */
436 	bits = xchg(&pcpu_devices[cpu].ec_mask, 0);
437 
438 	if (test_bit(ec_stop_cpu, &bits))
439 		smp_stop_cpu();
440 
441 	if (test_bit(ec_schedule, &bits))
442 		scheduler_ipi();
443 
444 	if (test_bit(ec_call_function, &bits))
445 		generic_smp_call_function_interrupt();
446 
447 	if (test_bit(ec_call_function_single, &bits))
448 		generic_smp_call_function_single_interrupt();
449 
450 }
451 
452 void arch_send_call_function_ipi_mask(const struct cpumask *mask)
453 {
454 	int cpu;
455 
456 	for_each_cpu(cpu, mask)
457 		pcpu_ec_call(pcpu_devices + cpu, ec_call_function);
458 }
459 
460 void arch_send_call_function_single_ipi(int cpu)
461 {
462 	pcpu_ec_call(pcpu_devices + cpu, ec_call_function_single);
463 }
464 
465 #ifndef CONFIG_64BIT
466 /*
467  * this function sends a 'purge tlb' signal to another CPU.
468  */
469 static void smp_ptlb_callback(void *info)
470 {
471 	__tlb_flush_local();
472 }
473 
474 void smp_ptlb_all(void)
475 {
476 	on_each_cpu(smp_ptlb_callback, NULL, 1);
477 }
478 EXPORT_SYMBOL(smp_ptlb_all);
479 #endif /* ! CONFIG_64BIT */
480 
481 /*
482  * this function sends a 'reschedule' IPI to another CPU.
483  * it goes straight through and wastes no time serializing
484  * anything. Worst case is that we lose a reschedule ...
485  */
486 void smp_send_reschedule(int cpu)
487 {
488 	pcpu_ec_call(pcpu_devices + cpu, ec_schedule);
489 }
490 
491 /*
492  * parameter area for the set/clear control bit callbacks
493  */
494 struct ec_creg_mask_parms {
495 	unsigned long orval;
496 	unsigned long andval;
497 	int cr;
498 };
499 
500 /*
501  * callback for setting/clearing control bits
502  */
503 static void smp_ctl_bit_callback(void *info)
504 {
505 	struct ec_creg_mask_parms *pp = info;
506 	unsigned long cregs[16];
507 
508 	__ctl_store(cregs, 0, 15);
509 	cregs[pp->cr] = (cregs[pp->cr] & pp->andval) | pp->orval;
510 	__ctl_load(cregs, 0, 15);
511 }
512 
513 /*
514  * Set a bit in a control register of all cpus
515  */
516 void smp_ctl_set_bit(int cr, int bit)
517 {
518 	struct ec_creg_mask_parms parms = { 1UL << bit, -1UL, cr };
519 
520 	on_each_cpu(smp_ctl_bit_callback, &parms, 1);
521 }
522 EXPORT_SYMBOL(smp_ctl_set_bit);
523 
524 /*
525  * Clear a bit in a control register of all cpus
526  */
527 void smp_ctl_clear_bit(int cr, int bit)
528 {
529 	struct ec_creg_mask_parms parms = { 0, ~(1UL << bit), cr };
530 
531 	on_each_cpu(smp_ctl_bit_callback, &parms, 1);
532 }
533 EXPORT_SYMBOL(smp_ctl_clear_bit);
534 
535 #if defined(CONFIG_ZFCPDUMP) || defined(CONFIG_CRASH_DUMP)
536 
537 struct save_area *zfcpdump_save_areas[NR_CPUS + 1];
538 EXPORT_SYMBOL_GPL(zfcpdump_save_areas);
539 
540 static void __init smp_get_save_area(int cpu, u16 address)
541 {
542 	void *lc = pcpu_devices[0].lowcore;
543 	struct save_area *save_area;
544 
545 	if (is_kdump_kernel())
546 		return;
547 	if (!OLDMEM_BASE && (address == boot_cpu_address ||
548 			     ipl_info.type != IPL_TYPE_FCP_DUMP))
549 		return;
550 	if (cpu >= NR_CPUS) {
551 		pr_warning("CPU %i exceeds the maximum %i and is excluded "
552 			   "from the dump\n", cpu, NR_CPUS - 1);
553 		return;
554 	}
555 	save_area = kmalloc(sizeof(struct save_area), GFP_KERNEL);
556 	if (!save_area)
557 		panic("could not allocate memory for save area\n");
558 	zfcpdump_save_areas[cpu] = save_area;
559 #ifdef CONFIG_CRASH_DUMP
560 	if (address == boot_cpu_address) {
561 		/* Copy the registers of the boot cpu. */
562 		copy_oldmem_page(1, (void *) save_area, sizeof(*save_area),
563 				 SAVE_AREA_BASE - PAGE_SIZE, 0);
564 		return;
565 	}
566 #endif
567 	/* Get the registers of a non-boot cpu. */
568 	__pcpu_sigp_relax(address, SIGP_STOP_AND_STORE_STATUS, 0, NULL);
569 	memcpy_real(save_area, lc + SAVE_AREA_BASE, sizeof(*save_area));
570 }
571 
572 int smp_store_status(int cpu)
573 {
574 	struct pcpu *pcpu;
575 
576 	pcpu = pcpu_devices + cpu;
577 	if (__pcpu_sigp_relax(pcpu->address, SIGP_STOP_AND_STORE_STATUS,
578 			      0, NULL) != SIGP_CC_ORDER_CODE_ACCEPTED)
579 		return -EIO;
580 	return 0;
581 }
582 
583 #else /* CONFIG_ZFCPDUMP || CONFIG_CRASH_DUMP */
584 
585 static inline void smp_get_save_area(int cpu, u16 address) { }
586 
587 #endif /* CONFIG_ZFCPDUMP || CONFIG_CRASH_DUMP */
588 
589 static struct sclp_cpu_info *smp_get_cpu_info(void)
590 {
591 	static int use_sigp_detection;
592 	struct sclp_cpu_info *info;
593 	int address;
594 
595 	info = kzalloc(sizeof(*info), GFP_KERNEL);
596 	if (info && (use_sigp_detection || sclp_get_cpu_info(info))) {
597 		use_sigp_detection = 1;
598 		for (address = 0; address <= MAX_CPU_ADDRESS; address++) {
599 			if (__pcpu_sigp_relax(address, SIGP_SENSE, 0, NULL) ==
600 			    SIGP_CC_NOT_OPERATIONAL)
601 				continue;
602 			info->cpu[info->configured].address = address;
603 			info->configured++;
604 		}
605 		info->combined = info->configured;
606 	}
607 	return info;
608 }
609 
610 static int __devinit smp_add_present_cpu(int cpu);
611 
612 static int __devinit __smp_rescan_cpus(struct sclp_cpu_info *info,
613 				       int sysfs_add)
614 {
615 	struct pcpu *pcpu;
616 	cpumask_t avail;
617 	int cpu, nr, i;
618 
619 	nr = 0;
620 	cpumask_xor(&avail, cpu_possible_mask, cpu_present_mask);
621 	cpu = cpumask_first(&avail);
622 	for (i = 0; (i < info->combined) && (cpu < nr_cpu_ids); i++) {
623 		if (info->has_cpu_type && info->cpu[i].type != boot_cpu_type)
624 			continue;
625 		if (pcpu_find_address(cpu_present_mask, info->cpu[i].address))
626 			continue;
627 		pcpu = pcpu_devices + cpu;
628 		pcpu->address = info->cpu[i].address;
629 		pcpu->state = (cpu >= info->configured) ?
630 			CPU_STATE_STANDBY : CPU_STATE_CONFIGURED;
631 		cpu_set_polarization(cpu, POLARIZATION_UNKNOWN);
632 		set_cpu_present(cpu, true);
633 		if (sysfs_add && smp_add_present_cpu(cpu) != 0)
634 			set_cpu_present(cpu, false);
635 		else
636 			nr++;
637 		cpu = cpumask_next(cpu, &avail);
638 	}
639 	return nr;
640 }
641 
642 static void __init smp_detect_cpus(void)
643 {
644 	unsigned int cpu, c_cpus, s_cpus;
645 	struct sclp_cpu_info *info;
646 
647 	info = smp_get_cpu_info();
648 	if (!info)
649 		panic("smp_detect_cpus failed to allocate memory\n");
650 	if (info->has_cpu_type) {
651 		for (cpu = 0; cpu < info->combined; cpu++) {
652 			if (info->cpu[cpu].address != boot_cpu_address)
653 				continue;
654 			/* The boot cpu dictates the cpu type. */
655 			boot_cpu_type = info->cpu[cpu].type;
656 			break;
657 		}
658 	}
659 	c_cpus = s_cpus = 0;
660 	for (cpu = 0; cpu < info->combined; cpu++) {
661 		if (info->has_cpu_type && info->cpu[cpu].type != boot_cpu_type)
662 			continue;
663 		if (cpu < info->configured) {
664 			smp_get_save_area(c_cpus, info->cpu[cpu].address);
665 			c_cpus++;
666 		} else
667 			s_cpus++;
668 	}
669 	pr_info("%d configured CPUs, %d standby CPUs\n", c_cpus, s_cpus);
670 	get_online_cpus();
671 	__smp_rescan_cpus(info, 0);
672 	put_online_cpus();
673 	kfree(info);
674 }
675 
676 /*
677  *	Activate a secondary processor.
678  */
679 static void __cpuinit smp_start_secondary(void *cpuvoid)
680 {
681 	S390_lowcore.last_update_clock = get_clock();
682 	S390_lowcore.restart_stack = (unsigned long) restart_stack;
683 	S390_lowcore.restart_fn = (unsigned long) do_restart;
684 	S390_lowcore.restart_data = 0;
685 	S390_lowcore.restart_source = -1UL;
686 	restore_access_regs(S390_lowcore.access_regs_save_area);
687 	__ctl_load(S390_lowcore.cregs_save_area, 0, 15);
688 	__load_psw_mask(psw_kernel_bits | PSW_MASK_DAT);
689 	cpu_init();
690 	preempt_disable();
691 	init_cpu_timer();
692 	init_cpu_vtimer();
693 	pfault_init();
694 	notify_cpu_starting(smp_processor_id());
695 	set_cpu_online(smp_processor_id(), true);
696 	local_irq_enable();
697 	/* cpu_idle will call schedule for us */
698 	cpu_idle();
699 }
700 
701 /* Upping and downing of CPUs */
702 int __cpuinit __cpu_up(unsigned int cpu, struct task_struct *tidle)
703 {
704 	struct pcpu *pcpu;
705 	int rc;
706 
707 	pcpu = pcpu_devices + cpu;
708 	if (pcpu->state != CPU_STATE_CONFIGURED)
709 		return -EIO;
710 	if (pcpu_sigp_retry(pcpu, SIGP_INITIAL_CPU_RESET, 0) !=
711 	    SIGP_CC_ORDER_CODE_ACCEPTED)
712 		return -EIO;
713 
714 	rc = pcpu_alloc_lowcore(pcpu, cpu);
715 	if (rc)
716 		return rc;
717 	pcpu_prepare_secondary(pcpu, cpu);
718 	pcpu_attach_task(pcpu, tidle);
719 	pcpu_start_fn(pcpu, smp_start_secondary, NULL);
720 	while (!cpu_online(cpu))
721 		cpu_relax();
722 	return 0;
723 }
724 
725 static int __init setup_possible_cpus(char *s)
726 {
727 	int max, cpu;
728 
729 	if (kstrtoint(s, 0, &max) < 0)
730 		return 0;
731 	init_cpu_possible(cpumask_of(0));
732 	for (cpu = 1; cpu < max && cpu < nr_cpu_ids; cpu++)
733 		set_cpu_possible(cpu, true);
734 	return 0;
735 }
736 early_param("possible_cpus", setup_possible_cpus);
737 
738 #ifdef CONFIG_HOTPLUG_CPU
739 
740 int __cpu_disable(void)
741 {
742 	unsigned long cregs[16];
743 
744 	set_cpu_online(smp_processor_id(), false);
745 	/* Disable pseudo page faults on this cpu. */
746 	pfault_fini();
747 	/* Disable interrupt sources via control register. */
748 	__ctl_store(cregs, 0, 15);
749 	cregs[0]  &= ~0x0000ee70UL;	/* disable all external interrupts */
750 	cregs[6]  &= ~0xff000000UL;	/* disable all I/O interrupts */
751 	cregs[14] &= ~0x1f000000UL;	/* disable most machine checks */
752 	__ctl_load(cregs, 0, 15);
753 	return 0;
754 }
755 
756 void __cpu_die(unsigned int cpu)
757 {
758 	struct pcpu *pcpu;
759 
760 	/* Wait until target cpu is down */
761 	pcpu = pcpu_devices + cpu;
762 	while (!pcpu_stopped(pcpu))
763 		cpu_relax();
764 	pcpu_free_lowcore(pcpu);
765 	atomic_dec(&init_mm.context.attach_count);
766 }
767 
768 void __noreturn cpu_die(void)
769 {
770 	idle_task_exit();
771 	pcpu_sigp_retry(pcpu_devices + smp_processor_id(), SIGP_STOP, 0);
772 	for (;;) ;
773 }
774 
775 #endif /* CONFIG_HOTPLUG_CPU */
776 
777 void __init smp_prepare_cpus(unsigned int max_cpus)
778 {
779 	/* request the 0x1201 emergency signal external interrupt */
780 	if (register_external_interrupt(0x1201, do_ext_call_interrupt) != 0)
781 		panic("Couldn't request external interrupt 0x1201");
782 	/* request the 0x1202 external call external interrupt */
783 	if (register_external_interrupt(0x1202, do_ext_call_interrupt) != 0)
784 		panic("Couldn't request external interrupt 0x1202");
785 	smp_detect_cpus();
786 }
787 
788 void __init smp_prepare_boot_cpu(void)
789 {
790 	struct pcpu *pcpu = pcpu_devices;
791 
792 	boot_cpu_address = stap();
793 	pcpu->state = CPU_STATE_CONFIGURED;
794 	pcpu->address = boot_cpu_address;
795 	pcpu->lowcore = (struct _lowcore *)(unsigned long) store_prefix();
796 	pcpu->async_stack = S390_lowcore.async_stack - ASYNC_SIZE;
797 	pcpu->panic_stack = S390_lowcore.panic_stack - PAGE_SIZE;
798 	S390_lowcore.percpu_offset = __per_cpu_offset[0];
799 	cpu_set_polarization(0, POLARIZATION_UNKNOWN);
800 	set_cpu_present(0, true);
801 	set_cpu_online(0, true);
802 }
803 
804 void __init smp_cpus_done(unsigned int max_cpus)
805 {
806 }
807 
808 void __init smp_setup_processor_id(void)
809 {
810 	S390_lowcore.cpu_nr = 0;
811 }
812 
813 /*
814  * the frequency of the profiling timer can be changed
815  * by writing a multiplier value into /proc/profile.
816  *
817  * usually you want to run this on all CPUs ;)
818  */
819 int setup_profiling_timer(unsigned int multiplier)
820 {
821 	return 0;
822 }
823 
824 #ifdef CONFIG_HOTPLUG_CPU
825 static ssize_t cpu_configure_show(struct device *dev,
826 				  struct device_attribute *attr, char *buf)
827 {
828 	ssize_t count;
829 
830 	mutex_lock(&smp_cpu_state_mutex);
831 	count = sprintf(buf, "%d\n", pcpu_devices[dev->id].state);
832 	mutex_unlock(&smp_cpu_state_mutex);
833 	return count;
834 }
835 
836 static ssize_t cpu_configure_store(struct device *dev,
837 				   struct device_attribute *attr,
838 				   const char *buf, size_t count)
839 {
840 	struct pcpu *pcpu;
841 	int cpu, val, rc;
842 	char delim;
843 
844 	if (sscanf(buf, "%d %c", &val, &delim) != 1)
845 		return -EINVAL;
846 	if (val != 0 && val != 1)
847 		return -EINVAL;
848 	get_online_cpus();
849 	mutex_lock(&smp_cpu_state_mutex);
850 	rc = -EBUSY;
851 	/* disallow configuration changes of online cpus and cpu 0 */
852 	cpu = dev->id;
853 	if (cpu_online(cpu) || cpu == 0)
854 		goto out;
855 	pcpu = pcpu_devices + cpu;
856 	rc = 0;
857 	switch (val) {
858 	case 0:
859 		if (pcpu->state != CPU_STATE_CONFIGURED)
860 			break;
861 		rc = sclp_cpu_deconfigure(pcpu->address);
862 		if (rc)
863 			break;
864 		pcpu->state = CPU_STATE_STANDBY;
865 		cpu_set_polarization(cpu, POLARIZATION_UNKNOWN);
866 		topology_expect_change();
867 		break;
868 	case 1:
869 		if (pcpu->state != CPU_STATE_STANDBY)
870 			break;
871 		rc = sclp_cpu_configure(pcpu->address);
872 		if (rc)
873 			break;
874 		pcpu->state = CPU_STATE_CONFIGURED;
875 		cpu_set_polarization(cpu, POLARIZATION_UNKNOWN);
876 		topology_expect_change();
877 		break;
878 	default:
879 		break;
880 	}
881 out:
882 	mutex_unlock(&smp_cpu_state_mutex);
883 	put_online_cpus();
884 	return rc ? rc : count;
885 }
886 static DEVICE_ATTR(configure, 0644, cpu_configure_show, cpu_configure_store);
887 #endif /* CONFIG_HOTPLUG_CPU */
888 
889 static ssize_t show_cpu_address(struct device *dev,
890 				struct device_attribute *attr, char *buf)
891 {
892 	return sprintf(buf, "%d\n", pcpu_devices[dev->id].address);
893 }
894 static DEVICE_ATTR(address, 0444, show_cpu_address, NULL);
895 
896 static struct attribute *cpu_common_attrs[] = {
897 #ifdef CONFIG_HOTPLUG_CPU
898 	&dev_attr_configure.attr,
899 #endif
900 	&dev_attr_address.attr,
901 	NULL,
902 };
903 
904 static struct attribute_group cpu_common_attr_group = {
905 	.attrs = cpu_common_attrs,
906 };
907 
908 static ssize_t show_idle_count(struct device *dev,
909 				struct device_attribute *attr, char *buf)
910 {
911 	struct s390_idle_data *idle = &per_cpu(s390_idle, dev->id);
912 	unsigned long long idle_count;
913 	unsigned int sequence;
914 
915 	do {
916 		sequence = ACCESS_ONCE(idle->sequence);
917 		idle_count = ACCESS_ONCE(idle->idle_count);
918 		if (ACCESS_ONCE(idle->clock_idle_enter))
919 			idle_count++;
920 	} while ((sequence & 1) || (idle->sequence != sequence));
921 	return sprintf(buf, "%llu\n", idle_count);
922 }
923 static DEVICE_ATTR(idle_count, 0444, show_idle_count, NULL);
924 
925 static ssize_t show_idle_time(struct device *dev,
926 				struct device_attribute *attr, char *buf)
927 {
928 	struct s390_idle_data *idle = &per_cpu(s390_idle, dev->id);
929 	unsigned long long now, idle_time, idle_enter, idle_exit;
930 	unsigned int sequence;
931 
932 	do {
933 		now = get_clock();
934 		sequence = ACCESS_ONCE(idle->sequence);
935 		idle_time = ACCESS_ONCE(idle->idle_time);
936 		idle_enter = ACCESS_ONCE(idle->clock_idle_enter);
937 		idle_exit = ACCESS_ONCE(idle->clock_idle_exit);
938 	} while ((sequence & 1) || (idle->sequence != sequence));
939 	idle_time += idle_enter ? ((idle_exit ? : now) - idle_enter) : 0;
940 	return sprintf(buf, "%llu\n", idle_time >> 12);
941 }
942 static DEVICE_ATTR(idle_time_us, 0444, show_idle_time, NULL);
943 
944 static struct attribute *cpu_online_attrs[] = {
945 	&dev_attr_idle_count.attr,
946 	&dev_attr_idle_time_us.attr,
947 	NULL,
948 };
949 
950 static struct attribute_group cpu_online_attr_group = {
951 	.attrs = cpu_online_attrs,
952 };
953 
954 static int __cpuinit smp_cpu_notify(struct notifier_block *self,
955 				    unsigned long action, void *hcpu)
956 {
957 	unsigned int cpu = (unsigned int)(long)hcpu;
958 	struct cpu *c = &pcpu_devices[cpu].cpu;
959 	struct device *s = &c->dev;
960 	int err = 0;
961 
962 	switch (action) {
963 	case CPU_ONLINE:
964 	case CPU_ONLINE_FROZEN:
965 		err = sysfs_create_group(&s->kobj, &cpu_online_attr_group);
966 		break;
967 	case CPU_DEAD:
968 	case CPU_DEAD_FROZEN:
969 		sysfs_remove_group(&s->kobj, &cpu_online_attr_group);
970 		break;
971 	}
972 	return notifier_from_errno(err);
973 }
974 
975 static struct notifier_block __cpuinitdata smp_cpu_nb = {
976 	.notifier_call = smp_cpu_notify,
977 };
978 
979 static int __devinit smp_add_present_cpu(int cpu)
980 {
981 	struct cpu *c = &pcpu_devices[cpu].cpu;
982 	struct device *s = &c->dev;
983 	int rc;
984 
985 	c->hotpluggable = 1;
986 	rc = register_cpu(c, cpu);
987 	if (rc)
988 		goto out;
989 	rc = sysfs_create_group(&s->kobj, &cpu_common_attr_group);
990 	if (rc)
991 		goto out_cpu;
992 	if (cpu_online(cpu)) {
993 		rc = sysfs_create_group(&s->kobj, &cpu_online_attr_group);
994 		if (rc)
995 			goto out_online;
996 	}
997 	rc = topology_cpu_init(c);
998 	if (rc)
999 		goto out_topology;
1000 	return 0;
1001 
1002 out_topology:
1003 	if (cpu_online(cpu))
1004 		sysfs_remove_group(&s->kobj, &cpu_online_attr_group);
1005 out_online:
1006 	sysfs_remove_group(&s->kobj, &cpu_common_attr_group);
1007 out_cpu:
1008 #ifdef CONFIG_HOTPLUG_CPU
1009 	unregister_cpu(c);
1010 #endif
1011 out:
1012 	return rc;
1013 }
1014 
1015 #ifdef CONFIG_HOTPLUG_CPU
1016 
1017 int __ref smp_rescan_cpus(void)
1018 {
1019 	struct sclp_cpu_info *info;
1020 	int nr;
1021 
1022 	info = smp_get_cpu_info();
1023 	if (!info)
1024 		return -ENOMEM;
1025 	get_online_cpus();
1026 	mutex_lock(&smp_cpu_state_mutex);
1027 	nr = __smp_rescan_cpus(info, 1);
1028 	mutex_unlock(&smp_cpu_state_mutex);
1029 	put_online_cpus();
1030 	kfree(info);
1031 	if (nr)
1032 		topology_schedule_update();
1033 	return 0;
1034 }
1035 
1036 static ssize_t __ref rescan_store(struct device *dev,
1037 				  struct device_attribute *attr,
1038 				  const char *buf,
1039 				  size_t count)
1040 {
1041 	int rc;
1042 
1043 	rc = smp_rescan_cpus();
1044 	return rc ? rc : count;
1045 }
1046 static DEVICE_ATTR(rescan, 0200, NULL, rescan_store);
1047 #endif /* CONFIG_HOTPLUG_CPU */
1048 
1049 static int __init s390_smp_init(void)
1050 {
1051 	int cpu, rc;
1052 
1053 	register_cpu_notifier(&smp_cpu_nb);
1054 #ifdef CONFIG_HOTPLUG_CPU
1055 	rc = device_create_file(cpu_subsys.dev_root, &dev_attr_rescan);
1056 	if (rc)
1057 		return rc;
1058 #endif
1059 	for_each_present_cpu(cpu) {
1060 		rc = smp_add_present_cpu(cpu);
1061 		if (rc)
1062 			return rc;
1063 	}
1064 	return 0;
1065 }
1066 subsys_initcall(s390_smp_init);
1067