xref: /linux/arch/s390/kernel/smp.c (revision 26b0d14106954ae46d2f4f7eec3481828a210f7d)
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/timer.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 "entry.h"
48 
49 enum {
50 	sigp_sense = 1,
51 	sigp_external_call = 2,
52 	sigp_emergency_signal = 3,
53 	sigp_start = 4,
54 	sigp_stop = 5,
55 	sigp_restart = 6,
56 	sigp_stop_and_store_status = 9,
57 	sigp_initial_cpu_reset = 11,
58 	sigp_cpu_reset = 12,
59 	sigp_set_prefix = 13,
60 	sigp_store_status_at_address = 14,
61 	sigp_store_extended_status_at_address = 15,
62 	sigp_set_architecture = 18,
63 	sigp_conditional_emergency_signal = 19,
64 	sigp_sense_running = 21,
65 };
66 
67 enum {
68 	sigp_order_code_accepted = 0,
69 	sigp_status_stored = 1,
70 	sigp_busy = 2,
71 	sigp_not_operational = 3,
72 };
73 
74 enum {
75 	ec_schedule = 0,
76 	ec_call_function,
77 	ec_call_function_single,
78 	ec_stop_cpu,
79 };
80 
81 enum {
82 	CPU_STATE_STANDBY,
83 	CPU_STATE_CONFIGURED,
84 };
85 
86 struct pcpu {
87 	struct cpu cpu;
88 	struct _lowcore *lowcore;	/* lowcore page(s) for the cpu */
89 	unsigned long async_stack;	/* async stack for the cpu */
90 	unsigned long panic_stack;	/* panic stack for the cpu */
91 	unsigned long ec_mask;		/* bit mask for ec_xxx functions */
92 	int state;			/* physical cpu state */
93 	u32 status;			/* last status received via sigp */
94 	u16 address;			/* physical cpu address */
95 };
96 
97 static u8 boot_cpu_type;
98 static u16 boot_cpu_address;
99 static struct pcpu pcpu_devices[NR_CPUS];
100 
101 DEFINE_MUTEX(smp_cpu_state_mutex);
102 
103 /*
104  * Signal processor helper functions.
105  */
106 static inline int __pcpu_sigp(u16 addr, u8 order, u32 parm, u32 *status)
107 {
108 	register unsigned int reg1 asm ("1") = parm;
109 	int cc;
110 
111 	asm volatile(
112 		"	sigp	%1,%2,0(%3)\n"
113 		"	ipm	%0\n"
114 		"	srl	%0,28\n"
115 		: "=d" (cc), "+d" (reg1) : "d" (addr), "a" (order) : "cc");
116 	if (status && cc == 1)
117 		*status = reg1;
118 	return cc;
119 }
120 
121 static inline int __pcpu_sigp_relax(u16 addr, u8 order, u32 parm, u32 *status)
122 {
123 	int cc;
124 
125 	while (1) {
126 		cc = __pcpu_sigp(addr, order, parm, status);
127 		if (cc != sigp_busy)
128 			return cc;
129 		cpu_relax();
130 	}
131 }
132 
133 static int pcpu_sigp_retry(struct pcpu *pcpu, u8 order, u32 parm)
134 {
135 	int cc, retry;
136 
137 	for (retry = 0; ; retry++) {
138 		cc = __pcpu_sigp(pcpu->address, order, parm, &pcpu->status);
139 		if (cc != sigp_busy)
140 			break;
141 		if (retry >= 3)
142 			udelay(10);
143 	}
144 	return cc;
145 }
146 
147 static inline int pcpu_stopped(struct pcpu *pcpu)
148 {
149 	if (__pcpu_sigp(pcpu->address, sigp_sense,
150 			0, &pcpu->status) != sigp_status_stored)
151 		return 0;
152 	/* Check for stopped and check stop state */
153 	return !!(pcpu->status & 0x50);
154 }
155 
156 static inline int pcpu_running(struct pcpu *pcpu)
157 {
158 	if (__pcpu_sigp(pcpu->address, sigp_sense_running,
159 			0, &pcpu->status) != sigp_status_stored)
160 		return 1;
161 	/* Check for running status */
162 	return !(pcpu->status & 0x400);
163 }
164 
165 /*
166  * Find struct pcpu by cpu address.
167  */
168 static struct pcpu *pcpu_find_address(const struct cpumask *mask, int address)
169 {
170 	int cpu;
171 
172 	for_each_cpu(cpu, mask)
173 		if (pcpu_devices[cpu].address == address)
174 			return pcpu_devices + cpu;
175 	return NULL;
176 }
177 
178 static void pcpu_ec_call(struct pcpu *pcpu, int ec_bit)
179 {
180 	int order;
181 
182 	set_bit(ec_bit, &pcpu->ec_mask);
183 	order = pcpu_running(pcpu) ?
184 		sigp_external_call : sigp_emergency_signal;
185 	pcpu_sigp_retry(pcpu, order, 0);
186 }
187 
188 static int __cpuinit pcpu_alloc_lowcore(struct pcpu *pcpu, int cpu)
189 {
190 	struct _lowcore *lc;
191 
192 	if (pcpu != &pcpu_devices[0]) {
193 		pcpu->lowcore =	(struct _lowcore *)
194 			__get_free_pages(GFP_KERNEL | GFP_DMA, LC_ORDER);
195 		pcpu->async_stack = __get_free_pages(GFP_KERNEL, ASYNC_ORDER);
196 		pcpu->panic_stack = __get_free_page(GFP_KERNEL);
197 		if (!pcpu->lowcore || !pcpu->panic_stack || !pcpu->async_stack)
198 			goto out;
199 	}
200 	lc = pcpu->lowcore;
201 	memcpy(lc, &S390_lowcore, 512);
202 	memset((char *) lc + 512, 0, sizeof(*lc) - 512);
203 	lc->async_stack = pcpu->async_stack + ASYNC_SIZE;
204 	lc->panic_stack = pcpu->panic_stack + PAGE_SIZE;
205 	lc->cpu_nr = cpu;
206 #ifndef CONFIG_64BIT
207 	if (MACHINE_HAS_IEEE) {
208 		lc->extended_save_area_addr = get_zeroed_page(GFP_KERNEL);
209 		if (!lc->extended_save_area_addr)
210 			goto out;
211 	}
212 #else
213 	if (vdso_alloc_per_cpu(lc))
214 		goto out;
215 #endif
216 	lowcore_ptr[cpu] = lc;
217 	pcpu_sigp_retry(pcpu, sigp_set_prefix, (u32)(unsigned long) lc);
218 	return 0;
219 out:
220 	if (pcpu != &pcpu_devices[0]) {
221 		free_page(pcpu->panic_stack);
222 		free_pages(pcpu->async_stack, ASYNC_ORDER);
223 		free_pages((unsigned long) pcpu->lowcore, LC_ORDER);
224 	}
225 	return -ENOMEM;
226 }
227 
228 #ifdef CONFIG_HOTPLUG_CPU
229 
230 static void pcpu_free_lowcore(struct pcpu *pcpu)
231 {
232 	pcpu_sigp_retry(pcpu, sigp_set_prefix, 0);
233 	lowcore_ptr[pcpu - pcpu_devices] = NULL;
234 #ifndef CONFIG_64BIT
235 	if (MACHINE_HAS_IEEE) {
236 		struct _lowcore *lc = pcpu->lowcore;
237 
238 		free_page((unsigned long) lc->extended_save_area_addr);
239 		lc->extended_save_area_addr = 0;
240 	}
241 #else
242 	vdso_free_per_cpu(pcpu->lowcore);
243 #endif
244 	if (pcpu != &pcpu_devices[0]) {
245 		free_page(pcpu->panic_stack);
246 		free_pages(pcpu->async_stack, ASYNC_ORDER);
247 		free_pages((unsigned long) pcpu->lowcore, LC_ORDER);
248 	}
249 }
250 
251 #endif /* CONFIG_HOTPLUG_CPU */
252 
253 static void pcpu_prepare_secondary(struct pcpu *pcpu, int cpu)
254 {
255 	struct _lowcore *lc = pcpu->lowcore;
256 
257 	atomic_inc(&init_mm.context.attach_count);
258 	lc->cpu_nr = cpu;
259 	lc->percpu_offset = __per_cpu_offset[cpu];
260 	lc->kernel_asce = S390_lowcore.kernel_asce;
261 	lc->machine_flags = S390_lowcore.machine_flags;
262 	lc->ftrace_func = S390_lowcore.ftrace_func;
263 	lc->user_timer = lc->system_timer = lc->steal_timer = 0;
264 	__ctl_store(lc->cregs_save_area, 0, 15);
265 	save_access_regs((unsigned int *) lc->access_regs_save_area);
266 	memcpy(lc->stfle_fac_list, S390_lowcore.stfle_fac_list,
267 	       MAX_FACILITY_BIT/8);
268 }
269 
270 static void pcpu_attach_task(struct pcpu *pcpu, struct task_struct *tsk)
271 {
272 	struct _lowcore *lc = pcpu->lowcore;
273 	struct thread_info *ti = task_thread_info(tsk);
274 
275 	lc->kernel_stack = (unsigned long) task_stack_page(tsk) + THREAD_SIZE;
276 	lc->thread_info = (unsigned long) task_thread_info(tsk);
277 	lc->current_task = (unsigned long) tsk;
278 	lc->user_timer = ti->user_timer;
279 	lc->system_timer = ti->system_timer;
280 	lc->steal_timer = 0;
281 }
282 
283 static void pcpu_start_fn(struct pcpu *pcpu, void (*func)(void *), void *data)
284 {
285 	struct _lowcore *lc = pcpu->lowcore;
286 
287 	lc->restart_stack = lc->kernel_stack;
288 	lc->restart_fn = (unsigned long) func;
289 	lc->restart_data = (unsigned long) data;
290 	lc->restart_source = -1UL;
291 	pcpu_sigp_retry(pcpu, sigp_restart, 0);
292 }
293 
294 /*
295  * Call function via PSW restart on pcpu and stop the current cpu.
296  */
297 static void pcpu_delegate(struct pcpu *pcpu, void (*func)(void *),
298 			  void *data, unsigned long stack)
299 {
300 	struct _lowcore *lc = lowcore_ptr[pcpu - pcpu_devices];
301 	struct {
302 		unsigned long	stack;
303 		void		*func;
304 		void		*data;
305 		unsigned long	source;
306 	} restart = { stack, func, data, stap() };
307 
308 	__load_psw_mask(psw_kernel_bits);
309 	if (pcpu->address == restart.source)
310 		func(data);	/* should not return */
311 	/* Stop target cpu (if func returns this stops the current cpu). */
312 	pcpu_sigp_retry(pcpu, sigp_stop, 0);
313 	/* Restart func on the target cpu and stop the current cpu. */
314 	memcpy_absolute(&lc->restart_stack, &restart, sizeof(restart));
315 	asm volatile(
316 		"0:	sigp	0,%0,6	# sigp restart to target cpu\n"
317 		"	brc	2,0b	# busy, try again\n"
318 		"1:	sigp	0,%1,5	# sigp stop to current cpu\n"
319 		"	brc	2,1b	# busy, try again\n"
320 		: : "d" (pcpu->address), "d" (restart.source) : "0", "1", "cc");
321 	for (;;) ;
322 }
323 
324 /*
325  * Call function on an online CPU.
326  */
327 void smp_call_online_cpu(void (*func)(void *), void *data)
328 {
329 	struct pcpu *pcpu;
330 
331 	/* Use the current cpu if it is online. */
332 	pcpu = pcpu_find_address(cpu_online_mask, stap());
333 	if (!pcpu)
334 		/* Use the first online cpu. */
335 		pcpu = pcpu_devices + cpumask_first(cpu_online_mask);
336 	pcpu_delegate(pcpu, func, data, (unsigned long) restart_stack);
337 }
338 
339 /*
340  * Call function on the ipl CPU.
341  */
342 void smp_call_ipl_cpu(void (*func)(void *), void *data)
343 {
344 	pcpu_delegate(&pcpu_devices[0], func, data,
345 		      pcpu_devices->panic_stack + PAGE_SIZE);
346 }
347 
348 int smp_find_processor_id(u16 address)
349 {
350 	int cpu;
351 
352 	for_each_present_cpu(cpu)
353 		if (pcpu_devices[cpu].address == address)
354 			return cpu;
355 	return -1;
356 }
357 
358 int smp_vcpu_scheduled(int cpu)
359 {
360 	return pcpu_running(pcpu_devices + cpu);
361 }
362 
363 void smp_yield(void)
364 {
365 	if (MACHINE_HAS_DIAG44)
366 		asm volatile("diag 0,0,0x44");
367 }
368 
369 void smp_yield_cpu(int cpu)
370 {
371 	if (MACHINE_HAS_DIAG9C)
372 		asm volatile("diag %0,0,0x9c"
373 			     : : "d" (pcpu_devices[cpu].address));
374 	else if (MACHINE_HAS_DIAG44)
375 		asm volatile("diag 0,0,0x44");
376 }
377 
378 /*
379  * Send cpus emergency shutdown signal. This gives the cpus the
380  * opportunity to complete outstanding interrupts.
381  */
382 void smp_emergency_stop(cpumask_t *cpumask)
383 {
384 	u64 end;
385 	int cpu;
386 
387 	end = get_clock() + (1000000UL << 12);
388 	for_each_cpu(cpu, cpumask) {
389 		struct pcpu *pcpu = pcpu_devices + cpu;
390 		set_bit(ec_stop_cpu, &pcpu->ec_mask);
391 		while (__pcpu_sigp(pcpu->address, sigp_emergency_signal,
392 				   0, NULL) == sigp_busy &&
393 		       get_clock() < end)
394 			cpu_relax();
395 	}
396 	while (get_clock() < end) {
397 		for_each_cpu(cpu, cpumask)
398 			if (pcpu_stopped(pcpu_devices + cpu))
399 				cpumask_clear_cpu(cpu, cpumask);
400 		if (cpumask_empty(cpumask))
401 			break;
402 		cpu_relax();
403 	}
404 }
405 
406 /*
407  * Stop all cpus but the current one.
408  */
409 void smp_send_stop(void)
410 {
411 	cpumask_t cpumask;
412 	int cpu;
413 
414 	/* Disable all interrupts/machine checks */
415 	__load_psw_mask(psw_kernel_bits | PSW_MASK_DAT);
416 	trace_hardirqs_off();
417 
418 	debug_set_critical();
419 	cpumask_copy(&cpumask, cpu_online_mask);
420 	cpumask_clear_cpu(smp_processor_id(), &cpumask);
421 
422 	if (oops_in_progress)
423 		smp_emergency_stop(&cpumask);
424 
425 	/* stop all processors */
426 	for_each_cpu(cpu, &cpumask) {
427 		struct pcpu *pcpu = pcpu_devices + cpu;
428 		pcpu_sigp_retry(pcpu, sigp_stop, 0);
429 		while (!pcpu_stopped(pcpu))
430 			cpu_relax();
431 	}
432 }
433 
434 /*
435  * Stop the current cpu.
436  */
437 void smp_stop_cpu(void)
438 {
439 	pcpu_sigp_retry(pcpu_devices + smp_processor_id(), sigp_stop, 0);
440 	for (;;) ;
441 }
442 
443 /*
444  * This is the main routine where commands issued by other
445  * cpus are handled.
446  */
447 static void do_ext_call_interrupt(struct ext_code ext_code,
448 				  unsigned int param32, unsigned long param64)
449 {
450 	unsigned long bits;
451 	int cpu;
452 
453 	cpu = smp_processor_id();
454 	if (ext_code.code == 0x1202)
455 		kstat_cpu(cpu).irqs[EXTINT_EXC]++;
456 	else
457 		kstat_cpu(cpu).irqs[EXTINT_EMS]++;
458 	/*
459 	 * handle bit signal external calls
460 	 */
461 	bits = xchg(&pcpu_devices[cpu].ec_mask, 0);
462 
463 	if (test_bit(ec_stop_cpu, &bits))
464 		smp_stop_cpu();
465 
466 	if (test_bit(ec_schedule, &bits))
467 		scheduler_ipi();
468 
469 	if (test_bit(ec_call_function, &bits))
470 		generic_smp_call_function_interrupt();
471 
472 	if (test_bit(ec_call_function_single, &bits))
473 		generic_smp_call_function_single_interrupt();
474 
475 }
476 
477 void arch_send_call_function_ipi_mask(const struct cpumask *mask)
478 {
479 	int cpu;
480 
481 	for_each_cpu(cpu, mask)
482 		pcpu_ec_call(pcpu_devices + cpu, ec_call_function);
483 }
484 
485 void arch_send_call_function_single_ipi(int cpu)
486 {
487 	pcpu_ec_call(pcpu_devices + cpu, ec_call_function_single);
488 }
489 
490 #ifndef CONFIG_64BIT
491 /*
492  * this function sends a 'purge tlb' signal to another CPU.
493  */
494 static void smp_ptlb_callback(void *info)
495 {
496 	__tlb_flush_local();
497 }
498 
499 void smp_ptlb_all(void)
500 {
501 	on_each_cpu(smp_ptlb_callback, NULL, 1);
502 }
503 EXPORT_SYMBOL(smp_ptlb_all);
504 #endif /* ! CONFIG_64BIT */
505 
506 /*
507  * this function sends a 'reschedule' IPI to another CPU.
508  * it goes straight through and wastes no time serializing
509  * anything. Worst case is that we lose a reschedule ...
510  */
511 void smp_send_reschedule(int cpu)
512 {
513 	pcpu_ec_call(pcpu_devices + cpu, ec_schedule);
514 }
515 
516 /*
517  * parameter area for the set/clear control bit callbacks
518  */
519 struct ec_creg_mask_parms {
520 	unsigned long orval;
521 	unsigned long andval;
522 	int cr;
523 };
524 
525 /*
526  * callback for setting/clearing control bits
527  */
528 static void smp_ctl_bit_callback(void *info)
529 {
530 	struct ec_creg_mask_parms *pp = info;
531 	unsigned long cregs[16];
532 
533 	__ctl_store(cregs, 0, 15);
534 	cregs[pp->cr] = (cregs[pp->cr] & pp->andval) | pp->orval;
535 	__ctl_load(cregs, 0, 15);
536 }
537 
538 /*
539  * Set a bit in a control register of all cpus
540  */
541 void smp_ctl_set_bit(int cr, int bit)
542 {
543 	struct ec_creg_mask_parms parms = { 1UL << bit, -1UL, cr };
544 
545 	on_each_cpu(smp_ctl_bit_callback, &parms, 1);
546 }
547 EXPORT_SYMBOL(smp_ctl_set_bit);
548 
549 /*
550  * Clear a bit in a control register of all cpus
551  */
552 void smp_ctl_clear_bit(int cr, int bit)
553 {
554 	struct ec_creg_mask_parms parms = { 0, ~(1UL << bit), cr };
555 
556 	on_each_cpu(smp_ctl_bit_callback, &parms, 1);
557 }
558 EXPORT_SYMBOL(smp_ctl_clear_bit);
559 
560 #if defined(CONFIG_ZFCPDUMP) || defined(CONFIG_CRASH_DUMP)
561 
562 struct save_area *zfcpdump_save_areas[NR_CPUS + 1];
563 EXPORT_SYMBOL_GPL(zfcpdump_save_areas);
564 
565 static void __init smp_get_save_area(int cpu, u16 address)
566 {
567 	void *lc = pcpu_devices[0].lowcore;
568 	struct save_area *save_area;
569 
570 	if (is_kdump_kernel())
571 		return;
572 	if (!OLDMEM_BASE && (address == boot_cpu_address ||
573 			     ipl_info.type != IPL_TYPE_FCP_DUMP))
574 		return;
575 	if (cpu >= NR_CPUS) {
576 		pr_warning("CPU %i exceeds the maximum %i and is excluded "
577 			   "from the dump\n", cpu, NR_CPUS - 1);
578 		return;
579 	}
580 	save_area = kmalloc(sizeof(struct save_area), GFP_KERNEL);
581 	if (!save_area)
582 		panic("could not allocate memory for save area\n");
583 	zfcpdump_save_areas[cpu] = save_area;
584 #ifdef CONFIG_CRASH_DUMP
585 	if (address == boot_cpu_address) {
586 		/* Copy the registers of the boot cpu. */
587 		copy_oldmem_page(1, (void *) save_area, sizeof(*save_area),
588 				 SAVE_AREA_BASE - PAGE_SIZE, 0);
589 		return;
590 	}
591 #endif
592 	/* Get the registers of a non-boot cpu. */
593 	__pcpu_sigp_relax(address, sigp_stop_and_store_status, 0, NULL);
594 	memcpy_real(save_area, lc + SAVE_AREA_BASE, sizeof(*save_area));
595 }
596 
597 int smp_store_status(int cpu)
598 {
599 	struct pcpu *pcpu;
600 
601 	pcpu = pcpu_devices + cpu;
602 	if (__pcpu_sigp_relax(pcpu->address, sigp_stop_and_store_status,
603 			      0, NULL) != sigp_order_code_accepted)
604 		return -EIO;
605 	return 0;
606 }
607 
608 #else /* CONFIG_ZFCPDUMP || CONFIG_CRASH_DUMP */
609 
610 static inline void smp_get_save_area(int cpu, u16 address) { }
611 
612 #endif /* CONFIG_ZFCPDUMP || CONFIG_CRASH_DUMP */
613 
614 static struct sclp_cpu_info *smp_get_cpu_info(void)
615 {
616 	static int use_sigp_detection;
617 	struct sclp_cpu_info *info;
618 	int address;
619 
620 	info = kzalloc(sizeof(*info), GFP_KERNEL);
621 	if (info && (use_sigp_detection || sclp_get_cpu_info(info))) {
622 		use_sigp_detection = 1;
623 		for (address = 0; address <= MAX_CPU_ADDRESS; address++) {
624 			if (__pcpu_sigp_relax(address, sigp_sense, 0, NULL) ==
625 			    sigp_not_operational)
626 				continue;
627 			info->cpu[info->configured].address = address;
628 			info->configured++;
629 		}
630 		info->combined = info->configured;
631 	}
632 	return info;
633 }
634 
635 static int __devinit smp_add_present_cpu(int cpu);
636 
637 static int __devinit __smp_rescan_cpus(struct sclp_cpu_info *info,
638 				       int sysfs_add)
639 {
640 	struct pcpu *pcpu;
641 	cpumask_t avail;
642 	int cpu, nr, i;
643 
644 	nr = 0;
645 	cpumask_xor(&avail, cpu_possible_mask, cpu_present_mask);
646 	cpu = cpumask_first(&avail);
647 	for (i = 0; (i < info->combined) && (cpu < nr_cpu_ids); i++) {
648 		if (info->has_cpu_type && info->cpu[i].type != boot_cpu_type)
649 			continue;
650 		if (pcpu_find_address(cpu_present_mask, info->cpu[i].address))
651 			continue;
652 		pcpu = pcpu_devices + cpu;
653 		pcpu->address = info->cpu[i].address;
654 		pcpu->state = (cpu >= info->configured) ?
655 			CPU_STATE_STANDBY : CPU_STATE_CONFIGURED;
656 		cpu_set_polarization(cpu, POLARIZATION_UNKNOWN);
657 		set_cpu_present(cpu, true);
658 		if (sysfs_add && smp_add_present_cpu(cpu) != 0)
659 			set_cpu_present(cpu, false);
660 		else
661 			nr++;
662 		cpu = cpumask_next(cpu, &avail);
663 	}
664 	return nr;
665 }
666 
667 static void __init smp_detect_cpus(void)
668 {
669 	unsigned int cpu, c_cpus, s_cpus;
670 	struct sclp_cpu_info *info;
671 
672 	info = smp_get_cpu_info();
673 	if (!info)
674 		panic("smp_detect_cpus failed to allocate memory\n");
675 	if (info->has_cpu_type) {
676 		for (cpu = 0; cpu < info->combined; cpu++) {
677 			if (info->cpu[cpu].address != boot_cpu_address)
678 				continue;
679 			/* The boot cpu dictates the cpu type. */
680 			boot_cpu_type = info->cpu[cpu].type;
681 			break;
682 		}
683 	}
684 	c_cpus = s_cpus = 0;
685 	for (cpu = 0; cpu < info->combined; cpu++) {
686 		if (info->has_cpu_type && info->cpu[cpu].type != boot_cpu_type)
687 			continue;
688 		if (cpu < info->configured) {
689 			smp_get_save_area(c_cpus, info->cpu[cpu].address);
690 			c_cpus++;
691 		} else
692 			s_cpus++;
693 	}
694 	pr_info("%d configured CPUs, %d standby CPUs\n", c_cpus, s_cpus);
695 	get_online_cpus();
696 	__smp_rescan_cpus(info, 0);
697 	put_online_cpus();
698 	kfree(info);
699 }
700 
701 /*
702  *	Activate a secondary processor.
703  */
704 static void __cpuinit smp_start_secondary(void *cpuvoid)
705 {
706 	S390_lowcore.last_update_clock = get_clock();
707 	S390_lowcore.restart_stack = (unsigned long) restart_stack;
708 	S390_lowcore.restart_fn = (unsigned long) do_restart;
709 	S390_lowcore.restart_data = 0;
710 	S390_lowcore.restart_source = -1UL;
711 	restore_access_regs(S390_lowcore.access_regs_save_area);
712 	__ctl_load(S390_lowcore.cregs_save_area, 0, 15);
713 	__load_psw_mask(psw_kernel_bits | PSW_MASK_DAT);
714 	cpu_init();
715 	preempt_disable();
716 	init_cpu_timer();
717 	init_cpu_vtimer();
718 	pfault_init();
719 	notify_cpu_starting(smp_processor_id());
720 	ipi_call_lock();
721 	set_cpu_online(smp_processor_id(), true);
722 	ipi_call_unlock();
723 	local_irq_enable();
724 	/* cpu_idle will call schedule for us */
725 	cpu_idle();
726 }
727 
728 /* Upping and downing of CPUs */
729 int __cpuinit __cpu_up(unsigned int cpu, struct task_struct *tidle)
730 {
731 	struct pcpu *pcpu;
732 	int rc;
733 
734 	pcpu = pcpu_devices + cpu;
735 	if (pcpu->state != CPU_STATE_CONFIGURED)
736 		return -EIO;
737 	if (pcpu_sigp_retry(pcpu, sigp_initial_cpu_reset, 0) !=
738 	    sigp_order_code_accepted)
739 		return -EIO;
740 
741 	rc = pcpu_alloc_lowcore(pcpu, cpu);
742 	if (rc)
743 		return rc;
744 	pcpu_prepare_secondary(pcpu, cpu);
745 	pcpu_attach_task(pcpu, tidle);
746 	pcpu_start_fn(pcpu, smp_start_secondary, NULL);
747 	while (!cpu_online(cpu))
748 		cpu_relax();
749 	return 0;
750 }
751 
752 static int __init setup_possible_cpus(char *s)
753 {
754 	int max, cpu;
755 
756 	if (kstrtoint(s, 0, &max) < 0)
757 		return 0;
758 	init_cpu_possible(cpumask_of(0));
759 	for (cpu = 1; cpu < max && cpu < nr_cpu_ids; cpu++)
760 		set_cpu_possible(cpu, true);
761 	return 0;
762 }
763 early_param("possible_cpus", setup_possible_cpus);
764 
765 #ifdef CONFIG_HOTPLUG_CPU
766 
767 int __cpu_disable(void)
768 {
769 	unsigned long cregs[16];
770 
771 	set_cpu_online(smp_processor_id(), false);
772 	/* Disable pseudo page faults on this cpu. */
773 	pfault_fini();
774 	/* Disable interrupt sources via control register. */
775 	__ctl_store(cregs, 0, 15);
776 	cregs[0]  &= ~0x0000ee70UL;	/* disable all external interrupts */
777 	cregs[6]  &= ~0xff000000UL;	/* disable all I/O interrupts */
778 	cregs[14] &= ~0x1f000000UL;	/* disable most machine checks */
779 	__ctl_load(cregs, 0, 15);
780 	return 0;
781 }
782 
783 void __cpu_die(unsigned int cpu)
784 {
785 	struct pcpu *pcpu;
786 
787 	/* Wait until target cpu is down */
788 	pcpu = pcpu_devices + cpu;
789 	while (!pcpu_stopped(pcpu))
790 		cpu_relax();
791 	pcpu_free_lowcore(pcpu);
792 	atomic_dec(&init_mm.context.attach_count);
793 }
794 
795 void __noreturn cpu_die(void)
796 {
797 	idle_task_exit();
798 	pcpu_sigp_retry(pcpu_devices + smp_processor_id(), sigp_stop, 0);
799 	for (;;) ;
800 }
801 
802 #endif /* CONFIG_HOTPLUG_CPU */
803 
804 void __init smp_prepare_cpus(unsigned int max_cpus)
805 {
806 	/* request the 0x1201 emergency signal external interrupt */
807 	if (register_external_interrupt(0x1201, do_ext_call_interrupt) != 0)
808 		panic("Couldn't request external interrupt 0x1201");
809 	/* request the 0x1202 external call external interrupt */
810 	if (register_external_interrupt(0x1202, do_ext_call_interrupt) != 0)
811 		panic("Couldn't request external interrupt 0x1202");
812 	smp_detect_cpus();
813 }
814 
815 void __init smp_prepare_boot_cpu(void)
816 {
817 	struct pcpu *pcpu = pcpu_devices;
818 
819 	boot_cpu_address = stap();
820 	pcpu->state = CPU_STATE_CONFIGURED;
821 	pcpu->address = boot_cpu_address;
822 	pcpu->lowcore = (struct _lowcore *)(unsigned long) store_prefix();
823 	pcpu->async_stack = S390_lowcore.async_stack - ASYNC_SIZE;
824 	pcpu->panic_stack = S390_lowcore.panic_stack - PAGE_SIZE;
825 	S390_lowcore.percpu_offset = __per_cpu_offset[0];
826 	cpu_set_polarization(0, POLARIZATION_UNKNOWN);
827 	set_cpu_present(0, true);
828 	set_cpu_online(0, true);
829 }
830 
831 void __init smp_cpus_done(unsigned int max_cpus)
832 {
833 }
834 
835 void __init smp_setup_processor_id(void)
836 {
837 	S390_lowcore.cpu_nr = 0;
838 }
839 
840 /*
841  * the frequency of the profiling timer can be changed
842  * by writing a multiplier value into /proc/profile.
843  *
844  * usually you want to run this on all CPUs ;)
845  */
846 int setup_profiling_timer(unsigned int multiplier)
847 {
848 	return 0;
849 }
850 
851 #ifdef CONFIG_HOTPLUG_CPU
852 static ssize_t cpu_configure_show(struct device *dev,
853 				  struct device_attribute *attr, char *buf)
854 {
855 	ssize_t count;
856 
857 	mutex_lock(&smp_cpu_state_mutex);
858 	count = sprintf(buf, "%d\n", pcpu_devices[dev->id].state);
859 	mutex_unlock(&smp_cpu_state_mutex);
860 	return count;
861 }
862 
863 static ssize_t cpu_configure_store(struct device *dev,
864 				   struct device_attribute *attr,
865 				   const char *buf, size_t count)
866 {
867 	struct pcpu *pcpu;
868 	int cpu, val, rc;
869 	char delim;
870 
871 	if (sscanf(buf, "%d %c", &val, &delim) != 1)
872 		return -EINVAL;
873 	if (val != 0 && val != 1)
874 		return -EINVAL;
875 	get_online_cpus();
876 	mutex_lock(&smp_cpu_state_mutex);
877 	rc = -EBUSY;
878 	/* disallow configuration changes of online cpus and cpu 0 */
879 	cpu = dev->id;
880 	if (cpu_online(cpu) || cpu == 0)
881 		goto out;
882 	pcpu = pcpu_devices + cpu;
883 	rc = 0;
884 	switch (val) {
885 	case 0:
886 		if (pcpu->state != CPU_STATE_CONFIGURED)
887 			break;
888 		rc = sclp_cpu_deconfigure(pcpu->address);
889 		if (rc)
890 			break;
891 		pcpu->state = CPU_STATE_STANDBY;
892 		cpu_set_polarization(cpu, POLARIZATION_UNKNOWN);
893 		topology_expect_change();
894 		break;
895 	case 1:
896 		if (pcpu->state != CPU_STATE_STANDBY)
897 			break;
898 		rc = sclp_cpu_configure(pcpu->address);
899 		if (rc)
900 			break;
901 		pcpu->state = CPU_STATE_CONFIGURED;
902 		cpu_set_polarization(cpu, POLARIZATION_UNKNOWN);
903 		topology_expect_change();
904 		break;
905 	default:
906 		break;
907 	}
908 out:
909 	mutex_unlock(&smp_cpu_state_mutex);
910 	put_online_cpus();
911 	return rc ? rc : count;
912 }
913 static DEVICE_ATTR(configure, 0644, cpu_configure_show, cpu_configure_store);
914 #endif /* CONFIG_HOTPLUG_CPU */
915 
916 static ssize_t show_cpu_address(struct device *dev,
917 				struct device_attribute *attr, char *buf)
918 {
919 	return sprintf(buf, "%d\n", pcpu_devices[dev->id].address);
920 }
921 static DEVICE_ATTR(address, 0444, show_cpu_address, NULL);
922 
923 static struct attribute *cpu_common_attrs[] = {
924 #ifdef CONFIG_HOTPLUG_CPU
925 	&dev_attr_configure.attr,
926 #endif
927 	&dev_attr_address.attr,
928 	NULL,
929 };
930 
931 static struct attribute_group cpu_common_attr_group = {
932 	.attrs = cpu_common_attrs,
933 };
934 
935 static ssize_t show_idle_count(struct device *dev,
936 				struct device_attribute *attr, char *buf)
937 {
938 	struct s390_idle_data *idle = &per_cpu(s390_idle, dev->id);
939 	unsigned long long idle_count;
940 	unsigned int sequence;
941 
942 	do {
943 		sequence = ACCESS_ONCE(idle->sequence);
944 		idle_count = ACCESS_ONCE(idle->idle_count);
945 		if (ACCESS_ONCE(idle->idle_enter))
946 			idle_count++;
947 	} while ((sequence & 1) || (idle->sequence != sequence));
948 	return sprintf(buf, "%llu\n", idle_count);
949 }
950 static DEVICE_ATTR(idle_count, 0444, show_idle_count, NULL);
951 
952 static ssize_t show_idle_time(struct device *dev,
953 				struct device_attribute *attr, char *buf)
954 {
955 	struct s390_idle_data *idle = &per_cpu(s390_idle, dev->id);
956 	unsigned long long now, idle_time, idle_enter, idle_exit;
957 	unsigned int sequence;
958 
959 	do {
960 		now = get_clock();
961 		sequence = ACCESS_ONCE(idle->sequence);
962 		idle_time = ACCESS_ONCE(idle->idle_time);
963 		idle_enter = ACCESS_ONCE(idle->idle_enter);
964 		idle_exit = ACCESS_ONCE(idle->idle_exit);
965 	} while ((sequence & 1) || (idle->sequence != sequence));
966 	idle_time += idle_enter ? ((idle_exit ? : now) - idle_enter) : 0;
967 	return sprintf(buf, "%llu\n", idle_time >> 12);
968 }
969 static DEVICE_ATTR(idle_time_us, 0444, show_idle_time, NULL);
970 
971 static struct attribute *cpu_online_attrs[] = {
972 	&dev_attr_idle_count.attr,
973 	&dev_attr_idle_time_us.attr,
974 	NULL,
975 };
976 
977 static struct attribute_group cpu_online_attr_group = {
978 	.attrs = cpu_online_attrs,
979 };
980 
981 static int __cpuinit smp_cpu_notify(struct notifier_block *self,
982 				    unsigned long action, void *hcpu)
983 {
984 	unsigned int cpu = (unsigned int)(long)hcpu;
985 	struct cpu *c = &pcpu_devices[cpu].cpu;
986 	struct device *s = &c->dev;
987 	struct s390_idle_data *idle;
988 	int err = 0;
989 
990 	switch (action) {
991 	case CPU_ONLINE:
992 	case CPU_ONLINE_FROZEN:
993 		idle = &per_cpu(s390_idle, cpu);
994 		memset(idle, 0, sizeof(struct s390_idle_data));
995 		err = sysfs_create_group(&s->kobj, &cpu_online_attr_group);
996 		break;
997 	case CPU_DEAD:
998 	case CPU_DEAD_FROZEN:
999 		sysfs_remove_group(&s->kobj, &cpu_online_attr_group);
1000 		break;
1001 	}
1002 	return notifier_from_errno(err);
1003 }
1004 
1005 static struct notifier_block __cpuinitdata smp_cpu_nb = {
1006 	.notifier_call = smp_cpu_notify,
1007 };
1008 
1009 static int __devinit smp_add_present_cpu(int cpu)
1010 {
1011 	struct cpu *c = &pcpu_devices[cpu].cpu;
1012 	struct device *s = &c->dev;
1013 	int rc;
1014 
1015 	c->hotpluggable = 1;
1016 	rc = register_cpu(c, cpu);
1017 	if (rc)
1018 		goto out;
1019 	rc = sysfs_create_group(&s->kobj, &cpu_common_attr_group);
1020 	if (rc)
1021 		goto out_cpu;
1022 	if (cpu_online(cpu)) {
1023 		rc = sysfs_create_group(&s->kobj, &cpu_online_attr_group);
1024 		if (rc)
1025 			goto out_online;
1026 	}
1027 	rc = topology_cpu_init(c);
1028 	if (rc)
1029 		goto out_topology;
1030 	return 0;
1031 
1032 out_topology:
1033 	if (cpu_online(cpu))
1034 		sysfs_remove_group(&s->kobj, &cpu_online_attr_group);
1035 out_online:
1036 	sysfs_remove_group(&s->kobj, &cpu_common_attr_group);
1037 out_cpu:
1038 #ifdef CONFIG_HOTPLUG_CPU
1039 	unregister_cpu(c);
1040 #endif
1041 out:
1042 	return rc;
1043 }
1044 
1045 #ifdef CONFIG_HOTPLUG_CPU
1046 
1047 int __ref smp_rescan_cpus(void)
1048 {
1049 	struct sclp_cpu_info *info;
1050 	int nr;
1051 
1052 	info = smp_get_cpu_info();
1053 	if (!info)
1054 		return -ENOMEM;
1055 	get_online_cpus();
1056 	mutex_lock(&smp_cpu_state_mutex);
1057 	nr = __smp_rescan_cpus(info, 1);
1058 	mutex_unlock(&smp_cpu_state_mutex);
1059 	put_online_cpus();
1060 	kfree(info);
1061 	if (nr)
1062 		topology_schedule_update();
1063 	return 0;
1064 }
1065 
1066 static ssize_t __ref rescan_store(struct device *dev,
1067 				  struct device_attribute *attr,
1068 				  const char *buf,
1069 				  size_t count)
1070 {
1071 	int rc;
1072 
1073 	rc = smp_rescan_cpus();
1074 	return rc ? rc : count;
1075 }
1076 static DEVICE_ATTR(rescan, 0200, NULL, rescan_store);
1077 #endif /* CONFIG_HOTPLUG_CPU */
1078 
1079 static int __init s390_smp_init(void)
1080 {
1081 	int cpu, rc;
1082 
1083 	register_cpu_notifier(&smp_cpu_nb);
1084 #ifdef CONFIG_HOTPLUG_CPU
1085 	rc = device_create_file(cpu_subsys.dev_root, &dev_attr_rescan);
1086 	if (rc)
1087 		return rc;
1088 #endif
1089 	for_each_present_cpu(cpu) {
1090 		rc = smp_add_present_cpu(cpu);
1091 		if (rc)
1092 			return rc;
1093 	}
1094 	return 0;
1095 }
1096 subsys_initcall(s390_smp_init);
1097