xref: /openbmc/linux/arch/s390/kernel/smp.c (revision ca460cc2)
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 <asm/idle.h>
49 #include "entry.h"
50 
51 enum {
52 	ec_schedule = 0,
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 	int polarization;		/* physical polarization */
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 /*
78  * The smp_cpu_state_mutex must be held when changing the state or polarization
79  * member of a pcpu data structure within the pcpu_devices arreay.
80  */
81 DEFINE_MUTEX(smp_cpu_state_mutex);
82 
83 /*
84  * Signal processor helper functions.
85  */
86 static inline int __pcpu_sigp_relax(u16 addr, u8 order, unsigned long parm,
87 				    u32 *status)
88 {
89 	int cc;
90 
91 	while (1) {
92 		cc = __pcpu_sigp(addr, order, parm, NULL);
93 		if (cc != SIGP_CC_BUSY)
94 			return cc;
95 		cpu_relax();
96 	}
97 }
98 
99 static int pcpu_sigp_retry(struct pcpu *pcpu, u8 order, u32 parm)
100 {
101 	int cc, retry;
102 
103 	for (retry = 0; ; retry++) {
104 		cc = __pcpu_sigp(pcpu->address, order, parm, NULL);
105 		if (cc != SIGP_CC_BUSY)
106 			break;
107 		if (retry >= 3)
108 			udelay(10);
109 	}
110 	return cc;
111 }
112 
113 static inline int pcpu_stopped(struct pcpu *pcpu)
114 {
115 	u32 uninitialized_var(status);
116 
117 	if (__pcpu_sigp(pcpu->address, SIGP_SENSE,
118 			0, &status) != SIGP_CC_STATUS_STORED)
119 		return 0;
120 	return !!(status & (SIGP_STATUS_CHECK_STOP|SIGP_STATUS_STOPPED));
121 }
122 
123 static inline int pcpu_running(struct pcpu *pcpu)
124 {
125 	if (__pcpu_sigp(pcpu->address, SIGP_SENSE_RUNNING,
126 			0, NULL) != SIGP_CC_STATUS_STORED)
127 		return 1;
128 	/* Status stored condition code is equivalent to cpu not running. */
129 	return 0;
130 }
131 
132 /*
133  * Find struct pcpu by cpu address.
134  */
135 static struct pcpu *pcpu_find_address(const struct cpumask *mask, int address)
136 {
137 	int cpu;
138 
139 	for_each_cpu(cpu, mask)
140 		if (pcpu_devices[cpu].address == address)
141 			return pcpu_devices + cpu;
142 	return NULL;
143 }
144 
145 static void pcpu_ec_call(struct pcpu *pcpu, int ec_bit)
146 {
147 	int order;
148 
149 	if (test_and_set_bit(ec_bit, &pcpu->ec_mask))
150 		return;
151 	order = pcpu_running(pcpu) ? SIGP_EXTERNAL_CALL : SIGP_EMERGENCY_SIGNAL;
152 	pcpu_sigp_retry(pcpu, order, 0);
153 }
154 
155 static int pcpu_alloc_lowcore(struct pcpu *pcpu, int cpu)
156 {
157 	struct _lowcore *lc;
158 
159 	if (pcpu != &pcpu_devices[0]) {
160 		pcpu->lowcore =	(struct _lowcore *)
161 			__get_free_pages(GFP_KERNEL | GFP_DMA, LC_ORDER);
162 		pcpu->async_stack = __get_free_pages(GFP_KERNEL, ASYNC_ORDER);
163 		pcpu->panic_stack = __get_free_page(GFP_KERNEL);
164 		if (!pcpu->lowcore || !pcpu->panic_stack || !pcpu->async_stack)
165 			goto out;
166 	}
167 	lc = pcpu->lowcore;
168 	memcpy(lc, &S390_lowcore, 512);
169 	memset((char *) lc + 512, 0, sizeof(*lc) - 512);
170 	lc->async_stack = pcpu->async_stack + ASYNC_SIZE
171 		- STACK_FRAME_OVERHEAD - sizeof(struct pt_regs);
172 	lc->panic_stack = pcpu->panic_stack + PAGE_SIZE
173 		- STACK_FRAME_OVERHEAD - sizeof(struct pt_regs);
174 	lc->cpu_nr = cpu;
175 	lc->spinlock_lockval = arch_spin_lockval(cpu);
176 #ifndef CONFIG_64BIT
177 	if (MACHINE_HAS_IEEE) {
178 		lc->extended_save_area_addr = get_zeroed_page(GFP_KERNEL);
179 		if (!lc->extended_save_area_addr)
180 			goto out;
181 	}
182 #else
183 	if (MACHINE_HAS_VX)
184 		lc->vector_save_area_addr =
185 			(unsigned long) &lc->vector_save_area;
186 	if (vdso_alloc_per_cpu(lc))
187 		goto out;
188 #endif
189 	lowcore_ptr[cpu] = lc;
190 	pcpu_sigp_retry(pcpu, SIGP_SET_PREFIX, (u32)(unsigned long) lc);
191 	return 0;
192 out:
193 	if (pcpu != &pcpu_devices[0]) {
194 		free_page(pcpu->panic_stack);
195 		free_pages(pcpu->async_stack, ASYNC_ORDER);
196 		free_pages((unsigned long) pcpu->lowcore, LC_ORDER);
197 	}
198 	return -ENOMEM;
199 }
200 
201 #ifdef CONFIG_HOTPLUG_CPU
202 
203 static void pcpu_free_lowcore(struct pcpu *pcpu)
204 {
205 	pcpu_sigp_retry(pcpu, SIGP_SET_PREFIX, 0);
206 	lowcore_ptr[pcpu - pcpu_devices] = NULL;
207 #ifndef CONFIG_64BIT
208 	if (MACHINE_HAS_IEEE) {
209 		struct _lowcore *lc = pcpu->lowcore;
210 
211 		free_page((unsigned long) lc->extended_save_area_addr);
212 		lc->extended_save_area_addr = 0;
213 	}
214 #else
215 	vdso_free_per_cpu(pcpu->lowcore);
216 #endif
217 	if (pcpu != &pcpu_devices[0]) {
218 		free_page(pcpu->panic_stack);
219 		free_pages(pcpu->async_stack, ASYNC_ORDER);
220 		free_pages((unsigned long) pcpu->lowcore, LC_ORDER);
221 	}
222 }
223 
224 #endif /* CONFIG_HOTPLUG_CPU */
225 
226 static void pcpu_prepare_secondary(struct pcpu *pcpu, int cpu)
227 {
228 	struct _lowcore *lc = pcpu->lowcore;
229 
230 	if (MACHINE_HAS_TLB_LC)
231 		cpumask_set_cpu(cpu, &init_mm.context.cpu_attach_mask);
232 	cpumask_set_cpu(cpu, mm_cpumask(&init_mm));
233 	atomic_inc(&init_mm.context.attach_count);
234 	lc->cpu_nr = cpu;
235 	lc->spinlock_lockval = arch_spin_lockval(cpu);
236 	lc->percpu_offset = __per_cpu_offset[cpu];
237 	lc->kernel_asce = S390_lowcore.kernel_asce;
238 	lc->machine_flags = S390_lowcore.machine_flags;
239 	lc->ftrace_func = S390_lowcore.ftrace_func;
240 	lc->user_timer = lc->system_timer = lc->steal_timer = 0;
241 	__ctl_store(lc->cregs_save_area, 0, 15);
242 	save_access_regs((unsigned int *) lc->access_regs_save_area);
243 	memcpy(lc->stfle_fac_list, S390_lowcore.stfle_fac_list,
244 	       MAX_FACILITY_BIT/8);
245 }
246 
247 static void pcpu_attach_task(struct pcpu *pcpu, struct task_struct *tsk)
248 {
249 	struct _lowcore *lc = pcpu->lowcore;
250 	struct thread_info *ti = task_thread_info(tsk);
251 
252 	lc->kernel_stack = (unsigned long) task_stack_page(tsk)
253 		+ THREAD_SIZE - STACK_FRAME_OVERHEAD - sizeof(struct pt_regs);
254 	lc->thread_info = (unsigned long) task_thread_info(tsk);
255 	lc->current_task = (unsigned long) tsk;
256 	lc->user_timer = ti->user_timer;
257 	lc->system_timer = ti->system_timer;
258 	lc->steal_timer = 0;
259 }
260 
261 static void pcpu_start_fn(struct pcpu *pcpu, void (*func)(void *), void *data)
262 {
263 	struct _lowcore *lc = pcpu->lowcore;
264 
265 	lc->restart_stack = lc->kernel_stack;
266 	lc->restart_fn = (unsigned long) func;
267 	lc->restart_data = (unsigned long) data;
268 	lc->restart_source = -1UL;
269 	pcpu_sigp_retry(pcpu, SIGP_RESTART, 0);
270 }
271 
272 /*
273  * Call function via PSW restart on pcpu and stop the current cpu.
274  */
275 static void pcpu_delegate(struct pcpu *pcpu, void (*func)(void *),
276 			  void *data, unsigned long stack)
277 {
278 	struct _lowcore *lc = lowcore_ptr[pcpu - pcpu_devices];
279 	unsigned long source_cpu = stap();
280 
281 	__load_psw_mask(PSW_KERNEL_BITS);
282 	if (pcpu->address == source_cpu)
283 		func(data);	/* should not return */
284 	/* Stop target cpu (if func returns this stops the current cpu). */
285 	pcpu_sigp_retry(pcpu, SIGP_STOP, 0);
286 	/* Restart func on the target cpu and stop the current cpu. */
287 	mem_assign_absolute(lc->restart_stack, stack);
288 	mem_assign_absolute(lc->restart_fn, (unsigned long) func);
289 	mem_assign_absolute(lc->restart_data, (unsigned long) data);
290 	mem_assign_absolute(lc->restart_source, source_cpu);
291 	asm volatile(
292 		"0:	sigp	0,%0,%2	# sigp restart to target cpu\n"
293 		"	brc	2,0b	# busy, try again\n"
294 		"1:	sigp	0,%1,%3	# sigp stop to current cpu\n"
295 		"	brc	2,1b	# busy, try again\n"
296 		: : "d" (pcpu->address), "d" (source_cpu),
297 		    "K" (SIGP_RESTART), "K" (SIGP_STOP)
298 		: "0", "1", "cc");
299 	for (;;) ;
300 }
301 
302 /*
303  * Call function on an online CPU.
304  */
305 void smp_call_online_cpu(void (*func)(void *), void *data)
306 {
307 	struct pcpu *pcpu;
308 
309 	/* Use the current cpu if it is online. */
310 	pcpu = pcpu_find_address(cpu_online_mask, stap());
311 	if (!pcpu)
312 		/* Use the first online cpu. */
313 		pcpu = pcpu_devices + cpumask_first(cpu_online_mask);
314 	pcpu_delegate(pcpu, func, data, (unsigned long) restart_stack);
315 }
316 
317 /*
318  * Call function on the ipl CPU.
319  */
320 void smp_call_ipl_cpu(void (*func)(void *), void *data)
321 {
322 	pcpu_delegate(&pcpu_devices[0], func, data,
323 		      pcpu_devices->panic_stack + PAGE_SIZE);
324 }
325 
326 int smp_find_processor_id(u16 address)
327 {
328 	int cpu;
329 
330 	for_each_present_cpu(cpu)
331 		if (pcpu_devices[cpu].address == address)
332 			return cpu;
333 	return -1;
334 }
335 
336 int smp_vcpu_scheduled(int cpu)
337 {
338 	return pcpu_running(pcpu_devices + cpu);
339 }
340 
341 void smp_yield_cpu(int cpu)
342 {
343 	if (MACHINE_HAS_DIAG9C)
344 		asm volatile("diag %0,0,0x9c"
345 			     : : "d" (pcpu_devices[cpu].address));
346 	else if (MACHINE_HAS_DIAG44)
347 		asm volatile("diag 0,0,0x44");
348 }
349 
350 /*
351  * Send cpus emergency shutdown signal. This gives the cpus the
352  * opportunity to complete outstanding interrupts.
353  */
354 static void smp_emergency_stop(cpumask_t *cpumask)
355 {
356 	u64 end;
357 	int cpu;
358 
359 	end = get_tod_clock() + (1000000UL << 12);
360 	for_each_cpu(cpu, cpumask) {
361 		struct pcpu *pcpu = pcpu_devices + cpu;
362 		set_bit(ec_stop_cpu, &pcpu->ec_mask);
363 		while (__pcpu_sigp(pcpu->address, SIGP_EMERGENCY_SIGNAL,
364 				   0, NULL) == SIGP_CC_BUSY &&
365 		       get_tod_clock() < end)
366 			cpu_relax();
367 	}
368 	while (get_tod_clock() < end) {
369 		for_each_cpu(cpu, cpumask)
370 			if (pcpu_stopped(pcpu_devices + cpu))
371 				cpumask_clear_cpu(cpu, cpumask);
372 		if (cpumask_empty(cpumask))
373 			break;
374 		cpu_relax();
375 	}
376 }
377 
378 /*
379  * Stop all cpus but the current one.
380  */
381 void smp_send_stop(void)
382 {
383 	cpumask_t cpumask;
384 	int cpu;
385 
386 	/* Disable all interrupts/machine checks */
387 	__load_psw_mask(PSW_KERNEL_BITS | PSW_MASK_DAT);
388 	trace_hardirqs_off();
389 
390 	debug_set_critical();
391 	cpumask_copy(&cpumask, cpu_online_mask);
392 	cpumask_clear_cpu(smp_processor_id(), &cpumask);
393 
394 	if (oops_in_progress)
395 		smp_emergency_stop(&cpumask);
396 
397 	/* stop all processors */
398 	for_each_cpu(cpu, &cpumask) {
399 		struct pcpu *pcpu = pcpu_devices + cpu;
400 		pcpu_sigp_retry(pcpu, SIGP_STOP, 0);
401 		while (!pcpu_stopped(pcpu))
402 			cpu_relax();
403 	}
404 }
405 
406 /*
407  * This is the main routine where commands issued by other
408  * cpus are handled.
409  */
410 static void smp_handle_ext_call(void)
411 {
412 	unsigned long bits;
413 
414 	/* handle bit signal external calls */
415 	bits = xchg(&pcpu_devices[smp_processor_id()].ec_mask, 0);
416 	if (test_bit(ec_stop_cpu, &bits))
417 		smp_stop_cpu();
418 	if (test_bit(ec_schedule, &bits))
419 		scheduler_ipi();
420 	if (test_bit(ec_call_function_single, &bits))
421 		generic_smp_call_function_single_interrupt();
422 }
423 
424 static void do_ext_call_interrupt(struct ext_code ext_code,
425 				  unsigned int param32, unsigned long param64)
426 {
427 	inc_irq_stat(ext_code.code == 0x1202 ? IRQEXT_EXC : IRQEXT_EMS);
428 	smp_handle_ext_call();
429 }
430 
431 void arch_send_call_function_ipi_mask(const struct cpumask *mask)
432 {
433 	int cpu;
434 
435 	for_each_cpu(cpu, mask)
436 		pcpu_ec_call(pcpu_devices + cpu, ec_call_function_single);
437 }
438 
439 void arch_send_call_function_single_ipi(int cpu)
440 {
441 	pcpu_ec_call(pcpu_devices + cpu, ec_call_function_single);
442 }
443 
444 #ifndef CONFIG_64BIT
445 /*
446  * this function sends a 'purge tlb' signal to another CPU.
447  */
448 static void smp_ptlb_callback(void *info)
449 {
450 	__tlb_flush_local();
451 }
452 
453 void smp_ptlb_all(void)
454 {
455 	on_each_cpu(smp_ptlb_callback, NULL, 1);
456 }
457 EXPORT_SYMBOL(smp_ptlb_all);
458 #endif /* ! CONFIG_64BIT */
459 
460 /*
461  * this function sends a 'reschedule' IPI to another CPU.
462  * it goes straight through and wastes no time serializing
463  * anything. Worst case is that we lose a reschedule ...
464  */
465 void smp_send_reschedule(int cpu)
466 {
467 	pcpu_ec_call(pcpu_devices + cpu, ec_schedule);
468 }
469 
470 /*
471  * parameter area for the set/clear control bit callbacks
472  */
473 struct ec_creg_mask_parms {
474 	unsigned long orval;
475 	unsigned long andval;
476 	int cr;
477 };
478 
479 /*
480  * callback for setting/clearing control bits
481  */
482 static void smp_ctl_bit_callback(void *info)
483 {
484 	struct ec_creg_mask_parms *pp = info;
485 	unsigned long cregs[16];
486 
487 	__ctl_store(cregs, 0, 15);
488 	cregs[pp->cr] = (cregs[pp->cr] & pp->andval) | pp->orval;
489 	__ctl_load(cregs, 0, 15);
490 }
491 
492 /*
493  * Set a bit in a control register of all cpus
494  */
495 void smp_ctl_set_bit(int cr, int bit)
496 {
497 	struct ec_creg_mask_parms parms = { 1UL << bit, -1UL, cr };
498 
499 	on_each_cpu(smp_ctl_bit_callback, &parms, 1);
500 }
501 EXPORT_SYMBOL(smp_ctl_set_bit);
502 
503 /*
504  * Clear a bit in a control register of all cpus
505  */
506 void smp_ctl_clear_bit(int cr, int bit)
507 {
508 	struct ec_creg_mask_parms parms = { 0, ~(1UL << bit), cr };
509 
510 	on_each_cpu(smp_ctl_bit_callback, &parms, 1);
511 }
512 EXPORT_SYMBOL(smp_ctl_clear_bit);
513 
514 #ifdef CONFIG_CRASH_DUMP
515 
516 static void __init smp_get_save_area(int cpu, u16 address)
517 {
518 	void *lc = pcpu_devices[0].lowcore;
519 	struct save_area_ext *sa_ext;
520 	unsigned long vx_sa;
521 
522 	if (is_kdump_kernel())
523 		return;
524 	if (!OLDMEM_BASE && (address == boot_cpu_address ||
525 			     ipl_info.type != IPL_TYPE_FCP_DUMP))
526 		return;
527 	sa_ext = dump_save_area_create(cpu);
528 	if (!sa_ext)
529 		panic("could not allocate memory for save area\n");
530 	if (address == boot_cpu_address) {
531 		/* Copy the registers of the boot cpu. */
532 		copy_oldmem_page(1, (void *) &sa_ext->sa, sizeof(sa_ext->sa),
533 				 SAVE_AREA_BASE - PAGE_SIZE, 0);
534 		if (MACHINE_HAS_VX)
535 			save_vx_regs_safe(sa_ext->vx_regs);
536 		return;
537 	}
538 	/* Get the registers of a non-boot cpu. */
539 	__pcpu_sigp_relax(address, SIGP_STOP_AND_STORE_STATUS, 0, NULL);
540 	memcpy_real(&sa_ext->sa, lc + SAVE_AREA_BASE, sizeof(sa_ext->sa));
541 	if (!MACHINE_HAS_VX)
542 		return;
543 	/* Get the VX registers */
544 	vx_sa = __get_free_page(GFP_KERNEL);
545 	if (!vx_sa)
546 		panic("could not allocate memory for VX save area\n");
547 	__pcpu_sigp_relax(address, SIGP_STORE_ADDITIONAL_STATUS, vx_sa, NULL);
548 	memcpy(sa_ext->vx_regs, (void *) vx_sa, sizeof(sa_ext->vx_regs));
549 	free_page(vx_sa);
550 }
551 
552 int smp_store_status(int cpu)
553 {
554 	unsigned long vx_sa;
555 	struct pcpu *pcpu;
556 
557 	pcpu = pcpu_devices + cpu;
558 	if (__pcpu_sigp_relax(pcpu->address, SIGP_STOP_AND_STORE_STATUS,
559 			      0, NULL) != SIGP_CC_ORDER_CODE_ACCEPTED)
560 		return -EIO;
561 	if (!MACHINE_HAS_VX)
562 		return 0;
563 	vx_sa = __pa(pcpu->lowcore->vector_save_area_addr);
564 	__pcpu_sigp_relax(pcpu->address, SIGP_STORE_ADDITIONAL_STATUS,
565 			  vx_sa, NULL);
566 	return 0;
567 }
568 
569 #else /* CONFIG_CRASH_DUMP */
570 
571 static inline void smp_get_save_area(int cpu, u16 address) { }
572 
573 #endif /* CONFIG_CRASH_DUMP */
574 
575 void smp_cpu_set_polarization(int cpu, int val)
576 {
577 	pcpu_devices[cpu].polarization = val;
578 }
579 
580 int smp_cpu_get_polarization(int cpu)
581 {
582 	return pcpu_devices[cpu].polarization;
583 }
584 
585 static struct sclp_cpu_info *smp_get_cpu_info(void)
586 {
587 	static int use_sigp_detection;
588 	struct sclp_cpu_info *info;
589 	int address;
590 
591 	info = kzalloc(sizeof(*info), GFP_KERNEL);
592 	if (info && (use_sigp_detection || sclp_get_cpu_info(info))) {
593 		use_sigp_detection = 1;
594 		for (address = 0; address <= MAX_CPU_ADDRESS; address++) {
595 			if (__pcpu_sigp_relax(address, SIGP_SENSE, 0, NULL) ==
596 			    SIGP_CC_NOT_OPERATIONAL)
597 				continue;
598 			info->cpu[info->configured].address = address;
599 			info->configured++;
600 		}
601 		info->combined = info->configured;
602 	}
603 	return info;
604 }
605 
606 static int smp_add_present_cpu(int cpu);
607 
608 static int __smp_rescan_cpus(struct sclp_cpu_info *info, int sysfs_add)
609 {
610 	struct pcpu *pcpu;
611 	cpumask_t avail;
612 	int cpu, nr, i;
613 
614 	nr = 0;
615 	cpumask_xor(&avail, cpu_possible_mask, cpu_present_mask);
616 	cpu = cpumask_first(&avail);
617 	for (i = 0; (i < info->combined) && (cpu < nr_cpu_ids); i++) {
618 		if (info->has_cpu_type && info->cpu[i].type != boot_cpu_type)
619 			continue;
620 		if (pcpu_find_address(cpu_present_mask, info->cpu[i].address))
621 			continue;
622 		pcpu = pcpu_devices + cpu;
623 		pcpu->address = info->cpu[i].address;
624 		pcpu->state = (i >= info->configured) ?
625 			CPU_STATE_STANDBY : CPU_STATE_CONFIGURED;
626 		smp_cpu_set_polarization(cpu, POLARIZATION_UNKNOWN);
627 		set_cpu_present(cpu, true);
628 		if (sysfs_add && smp_add_present_cpu(cpu) != 0)
629 			set_cpu_present(cpu, false);
630 		else
631 			nr++;
632 		cpu = cpumask_next(cpu, &avail);
633 	}
634 	return nr;
635 }
636 
637 static void __init smp_detect_cpus(void)
638 {
639 	unsigned int cpu, c_cpus, s_cpus;
640 	struct sclp_cpu_info *info;
641 
642 	info = smp_get_cpu_info();
643 	if (!info)
644 		panic("smp_detect_cpus failed to allocate memory\n");
645 	if (info->has_cpu_type) {
646 		for (cpu = 0; cpu < info->combined; cpu++) {
647 			if (info->cpu[cpu].address != boot_cpu_address)
648 				continue;
649 			/* The boot cpu dictates the cpu type. */
650 			boot_cpu_type = info->cpu[cpu].type;
651 			break;
652 		}
653 	}
654 	c_cpus = s_cpus = 0;
655 	for (cpu = 0; cpu < info->combined; cpu++) {
656 		if (info->has_cpu_type && info->cpu[cpu].type != boot_cpu_type)
657 			continue;
658 		if (cpu < info->configured) {
659 			smp_get_save_area(c_cpus, info->cpu[cpu].address);
660 			c_cpus++;
661 		} else
662 			s_cpus++;
663 	}
664 	pr_info("%d configured CPUs, %d standby CPUs\n", c_cpus, s_cpus);
665 	get_online_cpus();
666 	__smp_rescan_cpus(info, 0);
667 	put_online_cpus();
668 	kfree(info);
669 }
670 
671 /*
672  *	Activate a secondary processor.
673  */
674 static void smp_start_secondary(void *cpuvoid)
675 {
676 	S390_lowcore.last_update_clock = get_tod_clock();
677 	S390_lowcore.restart_stack = (unsigned long) restart_stack;
678 	S390_lowcore.restart_fn = (unsigned long) do_restart;
679 	S390_lowcore.restart_data = 0;
680 	S390_lowcore.restart_source = -1UL;
681 	restore_access_regs(S390_lowcore.access_regs_save_area);
682 	__ctl_load(S390_lowcore.cregs_save_area, 0, 15);
683 	__load_psw_mask(PSW_KERNEL_BITS | PSW_MASK_DAT);
684 	cpu_init();
685 	preempt_disable();
686 	init_cpu_timer();
687 	vtime_init();
688 	pfault_init();
689 	notify_cpu_starting(smp_processor_id());
690 	set_cpu_online(smp_processor_id(), true);
691 	inc_irq_stat(CPU_RST);
692 	local_irq_enable();
693 	cpu_startup_entry(CPUHP_ONLINE);
694 }
695 
696 /* Upping and downing of CPUs */
697 int __cpu_up(unsigned int cpu, struct task_struct *tidle)
698 {
699 	struct pcpu *pcpu;
700 	int rc;
701 
702 	pcpu = pcpu_devices + cpu;
703 	if (pcpu->state != CPU_STATE_CONFIGURED)
704 		return -EIO;
705 	if (pcpu_sigp_retry(pcpu, SIGP_INITIAL_CPU_RESET, 0) !=
706 	    SIGP_CC_ORDER_CODE_ACCEPTED)
707 		return -EIO;
708 
709 	rc = pcpu_alloc_lowcore(pcpu, cpu);
710 	if (rc)
711 		return rc;
712 	pcpu_prepare_secondary(pcpu, cpu);
713 	pcpu_attach_task(pcpu, tidle);
714 	pcpu_start_fn(pcpu, smp_start_secondary, NULL);
715 	while (!cpu_online(cpu))
716 		cpu_relax();
717 	return 0;
718 }
719 
720 static unsigned int setup_possible_cpus __initdata;
721 
722 static int __init _setup_possible_cpus(char *s)
723 {
724 	get_option(&s, &setup_possible_cpus);
725 	return 0;
726 }
727 early_param("possible_cpus", _setup_possible_cpus);
728 
729 #ifdef CONFIG_HOTPLUG_CPU
730 
731 int __cpu_disable(void)
732 {
733 	unsigned long cregs[16];
734 
735 	/* Handle possible pending IPIs */
736 	smp_handle_ext_call();
737 	set_cpu_online(smp_processor_id(), false);
738 	/* Disable pseudo page faults on this cpu. */
739 	pfault_fini();
740 	/* Disable interrupt sources via control register. */
741 	__ctl_store(cregs, 0, 15);
742 	cregs[0]  &= ~0x0000ee70UL;	/* disable all external interrupts */
743 	cregs[6]  &= ~0xff000000UL;	/* disable all I/O interrupts */
744 	cregs[14] &= ~0x1f000000UL;	/* disable most machine checks */
745 	__ctl_load(cregs, 0, 15);
746 	clear_cpu_flag(CIF_NOHZ_DELAY);
747 	return 0;
748 }
749 
750 void __cpu_die(unsigned int cpu)
751 {
752 	struct pcpu *pcpu;
753 
754 	/* Wait until target cpu is down */
755 	pcpu = pcpu_devices + cpu;
756 	while (!pcpu_stopped(pcpu))
757 		cpu_relax();
758 	pcpu_free_lowcore(pcpu);
759 	atomic_dec(&init_mm.context.attach_count);
760 	cpumask_clear_cpu(cpu, mm_cpumask(&init_mm));
761 	if (MACHINE_HAS_TLB_LC)
762 		cpumask_clear_cpu(cpu, &init_mm.context.cpu_attach_mask);
763 }
764 
765 void __noreturn cpu_die(void)
766 {
767 	idle_task_exit();
768 	pcpu_sigp_retry(pcpu_devices + smp_processor_id(), SIGP_STOP, 0);
769 	for (;;) ;
770 }
771 
772 #endif /* CONFIG_HOTPLUG_CPU */
773 
774 void __init smp_fill_possible_mask(void)
775 {
776 	unsigned int possible, sclp, cpu;
777 
778 	sclp = sclp_get_max_cpu() ?: nr_cpu_ids;
779 	possible = setup_possible_cpus ?: nr_cpu_ids;
780 	possible = min(possible, sclp);
781 	for (cpu = 0; cpu < possible && cpu < nr_cpu_ids; cpu++)
782 		set_cpu_possible(cpu, true);
783 }
784 
785 void __init smp_prepare_cpus(unsigned int max_cpus)
786 {
787 	/* request the 0x1201 emergency signal external interrupt */
788 	if (register_external_irq(EXT_IRQ_EMERGENCY_SIG, do_ext_call_interrupt))
789 		panic("Couldn't request external interrupt 0x1201");
790 	/* request the 0x1202 external call external interrupt */
791 	if (register_external_irq(EXT_IRQ_EXTERNAL_CALL, do_ext_call_interrupt))
792 		panic("Couldn't request external interrupt 0x1202");
793 	smp_detect_cpus();
794 }
795 
796 void __init smp_prepare_boot_cpu(void)
797 {
798 	struct pcpu *pcpu = pcpu_devices;
799 
800 	boot_cpu_address = stap();
801 	pcpu->state = CPU_STATE_CONFIGURED;
802 	pcpu->address = boot_cpu_address;
803 	pcpu->lowcore = (struct _lowcore *)(unsigned long) store_prefix();
804 	pcpu->async_stack = S390_lowcore.async_stack - ASYNC_SIZE
805 		+ STACK_FRAME_OVERHEAD + sizeof(struct pt_regs);
806 	pcpu->panic_stack = S390_lowcore.panic_stack - PAGE_SIZE
807 		+ STACK_FRAME_OVERHEAD + sizeof(struct pt_regs);
808 	S390_lowcore.percpu_offset = __per_cpu_offset[0];
809 	smp_cpu_set_polarization(0, POLARIZATION_UNKNOWN);
810 	set_cpu_present(0, true);
811 	set_cpu_online(0, true);
812 }
813 
814 void __init smp_cpus_done(unsigned int max_cpus)
815 {
816 }
817 
818 void __init smp_setup_processor_id(void)
819 {
820 	S390_lowcore.cpu_nr = 0;
821 	S390_lowcore.spinlock_lockval = arch_spin_lockval(0);
822 }
823 
824 /*
825  * the frequency of the profiling timer can be changed
826  * by writing a multiplier value into /proc/profile.
827  *
828  * usually you want to run this on all CPUs ;)
829  */
830 int setup_profiling_timer(unsigned int multiplier)
831 {
832 	return 0;
833 }
834 
835 #ifdef CONFIG_HOTPLUG_CPU
836 static ssize_t cpu_configure_show(struct device *dev,
837 				  struct device_attribute *attr, char *buf)
838 {
839 	ssize_t count;
840 
841 	mutex_lock(&smp_cpu_state_mutex);
842 	count = sprintf(buf, "%d\n", pcpu_devices[dev->id].state);
843 	mutex_unlock(&smp_cpu_state_mutex);
844 	return count;
845 }
846 
847 static ssize_t cpu_configure_store(struct device *dev,
848 				   struct device_attribute *attr,
849 				   const char *buf, size_t count)
850 {
851 	struct pcpu *pcpu;
852 	int cpu, val, rc;
853 	char delim;
854 
855 	if (sscanf(buf, "%d %c", &val, &delim) != 1)
856 		return -EINVAL;
857 	if (val != 0 && val != 1)
858 		return -EINVAL;
859 	get_online_cpus();
860 	mutex_lock(&smp_cpu_state_mutex);
861 	rc = -EBUSY;
862 	/* disallow configuration changes of online cpus and cpu 0 */
863 	cpu = dev->id;
864 	if (cpu_online(cpu) || cpu == 0)
865 		goto out;
866 	pcpu = pcpu_devices + cpu;
867 	rc = 0;
868 	switch (val) {
869 	case 0:
870 		if (pcpu->state != CPU_STATE_CONFIGURED)
871 			break;
872 		rc = sclp_cpu_deconfigure(pcpu->address);
873 		if (rc)
874 			break;
875 		pcpu->state = CPU_STATE_STANDBY;
876 		smp_cpu_set_polarization(cpu, POLARIZATION_UNKNOWN);
877 		topology_expect_change();
878 		break;
879 	case 1:
880 		if (pcpu->state != CPU_STATE_STANDBY)
881 			break;
882 		rc = sclp_cpu_configure(pcpu->address);
883 		if (rc)
884 			break;
885 		pcpu->state = CPU_STATE_CONFIGURED;
886 		smp_cpu_set_polarization(cpu, POLARIZATION_UNKNOWN);
887 		topology_expect_change();
888 		break;
889 	default:
890 		break;
891 	}
892 out:
893 	mutex_unlock(&smp_cpu_state_mutex);
894 	put_online_cpus();
895 	return rc ? rc : count;
896 }
897 static DEVICE_ATTR(configure, 0644, cpu_configure_show, cpu_configure_store);
898 #endif /* CONFIG_HOTPLUG_CPU */
899 
900 static ssize_t show_cpu_address(struct device *dev,
901 				struct device_attribute *attr, char *buf)
902 {
903 	return sprintf(buf, "%d\n", pcpu_devices[dev->id].address);
904 }
905 static DEVICE_ATTR(address, 0444, show_cpu_address, NULL);
906 
907 static struct attribute *cpu_common_attrs[] = {
908 #ifdef CONFIG_HOTPLUG_CPU
909 	&dev_attr_configure.attr,
910 #endif
911 	&dev_attr_address.attr,
912 	NULL,
913 };
914 
915 static struct attribute_group cpu_common_attr_group = {
916 	.attrs = cpu_common_attrs,
917 };
918 
919 static struct attribute *cpu_online_attrs[] = {
920 	&dev_attr_idle_count.attr,
921 	&dev_attr_idle_time_us.attr,
922 	NULL,
923 };
924 
925 static struct attribute_group cpu_online_attr_group = {
926 	.attrs = cpu_online_attrs,
927 };
928 
929 static int smp_cpu_notify(struct notifier_block *self, unsigned long action,
930 			  void *hcpu)
931 {
932 	unsigned int cpu = (unsigned int)(long)hcpu;
933 	struct cpu *c = pcpu_devices[cpu].cpu;
934 	struct device *s = &c->dev;
935 	int err = 0;
936 
937 	switch (action & ~CPU_TASKS_FROZEN) {
938 	case CPU_ONLINE:
939 		err = sysfs_create_group(&s->kobj, &cpu_online_attr_group);
940 		break;
941 	case CPU_DEAD:
942 		sysfs_remove_group(&s->kobj, &cpu_online_attr_group);
943 		break;
944 	}
945 	return notifier_from_errno(err);
946 }
947 
948 static int smp_add_present_cpu(int cpu)
949 {
950 	struct device *s;
951 	struct cpu *c;
952 	int rc;
953 
954 	c = kzalloc(sizeof(*c), GFP_KERNEL);
955 	if (!c)
956 		return -ENOMEM;
957 	pcpu_devices[cpu].cpu = c;
958 	s = &c->dev;
959 	c->hotpluggable = 1;
960 	rc = register_cpu(c, cpu);
961 	if (rc)
962 		goto out;
963 	rc = sysfs_create_group(&s->kobj, &cpu_common_attr_group);
964 	if (rc)
965 		goto out_cpu;
966 	if (cpu_online(cpu)) {
967 		rc = sysfs_create_group(&s->kobj, &cpu_online_attr_group);
968 		if (rc)
969 			goto out_online;
970 	}
971 	rc = topology_cpu_init(c);
972 	if (rc)
973 		goto out_topology;
974 	return 0;
975 
976 out_topology:
977 	if (cpu_online(cpu))
978 		sysfs_remove_group(&s->kobj, &cpu_online_attr_group);
979 out_online:
980 	sysfs_remove_group(&s->kobj, &cpu_common_attr_group);
981 out_cpu:
982 #ifdef CONFIG_HOTPLUG_CPU
983 	unregister_cpu(c);
984 #endif
985 out:
986 	return rc;
987 }
988 
989 #ifdef CONFIG_HOTPLUG_CPU
990 
991 int __ref smp_rescan_cpus(void)
992 {
993 	struct sclp_cpu_info *info;
994 	int nr;
995 
996 	info = smp_get_cpu_info();
997 	if (!info)
998 		return -ENOMEM;
999 	get_online_cpus();
1000 	mutex_lock(&smp_cpu_state_mutex);
1001 	nr = __smp_rescan_cpus(info, 1);
1002 	mutex_unlock(&smp_cpu_state_mutex);
1003 	put_online_cpus();
1004 	kfree(info);
1005 	if (nr)
1006 		topology_schedule_update();
1007 	return 0;
1008 }
1009 
1010 static ssize_t __ref rescan_store(struct device *dev,
1011 				  struct device_attribute *attr,
1012 				  const char *buf,
1013 				  size_t count)
1014 {
1015 	int rc;
1016 
1017 	rc = smp_rescan_cpus();
1018 	return rc ? rc : count;
1019 }
1020 static DEVICE_ATTR(rescan, 0200, NULL, rescan_store);
1021 #endif /* CONFIG_HOTPLUG_CPU */
1022 
1023 static int __init s390_smp_init(void)
1024 {
1025 	int cpu, rc = 0;
1026 
1027 #ifdef CONFIG_HOTPLUG_CPU
1028 	rc = device_create_file(cpu_subsys.dev_root, &dev_attr_rescan);
1029 	if (rc)
1030 		return rc;
1031 #endif
1032 	cpu_notifier_register_begin();
1033 	for_each_present_cpu(cpu) {
1034 		rc = smp_add_present_cpu(cpu);
1035 		if (rc)
1036 			goto out;
1037 	}
1038 
1039 	__hotcpu_notifier(smp_cpu_notify, 0);
1040 
1041 out:
1042 	cpu_notifier_register_done();
1043 	return rc;
1044 }
1045 subsys_initcall(s390_smp_init);
1046