xref: /openbmc/linux/arch/s390/kernel/smp.c (revision 63705da3)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  *  SMP related functions
4  *
5  *    Copyright IBM Corp. 1999, 2012
6  *    Author(s): Denis Joseph Barrow,
7  *		 Martin Schwidefsky <schwidefsky@de.ibm.com>,
8  *		 Heiko Carstens <heiko.carstens@de.ibm.com>,
9  *
10  *  based on other smp stuff by
11  *    (c) 1995 Alan Cox, CymruNET Ltd  <alan@cymru.net>
12  *    (c) 1998 Ingo Molnar
13  *
14  * The code outside of smp.c uses logical cpu numbers, only smp.c does
15  * the translation of logical to physical cpu ids. All new code that
16  * operates on physical cpu numbers needs to go into smp.c.
17  */
18 
19 #define KMSG_COMPONENT "cpu"
20 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
21 
22 #include <linux/workqueue.h>
23 #include <linux/memblock.h>
24 #include <linux/export.h>
25 #include <linux/init.h>
26 #include <linux/mm.h>
27 #include <linux/err.h>
28 #include <linux/spinlock.h>
29 #include <linux/kernel_stat.h>
30 #include <linux/delay.h>
31 #include <linux/interrupt.h>
32 #include <linux/irqflags.h>
33 #include <linux/irq_work.h>
34 #include <linux/cpu.h>
35 #include <linux/slab.h>
36 #include <linux/sched/hotplug.h>
37 #include <linux/sched/task_stack.h>
38 #include <linux/crash_dump.h>
39 #include <linux/kprobes.h>
40 #include <asm/asm-offsets.h>
41 #include <asm/diag.h>
42 #include <asm/switch_to.h>
43 #include <asm/facility.h>
44 #include <asm/ipl.h>
45 #include <asm/setup.h>
46 #include <asm/irq.h>
47 #include <asm/tlbflush.h>
48 #include <asm/vtimer.h>
49 #include <asm/lowcore.h>
50 #include <asm/sclp.h>
51 #include <asm/debug.h>
52 #include <asm/os_info.h>
53 #include <asm/sigp.h>
54 #include <asm/idle.h>
55 #include <asm/nmi.h>
56 #include <asm/stacktrace.h>
57 #include <asm/topology.h>
58 #include <asm/vdso.h>
59 #include "entry.h"
60 
61 enum {
62 	ec_schedule = 0,
63 	ec_call_function_single,
64 	ec_stop_cpu,
65 	ec_mcck_pending,
66 	ec_irq_work,
67 };
68 
69 enum {
70 	CPU_STATE_STANDBY,
71 	CPU_STATE_CONFIGURED,
72 };
73 
74 static DEFINE_PER_CPU(struct cpu *, cpu_device);
75 
76 struct pcpu {
77 	unsigned long ec_mask;		/* bit mask for ec_xxx functions */
78 	unsigned long ec_clk;		/* sigp timestamp for ec_xxx */
79 	signed char state;		/* physical cpu state */
80 	signed char polarization;	/* physical polarization */
81 	u16 address;			/* physical cpu address */
82 };
83 
84 static u8 boot_core_type;
85 static struct pcpu pcpu_devices[NR_CPUS];
86 
87 unsigned int smp_cpu_mt_shift;
88 EXPORT_SYMBOL(smp_cpu_mt_shift);
89 
90 unsigned int smp_cpu_mtid;
91 EXPORT_SYMBOL(smp_cpu_mtid);
92 
93 #ifdef CONFIG_CRASH_DUMP
94 __vector128 __initdata boot_cpu_vector_save_area[__NUM_VXRS];
95 #endif
96 
97 static unsigned int smp_max_threads __initdata = -1U;
98 cpumask_t cpu_setup_mask;
99 
100 static int __init early_nosmt(char *s)
101 {
102 	smp_max_threads = 1;
103 	return 0;
104 }
105 early_param("nosmt", early_nosmt);
106 
107 static int __init early_smt(char *s)
108 {
109 	get_option(&s, &smp_max_threads);
110 	return 0;
111 }
112 early_param("smt", early_smt);
113 
114 /*
115  * The smp_cpu_state_mutex must be held when changing the state or polarization
116  * member of a pcpu data structure within the pcpu_devices arreay.
117  */
118 DEFINE_MUTEX(smp_cpu_state_mutex);
119 
120 /*
121  * Signal processor helper functions.
122  */
123 static inline int __pcpu_sigp_relax(u16 addr, u8 order, unsigned long parm)
124 {
125 	int cc;
126 
127 	while (1) {
128 		cc = __pcpu_sigp(addr, order, parm, NULL);
129 		if (cc != SIGP_CC_BUSY)
130 			return cc;
131 		cpu_relax();
132 	}
133 }
134 
135 static int pcpu_sigp_retry(struct pcpu *pcpu, u8 order, u32 parm)
136 {
137 	int cc, retry;
138 
139 	for (retry = 0; ; retry++) {
140 		cc = __pcpu_sigp(pcpu->address, order, parm, NULL);
141 		if (cc != SIGP_CC_BUSY)
142 			break;
143 		if (retry >= 3)
144 			udelay(10);
145 	}
146 	return cc;
147 }
148 
149 static inline int pcpu_stopped(struct pcpu *pcpu)
150 {
151 	u32 status;
152 
153 	if (__pcpu_sigp(pcpu->address, SIGP_SENSE,
154 			0, &status) != SIGP_CC_STATUS_STORED)
155 		return 0;
156 	return !!(status & (SIGP_STATUS_CHECK_STOP|SIGP_STATUS_STOPPED));
157 }
158 
159 static inline int pcpu_running(struct pcpu *pcpu)
160 {
161 	if (__pcpu_sigp(pcpu->address, SIGP_SENSE_RUNNING,
162 			0, NULL) != SIGP_CC_STATUS_STORED)
163 		return 1;
164 	/* Status stored condition code is equivalent to cpu not running. */
165 	return 0;
166 }
167 
168 /*
169  * Find struct pcpu by cpu address.
170  */
171 static struct pcpu *pcpu_find_address(const struct cpumask *mask, u16 address)
172 {
173 	int cpu;
174 
175 	for_each_cpu(cpu, mask)
176 		if (pcpu_devices[cpu].address == address)
177 			return pcpu_devices + cpu;
178 	return NULL;
179 }
180 
181 static void pcpu_ec_call(struct pcpu *pcpu, int ec_bit)
182 {
183 	int order;
184 
185 	if (test_and_set_bit(ec_bit, &pcpu->ec_mask))
186 		return;
187 	order = pcpu_running(pcpu) ? SIGP_EXTERNAL_CALL : SIGP_EMERGENCY_SIGNAL;
188 	pcpu->ec_clk = get_tod_clock_fast();
189 	pcpu_sigp_retry(pcpu, order, 0);
190 }
191 
192 static int pcpu_alloc_lowcore(struct pcpu *pcpu, int cpu)
193 {
194 	unsigned long async_stack, nodat_stack, mcck_stack;
195 	struct lowcore *lc;
196 
197 	lc = (struct lowcore *) __get_free_pages(GFP_KERNEL | GFP_DMA, LC_ORDER);
198 	nodat_stack = __get_free_pages(GFP_KERNEL, THREAD_SIZE_ORDER);
199 	async_stack = stack_alloc();
200 	mcck_stack = stack_alloc();
201 	if (!lc || !nodat_stack || !async_stack || !mcck_stack)
202 		goto out;
203 	memcpy(lc, &S390_lowcore, 512);
204 	memset((char *) lc + 512, 0, sizeof(*lc) - 512);
205 	lc->async_stack = async_stack + STACK_INIT_OFFSET;
206 	lc->nodat_stack = nodat_stack + STACK_INIT_OFFSET;
207 	lc->mcck_stack = mcck_stack + STACK_INIT_OFFSET;
208 	lc->cpu_nr = cpu;
209 	lc->spinlock_lockval = arch_spin_lockval(cpu);
210 	lc->spinlock_index = 0;
211 	lc->br_r1_trampoline = 0x07f1;	/* br %r1 */
212 	lc->return_lpswe = gen_lpswe(__LC_RETURN_PSW);
213 	lc->return_mcck_lpswe = gen_lpswe(__LC_RETURN_MCCK_PSW);
214 	lc->preempt_count = PREEMPT_DISABLED;
215 	if (nmi_alloc_per_cpu(lc))
216 		goto out;
217 	lowcore_ptr[cpu] = lc;
218 	pcpu_sigp_retry(pcpu, SIGP_SET_PREFIX, (u32)(unsigned long) lc);
219 	return 0;
220 
221 out:
222 	stack_free(mcck_stack);
223 	stack_free(async_stack);
224 	free_pages(nodat_stack, THREAD_SIZE_ORDER);
225 	free_pages((unsigned long) lc, LC_ORDER);
226 	return -ENOMEM;
227 }
228 
229 static void pcpu_free_lowcore(struct pcpu *pcpu)
230 {
231 	unsigned long async_stack, nodat_stack, mcck_stack;
232 	struct lowcore *lc;
233 	int cpu;
234 
235 	cpu = pcpu - pcpu_devices;
236 	lc = lowcore_ptr[cpu];
237 	nodat_stack = lc->nodat_stack - STACK_INIT_OFFSET;
238 	async_stack = lc->async_stack - STACK_INIT_OFFSET;
239 	mcck_stack = lc->mcck_stack - STACK_INIT_OFFSET;
240 	pcpu_sigp_retry(pcpu, SIGP_SET_PREFIX, 0);
241 	lowcore_ptr[cpu] = NULL;
242 	nmi_free_per_cpu(lc);
243 	stack_free(async_stack);
244 	stack_free(mcck_stack);
245 	free_pages(nodat_stack, THREAD_SIZE_ORDER);
246 	free_pages((unsigned long) lc, LC_ORDER);
247 }
248 
249 static void pcpu_prepare_secondary(struct pcpu *pcpu, int cpu)
250 {
251 	struct lowcore *lc = lowcore_ptr[cpu];
252 
253 	cpumask_set_cpu(cpu, &init_mm.context.cpu_attach_mask);
254 	cpumask_set_cpu(cpu, mm_cpumask(&init_mm));
255 	lc->cpu_nr = cpu;
256 	lc->restart_flags = RESTART_FLAG_CTLREGS;
257 	lc->spinlock_lockval = arch_spin_lockval(cpu);
258 	lc->spinlock_index = 0;
259 	lc->percpu_offset = __per_cpu_offset[cpu];
260 	lc->kernel_asce = S390_lowcore.kernel_asce;
261 	lc->user_asce = s390_invalid_asce;
262 	lc->machine_flags = S390_lowcore.machine_flags;
263 	lc->user_timer = lc->system_timer =
264 		lc->steal_timer = lc->avg_steal_timer = 0;
265 	__ctl_store(lc->cregs_save_area, 0, 15);
266 	lc->cregs_save_area[1] = lc->kernel_asce;
267 	lc->cregs_save_area[7] = lc->user_asce;
268 	save_access_regs((unsigned int *) lc->access_regs_save_area);
269 	arch_spin_lock_setup(cpu);
270 }
271 
272 static void pcpu_attach_task(struct pcpu *pcpu, struct task_struct *tsk)
273 {
274 	struct lowcore *lc;
275 	int cpu;
276 
277 	cpu = pcpu - pcpu_devices;
278 	lc = lowcore_ptr[cpu];
279 	lc->kernel_stack = (unsigned long) task_stack_page(tsk)
280 		+ THREAD_SIZE - STACK_FRAME_OVERHEAD - sizeof(struct pt_regs);
281 	lc->current_task = (unsigned long) tsk;
282 	lc->lpp = LPP_MAGIC;
283 	lc->current_pid = tsk->pid;
284 	lc->user_timer = tsk->thread.user_timer;
285 	lc->guest_timer = tsk->thread.guest_timer;
286 	lc->system_timer = tsk->thread.system_timer;
287 	lc->hardirq_timer = tsk->thread.hardirq_timer;
288 	lc->softirq_timer = tsk->thread.softirq_timer;
289 	lc->steal_timer = 0;
290 }
291 
292 static void pcpu_start_fn(struct pcpu *pcpu, void (*func)(void *), void *data)
293 {
294 	struct lowcore *lc;
295 	int cpu;
296 
297 	cpu = pcpu - pcpu_devices;
298 	lc = lowcore_ptr[cpu];
299 	lc->restart_stack = lc->kernel_stack;
300 	lc->restart_fn = (unsigned long) func;
301 	lc->restart_data = (unsigned long) data;
302 	lc->restart_source = -1U;
303 	pcpu_sigp_retry(pcpu, SIGP_RESTART, 0);
304 }
305 
306 typedef void (pcpu_delegate_fn)(void *);
307 
308 /*
309  * Call function via PSW restart on pcpu and stop the current cpu.
310  */
311 static void __pcpu_delegate(pcpu_delegate_fn *func, void *data)
312 {
313 	func(data);	/* should not return */
314 }
315 
316 static void pcpu_delegate(struct pcpu *pcpu,
317 			  pcpu_delegate_fn *func,
318 			  void *data, unsigned long stack)
319 {
320 	struct lowcore *lc = lowcore_ptr[pcpu - pcpu_devices];
321 	unsigned int source_cpu = stap();
322 
323 	__load_psw_mask(PSW_KERNEL_BITS | PSW_MASK_DAT);
324 	if (pcpu->address == source_cpu) {
325 		call_on_stack(2, stack, void, __pcpu_delegate,
326 			      pcpu_delegate_fn *, func, void *, data);
327 	}
328 	/* Stop target cpu (if func returns this stops the current cpu). */
329 	pcpu_sigp_retry(pcpu, SIGP_STOP, 0);
330 	/* Restart func on the target cpu and stop the current cpu. */
331 	mem_assign_absolute(lc->restart_stack, stack);
332 	mem_assign_absolute(lc->restart_fn, (unsigned long) func);
333 	mem_assign_absolute(lc->restart_data, (unsigned long) data);
334 	mem_assign_absolute(lc->restart_source, source_cpu);
335 	__bpon();
336 	asm volatile(
337 		"0:	sigp	0,%0,%2	# sigp restart to target cpu\n"
338 		"	brc	2,0b	# busy, try again\n"
339 		"1:	sigp	0,%1,%3	# sigp stop to current cpu\n"
340 		"	brc	2,1b	# busy, try again\n"
341 		: : "d" (pcpu->address), "d" (source_cpu),
342 		    "K" (SIGP_RESTART), "K" (SIGP_STOP)
343 		: "0", "1", "cc");
344 	for (;;) ;
345 }
346 
347 /*
348  * Enable additional logical cpus for multi-threading.
349  */
350 static int pcpu_set_smt(unsigned int mtid)
351 {
352 	int cc;
353 
354 	if (smp_cpu_mtid == mtid)
355 		return 0;
356 	cc = __pcpu_sigp(0, SIGP_SET_MULTI_THREADING, mtid, NULL);
357 	if (cc == 0) {
358 		smp_cpu_mtid = mtid;
359 		smp_cpu_mt_shift = 0;
360 		while (smp_cpu_mtid >= (1U << smp_cpu_mt_shift))
361 			smp_cpu_mt_shift++;
362 		pcpu_devices[0].address = stap();
363 	}
364 	return cc;
365 }
366 
367 /*
368  * Call function on an online CPU.
369  */
370 void smp_call_online_cpu(void (*func)(void *), void *data)
371 {
372 	struct pcpu *pcpu;
373 
374 	/* Use the current cpu if it is online. */
375 	pcpu = pcpu_find_address(cpu_online_mask, stap());
376 	if (!pcpu)
377 		/* Use the first online cpu. */
378 		pcpu = pcpu_devices + cpumask_first(cpu_online_mask);
379 	pcpu_delegate(pcpu, func, data, (unsigned long) restart_stack);
380 }
381 
382 /*
383  * Call function on the ipl CPU.
384  */
385 void smp_call_ipl_cpu(void (*func)(void *), void *data)
386 {
387 	struct lowcore *lc = lowcore_ptr[0];
388 
389 	if (pcpu_devices[0].address == stap())
390 		lc = &S390_lowcore;
391 
392 	pcpu_delegate(&pcpu_devices[0], func, data,
393 		      lc->nodat_stack);
394 }
395 
396 int smp_find_processor_id(u16 address)
397 {
398 	int cpu;
399 
400 	for_each_present_cpu(cpu)
401 		if (pcpu_devices[cpu].address == address)
402 			return cpu;
403 	return -1;
404 }
405 
406 void schedule_mcck_handler(void)
407 {
408 	pcpu_ec_call(pcpu_devices + smp_processor_id(), ec_mcck_pending);
409 }
410 
411 bool notrace arch_vcpu_is_preempted(int cpu)
412 {
413 	if (test_cpu_flag_of(CIF_ENABLED_WAIT, cpu))
414 		return false;
415 	if (pcpu_running(pcpu_devices + cpu))
416 		return false;
417 	return true;
418 }
419 EXPORT_SYMBOL(arch_vcpu_is_preempted);
420 
421 void notrace smp_yield_cpu(int cpu)
422 {
423 	if (!MACHINE_HAS_DIAG9C)
424 		return;
425 	diag_stat_inc_norecursion(DIAG_STAT_X09C);
426 	asm volatile("diag %0,0,0x9c"
427 		     : : "d" (pcpu_devices[cpu].address));
428 }
429 EXPORT_SYMBOL_GPL(smp_yield_cpu);
430 
431 /*
432  * Send cpus emergency shutdown signal. This gives the cpus the
433  * opportunity to complete outstanding interrupts.
434  */
435 void notrace smp_emergency_stop(void)
436 {
437 	static arch_spinlock_t lock = __ARCH_SPIN_LOCK_UNLOCKED;
438 	static cpumask_t cpumask;
439 	u64 end;
440 	int cpu;
441 
442 	arch_spin_lock(&lock);
443 	cpumask_copy(&cpumask, cpu_online_mask);
444 	cpumask_clear_cpu(smp_processor_id(), &cpumask);
445 
446 	end = get_tod_clock() + (1000000UL << 12);
447 	for_each_cpu(cpu, &cpumask) {
448 		struct pcpu *pcpu = pcpu_devices + cpu;
449 		set_bit(ec_stop_cpu, &pcpu->ec_mask);
450 		while (__pcpu_sigp(pcpu->address, SIGP_EMERGENCY_SIGNAL,
451 				   0, NULL) == SIGP_CC_BUSY &&
452 		       get_tod_clock() < end)
453 			cpu_relax();
454 	}
455 	while (get_tod_clock() < end) {
456 		for_each_cpu(cpu, &cpumask)
457 			if (pcpu_stopped(pcpu_devices + cpu))
458 				cpumask_clear_cpu(cpu, &cpumask);
459 		if (cpumask_empty(&cpumask))
460 			break;
461 		cpu_relax();
462 	}
463 	arch_spin_unlock(&lock);
464 }
465 NOKPROBE_SYMBOL(smp_emergency_stop);
466 
467 /*
468  * Stop all cpus but the current one.
469  */
470 void smp_send_stop(void)
471 {
472 	int cpu;
473 
474 	/* Disable all interrupts/machine checks */
475 	__load_psw_mask(PSW_KERNEL_BITS | PSW_MASK_DAT);
476 	trace_hardirqs_off();
477 
478 	debug_set_critical();
479 
480 	if (oops_in_progress)
481 		smp_emergency_stop();
482 
483 	/* stop all processors */
484 	for_each_online_cpu(cpu) {
485 		if (cpu == smp_processor_id())
486 			continue;
487 		pcpu_sigp_retry(pcpu_devices + cpu, SIGP_STOP, 0);
488 		while (!pcpu_stopped(pcpu_devices + cpu))
489 			cpu_relax();
490 	}
491 }
492 
493 /*
494  * This is the main routine where commands issued by other
495  * cpus are handled.
496  */
497 static void smp_handle_ext_call(void)
498 {
499 	unsigned long bits;
500 
501 	/* handle bit signal external calls */
502 	bits = xchg(&pcpu_devices[smp_processor_id()].ec_mask, 0);
503 	if (test_bit(ec_stop_cpu, &bits))
504 		smp_stop_cpu();
505 	if (test_bit(ec_schedule, &bits))
506 		scheduler_ipi();
507 	if (test_bit(ec_call_function_single, &bits))
508 		generic_smp_call_function_single_interrupt();
509 	if (test_bit(ec_mcck_pending, &bits))
510 		__s390_handle_mcck();
511 	if (test_bit(ec_irq_work, &bits))
512 		irq_work_run();
513 }
514 
515 static void do_ext_call_interrupt(struct ext_code ext_code,
516 				  unsigned int param32, unsigned long param64)
517 {
518 	inc_irq_stat(ext_code.code == 0x1202 ? IRQEXT_EXC : IRQEXT_EMS);
519 	smp_handle_ext_call();
520 }
521 
522 void arch_send_call_function_ipi_mask(const struct cpumask *mask)
523 {
524 	int cpu;
525 
526 	for_each_cpu(cpu, mask)
527 		pcpu_ec_call(pcpu_devices + cpu, ec_call_function_single);
528 }
529 
530 void arch_send_call_function_single_ipi(int cpu)
531 {
532 	pcpu_ec_call(pcpu_devices + cpu, ec_call_function_single);
533 }
534 
535 /*
536  * this function sends a 'reschedule' IPI to another CPU.
537  * it goes straight through and wastes no time serializing
538  * anything. Worst case is that we lose a reschedule ...
539  */
540 void smp_send_reschedule(int cpu)
541 {
542 	pcpu_ec_call(pcpu_devices + cpu, ec_schedule);
543 }
544 
545 #ifdef CONFIG_IRQ_WORK
546 void arch_irq_work_raise(void)
547 {
548 	pcpu_ec_call(pcpu_devices + smp_processor_id(), ec_irq_work);
549 }
550 #endif
551 
552 /*
553  * parameter area for the set/clear control bit callbacks
554  */
555 struct ec_creg_mask_parms {
556 	unsigned long orval;
557 	unsigned long andval;
558 	int cr;
559 };
560 
561 /*
562  * callback for setting/clearing control bits
563  */
564 static void smp_ctl_bit_callback(void *info)
565 {
566 	struct ec_creg_mask_parms *pp = info;
567 	unsigned long cregs[16];
568 
569 	__ctl_store(cregs, 0, 15);
570 	cregs[pp->cr] = (cregs[pp->cr] & pp->andval) | pp->orval;
571 	__ctl_load(cregs, 0, 15);
572 }
573 
574 static DEFINE_SPINLOCK(ctl_lock);
575 static unsigned long ctlreg;
576 
577 /*
578  * Set a bit in a control register of all cpus
579  */
580 void smp_ctl_set_bit(int cr, int bit)
581 {
582 	struct ec_creg_mask_parms parms = { 1UL << bit, -1UL, cr };
583 
584 	spin_lock(&ctl_lock);
585 	memcpy_absolute(&ctlreg, &S390_lowcore.cregs_save_area[cr], sizeof(ctlreg));
586 	__set_bit(bit, &ctlreg);
587 	memcpy_absolute(&S390_lowcore.cregs_save_area[cr], &ctlreg, sizeof(ctlreg));
588 	spin_unlock(&ctl_lock);
589 	on_each_cpu(smp_ctl_bit_callback, &parms, 1);
590 }
591 EXPORT_SYMBOL(smp_ctl_set_bit);
592 
593 /*
594  * Clear a bit in a control register of all cpus
595  */
596 void smp_ctl_clear_bit(int cr, int bit)
597 {
598 	struct ec_creg_mask_parms parms = { 0, ~(1UL << bit), cr };
599 
600 	spin_lock(&ctl_lock);
601 	memcpy_absolute(&ctlreg, &S390_lowcore.cregs_save_area[cr], sizeof(ctlreg));
602 	__clear_bit(bit, &ctlreg);
603 	memcpy_absolute(&S390_lowcore.cregs_save_area[cr], &ctlreg, sizeof(ctlreg));
604 	spin_unlock(&ctl_lock);
605 	on_each_cpu(smp_ctl_bit_callback, &parms, 1);
606 }
607 EXPORT_SYMBOL(smp_ctl_clear_bit);
608 
609 #ifdef CONFIG_CRASH_DUMP
610 
611 int smp_store_status(int cpu)
612 {
613 	struct lowcore *lc;
614 	struct pcpu *pcpu;
615 	unsigned long pa;
616 
617 	pcpu = pcpu_devices + cpu;
618 	lc = lowcore_ptr[cpu];
619 	pa = __pa(&lc->floating_pt_save_area);
620 	if (__pcpu_sigp_relax(pcpu->address, SIGP_STORE_STATUS_AT_ADDRESS,
621 			      pa) != SIGP_CC_ORDER_CODE_ACCEPTED)
622 		return -EIO;
623 	if (!MACHINE_HAS_VX && !MACHINE_HAS_GS)
624 		return 0;
625 	pa = __pa(lc->mcesad & MCESA_ORIGIN_MASK);
626 	if (MACHINE_HAS_GS)
627 		pa |= lc->mcesad & MCESA_LC_MASK;
628 	if (__pcpu_sigp_relax(pcpu->address, SIGP_STORE_ADDITIONAL_STATUS,
629 			      pa) != SIGP_CC_ORDER_CODE_ACCEPTED)
630 		return -EIO;
631 	return 0;
632 }
633 
634 /*
635  * Collect CPU state of the previous, crashed system.
636  * There are four cases:
637  * 1) standard zfcp/nvme dump
638  *    condition: OLDMEM_BASE == NULL && is_ipl_type_dump() == true
639  *    The state for all CPUs except the boot CPU needs to be collected
640  *    with sigp stop-and-store-status. The boot CPU state is located in
641  *    the absolute lowcore of the memory stored in the HSA. The zcore code
642  *    will copy the boot CPU state from the HSA.
643  * 2) stand-alone kdump for SCSI/NVMe (zfcp/nvme dump with swapped memory)
644  *    condition: OLDMEM_BASE != NULL && is_ipl_type_dump() == true
645  *    The state for all CPUs except the boot CPU needs to be collected
646  *    with sigp stop-and-store-status. The firmware or the boot-loader
647  *    stored the registers of the boot CPU in the absolute lowcore in the
648  *    memory of the old system.
649  * 3) kdump and the old kernel did not store the CPU state,
650  *    or stand-alone kdump for DASD
651  *    condition: OLDMEM_BASE != NULL && !is_kdump_kernel()
652  *    The state for all CPUs except the boot CPU needs to be collected
653  *    with sigp stop-and-store-status. The kexec code or the boot-loader
654  *    stored the registers of the boot CPU in the memory of the old system.
655  * 4) kdump and the old kernel stored the CPU state
656  *    condition: OLDMEM_BASE != NULL && is_kdump_kernel()
657  *    This case does not exist for s390 anymore, setup_arch explicitly
658  *    deactivates the elfcorehdr= kernel parameter
659  */
660 static __init void smp_save_cpu_vxrs(struct save_area *sa, u16 addr,
661 				     bool is_boot_cpu, unsigned long page)
662 {
663 	__vector128 *vxrs = (__vector128 *) page;
664 
665 	if (is_boot_cpu)
666 		vxrs = boot_cpu_vector_save_area;
667 	else
668 		__pcpu_sigp_relax(addr, SIGP_STORE_ADDITIONAL_STATUS, page);
669 	save_area_add_vxrs(sa, vxrs);
670 }
671 
672 static __init void smp_save_cpu_regs(struct save_area *sa, u16 addr,
673 				     bool is_boot_cpu, unsigned long page)
674 {
675 	void *regs = (void *) page;
676 
677 	if (is_boot_cpu)
678 		copy_oldmem_kernel(regs, (void *) __LC_FPREGS_SAVE_AREA, 512);
679 	else
680 		__pcpu_sigp_relax(addr, SIGP_STORE_STATUS_AT_ADDRESS, page);
681 	save_area_add_regs(sa, regs);
682 }
683 
684 void __init smp_save_dump_cpus(void)
685 {
686 	int addr, boot_cpu_addr, max_cpu_addr;
687 	struct save_area *sa;
688 	unsigned long page;
689 	bool is_boot_cpu;
690 
691 	if (!(oldmem_data.start || is_ipl_type_dump()))
692 		/* No previous system present, normal boot. */
693 		return;
694 	/* Allocate a page as dumping area for the store status sigps */
695 	page = memblock_phys_alloc_range(PAGE_SIZE, PAGE_SIZE, 0, 1UL << 31);
696 	if (!page)
697 		panic("ERROR: Failed to allocate %lx bytes below %lx\n",
698 		      PAGE_SIZE, 1UL << 31);
699 
700 	/* Set multi-threading state to the previous system. */
701 	pcpu_set_smt(sclp.mtid_prev);
702 	boot_cpu_addr = stap();
703 	max_cpu_addr = SCLP_MAX_CORES << sclp.mtid_prev;
704 	for (addr = 0; addr <= max_cpu_addr; addr++) {
705 		if (__pcpu_sigp_relax(addr, SIGP_SENSE, 0) ==
706 		    SIGP_CC_NOT_OPERATIONAL)
707 			continue;
708 		is_boot_cpu = (addr == boot_cpu_addr);
709 		/* Allocate save area */
710 		sa = save_area_alloc(is_boot_cpu);
711 		if (!sa)
712 			panic("could not allocate memory for save area\n");
713 		if (MACHINE_HAS_VX)
714 			/* Get the vector registers */
715 			smp_save_cpu_vxrs(sa, addr, is_boot_cpu, page);
716 		/*
717 		 * For a zfcp/nvme dump OLDMEM_BASE == NULL and the registers
718 		 * of the boot CPU are stored in the HSA. To retrieve
719 		 * these registers an SCLP request is required which is
720 		 * done by drivers/s390/char/zcore.c:init_cpu_info()
721 		 */
722 		if (!is_boot_cpu || oldmem_data.start)
723 			/* Get the CPU registers */
724 			smp_save_cpu_regs(sa, addr, is_boot_cpu, page);
725 	}
726 	memblock_phys_free(page, PAGE_SIZE);
727 	diag_amode31_ops.diag308_reset();
728 	pcpu_set_smt(0);
729 }
730 #endif /* CONFIG_CRASH_DUMP */
731 
732 void smp_cpu_set_polarization(int cpu, int val)
733 {
734 	pcpu_devices[cpu].polarization = val;
735 }
736 
737 int smp_cpu_get_polarization(int cpu)
738 {
739 	return pcpu_devices[cpu].polarization;
740 }
741 
742 int smp_cpu_get_cpu_address(int cpu)
743 {
744 	return pcpu_devices[cpu].address;
745 }
746 
747 static void __ref smp_get_core_info(struct sclp_core_info *info, int early)
748 {
749 	static int use_sigp_detection;
750 	int address;
751 
752 	if (use_sigp_detection || sclp_get_core_info(info, early)) {
753 		use_sigp_detection = 1;
754 		for (address = 0;
755 		     address < (SCLP_MAX_CORES << smp_cpu_mt_shift);
756 		     address += (1U << smp_cpu_mt_shift)) {
757 			if (__pcpu_sigp_relax(address, SIGP_SENSE, 0) ==
758 			    SIGP_CC_NOT_OPERATIONAL)
759 				continue;
760 			info->core[info->configured].core_id =
761 				address >> smp_cpu_mt_shift;
762 			info->configured++;
763 		}
764 		info->combined = info->configured;
765 	}
766 }
767 
768 static int smp_add_present_cpu(int cpu);
769 
770 static int smp_add_core(struct sclp_core_entry *core, cpumask_t *avail,
771 			bool configured, bool early)
772 {
773 	struct pcpu *pcpu;
774 	int cpu, nr, i;
775 	u16 address;
776 
777 	nr = 0;
778 	if (sclp.has_core_type && core->type != boot_core_type)
779 		return nr;
780 	cpu = cpumask_first(avail);
781 	address = core->core_id << smp_cpu_mt_shift;
782 	for (i = 0; (i <= smp_cpu_mtid) && (cpu < nr_cpu_ids); i++) {
783 		if (pcpu_find_address(cpu_present_mask, address + i))
784 			continue;
785 		pcpu = pcpu_devices + cpu;
786 		pcpu->address = address + i;
787 		if (configured)
788 			pcpu->state = CPU_STATE_CONFIGURED;
789 		else
790 			pcpu->state = CPU_STATE_STANDBY;
791 		smp_cpu_set_polarization(cpu, POLARIZATION_UNKNOWN);
792 		set_cpu_present(cpu, true);
793 		if (!early && smp_add_present_cpu(cpu) != 0)
794 			set_cpu_present(cpu, false);
795 		else
796 			nr++;
797 		cpumask_clear_cpu(cpu, avail);
798 		cpu = cpumask_next(cpu, avail);
799 	}
800 	return nr;
801 }
802 
803 static int __smp_rescan_cpus(struct sclp_core_info *info, bool early)
804 {
805 	struct sclp_core_entry *core;
806 	static cpumask_t avail;
807 	bool configured;
808 	u16 core_id;
809 	int nr, i;
810 
811 	cpus_read_lock();
812 	mutex_lock(&smp_cpu_state_mutex);
813 	nr = 0;
814 	cpumask_xor(&avail, cpu_possible_mask, cpu_present_mask);
815 	/*
816 	 * Add IPL core first (which got logical CPU number 0) to make sure
817 	 * that all SMT threads get subsequent logical CPU numbers.
818 	 */
819 	if (early) {
820 		core_id = pcpu_devices[0].address >> smp_cpu_mt_shift;
821 		for (i = 0; i < info->configured; i++) {
822 			core = &info->core[i];
823 			if (core->core_id == core_id) {
824 				nr += smp_add_core(core, &avail, true, early);
825 				break;
826 			}
827 		}
828 	}
829 	for (i = 0; i < info->combined; i++) {
830 		configured = i < info->configured;
831 		nr += smp_add_core(&info->core[i], &avail, configured, early);
832 	}
833 	mutex_unlock(&smp_cpu_state_mutex);
834 	cpus_read_unlock();
835 	return nr;
836 }
837 
838 void __init smp_detect_cpus(void)
839 {
840 	unsigned int cpu, mtid, c_cpus, s_cpus;
841 	struct sclp_core_info *info;
842 	u16 address;
843 
844 	/* Get CPU information */
845 	info = memblock_alloc(sizeof(*info), 8);
846 	if (!info)
847 		panic("%s: Failed to allocate %zu bytes align=0x%x\n",
848 		      __func__, sizeof(*info), 8);
849 	smp_get_core_info(info, 1);
850 	/* Find boot CPU type */
851 	if (sclp.has_core_type) {
852 		address = stap();
853 		for (cpu = 0; cpu < info->combined; cpu++)
854 			if (info->core[cpu].core_id == address) {
855 				/* The boot cpu dictates the cpu type. */
856 				boot_core_type = info->core[cpu].type;
857 				break;
858 			}
859 		if (cpu >= info->combined)
860 			panic("Could not find boot CPU type");
861 	}
862 
863 	/* Set multi-threading state for the current system */
864 	mtid = boot_core_type ? sclp.mtid : sclp.mtid_cp;
865 	mtid = (mtid < smp_max_threads) ? mtid : smp_max_threads - 1;
866 	pcpu_set_smt(mtid);
867 
868 	/* Print number of CPUs */
869 	c_cpus = s_cpus = 0;
870 	for (cpu = 0; cpu < info->combined; cpu++) {
871 		if (sclp.has_core_type &&
872 		    info->core[cpu].type != boot_core_type)
873 			continue;
874 		if (cpu < info->configured)
875 			c_cpus += smp_cpu_mtid + 1;
876 		else
877 			s_cpus += smp_cpu_mtid + 1;
878 	}
879 	pr_info("%d configured CPUs, %d standby CPUs\n", c_cpus, s_cpus);
880 
881 	/* Add CPUs present at boot */
882 	__smp_rescan_cpus(info, true);
883 	memblock_phys_free((unsigned long)info, sizeof(*info));
884 }
885 
886 /*
887  *	Activate a secondary processor.
888  */
889 static void smp_start_secondary(void *cpuvoid)
890 {
891 	int cpu = raw_smp_processor_id();
892 
893 	S390_lowcore.last_update_clock = get_tod_clock();
894 	S390_lowcore.restart_stack = (unsigned long)restart_stack;
895 	S390_lowcore.restart_fn = (unsigned long)do_restart;
896 	S390_lowcore.restart_data = 0;
897 	S390_lowcore.restart_source = -1U;
898 	S390_lowcore.restart_flags = 0;
899 	restore_access_regs(S390_lowcore.access_regs_save_area);
900 	cpu_init();
901 	rcu_cpu_starting(cpu);
902 	init_cpu_timer();
903 	vtime_init();
904 	vdso_getcpu_init();
905 	pfault_init();
906 	cpumask_set_cpu(cpu, &cpu_setup_mask);
907 	update_cpu_masks();
908 	notify_cpu_starting(cpu);
909 	if (topology_cpu_dedicated(cpu))
910 		set_cpu_flag(CIF_DEDICATED_CPU);
911 	else
912 		clear_cpu_flag(CIF_DEDICATED_CPU);
913 	set_cpu_online(cpu, true);
914 	inc_irq_stat(CPU_RST);
915 	local_irq_enable();
916 	cpu_startup_entry(CPUHP_AP_ONLINE_IDLE);
917 }
918 
919 /* Upping and downing of CPUs */
920 int __cpu_up(unsigned int cpu, struct task_struct *tidle)
921 {
922 	struct pcpu *pcpu = pcpu_devices + cpu;
923 	int rc;
924 
925 	if (pcpu->state != CPU_STATE_CONFIGURED)
926 		return -EIO;
927 	if (pcpu_sigp_retry(pcpu, SIGP_INITIAL_CPU_RESET, 0) !=
928 	    SIGP_CC_ORDER_CODE_ACCEPTED)
929 		return -EIO;
930 
931 	rc = pcpu_alloc_lowcore(pcpu, cpu);
932 	if (rc)
933 		return rc;
934 	pcpu_prepare_secondary(pcpu, cpu);
935 	pcpu_attach_task(pcpu, tidle);
936 	pcpu_start_fn(pcpu, smp_start_secondary, NULL);
937 	/* Wait until cpu puts itself in the online & active maps */
938 	while (!cpu_online(cpu))
939 		cpu_relax();
940 	return 0;
941 }
942 
943 static unsigned int setup_possible_cpus __initdata;
944 
945 static int __init _setup_possible_cpus(char *s)
946 {
947 	get_option(&s, &setup_possible_cpus);
948 	return 0;
949 }
950 early_param("possible_cpus", _setup_possible_cpus);
951 
952 int __cpu_disable(void)
953 {
954 	unsigned long cregs[16];
955 	int cpu;
956 
957 	/* Handle possible pending IPIs */
958 	smp_handle_ext_call();
959 	cpu = smp_processor_id();
960 	set_cpu_online(cpu, false);
961 	cpumask_clear_cpu(cpu, &cpu_setup_mask);
962 	update_cpu_masks();
963 	/* Disable pseudo page faults on this cpu. */
964 	pfault_fini();
965 	/* Disable interrupt sources via control register. */
966 	__ctl_store(cregs, 0, 15);
967 	cregs[0]  &= ~0x0000ee70UL;	/* disable all external interrupts */
968 	cregs[6]  &= ~0xff000000UL;	/* disable all I/O interrupts */
969 	cregs[14] &= ~0x1f000000UL;	/* disable most machine checks */
970 	__ctl_load(cregs, 0, 15);
971 	clear_cpu_flag(CIF_NOHZ_DELAY);
972 	return 0;
973 }
974 
975 void __cpu_die(unsigned int cpu)
976 {
977 	struct pcpu *pcpu;
978 
979 	/* Wait until target cpu is down */
980 	pcpu = pcpu_devices + cpu;
981 	while (!pcpu_stopped(pcpu))
982 		cpu_relax();
983 	pcpu_free_lowcore(pcpu);
984 	cpumask_clear_cpu(cpu, mm_cpumask(&init_mm));
985 	cpumask_clear_cpu(cpu, &init_mm.context.cpu_attach_mask);
986 }
987 
988 void __noreturn cpu_die(void)
989 {
990 	idle_task_exit();
991 	__bpon();
992 	pcpu_sigp_retry(pcpu_devices + smp_processor_id(), SIGP_STOP, 0);
993 	for (;;) ;
994 }
995 
996 void __init smp_fill_possible_mask(void)
997 {
998 	unsigned int possible, sclp_max, cpu;
999 
1000 	sclp_max = max(sclp.mtid, sclp.mtid_cp) + 1;
1001 	sclp_max = min(smp_max_threads, sclp_max);
1002 	sclp_max = (sclp.max_cores * sclp_max) ?: nr_cpu_ids;
1003 	possible = setup_possible_cpus ?: nr_cpu_ids;
1004 	possible = min(possible, sclp_max);
1005 	for (cpu = 0; cpu < possible && cpu < nr_cpu_ids; cpu++)
1006 		set_cpu_possible(cpu, true);
1007 }
1008 
1009 void __init smp_prepare_cpus(unsigned int max_cpus)
1010 {
1011 	/* request the 0x1201 emergency signal external interrupt */
1012 	if (register_external_irq(EXT_IRQ_EMERGENCY_SIG, do_ext_call_interrupt))
1013 		panic("Couldn't request external interrupt 0x1201");
1014 	/* request the 0x1202 external call external interrupt */
1015 	if (register_external_irq(EXT_IRQ_EXTERNAL_CALL, do_ext_call_interrupt))
1016 		panic("Couldn't request external interrupt 0x1202");
1017 }
1018 
1019 void __init smp_prepare_boot_cpu(void)
1020 {
1021 	struct pcpu *pcpu = pcpu_devices;
1022 
1023 	WARN_ON(!cpu_present(0) || !cpu_online(0));
1024 	pcpu->state = CPU_STATE_CONFIGURED;
1025 	S390_lowcore.percpu_offset = __per_cpu_offset[0];
1026 	smp_cpu_set_polarization(0, POLARIZATION_UNKNOWN);
1027 }
1028 
1029 void __init smp_setup_processor_id(void)
1030 {
1031 	pcpu_devices[0].address = stap();
1032 	S390_lowcore.cpu_nr = 0;
1033 	S390_lowcore.spinlock_lockval = arch_spin_lockval(0);
1034 	S390_lowcore.spinlock_index = 0;
1035 }
1036 
1037 /*
1038  * the frequency of the profiling timer can be changed
1039  * by writing a multiplier value into /proc/profile.
1040  *
1041  * usually you want to run this on all CPUs ;)
1042  */
1043 int setup_profiling_timer(unsigned int multiplier)
1044 {
1045 	return 0;
1046 }
1047 
1048 static ssize_t cpu_configure_show(struct device *dev,
1049 				  struct device_attribute *attr, char *buf)
1050 {
1051 	ssize_t count;
1052 
1053 	mutex_lock(&smp_cpu_state_mutex);
1054 	count = sprintf(buf, "%d\n", pcpu_devices[dev->id].state);
1055 	mutex_unlock(&smp_cpu_state_mutex);
1056 	return count;
1057 }
1058 
1059 static ssize_t cpu_configure_store(struct device *dev,
1060 				   struct device_attribute *attr,
1061 				   const char *buf, size_t count)
1062 {
1063 	struct pcpu *pcpu;
1064 	int cpu, val, rc, i;
1065 	char delim;
1066 
1067 	if (sscanf(buf, "%d %c", &val, &delim) != 1)
1068 		return -EINVAL;
1069 	if (val != 0 && val != 1)
1070 		return -EINVAL;
1071 	cpus_read_lock();
1072 	mutex_lock(&smp_cpu_state_mutex);
1073 	rc = -EBUSY;
1074 	/* disallow configuration changes of online cpus and cpu 0 */
1075 	cpu = dev->id;
1076 	cpu = smp_get_base_cpu(cpu);
1077 	if (cpu == 0)
1078 		goto out;
1079 	for (i = 0; i <= smp_cpu_mtid; i++)
1080 		if (cpu_online(cpu + i))
1081 			goto out;
1082 	pcpu = pcpu_devices + cpu;
1083 	rc = 0;
1084 	switch (val) {
1085 	case 0:
1086 		if (pcpu->state != CPU_STATE_CONFIGURED)
1087 			break;
1088 		rc = sclp_core_deconfigure(pcpu->address >> smp_cpu_mt_shift);
1089 		if (rc)
1090 			break;
1091 		for (i = 0; i <= smp_cpu_mtid; i++) {
1092 			if (cpu + i >= nr_cpu_ids || !cpu_present(cpu + i))
1093 				continue;
1094 			pcpu[i].state = CPU_STATE_STANDBY;
1095 			smp_cpu_set_polarization(cpu + i,
1096 						 POLARIZATION_UNKNOWN);
1097 		}
1098 		topology_expect_change();
1099 		break;
1100 	case 1:
1101 		if (pcpu->state != CPU_STATE_STANDBY)
1102 			break;
1103 		rc = sclp_core_configure(pcpu->address >> smp_cpu_mt_shift);
1104 		if (rc)
1105 			break;
1106 		for (i = 0; i <= smp_cpu_mtid; i++) {
1107 			if (cpu + i >= nr_cpu_ids || !cpu_present(cpu + i))
1108 				continue;
1109 			pcpu[i].state = CPU_STATE_CONFIGURED;
1110 			smp_cpu_set_polarization(cpu + i,
1111 						 POLARIZATION_UNKNOWN);
1112 		}
1113 		topology_expect_change();
1114 		break;
1115 	default:
1116 		break;
1117 	}
1118 out:
1119 	mutex_unlock(&smp_cpu_state_mutex);
1120 	cpus_read_unlock();
1121 	return rc ? rc : count;
1122 }
1123 static DEVICE_ATTR(configure, 0644, cpu_configure_show, cpu_configure_store);
1124 
1125 static ssize_t show_cpu_address(struct device *dev,
1126 				struct device_attribute *attr, char *buf)
1127 {
1128 	return sprintf(buf, "%d\n", pcpu_devices[dev->id].address);
1129 }
1130 static DEVICE_ATTR(address, 0444, show_cpu_address, NULL);
1131 
1132 static struct attribute *cpu_common_attrs[] = {
1133 	&dev_attr_configure.attr,
1134 	&dev_attr_address.attr,
1135 	NULL,
1136 };
1137 
1138 static struct attribute_group cpu_common_attr_group = {
1139 	.attrs = cpu_common_attrs,
1140 };
1141 
1142 static struct attribute *cpu_online_attrs[] = {
1143 	&dev_attr_idle_count.attr,
1144 	&dev_attr_idle_time_us.attr,
1145 	NULL,
1146 };
1147 
1148 static struct attribute_group cpu_online_attr_group = {
1149 	.attrs = cpu_online_attrs,
1150 };
1151 
1152 static int smp_cpu_online(unsigned int cpu)
1153 {
1154 	struct device *s = &per_cpu(cpu_device, cpu)->dev;
1155 
1156 	return sysfs_create_group(&s->kobj, &cpu_online_attr_group);
1157 }
1158 
1159 static int smp_cpu_pre_down(unsigned int cpu)
1160 {
1161 	struct device *s = &per_cpu(cpu_device, cpu)->dev;
1162 
1163 	sysfs_remove_group(&s->kobj, &cpu_online_attr_group);
1164 	return 0;
1165 }
1166 
1167 static int smp_add_present_cpu(int cpu)
1168 {
1169 	struct device *s;
1170 	struct cpu *c;
1171 	int rc;
1172 
1173 	c = kzalloc(sizeof(*c), GFP_KERNEL);
1174 	if (!c)
1175 		return -ENOMEM;
1176 	per_cpu(cpu_device, cpu) = c;
1177 	s = &c->dev;
1178 	c->hotpluggable = 1;
1179 	rc = register_cpu(c, cpu);
1180 	if (rc)
1181 		goto out;
1182 	rc = sysfs_create_group(&s->kobj, &cpu_common_attr_group);
1183 	if (rc)
1184 		goto out_cpu;
1185 	rc = topology_cpu_init(c);
1186 	if (rc)
1187 		goto out_topology;
1188 	return 0;
1189 
1190 out_topology:
1191 	sysfs_remove_group(&s->kobj, &cpu_common_attr_group);
1192 out_cpu:
1193 	unregister_cpu(c);
1194 out:
1195 	return rc;
1196 }
1197 
1198 int __ref smp_rescan_cpus(void)
1199 {
1200 	struct sclp_core_info *info;
1201 	int nr;
1202 
1203 	info = kzalloc(sizeof(*info), GFP_KERNEL);
1204 	if (!info)
1205 		return -ENOMEM;
1206 	smp_get_core_info(info, 0);
1207 	nr = __smp_rescan_cpus(info, false);
1208 	kfree(info);
1209 	if (nr)
1210 		topology_schedule_update();
1211 	return 0;
1212 }
1213 
1214 static ssize_t __ref rescan_store(struct device *dev,
1215 				  struct device_attribute *attr,
1216 				  const char *buf,
1217 				  size_t count)
1218 {
1219 	int rc;
1220 
1221 	rc = lock_device_hotplug_sysfs();
1222 	if (rc)
1223 		return rc;
1224 	rc = smp_rescan_cpus();
1225 	unlock_device_hotplug();
1226 	return rc ? rc : count;
1227 }
1228 static DEVICE_ATTR_WO(rescan);
1229 
1230 static int __init s390_smp_init(void)
1231 {
1232 	int cpu, rc = 0;
1233 
1234 	rc = device_create_file(cpu_subsys.dev_root, &dev_attr_rescan);
1235 	if (rc)
1236 		return rc;
1237 	for_each_present_cpu(cpu) {
1238 		rc = smp_add_present_cpu(cpu);
1239 		if (rc)
1240 			goto out;
1241 	}
1242 
1243 	rc = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "s390/smp:online",
1244 			       smp_cpu_online, smp_cpu_pre_down);
1245 	rc = rc <= 0 ? rc : 0;
1246 out:
1247 	return rc;
1248 }
1249 subsys_initcall(s390_smp_init);
1250 
1251 static __always_inline void set_new_lowcore(struct lowcore *lc)
1252 {
1253 	union register_pair dst, src;
1254 	u32 pfx;
1255 
1256 	src.even = (unsigned long) &S390_lowcore;
1257 	src.odd  = sizeof(S390_lowcore);
1258 	dst.even = (unsigned long) lc;
1259 	dst.odd  = sizeof(*lc);
1260 	pfx = (unsigned long) lc;
1261 
1262 	asm volatile(
1263 		"	mvcl	%[dst],%[src]\n"
1264 		"	spx	%[pfx]\n"
1265 		: [dst] "+&d" (dst.pair), [src] "+&d" (src.pair)
1266 		: [pfx] "Q" (pfx)
1267 		: "memory", "cc");
1268 }
1269 
1270 static int __init smp_reinit_ipl_cpu(void)
1271 {
1272 	unsigned long async_stack, nodat_stack, mcck_stack;
1273 	struct lowcore *lc, *lc_ipl;
1274 	unsigned long flags;
1275 
1276 	lc_ipl = lowcore_ptr[0];
1277 	lc = (struct lowcore *)	__get_free_pages(GFP_KERNEL | GFP_DMA, LC_ORDER);
1278 	nodat_stack = __get_free_pages(GFP_KERNEL, THREAD_SIZE_ORDER);
1279 	async_stack = stack_alloc();
1280 	mcck_stack = stack_alloc();
1281 	if (!lc || !nodat_stack || !async_stack || !mcck_stack)
1282 		panic("Couldn't allocate memory");
1283 
1284 	local_irq_save(flags);
1285 	local_mcck_disable();
1286 	set_new_lowcore(lc);
1287 	S390_lowcore.nodat_stack = nodat_stack + STACK_INIT_OFFSET;
1288 	S390_lowcore.async_stack = async_stack + STACK_INIT_OFFSET;
1289 	S390_lowcore.mcck_stack = mcck_stack + STACK_INIT_OFFSET;
1290 	lowcore_ptr[0] = lc;
1291 	local_mcck_enable();
1292 	local_irq_restore(flags);
1293 
1294 	free_pages(lc_ipl->async_stack - STACK_INIT_OFFSET, THREAD_SIZE_ORDER);
1295 	memblock_free_late(lc_ipl->mcck_stack - STACK_INIT_OFFSET, THREAD_SIZE);
1296 	memblock_free_late((unsigned long) lc_ipl, sizeof(*lc_ipl));
1297 
1298 	return 0;
1299 }
1300 early_initcall(smp_reinit_ipl_cpu);
1301