xref: /openbmc/linux/arch/s390/kernel/smp.c (revision 55b37d9c)
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  *
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/memblock.h>
23 #include <linux/export.h>
24 #include <linux/init.h>
25 #include <linux/mm.h>
26 #include <linux/err.h>
27 #include <linux/spinlock.h>
28 #include <linux/kernel_stat.h>
29 #include <linux/delay.h>
30 #include <linux/interrupt.h>
31 #include <linux/irqflags.h>
32 #include <linux/irq_work.h>
33 #include <linux/cpu.h>
34 #include <linux/slab.h>
35 #include <linux/sched/hotplug.h>
36 #include <linux/sched/task_stack.h>
37 #include <linux/crash_dump.h>
38 #include <linux/kprobes.h>
39 #include <asm/asm-offsets.h>
40 #include <asm/diag.h>
41 #include <asm/switch_to.h>
42 #include <asm/facility.h>
43 #include <asm/ipl.h>
44 #include <asm/setup.h>
45 #include <asm/irq.h>
46 #include <asm/tlbflush.h>
47 #include <asm/vtimer.h>
48 #include <asm/abs_lowcore.h>
49 #include <asm/sclp.h>
50 #include <asm/debug.h>
51 #include <asm/os_info.h>
52 #include <asm/sigp.h>
53 #include <asm/idle.h>
54 #include <asm/nmi.h>
55 #include <asm/stacktrace.h>
56 #include <asm/topology.h>
57 #include <asm/vdso.h>
58 #include <asm/maccess.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->return_lpswe = gen_lpswe(__LC_RETURN_PSW);
212 	lc->return_mcck_lpswe = gen_lpswe(__LC_RETURN_MCCK_PSW);
213 	lc->preempt_count = PREEMPT_DISABLED;
214 	if (nmi_alloc_mcesa(&lc->mcesad))
215 		goto out;
216 	if (abs_lowcore_map(cpu, lc, true))
217 		goto out_mcesa;
218 	lowcore_ptr[cpu] = lc;
219 	pcpu_sigp_retry(pcpu, SIGP_SET_PREFIX, __pa(lc));
220 	return 0;
221 
222 out_mcesa:
223 	nmi_free_mcesa(&lc->mcesad);
224 out:
225 	stack_free(mcck_stack);
226 	stack_free(async_stack);
227 	free_pages(nodat_stack, THREAD_SIZE_ORDER);
228 	free_pages((unsigned long) lc, LC_ORDER);
229 	return -ENOMEM;
230 }
231 
232 static void pcpu_free_lowcore(struct pcpu *pcpu)
233 {
234 	unsigned long async_stack, nodat_stack, mcck_stack;
235 	struct lowcore *lc;
236 	int cpu;
237 
238 	cpu = pcpu - pcpu_devices;
239 	lc = lowcore_ptr[cpu];
240 	nodat_stack = lc->nodat_stack - STACK_INIT_OFFSET;
241 	async_stack = lc->async_stack - STACK_INIT_OFFSET;
242 	mcck_stack = lc->mcck_stack - STACK_INIT_OFFSET;
243 	pcpu_sigp_retry(pcpu, SIGP_SET_PREFIX, 0);
244 	lowcore_ptr[cpu] = NULL;
245 	abs_lowcore_unmap(cpu);
246 	nmi_free_mcesa(&lc->mcesad);
247 	stack_free(async_stack);
248 	stack_free(mcck_stack);
249 	free_pages(nodat_stack, THREAD_SIZE_ORDER);
250 	free_pages((unsigned long) lc, LC_ORDER);
251 }
252 
253 static void pcpu_prepare_secondary(struct pcpu *pcpu, int cpu)
254 {
255 	struct lowcore *lc = lowcore_ptr[cpu];
256 
257 	cpumask_set_cpu(cpu, &init_mm.context.cpu_attach_mask);
258 	cpumask_set_cpu(cpu, mm_cpumask(&init_mm));
259 	lc->cpu_nr = cpu;
260 	lc->restart_flags = RESTART_FLAG_CTLREGS;
261 	lc->spinlock_lockval = arch_spin_lockval(cpu);
262 	lc->spinlock_index = 0;
263 	lc->percpu_offset = __per_cpu_offset[cpu];
264 	lc->kernel_asce = S390_lowcore.kernel_asce;
265 	lc->user_asce = s390_invalid_asce;
266 	lc->machine_flags = S390_lowcore.machine_flags;
267 	lc->user_timer = lc->system_timer =
268 		lc->steal_timer = lc->avg_steal_timer = 0;
269 	__ctl_store(lc->cregs_save_area, 0, 15);
270 	lc->cregs_save_area[1] = lc->kernel_asce;
271 	lc->cregs_save_area[7] = lc->user_asce;
272 	save_access_regs((unsigned int *) lc->access_regs_save_area);
273 	arch_spin_lock_setup(cpu);
274 }
275 
276 static void pcpu_attach_task(struct pcpu *pcpu, struct task_struct *tsk)
277 {
278 	struct lowcore *lc;
279 	int cpu;
280 
281 	cpu = pcpu - pcpu_devices;
282 	lc = lowcore_ptr[cpu];
283 	lc->kernel_stack = (unsigned long)task_stack_page(tsk) + STACK_INIT_OFFSET;
284 	lc->current_task = (unsigned long)tsk;
285 	lc->lpp = LPP_MAGIC;
286 	lc->current_pid = tsk->pid;
287 	lc->user_timer = tsk->thread.user_timer;
288 	lc->guest_timer = tsk->thread.guest_timer;
289 	lc->system_timer = tsk->thread.system_timer;
290 	lc->hardirq_timer = tsk->thread.hardirq_timer;
291 	lc->softirq_timer = tsk->thread.softirq_timer;
292 	lc->steal_timer = 0;
293 }
294 
295 static void pcpu_start_fn(struct pcpu *pcpu, void (*func)(void *), void *data)
296 {
297 	struct lowcore *lc;
298 	int cpu;
299 
300 	cpu = pcpu - pcpu_devices;
301 	lc = lowcore_ptr[cpu];
302 	lc->restart_stack = lc->kernel_stack;
303 	lc->restart_fn = (unsigned long) func;
304 	lc->restart_data = (unsigned long) data;
305 	lc->restart_source = -1U;
306 	pcpu_sigp_retry(pcpu, SIGP_RESTART, 0);
307 }
308 
309 typedef void (pcpu_delegate_fn)(void *);
310 
311 /*
312  * Call function via PSW restart on pcpu and stop the current cpu.
313  */
314 static void __pcpu_delegate(pcpu_delegate_fn *func, void *data)
315 {
316 	func(data);	/* should not return */
317 }
318 
319 static void pcpu_delegate(struct pcpu *pcpu,
320 			  pcpu_delegate_fn *func,
321 			  void *data, unsigned long stack)
322 {
323 	struct lowcore *lc, *abs_lc;
324 	unsigned int source_cpu;
325 
326 	lc = lowcore_ptr[pcpu - pcpu_devices];
327 	source_cpu = stap();
328 
329 	if (pcpu->address == source_cpu) {
330 		call_on_stack(2, stack, void, __pcpu_delegate,
331 			      pcpu_delegate_fn *, func, void *, data);
332 	}
333 	/* Stop target cpu (if func returns this stops the current cpu). */
334 	pcpu_sigp_retry(pcpu, SIGP_STOP, 0);
335 	pcpu_sigp_retry(pcpu, SIGP_CPU_RESET, 0);
336 	/* Restart func on the target cpu and stop the current cpu. */
337 	if (lc) {
338 		lc->restart_stack = stack;
339 		lc->restart_fn = (unsigned long)func;
340 		lc->restart_data = (unsigned long)data;
341 		lc->restart_source = source_cpu;
342 	} else {
343 		abs_lc = get_abs_lowcore();
344 		abs_lc->restart_stack = stack;
345 		abs_lc->restart_fn = (unsigned long)func;
346 		abs_lc->restart_data = (unsigned long)data;
347 		abs_lc->restart_source = source_cpu;
348 		put_abs_lowcore(abs_lc);
349 	}
350 	asm volatile(
351 		"0:	sigp	0,%0,%2	# sigp restart to target cpu\n"
352 		"	brc	2,0b	# busy, try again\n"
353 		"1:	sigp	0,%1,%3	# sigp stop to current cpu\n"
354 		"	brc	2,1b	# busy, try again\n"
355 		: : "d" (pcpu->address), "d" (source_cpu),
356 		    "K" (SIGP_RESTART), "K" (SIGP_STOP)
357 		: "0", "1", "cc");
358 	for (;;) ;
359 }
360 
361 /*
362  * Enable additional logical cpus for multi-threading.
363  */
364 static int pcpu_set_smt(unsigned int mtid)
365 {
366 	int cc;
367 
368 	if (smp_cpu_mtid == mtid)
369 		return 0;
370 	cc = __pcpu_sigp(0, SIGP_SET_MULTI_THREADING, mtid, NULL);
371 	if (cc == 0) {
372 		smp_cpu_mtid = mtid;
373 		smp_cpu_mt_shift = 0;
374 		while (smp_cpu_mtid >= (1U << smp_cpu_mt_shift))
375 			smp_cpu_mt_shift++;
376 		pcpu_devices[0].address = stap();
377 	}
378 	return cc;
379 }
380 
381 /*
382  * Call function on an online CPU.
383  */
384 void smp_call_online_cpu(void (*func)(void *), void *data)
385 {
386 	struct pcpu *pcpu;
387 
388 	/* Use the current cpu if it is online. */
389 	pcpu = pcpu_find_address(cpu_online_mask, stap());
390 	if (!pcpu)
391 		/* Use the first online cpu. */
392 		pcpu = pcpu_devices + cpumask_first(cpu_online_mask);
393 	pcpu_delegate(pcpu, func, data, (unsigned long) restart_stack);
394 }
395 
396 /*
397  * Call function on the ipl CPU.
398  */
399 void smp_call_ipl_cpu(void (*func)(void *), void *data)
400 {
401 	struct lowcore *lc = lowcore_ptr[0];
402 
403 	if (pcpu_devices[0].address == stap())
404 		lc = &S390_lowcore;
405 
406 	pcpu_delegate(&pcpu_devices[0], func, data,
407 		      lc->nodat_stack);
408 }
409 
410 int smp_find_processor_id(u16 address)
411 {
412 	int cpu;
413 
414 	for_each_present_cpu(cpu)
415 		if (pcpu_devices[cpu].address == address)
416 			return cpu;
417 	return -1;
418 }
419 
420 void schedule_mcck_handler(void)
421 {
422 	pcpu_ec_call(pcpu_devices + smp_processor_id(), ec_mcck_pending);
423 }
424 
425 bool notrace arch_vcpu_is_preempted(int cpu)
426 {
427 	if (test_cpu_flag_of(CIF_ENABLED_WAIT, cpu))
428 		return false;
429 	if (pcpu_running(pcpu_devices + cpu))
430 		return false;
431 	return true;
432 }
433 EXPORT_SYMBOL(arch_vcpu_is_preempted);
434 
435 void notrace smp_yield_cpu(int cpu)
436 {
437 	if (!MACHINE_HAS_DIAG9C)
438 		return;
439 	diag_stat_inc_norecursion(DIAG_STAT_X09C);
440 	asm volatile("diag %0,0,0x9c"
441 		     : : "d" (pcpu_devices[cpu].address));
442 }
443 EXPORT_SYMBOL_GPL(smp_yield_cpu);
444 
445 /*
446  * Send cpus emergency shutdown signal. This gives the cpus the
447  * opportunity to complete outstanding interrupts.
448  */
449 void notrace smp_emergency_stop(void)
450 {
451 	static arch_spinlock_t lock = __ARCH_SPIN_LOCK_UNLOCKED;
452 	static cpumask_t cpumask;
453 	u64 end;
454 	int cpu;
455 
456 	arch_spin_lock(&lock);
457 	cpumask_copy(&cpumask, cpu_online_mask);
458 	cpumask_clear_cpu(smp_processor_id(), &cpumask);
459 
460 	end = get_tod_clock() + (1000000UL << 12);
461 	for_each_cpu(cpu, &cpumask) {
462 		struct pcpu *pcpu = pcpu_devices + cpu;
463 		set_bit(ec_stop_cpu, &pcpu->ec_mask);
464 		while (__pcpu_sigp(pcpu->address, SIGP_EMERGENCY_SIGNAL,
465 				   0, NULL) == SIGP_CC_BUSY &&
466 		       get_tod_clock() < end)
467 			cpu_relax();
468 	}
469 	while (get_tod_clock() < end) {
470 		for_each_cpu(cpu, &cpumask)
471 			if (pcpu_stopped(pcpu_devices + cpu))
472 				cpumask_clear_cpu(cpu, &cpumask);
473 		if (cpumask_empty(&cpumask))
474 			break;
475 		cpu_relax();
476 	}
477 	arch_spin_unlock(&lock);
478 }
479 NOKPROBE_SYMBOL(smp_emergency_stop);
480 
481 /*
482  * Stop all cpus but the current one.
483  */
484 void smp_send_stop(void)
485 {
486 	int cpu;
487 
488 	/* Disable all interrupts/machine checks */
489 	__load_psw_mask(PSW_KERNEL_BITS);
490 	trace_hardirqs_off();
491 
492 	debug_set_critical();
493 
494 	if (oops_in_progress)
495 		smp_emergency_stop();
496 
497 	/* stop all processors */
498 	for_each_online_cpu(cpu) {
499 		if (cpu == smp_processor_id())
500 			continue;
501 		pcpu_sigp_retry(pcpu_devices + cpu, SIGP_STOP, 0);
502 		while (!pcpu_stopped(pcpu_devices + cpu))
503 			cpu_relax();
504 	}
505 }
506 
507 /*
508  * This is the main routine where commands issued by other
509  * cpus are handled.
510  */
511 static void smp_handle_ext_call(void)
512 {
513 	unsigned long bits;
514 
515 	/* handle bit signal external calls */
516 	bits = xchg(&pcpu_devices[smp_processor_id()].ec_mask, 0);
517 	if (test_bit(ec_stop_cpu, &bits))
518 		smp_stop_cpu();
519 	if (test_bit(ec_schedule, &bits))
520 		scheduler_ipi();
521 	if (test_bit(ec_call_function_single, &bits))
522 		generic_smp_call_function_single_interrupt();
523 	if (test_bit(ec_mcck_pending, &bits))
524 		s390_handle_mcck();
525 	if (test_bit(ec_irq_work, &bits))
526 		irq_work_run();
527 }
528 
529 static void do_ext_call_interrupt(struct ext_code ext_code,
530 				  unsigned int param32, unsigned long param64)
531 {
532 	inc_irq_stat(ext_code.code == 0x1202 ? IRQEXT_EXC : IRQEXT_EMS);
533 	smp_handle_ext_call();
534 }
535 
536 void arch_send_call_function_ipi_mask(const struct cpumask *mask)
537 {
538 	int cpu;
539 
540 	for_each_cpu(cpu, mask)
541 		pcpu_ec_call(pcpu_devices + cpu, ec_call_function_single);
542 }
543 
544 void arch_send_call_function_single_ipi(int cpu)
545 {
546 	pcpu_ec_call(pcpu_devices + cpu, ec_call_function_single);
547 }
548 
549 /*
550  * this function sends a 'reschedule' IPI to another CPU.
551  * it goes straight through and wastes no time serializing
552  * anything. Worst case is that we lose a reschedule ...
553  */
554 void arch_smp_send_reschedule(int cpu)
555 {
556 	pcpu_ec_call(pcpu_devices + cpu, ec_schedule);
557 }
558 
559 #ifdef CONFIG_IRQ_WORK
560 void arch_irq_work_raise(void)
561 {
562 	pcpu_ec_call(pcpu_devices + smp_processor_id(), ec_irq_work);
563 }
564 #endif
565 
566 /*
567  * parameter area for the set/clear control bit callbacks
568  */
569 struct ec_creg_mask_parms {
570 	unsigned long orval;
571 	unsigned long andval;
572 	int cr;
573 };
574 
575 /*
576  * callback for setting/clearing control bits
577  */
578 static void smp_ctl_bit_callback(void *info)
579 {
580 	struct ec_creg_mask_parms *pp = info;
581 	unsigned long cregs[16];
582 
583 	__ctl_store(cregs, 0, 15);
584 	cregs[pp->cr] = (cregs[pp->cr] & pp->andval) | pp->orval;
585 	__ctl_load(cregs, 0, 15);
586 }
587 
588 static DEFINE_SPINLOCK(ctl_lock);
589 
590 void smp_ctl_set_clear_bit(int cr, int bit, bool set)
591 {
592 	struct ec_creg_mask_parms parms = { .cr = cr, };
593 	struct lowcore *abs_lc;
594 	u64 ctlreg;
595 
596 	if (set) {
597 		parms.orval = 1UL << bit;
598 		parms.andval = -1UL;
599 	} else {
600 		parms.orval = 0;
601 		parms.andval = ~(1UL << bit);
602 	}
603 	spin_lock(&ctl_lock);
604 	abs_lc = get_abs_lowcore();
605 	ctlreg = abs_lc->cregs_save_area[cr];
606 	ctlreg = (ctlreg & parms.andval) | parms.orval;
607 	abs_lc->cregs_save_area[cr] = ctlreg;
608 	put_abs_lowcore(abs_lc);
609 	spin_unlock(&ctl_lock);
610 	on_each_cpu(smp_ctl_bit_callback, &parms, 1);
611 }
612 EXPORT_SYMBOL(smp_ctl_set_clear_bit);
613 
614 #ifdef CONFIG_CRASH_DUMP
615 
616 int smp_store_status(int cpu)
617 {
618 	struct lowcore *lc;
619 	struct pcpu *pcpu;
620 	unsigned long pa;
621 
622 	pcpu = pcpu_devices + cpu;
623 	lc = lowcore_ptr[cpu];
624 	pa = __pa(&lc->floating_pt_save_area);
625 	if (__pcpu_sigp_relax(pcpu->address, SIGP_STORE_STATUS_AT_ADDRESS,
626 			      pa) != SIGP_CC_ORDER_CODE_ACCEPTED)
627 		return -EIO;
628 	if (!MACHINE_HAS_VX && !MACHINE_HAS_GS)
629 		return 0;
630 	pa = lc->mcesad & MCESA_ORIGIN_MASK;
631 	if (MACHINE_HAS_GS)
632 		pa |= lc->mcesad & MCESA_LC_MASK;
633 	if (__pcpu_sigp_relax(pcpu->address, SIGP_STORE_ADDITIONAL_STATUS,
634 			      pa) != SIGP_CC_ORDER_CODE_ACCEPTED)
635 		return -EIO;
636 	return 0;
637 }
638 
639 /*
640  * Collect CPU state of the previous, crashed system.
641  * There are four cases:
642  * 1) standard zfcp/nvme dump
643  *    condition: OLDMEM_BASE == NULL && is_ipl_type_dump() == true
644  *    The state for all CPUs except the boot CPU needs to be collected
645  *    with sigp stop-and-store-status. The boot CPU state is located in
646  *    the absolute lowcore of the memory stored in the HSA. The zcore code
647  *    will copy the boot CPU state from the HSA.
648  * 2) stand-alone kdump for SCSI/NVMe (zfcp/nvme dump with swapped memory)
649  *    condition: OLDMEM_BASE != NULL && is_ipl_type_dump() == true
650  *    The state for all CPUs except the boot CPU needs to be collected
651  *    with sigp stop-and-store-status. The firmware or the boot-loader
652  *    stored the registers of the boot CPU in the absolute lowcore in the
653  *    memory of the old system.
654  * 3) kdump and the old kernel did not store the CPU state,
655  *    or stand-alone kdump for DASD
656  *    condition: OLDMEM_BASE != NULL && !is_kdump_kernel()
657  *    The state for all CPUs except the boot CPU needs to be collected
658  *    with sigp stop-and-store-status. The kexec code or the boot-loader
659  *    stored the registers of the boot CPU in the memory of the old system.
660  * 4) kdump and the old kernel stored the CPU state
661  *    condition: OLDMEM_BASE != NULL && is_kdump_kernel()
662  *    This case does not exist for s390 anymore, setup_arch explicitly
663  *    deactivates the elfcorehdr= kernel parameter
664  */
665 static bool dump_available(void)
666 {
667 	return oldmem_data.start || is_ipl_type_dump();
668 }
669 
670 void __init smp_save_dump_ipl_cpu(void)
671 {
672 	struct save_area *sa;
673 	void *regs;
674 
675 	if (!dump_available())
676 		return;
677 	sa = save_area_alloc(true);
678 	regs = memblock_alloc(512, 8);
679 	if (!sa || !regs)
680 		panic("could not allocate memory for boot CPU save area\n");
681 	copy_oldmem_kernel(regs, __LC_FPREGS_SAVE_AREA, 512);
682 	save_area_add_regs(sa, regs);
683 	memblock_free(regs, 512);
684 	if (MACHINE_HAS_VX)
685 		save_area_add_vxrs(sa, boot_cpu_vector_save_area);
686 }
687 
688 void __init smp_save_dump_secondary_cpus(void)
689 {
690 	int addr, boot_cpu_addr, max_cpu_addr;
691 	struct save_area *sa;
692 	void *page;
693 
694 	if (!dump_available())
695 		return;
696 	/* Allocate a page as dumping area for the store status sigps */
697 	page = memblock_alloc_low(PAGE_SIZE, PAGE_SIZE);
698 	if (!page)
699 		panic("ERROR: Failed to allocate %lx bytes below %lx\n",
700 		      PAGE_SIZE, 1UL << 31);
701 
702 	/* Set multi-threading state to the previous system. */
703 	pcpu_set_smt(sclp.mtid_prev);
704 	boot_cpu_addr = stap();
705 	max_cpu_addr = SCLP_MAX_CORES << sclp.mtid_prev;
706 	for (addr = 0; addr <= max_cpu_addr; addr++) {
707 		if (addr == boot_cpu_addr)
708 			continue;
709 		if (__pcpu_sigp_relax(addr, SIGP_SENSE, 0) ==
710 		    SIGP_CC_NOT_OPERATIONAL)
711 			continue;
712 		sa = save_area_alloc(false);
713 		if (!sa)
714 			panic("could not allocate memory for save area\n");
715 		__pcpu_sigp_relax(addr, SIGP_STORE_STATUS_AT_ADDRESS, __pa(page));
716 		save_area_add_regs(sa, page);
717 		if (MACHINE_HAS_VX) {
718 			__pcpu_sigp_relax(addr, SIGP_STORE_ADDITIONAL_STATUS, __pa(page));
719 			save_area_add_vxrs(sa, page);
720 		}
721 	}
722 	memblock_free(page, PAGE_SIZE);
723 	diag_amode31_ops.diag308_reset();
724 	pcpu_set_smt(0);
725 }
726 #endif /* CONFIG_CRASH_DUMP */
727 
728 void smp_cpu_set_polarization(int cpu, int val)
729 {
730 	pcpu_devices[cpu].polarization = val;
731 }
732 
733 int smp_cpu_get_polarization(int cpu)
734 {
735 	return pcpu_devices[cpu].polarization;
736 }
737 
738 int smp_cpu_get_cpu_address(int cpu)
739 {
740 	return pcpu_devices[cpu].address;
741 }
742 
743 static void __ref smp_get_core_info(struct sclp_core_info *info, int early)
744 {
745 	static int use_sigp_detection;
746 	int address;
747 
748 	if (use_sigp_detection || sclp_get_core_info(info, early)) {
749 		use_sigp_detection = 1;
750 		for (address = 0;
751 		     address < (SCLP_MAX_CORES << smp_cpu_mt_shift);
752 		     address += (1U << smp_cpu_mt_shift)) {
753 			if (__pcpu_sigp_relax(address, SIGP_SENSE, 0) ==
754 			    SIGP_CC_NOT_OPERATIONAL)
755 				continue;
756 			info->core[info->configured].core_id =
757 				address >> smp_cpu_mt_shift;
758 			info->configured++;
759 		}
760 		info->combined = info->configured;
761 	}
762 }
763 
764 static int smp_add_present_cpu(int cpu);
765 
766 static int smp_add_core(struct sclp_core_entry *core, cpumask_t *avail,
767 			bool configured, bool early)
768 {
769 	struct pcpu *pcpu;
770 	int cpu, nr, i;
771 	u16 address;
772 
773 	nr = 0;
774 	if (sclp.has_core_type && core->type != boot_core_type)
775 		return nr;
776 	cpu = cpumask_first(avail);
777 	address = core->core_id << smp_cpu_mt_shift;
778 	for (i = 0; (i <= smp_cpu_mtid) && (cpu < nr_cpu_ids); i++) {
779 		if (pcpu_find_address(cpu_present_mask, address + i))
780 			continue;
781 		pcpu = pcpu_devices + cpu;
782 		pcpu->address = address + i;
783 		if (configured)
784 			pcpu->state = CPU_STATE_CONFIGURED;
785 		else
786 			pcpu->state = CPU_STATE_STANDBY;
787 		smp_cpu_set_polarization(cpu, POLARIZATION_UNKNOWN);
788 		set_cpu_present(cpu, true);
789 		if (!early && smp_add_present_cpu(cpu) != 0)
790 			set_cpu_present(cpu, false);
791 		else
792 			nr++;
793 		cpumask_clear_cpu(cpu, avail);
794 		cpu = cpumask_next(cpu, avail);
795 	}
796 	return nr;
797 }
798 
799 static int __smp_rescan_cpus(struct sclp_core_info *info, bool early)
800 {
801 	struct sclp_core_entry *core;
802 	static cpumask_t avail;
803 	bool configured;
804 	u16 core_id;
805 	int nr, i;
806 
807 	cpus_read_lock();
808 	mutex_lock(&smp_cpu_state_mutex);
809 	nr = 0;
810 	cpumask_xor(&avail, cpu_possible_mask, cpu_present_mask);
811 	/*
812 	 * Add IPL core first (which got logical CPU number 0) to make sure
813 	 * that all SMT threads get subsequent logical CPU numbers.
814 	 */
815 	if (early) {
816 		core_id = pcpu_devices[0].address >> smp_cpu_mt_shift;
817 		for (i = 0; i < info->configured; i++) {
818 			core = &info->core[i];
819 			if (core->core_id == core_id) {
820 				nr += smp_add_core(core, &avail, true, early);
821 				break;
822 			}
823 		}
824 	}
825 	for (i = 0; i < info->combined; i++) {
826 		configured = i < info->configured;
827 		nr += smp_add_core(&info->core[i], &avail, configured, early);
828 	}
829 	mutex_unlock(&smp_cpu_state_mutex);
830 	cpus_read_unlock();
831 	return nr;
832 }
833 
834 void __init smp_detect_cpus(void)
835 {
836 	unsigned int cpu, mtid, c_cpus, s_cpus;
837 	struct sclp_core_info *info;
838 	u16 address;
839 
840 	/* Get CPU information */
841 	info = memblock_alloc(sizeof(*info), 8);
842 	if (!info)
843 		panic("%s: Failed to allocate %zu bytes align=0x%x\n",
844 		      __func__, sizeof(*info), 8);
845 	smp_get_core_info(info, 1);
846 	/* Find boot CPU type */
847 	if (sclp.has_core_type) {
848 		address = stap();
849 		for (cpu = 0; cpu < info->combined; cpu++)
850 			if (info->core[cpu].core_id == address) {
851 				/* The boot cpu dictates the cpu type. */
852 				boot_core_type = info->core[cpu].type;
853 				break;
854 			}
855 		if (cpu >= info->combined)
856 			panic("Could not find boot CPU type");
857 	}
858 
859 	/* Set multi-threading state for the current system */
860 	mtid = boot_core_type ? sclp.mtid : sclp.mtid_cp;
861 	mtid = (mtid < smp_max_threads) ? mtid : smp_max_threads - 1;
862 	pcpu_set_smt(mtid);
863 
864 	/* Print number of CPUs */
865 	c_cpus = s_cpus = 0;
866 	for (cpu = 0; cpu < info->combined; cpu++) {
867 		if (sclp.has_core_type &&
868 		    info->core[cpu].type != boot_core_type)
869 			continue;
870 		if (cpu < info->configured)
871 			c_cpus += smp_cpu_mtid + 1;
872 		else
873 			s_cpus += smp_cpu_mtid + 1;
874 	}
875 	pr_info("%d configured CPUs, %d standby CPUs\n", c_cpus, s_cpus);
876 
877 	/* Add CPUs present at boot */
878 	__smp_rescan_cpus(info, true);
879 	memblock_free(info, sizeof(*info));
880 }
881 
882 /*
883  *	Activate a secondary processor.
884  */
885 static void smp_start_secondary(void *cpuvoid)
886 {
887 	int cpu = raw_smp_processor_id();
888 
889 	S390_lowcore.last_update_clock = get_tod_clock();
890 	S390_lowcore.restart_stack = (unsigned long)restart_stack;
891 	S390_lowcore.restart_fn = (unsigned long)do_restart;
892 	S390_lowcore.restart_data = 0;
893 	S390_lowcore.restart_source = -1U;
894 	S390_lowcore.restart_flags = 0;
895 	restore_access_regs(S390_lowcore.access_regs_save_area);
896 	cpu_init();
897 	rcu_cpu_starting(cpu);
898 	init_cpu_timer();
899 	vtime_init();
900 	vdso_getcpu_init();
901 	pfault_init();
902 	cpumask_set_cpu(cpu, &cpu_setup_mask);
903 	update_cpu_masks();
904 	notify_cpu_starting(cpu);
905 	if (topology_cpu_dedicated(cpu))
906 		set_cpu_flag(CIF_DEDICATED_CPU);
907 	else
908 		clear_cpu_flag(CIF_DEDICATED_CPU);
909 	set_cpu_online(cpu, true);
910 	inc_irq_stat(CPU_RST);
911 	local_irq_enable();
912 	cpu_startup_entry(CPUHP_AP_ONLINE_IDLE);
913 }
914 
915 /* Upping and downing of CPUs */
916 int __cpu_up(unsigned int cpu, struct task_struct *tidle)
917 {
918 	struct pcpu *pcpu = pcpu_devices + cpu;
919 	int rc;
920 
921 	if (pcpu->state != CPU_STATE_CONFIGURED)
922 		return -EIO;
923 	if (pcpu_sigp_retry(pcpu, SIGP_INITIAL_CPU_RESET, 0) !=
924 	    SIGP_CC_ORDER_CODE_ACCEPTED)
925 		return -EIO;
926 
927 	rc = pcpu_alloc_lowcore(pcpu, cpu);
928 	if (rc)
929 		return rc;
930 	pcpu_prepare_secondary(pcpu, cpu);
931 	pcpu_attach_task(pcpu, tidle);
932 	pcpu_start_fn(pcpu, smp_start_secondary, NULL);
933 	/* Wait until cpu puts itself in the online & active maps */
934 	while (!cpu_online(cpu))
935 		cpu_relax();
936 	return 0;
937 }
938 
939 static unsigned int setup_possible_cpus __initdata;
940 
941 static int __init _setup_possible_cpus(char *s)
942 {
943 	get_option(&s, &setup_possible_cpus);
944 	return 0;
945 }
946 early_param("possible_cpus", _setup_possible_cpus);
947 
948 int __cpu_disable(void)
949 {
950 	unsigned long cregs[16];
951 	int cpu;
952 
953 	/* Handle possible pending IPIs */
954 	smp_handle_ext_call();
955 	cpu = smp_processor_id();
956 	set_cpu_online(cpu, false);
957 	cpumask_clear_cpu(cpu, &cpu_setup_mask);
958 	update_cpu_masks();
959 	/* Disable pseudo page faults on this cpu. */
960 	pfault_fini();
961 	/* Disable interrupt sources via control register. */
962 	__ctl_store(cregs, 0, 15);
963 	cregs[0]  &= ~0x0000ee70UL;	/* disable all external interrupts */
964 	cregs[6]  &= ~0xff000000UL;	/* disable all I/O interrupts */
965 	cregs[14] &= ~0x1f000000UL;	/* disable most machine checks */
966 	__ctl_load(cregs, 0, 15);
967 	clear_cpu_flag(CIF_NOHZ_DELAY);
968 	return 0;
969 }
970 
971 void __cpu_die(unsigned int cpu)
972 {
973 	struct pcpu *pcpu;
974 
975 	/* Wait until target cpu is down */
976 	pcpu = pcpu_devices + cpu;
977 	while (!pcpu_stopped(pcpu))
978 		cpu_relax();
979 	pcpu_free_lowcore(pcpu);
980 	cpumask_clear_cpu(cpu, mm_cpumask(&init_mm));
981 	cpumask_clear_cpu(cpu, &init_mm.context.cpu_attach_mask);
982 }
983 
984 void __noreturn cpu_die(void)
985 {
986 	idle_task_exit();
987 	pcpu_sigp_retry(pcpu_devices + smp_processor_id(), SIGP_STOP, 0);
988 	for (;;) ;
989 }
990 
991 void __init smp_fill_possible_mask(void)
992 {
993 	unsigned int possible, sclp_max, cpu;
994 
995 	sclp_max = max(sclp.mtid, sclp.mtid_cp) + 1;
996 	sclp_max = min(smp_max_threads, sclp_max);
997 	sclp_max = (sclp.max_cores * sclp_max) ?: nr_cpu_ids;
998 	possible = setup_possible_cpus ?: nr_cpu_ids;
999 	possible = min(possible, sclp_max);
1000 	for (cpu = 0; cpu < possible && cpu < nr_cpu_ids; cpu++)
1001 		set_cpu_possible(cpu, true);
1002 }
1003 
1004 void __init smp_prepare_cpus(unsigned int max_cpus)
1005 {
1006 	/* request the 0x1201 emergency signal external interrupt */
1007 	if (register_external_irq(EXT_IRQ_EMERGENCY_SIG, do_ext_call_interrupt))
1008 		panic("Couldn't request external interrupt 0x1201");
1009 	/* request the 0x1202 external call external interrupt */
1010 	if (register_external_irq(EXT_IRQ_EXTERNAL_CALL, do_ext_call_interrupt))
1011 		panic("Couldn't request external interrupt 0x1202");
1012 }
1013 
1014 void __init smp_prepare_boot_cpu(void)
1015 {
1016 	struct pcpu *pcpu = pcpu_devices;
1017 
1018 	WARN_ON(!cpu_present(0) || !cpu_online(0));
1019 	pcpu->state = CPU_STATE_CONFIGURED;
1020 	S390_lowcore.percpu_offset = __per_cpu_offset[0];
1021 	smp_cpu_set_polarization(0, POLARIZATION_UNKNOWN);
1022 }
1023 
1024 void __init smp_setup_processor_id(void)
1025 {
1026 	pcpu_devices[0].address = stap();
1027 	S390_lowcore.cpu_nr = 0;
1028 	S390_lowcore.spinlock_lockval = arch_spin_lockval(0);
1029 	S390_lowcore.spinlock_index = 0;
1030 }
1031 
1032 /*
1033  * the frequency of the profiling timer can be changed
1034  * by writing a multiplier value into /proc/profile.
1035  *
1036  * usually you want to run this on all CPUs ;)
1037  */
1038 int setup_profiling_timer(unsigned int multiplier)
1039 {
1040 	return 0;
1041 }
1042 
1043 static ssize_t cpu_configure_show(struct device *dev,
1044 				  struct device_attribute *attr, char *buf)
1045 {
1046 	ssize_t count;
1047 
1048 	mutex_lock(&smp_cpu_state_mutex);
1049 	count = sprintf(buf, "%d\n", pcpu_devices[dev->id].state);
1050 	mutex_unlock(&smp_cpu_state_mutex);
1051 	return count;
1052 }
1053 
1054 static ssize_t cpu_configure_store(struct device *dev,
1055 				   struct device_attribute *attr,
1056 				   const char *buf, size_t count)
1057 {
1058 	struct pcpu *pcpu;
1059 	int cpu, val, rc, i;
1060 	char delim;
1061 
1062 	if (sscanf(buf, "%d %c", &val, &delim) != 1)
1063 		return -EINVAL;
1064 	if (val != 0 && val != 1)
1065 		return -EINVAL;
1066 	cpus_read_lock();
1067 	mutex_lock(&smp_cpu_state_mutex);
1068 	rc = -EBUSY;
1069 	/* disallow configuration changes of online cpus and cpu 0 */
1070 	cpu = dev->id;
1071 	cpu = smp_get_base_cpu(cpu);
1072 	if (cpu == 0)
1073 		goto out;
1074 	for (i = 0; i <= smp_cpu_mtid; i++)
1075 		if (cpu_online(cpu + i))
1076 			goto out;
1077 	pcpu = pcpu_devices + cpu;
1078 	rc = 0;
1079 	switch (val) {
1080 	case 0:
1081 		if (pcpu->state != CPU_STATE_CONFIGURED)
1082 			break;
1083 		rc = sclp_core_deconfigure(pcpu->address >> smp_cpu_mt_shift);
1084 		if (rc)
1085 			break;
1086 		for (i = 0; i <= smp_cpu_mtid; i++) {
1087 			if (cpu + i >= nr_cpu_ids || !cpu_present(cpu + i))
1088 				continue;
1089 			pcpu[i].state = CPU_STATE_STANDBY;
1090 			smp_cpu_set_polarization(cpu + i,
1091 						 POLARIZATION_UNKNOWN);
1092 		}
1093 		topology_expect_change();
1094 		break;
1095 	case 1:
1096 		if (pcpu->state != CPU_STATE_STANDBY)
1097 			break;
1098 		rc = sclp_core_configure(pcpu->address >> smp_cpu_mt_shift);
1099 		if (rc)
1100 			break;
1101 		for (i = 0; i <= smp_cpu_mtid; i++) {
1102 			if (cpu + i >= nr_cpu_ids || !cpu_present(cpu + i))
1103 				continue;
1104 			pcpu[i].state = CPU_STATE_CONFIGURED;
1105 			smp_cpu_set_polarization(cpu + i,
1106 						 POLARIZATION_UNKNOWN);
1107 		}
1108 		topology_expect_change();
1109 		break;
1110 	default:
1111 		break;
1112 	}
1113 out:
1114 	mutex_unlock(&smp_cpu_state_mutex);
1115 	cpus_read_unlock();
1116 	return rc ? rc : count;
1117 }
1118 static DEVICE_ATTR(configure, 0644, cpu_configure_show, cpu_configure_store);
1119 
1120 static ssize_t show_cpu_address(struct device *dev,
1121 				struct device_attribute *attr, char *buf)
1122 {
1123 	return sprintf(buf, "%d\n", pcpu_devices[dev->id].address);
1124 }
1125 static DEVICE_ATTR(address, 0444, show_cpu_address, NULL);
1126 
1127 static struct attribute *cpu_common_attrs[] = {
1128 	&dev_attr_configure.attr,
1129 	&dev_attr_address.attr,
1130 	NULL,
1131 };
1132 
1133 static struct attribute_group cpu_common_attr_group = {
1134 	.attrs = cpu_common_attrs,
1135 };
1136 
1137 static struct attribute *cpu_online_attrs[] = {
1138 	&dev_attr_idle_count.attr,
1139 	&dev_attr_idle_time_us.attr,
1140 	NULL,
1141 };
1142 
1143 static struct attribute_group cpu_online_attr_group = {
1144 	.attrs = cpu_online_attrs,
1145 };
1146 
1147 static int smp_cpu_online(unsigned int cpu)
1148 {
1149 	struct device *s = &per_cpu(cpu_device, cpu)->dev;
1150 
1151 	return sysfs_create_group(&s->kobj, &cpu_online_attr_group);
1152 }
1153 
1154 static int smp_cpu_pre_down(unsigned int cpu)
1155 {
1156 	struct device *s = &per_cpu(cpu_device, cpu)->dev;
1157 
1158 	sysfs_remove_group(&s->kobj, &cpu_online_attr_group);
1159 	return 0;
1160 }
1161 
1162 static int smp_add_present_cpu(int cpu)
1163 {
1164 	struct device *s;
1165 	struct cpu *c;
1166 	int rc;
1167 
1168 	c = kzalloc(sizeof(*c), GFP_KERNEL);
1169 	if (!c)
1170 		return -ENOMEM;
1171 	per_cpu(cpu_device, cpu) = c;
1172 	s = &c->dev;
1173 	c->hotpluggable = 1;
1174 	rc = register_cpu(c, cpu);
1175 	if (rc)
1176 		goto out;
1177 	rc = sysfs_create_group(&s->kobj, &cpu_common_attr_group);
1178 	if (rc)
1179 		goto out_cpu;
1180 	rc = topology_cpu_init(c);
1181 	if (rc)
1182 		goto out_topology;
1183 	return 0;
1184 
1185 out_topology:
1186 	sysfs_remove_group(&s->kobj, &cpu_common_attr_group);
1187 out_cpu:
1188 	unregister_cpu(c);
1189 out:
1190 	return rc;
1191 }
1192 
1193 int __ref smp_rescan_cpus(void)
1194 {
1195 	struct sclp_core_info *info;
1196 	int nr;
1197 
1198 	info = kzalloc(sizeof(*info), GFP_KERNEL);
1199 	if (!info)
1200 		return -ENOMEM;
1201 	smp_get_core_info(info, 0);
1202 	nr = __smp_rescan_cpus(info, false);
1203 	kfree(info);
1204 	if (nr)
1205 		topology_schedule_update();
1206 	return 0;
1207 }
1208 
1209 static ssize_t __ref rescan_store(struct device *dev,
1210 				  struct device_attribute *attr,
1211 				  const char *buf,
1212 				  size_t count)
1213 {
1214 	int rc;
1215 
1216 	rc = lock_device_hotplug_sysfs();
1217 	if (rc)
1218 		return rc;
1219 	rc = smp_rescan_cpus();
1220 	unlock_device_hotplug();
1221 	return rc ? rc : count;
1222 }
1223 static DEVICE_ATTR_WO(rescan);
1224 
1225 static int __init s390_smp_init(void)
1226 {
1227 	struct device *dev_root;
1228 	int cpu, rc = 0;
1229 
1230 	dev_root = bus_get_dev_root(&cpu_subsys);
1231 	if (dev_root) {
1232 		rc = device_create_file(dev_root, &dev_attr_rescan);
1233 		put_device(dev_root);
1234 		if (rc)
1235 			return rc;
1236 	}
1237 
1238 	for_each_present_cpu(cpu) {
1239 		rc = smp_add_present_cpu(cpu);
1240 		if (rc)
1241 			goto out;
1242 	}
1243 
1244 	rc = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "s390/smp:online",
1245 			       smp_cpu_online, smp_cpu_pre_down);
1246 	rc = rc <= 0 ? rc : 0;
1247 out:
1248 	return rc;
1249 }
1250 subsys_initcall(s390_smp_init);
1251 
1252 static __always_inline void set_new_lowcore(struct lowcore *lc)
1253 {
1254 	union register_pair dst, src;
1255 	u32 pfx;
1256 
1257 	src.even = (unsigned long) &S390_lowcore;
1258 	src.odd  = sizeof(S390_lowcore);
1259 	dst.even = (unsigned long) lc;
1260 	dst.odd  = sizeof(*lc);
1261 	pfx = __pa(lc);
1262 
1263 	asm volatile(
1264 		"	mvcl	%[dst],%[src]\n"
1265 		"	spx	%[pfx]\n"
1266 		: [dst] "+&d" (dst.pair), [src] "+&d" (src.pair)
1267 		: [pfx] "Q" (pfx)
1268 		: "memory", "cc");
1269 }
1270 
1271 int __init smp_reinit_ipl_cpu(void)
1272 {
1273 	unsigned long async_stack, nodat_stack, mcck_stack;
1274 	struct lowcore *lc, *lc_ipl;
1275 	unsigned long flags, cr0;
1276 	u64 mcesad;
1277 
1278 	lc_ipl = lowcore_ptr[0];
1279 	lc = (struct lowcore *)	__get_free_pages(GFP_KERNEL | GFP_DMA, LC_ORDER);
1280 	nodat_stack = __get_free_pages(GFP_KERNEL, THREAD_SIZE_ORDER);
1281 	async_stack = stack_alloc();
1282 	mcck_stack = stack_alloc();
1283 	if (!lc || !nodat_stack || !async_stack || !mcck_stack || nmi_alloc_mcesa(&mcesad))
1284 		panic("Couldn't allocate memory");
1285 
1286 	local_irq_save(flags);
1287 	local_mcck_disable();
1288 	set_new_lowcore(lc);
1289 	S390_lowcore.nodat_stack = nodat_stack + STACK_INIT_OFFSET;
1290 	S390_lowcore.async_stack = async_stack + STACK_INIT_OFFSET;
1291 	S390_lowcore.mcck_stack = mcck_stack + STACK_INIT_OFFSET;
1292 	__ctl_store(cr0, 0, 0);
1293 	__ctl_clear_bit(0, 28); /* disable lowcore protection */
1294 	S390_lowcore.mcesad = mcesad;
1295 	__ctl_load(cr0, 0, 0);
1296 	if (abs_lowcore_map(0, lc, false))
1297 		panic("Couldn't remap absolute lowcore");
1298 	lowcore_ptr[0] = lc;
1299 	local_mcck_enable();
1300 	local_irq_restore(flags);
1301 
1302 	memblock_free_late(__pa(lc_ipl->mcck_stack - STACK_INIT_OFFSET), THREAD_SIZE);
1303 	memblock_free_late(__pa(lc_ipl->async_stack - STACK_INIT_OFFSET), THREAD_SIZE);
1304 	memblock_free_late(__pa(lc_ipl->nodat_stack - STACK_INIT_OFFSET), THREAD_SIZE);
1305 	memblock_free_late(__pa(lc_ipl), sizeof(*lc_ipl));
1306 	return 0;
1307 }
1308