xref: /linux/arch/s390/kernel/smp.c (revision ef69f8d2ff09518657c3ecaf2db8408c16549829)
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/bootmem.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/kmemleak.h>
31 #include <linux/delay.h>
32 #include <linux/interrupt.h>
33 #include <linux/irqflags.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/memblock.h>
40 #include <linux/kprobes.h>
41 #include <asm/asm-offsets.h>
42 #include <asm/diag.h>
43 #include <asm/switch_to.h>
44 #include <asm/facility.h>
45 #include <asm/ipl.h>
46 #include <asm/setup.h>
47 #include <asm/irq.h>
48 #include <asm/tlbflush.h>
49 #include <asm/vtimer.h>
50 #include <asm/lowcore.h>
51 #include <asm/sclp.h>
52 #include <asm/vdso.h>
53 #include <asm/debug.h>
54 #include <asm/os_info.h>
55 #include <asm/sigp.h>
56 #include <asm/idle.h>
57 #include <asm/nmi.h>
58 #include "entry.h"
59 
60 enum {
61 	ec_schedule = 0,
62 	ec_call_function_single,
63 	ec_stop_cpu,
64 };
65 
66 enum {
67 	CPU_STATE_STANDBY,
68 	CPU_STATE_CONFIGURED,
69 };
70 
71 static DEFINE_PER_CPU(struct cpu *, cpu_device);
72 
73 struct pcpu {
74 	struct lowcore *lowcore;	/* lowcore page(s) for the cpu */
75 	unsigned long ec_mask;		/* bit mask for ec_xxx functions */
76 	unsigned long ec_clk;		/* sigp timestamp for ec_xxx */
77 	signed char state;		/* physical cpu state */
78 	signed char polarization;	/* physical polarization */
79 	u16 address;			/* physical cpu address */
80 };
81 
82 static u8 boot_core_type;
83 static struct pcpu pcpu_devices[NR_CPUS];
84 
85 unsigned int smp_cpu_mt_shift;
86 EXPORT_SYMBOL(smp_cpu_mt_shift);
87 
88 unsigned int smp_cpu_mtid;
89 EXPORT_SYMBOL(smp_cpu_mtid);
90 
91 #ifdef CONFIG_CRASH_DUMP
92 __vector128 __initdata boot_cpu_vector_save_area[__NUM_VXRS];
93 #endif
94 
95 static unsigned int smp_max_threads __initdata = -1U;
96 
97 static int __init early_nosmt(char *s)
98 {
99 	smp_max_threads = 1;
100 	return 0;
101 }
102 early_param("nosmt", early_nosmt);
103 
104 static int __init early_smt(char *s)
105 {
106 	get_option(&s, &smp_max_threads);
107 	return 0;
108 }
109 early_param("smt", early_smt);
110 
111 /*
112  * The smp_cpu_state_mutex must be held when changing the state or polarization
113  * member of a pcpu data structure within the pcpu_devices arreay.
114  */
115 DEFINE_MUTEX(smp_cpu_state_mutex);
116 
117 /*
118  * Signal processor helper functions.
119  */
120 static inline int __pcpu_sigp_relax(u16 addr, u8 order, unsigned long parm)
121 {
122 	int cc;
123 
124 	while (1) {
125 		cc = __pcpu_sigp(addr, order, parm, NULL);
126 		if (cc != SIGP_CC_BUSY)
127 			return cc;
128 		cpu_relax();
129 	}
130 }
131 
132 static int pcpu_sigp_retry(struct pcpu *pcpu, u8 order, u32 parm)
133 {
134 	int cc, retry;
135 
136 	for (retry = 0; ; retry++) {
137 		cc = __pcpu_sigp(pcpu->address, order, parm, NULL);
138 		if (cc != SIGP_CC_BUSY)
139 			break;
140 		if (retry >= 3)
141 			udelay(10);
142 	}
143 	return cc;
144 }
145 
146 static inline int pcpu_stopped(struct pcpu *pcpu)
147 {
148 	u32 uninitialized_var(status);
149 
150 	if (__pcpu_sigp(pcpu->address, SIGP_SENSE,
151 			0, &status) != SIGP_CC_STATUS_STORED)
152 		return 0;
153 	return !!(status & (SIGP_STATUS_CHECK_STOP|SIGP_STATUS_STOPPED));
154 }
155 
156 static inline int pcpu_running(struct pcpu *pcpu)
157 {
158 	if (__pcpu_sigp(pcpu->address, SIGP_SENSE_RUNNING,
159 			0, NULL) != SIGP_CC_STATUS_STORED)
160 		return 1;
161 	/* Status stored condition code is equivalent to cpu not running. */
162 	return 0;
163 }
164 
165 /*
166  * Find struct pcpu by cpu address.
167  */
168 static struct pcpu *pcpu_find_address(const struct cpumask *mask, u16 address)
169 {
170 	int cpu;
171 
172 	for_each_cpu(cpu, mask)
173 		if (pcpu_devices[cpu].address == address)
174 			return pcpu_devices + cpu;
175 	return NULL;
176 }
177 
178 static void pcpu_ec_call(struct pcpu *pcpu, int ec_bit)
179 {
180 	int order;
181 
182 	if (test_and_set_bit(ec_bit, &pcpu->ec_mask))
183 		return;
184 	order = pcpu_running(pcpu) ? SIGP_EXTERNAL_CALL : SIGP_EMERGENCY_SIGNAL;
185 	pcpu->ec_clk = get_tod_clock_fast();
186 	pcpu_sigp_retry(pcpu, order, 0);
187 }
188 
189 #define ASYNC_FRAME_OFFSET (ASYNC_SIZE - STACK_FRAME_OVERHEAD - __PT_SIZE)
190 #define PANIC_FRAME_OFFSET (PAGE_SIZE - STACK_FRAME_OVERHEAD - __PT_SIZE)
191 
192 static int pcpu_alloc_lowcore(struct pcpu *pcpu, int cpu)
193 {
194 	unsigned long async_stack, panic_stack;
195 	struct lowcore *lc;
196 
197 	if (pcpu != &pcpu_devices[0]) {
198 		pcpu->lowcore =	(struct lowcore *)
199 			__get_free_pages(GFP_KERNEL | GFP_DMA, LC_ORDER);
200 		async_stack = __get_free_pages(GFP_KERNEL, ASYNC_ORDER);
201 		panic_stack = __get_free_page(GFP_KERNEL);
202 		if (!pcpu->lowcore || !panic_stack || !async_stack)
203 			goto out;
204 	} else {
205 		async_stack = pcpu->lowcore->async_stack - ASYNC_FRAME_OFFSET;
206 		panic_stack = pcpu->lowcore->panic_stack - PANIC_FRAME_OFFSET;
207 	}
208 	lc = pcpu->lowcore;
209 	memcpy(lc, &S390_lowcore, 512);
210 	memset((char *) lc + 512, 0, sizeof(*lc) - 512);
211 	lc->async_stack = async_stack + ASYNC_FRAME_OFFSET;
212 	lc->panic_stack = panic_stack + PANIC_FRAME_OFFSET;
213 	lc->cpu_nr = cpu;
214 	lc->spinlock_lockval = arch_spin_lockval(cpu);
215 	lc->spinlock_index = 0;
216 	if (nmi_alloc_per_cpu(lc))
217 		goto out;
218 	if (vdso_alloc_per_cpu(lc))
219 		goto out_mcesa;
220 	lowcore_ptr[cpu] = lc;
221 	pcpu_sigp_retry(pcpu, SIGP_SET_PREFIX, (u32)(unsigned long) lc);
222 	return 0;
223 
224 out_mcesa:
225 	nmi_free_per_cpu(lc);
226 out:
227 	if (pcpu != &pcpu_devices[0]) {
228 		free_page(panic_stack);
229 		free_pages(async_stack, ASYNC_ORDER);
230 		free_pages((unsigned long) pcpu->lowcore, LC_ORDER);
231 	}
232 	return -ENOMEM;
233 }
234 
235 #ifdef CONFIG_HOTPLUG_CPU
236 
237 static void pcpu_free_lowcore(struct pcpu *pcpu)
238 {
239 	pcpu_sigp_retry(pcpu, SIGP_SET_PREFIX, 0);
240 	lowcore_ptr[pcpu - pcpu_devices] = NULL;
241 	vdso_free_per_cpu(pcpu->lowcore);
242 	nmi_free_per_cpu(pcpu->lowcore);
243 	if (pcpu == &pcpu_devices[0])
244 		return;
245 	free_page(pcpu->lowcore->panic_stack-PANIC_FRAME_OFFSET);
246 	free_pages(pcpu->lowcore->async_stack-ASYNC_FRAME_OFFSET, ASYNC_ORDER);
247 	free_pages((unsigned long) pcpu->lowcore, LC_ORDER);
248 }
249 
250 #endif /* CONFIG_HOTPLUG_CPU */
251 
252 static void pcpu_prepare_secondary(struct pcpu *pcpu, int cpu)
253 {
254 	struct lowcore *lc = pcpu->lowcore;
255 
256 	cpumask_set_cpu(cpu, &init_mm.context.cpu_attach_mask);
257 	cpumask_set_cpu(cpu, mm_cpumask(&init_mm));
258 	lc->cpu_nr = cpu;
259 	lc->spinlock_lockval = arch_spin_lockval(cpu);
260 	lc->spinlock_index = 0;
261 	lc->percpu_offset = __per_cpu_offset[cpu];
262 	lc->kernel_asce = S390_lowcore.kernel_asce;
263 	lc->machine_flags = S390_lowcore.machine_flags;
264 	lc->user_timer = lc->system_timer = lc->steal_timer = 0;
265 	__ctl_store(lc->cregs_save_area, 0, 15);
266 	save_access_regs((unsigned int *) lc->access_regs_save_area);
267 	memcpy(lc->stfle_fac_list, S390_lowcore.stfle_fac_list,
268 	       MAX_FACILITY_BIT/8);
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 = pcpu->lowcore;
275 
276 	lc->kernel_stack = (unsigned long) task_stack_page(tsk)
277 		+ THREAD_SIZE - STACK_FRAME_OVERHEAD - sizeof(struct pt_regs);
278 	lc->current_task = (unsigned long) tsk;
279 	lc->lpp = LPP_MAGIC;
280 	lc->current_pid = tsk->pid;
281 	lc->user_timer = tsk->thread.user_timer;
282 	lc->guest_timer = tsk->thread.guest_timer;
283 	lc->system_timer = tsk->thread.system_timer;
284 	lc->hardirq_timer = tsk->thread.hardirq_timer;
285 	lc->softirq_timer = tsk->thread.softirq_timer;
286 	lc->steal_timer = 0;
287 }
288 
289 static void pcpu_start_fn(struct pcpu *pcpu, void (*func)(void *), void *data)
290 {
291 	struct lowcore *lc = pcpu->lowcore;
292 
293 	lc->restart_stack = lc->kernel_stack;
294 	lc->restart_fn = (unsigned long) func;
295 	lc->restart_data = (unsigned long) data;
296 	lc->restart_source = -1UL;
297 	pcpu_sigp_retry(pcpu, SIGP_RESTART, 0);
298 }
299 
300 /*
301  * Call function via PSW restart on pcpu and stop the current cpu.
302  */
303 static void pcpu_delegate(struct pcpu *pcpu, void (*func)(void *),
304 			  void *data, unsigned long stack)
305 {
306 	struct lowcore *lc = lowcore_ptr[pcpu - pcpu_devices];
307 	unsigned long source_cpu = stap();
308 
309 	__load_psw_mask(PSW_KERNEL_BITS);
310 	if (pcpu->address == source_cpu)
311 		func(data);	/* should not return */
312 	/* Stop target cpu (if func returns this stops the current cpu). */
313 	pcpu_sigp_retry(pcpu, SIGP_STOP, 0);
314 	/* Restart func on the target cpu and stop the current cpu. */
315 	mem_assign_absolute(lc->restart_stack, stack);
316 	mem_assign_absolute(lc->restart_fn, (unsigned long) func);
317 	mem_assign_absolute(lc->restart_data, (unsigned long) data);
318 	mem_assign_absolute(lc->restart_source, source_cpu);
319 	asm volatile(
320 		"0:	sigp	0,%0,%2	# sigp restart to target cpu\n"
321 		"	brc	2,0b	# busy, try again\n"
322 		"1:	sigp	0,%1,%3	# sigp stop to current cpu\n"
323 		"	brc	2,1b	# busy, try again\n"
324 		: : "d" (pcpu->address), "d" (source_cpu),
325 		    "K" (SIGP_RESTART), "K" (SIGP_STOP)
326 		: "0", "1", "cc");
327 	for (;;) ;
328 }
329 
330 /*
331  * Enable additional logical cpus for multi-threading.
332  */
333 static int pcpu_set_smt(unsigned int mtid)
334 {
335 	int cc;
336 
337 	if (smp_cpu_mtid == mtid)
338 		return 0;
339 	cc = __pcpu_sigp(0, SIGP_SET_MULTI_THREADING, mtid, NULL);
340 	if (cc == 0) {
341 		smp_cpu_mtid = mtid;
342 		smp_cpu_mt_shift = 0;
343 		while (smp_cpu_mtid >= (1U << smp_cpu_mt_shift))
344 			smp_cpu_mt_shift++;
345 		pcpu_devices[0].address = stap();
346 	}
347 	return cc;
348 }
349 
350 /*
351  * Call function on an online CPU.
352  */
353 void smp_call_online_cpu(void (*func)(void *), void *data)
354 {
355 	struct pcpu *pcpu;
356 
357 	/* Use the current cpu if it is online. */
358 	pcpu = pcpu_find_address(cpu_online_mask, stap());
359 	if (!pcpu)
360 		/* Use the first online cpu. */
361 		pcpu = pcpu_devices + cpumask_first(cpu_online_mask);
362 	pcpu_delegate(pcpu, func, data, (unsigned long) restart_stack);
363 }
364 
365 /*
366  * Call function on the ipl CPU.
367  */
368 void smp_call_ipl_cpu(void (*func)(void *), void *data)
369 {
370 	pcpu_delegate(&pcpu_devices[0], func, data,
371 		      pcpu_devices->lowcore->panic_stack -
372 		      PANIC_FRAME_OFFSET + PAGE_SIZE);
373 }
374 
375 int smp_find_processor_id(u16 address)
376 {
377 	int cpu;
378 
379 	for_each_present_cpu(cpu)
380 		if (pcpu_devices[cpu].address == address)
381 			return cpu;
382 	return -1;
383 }
384 
385 bool arch_vcpu_is_preempted(int cpu)
386 {
387 	if (test_cpu_flag_of(CIF_ENABLED_WAIT, cpu))
388 		return false;
389 	if (pcpu_running(pcpu_devices + cpu))
390 		return false;
391 	return true;
392 }
393 EXPORT_SYMBOL(arch_vcpu_is_preempted);
394 
395 void smp_yield_cpu(int cpu)
396 {
397 	if (MACHINE_HAS_DIAG9C) {
398 		diag_stat_inc_norecursion(DIAG_STAT_X09C);
399 		asm volatile("diag %0,0,0x9c"
400 			     : : "d" (pcpu_devices[cpu].address));
401 	} else if (MACHINE_HAS_DIAG44) {
402 		diag_stat_inc_norecursion(DIAG_STAT_X044);
403 		asm volatile("diag 0,0,0x44");
404 	}
405 }
406 
407 /*
408  * Send cpus emergency shutdown signal. This gives the cpus the
409  * opportunity to complete outstanding interrupts.
410  */
411 void notrace smp_emergency_stop(void)
412 {
413 	cpumask_t cpumask;
414 	u64 end;
415 	int cpu;
416 
417 	cpumask_copy(&cpumask, cpu_online_mask);
418 	cpumask_clear_cpu(smp_processor_id(), &cpumask);
419 
420 	end = get_tod_clock() + (1000000UL << 12);
421 	for_each_cpu(cpu, &cpumask) {
422 		struct pcpu *pcpu = pcpu_devices + cpu;
423 		set_bit(ec_stop_cpu, &pcpu->ec_mask);
424 		while (__pcpu_sigp(pcpu->address, SIGP_EMERGENCY_SIGNAL,
425 				   0, NULL) == SIGP_CC_BUSY &&
426 		       get_tod_clock() < end)
427 			cpu_relax();
428 	}
429 	while (get_tod_clock() < end) {
430 		for_each_cpu(cpu, &cpumask)
431 			if (pcpu_stopped(pcpu_devices + cpu))
432 				cpumask_clear_cpu(cpu, &cpumask);
433 		if (cpumask_empty(&cpumask))
434 			break;
435 		cpu_relax();
436 	}
437 }
438 NOKPROBE_SYMBOL(smp_emergency_stop);
439 
440 /*
441  * Stop all cpus but the current one.
442  */
443 void smp_send_stop(void)
444 {
445 	int cpu;
446 
447 	/* Disable all interrupts/machine checks */
448 	__load_psw_mask(PSW_KERNEL_BITS | PSW_MASK_DAT);
449 	trace_hardirqs_off();
450 
451 	debug_set_critical();
452 
453 	if (oops_in_progress)
454 		smp_emergency_stop();
455 
456 	/* stop all processors */
457 	for_each_online_cpu(cpu) {
458 		if (cpu == smp_processor_id())
459 			continue;
460 		pcpu_sigp_retry(pcpu_devices + cpu, SIGP_STOP, 0);
461 		while (!pcpu_stopped(pcpu_devices + cpu))
462 			cpu_relax();
463 	}
464 }
465 
466 /*
467  * This is the main routine where commands issued by other
468  * cpus are handled.
469  */
470 static void smp_handle_ext_call(void)
471 {
472 	unsigned long bits;
473 
474 	/* handle bit signal external calls */
475 	bits = xchg(&pcpu_devices[smp_processor_id()].ec_mask, 0);
476 	if (test_bit(ec_stop_cpu, &bits))
477 		smp_stop_cpu();
478 	if (test_bit(ec_schedule, &bits))
479 		scheduler_ipi();
480 	if (test_bit(ec_call_function_single, &bits))
481 		generic_smp_call_function_single_interrupt();
482 }
483 
484 static void do_ext_call_interrupt(struct ext_code ext_code,
485 				  unsigned int param32, unsigned long param64)
486 {
487 	inc_irq_stat(ext_code.code == 0x1202 ? IRQEXT_EXC : IRQEXT_EMS);
488 	smp_handle_ext_call();
489 }
490 
491 void arch_send_call_function_ipi_mask(const struct cpumask *mask)
492 {
493 	int cpu;
494 
495 	for_each_cpu(cpu, mask)
496 		pcpu_ec_call(pcpu_devices + cpu, ec_call_function_single);
497 }
498 
499 void arch_send_call_function_single_ipi(int cpu)
500 {
501 	pcpu_ec_call(pcpu_devices + cpu, ec_call_function_single);
502 }
503 
504 /*
505  * this function sends a 'reschedule' IPI to another CPU.
506  * it goes straight through and wastes no time serializing
507  * anything. Worst case is that we lose a reschedule ...
508  */
509 void smp_send_reschedule(int cpu)
510 {
511 	pcpu_ec_call(pcpu_devices + cpu, ec_schedule);
512 }
513 
514 /*
515  * parameter area for the set/clear control bit callbacks
516  */
517 struct ec_creg_mask_parms {
518 	unsigned long orval;
519 	unsigned long andval;
520 	int cr;
521 };
522 
523 /*
524  * callback for setting/clearing control bits
525  */
526 static void smp_ctl_bit_callback(void *info)
527 {
528 	struct ec_creg_mask_parms *pp = info;
529 	unsigned long cregs[16];
530 
531 	__ctl_store(cregs, 0, 15);
532 	cregs[pp->cr] = (cregs[pp->cr] & pp->andval) | pp->orval;
533 	__ctl_load(cregs, 0, 15);
534 }
535 
536 /*
537  * Set a bit in a control register of all cpus
538  */
539 void smp_ctl_set_bit(int cr, int bit)
540 {
541 	struct ec_creg_mask_parms parms = { 1UL << bit, -1UL, cr };
542 
543 	on_each_cpu(smp_ctl_bit_callback, &parms, 1);
544 }
545 EXPORT_SYMBOL(smp_ctl_set_bit);
546 
547 /*
548  * Clear a bit in a control register of all cpus
549  */
550 void smp_ctl_clear_bit(int cr, int bit)
551 {
552 	struct ec_creg_mask_parms parms = { 0, ~(1UL << bit), cr };
553 
554 	on_each_cpu(smp_ctl_bit_callback, &parms, 1);
555 }
556 EXPORT_SYMBOL(smp_ctl_clear_bit);
557 
558 #ifdef CONFIG_CRASH_DUMP
559 
560 int smp_store_status(int cpu)
561 {
562 	struct pcpu *pcpu = pcpu_devices + cpu;
563 	unsigned long pa;
564 
565 	pa = __pa(&pcpu->lowcore->floating_pt_save_area);
566 	if (__pcpu_sigp_relax(pcpu->address, SIGP_STORE_STATUS_AT_ADDRESS,
567 			      pa) != SIGP_CC_ORDER_CODE_ACCEPTED)
568 		return -EIO;
569 	if (!MACHINE_HAS_VX && !MACHINE_HAS_GS)
570 		return 0;
571 	pa = __pa(pcpu->lowcore->mcesad & MCESA_ORIGIN_MASK);
572 	if (MACHINE_HAS_GS)
573 		pa |= pcpu->lowcore->mcesad & MCESA_LC_MASK;
574 	if (__pcpu_sigp_relax(pcpu->address, SIGP_STORE_ADDITIONAL_STATUS,
575 			      pa) != SIGP_CC_ORDER_CODE_ACCEPTED)
576 		return -EIO;
577 	return 0;
578 }
579 
580 /*
581  * Collect CPU state of the previous, crashed system.
582  * There are four cases:
583  * 1) standard zfcp dump
584  *    condition: OLDMEM_BASE == NULL && ipl_info.type == IPL_TYPE_FCP_DUMP
585  *    The state for all CPUs except the boot CPU needs to be collected
586  *    with sigp stop-and-store-status. The boot CPU state is located in
587  *    the absolute lowcore of the memory stored in the HSA. The zcore code
588  *    will copy the boot CPU state from the HSA.
589  * 2) stand-alone kdump for SCSI (zfcp dump with swapped memory)
590  *    condition: OLDMEM_BASE != NULL && ipl_info.type == IPL_TYPE_FCP_DUMP
591  *    The state for all CPUs except the boot CPU needs to be collected
592  *    with sigp stop-and-store-status. The firmware or the boot-loader
593  *    stored the registers of the boot CPU in the absolute lowcore in the
594  *    memory of the old system.
595  * 3) kdump and the old kernel did not store the CPU state,
596  *    or stand-alone kdump for DASD
597  *    condition: OLDMEM_BASE != NULL && !is_kdump_kernel()
598  *    The state for all CPUs except the boot CPU needs to be collected
599  *    with sigp stop-and-store-status. The kexec code or the boot-loader
600  *    stored the registers of the boot CPU in the memory of the old system.
601  * 4) kdump and the old kernel stored the CPU state
602  *    condition: OLDMEM_BASE != NULL && is_kdump_kernel()
603  *    This case does not exist for s390 anymore, setup_arch explicitly
604  *    deactivates the elfcorehdr= kernel parameter
605  */
606 static __init void smp_save_cpu_vxrs(struct save_area *sa, u16 addr,
607 				     bool is_boot_cpu, unsigned long page)
608 {
609 	__vector128 *vxrs = (__vector128 *) page;
610 
611 	if (is_boot_cpu)
612 		vxrs = boot_cpu_vector_save_area;
613 	else
614 		__pcpu_sigp_relax(addr, SIGP_STORE_ADDITIONAL_STATUS, page);
615 	save_area_add_vxrs(sa, vxrs);
616 }
617 
618 static __init void smp_save_cpu_regs(struct save_area *sa, u16 addr,
619 				     bool is_boot_cpu, unsigned long page)
620 {
621 	void *regs = (void *) page;
622 
623 	if (is_boot_cpu)
624 		copy_oldmem_kernel(regs, (void *) __LC_FPREGS_SAVE_AREA, 512);
625 	else
626 		__pcpu_sigp_relax(addr, SIGP_STORE_STATUS_AT_ADDRESS, page);
627 	save_area_add_regs(sa, regs);
628 }
629 
630 void __init smp_save_dump_cpus(void)
631 {
632 	int addr, boot_cpu_addr, max_cpu_addr;
633 	struct save_area *sa;
634 	unsigned long page;
635 	bool is_boot_cpu;
636 
637 	if (!(OLDMEM_BASE || ipl_info.type == IPL_TYPE_FCP_DUMP))
638 		/* No previous system present, normal boot. */
639 		return;
640 	/* Allocate a page as dumping area for the store status sigps */
641 	page = memblock_alloc_base(PAGE_SIZE, PAGE_SIZE, 1UL << 31);
642 	/* Set multi-threading state to the previous system. */
643 	pcpu_set_smt(sclp.mtid_prev);
644 	boot_cpu_addr = stap();
645 	max_cpu_addr = SCLP_MAX_CORES << sclp.mtid_prev;
646 	for (addr = 0; addr <= max_cpu_addr; addr++) {
647 		if (__pcpu_sigp_relax(addr, SIGP_SENSE, 0) ==
648 		    SIGP_CC_NOT_OPERATIONAL)
649 			continue;
650 		is_boot_cpu = (addr == boot_cpu_addr);
651 		/* Allocate save area */
652 		sa = save_area_alloc(is_boot_cpu);
653 		if (!sa)
654 			panic("could not allocate memory for save area\n");
655 		if (MACHINE_HAS_VX)
656 			/* Get the vector registers */
657 			smp_save_cpu_vxrs(sa, addr, is_boot_cpu, page);
658 		/*
659 		 * For a zfcp dump OLDMEM_BASE == NULL and the registers
660 		 * of the boot CPU are stored in the HSA. To retrieve
661 		 * these registers an SCLP request is required which is
662 		 * done by drivers/s390/char/zcore.c:init_cpu_info()
663 		 */
664 		if (!is_boot_cpu || OLDMEM_BASE)
665 			/* Get the CPU registers */
666 			smp_save_cpu_regs(sa, addr, is_boot_cpu, page);
667 	}
668 	memblock_free(page, PAGE_SIZE);
669 	diag308_reset();
670 	pcpu_set_smt(0);
671 }
672 #endif /* CONFIG_CRASH_DUMP */
673 
674 void smp_cpu_set_polarization(int cpu, int val)
675 {
676 	pcpu_devices[cpu].polarization = val;
677 }
678 
679 int smp_cpu_get_polarization(int cpu)
680 {
681 	return pcpu_devices[cpu].polarization;
682 }
683 
684 static void __ref smp_get_core_info(struct sclp_core_info *info, int early)
685 {
686 	static int use_sigp_detection;
687 	int address;
688 
689 	if (use_sigp_detection || sclp_get_core_info(info, early)) {
690 		use_sigp_detection = 1;
691 		for (address = 0;
692 		     address < (SCLP_MAX_CORES << smp_cpu_mt_shift);
693 		     address += (1U << smp_cpu_mt_shift)) {
694 			if (__pcpu_sigp_relax(address, SIGP_SENSE, 0) ==
695 			    SIGP_CC_NOT_OPERATIONAL)
696 				continue;
697 			info->core[info->configured].core_id =
698 				address >> smp_cpu_mt_shift;
699 			info->configured++;
700 		}
701 		info->combined = info->configured;
702 	}
703 }
704 
705 static int smp_add_present_cpu(int cpu);
706 
707 static int __smp_rescan_cpus(struct sclp_core_info *info, int sysfs_add)
708 {
709 	struct pcpu *pcpu;
710 	cpumask_t avail;
711 	int cpu, nr, i, j;
712 	u16 address;
713 
714 	nr = 0;
715 	cpumask_xor(&avail, cpu_possible_mask, cpu_present_mask);
716 	cpu = cpumask_first(&avail);
717 	for (i = 0; (i < info->combined) && (cpu < nr_cpu_ids); i++) {
718 		if (sclp.has_core_type && info->core[i].type != boot_core_type)
719 			continue;
720 		address = info->core[i].core_id << smp_cpu_mt_shift;
721 		for (j = 0; j <= smp_cpu_mtid; j++) {
722 			if (pcpu_find_address(cpu_present_mask, address + j))
723 				continue;
724 			pcpu = pcpu_devices + cpu;
725 			pcpu->address = address + j;
726 			pcpu->state =
727 				(cpu >= info->configured*(smp_cpu_mtid + 1)) ?
728 				CPU_STATE_STANDBY : CPU_STATE_CONFIGURED;
729 			smp_cpu_set_polarization(cpu, POLARIZATION_UNKNOWN);
730 			set_cpu_present(cpu, true);
731 			if (sysfs_add && smp_add_present_cpu(cpu) != 0)
732 				set_cpu_present(cpu, false);
733 			else
734 				nr++;
735 			cpu = cpumask_next(cpu, &avail);
736 			if (cpu >= nr_cpu_ids)
737 				break;
738 		}
739 	}
740 	return nr;
741 }
742 
743 void __init smp_detect_cpus(void)
744 {
745 	unsigned int cpu, mtid, c_cpus, s_cpus;
746 	struct sclp_core_info *info;
747 	u16 address;
748 
749 	/* Get CPU information */
750 	info = memblock_virt_alloc(sizeof(*info), 8);
751 	smp_get_core_info(info, 1);
752 	/* Find boot CPU type */
753 	if (sclp.has_core_type) {
754 		address = stap();
755 		for (cpu = 0; cpu < info->combined; cpu++)
756 			if (info->core[cpu].core_id == address) {
757 				/* The boot cpu dictates the cpu type. */
758 				boot_core_type = info->core[cpu].type;
759 				break;
760 			}
761 		if (cpu >= info->combined)
762 			panic("Could not find boot CPU type");
763 	}
764 
765 	/* Set multi-threading state for the current system */
766 	mtid = boot_core_type ? sclp.mtid : sclp.mtid_cp;
767 	mtid = (mtid < smp_max_threads) ? mtid : smp_max_threads - 1;
768 	pcpu_set_smt(mtid);
769 
770 	/* Print number of CPUs */
771 	c_cpus = s_cpus = 0;
772 	for (cpu = 0; cpu < info->combined; cpu++) {
773 		if (sclp.has_core_type &&
774 		    info->core[cpu].type != boot_core_type)
775 			continue;
776 		if (cpu < info->configured)
777 			c_cpus += smp_cpu_mtid + 1;
778 		else
779 			s_cpus += smp_cpu_mtid + 1;
780 	}
781 	pr_info("%d configured CPUs, %d standby CPUs\n", c_cpus, s_cpus);
782 
783 	/* Add CPUs present at boot */
784 	get_online_cpus();
785 	__smp_rescan_cpus(info, 0);
786 	put_online_cpus();
787 	memblock_free_early((unsigned long)info, sizeof(*info));
788 }
789 
790 /*
791  *	Activate a secondary processor.
792  */
793 static void smp_start_secondary(void *cpuvoid)
794 {
795 	int cpu = smp_processor_id();
796 
797 	S390_lowcore.last_update_clock = get_tod_clock();
798 	S390_lowcore.restart_stack = (unsigned long) restart_stack;
799 	S390_lowcore.restart_fn = (unsigned long) do_restart;
800 	S390_lowcore.restart_data = 0;
801 	S390_lowcore.restart_source = -1UL;
802 	restore_access_regs(S390_lowcore.access_regs_save_area);
803 	__ctl_load(S390_lowcore.cregs_save_area, 0, 15);
804 	__load_psw_mask(PSW_KERNEL_BITS | PSW_MASK_DAT);
805 	cpu_init();
806 	preempt_disable();
807 	init_cpu_timer();
808 	vtime_init();
809 	pfault_init();
810 	notify_cpu_starting(cpu);
811 	if (topology_cpu_dedicated(cpu))
812 		set_cpu_flag(CIF_DEDICATED_CPU);
813 	else
814 		clear_cpu_flag(CIF_DEDICATED_CPU);
815 	set_cpu_online(cpu, true);
816 	inc_irq_stat(CPU_RST);
817 	local_irq_enable();
818 	cpu_startup_entry(CPUHP_AP_ONLINE_IDLE);
819 }
820 
821 /* Upping and downing of CPUs */
822 int __cpu_up(unsigned int cpu, struct task_struct *tidle)
823 {
824 	struct pcpu *pcpu;
825 	int base, i, rc;
826 
827 	pcpu = pcpu_devices + cpu;
828 	if (pcpu->state != CPU_STATE_CONFIGURED)
829 		return -EIO;
830 	base = smp_get_base_cpu(cpu);
831 	for (i = 0; i <= smp_cpu_mtid; i++) {
832 		if (base + i < nr_cpu_ids)
833 			if (cpu_online(base + i))
834 				break;
835 	}
836 	/*
837 	 * If this is the first CPU of the core to get online
838 	 * do an initial CPU reset.
839 	 */
840 	if (i > smp_cpu_mtid &&
841 	    pcpu_sigp_retry(pcpu_devices + base, SIGP_INITIAL_CPU_RESET, 0) !=
842 	    SIGP_CC_ORDER_CODE_ACCEPTED)
843 		return -EIO;
844 
845 	rc = pcpu_alloc_lowcore(pcpu, cpu);
846 	if (rc)
847 		return rc;
848 	pcpu_prepare_secondary(pcpu, cpu);
849 	pcpu_attach_task(pcpu, tidle);
850 	pcpu_start_fn(pcpu, smp_start_secondary, NULL);
851 	/* Wait until cpu puts itself in the online & active maps */
852 	while (!cpu_online(cpu))
853 		cpu_relax();
854 	return 0;
855 }
856 
857 static unsigned int setup_possible_cpus __initdata;
858 
859 static int __init _setup_possible_cpus(char *s)
860 {
861 	get_option(&s, &setup_possible_cpus);
862 	return 0;
863 }
864 early_param("possible_cpus", _setup_possible_cpus);
865 
866 #ifdef CONFIG_HOTPLUG_CPU
867 
868 int __cpu_disable(void)
869 {
870 	unsigned long cregs[16];
871 
872 	/* Handle possible pending IPIs */
873 	smp_handle_ext_call();
874 	set_cpu_online(smp_processor_id(), false);
875 	/* Disable pseudo page faults on this cpu. */
876 	pfault_fini();
877 	/* Disable interrupt sources via control register. */
878 	__ctl_store(cregs, 0, 15);
879 	cregs[0]  &= ~0x0000ee70UL;	/* disable all external interrupts */
880 	cregs[6]  &= ~0xff000000UL;	/* disable all I/O interrupts */
881 	cregs[14] &= ~0x1f000000UL;	/* disable most machine checks */
882 	__ctl_load(cregs, 0, 15);
883 	clear_cpu_flag(CIF_NOHZ_DELAY);
884 	return 0;
885 }
886 
887 void __cpu_die(unsigned int cpu)
888 {
889 	struct pcpu *pcpu;
890 
891 	/* Wait until target cpu is down */
892 	pcpu = pcpu_devices + cpu;
893 	while (!pcpu_stopped(pcpu))
894 		cpu_relax();
895 	pcpu_free_lowcore(pcpu);
896 	cpumask_clear_cpu(cpu, mm_cpumask(&init_mm));
897 	cpumask_clear_cpu(cpu, &init_mm.context.cpu_attach_mask);
898 }
899 
900 void __noreturn cpu_die(void)
901 {
902 	idle_task_exit();
903 	pcpu_sigp_retry(pcpu_devices + smp_processor_id(), SIGP_STOP, 0);
904 	for (;;) ;
905 }
906 
907 #endif /* CONFIG_HOTPLUG_CPU */
908 
909 void __init smp_fill_possible_mask(void)
910 {
911 	unsigned int possible, sclp_max, cpu;
912 
913 	sclp_max = max(sclp.mtid, sclp.mtid_cp) + 1;
914 	sclp_max = min(smp_max_threads, sclp_max);
915 	sclp_max = (sclp.max_cores * sclp_max) ?: nr_cpu_ids;
916 	possible = setup_possible_cpus ?: nr_cpu_ids;
917 	possible = min(possible, sclp_max);
918 	for (cpu = 0; cpu < possible && cpu < nr_cpu_ids; cpu++)
919 		set_cpu_possible(cpu, true);
920 }
921 
922 void __init smp_prepare_cpus(unsigned int max_cpus)
923 {
924 	/* request the 0x1201 emergency signal external interrupt */
925 	if (register_external_irq(EXT_IRQ_EMERGENCY_SIG, do_ext_call_interrupt))
926 		panic("Couldn't request external interrupt 0x1201");
927 	/* request the 0x1202 external call external interrupt */
928 	if (register_external_irq(EXT_IRQ_EXTERNAL_CALL, do_ext_call_interrupt))
929 		panic("Couldn't request external interrupt 0x1202");
930 }
931 
932 void __init smp_prepare_boot_cpu(void)
933 {
934 	struct pcpu *pcpu = pcpu_devices;
935 
936 	WARN_ON(!cpu_present(0) || !cpu_online(0));
937 	pcpu->state = CPU_STATE_CONFIGURED;
938 	pcpu->lowcore = (struct lowcore *)(unsigned long) store_prefix();
939 	S390_lowcore.percpu_offset = __per_cpu_offset[0];
940 	smp_cpu_set_polarization(0, POLARIZATION_UNKNOWN);
941 }
942 
943 void __init smp_cpus_done(unsigned int max_cpus)
944 {
945 }
946 
947 void __init smp_setup_processor_id(void)
948 {
949 	pcpu_devices[0].address = stap();
950 	S390_lowcore.cpu_nr = 0;
951 	S390_lowcore.spinlock_lockval = arch_spin_lockval(0);
952 	S390_lowcore.spinlock_index = 0;
953 }
954 
955 /*
956  * the frequency of the profiling timer can be changed
957  * by writing a multiplier value into /proc/profile.
958  *
959  * usually you want to run this on all CPUs ;)
960  */
961 int setup_profiling_timer(unsigned int multiplier)
962 {
963 	return 0;
964 }
965 
966 #ifdef CONFIG_HOTPLUG_CPU
967 static ssize_t cpu_configure_show(struct device *dev,
968 				  struct device_attribute *attr, char *buf)
969 {
970 	ssize_t count;
971 
972 	mutex_lock(&smp_cpu_state_mutex);
973 	count = sprintf(buf, "%d\n", pcpu_devices[dev->id].state);
974 	mutex_unlock(&smp_cpu_state_mutex);
975 	return count;
976 }
977 
978 static ssize_t cpu_configure_store(struct device *dev,
979 				   struct device_attribute *attr,
980 				   const char *buf, size_t count)
981 {
982 	struct pcpu *pcpu;
983 	int cpu, val, rc, i;
984 	char delim;
985 
986 	if (sscanf(buf, "%d %c", &val, &delim) != 1)
987 		return -EINVAL;
988 	if (val != 0 && val != 1)
989 		return -EINVAL;
990 	get_online_cpus();
991 	mutex_lock(&smp_cpu_state_mutex);
992 	rc = -EBUSY;
993 	/* disallow configuration changes of online cpus and cpu 0 */
994 	cpu = dev->id;
995 	cpu = smp_get_base_cpu(cpu);
996 	if (cpu == 0)
997 		goto out;
998 	for (i = 0; i <= smp_cpu_mtid; i++)
999 		if (cpu_online(cpu + i))
1000 			goto out;
1001 	pcpu = pcpu_devices + cpu;
1002 	rc = 0;
1003 	switch (val) {
1004 	case 0:
1005 		if (pcpu->state != CPU_STATE_CONFIGURED)
1006 			break;
1007 		rc = sclp_core_deconfigure(pcpu->address >> smp_cpu_mt_shift);
1008 		if (rc)
1009 			break;
1010 		for (i = 0; i <= smp_cpu_mtid; i++) {
1011 			if (cpu + i >= nr_cpu_ids || !cpu_present(cpu + i))
1012 				continue;
1013 			pcpu[i].state = CPU_STATE_STANDBY;
1014 			smp_cpu_set_polarization(cpu + i,
1015 						 POLARIZATION_UNKNOWN);
1016 		}
1017 		topology_expect_change();
1018 		break;
1019 	case 1:
1020 		if (pcpu->state != CPU_STATE_STANDBY)
1021 			break;
1022 		rc = sclp_core_configure(pcpu->address >> smp_cpu_mt_shift);
1023 		if (rc)
1024 			break;
1025 		for (i = 0; i <= smp_cpu_mtid; i++) {
1026 			if (cpu + i >= nr_cpu_ids || !cpu_present(cpu + i))
1027 				continue;
1028 			pcpu[i].state = CPU_STATE_CONFIGURED;
1029 			smp_cpu_set_polarization(cpu + i,
1030 						 POLARIZATION_UNKNOWN);
1031 		}
1032 		topology_expect_change();
1033 		break;
1034 	default:
1035 		break;
1036 	}
1037 out:
1038 	mutex_unlock(&smp_cpu_state_mutex);
1039 	put_online_cpus();
1040 	return rc ? rc : count;
1041 }
1042 static DEVICE_ATTR(configure, 0644, cpu_configure_show, cpu_configure_store);
1043 #endif /* CONFIG_HOTPLUG_CPU */
1044 
1045 static ssize_t show_cpu_address(struct device *dev,
1046 				struct device_attribute *attr, char *buf)
1047 {
1048 	return sprintf(buf, "%d\n", pcpu_devices[dev->id].address);
1049 }
1050 static DEVICE_ATTR(address, 0444, show_cpu_address, NULL);
1051 
1052 static struct attribute *cpu_common_attrs[] = {
1053 #ifdef CONFIG_HOTPLUG_CPU
1054 	&dev_attr_configure.attr,
1055 #endif
1056 	&dev_attr_address.attr,
1057 	NULL,
1058 };
1059 
1060 static struct attribute_group cpu_common_attr_group = {
1061 	.attrs = cpu_common_attrs,
1062 };
1063 
1064 static struct attribute *cpu_online_attrs[] = {
1065 	&dev_attr_idle_count.attr,
1066 	&dev_attr_idle_time_us.attr,
1067 	NULL,
1068 };
1069 
1070 static struct attribute_group cpu_online_attr_group = {
1071 	.attrs = cpu_online_attrs,
1072 };
1073 
1074 static int smp_cpu_online(unsigned int cpu)
1075 {
1076 	struct device *s = &per_cpu(cpu_device, cpu)->dev;
1077 
1078 	return sysfs_create_group(&s->kobj, &cpu_online_attr_group);
1079 }
1080 static int smp_cpu_pre_down(unsigned int cpu)
1081 {
1082 	struct device *s = &per_cpu(cpu_device, cpu)->dev;
1083 
1084 	sysfs_remove_group(&s->kobj, &cpu_online_attr_group);
1085 	return 0;
1086 }
1087 
1088 static int smp_add_present_cpu(int cpu)
1089 {
1090 	struct device *s;
1091 	struct cpu *c;
1092 	int rc;
1093 
1094 	c = kzalloc(sizeof(*c), GFP_KERNEL);
1095 	if (!c)
1096 		return -ENOMEM;
1097 	per_cpu(cpu_device, cpu) = c;
1098 	s = &c->dev;
1099 	c->hotpluggable = 1;
1100 	rc = register_cpu(c, cpu);
1101 	if (rc)
1102 		goto out;
1103 	rc = sysfs_create_group(&s->kobj, &cpu_common_attr_group);
1104 	if (rc)
1105 		goto out_cpu;
1106 	rc = topology_cpu_init(c);
1107 	if (rc)
1108 		goto out_topology;
1109 	return 0;
1110 
1111 out_topology:
1112 	sysfs_remove_group(&s->kobj, &cpu_common_attr_group);
1113 out_cpu:
1114 #ifdef CONFIG_HOTPLUG_CPU
1115 	unregister_cpu(c);
1116 #endif
1117 out:
1118 	return rc;
1119 }
1120 
1121 #ifdef CONFIG_HOTPLUG_CPU
1122 
1123 int __ref smp_rescan_cpus(void)
1124 {
1125 	struct sclp_core_info *info;
1126 	int nr;
1127 
1128 	info = kzalloc(sizeof(*info), GFP_KERNEL);
1129 	if (!info)
1130 		return -ENOMEM;
1131 	smp_get_core_info(info, 0);
1132 	get_online_cpus();
1133 	mutex_lock(&smp_cpu_state_mutex);
1134 	nr = __smp_rescan_cpus(info, 1);
1135 	mutex_unlock(&smp_cpu_state_mutex);
1136 	put_online_cpus();
1137 	kfree(info);
1138 	if (nr)
1139 		topology_schedule_update();
1140 	return 0;
1141 }
1142 
1143 static ssize_t __ref rescan_store(struct device *dev,
1144 				  struct device_attribute *attr,
1145 				  const char *buf,
1146 				  size_t count)
1147 {
1148 	int rc;
1149 
1150 	rc = smp_rescan_cpus();
1151 	return rc ? rc : count;
1152 }
1153 static DEVICE_ATTR(rescan, 0200, NULL, rescan_store);
1154 #endif /* CONFIG_HOTPLUG_CPU */
1155 
1156 static int __init s390_smp_init(void)
1157 {
1158 	int cpu, rc = 0;
1159 
1160 #ifdef CONFIG_HOTPLUG_CPU
1161 	rc = device_create_file(cpu_subsys.dev_root, &dev_attr_rescan);
1162 	if (rc)
1163 		return rc;
1164 #endif
1165 	for_each_present_cpu(cpu) {
1166 		rc = smp_add_present_cpu(cpu);
1167 		if (rc)
1168 			goto out;
1169 	}
1170 
1171 	rc = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "s390/smp:online",
1172 			       smp_cpu_online, smp_cpu_pre_down);
1173 	rc = rc <= 0 ? rc : 0;
1174 out:
1175 	return rc;
1176 }
1177 subsys_initcall(s390_smp_init);
1178