xref: /openbmc/linux/arch/s390/kernel/smp.c (revision 5b4cb650)
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/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/lowcore.h>
49 #include <asm/sclp.h>
50 #include <asm/vdso.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/topology.h>
57 #include "entry.h"
58 
59 enum {
60 	ec_schedule = 0,
61 	ec_call_function_single,
62 	ec_stop_cpu,
63 };
64 
65 enum {
66 	CPU_STATE_STANDBY,
67 	CPU_STATE_CONFIGURED,
68 };
69 
70 static DEFINE_PER_CPU(struct cpu *, cpu_device);
71 
72 struct pcpu {
73 	struct lowcore *lowcore;	/* lowcore page(s) for the cpu */
74 	unsigned long ec_mask;		/* bit mask for ec_xxx functions */
75 	unsigned long ec_clk;		/* sigp timestamp for ec_xxx */
76 	signed char state;		/* physical cpu state */
77 	signed char polarization;	/* physical polarization */
78 	u16 address;			/* physical cpu address */
79 };
80 
81 static u8 boot_core_type;
82 static struct pcpu pcpu_devices[NR_CPUS];
83 
84 unsigned int smp_cpu_mt_shift;
85 EXPORT_SYMBOL(smp_cpu_mt_shift);
86 
87 unsigned int smp_cpu_mtid;
88 EXPORT_SYMBOL(smp_cpu_mtid);
89 
90 #ifdef CONFIG_CRASH_DUMP
91 __vector128 __initdata boot_cpu_vector_save_area[__NUM_VXRS];
92 #endif
93 
94 static unsigned int smp_max_threads __initdata = -1U;
95 
96 static int __init early_nosmt(char *s)
97 {
98 	smp_max_threads = 1;
99 	return 0;
100 }
101 early_param("nosmt", early_nosmt);
102 
103 static int __init early_smt(char *s)
104 {
105 	get_option(&s, &smp_max_threads);
106 	return 0;
107 }
108 early_param("smt", early_smt);
109 
110 /*
111  * The smp_cpu_state_mutex must be held when changing the state or polarization
112  * member of a pcpu data structure within the pcpu_devices arreay.
113  */
114 DEFINE_MUTEX(smp_cpu_state_mutex);
115 
116 /*
117  * Signal processor helper functions.
118  */
119 static inline int __pcpu_sigp_relax(u16 addr, u8 order, unsigned long parm)
120 {
121 	int cc;
122 
123 	while (1) {
124 		cc = __pcpu_sigp(addr, order, parm, NULL);
125 		if (cc != SIGP_CC_BUSY)
126 			return cc;
127 		cpu_relax();
128 	}
129 }
130 
131 static int pcpu_sigp_retry(struct pcpu *pcpu, u8 order, u32 parm)
132 {
133 	int cc, retry;
134 
135 	for (retry = 0; ; retry++) {
136 		cc = __pcpu_sigp(pcpu->address, order, parm, NULL);
137 		if (cc != SIGP_CC_BUSY)
138 			break;
139 		if (retry >= 3)
140 			udelay(10);
141 	}
142 	return cc;
143 }
144 
145 static inline int pcpu_stopped(struct pcpu *pcpu)
146 {
147 	u32 uninitialized_var(status);
148 
149 	if (__pcpu_sigp(pcpu->address, SIGP_SENSE,
150 			0, &status) != SIGP_CC_STATUS_STORED)
151 		return 0;
152 	return !!(status & (SIGP_STATUS_CHECK_STOP|SIGP_STATUS_STOPPED));
153 }
154 
155 static inline int pcpu_running(struct pcpu *pcpu)
156 {
157 	if (__pcpu_sigp(pcpu->address, SIGP_SENSE_RUNNING,
158 			0, NULL) != SIGP_CC_STATUS_STORED)
159 		return 1;
160 	/* Status stored condition code is equivalent to cpu not running. */
161 	return 0;
162 }
163 
164 /*
165  * Find struct pcpu by cpu address.
166  */
167 static struct pcpu *pcpu_find_address(const struct cpumask *mask, u16 address)
168 {
169 	int cpu;
170 
171 	for_each_cpu(cpu, mask)
172 		if (pcpu_devices[cpu].address == address)
173 			return pcpu_devices + cpu;
174 	return NULL;
175 }
176 
177 static void pcpu_ec_call(struct pcpu *pcpu, int ec_bit)
178 {
179 	int order;
180 
181 	if (test_and_set_bit(ec_bit, &pcpu->ec_mask))
182 		return;
183 	order = pcpu_running(pcpu) ? SIGP_EXTERNAL_CALL : SIGP_EMERGENCY_SIGNAL;
184 	pcpu->ec_clk = get_tod_clock_fast();
185 	pcpu_sigp_retry(pcpu, order, 0);
186 }
187 
188 static int pcpu_alloc_lowcore(struct pcpu *pcpu, int cpu)
189 {
190 	unsigned long async_stack, nodat_stack;
191 	struct lowcore *lc;
192 
193 	if (pcpu != &pcpu_devices[0]) {
194 		pcpu->lowcore =	(struct lowcore *)
195 			__get_free_pages(GFP_KERNEL | GFP_DMA, LC_ORDER);
196 		nodat_stack = __get_free_pages(GFP_KERNEL, THREAD_SIZE_ORDER);
197 		if (!pcpu->lowcore || !nodat_stack)
198 			goto out;
199 	} else {
200 		nodat_stack = pcpu->lowcore->nodat_stack - STACK_INIT_OFFSET;
201 	}
202 	async_stack = stack_alloc();
203 	if (!async_stack)
204 		goto out;
205 	lc = pcpu->lowcore;
206 	memcpy(lc, &S390_lowcore, 512);
207 	memset((char *) lc + 512, 0, sizeof(*lc) - 512);
208 	lc->async_stack = async_stack + STACK_INIT_OFFSET;
209 	lc->nodat_stack = nodat_stack + STACK_INIT_OFFSET;
210 	lc->cpu_nr = cpu;
211 	lc->spinlock_lockval = arch_spin_lockval(cpu);
212 	lc->spinlock_index = 0;
213 	lc->br_r1_trampoline = 0x07f1;	/* br %r1 */
214 	if (nmi_alloc_per_cpu(lc))
215 		goto out_async;
216 	if (vdso_alloc_per_cpu(lc))
217 		goto out_mcesa;
218 	lowcore_ptr[cpu] = lc;
219 	pcpu_sigp_retry(pcpu, SIGP_SET_PREFIX, (u32)(unsigned long) lc);
220 	return 0;
221 
222 out_mcesa:
223 	nmi_free_per_cpu(lc);
224 out_async:
225 	stack_free(async_stack);
226 out:
227 	if (pcpu != &pcpu_devices[0]) {
228 		free_pages(nodat_stack, THREAD_SIZE_ORDER);
229 		free_pages((unsigned long) pcpu->lowcore, LC_ORDER);
230 	}
231 	return -ENOMEM;
232 }
233 
234 #ifdef CONFIG_HOTPLUG_CPU
235 
236 static void pcpu_free_lowcore(struct pcpu *pcpu)
237 {
238 	unsigned long async_stack, nodat_stack, lowcore;
239 
240 	nodat_stack = pcpu->lowcore->nodat_stack - STACK_INIT_OFFSET;
241 	async_stack = pcpu->lowcore->async_stack - STACK_INIT_OFFSET;
242 	lowcore = (unsigned long) pcpu->lowcore;
243 
244 	pcpu_sigp_retry(pcpu, SIGP_SET_PREFIX, 0);
245 	lowcore_ptr[pcpu - pcpu_devices] = NULL;
246 	vdso_free_per_cpu(pcpu->lowcore);
247 	nmi_free_per_cpu(pcpu->lowcore);
248 	stack_free(async_stack);
249 	if (pcpu == &pcpu_devices[0])
250 		return;
251 	free_pages(nodat_stack, THREAD_SIZE_ORDER);
252 	free_pages(lowcore, LC_ORDER);
253 }
254 
255 #endif /* CONFIG_HOTPLUG_CPU */
256 
257 static void pcpu_prepare_secondary(struct pcpu *pcpu, int cpu)
258 {
259 	struct lowcore *lc = pcpu->lowcore;
260 
261 	cpumask_set_cpu(cpu, &init_mm.context.cpu_attach_mask);
262 	cpumask_set_cpu(cpu, mm_cpumask(&init_mm));
263 	lc->cpu_nr = cpu;
264 	lc->spinlock_lockval = arch_spin_lockval(cpu);
265 	lc->spinlock_index = 0;
266 	lc->percpu_offset = __per_cpu_offset[cpu];
267 	lc->kernel_asce = S390_lowcore.kernel_asce;
268 	lc->machine_flags = S390_lowcore.machine_flags;
269 	lc->user_timer = lc->system_timer = lc->steal_timer = 0;
270 	__ctl_store(lc->cregs_save_area, 0, 15);
271 	save_access_regs((unsigned int *) lc->access_regs_save_area);
272 	memcpy(lc->stfle_fac_list, S390_lowcore.stfle_fac_list,
273 	       sizeof(lc->stfle_fac_list));
274 	memcpy(lc->alt_stfle_fac_list, S390_lowcore.alt_stfle_fac_list,
275 	       sizeof(lc->alt_stfle_fac_list));
276 	arch_spin_lock_setup(cpu);
277 }
278 
279 static void pcpu_attach_task(struct pcpu *pcpu, struct task_struct *tsk)
280 {
281 	struct lowcore *lc = pcpu->lowcore;
282 
283 	lc->kernel_stack = (unsigned long) task_stack_page(tsk)
284 		+ THREAD_SIZE - STACK_FRAME_OVERHEAD - sizeof(struct pt_regs);
285 	lc->current_task = (unsigned long) tsk;
286 	lc->lpp = LPP_MAGIC;
287 	lc->current_pid = tsk->pid;
288 	lc->user_timer = tsk->thread.user_timer;
289 	lc->guest_timer = tsk->thread.guest_timer;
290 	lc->system_timer = tsk->thread.system_timer;
291 	lc->hardirq_timer = tsk->thread.hardirq_timer;
292 	lc->softirq_timer = tsk->thread.softirq_timer;
293 	lc->steal_timer = 0;
294 }
295 
296 static void pcpu_start_fn(struct pcpu *pcpu, void (*func)(void *), void *data)
297 {
298 	struct lowcore *lc = pcpu->lowcore;
299 
300 	lc->restart_stack = lc->nodat_stack;
301 	lc->restart_fn = (unsigned long) func;
302 	lc->restart_data = (unsigned long) data;
303 	lc->restart_source = -1UL;
304 	pcpu_sigp_retry(pcpu, SIGP_RESTART, 0);
305 }
306 
307 /*
308  * Call function via PSW restart on pcpu and stop the current cpu.
309  */
310 static void __pcpu_delegate(void (*func)(void*), void *data)
311 {
312 	func(data);	/* should not return */
313 }
314 
315 static void __no_sanitize_address pcpu_delegate(struct pcpu *pcpu,
316 						void (*func)(void *),
317 						void *data, unsigned long stack)
318 {
319 	struct lowcore *lc = lowcore_ptr[pcpu - pcpu_devices];
320 	unsigned long source_cpu = stap();
321 
322 	__load_psw_mask(PSW_KERNEL_BITS | PSW_MASK_DAT);
323 	if (pcpu->address == source_cpu)
324 		CALL_ON_STACK(__pcpu_delegate, stack, 2, func, data);
325 	/* Stop target cpu (if func returns this stops the current cpu). */
326 	pcpu_sigp_retry(pcpu, SIGP_STOP, 0);
327 	/* Restart func on the target cpu and stop the current cpu. */
328 	mem_assign_absolute(lc->restart_stack, stack);
329 	mem_assign_absolute(lc->restart_fn, (unsigned long) func);
330 	mem_assign_absolute(lc->restart_data, (unsigned long) data);
331 	mem_assign_absolute(lc->restart_source, source_cpu);
332 	__bpon();
333 	asm volatile(
334 		"0:	sigp	0,%0,%2	# sigp restart to target cpu\n"
335 		"	brc	2,0b	# busy, try again\n"
336 		"1:	sigp	0,%1,%3	# sigp stop to current cpu\n"
337 		"	brc	2,1b	# busy, try again\n"
338 		: : "d" (pcpu->address), "d" (source_cpu),
339 		    "K" (SIGP_RESTART), "K" (SIGP_STOP)
340 		: "0", "1", "cc");
341 	for (;;) ;
342 }
343 
344 /*
345  * Enable additional logical cpus for multi-threading.
346  */
347 static int pcpu_set_smt(unsigned int mtid)
348 {
349 	int cc;
350 
351 	if (smp_cpu_mtid == mtid)
352 		return 0;
353 	cc = __pcpu_sigp(0, SIGP_SET_MULTI_THREADING, mtid, NULL);
354 	if (cc == 0) {
355 		smp_cpu_mtid = mtid;
356 		smp_cpu_mt_shift = 0;
357 		while (smp_cpu_mtid >= (1U << smp_cpu_mt_shift))
358 			smp_cpu_mt_shift++;
359 		pcpu_devices[0].address = stap();
360 	}
361 	return cc;
362 }
363 
364 /*
365  * Call function on an online CPU.
366  */
367 void smp_call_online_cpu(void (*func)(void *), void *data)
368 {
369 	struct pcpu *pcpu;
370 
371 	/* Use the current cpu if it is online. */
372 	pcpu = pcpu_find_address(cpu_online_mask, stap());
373 	if (!pcpu)
374 		/* Use the first online cpu. */
375 		pcpu = pcpu_devices + cpumask_first(cpu_online_mask);
376 	pcpu_delegate(pcpu, func, data, (unsigned long) restart_stack);
377 }
378 
379 /*
380  * Call function on the ipl CPU.
381  */
382 void smp_call_ipl_cpu(void (*func)(void *), void *data)
383 {
384 	pcpu_delegate(&pcpu_devices[0], func, data,
385 		      pcpu_devices->lowcore->nodat_stack);
386 }
387 
388 int smp_find_processor_id(u16 address)
389 {
390 	int cpu;
391 
392 	for_each_present_cpu(cpu)
393 		if (pcpu_devices[cpu].address == address)
394 			return cpu;
395 	return -1;
396 }
397 
398 bool arch_vcpu_is_preempted(int cpu)
399 {
400 	if (test_cpu_flag_of(CIF_ENABLED_WAIT, cpu))
401 		return false;
402 	if (pcpu_running(pcpu_devices + cpu))
403 		return false;
404 	return true;
405 }
406 EXPORT_SYMBOL(arch_vcpu_is_preempted);
407 
408 void smp_yield_cpu(int cpu)
409 {
410 	if (MACHINE_HAS_DIAG9C) {
411 		diag_stat_inc_norecursion(DIAG_STAT_X09C);
412 		asm volatile("diag %0,0,0x9c"
413 			     : : "d" (pcpu_devices[cpu].address));
414 	} else if (MACHINE_HAS_DIAG44) {
415 		diag_stat_inc_norecursion(DIAG_STAT_X044);
416 		asm volatile("diag 0,0,0x44");
417 	}
418 }
419 
420 /*
421  * Send cpus emergency shutdown signal. This gives the cpus the
422  * opportunity to complete outstanding interrupts.
423  */
424 void notrace smp_emergency_stop(void)
425 {
426 	cpumask_t cpumask;
427 	u64 end;
428 	int cpu;
429 
430 	cpumask_copy(&cpumask, cpu_online_mask);
431 	cpumask_clear_cpu(smp_processor_id(), &cpumask);
432 
433 	end = get_tod_clock() + (1000000UL << 12);
434 	for_each_cpu(cpu, &cpumask) {
435 		struct pcpu *pcpu = pcpu_devices + cpu;
436 		set_bit(ec_stop_cpu, &pcpu->ec_mask);
437 		while (__pcpu_sigp(pcpu->address, SIGP_EMERGENCY_SIGNAL,
438 				   0, NULL) == SIGP_CC_BUSY &&
439 		       get_tod_clock() < end)
440 			cpu_relax();
441 	}
442 	while (get_tod_clock() < end) {
443 		for_each_cpu(cpu, &cpumask)
444 			if (pcpu_stopped(pcpu_devices + cpu))
445 				cpumask_clear_cpu(cpu, &cpumask);
446 		if (cpumask_empty(&cpumask))
447 			break;
448 		cpu_relax();
449 	}
450 }
451 NOKPROBE_SYMBOL(smp_emergency_stop);
452 
453 /*
454  * Stop all cpus but the current one.
455  */
456 void smp_send_stop(void)
457 {
458 	int cpu;
459 
460 	/* Disable all interrupts/machine checks */
461 	__load_psw_mask(PSW_KERNEL_BITS | PSW_MASK_DAT);
462 	trace_hardirqs_off();
463 
464 	debug_set_critical();
465 
466 	if (oops_in_progress)
467 		smp_emergency_stop();
468 
469 	/* stop all processors */
470 	for_each_online_cpu(cpu) {
471 		if (cpu == smp_processor_id())
472 			continue;
473 		pcpu_sigp_retry(pcpu_devices + cpu, SIGP_STOP, 0);
474 		while (!pcpu_stopped(pcpu_devices + cpu))
475 			cpu_relax();
476 	}
477 }
478 
479 /*
480  * This is the main routine where commands issued by other
481  * cpus are handled.
482  */
483 static void smp_handle_ext_call(void)
484 {
485 	unsigned long bits;
486 
487 	/* handle bit signal external calls */
488 	bits = xchg(&pcpu_devices[smp_processor_id()].ec_mask, 0);
489 	if (test_bit(ec_stop_cpu, &bits))
490 		smp_stop_cpu();
491 	if (test_bit(ec_schedule, &bits))
492 		scheduler_ipi();
493 	if (test_bit(ec_call_function_single, &bits))
494 		generic_smp_call_function_single_interrupt();
495 }
496 
497 static void do_ext_call_interrupt(struct ext_code ext_code,
498 				  unsigned int param32, unsigned long param64)
499 {
500 	inc_irq_stat(ext_code.code == 0x1202 ? IRQEXT_EXC : IRQEXT_EMS);
501 	smp_handle_ext_call();
502 }
503 
504 void arch_send_call_function_ipi_mask(const struct cpumask *mask)
505 {
506 	int cpu;
507 
508 	for_each_cpu(cpu, mask)
509 		pcpu_ec_call(pcpu_devices + cpu, ec_call_function_single);
510 }
511 
512 void arch_send_call_function_single_ipi(int cpu)
513 {
514 	pcpu_ec_call(pcpu_devices + cpu, ec_call_function_single);
515 }
516 
517 /*
518  * this function sends a 'reschedule' IPI to another CPU.
519  * it goes straight through and wastes no time serializing
520  * anything. Worst case is that we lose a reschedule ...
521  */
522 void smp_send_reschedule(int cpu)
523 {
524 	pcpu_ec_call(pcpu_devices + cpu, ec_schedule);
525 }
526 
527 /*
528  * parameter area for the set/clear control bit callbacks
529  */
530 struct ec_creg_mask_parms {
531 	unsigned long orval;
532 	unsigned long andval;
533 	int cr;
534 };
535 
536 /*
537  * callback for setting/clearing control bits
538  */
539 static void smp_ctl_bit_callback(void *info)
540 {
541 	struct ec_creg_mask_parms *pp = info;
542 	unsigned long cregs[16];
543 
544 	__ctl_store(cregs, 0, 15);
545 	cregs[pp->cr] = (cregs[pp->cr] & pp->andval) | pp->orval;
546 	__ctl_load(cregs, 0, 15);
547 }
548 
549 /*
550  * Set a bit in a control register of all cpus
551  */
552 void smp_ctl_set_bit(int cr, int bit)
553 {
554 	struct ec_creg_mask_parms parms = { 1UL << bit, -1UL, cr };
555 
556 	on_each_cpu(smp_ctl_bit_callback, &parms, 1);
557 }
558 EXPORT_SYMBOL(smp_ctl_set_bit);
559 
560 /*
561  * Clear a bit in a control register of all cpus
562  */
563 void smp_ctl_clear_bit(int cr, int bit)
564 {
565 	struct ec_creg_mask_parms parms = { 0, ~(1UL << bit), cr };
566 
567 	on_each_cpu(smp_ctl_bit_callback, &parms, 1);
568 }
569 EXPORT_SYMBOL(smp_ctl_clear_bit);
570 
571 #ifdef CONFIG_CRASH_DUMP
572 
573 int smp_store_status(int cpu)
574 {
575 	struct pcpu *pcpu = pcpu_devices + cpu;
576 	unsigned long pa;
577 
578 	pa = __pa(&pcpu->lowcore->floating_pt_save_area);
579 	if (__pcpu_sigp_relax(pcpu->address, SIGP_STORE_STATUS_AT_ADDRESS,
580 			      pa) != SIGP_CC_ORDER_CODE_ACCEPTED)
581 		return -EIO;
582 	if (!MACHINE_HAS_VX && !MACHINE_HAS_GS)
583 		return 0;
584 	pa = __pa(pcpu->lowcore->mcesad & MCESA_ORIGIN_MASK);
585 	if (MACHINE_HAS_GS)
586 		pa |= pcpu->lowcore->mcesad & MCESA_LC_MASK;
587 	if (__pcpu_sigp_relax(pcpu->address, SIGP_STORE_ADDITIONAL_STATUS,
588 			      pa) != SIGP_CC_ORDER_CODE_ACCEPTED)
589 		return -EIO;
590 	return 0;
591 }
592 
593 /*
594  * Collect CPU state of the previous, crashed system.
595  * There are four cases:
596  * 1) standard zfcp dump
597  *    condition: OLDMEM_BASE == NULL && ipl_info.type == IPL_TYPE_FCP_DUMP
598  *    The state for all CPUs except the boot CPU needs to be collected
599  *    with sigp stop-and-store-status. The boot CPU state is located in
600  *    the absolute lowcore of the memory stored in the HSA. The zcore code
601  *    will copy the boot CPU state from the HSA.
602  * 2) stand-alone kdump for SCSI (zfcp dump with swapped memory)
603  *    condition: OLDMEM_BASE != NULL && ipl_info.type == IPL_TYPE_FCP_DUMP
604  *    The state for all CPUs except the boot CPU needs to be collected
605  *    with sigp stop-and-store-status. The firmware or the boot-loader
606  *    stored the registers of the boot CPU in the absolute lowcore in the
607  *    memory of the old system.
608  * 3) kdump and the old kernel did not store the CPU state,
609  *    or stand-alone kdump for DASD
610  *    condition: OLDMEM_BASE != NULL && !is_kdump_kernel()
611  *    The state for all CPUs except the boot CPU needs to be collected
612  *    with sigp stop-and-store-status. The kexec code or the boot-loader
613  *    stored the registers of the boot CPU in the memory of the old system.
614  * 4) kdump and the old kernel stored the CPU state
615  *    condition: OLDMEM_BASE != NULL && is_kdump_kernel()
616  *    This case does not exist for s390 anymore, setup_arch explicitly
617  *    deactivates the elfcorehdr= kernel parameter
618  */
619 static __init void smp_save_cpu_vxrs(struct save_area *sa, u16 addr,
620 				     bool is_boot_cpu, unsigned long page)
621 {
622 	__vector128 *vxrs = (__vector128 *) page;
623 
624 	if (is_boot_cpu)
625 		vxrs = boot_cpu_vector_save_area;
626 	else
627 		__pcpu_sigp_relax(addr, SIGP_STORE_ADDITIONAL_STATUS, page);
628 	save_area_add_vxrs(sa, vxrs);
629 }
630 
631 static __init void smp_save_cpu_regs(struct save_area *sa, u16 addr,
632 				     bool is_boot_cpu, unsigned long page)
633 {
634 	void *regs = (void *) page;
635 
636 	if (is_boot_cpu)
637 		copy_oldmem_kernel(regs, (void *) __LC_FPREGS_SAVE_AREA, 512);
638 	else
639 		__pcpu_sigp_relax(addr, SIGP_STORE_STATUS_AT_ADDRESS, page);
640 	save_area_add_regs(sa, regs);
641 }
642 
643 void __init smp_save_dump_cpus(void)
644 {
645 	int addr, boot_cpu_addr, max_cpu_addr;
646 	struct save_area *sa;
647 	unsigned long page;
648 	bool is_boot_cpu;
649 
650 	if (!(OLDMEM_BASE || ipl_info.type == IPL_TYPE_FCP_DUMP))
651 		/* No previous system present, normal boot. */
652 		return;
653 	/* Allocate a page as dumping area for the store status sigps */
654 	page = memblock_alloc_base(PAGE_SIZE, PAGE_SIZE, 1UL << 31);
655 	/* Set multi-threading state to the previous system. */
656 	pcpu_set_smt(sclp.mtid_prev);
657 	boot_cpu_addr = stap();
658 	max_cpu_addr = SCLP_MAX_CORES << sclp.mtid_prev;
659 	for (addr = 0; addr <= max_cpu_addr; addr++) {
660 		if (__pcpu_sigp_relax(addr, SIGP_SENSE, 0) ==
661 		    SIGP_CC_NOT_OPERATIONAL)
662 			continue;
663 		is_boot_cpu = (addr == boot_cpu_addr);
664 		/* Allocate save area */
665 		sa = save_area_alloc(is_boot_cpu);
666 		if (!sa)
667 			panic("could not allocate memory for save area\n");
668 		if (MACHINE_HAS_VX)
669 			/* Get the vector registers */
670 			smp_save_cpu_vxrs(sa, addr, is_boot_cpu, page);
671 		/*
672 		 * For a zfcp dump OLDMEM_BASE == NULL and the registers
673 		 * of the boot CPU are stored in the HSA. To retrieve
674 		 * these registers an SCLP request is required which is
675 		 * done by drivers/s390/char/zcore.c:init_cpu_info()
676 		 */
677 		if (!is_boot_cpu || OLDMEM_BASE)
678 			/* Get the CPU registers */
679 			smp_save_cpu_regs(sa, addr, is_boot_cpu, page);
680 	}
681 	memblock_free(page, PAGE_SIZE);
682 	diag308_reset();
683 	pcpu_set_smt(0);
684 }
685 #endif /* CONFIG_CRASH_DUMP */
686 
687 void smp_cpu_set_polarization(int cpu, int val)
688 {
689 	pcpu_devices[cpu].polarization = val;
690 }
691 
692 int smp_cpu_get_polarization(int cpu)
693 {
694 	return pcpu_devices[cpu].polarization;
695 }
696 
697 static void __ref smp_get_core_info(struct sclp_core_info *info, int early)
698 {
699 	static int use_sigp_detection;
700 	int address;
701 
702 	if (use_sigp_detection || sclp_get_core_info(info, early)) {
703 		use_sigp_detection = 1;
704 		for (address = 0;
705 		     address < (SCLP_MAX_CORES << smp_cpu_mt_shift);
706 		     address += (1U << smp_cpu_mt_shift)) {
707 			if (__pcpu_sigp_relax(address, SIGP_SENSE, 0) ==
708 			    SIGP_CC_NOT_OPERATIONAL)
709 				continue;
710 			info->core[info->configured].core_id =
711 				address >> smp_cpu_mt_shift;
712 			info->configured++;
713 		}
714 		info->combined = info->configured;
715 	}
716 }
717 
718 static int smp_add_present_cpu(int cpu);
719 
720 static int __smp_rescan_cpus(struct sclp_core_info *info, int sysfs_add)
721 {
722 	struct pcpu *pcpu;
723 	cpumask_t avail;
724 	int cpu, nr, i, j;
725 	u16 address;
726 
727 	nr = 0;
728 	cpumask_xor(&avail, cpu_possible_mask, cpu_present_mask);
729 	cpu = cpumask_first(&avail);
730 	for (i = 0; (i < info->combined) && (cpu < nr_cpu_ids); i++) {
731 		if (sclp.has_core_type && info->core[i].type != boot_core_type)
732 			continue;
733 		address = info->core[i].core_id << smp_cpu_mt_shift;
734 		for (j = 0; j <= smp_cpu_mtid; j++) {
735 			if (pcpu_find_address(cpu_present_mask, address + j))
736 				continue;
737 			pcpu = pcpu_devices + cpu;
738 			pcpu->address = address + j;
739 			pcpu->state =
740 				(cpu >= info->configured*(smp_cpu_mtid + 1)) ?
741 				CPU_STATE_STANDBY : CPU_STATE_CONFIGURED;
742 			smp_cpu_set_polarization(cpu, POLARIZATION_UNKNOWN);
743 			set_cpu_present(cpu, true);
744 			if (sysfs_add && smp_add_present_cpu(cpu) != 0)
745 				set_cpu_present(cpu, false);
746 			else
747 				nr++;
748 			cpu = cpumask_next(cpu, &avail);
749 			if (cpu >= nr_cpu_ids)
750 				break;
751 		}
752 	}
753 	return nr;
754 }
755 
756 void __init smp_detect_cpus(void)
757 {
758 	unsigned int cpu, mtid, c_cpus, s_cpus;
759 	struct sclp_core_info *info;
760 	u16 address;
761 
762 	/* Get CPU information */
763 	info = memblock_alloc(sizeof(*info), 8);
764 	smp_get_core_info(info, 1);
765 	/* Find boot CPU type */
766 	if (sclp.has_core_type) {
767 		address = stap();
768 		for (cpu = 0; cpu < info->combined; cpu++)
769 			if (info->core[cpu].core_id == address) {
770 				/* The boot cpu dictates the cpu type. */
771 				boot_core_type = info->core[cpu].type;
772 				break;
773 			}
774 		if (cpu >= info->combined)
775 			panic("Could not find boot CPU type");
776 	}
777 
778 	/* Set multi-threading state for the current system */
779 	mtid = boot_core_type ? sclp.mtid : sclp.mtid_cp;
780 	mtid = (mtid < smp_max_threads) ? mtid : smp_max_threads - 1;
781 	pcpu_set_smt(mtid);
782 
783 	/* Print number of CPUs */
784 	c_cpus = s_cpus = 0;
785 	for (cpu = 0; cpu < info->combined; cpu++) {
786 		if (sclp.has_core_type &&
787 		    info->core[cpu].type != boot_core_type)
788 			continue;
789 		if (cpu < info->configured)
790 			c_cpus += smp_cpu_mtid + 1;
791 		else
792 			s_cpus += smp_cpu_mtid + 1;
793 	}
794 	pr_info("%d configured CPUs, %d standby CPUs\n", c_cpus, s_cpus);
795 
796 	/* Add CPUs present at boot */
797 	get_online_cpus();
798 	__smp_rescan_cpus(info, 0);
799 	put_online_cpus();
800 	memblock_free_early((unsigned long)info, sizeof(*info));
801 }
802 
803 static void smp_init_secondary(void)
804 {
805 	int cpu = smp_processor_id();
806 
807 	S390_lowcore.last_update_clock = get_tod_clock();
808 	restore_access_regs(S390_lowcore.access_regs_save_area);
809 	cpu_init();
810 	preempt_disable();
811 	init_cpu_timer();
812 	vtime_init();
813 	pfault_init();
814 	notify_cpu_starting(smp_processor_id());
815 	if (topology_cpu_dedicated(cpu))
816 		set_cpu_flag(CIF_DEDICATED_CPU);
817 	else
818 		clear_cpu_flag(CIF_DEDICATED_CPU);
819 	set_cpu_online(smp_processor_id(), true);
820 	inc_irq_stat(CPU_RST);
821 	local_irq_enable();
822 	cpu_startup_entry(CPUHP_AP_ONLINE_IDLE);
823 }
824 
825 /*
826  *	Activate a secondary processor.
827  */
828 static void __no_sanitize_address smp_start_secondary(void *cpuvoid)
829 {
830 	S390_lowcore.restart_stack = (unsigned long) restart_stack;
831 	S390_lowcore.restart_fn = (unsigned long) do_restart;
832 	S390_lowcore.restart_data = 0;
833 	S390_lowcore.restart_source = -1UL;
834 	__ctl_load(S390_lowcore.cregs_save_area, 0, 15);
835 	__load_psw_mask(PSW_KERNEL_BITS | PSW_MASK_DAT);
836 	CALL_ON_STACK(smp_init_secondary, S390_lowcore.kernel_stack, 0);
837 }
838 
839 /* Upping and downing of CPUs */
840 int __cpu_up(unsigned int cpu, struct task_struct *tidle)
841 {
842 	struct pcpu *pcpu;
843 	int base, i, rc;
844 
845 	pcpu = pcpu_devices + cpu;
846 	if (pcpu->state != CPU_STATE_CONFIGURED)
847 		return -EIO;
848 	base = smp_get_base_cpu(cpu);
849 	for (i = 0; i <= smp_cpu_mtid; i++) {
850 		if (base + i < nr_cpu_ids)
851 			if (cpu_online(base + i))
852 				break;
853 	}
854 	/*
855 	 * If this is the first CPU of the core to get online
856 	 * do an initial CPU reset.
857 	 */
858 	if (i > smp_cpu_mtid &&
859 	    pcpu_sigp_retry(pcpu_devices + base, SIGP_INITIAL_CPU_RESET, 0) !=
860 	    SIGP_CC_ORDER_CODE_ACCEPTED)
861 		return -EIO;
862 
863 	rc = pcpu_alloc_lowcore(pcpu, cpu);
864 	if (rc)
865 		return rc;
866 	pcpu_prepare_secondary(pcpu, cpu);
867 	pcpu_attach_task(pcpu, tidle);
868 	pcpu_start_fn(pcpu, smp_start_secondary, NULL);
869 	/* Wait until cpu puts itself in the online & active maps */
870 	while (!cpu_online(cpu))
871 		cpu_relax();
872 	return 0;
873 }
874 
875 static unsigned int setup_possible_cpus __initdata;
876 
877 static int __init _setup_possible_cpus(char *s)
878 {
879 	get_option(&s, &setup_possible_cpus);
880 	return 0;
881 }
882 early_param("possible_cpus", _setup_possible_cpus);
883 
884 #ifdef CONFIG_HOTPLUG_CPU
885 
886 int __cpu_disable(void)
887 {
888 	unsigned long cregs[16];
889 
890 	/* Handle possible pending IPIs */
891 	smp_handle_ext_call();
892 	set_cpu_online(smp_processor_id(), false);
893 	/* Disable pseudo page faults on this cpu. */
894 	pfault_fini();
895 	/* Disable interrupt sources via control register. */
896 	__ctl_store(cregs, 0, 15);
897 	cregs[0]  &= ~0x0000ee70UL;	/* disable all external interrupts */
898 	cregs[6]  &= ~0xff000000UL;	/* disable all I/O interrupts */
899 	cregs[14] &= ~0x1f000000UL;	/* disable most machine checks */
900 	__ctl_load(cregs, 0, 15);
901 	clear_cpu_flag(CIF_NOHZ_DELAY);
902 	return 0;
903 }
904 
905 void __cpu_die(unsigned int cpu)
906 {
907 	struct pcpu *pcpu;
908 
909 	/* Wait until target cpu is down */
910 	pcpu = pcpu_devices + cpu;
911 	while (!pcpu_stopped(pcpu))
912 		cpu_relax();
913 	pcpu_free_lowcore(pcpu);
914 	cpumask_clear_cpu(cpu, mm_cpumask(&init_mm));
915 	cpumask_clear_cpu(cpu, &init_mm.context.cpu_attach_mask);
916 }
917 
918 void __noreturn cpu_die(void)
919 {
920 	idle_task_exit();
921 	__bpon();
922 	pcpu_sigp_retry(pcpu_devices + smp_processor_id(), SIGP_STOP, 0);
923 	for (;;) ;
924 }
925 
926 #endif /* CONFIG_HOTPLUG_CPU */
927 
928 void __init smp_fill_possible_mask(void)
929 {
930 	unsigned int possible, sclp_max, cpu;
931 
932 	sclp_max = max(sclp.mtid, sclp.mtid_cp) + 1;
933 	sclp_max = min(smp_max_threads, sclp_max);
934 	sclp_max = (sclp.max_cores * sclp_max) ?: nr_cpu_ids;
935 	possible = setup_possible_cpus ?: nr_cpu_ids;
936 	possible = min(possible, sclp_max);
937 	for (cpu = 0; cpu < possible && cpu < nr_cpu_ids; cpu++)
938 		set_cpu_possible(cpu, true);
939 }
940 
941 void __init smp_prepare_cpus(unsigned int max_cpus)
942 {
943 	/* request the 0x1201 emergency signal external interrupt */
944 	if (register_external_irq(EXT_IRQ_EMERGENCY_SIG, do_ext_call_interrupt))
945 		panic("Couldn't request external interrupt 0x1201");
946 	/* request the 0x1202 external call external interrupt */
947 	if (register_external_irq(EXT_IRQ_EXTERNAL_CALL, do_ext_call_interrupt))
948 		panic("Couldn't request external interrupt 0x1202");
949 }
950 
951 void __init smp_prepare_boot_cpu(void)
952 {
953 	struct pcpu *pcpu = pcpu_devices;
954 
955 	WARN_ON(!cpu_present(0) || !cpu_online(0));
956 	pcpu->state = CPU_STATE_CONFIGURED;
957 	pcpu->lowcore = (struct lowcore *)(unsigned long) store_prefix();
958 	S390_lowcore.percpu_offset = __per_cpu_offset[0];
959 	smp_cpu_set_polarization(0, POLARIZATION_UNKNOWN);
960 }
961 
962 void __init smp_cpus_done(unsigned int max_cpus)
963 {
964 }
965 
966 void __init smp_setup_processor_id(void)
967 {
968 	pcpu_devices[0].address = stap();
969 	S390_lowcore.cpu_nr = 0;
970 	S390_lowcore.spinlock_lockval = arch_spin_lockval(0);
971 	S390_lowcore.spinlock_index = 0;
972 }
973 
974 /*
975  * the frequency of the profiling timer can be changed
976  * by writing a multiplier value into /proc/profile.
977  *
978  * usually you want to run this on all CPUs ;)
979  */
980 int setup_profiling_timer(unsigned int multiplier)
981 {
982 	return 0;
983 }
984 
985 #ifdef CONFIG_HOTPLUG_CPU
986 static ssize_t cpu_configure_show(struct device *dev,
987 				  struct device_attribute *attr, char *buf)
988 {
989 	ssize_t count;
990 
991 	mutex_lock(&smp_cpu_state_mutex);
992 	count = sprintf(buf, "%d\n", pcpu_devices[dev->id].state);
993 	mutex_unlock(&smp_cpu_state_mutex);
994 	return count;
995 }
996 
997 static ssize_t cpu_configure_store(struct device *dev,
998 				   struct device_attribute *attr,
999 				   const char *buf, size_t count)
1000 {
1001 	struct pcpu *pcpu;
1002 	int cpu, val, rc, i;
1003 	char delim;
1004 
1005 	if (sscanf(buf, "%d %c", &val, &delim) != 1)
1006 		return -EINVAL;
1007 	if (val != 0 && val != 1)
1008 		return -EINVAL;
1009 	get_online_cpus();
1010 	mutex_lock(&smp_cpu_state_mutex);
1011 	rc = -EBUSY;
1012 	/* disallow configuration changes of online cpus and cpu 0 */
1013 	cpu = dev->id;
1014 	cpu = smp_get_base_cpu(cpu);
1015 	if (cpu == 0)
1016 		goto out;
1017 	for (i = 0; i <= smp_cpu_mtid; i++)
1018 		if (cpu_online(cpu + i))
1019 			goto out;
1020 	pcpu = pcpu_devices + cpu;
1021 	rc = 0;
1022 	switch (val) {
1023 	case 0:
1024 		if (pcpu->state != CPU_STATE_CONFIGURED)
1025 			break;
1026 		rc = sclp_core_deconfigure(pcpu->address >> smp_cpu_mt_shift);
1027 		if (rc)
1028 			break;
1029 		for (i = 0; i <= smp_cpu_mtid; i++) {
1030 			if (cpu + i >= nr_cpu_ids || !cpu_present(cpu + i))
1031 				continue;
1032 			pcpu[i].state = CPU_STATE_STANDBY;
1033 			smp_cpu_set_polarization(cpu + i,
1034 						 POLARIZATION_UNKNOWN);
1035 		}
1036 		topology_expect_change();
1037 		break;
1038 	case 1:
1039 		if (pcpu->state != CPU_STATE_STANDBY)
1040 			break;
1041 		rc = sclp_core_configure(pcpu->address >> smp_cpu_mt_shift);
1042 		if (rc)
1043 			break;
1044 		for (i = 0; i <= smp_cpu_mtid; i++) {
1045 			if (cpu + i >= nr_cpu_ids || !cpu_present(cpu + i))
1046 				continue;
1047 			pcpu[i].state = CPU_STATE_CONFIGURED;
1048 			smp_cpu_set_polarization(cpu + i,
1049 						 POLARIZATION_UNKNOWN);
1050 		}
1051 		topology_expect_change();
1052 		break;
1053 	default:
1054 		break;
1055 	}
1056 out:
1057 	mutex_unlock(&smp_cpu_state_mutex);
1058 	put_online_cpus();
1059 	return rc ? rc : count;
1060 }
1061 static DEVICE_ATTR(configure, 0644, cpu_configure_show, cpu_configure_store);
1062 #endif /* CONFIG_HOTPLUG_CPU */
1063 
1064 static ssize_t show_cpu_address(struct device *dev,
1065 				struct device_attribute *attr, char *buf)
1066 {
1067 	return sprintf(buf, "%d\n", pcpu_devices[dev->id].address);
1068 }
1069 static DEVICE_ATTR(address, 0444, show_cpu_address, NULL);
1070 
1071 static struct attribute *cpu_common_attrs[] = {
1072 #ifdef CONFIG_HOTPLUG_CPU
1073 	&dev_attr_configure.attr,
1074 #endif
1075 	&dev_attr_address.attr,
1076 	NULL,
1077 };
1078 
1079 static struct attribute_group cpu_common_attr_group = {
1080 	.attrs = cpu_common_attrs,
1081 };
1082 
1083 static struct attribute *cpu_online_attrs[] = {
1084 	&dev_attr_idle_count.attr,
1085 	&dev_attr_idle_time_us.attr,
1086 	NULL,
1087 };
1088 
1089 static struct attribute_group cpu_online_attr_group = {
1090 	.attrs = cpu_online_attrs,
1091 };
1092 
1093 static int smp_cpu_online(unsigned int cpu)
1094 {
1095 	struct device *s = &per_cpu(cpu_device, cpu)->dev;
1096 
1097 	return sysfs_create_group(&s->kobj, &cpu_online_attr_group);
1098 }
1099 static int smp_cpu_pre_down(unsigned int cpu)
1100 {
1101 	struct device *s = &per_cpu(cpu_device, cpu)->dev;
1102 
1103 	sysfs_remove_group(&s->kobj, &cpu_online_attr_group);
1104 	return 0;
1105 }
1106 
1107 static int smp_add_present_cpu(int cpu)
1108 {
1109 	struct device *s;
1110 	struct cpu *c;
1111 	int rc;
1112 
1113 	c = kzalloc(sizeof(*c), GFP_KERNEL);
1114 	if (!c)
1115 		return -ENOMEM;
1116 	per_cpu(cpu_device, cpu) = c;
1117 	s = &c->dev;
1118 	c->hotpluggable = 1;
1119 	rc = register_cpu(c, cpu);
1120 	if (rc)
1121 		goto out;
1122 	rc = sysfs_create_group(&s->kobj, &cpu_common_attr_group);
1123 	if (rc)
1124 		goto out_cpu;
1125 	rc = topology_cpu_init(c);
1126 	if (rc)
1127 		goto out_topology;
1128 	return 0;
1129 
1130 out_topology:
1131 	sysfs_remove_group(&s->kobj, &cpu_common_attr_group);
1132 out_cpu:
1133 #ifdef CONFIG_HOTPLUG_CPU
1134 	unregister_cpu(c);
1135 #endif
1136 out:
1137 	return rc;
1138 }
1139 
1140 #ifdef CONFIG_HOTPLUG_CPU
1141 
1142 int __ref smp_rescan_cpus(void)
1143 {
1144 	struct sclp_core_info *info;
1145 	int nr;
1146 
1147 	info = kzalloc(sizeof(*info), GFP_KERNEL);
1148 	if (!info)
1149 		return -ENOMEM;
1150 	smp_get_core_info(info, 0);
1151 	get_online_cpus();
1152 	mutex_lock(&smp_cpu_state_mutex);
1153 	nr = __smp_rescan_cpus(info, 1);
1154 	mutex_unlock(&smp_cpu_state_mutex);
1155 	put_online_cpus();
1156 	kfree(info);
1157 	if (nr)
1158 		topology_schedule_update();
1159 	return 0;
1160 }
1161 
1162 static ssize_t __ref rescan_store(struct device *dev,
1163 				  struct device_attribute *attr,
1164 				  const char *buf,
1165 				  size_t count)
1166 {
1167 	int rc;
1168 
1169 	rc = smp_rescan_cpus();
1170 	return rc ? rc : count;
1171 }
1172 static DEVICE_ATTR_WO(rescan);
1173 #endif /* CONFIG_HOTPLUG_CPU */
1174 
1175 static int __init s390_smp_init(void)
1176 {
1177 	int cpu, rc = 0;
1178 
1179 #ifdef CONFIG_HOTPLUG_CPU
1180 	rc = device_create_file(cpu_subsys.dev_root, &dev_attr_rescan);
1181 	if (rc)
1182 		return rc;
1183 #endif
1184 	for_each_present_cpu(cpu) {
1185 		rc = smp_add_present_cpu(cpu);
1186 		if (rc)
1187 			goto out;
1188 	}
1189 
1190 	rc = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "s390/smp:online",
1191 			       smp_cpu_online, smp_cpu_pre_down);
1192 	rc = rc <= 0 ? rc : 0;
1193 out:
1194 	return rc;
1195 }
1196 subsys_initcall(s390_smp_init);
1197