xref: /openbmc/linux/arch/s390/kernel/smp.c (revision c819e2cf)
1 /*
2  *  SMP related functions
3  *
4  *    Copyright IBM Corp. 1999, 2012
5  *    Author(s): Denis Joseph Barrow,
6  *		 Martin Schwidefsky <schwidefsky@de.ibm.com>,
7  *		 Heiko Carstens <heiko.carstens@de.ibm.com>,
8  *
9  *  based on other smp stuff by
10  *    (c) 1995 Alan Cox, CymruNET Ltd  <alan@cymru.net>
11  *    (c) 1998 Ingo Molnar
12  *
13  * The code outside of smp.c uses logical cpu numbers, only smp.c does
14  * the translation of logical to physical cpu ids. All new code that
15  * operates on physical cpu numbers needs to go into smp.c.
16  */
17 
18 #define KMSG_COMPONENT "cpu"
19 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
20 
21 #include <linux/workqueue.h>
22 #include <linux/module.h>
23 #include <linux/init.h>
24 #include <linux/mm.h>
25 #include <linux/err.h>
26 #include <linux/spinlock.h>
27 #include <linux/kernel_stat.h>
28 #include <linux/delay.h>
29 #include <linux/interrupt.h>
30 #include <linux/irqflags.h>
31 #include <linux/cpu.h>
32 #include <linux/slab.h>
33 #include <linux/crash_dump.h>
34 #include <asm/asm-offsets.h>
35 #include <asm/switch_to.h>
36 #include <asm/facility.h>
37 #include <asm/ipl.h>
38 #include <asm/setup.h>
39 #include <asm/irq.h>
40 #include <asm/tlbflush.h>
41 #include <asm/vtimer.h>
42 #include <asm/lowcore.h>
43 #include <asm/sclp.h>
44 #include <asm/vdso.h>
45 #include <asm/debug.h>
46 #include <asm/os_info.h>
47 #include <asm/sigp.h>
48 #include <asm/idle.h>
49 #include "entry.h"
50 
51 enum {
52 	ec_schedule = 0,
53 	ec_call_function_single,
54 	ec_stop_cpu,
55 };
56 
57 enum {
58 	CPU_STATE_STANDBY,
59 	CPU_STATE_CONFIGURED,
60 };
61 
62 struct pcpu {
63 	struct cpu *cpu;
64 	struct _lowcore *lowcore;	/* lowcore page(s) for the cpu */
65 	unsigned long async_stack;	/* async stack for the cpu */
66 	unsigned long panic_stack;	/* panic stack for the cpu */
67 	unsigned long ec_mask;		/* bit mask for ec_xxx functions */
68 	int state;			/* physical cpu state */
69 	int polarization;		/* physical polarization */
70 	u16 address;			/* physical cpu address */
71 };
72 
73 static u8 boot_cpu_type;
74 static u16 boot_cpu_address;
75 static struct pcpu pcpu_devices[NR_CPUS];
76 
77 /*
78  * The smp_cpu_state_mutex must be held when changing the state or polarization
79  * member of a pcpu data structure within the pcpu_devices arreay.
80  */
81 DEFINE_MUTEX(smp_cpu_state_mutex);
82 
83 /*
84  * Signal processor helper functions.
85  */
86 static inline int __pcpu_sigp_relax(u16 addr, u8 order, unsigned long parm,
87 				    u32 *status)
88 {
89 	int cc;
90 
91 	while (1) {
92 		cc = __pcpu_sigp(addr, order, parm, NULL);
93 		if (cc != SIGP_CC_BUSY)
94 			return cc;
95 		cpu_relax();
96 	}
97 }
98 
99 static int pcpu_sigp_retry(struct pcpu *pcpu, u8 order, u32 parm)
100 {
101 	int cc, retry;
102 
103 	for (retry = 0; ; retry++) {
104 		cc = __pcpu_sigp(pcpu->address, order, parm, NULL);
105 		if (cc != SIGP_CC_BUSY)
106 			break;
107 		if (retry >= 3)
108 			udelay(10);
109 	}
110 	return cc;
111 }
112 
113 static inline int pcpu_stopped(struct pcpu *pcpu)
114 {
115 	u32 uninitialized_var(status);
116 
117 	if (__pcpu_sigp(pcpu->address, SIGP_SENSE,
118 			0, &status) != SIGP_CC_STATUS_STORED)
119 		return 0;
120 	return !!(status & (SIGP_STATUS_CHECK_STOP|SIGP_STATUS_STOPPED));
121 }
122 
123 static inline int pcpu_running(struct pcpu *pcpu)
124 {
125 	if (__pcpu_sigp(pcpu->address, SIGP_SENSE_RUNNING,
126 			0, NULL) != SIGP_CC_STATUS_STORED)
127 		return 1;
128 	/* Status stored condition code is equivalent to cpu not running. */
129 	return 0;
130 }
131 
132 /*
133  * Find struct pcpu by cpu address.
134  */
135 static struct pcpu *pcpu_find_address(const struct cpumask *mask, int address)
136 {
137 	int cpu;
138 
139 	for_each_cpu(cpu, mask)
140 		if (pcpu_devices[cpu].address == address)
141 			return pcpu_devices + cpu;
142 	return NULL;
143 }
144 
145 static void pcpu_ec_call(struct pcpu *pcpu, int ec_bit)
146 {
147 	int order;
148 
149 	if (test_and_set_bit(ec_bit, &pcpu->ec_mask))
150 		return;
151 	order = pcpu_running(pcpu) ? SIGP_EXTERNAL_CALL : SIGP_EMERGENCY_SIGNAL;
152 	pcpu_sigp_retry(pcpu, order, 0);
153 }
154 
155 static int pcpu_alloc_lowcore(struct pcpu *pcpu, int cpu)
156 {
157 	struct _lowcore *lc;
158 
159 	if (pcpu != &pcpu_devices[0]) {
160 		pcpu->lowcore =	(struct _lowcore *)
161 			__get_free_pages(GFP_KERNEL | GFP_DMA, LC_ORDER);
162 		pcpu->async_stack = __get_free_pages(GFP_KERNEL, ASYNC_ORDER);
163 		pcpu->panic_stack = __get_free_page(GFP_KERNEL);
164 		if (!pcpu->lowcore || !pcpu->panic_stack || !pcpu->async_stack)
165 			goto out;
166 	}
167 	lc = pcpu->lowcore;
168 	memcpy(lc, &S390_lowcore, 512);
169 	memset((char *) lc + 512, 0, sizeof(*lc) - 512);
170 	lc->async_stack = pcpu->async_stack + ASYNC_SIZE
171 		- STACK_FRAME_OVERHEAD - sizeof(struct pt_regs);
172 	lc->panic_stack = pcpu->panic_stack + PAGE_SIZE
173 		- STACK_FRAME_OVERHEAD - sizeof(struct pt_regs);
174 	lc->cpu_nr = cpu;
175 	lc->spinlock_lockval = arch_spin_lockval(cpu);
176 #ifndef CONFIG_64BIT
177 	if (MACHINE_HAS_IEEE) {
178 		lc->extended_save_area_addr = get_zeroed_page(GFP_KERNEL);
179 		if (!lc->extended_save_area_addr)
180 			goto out;
181 	}
182 #else
183 	if (MACHINE_HAS_VX)
184 		lc->vector_save_area_addr =
185 			(unsigned long) &lc->vector_save_area;
186 	if (vdso_alloc_per_cpu(lc))
187 		goto out;
188 #endif
189 	lowcore_ptr[cpu] = lc;
190 	pcpu_sigp_retry(pcpu, SIGP_SET_PREFIX, (u32)(unsigned long) lc);
191 	return 0;
192 out:
193 	if (pcpu != &pcpu_devices[0]) {
194 		free_page(pcpu->panic_stack);
195 		free_pages(pcpu->async_stack, ASYNC_ORDER);
196 		free_pages((unsigned long) pcpu->lowcore, LC_ORDER);
197 	}
198 	return -ENOMEM;
199 }
200 
201 #ifdef CONFIG_HOTPLUG_CPU
202 
203 static void pcpu_free_lowcore(struct pcpu *pcpu)
204 {
205 	pcpu_sigp_retry(pcpu, SIGP_SET_PREFIX, 0);
206 	lowcore_ptr[pcpu - pcpu_devices] = NULL;
207 #ifndef CONFIG_64BIT
208 	if (MACHINE_HAS_IEEE) {
209 		struct _lowcore *lc = pcpu->lowcore;
210 
211 		free_page((unsigned long) lc->extended_save_area_addr);
212 		lc->extended_save_area_addr = 0;
213 	}
214 #else
215 	vdso_free_per_cpu(pcpu->lowcore);
216 #endif
217 	if (pcpu != &pcpu_devices[0]) {
218 		free_page(pcpu->panic_stack);
219 		free_pages(pcpu->async_stack, ASYNC_ORDER);
220 		free_pages((unsigned long) pcpu->lowcore, LC_ORDER);
221 	}
222 }
223 
224 #endif /* CONFIG_HOTPLUG_CPU */
225 
226 static void pcpu_prepare_secondary(struct pcpu *pcpu, int cpu)
227 {
228 	struct _lowcore *lc = pcpu->lowcore;
229 
230 	if (MACHINE_HAS_TLB_LC)
231 		cpumask_set_cpu(cpu, &init_mm.context.cpu_attach_mask);
232 	cpumask_set_cpu(cpu, mm_cpumask(&init_mm));
233 	atomic_inc(&init_mm.context.attach_count);
234 	lc->cpu_nr = cpu;
235 	lc->spinlock_lockval = arch_spin_lockval(cpu);
236 	lc->percpu_offset = __per_cpu_offset[cpu];
237 	lc->kernel_asce = S390_lowcore.kernel_asce;
238 	lc->machine_flags = S390_lowcore.machine_flags;
239 	lc->user_timer = lc->system_timer = lc->steal_timer = 0;
240 	__ctl_store(lc->cregs_save_area, 0, 15);
241 	save_access_regs((unsigned int *) lc->access_regs_save_area);
242 	memcpy(lc->stfle_fac_list, S390_lowcore.stfle_fac_list,
243 	       MAX_FACILITY_BIT/8);
244 }
245 
246 static void pcpu_attach_task(struct pcpu *pcpu, struct task_struct *tsk)
247 {
248 	struct _lowcore *lc = pcpu->lowcore;
249 	struct thread_info *ti = task_thread_info(tsk);
250 
251 	lc->kernel_stack = (unsigned long) task_stack_page(tsk)
252 		+ THREAD_SIZE - STACK_FRAME_OVERHEAD - sizeof(struct pt_regs);
253 	lc->thread_info = (unsigned long) task_thread_info(tsk);
254 	lc->current_task = (unsigned long) tsk;
255 	lc->user_timer = ti->user_timer;
256 	lc->system_timer = ti->system_timer;
257 	lc->steal_timer = 0;
258 }
259 
260 static void pcpu_start_fn(struct pcpu *pcpu, void (*func)(void *), void *data)
261 {
262 	struct _lowcore *lc = pcpu->lowcore;
263 
264 	lc->restart_stack = lc->kernel_stack;
265 	lc->restart_fn = (unsigned long) func;
266 	lc->restart_data = (unsigned long) data;
267 	lc->restart_source = -1UL;
268 	pcpu_sigp_retry(pcpu, SIGP_RESTART, 0);
269 }
270 
271 /*
272  * Call function via PSW restart on pcpu and stop the current cpu.
273  */
274 static void pcpu_delegate(struct pcpu *pcpu, void (*func)(void *),
275 			  void *data, unsigned long stack)
276 {
277 	struct _lowcore *lc = lowcore_ptr[pcpu - pcpu_devices];
278 	unsigned long source_cpu = stap();
279 
280 	__load_psw_mask(PSW_KERNEL_BITS);
281 	if (pcpu->address == source_cpu)
282 		func(data);	/* should not return */
283 	/* Stop target cpu (if func returns this stops the current cpu). */
284 	pcpu_sigp_retry(pcpu, SIGP_STOP, 0);
285 	/* Restart func on the target cpu and stop the current cpu. */
286 	mem_assign_absolute(lc->restart_stack, stack);
287 	mem_assign_absolute(lc->restart_fn, (unsigned long) func);
288 	mem_assign_absolute(lc->restart_data, (unsigned long) data);
289 	mem_assign_absolute(lc->restart_source, source_cpu);
290 	asm volatile(
291 		"0:	sigp	0,%0,%2	# sigp restart to target cpu\n"
292 		"	brc	2,0b	# busy, try again\n"
293 		"1:	sigp	0,%1,%3	# sigp stop to current cpu\n"
294 		"	brc	2,1b	# busy, try again\n"
295 		: : "d" (pcpu->address), "d" (source_cpu),
296 		    "K" (SIGP_RESTART), "K" (SIGP_STOP)
297 		: "0", "1", "cc");
298 	for (;;) ;
299 }
300 
301 /*
302  * Call function on an online CPU.
303  */
304 void smp_call_online_cpu(void (*func)(void *), void *data)
305 {
306 	struct pcpu *pcpu;
307 
308 	/* Use the current cpu if it is online. */
309 	pcpu = pcpu_find_address(cpu_online_mask, stap());
310 	if (!pcpu)
311 		/* Use the first online cpu. */
312 		pcpu = pcpu_devices + cpumask_first(cpu_online_mask);
313 	pcpu_delegate(pcpu, func, data, (unsigned long) restart_stack);
314 }
315 
316 /*
317  * Call function on the ipl CPU.
318  */
319 void smp_call_ipl_cpu(void (*func)(void *), void *data)
320 {
321 	pcpu_delegate(&pcpu_devices[0], func, data,
322 		      pcpu_devices->panic_stack + PAGE_SIZE);
323 }
324 
325 int smp_find_processor_id(u16 address)
326 {
327 	int cpu;
328 
329 	for_each_present_cpu(cpu)
330 		if (pcpu_devices[cpu].address == address)
331 			return cpu;
332 	return -1;
333 }
334 
335 int smp_vcpu_scheduled(int cpu)
336 {
337 	return pcpu_running(pcpu_devices + cpu);
338 }
339 
340 void smp_yield_cpu(int cpu)
341 {
342 	if (MACHINE_HAS_DIAG9C)
343 		asm volatile("diag %0,0,0x9c"
344 			     : : "d" (pcpu_devices[cpu].address));
345 	else if (MACHINE_HAS_DIAG44)
346 		asm volatile("diag 0,0,0x44");
347 }
348 
349 /*
350  * Send cpus emergency shutdown signal. This gives the cpus the
351  * opportunity to complete outstanding interrupts.
352  */
353 static void smp_emergency_stop(cpumask_t *cpumask)
354 {
355 	u64 end;
356 	int cpu;
357 
358 	end = get_tod_clock() + (1000000UL << 12);
359 	for_each_cpu(cpu, cpumask) {
360 		struct pcpu *pcpu = pcpu_devices + cpu;
361 		set_bit(ec_stop_cpu, &pcpu->ec_mask);
362 		while (__pcpu_sigp(pcpu->address, SIGP_EMERGENCY_SIGNAL,
363 				   0, NULL) == SIGP_CC_BUSY &&
364 		       get_tod_clock() < end)
365 			cpu_relax();
366 	}
367 	while (get_tod_clock() < end) {
368 		for_each_cpu(cpu, cpumask)
369 			if (pcpu_stopped(pcpu_devices + cpu))
370 				cpumask_clear_cpu(cpu, cpumask);
371 		if (cpumask_empty(cpumask))
372 			break;
373 		cpu_relax();
374 	}
375 }
376 
377 /*
378  * Stop all cpus but the current one.
379  */
380 void smp_send_stop(void)
381 {
382 	cpumask_t cpumask;
383 	int cpu;
384 
385 	/* Disable all interrupts/machine checks */
386 	__load_psw_mask(PSW_KERNEL_BITS | PSW_MASK_DAT);
387 	trace_hardirqs_off();
388 
389 	debug_set_critical();
390 	cpumask_copy(&cpumask, cpu_online_mask);
391 	cpumask_clear_cpu(smp_processor_id(), &cpumask);
392 
393 	if (oops_in_progress)
394 		smp_emergency_stop(&cpumask);
395 
396 	/* stop all processors */
397 	for_each_cpu(cpu, &cpumask) {
398 		struct pcpu *pcpu = pcpu_devices + cpu;
399 		pcpu_sigp_retry(pcpu, SIGP_STOP, 0);
400 		while (!pcpu_stopped(pcpu))
401 			cpu_relax();
402 	}
403 }
404 
405 /*
406  * This is the main routine where commands issued by other
407  * cpus are handled.
408  */
409 static void smp_handle_ext_call(void)
410 {
411 	unsigned long bits;
412 
413 	/* handle bit signal external calls */
414 	bits = xchg(&pcpu_devices[smp_processor_id()].ec_mask, 0);
415 	if (test_bit(ec_stop_cpu, &bits))
416 		smp_stop_cpu();
417 	if (test_bit(ec_schedule, &bits))
418 		scheduler_ipi();
419 	if (test_bit(ec_call_function_single, &bits))
420 		generic_smp_call_function_single_interrupt();
421 }
422 
423 static void do_ext_call_interrupt(struct ext_code ext_code,
424 				  unsigned int param32, unsigned long param64)
425 {
426 	inc_irq_stat(ext_code.code == 0x1202 ? IRQEXT_EXC : IRQEXT_EMS);
427 	smp_handle_ext_call();
428 }
429 
430 void arch_send_call_function_ipi_mask(const struct cpumask *mask)
431 {
432 	int cpu;
433 
434 	for_each_cpu(cpu, mask)
435 		pcpu_ec_call(pcpu_devices + cpu, ec_call_function_single);
436 }
437 
438 void arch_send_call_function_single_ipi(int cpu)
439 {
440 	pcpu_ec_call(pcpu_devices + cpu, ec_call_function_single);
441 }
442 
443 #ifndef CONFIG_64BIT
444 /*
445  * this function sends a 'purge tlb' signal to another CPU.
446  */
447 static void smp_ptlb_callback(void *info)
448 {
449 	__tlb_flush_local();
450 }
451 
452 void smp_ptlb_all(void)
453 {
454 	on_each_cpu(smp_ptlb_callback, NULL, 1);
455 }
456 EXPORT_SYMBOL(smp_ptlb_all);
457 #endif /* ! CONFIG_64BIT */
458 
459 /*
460  * this function sends a 'reschedule' IPI to another CPU.
461  * it goes straight through and wastes no time serializing
462  * anything. Worst case is that we lose a reschedule ...
463  */
464 void smp_send_reschedule(int cpu)
465 {
466 	pcpu_ec_call(pcpu_devices + cpu, ec_schedule);
467 }
468 
469 /*
470  * parameter area for the set/clear control bit callbacks
471  */
472 struct ec_creg_mask_parms {
473 	unsigned long orval;
474 	unsigned long andval;
475 	int cr;
476 };
477 
478 /*
479  * callback for setting/clearing control bits
480  */
481 static void smp_ctl_bit_callback(void *info)
482 {
483 	struct ec_creg_mask_parms *pp = info;
484 	unsigned long cregs[16];
485 
486 	__ctl_store(cregs, 0, 15);
487 	cregs[pp->cr] = (cregs[pp->cr] & pp->andval) | pp->orval;
488 	__ctl_load(cregs, 0, 15);
489 }
490 
491 /*
492  * Set a bit in a control register of all cpus
493  */
494 void smp_ctl_set_bit(int cr, int bit)
495 {
496 	struct ec_creg_mask_parms parms = { 1UL << bit, -1UL, cr };
497 
498 	on_each_cpu(smp_ctl_bit_callback, &parms, 1);
499 }
500 EXPORT_SYMBOL(smp_ctl_set_bit);
501 
502 /*
503  * Clear a bit in a control register of all cpus
504  */
505 void smp_ctl_clear_bit(int cr, int bit)
506 {
507 	struct ec_creg_mask_parms parms = { 0, ~(1UL << bit), cr };
508 
509 	on_each_cpu(smp_ctl_bit_callback, &parms, 1);
510 }
511 EXPORT_SYMBOL(smp_ctl_clear_bit);
512 
513 #ifdef CONFIG_CRASH_DUMP
514 
515 static void __init smp_get_save_area(int cpu, u16 address)
516 {
517 	void *lc = pcpu_devices[0].lowcore;
518 	struct save_area_ext *sa_ext;
519 	unsigned long vx_sa;
520 
521 	if (is_kdump_kernel())
522 		return;
523 	if (!OLDMEM_BASE && (address == boot_cpu_address ||
524 			     ipl_info.type != IPL_TYPE_FCP_DUMP))
525 		return;
526 	sa_ext = dump_save_area_create(cpu);
527 	if (!sa_ext)
528 		panic("could not allocate memory for save area\n");
529 	if (address == boot_cpu_address) {
530 		/* Copy the registers of the boot cpu. */
531 		copy_oldmem_page(1, (void *) &sa_ext->sa, sizeof(sa_ext->sa),
532 				 SAVE_AREA_BASE - PAGE_SIZE, 0);
533 		if (MACHINE_HAS_VX)
534 			save_vx_regs_safe(sa_ext->vx_regs);
535 		return;
536 	}
537 	/* Get the registers of a non-boot cpu. */
538 	__pcpu_sigp_relax(address, SIGP_STOP_AND_STORE_STATUS, 0, NULL);
539 	memcpy_real(&sa_ext->sa, lc + SAVE_AREA_BASE, sizeof(sa_ext->sa));
540 	if (!MACHINE_HAS_VX)
541 		return;
542 	/* Get the VX registers */
543 	vx_sa = __get_free_page(GFP_KERNEL);
544 	if (!vx_sa)
545 		panic("could not allocate memory for VX save area\n");
546 	__pcpu_sigp_relax(address, SIGP_STORE_ADDITIONAL_STATUS, vx_sa, NULL);
547 	memcpy(sa_ext->vx_regs, (void *) vx_sa, sizeof(sa_ext->vx_regs));
548 	free_page(vx_sa);
549 }
550 
551 int smp_store_status(int cpu)
552 {
553 	unsigned long vx_sa;
554 	struct pcpu *pcpu;
555 
556 	pcpu = pcpu_devices + cpu;
557 	if (__pcpu_sigp_relax(pcpu->address, SIGP_STOP_AND_STORE_STATUS,
558 			      0, NULL) != SIGP_CC_ORDER_CODE_ACCEPTED)
559 		return -EIO;
560 	if (!MACHINE_HAS_VX)
561 		return 0;
562 	vx_sa = __pa(pcpu->lowcore->vector_save_area_addr);
563 	__pcpu_sigp_relax(pcpu->address, SIGP_STORE_ADDITIONAL_STATUS,
564 			  vx_sa, NULL);
565 	return 0;
566 }
567 
568 #else /* CONFIG_CRASH_DUMP */
569 
570 static inline void smp_get_save_area(int cpu, u16 address) { }
571 
572 #endif /* CONFIG_CRASH_DUMP */
573 
574 void smp_cpu_set_polarization(int cpu, int val)
575 {
576 	pcpu_devices[cpu].polarization = val;
577 }
578 
579 int smp_cpu_get_polarization(int cpu)
580 {
581 	return pcpu_devices[cpu].polarization;
582 }
583 
584 static struct sclp_cpu_info *smp_get_cpu_info(void)
585 {
586 	static int use_sigp_detection;
587 	struct sclp_cpu_info *info;
588 	int address;
589 
590 	info = kzalloc(sizeof(*info), GFP_KERNEL);
591 	if (info && (use_sigp_detection || sclp_get_cpu_info(info))) {
592 		use_sigp_detection = 1;
593 		for (address = 0; address <= MAX_CPU_ADDRESS; address++) {
594 			if (__pcpu_sigp_relax(address, SIGP_SENSE, 0, NULL) ==
595 			    SIGP_CC_NOT_OPERATIONAL)
596 				continue;
597 			info->cpu[info->configured].address = address;
598 			info->configured++;
599 		}
600 		info->combined = info->configured;
601 	}
602 	return info;
603 }
604 
605 static int smp_add_present_cpu(int cpu);
606 
607 static int __smp_rescan_cpus(struct sclp_cpu_info *info, int sysfs_add)
608 {
609 	struct pcpu *pcpu;
610 	cpumask_t avail;
611 	int cpu, nr, i;
612 
613 	nr = 0;
614 	cpumask_xor(&avail, cpu_possible_mask, cpu_present_mask);
615 	cpu = cpumask_first(&avail);
616 	for (i = 0; (i < info->combined) && (cpu < nr_cpu_ids); i++) {
617 		if (info->has_cpu_type && info->cpu[i].type != boot_cpu_type)
618 			continue;
619 		if (pcpu_find_address(cpu_present_mask, info->cpu[i].address))
620 			continue;
621 		pcpu = pcpu_devices + cpu;
622 		pcpu->address = info->cpu[i].address;
623 		pcpu->state = (i >= info->configured) ?
624 			CPU_STATE_STANDBY : CPU_STATE_CONFIGURED;
625 		smp_cpu_set_polarization(cpu, POLARIZATION_UNKNOWN);
626 		set_cpu_present(cpu, true);
627 		if (sysfs_add && smp_add_present_cpu(cpu) != 0)
628 			set_cpu_present(cpu, false);
629 		else
630 			nr++;
631 		cpu = cpumask_next(cpu, &avail);
632 	}
633 	return nr;
634 }
635 
636 static void __init smp_detect_cpus(void)
637 {
638 	unsigned int cpu, c_cpus, s_cpus;
639 	struct sclp_cpu_info *info;
640 
641 	info = smp_get_cpu_info();
642 	if (!info)
643 		panic("smp_detect_cpus failed to allocate memory\n");
644 	if (info->has_cpu_type) {
645 		for (cpu = 0; cpu < info->combined; cpu++) {
646 			if (info->cpu[cpu].address != boot_cpu_address)
647 				continue;
648 			/* The boot cpu dictates the cpu type. */
649 			boot_cpu_type = info->cpu[cpu].type;
650 			break;
651 		}
652 	}
653 	c_cpus = s_cpus = 0;
654 	for (cpu = 0; cpu < info->combined; cpu++) {
655 		if (info->has_cpu_type && info->cpu[cpu].type != boot_cpu_type)
656 			continue;
657 		if (cpu < info->configured) {
658 			smp_get_save_area(c_cpus, info->cpu[cpu].address);
659 			c_cpus++;
660 		} else
661 			s_cpus++;
662 	}
663 	pr_info("%d configured CPUs, %d standby CPUs\n", c_cpus, s_cpus);
664 	get_online_cpus();
665 	__smp_rescan_cpus(info, 0);
666 	put_online_cpus();
667 	kfree(info);
668 }
669 
670 /*
671  *	Activate a secondary processor.
672  */
673 static void smp_start_secondary(void *cpuvoid)
674 {
675 	S390_lowcore.last_update_clock = get_tod_clock();
676 	S390_lowcore.restart_stack = (unsigned long) restart_stack;
677 	S390_lowcore.restart_fn = (unsigned long) do_restart;
678 	S390_lowcore.restart_data = 0;
679 	S390_lowcore.restart_source = -1UL;
680 	restore_access_regs(S390_lowcore.access_regs_save_area);
681 	__ctl_load(S390_lowcore.cregs_save_area, 0, 15);
682 	__load_psw_mask(PSW_KERNEL_BITS | PSW_MASK_DAT);
683 	cpu_init();
684 	preempt_disable();
685 	init_cpu_timer();
686 	vtime_init();
687 	pfault_init();
688 	notify_cpu_starting(smp_processor_id());
689 	set_cpu_online(smp_processor_id(), true);
690 	inc_irq_stat(CPU_RST);
691 	local_irq_enable();
692 	cpu_startup_entry(CPUHP_ONLINE);
693 }
694 
695 /* Upping and downing of CPUs */
696 int __cpu_up(unsigned int cpu, struct task_struct *tidle)
697 {
698 	struct pcpu *pcpu;
699 	int rc;
700 
701 	pcpu = pcpu_devices + cpu;
702 	if (pcpu->state != CPU_STATE_CONFIGURED)
703 		return -EIO;
704 	if (pcpu_sigp_retry(pcpu, SIGP_INITIAL_CPU_RESET, 0) !=
705 	    SIGP_CC_ORDER_CODE_ACCEPTED)
706 		return -EIO;
707 
708 	rc = pcpu_alloc_lowcore(pcpu, cpu);
709 	if (rc)
710 		return rc;
711 	pcpu_prepare_secondary(pcpu, cpu);
712 	pcpu_attach_task(pcpu, tidle);
713 	pcpu_start_fn(pcpu, smp_start_secondary, NULL);
714 	while (!cpu_online(cpu))
715 		cpu_relax();
716 	return 0;
717 }
718 
719 static unsigned int setup_possible_cpus __initdata;
720 
721 static int __init _setup_possible_cpus(char *s)
722 {
723 	get_option(&s, &setup_possible_cpus);
724 	return 0;
725 }
726 early_param("possible_cpus", _setup_possible_cpus);
727 
728 #ifdef CONFIG_HOTPLUG_CPU
729 
730 int __cpu_disable(void)
731 {
732 	unsigned long cregs[16];
733 
734 	/* Handle possible pending IPIs */
735 	smp_handle_ext_call();
736 	set_cpu_online(smp_processor_id(), false);
737 	/* Disable pseudo page faults on this cpu. */
738 	pfault_fini();
739 	/* Disable interrupt sources via control register. */
740 	__ctl_store(cregs, 0, 15);
741 	cregs[0]  &= ~0x0000ee70UL;	/* disable all external interrupts */
742 	cregs[6]  &= ~0xff000000UL;	/* disable all I/O interrupts */
743 	cregs[14] &= ~0x1f000000UL;	/* disable most machine checks */
744 	__ctl_load(cregs, 0, 15);
745 	clear_cpu_flag(CIF_NOHZ_DELAY);
746 	return 0;
747 }
748 
749 void __cpu_die(unsigned int cpu)
750 {
751 	struct pcpu *pcpu;
752 
753 	/* Wait until target cpu is down */
754 	pcpu = pcpu_devices + cpu;
755 	while (!pcpu_stopped(pcpu))
756 		cpu_relax();
757 	pcpu_free_lowcore(pcpu);
758 	atomic_dec(&init_mm.context.attach_count);
759 	cpumask_clear_cpu(cpu, mm_cpumask(&init_mm));
760 	if (MACHINE_HAS_TLB_LC)
761 		cpumask_clear_cpu(cpu, &init_mm.context.cpu_attach_mask);
762 }
763 
764 void __noreturn cpu_die(void)
765 {
766 	idle_task_exit();
767 	pcpu_sigp_retry(pcpu_devices + smp_processor_id(), SIGP_STOP, 0);
768 	for (;;) ;
769 }
770 
771 #endif /* CONFIG_HOTPLUG_CPU */
772 
773 void __init smp_fill_possible_mask(void)
774 {
775 	unsigned int possible, sclp, cpu;
776 
777 	sclp = sclp_get_max_cpu() ?: nr_cpu_ids;
778 	possible = setup_possible_cpus ?: nr_cpu_ids;
779 	possible = min(possible, sclp);
780 	for (cpu = 0; cpu < possible && cpu < nr_cpu_ids; cpu++)
781 		set_cpu_possible(cpu, true);
782 }
783 
784 void __init smp_prepare_cpus(unsigned int max_cpus)
785 {
786 	/* request the 0x1201 emergency signal external interrupt */
787 	if (register_external_irq(EXT_IRQ_EMERGENCY_SIG, do_ext_call_interrupt))
788 		panic("Couldn't request external interrupt 0x1201");
789 	/* request the 0x1202 external call external interrupt */
790 	if (register_external_irq(EXT_IRQ_EXTERNAL_CALL, do_ext_call_interrupt))
791 		panic("Couldn't request external interrupt 0x1202");
792 	smp_detect_cpus();
793 }
794 
795 void __init smp_prepare_boot_cpu(void)
796 {
797 	struct pcpu *pcpu = pcpu_devices;
798 
799 	boot_cpu_address = stap();
800 	pcpu->state = CPU_STATE_CONFIGURED;
801 	pcpu->address = boot_cpu_address;
802 	pcpu->lowcore = (struct _lowcore *)(unsigned long) store_prefix();
803 	pcpu->async_stack = S390_lowcore.async_stack - ASYNC_SIZE
804 		+ STACK_FRAME_OVERHEAD + sizeof(struct pt_regs);
805 	pcpu->panic_stack = S390_lowcore.panic_stack - PAGE_SIZE
806 		+ STACK_FRAME_OVERHEAD + sizeof(struct pt_regs);
807 	S390_lowcore.percpu_offset = __per_cpu_offset[0];
808 	smp_cpu_set_polarization(0, POLARIZATION_UNKNOWN);
809 	set_cpu_present(0, true);
810 	set_cpu_online(0, true);
811 }
812 
813 void __init smp_cpus_done(unsigned int max_cpus)
814 {
815 }
816 
817 void __init smp_setup_processor_id(void)
818 {
819 	S390_lowcore.cpu_nr = 0;
820 	S390_lowcore.spinlock_lockval = arch_spin_lockval(0);
821 }
822 
823 /*
824  * the frequency of the profiling timer can be changed
825  * by writing a multiplier value into /proc/profile.
826  *
827  * usually you want to run this on all CPUs ;)
828  */
829 int setup_profiling_timer(unsigned int multiplier)
830 {
831 	return 0;
832 }
833 
834 #ifdef CONFIG_HOTPLUG_CPU
835 static ssize_t cpu_configure_show(struct device *dev,
836 				  struct device_attribute *attr, char *buf)
837 {
838 	ssize_t count;
839 
840 	mutex_lock(&smp_cpu_state_mutex);
841 	count = sprintf(buf, "%d\n", pcpu_devices[dev->id].state);
842 	mutex_unlock(&smp_cpu_state_mutex);
843 	return count;
844 }
845 
846 static ssize_t cpu_configure_store(struct device *dev,
847 				   struct device_attribute *attr,
848 				   const char *buf, size_t count)
849 {
850 	struct pcpu *pcpu;
851 	int cpu, val, rc;
852 	char delim;
853 
854 	if (sscanf(buf, "%d %c", &val, &delim) != 1)
855 		return -EINVAL;
856 	if (val != 0 && val != 1)
857 		return -EINVAL;
858 	get_online_cpus();
859 	mutex_lock(&smp_cpu_state_mutex);
860 	rc = -EBUSY;
861 	/* disallow configuration changes of online cpus and cpu 0 */
862 	cpu = dev->id;
863 	if (cpu_online(cpu) || cpu == 0)
864 		goto out;
865 	pcpu = pcpu_devices + cpu;
866 	rc = 0;
867 	switch (val) {
868 	case 0:
869 		if (pcpu->state != CPU_STATE_CONFIGURED)
870 			break;
871 		rc = sclp_cpu_deconfigure(pcpu->address);
872 		if (rc)
873 			break;
874 		pcpu->state = CPU_STATE_STANDBY;
875 		smp_cpu_set_polarization(cpu, POLARIZATION_UNKNOWN);
876 		topology_expect_change();
877 		break;
878 	case 1:
879 		if (pcpu->state != CPU_STATE_STANDBY)
880 			break;
881 		rc = sclp_cpu_configure(pcpu->address);
882 		if (rc)
883 			break;
884 		pcpu->state = CPU_STATE_CONFIGURED;
885 		smp_cpu_set_polarization(cpu, POLARIZATION_UNKNOWN);
886 		topology_expect_change();
887 		break;
888 	default:
889 		break;
890 	}
891 out:
892 	mutex_unlock(&smp_cpu_state_mutex);
893 	put_online_cpus();
894 	return rc ? rc : count;
895 }
896 static DEVICE_ATTR(configure, 0644, cpu_configure_show, cpu_configure_store);
897 #endif /* CONFIG_HOTPLUG_CPU */
898 
899 static ssize_t show_cpu_address(struct device *dev,
900 				struct device_attribute *attr, char *buf)
901 {
902 	return sprintf(buf, "%d\n", pcpu_devices[dev->id].address);
903 }
904 static DEVICE_ATTR(address, 0444, show_cpu_address, NULL);
905 
906 static struct attribute *cpu_common_attrs[] = {
907 #ifdef CONFIG_HOTPLUG_CPU
908 	&dev_attr_configure.attr,
909 #endif
910 	&dev_attr_address.attr,
911 	NULL,
912 };
913 
914 static struct attribute_group cpu_common_attr_group = {
915 	.attrs = cpu_common_attrs,
916 };
917 
918 static struct attribute *cpu_online_attrs[] = {
919 	&dev_attr_idle_count.attr,
920 	&dev_attr_idle_time_us.attr,
921 	NULL,
922 };
923 
924 static struct attribute_group cpu_online_attr_group = {
925 	.attrs = cpu_online_attrs,
926 };
927 
928 static int smp_cpu_notify(struct notifier_block *self, unsigned long action,
929 			  void *hcpu)
930 {
931 	unsigned int cpu = (unsigned int)(long)hcpu;
932 	struct cpu *c = pcpu_devices[cpu].cpu;
933 	struct device *s = &c->dev;
934 	int err = 0;
935 
936 	switch (action & ~CPU_TASKS_FROZEN) {
937 	case CPU_ONLINE:
938 		err = sysfs_create_group(&s->kobj, &cpu_online_attr_group);
939 		break;
940 	case CPU_DEAD:
941 		sysfs_remove_group(&s->kobj, &cpu_online_attr_group);
942 		break;
943 	}
944 	return notifier_from_errno(err);
945 }
946 
947 static int smp_add_present_cpu(int cpu)
948 {
949 	struct device *s;
950 	struct cpu *c;
951 	int rc;
952 
953 	c = kzalloc(sizeof(*c), GFP_KERNEL);
954 	if (!c)
955 		return -ENOMEM;
956 	pcpu_devices[cpu].cpu = c;
957 	s = &c->dev;
958 	c->hotpluggable = 1;
959 	rc = register_cpu(c, cpu);
960 	if (rc)
961 		goto out;
962 	rc = sysfs_create_group(&s->kobj, &cpu_common_attr_group);
963 	if (rc)
964 		goto out_cpu;
965 	if (cpu_online(cpu)) {
966 		rc = sysfs_create_group(&s->kobj, &cpu_online_attr_group);
967 		if (rc)
968 			goto out_online;
969 	}
970 	rc = topology_cpu_init(c);
971 	if (rc)
972 		goto out_topology;
973 	return 0;
974 
975 out_topology:
976 	if (cpu_online(cpu))
977 		sysfs_remove_group(&s->kobj, &cpu_online_attr_group);
978 out_online:
979 	sysfs_remove_group(&s->kobj, &cpu_common_attr_group);
980 out_cpu:
981 #ifdef CONFIG_HOTPLUG_CPU
982 	unregister_cpu(c);
983 #endif
984 out:
985 	return rc;
986 }
987 
988 #ifdef CONFIG_HOTPLUG_CPU
989 
990 int __ref smp_rescan_cpus(void)
991 {
992 	struct sclp_cpu_info *info;
993 	int nr;
994 
995 	info = smp_get_cpu_info();
996 	if (!info)
997 		return -ENOMEM;
998 	get_online_cpus();
999 	mutex_lock(&smp_cpu_state_mutex);
1000 	nr = __smp_rescan_cpus(info, 1);
1001 	mutex_unlock(&smp_cpu_state_mutex);
1002 	put_online_cpus();
1003 	kfree(info);
1004 	if (nr)
1005 		topology_schedule_update();
1006 	return 0;
1007 }
1008 
1009 static ssize_t __ref rescan_store(struct device *dev,
1010 				  struct device_attribute *attr,
1011 				  const char *buf,
1012 				  size_t count)
1013 {
1014 	int rc;
1015 
1016 	rc = smp_rescan_cpus();
1017 	return rc ? rc : count;
1018 }
1019 static DEVICE_ATTR(rescan, 0200, NULL, rescan_store);
1020 #endif /* CONFIG_HOTPLUG_CPU */
1021 
1022 static int __init s390_smp_init(void)
1023 {
1024 	int cpu, rc = 0;
1025 
1026 #ifdef CONFIG_HOTPLUG_CPU
1027 	rc = device_create_file(cpu_subsys.dev_root, &dev_attr_rescan);
1028 	if (rc)
1029 		return rc;
1030 #endif
1031 	cpu_notifier_register_begin();
1032 	for_each_present_cpu(cpu) {
1033 		rc = smp_add_present_cpu(cpu);
1034 		if (rc)
1035 			goto out;
1036 	}
1037 
1038 	__hotcpu_notifier(smp_cpu_notify, 0);
1039 
1040 out:
1041 	cpu_notifier_register_done();
1042 	return rc;
1043 }
1044 subsys_initcall(s390_smp_init);
1045