xref: /openbmc/linux/arch/s390/kernel/smp.c (revision 8e9356c6)
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 "entry.h"
49 
50 enum {
51 	ec_schedule = 0,
52 	ec_call_function_single,
53 	ec_stop_cpu,
54 };
55 
56 enum {
57 	CPU_STATE_STANDBY,
58 	CPU_STATE_CONFIGURED,
59 };
60 
61 struct pcpu {
62 	struct cpu *cpu;
63 	struct _lowcore *lowcore;	/* lowcore page(s) for the cpu */
64 	unsigned long async_stack;	/* async stack for the cpu */
65 	unsigned long panic_stack;	/* panic stack for the cpu */
66 	unsigned long ec_mask;		/* bit mask for ec_xxx functions */
67 	int state;			/* physical cpu state */
68 	int polarization;		/* physical polarization */
69 	u16 address;			/* physical cpu address */
70 };
71 
72 static u8 boot_cpu_type;
73 static u16 boot_cpu_address;
74 static struct pcpu pcpu_devices[NR_CPUS];
75 
76 /*
77  * The smp_cpu_state_mutex must be held when changing the state or polarization
78  * member of a pcpu data structure within the pcpu_devices arreay.
79  */
80 DEFINE_MUTEX(smp_cpu_state_mutex);
81 
82 /*
83  * Signal processor helper functions.
84  */
85 static inline int __pcpu_sigp(u16 addr, u8 order, u32 parm, u32 *status)
86 {
87 	register unsigned int reg1 asm ("1") = parm;
88 	int cc;
89 
90 	asm volatile(
91 		"	sigp	%1,%2,0(%3)\n"
92 		"	ipm	%0\n"
93 		"	srl	%0,28\n"
94 		: "=d" (cc), "+d" (reg1) : "d" (addr), "a" (order) : "cc");
95 	if (status && cc == 1)
96 		*status = reg1;
97 	return cc;
98 }
99 
100 static inline int __pcpu_sigp_relax(u16 addr, u8 order, u32 parm, u32 *status)
101 {
102 	int cc;
103 
104 	while (1) {
105 		cc = __pcpu_sigp(addr, order, parm, NULL);
106 		if (cc != SIGP_CC_BUSY)
107 			return cc;
108 		cpu_relax();
109 	}
110 }
111 
112 static int pcpu_sigp_retry(struct pcpu *pcpu, u8 order, u32 parm)
113 {
114 	int cc, retry;
115 
116 	for (retry = 0; ; retry++) {
117 		cc = __pcpu_sigp(pcpu->address, order, parm, NULL);
118 		if (cc != SIGP_CC_BUSY)
119 			break;
120 		if (retry >= 3)
121 			udelay(10);
122 	}
123 	return cc;
124 }
125 
126 static inline int pcpu_stopped(struct pcpu *pcpu)
127 {
128 	u32 uninitialized_var(status);
129 
130 	if (__pcpu_sigp(pcpu->address, SIGP_SENSE,
131 			0, &status) != SIGP_CC_STATUS_STORED)
132 		return 0;
133 	return !!(status & (SIGP_STATUS_CHECK_STOP|SIGP_STATUS_STOPPED));
134 }
135 
136 static inline int pcpu_running(struct pcpu *pcpu)
137 {
138 	if (__pcpu_sigp(pcpu->address, SIGP_SENSE_RUNNING,
139 			0, NULL) != SIGP_CC_STATUS_STORED)
140 		return 1;
141 	/* Status stored condition code is equivalent to cpu not running. */
142 	return 0;
143 }
144 
145 /*
146  * Find struct pcpu by cpu address.
147  */
148 static struct pcpu *pcpu_find_address(const struct cpumask *mask, int address)
149 {
150 	int cpu;
151 
152 	for_each_cpu(cpu, mask)
153 		if (pcpu_devices[cpu].address == address)
154 			return pcpu_devices + cpu;
155 	return NULL;
156 }
157 
158 static void pcpu_ec_call(struct pcpu *pcpu, int ec_bit)
159 {
160 	int order;
161 
162 	if (test_and_set_bit(ec_bit, &pcpu->ec_mask))
163 		return;
164 	order = pcpu_running(pcpu) ? SIGP_EXTERNAL_CALL : SIGP_EMERGENCY_SIGNAL;
165 	pcpu_sigp_retry(pcpu, order, 0);
166 }
167 
168 static int pcpu_alloc_lowcore(struct pcpu *pcpu, int cpu)
169 {
170 	struct _lowcore *lc;
171 
172 	if (pcpu != &pcpu_devices[0]) {
173 		pcpu->lowcore =	(struct _lowcore *)
174 			__get_free_pages(GFP_KERNEL | GFP_DMA, LC_ORDER);
175 		pcpu->async_stack = __get_free_pages(GFP_KERNEL, ASYNC_ORDER);
176 		pcpu->panic_stack = __get_free_page(GFP_KERNEL);
177 		if (!pcpu->lowcore || !pcpu->panic_stack || !pcpu->async_stack)
178 			goto out;
179 	}
180 	lc = pcpu->lowcore;
181 	memcpy(lc, &S390_lowcore, 512);
182 	memset((char *) lc + 512, 0, sizeof(*lc) - 512);
183 	lc->async_stack = pcpu->async_stack + ASYNC_SIZE
184 		- STACK_FRAME_OVERHEAD - sizeof(struct pt_regs);
185 	lc->panic_stack = pcpu->panic_stack + PAGE_SIZE
186 		- STACK_FRAME_OVERHEAD - sizeof(struct pt_regs);
187 	lc->cpu_nr = cpu;
188 #ifndef CONFIG_64BIT
189 	if (MACHINE_HAS_IEEE) {
190 		lc->extended_save_area_addr = get_zeroed_page(GFP_KERNEL);
191 		if (!lc->extended_save_area_addr)
192 			goto out;
193 	}
194 #else
195 	if (vdso_alloc_per_cpu(lc))
196 		goto out;
197 #endif
198 	lowcore_ptr[cpu] = lc;
199 	pcpu_sigp_retry(pcpu, SIGP_SET_PREFIX, (u32)(unsigned long) lc);
200 	return 0;
201 out:
202 	if (pcpu != &pcpu_devices[0]) {
203 		free_page(pcpu->panic_stack);
204 		free_pages(pcpu->async_stack, ASYNC_ORDER);
205 		free_pages((unsigned long) pcpu->lowcore, LC_ORDER);
206 	}
207 	return -ENOMEM;
208 }
209 
210 #ifdef CONFIG_HOTPLUG_CPU
211 
212 static void pcpu_free_lowcore(struct pcpu *pcpu)
213 {
214 	pcpu_sigp_retry(pcpu, SIGP_SET_PREFIX, 0);
215 	lowcore_ptr[pcpu - pcpu_devices] = NULL;
216 #ifndef CONFIG_64BIT
217 	if (MACHINE_HAS_IEEE) {
218 		struct _lowcore *lc = pcpu->lowcore;
219 
220 		free_page((unsigned long) lc->extended_save_area_addr);
221 		lc->extended_save_area_addr = 0;
222 	}
223 #else
224 	vdso_free_per_cpu(pcpu->lowcore);
225 #endif
226 	if (pcpu != &pcpu_devices[0]) {
227 		free_page(pcpu->panic_stack);
228 		free_pages(pcpu->async_stack, ASYNC_ORDER);
229 		free_pages((unsigned long) pcpu->lowcore, LC_ORDER);
230 	}
231 }
232 
233 #endif /* CONFIG_HOTPLUG_CPU */
234 
235 static void pcpu_prepare_secondary(struct pcpu *pcpu, int cpu)
236 {
237 	struct _lowcore *lc = pcpu->lowcore;
238 
239 	atomic_inc(&init_mm.context.attach_count);
240 	lc->cpu_nr = cpu;
241 	lc->percpu_offset = __per_cpu_offset[cpu];
242 	lc->kernel_asce = S390_lowcore.kernel_asce;
243 	lc->machine_flags = S390_lowcore.machine_flags;
244 	lc->ftrace_func = S390_lowcore.ftrace_func;
245 	lc->user_timer = lc->system_timer = lc->steal_timer = 0;
246 	__ctl_store(lc->cregs_save_area, 0, 15);
247 	save_access_regs((unsigned int *) lc->access_regs_save_area);
248 	memcpy(lc->stfle_fac_list, S390_lowcore.stfle_fac_list,
249 	       MAX_FACILITY_BIT/8);
250 }
251 
252 static void pcpu_attach_task(struct pcpu *pcpu, struct task_struct *tsk)
253 {
254 	struct _lowcore *lc = pcpu->lowcore;
255 	struct thread_info *ti = task_thread_info(tsk);
256 
257 	lc->kernel_stack = (unsigned long) task_stack_page(tsk)
258 		+ THREAD_SIZE - STACK_FRAME_OVERHEAD - sizeof(struct pt_regs);
259 	lc->thread_info = (unsigned long) task_thread_info(tsk);
260 	lc->current_task = (unsigned long) tsk;
261 	lc->user_timer = ti->user_timer;
262 	lc->system_timer = ti->system_timer;
263 	lc->steal_timer = 0;
264 }
265 
266 static void pcpu_start_fn(struct pcpu *pcpu, void (*func)(void *), void *data)
267 {
268 	struct _lowcore *lc = pcpu->lowcore;
269 
270 	lc->restart_stack = lc->kernel_stack;
271 	lc->restart_fn = (unsigned long) func;
272 	lc->restart_data = (unsigned long) data;
273 	lc->restart_source = -1UL;
274 	pcpu_sigp_retry(pcpu, SIGP_RESTART, 0);
275 }
276 
277 /*
278  * Call function via PSW restart on pcpu and stop the current cpu.
279  */
280 static void pcpu_delegate(struct pcpu *pcpu, void (*func)(void *),
281 			  void *data, unsigned long stack)
282 {
283 	struct _lowcore *lc = lowcore_ptr[pcpu - pcpu_devices];
284 	unsigned long source_cpu = stap();
285 
286 	__load_psw_mask(PSW_KERNEL_BITS);
287 	if (pcpu->address == source_cpu)
288 		func(data);	/* should not return */
289 	/* Stop target cpu (if func returns this stops the current cpu). */
290 	pcpu_sigp_retry(pcpu, SIGP_STOP, 0);
291 	/* Restart func on the target cpu and stop the current cpu. */
292 	mem_assign_absolute(lc->restart_stack, stack);
293 	mem_assign_absolute(lc->restart_fn, (unsigned long) func);
294 	mem_assign_absolute(lc->restart_data, (unsigned long) data);
295 	mem_assign_absolute(lc->restart_source, source_cpu);
296 	asm volatile(
297 		"0:	sigp	0,%0,%2	# sigp restart to target cpu\n"
298 		"	brc	2,0b	# busy, try again\n"
299 		"1:	sigp	0,%1,%3	# sigp stop to current cpu\n"
300 		"	brc	2,1b	# busy, try again\n"
301 		: : "d" (pcpu->address), "d" (source_cpu),
302 		    "K" (SIGP_RESTART), "K" (SIGP_STOP)
303 		: "0", "1", "cc");
304 	for (;;) ;
305 }
306 
307 /*
308  * Call function on an online CPU.
309  */
310 void smp_call_online_cpu(void (*func)(void *), void *data)
311 {
312 	struct pcpu *pcpu;
313 
314 	/* Use the current cpu if it is online. */
315 	pcpu = pcpu_find_address(cpu_online_mask, stap());
316 	if (!pcpu)
317 		/* Use the first online cpu. */
318 		pcpu = pcpu_devices + cpumask_first(cpu_online_mask);
319 	pcpu_delegate(pcpu, func, data, (unsigned long) restart_stack);
320 }
321 
322 /*
323  * Call function on the ipl CPU.
324  */
325 void smp_call_ipl_cpu(void (*func)(void *), void *data)
326 {
327 	pcpu_delegate(&pcpu_devices[0], func, data,
328 		      pcpu_devices->panic_stack + PAGE_SIZE);
329 }
330 
331 int smp_find_processor_id(u16 address)
332 {
333 	int cpu;
334 
335 	for_each_present_cpu(cpu)
336 		if (pcpu_devices[cpu].address == address)
337 			return cpu;
338 	return -1;
339 }
340 
341 int smp_vcpu_scheduled(int cpu)
342 {
343 	return pcpu_running(pcpu_devices + cpu);
344 }
345 
346 void smp_yield(void)
347 {
348 	if (MACHINE_HAS_DIAG44)
349 		asm volatile("diag 0,0,0x44");
350 }
351 
352 void smp_yield_cpu(int cpu)
353 {
354 	if (MACHINE_HAS_DIAG9C)
355 		asm volatile("diag %0,0,0x9c"
356 			     : : "d" (pcpu_devices[cpu].address));
357 	else if (MACHINE_HAS_DIAG44)
358 		asm volatile("diag 0,0,0x44");
359 }
360 
361 /*
362  * Send cpus emergency shutdown signal. This gives the cpus the
363  * opportunity to complete outstanding interrupts.
364  */
365 static void smp_emergency_stop(cpumask_t *cpumask)
366 {
367 	u64 end;
368 	int cpu;
369 
370 	end = get_tod_clock() + (1000000UL << 12);
371 	for_each_cpu(cpu, cpumask) {
372 		struct pcpu *pcpu = pcpu_devices + cpu;
373 		set_bit(ec_stop_cpu, &pcpu->ec_mask);
374 		while (__pcpu_sigp(pcpu->address, SIGP_EMERGENCY_SIGNAL,
375 				   0, NULL) == SIGP_CC_BUSY &&
376 		       get_tod_clock() < end)
377 			cpu_relax();
378 	}
379 	while (get_tod_clock() < end) {
380 		for_each_cpu(cpu, cpumask)
381 			if (pcpu_stopped(pcpu_devices + cpu))
382 				cpumask_clear_cpu(cpu, cpumask);
383 		if (cpumask_empty(cpumask))
384 			break;
385 		cpu_relax();
386 	}
387 }
388 
389 /*
390  * Stop all cpus but the current one.
391  */
392 void smp_send_stop(void)
393 {
394 	cpumask_t cpumask;
395 	int cpu;
396 
397 	/* Disable all interrupts/machine checks */
398 	__load_psw_mask(PSW_KERNEL_BITS | PSW_MASK_DAT);
399 	trace_hardirqs_off();
400 
401 	debug_set_critical();
402 	cpumask_copy(&cpumask, cpu_online_mask);
403 	cpumask_clear_cpu(smp_processor_id(), &cpumask);
404 
405 	if (oops_in_progress)
406 		smp_emergency_stop(&cpumask);
407 
408 	/* stop all processors */
409 	for_each_cpu(cpu, &cpumask) {
410 		struct pcpu *pcpu = pcpu_devices + cpu;
411 		pcpu_sigp_retry(pcpu, SIGP_STOP, 0);
412 		while (!pcpu_stopped(pcpu))
413 			cpu_relax();
414 	}
415 }
416 
417 /*
418  * Stop the current cpu.
419  */
420 void smp_stop_cpu(void)
421 {
422 	pcpu_sigp_retry(pcpu_devices + smp_processor_id(), SIGP_STOP, 0);
423 	for (;;) ;
424 }
425 
426 /*
427  * This is the main routine where commands issued by other
428  * cpus are handled.
429  */
430 static void smp_handle_ext_call(void)
431 {
432 	unsigned long bits;
433 
434 	/* handle bit signal external calls */
435 	bits = xchg(&pcpu_devices[smp_processor_id()].ec_mask, 0);
436 	if (test_bit(ec_stop_cpu, &bits))
437 		smp_stop_cpu();
438 	if (test_bit(ec_schedule, &bits))
439 		scheduler_ipi();
440 	if (test_bit(ec_call_function_single, &bits))
441 		generic_smp_call_function_single_interrupt();
442 }
443 
444 static void do_ext_call_interrupt(struct ext_code ext_code,
445 				  unsigned int param32, unsigned long param64)
446 {
447 	inc_irq_stat(ext_code.code == 0x1202 ? IRQEXT_EXC : IRQEXT_EMS);
448 	smp_handle_ext_call();
449 }
450 
451 void arch_send_call_function_ipi_mask(const struct cpumask *mask)
452 {
453 	int cpu;
454 
455 	for_each_cpu(cpu, mask)
456 		pcpu_ec_call(pcpu_devices + cpu, ec_call_function_single);
457 }
458 
459 void arch_send_call_function_single_ipi(int cpu)
460 {
461 	pcpu_ec_call(pcpu_devices + cpu, ec_call_function_single);
462 }
463 
464 #ifndef CONFIG_64BIT
465 /*
466  * this function sends a 'purge tlb' signal to another CPU.
467  */
468 static void smp_ptlb_callback(void *info)
469 {
470 	__tlb_flush_local();
471 }
472 
473 void smp_ptlb_all(void)
474 {
475 	on_each_cpu(smp_ptlb_callback, NULL, 1);
476 }
477 EXPORT_SYMBOL(smp_ptlb_all);
478 #endif /* ! CONFIG_64BIT */
479 
480 /*
481  * this function sends a 'reschedule' IPI to another CPU.
482  * it goes straight through and wastes no time serializing
483  * anything. Worst case is that we lose a reschedule ...
484  */
485 void smp_send_reschedule(int cpu)
486 {
487 	pcpu_ec_call(pcpu_devices + cpu, ec_schedule);
488 }
489 
490 /*
491  * parameter area for the set/clear control bit callbacks
492  */
493 struct ec_creg_mask_parms {
494 	unsigned long orval;
495 	unsigned long andval;
496 	int cr;
497 };
498 
499 /*
500  * callback for setting/clearing control bits
501  */
502 static void smp_ctl_bit_callback(void *info)
503 {
504 	struct ec_creg_mask_parms *pp = info;
505 	unsigned long cregs[16];
506 
507 	__ctl_store(cregs, 0, 15);
508 	cregs[pp->cr] = (cregs[pp->cr] & pp->andval) | pp->orval;
509 	__ctl_load(cregs, 0, 15);
510 }
511 
512 /*
513  * Set a bit in a control register of all cpus
514  */
515 void smp_ctl_set_bit(int cr, int bit)
516 {
517 	struct ec_creg_mask_parms parms = { 1UL << bit, -1UL, cr };
518 
519 	on_each_cpu(smp_ctl_bit_callback, &parms, 1);
520 }
521 EXPORT_SYMBOL(smp_ctl_set_bit);
522 
523 /*
524  * Clear a bit in a control register of all cpus
525  */
526 void smp_ctl_clear_bit(int cr, int bit)
527 {
528 	struct ec_creg_mask_parms parms = { 0, ~(1UL << bit), cr };
529 
530 	on_each_cpu(smp_ctl_bit_callback, &parms, 1);
531 }
532 EXPORT_SYMBOL(smp_ctl_clear_bit);
533 
534 #if defined(CONFIG_ZFCPDUMP) || defined(CONFIG_CRASH_DUMP)
535 
536 static void __init smp_get_save_area(int cpu, u16 address)
537 {
538 	void *lc = pcpu_devices[0].lowcore;
539 	struct save_area *save_area;
540 
541 	if (is_kdump_kernel())
542 		return;
543 	if (!OLDMEM_BASE && (address == boot_cpu_address ||
544 			     ipl_info.type != IPL_TYPE_FCP_DUMP))
545 		return;
546 	save_area = dump_save_area_create(cpu);
547 	if (!save_area)
548 		panic("could not allocate memory for save area\n");
549 #ifdef CONFIG_CRASH_DUMP
550 	if (address == boot_cpu_address) {
551 		/* Copy the registers of the boot cpu. */
552 		copy_oldmem_page(1, (void *) save_area, sizeof(*save_area),
553 				 SAVE_AREA_BASE - PAGE_SIZE, 0);
554 		return;
555 	}
556 #endif
557 	/* Get the registers of a non-boot cpu. */
558 	__pcpu_sigp_relax(address, SIGP_STOP_AND_STORE_STATUS, 0, NULL);
559 	memcpy_real(save_area, lc + SAVE_AREA_BASE, sizeof(*save_area));
560 }
561 
562 int smp_store_status(int cpu)
563 {
564 	struct pcpu *pcpu;
565 
566 	pcpu = pcpu_devices + cpu;
567 	if (__pcpu_sigp_relax(pcpu->address, SIGP_STOP_AND_STORE_STATUS,
568 			      0, NULL) != SIGP_CC_ORDER_CODE_ACCEPTED)
569 		return -EIO;
570 	return 0;
571 }
572 
573 #else /* CONFIG_ZFCPDUMP || CONFIG_CRASH_DUMP */
574 
575 static inline void smp_get_save_area(int cpu, u16 address) { }
576 
577 #endif /* CONFIG_ZFCPDUMP || CONFIG_CRASH_DUMP */
578 
579 void smp_cpu_set_polarization(int cpu, int val)
580 {
581 	pcpu_devices[cpu].polarization = val;
582 }
583 
584 int smp_cpu_get_polarization(int cpu)
585 {
586 	return pcpu_devices[cpu].polarization;
587 }
588 
589 static struct sclp_cpu_info *smp_get_cpu_info(void)
590 {
591 	static int use_sigp_detection;
592 	struct sclp_cpu_info *info;
593 	int address;
594 
595 	info = kzalloc(sizeof(*info), GFP_KERNEL);
596 	if (info && (use_sigp_detection || sclp_get_cpu_info(info))) {
597 		use_sigp_detection = 1;
598 		for (address = 0; address <= MAX_CPU_ADDRESS; address++) {
599 			if (__pcpu_sigp_relax(address, SIGP_SENSE, 0, NULL) ==
600 			    SIGP_CC_NOT_OPERATIONAL)
601 				continue;
602 			info->cpu[info->configured].address = address;
603 			info->configured++;
604 		}
605 		info->combined = info->configured;
606 	}
607 	return info;
608 }
609 
610 static int smp_add_present_cpu(int cpu);
611 
612 static int __smp_rescan_cpus(struct sclp_cpu_info *info, int sysfs_add)
613 {
614 	struct pcpu *pcpu;
615 	cpumask_t avail;
616 	int cpu, nr, i;
617 
618 	nr = 0;
619 	cpumask_xor(&avail, cpu_possible_mask, cpu_present_mask);
620 	cpu = cpumask_first(&avail);
621 	for (i = 0; (i < info->combined) && (cpu < nr_cpu_ids); i++) {
622 		if (info->has_cpu_type && info->cpu[i].type != boot_cpu_type)
623 			continue;
624 		if (pcpu_find_address(cpu_present_mask, info->cpu[i].address))
625 			continue;
626 		pcpu = pcpu_devices + cpu;
627 		pcpu->address = info->cpu[i].address;
628 		pcpu->state = (i >= info->configured) ?
629 			CPU_STATE_STANDBY : CPU_STATE_CONFIGURED;
630 		smp_cpu_set_polarization(cpu, POLARIZATION_UNKNOWN);
631 		set_cpu_present(cpu, true);
632 		if (sysfs_add && smp_add_present_cpu(cpu) != 0)
633 			set_cpu_present(cpu, false);
634 		else
635 			nr++;
636 		cpu = cpumask_next(cpu, &avail);
637 	}
638 	return nr;
639 }
640 
641 static void __init smp_detect_cpus(void)
642 {
643 	unsigned int cpu, c_cpus, s_cpus;
644 	struct sclp_cpu_info *info;
645 
646 	info = smp_get_cpu_info();
647 	if (!info)
648 		panic("smp_detect_cpus failed to allocate memory\n");
649 	if (info->has_cpu_type) {
650 		for (cpu = 0; cpu < info->combined; cpu++) {
651 			if (info->cpu[cpu].address != boot_cpu_address)
652 				continue;
653 			/* The boot cpu dictates the cpu type. */
654 			boot_cpu_type = info->cpu[cpu].type;
655 			break;
656 		}
657 	}
658 	c_cpus = s_cpus = 0;
659 	for (cpu = 0; cpu < info->combined; cpu++) {
660 		if (info->has_cpu_type && info->cpu[cpu].type != boot_cpu_type)
661 			continue;
662 		if (cpu < info->configured) {
663 			smp_get_save_area(c_cpus, info->cpu[cpu].address);
664 			c_cpus++;
665 		} else
666 			s_cpus++;
667 	}
668 	pr_info("%d configured CPUs, %d standby CPUs\n", c_cpus, s_cpus);
669 	get_online_cpus();
670 	__smp_rescan_cpus(info, 0);
671 	put_online_cpus();
672 	kfree(info);
673 }
674 
675 /*
676  *	Activate a secondary processor.
677  */
678 static void smp_start_secondary(void *cpuvoid)
679 {
680 	S390_lowcore.last_update_clock = get_tod_clock();
681 	S390_lowcore.restart_stack = (unsigned long) restart_stack;
682 	S390_lowcore.restart_fn = (unsigned long) do_restart;
683 	S390_lowcore.restart_data = 0;
684 	S390_lowcore.restart_source = -1UL;
685 	restore_access_regs(S390_lowcore.access_regs_save_area);
686 	__ctl_load(S390_lowcore.cregs_save_area, 0, 15);
687 	__load_psw_mask(PSW_KERNEL_BITS | PSW_MASK_DAT);
688 	cpu_init();
689 	preempt_disable();
690 	init_cpu_timer();
691 	init_cpu_vtimer();
692 	pfault_init();
693 	notify_cpu_starting(smp_processor_id());
694 	set_cpu_online(smp_processor_id(), true);
695 	inc_irq_stat(CPU_RST);
696 	local_irq_enable();
697 	cpu_startup_entry(CPUHP_ONLINE);
698 }
699 
700 /* Upping and downing of CPUs */
701 int __cpu_up(unsigned int cpu, struct task_struct *tidle)
702 {
703 	struct pcpu *pcpu;
704 	int rc;
705 
706 	pcpu = pcpu_devices + cpu;
707 	if (pcpu->state != CPU_STATE_CONFIGURED)
708 		return -EIO;
709 	if (pcpu_sigp_retry(pcpu, SIGP_INITIAL_CPU_RESET, 0) !=
710 	    SIGP_CC_ORDER_CODE_ACCEPTED)
711 		return -EIO;
712 
713 	rc = pcpu_alloc_lowcore(pcpu, cpu);
714 	if (rc)
715 		return rc;
716 	pcpu_prepare_secondary(pcpu, cpu);
717 	pcpu_attach_task(pcpu, tidle);
718 	pcpu_start_fn(pcpu, smp_start_secondary, NULL);
719 	while (!cpu_online(cpu))
720 		cpu_relax();
721 	return 0;
722 }
723 
724 static unsigned int setup_possible_cpus __initdata;
725 
726 static int __init _setup_possible_cpus(char *s)
727 {
728 	get_option(&s, &setup_possible_cpus);
729 	return 0;
730 }
731 early_param("possible_cpus", _setup_possible_cpus);
732 
733 #ifdef CONFIG_HOTPLUG_CPU
734 
735 int __cpu_disable(void)
736 {
737 	unsigned long cregs[16];
738 
739 	/* Handle possible pending IPIs */
740 	smp_handle_ext_call();
741 	set_cpu_online(smp_processor_id(), false);
742 	/* Disable pseudo page faults on this cpu. */
743 	pfault_fini();
744 	/* Disable interrupt sources via control register. */
745 	__ctl_store(cregs, 0, 15);
746 	cregs[0]  &= ~0x0000ee70UL;	/* disable all external interrupts */
747 	cregs[6]  &= ~0xff000000UL;	/* disable all I/O interrupts */
748 	cregs[14] &= ~0x1f000000UL;	/* disable most machine checks */
749 	__ctl_load(cregs, 0, 15);
750 	return 0;
751 }
752 
753 void __cpu_die(unsigned int cpu)
754 {
755 	struct pcpu *pcpu;
756 
757 	/* Wait until target cpu is down */
758 	pcpu = pcpu_devices + cpu;
759 	while (!pcpu_stopped(pcpu))
760 		cpu_relax();
761 	pcpu_free_lowcore(pcpu);
762 	atomic_dec(&init_mm.context.attach_count);
763 }
764 
765 void __noreturn cpu_die(void)
766 {
767 	idle_task_exit();
768 	pcpu_sigp_retry(pcpu_devices + smp_processor_id(), SIGP_STOP, 0);
769 	for (;;) ;
770 }
771 
772 #endif /* CONFIG_HOTPLUG_CPU */
773 
774 void __init smp_fill_possible_mask(void)
775 {
776 	unsigned int possible, cpu;
777 
778 	possible = setup_possible_cpus;
779 	if (!possible)
780 		possible = MACHINE_IS_VM ? 64 : nr_cpu_ids;
781 	for (cpu = 0; cpu < possible && cpu < nr_cpu_ids; cpu++)
782 		set_cpu_possible(cpu, true);
783 }
784 
785 void __init smp_prepare_cpus(unsigned int max_cpus)
786 {
787 	/* request the 0x1201 emergency signal external interrupt */
788 	if (register_external_interrupt(0x1201, do_ext_call_interrupt) != 0)
789 		panic("Couldn't request external interrupt 0x1201");
790 	/* request the 0x1202 external call external interrupt */
791 	if (register_external_interrupt(0x1202, do_ext_call_interrupt) != 0)
792 		panic("Couldn't request external interrupt 0x1202");
793 	smp_detect_cpus();
794 }
795 
796 void __init smp_prepare_boot_cpu(void)
797 {
798 	struct pcpu *pcpu = pcpu_devices;
799 
800 	boot_cpu_address = stap();
801 	pcpu->state = CPU_STATE_CONFIGURED;
802 	pcpu->address = boot_cpu_address;
803 	pcpu->lowcore = (struct _lowcore *)(unsigned long) store_prefix();
804 	pcpu->async_stack = S390_lowcore.async_stack - ASYNC_SIZE
805 		+ STACK_FRAME_OVERHEAD + sizeof(struct pt_regs);
806 	pcpu->panic_stack = S390_lowcore.panic_stack - PAGE_SIZE
807 		+ STACK_FRAME_OVERHEAD + sizeof(struct pt_regs);
808 	S390_lowcore.percpu_offset = __per_cpu_offset[0];
809 	smp_cpu_set_polarization(0, POLARIZATION_UNKNOWN);
810 	set_cpu_present(0, true);
811 	set_cpu_online(0, true);
812 }
813 
814 void __init smp_cpus_done(unsigned int max_cpus)
815 {
816 }
817 
818 void __init smp_setup_processor_id(void)
819 {
820 	S390_lowcore.cpu_nr = 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 ssize_t show_idle_count(struct device *dev,
919 				struct device_attribute *attr, char *buf)
920 {
921 	struct s390_idle_data *idle = &per_cpu(s390_idle, dev->id);
922 	unsigned long long idle_count;
923 	unsigned int sequence;
924 
925 	do {
926 		sequence = ACCESS_ONCE(idle->sequence);
927 		idle_count = ACCESS_ONCE(idle->idle_count);
928 		if (ACCESS_ONCE(idle->clock_idle_enter))
929 			idle_count++;
930 	} while ((sequence & 1) || (ACCESS_ONCE(idle->sequence) != sequence));
931 	return sprintf(buf, "%llu\n", idle_count);
932 }
933 static DEVICE_ATTR(idle_count, 0444, show_idle_count, NULL);
934 
935 static ssize_t show_idle_time(struct device *dev,
936 				struct device_attribute *attr, char *buf)
937 {
938 	struct s390_idle_data *idle = &per_cpu(s390_idle, dev->id);
939 	unsigned long long now, idle_time, idle_enter, idle_exit;
940 	unsigned int sequence;
941 
942 	do {
943 		now = get_tod_clock();
944 		sequence = ACCESS_ONCE(idle->sequence);
945 		idle_time = ACCESS_ONCE(idle->idle_time);
946 		idle_enter = ACCESS_ONCE(idle->clock_idle_enter);
947 		idle_exit = ACCESS_ONCE(idle->clock_idle_exit);
948 	} while ((sequence & 1) || (ACCESS_ONCE(idle->sequence) != sequence));
949 	idle_time += idle_enter ? ((idle_exit ? : now) - idle_enter) : 0;
950 	return sprintf(buf, "%llu\n", idle_time >> 12);
951 }
952 static DEVICE_ATTR(idle_time_us, 0444, show_idle_time, NULL);
953 
954 static struct attribute *cpu_online_attrs[] = {
955 	&dev_attr_idle_count.attr,
956 	&dev_attr_idle_time_us.attr,
957 	NULL,
958 };
959 
960 static struct attribute_group cpu_online_attr_group = {
961 	.attrs = cpu_online_attrs,
962 };
963 
964 static int smp_cpu_notify(struct notifier_block *self, unsigned long action,
965 			  void *hcpu)
966 {
967 	unsigned int cpu = (unsigned int)(long)hcpu;
968 	struct cpu *c = pcpu_devices[cpu].cpu;
969 	struct device *s = &c->dev;
970 	int err = 0;
971 
972 	switch (action & ~CPU_TASKS_FROZEN) {
973 	case CPU_ONLINE:
974 		err = sysfs_create_group(&s->kobj, &cpu_online_attr_group);
975 		break;
976 	case CPU_DEAD:
977 		sysfs_remove_group(&s->kobj, &cpu_online_attr_group);
978 		break;
979 	}
980 	return notifier_from_errno(err);
981 }
982 
983 static int smp_add_present_cpu(int cpu)
984 {
985 	struct device *s;
986 	struct cpu *c;
987 	int rc;
988 
989 	c = kzalloc(sizeof(*c), GFP_KERNEL);
990 	if (!c)
991 		return -ENOMEM;
992 	pcpu_devices[cpu].cpu = c;
993 	s = &c->dev;
994 	c->hotpluggable = 1;
995 	rc = register_cpu(c, cpu);
996 	if (rc)
997 		goto out;
998 	rc = sysfs_create_group(&s->kobj, &cpu_common_attr_group);
999 	if (rc)
1000 		goto out_cpu;
1001 	if (cpu_online(cpu)) {
1002 		rc = sysfs_create_group(&s->kobj, &cpu_online_attr_group);
1003 		if (rc)
1004 			goto out_online;
1005 	}
1006 	rc = topology_cpu_init(c);
1007 	if (rc)
1008 		goto out_topology;
1009 	return 0;
1010 
1011 out_topology:
1012 	if (cpu_online(cpu))
1013 		sysfs_remove_group(&s->kobj, &cpu_online_attr_group);
1014 out_online:
1015 	sysfs_remove_group(&s->kobj, &cpu_common_attr_group);
1016 out_cpu:
1017 #ifdef CONFIG_HOTPLUG_CPU
1018 	unregister_cpu(c);
1019 #endif
1020 out:
1021 	return rc;
1022 }
1023 
1024 #ifdef CONFIG_HOTPLUG_CPU
1025 
1026 int __ref smp_rescan_cpus(void)
1027 {
1028 	struct sclp_cpu_info *info;
1029 	int nr;
1030 
1031 	info = smp_get_cpu_info();
1032 	if (!info)
1033 		return -ENOMEM;
1034 	get_online_cpus();
1035 	mutex_lock(&smp_cpu_state_mutex);
1036 	nr = __smp_rescan_cpus(info, 1);
1037 	mutex_unlock(&smp_cpu_state_mutex);
1038 	put_online_cpus();
1039 	kfree(info);
1040 	if (nr)
1041 		topology_schedule_update();
1042 	return 0;
1043 }
1044 
1045 static ssize_t __ref rescan_store(struct device *dev,
1046 				  struct device_attribute *attr,
1047 				  const char *buf,
1048 				  size_t count)
1049 {
1050 	int rc;
1051 
1052 	rc = smp_rescan_cpus();
1053 	return rc ? rc : count;
1054 }
1055 static DEVICE_ATTR(rescan, 0200, NULL, rescan_store);
1056 #endif /* CONFIG_HOTPLUG_CPU */
1057 
1058 static int __init s390_smp_init(void)
1059 {
1060 	int cpu, rc;
1061 
1062 	hotcpu_notifier(smp_cpu_notify, 0);
1063 #ifdef CONFIG_HOTPLUG_CPU
1064 	rc = device_create_file(cpu_subsys.dev_root, &dev_attr_rescan);
1065 	if (rc)
1066 		return rc;
1067 #endif
1068 	for_each_present_cpu(cpu) {
1069 		rc = smp_add_present_cpu(cpu);
1070 		if (rc)
1071 			return rc;
1072 	}
1073 	return 0;
1074 }
1075 subsys_initcall(s390_smp_init);
1076