xref: /linux/arch/s390/kernel/smp.c (revision e5c86679d5e864947a52fb31e45a425dea3e7fa9)
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/bootmem.h>
23 #include <linux/export.h>
24 #include <linux/init.h>
25 #include <linux/mm.h>
26 #include <linux/err.h>
27 #include <linux/spinlock.h>
28 #include <linux/kernel_stat.h>
29 #include <linux/delay.h>
30 #include <linux/interrupt.h>
31 #include <linux/irqflags.h>
32 #include <linux/cpu.h>
33 #include <linux/slab.h>
34 #include <linux/sched/hotplug.h>
35 #include <linux/sched/task_stack.h>
36 #include <linux/crash_dump.h>
37 #include <linux/memblock.h>
38 #include <asm/asm-offsets.h>
39 #include <asm/diag.h>
40 #include <asm/switch_to.h>
41 #include <asm/facility.h>
42 #include <asm/ipl.h>
43 #include <asm/setup.h>
44 #include <asm/irq.h>
45 #include <asm/tlbflush.h>
46 #include <asm/vtimer.h>
47 #include <asm/lowcore.h>
48 #include <asm/sclp.h>
49 #include <asm/vdso.h>
50 #include <asm/debug.h>
51 #include <asm/os_info.h>
52 #include <asm/sigp.h>
53 #include <asm/idle.h>
54 #include "entry.h"
55 
56 enum {
57 	ec_schedule = 0,
58 	ec_call_function_single,
59 	ec_stop_cpu,
60 };
61 
62 enum {
63 	CPU_STATE_STANDBY,
64 	CPU_STATE_CONFIGURED,
65 };
66 
67 static DEFINE_PER_CPU(struct cpu *, cpu_device);
68 
69 struct pcpu {
70 	struct lowcore *lowcore;	/* lowcore page(s) for the cpu */
71 	unsigned long ec_mask;		/* bit mask for ec_xxx functions */
72 	unsigned long ec_clk;		/* sigp timestamp for ec_xxx */
73 	signed char state;		/* physical cpu state */
74 	signed char polarization;	/* physical polarization */
75 	u16 address;			/* physical cpu address */
76 };
77 
78 static u8 boot_core_type;
79 static struct pcpu pcpu_devices[NR_CPUS];
80 
81 unsigned int smp_cpu_mt_shift;
82 EXPORT_SYMBOL(smp_cpu_mt_shift);
83 
84 unsigned int smp_cpu_mtid;
85 EXPORT_SYMBOL(smp_cpu_mtid);
86 
87 #ifdef CONFIG_CRASH_DUMP
88 __vector128 __initdata boot_cpu_vector_save_area[__NUM_VXRS];
89 #endif
90 
91 static unsigned int smp_max_threads __initdata = -1U;
92 
93 static int __init early_nosmt(char *s)
94 {
95 	smp_max_threads = 1;
96 	return 0;
97 }
98 early_param("nosmt", early_nosmt);
99 
100 static int __init early_smt(char *s)
101 {
102 	get_option(&s, &smp_max_threads);
103 	return 0;
104 }
105 early_param("smt", early_smt);
106 
107 /*
108  * The smp_cpu_state_mutex must be held when changing the state or polarization
109  * member of a pcpu data structure within the pcpu_devices arreay.
110  */
111 DEFINE_MUTEX(smp_cpu_state_mutex);
112 
113 /*
114  * Signal processor helper functions.
115  */
116 static inline int __pcpu_sigp_relax(u16 addr, u8 order, unsigned long parm)
117 {
118 	int cc;
119 
120 	while (1) {
121 		cc = __pcpu_sigp(addr, order, parm, NULL);
122 		if (cc != SIGP_CC_BUSY)
123 			return cc;
124 		cpu_relax();
125 	}
126 }
127 
128 static int pcpu_sigp_retry(struct pcpu *pcpu, u8 order, u32 parm)
129 {
130 	int cc, retry;
131 
132 	for (retry = 0; ; retry++) {
133 		cc = __pcpu_sigp(pcpu->address, order, parm, NULL);
134 		if (cc != SIGP_CC_BUSY)
135 			break;
136 		if (retry >= 3)
137 			udelay(10);
138 	}
139 	return cc;
140 }
141 
142 static inline int pcpu_stopped(struct pcpu *pcpu)
143 {
144 	u32 uninitialized_var(status);
145 
146 	if (__pcpu_sigp(pcpu->address, SIGP_SENSE,
147 			0, &status) != SIGP_CC_STATUS_STORED)
148 		return 0;
149 	return !!(status & (SIGP_STATUS_CHECK_STOP|SIGP_STATUS_STOPPED));
150 }
151 
152 static inline int pcpu_running(struct pcpu *pcpu)
153 {
154 	if (__pcpu_sigp(pcpu->address, SIGP_SENSE_RUNNING,
155 			0, NULL) != SIGP_CC_STATUS_STORED)
156 		return 1;
157 	/* Status stored condition code is equivalent to cpu not running. */
158 	return 0;
159 }
160 
161 /*
162  * Find struct pcpu by cpu address.
163  */
164 static struct pcpu *pcpu_find_address(const struct cpumask *mask, u16 address)
165 {
166 	int cpu;
167 
168 	for_each_cpu(cpu, mask)
169 		if (pcpu_devices[cpu].address == address)
170 			return pcpu_devices + cpu;
171 	return NULL;
172 }
173 
174 static void pcpu_ec_call(struct pcpu *pcpu, int ec_bit)
175 {
176 	int order;
177 
178 	if (test_and_set_bit(ec_bit, &pcpu->ec_mask))
179 		return;
180 	order = pcpu_running(pcpu) ? SIGP_EXTERNAL_CALL : SIGP_EMERGENCY_SIGNAL;
181 	pcpu->ec_clk = get_tod_clock_fast();
182 	pcpu_sigp_retry(pcpu, order, 0);
183 }
184 
185 #define ASYNC_FRAME_OFFSET (ASYNC_SIZE - STACK_FRAME_OVERHEAD - __PT_SIZE)
186 #define PANIC_FRAME_OFFSET (PAGE_SIZE - STACK_FRAME_OVERHEAD - __PT_SIZE)
187 
188 static int pcpu_alloc_lowcore(struct pcpu *pcpu, int cpu)
189 {
190 	unsigned long async_stack, panic_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 		async_stack = __get_free_pages(GFP_KERNEL, ASYNC_ORDER);
197 		panic_stack = __get_free_page(GFP_KERNEL);
198 		if (!pcpu->lowcore || !panic_stack || !async_stack)
199 			goto out;
200 	} else {
201 		async_stack = pcpu->lowcore->async_stack - ASYNC_FRAME_OFFSET;
202 		panic_stack = pcpu->lowcore->panic_stack - PANIC_FRAME_OFFSET;
203 	}
204 	lc = pcpu->lowcore;
205 	memcpy(lc, &S390_lowcore, 512);
206 	memset((char *) lc + 512, 0, sizeof(*lc) - 512);
207 	lc->async_stack = async_stack + ASYNC_FRAME_OFFSET;
208 	lc->panic_stack = panic_stack + PANIC_FRAME_OFFSET;
209 	lc->cpu_nr = cpu;
210 	lc->spinlock_lockval = arch_spin_lockval(cpu);
211 	if (MACHINE_HAS_VX)
212 		lc->vector_save_area_addr =
213 			(unsigned long) &lc->vector_save_area;
214 	if (vdso_alloc_per_cpu(lc))
215 		goto out;
216 	lowcore_ptr[cpu] = lc;
217 	pcpu_sigp_retry(pcpu, SIGP_SET_PREFIX, (u32)(unsigned long) lc);
218 	return 0;
219 out:
220 	if (pcpu != &pcpu_devices[0]) {
221 		free_page(panic_stack);
222 		free_pages(async_stack, ASYNC_ORDER);
223 		free_pages((unsigned long) pcpu->lowcore, LC_ORDER);
224 	}
225 	return -ENOMEM;
226 }
227 
228 #ifdef CONFIG_HOTPLUG_CPU
229 
230 static void pcpu_free_lowcore(struct pcpu *pcpu)
231 {
232 	pcpu_sigp_retry(pcpu, SIGP_SET_PREFIX, 0);
233 	lowcore_ptr[pcpu - pcpu_devices] = NULL;
234 	vdso_free_per_cpu(pcpu->lowcore);
235 	if (pcpu == &pcpu_devices[0])
236 		return;
237 	free_page(pcpu->lowcore->panic_stack-PANIC_FRAME_OFFSET);
238 	free_pages(pcpu->lowcore->async_stack-ASYNC_FRAME_OFFSET, ASYNC_ORDER);
239 	free_pages((unsigned long) pcpu->lowcore, LC_ORDER);
240 }
241 
242 #endif /* CONFIG_HOTPLUG_CPU */
243 
244 static void pcpu_prepare_secondary(struct pcpu *pcpu, int cpu)
245 {
246 	struct lowcore *lc = pcpu->lowcore;
247 
248 	cpumask_set_cpu(cpu, &init_mm.context.cpu_attach_mask);
249 	cpumask_set_cpu(cpu, mm_cpumask(&init_mm));
250 	lc->cpu_nr = cpu;
251 	lc->spinlock_lockval = arch_spin_lockval(cpu);
252 	lc->percpu_offset = __per_cpu_offset[cpu];
253 	lc->kernel_asce = S390_lowcore.kernel_asce;
254 	lc->machine_flags = S390_lowcore.machine_flags;
255 	lc->user_timer = lc->system_timer = lc->steal_timer = 0;
256 	__ctl_store(lc->cregs_save_area, 0, 15);
257 	save_access_regs((unsigned int *) lc->access_regs_save_area);
258 	memcpy(lc->stfle_fac_list, S390_lowcore.stfle_fac_list,
259 	       MAX_FACILITY_BIT/8);
260 }
261 
262 static void pcpu_attach_task(struct pcpu *pcpu, struct task_struct *tsk)
263 {
264 	struct lowcore *lc = pcpu->lowcore;
265 
266 	lc->kernel_stack = (unsigned long) task_stack_page(tsk)
267 		+ THREAD_SIZE - STACK_FRAME_OVERHEAD - sizeof(struct pt_regs);
268 	lc->current_task = (unsigned long) tsk;
269 	lc->lpp = LPP_MAGIC;
270 	lc->current_pid = tsk->pid;
271 	lc->user_timer = tsk->thread.user_timer;
272 	lc->system_timer = tsk->thread.system_timer;
273 	lc->steal_timer = 0;
274 }
275 
276 static void pcpu_start_fn(struct pcpu *pcpu, void (*func)(void *), void *data)
277 {
278 	struct lowcore *lc = pcpu->lowcore;
279 
280 	lc->restart_stack = lc->kernel_stack;
281 	lc->restart_fn = (unsigned long) func;
282 	lc->restart_data = (unsigned long) data;
283 	lc->restart_source = -1UL;
284 	pcpu_sigp_retry(pcpu, SIGP_RESTART, 0);
285 }
286 
287 /*
288  * Call function via PSW restart on pcpu and stop the current cpu.
289  */
290 static void pcpu_delegate(struct pcpu *pcpu, void (*func)(void *),
291 			  void *data, unsigned long stack)
292 {
293 	struct lowcore *lc = lowcore_ptr[pcpu - pcpu_devices];
294 	unsigned long source_cpu = stap();
295 
296 	__load_psw_mask(PSW_KERNEL_BITS);
297 	if (pcpu->address == source_cpu)
298 		func(data);	/* should not return */
299 	/* Stop target cpu (if func returns this stops the current cpu). */
300 	pcpu_sigp_retry(pcpu, SIGP_STOP, 0);
301 	/* Restart func on the target cpu and stop the current cpu. */
302 	mem_assign_absolute(lc->restart_stack, stack);
303 	mem_assign_absolute(lc->restart_fn, (unsigned long) func);
304 	mem_assign_absolute(lc->restart_data, (unsigned long) data);
305 	mem_assign_absolute(lc->restart_source, source_cpu);
306 	asm volatile(
307 		"0:	sigp	0,%0,%2	# sigp restart to target cpu\n"
308 		"	brc	2,0b	# busy, try again\n"
309 		"1:	sigp	0,%1,%3	# sigp stop to current cpu\n"
310 		"	brc	2,1b	# busy, try again\n"
311 		: : "d" (pcpu->address), "d" (source_cpu),
312 		    "K" (SIGP_RESTART), "K" (SIGP_STOP)
313 		: "0", "1", "cc");
314 	for (;;) ;
315 }
316 
317 /*
318  * Enable additional logical cpus for multi-threading.
319  */
320 static int pcpu_set_smt(unsigned int mtid)
321 {
322 	int cc;
323 
324 	if (smp_cpu_mtid == mtid)
325 		return 0;
326 	cc = __pcpu_sigp(0, SIGP_SET_MULTI_THREADING, mtid, NULL);
327 	if (cc == 0) {
328 		smp_cpu_mtid = mtid;
329 		smp_cpu_mt_shift = 0;
330 		while (smp_cpu_mtid >= (1U << smp_cpu_mt_shift))
331 			smp_cpu_mt_shift++;
332 		pcpu_devices[0].address = stap();
333 	}
334 	return cc;
335 }
336 
337 /*
338  * Call function on an online CPU.
339  */
340 void smp_call_online_cpu(void (*func)(void *), void *data)
341 {
342 	struct pcpu *pcpu;
343 
344 	/* Use the current cpu if it is online. */
345 	pcpu = pcpu_find_address(cpu_online_mask, stap());
346 	if (!pcpu)
347 		/* Use the first online cpu. */
348 		pcpu = pcpu_devices + cpumask_first(cpu_online_mask);
349 	pcpu_delegate(pcpu, func, data, (unsigned long) restart_stack);
350 }
351 
352 /*
353  * Call function on the ipl CPU.
354  */
355 void smp_call_ipl_cpu(void (*func)(void *), void *data)
356 {
357 	pcpu_delegate(&pcpu_devices[0], func, data,
358 		      pcpu_devices->lowcore->panic_stack -
359 		      PANIC_FRAME_OFFSET + PAGE_SIZE);
360 }
361 
362 int smp_find_processor_id(u16 address)
363 {
364 	int cpu;
365 
366 	for_each_present_cpu(cpu)
367 		if (pcpu_devices[cpu].address == address)
368 			return cpu;
369 	return -1;
370 }
371 
372 bool arch_vcpu_is_preempted(int cpu)
373 {
374 	if (test_cpu_flag_of(CIF_ENABLED_WAIT, cpu))
375 		return false;
376 	if (pcpu_running(pcpu_devices + cpu))
377 		return false;
378 	return true;
379 }
380 EXPORT_SYMBOL(arch_vcpu_is_preempted);
381 
382 void smp_yield_cpu(int cpu)
383 {
384 	if (MACHINE_HAS_DIAG9C) {
385 		diag_stat_inc_norecursion(DIAG_STAT_X09C);
386 		asm volatile("diag %0,0,0x9c"
387 			     : : "d" (pcpu_devices[cpu].address));
388 	} else if (MACHINE_HAS_DIAG44) {
389 		diag_stat_inc_norecursion(DIAG_STAT_X044);
390 		asm volatile("diag 0,0,0x44");
391 	}
392 }
393 
394 /*
395  * Send cpus emergency shutdown signal. This gives the cpus the
396  * opportunity to complete outstanding interrupts.
397  */
398 static void smp_emergency_stop(cpumask_t *cpumask)
399 {
400 	u64 end;
401 	int cpu;
402 
403 	end = get_tod_clock() + (1000000UL << 12);
404 	for_each_cpu(cpu, cpumask) {
405 		struct pcpu *pcpu = pcpu_devices + cpu;
406 		set_bit(ec_stop_cpu, &pcpu->ec_mask);
407 		while (__pcpu_sigp(pcpu->address, SIGP_EMERGENCY_SIGNAL,
408 				   0, NULL) == SIGP_CC_BUSY &&
409 		       get_tod_clock() < end)
410 			cpu_relax();
411 	}
412 	while (get_tod_clock() < end) {
413 		for_each_cpu(cpu, cpumask)
414 			if (pcpu_stopped(pcpu_devices + cpu))
415 				cpumask_clear_cpu(cpu, cpumask);
416 		if (cpumask_empty(cpumask))
417 			break;
418 		cpu_relax();
419 	}
420 }
421 
422 /*
423  * Stop all cpus but the current one.
424  */
425 void smp_send_stop(void)
426 {
427 	cpumask_t cpumask;
428 	int cpu;
429 
430 	/* Disable all interrupts/machine checks */
431 	__load_psw_mask(PSW_KERNEL_BITS | PSW_MASK_DAT);
432 	trace_hardirqs_off();
433 
434 	debug_set_critical();
435 	cpumask_copy(&cpumask, cpu_online_mask);
436 	cpumask_clear_cpu(smp_processor_id(), &cpumask);
437 
438 	if (oops_in_progress)
439 		smp_emergency_stop(&cpumask);
440 
441 	/* stop all processors */
442 	for_each_cpu(cpu, &cpumask) {
443 		struct pcpu *pcpu = pcpu_devices + cpu;
444 		pcpu_sigp_retry(pcpu, SIGP_STOP, 0);
445 		while (!pcpu_stopped(pcpu))
446 			cpu_relax();
447 	}
448 }
449 
450 /*
451  * This is the main routine where commands issued by other
452  * cpus are handled.
453  */
454 static void smp_handle_ext_call(void)
455 {
456 	unsigned long bits;
457 
458 	/* handle bit signal external calls */
459 	bits = xchg(&pcpu_devices[smp_processor_id()].ec_mask, 0);
460 	if (test_bit(ec_stop_cpu, &bits))
461 		smp_stop_cpu();
462 	if (test_bit(ec_schedule, &bits))
463 		scheduler_ipi();
464 	if (test_bit(ec_call_function_single, &bits))
465 		generic_smp_call_function_single_interrupt();
466 }
467 
468 static void do_ext_call_interrupt(struct ext_code ext_code,
469 				  unsigned int param32, unsigned long param64)
470 {
471 	inc_irq_stat(ext_code.code == 0x1202 ? IRQEXT_EXC : IRQEXT_EMS);
472 	smp_handle_ext_call();
473 }
474 
475 void arch_send_call_function_ipi_mask(const struct cpumask *mask)
476 {
477 	int cpu;
478 
479 	for_each_cpu(cpu, mask)
480 		pcpu_ec_call(pcpu_devices + cpu, ec_call_function_single);
481 }
482 
483 void arch_send_call_function_single_ipi(int cpu)
484 {
485 	pcpu_ec_call(pcpu_devices + cpu, ec_call_function_single);
486 }
487 
488 /*
489  * this function sends a 'reschedule' IPI to another CPU.
490  * it goes straight through and wastes no time serializing
491  * anything. Worst case is that we lose a reschedule ...
492  */
493 void smp_send_reschedule(int cpu)
494 {
495 	pcpu_ec_call(pcpu_devices + cpu, ec_schedule);
496 }
497 
498 /*
499  * parameter area for the set/clear control bit callbacks
500  */
501 struct ec_creg_mask_parms {
502 	unsigned long orval;
503 	unsigned long andval;
504 	int cr;
505 };
506 
507 /*
508  * callback for setting/clearing control bits
509  */
510 static void smp_ctl_bit_callback(void *info)
511 {
512 	struct ec_creg_mask_parms *pp = info;
513 	unsigned long cregs[16];
514 
515 	__ctl_store(cregs, 0, 15);
516 	cregs[pp->cr] = (cregs[pp->cr] & pp->andval) | pp->orval;
517 	__ctl_load(cregs, 0, 15);
518 }
519 
520 /*
521  * Set a bit in a control register of all cpus
522  */
523 void smp_ctl_set_bit(int cr, int bit)
524 {
525 	struct ec_creg_mask_parms parms = { 1UL << bit, -1UL, cr };
526 
527 	on_each_cpu(smp_ctl_bit_callback, &parms, 1);
528 }
529 EXPORT_SYMBOL(smp_ctl_set_bit);
530 
531 /*
532  * Clear a bit in a control register of all cpus
533  */
534 void smp_ctl_clear_bit(int cr, int bit)
535 {
536 	struct ec_creg_mask_parms parms = { 0, ~(1UL << bit), cr };
537 
538 	on_each_cpu(smp_ctl_bit_callback, &parms, 1);
539 }
540 EXPORT_SYMBOL(smp_ctl_clear_bit);
541 
542 #ifdef CONFIG_CRASH_DUMP
543 
544 int smp_store_status(int cpu)
545 {
546 	struct pcpu *pcpu = pcpu_devices + cpu;
547 	unsigned long pa;
548 
549 	pa = __pa(&pcpu->lowcore->floating_pt_save_area);
550 	if (__pcpu_sigp_relax(pcpu->address, SIGP_STORE_STATUS_AT_ADDRESS,
551 			      pa) != SIGP_CC_ORDER_CODE_ACCEPTED)
552 		return -EIO;
553 	if (!MACHINE_HAS_VX)
554 		return 0;
555 	pa = __pa(pcpu->lowcore->vector_save_area_addr);
556 	if (__pcpu_sigp_relax(pcpu->address, SIGP_STORE_ADDITIONAL_STATUS,
557 			      pa) != SIGP_CC_ORDER_CODE_ACCEPTED)
558 		return -EIO;
559 	return 0;
560 }
561 
562 /*
563  * Collect CPU state of the previous, crashed system.
564  * There are four cases:
565  * 1) standard zfcp dump
566  *    condition: OLDMEM_BASE == NULL && ipl_info.type == IPL_TYPE_FCP_DUMP
567  *    The state for all CPUs except the boot CPU needs to be collected
568  *    with sigp stop-and-store-status. The boot CPU state is located in
569  *    the absolute lowcore of the memory stored in the HSA. The zcore code
570  *    will copy the boot CPU state from the HSA.
571  * 2) stand-alone kdump for SCSI (zfcp dump with swapped memory)
572  *    condition: OLDMEM_BASE != NULL && ipl_info.type == IPL_TYPE_FCP_DUMP
573  *    The state for all CPUs except the boot CPU needs to be collected
574  *    with sigp stop-and-store-status. The firmware or the boot-loader
575  *    stored the registers of the boot CPU in the absolute lowcore in the
576  *    memory of the old system.
577  * 3) kdump and the old kernel did not store the CPU state,
578  *    or stand-alone kdump for DASD
579  *    condition: OLDMEM_BASE != NULL && !is_kdump_kernel()
580  *    The state for all CPUs except the boot CPU needs to be collected
581  *    with sigp stop-and-store-status. The kexec code or the boot-loader
582  *    stored the registers of the boot CPU in the memory of the old system.
583  * 4) kdump and the old kernel stored the CPU state
584  *    condition: OLDMEM_BASE != NULL && is_kdump_kernel()
585  *    This case does not exist for s390 anymore, setup_arch explicitly
586  *    deactivates the elfcorehdr= kernel parameter
587  */
588 static __init void smp_save_cpu_vxrs(struct save_area *sa, u16 addr,
589 				     bool is_boot_cpu, unsigned long page)
590 {
591 	__vector128 *vxrs = (__vector128 *) page;
592 
593 	if (is_boot_cpu)
594 		vxrs = boot_cpu_vector_save_area;
595 	else
596 		__pcpu_sigp_relax(addr, SIGP_STORE_ADDITIONAL_STATUS, page);
597 	save_area_add_vxrs(sa, vxrs);
598 }
599 
600 static __init void smp_save_cpu_regs(struct save_area *sa, u16 addr,
601 				     bool is_boot_cpu, unsigned long page)
602 {
603 	void *regs = (void *) page;
604 
605 	if (is_boot_cpu)
606 		copy_oldmem_kernel(regs, (void *) __LC_FPREGS_SAVE_AREA, 512);
607 	else
608 		__pcpu_sigp_relax(addr, SIGP_STORE_STATUS_AT_ADDRESS, page);
609 	save_area_add_regs(sa, regs);
610 }
611 
612 void __init smp_save_dump_cpus(void)
613 {
614 	int addr, boot_cpu_addr, max_cpu_addr;
615 	struct save_area *sa;
616 	unsigned long page;
617 	bool is_boot_cpu;
618 
619 	if (!(OLDMEM_BASE || ipl_info.type == IPL_TYPE_FCP_DUMP))
620 		/* No previous system present, normal boot. */
621 		return;
622 	/* Allocate a page as dumping area for the store status sigps */
623 	page = memblock_alloc_base(PAGE_SIZE, PAGE_SIZE, 1UL << 31);
624 	/* Set multi-threading state to the previous system. */
625 	pcpu_set_smt(sclp.mtid_prev);
626 	boot_cpu_addr = stap();
627 	max_cpu_addr = SCLP_MAX_CORES << sclp.mtid_prev;
628 	for (addr = 0; addr <= max_cpu_addr; addr++) {
629 		if (__pcpu_sigp_relax(addr, SIGP_SENSE, 0) ==
630 		    SIGP_CC_NOT_OPERATIONAL)
631 			continue;
632 		is_boot_cpu = (addr == boot_cpu_addr);
633 		/* Allocate save area */
634 		sa = save_area_alloc(is_boot_cpu);
635 		if (!sa)
636 			panic("could not allocate memory for save area\n");
637 		if (MACHINE_HAS_VX)
638 			/* Get the vector registers */
639 			smp_save_cpu_vxrs(sa, addr, is_boot_cpu, page);
640 		/*
641 		 * For a zfcp dump OLDMEM_BASE == NULL and the registers
642 		 * of the boot CPU are stored in the HSA. To retrieve
643 		 * these registers an SCLP request is required which is
644 		 * done by drivers/s390/char/zcore.c:init_cpu_info()
645 		 */
646 		if (!is_boot_cpu || OLDMEM_BASE)
647 			/* Get the CPU registers */
648 			smp_save_cpu_regs(sa, addr, is_boot_cpu, page);
649 	}
650 	memblock_free(page, PAGE_SIZE);
651 	diag308_reset();
652 	pcpu_set_smt(0);
653 }
654 #endif /* CONFIG_CRASH_DUMP */
655 
656 void smp_cpu_set_polarization(int cpu, int val)
657 {
658 	pcpu_devices[cpu].polarization = val;
659 }
660 
661 int smp_cpu_get_polarization(int cpu)
662 {
663 	return pcpu_devices[cpu].polarization;
664 }
665 
666 static void __ref smp_get_core_info(struct sclp_core_info *info, int early)
667 {
668 	static int use_sigp_detection;
669 	int address;
670 
671 	if (use_sigp_detection || sclp_get_core_info(info, early)) {
672 		use_sigp_detection = 1;
673 		for (address = 0;
674 		     address < (SCLP_MAX_CORES << smp_cpu_mt_shift);
675 		     address += (1U << smp_cpu_mt_shift)) {
676 			if (__pcpu_sigp_relax(address, SIGP_SENSE, 0) ==
677 			    SIGP_CC_NOT_OPERATIONAL)
678 				continue;
679 			info->core[info->configured].core_id =
680 				address >> smp_cpu_mt_shift;
681 			info->configured++;
682 		}
683 		info->combined = info->configured;
684 	}
685 }
686 
687 static int smp_add_present_cpu(int cpu);
688 
689 static int __smp_rescan_cpus(struct sclp_core_info *info, int sysfs_add)
690 {
691 	struct pcpu *pcpu;
692 	cpumask_t avail;
693 	int cpu, nr, i, j;
694 	u16 address;
695 
696 	nr = 0;
697 	cpumask_xor(&avail, cpu_possible_mask, cpu_present_mask);
698 	cpu = cpumask_first(&avail);
699 	for (i = 0; (i < info->combined) && (cpu < nr_cpu_ids); i++) {
700 		if (sclp.has_core_type && info->core[i].type != boot_core_type)
701 			continue;
702 		address = info->core[i].core_id << smp_cpu_mt_shift;
703 		for (j = 0; j <= smp_cpu_mtid; j++) {
704 			if (pcpu_find_address(cpu_present_mask, address + j))
705 				continue;
706 			pcpu = pcpu_devices + cpu;
707 			pcpu->address = address + j;
708 			pcpu->state =
709 				(cpu >= info->configured*(smp_cpu_mtid + 1)) ?
710 				CPU_STATE_STANDBY : CPU_STATE_CONFIGURED;
711 			smp_cpu_set_polarization(cpu, POLARIZATION_UNKNOWN);
712 			set_cpu_present(cpu, true);
713 			if (sysfs_add && smp_add_present_cpu(cpu) != 0)
714 				set_cpu_present(cpu, false);
715 			else
716 				nr++;
717 			cpu = cpumask_next(cpu, &avail);
718 			if (cpu >= nr_cpu_ids)
719 				break;
720 		}
721 	}
722 	return nr;
723 }
724 
725 void __init smp_detect_cpus(void)
726 {
727 	unsigned int cpu, mtid, c_cpus, s_cpus;
728 	struct sclp_core_info *info;
729 	u16 address;
730 
731 	/* Get CPU information */
732 	info = memblock_virt_alloc(sizeof(*info), 8);
733 	smp_get_core_info(info, 1);
734 	/* Find boot CPU type */
735 	if (sclp.has_core_type) {
736 		address = stap();
737 		for (cpu = 0; cpu < info->combined; cpu++)
738 			if (info->core[cpu].core_id == address) {
739 				/* The boot cpu dictates the cpu type. */
740 				boot_core_type = info->core[cpu].type;
741 				break;
742 			}
743 		if (cpu >= info->combined)
744 			panic("Could not find boot CPU type");
745 	}
746 
747 	/* Set multi-threading state for the current system */
748 	mtid = boot_core_type ? sclp.mtid : sclp.mtid_cp;
749 	mtid = (mtid < smp_max_threads) ? mtid : smp_max_threads - 1;
750 	pcpu_set_smt(mtid);
751 
752 	/* Print number of CPUs */
753 	c_cpus = s_cpus = 0;
754 	for (cpu = 0; cpu < info->combined; cpu++) {
755 		if (sclp.has_core_type &&
756 		    info->core[cpu].type != boot_core_type)
757 			continue;
758 		if (cpu < info->configured)
759 			c_cpus += smp_cpu_mtid + 1;
760 		else
761 			s_cpus += smp_cpu_mtid + 1;
762 	}
763 	pr_info("%d configured CPUs, %d standby CPUs\n", c_cpus, s_cpus);
764 
765 	/* Add CPUs present at boot */
766 	get_online_cpus();
767 	__smp_rescan_cpus(info, 0);
768 	put_online_cpus();
769 	memblock_free_early((unsigned long)info, sizeof(*info));
770 }
771 
772 /*
773  *	Activate a secondary processor.
774  */
775 static void smp_start_secondary(void *cpuvoid)
776 {
777 	S390_lowcore.last_update_clock = get_tod_clock();
778 	S390_lowcore.restart_stack = (unsigned long) restart_stack;
779 	S390_lowcore.restart_fn = (unsigned long) do_restart;
780 	S390_lowcore.restart_data = 0;
781 	S390_lowcore.restart_source = -1UL;
782 	restore_access_regs(S390_lowcore.access_regs_save_area);
783 	__ctl_load(S390_lowcore.cregs_save_area, 0, 15);
784 	__load_psw_mask(PSW_KERNEL_BITS | PSW_MASK_DAT);
785 	cpu_init();
786 	preempt_disable();
787 	init_cpu_timer();
788 	vtime_init();
789 	pfault_init();
790 	notify_cpu_starting(smp_processor_id());
791 	set_cpu_online(smp_processor_id(), true);
792 	inc_irq_stat(CPU_RST);
793 	local_irq_enable();
794 	cpu_startup_entry(CPUHP_AP_ONLINE_IDLE);
795 }
796 
797 /* Upping and downing of CPUs */
798 int __cpu_up(unsigned int cpu, struct task_struct *tidle)
799 {
800 	struct pcpu *pcpu;
801 	int base, i, rc;
802 
803 	pcpu = pcpu_devices + cpu;
804 	if (pcpu->state != CPU_STATE_CONFIGURED)
805 		return -EIO;
806 	base = smp_get_base_cpu(cpu);
807 	for (i = 0; i <= smp_cpu_mtid; i++) {
808 		if (base + i < nr_cpu_ids)
809 			if (cpu_online(base + i))
810 				break;
811 	}
812 	/*
813 	 * If this is the first CPU of the core to get online
814 	 * do an initial CPU reset.
815 	 */
816 	if (i > smp_cpu_mtid &&
817 	    pcpu_sigp_retry(pcpu_devices + base, SIGP_INITIAL_CPU_RESET, 0) !=
818 	    SIGP_CC_ORDER_CODE_ACCEPTED)
819 		return -EIO;
820 
821 	rc = pcpu_alloc_lowcore(pcpu, cpu);
822 	if (rc)
823 		return rc;
824 	pcpu_prepare_secondary(pcpu, cpu);
825 	pcpu_attach_task(pcpu, tidle);
826 	pcpu_start_fn(pcpu, smp_start_secondary, NULL);
827 	/* Wait until cpu puts itself in the online & active maps */
828 	while (!cpu_online(cpu))
829 		cpu_relax();
830 	return 0;
831 }
832 
833 static unsigned int setup_possible_cpus __initdata;
834 
835 static int __init _setup_possible_cpus(char *s)
836 {
837 	get_option(&s, &setup_possible_cpus);
838 	return 0;
839 }
840 early_param("possible_cpus", _setup_possible_cpus);
841 
842 #ifdef CONFIG_HOTPLUG_CPU
843 
844 int __cpu_disable(void)
845 {
846 	unsigned long cregs[16];
847 
848 	/* Handle possible pending IPIs */
849 	smp_handle_ext_call();
850 	set_cpu_online(smp_processor_id(), false);
851 	/* Disable pseudo page faults on this cpu. */
852 	pfault_fini();
853 	/* Disable interrupt sources via control register. */
854 	__ctl_store(cregs, 0, 15);
855 	cregs[0]  &= ~0x0000ee70UL;	/* disable all external interrupts */
856 	cregs[6]  &= ~0xff000000UL;	/* disable all I/O interrupts */
857 	cregs[14] &= ~0x1f000000UL;	/* disable most machine checks */
858 	__ctl_load(cregs, 0, 15);
859 	clear_cpu_flag(CIF_NOHZ_DELAY);
860 	return 0;
861 }
862 
863 void __cpu_die(unsigned int cpu)
864 {
865 	struct pcpu *pcpu;
866 
867 	/* Wait until target cpu is down */
868 	pcpu = pcpu_devices + cpu;
869 	while (!pcpu_stopped(pcpu))
870 		cpu_relax();
871 	pcpu_free_lowcore(pcpu);
872 	cpumask_clear_cpu(cpu, mm_cpumask(&init_mm));
873 	cpumask_clear_cpu(cpu, &init_mm.context.cpu_attach_mask);
874 }
875 
876 void __noreturn cpu_die(void)
877 {
878 	idle_task_exit();
879 	pcpu_sigp_retry(pcpu_devices + smp_processor_id(), SIGP_STOP, 0);
880 	for (;;) ;
881 }
882 
883 #endif /* CONFIG_HOTPLUG_CPU */
884 
885 void __init smp_fill_possible_mask(void)
886 {
887 	unsigned int possible, sclp_max, cpu;
888 
889 	sclp_max = max(sclp.mtid, sclp.mtid_cp) + 1;
890 	sclp_max = min(smp_max_threads, sclp_max);
891 	sclp_max = (sclp.max_cores * sclp_max) ?: nr_cpu_ids;
892 	possible = setup_possible_cpus ?: nr_cpu_ids;
893 	possible = min(possible, sclp_max);
894 	for (cpu = 0; cpu < possible && cpu < nr_cpu_ids; cpu++)
895 		set_cpu_possible(cpu, true);
896 }
897 
898 void __init smp_prepare_cpus(unsigned int max_cpus)
899 {
900 	/* request the 0x1201 emergency signal external interrupt */
901 	if (register_external_irq(EXT_IRQ_EMERGENCY_SIG, do_ext_call_interrupt))
902 		panic("Couldn't request external interrupt 0x1201");
903 	/* request the 0x1202 external call external interrupt */
904 	if (register_external_irq(EXT_IRQ_EXTERNAL_CALL, do_ext_call_interrupt))
905 		panic("Couldn't request external interrupt 0x1202");
906 }
907 
908 void __init smp_prepare_boot_cpu(void)
909 {
910 	struct pcpu *pcpu = pcpu_devices;
911 
912 	WARN_ON(!cpu_present(0) || !cpu_online(0));
913 	pcpu->state = CPU_STATE_CONFIGURED;
914 	pcpu->lowcore = (struct lowcore *)(unsigned long) store_prefix();
915 	S390_lowcore.percpu_offset = __per_cpu_offset[0];
916 	smp_cpu_set_polarization(0, POLARIZATION_UNKNOWN);
917 }
918 
919 void __init smp_cpus_done(unsigned int max_cpus)
920 {
921 }
922 
923 void __init smp_setup_processor_id(void)
924 {
925 	pcpu_devices[0].address = stap();
926 	S390_lowcore.cpu_nr = 0;
927 	S390_lowcore.spinlock_lockval = arch_spin_lockval(0);
928 }
929 
930 /*
931  * the frequency of the profiling timer can be changed
932  * by writing a multiplier value into /proc/profile.
933  *
934  * usually you want to run this on all CPUs ;)
935  */
936 int setup_profiling_timer(unsigned int multiplier)
937 {
938 	return 0;
939 }
940 
941 #ifdef CONFIG_HOTPLUG_CPU
942 static ssize_t cpu_configure_show(struct device *dev,
943 				  struct device_attribute *attr, char *buf)
944 {
945 	ssize_t count;
946 
947 	mutex_lock(&smp_cpu_state_mutex);
948 	count = sprintf(buf, "%d\n", pcpu_devices[dev->id].state);
949 	mutex_unlock(&smp_cpu_state_mutex);
950 	return count;
951 }
952 
953 static ssize_t cpu_configure_store(struct device *dev,
954 				   struct device_attribute *attr,
955 				   const char *buf, size_t count)
956 {
957 	struct pcpu *pcpu;
958 	int cpu, val, rc, i;
959 	char delim;
960 
961 	if (sscanf(buf, "%d %c", &val, &delim) != 1)
962 		return -EINVAL;
963 	if (val != 0 && val != 1)
964 		return -EINVAL;
965 	get_online_cpus();
966 	mutex_lock(&smp_cpu_state_mutex);
967 	rc = -EBUSY;
968 	/* disallow configuration changes of online cpus and cpu 0 */
969 	cpu = dev->id;
970 	cpu = smp_get_base_cpu(cpu);
971 	if (cpu == 0)
972 		goto out;
973 	for (i = 0; i <= smp_cpu_mtid; i++)
974 		if (cpu_online(cpu + i))
975 			goto out;
976 	pcpu = pcpu_devices + cpu;
977 	rc = 0;
978 	switch (val) {
979 	case 0:
980 		if (pcpu->state != CPU_STATE_CONFIGURED)
981 			break;
982 		rc = sclp_core_deconfigure(pcpu->address >> smp_cpu_mt_shift);
983 		if (rc)
984 			break;
985 		for (i = 0; i <= smp_cpu_mtid; i++) {
986 			if (cpu + i >= nr_cpu_ids || !cpu_present(cpu + i))
987 				continue;
988 			pcpu[i].state = CPU_STATE_STANDBY;
989 			smp_cpu_set_polarization(cpu + i,
990 						 POLARIZATION_UNKNOWN);
991 		}
992 		topology_expect_change();
993 		break;
994 	case 1:
995 		if (pcpu->state != CPU_STATE_STANDBY)
996 			break;
997 		rc = sclp_core_configure(pcpu->address >> smp_cpu_mt_shift);
998 		if (rc)
999 			break;
1000 		for (i = 0; i <= smp_cpu_mtid; i++) {
1001 			if (cpu + i >= nr_cpu_ids || !cpu_present(cpu + i))
1002 				continue;
1003 			pcpu[i].state = CPU_STATE_CONFIGURED;
1004 			smp_cpu_set_polarization(cpu + i,
1005 						 POLARIZATION_UNKNOWN);
1006 		}
1007 		topology_expect_change();
1008 		break;
1009 	default:
1010 		break;
1011 	}
1012 out:
1013 	mutex_unlock(&smp_cpu_state_mutex);
1014 	put_online_cpus();
1015 	return rc ? rc : count;
1016 }
1017 static DEVICE_ATTR(configure, 0644, cpu_configure_show, cpu_configure_store);
1018 #endif /* CONFIG_HOTPLUG_CPU */
1019 
1020 static ssize_t show_cpu_address(struct device *dev,
1021 				struct device_attribute *attr, char *buf)
1022 {
1023 	return sprintf(buf, "%d\n", pcpu_devices[dev->id].address);
1024 }
1025 static DEVICE_ATTR(address, 0444, show_cpu_address, NULL);
1026 
1027 static struct attribute *cpu_common_attrs[] = {
1028 #ifdef CONFIG_HOTPLUG_CPU
1029 	&dev_attr_configure.attr,
1030 #endif
1031 	&dev_attr_address.attr,
1032 	NULL,
1033 };
1034 
1035 static struct attribute_group cpu_common_attr_group = {
1036 	.attrs = cpu_common_attrs,
1037 };
1038 
1039 static struct attribute *cpu_online_attrs[] = {
1040 	&dev_attr_idle_count.attr,
1041 	&dev_attr_idle_time_us.attr,
1042 	NULL,
1043 };
1044 
1045 static struct attribute_group cpu_online_attr_group = {
1046 	.attrs = cpu_online_attrs,
1047 };
1048 
1049 static int smp_cpu_online(unsigned int cpu)
1050 {
1051 	struct device *s = &per_cpu(cpu_device, cpu)->dev;
1052 
1053 	return sysfs_create_group(&s->kobj, &cpu_online_attr_group);
1054 }
1055 static int smp_cpu_pre_down(unsigned int cpu)
1056 {
1057 	struct device *s = &per_cpu(cpu_device, cpu)->dev;
1058 
1059 	sysfs_remove_group(&s->kobj, &cpu_online_attr_group);
1060 	return 0;
1061 }
1062 
1063 static int smp_add_present_cpu(int cpu)
1064 {
1065 	struct device *s;
1066 	struct cpu *c;
1067 	int rc;
1068 
1069 	c = kzalloc(sizeof(*c), GFP_KERNEL);
1070 	if (!c)
1071 		return -ENOMEM;
1072 	per_cpu(cpu_device, cpu) = c;
1073 	s = &c->dev;
1074 	c->hotpluggable = 1;
1075 	rc = register_cpu(c, cpu);
1076 	if (rc)
1077 		goto out;
1078 	rc = sysfs_create_group(&s->kobj, &cpu_common_attr_group);
1079 	if (rc)
1080 		goto out_cpu;
1081 	rc = topology_cpu_init(c);
1082 	if (rc)
1083 		goto out_topology;
1084 	return 0;
1085 
1086 out_topology:
1087 	sysfs_remove_group(&s->kobj, &cpu_common_attr_group);
1088 out_cpu:
1089 #ifdef CONFIG_HOTPLUG_CPU
1090 	unregister_cpu(c);
1091 #endif
1092 out:
1093 	return rc;
1094 }
1095 
1096 #ifdef CONFIG_HOTPLUG_CPU
1097 
1098 int __ref smp_rescan_cpus(void)
1099 {
1100 	struct sclp_core_info *info;
1101 	int nr;
1102 
1103 	info = kzalloc(sizeof(*info), GFP_KERNEL);
1104 	if (!info)
1105 		return -ENOMEM;
1106 	smp_get_core_info(info, 0);
1107 	get_online_cpus();
1108 	mutex_lock(&smp_cpu_state_mutex);
1109 	nr = __smp_rescan_cpus(info, 1);
1110 	mutex_unlock(&smp_cpu_state_mutex);
1111 	put_online_cpus();
1112 	kfree(info);
1113 	if (nr)
1114 		topology_schedule_update();
1115 	return 0;
1116 }
1117 
1118 static ssize_t __ref rescan_store(struct device *dev,
1119 				  struct device_attribute *attr,
1120 				  const char *buf,
1121 				  size_t count)
1122 {
1123 	int rc;
1124 
1125 	rc = smp_rescan_cpus();
1126 	return rc ? rc : count;
1127 }
1128 static DEVICE_ATTR(rescan, 0200, NULL, rescan_store);
1129 #endif /* CONFIG_HOTPLUG_CPU */
1130 
1131 static int __init s390_smp_init(void)
1132 {
1133 	int cpu, rc = 0;
1134 
1135 #ifdef CONFIG_HOTPLUG_CPU
1136 	rc = device_create_file(cpu_subsys.dev_root, &dev_attr_rescan);
1137 	if (rc)
1138 		return rc;
1139 #endif
1140 	for_each_present_cpu(cpu) {
1141 		rc = smp_add_present_cpu(cpu);
1142 		if (rc)
1143 			goto out;
1144 	}
1145 
1146 	rc = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "s390/smp:online",
1147 			       smp_cpu_online, smp_cpu_pre_down);
1148 out:
1149 	return rc;
1150 }
1151 subsys_initcall(s390_smp_init);
1152