xref: /openbmc/linux/arch/s390/kernel/smp.c (revision bbecb07f)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  *  SMP related functions
4  *
5  *    Copyright IBM Corp. 1999, 2012
6  *    Author(s): Denis Joseph Barrow,
7  *		 Martin Schwidefsky <schwidefsky@de.ibm.com>,
8  *		 Heiko Carstens <heiko.carstens@de.ibm.com>,
9  *
10  *  based on other smp stuff by
11  *    (c) 1995 Alan Cox, CymruNET Ltd  <alan@cymru.net>
12  *    (c) 1998 Ingo Molnar
13  *
14  * The code outside of smp.c uses logical cpu numbers, only smp.c does
15  * the translation of logical to physical cpu ids. All new code that
16  * operates on physical cpu numbers needs to go into smp.c.
17  */
18 
19 #define KMSG_COMPONENT "cpu"
20 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
21 
22 #include <linux/workqueue.h>
23 #include <linux/bootmem.h>
24 #include <linux/export.h>
25 #include <linux/init.h>
26 #include <linux/mm.h>
27 #include <linux/err.h>
28 #include <linux/spinlock.h>
29 #include <linux/kernel_stat.h>
30 #include <linux/kmemleak.h>
31 #include <linux/delay.h>
32 #include <linux/interrupt.h>
33 #include <linux/irqflags.h>
34 #include <linux/cpu.h>
35 #include <linux/slab.h>
36 #include <linux/sched/hotplug.h>
37 #include <linux/sched/task_stack.h>
38 #include <linux/crash_dump.h>
39 #include <linux/memblock.h>
40 #include <linux/kprobes.h>
41 #include <asm/asm-offsets.h>
42 #include <asm/diag.h>
43 #include <asm/switch_to.h>
44 #include <asm/facility.h>
45 #include <asm/ipl.h>
46 #include <asm/setup.h>
47 #include <asm/irq.h>
48 #include <asm/tlbflush.h>
49 #include <asm/vtimer.h>
50 #include <asm/lowcore.h>
51 #include <asm/sclp.h>
52 #include <asm/vdso.h>
53 #include <asm/debug.h>
54 #include <asm/os_info.h>
55 #include <asm/sigp.h>
56 #include <asm/idle.h>
57 #include <asm/nmi.h>
58 #include <asm/topology.h>
59 #include "entry.h"
60 
61 enum {
62 	ec_schedule = 0,
63 	ec_call_function_single,
64 	ec_stop_cpu,
65 };
66 
67 enum {
68 	CPU_STATE_STANDBY,
69 	CPU_STATE_CONFIGURED,
70 };
71 
72 static DEFINE_PER_CPU(struct cpu *, cpu_device);
73 
74 struct pcpu {
75 	struct lowcore *lowcore;	/* lowcore page(s) for the cpu */
76 	unsigned long ec_mask;		/* bit mask for ec_xxx functions */
77 	unsigned long ec_clk;		/* sigp timestamp for ec_xxx */
78 	signed char state;		/* physical cpu state */
79 	signed char polarization;	/* physical polarization */
80 	u16 address;			/* physical cpu address */
81 };
82 
83 static u8 boot_core_type;
84 static struct pcpu pcpu_devices[NR_CPUS];
85 
86 unsigned int smp_cpu_mt_shift;
87 EXPORT_SYMBOL(smp_cpu_mt_shift);
88 
89 unsigned int smp_cpu_mtid;
90 EXPORT_SYMBOL(smp_cpu_mtid);
91 
92 #ifdef CONFIG_CRASH_DUMP
93 __vector128 __initdata boot_cpu_vector_save_area[__NUM_VXRS];
94 #endif
95 
96 static unsigned int smp_max_threads __initdata = -1U;
97 
98 static int __init early_nosmt(char *s)
99 {
100 	smp_max_threads = 1;
101 	return 0;
102 }
103 early_param("nosmt", early_nosmt);
104 
105 static int __init early_smt(char *s)
106 {
107 	get_option(&s, &smp_max_threads);
108 	return 0;
109 }
110 early_param("smt", early_smt);
111 
112 /*
113  * The smp_cpu_state_mutex must be held when changing the state or polarization
114  * member of a pcpu data structure within the pcpu_devices arreay.
115  */
116 DEFINE_MUTEX(smp_cpu_state_mutex);
117 
118 /*
119  * Signal processor helper functions.
120  */
121 static inline int __pcpu_sigp_relax(u16 addr, u8 order, unsigned long parm)
122 {
123 	int cc;
124 
125 	while (1) {
126 		cc = __pcpu_sigp(addr, order, parm, NULL);
127 		if (cc != SIGP_CC_BUSY)
128 			return cc;
129 		cpu_relax();
130 	}
131 }
132 
133 static int pcpu_sigp_retry(struct pcpu *pcpu, u8 order, u32 parm)
134 {
135 	int cc, retry;
136 
137 	for (retry = 0; ; retry++) {
138 		cc = __pcpu_sigp(pcpu->address, order, parm, NULL);
139 		if (cc != SIGP_CC_BUSY)
140 			break;
141 		if (retry >= 3)
142 			udelay(10);
143 	}
144 	return cc;
145 }
146 
147 static inline int pcpu_stopped(struct pcpu *pcpu)
148 {
149 	u32 uninitialized_var(status);
150 
151 	if (__pcpu_sigp(pcpu->address, SIGP_SENSE,
152 			0, &status) != SIGP_CC_STATUS_STORED)
153 		return 0;
154 	return !!(status & (SIGP_STATUS_CHECK_STOP|SIGP_STATUS_STOPPED));
155 }
156 
157 static inline int pcpu_running(struct pcpu *pcpu)
158 {
159 	if (__pcpu_sigp(pcpu->address, SIGP_SENSE_RUNNING,
160 			0, NULL) != SIGP_CC_STATUS_STORED)
161 		return 1;
162 	/* Status stored condition code is equivalent to cpu not running. */
163 	return 0;
164 }
165 
166 /*
167  * Find struct pcpu by cpu address.
168  */
169 static struct pcpu *pcpu_find_address(const struct cpumask *mask, u16 address)
170 {
171 	int cpu;
172 
173 	for_each_cpu(cpu, mask)
174 		if (pcpu_devices[cpu].address == address)
175 			return pcpu_devices + cpu;
176 	return NULL;
177 }
178 
179 static void pcpu_ec_call(struct pcpu *pcpu, int ec_bit)
180 {
181 	int order;
182 
183 	if (test_and_set_bit(ec_bit, &pcpu->ec_mask))
184 		return;
185 	order = pcpu_running(pcpu) ? SIGP_EXTERNAL_CALL : SIGP_EMERGENCY_SIGNAL;
186 	pcpu->ec_clk = get_tod_clock_fast();
187 	pcpu_sigp_retry(pcpu, order, 0);
188 }
189 
190 #define ASYNC_FRAME_OFFSET (ASYNC_SIZE - STACK_FRAME_OVERHEAD - __PT_SIZE)
191 #define PANIC_FRAME_OFFSET (PAGE_SIZE - STACK_FRAME_OVERHEAD - __PT_SIZE)
192 
193 static int pcpu_alloc_lowcore(struct pcpu *pcpu, int cpu)
194 {
195 	unsigned long async_stack, panic_stack;
196 	struct lowcore *lc;
197 
198 	if (pcpu != &pcpu_devices[0]) {
199 		pcpu->lowcore =	(struct lowcore *)
200 			__get_free_pages(GFP_KERNEL | GFP_DMA, LC_ORDER);
201 		async_stack = __get_free_pages(GFP_KERNEL, ASYNC_ORDER);
202 		panic_stack = __get_free_page(GFP_KERNEL);
203 		if (!pcpu->lowcore || !panic_stack || !async_stack)
204 			goto out;
205 	} else {
206 		async_stack = pcpu->lowcore->async_stack - ASYNC_FRAME_OFFSET;
207 		panic_stack = pcpu->lowcore->panic_stack - PANIC_FRAME_OFFSET;
208 	}
209 	lc = pcpu->lowcore;
210 	memcpy(lc, &S390_lowcore, 512);
211 	memset((char *) lc + 512, 0, sizeof(*lc) - 512);
212 	lc->async_stack = async_stack + ASYNC_FRAME_OFFSET;
213 	lc->panic_stack = panic_stack + PANIC_FRAME_OFFSET;
214 	lc->cpu_nr = cpu;
215 	lc->spinlock_lockval = arch_spin_lockval(cpu);
216 	lc->spinlock_index = 0;
217 	if (nmi_alloc_per_cpu(lc))
218 		goto out;
219 	if (vdso_alloc_per_cpu(lc))
220 		goto out_mcesa;
221 	lowcore_ptr[cpu] = lc;
222 	pcpu_sigp_retry(pcpu, SIGP_SET_PREFIX, (u32)(unsigned long) lc);
223 	return 0;
224 
225 out_mcesa:
226 	nmi_free_per_cpu(lc);
227 out:
228 	if (pcpu != &pcpu_devices[0]) {
229 		free_page(panic_stack);
230 		free_pages(async_stack, ASYNC_ORDER);
231 		free_pages((unsigned long) pcpu->lowcore, LC_ORDER);
232 	}
233 	return -ENOMEM;
234 }
235 
236 #ifdef CONFIG_HOTPLUG_CPU
237 
238 static void pcpu_free_lowcore(struct pcpu *pcpu)
239 {
240 	pcpu_sigp_retry(pcpu, SIGP_SET_PREFIX, 0);
241 	lowcore_ptr[pcpu - pcpu_devices] = NULL;
242 	vdso_free_per_cpu(pcpu->lowcore);
243 	nmi_free_per_cpu(pcpu->lowcore);
244 	if (pcpu == &pcpu_devices[0])
245 		return;
246 	free_page(pcpu->lowcore->panic_stack-PANIC_FRAME_OFFSET);
247 	free_pages(pcpu->lowcore->async_stack-ASYNC_FRAME_OFFSET, ASYNC_ORDER);
248 	free_pages((unsigned long) pcpu->lowcore, LC_ORDER);
249 }
250 
251 #endif /* CONFIG_HOTPLUG_CPU */
252 
253 static void pcpu_prepare_secondary(struct pcpu *pcpu, int cpu)
254 {
255 	struct lowcore *lc = pcpu->lowcore;
256 
257 	cpumask_set_cpu(cpu, &init_mm.context.cpu_attach_mask);
258 	cpumask_set_cpu(cpu, mm_cpumask(&init_mm));
259 	lc->cpu_nr = cpu;
260 	lc->spinlock_lockval = arch_spin_lockval(cpu);
261 	lc->spinlock_index = 0;
262 	lc->percpu_offset = __per_cpu_offset[cpu];
263 	lc->kernel_asce = S390_lowcore.kernel_asce;
264 	lc->machine_flags = S390_lowcore.machine_flags;
265 	lc->user_timer = lc->system_timer = lc->steal_timer = 0;
266 	__ctl_store(lc->cregs_save_area, 0, 15);
267 	save_access_regs((unsigned int *) lc->access_regs_save_area);
268 	memcpy(lc->stfle_fac_list, S390_lowcore.stfle_fac_list,
269 	       MAX_FACILITY_BIT/8);
270 	arch_spin_lock_setup(cpu);
271 }
272 
273 static void pcpu_attach_task(struct pcpu *pcpu, struct task_struct *tsk)
274 {
275 	struct lowcore *lc = pcpu->lowcore;
276 
277 	lc->kernel_stack = (unsigned long) task_stack_page(tsk)
278 		+ THREAD_SIZE - STACK_FRAME_OVERHEAD - sizeof(struct pt_regs);
279 	lc->current_task = (unsigned long) tsk;
280 	lc->lpp = LPP_MAGIC;
281 	lc->current_pid = tsk->pid;
282 	lc->user_timer = tsk->thread.user_timer;
283 	lc->guest_timer = tsk->thread.guest_timer;
284 	lc->system_timer = tsk->thread.system_timer;
285 	lc->hardirq_timer = tsk->thread.hardirq_timer;
286 	lc->softirq_timer = tsk->thread.softirq_timer;
287 	lc->steal_timer = 0;
288 }
289 
290 static void pcpu_start_fn(struct pcpu *pcpu, void (*func)(void *), void *data)
291 {
292 	struct lowcore *lc = pcpu->lowcore;
293 
294 	lc->restart_stack = lc->kernel_stack;
295 	lc->restart_fn = (unsigned long) func;
296 	lc->restart_data = (unsigned long) data;
297 	lc->restart_source = -1UL;
298 	pcpu_sigp_retry(pcpu, SIGP_RESTART, 0);
299 }
300 
301 /*
302  * Call function via PSW restart on pcpu and stop the current cpu.
303  */
304 static void pcpu_delegate(struct pcpu *pcpu, void (*func)(void *),
305 			  void *data, unsigned long stack)
306 {
307 	struct lowcore *lc = lowcore_ptr[pcpu - pcpu_devices];
308 	unsigned long source_cpu = stap();
309 
310 	__load_psw_mask(PSW_KERNEL_BITS);
311 	if (pcpu->address == source_cpu)
312 		func(data);	/* should not return */
313 	/* Stop target cpu (if func returns this stops the current cpu). */
314 	pcpu_sigp_retry(pcpu, SIGP_STOP, 0);
315 	/* Restart func on the target cpu and stop the current cpu. */
316 	mem_assign_absolute(lc->restart_stack, stack);
317 	mem_assign_absolute(lc->restart_fn, (unsigned long) func);
318 	mem_assign_absolute(lc->restart_data, (unsigned long) data);
319 	mem_assign_absolute(lc->restart_source, source_cpu);
320 	asm volatile(
321 		"0:	sigp	0,%0,%2	# sigp restart to target cpu\n"
322 		"	brc	2,0b	# busy, try again\n"
323 		"1:	sigp	0,%1,%3	# sigp stop to current cpu\n"
324 		"	brc	2,1b	# busy, try again\n"
325 		: : "d" (pcpu->address), "d" (source_cpu),
326 		    "K" (SIGP_RESTART), "K" (SIGP_STOP)
327 		: "0", "1", "cc");
328 	for (;;) ;
329 }
330 
331 /*
332  * Enable additional logical cpus for multi-threading.
333  */
334 static int pcpu_set_smt(unsigned int mtid)
335 {
336 	int cc;
337 
338 	if (smp_cpu_mtid == mtid)
339 		return 0;
340 	cc = __pcpu_sigp(0, SIGP_SET_MULTI_THREADING, mtid, NULL);
341 	if (cc == 0) {
342 		smp_cpu_mtid = mtid;
343 		smp_cpu_mt_shift = 0;
344 		while (smp_cpu_mtid >= (1U << smp_cpu_mt_shift))
345 			smp_cpu_mt_shift++;
346 		pcpu_devices[0].address = stap();
347 	}
348 	return cc;
349 }
350 
351 /*
352  * Call function on an online CPU.
353  */
354 void smp_call_online_cpu(void (*func)(void *), void *data)
355 {
356 	struct pcpu *pcpu;
357 
358 	/* Use the current cpu if it is online. */
359 	pcpu = pcpu_find_address(cpu_online_mask, stap());
360 	if (!pcpu)
361 		/* Use the first online cpu. */
362 		pcpu = pcpu_devices + cpumask_first(cpu_online_mask);
363 	pcpu_delegate(pcpu, func, data, (unsigned long) restart_stack);
364 }
365 
366 /*
367  * Call function on the ipl CPU.
368  */
369 void smp_call_ipl_cpu(void (*func)(void *), void *data)
370 {
371 	pcpu_delegate(&pcpu_devices[0], func, data,
372 		      pcpu_devices->lowcore->panic_stack -
373 		      PANIC_FRAME_OFFSET + PAGE_SIZE);
374 }
375 
376 int smp_find_processor_id(u16 address)
377 {
378 	int cpu;
379 
380 	for_each_present_cpu(cpu)
381 		if (pcpu_devices[cpu].address == address)
382 			return cpu;
383 	return -1;
384 }
385 
386 bool arch_vcpu_is_preempted(int cpu)
387 {
388 	if (test_cpu_flag_of(CIF_ENABLED_WAIT, cpu))
389 		return false;
390 	if (pcpu_running(pcpu_devices + cpu))
391 		return false;
392 	return true;
393 }
394 EXPORT_SYMBOL(arch_vcpu_is_preempted);
395 
396 void smp_yield_cpu(int cpu)
397 {
398 	if (MACHINE_HAS_DIAG9C) {
399 		diag_stat_inc_norecursion(DIAG_STAT_X09C);
400 		asm volatile("diag %0,0,0x9c"
401 			     : : "d" (pcpu_devices[cpu].address));
402 	} else if (MACHINE_HAS_DIAG44) {
403 		diag_stat_inc_norecursion(DIAG_STAT_X044);
404 		asm volatile("diag 0,0,0x44");
405 	}
406 }
407 
408 /*
409  * Send cpus emergency shutdown signal. This gives the cpus the
410  * opportunity to complete outstanding interrupts.
411  */
412 void notrace smp_emergency_stop(void)
413 {
414 	cpumask_t cpumask;
415 	u64 end;
416 	int cpu;
417 
418 	cpumask_copy(&cpumask, cpu_online_mask);
419 	cpumask_clear_cpu(smp_processor_id(), &cpumask);
420 
421 	end = get_tod_clock() + (1000000UL << 12);
422 	for_each_cpu(cpu, &cpumask) {
423 		struct pcpu *pcpu = pcpu_devices + cpu;
424 		set_bit(ec_stop_cpu, &pcpu->ec_mask);
425 		while (__pcpu_sigp(pcpu->address, SIGP_EMERGENCY_SIGNAL,
426 				   0, NULL) == SIGP_CC_BUSY &&
427 		       get_tod_clock() < end)
428 			cpu_relax();
429 	}
430 	while (get_tod_clock() < end) {
431 		for_each_cpu(cpu, &cpumask)
432 			if (pcpu_stopped(pcpu_devices + cpu))
433 				cpumask_clear_cpu(cpu, &cpumask);
434 		if (cpumask_empty(&cpumask))
435 			break;
436 		cpu_relax();
437 	}
438 }
439 NOKPROBE_SYMBOL(smp_emergency_stop);
440 
441 /*
442  * Stop all cpus but the current one.
443  */
444 void smp_send_stop(void)
445 {
446 	int cpu;
447 
448 	/* Disable all interrupts/machine checks */
449 	__load_psw_mask(PSW_KERNEL_BITS | PSW_MASK_DAT);
450 	trace_hardirqs_off();
451 
452 	debug_set_critical();
453 
454 	if (oops_in_progress)
455 		smp_emergency_stop();
456 
457 	/* stop all processors */
458 	for_each_online_cpu(cpu) {
459 		if (cpu == smp_processor_id())
460 			continue;
461 		pcpu_sigp_retry(pcpu_devices + cpu, SIGP_STOP, 0);
462 		while (!pcpu_stopped(pcpu_devices + cpu))
463 			cpu_relax();
464 	}
465 }
466 
467 /*
468  * This is the main routine where commands issued by other
469  * cpus are handled.
470  */
471 static void smp_handle_ext_call(void)
472 {
473 	unsigned long bits;
474 
475 	/* handle bit signal external calls */
476 	bits = xchg(&pcpu_devices[smp_processor_id()].ec_mask, 0);
477 	if (test_bit(ec_stop_cpu, &bits))
478 		smp_stop_cpu();
479 	if (test_bit(ec_schedule, &bits))
480 		scheduler_ipi();
481 	if (test_bit(ec_call_function_single, &bits))
482 		generic_smp_call_function_single_interrupt();
483 }
484 
485 static void do_ext_call_interrupt(struct ext_code ext_code,
486 				  unsigned int param32, unsigned long param64)
487 {
488 	inc_irq_stat(ext_code.code == 0x1202 ? IRQEXT_EXC : IRQEXT_EMS);
489 	smp_handle_ext_call();
490 }
491 
492 void arch_send_call_function_ipi_mask(const struct cpumask *mask)
493 {
494 	int cpu;
495 
496 	for_each_cpu(cpu, mask)
497 		pcpu_ec_call(pcpu_devices + cpu, ec_call_function_single);
498 }
499 
500 void arch_send_call_function_single_ipi(int cpu)
501 {
502 	pcpu_ec_call(pcpu_devices + cpu, ec_call_function_single);
503 }
504 
505 /*
506  * this function sends a 'reschedule' IPI to another CPU.
507  * it goes straight through and wastes no time serializing
508  * anything. Worst case is that we lose a reschedule ...
509  */
510 void smp_send_reschedule(int cpu)
511 {
512 	pcpu_ec_call(pcpu_devices + cpu, ec_schedule);
513 }
514 
515 /*
516  * parameter area for the set/clear control bit callbacks
517  */
518 struct ec_creg_mask_parms {
519 	unsigned long orval;
520 	unsigned long andval;
521 	int cr;
522 };
523 
524 /*
525  * callback for setting/clearing control bits
526  */
527 static void smp_ctl_bit_callback(void *info)
528 {
529 	struct ec_creg_mask_parms *pp = info;
530 	unsigned long cregs[16];
531 
532 	__ctl_store(cregs, 0, 15);
533 	cregs[pp->cr] = (cregs[pp->cr] & pp->andval) | pp->orval;
534 	__ctl_load(cregs, 0, 15);
535 }
536 
537 /*
538  * Set a bit in a control register of all cpus
539  */
540 void smp_ctl_set_bit(int cr, int bit)
541 {
542 	struct ec_creg_mask_parms parms = { 1UL << bit, -1UL, cr };
543 
544 	on_each_cpu(smp_ctl_bit_callback, &parms, 1);
545 }
546 EXPORT_SYMBOL(smp_ctl_set_bit);
547 
548 /*
549  * Clear a bit in a control register of all cpus
550  */
551 void smp_ctl_clear_bit(int cr, int bit)
552 {
553 	struct ec_creg_mask_parms parms = { 0, ~(1UL << bit), cr };
554 
555 	on_each_cpu(smp_ctl_bit_callback, &parms, 1);
556 }
557 EXPORT_SYMBOL(smp_ctl_clear_bit);
558 
559 #ifdef CONFIG_CRASH_DUMP
560 
561 int smp_store_status(int cpu)
562 {
563 	struct pcpu *pcpu = pcpu_devices + cpu;
564 	unsigned long pa;
565 
566 	pa = __pa(&pcpu->lowcore->floating_pt_save_area);
567 	if (__pcpu_sigp_relax(pcpu->address, SIGP_STORE_STATUS_AT_ADDRESS,
568 			      pa) != SIGP_CC_ORDER_CODE_ACCEPTED)
569 		return -EIO;
570 	if (!MACHINE_HAS_VX && !MACHINE_HAS_GS)
571 		return 0;
572 	pa = __pa(pcpu->lowcore->mcesad & MCESA_ORIGIN_MASK);
573 	if (MACHINE_HAS_GS)
574 		pa |= pcpu->lowcore->mcesad & MCESA_LC_MASK;
575 	if (__pcpu_sigp_relax(pcpu->address, SIGP_STORE_ADDITIONAL_STATUS,
576 			      pa) != SIGP_CC_ORDER_CODE_ACCEPTED)
577 		return -EIO;
578 	return 0;
579 }
580 
581 /*
582  * Collect CPU state of the previous, crashed system.
583  * There are four cases:
584  * 1) standard zfcp dump
585  *    condition: OLDMEM_BASE == NULL && ipl_info.type == IPL_TYPE_FCP_DUMP
586  *    The state for all CPUs except the boot CPU needs to be collected
587  *    with sigp stop-and-store-status. The boot CPU state is located in
588  *    the absolute lowcore of the memory stored in the HSA. The zcore code
589  *    will copy the boot CPU state from the HSA.
590  * 2) stand-alone kdump for SCSI (zfcp dump with swapped memory)
591  *    condition: OLDMEM_BASE != NULL && ipl_info.type == IPL_TYPE_FCP_DUMP
592  *    The state for all CPUs except the boot CPU needs to be collected
593  *    with sigp stop-and-store-status. The firmware or the boot-loader
594  *    stored the registers of the boot CPU in the absolute lowcore in the
595  *    memory of the old system.
596  * 3) kdump and the old kernel did not store the CPU state,
597  *    or stand-alone kdump for DASD
598  *    condition: OLDMEM_BASE != NULL && !is_kdump_kernel()
599  *    The state for all CPUs except the boot CPU needs to be collected
600  *    with sigp stop-and-store-status. The kexec code or the boot-loader
601  *    stored the registers of the boot CPU in the memory of the old system.
602  * 4) kdump and the old kernel stored the CPU state
603  *    condition: OLDMEM_BASE != NULL && is_kdump_kernel()
604  *    This case does not exist for s390 anymore, setup_arch explicitly
605  *    deactivates the elfcorehdr= kernel parameter
606  */
607 static __init void smp_save_cpu_vxrs(struct save_area *sa, u16 addr,
608 				     bool is_boot_cpu, unsigned long page)
609 {
610 	__vector128 *vxrs = (__vector128 *) page;
611 
612 	if (is_boot_cpu)
613 		vxrs = boot_cpu_vector_save_area;
614 	else
615 		__pcpu_sigp_relax(addr, SIGP_STORE_ADDITIONAL_STATUS, page);
616 	save_area_add_vxrs(sa, vxrs);
617 }
618 
619 static __init void smp_save_cpu_regs(struct save_area *sa, u16 addr,
620 				     bool is_boot_cpu, unsigned long page)
621 {
622 	void *regs = (void *) page;
623 
624 	if (is_boot_cpu)
625 		copy_oldmem_kernel(regs, (void *) __LC_FPREGS_SAVE_AREA, 512);
626 	else
627 		__pcpu_sigp_relax(addr, SIGP_STORE_STATUS_AT_ADDRESS, page);
628 	save_area_add_regs(sa, regs);
629 }
630 
631 void __init smp_save_dump_cpus(void)
632 {
633 	int addr, boot_cpu_addr, max_cpu_addr;
634 	struct save_area *sa;
635 	unsigned long page;
636 	bool is_boot_cpu;
637 
638 	if (!(OLDMEM_BASE || ipl_info.type == IPL_TYPE_FCP_DUMP))
639 		/* No previous system present, normal boot. */
640 		return;
641 	/* Allocate a page as dumping area for the store status sigps */
642 	page = memblock_alloc_base(PAGE_SIZE, PAGE_SIZE, 1UL << 31);
643 	/* Set multi-threading state to the previous system. */
644 	pcpu_set_smt(sclp.mtid_prev);
645 	boot_cpu_addr = stap();
646 	max_cpu_addr = SCLP_MAX_CORES << sclp.mtid_prev;
647 	for (addr = 0; addr <= max_cpu_addr; addr++) {
648 		if (__pcpu_sigp_relax(addr, SIGP_SENSE, 0) ==
649 		    SIGP_CC_NOT_OPERATIONAL)
650 			continue;
651 		is_boot_cpu = (addr == boot_cpu_addr);
652 		/* Allocate save area */
653 		sa = save_area_alloc(is_boot_cpu);
654 		if (!sa)
655 			panic("could not allocate memory for save area\n");
656 		if (MACHINE_HAS_VX)
657 			/* Get the vector registers */
658 			smp_save_cpu_vxrs(sa, addr, is_boot_cpu, page);
659 		/*
660 		 * For a zfcp dump OLDMEM_BASE == NULL and the registers
661 		 * of the boot CPU are stored in the HSA. To retrieve
662 		 * these registers an SCLP request is required which is
663 		 * done by drivers/s390/char/zcore.c:init_cpu_info()
664 		 */
665 		if (!is_boot_cpu || OLDMEM_BASE)
666 			/* Get the CPU registers */
667 			smp_save_cpu_regs(sa, addr, is_boot_cpu, page);
668 	}
669 	memblock_free(page, PAGE_SIZE);
670 	diag308_reset();
671 	pcpu_set_smt(0);
672 }
673 #endif /* CONFIG_CRASH_DUMP */
674 
675 void smp_cpu_set_polarization(int cpu, int val)
676 {
677 	pcpu_devices[cpu].polarization = val;
678 }
679 
680 int smp_cpu_get_polarization(int cpu)
681 {
682 	return pcpu_devices[cpu].polarization;
683 }
684 
685 static void __ref smp_get_core_info(struct sclp_core_info *info, int early)
686 {
687 	static int use_sigp_detection;
688 	int address;
689 
690 	if (use_sigp_detection || sclp_get_core_info(info, early)) {
691 		use_sigp_detection = 1;
692 		for (address = 0;
693 		     address < (SCLP_MAX_CORES << smp_cpu_mt_shift);
694 		     address += (1U << smp_cpu_mt_shift)) {
695 			if (__pcpu_sigp_relax(address, SIGP_SENSE, 0) ==
696 			    SIGP_CC_NOT_OPERATIONAL)
697 				continue;
698 			info->core[info->configured].core_id =
699 				address >> smp_cpu_mt_shift;
700 			info->configured++;
701 		}
702 		info->combined = info->configured;
703 	}
704 }
705 
706 static int smp_add_present_cpu(int cpu);
707 
708 static int __smp_rescan_cpus(struct sclp_core_info *info, int sysfs_add)
709 {
710 	struct pcpu *pcpu;
711 	cpumask_t avail;
712 	int cpu, nr, i, j;
713 	u16 address;
714 
715 	nr = 0;
716 	cpumask_xor(&avail, cpu_possible_mask, cpu_present_mask);
717 	cpu = cpumask_first(&avail);
718 	for (i = 0; (i < info->combined) && (cpu < nr_cpu_ids); i++) {
719 		if (sclp.has_core_type && info->core[i].type != boot_core_type)
720 			continue;
721 		address = info->core[i].core_id << smp_cpu_mt_shift;
722 		for (j = 0; j <= smp_cpu_mtid; j++) {
723 			if (pcpu_find_address(cpu_present_mask, address + j))
724 				continue;
725 			pcpu = pcpu_devices + cpu;
726 			pcpu->address = address + j;
727 			pcpu->state =
728 				(cpu >= info->configured*(smp_cpu_mtid + 1)) ?
729 				CPU_STATE_STANDBY : CPU_STATE_CONFIGURED;
730 			smp_cpu_set_polarization(cpu, POLARIZATION_UNKNOWN);
731 			set_cpu_present(cpu, true);
732 			if (sysfs_add && smp_add_present_cpu(cpu) != 0)
733 				set_cpu_present(cpu, false);
734 			else
735 				nr++;
736 			cpu = cpumask_next(cpu, &avail);
737 			if (cpu >= nr_cpu_ids)
738 				break;
739 		}
740 	}
741 	return nr;
742 }
743 
744 void __init smp_detect_cpus(void)
745 {
746 	unsigned int cpu, mtid, c_cpus, s_cpus;
747 	struct sclp_core_info *info;
748 	u16 address;
749 
750 	/* Get CPU information */
751 	info = memblock_virt_alloc(sizeof(*info), 8);
752 	smp_get_core_info(info, 1);
753 	/* Find boot CPU type */
754 	if (sclp.has_core_type) {
755 		address = stap();
756 		for (cpu = 0; cpu < info->combined; cpu++)
757 			if (info->core[cpu].core_id == address) {
758 				/* The boot cpu dictates the cpu type. */
759 				boot_core_type = info->core[cpu].type;
760 				break;
761 			}
762 		if (cpu >= info->combined)
763 			panic("Could not find boot CPU type");
764 	}
765 
766 	/* Set multi-threading state for the current system */
767 	mtid = boot_core_type ? sclp.mtid : sclp.mtid_cp;
768 	mtid = (mtid < smp_max_threads) ? mtid : smp_max_threads - 1;
769 	pcpu_set_smt(mtid);
770 
771 	/* Print number of CPUs */
772 	c_cpus = s_cpus = 0;
773 	for (cpu = 0; cpu < info->combined; cpu++) {
774 		if (sclp.has_core_type &&
775 		    info->core[cpu].type != boot_core_type)
776 			continue;
777 		if (cpu < info->configured)
778 			c_cpus += smp_cpu_mtid + 1;
779 		else
780 			s_cpus += smp_cpu_mtid + 1;
781 	}
782 	pr_info("%d configured CPUs, %d standby CPUs\n", c_cpus, s_cpus);
783 
784 	/* Add CPUs present at boot */
785 	get_online_cpus();
786 	__smp_rescan_cpus(info, 0);
787 	put_online_cpus();
788 	memblock_free_early((unsigned long)info, sizeof(*info));
789 }
790 
791 /*
792  *	Activate a secondary processor.
793  */
794 static void smp_start_secondary(void *cpuvoid)
795 {
796 	int cpu = smp_processor_id();
797 
798 	S390_lowcore.last_update_clock = get_tod_clock();
799 	S390_lowcore.restart_stack = (unsigned long) restart_stack;
800 	S390_lowcore.restart_fn = (unsigned long) do_restart;
801 	S390_lowcore.restart_data = 0;
802 	S390_lowcore.restart_source = -1UL;
803 	restore_access_regs(S390_lowcore.access_regs_save_area);
804 	__ctl_load(S390_lowcore.cregs_save_area, 0, 15);
805 	__load_psw_mask(PSW_KERNEL_BITS | PSW_MASK_DAT);
806 	cpu_init();
807 	preempt_disable();
808 	init_cpu_timer();
809 	vtime_init();
810 	pfault_init();
811 	notify_cpu_starting(cpu);
812 	if (topology_cpu_dedicated(cpu))
813 		set_cpu_flag(CIF_DEDICATED_CPU);
814 	else
815 		clear_cpu_flag(CIF_DEDICATED_CPU);
816 	set_cpu_online(cpu, true);
817 	inc_irq_stat(CPU_RST);
818 	local_irq_enable();
819 	cpu_startup_entry(CPUHP_AP_ONLINE_IDLE);
820 }
821 
822 /* Upping and downing of CPUs */
823 int __cpu_up(unsigned int cpu, struct task_struct *tidle)
824 {
825 	struct pcpu *pcpu;
826 	int base, i, rc;
827 
828 	pcpu = pcpu_devices + cpu;
829 	if (pcpu->state != CPU_STATE_CONFIGURED)
830 		return -EIO;
831 	base = smp_get_base_cpu(cpu);
832 	for (i = 0; i <= smp_cpu_mtid; i++) {
833 		if (base + i < nr_cpu_ids)
834 			if (cpu_online(base + i))
835 				break;
836 	}
837 	/*
838 	 * If this is the first CPU of the core to get online
839 	 * do an initial CPU reset.
840 	 */
841 	if (i > smp_cpu_mtid &&
842 	    pcpu_sigp_retry(pcpu_devices + base, SIGP_INITIAL_CPU_RESET, 0) !=
843 	    SIGP_CC_ORDER_CODE_ACCEPTED)
844 		return -EIO;
845 
846 	rc = pcpu_alloc_lowcore(pcpu, cpu);
847 	if (rc)
848 		return rc;
849 	pcpu_prepare_secondary(pcpu, cpu);
850 	pcpu_attach_task(pcpu, tidle);
851 	pcpu_start_fn(pcpu, smp_start_secondary, NULL);
852 	/* Wait until cpu puts itself in the online & active maps */
853 	while (!cpu_online(cpu))
854 		cpu_relax();
855 	return 0;
856 }
857 
858 static unsigned int setup_possible_cpus __initdata;
859 
860 static int __init _setup_possible_cpus(char *s)
861 {
862 	get_option(&s, &setup_possible_cpus);
863 	return 0;
864 }
865 early_param("possible_cpus", _setup_possible_cpus);
866 
867 #ifdef CONFIG_HOTPLUG_CPU
868 
869 int __cpu_disable(void)
870 {
871 	unsigned long cregs[16];
872 
873 	/* Handle possible pending IPIs */
874 	smp_handle_ext_call();
875 	set_cpu_online(smp_processor_id(), false);
876 	/* Disable pseudo page faults on this cpu. */
877 	pfault_fini();
878 	/* Disable interrupt sources via control register. */
879 	__ctl_store(cregs, 0, 15);
880 	cregs[0]  &= ~0x0000ee70UL;	/* disable all external interrupts */
881 	cregs[6]  &= ~0xff000000UL;	/* disable all I/O interrupts */
882 	cregs[14] &= ~0x1f000000UL;	/* disable most machine checks */
883 	__ctl_load(cregs, 0, 15);
884 	clear_cpu_flag(CIF_NOHZ_DELAY);
885 	return 0;
886 }
887 
888 void __cpu_die(unsigned int cpu)
889 {
890 	struct pcpu *pcpu;
891 
892 	/* Wait until target cpu is down */
893 	pcpu = pcpu_devices + cpu;
894 	while (!pcpu_stopped(pcpu))
895 		cpu_relax();
896 	pcpu_free_lowcore(pcpu);
897 	cpumask_clear_cpu(cpu, mm_cpumask(&init_mm));
898 	cpumask_clear_cpu(cpu, &init_mm.context.cpu_attach_mask);
899 }
900 
901 void __noreturn cpu_die(void)
902 {
903 	idle_task_exit();
904 	pcpu_sigp_retry(pcpu_devices + smp_processor_id(), SIGP_STOP, 0);
905 	for (;;) ;
906 }
907 
908 #endif /* CONFIG_HOTPLUG_CPU */
909 
910 void __init smp_fill_possible_mask(void)
911 {
912 	unsigned int possible, sclp_max, cpu;
913 
914 	sclp_max = max(sclp.mtid, sclp.mtid_cp) + 1;
915 	sclp_max = min(smp_max_threads, sclp_max);
916 	sclp_max = (sclp.max_cores * sclp_max) ?: nr_cpu_ids;
917 	possible = setup_possible_cpus ?: nr_cpu_ids;
918 	possible = min(possible, sclp_max);
919 	for (cpu = 0; cpu < possible && cpu < nr_cpu_ids; cpu++)
920 		set_cpu_possible(cpu, true);
921 }
922 
923 void __init smp_prepare_cpus(unsigned int max_cpus)
924 {
925 	/* request the 0x1201 emergency signal external interrupt */
926 	if (register_external_irq(EXT_IRQ_EMERGENCY_SIG, do_ext_call_interrupt))
927 		panic("Couldn't request external interrupt 0x1201");
928 	/* request the 0x1202 external call external interrupt */
929 	if (register_external_irq(EXT_IRQ_EXTERNAL_CALL, do_ext_call_interrupt))
930 		panic("Couldn't request external interrupt 0x1202");
931 }
932 
933 void __init smp_prepare_boot_cpu(void)
934 {
935 	struct pcpu *pcpu = pcpu_devices;
936 
937 	WARN_ON(!cpu_present(0) || !cpu_online(0));
938 	pcpu->state = CPU_STATE_CONFIGURED;
939 	pcpu->lowcore = (struct lowcore *)(unsigned long) store_prefix();
940 	S390_lowcore.percpu_offset = __per_cpu_offset[0];
941 	smp_cpu_set_polarization(0, POLARIZATION_UNKNOWN);
942 }
943 
944 void __init smp_cpus_done(unsigned int max_cpus)
945 {
946 }
947 
948 void __init smp_setup_processor_id(void)
949 {
950 	pcpu_devices[0].address = stap();
951 	S390_lowcore.cpu_nr = 0;
952 	S390_lowcore.spinlock_lockval = arch_spin_lockval(0);
953 	S390_lowcore.spinlock_index = 0;
954 }
955 
956 /*
957  * the frequency of the profiling timer can be changed
958  * by writing a multiplier value into /proc/profile.
959  *
960  * usually you want to run this on all CPUs ;)
961  */
962 int setup_profiling_timer(unsigned int multiplier)
963 {
964 	return 0;
965 }
966 
967 #ifdef CONFIG_HOTPLUG_CPU
968 static ssize_t cpu_configure_show(struct device *dev,
969 				  struct device_attribute *attr, char *buf)
970 {
971 	ssize_t count;
972 
973 	mutex_lock(&smp_cpu_state_mutex);
974 	count = sprintf(buf, "%d\n", pcpu_devices[dev->id].state);
975 	mutex_unlock(&smp_cpu_state_mutex);
976 	return count;
977 }
978 
979 static ssize_t cpu_configure_store(struct device *dev,
980 				   struct device_attribute *attr,
981 				   const char *buf, size_t count)
982 {
983 	struct pcpu *pcpu;
984 	int cpu, val, rc, i;
985 	char delim;
986 
987 	if (sscanf(buf, "%d %c", &val, &delim) != 1)
988 		return -EINVAL;
989 	if (val != 0 && val != 1)
990 		return -EINVAL;
991 	get_online_cpus();
992 	mutex_lock(&smp_cpu_state_mutex);
993 	rc = -EBUSY;
994 	/* disallow configuration changes of online cpus and cpu 0 */
995 	cpu = dev->id;
996 	cpu = smp_get_base_cpu(cpu);
997 	if (cpu == 0)
998 		goto out;
999 	for (i = 0; i <= smp_cpu_mtid; i++)
1000 		if (cpu_online(cpu + i))
1001 			goto out;
1002 	pcpu = pcpu_devices + cpu;
1003 	rc = 0;
1004 	switch (val) {
1005 	case 0:
1006 		if (pcpu->state != CPU_STATE_CONFIGURED)
1007 			break;
1008 		rc = sclp_core_deconfigure(pcpu->address >> smp_cpu_mt_shift);
1009 		if (rc)
1010 			break;
1011 		for (i = 0; i <= smp_cpu_mtid; i++) {
1012 			if (cpu + i >= nr_cpu_ids || !cpu_present(cpu + i))
1013 				continue;
1014 			pcpu[i].state = CPU_STATE_STANDBY;
1015 			smp_cpu_set_polarization(cpu + i,
1016 						 POLARIZATION_UNKNOWN);
1017 		}
1018 		topology_expect_change();
1019 		break;
1020 	case 1:
1021 		if (pcpu->state != CPU_STATE_STANDBY)
1022 			break;
1023 		rc = sclp_core_configure(pcpu->address >> smp_cpu_mt_shift);
1024 		if (rc)
1025 			break;
1026 		for (i = 0; i <= smp_cpu_mtid; i++) {
1027 			if (cpu + i >= nr_cpu_ids || !cpu_present(cpu + i))
1028 				continue;
1029 			pcpu[i].state = CPU_STATE_CONFIGURED;
1030 			smp_cpu_set_polarization(cpu + i,
1031 						 POLARIZATION_UNKNOWN);
1032 		}
1033 		topology_expect_change();
1034 		break;
1035 	default:
1036 		break;
1037 	}
1038 out:
1039 	mutex_unlock(&smp_cpu_state_mutex);
1040 	put_online_cpus();
1041 	return rc ? rc : count;
1042 }
1043 static DEVICE_ATTR(configure, 0644, cpu_configure_show, cpu_configure_store);
1044 #endif /* CONFIG_HOTPLUG_CPU */
1045 
1046 static ssize_t show_cpu_address(struct device *dev,
1047 				struct device_attribute *attr, char *buf)
1048 {
1049 	return sprintf(buf, "%d\n", pcpu_devices[dev->id].address);
1050 }
1051 static DEVICE_ATTR(address, 0444, show_cpu_address, NULL);
1052 
1053 static struct attribute *cpu_common_attrs[] = {
1054 #ifdef CONFIG_HOTPLUG_CPU
1055 	&dev_attr_configure.attr,
1056 #endif
1057 	&dev_attr_address.attr,
1058 	NULL,
1059 };
1060 
1061 static struct attribute_group cpu_common_attr_group = {
1062 	.attrs = cpu_common_attrs,
1063 };
1064 
1065 static struct attribute *cpu_online_attrs[] = {
1066 	&dev_attr_idle_count.attr,
1067 	&dev_attr_idle_time_us.attr,
1068 	NULL,
1069 };
1070 
1071 static struct attribute_group cpu_online_attr_group = {
1072 	.attrs = cpu_online_attrs,
1073 };
1074 
1075 static int smp_cpu_online(unsigned int cpu)
1076 {
1077 	struct device *s = &per_cpu(cpu_device, cpu)->dev;
1078 
1079 	return sysfs_create_group(&s->kobj, &cpu_online_attr_group);
1080 }
1081 static int smp_cpu_pre_down(unsigned int cpu)
1082 {
1083 	struct device *s = &per_cpu(cpu_device, cpu)->dev;
1084 
1085 	sysfs_remove_group(&s->kobj, &cpu_online_attr_group);
1086 	return 0;
1087 }
1088 
1089 static int smp_add_present_cpu(int cpu)
1090 {
1091 	struct device *s;
1092 	struct cpu *c;
1093 	int rc;
1094 
1095 	c = kzalloc(sizeof(*c), GFP_KERNEL);
1096 	if (!c)
1097 		return -ENOMEM;
1098 	per_cpu(cpu_device, cpu) = c;
1099 	s = &c->dev;
1100 	c->hotpluggable = 1;
1101 	rc = register_cpu(c, cpu);
1102 	if (rc)
1103 		goto out;
1104 	rc = sysfs_create_group(&s->kobj, &cpu_common_attr_group);
1105 	if (rc)
1106 		goto out_cpu;
1107 	rc = topology_cpu_init(c);
1108 	if (rc)
1109 		goto out_topology;
1110 	return 0;
1111 
1112 out_topology:
1113 	sysfs_remove_group(&s->kobj, &cpu_common_attr_group);
1114 out_cpu:
1115 #ifdef CONFIG_HOTPLUG_CPU
1116 	unregister_cpu(c);
1117 #endif
1118 out:
1119 	return rc;
1120 }
1121 
1122 #ifdef CONFIG_HOTPLUG_CPU
1123 
1124 int __ref smp_rescan_cpus(void)
1125 {
1126 	struct sclp_core_info *info;
1127 	int nr;
1128 
1129 	info = kzalloc(sizeof(*info), GFP_KERNEL);
1130 	if (!info)
1131 		return -ENOMEM;
1132 	smp_get_core_info(info, 0);
1133 	get_online_cpus();
1134 	mutex_lock(&smp_cpu_state_mutex);
1135 	nr = __smp_rescan_cpus(info, 1);
1136 	mutex_unlock(&smp_cpu_state_mutex);
1137 	put_online_cpus();
1138 	kfree(info);
1139 	if (nr)
1140 		topology_schedule_update();
1141 	return 0;
1142 }
1143 
1144 static ssize_t __ref rescan_store(struct device *dev,
1145 				  struct device_attribute *attr,
1146 				  const char *buf,
1147 				  size_t count)
1148 {
1149 	int rc;
1150 
1151 	rc = smp_rescan_cpus();
1152 	return rc ? rc : count;
1153 }
1154 static DEVICE_ATTR(rescan, 0200, NULL, rescan_store);
1155 #endif /* CONFIG_HOTPLUG_CPU */
1156 
1157 static int __init s390_smp_init(void)
1158 {
1159 	int cpu, rc = 0;
1160 
1161 #ifdef CONFIG_HOTPLUG_CPU
1162 	rc = device_create_file(cpu_subsys.dev_root, &dev_attr_rescan);
1163 	if (rc)
1164 		return rc;
1165 #endif
1166 	for_each_present_cpu(cpu) {
1167 		rc = smp_add_present_cpu(cpu);
1168 		if (rc)
1169 			goto out;
1170 	}
1171 
1172 	rc = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "s390/smp:online",
1173 			       smp_cpu_online, smp_cpu_pre_down);
1174 	rc = rc <= 0 ? rc : 0;
1175 out:
1176 	return rc;
1177 }
1178 subsys_initcall(s390_smp_init);
1179