xref: /linux/arch/s390/kernel/smp.c (revision d2912cb15bdda8ba4a5dd73396ad62641af2f520)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  *  SMP related functions
4  *
5  *    Copyright IBM Corp. 1999, 2012
6  *    Author(s): Denis Joseph Barrow,
7  *		 Martin Schwidefsky <schwidefsky@de.ibm.com>,
8  *		 Heiko Carstens <heiko.carstens@de.ibm.com>,
9  *
10  *  based on other smp stuff by
11  *    (c) 1995 Alan Cox, CymruNET Ltd  <alan@cymru.net>
12  *    (c) 1998 Ingo Molnar
13  *
14  * The code outside of smp.c uses logical cpu numbers, only smp.c does
15  * the translation of logical to physical cpu ids. All new code that
16  * operates on physical cpu numbers needs to go into smp.c.
17  */
18 
19 #define KMSG_COMPONENT "cpu"
20 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
21 
22 #include <linux/workqueue.h>
23 #include <linux/memblock.h>
24 #include <linux/export.h>
25 #include <linux/init.h>
26 #include <linux/mm.h>
27 #include <linux/err.h>
28 #include <linux/spinlock.h>
29 #include <linux/kernel_stat.h>
30 #include <linux/delay.h>
31 #include <linux/interrupt.h>
32 #include <linux/irqflags.h>
33 #include <linux/cpu.h>
34 #include <linux/slab.h>
35 #include <linux/sched/hotplug.h>
36 #include <linux/sched/task_stack.h>
37 #include <linux/crash_dump.h>
38 #include <linux/kprobes.h>
39 #include <asm/asm-offsets.h>
40 #include <asm/diag.h>
41 #include <asm/switch_to.h>
42 #include <asm/facility.h>
43 #include <asm/ipl.h>
44 #include <asm/setup.h>
45 #include <asm/irq.h>
46 #include <asm/tlbflush.h>
47 #include <asm/vtimer.h>
48 #include <asm/lowcore.h>
49 #include <asm/sclp.h>
50 #include <asm/vdso.h>
51 #include <asm/debug.h>
52 #include <asm/os_info.h>
53 #include <asm/sigp.h>
54 #include <asm/idle.h>
55 #include <asm/nmi.h>
56 #include <asm/stacktrace.h>
57 #include <asm/topology.h>
58 #include "entry.h"
59 
60 enum {
61 	ec_schedule = 0,
62 	ec_call_function_single,
63 	ec_stop_cpu,
64 };
65 
66 enum {
67 	CPU_STATE_STANDBY,
68 	CPU_STATE_CONFIGURED,
69 };
70 
71 static DEFINE_PER_CPU(struct cpu *, cpu_device);
72 
73 struct pcpu {
74 	struct lowcore *lowcore;	/* lowcore page(s) for the cpu */
75 	unsigned long ec_mask;		/* bit mask for ec_xxx functions */
76 	unsigned long ec_clk;		/* sigp timestamp for ec_xxx */
77 	signed char state;		/* physical cpu state */
78 	signed char polarization;	/* physical polarization */
79 	u16 address;			/* physical cpu address */
80 };
81 
82 static u8 boot_core_type;
83 static struct pcpu pcpu_devices[NR_CPUS];
84 
85 unsigned int smp_cpu_mt_shift;
86 EXPORT_SYMBOL(smp_cpu_mt_shift);
87 
88 unsigned int smp_cpu_mtid;
89 EXPORT_SYMBOL(smp_cpu_mtid);
90 
91 #ifdef CONFIG_CRASH_DUMP
92 __vector128 __initdata boot_cpu_vector_save_area[__NUM_VXRS];
93 #endif
94 
95 static unsigned int smp_max_threads __initdata = -1U;
96 
97 static int __init early_nosmt(char *s)
98 {
99 	smp_max_threads = 1;
100 	return 0;
101 }
102 early_param("nosmt", early_nosmt);
103 
104 static int __init early_smt(char *s)
105 {
106 	get_option(&s, &smp_max_threads);
107 	return 0;
108 }
109 early_param("smt", early_smt);
110 
111 /*
112  * The smp_cpu_state_mutex must be held when changing the state or polarization
113  * member of a pcpu data structure within the pcpu_devices arreay.
114  */
115 DEFINE_MUTEX(smp_cpu_state_mutex);
116 
117 /*
118  * Signal processor helper functions.
119  */
120 static inline int __pcpu_sigp_relax(u16 addr, u8 order, unsigned long parm)
121 {
122 	int cc;
123 
124 	while (1) {
125 		cc = __pcpu_sigp(addr, order, parm, NULL);
126 		if (cc != SIGP_CC_BUSY)
127 			return cc;
128 		cpu_relax();
129 	}
130 }
131 
132 static int pcpu_sigp_retry(struct pcpu *pcpu, u8 order, u32 parm)
133 {
134 	int cc, retry;
135 
136 	for (retry = 0; ; retry++) {
137 		cc = __pcpu_sigp(pcpu->address, order, parm, NULL);
138 		if (cc != SIGP_CC_BUSY)
139 			break;
140 		if (retry >= 3)
141 			udelay(10);
142 	}
143 	return cc;
144 }
145 
146 static inline int pcpu_stopped(struct pcpu *pcpu)
147 {
148 	u32 uninitialized_var(status);
149 
150 	if (__pcpu_sigp(pcpu->address, SIGP_SENSE,
151 			0, &status) != SIGP_CC_STATUS_STORED)
152 		return 0;
153 	return !!(status & (SIGP_STATUS_CHECK_STOP|SIGP_STATUS_STOPPED));
154 }
155 
156 static inline int pcpu_running(struct pcpu *pcpu)
157 {
158 	if (__pcpu_sigp(pcpu->address, SIGP_SENSE_RUNNING,
159 			0, NULL) != SIGP_CC_STATUS_STORED)
160 		return 1;
161 	/* Status stored condition code is equivalent to cpu not running. */
162 	return 0;
163 }
164 
165 /*
166  * Find struct pcpu by cpu address.
167  */
168 static struct pcpu *pcpu_find_address(const struct cpumask *mask, u16 address)
169 {
170 	int cpu;
171 
172 	for_each_cpu(cpu, mask)
173 		if (pcpu_devices[cpu].address == address)
174 			return pcpu_devices + cpu;
175 	return NULL;
176 }
177 
178 static void pcpu_ec_call(struct pcpu *pcpu, int ec_bit)
179 {
180 	int order;
181 
182 	if (test_and_set_bit(ec_bit, &pcpu->ec_mask))
183 		return;
184 	order = pcpu_running(pcpu) ? SIGP_EXTERNAL_CALL : SIGP_EMERGENCY_SIGNAL;
185 	pcpu->ec_clk = get_tod_clock_fast();
186 	pcpu_sigp_retry(pcpu, order, 0);
187 }
188 
189 static int pcpu_alloc_lowcore(struct pcpu *pcpu, int cpu)
190 {
191 	unsigned long async_stack, nodat_stack;
192 	struct lowcore *lc;
193 
194 	if (pcpu != &pcpu_devices[0]) {
195 		pcpu->lowcore =	(struct lowcore *)
196 			__get_free_pages(GFP_KERNEL | GFP_DMA, LC_ORDER);
197 		nodat_stack = __get_free_pages(GFP_KERNEL, THREAD_SIZE_ORDER);
198 		if (!pcpu->lowcore || !nodat_stack)
199 			goto out;
200 	} else {
201 		nodat_stack = pcpu->lowcore->nodat_stack - STACK_INIT_OFFSET;
202 	}
203 	async_stack = stack_alloc();
204 	if (!async_stack)
205 		goto out;
206 	lc = pcpu->lowcore;
207 	memcpy(lc, &S390_lowcore, 512);
208 	memset((char *) lc + 512, 0, sizeof(*lc) - 512);
209 	lc->async_stack = async_stack + STACK_INIT_OFFSET;
210 	lc->nodat_stack = nodat_stack + STACK_INIT_OFFSET;
211 	lc->cpu_nr = cpu;
212 	lc->spinlock_lockval = arch_spin_lockval(cpu);
213 	lc->spinlock_index = 0;
214 	lc->br_r1_trampoline = 0x07f1;	/* br %r1 */
215 	if (nmi_alloc_per_cpu(lc))
216 		goto out_async;
217 	if (vdso_alloc_per_cpu(lc))
218 		goto out_mcesa;
219 	lowcore_ptr[cpu] = lc;
220 	pcpu_sigp_retry(pcpu, SIGP_SET_PREFIX, (u32)(unsigned long) lc);
221 	return 0;
222 
223 out_mcesa:
224 	nmi_free_per_cpu(lc);
225 out_async:
226 	stack_free(async_stack);
227 out:
228 	if (pcpu != &pcpu_devices[0]) {
229 		free_pages(nodat_stack, THREAD_SIZE_ORDER);
230 		free_pages((unsigned long) pcpu->lowcore, LC_ORDER);
231 	}
232 	return -ENOMEM;
233 }
234 
235 #ifdef CONFIG_HOTPLUG_CPU
236 
237 static void pcpu_free_lowcore(struct pcpu *pcpu)
238 {
239 	unsigned long async_stack, nodat_stack, lowcore;
240 
241 	nodat_stack = pcpu->lowcore->nodat_stack - STACK_INIT_OFFSET;
242 	async_stack = pcpu->lowcore->async_stack - STACK_INIT_OFFSET;
243 	lowcore = (unsigned long) pcpu->lowcore;
244 
245 	pcpu_sigp_retry(pcpu, SIGP_SET_PREFIX, 0);
246 	lowcore_ptr[pcpu - pcpu_devices] = NULL;
247 	vdso_free_per_cpu(pcpu->lowcore);
248 	nmi_free_per_cpu(pcpu->lowcore);
249 	stack_free(async_stack);
250 	if (pcpu == &pcpu_devices[0])
251 		return;
252 	free_pages(nodat_stack, THREAD_SIZE_ORDER);
253 	free_pages(lowcore, LC_ORDER);
254 }
255 
256 #endif /* CONFIG_HOTPLUG_CPU */
257 
258 static void pcpu_prepare_secondary(struct pcpu *pcpu, int cpu)
259 {
260 	struct lowcore *lc = pcpu->lowcore;
261 
262 	cpumask_set_cpu(cpu, &init_mm.context.cpu_attach_mask);
263 	cpumask_set_cpu(cpu, mm_cpumask(&init_mm));
264 	lc->cpu_nr = cpu;
265 	lc->spinlock_lockval = arch_spin_lockval(cpu);
266 	lc->spinlock_index = 0;
267 	lc->percpu_offset = __per_cpu_offset[cpu];
268 	lc->kernel_asce = S390_lowcore.kernel_asce;
269 	lc->machine_flags = S390_lowcore.machine_flags;
270 	lc->user_timer = lc->system_timer =
271 		lc->steal_timer = lc->avg_steal_timer = 0;
272 	__ctl_store(lc->cregs_save_area, 0, 15);
273 	save_access_regs((unsigned int *) lc->access_regs_save_area);
274 	memcpy(lc->stfle_fac_list, S390_lowcore.stfle_fac_list,
275 	       sizeof(lc->stfle_fac_list));
276 	memcpy(lc->alt_stfle_fac_list, S390_lowcore.alt_stfle_fac_list,
277 	       sizeof(lc->alt_stfle_fac_list));
278 	arch_spin_lock_setup(cpu);
279 }
280 
281 static void pcpu_attach_task(struct pcpu *pcpu, struct task_struct *tsk)
282 {
283 	struct lowcore *lc = pcpu->lowcore;
284 
285 	lc->kernel_stack = (unsigned long) task_stack_page(tsk)
286 		+ THREAD_SIZE - STACK_FRAME_OVERHEAD - sizeof(struct pt_regs);
287 	lc->current_task = (unsigned long) tsk;
288 	lc->lpp = LPP_MAGIC;
289 	lc->current_pid = tsk->pid;
290 	lc->user_timer = tsk->thread.user_timer;
291 	lc->guest_timer = tsk->thread.guest_timer;
292 	lc->system_timer = tsk->thread.system_timer;
293 	lc->hardirq_timer = tsk->thread.hardirq_timer;
294 	lc->softirq_timer = tsk->thread.softirq_timer;
295 	lc->steal_timer = 0;
296 }
297 
298 static void pcpu_start_fn(struct pcpu *pcpu, void (*func)(void *), void *data)
299 {
300 	struct lowcore *lc = pcpu->lowcore;
301 
302 	lc->restart_stack = lc->nodat_stack;
303 	lc->restart_fn = (unsigned long) func;
304 	lc->restart_data = (unsigned long) data;
305 	lc->restart_source = -1UL;
306 	pcpu_sigp_retry(pcpu, SIGP_RESTART, 0);
307 }
308 
309 /*
310  * Call function via PSW restart on pcpu and stop the current cpu.
311  */
312 static void __pcpu_delegate(void (*func)(void*), void *data)
313 {
314 	func(data);	/* should not return */
315 }
316 
317 static void __no_sanitize_address pcpu_delegate(struct pcpu *pcpu,
318 						void (*func)(void *),
319 						void *data, unsigned long stack)
320 {
321 	struct lowcore *lc = lowcore_ptr[pcpu - pcpu_devices];
322 	unsigned long source_cpu = stap();
323 
324 	__load_psw_mask(PSW_KERNEL_BITS | PSW_MASK_DAT);
325 	if (pcpu->address == source_cpu)
326 		CALL_ON_STACK(__pcpu_delegate, stack, 2, func, data);
327 	/* Stop target cpu (if func returns this stops the current cpu). */
328 	pcpu_sigp_retry(pcpu, SIGP_STOP, 0);
329 	/* Restart func on the target cpu and stop the current cpu. */
330 	mem_assign_absolute(lc->restart_stack, stack);
331 	mem_assign_absolute(lc->restart_fn, (unsigned long) func);
332 	mem_assign_absolute(lc->restart_data, (unsigned long) data);
333 	mem_assign_absolute(lc->restart_source, source_cpu);
334 	__bpon();
335 	asm volatile(
336 		"0:	sigp	0,%0,%2	# sigp restart to target cpu\n"
337 		"	brc	2,0b	# busy, try again\n"
338 		"1:	sigp	0,%1,%3	# sigp stop to current cpu\n"
339 		"	brc	2,1b	# busy, try again\n"
340 		: : "d" (pcpu->address), "d" (source_cpu),
341 		    "K" (SIGP_RESTART), "K" (SIGP_STOP)
342 		: "0", "1", "cc");
343 	for (;;) ;
344 }
345 
346 /*
347  * Enable additional logical cpus for multi-threading.
348  */
349 static int pcpu_set_smt(unsigned int mtid)
350 {
351 	int cc;
352 
353 	if (smp_cpu_mtid == mtid)
354 		return 0;
355 	cc = __pcpu_sigp(0, SIGP_SET_MULTI_THREADING, mtid, NULL);
356 	if (cc == 0) {
357 		smp_cpu_mtid = mtid;
358 		smp_cpu_mt_shift = 0;
359 		while (smp_cpu_mtid >= (1U << smp_cpu_mt_shift))
360 			smp_cpu_mt_shift++;
361 		pcpu_devices[0].address = stap();
362 	}
363 	return cc;
364 }
365 
366 /*
367  * Call function on an online CPU.
368  */
369 void smp_call_online_cpu(void (*func)(void *), void *data)
370 {
371 	struct pcpu *pcpu;
372 
373 	/* Use the current cpu if it is online. */
374 	pcpu = pcpu_find_address(cpu_online_mask, stap());
375 	if (!pcpu)
376 		/* Use the first online cpu. */
377 		pcpu = pcpu_devices + cpumask_first(cpu_online_mask);
378 	pcpu_delegate(pcpu, func, data, (unsigned long) restart_stack);
379 }
380 
381 /*
382  * Call function on the ipl CPU.
383  */
384 void smp_call_ipl_cpu(void (*func)(void *), void *data)
385 {
386 	struct lowcore *lc = pcpu_devices->lowcore;
387 
388 	if (pcpu_devices[0].address == stap())
389 		lc = &S390_lowcore;
390 
391 	pcpu_delegate(&pcpu_devices[0], func, data,
392 		      lc->nodat_stack);
393 }
394 
395 int smp_find_processor_id(u16 address)
396 {
397 	int cpu;
398 
399 	for_each_present_cpu(cpu)
400 		if (pcpu_devices[cpu].address == address)
401 			return cpu;
402 	return -1;
403 }
404 
405 bool arch_vcpu_is_preempted(int cpu)
406 {
407 	if (test_cpu_flag_of(CIF_ENABLED_WAIT, cpu))
408 		return false;
409 	if (pcpu_running(pcpu_devices + cpu))
410 		return false;
411 	return true;
412 }
413 EXPORT_SYMBOL(arch_vcpu_is_preempted);
414 
415 void smp_yield_cpu(int cpu)
416 {
417 	if (MACHINE_HAS_DIAG9C) {
418 		diag_stat_inc_norecursion(DIAG_STAT_X09C);
419 		asm volatile("diag %0,0,0x9c"
420 			     : : "d" (pcpu_devices[cpu].address));
421 	} else if (MACHINE_HAS_DIAG44) {
422 		diag_stat_inc_norecursion(DIAG_STAT_X044);
423 		asm volatile("diag 0,0,0x44");
424 	}
425 }
426 
427 /*
428  * Send cpus emergency shutdown signal. This gives the cpus the
429  * opportunity to complete outstanding interrupts.
430  */
431 void notrace smp_emergency_stop(void)
432 {
433 	cpumask_t cpumask;
434 	u64 end;
435 	int cpu;
436 
437 	cpumask_copy(&cpumask, cpu_online_mask);
438 	cpumask_clear_cpu(smp_processor_id(), &cpumask);
439 
440 	end = get_tod_clock() + (1000000UL << 12);
441 	for_each_cpu(cpu, &cpumask) {
442 		struct pcpu *pcpu = pcpu_devices + cpu;
443 		set_bit(ec_stop_cpu, &pcpu->ec_mask);
444 		while (__pcpu_sigp(pcpu->address, SIGP_EMERGENCY_SIGNAL,
445 				   0, NULL) == SIGP_CC_BUSY &&
446 		       get_tod_clock() < end)
447 			cpu_relax();
448 	}
449 	while (get_tod_clock() < end) {
450 		for_each_cpu(cpu, &cpumask)
451 			if (pcpu_stopped(pcpu_devices + cpu))
452 				cpumask_clear_cpu(cpu, &cpumask);
453 		if (cpumask_empty(&cpumask))
454 			break;
455 		cpu_relax();
456 	}
457 }
458 NOKPROBE_SYMBOL(smp_emergency_stop);
459 
460 /*
461  * Stop all cpus but the current one.
462  */
463 void smp_send_stop(void)
464 {
465 	int cpu;
466 
467 	/* Disable all interrupts/machine checks */
468 	__load_psw_mask(PSW_KERNEL_BITS | PSW_MASK_DAT);
469 	trace_hardirqs_off();
470 
471 	debug_set_critical();
472 
473 	if (oops_in_progress)
474 		smp_emergency_stop();
475 
476 	/* stop all processors */
477 	for_each_online_cpu(cpu) {
478 		if (cpu == smp_processor_id())
479 			continue;
480 		pcpu_sigp_retry(pcpu_devices + cpu, SIGP_STOP, 0);
481 		while (!pcpu_stopped(pcpu_devices + cpu))
482 			cpu_relax();
483 	}
484 }
485 
486 /*
487  * This is the main routine where commands issued by other
488  * cpus are handled.
489  */
490 static void smp_handle_ext_call(void)
491 {
492 	unsigned long bits;
493 
494 	/* handle bit signal external calls */
495 	bits = xchg(&pcpu_devices[smp_processor_id()].ec_mask, 0);
496 	if (test_bit(ec_stop_cpu, &bits))
497 		smp_stop_cpu();
498 	if (test_bit(ec_schedule, &bits))
499 		scheduler_ipi();
500 	if (test_bit(ec_call_function_single, &bits))
501 		generic_smp_call_function_single_interrupt();
502 }
503 
504 static void do_ext_call_interrupt(struct ext_code ext_code,
505 				  unsigned int param32, unsigned long param64)
506 {
507 	inc_irq_stat(ext_code.code == 0x1202 ? IRQEXT_EXC : IRQEXT_EMS);
508 	smp_handle_ext_call();
509 }
510 
511 void arch_send_call_function_ipi_mask(const struct cpumask *mask)
512 {
513 	int cpu;
514 
515 	for_each_cpu(cpu, mask)
516 		pcpu_ec_call(pcpu_devices + cpu, ec_call_function_single);
517 }
518 
519 void arch_send_call_function_single_ipi(int cpu)
520 {
521 	pcpu_ec_call(pcpu_devices + cpu, ec_call_function_single);
522 }
523 
524 /*
525  * this function sends a 'reschedule' IPI to another CPU.
526  * it goes straight through and wastes no time serializing
527  * anything. Worst case is that we lose a reschedule ...
528  */
529 void smp_send_reschedule(int cpu)
530 {
531 	pcpu_ec_call(pcpu_devices + cpu, ec_schedule);
532 }
533 
534 /*
535  * parameter area for the set/clear control bit callbacks
536  */
537 struct ec_creg_mask_parms {
538 	unsigned long orval;
539 	unsigned long andval;
540 	int cr;
541 };
542 
543 /*
544  * callback for setting/clearing control bits
545  */
546 static void smp_ctl_bit_callback(void *info)
547 {
548 	struct ec_creg_mask_parms *pp = info;
549 	unsigned long cregs[16];
550 
551 	__ctl_store(cregs, 0, 15);
552 	cregs[pp->cr] = (cregs[pp->cr] & pp->andval) | pp->orval;
553 	__ctl_load(cregs, 0, 15);
554 }
555 
556 /*
557  * Set a bit in a control register of all cpus
558  */
559 void smp_ctl_set_bit(int cr, int bit)
560 {
561 	struct ec_creg_mask_parms parms = { 1UL << bit, -1UL, cr };
562 
563 	on_each_cpu(smp_ctl_bit_callback, &parms, 1);
564 }
565 EXPORT_SYMBOL(smp_ctl_set_bit);
566 
567 /*
568  * Clear a bit in a control register of all cpus
569  */
570 void smp_ctl_clear_bit(int cr, int bit)
571 {
572 	struct ec_creg_mask_parms parms = { 0, ~(1UL << bit), cr };
573 
574 	on_each_cpu(smp_ctl_bit_callback, &parms, 1);
575 }
576 EXPORT_SYMBOL(smp_ctl_clear_bit);
577 
578 #ifdef CONFIG_CRASH_DUMP
579 
580 int smp_store_status(int cpu)
581 {
582 	struct pcpu *pcpu = pcpu_devices + cpu;
583 	unsigned long pa;
584 
585 	pa = __pa(&pcpu->lowcore->floating_pt_save_area);
586 	if (__pcpu_sigp_relax(pcpu->address, SIGP_STORE_STATUS_AT_ADDRESS,
587 			      pa) != SIGP_CC_ORDER_CODE_ACCEPTED)
588 		return -EIO;
589 	if (!MACHINE_HAS_VX && !MACHINE_HAS_GS)
590 		return 0;
591 	pa = __pa(pcpu->lowcore->mcesad & MCESA_ORIGIN_MASK);
592 	if (MACHINE_HAS_GS)
593 		pa |= pcpu->lowcore->mcesad & MCESA_LC_MASK;
594 	if (__pcpu_sigp_relax(pcpu->address, SIGP_STORE_ADDITIONAL_STATUS,
595 			      pa) != SIGP_CC_ORDER_CODE_ACCEPTED)
596 		return -EIO;
597 	return 0;
598 }
599 
600 /*
601  * Collect CPU state of the previous, crashed system.
602  * There are four cases:
603  * 1) standard zfcp dump
604  *    condition: OLDMEM_BASE == NULL && ipl_info.type == IPL_TYPE_FCP_DUMP
605  *    The state for all CPUs except the boot CPU needs to be collected
606  *    with sigp stop-and-store-status. The boot CPU state is located in
607  *    the absolute lowcore of the memory stored in the HSA. The zcore code
608  *    will copy the boot CPU state from the HSA.
609  * 2) stand-alone kdump for SCSI (zfcp dump with swapped memory)
610  *    condition: OLDMEM_BASE != NULL && ipl_info.type == IPL_TYPE_FCP_DUMP
611  *    The state for all CPUs except the boot CPU needs to be collected
612  *    with sigp stop-and-store-status. The firmware or the boot-loader
613  *    stored the registers of the boot CPU in the absolute lowcore in the
614  *    memory of the old system.
615  * 3) kdump and the old kernel did not store the CPU state,
616  *    or stand-alone kdump for DASD
617  *    condition: OLDMEM_BASE != NULL && !is_kdump_kernel()
618  *    The state for all CPUs except the boot CPU needs to be collected
619  *    with sigp stop-and-store-status. The kexec code or the boot-loader
620  *    stored the registers of the boot CPU in the memory of the old system.
621  * 4) kdump and the old kernel stored the CPU state
622  *    condition: OLDMEM_BASE != NULL && is_kdump_kernel()
623  *    This case does not exist for s390 anymore, setup_arch explicitly
624  *    deactivates the elfcorehdr= kernel parameter
625  */
626 static __init void smp_save_cpu_vxrs(struct save_area *sa, u16 addr,
627 				     bool is_boot_cpu, unsigned long page)
628 {
629 	__vector128 *vxrs = (__vector128 *) page;
630 
631 	if (is_boot_cpu)
632 		vxrs = boot_cpu_vector_save_area;
633 	else
634 		__pcpu_sigp_relax(addr, SIGP_STORE_ADDITIONAL_STATUS, page);
635 	save_area_add_vxrs(sa, vxrs);
636 }
637 
638 static __init void smp_save_cpu_regs(struct save_area *sa, u16 addr,
639 				     bool is_boot_cpu, unsigned long page)
640 {
641 	void *regs = (void *) page;
642 
643 	if (is_boot_cpu)
644 		copy_oldmem_kernel(regs, (void *) __LC_FPREGS_SAVE_AREA, 512);
645 	else
646 		__pcpu_sigp_relax(addr, SIGP_STORE_STATUS_AT_ADDRESS, page);
647 	save_area_add_regs(sa, regs);
648 }
649 
650 void __init smp_save_dump_cpus(void)
651 {
652 	int addr, boot_cpu_addr, max_cpu_addr;
653 	struct save_area *sa;
654 	unsigned long page;
655 	bool is_boot_cpu;
656 
657 	if (!(OLDMEM_BASE || ipl_info.type == IPL_TYPE_FCP_DUMP))
658 		/* No previous system present, normal boot. */
659 		return;
660 	/* Allocate a page as dumping area for the store status sigps */
661 	page = memblock_phys_alloc_range(PAGE_SIZE, PAGE_SIZE, 0, 1UL << 31);
662 	if (!page)
663 		panic("ERROR: Failed to allocate %lx bytes below %lx\n",
664 		      PAGE_SIZE, 1UL << 31);
665 
666 	/* Set multi-threading state to the previous system. */
667 	pcpu_set_smt(sclp.mtid_prev);
668 	boot_cpu_addr = stap();
669 	max_cpu_addr = SCLP_MAX_CORES << sclp.mtid_prev;
670 	for (addr = 0; addr <= max_cpu_addr; addr++) {
671 		if (__pcpu_sigp_relax(addr, SIGP_SENSE, 0) ==
672 		    SIGP_CC_NOT_OPERATIONAL)
673 			continue;
674 		is_boot_cpu = (addr == boot_cpu_addr);
675 		/* Allocate save area */
676 		sa = save_area_alloc(is_boot_cpu);
677 		if (!sa)
678 			panic("could not allocate memory for save area\n");
679 		if (MACHINE_HAS_VX)
680 			/* Get the vector registers */
681 			smp_save_cpu_vxrs(sa, addr, is_boot_cpu, page);
682 		/*
683 		 * For a zfcp dump OLDMEM_BASE == NULL and the registers
684 		 * of the boot CPU are stored in the HSA. To retrieve
685 		 * these registers an SCLP request is required which is
686 		 * done by drivers/s390/char/zcore.c:init_cpu_info()
687 		 */
688 		if (!is_boot_cpu || OLDMEM_BASE)
689 			/* Get the CPU registers */
690 			smp_save_cpu_regs(sa, addr, is_boot_cpu, page);
691 	}
692 	memblock_free(page, PAGE_SIZE);
693 	diag_dma_ops.diag308_reset();
694 	pcpu_set_smt(0);
695 }
696 #endif /* CONFIG_CRASH_DUMP */
697 
698 void smp_cpu_set_polarization(int cpu, int val)
699 {
700 	pcpu_devices[cpu].polarization = val;
701 }
702 
703 int smp_cpu_get_polarization(int cpu)
704 {
705 	return pcpu_devices[cpu].polarization;
706 }
707 
708 static void __ref smp_get_core_info(struct sclp_core_info *info, int early)
709 {
710 	static int use_sigp_detection;
711 	int address;
712 
713 	if (use_sigp_detection || sclp_get_core_info(info, early)) {
714 		use_sigp_detection = 1;
715 		for (address = 0;
716 		     address < (SCLP_MAX_CORES << smp_cpu_mt_shift);
717 		     address += (1U << smp_cpu_mt_shift)) {
718 			if (__pcpu_sigp_relax(address, SIGP_SENSE, 0) ==
719 			    SIGP_CC_NOT_OPERATIONAL)
720 				continue;
721 			info->core[info->configured].core_id =
722 				address >> smp_cpu_mt_shift;
723 			info->configured++;
724 		}
725 		info->combined = info->configured;
726 	}
727 }
728 
729 static int smp_add_present_cpu(int cpu);
730 
731 static int __smp_rescan_cpus(struct sclp_core_info *info, int sysfs_add)
732 {
733 	struct pcpu *pcpu;
734 	cpumask_t avail;
735 	int cpu, nr, i, j;
736 	u16 address;
737 
738 	nr = 0;
739 	cpumask_xor(&avail, cpu_possible_mask, cpu_present_mask);
740 	cpu = cpumask_first(&avail);
741 	for (i = 0; (i < info->combined) && (cpu < nr_cpu_ids); i++) {
742 		if (sclp.has_core_type && info->core[i].type != boot_core_type)
743 			continue;
744 		address = info->core[i].core_id << smp_cpu_mt_shift;
745 		for (j = 0; j <= smp_cpu_mtid; j++) {
746 			if (pcpu_find_address(cpu_present_mask, address + j))
747 				continue;
748 			pcpu = pcpu_devices + cpu;
749 			pcpu->address = address + j;
750 			pcpu->state =
751 				(cpu >= info->configured*(smp_cpu_mtid + 1)) ?
752 				CPU_STATE_STANDBY : CPU_STATE_CONFIGURED;
753 			smp_cpu_set_polarization(cpu, POLARIZATION_UNKNOWN);
754 			set_cpu_present(cpu, true);
755 			if (sysfs_add && smp_add_present_cpu(cpu) != 0)
756 				set_cpu_present(cpu, false);
757 			else
758 				nr++;
759 			cpu = cpumask_next(cpu, &avail);
760 			if (cpu >= nr_cpu_ids)
761 				break;
762 		}
763 	}
764 	return nr;
765 }
766 
767 void __init smp_detect_cpus(void)
768 {
769 	unsigned int cpu, mtid, c_cpus, s_cpus;
770 	struct sclp_core_info *info;
771 	u16 address;
772 
773 	/* Get CPU information */
774 	info = memblock_alloc(sizeof(*info), 8);
775 	if (!info)
776 		panic("%s: Failed to allocate %zu bytes align=0x%x\n",
777 		      __func__, sizeof(*info), 8);
778 	smp_get_core_info(info, 1);
779 	/* Find boot CPU type */
780 	if (sclp.has_core_type) {
781 		address = stap();
782 		for (cpu = 0; cpu < info->combined; cpu++)
783 			if (info->core[cpu].core_id == address) {
784 				/* The boot cpu dictates the cpu type. */
785 				boot_core_type = info->core[cpu].type;
786 				break;
787 			}
788 		if (cpu >= info->combined)
789 			panic("Could not find boot CPU type");
790 	}
791 
792 	/* Set multi-threading state for the current system */
793 	mtid = boot_core_type ? sclp.mtid : sclp.mtid_cp;
794 	mtid = (mtid < smp_max_threads) ? mtid : smp_max_threads - 1;
795 	pcpu_set_smt(mtid);
796 
797 	/* Print number of CPUs */
798 	c_cpus = s_cpus = 0;
799 	for (cpu = 0; cpu < info->combined; cpu++) {
800 		if (sclp.has_core_type &&
801 		    info->core[cpu].type != boot_core_type)
802 			continue;
803 		if (cpu < info->configured)
804 			c_cpus += smp_cpu_mtid + 1;
805 		else
806 			s_cpus += smp_cpu_mtid + 1;
807 	}
808 	pr_info("%d configured CPUs, %d standby CPUs\n", c_cpus, s_cpus);
809 
810 	/* Add CPUs present at boot */
811 	get_online_cpus();
812 	__smp_rescan_cpus(info, 0);
813 	put_online_cpus();
814 	memblock_free_early((unsigned long)info, sizeof(*info));
815 }
816 
817 static void smp_init_secondary(void)
818 {
819 	int cpu = smp_processor_id();
820 
821 	S390_lowcore.last_update_clock = get_tod_clock();
822 	restore_access_regs(S390_lowcore.access_regs_save_area);
823 	cpu_init();
824 	preempt_disable();
825 	init_cpu_timer();
826 	vtime_init();
827 	pfault_init();
828 	notify_cpu_starting(smp_processor_id());
829 	if (topology_cpu_dedicated(cpu))
830 		set_cpu_flag(CIF_DEDICATED_CPU);
831 	else
832 		clear_cpu_flag(CIF_DEDICATED_CPU);
833 	set_cpu_online(smp_processor_id(), true);
834 	inc_irq_stat(CPU_RST);
835 	local_irq_enable();
836 	cpu_startup_entry(CPUHP_AP_ONLINE_IDLE);
837 }
838 
839 /*
840  *	Activate a secondary processor.
841  */
842 static void __no_sanitize_address smp_start_secondary(void *cpuvoid)
843 {
844 	S390_lowcore.restart_stack = (unsigned long) restart_stack;
845 	S390_lowcore.restart_fn = (unsigned long) do_restart;
846 	S390_lowcore.restart_data = 0;
847 	S390_lowcore.restart_source = -1UL;
848 	__ctl_load(S390_lowcore.cregs_save_area, 0, 15);
849 	__load_psw_mask(PSW_KERNEL_BITS | PSW_MASK_DAT);
850 	CALL_ON_STACK(smp_init_secondary, S390_lowcore.kernel_stack, 0);
851 }
852 
853 /* Upping and downing of CPUs */
854 int __cpu_up(unsigned int cpu, struct task_struct *tidle)
855 {
856 	struct pcpu *pcpu;
857 	int base, i, rc;
858 
859 	pcpu = pcpu_devices + cpu;
860 	if (pcpu->state != CPU_STATE_CONFIGURED)
861 		return -EIO;
862 	base = smp_get_base_cpu(cpu);
863 	for (i = 0; i <= smp_cpu_mtid; i++) {
864 		if (base + i < nr_cpu_ids)
865 			if (cpu_online(base + i))
866 				break;
867 	}
868 	/*
869 	 * If this is the first CPU of the core to get online
870 	 * do an initial CPU reset.
871 	 */
872 	if (i > smp_cpu_mtid &&
873 	    pcpu_sigp_retry(pcpu_devices + base, SIGP_INITIAL_CPU_RESET, 0) !=
874 	    SIGP_CC_ORDER_CODE_ACCEPTED)
875 		return -EIO;
876 
877 	rc = pcpu_alloc_lowcore(pcpu, cpu);
878 	if (rc)
879 		return rc;
880 	pcpu_prepare_secondary(pcpu, cpu);
881 	pcpu_attach_task(pcpu, tidle);
882 	pcpu_start_fn(pcpu, smp_start_secondary, NULL);
883 	/* Wait until cpu puts itself in the online & active maps */
884 	while (!cpu_online(cpu))
885 		cpu_relax();
886 	return 0;
887 }
888 
889 static unsigned int setup_possible_cpus __initdata;
890 
891 static int __init _setup_possible_cpus(char *s)
892 {
893 	get_option(&s, &setup_possible_cpus);
894 	return 0;
895 }
896 early_param("possible_cpus", _setup_possible_cpus);
897 
898 #ifdef CONFIG_HOTPLUG_CPU
899 
900 int __cpu_disable(void)
901 {
902 	unsigned long cregs[16];
903 
904 	/* Handle possible pending IPIs */
905 	smp_handle_ext_call();
906 	set_cpu_online(smp_processor_id(), false);
907 	/* Disable pseudo page faults on this cpu. */
908 	pfault_fini();
909 	/* Disable interrupt sources via control register. */
910 	__ctl_store(cregs, 0, 15);
911 	cregs[0]  &= ~0x0000ee70UL;	/* disable all external interrupts */
912 	cregs[6]  &= ~0xff000000UL;	/* disable all I/O interrupts */
913 	cregs[14] &= ~0x1f000000UL;	/* disable most machine checks */
914 	__ctl_load(cregs, 0, 15);
915 	clear_cpu_flag(CIF_NOHZ_DELAY);
916 	return 0;
917 }
918 
919 void __cpu_die(unsigned int cpu)
920 {
921 	struct pcpu *pcpu;
922 
923 	/* Wait until target cpu is down */
924 	pcpu = pcpu_devices + cpu;
925 	while (!pcpu_stopped(pcpu))
926 		cpu_relax();
927 	pcpu_free_lowcore(pcpu);
928 	cpumask_clear_cpu(cpu, mm_cpumask(&init_mm));
929 	cpumask_clear_cpu(cpu, &init_mm.context.cpu_attach_mask);
930 }
931 
932 void __noreturn cpu_die(void)
933 {
934 	idle_task_exit();
935 	__bpon();
936 	pcpu_sigp_retry(pcpu_devices + smp_processor_id(), SIGP_STOP, 0);
937 	for (;;) ;
938 }
939 
940 #endif /* CONFIG_HOTPLUG_CPU */
941 
942 void __init smp_fill_possible_mask(void)
943 {
944 	unsigned int possible, sclp_max, cpu;
945 
946 	sclp_max = max(sclp.mtid, sclp.mtid_cp) + 1;
947 	sclp_max = min(smp_max_threads, sclp_max);
948 	sclp_max = (sclp.max_cores * sclp_max) ?: nr_cpu_ids;
949 	possible = setup_possible_cpus ?: nr_cpu_ids;
950 	possible = min(possible, sclp_max);
951 	for (cpu = 0; cpu < possible && cpu < nr_cpu_ids; cpu++)
952 		set_cpu_possible(cpu, true);
953 }
954 
955 void __init smp_prepare_cpus(unsigned int max_cpus)
956 {
957 	/* request the 0x1201 emergency signal external interrupt */
958 	if (register_external_irq(EXT_IRQ_EMERGENCY_SIG, do_ext_call_interrupt))
959 		panic("Couldn't request external interrupt 0x1201");
960 	/* request the 0x1202 external call external interrupt */
961 	if (register_external_irq(EXT_IRQ_EXTERNAL_CALL, do_ext_call_interrupt))
962 		panic("Couldn't request external interrupt 0x1202");
963 }
964 
965 void __init smp_prepare_boot_cpu(void)
966 {
967 	struct pcpu *pcpu = pcpu_devices;
968 
969 	WARN_ON(!cpu_present(0) || !cpu_online(0));
970 	pcpu->state = CPU_STATE_CONFIGURED;
971 	pcpu->lowcore = (struct lowcore *)(unsigned long) store_prefix();
972 	S390_lowcore.percpu_offset = __per_cpu_offset[0];
973 	smp_cpu_set_polarization(0, POLARIZATION_UNKNOWN);
974 }
975 
976 void __init smp_cpus_done(unsigned int max_cpus)
977 {
978 }
979 
980 void __init smp_setup_processor_id(void)
981 {
982 	pcpu_devices[0].address = stap();
983 	S390_lowcore.cpu_nr = 0;
984 	S390_lowcore.spinlock_lockval = arch_spin_lockval(0);
985 	S390_lowcore.spinlock_index = 0;
986 }
987 
988 /*
989  * the frequency of the profiling timer can be changed
990  * by writing a multiplier value into /proc/profile.
991  *
992  * usually you want to run this on all CPUs ;)
993  */
994 int setup_profiling_timer(unsigned int multiplier)
995 {
996 	return 0;
997 }
998 
999 #ifdef CONFIG_HOTPLUG_CPU
1000 static ssize_t cpu_configure_show(struct device *dev,
1001 				  struct device_attribute *attr, char *buf)
1002 {
1003 	ssize_t count;
1004 
1005 	mutex_lock(&smp_cpu_state_mutex);
1006 	count = sprintf(buf, "%d\n", pcpu_devices[dev->id].state);
1007 	mutex_unlock(&smp_cpu_state_mutex);
1008 	return count;
1009 }
1010 
1011 static ssize_t cpu_configure_store(struct device *dev,
1012 				   struct device_attribute *attr,
1013 				   const char *buf, size_t count)
1014 {
1015 	struct pcpu *pcpu;
1016 	int cpu, val, rc, i;
1017 	char delim;
1018 
1019 	if (sscanf(buf, "%d %c", &val, &delim) != 1)
1020 		return -EINVAL;
1021 	if (val != 0 && val != 1)
1022 		return -EINVAL;
1023 	get_online_cpus();
1024 	mutex_lock(&smp_cpu_state_mutex);
1025 	rc = -EBUSY;
1026 	/* disallow configuration changes of online cpus and cpu 0 */
1027 	cpu = dev->id;
1028 	cpu = smp_get_base_cpu(cpu);
1029 	if (cpu == 0)
1030 		goto out;
1031 	for (i = 0; i <= smp_cpu_mtid; i++)
1032 		if (cpu_online(cpu + i))
1033 			goto out;
1034 	pcpu = pcpu_devices + cpu;
1035 	rc = 0;
1036 	switch (val) {
1037 	case 0:
1038 		if (pcpu->state != CPU_STATE_CONFIGURED)
1039 			break;
1040 		rc = sclp_core_deconfigure(pcpu->address >> smp_cpu_mt_shift);
1041 		if (rc)
1042 			break;
1043 		for (i = 0; i <= smp_cpu_mtid; i++) {
1044 			if (cpu + i >= nr_cpu_ids || !cpu_present(cpu + i))
1045 				continue;
1046 			pcpu[i].state = CPU_STATE_STANDBY;
1047 			smp_cpu_set_polarization(cpu + i,
1048 						 POLARIZATION_UNKNOWN);
1049 		}
1050 		topology_expect_change();
1051 		break;
1052 	case 1:
1053 		if (pcpu->state != CPU_STATE_STANDBY)
1054 			break;
1055 		rc = sclp_core_configure(pcpu->address >> smp_cpu_mt_shift);
1056 		if (rc)
1057 			break;
1058 		for (i = 0; i <= smp_cpu_mtid; i++) {
1059 			if (cpu + i >= nr_cpu_ids || !cpu_present(cpu + i))
1060 				continue;
1061 			pcpu[i].state = CPU_STATE_CONFIGURED;
1062 			smp_cpu_set_polarization(cpu + i,
1063 						 POLARIZATION_UNKNOWN);
1064 		}
1065 		topology_expect_change();
1066 		break;
1067 	default:
1068 		break;
1069 	}
1070 out:
1071 	mutex_unlock(&smp_cpu_state_mutex);
1072 	put_online_cpus();
1073 	return rc ? rc : count;
1074 }
1075 static DEVICE_ATTR(configure, 0644, cpu_configure_show, cpu_configure_store);
1076 #endif /* CONFIG_HOTPLUG_CPU */
1077 
1078 static ssize_t show_cpu_address(struct device *dev,
1079 				struct device_attribute *attr, char *buf)
1080 {
1081 	return sprintf(buf, "%d\n", pcpu_devices[dev->id].address);
1082 }
1083 static DEVICE_ATTR(address, 0444, show_cpu_address, NULL);
1084 
1085 static struct attribute *cpu_common_attrs[] = {
1086 #ifdef CONFIG_HOTPLUG_CPU
1087 	&dev_attr_configure.attr,
1088 #endif
1089 	&dev_attr_address.attr,
1090 	NULL,
1091 };
1092 
1093 static struct attribute_group cpu_common_attr_group = {
1094 	.attrs = cpu_common_attrs,
1095 };
1096 
1097 static struct attribute *cpu_online_attrs[] = {
1098 	&dev_attr_idle_count.attr,
1099 	&dev_attr_idle_time_us.attr,
1100 	NULL,
1101 };
1102 
1103 static struct attribute_group cpu_online_attr_group = {
1104 	.attrs = cpu_online_attrs,
1105 };
1106 
1107 static int smp_cpu_online(unsigned int cpu)
1108 {
1109 	struct device *s = &per_cpu(cpu_device, cpu)->dev;
1110 
1111 	return sysfs_create_group(&s->kobj, &cpu_online_attr_group);
1112 }
1113 static int smp_cpu_pre_down(unsigned int cpu)
1114 {
1115 	struct device *s = &per_cpu(cpu_device, cpu)->dev;
1116 
1117 	sysfs_remove_group(&s->kobj, &cpu_online_attr_group);
1118 	return 0;
1119 }
1120 
1121 static int smp_add_present_cpu(int cpu)
1122 {
1123 	struct device *s;
1124 	struct cpu *c;
1125 	int rc;
1126 
1127 	c = kzalloc(sizeof(*c), GFP_KERNEL);
1128 	if (!c)
1129 		return -ENOMEM;
1130 	per_cpu(cpu_device, cpu) = c;
1131 	s = &c->dev;
1132 	c->hotpluggable = 1;
1133 	rc = register_cpu(c, cpu);
1134 	if (rc)
1135 		goto out;
1136 	rc = sysfs_create_group(&s->kobj, &cpu_common_attr_group);
1137 	if (rc)
1138 		goto out_cpu;
1139 	rc = topology_cpu_init(c);
1140 	if (rc)
1141 		goto out_topology;
1142 	return 0;
1143 
1144 out_topology:
1145 	sysfs_remove_group(&s->kobj, &cpu_common_attr_group);
1146 out_cpu:
1147 #ifdef CONFIG_HOTPLUG_CPU
1148 	unregister_cpu(c);
1149 #endif
1150 out:
1151 	return rc;
1152 }
1153 
1154 #ifdef CONFIG_HOTPLUG_CPU
1155 
1156 int __ref smp_rescan_cpus(void)
1157 {
1158 	struct sclp_core_info *info;
1159 	int nr;
1160 
1161 	info = kzalloc(sizeof(*info), GFP_KERNEL);
1162 	if (!info)
1163 		return -ENOMEM;
1164 	smp_get_core_info(info, 0);
1165 	get_online_cpus();
1166 	mutex_lock(&smp_cpu_state_mutex);
1167 	nr = __smp_rescan_cpus(info, 1);
1168 	mutex_unlock(&smp_cpu_state_mutex);
1169 	put_online_cpus();
1170 	kfree(info);
1171 	if (nr)
1172 		topology_schedule_update();
1173 	return 0;
1174 }
1175 
1176 static ssize_t __ref rescan_store(struct device *dev,
1177 				  struct device_attribute *attr,
1178 				  const char *buf,
1179 				  size_t count)
1180 {
1181 	int rc;
1182 
1183 	rc = lock_device_hotplug_sysfs();
1184 	if (rc)
1185 		return rc;
1186 	rc = smp_rescan_cpus();
1187 	unlock_device_hotplug();
1188 	return rc ? rc : count;
1189 }
1190 static DEVICE_ATTR_WO(rescan);
1191 #endif /* CONFIG_HOTPLUG_CPU */
1192 
1193 static int __init s390_smp_init(void)
1194 {
1195 	int cpu, rc = 0;
1196 
1197 #ifdef CONFIG_HOTPLUG_CPU
1198 	rc = device_create_file(cpu_subsys.dev_root, &dev_attr_rescan);
1199 	if (rc)
1200 		return rc;
1201 #endif
1202 	for_each_present_cpu(cpu) {
1203 		rc = smp_add_present_cpu(cpu);
1204 		if (rc)
1205 			goto out;
1206 	}
1207 
1208 	rc = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "s390/smp:online",
1209 			       smp_cpu_online, smp_cpu_pre_down);
1210 	rc = rc <= 0 ? rc : 0;
1211 out:
1212 	return rc;
1213 }
1214 subsys_initcall(s390_smp_init);
1215