xref: /linux/arch/s390/kernel/smp.c (revision e21f9e2e862e9eb3dd64eaddb6256b3e5098660f)
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/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/memblock.h>
39 #include <linux/kprobes.h>
40 #include <asm/asm-offsets.h>
41 #include <asm/diag.h>
42 #include <asm/switch_to.h>
43 #include <asm/facility.h>
44 #include <asm/ipl.h>
45 #include <asm/setup.h>
46 #include <asm/irq.h>
47 #include <asm/tlbflush.h>
48 #include <asm/vtimer.h>
49 #include <asm/lowcore.h>
50 #include <asm/sclp.h>
51 #include <asm/vdso.h>
52 #include <asm/debug.h>
53 #include <asm/os_info.h>
54 #include <asm/sigp.h>
55 #include <asm/idle.h>
56 #include <asm/nmi.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 #define ASYNC_FRAME_OFFSET (ASYNC_SIZE - STACK_FRAME_OVERHEAD - __PT_SIZE)
190 #define PANIC_FRAME_OFFSET (PAGE_SIZE - STACK_FRAME_OVERHEAD - __PT_SIZE)
191 
192 static int pcpu_alloc_lowcore(struct pcpu *pcpu, int cpu)
193 {
194 	unsigned long async_stack, panic_stack;
195 	struct lowcore *lc;
196 
197 	if (pcpu != &pcpu_devices[0]) {
198 		pcpu->lowcore =	(struct lowcore *)
199 			__get_free_pages(GFP_KERNEL | GFP_DMA, LC_ORDER);
200 		async_stack = __get_free_pages(GFP_KERNEL, ASYNC_ORDER);
201 		panic_stack = __get_free_page(GFP_KERNEL);
202 		if (!pcpu->lowcore || !panic_stack || !async_stack)
203 			goto out;
204 	} else {
205 		async_stack = pcpu->lowcore->async_stack - ASYNC_FRAME_OFFSET;
206 		panic_stack = pcpu->lowcore->panic_stack - PANIC_FRAME_OFFSET;
207 	}
208 	lc = pcpu->lowcore;
209 	memcpy(lc, &S390_lowcore, 512);
210 	memset((char *) lc + 512, 0, sizeof(*lc) - 512);
211 	lc->async_stack = async_stack + ASYNC_FRAME_OFFSET;
212 	lc->panic_stack = panic_stack + PANIC_FRAME_OFFSET;
213 	lc->cpu_nr = cpu;
214 	lc->spinlock_lockval = arch_spin_lockval(cpu);
215 	lc->spinlock_index = 0;
216 	lc->br_r1_trampoline = 0x07f1;	/* br %r1 */
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 	       sizeof(lc->stfle_fac_list));
270 	memcpy(lc->alt_stfle_fac_list, S390_lowcore.alt_stfle_fac_list,
271 	       sizeof(lc->alt_stfle_fac_list));
272 	arch_spin_lock_setup(cpu);
273 }
274 
275 static void pcpu_attach_task(struct pcpu *pcpu, struct task_struct *tsk)
276 {
277 	struct lowcore *lc = pcpu->lowcore;
278 
279 	lc->kernel_stack = (unsigned long) task_stack_page(tsk)
280 		+ THREAD_SIZE - STACK_FRAME_OVERHEAD - sizeof(struct pt_regs);
281 	lc->current_task = (unsigned long) tsk;
282 	lc->lpp = LPP_MAGIC;
283 	lc->current_pid = tsk->pid;
284 	lc->user_timer = tsk->thread.user_timer;
285 	lc->guest_timer = tsk->thread.guest_timer;
286 	lc->system_timer = tsk->thread.system_timer;
287 	lc->hardirq_timer = tsk->thread.hardirq_timer;
288 	lc->softirq_timer = tsk->thread.softirq_timer;
289 	lc->steal_timer = 0;
290 }
291 
292 static void pcpu_start_fn(struct pcpu *pcpu, void (*func)(void *), void *data)
293 {
294 	struct lowcore *lc = pcpu->lowcore;
295 
296 	lc->restart_stack = lc->kernel_stack;
297 	lc->restart_fn = (unsigned long) func;
298 	lc->restart_data = (unsigned long) data;
299 	lc->restart_source = -1UL;
300 	pcpu_sigp_retry(pcpu, SIGP_RESTART, 0);
301 }
302 
303 /*
304  * Call function via PSW restart on pcpu and stop the current cpu.
305  */
306 static void pcpu_delegate(struct pcpu *pcpu, void (*func)(void *),
307 			  void *data, unsigned long stack)
308 {
309 	struct lowcore *lc = lowcore_ptr[pcpu - pcpu_devices];
310 	unsigned long source_cpu = stap();
311 
312 	__load_psw_mask(PSW_KERNEL_BITS);
313 	if (pcpu->address == source_cpu)
314 		func(data);	/* should not return */
315 	/* Stop target cpu (if func returns this stops the current cpu). */
316 	pcpu_sigp_retry(pcpu, SIGP_STOP, 0);
317 	/* Restart func on the target cpu and stop the current cpu. */
318 	mem_assign_absolute(lc->restart_stack, stack);
319 	mem_assign_absolute(lc->restart_fn, (unsigned long) func);
320 	mem_assign_absolute(lc->restart_data, (unsigned long) data);
321 	mem_assign_absolute(lc->restart_source, source_cpu);
322 	__bpon();
323 	asm volatile(
324 		"0:	sigp	0,%0,%2	# sigp restart to target cpu\n"
325 		"	brc	2,0b	# busy, try again\n"
326 		"1:	sigp	0,%1,%3	# sigp stop to current cpu\n"
327 		"	brc	2,1b	# busy, try again\n"
328 		: : "d" (pcpu->address), "d" (source_cpu),
329 		    "K" (SIGP_RESTART), "K" (SIGP_STOP)
330 		: "0", "1", "cc");
331 	for (;;) ;
332 }
333 
334 /*
335  * Enable additional logical cpus for multi-threading.
336  */
337 static int pcpu_set_smt(unsigned int mtid)
338 {
339 	int cc;
340 
341 	if (smp_cpu_mtid == mtid)
342 		return 0;
343 	cc = __pcpu_sigp(0, SIGP_SET_MULTI_THREADING, mtid, NULL);
344 	if (cc == 0) {
345 		smp_cpu_mtid = mtid;
346 		smp_cpu_mt_shift = 0;
347 		while (smp_cpu_mtid >= (1U << smp_cpu_mt_shift))
348 			smp_cpu_mt_shift++;
349 		pcpu_devices[0].address = stap();
350 	}
351 	return cc;
352 }
353 
354 /*
355  * Call function on an online CPU.
356  */
357 void smp_call_online_cpu(void (*func)(void *), void *data)
358 {
359 	struct pcpu *pcpu;
360 
361 	/* Use the current cpu if it is online. */
362 	pcpu = pcpu_find_address(cpu_online_mask, stap());
363 	if (!pcpu)
364 		/* Use the first online cpu. */
365 		pcpu = pcpu_devices + cpumask_first(cpu_online_mask);
366 	pcpu_delegate(pcpu, func, data, (unsigned long) restart_stack);
367 }
368 
369 /*
370  * Call function on the ipl CPU.
371  */
372 void smp_call_ipl_cpu(void (*func)(void *), void *data)
373 {
374 	pcpu_delegate(&pcpu_devices[0], func, data,
375 		      pcpu_devices->lowcore->panic_stack -
376 		      PANIC_FRAME_OFFSET + PAGE_SIZE);
377 }
378 
379 int smp_find_processor_id(u16 address)
380 {
381 	int cpu;
382 
383 	for_each_present_cpu(cpu)
384 		if (pcpu_devices[cpu].address == address)
385 			return cpu;
386 	return -1;
387 }
388 
389 bool arch_vcpu_is_preempted(int cpu)
390 {
391 	if (test_cpu_flag_of(CIF_ENABLED_WAIT, cpu))
392 		return false;
393 	if (pcpu_running(pcpu_devices + cpu))
394 		return false;
395 	return true;
396 }
397 EXPORT_SYMBOL(arch_vcpu_is_preempted);
398 
399 void smp_yield_cpu(int cpu)
400 {
401 	if (MACHINE_HAS_DIAG9C) {
402 		diag_stat_inc_norecursion(DIAG_STAT_X09C);
403 		asm volatile("diag %0,0,0x9c"
404 			     : : "d" (pcpu_devices[cpu].address));
405 	} else if (MACHINE_HAS_DIAG44) {
406 		diag_stat_inc_norecursion(DIAG_STAT_X044);
407 		asm volatile("diag 0,0,0x44");
408 	}
409 }
410 
411 /*
412  * Send cpus emergency shutdown signal. This gives the cpus the
413  * opportunity to complete outstanding interrupts.
414  */
415 void notrace smp_emergency_stop(void)
416 {
417 	cpumask_t cpumask;
418 	u64 end;
419 	int cpu;
420 
421 	cpumask_copy(&cpumask, cpu_online_mask);
422 	cpumask_clear_cpu(smp_processor_id(), &cpumask);
423 
424 	end = get_tod_clock() + (1000000UL << 12);
425 	for_each_cpu(cpu, &cpumask) {
426 		struct pcpu *pcpu = pcpu_devices + cpu;
427 		set_bit(ec_stop_cpu, &pcpu->ec_mask);
428 		while (__pcpu_sigp(pcpu->address, SIGP_EMERGENCY_SIGNAL,
429 				   0, NULL) == SIGP_CC_BUSY &&
430 		       get_tod_clock() < end)
431 			cpu_relax();
432 	}
433 	while (get_tod_clock() < end) {
434 		for_each_cpu(cpu, &cpumask)
435 			if (pcpu_stopped(pcpu_devices + cpu))
436 				cpumask_clear_cpu(cpu, &cpumask);
437 		if (cpumask_empty(&cpumask))
438 			break;
439 		cpu_relax();
440 	}
441 }
442 NOKPROBE_SYMBOL(smp_emergency_stop);
443 
444 /*
445  * Stop all cpus but the current one.
446  */
447 void smp_send_stop(void)
448 {
449 	int cpu;
450 
451 	/* Disable all interrupts/machine checks */
452 	__load_psw_mask(PSW_KERNEL_BITS | PSW_MASK_DAT);
453 	trace_hardirqs_off();
454 
455 	debug_set_critical();
456 
457 	if (oops_in_progress)
458 		smp_emergency_stop();
459 
460 	/* stop all processors */
461 	for_each_online_cpu(cpu) {
462 		if (cpu == smp_processor_id())
463 			continue;
464 		pcpu_sigp_retry(pcpu_devices + cpu, SIGP_STOP, 0);
465 		while (!pcpu_stopped(pcpu_devices + cpu))
466 			cpu_relax();
467 	}
468 }
469 
470 /*
471  * This is the main routine where commands issued by other
472  * cpus are handled.
473  */
474 static void smp_handle_ext_call(void)
475 {
476 	unsigned long bits;
477 
478 	/* handle bit signal external calls */
479 	bits = xchg(&pcpu_devices[smp_processor_id()].ec_mask, 0);
480 	if (test_bit(ec_stop_cpu, &bits))
481 		smp_stop_cpu();
482 	if (test_bit(ec_schedule, &bits))
483 		scheduler_ipi();
484 	if (test_bit(ec_call_function_single, &bits))
485 		generic_smp_call_function_single_interrupt();
486 }
487 
488 static void do_ext_call_interrupt(struct ext_code ext_code,
489 				  unsigned int param32, unsigned long param64)
490 {
491 	inc_irq_stat(ext_code.code == 0x1202 ? IRQEXT_EXC : IRQEXT_EMS);
492 	smp_handle_ext_call();
493 }
494 
495 void arch_send_call_function_ipi_mask(const struct cpumask *mask)
496 {
497 	int cpu;
498 
499 	for_each_cpu(cpu, mask)
500 		pcpu_ec_call(pcpu_devices + cpu, ec_call_function_single);
501 }
502 
503 void arch_send_call_function_single_ipi(int cpu)
504 {
505 	pcpu_ec_call(pcpu_devices + cpu, ec_call_function_single);
506 }
507 
508 /*
509  * this function sends a 'reschedule' IPI to another CPU.
510  * it goes straight through and wastes no time serializing
511  * anything. Worst case is that we lose a reschedule ...
512  */
513 void smp_send_reschedule(int cpu)
514 {
515 	pcpu_ec_call(pcpu_devices + cpu, ec_schedule);
516 }
517 
518 /*
519  * parameter area for the set/clear control bit callbacks
520  */
521 struct ec_creg_mask_parms {
522 	unsigned long orval;
523 	unsigned long andval;
524 	int cr;
525 };
526 
527 /*
528  * callback for setting/clearing control bits
529  */
530 static void smp_ctl_bit_callback(void *info)
531 {
532 	struct ec_creg_mask_parms *pp = info;
533 	unsigned long cregs[16];
534 
535 	__ctl_store(cregs, 0, 15);
536 	cregs[pp->cr] = (cregs[pp->cr] & pp->andval) | pp->orval;
537 	__ctl_load(cregs, 0, 15);
538 }
539 
540 /*
541  * Set a bit in a control register of all cpus
542  */
543 void smp_ctl_set_bit(int cr, int bit)
544 {
545 	struct ec_creg_mask_parms parms = { 1UL << bit, -1UL, cr };
546 
547 	on_each_cpu(smp_ctl_bit_callback, &parms, 1);
548 }
549 EXPORT_SYMBOL(smp_ctl_set_bit);
550 
551 /*
552  * Clear a bit in a control register of all cpus
553  */
554 void smp_ctl_clear_bit(int cr, int bit)
555 {
556 	struct ec_creg_mask_parms parms = { 0, ~(1UL << bit), cr };
557 
558 	on_each_cpu(smp_ctl_bit_callback, &parms, 1);
559 }
560 EXPORT_SYMBOL(smp_ctl_clear_bit);
561 
562 #ifdef CONFIG_CRASH_DUMP
563 
564 int smp_store_status(int cpu)
565 {
566 	struct pcpu *pcpu = pcpu_devices + cpu;
567 	unsigned long pa;
568 
569 	pa = __pa(&pcpu->lowcore->floating_pt_save_area);
570 	if (__pcpu_sigp_relax(pcpu->address, SIGP_STORE_STATUS_AT_ADDRESS,
571 			      pa) != SIGP_CC_ORDER_CODE_ACCEPTED)
572 		return -EIO;
573 	if (!MACHINE_HAS_VX && !MACHINE_HAS_GS)
574 		return 0;
575 	pa = __pa(pcpu->lowcore->mcesad & MCESA_ORIGIN_MASK);
576 	if (MACHINE_HAS_GS)
577 		pa |= pcpu->lowcore->mcesad & MCESA_LC_MASK;
578 	if (__pcpu_sigp_relax(pcpu->address, SIGP_STORE_ADDITIONAL_STATUS,
579 			      pa) != SIGP_CC_ORDER_CODE_ACCEPTED)
580 		return -EIO;
581 	return 0;
582 }
583 
584 /*
585  * Collect CPU state of the previous, crashed system.
586  * There are four cases:
587  * 1) standard zfcp dump
588  *    condition: OLDMEM_BASE == NULL && ipl_info.type == IPL_TYPE_FCP_DUMP
589  *    The state for all CPUs except the boot CPU needs to be collected
590  *    with sigp stop-and-store-status. The boot CPU state is located in
591  *    the absolute lowcore of the memory stored in the HSA. The zcore code
592  *    will copy the boot CPU state from the HSA.
593  * 2) stand-alone kdump for SCSI (zfcp dump with swapped memory)
594  *    condition: OLDMEM_BASE != NULL && ipl_info.type == IPL_TYPE_FCP_DUMP
595  *    The state for all CPUs except the boot CPU needs to be collected
596  *    with sigp stop-and-store-status. The firmware or the boot-loader
597  *    stored the registers of the boot CPU in the absolute lowcore in the
598  *    memory of the old system.
599  * 3) kdump and the old kernel did not store the CPU state,
600  *    or stand-alone kdump for DASD
601  *    condition: OLDMEM_BASE != NULL && !is_kdump_kernel()
602  *    The state for all CPUs except the boot CPU needs to be collected
603  *    with sigp stop-and-store-status. The kexec code or the boot-loader
604  *    stored the registers of the boot CPU in the memory of the old system.
605  * 4) kdump and the old kernel stored the CPU state
606  *    condition: OLDMEM_BASE != NULL && is_kdump_kernel()
607  *    This case does not exist for s390 anymore, setup_arch explicitly
608  *    deactivates the elfcorehdr= kernel parameter
609  */
610 static __init void smp_save_cpu_vxrs(struct save_area *sa, u16 addr,
611 				     bool is_boot_cpu, unsigned long page)
612 {
613 	__vector128 *vxrs = (__vector128 *) page;
614 
615 	if (is_boot_cpu)
616 		vxrs = boot_cpu_vector_save_area;
617 	else
618 		__pcpu_sigp_relax(addr, SIGP_STORE_ADDITIONAL_STATUS, page);
619 	save_area_add_vxrs(sa, vxrs);
620 }
621 
622 static __init void smp_save_cpu_regs(struct save_area *sa, u16 addr,
623 				     bool is_boot_cpu, unsigned long page)
624 {
625 	void *regs = (void *) page;
626 
627 	if (is_boot_cpu)
628 		copy_oldmem_kernel(regs, (void *) __LC_FPREGS_SAVE_AREA, 512);
629 	else
630 		__pcpu_sigp_relax(addr, SIGP_STORE_STATUS_AT_ADDRESS, page);
631 	save_area_add_regs(sa, regs);
632 }
633 
634 void __init smp_save_dump_cpus(void)
635 {
636 	int addr, boot_cpu_addr, max_cpu_addr;
637 	struct save_area *sa;
638 	unsigned long page;
639 	bool is_boot_cpu;
640 
641 	if (!(OLDMEM_BASE || ipl_info.type == IPL_TYPE_FCP_DUMP))
642 		/* No previous system present, normal boot. */
643 		return;
644 	/* Allocate a page as dumping area for the store status sigps */
645 	page = memblock_alloc_base(PAGE_SIZE, PAGE_SIZE, 1UL << 31);
646 	/* Set multi-threading state to the previous system. */
647 	pcpu_set_smt(sclp.mtid_prev);
648 	boot_cpu_addr = stap();
649 	max_cpu_addr = SCLP_MAX_CORES << sclp.mtid_prev;
650 	for (addr = 0; addr <= max_cpu_addr; addr++) {
651 		if (__pcpu_sigp_relax(addr, SIGP_SENSE, 0) ==
652 		    SIGP_CC_NOT_OPERATIONAL)
653 			continue;
654 		is_boot_cpu = (addr == boot_cpu_addr);
655 		/* Allocate save area */
656 		sa = save_area_alloc(is_boot_cpu);
657 		if (!sa)
658 			panic("could not allocate memory for save area\n");
659 		if (MACHINE_HAS_VX)
660 			/* Get the vector registers */
661 			smp_save_cpu_vxrs(sa, addr, is_boot_cpu, page);
662 		/*
663 		 * For a zfcp dump OLDMEM_BASE == NULL and the registers
664 		 * of the boot CPU are stored in the HSA. To retrieve
665 		 * these registers an SCLP request is required which is
666 		 * done by drivers/s390/char/zcore.c:init_cpu_info()
667 		 */
668 		if (!is_boot_cpu || OLDMEM_BASE)
669 			/* Get the CPU registers */
670 			smp_save_cpu_regs(sa, addr, is_boot_cpu, page);
671 	}
672 	memblock_free(page, PAGE_SIZE);
673 	diag308_reset();
674 	pcpu_set_smt(0);
675 }
676 #endif /* CONFIG_CRASH_DUMP */
677 
678 void smp_cpu_set_polarization(int cpu, int val)
679 {
680 	pcpu_devices[cpu].polarization = val;
681 }
682 
683 int smp_cpu_get_polarization(int cpu)
684 {
685 	return pcpu_devices[cpu].polarization;
686 }
687 
688 static void __ref smp_get_core_info(struct sclp_core_info *info, int early)
689 {
690 	static int use_sigp_detection;
691 	int address;
692 
693 	if (use_sigp_detection || sclp_get_core_info(info, early)) {
694 		use_sigp_detection = 1;
695 		for (address = 0;
696 		     address < (SCLP_MAX_CORES << smp_cpu_mt_shift);
697 		     address += (1U << smp_cpu_mt_shift)) {
698 			if (__pcpu_sigp_relax(address, SIGP_SENSE, 0) ==
699 			    SIGP_CC_NOT_OPERATIONAL)
700 				continue;
701 			info->core[info->configured].core_id =
702 				address >> smp_cpu_mt_shift;
703 			info->configured++;
704 		}
705 		info->combined = info->configured;
706 	}
707 }
708 
709 static int smp_add_present_cpu(int cpu);
710 
711 static int __smp_rescan_cpus(struct sclp_core_info *info, int sysfs_add)
712 {
713 	struct pcpu *pcpu;
714 	cpumask_t avail;
715 	int cpu, nr, i, j;
716 	u16 address;
717 
718 	nr = 0;
719 	cpumask_xor(&avail, cpu_possible_mask, cpu_present_mask);
720 	cpu = cpumask_first(&avail);
721 	for (i = 0; (i < info->combined) && (cpu < nr_cpu_ids); i++) {
722 		if (sclp.has_core_type && info->core[i].type != boot_core_type)
723 			continue;
724 		address = info->core[i].core_id << smp_cpu_mt_shift;
725 		for (j = 0; j <= smp_cpu_mtid; j++) {
726 			if (pcpu_find_address(cpu_present_mask, address + j))
727 				continue;
728 			pcpu = pcpu_devices + cpu;
729 			pcpu->address = address + j;
730 			pcpu->state =
731 				(cpu >= info->configured*(smp_cpu_mtid + 1)) ?
732 				CPU_STATE_STANDBY : CPU_STATE_CONFIGURED;
733 			smp_cpu_set_polarization(cpu, POLARIZATION_UNKNOWN);
734 			set_cpu_present(cpu, true);
735 			if (sysfs_add && smp_add_present_cpu(cpu) != 0)
736 				set_cpu_present(cpu, false);
737 			else
738 				nr++;
739 			cpu = cpumask_next(cpu, &avail);
740 			if (cpu >= nr_cpu_ids)
741 				break;
742 		}
743 	}
744 	return nr;
745 }
746 
747 void __init smp_detect_cpus(void)
748 {
749 	unsigned int cpu, mtid, c_cpus, s_cpus;
750 	struct sclp_core_info *info;
751 	u16 address;
752 
753 	/* Get CPU information */
754 	info = memblock_virt_alloc(sizeof(*info), 8);
755 	smp_get_core_info(info, 1);
756 	/* Find boot CPU type */
757 	if (sclp.has_core_type) {
758 		address = stap();
759 		for (cpu = 0; cpu < info->combined; cpu++)
760 			if (info->core[cpu].core_id == address) {
761 				/* The boot cpu dictates the cpu type. */
762 				boot_core_type = info->core[cpu].type;
763 				break;
764 			}
765 		if (cpu >= info->combined)
766 			panic("Could not find boot CPU type");
767 	}
768 
769 	/* Set multi-threading state for the current system */
770 	mtid = boot_core_type ? sclp.mtid : sclp.mtid_cp;
771 	mtid = (mtid < smp_max_threads) ? mtid : smp_max_threads - 1;
772 	pcpu_set_smt(mtid);
773 
774 	/* Print number of CPUs */
775 	c_cpus = s_cpus = 0;
776 	for (cpu = 0; cpu < info->combined; cpu++) {
777 		if (sclp.has_core_type &&
778 		    info->core[cpu].type != boot_core_type)
779 			continue;
780 		if (cpu < info->configured)
781 			c_cpus += smp_cpu_mtid + 1;
782 		else
783 			s_cpus += smp_cpu_mtid + 1;
784 	}
785 	pr_info("%d configured CPUs, %d standby CPUs\n", c_cpus, s_cpus);
786 
787 	/* Add CPUs present at boot */
788 	get_online_cpus();
789 	__smp_rescan_cpus(info, 0);
790 	put_online_cpus();
791 	memblock_free_early((unsigned long)info, sizeof(*info));
792 }
793 
794 /*
795  *	Activate a secondary processor.
796  */
797 static void smp_start_secondary(void *cpuvoid)
798 {
799 	int cpu = smp_processor_id();
800 
801 	S390_lowcore.last_update_clock = get_tod_clock();
802 	S390_lowcore.restart_stack = (unsigned long) restart_stack;
803 	S390_lowcore.restart_fn = (unsigned long) do_restart;
804 	S390_lowcore.restart_data = 0;
805 	S390_lowcore.restart_source = -1UL;
806 	restore_access_regs(S390_lowcore.access_regs_save_area);
807 	__ctl_load(S390_lowcore.cregs_save_area, 0, 15);
808 	__load_psw_mask(PSW_KERNEL_BITS | PSW_MASK_DAT);
809 	cpu_init();
810 	preempt_disable();
811 	init_cpu_timer();
812 	vtime_init();
813 	pfault_init();
814 	notify_cpu_starting(cpu);
815 	if (topology_cpu_dedicated(cpu))
816 		set_cpu_flag(CIF_DEDICATED_CPU);
817 	else
818 		clear_cpu_flag(CIF_DEDICATED_CPU);
819 	set_cpu_online(cpu, true);
820 	inc_irq_stat(CPU_RST);
821 	local_irq_enable();
822 	cpu_startup_entry(CPUHP_AP_ONLINE_IDLE);
823 }
824 
825 /* Upping and downing of CPUs */
826 int __cpu_up(unsigned int cpu, struct task_struct *tidle)
827 {
828 	struct pcpu *pcpu;
829 	int base, i, rc;
830 
831 	pcpu = pcpu_devices + cpu;
832 	if (pcpu->state != CPU_STATE_CONFIGURED)
833 		return -EIO;
834 	base = smp_get_base_cpu(cpu);
835 	for (i = 0; i <= smp_cpu_mtid; i++) {
836 		if (base + i < nr_cpu_ids)
837 			if (cpu_online(base + i))
838 				break;
839 	}
840 	/*
841 	 * If this is the first CPU of the core to get online
842 	 * do an initial CPU reset.
843 	 */
844 	if (i > smp_cpu_mtid &&
845 	    pcpu_sigp_retry(pcpu_devices + base, SIGP_INITIAL_CPU_RESET, 0) !=
846 	    SIGP_CC_ORDER_CODE_ACCEPTED)
847 		return -EIO;
848 
849 	rc = pcpu_alloc_lowcore(pcpu, cpu);
850 	if (rc)
851 		return rc;
852 	pcpu_prepare_secondary(pcpu, cpu);
853 	pcpu_attach_task(pcpu, tidle);
854 	pcpu_start_fn(pcpu, smp_start_secondary, NULL);
855 	/* Wait until cpu puts itself in the online & active maps */
856 	while (!cpu_online(cpu))
857 		cpu_relax();
858 	return 0;
859 }
860 
861 static unsigned int setup_possible_cpus __initdata;
862 
863 static int __init _setup_possible_cpus(char *s)
864 {
865 	get_option(&s, &setup_possible_cpus);
866 	return 0;
867 }
868 early_param("possible_cpus", _setup_possible_cpus);
869 
870 #ifdef CONFIG_HOTPLUG_CPU
871 
872 int __cpu_disable(void)
873 {
874 	unsigned long cregs[16];
875 
876 	/* Handle possible pending IPIs */
877 	smp_handle_ext_call();
878 	set_cpu_online(smp_processor_id(), false);
879 	/* Disable pseudo page faults on this cpu. */
880 	pfault_fini();
881 	/* Disable interrupt sources via control register. */
882 	__ctl_store(cregs, 0, 15);
883 	cregs[0]  &= ~0x0000ee70UL;	/* disable all external interrupts */
884 	cregs[6]  &= ~0xff000000UL;	/* disable all I/O interrupts */
885 	cregs[14] &= ~0x1f000000UL;	/* disable most machine checks */
886 	__ctl_load(cregs, 0, 15);
887 	clear_cpu_flag(CIF_NOHZ_DELAY);
888 	return 0;
889 }
890 
891 void __cpu_die(unsigned int cpu)
892 {
893 	struct pcpu *pcpu;
894 
895 	/* Wait until target cpu is down */
896 	pcpu = pcpu_devices + cpu;
897 	while (!pcpu_stopped(pcpu))
898 		cpu_relax();
899 	pcpu_free_lowcore(pcpu);
900 	cpumask_clear_cpu(cpu, mm_cpumask(&init_mm));
901 	cpumask_clear_cpu(cpu, &init_mm.context.cpu_attach_mask);
902 }
903 
904 void __noreturn cpu_die(void)
905 {
906 	idle_task_exit();
907 	__bpon();
908 	pcpu_sigp_retry(pcpu_devices + smp_processor_id(), SIGP_STOP, 0);
909 	for (;;) ;
910 }
911 
912 #endif /* CONFIG_HOTPLUG_CPU */
913 
914 void __init smp_fill_possible_mask(void)
915 {
916 	unsigned int possible, sclp_max, cpu;
917 
918 	sclp_max = max(sclp.mtid, sclp.mtid_cp) + 1;
919 	sclp_max = min(smp_max_threads, sclp_max);
920 	sclp_max = (sclp.max_cores * sclp_max) ?: nr_cpu_ids;
921 	possible = setup_possible_cpus ?: nr_cpu_ids;
922 	possible = min(possible, sclp_max);
923 	for (cpu = 0; cpu < possible && cpu < nr_cpu_ids; cpu++)
924 		set_cpu_possible(cpu, true);
925 }
926 
927 void __init smp_prepare_cpus(unsigned int max_cpus)
928 {
929 	/* request the 0x1201 emergency signal external interrupt */
930 	if (register_external_irq(EXT_IRQ_EMERGENCY_SIG, do_ext_call_interrupt))
931 		panic("Couldn't request external interrupt 0x1201");
932 	/* request the 0x1202 external call external interrupt */
933 	if (register_external_irq(EXT_IRQ_EXTERNAL_CALL, do_ext_call_interrupt))
934 		panic("Couldn't request external interrupt 0x1202");
935 }
936 
937 void __init smp_prepare_boot_cpu(void)
938 {
939 	struct pcpu *pcpu = pcpu_devices;
940 
941 	WARN_ON(!cpu_present(0) || !cpu_online(0));
942 	pcpu->state = CPU_STATE_CONFIGURED;
943 	pcpu->lowcore = (struct lowcore *)(unsigned long) store_prefix();
944 	S390_lowcore.percpu_offset = __per_cpu_offset[0];
945 	smp_cpu_set_polarization(0, POLARIZATION_UNKNOWN);
946 }
947 
948 void __init smp_cpus_done(unsigned int max_cpus)
949 {
950 }
951 
952 void __init smp_setup_processor_id(void)
953 {
954 	pcpu_devices[0].address = stap();
955 	S390_lowcore.cpu_nr = 0;
956 	S390_lowcore.spinlock_lockval = arch_spin_lockval(0);
957 	S390_lowcore.spinlock_index = 0;
958 }
959 
960 /*
961  * the frequency of the profiling timer can be changed
962  * by writing a multiplier value into /proc/profile.
963  *
964  * usually you want to run this on all CPUs ;)
965  */
966 int setup_profiling_timer(unsigned int multiplier)
967 {
968 	return 0;
969 }
970 
971 #ifdef CONFIG_HOTPLUG_CPU
972 static ssize_t cpu_configure_show(struct device *dev,
973 				  struct device_attribute *attr, char *buf)
974 {
975 	ssize_t count;
976 
977 	mutex_lock(&smp_cpu_state_mutex);
978 	count = sprintf(buf, "%d\n", pcpu_devices[dev->id].state);
979 	mutex_unlock(&smp_cpu_state_mutex);
980 	return count;
981 }
982 
983 static ssize_t cpu_configure_store(struct device *dev,
984 				   struct device_attribute *attr,
985 				   const char *buf, size_t count)
986 {
987 	struct pcpu *pcpu;
988 	int cpu, val, rc, i;
989 	char delim;
990 
991 	if (sscanf(buf, "%d %c", &val, &delim) != 1)
992 		return -EINVAL;
993 	if (val != 0 && val != 1)
994 		return -EINVAL;
995 	get_online_cpus();
996 	mutex_lock(&smp_cpu_state_mutex);
997 	rc = -EBUSY;
998 	/* disallow configuration changes of online cpus and cpu 0 */
999 	cpu = dev->id;
1000 	cpu = smp_get_base_cpu(cpu);
1001 	if (cpu == 0)
1002 		goto out;
1003 	for (i = 0; i <= smp_cpu_mtid; i++)
1004 		if (cpu_online(cpu + i))
1005 			goto out;
1006 	pcpu = pcpu_devices + cpu;
1007 	rc = 0;
1008 	switch (val) {
1009 	case 0:
1010 		if (pcpu->state != CPU_STATE_CONFIGURED)
1011 			break;
1012 		rc = sclp_core_deconfigure(pcpu->address >> smp_cpu_mt_shift);
1013 		if (rc)
1014 			break;
1015 		for (i = 0; i <= smp_cpu_mtid; i++) {
1016 			if (cpu + i >= nr_cpu_ids || !cpu_present(cpu + i))
1017 				continue;
1018 			pcpu[i].state = CPU_STATE_STANDBY;
1019 			smp_cpu_set_polarization(cpu + i,
1020 						 POLARIZATION_UNKNOWN);
1021 		}
1022 		topology_expect_change();
1023 		break;
1024 	case 1:
1025 		if (pcpu->state != CPU_STATE_STANDBY)
1026 			break;
1027 		rc = sclp_core_configure(pcpu->address >> smp_cpu_mt_shift);
1028 		if (rc)
1029 			break;
1030 		for (i = 0; i <= smp_cpu_mtid; i++) {
1031 			if (cpu + i >= nr_cpu_ids || !cpu_present(cpu + i))
1032 				continue;
1033 			pcpu[i].state = CPU_STATE_CONFIGURED;
1034 			smp_cpu_set_polarization(cpu + i,
1035 						 POLARIZATION_UNKNOWN);
1036 		}
1037 		topology_expect_change();
1038 		break;
1039 	default:
1040 		break;
1041 	}
1042 out:
1043 	mutex_unlock(&smp_cpu_state_mutex);
1044 	put_online_cpus();
1045 	return rc ? rc : count;
1046 }
1047 static DEVICE_ATTR(configure, 0644, cpu_configure_show, cpu_configure_store);
1048 #endif /* CONFIG_HOTPLUG_CPU */
1049 
1050 static ssize_t show_cpu_address(struct device *dev,
1051 				struct device_attribute *attr, char *buf)
1052 {
1053 	return sprintf(buf, "%d\n", pcpu_devices[dev->id].address);
1054 }
1055 static DEVICE_ATTR(address, 0444, show_cpu_address, NULL);
1056 
1057 static struct attribute *cpu_common_attrs[] = {
1058 #ifdef CONFIG_HOTPLUG_CPU
1059 	&dev_attr_configure.attr,
1060 #endif
1061 	&dev_attr_address.attr,
1062 	NULL,
1063 };
1064 
1065 static struct attribute_group cpu_common_attr_group = {
1066 	.attrs = cpu_common_attrs,
1067 };
1068 
1069 static struct attribute *cpu_online_attrs[] = {
1070 	&dev_attr_idle_count.attr,
1071 	&dev_attr_idle_time_us.attr,
1072 	NULL,
1073 };
1074 
1075 static struct attribute_group cpu_online_attr_group = {
1076 	.attrs = cpu_online_attrs,
1077 };
1078 
1079 static int smp_cpu_online(unsigned int cpu)
1080 {
1081 	struct device *s = &per_cpu(cpu_device, cpu)->dev;
1082 
1083 	return sysfs_create_group(&s->kobj, &cpu_online_attr_group);
1084 }
1085 static int smp_cpu_pre_down(unsigned int cpu)
1086 {
1087 	struct device *s = &per_cpu(cpu_device, cpu)->dev;
1088 
1089 	sysfs_remove_group(&s->kobj, &cpu_online_attr_group);
1090 	return 0;
1091 }
1092 
1093 static int smp_add_present_cpu(int cpu)
1094 {
1095 	struct device *s;
1096 	struct cpu *c;
1097 	int rc;
1098 
1099 	c = kzalloc(sizeof(*c), GFP_KERNEL);
1100 	if (!c)
1101 		return -ENOMEM;
1102 	per_cpu(cpu_device, cpu) = c;
1103 	s = &c->dev;
1104 	c->hotpluggable = 1;
1105 	rc = register_cpu(c, cpu);
1106 	if (rc)
1107 		goto out;
1108 	rc = sysfs_create_group(&s->kobj, &cpu_common_attr_group);
1109 	if (rc)
1110 		goto out_cpu;
1111 	rc = topology_cpu_init(c);
1112 	if (rc)
1113 		goto out_topology;
1114 	return 0;
1115 
1116 out_topology:
1117 	sysfs_remove_group(&s->kobj, &cpu_common_attr_group);
1118 out_cpu:
1119 #ifdef CONFIG_HOTPLUG_CPU
1120 	unregister_cpu(c);
1121 #endif
1122 out:
1123 	return rc;
1124 }
1125 
1126 #ifdef CONFIG_HOTPLUG_CPU
1127 
1128 int __ref smp_rescan_cpus(void)
1129 {
1130 	struct sclp_core_info *info;
1131 	int nr;
1132 
1133 	info = kzalloc(sizeof(*info), GFP_KERNEL);
1134 	if (!info)
1135 		return -ENOMEM;
1136 	smp_get_core_info(info, 0);
1137 	get_online_cpus();
1138 	mutex_lock(&smp_cpu_state_mutex);
1139 	nr = __smp_rescan_cpus(info, 1);
1140 	mutex_unlock(&smp_cpu_state_mutex);
1141 	put_online_cpus();
1142 	kfree(info);
1143 	if (nr)
1144 		topology_schedule_update();
1145 	return 0;
1146 }
1147 
1148 static ssize_t __ref rescan_store(struct device *dev,
1149 				  struct device_attribute *attr,
1150 				  const char *buf,
1151 				  size_t count)
1152 {
1153 	int rc;
1154 
1155 	rc = smp_rescan_cpus();
1156 	return rc ? rc : count;
1157 }
1158 static DEVICE_ATTR_WO(rescan);
1159 #endif /* CONFIG_HOTPLUG_CPU */
1160 
1161 static int __init s390_smp_init(void)
1162 {
1163 	int cpu, rc = 0;
1164 
1165 #ifdef CONFIG_HOTPLUG_CPU
1166 	rc = device_create_file(cpu_subsys.dev_root, &dev_attr_rescan);
1167 	if (rc)
1168 		return rc;
1169 #endif
1170 	for_each_present_cpu(cpu) {
1171 		rc = smp_add_present_cpu(cpu);
1172 		if (rc)
1173 			goto out;
1174 	}
1175 
1176 	rc = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "s390/smp:online",
1177 			       smp_cpu_online, smp_cpu_pre_down);
1178 	rc = rc <= 0 ? rc : 0;
1179 out:
1180 	return rc;
1181 }
1182 subsys_initcall(s390_smp_init);
1183