xref: /linux/arch/s390/kernel/smp.c (revision 26bff9eb49201aeb4e1b32d698c191831a39f5d4)
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 	lc->return_lpswe = gen_lpswe(__LC_RETURN_PSW);
216 	lc->return_mcck_lpswe = gen_lpswe(__LC_RETURN_MCCK_PSW);
217 	if (nmi_alloc_per_cpu(lc))
218 		goto out_async;
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_async:
228 	stack_free(async_stack);
229 out:
230 	if (pcpu != &pcpu_devices[0]) {
231 		free_pages(nodat_stack, THREAD_SIZE_ORDER);
232 		free_pages((unsigned long) pcpu->lowcore, LC_ORDER);
233 	}
234 	return -ENOMEM;
235 }
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 static void pcpu_prepare_secondary(struct pcpu *pcpu, int cpu)
257 {
258 	struct lowcore *lc = pcpu->lowcore;
259 
260 	cpumask_set_cpu(cpu, &init_mm.context.cpu_attach_mask);
261 	cpumask_set_cpu(cpu, mm_cpumask(&init_mm));
262 	lc->cpu_nr = cpu;
263 	lc->spinlock_lockval = arch_spin_lockval(cpu);
264 	lc->spinlock_index = 0;
265 	lc->percpu_offset = __per_cpu_offset[cpu];
266 	lc->kernel_asce = S390_lowcore.kernel_asce;
267 	lc->user_asce = S390_lowcore.kernel_asce;
268 	lc->machine_flags = S390_lowcore.machine_flags;
269 	lc->user_timer = lc->system_timer =
270 		lc->steal_timer = lc->avg_steal_timer = 0;
271 	__ctl_store(lc->cregs_save_area, 0, 15);
272 	lc->cregs_save_area[1] = lc->kernel_asce;
273 	lc->cregs_save_area[7] = lc->vdso_asce;
274 	save_access_regs((unsigned int *) lc->access_regs_save_area);
275 	memcpy(lc->stfle_fac_list, S390_lowcore.stfle_fac_list,
276 	       sizeof(lc->stfle_fac_list));
277 	memcpy(lc->alt_stfle_fac_list, S390_lowcore.alt_stfle_fac_list,
278 	       sizeof(lc->alt_stfle_fac_list));
279 	arch_spin_lock_setup(cpu);
280 }
281 
282 static void pcpu_attach_task(struct pcpu *pcpu, struct task_struct *tsk)
283 {
284 	struct lowcore *lc = pcpu->lowcore;
285 
286 	lc->kernel_stack = (unsigned long) task_stack_page(tsk)
287 		+ THREAD_SIZE - STACK_FRAME_OVERHEAD - sizeof(struct pt_regs);
288 	lc->current_task = (unsigned long) tsk;
289 	lc->lpp = LPP_MAGIC;
290 	lc->current_pid = tsk->pid;
291 	lc->user_timer = tsk->thread.user_timer;
292 	lc->guest_timer = tsk->thread.guest_timer;
293 	lc->system_timer = tsk->thread.system_timer;
294 	lc->hardirq_timer = tsk->thread.hardirq_timer;
295 	lc->softirq_timer = tsk->thread.softirq_timer;
296 	lc->steal_timer = 0;
297 }
298 
299 static void pcpu_start_fn(struct pcpu *pcpu, void (*func)(void *), void *data)
300 {
301 	struct lowcore *lc = pcpu->lowcore;
302 
303 	lc->restart_stack = lc->nodat_stack;
304 	lc->restart_fn = (unsigned long) func;
305 	lc->restart_data = (unsigned long) data;
306 	lc->restart_source = -1UL;
307 	pcpu_sigp_retry(pcpu, SIGP_RESTART, 0);
308 }
309 
310 /*
311  * Call function via PSW restart on pcpu and stop the current cpu.
312  */
313 static void __pcpu_delegate(void (*func)(void*), void *data)
314 {
315 	func(data);	/* should not return */
316 }
317 
318 static void __no_sanitize_address pcpu_delegate(struct pcpu *pcpu,
319 						void (*func)(void *),
320 						void *data, unsigned long stack)
321 {
322 	struct lowcore *lc = lowcore_ptr[pcpu - pcpu_devices];
323 	unsigned long source_cpu = stap();
324 
325 	__load_psw_mask(PSW_KERNEL_BITS | PSW_MASK_DAT);
326 	if (pcpu->address == source_cpu)
327 		CALL_ON_STACK(__pcpu_delegate, stack, 2, func, data);
328 	/* Stop target cpu (if func returns this stops the current cpu). */
329 	pcpu_sigp_retry(pcpu, SIGP_STOP, 0);
330 	/* Restart func on the target cpu and stop the current cpu. */
331 	mem_assign_absolute(lc->restart_stack, stack);
332 	mem_assign_absolute(lc->restart_fn, (unsigned long) func);
333 	mem_assign_absolute(lc->restart_data, (unsigned long) data);
334 	mem_assign_absolute(lc->restart_source, source_cpu);
335 	__bpon();
336 	asm volatile(
337 		"0:	sigp	0,%0,%2	# sigp restart to target cpu\n"
338 		"	brc	2,0b	# busy, try again\n"
339 		"1:	sigp	0,%1,%3	# sigp stop to current cpu\n"
340 		"	brc	2,1b	# busy, try again\n"
341 		: : "d" (pcpu->address), "d" (source_cpu),
342 		    "K" (SIGP_RESTART), "K" (SIGP_STOP)
343 		: "0", "1", "cc");
344 	for (;;) ;
345 }
346 
347 /*
348  * Enable additional logical cpus for multi-threading.
349  */
350 static int pcpu_set_smt(unsigned int mtid)
351 {
352 	int cc;
353 
354 	if (smp_cpu_mtid == mtid)
355 		return 0;
356 	cc = __pcpu_sigp(0, SIGP_SET_MULTI_THREADING, mtid, NULL);
357 	if (cc == 0) {
358 		smp_cpu_mtid = mtid;
359 		smp_cpu_mt_shift = 0;
360 		while (smp_cpu_mtid >= (1U << smp_cpu_mt_shift))
361 			smp_cpu_mt_shift++;
362 		pcpu_devices[0].address = stap();
363 	}
364 	return cc;
365 }
366 
367 /*
368  * Call function on an online CPU.
369  */
370 void smp_call_online_cpu(void (*func)(void *), void *data)
371 {
372 	struct pcpu *pcpu;
373 
374 	/* Use the current cpu if it is online. */
375 	pcpu = pcpu_find_address(cpu_online_mask, stap());
376 	if (!pcpu)
377 		/* Use the first online cpu. */
378 		pcpu = pcpu_devices + cpumask_first(cpu_online_mask);
379 	pcpu_delegate(pcpu, func, data, (unsigned long) restart_stack);
380 }
381 
382 /*
383  * Call function on the ipl CPU.
384  */
385 void smp_call_ipl_cpu(void (*func)(void *), void *data)
386 {
387 	struct lowcore *lc = pcpu_devices->lowcore;
388 
389 	if (pcpu_devices[0].address == stap())
390 		lc = &S390_lowcore;
391 
392 	pcpu_delegate(&pcpu_devices[0], func, data,
393 		      lc->nodat_stack);
394 }
395 
396 int smp_find_processor_id(u16 address)
397 {
398 	int cpu;
399 
400 	for_each_present_cpu(cpu)
401 		if (pcpu_devices[cpu].address == address)
402 			return cpu;
403 	return -1;
404 }
405 
406 bool arch_vcpu_is_preempted(int cpu)
407 {
408 	if (test_cpu_flag_of(CIF_ENABLED_WAIT, cpu))
409 		return false;
410 	if (pcpu_running(pcpu_devices + cpu))
411 		return false;
412 	return true;
413 }
414 EXPORT_SYMBOL(arch_vcpu_is_preempted);
415 
416 void smp_yield_cpu(int cpu)
417 {
418 	if (!MACHINE_HAS_DIAG9C)
419 		return;
420 	diag_stat_inc_norecursion(DIAG_STAT_X09C);
421 	asm volatile("diag %0,0,0x9c"
422 		     : : "d" (pcpu_devices[cpu].address));
423 }
424 
425 /*
426  * Send cpus emergency shutdown signal. This gives the cpus the
427  * opportunity to complete outstanding interrupts.
428  */
429 void notrace smp_emergency_stop(void)
430 {
431 	cpumask_t cpumask;
432 	u64 end;
433 	int cpu;
434 
435 	cpumask_copy(&cpumask, cpu_online_mask);
436 	cpumask_clear_cpu(smp_processor_id(), &cpumask);
437 
438 	end = get_tod_clock() + (1000000UL << 12);
439 	for_each_cpu(cpu, &cpumask) {
440 		struct pcpu *pcpu = pcpu_devices + cpu;
441 		set_bit(ec_stop_cpu, &pcpu->ec_mask);
442 		while (__pcpu_sigp(pcpu->address, SIGP_EMERGENCY_SIGNAL,
443 				   0, NULL) == SIGP_CC_BUSY &&
444 		       get_tod_clock() < end)
445 			cpu_relax();
446 	}
447 	while (get_tod_clock() < end) {
448 		for_each_cpu(cpu, &cpumask)
449 			if (pcpu_stopped(pcpu_devices + cpu))
450 				cpumask_clear_cpu(cpu, &cpumask);
451 		if (cpumask_empty(&cpumask))
452 			break;
453 		cpu_relax();
454 	}
455 }
456 NOKPROBE_SYMBOL(smp_emergency_stop);
457 
458 /*
459  * Stop all cpus but the current one.
460  */
461 void smp_send_stop(void)
462 {
463 	int cpu;
464 
465 	/* Disable all interrupts/machine checks */
466 	__load_psw_mask(PSW_KERNEL_BITS | PSW_MASK_DAT);
467 	trace_hardirqs_off();
468 
469 	debug_set_critical();
470 
471 	if (oops_in_progress)
472 		smp_emergency_stop();
473 
474 	/* stop all processors */
475 	for_each_online_cpu(cpu) {
476 		if (cpu == smp_processor_id())
477 			continue;
478 		pcpu_sigp_retry(pcpu_devices + cpu, SIGP_STOP, 0);
479 		while (!pcpu_stopped(pcpu_devices + cpu))
480 			cpu_relax();
481 	}
482 }
483 
484 /*
485  * This is the main routine where commands issued by other
486  * cpus are handled.
487  */
488 static void smp_handle_ext_call(void)
489 {
490 	unsigned long bits;
491 
492 	/* handle bit signal external calls */
493 	bits = xchg(&pcpu_devices[smp_processor_id()].ec_mask, 0);
494 	if (test_bit(ec_stop_cpu, &bits))
495 		smp_stop_cpu();
496 	if (test_bit(ec_schedule, &bits))
497 		scheduler_ipi();
498 	if (test_bit(ec_call_function_single, &bits))
499 		generic_smp_call_function_single_interrupt();
500 }
501 
502 static void do_ext_call_interrupt(struct ext_code ext_code,
503 				  unsigned int param32, unsigned long param64)
504 {
505 	inc_irq_stat(ext_code.code == 0x1202 ? IRQEXT_EXC : IRQEXT_EMS);
506 	smp_handle_ext_call();
507 }
508 
509 void arch_send_call_function_ipi_mask(const struct cpumask *mask)
510 {
511 	int cpu;
512 
513 	for_each_cpu(cpu, mask)
514 		pcpu_ec_call(pcpu_devices + cpu, ec_call_function_single);
515 }
516 
517 void arch_send_call_function_single_ipi(int cpu)
518 {
519 	pcpu_ec_call(pcpu_devices + cpu, ec_call_function_single);
520 }
521 
522 /*
523  * this function sends a 'reschedule' IPI to another CPU.
524  * it goes straight through and wastes no time serializing
525  * anything. Worst case is that we lose a reschedule ...
526  */
527 void smp_send_reschedule(int cpu)
528 {
529 	pcpu_ec_call(pcpu_devices + cpu, ec_schedule);
530 }
531 
532 /*
533  * parameter area for the set/clear control bit callbacks
534  */
535 struct ec_creg_mask_parms {
536 	unsigned long orval;
537 	unsigned long andval;
538 	int cr;
539 };
540 
541 /*
542  * callback for setting/clearing control bits
543  */
544 static void smp_ctl_bit_callback(void *info)
545 {
546 	struct ec_creg_mask_parms *pp = info;
547 	unsigned long cregs[16];
548 
549 	__ctl_store(cregs, 0, 15);
550 	cregs[pp->cr] = (cregs[pp->cr] & pp->andval) | pp->orval;
551 	__ctl_load(cregs, 0, 15);
552 }
553 
554 /*
555  * Set a bit in a control register of all cpus
556  */
557 void smp_ctl_set_bit(int cr, int bit)
558 {
559 	struct ec_creg_mask_parms parms = { 1UL << bit, -1UL, cr };
560 
561 	on_each_cpu(smp_ctl_bit_callback, &parms, 1);
562 }
563 EXPORT_SYMBOL(smp_ctl_set_bit);
564 
565 /*
566  * Clear a bit in a control register of all cpus
567  */
568 void smp_ctl_clear_bit(int cr, int bit)
569 {
570 	struct ec_creg_mask_parms parms = { 0, ~(1UL << bit), cr };
571 
572 	on_each_cpu(smp_ctl_bit_callback, &parms, 1);
573 }
574 EXPORT_SYMBOL(smp_ctl_clear_bit);
575 
576 #ifdef CONFIG_CRASH_DUMP
577 
578 int smp_store_status(int cpu)
579 {
580 	struct pcpu *pcpu = pcpu_devices + cpu;
581 	unsigned long pa;
582 
583 	pa = __pa(&pcpu->lowcore->floating_pt_save_area);
584 	if (__pcpu_sigp_relax(pcpu->address, SIGP_STORE_STATUS_AT_ADDRESS,
585 			      pa) != SIGP_CC_ORDER_CODE_ACCEPTED)
586 		return -EIO;
587 	if (!MACHINE_HAS_VX && !MACHINE_HAS_GS)
588 		return 0;
589 	pa = __pa(pcpu->lowcore->mcesad & MCESA_ORIGIN_MASK);
590 	if (MACHINE_HAS_GS)
591 		pa |= pcpu->lowcore->mcesad & MCESA_LC_MASK;
592 	if (__pcpu_sigp_relax(pcpu->address, SIGP_STORE_ADDITIONAL_STATUS,
593 			      pa) != SIGP_CC_ORDER_CODE_ACCEPTED)
594 		return -EIO;
595 	return 0;
596 }
597 
598 /*
599  * Collect CPU state of the previous, crashed system.
600  * There are four cases:
601  * 1) standard zfcp dump
602  *    condition: OLDMEM_BASE == NULL && ipl_info.type == IPL_TYPE_FCP_DUMP
603  *    The state for all CPUs except the boot CPU needs to be collected
604  *    with sigp stop-and-store-status. The boot CPU state is located in
605  *    the absolute lowcore of the memory stored in the HSA. The zcore code
606  *    will copy the boot CPU state from the HSA.
607  * 2) stand-alone kdump for SCSI (zfcp dump with swapped memory)
608  *    condition: OLDMEM_BASE != NULL && ipl_info.type == IPL_TYPE_FCP_DUMP
609  *    The state for all CPUs except the boot CPU needs to be collected
610  *    with sigp stop-and-store-status. The firmware or the boot-loader
611  *    stored the registers of the boot CPU in the absolute lowcore in the
612  *    memory of the old system.
613  * 3) kdump and the old kernel did not store the CPU state,
614  *    or stand-alone kdump for DASD
615  *    condition: OLDMEM_BASE != NULL && !is_kdump_kernel()
616  *    The state for all CPUs except the boot CPU needs to be collected
617  *    with sigp stop-and-store-status. The kexec code or the boot-loader
618  *    stored the registers of the boot CPU in the memory of the old system.
619  * 4) kdump and the old kernel stored the CPU state
620  *    condition: OLDMEM_BASE != NULL && is_kdump_kernel()
621  *    This case does not exist for s390 anymore, setup_arch explicitly
622  *    deactivates the elfcorehdr= kernel parameter
623  */
624 static __init void smp_save_cpu_vxrs(struct save_area *sa, u16 addr,
625 				     bool is_boot_cpu, unsigned long page)
626 {
627 	__vector128 *vxrs = (__vector128 *) page;
628 
629 	if (is_boot_cpu)
630 		vxrs = boot_cpu_vector_save_area;
631 	else
632 		__pcpu_sigp_relax(addr, SIGP_STORE_ADDITIONAL_STATUS, page);
633 	save_area_add_vxrs(sa, vxrs);
634 }
635 
636 static __init void smp_save_cpu_regs(struct save_area *sa, u16 addr,
637 				     bool is_boot_cpu, unsigned long page)
638 {
639 	void *regs = (void *) page;
640 
641 	if (is_boot_cpu)
642 		copy_oldmem_kernel(regs, (void *) __LC_FPREGS_SAVE_AREA, 512);
643 	else
644 		__pcpu_sigp_relax(addr, SIGP_STORE_STATUS_AT_ADDRESS, page);
645 	save_area_add_regs(sa, regs);
646 }
647 
648 void __init smp_save_dump_cpus(void)
649 {
650 	int addr, boot_cpu_addr, max_cpu_addr;
651 	struct save_area *sa;
652 	unsigned long page;
653 	bool is_boot_cpu;
654 
655 	if (!(OLDMEM_BASE || ipl_info.type == IPL_TYPE_FCP_DUMP))
656 		/* No previous system present, normal boot. */
657 		return;
658 	/* Allocate a page as dumping area for the store status sigps */
659 	page = memblock_phys_alloc_range(PAGE_SIZE, PAGE_SIZE, 0, 1UL << 31);
660 	if (!page)
661 		panic("ERROR: Failed to allocate %lx bytes below %lx\n",
662 		      PAGE_SIZE, 1UL << 31);
663 
664 	/* Set multi-threading state to the previous system. */
665 	pcpu_set_smt(sclp.mtid_prev);
666 	boot_cpu_addr = stap();
667 	max_cpu_addr = SCLP_MAX_CORES << sclp.mtid_prev;
668 	for (addr = 0; addr <= max_cpu_addr; addr++) {
669 		if (__pcpu_sigp_relax(addr, SIGP_SENSE, 0) ==
670 		    SIGP_CC_NOT_OPERATIONAL)
671 			continue;
672 		is_boot_cpu = (addr == boot_cpu_addr);
673 		/* Allocate save area */
674 		sa = save_area_alloc(is_boot_cpu);
675 		if (!sa)
676 			panic("could not allocate memory for save area\n");
677 		if (MACHINE_HAS_VX)
678 			/* Get the vector registers */
679 			smp_save_cpu_vxrs(sa, addr, is_boot_cpu, page);
680 		/*
681 		 * For a zfcp dump OLDMEM_BASE == NULL and the registers
682 		 * of the boot CPU are stored in the HSA. To retrieve
683 		 * these registers an SCLP request is required which is
684 		 * done by drivers/s390/char/zcore.c:init_cpu_info()
685 		 */
686 		if (!is_boot_cpu || OLDMEM_BASE)
687 			/* Get the CPU registers */
688 			smp_save_cpu_regs(sa, addr, is_boot_cpu, page);
689 	}
690 	memblock_free(page, PAGE_SIZE);
691 	diag_dma_ops.diag308_reset();
692 	pcpu_set_smt(0);
693 }
694 #endif /* CONFIG_CRASH_DUMP */
695 
696 void smp_cpu_set_polarization(int cpu, int val)
697 {
698 	pcpu_devices[cpu].polarization = val;
699 }
700 
701 int smp_cpu_get_polarization(int cpu)
702 {
703 	return pcpu_devices[cpu].polarization;
704 }
705 
706 int smp_cpu_get_cpu_address(int cpu)
707 {
708 	return pcpu_devices[cpu].address;
709 }
710 
711 static void __ref smp_get_core_info(struct sclp_core_info *info, int early)
712 {
713 	static int use_sigp_detection;
714 	int address;
715 
716 	if (use_sigp_detection || sclp_get_core_info(info, early)) {
717 		use_sigp_detection = 1;
718 		for (address = 0;
719 		     address < (SCLP_MAX_CORES << smp_cpu_mt_shift);
720 		     address += (1U << smp_cpu_mt_shift)) {
721 			if (__pcpu_sigp_relax(address, SIGP_SENSE, 0) ==
722 			    SIGP_CC_NOT_OPERATIONAL)
723 				continue;
724 			info->core[info->configured].core_id =
725 				address >> smp_cpu_mt_shift;
726 			info->configured++;
727 		}
728 		info->combined = info->configured;
729 	}
730 }
731 
732 static int smp_add_present_cpu(int cpu);
733 
734 static int smp_add_core(struct sclp_core_entry *core, cpumask_t *avail,
735 			bool configured, bool early)
736 {
737 	struct pcpu *pcpu;
738 	int cpu, nr, i;
739 	u16 address;
740 
741 	nr = 0;
742 	if (sclp.has_core_type && core->type != boot_core_type)
743 		return nr;
744 	cpu = cpumask_first(avail);
745 	address = core->core_id << smp_cpu_mt_shift;
746 	for (i = 0; (i <= smp_cpu_mtid) && (cpu < nr_cpu_ids); i++) {
747 		if (pcpu_find_address(cpu_present_mask, address + i))
748 			continue;
749 		pcpu = pcpu_devices + cpu;
750 		pcpu->address = address + i;
751 		if (configured)
752 			pcpu->state = CPU_STATE_CONFIGURED;
753 		else
754 			pcpu->state = CPU_STATE_STANDBY;
755 		smp_cpu_set_polarization(cpu, POLARIZATION_UNKNOWN);
756 		set_cpu_present(cpu, true);
757 		if (!early && smp_add_present_cpu(cpu) != 0)
758 			set_cpu_present(cpu, false);
759 		else
760 			nr++;
761 		cpumask_clear_cpu(cpu, avail);
762 		cpu = cpumask_next(cpu, avail);
763 	}
764 	return nr;
765 }
766 
767 static int __smp_rescan_cpus(struct sclp_core_info *info, bool early)
768 {
769 	struct sclp_core_entry *core;
770 	cpumask_t avail;
771 	bool configured;
772 	u16 core_id;
773 	int nr, i;
774 
775 	nr = 0;
776 	cpumask_xor(&avail, cpu_possible_mask, cpu_present_mask);
777 	/*
778 	 * Add IPL core first (which got logical CPU number 0) to make sure
779 	 * that all SMT threads get subsequent logical CPU numbers.
780 	 */
781 	if (early) {
782 		core_id = pcpu_devices[0].address >> smp_cpu_mt_shift;
783 		for (i = 0; i < info->configured; i++) {
784 			core = &info->core[i];
785 			if (core->core_id == core_id) {
786 				nr += smp_add_core(core, &avail, true, early);
787 				break;
788 			}
789 		}
790 	}
791 	for (i = 0; i < info->combined; i++) {
792 		configured = i < info->configured;
793 		nr += smp_add_core(&info->core[i], &avail, configured, early);
794 	}
795 	return nr;
796 }
797 
798 void __init smp_detect_cpus(void)
799 {
800 	unsigned int cpu, mtid, c_cpus, s_cpus;
801 	struct sclp_core_info *info;
802 	u16 address;
803 
804 	/* Get CPU information */
805 	info = memblock_alloc(sizeof(*info), 8);
806 	if (!info)
807 		panic("%s: Failed to allocate %zu bytes align=0x%x\n",
808 		      __func__, sizeof(*info), 8);
809 	smp_get_core_info(info, 1);
810 	/* Find boot CPU type */
811 	if (sclp.has_core_type) {
812 		address = stap();
813 		for (cpu = 0; cpu < info->combined; cpu++)
814 			if (info->core[cpu].core_id == address) {
815 				/* The boot cpu dictates the cpu type. */
816 				boot_core_type = info->core[cpu].type;
817 				break;
818 			}
819 		if (cpu >= info->combined)
820 			panic("Could not find boot CPU type");
821 	}
822 
823 	/* Set multi-threading state for the current system */
824 	mtid = boot_core_type ? sclp.mtid : sclp.mtid_cp;
825 	mtid = (mtid < smp_max_threads) ? mtid : smp_max_threads - 1;
826 	pcpu_set_smt(mtid);
827 
828 	/* Print number of CPUs */
829 	c_cpus = s_cpus = 0;
830 	for (cpu = 0; cpu < info->combined; cpu++) {
831 		if (sclp.has_core_type &&
832 		    info->core[cpu].type != boot_core_type)
833 			continue;
834 		if (cpu < info->configured)
835 			c_cpus += smp_cpu_mtid + 1;
836 		else
837 			s_cpus += smp_cpu_mtid + 1;
838 	}
839 	pr_info("%d configured CPUs, %d standby CPUs\n", c_cpus, s_cpus);
840 
841 	/* Add CPUs present at boot */
842 	get_online_cpus();
843 	__smp_rescan_cpus(info, true);
844 	put_online_cpus();
845 	memblock_free_early((unsigned long)info, sizeof(*info));
846 }
847 
848 static void smp_init_secondary(void)
849 {
850 	int cpu = smp_processor_id();
851 
852 	S390_lowcore.last_update_clock = get_tod_clock();
853 	restore_access_regs(S390_lowcore.access_regs_save_area);
854 	set_cpu_flag(CIF_ASCE_PRIMARY);
855 	set_cpu_flag(CIF_ASCE_SECONDARY);
856 	cpu_init();
857 	preempt_disable();
858 	init_cpu_timer();
859 	vtime_init();
860 	pfault_init();
861 	notify_cpu_starting(cpu);
862 	if (topology_cpu_dedicated(cpu))
863 		set_cpu_flag(CIF_DEDICATED_CPU);
864 	else
865 		clear_cpu_flag(CIF_DEDICATED_CPU);
866 	set_cpu_online(cpu, true);
867 	update_cpu_masks();
868 	inc_irq_stat(CPU_RST);
869 	local_irq_enable();
870 	cpu_startup_entry(CPUHP_AP_ONLINE_IDLE);
871 }
872 
873 /*
874  *	Activate a secondary processor.
875  */
876 static void __no_sanitize_address smp_start_secondary(void *cpuvoid)
877 {
878 	S390_lowcore.restart_stack = (unsigned long) restart_stack;
879 	S390_lowcore.restart_fn = (unsigned long) do_restart;
880 	S390_lowcore.restart_data = 0;
881 	S390_lowcore.restart_source = -1UL;
882 	__ctl_load(S390_lowcore.cregs_save_area, 0, 15);
883 	__load_psw_mask(PSW_KERNEL_BITS | PSW_MASK_DAT);
884 	CALL_ON_STACK_NORETURN(smp_init_secondary, S390_lowcore.kernel_stack);
885 }
886 
887 /* Upping and downing of CPUs */
888 int __cpu_up(unsigned int cpu, struct task_struct *tidle)
889 {
890 	struct pcpu *pcpu;
891 	int base, i, rc;
892 
893 	pcpu = pcpu_devices + cpu;
894 	if (pcpu->state != CPU_STATE_CONFIGURED)
895 		return -EIO;
896 	base = smp_get_base_cpu(cpu);
897 	for (i = 0; i <= smp_cpu_mtid; i++) {
898 		if (base + i < nr_cpu_ids)
899 			if (cpu_online(base + i))
900 				break;
901 	}
902 	/*
903 	 * If this is the first CPU of the core to get online
904 	 * do an initial CPU reset.
905 	 */
906 	if (i > smp_cpu_mtid &&
907 	    pcpu_sigp_retry(pcpu_devices + base, SIGP_INITIAL_CPU_RESET, 0) !=
908 	    SIGP_CC_ORDER_CODE_ACCEPTED)
909 		return -EIO;
910 
911 	rc = pcpu_alloc_lowcore(pcpu, cpu);
912 	if (rc)
913 		return rc;
914 	pcpu_prepare_secondary(pcpu, cpu);
915 	pcpu_attach_task(pcpu, tidle);
916 	pcpu_start_fn(pcpu, smp_start_secondary, NULL);
917 	/* Wait until cpu puts itself in the online & active maps */
918 	while (!cpu_online(cpu))
919 		cpu_relax();
920 	return 0;
921 }
922 
923 static unsigned int setup_possible_cpus __initdata;
924 
925 static int __init _setup_possible_cpus(char *s)
926 {
927 	get_option(&s, &setup_possible_cpus);
928 	return 0;
929 }
930 early_param("possible_cpus", _setup_possible_cpus);
931 
932 int __cpu_disable(void)
933 {
934 	unsigned long cregs[16];
935 
936 	/* Handle possible pending IPIs */
937 	smp_handle_ext_call();
938 	set_cpu_online(smp_processor_id(), false);
939 	update_cpu_masks();
940 	/* Disable pseudo page faults on this cpu. */
941 	pfault_fini();
942 	/* Disable interrupt sources via control register. */
943 	__ctl_store(cregs, 0, 15);
944 	cregs[0]  &= ~0x0000ee70UL;	/* disable all external interrupts */
945 	cregs[6]  &= ~0xff000000UL;	/* disable all I/O interrupts */
946 	cregs[14] &= ~0x1f000000UL;	/* disable most machine checks */
947 	__ctl_load(cregs, 0, 15);
948 	clear_cpu_flag(CIF_NOHZ_DELAY);
949 	return 0;
950 }
951 
952 void __cpu_die(unsigned int cpu)
953 {
954 	struct pcpu *pcpu;
955 
956 	/* Wait until target cpu is down */
957 	pcpu = pcpu_devices + cpu;
958 	while (!pcpu_stopped(pcpu))
959 		cpu_relax();
960 	pcpu_free_lowcore(pcpu);
961 	cpumask_clear_cpu(cpu, mm_cpumask(&init_mm));
962 	cpumask_clear_cpu(cpu, &init_mm.context.cpu_attach_mask);
963 }
964 
965 void __noreturn cpu_die(void)
966 {
967 	idle_task_exit();
968 	__bpon();
969 	pcpu_sigp_retry(pcpu_devices + smp_processor_id(), SIGP_STOP, 0);
970 	for (;;) ;
971 }
972 
973 void __init smp_fill_possible_mask(void)
974 {
975 	unsigned int possible, sclp_max, cpu;
976 
977 	sclp_max = max(sclp.mtid, sclp.mtid_cp) + 1;
978 	sclp_max = min(smp_max_threads, sclp_max);
979 	sclp_max = (sclp.max_cores * sclp_max) ?: nr_cpu_ids;
980 	possible = setup_possible_cpus ?: nr_cpu_ids;
981 	possible = min(possible, sclp_max);
982 	for (cpu = 0; cpu < possible && cpu < nr_cpu_ids; cpu++)
983 		set_cpu_possible(cpu, true);
984 }
985 
986 void __init smp_prepare_cpus(unsigned int max_cpus)
987 {
988 	/* request the 0x1201 emergency signal external interrupt */
989 	if (register_external_irq(EXT_IRQ_EMERGENCY_SIG, do_ext_call_interrupt))
990 		panic("Couldn't request external interrupt 0x1201");
991 	/* request the 0x1202 external call external interrupt */
992 	if (register_external_irq(EXT_IRQ_EXTERNAL_CALL, do_ext_call_interrupt))
993 		panic("Couldn't request external interrupt 0x1202");
994 }
995 
996 void __init smp_prepare_boot_cpu(void)
997 {
998 	struct pcpu *pcpu = pcpu_devices;
999 
1000 	WARN_ON(!cpu_present(0) || !cpu_online(0));
1001 	pcpu->state = CPU_STATE_CONFIGURED;
1002 	pcpu->lowcore = (struct lowcore *)(unsigned long) store_prefix();
1003 	S390_lowcore.percpu_offset = __per_cpu_offset[0];
1004 	smp_cpu_set_polarization(0, POLARIZATION_UNKNOWN);
1005 }
1006 
1007 void __init smp_cpus_done(unsigned int max_cpus)
1008 {
1009 }
1010 
1011 void __init smp_setup_processor_id(void)
1012 {
1013 	pcpu_devices[0].address = stap();
1014 	S390_lowcore.cpu_nr = 0;
1015 	S390_lowcore.spinlock_lockval = arch_spin_lockval(0);
1016 	S390_lowcore.spinlock_index = 0;
1017 }
1018 
1019 /*
1020  * the frequency of the profiling timer can be changed
1021  * by writing a multiplier value into /proc/profile.
1022  *
1023  * usually you want to run this on all CPUs ;)
1024  */
1025 int setup_profiling_timer(unsigned int multiplier)
1026 {
1027 	return 0;
1028 }
1029 
1030 static ssize_t cpu_configure_show(struct device *dev,
1031 				  struct device_attribute *attr, char *buf)
1032 {
1033 	ssize_t count;
1034 
1035 	mutex_lock(&smp_cpu_state_mutex);
1036 	count = sprintf(buf, "%d\n", pcpu_devices[dev->id].state);
1037 	mutex_unlock(&smp_cpu_state_mutex);
1038 	return count;
1039 }
1040 
1041 static ssize_t cpu_configure_store(struct device *dev,
1042 				   struct device_attribute *attr,
1043 				   const char *buf, size_t count)
1044 {
1045 	struct pcpu *pcpu;
1046 	int cpu, val, rc, i;
1047 	char delim;
1048 
1049 	if (sscanf(buf, "%d %c", &val, &delim) != 1)
1050 		return -EINVAL;
1051 	if (val != 0 && val != 1)
1052 		return -EINVAL;
1053 	get_online_cpus();
1054 	mutex_lock(&smp_cpu_state_mutex);
1055 	rc = -EBUSY;
1056 	/* disallow configuration changes of online cpus and cpu 0 */
1057 	cpu = dev->id;
1058 	cpu = smp_get_base_cpu(cpu);
1059 	if (cpu == 0)
1060 		goto out;
1061 	for (i = 0; i <= smp_cpu_mtid; i++)
1062 		if (cpu_online(cpu + i))
1063 			goto out;
1064 	pcpu = pcpu_devices + cpu;
1065 	rc = 0;
1066 	switch (val) {
1067 	case 0:
1068 		if (pcpu->state != CPU_STATE_CONFIGURED)
1069 			break;
1070 		rc = sclp_core_deconfigure(pcpu->address >> smp_cpu_mt_shift);
1071 		if (rc)
1072 			break;
1073 		for (i = 0; i <= smp_cpu_mtid; i++) {
1074 			if (cpu + i >= nr_cpu_ids || !cpu_present(cpu + i))
1075 				continue;
1076 			pcpu[i].state = CPU_STATE_STANDBY;
1077 			smp_cpu_set_polarization(cpu + i,
1078 						 POLARIZATION_UNKNOWN);
1079 		}
1080 		topology_expect_change();
1081 		break;
1082 	case 1:
1083 		if (pcpu->state != CPU_STATE_STANDBY)
1084 			break;
1085 		rc = sclp_core_configure(pcpu->address >> smp_cpu_mt_shift);
1086 		if (rc)
1087 			break;
1088 		for (i = 0; i <= smp_cpu_mtid; i++) {
1089 			if (cpu + i >= nr_cpu_ids || !cpu_present(cpu + i))
1090 				continue;
1091 			pcpu[i].state = CPU_STATE_CONFIGURED;
1092 			smp_cpu_set_polarization(cpu + i,
1093 						 POLARIZATION_UNKNOWN);
1094 		}
1095 		topology_expect_change();
1096 		break;
1097 	default:
1098 		break;
1099 	}
1100 out:
1101 	mutex_unlock(&smp_cpu_state_mutex);
1102 	put_online_cpus();
1103 	return rc ? rc : count;
1104 }
1105 static DEVICE_ATTR(configure, 0644, cpu_configure_show, cpu_configure_store);
1106 
1107 static ssize_t show_cpu_address(struct device *dev,
1108 				struct device_attribute *attr, char *buf)
1109 {
1110 	return sprintf(buf, "%d\n", pcpu_devices[dev->id].address);
1111 }
1112 static DEVICE_ATTR(address, 0444, show_cpu_address, NULL);
1113 
1114 static struct attribute *cpu_common_attrs[] = {
1115 	&dev_attr_configure.attr,
1116 	&dev_attr_address.attr,
1117 	NULL,
1118 };
1119 
1120 static struct attribute_group cpu_common_attr_group = {
1121 	.attrs = cpu_common_attrs,
1122 };
1123 
1124 static struct attribute *cpu_online_attrs[] = {
1125 	&dev_attr_idle_count.attr,
1126 	&dev_attr_idle_time_us.attr,
1127 	NULL,
1128 };
1129 
1130 static struct attribute_group cpu_online_attr_group = {
1131 	.attrs = cpu_online_attrs,
1132 };
1133 
1134 static int smp_cpu_online(unsigned int cpu)
1135 {
1136 	struct device *s = &per_cpu(cpu_device, cpu)->dev;
1137 
1138 	return sysfs_create_group(&s->kobj, &cpu_online_attr_group);
1139 }
1140 static int smp_cpu_pre_down(unsigned int cpu)
1141 {
1142 	struct device *s = &per_cpu(cpu_device, cpu)->dev;
1143 
1144 	sysfs_remove_group(&s->kobj, &cpu_online_attr_group);
1145 	return 0;
1146 }
1147 
1148 static int smp_add_present_cpu(int cpu)
1149 {
1150 	struct device *s;
1151 	struct cpu *c;
1152 	int rc;
1153 
1154 	c = kzalloc(sizeof(*c), GFP_KERNEL);
1155 	if (!c)
1156 		return -ENOMEM;
1157 	per_cpu(cpu_device, cpu) = c;
1158 	s = &c->dev;
1159 	c->hotpluggable = 1;
1160 	rc = register_cpu(c, cpu);
1161 	if (rc)
1162 		goto out;
1163 	rc = sysfs_create_group(&s->kobj, &cpu_common_attr_group);
1164 	if (rc)
1165 		goto out_cpu;
1166 	rc = topology_cpu_init(c);
1167 	if (rc)
1168 		goto out_topology;
1169 	return 0;
1170 
1171 out_topology:
1172 	sysfs_remove_group(&s->kobj, &cpu_common_attr_group);
1173 out_cpu:
1174 	unregister_cpu(c);
1175 out:
1176 	return rc;
1177 }
1178 
1179 int __ref smp_rescan_cpus(void)
1180 {
1181 	struct sclp_core_info *info;
1182 	int nr;
1183 
1184 	info = kzalloc(sizeof(*info), GFP_KERNEL);
1185 	if (!info)
1186 		return -ENOMEM;
1187 	smp_get_core_info(info, 0);
1188 	get_online_cpus();
1189 	mutex_lock(&smp_cpu_state_mutex);
1190 	nr = __smp_rescan_cpus(info, false);
1191 	mutex_unlock(&smp_cpu_state_mutex);
1192 	put_online_cpus();
1193 	kfree(info);
1194 	if (nr)
1195 		topology_schedule_update();
1196 	return 0;
1197 }
1198 
1199 static ssize_t __ref rescan_store(struct device *dev,
1200 				  struct device_attribute *attr,
1201 				  const char *buf,
1202 				  size_t count)
1203 {
1204 	int rc;
1205 
1206 	rc = lock_device_hotplug_sysfs();
1207 	if (rc)
1208 		return rc;
1209 	rc = smp_rescan_cpus();
1210 	unlock_device_hotplug();
1211 	return rc ? rc : count;
1212 }
1213 static DEVICE_ATTR_WO(rescan);
1214 
1215 static int __init s390_smp_init(void)
1216 {
1217 	int cpu, rc = 0;
1218 
1219 	rc = device_create_file(cpu_subsys.dev_root, &dev_attr_rescan);
1220 	if (rc)
1221 		return rc;
1222 	for_each_present_cpu(cpu) {
1223 		rc = smp_add_present_cpu(cpu);
1224 		if (rc)
1225 			goto out;
1226 	}
1227 
1228 	rc = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "s390/smp:online",
1229 			       smp_cpu_online, smp_cpu_pre_down);
1230 	rc = rc <= 0 ? rc : 0;
1231 out:
1232 	return rc;
1233 }
1234 subsys_initcall(s390_smp_init);
1235