xref: /openbmc/linux/arch/s390/kernel/smp.c (revision 2208f39c)
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 	ec_mcck_pending,
65 };
66 
67 enum {
68 	CPU_STATE_STANDBY,
69 	CPU_STATE_CONFIGURED,
70 };
71 
72 static DEFINE_PER_CPU(struct cpu *, cpu_device);
73 
74 struct pcpu {
75 	struct lowcore *lowcore;	/* lowcore page(s) for the cpu */
76 	unsigned long ec_mask;		/* bit mask for ec_xxx functions */
77 	unsigned long ec_clk;		/* sigp timestamp for ec_xxx */
78 	signed char state;		/* physical cpu state */
79 	signed char polarization;	/* physical polarization */
80 	u16 address;			/* physical cpu address */
81 };
82 
83 static u8 boot_core_type;
84 static struct pcpu pcpu_devices[NR_CPUS];
85 
86 unsigned int smp_cpu_mt_shift;
87 EXPORT_SYMBOL(smp_cpu_mt_shift);
88 
89 unsigned int smp_cpu_mtid;
90 EXPORT_SYMBOL(smp_cpu_mtid);
91 
92 #ifdef CONFIG_CRASH_DUMP
93 __vector128 __initdata boot_cpu_vector_save_area[__NUM_VXRS];
94 #endif
95 
96 static unsigned int smp_max_threads __initdata = -1U;
97 
98 static int __init early_nosmt(char *s)
99 {
100 	smp_max_threads = 1;
101 	return 0;
102 }
103 early_param("nosmt", early_nosmt);
104 
105 static int __init early_smt(char *s)
106 {
107 	get_option(&s, &smp_max_threads);
108 	return 0;
109 }
110 early_param("smt", early_smt);
111 
112 /*
113  * The smp_cpu_state_mutex must be held when changing the state or polarization
114  * member of a pcpu data structure within the pcpu_devices arreay.
115  */
116 DEFINE_MUTEX(smp_cpu_state_mutex);
117 
118 /*
119  * Signal processor helper functions.
120  */
121 static inline int __pcpu_sigp_relax(u16 addr, u8 order, unsigned long parm)
122 {
123 	int cc;
124 
125 	while (1) {
126 		cc = __pcpu_sigp(addr, order, parm, NULL);
127 		if (cc != SIGP_CC_BUSY)
128 			return cc;
129 		cpu_relax();
130 	}
131 }
132 
133 static int pcpu_sigp_retry(struct pcpu *pcpu, u8 order, u32 parm)
134 {
135 	int cc, retry;
136 
137 	for (retry = 0; ; retry++) {
138 		cc = __pcpu_sigp(pcpu->address, order, parm, NULL);
139 		if (cc != SIGP_CC_BUSY)
140 			break;
141 		if (retry >= 3)
142 			udelay(10);
143 	}
144 	return cc;
145 }
146 
147 static inline int pcpu_stopped(struct pcpu *pcpu)
148 {
149 	u32 status;
150 
151 	if (__pcpu_sigp(pcpu->address, SIGP_SENSE,
152 			0, &status) != SIGP_CC_STATUS_STORED)
153 		return 0;
154 	return !!(status & (SIGP_STATUS_CHECK_STOP|SIGP_STATUS_STOPPED));
155 }
156 
157 static inline int pcpu_running(struct pcpu *pcpu)
158 {
159 	if (__pcpu_sigp(pcpu->address, SIGP_SENSE_RUNNING,
160 			0, NULL) != SIGP_CC_STATUS_STORED)
161 		return 1;
162 	/* Status stored condition code is equivalent to cpu not running. */
163 	return 0;
164 }
165 
166 /*
167  * Find struct pcpu by cpu address.
168  */
169 static struct pcpu *pcpu_find_address(const struct cpumask *mask, u16 address)
170 {
171 	int cpu;
172 
173 	for_each_cpu(cpu, mask)
174 		if (pcpu_devices[cpu].address == address)
175 			return pcpu_devices + cpu;
176 	return NULL;
177 }
178 
179 static void pcpu_ec_call(struct pcpu *pcpu, int ec_bit)
180 {
181 	int order;
182 
183 	if (test_and_set_bit(ec_bit, &pcpu->ec_mask))
184 		return;
185 	order = pcpu_running(pcpu) ? SIGP_EXTERNAL_CALL : SIGP_EMERGENCY_SIGNAL;
186 	pcpu->ec_clk = get_tod_clock_fast();
187 	pcpu_sigp_retry(pcpu, order, 0);
188 }
189 
190 static int pcpu_alloc_lowcore(struct pcpu *pcpu, int cpu)
191 {
192 	unsigned long async_stack, nodat_stack;
193 	struct lowcore *lc;
194 
195 	if (pcpu != &pcpu_devices[0]) {
196 		pcpu->lowcore =	(struct lowcore *)
197 			__get_free_pages(GFP_KERNEL | GFP_DMA, LC_ORDER);
198 		nodat_stack = __get_free_pages(GFP_KERNEL, THREAD_SIZE_ORDER);
199 		if (!pcpu->lowcore || !nodat_stack)
200 			goto out;
201 	} else {
202 		nodat_stack = pcpu->lowcore->nodat_stack - STACK_INIT_OFFSET;
203 	}
204 	async_stack = stack_alloc();
205 	if (!async_stack)
206 		goto out;
207 	lc = pcpu->lowcore;
208 	memcpy(lc, &S390_lowcore, 512);
209 	memset((char *) lc + 512, 0, sizeof(*lc) - 512);
210 	lc->async_stack = async_stack + STACK_INIT_OFFSET;
211 	lc->nodat_stack = nodat_stack + STACK_INIT_OFFSET;
212 	lc->cpu_nr = cpu;
213 	lc->spinlock_lockval = arch_spin_lockval(cpu);
214 	lc->spinlock_index = 0;
215 	lc->br_r1_trampoline = 0x07f1;	/* br %r1 */
216 	lc->return_lpswe = gen_lpswe(__LC_RETURN_PSW);
217 	lc->return_mcck_lpswe = gen_lpswe(__LC_RETURN_MCCK_PSW);
218 	if (nmi_alloc_per_cpu(lc))
219 		goto out_async;
220 	if (vdso_alloc_per_cpu(lc))
221 		goto out_mcesa;
222 	lowcore_ptr[cpu] = lc;
223 	pcpu_sigp_retry(pcpu, SIGP_SET_PREFIX, (u32)(unsigned long) lc);
224 	return 0;
225 
226 out_mcesa:
227 	nmi_free_per_cpu(lc);
228 out_async:
229 	stack_free(async_stack);
230 out:
231 	if (pcpu != &pcpu_devices[0]) {
232 		free_pages(nodat_stack, THREAD_SIZE_ORDER);
233 		free_pages((unsigned long) pcpu->lowcore, LC_ORDER);
234 	}
235 	return -ENOMEM;
236 }
237 
238 static void pcpu_free_lowcore(struct pcpu *pcpu)
239 {
240 	unsigned long async_stack, nodat_stack, lowcore;
241 
242 	nodat_stack = pcpu->lowcore->nodat_stack - STACK_INIT_OFFSET;
243 	async_stack = pcpu->lowcore->async_stack - STACK_INIT_OFFSET;
244 	lowcore = (unsigned long) pcpu->lowcore;
245 
246 	pcpu_sigp_retry(pcpu, SIGP_SET_PREFIX, 0);
247 	lowcore_ptr[pcpu - pcpu_devices] = NULL;
248 	vdso_free_per_cpu(pcpu->lowcore);
249 	nmi_free_per_cpu(pcpu->lowcore);
250 	stack_free(async_stack);
251 	if (pcpu == &pcpu_devices[0])
252 		return;
253 	free_pages(nodat_stack, THREAD_SIZE_ORDER);
254 	free_pages(lowcore, LC_ORDER);
255 }
256 
257 static void pcpu_prepare_secondary(struct pcpu *pcpu, int cpu)
258 {
259 	struct lowcore *lc = pcpu->lowcore;
260 
261 	cpumask_set_cpu(cpu, &init_mm.context.cpu_attach_mask);
262 	cpumask_set_cpu(cpu, mm_cpumask(&init_mm));
263 	lc->cpu_nr = cpu;
264 	lc->spinlock_lockval = arch_spin_lockval(cpu);
265 	lc->spinlock_index = 0;
266 	lc->percpu_offset = __per_cpu_offset[cpu];
267 	lc->kernel_asce = S390_lowcore.kernel_asce;
268 	lc->user_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 	lc->cregs_save_area[1] = lc->kernel_asce;
274 	lc->cregs_save_area[7] = lc->vdso_asce;
275 	save_access_regs((unsigned int *) lc->access_regs_save_area);
276 	memcpy(lc->stfle_fac_list, S390_lowcore.stfle_fac_list,
277 	       sizeof(lc->stfle_fac_list));
278 	memcpy(lc->alt_stfle_fac_list, S390_lowcore.alt_stfle_fac_list,
279 	       sizeof(lc->alt_stfle_fac_list));
280 	arch_spin_lock_setup(cpu);
281 }
282 
283 static void pcpu_attach_task(struct pcpu *pcpu, struct task_struct *tsk)
284 {
285 	struct lowcore *lc = pcpu->lowcore;
286 
287 	lc->kernel_stack = (unsigned long) task_stack_page(tsk)
288 		+ THREAD_SIZE - STACK_FRAME_OVERHEAD - sizeof(struct pt_regs);
289 	lc->current_task = (unsigned long) tsk;
290 	lc->lpp = LPP_MAGIC;
291 	lc->current_pid = tsk->pid;
292 	lc->user_timer = tsk->thread.user_timer;
293 	lc->guest_timer = tsk->thread.guest_timer;
294 	lc->system_timer = tsk->thread.system_timer;
295 	lc->hardirq_timer = tsk->thread.hardirq_timer;
296 	lc->softirq_timer = tsk->thread.softirq_timer;
297 	lc->steal_timer = 0;
298 }
299 
300 static void pcpu_start_fn(struct pcpu *pcpu, void (*func)(void *), void *data)
301 {
302 	struct lowcore *lc = pcpu->lowcore;
303 
304 	lc->restart_stack = lc->nodat_stack;
305 	lc->restart_fn = (unsigned long) func;
306 	lc->restart_data = (unsigned long) data;
307 	lc->restart_source = -1UL;
308 	pcpu_sigp_retry(pcpu, SIGP_RESTART, 0);
309 }
310 
311 /*
312  * Call function via PSW restart on pcpu and stop the current cpu.
313  */
314 static void __pcpu_delegate(void (*func)(void*), void *data)
315 {
316 	func(data);	/* should not return */
317 }
318 
319 static void __no_sanitize_address pcpu_delegate(struct pcpu *pcpu,
320 						void (*func)(void *),
321 						void *data, unsigned long stack)
322 {
323 	struct lowcore *lc = lowcore_ptr[pcpu - pcpu_devices];
324 	unsigned long source_cpu = stap();
325 
326 	__load_psw_mask(PSW_KERNEL_BITS | PSW_MASK_DAT);
327 	if (pcpu->address == source_cpu)
328 		CALL_ON_STACK(__pcpu_delegate, stack, 2, func, data);
329 	/* Stop target cpu (if func returns this stops the current cpu). */
330 	pcpu_sigp_retry(pcpu, SIGP_STOP, 0);
331 	/* Restart func on the target cpu and stop the current cpu. */
332 	mem_assign_absolute(lc->restart_stack, stack);
333 	mem_assign_absolute(lc->restart_fn, (unsigned long) func);
334 	mem_assign_absolute(lc->restart_data, (unsigned long) data);
335 	mem_assign_absolute(lc->restart_source, source_cpu);
336 	__bpon();
337 	asm volatile(
338 		"0:	sigp	0,%0,%2	# sigp restart to target cpu\n"
339 		"	brc	2,0b	# busy, try again\n"
340 		"1:	sigp	0,%1,%3	# sigp stop to current cpu\n"
341 		"	brc	2,1b	# busy, try again\n"
342 		: : "d" (pcpu->address), "d" (source_cpu),
343 		    "K" (SIGP_RESTART), "K" (SIGP_STOP)
344 		: "0", "1", "cc");
345 	for (;;) ;
346 }
347 
348 /*
349  * Enable additional logical cpus for multi-threading.
350  */
351 static int pcpu_set_smt(unsigned int mtid)
352 {
353 	int cc;
354 
355 	if (smp_cpu_mtid == mtid)
356 		return 0;
357 	cc = __pcpu_sigp(0, SIGP_SET_MULTI_THREADING, mtid, NULL);
358 	if (cc == 0) {
359 		smp_cpu_mtid = mtid;
360 		smp_cpu_mt_shift = 0;
361 		while (smp_cpu_mtid >= (1U << smp_cpu_mt_shift))
362 			smp_cpu_mt_shift++;
363 		pcpu_devices[0].address = stap();
364 	}
365 	return cc;
366 }
367 
368 /*
369  * Call function on an online CPU.
370  */
371 void smp_call_online_cpu(void (*func)(void *), void *data)
372 {
373 	struct pcpu *pcpu;
374 
375 	/* Use the current cpu if it is online. */
376 	pcpu = pcpu_find_address(cpu_online_mask, stap());
377 	if (!pcpu)
378 		/* Use the first online cpu. */
379 		pcpu = pcpu_devices + cpumask_first(cpu_online_mask);
380 	pcpu_delegate(pcpu, func, data, (unsigned long) restart_stack);
381 }
382 
383 /*
384  * Call function on the ipl CPU.
385  */
386 void smp_call_ipl_cpu(void (*func)(void *), void *data)
387 {
388 	struct lowcore *lc = pcpu_devices->lowcore;
389 
390 	if (pcpu_devices[0].address == stap())
391 		lc = &S390_lowcore;
392 
393 	pcpu_delegate(&pcpu_devices[0], func, data,
394 		      lc->nodat_stack);
395 }
396 
397 int smp_find_processor_id(u16 address)
398 {
399 	int cpu;
400 
401 	for_each_present_cpu(cpu)
402 		if (pcpu_devices[cpu].address == address)
403 			return cpu;
404 	return -1;
405 }
406 
407 void schedule_mcck_handler(void)
408 {
409 	pcpu_ec_call(pcpu_devices + smp_processor_id(), ec_mcck_pending);
410 }
411 
412 bool notrace arch_vcpu_is_preempted(int cpu)
413 {
414 	if (test_cpu_flag_of(CIF_ENABLED_WAIT, cpu))
415 		return false;
416 	if (pcpu_running(pcpu_devices + cpu))
417 		return false;
418 	return true;
419 }
420 EXPORT_SYMBOL(arch_vcpu_is_preempted);
421 
422 void notrace smp_yield_cpu(int cpu)
423 {
424 	if (!MACHINE_HAS_DIAG9C)
425 		return;
426 	diag_stat_inc_norecursion(DIAG_STAT_X09C);
427 	asm volatile("diag %0,0,0x9c"
428 		     : : "d" (pcpu_devices[cpu].address));
429 }
430 
431 /*
432  * Send cpus emergency shutdown signal. This gives the cpus the
433  * opportunity to complete outstanding interrupts.
434  */
435 void notrace smp_emergency_stop(void)
436 {
437 	cpumask_t cpumask;
438 	u64 end;
439 	int cpu;
440 
441 	cpumask_copy(&cpumask, cpu_online_mask);
442 	cpumask_clear_cpu(smp_processor_id(), &cpumask);
443 
444 	end = get_tod_clock() + (1000000UL << 12);
445 	for_each_cpu(cpu, &cpumask) {
446 		struct pcpu *pcpu = pcpu_devices + cpu;
447 		set_bit(ec_stop_cpu, &pcpu->ec_mask);
448 		while (__pcpu_sigp(pcpu->address, SIGP_EMERGENCY_SIGNAL,
449 				   0, NULL) == SIGP_CC_BUSY &&
450 		       get_tod_clock() < end)
451 			cpu_relax();
452 	}
453 	while (get_tod_clock() < end) {
454 		for_each_cpu(cpu, &cpumask)
455 			if (pcpu_stopped(pcpu_devices + cpu))
456 				cpumask_clear_cpu(cpu, &cpumask);
457 		if (cpumask_empty(&cpumask))
458 			break;
459 		cpu_relax();
460 	}
461 }
462 NOKPROBE_SYMBOL(smp_emergency_stop);
463 
464 /*
465  * Stop all cpus but the current one.
466  */
467 void smp_send_stop(void)
468 {
469 	int cpu;
470 
471 	/* Disable all interrupts/machine checks */
472 	__load_psw_mask(PSW_KERNEL_BITS | PSW_MASK_DAT);
473 	trace_hardirqs_off();
474 
475 	debug_set_critical();
476 
477 	if (oops_in_progress)
478 		smp_emergency_stop();
479 
480 	/* stop all processors */
481 	for_each_online_cpu(cpu) {
482 		if (cpu == smp_processor_id())
483 			continue;
484 		pcpu_sigp_retry(pcpu_devices + cpu, SIGP_STOP, 0);
485 		while (!pcpu_stopped(pcpu_devices + cpu))
486 			cpu_relax();
487 	}
488 }
489 
490 /*
491  * This is the main routine where commands issued by other
492  * cpus are handled.
493  */
494 static void smp_handle_ext_call(void)
495 {
496 	unsigned long bits;
497 
498 	/* handle bit signal external calls */
499 	bits = xchg(&pcpu_devices[smp_processor_id()].ec_mask, 0);
500 	if (test_bit(ec_stop_cpu, &bits))
501 		smp_stop_cpu();
502 	if (test_bit(ec_schedule, &bits))
503 		scheduler_ipi();
504 	if (test_bit(ec_call_function_single, &bits))
505 		generic_smp_call_function_single_interrupt();
506 	if (test_bit(ec_mcck_pending, &bits))
507 		s390_handle_mcck();
508 }
509 
510 static void do_ext_call_interrupt(struct ext_code ext_code,
511 				  unsigned int param32, unsigned long param64)
512 {
513 	inc_irq_stat(ext_code.code == 0x1202 ? IRQEXT_EXC : IRQEXT_EMS);
514 	smp_handle_ext_call();
515 }
516 
517 void arch_send_call_function_ipi_mask(const struct cpumask *mask)
518 {
519 	int cpu;
520 
521 	for_each_cpu(cpu, mask)
522 		pcpu_ec_call(pcpu_devices + cpu, ec_call_function_single);
523 }
524 
525 void arch_send_call_function_single_ipi(int cpu)
526 {
527 	pcpu_ec_call(pcpu_devices + cpu, ec_call_function_single);
528 }
529 
530 /*
531  * this function sends a 'reschedule' IPI to another CPU.
532  * it goes straight through and wastes no time serializing
533  * anything. Worst case is that we lose a reschedule ...
534  */
535 void smp_send_reschedule(int cpu)
536 {
537 	pcpu_ec_call(pcpu_devices + cpu, ec_schedule);
538 }
539 
540 /*
541  * parameter area for the set/clear control bit callbacks
542  */
543 struct ec_creg_mask_parms {
544 	unsigned long orval;
545 	unsigned long andval;
546 	int cr;
547 };
548 
549 /*
550  * callback for setting/clearing control bits
551  */
552 static void smp_ctl_bit_callback(void *info)
553 {
554 	struct ec_creg_mask_parms *pp = info;
555 	unsigned long cregs[16];
556 
557 	__ctl_store(cregs, 0, 15);
558 	cregs[pp->cr] = (cregs[pp->cr] & pp->andval) | pp->orval;
559 	__ctl_load(cregs, 0, 15);
560 }
561 
562 /*
563  * Set a bit in a control register of all cpus
564  */
565 void smp_ctl_set_bit(int cr, int bit)
566 {
567 	struct ec_creg_mask_parms parms = { 1UL << bit, -1UL, cr };
568 
569 	on_each_cpu(smp_ctl_bit_callback, &parms, 1);
570 }
571 EXPORT_SYMBOL(smp_ctl_set_bit);
572 
573 /*
574  * Clear a bit in a control register of all cpus
575  */
576 void smp_ctl_clear_bit(int cr, int bit)
577 {
578 	struct ec_creg_mask_parms parms = { 0, ~(1UL << bit), cr };
579 
580 	on_each_cpu(smp_ctl_bit_callback, &parms, 1);
581 }
582 EXPORT_SYMBOL(smp_ctl_clear_bit);
583 
584 #ifdef CONFIG_CRASH_DUMP
585 
586 int smp_store_status(int cpu)
587 {
588 	struct pcpu *pcpu = pcpu_devices + cpu;
589 	unsigned long pa;
590 
591 	pa = __pa(&pcpu->lowcore->floating_pt_save_area);
592 	if (__pcpu_sigp_relax(pcpu->address, SIGP_STORE_STATUS_AT_ADDRESS,
593 			      pa) != SIGP_CC_ORDER_CODE_ACCEPTED)
594 		return -EIO;
595 	if (!MACHINE_HAS_VX && !MACHINE_HAS_GS)
596 		return 0;
597 	pa = __pa(pcpu->lowcore->mcesad & MCESA_ORIGIN_MASK);
598 	if (MACHINE_HAS_GS)
599 		pa |= pcpu->lowcore->mcesad & MCESA_LC_MASK;
600 	if (__pcpu_sigp_relax(pcpu->address, SIGP_STORE_ADDITIONAL_STATUS,
601 			      pa) != SIGP_CC_ORDER_CODE_ACCEPTED)
602 		return -EIO;
603 	return 0;
604 }
605 
606 /*
607  * Collect CPU state of the previous, crashed system.
608  * There are four cases:
609  * 1) standard zfcp/nvme dump
610  *    condition: OLDMEM_BASE == NULL && is_ipl_type_dump() == true
611  *    The state for all CPUs except the boot CPU needs to be collected
612  *    with sigp stop-and-store-status. The boot CPU state is located in
613  *    the absolute lowcore of the memory stored in the HSA. The zcore code
614  *    will copy the boot CPU state from the HSA.
615  * 2) stand-alone kdump for SCSI/NVMe (zfcp/nvme dump with swapped memory)
616  *    condition: OLDMEM_BASE != NULL && is_ipl_type_dump() == true
617  *    The state for all CPUs except the boot CPU needs to be collected
618  *    with sigp stop-and-store-status. The firmware or the boot-loader
619  *    stored the registers of the boot CPU in the absolute lowcore in the
620  *    memory of the old system.
621  * 3) kdump and the old kernel did not store the CPU state,
622  *    or stand-alone kdump for DASD
623  *    condition: OLDMEM_BASE != NULL && !is_kdump_kernel()
624  *    The state for all CPUs except the boot CPU needs to be collected
625  *    with sigp stop-and-store-status. The kexec code or the boot-loader
626  *    stored the registers of the boot CPU in the memory of the old system.
627  * 4) kdump and the old kernel stored the CPU state
628  *    condition: OLDMEM_BASE != NULL && is_kdump_kernel()
629  *    This case does not exist for s390 anymore, setup_arch explicitly
630  *    deactivates the elfcorehdr= kernel parameter
631  */
632 static __init void smp_save_cpu_vxrs(struct save_area *sa, u16 addr,
633 				     bool is_boot_cpu, unsigned long page)
634 {
635 	__vector128 *vxrs = (__vector128 *) page;
636 
637 	if (is_boot_cpu)
638 		vxrs = boot_cpu_vector_save_area;
639 	else
640 		__pcpu_sigp_relax(addr, SIGP_STORE_ADDITIONAL_STATUS, page);
641 	save_area_add_vxrs(sa, vxrs);
642 }
643 
644 static __init void smp_save_cpu_regs(struct save_area *sa, u16 addr,
645 				     bool is_boot_cpu, unsigned long page)
646 {
647 	void *regs = (void *) page;
648 
649 	if (is_boot_cpu)
650 		copy_oldmem_kernel(regs, (void *) __LC_FPREGS_SAVE_AREA, 512);
651 	else
652 		__pcpu_sigp_relax(addr, SIGP_STORE_STATUS_AT_ADDRESS, page);
653 	save_area_add_regs(sa, regs);
654 }
655 
656 void __init smp_save_dump_cpus(void)
657 {
658 	int addr, boot_cpu_addr, max_cpu_addr;
659 	struct save_area *sa;
660 	unsigned long page;
661 	bool is_boot_cpu;
662 
663 	if (!(OLDMEM_BASE || is_ipl_type_dump()))
664 		/* No previous system present, normal boot. */
665 		return;
666 	/* Allocate a page as dumping area for the store status sigps */
667 	page = memblock_phys_alloc_range(PAGE_SIZE, PAGE_SIZE, 0, 1UL << 31);
668 	if (!page)
669 		panic("ERROR: Failed to allocate %lx bytes below %lx\n",
670 		      PAGE_SIZE, 1UL << 31);
671 
672 	/* Set multi-threading state to the previous system. */
673 	pcpu_set_smt(sclp.mtid_prev);
674 	boot_cpu_addr = stap();
675 	max_cpu_addr = SCLP_MAX_CORES << sclp.mtid_prev;
676 	for (addr = 0; addr <= max_cpu_addr; addr++) {
677 		if (__pcpu_sigp_relax(addr, SIGP_SENSE, 0) ==
678 		    SIGP_CC_NOT_OPERATIONAL)
679 			continue;
680 		is_boot_cpu = (addr == boot_cpu_addr);
681 		/* Allocate save area */
682 		sa = save_area_alloc(is_boot_cpu);
683 		if (!sa)
684 			panic("could not allocate memory for save area\n");
685 		if (MACHINE_HAS_VX)
686 			/* Get the vector registers */
687 			smp_save_cpu_vxrs(sa, addr, is_boot_cpu, page);
688 		/*
689 		 * For a zfcp/nvme dump OLDMEM_BASE == NULL and the registers
690 		 * of the boot CPU are stored in the HSA. To retrieve
691 		 * these registers an SCLP request is required which is
692 		 * done by drivers/s390/char/zcore.c:init_cpu_info()
693 		 */
694 		if (!is_boot_cpu || OLDMEM_BASE)
695 			/* Get the CPU registers */
696 			smp_save_cpu_regs(sa, addr, is_boot_cpu, page);
697 	}
698 	memblock_free(page, PAGE_SIZE);
699 	diag_dma_ops.diag308_reset();
700 	pcpu_set_smt(0);
701 }
702 #endif /* CONFIG_CRASH_DUMP */
703 
704 void smp_cpu_set_polarization(int cpu, int val)
705 {
706 	pcpu_devices[cpu].polarization = val;
707 }
708 
709 int smp_cpu_get_polarization(int cpu)
710 {
711 	return pcpu_devices[cpu].polarization;
712 }
713 
714 int smp_cpu_get_cpu_address(int cpu)
715 {
716 	return pcpu_devices[cpu].address;
717 }
718 
719 static void __ref smp_get_core_info(struct sclp_core_info *info, int early)
720 {
721 	static int use_sigp_detection;
722 	int address;
723 
724 	if (use_sigp_detection || sclp_get_core_info(info, early)) {
725 		use_sigp_detection = 1;
726 		for (address = 0;
727 		     address < (SCLP_MAX_CORES << smp_cpu_mt_shift);
728 		     address += (1U << smp_cpu_mt_shift)) {
729 			if (__pcpu_sigp_relax(address, SIGP_SENSE, 0) ==
730 			    SIGP_CC_NOT_OPERATIONAL)
731 				continue;
732 			info->core[info->configured].core_id =
733 				address >> smp_cpu_mt_shift;
734 			info->configured++;
735 		}
736 		info->combined = info->configured;
737 	}
738 }
739 
740 static int smp_add_present_cpu(int cpu);
741 
742 static int smp_add_core(struct sclp_core_entry *core, cpumask_t *avail,
743 			bool configured, bool early)
744 {
745 	struct pcpu *pcpu;
746 	int cpu, nr, i;
747 	u16 address;
748 
749 	nr = 0;
750 	if (sclp.has_core_type && core->type != boot_core_type)
751 		return nr;
752 	cpu = cpumask_first(avail);
753 	address = core->core_id << smp_cpu_mt_shift;
754 	for (i = 0; (i <= smp_cpu_mtid) && (cpu < nr_cpu_ids); i++) {
755 		if (pcpu_find_address(cpu_present_mask, address + i))
756 			continue;
757 		pcpu = pcpu_devices + cpu;
758 		pcpu->address = address + i;
759 		if (configured)
760 			pcpu->state = CPU_STATE_CONFIGURED;
761 		else
762 			pcpu->state = CPU_STATE_STANDBY;
763 		smp_cpu_set_polarization(cpu, POLARIZATION_UNKNOWN);
764 		set_cpu_present(cpu, true);
765 		if (!early && smp_add_present_cpu(cpu) != 0)
766 			set_cpu_present(cpu, false);
767 		else
768 			nr++;
769 		cpumask_clear_cpu(cpu, avail);
770 		cpu = cpumask_next(cpu, avail);
771 	}
772 	return nr;
773 }
774 
775 static int __smp_rescan_cpus(struct sclp_core_info *info, bool early)
776 {
777 	struct sclp_core_entry *core;
778 	cpumask_t avail;
779 	bool configured;
780 	u16 core_id;
781 	int nr, i;
782 
783 	nr = 0;
784 	cpumask_xor(&avail, cpu_possible_mask, cpu_present_mask);
785 	/*
786 	 * Add IPL core first (which got logical CPU number 0) to make sure
787 	 * that all SMT threads get subsequent logical CPU numbers.
788 	 */
789 	if (early) {
790 		core_id = pcpu_devices[0].address >> smp_cpu_mt_shift;
791 		for (i = 0; i < info->configured; i++) {
792 			core = &info->core[i];
793 			if (core->core_id == core_id) {
794 				nr += smp_add_core(core, &avail, true, early);
795 				break;
796 			}
797 		}
798 	}
799 	for (i = 0; i < info->combined; i++) {
800 		configured = i < info->configured;
801 		nr += smp_add_core(&info->core[i], &avail, configured, early);
802 	}
803 	return nr;
804 }
805 
806 void __init smp_detect_cpus(void)
807 {
808 	unsigned int cpu, mtid, c_cpus, s_cpus;
809 	struct sclp_core_info *info;
810 	u16 address;
811 
812 	/* Get CPU information */
813 	info = memblock_alloc(sizeof(*info), 8);
814 	if (!info)
815 		panic("%s: Failed to allocate %zu bytes align=0x%x\n",
816 		      __func__, sizeof(*info), 8);
817 	smp_get_core_info(info, 1);
818 	/* Find boot CPU type */
819 	if (sclp.has_core_type) {
820 		address = stap();
821 		for (cpu = 0; cpu < info->combined; cpu++)
822 			if (info->core[cpu].core_id == address) {
823 				/* The boot cpu dictates the cpu type. */
824 				boot_core_type = info->core[cpu].type;
825 				break;
826 			}
827 		if (cpu >= info->combined)
828 			panic("Could not find boot CPU type");
829 	}
830 
831 	/* Set multi-threading state for the current system */
832 	mtid = boot_core_type ? sclp.mtid : sclp.mtid_cp;
833 	mtid = (mtid < smp_max_threads) ? mtid : smp_max_threads - 1;
834 	pcpu_set_smt(mtid);
835 
836 	/* Print number of CPUs */
837 	c_cpus = s_cpus = 0;
838 	for (cpu = 0; cpu < info->combined; cpu++) {
839 		if (sclp.has_core_type &&
840 		    info->core[cpu].type != boot_core_type)
841 			continue;
842 		if (cpu < info->configured)
843 			c_cpus += smp_cpu_mtid + 1;
844 		else
845 			s_cpus += smp_cpu_mtid + 1;
846 	}
847 	pr_info("%d configured CPUs, %d standby CPUs\n", c_cpus, s_cpus);
848 
849 	/* Add CPUs present at boot */
850 	get_online_cpus();
851 	__smp_rescan_cpus(info, true);
852 	put_online_cpus();
853 	memblock_free_early((unsigned long)info, sizeof(*info));
854 }
855 
856 static void smp_init_secondary(void)
857 {
858 	int cpu = raw_smp_processor_id();
859 
860 	S390_lowcore.last_update_clock = get_tod_clock();
861 	restore_access_regs(S390_lowcore.access_regs_save_area);
862 	set_cpu_flag(CIF_ASCE_PRIMARY);
863 	set_cpu_flag(CIF_ASCE_SECONDARY);
864 	cpu_init();
865 	rcu_cpu_starting(cpu);
866 	preempt_disable();
867 	init_cpu_timer();
868 	vtime_init();
869 	pfault_init();
870 	notify_cpu_starting(cpu);
871 	if (topology_cpu_dedicated(cpu))
872 		set_cpu_flag(CIF_DEDICATED_CPU);
873 	else
874 		clear_cpu_flag(CIF_DEDICATED_CPU);
875 	set_cpu_online(cpu, true);
876 	update_cpu_masks();
877 	inc_irq_stat(CPU_RST);
878 	local_irq_enable();
879 	cpu_startup_entry(CPUHP_AP_ONLINE_IDLE);
880 }
881 
882 /*
883  *	Activate a secondary processor.
884  */
885 static void __no_sanitize_address smp_start_secondary(void *cpuvoid)
886 {
887 	S390_lowcore.restart_stack = (unsigned long) restart_stack;
888 	S390_lowcore.restart_fn = (unsigned long) do_restart;
889 	S390_lowcore.restart_data = 0;
890 	S390_lowcore.restart_source = -1UL;
891 	__ctl_load(S390_lowcore.cregs_save_area, 0, 15);
892 	__load_psw_mask(PSW_KERNEL_BITS | PSW_MASK_DAT);
893 	CALL_ON_STACK_NORETURN(smp_init_secondary, S390_lowcore.kernel_stack);
894 }
895 
896 /* Upping and downing of CPUs */
897 int __cpu_up(unsigned int cpu, struct task_struct *tidle)
898 {
899 	struct pcpu *pcpu;
900 	int base, i, rc;
901 
902 	pcpu = pcpu_devices + cpu;
903 	if (pcpu->state != CPU_STATE_CONFIGURED)
904 		return -EIO;
905 	base = smp_get_base_cpu(cpu);
906 	for (i = 0; i <= smp_cpu_mtid; i++) {
907 		if (base + i < nr_cpu_ids)
908 			if (cpu_online(base + i))
909 				break;
910 	}
911 	/*
912 	 * If this is the first CPU of the core to get online
913 	 * do an initial CPU reset.
914 	 */
915 	if (i > smp_cpu_mtid &&
916 	    pcpu_sigp_retry(pcpu_devices + base, SIGP_INITIAL_CPU_RESET, 0) !=
917 	    SIGP_CC_ORDER_CODE_ACCEPTED)
918 		return -EIO;
919 
920 	rc = pcpu_alloc_lowcore(pcpu, cpu);
921 	if (rc)
922 		return rc;
923 	pcpu_prepare_secondary(pcpu, cpu);
924 	pcpu_attach_task(pcpu, tidle);
925 	pcpu_start_fn(pcpu, smp_start_secondary, NULL);
926 	/* Wait until cpu puts itself in the online & active maps */
927 	while (!cpu_online(cpu))
928 		cpu_relax();
929 	return 0;
930 }
931 
932 static unsigned int setup_possible_cpus __initdata;
933 
934 static int __init _setup_possible_cpus(char *s)
935 {
936 	get_option(&s, &setup_possible_cpus);
937 	return 0;
938 }
939 early_param("possible_cpus", _setup_possible_cpus);
940 
941 int __cpu_disable(void)
942 {
943 	unsigned long cregs[16];
944 
945 	/* Handle possible pending IPIs */
946 	smp_handle_ext_call();
947 	set_cpu_online(smp_processor_id(), false);
948 	update_cpu_masks();
949 	/* Disable pseudo page faults on this cpu. */
950 	pfault_fini();
951 	/* Disable interrupt sources via control register. */
952 	__ctl_store(cregs, 0, 15);
953 	cregs[0]  &= ~0x0000ee70UL;	/* disable all external interrupts */
954 	cregs[6]  &= ~0xff000000UL;	/* disable all I/O interrupts */
955 	cregs[14] &= ~0x1f000000UL;	/* disable most machine checks */
956 	__ctl_load(cregs, 0, 15);
957 	clear_cpu_flag(CIF_NOHZ_DELAY);
958 	return 0;
959 }
960 
961 void __cpu_die(unsigned int cpu)
962 {
963 	struct pcpu *pcpu;
964 
965 	/* Wait until target cpu is down */
966 	pcpu = pcpu_devices + cpu;
967 	while (!pcpu_stopped(pcpu))
968 		cpu_relax();
969 	pcpu_free_lowcore(pcpu);
970 	cpumask_clear_cpu(cpu, mm_cpumask(&init_mm));
971 	cpumask_clear_cpu(cpu, &init_mm.context.cpu_attach_mask);
972 }
973 
974 void __noreturn cpu_die(void)
975 {
976 	idle_task_exit();
977 	__bpon();
978 	pcpu_sigp_retry(pcpu_devices + smp_processor_id(), SIGP_STOP, 0);
979 	for (;;) ;
980 }
981 
982 void __init smp_fill_possible_mask(void)
983 {
984 	unsigned int possible, sclp_max, cpu;
985 
986 	sclp_max = max(sclp.mtid, sclp.mtid_cp) + 1;
987 	sclp_max = min(smp_max_threads, sclp_max);
988 	sclp_max = (sclp.max_cores * sclp_max) ?: nr_cpu_ids;
989 	possible = setup_possible_cpus ?: nr_cpu_ids;
990 	possible = min(possible, sclp_max);
991 	for (cpu = 0; cpu < possible && cpu < nr_cpu_ids; cpu++)
992 		set_cpu_possible(cpu, true);
993 }
994 
995 void __init smp_prepare_cpus(unsigned int max_cpus)
996 {
997 	/* request the 0x1201 emergency signal external interrupt */
998 	if (register_external_irq(EXT_IRQ_EMERGENCY_SIG, do_ext_call_interrupt))
999 		panic("Couldn't request external interrupt 0x1201");
1000 	/* request the 0x1202 external call external interrupt */
1001 	if (register_external_irq(EXT_IRQ_EXTERNAL_CALL, do_ext_call_interrupt))
1002 		panic("Couldn't request external interrupt 0x1202");
1003 }
1004 
1005 void __init smp_prepare_boot_cpu(void)
1006 {
1007 	struct pcpu *pcpu = pcpu_devices;
1008 
1009 	WARN_ON(!cpu_present(0) || !cpu_online(0));
1010 	pcpu->state = CPU_STATE_CONFIGURED;
1011 	pcpu->lowcore = (struct lowcore *)(unsigned long) store_prefix();
1012 	S390_lowcore.percpu_offset = __per_cpu_offset[0];
1013 	smp_cpu_set_polarization(0, POLARIZATION_UNKNOWN);
1014 }
1015 
1016 void __init smp_setup_processor_id(void)
1017 {
1018 	pcpu_devices[0].address = stap();
1019 	S390_lowcore.cpu_nr = 0;
1020 	S390_lowcore.spinlock_lockval = arch_spin_lockval(0);
1021 	S390_lowcore.spinlock_index = 0;
1022 }
1023 
1024 /*
1025  * the frequency of the profiling timer can be changed
1026  * by writing a multiplier value into /proc/profile.
1027  *
1028  * usually you want to run this on all CPUs ;)
1029  */
1030 int setup_profiling_timer(unsigned int multiplier)
1031 {
1032 	return 0;
1033 }
1034 
1035 static ssize_t cpu_configure_show(struct device *dev,
1036 				  struct device_attribute *attr, char *buf)
1037 {
1038 	ssize_t count;
1039 
1040 	mutex_lock(&smp_cpu_state_mutex);
1041 	count = sprintf(buf, "%d\n", pcpu_devices[dev->id].state);
1042 	mutex_unlock(&smp_cpu_state_mutex);
1043 	return count;
1044 }
1045 
1046 static ssize_t cpu_configure_store(struct device *dev,
1047 				   struct device_attribute *attr,
1048 				   const char *buf, size_t count)
1049 {
1050 	struct pcpu *pcpu;
1051 	int cpu, val, rc, i;
1052 	char delim;
1053 
1054 	if (sscanf(buf, "%d %c", &val, &delim) != 1)
1055 		return -EINVAL;
1056 	if (val != 0 && val != 1)
1057 		return -EINVAL;
1058 	get_online_cpus();
1059 	mutex_lock(&smp_cpu_state_mutex);
1060 	rc = -EBUSY;
1061 	/* disallow configuration changes of online cpus and cpu 0 */
1062 	cpu = dev->id;
1063 	cpu = smp_get_base_cpu(cpu);
1064 	if (cpu == 0)
1065 		goto out;
1066 	for (i = 0; i <= smp_cpu_mtid; i++)
1067 		if (cpu_online(cpu + i))
1068 			goto out;
1069 	pcpu = pcpu_devices + cpu;
1070 	rc = 0;
1071 	switch (val) {
1072 	case 0:
1073 		if (pcpu->state != CPU_STATE_CONFIGURED)
1074 			break;
1075 		rc = sclp_core_deconfigure(pcpu->address >> smp_cpu_mt_shift);
1076 		if (rc)
1077 			break;
1078 		for (i = 0; i <= smp_cpu_mtid; i++) {
1079 			if (cpu + i >= nr_cpu_ids || !cpu_present(cpu + i))
1080 				continue;
1081 			pcpu[i].state = CPU_STATE_STANDBY;
1082 			smp_cpu_set_polarization(cpu + i,
1083 						 POLARIZATION_UNKNOWN);
1084 		}
1085 		topology_expect_change();
1086 		break;
1087 	case 1:
1088 		if (pcpu->state != CPU_STATE_STANDBY)
1089 			break;
1090 		rc = sclp_core_configure(pcpu->address >> smp_cpu_mt_shift);
1091 		if (rc)
1092 			break;
1093 		for (i = 0; i <= smp_cpu_mtid; i++) {
1094 			if (cpu + i >= nr_cpu_ids || !cpu_present(cpu + i))
1095 				continue;
1096 			pcpu[i].state = CPU_STATE_CONFIGURED;
1097 			smp_cpu_set_polarization(cpu + i,
1098 						 POLARIZATION_UNKNOWN);
1099 		}
1100 		topology_expect_change();
1101 		break;
1102 	default:
1103 		break;
1104 	}
1105 out:
1106 	mutex_unlock(&smp_cpu_state_mutex);
1107 	put_online_cpus();
1108 	return rc ? rc : count;
1109 }
1110 static DEVICE_ATTR(configure, 0644, cpu_configure_show, cpu_configure_store);
1111 
1112 static ssize_t show_cpu_address(struct device *dev,
1113 				struct device_attribute *attr, char *buf)
1114 {
1115 	return sprintf(buf, "%d\n", pcpu_devices[dev->id].address);
1116 }
1117 static DEVICE_ATTR(address, 0444, show_cpu_address, NULL);
1118 
1119 static struct attribute *cpu_common_attrs[] = {
1120 	&dev_attr_configure.attr,
1121 	&dev_attr_address.attr,
1122 	NULL,
1123 };
1124 
1125 static struct attribute_group cpu_common_attr_group = {
1126 	.attrs = cpu_common_attrs,
1127 };
1128 
1129 static struct attribute *cpu_online_attrs[] = {
1130 	&dev_attr_idle_count.attr,
1131 	&dev_attr_idle_time_us.attr,
1132 	NULL,
1133 };
1134 
1135 static struct attribute_group cpu_online_attr_group = {
1136 	.attrs = cpu_online_attrs,
1137 };
1138 
1139 static int smp_cpu_online(unsigned int cpu)
1140 {
1141 	struct device *s = &per_cpu(cpu_device, cpu)->dev;
1142 
1143 	return sysfs_create_group(&s->kobj, &cpu_online_attr_group);
1144 }
1145 
1146 static int smp_cpu_pre_down(unsigned int cpu)
1147 {
1148 	struct device *s = &per_cpu(cpu_device, cpu)->dev;
1149 
1150 	sysfs_remove_group(&s->kobj, &cpu_online_attr_group);
1151 	return 0;
1152 }
1153 
1154 static int smp_add_present_cpu(int cpu)
1155 {
1156 	struct device *s;
1157 	struct cpu *c;
1158 	int rc;
1159 
1160 	c = kzalloc(sizeof(*c), GFP_KERNEL);
1161 	if (!c)
1162 		return -ENOMEM;
1163 	per_cpu(cpu_device, cpu) = c;
1164 	s = &c->dev;
1165 	c->hotpluggable = 1;
1166 	rc = register_cpu(c, cpu);
1167 	if (rc)
1168 		goto out;
1169 	rc = sysfs_create_group(&s->kobj, &cpu_common_attr_group);
1170 	if (rc)
1171 		goto out_cpu;
1172 	rc = topology_cpu_init(c);
1173 	if (rc)
1174 		goto out_topology;
1175 	return 0;
1176 
1177 out_topology:
1178 	sysfs_remove_group(&s->kobj, &cpu_common_attr_group);
1179 out_cpu:
1180 	unregister_cpu(c);
1181 out:
1182 	return rc;
1183 }
1184 
1185 int __ref smp_rescan_cpus(void)
1186 {
1187 	struct sclp_core_info *info;
1188 	int nr;
1189 
1190 	info = kzalloc(sizeof(*info), GFP_KERNEL);
1191 	if (!info)
1192 		return -ENOMEM;
1193 	smp_get_core_info(info, 0);
1194 	get_online_cpus();
1195 	mutex_lock(&smp_cpu_state_mutex);
1196 	nr = __smp_rescan_cpus(info, false);
1197 	mutex_unlock(&smp_cpu_state_mutex);
1198 	put_online_cpus();
1199 	kfree(info);
1200 	if (nr)
1201 		topology_schedule_update();
1202 	return 0;
1203 }
1204 
1205 static ssize_t __ref rescan_store(struct device *dev,
1206 				  struct device_attribute *attr,
1207 				  const char *buf,
1208 				  size_t count)
1209 {
1210 	int rc;
1211 
1212 	rc = lock_device_hotplug_sysfs();
1213 	if (rc)
1214 		return rc;
1215 	rc = smp_rescan_cpus();
1216 	unlock_device_hotplug();
1217 	return rc ? rc : count;
1218 }
1219 static DEVICE_ATTR_WO(rescan);
1220 
1221 static int __init s390_smp_init(void)
1222 {
1223 	int cpu, rc = 0;
1224 
1225 	rc = device_create_file(cpu_subsys.dev_root, &dev_attr_rescan);
1226 	if (rc)
1227 		return rc;
1228 	for_each_present_cpu(cpu) {
1229 		rc = smp_add_present_cpu(cpu);
1230 		if (rc)
1231 			goto out;
1232 	}
1233 
1234 	rc = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "s390/smp:online",
1235 			       smp_cpu_online, smp_cpu_pre_down);
1236 	rc = rc <= 0 ? rc : 0;
1237 out:
1238 	return rc;
1239 }
1240 subsys_initcall(s390_smp_init);
1241