xref: /openbmc/linux/arch/s390/kernel/smp.c (revision 5e012745)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  *  SMP related functions
4  *
5  *    Copyright IBM Corp. 1999, 2012
6  *    Author(s): Denis Joseph Barrow,
7  *		 Martin Schwidefsky <schwidefsky@de.ibm.com>,
8  *		 Heiko Carstens <heiko.carstens@de.ibm.com>,
9  *
10  *  based on other smp stuff by
11  *    (c) 1995 Alan Cox, CymruNET Ltd  <alan@cymru.net>
12  *    (c) 1998 Ingo Molnar
13  *
14  * The code outside of smp.c uses logical cpu numbers, only smp.c does
15  * the translation of logical to physical cpu ids. All new code that
16  * operates on physical cpu numbers needs to go into smp.c.
17  */
18 
19 #define KMSG_COMPONENT "cpu"
20 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
21 
22 #include <linux/workqueue.h>
23 #include <linux/memblock.h>
24 #include <linux/export.h>
25 #include <linux/init.h>
26 #include <linux/mm.h>
27 #include <linux/err.h>
28 #include <linux/spinlock.h>
29 #include <linux/kernel_stat.h>
30 #include <linux/delay.h>
31 #include <linux/interrupt.h>
32 #include <linux/irqflags.h>
33 #include <linux/cpu.h>
34 #include <linux/slab.h>
35 #include <linux/sched/hotplug.h>
36 #include <linux/sched/task_stack.h>
37 #include <linux/crash_dump.h>
38 #include <linux/kprobes.h>
39 #include <asm/asm-offsets.h>
40 #include <asm/diag.h>
41 #include <asm/switch_to.h>
42 #include <asm/facility.h>
43 #include <asm/ipl.h>
44 #include <asm/setup.h>
45 #include <asm/irq.h>
46 #include <asm/tlbflush.h>
47 #include <asm/vtimer.h>
48 #include <asm/lowcore.h>
49 #include <asm/sclp.h>
50 #include <asm/vdso.h>
51 #include <asm/debug.h>
52 #include <asm/os_info.h>
53 #include <asm/sigp.h>
54 #include <asm/idle.h>
55 #include <asm/nmi.h>
56 #include <asm/stacktrace.h>
57 #include <asm/topology.h>
58 #include "entry.h"
59 
60 enum {
61 	ec_schedule = 0,
62 	ec_call_function_single,
63 	ec_stop_cpu,
64 };
65 
66 enum {
67 	CPU_STATE_STANDBY,
68 	CPU_STATE_CONFIGURED,
69 };
70 
71 static DEFINE_PER_CPU(struct cpu *, cpu_device);
72 
73 struct pcpu {
74 	struct lowcore *lowcore;	/* lowcore page(s) for the cpu */
75 	unsigned long ec_mask;		/* bit mask for ec_xxx functions */
76 	unsigned long ec_clk;		/* sigp timestamp for ec_xxx */
77 	signed char state;		/* physical cpu state */
78 	signed char polarization;	/* physical polarization */
79 	u16 address;			/* physical cpu address */
80 };
81 
82 static u8 boot_core_type;
83 static struct pcpu pcpu_devices[NR_CPUS];
84 
85 unsigned int smp_cpu_mt_shift;
86 EXPORT_SYMBOL(smp_cpu_mt_shift);
87 
88 unsigned int smp_cpu_mtid;
89 EXPORT_SYMBOL(smp_cpu_mtid);
90 
91 #ifdef CONFIG_CRASH_DUMP
92 __vector128 __initdata boot_cpu_vector_save_area[__NUM_VXRS];
93 #endif
94 
95 static unsigned int smp_max_threads __initdata = -1U;
96 
97 static int __init early_nosmt(char *s)
98 {
99 	smp_max_threads = 1;
100 	return 0;
101 }
102 early_param("nosmt", early_nosmt);
103 
104 static int __init early_smt(char *s)
105 {
106 	get_option(&s, &smp_max_threads);
107 	return 0;
108 }
109 early_param("smt", early_smt);
110 
111 /*
112  * The smp_cpu_state_mutex must be held when changing the state or polarization
113  * member of a pcpu data structure within the pcpu_devices arreay.
114  */
115 DEFINE_MUTEX(smp_cpu_state_mutex);
116 
117 /*
118  * Signal processor helper functions.
119  */
120 static inline int __pcpu_sigp_relax(u16 addr, u8 order, unsigned long parm)
121 {
122 	int cc;
123 
124 	while (1) {
125 		cc = __pcpu_sigp(addr, order, parm, NULL);
126 		if (cc != SIGP_CC_BUSY)
127 			return cc;
128 		cpu_relax();
129 	}
130 }
131 
132 static int pcpu_sigp_retry(struct pcpu *pcpu, u8 order, u32 parm)
133 {
134 	int cc, retry;
135 
136 	for (retry = 0; ; retry++) {
137 		cc = __pcpu_sigp(pcpu->address, order, parm, NULL);
138 		if (cc != SIGP_CC_BUSY)
139 			break;
140 		if (retry >= 3)
141 			udelay(10);
142 	}
143 	return cc;
144 }
145 
146 static inline int pcpu_stopped(struct pcpu *pcpu)
147 {
148 	u32 uninitialized_var(status);
149 
150 	if (__pcpu_sigp(pcpu->address, SIGP_SENSE,
151 			0, &status) != SIGP_CC_STATUS_STORED)
152 		return 0;
153 	return !!(status & (SIGP_STATUS_CHECK_STOP|SIGP_STATUS_STOPPED));
154 }
155 
156 static inline int pcpu_running(struct pcpu *pcpu)
157 {
158 	if (__pcpu_sigp(pcpu->address, SIGP_SENSE_RUNNING,
159 			0, NULL) != SIGP_CC_STATUS_STORED)
160 		return 1;
161 	/* Status stored condition code is equivalent to cpu not running. */
162 	return 0;
163 }
164 
165 /*
166  * Find struct pcpu by cpu address.
167  */
168 static struct pcpu *pcpu_find_address(const struct cpumask *mask, u16 address)
169 {
170 	int cpu;
171 
172 	for_each_cpu(cpu, mask)
173 		if (pcpu_devices[cpu].address == address)
174 			return pcpu_devices + cpu;
175 	return NULL;
176 }
177 
178 static void pcpu_ec_call(struct pcpu *pcpu, int ec_bit)
179 {
180 	int order;
181 
182 	if (test_and_set_bit(ec_bit, &pcpu->ec_mask))
183 		return;
184 	order = pcpu_running(pcpu) ? SIGP_EXTERNAL_CALL : SIGP_EMERGENCY_SIGNAL;
185 	pcpu->ec_clk = get_tod_clock_fast();
186 	pcpu_sigp_retry(pcpu, order, 0);
187 }
188 
189 static int pcpu_alloc_lowcore(struct pcpu *pcpu, int cpu)
190 {
191 	unsigned long async_stack, nodat_stack;
192 	struct lowcore *lc;
193 
194 	if (pcpu != &pcpu_devices[0]) {
195 		pcpu->lowcore =	(struct lowcore *)
196 			__get_free_pages(GFP_KERNEL | GFP_DMA, LC_ORDER);
197 		nodat_stack = __get_free_pages(GFP_KERNEL, THREAD_SIZE_ORDER);
198 		if (!pcpu->lowcore || !nodat_stack)
199 			goto out;
200 	} else {
201 		nodat_stack = pcpu->lowcore->nodat_stack - STACK_INIT_OFFSET;
202 	}
203 	async_stack = stack_alloc();
204 	if (!async_stack)
205 		goto out;
206 	lc = pcpu->lowcore;
207 	memcpy(lc, &S390_lowcore, 512);
208 	memset((char *) lc + 512, 0, sizeof(*lc) - 512);
209 	lc->async_stack = async_stack + STACK_INIT_OFFSET;
210 	lc->nodat_stack = nodat_stack + STACK_INIT_OFFSET;
211 	lc->cpu_nr = cpu;
212 	lc->spinlock_lockval = arch_spin_lockval(cpu);
213 	lc->spinlock_index = 0;
214 	lc->br_r1_trampoline = 0x07f1;	/* br %r1 */
215 	if (nmi_alloc_per_cpu(lc))
216 		goto out_async;
217 	if (vdso_alloc_per_cpu(lc))
218 		goto out_mcesa;
219 	lowcore_ptr[cpu] = lc;
220 	pcpu_sigp_retry(pcpu, SIGP_SET_PREFIX, (u32)(unsigned long) lc);
221 	return 0;
222 
223 out_mcesa:
224 	nmi_free_per_cpu(lc);
225 out_async:
226 	stack_free(async_stack);
227 out:
228 	if (pcpu != &pcpu_devices[0]) {
229 		free_pages(nodat_stack, THREAD_SIZE_ORDER);
230 		free_pages((unsigned long) pcpu->lowcore, LC_ORDER);
231 	}
232 	return -ENOMEM;
233 }
234 
235 static void pcpu_free_lowcore(struct pcpu *pcpu)
236 {
237 	unsigned long async_stack, nodat_stack, lowcore;
238 
239 	nodat_stack = pcpu->lowcore->nodat_stack - STACK_INIT_OFFSET;
240 	async_stack = pcpu->lowcore->async_stack - STACK_INIT_OFFSET;
241 	lowcore = (unsigned long) pcpu->lowcore;
242 
243 	pcpu_sigp_retry(pcpu, SIGP_SET_PREFIX, 0);
244 	lowcore_ptr[pcpu - pcpu_devices] = NULL;
245 	vdso_free_per_cpu(pcpu->lowcore);
246 	nmi_free_per_cpu(pcpu->lowcore);
247 	stack_free(async_stack);
248 	if (pcpu == &pcpu_devices[0])
249 		return;
250 	free_pages(nodat_stack, THREAD_SIZE_ORDER);
251 	free_pages(lowcore, LC_ORDER);
252 }
253 
254 static void pcpu_prepare_secondary(struct pcpu *pcpu, int cpu)
255 {
256 	struct lowcore *lc = pcpu->lowcore;
257 
258 	cpumask_set_cpu(cpu, &init_mm.context.cpu_attach_mask);
259 	cpumask_set_cpu(cpu, mm_cpumask(&init_mm));
260 	lc->cpu_nr = cpu;
261 	lc->spinlock_lockval = arch_spin_lockval(cpu);
262 	lc->spinlock_index = 0;
263 	lc->percpu_offset = __per_cpu_offset[cpu];
264 	lc->kernel_asce = S390_lowcore.kernel_asce;
265 	lc->machine_flags = S390_lowcore.machine_flags;
266 	lc->user_timer = lc->system_timer =
267 		lc->steal_timer = lc->avg_steal_timer = 0;
268 	__ctl_store(lc->cregs_save_area, 0, 15);
269 	save_access_regs((unsigned int *) lc->access_regs_save_area);
270 	memcpy(lc->stfle_fac_list, S390_lowcore.stfle_fac_list,
271 	       sizeof(lc->stfle_fac_list));
272 	memcpy(lc->alt_stfle_fac_list, S390_lowcore.alt_stfle_fac_list,
273 	       sizeof(lc->alt_stfle_fac_list));
274 	arch_spin_lock_setup(cpu);
275 }
276 
277 static void pcpu_attach_task(struct pcpu *pcpu, struct task_struct *tsk)
278 {
279 	struct lowcore *lc = pcpu->lowcore;
280 
281 	lc->kernel_stack = (unsigned long) task_stack_page(tsk)
282 		+ THREAD_SIZE - STACK_FRAME_OVERHEAD - sizeof(struct pt_regs);
283 	lc->current_task = (unsigned long) tsk;
284 	lc->lpp = LPP_MAGIC;
285 	lc->current_pid = tsk->pid;
286 	lc->user_timer = tsk->thread.user_timer;
287 	lc->guest_timer = tsk->thread.guest_timer;
288 	lc->system_timer = tsk->thread.system_timer;
289 	lc->hardirq_timer = tsk->thread.hardirq_timer;
290 	lc->softirq_timer = tsk->thread.softirq_timer;
291 	lc->steal_timer = 0;
292 }
293 
294 static void pcpu_start_fn(struct pcpu *pcpu, void (*func)(void *), void *data)
295 {
296 	struct lowcore *lc = pcpu->lowcore;
297 
298 	lc->restart_stack = lc->nodat_stack;
299 	lc->restart_fn = (unsigned long) func;
300 	lc->restart_data = (unsigned long) data;
301 	lc->restart_source = -1UL;
302 	pcpu_sigp_retry(pcpu, SIGP_RESTART, 0);
303 }
304 
305 /*
306  * Call function via PSW restart on pcpu and stop the current cpu.
307  */
308 static void __pcpu_delegate(void (*func)(void*), void *data)
309 {
310 	func(data);	/* should not return */
311 }
312 
313 static void __no_sanitize_address pcpu_delegate(struct pcpu *pcpu,
314 						void (*func)(void *),
315 						void *data, unsigned long stack)
316 {
317 	struct lowcore *lc = lowcore_ptr[pcpu - pcpu_devices];
318 	unsigned long source_cpu = stap();
319 
320 	__load_psw_mask(PSW_KERNEL_BITS | PSW_MASK_DAT);
321 	if (pcpu->address == source_cpu)
322 		CALL_ON_STACK(__pcpu_delegate, stack, 2, func, data);
323 	/* Stop target cpu (if func returns this stops the current cpu). */
324 	pcpu_sigp_retry(pcpu, SIGP_STOP, 0);
325 	/* Restart func on the target cpu and stop the current cpu. */
326 	mem_assign_absolute(lc->restart_stack, stack);
327 	mem_assign_absolute(lc->restart_fn, (unsigned long) func);
328 	mem_assign_absolute(lc->restart_data, (unsigned long) data);
329 	mem_assign_absolute(lc->restart_source, source_cpu);
330 	__bpon();
331 	asm volatile(
332 		"0:	sigp	0,%0,%2	# sigp restart to target cpu\n"
333 		"	brc	2,0b	# busy, try again\n"
334 		"1:	sigp	0,%1,%3	# sigp stop to current cpu\n"
335 		"	brc	2,1b	# busy, try again\n"
336 		: : "d" (pcpu->address), "d" (source_cpu),
337 		    "K" (SIGP_RESTART), "K" (SIGP_STOP)
338 		: "0", "1", "cc");
339 	for (;;) ;
340 }
341 
342 /*
343  * Enable additional logical cpus for multi-threading.
344  */
345 static int pcpu_set_smt(unsigned int mtid)
346 {
347 	int cc;
348 
349 	if (smp_cpu_mtid == mtid)
350 		return 0;
351 	cc = __pcpu_sigp(0, SIGP_SET_MULTI_THREADING, mtid, NULL);
352 	if (cc == 0) {
353 		smp_cpu_mtid = mtid;
354 		smp_cpu_mt_shift = 0;
355 		while (smp_cpu_mtid >= (1U << smp_cpu_mt_shift))
356 			smp_cpu_mt_shift++;
357 		pcpu_devices[0].address = stap();
358 	}
359 	return cc;
360 }
361 
362 /*
363  * Call function on an online CPU.
364  */
365 void smp_call_online_cpu(void (*func)(void *), void *data)
366 {
367 	struct pcpu *pcpu;
368 
369 	/* Use the current cpu if it is online. */
370 	pcpu = pcpu_find_address(cpu_online_mask, stap());
371 	if (!pcpu)
372 		/* Use the first online cpu. */
373 		pcpu = pcpu_devices + cpumask_first(cpu_online_mask);
374 	pcpu_delegate(pcpu, func, data, (unsigned long) restart_stack);
375 }
376 
377 /*
378  * Call function on the ipl CPU.
379  */
380 void smp_call_ipl_cpu(void (*func)(void *), void *data)
381 {
382 	struct lowcore *lc = pcpu_devices->lowcore;
383 
384 	if (pcpu_devices[0].address == stap())
385 		lc = &S390_lowcore;
386 
387 	pcpu_delegate(&pcpu_devices[0], func, data,
388 		      lc->nodat_stack);
389 }
390 
391 int smp_find_processor_id(u16 address)
392 {
393 	int cpu;
394 
395 	for_each_present_cpu(cpu)
396 		if (pcpu_devices[cpu].address == address)
397 			return cpu;
398 	return -1;
399 }
400 
401 bool arch_vcpu_is_preempted(int cpu)
402 {
403 	if (test_cpu_flag_of(CIF_ENABLED_WAIT, cpu))
404 		return false;
405 	if (pcpu_running(pcpu_devices + cpu))
406 		return false;
407 	return true;
408 }
409 EXPORT_SYMBOL(arch_vcpu_is_preempted);
410 
411 void smp_yield_cpu(int cpu)
412 {
413 	if (MACHINE_HAS_DIAG9C) {
414 		diag_stat_inc_norecursion(DIAG_STAT_X09C);
415 		asm volatile("diag %0,0,0x9c"
416 			     : : "d" (pcpu_devices[cpu].address));
417 	} else if (MACHINE_HAS_DIAG44 && !smp_cpu_mtid) {
418 		diag_stat_inc_norecursion(DIAG_STAT_X044);
419 		asm volatile("diag 0,0,0x44");
420 	}
421 }
422 
423 /*
424  * Send cpus emergency shutdown signal. This gives the cpus the
425  * opportunity to complete outstanding interrupts.
426  */
427 void notrace smp_emergency_stop(void)
428 {
429 	cpumask_t cpumask;
430 	u64 end;
431 	int cpu;
432 
433 	cpumask_copy(&cpumask, cpu_online_mask);
434 	cpumask_clear_cpu(smp_processor_id(), &cpumask);
435 
436 	end = get_tod_clock() + (1000000UL << 12);
437 	for_each_cpu(cpu, &cpumask) {
438 		struct pcpu *pcpu = pcpu_devices + cpu;
439 		set_bit(ec_stop_cpu, &pcpu->ec_mask);
440 		while (__pcpu_sigp(pcpu->address, SIGP_EMERGENCY_SIGNAL,
441 				   0, NULL) == SIGP_CC_BUSY &&
442 		       get_tod_clock() < end)
443 			cpu_relax();
444 	}
445 	while (get_tod_clock() < end) {
446 		for_each_cpu(cpu, &cpumask)
447 			if (pcpu_stopped(pcpu_devices + cpu))
448 				cpumask_clear_cpu(cpu, &cpumask);
449 		if (cpumask_empty(&cpumask))
450 			break;
451 		cpu_relax();
452 	}
453 }
454 NOKPROBE_SYMBOL(smp_emergency_stop);
455 
456 /*
457  * Stop all cpus but the current one.
458  */
459 void smp_send_stop(void)
460 {
461 	int cpu;
462 
463 	/* Disable all interrupts/machine checks */
464 	__load_psw_mask(PSW_KERNEL_BITS | PSW_MASK_DAT);
465 	trace_hardirqs_off();
466 
467 	debug_set_critical();
468 
469 	if (oops_in_progress)
470 		smp_emergency_stop();
471 
472 	/* stop all processors */
473 	for_each_online_cpu(cpu) {
474 		if (cpu == smp_processor_id())
475 			continue;
476 		pcpu_sigp_retry(pcpu_devices + cpu, SIGP_STOP, 0);
477 		while (!pcpu_stopped(pcpu_devices + cpu))
478 			cpu_relax();
479 	}
480 }
481 
482 /*
483  * This is the main routine where commands issued by other
484  * cpus are handled.
485  */
486 static void smp_handle_ext_call(void)
487 {
488 	unsigned long bits;
489 
490 	/* handle bit signal external calls */
491 	bits = xchg(&pcpu_devices[smp_processor_id()].ec_mask, 0);
492 	if (test_bit(ec_stop_cpu, &bits))
493 		smp_stop_cpu();
494 	if (test_bit(ec_schedule, &bits))
495 		scheduler_ipi();
496 	if (test_bit(ec_call_function_single, &bits))
497 		generic_smp_call_function_single_interrupt();
498 }
499 
500 static void do_ext_call_interrupt(struct ext_code ext_code,
501 				  unsigned int param32, unsigned long param64)
502 {
503 	inc_irq_stat(ext_code.code == 0x1202 ? IRQEXT_EXC : IRQEXT_EMS);
504 	smp_handle_ext_call();
505 }
506 
507 void arch_send_call_function_ipi_mask(const struct cpumask *mask)
508 {
509 	int cpu;
510 
511 	for_each_cpu(cpu, mask)
512 		pcpu_ec_call(pcpu_devices + cpu, ec_call_function_single);
513 }
514 
515 void arch_send_call_function_single_ipi(int cpu)
516 {
517 	pcpu_ec_call(pcpu_devices + cpu, ec_call_function_single);
518 }
519 
520 /*
521  * this function sends a 'reschedule' IPI to another CPU.
522  * it goes straight through and wastes no time serializing
523  * anything. Worst case is that we lose a reschedule ...
524  */
525 void smp_send_reschedule(int cpu)
526 {
527 	pcpu_ec_call(pcpu_devices + cpu, ec_schedule);
528 }
529 
530 /*
531  * parameter area for the set/clear control bit callbacks
532  */
533 struct ec_creg_mask_parms {
534 	unsigned long orval;
535 	unsigned long andval;
536 	int cr;
537 };
538 
539 /*
540  * callback for setting/clearing control bits
541  */
542 static void smp_ctl_bit_callback(void *info)
543 {
544 	struct ec_creg_mask_parms *pp = info;
545 	unsigned long cregs[16];
546 
547 	__ctl_store(cregs, 0, 15);
548 	cregs[pp->cr] = (cregs[pp->cr] & pp->andval) | pp->orval;
549 	__ctl_load(cregs, 0, 15);
550 }
551 
552 /*
553  * Set a bit in a control register of all cpus
554  */
555 void smp_ctl_set_bit(int cr, int bit)
556 {
557 	struct ec_creg_mask_parms parms = { 1UL << bit, -1UL, cr };
558 
559 	on_each_cpu(smp_ctl_bit_callback, &parms, 1);
560 }
561 EXPORT_SYMBOL(smp_ctl_set_bit);
562 
563 /*
564  * Clear a bit in a control register of all cpus
565  */
566 void smp_ctl_clear_bit(int cr, int bit)
567 {
568 	struct ec_creg_mask_parms parms = { 0, ~(1UL << bit), cr };
569 
570 	on_each_cpu(smp_ctl_bit_callback, &parms, 1);
571 }
572 EXPORT_SYMBOL(smp_ctl_clear_bit);
573 
574 #ifdef CONFIG_CRASH_DUMP
575 
576 int smp_store_status(int cpu)
577 {
578 	struct pcpu *pcpu = pcpu_devices + cpu;
579 	unsigned long pa;
580 
581 	pa = __pa(&pcpu->lowcore->floating_pt_save_area);
582 	if (__pcpu_sigp_relax(pcpu->address, SIGP_STORE_STATUS_AT_ADDRESS,
583 			      pa) != SIGP_CC_ORDER_CODE_ACCEPTED)
584 		return -EIO;
585 	if (!MACHINE_HAS_VX && !MACHINE_HAS_GS)
586 		return 0;
587 	pa = __pa(pcpu->lowcore->mcesad & MCESA_ORIGIN_MASK);
588 	if (MACHINE_HAS_GS)
589 		pa |= pcpu->lowcore->mcesad & MCESA_LC_MASK;
590 	if (__pcpu_sigp_relax(pcpu->address, SIGP_STORE_ADDITIONAL_STATUS,
591 			      pa) != SIGP_CC_ORDER_CODE_ACCEPTED)
592 		return -EIO;
593 	return 0;
594 }
595 
596 /*
597  * Collect CPU state of the previous, crashed system.
598  * There are four cases:
599  * 1) standard zfcp dump
600  *    condition: OLDMEM_BASE == NULL && ipl_info.type == IPL_TYPE_FCP_DUMP
601  *    The state for all CPUs except the boot CPU needs to be collected
602  *    with sigp stop-and-store-status. The boot CPU state is located in
603  *    the absolute lowcore of the memory stored in the HSA. The zcore code
604  *    will copy the boot CPU state from the HSA.
605  * 2) stand-alone kdump for SCSI (zfcp dump with swapped memory)
606  *    condition: OLDMEM_BASE != NULL && ipl_info.type == IPL_TYPE_FCP_DUMP
607  *    The state for all CPUs except the boot CPU needs to be collected
608  *    with sigp stop-and-store-status. The firmware or the boot-loader
609  *    stored the registers of the boot CPU in the absolute lowcore in the
610  *    memory of the old system.
611  * 3) kdump and the old kernel did not store the CPU state,
612  *    or stand-alone kdump for DASD
613  *    condition: OLDMEM_BASE != NULL && !is_kdump_kernel()
614  *    The state for all CPUs except the boot CPU needs to be collected
615  *    with sigp stop-and-store-status. The kexec code or the boot-loader
616  *    stored the registers of the boot CPU in the memory of the old system.
617  * 4) kdump and the old kernel stored the CPU state
618  *    condition: OLDMEM_BASE != NULL && is_kdump_kernel()
619  *    This case does not exist for s390 anymore, setup_arch explicitly
620  *    deactivates the elfcorehdr= kernel parameter
621  */
622 static __init void smp_save_cpu_vxrs(struct save_area *sa, u16 addr,
623 				     bool is_boot_cpu, unsigned long page)
624 {
625 	__vector128 *vxrs = (__vector128 *) page;
626 
627 	if (is_boot_cpu)
628 		vxrs = boot_cpu_vector_save_area;
629 	else
630 		__pcpu_sigp_relax(addr, SIGP_STORE_ADDITIONAL_STATUS, page);
631 	save_area_add_vxrs(sa, vxrs);
632 }
633 
634 static __init void smp_save_cpu_regs(struct save_area *sa, u16 addr,
635 				     bool is_boot_cpu, unsigned long page)
636 {
637 	void *regs = (void *) page;
638 
639 	if (is_boot_cpu)
640 		copy_oldmem_kernel(regs, (void *) __LC_FPREGS_SAVE_AREA, 512);
641 	else
642 		__pcpu_sigp_relax(addr, SIGP_STORE_STATUS_AT_ADDRESS, page);
643 	save_area_add_regs(sa, regs);
644 }
645 
646 void __init smp_save_dump_cpus(void)
647 {
648 	int addr, boot_cpu_addr, max_cpu_addr;
649 	struct save_area *sa;
650 	unsigned long page;
651 	bool is_boot_cpu;
652 
653 	if (!(OLDMEM_BASE || ipl_info.type == IPL_TYPE_FCP_DUMP))
654 		/* No previous system present, normal boot. */
655 		return;
656 	/* Allocate a page as dumping area for the store status sigps */
657 	page = memblock_phys_alloc_range(PAGE_SIZE, PAGE_SIZE, 0, 1UL << 31);
658 	if (!page)
659 		panic("ERROR: Failed to allocate %lx bytes below %lx\n",
660 		      PAGE_SIZE, 1UL << 31);
661 
662 	/* Set multi-threading state to the previous system. */
663 	pcpu_set_smt(sclp.mtid_prev);
664 	boot_cpu_addr = stap();
665 	max_cpu_addr = SCLP_MAX_CORES << sclp.mtid_prev;
666 	for (addr = 0; addr <= max_cpu_addr; addr++) {
667 		if (__pcpu_sigp_relax(addr, SIGP_SENSE, 0) ==
668 		    SIGP_CC_NOT_OPERATIONAL)
669 			continue;
670 		is_boot_cpu = (addr == boot_cpu_addr);
671 		/* Allocate save area */
672 		sa = save_area_alloc(is_boot_cpu);
673 		if (!sa)
674 			panic("could not allocate memory for save area\n");
675 		if (MACHINE_HAS_VX)
676 			/* Get the vector registers */
677 			smp_save_cpu_vxrs(sa, addr, is_boot_cpu, page);
678 		/*
679 		 * For a zfcp dump OLDMEM_BASE == NULL and the registers
680 		 * of the boot CPU are stored in the HSA. To retrieve
681 		 * these registers an SCLP request is required which is
682 		 * done by drivers/s390/char/zcore.c:init_cpu_info()
683 		 */
684 		if (!is_boot_cpu || OLDMEM_BASE)
685 			/* Get the CPU registers */
686 			smp_save_cpu_regs(sa, addr, is_boot_cpu, page);
687 	}
688 	memblock_free(page, PAGE_SIZE);
689 	diag_dma_ops.diag308_reset();
690 	pcpu_set_smt(0);
691 }
692 #endif /* CONFIG_CRASH_DUMP */
693 
694 void smp_cpu_set_polarization(int cpu, int val)
695 {
696 	pcpu_devices[cpu].polarization = val;
697 }
698 
699 int smp_cpu_get_polarization(int cpu)
700 {
701 	return pcpu_devices[cpu].polarization;
702 }
703 
704 static void __ref smp_get_core_info(struct sclp_core_info *info, int early)
705 {
706 	static int use_sigp_detection;
707 	int address;
708 
709 	if (use_sigp_detection || sclp_get_core_info(info, early)) {
710 		use_sigp_detection = 1;
711 		for (address = 0;
712 		     address < (SCLP_MAX_CORES << smp_cpu_mt_shift);
713 		     address += (1U << smp_cpu_mt_shift)) {
714 			if (__pcpu_sigp_relax(address, SIGP_SENSE, 0) ==
715 			    SIGP_CC_NOT_OPERATIONAL)
716 				continue;
717 			info->core[info->configured].core_id =
718 				address >> smp_cpu_mt_shift;
719 			info->configured++;
720 		}
721 		info->combined = info->configured;
722 	}
723 }
724 
725 static int smp_add_present_cpu(int cpu);
726 
727 static int __smp_rescan_cpus(struct sclp_core_info *info, int sysfs_add)
728 {
729 	struct pcpu *pcpu;
730 	cpumask_t avail;
731 	int cpu, nr, i, j;
732 	u16 address;
733 
734 	nr = 0;
735 	cpumask_xor(&avail, cpu_possible_mask, cpu_present_mask);
736 	cpu = cpumask_first(&avail);
737 	for (i = 0; (i < info->combined) && (cpu < nr_cpu_ids); i++) {
738 		if (sclp.has_core_type && info->core[i].type != boot_core_type)
739 			continue;
740 		address = info->core[i].core_id << smp_cpu_mt_shift;
741 		for (j = 0; j <= smp_cpu_mtid; j++) {
742 			if (pcpu_find_address(cpu_present_mask, address + j))
743 				continue;
744 			pcpu = pcpu_devices + cpu;
745 			pcpu->address = address + j;
746 			pcpu->state =
747 				(cpu >= info->configured*(smp_cpu_mtid + 1)) ?
748 				CPU_STATE_STANDBY : CPU_STATE_CONFIGURED;
749 			smp_cpu_set_polarization(cpu, POLARIZATION_UNKNOWN);
750 			set_cpu_present(cpu, true);
751 			if (sysfs_add && smp_add_present_cpu(cpu) != 0)
752 				set_cpu_present(cpu, false);
753 			else
754 				nr++;
755 			cpu = cpumask_next(cpu, &avail);
756 			if (cpu >= nr_cpu_ids)
757 				break;
758 		}
759 	}
760 	return nr;
761 }
762 
763 void __init smp_detect_cpus(void)
764 {
765 	unsigned int cpu, mtid, c_cpus, s_cpus;
766 	struct sclp_core_info *info;
767 	u16 address;
768 
769 	/* Get CPU information */
770 	info = memblock_alloc(sizeof(*info), 8);
771 	if (!info)
772 		panic("%s: Failed to allocate %zu bytes align=0x%x\n",
773 		      __func__, sizeof(*info), 8);
774 	smp_get_core_info(info, 1);
775 	/* Find boot CPU type */
776 	if (sclp.has_core_type) {
777 		address = stap();
778 		for (cpu = 0; cpu < info->combined; cpu++)
779 			if (info->core[cpu].core_id == address) {
780 				/* The boot cpu dictates the cpu type. */
781 				boot_core_type = info->core[cpu].type;
782 				break;
783 			}
784 		if (cpu >= info->combined)
785 			panic("Could not find boot CPU type");
786 	}
787 
788 	/* Set multi-threading state for the current system */
789 	mtid = boot_core_type ? sclp.mtid : sclp.mtid_cp;
790 	mtid = (mtid < smp_max_threads) ? mtid : smp_max_threads - 1;
791 	pcpu_set_smt(mtid);
792 
793 	/* Print number of CPUs */
794 	c_cpus = s_cpus = 0;
795 	for (cpu = 0; cpu < info->combined; cpu++) {
796 		if (sclp.has_core_type &&
797 		    info->core[cpu].type != boot_core_type)
798 			continue;
799 		if (cpu < info->configured)
800 			c_cpus += smp_cpu_mtid + 1;
801 		else
802 			s_cpus += smp_cpu_mtid + 1;
803 	}
804 	pr_info("%d configured CPUs, %d standby CPUs\n", c_cpus, s_cpus);
805 
806 	/* Add CPUs present at boot */
807 	get_online_cpus();
808 	__smp_rescan_cpus(info, 0);
809 	put_online_cpus();
810 	memblock_free_early((unsigned long)info, sizeof(*info));
811 }
812 
813 static void smp_init_secondary(void)
814 {
815 	int cpu = smp_processor_id();
816 
817 	S390_lowcore.last_update_clock = get_tod_clock();
818 	restore_access_regs(S390_lowcore.access_regs_save_area);
819 	cpu_init();
820 	preempt_disable();
821 	init_cpu_timer();
822 	vtime_init();
823 	pfault_init();
824 	notify_cpu_starting(smp_processor_id());
825 	if (topology_cpu_dedicated(cpu))
826 		set_cpu_flag(CIF_DEDICATED_CPU);
827 	else
828 		clear_cpu_flag(CIF_DEDICATED_CPU);
829 	set_cpu_online(smp_processor_id(), true);
830 	inc_irq_stat(CPU_RST);
831 	local_irq_enable();
832 	cpu_startup_entry(CPUHP_AP_ONLINE_IDLE);
833 }
834 
835 /*
836  *	Activate a secondary processor.
837  */
838 static void __no_sanitize_address smp_start_secondary(void *cpuvoid)
839 {
840 	S390_lowcore.restart_stack = (unsigned long) restart_stack;
841 	S390_lowcore.restart_fn = (unsigned long) do_restart;
842 	S390_lowcore.restart_data = 0;
843 	S390_lowcore.restart_source = -1UL;
844 	__ctl_load(S390_lowcore.cregs_save_area, 0, 15);
845 	__load_psw_mask(PSW_KERNEL_BITS | PSW_MASK_DAT);
846 	CALL_ON_STACK(smp_init_secondary, S390_lowcore.kernel_stack, 0);
847 }
848 
849 /* Upping and downing of CPUs */
850 int __cpu_up(unsigned int cpu, struct task_struct *tidle)
851 {
852 	struct pcpu *pcpu;
853 	int base, i, rc;
854 
855 	pcpu = pcpu_devices + cpu;
856 	if (pcpu->state != CPU_STATE_CONFIGURED)
857 		return -EIO;
858 	base = smp_get_base_cpu(cpu);
859 	for (i = 0; i <= smp_cpu_mtid; i++) {
860 		if (base + i < nr_cpu_ids)
861 			if (cpu_online(base + i))
862 				break;
863 	}
864 	/*
865 	 * If this is the first CPU of the core to get online
866 	 * do an initial CPU reset.
867 	 */
868 	if (i > smp_cpu_mtid &&
869 	    pcpu_sigp_retry(pcpu_devices + base, SIGP_INITIAL_CPU_RESET, 0) !=
870 	    SIGP_CC_ORDER_CODE_ACCEPTED)
871 		return -EIO;
872 
873 	rc = pcpu_alloc_lowcore(pcpu, cpu);
874 	if (rc)
875 		return rc;
876 	pcpu_prepare_secondary(pcpu, cpu);
877 	pcpu_attach_task(pcpu, tidle);
878 	pcpu_start_fn(pcpu, smp_start_secondary, NULL);
879 	/* Wait until cpu puts itself in the online & active maps */
880 	while (!cpu_online(cpu))
881 		cpu_relax();
882 	return 0;
883 }
884 
885 static unsigned int setup_possible_cpus __initdata;
886 
887 static int __init _setup_possible_cpus(char *s)
888 {
889 	get_option(&s, &setup_possible_cpus);
890 	return 0;
891 }
892 early_param("possible_cpus", _setup_possible_cpus);
893 
894 int __cpu_disable(void)
895 {
896 	unsigned long cregs[16];
897 
898 	/* Handle possible pending IPIs */
899 	smp_handle_ext_call();
900 	set_cpu_online(smp_processor_id(), false);
901 	/* Disable pseudo page faults on this cpu. */
902 	pfault_fini();
903 	/* Disable interrupt sources via control register. */
904 	__ctl_store(cregs, 0, 15);
905 	cregs[0]  &= ~0x0000ee70UL;	/* disable all external interrupts */
906 	cregs[6]  &= ~0xff000000UL;	/* disable all I/O interrupts */
907 	cregs[14] &= ~0x1f000000UL;	/* disable most machine checks */
908 	__ctl_load(cregs, 0, 15);
909 	clear_cpu_flag(CIF_NOHZ_DELAY);
910 	return 0;
911 }
912 
913 void __cpu_die(unsigned int cpu)
914 {
915 	struct pcpu *pcpu;
916 
917 	/* Wait until target cpu is down */
918 	pcpu = pcpu_devices + cpu;
919 	while (!pcpu_stopped(pcpu))
920 		cpu_relax();
921 	pcpu_free_lowcore(pcpu);
922 	cpumask_clear_cpu(cpu, mm_cpumask(&init_mm));
923 	cpumask_clear_cpu(cpu, &init_mm.context.cpu_attach_mask);
924 }
925 
926 void __noreturn cpu_die(void)
927 {
928 	idle_task_exit();
929 	__bpon();
930 	pcpu_sigp_retry(pcpu_devices + smp_processor_id(), SIGP_STOP, 0);
931 	for (;;) ;
932 }
933 
934 void __init smp_fill_possible_mask(void)
935 {
936 	unsigned int possible, sclp_max, cpu;
937 
938 	sclp_max = max(sclp.mtid, sclp.mtid_cp) + 1;
939 	sclp_max = min(smp_max_threads, sclp_max);
940 	sclp_max = (sclp.max_cores * sclp_max) ?: nr_cpu_ids;
941 	possible = setup_possible_cpus ?: nr_cpu_ids;
942 	possible = min(possible, sclp_max);
943 	for (cpu = 0; cpu < possible && cpu < nr_cpu_ids; cpu++)
944 		set_cpu_possible(cpu, true);
945 }
946 
947 void __init smp_prepare_cpus(unsigned int max_cpus)
948 {
949 	/* request the 0x1201 emergency signal external interrupt */
950 	if (register_external_irq(EXT_IRQ_EMERGENCY_SIG, do_ext_call_interrupt))
951 		panic("Couldn't request external interrupt 0x1201");
952 	/* request the 0x1202 external call external interrupt */
953 	if (register_external_irq(EXT_IRQ_EXTERNAL_CALL, do_ext_call_interrupt))
954 		panic("Couldn't request external interrupt 0x1202");
955 }
956 
957 void __init smp_prepare_boot_cpu(void)
958 {
959 	struct pcpu *pcpu = pcpu_devices;
960 
961 	WARN_ON(!cpu_present(0) || !cpu_online(0));
962 	pcpu->state = CPU_STATE_CONFIGURED;
963 	pcpu->lowcore = (struct lowcore *)(unsigned long) store_prefix();
964 	S390_lowcore.percpu_offset = __per_cpu_offset[0];
965 	smp_cpu_set_polarization(0, POLARIZATION_UNKNOWN);
966 }
967 
968 void __init smp_cpus_done(unsigned int max_cpus)
969 {
970 }
971 
972 void __init smp_setup_processor_id(void)
973 {
974 	pcpu_devices[0].address = stap();
975 	S390_lowcore.cpu_nr = 0;
976 	S390_lowcore.spinlock_lockval = arch_spin_lockval(0);
977 	S390_lowcore.spinlock_index = 0;
978 }
979 
980 /*
981  * the frequency of the profiling timer can be changed
982  * by writing a multiplier value into /proc/profile.
983  *
984  * usually you want to run this on all CPUs ;)
985  */
986 int setup_profiling_timer(unsigned int multiplier)
987 {
988 	return 0;
989 }
990 
991 static ssize_t cpu_configure_show(struct device *dev,
992 				  struct device_attribute *attr, char *buf)
993 {
994 	ssize_t count;
995 
996 	mutex_lock(&smp_cpu_state_mutex);
997 	count = sprintf(buf, "%d\n", pcpu_devices[dev->id].state);
998 	mutex_unlock(&smp_cpu_state_mutex);
999 	return count;
1000 }
1001 
1002 static ssize_t cpu_configure_store(struct device *dev,
1003 				   struct device_attribute *attr,
1004 				   const char *buf, size_t count)
1005 {
1006 	struct pcpu *pcpu;
1007 	int cpu, val, rc, i;
1008 	char delim;
1009 
1010 	if (sscanf(buf, "%d %c", &val, &delim) != 1)
1011 		return -EINVAL;
1012 	if (val != 0 && val != 1)
1013 		return -EINVAL;
1014 	get_online_cpus();
1015 	mutex_lock(&smp_cpu_state_mutex);
1016 	rc = -EBUSY;
1017 	/* disallow configuration changes of online cpus and cpu 0 */
1018 	cpu = dev->id;
1019 	cpu = smp_get_base_cpu(cpu);
1020 	if (cpu == 0)
1021 		goto out;
1022 	for (i = 0; i <= smp_cpu_mtid; i++)
1023 		if (cpu_online(cpu + i))
1024 			goto out;
1025 	pcpu = pcpu_devices + cpu;
1026 	rc = 0;
1027 	switch (val) {
1028 	case 0:
1029 		if (pcpu->state != CPU_STATE_CONFIGURED)
1030 			break;
1031 		rc = sclp_core_deconfigure(pcpu->address >> smp_cpu_mt_shift);
1032 		if (rc)
1033 			break;
1034 		for (i = 0; i <= smp_cpu_mtid; i++) {
1035 			if (cpu + i >= nr_cpu_ids || !cpu_present(cpu + i))
1036 				continue;
1037 			pcpu[i].state = CPU_STATE_STANDBY;
1038 			smp_cpu_set_polarization(cpu + i,
1039 						 POLARIZATION_UNKNOWN);
1040 		}
1041 		topology_expect_change();
1042 		break;
1043 	case 1:
1044 		if (pcpu->state != CPU_STATE_STANDBY)
1045 			break;
1046 		rc = sclp_core_configure(pcpu->address >> smp_cpu_mt_shift);
1047 		if (rc)
1048 			break;
1049 		for (i = 0; i <= smp_cpu_mtid; i++) {
1050 			if (cpu + i >= nr_cpu_ids || !cpu_present(cpu + i))
1051 				continue;
1052 			pcpu[i].state = CPU_STATE_CONFIGURED;
1053 			smp_cpu_set_polarization(cpu + i,
1054 						 POLARIZATION_UNKNOWN);
1055 		}
1056 		topology_expect_change();
1057 		break;
1058 	default:
1059 		break;
1060 	}
1061 out:
1062 	mutex_unlock(&smp_cpu_state_mutex);
1063 	put_online_cpus();
1064 	return rc ? rc : count;
1065 }
1066 static DEVICE_ATTR(configure, 0644, cpu_configure_show, cpu_configure_store);
1067 
1068 static ssize_t show_cpu_address(struct device *dev,
1069 				struct device_attribute *attr, char *buf)
1070 {
1071 	return sprintf(buf, "%d\n", pcpu_devices[dev->id].address);
1072 }
1073 static DEVICE_ATTR(address, 0444, show_cpu_address, NULL);
1074 
1075 static struct attribute *cpu_common_attrs[] = {
1076 	&dev_attr_configure.attr,
1077 	&dev_attr_address.attr,
1078 	NULL,
1079 };
1080 
1081 static struct attribute_group cpu_common_attr_group = {
1082 	.attrs = cpu_common_attrs,
1083 };
1084 
1085 static struct attribute *cpu_online_attrs[] = {
1086 	&dev_attr_idle_count.attr,
1087 	&dev_attr_idle_time_us.attr,
1088 	NULL,
1089 };
1090 
1091 static struct attribute_group cpu_online_attr_group = {
1092 	.attrs = cpu_online_attrs,
1093 };
1094 
1095 static int smp_cpu_online(unsigned int cpu)
1096 {
1097 	struct device *s = &per_cpu(cpu_device, cpu)->dev;
1098 
1099 	return sysfs_create_group(&s->kobj, &cpu_online_attr_group);
1100 }
1101 static int smp_cpu_pre_down(unsigned int cpu)
1102 {
1103 	struct device *s = &per_cpu(cpu_device, cpu)->dev;
1104 
1105 	sysfs_remove_group(&s->kobj, &cpu_online_attr_group);
1106 	return 0;
1107 }
1108 
1109 static int smp_add_present_cpu(int cpu)
1110 {
1111 	struct device *s;
1112 	struct cpu *c;
1113 	int rc;
1114 
1115 	c = kzalloc(sizeof(*c), GFP_KERNEL);
1116 	if (!c)
1117 		return -ENOMEM;
1118 	per_cpu(cpu_device, cpu) = c;
1119 	s = &c->dev;
1120 	c->hotpluggable = 1;
1121 	rc = register_cpu(c, cpu);
1122 	if (rc)
1123 		goto out;
1124 	rc = sysfs_create_group(&s->kobj, &cpu_common_attr_group);
1125 	if (rc)
1126 		goto out_cpu;
1127 	rc = topology_cpu_init(c);
1128 	if (rc)
1129 		goto out_topology;
1130 	return 0;
1131 
1132 out_topology:
1133 	sysfs_remove_group(&s->kobj, &cpu_common_attr_group);
1134 out_cpu:
1135 	unregister_cpu(c);
1136 out:
1137 	return rc;
1138 }
1139 
1140 int __ref smp_rescan_cpus(void)
1141 {
1142 	struct sclp_core_info *info;
1143 	int nr;
1144 
1145 	info = kzalloc(sizeof(*info), GFP_KERNEL);
1146 	if (!info)
1147 		return -ENOMEM;
1148 	smp_get_core_info(info, 0);
1149 	get_online_cpus();
1150 	mutex_lock(&smp_cpu_state_mutex);
1151 	nr = __smp_rescan_cpus(info, 1);
1152 	mutex_unlock(&smp_cpu_state_mutex);
1153 	put_online_cpus();
1154 	kfree(info);
1155 	if (nr)
1156 		topology_schedule_update();
1157 	return 0;
1158 }
1159 
1160 static ssize_t __ref rescan_store(struct device *dev,
1161 				  struct device_attribute *attr,
1162 				  const char *buf,
1163 				  size_t count)
1164 {
1165 	int rc;
1166 
1167 	rc = lock_device_hotplug_sysfs();
1168 	if (rc)
1169 		return rc;
1170 	rc = smp_rescan_cpus();
1171 	unlock_device_hotplug();
1172 	return rc ? rc : count;
1173 }
1174 static DEVICE_ATTR_WO(rescan);
1175 
1176 static int __init s390_smp_init(void)
1177 {
1178 	int cpu, rc = 0;
1179 
1180 	rc = device_create_file(cpu_subsys.dev_root, &dev_attr_rescan);
1181 	if (rc)
1182 		return rc;
1183 	for_each_present_cpu(cpu) {
1184 		rc = smp_add_present_cpu(cpu);
1185 		if (rc)
1186 			goto out;
1187 	}
1188 
1189 	rc = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "s390/smp:online",
1190 			       smp_cpu_online, smp_cpu_pre_down);
1191 	rc = rc <= 0 ? rc : 0;
1192 out:
1193 	return rc;
1194 }
1195 subsys_initcall(s390_smp_init);
1196