xref: /openbmc/linux/arch/s390/kernel/smp.c (revision 293d5b43)
1 /*
2  *  SMP related functions
3  *
4  *    Copyright IBM Corp. 1999, 2012
5  *    Author(s): Denis Joseph Barrow,
6  *		 Martin Schwidefsky <schwidefsky@de.ibm.com>,
7  *		 Heiko Carstens <heiko.carstens@de.ibm.com>,
8  *
9  *  based on other smp stuff by
10  *    (c) 1995 Alan Cox, CymruNET Ltd  <alan@cymru.net>
11  *    (c) 1998 Ingo Molnar
12  *
13  * The code outside of smp.c uses logical cpu numbers, only smp.c does
14  * the translation of logical to physical cpu ids. All new code that
15  * operates on physical cpu numbers needs to go into smp.c.
16  */
17 
18 #define KMSG_COMPONENT "cpu"
19 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
20 
21 #include <linux/workqueue.h>
22 #include <linux/module.h>
23 #include <linux/init.h>
24 #include <linux/mm.h>
25 #include <linux/err.h>
26 #include <linux/spinlock.h>
27 #include <linux/kernel_stat.h>
28 #include <linux/delay.h>
29 #include <linux/interrupt.h>
30 #include <linux/irqflags.h>
31 #include <linux/cpu.h>
32 #include <linux/slab.h>
33 #include <linux/crash_dump.h>
34 #include <linux/memblock.h>
35 #include <asm/asm-offsets.h>
36 #include <asm/diag.h>
37 #include <asm/switch_to.h>
38 #include <asm/facility.h>
39 #include <asm/ipl.h>
40 #include <asm/setup.h>
41 #include <asm/irq.h>
42 #include <asm/tlbflush.h>
43 #include <asm/vtimer.h>
44 #include <asm/lowcore.h>
45 #include <asm/sclp.h>
46 #include <asm/vdso.h>
47 #include <asm/debug.h>
48 #include <asm/os_info.h>
49 #include <asm/sigp.h>
50 #include <asm/idle.h>
51 #include "entry.h"
52 
53 enum {
54 	ec_schedule = 0,
55 	ec_call_function_single,
56 	ec_stop_cpu,
57 };
58 
59 enum {
60 	CPU_STATE_STANDBY,
61 	CPU_STATE_CONFIGURED,
62 };
63 
64 static DEFINE_PER_CPU(struct cpu *, cpu_device);
65 
66 struct pcpu {
67 	struct lowcore *lowcore;	/* lowcore page(s) for the cpu */
68 	unsigned long ec_mask;		/* bit mask for ec_xxx functions */
69 	unsigned long ec_clk;		/* sigp timestamp for ec_xxx */
70 	signed char state;		/* physical cpu state */
71 	signed char polarization;	/* physical polarization */
72 	u16 address;			/* physical cpu address */
73 };
74 
75 static u8 boot_core_type;
76 static struct pcpu pcpu_devices[NR_CPUS];
77 
78 unsigned int smp_cpu_mt_shift;
79 EXPORT_SYMBOL(smp_cpu_mt_shift);
80 
81 unsigned int smp_cpu_mtid;
82 EXPORT_SYMBOL(smp_cpu_mtid);
83 
84 #ifdef CONFIG_CRASH_DUMP
85 __vector128 __initdata boot_cpu_vector_save_area[__NUM_VXRS];
86 #endif
87 
88 static unsigned int smp_max_threads __initdata = -1U;
89 
90 static int __init early_nosmt(char *s)
91 {
92 	smp_max_threads = 1;
93 	return 0;
94 }
95 early_param("nosmt", early_nosmt);
96 
97 static int __init early_smt(char *s)
98 {
99 	get_option(&s, &smp_max_threads);
100 	return 0;
101 }
102 early_param("smt", early_smt);
103 
104 /*
105  * The smp_cpu_state_mutex must be held when changing the state or polarization
106  * member of a pcpu data structure within the pcpu_devices arreay.
107  */
108 DEFINE_MUTEX(smp_cpu_state_mutex);
109 
110 /*
111  * Signal processor helper functions.
112  */
113 static inline int __pcpu_sigp_relax(u16 addr, u8 order, unsigned long parm)
114 {
115 	int cc;
116 
117 	while (1) {
118 		cc = __pcpu_sigp(addr, order, parm, NULL);
119 		if (cc != SIGP_CC_BUSY)
120 			return cc;
121 		cpu_relax();
122 	}
123 }
124 
125 static int pcpu_sigp_retry(struct pcpu *pcpu, u8 order, u32 parm)
126 {
127 	int cc, retry;
128 
129 	for (retry = 0; ; retry++) {
130 		cc = __pcpu_sigp(pcpu->address, order, parm, NULL);
131 		if (cc != SIGP_CC_BUSY)
132 			break;
133 		if (retry >= 3)
134 			udelay(10);
135 	}
136 	return cc;
137 }
138 
139 static inline int pcpu_stopped(struct pcpu *pcpu)
140 {
141 	u32 uninitialized_var(status);
142 
143 	if (__pcpu_sigp(pcpu->address, SIGP_SENSE,
144 			0, &status) != SIGP_CC_STATUS_STORED)
145 		return 0;
146 	return !!(status & (SIGP_STATUS_CHECK_STOP|SIGP_STATUS_STOPPED));
147 }
148 
149 static inline int pcpu_running(struct pcpu *pcpu)
150 {
151 	if (__pcpu_sigp(pcpu->address, SIGP_SENSE_RUNNING,
152 			0, NULL) != SIGP_CC_STATUS_STORED)
153 		return 1;
154 	/* Status stored condition code is equivalent to cpu not running. */
155 	return 0;
156 }
157 
158 /*
159  * Find struct pcpu by cpu address.
160  */
161 static struct pcpu *pcpu_find_address(const struct cpumask *mask, u16 address)
162 {
163 	int cpu;
164 
165 	for_each_cpu(cpu, mask)
166 		if (pcpu_devices[cpu].address == address)
167 			return pcpu_devices + cpu;
168 	return NULL;
169 }
170 
171 static void pcpu_ec_call(struct pcpu *pcpu, int ec_bit)
172 {
173 	int order;
174 
175 	if (test_and_set_bit(ec_bit, &pcpu->ec_mask))
176 		return;
177 	order = pcpu_running(pcpu) ? SIGP_EXTERNAL_CALL : SIGP_EMERGENCY_SIGNAL;
178 	pcpu->ec_clk = get_tod_clock_fast();
179 	pcpu_sigp_retry(pcpu, order, 0);
180 }
181 
182 #define ASYNC_FRAME_OFFSET (ASYNC_SIZE - STACK_FRAME_OVERHEAD - __PT_SIZE)
183 #define PANIC_FRAME_OFFSET (PAGE_SIZE - STACK_FRAME_OVERHEAD - __PT_SIZE)
184 
185 static int pcpu_alloc_lowcore(struct pcpu *pcpu, int cpu)
186 {
187 	unsigned long async_stack, panic_stack;
188 	struct lowcore *lc;
189 
190 	if (pcpu != &pcpu_devices[0]) {
191 		pcpu->lowcore =	(struct lowcore *)
192 			__get_free_pages(GFP_KERNEL | GFP_DMA, LC_ORDER);
193 		async_stack = __get_free_pages(GFP_KERNEL, ASYNC_ORDER);
194 		panic_stack = __get_free_page(GFP_KERNEL);
195 		if (!pcpu->lowcore || !panic_stack || !async_stack)
196 			goto out;
197 	} else {
198 		async_stack = pcpu->lowcore->async_stack - ASYNC_FRAME_OFFSET;
199 		panic_stack = pcpu->lowcore->panic_stack - PANIC_FRAME_OFFSET;
200 	}
201 	lc = pcpu->lowcore;
202 	memcpy(lc, &S390_lowcore, 512);
203 	memset((char *) lc + 512, 0, sizeof(*lc) - 512);
204 	lc->async_stack = async_stack + ASYNC_FRAME_OFFSET;
205 	lc->panic_stack = panic_stack + PANIC_FRAME_OFFSET;
206 	lc->cpu_nr = cpu;
207 	lc->spinlock_lockval = arch_spin_lockval(cpu);
208 	if (MACHINE_HAS_VX)
209 		lc->vector_save_area_addr =
210 			(unsigned long) &lc->vector_save_area;
211 	if (vdso_alloc_per_cpu(lc))
212 		goto out;
213 	lowcore_ptr[cpu] = lc;
214 	pcpu_sigp_retry(pcpu, SIGP_SET_PREFIX, (u32)(unsigned long) lc);
215 	return 0;
216 out:
217 	if (pcpu != &pcpu_devices[0]) {
218 		free_page(panic_stack);
219 		free_pages(async_stack, ASYNC_ORDER);
220 		free_pages((unsigned long) pcpu->lowcore, LC_ORDER);
221 	}
222 	return -ENOMEM;
223 }
224 
225 #ifdef CONFIG_HOTPLUG_CPU
226 
227 static void pcpu_free_lowcore(struct pcpu *pcpu)
228 {
229 	pcpu_sigp_retry(pcpu, SIGP_SET_PREFIX, 0);
230 	lowcore_ptr[pcpu - pcpu_devices] = NULL;
231 	vdso_free_per_cpu(pcpu->lowcore);
232 	if (pcpu == &pcpu_devices[0])
233 		return;
234 	free_page(pcpu->lowcore->panic_stack-PANIC_FRAME_OFFSET);
235 	free_pages(pcpu->lowcore->async_stack-ASYNC_FRAME_OFFSET, ASYNC_ORDER);
236 	free_pages((unsigned long) pcpu->lowcore, LC_ORDER);
237 }
238 
239 #endif /* CONFIG_HOTPLUG_CPU */
240 
241 static void pcpu_prepare_secondary(struct pcpu *pcpu, int cpu)
242 {
243 	struct lowcore *lc = pcpu->lowcore;
244 
245 	cpumask_set_cpu(cpu, &init_mm.context.cpu_attach_mask);
246 	cpumask_set_cpu(cpu, mm_cpumask(&init_mm));
247 	lc->cpu_nr = cpu;
248 	lc->spinlock_lockval = arch_spin_lockval(cpu);
249 	lc->percpu_offset = __per_cpu_offset[cpu];
250 	lc->kernel_asce = S390_lowcore.kernel_asce;
251 	lc->machine_flags = S390_lowcore.machine_flags;
252 	lc->user_timer = lc->system_timer = lc->steal_timer = 0;
253 	__ctl_store(lc->cregs_save_area, 0, 15);
254 	save_access_regs((unsigned int *) lc->access_regs_save_area);
255 	memcpy(lc->stfle_fac_list, S390_lowcore.stfle_fac_list,
256 	       MAX_FACILITY_BIT/8);
257 }
258 
259 static void pcpu_attach_task(struct pcpu *pcpu, struct task_struct *tsk)
260 {
261 	struct lowcore *lc = pcpu->lowcore;
262 	struct thread_info *ti = task_thread_info(tsk);
263 
264 	lc->kernel_stack = (unsigned long) task_stack_page(tsk)
265 		+ THREAD_SIZE - STACK_FRAME_OVERHEAD - sizeof(struct pt_regs);
266 	lc->thread_info = (unsigned long) task_thread_info(tsk);
267 	lc->current_task = (unsigned long) tsk;
268 	lc->lpp = LPP_MAGIC;
269 	lc->current_pid = tsk->pid;
270 	lc->user_timer = ti->user_timer;
271 	lc->system_timer = ti->system_timer;
272 	lc->steal_timer = 0;
273 }
274 
275 static void pcpu_start_fn(struct pcpu *pcpu, void (*func)(void *), void *data)
276 {
277 	struct lowcore *lc = pcpu->lowcore;
278 
279 	lc->restart_stack = lc->kernel_stack;
280 	lc->restart_fn = (unsigned long) func;
281 	lc->restart_data = (unsigned long) data;
282 	lc->restart_source = -1UL;
283 	pcpu_sigp_retry(pcpu, SIGP_RESTART, 0);
284 }
285 
286 /*
287  * Call function via PSW restart on pcpu and stop the current cpu.
288  */
289 static void pcpu_delegate(struct pcpu *pcpu, void (*func)(void *),
290 			  void *data, unsigned long stack)
291 {
292 	struct lowcore *lc = lowcore_ptr[pcpu - pcpu_devices];
293 	unsigned long source_cpu = stap();
294 
295 	__load_psw_mask(PSW_KERNEL_BITS);
296 	if (pcpu->address == source_cpu)
297 		func(data);	/* should not return */
298 	/* Stop target cpu (if func returns this stops the current cpu). */
299 	pcpu_sigp_retry(pcpu, SIGP_STOP, 0);
300 	/* Restart func on the target cpu and stop the current cpu. */
301 	mem_assign_absolute(lc->restart_stack, stack);
302 	mem_assign_absolute(lc->restart_fn, (unsigned long) func);
303 	mem_assign_absolute(lc->restart_data, (unsigned long) data);
304 	mem_assign_absolute(lc->restart_source, source_cpu);
305 	asm volatile(
306 		"0:	sigp	0,%0,%2	# sigp restart to target cpu\n"
307 		"	brc	2,0b	# busy, try again\n"
308 		"1:	sigp	0,%1,%3	# sigp stop to current cpu\n"
309 		"	brc	2,1b	# busy, try again\n"
310 		: : "d" (pcpu->address), "d" (source_cpu),
311 		    "K" (SIGP_RESTART), "K" (SIGP_STOP)
312 		: "0", "1", "cc");
313 	for (;;) ;
314 }
315 
316 /*
317  * Enable additional logical cpus for multi-threading.
318  */
319 static int pcpu_set_smt(unsigned int mtid)
320 {
321 	int cc;
322 
323 	if (smp_cpu_mtid == mtid)
324 		return 0;
325 	cc = __pcpu_sigp(0, SIGP_SET_MULTI_THREADING, mtid, NULL);
326 	if (cc == 0) {
327 		smp_cpu_mtid = mtid;
328 		smp_cpu_mt_shift = 0;
329 		while (smp_cpu_mtid >= (1U << smp_cpu_mt_shift))
330 			smp_cpu_mt_shift++;
331 		pcpu_devices[0].address = stap();
332 	}
333 	return cc;
334 }
335 
336 /*
337  * Call function on an online CPU.
338  */
339 void smp_call_online_cpu(void (*func)(void *), void *data)
340 {
341 	struct pcpu *pcpu;
342 
343 	/* Use the current cpu if it is online. */
344 	pcpu = pcpu_find_address(cpu_online_mask, stap());
345 	if (!pcpu)
346 		/* Use the first online cpu. */
347 		pcpu = pcpu_devices + cpumask_first(cpu_online_mask);
348 	pcpu_delegate(pcpu, func, data, (unsigned long) restart_stack);
349 }
350 
351 /*
352  * Call function on the ipl CPU.
353  */
354 void smp_call_ipl_cpu(void (*func)(void *), void *data)
355 {
356 	pcpu_delegate(&pcpu_devices[0], func, data,
357 		      pcpu_devices->lowcore->panic_stack -
358 		      PANIC_FRAME_OFFSET + PAGE_SIZE);
359 }
360 
361 int smp_find_processor_id(u16 address)
362 {
363 	int cpu;
364 
365 	for_each_present_cpu(cpu)
366 		if (pcpu_devices[cpu].address == address)
367 			return cpu;
368 	return -1;
369 }
370 
371 int smp_vcpu_scheduled(int cpu)
372 {
373 	return pcpu_running(pcpu_devices + cpu);
374 }
375 
376 void smp_yield_cpu(int cpu)
377 {
378 	if (MACHINE_HAS_DIAG9C) {
379 		diag_stat_inc_norecursion(DIAG_STAT_X09C);
380 		asm volatile("diag %0,0,0x9c"
381 			     : : "d" (pcpu_devices[cpu].address));
382 	} else if (MACHINE_HAS_DIAG44) {
383 		diag_stat_inc_norecursion(DIAG_STAT_X044);
384 		asm volatile("diag 0,0,0x44");
385 	}
386 }
387 
388 /*
389  * Send cpus emergency shutdown signal. This gives the cpus the
390  * opportunity to complete outstanding interrupts.
391  */
392 static void smp_emergency_stop(cpumask_t *cpumask)
393 {
394 	u64 end;
395 	int cpu;
396 
397 	end = get_tod_clock() + (1000000UL << 12);
398 	for_each_cpu(cpu, cpumask) {
399 		struct pcpu *pcpu = pcpu_devices + cpu;
400 		set_bit(ec_stop_cpu, &pcpu->ec_mask);
401 		while (__pcpu_sigp(pcpu->address, SIGP_EMERGENCY_SIGNAL,
402 				   0, NULL) == SIGP_CC_BUSY &&
403 		       get_tod_clock() < end)
404 			cpu_relax();
405 	}
406 	while (get_tod_clock() < end) {
407 		for_each_cpu(cpu, cpumask)
408 			if (pcpu_stopped(pcpu_devices + cpu))
409 				cpumask_clear_cpu(cpu, cpumask);
410 		if (cpumask_empty(cpumask))
411 			break;
412 		cpu_relax();
413 	}
414 }
415 
416 /*
417  * Stop all cpus but the current one.
418  */
419 void smp_send_stop(void)
420 {
421 	cpumask_t cpumask;
422 	int cpu;
423 
424 	/* Disable all interrupts/machine checks */
425 	__load_psw_mask(PSW_KERNEL_BITS | PSW_MASK_DAT);
426 	trace_hardirqs_off();
427 
428 	debug_set_critical();
429 	cpumask_copy(&cpumask, cpu_online_mask);
430 	cpumask_clear_cpu(smp_processor_id(), &cpumask);
431 
432 	if (oops_in_progress)
433 		smp_emergency_stop(&cpumask);
434 
435 	/* stop all processors */
436 	for_each_cpu(cpu, &cpumask) {
437 		struct pcpu *pcpu = pcpu_devices + cpu;
438 		pcpu_sigp_retry(pcpu, SIGP_STOP, 0);
439 		while (!pcpu_stopped(pcpu))
440 			cpu_relax();
441 	}
442 }
443 
444 /*
445  * This is the main routine where commands issued by other
446  * cpus are handled.
447  */
448 static void smp_handle_ext_call(void)
449 {
450 	unsigned long bits;
451 
452 	/* handle bit signal external calls */
453 	bits = xchg(&pcpu_devices[smp_processor_id()].ec_mask, 0);
454 	if (test_bit(ec_stop_cpu, &bits))
455 		smp_stop_cpu();
456 	if (test_bit(ec_schedule, &bits))
457 		scheduler_ipi();
458 	if (test_bit(ec_call_function_single, &bits))
459 		generic_smp_call_function_single_interrupt();
460 }
461 
462 static void do_ext_call_interrupt(struct ext_code ext_code,
463 				  unsigned int param32, unsigned long param64)
464 {
465 	inc_irq_stat(ext_code.code == 0x1202 ? IRQEXT_EXC : IRQEXT_EMS);
466 	smp_handle_ext_call();
467 }
468 
469 void arch_send_call_function_ipi_mask(const struct cpumask *mask)
470 {
471 	int cpu;
472 
473 	for_each_cpu(cpu, mask)
474 		pcpu_ec_call(pcpu_devices + cpu, ec_call_function_single);
475 }
476 
477 void arch_send_call_function_single_ipi(int cpu)
478 {
479 	pcpu_ec_call(pcpu_devices + cpu, ec_call_function_single);
480 }
481 
482 /*
483  * this function sends a 'reschedule' IPI to another CPU.
484  * it goes straight through and wastes no time serializing
485  * anything. Worst case is that we lose a reschedule ...
486  */
487 void smp_send_reschedule(int cpu)
488 {
489 	pcpu_ec_call(pcpu_devices + cpu, ec_schedule);
490 }
491 
492 /*
493  * parameter area for the set/clear control bit callbacks
494  */
495 struct ec_creg_mask_parms {
496 	unsigned long orval;
497 	unsigned long andval;
498 	int cr;
499 };
500 
501 /*
502  * callback for setting/clearing control bits
503  */
504 static void smp_ctl_bit_callback(void *info)
505 {
506 	struct ec_creg_mask_parms *pp = info;
507 	unsigned long cregs[16];
508 
509 	__ctl_store(cregs, 0, 15);
510 	cregs[pp->cr] = (cregs[pp->cr] & pp->andval) | pp->orval;
511 	__ctl_load(cregs, 0, 15);
512 }
513 
514 /*
515  * Set a bit in a control register of all cpus
516  */
517 void smp_ctl_set_bit(int cr, int bit)
518 {
519 	struct ec_creg_mask_parms parms = { 1UL << bit, -1UL, cr };
520 
521 	on_each_cpu(smp_ctl_bit_callback, &parms, 1);
522 }
523 EXPORT_SYMBOL(smp_ctl_set_bit);
524 
525 /*
526  * Clear a bit in a control register of all cpus
527  */
528 void smp_ctl_clear_bit(int cr, int bit)
529 {
530 	struct ec_creg_mask_parms parms = { 0, ~(1UL << bit), cr };
531 
532 	on_each_cpu(smp_ctl_bit_callback, &parms, 1);
533 }
534 EXPORT_SYMBOL(smp_ctl_clear_bit);
535 
536 #ifdef CONFIG_CRASH_DUMP
537 
538 int smp_store_status(int cpu)
539 {
540 	struct pcpu *pcpu = pcpu_devices + cpu;
541 	unsigned long pa;
542 
543 	pa = __pa(&pcpu->lowcore->floating_pt_save_area);
544 	if (__pcpu_sigp_relax(pcpu->address, SIGP_STORE_STATUS_AT_ADDRESS,
545 			      pa) != SIGP_CC_ORDER_CODE_ACCEPTED)
546 		return -EIO;
547 	if (!MACHINE_HAS_VX)
548 		return 0;
549 	pa = __pa(pcpu->lowcore->vector_save_area_addr);
550 	if (__pcpu_sigp_relax(pcpu->address, SIGP_STORE_ADDITIONAL_STATUS,
551 			      pa) != SIGP_CC_ORDER_CODE_ACCEPTED)
552 		return -EIO;
553 	return 0;
554 }
555 
556 /*
557  * Collect CPU state of the previous, crashed system.
558  * There are four cases:
559  * 1) standard zfcp dump
560  *    condition: OLDMEM_BASE == NULL && ipl_info.type == IPL_TYPE_FCP_DUMP
561  *    The state for all CPUs except the boot CPU needs to be collected
562  *    with sigp stop-and-store-status. The boot CPU state is located in
563  *    the absolute lowcore of the memory stored in the HSA. The zcore code
564  *    will copy the boot CPU state from the HSA.
565  * 2) stand-alone kdump for SCSI (zfcp dump with swapped memory)
566  *    condition: OLDMEM_BASE != NULL && ipl_info.type == IPL_TYPE_FCP_DUMP
567  *    The state for all CPUs except the boot CPU needs to be collected
568  *    with sigp stop-and-store-status. The firmware or the boot-loader
569  *    stored the registers of the boot CPU in the absolute lowcore in the
570  *    memory of the old system.
571  * 3) kdump and the old kernel did not store the CPU state,
572  *    or stand-alone kdump for DASD
573  *    condition: OLDMEM_BASE != NULL && !is_kdump_kernel()
574  *    The state for all CPUs except the boot CPU needs to be collected
575  *    with sigp stop-and-store-status. The kexec code or the boot-loader
576  *    stored the registers of the boot CPU in the memory of the old system.
577  * 4) kdump and the old kernel stored the CPU state
578  *    condition: OLDMEM_BASE != NULL && is_kdump_kernel()
579  *    This case does not exist for s390 anymore, setup_arch explicitly
580  *    deactivates the elfcorehdr= kernel parameter
581  */
582 static __init void smp_save_cpu_vxrs(struct save_area *sa, u16 addr,
583 				     bool is_boot_cpu, unsigned long page)
584 {
585 	__vector128 *vxrs = (__vector128 *) page;
586 
587 	if (is_boot_cpu)
588 		vxrs = boot_cpu_vector_save_area;
589 	else
590 		__pcpu_sigp_relax(addr, SIGP_STORE_ADDITIONAL_STATUS, page);
591 	save_area_add_vxrs(sa, vxrs);
592 }
593 
594 static __init void smp_save_cpu_regs(struct save_area *sa, u16 addr,
595 				     bool is_boot_cpu, unsigned long page)
596 {
597 	void *regs = (void *) page;
598 
599 	if (is_boot_cpu)
600 		copy_oldmem_kernel(regs, (void *) __LC_FPREGS_SAVE_AREA, 512);
601 	else
602 		__pcpu_sigp_relax(addr, SIGP_STORE_STATUS_AT_ADDRESS, page);
603 	save_area_add_regs(sa, regs);
604 }
605 
606 void __init smp_save_dump_cpus(void)
607 {
608 	int addr, boot_cpu_addr, max_cpu_addr;
609 	struct save_area *sa;
610 	unsigned long page;
611 	bool is_boot_cpu;
612 
613 	if (!(OLDMEM_BASE || ipl_info.type == IPL_TYPE_FCP_DUMP))
614 		/* No previous system present, normal boot. */
615 		return;
616 	/* Allocate a page as dumping area for the store status sigps */
617 	page = memblock_alloc_base(PAGE_SIZE, PAGE_SIZE, 1UL << 31);
618 	/* Set multi-threading state to the previous system. */
619 	pcpu_set_smt(sclp.mtid_prev);
620 	boot_cpu_addr = stap();
621 	max_cpu_addr = SCLP_MAX_CORES << sclp.mtid_prev;
622 	for (addr = 0; addr <= max_cpu_addr; addr++) {
623 		if (__pcpu_sigp_relax(addr, SIGP_SENSE, 0) ==
624 		    SIGP_CC_NOT_OPERATIONAL)
625 			continue;
626 		is_boot_cpu = (addr == boot_cpu_addr);
627 		/* Allocate save area */
628 		sa = save_area_alloc(is_boot_cpu);
629 		if (!sa)
630 			panic("could not allocate memory for save area\n");
631 		if (MACHINE_HAS_VX)
632 			/* Get the vector registers */
633 			smp_save_cpu_vxrs(sa, addr, is_boot_cpu, page);
634 		/*
635 		 * For a zfcp dump OLDMEM_BASE == NULL and the registers
636 		 * of the boot CPU are stored in the HSA. To retrieve
637 		 * these registers an SCLP request is required which is
638 		 * done by drivers/s390/char/zcore.c:init_cpu_info()
639 		 */
640 		if (!is_boot_cpu || OLDMEM_BASE)
641 			/* Get the CPU registers */
642 			smp_save_cpu_regs(sa, addr, is_boot_cpu, page);
643 	}
644 	memblock_free(page, PAGE_SIZE);
645 	diag308_reset();
646 	pcpu_set_smt(0);
647 }
648 #endif /* CONFIG_CRASH_DUMP */
649 
650 void smp_cpu_set_polarization(int cpu, int val)
651 {
652 	pcpu_devices[cpu].polarization = val;
653 }
654 
655 int smp_cpu_get_polarization(int cpu)
656 {
657 	return pcpu_devices[cpu].polarization;
658 }
659 
660 static struct sclp_core_info *smp_get_core_info(void)
661 {
662 	static int use_sigp_detection;
663 	struct sclp_core_info *info;
664 	int address;
665 
666 	info = kzalloc(sizeof(*info), GFP_KERNEL);
667 	if (info && (use_sigp_detection || sclp_get_core_info(info))) {
668 		use_sigp_detection = 1;
669 		for (address = 0;
670 		     address < (SCLP_MAX_CORES << smp_cpu_mt_shift);
671 		     address += (1U << smp_cpu_mt_shift)) {
672 			if (__pcpu_sigp_relax(address, SIGP_SENSE, 0) ==
673 			    SIGP_CC_NOT_OPERATIONAL)
674 				continue;
675 			info->core[info->configured].core_id =
676 				address >> smp_cpu_mt_shift;
677 			info->configured++;
678 		}
679 		info->combined = info->configured;
680 	}
681 	return info;
682 }
683 
684 static int smp_add_present_cpu(int cpu);
685 
686 static int __smp_rescan_cpus(struct sclp_core_info *info, int sysfs_add)
687 {
688 	struct pcpu *pcpu;
689 	cpumask_t avail;
690 	int cpu, nr, i, j;
691 	u16 address;
692 
693 	nr = 0;
694 	cpumask_xor(&avail, cpu_possible_mask, cpu_present_mask);
695 	cpu = cpumask_first(&avail);
696 	for (i = 0; (i < info->combined) && (cpu < nr_cpu_ids); i++) {
697 		if (sclp.has_core_type && info->core[i].type != boot_core_type)
698 			continue;
699 		address = info->core[i].core_id << smp_cpu_mt_shift;
700 		for (j = 0; j <= smp_cpu_mtid; j++) {
701 			if (pcpu_find_address(cpu_present_mask, address + j))
702 				continue;
703 			pcpu = pcpu_devices + cpu;
704 			pcpu->address = address + j;
705 			pcpu->state =
706 				(cpu >= info->configured*(smp_cpu_mtid + 1)) ?
707 				CPU_STATE_STANDBY : CPU_STATE_CONFIGURED;
708 			smp_cpu_set_polarization(cpu, POLARIZATION_UNKNOWN);
709 			set_cpu_present(cpu, true);
710 			if (sysfs_add && smp_add_present_cpu(cpu) != 0)
711 				set_cpu_present(cpu, false);
712 			else
713 				nr++;
714 			cpu = cpumask_next(cpu, &avail);
715 			if (cpu >= nr_cpu_ids)
716 				break;
717 		}
718 	}
719 	return nr;
720 }
721 
722 static void __init smp_detect_cpus(void)
723 {
724 	unsigned int cpu, mtid, c_cpus, s_cpus;
725 	struct sclp_core_info *info;
726 	u16 address;
727 
728 	/* Get CPU information */
729 	info = smp_get_core_info();
730 	if (!info)
731 		panic("smp_detect_cpus failed to allocate memory\n");
732 
733 	/* Find boot CPU type */
734 	if (sclp.has_core_type) {
735 		address = stap();
736 		for (cpu = 0; cpu < info->combined; cpu++)
737 			if (info->core[cpu].core_id == address) {
738 				/* The boot cpu dictates the cpu type. */
739 				boot_core_type = info->core[cpu].type;
740 				break;
741 			}
742 		if (cpu >= info->combined)
743 			panic("Could not find boot CPU type");
744 	}
745 
746 	/* Set multi-threading state for the current system */
747 	mtid = boot_core_type ? sclp.mtid : sclp.mtid_cp;
748 	mtid = (mtid < smp_max_threads) ? mtid : smp_max_threads - 1;
749 	pcpu_set_smt(mtid);
750 
751 	/* Print number of CPUs */
752 	c_cpus = s_cpus = 0;
753 	for (cpu = 0; cpu < info->combined; cpu++) {
754 		if (sclp.has_core_type &&
755 		    info->core[cpu].type != boot_core_type)
756 			continue;
757 		if (cpu < info->configured)
758 			c_cpus += smp_cpu_mtid + 1;
759 		else
760 			s_cpus += smp_cpu_mtid + 1;
761 	}
762 	pr_info("%d configured CPUs, %d standby CPUs\n", c_cpus, s_cpus);
763 
764 	/* Add CPUs present at boot */
765 	get_online_cpus();
766 	__smp_rescan_cpus(info, 0);
767 	put_online_cpus();
768 	kfree(info);
769 }
770 
771 /*
772  *	Activate a secondary processor.
773  */
774 static void smp_start_secondary(void *cpuvoid)
775 {
776 	S390_lowcore.last_update_clock = get_tod_clock();
777 	S390_lowcore.restart_stack = (unsigned long) restart_stack;
778 	S390_lowcore.restart_fn = (unsigned long) do_restart;
779 	S390_lowcore.restart_data = 0;
780 	S390_lowcore.restart_source = -1UL;
781 	restore_access_regs(S390_lowcore.access_regs_save_area);
782 	__ctl_load(S390_lowcore.cregs_save_area, 0, 15);
783 	__load_psw_mask(PSW_KERNEL_BITS | PSW_MASK_DAT);
784 	cpu_init();
785 	preempt_disable();
786 	init_cpu_timer();
787 	vtime_init();
788 	pfault_init();
789 	notify_cpu_starting(smp_processor_id());
790 	set_cpu_online(smp_processor_id(), true);
791 	inc_irq_stat(CPU_RST);
792 	local_irq_enable();
793 	cpu_startup_entry(CPUHP_AP_ONLINE_IDLE);
794 }
795 
796 /* Upping and downing of CPUs */
797 int __cpu_up(unsigned int cpu, struct task_struct *tidle)
798 {
799 	struct pcpu *pcpu;
800 	int base, i, rc;
801 
802 	pcpu = pcpu_devices + cpu;
803 	if (pcpu->state != CPU_STATE_CONFIGURED)
804 		return -EIO;
805 	base = cpu - (cpu % (smp_cpu_mtid + 1));
806 	for (i = 0; i <= smp_cpu_mtid; i++) {
807 		if (base + i < nr_cpu_ids)
808 			if (cpu_online(base + i))
809 				break;
810 	}
811 	/*
812 	 * If this is the first CPU of the core to get online
813 	 * do an initial CPU reset.
814 	 */
815 	if (i > smp_cpu_mtid &&
816 	    pcpu_sigp_retry(pcpu_devices + base, SIGP_INITIAL_CPU_RESET, 0) !=
817 	    SIGP_CC_ORDER_CODE_ACCEPTED)
818 		return -EIO;
819 
820 	rc = pcpu_alloc_lowcore(pcpu, cpu);
821 	if (rc)
822 		return rc;
823 	pcpu_prepare_secondary(pcpu, cpu);
824 	pcpu_attach_task(pcpu, tidle);
825 	pcpu_start_fn(pcpu, smp_start_secondary, NULL);
826 	/* Wait until cpu puts itself in the online & active maps */
827 	while (!cpu_online(cpu))
828 		cpu_relax();
829 	return 0;
830 }
831 
832 static unsigned int setup_possible_cpus __initdata;
833 
834 static int __init _setup_possible_cpus(char *s)
835 {
836 	get_option(&s, &setup_possible_cpus);
837 	return 0;
838 }
839 early_param("possible_cpus", _setup_possible_cpus);
840 
841 #ifdef CONFIG_HOTPLUG_CPU
842 
843 int __cpu_disable(void)
844 {
845 	unsigned long cregs[16];
846 
847 	/* Handle possible pending IPIs */
848 	smp_handle_ext_call();
849 	set_cpu_online(smp_processor_id(), false);
850 	/* Disable pseudo page faults on this cpu. */
851 	pfault_fini();
852 	/* Disable interrupt sources via control register. */
853 	__ctl_store(cregs, 0, 15);
854 	cregs[0]  &= ~0x0000ee70UL;	/* disable all external interrupts */
855 	cregs[6]  &= ~0xff000000UL;	/* disable all I/O interrupts */
856 	cregs[14] &= ~0x1f000000UL;	/* disable most machine checks */
857 	__ctl_load(cregs, 0, 15);
858 	clear_cpu_flag(CIF_NOHZ_DELAY);
859 	return 0;
860 }
861 
862 void __cpu_die(unsigned int cpu)
863 {
864 	struct pcpu *pcpu;
865 
866 	/* Wait until target cpu is down */
867 	pcpu = pcpu_devices + cpu;
868 	while (!pcpu_stopped(pcpu))
869 		cpu_relax();
870 	pcpu_free_lowcore(pcpu);
871 	cpumask_clear_cpu(cpu, mm_cpumask(&init_mm));
872 	cpumask_clear_cpu(cpu, &init_mm.context.cpu_attach_mask);
873 }
874 
875 void __noreturn cpu_die(void)
876 {
877 	idle_task_exit();
878 	pcpu_sigp_retry(pcpu_devices + smp_processor_id(), SIGP_STOP, 0);
879 	for (;;) ;
880 }
881 
882 #endif /* CONFIG_HOTPLUG_CPU */
883 
884 void __init smp_fill_possible_mask(void)
885 {
886 	unsigned int possible, sclp_max, cpu;
887 
888 	sclp_max = max(sclp.mtid, sclp.mtid_cp) + 1;
889 	sclp_max = min(smp_max_threads, sclp_max);
890 	sclp_max = (sclp.max_cores * sclp_max) ?: nr_cpu_ids;
891 	possible = setup_possible_cpus ?: nr_cpu_ids;
892 	possible = min(possible, sclp_max);
893 	for (cpu = 0; cpu < possible && cpu < nr_cpu_ids; cpu++)
894 		set_cpu_possible(cpu, true);
895 }
896 
897 void __init smp_prepare_cpus(unsigned int max_cpus)
898 {
899 	/* request the 0x1201 emergency signal external interrupt */
900 	if (register_external_irq(EXT_IRQ_EMERGENCY_SIG, do_ext_call_interrupt))
901 		panic("Couldn't request external interrupt 0x1201");
902 	/* request the 0x1202 external call external interrupt */
903 	if (register_external_irq(EXT_IRQ_EXTERNAL_CALL, do_ext_call_interrupt))
904 		panic("Couldn't request external interrupt 0x1202");
905 	smp_detect_cpus();
906 }
907 
908 void __init smp_prepare_boot_cpu(void)
909 {
910 	struct pcpu *pcpu = pcpu_devices;
911 
912 	pcpu->state = CPU_STATE_CONFIGURED;
913 	pcpu->address = stap();
914 	pcpu->lowcore = (struct lowcore *)(unsigned long) store_prefix();
915 	S390_lowcore.percpu_offset = __per_cpu_offset[0];
916 	smp_cpu_set_polarization(0, POLARIZATION_UNKNOWN);
917 	set_cpu_present(0, true);
918 	set_cpu_online(0, true);
919 }
920 
921 void __init smp_cpus_done(unsigned int max_cpus)
922 {
923 }
924 
925 void __init smp_setup_processor_id(void)
926 {
927 	S390_lowcore.cpu_nr = 0;
928 	S390_lowcore.spinlock_lockval = arch_spin_lockval(0);
929 }
930 
931 /*
932  * the frequency of the profiling timer can be changed
933  * by writing a multiplier value into /proc/profile.
934  *
935  * usually you want to run this on all CPUs ;)
936  */
937 int setup_profiling_timer(unsigned int multiplier)
938 {
939 	return 0;
940 }
941 
942 #ifdef CONFIG_HOTPLUG_CPU
943 static ssize_t cpu_configure_show(struct device *dev,
944 				  struct device_attribute *attr, char *buf)
945 {
946 	ssize_t count;
947 
948 	mutex_lock(&smp_cpu_state_mutex);
949 	count = sprintf(buf, "%d\n", pcpu_devices[dev->id].state);
950 	mutex_unlock(&smp_cpu_state_mutex);
951 	return count;
952 }
953 
954 static ssize_t cpu_configure_store(struct device *dev,
955 				   struct device_attribute *attr,
956 				   const char *buf, size_t count)
957 {
958 	struct pcpu *pcpu;
959 	int cpu, val, rc, i;
960 	char delim;
961 
962 	if (sscanf(buf, "%d %c", &val, &delim) != 1)
963 		return -EINVAL;
964 	if (val != 0 && val != 1)
965 		return -EINVAL;
966 	get_online_cpus();
967 	mutex_lock(&smp_cpu_state_mutex);
968 	rc = -EBUSY;
969 	/* disallow configuration changes of online cpus and cpu 0 */
970 	cpu = dev->id;
971 	cpu -= cpu % (smp_cpu_mtid + 1);
972 	if (cpu == 0)
973 		goto out;
974 	for (i = 0; i <= smp_cpu_mtid; i++)
975 		if (cpu_online(cpu + i))
976 			goto out;
977 	pcpu = pcpu_devices + cpu;
978 	rc = 0;
979 	switch (val) {
980 	case 0:
981 		if (pcpu->state != CPU_STATE_CONFIGURED)
982 			break;
983 		rc = sclp_core_deconfigure(pcpu->address >> smp_cpu_mt_shift);
984 		if (rc)
985 			break;
986 		for (i = 0; i <= smp_cpu_mtid; i++) {
987 			if (cpu + i >= nr_cpu_ids || !cpu_present(cpu + i))
988 				continue;
989 			pcpu[i].state = CPU_STATE_STANDBY;
990 			smp_cpu_set_polarization(cpu + i,
991 						 POLARIZATION_UNKNOWN);
992 		}
993 		topology_expect_change();
994 		break;
995 	case 1:
996 		if (pcpu->state != CPU_STATE_STANDBY)
997 			break;
998 		rc = sclp_core_configure(pcpu->address >> smp_cpu_mt_shift);
999 		if (rc)
1000 			break;
1001 		for (i = 0; i <= smp_cpu_mtid; i++) {
1002 			if (cpu + i >= nr_cpu_ids || !cpu_present(cpu + i))
1003 				continue;
1004 			pcpu[i].state = CPU_STATE_CONFIGURED;
1005 			smp_cpu_set_polarization(cpu + i,
1006 						 POLARIZATION_UNKNOWN);
1007 		}
1008 		topology_expect_change();
1009 		break;
1010 	default:
1011 		break;
1012 	}
1013 out:
1014 	mutex_unlock(&smp_cpu_state_mutex);
1015 	put_online_cpus();
1016 	return rc ? rc : count;
1017 }
1018 static DEVICE_ATTR(configure, 0644, cpu_configure_show, cpu_configure_store);
1019 #endif /* CONFIG_HOTPLUG_CPU */
1020 
1021 static ssize_t show_cpu_address(struct device *dev,
1022 				struct device_attribute *attr, char *buf)
1023 {
1024 	return sprintf(buf, "%d\n", pcpu_devices[dev->id].address);
1025 }
1026 static DEVICE_ATTR(address, 0444, show_cpu_address, NULL);
1027 
1028 static struct attribute *cpu_common_attrs[] = {
1029 #ifdef CONFIG_HOTPLUG_CPU
1030 	&dev_attr_configure.attr,
1031 #endif
1032 	&dev_attr_address.attr,
1033 	NULL,
1034 };
1035 
1036 static struct attribute_group cpu_common_attr_group = {
1037 	.attrs = cpu_common_attrs,
1038 };
1039 
1040 static struct attribute *cpu_online_attrs[] = {
1041 	&dev_attr_idle_count.attr,
1042 	&dev_attr_idle_time_us.attr,
1043 	NULL,
1044 };
1045 
1046 static struct attribute_group cpu_online_attr_group = {
1047 	.attrs = cpu_online_attrs,
1048 };
1049 
1050 static int smp_cpu_notify(struct notifier_block *self, unsigned long action,
1051 			  void *hcpu)
1052 {
1053 	unsigned int cpu = (unsigned int)(long)hcpu;
1054 	struct device *s = &per_cpu(cpu_device, cpu)->dev;
1055 	int err = 0;
1056 
1057 	switch (action & ~CPU_TASKS_FROZEN) {
1058 	case CPU_ONLINE:
1059 		err = sysfs_create_group(&s->kobj, &cpu_online_attr_group);
1060 		break;
1061 	case CPU_DEAD:
1062 		sysfs_remove_group(&s->kobj, &cpu_online_attr_group);
1063 		break;
1064 	}
1065 	return notifier_from_errno(err);
1066 }
1067 
1068 static int smp_add_present_cpu(int cpu)
1069 {
1070 	struct device *s;
1071 	struct cpu *c;
1072 	int rc;
1073 
1074 	c = kzalloc(sizeof(*c), GFP_KERNEL);
1075 	if (!c)
1076 		return -ENOMEM;
1077 	per_cpu(cpu_device, cpu) = c;
1078 	s = &c->dev;
1079 	c->hotpluggable = 1;
1080 	rc = register_cpu(c, cpu);
1081 	if (rc)
1082 		goto out;
1083 	rc = sysfs_create_group(&s->kobj, &cpu_common_attr_group);
1084 	if (rc)
1085 		goto out_cpu;
1086 	if (cpu_online(cpu)) {
1087 		rc = sysfs_create_group(&s->kobj, &cpu_online_attr_group);
1088 		if (rc)
1089 			goto out_online;
1090 	}
1091 	rc = topology_cpu_init(c);
1092 	if (rc)
1093 		goto out_topology;
1094 	return 0;
1095 
1096 out_topology:
1097 	if (cpu_online(cpu))
1098 		sysfs_remove_group(&s->kobj, &cpu_online_attr_group);
1099 out_online:
1100 	sysfs_remove_group(&s->kobj, &cpu_common_attr_group);
1101 out_cpu:
1102 #ifdef CONFIG_HOTPLUG_CPU
1103 	unregister_cpu(c);
1104 #endif
1105 out:
1106 	return rc;
1107 }
1108 
1109 #ifdef CONFIG_HOTPLUG_CPU
1110 
1111 int __ref smp_rescan_cpus(void)
1112 {
1113 	struct sclp_core_info *info;
1114 	int nr;
1115 
1116 	info = smp_get_core_info();
1117 	if (!info)
1118 		return -ENOMEM;
1119 	get_online_cpus();
1120 	mutex_lock(&smp_cpu_state_mutex);
1121 	nr = __smp_rescan_cpus(info, 1);
1122 	mutex_unlock(&smp_cpu_state_mutex);
1123 	put_online_cpus();
1124 	kfree(info);
1125 	if (nr)
1126 		topology_schedule_update();
1127 	return 0;
1128 }
1129 
1130 static ssize_t __ref rescan_store(struct device *dev,
1131 				  struct device_attribute *attr,
1132 				  const char *buf,
1133 				  size_t count)
1134 {
1135 	int rc;
1136 
1137 	rc = smp_rescan_cpus();
1138 	return rc ? rc : count;
1139 }
1140 static DEVICE_ATTR(rescan, 0200, NULL, rescan_store);
1141 #endif /* CONFIG_HOTPLUG_CPU */
1142 
1143 static int __init s390_smp_init(void)
1144 {
1145 	int cpu, rc = 0;
1146 
1147 #ifdef CONFIG_HOTPLUG_CPU
1148 	rc = device_create_file(cpu_subsys.dev_root, &dev_attr_rescan);
1149 	if (rc)
1150 		return rc;
1151 #endif
1152 	cpu_notifier_register_begin();
1153 	for_each_present_cpu(cpu) {
1154 		rc = smp_add_present_cpu(cpu);
1155 		if (rc)
1156 			goto out;
1157 	}
1158 
1159 	__hotcpu_notifier(smp_cpu_notify, 0);
1160 
1161 out:
1162 	cpu_notifier_register_done();
1163 	return rc;
1164 }
1165 subsys_initcall(s390_smp_init);
1166