xref: /openbmc/linux/arch/s390/kernel/smp.c (revision 92a2c6b2)
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 <asm/asm-offsets.h>
35 #include <asm/switch_to.h>
36 #include <asm/facility.h>
37 #include <asm/ipl.h>
38 #include <asm/setup.h>
39 #include <asm/irq.h>
40 #include <asm/tlbflush.h>
41 #include <asm/vtimer.h>
42 #include <asm/lowcore.h>
43 #include <asm/sclp.h>
44 #include <asm/vdso.h>
45 #include <asm/debug.h>
46 #include <asm/os_info.h>
47 #include <asm/sigp.h>
48 #include <asm/idle.h>
49 #include "entry.h"
50 
51 enum {
52 	ec_schedule = 0,
53 	ec_call_function_single,
54 	ec_stop_cpu,
55 };
56 
57 enum {
58 	CPU_STATE_STANDBY,
59 	CPU_STATE_CONFIGURED,
60 };
61 
62 static DEFINE_PER_CPU(struct cpu *, cpu_device);
63 
64 struct pcpu {
65 	struct _lowcore *lowcore;	/* lowcore page(s) for the cpu */
66 	unsigned long ec_mask;		/* bit mask for ec_xxx functions */
67 	signed char state;		/* physical cpu state */
68 	signed char polarization;	/* physical polarization */
69 	u16 address;			/* physical cpu address */
70 };
71 
72 static u8 boot_cpu_type;
73 static struct pcpu pcpu_devices[NR_CPUS];
74 
75 unsigned int smp_cpu_mt_shift;
76 EXPORT_SYMBOL(smp_cpu_mt_shift);
77 
78 unsigned int smp_cpu_mtid;
79 EXPORT_SYMBOL(smp_cpu_mtid);
80 
81 static unsigned int smp_max_threads __initdata = -1U;
82 
83 static int __init early_nosmt(char *s)
84 {
85 	smp_max_threads = 1;
86 	return 0;
87 }
88 early_param("nosmt", early_nosmt);
89 
90 static int __init early_smt(char *s)
91 {
92 	get_option(&s, &smp_max_threads);
93 	return 0;
94 }
95 early_param("smt", early_smt);
96 
97 /*
98  * The smp_cpu_state_mutex must be held when changing the state or polarization
99  * member of a pcpu data structure within the pcpu_devices arreay.
100  */
101 DEFINE_MUTEX(smp_cpu_state_mutex);
102 
103 /*
104  * Signal processor helper functions.
105  */
106 static inline int __pcpu_sigp_relax(u16 addr, u8 order, unsigned long parm,
107 				    u32 *status)
108 {
109 	int cc;
110 
111 	while (1) {
112 		cc = __pcpu_sigp(addr, order, parm, NULL);
113 		if (cc != SIGP_CC_BUSY)
114 			return cc;
115 		cpu_relax();
116 	}
117 }
118 
119 static int pcpu_sigp_retry(struct pcpu *pcpu, u8 order, u32 parm)
120 {
121 	int cc, retry;
122 
123 	for (retry = 0; ; retry++) {
124 		cc = __pcpu_sigp(pcpu->address, order, parm, NULL);
125 		if (cc != SIGP_CC_BUSY)
126 			break;
127 		if (retry >= 3)
128 			udelay(10);
129 	}
130 	return cc;
131 }
132 
133 static inline int pcpu_stopped(struct pcpu *pcpu)
134 {
135 	u32 uninitialized_var(status);
136 
137 	if (__pcpu_sigp(pcpu->address, SIGP_SENSE,
138 			0, &status) != SIGP_CC_STATUS_STORED)
139 		return 0;
140 	return !!(status & (SIGP_STATUS_CHECK_STOP|SIGP_STATUS_STOPPED));
141 }
142 
143 static inline int pcpu_running(struct pcpu *pcpu)
144 {
145 	if (__pcpu_sigp(pcpu->address, SIGP_SENSE_RUNNING,
146 			0, NULL) != SIGP_CC_STATUS_STORED)
147 		return 1;
148 	/* Status stored condition code is equivalent to cpu not running. */
149 	return 0;
150 }
151 
152 /*
153  * Find struct pcpu by cpu address.
154  */
155 static struct pcpu *pcpu_find_address(const struct cpumask *mask, u16 address)
156 {
157 	int cpu;
158 
159 	for_each_cpu(cpu, mask)
160 		if (pcpu_devices[cpu].address == address)
161 			return pcpu_devices + cpu;
162 	return NULL;
163 }
164 
165 static void pcpu_ec_call(struct pcpu *pcpu, int ec_bit)
166 {
167 	int order;
168 
169 	if (test_and_set_bit(ec_bit, &pcpu->ec_mask))
170 		return;
171 	order = pcpu_running(pcpu) ? SIGP_EXTERNAL_CALL : SIGP_EMERGENCY_SIGNAL;
172 	pcpu_sigp_retry(pcpu, order, 0);
173 }
174 
175 #define ASYNC_FRAME_OFFSET (ASYNC_SIZE - STACK_FRAME_OVERHEAD - __PT_SIZE)
176 #define PANIC_FRAME_OFFSET (PAGE_SIZE - STACK_FRAME_OVERHEAD - __PT_SIZE)
177 
178 static int pcpu_alloc_lowcore(struct pcpu *pcpu, int cpu)
179 {
180 	unsigned long async_stack, panic_stack;
181 	struct _lowcore *lc;
182 
183 	if (pcpu != &pcpu_devices[0]) {
184 		pcpu->lowcore =	(struct _lowcore *)
185 			__get_free_pages(GFP_KERNEL | GFP_DMA, LC_ORDER);
186 		async_stack = __get_free_pages(GFP_KERNEL, ASYNC_ORDER);
187 		panic_stack = __get_free_page(GFP_KERNEL);
188 		if (!pcpu->lowcore || !panic_stack || !async_stack)
189 			goto out;
190 	} else {
191 		async_stack = pcpu->lowcore->async_stack - ASYNC_FRAME_OFFSET;
192 		panic_stack = pcpu->lowcore->panic_stack - PANIC_FRAME_OFFSET;
193 	}
194 	lc = pcpu->lowcore;
195 	memcpy(lc, &S390_lowcore, 512);
196 	memset((char *) lc + 512, 0, sizeof(*lc) - 512);
197 	lc->async_stack = async_stack + ASYNC_FRAME_OFFSET;
198 	lc->panic_stack = panic_stack + PANIC_FRAME_OFFSET;
199 	lc->cpu_nr = cpu;
200 	lc->spinlock_lockval = arch_spin_lockval(cpu);
201 #ifndef CONFIG_64BIT
202 	if (MACHINE_HAS_IEEE) {
203 		lc->extended_save_area_addr = get_zeroed_page(GFP_KERNEL);
204 		if (!lc->extended_save_area_addr)
205 			goto out;
206 	}
207 #else
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 #endif
214 	lowcore_ptr[cpu] = lc;
215 	pcpu_sigp_retry(pcpu, SIGP_SET_PREFIX, (u32)(unsigned long) lc);
216 	return 0;
217 out:
218 	if (pcpu != &pcpu_devices[0]) {
219 		free_page(panic_stack);
220 		free_pages(async_stack, ASYNC_ORDER);
221 		free_pages((unsigned long) pcpu->lowcore, LC_ORDER);
222 	}
223 	return -ENOMEM;
224 }
225 
226 #ifdef CONFIG_HOTPLUG_CPU
227 
228 static void pcpu_free_lowcore(struct pcpu *pcpu)
229 {
230 	pcpu_sigp_retry(pcpu, SIGP_SET_PREFIX, 0);
231 	lowcore_ptr[pcpu - pcpu_devices] = NULL;
232 #ifndef CONFIG_64BIT
233 	if (MACHINE_HAS_IEEE) {
234 		struct _lowcore *lc = pcpu->lowcore;
235 
236 		free_page((unsigned long) lc->extended_save_area_addr);
237 		lc->extended_save_area_addr = 0;
238 	}
239 #else
240 	vdso_free_per_cpu(pcpu->lowcore);
241 #endif
242 	if (pcpu == &pcpu_devices[0])
243 		return;
244 	free_page(pcpu->lowcore->panic_stack-PANIC_FRAME_OFFSET);
245 	free_pages(pcpu->lowcore->async_stack-ASYNC_FRAME_OFFSET, ASYNC_ORDER);
246 	free_pages((unsigned long) pcpu->lowcore, LC_ORDER);
247 }
248 
249 #endif /* CONFIG_HOTPLUG_CPU */
250 
251 static void pcpu_prepare_secondary(struct pcpu *pcpu, int cpu)
252 {
253 	struct _lowcore *lc = pcpu->lowcore;
254 
255 	if (MACHINE_HAS_TLB_LC)
256 		cpumask_set_cpu(cpu, &init_mm.context.cpu_attach_mask);
257 	cpumask_set_cpu(cpu, mm_cpumask(&init_mm));
258 	atomic_inc(&init_mm.context.attach_count);
259 	lc->cpu_nr = cpu;
260 	lc->spinlock_lockval = arch_spin_lockval(cpu);
261 	lc->percpu_offset = __per_cpu_offset[cpu];
262 	lc->kernel_asce = S390_lowcore.kernel_asce;
263 	lc->machine_flags = S390_lowcore.machine_flags;
264 	lc->user_timer = lc->system_timer = lc->steal_timer = 0;
265 	__ctl_store(lc->cregs_save_area, 0, 15);
266 	save_access_regs((unsigned int *) lc->access_regs_save_area);
267 	memcpy(lc->stfle_fac_list, S390_lowcore.stfle_fac_list,
268 	       MAX_FACILITY_BIT/8);
269 }
270 
271 static void pcpu_attach_task(struct pcpu *pcpu, struct task_struct *tsk)
272 {
273 	struct _lowcore *lc = pcpu->lowcore;
274 	struct thread_info *ti = task_thread_info(tsk);
275 
276 	lc->kernel_stack = (unsigned long) task_stack_page(tsk)
277 		+ THREAD_SIZE - STACK_FRAME_OVERHEAD - sizeof(struct pt_regs);
278 	lc->thread_info = (unsigned long) task_thread_info(tsk);
279 	lc->current_task = (unsigned long) tsk;
280 	lc->user_timer = ti->user_timer;
281 	lc->system_timer = ti->system_timer;
282 	lc->steal_timer = 0;
283 }
284 
285 static void pcpu_start_fn(struct pcpu *pcpu, void (*func)(void *), void *data)
286 {
287 	struct _lowcore *lc = pcpu->lowcore;
288 
289 	lc->restart_stack = lc->kernel_stack;
290 	lc->restart_fn = (unsigned long) func;
291 	lc->restart_data = (unsigned long) data;
292 	lc->restart_source = -1UL;
293 	pcpu_sigp_retry(pcpu, SIGP_RESTART, 0);
294 }
295 
296 /*
297  * Call function via PSW restart on pcpu and stop the current cpu.
298  */
299 static void pcpu_delegate(struct pcpu *pcpu, void (*func)(void *),
300 			  void *data, unsigned long stack)
301 {
302 	struct _lowcore *lc = lowcore_ptr[pcpu - pcpu_devices];
303 	unsigned long source_cpu = stap();
304 
305 	__load_psw_mask(PSW_KERNEL_BITS);
306 	if (pcpu->address == source_cpu)
307 		func(data);	/* should not return */
308 	/* Stop target cpu (if func returns this stops the current cpu). */
309 	pcpu_sigp_retry(pcpu, SIGP_STOP, 0);
310 	/* Restart func on the target cpu and stop the current cpu. */
311 	mem_assign_absolute(lc->restart_stack, stack);
312 	mem_assign_absolute(lc->restart_fn, (unsigned long) func);
313 	mem_assign_absolute(lc->restart_data, (unsigned long) data);
314 	mem_assign_absolute(lc->restart_source, source_cpu);
315 	asm volatile(
316 		"0:	sigp	0,%0,%2	# sigp restart to target cpu\n"
317 		"	brc	2,0b	# busy, try again\n"
318 		"1:	sigp	0,%1,%3	# sigp stop to current cpu\n"
319 		"	brc	2,1b	# busy, try again\n"
320 		: : "d" (pcpu->address), "d" (source_cpu),
321 		    "K" (SIGP_RESTART), "K" (SIGP_STOP)
322 		: "0", "1", "cc");
323 	for (;;) ;
324 }
325 
326 /*
327  * Enable additional logical cpus for multi-threading.
328  */
329 static int pcpu_set_smt(unsigned int mtid)
330 {
331 	register unsigned long reg1 asm ("1") = (unsigned long) mtid;
332 	int cc;
333 
334 	if (smp_cpu_mtid == mtid)
335 		return 0;
336 	asm volatile(
337 		"	sigp	%1,0,%2	# sigp set multi-threading\n"
338 		"	ipm	%0\n"
339 		"	srl	%0,28\n"
340 		: "=d" (cc) : "d" (reg1), "K" (SIGP_SET_MULTI_THREADING)
341 		: "cc");
342 	if (cc == 0) {
343 		smp_cpu_mtid = mtid;
344 		smp_cpu_mt_shift = 0;
345 		while (smp_cpu_mtid >= (1U << smp_cpu_mt_shift))
346 			smp_cpu_mt_shift++;
347 		pcpu_devices[0].address = stap();
348 	}
349 	return cc;
350 }
351 
352 /*
353  * Call function on an online CPU.
354  */
355 void smp_call_online_cpu(void (*func)(void *), void *data)
356 {
357 	struct pcpu *pcpu;
358 
359 	/* Use the current cpu if it is online. */
360 	pcpu = pcpu_find_address(cpu_online_mask, stap());
361 	if (!pcpu)
362 		/* Use the first online cpu. */
363 		pcpu = pcpu_devices + cpumask_first(cpu_online_mask);
364 	pcpu_delegate(pcpu, func, data, (unsigned long) restart_stack);
365 }
366 
367 /*
368  * Call function on the ipl CPU.
369  */
370 void smp_call_ipl_cpu(void (*func)(void *), void *data)
371 {
372 	pcpu_delegate(&pcpu_devices[0], func, data,
373 		      pcpu_devices->lowcore->panic_stack -
374 		      PANIC_FRAME_OFFSET + PAGE_SIZE);
375 }
376 
377 int smp_find_processor_id(u16 address)
378 {
379 	int cpu;
380 
381 	for_each_present_cpu(cpu)
382 		if (pcpu_devices[cpu].address == address)
383 			return cpu;
384 	return -1;
385 }
386 
387 int smp_vcpu_scheduled(int cpu)
388 {
389 	return pcpu_running(pcpu_devices + cpu);
390 }
391 
392 void smp_yield_cpu(int cpu)
393 {
394 	if (MACHINE_HAS_DIAG9C)
395 		asm volatile("diag %0,0,0x9c"
396 			     : : "d" (pcpu_devices[cpu].address));
397 	else if (MACHINE_HAS_DIAG44)
398 		asm volatile("diag 0,0,0x44");
399 }
400 
401 /*
402  * Send cpus emergency shutdown signal. This gives the cpus the
403  * opportunity to complete outstanding interrupts.
404  */
405 static void smp_emergency_stop(cpumask_t *cpumask)
406 {
407 	u64 end;
408 	int cpu;
409 
410 	end = get_tod_clock() + (1000000UL << 12);
411 	for_each_cpu(cpu, cpumask) {
412 		struct pcpu *pcpu = pcpu_devices + cpu;
413 		set_bit(ec_stop_cpu, &pcpu->ec_mask);
414 		while (__pcpu_sigp(pcpu->address, SIGP_EMERGENCY_SIGNAL,
415 				   0, NULL) == SIGP_CC_BUSY &&
416 		       get_tod_clock() < end)
417 			cpu_relax();
418 	}
419 	while (get_tod_clock() < end) {
420 		for_each_cpu(cpu, cpumask)
421 			if (pcpu_stopped(pcpu_devices + cpu))
422 				cpumask_clear_cpu(cpu, cpumask);
423 		if (cpumask_empty(cpumask))
424 			break;
425 		cpu_relax();
426 	}
427 }
428 
429 /*
430  * Stop all cpus but the current one.
431  */
432 void smp_send_stop(void)
433 {
434 	cpumask_t cpumask;
435 	int cpu;
436 
437 	/* Disable all interrupts/machine checks */
438 	__load_psw_mask(PSW_KERNEL_BITS | PSW_MASK_DAT);
439 	trace_hardirqs_off();
440 
441 	debug_set_critical();
442 	cpumask_copy(&cpumask, cpu_online_mask);
443 	cpumask_clear_cpu(smp_processor_id(), &cpumask);
444 
445 	if (oops_in_progress)
446 		smp_emergency_stop(&cpumask);
447 
448 	/* stop all processors */
449 	for_each_cpu(cpu, &cpumask) {
450 		struct pcpu *pcpu = pcpu_devices + cpu;
451 		pcpu_sigp_retry(pcpu, SIGP_STOP, 0);
452 		while (!pcpu_stopped(pcpu))
453 			cpu_relax();
454 	}
455 }
456 
457 /*
458  * This is the main routine where commands issued by other
459  * cpus are handled.
460  */
461 static void smp_handle_ext_call(void)
462 {
463 	unsigned long bits;
464 
465 	/* handle bit signal external calls */
466 	bits = xchg(&pcpu_devices[smp_processor_id()].ec_mask, 0);
467 	if (test_bit(ec_stop_cpu, &bits))
468 		smp_stop_cpu();
469 	if (test_bit(ec_schedule, &bits))
470 		scheduler_ipi();
471 	if (test_bit(ec_call_function_single, &bits))
472 		generic_smp_call_function_single_interrupt();
473 }
474 
475 static void do_ext_call_interrupt(struct ext_code ext_code,
476 				  unsigned int param32, unsigned long param64)
477 {
478 	inc_irq_stat(ext_code.code == 0x1202 ? IRQEXT_EXC : IRQEXT_EMS);
479 	smp_handle_ext_call();
480 }
481 
482 void arch_send_call_function_ipi_mask(const struct cpumask *mask)
483 {
484 	int cpu;
485 
486 	for_each_cpu(cpu, mask)
487 		pcpu_ec_call(pcpu_devices + cpu, ec_call_function_single);
488 }
489 
490 void arch_send_call_function_single_ipi(int cpu)
491 {
492 	pcpu_ec_call(pcpu_devices + cpu, ec_call_function_single);
493 }
494 
495 #ifndef CONFIG_64BIT
496 /*
497  * this function sends a 'purge tlb' signal to another CPU.
498  */
499 static void smp_ptlb_callback(void *info)
500 {
501 	__tlb_flush_local();
502 }
503 
504 void smp_ptlb_all(void)
505 {
506 	on_each_cpu(smp_ptlb_callback, NULL, 1);
507 }
508 EXPORT_SYMBOL(smp_ptlb_all);
509 #endif /* ! CONFIG_64BIT */
510 
511 /*
512  * this function sends a 'reschedule' IPI to another CPU.
513  * it goes straight through and wastes no time serializing
514  * anything. Worst case is that we lose a reschedule ...
515  */
516 void smp_send_reschedule(int cpu)
517 {
518 	pcpu_ec_call(pcpu_devices + cpu, ec_schedule);
519 }
520 
521 /*
522  * parameter area for the set/clear control bit callbacks
523  */
524 struct ec_creg_mask_parms {
525 	unsigned long orval;
526 	unsigned long andval;
527 	int cr;
528 };
529 
530 /*
531  * callback for setting/clearing control bits
532  */
533 static void smp_ctl_bit_callback(void *info)
534 {
535 	struct ec_creg_mask_parms *pp = info;
536 	unsigned long cregs[16];
537 
538 	__ctl_store(cregs, 0, 15);
539 	cregs[pp->cr] = (cregs[pp->cr] & pp->andval) | pp->orval;
540 	__ctl_load(cregs, 0, 15);
541 }
542 
543 /*
544  * Set a bit in a control register of all cpus
545  */
546 void smp_ctl_set_bit(int cr, int bit)
547 {
548 	struct ec_creg_mask_parms parms = { 1UL << bit, -1UL, cr };
549 
550 	on_each_cpu(smp_ctl_bit_callback, &parms, 1);
551 }
552 EXPORT_SYMBOL(smp_ctl_set_bit);
553 
554 /*
555  * Clear a bit in a control register of all cpus
556  */
557 void smp_ctl_clear_bit(int cr, int bit)
558 {
559 	struct ec_creg_mask_parms parms = { 0, ~(1UL << bit), cr };
560 
561 	on_each_cpu(smp_ctl_bit_callback, &parms, 1);
562 }
563 EXPORT_SYMBOL(smp_ctl_clear_bit);
564 
565 #ifdef CONFIG_CRASH_DUMP
566 
567 static inline void __smp_store_cpu_state(int cpu, u16 address, int is_boot_cpu)
568 {
569 	void *lc = pcpu_devices[0].lowcore;
570 	struct save_area_ext *sa_ext;
571 	unsigned long vx_sa;
572 
573 	sa_ext = dump_save_area_create(cpu);
574 	if (!sa_ext)
575 		panic("could not allocate memory for save area\n");
576 	if (is_boot_cpu) {
577 		/* Copy the registers of the boot CPU. */
578 		copy_oldmem_page(1, (void *) &sa_ext->sa, sizeof(sa_ext->sa),
579 				 SAVE_AREA_BASE - PAGE_SIZE, 0);
580 		if (MACHINE_HAS_VX)
581 			save_vx_regs_safe(sa_ext->vx_regs);
582 		return;
583 	}
584 	/* Get the registers of a non-boot cpu. */
585 	__pcpu_sigp_relax(address, SIGP_STOP_AND_STORE_STATUS, 0, NULL);
586 	memcpy_real(&sa_ext->sa, lc + SAVE_AREA_BASE, sizeof(sa_ext->sa));
587 	if (!MACHINE_HAS_VX)
588 		return;
589 	/* Get the VX registers */
590 	vx_sa = __get_free_page(GFP_KERNEL);
591 	if (!vx_sa)
592 		panic("could not allocate memory for VX save area\n");
593 	__pcpu_sigp_relax(address, SIGP_STORE_ADDITIONAL_STATUS, vx_sa, NULL);
594 	memcpy(sa_ext->vx_regs, (void *) vx_sa, sizeof(sa_ext->vx_regs));
595 	free_page(vx_sa);
596 }
597 
598 /*
599  * Collect CPU state of the previous, crashed system.
600  * There are four cases:
601  * 1) standard zfcp dump
602  *    condition: OLDMEM_BASE == NULL && ipl_info.type == IPL_TYPE_FCP_DUMP
603  *    The state for all CPUs except the boot CPU needs to be collected
604  *    with sigp stop-and-store-status. The boot CPU state is located in
605  *    the absolute lowcore of the memory stored in the HSA. The zcore code
606  *    will allocate the save area and copy the boot CPU state from the HSA.
607  * 2) stand-alone kdump for SCSI (zfcp dump with swapped memory)
608  *    condition: OLDMEM_BASE != NULL && ipl_info.type == IPL_TYPE_FCP_DUMP
609  *    The state for all CPUs except the boot CPU needs to be collected
610  *    with sigp stop-and-store-status. The firmware or the boot-loader
611  *    stored the registers of the boot CPU in the absolute lowcore in the
612  *    memory of the old system.
613  * 3) kdump and the old kernel did not store the CPU state,
614  *    or stand-alone kdump for DASD
615  *    condition: OLDMEM_BASE != NULL && !is_kdump_kernel()
616  *    The state for all CPUs except the boot CPU needs to be collected
617  *    with sigp stop-and-store-status. The kexec code or the boot-loader
618  *    stored the registers of the boot CPU in the memory of the old system.
619  * 4) kdump and the old kernel stored the CPU state
620  *    condition: OLDMEM_BASE != NULL && is_kdump_kernel()
621  *    The state of all CPUs is stored in ELF sections in the memory of the
622  *    old system. The ELF sections are picked up by the crash_dump code
623  *    via elfcorehdr_addr.
624  */
625 static void __init smp_store_cpu_states(struct sclp_cpu_info *info)
626 {
627 	unsigned int cpu, address, i, j;
628 	int is_boot_cpu;
629 
630 	if (is_kdump_kernel())
631 		/* Previous system stored the CPU states. Nothing to do. */
632 		return;
633 	if (!(OLDMEM_BASE || ipl_info.type == IPL_TYPE_FCP_DUMP))
634 		/* No previous system present, normal boot. */
635 		return;
636 	/* Set multi-threading state to the previous system. */
637 	pcpu_set_smt(sclp_get_mtid_prev());
638 	/* Collect CPU states. */
639 	cpu = 0;
640 	for (i = 0; i < info->configured; i++) {
641 		/* Skip CPUs with different CPU type. */
642 		if (info->has_cpu_type && info->cpu[i].type != boot_cpu_type)
643 			continue;
644 		for (j = 0; j <= smp_cpu_mtid; j++, cpu++) {
645 			address = (info->cpu[i].core_id << smp_cpu_mt_shift) + j;
646 			is_boot_cpu = (address == pcpu_devices[0].address);
647 			if (is_boot_cpu && !OLDMEM_BASE)
648 				/* Skip boot CPU for standard zfcp dump. */
649 				continue;
650 			/* Get state for this CPu. */
651 			__smp_store_cpu_state(cpu, address, is_boot_cpu);
652 		}
653 	}
654 }
655 
656 int smp_store_status(int cpu)
657 {
658 	unsigned long vx_sa;
659 	struct pcpu *pcpu;
660 
661 	pcpu = pcpu_devices + cpu;
662 	if (__pcpu_sigp_relax(pcpu->address, SIGP_STOP_AND_STORE_STATUS,
663 			      0, NULL) != SIGP_CC_ORDER_CODE_ACCEPTED)
664 		return -EIO;
665 	if (!MACHINE_HAS_VX)
666 		return 0;
667 	vx_sa = __pa(pcpu->lowcore->vector_save_area_addr);
668 	__pcpu_sigp_relax(pcpu->address, SIGP_STORE_ADDITIONAL_STATUS,
669 			  vx_sa, NULL);
670 	return 0;
671 }
672 
673 #endif /* CONFIG_CRASH_DUMP */
674 
675 void smp_cpu_set_polarization(int cpu, int val)
676 {
677 	pcpu_devices[cpu].polarization = val;
678 }
679 
680 int smp_cpu_get_polarization(int cpu)
681 {
682 	return pcpu_devices[cpu].polarization;
683 }
684 
685 static struct sclp_cpu_info *smp_get_cpu_info(void)
686 {
687 	static int use_sigp_detection;
688 	struct sclp_cpu_info *info;
689 	int address;
690 
691 	info = kzalloc(sizeof(*info), GFP_KERNEL);
692 	if (info && (use_sigp_detection || sclp_get_cpu_info(info))) {
693 		use_sigp_detection = 1;
694 		for (address = 0; address <= MAX_CPU_ADDRESS;
695 		     address += (1U << smp_cpu_mt_shift)) {
696 			if (__pcpu_sigp_relax(address, SIGP_SENSE, 0, NULL) ==
697 			    SIGP_CC_NOT_OPERATIONAL)
698 				continue;
699 			info->cpu[info->configured].core_id =
700 				address >> smp_cpu_mt_shift;
701 			info->configured++;
702 		}
703 		info->combined = info->configured;
704 	}
705 	return info;
706 }
707 
708 static int smp_add_present_cpu(int cpu);
709 
710 static int __smp_rescan_cpus(struct sclp_cpu_info *info, int sysfs_add)
711 {
712 	struct pcpu *pcpu;
713 	cpumask_t avail;
714 	int cpu, nr, i, j;
715 	u16 address;
716 
717 	nr = 0;
718 	cpumask_xor(&avail, cpu_possible_mask, cpu_present_mask);
719 	cpu = cpumask_first(&avail);
720 	for (i = 0; (i < info->combined) && (cpu < nr_cpu_ids); i++) {
721 		if (info->has_cpu_type && info->cpu[i].type != boot_cpu_type)
722 			continue;
723 		address = info->cpu[i].core_id << smp_cpu_mt_shift;
724 		for (j = 0; j <= smp_cpu_mtid; j++) {
725 			if (pcpu_find_address(cpu_present_mask, address + j))
726 				continue;
727 			pcpu = pcpu_devices + cpu;
728 			pcpu->address = address + j;
729 			pcpu->state =
730 				(cpu >= info->configured*(smp_cpu_mtid + 1)) ?
731 				CPU_STATE_STANDBY : CPU_STATE_CONFIGURED;
732 			smp_cpu_set_polarization(cpu, POLARIZATION_UNKNOWN);
733 			set_cpu_present(cpu, true);
734 			if (sysfs_add && smp_add_present_cpu(cpu) != 0)
735 				set_cpu_present(cpu, false);
736 			else
737 				nr++;
738 			cpu = cpumask_next(cpu, &avail);
739 			if (cpu >= nr_cpu_ids)
740 				break;
741 		}
742 	}
743 	return nr;
744 }
745 
746 static void __init smp_detect_cpus(void)
747 {
748 	unsigned int cpu, mtid, c_cpus, s_cpus;
749 	struct sclp_cpu_info *info;
750 	u16 address;
751 
752 	/* Get CPU information */
753 	info = smp_get_cpu_info();
754 	if (!info)
755 		panic("smp_detect_cpus failed to allocate memory\n");
756 
757 	/* Find boot CPU type */
758 	if (info->has_cpu_type) {
759 		address = stap();
760 		for (cpu = 0; cpu < info->combined; cpu++)
761 			if (info->cpu[cpu].core_id == address) {
762 				/* The boot cpu dictates the cpu type. */
763 				boot_cpu_type = info->cpu[cpu].type;
764 				break;
765 			}
766 		if (cpu >= info->combined)
767 			panic("Could not find boot CPU type");
768 	}
769 
770 #ifdef CONFIG_CRASH_DUMP
771 	/* Collect CPU state of previous system */
772 	smp_store_cpu_states(info);
773 #endif
774 
775 	/* Set multi-threading state for the current system */
776 	mtid = sclp_get_mtid(boot_cpu_type);
777 	mtid = (mtid < smp_max_threads) ? mtid : smp_max_threads - 1;
778 	pcpu_set_smt(mtid);
779 
780 	/* Print number of CPUs */
781 	c_cpus = s_cpus = 0;
782 	for (cpu = 0; cpu < info->combined; cpu++) {
783 		if (info->has_cpu_type && info->cpu[cpu].type != boot_cpu_type)
784 			continue;
785 		if (cpu < info->configured)
786 			c_cpus += smp_cpu_mtid + 1;
787 		else
788 			s_cpus += smp_cpu_mtid + 1;
789 	}
790 	pr_info("%d configured CPUs, %d standby CPUs\n", c_cpus, s_cpus);
791 
792 	/* Add CPUs present at boot */
793 	get_online_cpus();
794 	__smp_rescan_cpus(info, 0);
795 	put_online_cpus();
796 	kfree(info);
797 }
798 
799 /*
800  *	Activate a secondary processor.
801  */
802 static void smp_start_secondary(void *cpuvoid)
803 {
804 	S390_lowcore.last_update_clock = get_tod_clock();
805 	S390_lowcore.restart_stack = (unsigned long) restart_stack;
806 	S390_lowcore.restart_fn = (unsigned long) do_restart;
807 	S390_lowcore.restart_data = 0;
808 	S390_lowcore.restart_source = -1UL;
809 	restore_access_regs(S390_lowcore.access_regs_save_area);
810 	__ctl_load(S390_lowcore.cregs_save_area, 0, 15);
811 	__load_psw_mask(PSW_KERNEL_BITS | PSW_MASK_DAT);
812 	cpu_init();
813 	preempt_disable();
814 	init_cpu_timer();
815 	vtime_init();
816 	pfault_init();
817 	notify_cpu_starting(smp_processor_id());
818 	set_cpu_online(smp_processor_id(), true);
819 	inc_irq_stat(CPU_RST);
820 	local_irq_enable();
821 	cpu_startup_entry(CPUHP_ONLINE);
822 }
823 
824 /* Upping and downing of CPUs */
825 int __cpu_up(unsigned int cpu, struct task_struct *tidle)
826 {
827 	struct pcpu *pcpu;
828 	int base, i, rc;
829 
830 	pcpu = pcpu_devices + cpu;
831 	if (pcpu->state != CPU_STATE_CONFIGURED)
832 		return -EIO;
833 	base = cpu - (cpu % (smp_cpu_mtid + 1));
834 	for (i = 0; i <= smp_cpu_mtid; i++) {
835 		if (base + i < nr_cpu_ids)
836 			if (cpu_online(base + i))
837 				break;
838 	}
839 	/*
840 	 * If this is the first CPU of the core to get online
841 	 * do an initial CPU reset.
842 	 */
843 	if (i > smp_cpu_mtid &&
844 	    pcpu_sigp_retry(pcpu_devices + base, SIGP_INITIAL_CPU_RESET, 0) !=
845 	    SIGP_CC_ORDER_CODE_ACCEPTED)
846 		return -EIO;
847 
848 	rc = pcpu_alloc_lowcore(pcpu, cpu);
849 	if (rc)
850 		return rc;
851 	pcpu_prepare_secondary(pcpu, cpu);
852 	pcpu_attach_task(pcpu, tidle);
853 	pcpu_start_fn(pcpu, smp_start_secondary, NULL);
854 	while (!cpu_online(cpu))
855 		cpu_relax();
856 	return 0;
857 }
858 
859 static unsigned int setup_possible_cpus __initdata;
860 
861 static int __init _setup_possible_cpus(char *s)
862 {
863 	get_option(&s, &setup_possible_cpus);
864 	return 0;
865 }
866 early_param("possible_cpus", _setup_possible_cpus);
867 
868 #ifdef CONFIG_HOTPLUG_CPU
869 
870 int __cpu_disable(void)
871 {
872 	unsigned long cregs[16];
873 
874 	/* Handle possible pending IPIs */
875 	smp_handle_ext_call();
876 	set_cpu_online(smp_processor_id(), false);
877 	/* Disable pseudo page faults on this cpu. */
878 	pfault_fini();
879 	/* Disable interrupt sources via control register. */
880 	__ctl_store(cregs, 0, 15);
881 	cregs[0]  &= ~0x0000ee70UL;	/* disable all external interrupts */
882 	cregs[6]  &= ~0xff000000UL;	/* disable all I/O interrupts */
883 	cregs[14] &= ~0x1f000000UL;	/* disable most machine checks */
884 	__ctl_load(cregs, 0, 15);
885 	clear_cpu_flag(CIF_NOHZ_DELAY);
886 	return 0;
887 }
888 
889 void __cpu_die(unsigned int cpu)
890 {
891 	struct pcpu *pcpu;
892 
893 	/* Wait until target cpu is down */
894 	pcpu = pcpu_devices + cpu;
895 	while (!pcpu_stopped(pcpu))
896 		cpu_relax();
897 	pcpu_free_lowcore(pcpu);
898 	atomic_dec(&init_mm.context.attach_count);
899 	cpumask_clear_cpu(cpu, mm_cpumask(&init_mm));
900 	if (MACHINE_HAS_TLB_LC)
901 		cpumask_clear_cpu(cpu, &init_mm.context.cpu_attach_mask);
902 }
903 
904 void __noreturn cpu_die(void)
905 {
906 	idle_task_exit();
907 	pcpu_sigp_retry(pcpu_devices + smp_processor_id(), SIGP_STOP, 0);
908 	for (;;) ;
909 }
910 
911 #endif /* CONFIG_HOTPLUG_CPU */
912 
913 void __init smp_fill_possible_mask(void)
914 {
915 	unsigned int possible, sclp, cpu;
916 
917 	sclp = min(smp_max_threads, sclp_get_mtid_max() + 1);
918 	sclp = sclp_get_max_cpu()*sclp ?: nr_cpu_ids;
919 	possible = setup_possible_cpus ?: nr_cpu_ids;
920 	possible = min(possible, sclp);
921 	for (cpu = 0; cpu < possible && cpu < nr_cpu_ids; cpu++)
922 		set_cpu_possible(cpu, true);
923 }
924 
925 void __init smp_prepare_cpus(unsigned int max_cpus)
926 {
927 	/* request the 0x1201 emergency signal external interrupt */
928 	if (register_external_irq(EXT_IRQ_EMERGENCY_SIG, do_ext_call_interrupt))
929 		panic("Couldn't request external interrupt 0x1201");
930 	/* request the 0x1202 external call external interrupt */
931 	if (register_external_irq(EXT_IRQ_EXTERNAL_CALL, do_ext_call_interrupt))
932 		panic("Couldn't request external interrupt 0x1202");
933 	smp_detect_cpus();
934 }
935 
936 void __init smp_prepare_boot_cpu(void)
937 {
938 	struct pcpu *pcpu = pcpu_devices;
939 
940 	pcpu->state = CPU_STATE_CONFIGURED;
941 	pcpu->address = stap();
942 	pcpu->lowcore = (struct _lowcore *)(unsigned long) store_prefix();
943 	S390_lowcore.percpu_offset = __per_cpu_offset[0];
944 	smp_cpu_set_polarization(0, POLARIZATION_UNKNOWN);
945 	set_cpu_present(0, true);
946 	set_cpu_online(0, true);
947 }
948 
949 void __init smp_cpus_done(unsigned int max_cpus)
950 {
951 }
952 
953 void __init smp_setup_processor_id(void)
954 {
955 	S390_lowcore.cpu_nr = 0;
956 	S390_lowcore.spinlock_lockval = arch_spin_lockval(0);
957 }
958 
959 /*
960  * the frequency of the profiling timer can be changed
961  * by writing a multiplier value into /proc/profile.
962  *
963  * usually you want to run this on all CPUs ;)
964  */
965 int setup_profiling_timer(unsigned int multiplier)
966 {
967 	return 0;
968 }
969 
970 #ifdef CONFIG_HOTPLUG_CPU
971 static ssize_t cpu_configure_show(struct device *dev,
972 				  struct device_attribute *attr, char *buf)
973 {
974 	ssize_t count;
975 
976 	mutex_lock(&smp_cpu_state_mutex);
977 	count = sprintf(buf, "%d\n", pcpu_devices[dev->id].state);
978 	mutex_unlock(&smp_cpu_state_mutex);
979 	return count;
980 }
981 
982 static ssize_t cpu_configure_store(struct device *dev,
983 				   struct device_attribute *attr,
984 				   const char *buf, size_t count)
985 {
986 	struct pcpu *pcpu;
987 	int cpu, val, rc, i;
988 	char delim;
989 
990 	if (sscanf(buf, "%d %c", &val, &delim) != 1)
991 		return -EINVAL;
992 	if (val != 0 && val != 1)
993 		return -EINVAL;
994 	get_online_cpus();
995 	mutex_lock(&smp_cpu_state_mutex);
996 	rc = -EBUSY;
997 	/* disallow configuration changes of online cpus and cpu 0 */
998 	cpu = dev->id;
999 	cpu -= cpu % (smp_cpu_mtid + 1);
1000 	if (cpu == 0)
1001 		goto out;
1002 	for (i = 0; i <= smp_cpu_mtid; i++)
1003 		if (cpu_online(cpu + i))
1004 			goto out;
1005 	pcpu = pcpu_devices + cpu;
1006 	rc = 0;
1007 	switch (val) {
1008 	case 0:
1009 		if (pcpu->state != CPU_STATE_CONFIGURED)
1010 			break;
1011 		rc = sclp_cpu_deconfigure(pcpu->address >> smp_cpu_mt_shift);
1012 		if (rc)
1013 			break;
1014 		for (i = 0; i <= smp_cpu_mtid; i++) {
1015 			if (cpu + i >= nr_cpu_ids || !cpu_present(cpu + i))
1016 				continue;
1017 			pcpu[i].state = CPU_STATE_STANDBY;
1018 			smp_cpu_set_polarization(cpu + i,
1019 						 POLARIZATION_UNKNOWN);
1020 		}
1021 		topology_expect_change();
1022 		break;
1023 	case 1:
1024 		if (pcpu->state != CPU_STATE_STANDBY)
1025 			break;
1026 		rc = sclp_cpu_configure(pcpu->address >> smp_cpu_mt_shift);
1027 		if (rc)
1028 			break;
1029 		for (i = 0; i <= smp_cpu_mtid; i++) {
1030 			if (cpu + i >= nr_cpu_ids || !cpu_present(cpu + i))
1031 				continue;
1032 			pcpu[i].state = CPU_STATE_CONFIGURED;
1033 			smp_cpu_set_polarization(cpu + i,
1034 						 POLARIZATION_UNKNOWN);
1035 		}
1036 		topology_expect_change();
1037 		break;
1038 	default:
1039 		break;
1040 	}
1041 out:
1042 	mutex_unlock(&smp_cpu_state_mutex);
1043 	put_online_cpus();
1044 	return rc ? rc : count;
1045 }
1046 static DEVICE_ATTR(configure, 0644, cpu_configure_show, cpu_configure_store);
1047 #endif /* CONFIG_HOTPLUG_CPU */
1048 
1049 static ssize_t show_cpu_address(struct device *dev,
1050 				struct device_attribute *attr, char *buf)
1051 {
1052 	return sprintf(buf, "%d\n", pcpu_devices[dev->id].address);
1053 }
1054 static DEVICE_ATTR(address, 0444, show_cpu_address, NULL);
1055 
1056 static struct attribute *cpu_common_attrs[] = {
1057 #ifdef CONFIG_HOTPLUG_CPU
1058 	&dev_attr_configure.attr,
1059 #endif
1060 	&dev_attr_address.attr,
1061 	NULL,
1062 };
1063 
1064 static struct attribute_group cpu_common_attr_group = {
1065 	.attrs = cpu_common_attrs,
1066 };
1067 
1068 static struct attribute *cpu_online_attrs[] = {
1069 	&dev_attr_idle_count.attr,
1070 	&dev_attr_idle_time_us.attr,
1071 	NULL,
1072 };
1073 
1074 static struct attribute_group cpu_online_attr_group = {
1075 	.attrs = cpu_online_attrs,
1076 };
1077 
1078 static int smp_cpu_notify(struct notifier_block *self, unsigned long action,
1079 			  void *hcpu)
1080 {
1081 	unsigned int cpu = (unsigned int)(long)hcpu;
1082 	struct device *s = &per_cpu(cpu_device, cpu)->dev;
1083 	int err = 0;
1084 
1085 	switch (action & ~CPU_TASKS_FROZEN) {
1086 	case CPU_ONLINE:
1087 		err = sysfs_create_group(&s->kobj, &cpu_online_attr_group);
1088 		break;
1089 	case CPU_DEAD:
1090 		sysfs_remove_group(&s->kobj, &cpu_online_attr_group);
1091 		break;
1092 	}
1093 	return notifier_from_errno(err);
1094 }
1095 
1096 static int smp_add_present_cpu(int cpu)
1097 {
1098 	struct device *s;
1099 	struct cpu *c;
1100 	int rc;
1101 
1102 	c = kzalloc(sizeof(*c), GFP_KERNEL);
1103 	if (!c)
1104 		return -ENOMEM;
1105 	per_cpu(cpu_device, cpu) = c;
1106 	s = &c->dev;
1107 	c->hotpluggable = 1;
1108 	rc = register_cpu(c, cpu);
1109 	if (rc)
1110 		goto out;
1111 	rc = sysfs_create_group(&s->kobj, &cpu_common_attr_group);
1112 	if (rc)
1113 		goto out_cpu;
1114 	if (cpu_online(cpu)) {
1115 		rc = sysfs_create_group(&s->kobj, &cpu_online_attr_group);
1116 		if (rc)
1117 			goto out_online;
1118 	}
1119 	rc = topology_cpu_init(c);
1120 	if (rc)
1121 		goto out_topology;
1122 	return 0;
1123 
1124 out_topology:
1125 	if (cpu_online(cpu))
1126 		sysfs_remove_group(&s->kobj, &cpu_online_attr_group);
1127 out_online:
1128 	sysfs_remove_group(&s->kobj, &cpu_common_attr_group);
1129 out_cpu:
1130 #ifdef CONFIG_HOTPLUG_CPU
1131 	unregister_cpu(c);
1132 #endif
1133 out:
1134 	return rc;
1135 }
1136 
1137 #ifdef CONFIG_HOTPLUG_CPU
1138 
1139 int __ref smp_rescan_cpus(void)
1140 {
1141 	struct sclp_cpu_info *info;
1142 	int nr;
1143 
1144 	info = smp_get_cpu_info();
1145 	if (!info)
1146 		return -ENOMEM;
1147 	get_online_cpus();
1148 	mutex_lock(&smp_cpu_state_mutex);
1149 	nr = __smp_rescan_cpus(info, 1);
1150 	mutex_unlock(&smp_cpu_state_mutex);
1151 	put_online_cpus();
1152 	kfree(info);
1153 	if (nr)
1154 		topology_schedule_update();
1155 	return 0;
1156 }
1157 
1158 static ssize_t __ref rescan_store(struct device *dev,
1159 				  struct device_attribute *attr,
1160 				  const char *buf,
1161 				  size_t count)
1162 {
1163 	int rc;
1164 
1165 	rc = smp_rescan_cpus();
1166 	return rc ? rc : count;
1167 }
1168 static DEVICE_ATTR(rescan, 0200, NULL, rescan_store);
1169 #endif /* CONFIG_HOTPLUG_CPU */
1170 
1171 static int __init s390_smp_init(void)
1172 {
1173 	int cpu, rc = 0;
1174 
1175 #ifdef CONFIG_HOTPLUG_CPU
1176 	rc = device_create_file(cpu_subsys.dev_root, &dev_attr_rescan);
1177 	if (rc)
1178 		return rc;
1179 #endif
1180 	cpu_notifier_register_begin();
1181 	for_each_present_cpu(cpu) {
1182 		rc = smp_add_present_cpu(cpu);
1183 		if (rc)
1184 			goto out;
1185 	}
1186 
1187 	__hotcpu_notifier(smp_cpu_notify, 0);
1188 
1189 out:
1190 	cpu_notifier_register_done();
1191 	return rc;
1192 }
1193 subsys_initcall(s390_smp_init);
1194