xref: /openbmc/linux/arch/s390/kernel/smp.c (revision 976fa9a3)
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 	if (MACHINE_HAS_VX)
202 		lc->vector_save_area_addr =
203 			(unsigned long) &lc->vector_save_area;
204 	if (vdso_alloc_per_cpu(lc))
205 		goto out;
206 	lowcore_ptr[cpu] = lc;
207 	pcpu_sigp_retry(pcpu, SIGP_SET_PREFIX, (u32)(unsigned long) lc);
208 	return 0;
209 out:
210 	if (pcpu != &pcpu_devices[0]) {
211 		free_page(panic_stack);
212 		free_pages(async_stack, ASYNC_ORDER);
213 		free_pages((unsigned long) pcpu->lowcore, LC_ORDER);
214 	}
215 	return -ENOMEM;
216 }
217 
218 #ifdef CONFIG_HOTPLUG_CPU
219 
220 static void pcpu_free_lowcore(struct pcpu *pcpu)
221 {
222 	pcpu_sigp_retry(pcpu, SIGP_SET_PREFIX, 0);
223 	lowcore_ptr[pcpu - pcpu_devices] = NULL;
224 	vdso_free_per_cpu(pcpu->lowcore);
225 	if (pcpu == &pcpu_devices[0])
226 		return;
227 	free_page(pcpu->lowcore->panic_stack-PANIC_FRAME_OFFSET);
228 	free_pages(pcpu->lowcore->async_stack-ASYNC_FRAME_OFFSET, ASYNC_ORDER);
229 	free_pages((unsigned long) pcpu->lowcore, LC_ORDER);
230 }
231 
232 #endif /* CONFIG_HOTPLUG_CPU */
233 
234 static void pcpu_prepare_secondary(struct pcpu *pcpu, int cpu)
235 {
236 	struct _lowcore *lc = pcpu->lowcore;
237 
238 	if (MACHINE_HAS_TLB_LC)
239 		cpumask_set_cpu(cpu, &init_mm.context.cpu_attach_mask);
240 	cpumask_set_cpu(cpu, mm_cpumask(&init_mm));
241 	atomic_inc(&init_mm.context.attach_count);
242 	lc->cpu_nr = cpu;
243 	lc->spinlock_lockval = arch_spin_lockval(cpu);
244 	lc->percpu_offset = __per_cpu_offset[cpu];
245 	lc->kernel_asce = S390_lowcore.kernel_asce;
246 	lc->machine_flags = S390_lowcore.machine_flags;
247 	lc->user_timer = lc->system_timer = lc->steal_timer = 0;
248 	__ctl_store(lc->cregs_save_area, 0, 15);
249 	save_access_regs((unsigned int *) lc->access_regs_save_area);
250 	memcpy(lc->stfle_fac_list, S390_lowcore.stfle_fac_list,
251 	       MAX_FACILITY_BIT/8);
252 }
253 
254 static void pcpu_attach_task(struct pcpu *pcpu, struct task_struct *tsk)
255 {
256 	struct _lowcore *lc = pcpu->lowcore;
257 	struct thread_info *ti = task_thread_info(tsk);
258 
259 	lc->kernel_stack = (unsigned long) task_stack_page(tsk)
260 		+ THREAD_SIZE - STACK_FRAME_OVERHEAD - sizeof(struct pt_regs);
261 	lc->thread_info = (unsigned long) task_thread_info(tsk);
262 	lc->current_task = (unsigned long) tsk;
263 	lc->user_timer = ti->user_timer;
264 	lc->system_timer = ti->system_timer;
265 	lc->steal_timer = 0;
266 }
267 
268 static void pcpu_start_fn(struct pcpu *pcpu, void (*func)(void *), void *data)
269 {
270 	struct _lowcore *lc = pcpu->lowcore;
271 
272 	lc->restart_stack = lc->kernel_stack;
273 	lc->restart_fn = (unsigned long) func;
274 	lc->restart_data = (unsigned long) data;
275 	lc->restart_source = -1UL;
276 	pcpu_sigp_retry(pcpu, SIGP_RESTART, 0);
277 }
278 
279 /*
280  * Call function via PSW restart on pcpu and stop the current cpu.
281  */
282 static void pcpu_delegate(struct pcpu *pcpu, void (*func)(void *),
283 			  void *data, unsigned long stack)
284 {
285 	struct _lowcore *lc = lowcore_ptr[pcpu - pcpu_devices];
286 	unsigned long source_cpu = stap();
287 
288 	__load_psw_mask(PSW_KERNEL_BITS);
289 	if (pcpu->address == source_cpu)
290 		func(data);	/* should not return */
291 	/* Stop target cpu (if func returns this stops the current cpu). */
292 	pcpu_sigp_retry(pcpu, SIGP_STOP, 0);
293 	/* Restart func on the target cpu and stop the current cpu. */
294 	mem_assign_absolute(lc->restart_stack, stack);
295 	mem_assign_absolute(lc->restart_fn, (unsigned long) func);
296 	mem_assign_absolute(lc->restart_data, (unsigned long) data);
297 	mem_assign_absolute(lc->restart_source, source_cpu);
298 	asm volatile(
299 		"0:	sigp	0,%0,%2	# sigp restart to target cpu\n"
300 		"	brc	2,0b	# busy, try again\n"
301 		"1:	sigp	0,%1,%3	# sigp stop to current cpu\n"
302 		"	brc	2,1b	# busy, try again\n"
303 		: : "d" (pcpu->address), "d" (source_cpu),
304 		    "K" (SIGP_RESTART), "K" (SIGP_STOP)
305 		: "0", "1", "cc");
306 	for (;;) ;
307 }
308 
309 /*
310  * Enable additional logical cpus for multi-threading.
311  */
312 static int pcpu_set_smt(unsigned int mtid)
313 {
314 	register unsigned long reg1 asm ("1") = (unsigned long) mtid;
315 	int cc;
316 
317 	if (smp_cpu_mtid == mtid)
318 		return 0;
319 	asm volatile(
320 		"	sigp	%1,0,%2	# sigp set multi-threading\n"
321 		"	ipm	%0\n"
322 		"	srl	%0,28\n"
323 		: "=d" (cc) : "d" (reg1), "K" (SIGP_SET_MULTI_THREADING)
324 		: "cc");
325 	if (cc == 0) {
326 		smp_cpu_mtid = mtid;
327 		smp_cpu_mt_shift = 0;
328 		while (smp_cpu_mtid >= (1U << smp_cpu_mt_shift))
329 			smp_cpu_mt_shift++;
330 		pcpu_devices[0].address = stap();
331 	}
332 	return cc;
333 }
334 
335 /*
336  * Call function on an online CPU.
337  */
338 void smp_call_online_cpu(void (*func)(void *), void *data)
339 {
340 	struct pcpu *pcpu;
341 
342 	/* Use the current cpu if it is online. */
343 	pcpu = pcpu_find_address(cpu_online_mask, stap());
344 	if (!pcpu)
345 		/* Use the first online cpu. */
346 		pcpu = pcpu_devices + cpumask_first(cpu_online_mask);
347 	pcpu_delegate(pcpu, func, data, (unsigned long) restart_stack);
348 }
349 
350 /*
351  * Call function on the ipl CPU.
352  */
353 void smp_call_ipl_cpu(void (*func)(void *), void *data)
354 {
355 	pcpu_delegate(&pcpu_devices[0], func, data,
356 		      pcpu_devices->lowcore->panic_stack -
357 		      PANIC_FRAME_OFFSET + PAGE_SIZE);
358 }
359 
360 int smp_find_processor_id(u16 address)
361 {
362 	int cpu;
363 
364 	for_each_present_cpu(cpu)
365 		if (pcpu_devices[cpu].address == address)
366 			return cpu;
367 	return -1;
368 }
369 
370 int smp_vcpu_scheduled(int cpu)
371 {
372 	return pcpu_running(pcpu_devices + cpu);
373 }
374 
375 void smp_yield_cpu(int cpu)
376 {
377 	if (MACHINE_HAS_DIAG9C)
378 		asm volatile("diag %0,0,0x9c"
379 			     : : "d" (pcpu_devices[cpu].address));
380 	else if (MACHINE_HAS_DIAG44)
381 		asm volatile("diag 0,0,0x44");
382 }
383 
384 /*
385  * Send cpus emergency shutdown signal. This gives the cpus the
386  * opportunity to complete outstanding interrupts.
387  */
388 static void smp_emergency_stop(cpumask_t *cpumask)
389 {
390 	u64 end;
391 	int cpu;
392 
393 	end = get_tod_clock() + (1000000UL << 12);
394 	for_each_cpu(cpu, cpumask) {
395 		struct pcpu *pcpu = pcpu_devices + cpu;
396 		set_bit(ec_stop_cpu, &pcpu->ec_mask);
397 		while (__pcpu_sigp(pcpu->address, SIGP_EMERGENCY_SIGNAL,
398 				   0, NULL) == SIGP_CC_BUSY &&
399 		       get_tod_clock() < end)
400 			cpu_relax();
401 	}
402 	while (get_tod_clock() < end) {
403 		for_each_cpu(cpu, cpumask)
404 			if (pcpu_stopped(pcpu_devices + cpu))
405 				cpumask_clear_cpu(cpu, cpumask);
406 		if (cpumask_empty(cpumask))
407 			break;
408 		cpu_relax();
409 	}
410 }
411 
412 /*
413  * Stop all cpus but the current one.
414  */
415 void smp_send_stop(void)
416 {
417 	cpumask_t cpumask;
418 	int cpu;
419 
420 	/* Disable all interrupts/machine checks */
421 	__load_psw_mask(PSW_KERNEL_BITS | PSW_MASK_DAT);
422 	trace_hardirqs_off();
423 
424 	debug_set_critical();
425 	cpumask_copy(&cpumask, cpu_online_mask);
426 	cpumask_clear_cpu(smp_processor_id(), &cpumask);
427 
428 	if (oops_in_progress)
429 		smp_emergency_stop(&cpumask);
430 
431 	/* stop all processors */
432 	for_each_cpu(cpu, &cpumask) {
433 		struct pcpu *pcpu = pcpu_devices + cpu;
434 		pcpu_sigp_retry(pcpu, SIGP_STOP, 0);
435 		while (!pcpu_stopped(pcpu))
436 			cpu_relax();
437 	}
438 }
439 
440 /*
441  * This is the main routine where commands issued by other
442  * cpus are handled.
443  */
444 static void smp_handle_ext_call(void)
445 {
446 	unsigned long bits;
447 
448 	/* handle bit signal external calls */
449 	bits = xchg(&pcpu_devices[smp_processor_id()].ec_mask, 0);
450 	if (test_bit(ec_stop_cpu, &bits))
451 		smp_stop_cpu();
452 	if (test_bit(ec_schedule, &bits))
453 		scheduler_ipi();
454 	if (test_bit(ec_call_function_single, &bits))
455 		generic_smp_call_function_single_interrupt();
456 }
457 
458 static void do_ext_call_interrupt(struct ext_code ext_code,
459 				  unsigned int param32, unsigned long param64)
460 {
461 	inc_irq_stat(ext_code.code == 0x1202 ? IRQEXT_EXC : IRQEXT_EMS);
462 	smp_handle_ext_call();
463 }
464 
465 void arch_send_call_function_ipi_mask(const struct cpumask *mask)
466 {
467 	int cpu;
468 
469 	for_each_cpu(cpu, mask)
470 		pcpu_ec_call(pcpu_devices + cpu, ec_call_function_single);
471 }
472 
473 void arch_send_call_function_single_ipi(int cpu)
474 {
475 	pcpu_ec_call(pcpu_devices + cpu, ec_call_function_single);
476 }
477 
478 /*
479  * this function sends a 'reschedule' IPI to another CPU.
480  * it goes straight through and wastes no time serializing
481  * anything. Worst case is that we lose a reschedule ...
482  */
483 void smp_send_reschedule(int cpu)
484 {
485 	pcpu_ec_call(pcpu_devices + cpu, ec_schedule);
486 }
487 
488 /*
489  * parameter area for the set/clear control bit callbacks
490  */
491 struct ec_creg_mask_parms {
492 	unsigned long orval;
493 	unsigned long andval;
494 	int cr;
495 };
496 
497 /*
498  * callback for setting/clearing control bits
499  */
500 static void smp_ctl_bit_callback(void *info)
501 {
502 	struct ec_creg_mask_parms *pp = info;
503 	unsigned long cregs[16];
504 
505 	__ctl_store(cregs, 0, 15);
506 	cregs[pp->cr] = (cregs[pp->cr] & pp->andval) | pp->orval;
507 	__ctl_load(cregs, 0, 15);
508 }
509 
510 /*
511  * Set a bit in a control register of all cpus
512  */
513 void smp_ctl_set_bit(int cr, int bit)
514 {
515 	struct ec_creg_mask_parms parms = { 1UL << bit, -1UL, cr };
516 
517 	on_each_cpu(smp_ctl_bit_callback, &parms, 1);
518 }
519 EXPORT_SYMBOL(smp_ctl_set_bit);
520 
521 /*
522  * Clear a bit in a control register of all cpus
523  */
524 void smp_ctl_clear_bit(int cr, int bit)
525 {
526 	struct ec_creg_mask_parms parms = { 0, ~(1UL << bit), cr };
527 
528 	on_each_cpu(smp_ctl_bit_callback, &parms, 1);
529 }
530 EXPORT_SYMBOL(smp_ctl_clear_bit);
531 
532 #ifdef CONFIG_CRASH_DUMP
533 
534 static inline void __smp_store_cpu_state(int cpu, u16 address, int is_boot_cpu)
535 {
536 	void *lc = pcpu_devices[0].lowcore;
537 	struct save_area_ext *sa_ext;
538 	unsigned long vx_sa;
539 
540 	sa_ext = dump_save_area_create(cpu);
541 	if (!sa_ext)
542 		panic("could not allocate memory for save area\n");
543 	if (is_boot_cpu) {
544 		/* Copy the registers of the boot CPU. */
545 		copy_oldmem_page(1, (void *) &sa_ext->sa, sizeof(sa_ext->sa),
546 				 SAVE_AREA_BASE - PAGE_SIZE, 0);
547 		if (MACHINE_HAS_VX)
548 			save_vx_regs_safe(sa_ext->vx_regs);
549 		return;
550 	}
551 	/* Get the registers of a non-boot cpu. */
552 	__pcpu_sigp_relax(address, SIGP_STOP_AND_STORE_STATUS, 0, NULL);
553 	memcpy_real(&sa_ext->sa, lc + SAVE_AREA_BASE, sizeof(sa_ext->sa));
554 	if (!MACHINE_HAS_VX)
555 		return;
556 	/* Get the VX registers */
557 	vx_sa = __get_free_page(GFP_KERNEL);
558 	if (!vx_sa)
559 		panic("could not allocate memory for VX save area\n");
560 	__pcpu_sigp_relax(address, SIGP_STORE_ADDITIONAL_STATUS, vx_sa, NULL);
561 	memcpy(sa_ext->vx_regs, (void *) vx_sa, sizeof(sa_ext->vx_regs));
562 	free_page(vx_sa);
563 }
564 
565 /*
566  * Collect CPU state of the previous, crashed system.
567  * There are four cases:
568  * 1) standard zfcp dump
569  *    condition: OLDMEM_BASE == NULL && ipl_info.type == IPL_TYPE_FCP_DUMP
570  *    The state for all CPUs except the boot CPU needs to be collected
571  *    with sigp stop-and-store-status. The boot CPU state is located in
572  *    the absolute lowcore of the memory stored in the HSA. The zcore code
573  *    will allocate the save area and copy the boot CPU state from the HSA.
574  * 2) stand-alone kdump for SCSI (zfcp dump with swapped memory)
575  *    condition: OLDMEM_BASE != NULL && ipl_info.type == IPL_TYPE_FCP_DUMP
576  *    The state for all CPUs except the boot CPU needs to be collected
577  *    with sigp stop-and-store-status. The firmware or the boot-loader
578  *    stored the registers of the boot CPU in the absolute lowcore in the
579  *    memory of the old system.
580  * 3) kdump and the old kernel did not store the CPU state,
581  *    or stand-alone kdump for DASD
582  *    condition: OLDMEM_BASE != NULL && !is_kdump_kernel()
583  *    The state for all CPUs except the boot CPU needs to be collected
584  *    with sigp stop-and-store-status. The kexec code or the boot-loader
585  *    stored the registers of the boot CPU in the memory of the old system.
586  * 4) kdump and the old kernel stored the CPU state
587  *    condition: OLDMEM_BASE != NULL && is_kdump_kernel()
588  *    The state of all CPUs is stored in ELF sections in the memory of the
589  *    old system. The ELF sections are picked up by the crash_dump code
590  *    via elfcorehdr_addr.
591  */
592 static void __init smp_store_cpu_states(struct sclp_cpu_info *info)
593 {
594 	unsigned int cpu, address, i, j;
595 	int is_boot_cpu;
596 
597 	if (is_kdump_kernel())
598 		/* Previous system stored the CPU states. Nothing to do. */
599 		return;
600 	if (!(OLDMEM_BASE || ipl_info.type == IPL_TYPE_FCP_DUMP))
601 		/* No previous system present, normal boot. */
602 		return;
603 	/* Set multi-threading state to the previous system. */
604 	pcpu_set_smt(sclp_get_mtid_prev());
605 	/* Collect CPU states. */
606 	cpu = 0;
607 	for (i = 0; i < info->configured; i++) {
608 		/* Skip CPUs with different CPU type. */
609 		if (info->has_cpu_type && info->cpu[i].type != boot_cpu_type)
610 			continue;
611 		for (j = 0; j <= smp_cpu_mtid; j++, cpu++) {
612 			address = (info->cpu[i].core_id << smp_cpu_mt_shift) + j;
613 			is_boot_cpu = (address == pcpu_devices[0].address);
614 			if (is_boot_cpu && !OLDMEM_BASE)
615 				/* Skip boot CPU for standard zfcp dump. */
616 				continue;
617 			/* Get state for this CPu. */
618 			__smp_store_cpu_state(cpu, address, is_boot_cpu);
619 		}
620 	}
621 }
622 
623 int smp_store_status(int cpu)
624 {
625 	unsigned long vx_sa;
626 	struct pcpu *pcpu;
627 
628 	pcpu = pcpu_devices + cpu;
629 	if (__pcpu_sigp_relax(pcpu->address, SIGP_STOP_AND_STORE_STATUS,
630 			      0, NULL) != SIGP_CC_ORDER_CODE_ACCEPTED)
631 		return -EIO;
632 	if (!MACHINE_HAS_VX)
633 		return 0;
634 	vx_sa = __pa(pcpu->lowcore->vector_save_area_addr);
635 	__pcpu_sigp_relax(pcpu->address, SIGP_STORE_ADDITIONAL_STATUS,
636 			  vx_sa, NULL);
637 	return 0;
638 }
639 
640 #endif /* CONFIG_CRASH_DUMP */
641 
642 void smp_cpu_set_polarization(int cpu, int val)
643 {
644 	pcpu_devices[cpu].polarization = val;
645 }
646 
647 int smp_cpu_get_polarization(int cpu)
648 {
649 	return pcpu_devices[cpu].polarization;
650 }
651 
652 static struct sclp_cpu_info *smp_get_cpu_info(void)
653 {
654 	static int use_sigp_detection;
655 	struct sclp_cpu_info *info;
656 	int address;
657 
658 	info = kzalloc(sizeof(*info), GFP_KERNEL);
659 	if (info && (use_sigp_detection || sclp_get_cpu_info(info))) {
660 		use_sigp_detection = 1;
661 		for (address = 0; address <= MAX_CPU_ADDRESS;
662 		     address += (1U << smp_cpu_mt_shift)) {
663 			if (__pcpu_sigp_relax(address, SIGP_SENSE, 0, NULL) ==
664 			    SIGP_CC_NOT_OPERATIONAL)
665 				continue;
666 			info->cpu[info->configured].core_id =
667 				address >> smp_cpu_mt_shift;
668 			info->configured++;
669 		}
670 		info->combined = info->configured;
671 	}
672 	return info;
673 }
674 
675 static int smp_add_present_cpu(int cpu);
676 
677 static int __smp_rescan_cpus(struct sclp_cpu_info *info, int sysfs_add)
678 {
679 	struct pcpu *pcpu;
680 	cpumask_t avail;
681 	int cpu, nr, i, j;
682 	u16 address;
683 
684 	nr = 0;
685 	cpumask_xor(&avail, cpu_possible_mask, cpu_present_mask);
686 	cpu = cpumask_first(&avail);
687 	for (i = 0; (i < info->combined) && (cpu < nr_cpu_ids); i++) {
688 		if (info->has_cpu_type && info->cpu[i].type != boot_cpu_type)
689 			continue;
690 		address = info->cpu[i].core_id << smp_cpu_mt_shift;
691 		for (j = 0; j <= smp_cpu_mtid; j++) {
692 			if (pcpu_find_address(cpu_present_mask, address + j))
693 				continue;
694 			pcpu = pcpu_devices + cpu;
695 			pcpu->address = address + j;
696 			pcpu->state =
697 				(cpu >= info->configured*(smp_cpu_mtid + 1)) ?
698 				CPU_STATE_STANDBY : CPU_STATE_CONFIGURED;
699 			smp_cpu_set_polarization(cpu, POLARIZATION_UNKNOWN);
700 			set_cpu_present(cpu, true);
701 			if (sysfs_add && smp_add_present_cpu(cpu) != 0)
702 				set_cpu_present(cpu, false);
703 			else
704 				nr++;
705 			cpu = cpumask_next(cpu, &avail);
706 			if (cpu >= nr_cpu_ids)
707 				break;
708 		}
709 	}
710 	return nr;
711 }
712 
713 static void __init smp_detect_cpus(void)
714 {
715 	unsigned int cpu, mtid, c_cpus, s_cpus;
716 	struct sclp_cpu_info *info;
717 	u16 address;
718 
719 	/* Get CPU information */
720 	info = smp_get_cpu_info();
721 	if (!info)
722 		panic("smp_detect_cpus failed to allocate memory\n");
723 
724 	/* Find boot CPU type */
725 	if (info->has_cpu_type) {
726 		address = stap();
727 		for (cpu = 0; cpu < info->combined; cpu++)
728 			if (info->cpu[cpu].core_id == address) {
729 				/* The boot cpu dictates the cpu type. */
730 				boot_cpu_type = info->cpu[cpu].type;
731 				break;
732 			}
733 		if (cpu >= info->combined)
734 			panic("Could not find boot CPU type");
735 	}
736 
737 #ifdef CONFIG_CRASH_DUMP
738 	/* Collect CPU state of previous system */
739 	smp_store_cpu_states(info);
740 #endif
741 
742 	/* Set multi-threading state for the current system */
743 	mtid = sclp_get_mtid(boot_cpu_type);
744 	mtid = (mtid < smp_max_threads) ? mtid : smp_max_threads - 1;
745 	pcpu_set_smt(mtid);
746 
747 	/* Print number of CPUs */
748 	c_cpus = s_cpus = 0;
749 	for (cpu = 0; cpu < info->combined; cpu++) {
750 		if (info->has_cpu_type && info->cpu[cpu].type != boot_cpu_type)
751 			continue;
752 		if (cpu < info->configured)
753 			c_cpus += smp_cpu_mtid + 1;
754 		else
755 			s_cpus += smp_cpu_mtid + 1;
756 	}
757 	pr_info("%d configured CPUs, %d standby CPUs\n", c_cpus, s_cpus);
758 
759 	/* Add CPUs present at boot */
760 	get_online_cpus();
761 	__smp_rescan_cpus(info, 0);
762 	put_online_cpus();
763 	kfree(info);
764 }
765 
766 /*
767  *	Activate a secondary processor.
768  */
769 static void smp_start_secondary(void *cpuvoid)
770 {
771 	S390_lowcore.last_update_clock = get_tod_clock();
772 	S390_lowcore.restart_stack = (unsigned long) restart_stack;
773 	S390_lowcore.restart_fn = (unsigned long) do_restart;
774 	S390_lowcore.restart_data = 0;
775 	S390_lowcore.restart_source = -1UL;
776 	restore_access_regs(S390_lowcore.access_regs_save_area);
777 	__ctl_load(S390_lowcore.cregs_save_area, 0, 15);
778 	__load_psw_mask(PSW_KERNEL_BITS | PSW_MASK_DAT);
779 	cpu_init();
780 	preempt_disable();
781 	init_cpu_timer();
782 	vtime_init();
783 	pfault_init();
784 	notify_cpu_starting(smp_processor_id());
785 	set_cpu_online(smp_processor_id(), true);
786 	inc_irq_stat(CPU_RST);
787 	local_irq_enable();
788 	cpu_startup_entry(CPUHP_ONLINE);
789 }
790 
791 /* Upping and downing of CPUs */
792 int __cpu_up(unsigned int cpu, struct task_struct *tidle)
793 {
794 	struct pcpu *pcpu;
795 	int base, i, rc;
796 
797 	pcpu = pcpu_devices + cpu;
798 	if (pcpu->state != CPU_STATE_CONFIGURED)
799 		return -EIO;
800 	base = cpu - (cpu % (smp_cpu_mtid + 1));
801 	for (i = 0; i <= smp_cpu_mtid; i++) {
802 		if (base + i < nr_cpu_ids)
803 			if (cpu_online(base + i))
804 				break;
805 	}
806 	/*
807 	 * If this is the first CPU of the core to get online
808 	 * do an initial CPU reset.
809 	 */
810 	if (i > smp_cpu_mtid &&
811 	    pcpu_sigp_retry(pcpu_devices + base, SIGP_INITIAL_CPU_RESET, 0) !=
812 	    SIGP_CC_ORDER_CODE_ACCEPTED)
813 		return -EIO;
814 
815 	rc = pcpu_alloc_lowcore(pcpu, cpu);
816 	if (rc)
817 		return rc;
818 	pcpu_prepare_secondary(pcpu, cpu);
819 	pcpu_attach_task(pcpu, tidle);
820 	pcpu_start_fn(pcpu, smp_start_secondary, NULL);
821 	/* Wait until cpu puts itself in the online & active maps */
822 	while (!cpu_online(cpu) || !cpu_active(cpu))
823 		cpu_relax();
824 	return 0;
825 }
826 
827 static unsigned int setup_possible_cpus __initdata;
828 
829 static int __init _setup_possible_cpus(char *s)
830 {
831 	get_option(&s, &setup_possible_cpus);
832 	return 0;
833 }
834 early_param("possible_cpus", _setup_possible_cpus);
835 
836 #ifdef CONFIG_HOTPLUG_CPU
837 
838 int __cpu_disable(void)
839 {
840 	unsigned long cregs[16];
841 
842 	/* Handle possible pending IPIs */
843 	smp_handle_ext_call();
844 	set_cpu_online(smp_processor_id(), false);
845 	/* Disable pseudo page faults on this cpu. */
846 	pfault_fini();
847 	/* Disable interrupt sources via control register. */
848 	__ctl_store(cregs, 0, 15);
849 	cregs[0]  &= ~0x0000ee70UL;	/* disable all external interrupts */
850 	cregs[6]  &= ~0xff000000UL;	/* disable all I/O interrupts */
851 	cregs[14] &= ~0x1f000000UL;	/* disable most machine checks */
852 	__ctl_load(cregs, 0, 15);
853 	clear_cpu_flag(CIF_NOHZ_DELAY);
854 	return 0;
855 }
856 
857 void __cpu_die(unsigned int cpu)
858 {
859 	struct pcpu *pcpu;
860 
861 	/* Wait until target cpu is down */
862 	pcpu = pcpu_devices + cpu;
863 	while (!pcpu_stopped(pcpu))
864 		cpu_relax();
865 	pcpu_free_lowcore(pcpu);
866 	atomic_dec(&init_mm.context.attach_count);
867 	cpumask_clear_cpu(cpu, mm_cpumask(&init_mm));
868 	if (MACHINE_HAS_TLB_LC)
869 		cpumask_clear_cpu(cpu, &init_mm.context.cpu_attach_mask);
870 }
871 
872 void __noreturn cpu_die(void)
873 {
874 	idle_task_exit();
875 	pcpu_sigp_retry(pcpu_devices + smp_processor_id(), SIGP_STOP, 0);
876 	for (;;) ;
877 }
878 
879 #endif /* CONFIG_HOTPLUG_CPU */
880 
881 void __init smp_fill_possible_mask(void)
882 {
883 	unsigned int possible, sclp, cpu;
884 
885 	sclp = min(smp_max_threads, sclp_get_mtid_max() + 1);
886 	sclp = sclp_get_max_cpu()*sclp ?: nr_cpu_ids;
887 	possible = setup_possible_cpus ?: nr_cpu_ids;
888 	possible = min(possible, sclp);
889 	for (cpu = 0; cpu < possible && cpu < nr_cpu_ids; cpu++)
890 		set_cpu_possible(cpu, true);
891 }
892 
893 void __init smp_prepare_cpus(unsigned int max_cpus)
894 {
895 	/* request the 0x1201 emergency signal external interrupt */
896 	if (register_external_irq(EXT_IRQ_EMERGENCY_SIG, do_ext_call_interrupt))
897 		panic("Couldn't request external interrupt 0x1201");
898 	/* request the 0x1202 external call external interrupt */
899 	if (register_external_irq(EXT_IRQ_EXTERNAL_CALL, do_ext_call_interrupt))
900 		panic("Couldn't request external interrupt 0x1202");
901 	smp_detect_cpus();
902 }
903 
904 void __init smp_prepare_boot_cpu(void)
905 {
906 	struct pcpu *pcpu = pcpu_devices;
907 
908 	pcpu->state = CPU_STATE_CONFIGURED;
909 	pcpu->address = stap();
910 	pcpu->lowcore = (struct _lowcore *)(unsigned long) store_prefix();
911 	S390_lowcore.percpu_offset = __per_cpu_offset[0];
912 	smp_cpu_set_polarization(0, POLARIZATION_UNKNOWN);
913 	set_cpu_present(0, true);
914 	set_cpu_online(0, true);
915 }
916 
917 void __init smp_cpus_done(unsigned int max_cpus)
918 {
919 }
920 
921 void __init smp_setup_processor_id(void)
922 {
923 	S390_lowcore.cpu_nr = 0;
924 	S390_lowcore.spinlock_lockval = arch_spin_lockval(0);
925 }
926 
927 /*
928  * the frequency of the profiling timer can be changed
929  * by writing a multiplier value into /proc/profile.
930  *
931  * usually you want to run this on all CPUs ;)
932  */
933 int setup_profiling_timer(unsigned int multiplier)
934 {
935 	return 0;
936 }
937 
938 #ifdef CONFIG_HOTPLUG_CPU
939 static ssize_t cpu_configure_show(struct device *dev,
940 				  struct device_attribute *attr, char *buf)
941 {
942 	ssize_t count;
943 
944 	mutex_lock(&smp_cpu_state_mutex);
945 	count = sprintf(buf, "%d\n", pcpu_devices[dev->id].state);
946 	mutex_unlock(&smp_cpu_state_mutex);
947 	return count;
948 }
949 
950 static ssize_t cpu_configure_store(struct device *dev,
951 				   struct device_attribute *attr,
952 				   const char *buf, size_t count)
953 {
954 	struct pcpu *pcpu;
955 	int cpu, val, rc, i;
956 	char delim;
957 
958 	if (sscanf(buf, "%d %c", &val, &delim) != 1)
959 		return -EINVAL;
960 	if (val != 0 && val != 1)
961 		return -EINVAL;
962 	get_online_cpus();
963 	mutex_lock(&smp_cpu_state_mutex);
964 	rc = -EBUSY;
965 	/* disallow configuration changes of online cpus and cpu 0 */
966 	cpu = dev->id;
967 	cpu -= cpu % (smp_cpu_mtid + 1);
968 	if (cpu == 0)
969 		goto out;
970 	for (i = 0; i <= smp_cpu_mtid; i++)
971 		if (cpu_online(cpu + i))
972 			goto out;
973 	pcpu = pcpu_devices + cpu;
974 	rc = 0;
975 	switch (val) {
976 	case 0:
977 		if (pcpu->state != CPU_STATE_CONFIGURED)
978 			break;
979 		rc = sclp_cpu_deconfigure(pcpu->address >> smp_cpu_mt_shift);
980 		if (rc)
981 			break;
982 		for (i = 0; i <= smp_cpu_mtid; i++) {
983 			if (cpu + i >= nr_cpu_ids || !cpu_present(cpu + i))
984 				continue;
985 			pcpu[i].state = CPU_STATE_STANDBY;
986 			smp_cpu_set_polarization(cpu + i,
987 						 POLARIZATION_UNKNOWN);
988 		}
989 		topology_expect_change();
990 		break;
991 	case 1:
992 		if (pcpu->state != CPU_STATE_STANDBY)
993 			break;
994 		rc = sclp_cpu_configure(pcpu->address >> smp_cpu_mt_shift);
995 		if (rc)
996 			break;
997 		for (i = 0; i <= smp_cpu_mtid; i++) {
998 			if (cpu + i >= nr_cpu_ids || !cpu_present(cpu + i))
999 				continue;
1000 			pcpu[i].state = CPU_STATE_CONFIGURED;
1001 			smp_cpu_set_polarization(cpu + i,
1002 						 POLARIZATION_UNKNOWN);
1003 		}
1004 		topology_expect_change();
1005 		break;
1006 	default:
1007 		break;
1008 	}
1009 out:
1010 	mutex_unlock(&smp_cpu_state_mutex);
1011 	put_online_cpus();
1012 	return rc ? rc : count;
1013 }
1014 static DEVICE_ATTR(configure, 0644, cpu_configure_show, cpu_configure_store);
1015 #endif /* CONFIG_HOTPLUG_CPU */
1016 
1017 static ssize_t show_cpu_address(struct device *dev,
1018 				struct device_attribute *attr, char *buf)
1019 {
1020 	return sprintf(buf, "%d\n", pcpu_devices[dev->id].address);
1021 }
1022 static DEVICE_ATTR(address, 0444, show_cpu_address, NULL);
1023 
1024 static struct attribute *cpu_common_attrs[] = {
1025 #ifdef CONFIG_HOTPLUG_CPU
1026 	&dev_attr_configure.attr,
1027 #endif
1028 	&dev_attr_address.attr,
1029 	NULL,
1030 };
1031 
1032 static struct attribute_group cpu_common_attr_group = {
1033 	.attrs = cpu_common_attrs,
1034 };
1035 
1036 static struct attribute *cpu_online_attrs[] = {
1037 	&dev_attr_idle_count.attr,
1038 	&dev_attr_idle_time_us.attr,
1039 	NULL,
1040 };
1041 
1042 static struct attribute_group cpu_online_attr_group = {
1043 	.attrs = cpu_online_attrs,
1044 };
1045 
1046 static int smp_cpu_notify(struct notifier_block *self, unsigned long action,
1047 			  void *hcpu)
1048 {
1049 	unsigned int cpu = (unsigned int)(long)hcpu;
1050 	struct device *s = &per_cpu(cpu_device, cpu)->dev;
1051 	int err = 0;
1052 
1053 	switch (action & ~CPU_TASKS_FROZEN) {
1054 	case CPU_ONLINE:
1055 		err = sysfs_create_group(&s->kobj, &cpu_online_attr_group);
1056 		break;
1057 	case CPU_DEAD:
1058 		sysfs_remove_group(&s->kobj, &cpu_online_attr_group);
1059 		break;
1060 	}
1061 	return notifier_from_errno(err);
1062 }
1063 
1064 static int smp_add_present_cpu(int cpu)
1065 {
1066 	struct device *s;
1067 	struct cpu *c;
1068 	int rc;
1069 
1070 	c = kzalloc(sizeof(*c), GFP_KERNEL);
1071 	if (!c)
1072 		return -ENOMEM;
1073 	per_cpu(cpu_device, cpu) = c;
1074 	s = &c->dev;
1075 	c->hotpluggable = 1;
1076 	rc = register_cpu(c, cpu);
1077 	if (rc)
1078 		goto out;
1079 	rc = sysfs_create_group(&s->kobj, &cpu_common_attr_group);
1080 	if (rc)
1081 		goto out_cpu;
1082 	if (cpu_online(cpu)) {
1083 		rc = sysfs_create_group(&s->kobj, &cpu_online_attr_group);
1084 		if (rc)
1085 			goto out_online;
1086 	}
1087 	rc = topology_cpu_init(c);
1088 	if (rc)
1089 		goto out_topology;
1090 	return 0;
1091 
1092 out_topology:
1093 	if (cpu_online(cpu))
1094 		sysfs_remove_group(&s->kobj, &cpu_online_attr_group);
1095 out_online:
1096 	sysfs_remove_group(&s->kobj, &cpu_common_attr_group);
1097 out_cpu:
1098 #ifdef CONFIG_HOTPLUG_CPU
1099 	unregister_cpu(c);
1100 #endif
1101 out:
1102 	return rc;
1103 }
1104 
1105 #ifdef CONFIG_HOTPLUG_CPU
1106 
1107 int __ref smp_rescan_cpus(void)
1108 {
1109 	struct sclp_cpu_info *info;
1110 	int nr;
1111 
1112 	info = smp_get_cpu_info();
1113 	if (!info)
1114 		return -ENOMEM;
1115 	get_online_cpus();
1116 	mutex_lock(&smp_cpu_state_mutex);
1117 	nr = __smp_rescan_cpus(info, 1);
1118 	mutex_unlock(&smp_cpu_state_mutex);
1119 	put_online_cpus();
1120 	kfree(info);
1121 	if (nr)
1122 		topology_schedule_update();
1123 	return 0;
1124 }
1125 
1126 static ssize_t __ref rescan_store(struct device *dev,
1127 				  struct device_attribute *attr,
1128 				  const char *buf,
1129 				  size_t count)
1130 {
1131 	int rc;
1132 
1133 	rc = smp_rescan_cpus();
1134 	return rc ? rc : count;
1135 }
1136 static DEVICE_ATTR(rescan, 0200, NULL, rescan_store);
1137 #endif /* CONFIG_HOTPLUG_CPU */
1138 
1139 static int __init s390_smp_init(void)
1140 {
1141 	int cpu, rc = 0;
1142 
1143 #ifdef CONFIG_HOTPLUG_CPU
1144 	rc = device_create_file(cpu_subsys.dev_root, &dev_attr_rescan);
1145 	if (rc)
1146 		return rc;
1147 #endif
1148 	cpu_notifier_register_begin();
1149 	for_each_present_cpu(cpu) {
1150 		rc = smp_add_present_cpu(cpu);
1151 		if (rc)
1152 			goto out;
1153 	}
1154 
1155 	__hotcpu_notifier(smp_cpu_notify, 0);
1156 
1157 out:
1158 	cpu_notifier_register_done();
1159 	return rc;
1160 }
1161 subsys_initcall(s390_smp_init);
1162