xref: /openbmc/linux/arch/s390/kernel/smp.c (revision a36954f5)
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/bootmem.h>
23 #include <linux/export.h>
24 #include <linux/init.h>
25 #include <linux/mm.h>
26 #include <linux/err.h>
27 #include <linux/spinlock.h>
28 #include <linux/kernel_stat.h>
29 #include <linux/delay.h>
30 #include <linux/interrupt.h>
31 #include <linux/irqflags.h>
32 #include <linux/cpu.h>
33 #include <linux/slab.h>
34 #include <linux/sched/hotplug.h>
35 #include <linux/sched/task_stack.h>
36 #include <linux/crash_dump.h>
37 #include <linux/memblock.h>
38 #include <asm/asm-offsets.h>
39 #include <asm/diag.h>
40 #include <asm/switch_to.h>
41 #include <asm/facility.h>
42 #include <asm/ipl.h>
43 #include <asm/setup.h>
44 #include <asm/irq.h>
45 #include <asm/tlbflush.h>
46 #include <asm/vtimer.h>
47 #include <asm/lowcore.h>
48 #include <asm/sclp.h>
49 #include <asm/vdso.h>
50 #include <asm/debug.h>
51 #include <asm/os_info.h>
52 #include <asm/sigp.h>
53 #include <asm/idle.h>
54 #include <asm/nmi.h>
55 #include "entry.h"
56 
57 enum {
58 	ec_schedule = 0,
59 	ec_call_function_single,
60 	ec_stop_cpu,
61 };
62 
63 enum {
64 	CPU_STATE_STANDBY,
65 	CPU_STATE_CONFIGURED,
66 };
67 
68 static DEFINE_PER_CPU(struct cpu *, cpu_device);
69 
70 struct pcpu {
71 	struct lowcore *lowcore;	/* lowcore page(s) for the cpu */
72 	unsigned long ec_mask;		/* bit mask for ec_xxx functions */
73 	unsigned long ec_clk;		/* sigp timestamp for ec_xxx */
74 	signed char state;		/* physical cpu state */
75 	signed char polarization;	/* physical polarization */
76 	u16 address;			/* physical cpu address */
77 };
78 
79 static u8 boot_core_type;
80 static struct pcpu pcpu_devices[NR_CPUS];
81 
82 static struct kmem_cache *pcpu_mcesa_cache;
83 
84 unsigned int smp_cpu_mt_shift;
85 EXPORT_SYMBOL(smp_cpu_mt_shift);
86 
87 unsigned int smp_cpu_mtid;
88 EXPORT_SYMBOL(smp_cpu_mtid);
89 
90 #ifdef CONFIG_CRASH_DUMP
91 __vector128 __initdata boot_cpu_vector_save_area[__NUM_VXRS];
92 #endif
93 
94 static unsigned int smp_max_threads __initdata = -1U;
95 
96 static int __init early_nosmt(char *s)
97 {
98 	smp_max_threads = 1;
99 	return 0;
100 }
101 early_param("nosmt", early_nosmt);
102 
103 static int __init early_smt(char *s)
104 {
105 	get_option(&s, &smp_max_threads);
106 	return 0;
107 }
108 early_param("smt", early_smt);
109 
110 /*
111  * The smp_cpu_state_mutex must be held when changing the state or polarization
112  * member of a pcpu data structure within the pcpu_devices arreay.
113  */
114 DEFINE_MUTEX(smp_cpu_state_mutex);
115 
116 /*
117  * Signal processor helper functions.
118  */
119 static inline int __pcpu_sigp_relax(u16 addr, u8 order, unsigned long parm)
120 {
121 	int cc;
122 
123 	while (1) {
124 		cc = __pcpu_sigp(addr, order, parm, NULL);
125 		if (cc != SIGP_CC_BUSY)
126 			return cc;
127 		cpu_relax();
128 	}
129 }
130 
131 static int pcpu_sigp_retry(struct pcpu *pcpu, u8 order, u32 parm)
132 {
133 	int cc, retry;
134 
135 	for (retry = 0; ; retry++) {
136 		cc = __pcpu_sigp(pcpu->address, order, parm, NULL);
137 		if (cc != SIGP_CC_BUSY)
138 			break;
139 		if (retry >= 3)
140 			udelay(10);
141 	}
142 	return cc;
143 }
144 
145 static inline int pcpu_stopped(struct pcpu *pcpu)
146 {
147 	u32 uninitialized_var(status);
148 
149 	if (__pcpu_sigp(pcpu->address, SIGP_SENSE,
150 			0, &status) != SIGP_CC_STATUS_STORED)
151 		return 0;
152 	return !!(status & (SIGP_STATUS_CHECK_STOP|SIGP_STATUS_STOPPED));
153 }
154 
155 static inline int pcpu_running(struct pcpu *pcpu)
156 {
157 	if (__pcpu_sigp(pcpu->address, SIGP_SENSE_RUNNING,
158 			0, NULL) != SIGP_CC_STATUS_STORED)
159 		return 1;
160 	/* Status stored condition code is equivalent to cpu not running. */
161 	return 0;
162 }
163 
164 /*
165  * Find struct pcpu by cpu address.
166  */
167 static struct pcpu *pcpu_find_address(const struct cpumask *mask, u16 address)
168 {
169 	int cpu;
170 
171 	for_each_cpu(cpu, mask)
172 		if (pcpu_devices[cpu].address == address)
173 			return pcpu_devices + cpu;
174 	return NULL;
175 }
176 
177 static void pcpu_ec_call(struct pcpu *pcpu, int ec_bit)
178 {
179 	int order;
180 
181 	if (test_and_set_bit(ec_bit, &pcpu->ec_mask))
182 		return;
183 	order = pcpu_running(pcpu) ? SIGP_EXTERNAL_CALL : SIGP_EMERGENCY_SIGNAL;
184 	pcpu->ec_clk = get_tod_clock_fast();
185 	pcpu_sigp_retry(pcpu, order, 0);
186 }
187 
188 #define ASYNC_FRAME_OFFSET (ASYNC_SIZE - STACK_FRAME_OVERHEAD - __PT_SIZE)
189 #define PANIC_FRAME_OFFSET (PAGE_SIZE - STACK_FRAME_OVERHEAD - __PT_SIZE)
190 
191 static int pcpu_alloc_lowcore(struct pcpu *pcpu, int cpu)
192 {
193 	unsigned long async_stack, panic_stack;
194 	unsigned long mcesa_origin, mcesa_bits;
195 	struct lowcore *lc;
196 
197 	mcesa_origin = mcesa_bits = 0;
198 	if (pcpu != &pcpu_devices[0]) {
199 		pcpu->lowcore =	(struct lowcore *)
200 			__get_free_pages(GFP_KERNEL | GFP_DMA, LC_ORDER);
201 		async_stack = __get_free_pages(GFP_KERNEL, ASYNC_ORDER);
202 		panic_stack = __get_free_page(GFP_KERNEL);
203 		if (!pcpu->lowcore || !panic_stack || !async_stack)
204 			goto out;
205 		if (MACHINE_HAS_VX || MACHINE_HAS_GS) {
206 			mcesa_origin = (unsigned long)
207 				kmem_cache_alloc(pcpu_mcesa_cache, GFP_KERNEL);
208 			if (!mcesa_origin)
209 				goto out;
210 			mcesa_bits = MACHINE_HAS_GS ? 11 : 0;
211 		}
212 	} else {
213 		async_stack = pcpu->lowcore->async_stack - ASYNC_FRAME_OFFSET;
214 		panic_stack = pcpu->lowcore->panic_stack - PANIC_FRAME_OFFSET;
215 		mcesa_origin = pcpu->lowcore->mcesad & MCESA_ORIGIN_MASK;
216 		mcesa_bits = pcpu->lowcore->mcesad & MCESA_LC_MASK;
217 	}
218 	lc = pcpu->lowcore;
219 	memcpy(lc, &S390_lowcore, 512);
220 	memset((char *) lc + 512, 0, sizeof(*lc) - 512);
221 	lc->async_stack = async_stack + ASYNC_FRAME_OFFSET;
222 	lc->panic_stack = panic_stack + PANIC_FRAME_OFFSET;
223 	lc->mcesad = mcesa_origin | mcesa_bits;
224 	lc->cpu_nr = cpu;
225 	lc->spinlock_lockval = arch_spin_lockval(cpu);
226 	if (vdso_alloc_per_cpu(lc))
227 		goto out;
228 	lowcore_ptr[cpu] = lc;
229 	pcpu_sigp_retry(pcpu, SIGP_SET_PREFIX, (u32)(unsigned long) lc);
230 	return 0;
231 out:
232 	if (pcpu != &pcpu_devices[0]) {
233 		if (mcesa_origin)
234 			kmem_cache_free(pcpu_mcesa_cache,
235 					(void *) mcesa_origin);
236 		free_page(panic_stack);
237 		free_pages(async_stack, ASYNC_ORDER);
238 		free_pages((unsigned long) pcpu->lowcore, LC_ORDER);
239 	}
240 	return -ENOMEM;
241 }
242 
243 #ifdef CONFIG_HOTPLUG_CPU
244 
245 static void pcpu_free_lowcore(struct pcpu *pcpu)
246 {
247 	unsigned long mcesa_origin;
248 
249 	pcpu_sigp_retry(pcpu, SIGP_SET_PREFIX, 0);
250 	lowcore_ptr[pcpu - pcpu_devices] = NULL;
251 	vdso_free_per_cpu(pcpu->lowcore);
252 	if (pcpu == &pcpu_devices[0])
253 		return;
254 	if (MACHINE_HAS_VX || MACHINE_HAS_GS) {
255 		mcesa_origin = pcpu->lowcore->mcesad & MCESA_ORIGIN_MASK;
256 		kmem_cache_free(pcpu_mcesa_cache, (void *) mcesa_origin);
257 	}
258 	free_page(pcpu->lowcore->panic_stack-PANIC_FRAME_OFFSET);
259 	free_pages(pcpu->lowcore->async_stack-ASYNC_FRAME_OFFSET, ASYNC_ORDER);
260 	free_pages((unsigned long) pcpu->lowcore, LC_ORDER);
261 }
262 
263 #endif /* CONFIG_HOTPLUG_CPU */
264 
265 static void pcpu_prepare_secondary(struct pcpu *pcpu, int cpu)
266 {
267 	struct lowcore *lc = pcpu->lowcore;
268 
269 	cpumask_set_cpu(cpu, &init_mm.context.cpu_attach_mask);
270 	cpumask_set_cpu(cpu, mm_cpumask(&init_mm));
271 	lc->cpu_nr = cpu;
272 	lc->spinlock_lockval = arch_spin_lockval(cpu);
273 	lc->percpu_offset = __per_cpu_offset[cpu];
274 	lc->kernel_asce = S390_lowcore.kernel_asce;
275 	lc->machine_flags = S390_lowcore.machine_flags;
276 	lc->user_timer = lc->system_timer = lc->steal_timer = 0;
277 	__ctl_store(lc->cregs_save_area, 0, 15);
278 	save_access_regs((unsigned int *) lc->access_regs_save_area);
279 	memcpy(lc->stfle_fac_list, S390_lowcore.stfle_fac_list,
280 	       MAX_FACILITY_BIT/8);
281 }
282 
283 static void pcpu_attach_task(struct pcpu *pcpu, struct task_struct *tsk)
284 {
285 	struct lowcore *lc = pcpu->lowcore;
286 
287 	lc->kernel_stack = (unsigned long) task_stack_page(tsk)
288 		+ THREAD_SIZE - STACK_FRAME_OVERHEAD - sizeof(struct pt_regs);
289 	lc->current_task = (unsigned long) tsk;
290 	lc->lpp = LPP_MAGIC;
291 	lc->current_pid = tsk->pid;
292 	lc->user_timer = tsk->thread.user_timer;
293 	lc->system_timer = tsk->thread.system_timer;
294 	lc->steal_timer = 0;
295 }
296 
297 static void pcpu_start_fn(struct pcpu *pcpu, void (*func)(void *), void *data)
298 {
299 	struct lowcore *lc = pcpu->lowcore;
300 
301 	lc->restart_stack = lc->kernel_stack;
302 	lc->restart_fn = (unsigned long) func;
303 	lc->restart_data = (unsigned long) data;
304 	lc->restart_source = -1UL;
305 	pcpu_sigp_retry(pcpu, SIGP_RESTART, 0);
306 }
307 
308 /*
309  * Call function via PSW restart on pcpu and stop the current cpu.
310  */
311 static void pcpu_delegate(struct pcpu *pcpu, void (*func)(void *),
312 			  void *data, unsigned long stack)
313 {
314 	struct lowcore *lc = lowcore_ptr[pcpu - pcpu_devices];
315 	unsigned long source_cpu = stap();
316 
317 	__load_psw_mask(PSW_KERNEL_BITS);
318 	if (pcpu->address == source_cpu)
319 		func(data);	/* should not return */
320 	/* Stop target cpu (if func returns this stops the current cpu). */
321 	pcpu_sigp_retry(pcpu, SIGP_STOP, 0);
322 	/* Restart func on the target cpu and stop the current cpu. */
323 	mem_assign_absolute(lc->restart_stack, stack);
324 	mem_assign_absolute(lc->restart_fn, (unsigned long) func);
325 	mem_assign_absolute(lc->restart_data, (unsigned long) data);
326 	mem_assign_absolute(lc->restart_source, source_cpu);
327 	asm volatile(
328 		"0:	sigp	0,%0,%2	# sigp restart to target cpu\n"
329 		"	brc	2,0b	# busy, try again\n"
330 		"1:	sigp	0,%1,%3	# sigp stop to current cpu\n"
331 		"	brc	2,1b	# busy, try again\n"
332 		: : "d" (pcpu->address), "d" (source_cpu),
333 		    "K" (SIGP_RESTART), "K" (SIGP_STOP)
334 		: "0", "1", "cc");
335 	for (;;) ;
336 }
337 
338 /*
339  * Enable additional logical cpus for multi-threading.
340  */
341 static int pcpu_set_smt(unsigned int mtid)
342 {
343 	int cc;
344 
345 	if (smp_cpu_mtid == mtid)
346 		return 0;
347 	cc = __pcpu_sigp(0, SIGP_SET_MULTI_THREADING, mtid, NULL);
348 	if (cc == 0) {
349 		smp_cpu_mtid = mtid;
350 		smp_cpu_mt_shift = 0;
351 		while (smp_cpu_mtid >= (1U << smp_cpu_mt_shift))
352 			smp_cpu_mt_shift++;
353 		pcpu_devices[0].address = stap();
354 	}
355 	return cc;
356 }
357 
358 /*
359  * Call function on an online CPU.
360  */
361 void smp_call_online_cpu(void (*func)(void *), void *data)
362 {
363 	struct pcpu *pcpu;
364 
365 	/* Use the current cpu if it is online. */
366 	pcpu = pcpu_find_address(cpu_online_mask, stap());
367 	if (!pcpu)
368 		/* Use the first online cpu. */
369 		pcpu = pcpu_devices + cpumask_first(cpu_online_mask);
370 	pcpu_delegate(pcpu, func, data, (unsigned long) restart_stack);
371 }
372 
373 /*
374  * Call function on the ipl CPU.
375  */
376 void smp_call_ipl_cpu(void (*func)(void *), void *data)
377 {
378 	pcpu_delegate(&pcpu_devices[0], func, data,
379 		      pcpu_devices->lowcore->panic_stack -
380 		      PANIC_FRAME_OFFSET + PAGE_SIZE);
381 }
382 
383 int smp_find_processor_id(u16 address)
384 {
385 	int cpu;
386 
387 	for_each_present_cpu(cpu)
388 		if (pcpu_devices[cpu].address == address)
389 			return cpu;
390 	return -1;
391 }
392 
393 bool arch_vcpu_is_preempted(int cpu)
394 {
395 	if (test_cpu_flag_of(CIF_ENABLED_WAIT, cpu))
396 		return false;
397 	if (pcpu_running(pcpu_devices + cpu))
398 		return false;
399 	return true;
400 }
401 EXPORT_SYMBOL(arch_vcpu_is_preempted);
402 
403 void smp_yield_cpu(int cpu)
404 {
405 	if (MACHINE_HAS_DIAG9C) {
406 		diag_stat_inc_norecursion(DIAG_STAT_X09C);
407 		asm volatile("diag %0,0,0x9c"
408 			     : : "d" (pcpu_devices[cpu].address));
409 	} else if (MACHINE_HAS_DIAG44) {
410 		diag_stat_inc_norecursion(DIAG_STAT_X044);
411 		asm volatile("diag 0,0,0x44");
412 	}
413 }
414 
415 /*
416  * Send cpus emergency shutdown signal. This gives the cpus the
417  * opportunity to complete outstanding interrupts.
418  */
419 static void smp_emergency_stop(cpumask_t *cpumask)
420 {
421 	u64 end;
422 	int cpu;
423 
424 	end = get_tod_clock() + (1000000UL << 12);
425 	for_each_cpu(cpu, cpumask) {
426 		struct pcpu *pcpu = pcpu_devices + cpu;
427 		set_bit(ec_stop_cpu, &pcpu->ec_mask);
428 		while (__pcpu_sigp(pcpu->address, SIGP_EMERGENCY_SIGNAL,
429 				   0, NULL) == SIGP_CC_BUSY &&
430 		       get_tod_clock() < end)
431 			cpu_relax();
432 	}
433 	while (get_tod_clock() < end) {
434 		for_each_cpu(cpu, cpumask)
435 			if (pcpu_stopped(pcpu_devices + cpu))
436 				cpumask_clear_cpu(cpu, cpumask);
437 		if (cpumask_empty(cpumask))
438 			break;
439 		cpu_relax();
440 	}
441 }
442 
443 /*
444  * Stop all cpus but the current one.
445  */
446 void smp_send_stop(void)
447 {
448 	cpumask_t cpumask;
449 	int cpu;
450 
451 	/* Disable all interrupts/machine checks */
452 	__load_psw_mask(PSW_KERNEL_BITS | PSW_MASK_DAT);
453 	trace_hardirqs_off();
454 
455 	debug_set_critical();
456 	cpumask_copy(&cpumask, cpu_online_mask);
457 	cpumask_clear_cpu(smp_processor_id(), &cpumask);
458 
459 	if (oops_in_progress)
460 		smp_emergency_stop(&cpumask);
461 
462 	/* stop all processors */
463 	for_each_cpu(cpu, &cpumask) {
464 		struct pcpu *pcpu = pcpu_devices + cpu;
465 		pcpu_sigp_retry(pcpu, SIGP_STOP, 0);
466 		while (!pcpu_stopped(pcpu))
467 			cpu_relax();
468 	}
469 }
470 
471 /*
472  * This is the main routine where commands issued by other
473  * cpus are handled.
474  */
475 static void smp_handle_ext_call(void)
476 {
477 	unsigned long bits;
478 
479 	/* handle bit signal external calls */
480 	bits = xchg(&pcpu_devices[smp_processor_id()].ec_mask, 0);
481 	if (test_bit(ec_stop_cpu, &bits))
482 		smp_stop_cpu();
483 	if (test_bit(ec_schedule, &bits))
484 		scheduler_ipi();
485 	if (test_bit(ec_call_function_single, &bits))
486 		generic_smp_call_function_single_interrupt();
487 }
488 
489 static void do_ext_call_interrupt(struct ext_code ext_code,
490 				  unsigned int param32, unsigned long param64)
491 {
492 	inc_irq_stat(ext_code.code == 0x1202 ? IRQEXT_EXC : IRQEXT_EMS);
493 	smp_handle_ext_call();
494 }
495 
496 void arch_send_call_function_ipi_mask(const struct cpumask *mask)
497 {
498 	int cpu;
499 
500 	for_each_cpu(cpu, mask)
501 		pcpu_ec_call(pcpu_devices + cpu, ec_call_function_single);
502 }
503 
504 void arch_send_call_function_single_ipi(int cpu)
505 {
506 	pcpu_ec_call(pcpu_devices + cpu, ec_call_function_single);
507 }
508 
509 /*
510  * this function sends a 'reschedule' IPI to another CPU.
511  * it goes straight through and wastes no time serializing
512  * anything. Worst case is that we lose a reschedule ...
513  */
514 void smp_send_reschedule(int cpu)
515 {
516 	pcpu_ec_call(pcpu_devices + cpu, ec_schedule);
517 }
518 
519 /*
520  * parameter area for the set/clear control bit callbacks
521  */
522 struct ec_creg_mask_parms {
523 	unsigned long orval;
524 	unsigned long andval;
525 	int cr;
526 };
527 
528 /*
529  * callback for setting/clearing control bits
530  */
531 static void smp_ctl_bit_callback(void *info)
532 {
533 	struct ec_creg_mask_parms *pp = info;
534 	unsigned long cregs[16];
535 
536 	__ctl_store(cregs, 0, 15);
537 	cregs[pp->cr] = (cregs[pp->cr] & pp->andval) | pp->orval;
538 	__ctl_load(cregs, 0, 15);
539 }
540 
541 /*
542  * Set a bit in a control register of all cpus
543  */
544 void smp_ctl_set_bit(int cr, int bit)
545 {
546 	struct ec_creg_mask_parms parms = { 1UL << bit, -1UL, cr };
547 
548 	on_each_cpu(smp_ctl_bit_callback, &parms, 1);
549 }
550 EXPORT_SYMBOL(smp_ctl_set_bit);
551 
552 /*
553  * Clear a bit in a control register of all cpus
554  */
555 void smp_ctl_clear_bit(int cr, int bit)
556 {
557 	struct ec_creg_mask_parms parms = { 0, ~(1UL << bit), cr };
558 
559 	on_each_cpu(smp_ctl_bit_callback, &parms, 1);
560 }
561 EXPORT_SYMBOL(smp_ctl_clear_bit);
562 
563 #ifdef CONFIG_CRASH_DUMP
564 
565 int smp_store_status(int cpu)
566 {
567 	struct pcpu *pcpu = pcpu_devices + cpu;
568 	unsigned long pa;
569 
570 	pa = __pa(&pcpu->lowcore->floating_pt_save_area);
571 	if (__pcpu_sigp_relax(pcpu->address, SIGP_STORE_STATUS_AT_ADDRESS,
572 			      pa) != SIGP_CC_ORDER_CODE_ACCEPTED)
573 		return -EIO;
574 	if (!MACHINE_HAS_VX && !MACHINE_HAS_GS)
575 		return 0;
576 	pa = __pa(pcpu->lowcore->mcesad & MCESA_ORIGIN_MASK);
577 	if (MACHINE_HAS_GS)
578 		pa |= pcpu->lowcore->mcesad & MCESA_LC_MASK;
579 	if (__pcpu_sigp_relax(pcpu->address, SIGP_STORE_ADDITIONAL_STATUS,
580 			      pa) != SIGP_CC_ORDER_CODE_ACCEPTED)
581 		return -EIO;
582 	return 0;
583 }
584 
585 /*
586  * Collect CPU state of the previous, crashed system.
587  * There are four cases:
588  * 1) standard zfcp dump
589  *    condition: OLDMEM_BASE == NULL && ipl_info.type == IPL_TYPE_FCP_DUMP
590  *    The state for all CPUs except the boot CPU needs to be collected
591  *    with sigp stop-and-store-status. The boot CPU state is located in
592  *    the absolute lowcore of the memory stored in the HSA. The zcore code
593  *    will copy the boot CPU state from the HSA.
594  * 2) stand-alone kdump for SCSI (zfcp dump with swapped memory)
595  *    condition: OLDMEM_BASE != NULL && ipl_info.type == IPL_TYPE_FCP_DUMP
596  *    The state for all CPUs except the boot CPU needs to be collected
597  *    with sigp stop-and-store-status. The firmware or the boot-loader
598  *    stored the registers of the boot CPU in the absolute lowcore in the
599  *    memory of the old system.
600  * 3) kdump and the old kernel did not store the CPU state,
601  *    or stand-alone kdump for DASD
602  *    condition: OLDMEM_BASE != NULL && !is_kdump_kernel()
603  *    The state for all CPUs except the boot CPU needs to be collected
604  *    with sigp stop-and-store-status. The kexec code or the boot-loader
605  *    stored the registers of the boot CPU in the memory of the old system.
606  * 4) kdump and the old kernel stored the CPU state
607  *    condition: OLDMEM_BASE != NULL && is_kdump_kernel()
608  *    This case does not exist for s390 anymore, setup_arch explicitly
609  *    deactivates the elfcorehdr= kernel parameter
610  */
611 static __init void smp_save_cpu_vxrs(struct save_area *sa, u16 addr,
612 				     bool is_boot_cpu, unsigned long page)
613 {
614 	__vector128 *vxrs = (__vector128 *) page;
615 
616 	if (is_boot_cpu)
617 		vxrs = boot_cpu_vector_save_area;
618 	else
619 		__pcpu_sigp_relax(addr, SIGP_STORE_ADDITIONAL_STATUS, page);
620 	save_area_add_vxrs(sa, vxrs);
621 }
622 
623 static __init void smp_save_cpu_regs(struct save_area *sa, u16 addr,
624 				     bool is_boot_cpu, unsigned long page)
625 {
626 	void *regs = (void *) page;
627 
628 	if (is_boot_cpu)
629 		copy_oldmem_kernel(regs, (void *) __LC_FPREGS_SAVE_AREA, 512);
630 	else
631 		__pcpu_sigp_relax(addr, SIGP_STORE_STATUS_AT_ADDRESS, page);
632 	save_area_add_regs(sa, regs);
633 }
634 
635 void __init smp_save_dump_cpus(void)
636 {
637 	int addr, boot_cpu_addr, max_cpu_addr;
638 	struct save_area *sa;
639 	unsigned long page;
640 	bool is_boot_cpu;
641 
642 	if (!(OLDMEM_BASE || ipl_info.type == IPL_TYPE_FCP_DUMP))
643 		/* No previous system present, normal boot. */
644 		return;
645 	/* Allocate a page as dumping area for the store status sigps */
646 	page = memblock_alloc_base(PAGE_SIZE, PAGE_SIZE, 1UL << 31);
647 	/* Set multi-threading state to the previous system. */
648 	pcpu_set_smt(sclp.mtid_prev);
649 	boot_cpu_addr = stap();
650 	max_cpu_addr = SCLP_MAX_CORES << sclp.mtid_prev;
651 	for (addr = 0; addr <= max_cpu_addr; addr++) {
652 		if (__pcpu_sigp_relax(addr, SIGP_SENSE, 0) ==
653 		    SIGP_CC_NOT_OPERATIONAL)
654 			continue;
655 		is_boot_cpu = (addr == boot_cpu_addr);
656 		/* Allocate save area */
657 		sa = save_area_alloc(is_boot_cpu);
658 		if (!sa)
659 			panic("could not allocate memory for save area\n");
660 		if (MACHINE_HAS_VX)
661 			/* Get the vector registers */
662 			smp_save_cpu_vxrs(sa, addr, is_boot_cpu, page);
663 		/*
664 		 * For a zfcp dump OLDMEM_BASE == NULL and the registers
665 		 * of the boot CPU are stored in the HSA. To retrieve
666 		 * these registers an SCLP request is required which is
667 		 * done by drivers/s390/char/zcore.c:init_cpu_info()
668 		 */
669 		if (!is_boot_cpu || OLDMEM_BASE)
670 			/* Get the CPU registers */
671 			smp_save_cpu_regs(sa, addr, is_boot_cpu, page);
672 	}
673 	memblock_free(page, PAGE_SIZE);
674 	diag308_reset();
675 	pcpu_set_smt(0);
676 }
677 #endif /* CONFIG_CRASH_DUMP */
678 
679 void smp_cpu_set_polarization(int cpu, int val)
680 {
681 	pcpu_devices[cpu].polarization = val;
682 }
683 
684 int smp_cpu_get_polarization(int cpu)
685 {
686 	return pcpu_devices[cpu].polarization;
687 }
688 
689 static void __ref smp_get_core_info(struct sclp_core_info *info, int early)
690 {
691 	static int use_sigp_detection;
692 	int address;
693 
694 	if (use_sigp_detection || sclp_get_core_info(info, early)) {
695 		use_sigp_detection = 1;
696 		for (address = 0;
697 		     address < (SCLP_MAX_CORES << smp_cpu_mt_shift);
698 		     address += (1U << smp_cpu_mt_shift)) {
699 			if (__pcpu_sigp_relax(address, SIGP_SENSE, 0) ==
700 			    SIGP_CC_NOT_OPERATIONAL)
701 				continue;
702 			info->core[info->configured].core_id =
703 				address >> smp_cpu_mt_shift;
704 			info->configured++;
705 		}
706 		info->combined = info->configured;
707 	}
708 }
709 
710 static int smp_add_present_cpu(int cpu);
711 
712 static int __smp_rescan_cpus(struct sclp_core_info *info, int sysfs_add)
713 {
714 	struct pcpu *pcpu;
715 	cpumask_t avail;
716 	int cpu, nr, i, j;
717 	u16 address;
718 
719 	nr = 0;
720 	cpumask_xor(&avail, cpu_possible_mask, cpu_present_mask);
721 	cpu = cpumask_first(&avail);
722 	for (i = 0; (i < info->combined) && (cpu < nr_cpu_ids); i++) {
723 		if (sclp.has_core_type && info->core[i].type != boot_core_type)
724 			continue;
725 		address = info->core[i].core_id << smp_cpu_mt_shift;
726 		for (j = 0; j <= smp_cpu_mtid; j++) {
727 			if (pcpu_find_address(cpu_present_mask, address + j))
728 				continue;
729 			pcpu = pcpu_devices + cpu;
730 			pcpu->address = address + j;
731 			pcpu->state =
732 				(cpu >= info->configured*(smp_cpu_mtid + 1)) ?
733 				CPU_STATE_STANDBY : CPU_STATE_CONFIGURED;
734 			smp_cpu_set_polarization(cpu, POLARIZATION_UNKNOWN);
735 			set_cpu_present(cpu, true);
736 			if (sysfs_add && smp_add_present_cpu(cpu) != 0)
737 				set_cpu_present(cpu, false);
738 			else
739 				nr++;
740 			cpu = cpumask_next(cpu, &avail);
741 			if (cpu >= nr_cpu_ids)
742 				break;
743 		}
744 	}
745 	return nr;
746 }
747 
748 void __init smp_detect_cpus(void)
749 {
750 	unsigned int cpu, mtid, c_cpus, s_cpus;
751 	struct sclp_core_info *info;
752 	u16 address;
753 
754 	/* Get CPU information */
755 	info = memblock_virt_alloc(sizeof(*info), 8);
756 	smp_get_core_info(info, 1);
757 	/* Find boot CPU type */
758 	if (sclp.has_core_type) {
759 		address = stap();
760 		for (cpu = 0; cpu < info->combined; cpu++)
761 			if (info->core[cpu].core_id == address) {
762 				/* The boot cpu dictates the cpu type. */
763 				boot_core_type = info->core[cpu].type;
764 				break;
765 			}
766 		if (cpu >= info->combined)
767 			panic("Could not find boot CPU type");
768 	}
769 
770 	/* Set multi-threading state for the current system */
771 	mtid = boot_core_type ? sclp.mtid : sclp.mtid_cp;
772 	mtid = (mtid < smp_max_threads) ? mtid : smp_max_threads - 1;
773 	pcpu_set_smt(mtid);
774 
775 	/* Print number of CPUs */
776 	c_cpus = s_cpus = 0;
777 	for (cpu = 0; cpu < info->combined; cpu++) {
778 		if (sclp.has_core_type &&
779 		    info->core[cpu].type != boot_core_type)
780 			continue;
781 		if (cpu < info->configured)
782 			c_cpus += smp_cpu_mtid + 1;
783 		else
784 			s_cpus += smp_cpu_mtid + 1;
785 	}
786 	pr_info("%d configured CPUs, %d standby CPUs\n", c_cpus, s_cpus);
787 
788 	/* Add CPUs present at boot */
789 	get_online_cpus();
790 	__smp_rescan_cpus(info, 0);
791 	put_online_cpus();
792 	memblock_free_early((unsigned long)info, sizeof(*info));
793 }
794 
795 /*
796  *	Activate a secondary processor.
797  */
798 static void smp_start_secondary(void *cpuvoid)
799 {
800 	S390_lowcore.last_update_clock = get_tod_clock();
801 	S390_lowcore.restart_stack = (unsigned long) restart_stack;
802 	S390_lowcore.restart_fn = (unsigned long) do_restart;
803 	S390_lowcore.restart_data = 0;
804 	S390_lowcore.restart_source = -1UL;
805 	restore_access_regs(S390_lowcore.access_regs_save_area);
806 	__ctl_load(S390_lowcore.cregs_save_area, 0, 15);
807 	__load_psw_mask(PSW_KERNEL_BITS | PSW_MASK_DAT);
808 	cpu_init();
809 	preempt_disable();
810 	init_cpu_timer();
811 	vtime_init();
812 	pfault_init();
813 	notify_cpu_starting(smp_processor_id());
814 	set_cpu_online(smp_processor_id(), true);
815 	inc_irq_stat(CPU_RST);
816 	local_irq_enable();
817 	cpu_startup_entry(CPUHP_AP_ONLINE_IDLE);
818 }
819 
820 /* Upping and downing of CPUs */
821 int __cpu_up(unsigned int cpu, struct task_struct *tidle)
822 {
823 	struct pcpu *pcpu;
824 	int base, i, rc;
825 
826 	pcpu = pcpu_devices + cpu;
827 	if (pcpu->state != CPU_STATE_CONFIGURED)
828 		return -EIO;
829 	base = smp_get_base_cpu(cpu);
830 	for (i = 0; i <= smp_cpu_mtid; i++) {
831 		if (base + i < nr_cpu_ids)
832 			if (cpu_online(base + i))
833 				break;
834 	}
835 	/*
836 	 * If this is the first CPU of the core to get online
837 	 * do an initial CPU reset.
838 	 */
839 	if (i > smp_cpu_mtid &&
840 	    pcpu_sigp_retry(pcpu_devices + base, SIGP_INITIAL_CPU_RESET, 0) !=
841 	    SIGP_CC_ORDER_CODE_ACCEPTED)
842 		return -EIO;
843 
844 	rc = pcpu_alloc_lowcore(pcpu, cpu);
845 	if (rc)
846 		return rc;
847 	pcpu_prepare_secondary(pcpu, cpu);
848 	pcpu_attach_task(pcpu, tidle);
849 	pcpu_start_fn(pcpu, smp_start_secondary, NULL);
850 	/* Wait until cpu puts itself in the online & active maps */
851 	while (!cpu_online(cpu))
852 		cpu_relax();
853 	return 0;
854 }
855 
856 static unsigned int setup_possible_cpus __initdata;
857 
858 static int __init _setup_possible_cpus(char *s)
859 {
860 	get_option(&s, &setup_possible_cpus);
861 	return 0;
862 }
863 early_param("possible_cpus", _setup_possible_cpus);
864 
865 #ifdef CONFIG_HOTPLUG_CPU
866 
867 int __cpu_disable(void)
868 {
869 	unsigned long cregs[16];
870 
871 	/* Handle possible pending IPIs */
872 	smp_handle_ext_call();
873 	set_cpu_online(smp_processor_id(), false);
874 	/* Disable pseudo page faults on this cpu. */
875 	pfault_fini();
876 	/* Disable interrupt sources via control register. */
877 	__ctl_store(cregs, 0, 15);
878 	cregs[0]  &= ~0x0000ee70UL;	/* disable all external interrupts */
879 	cregs[6]  &= ~0xff000000UL;	/* disable all I/O interrupts */
880 	cregs[14] &= ~0x1f000000UL;	/* disable most machine checks */
881 	__ctl_load(cregs, 0, 15);
882 	clear_cpu_flag(CIF_NOHZ_DELAY);
883 	return 0;
884 }
885 
886 void __cpu_die(unsigned int cpu)
887 {
888 	struct pcpu *pcpu;
889 
890 	/* Wait until target cpu is down */
891 	pcpu = pcpu_devices + cpu;
892 	while (!pcpu_stopped(pcpu))
893 		cpu_relax();
894 	pcpu_free_lowcore(pcpu);
895 	cpumask_clear_cpu(cpu, mm_cpumask(&init_mm));
896 	cpumask_clear_cpu(cpu, &init_mm.context.cpu_attach_mask);
897 }
898 
899 void __noreturn cpu_die(void)
900 {
901 	idle_task_exit();
902 	pcpu_sigp_retry(pcpu_devices + smp_processor_id(), SIGP_STOP, 0);
903 	for (;;) ;
904 }
905 
906 #endif /* CONFIG_HOTPLUG_CPU */
907 
908 void __init smp_fill_possible_mask(void)
909 {
910 	unsigned int possible, sclp_max, cpu;
911 
912 	sclp_max = max(sclp.mtid, sclp.mtid_cp) + 1;
913 	sclp_max = min(smp_max_threads, sclp_max);
914 	sclp_max = (sclp.max_cores * sclp_max) ?: nr_cpu_ids;
915 	possible = setup_possible_cpus ?: nr_cpu_ids;
916 	possible = min(possible, sclp_max);
917 	for (cpu = 0; cpu < possible && cpu < nr_cpu_ids; cpu++)
918 		set_cpu_possible(cpu, true);
919 }
920 
921 void __init smp_prepare_cpus(unsigned int max_cpus)
922 {
923 	unsigned long size;
924 
925 	/* request the 0x1201 emergency signal external interrupt */
926 	if (register_external_irq(EXT_IRQ_EMERGENCY_SIG, do_ext_call_interrupt))
927 		panic("Couldn't request external interrupt 0x1201");
928 	/* request the 0x1202 external call external interrupt */
929 	if (register_external_irq(EXT_IRQ_EXTERNAL_CALL, do_ext_call_interrupt))
930 		panic("Couldn't request external interrupt 0x1202");
931 	/* create slab cache for the machine-check-extended-save-areas */
932 	if (MACHINE_HAS_VX || MACHINE_HAS_GS) {
933 		size = 1UL << (MACHINE_HAS_GS ? 11 : 10);
934 		pcpu_mcesa_cache = kmem_cache_create("nmi_save_areas",
935 						     size, size, 0, NULL);
936 		if (!pcpu_mcesa_cache)
937 			panic("Couldn't create nmi save area cache");
938 	}
939 }
940 
941 void __init smp_prepare_boot_cpu(void)
942 {
943 	struct pcpu *pcpu = pcpu_devices;
944 
945 	WARN_ON(!cpu_present(0) || !cpu_online(0));
946 	pcpu->state = CPU_STATE_CONFIGURED;
947 	pcpu->lowcore = (struct lowcore *)(unsigned long) store_prefix();
948 	S390_lowcore.percpu_offset = __per_cpu_offset[0];
949 	smp_cpu_set_polarization(0, POLARIZATION_UNKNOWN);
950 }
951 
952 void __init smp_cpus_done(unsigned int max_cpus)
953 {
954 }
955 
956 void __init smp_setup_processor_id(void)
957 {
958 	pcpu_devices[0].address = stap();
959 	S390_lowcore.cpu_nr = 0;
960 	S390_lowcore.spinlock_lockval = arch_spin_lockval(0);
961 }
962 
963 /*
964  * the frequency of the profiling timer can be changed
965  * by writing a multiplier value into /proc/profile.
966  *
967  * usually you want to run this on all CPUs ;)
968  */
969 int setup_profiling_timer(unsigned int multiplier)
970 {
971 	return 0;
972 }
973 
974 #ifdef CONFIG_HOTPLUG_CPU
975 static ssize_t cpu_configure_show(struct device *dev,
976 				  struct device_attribute *attr, char *buf)
977 {
978 	ssize_t count;
979 
980 	mutex_lock(&smp_cpu_state_mutex);
981 	count = sprintf(buf, "%d\n", pcpu_devices[dev->id].state);
982 	mutex_unlock(&smp_cpu_state_mutex);
983 	return count;
984 }
985 
986 static ssize_t cpu_configure_store(struct device *dev,
987 				   struct device_attribute *attr,
988 				   const char *buf, size_t count)
989 {
990 	struct pcpu *pcpu;
991 	int cpu, val, rc, i;
992 	char delim;
993 
994 	if (sscanf(buf, "%d %c", &val, &delim) != 1)
995 		return -EINVAL;
996 	if (val != 0 && val != 1)
997 		return -EINVAL;
998 	get_online_cpus();
999 	mutex_lock(&smp_cpu_state_mutex);
1000 	rc = -EBUSY;
1001 	/* disallow configuration changes of online cpus and cpu 0 */
1002 	cpu = dev->id;
1003 	cpu = smp_get_base_cpu(cpu);
1004 	if (cpu == 0)
1005 		goto out;
1006 	for (i = 0; i <= smp_cpu_mtid; i++)
1007 		if (cpu_online(cpu + i))
1008 			goto out;
1009 	pcpu = pcpu_devices + cpu;
1010 	rc = 0;
1011 	switch (val) {
1012 	case 0:
1013 		if (pcpu->state != CPU_STATE_CONFIGURED)
1014 			break;
1015 		rc = sclp_core_deconfigure(pcpu->address >> smp_cpu_mt_shift);
1016 		if (rc)
1017 			break;
1018 		for (i = 0; i <= smp_cpu_mtid; i++) {
1019 			if (cpu + i >= nr_cpu_ids || !cpu_present(cpu + i))
1020 				continue;
1021 			pcpu[i].state = CPU_STATE_STANDBY;
1022 			smp_cpu_set_polarization(cpu + i,
1023 						 POLARIZATION_UNKNOWN);
1024 		}
1025 		topology_expect_change();
1026 		break;
1027 	case 1:
1028 		if (pcpu->state != CPU_STATE_STANDBY)
1029 			break;
1030 		rc = sclp_core_configure(pcpu->address >> smp_cpu_mt_shift);
1031 		if (rc)
1032 			break;
1033 		for (i = 0; i <= smp_cpu_mtid; i++) {
1034 			if (cpu + i >= nr_cpu_ids || !cpu_present(cpu + i))
1035 				continue;
1036 			pcpu[i].state = CPU_STATE_CONFIGURED;
1037 			smp_cpu_set_polarization(cpu + i,
1038 						 POLARIZATION_UNKNOWN);
1039 		}
1040 		topology_expect_change();
1041 		break;
1042 	default:
1043 		break;
1044 	}
1045 out:
1046 	mutex_unlock(&smp_cpu_state_mutex);
1047 	put_online_cpus();
1048 	return rc ? rc : count;
1049 }
1050 static DEVICE_ATTR(configure, 0644, cpu_configure_show, cpu_configure_store);
1051 #endif /* CONFIG_HOTPLUG_CPU */
1052 
1053 static ssize_t show_cpu_address(struct device *dev,
1054 				struct device_attribute *attr, char *buf)
1055 {
1056 	return sprintf(buf, "%d\n", pcpu_devices[dev->id].address);
1057 }
1058 static DEVICE_ATTR(address, 0444, show_cpu_address, NULL);
1059 
1060 static struct attribute *cpu_common_attrs[] = {
1061 #ifdef CONFIG_HOTPLUG_CPU
1062 	&dev_attr_configure.attr,
1063 #endif
1064 	&dev_attr_address.attr,
1065 	NULL,
1066 };
1067 
1068 static struct attribute_group cpu_common_attr_group = {
1069 	.attrs = cpu_common_attrs,
1070 };
1071 
1072 static struct attribute *cpu_online_attrs[] = {
1073 	&dev_attr_idle_count.attr,
1074 	&dev_attr_idle_time_us.attr,
1075 	NULL,
1076 };
1077 
1078 static struct attribute_group cpu_online_attr_group = {
1079 	.attrs = cpu_online_attrs,
1080 };
1081 
1082 static int smp_cpu_online(unsigned int cpu)
1083 {
1084 	struct device *s = &per_cpu(cpu_device, cpu)->dev;
1085 
1086 	return sysfs_create_group(&s->kobj, &cpu_online_attr_group);
1087 }
1088 static int smp_cpu_pre_down(unsigned int cpu)
1089 {
1090 	struct device *s = &per_cpu(cpu_device, cpu)->dev;
1091 
1092 	sysfs_remove_group(&s->kobj, &cpu_online_attr_group);
1093 	return 0;
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 	rc = topology_cpu_init(c);
1115 	if (rc)
1116 		goto out_topology;
1117 	return 0;
1118 
1119 out_topology:
1120 	sysfs_remove_group(&s->kobj, &cpu_common_attr_group);
1121 out_cpu:
1122 #ifdef CONFIG_HOTPLUG_CPU
1123 	unregister_cpu(c);
1124 #endif
1125 out:
1126 	return rc;
1127 }
1128 
1129 #ifdef CONFIG_HOTPLUG_CPU
1130 
1131 int __ref smp_rescan_cpus(void)
1132 {
1133 	struct sclp_core_info *info;
1134 	int nr;
1135 
1136 	info = kzalloc(sizeof(*info), GFP_KERNEL);
1137 	if (!info)
1138 		return -ENOMEM;
1139 	smp_get_core_info(info, 0);
1140 	get_online_cpus();
1141 	mutex_lock(&smp_cpu_state_mutex);
1142 	nr = __smp_rescan_cpus(info, 1);
1143 	mutex_unlock(&smp_cpu_state_mutex);
1144 	put_online_cpus();
1145 	kfree(info);
1146 	if (nr)
1147 		topology_schedule_update();
1148 	return 0;
1149 }
1150 
1151 static ssize_t __ref rescan_store(struct device *dev,
1152 				  struct device_attribute *attr,
1153 				  const char *buf,
1154 				  size_t count)
1155 {
1156 	int rc;
1157 
1158 	rc = smp_rescan_cpus();
1159 	return rc ? rc : count;
1160 }
1161 static DEVICE_ATTR(rescan, 0200, NULL, rescan_store);
1162 #endif /* CONFIG_HOTPLUG_CPU */
1163 
1164 static int __init s390_smp_init(void)
1165 {
1166 	int cpu, rc = 0;
1167 
1168 #ifdef CONFIG_HOTPLUG_CPU
1169 	rc = device_create_file(cpu_subsys.dev_root, &dev_attr_rescan);
1170 	if (rc)
1171 		return rc;
1172 #endif
1173 	for_each_present_cpu(cpu) {
1174 		rc = smp_add_present_cpu(cpu);
1175 		if (rc)
1176 			goto out;
1177 	}
1178 
1179 	rc = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "s390/smp:online",
1180 			       smp_cpu_online, smp_cpu_pre_down);
1181 out:
1182 	return rc;
1183 }
1184 subsys_initcall(s390_smp_init);
1185