xref: /openbmc/linux/arch/s390/kernel/smp.c (revision 7211ec63)
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/kmemleak.h>
30 #include <linux/delay.h>
31 #include <linux/interrupt.h>
32 #include <linux/irqflags.h>
33 #include <linux/cpu.h>
34 #include <linux/slab.h>
35 #include <linux/sched/hotplug.h>
36 #include <linux/sched/task_stack.h>
37 #include <linux/crash_dump.h>
38 #include <linux/memblock.h>
39 #include <asm/asm-offsets.h>
40 #include <asm/diag.h>
41 #include <asm/switch_to.h>
42 #include <asm/facility.h>
43 #include <asm/ipl.h>
44 #include <asm/setup.h>
45 #include <asm/irq.h>
46 #include <asm/tlbflush.h>
47 #include <asm/vtimer.h>
48 #include <asm/lowcore.h>
49 #include <asm/sclp.h>
50 #include <asm/vdso.h>
51 #include <asm/debug.h>
52 #include <asm/os_info.h>
53 #include <asm/sigp.h>
54 #include <asm/idle.h>
55 #include <asm/nmi.h>
56 #include "entry.h"
57 
58 enum {
59 	ec_schedule = 0,
60 	ec_call_function_single,
61 	ec_stop_cpu,
62 };
63 
64 enum {
65 	CPU_STATE_STANDBY,
66 	CPU_STATE_CONFIGURED,
67 };
68 
69 static DEFINE_PER_CPU(struct cpu *, cpu_device);
70 
71 struct pcpu {
72 	struct lowcore *lowcore;	/* lowcore page(s) for the cpu */
73 	unsigned long ec_mask;		/* bit mask for ec_xxx functions */
74 	unsigned long ec_clk;		/* sigp timestamp for ec_xxx */
75 	signed char state;		/* physical cpu state */
76 	signed char polarization;	/* physical polarization */
77 	u16 address;			/* physical cpu address */
78 };
79 
80 static u8 boot_core_type;
81 static struct pcpu pcpu_devices[NR_CPUS];
82 
83 static struct kmem_cache *pcpu_mcesa_cache;
84 
85 unsigned int smp_cpu_mt_shift;
86 EXPORT_SYMBOL(smp_cpu_mt_shift);
87 
88 unsigned int smp_cpu_mtid;
89 EXPORT_SYMBOL(smp_cpu_mtid);
90 
91 #ifdef CONFIG_CRASH_DUMP
92 __vector128 __initdata boot_cpu_vector_save_area[__NUM_VXRS];
93 #endif
94 
95 static unsigned int smp_max_threads __initdata = -1U;
96 
97 static int __init early_nosmt(char *s)
98 {
99 	smp_max_threads = 1;
100 	return 0;
101 }
102 early_param("nosmt", early_nosmt);
103 
104 static int __init early_smt(char *s)
105 {
106 	get_option(&s, &smp_max_threads);
107 	return 0;
108 }
109 early_param("smt", early_smt);
110 
111 /*
112  * The smp_cpu_state_mutex must be held when changing the state or polarization
113  * member of a pcpu data structure within the pcpu_devices arreay.
114  */
115 DEFINE_MUTEX(smp_cpu_state_mutex);
116 
117 /*
118  * Signal processor helper functions.
119  */
120 static inline int __pcpu_sigp_relax(u16 addr, u8 order, unsigned long parm)
121 {
122 	int cc;
123 
124 	while (1) {
125 		cc = __pcpu_sigp(addr, order, parm, NULL);
126 		if (cc != SIGP_CC_BUSY)
127 			return cc;
128 		cpu_relax();
129 	}
130 }
131 
132 static int pcpu_sigp_retry(struct pcpu *pcpu, u8 order, u32 parm)
133 {
134 	int cc, retry;
135 
136 	for (retry = 0; ; retry++) {
137 		cc = __pcpu_sigp(pcpu->address, order, parm, NULL);
138 		if (cc != SIGP_CC_BUSY)
139 			break;
140 		if (retry >= 3)
141 			udelay(10);
142 	}
143 	return cc;
144 }
145 
146 static inline int pcpu_stopped(struct pcpu *pcpu)
147 {
148 	u32 uninitialized_var(status);
149 
150 	if (__pcpu_sigp(pcpu->address, SIGP_SENSE,
151 			0, &status) != SIGP_CC_STATUS_STORED)
152 		return 0;
153 	return !!(status & (SIGP_STATUS_CHECK_STOP|SIGP_STATUS_STOPPED));
154 }
155 
156 static inline int pcpu_running(struct pcpu *pcpu)
157 {
158 	if (__pcpu_sigp(pcpu->address, SIGP_SENSE_RUNNING,
159 			0, NULL) != SIGP_CC_STATUS_STORED)
160 		return 1;
161 	/* Status stored condition code is equivalent to cpu not running. */
162 	return 0;
163 }
164 
165 /*
166  * Find struct pcpu by cpu address.
167  */
168 static struct pcpu *pcpu_find_address(const struct cpumask *mask, u16 address)
169 {
170 	int cpu;
171 
172 	for_each_cpu(cpu, mask)
173 		if (pcpu_devices[cpu].address == address)
174 			return pcpu_devices + cpu;
175 	return NULL;
176 }
177 
178 static void pcpu_ec_call(struct pcpu *pcpu, int ec_bit)
179 {
180 	int order;
181 
182 	if (test_and_set_bit(ec_bit, &pcpu->ec_mask))
183 		return;
184 	order = pcpu_running(pcpu) ? SIGP_EXTERNAL_CALL : SIGP_EMERGENCY_SIGNAL;
185 	pcpu->ec_clk = get_tod_clock_fast();
186 	pcpu_sigp_retry(pcpu, order, 0);
187 }
188 
189 #define ASYNC_FRAME_OFFSET (ASYNC_SIZE - STACK_FRAME_OVERHEAD - __PT_SIZE)
190 #define PANIC_FRAME_OFFSET (PAGE_SIZE - STACK_FRAME_OVERHEAD - __PT_SIZE)
191 
192 static int pcpu_alloc_lowcore(struct pcpu *pcpu, int cpu)
193 {
194 	unsigned long async_stack, panic_stack;
195 	unsigned long mcesa_origin, mcesa_bits;
196 	struct lowcore *lc;
197 
198 	mcesa_origin = mcesa_bits = 0;
199 	if (pcpu != &pcpu_devices[0]) {
200 		pcpu->lowcore =	(struct lowcore *)
201 			__get_free_pages(GFP_KERNEL | GFP_DMA, LC_ORDER);
202 		async_stack = __get_free_pages(GFP_KERNEL, ASYNC_ORDER);
203 		panic_stack = __get_free_page(GFP_KERNEL);
204 		if (!pcpu->lowcore || !panic_stack || !async_stack)
205 			goto out;
206 		if (MACHINE_HAS_VX || MACHINE_HAS_GS) {
207 			mcesa_origin = (unsigned long)
208 				kmem_cache_alloc(pcpu_mcesa_cache, GFP_KERNEL);
209 			if (!mcesa_origin)
210 				goto out;
211 			/* The pointer is stored with mcesa_bits ORed in */
212 			kmemleak_not_leak((void *) mcesa_origin);
213 			mcesa_bits = MACHINE_HAS_GS ? 11 : 0;
214 		}
215 	} else {
216 		async_stack = pcpu->lowcore->async_stack - ASYNC_FRAME_OFFSET;
217 		panic_stack = pcpu->lowcore->panic_stack - PANIC_FRAME_OFFSET;
218 		mcesa_origin = pcpu->lowcore->mcesad & MCESA_ORIGIN_MASK;
219 		mcesa_bits = pcpu->lowcore->mcesad & MCESA_LC_MASK;
220 	}
221 	lc = pcpu->lowcore;
222 	memcpy(lc, &S390_lowcore, 512);
223 	memset((char *) lc + 512, 0, sizeof(*lc) - 512);
224 	lc->async_stack = async_stack + ASYNC_FRAME_OFFSET;
225 	lc->panic_stack = panic_stack + PANIC_FRAME_OFFSET;
226 	lc->mcesad = mcesa_origin | mcesa_bits;
227 	lc->cpu_nr = cpu;
228 	lc->spinlock_lockval = arch_spin_lockval(cpu);
229 	if (vdso_alloc_per_cpu(lc))
230 		goto out;
231 	lowcore_ptr[cpu] = lc;
232 	pcpu_sigp_retry(pcpu, SIGP_SET_PREFIX, (u32)(unsigned long) lc);
233 	return 0;
234 out:
235 	if (pcpu != &pcpu_devices[0]) {
236 		if (mcesa_origin)
237 			kmem_cache_free(pcpu_mcesa_cache,
238 					(void *) mcesa_origin);
239 		free_page(panic_stack);
240 		free_pages(async_stack, ASYNC_ORDER);
241 		free_pages((unsigned long) pcpu->lowcore, LC_ORDER);
242 	}
243 	return -ENOMEM;
244 }
245 
246 #ifdef CONFIG_HOTPLUG_CPU
247 
248 static void pcpu_free_lowcore(struct pcpu *pcpu)
249 {
250 	unsigned long mcesa_origin;
251 
252 	pcpu_sigp_retry(pcpu, SIGP_SET_PREFIX, 0);
253 	lowcore_ptr[pcpu - pcpu_devices] = NULL;
254 	vdso_free_per_cpu(pcpu->lowcore);
255 	if (pcpu == &pcpu_devices[0])
256 		return;
257 	if (MACHINE_HAS_VX || MACHINE_HAS_GS) {
258 		mcesa_origin = pcpu->lowcore->mcesad & MCESA_ORIGIN_MASK;
259 		kmem_cache_free(pcpu_mcesa_cache, (void *) mcesa_origin);
260 	}
261 	free_page(pcpu->lowcore->panic_stack-PANIC_FRAME_OFFSET);
262 	free_pages(pcpu->lowcore->async_stack-ASYNC_FRAME_OFFSET, ASYNC_ORDER);
263 	free_pages((unsigned long) pcpu->lowcore, LC_ORDER);
264 }
265 
266 #endif /* CONFIG_HOTPLUG_CPU */
267 
268 static void pcpu_prepare_secondary(struct pcpu *pcpu, int cpu)
269 {
270 	struct lowcore *lc = pcpu->lowcore;
271 
272 	cpumask_set_cpu(cpu, &init_mm.context.cpu_attach_mask);
273 	cpumask_set_cpu(cpu, mm_cpumask(&init_mm));
274 	lc->cpu_nr = cpu;
275 	lc->spinlock_lockval = arch_spin_lockval(cpu);
276 	lc->percpu_offset = __per_cpu_offset[cpu];
277 	lc->kernel_asce = S390_lowcore.kernel_asce;
278 	lc->machine_flags = S390_lowcore.machine_flags;
279 	lc->user_timer = lc->system_timer = lc->steal_timer = 0;
280 	__ctl_store(lc->cregs_save_area, 0, 15);
281 	save_access_regs((unsigned int *) lc->access_regs_save_area);
282 	memcpy(lc->stfle_fac_list, S390_lowcore.stfle_fac_list,
283 	       MAX_FACILITY_BIT/8);
284 }
285 
286 static void pcpu_attach_task(struct pcpu *pcpu, struct task_struct *tsk)
287 {
288 	struct lowcore *lc = pcpu->lowcore;
289 
290 	lc->kernel_stack = (unsigned long) task_stack_page(tsk)
291 		+ THREAD_SIZE - STACK_FRAME_OVERHEAD - sizeof(struct pt_regs);
292 	lc->current_task = (unsigned long) tsk;
293 	lc->lpp = LPP_MAGIC;
294 	lc->current_pid = tsk->pid;
295 	lc->user_timer = tsk->thread.user_timer;
296 	lc->system_timer = tsk->thread.system_timer;
297 	lc->steal_timer = 0;
298 }
299 
300 static void pcpu_start_fn(struct pcpu *pcpu, void (*func)(void *), void *data)
301 {
302 	struct lowcore *lc = pcpu->lowcore;
303 
304 	lc->restart_stack = lc->kernel_stack;
305 	lc->restart_fn = (unsigned long) func;
306 	lc->restart_data = (unsigned long) data;
307 	lc->restart_source = -1UL;
308 	pcpu_sigp_retry(pcpu, SIGP_RESTART, 0);
309 }
310 
311 /*
312  * Call function via PSW restart on pcpu and stop the current cpu.
313  */
314 static void pcpu_delegate(struct pcpu *pcpu, void (*func)(void *),
315 			  void *data, unsigned long stack)
316 {
317 	struct lowcore *lc = lowcore_ptr[pcpu - pcpu_devices];
318 	unsigned long source_cpu = stap();
319 
320 	__load_psw_mask(PSW_KERNEL_BITS);
321 	if (pcpu->address == source_cpu)
322 		func(data);	/* should not return */
323 	/* Stop target cpu (if func returns this stops the current cpu). */
324 	pcpu_sigp_retry(pcpu, SIGP_STOP, 0);
325 	/* Restart func on the target cpu and stop the current cpu. */
326 	mem_assign_absolute(lc->restart_stack, stack);
327 	mem_assign_absolute(lc->restart_fn, (unsigned long) func);
328 	mem_assign_absolute(lc->restart_data, (unsigned long) data);
329 	mem_assign_absolute(lc->restart_source, source_cpu);
330 	asm volatile(
331 		"0:	sigp	0,%0,%2	# sigp restart to target cpu\n"
332 		"	brc	2,0b	# busy, try again\n"
333 		"1:	sigp	0,%1,%3	# sigp stop to current cpu\n"
334 		"	brc	2,1b	# busy, try again\n"
335 		: : "d" (pcpu->address), "d" (source_cpu),
336 		    "K" (SIGP_RESTART), "K" (SIGP_STOP)
337 		: "0", "1", "cc");
338 	for (;;) ;
339 }
340 
341 /*
342  * Enable additional logical cpus for multi-threading.
343  */
344 static int pcpu_set_smt(unsigned int mtid)
345 {
346 	int cc;
347 
348 	if (smp_cpu_mtid == mtid)
349 		return 0;
350 	cc = __pcpu_sigp(0, SIGP_SET_MULTI_THREADING, mtid, NULL);
351 	if (cc == 0) {
352 		smp_cpu_mtid = mtid;
353 		smp_cpu_mt_shift = 0;
354 		while (smp_cpu_mtid >= (1U << smp_cpu_mt_shift))
355 			smp_cpu_mt_shift++;
356 		pcpu_devices[0].address = stap();
357 	}
358 	return cc;
359 }
360 
361 /*
362  * Call function on an online CPU.
363  */
364 void smp_call_online_cpu(void (*func)(void *), void *data)
365 {
366 	struct pcpu *pcpu;
367 
368 	/* Use the current cpu if it is online. */
369 	pcpu = pcpu_find_address(cpu_online_mask, stap());
370 	if (!pcpu)
371 		/* Use the first online cpu. */
372 		pcpu = pcpu_devices + cpumask_first(cpu_online_mask);
373 	pcpu_delegate(pcpu, func, data, (unsigned long) restart_stack);
374 }
375 
376 /*
377  * Call function on the ipl CPU.
378  */
379 void smp_call_ipl_cpu(void (*func)(void *), void *data)
380 {
381 	pcpu_delegate(&pcpu_devices[0], func, data,
382 		      pcpu_devices->lowcore->panic_stack -
383 		      PANIC_FRAME_OFFSET + PAGE_SIZE);
384 }
385 
386 int smp_find_processor_id(u16 address)
387 {
388 	int cpu;
389 
390 	for_each_present_cpu(cpu)
391 		if (pcpu_devices[cpu].address == address)
392 			return cpu;
393 	return -1;
394 }
395 
396 bool arch_vcpu_is_preempted(int cpu)
397 {
398 	if (test_cpu_flag_of(CIF_ENABLED_WAIT, cpu))
399 		return false;
400 	if (pcpu_running(pcpu_devices + cpu))
401 		return false;
402 	return true;
403 }
404 EXPORT_SYMBOL(arch_vcpu_is_preempted);
405 
406 void smp_yield_cpu(int cpu)
407 {
408 	if (MACHINE_HAS_DIAG9C) {
409 		diag_stat_inc_norecursion(DIAG_STAT_X09C);
410 		asm volatile("diag %0,0,0x9c"
411 			     : : "d" (pcpu_devices[cpu].address));
412 	} else if (MACHINE_HAS_DIAG44) {
413 		diag_stat_inc_norecursion(DIAG_STAT_X044);
414 		asm volatile("diag 0,0,0x44");
415 	}
416 }
417 
418 /*
419  * Send cpus emergency shutdown signal. This gives the cpus the
420  * opportunity to complete outstanding interrupts.
421  */
422 static void smp_emergency_stop(cpumask_t *cpumask)
423 {
424 	u64 end;
425 	int cpu;
426 
427 	end = get_tod_clock() + (1000000UL << 12);
428 	for_each_cpu(cpu, cpumask) {
429 		struct pcpu *pcpu = pcpu_devices + cpu;
430 		set_bit(ec_stop_cpu, &pcpu->ec_mask);
431 		while (__pcpu_sigp(pcpu->address, SIGP_EMERGENCY_SIGNAL,
432 				   0, NULL) == SIGP_CC_BUSY &&
433 		       get_tod_clock() < end)
434 			cpu_relax();
435 	}
436 	while (get_tod_clock() < end) {
437 		for_each_cpu(cpu, cpumask)
438 			if (pcpu_stopped(pcpu_devices + cpu))
439 				cpumask_clear_cpu(cpu, cpumask);
440 		if (cpumask_empty(cpumask))
441 			break;
442 		cpu_relax();
443 	}
444 }
445 
446 /*
447  * Stop all cpus but the current one.
448  */
449 void smp_send_stop(void)
450 {
451 	cpumask_t cpumask;
452 	int cpu;
453 
454 	/* Disable all interrupts/machine checks */
455 	__load_psw_mask(PSW_KERNEL_BITS | PSW_MASK_DAT);
456 	trace_hardirqs_off();
457 
458 	debug_set_critical();
459 	cpumask_copy(&cpumask, cpu_online_mask);
460 	cpumask_clear_cpu(smp_processor_id(), &cpumask);
461 
462 	if (oops_in_progress)
463 		smp_emergency_stop(&cpumask);
464 
465 	/* stop all processors */
466 	for_each_cpu(cpu, &cpumask) {
467 		struct pcpu *pcpu = pcpu_devices + cpu;
468 		pcpu_sigp_retry(pcpu, SIGP_STOP, 0);
469 		while (!pcpu_stopped(pcpu))
470 			cpu_relax();
471 	}
472 }
473 
474 /*
475  * This is the main routine where commands issued by other
476  * cpus are handled.
477  */
478 static void smp_handle_ext_call(void)
479 {
480 	unsigned long bits;
481 
482 	/* handle bit signal external calls */
483 	bits = xchg(&pcpu_devices[smp_processor_id()].ec_mask, 0);
484 	if (test_bit(ec_stop_cpu, &bits))
485 		smp_stop_cpu();
486 	if (test_bit(ec_schedule, &bits))
487 		scheduler_ipi();
488 	if (test_bit(ec_call_function_single, &bits))
489 		generic_smp_call_function_single_interrupt();
490 }
491 
492 static void do_ext_call_interrupt(struct ext_code ext_code,
493 				  unsigned int param32, unsigned long param64)
494 {
495 	inc_irq_stat(ext_code.code == 0x1202 ? IRQEXT_EXC : IRQEXT_EMS);
496 	smp_handle_ext_call();
497 }
498 
499 void arch_send_call_function_ipi_mask(const struct cpumask *mask)
500 {
501 	int cpu;
502 
503 	for_each_cpu(cpu, mask)
504 		pcpu_ec_call(pcpu_devices + cpu, ec_call_function_single);
505 }
506 
507 void arch_send_call_function_single_ipi(int cpu)
508 {
509 	pcpu_ec_call(pcpu_devices + cpu, ec_call_function_single);
510 }
511 
512 /*
513  * this function sends a 'reschedule' IPI to another CPU.
514  * it goes straight through and wastes no time serializing
515  * anything. Worst case is that we lose a reschedule ...
516  */
517 void smp_send_reschedule(int cpu)
518 {
519 	pcpu_ec_call(pcpu_devices + cpu, ec_schedule);
520 }
521 
522 /*
523  * parameter area for the set/clear control bit callbacks
524  */
525 struct ec_creg_mask_parms {
526 	unsigned long orval;
527 	unsigned long andval;
528 	int cr;
529 };
530 
531 /*
532  * callback for setting/clearing control bits
533  */
534 static void smp_ctl_bit_callback(void *info)
535 {
536 	struct ec_creg_mask_parms *pp = info;
537 	unsigned long cregs[16];
538 
539 	__ctl_store(cregs, 0, 15);
540 	cregs[pp->cr] = (cregs[pp->cr] & pp->andval) | pp->orval;
541 	__ctl_load(cregs, 0, 15);
542 }
543 
544 /*
545  * Set a bit in a control register of all cpus
546  */
547 void smp_ctl_set_bit(int cr, int bit)
548 {
549 	struct ec_creg_mask_parms parms = { 1UL << bit, -1UL, cr };
550 
551 	on_each_cpu(smp_ctl_bit_callback, &parms, 1);
552 }
553 EXPORT_SYMBOL(smp_ctl_set_bit);
554 
555 /*
556  * Clear a bit in a control register of all cpus
557  */
558 void smp_ctl_clear_bit(int cr, int bit)
559 {
560 	struct ec_creg_mask_parms parms = { 0, ~(1UL << bit), cr };
561 
562 	on_each_cpu(smp_ctl_bit_callback, &parms, 1);
563 }
564 EXPORT_SYMBOL(smp_ctl_clear_bit);
565 
566 #ifdef CONFIG_CRASH_DUMP
567 
568 int smp_store_status(int cpu)
569 {
570 	struct pcpu *pcpu = pcpu_devices + cpu;
571 	unsigned long pa;
572 
573 	pa = __pa(&pcpu->lowcore->floating_pt_save_area);
574 	if (__pcpu_sigp_relax(pcpu->address, SIGP_STORE_STATUS_AT_ADDRESS,
575 			      pa) != SIGP_CC_ORDER_CODE_ACCEPTED)
576 		return -EIO;
577 	if (!MACHINE_HAS_VX && !MACHINE_HAS_GS)
578 		return 0;
579 	pa = __pa(pcpu->lowcore->mcesad & MCESA_ORIGIN_MASK);
580 	if (MACHINE_HAS_GS)
581 		pa |= pcpu->lowcore->mcesad & MCESA_LC_MASK;
582 	if (__pcpu_sigp_relax(pcpu->address, SIGP_STORE_ADDITIONAL_STATUS,
583 			      pa) != SIGP_CC_ORDER_CODE_ACCEPTED)
584 		return -EIO;
585 	return 0;
586 }
587 
588 /*
589  * Collect CPU state of the previous, crashed system.
590  * There are four cases:
591  * 1) standard zfcp dump
592  *    condition: OLDMEM_BASE == NULL && ipl_info.type == IPL_TYPE_FCP_DUMP
593  *    The state for all CPUs except the boot CPU needs to be collected
594  *    with sigp stop-and-store-status. The boot CPU state is located in
595  *    the absolute lowcore of the memory stored in the HSA. The zcore code
596  *    will copy the boot CPU state from the HSA.
597  * 2) stand-alone kdump for SCSI (zfcp dump with swapped memory)
598  *    condition: OLDMEM_BASE != NULL && ipl_info.type == IPL_TYPE_FCP_DUMP
599  *    The state for all CPUs except the boot CPU needs to be collected
600  *    with sigp stop-and-store-status. The firmware or the boot-loader
601  *    stored the registers of the boot CPU in the absolute lowcore in the
602  *    memory of the old system.
603  * 3) kdump and the old kernel did not store the CPU state,
604  *    or stand-alone kdump for DASD
605  *    condition: OLDMEM_BASE != NULL && !is_kdump_kernel()
606  *    The state for all CPUs except the boot CPU needs to be collected
607  *    with sigp stop-and-store-status. The kexec code or the boot-loader
608  *    stored the registers of the boot CPU in the memory of the old system.
609  * 4) kdump and the old kernel stored the CPU state
610  *    condition: OLDMEM_BASE != NULL && is_kdump_kernel()
611  *    This case does not exist for s390 anymore, setup_arch explicitly
612  *    deactivates the elfcorehdr= kernel parameter
613  */
614 static __init void smp_save_cpu_vxrs(struct save_area *sa, u16 addr,
615 				     bool is_boot_cpu, unsigned long page)
616 {
617 	__vector128 *vxrs = (__vector128 *) page;
618 
619 	if (is_boot_cpu)
620 		vxrs = boot_cpu_vector_save_area;
621 	else
622 		__pcpu_sigp_relax(addr, SIGP_STORE_ADDITIONAL_STATUS, page);
623 	save_area_add_vxrs(sa, vxrs);
624 }
625 
626 static __init void smp_save_cpu_regs(struct save_area *sa, u16 addr,
627 				     bool is_boot_cpu, unsigned long page)
628 {
629 	void *regs = (void *) page;
630 
631 	if (is_boot_cpu)
632 		copy_oldmem_kernel(regs, (void *) __LC_FPREGS_SAVE_AREA, 512);
633 	else
634 		__pcpu_sigp_relax(addr, SIGP_STORE_STATUS_AT_ADDRESS, page);
635 	save_area_add_regs(sa, regs);
636 }
637 
638 void __init smp_save_dump_cpus(void)
639 {
640 	int addr, boot_cpu_addr, max_cpu_addr;
641 	struct save_area *sa;
642 	unsigned long page;
643 	bool is_boot_cpu;
644 
645 	if (!(OLDMEM_BASE || ipl_info.type == IPL_TYPE_FCP_DUMP))
646 		/* No previous system present, normal boot. */
647 		return;
648 	/* Allocate a page as dumping area for the store status sigps */
649 	page = memblock_alloc_base(PAGE_SIZE, PAGE_SIZE, 1UL << 31);
650 	/* Set multi-threading state to the previous system. */
651 	pcpu_set_smt(sclp.mtid_prev);
652 	boot_cpu_addr = stap();
653 	max_cpu_addr = SCLP_MAX_CORES << sclp.mtid_prev;
654 	for (addr = 0; addr <= max_cpu_addr; addr++) {
655 		if (__pcpu_sigp_relax(addr, SIGP_SENSE, 0) ==
656 		    SIGP_CC_NOT_OPERATIONAL)
657 			continue;
658 		is_boot_cpu = (addr == boot_cpu_addr);
659 		/* Allocate save area */
660 		sa = save_area_alloc(is_boot_cpu);
661 		if (!sa)
662 			panic("could not allocate memory for save area\n");
663 		if (MACHINE_HAS_VX)
664 			/* Get the vector registers */
665 			smp_save_cpu_vxrs(sa, addr, is_boot_cpu, page);
666 		/*
667 		 * For a zfcp dump OLDMEM_BASE == NULL and the registers
668 		 * of the boot CPU are stored in the HSA. To retrieve
669 		 * these registers an SCLP request is required which is
670 		 * done by drivers/s390/char/zcore.c:init_cpu_info()
671 		 */
672 		if (!is_boot_cpu || OLDMEM_BASE)
673 			/* Get the CPU registers */
674 			smp_save_cpu_regs(sa, addr, is_boot_cpu, page);
675 	}
676 	memblock_free(page, PAGE_SIZE);
677 	diag308_reset();
678 	pcpu_set_smt(0);
679 }
680 #endif /* CONFIG_CRASH_DUMP */
681 
682 void smp_cpu_set_polarization(int cpu, int val)
683 {
684 	pcpu_devices[cpu].polarization = val;
685 }
686 
687 int smp_cpu_get_polarization(int cpu)
688 {
689 	return pcpu_devices[cpu].polarization;
690 }
691 
692 static void __ref smp_get_core_info(struct sclp_core_info *info, int early)
693 {
694 	static int use_sigp_detection;
695 	int address;
696 
697 	if (use_sigp_detection || sclp_get_core_info(info, early)) {
698 		use_sigp_detection = 1;
699 		for (address = 0;
700 		     address < (SCLP_MAX_CORES << smp_cpu_mt_shift);
701 		     address += (1U << smp_cpu_mt_shift)) {
702 			if (__pcpu_sigp_relax(address, SIGP_SENSE, 0) ==
703 			    SIGP_CC_NOT_OPERATIONAL)
704 				continue;
705 			info->core[info->configured].core_id =
706 				address >> smp_cpu_mt_shift;
707 			info->configured++;
708 		}
709 		info->combined = info->configured;
710 	}
711 }
712 
713 static int smp_add_present_cpu(int cpu);
714 
715 static int __smp_rescan_cpus(struct sclp_core_info *info, int sysfs_add)
716 {
717 	struct pcpu *pcpu;
718 	cpumask_t avail;
719 	int cpu, nr, i, j;
720 	u16 address;
721 
722 	nr = 0;
723 	cpumask_xor(&avail, cpu_possible_mask, cpu_present_mask);
724 	cpu = cpumask_first(&avail);
725 	for (i = 0; (i < info->combined) && (cpu < nr_cpu_ids); i++) {
726 		if (sclp.has_core_type && info->core[i].type != boot_core_type)
727 			continue;
728 		address = info->core[i].core_id << smp_cpu_mt_shift;
729 		for (j = 0; j <= smp_cpu_mtid; j++) {
730 			if (pcpu_find_address(cpu_present_mask, address + j))
731 				continue;
732 			pcpu = pcpu_devices + cpu;
733 			pcpu->address = address + j;
734 			pcpu->state =
735 				(cpu >= info->configured*(smp_cpu_mtid + 1)) ?
736 				CPU_STATE_STANDBY : CPU_STATE_CONFIGURED;
737 			smp_cpu_set_polarization(cpu, POLARIZATION_UNKNOWN);
738 			set_cpu_present(cpu, true);
739 			if (sysfs_add && smp_add_present_cpu(cpu) != 0)
740 				set_cpu_present(cpu, false);
741 			else
742 				nr++;
743 			cpu = cpumask_next(cpu, &avail);
744 			if (cpu >= nr_cpu_ids)
745 				break;
746 		}
747 	}
748 	return nr;
749 }
750 
751 void __init smp_detect_cpus(void)
752 {
753 	unsigned int cpu, mtid, c_cpus, s_cpus;
754 	struct sclp_core_info *info;
755 	u16 address;
756 
757 	/* Get CPU information */
758 	info = memblock_virt_alloc(sizeof(*info), 8);
759 	smp_get_core_info(info, 1);
760 	/* Find boot CPU type */
761 	if (sclp.has_core_type) {
762 		address = stap();
763 		for (cpu = 0; cpu < info->combined; cpu++)
764 			if (info->core[cpu].core_id == address) {
765 				/* The boot cpu dictates the cpu type. */
766 				boot_core_type = info->core[cpu].type;
767 				break;
768 			}
769 		if (cpu >= info->combined)
770 			panic("Could not find boot CPU type");
771 	}
772 
773 	/* Set multi-threading state for the current system */
774 	mtid = boot_core_type ? sclp.mtid : sclp.mtid_cp;
775 	mtid = (mtid < smp_max_threads) ? mtid : smp_max_threads - 1;
776 	pcpu_set_smt(mtid);
777 
778 	/* Print number of CPUs */
779 	c_cpus = s_cpus = 0;
780 	for (cpu = 0; cpu < info->combined; cpu++) {
781 		if (sclp.has_core_type &&
782 		    info->core[cpu].type != boot_core_type)
783 			continue;
784 		if (cpu < info->configured)
785 			c_cpus += smp_cpu_mtid + 1;
786 		else
787 			s_cpus += smp_cpu_mtid + 1;
788 	}
789 	pr_info("%d configured CPUs, %d standby CPUs\n", c_cpus, s_cpus);
790 
791 	/* Add CPUs present at boot */
792 	get_online_cpus();
793 	__smp_rescan_cpus(info, 0);
794 	put_online_cpus();
795 	memblock_free_early((unsigned long)info, sizeof(*info));
796 }
797 
798 /*
799  *	Activate a secondary processor.
800  */
801 static void smp_start_secondary(void *cpuvoid)
802 {
803 	S390_lowcore.last_update_clock = get_tod_clock();
804 	S390_lowcore.restart_stack = (unsigned long) restart_stack;
805 	S390_lowcore.restart_fn = (unsigned long) do_restart;
806 	S390_lowcore.restart_data = 0;
807 	S390_lowcore.restart_source = -1UL;
808 	restore_access_regs(S390_lowcore.access_regs_save_area);
809 	__ctl_load(S390_lowcore.cregs_save_area, 0, 15);
810 	__load_psw_mask(PSW_KERNEL_BITS | PSW_MASK_DAT);
811 	cpu_init();
812 	preempt_disable();
813 	init_cpu_timer();
814 	vtime_init();
815 	pfault_init();
816 	notify_cpu_starting(smp_processor_id());
817 	set_cpu_online(smp_processor_id(), true);
818 	inc_irq_stat(CPU_RST);
819 	local_irq_enable();
820 	cpu_startup_entry(CPUHP_AP_ONLINE_IDLE);
821 }
822 
823 /* Upping and downing of CPUs */
824 int __cpu_up(unsigned int cpu, struct task_struct *tidle)
825 {
826 	struct pcpu *pcpu;
827 	int base, i, rc;
828 
829 	pcpu = pcpu_devices + cpu;
830 	if (pcpu->state != CPU_STATE_CONFIGURED)
831 		return -EIO;
832 	base = smp_get_base_cpu(cpu);
833 	for (i = 0; i <= smp_cpu_mtid; i++) {
834 		if (base + i < nr_cpu_ids)
835 			if (cpu_online(base + i))
836 				break;
837 	}
838 	/*
839 	 * If this is the first CPU of the core to get online
840 	 * do an initial CPU reset.
841 	 */
842 	if (i > smp_cpu_mtid &&
843 	    pcpu_sigp_retry(pcpu_devices + base, SIGP_INITIAL_CPU_RESET, 0) !=
844 	    SIGP_CC_ORDER_CODE_ACCEPTED)
845 		return -EIO;
846 
847 	rc = pcpu_alloc_lowcore(pcpu, cpu);
848 	if (rc)
849 		return rc;
850 	pcpu_prepare_secondary(pcpu, cpu);
851 	pcpu_attach_task(pcpu, tidle);
852 	pcpu_start_fn(pcpu, smp_start_secondary, NULL);
853 	/* Wait until cpu puts itself in the online & active maps */
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 	cpumask_clear_cpu(cpu, mm_cpumask(&init_mm));
899 	cpumask_clear_cpu(cpu, &init_mm.context.cpu_attach_mask);
900 }
901 
902 void __noreturn cpu_die(void)
903 {
904 	idle_task_exit();
905 	pcpu_sigp_retry(pcpu_devices + smp_processor_id(), SIGP_STOP, 0);
906 	for (;;) ;
907 }
908 
909 #endif /* CONFIG_HOTPLUG_CPU */
910 
911 void __init smp_fill_possible_mask(void)
912 {
913 	unsigned int possible, sclp_max, cpu;
914 
915 	sclp_max = max(sclp.mtid, sclp.mtid_cp) + 1;
916 	sclp_max = min(smp_max_threads, sclp_max);
917 	sclp_max = (sclp.max_cores * sclp_max) ?: nr_cpu_ids;
918 	possible = setup_possible_cpus ?: nr_cpu_ids;
919 	possible = min(possible, sclp_max);
920 	for (cpu = 0; cpu < possible && cpu < nr_cpu_ids; cpu++)
921 		set_cpu_possible(cpu, true);
922 }
923 
924 void __init smp_prepare_cpus(unsigned int max_cpus)
925 {
926 	unsigned long size;
927 
928 	/* request the 0x1201 emergency signal external interrupt */
929 	if (register_external_irq(EXT_IRQ_EMERGENCY_SIG, do_ext_call_interrupt))
930 		panic("Couldn't request external interrupt 0x1201");
931 	/* request the 0x1202 external call external interrupt */
932 	if (register_external_irq(EXT_IRQ_EXTERNAL_CALL, do_ext_call_interrupt))
933 		panic("Couldn't request external interrupt 0x1202");
934 	/* create slab cache for the machine-check-extended-save-areas */
935 	if (MACHINE_HAS_VX || MACHINE_HAS_GS) {
936 		size = 1UL << (MACHINE_HAS_GS ? 11 : 10);
937 		pcpu_mcesa_cache = kmem_cache_create("nmi_save_areas",
938 						     size, size, 0, NULL);
939 		if (!pcpu_mcesa_cache)
940 			panic("Couldn't create nmi save area cache");
941 	}
942 }
943 
944 void __init smp_prepare_boot_cpu(void)
945 {
946 	struct pcpu *pcpu = pcpu_devices;
947 
948 	WARN_ON(!cpu_present(0) || !cpu_online(0));
949 	pcpu->state = CPU_STATE_CONFIGURED;
950 	pcpu->lowcore = (struct lowcore *)(unsigned long) store_prefix();
951 	S390_lowcore.percpu_offset = __per_cpu_offset[0];
952 	smp_cpu_set_polarization(0, POLARIZATION_UNKNOWN);
953 }
954 
955 void __init smp_cpus_done(unsigned int max_cpus)
956 {
957 }
958 
959 void __init smp_setup_processor_id(void)
960 {
961 	pcpu_devices[0].address = stap();
962 	S390_lowcore.cpu_nr = 0;
963 	S390_lowcore.spinlock_lockval = arch_spin_lockval(0);
964 }
965 
966 /*
967  * the frequency of the profiling timer can be changed
968  * by writing a multiplier value into /proc/profile.
969  *
970  * usually you want to run this on all CPUs ;)
971  */
972 int setup_profiling_timer(unsigned int multiplier)
973 {
974 	return 0;
975 }
976 
977 #ifdef CONFIG_HOTPLUG_CPU
978 static ssize_t cpu_configure_show(struct device *dev,
979 				  struct device_attribute *attr, char *buf)
980 {
981 	ssize_t count;
982 
983 	mutex_lock(&smp_cpu_state_mutex);
984 	count = sprintf(buf, "%d\n", pcpu_devices[dev->id].state);
985 	mutex_unlock(&smp_cpu_state_mutex);
986 	return count;
987 }
988 
989 static ssize_t cpu_configure_store(struct device *dev,
990 				   struct device_attribute *attr,
991 				   const char *buf, size_t count)
992 {
993 	struct pcpu *pcpu;
994 	int cpu, val, rc, i;
995 	char delim;
996 
997 	if (sscanf(buf, "%d %c", &val, &delim) != 1)
998 		return -EINVAL;
999 	if (val != 0 && val != 1)
1000 		return -EINVAL;
1001 	get_online_cpus();
1002 	mutex_lock(&smp_cpu_state_mutex);
1003 	rc = -EBUSY;
1004 	/* disallow configuration changes of online cpus and cpu 0 */
1005 	cpu = dev->id;
1006 	cpu = smp_get_base_cpu(cpu);
1007 	if (cpu == 0)
1008 		goto out;
1009 	for (i = 0; i <= smp_cpu_mtid; i++)
1010 		if (cpu_online(cpu + i))
1011 			goto out;
1012 	pcpu = pcpu_devices + cpu;
1013 	rc = 0;
1014 	switch (val) {
1015 	case 0:
1016 		if (pcpu->state != CPU_STATE_CONFIGURED)
1017 			break;
1018 		rc = sclp_core_deconfigure(pcpu->address >> smp_cpu_mt_shift);
1019 		if (rc)
1020 			break;
1021 		for (i = 0; i <= smp_cpu_mtid; i++) {
1022 			if (cpu + i >= nr_cpu_ids || !cpu_present(cpu + i))
1023 				continue;
1024 			pcpu[i].state = CPU_STATE_STANDBY;
1025 			smp_cpu_set_polarization(cpu + i,
1026 						 POLARIZATION_UNKNOWN);
1027 		}
1028 		topology_expect_change();
1029 		break;
1030 	case 1:
1031 		if (pcpu->state != CPU_STATE_STANDBY)
1032 			break;
1033 		rc = sclp_core_configure(pcpu->address >> smp_cpu_mt_shift);
1034 		if (rc)
1035 			break;
1036 		for (i = 0; i <= smp_cpu_mtid; i++) {
1037 			if (cpu + i >= nr_cpu_ids || !cpu_present(cpu + i))
1038 				continue;
1039 			pcpu[i].state = CPU_STATE_CONFIGURED;
1040 			smp_cpu_set_polarization(cpu + i,
1041 						 POLARIZATION_UNKNOWN);
1042 		}
1043 		topology_expect_change();
1044 		break;
1045 	default:
1046 		break;
1047 	}
1048 out:
1049 	mutex_unlock(&smp_cpu_state_mutex);
1050 	put_online_cpus();
1051 	return rc ? rc : count;
1052 }
1053 static DEVICE_ATTR(configure, 0644, cpu_configure_show, cpu_configure_store);
1054 #endif /* CONFIG_HOTPLUG_CPU */
1055 
1056 static ssize_t show_cpu_address(struct device *dev,
1057 				struct device_attribute *attr, char *buf)
1058 {
1059 	return sprintf(buf, "%d\n", pcpu_devices[dev->id].address);
1060 }
1061 static DEVICE_ATTR(address, 0444, show_cpu_address, NULL);
1062 
1063 static struct attribute *cpu_common_attrs[] = {
1064 #ifdef CONFIG_HOTPLUG_CPU
1065 	&dev_attr_configure.attr,
1066 #endif
1067 	&dev_attr_address.attr,
1068 	NULL,
1069 };
1070 
1071 static struct attribute_group cpu_common_attr_group = {
1072 	.attrs = cpu_common_attrs,
1073 };
1074 
1075 static struct attribute *cpu_online_attrs[] = {
1076 	&dev_attr_idle_count.attr,
1077 	&dev_attr_idle_time_us.attr,
1078 	NULL,
1079 };
1080 
1081 static struct attribute_group cpu_online_attr_group = {
1082 	.attrs = cpu_online_attrs,
1083 };
1084 
1085 static int smp_cpu_online(unsigned int cpu)
1086 {
1087 	struct device *s = &per_cpu(cpu_device, cpu)->dev;
1088 
1089 	return sysfs_create_group(&s->kobj, &cpu_online_attr_group);
1090 }
1091 static int smp_cpu_pre_down(unsigned int cpu)
1092 {
1093 	struct device *s = &per_cpu(cpu_device, cpu)->dev;
1094 
1095 	sysfs_remove_group(&s->kobj, &cpu_online_attr_group);
1096 	return 0;
1097 }
1098 
1099 static int smp_add_present_cpu(int cpu)
1100 {
1101 	struct device *s;
1102 	struct cpu *c;
1103 	int rc;
1104 
1105 	c = kzalloc(sizeof(*c), GFP_KERNEL);
1106 	if (!c)
1107 		return -ENOMEM;
1108 	per_cpu(cpu_device, cpu) = c;
1109 	s = &c->dev;
1110 	c->hotpluggable = 1;
1111 	rc = register_cpu(c, cpu);
1112 	if (rc)
1113 		goto out;
1114 	rc = sysfs_create_group(&s->kobj, &cpu_common_attr_group);
1115 	if (rc)
1116 		goto out_cpu;
1117 	rc = topology_cpu_init(c);
1118 	if (rc)
1119 		goto out_topology;
1120 	return 0;
1121 
1122 out_topology:
1123 	sysfs_remove_group(&s->kobj, &cpu_common_attr_group);
1124 out_cpu:
1125 #ifdef CONFIG_HOTPLUG_CPU
1126 	unregister_cpu(c);
1127 #endif
1128 out:
1129 	return rc;
1130 }
1131 
1132 #ifdef CONFIG_HOTPLUG_CPU
1133 
1134 int __ref smp_rescan_cpus(void)
1135 {
1136 	struct sclp_core_info *info;
1137 	int nr;
1138 
1139 	info = kzalloc(sizeof(*info), GFP_KERNEL);
1140 	if (!info)
1141 		return -ENOMEM;
1142 	smp_get_core_info(info, 0);
1143 	get_online_cpus();
1144 	mutex_lock(&smp_cpu_state_mutex);
1145 	nr = __smp_rescan_cpus(info, 1);
1146 	mutex_unlock(&smp_cpu_state_mutex);
1147 	put_online_cpus();
1148 	kfree(info);
1149 	if (nr)
1150 		topology_schedule_update();
1151 	return 0;
1152 }
1153 
1154 static ssize_t __ref rescan_store(struct device *dev,
1155 				  struct device_attribute *attr,
1156 				  const char *buf,
1157 				  size_t count)
1158 {
1159 	int rc;
1160 
1161 	rc = smp_rescan_cpus();
1162 	return rc ? rc : count;
1163 }
1164 static DEVICE_ATTR(rescan, 0200, NULL, rescan_store);
1165 #endif /* CONFIG_HOTPLUG_CPU */
1166 
1167 static int __init s390_smp_init(void)
1168 {
1169 	int cpu, rc = 0;
1170 
1171 #ifdef CONFIG_HOTPLUG_CPU
1172 	rc = device_create_file(cpu_subsys.dev_root, &dev_attr_rescan);
1173 	if (rc)
1174 		return rc;
1175 #endif
1176 	for_each_present_cpu(cpu) {
1177 		rc = smp_add_present_cpu(cpu);
1178 		if (rc)
1179 			goto out;
1180 	}
1181 
1182 	rc = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "s390/smp:online",
1183 			       smp_cpu_online, smp_cpu_pre_down);
1184 	rc = rc <= 0 ? rc : 0;
1185 out:
1186 	return rc;
1187 }
1188 subsys_initcall(s390_smp_init);
1189