Loading...
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 "entry.h"
49
50enum {
51 ec_schedule = 0,
52 ec_call_function_single,
53 ec_stop_cpu,
54};
55
56enum {
57 CPU_STATE_STANDBY,
58 CPU_STATE_CONFIGURED,
59};
60
61struct pcpu {
62 struct cpu *cpu;
63 struct _lowcore *lowcore; /* lowcore page(s) for the cpu */
64 unsigned long async_stack; /* async stack for the cpu */
65 unsigned long panic_stack; /* panic stack for the cpu */
66 unsigned long ec_mask; /* bit mask for ec_xxx functions */
67 int state; /* physical cpu state */
68 int polarization; /* physical polarization */
69 u16 address; /* physical cpu address */
70};
71
72static u8 boot_cpu_type;
73static u16 boot_cpu_address;
74static struct pcpu pcpu_devices[NR_CPUS];
75
76/*
77 * The smp_cpu_state_mutex must be held when changing the state or polarization
78 * member of a pcpu data structure within the pcpu_devices arreay.
79 */
80DEFINE_MUTEX(smp_cpu_state_mutex);
81
82/*
83 * Signal processor helper functions.
84 */
85static inline int __pcpu_sigp_relax(u16 addr, u8 order, u32 parm, u32 *status)
86{
87 int cc;
88
89 while (1) {
90 cc = __pcpu_sigp(addr, order, parm, NULL);
91 if (cc != SIGP_CC_BUSY)
92 return cc;
93 cpu_relax();
94 }
95}
96
97static int pcpu_sigp_retry(struct pcpu *pcpu, u8 order, u32 parm)
98{
99 int cc, retry;
100
101 for (retry = 0; ; retry++) {
102 cc = __pcpu_sigp(pcpu->address, order, parm, NULL);
103 if (cc != SIGP_CC_BUSY)
104 break;
105 if (retry >= 3)
106 udelay(10);
107 }
108 return cc;
109}
110
111static inline int pcpu_stopped(struct pcpu *pcpu)
112{
113 u32 uninitialized_var(status);
114
115 if (__pcpu_sigp(pcpu->address, SIGP_SENSE,
116 0, &status) != SIGP_CC_STATUS_STORED)
117 return 0;
118 return !!(status & (SIGP_STATUS_CHECK_STOP|SIGP_STATUS_STOPPED));
119}
120
121static inline int pcpu_running(struct pcpu *pcpu)
122{
123 if (__pcpu_sigp(pcpu->address, SIGP_SENSE_RUNNING,
124 0, NULL) != SIGP_CC_STATUS_STORED)
125 return 1;
126 /* Status stored condition code is equivalent to cpu not running. */
127 return 0;
128}
129
130/*
131 * Find struct pcpu by cpu address.
132 */
133static struct pcpu *pcpu_find_address(const struct cpumask *mask, int address)
134{
135 int cpu;
136
137 for_each_cpu(cpu, mask)
138 if (pcpu_devices[cpu].address == address)
139 return pcpu_devices + cpu;
140 return NULL;
141}
142
143static void pcpu_ec_call(struct pcpu *pcpu, int ec_bit)
144{
145 int order;
146
147 if (test_and_set_bit(ec_bit, &pcpu->ec_mask))
148 return;
149 order = pcpu_running(pcpu) ? SIGP_EXTERNAL_CALL : SIGP_EMERGENCY_SIGNAL;
150 pcpu_sigp_retry(pcpu, order, 0);
151}
152
153static int pcpu_alloc_lowcore(struct pcpu *pcpu, int cpu)
154{
155 struct _lowcore *lc;
156
157 if (pcpu != &pcpu_devices[0]) {
158 pcpu->lowcore = (struct _lowcore *)
159 __get_free_pages(GFP_KERNEL | GFP_DMA, LC_ORDER);
160 pcpu->async_stack = __get_free_pages(GFP_KERNEL, ASYNC_ORDER);
161 pcpu->panic_stack = __get_free_page(GFP_KERNEL);
162 if (!pcpu->lowcore || !pcpu->panic_stack || !pcpu->async_stack)
163 goto out;
164 }
165 lc = pcpu->lowcore;
166 memcpy(lc, &S390_lowcore, 512);
167 memset((char *) lc + 512, 0, sizeof(*lc) - 512);
168 lc->async_stack = pcpu->async_stack + ASYNC_SIZE
169 - STACK_FRAME_OVERHEAD - sizeof(struct pt_regs);
170 lc->panic_stack = pcpu->panic_stack + PAGE_SIZE
171 - STACK_FRAME_OVERHEAD - sizeof(struct pt_regs);
172 lc->cpu_nr = cpu;
173#ifndef CONFIG_64BIT
174 if (MACHINE_HAS_IEEE) {
175 lc->extended_save_area_addr = get_zeroed_page(GFP_KERNEL);
176 if (!lc->extended_save_area_addr)
177 goto out;
178 }
179#else
180 if (vdso_alloc_per_cpu(lc))
181 goto out;
182#endif
183 lowcore_ptr[cpu] = lc;
184 pcpu_sigp_retry(pcpu, SIGP_SET_PREFIX, (u32)(unsigned long) lc);
185 return 0;
186out:
187 if (pcpu != &pcpu_devices[0]) {
188 free_page(pcpu->panic_stack);
189 free_pages(pcpu->async_stack, ASYNC_ORDER);
190 free_pages((unsigned long) pcpu->lowcore, LC_ORDER);
191 }
192 return -ENOMEM;
193}
194
195#ifdef CONFIG_HOTPLUG_CPU
196
197static void pcpu_free_lowcore(struct pcpu *pcpu)
198{
199 pcpu_sigp_retry(pcpu, SIGP_SET_PREFIX, 0);
200 lowcore_ptr[pcpu - pcpu_devices] = NULL;
201#ifndef CONFIG_64BIT
202 if (MACHINE_HAS_IEEE) {
203 struct _lowcore *lc = pcpu->lowcore;
204
205 free_page((unsigned long) lc->extended_save_area_addr);
206 lc->extended_save_area_addr = 0;
207 }
208#else
209 vdso_free_per_cpu(pcpu->lowcore);
210#endif
211 if (pcpu != &pcpu_devices[0]) {
212 free_page(pcpu->panic_stack);
213 free_pages(pcpu->async_stack, ASYNC_ORDER);
214 free_pages((unsigned long) pcpu->lowcore, LC_ORDER);
215 }
216}
217
218#endif /* CONFIG_HOTPLUG_CPU */
219
220static void pcpu_prepare_secondary(struct pcpu *pcpu, int cpu)
221{
222 struct _lowcore *lc = pcpu->lowcore;
223
224 if (MACHINE_HAS_TLB_LC)
225 cpumask_set_cpu(cpu, &init_mm.context.cpu_attach_mask);
226 cpumask_set_cpu(cpu, mm_cpumask(&init_mm));
227 atomic_inc(&init_mm.context.attach_count);
228 lc->cpu_nr = cpu;
229 lc->percpu_offset = __per_cpu_offset[cpu];
230 lc->kernel_asce = S390_lowcore.kernel_asce;
231 lc->machine_flags = S390_lowcore.machine_flags;
232 lc->ftrace_func = S390_lowcore.ftrace_func;
233 lc->user_timer = lc->system_timer = lc->steal_timer = 0;
234 __ctl_store(lc->cregs_save_area, 0, 15);
235 save_access_regs((unsigned int *) lc->access_regs_save_area);
236 memcpy(lc->stfle_fac_list, S390_lowcore.stfle_fac_list,
237 MAX_FACILITY_BIT/8);
238}
239
240static void pcpu_attach_task(struct pcpu *pcpu, struct task_struct *tsk)
241{
242 struct _lowcore *lc = pcpu->lowcore;
243 struct thread_info *ti = task_thread_info(tsk);
244
245 lc->kernel_stack = (unsigned long) task_stack_page(tsk)
246 + THREAD_SIZE - STACK_FRAME_OVERHEAD - sizeof(struct pt_regs);
247 lc->thread_info = (unsigned long) task_thread_info(tsk);
248 lc->current_task = (unsigned long) tsk;
249 lc->user_timer = ti->user_timer;
250 lc->system_timer = ti->system_timer;
251 lc->steal_timer = 0;
252}
253
254static void pcpu_start_fn(struct pcpu *pcpu, void (*func)(void *), void *data)
255{
256 struct _lowcore *lc = pcpu->lowcore;
257
258 lc->restart_stack = lc->kernel_stack;
259 lc->restart_fn = (unsigned long) func;
260 lc->restart_data = (unsigned long) data;
261 lc->restart_source = -1UL;
262 pcpu_sigp_retry(pcpu, SIGP_RESTART, 0);
263}
264
265/*
266 * Call function via PSW restart on pcpu and stop the current cpu.
267 */
268static void pcpu_delegate(struct pcpu *pcpu, void (*func)(void *),
269 void *data, unsigned long stack)
270{
271 struct _lowcore *lc = lowcore_ptr[pcpu - pcpu_devices];
272 unsigned long source_cpu = stap();
273
274 __load_psw_mask(PSW_KERNEL_BITS);
275 if (pcpu->address == source_cpu)
276 func(data); /* should not return */
277 /* Stop target cpu (if func returns this stops the current cpu). */
278 pcpu_sigp_retry(pcpu, SIGP_STOP, 0);
279 /* Restart func on the target cpu and stop the current cpu. */
280 mem_assign_absolute(lc->restart_stack, stack);
281 mem_assign_absolute(lc->restart_fn, (unsigned long) func);
282 mem_assign_absolute(lc->restart_data, (unsigned long) data);
283 mem_assign_absolute(lc->restart_source, source_cpu);
284 asm volatile(
285 "0: sigp 0,%0,%2 # sigp restart to target cpu\n"
286 " brc 2,0b # busy, try again\n"
287 "1: sigp 0,%1,%3 # sigp stop to current cpu\n"
288 " brc 2,1b # busy, try again\n"
289 : : "d" (pcpu->address), "d" (source_cpu),
290 "K" (SIGP_RESTART), "K" (SIGP_STOP)
291 : "0", "1", "cc");
292 for (;;) ;
293}
294
295/*
296 * Call function on an online CPU.
297 */
298void smp_call_online_cpu(void (*func)(void *), void *data)
299{
300 struct pcpu *pcpu;
301
302 /* Use the current cpu if it is online. */
303 pcpu = pcpu_find_address(cpu_online_mask, stap());
304 if (!pcpu)
305 /* Use the first online cpu. */
306 pcpu = pcpu_devices + cpumask_first(cpu_online_mask);
307 pcpu_delegate(pcpu, func, data, (unsigned long) restart_stack);
308}
309
310/*
311 * Call function on the ipl CPU.
312 */
313void smp_call_ipl_cpu(void (*func)(void *), void *data)
314{
315 pcpu_delegate(&pcpu_devices[0], func, data,
316 pcpu_devices->panic_stack + PAGE_SIZE);
317}
318
319int smp_find_processor_id(u16 address)
320{
321 int cpu;
322
323 for_each_present_cpu(cpu)
324 if (pcpu_devices[cpu].address == address)
325 return cpu;
326 return -1;
327}
328
329int smp_vcpu_scheduled(int cpu)
330{
331 return pcpu_running(pcpu_devices + cpu);
332}
333
334void smp_yield(void)
335{
336 if (MACHINE_HAS_DIAG44)
337 asm volatile("diag 0,0,0x44");
338}
339
340void smp_yield_cpu(int cpu)
341{
342 if (MACHINE_HAS_DIAG9C)
343 asm volatile("diag %0,0,0x9c"
344 : : "d" (pcpu_devices[cpu].address));
345 else if (MACHINE_HAS_DIAG44)
346 asm volatile("diag 0,0,0x44");
347}
348
349/*
350 * Send cpus emergency shutdown signal. This gives the cpus the
351 * opportunity to complete outstanding interrupts.
352 */
353static void smp_emergency_stop(cpumask_t *cpumask)
354{
355 u64 end;
356 int cpu;
357
358 end = get_tod_clock() + (1000000UL << 12);
359 for_each_cpu(cpu, cpumask) {
360 struct pcpu *pcpu = pcpu_devices + cpu;
361 set_bit(ec_stop_cpu, &pcpu->ec_mask);
362 while (__pcpu_sigp(pcpu->address, SIGP_EMERGENCY_SIGNAL,
363 0, NULL) == SIGP_CC_BUSY &&
364 get_tod_clock() < end)
365 cpu_relax();
366 }
367 while (get_tod_clock() < end) {
368 for_each_cpu(cpu, cpumask)
369 if (pcpu_stopped(pcpu_devices + cpu))
370 cpumask_clear_cpu(cpu, cpumask);
371 if (cpumask_empty(cpumask))
372 break;
373 cpu_relax();
374 }
375}
376
377/*
378 * Stop all cpus but the current one.
379 */
380void smp_send_stop(void)
381{
382 cpumask_t cpumask;
383 int cpu;
384
385 /* Disable all interrupts/machine checks */
386 __load_psw_mask(PSW_KERNEL_BITS | PSW_MASK_DAT);
387 trace_hardirqs_off();
388
389 debug_set_critical();
390 cpumask_copy(&cpumask, cpu_online_mask);
391 cpumask_clear_cpu(smp_processor_id(), &cpumask);
392
393 if (oops_in_progress)
394 smp_emergency_stop(&cpumask);
395
396 /* stop all processors */
397 for_each_cpu(cpu, &cpumask) {
398 struct pcpu *pcpu = pcpu_devices + cpu;
399 pcpu_sigp_retry(pcpu, SIGP_STOP, 0);
400 while (!pcpu_stopped(pcpu))
401 cpu_relax();
402 }
403}
404
405/*
406 * Stop the current cpu.
407 */
408void smp_stop_cpu(void)
409{
410 pcpu_sigp_retry(pcpu_devices + smp_processor_id(), SIGP_STOP, 0);
411 for (;;) ;
412}
413
414/*
415 * This is the main routine where commands issued by other
416 * cpus are handled.
417 */
418static void smp_handle_ext_call(void)
419{
420 unsigned long bits;
421
422 /* handle bit signal external calls */
423 bits = xchg(&pcpu_devices[smp_processor_id()].ec_mask, 0);
424 if (test_bit(ec_stop_cpu, &bits))
425 smp_stop_cpu();
426 if (test_bit(ec_schedule, &bits))
427 scheduler_ipi();
428 if (test_bit(ec_call_function_single, &bits))
429 generic_smp_call_function_single_interrupt();
430}
431
432static void do_ext_call_interrupt(struct ext_code ext_code,
433 unsigned int param32, unsigned long param64)
434{
435 inc_irq_stat(ext_code.code == 0x1202 ? IRQEXT_EXC : IRQEXT_EMS);
436 smp_handle_ext_call();
437}
438
439void arch_send_call_function_ipi_mask(const struct cpumask *mask)
440{
441 int cpu;
442
443 for_each_cpu(cpu, mask)
444 pcpu_ec_call(pcpu_devices + cpu, ec_call_function_single);
445}
446
447void arch_send_call_function_single_ipi(int cpu)
448{
449 pcpu_ec_call(pcpu_devices + cpu, ec_call_function_single);
450}
451
452#ifndef CONFIG_64BIT
453/*
454 * this function sends a 'purge tlb' signal to another CPU.
455 */
456static void smp_ptlb_callback(void *info)
457{
458 __tlb_flush_local();
459}
460
461void smp_ptlb_all(void)
462{
463 on_each_cpu(smp_ptlb_callback, NULL, 1);
464}
465EXPORT_SYMBOL(smp_ptlb_all);
466#endif /* ! CONFIG_64BIT */
467
468/*
469 * this function sends a 'reschedule' IPI to another CPU.
470 * it goes straight through and wastes no time serializing
471 * anything. Worst case is that we lose a reschedule ...
472 */
473void smp_send_reschedule(int cpu)
474{
475 pcpu_ec_call(pcpu_devices + cpu, ec_schedule);
476}
477
478/*
479 * parameter area for the set/clear control bit callbacks
480 */
481struct ec_creg_mask_parms {
482 unsigned long orval;
483 unsigned long andval;
484 int cr;
485};
486
487/*
488 * callback for setting/clearing control bits
489 */
490static void smp_ctl_bit_callback(void *info)
491{
492 struct ec_creg_mask_parms *pp = info;
493 unsigned long cregs[16];
494
495 __ctl_store(cregs, 0, 15);
496 cregs[pp->cr] = (cregs[pp->cr] & pp->andval) | pp->orval;
497 __ctl_load(cregs, 0, 15);
498}
499
500/*
501 * Set a bit in a control register of all cpus
502 */
503void smp_ctl_set_bit(int cr, int bit)
504{
505 struct ec_creg_mask_parms parms = { 1UL << bit, -1UL, cr };
506
507 on_each_cpu(smp_ctl_bit_callback, &parms, 1);
508}
509EXPORT_SYMBOL(smp_ctl_set_bit);
510
511/*
512 * Clear a bit in a control register of all cpus
513 */
514void smp_ctl_clear_bit(int cr, int bit)
515{
516 struct ec_creg_mask_parms parms = { 0, ~(1UL << bit), cr };
517
518 on_each_cpu(smp_ctl_bit_callback, &parms, 1);
519}
520EXPORT_SYMBOL(smp_ctl_clear_bit);
521
522#if defined(CONFIG_ZFCPDUMP) || defined(CONFIG_CRASH_DUMP)
523
524static void __init smp_get_save_area(int cpu, u16 address)
525{
526 void *lc = pcpu_devices[0].lowcore;
527 struct save_area *save_area;
528
529 if (is_kdump_kernel())
530 return;
531 if (!OLDMEM_BASE && (address == boot_cpu_address ||
532 ipl_info.type != IPL_TYPE_FCP_DUMP))
533 return;
534 save_area = dump_save_area_create(cpu);
535 if (!save_area)
536 panic("could not allocate memory for save area\n");
537#ifdef CONFIG_CRASH_DUMP
538 if (address == boot_cpu_address) {
539 /* Copy the registers of the boot cpu. */
540 copy_oldmem_page(1, (void *) save_area, sizeof(*save_area),
541 SAVE_AREA_BASE - PAGE_SIZE, 0);
542 return;
543 }
544#endif
545 /* Get the registers of a non-boot cpu. */
546 __pcpu_sigp_relax(address, SIGP_STOP_AND_STORE_STATUS, 0, NULL);
547 memcpy_real(save_area, lc + SAVE_AREA_BASE, sizeof(*save_area));
548}
549
550int smp_store_status(int cpu)
551{
552 struct pcpu *pcpu;
553
554 pcpu = pcpu_devices + cpu;
555 if (__pcpu_sigp_relax(pcpu->address, SIGP_STOP_AND_STORE_STATUS,
556 0, NULL) != SIGP_CC_ORDER_CODE_ACCEPTED)
557 return -EIO;
558 return 0;
559}
560
561#else /* CONFIG_ZFCPDUMP || CONFIG_CRASH_DUMP */
562
563static inline void smp_get_save_area(int cpu, u16 address) { }
564
565#endif /* CONFIG_ZFCPDUMP || CONFIG_CRASH_DUMP */
566
567void smp_cpu_set_polarization(int cpu, int val)
568{
569 pcpu_devices[cpu].polarization = val;
570}
571
572int smp_cpu_get_polarization(int cpu)
573{
574 return pcpu_devices[cpu].polarization;
575}
576
577static struct sclp_cpu_info *smp_get_cpu_info(void)
578{
579 static int use_sigp_detection;
580 struct sclp_cpu_info *info;
581 int address;
582
583 info = kzalloc(sizeof(*info), GFP_KERNEL);
584 if (info && (use_sigp_detection || sclp_get_cpu_info(info))) {
585 use_sigp_detection = 1;
586 for (address = 0; address <= MAX_CPU_ADDRESS; address++) {
587 if (__pcpu_sigp_relax(address, SIGP_SENSE, 0, NULL) ==
588 SIGP_CC_NOT_OPERATIONAL)
589 continue;
590 info->cpu[info->configured].address = address;
591 info->configured++;
592 }
593 info->combined = info->configured;
594 }
595 return info;
596}
597
598static int smp_add_present_cpu(int cpu);
599
600static int __smp_rescan_cpus(struct sclp_cpu_info *info, int sysfs_add)
601{
602 struct pcpu *pcpu;
603 cpumask_t avail;
604 int cpu, nr, i;
605
606 nr = 0;
607 cpumask_xor(&avail, cpu_possible_mask, cpu_present_mask);
608 cpu = cpumask_first(&avail);
609 for (i = 0; (i < info->combined) && (cpu < nr_cpu_ids); i++) {
610 if (info->has_cpu_type && info->cpu[i].type != boot_cpu_type)
611 continue;
612 if (pcpu_find_address(cpu_present_mask, info->cpu[i].address))
613 continue;
614 pcpu = pcpu_devices + cpu;
615 pcpu->address = info->cpu[i].address;
616 pcpu->state = (i >= info->configured) ?
617 CPU_STATE_STANDBY : CPU_STATE_CONFIGURED;
618 smp_cpu_set_polarization(cpu, POLARIZATION_UNKNOWN);
619 set_cpu_present(cpu, true);
620 if (sysfs_add && smp_add_present_cpu(cpu) != 0)
621 set_cpu_present(cpu, false);
622 else
623 nr++;
624 cpu = cpumask_next(cpu, &avail);
625 }
626 return nr;
627}
628
629static void __init smp_detect_cpus(void)
630{
631 unsigned int cpu, c_cpus, s_cpus;
632 struct sclp_cpu_info *info;
633
634 info = smp_get_cpu_info();
635 if (!info)
636 panic("smp_detect_cpus failed to allocate memory\n");
637 if (info->has_cpu_type) {
638 for (cpu = 0; cpu < info->combined; cpu++) {
639 if (info->cpu[cpu].address != boot_cpu_address)
640 continue;
641 /* The boot cpu dictates the cpu type. */
642 boot_cpu_type = info->cpu[cpu].type;
643 break;
644 }
645 }
646 c_cpus = s_cpus = 0;
647 for (cpu = 0; cpu < info->combined; cpu++) {
648 if (info->has_cpu_type && info->cpu[cpu].type != boot_cpu_type)
649 continue;
650 if (cpu < info->configured) {
651 smp_get_save_area(c_cpus, info->cpu[cpu].address);
652 c_cpus++;
653 } else
654 s_cpus++;
655 }
656 pr_info("%d configured CPUs, %d standby CPUs\n", c_cpus, s_cpus);
657 get_online_cpus();
658 __smp_rescan_cpus(info, 0);
659 put_online_cpus();
660 kfree(info);
661}
662
663/*
664 * Activate a secondary processor.
665 */
666static void smp_start_secondary(void *cpuvoid)
667{
668 S390_lowcore.last_update_clock = get_tod_clock();
669 S390_lowcore.restart_stack = (unsigned long) restart_stack;
670 S390_lowcore.restart_fn = (unsigned long) do_restart;
671 S390_lowcore.restart_data = 0;
672 S390_lowcore.restart_source = -1UL;
673 restore_access_regs(S390_lowcore.access_regs_save_area);
674 __ctl_load(S390_lowcore.cregs_save_area, 0, 15);
675 __load_psw_mask(PSW_KERNEL_BITS | PSW_MASK_DAT);
676 cpu_init();
677 preempt_disable();
678 init_cpu_timer();
679 init_cpu_vtimer();
680 pfault_init();
681 notify_cpu_starting(smp_processor_id());
682 set_cpu_online(smp_processor_id(), true);
683 inc_irq_stat(CPU_RST);
684 local_irq_enable();
685 cpu_startup_entry(CPUHP_ONLINE);
686}
687
688/* Upping and downing of CPUs */
689int __cpu_up(unsigned int cpu, struct task_struct *tidle)
690{
691 struct pcpu *pcpu;
692 int rc;
693
694 pcpu = pcpu_devices + cpu;
695 if (pcpu->state != CPU_STATE_CONFIGURED)
696 return -EIO;
697 if (pcpu_sigp_retry(pcpu, SIGP_INITIAL_CPU_RESET, 0) !=
698 SIGP_CC_ORDER_CODE_ACCEPTED)
699 return -EIO;
700
701 rc = pcpu_alloc_lowcore(pcpu, cpu);
702 if (rc)
703 return rc;
704 pcpu_prepare_secondary(pcpu, cpu);
705 pcpu_attach_task(pcpu, tidle);
706 pcpu_start_fn(pcpu, smp_start_secondary, NULL);
707 while (!cpu_online(cpu))
708 cpu_relax();
709 return 0;
710}
711
712static unsigned int setup_possible_cpus __initdata;
713
714static int __init _setup_possible_cpus(char *s)
715{
716 get_option(&s, &setup_possible_cpus);
717 return 0;
718}
719early_param("possible_cpus", _setup_possible_cpus);
720
721#ifdef CONFIG_HOTPLUG_CPU
722
723int __cpu_disable(void)
724{
725 unsigned long cregs[16];
726
727 /* Handle possible pending IPIs */
728 smp_handle_ext_call();
729 set_cpu_online(smp_processor_id(), false);
730 /* Disable pseudo page faults on this cpu. */
731 pfault_fini();
732 /* Disable interrupt sources via control register. */
733 __ctl_store(cregs, 0, 15);
734 cregs[0] &= ~0x0000ee70UL; /* disable all external interrupts */
735 cregs[6] &= ~0xff000000UL; /* disable all I/O interrupts */
736 cregs[14] &= ~0x1f000000UL; /* disable most machine checks */
737 __ctl_load(cregs, 0, 15);
738 return 0;
739}
740
741void __cpu_die(unsigned int cpu)
742{
743 struct pcpu *pcpu;
744
745 /* Wait until target cpu is down */
746 pcpu = pcpu_devices + cpu;
747 while (!pcpu_stopped(pcpu))
748 cpu_relax();
749 pcpu_free_lowcore(pcpu);
750 atomic_dec(&init_mm.context.attach_count);
751 cpumask_clear_cpu(cpu, mm_cpumask(&init_mm));
752 if (MACHINE_HAS_TLB_LC)
753 cpumask_clear_cpu(cpu, &init_mm.context.cpu_attach_mask);
754}
755
756void __noreturn cpu_die(void)
757{
758 idle_task_exit();
759 pcpu_sigp_retry(pcpu_devices + smp_processor_id(), SIGP_STOP, 0);
760 for (;;) ;
761}
762
763#endif /* CONFIG_HOTPLUG_CPU */
764
765void __init smp_fill_possible_mask(void)
766{
767 unsigned int possible, sclp, cpu;
768
769 sclp = sclp_get_max_cpu() ?: nr_cpu_ids;
770 possible = setup_possible_cpus ?: nr_cpu_ids;
771 possible = min(possible, sclp);
772 for (cpu = 0; cpu < possible && cpu < nr_cpu_ids; cpu++)
773 set_cpu_possible(cpu, true);
774}
775
776void __init smp_prepare_cpus(unsigned int max_cpus)
777{
778 /* request the 0x1201 emergency signal external interrupt */
779 if (register_external_irq(EXT_IRQ_EMERGENCY_SIG, do_ext_call_interrupt))
780 panic("Couldn't request external interrupt 0x1201");
781 /* request the 0x1202 external call external interrupt */
782 if (register_external_irq(EXT_IRQ_EXTERNAL_CALL, do_ext_call_interrupt))
783 panic("Couldn't request external interrupt 0x1202");
784 smp_detect_cpus();
785}
786
787void __init smp_prepare_boot_cpu(void)
788{
789 struct pcpu *pcpu = pcpu_devices;
790
791 boot_cpu_address = stap();
792 pcpu->state = CPU_STATE_CONFIGURED;
793 pcpu->address = boot_cpu_address;
794 pcpu->lowcore = (struct _lowcore *)(unsigned long) store_prefix();
795 pcpu->async_stack = S390_lowcore.async_stack - ASYNC_SIZE
796 + STACK_FRAME_OVERHEAD + sizeof(struct pt_regs);
797 pcpu->panic_stack = S390_lowcore.panic_stack - PAGE_SIZE
798 + STACK_FRAME_OVERHEAD + sizeof(struct pt_regs);
799 S390_lowcore.percpu_offset = __per_cpu_offset[0];
800 smp_cpu_set_polarization(0, POLARIZATION_UNKNOWN);
801 set_cpu_present(0, true);
802 set_cpu_online(0, true);
803}
804
805void __init smp_cpus_done(unsigned int max_cpus)
806{
807}
808
809void __init smp_setup_processor_id(void)
810{
811 S390_lowcore.cpu_nr = 0;
812}
813
814/*
815 * the frequency of the profiling timer can be changed
816 * by writing a multiplier value into /proc/profile.
817 *
818 * usually you want to run this on all CPUs ;)
819 */
820int setup_profiling_timer(unsigned int multiplier)
821{
822 return 0;
823}
824
825#ifdef CONFIG_HOTPLUG_CPU
826static ssize_t cpu_configure_show(struct device *dev,
827 struct device_attribute *attr, char *buf)
828{
829 ssize_t count;
830
831 mutex_lock(&smp_cpu_state_mutex);
832 count = sprintf(buf, "%d\n", pcpu_devices[dev->id].state);
833 mutex_unlock(&smp_cpu_state_mutex);
834 return count;
835}
836
837static ssize_t cpu_configure_store(struct device *dev,
838 struct device_attribute *attr,
839 const char *buf, size_t count)
840{
841 struct pcpu *pcpu;
842 int cpu, val, rc;
843 char delim;
844
845 if (sscanf(buf, "%d %c", &val, &delim) != 1)
846 return -EINVAL;
847 if (val != 0 && val != 1)
848 return -EINVAL;
849 get_online_cpus();
850 mutex_lock(&smp_cpu_state_mutex);
851 rc = -EBUSY;
852 /* disallow configuration changes of online cpus and cpu 0 */
853 cpu = dev->id;
854 if (cpu_online(cpu) || cpu == 0)
855 goto out;
856 pcpu = pcpu_devices + cpu;
857 rc = 0;
858 switch (val) {
859 case 0:
860 if (pcpu->state != CPU_STATE_CONFIGURED)
861 break;
862 rc = sclp_cpu_deconfigure(pcpu->address);
863 if (rc)
864 break;
865 pcpu->state = CPU_STATE_STANDBY;
866 smp_cpu_set_polarization(cpu, POLARIZATION_UNKNOWN);
867 topology_expect_change();
868 break;
869 case 1:
870 if (pcpu->state != CPU_STATE_STANDBY)
871 break;
872 rc = sclp_cpu_configure(pcpu->address);
873 if (rc)
874 break;
875 pcpu->state = CPU_STATE_CONFIGURED;
876 smp_cpu_set_polarization(cpu, POLARIZATION_UNKNOWN);
877 topology_expect_change();
878 break;
879 default:
880 break;
881 }
882out:
883 mutex_unlock(&smp_cpu_state_mutex);
884 put_online_cpus();
885 return rc ? rc : count;
886}
887static DEVICE_ATTR(configure, 0644, cpu_configure_show, cpu_configure_store);
888#endif /* CONFIG_HOTPLUG_CPU */
889
890static ssize_t show_cpu_address(struct device *dev,
891 struct device_attribute *attr, char *buf)
892{
893 return sprintf(buf, "%d\n", pcpu_devices[dev->id].address);
894}
895static DEVICE_ATTR(address, 0444, show_cpu_address, NULL);
896
897static struct attribute *cpu_common_attrs[] = {
898#ifdef CONFIG_HOTPLUG_CPU
899 &dev_attr_configure.attr,
900#endif
901 &dev_attr_address.attr,
902 NULL,
903};
904
905static struct attribute_group cpu_common_attr_group = {
906 .attrs = cpu_common_attrs,
907};
908
909static ssize_t show_idle_count(struct device *dev,
910 struct device_attribute *attr, char *buf)
911{
912 struct s390_idle_data *idle = &per_cpu(s390_idle, dev->id);
913 unsigned long long idle_count;
914 unsigned int sequence;
915
916 do {
917 sequence = ACCESS_ONCE(idle->sequence);
918 idle_count = ACCESS_ONCE(idle->idle_count);
919 if (ACCESS_ONCE(idle->clock_idle_enter))
920 idle_count++;
921 } while ((sequence & 1) || (ACCESS_ONCE(idle->sequence) != sequence));
922 return sprintf(buf, "%llu\n", idle_count);
923}
924static DEVICE_ATTR(idle_count, 0444, show_idle_count, NULL);
925
926static ssize_t show_idle_time(struct device *dev,
927 struct device_attribute *attr, char *buf)
928{
929 struct s390_idle_data *idle = &per_cpu(s390_idle, dev->id);
930 unsigned long long now, idle_time, idle_enter, idle_exit;
931 unsigned int sequence;
932
933 do {
934 now = get_tod_clock();
935 sequence = ACCESS_ONCE(idle->sequence);
936 idle_time = ACCESS_ONCE(idle->idle_time);
937 idle_enter = ACCESS_ONCE(idle->clock_idle_enter);
938 idle_exit = ACCESS_ONCE(idle->clock_idle_exit);
939 } while ((sequence & 1) || (ACCESS_ONCE(idle->sequence) != sequence));
940 idle_time += idle_enter ? ((idle_exit ? : now) - idle_enter) : 0;
941 return sprintf(buf, "%llu\n", idle_time >> 12);
942}
943static DEVICE_ATTR(idle_time_us, 0444, show_idle_time, NULL);
944
945static struct attribute *cpu_online_attrs[] = {
946 &dev_attr_idle_count.attr,
947 &dev_attr_idle_time_us.attr,
948 NULL,
949};
950
951static struct attribute_group cpu_online_attr_group = {
952 .attrs = cpu_online_attrs,
953};
954
955static int smp_cpu_notify(struct notifier_block *self, unsigned long action,
956 void *hcpu)
957{
958 unsigned int cpu = (unsigned int)(long)hcpu;
959 struct cpu *c = pcpu_devices[cpu].cpu;
960 struct device *s = &c->dev;
961 int err = 0;
962
963 switch (action & ~CPU_TASKS_FROZEN) {
964 case CPU_ONLINE:
965 err = sysfs_create_group(&s->kobj, &cpu_online_attr_group);
966 break;
967 case CPU_DEAD:
968 sysfs_remove_group(&s->kobj, &cpu_online_attr_group);
969 break;
970 }
971 return notifier_from_errno(err);
972}
973
974static int smp_add_present_cpu(int cpu)
975{
976 struct device *s;
977 struct cpu *c;
978 int rc;
979
980 c = kzalloc(sizeof(*c), GFP_KERNEL);
981 if (!c)
982 return -ENOMEM;
983 pcpu_devices[cpu].cpu = c;
984 s = &c->dev;
985 c->hotpluggable = 1;
986 rc = register_cpu(c, cpu);
987 if (rc)
988 goto out;
989 rc = sysfs_create_group(&s->kobj, &cpu_common_attr_group);
990 if (rc)
991 goto out_cpu;
992 if (cpu_online(cpu)) {
993 rc = sysfs_create_group(&s->kobj, &cpu_online_attr_group);
994 if (rc)
995 goto out_online;
996 }
997 rc = topology_cpu_init(c);
998 if (rc)
999 goto out_topology;
1000 return 0;
1001
1002out_topology:
1003 if (cpu_online(cpu))
1004 sysfs_remove_group(&s->kobj, &cpu_online_attr_group);
1005out_online:
1006 sysfs_remove_group(&s->kobj, &cpu_common_attr_group);
1007out_cpu:
1008#ifdef CONFIG_HOTPLUG_CPU
1009 unregister_cpu(c);
1010#endif
1011out:
1012 return rc;
1013}
1014
1015#ifdef CONFIG_HOTPLUG_CPU
1016
1017int __ref smp_rescan_cpus(void)
1018{
1019 struct sclp_cpu_info *info;
1020 int nr;
1021
1022 info = smp_get_cpu_info();
1023 if (!info)
1024 return -ENOMEM;
1025 get_online_cpus();
1026 mutex_lock(&smp_cpu_state_mutex);
1027 nr = __smp_rescan_cpus(info, 1);
1028 mutex_unlock(&smp_cpu_state_mutex);
1029 put_online_cpus();
1030 kfree(info);
1031 if (nr)
1032 topology_schedule_update();
1033 return 0;
1034}
1035
1036static ssize_t __ref rescan_store(struct device *dev,
1037 struct device_attribute *attr,
1038 const char *buf,
1039 size_t count)
1040{
1041 int rc;
1042
1043 rc = smp_rescan_cpus();
1044 return rc ? rc : count;
1045}
1046static DEVICE_ATTR(rescan, 0200, NULL, rescan_store);
1047#endif /* CONFIG_HOTPLUG_CPU */
1048
1049static int __init s390_smp_init(void)
1050{
1051 int cpu, rc = 0;
1052
1053#ifdef CONFIG_HOTPLUG_CPU
1054 rc = device_create_file(cpu_subsys.dev_root, &dev_attr_rescan);
1055 if (rc)
1056 return rc;
1057#endif
1058 cpu_notifier_register_begin();
1059 for_each_present_cpu(cpu) {
1060 rc = smp_add_present_cpu(cpu);
1061 if (rc)
1062 goto out;
1063 }
1064
1065 __hotcpu_notifier(smp_cpu_notify, 0);
1066
1067out:
1068 cpu_notifier_register_done();
1069 return rc;
1070}
1071subsys_initcall(s390_smp_init);
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * SMP related functions
4 *
5 * Copyright IBM Corp. 1999, 2012
6 * Author(s): Denis Joseph Barrow,
7 * Martin Schwidefsky <schwidefsky@de.ibm.com>,
8 * Heiko Carstens <heiko.carstens@de.ibm.com>,
9 *
10 * based on other smp stuff by
11 * (c) 1995 Alan Cox, CymruNET Ltd <alan@cymru.net>
12 * (c) 1998 Ingo Molnar
13 *
14 * The code outside of smp.c uses logical cpu numbers, only smp.c does
15 * the translation of logical to physical cpu ids. All new code that
16 * operates on physical cpu numbers needs to go into smp.c.
17 */
18
19#define KMSG_COMPONENT "cpu"
20#define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
21
22#include <linux/workqueue.h>
23#include <linux/memblock.h>
24#include <linux/export.h>
25#include <linux/init.h>
26#include <linux/mm.h>
27#include <linux/err.h>
28#include <linux/spinlock.h>
29#include <linux/kernel_stat.h>
30#include <linux/delay.h>
31#include <linux/interrupt.h>
32#include <linux/irqflags.h>
33#include <linux/irq_work.h>
34#include <linux/cpu.h>
35#include <linux/slab.h>
36#include <linux/sched/hotplug.h>
37#include <linux/sched/task_stack.h>
38#include <linux/crash_dump.h>
39#include <linux/kprobes.h>
40#include <asm/asm-offsets.h>
41#include <asm/diag.h>
42#include <asm/switch_to.h>
43#include <asm/facility.h>
44#include <asm/ipl.h>
45#include <asm/setup.h>
46#include <asm/irq.h>
47#include <asm/tlbflush.h>
48#include <asm/vtimer.h>
49#include <asm/lowcore.h>
50#include <asm/sclp.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 <asm/stacktrace.h>
57#include <asm/topology.h>
58#include <asm/vdso.h>
59#include "entry.h"
60
61enum {
62 ec_schedule = 0,
63 ec_call_function_single,
64 ec_stop_cpu,
65 ec_mcck_pending,
66 ec_irq_work,
67};
68
69enum {
70 CPU_STATE_STANDBY,
71 CPU_STATE_CONFIGURED,
72};
73
74static DEFINE_PER_CPU(struct cpu *, cpu_device);
75
76struct pcpu {
77 unsigned long ec_mask; /* bit mask for ec_xxx functions */
78 unsigned long ec_clk; /* sigp timestamp for ec_xxx */
79 signed char state; /* physical cpu state */
80 signed char polarization; /* physical polarization */
81 u16 address; /* physical cpu address */
82};
83
84static u8 boot_core_type;
85static struct pcpu pcpu_devices[NR_CPUS];
86
87unsigned int smp_cpu_mt_shift;
88EXPORT_SYMBOL(smp_cpu_mt_shift);
89
90unsigned int smp_cpu_mtid;
91EXPORT_SYMBOL(smp_cpu_mtid);
92
93#ifdef CONFIG_CRASH_DUMP
94__vector128 __initdata boot_cpu_vector_save_area[__NUM_VXRS];
95#endif
96
97static unsigned int smp_max_threads __initdata = -1U;
98cpumask_t cpu_setup_mask;
99
100static int __init early_nosmt(char *s)
101{
102 smp_max_threads = 1;
103 return 0;
104}
105early_param("nosmt", early_nosmt);
106
107static int __init early_smt(char *s)
108{
109 get_option(&s, &smp_max_threads);
110 return 0;
111}
112early_param("smt", early_smt);
113
114/*
115 * The smp_cpu_state_mutex must be held when changing the state or polarization
116 * member of a pcpu data structure within the pcpu_devices arreay.
117 */
118DEFINE_MUTEX(smp_cpu_state_mutex);
119
120/*
121 * Signal processor helper functions.
122 */
123static inline int __pcpu_sigp_relax(u16 addr, u8 order, unsigned long parm)
124{
125 int cc;
126
127 while (1) {
128 cc = __pcpu_sigp(addr, order, parm, NULL);
129 if (cc != SIGP_CC_BUSY)
130 return cc;
131 cpu_relax();
132 }
133}
134
135static int pcpu_sigp_retry(struct pcpu *pcpu, u8 order, u32 parm)
136{
137 int cc, retry;
138
139 for (retry = 0; ; retry++) {
140 cc = __pcpu_sigp(pcpu->address, order, parm, NULL);
141 if (cc != SIGP_CC_BUSY)
142 break;
143 if (retry >= 3)
144 udelay(10);
145 }
146 return cc;
147}
148
149static inline int pcpu_stopped(struct pcpu *pcpu)
150{
151 u32 status;
152
153 if (__pcpu_sigp(pcpu->address, SIGP_SENSE,
154 0, &status) != SIGP_CC_STATUS_STORED)
155 return 0;
156 return !!(status & (SIGP_STATUS_CHECK_STOP|SIGP_STATUS_STOPPED));
157}
158
159static inline int pcpu_running(struct pcpu *pcpu)
160{
161 if (__pcpu_sigp(pcpu->address, SIGP_SENSE_RUNNING,
162 0, NULL) != SIGP_CC_STATUS_STORED)
163 return 1;
164 /* Status stored condition code is equivalent to cpu not running. */
165 return 0;
166}
167
168/*
169 * Find struct pcpu by cpu address.
170 */
171static struct pcpu *pcpu_find_address(const struct cpumask *mask, u16 address)
172{
173 int cpu;
174
175 for_each_cpu(cpu, mask)
176 if (pcpu_devices[cpu].address == address)
177 return pcpu_devices + cpu;
178 return NULL;
179}
180
181static void pcpu_ec_call(struct pcpu *pcpu, int ec_bit)
182{
183 int order;
184
185 if (test_and_set_bit(ec_bit, &pcpu->ec_mask))
186 return;
187 order = pcpu_running(pcpu) ? SIGP_EXTERNAL_CALL : SIGP_EMERGENCY_SIGNAL;
188 pcpu->ec_clk = get_tod_clock_fast();
189 pcpu_sigp_retry(pcpu, order, 0);
190}
191
192static int pcpu_alloc_lowcore(struct pcpu *pcpu, int cpu)
193{
194 unsigned long async_stack, nodat_stack, mcck_stack;
195 struct lowcore *lc;
196
197 lc = (struct lowcore *) __get_free_pages(GFP_KERNEL | GFP_DMA, LC_ORDER);
198 nodat_stack = __get_free_pages(GFP_KERNEL, THREAD_SIZE_ORDER);
199 async_stack = stack_alloc();
200 mcck_stack = stack_alloc();
201 if (!lc || !nodat_stack || !async_stack || !mcck_stack)
202 goto out;
203 memcpy(lc, &S390_lowcore, 512);
204 memset((char *) lc + 512, 0, sizeof(*lc) - 512);
205 lc->async_stack = async_stack + STACK_INIT_OFFSET;
206 lc->nodat_stack = nodat_stack + STACK_INIT_OFFSET;
207 lc->mcck_stack = mcck_stack + STACK_INIT_OFFSET;
208 lc->cpu_nr = cpu;
209 lc->spinlock_lockval = arch_spin_lockval(cpu);
210 lc->spinlock_index = 0;
211 lc->br_r1_trampoline = 0x07f1; /* br %r1 */
212 lc->return_lpswe = gen_lpswe(__LC_RETURN_PSW);
213 lc->return_mcck_lpswe = gen_lpswe(__LC_RETURN_MCCK_PSW);
214 lc->preempt_count = PREEMPT_DISABLED;
215 if (nmi_alloc_per_cpu(lc))
216 goto out;
217 lowcore_ptr[cpu] = lc;
218 pcpu_sigp_retry(pcpu, SIGP_SET_PREFIX, (u32)(unsigned long) lc);
219 return 0;
220
221out:
222 stack_free(mcck_stack);
223 stack_free(async_stack);
224 free_pages(nodat_stack, THREAD_SIZE_ORDER);
225 free_pages((unsigned long) lc, LC_ORDER);
226 return -ENOMEM;
227}
228
229static void pcpu_free_lowcore(struct pcpu *pcpu)
230{
231 unsigned long async_stack, nodat_stack, mcck_stack;
232 struct lowcore *lc;
233 int cpu;
234
235 cpu = pcpu - pcpu_devices;
236 lc = lowcore_ptr[cpu];
237 nodat_stack = lc->nodat_stack - STACK_INIT_OFFSET;
238 async_stack = lc->async_stack - STACK_INIT_OFFSET;
239 mcck_stack = lc->mcck_stack - STACK_INIT_OFFSET;
240 pcpu_sigp_retry(pcpu, SIGP_SET_PREFIX, 0);
241 lowcore_ptr[cpu] = NULL;
242 nmi_free_per_cpu(lc);
243 stack_free(async_stack);
244 stack_free(mcck_stack);
245 free_pages(nodat_stack, THREAD_SIZE_ORDER);
246 free_pages((unsigned long) lc, LC_ORDER);
247}
248
249static void pcpu_prepare_secondary(struct pcpu *pcpu, int cpu)
250{
251 struct lowcore *lc = lowcore_ptr[cpu];
252
253 cpumask_set_cpu(cpu, &init_mm.context.cpu_attach_mask);
254 cpumask_set_cpu(cpu, mm_cpumask(&init_mm));
255 lc->cpu_nr = cpu;
256 lc->restart_flags = RESTART_FLAG_CTLREGS;
257 lc->spinlock_lockval = arch_spin_lockval(cpu);
258 lc->spinlock_index = 0;
259 lc->percpu_offset = __per_cpu_offset[cpu];
260 lc->kernel_asce = S390_lowcore.kernel_asce;
261 lc->user_asce = s390_invalid_asce;
262 lc->machine_flags = S390_lowcore.machine_flags;
263 lc->user_timer = lc->system_timer =
264 lc->steal_timer = lc->avg_steal_timer = 0;
265 __ctl_store(lc->cregs_save_area, 0, 15);
266 lc->cregs_save_area[1] = lc->kernel_asce;
267 lc->cregs_save_area[7] = lc->user_asce;
268 save_access_regs((unsigned int *) lc->access_regs_save_area);
269 arch_spin_lock_setup(cpu);
270}
271
272static void pcpu_attach_task(struct pcpu *pcpu, struct task_struct *tsk)
273{
274 struct lowcore *lc;
275 int cpu;
276
277 cpu = pcpu - pcpu_devices;
278 lc = lowcore_ptr[cpu];
279 lc->kernel_stack = (unsigned long) task_stack_page(tsk)
280 + THREAD_SIZE - STACK_FRAME_OVERHEAD - sizeof(struct pt_regs);
281 lc->current_task = (unsigned long) tsk;
282 lc->lpp = LPP_MAGIC;
283 lc->current_pid = tsk->pid;
284 lc->user_timer = tsk->thread.user_timer;
285 lc->guest_timer = tsk->thread.guest_timer;
286 lc->system_timer = tsk->thread.system_timer;
287 lc->hardirq_timer = tsk->thread.hardirq_timer;
288 lc->softirq_timer = tsk->thread.softirq_timer;
289 lc->steal_timer = 0;
290}
291
292static void pcpu_start_fn(struct pcpu *pcpu, void (*func)(void *), void *data)
293{
294 struct lowcore *lc;
295 int cpu;
296
297 cpu = pcpu - pcpu_devices;
298 lc = lowcore_ptr[cpu];
299 lc->restart_stack = lc->nodat_stack;
300 lc->restart_fn = (unsigned long) func;
301 lc->restart_data = (unsigned long) data;
302 lc->restart_source = -1U;
303 pcpu_sigp_retry(pcpu, SIGP_RESTART, 0);
304}
305
306typedef void (pcpu_delegate_fn)(void *);
307
308/*
309 * Call function via PSW restart on pcpu and stop the current cpu.
310 */
311static void __pcpu_delegate(pcpu_delegate_fn *func, void *data)
312{
313 func(data); /* should not return */
314}
315
316static void pcpu_delegate(struct pcpu *pcpu,
317 pcpu_delegate_fn *func,
318 void *data, unsigned long stack)
319{
320 struct lowcore *lc = lowcore_ptr[pcpu - pcpu_devices];
321 unsigned int source_cpu = stap();
322
323 __load_psw_mask(PSW_KERNEL_BITS | PSW_MASK_DAT);
324 if (pcpu->address == source_cpu) {
325 call_on_stack(2, stack, void, __pcpu_delegate,
326 pcpu_delegate_fn *, func, void *, data);
327 }
328 /* Stop target cpu (if func returns this stops the current cpu). */
329 pcpu_sigp_retry(pcpu, SIGP_STOP, 0);
330 /* Restart func on the target cpu and stop the current cpu. */
331 mem_assign_absolute(lc->restart_stack, stack);
332 mem_assign_absolute(lc->restart_fn, (unsigned long) func);
333 mem_assign_absolute(lc->restart_data, (unsigned long) data);
334 mem_assign_absolute(lc->restart_source, source_cpu);
335 __bpon();
336 asm volatile(
337 "0: sigp 0,%0,%2 # sigp restart to target cpu\n"
338 " brc 2,0b # busy, try again\n"
339 "1: sigp 0,%1,%3 # sigp stop to current cpu\n"
340 " brc 2,1b # busy, try again\n"
341 : : "d" (pcpu->address), "d" (source_cpu),
342 "K" (SIGP_RESTART), "K" (SIGP_STOP)
343 : "0", "1", "cc");
344 for (;;) ;
345}
346
347/*
348 * Enable additional logical cpus for multi-threading.
349 */
350static int pcpu_set_smt(unsigned int mtid)
351{
352 int cc;
353
354 if (smp_cpu_mtid == mtid)
355 return 0;
356 cc = __pcpu_sigp(0, SIGP_SET_MULTI_THREADING, mtid, NULL);
357 if (cc == 0) {
358 smp_cpu_mtid = mtid;
359 smp_cpu_mt_shift = 0;
360 while (smp_cpu_mtid >= (1U << smp_cpu_mt_shift))
361 smp_cpu_mt_shift++;
362 pcpu_devices[0].address = stap();
363 }
364 return cc;
365}
366
367/*
368 * Call function on an online CPU.
369 */
370void smp_call_online_cpu(void (*func)(void *), void *data)
371{
372 struct pcpu *pcpu;
373
374 /* Use the current cpu if it is online. */
375 pcpu = pcpu_find_address(cpu_online_mask, stap());
376 if (!pcpu)
377 /* Use the first online cpu. */
378 pcpu = pcpu_devices + cpumask_first(cpu_online_mask);
379 pcpu_delegate(pcpu, func, data, (unsigned long) restart_stack);
380}
381
382/*
383 * Call function on the ipl CPU.
384 */
385void smp_call_ipl_cpu(void (*func)(void *), void *data)
386{
387 struct lowcore *lc = lowcore_ptr[0];
388
389 if (pcpu_devices[0].address == stap())
390 lc = &S390_lowcore;
391
392 pcpu_delegate(&pcpu_devices[0], func, data,
393 lc->nodat_stack);
394}
395
396int smp_find_processor_id(u16 address)
397{
398 int cpu;
399
400 for_each_present_cpu(cpu)
401 if (pcpu_devices[cpu].address == address)
402 return cpu;
403 return -1;
404}
405
406void schedule_mcck_handler(void)
407{
408 pcpu_ec_call(pcpu_devices + smp_processor_id(), ec_mcck_pending);
409}
410
411bool notrace arch_vcpu_is_preempted(int cpu)
412{
413 if (test_cpu_flag_of(CIF_ENABLED_WAIT, cpu))
414 return false;
415 if (pcpu_running(pcpu_devices + cpu))
416 return false;
417 return true;
418}
419EXPORT_SYMBOL(arch_vcpu_is_preempted);
420
421void notrace smp_yield_cpu(int cpu)
422{
423 if (!MACHINE_HAS_DIAG9C)
424 return;
425 diag_stat_inc_norecursion(DIAG_STAT_X09C);
426 asm volatile("diag %0,0,0x9c"
427 : : "d" (pcpu_devices[cpu].address));
428}
429EXPORT_SYMBOL_GPL(smp_yield_cpu);
430
431/*
432 * Send cpus emergency shutdown signal. This gives the cpus the
433 * opportunity to complete outstanding interrupts.
434 */
435void notrace smp_emergency_stop(void)
436{
437 static arch_spinlock_t lock = __ARCH_SPIN_LOCK_UNLOCKED;
438 static cpumask_t cpumask;
439 u64 end;
440 int cpu;
441
442 arch_spin_lock(&lock);
443 cpumask_copy(&cpumask, cpu_online_mask);
444 cpumask_clear_cpu(smp_processor_id(), &cpumask);
445
446 end = get_tod_clock() + (1000000UL << 12);
447 for_each_cpu(cpu, &cpumask) {
448 struct pcpu *pcpu = pcpu_devices + cpu;
449 set_bit(ec_stop_cpu, &pcpu->ec_mask);
450 while (__pcpu_sigp(pcpu->address, SIGP_EMERGENCY_SIGNAL,
451 0, NULL) == SIGP_CC_BUSY &&
452 get_tod_clock() < end)
453 cpu_relax();
454 }
455 while (get_tod_clock() < end) {
456 for_each_cpu(cpu, &cpumask)
457 if (pcpu_stopped(pcpu_devices + cpu))
458 cpumask_clear_cpu(cpu, &cpumask);
459 if (cpumask_empty(&cpumask))
460 break;
461 cpu_relax();
462 }
463 arch_spin_unlock(&lock);
464}
465NOKPROBE_SYMBOL(smp_emergency_stop);
466
467/*
468 * Stop all cpus but the current one.
469 */
470void smp_send_stop(void)
471{
472 int cpu;
473
474 /* Disable all interrupts/machine checks */
475 __load_psw_mask(PSW_KERNEL_BITS | PSW_MASK_DAT);
476 trace_hardirqs_off();
477
478 debug_set_critical();
479
480 if (oops_in_progress)
481 smp_emergency_stop();
482
483 /* stop all processors */
484 for_each_online_cpu(cpu) {
485 if (cpu == smp_processor_id())
486 continue;
487 pcpu_sigp_retry(pcpu_devices + cpu, SIGP_STOP, 0);
488 while (!pcpu_stopped(pcpu_devices + cpu))
489 cpu_relax();
490 }
491}
492
493/*
494 * This is the main routine where commands issued by other
495 * cpus are handled.
496 */
497static void smp_handle_ext_call(void)
498{
499 unsigned long bits;
500
501 /* handle bit signal external calls */
502 bits = xchg(&pcpu_devices[smp_processor_id()].ec_mask, 0);
503 if (test_bit(ec_stop_cpu, &bits))
504 smp_stop_cpu();
505 if (test_bit(ec_schedule, &bits))
506 scheduler_ipi();
507 if (test_bit(ec_call_function_single, &bits))
508 generic_smp_call_function_single_interrupt();
509 if (test_bit(ec_mcck_pending, &bits))
510 __s390_handle_mcck();
511 if (test_bit(ec_irq_work, &bits))
512 irq_work_run();
513}
514
515static void do_ext_call_interrupt(struct ext_code ext_code,
516 unsigned int param32, unsigned long param64)
517{
518 inc_irq_stat(ext_code.code == 0x1202 ? IRQEXT_EXC : IRQEXT_EMS);
519 smp_handle_ext_call();
520}
521
522void arch_send_call_function_ipi_mask(const struct cpumask *mask)
523{
524 int cpu;
525
526 for_each_cpu(cpu, mask)
527 pcpu_ec_call(pcpu_devices + cpu, ec_call_function_single);
528}
529
530void arch_send_call_function_single_ipi(int cpu)
531{
532 pcpu_ec_call(pcpu_devices + cpu, ec_call_function_single);
533}
534
535/*
536 * this function sends a 'reschedule' IPI to another CPU.
537 * it goes straight through and wastes no time serializing
538 * anything. Worst case is that we lose a reschedule ...
539 */
540void smp_send_reschedule(int cpu)
541{
542 pcpu_ec_call(pcpu_devices + cpu, ec_schedule);
543}
544
545#ifdef CONFIG_IRQ_WORK
546void arch_irq_work_raise(void)
547{
548 pcpu_ec_call(pcpu_devices + smp_processor_id(), ec_irq_work);
549}
550#endif
551
552/*
553 * parameter area for the set/clear control bit callbacks
554 */
555struct ec_creg_mask_parms {
556 unsigned long orval;
557 unsigned long andval;
558 int cr;
559};
560
561/*
562 * callback for setting/clearing control bits
563 */
564static void smp_ctl_bit_callback(void *info)
565{
566 struct ec_creg_mask_parms *pp = info;
567 unsigned long cregs[16];
568
569 __ctl_store(cregs, 0, 15);
570 cregs[pp->cr] = (cregs[pp->cr] & pp->andval) | pp->orval;
571 __ctl_load(cregs, 0, 15);
572}
573
574static DEFINE_SPINLOCK(ctl_lock);
575static unsigned long ctlreg;
576
577/*
578 * Set a bit in a control register of all cpus
579 */
580void smp_ctl_set_bit(int cr, int bit)
581{
582 struct ec_creg_mask_parms parms = { 1UL << bit, -1UL, cr };
583
584 spin_lock(&ctl_lock);
585 memcpy_absolute(&ctlreg, &S390_lowcore.cregs_save_area[cr], sizeof(ctlreg));
586 __set_bit(bit, &ctlreg);
587 memcpy_absolute(&S390_lowcore.cregs_save_area[cr], &ctlreg, sizeof(ctlreg));
588 spin_unlock(&ctl_lock);
589 on_each_cpu(smp_ctl_bit_callback, &parms, 1);
590}
591EXPORT_SYMBOL(smp_ctl_set_bit);
592
593/*
594 * Clear a bit in a control register of all cpus
595 */
596void smp_ctl_clear_bit(int cr, int bit)
597{
598 struct ec_creg_mask_parms parms = { 0, ~(1UL << bit), cr };
599
600 spin_lock(&ctl_lock);
601 memcpy_absolute(&ctlreg, &S390_lowcore.cregs_save_area[cr], sizeof(ctlreg));
602 __clear_bit(bit, &ctlreg);
603 memcpy_absolute(&S390_lowcore.cregs_save_area[cr], &ctlreg, sizeof(ctlreg));
604 spin_unlock(&ctl_lock);
605 on_each_cpu(smp_ctl_bit_callback, &parms, 1);
606}
607EXPORT_SYMBOL(smp_ctl_clear_bit);
608
609#ifdef CONFIG_CRASH_DUMP
610
611int smp_store_status(int cpu)
612{
613 struct lowcore *lc;
614 struct pcpu *pcpu;
615 unsigned long pa;
616
617 pcpu = pcpu_devices + cpu;
618 lc = lowcore_ptr[cpu];
619 pa = __pa(&lc->floating_pt_save_area);
620 if (__pcpu_sigp_relax(pcpu->address, SIGP_STORE_STATUS_AT_ADDRESS,
621 pa) != SIGP_CC_ORDER_CODE_ACCEPTED)
622 return -EIO;
623 if (!MACHINE_HAS_VX && !MACHINE_HAS_GS)
624 return 0;
625 pa = __pa(lc->mcesad & MCESA_ORIGIN_MASK);
626 if (MACHINE_HAS_GS)
627 pa |= lc->mcesad & MCESA_LC_MASK;
628 if (__pcpu_sigp_relax(pcpu->address, SIGP_STORE_ADDITIONAL_STATUS,
629 pa) != SIGP_CC_ORDER_CODE_ACCEPTED)
630 return -EIO;
631 return 0;
632}
633
634/*
635 * Collect CPU state of the previous, crashed system.
636 * There are four cases:
637 * 1) standard zfcp/nvme dump
638 * condition: OLDMEM_BASE == NULL && is_ipl_type_dump() == true
639 * The state for all CPUs except the boot CPU needs to be collected
640 * with sigp stop-and-store-status. The boot CPU state is located in
641 * the absolute lowcore of the memory stored in the HSA. The zcore code
642 * will copy the boot CPU state from the HSA.
643 * 2) stand-alone kdump for SCSI/NVMe (zfcp/nvme dump with swapped memory)
644 * condition: OLDMEM_BASE != NULL && is_ipl_type_dump() == true
645 * The state for all CPUs except the boot CPU needs to be collected
646 * with sigp stop-and-store-status. The firmware or the boot-loader
647 * stored the registers of the boot CPU in the absolute lowcore in the
648 * memory of the old system.
649 * 3) kdump and the old kernel did not store the CPU state,
650 * or stand-alone kdump for DASD
651 * condition: OLDMEM_BASE != NULL && !is_kdump_kernel()
652 * The state for all CPUs except the boot CPU needs to be collected
653 * with sigp stop-and-store-status. The kexec code or the boot-loader
654 * stored the registers of the boot CPU in the memory of the old system.
655 * 4) kdump and the old kernel stored the CPU state
656 * condition: OLDMEM_BASE != NULL && is_kdump_kernel()
657 * This case does not exist for s390 anymore, setup_arch explicitly
658 * deactivates the elfcorehdr= kernel parameter
659 */
660static __init void smp_save_cpu_vxrs(struct save_area *sa, u16 addr,
661 bool is_boot_cpu, unsigned long page)
662{
663 __vector128 *vxrs = (__vector128 *) page;
664
665 if (is_boot_cpu)
666 vxrs = boot_cpu_vector_save_area;
667 else
668 __pcpu_sigp_relax(addr, SIGP_STORE_ADDITIONAL_STATUS, page);
669 save_area_add_vxrs(sa, vxrs);
670}
671
672static __init void smp_save_cpu_regs(struct save_area *sa, u16 addr,
673 bool is_boot_cpu, unsigned long page)
674{
675 void *regs = (void *) page;
676
677 if (is_boot_cpu)
678 copy_oldmem_kernel(regs, (void *) __LC_FPREGS_SAVE_AREA, 512);
679 else
680 __pcpu_sigp_relax(addr, SIGP_STORE_STATUS_AT_ADDRESS, page);
681 save_area_add_regs(sa, regs);
682}
683
684void __init smp_save_dump_cpus(void)
685{
686 int addr, boot_cpu_addr, max_cpu_addr;
687 struct save_area *sa;
688 unsigned long page;
689 bool is_boot_cpu;
690
691 if (!(OLDMEM_BASE || is_ipl_type_dump()))
692 /* No previous system present, normal boot. */
693 return;
694 /* Allocate a page as dumping area for the store status sigps */
695 page = memblock_phys_alloc_range(PAGE_SIZE, PAGE_SIZE, 0, 1UL << 31);
696 if (!page)
697 panic("ERROR: Failed to allocate %lx bytes below %lx\n",
698 PAGE_SIZE, 1UL << 31);
699
700 /* Set multi-threading state to the previous system. */
701 pcpu_set_smt(sclp.mtid_prev);
702 boot_cpu_addr = stap();
703 max_cpu_addr = SCLP_MAX_CORES << sclp.mtid_prev;
704 for (addr = 0; addr <= max_cpu_addr; addr++) {
705 if (__pcpu_sigp_relax(addr, SIGP_SENSE, 0) ==
706 SIGP_CC_NOT_OPERATIONAL)
707 continue;
708 is_boot_cpu = (addr == boot_cpu_addr);
709 /* Allocate save area */
710 sa = save_area_alloc(is_boot_cpu);
711 if (!sa)
712 panic("could not allocate memory for save area\n");
713 if (MACHINE_HAS_VX)
714 /* Get the vector registers */
715 smp_save_cpu_vxrs(sa, addr, is_boot_cpu, page);
716 /*
717 * For a zfcp/nvme dump OLDMEM_BASE == NULL and the registers
718 * of the boot CPU are stored in the HSA. To retrieve
719 * these registers an SCLP request is required which is
720 * done by drivers/s390/char/zcore.c:init_cpu_info()
721 */
722 if (!is_boot_cpu || OLDMEM_BASE)
723 /* Get the CPU registers */
724 smp_save_cpu_regs(sa, addr, is_boot_cpu, page);
725 }
726 memblock_free(page, PAGE_SIZE);
727 diag_dma_ops.diag308_reset();
728 pcpu_set_smt(0);
729}
730#endif /* CONFIG_CRASH_DUMP */
731
732void smp_cpu_set_polarization(int cpu, int val)
733{
734 pcpu_devices[cpu].polarization = val;
735}
736
737int smp_cpu_get_polarization(int cpu)
738{
739 return pcpu_devices[cpu].polarization;
740}
741
742int smp_cpu_get_cpu_address(int cpu)
743{
744 return pcpu_devices[cpu].address;
745}
746
747static void __ref smp_get_core_info(struct sclp_core_info *info, int early)
748{
749 static int use_sigp_detection;
750 int address;
751
752 if (use_sigp_detection || sclp_get_core_info(info, early)) {
753 use_sigp_detection = 1;
754 for (address = 0;
755 address < (SCLP_MAX_CORES << smp_cpu_mt_shift);
756 address += (1U << smp_cpu_mt_shift)) {
757 if (__pcpu_sigp_relax(address, SIGP_SENSE, 0) ==
758 SIGP_CC_NOT_OPERATIONAL)
759 continue;
760 info->core[info->configured].core_id =
761 address >> smp_cpu_mt_shift;
762 info->configured++;
763 }
764 info->combined = info->configured;
765 }
766}
767
768static int smp_add_present_cpu(int cpu);
769
770static int smp_add_core(struct sclp_core_entry *core, cpumask_t *avail,
771 bool configured, bool early)
772{
773 struct pcpu *pcpu;
774 int cpu, nr, i;
775 u16 address;
776
777 nr = 0;
778 if (sclp.has_core_type && core->type != boot_core_type)
779 return nr;
780 cpu = cpumask_first(avail);
781 address = core->core_id << smp_cpu_mt_shift;
782 for (i = 0; (i <= smp_cpu_mtid) && (cpu < nr_cpu_ids); i++) {
783 if (pcpu_find_address(cpu_present_mask, address + i))
784 continue;
785 pcpu = pcpu_devices + cpu;
786 pcpu->address = address + i;
787 if (configured)
788 pcpu->state = CPU_STATE_CONFIGURED;
789 else
790 pcpu->state = CPU_STATE_STANDBY;
791 smp_cpu_set_polarization(cpu, POLARIZATION_UNKNOWN);
792 set_cpu_present(cpu, true);
793 if (!early && smp_add_present_cpu(cpu) != 0)
794 set_cpu_present(cpu, false);
795 else
796 nr++;
797 cpumask_clear_cpu(cpu, avail);
798 cpu = cpumask_next(cpu, avail);
799 }
800 return nr;
801}
802
803static int __smp_rescan_cpus(struct sclp_core_info *info, bool early)
804{
805 struct sclp_core_entry *core;
806 static cpumask_t avail;
807 bool configured;
808 u16 core_id;
809 int nr, i;
810
811 get_online_cpus();
812 mutex_lock(&smp_cpu_state_mutex);
813 nr = 0;
814 cpumask_xor(&avail, cpu_possible_mask, cpu_present_mask);
815 /*
816 * Add IPL core first (which got logical CPU number 0) to make sure
817 * that all SMT threads get subsequent logical CPU numbers.
818 */
819 if (early) {
820 core_id = pcpu_devices[0].address >> smp_cpu_mt_shift;
821 for (i = 0; i < info->configured; i++) {
822 core = &info->core[i];
823 if (core->core_id == core_id) {
824 nr += smp_add_core(core, &avail, true, early);
825 break;
826 }
827 }
828 }
829 for (i = 0; i < info->combined; i++) {
830 configured = i < info->configured;
831 nr += smp_add_core(&info->core[i], &avail, configured, early);
832 }
833 mutex_unlock(&smp_cpu_state_mutex);
834 put_online_cpus();
835 return nr;
836}
837
838void __init smp_detect_cpus(void)
839{
840 unsigned int cpu, mtid, c_cpus, s_cpus;
841 struct sclp_core_info *info;
842 u16 address;
843
844 /* Get CPU information */
845 info = memblock_alloc(sizeof(*info), 8);
846 if (!info)
847 panic("%s: Failed to allocate %zu bytes align=0x%x\n",
848 __func__, sizeof(*info), 8);
849 smp_get_core_info(info, 1);
850 /* Find boot CPU type */
851 if (sclp.has_core_type) {
852 address = stap();
853 for (cpu = 0; cpu < info->combined; cpu++)
854 if (info->core[cpu].core_id == address) {
855 /* The boot cpu dictates the cpu type. */
856 boot_core_type = info->core[cpu].type;
857 break;
858 }
859 if (cpu >= info->combined)
860 panic("Could not find boot CPU type");
861 }
862
863 /* Set multi-threading state for the current system */
864 mtid = boot_core_type ? sclp.mtid : sclp.mtid_cp;
865 mtid = (mtid < smp_max_threads) ? mtid : smp_max_threads - 1;
866 pcpu_set_smt(mtid);
867
868 /* Print number of CPUs */
869 c_cpus = s_cpus = 0;
870 for (cpu = 0; cpu < info->combined; cpu++) {
871 if (sclp.has_core_type &&
872 info->core[cpu].type != boot_core_type)
873 continue;
874 if (cpu < info->configured)
875 c_cpus += smp_cpu_mtid + 1;
876 else
877 s_cpus += smp_cpu_mtid + 1;
878 }
879 pr_info("%d configured CPUs, %d standby CPUs\n", c_cpus, s_cpus);
880
881 /* Add CPUs present at boot */
882 __smp_rescan_cpus(info, true);
883 memblock_free_early((unsigned long)info, sizeof(*info));
884}
885
886static void smp_init_secondary(void)
887{
888 int cpu = raw_smp_processor_id();
889
890 S390_lowcore.last_update_clock = get_tod_clock();
891 restore_access_regs(S390_lowcore.access_regs_save_area);
892 cpu_init();
893 rcu_cpu_starting(cpu);
894 init_cpu_timer();
895 vtime_init();
896 vdso_getcpu_init();
897 pfault_init();
898 cpumask_set_cpu(cpu, &cpu_setup_mask);
899 update_cpu_masks();
900 notify_cpu_starting(cpu);
901 if (topology_cpu_dedicated(cpu))
902 set_cpu_flag(CIF_DEDICATED_CPU);
903 else
904 clear_cpu_flag(CIF_DEDICATED_CPU);
905 set_cpu_online(cpu, true);
906 inc_irq_stat(CPU_RST);
907 local_irq_enable();
908 cpu_startup_entry(CPUHP_AP_ONLINE_IDLE);
909}
910
911/*
912 * Activate a secondary processor.
913 */
914static void smp_start_secondary(void *cpuvoid)
915{
916 S390_lowcore.restart_stack = (unsigned long) restart_stack;
917 S390_lowcore.restart_fn = (unsigned long) do_restart;
918 S390_lowcore.restart_data = 0;
919 S390_lowcore.restart_source = -1U;
920 S390_lowcore.restart_flags = 0;
921 call_on_stack_noreturn(smp_init_secondary, S390_lowcore.kernel_stack);
922}
923
924/* Upping and downing of CPUs */
925int __cpu_up(unsigned int cpu, struct task_struct *tidle)
926{
927 struct pcpu *pcpu = pcpu_devices + cpu;
928 int rc;
929
930 if (pcpu->state != CPU_STATE_CONFIGURED)
931 return -EIO;
932 if (pcpu_sigp_retry(pcpu, SIGP_INITIAL_CPU_RESET, 0) !=
933 SIGP_CC_ORDER_CODE_ACCEPTED)
934 return -EIO;
935
936 rc = pcpu_alloc_lowcore(pcpu, cpu);
937 if (rc)
938 return rc;
939 pcpu_prepare_secondary(pcpu, cpu);
940 pcpu_attach_task(pcpu, tidle);
941 pcpu_start_fn(pcpu, smp_start_secondary, NULL);
942 /* Wait until cpu puts itself in the online & active maps */
943 while (!cpu_online(cpu))
944 cpu_relax();
945 return 0;
946}
947
948static unsigned int setup_possible_cpus __initdata;
949
950static int __init _setup_possible_cpus(char *s)
951{
952 get_option(&s, &setup_possible_cpus);
953 return 0;
954}
955early_param("possible_cpus", _setup_possible_cpus);
956
957int __cpu_disable(void)
958{
959 unsigned long cregs[16];
960 int cpu;
961
962 /* Handle possible pending IPIs */
963 smp_handle_ext_call();
964 cpu = smp_processor_id();
965 set_cpu_online(cpu, false);
966 cpumask_clear_cpu(cpu, &cpu_setup_mask);
967 update_cpu_masks();
968 /* Disable pseudo page faults on this cpu. */
969 pfault_fini();
970 /* Disable interrupt sources via control register. */
971 __ctl_store(cregs, 0, 15);
972 cregs[0] &= ~0x0000ee70UL; /* disable all external interrupts */
973 cregs[6] &= ~0xff000000UL; /* disable all I/O interrupts */
974 cregs[14] &= ~0x1f000000UL; /* disable most machine checks */
975 __ctl_load(cregs, 0, 15);
976 clear_cpu_flag(CIF_NOHZ_DELAY);
977 return 0;
978}
979
980void __cpu_die(unsigned int cpu)
981{
982 struct pcpu *pcpu;
983
984 /* Wait until target cpu is down */
985 pcpu = pcpu_devices + cpu;
986 while (!pcpu_stopped(pcpu))
987 cpu_relax();
988 pcpu_free_lowcore(pcpu);
989 cpumask_clear_cpu(cpu, mm_cpumask(&init_mm));
990 cpumask_clear_cpu(cpu, &init_mm.context.cpu_attach_mask);
991}
992
993void __noreturn cpu_die(void)
994{
995 idle_task_exit();
996 __bpon();
997 pcpu_sigp_retry(pcpu_devices + smp_processor_id(), SIGP_STOP, 0);
998 for (;;) ;
999}
1000
1001void __init smp_fill_possible_mask(void)
1002{
1003 unsigned int possible, sclp_max, cpu;
1004
1005 sclp_max = max(sclp.mtid, sclp.mtid_cp) + 1;
1006 sclp_max = min(smp_max_threads, sclp_max);
1007 sclp_max = (sclp.max_cores * sclp_max) ?: nr_cpu_ids;
1008 possible = setup_possible_cpus ?: nr_cpu_ids;
1009 possible = min(possible, sclp_max);
1010 for (cpu = 0; cpu < possible && cpu < nr_cpu_ids; cpu++)
1011 set_cpu_possible(cpu, true);
1012}
1013
1014void __init smp_prepare_cpus(unsigned int max_cpus)
1015{
1016 /* request the 0x1201 emergency signal external interrupt */
1017 if (register_external_irq(EXT_IRQ_EMERGENCY_SIG, do_ext_call_interrupt))
1018 panic("Couldn't request external interrupt 0x1201");
1019 /* request the 0x1202 external call external interrupt */
1020 if (register_external_irq(EXT_IRQ_EXTERNAL_CALL, do_ext_call_interrupt))
1021 panic("Couldn't request external interrupt 0x1202");
1022}
1023
1024void __init smp_prepare_boot_cpu(void)
1025{
1026 struct pcpu *pcpu = pcpu_devices;
1027
1028 WARN_ON(!cpu_present(0) || !cpu_online(0));
1029 pcpu->state = CPU_STATE_CONFIGURED;
1030 S390_lowcore.percpu_offset = __per_cpu_offset[0];
1031 smp_cpu_set_polarization(0, POLARIZATION_UNKNOWN);
1032}
1033
1034void __init smp_setup_processor_id(void)
1035{
1036 pcpu_devices[0].address = stap();
1037 S390_lowcore.cpu_nr = 0;
1038 S390_lowcore.spinlock_lockval = arch_spin_lockval(0);
1039 S390_lowcore.spinlock_index = 0;
1040}
1041
1042/*
1043 * the frequency of the profiling timer can be changed
1044 * by writing a multiplier value into /proc/profile.
1045 *
1046 * usually you want to run this on all CPUs ;)
1047 */
1048int setup_profiling_timer(unsigned int multiplier)
1049{
1050 return 0;
1051}
1052
1053static ssize_t cpu_configure_show(struct device *dev,
1054 struct device_attribute *attr, char *buf)
1055{
1056 ssize_t count;
1057
1058 mutex_lock(&smp_cpu_state_mutex);
1059 count = sprintf(buf, "%d\n", pcpu_devices[dev->id].state);
1060 mutex_unlock(&smp_cpu_state_mutex);
1061 return count;
1062}
1063
1064static ssize_t cpu_configure_store(struct device *dev,
1065 struct device_attribute *attr,
1066 const char *buf, size_t count)
1067{
1068 struct pcpu *pcpu;
1069 int cpu, val, rc, i;
1070 char delim;
1071
1072 if (sscanf(buf, "%d %c", &val, &delim) != 1)
1073 return -EINVAL;
1074 if (val != 0 && val != 1)
1075 return -EINVAL;
1076 get_online_cpus();
1077 mutex_lock(&smp_cpu_state_mutex);
1078 rc = -EBUSY;
1079 /* disallow configuration changes of online cpus and cpu 0 */
1080 cpu = dev->id;
1081 cpu = smp_get_base_cpu(cpu);
1082 if (cpu == 0)
1083 goto out;
1084 for (i = 0; i <= smp_cpu_mtid; i++)
1085 if (cpu_online(cpu + i))
1086 goto out;
1087 pcpu = pcpu_devices + cpu;
1088 rc = 0;
1089 switch (val) {
1090 case 0:
1091 if (pcpu->state != CPU_STATE_CONFIGURED)
1092 break;
1093 rc = sclp_core_deconfigure(pcpu->address >> smp_cpu_mt_shift);
1094 if (rc)
1095 break;
1096 for (i = 0; i <= smp_cpu_mtid; i++) {
1097 if (cpu + i >= nr_cpu_ids || !cpu_present(cpu + i))
1098 continue;
1099 pcpu[i].state = CPU_STATE_STANDBY;
1100 smp_cpu_set_polarization(cpu + i,
1101 POLARIZATION_UNKNOWN);
1102 }
1103 topology_expect_change();
1104 break;
1105 case 1:
1106 if (pcpu->state != CPU_STATE_STANDBY)
1107 break;
1108 rc = sclp_core_configure(pcpu->address >> smp_cpu_mt_shift);
1109 if (rc)
1110 break;
1111 for (i = 0; i <= smp_cpu_mtid; i++) {
1112 if (cpu + i >= nr_cpu_ids || !cpu_present(cpu + i))
1113 continue;
1114 pcpu[i].state = CPU_STATE_CONFIGURED;
1115 smp_cpu_set_polarization(cpu + i,
1116 POLARIZATION_UNKNOWN);
1117 }
1118 topology_expect_change();
1119 break;
1120 default:
1121 break;
1122 }
1123out:
1124 mutex_unlock(&smp_cpu_state_mutex);
1125 put_online_cpus();
1126 return rc ? rc : count;
1127}
1128static DEVICE_ATTR(configure, 0644, cpu_configure_show, cpu_configure_store);
1129
1130static ssize_t show_cpu_address(struct device *dev,
1131 struct device_attribute *attr, char *buf)
1132{
1133 return sprintf(buf, "%d\n", pcpu_devices[dev->id].address);
1134}
1135static DEVICE_ATTR(address, 0444, show_cpu_address, NULL);
1136
1137static struct attribute *cpu_common_attrs[] = {
1138 &dev_attr_configure.attr,
1139 &dev_attr_address.attr,
1140 NULL,
1141};
1142
1143static struct attribute_group cpu_common_attr_group = {
1144 .attrs = cpu_common_attrs,
1145};
1146
1147static struct attribute *cpu_online_attrs[] = {
1148 &dev_attr_idle_count.attr,
1149 &dev_attr_idle_time_us.attr,
1150 NULL,
1151};
1152
1153static struct attribute_group cpu_online_attr_group = {
1154 .attrs = cpu_online_attrs,
1155};
1156
1157static int smp_cpu_online(unsigned int cpu)
1158{
1159 struct device *s = &per_cpu(cpu_device, cpu)->dev;
1160
1161 return sysfs_create_group(&s->kobj, &cpu_online_attr_group);
1162}
1163
1164static int smp_cpu_pre_down(unsigned int cpu)
1165{
1166 struct device *s = &per_cpu(cpu_device, cpu)->dev;
1167
1168 sysfs_remove_group(&s->kobj, &cpu_online_attr_group);
1169 return 0;
1170}
1171
1172static int smp_add_present_cpu(int cpu)
1173{
1174 struct device *s;
1175 struct cpu *c;
1176 int rc;
1177
1178 c = kzalloc(sizeof(*c), GFP_KERNEL);
1179 if (!c)
1180 return -ENOMEM;
1181 per_cpu(cpu_device, cpu) = c;
1182 s = &c->dev;
1183 c->hotpluggable = 1;
1184 rc = register_cpu(c, cpu);
1185 if (rc)
1186 goto out;
1187 rc = sysfs_create_group(&s->kobj, &cpu_common_attr_group);
1188 if (rc)
1189 goto out_cpu;
1190 rc = topology_cpu_init(c);
1191 if (rc)
1192 goto out_topology;
1193 return 0;
1194
1195out_topology:
1196 sysfs_remove_group(&s->kobj, &cpu_common_attr_group);
1197out_cpu:
1198 unregister_cpu(c);
1199out:
1200 return rc;
1201}
1202
1203int __ref smp_rescan_cpus(void)
1204{
1205 struct sclp_core_info *info;
1206 int nr;
1207
1208 info = kzalloc(sizeof(*info), GFP_KERNEL);
1209 if (!info)
1210 return -ENOMEM;
1211 smp_get_core_info(info, 0);
1212 nr = __smp_rescan_cpus(info, false);
1213 kfree(info);
1214 if (nr)
1215 topology_schedule_update();
1216 return 0;
1217}
1218
1219static ssize_t __ref rescan_store(struct device *dev,
1220 struct device_attribute *attr,
1221 const char *buf,
1222 size_t count)
1223{
1224 int rc;
1225
1226 rc = lock_device_hotplug_sysfs();
1227 if (rc)
1228 return rc;
1229 rc = smp_rescan_cpus();
1230 unlock_device_hotplug();
1231 return rc ? rc : count;
1232}
1233static DEVICE_ATTR_WO(rescan);
1234
1235static int __init s390_smp_init(void)
1236{
1237 int cpu, rc = 0;
1238
1239 rc = device_create_file(cpu_subsys.dev_root, &dev_attr_rescan);
1240 if (rc)
1241 return rc;
1242 for_each_present_cpu(cpu) {
1243 rc = smp_add_present_cpu(cpu);
1244 if (rc)
1245 goto out;
1246 }
1247
1248 rc = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "s390/smp:online",
1249 smp_cpu_online, smp_cpu_pre_down);
1250 rc = rc <= 0 ? rc : 0;
1251out:
1252 return rc;
1253}
1254subsys_initcall(s390_smp_init);
1255
1256static __always_inline void set_new_lowcore(struct lowcore *lc)
1257{
1258 union register_pair dst, src;
1259 u32 pfx;
1260
1261 src.even = (unsigned long) &S390_lowcore;
1262 src.odd = sizeof(S390_lowcore);
1263 dst.even = (unsigned long) lc;
1264 dst.odd = sizeof(*lc);
1265 pfx = (unsigned long) lc;
1266
1267 asm volatile(
1268 " mvcl %[dst],%[src]\n"
1269 " spx %[pfx]\n"
1270 : [dst] "+&d" (dst.pair), [src] "+&d" (src.pair)
1271 : [pfx] "Q" (pfx)
1272 : "memory", "cc");
1273}
1274
1275static int __init smp_reinit_ipl_cpu(void)
1276{
1277 unsigned long async_stack, nodat_stack, mcck_stack;
1278 struct lowcore *lc, *lc_ipl;
1279 unsigned long flags;
1280
1281 lc_ipl = lowcore_ptr[0];
1282 lc = (struct lowcore *) __get_free_pages(GFP_KERNEL | GFP_DMA, LC_ORDER);
1283 nodat_stack = __get_free_pages(GFP_KERNEL, THREAD_SIZE_ORDER);
1284 async_stack = stack_alloc();
1285 mcck_stack = stack_alloc();
1286 if (!lc || !nodat_stack || !async_stack || !mcck_stack)
1287 panic("Couldn't allocate memory");
1288
1289 local_irq_save(flags);
1290 local_mcck_disable();
1291 set_new_lowcore(lc);
1292 S390_lowcore.nodat_stack = nodat_stack + STACK_INIT_OFFSET;
1293 S390_lowcore.async_stack = async_stack + STACK_INIT_OFFSET;
1294 S390_lowcore.mcck_stack = mcck_stack + STACK_INIT_OFFSET;
1295 lowcore_ptr[0] = lc;
1296 local_mcck_enable();
1297 local_irq_restore(flags);
1298
1299 free_pages(lc_ipl->async_stack - STACK_INIT_OFFSET, THREAD_SIZE_ORDER);
1300 memblock_free_late(lc_ipl->mcck_stack - STACK_INIT_OFFSET, THREAD_SIZE);
1301 memblock_free_late((unsigned long) lc_ipl, sizeof(*lc_ipl));
1302
1303 return 0;
1304}
1305early_initcall(smp_reinit_ipl_cpu);