GNU Linux-libre 5.4.274-gnu1
[releases.git] / arch / s390 / kernel / smp.c
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/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/kprobes.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 <asm/stacktrace.h>
57 #include <asm/topology.h>
58 #include "entry.h"
59
60 enum {
61         ec_schedule = 0,
62         ec_call_function_single,
63         ec_stop_cpu,
64 };
65
66 enum {
67         CPU_STATE_STANDBY,
68         CPU_STATE_CONFIGURED,
69 };
70
71 static DEFINE_PER_CPU(struct cpu *, cpu_device);
72
73 struct pcpu {
74         struct lowcore *lowcore;        /* lowcore page(s) for the cpu */
75         unsigned long ec_mask;          /* bit mask for ec_xxx functions */
76         unsigned long ec_clk;           /* sigp timestamp for ec_xxx */
77         signed char state;              /* physical cpu state */
78         signed char polarization;       /* physical polarization */
79         u16 address;                    /* physical cpu address */
80 };
81
82 static u8 boot_core_type;
83 static struct pcpu pcpu_devices[NR_CPUS];
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 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 static int pcpu_alloc_lowcore(struct pcpu *pcpu, int cpu)
190 {
191         unsigned long async_stack, nodat_stack;
192         struct lowcore *lc;
193
194         if (pcpu != &pcpu_devices[0]) {
195                 pcpu->lowcore = (struct lowcore *)
196                         __get_free_pages(GFP_KERNEL | GFP_DMA, LC_ORDER);
197                 nodat_stack = __get_free_pages(GFP_KERNEL, THREAD_SIZE_ORDER);
198                 if (!pcpu->lowcore || !nodat_stack)
199                         goto out;
200         } else {
201                 nodat_stack = pcpu->lowcore->nodat_stack - STACK_INIT_OFFSET;
202         }
203         async_stack = stack_alloc();
204         if (!async_stack)
205                 goto out;
206         lc = pcpu->lowcore;
207         memcpy(lc, &S390_lowcore, 512);
208         memset((char *) lc + 512, 0, sizeof(*lc) - 512);
209         lc->async_stack = async_stack + STACK_INIT_OFFSET;
210         lc->nodat_stack = nodat_stack + STACK_INIT_OFFSET;
211         lc->cpu_nr = cpu;
212         lc->spinlock_lockval = arch_spin_lockval(cpu);
213         lc->spinlock_index = 0;
214         lc->br_r1_trampoline = 0x07f1;  /* br %r1 */
215         lc->return_lpswe = gen_lpswe(__LC_RETURN_PSW);
216         lc->return_mcck_lpswe = gen_lpswe(__LC_RETURN_MCCK_PSW);
217         if (nmi_alloc_per_cpu(lc))
218                 goto out_async;
219         if (vdso_alloc_per_cpu(lc))
220                 goto out_mcesa;
221         lowcore_ptr[cpu] = lc;
222         pcpu_sigp_retry(pcpu, SIGP_SET_PREFIX, (u32)(unsigned long) lc);
223         return 0;
224
225 out_mcesa:
226         nmi_free_per_cpu(lc);
227 out_async:
228         stack_free(async_stack);
229 out:
230         if (pcpu != &pcpu_devices[0]) {
231                 free_pages(nodat_stack, THREAD_SIZE_ORDER);
232                 free_pages((unsigned long) pcpu->lowcore, LC_ORDER);
233         }
234         return -ENOMEM;
235 }
236
237 static void pcpu_free_lowcore(struct pcpu *pcpu)
238 {
239         unsigned long async_stack, nodat_stack, lowcore;
240
241         nodat_stack = pcpu->lowcore->nodat_stack - STACK_INIT_OFFSET;
242         async_stack = pcpu->lowcore->async_stack - STACK_INIT_OFFSET;
243         lowcore = (unsigned long) pcpu->lowcore;
244
245         pcpu_sigp_retry(pcpu, SIGP_SET_PREFIX, 0);
246         lowcore_ptr[pcpu - pcpu_devices] = NULL;
247         vdso_free_per_cpu(pcpu->lowcore);
248         nmi_free_per_cpu(pcpu->lowcore);
249         stack_free(async_stack);
250         if (pcpu == &pcpu_devices[0])
251                 return;
252         free_pages(nodat_stack, THREAD_SIZE_ORDER);
253         free_pages(lowcore, LC_ORDER);
254 }
255
256 static void pcpu_prepare_secondary(struct pcpu *pcpu, int cpu)
257 {
258         struct lowcore *lc = pcpu->lowcore;
259
260         cpumask_set_cpu(cpu, &init_mm.context.cpu_attach_mask);
261         cpumask_set_cpu(cpu, mm_cpumask(&init_mm));
262         lc->cpu_nr = cpu;
263         lc->spinlock_lockval = arch_spin_lockval(cpu);
264         lc->spinlock_index = 0;
265         lc->percpu_offset = __per_cpu_offset[cpu];
266         lc->kernel_asce = S390_lowcore.kernel_asce;
267         lc->user_asce = S390_lowcore.kernel_asce;
268         lc->machine_flags = S390_lowcore.machine_flags;
269         lc->user_timer = lc->system_timer =
270                 lc->steal_timer = lc->avg_steal_timer = 0;
271         __ctl_store(lc->cregs_save_area, 0, 15);
272         lc->cregs_save_area[1] = lc->kernel_asce;
273         lc->cregs_save_area[7] = lc->vdso_asce;
274         save_access_regs((unsigned int *) lc->access_regs_save_area);
275         memcpy(lc->stfle_fac_list, S390_lowcore.stfle_fac_list,
276                sizeof(lc->stfle_fac_list));
277         memcpy(lc->alt_stfle_fac_list, S390_lowcore.alt_stfle_fac_list,
278                sizeof(lc->alt_stfle_fac_list));
279         arch_spin_lock_setup(cpu);
280 }
281
282 static void pcpu_attach_task(struct pcpu *pcpu, struct task_struct *tsk)
283 {
284         struct lowcore *lc = pcpu->lowcore;
285
286         lc->kernel_stack = (unsigned long) task_stack_page(tsk)
287                 + THREAD_SIZE - STACK_FRAME_OVERHEAD - sizeof(struct pt_regs);
288         lc->current_task = (unsigned long) tsk;
289         lc->lpp = LPP_MAGIC;
290         lc->current_pid = tsk->pid;
291         lc->user_timer = tsk->thread.user_timer;
292         lc->guest_timer = tsk->thread.guest_timer;
293         lc->system_timer = tsk->thread.system_timer;
294         lc->hardirq_timer = tsk->thread.hardirq_timer;
295         lc->softirq_timer = tsk->thread.softirq_timer;
296         lc->steal_timer = 0;
297 }
298
299 static void pcpu_start_fn(struct pcpu *pcpu, void (*func)(void *), void *data)
300 {
301         struct lowcore *lc = pcpu->lowcore;
302
303         lc->restart_stack = lc->nodat_stack;
304         lc->restart_fn = (unsigned long) func;
305         lc->restart_data = (unsigned long) data;
306         lc->restart_source = -1UL;
307         pcpu_sigp_retry(pcpu, SIGP_RESTART, 0);
308 }
309
310 /*
311  * Call function via PSW restart on pcpu and stop the current cpu.
312  */
313 static void __pcpu_delegate(void (*func)(void*), void *data)
314 {
315         func(data);     /* should not return */
316 }
317
318 static void __no_sanitize_address pcpu_delegate(struct pcpu *pcpu,
319                                                 void (*func)(void *),
320                                                 void *data, unsigned long stack)
321 {
322         struct lowcore *lc = lowcore_ptr[pcpu - pcpu_devices];
323         unsigned long source_cpu = stap();
324
325         __load_psw_mask(PSW_KERNEL_BITS | PSW_MASK_DAT);
326         if (pcpu->address == source_cpu)
327                 CALL_ON_STACK(__pcpu_delegate, stack, 2, func, data);
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  */
350 static 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  */
370 void 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  */
385 void smp_call_ipl_cpu(void (*func)(void *), void *data)
386 {
387         struct lowcore *lc = pcpu_devices->lowcore;
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
396 int 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
406 bool notrace arch_vcpu_is_preempted(int cpu)
407 {
408         if (test_cpu_flag_of(CIF_ENABLED_WAIT, cpu))
409                 return false;
410         if (pcpu_running(pcpu_devices + cpu))
411                 return false;
412         return true;
413 }
414 EXPORT_SYMBOL(arch_vcpu_is_preempted);
415
416 void notrace smp_yield_cpu(int cpu)
417 {
418         if (MACHINE_HAS_DIAG9C) {
419                 diag_stat_inc_norecursion(DIAG_STAT_X09C);
420                 asm volatile("diag %0,0,0x9c"
421                              : : "d" (pcpu_devices[cpu].address));
422         } else if (MACHINE_HAS_DIAG44 && !smp_cpu_mtid) {
423                 diag_stat_inc_norecursion(DIAG_STAT_X044);
424                 asm volatile("diag 0,0,0x44");
425         }
426 }
427
428 /*
429  * Send cpus emergency shutdown signal. This gives the cpus the
430  * opportunity to complete outstanding interrupts.
431  */
432 void notrace smp_emergency_stop(void)
433 {
434         cpumask_t cpumask;
435         u64 end;
436         int cpu;
437
438         cpumask_copy(&cpumask, cpu_online_mask);
439         cpumask_clear_cpu(smp_processor_id(), &cpumask);
440
441         end = get_tod_clock() + (1000000UL << 12);
442         for_each_cpu(cpu, &cpumask) {
443                 struct pcpu *pcpu = pcpu_devices + cpu;
444                 set_bit(ec_stop_cpu, &pcpu->ec_mask);
445                 while (__pcpu_sigp(pcpu->address, SIGP_EMERGENCY_SIGNAL,
446                                    0, NULL) == SIGP_CC_BUSY &&
447                        get_tod_clock() < end)
448                         cpu_relax();
449         }
450         while (get_tod_clock() < end) {
451                 for_each_cpu(cpu, &cpumask)
452                         if (pcpu_stopped(pcpu_devices + cpu))
453                                 cpumask_clear_cpu(cpu, &cpumask);
454                 if (cpumask_empty(&cpumask))
455                         break;
456                 cpu_relax();
457         }
458 }
459 NOKPROBE_SYMBOL(smp_emergency_stop);
460
461 /*
462  * Stop all cpus but the current one.
463  */
464 void smp_send_stop(void)
465 {
466         int cpu;
467
468         /* Disable all interrupts/machine checks */
469         __load_psw_mask(PSW_KERNEL_BITS | PSW_MASK_DAT);
470         trace_hardirqs_off();
471
472         debug_set_critical();
473
474         if (oops_in_progress)
475                 smp_emergency_stop();
476
477         /* stop all processors */
478         for_each_online_cpu(cpu) {
479                 if (cpu == smp_processor_id())
480                         continue;
481                 pcpu_sigp_retry(pcpu_devices + cpu, SIGP_STOP, 0);
482                 while (!pcpu_stopped(pcpu_devices + cpu))
483                         cpu_relax();
484         }
485 }
486
487 /*
488  * This is the main routine where commands issued by other
489  * cpus are handled.
490  */
491 static void smp_handle_ext_call(void)
492 {
493         unsigned long bits;
494
495         /* handle bit signal external calls */
496         bits = xchg(&pcpu_devices[smp_processor_id()].ec_mask, 0);
497         if (test_bit(ec_stop_cpu, &bits))
498                 smp_stop_cpu();
499         if (test_bit(ec_schedule, &bits))
500                 scheduler_ipi();
501         if (test_bit(ec_call_function_single, &bits))
502                 generic_smp_call_function_single_interrupt();
503 }
504
505 static void do_ext_call_interrupt(struct ext_code ext_code,
506                                   unsigned int param32, unsigned long param64)
507 {
508         inc_irq_stat(ext_code.code == 0x1202 ? IRQEXT_EXC : IRQEXT_EMS);
509         smp_handle_ext_call();
510 }
511
512 void arch_send_call_function_ipi_mask(const struct cpumask *mask)
513 {
514         int cpu;
515
516         for_each_cpu(cpu, mask)
517                 pcpu_ec_call(pcpu_devices + cpu, ec_call_function_single);
518 }
519
520 void arch_send_call_function_single_ipi(int cpu)
521 {
522         pcpu_ec_call(pcpu_devices + cpu, ec_call_function_single);
523 }
524
525 /*
526  * this function sends a 'reschedule' IPI to another CPU.
527  * it goes straight through and wastes no time serializing
528  * anything. Worst case is that we lose a reschedule ...
529  */
530 void smp_send_reschedule(int cpu)
531 {
532         pcpu_ec_call(pcpu_devices + cpu, ec_schedule);
533 }
534
535 /*
536  * parameter area for the set/clear control bit callbacks
537  */
538 struct ec_creg_mask_parms {
539         unsigned long orval;
540         unsigned long andval;
541         int cr;
542 };
543
544 /*
545  * callback for setting/clearing control bits
546  */
547 static void smp_ctl_bit_callback(void *info)
548 {
549         struct ec_creg_mask_parms *pp = info;
550         unsigned long cregs[16];
551
552         __ctl_store(cregs, 0, 15);
553         cregs[pp->cr] = (cregs[pp->cr] & pp->andval) | pp->orval;
554         __ctl_load(cregs, 0, 15);
555 }
556
557 /*
558  * Set a bit in a control register of all cpus
559  */
560 void smp_ctl_set_bit(int cr, int bit)
561 {
562         struct ec_creg_mask_parms parms = { 1UL << bit, -1UL, cr };
563
564         on_each_cpu(smp_ctl_bit_callback, &parms, 1);
565 }
566 EXPORT_SYMBOL(smp_ctl_set_bit);
567
568 /*
569  * Clear a bit in a control register of all cpus
570  */
571 void smp_ctl_clear_bit(int cr, int bit)
572 {
573         struct ec_creg_mask_parms parms = { 0, ~(1UL << bit), cr };
574
575         on_each_cpu(smp_ctl_bit_callback, &parms, 1);
576 }
577 EXPORT_SYMBOL(smp_ctl_clear_bit);
578
579 #ifdef CONFIG_CRASH_DUMP
580
581 int smp_store_status(int cpu)
582 {
583         struct pcpu *pcpu = pcpu_devices + cpu;
584         unsigned long pa;
585
586         pa = __pa(&pcpu->lowcore->floating_pt_save_area);
587         if (__pcpu_sigp_relax(pcpu->address, SIGP_STORE_STATUS_AT_ADDRESS,
588                               pa) != SIGP_CC_ORDER_CODE_ACCEPTED)
589                 return -EIO;
590         if (!MACHINE_HAS_VX && !MACHINE_HAS_GS)
591                 return 0;
592         pa = __pa(pcpu->lowcore->mcesad & MCESA_ORIGIN_MASK);
593         if (MACHINE_HAS_GS)
594                 pa |= pcpu->lowcore->mcesad & MCESA_LC_MASK;
595         if (__pcpu_sigp_relax(pcpu->address, SIGP_STORE_ADDITIONAL_STATUS,
596                               pa) != SIGP_CC_ORDER_CODE_ACCEPTED)
597                 return -EIO;
598         return 0;
599 }
600
601 /*
602  * Collect CPU state of the previous, crashed system.
603  * There are four cases:
604  * 1) standard zfcp dump
605  *    condition: OLDMEM_BASE == NULL && ipl_info.type == IPL_TYPE_FCP_DUMP
606  *    The state for all CPUs except the boot CPU needs to be collected
607  *    with sigp stop-and-store-status. The boot CPU state is located in
608  *    the absolute lowcore of the memory stored in the HSA. The zcore code
609  *    will copy the boot CPU state from the HSA.
610  * 2) stand-alone kdump for SCSI (zfcp dump with swapped memory)
611  *    condition: OLDMEM_BASE != NULL && ipl_info.type == IPL_TYPE_FCP_DUMP
612  *    The state for all CPUs except the boot CPU needs to be collected
613  *    with sigp stop-and-store-status. The firmware or the boot-loader
614  *    stored the registers of the boot CPU in the absolute lowcore in the
615  *    memory of the old system.
616  * 3) kdump and the old kernel did not store the CPU state,
617  *    or stand-alone kdump for DASD
618  *    condition: OLDMEM_BASE != NULL && !is_kdump_kernel()
619  *    The state for all CPUs except the boot CPU needs to be collected
620  *    with sigp stop-and-store-status. The kexec code or the boot-loader
621  *    stored the registers of the boot CPU in the memory of the old system.
622  * 4) kdump and the old kernel stored the CPU state
623  *    condition: OLDMEM_BASE != NULL && is_kdump_kernel()
624  *    This case does not exist for s390 anymore, setup_arch explicitly
625  *    deactivates the elfcorehdr= kernel parameter
626  */
627 static __init void smp_save_cpu_vxrs(struct save_area *sa, u16 addr,
628                                      bool is_boot_cpu, unsigned long page)
629 {
630         __vector128 *vxrs = (__vector128 *) page;
631
632         if (is_boot_cpu)
633                 vxrs = boot_cpu_vector_save_area;
634         else
635                 __pcpu_sigp_relax(addr, SIGP_STORE_ADDITIONAL_STATUS, page);
636         save_area_add_vxrs(sa, vxrs);
637 }
638
639 static __init void smp_save_cpu_regs(struct save_area *sa, u16 addr,
640                                      bool is_boot_cpu, unsigned long page)
641 {
642         void *regs = (void *) page;
643
644         if (is_boot_cpu)
645                 copy_oldmem_kernel(regs, (void *) __LC_FPREGS_SAVE_AREA, 512);
646         else
647                 __pcpu_sigp_relax(addr, SIGP_STORE_STATUS_AT_ADDRESS, page);
648         save_area_add_regs(sa, regs);
649 }
650
651 void __init smp_save_dump_cpus(void)
652 {
653         int addr, boot_cpu_addr, max_cpu_addr;
654         struct save_area *sa;
655         unsigned long page;
656         bool is_boot_cpu;
657
658         if (!(OLDMEM_BASE || ipl_info.type == IPL_TYPE_FCP_DUMP))
659                 /* No previous system present, normal boot. */
660                 return;
661         /* Allocate a page as dumping area for the store status sigps */
662         page = memblock_phys_alloc_range(PAGE_SIZE, PAGE_SIZE, 0, 1UL << 31);
663         if (!page)
664                 panic("ERROR: Failed to allocate %lx bytes below %lx\n",
665                       PAGE_SIZE, 1UL << 31);
666
667         /* Set multi-threading state to the previous system. */
668         pcpu_set_smt(sclp.mtid_prev);
669         boot_cpu_addr = stap();
670         max_cpu_addr = SCLP_MAX_CORES << sclp.mtid_prev;
671         for (addr = 0; addr <= max_cpu_addr; addr++) {
672                 if (__pcpu_sigp_relax(addr, SIGP_SENSE, 0) ==
673                     SIGP_CC_NOT_OPERATIONAL)
674                         continue;
675                 is_boot_cpu = (addr == boot_cpu_addr);
676                 /* Allocate save area */
677                 sa = save_area_alloc(is_boot_cpu);
678                 if (!sa)
679                         panic("could not allocate memory for save area\n");
680                 if (MACHINE_HAS_VX)
681                         /* Get the vector registers */
682                         smp_save_cpu_vxrs(sa, addr, is_boot_cpu, page);
683                 /*
684                  * For a zfcp dump OLDMEM_BASE == NULL and the registers
685                  * of the boot CPU are stored in the HSA. To retrieve
686                  * these registers an SCLP request is required which is
687                  * done by drivers/s390/char/zcore.c:init_cpu_info()
688                  */
689                 if (!is_boot_cpu || OLDMEM_BASE)
690                         /* Get the CPU registers */
691                         smp_save_cpu_regs(sa, addr, is_boot_cpu, page);
692         }
693         memblock_free(page, PAGE_SIZE);
694         diag_dma_ops.diag308_reset();
695         pcpu_set_smt(0);
696 }
697 #endif /* CONFIG_CRASH_DUMP */
698
699 void smp_cpu_set_polarization(int cpu, int val)
700 {
701         pcpu_devices[cpu].polarization = val;
702 }
703
704 int smp_cpu_get_polarization(int cpu)
705 {
706         return pcpu_devices[cpu].polarization;
707 }
708
709 static void __ref smp_get_core_info(struct sclp_core_info *info, int early)
710 {
711         static int use_sigp_detection;
712         int address;
713
714         if (use_sigp_detection || sclp_get_core_info(info, early)) {
715                 use_sigp_detection = 1;
716                 for (address = 0;
717                      address < (SCLP_MAX_CORES << smp_cpu_mt_shift);
718                      address += (1U << smp_cpu_mt_shift)) {
719                         if (__pcpu_sigp_relax(address, SIGP_SENSE, 0) ==
720                             SIGP_CC_NOT_OPERATIONAL)
721                                 continue;
722                         info->core[info->configured].core_id =
723                                 address >> smp_cpu_mt_shift;
724                         info->configured++;
725                 }
726                 info->combined = info->configured;
727         }
728 }
729
730 static int smp_add_present_cpu(int cpu);
731
732 static int smp_add_core(struct sclp_core_entry *core, cpumask_t *avail,
733                         bool configured, bool early)
734 {
735         struct pcpu *pcpu;
736         int cpu, nr, i;
737         u16 address;
738
739         nr = 0;
740         if (sclp.has_core_type && core->type != boot_core_type)
741                 return nr;
742         cpu = cpumask_first(avail);
743         address = core->core_id << smp_cpu_mt_shift;
744         for (i = 0; (i <= smp_cpu_mtid) && (cpu < nr_cpu_ids); i++) {
745                 if (pcpu_find_address(cpu_present_mask, address + i))
746                         continue;
747                 pcpu = pcpu_devices + cpu;
748                 pcpu->address = address + i;
749                 if (configured)
750                         pcpu->state = CPU_STATE_CONFIGURED;
751                 else
752                         pcpu->state = CPU_STATE_STANDBY;
753                 smp_cpu_set_polarization(cpu, POLARIZATION_UNKNOWN);
754                 set_cpu_present(cpu, true);
755                 if (!early && smp_add_present_cpu(cpu) != 0)
756                         set_cpu_present(cpu, false);
757                 else
758                         nr++;
759                 cpumask_clear_cpu(cpu, avail);
760                 cpu = cpumask_next(cpu, avail);
761         }
762         return nr;
763 }
764
765 static int __smp_rescan_cpus(struct sclp_core_info *info, bool early)
766 {
767         struct sclp_core_entry *core;
768         static cpumask_t avail;
769         bool configured;
770         u16 core_id;
771         int nr, i;
772
773         nr = 0;
774         cpumask_xor(&avail, cpu_possible_mask, cpu_present_mask);
775         /*
776          * Add IPL core first (which got logical CPU number 0) to make sure
777          * that all SMT threads get subsequent logical CPU numbers.
778          */
779         if (early) {
780                 core_id = pcpu_devices[0].address >> smp_cpu_mt_shift;
781                 for (i = 0; i < info->configured; i++) {
782                         core = &info->core[i];
783                         if (core->core_id == core_id) {
784                                 nr += smp_add_core(core, &avail, true, early);
785                                 break;
786                         }
787                 }
788         }
789         for (i = 0; i < info->combined; i++) {
790                 configured = i < info->configured;
791                 nr += smp_add_core(&info->core[i], &avail, configured, early);
792         }
793         return nr;
794 }
795
796 void __init smp_detect_cpus(void)
797 {
798         unsigned int cpu, mtid, c_cpus, s_cpus;
799         struct sclp_core_info *info;
800         u16 address;
801
802         /* Get CPU information */
803         info = memblock_alloc(sizeof(*info), 8);
804         if (!info)
805                 panic("%s: Failed to allocate %zu bytes align=0x%x\n",
806                       __func__, sizeof(*info), 8);
807         smp_get_core_info(info, 1);
808         /* Find boot CPU type */
809         if (sclp.has_core_type) {
810                 address = stap();
811                 for (cpu = 0; cpu < info->combined; cpu++)
812                         if (info->core[cpu].core_id == address) {
813                                 /* The boot cpu dictates the cpu type. */
814                                 boot_core_type = info->core[cpu].type;
815                                 break;
816                         }
817                 if (cpu >= info->combined)
818                         panic("Could not find boot CPU type");
819         }
820
821         /* Set multi-threading state for the current system */
822         mtid = boot_core_type ? sclp.mtid : sclp.mtid_cp;
823         mtid = (mtid < smp_max_threads) ? mtid : smp_max_threads - 1;
824         pcpu_set_smt(mtid);
825
826         /* Print number of CPUs */
827         c_cpus = s_cpus = 0;
828         for (cpu = 0; cpu < info->combined; cpu++) {
829                 if (sclp.has_core_type &&
830                     info->core[cpu].type != boot_core_type)
831                         continue;
832                 if (cpu < info->configured)
833                         c_cpus += smp_cpu_mtid + 1;
834                 else
835                         s_cpus += smp_cpu_mtid + 1;
836         }
837         pr_info("%d configured CPUs, %d standby CPUs\n", c_cpus, s_cpus);
838
839         /* Add CPUs present at boot */
840         get_online_cpus();
841         __smp_rescan_cpus(info, true);
842         put_online_cpus();
843         memblock_free_early((unsigned long)info, sizeof(*info));
844 }
845
846 static void smp_init_secondary(void)
847 {
848         int cpu = raw_smp_processor_id();
849
850         S390_lowcore.last_update_clock = get_tod_clock();
851         restore_access_regs(S390_lowcore.access_regs_save_area);
852         set_cpu_flag(CIF_ASCE_PRIMARY);
853         set_cpu_flag(CIF_ASCE_SECONDARY);
854         cpu_init();
855         rcu_cpu_starting(cpu);
856         preempt_disable();
857         init_cpu_timer();
858         vtime_init();
859         pfault_init();
860         notify_cpu_starting(smp_processor_id());
861         if (topology_cpu_dedicated(cpu))
862                 set_cpu_flag(CIF_DEDICATED_CPU);
863         else
864                 clear_cpu_flag(CIF_DEDICATED_CPU);
865         set_cpu_online(smp_processor_id(), true);
866         inc_irq_stat(CPU_RST);
867         local_irq_enable();
868         cpu_startup_entry(CPUHP_AP_ONLINE_IDLE);
869 }
870
871 /*
872  *      Activate a secondary processor.
873  */
874 static void __no_sanitize_address smp_start_secondary(void *cpuvoid)
875 {
876         S390_lowcore.restart_stack = (unsigned long) restart_stack;
877         S390_lowcore.restart_fn = (unsigned long) do_restart;
878         S390_lowcore.restart_data = 0;
879         S390_lowcore.restart_source = -1UL;
880         __ctl_load(S390_lowcore.cregs_save_area, 0, 15);
881         __load_psw_mask(PSW_KERNEL_BITS | PSW_MASK_DAT);
882         CALL_ON_STACK_NORETURN(smp_init_secondary, S390_lowcore.kernel_stack);
883 }
884
885 /* Upping and downing of CPUs */
886 int __cpu_up(unsigned int cpu, struct task_struct *tidle)
887 {
888         struct pcpu *pcpu = pcpu_devices + cpu;
889         int rc;
890
891         if (pcpu->state != CPU_STATE_CONFIGURED)
892                 return -EIO;
893         if (pcpu_sigp_retry(pcpu, SIGP_INITIAL_CPU_RESET, 0) !=
894             SIGP_CC_ORDER_CODE_ACCEPTED)
895                 return -EIO;
896
897         rc = pcpu_alloc_lowcore(pcpu, cpu);
898         if (rc)
899                 return rc;
900         pcpu_prepare_secondary(pcpu, cpu);
901         pcpu_attach_task(pcpu, tidle);
902         pcpu_start_fn(pcpu, smp_start_secondary, NULL);
903         /* Wait until cpu puts itself in the online & active maps */
904         while (!cpu_online(cpu))
905                 cpu_relax();
906         return 0;
907 }
908
909 static unsigned int setup_possible_cpus __initdata;
910
911 static int __init _setup_possible_cpus(char *s)
912 {
913         get_option(&s, &setup_possible_cpus);
914         return 0;
915 }
916 early_param("possible_cpus", _setup_possible_cpus);
917
918 int __cpu_disable(void)
919 {
920         unsigned long cregs[16];
921
922         /* Handle possible pending IPIs */
923         smp_handle_ext_call();
924         set_cpu_online(smp_processor_id(), false);
925         /* Disable pseudo page faults on this cpu. */
926         pfault_fini();
927         /* Disable interrupt sources via control register. */
928         __ctl_store(cregs, 0, 15);
929         cregs[0]  &= ~0x0000ee70UL;     /* disable all external interrupts */
930         cregs[6]  &= ~0xff000000UL;     /* disable all I/O interrupts */
931         cregs[14] &= ~0x1f000000UL;     /* disable most machine checks */
932         __ctl_load(cregs, 0, 15);
933         clear_cpu_flag(CIF_NOHZ_DELAY);
934         return 0;
935 }
936
937 void __cpu_die(unsigned int cpu)
938 {
939         struct pcpu *pcpu;
940
941         /* Wait until target cpu is down */
942         pcpu = pcpu_devices + cpu;
943         while (!pcpu_stopped(pcpu))
944                 cpu_relax();
945         pcpu_free_lowcore(pcpu);
946         cpumask_clear_cpu(cpu, mm_cpumask(&init_mm));
947         cpumask_clear_cpu(cpu, &init_mm.context.cpu_attach_mask);
948 }
949
950 void __noreturn cpu_die(void)
951 {
952         idle_task_exit();
953         __bpon();
954         pcpu_sigp_retry(pcpu_devices + smp_processor_id(), SIGP_STOP, 0);
955         for (;;) ;
956 }
957
958 void __init smp_fill_possible_mask(void)
959 {
960         unsigned int possible, sclp_max, cpu;
961
962         sclp_max = max(sclp.mtid, sclp.mtid_cp) + 1;
963         sclp_max = min(smp_max_threads, sclp_max);
964         sclp_max = (sclp.max_cores * sclp_max) ?: nr_cpu_ids;
965         possible = setup_possible_cpus ?: nr_cpu_ids;
966         possible = min(possible, sclp_max);
967         for (cpu = 0; cpu < possible && cpu < nr_cpu_ids; cpu++)
968                 set_cpu_possible(cpu, true);
969 }
970
971 void __init smp_prepare_cpus(unsigned int max_cpus)
972 {
973         /* request the 0x1201 emergency signal external interrupt */
974         if (register_external_irq(EXT_IRQ_EMERGENCY_SIG, do_ext_call_interrupt))
975                 panic("Couldn't request external interrupt 0x1201");
976         /* request the 0x1202 external call external interrupt */
977         if (register_external_irq(EXT_IRQ_EXTERNAL_CALL, do_ext_call_interrupt))
978                 panic("Couldn't request external interrupt 0x1202");
979 }
980
981 void __init smp_prepare_boot_cpu(void)
982 {
983         struct pcpu *pcpu = pcpu_devices;
984
985         WARN_ON(!cpu_present(0) || !cpu_online(0));
986         pcpu->state = CPU_STATE_CONFIGURED;
987         pcpu->lowcore = (struct lowcore *)(unsigned long) store_prefix();
988         S390_lowcore.percpu_offset = __per_cpu_offset[0];
989         smp_cpu_set_polarization(0, POLARIZATION_UNKNOWN);
990 }
991
992 void __init smp_cpus_done(unsigned int max_cpus)
993 {
994 }
995
996 void __init smp_setup_processor_id(void)
997 {
998         pcpu_devices[0].address = stap();
999         S390_lowcore.cpu_nr = 0;
1000         S390_lowcore.spinlock_lockval = arch_spin_lockval(0);
1001         S390_lowcore.spinlock_index = 0;
1002 }
1003
1004 /*
1005  * the frequency of the profiling timer can be changed
1006  * by writing a multiplier value into /proc/profile.
1007  *
1008  * usually you want to run this on all CPUs ;)
1009  */
1010 int setup_profiling_timer(unsigned int multiplier)
1011 {
1012         return 0;
1013 }
1014
1015 static ssize_t cpu_configure_show(struct device *dev,
1016                                   struct device_attribute *attr, char *buf)
1017 {
1018         ssize_t count;
1019
1020         mutex_lock(&smp_cpu_state_mutex);
1021         count = sprintf(buf, "%d\n", pcpu_devices[dev->id].state);
1022         mutex_unlock(&smp_cpu_state_mutex);
1023         return count;
1024 }
1025
1026 static ssize_t cpu_configure_store(struct device *dev,
1027                                    struct device_attribute *attr,
1028                                    const char *buf, size_t count)
1029 {
1030         struct pcpu *pcpu;
1031         int cpu, val, rc, i;
1032         char delim;
1033
1034         if (sscanf(buf, "%d %c", &val, &delim) != 1)
1035                 return -EINVAL;
1036         if (val != 0 && val != 1)
1037                 return -EINVAL;
1038         get_online_cpus();
1039         mutex_lock(&smp_cpu_state_mutex);
1040         rc = -EBUSY;
1041         /* disallow configuration changes of online cpus and cpu 0 */
1042         cpu = dev->id;
1043         cpu = smp_get_base_cpu(cpu);
1044         if (cpu == 0)
1045                 goto out;
1046         for (i = 0; i <= smp_cpu_mtid; i++)
1047                 if (cpu_online(cpu + i))
1048                         goto out;
1049         pcpu = pcpu_devices + cpu;
1050         rc = 0;
1051         switch (val) {
1052         case 0:
1053                 if (pcpu->state != CPU_STATE_CONFIGURED)
1054                         break;
1055                 rc = sclp_core_deconfigure(pcpu->address >> smp_cpu_mt_shift);
1056                 if (rc)
1057                         break;
1058                 for (i = 0; i <= smp_cpu_mtid; i++) {
1059                         if (cpu + i >= nr_cpu_ids || !cpu_present(cpu + i))
1060                                 continue;
1061                         pcpu[i].state = CPU_STATE_STANDBY;
1062                         smp_cpu_set_polarization(cpu + i,
1063                                                  POLARIZATION_UNKNOWN);
1064                 }
1065                 topology_expect_change();
1066                 break;
1067         case 1:
1068                 if (pcpu->state != CPU_STATE_STANDBY)
1069                         break;
1070                 rc = sclp_core_configure(pcpu->address >> smp_cpu_mt_shift);
1071                 if (rc)
1072                         break;
1073                 for (i = 0; i <= smp_cpu_mtid; i++) {
1074                         if (cpu + i >= nr_cpu_ids || !cpu_present(cpu + i))
1075                                 continue;
1076                         pcpu[i].state = CPU_STATE_CONFIGURED;
1077                         smp_cpu_set_polarization(cpu + i,
1078                                                  POLARIZATION_UNKNOWN);
1079                 }
1080                 topology_expect_change();
1081                 break;
1082         default:
1083                 break;
1084         }
1085 out:
1086         mutex_unlock(&smp_cpu_state_mutex);
1087         put_online_cpus();
1088         return rc ? rc : count;
1089 }
1090 static DEVICE_ATTR(configure, 0644, cpu_configure_show, cpu_configure_store);
1091
1092 static ssize_t show_cpu_address(struct device *dev,
1093                                 struct device_attribute *attr, char *buf)
1094 {
1095         return sprintf(buf, "%d\n", pcpu_devices[dev->id].address);
1096 }
1097 static DEVICE_ATTR(address, 0444, show_cpu_address, NULL);
1098
1099 static struct attribute *cpu_common_attrs[] = {
1100         &dev_attr_configure.attr,
1101         &dev_attr_address.attr,
1102         NULL,
1103 };
1104
1105 static struct attribute_group cpu_common_attr_group = {
1106         .attrs = cpu_common_attrs,
1107 };
1108
1109 static struct attribute *cpu_online_attrs[] = {
1110         &dev_attr_idle_count.attr,
1111         &dev_attr_idle_time_us.attr,
1112         NULL,
1113 };
1114
1115 static struct attribute_group cpu_online_attr_group = {
1116         .attrs = cpu_online_attrs,
1117 };
1118
1119 static int smp_cpu_online(unsigned int cpu)
1120 {
1121         struct device *s = &per_cpu(cpu_device, cpu)->dev;
1122
1123         return sysfs_create_group(&s->kobj, &cpu_online_attr_group);
1124 }
1125 static int smp_cpu_pre_down(unsigned int cpu)
1126 {
1127         struct device *s = &per_cpu(cpu_device, cpu)->dev;
1128
1129         sysfs_remove_group(&s->kobj, &cpu_online_attr_group);
1130         return 0;
1131 }
1132
1133 static int smp_add_present_cpu(int cpu)
1134 {
1135         struct device *s;
1136         struct cpu *c;
1137         int rc;
1138
1139         c = kzalloc(sizeof(*c), GFP_KERNEL);
1140         if (!c)
1141                 return -ENOMEM;
1142         per_cpu(cpu_device, cpu) = c;
1143         s = &c->dev;
1144         c->hotpluggable = 1;
1145         rc = register_cpu(c, cpu);
1146         if (rc)
1147                 goto out;
1148         rc = sysfs_create_group(&s->kobj, &cpu_common_attr_group);
1149         if (rc)
1150                 goto out_cpu;
1151         rc = topology_cpu_init(c);
1152         if (rc)
1153                 goto out_topology;
1154         return 0;
1155
1156 out_topology:
1157         sysfs_remove_group(&s->kobj, &cpu_common_attr_group);
1158 out_cpu:
1159         unregister_cpu(c);
1160 out:
1161         return rc;
1162 }
1163
1164 int __ref smp_rescan_cpus(void)
1165 {
1166         struct sclp_core_info *info;
1167         int nr;
1168
1169         info = kzalloc(sizeof(*info), GFP_KERNEL);
1170         if (!info)
1171                 return -ENOMEM;
1172         smp_get_core_info(info, 0);
1173         get_online_cpus();
1174         mutex_lock(&smp_cpu_state_mutex);
1175         nr = __smp_rescan_cpus(info, false);
1176         mutex_unlock(&smp_cpu_state_mutex);
1177         put_online_cpus();
1178         kfree(info);
1179         if (nr)
1180                 topology_schedule_update();
1181         return 0;
1182 }
1183
1184 static ssize_t __ref rescan_store(struct device *dev,
1185                                   struct device_attribute *attr,
1186                                   const char *buf,
1187                                   size_t count)
1188 {
1189         int rc;
1190
1191         rc = lock_device_hotplug_sysfs();
1192         if (rc)
1193                 return rc;
1194         rc = smp_rescan_cpus();
1195         unlock_device_hotplug();
1196         return rc ? rc : count;
1197 }
1198 static DEVICE_ATTR_WO(rescan);
1199
1200 static int __init s390_smp_init(void)
1201 {
1202         int cpu, rc = 0;
1203
1204         rc = device_create_file(cpu_subsys.dev_root, &dev_attr_rescan);
1205         if (rc)
1206                 return rc;
1207         for_each_present_cpu(cpu) {
1208                 rc = smp_add_present_cpu(cpu);
1209                 if (rc)
1210                         goto out;
1211         }
1212
1213         rc = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "s390/smp:online",
1214                                smp_cpu_online, smp_cpu_pre_down);
1215         rc = rc <= 0 ? rc : 0;
1216 out:
1217         return rc;
1218 }
1219 subsys_initcall(s390_smp_init);