GNU Linux-libre 4.14.290-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/bootmem.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/kmemleak.h>
31 #include <linux/delay.h>
32 #include <linux/interrupt.h>
33 #include <linux/irqflags.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/memblock.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/vdso.h>
52 #include <asm/debug.h>
53 #include <asm/os_info.h>
54 #include <asm/sigp.h>
55 #include <asm/idle.h>
56 #include <asm/nmi.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 static struct kmem_cache *pcpu_mcesa_cache;
86
87 unsigned int smp_cpu_mt_shift;
88 EXPORT_SYMBOL(smp_cpu_mt_shift);
89
90 unsigned int smp_cpu_mtid;
91 EXPORT_SYMBOL(smp_cpu_mtid);
92
93 #ifdef CONFIG_CRASH_DUMP
94 __vector128 __initdata boot_cpu_vector_save_area[__NUM_VXRS];
95 #endif
96
97 static unsigned int smp_max_threads __initdata = -1U;
98
99 static int __init early_nosmt(char *s)
100 {
101         smp_max_threads = 1;
102         return 0;
103 }
104 early_param("nosmt", early_nosmt);
105
106 static int __init early_smt(char *s)
107 {
108         get_option(&s, &smp_max_threads);
109         return 0;
110 }
111 early_param("smt", early_smt);
112
113 /*
114  * The smp_cpu_state_mutex must be held when changing the state or polarization
115  * member of a pcpu data structure within the pcpu_devices arreay.
116  */
117 DEFINE_MUTEX(smp_cpu_state_mutex);
118
119 /*
120  * Signal processor helper functions.
121  */
122 static inline int __pcpu_sigp_relax(u16 addr, u8 order, unsigned long parm)
123 {
124         int cc;
125
126         while (1) {
127                 cc = __pcpu_sigp(addr, order, parm, NULL);
128                 if (cc != SIGP_CC_BUSY)
129                         return cc;
130                 cpu_relax();
131         }
132 }
133
134 static int pcpu_sigp_retry(struct pcpu *pcpu, u8 order, u32 parm)
135 {
136         int cc, retry;
137
138         for (retry = 0; ; retry++) {
139                 cc = __pcpu_sigp(pcpu->address, order, parm, NULL);
140                 if (cc != SIGP_CC_BUSY)
141                         break;
142                 if (retry >= 3)
143                         udelay(10);
144         }
145         return cc;
146 }
147
148 static inline int pcpu_stopped(struct pcpu *pcpu)
149 {
150         u32 uninitialized_var(status);
151
152         if (__pcpu_sigp(pcpu->address, SIGP_SENSE,
153                         0, &status) != SIGP_CC_STATUS_STORED)
154                 return 0;
155         return !!(status & (SIGP_STATUS_CHECK_STOP|SIGP_STATUS_STOPPED));
156 }
157
158 static inline int pcpu_running(struct pcpu *pcpu)
159 {
160         if (__pcpu_sigp(pcpu->address, SIGP_SENSE_RUNNING,
161                         0, NULL) != SIGP_CC_STATUS_STORED)
162                 return 1;
163         /* Status stored condition code is equivalent to cpu not running. */
164         return 0;
165 }
166
167 /*
168  * Find struct pcpu by cpu address.
169  */
170 static struct pcpu *pcpu_find_address(const struct cpumask *mask, u16 address)
171 {
172         int cpu;
173
174         for_each_cpu(cpu, mask)
175                 if (pcpu_devices[cpu].address == address)
176                         return pcpu_devices + cpu;
177         return NULL;
178 }
179
180 static void pcpu_ec_call(struct pcpu *pcpu, int ec_bit)
181 {
182         int order;
183
184         if (test_and_set_bit(ec_bit, &pcpu->ec_mask))
185                 return;
186         order = pcpu_running(pcpu) ? SIGP_EXTERNAL_CALL : SIGP_EMERGENCY_SIGNAL;
187         pcpu->ec_clk = get_tod_clock_fast();
188         pcpu_sigp_retry(pcpu, order, 0);
189 }
190
191 #define ASYNC_FRAME_OFFSET (ASYNC_SIZE - STACK_FRAME_OVERHEAD - __PT_SIZE)
192 #define PANIC_FRAME_OFFSET (PAGE_SIZE - STACK_FRAME_OVERHEAD - __PT_SIZE)
193
194 static int pcpu_alloc_lowcore(struct pcpu *pcpu, int cpu)
195 {
196         unsigned long async_stack, panic_stack;
197         unsigned long mcesa_origin, mcesa_bits;
198         struct lowcore *lc;
199
200         mcesa_origin = mcesa_bits = 0;
201         if (pcpu != &pcpu_devices[0]) {
202                 pcpu->lowcore = (struct lowcore *)
203                         __get_free_pages(GFP_KERNEL | GFP_DMA, LC_ORDER);
204                 async_stack = __get_free_pages(GFP_KERNEL, ASYNC_ORDER);
205                 panic_stack = __get_free_page(GFP_KERNEL);
206                 if (!pcpu->lowcore || !panic_stack || !async_stack)
207                         goto out;
208                 if (MACHINE_HAS_VX || MACHINE_HAS_GS) {
209                         mcesa_origin = (unsigned long)
210                                 kmem_cache_alloc(pcpu_mcesa_cache, GFP_KERNEL);
211                         if (!mcesa_origin)
212                                 goto out;
213                         /* The pointer is stored with mcesa_bits ORed in */
214                         kmemleak_not_leak((void *) mcesa_origin);
215                         mcesa_bits = MACHINE_HAS_GS ? 11 : 0;
216                 }
217         } else {
218                 async_stack = pcpu->lowcore->async_stack - ASYNC_FRAME_OFFSET;
219                 panic_stack = pcpu->lowcore->panic_stack - PANIC_FRAME_OFFSET;
220                 mcesa_origin = pcpu->lowcore->mcesad & MCESA_ORIGIN_MASK;
221                 mcesa_bits = pcpu->lowcore->mcesad & MCESA_LC_MASK;
222         }
223         lc = pcpu->lowcore;
224         memcpy(lc, &S390_lowcore, 512);
225         memset((char *) lc + 512, 0, sizeof(*lc) - 512);
226         lc->async_stack = async_stack + ASYNC_FRAME_OFFSET;
227         lc->panic_stack = panic_stack + PANIC_FRAME_OFFSET;
228         lc->mcesad = mcesa_origin | mcesa_bits;
229         lc->cpu_nr = cpu;
230         lc->spinlock_lockval = arch_spin_lockval(cpu);
231         lc->br_r1_trampoline = 0x07f1;  /* br %r1 */
232         if (vdso_alloc_per_cpu(lc))
233                 goto out;
234         lowcore_ptr[cpu] = lc;
235         pcpu_sigp_retry(pcpu, SIGP_SET_PREFIX, (u32)(unsigned long) lc);
236         return 0;
237 out:
238         if (pcpu != &pcpu_devices[0]) {
239                 if (mcesa_origin)
240                         kmem_cache_free(pcpu_mcesa_cache,
241                                         (void *) mcesa_origin);
242                 free_page(panic_stack);
243                 free_pages(async_stack, ASYNC_ORDER);
244                 free_pages((unsigned long) pcpu->lowcore, LC_ORDER);
245         }
246         return -ENOMEM;
247 }
248
249 #ifdef CONFIG_HOTPLUG_CPU
250
251 static void pcpu_free_lowcore(struct pcpu *pcpu)
252 {
253         unsigned long mcesa_origin;
254
255         pcpu_sigp_retry(pcpu, SIGP_SET_PREFIX, 0);
256         lowcore_ptr[pcpu - pcpu_devices] = NULL;
257         vdso_free_per_cpu(pcpu->lowcore);
258         if (pcpu == &pcpu_devices[0])
259                 return;
260         if (MACHINE_HAS_VX || MACHINE_HAS_GS) {
261                 mcesa_origin = pcpu->lowcore->mcesad & MCESA_ORIGIN_MASK;
262                 kmem_cache_free(pcpu_mcesa_cache, (void *) mcesa_origin);
263         }
264         free_page(pcpu->lowcore->panic_stack-PANIC_FRAME_OFFSET);
265         free_pages(pcpu->lowcore->async_stack-ASYNC_FRAME_OFFSET, ASYNC_ORDER);
266         free_pages((unsigned long) pcpu->lowcore, LC_ORDER);
267 }
268
269 #endif /* CONFIG_HOTPLUG_CPU */
270
271 static void pcpu_prepare_secondary(struct pcpu *pcpu, int cpu)
272 {
273         struct lowcore *lc = pcpu->lowcore;
274
275         cpumask_set_cpu(cpu, &init_mm.context.cpu_attach_mask);
276         cpumask_set_cpu(cpu, mm_cpumask(&init_mm));
277         lc->cpu_nr = cpu;
278         lc->spinlock_lockval = arch_spin_lockval(cpu);
279         lc->percpu_offset = __per_cpu_offset[cpu];
280         lc->kernel_asce = S390_lowcore.kernel_asce;
281         lc->machine_flags = S390_lowcore.machine_flags;
282         lc->user_timer = lc->system_timer = lc->steal_timer = 0;
283         __ctl_store(lc->cregs_save_area, 0, 15);
284         save_access_regs((unsigned int *) lc->access_regs_save_area);
285         memcpy(lc->stfle_fac_list, S390_lowcore.stfle_fac_list,
286                sizeof(lc->stfle_fac_list));
287         memcpy(lc->alt_stfle_fac_list, S390_lowcore.alt_stfle_fac_list,
288                sizeof(lc->alt_stfle_fac_list));
289 }
290
291 static void pcpu_attach_task(struct pcpu *pcpu, struct task_struct *tsk)
292 {
293         struct lowcore *lc = pcpu->lowcore;
294
295         lc->kernel_stack = (unsigned long) task_stack_page(tsk)
296                 + THREAD_SIZE - STACK_FRAME_OVERHEAD - sizeof(struct pt_regs);
297         lc->current_task = (unsigned long) tsk;
298         lc->lpp = LPP_MAGIC;
299         lc->current_pid = tsk->pid;
300         lc->user_timer = tsk->thread.user_timer;
301         lc->guest_timer = tsk->thread.guest_timer;
302         lc->system_timer = tsk->thread.system_timer;
303         lc->hardirq_timer = tsk->thread.hardirq_timer;
304         lc->softirq_timer = tsk->thread.softirq_timer;
305         lc->steal_timer = 0;
306 }
307
308 static void pcpu_start_fn(struct pcpu *pcpu, void (*func)(void *), void *data)
309 {
310         struct lowcore *lc = pcpu->lowcore;
311
312         lc->restart_stack = lc->kernel_stack;
313         lc->restart_fn = (unsigned long) func;
314         lc->restart_data = (unsigned long) data;
315         lc->restart_source = -1UL;
316         pcpu_sigp_retry(pcpu, SIGP_RESTART, 0);
317 }
318
319 /*
320  * Call function via PSW restart on pcpu and stop the current cpu.
321  */
322 static void pcpu_delegate(struct pcpu *pcpu, void (*func)(void *),
323                           void *data, unsigned long stack)
324 {
325         struct lowcore *lc = lowcore_ptr[pcpu - pcpu_devices];
326         unsigned long source_cpu = stap();
327
328         __load_psw_mask(PSW_KERNEL_BITS);
329         if (pcpu->address == source_cpu)
330                 func(data);     /* should not return */
331         /* Stop target cpu (if func returns this stops the current cpu). */
332         pcpu_sigp_retry(pcpu, SIGP_STOP, 0);
333         /* Restart func on the target cpu and stop the current cpu. */
334         mem_assign_absolute(lc->restart_stack, stack);
335         mem_assign_absolute(lc->restart_fn, (unsigned long) func);
336         mem_assign_absolute(lc->restart_data, (unsigned long) data);
337         mem_assign_absolute(lc->restart_source, source_cpu);
338         __bpon();
339         asm volatile(
340                 "0:     sigp    0,%0,%2 # sigp restart to target cpu\n"
341                 "       brc     2,0b    # busy, try again\n"
342                 "1:     sigp    0,%1,%3 # sigp stop to current cpu\n"
343                 "       brc     2,1b    # busy, try again\n"
344                 : : "d" (pcpu->address), "d" (source_cpu),
345                     "K" (SIGP_RESTART), "K" (SIGP_STOP)
346                 : "0", "1", "cc");
347         for (;;) ;
348 }
349
350 /*
351  * Enable additional logical cpus for multi-threading.
352  */
353 static int pcpu_set_smt(unsigned int mtid)
354 {
355         int cc;
356
357         if (smp_cpu_mtid == mtid)
358                 return 0;
359         cc = __pcpu_sigp(0, SIGP_SET_MULTI_THREADING, mtid, NULL);
360         if (cc == 0) {
361                 smp_cpu_mtid = mtid;
362                 smp_cpu_mt_shift = 0;
363                 while (smp_cpu_mtid >= (1U << smp_cpu_mt_shift))
364                         smp_cpu_mt_shift++;
365                 pcpu_devices[0].address = stap();
366         }
367         return cc;
368 }
369
370 /*
371  * Call function on an online CPU.
372  */
373 void smp_call_online_cpu(void (*func)(void *), void *data)
374 {
375         struct pcpu *pcpu;
376
377         /* Use the current cpu if it is online. */
378         pcpu = pcpu_find_address(cpu_online_mask, stap());
379         if (!pcpu)
380                 /* Use the first online cpu. */
381                 pcpu = pcpu_devices + cpumask_first(cpu_online_mask);
382         pcpu_delegate(pcpu, func, data, (unsigned long) restart_stack);
383 }
384
385 /*
386  * Call function on the ipl CPU.
387  */
388 void smp_call_ipl_cpu(void (*func)(void *), void *data)
389 {
390         struct lowcore *lc = pcpu_devices->lowcore;
391
392         if (pcpu_devices[0].address == stap())
393                 lc = &S390_lowcore;
394
395         pcpu_delegate(&pcpu_devices[0], func, data,
396                       lc->panic_stack - PANIC_FRAME_OFFSET + PAGE_SIZE);
397 }
398
399 int smp_find_processor_id(u16 address)
400 {
401         int cpu;
402
403         for_each_present_cpu(cpu)
404                 if (pcpu_devices[cpu].address == address)
405                         return cpu;
406         return -1;
407 }
408
409 bool notrace arch_vcpu_is_preempted(int cpu)
410 {
411         if (test_cpu_flag_of(CIF_ENABLED_WAIT, cpu))
412                 return false;
413         if (pcpu_running(pcpu_devices + cpu))
414                 return false;
415         return true;
416 }
417 EXPORT_SYMBOL(arch_vcpu_is_preempted);
418
419 void notrace smp_yield_cpu(int cpu)
420 {
421         if (MACHINE_HAS_DIAG9C) {
422                 diag_stat_inc_norecursion(DIAG_STAT_X09C);
423                 asm volatile("diag %0,0,0x9c"
424                              : : "d" (pcpu_devices[cpu].address));
425         } else if (MACHINE_HAS_DIAG44) {
426                 diag_stat_inc_norecursion(DIAG_STAT_X044);
427                 asm volatile("diag 0,0,0x44");
428         }
429 }
430
431 /*
432  * Send cpus emergency shutdown signal. This gives the cpus the
433  * opportunity to complete outstanding interrupts.
434  */
435 static void smp_emergency_stop(cpumask_t *cpumask)
436 {
437         u64 end;
438         int cpu;
439
440         end = get_tod_clock() + (1000000UL << 12);
441         for_each_cpu(cpu, cpumask) {
442                 struct pcpu *pcpu = pcpu_devices + cpu;
443                 set_bit(ec_stop_cpu, &pcpu->ec_mask);
444                 while (__pcpu_sigp(pcpu->address, SIGP_EMERGENCY_SIGNAL,
445                                    0, NULL) == SIGP_CC_BUSY &&
446                        get_tod_clock() < end)
447                         cpu_relax();
448         }
449         while (get_tod_clock() < end) {
450                 for_each_cpu(cpu, cpumask)
451                         if (pcpu_stopped(pcpu_devices + cpu))
452                                 cpumask_clear_cpu(cpu, cpumask);
453                 if (cpumask_empty(cpumask))
454                         break;
455                 cpu_relax();
456         }
457 }
458
459 /*
460  * Stop all cpus but the current one.
461  */
462 void smp_send_stop(void)
463 {
464         cpumask_t cpumask;
465         int cpu;
466
467         /* Disable all interrupts/machine checks */
468         __load_psw_mask(PSW_KERNEL_BITS | PSW_MASK_DAT);
469         trace_hardirqs_off();
470
471         debug_set_critical();
472         cpumask_copy(&cpumask, cpu_online_mask);
473         cpumask_clear_cpu(smp_processor_id(), &cpumask);
474
475         if (oops_in_progress)
476                 smp_emergency_stop(&cpumask);
477
478         /* stop all processors */
479         for_each_cpu(cpu, &cpumask) {
480                 struct pcpu *pcpu = pcpu_devices + cpu;
481                 pcpu_sigp_retry(pcpu, SIGP_STOP, 0);
482                 while (!pcpu_stopped(pcpu))
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_alloc_base(PAGE_SIZE, PAGE_SIZE, 1UL << 31);
663         /* Set multi-threading state to the previous system. */
664         pcpu_set_smt(sclp.mtid_prev);
665         boot_cpu_addr = stap();
666         max_cpu_addr = SCLP_MAX_CORES << sclp.mtid_prev;
667         for (addr = 0; addr <= max_cpu_addr; addr++) {
668                 if (__pcpu_sigp_relax(addr, SIGP_SENSE, 0) ==
669                     SIGP_CC_NOT_OPERATIONAL)
670                         continue;
671                 is_boot_cpu = (addr == boot_cpu_addr);
672                 /* Allocate save area */
673                 sa = save_area_alloc(is_boot_cpu);
674                 if (!sa)
675                         panic("could not allocate memory for save area\n");
676                 if (MACHINE_HAS_VX)
677                         /* Get the vector registers */
678                         smp_save_cpu_vxrs(sa, addr, is_boot_cpu, page);
679                 /*
680                  * For a zfcp dump OLDMEM_BASE == NULL and the registers
681                  * of the boot CPU are stored in the HSA. To retrieve
682                  * these registers an SCLP request is required which is
683                  * done by drivers/s390/char/zcore.c:init_cpu_info()
684                  */
685                 if (!is_boot_cpu || OLDMEM_BASE)
686                         /* Get the CPU registers */
687                         smp_save_cpu_regs(sa, addr, is_boot_cpu, page);
688         }
689         memblock_free(page, PAGE_SIZE);
690         diag308_reset();
691         pcpu_set_smt(0);
692 }
693 #endif /* CONFIG_CRASH_DUMP */
694
695 void smp_cpu_set_polarization(int cpu, int val)
696 {
697         pcpu_devices[cpu].polarization = val;
698 }
699
700 int smp_cpu_get_polarization(int cpu)
701 {
702         return pcpu_devices[cpu].polarization;
703 }
704
705 static void __ref smp_get_core_info(struct sclp_core_info *info, int early)
706 {
707         static int use_sigp_detection;
708         int address;
709
710         if (use_sigp_detection || sclp_get_core_info(info, early)) {
711                 use_sigp_detection = 1;
712                 for (address = 0;
713                      address < (SCLP_MAX_CORES << smp_cpu_mt_shift);
714                      address += (1U << smp_cpu_mt_shift)) {
715                         if (__pcpu_sigp_relax(address, SIGP_SENSE, 0) ==
716                             SIGP_CC_NOT_OPERATIONAL)
717                                 continue;
718                         info->core[info->configured].core_id =
719                                 address >> smp_cpu_mt_shift;
720                         info->configured++;
721                 }
722                 info->combined = info->configured;
723         }
724 }
725
726 static int smp_add_present_cpu(int cpu);
727
728 static int smp_add_core(struct sclp_core_entry *core, cpumask_t *avail,
729                         bool configured, bool early)
730 {
731         struct pcpu *pcpu;
732         int cpu, nr, i;
733         u16 address;
734
735         nr = 0;
736         if (sclp.has_core_type && core->type != boot_core_type)
737                 return nr;
738         cpu = cpumask_first(avail);
739         address = core->core_id << smp_cpu_mt_shift;
740         for (i = 0; (i <= smp_cpu_mtid) && (cpu < nr_cpu_ids); i++) {
741                 if (pcpu_find_address(cpu_present_mask, address + i))
742                         continue;
743                 pcpu = pcpu_devices + cpu;
744                 pcpu->address = address + i;
745                 if (configured)
746                         pcpu->state = CPU_STATE_CONFIGURED;
747                 else
748                         pcpu->state = CPU_STATE_STANDBY;
749                 smp_cpu_set_polarization(cpu, POLARIZATION_UNKNOWN);
750                 set_cpu_present(cpu, true);
751                 if (!early && smp_add_present_cpu(cpu) != 0)
752                         set_cpu_present(cpu, false);
753                 else
754                         nr++;
755                 cpumask_clear_cpu(cpu, avail);
756                 cpu = cpumask_next(cpu, avail);
757         }
758         return nr;
759 }
760
761 static int __smp_rescan_cpus(struct sclp_core_info *info, bool early)
762 {
763         struct sclp_core_entry *core;
764         static cpumask_t avail;
765         bool configured;
766         u16 core_id;
767         int nr, i;
768
769         nr = 0;
770         cpumask_xor(&avail, cpu_possible_mask, cpu_present_mask);
771         /*
772          * Add IPL core first (which got logical CPU number 0) to make sure
773          * that all SMT threads get subsequent logical CPU numbers.
774          */
775         if (early) {
776                 core_id = pcpu_devices[0].address >> smp_cpu_mt_shift;
777                 for (i = 0; i < info->configured; i++) {
778                         core = &info->core[i];
779                         if (core->core_id == core_id) {
780                                 nr += smp_add_core(core, &avail, true, early);
781                                 break;
782                         }
783                 }
784         }
785         for (i = 0; i < info->combined; i++) {
786                 configured = i < info->configured;
787                 nr += smp_add_core(&info->core[i], &avail, configured, early);
788         }
789         return nr;
790 }
791
792 void __init smp_detect_cpus(void)
793 {
794         unsigned int cpu, mtid, c_cpus, s_cpus;
795         struct sclp_core_info *info;
796         u16 address;
797
798         /* Get CPU information */
799         info = memblock_virt_alloc(sizeof(*info), 8);
800         smp_get_core_info(info, 1);
801         /* Find boot CPU type */
802         if (sclp.has_core_type) {
803                 address = stap();
804                 for (cpu = 0; cpu < info->combined; cpu++)
805                         if (info->core[cpu].core_id == address) {
806                                 /* The boot cpu dictates the cpu type. */
807                                 boot_core_type = info->core[cpu].type;
808                                 break;
809                         }
810                 if (cpu >= info->combined)
811                         panic("Could not find boot CPU type");
812         }
813
814         /* Set multi-threading state for the current system */
815         mtid = boot_core_type ? sclp.mtid : sclp.mtid_cp;
816         mtid = (mtid < smp_max_threads) ? mtid : smp_max_threads - 1;
817         pcpu_set_smt(mtid);
818
819         /* Print number of CPUs */
820         c_cpus = s_cpus = 0;
821         for (cpu = 0; cpu < info->combined; cpu++) {
822                 if (sclp.has_core_type &&
823                     info->core[cpu].type != boot_core_type)
824                         continue;
825                 if (cpu < info->configured)
826                         c_cpus += smp_cpu_mtid + 1;
827                 else
828                         s_cpus += smp_cpu_mtid + 1;
829         }
830         pr_info("%d configured CPUs, %d standby CPUs\n", c_cpus, s_cpus);
831
832         /* Add CPUs present at boot */
833         get_online_cpus();
834         __smp_rescan_cpus(info, true);
835         put_online_cpus();
836         memblock_free_early((unsigned long)info, sizeof(*info));
837 }
838
839 /*
840  *      Activate a secondary processor.
841  */
842 static void smp_start_secondary(void *cpuvoid)
843 {
844         S390_lowcore.last_update_clock = get_tod_clock();
845         S390_lowcore.restart_stack = (unsigned long) restart_stack;
846         S390_lowcore.restart_fn = (unsigned long) do_restart;
847         S390_lowcore.restart_data = 0;
848         S390_lowcore.restart_source = -1UL;
849         restore_access_regs(S390_lowcore.access_regs_save_area);
850         __ctl_load(S390_lowcore.cregs_save_area, 0, 15);
851         __load_psw_mask(PSW_KERNEL_BITS | PSW_MASK_DAT);
852         cpu_init();
853         preempt_disable();
854         init_cpu_timer();
855         vtime_init();
856         pfault_init();
857         notify_cpu_starting(smp_processor_id());
858         set_cpu_online(smp_processor_id(), true);
859         inc_irq_stat(CPU_RST);
860         local_irq_enable();
861         cpu_startup_entry(CPUHP_AP_ONLINE_IDLE);
862 }
863
864 /* Upping and downing of CPUs */
865 int __cpu_up(unsigned int cpu, struct task_struct *tidle)
866 {
867         struct pcpu *pcpu = pcpu_devices + cpu;
868         int rc;
869
870         if (pcpu->state != CPU_STATE_CONFIGURED)
871                 return -EIO;
872         if (pcpu_sigp_retry(pcpu, SIGP_INITIAL_CPU_RESET, 0) !=
873             SIGP_CC_ORDER_CODE_ACCEPTED)
874                 return -EIO;
875
876         rc = pcpu_alloc_lowcore(pcpu, cpu);
877         if (rc)
878                 return rc;
879         pcpu_prepare_secondary(pcpu, cpu);
880         pcpu_attach_task(pcpu, tidle);
881         pcpu_start_fn(pcpu, smp_start_secondary, NULL);
882         /* Wait until cpu puts itself in the online & active maps */
883         while (!cpu_online(cpu))
884                 cpu_relax();
885         return 0;
886 }
887
888 static unsigned int setup_possible_cpus __initdata;
889
890 static int __init _setup_possible_cpus(char *s)
891 {
892         get_option(&s, &setup_possible_cpus);
893         return 0;
894 }
895 early_param("possible_cpus", _setup_possible_cpus);
896
897 #ifdef CONFIG_HOTPLUG_CPU
898
899 int __cpu_disable(void)
900 {
901         unsigned long cregs[16];
902
903         /* Handle possible pending IPIs */
904         smp_handle_ext_call();
905         set_cpu_online(smp_processor_id(), false);
906         /* Disable pseudo page faults on this cpu. */
907         pfault_fini();
908         /* Disable interrupt sources via control register. */
909         __ctl_store(cregs, 0, 15);
910         cregs[0]  &= ~0x0000ee70UL;     /* disable all external interrupts */
911         cregs[6]  &= ~0xff000000UL;     /* disable all I/O interrupts */
912         cregs[14] &= ~0x1f000000UL;     /* disable most machine checks */
913         __ctl_load(cregs, 0, 15);
914         clear_cpu_flag(CIF_NOHZ_DELAY);
915         return 0;
916 }
917
918 void __cpu_die(unsigned int cpu)
919 {
920         struct pcpu *pcpu;
921
922         /* Wait until target cpu is down */
923         pcpu = pcpu_devices + cpu;
924         while (!pcpu_stopped(pcpu))
925                 cpu_relax();
926         pcpu_free_lowcore(pcpu);
927         cpumask_clear_cpu(cpu, mm_cpumask(&init_mm));
928         cpumask_clear_cpu(cpu, &init_mm.context.cpu_attach_mask);
929 }
930
931 void __noreturn cpu_die(void)
932 {
933         idle_task_exit();
934         __bpon();
935         pcpu_sigp_retry(pcpu_devices + smp_processor_id(), SIGP_STOP, 0);
936         for (;;) ;
937 }
938
939 #endif /* CONFIG_HOTPLUG_CPU */
940
941 void __init smp_fill_possible_mask(void)
942 {
943         unsigned int possible, sclp_max, cpu;
944
945         sclp_max = max(sclp.mtid, sclp.mtid_cp) + 1;
946         sclp_max = min(smp_max_threads, sclp_max);
947         sclp_max = (sclp.max_cores * sclp_max) ?: nr_cpu_ids;
948         possible = setup_possible_cpus ?: nr_cpu_ids;
949         possible = min(possible, sclp_max);
950         for (cpu = 0; cpu < possible && cpu < nr_cpu_ids; cpu++)
951                 set_cpu_possible(cpu, true);
952 }
953
954 void __init smp_prepare_cpus(unsigned int max_cpus)
955 {
956         unsigned long size;
957
958         /* request the 0x1201 emergency signal external interrupt */
959         if (register_external_irq(EXT_IRQ_EMERGENCY_SIG, do_ext_call_interrupt))
960                 panic("Couldn't request external interrupt 0x1201");
961         /* request the 0x1202 external call external interrupt */
962         if (register_external_irq(EXT_IRQ_EXTERNAL_CALL, do_ext_call_interrupt))
963                 panic("Couldn't request external interrupt 0x1202");
964         /* create slab cache for the machine-check-extended-save-areas */
965         if (MACHINE_HAS_VX || MACHINE_HAS_GS) {
966                 size = 1UL << (MACHINE_HAS_GS ? 11 : 10);
967                 pcpu_mcesa_cache = kmem_cache_create("nmi_save_areas",
968                                                      size, size, 0, NULL);
969                 if (!pcpu_mcesa_cache)
970                         panic("Couldn't create nmi save area cache");
971         }
972 }
973
974 void __init smp_prepare_boot_cpu(void)
975 {
976         struct pcpu *pcpu = pcpu_devices;
977
978         WARN_ON(!cpu_present(0) || !cpu_online(0));
979         pcpu->state = CPU_STATE_CONFIGURED;
980         pcpu->lowcore = (struct lowcore *)(unsigned long) store_prefix();
981         S390_lowcore.percpu_offset = __per_cpu_offset[0];
982         smp_cpu_set_polarization(0, POLARIZATION_UNKNOWN);
983 }
984
985 void __init smp_cpus_done(unsigned int max_cpus)
986 {
987 }
988
989 void __init smp_setup_processor_id(void)
990 {
991         pcpu_devices[0].address = stap();
992         S390_lowcore.cpu_nr = 0;
993         S390_lowcore.spinlock_lockval = arch_spin_lockval(0);
994 }
995
996 /*
997  * the frequency of the profiling timer can be changed
998  * by writing a multiplier value into /proc/profile.
999  *
1000  * usually you want to run this on all CPUs ;)
1001  */
1002 int setup_profiling_timer(unsigned int multiplier)
1003 {
1004         return 0;
1005 }
1006
1007 #ifdef CONFIG_HOTPLUG_CPU
1008 static ssize_t cpu_configure_show(struct device *dev,
1009                                   struct device_attribute *attr, char *buf)
1010 {
1011         ssize_t count;
1012
1013         mutex_lock(&smp_cpu_state_mutex);
1014         count = sprintf(buf, "%d\n", pcpu_devices[dev->id].state);
1015         mutex_unlock(&smp_cpu_state_mutex);
1016         return count;
1017 }
1018
1019 static ssize_t cpu_configure_store(struct device *dev,
1020                                    struct device_attribute *attr,
1021                                    const char *buf, size_t count)
1022 {
1023         struct pcpu *pcpu;
1024         int cpu, val, rc, i;
1025         char delim;
1026
1027         if (sscanf(buf, "%d %c", &val, &delim) != 1)
1028                 return -EINVAL;
1029         if (val != 0 && val != 1)
1030                 return -EINVAL;
1031         get_online_cpus();
1032         mutex_lock(&smp_cpu_state_mutex);
1033         rc = -EBUSY;
1034         /* disallow configuration changes of online cpus and cpu 0 */
1035         cpu = dev->id;
1036         cpu = smp_get_base_cpu(cpu);
1037         if (cpu == 0)
1038                 goto out;
1039         for (i = 0; i <= smp_cpu_mtid; i++)
1040                 if (cpu_online(cpu + i))
1041                         goto out;
1042         pcpu = pcpu_devices + cpu;
1043         rc = 0;
1044         switch (val) {
1045         case 0:
1046                 if (pcpu->state != CPU_STATE_CONFIGURED)
1047                         break;
1048                 rc = sclp_core_deconfigure(pcpu->address >> smp_cpu_mt_shift);
1049                 if (rc)
1050                         break;
1051                 for (i = 0; i <= smp_cpu_mtid; i++) {
1052                         if (cpu + i >= nr_cpu_ids || !cpu_present(cpu + i))
1053                                 continue;
1054                         pcpu[i].state = CPU_STATE_STANDBY;
1055                         smp_cpu_set_polarization(cpu + i,
1056                                                  POLARIZATION_UNKNOWN);
1057                 }
1058                 topology_expect_change();
1059                 break;
1060         case 1:
1061                 if (pcpu->state != CPU_STATE_STANDBY)
1062                         break;
1063                 rc = sclp_core_configure(pcpu->address >> smp_cpu_mt_shift);
1064                 if (rc)
1065                         break;
1066                 for (i = 0; i <= smp_cpu_mtid; i++) {
1067                         if (cpu + i >= nr_cpu_ids || !cpu_present(cpu + i))
1068                                 continue;
1069                         pcpu[i].state = CPU_STATE_CONFIGURED;
1070                         smp_cpu_set_polarization(cpu + i,
1071                                                  POLARIZATION_UNKNOWN);
1072                 }
1073                 topology_expect_change();
1074                 break;
1075         default:
1076                 break;
1077         }
1078 out:
1079         mutex_unlock(&smp_cpu_state_mutex);
1080         put_online_cpus();
1081         return rc ? rc : count;
1082 }
1083 static DEVICE_ATTR(configure, 0644, cpu_configure_show, cpu_configure_store);
1084 #endif /* CONFIG_HOTPLUG_CPU */
1085
1086 static ssize_t show_cpu_address(struct device *dev,
1087                                 struct device_attribute *attr, char *buf)
1088 {
1089         return sprintf(buf, "%d\n", pcpu_devices[dev->id].address);
1090 }
1091 static DEVICE_ATTR(address, 0444, show_cpu_address, NULL);
1092
1093 static struct attribute *cpu_common_attrs[] = {
1094 #ifdef CONFIG_HOTPLUG_CPU
1095         &dev_attr_configure.attr,
1096 #endif
1097         &dev_attr_address.attr,
1098         NULL,
1099 };
1100
1101 static struct attribute_group cpu_common_attr_group = {
1102         .attrs = cpu_common_attrs,
1103 };
1104
1105 static struct attribute *cpu_online_attrs[] = {
1106         &dev_attr_idle_count.attr,
1107         &dev_attr_idle_time_us.attr,
1108         NULL,
1109 };
1110
1111 static struct attribute_group cpu_online_attr_group = {
1112         .attrs = cpu_online_attrs,
1113 };
1114
1115 static int smp_cpu_online(unsigned int cpu)
1116 {
1117         struct device *s = &per_cpu(cpu_device, cpu)->dev;
1118
1119         return sysfs_create_group(&s->kobj, &cpu_online_attr_group);
1120 }
1121 static int smp_cpu_pre_down(unsigned int cpu)
1122 {
1123         struct device *s = &per_cpu(cpu_device, cpu)->dev;
1124
1125         sysfs_remove_group(&s->kobj, &cpu_online_attr_group);
1126         return 0;
1127 }
1128
1129 static int smp_add_present_cpu(int cpu)
1130 {
1131         struct device *s;
1132         struct cpu *c;
1133         int rc;
1134
1135         c = kzalloc(sizeof(*c), GFP_KERNEL);
1136         if (!c)
1137                 return -ENOMEM;
1138         per_cpu(cpu_device, cpu) = c;
1139         s = &c->dev;
1140         c->hotpluggable = 1;
1141         rc = register_cpu(c, cpu);
1142         if (rc)
1143                 goto out;
1144         rc = sysfs_create_group(&s->kobj, &cpu_common_attr_group);
1145         if (rc)
1146                 goto out_cpu;
1147         rc = topology_cpu_init(c);
1148         if (rc)
1149                 goto out_topology;
1150         return 0;
1151
1152 out_topology:
1153         sysfs_remove_group(&s->kobj, &cpu_common_attr_group);
1154 out_cpu:
1155 #ifdef CONFIG_HOTPLUG_CPU
1156         unregister_cpu(c);
1157 #endif
1158 out:
1159         return rc;
1160 }
1161
1162 #ifdef CONFIG_HOTPLUG_CPU
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(rescan, 0200, NULL, rescan_store);
1199 #endif /* CONFIG_HOTPLUG_CPU */
1200
1201 static int __init s390_smp_init(void)
1202 {
1203         int cpu, rc = 0;
1204
1205 #ifdef CONFIG_HOTPLUG_CPU
1206         rc = device_create_file(cpu_subsys.dev_root, &dev_attr_rescan);
1207         if (rc)
1208                 return rc;
1209 #endif
1210         for_each_present_cpu(cpu) {
1211                 rc = smp_add_present_cpu(cpu);
1212                 if (rc)
1213                         goto out;
1214         }
1215
1216         rc = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "s390/smp:online",
1217                                smp_cpu_online, smp_cpu_pre_down);
1218         rc = rc <= 0 ? rc : 0;
1219 out:
1220         return rc;
1221 }
1222 subsys_initcall(s390_smp_init);