tcp: highest_sack fix
[sfrench/cifs-2.6.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 "entry.h"
58
59 enum {
60         ec_schedule = 0,
61         ec_call_function_single,
62         ec_stop_cpu,
63 };
64
65 enum {
66         CPU_STATE_STANDBY,
67         CPU_STATE_CONFIGURED,
68 };
69
70 static DEFINE_PER_CPU(struct cpu *, cpu_device);
71
72 struct pcpu {
73         struct lowcore *lowcore;        /* lowcore page(s) for the cpu */
74         unsigned long ec_mask;          /* bit mask for ec_xxx functions */
75         unsigned long ec_clk;           /* sigp timestamp for ec_xxx */
76         signed char state;              /* physical cpu state */
77         signed char polarization;       /* physical polarization */
78         u16 address;                    /* physical cpu address */
79 };
80
81 static u8 boot_core_type;
82 static struct pcpu pcpu_devices[NR_CPUS];
83
84 static struct kmem_cache *pcpu_mcesa_cache;
85
86 unsigned int smp_cpu_mt_shift;
87 EXPORT_SYMBOL(smp_cpu_mt_shift);
88
89 unsigned int smp_cpu_mtid;
90 EXPORT_SYMBOL(smp_cpu_mtid);
91
92 #ifdef CONFIG_CRASH_DUMP
93 __vector128 __initdata boot_cpu_vector_save_area[__NUM_VXRS];
94 #endif
95
96 static unsigned int smp_max_threads __initdata = -1U;
97
98 static int __init early_nosmt(char *s)
99 {
100         smp_max_threads = 1;
101         return 0;
102 }
103 early_param("nosmt", early_nosmt);
104
105 static int __init early_smt(char *s)
106 {
107         get_option(&s, &smp_max_threads);
108         return 0;
109 }
110 early_param("smt", early_smt);
111
112 /*
113  * The smp_cpu_state_mutex must be held when changing the state or polarization
114  * member of a pcpu data structure within the pcpu_devices arreay.
115  */
116 DEFINE_MUTEX(smp_cpu_state_mutex);
117
118 /*
119  * Signal processor helper functions.
120  */
121 static inline int __pcpu_sigp_relax(u16 addr, u8 order, unsigned long parm)
122 {
123         int cc;
124
125         while (1) {
126                 cc = __pcpu_sigp(addr, order, parm, NULL);
127                 if (cc != SIGP_CC_BUSY)
128                         return cc;
129                 cpu_relax();
130         }
131 }
132
133 static int pcpu_sigp_retry(struct pcpu *pcpu, u8 order, u32 parm)
134 {
135         int cc, retry;
136
137         for (retry = 0; ; retry++) {
138                 cc = __pcpu_sigp(pcpu->address, order, parm, NULL);
139                 if (cc != SIGP_CC_BUSY)
140                         break;
141                 if (retry >= 3)
142                         udelay(10);
143         }
144         return cc;
145 }
146
147 static inline int pcpu_stopped(struct pcpu *pcpu)
148 {
149         u32 uninitialized_var(status);
150
151         if (__pcpu_sigp(pcpu->address, SIGP_SENSE,
152                         0, &status) != SIGP_CC_STATUS_STORED)
153                 return 0;
154         return !!(status & (SIGP_STATUS_CHECK_STOP|SIGP_STATUS_STOPPED));
155 }
156
157 static inline int pcpu_running(struct pcpu *pcpu)
158 {
159         if (__pcpu_sigp(pcpu->address, SIGP_SENSE_RUNNING,
160                         0, NULL) != SIGP_CC_STATUS_STORED)
161                 return 1;
162         /* Status stored condition code is equivalent to cpu not running. */
163         return 0;
164 }
165
166 /*
167  * Find struct pcpu by cpu address.
168  */
169 static struct pcpu *pcpu_find_address(const struct cpumask *mask, u16 address)
170 {
171         int cpu;
172
173         for_each_cpu(cpu, mask)
174                 if (pcpu_devices[cpu].address == address)
175                         return pcpu_devices + cpu;
176         return NULL;
177 }
178
179 static void pcpu_ec_call(struct pcpu *pcpu, int ec_bit)
180 {
181         int order;
182
183         if (test_and_set_bit(ec_bit, &pcpu->ec_mask))
184                 return;
185         order = pcpu_running(pcpu) ? SIGP_EXTERNAL_CALL : SIGP_EMERGENCY_SIGNAL;
186         pcpu->ec_clk = get_tod_clock_fast();
187         pcpu_sigp_retry(pcpu, order, 0);
188 }
189
190 #define ASYNC_FRAME_OFFSET (ASYNC_SIZE - STACK_FRAME_OVERHEAD - __PT_SIZE)
191 #define PANIC_FRAME_OFFSET (PAGE_SIZE - STACK_FRAME_OVERHEAD - __PT_SIZE)
192
193 static int pcpu_alloc_lowcore(struct pcpu *pcpu, int cpu)
194 {
195         unsigned long async_stack, panic_stack;
196         unsigned long mcesa_origin, mcesa_bits;
197         struct lowcore *lc;
198
199         mcesa_origin = mcesa_bits = 0;
200         if (pcpu != &pcpu_devices[0]) {
201                 pcpu->lowcore = (struct lowcore *)
202                         __get_free_pages(GFP_KERNEL | GFP_DMA, LC_ORDER);
203                 async_stack = __get_free_pages(GFP_KERNEL, ASYNC_ORDER);
204                 panic_stack = __get_free_page(GFP_KERNEL);
205                 if (!pcpu->lowcore || !panic_stack || !async_stack)
206                         goto out;
207                 if (MACHINE_HAS_VX || MACHINE_HAS_GS) {
208                         mcesa_origin = (unsigned long)
209                                 kmem_cache_alloc(pcpu_mcesa_cache, GFP_KERNEL);
210                         if (!mcesa_origin)
211                                 goto out;
212                         /* The pointer is stored with mcesa_bits ORed in */
213                         kmemleak_not_leak((void *) mcesa_origin);
214                         mcesa_bits = MACHINE_HAS_GS ? 11 : 0;
215                 }
216         } else {
217                 async_stack = pcpu->lowcore->async_stack - ASYNC_FRAME_OFFSET;
218                 panic_stack = pcpu->lowcore->panic_stack - PANIC_FRAME_OFFSET;
219                 mcesa_origin = pcpu->lowcore->mcesad & MCESA_ORIGIN_MASK;
220                 mcesa_bits = pcpu->lowcore->mcesad & MCESA_LC_MASK;
221         }
222         lc = pcpu->lowcore;
223         memcpy(lc, &S390_lowcore, 512);
224         memset((char *) lc + 512, 0, sizeof(*lc) - 512);
225         lc->async_stack = async_stack + ASYNC_FRAME_OFFSET;
226         lc->panic_stack = panic_stack + PANIC_FRAME_OFFSET;
227         lc->mcesad = mcesa_origin | mcesa_bits;
228         lc->cpu_nr = cpu;
229         lc->spinlock_lockval = arch_spin_lockval(cpu);
230         if (vdso_alloc_per_cpu(lc))
231                 goto out;
232         lowcore_ptr[cpu] = lc;
233         pcpu_sigp_retry(pcpu, SIGP_SET_PREFIX, (u32)(unsigned long) lc);
234         return 0;
235 out:
236         if (pcpu != &pcpu_devices[0]) {
237                 if (mcesa_origin)
238                         kmem_cache_free(pcpu_mcesa_cache,
239                                         (void *) mcesa_origin);
240                 free_page(panic_stack);
241                 free_pages(async_stack, ASYNC_ORDER);
242                 free_pages((unsigned long) pcpu->lowcore, LC_ORDER);
243         }
244         return -ENOMEM;
245 }
246
247 #ifdef CONFIG_HOTPLUG_CPU
248
249 static void pcpu_free_lowcore(struct pcpu *pcpu)
250 {
251         unsigned long mcesa_origin;
252
253         pcpu_sigp_retry(pcpu, SIGP_SET_PREFIX, 0);
254         lowcore_ptr[pcpu - pcpu_devices] = NULL;
255         vdso_free_per_cpu(pcpu->lowcore);
256         if (pcpu == &pcpu_devices[0])
257                 return;
258         if (MACHINE_HAS_VX || MACHINE_HAS_GS) {
259                 mcesa_origin = pcpu->lowcore->mcesad & MCESA_ORIGIN_MASK;
260                 kmem_cache_free(pcpu_mcesa_cache, (void *) mcesa_origin);
261         }
262         free_page(pcpu->lowcore->panic_stack-PANIC_FRAME_OFFSET);
263         free_pages(pcpu->lowcore->async_stack-ASYNC_FRAME_OFFSET, ASYNC_ORDER);
264         free_pages((unsigned long) pcpu->lowcore, LC_ORDER);
265 }
266
267 #endif /* CONFIG_HOTPLUG_CPU */
268
269 static void pcpu_prepare_secondary(struct pcpu *pcpu, int cpu)
270 {
271         struct lowcore *lc = pcpu->lowcore;
272
273         cpumask_set_cpu(cpu, &init_mm.context.cpu_attach_mask);
274         cpumask_set_cpu(cpu, mm_cpumask(&init_mm));
275         lc->cpu_nr = cpu;
276         lc->spinlock_lockval = arch_spin_lockval(cpu);
277         lc->percpu_offset = __per_cpu_offset[cpu];
278         lc->kernel_asce = S390_lowcore.kernel_asce;
279         lc->machine_flags = S390_lowcore.machine_flags;
280         lc->user_timer = lc->system_timer = lc->steal_timer = 0;
281         __ctl_store(lc->cregs_save_area, 0, 15);
282         save_access_regs((unsigned int *) lc->access_regs_save_area);
283         memcpy(lc->stfle_fac_list, S390_lowcore.stfle_fac_list,
284                MAX_FACILITY_BIT/8);
285 }
286
287 static void pcpu_attach_task(struct pcpu *pcpu, struct task_struct *tsk)
288 {
289         struct lowcore *lc = pcpu->lowcore;
290
291         lc->kernel_stack = (unsigned long) task_stack_page(tsk)
292                 + THREAD_SIZE - STACK_FRAME_OVERHEAD - sizeof(struct pt_regs);
293         lc->current_task = (unsigned long) tsk;
294         lc->lpp = LPP_MAGIC;
295         lc->current_pid = tsk->pid;
296         lc->user_timer = tsk->thread.user_timer;
297         lc->guest_timer = tsk->thread.guest_timer;
298         lc->system_timer = tsk->thread.system_timer;
299         lc->hardirq_timer = tsk->thread.hardirq_timer;
300         lc->softirq_timer = tsk->thread.softirq_timer;
301         lc->steal_timer = 0;
302 }
303
304 static void pcpu_start_fn(struct pcpu *pcpu, void (*func)(void *), void *data)
305 {
306         struct lowcore *lc = pcpu->lowcore;
307
308         lc->restart_stack = lc->kernel_stack;
309         lc->restart_fn = (unsigned long) func;
310         lc->restart_data = (unsigned long) data;
311         lc->restart_source = -1UL;
312         pcpu_sigp_retry(pcpu, SIGP_RESTART, 0);
313 }
314
315 /*
316  * Call function via PSW restart on pcpu and stop the current cpu.
317  */
318 static void pcpu_delegate(struct pcpu *pcpu, void (*func)(void *),
319                           void *data, unsigned long stack)
320 {
321         struct lowcore *lc = lowcore_ptr[pcpu - pcpu_devices];
322         unsigned long source_cpu = stap();
323
324         __load_psw_mask(PSW_KERNEL_BITS);
325         if (pcpu->address == source_cpu)
326                 func(data);     /* should not return */
327         /* Stop target cpu (if func returns this stops the current cpu). */
328         pcpu_sigp_retry(pcpu, SIGP_STOP, 0);
329         /* Restart func on the target cpu and stop the current cpu. */
330         mem_assign_absolute(lc->restart_stack, stack);
331         mem_assign_absolute(lc->restart_fn, (unsigned long) func);
332         mem_assign_absolute(lc->restart_data, (unsigned long) data);
333         mem_assign_absolute(lc->restart_source, source_cpu);
334         asm volatile(
335                 "0:     sigp    0,%0,%2 # sigp restart to target cpu\n"
336                 "       brc     2,0b    # busy, try again\n"
337                 "1:     sigp    0,%1,%3 # sigp stop to current cpu\n"
338                 "       brc     2,1b    # busy, try again\n"
339                 : : "d" (pcpu->address), "d" (source_cpu),
340                     "K" (SIGP_RESTART), "K" (SIGP_STOP)
341                 : "0", "1", "cc");
342         for (;;) ;
343 }
344
345 /*
346  * Enable additional logical cpus for multi-threading.
347  */
348 static int pcpu_set_smt(unsigned int mtid)
349 {
350         int cc;
351
352         if (smp_cpu_mtid == mtid)
353                 return 0;
354         cc = __pcpu_sigp(0, SIGP_SET_MULTI_THREADING, mtid, NULL);
355         if (cc == 0) {
356                 smp_cpu_mtid = mtid;
357                 smp_cpu_mt_shift = 0;
358                 while (smp_cpu_mtid >= (1U << smp_cpu_mt_shift))
359                         smp_cpu_mt_shift++;
360                 pcpu_devices[0].address = stap();
361         }
362         return cc;
363 }
364
365 /*
366  * Call function on an online CPU.
367  */
368 void smp_call_online_cpu(void (*func)(void *), void *data)
369 {
370         struct pcpu *pcpu;
371
372         /* Use the current cpu if it is online. */
373         pcpu = pcpu_find_address(cpu_online_mask, stap());
374         if (!pcpu)
375                 /* Use the first online cpu. */
376                 pcpu = pcpu_devices + cpumask_first(cpu_online_mask);
377         pcpu_delegate(pcpu, func, data, (unsigned long) restart_stack);
378 }
379
380 /*
381  * Call function on the ipl CPU.
382  */
383 void smp_call_ipl_cpu(void (*func)(void *), void *data)
384 {
385         pcpu_delegate(&pcpu_devices[0], func, data,
386                       pcpu_devices->lowcore->panic_stack -
387                       PANIC_FRAME_OFFSET + PAGE_SIZE);
388 }
389
390 int smp_find_processor_id(u16 address)
391 {
392         int cpu;
393
394         for_each_present_cpu(cpu)
395                 if (pcpu_devices[cpu].address == address)
396                         return cpu;
397         return -1;
398 }
399
400 bool arch_vcpu_is_preempted(int cpu)
401 {
402         if (test_cpu_flag_of(CIF_ENABLED_WAIT, cpu))
403                 return false;
404         if (pcpu_running(pcpu_devices + cpu))
405                 return false;
406         return true;
407 }
408 EXPORT_SYMBOL(arch_vcpu_is_preempted);
409
410 void smp_yield_cpu(int cpu)
411 {
412         if (MACHINE_HAS_DIAG9C) {
413                 diag_stat_inc_norecursion(DIAG_STAT_X09C);
414                 asm volatile("diag %0,0,0x9c"
415                              : : "d" (pcpu_devices[cpu].address));
416         } else if (MACHINE_HAS_DIAG44) {
417                 diag_stat_inc_norecursion(DIAG_STAT_X044);
418                 asm volatile("diag 0,0,0x44");
419         }
420 }
421
422 /*
423  * Send cpus emergency shutdown signal. This gives the cpus the
424  * opportunity to complete outstanding interrupts.
425  */
426 static void smp_emergency_stop(cpumask_t *cpumask)
427 {
428         u64 end;
429         int cpu;
430
431         end = get_tod_clock() + (1000000UL << 12);
432         for_each_cpu(cpu, cpumask) {
433                 struct pcpu *pcpu = pcpu_devices + cpu;
434                 set_bit(ec_stop_cpu, &pcpu->ec_mask);
435                 while (__pcpu_sigp(pcpu->address, SIGP_EMERGENCY_SIGNAL,
436                                    0, NULL) == SIGP_CC_BUSY &&
437                        get_tod_clock() < end)
438                         cpu_relax();
439         }
440         while (get_tod_clock() < end) {
441                 for_each_cpu(cpu, cpumask)
442                         if (pcpu_stopped(pcpu_devices + cpu))
443                                 cpumask_clear_cpu(cpu, cpumask);
444                 if (cpumask_empty(cpumask))
445                         break;
446                 cpu_relax();
447         }
448 }
449
450 /*
451  * Stop all cpus but the current one.
452  */
453 void smp_send_stop(void)
454 {
455         cpumask_t cpumask;
456         int cpu;
457
458         /* Disable all interrupts/machine checks */
459         __load_psw_mask(PSW_KERNEL_BITS | PSW_MASK_DAT);
460         trace_hardirqs_off();
461
462         debug_set_critical();
463         cpumask_copy(&cpumask, cpu_online_mask);
464         cpumask_clear_cpu(smp_processor_id(), &cpumask);
465
466         if (oops_in_progress)
467                 smp_emergency_stop(&cpumask);
468
469         /* stop all processors */
470         for_each_cpu(cpu, &cpumask) {
471                 struct pcpu *pcpu = pcpu_devices + cpu;
472                 pcpu_sigp_retry(pcpu, SIGP_STOP, 0);
473                 while (!pcpu_stopped(pcpu))
474                         cpu_relax();
475         }
476 }
477
478 /*
479  * This is the main routine where commands issued by other
480  * cpus are handled.
481  */
482 static void smp_handle_ext_call(void)
483 {
484         unsigned long bits;
485
486         /* handle bit signal external calls */
487         bits = xchg(&pcpu_devices[smp_processor_id()].ec_mask, 0);
488         if (test_bit(ec_stop_cpu, &bits))
489                 smp_stop_cpu();
490         if (test_bit(ec_schedule, &bits))
491                 scheduler_ipi();
492         if (test_bit(ec_call_function_single, &bits))
493                 generic_smp_call_function_single_interrupt();
494 }
495
496 static void do_ext_call_interrupt(struct ext_code ext_code,
497                                   unsigned int param32, unsigned long param64)
498 {
499         inc_irq_stat(ext_code.code == 0x1202 ? IRQEXT_EXC : IRQEXT_EMS);
500         smp_handle_ext_call();
501 }
502
503 void arch_send_call_function_ipi_mask(const struct cpumask *mask)
504 {
505         int cpu;
506
507         for_each_cpu(cpu, mask)
508                 pcpu_ec_call(pcpu_devices + cpu, ec_call_function_single);
509 }
510
511 void arch_send_call_function_single_ipi(int cpu)
512 {
513         pcpu_ec_call(pcpu_devices + cpu, ec_call_function_single);
514 }
515
516 /*
517  * this function sends a 'reschedule' IPI to another CPU.
518  * it goes straight through and wastes no time serializing
519  * anything. Worst case is that we lose a reschedule ...
520  */
521 void smp_send_reschedule(int cpu)
522 {
523         pcpu_ec_call(pcpu_devices + cpu, ec_schedule);
524 }
525
526 /*
527  * parameter area for the set/clear control bit callbacks
528  */
529 struct ec_creg_mask_parms {
530         unsigned long orval;
531         unsigned long andval;
532         int cr;
533 };
534
535 /*
536  * callback for setting/clearing control bits
537  */
538 static void smp_ctl_bit_callback(void *info)
539 {
540         struct ec_creg_mask_parms *pp = info;
541         unsigned long cregs[16];
542
543         __ctl_store(cregs, 0, 15);
544         cregs[pp->cr] = (cregs[pp->cr] & pp->andval) | pp->orval;
545         __ctl_load(cregs, 0, 15);
546 }
547
548 /*
549  * Set a bit in a control register of all cpus
550  */
551 void smp_ctl_set_bit(int cr, int bit)
552 {
553         struct ec_creg_mask_parms parms = { 1UL << bit, -1UL, cr };
554
555         on_each_cpu(smp_ctl_bit_callback, &parms, 1);
556 }
557 EXPORT_SYMBOL(smp_ctl_set_bit);
558
559 /*
560  * Clear a bit in a control register of all cpus
561  */
562 void smp_ctl_clear_bit(int cr, int bit)
563 {
564         struct ec_creg_mask_parms parms = { 0, ~(1UL << bit), cr };
565
566         on_each_cpu(smp_ctl_bit_callback, &parms, 1);
567 }
568 EXPORT_SYMBOL(smp_ctl_clear_bit);
569
570 #ifdef CONFIG_CRASH_DUMP
571
572 int smp_store_status(int cpu)
573 {
574         struct pcpu *pcpu = pcpu_devices + cpu;
575         unsigned long pa;
576
577         pa = __pa(&pcpu->lowcore->floating_pt_save_area);
578         if (__pcpu_sigp_relax(pcpu->address, SIGP_STORE_STATUS_AT_ADDRESS,
579                               pa) != SIGP_CC_ORDER_CODE_ACCEPTED)
580                 return -EIO;
581         if (!MACHINE_HAS_VX && !MACHINE_HAS_GS)
582                 return 0;
583         pa = __pa(pcpu->lowcore->mcesad & MCESA_ORIGIN_MASK);
584         if (MACHINE_HAS_GS)
585                 pa |= pcpu->lowcore->mcesad & MCESA_LC_MASK;
586         if (__pcpu_sigp_relax(pcpu->address, SIGP_STORE_ADDITIONAL_STATUS,
587                               pa) != SIGP_CC_ORDER_CODE_ACCEPTED)
588                 return -EIO;
589         return 0;
590 }
591
592 /*
593  * Collect CPU state of the previous, crashed system.
594  * There are four cases:
595  * 1) standard zfcp dump
596  *    condition: OLDMEM_BASE == NULL && ipl_info.type == IPL_TYPE_FCP_DUMP
597  *    The state for all CPUs except the boot CPU needs to be collected
598  *    with sigp stop-and-store-status. The boot CPU state is located in
599  *    the absolute lowcore of the memory stored in the HSA. The zcore code
600  *    will copy the boot CPU state from the HSA.
601  * 2) stand-alone kdump for SCSI (zfcp dump with swapped memory)
602  *    condition: OLDMEM_BASE != NULL && ipl_info.type == IPL_TYPE_FCP_DUMP
603  *    The state for all CPUs except the boot CPU needs to be collected
604  *    with sigp stop-and-store-status. The firmware or the boot-loader
605  *    stored the registers of the boot CPU in the absolute lowcore in the
606  *    memory of the old system.
607  * 3) kdump and the old kernel did not store the CPU state,
608  *    or stand-alone kdump for DASD
609  *    condition: OLDMEM_BASE != NULL && !is_kdump_kernel()
610  *    The state for all CPUs except the boot CPU needs to be collected
611  *    with sigp stop-and-store-status. The kexec code or the boot-loader
612  *    stored the registers of the boot CPU in the memory of the old system.
613  * 4) kdump and the old kernel stored the CPU state
614  *    condition: OLDMEM_BASE != NULL && is_kdump_kernel()
615  *    This case does not exist for s390 anymore, setup_arch explicitly
616  *    deactivates the elfcorehdr= kernel parameter
617  */
618 static __init void smp_save_cpu_vxrs(struct save_area *sa, u16 addr,
619                                      bool is_boot_cpu, unsigned long page)
620 {
621         __vector128 *vxrs = (__vector128 *) page;
622
623         if (is_boot_cpu)
624                 vxrs = boot_cpu_vector_save_area;
625         else
626                 __pcpu_sigp_relax(addr, SIGP_STORE_ADDITIONAL_STATUS, page);
627         save_area_add_vxrs(sa, vxrs);
628 }
629
630 static __init void smp_save_cpu_regs(struct save_area *sa, u16 addr,
631                                      bool is_boot_cpu, unsigned long page)
632 {
633         void *regs = (void *) page;
634
635         if (is_boot_cpu)
636                 copy_oldmem_kernel(regs, (void *) __LC_FPREGS_SAVE_AREA, 512);
637         else
638                 __pcpu_sigp_relax(addr, SIGP_STORE_STATUS_AT_ADDRESS, page);
639         save_area_add_regs(sa, regs);
640 }
641
642 void __init smp_save_dump_cpus(void)
643 {
644         int addr, boot_cpu_addr, max_cpu_addr;
645         struct save_area *sa;
646         unsigned long page;
647         bool is_boot_cpu;
648
649         if (!(OLDMEM_BASE || ipl_info.type == IPL_TYPE_FCP_DUMP))
650                 /* No previous system present, normal boot. */
651                 return;
652         /* Allocate a page as dumping area for the store status sigps */
653         page = memblock_alloc_base(PAGE_SIZE, PAGE_SIZE, 1UL << 31);
654         /* Set multi-threading state to the previous system. */
655         pcpu_set_smt(sclp.mtid_prev);
656         boot_cpu_addr = stap();
657         max_cpu_addr = SCLP_MAX_CORES << sclp.mtid_prev;
658         for (addr = 0; addr <= max_cpu_addr; addr++) {
659                 if (__pcpu_sigp_relax(addr, SIGP_SENSE, 0) ==
660                     SIGP_CC_NOT_OPERATIONAL)
661                         continue;
662                 is_boot_cpu = (addr == boot_cpu_addr);
663                 /* Allocate save area */
664                 sa = save_area_alloc(is_boot_cpu);
665                 if (!sa)
666                         panic("could not allocate memory for save area\n");
667                 if (MACHINE_HAS_VX)
668                         /* Get the vector registers */
669                         smp_save_cpu_vxrs(sa, addr, is_boot_cpu, page);
670                 /*
671                  * For a zfcp dump OLDMEM_BASE == NULL and the registers
672                  * of the boot CPU are stored in the HSA. To retrieve
673                  * these registers an SCLP request is required which is
674                  * done by drivers/s390/char/zcore.c:init_cpu_info()
675                  */
676                 if (!is_boot_cpu || OLDMEM_BASE)
677                         /* Get the CPU registers */
678                         smp_save_cpu_regs(sa, addr, is_boot_cpu, page);
679         }
680         memblock_free(page, PAGE_SIZE);
681         diag308_reset();
682         pcpu_set_smt(0);
683 }
684 #endif /* CONFIG_CRASH_DUMP */
685
686 void smp_cpu_set_polarization(int cpu, int val)
687 {
688         pcpu_devices[cpu].polarization = val;
689 }
690
691 int smp_cpu_get_polarization(int cpu)
692 {
693         return pcpu_devices[cpu].polarization;
694 }
695
696 static void __ref smp_get_core_info(struct sclp_core_info *info, int early)
697 {
698         static int use_sigp_detection;
699         int address;
700
701         if (use_sigp_detection || sclp_get_core_info(info, early)) {
702                 use_sigp_detection = 1;
703                 for (address = 0;
704                      address < (SCLP_MAX_CORES << smp_cpu_mt_shift);
705                      address += (1U << smp_cpu_mt_shift)) {
706                         if (__pcpu_sigp_relax(address, SIGP_SENSE, 0) ==
707                             SIGP_CC_NOT_OPERATIONAL)
708                                 continue;
709                         info->core[info->configured].core_id =
710                                 address >> smp_cpu_mt_shift;
711                         info->configured++;
712                 }
713                 info->combined = info->configured;
714         }
715 }
716
717 static int smp_add_present_cpu(int cpu);
718
719 static int __smp_rescan_cpus(struct sclp_core_info *info, int sysfs_add)
720 {
721         struct pcpu *pcpu;
722         cpumask_t avail;
723         int cpu, nr, i, j;
724         u16 address;
725
726         nr = 0;
727         cpumask_xor(&avail, cpu_possible_mask, cpu_present_mask);
728         cpu = cpumask_first(&avail);
729         for (i = 0; (i < info->combined) && (cpu < nr_cpu_ids); i++) {
730                 if (sclp.has_core_type && info->core[i].type != boot_core_type)
731                         continue;
732                 address = info->core[i].core_id << smp_cpu_mt_shift;
733                 for (j = 0; j <= smp_cpu_mtid; j++) {
734                         if (pcpu_find_address(cpu_present_mask, address + j))
735                                 continue;
736                         pcpu = pcpu_devices + cpu;
737                         pcpu->address = address + j;
738                         pcpu->state =
739                                 (cpu >= info->configured*(smp_cpu_mtid + 1)) ?
740                                 CPU_STATE_STANDBY : CPU_STATE_CONFIGURED;
741                         smp_cpu_set_polarization(cpu, POLARIZATION_UNKNOWN);
742                         set_cpu_present(cpu, true);
743                         if (sysfs_add && smp_add_present_cpu(cpu) != 0)
744                                 set_cpu_present(cpu, false);
745                         else
746                                 nr++;
747                         cpu = cpumask_next(cpu, &avail);
748                         if (cpu >= nr_cpu_ids)
749                                 break;
750                 }
751         }
752         return nr;
753 }
754
755 void __init smp_detect_cpus(void)
756 {
757         unsigned int cpu, mtid, c_cpus, s_cpus;
758         struct sclp_core_info *info;
759         u16 address;
760
761         /* Get CPU information */
762         info = memblock_virt_alloc(sizeof(*info), 8);
763         smp_get_core_info(info, 1);
764         /* Find boot CPU type */
765         if (sclp.has_core_type) {
766                 address = stap();
767                 for (cpu = 0; cpu < info->combined; cpu++)
768                         if (info->core[cpu].core_id == address) {
769                                 /* The boot cpu dictates the cpu type. */
770                                 boot_core_type = info->core[cpu].type;
771                                 break;
772                         }
773                 if (cpu >= info->combined)
774                         panic("Could not find boot CPU type");
775         }
776
777         /* Set multi-threading state for the current system */
778         mtid = boot_core_type ? sclp.mtid : sclp.mtid_cp;
779         mtid = (mtid < smp_max_threads) ? mtid : smp_max_threads - 1;
780         pcpu_set_smt(mtid);
781
782         /* Print number of CPUs */
783         c_cpus = s_cpus = 0;
784         for (cpu = 0; cpu < info->combined; cpu++) {
785                 if (sclp.has_core_type &&
786                     info->core[cpu].type != boot_core_type)
787                         continue;
788                 if (cpu < info->configured)
789                         c_cpus += smp_cpu_mtid + 1;
790                 else
791                         s_cpus += smp_cpu_mtid + 1;
792         }
793         pr_info("%d configured CPUs, %d standby CPUs\n", c_cpus, s_cpus);
794
795         /* Add CPUs present at boot */
796         get_online_cpus();
797         __smp_rescan_cpus(info, 0);
798         put_online_cpus();
799         memblock_free_early((unsigned long)info, sizeof(*info));
800 }
801
802 /*
803  *      Activate a secondary processor.
804  */
805 static void smp_start_secondary(void *cpuvoid)
806 {
807         S390_lowcore.last_update_clock = get_tod_clock();
808         S390_lowcore.restart_stack = (unsigned long) restart_stack;
809         S390_lowcore.restart_fn = (unsigned long) do_restart;
810         S390_lowcore.restart_data = 0;
811         S390_lowcore.restart_source = -1UL;
812         restore_access_regs(S390_lowcore.access_regs_save_area);
813         __ctl_load(S390_lowcore.cregs_save_area, 0, 15);
814         __load_psw_mask(PSW_KERNEL_BITS | PSW_MASK_DAT);
815         cpu_init();
816         preempt_disable();
817         init_cpu_timer();
818         vtime_init();
819         pfault_init();
820         notify_cpu_starting(smp_processor_id());
821         set_cpu_online(smp_processor_id(), true);
822         inc_irq_stat(CPU_RST);
823         local_irq_enable();
824         cpu_startup_entry(CPUHP_AP_ONLINE_IDLE);
825 }
826
827 /* Upping and downing of CPUs */
828 int __cpu_up(unsigned int cpu, struct task_struct *tidle)
829 {
830         struct pcpu *pcpu;
831         int base, i, rc;
832
833         pcpu = pcpu_devices + cpu;
834         if (pcpu->state != CPU_STATE_CONFIGURED)
835                 return -EIO;
836         base = smp_get_base_cpu(cpu);
837         for (i = 0; i <= smp_cpu_mtid; i++) {
838                 if (base + i < nr_cpu_ids)
839                         if (cpu_online(base + i))
840                                 break;
841         }
842         /*
843          * If this is the first CPU of the core to get online
844          * do an initial CPU reset.
845          */
846         if (i > smp_cpu_mtid &&
847             pcpu_sigp_retry(pcpu_devices + base, SIGP_INITIAL_CPU_RESET, 0) !=
848             SIGP_CC_ORDER_CODE_ACCEPTED)
849                 return -EIO;
850
851         rc = pcpu_alloc_lowcore(pcpu, cpu);
852         if (rc)
853                 return rc;
854         pcpu_prepare_secondary(pcpu, cpu);
855         pcpu_attach_task(pcpu, tidle);
856         pcpu_start_fn(pcpu, smp_start_secondary, NULL);
857         /* Wait until cpu puts itself in the online & active maps */
858         while (!cpu_online(cpu))
859                 cpu_relax();
860         return 0;
861 }
862
863 static unsigned int setup_possible_cpus __initdata;
864
865 static int __init _setup_possible_cpus(char *s)
866 {
867         get_option(&s, &setup_possible_cpus);
868         return 0;
869 }
870 early_param("possible_cpus", _setup_possible_cpus);
871
872 #ifdef CONFIG_HOTPLUG_CPU
873
874 int __cpu_disable(void)
875 {
876         unsigned long cregs[16];
877
878         /* Handle possible pending IPIs */
879         smp_handle_ext_call();
880         set_cpu_online(smp_processor_id(), false);
881         /* Disable pseudo page faults on this cpu. */
882         pfault_fini();
883         /* Disable interrupt sources via control register. */
884         __ctl_store(cregs, 0, 15);
885         cregs[0]  &= ~0x0000ee70UL;     /* disable all external interrupts */
886         cregs[6]  &= ~0xff000000UL;     /* disable all I/O interrupts */
887         cregs[14] &= ~0x1f000000UL;     /* disable most machine checks */
888         __ctl_load(cregs, 0, 15);
889         clear_cpu_flag(CIF_NOHZ_DELAY);
890         return 0;
891 }
892
893 void __cpu_die(unsigned int cpu)
894 {
895         struct pcpu *pcpu;
896
897         /* Wait until target cpu is down */
898         pcpu = pcpu_devices + cpu;
899         while (!pcpu_stopped(pcpu))
900                 cpu_relax();
901         pcpu_free_lowcore(pcpu);
902         cpumask_clear_cpu(cpu, mm_cpumask(&init_mm));
903         cpumask_clear_cpu(cpu, &init_mm.context.cpu_attach_mask);
904 }
905
906 void __noreturn cpu_die(void)
907 {
908         idle_task_exit();
909         pcpu_sigp_retry(pcpu_devices + smp_processor_id(), SIGP_STOP, 0);
910         for (;;) ;
911 }
912
913 #endif /* CONFIG_HOTPLUG_CPU */
914
915 void __init smp_fill_possible_mask(void)
916 {
917         unsigned int possible, sclp_max, cpu;
918
919         sclp_max = max(sclp.mtid, sclp.mtid_cp) + 1;
920         sclp_max = min(smp_max_threads, sclp_max);
921         sclp_max = (sclp.max_cores * sclp_max) ?: nr_cpu_ids;
922         possible = setup_possible_cpus ?: nr_cpu_ids;
923         possible = min(possible, sclp_max);
924         for (cpu = 0; cpu < possible && cpu < nr_cpu_ids; cpu++)
925                 set_cpu_possible(cpu, true);
926 }
927
928 void __init smp_prepare_cpus(unsigned int max_cpus)
929 {
930         unsigned long size;
931
932         /* request the 0x1201 emergency signal external interrupt */
933         if (register_external_irq(EXT_IRQ_EMERGENCY_SIG, do_ext_call_interrupt))
934                 panic("Couldn't request external interrupt 0x1201");
935         /* request the 0x1202 external call external interrupt */
936         if (register_external_irq(EXT_IRQ_EXTERNAL_CALL, do_ext_call_interrupt))
937                 panic("Couldn't request external interrupt 0x1202");
938         /* create slab cache for the machine-check-extended-save-areas */
939         if (MACHINE_HAS_VX || MACHINE_HAS_GS) {
940                 size = 1UL << (MACHINE_HAS_GS ? 11 : 10);
941                 pcpu_mcesa_cache = kmem_cache_create("nmi_save_areas",
942                                                      size, size, 0, NULL);
943                 if (!pcpu_mcesa_cache)
944                         panic("Couldn't create nmi save area cache");
945         }
946 }
947
948 void __init smp_prepare_boot_cpu(void)
949 {
950         struct pcpu *pcpu = pcpu_devices;
951
952         WARN_ON(!cpu_present(0) || !cpu_online(0));
953         pcpu->state = CPU_STATE_CONFIGURED;
954         pcpu->lowcore = (struct lowcore *)(unsigned long) store_prefix();
955         S390_lowcore.percpu_offset = __per_cpu_offset[0];
956         smp_cpu_set_polarization(0, POLARIZATION_UNKNOWN);
957 }
958
959 void __init smp_cpus_done(unsigned int max_cpus)
960 {
961 }
962
963 void __init smp_setup_processor_id(void)
964 {
965         pcpu_devices[0].address = stap();
966         S390_lowcore.cpu_nr = 0;
967         S390_lowcore.spinlock_lockval = arch_spin_lockval(0);
968 }
969
970 /*
971  * the frequency of the profiling timer can be changed
972  * by writing a multiplier value into /proc/profile.
973  *
974  * usually you want to run this on all CPUs ;)
975  */
976 int setup_profiling_timer(unsigned int multiplier)
977 {
978         return 0;
979 }
980
981 #ifdef CONFIG_HOTPLUG_CPU
982 static ssize_t cpu_configure_show(struct device *dev,
983                                   struct device_attribute *attr, char *buf)
984 {
985         ssize_t count;
986
987         mutex_lock(&smp_cpu_state_mutex);
988         count = sprintf(buf, "%d\n", pcpu_devices[dev->id].state);
989         mutex_unlock(&smp_cpu_state_mutex);
990         return count;
991 }
992
993 static ssize_t cpu_configure_store(struct device *dev,
994                                    struct device_attribute *attr,
995                                    const char *buf, size_t count)
996 {
997         struct pcpu *pcpu;
998         int cpu, val, rc, i;
999         char delim;
1000
1001         if (sscanf(buf, "%d %c", &val, &delim) != 1)
1002                 return -EINVAL;
1003         if (val != 0 && val != 1)
1004                 return -EINVAL;
1005         get_online_cpus();
1006         mutex_lock(&smp_cpu_state_mutex);
1007         rc = -EBUSY;
1008         /* disallow configuration changes of online cpus and cpu 0 */
1009         cpu = dev->id;
1010         cpu = smp_get_base_cpu(cpu);
1011         if (cpu == 0)
1012                 goto out;
1013         for (i = 0; i <= smp_cpu_mtid; i++)
1014                 if (cpu_online(cpu + i))
1015                         goto out;
1016         pcpu = pcpu_devices + cpu;
1017         rc = 0;
1018         switch (val) {
1019         case 0:
1020                 if (pcpu->state != CPU_STATE_CONFIGURED)
1021                         break;
1022                 rc = sclp_core_deconfigure(pcpu->address >> smp_cpu_mt_shift);
1023                 if (rc)
1024                         break;
1025                 for (i = 0; i <= smp_cpu_mtid; i++) {
1026                         if (cpu + i >= nr_cpu_ids || !cpu_present(cpu + i))
1027                                 continue;
1028                         pcpu[i].state = CPU_STATE_STANDBY;
1029                         smp_cpu_set_polarization(cpu + i,
1030                                                  POLARIZATION_UNKNOWN);
1031                 }
1032                 topology_expect_change();
1033                 break;
1034         case 1:
1035                 if (pcpu->state != CPU_STATE_STANDBY)
1036                         break;
1037                 rc = sclp_core_configure(pcpu->address >> smp_cpu_mt_shift);
1038                 if (rc)
1039                         break;
1040                 for (i = 0; i <= smp_cpu_mtid; i++) {
1041                         if (cpu + i >= nr_cpu_ids || !cpu_present(cpu + i))
1042                                 continue;
1043                         pcpu[i].state = CPU_STATE_CONFIGURED;
1044                         smp_cpu_set_polarization(cpu + i,
1045                                                  POLARIZATION_UNKNOWN);
1046                 }
1047                 topology_expect_change();
1048                 break;
1049         default:
1050                 break;
1051         }
1052 out:
1053         mutex_unlock(&smp_cpu_state_mutex);
1054         put_online_cpus();
1055         return rc ? rc : count;
1056 }
1057 static DEVICE_ATTR(configure, 0644, cpu_configure_show, cpu_configure_store);
1058 #endif /* CONFIG_HOTPLUG_CPU */
1059
1060 static ssize_t show_cpu_address(struct device *dev,
1061                                 struct device_attribute *attr, char *buf)
1062 {
1063         return sprintf(buf, "%d\n", pcpu_devices[dev->id].address);
1064 }
1065 static DEVICE_ATTR(address, 0444, show_cpu_address, NULL);
1066
1067 static struct attribute *cpu_common_attrs[] = {
1068 #ifdef CONFIG_HOTPLUG_CPU
1069         &dev_attr_configure.attr,
1070 #endif
1071         &dev_attr_address.attr,
1072         NULL,
1073 };
1074
1075 static struct attribute_group cpu_common_attr_group = {
1076         .attrs = cpu_common_attrs,
1077 };
1078
1079 static struct attribute *cpu_online_attrs[] = {
1080         &dev_attr_idle_count.attr,
1081         &dev_attr_idle_time_us.attr,
1082         NULL,
1083 };
1084
1085 static struct attribute_group cpu_online_attr_group = {
1086         .attrs = cpu_online_attrs,
1087 };
1088
1089 static int smp_cpu_online(unsigned int cpu)
1090 {
1091         struct device *s = &per_cpu(cpu_device, cpu)->dev;
1092
1093         return sysfs_create_group(&s->kobj, &cpu_online_attr_group);
1094 }
1095 static int smp_cpu_pre_down(unsigned int cpu)
1096 {
1097         struct device *s = &per_cpu(cpu_device, cpu)->dev;
1098
1099         sysfs_remove_group(&s->kobj, &cpu_online_attr_group);
1100         return 0;
1101 }
1102
1103 static int smp_add_present_cpu(int cpu)
1104 {
1105         struct device *s;
1106         struct cpu *c;
1107         int rc;
1108
1109         c = kzalloc(sizeof(*c), GFP_KERNEL);
1110         if (!c)
1111                 return -ENOMEM;
1112         per_cpu(cpu_device, cpu) = c;
1113         s = &c->dev;
1114         c->hotpluggable = 1;
1115         rc = register_cpu(c, cpu);
1116         if (rc)
1117                 goto out;
1118         rc = sysfs_create_group(&s->kobj, &cpu_common_attr_group);
1119         if (rc)
1120                 goto out_cpu;
1121         rc = topology_cpu_init(c);
1122         if (rc)
1123                 goto out_topology;
1124         return 0;
1125
1126 out_topology:
1127         sysfs_remove_group(&s->kobj, &cpu_common_attr_group);
1128 out_cpu:
1129 #ifdef CONFIG_HOTPLUG_CPU
1130         unregister_cpu(c);
1131 #endif
1132 out:
1133         return rc;
1134 }
1135
1136 #ifdef CONFIG_HOTPLUG_CPU
1137
1138 int __ref smp_rescan_cpus(void)
1139 {
1140         struct sclp_core_info *info;
1141         int nr;
1142
1143         info = kzalloc(sizeof(*info), GFP_KERNEL);
1144         if (!info)
1145                 return -ENOMEM;
1146         smp_get_core_info(info, 0);
1147         get_online_cpus();
1148         mutex_lock(&smp_cpu_state_mutex);
1149         nr = __smp_rescan_cpus(info, 1);
1150         mutex_unlock(&smp_cpu_state_mutex);
1151         put_online_cpus();
1152         kfree(info);
1153         if (nr)
1154                 topology_schedule_update();
1155         return 0;
1156 }
1157
1158 static ssize_t __ref rescan_store(struct device *dev,
1159                                   struct device_attribute *attr,
1160                                   const char *buf,
1161                                   size_t count)
1162 {
1163         int rc;
1164
1165         rc = smp_rescan_cpus();
1166         return rc ? rc : count;
1167 }
1168 static DEVICE_ATTR(rescan, 0200, NULL, rescan_store);
1169 #endif /* CONFIG_HOTPLUG_CPU */
1170
1171 static int __init s390_smp_init(void)
1172 {
1173         int cpu, rc = 0;
1174
1175 #ifdef CONFIG_HOTPLUG_CPU
1176         rc = device_create_file(cpu_subsys.dev_root, &dev_attr_rescan);
1177         if (rc)
1178                 return rc;
1179 #endif
1180         for_each_present_cpu(cpu) {
1181                 rc = smp_add_present_cpu(cpu);
1182                 if (rc)
1183                         goto out;
1184         }
1185
1186         rc = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "s390/smp:online",
1187                                smp_cpu_online, smp_cpu_pre_down);
1188         rc = rc <= 0 ? rc : 0;
1189 out:
1190         return rc;
1191 }
1192 subsys_initcall(s390_smp_init);