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/delay.h>
  31 #include <linux/interrupt.h>
  32 #include <linux/irqflags.h>
  33 #include <linux/cpu.h>
  34 #include <linux/slab.h>
  35 #include <linux/sched/hotplug.h>
  36 #include <linux/sched/task_stack.h>
  37 #include <linux/crash_dump.h>
  38 #include <linux/memblock.h>
  39 #include <linux/kprobes.h>
  40 #include <asm/asm-offsets.h>
  41 #include <asm/diag.h>
  42 #include <asm/switch_to.h>
  43 #include <asm/facility.h>
  44 #include <asm/ipl.h>
  45 #include <asm/setup.h>
  46 #include <asm/irq.h>
  47 #include <asm/tlbflush.h>
  48 #include <asm/vtimer.h>
  49 #include <asm/lowcore.h>
  50 #include <asm/sclp.h>
  51 #include <asm/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 unsigned int smp_cpu_mt_shift;
  86 EXPORT_SYMBOL(smp_cpu_mt_shift);
  87
  88 unsigned int smp_cpu_mtid;
  89 EXPORT_SYMBOL(smp_cpu_mtid);
  90
  91 #ifdef CONFIG_CRASH_DUMP
  92 __vector128 __initdata boot_cpu_vector_save_area[__NUM_VXRS];
  93 #endif
  94
  95 static unsigned int smp_max_threads __initdata = -1U;
  96
  97 static int __init early_nosmt(char *s)
  98 {
  99         smp_max_threads = 1;
 100         return 0;
 101 }
 102 early_param("nosmt", early_nosmt);
 103
 104 static int __init early_smt(char *s)
 105 {
 106         get_option(&s, &smp_max_threads);
 107         return 0;
 108 }
 109 early_param("smt", early_smt);
 110
 111 /*
 112  * The smp_cpu_state_mutex must be held when changing the state or polarization
 113  * member of a pcpu data structure within the pcpu_devices arreay.
 114  */
 115 DEFINE_MUTEX(smp_cpu_state_mutex);
 116
 117 /*
 118  * Signal processor helper functions.
 119  */
 120 static inline int __pcpu_sigp_relax(u16 addr, u8 order, unsigned long parm)
 121 {
 122         int cc;
 123
 124         while (1) {
 125                 cc = __pcpu_sigp(addr, order, parm, NULL);
 126                 if (cc != SIGP_CC_BUSY)
 127                         return cc;
 128                 cpu_relax();
 129         }
 130 }
 131
 132 static int pcpu_sigp_retry(struct pcpu *pcpu, u8 order, u32 parm)
 133 {
 134         int cc, retry;
 135
 136         for (retry = 0; ; retry++) {
 137                 cc = __pcpu_sigp(pcpu->address, order, parm, NULL);
 138                 if (cc != SIGP_CC_BUSY)
 139                         break;
 140                 if (retry >= 3)
 141                         udelay(10);
 142         }
 143         return cc;
 144 }
 145
 146 static inline int pcpu_stopped(struct pcpu *pcpu)
 147 {
 148         u32 uninitialized_var(status);
 149
 150         if (__pcpu_sigp(pcpu->address, SIGP_SENSE,
 151                         0, &status) != SIGP_CC_STATUS_STORED)
 152                 return 0;
 153         return !!(status & (SIGP_STATUS_CHECK_STOP|SIGP_STATUS_STOPPED));
 154 }
 155
 156 static inline int pcpu_running(struct pcpu *pcpu)
 157 {
 158         if (__pcpu_sigp(pcpu->address, SIGP_SENSE_RUNNING,
 159                         0, NULL) != SIGP_CC_STATUS_STORED)
 160                 return 1;
 161         /* Status stored condition code is equivalent to cpu not running. */
 162         return 0;
 163 }
 164
 165 /*
 166  * Find struct pcpu by cpu address.
 167  */
 168 static struct pcpu *pcpu_find_address(const struct cpumask *mask, u16 address)
 169 {
 170         int cpu;
 171
 172         for_each_cpu(cpu, mask)
 173                 if (pcpu_devices[cpu].address == address)
 174                         return pcpu_devices + cpu;
 175         return NULL;
 176 }
 177
 178 static void pcpu_ec_call(struct pcpu *pcpu, int ec_bit)
 179 {
 180         int order;
 181
 182         if (test_and_set_bit(ec_bit, &pcpu->ec_mask))
 183                 return;
 184         order = pcpu_running(pcpu) ? SIGP_EXTERNAL_CALL : SIGP_EMERGENCY_SIGNAL;
 185         pcpu->ec_clk = get_tod_clock_fast();
 186         pcpu_sigp_retry(pcpu, order, 0);
 187 }
 188
 189 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         if (nmi_alloc_per_cpu(lc))
 216                 goto out_async;
 217         if (vdso_alloc_per_cpu(lc))
 218                 goto out_mcesa;
 219         lowcore_ptr[cpu] = lc;
 220         pcpu_sigp_retry(pcpu, SIGP_SET_PREFIX, (u32)(unsigned long) lc);
 221         return 0;
 222
 223 out_mcesa:
 224         nmi_free_per_cpu(lc);
 225 out_async:
 226         stack_free(async_stack);
 227 out:
 228         if (pcpu != &pcpu_devices[0]) {
 229                 free_pages(nodat_stack, THREAD_SIZE_ORDER);
 230                 free_pages((unsigned long) pcpu->lowcore, LC_ORDER);
 231         }
 232         return -ENOMEM;
 233 }
 234
 235 #ifdef CONFIG_HOTPLUG_CPU
 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 #endif /* CONFIG_HOTPLUG_CPU */
 257
 258 static void pcpu_prepare_secondary(struct pcpu *pcpu, int cpu)
 259 {
 260         struct lowcore *lc = pcpu->lowcore;
 261
 262         cpumask_set_cpu(cpu, &init_mm.context.cpu_attach_mask);
 263         cpumask_set_cpu(cpu, mm_cpumask(&init_mm));
 264         lc->cpu_nr = cpu;
 265         lc->spinlock_lockval = arch_spin_lockval(cpu);
 266         lc->spinlock_index = 0;
 267         lc->percpu_offset = __per_cpu_offset[cpu];
 268         lc->kernel_asce = S390_lowcore.kernel_asce;
 269         lc->machine_flags = S390_lowcore.machine_flags;
 270         lc->user_timer = lc->system_timer = lc->steal_timer = 0;
 271         __ctl_store(lc->cregs_save_area, 0, 15);
 272         save_access_regs((unsigned int *) lc->access_regs_save_area);
 273         memcpy(lc->stfle_fac_list, S390_lowcore.stfle_fac_list,
 274                sizeof(lc->stfle_fac_list));
 275         memcpy(lc->alt_stfle_fac_list, S390_lowcore.alt_stfle_fac_list,
 276                sizeof(lc->alt_stfle_fac_list));
 277         arch_spin_lock_setup(cpu);
 278 }
 279
 280 static void pcpu_attach_task(struct pcpu *pcpu, struct task_struct *tsk)
 281 {
 282         struct lowcore *lc = pcpu->lowcore;
 283
 284         lc->kernel_stack = (unsigned long) task_stack_page(tsk)
 285                 + THREAD_SIZE - STACK_FRAME_OVERHEAD - sizeof(struct pt_regs);
 286         lc->current_task = (unsigned long) tsk;
 287         lc->lpp = LPP_MAGIC;
 288         lc->current_pid = tsk->pid;
 289         lc->user_timer = tsk->thread.user_timer;
 290         lc->guest_timer = tsk->thread.guest_timer;
 291         lc->system_timer = tsk->thread.system_timer;
 292         lc->hardirq_timer = tsk->thread.hardirq_timer;
 293         lc->softirq_timer = tsk->thread.softirq_timer;
 294         lc->steal_timer = 0;
 295 }
 296
 297 static void pcpu_start_fn(struct pcpu *pcpu, void (*func)(void *), void *data)
 298 {
 299         struct lowcore *lc = pcpu->lowcore;
 300
 301         lc->restart_stack = lc->nodat_stack;
 302         lc->restart_fn = (unsigned long) func;
 303         lc->restart_data = (unsigned long) data;
 304         lc->restart_source = -1UL;
 305         pcpu_sigp_retry(pcpu, SIGP_RESTART, 0);
 306 }
 307
 308 /*
 309  * Call function via PSW restart on pcpu and stop the current cpu.
 310  */
 311 static void __pcpu_delegate(void (*func)(void*), void *data)
 312 {
 313         func(data);     /* should not return */
 314 }
 315
 316 static void __no_sanitize_address pcpu_delegate(struct pcpu *pcpu,
 317                                                 void (*func)(void *),
 318                                                 void *data, unsigned long stack)
 319 {
 320         struct lowcore *lc = lowcore_ptr[pcpu - pcpu_devices];
 321         unsigned long source_cpu = stap();
 322
 323         __load_psw_mask(PSW_KERNEL_BITS | PSW_MASK_DAT);
 324         if (pcpu->address == source_cpu)
 325                 CALL_ON_STACK(__pcpu_delegate, stack, 2, func, data);
 326         /* Stop target cpu (if func returns this stops the current cpu). */
 327         pcpu_sigp_retry(pcpu, SIGP_STOP, 0);
 328         /* Restart func on the target cpu and stop the current cpu. */
 329         mem_assign_absolute(lc->restart_stack, stack);
 330         mem_assign_absolute(lc->restart_fn, (unsigned long) func);
 331         mem_assign_absolute(lc->restart_data, (unsigned long) data);
 332         mem_assign_absolute(lc->restart_source, source_cpu);
 333         __bpon();
 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->nodat_stack);
 387 }
 388
 389 int smp_find_processor_id(u16 address)
 390 {
 391         int cpu;
 392
 393         for_each_present_cpu(cpu)
 394                 if (pcpu_devices[cpu].address == address)
 395                         return cpu;
 396         return -1;
 397 }
 398
 399 bool arch_vcpu_is_preempted(int cpu)
 400 {
 401         if (test_cpu_flag_of(CIF_ENABLED_WAIT, cpu))
 402                 return false;
 403         if (pcpu_running(pcpu_devices + cpu))
 404                 return false;
 405         return true;
 406 }
 407 EXPORT_SYMBOL(arch_vcpu_is_preempted);
 408
 409 void smp_yield_cpu(int cpu)
 410 {
 411         if (MACHINE_HAS_DIAG9C) {
 412                 diag_stat_inc_norecursion(DIAG_STAT_X09C);
 413                 asm volatile("diag %0,0,0x9c"
 414                              : : "d" (pcpu_devices[cpu].address));
 415         } else if (MACHINE_HAS_DIAG44) {
 416                 diag_stat_inc_norecursion(DIAG_STAT_X044);
 417                 asm volatile("diag 0,0,0x44");
 418         }
 419 }
 420
 421 /*
 422  * Send cpus emergency shutdown signal. This gives the cpus the
 423  * opportunity to complete outstanding interrupts.
 424  */
 425 void notrace smp_emergency_stop(void)
 426 {
 427         cpumask_t cpumask;
 428         u64 end;
 429         int cpu;
 430
 431         cpumask_copy(&cpumask, cpu_online_mask);
 432         cpumask_clear_cpu(smp_processor_id(), &cpumask);
 433
 434         end = get_tod_clock() + (1000000UL << 12);
 435         for_each_cpu(cpu, &cpumask) {
 436                 struct pcpu *pcpu = pcpu_devices + cpu;
 437                 set_bit(ec_stop_cpu, &pcpu->ec_mask);
 438                 while (__pcpu_sigp(pcpu->address, SIGP_EMERGENCY_SIGNAL,
 439                                    0, NULL) == SIGP_CC_BUSY &&
 440                        get_tod_clock() < end)
 441                         cpu_relax();
 442         }
 443         while (get_tod_clock() < end) {
 444                 for_each_cpu(cpu, &cpumask)
 445                         if (pcpu_stopped(pcpu_devices + cpu))
 446                                 cpumask_clear_cpu(cpu, &cpumask);
 447                 if (cpumask_empty(&cpumask))
 448                         break;
 449                 cpu_relax();
 450         }
 451 }
 452 NOKPROBE_SYMBOL(smp_emergency_stop);
 453
 454 /*
 455  * Stop all cpus but the current one.
 456  */
 457 void smp_send_stop(void)
 458 {
 459         int cpu;
 460
 461         /* Disable all interrupts/machine checks */
 462         __load_psw_mask(PSW_KERNEL_BITS | PSW_MASK_DAT);
 463         trace_hardirqs_off();
 464
 465         debug_set_critical();
 466
 467         if (oops_in_progress)
 468                 smp_emergency_stop();
 469
 470         /* stop all processors */
 471         for_each_online_cpu(cpu) {
 472                 if (cpu == smp_processor_id())
 473                         continue;
 474                 pcpu_sigp_retry(pcpu_devices + cpu, SIGP_STOP, 0);
 475                 while (!pcpu_stopped(pcpu_devices + cpu))
 476                         cpu_relax();
 477         }
 478 }
 479
 480 /*
 481  * This is the main routine where commands issued by other
 482  * cpus are handled.
 483  */
 484 static void smp_handle_ext_call(void)
 485 {
 486         unsigned long bits;
 487
 488         /* handle bit signal external calls */
 489         bits = xchg(&pcpu_devices[smp_processor_id()].ec_mask, 0);
 490         if (test_bit(ec_stop_cpu, &bits))
 491                 smp_stop_cpu();
 492         if (test_bit(ec_schedule, &bits))
 493                 scheduler_ipi();
 494         if (test_bit(ec_call_function_single, &bits))
 495                 generic_smp_call_function_single_interrupt();
 496 }
 497
 498 static void do_ext_call_interrupt(struct ext_code ext_code,
 499                                   unsigned int param32, unsigned long param64)
 500 {
 501         inc_irq_stat(ext_code.code == 0x1202 ? IRQEXT_EXC : IRQEXT_EMS);
 502         smp_handle_ext_call();
 503 }
 504
 505 void arch_send_call_function_ipi_mask(const struct cpumask *mask)
 506 {
 507         int cpu;
 508
 509         for_each_cpu(cpu, mask)
 510                 pcpu_ec_call(pcpu_devices + cpu, ec_call_function_single);
 511 }
 512
 513 void arch_send_call_function_single_ipi(int cpu)
 514 {
 515         pcpu_ec_call(pcpu_devices + cpu, ec_call_function_single);
 516 }
 517
 518 /*
 519  * this function sends a 'reschedule' IPI to another CPU.
 520  * it goes straight through and wastes no time serializing
 521  * anything. Worst case is that we lose a reschedule ...
 522  */
 523 void smp_send_reschedule(int cpu)
 524 {
 525         pcpu_ec_call(pcpu_devices + cpu, ec_schedule);
 526 }
 527
 528 /*
 529  * parameter area for the set/clear control bit callbacks
 530  */
 531 struct ec_creg_mask_parms {
 532         unsigned long orval;
 533         unsigned long andval;
 534         int cr;
 535 };
 536
 537 /*
 538  * callback for setting/clearing control bits
 539  */
 540 static void smp_ctl_bit_callback(void *info)
 541 {
 542         struct ec_creg_mask_parms *pp = info;
 543         unsigned long cregs[16];
 544
 545         __ctl_store(cregs, 0, 15);
 546         cregs[pp->cr] = (cregs[pp->cr] & pp->andval) | pp->orval;
 547         __ctl_load(cregs, 0, 15);
 548 }
 549
 550 /*
 551  * Set a bit in a control register of all cpus
 552  */
 553 void smp_ctl_set_bit(int cr, int bit)
 554 {
 555         struct ec_creg_mask_parms parms = { 1UL << bit, -1UL, cr };
 556
 557         on_each_cpu(smp_ctl_bit_callback, &parms, 1);
 558 }
 559 EXPORT_SYMBOL(smp_ctl_set_bit);
 560
 561 /*
 562  * Clear a bit in a control register of all cpus
 563  */
 564 void smp_ctl_clear_bit(int cr, int bit)
 565 {
 566         struct ec_creg_mask_parms parms = { 0, ~(1UL << bit), cr };
 567
 568         on_each_cpu(smp_ctl_bit_callback, &parms, 1);
 569 }
 570 EXPORT_SYMBOL(smp_ctl_clear_bit);
 571
 572 #ifdef CONFIG_CRASH_DUMP
 573
 574 int smp_store_status(int cpu)
 575 {
 576         struct pcpu *pcpu = pcpu_devices + cpu;
 577         unsigned long pa;
 578
 579         pa = __pa(&pcpu->lowcore->floating_pt_save_area);
 580         if (__pcpu_sigp_relax(pcpu->address, SIGP_STORE_STATUS_AT_ADDRESS,
 581                               pa) != SIGP_CC_ORDER_CODE_ACCEPTED)
 582                 return -EIO;
 583         if (!MACHINE_HAS_VX && !MACHINE_HAS_GS)
 584                 return 0;
 585         pa = __pa(pcpu->lowcore->mcesad & MCESA_ORIGIN_MASK);
 586         if (MACHINE_HAS_GS)
 587                 pa |= pcpu->lowcore->mcesad & MCESA_LC_MASK;
 588         if (__pcpu_sigp_relax(pcpu->address, SIGP_STORE_ADDITIONAL_STATUS,
 589                               pa) != SIGP_CC_ORDER_CODE_ACCEPTED)
 590                 return -EIO;
 591         return 0;
 592 }
 593
 594 /*
 595  * Collect CPU state of the previous, crashed system.
 596  * There are four cases:
 597  * 1) standard zfcp dump
 598  *    condition: OLDMEM_BASE == NULL && ipl_info.type == IPL_TYPE_FCP_DUMP
 599  *    The state for all CPUs except the boot CPU needs to be collected
 600  *    with sigp stop-and-store-status. The boot CPU state is located in
 601  *    the absolute lowcore of the memory stored in the HSA. The zcore code
 602  *    will copy the boot CPU state from the HSA.
 603  * 2) stand-alone kdump for SCSI (zfcp dump with swapped memory)
 604  *    condition: OLDMEM_BASE != NULL && ipl_info.type == IPL_TYPE_FCP_DUMP
 605  *    The state for all CPUs except the boot CPU needs to be collected
 606  *    with sigp stop-and-store-status. The firmware or the boot-loader
 607  *    stored the registers of the boot CPU in the absolute lowcore in the
 608  *    memory of the old system.
 609  * 3) kdump and the old kernel did not store the CPU state,
 610  *    or stand-alone kdump for DASD
 611  *    condition: OLDMEM_BASE != NULL && !is_kdump_kernel()
 612  *    The state for all CPUs except the boot CPU needs to be collected
 613  *    with sigp stop-and-store-status. The kexec code or the boot-loader
 614  *    stored the registers of the boot CPU in the memory of the old system.
 615  * 4) kdump and the old kernel stored the CPU state
 616  *    condition: OLDMEM_BASE != NULL && is_kdump_kernel()
 617  *    This case does not exist for s390 anymore, setup_arch explicitly
 618  *    deactivates the elfcorehdr= kernel parameter
 619  */
 620 static __init void smp_save_cpu_vxrs(struct save_area *sa, u16 addr,
 621                                      bool is_boot_cpu, unsigned long page)
 622 {
 623         __vector128 *vxrs = (__vector128 *) page;
 624
 625         if (is_boot_cpu)
 626                 vxrs = boot_cpu_vector_save_area;
 627         else
 628                 __pcpu_sigp_relax(addr, SIGP_STORE_ADDITIONAL_STATUS, page);
 629         save_area_add_vxrs(sa, vxrs);
 630 }
 631
 632 static __init void smp_save_cpu_regs(struct save_area *sa, u16 addr,
 633                                      bool is_boot_cpu, unsigned long page)
 634 {
 635         void *regs = (void *) page;
 636
 637         if (is_boot_cpu)
 638                 copy_oldmem_kernel(regs, (void *) __LC_FPREGS_SAVE_AREA, 512);
 639         else
 640                 __pcpu_sigp_relax(addr, SIGP_STORE_STATUS_AT_ADDRESS, page);
 641         save_area_add_regs(sa, regs);
 642 }
 643
 644 void __init smp_save_dump_cpus(void)
 645 {
 646         int addr, boot_cpu_addr, max_cpu_addr;
 647         struct save_area *sa;
 648         unsigned long page;
 649         bool is_boot_cpu;
 650
 651         if (!(OLDMEM_BASE || ipl_info.type == IPL_TYPE_FCP_DUMP))
 652                 /* No previous system present, normal boot. */
 653                 return;
 654         /* Allocate a page as dumping area for the store status sigps */
 655         page = memblock_alloc_base(PAGE_SIZE, PAGE_SIZE, 1UL << 31);
 656         /* Set multi-threading state to the previous system. */
 657         pcpu_set_smt(sclp.mtid_prev);
 658         boot_cpu_addr = stap();
 659         max_cpu_addr = SCLP_MAX_CORES << sclp.mtid_prev;
 660         for (addr = 0; addr <= max_cpu_addr; addr++) {
 661                 if (__pcpu_sigp_relax(addr, SIGP_SENSE, 0) ==
 662                     SIGP_CC_NOT_OPERATIONAL)
 663                         continue;
 664                 is_boot_cpu = (addr == boot_cpu_addr);
 665                 /* Allocate save area */
 666                 sa = save_area_alloc(is_boot_cpu);
 667                 if (!sa)
 668                         panic("could not allocate memory for save area\n");
 669                 if (MACHINE_HAS_VX)
 670                         /* Get the vector registers */
 671                         smp_save_cpu_vxrs(sa, addr, is_boot_cpu, page);
 672                 /*
 673                  * For a zfcp dump OLDMEM_BASE == NULL and the registers
 674                  * of the boot CPU are stored in the HSA. To retrieve
 675                  * these registers an SCLP request is required which is
 676                  * done by drivers/s390/char/zcore.c:init_cpu_info()
 677                  */
 678                 if (!is_boot_cpu || OLDMEM_BASE)
 679                         /* Get the CPU registers */
 680                         smp_save_cpu_regs(sa, addr, is_boot_cpu, page);
 681         }
 682         memblock_free(page, PAGE_SIZE);
 683         diag308_reset();
 684         pcpu_set_smt(0);
 685 }
 686 #endif /* CONFIG_CRASH_DUMP */
 687
 688 void smp_cpu_set_polarization(int cpu, int val)
 689 {
 690         pcpu_devices[cpu].polarization = val;
 691 }
 692
 693 int smp_cpu_get_polarization(int cpu)
 694 {
 695         return pcpu_devices[cpu].polarization;
 696 }
 697
 698 static void __ref smp_get_core_info(struct sclp_core_info *info, int early)
 699 {
 700         static int use_sigp_detection;
 701         int address;
 702
 703         if (use_sigp_detection || sclp_get_core_info(info, early)) {
 704                 use_sigp_detection = 1;
 705                 for (address = 0;
 706                      address < (SCLP_MAX_CORES << smp_cpu_mt_shift);
 707                      address += (1U << smp_cpu_mt_shift)) {
 708                         if (__pcpu_sigp_relax(address, SIGP_SENSE, 0) ==
 709                             SIGP_CC_NOT_OPERATIONAL)
 710                                 continue;
 711                         info->core[info->configured].core_id =
 712                                 address >> smp_cpu_mt_shift;
 713                         info->configured++;
 714                 }
 715                 info->combined = info->configured;
 716         }
 717 }
 718
 719 static int smp_add_present_cpu(int cpu);
 720
 721 static int __smp_rescan_cpus(struct sclp_core_info *info, int sysfs_add)
 722 {
 723         struct pcpu *pcpu;
 724         cpumask_t avail;
 725         int cpu, nr, i, j;
 726         u16 address;
 727
 728         nr = 0;
 729         cpumask_xor(&avail, cpu_possible_mask, cpu_present_mask);
 730         cpu = cpumask_first(&avail);
 731         for (i = 0; (i < info->combined) && (cpu < nr_cpu_ids); i++) {
 732                 if (sclp.has_core_type && info->core[i].type != boot_core_type)
 733                         continue;
 734                 address = info->core[i].core_id << smp_cpu_mt_shift;
 735                 for (j = 0; j <= smp_cpu_mtid; j++) {
 736                         if (pcpu_find_address(cpu_present_mask, address + j))
 737                                 continue;
 738                         pcpu = pcpu_devices + cpu;
 739                         pcpu->address = address + j;
 740                         pcpu->state =
 741                                 (cpu >= info->configured*(smp_cpu_mtid + 1)) ?
 742                                 CPU_STATE_STANDBY : CPU_STATE_CONFIGURED;
 743                         smp_cpu_set_polarization(cpu, POLARIZATION_UNKNOWN);
 744                         set_cpu_present(cpu, true);
 745                         if (sysfs_add && smp_add_present_cpu(cpu) != 0)
 746                                 set_cpu_present(cpu, false);
 747                         else
 748                                 nr++;
 749                         cpu = cpumask_next(cpu, &avail);
 750                         if (cpu >= nr_cpu_ids)
 751                                 break;
 752                 }
 753         }
 754         return nr;
 755 }
 756
 757 void __init smp_detect_cpus(void)
 758 {
 759         unsigned int cpu, mtid, c_cpus, s_cpus;
 760         struct sclp_core_info *info;
 761         u16 address;
 762
 763         /* Get CPU information */
 764         info = memblock_virt_alloc(sizeof(*info), 8);
 765         smp_get_core_info(info, 1);
 766         /* Find boot CPU type */
 767         if (sclp.has_core_type) {
 768                 address = stap();
 769                 for (cpu = 0; cpu < info->combined; cpu++)
 770                         if (info->core[cpu].core_id == address) {
 771                                 /* The boot cpu dictates the cpu type. */
 772                                 boot_core_type = info->core[cpu].type;
 773                                 break;
 774                         }
 775                 if (cpu >= info->combined)
 776                         panic("Could not find boot CPU type");
 777         }
 778
 779         /* Set multi-threading state for the current system */
 780         mtid = boot_core_type ? sclp.mtid : sclp.mtid_cp;
 781         mtid = (mtid < smp_max_threads) ? mtid : smp_max_threads - 1;
 782         pcpu_set_smt(mtid);
 783
 784         /* Print number of CPUs */
 785         c_cpus = s_cpus = 0;
 786         for (cpu = 0; cpu < info->combined; cpu++) {
 787                 if (sclp.has_core_type &&
 788                     info->core[cpu].type != boot_core_type)
 789                         continue;
 790                 if (cpu < info->configured)
 791                         c_cpus += smp_cpu_mtid + 1;
 792                 else
 793                         s_cpus += smp_cpu_mtid + 1;
 794         }
 795         pr_info("%d configured CPUs, %d standby CPUs\n", c_cpus, s_cpus);
 796
 797         /* Add CPUs present at boot */
 798         get_online_cpus();
 799         __smp_rescan_cpus(info, 0);
 800         put_online_cpus();
 801         memblock_free_early((unsigned long)info, sizeof(*info));
 802 }
 803
 804 static void smp_init_secondary(void)
 805 {
 806         int cpu = smp_processor_id();
 807
 808         S390_lowcore.last_update_clock = get_tod_clock();
 809         restore_access_regs(S390_lowcore.access_regs_save_area);
 810         cpu_init();
 811         preempt_disable();
 812         init_cpu_timer();
 813         vtime_init();
 814         pfault_init();
 815         notify_cpu_starting(smp_processor_id());
 816         if (topology_cpu_dedicated(cpu))
 817                 set_cpu_flag(CIF_DEDICATED_CPU);
 818         else
 819                 clear_cpu_flag(CIF_DEDICATED_CPU);
 820         set_cpu_online(smp_processor_id(), true);
 821         inc_irq_stat(CPU_RST);
 822         local_irq_enable();
 823         cpu_startup_entry(CPUHP_AP_ONLINE_IDLE);
 824 }
 825
 826 /*
 827  *      Activate a secondary processor.
 828  */
 829 static void __no_sanitize_address smp_start_secondary(void *cpuvoid)
 830 {
 831         S390_lowcore.restart_stack = (unsigned long) restart_stack;
 832         S390_lowcore.restart_fn = (unsigned long) do_restart;
 833         S390_lowcore.restart_data = 0;
 834         S390_lowcore.restart_source = -1UL;
 835         __ctl_load(S390_lowcore.cregs_save_area, 0, 15);
 836         __load_psw_mask(PSW_KERNEL_BITS | PSW_MASK_DAT);
 837         CALL_ON_STACK(smp_init_secondary, S390_lowcore.kernel_stack, 0);
 838 }
 839
 840 /* Upping and downing of CPUs */
 841 int __cpu_up(unsigned int cpu, struct task_struct *tidle)
 842 {
 843         struct pcpu *pcpu;
 844         int base, i, rc;
 845
 846         pcpu = pcpu_devices + cpu;
 847         if (pcpu->state != CPU_STATE_CONFIGURED)
 848                 return -EIO;
 849         base = smp_get_base_cpu(cpu);
 850         for (i = 0; i <= smp_cpu_mtid; i++) {
 851                 if (base + i < nr_cpu_ids)
 852                         if (cpu_online(base + i))
 853                                 break;
 854         }
 855         /*
 856          * If this is the first CPU of the core to get online
 857          * do an initial CPU reset.
 858          */
 859         if (i > smp_cpu_mtid &&
 860             pcpu_sigp_retry(pcpu_devices + base, SIGP_INITIAL_CPU_RESET, 0) !=
 861             SIGP_CC_ORDER_CODE_ACCEPTED)
 862                 return -EIO;
 863
 864         rc = pcpu_alloc_lowcore(pcpu, cpu);
 865         if (rc)
 866                 return rc;
 867         pcpu_prepare_secondary(pcpu, cpu);
 868         pcpu_attach_task(pcpu, tidle);
 869         pcpu_start_fn(pcpu, smp_start_secondary, NULL);
 870         /* Wait until cpu puts itself in the online & active maps */
 871         while (!cpu_online(cpu))
 872                 cpu_relax();
 873         return 0;
 874 }
 875
 876 static unsigned int setup_possible_cpus __initdata;
 877
 878 static int __init _setup_possible_cpus(char *s)
 879 {
 880         get_option(&s, &setup_possible_cpus);
 881         return 0;
 882 }
 883 early_param("possible_cpus", _setup_possible_cpus);
 884
 885 #ifdef CONFIG_HOTPLUG_CPU
 886
 887 int __cpu_disable(void)
 888 {
 889         unsigned long cregs[16];
 890
 891         /* Handle possible pending IPIs */
 892         smp_handle_ext_call();
 893         set_cpu_online(smp_processor_id(), false);
 894         /* Disable pseudo page faults on this cpu. */
 895         pfault_fini();
 896         /* Disable interrupt sources via control register. */
 897         __ctl_store(cregs, 0, 15);
 898         cregs[0]  &= ~0x0000ee70UL;     /* disable all external interrupts */
 899         cregs[6]  &= ~0xff000000UL;     /* disable all I/O interrupts */
 900         cregs[14] &= ~0x1f000000UL;     /* disable most machine checks */
 901         __ctl_load(cregs, 0, 15);
 902         clear_cpu_flag(CIF_NOHZ_DELAY);
 903         return 0;
 904 }
 905
 906 void __cpu_die(unsigned int cpu)
 907 {
 908         struct pcpu *pcpu;
 909
 910         /* Wait until target cpu is down */
 911         pcpu = pcpu_devices + cpu;
 912         while (!pcpu_stopped(pcpu))
 913                 cpu_relax();
 914         pcpu_free_lowcore(pcpu);
 915         cpumask_clear_cpu(cpu, mm_cpumask(&init_mm));
 916         cpumask_clear_cpu(cpu, &init_mm.context.cpu_attach_mask);
 917 }
 918
 919 void __noreturn cpu_die(void)
 920 {
 921         idle_task_exit();
 922         __bpon();
 923         pcpu_sigp_retry(pcpu_devices + smp_processor_id(), SIGP_STOP, 0);
 924         for (;;) ;
 925 }
 926
 927 #endif /* CONFIG_HOTPLUG_CPU */
 928
 929 void __init smp_fill_possible_mask(void)
 930 {
 931         unsigned int possible, sclp_max, cpu;
 932
 933         sclp_max = max(sclp.mtid, sclp.mtid_cp) + 1;
 934         sclp_max = min(smp_max_threads, sclp_max);
 935         sclp_max = (sclp.max_cores * sclp_max) ?: nr_cpu_ids;
 936         possible = setup_possible_cpus ?: nr_cpu_ids;
 937         possible = min(possible, sclp_max);
 938         for (cpu = 0; cpu < possible && cpu < nr_cpu_ids; cpu++)
 939                 set_cpu_possible(cpu, true);
 940 }
 941
 942 void __init smp_prepare_cpus(unsigned int max_cpus)
 943 {
 944         /* request the 0x1201 emergency signal external interrupt */
 945         if (register_external_irq(EXT_IRQ_EMERGENCY_SIG, do_ext_call_interrupt))
 946                 panic("Couldn't request external interrupt 0x1201");
 947         /* request the 0x1202 external call external interrupt */
 948         if (register_external_irq(EXT_IRQ_EXTERNAL_CALL, do_ext_call_interrupt))
 949                 panic("Couldn't request external interrupt 0x1202");
 950 }
 951
 952 void __init smp_prepare_boot_cpu(void)
 953 {
 954         struct pcpu *pcpu = pcpu_devices;
 955
 956         WARN_ON(!cpu_present(0) || !cpu_online(0));
 957         pcpu->state = CPU_STATE_CONFIGURED;
 958         pcpu->lowcore = (struct lowcore *)(unsigned long) store_prefix();
 959         S390_lowcore.percpu_offset = __per_cpu_offset[0];
 960         smp_cpu_set_polarization(0, POLARIZATION_UNKNOWN);
 961 }
 962
 963 void __init smp_cpus_done(unsigned int max_cpus)
 964 {
 965 }
 966
 967 void __init smp_setup_processor_id(void)
 968 {
 969         pcpu_devices[0].address = stap();
 970         S390_lowcore.cpu_nr = 0;
 971         S390_lowcore.spinlock_lockval = arch_spin_lockval(0);
 972         S390_lowcore.spinlock_index = 0;
 973 }
 974
 975 /*
 976  * the frequency of the profiling timer can be changed
 977  * by writing a multiplier value into /proc/profile.
 978  *
 979  * usually you want to run this on all CPUs ;)
 980  */
 981 int setup_profiling_timer(unsigned int multiplier)
 982 {
 983         return 0;
 984 }
 985
 986 #ifdef CONFIG_HOTPLUG_CPU
 987 static ssize_t cpu_configure_show(struct device *dev,
 988                                   struct device_attribute *attr, char *buf)
 989 {
 990         ssize_t count;
 991
 992         mutex_lock(&smp_cpu_state_mutex);
 993         count = sprintf(buf, "%d\n", pcpu_devices[dev->id].state);
 994         mutex_unlock(&smp_cpu_state_mutex);
 995         return count;
 996 }
 997
 998 static ssize_t cpu_configure_store(struct device *dev,
 999                                    struct device_attribute *attr,
1000                                    const char *buf, size_t count)
1001 {
1002         struct pcpu *pcpu;
1003         int cpu, val, rc, i;
1004         char delim;
1005
1006         if (sscanf(buf, "%d %c", &val, &delim) != 1)
1007                 return -EINVAL;
1008         if (val != 0 && val != 1)
1009                 return -EINVAL;
1010         get_online_cpus();
1011         mutex_lock(&smp_cpu_state_mutex);
1012         rc = -EBUSY;
1013         /* disallow configuration changes of online cpus and cpu 0 */
1014         cpu = dev->id;
1015         cpu = smp_get_base_cpu(cpu);
1016         if (cpu == 0)
1017                 goto out;
1018         for (i = 0; i <= smp_cpu_mtid; i++)
1019                 if (cpu_online(cpu + i))
1020                         goto out;
1021         pcpu = pcpu_devices + cpu;
1022         rc = 0;
1023         switch (val) {
1024         case 0:
1025                 if (pcpu->state != CPU_STATE_CONFIGURED)
1026                         break;
1027                 rc = sclp_core_deconfigure(pcpu->address >> smp_cpu_mt_shift);
1028                 if (rc)
1029                         break;
1030                 for (i = 0; i <= smp_cpu_mtid; i++) {
1031                         if (cpu + i >= nr_cpu_ids || !cpu_present(cpu + i))
1032                                 continue;
1033                         pcpu[i].state = CPU_STATE_STANDBY;
1034                         smp_cpu_set_polarization(cpu + i,
1035                                                  POLARIZATION_UNKNOWN);
1036                 }
1037                 topology_expect_change();
1038                 break;
1039         case 1:
1040                 if (pcpu->state != CPU_STATE_STANDBY)
1041                         break;
1042                 rc = sclp_core_configure(pcpu->address >> smp_cpu_mt_shift);
1043                 if (rc)
1044                         break;
1045                 for (i = 0; i <= smp_cpu_mtid; i++) {
1046                         if (cpu + i >= nr_cpu_ids || !cpu_present(cpu + i))
1047                                 continue;
1048                         pcpu[i].state = CPU_STATE_CONFIGURED;
1049                         smp_cpu_set_polarization(cpu + i,
1050                                                  POLARIZATION_UNKNOWN);
1051                 }
1052                 topology_expect_change();
1053                 break;
1054         default:
1055                 break;
1056         }
1057 out:
1058         mutex_unlock(&smp_cpu_state_mutex);
1059         put_online_cpus();
1060         return rc ? rc : count;
1061 }
1062 static DEVICE_ATTR(configure, 0644, cpu_configure_show, cpu_configure_store);
1063 #endif /* CONFIG_HOTPLUG_CPU */
1064
1065 static ssize_t show_cpu_address(struct device *dev,
1066                                 struct device_attribute *attr, char *buf)
1067 {
1068         return sprintf(buf, "%d\n", pcpu_devices[dev->id].address);
1069 }
1070 static DEVICE_ATTR(address, 0444, show_cpu_address, NULL);
1071
1072 static struct attribute *cpu_common_attrs[] = {
1073 #ifdef CONFIG_HOTPLUG_CPU
1074         &dev_attr_configure.attr,
1075 #endif
1076         &dev_attr_address.attr,
1077         NULL,
1078 };
1079
1080 static struct attribute_group cpu_common_attr_group = {
1081         .attrs = cpu_common_attrs,
1082 };
1083
1084 static struct attribute *cpu_online_attrs[] = {
1085         &dev_attr_idle_count.attr,
1086         &dev_attr_idle_time_us.attr,
1087         NULL,
1088 };
1089
1090 static struct attribute_group cpu_online_attr_group = {
1091         .attrs = cpu_online_attrs,
1092 };
1093
1094 static int smp_cpu_online(unsigned int cpu)
1095 {
1096         struct device *s = &per_cpu(cpu_device, cpu)->dev;
1097
1098         return sysfs_create_group(&s->kobj, &cpu_online_attr_group);
1099 }
1100 static int smp_cpu_pre_down(unsigned int cpu)
1101 {
1102         struct device *s = &per_cpu(cpu_device, cpu)->dev;
1103
1104         sysfs_remove_group(&s->kobj, &cpu_online_attr_group);
1105         return 0;
1106 }
1107
1108 static int smp_add_present_cpu(int cpu)
1109 {
1110         struct device *s;
1111         struct cpu *c;
1112         int rc;
1113
1114         c = kzalloc(sizeof(*c), GFP_KERNEL);
1115         if (!c)
1116                 return -ENOMEM;
1117         per_cpu(cpu_device, cpu) = c;
1118         s = &c->dev;
1119         c->hotpluggable = 1;
1120         rc = register_cpu(c, cpu);
1121         if (rc)
1122                 goto out;
1123         rc = sysfs_create_group(&s->kobj, &cpu_common_attr_group);
1124         if (rc)
1125                 goto out_cpu;
1126         rc = topology_cpu_init(c);
1127         if (rc)
1128                 goto out_topology;
1129         return 0;
1130
1131 out_topology:
1132         sysfs_remove_group(&s->kobj, &cpu_common_attr_group);
1133 out_cpu:
1134 #ifdef CONFIG_HOTPLUG_CPU
1135         unregister_cpu(c);
1136 #endif
1137 out:
1138         return rc;
1139 }
1140
1141 #ifdef CONFIG_HOTPLUG_CPU
1142
1143 int __ref smp_rescan_cpus(void)
1144 {
1145         struct sclp_core_info *info;
1146         int nr;
1147
1148         info = kzalloc(sizeof(*info), GFP_KERNEL);
1149         if (!info)
1150                 return -ENOMEM;
1151         smp_get_core_info(info, 0);
1152         get_online_cpus();
1153         mutex_lock(&smp_cpu_state_mutex);
1154         nr = __smp_rescan_cpus(info, 1);
1155         mutex_unlock(&smp_cpu_state_mutex);
1156         put_online_cpus();
1157         kfree(info);
1158         if (nr)
1159                 topology_schedule_update();
1160         return 0;
1161 }
1162
1163 static ssize_t __ref rescan_store(struct device *dev,
1164                                   struct device_attribute *attr,
1165                                   const char *buf,
1166                                   size_t count)
1167 {
1168         int rc;
1169
1170         rc = smp_rescan_cpus();
1171         return rc ? rc : count;
1172 }
1173 static DEVICE_ATTR_WO(rescan);
1174 #endif /* CONFIG_HOTPLUG_CPU */
1175
1176 static int __init s390_smp_init(void)
1177 {
1178         int cpu, rc = 0;
1179
1180 #ifdef CONFIG_HOTPLUG_CPU
1181         rc = device_create_file(cpu_subsys.dev_root, &dev_attr_rescan);
1182         if (rc)
1183                 return rc;
1184 #endif
1185         for_each_present_cpu(cpu) {
1186                 rc = smp_add_present_cpu(cpu);
1187                 if (rc)
1188                         goto out;
1189         }
1190
1191         rc = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "s390/smp:online",
1192                                smp_cpu_online, smp_cpu_pre_down);
1193         rc = rc <= 0 ? rc : 0;
1194 out:
1195         return rc;
1196 }
1197 subsys_initcall(s390_smp_init);