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  *
   9  *  based on other smp stuff by
  10  *    (c) 1995 Alan Cox, CymruNET Ltd  <alan@cymru.net>
  11  *    (c) 1998 Ingo Molnar
  12  *
  13  * The code outside of smp.c uses logical cpu numbers, only smp.c does
  14  * the translation of logical to physical cpu ids. All new code that
  15  * operates on physical cpu numbers needs to go into smp.c.
  16  */
  17
  18 #define KMSG_COMPONENT "cpu"
  19 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
  20
  21 #include <linux/workqueue.h>
  22 #include <linux/memblock.h>
  23 #include <linux/export.h>
  24 #include <linux/init.h>
  25 #include <linux/mm.h>
  26 #include <linux/err.h>
  27 #include <linux/spinlock.h>
  28 #include <linux/kernel_stat.h>
  29 #include <linux/delay.h>
  30 #include <linux/interrupt.h>
  31 #include <linux/irqflags.h>
  32 #include <linux/irq_work.h>
  33 #include <linux/cpu.h>
  34 #include <linux/slab.h>
  35 #include <linux/sched/hotplug.h>
  36 #include <linux/sched/task_stack.h>
  37 #include <linux/crash_dump.h>
  38 #include <linux/kprobes.h>
  39 #include <asm/asm-offsets.h>
  40 #include <asm/ctlreg.h>
  41 #include <asm/pfault.h>
  42 #include <asm/diag.h>
  43 #include <asm/switch_to.h>
  44 #include <asm/facility.h>
  45 #include <asm/ipl.h>
  46 #include <asm/setup.h>
  47 #include <asm/irq.h>
  48 #include <asm/tlbflush.h>
  49 #include <asm/vtimer.h>
  50 #include <asm/abs_lowcore.h>
  51 #include <asm/sclp.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/stacktrace.h>
  58 #include <asm/topology.h>
  59 #include <asm/vdso.h>
  60 #include <asm/maccess.h>
  61 #include "entry.h"
  62
  63 enum {
  64         ec_schedule = 0,
  65         ec_call_function_single,
  66         ec_stop_cpu,
  67         ec_mcck_pending,
  68         ec_irq_work,
  69 };
  70
  71 enum {
  72         CPU_STATE_STANDBY,
  73         CPU_STATE_CONFIGURED,
  74 };
  75
  76 static DEFINE_PER_CPU(struct cpu *, cpu_device);
  77
  78 struct pcpu {
  79         unsigned long ec_mask;          /* bit mask for ec_xxx functions */
  80         unsigned long ec_clk;           /* sigp timestamp for ec_xxx */
  81         signed char state;              /* physical cpu state */
  82         signed char polarization;       /* physical polarization */
  83         u16 address;                    /* physical cpu address */
  84 };
  85
  86 static u8 boot_core_type;
  87 static struct pcpu pcpu_devices[NR_CPUS];
  88
  89 unsigned int smp_cpu_mt_shift;
  90 EXPORT_SYMBOL(smp_cpu_mt_shift);
  91
  92 unsigned int smp_cpu_mtid;
  93 EXPORT_SYMBOL(smp_cpu_mtid);
  94
  95 #ifdef CONFIG_CRASH_DUMP
  96 __vector128 __initdata boot_cpu_vector_save_area[__NUM_VXRS];
  97 #endif
  98
  99 static unsigned int smp_max_threads __initdata = -1U;
 100 cpumask_t cpu_setup_mask;
 101
 102 static int __init early_nosmt(char *s)
 103 {
 104         smp_max_threads = 1;
 105         return 0;
 106 }
 107 early_param("nosmt", early_nosmt);
 108
 109 static int __init early_smt(char *s)
 110 {
 111         get_option(&s, &smp_max_threads);
 112         return 0;
 113 }
 114 early_param("smt", early_smt);
 115
 116 /*
 117  * The smp_cpu_state_mutex must be held when changing the state or polarization
 118  * member of a pcpu data structure within the pcpu_devices array.
 119  */
 120 DEFINE_MUTEX(smp_cpu_state_mutex);
 121
 122 /*
 123  * Signal processor helper functions.
 124  */
 125 static inline int __pcpu_sigp_relax(u16 addr, u8 order, unsigned long parm)
 126 {
 127         int cc;
 128
 129         while (1) {
 130                 cc = __pcpu_sigp(addr, order, parm, NULL);
 131                 if (cc != SIGP_CC_BUSY)
 132                         return cc;
 133                 cpu_relax();
 134         }
 135 }
 136
 137 static int pcpu_sigp_retry(struct pcpu *pcpu, u8 order, u32 parm)
 138 {
 139         int cc, retry;
 140
 141         for (retry = 0; ; retry++) {
 142                 cc = __pcpu_sigp(pcpu->address, order, parm, NULL);
 143                 if (cc != SIGP_CC_BUSY)
 144                         break;
 145                 if (retry >= 3)
 146                         udelay(10);
 147         }
 148         return cc;
 149 }
 150
 151 static inline int pcpu_stopped(struct pcpu *pcpu)
 152 {
 153         u32 status;
 154
 155         if (__pcpu_sigp(pcpu->address, SIGP_SENSE,
 156                         0, &status) != SIGP_CC_STATUS_STORED)
 157                 return 0;
 158         return !!(status & (SIGP_STATUS_CHECK_STOP|SIGP_STATUS_STOPPED));
 159 }
 160
 161 static inline int pcpu_running(struct pcpu *pcpu)
 162 {
 163         if (__pcpu_sigp(pcpu->address, SIGP_SENSE_RUNNING,
 164                         0, NULL) != SIGP_CC_STATUS_STORED)
 165                 return 1;
 166         /* Status stored condition code is equivalent to cpu not running. */
 167         return 0;
 168 }
 169
 170 /*
 171  * Find struct pcpu by cpu address.
 172  */
 173 static struct pcpu *pcpu_find_address(const struct cpumask *mask, u16 address)
 174 {
 175         int cpu;
 176
 177         for_each_cpu(cpu, mask)
 178                 if (pcpu_devices[cpu].address == address)
 179                         return pcpu_devices + cpu;
 180         return NULL;
 181 }
 182
 183 static void pcpu_ec_call(struct pcpu *pcpu, int ec_bit)
 184 {
 185         int order;
 186
 187         if (test_and_set_bit(ec_bit, &pcpu->ec_mask))
 188                 return;
 189         order = pcpu_running(pcpu) ? SIGP_EXTERNAL_CALL : SIGP_EMERGENCY_SIGNAL;
 190         pcpu->ec_clk = get_tod_clock_fast();
 191         pcpu_sigp_retry(pcpu, order, 0);
 192 }
 193
 194 static int pcpu_alloc_lowcore(struct pcpu *pcpu, int cpu)
 195 {
 196         unsigned long async_stack, nodat_stack, mcck_stack;
 197         struct lowcore *lc;
 198
 199         lc = (struct lowcore *) __get_free_pages(GFP_KERNEL | GFP_DMA, LC_ORDER);
 200         nodat_stack = __get_free_pages(GFP_KERNEL, THREAD_SIZE_ORDER);
 201         async_stack = stack_alloc();
 202         mcck_stack = stack_alloc();
 203         if (!lc || !nodat_stack || !async_stack || !mcck_stack)
 204                 goto out;
 205         memcpy(lc, &S390_lowcore, 512);
 206         memset((char *) lc + 512, 0, sizeof(*lc) - 512);
 207         lc->async_stack = async_stack + STACK_INIT_OFFSET;
 208         lc->nodat_stack = nodat_stack + STACK_INIT_OFFSET;
 209         lc->mcck_stack = mcck_stack + STACK_INIT_OFFSET;
 210         lc->cpu_nr = cpu;
 211         lc->spinlock_lockval = arch_spin_lockval(cpu);
 212         lc->spinlock_index = 0;
 213         lc->return_lpswe = gen_lpswe(__LC_RETURN_PSW);
 214         lc->return_mcck_lpswe = gen_lpswe(__LC_RETURN_MCCK_PSW);
 215         lc->preempt_count = PREEMPT_DISABLED;
 216         if (nmi_alloc_mcesa(&lc->mcesad))
 217                 goto out;
 218         if (abs_lowcore_map(cpu, lc, true))
 219                 goto out_mcesa;
 220         lowcore_ptr[cpu] = lc;
 221         pcpu_sigp_retry(pcpu, SIGP_SET_PREFIX, __pa(lc));
 222         return 0;
 223
 224 out_mcesa:
 225         nmi_free_mcesa(&lc->mcesad);
 226 out:
 227         stack_free(mcck_stack);
 228         stack_free(async_stack);
 229         free_pages(nodat_stack, THREAD_SIZE_ORDER);
 230         free_pages((unsigned long) lc, LC_ORDER);
 231         return -ENOMEM;
 232 }
 233
 234 static void pcpu_free_lowcore(struct pcpu *pcpu)
 235 {
 236         unsigned long async_stack, nodat_stack, mcck_stack;
 237         struct lowcore *lc;
 238         int cpu;
 239
 240         cpu = pcpu - pcpu_devices;
 241         lc = lowcore_ptr[cpu];
 242         nodat_stack = lc->nodat_stack - STACK_INIT_OFFSET;
 243         async_stack = lc->async_stack - STACK_INIT_OFFSET;
 244         mcck_stack = lc->mcck_stack - STACK_INIT_OFFSET;
 245         pcpu_sigp_retry(pcpu, SIGP_SET_PREFIX, 0);
 246         lowcore_ptr[cpu] = NULL;
 247         abs_lowcore_unmap(cpu);
 248         nmi_free_mcesa(&lc->mcesad);
 249         stack_free(async_stack);
 250         stack_free(mcck_stack);
 251         free_pages(nodat_stack, THREAD_SIZE_ORDER);
 252         free_pages((unsigned long) lc, LC_ORDER);
 253 }
 254
 255 static void pcpu_prepare_secondary(struct pcpu *pcpu, int cpu)
 256 {
 257         struct lowcore *lc, *abs_lc;
 258
 259         lc = lowcore_ptr[cpu];
 260         cpumask_set_cpu(cpu, &init_mm.context.cpu_attach_mask);
 261         cpumask_set_cpu(cpu, mm_cpumask(&init_mm));
 262         lc->cpu_nr = cpu;
 263         lc->restart_flags = RESTART_FLAG_CTLREGS;
 264         lc->spinlock_lockval = arch_spin_lockval(cpu);
 265         lc->spinlock_index = 0;
 266         lc->percpu_offset = __per_cpu_offset[cpu];
 267         lc->kernel_asce = S390_lowcore.kernel_asce;
 268         lc->user_asce = s390_invalid_asce;
 269         lc->machine_flags = S390_lowcore.machine_flags;
 270         lc->user_timer = lc->system_timer =
 271                 lc->steal_timer = lc->avg_steal_timer = 0;
 272         abs_lc = get_abs_lowcore();
 273         memcpy(lc->cregs_save_area, abs_lc->cregs_save_area, sizeof(lc->cregs_save_area));
 274         put_abs_lowcore(abs_lc);
 275         lc->cregs_save_area[1] = lc->kernel_asce;
 276         lc->cregs_save_area[7] = lc->user_asce;
 277         save_access_regs((unsigned int *) lc->access_regs_save_area);
 278         arch_spin_lock_setup(cpu);
 279 }
 280
 281 static void pcpu_attach_task(struct pcpu *pcpu, struct task_struct *tsk)
 282 {
 283         struct lowcore *lc;
 284         int cpu;
 285
 286         cpu = pcpu - pcpu_devices;
 287         lc = lowcore_ptr[cpu];
 288         lc->kernel_stack = (unsigned long)task_stack_page(tsk) + STACK_INIT_OFFSET;
 289         lc->current_task = (unsigned long)tsk;
 290         lc->lpp = LPP_MAGIC;
 291         lc->current_pid = tsk->pid;
 292         lc->user_timer = tsk->thread.user_timer;
 293         lc->guest_timer = tsk->thread.guest_timer;
 294         lc->system_timer = tsk->thread.system_timer;
 295         lc->hardirq_timer = tsk->thread.hardirq_timer;
 296         lc->softirq_timer = tsk->thread.softirq_timer;
 297         lc->steal_timer = 0;
 298 }
 299
 300 static void pcpu_start_fn(struct pcpu *pcpu, void (*func)(void *), void *data)
 301 {
 302         struct lowcore *lc;
 303         int cpu;
 304
 305         cpu = pcpu - pcpu_devices;
 306         lc = lowcore_ptr[cpu];
 307         lc->restart_stack = lc->kernel_stack;
 308         lc->restart_fn = (unsigned long) func;
 309         lc->restart_data = (unsigned long) data;
 310         lc->restart_source = -1U;
 311         pcpu_sigp_retry(pcpu, SIGP_RESTART, 0);
 312 }
 313
 314 typedef void (pcpu_delegate_fn)(void *);
 315
 316 /*
 317  * Call function via PSW restart on pcpu and stop the current cpu.
 318  */
 319 static void __pcpu_delegate(pcpu_delegate_fn *func, void *data)
 320 {
 321         func(data);     /* should not return */
 322 }
 323
 324 static void pcpu_delegate(struct pcpu *pcpu,
 325                           pcpu_delegate_fn *func,
 326                           void *data, unsigned long stack)
 327 {
 328         struct lowcore *lc, *abs_lc;
 329         unsigned int source_cpu;
 330
 331         lc = lowcore_ptr[pcpu - pcpu_devices];
 332         source_cpu = stap();
 333
 334         if (pcpu->address == source_cpu) {
 335                 call_on_stack(2, stack, void, __pcpu_delegate,
 336                               pcpu_delegate_fn *, func, void *, data);
 337         }
 338         /* Stop target cpu (if func returns this stops the current cpu). */
 339         pcpu_sigp_retry(pcpu, SIGP_STOP, 0);
 340         pcpu_sigp_retry(pcpu, SIGP_CPU_RESET, 0);
 341         /* Restart func on the target cpu and stop the current cpu. */
 342         if (lc) {
 343                 lc->restart_stack = stack;
 344                 lc->restart_fn = (unsigned long)func;
 345                 lc->restart_data = (unsigned long)data;
 346                 lc->restart_source = source_cpu;
 347         } else {
 348                 abs_lc = get_abs_lowcore();
 349                 abs_lc->restart_stack = stack;
 350                 abs_lc->restart_fn = (unsigned long)func;
 351                 abs_lc->restart_data = (unsigned long)data;
 352                 abs_lc->restart_source = source_cpu;
 353                 put_abs_lowcore(abs_lc);
 354         }
 355         asm volatile(
 356                 "0:     sigp    0,%0,%2 # sigp restart to target cpu\n"
 357                 "       brc     2,0b    # busy, try again\n"
 358                 "1:     sigp    0,%1,%3 # sigp stop to current cpu\n"
 359                 "       brc     2,1b    # busy, try again\n"
 360                 : : "d" (pcpu->address), "d" (source_cpu),
 361                     "K" (SIGP_RESTART), "K" (SIGP_STOP)
 362                 : "0", "1", "cc");
 363         for (;;) ;
 364 }
 365
 366 /*
 367  * Enable additional logical cpus for multi-threading.
 368  */
 369 static int pcpu_set_smt(unsigned int mtid)
 370 {
 371         int cc;
 372
 373         if (smp_cpu_mtid == mtid)
 374                 return 0;
 375         cc = __pcpu_sigp(0, SIGP_SET_MULTI_THREADING, mtid, NULL);
 376         if (cc == 0) {
 377                 smp_cpu_mtid = mtid;
 378                 smp_cpu_mt_shift = 0;
 379                 while (smp_cpu_mtid >= (1U << smp_cpu_mt_shift))
 380                         smp_cpu_mt_shift++;
 381                 pcpu_devices[0].address = stap();
 382         }
 383         return cc;
 384 }
 385
 386 /*
 387  * Call function on an online CPU.
 388  */
 389 void smp_call_online_cpu(void (*func)(void *), void *data)
 390 {
 391         struct pcpu *pcpu;
 392
 393         /* Use the current cpu if it is online. */
 394         pcpu = pcpu_find_address(cpu_online_mask, stap());
 395         if (!pcpu)
 396                 /* Use the first online cpu. */
 397                 pcpu = pcpu_devices + cpumask_first(cpu_online_mask);
 398         pcpu_delegate(pcpu, func, data, (unsigned long) restart_stack);
 399 }
 400
 401 /*
 402  * Call function on the ipl CPU.
 403  */
 404 void smp_call_ipl_cpu(void (*func)(void *), void *data)
 405 {
 406         struct lowcore *lc = lowcore_ptr[0];
 407
 408         if (pcpu_devices[0].address == stap())
 409                 lc = &S390_lowcore;
 410
 411         pcpu_delegate(&pcpu_devices[0], func, data,
 412                       lc->nodat_stack);
 413 }
 414
 415 int smp_find_processor_id(u16 address)
 416 {
 417         int cpu;
 418
 419         for_each_present_cpu(cpu)
 420                 if (pcpu_devices[cpu].address == address)
 421                         return cpu;
 422         return -1;
 423 }
 424
 425 void schedule_mcck_handler(void)
 426 {
 427         pcpu_ec_call(pcpu_devices + smp_processor_id(), ec_mcck_pending);
 428 }
 429
 430 bool notrace arch_vcpu_is_preempted(int cpu)
 431 {
 432         if (test_cpu_flag_of(CIF_ENABLED_WAIT, cpu))
 433                 return false;
 434         if (pcpu_running(pcpu_devices + cpu))
 435                 return false;
 436         return true;
 437 }
 438 EXPORT_SYMBOL(arch_vcpu_is_preempted);
 439
 440 void notrace smp_yield_cpu(int cpu)
 441 {
 442         if (!MACHINE_HAS_DIAG9C)
 443                 return;
 444         diag_stat_inc_norecursion(DIAG_STAT_X09C);
 445         asm volatile("diag %0,0,0x9c"
 446                      : : "d" (pcpu_devices[cpu].address));
 447 }
 448 EXPORT_SYMBOL_GPL(smp_yield_cpu);
 449
 450 /*
 451  * Send cpus emergency shutdown signal. This gives the cpus the
 452  * opportunity to complete outstanding interrupts.
 453  */
 454 void notrace smp_emergency_stop(void)
 455 {
 456         static arch_spinlock_t lock = __ARCH_SPIN_LOCK_UNLOCKED;
 457         static cpumask_t cpumask;
 458         u64 end;
 459         int cpu;
 460
 461         arch_spin_lock(&lock);
 462         cpumask_copy(&cpumask, cpu_online_mask);
 463         cpumask_clear_cpu(smp_processor_id(), &cpumask);
 464
 465         end = get_tod_clock() + (1000000UL << 12);
 466         for_each_cpu(cpu, &cpumask) {
 467                 struct pcpu *pcpu = pcpu_devices + cpu;
 468                 set_bit(ec_stop_cpu, &pcpu->ec_mask);
 469                 while (__pcpu_sigp(pcpu->address, SIGP_EMERGENCY_SIGNAL,
 470                                    0, NULL) == SIGP_CC_BUSY &&
 471                        get_tod_clock() < end)
 472                         cpu_relax();
 473         }
 474         while (get_tod_clock() < end) {
 475                 for_each_cpu(cpu, &cpumask)
 476                         if (pcpu_stopped(pcpu_devices + cpu))
 477                                 cpumask_clear_cpu(cpu, &cpumask);
 478                 if (cpumask_empty(&cpumask))
 479                         break;
 480                 cpu_relax();
 481         }
 482         arch_spin_unlock(&lock);
 483 }
 484 NOKPROBE_SYMBOL(smp_emergency_stop);
 485
 486 /*
 487  * Stop all cpus but the current one.
 488  */
 489 void smp_send_stop(void)
 490 {
 491         int cpu;
 492
 493         /* Disable all interrupts/machine checks */
 494         __load_psw_mask(PSW_KERNEL_BITS);
 495         trace_hardirqs_off();
 496
 497         debug_set_critical();
 498
 499         if (oops_in_progress)
 500                 smp_emergency_stop();
 501
 502         /* stop all processors */
 503         for_each_online_cpu(cpu) {
 504                 if (cpu == smp_processor_id())
 505                         continue;
 506                 pcpu_sigp_retry(pcpu_devices + cpu, SIGP_STOP, 0);
 507                 while (!pcpu_stopped(pcpu_devices + cpu))
 508                         cpu_relax();
 509         }
 510 }
 511
 512 /*
 513  * This is the main routine where commands issued by other
 514  * cpus are handled.
 515  */
 516 static void smp_handle_ext_call(void)
 517 {
 518         unsigned long bits;
 519
 520         /* handle bit signal external calls */
 521         bits = xchg(&pcpu_devices[smp_processor_id()].ec_mask, 0);
 522         if (test_bit(ec_stop_cpu, &bits))
 523                 smp_stop_cpu();
 524         if (test_bit(ec_schedule, &bits))
 525                 scheduler_ipi();
 526         if (test_bit(ec_call_function_single, &bits))
 527                 generic_smp_call_function_single_interrupt();
 528         if (test_bit(ec_mcck_pending, &bits))
 529                 s390_handle_mcck();
 530         if (test_bit(ec_irq_work, &bits))
 531                 irq_work_run();
 532 }
 533
 534 static void do_ext_call_interrupt(struct ext_code ext_code,
 535                                   unsigned int param32, unsigned long param64)
 536 {
 537         inc_irq_stat(ext_code.code == 0x1202 ? IRQEXT_EXC : IRQEXT_EMS);
 538         smp_handle_ext_call();
 539 }
 540
 541 void arch_send_call_function_ipi_mask(const struct cpumask *mask)
 542 {
 543         int cpu;
 544
 545         for_each_cpu(cpu, mask)
 546                 pcpu_ec_call(pcpu_devices + cpu, ec_call_function_single);
 547 }
 548
 549 void arch_send_call_function_single_ipi(int cpu)
 550 {
 551         pcpu_ec_call(pcpu_devices + cpu, ec_call_function_single);
 552 }
 553
 554 /*
 555  * this function sends a 'reschedule' IPI to another CPU.
 556  * it goes straight through and wastes no time serializing
 557  * anything. Worst case is that we lose a reschedule ...
 558  */
 559 void arch_smp_send_reschedule(int cpu)
 560 {
 561         pcpu_ec_call(pcpu_devices + cpu, ec_schedule);
 562 }
 563
 564 #ifdef CONFIG_IRQ_WORK
 565 void arch_irq_work_raise(void)
 566 {
 567         pcpu_ec_call(pcpu_devices + smp_processor_id(), ec_irq_work);
 568 }
 569 #endif
 570
 571 #ifdef CONFIG_CRASH_DUMP
 572
 573 int smp_store_status(int cpu)
 574 {
 575         struct lowcore *lc;
 576         struct pcpu *pcpu;
 577         unsigned long pa;
 578
 579         pcpu = pcpu_devices + cpu;
 580         lc = lowcore_ptr[cpu];
 581         pa = __pa(&lc->floating_pt_save_area);
 582         if (__pcpu_sigp_relax(pcpu->address, SIGP_STORE_STATUS_AT_ADDRESS,
 583                               pa) != SIGP_CC_ORDER_CODE_ACCEPTED)
 584                 return -EIO;
 585         if (!cpu_has_vx() && !MACHINE_HAS_GS)
 586                 return 0;
 587         pa = lc->mcesad & MCESA_ORIGIN_MASK;
 588         if (MACHINE_HAS_GS)
 589                 pa |= lc->mcesad & MCESA_LC_MASK;
 590         if (__pcpu_sigp_relax(pcpu->address, SIGP_STORE_ADDITIONAL_STATUS,
 591                               pa) != SIGP_CC_ORDER_CODE_ACCEPTED)
 592                 return -EIO;
 593         return 0;
 594 }
 595
 596 /*
 597  * Collect CPU state of the previous, crashed system.
 598  * There are four cases:
 599  * 1) standard zfcp/nvme dump
 600  *    condition: OLDMEM_BASE == NULL && is_ipl_type_dump() == true
 601  *    The state for all CPUs except the boot CPU needs to be collected
 602  *    with sigp stop-and-store-status. The boot CPU state is located in
 603  *    the absolute lowcore of the memory stored in the HSA. The zcore code
 604  *    will copy the boot CPU state from the HSA.
 605  * 2) stand-alone kdump for SCSI/NVMe (zfcp/nvme dump with swapped memory)
 606  *    condition: OLDMEM_BASE != NULL && is_ipl_type_dump() == true
 607  *    The state for all CPUs except the boot CPU needs to be collected
 608  *    with sigp stop-and-store-status. The firmware or the boot-loader
 609  *    stored the registers of the boot CPU in the absolute lowcore in the
 610  *    memory of the old system.
 611  * 3) kdump and the old kernel did not store the CPU state,
 612  *    or stand-alone kdump for DASD
 613  *    condition: OLDMEM_BASE != NULL && !is_kdump_kernel()
 614  *    The state for all CPUs except the boot CPU needs to be collected
 615  *    with sigp stop-and-store-status. The kexec code or the boot-loader
 616  *    stored the registers of the boot CPU in the memory of the old system.
 617  * 4) kdump and the old kernel stored the CPU state
 618  *    condition: OLDMEM_BASE != NULL && is_kdump_kernel()
 619  *    This case does not exist for s390 anymore, setup_arch explicitly
 620  *    deactivates the elfcorehdr= kernel parameter
 621  */
 622 static bool dump_available(void)
 623 {
 624         return oldmem_data.start || is_ipl_type_dump();
 625 }
 626
 627 void __init smp_save_dump_ipl_cpu(void)
 628 {
 629         struct save_area *sa;
 630         void *regs;
 631
 632         if (!dump_available())
 633                 return;
 634         sa = save_area_alloc(true);
 635         regs = memblock_alloc(512, 8);
 636         if (!sa || !regs)
 637                 panic("could not allocate memory for boot CPU save area\n");
 638         copy_oldmem_kernel(regs, __LC_FPREGS_SAVE_AREA, 512);
 639         save_area_add_regs(sa, regs);
 640         memblock_free(regs, 512);
 641         if (cpu_has_vx())
 642                 save_area_add_vxrs(sa, boot_cpu_vector_save_area);
 643 }
 644
 645 void __init smp_save_dump_secondary_cpus(void)
 646 {
 647         int addr, boot_cpu_addr, max_cpu_addr;
 648         struct save_area *sa;
 649         void *page;
 650
 651         if (!dump_available())
 652                 return;
 653         /* Allocate a page as dumping area for the store status sigps */
 654         page = memblock_alloc_low(PAGE_SIZE, PAGE_SIZE);
 655         if (!page)
 656                 panic("ERROR: Failed to allocate %lx bytes below %lx\n",
 657                       PAGE_SIZE, 1UL << 31);
 658
 659         /* Set multi-threading state to the previous system. */
 660         pcpu_set_smt(sclp.mtid_prev);
 661         boot_cpu_addr = stap();
 662         max_cpu_addr = SCLP_MAX_CORES << sclp.mtid_prev;
 663         for (addr = 0; addr <= max_cpu_addr; addr++) {
 664                 if (addr == boot_cpu_addr)
 665                         continue;
 666                 if (__pcpu_sigp_relax(addr, SIGP_SENSE, 0) ==
 667                     SIGP_CC_NOT_OPERATIONAL)
 668                         continue;
 669                 sa = save_area_alloc(false);
 670                 if (!sa)
 671                         panic("could not allocate memory for save area\n");
 672                 __pcpu_sigp_relax(addr, SIGP_STORE_STATUS_AT_ADDRESS, __pa(page));
 673                 save_area_add_regs(sa, page);
 674                 if (cpu_has_vx()) {
 675                         __pcpu_sigp_relax(addr, SIGP_STORE_ADDITIONAL_STATUS, __pa(page));
 676                         save_area_add_vxrs(sa, page);
 677                 }
 678         }
 679         memblock_free(page, PAGE_SIZE);
 680         diag_amode31_ops.diag308_reset();
 681         pcpu_set_smt(0);
 682 }
 683 #endif /* CONFIG_CRASH_DUMP */
 684
 685 void smp_cpu_set_polarization(int cpu, int val)
 686 {
 687         pcpu_devices[cpu].polarization = val;
 688 }
 689
 690 int smp_cpu_get_polarization(int cpu)
 691 {
 692         return pcpu_devices[cpu].polarization;
 693 }
 694
 695 int smp_cpu_get_cpu_address(int cpu)
 696 {
 697         return pcpu_devices[cpu].address;
 698 }
 699
 700 static void __ref smp_get_core_info(struct sclp_core_info *info, int early)
 701 {
 702         static int use_sigp_detection;
 703         int address;
 704
 705         if (use_sigp_detection || sclp_get_core_info(info, early)) {
 706                 use_sigp_detection = 1;
 707                 for (address = 0;
 708                      address < (SCLP_MAX_CORES << smp_cpu_mt_shift);
 709                      address += (1U << smp_cpu_mt_shift)) {
 710                         if (__pcpu_sigp_relax(address, SIGP_SENSE, 0) ==
 711                             SIGP_CC_NOT_OPERATIONAL)
 712                                 continue;
 713                         info->core[info->configured].core_id =
 714                                 address >> smp_cpu_mt_shift;
 715                         info->configured++;
 716                 }
 717                 info->combined = info->configured;
 718         }
 719 }
 720
 721 static int smp_add_present_cpu(int cpu);
 722
 723 static int smp_add_core(struct sclp_core_entry *core, cpumask_t *avail,
 724                         bool configured, bool early)
 725 {
 726         struct pcpu *pcpu;
 727         int cpu, nr, i;
 728         u16 address;
 729
 730         nr = 0;
 731         if (sclp.has_core_type && core->type != boot_core_type)
 732                 return nr;
 733         cpu = cpumask_first(avail);
 734         address = core->core_id << smp_cpu_mt_shift;
 735         for (i = 0; (i <= smp_cpu_mtid) && (cpu < nr_cpu_ids); i++) {
 736                 if (pcpu_find_address(cpu_present_mask, address + i))
 737                         continue;
 738                 pcpu = pcpu_devices + cpu;
 739                 pcpu->address = address + i;
 740                 if (configured)
 741                         pcpu->state = CPU_STATE_CONFIGURED;
 742                 else
 743                         pcpu->state = CPU_STATE_STANDBY;
 744                 smp_cpu_set_polarization(cpu, POLARIZATION_UNKNOWN);
 745                 set_cpu_present(cpu, true);
 746                 if (!early && smp_add_present_cpu(cpu) != 0)
 747                         set_cpu_present(cpu, false);
 748                 else
 749                         nr++;
 750                 cpumask_clear_cpu(cpu, avail);
 751                 cpu = cpumask_next(cpu, avail);
 752         }
 753         return nr;
 754 }
 755
 756 static int __smp_rescan_cpus(struct sclp_core_info *info, bool early)
 757 {
 758         struct sclp_core_entry *core;
 759         static cpumask_t avail;
 760         bool configured;
 761         u16 core_id;
 762         int nr, i;
 763
 764         cpus_read_lock();
 765         mutex_lock(&smp_cpu_state_mutex);
 766         nr = 0;
 767         cpumask_xor(&avail, cpu_possible_mask, cpu_present_mask);
 768         /*
 769          * Add IPL core first (which got logical CPU number 0) to make sure
 770          * that all SMT threads get subsequent logical CPU numbers.
 771          */
 772         if (early) {
 773                 core_id = pcpu_devices[0].address >> smp_cpu_mt_shift;
 774                 for (i = 0; i < info->configured; i++) {
 775                         core = &info->core[i];
 776                         if (core->core_id == core_id) {
 777                                 nr += smp_add_core(core, &avail, true, early);
 778                                 break;
 779                         }
 780                 }
 781         }
 782         for (i = 0; i < info->combined; i++) {
 783                 configured = i < info->configured;
 784                 nr += smp_add_core(&info->core[i], &avail, configured, early);
 785         }
 786         mutex_unlock(&smp_cpu_state_mutex);
 787         cpus_read_unlock();
 788         return nr;
 789 }
 790
 791 void __init smp_detect_cpus(void)
 792 {
 793         unsigned int cpu, mtid, c_cpus, s_cpus;
 794         struct sclp_core_info *info;
 795         u16 address;
 796
 797         /* Get CPU information */
 798         info = memblock_alloc(sizeof(*info), 8);
 799         if (!info)
 800                 panic("%s: Failed to allocate %zu bytes align=0x%x\n",
 801                       __func__, sizeof(*info), 8);
 802         smp_get_core_info(info, 1);
 803         /* Find boot CPU type */
 804         if (sclp.has_core_type) {
 805                 address = stap();
 806                 for (cpu = 0; cpu < info->combined; cpu++)
 807                         if (info->core[cpu].core_id == address) {
 808                                 /* The boot cpu dictates the cpu type. */
 809                                 boot_core_type = info->core[cpu].type;
 810                                 break;
 811                         }
 812                 if (cpu >= info->combined)
 813                         panic("Could not find boot CPU type");
 814         }
 815
 816         /* Set multi-threading state for the current system */
 817         mtid = boot_core_type ? sclp.mtid : sclp.mtid_cp;
 818         mtid = (mtid < smp_max_threads) ? mtid : smp_max_threads - 1;
 819         pcpu_set_smt(mtid);
 820
 821         /* Print number of CPUs */
 822         c_cpus = s_cpus = 0;
 823         for (cpu = 0; cpu < info->combined; cpu++) {
 824                 if (sclp.has_core_type &&
 825                     info->core[cpu].type != boot_core_type)
 826                         continue;
 827                 if (cpu < info->configured)
 828                         c_cpus += smp_cpu_mtid + 1;
 829                 else
 830                         s_cpus += smp_cpu_mtid + 1;
 831         }
 832         pr_info("%d configured CPUs, %d standby CPUs\n", c_cpus, s_cpus);
 833
 834         /* Add CPUs present at boot */
 835         __smp_rescan_cpus(info, true);
 836         memblock_free(info, sizeof(*info));
 837 }
 838
 839 /*
 840  *      Activate a secondary processor.
 841  */
 842 static void smp_start_secondary(void *cpuvoid)
 843 {
 844         int cpu = raw_smp_processor_id();
 845
 846         S390_lowcore.last_update_clock = get_tod_clock();
 847         S390_lowcore.restart_stack = (unsigned long)restart_stack;
 848         S390_lowcore.restart_fn = (unsigned long)do_restart;
 849         S390_lowcore.restart_data = 0;
 850         S390_lowcore.restart_source = -1U;
 851         S390_lowcore.restart_flags = 0;
 852         restore_access_regs(S390_lowcore.access_regs_save_area);
 853         cpu_init();
 854         rcutree_report_cpu_starting(cpu);
 855         init_cpu_timer();
 856         vtime_init();
 857         vdso_getcpu_init();
 858         pfault_init();
 859         cpumask_set_cpu(cpu, &cpu_setup_mask);
 860         update_cpu_masks();
 861         notify_cpu_starting(cpu);
 862         if (topology_cpu_dedicated(cpu))
 863                 set_cpu_flag(CIF_DEDICATED_CPU);
 864         else
 865                 clear_cpu_flag(CIF_DEDICATED_CPU);
 866         set_cpu_online(cpu, true);
 867         inc_irq_stat(CPU_RST);
 868         local_irq_enable();
 869         cpu_startup_entry(CPUHP_AP_ONLINE_IDLE);
 870 }
 871
 872 /* Upping and downing of CPUs */
 873 int __cpu_up(unsigned int cpu, struct task_struct *tidle)
 874 {
 875         struct pcpu *pcpu = pcpu_devices + cpu;
 876         int rc;
 877
 878         if (pcpu->state != CPU_STATE_CONFIGURED)
 879                 return -EIO;
 880         if (pcpu_sigp_retry(pcpu, SIGP_INITIAL_CPU_RESET, 0) !=
 881             SIGP_CC_ORDER_CODE_ACCEPTED)
 882                 return -EIO;
 883
 884         rc = pcpu_alloc_lowcore(pcpu, cpu);
 885         if (rc)
 886                 return rc;
 887         /*
 888          * Make sure global control register contents do not change
 889          * until new CPU has initialized control registers.
 890          */
 891         system_ctlreg_lock();
 892         pcpu_prepare_secondary(pcpu, cpu);
 893         pcpu_attach_task(pcpu, tidle);
 894         pcpu_start_fn(pcpu, smp_start_secondary, NULL);
 895         /* Wait until cpu puts itself in the online & active maps */
 896         while (!cpu_online(cpu))
 897                 cpu_relax();
 898         system_ctlreg_unlock();
 899         return 0;
 900 }
 901
 902 static unsigned int setup_possible_cpus __initdata;
 903
 904 static int __init _setup_possible_cpus(char *s)
 905 {
 906         get_option(&s, &setup_possible_cpus);
 907         return 0;
 908 }
 909 early_param("possible_cpus", _setup_possible_cpus);
 910
 911 int __cpu_disable(void)
 912 {
 913         struct ctlreg cregs[16];
 914         int cpu;
 915
 916         /* Handle possible pending IPIs */
 917         smp_handle_ext_call();
 918         cpu = smp_processor_id();
 919         set_cpu_online(cpu, false);
 920         cpumask_clear_cpu(cpu, &cpu_setup_mask);
 921         update_cpu_masks();
 922         /* Disable pseudo page faults on this cpu. */
 923         pfault_fini();
 924         /* Disable interrupt sources via control register. */
 925         __local_ctl_store(0, 15, cregs);
 926         cregs[0].val  &= ~0x0000ee70UL; /* disable all external interrupts */
 927         cregs[6].val  &= ~0xff000000UL; /* disable all I/O interrupts */
 928         cregs[14].val &= ~0x1f000000UL; /* disable most machine checks */
 929         __local_ctl_load(0, 15, cregs);
 930         clear_cpu_flag(CIF_NOHZ_DELAY);
 931         return 0;
 932 }
 933
 934 void __cpu_die(unsigned int cpu)
 935 {
 936         struct pcpu *pcpu;
 937
 938         /* Wait until target cpu is down */
 939         pcpu = pcpu_devices + cpu;
 940         while (!pcpu_stopped(pcpu))
 941                 cpu_relax();
 942         pcpu_free_lowcore(pcpu);
 943         cpumask_clear_cpu(cpu, mm_cpumask(&init_mm));
 944         cpumask_clear_cpu(cpu, &init_mm.context.cpu_attach_mask);
 945 }
 946
 947 void __noreturn cpu_die(void)
 948 {
 949         idle_task_exit();
 950         pcpu_sigp_retry(pcpu_devices + smp_processor_id(), SIGP_STOP, 0);
 951         for (;;) ;
 952 }
 953
 954 void __init smp_fill_possible_mask(void)
 955 {
 956         unsigned int possible, sclp_max, cpu;
 957
 958         sclp_max = max(sclp.mtid, sclp.mtid_cp) + 1;
 959         sclp_max = min(smp_max_threads, sclp_max);
 960         sclp_max = (sclp.max_cores * sclp_max) ?: nr_cpu_ids;
 961         possible = setup_possible_cpus ?: nr_cpu_ids;
 962         possible = min(possible, sclp_max);
 963         for (cpu = 0; cpu < possible && cpu < nr_cpu_ids; cpu++)
 964                 set_cpu_possible(cpu, true);
 965 }
 966
 967 void __init smp_prepare_cpus(unsigned int max_cpus)
 968 {
 969         if (register_external_irq(EXT_IRQ_EMERGENCY_SIG, do_ext_call_interrupt))
 970                 panic("Couldn't request external interrupt 0x1201");
 971         system_ctl_set_bit(0, 14);
 972         if (register_external_irq(EXT_IRQ_EXTERNAL_CALL, do_ext_call_interrupt))
 973                 panic("Couldn't request external interrupt 0x1202");
 974         system_ctl_set_bit(0, 13);
 975 }
 976
 977 void __init smp_prepare_boot_cpu(void)
 978 {
 979         struct pcpu *pcpu = pcpu_devices;
 980
 981         WARN_ON(!cpu_present(0) || !cpu_online(0));
 982         pcpu->state = CPU_STATE_CONFIGURED;
 983         S390_lowcore.percpu_offset = __per_cpu_offset[0];
 984         smp_cpu_set_polarization(0, POLARIZATION_UNKNOWN);
 985 }
 986
 987 void __init smp_setup_processor_id(void)
 988 {
 989         pcpu_devices[0].address = stap();
 990         S390_lowcore.cpu_nr = 0;
 991         S390_lowcore.spinlock_lockval = arch_spin_lockval(0);
 992         S390_lowcore.spinlock_index = 0;
 993 }
 994
 995 /*
 996  * the frequency of the profiling timer can be changed
 997  * by writing a multiplier value into /proc/profile.
 998  *
 999  * usually you want to run this on all CPUs ;)
1000  */
1001 int setup_profiling_timer(unsigned int multiplier)
1002 {
1003         return 0;
1004 }
1005
1006 static ssize_t cpu_configure_show(struct device *dev,
1007                                   struct device_attribute *attr, char *buf)
1008 {
1009         ssize_t count;
1010
1011         mutex_lock(&smp_cpu_state_mutex);
1012         count = sprintf(buf, "%d\n", pcpu_devices[dev->id].state);
1013         mutex_unlock(&smp_cpu_state_mutex);
1014         return count;
1015 }
1016
1017 static ssize_t cpu_configure_store(struct device *dev,
1018                                    struct device_attribute *attr,
1019                                    const char *buf, size_t count)
1020 {
1021         struct pcpu *pcpu;
1022         int cpu, val, rc, i;
1023         char delim;
1024
1025         if (sscanf(buf, "%d %c", &val, &delim) != 1)
1026                 return -EINVAL;
1027         if (val != 0 && val != 1)
1028                 return -EINVAL;
1029         cpus_read_lock();
1030         mutex_lock(&smp_cpu_state_mutex);
1031         rc = -EBUSY;
1032         /* disallow configuration changes of online cpus */
1033         cpu = dev->id;
1034         cpu = smp_get_base_cpu(cpu);
1035         for (i = 0; i <= smp_cpu_mtid; i++)
1036                 if (cpu_online(cpu + i))
1037                         goto out;
1038         pcpu = pcpu_devices + cpu;
1039         rc = 0;
1040         switch (val) {
1041         case 0:
1042                 if (pcpu->state != CPU_STATE_CONFIGURED)
1043                         break;
1044                 rc = sclp_core_deconfigure(pcpu->address >> smp_cpu_mt_shift);
1045                 if (rc)
1046                         break;
1047                 for (i = 0; i <= smp_cpu_mtid; i++) {
1048                         if (cpu + i >= nr_cpu_ids || !cpu_present(cpu + i))
1049                                 continue;
1050                         pcpu[i].state = CPU_STATE_STANDBY;
1051                         smp_cpu_set_polarization(cpu + i,
1052                                                  POLARIZATION_UNKNOWN);
1053                 }
1054                 topology_expect_change();
1055                 break;
1056         case 1:
1057                 if (pcpu->state != CPU_STATE_STANDBY)
1058                         break;
1059                 rc = sclp_core_configure(pcpu->address >> smp_cpu_mt_shift);
1060                 if (rc)
1061                         break;
1062                 for (i = 0; i <= smp_cpu_mtid; i++) {
1063                         if (cpu + i >= nr_cpu_ids || !cpu_present(cpu + i))
1064                                 continue;
1065                         pcpu[i].state = CPU_STATE_CONFIGURED;
1066                         smp_cpu_set_polarization(cpu + i,
1067                                                  POLARIZATION_UNKNOWN);
1068                 }
1069                 topology_expect_change();
1070                 break;
1071         default:
1072                 break;
1073         }
1074 out:
1075         mutex_unlock(&smp_cpu_state_mutex);
1076         cpus_read_unlock();
1077         return rc ? rc : count;
1078 }
1079 static DEVICE_ATTR(configure, 0644, cpu_configure_show, cpu_configure_store);
1080
1081 static ssize_t show_cpu_address(struct device *dev,
1082                                 struct device_attribute *attr, char *buf)
1083 {
1084         return sprintf(buf, "%d\n", pcpu_devices[dev->id].address);
1085 }
1086 static DEVICE_ATTR(address, 0444, show_cpu_address, NULL);
1087
1088 static struct attribute *cpu_common_attrs[] = {
1089         &dev_attr_configure.attr,
1090         &dev_attr_address.attr,
1091         NULL,
1092 };
1093
1094 static struct attribute_group cpu_common_attr_group = {
1095         .attrs = cpu_common_attrs,
1096 };
1097
1098 static struct attribute *cpu_online_attrs[] = {
1099         &dev_attr_idle_count.attr,
1100         &dev_attr_idle_time_us.attr,
1101         NULL,
1102 };
1103
1104 static struct attribute_group cpu_online_attr_group = {
1105         .attrs = cpu_online_attrs,
1106 };
1107
1108 static int smp_cpu_online(unsigned int cpu)
1109 {
1110         struct device *s = &per_cpu(cpu_device, cpu)->dev;
1111
1112         return sysfs_create_group(&s->kobj, &cpu_online_attr_group);
1113 }
1114
1115 static int smp_cpu_pre_down(unsigned int cpu)
1116 {
1117         struct device *s = &per_cpu(cpu_device, cpu)->dev;
1118
1119         sysfs_remove_group(&s->kobj, &cpu_online_attr_group);
1120         return 0;
1121 }
1122
1123 static int smp_add_present_cpu(int cpu)
1124 {
1125         struct device *s;
1126         struct cpu *c;
1127         int rc;
1128
1129         c = kzalloc(sizeof(*c), GFP_KERNEL);
1130         if (!c)
1131                 return -ENOMEM;
1132         per_cpu(cpu_device, cpu) = c;
1133         s = &c->dev;
1134         c->hotpluggable = !!cpu;
1135         rc = register_cpu(c, cpu);
1136         if (rc)
1137                 goto out;
1138         rc = sysfs_create_group(&s->kobj, &cpu_common_attr_group);
1139         if (rc)
1140                 goto out_cpu;
1141         rc = topology_cpu_init(c);
1142         if (rc)
1143                 goto out_topology;
1144         return 0;
1145
1146 out_topology:
1147         sysfs_remove_group(&s->kobj, &cpu_common_attr_group);
1148 out_cpu:
1149         unregister_cpu(c);
1150 out:
1151         return rc;
1152 }
1153
1154 int __ref smp_rescan_cpus(void)
1155 {
1156         struct sclp_core_info *info;
1157         int nr;
1158
1159         info = kzalloc(sizeof(*info), GFP_KERNEL);
1160         if (!info)
1161                 return -ENOMEM;
1162         smp_get_core_info(info, 0);
1163         nr = __smp_rescan_cpus(info, false);
1164         kfree(info);
1165         if (nr)
1166                 topology_schedule_update();
1167         return 0;
1168 }
1169
1170 static ssize_t __ref rescan_store(struct device *dev,
1171                                   struct device_attribute *attr,
1172                                   const char *buf,
1173                                   size_t count)
1174 {
1175         int rc;
1176
1177         rc = lock_device_hotplug_sysfs();
1178         if (rc)
1179                 return rc;
1180         rc = smp_rescan_cpus();
1181         unlock_device_hotplug();
1182         return rc ? rc : count;
1183 }
1184 static DEVICE_ATTR_WO(rescan);
1185
1186 static int __init s390_smp_init(void)
1187 {
1188         struct device *dev_root;
1189         int cpu, rc = 0;
1190
1191         dev_root = bus_get_dev_root(&cpu_subsys);
1192         if (dev_root) {
1193                 rc = device_create_file(dev_root, &dev_attr_rescan);
1194                 put_device(dev_root);
1195                 if (rc)
1196                         return rc;
1197         }
1198
1199         for_each_present_cpu(cpu) {
1200                 rc = smp_add_present_cpu(cpu);
1201                 if (rc)
1202                         goto out;
1203         }
1204
1205         rc = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "s390/smp:online",
1206                                smp_cpu_online, smp_cpu_pre_down);
1207         rc = rc <= 0 ? rc : 0;
1208 out:
1209         return rc;
1210 }
1211 subsys_initcall(s390_smp_init);