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 <[email protected]>,
   8  *
   9  *  based on other smp stuff by
  10  *    (c) 1995 Alan Cox, CymruNET Ltd  <[email protected]>
  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/access-regs.h>
  40 #include <asm/asm-offsets.h>
  41 #include <asm/ctlreg.h>
  42 #include <asm/pfault.h>
  43 #include <asm/diag.h>
  44 #include <asm/facility.h>
  45 #include <asm/fpu.h>
  46 #include <asm/ipl.h>
  47 #include <asm/setup.h>
  48 #include <asm/irq.h>
  49 #include <asm/tlbflush.h>
  50 #include <asm/vtimer.h>
  51 #include <asm/abs_lowcore.h>
  52 #include <asm/sclp.h>
  53 #include <asm/debug.h>
  54 #include <asm/os_info.h>
  55 #include <asm/sigp.h>
  56 #include <asm/idle.h>
  57 #include <asm/nmi.h>
  58 #include <asm/stacktrace.h>
  59 #include <asm/topology.h>
  60 #include <asm/vdso.h>
  61 #include <asm/maccess.h>
  62 #include "entry.h"
  63
  64 enum {
  65         ec_schedule = 0,
  66         ec_call_function_single,
  67         ec_stop_cpu,
  68         ec_mcck_pending,
  69         ec_irq_work,
  70 };
  71
  72 enum {
  73         CPU_STATE_STANDBY,
  74         CPU_STATE_CONFIGURED,
  75 };
  76
  77 static DEFINE_PER_CPU(struct cpu *, cpu_device);
  78
  79 struct pcpu {
  80         unsigned long ec_mask;          /* bit mask for ec_xxx functions */
  81         unsigned long ec_clk;           /* sigp timestamp for ec_xxx */
  82         signed char state;              /* physical cpu state */
  83         signed char polarization;       /* physical polarization */
  84         u16 address;                    /* physical cpu address */
  85 };
  86
  87 static u8 boot_core_type;
  88 static struct pcpu pcpu_devices[NR_CPUS];
  89
  90 unsigned int smp_cpu_mt_shift;
  91 EXPORT_SYMBOL(smp_cpu_mt_shift);
  92
  93 unsigned int smp_cpu_mtid;
  94 EXPORT_SYMBOL(smp_cpu_mtid);
  95
  96 #ifdef CONFIG_CRASH_DUMP
  97 __vector128 __initdata boot_cpu_vector_save_area[__NUM_VXRS];
  98 #endif
  99
 100 static unsigned int smp_max_threads __initdata = -1U;
 101 cpumask_t cpu_setup_mask;
 102
 103 static int __init early_nosmt(char *s)
 104 {
 105         smp_max_threads = 1;
 106         return 0;
 107 }
 108 early_param("nosmt", early_nosmt);
 109
 110 static int __init early_smt(char *s)
 111 {
 112         get_option(&s, &smp_max_threads);
 113         return 0;
 114 }
 115 early_param("smt", early_smt);
 116
 117 /*
 118  * The smp_cpu_state_mutex must be held when changing the state or polarization
 119  * member of a pcpu data structure within the pcpu_devices array.
 120  */
 121 DEFINE_MUTEX(smp_cpu_state_mutex);
 122
 123 /*
 124  * Signal processor helper functions.
 125  */
 126 static inline int __pcpu_sigp_relax(u16 addr, u8 order, unsigned long parm)
 127 {
 128         int cc;
 129
 130         while (1) {
 131                 cc = __pcpu_sigp(addr, order, parm, NULL);
 132                 if (cc != SIGP_CC_BUSY)
 133                         return cc;
 134                 cpu_relax();
 135         }
 136 }
 137
 138 static int pcpu_sigp_retry(struct pcpu *pcpu, u8 order, u32 parm)
 139 {
 140         int cc, retry;
 141
 142         for (retry = 0; ; retry++) {
 143                 cc = __pcpu_sigp(pcpu->address, order, parm, NULL);
 144                 if (cc != SIGP_CC_BUSY)
 145                         break;
 146                 if (retry >= 3)
 147                         udelay(10);
 148         }
 149         return cc;
 150 }
 151
 152 static inline int pcpu_stopped(struct pcpu *pcpu)
 153 {
 154         u32 status;
 155
 156         if (__pcpu_sigp(pcpu->address, SIGP_SENSE,
 157                         0, &status) != SIGP_CC_STATUS_STORED)
 158                 return 0;
 159         return !!(status & (SIGP_STATUS_CHECK_STOP|SIGP_STATUS_STOPPED));
 160 }
 161
 162 static inline int pcpu_running(struct pcpu *pcpu)
 163 {
 164         if (__pcpu_sigp(pcpu->address, SIGP_SENSE_RUNNING,
 165                         0, NULL) != SIGP_CC_STATUS_STORED)
 166                 return 1;
 167         /* Status stored condition code is equivalent to cpu not running. */
 168         return 0;
 169 }
 170
 171 /*
 172  * Find struct pcpu by cpu address.
 173  */
 174 static struct pcpu *pcpu_find_address(const struct cpumask *mask, u16 address)
 175 {
 176         int cpu;
 177
 178         for_each_cpu(cpu, mask)
 179                 if (pcpu_devices[cpu].address == address)
 180                         return pcpu_devices + cpu;
 181         return NULL;
 182 }
 183
 184 static void pcpu_ec_call(struct pcpu *pcpu, int ec_bit)
 185 {
 186         int order;
 187
 188         if (test_and_set_bit(ec_bit, &pcpu->ec_mask))
 189                 return;
 190         order = pcpu_running(pcpu) ? SIGP_EXTERNAL_CALL : SIGP_EMERGENCY_SIGNAL;
 191         pcpu->ec_clk = get_tod_clock_fast();
 192         pcpu_sigp_retry(pcpu, order, 0);
 193 }
 194
 195 static int pcpu_alloc_lowcore(struct pcpu *pcpu, int cpu)
 196 {
 197         unsigned long async_stack, nodat_stack, mcck_stack;
 198         struct lowcore *lc;
 199
 200         lc = (struct lowcore *) __get_free_pages(GFP_KERNEL | GFP_DMA, LC_ORDER);
 201         nodat_stack = __get_free_pages(GFP_KERNEL, THREAD_SIZE_ORDER);
 202         async_stack = stack_alloc();
 203         mcck_stack = stack_alloc();
 204         if (!lc || !nodat_stack || !async_stack || !mcck_stack)
 205                 goto out;
 206         memcpy(lc, &S390_lowcore, 512);
 207         memset((char *) lc + 512, 0, sizeof(*lc) - 512);
 208         lc->async_stack = async_stack + STACK_INIT_OFFSET;
 209         lc->nodat_stack = nodat_stack + STACK_INIT_OFFSET;
 210         lc->mcck_stack = mcck_stack + STACK_INIT_OFFSET;
 211         lc->cpu_nr = cpu;
 212         lc->spinlock_lockval = arch_spin_lockval(cpu);
 213         lc->spinlock_index = 0;
 214         lc->return_lpswe = gen_lpswe(__LC_RETURN_PSW);
 215         lc->return_mcck_lpswe = gen_lpswe(__LC_RETURN_MCCK_PSW);
 216         lc->preempt_count = PREEMPT_DISABLED;
 217         if (nmi_alloc_mcesa(&lc->mcesad))
 218                 goto out;
 219         if (abs_lowcore_map(cpu, lc, true))
 220                 goto out_mcesa;
 221         lowcore_ptr[cpu] = lc;
 222         pcpu_sigp_retry(pcpu, SIGP_SET_PREFIX, __pa(lc));
 223         return 0;
 224
 225 out_mcesa:
 226         nmi_free_mcesa(&lc->mcesad);
 227 out:
 228         stack_free(mcck_stack);
 229         stack_free(async_stack);
 230         free_pages(nodat_stack, THREAD_SIZE_ORDER);
 231         free_pages((unsigned long) lc, LC_ORDER);
 232         return -ENOMEM;
 233 }
 234
 235 static void pcpu_free_lowcore(struct pcpu *pcpu)
 236 {
 237         unsigned long async_stack, nodat_stack, mcck_stack;
 238         struct lowcore *lc;
 239         int cpu;
 240
 241         cpu = pcpu - pcpu_devices;
 242         lc = lowcore_ptr[cpu];
 243         nodat_stack = lc->nodat_stack - STACK_INIT_OFFSET;
 244         async_stack = lc->async_stack - STACK_INIT_OFFSET;
 245         mcck_stack = lc->mcck_stack - STACK_INIT_OFFSET;
 246         pcpu_sigp_retry(pcpu, SIGP_SET_PREFIX, 0);
 247         lowcore_ptr[cpu] = NULL;
 248         abs_lowcore_unmap(cpu);
 249         nmi_free_mcesa(&lc->mcesad);
 250         stack_free(async_stack);
 251         stack_free(mcck_stack);
 252         free_pages(nodat_stack, THREAD_SIZE_ORDER);
 253         free_pages((unsigned long) lc, LC_ORDER);
 254 }
 255
 256 static void pcpu_prepare_secondary(struct pcpu *pcpu, int cpu)
 257 {
 258         struct lowcore *lc, *abs_lc;
 259
 260         lc = lowcore_ptr[cpu];
 261         cpumask_set_cpu(cpu, &init_mm.context.cpu_attach_mask);
 262         cpumask_set_cpu(cpu, mm_cpumask(&init_mm));
 263         lc->cpu_nr = cpu;
 264         lc->restart_flags = RESTART_FLAG_CTLREGS;
 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->user_asce = s390_invalid_asce;
 270         lc->machine_flags = S390_lowcore.machine_flags;
 271         lc->user_timer = lc->system_timer =
 272                 lc->steal_timer = lc->avg_steal_timer = 0;
 273         abs_lc = get_abs_lowcore();
 274         memcpy(lc->cregs_save_area, abs_lc->cregs_save_area, sizeof(lc->cregs_save_area));
 275         put_abs_lowcore(abs_lc);
 276         lc->cregs_save_area[1] = lc->kernel_asce;
 277         lc->cregs_save_area[7] = lc->user_asce;
 278         save_access_regs((unsigned int *) lc->access_regs_save_area);
 279         arch_spin_lock_setup(cpu);
 280 }
 281
 282 static void pcpu_attach_task(struct pcpu *pcpu, struct task_struct *tsk)
 283 {
 284         struct lowcore *lc;
 285         int cpu;
 286
 287         cpu = pcpu - pcpu_devices;
 288         lc = lowcore_ptr[cpu];
 289         lc->kernel_stack = (unsigned long)task_stack_page(tsk) + STACK_INIT_OFFSET;
 290         lc->current_task = (unsigned long)tsk;
 291         lc->lpp = LPP_MAGIC;
 292         lc->current_pid = tsk->pid;
 293         lc->user_timer = tsk->thread.user_timer;
 294         lc->guest_timer = tsk->thread.guest_timer;
 295         lc->system_timer = tsk->thread.system_timer;
 296         lc->hardirq_timer = tsk->thread.hardirq_timer;
 297         lc->softirq_timer = tsk->thread.softirq_timer;
 298         lc->steal_timer = 0;
 299 }
 300
 301 static void pcpu_start_fn(struct pcpu *pcpu, void (*func)(void *), void *data)
 302 {
 303         struct lowcore *lc;
 304         int cpu;
 305
 306         cpu = pcpu - pcpu_devices;
 307         lc = lowcore_ptr[cpu];
 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 = -1U;
 312         pcpu_sigp_retry(pcpu, SIGP_RESTART, 0);
 313 }
 314
 315 typedef void (pcpu_delegate_fn)(void *);
 316
 317 /*
 318  * Call function via PSW restart on pcpu and stop the current cpu.
 319  */
 320 static void __pcpu_delegate(pcpu_delegate_fn *func, void *data)
 321 {
 322         func(data);     /* should not return */
 323 }
 324
 325 static void pcpu_delegate(struct pcpu *pcpu,
 326                           pcpu_delegate_fn *func,
 327                           void *data, unsigned long stack)
 328 {
 329         struct lowcore *lc, *abs_lc;
 330         unsigned int source_cpu;
 331
 332         lc = lowcore_ptr[pcpu - pcpu_devices];
 333         source_cpu = stap();
 334
 335         if (pcpu->address == source_cpu) {
 336                 call_on_stack(2, stack, void, __pcpu_delegate,
 337                               pcpu_delegate_fn *, func, void *, data);
 338         }
 339         /* Stop target cpu (if func returns this stops the current cpu). */
 340         pcpu_sigp_retry(pcpu, SIGP_STOP, 0);
 341         pcpu_sigp_retry(pcpu, SIGP_CPU_RESET, 0);
 342         /* Restart func on the target cpu and stop the current cpu. */
 343         if (lc) {
 344                 lc->restart_stack = stack;
 345                 lc->restart_fn = (unsigned long)func;
 346                 lc->restart_data = (unsigned long)data;
 347                 lc->restart_source = source_cpu;
 348         } else {
 349                 abs_lc = get_abs_lowcore();
 350                 abs_lc->restart_stack = stack;
 351                 abs_lc->restart_fn = (unsigned long)func;
 352                 abs_lc->restart_data = (unsigned long)data;
 353                 abs_lc->restart_source = source_cpu;
 354                 put_abs_lowcore(abs_lc);
 355         }
 356         asm volatile(
 357                 "0:     sigp    0,%0,%2 # sigp restart to target cpu\n"
 358                 "       brc     2,0b    # busy, try again\n"
 359                 "1:     sigp    0,%1,%3 # sigp stop to current cpu\n"
 360                 "       brc     2,1b    # busy, try again\n"
 361                 : : "d" (pcpu->address), "d" (source_cpu),
 362                     "K" (SIGP_RESTART), "K" (SIGP_STOP)
 363                 : "0", "1", "cc");
 364         for (;;) ;
 365 }
 366
 367 /*
 368  * Enable additional logical cpus for multi-threading.
 369  */
 370 static int pcpu_set_smt(unsigned int mtid)
 371 {
 372         int cc;
 373
 374         if (smp_cpu_mtid == mtid)
 375                 return 0;
 376         cc = __pcpu_sigp(0, SIGP_SET_MULTI_THREADING, mtid, NULL);
 377         if (cc == 0) {
 378                 smp_cpu_mtid = mtid;
 379                 smp_cpu_mt_shift = 0;
 380                 while (smp_cpu_mtid >= (1U << smp_cpu_mt_shift))
 381                         smp_cpu_mt_shift++;
 382                 pcpu_devices[0].address = stap();
 383         }
 384         return cc;
 385 }
 386
 387 /*
 388  * Call function on an online CPU.
 389  */
 390 void smp_call_online_cpu(void (*func)(void *), void *data)
 391 {
 392         struct pcpu *pcpu;
 393
 394         /* Use the current cpu if it is online. */
 395         pcpu = pcpu_find_address(cpu_online_mask, stap());
 396         if (!pcpu)
 397                 /* Use the first online cpu. */
 398                 pcpu = pcpu_devices + cpumask_first(cpu_online_mask);
 399         pcpu_delegate(pcpu, func, data, (unsigned long) restart_stack);
 400 }
 401
 402 /*
 403  * Call function on the ipl CPU.
 404  */
 405 void smp_call_ipl_cpu(void (*func)(void *), void *data)
 406 {
 407         struct lowcore *lc = lowcore_ptr[0];
 408
 409         if (pcpu_devices[0].address == stap())
 410                 lc = &S390_lowcore;
 411
 412         pcpu_delegate(&pcpu_devices[0], func, data,
 413                       lc->nodat_stack);
 414 }
 415
 416 int smp_find_processor_id(u16 address)
 417 {
 418         int cpu;
 419
 420         for_each_present_cpu(cpu)
 421                 if (pcpu_devices[cpu].address == address)
 422                         return cpu;
 423         return -1;
 424 }
 425
 426 void schedule_mcck_handler(void)
 427 {
 428         pcpu_ec_call(pcpu_devices + smp_processor_id(), ec_mcck_pending);
 429 }
 430
 431 bool notrace arch_vcpu_is_preempted(int cpu)
 432 {
 433         if (test_cpu_flag_of(CIF_ENABLED_WAIT, cpu))
 434                 return false;
 435         if (pcpu_running(pcpu_devices + cpu))
 436                 return false;
 437         return true;
 438 }
 439 EXPORT_SYMBOL(arch_vcpu_is_preempted);
 440
 441 void notrace smp_yield_cpu(int cpu)
 442 {
 443         if (!MACHINE_HAS_DIAG9C)
 444                 return;
 445         diag_stat_inc_norecursion(DIAG_STAT_X09C);
 446         asm volatile("diag %0,0,0x9c"
 447                      : : "d" (pcpu_devices[cpu].address));
 448 }
 449 EXPORT_SYMBOL_GPL(smp_yield_cpu);
 450
 451 /*
 452  * Send cpus emergency shutdown signal. This gives the cpus the
 453  * opportunity to complete outstanding interrupts.
 454  */
 455 void notrace smp_emergency_stop(void)
 456 {
 457         static arch_spinlock_t lock = __ARCH_SPIN_LOCK_UNLOCKED;
 458         static cpumask_t cpumask;
 459         u64 end;
 460         int cpu;
 461
 462         arch_spin_lock(&lock);
 463         cpumask_copy(&cpumask, cpu_online_mask);
 464         cpumask_clear_cpu(smp_processor_id(), &cpumask);
 465
 466         end = get_tod_clock() + (1000000UL << 12);
 467         for_each_cpu(cpu, &cpumask) {
 468                 struct pcpu *pcpu = pcpu_devices + cpu;
 469                 set_bit(ec_stop_cpu, &pcpu->ec_mask);
 470                 while (__pcpu_sigp(pcpu->address, SIGP_EMERGENCY_SIGNAL,
 471                                    0, NULL) == SIGP_CC_BUSY &&
 472                        get_tod_clock() < end)
 473                         cpu_relax();
 474         }
 475         while (get_tod_clock() < end) {
 476                 for_each_cpu(cpu, &cpumask)
 477                         if (pcpu_stopped(pcpu_devices + cpu))
 478                                 cpumask_clear_cpu(cpu, &cpumask);
 479                 if (cpumask_empty(&cpumask))
 480                         break;
 481                 cpu_relax();
 482         }
 483         arch_spin_unlock(&lock);
 484 }
 485 NOKPROBE_SYMBOL(smp_emergency_stop);
 486
 487 /*
 488  * Stop all cpus but the current one.
 489  */
 490 void smp_send_stop(void)
 491 {
 492         int cpu;
 493
 494         /* Disable all interrupts/machine checks */
 495         __load_psw_mask(PSW_KERNEL_BITS);
 496         trace_hardirqs_off();
 497
 498         debug_set_critical();
 499
 500         if (oops_in_progress)
 501                 smp_emergency_stop();
 502
 503         /* stop all processors */
 504         for_each_online_cpu(cpu) {
 505                 if (cpu == smp_processor_id())
 506                         continue;
 507                 pcpu_sigp_retry(pcpu_devices + cpu, SIGP_STOP, 0);
 508                 while (!pcpu_stopped(pcpu_devices + cpu))
 509                         cpu_relax();
 510         }
 511 }
 512
 513 /*
 514  * This is the main routine where commands issued by other
 515  * cpus are handled.
 516  */
 517 static void smp_handle_ext_call(void)
 518 {
 519         unsigned long bits;
 520
 521         /* handle bit signal external calls */
 522         bits = xchg(&pcpu_devices[smp_processor_id()].ec_mask, 0);
 523         if (test_bit(ec_stop_cpu, &bits))
 524                 smp_stop_cpu();
 525         if (test_bit(ec_schedule, &bits))
 526                 scheduler_ipi();
 527         if (test_bit(ec_call_function_single, &bits))
 528                 generic_smp_call_function_single_interrupt();
 529         if (test_bit(ec_mcck_pending, &bits))
 530                 s390_handle_mcck();
 531         if (test_bit(ec_irq_work, &bits))
 532                 irq_work_run();
 533 }
 534
 535 static void do_ext_call_interrupt(struct ext_code ext_code,
 536                                   unsigned int param32, unsigned long param64)
 537 {
 538         inc_irq_stat(ext_code.code == 0x1202 ? IRQEXT_EXC : IRQEXT_EMS);
 539         smp_handle_ext_call();
 540 }
 541
 542 void arch_send_call_function_ipi_mask(const struct cpumask *mask)
 543 {
 544         int cpu;
 545
 546         for_each_cpu(cpu, mask)
 547                 pcpu_ec_call(pcpu_devices + cpu, ec_call_function_single);
 548 }
 549
 550 void arch_send_call_function_single_ipi(int cpu)
 551 {
 552         pcpu_ec_call(pcpu_devices + cpu, ec_call_function_single);
 553 }
 554
 555 /*
 556  * this function sends a 'reschedule' IPI to another CPU.
 557  * it goes straight through and wastes no time serializing
 558  * anything. Worst case is that we lose a reschedule ...
 559  */
 560 void arch_smp_send_reschedule(int cpu)
 561 {
 562         pcpu_ec_call(pcpu_devices + cpu, ec_schedule);
 563 }
 564
 565 #ifdef CONFIG_IRQ_WORK
 566 void arch_irq_work_raise(void)
 567 {
 568         pcpu_ec_call(pcpu_devices + smp_processor_id(), ec_irq_work);
 569 }
 570 #endif
 571
 572 #ifdef CONFIG_CRASH_DUMP
 573
 574 int smp_store_status(int cpu)
 575 {
 576         struct lowcore *lc;
 577         struct pcpu *pcpu;
 578         unsigned long pa;
 579
 580         pcpu = pcpu_devices + cpu;
 581         lc = lowcore_ptr[cpu];
 582         pa = __pa(&lc->floating_pt_save_area);
 583         if (__pcpu_sigp_relax(pcpu->address, SIGP_STORE_STATUS_AT_ADDRESS,
 584                               pa) != SIGP_CC_ORDER_CODE_ACCEPTED)
 585                 return -EIO;
 586         if (!cpu_has_vx() && !MACHINE_HAS_GS)
 587                 return 0;
 588         pa = lc->mcesad & MCESA_ORIGIN_MASK;
 589         if (MACHINE_HAS_GS)
 590                 pa |= lc->mcesad & MCESA_LC_MASK;
 591         if (__pcpu_sigp_relax(pcpu->address, SIGP_STORE_ADDITIONAL_STATUS,
 592                               pa) != SIGP_CC_ORDER_CODE_ACCEPTED)
 593                 return -EIO;
 594         return 0;
 595 }
 596
 597 /*
 598  * Collect CPU state of the previous, crashed system.
 599  * There are four cases:
 600  * 1) standard zfcp/nvme dump
 601  *    condition: OLDMEM_BASE == NULL && is_ipl_type_dump() == true
 602  *    The state for all CPUs except the boot CPU needs to be collected
 603  *    with sigp stop-and-store-status. The boot CPU state is located in
 604  *    the absolute lowcore of the memory stored in the HSA. The zcore code
 605  *    will copy the boot CPU state from the HSA.
 606  * 2) stand-alone kdump for SCSI/NVMe (zfcp/nvme dump with swapped memory)
 607  *    condition: OLDMEM_BASE != NULL && is_ipl_type_dump() == true
 608  *    The state for all CPUs except the boot CPU needs to be collected
 609  *    with sigp stop-and-store-status. The firmware or the boot-loader
 610  *    stored the registers of the boot CPU in the absolute lowcore in the
 611  *    memory of the old system.
 612  * 3) kdump and the old kernel did not store the CPU state,
 613  *    or stand-alone kdump for DASD
 614  *    condition: OLDMEM_BASE != NULL && !is_kdump_kernel()
 615  *    The state for all CPUs except the boot CPU needs to be collected
 616  *    with sigp stop-and-store-status. The kexec code or the boot-loader
 617  *    stored the registers of the boot CPU in the memory of the old system.
 618  * 4) kdump and the old kernel stored the CPU state
 619  *    condition: OLDMEM_BASE != NULL && is_kdump_kernel()
 620  *    This case does not exist for s390 anymore, setup_arch explicitly
 621  *    deactivates the elfcorehdr= kernel parameter
 622  */
 623 static bool dump_available(void)
 624 {
 625         return oldmem_data.start || is_ipl_type_dump();
 626 }
 627
 628 void __init smp_save_dump_ipl_cpu(void)
 629 {
 630         struct save_area *sa;
 631         void *regs;
 632
 633         if (!dump_available())
 634                 return;
 635         sa = save_area_alloc(true);
 636         regs = memblock_alloc(512, 8);
 637         if (!sa || !regs)
 638                 panic("could not allocate memory for boot CPU save area\n");
 639         copy_oldmem_kernel(regs, __LC_FPREGS_SAVE_AREA, 512);
 640         save_area_add_regs(sa, regs);
 641         memblock_free(regs, 512);
 642         if (cpu_has_vx())
 643                 save_area_add_vxrs(sa, boot_cpu_vector_save_area);
 644 }
 645
 646 void __init smp_save_dump_secondary_cpus(void)
 647 {
 648         int addr, boot_cpu_addr, max_cpu_addr;
 649         struct save_area *sa;
 650         void *page;
 651
 652         if (!dump_available())
 653                 return;
 654         /* Allocate a page as dumping area for the store status sigps */
 655         page = memblock_alloc_low(PAGE_SIZE, PAGE_SIZE);
 656         if (!page)
 657                 panic("ERROR: Failed to allocate %lx bytes below %lx\n",
 658                       PAGE_SIZE, 1UL << 31);
 659
 660         /* Set multi-threading state to the previous system. */
 661         pcpu_set_smt(sclp.mtid_prev);
 662         boot_cpu_addr = stap();
 663         max_cpu_addr = SCLP_MAX_CORES << sclp.mtid_prev;
 664         for (addr = 0; addr <= max_cpu_addr; addr++) {
 665                 if (addr == boot_cpu_addr)
 666                         continue;
 667                 if (__pcpu_sigp_relax(addr, SIGP_SENSE, 0) ==
 668                     SIGP_CC_NOT_OPERATIONAL)
 669                         continue;
 670                 sa = save_area_alloc(false);
 671                 if (!sa)
 672                         panic("could not allocate memory for save area\n");
 673                 __pcpu_sigp_relax(addr, SIGP_STORE_STATUS_AT_ADDRESS, __pa(page));
 674                 save_area_add_regs(sa, page);
 675                 if (cpu_has_vx()) {
 676                         __pcpu_sigp_relax(addr, SIGP_STORE_ADDITIONAL_STATUS, __pa(page));
 677                         save_area_add_vxrs(sa, page);
 678                 }
 679         }
 680         memblock_free(page, PAGE_SIZE);
 681         diag_amode31_ops.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 int smp_cpu_get_cpu_address(int cpu)
 697 {
 698         return pcpu_devices[cpu].address;
 699 }
 700
 701 static void __ref smp_get_core_info(struct sclp_core_info *info, int early)
 702 {
 703         static int use_sigp_detection;
 704         int address;
 705
 706         if (use_sigp_detection || sclp_get_core_info(info, early)) {
 707                 use_sigp_detection = 1;
 708                 for (address = 0;
 709                      address < (SCLP_MAX_CORES << smp_cpu_mt_shift);
 710                      address += (1U << smp_cpu_mt_shift)) {
 711                         if (__pcpu_sigp_relax(address, SIGP_SENSE, 0) ==
 712                             SIGP_CC_NOT_OPERATIONAL)
 713                                 continue;
 714                         info->core[info->configured].core_id =
 715                                 address >> smp_cpu_mt_shift;
 716                         info->configured++;
 717                 }
 718                 info->combined = info->configured;
 719         }
 720 }
 721
 722 static int smp_add_present_cpu(int cpu);
 723
 724 static int smp_add_core(struct sclp_core_entry *core, cpumask_t *avail,
 725                         bool configured, bool early)
 726 {
 727         struct pcpu *pcpu;
 728         int cpu, nr, i;
 729         u16 address;
 730
 731         nr = 0;
 732         if (sclp.has_core_type && core->type != boot_core_type)
 733                 return nr;
 734         cpu = cpumask_first(avail);
 735         address = core->core_id << smp_cpu_mt_shift;
 736         for (i = 0; (i <= smp_cpu_mtid) && (cpu < nr_cpu_ids); i++) {
 737                 if (pcpu_find_address(cpu_present_mask, address + i))
 738                         continue;
 739                 pcpu = pcpu_devices + cpu;
 740                 pcpu->address = address + i;
 741                 if (configured)
 742                         pcpu->state = CPU_STATE_CONFIGURED;
 743                 else
 744                         pcpu->state = CPU_STATE_STANDBY;
 745                 smp_cpu_set_polarization(cpu, POLARIZATION_UNKNOWN);
 746                 set_cpu_present(cpu, true);
 747                 if (!early && smp_add_present_cpu(cpu) != 0)
 748                         set_cpu_present(cpu, false);
 749                 else
 750                         nr++;
 751                 cpumask_clear_cpu(cpu, avail);
 752                 cpu = cpumask_next(cpu, avail);
 753         }
 754         return nr;
 755 }
 756
 757 static int __smp_rescan_cpus(struct sclp_core_info *info, bool early)
 758 {
 759         struct sclp_core_entry *core;
 760         static cpumask_t avail;
 761         bool configured;
 762         u16 core_id;
 763         int nr, i;
 764
 765         cpus_read_lock();
 766         mutex_lock(&smp_cpu_state_mutex);
 767         nr = 0;
 768         cpumask_xor(&avail, cpu_possible_mask, cpu_present_mask);
 769         /*
 770          * Add IPL core first (which got logical CPU number 0) to make sure
 771          * that all SMT threads get subsequent logical CPU numbers.
 772          */
 773         if (early) {
 774                 core_id = pcpu_devices[0].address >> smp_cpu_mt_shift;
 775                 for (i = 0; i < info->configured; i++) {
 776                         core = &info->core[i];
 777                         if (core->core_id == core_id) {
 778                                 nr += smp_add_core(core, &avail, true, early);
 779                                 break;
 780                         }
 781                 }
 782         }
 783         for (i = 0; i < info->combined; i++) {
 784                 configured = i < info->configured;
 785                 nr += smp_add_core(&info->core[i], &avail, configured, early);
 786         }
 787         mutex_unlock(&smp_cpu_state_mutex);
 788         cpus_read_unlock();
 789         return nr;
 790 }
 791
 792 void __init smp_detect_cpus(void)
 793 {
 794         unsigned int cpu, mtid, c_cpus, s_cpus;
 795         struct sclp_core_info *info;
 796         u16 address;
 797
 798         /* Get CPU information */
 799         info = memblock_alloc(sizeof(*info), 8);
 800         if (!info)
 801                 panic("%s: Failed to allocate %zu bytes align=0x%x\n",
 802                       __func__, sizeof(*info), 8);
 803         smp_get_core_info(info, 1);
 804         /* Find boot CPU type */
 805         if (sclp.has_core_type) {
 806                 address = stap();
 807                 for (cpu = 0; cpu < info->combined; cpu++)
 808                         if (info->core[cpu].core_id == address) {
 809                                 /* The boot cpu dictates the cpu type. */
 810                                 boot_core_type = info->core[cpu].type;
 811                                 break;
 812                         }
 813                 if (cpu >= info->combined)
 814                         panic("Could not find boot CPU type");
 815         }
 816
 817         /* Set multi-threading state for the current system */
 818         mtid = boot_core_type ? sclp.mtid : sclp.mtid_cp;
 819         mtid = (mtid < smp_max_threads) ? mtid : smp_max_threads - 1;
 820         pcpu_set_smt(mtid);
 821
 822         /* Print number of CPUs */
 823         c_cpus = s_cpus = 0;
 824         for (cpu = 0; cpu < info->combined; cpu++) {
 825                 if (sclp.has_core_type &&
 826                     info->core[cpu].type != boot_core_type)
 827                         continue;
 828                 if (cpu < info->configured)
 829                         c_cpus += smp_cpu_mtid + 1;
 830                 else
 831                         s_cpus += smp_cpu_mtid + 1;
 832         }
 833         pr_info("%d configured CPUs, %d standby CPUs\n", c_cpus, s_cpus);
 834
 835         /* Add CPUs present at boot */
 836         __smp_rescan_cpus(info, true);
 837         memblock_free(info, sizeof(*info));
 838 }
 839
 840 /*
 841  *      Activate a secondary processor.
 842  */
 843 static void smp_start_secondary(void *cpuvoid)
 844 {
 845         int cpu = raw_smp_processor_id();
 846
 847         S390_lowcore.last_update_clock = get_tod_clock();
 848         S390_lowcore.restart_stack = (unsigned long)restart_stack;
 849         S390_lowcore.restart_fn = (unsigned long)do_restart;
 850         S390_lowcore.restart_data = 0;
 851         S390_lowcore.restart_source = -1U;
 852         S390_lowcore.restart_flags = 0;
 853         restore_access_regs(S390_lowcore.access_regs_save_area);
 854         cpu_init();
 855         rcutree_report_cpu_starting(cpu);
 856         init_cpu_timer();
 857         vtime_init();
 858         vdso_getcpu_init();
 859         pfault_init();
 860         cpumask_set_cpu(cpu, &cpu_setup_mask);
 861         update_cpu_masks();
 862         notify_cpu_starting(cpu);
 863         if (topology_cpu_dedicated(cpu))
 864                 set_cpu_flag(CIF_DEDICATED_CPU);
 865         else
 866                 clear_cpu_flag(CIF_DEDICATED_CPU);
 867         set_cpu_online(cpu, true);
 868         inc_irq_stat(CPU_RST);
 869         local_irq_enable();
 870         cpu_startup_entry(CPUHP_AP_ONLINE_IDLE);
 871 }
 872
 873 /* Upping and downing of CPUs */
 874 int __cpu_up(unsigned int cpu, struct task_struct *tidle)
 875 {
 876         struct pcpu *pcpu = pcpu_devices + cpu;
 877         int rc;
 878
 879         if (pcpu->state != CPU_STATE_CONFIGURED)
 880                 return -EIO;
 881         if (pcpu_sigp_retry(pcpu, SIGP_INITIAL_CPU_RESET, 0) !=
 882             SIGP_CC_ORDER_CODE_ACCEPTED)
 883                 return -EIO;
 884
 885         rc = pcpu_alloc_lowcore(pcpu, cpu);
 886         if (rc)
 887                 return rc;
 888         /*
 889          * Make sure global control register contents do not change
 890          * until new CPU has initialized control registers.
 891          */
 892         system_ctlreg_lock();
 893         pcpu_prepare_secondary(pcpu, cpu);
 894         pcpu_attach_task(pcpu, tidle);
 895         pcpu_start_fn(pcpu, smp_start_secondary, NULL);
 896         /* Wait until cpu puts itself in the online & active maps */
 897         while (!cpu_online(cpu))
 898                 cpu_relax();
 899         system_ctlreg_unlock();
 900         return 0;
 901 }
 902
 903 static unsigned int setup_possible_cpus __initdata;
 904
 905 static int __init _setup_possible_cpus(char *s)
 906 {
 907         get_option(&s, &setup_possible_cpus);
 908         return 0;
 909 }
 910 early_param("possible_cpus", _setup_possible_cpus);
 911
 912 int __cpu_disable(void)
 913 {
 914         struct ctlreg cregs[16];
 915         int cpu;
 916
 917         /* Handle possible pending IPIs */
 918         smp_handle_ext_call();
 919         cpu = smp_processor_id();
 920         set_cpu_online(cpu, false);
 921         cpumask_clear_cpu(cpu, &cpu_setup_mask);
 922         update_cpu_masks();
 923         /* Disable pseudo page faults on this cpu. */
 924         pfault_fini();
 925         /* Disable interrupt sources via control register. */
 926         __local_ctl_store(0, 15, cregs);
 927         cregs[0].val  &= ~0x0000ee70UL; /* disable all external interrupts */
 928         cregs[6].val  &= ~0xff000000UL; /* disable all I/O interrupts */
 929         cregs[14].val &= ~0x1f000000UL; /* disable most machine checks */
 930         __local_ctl_load(0, 15, cregs);
 931         clear_cpu_flag(CIF_NOHZ_DELAY);
 932         return 0;
 933 }
 934
 935 void __cpu_die(unsigned int cpu)
 936 {
 937         struct pcpu *pcpu;
 938
 939         /* Wait until target cpu is down */
 940         pcpu = pcpu_devices + cpu;
 941         while (!pcpu_stopped(pcpu))
 942                 cpu_relax();
 943         pcpu_free_lowcore(pcpu);
 944         cpumask_clear_cpu(cpu, mm_cpumask(&init_mm));
 945         cpumask_clear_cpu(cpu, &init_mm.context.cpu_attach_mask);
 946 }
 947
 948 void __noreturn cpu_die(void)
 949 {
 950         idle_task_exit();
 951         pcpu_sigp_retry(pcpu_devices + smp_processor_id(), SIGP_STOP, 0);
 952         for (;;) ;
 953 }
 954
 955 void __init smp_fill_possible_mask(void)
 956 {
 957         unsigned int possible, sclp_max, cpu;
 958
 959         sclp_max = max(sclp.mtid, sclp.mtid_cp) + 1;
 960         sclp_max = min(smp_max_threads, sclp_max);
 961         sclp_max = (sclp.max_cores * sclp_max) ?: nr_cpu_ids;
 962         possible = setup_possible_cpus ?: nr_cpu_ids;
 963         possible = min(possible, sclp_max);
 964         for (cpu = 0; cpu < possible && cpu < nr_cpu_ids; cpu++)
 965                 set_cpu_possible(cpu, true);
 966 }
 967
 968 void __init smp_prepare_cpus(unsigned int max_cpus)
 969 {
 970         if (register_external_irq(EXT_IRQ_EMERGENCY_SIG, do_ext_call_interrupt))
 971                 panic("Couldn't request external interrupt 0x1201");
 972         system_ctl_set_bit(0, 14);
 973         if (register_external_irq(EXT_IRQ_EXTERNAL_CALL, do_ext_call_interrupt))
 974                 panic("Couldn't request external interrupt 0x1202");
 975         system_ctl_set_bit(0, 13);
 976 }
 977
 978 void __init smp_prepare_boot_cpu(void)
 979 {
 980         struct pcpu *pcpu = pcpu_devices;
 981
 982         WARN_ON(!cpu_present(0) || !cpu_online(0));
 983         pcpu->state = CPU_STATE_CONFIGURED;
 984         S390_lowcore.percpu_offset = __per_cpu_offset[0];
 985         smp_cpu_set_polarization(0, POLARIZATION_UNKNOWN);
 986 }
 987
 988 void __init smp_setup_processor_id(void)
 989 {
 990         pcpu_devices[0].address = stap();
 991         S390_lowcore.cpu_nr = 0;
 992         S390_lowcore.spinlock_lockval = arch_spin_lockval(0);
 993         S390_lowcore.spinlock_index = 0;
 994 }
 995
 996 /*
 997  * the frequency of the profiling timer can be changed
 998  * by writing a multiplier value into /proc/profile.
 999  *
1000  * usually you want to run this on all CPUs ;)
1001  */
1002 int setup_profiling_timer(unsigned int multiplier)
1003 {
1004         return 0;
1005 }
1006
1007 static ssize_t cpu_configure_show(struct device *dev,
1008                                   struct device_attribute *attr, char *buf)
1009 {
1010         ssize_t count;
1011
1012         mutex_lock(&smp_cpu_state_mutex);
1013         count = sprintf(buf, "%d\n", pcpu_devices[dev->id].state);
1014         mutex_unlock(&smp_cpu_state_mutex);
1015         return count;
1016 }
1017
1018 static ssize_t cpu_configure_store(struct device *dev,
1019                                    struct device_attribute *attr,
1020                                    const char *buf, size_t count)
1021 {
1022         struct pcpu *pcpu;
1023         int cpu, val, rc, i;
1024         char delim;
1025
1026         if (sscanf(buf, "%d %c", &val, &delim) != 1)
1027                 return -EINVAL;
1028         if (val != 0 && val != 1)
1029                 return -EINVAL;
1030         cpus_read_lock();
1031         mutex_lock(&smp_cpu_state_mutex);
1032         rc = -EBUSY;
1033         /* disallow configuration changes of online cpus */
1034         cpu = dev->id;
1035         cpu = smp_get_base_cpu(cpu);
1036         for (i = 0; i <= smp_cpu_mtid; i++)
1037                 if (cpu_online(cpu + i))
1038                         goto out;
1039         pcpu = pcpu_devices + cpu;
1040         rc = 0;
1041         switch (val) {
1042         case 0:
1043                 if (pcpu->state != CPU_STATE_CONFIGURED)
1044                         break;
1045                 rc = sclp_core_deconfigure(pcpu->address >> smp_cpu_mt_shift);
1046                 if (rc)
1047                         break;
1048                 for (i = 0; i <= smp_cpu_mtid; i++) {
1049                         if (cpu + i >= nr_cpu_ids || !cpu_present(cpu + i))
1050                                 continue;
1051                         pcpu[i].state = CPU_STATE_STANDBY;
1052                         smp_cpu_set_polarization(cpu + i,
1053                                                  POLARIZATION_UNKNOWN);
1054                 }
1055                 topology_expect_change();
1056                 break;
1057         case 1:
1058                 if (pcpu->state != CPU_STATE_STANDBY)
1059                         break;
1060                 rc = sclp_core_configure(pcpu->address >> smp_cpu_mt_shift);
1061                 if (rc)
1062                         break;
1063                 for (i = 0; i <= smp_cpu_mtid; i++) {
1064                         if (cpu + i >= nr_cpu_ids || !cpu_present(cpu + i))
1065                                 continue;
1066                         pcpu[i].state = CPU_STATE_CONFIGURED;
1067                         smp_cpu_set_polarization(cpu + i,
1068                                                  POLARIZATION_UNKNOWN);
1069                 }
1070                 topology_expect_change();
1071                 break;
1072         default:
1073                 break;
1074         }
1075 out:
1076         mutex_unlock(&smp_cpu_state_mutex);
1077         cpus_read_unlock();
1078         return rc ? rc : count;
1079 }
1080 static DEVICE_ATTR(configure, 0644, cpu_configure_show, cpu_configure_store);
1081
1082 static ssize_t show_cpu_address(struct device *dev,
1083                                 struct device_attribute *attr, char *buf)
1084 {
1085         return sprintf(buf, "%d\n", pcpu_devices[dev->id].address);
1086 }
1087 static DEVICE_ATTR(address, 0444, show_cpu_address, NULL);
1088
1089 static struct attribute *cpu_common_attrs[] = {
1090         &dev_attr_configure.attr,
1091         &dev_attr_address.attr,
1092         NULL,
1093 };
1094
1095 static struct attribute_group cpu_common_attr_group = {
1096         .attrs = cpu_common_attrs,
1097 };
1098
1099 static struct attribute *cpu_online_attrs[] = {
1100         &dev_attr_idle_count.attr,
1101         &dev_attr_idle_time_us.attr,
1102         NULL,
1103 };
1104
1105 static struct attribute_group cpu_online_attr_group = {
1106         .attrs = cpu_online_attrs,
1107 };
1108
1109 static int smp_cpu_online(unsigned int cpu)
1110 {
1111         struct device *s = &per_cpu(cpu_device, cpu)->dev;
1112
1113         return sysfs_create_group(&s->kobj, &cpu_online_attr_group);
1114 }
1115
1116 static int smp_cpu_pre_down(unsigned int cpu)
1117 {
1118         struct device *s = &per_cpu(cpu_device, cpu)->dev;
1119
1120         sysfs_remove_group(&s->kobj, &cpu_online_attr_group);
1121         return 0;
1122 }
1123
1124 static int smp_add_present_cpu(int cpu)
1125 {
1126         struct device *s;
1127         struct cpu *c;
1128         int rc;
1129
1130         c = kzalloc(sizeof(*c), GFP_KERNEL);
1131         if (!c)
1132                 return -ENOMEM;
1133         per_cpu(cpu_device, cpu) = c;
1134         s = &c->dev;
1135         c->hotpluggable = !!cpu;
1136         rc = register_cpu(c, cpu);
1137         if (rc)
1138                 goto out;
1139         rc = sysfs_create_group(&s->kobj, &cpu_common_attr_group);
1140         if (rc)
1141                 goto out_cpu;
1142         rc = topology_cpu_init(c);
1143         if (rc)
1144                 goto out_topology;
1145         return 0;
1146
1147 out_topology:
1148         sysfs_remove_group(&s->kobj, &cpu_common_attr_group);
1149 out_cpu:
1150         unregister_cpu(c);
1151 out:
1152         return rc;
1153 }
1154
1155 int __ref smp_rescan_cpus(void)
1156 {
1157         struct sclp_core_info *info;
1158         int nr;
1159
1160         info = kzalloc(sizeof(*info), GFP_KERNEL);
1161         if (!info)
1162                 return -ENOMEM;
1163         smp_get_core_info(info, 0);
1164         nr = __smp_rescan_cpus(info, false);
1165         kfree(info);
1166         if (nr)
1167                 topology_schedule_update();
1168         return 0;
1169 }
1170
1171 static ssize_t __ref rescan_store(struct device *dev,
1172                                   struct device_attribute *attr,
1173                                   const char *buf,
1174                                   size_t count)
1175 {
1176         int rc;
1177
1178         rc = lock_device_hotplug_sysfs();
1179         if (rc)
1180                 return rc;
1181         rc = smp_rescan_cpus();
1182         unlock_device_hotplug();
1183         return rc ? rc : count;
1184 }
1185 static DEVICE_ATTR_WO(rescan);
1186
1187 static int __init s390_smp_init(void)
1188 {
1189         struct device *dev_root;
1190         int cpu, rc = 0;
1191
1192         dev_root = bus_get_dev_root(&cpu_subsys);
1193         if (dev_root) {
1194                 rc = device_create_file(dev_root, &dev_attr_rescan);
1195                 put_device(dev_root);
1196                 if (rc)
1197                         return rc;
1198         }
1199
1200         for_each_present_cpu(cpu) {
1201                 rc = smp_add_present_cpu(cpu);
1202                 if (rc)
1203                         goto out;
1204         }
1205
1206         rc = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "s390/smp:online",
1207                                smp_cpu_online, smp_cpu_pre_down);
1208         rc = rc <= 0 ? rc : 0;
1209 out:
1210         return rc;
1211 }
1212 subsys_initcall(s390_smp_init);