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/memblock.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/kprobes.h>
  39 #include <asm/asm-offsets.h>
  40 #include <asm/diag.h>
  41 #include <asm/switch_to.h>
  42 #include <asm/facility.h>
  43 #include <asm/ipl.h>
  44 #include <asm/setup.h>
  45 #include <asm/irq.h>
  46 #include <asm/tlbflush.h>
  47 #include <asm/vtimer.h>
  48 #include <asm/lowcore.h>
  49 #include <asm/sclp.h>
  50 #include <asm/debug.h>
  51 #include <asm/os_info.h>
  52 #include <asm/sigp.h>
  53 #include <asm/idle.h>
  54 #include <asm/nmi.h>
  55 #include <asm/stacktrace.h>
  56 #include <asm/topology.h>
  57 #include <asm/vdso.h>
  58 #include "entry.h"
  59
  60 enum {
  61         ec_schedule = 0,
  62         ec_call_function_single,
  63         ec_stop_cpu,
  64         ec_mcck_pending,
  65 };
  66
  67 enum {
  68         CPU_STATE_STANDBY,
  69         CPU_STATE_CONFIGURED,
  70 };
  71
  72 static DEFINE_PER_CPU(struct cpu *, cpu_device);
  73
  74 struct pcpu {
  75         struct lowcore *lowcore;        /* lowcore page(s) for the cpu */
  76         unsigned long ec_mask;          /* bit mask for ec_xxx functions */
  77         unsigned long ec_clk;           /* sigp timestamp for ec_xxx */
  78         signed char state;              /* physical cpu state */
  79         signed char polarization;       /* physical polarization */
  80         u16 address;                    /* physical cpu address */
  81 };
  82
  83 static u8 boot_core_type;
  84 static struct pcpu pcpu_devices[NR_CPUS];
  85
  86 unsigned int smp_cpu_mt_shift;
  87 EXPORT_SYMBOL(smp_cpu_mt_shift);
  88
  89 unsigned int smp_cpu_mtid;
  90 EXPORT_SYMBOL(smp_cpu_mtid);
  91
  92 #ifdef CONFIG_CRASH_DUMP
  93 __vector128 __initdata boot_cpu_vector_save_area[__NUM_VXRS];
  94 #endif
  95
  96 static unsigned int smp_max_threads __initdata = -1U;
  97
  98 static int __init early_nosmt(char *s)
  99 {
 100         smp_max_threads = 1;
 101         return 0;
 102 }
 103 early_param("nosmt", early_nosmt);
 104
 105 static int __init early_smt(char *s)
 106 {
 107         get_option(&s, &smp_max_threads);
 108         return 0;
 109 }
 110 early_param("smt", early_smt);
 111
 112 /*
 113  * The smp_cpu_state_mutex must be held when changing the state or polarization
 114  * member of a pcpu data structure within the pcpu_devices arreay.
 115  */
 116 DEFINE_MUTEX(smp_cpu_state_mutex);
 117
 118 /*
 119  * Signal processor helper functions.
 120  */
 121 static inline int __pcpu_sigp_relax(u16 addr, u8 order, unsigned long parm)
 122 {
 123         int cc;
 124
 125         while (1) {
 126                 cc = __pcpu_sigp(addr, order, parm, NULL);
 127                 if (cc != SIGP_CC_BUSY)
 128                         return cc;
 129                 cpu_relax();
 130         }
 131 }
 132
 133 static int pcpu_sigp_retry(struct pcpu *pcpu, u8 order, u32 parm)
 134 {
 135         int cc, retry;
 136
 137         for (retry = 0; ; retry++) {
 138                 cc = __pcpu_sigp(pcpu->address, order, parm, NULL);
 139                 if (cc != SIGP_CC_BUSY)
 140                         break;
 141                 if (retry >= 3)
 142                         udelay(10);
 143         }
 144         return cc;
 145 }
 146
 147 static inline int pcpu_stopped(struct pcpu *pcpu)
 148 {
 149         u32 status;
 150
 151         if (__pcpu_sigp(pcpu->address, SIGP_SENSE,
 152                         0, &status) != SIGP_CC_STATUS_STORED)
 153                 return 0;
 154         return !!(status & (SIGP_STATUS_CHECK_STOP|SIGP_STATUS_STOPPED));
 155 }
 156
 157 static inline int pcpu_running(struct pcpu *pcpu)
 158 {
 159         if (__pcpu_sigp(pcpu->address, SIGP_SENSE_RUNNING,
 160                         0, NULL) != SIGP_CC_STATUS_STORED)
 161                 return 1;
 162         /* Status stored condition code is equivalent to cpu not running. */
 163         return 0;
 164 }
 165
 166 /*
 167  * Find struct pcpu by cpu address.
 168  */
 169 static struct pcpu *pcpu_find_address(const struct cpumask *mask, u16 address)
 170 {
 171         int cpu;
 172
 173         for_each_cpu(cpu, mask)
 174                 if (pcpu_devices[cpu].address == address)
 175                         return pcpu_devices + cpu;
 176         return NULL;
 177 }
 178
 179 static void pcpu_ec_call(struct pcpu *pcpu, int ec_bit)
 180 {
 181         int order;
 182
 183         if (test_and_set_bit(ec_bit, &pcpu->ec_mask))
 184                 return;
 185         order = pcpu_running(pcpu) ? SIGP_EXTERNAL_CALL : SIGP_EMERGENCY_SIGNAL;
 186         pcpu->ec_clk = get_tod_clock_fast();
 187         pcpu_sigp_retry(pcpu, order, 0);
 188 }
 189
 190 static int pcpu_alloc_lowcore(struct pcpu *pcpu, int cpu)
 191 {
 192         unsigned long async_stack, nodat_stack;
 193         struct lowcore *lc;
 194
 195         if (pcpu != &pcpu_devices[0]) {
 196                 pcpu->lowcore = (struct lowcore *)
 197                         __get_free_pages(GFP_KERNEL | GFP_DMA, LC_ORDER);
 198                 nodat_stack = __get_free_pages(GFP_KERNEL, THREAD_SIZE_ORDER);
 199                 if (!pcpu->lowcore || !nodat_stack)
 200                         goto out;
 201         } else {
 202                 nodat_stack = pcpu->lowcore->nodat_stack - STACK_INIT_OFFSET;
 203         }
 204         async_stack = stack_alloc();
 205         if (!async_stack)
 206                 goto out;
 207         lc = pcpu->lowcore;
 208         memcpy(lc, &S390_lowcore, 512);
 209         memset((char *) lc + 512, 0, sizeof(*lc) - 512);
 210         lc->async_stack = async_stack + STACK_INIT_OFFSET;
 211         lc->nodat_stack = nodat_stack + STACK_INIT_OFFSET;
 212         lc->cpu_nr = cpu;
 213         lc->spinlock_lockval = arch_spin_lockval(cpu);
 214         lc->spinlock_index = 0;
 215         lc->br_r1_trampoline = 0x07f1;  /* br %r1 */
 216         lc->return_lpswe = gen_lpswe(__LC_RETURN_PSW);
 217         lc->return_mcck_lpswe = gen_lpswe(__LC_RETURN_MCCK_PSW);
 218         if (nmi_alloc_per_cpu(lc))
 219                 goto out_async;
 220         lowcore_ptr[cpu] = lc;
 221         pcpu_sigp_retry(pcpu, SIGP_SET_PREFIX, (u32)(unsigned long) lc);
 222         return 0;
 223
 224 out_async:
 225         stack_free(async_stack);
 226 out:
 227         if (pcpu != &pcpu_devices[0]) {
 228                 free_pages(nodat_stack, THREAD_SIZE_ORDER);
 229                 free_pages((unsigned long) pcpu->lowcore, LC_ORDER);
 230         }
 231         return -ENOMEM;
 232 }
 233
 234 static void pcpu_free_lowcore(struct pcpu *pcpu)
 235 {
 236         unsigned long async_stack, nodat_stack, lowcore;
 237
 238         nodat_stack = pcpu->lowcore->nodat_stack - STACK_INIT_OFFSET;
 239         async_stack = pcpu->lowcore->async_stack - STACK_INIT_OFFSET;
 240         lowcore = (unsigned long) pcpu->lowcore;
 241
 242         pcpu_sigp_retry(pcpu, SIGP_SET_PREFIX, 0);
 243         lowcore_ptr[pcpu - pcpu_devices] = NULL;
 244         nmi_free_per_cpu(pcpu->lowcore);
 245         stack_free(async_stack);
 246         if (pcpu == &pcpu_devices[0])
 247                 return;
 248         free_pages(nodat_stack, THREAD_SIZE_ORDER);
 249         free_pages(lowcore, LC_ORDER);
 250 }
 251
 252 static void pcpu_prepare_secondary(struct pcpu *pcpu, int cpu)
 253 {
 254         struct lowcore *lc = pcpu->lowcore;
 255
 256         cpumask_set_cpu(cpu, &init_mm.context.cpu_attach_mask);
 257         cpumask_set_cpu(cpu, mm_cpumask(&init_mm));
 258         lc->cpu_nr = cpu;
 259         lc->spinlock_lockval = arch_spin_lockval(cpu);
 260         lc->spinlock_index = 0;
 261         lc->percpu_offset = __per_cpu_offset[cpu];
 262         lc->kernel_asce = S390_lowcore.kernel_asce;
 263         lc->user_asce = s390_invalid_asce;
 264         lc->machine_flags = S390_lowcore.machine_flags;
 265         lc->user_timer = lc->system_timer =
 266                 lc->steal_timer = lc->avg_steal_timer = 0;
 267         __ctl_store(lc->cregs_save_area, 0, 15);
 268         lc->cregs_save_area[1] = lc->kernel_asce;
 269         lc->cregs_save_area[7] = lc->user_asce;
 270         save_access_regs((unsigned int *) lc->access_regs_save_area);
 271         memcpy(lc->stfle_fac_list, S390_lowcore.stfle_fac_list,
 272                sizeof(lc->stfle_fac_list));
 273         memcpy(lc->alt_stfle_fac_list, S390_lowcore.alt_stfle_fac_list,
 274                sizeof(lc->alt_stfle_fac_list));
 275         arch_spin_lock_setup(cpu);
 276 }
 277
 278 static void pcpu_attach_task(struct pcpu *pcpu, struct task_struct *tsk)
 279 {
 280         struct lowcore *lc = pcpu->lowcore;
 281
 282         lc->kernel_stack = (unsigned long) task_stack_page(tsk)
 283                 + THREAD_SIZE - STACK_FRAME_OVERHEAD - sizeof(struct pt_regs);
 284         lc->current_task = (unsigned long) tsk;
 285         lc->lpp = LPP_MAGIC;
 286         lc->current_pid = tsk->pid;
 287         lc->user_timer = tsk->thread.user_timer;
 288         lc->guest_timer = tsk->thread.guest_timer;
 289         lc->system_timer = tsk->thread.system_timer;
 290         lc->hardirq_timer = tsk->thread.hardirq_timer;
 291         lc->softirq_timer = tsk->thread.softirq_timer;
 292         lc->steal_timer = 0;
 293 }
 294
 295 static void pcpu_start_fn(struct pcpu *pcpu, void (*func)(void *), void *data)
 296 {
 297         struct lowcore *lc = pcpu->lowcore;
 298
 299         lc->restart_stack = lc->nodat_stack;
 300         lc->restart_fn = (unsigned long) func;
 301         lc->restart_data = (unsigned long) data;
 302         lc->restart_source = -1UL;
 303         pcpu_sigp_retry(pcpu, SIGP_RESTART, 0);
 304 }
 305
 306 /*
 307  * Call function via PSW restart on pcpu and stop the current cpu.
 308  */
 309 static void __pcpu_delegate(void (*func)(void*), void *data)
 310 {
 311         func(data);     /* should not return */
 312 }
 313
 314 static void __no_sanitize_address pcpu_delegate(struct pcpu *pcpu,
 315                                                 void (*func)(void *),
 316                                                 void *data, unsigned long stack)
 317 {
 318         struct lowcore *lc = lowcore_ptr[pcpu - pcpu_devices];
 319         unsigned long source_cpu = stap();
 320
 321         __load_psw_mask(PSW_KERNEL_BITS | PSW_MASK_DAT);
 322         if (pcpu->address == source_cpu)
 323                 CALL_ON_STACK(__pcpu_delegate, stack, 2, func, data);
 324         /* Stop target cpu (if func returns this stops the current cpu). */
 325         pcpu_sigp_retry(pcpu, SIGP_STOP, 0);
 326         /* Restart func on the target cpu and stop the current cpu. */
 327         mem_assign_absolute(lc->restart_stack, stack);
 328         mem_assign_absolute(lc->restart_fn, (unsigned long) func);
 329         mem_assign_absolute(lc->restart_data, (unsigned long) data);
 330         mem_assign_absolute(lc->restart_source, source_cpu);
 331         __bpon();
 332         asm volatile(
 333                 "0:     sigp    0,%0,%2 # sigp restart to target cpu\n"
 334                 "       brc     2,0b    # busy, try again\n"
 335                 "1:     sigp    0,%1,%3 # sigp stop to current cpu\n"
 336                 "       brc     2,1b    # busy, try again\n"
 337                 : : "d" (pcpu->address), "d" (source_cpu),
 338                     "K" (SIGP_RESTART), "K" (SIGP_STOP)
 339                 : "0", "1", "cc");
 340         for (;;) ;
 341 }
 342
 343 /*
 344  * Enable additional logical cpus for multi-threading.
 345  */
 346 static int pcpu_set_smt(unsigned int mtid)
 347 {
 348         int cc;
 349
 350         if (smp_cpu_mtid == mtid)
 351                 return 0;
 352         cc = __pcpu_sigp(0, SIGP_SET_MULTI_THREADING, mtid, NULL);
 353         if (cc == 0) {
 354                 smp_cpu_mtid = mtid;
 355                 smp_cpu_mt_shift = 0;
 356                 while (smp_cpu_mtid >= (1U << smp_cpu_mt_shift))
 357                         smp_cpu_mt_shift++;
 358                 pcpu_devices[0].address = stap();
 359         }
 360         return cc;
 361 }
 362
 363 /*
 364  * Call function on an online CPU.
 365  */
 366 void smp_call_online_cpu(void (*func)(void *), void *data)
 367 {
 368         struct pcpu *pcpu;
 369
 370         /* Use the current cpu if it is online. */
 371         pcpu = pcpu_find_address(cpu_online_mask, stap());
 372         if (!pcpu)
 373                 /* Use the first online cpu. */
 374                 pcpu = pcpu_devices + cpumask_first(cpu_online_mask);
 375         pcpu_delegate(pcpu, func, data, (unsigned long) restart_stack);
 376 }
 377
 378 /*
 379  * Call function on the ipl CPU.
 380  */
 381 void smp_call_ipl_cpu(void (*func)(void *), void *data)
 382 {
 383         struct lowcore *lc = pcpu_devices->lowcore;
 384
 385         if (pcpu_devices[0].address == stap())
 386                 lc = &S390_lowcore;
 387
 388         pcpu_delegate(&pcpu_devices[0], func, data,
 389                       lc->nodat_stack);
 390 }
 391
 392 int smp_find_processor_id(u16 address)
 393 {
 394         int cpu;
 395
 396         for_each_present_cpu(cpu)
 397                 if (pcpu_devices[cpu].address == address)
 398                         return cpu;
 399         return -1;
 400 }
 401
 402 void schedule_mcck_handler(void)
 403 {
 404         pcpu_ec_call(pcpu_devices + smp_processor_id(), ec_mcck_pending);
 405 }
 406
 407 bool notrace arch_vcpu_is_preempted(int cpu)
 408 {
 409         if (test_cpu_flag_of(CIF_ENABLED_WAIT, cpu))
 410                 return false;
 411         if (pcpu_running(pcpu_devices + cpu))
 412                 return false;
 413         return true;
 414 }
 415 EXPORT_SYMBOL(arch_vcpu_is_preempted);
 416
 417 void notrace smp_yield_cpu(int cpu)
 418 {
 419         if (!MACHINE_HAS_DIAG9C)
 420                 return;
 421         diag_stat_inc_norecursion(DIAG_STAT_X09C);
 422         asm volatile("diag %0,0,0x9c"
 423                      : : "d" (pcpu_devices[cpu].address));
 424 }
 425
 426 /*
 427  * Send cpus emergency shutdown signal. This gives the cpus the
 428  * opportunity to complete outstanding interrupts.
 429  */
 430 void notrace smp_emergency_stop(void)
 431 {
 432         cpumask_t cpumask;
 433         u64 end;
 434         int cpu;
 435
 436         cpumask_copy(&cpumask, cpu_online_mask);
 437         cpumask_clear_cpu(smp_processor_id(), &cpumask);
 438
 439         end = get_tod_clock() + (1000000UL << 12);
 440         for_each_cpu(cpu, &cpumask) {
 441                 struct pcpu *pcpu = pcpu_devices + cpu;
 442                 set_bit(ec_stop_cpu, &pcpu->ec_mask);
 443                 while (__pcpu_sigp(pcpu->address, SIGP_EMERGENCY_SIGNAL,
 444                                    0, NULL) == SIGP_CC_BUSY &&
 445                        get_tod_clock() < end)
 446                         cpu_relax();
 447         }
 448         while (get_tod_clock() < end) {
 449                 for_each_cpu(cpu, &cpumask)
 450                         if (pcpu_stopped(pcpu_devices + cpu))
 451                                 cpumask_clear_cpu(cpu, &cpumask);
 452                 if (cpumask_empty(&cpumask))
 453                         break;
 454                 cpu_relax();
 455         }
 456 }
 457 NOKPROBE_SYMBOL(smp_emergency_stop);
 458
 459 /*
 460  * Stop all cpus but the current one.
 461  */
 462 void smp_send_stop(void)
 463 {
 464         int cpu;
 465
 466         /* Disable all interrupts/machine checks */
 467         __load_psw_mask(PSW_KERNEL_BITS | PSW_MASK_DAT);
 468         trace_hardirqs_off();
 469
 470         debug_set_critical();
 471
 472         if (oops_in_progress)
 473                 smp_emergency_stop();
 474
 475         /* stop all processors */
 476         for_each_online_cpu(cpu) {
 477                 if (cpu == smp_processor_id())
 478                         continue;
 479                 pcpu_sigp_retry(pcpu_devices + cpu, SIGP_STOP, 0);
 480                 while (!pcpu_stopped(pcpu_devices + cpu))
 481                         cpu_relax();
 482         }
 483 }
 484
 485 /*
 486  * This is the main routine where commands issued by other
 487  * cpus are handled.
 488  */
 489 static void smp_handle_ext_call(void)
 490 {
 491         unsigned long bits;
 492
 493         /* handle bit signal external calls */
 494         bits = xchg(&pcpu_devices[smp_processor_id()].ec_mask, 0);
 495         if (test_bit(ec_stop_cpu, &bits))
 496                 smp_stop_cpu();
 497         if (test_bit(ec_schedule, &bits))
 498                 scheduler_ipi();
 499         if (test_bit(ec_call_function_single, &bits))
 500                 generic_smp_call_function_single_interrupt();
 501         if (test_bit(ec_mcck_pending, &bits))
 502                 __s390_handle_mcck();
 503 }
 504
 505 static void do_ext_call_interrupt(struct ext_code ext_code,
 506                                   unsigned int param32, unsigned long param64)
 507 {
 508         inc_irq_stat(ext_code.code == 0x1202 ? IRQEXT_EXC : IRQEXT_EMS);
 509         smp_handle_ext_call();
 510 }
 511
 512 void arch_send_call_function_ipi_mask(const struct cpumask *mask)
 513 {
 514         int cpu;
 515
 516         for_each_cpu(cpu, mask)
 517                 pcpu_ec_call(pcpu_devices + cpu, ec_call_function_single);
 518 }
 519
 520 void arch_send_call_function_single_ipi(int cpu)
 521 {
 522         pcpu_ec_call(pcpu_devices + cpu, ec_call_function_single);
 523 }
 524
 525 /*
 526  * this function sends a 'reschedule' IPI to another CPU.
 527  * it goes straight through and wastes no time serializing
 528  * anything. Worst case is that we lose a reschedule ...
 529  */
 530 void smp_send_reschedule(int cpu)
 531 {
 532         pcpu_ec_call(pcpu_devices + cpu, ec_schedule);
 533 }
 534
 535 /*
 536  * parameter area for the set/clear control bit callbacks
 537  */
 538 struct ec_creg_mask_parms {
 539         unsigned long orval;
 540         unsigned long andval;
 541         int cr;
 542 };
 543
 544 /*
 545  * callback for setting/clearing control bits
 546  */
 547 static void smp_ctl_bit_callback(void *info)
 548 {
 549         struct ec_creg_mask_parms *pp = info;
 550         unsigned long cregs[16];
 551
 552         __ctl_store(cregs, 0, 15);
 553         cregs[pp->cr] = (cregs[pp->cr] & pp->andval) | pp->orval;
 554         __ctl_load(cregs, 0, 15);
 555 }
 556
 557 /*
 558  * Set a bit in a control register of all cpus
 559  */
 560 void smp_ctl_set_bit(int cr, int bit)
 561 {
 562         struct ec_creg_mask_parms parms = { 1UL << bit, -1UL, cr };
 563
 564         on_each_cpu(smp_ctl_bit_callback, &parms, 1);
 565 }
 566 EXPORT_SYMBOL(smp_ctl_set_bit);
 567
 568 /*
 569  * Clear a bit in a control register of all cpus
 570  */
 571 void smp_ctl_clear_bit(int cr, int bit)
 572 {
 573         struct ec_creg_mask_parms parms = { 0, ~(1UL << bit), cr };
 574
 575         on_each_cpu(smp_ctl_bit_callback, &parms, 1);
 576 }
 577 EXPORT_SYMBOL(smp_ctl_clear_bit);
 578
 579 #ifdef CONFIG_CRASH_DUMP
 580
 581 int smp_store_status(int cpu)
 582 {
 583         struct pcpu *pcpu = pcpu_devices + cpu;
 584         unsigned long pa;
 585
 586         pa = __pa(&pcpu->lowcore->floating_pt_save_area);
 587         if (__pcpu_sigp_relax(pcpu->address, SIGP_STORE_STATUS_AT_ADDRESS,
 588                               pa) != SIGP_CC_ORDER_CODE_ACCEPTED)
 589                 return -EIO;
 590         if (!MACHINE_HAS_VX && !MACHINE_HAS_GS)
 591                 return 0;
 592         pa = __pa(pcpu->lowcore->mcesad & MCESA_ORIGIN_MASK);
 593         if (MACHINE_HAS_GS)
 594                 pa |= pcpu->lowcore->mcesad & MCESA_LC_MASK;
 595         if (__pcpu_sigp_relax(pcpu->address, SIGP_STORE_ADDITIONAL_STATUS,
 596                               pa) != SIGP_CC_ORDER_CODE_ACCEPTED)
 597                 return -EIO;
 598         return 0;
 599 }
 600
 601 /*
 602  * Collect CPU state of the previous, crashed system.
 603  * There are four cases:
 604  * 1) standard zfcp/nvme dump
 605  *    condition: OLDMEM_BASE == NULL && is_ipl_type_dump() == true
 606  *    The state for all CPUs except the boot CPU needs to be collected
 607  *    with sigp stop-and-store-status. The boot CPU state is located in
 608  *    the absolute lowcore of the memory stored in the HSA. The zcore code
 609  *    will copy the boot CPU state from the HSA.
 610  * 2) stand-alone kdump for SCSI/NVMe (zfcp/nvme dump with swapped memory)
 611  *    condition: OLDMEM_BASE != NULL && is_ipl_type_dump() == true
 612  *    The state for all CPUs except the boot CPU needs to be collected
 613  *    with sigp stop-and-store-status. The firmware or the boot-loader
 614  *    stored the registers of the boot CPU in the absolute lowcore in the
 615  *    memory of the old system.
 616  * 3) kdump and the old kernel did not store the CPU state,
 617  *    or stand-alone kdump for DASD
 618  *    condition: OLDMEM_BASE != NULL && !is_kdump_kernel()
 619  *    The state for all CPUs except the boot CPU needs to be collected
 620  *    with sigp stop-and-store-status. The kexec code or the boot-loader
 621  *    stored the registers of the boot CPU in the memory of the old system.
 622  * 4) kdump and the old kernel stored the CPU state
 623  *    condition: OLDMEM_BASE != NULL && is_kdump_kernel()
 624  *    This case does not exist for s390 anymore, setup_arch explicitly
 625  *    deactivates the elfcorehdr= kernel parameter
 626  */
 627 static __init void smp_save_cpu_vxrs(struct save_area *sa, u16 addr,
 628                                      bool is_boot_cpu, unsigned long page)
 629 {
 630         __vector128 *vxrs = (__vector128 *) page;
 631
 632         if (is_boot_cpu)
 633                 vxrs = boot_cpu_vector_save_area;
 634         else
 635                 __pcpu_sigp_relax(addr, SIGP_STORE_ADDITIONAL_STATUS, page);
 636         save_area_add_vxrs(sa, vxrs);
 637 }
 638
 639 static __init void smp_save_cpu_regs(struct save_area *sa, u16 addr,
 640                                      bool is_boot_cpu, unsigned long page)
 641 {
 642         void *regs = (void *) page;
 643
 644         if (is_boot_cpu)
 645                 copy_oldmem_kernel(regs, (void *) __LC_FPREGS_SAVE_AREA, 512);
 646         else
 647                 __pcpu_sigp_relax(addr, SIGP_STORE_STATUS_AT_ADDRESS, page);
 648         save_area_add_regs(sa, regs);
 649 }
 650
 651 void __init smp_save_dump_cpus(void)
 652 {
 653         int addr, boot_cpu_addr, max_cpu_addr;
 654         struct save_area *sa;
 655         unsigned long page;
 656         bool is_boot_cpu;
 657
 658         if (!(OLDMEM_BASE || is_ipl_type_dump()))
 659                 /* No previous system present, normal boot. */
 660                 return;
 661         /* Allocate a page as dumping area for the store status sigps */
 662         page = memblock_phys_alloc_range(PAGE_SIZE, PAGE_SIZE, 0, 1UL << 31);
 663         if (!page)
 664                 panic("ERROR: Failed to allocate %lx bytes below %lx\n",
 665                       PAGE_SIZE, 1UL << 31);
 666
 667         /* Set multi-threading state to the previous system. */
 668         pcpu_set_smt(sclp.mtid_prev);
 669         boot_cpu_addr = stap();
 670         max_cpu_addr = SCLP_MAX_CORES << sclp.mtid_prev;
 671         for (addr = 0; addr <= max_cpu_addr; addr++) {
 672                 if (__pcpu_sigp_relax(addr, SIGP_SENSE, 0) ==
 673                     SIGP_CC_NOT_OPERATIONAL)
 674                         continue;
 675                 is_boot_cpu = (addr == boot_cpu_addr);
 676                 /* Allocate save area */
 677                 sa = save_area_alloc(is_boot_cpu);
 678                 if (!sa)
 679                         panic("could not allocate memory for save area\n");
 680                 if (MACHINE_HAS_VX)
 681                         /* Get the vector registers */
 682                         smp_save_cpu_vxrs(sa, addr, is_boot_cpu, page);
 683                 /*
 684                  * For a zfcp/nvme dump OLDMEM_BASE == NULL and the registers
 685                  * of the boot CPU are stored in the HSA. To retrieve
 686                  * these registers an SCLP request is required which is
 687                  * done by drivers/s390/char/zcore.c:init_cpu_info()
 688                  */
 689                 if (!is_boot_cpu || OLDMEM_BASE)
 690                         /* Get the CPU registers */
 691                         smp_save_cpu_regs(sa, addr, is_boot_cpu, page);
 692         }
 693         memblock_free(page, PAGE_SIZE);
 694         diag_dma_ops.diag308_reset();
 695         pcpu_set_smt(0);
 696 }
 697 #endif /* CONFIG_CRASH_DUMP */
 698
 699 void smp_cpu_set_polarization(int cpu, int val)
 700 {
 701         pcpu_devices[cpu].polarization = val;
 702 }
 703
 704 int smp_cpu_get_polarization(int cpu)
 705 {
 706         return pcpu_devices[cpu].polarization;
 707 }
 708
 709 int smp_cpu_get_cpu_address(int cpu)
 710 {
 711         return pcpu_devices[cpu].address;
 712 }
 713
 714 static void __ref smp_get_core_info(struct sclp_core_info *info, int early)
 715 {
 716         static int use_sigp_detection;
 717         int address;
 718
 719         if (use_sigp_detection || sclp_get_core_info(info, early)) {
 720                 use_sigp_detection = 1;
 721                 for (address = 0;
 722                      address < (SCLP_MAX_CORES << smp_cpu_mt_shift);
 723                      address += (1U << smp_cpu_mt_shift)) {
 724                         if (__pcpu_sigp_relax(address, SIGP_SENSE, 0) ==
 725                             SIGP_CC_NOT_OPERATIONAL)
 726                                 continue;
 727                         info->core[info->configured].core_id =
 728                                 address >> smp_cpu_mt_shift;
 729                         info->configured++;
 730                 }
 731                 info->combined = info->configured;
 732         }
 733 }
 734
 735 static int smp_add_present_cpu(int cpu);
 736
 737 static int smp_add_core(struct sclp_core_entry *core, cpumask_t *avail,
 738                         bool configured, bool early)
 739 {
 740         struct pcpu *pcpu;
 741         int cpu, nr, i;
 742         u16 address;
 743
 744         nr = 0;
 745         if (sclp.has_core_type && core->type != boot_core_type)
 746                 return nr;
 747         cpu = cpumask_first(avail);
 748         address = core->core_id << smp_cpu_mt_shift;
 749         for (i = 0; (i <= smp_cpu_mtid) && (cpu < nr_cpu_ids); i++) {
 750                 if (pcpu_find_address(cpu_present_mask, address + i))
 751                         continue;
 752                 pcpu = pcpu_devices + cpu;
 753                 pcpu->address = address + i;
 754                 if (configured)
 755                         pcpu->state = CPU_STATE_CONFIGURED;
 756                 else
 757                         pcpu->state = CPU_STATE_STANDBY;
 758                 smp_cpu_set_polarization(cpu, POLARIZATION_UNKNOWN);
 759                 set_cpu_present(cpu, true);
 760                 if (!early && smp_add_present_cpu(cpu) != 0)
 761                         set_cpu_present(cpu, false);
 762                 else
 763                         nr++;
 764                 cpumask_clear_cpu(cpu, avail);
 765                 cpu = cpumask_next(cpu, avail);
 766         }
 767         return nr;
 768 }
 769
 770 static int __smp_rescan_cpus(struct sclp_core_info *info, bool early)
 771 {
 772         struct sclp_core_entry *core;
 773         cpumask_t avail;
 774         bool configured;
 775         u16 core_id;
 776         int nr, i;
 777
 778         nr = 0;
 779         cpumask_xor(&avail, cpu_possible_mask, cpu_present_mask);
 780         /*
 781          * Add IPL core first (which got logical CPU number 0) to make sure
 782          * that all SMT threads get subsequent logical CPU numbers.
 783          */
 784         if (early) {
 785                 core_id = pcpu_devices[0].address >> smp_cpu_mt_shift;
 786                 for (i = 0; i < info->configured; i++) {
 787                         core = &info->core[i];
 788                         if (core->core_id == core_id) {
 789                                 nr += smp_add_core(core, &avail, true, early);
 790                                 break;
 791                         }
 792                 }
 793         }
 794         for (i = 0; i < info->combined; i++) {
 795                 configured = i < info->configured;
 796                 nr += smp_add_core(&info->core[i], &avail, configured, early);
 797         }
 798         return nr;
 799 }
 800
 801 void __init smp_detect_cpus(void)
 802 {
 803         unsigned int cpu, mtid, c_cpus, s_cpus;
 804         struct sclp_core_info *info;
 805         u16 address;
 806
 807         /* Get CPU information */
 808         info = memblock_alloc(sizeof(*info), 8);
 809         if (!info)
 810                 panic("%s: Failed to allocate %zu bytes align=0x%x\n",
 811                       __func__, sizeof(*info), 8);
 812         smp_get_core_info(info, 1);
 813         /* Find boot CPU type */
 814         if (sclp.has_core_type) {
 815                 address = stap();
 816                 for (cpu = 0; cpu < info->combined; cpu++)
 817                         if (info->core[cpu].core_id == address) {
 818                                 /* The boot cpu dictates the cpu type. */
 819                                 boot_core_type = info->core[cpu].type;
 820                                 break;
 821                         }
 822                 if (cpu >= info->combined)
 823                         panic("Could not find boot CPU type");
 824         }
 825
 826         /* Set multi-threading state for the current system */
 827         mtid = boot_core_type ? sclp.mtid : sclp.mtid_cp;
 828         mtid = (mtid < smp_max_threads) ? mtid : smp_max_threads - 1;
 829         pcpu_set_smt(mtid);
 830
 831         /* Print number of CPUs */
 832         c_cpus = s_cpus = 0;
 833         for (cpu = 0; cpu < info->combined; cpu++) {
 834                 if (sclp.has_core_type &&
 835                     info->core[cpu].type != boot_core_type)
 836                         continue;
 837                 if (cpu < info->configured)
 838                         c_cpus += smp_cpu_mtid + 1;
 839                 else
 840                         s_cpus += smp_cpu_mtid + 1;
 841         }
 842         pr_info("%d configured CPUs, %d standby CPUs\n", c_cpus, s_cpus);
 843
 844         /* Add CPUs present at boot */
 845         get_online_cpus();
 846         __smp_rescan_cpus(info, true);
 847         put_online_cpus();
 848         memblock_free_early((unsigned long)info, sizeof(*info));
 849 }
 850
 851 static void smp_init_secondary(void)
 852 {
 853         int cpu = raw_smp_processor_id();
 854
 855         S390_lowcore.last_update_clock = get_tod_clock();
 856         restore_access_regs(S390_lowcore.access_regs_save_area);
 857         cpu_init();
 858         rcu_cpu_starting(cpu);
 859         preempt_disable();
 860         init_cpu_timer();
 861         vtime_init();
 862         vdso_getcpu_init();
 863         pfault_init();
 864         notify_cpu_starting(cpu);
 865         if (topology_cpu_dedicated(cpu))
 866                 set_cpu_flag(CIF_DEDICATED_CPU);
 867         else
 868                 clear_cpu_flag(CIF_DEDICATED_CPU);
 869         set_cpu_online(cpu, true);
 870         update_cpu_masks();
 871         inc_irq_stat(CPU_RST);
 872         local_irq_enable();
 873         cpu_startup_entry(CPUHP_AP_ONLINE_IDLE);
 874 }
 875
 876 /*
 877  *      Activate a secondary processor.
 878  */
 879 static void __no_sanitize_address smp_start_secondary(void *cpuvoid)
 880 {
 881         S390_lowcore.restart_stack = (unsigned long) restart_stack;
 882         S390_lowcore.restart_fn = (unsigned long) do_restart;
 883         S390_lowcore.restart_data = 0;
 884         S390_lowcore.restart_source = -1UL;
 885         __ctl_load(S390_lowcore.cregs_save_area, 0, 15);
 886         __load_psw_mask(PSW_KERNEL_BITS | PSW_MASK_DAT);
 887         CALL_ON_STACK_NORETURN(smp_init_secondary, S390_lowcore.kernel_stack);
 888 }
 889
 890 /* Upping and downing of CPUs */
 891 int __cpu_up(unsigned int cpu, struct task_struct *tidle)
 892 {
 893         struct pcpu *pcpu = pcpu_devices + cpu;
 894         int rc;
 895
 896         if (pcpu->state != CPU_STATE_CONFIGURED)
 897                 return -EIO;
 898         if (pcpu_sigp_retry(pcpu, SIGP_INITIAL_CPU_RESET, 0) !=
 899             SIGP_CC_ORDER_CODE_ACCEPTED)
 900                 return -EIO;
 901
 902         rc = pcpu_alloc_lowcore(pcpu, cpu);
 903         if (rc)
 904                 return rc;
 905         pcpu_prepare_secondary(pcpu, cpu);
 906         pcpu_attach_task(pcpu, tidle);
 907         pcpu_start_fn(pcpu, smp_start_secondary, NULL);
 908         /* Wait until cpu puts itself in the online & active maps */
 909         while (!cpu_online(cpu))
 910                 cpu_relax();
 911         return 0;
 912 }
 913
 914 static unsigned int setup_possible_cpus __initdata;
 915
 916 static int __init _setup_possible_cpus(char *s)
 917 {
 918         get_option(&s, &setup_possible_cpus);
 919         return 0;
 920 }
 921 early_param("possible_cpus", _setup_possible_cpus);
 922
 923 int __cpu_disable(void)
 924 {
 925         unsigned long cregs[16];
 926
 927         /* Handle possible pending IPIs */
 928         smp_handle_ext_call();
 929         set_cpu_online(smp_processor_id(), false);
 930         update_cpu_masks();
 931         /* Disable pseudo page faults on this cpu. */
 932         pfault_fini();
 933         /* Disable interrupt sources via control register. */
 934         __ctl_store(cregs, 0, 15);
 935         cregs[0]  &= ~0x0000ee70UL;     /* disable all external interrupts */
 936         cregs[6]  &= ~0xff000000UL;     /* disable all I/O interrupts */
 937         cregs[14] &= ~0x1f000000UL;     /* disable most machine checks */
 938         __ctl_load(cregs, 0, 15);
 939         clear_cpu_flag(CIF_NOHZ_DELAY);
 940         return 0;
 941 }
 942
 943 void __cpu_die(unsigned int cpu)
 944 {
 945         struct pcpu *pcpu;
 946
 947         /* Wait until target cpu is down */
 948         pcpu = pcpu_devices + cpu;
 949         while (!pcpu_stopped(pcpu))
 950                 cpu_relax();
 951         pcpu_free_lowcore(pcpu);
 952         cpumask_clear_cpu(cpu, mm_cpumask(&init_mm));
 953         cpumask_clear_cpu(cpu, &init_mm.context.cpu_attach_mask);
 954 }
 955
 956 void __noreturn cpu_die(void)
 957 {
 958         idle_task_exit();
 959         __bpon();
 960         pcpu_sigp_retry(pcpu_devices + smp_processor_id(), SIGP_STOP, 0);
 961         for (;;) ;
 962 }
 963
 964 void __init smp_fill_possible_mask(void)
 965 {
 966         unsigned int possible, sclp_max, cpu;
 967
 968         sclp_max = max(sclp.mtid, sclp.mtid_cp) + 1;
 969         sclp_max = min(smp_max_threads, sclp_max);
 970         sclp_max = (sclp.max_cores * sclp_max) ?: nr_cpu_ids;
 971         possible = setup_possible_cpus ?: nr_cpu_ids;
 972         possible = min(possible, sclp_max);
 973         for (cpu = 0; cpu < possible && cpu < nr_cpu_ids; cpu++)
 974                 set_cpu_possible(cpu, true);
 975 }
 976
 977 void __init smp_prepare_cpus(unsigned int max_cpus)
 978 {
 979         /* request the 0x1201 emergency signal external interrupt */
 980         if (register_external_irq(EXT_IRQ_EMERGENCY_SIG, do_ext_call_interrupt))
 981                 panic("Couldn't request external interrupt 0x1201");
 982         /* request the 0x1202 external call external interrupt */
 983         if (register_external_irq(EXT_IRQ_EXTERNAL_CALL, do_ext_call_interrupt))
 984                 panic("Couldn't request external interrupt 0x1202");
 985 }
 986
 987 void __init smp_prepare_boot_cpu(void)
 988 {
 989         struct pcpu *pcpu = pcpu_devices;
 990
 991         WARN_ON(!cpu_present(0) || !cpu_online(0));
 992         pcpu->state = CPU_STATE_CONFIGURED;
 993         pcpu->lowcore = (struct lowcore *)(unsigned long) store_prefix();
 994         S390_lowcore.percpu_offset = __per_cpu_offset[0];
 995         smp_cpu_set_polarization(0, POLARIZATION_UNKNOWN);
 996 }
 997
 998 void __init smp_setup_processor_id(void)
 999 {
1000         pcpu_devices[0].address = stap();
1001         S390_lowcore.cpu_nr = 0;
1002         S390_lowcore.spinlock_lockval = arch_spin_lockval(0);
1003         S390_lowcore.spinlock_index = 0;
1004 }
1005
1006 /*
1007  * the frequency of the profiling timer can be changed
1008  * by writing a multiplier value into /proc/profile.
1009  *
1010  * usually you want to run this on all CPUs ;)
1011  */
1012 int setup_profiling_timer(unsigned int multiplier)
1013 {
1014         return 0;
1015 }
1016
1017 static ssize_t cpu_configure_show(struct device *dev,
1018                                   struct device_attribute *attr, char *buf)
1019 {
1020         ssize_t count;
1021
1022         mutex_lock(&smp_cpu_state_mutex);
1023         count = sprintf(buf, "%d\n", pcpu_devices[dev->id].state);
1024         mutex_unlock(&smp_cpu_state_mutex);
1025         return count;
1026 }
1027
1028 static ssize_t cpu_configure_store(struct device *dev,
1029                                    struct device_attribute *attr,
1030                                    const char *buf, size_t count)
1031 {
1032         struct pcpu *pcpu;
1033         int cpu, val, rc, i;
1034         char delim;
1035
1036         if (sscanf(buf, "%d %c", &val, &delim) != 1)
1037                 return -EINVAL;
1038         if (val != 0 && val != 1)
1039                 return -EINVAL;
1040         get_online_cpus();
1041         mutex_lock(&smp_cpu_state_mutex);
1042         rc = -EBUSY;
1043         /* disallow configuration changes of online cpus and cpu 0 */
1044         cpu = dev->id;
1045         cpu = smp_get_base_cpu(cpu);
1046         if (cpu == 0)
1047                 goto out;
1048         for (i = 0; i <= smp_cpu_mtid; i++)
1049                 if (cpu_online(cpu + i))
1050                         goto out;
1051         pcpu = pcpu_devices + cpu;
1052         rc = 0;
1053         switch (val) {
1054         case 0:
1055                 if (pcpu->state != CPU_STATE_CONFIGURED)
1056                         break;
1057                 rc = sclp_core_deconfigure(pcpu->address >> smp_cpu_mt_shift);
1058                 if (rc)
1059                         break;
1060                 for (i = 0; i <= smp_cpu_mtid; i++) {
1061                         if (cpu + i >= nr_cpu_ids || !cpu_present(cpu + i))
1062                                 continue;
1063                         pcpu[i].state = CPU_STATE_STANDBY;
1064                         smp_cpu_set_polarization(cpu + i,
1065                                                  POLARIZATION_UNKNOWN);
1066                 }
1067                 topology_expect_change();
1068                 break;
1069         case 1:
1070                 if (pcpu->state != CPU_STATE_STANDBY)
1071                         break;
1072                 rc = sclp_core_configure(pcpu->address >> smp_cpu_mt_shift);
1073                 if (rc)
1074                         break;
1075                 for (i = 0; i <= smp_cpu_mtid; i++) {
1076                         if (cpu + i >= nr_cpu_ids || !cpu_present(cpu + i))
1077                                 continue;
1078                         pcpu[i].state = CPU_STATE_CONFIGURED;
1079                         smp_cpu_set_polarization(cpu + i,
1080                                                  POLARIZATION_UNKNOWN);
1081                 }
1082                 topology_expect_change();
1083                 break;
1084         default:
1085                 break;
1086         }
1087 out:
1088         mutex_unlock(&smp_cpu_state_mutex);
1089         put_online_cpus();
1090         return rc ? rc : count;
1091 }
1092 static DEVICE_ATTR(configure, 0644, cpu_configure_show, cpu_configure_store);
1093
1094 static ssize_t show_cpu_address(struct device *dev,
1095                                 struct device_attribute *attr, char *buf)
1096 {
1097         return sprintf(buf, "%d\n", pcpu_devices[dev->id].address);
1098 }
1099 static DEVICE_ATTR(address, 0444, show_cpu_address, NULL);
1100
1101 static struct attribute *cpu_common_attrs[] = {
1102         &dev_attr_configure.attr,
1103         &dev_attr_address.attr,
1104         NULL,
1105 };
1106
1107 static struct attribute_group cpu_common_attr_group = {
1108         .attrs = cpu_common_attrs,
1109 };
1110
1111 static struct attribute *cpu_online_attrs[] = {
1112         &dev_attr_idle_count.attr,
1113         &dev_attr_idle_time_us.attr,
1114         NULL,
1115 };
1116
1117 static struct attribute_group cpu_online_attr_group = {
1118         .attrs = cpu_online_attrs,
1119 };
1120
1121 static int smp_cpu_online(unsigned int cpu)
1122 {
1123         struct device *s = &per_cpu(cpu_device, cpu)->dev;
1124
1125         return sysfs_create_group(&s->kobj, &cpu_online_attr_group);
1126 }
1127
1128 static int smp_cpu_pre_down(unsigned int cpu)
1129 {
1130         struct device *s = &per_cpu(cpu_device, cpu)->dev;
1131
1132         sysfs_remove_group(&s->kobj, &cpu_online_attr_group);
1133         return 0;
1134 }
1135
1136 static int smp_add_present_cpu(int cpu)
1137 {
1138         struct device *s;
1139         struct cpu *c;
1140         int rc;
1141
1142         c = kzalloc(sizeof(*c), GFP_KERNEL);
1143         if (!c)
1144                 return -ENOMEM;
1145         per_cpu(cpu_device, cpu) = c;
1146         s = &c->dev;
1147         c->hotpluggable = 1;
1148         rc = register_cpu(c, cpu);
1149         if (rc)
1150                 goto out;
1151         rc = sysfs_create_group(&s->kobj, &cpu_common_attr_group);
1152         if (rc)
1153                 goto out_cpu;
1154         rc = topology_cpu_init(c);
1155         if (rc)
1156                 goto out_topology;
1157         return 0;
1158
1159 out_topology:
1160         sysfs_remove_group(&s->kobj, &cpu_common_attr_group);
1161 out_cpu:
1162         unregister_cpu(c);
1163 out:
1164         return rc;
1165 }
1166
1167 int __ref smp_rescan_cpus(void)
1168 {
1169         struct sclp_core_info *info;
1170         int nr;
1171
1172         info = kzalloc(sizeof(*info), GFP_KERNEL);
1173         if (!info)
1174                 return -ENOMEM;
1175         smp_get_core_info(info, 0);
1176         get_online_cpus();
1177         mutex_lock(&smp_cpu_state_mutex);
1178         nr = __smp_rescan_cpus(info, false);
1179         mutex_unlock(&smp_cpu_state_mutex);
1180         put_online_cpus();
1181         kfree(info);
1182         if (nr)
1183                 topology_schedule_update();
1184         return 0;
1185 }
1186
1187 static ssize_t __ref rescan_store(struct device *dev,
1188                                   struct device_attribute *attr,
1189                                   const char *buf,
1190                                   size_t count)
1191 {
1192         int rc;
1193
1194         rc = lock_device_hotplug_sysfs();
1195         if (rc)
1196                 return rc;
1197         rc = smp_rescan_cpus();
1198         unlock_device_hotplug();
1199         return rc ? rc : count;
1200 }
1201 static DEVICE_ATTR_WO(rescan);
1202
1203 static int __init s390_smp_init(void)
1204 {
1205         int cpu, rc = 0;
1206
1207         rc = device_create_file(cpu_subsys.dev_root, &dev_attr_rescan);
1208         if (rc)
1209                 return rc;
1210         for_each_present_cpu(cpu) {
1211                 rc = smp_add_present_cpu(cpu);
1212                 if (rc)
1213                         goto out;
1214         }
1215
1216         rc = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "s390/smp:online",
1217                                smp_cpu_online, smp_cpu_pre_down);
1218         rc = rc <= 0 ? rc : 0;
1219 out:
1220         return rc;
1221 }
1222 subsys_initcall(s390_smp_init);