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/vdso.h>
  51 #include <asm/debug.h>
  52 #include <asm/os_info.h>
  53 #include <asm/sigp.h>
  54 #include <asm/idle.h>
  55 #include <asm/nmi.h>
  56 #include <asm/topology.h>
  57 #include "entry.h"
  58
  59 enum {
  60         ec_schedule = 0,
  61         ec_call_function_single,
  62         ec_stop_cpu,
  63 };
  64
  65 enum {
  66         CPU_STATE_STANDBY,
  67         CPU_STATE_CONFIGURED,
  68 };
  69
  70 static DEFINE_PER_CPU(struct cpu *, cpu_device);
  71
  72 struct pcpu {
  73         struct lowcore *lowcore;        /* lowcore page(s) for the cpu */
  74         unsigned long ec_mask;          /* bit mask for ec_xxx functions */
  75         unsigned long ec_clk;           /* sigp timestamp for ec_xxx */
  76         signed char state;              /* physical cpu state */
  77         signed char polarization;       /* physical polarization */
  78         u16 address;                    /* physical cpu address */
  79 };
  80
  81 static u8 boot_core_type;
  82 static struct pcpu pcpu_devices[NR_CPUS];
  83
  84 unsigned int smp_cpu_mt_shift;
  85 EXPORT_SYMBOL(smp_cpu_mt_shift);
  86
  87 unsigned int smp_cpu_mtid;
  88 EXPORT_SYMBOL(smp_cpu_mtid);
  89
  90 #ifdef CONFIG_CRASH_DUMP
  91 __vector128 __initdata boot_cpu_vector_save_area[__NUM_VXRS];
  92 #endif
  93
  94 static unsigned int smp_max_threads __initdata = -1U;
  95
  96 static int __init early_nosmt(char *s)
  97 {
  98         smp_max_threads = 1;
  99         return 0;
 100 }
 101 early_param("nosmt", early_nosmt);
 102
 103 static int __init early_smt(char *s)
 104 {
 105         get_option(&s, &smp_max_threads);
 106         return 0;
 107 }
 108 early_param("smt", early_smt);
 109
 110 /*
 111  * The smp_cpu_state_mutex must be held when changing the state or polarization
 112  * member of a pcpu data structure within the pcpu_devices arreay.
 113  */
 114 DEFINE_MUTEX(smp_cpu_state_mutex);
 115
 116 /*
 117  * Signal processor helper functions.
 118  */
 119 static inline int __pcpu_sigp_relax(u16 addr, u8 order, unsigned long parm)
 120 {
 121         int cc;
 122
 123         while (1) {
 124                 cc = __pcpu_sigp(addr, order, parm, NULL);
 125                 if (cc != SIGP_CC_BUSY)
 126                         return cc;
 127                 cpu_relax();
 128         }
 129 }
 130
 131 static int pcpu_sigp_retry(struct pcpu *pcpu, u8 order, u32 parm)
 132 {
 133         int cc, retry;
 134
 135         for (retry = 0; ; retry++) {
 136                 cc = __pcpu_sigp(pcpu->address, order, parm, NULL);
 137                 if (cc != SIGP_CC_BUSY)
 138                         break;
 139                 if (retry >= 3)
 140                         udelay(10);
 141         }
 142         return cc;
 143 }
 144
 145 static inline int pcpu_stopped(struct pcpu *pcpu)
 146 {
 147         u32 uninitialized_var(status);
 148
 149         if (__pcpu_sigp(pcpu->address, SIGP_SENSE,
 150                         0, &status) != SIGP_CC_STATUS_STORED)
 151                 return 0;
 152         return !!(status & (SIGP_STATUS_CHECK_STOP|SIGP_STATUS_STOPPED));
 153 }
 154
 155 static inline int pcpu_running(struct pcpu *pcpu)
 156 {
 157         if (__pcpu_sigp(pcpu->address, SIGP_SENSE_RUNNING,
 158                         0, NULL) != SIGP_CC_STATUS_STORED)
 159                 return 1;
 160         /* Status stored condition code is equivalent to cpu not running. */
 161         return 0;
 162 }
 163
 164 /*
 165  * Find struct pcpu by cpu address.
 166  */
 167 static struct pcpu *pcpu_find_address(const struct cpumask *mask, u16 address)
 168 {
 169         int cpu;
 170
 171         for_each_cpu(cpu, mask)
 172                 if (pcpu_devices[cpu].address == address)
 173                         return pcpu_devices + cpu;
 174         return NULL;
 175 }
 176
 177 static void pcpu_ec_call(struct pcpu *pcpu, int ec_bit)
 178 {
 179         int order;
 180
 181         if (test_and_set_bit(ec_bit, &pcpu->ec_mask))
 182                 return;
 183         order = pcpu_running(pcpu) ? SIGP_EXTERNAL_CALL : SIGP_EMERGENCY_SIGNAL;
 184         pcpu->ec_clk = get_tod_clock_fast();
 185         pcpu_sigp_retry(pcpu, order, 0);
 186 }
 187
 188 static int pcpu_alloc_lowcore(struct pcpu *pcpu, int cpu)
 189 {
 190         unsigned long async_stack, nodat_stack;
 191         struct lowcore *lc;
 192
 193         if (pcpu != &pcpu_devices[0]) {
 194                 pcpu->lowcore = (struct lowcore *)
 195                         __get_free_pages(GFP_KERNEL | GFP_DMA, LC_ORDER);
 196                 nodat_stack = __get_free_pages(GFP_KERNEL, THREAD_SIZE_ORDER);
 197                 if (!pcpu->lowcore || !nodat_stack)
 198                         goto out;
 199         } else {
 200                 nodat_stack = pcpu->lowcore->nodat_stack - STACK_INIT_OFFSET;
 201         }
 202         async_stack = stack_alloc();
 203         if (!async_stack)
 204                 goto out;
 205         lc = pcpu->lowcore;
 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->cpu_nr = cpu;
 211         lc->spinlock_lockval = arch_spin_lockval(cpu);
 212         lc->spinlock_index = 0;
 213         lc->br_r1_trampoline = 0x07f1;  /* br %r1 */
 214         if (nmi_alloc_per_cpu(lc))
 215                 goto out_async;
 216         if (vdso_alloc_per_cpu(lc))
 217                 goto out_mcesa;
 218         lowcore_ptr[cpu] = lc;
 219         pcpu_sigp_retry(pcpu, SIGP_SET_PREFIX, (u32)(unsigned long) lc);
 220         return 0;
 221
 222 out_mcesa:
 223         nmi_free_per_cpu(lc);
 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 #ifdef CONFIG_HOTPLUG_CPU
 235
 236 static void pcpu_free_lowcore(struct pcpu *pcpu)
 237 {
 238         unsigned long async_stack, nodat_stack, lowcore;
 239
 240         nodat_stack = pcpu->lowcore->nodat_stack - STACK_INIT_OFFSET;
 241         async_stack = pcpu->lowcore->async_stack - STACK_INIT_OFFSET;
 242         lowcore = (unsigned long) pcpu->lowcore;
 243
 244         pcpu_sigp_retry(pcpu, SIGP_SET_PREFIX, 0);
 245         lowcore_ptr[pcpu - pcpu_devices] = NULL;
 246         vdso_free_per_cpu(pcpu->lowcore);
 247         nmi_free_per_cpu(pcpu->lowcore);
 248         stack_free(async_stack);
 249         if (pcpu == &pcpu_devices[0])
 250                 return;
 251         free_pages(nodat_stack, THREAD_SIZE_ORDER);
 252         free_pages(lowcore, LC_ORDER);
 253 }
 254
 255 #endif /* CONFIG_HOTPLUG_CPU */
 256
 257 static void pcpu_prepare_secondary(struct pcpu *pcpu, int cpu)
 258 {
 259         struct lowcore *lc = pcpu->lowcore;
 260
 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->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->machine_flags = S390_lowcore.machine_flags;
 269         lc->user_timer = lc->system_timer =
 270                 lc->steal_timer = lc->avg_steal_timer = 0;
 271         __ctl_store(lc->cregs_save_area, 0, 15);
 272         save_access_regs((unsigned int *) lc->access_regs_save_area);
 273         memcpy(lc->stfle_fac_list, S390_lowcore.stfle_fac_list,
 274                sizeof(lc->stfle_fac_list));
 275         memcpy(lc->alt_stfle_fac_list, S390_lowcore.alt_stfle_fac_list,
 276                sizeof(lc->alt_stfle_fac_list));
 277         arch_spin_lock_setup(cpu);
 278 }
 279
 280 static void pcpu_attach_task(struct pcpu *pcpu, struct task_struct *tsk)
 281 {
 282         struct lowcore *lc = pcpu->lowcore;
 283
 284         lc->kernel_stack = (unsigned long) task_stack_page(tsk)
 285                 + THREAD_SIZE - STACK_FRAME_OVERHEAD - sizeof(struct pt_regs);
 286         lc->current_task = (unsigned long) tsk;
 287         lc->lpp = LPP_MAGIC;
 288         lc->current_pid = tsk->pid;
 289         lc->user_timer = tsk->thread.user_timer;
 290         lc->guest_timer = tsk->thread.guest_timer;
 291         lc->system_timer = tsk->thread.system_timer;
 292         lc->hardirq_timer = tsk->thread.hardirq_timer;
 293         lc->softirq_timer = tsk->thread.softirq_timer;
 294         lc->steal_timer = 0;
 295 }
 296
 297 static void pcpu_start_fn(struct pcpu *pcpu, void (*func)(void *), void *data)
 298 {
 299         struct lowcore *lc = pcpu->lowcore;
 300
 301         lc->restart_stack = lc->nodat_stack;
 302         lc->restart_fn = (unsigned long) func;
 303         lc->restart_data = (unsigned long) data;
 304         lc->restart_source = -1UL;
 305         pcpu_sigp_retry(pcpu, SIGP_RESTART, 0);
 306 }
 307
 308 /*
 309  * Call function via PSW restart on pcpu and stop the current cpu.
 310  */
 311 static void __pcpu_delegate(void (*func)(void*), void *data)
 312 {
 313         func(data);     /* should not return */
 314 }
 315
 316 static void __no_sanitize_address pcpu_delegate(struct pcpu *pcpu,
 317                                                 void (*func)(void *),
 318                                                 void *data, unsigned long stack)
 319 {
 320         struct lowcore *lc = lowcore_ptr[pcpu - pcpu_devices];
 321         unsigned long source_cpu = stap();
 322
 323         __load_psw_mask(PSW_KERNEL_BITS | PSW_MASK_DAT);
 324         if (pcpu->address == source_cpu)
 325                 CALL_ON_STACK(__pcpu_delegate, stack, 2, func, data);
 326         /* Stop target cpu (if func returns this stops the current cpu). */
 327         pcpu_sigp_retry(pcpu, SIGP_STOP, 0);
 328         /* Restart func on the target cpu and stop the current cpu. */
 329         mem_assign_absolute(lc->restart_stack, stack);
 330         mem_assign_absolute(lc->restart_fn, (unsigned long) func);
 331         mem_assign_absolute(lc->restart_data, (unsigned long) data);
 332         mem_assign_absolute(lc->restart_source, source_cpu);
 333         __bpon();
 334         asm volatile(
 335                 "0:     sigp    0,%0,%2 # sigp restart to target cpu\n"
 336                 "       brc     2,0b    # busy, try again\n"
 337                 "1:     sigp    0,%1,%3 # sigp stop to current cpu\n"
 338                 "       brc     2,1b    # busy, try again\n"
 339                 : : "d" (pcpu->address), "d" (source_cpu),
 340                     "K" (SIGP_RESTART), "K" (SIGP_STOP)
 341                 : "0", "1", "cc");
 342         for (;;) ;
 343 }
 344
 345 /*
 346  * Enable additional logical cpus for multi-threading.
 347  */
 348 static int pcpu_set_smt(unsigned int mtid)
 349 {
 350         int cc;
 351
 352         if (smp_cpu_mtid == mtid)
 353                 return 0;
 354         cc = __pcpu_sigp(0, SIGP_SET_MULTI_THREADING, mtid, NULL);
 355         if (cc == 0) {
 356                 smp_cpu_mtid = mtid;
 357                 smp_cpu_mt_shift = 0;
 358                 while (smp_cpu_mtid >= (1U << smp_cpu_mt_shift))
 359                         smp_cpu_mt_shift++;
 360                 pcpu_devices[0].address = stap();
 361         }
 362         return cc;
 363 }
 364
 365 /*
 366  * Call function on an online CPU.
 367  */
 368 void smp_call_online_cpu(void (*func)(void *), void *data)
 369 {
 370         struct pcpu *pcpu;
 371
 372         /* Use the current cpu if it is online. */
 373         pcpu = pcpu_find_address(cpu_online_mask, stap());
 374         if (!pcpu)
 375                 /* Use the first online cpu. */
 376                 pcpu = pcpu_devices + cpumask_first(cpu_online_mask);
 377         pcpu_delegate(pcpu, func, data, (unsigned long) restart_stack);
 378 }
 379
 380 /*
 381  * Call function on the ipl CPU.
 382  */
 383 void smp_call_ipl_cpu(void (*func)(void *), void *data)
 384 {
 385         struct lowcore *lc = pcpu_devices->lowcore;
 386
 387         if (pcpu_devices[0].address == stap())
 388                 lc = &S390_lowcore;
 389
 390         pcpu_delegate(&pcpu_devices[0], func, data,
 391                       lc->nodat_stack);
 392 }
 393
 394 int smp_find_processor_id(u16 address)
 395 {
 396         int cpu;
 397
 398         for_each_present_cpu(cpu)
 399                 if (pcpu_devices[cpu].address == address)
 400                         return cpu;
 401         return -1;
 402 }
 403
 404 bool arch_vcpu_is_preempted(int cpu)
 405 {
 406         if (test_cpu_flag_of(CIF_ENABLED_WAIT, cpu))
 407                 return false;
 408         if (pcpu_running(pcpu_devices + cpu))
 409                 return false;
 410         return true;
 411 }
 412 EXPORT_SYMBOL(arch_vcpu_is_preempted);
 413
 414 void smp_yield_cpu(int cpu)
 415 {
 416         if (MACHINE_HAS_DIAG9C) {
 417                 diag_stat_inc_norecursion(DIAG_STAT_X09C);
 418                 asm volatile("diag %0,0,0x9c"
 419                              : : "d" (pcpu_devices[cpu].address));
 420         } else if (MACHINE_HAS_DIAG44) {
 421                 diag_stat_inc_norecursion(DIAG_STAT_X044);
 422                 asm volatile("diag 0,0,0x44");
 423         }
 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 }
 502
 503 static void do_ext_call_interrupt(struct ext_code ext_code,
 504                                   unsigned int param32, unsigned long param64)
 505 {
 506         inc_irq_stat(ext_code.code == 0x1202 ? IRQEXT_EXC : IRQEXT_EMS);
 507         smp_handle_ext_call();
 508 }
 509
 510 void arch_send_call_function_ipi_mask(const struct cpumask *mask)
 511 {
 512         int cpu;
 513
 514         for_each_cpu(cpu, mask)
 515                 pcpu_ec_call(pcpu_devices + cpu, ec_call_function_single);
 516 }
 517
 518 void arch_send_call_function_single_ipi(int cpu)
 519 {
 520         pcpu_ec_call(pcpu_devices + cpu, ec_call_function_single);
 521 }
 522
 523 /*
 524  * this function sends a 'reschedule' IPI to another CPU.
 525  * it goes straight through and wastes no time serializing
 526  * anything. Worst case is that we lose a reschedule ...
 527  */
 528 void smp_send_reschedule(int cpu)
 529 {
 530         pcpu_ec_call(pcpu_devices + cpu, ec_schedule);
 531 }
 532
 533 /*
 534  * parameter area for the set/clear control bit callbacks
 535  */
 536 struct ec_creg_mask_parms {
 537         unsigned long orval;
 538         unsigned long andval;
 539         int cr;
 540 };
 541
 542 /*
 543  * callback for setting/clearing control bits
 544  */
 545 static void smp_ctl_bit_callback(void *info)
 546 {
 547         struct ec_creg_mask_parms *pp = info;
 548         unsigned long cregs[16];
 549
 550         __ctl_store(cregs, 0, 15);
 551         cregs[pp->cr] = (cregs[pp->cr] & pp->andval) | pp->orval;
 552         __ctl_load(cregs, 0, 15);
 553 }
 554
 555 /*
 556  * Set a bit in a control register of all cpus
 557  */
 558 void smp_ctl_set_bit(int cr, int bit)
 559 {
 560         struct ec_creg_mask_parms parms = { 1UL << bit, -1UL, cr };
 561
 562         on_each_cpu(smp_ctl_bit_callback, &parms, 1);
 563 }
 564 EXPORT_SYMBOL(smp_ctl_set_bit);
 565
 566 /*
 567  * Clear a bit in a control register of all cpus
 568  */
 569 void smp_ctl_clear_bit(int cr, int bit)
 570 {
 571         struct ec_creg_mask_parms parms = { 0, ~(1UL << bit), cr };
 572
 573         on_each_cpu(smp_ctl_bit_callback, &parms, 1);
 574 }
 575 EXPORT_SYMBOL(smp_ctl_clear_bit);
 576
 577 #ifdef CONFIG_CRASH_DUMP
 578
 579 int smp_store_status(int cpu)
 580 {
 581         struct pcpu *pcpu = pcpu_devices + cpu;
 582         unsigned long pa;
 583
 584         pa = __pa(&pcpu->lowcore->floating_pt_save_area);
 585         if (__pcpu_sigp_relax(pcpu->address, SIGP_STORE_STATUS_AT_ADDRESS,
 586                               pa) != SIGP_CC_ORDER_CODE_ACCEPTED)
 587                 return -EIO;
 588         if (!MACHINE_HAS_VX && !MACHINE_HAS_GS)
 589                 return 0;
 590         pa = __pa(pcpu->lowcore->mcesad & MCESA_ORIGIN_MASK);
 591         if (MACHINE_HAS_GS)
 592                 pa |= pcpu->lowcore->mcesad & MCESA_LC_MASK;
 593         if (__pcpu_sigp_relax(pcpu->address, SIGP_STORE_ADDITIONAL_STATUS,
 594                               pa) != SIGP_CC_ORDER_CODE_ACCEPTED)
 595                 return -EIO;
 596         return 0;
 597 }
 598
 599 /*
 600  * Collect CPU state of the previous, crashed system.
 601  * There are four cases:
 602  * 1) standard zfcp dump
 603  *    condition: OLDMEM_BASE == NULL && ipl_info.type == IPL_TYPE_FCP_DUMP
 604  *    The state for all CPUs except the boot CPU needs to be collected
 605  *    with sigp stop-and-store-status. The boot CPU state is located in
 606  *    the absolute lowcore of the memory stored in the HSA. The zcore code
 607  *    will copy the boot CPU state from the HSA.
 608  * 2) stand-alone kdump for SCSI (zfcp dump with swapped memory)
 609  *    condition: OLDMEM_BASE != NULL && ipl_info.type == IPL_TYPE_FCP_DUMP
 610  *    The state for all CPUs except the boot CPU needs to be collected
 611  *    with sigp stop-and-store-status. The firmware or the boot-loader
 612  *    stored the registers of the boot CPU in the absolute lowcore in the
 613  *    memory of the old system.
 614  * 3) kdump and the old kernel did not store the CPU state,
 615  *    or stand-alone kdump for DASD
 616  *    condition: OLDMEM_BASE != NULL && !is_kdump_kernel()
 617  *    The state for all CPUs except the boot CPU needs to be collected
 618  *    with sigp stop-and-store-status. The kexec code or the boot-loader
 619  *    stored the registers of the boot CPU in the memory of the old system.
 620  * 4) kdump and the old kernel stored the CPU state
 621  *    condition: OLDMEM_BASE != NULL && is_kdump_kernel()
 622  *    This case does not exist for s390 anymore, setup_arch explicitly
 623  *    deactivates the elfcorehdr= kernel parameter
 624  */
 625 static __init void smp_save_cpu_vxrs(struct save_area *sa, u16 addr,
 626                                      bool is_boot_cpu, unsigned long page)
 627 {
 628         __vector128 *vxrs = (__vector128 *) page;
 629
 630         if (is_boot_cpu)
 631                 vxrs = boot_cpu_vector_save_area;
 632         else
 633                 __pcpu_sigp_relax(addr, SIGP_STORE_ADDITIONAL_STATUS, page);
 634         save_area_add_vxrs(sa, vxrs);
 635 }
 636
 637 static __init void smp_save_cpu_regs(struct save_area *sa, u16 addr,
 638                                      bool is_boot_cpu, unsigned long page)
 639 {
 640         void *regs = (void *) page;
 641
 642         if (is_boot_cpu)
 643                 copy_oldmem_kernel(regs, (void *) __LC_FPREGS_SAVE_AREA, 512);
 644         else
 645                 __pcpu_sigp_relax(addr, SIGP_STORE_STATUS_AT_ADDRESS, page);
 646         save_area_add_regs(sa, regs);
 647 }
 648
 649 void __init smp_save_dump_cpus(void)
 650 {
 651         int addr, boot_cpu_addr, max_cpu_addr;
 652         struct save_area *sa;
 653         unsigned long page;
 654         bool is_boot_cpu;
 655
 656         if (!(OLDMEM_BASE || ipl_info.type == IPL_TYPE_FCP_DUMP))
 657                 /* No previous system present, normal boot. */
 658                 return;
 659         /* Allocate a page as dumping area for the store status sigps */
 660         page = memblock_phys_alloc_range(PAGE_SIZE, PAGE_SIZE, 0, 1UL << 31);
 661         if (!page)
 662                 panic("ERROR: Failed to allocate %lx bytes below %lx\n",
 663                       PAGE_SIZE, 1UL << 31);
 664
 665         /* Set multi-threading state to the previous system. */
 666         pcpu_set_smt(sclp.mtid_prev);
 667         boot_cpu_addr = stap();
 668         max_cpu_addr = SCLP_MAX_CORES << sclp.mtid_prev;
 669         for (addr = 0; addr <= max_cpu_addr; addr++) {
 670                 if (__pcpu_sigp_relax(addr, SIGP_SENSE, 0) ==
 671                     SIGP_CC_NOT_OPERATIONAL)
 672                         continue;
 673                 is_boot_cpu = (addr == boot_cpu_addr);
 674                 /* Allocate save area */
 675                 sa = save_area_alloc(is_boot_cpu);
 676                 if (!sa)
 677                         panic("could not allocate memory for save area\n");
 678                 if (MACHINE_HAS_VX)
 679                         /* Get the vector registers */
 680                         smp_save_cpu_vxrs(sa, addr, is_boot_cpu, page);
 681                 /*
 682                  * For a zfcp dump OLDMEM_BASE == NULL and the registers
 683                  * of the boot CPU are stored in the HSA. To retrieve
 684                  * these registers an SCLP request is required which is
 685                  * done by drivers/s390/char/zcore.c:init_cpu_info()
 686                  */
 687                 if (!is_boot_cpu || OLDMEM_BASE)
 688                         /* Get the CPU registers */
 689                         smp_save_cpu_regs(sa, addr, is_boot_cpu, page);
 690         }
 691         memblock_free(page, PAGE_SIZE);
 692         diag308_reset();
 693         pcpu_set_smt(0);
 694 }
 695 #endif /* CONFIG_CRASH_DUMP */
 696
 697 void smp_cpu_set_polarization(int cpu, int val)
 698 {
 699         pcpu_devices[cpu].polarization = val;
 700 }
 701
 702 int smp_cpu_get_polarization(int cpu)
 703 {
 704         return pcpu_devices[cpu].polarization;
 705 }
 706
 707 static void __ref smp_get_core_info(struct sclp_core_info *info, int early)
 708 {
 709         static int use_sigp_detection;
 710         int address;
 711
 712         if (use_sigp_detection || sclp_get_core_info(info, early)) {
 713                 use_sigp_detection = 1;
 714                 for (address = 0;
 715                      address < (SCLP_MAX_CORES << smp_cpu_mt_shift);
 716                      address += (1U << smp_cpu_mt_shift)) {
 717                         if (__pcpu_sigp_relax(address, SIGP_SENSE, 0) ==
 718                             SIGP_CC_NOT_OPERATIONAL)
 719                                 continue;
 720                         info->core[info->configured].core_id =
 721                                 address >> smp_cpu_mt_shift;
 722                         info->configured++;
 723                 }
 724                 info->combined = info->configured;
 725         }
 726 }
 727
 728 static int smp_add_present_cpu(int cpu);
 729
 730 static int __smp_rescan_cpus(struct sclp_core_info *info, int sysfs_add)
 731 {
 732         struct pcpu *pcpu;
 733         cpumask_t avail;
 734         int cpu, nr, i, j;
 735         u16 address;
 736
 737         nr = 0;
 738         cpumask_xor(&avail, cpu_possible_mask, cpu_present_mask);
 739         cpu = cpumask_first(&avail);
 740         for (i = 0; (i < info->combined) && (cpu < nr_cpu_ids); i++) {
 741                 if (sclp.has_core_type && info->core[i].type != boot_core_type)
 742                         continue;
 743                 address = info->core[i].core_id << smp_cpu_mt_shift;
 744                 for (j = 0; j <= smp_cpu_mtid; j++) {
 745                         if (pcpu_find_address(cpu_present_mask, address + j))
 746                                 continue;
 747                         pcpu = pcpu_devices + cpu;
 748                         pcpu->address = address + j;
 749                         pcpu->state =
 750                                 (cpu >= info->configured*(smp_cpu_mtid + 1)) ?
 751                                 CPU_STATE_STANDBY : CPU_STATE_CONFIGURED;
 752                         smp_cpu_set_polarization(cpu, POLARIZATION_UNKNOWN);
 753                         set_cpu_present(cpu, true);
 754                         if (sysfs_add && smp_add_present_cpu(cpu) != 0)
 755                                 set_cpu_present(cpu, false);
 756                         else
 757                                 nr++;
 758                         cpu = cpumask_next(cpu, &avail);
 759                         if (cpu >= nr_cpu_ids)
 760                                 break;
 761                 }
 762         }
 763         return nr;
 764 }
 765
 766 void __init smp_detect_cpus(void)
 767 {
 768         unsigned int cpu, mtid, c_cpus, s_cpus;
 769         struct sclp_core_info *info;
 770         u16 address;
 771
 772         /* Get CPU information */
 773         info = memblock_alloc(sizeof(*info), 8);
 774         if (!info)
 775                 panic("%s: Failed to allocate %zu bytes align=0x%x\n",
 776                       __func__, sizeof(*info), 8);
 777         smp_get_core_info(info, 1);
 778         /* Find boot CPU type */
 779         if (sclp.has_core_type) {
 780                 address = stap();
 781                 for (cpu = 0; cpu < info->combined; cpu++)
 782                         if (info->core[cpu].core_id == address) {
 783                                 /* The boot cpu dictates the cpu type. */
 784                                 boot_core_type = info->core[cpu].type;
 785                                 break;
 786                         }
 787                 if (cpu >= info->combined)
 788                         panic("Could not find boot CPU type");
 789         }
 790
 791         /* Set multi-threading state for the current system */
 792         mtid = boot_core_type ? sclp.mtid : sclp.mtid_cp;
 793         mtid = (mtid < smp_max_threads) ? mtid : smp_max_threads - 1;
 794         pcpu_set_smt(mtid);
 795
 796         /* Print number of CPUs */
 797         c_cpus = s_cpus = 0;
 798         for (cpu = 0; cpu < info->combined; cpu++) {
 799                 if (sclp.has_core_type &&
 800                     info->core[cpu].type != boot_core_type)
 801                         continue;
 802                 if (cpu < info->configured)
 803                         c_cpus += smp_cpu_mtid + 1;
 804                 else
 805                         s_cpus += smp_cpu_mtid + 1;
 806         }
 807         pr_info("%d configured CPUs, %d standby CPUs\n", c_cpus, s_cpus);
 808
 809         /* Add CPUs present at boot */
 810         get_online_cpus();
 811         __smp_rescan_cpus(info, 0);
 812         put_online_cpus();
 813         memblock_free_early((unsigned long)info, sizeof(*info));
 814 }
 815
 816 static void smp_init_secondary(void)
 817 {
 818         int cpu = smp_processor_id();
 819
 820         S390_lowcore.last_update_clock = get_tod_clock();
 821         restore_access_regs(S390_lowcore.access_regs_save_area);
 822         cpu_init();
 823         preempt_disable();
 824         init_cpu_timer();
 825         vtime_init();
 826         pfault_init();
 827         notify_cpu_starting(smp_processor_id());
 828         if (topology_cpu_dedicated(cpu))
 829                 set_cpu_flag(CIF_DEDICATED_CPU);
 830         else
 831                 clear_cpu_flag(CIF_DEDICATED_CPU);
 832         set_cpu_online(smp_processor_id(), true);
 833         inc_irq_stat(CPU_RST);
 834         local_irq_enable();
 835         cpu_startup_entry(CPUHP_AP_ONLINE_IDLE);
 836 }
 837
 838 /*
 839  *      Activate a secondary processor.
 840  */
 841 static void __no_sanitize_address smp_start_secondary(void *cpuvoid)
 842 {
 843         S390_lowcore.restart_stack = (unsigned long) restart_stack;
 844         S390_lowcore.restart_fn = (unsigned long) do_restart;
 845         S390_lowcore.restart_data = 0;
 846         S390_lowcore.restart_source = -1UL;
 847         __ctl_load(S390_lowcore.cregs_save_area, 0, 15);
 848         __load_psw_mask(PSW_KERNEL_BITS | PSW_MASK_DAT);
 849         CALL_ON_STACK(smp_init_secondary, S390_lowcore.kernel_stack, 0);
 850 }
 851
 852 /* Upping and downing of CPUs */
 853 int __cpu_up(unsigned int cpu, struct task_struct *tidle)
 854 {
 855         struct pcpu *pcpu;
 856         int base, i, rc;
 857
 858         pcpu = pcpu_devices + cpu;
 859         if (pcpu->state != CPU_STATE_CONFIGURED)
 860                 return -EIO;
 861         base = smp_get_base_cpu(cpu);
 862         for (i = 0; i <= smp_cpu_mtid; i++) {
 863                 if (base + i < nr_cpu_ids)
 864                         if (cpu_online(base + i))
 865                                 break;
 866         }
 867         /*
 868          * If this is the first CPU of the core to get online
 869          * do an initial CPU reset.
 870          */
 871         if (i > smp_cpu_mtid &&
 872             pcpu_sigp_retry(pcpu_devices + base, SIGP_INITIAL_CPU_RESET, 0) !=
 873             SIGP_CC_ORDER_CODE_ACCEPTED)
 874                 return -EIO;
 875
 876         rc = pcpu_alloc_lowcore(pcpu, cpu);
 877         if (rc)
 878                 return rc;
 879         pcpu_prepare_secondary(pcpu, cpu);
 880         pcpu_attach_task(pcpu, tidle);
 881         pcpu_start_fn(pcpu, smp_start_secondary, NULL);
 882         /* Wait until cpu puts itself in the online & active maps */
 883         while (!cpu_online(cpu))
 884                 cpu_relax();
 885         return 0;
 886 }
 887
 888 static unsigned int setup_possible_cpus __initdata;
 889
 890 static int __init _setup_possible_cpus(char *s)
 891 {
 892         get_option(&s, &setup_possible_cpus);
 893         return 0;
 894 }
 895 early_param("possible_cpus", _setup_possible_cpus);
 896
 897 #ifdef CONFIG_HOTPLUG_CPU
 898
 899 int __cpu_disable(void)
 900 {
 901         unsigned long cregs[16];
 902
 903         /* Handle possible pending IPIs */
 904         smp_handle_ext_call();
 905         set_cpu_online(smp_processor_id(), false);
 906         /* Disable pseudo page faults on this cpu. */
 907         pfault_fini();
 908         /* Disable interrupt sources via control register. */
 909         __ctl_store(cregs, 0, 15);
 910         cregs[0]  &= ~0x0000ee70UL;     /* disable all external interrupts */
 911         cregs[6]  &= ~0xff000000UL;     /* disable all I/O interrupts */
 912         cregs[14] &= ~0x1f000000UL;     /* disable most machine checks */
 913         __ctl_load(cregs, 0, 15);
 914         clear_cpu_flag(CIF_NOHZ_DELAY);
 915         return 0;
 916 }
 917
 918 void __cpu_die(unsigned int cpu)
 919 {
 920         struct pcpu *pcpu;
 921
 922         /* Wait until target cpu is down */
 923         pcpu = pcpu_devices + cpu;
 924         while (!pcpu_stopped(pcpu))
 925                 cpu_relax();
 926         pcpu_free_lowcore(pcpu);
 927         cpumask_clear_cpu(cpu, mm_cpumask(&init_mm));
 928         cpumask_clear_cpu(cpu, &init_mm.context.cpu_attach_mask);
 929 }
 930
 931 void __noreturn cpu_die(void)
 932 {
 933         idle_task_exit();
 934         __bpon();
 935         pcpu_sigp_retry(pcpu_devices + smp_processor_id(), SIGP_STOP, 0);
 936         for (;;) ;
 937 }
 938
 939 #endif /* CONFIG_HOTPLUG_CPU */
 940
 941 void __init smp_fill_possible_mask(void)
 942 {
 943         unsigned int possible, sclp_max, cpu;
 944
 945         sclp_max = max(sclp.mtid, sclp.mtid_cp) + 1;
 946         sclp_max = min(smp_max_threads, sclp_max);
 947         sclp_max = (sclp.max_cores * sclp_max) ?: nr_cpu_ids;
 948         possible = setup_possible_cpus ?: nr_cpu_ids;
 949         possible = min(possible, sclp_max);
 950         for (cpu = 0; cpu < possible && cpu < nr_cpu_ids; cpu++)
 951                 set_cpu_possible(cpu, true);
 952 }
 953
 954 void __init smp_prepare_cpus(unsigned int max_cpus)
 955 {
 956         /* request the 0x1201 emergency signal external interrupt */
 957         if (register_external_irq(EXT_IRQ_EMERGENCY_SIG, do_ext_call_interrupt))
 958                 panic("Couldn't request external interrupt 0x1201");
 959         /* request the 0x1202 external call external interrupt */
 960         if (register_external_irq(EXT_IRQ_EXTERNAL_CALL, do_ext_call_interrupt))
 961                 panic("Couldn't request external interrupt 0x1202");
 962 }
 963
 964 void __init smp_prepare_boot_cpu(void)
 965 {
 966         struct pcpu *pcpu = pcpu_devices;
 967
 968         WARN_ON(!cpu_present(0) || !cpu_online(0));
 969         pcpu->state = CPU_STATE_CONFIGURED;
 970         pcpu->lowcore = (struct lowcore *)(unsigned long) store_prefix();
 971         S390_lowcore.percpu_offset = __per_cpu_offset[0];
 972         smp_cpu_set_polarization(0, POLARIZATION_UNKNOWN);
 973 }
 974
 975 void __init smp_cpus_done(unsigned int max_cpus)
 976 {
 977 }
 978
 979 void __init smp_setup_processor_id(void)
 980 {
 981         pcpu_devices[0].address = stap();
 982         S390_lowcore.cpu_nr = 0;
 983         S390_lowcore.spinlock_lockval = arch_spin_lockval(0);
 984         S390_lowcore.spinlock_index = 0;
 985 }
 986
 987 /*
 988  * the frequency of the profiling timer can be changed
 989  * by writing a multiplier value into /proc/profile.
 990  *
 991  * usually you want to run this on all CPUs ;)
 992  */
 993 int setup_profiling_timer(unsigned int multiplier)
 994 {
 995         return 0;
 996 }
 997
 998 #ifdef CONFIG_HOTPLUG_CPU
 999 static ssize_t cpu_configure_show(struct device *dev,
1000                                   struct device_attribute *attr, char *buf)
1001 {
1002         ssize_t count;
1003
1004         mutex_lock(&smp_cpu_state_mutex);
1005         count = sprintf(buf, "%d\n", pcpu_devices[dev->id].state);
1006         mutex_unlock(&smp_cpu_state_mutex);
1007         return count;
1008 }
1009
1010 static ssize_t cpu_configure_store(struct device *dev,
1011                                    struct device_attribute *attr,
1012                                    const char *buf, size_t count)
1013 {
1014         struct pcpu *pcpu;
1015         int cpu, val, rc, i;
1016         char delim;
1017
1018         if (sscanf(buf, "%d %c", &val, &delim) != 1)
1019                 return -EINVAL;
1020         if (val != 0 && val != 1)
1021                 return -EINVAL;
1022         get_online_cpus();
1023         mutex_lock(&smp_cpu_state_mutex);
1024         rc = -EBUSY;
1025         /* disallow configuration changes of online cpus and cpu 0 */
1026         cpu = dev->id;
1027         cpu = smp_get_base_cpu(cpu);
1028         if (cpu == 0)
1029                 goto out;
1030         for (i = 0; i <= smp_cpu_mtid; i++)
1031                 if (cpu_online(cpu + i))
1032                         goto out;
1033         pcpu = pcpu_devices + cpu;
1034         rc = 0;
1035         switch (val) {
1036         case 0:
1037                 if (pcpu->state != CPU_STATE_CONFIGURED)
1038                         break;
1039                 rc = sclp_core_deconfigure(pcpu->address >> smp_cpu_mt_shift);
1040                 if (rc)
1041                         break;
1042                 for (i = 0; i <= smp_cpu_mtid; i++) {
1043                         if (cpu + i >= nr_cpu_ids || !cpu_present(cpu + i))
1044                                 continue;
1045                         pcpu[i].state = CPU_STATE_STANDBY;
1046                         smp_cpu_set_polarization(cpu + i,
1047                                                  POLARIZATION_UNKNOWN);
1048                 }
1049                 topology_expect_change();
1050                 break;
1051         case 1:
1052                 if (pcpu->state != CPU_STATE_STANDBY)
1053                         break;
1054                 rc = sclp_core_configure(pcpu->address >> smp_cpu_mt_shift);
1055                 if (rc)
1056                         break;
1057                 for (i = 0; i <= smp_cpu_mtid; i++) {
1058                         if (cpu + i >= nr_cpu_ids || !cpu_present(cpu + i))
1059                                 continue;
1060                         pcpu[i].state = CPU_STATE_CONFIGURED;
1061                         smp_cpu_set_polarization(cpu + i,
1062                                                  POLARIZATION_UNKNOWN);
1063                 }
1064                 topology_expect_change();
1065                 break;
1066         default:
1067                 break;
1068         }
1069 out:
1070         mutex_unlock(&smp_cpu_state_mutex);
1071         put_online_cpus();
1072         return rc ? rc : count;
1073 }
1074 static DEVICE_ATTR(configure, 0644, cpu_configure_show, cpu_configure_store);
1075 #endif /* CONFIG_HOTPLUG_CPU */
1076
1077 static ssize_t show_cpu_address(struct device *dev,
1078                                 struct device_attribute *attr, char *buf)
1079 {
1080         return sprintf(buf, "%d\n", pcpu_devices[dev->id].address);
1081 }
1082 static DEVICE_ATTR(address, 0444, show_cpu_address, NULL);
1083
1084 static struct attribute *cpu_common_attrs[] = {
1085 #ifdef CONFIG_HOTPLUG_CPU
1086         &dev_attr_configure.attr,
1087 #endif
1088         &dev_attr_address.attr,
1089         NULL,
1090 };
1091
1092 static struct attribute_group cpu_common_attr_group = {
1093         .attrs = cpu_common_attrs,
1094 };
1095
1096 static struct attribute *cpu_online_attrs[] = {
1097         &dev_attr_idle_count.attr,
1098         &dev_attr_idle_time_us.attr,
1099         NULL,
1100 };
1101
1102 static struct attribute_group cpu_online_attr_group = {
1103         .attrs = cpu_online_attrs,
1104 };
1105
1106 static int smp_cpu_online(unsigned int cpu)
1107 {
1108         struct device *s = &per_cpu(cpu_device, cpu)->dev;
1109
1110         return sysfs_create_group(&s->kobj, &cpu_online_attr_group);
1111 }
1112 static int smp_cpu_pre_down(unsigned int cpu)
1113 {
1114         struct device *s = &per_cpu(cpu_device, cpu)->dev;
1115
1116         sysfs_remove_group(&s->kobj, &cpu_online_attr_group);
1117         return 0;
1118 }
1119
1120 static int smp_add_present_cpu(int cpu)
1121 {
1122         struct device *s;
1123         struct cpu *c;
1124         int rc;
1125
1126         c = kzalloc(sizeof(*c), GFP_KERNEL);
1127         if (!c)
1128                 return -ENOMEM;
1129         per_cpu(cpu_device, cpu) = c;
1130         s = &c->dev;
1131         c->hotpluggable = 1;
1132         rc = register_cpu(c, cpu);
1133         if (rc)
1134                 goto out;
1135         rc = sysfs_create_group(&s->kobj, &cpu_common_attr_group);
1136         if (rc)
1137                 goto out_cpu;
1138         rc = topology_cpu_init(c);
1139         if (rc)
1140                 goto out_topology;
1141         return 0;
1142
1143 out_topology:
1144         sysfs_remove_group(&s->kobj, &cpu_common_attr_group);
1145 out_cpu:
1146 #ifdef CONFIG_HOTPLUG_CPU
1147         unregister_cpu(c);
1148 #endif
1149 out:
1150         return rc;
1151 }
1152
1153 #ifdef CONFIG_HOTPLUG_CPU
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         get_online_cpus();
1165         mutex_lock(&smp_cpu_state_mutex);
1166         nr = __smp_rescan_cpus(info, 1);
1167         mutex_unlock(&smp_cpu_state_mutex);
1168         put_online_cpus();
1169         kfree(info);
1170         if (nr)
1171                 topology_schedule_update();
1172         return 0;
1173 }
1174
1175 static ssize_t __ref rescan_store(struct device *dev,
1176                                   struct device_attribute *attr,
1177                                   const char *buf,
1178                                   size_t count)
1179 {
1180         int rc;
1181
1182         rc = lock_device_hotplug_sysfs();
1183         if (rc)
1184                 return rc;
1185         rc = smp_rescan_cpus();
1186         unlock_device_hotplug();
1187         return rc ? rc : count;
1188 }
1189 static DEVICE_ATTR_WO(rescan);
1190 #endif /* CONFIG_HOTPLUG_CPU */
1191
1192 static int __init s390_smp_init(void)
1193 {
1194         int cpu, rc = 0;
1195
1196 #ifdef CONFIG_HOTPLUG_CPU
1197         rc = device_create_file(cpu_subsys.dev_root, &dev_attr_rescan);
1198         if (rc)
1199                 return rc;
1200 #endif
1201         for_each_present_cpu(cpu) {
1202                 rc = smp_add_present_cpu(cpu);
1203                 if (rc)
1204                         goto out;
1205         }
1206
1207         rc = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "s390/smp:online",
1208                                smp_cpu_online, smp_cpu_pre_down);
1209         rc = rc <= 0 ? rc : 0;
1210 out:
1211         return rc;
1212 }
1213 subsys_initcall(s390_smp_init);