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 = lc->steal_timer = 0;
 270         __ctl_store(lc->cregs_save_area, 0, 15);
 271         save_access_regs((unsigned int *) lc->access_regs_save_area);
 272         memcpy(lc->stfle_fac_list, S390_lowcore.stfle_fac_list,
 273                sizeof(lc->stfle_fac_list));
 274         memcpy(lc->alt_stfle_fac_list, S390_lowcore.alt_stfle_fac_list,
 275                sizeof(lc->alt_stfle_fac_list));
 276         arch_spin_lock_setup(cpu);
 277 }
 278
 279 static void pcpu_attach_task(struct pcpu *pcpu, struct task_struct *tsk)
 280 {
 281         struct lowcore *lc = pcpu->lowcore;
 282
 283         lc->kernel_stack = (unsigned long) task_stack_page(tsk)
 284                 + THREAD_SIZE - STACK_FRAME_OVERHEAD - sizeof(struct pt_regs);
 285         lc->current_task = (unsigned long) tsk;
 286         lc->lpp = LPP_MAGIC;
 287         lc->current_pid = tsk->pid;
 288         lc->user_timer = tsk->thread.user_timer;
 289         lc->guest_timer = tsk->thread.guest_timer;
 290         lc->system_timer = tsk->thread.system_timer;
 291         lc->hardirq_timer = tsk->thread.hardirq_timer;
 292         lc->softirq_timer = tsk->thread.softirq_timer;
 293         lc->steal_timer = 0;
 294 }
 295
 296 static void pcpu_start_fn(struct pcpu *pcpu, void (*func)(void *), void *data)
 297 {
 298         struct lowcore *lc = pcpu->lowcore;
 299
 300         lc->restart_stack = lc->nodat_stack;
 301         lc->restart_fn = (unsigned long) func;
 302         lc->restart_data = (unsigned long) data;
 303         lc->restart_source = -1UL;
 304         pcpu_sigp_retry(pcpu, SIGP_RESTART, 0);
 305 }
 306
 307 /*
 308  * Call function via PSW restart on pcpu and stop the current cpu.
 309  */
 310 static void __pcpu_delegate(void (*func)(void*), void *data)
 311 {
 312         func(data);     /* should not return */
 313 }
 314
 315 static void __no_sanitize_address pcpu_delegate(struct pcpu *pcpu,
 316                                                 void (*func)(void *),
 317                                                 void *data, unsigned long stack)
 318 {
 319         struct lowcore *lc = lowcore_ptr[pcpu - pcpu_devices];
 320         unsigned long source_cpu = stap();
 321
 322         __load_psw_mask(PSW_KERNEL_BITS | PSW_MASK_DAT);
 323         if (pcpu->address == source_cpu)
 324                 CALL_ON_STACK(__pcpu_delegate, stack, 2, func, data);
 325         /* Stop target cpu (if func returns this stops the current cpu). */
 326         pcpu_sigp_retry(pcpu, SIGP_STOP, 0);
 327         /* Restart func on the target cpu and stop the current cpu. */
 328         mem_assign_absolute(lc->restart_stack, stack);
 329         mem_assign_absolute(lc->restart_fn, (unsigned long) func);
 330         mem_assign_absolute(lc->restart_data, (unsigned long) data);
 331         mem_assign_absolute(lc->restart_source, source_cpu);
 332         __bpon();
 333         asm volatile(
 334                 "0:     sigp    0,%0,%2 # sigp restart to target cpu\n"
 335                 "       brc     2,0b    # busy, try again\n"
 336                 "1:     sigp    0,%1,%3 # sigp stop to current cpu\n"
 337                 "       brc     2,1b    # busy, try again\n"
 338                 : : "d" (pcpu->address), "d" (source_cpu),
 339                     "K" (SIGP_RESTART), "K" (SIGP_STOP)
 340                 : "0", "1", "cc");
 341         for (;;) ;
 342 }
 343
 344 /*
 345  * Enable additional logical cpus for multi-threading.
 346  */
 347 static int pcpu_set_smt(unsigned int mtid)
 348 {
 349         int cc;
 350
 351         if (smp_cpu_mtid == mtid)
 352                 return 0;
 353         cc = __pcpu_sigp(0, SIGP_SET_MULTI_THREADING, mtid, NULL);
 354         if (cc == 0) {
 355                 smp_cpu_mtid = mtid;
 356                 smp_cpu_mt_shift = 0;
 357                 while (smp_cpu_mtid >= (1U << smp_cpu_mt_shift))
 358                         smp_cpu_mt_shift++;
 359                 pcpu_devices[0].address = stap();
 360         }
 361         return cc;
 362 }
 363
 364 /*
 365  * Call function on an online CPU.
 366  */
 367 void smp_call_online_cpu(void (*func)(void *), void *data)
 368 {
 369         struct pcpu *pcpu;
 370
 371         /* Use the current cpu if it is online. */
 372         pcpu = pcpu_find_address(cpu_online_mask, stap());
 373         if (!pcpu)
 374                 /* Use the first online cpu. */
 375                 pcpu = pcpu_devices + cpumask_first(cpu_online_mask);
 376         pcpu_delegate(pcpu, func, data, (unsigned long) restart_stack);
 377 }
 378
 379 /*
 380  * Call function on the ipl CPU.
 381  */
 382 void smp_call_ipl_cpu(void (*func)(void *), void *data)
 383 {
 384         struct lowcore *lc = pcpu_devices->lowcore;
 385
 386         if (pcpu_devices[0].address == stap())
 387                 lc = &S390_lowcore;
 388
 389         pcpu_delegate(&pcpu_devices[0], func, data,
 390                       lc->nodat_stack);
 391 }
 392
 393 int smp_find_processor_id(u16 address)
 394 {
 395         int cpu;
 396
 397         for_each_present_cpu(cpu)
 398                 if (pcpu_devices[cpu].address == address)
 399                         return cpu;
 400         return -1;
 401 }
 402
 403 bool arch_vcpu_is_preempted(int cpu)
 404 {
 405         if (test_cpu_flag_of(CIF_ENABLED_WAIT, cpu))
 406                 return false;
 407         if (pcpu_running(pcpu_devices + cpu))
 408                 return false;
 409         return true;
 410 }
 411 EXPORT_SYMBOL(arch_vcpu_is_preempted);
 412
 413 void smp_yield_cpu(int cpu)
 414 {
 415         if (MACHINE_HAS_DIAG9C) {
 416                 diag_stat_inc_norecursion(DIAG_STAT_X09C);
 417                 asm volatile("diag %0,0,0x9c"
 418                              : : "d" (pcpu_devices[cpu].address));
 419         } else if (MACHINE_HAS_DIAG44) {
 420                 diag_stat_inc_norecursion(DIAG_STAT_X044);
 421                 asm volatile("diag 0,0,0x44");
 422         }
 423 }
 424
 425 /*
 426  * Send cpus emergency shutdown signal. This gives the cpus the
 427  * opportunity to complete outstanding interrupts.
 428  */
 429 void notrace smp_emergency_stop(void)
 430 {
 431         cpumask_t cpumask;
 432         u64 end;
 433         int cpu;
 434
 435         cpumask_copy(&cpumask, cpu_online_mask);
 436         cpumask_clear_cpu(smp_processor_id(), &cpumask);
 437
 438         end = get_tod_clock() + (1000000UL << 12);
 439         for_each_cpu(cpu, &cpumask) {
 440                 struct pcpu *pcpu = pcpu_devices + cpu;
 441                 set_bit(ec_stop_cpu, &pcpu->ec_mask);
 442                 while (__pcpu_sigp(pcpu->address, SIGP_EMERGENCY_SIGNAL,
 443                                    0, NULL) == SIGP_CC_BUSY &&
 444                        get_tod_clock() < end)
 445                         cpu_relax();
 446         }
 447         while (get_tod_clock() < end) {
 448                 for_each_cpu(cpu, &cpumask)
 449                         if (pcpu_stopped(pcpu_devices + cpu))
 450                                 cpumask_clear_cpu(cpu, &cpumask);
 451                 if (cpumask_empty(&cpumask))
 452                         break;
 453                 cpu_relax();
 454         }
 455 }
 456 NOKPROBE_SYMBOL(smp_emergency_stop);
 457
 458 /*
 459  * Stop all cpus but the current one.
 460  */
 461 void smp_send_stop(void)
 462 {
 463         int cpu;
 464
 465         /* Disable all interrupts/machine checks */
 466         __load_psw_mask(PSW_KERNEL_BITS | PSW_MASK_DAT);
 467         trace_hardirqs_off();
 468
 469         debug_set_critical();
 470
 471         if (oops_in_progress)
 472                 smp_emergency_stop();
 473
 474         /* stop all processors */
 475         for_each_online_cpu(cpu) {
 476                 if (cpu == smp_processor_id())
 477                         continue;
 478                 pcpu_sigp_retry(pcpu_devices + cpu, SIGP_STOP, 0);
 479                 while (!pcpu_stopped(pcpu_devices + cpu))
 480                         cpu_relax();
 481         }
 482 }
 483
 484 /*
 485  * This is the main routine where commands issued by other
 486  * cpus are handled.
 487  */
 488 static void smp_handle_ext_call(void)
 489 {
 490         unsigned long bits;
 491
 492         /* handle bit signal external calls */
 493         bits = xchg(&pcpu_devices[smp_processor_id()].ec_mask, 0);
 494         if (test_bit(ec_stop_cpu, &bits))
 495                 smp_stop_cpu();
 496         if (test_bit(ec_schedule, &bits))
 497                 scheduler_ipi();
 498         if (test_bit(ec_call_function_single, &bits))
 499                 generic_smp_call_function_single_interrupt();
 500 }
 501
 502 static void do_ext_call_interrupt(struct ext_code ext_code,
 503                                   unsigned int param32, unsigned long param64)
 504 {
 505         inc_irq_stat(ext_code.code == 0x1202 ? IRQEXT_EXC : IRQEXT_EMS);
 506         smp_handle_ext_call();
 507 }
 508
 509 void arch_send_call_function_ipi_mask(const struct cpumask *mask)
 510 {
 511         int cpu;
 512
 513         for_each_cpu(cpu, mask)
 514                 pcpu_ec_call(pcpu_devices + cpu, ec_call_function_single);
 515 }
 516
 517 void arch_send_call_function_single_ipi(int cpu)
 518 {
 519         pcpu_ec_call(pcpu_devices + cpu, ec_call_function_single);
 520 }
 521
 522 /*
 523  * this function sends a 'reschedule' IPI to another CPU.
 524  * it goes straight through and wastes no time serializing
 525  * anything. Worst case is that we lose a reschedule ...
 526  */
 527 void smp_send_reschedule(int cpu)
 528 {
 529         pcpu_ec_call(pcpu_devices + cpu, ec_schedule);
 530 }
 531
 532 /*
 533  * parameter area for the set/clear control bit callbacks
 534  */
 535 struct ec_creg_mask_parms {
 536         unsigned long orval;
 537         unsigned long andval;
 538         int cr;
 539 };
 540
 541 /*
 542  * callback for setting/clearing control bits
 543  */
 544 static void smp_ctl_bit_callback(void *info)
 545 {
 546         struct ec_creg_mask_parms *pp = info;
 547         unsigned long cregs[16];
 548
 549         __ctl_store(cregs, 0, 15);
 550         cregs[pp->cr] = (cregs[pp->cr] & pp->andval) | pp->orval;
 551         __ctl_load(cregs, 0, 15);
 552 }
 553
 554 /*
 555  * Set a bit in a control register of all cpus
 556  */
 557 void smp_ctl_set_bit(int cr, int bit)
 558 {
 559         struct ec_creg_mask_parms parms = { 1UL << bit, -1UL, cr };
 560
 561         on_each_cpu(smp_ctl_bit_callback, &parms, 1);
 562 }
 563 EXPORT_SYMBOL(smp_ctl_set_bit);
 564
 565 /*
 566  * Clear a bit in a control register of all cpus
 567  */
 568 void smp_ctl_clear_bit(int cr, int bit)
 569 {
 570         struct ec_creg_mask_parms parms = { 0, ~(1UL << bit), cr };
 571
 572         on_each_cpu(smp_ctl_bit_callback, &parms, 1);
 573 }
 574 EXPORT_SYMBOL(smp_ctl_clear_bit);
 575
 576 #ifdef CONFIG_CRASH_DUMP
 577
 578 int smp_store_status(int cpu)
 579 {
 580         struct pcpu *pcpu = pcpu_devices + cpu;
 581         unsigned long pa;
 582
 583         pa = __pa(&pcpu->lowcore->floating_pt_save_area);
 584         if (__pcpu_sigp_relax(pcpu->address, SIGP_STORE_STATUS_AT_ADDRESS,
 585                               pa) != SIGP_CC_ORDER_CODE_ACCEPTED)
 586                 return -EIO;
 587         if (!MACHINE_HAS_VX && !MACHINE_HAS_GS)
 588                 return 0;
 589         pa = __pa(pcpu->lowcore->mcesad & MCESA_ORIGIN_MASK);
 590         if (MACHINE_HAS_GS)
 591                 pa |= pcpu->lowcore->mcesad & MCESA_LC_MASK;
 592         if (__pcpu_sigp_relax(pcpu->address, SIGP_STORE_ADDITIONAL_STATUS,
 593                               pa) != SIGP_CC_ORDER_CODE_ACCEPTED)
 594                 return -EIO;
 595         return 0;
 596 }
 597
 598 /*
 599  * Collect CPU state of the previous, crashed system.
 600  * There are four cases:
 601  * 1) standard zfcp dump
 602  *    condition: OLDMEM_BASE == NULL && ipl_info.type == IPL_TYPE_FCP_DUMP
 603  *    The state for all CPUs except the boot CPU needs to be collected
 604  *    with sigp stop-and-store-status. The boot CPU state is located in
 605  *    the absolute lowcore of the memory stored in the HSA. The zcore code
 606  *    will copy the boot CPU state from the HSA.
 607  * 2) stand-alone kdump for SCSI (zfcp dump with swapped memory)
 608  *    condition: OLDMEM_BASE != NULL && ipl_info.type == IPL_TYPE_FCP_DUMP
 609  *    The state for all CPUs except the boot CPU needs to be collected
 610  *    with sigp stop-and-store-status. The firmware or the boot-loader
 611  *    stored the registers of the boot CPU in the absolute lowcore in the
 612  *    memory of the old system.
 613  * 3) kdump and the old kernel did not store the CPU state,
 614  *    or stand-alone kdump for DASD
 615  *    condition: OLDMEM_BASE != NULL && !is_kdump_kernel()
 616  *    The state for all CPUs except the boot CPU needs to be collected
 617  *    with sigp stop-and-store-status. The kexec code or the boot-loader
 618  *    stored the registers of the boot CPU in the memory of the old system.
 619  * 4) kdump and the old kernel stored the CPU state
 620  *    condition: OLDMEM_BASE != NULL && is_kdump_kernel()
 621  *    This case does not exist for s390 anymore, setup_arch explicitly
 622  *    deactivates the elfcorehdr= kernel parameter
 623  */
 624 static __init void smp_save_cpu_vxrs(struct save_area *sa, u16 addr,
 625                                      bool is_boot_cpu, unsigned long page)
 626 {
 627         __vector128 *vxrs = (__vector128 *) page;
 628
 629         if (is_boot_cpu)
 630                 vxrs = boot_cpu_vector_save_area;
 631         else
 632                 __pcpu_sigp_relax(addr, SIGP_STORE_ADDITIONAL_STATUS, page);
 633         save_area_add_vxrs(sa, vxrs);
 634 }
 635
 636 static __init void smp_save_cpu_regs(struct save_area *sa, u16 addr,
 637                                      bool is_boot_cpu, unsigned long page)
 638 {
 639         void *regs = (void *) page;
 640
 641         if (is_boot_cpu)
 642                 copy_oldmem_kernel(regs, (void *) __LC_FPREGS_SAVE_AREA, 512);
 643         else
 644                 __pcpu_sigp_relax(addr, SIGP_STORE_STATUS_AT_ADDRESS, page);
 645         save_area_add_regs(sa, regs);
 646 }
 647
 648 void __init smp_save_dump_cpus(void)
 649 {
 650         int addr, boot_cpu_addr, max_cpu_addr;
 651         struct save_area *sa;
 652         unsigned long page;
 653         bool is_boot_cpu;
 654
 655         if (!(OLDMEM_BASE || ipl_info.type == IPL_TYPE_FCP_DUMP))
 656                 /* No previous system present, normal boot. */
 657                 return;
 658         /* Allocate a page as dumping area for the store status sigps */
 659         page = memblock_alloc_base(PAGE_SIZE, PAGE_SIZE, 1UL << 31);
 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 (__pcpu_sigp_relax(addr, SIGP_SENSE, 0) ==
 666                     SIGP_CC_NOT_OPERATIONAL)
 667                         continue;
 668                 is_boot_cpu = (addr == boot_cpu_addr);
 669                 /* Allocate save area */
 670                 sa = save_area_alloc(is_boot_cpu);
 671                 if (!sa)
 672                         panic("could not allocate memory for save area\n");
 673                 if (MACHINE_HAS_VX)
 674                         /* Get the vector registers */
 675                         smp_save_cpu_vxrs(sa, addr, is_boot_cpu, page);
 676                 /*
 677                  * For a zfcp dump OLDMEM_BASE == NULL and the registers
 678                  * of the boot CPU are stored in the HSA. To retrieve
 679                  * these registers an SCLP request is required which is
 680                  * done by drivers/s390/char/zcore.c:init_cpu_info()
 681                  */
 682                 if (!is_boot_cpu || OLDMEM_BASE)
 683                         /* Get the CPU registers */
 684                         smp_save_cpu_regs(sa, addr, is_boot_cpu, page);
 685         }
 686         memblock_free(page, PAGE_SIZE);
 687         diag308_reset();
 688         pcpu_set_smt(0);
 689 }
 690 #endif /* CONFIG_CRASH_DUMP */
 691
 692 void smp_cpu_set_polarization(int cpu, int val)
 693 {
 694         pcpu_devices[cpu].polarization = val;
 695 }
 696
 697 int smp_cpu_get_polarization(int cpu)
 698 {
 699         return pcpu_devices[cpu].polarization;
 700 }
 701
 702 static void __ref smp_get_core_info(struct sclp_core_info *info, int early)
 703 {
 704         static int use_sigp_detection;
 705         int address;
 706
 707         if (use_sigp_detection || sclp_get_core_info(info, early)) {
 708                 use_sigp_detection = 1;
 709                 for (address = 0;
 710                      address < (SCLP_MAX_CORES << smp_cpu_mt_shift);
 711                      address += (1U << smp_cpu_mt_shift)) {
 712                         if (__pcpu_sigp_relax(address, SIGP_SENSE, 0) ==
 713                             SIGP_CC_NOT_OPERATIONAL)
 714                                 continue;
 715                         info->core[info->configured].core_id =
 716                                 address >> smp_cpu_mt_shift;
 717                         info->configured++;
 718                 }
 719                 info->combined = info->configured;
 720         }
 721 }
 722
 723 static int smp_add_present_cpu(int cpu);
 724
 725 static int __smp_rescan_cpus(struct sclp_core_info *info, int sysfs_add)
 726 {
 727         struct pcpu *pcpu;
 728         cpumask_t avail;
 729         int cpu, nr, i, j;
 730         u16 address;
 731
 732         nr = 0;
 733         cpumask_xor(&avail, cpu_possible_mask, cpu_present_mask);
 734         cpu = cpumask_first(&avail);
 735         for (i = 0; (i < info->combined) && (cpu < nr_cpu_ids); i++) {
 736                 if (sclp.has_core_type && info->core[i].type != boot_core_type)
 737                         continue;
 738                 address = info->core[i].core_id << smp_cpu_mt_shift;
 739                 for (j = 0; j <= smp_cpu_mtid; j++) {
 740                         if (pcpu_find_address(cpu_present_mask, address + j))
 741                                 continue;
 742                         pcpu = pcpu_devices + cpu;
 743                         pcpu->address = address + j;
 744                         pcpu->state =
 745                                 (cpu >= info->configured*(smp_cpu_mtid + 1)) ?
 746                                 CPU_STATE_STANDBY : CPU_STATE_CONFIGURED;
 747                         smp_cpu_set_polarization(cpu, POLARIZATION_UNKNOWN);
 748                         set_cpu_present(cpu, true);
 749                         if (sysfs_add && smp_add_present_cpu(cpu) != 0)
 750                                 set_cpu_present(cpu, false);
 751                         else
 752                                 nr++;
 753                         cpu = cpumask_next(cpu, &avail);
 754                         if (cpu >= nr_cpu_ids)
 755                                 break;
 756                 }
 757         }
 758         return nr;
 759 }
 760
 761 void __init smp_detect_cpus(void)
 762 {
 763         unsigned int cpu, mtid, c_cpus, s_cpus;
 764         struct sclp_core_info *info;
 765         u16 address;
 766
 767         /* Get CPU information */
 768         info = memblock_alloc(sizeof(*info), 8);
 769         smp_get_core_info(info, 1);
 770         /* Find boot CPU type */
 771         if (sclp.has_core_type) {
 772                 address = stap();
 773                 for (cpu = 0; cpu < info->combined; cpu++)
 774                         if (info->core[cpu].core_id == address) {
 775                                 /* The boot cpu dictates the cpu type. */
 776                                 boot_core_type = info->core[cpu].type;
 777                                 break;
 778                         }
 779                 if (cpu >= info->combined)
 780                         panic("Could not find boot CPU type");
 781         }
 782
 783         /* Set multi-threading state for the current system */
 784         mtid = boot_core_type ? sclp.mtid : sclp.mtid_cp;
 785         mtid = (mtid < smp_max_threads) ? mtid : smp_max_threads - 1;
 786         pcpu_set_smt(mtid);
 787
 788         /* Print number of CPUs */
 789         c_cpus = s_cpus = 0;
 790         for (cpu = 0; cpu < info->combined; cpu++) {
 791                 if (sclp.has_core_type &&
 792                     info->core[cpu].type != boot_core_type)
 793                         continue;
 794                 if (cpu < info->configured)
 795                         c_cpus += smp_cpu_mtid + 1;
 796                 else
 797                         s_cpus += smp_cpu_mtid + 1;
 798         }
 799         pr_info("%d configured CPUs, %d standby CPUs\n", c_cpus, s_cpus);
 800
 801         /* Add CPUs present at boot */
 802         get_online_cpus();
 803         __smp_rescan_cpus(info, 0);
 804         put_online_cpus();
 805         memblock_free_early((unsigned long)info, sizeof(*info));
 806 }
 807
 808 static void smp_init_secondary(void)
 809 {
 810         int cpu = smp_processor_id();
 811
 812         S390_lowcore.last_update_clock = get_tod_clock();
 813         restore_access_regs(S390_lowcore.access_regs_save_area);
 814         cpu_init();
 815         preempt_disable();
 816         init_cpu_timer();
 817         vtime_init();
 818         pfault_init();
 819         notify_cpu_starting(smp_processor_id());
 820         if (topology_cpu_dedicated(cpu))
 821                 set_cpu_flag(CIF_DEDICATED_CPU);
 822         else
 823                 clear_cpu_flag(CIF_DEDICATED_CPU);
 824         set_cpu_online(smp_processor_id(), true);
 825         inc_irq_stat(CPU_RST);
 826         local_irq_enable();
 827         cpu_startup_entry(CPUHP_AP_ONLINE_IDLE);
 828 }
 829
 830 /*
 831  *      Activate a secondary processor.
 832  */
 833 static void __no_sanitize_address smp_start_secondary(void *cpuvoid)
 834 {
 835         S390_lowcore.restart_stack = (unsigned long) restart_stack;
 836         S390_lowcore.restart_fn = (unsigned long) do_restart;
 837         S390_lowcore.restart_data = 0;
 838         S390_lowcore.restart_source = -1UL;
 839         __ctl_load(S390_lowcore.cregs_save_area, 0, 15);
 840         __load_psw_mask(PSW_KERNEL_BITS | PSW_MASK_DAT);
 841         CALL_ON_STACK(smp_init_secondary, S390_lowcore.kernel_stack, 0);
 842 }
 843
 844 /* Upping and downing of CPUs */
 845 int __cpu_up(unsigned int cpu, struct task_struct *tidle)
 846 {
 847         struct pcpu *pcpu;
 848         int base, i, rc;
 849
 850         pcpu = pcpu_devices + cpu;
 851         if (pcpu->state != CPU_STATE_CONFIGURED)
 852                 return -EIO;
 853         base = smp_get_base_cpu(cpu);
 854         for (i = 0; i <= smp_cpu_mtid; i++) {
 855                 if (base + i < nr_cpu_ids)
 856                         if (cpu_online(base + i))
 857                                 break;
 858         }
 859         /*
 860          * If this is the first CPU of the core to get online
 861          * do an initial CPU reset.
 862          */
 863         if (i > smp_cpu_mtid &&
 864             pcpu_sigp_retry(pcpu_devices + base, SIGP_INITIAL_CPU_RESET, 0) !=
 865             SIGP_CC_ORDER_CODE_ACCEPTED)
 866                 return -EIO;
 867
 868         rc = pcpu_alloc_lowcore(pcpu, cpu);
 869         if (rc)
 870                 return rc;
 871         pcpu_prepare_secondary(pcpu, cpu);
 872         pcpu_attach_task(pcpu, tidle);
 873         pcpu_start_fn(pcpu, smp_start_secondary, NULL);
 874         /* Wait until cpu puts itself in the online & active maps */
 875         while (!cpu_online(cpu))
 876                 cpu_relax();
 877         return 0;
 878 }
 879
 880 static unsigned int setup_possible_cpus __initdata;
 881
 882 static int __init _setup_possible_cpus(char *s)
 883 {
 884         get_option(&s, &setup_possible_cpus);
 885         return 0;
 886 }
 887 early_param("possible_cpus", _setup_possible_cpus);
 888
 889 #ifdef CONFIG_HOTPLUG_CPU
 890
 891 int __cpu_disable(void)
 892 {
 893         unsigned long cregs[16];
 894
 895         /* Handle possible pending IPIs */
 896         smp_handle_ext_call();
 897         set_cpu_online(smp_processor_id(), false);
 898         /* Disable pseudo page faults on this cpu. */
 899         pfault_fini();
 900         /* Disable interrupt sources via control register. */
 901         __ctl_store(cregs, 0, 15);
 902         cregs[0]  &= ~0x0000ee70UL;     /* disable all external interrupts */
 903         cregs[6]  &= ~0xff000000UL;     /* disable all I/O interrupts */
 904         cregs[14] &= ~0x1f000000UL;     /* disable most machine checks */
 905         __ctl_load(cregs, 0, 15);
 906         clear_cpu_flag(CIF_NOHZ_DELAY);
 907         return 0;
 908 }
 909
 910 void __cpu_die(unsigned int cpu)
 911 {
 912         struct pcpu *pcpu;
 913
 914         /* Wait until target cpu is down */
 915         pcpu = pcpu_devices + cpu;
 916         while (!pcpu_stopped(pcpu))
 917                 cpu_relax();
 918         pcpu_free_lowcore(pcpu);
 919         cpumask_clear_cpu(cpu, mm_cpumask(&init_mm));
 920         cpumask_clear_cpu(cpu, &init_mm.context.cpu_attach_mask);
 921 }
 922
 923 void __noreturn cpu_die(void)
 924 {
 925         idle_task_exit();
 926         __bpon();
 927         pcpu_sigp_retry(pcpu_devices + smp_processor_id(), SIGP_STOP, 0);
 928         for (;;) ;
 929 }
 930
 931 #endif /* CONFIG_HOTPLUG_CPU */
 932
 933 void __init smp_fill_possible_mask(void)
 934 {
 935         unsigned int possible, sclp_max, cpu;
 936
 937         sclp_max = max(sclp.mtid, sclp.mtid_cp) + 1;
 938         sclp_max = min(smp_max_threads, sclp_max);
 939         sclp_max = (sclp.max_cores * sclp_max) ?: nr_cpu_ids;
 940         possible = setup_possible_cpus ?: nr_cpu_ids;
 941         possible = min(possible, sclp_max);
 942         for (cpu = 0; cpu < possible && cpu < nr_cpu_ids; cpu++)
 943                 set_cpu_possible(cpu, true);
 944 }
 945
 946 void __init smp_prepare_cpus(unsigned int max_cpus)
 947 {
 948         /* request the 0x1201 emergency signal external interrupt */
 949         if (register_external_irq(EXT_IRQ_EMERGENCY_SIG, do_ext_call_interrupt))
 950                 panic("Couldn't request external interrupt 0x1201");
 951         /* request the 0x1202 external call external interrupt */
 952         if (register_external_irq(EXT_IRQ_EXTERNAL_CALL, do_ext_call_interrupt))
 953                 panic("Couldn't request external interrupt 0x1202");
 954 }
 955
 956 void __init smp_prepare_boot_cpu(void)
 957 {
 958         struct pcpu *pcpu = pcpu_devices;
 959
 960         WARN_ON(!cpu_present(0) || !cpu_online(0));
 961         pcpu->state = CPU_STATE_CONFIGURED;
 962         pcpu->lowcore = (struct lowcore *)(unsigned long) store_prefix();
 963         S390_lowcore.percpu_offset = __per_cpu_offset[0];
 964         smp_cpu_set_polarization(0, POLARIZATION_UNKNOWN);
 965 }
 966
 967 void __init smp_cpus_done(unsigned int max_cpus)
 968 {
 969 }
 970
 971 void __init smp_setup_processor_id(void)
 972 {
 973         pcpu_devices[0].address = stap();
 974         S390_lowcore.cpu_nr = 0;
 975         S390_lowcore.spinlock_lockval = arch_spin_lockval(0);
 976         S390_lowcore.spinlock_index = 0;
 977 }
 978
 979 /*
 980  * the frequency of the profiling timer can be changed
 981  * by writing a multiplier value into /proc/profile.
 982  *
 983  * usually you want to run this on all CPUs ;)
 984  */
 985 int setup_profiling_timer(unsigned int multiplier)
 986 {
 987         return 0;
 988 }
 989
 990 #ifdef CONFIG_HOTPLUG_CPU
 991 static ssize_t cpu_configure_show(struct device *dev,
 992                                   struct device_attribute *attr, char *buf)
 993 {
 994         ssize_t count;
 995
 996         mutex_lock(&smp_cpu_state_mutex);
 997         count = sprintf(buf, "%d\n", pcpu_devices[dev->id].state);
 998         mutex_unlock(&smp_cpu_state_mutex);
 999         return count;
1000 }
1001
1002 static ssize_t cpu_configure_store(struct device *dev,
1003                                    struct device_attribute *attr,
1004                                    const char *buf, size_t count)
1005 {
1006         struct pcpu *pcpu;
1007         int cpu, val, rc, i;
1008         char delim;
1009
1010         if (sscanf(buf, "%d %c", &val, &delim) != 1)
1011                 return -EINVAL;
1012         if (val != 0 && val != 1)
1013                 return -EINVAL;
1014         get_online_cpus();
1015         mutex_lock(&smp_cpu_state_mutex);
1016         rc = -EBUSY;
1017         /* disallow configuration changes of online cpus and cpu 0 */
1018         cpu = dev->id;
1019         cpu = smp_get_base_cpu(cpu);
1020         if (cpu == 0)
1021                 goto out;
1022         for (i = 0; i <= smp_cpu_mtid; i++)
1023                 if (cpu_online(cpu + i))
1024                         goto out;
1025         pcpu = pcpu_devices + cpu;
1026         rc = 0;
1027         switch (val) {
1028         case 0:
1029                 if (pcpu->state != CPU_STATE_CONFIGURED)
1030                         break;
1031                 rc = sclp_core_deconfigure(pcpu->address >> smp_cpu_mt_shift);
1032                 if (rc)
1033                         break;
1034                 for (i = 0; i <= smp_cpu_mtid; i++) {
1035                         if (cpu + i >= nr_cpu_ids || !cpu_present(cpu + i))
1036                                 continue;
1037                         pcpu[i].state = CPU_STATE_STANDBY;
1038                         smp_cpu_set_polarization(cpu + i,
1039                                                  POLARIZATION_UNKNOWN);
1040                 }
1041                 topology_expect_change();
1042                 break;
1043         case 1:
1044                 if (pcpu->state != CPU_STATE_STANDBY)
1045                         break;
1046                 rc = sclp_core_configure(pcpu->address >> smp_cpu_mt_shift);
1047                 if (rc)
1048                         break;
1049                 for (i = 0; i <= smp_cpu_mtid; i++) {
1050                         if (cpu + i >= nr_cpu_ids || !cpu_present(cpu + i))
1051                                 continue;
1052                         pcpu[i].state = CPU_STATE_CONFIGURED;
1053                         smp_cpu_set_polarization(cpu + i,
1054                                                  POLARIZATION_UNKNOWN);
1055                 }
1056                 topology_expect_change();
1057                 break;
1058         default:
1059                 break;
1060         }
1061 out:
1062         mutex_unlock(&smp_cpu_state_mutex);
1063         put_online_cpus();
1064         return rc ? rc : count;
1065 }
1066 static DEVICE_ATTR(configure, 0644, cpu_configure_show, cpu_configure_store);
1067 #endif /* CONFIG_HOTPLUG_CPU */
1068
1069 static ssize_t show_cpu_address(struct device *dev,
1070                                 struct device_attribute *attr, char *buf)
1071 {
1072         return sprintf(buf, "%d\n", pcpu_devices[dev->id].address);
1073 }
1074 static DEVICE_ATTR(address, 0444, show_cpu_address, NULL);
1075
1076 static struct attribute *cpu_common_attrs[] = {
1077 #ifdef CONFIG_HOTPLUG_CPU
1078         &dev_attr_configure.attr,
1079 #endif
1080         &dev_attr_address.attr,
1081         NULL,
1082 };
1083
1084 static struct attribute_group cpu_common_attr_group = {
1085         .attrs = cpu_common_attrs,
1086 };
1087
1088 static struct attribute *cpu_online_attrs[] = {
1089         &dev_attr_idle_count.attr,
1090         &dev_attr_idle_time_us.attr,
1091         NULL,
1092 };
1093
1094 static struct attribute_group cpu_online_attr_group = {
1095         .attrs = cpu_online_attrs,
1096 };
1097
1098 static int smp_cpu_online(unsigned int cpu)
1099 {
1100         struct device *s = &per_cpu(cpu_device, cpu)->dev;
1101
1102         return sysfs_create_group(&s->kobj, &cpu_online_attr_group);
1103 }
1104 static int smp_cpu_pre_down(unsigned int cpu)
1105 {
1106         struct device *s = &per_cpu(cpu_device, cpu)->dev;
1107
1108         sysfs_remove_group(&s->kobj, &cpu_online_attr_group);
1109         return 0;
1110 }
1111
1112 static int smp_add_present_cpu(int cpu)
1113 {
1114         struct device *s;
1115         struct cpu *c;
1116         int rc;
1117
1118         c = kzalloc(sizeof(*c), GFP_KERNEL);
1119         if (!c)
1120                 return -ENOMEM;
1121         per_cpu(cpu_device, cpu) = c;
1122         s = &c->dev;
1123         c->hotpluggable = 1;
1124         rc = register_cpu(c, cpu);
1125         if (rc)
1126                 goto out;
1127         rc = sysfs_create_group(&s->kobj, &cpu_common_attr_group);
1128         if (rc)
1129                 goto out_cpu;
1130         rc = topology_cpu_init(c);
1131         if (rc)
1132                 goto out_topology;
1133         return 0;
1134
1135 out_topology:
1136         sysfs_remove_group(&s->kobj, &cpu_common_attr_group);
1137 out_cpu:
1138 #ifdef CONFIG_HOTPLUG_CPU
1139         unregister_cpu(c);
1140 #endif
1141 out:
1142         return rc;
1143 }
1144
1145 #ifdef CONFIG_HOTPLUG_CPU
1146
1147 int __ref smp_rescan_cpus(void)
1148 {
1149         struct sclp_core_info *info;
1150         int nr;
1151
1152         info = kzalloc(sizeof(*info), GFP_KERNEL);
1153         if (!info)
1154                 return -ENOMEM;
1155         smp_get_core_info(info, 0);
1156         get_online_cpus();
1157         mutex_lock(&smp_cpu_state_mutex);
1158         nr = __smp_rescan_cpus(info, 1);
1159         mutex_unlock(&smp_cpu_state_mutex);
1160         put_online_cpus();
1161         kfree(info);
1162         if (nr)
1163                 topology_schedule_update();
1164         return 0;
1165 }
1166
1167 static ssize_t __ref rescan_store(struct device *dev,
1168                                   struct device_attribute *attr,
1169                                   const char *buf,
1170                                   size_t count)
1171 {
1172         int rc;
1173
1174         rc = lock_device_hotplug_sysfs();
1175         if (rc)
1176                 return rc;
1177         rc = smp_rescan_cpus();
1178         unlock_device_hotplug();
1179         return rc ? rc : count;
1180 }
1181 static DEVICE_ATTR_WO(rescan);
1182 #endif /* CONFIG_HOTPLUG_CPU */
1183
1184 static int __init s390_smp_init(void)
1185 {
1186         int cpu, rc = 0;
1187
1188 #ifdef CONFIG_HOTPLUG_CPU
1189         rc = device_create_file(cpu_subsys.dev_root, &dev_attr_rescan);
1190         if (rc)
1191                 return rc;
1192 #endif
1193         for_each_present_cpu(cpu) {
1194                 rc = smp_add_present_cpu(cpu);
1195                 if (rc)
1196                         goto out;
1197         }
1198
1199         rc = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "s390/smp:online",
1200                                smp_cpu_online, smp_cpu_pre_down);
1201         rc = rc <= 0 ? rc : 0;
1202 out:
1203         return rc;
1204 }
1205 subsys_initcall(s390_smp_init);