arch/s390/kernel/smp.c

   1 /*
   2  *  SMP related functions
   3  *
   4  *    Copyright IBM Corp. 1999, 2012
   5  *    Author(s): Denis Joseph Barrow,
   6  *               Martin Schwidefsky <schwidefsky@de.ibm.com>,
   7  *               Heiko Carstens <heiko.carstens@de.ibm.com>,
   8  *
   9  *  based on other smp stuff by
  10  *    (c) 1995 Alan Cox, CymruNET Ltd  <alan@cymru.net>
  11  *    (c) 1998 Ingo Molnar
  12  *
  13  * The code outside of smp.c uses logical cpu numbers, only smp.c does
  14  * the translation of logical to physical cpu ids. All new code that
  15  * operates on physical cpu numbers needs to go into smp.c.
  16  */
  17
  18 #define KMSG_COMPONENT "cpu"
  19 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
  20
  21 #include <linux/workqueue.h>
  22 #include <linux/module.h>
  23 #include <linux/init.h>
  24 #include <linux/mm.h>
  25 #include <linux/err.h>
  26 #include <linux/spinlock.h>
  27 #include <linux/kernel_stat.h>
  28 #include <linux/delay.h>
  29 #include <linux/interrupt.h>
  30 #include <linux/irqflags.h>
  31 #include <linux/cpu.h>
  32 #include <linux/slab.h>
  33 #include <linux/crash_dump.h>
  34 #include <linux/memblock.h>
  35 #include <asm/asm-offsets.h>
  36 #include <asm/diag.h>
  37 #include <asm/switch_to.h>
  38 #include <asm/facility.h>
  39 #include <asm/ipl.h>
  40 #include <asm/setup.h>
  41 #include <asm/irq.h>
  42 #include <asm/tlbflush.h>
  43 #include <asm/vtimer.h>
  44 #include <asm/lowcore.h>
  45 #include <asm/sclp.h>
  46 #include <asm/vdso.h>
  47 #include <asm/debug.h>
  48 #include <asm/os_info.h>
  49 #include <asm/sigp.h>
  50 #include <asm/idle.h>
  51 #include "entry.h"
  52
  53 enum {
  54         ec_schedule = 0,
  55         ec_call_function_single,
  56         ec_stop_cpu,
  57 };
  58
  59 enum {
  60         CPU_STATE_STANDBY,
  61         CPU_STATE_CONFIGURED,
  62 };
  63
  64 static DEFINE_PER_CPU(struct cpu *, cpu_device);
  65
  66 struct pcpu {
  67         struct _lowcore *lowcore;       /* lowcore page(s) for the cpu */
  68         unsigned long ec_mask;          /* bit mask for ec_xxx functions */
  69         signed char state;              /* physical cpu state */
  70         signed char polarization;       /* physical polarization */
  71         u16 address;                    /* physical cpu address */
  72 };
  73
  74 static u8 boot_core_type;
  75 static struct pcpu pcpu_devices[NR_CPUS];
  76
  77 unsigned int smp_cpu_mt_shift;
  78 EXPORT_SYMBOL(smp_cpu_mt_shift);
  79
  80 unsigned int smp_cpu_mtid;
  81 EXPORT_SYMBOL(smp_cpu_mtid);
  82
  83 static unsigned int smp_max_threads __initdata = -1U;
  84
  85 static int __init early_nosmt(char *s)
  86 {
  87         smp_max_threads = 1;
  88         return 0;
  89 }
  90 early_param("nosmt", early_nosmt);
  91
  92 static int __init early_smt(char *s)
  93 {
  94         get_option(&s, &smp_max_threads);
  95         return 0;
  96 }
  97 early_param("smt", early_smt);
  98
  99 /*
 100  * The smp_cpu_state_mutex must be held when changing the state or polarization
 101  * member of a pcpu data structure within the pcpu_devices arreay.
 102  */
 103 DEFINE_MUTEX(smp_cpu_state_mutex);
 104
 105 /*
 106  * Signal processor helper functions.
 107  */
 108 static inline int __pcpu_sigp_relax(u16 addr, u8 order, unsigned long parm,
 109                                     u32 *status)
 110 {
 111         int cc;
 112
 113         while (1) {
 114                 cc = __pcpu_sigp(addr, order, parm, NULL);
 115                 if (cc != SIGP_CC_BUSY)
 116                         return cc;
 117                 cpu_relax();
 118         }
 119 }
 120
 121 static int pcpu_sigp_retry(struct pcpu *pcpu, u8 order, u32 parm)
 122 {
 123         int cc, retry;
 124
 125         for (retry = 0; ; retry++) {
 126                 cc = __pcpu_sigp(pcpu->address, order, parm, NULL);
 127                 if (cc != SIGP_CC_BUSY)
 128                         break;
 129                 if (retry >= 3)
 130                         udelay(10);
 131         }
 132         return cc;
 133 }
 134
 135 static inline int pcpu_stopped(struct pcpu *pcpu)
 136 {
 137         u32 uninitialized_var(status);
 138
 139         if (__pcpu_sigp(pcpu->address, SIGP_SENSE,
 140                         0, &status) != SIGP_CC_STATUS_STORED)
 141                 return 0;
 142         return !!(status & (SIGP_STATUS_CHECK_STOP|SIGP_STATUS_STOPPED));
 143 }
 144
 145 static inline int pcpu_running(struct pcpu *pcpu)
 146 {
 147         if (__pcpu_sigp(pcpu->address, SIGP_SENSE_RUNNING,
 148                         0, NULL) != SIGP_CC_STATUS_STORED)
 149                 return 1;
 150         /* Status stored condition code is equivalent to cpu not running. */
 151         return 0;
 152 }
 153
 154 /*
 155  * Find struct pcpu by cpu address.
 156  */
 157 static struct pcpu *pcpu_find_address(const struct cpumask *mask, u16 address)
 158 {
 159         int cpu;
 160
 161         for_each_cpu(cpu, mask)
 162                 if (pcpu_devices[cpu].address == address)
 163                         return pcpu_devices + cpu;
 164         return NULL;
 165 }
 166
 167 static void pcpu_ec_call(struct pcpu *pcpu, int ec_bit)
 168 {
 169         int order;
 170
 171         if (test_and_set_bit(ec_bit, &pcpu->ec_mask))
 172                 return;
 173         order = pcpu_running(pcpu) ? SIGP_EXTERNAL_CALL : SIGP_EMERGENCY_SIGNAL;
 174         pcpu_sigp_retry(pcpu, order, 0);
 175 }
 176
 177 #define ASYNC_FRAME_OFFSET (ASYNC_SIZE - STACK_FRAME_OVERHEAD - __PT_SIZE)
 178 #define PANIC_FRAME_OFFSET (PAGE_SIZE - STACK_FRAME_OVERHEAD - __PT_SIZE)
 179
 180 static int pcpu_alloc_lowcore(struct pcpu *pcpu, int cpu)
 181 {
 182         unsigned long async_stack, panic_stack;
 183         struct _lowcore *lc;
 184
 185         if (pcpu != &pcpu_devices[0]) {
 186                 pcpu->lowcore = (struct _lowcore *)
 187                         __get_free_pages(GFP_KERNEL | GFP_DMA, LC_ORDER);
 188                 async_stack = __get_free_pages(GFP_KERNEL, ASYNC_ORDER);
 189                 panic_stack = __get_free_page(GFP_KERNEL);
 190                 if (!pcpu->lowcore || !panic_stack || !async_stack)
 191                         goto out;
 192         } else {
 193                 async_stack = pcpu->lowcore->async_stack - ASYNC_FRAME_OFFSET;
 194                 panic_stack = pcpu->lowcore->panic_stack - PANIC_FRAME_OFFSET;
 195         }
 196         lc = pcpu->lowcore;
 197         memcpy(lc, &S390_lowcore, 512);
 198         memset((char *) lc + 512, 0, sizeof(*lc) - 512);
 199         lc->async_stack = async_stack + ASYNC_FRAME_OFFSET;
 200         lc->panic_stack = panic_stack + PANIC_FRAME_OFFSET;
 201         lc->cpu_nr = cpu;
 202         lc->spinlock_lockval = arch_spin_lockval(cpu);
 203         if (MACHINE_HAS_VX)
 204                 lc->vector_save_area_addr =
 205                         (unsigned long) &lc->vector_save_area;
 206         if (vdso_alloc_per_cpu(lc))
 207                 goto out;
 208         lowcore_ptr[cpu] = lc;
 209         pcpu_sigp_retry(pcpu, SIGP_SET_PREFIX, (u32)(unsigned long) lc);
 210         return 0;
 211 out:
 212         if (pcpu != &pcpu_devices[0]) {
 213                 free_page(panic_stack);
 214                 free_pages(async_stack, ASYNC_ORDER);
 215                 free_pages((unsigned long) pcpu->lowcore, LC_ORDER);
 216         }
 217         return -ENOMEM;
 218 }
 219
 220 #ifdef CONFIG_HOTPLUG_CPU
 221
 222 static void pcpu_free_lowcore(struct pcpu *pcpu)
 223 {
 224         pcpu_sigp_retry(pcpu, SIGP_SET_PREFIX, 0);
 225         lowcore_ptr[pcpu - pcpu_devices] = NULL;
 226         vdso_free_per_cpu(pcpu->lowcore);
 227         if (pcpu == &pcpu_devices[0])
 228                 return;
 229         free_page(pcpu->lowcore->panic_stack-PANIC_FRAME_OFFSET);
 230         free_pages(pcpu->lowcore->async_stack-ASYNC_FRAME_OFFSET, ASYNC_ORDER);
 231         free_pages((unsigned long) pcpu->lowcore, LC_ORDER);
 232 }
 233
 234 #endif /* CONFIG_HOTPLUG_CPU */
 235
 236 static void pcpu_prepare_secondary(struct pcpu *pcpu, int cpu)
 237 {
 238         struct _lowcore *lc = pcpu->lowcore;
 239
 240         if (MACHINE_HAS_TLB_LC)
 241                 cpumask_set_cpu(cpu, &init_mm.context.cpu_attach_mask);
 242         cpumask_set_cpu(cpu, mm_cpumask(&init_mm));
 243         atomic_inc(&init_mm.context.attach_count);
 244         lc->cpu_nr = cpu;
 245         lc->spinlock_lockval = arch_spin_lockval(cpu);
 246         lc->percpu_offset = __per_cpu_offset[cpu];
 247         lc->kernel_asce = S390_lowcore.kernel_asce;
 248         lc->machine_flags = S390_lowcore.machine_flags;
 249         lc->user_timer = lc->system_timer = lc->steal_timer = 0;
 250         __ctl_store(lc->cregs_save_area, 0, 15);
 251         save_access_regs((unsigned int *) lc->access_regs_save_area);
 252         memcpy(lc->stfle_fac_list, S390_lowcore.stfle_fac_list,
 253                sizeof(lc->stfle_fac_list));
 254         memcpy(lc->alt_stfle_fac_list, S390_lowcore.alt_stfle_fac_list,
 255                sizeof(lc->alt_stfle_fac_list));
 256 }
 257
 258 static void pcpu_attach_task(struct pcpu *pcpu, struct task_struct *tsk)
 259 {
 260         struct _lowcore *lc = pcpu->lowcore;
 261         struct thread_info *ti = task_thread_info(tsk);
 262
 263         lc->kernel_stack = (unsigned long) task_stack_page(tsk)
 264                 + THREAD_SIZE - STACK_FRAME_OVERHEAD - sizeof(struct pt_regs);
 265         lc->thread_info = (unsigned long) task_thread_info(tsk);
 266         lc->current_task = (unsigned long) tsk;
 267         lc->lpp = LPP_MAGIC;
 268         lc->current_pid = tsk->pid;
 269         lc->user_timer = ti->user_timer;
 270         lc->system_timer = ti->system_timer;
 271         lc->steal_timer = 0;
 272 }
 273
 274 static void pcpu_start_fn(struct pcpu *pcpu, void (*func)(void *), void *data)
 275 {
 276         struct _lowcore *lc = pcpu->lowcore;
 277
 278         lc->restart_stack = lc->kernel_stack;
 279         lc->restart_fn = (unsigned long) func;
 280         lc->restart_data = (unsigned long) data;
 281         lc->restart_source = -1UL;
 282         pcpu_sigp_retry(pcpu, SIGP_RESTART, 0);
 283 }
 284
 285 /*
 286  * Call function via PSW restart on pcpu and stop the current cpu.
 287  */
 288 static void pcpu_delegate(struct pcpu *pcpu, void (*func)(void *),
 289                           void *data, unsigned long stack)
 290 {
 291         struct _lowcore *lc = lowcore_ptr[pcpu - pcpu_devices];
 292         unsigned long source_cpu = stap();
 293
 294         __load_psw_mask(PSW_KERNEL_BITS);
 295         if (pcpu->address == source_cpu)
 296                 func(data);     /* should not return */
 297         /* Stop target cpu (if func returns this stops the current cpu). */
 298         pcpu_sigp_retry(pcpu, SIGP_STOP, 0);
 299         /* Restart func on the target cpu and stop the current cpu. */
 300         mem_assign_absolute(lc->restart_stack, stack);
 301         mem_assign_absolute(lc->restart_fn, (unsigned long) func);
 302         mem_assign_absolute(lc->restart_data, (unsigned long) data);
 303         mem_assign_absolute(lc->restart_source, source_cpu);
 304         asm volatile(
 305                 "0:     sigp    0,%0,%2 # sigp restart to target cpu\n"
 306                 "       brc     2,0b    # busy, try again\n"
 307                 "1:     sigp    0,%1,%3 # sigp stop to current cpu\n"
 308                 "       brc     2,1b    # busy, try again\n"
 309                 : : "d" (pcpu->address), "d" (source_cpu),
 310                     "K" (SIGP_RESTART), "K" (SIGP_STOP)
 311                 : "0", "1", "cc");
 312         for (;;) ;
 313 }
 314
 315 /*
 316  * Enable additional logical cpus for multi-threading.
 317  */
 318 static int pcpu_set_smt(unsigned int mtid)
 319 {
 320         register unsigned long reg1 asm ("1") = (unsigned long) mtid;
 321         int cc;
 322
 323         if (smp_cpu_mtid == mtid)
 324                 return 0;
 325         asm volatile(
 326                 "       sigp    %1,0,%2 # sigp set multi-threading\n"
 327                 "       ipm     %0\n"
 328                 "       srl     %0,28\n"
 329                 : "=d" (cc) : "d" (reg1), "K" (SIGP_SET_MULTI_THREADING)
 330                 : "cc");
 331         if (cc == 0) {
 332                 smp_cpu_mtid = mtid;
 333                 smp_cpu_mt_shift = 0;
 334                 while (smp_cpu_mtid >= (1U << smp_cpu_mt_shift))
 335                         smp_cpu_mt_shift++;
 336                 pcpu_devices[0].address = stap();
 337         }
 338         return cc;
 339 }
 340
 341 /*
 342  * Call function on an online CPU.
 343  */
 344 void smp_call_online_cpu(void (*func)(void *), void *data)
 345 {
 346         struct pcpu *pcpu;
 347
 348         /* Use the current cpu if it is online. */
 349         pcpu = pcpu_find_address(cpu_online_mask, stap());
 350         if (!pcpu)
 351                 /* Use the first online cpu. */
 352                 pcpu = pcpu_devices + cpumask_first(cpu_online_mask);
 353         pcpu_delegate(pcpu, func, data, (unsigned long) restart_stack);
 354 }
 355
 356 /*
 357  * Call function on the ipl CPU.
 358  */
 359 void smp_call_ipl_cpu(void (*func)(void *), void *data)
 360 {
 361         pcpu_delegate(&pcpu_devices[0], func, data,
 362                       pcpu_devices->lowcore->panic_stack -
 363                       PANIC_FRAME_OFFSET + PAGE_SIZE);
 364 }
 365
 366 int smp_find_processor_id(u16 address)
 367 {
 368         int cpu;
 369
 370         for_each_present_cpu(cpu)
 371                 if (pcpu_devices[cpu].address == address)
 372                         return cpu;
 373         return -1;
 374 }
 375
 376 int smp_vcpu_scheduled(int cpu)
 377 {
 378         return pcpu_running(pcpu_devices + cpu);
 379 }
 380
 381 void smp_yield_cpu(int cpu)
 382 {
 383         if (MACHINE_HAS_DIAG9C) {
 384                 diag_stat_inc_norecursion(DIAG_STAT_X09C);
 385                 asm volatile("diag %0,0,0x9c"
 386                              : : "d" (pcpu_devices[cpu].address));
 387         } else if (MACHINE_HAS_DIAG44) {
 388                 diag_stat_inc_norecursion(DIAG_STAT_X044);
 389                 asm volatile("diag 0,0,0x44");
 390         }
 391 }
 392
 393 /*
 394  * Send cpus emergency shutdown signal. This gives the cpus the
 395  * opportunity to complete outstanding interrupts.
 396  */
 397 static void smp_emergency_stop(cpumask_t *cpumask)
 398 {
 399         u64 end;
 400         int cpu;
 401
 402         end = get_tod_clock() + (1000000UL << 12);
 403         for_each_cpu(cpu, cpumask) {
 404                 struct pcpu *pcpu = pcpu_devices + cpu;
 405                 set_bit(ec_stop_cpu, &pcpu->ec_mask);
 406                 while (__pcpu_sigp(pcpu->address, SIGP_EMERGENCY_SIGNAL,
 407                                    0, NULL) == SIGP_CC_BUSY &&
 408                        get_tod_clock() < end)
 409                         cpu_relax();
 410         }
 411         while (get_tod_clock() < end) {
 412                 for_each_cpu(cpu, cpumask)
 413                         if (pcpu_stopped(pcpu_devices + cpu))
 414                                 cpumask_clear_cpu(cpu, cpumask);
 415                 if (cpumask_empty(cpumask))
 416                         break;
 417                 cpu_relax();
 418         }
 419 }
 420
 421 /*
 422  * Stop all cpus but the current one.
 423  */
 424 void smp_send_stop(void)
 425 {
 426         cpumask_t cpumask;
 427         int cpu;
 428
 429         /* Disable all interrupts/machine checks */
 430         __load_psw_mask(PSW_KERNEL_BITS | PSW_MASK_DAT);
 431         trace_hardirqs_off();
 432
 433         debug_set_critical();
 434         cpumask_copy(&cpumask, cpu_online_mask);
 435         cpumask_clear_cpu(smp_processor_id(), &cpumask);
 436
 437         if (oops_in_progress)
 438                 smp_emergency_stop(&cpumask);
 439
 440         /* stop all processors */
 441         for_each_cpu(cpu, &cpumask) {
 442                 struct pcpu *pcpu = pcpu_devices + cpu;
 443                 pcpu_sigp_retry(pcpu, SIGP_STOP, 0);
 444                 while (!pcpu_stopped(pcpu))
 445                         cpu_relax();
 446         }
 447 }
 448
 449 /*
 450  * This is the main routine where commands issued by other
 451  * cpus are handled.
 452  */
 453 static void smp_handle_ext_call(void)
 454 {
 455         unsigned long bits;
 456
 457         /* handle bit signal external calls */
 458         bits = xchg(&pcpu_devices[smp_processor_id()].ec_mask, 0);
 459         if (test_bit(ec_stop_cpu, &bits))
 460                 smp_stop_cpu();
 461         if (test_bit(ec_schedule, &bits))
 462                 scheduler_ipi();
 463         if (test_bit(ec_call_function_single, &bits))
 464                 generic_smp_call_function_single_interrupt();
 465 }
 466
 467 static void do_ext_call_interrupt(struct ext_code ext_code,
 468                                   unsigned int param32, unsigned long param64)
 469 {
 470         inc_irq_stat(ext_code.code == 0x1202 ? IRQEXT_EXC : IRQEXT_EMS);
 471         smp_handle_ext_call();
 472 }
 473
 474 void arch_send_call_function_ipi_mask(const struct cpumask *mask)
 475 {
 476         int cpu;
 477
 478         for_each_cpu(cpu, mask)
 479                 pcpu_ec_call(pcpu_devices + cpu, ec_call_function_single);
 480 }
 481
 482 void arch_send_call_function_single_ipi(int cpu)
 483 {
 484         pcpu_ec_call(pcpu_devices + cpu, ec_call_function_single);
 485 }
 486
 487 /*
 488  * this function sends a 'reschedule' IPI to another CPU.
 489  * it goes straight through and wastes no time serializing
 490  * anything. Worst case is that we lose a reschedule ...
 491  */
 492 void smp_send_reschedule(int cpu)
 493 {
 494         pcpu_ec_call(pcpu_devices + cpu, ec_schedule);
 495 }
 496
 497 /*
 498  * parameter area for the set/clear control bit callbacks
 499  */
 500 struct ec_creg_mask_parms {
 501         unsigned long orval;
 502         unsigned long andval;
 503         int cr;
 504 };
 505
 506 /*
 507  * callback for setting/clearing control bits
 508  */
 509 static void smp_ctl_bit_callback(void *info)
 510 {
 511         struct ec_creg_mask_parms *pp = info;
 512         unsigned long cregs[16];
 513
 514         __ctl_store(cregs, 0, 15);
 515         cregs[pp->cr] = (cregs[pp->cr] & pp->andval) | pp->orval;
 516         __ctl_load(cregs, 0, 15);
 517 }
 518
 519 /*
 520  * Set a bit in a control register of all cpus
 521  */
 522 void smp_ctl_set_bit(int cr, int bit)
 523 {
 524         struct ec_creg_mask_parms parms = { 1UL << bit, -1UL, cr };
 525
 526         on_each_cpu(smp_ctl_bit_callback, &parms, 1);
 527 }
 528 EXPORT_SYMBOL(smp_ctl_set_bit);
 529
 530 /*
 531  * Clear a bit in a control register of all cpus
 532  */
 533 void smp_ctl_clear_bit(int cr, int bit)
 534 {
 535         struct ec_creg_mask_parms parms = { 0, ~(1UL << bit), cr };
 536
 537         on_each_cpu(smp_ctl_bit_callback, &parms, 1);
 538 }
 539 EXPORT_SYMBOL(smp_ctl_clear_bit);
 540
 541 #ifdef CONFIG_CRASH_DUMP
 542
 543 static void __init __smp_store_cpu_state(struct save_area_ext *sa_ext,
 544                                          u16 address, int is_boot_cpu)
 545 {
 546         void *lc = (void *)(unsigned long) store_prefix();
 547         unsigned long vx_sa;
 548
 549         if (is_boot_cpu) {
 550                 /* Copy the registers of the boot CPU. */
 551                 copy_oldmem_page(1, (void *) &sa_ext->sa, sizeof(sa_ext->sa),
 552                                  SAVE_AREA_BASE - PAGE_SIZE, 0);
 553                 if (MACHINE_HAS_VX)
 554                         save_vx_regs_safe(sa_ext->vx_regs);
 555                 return;
 556         }
 557         /* Get the registers of a non-boot cpu. */
 558         __pcpu_sigp_relax(address, SIGP_STOP_AND_STORE_STATUS, 0, NULL);
 559         memcpy_real(&sa_ext->sa, lc + SAVE_AREA_BASE, sizeof(sa_ext->sa));
 560         if (!MACHINE_HAS_VX)
 561                 return;
 562         /* Get the VX registers */
 563         vx_sa = memblock_alloc(PAGE_SIZE, PAGE_SIZE);
 564         if (!vx_sa)
 565                 panic("could not allocate memory for VX save area\n");
 566         __pcpu_sigp_relax(address, SIGP_STORE_ADDITIONAL_STATUS, vx_sa, NULL);
 567         memcpy(sa_ext->vx_regs, (void *) vx_sa, sizeof(sa_ext->vx_regs));
 568         memblock_free(vx_sa, PAGE_SIZE);
 569 }
 570
 571 int smp_store_status(int cpu)
 572 {
 573         unsigned long vx_sa;
 574         struct pcpu *pcpu;
 575
 576         pcpu = pcpu_devices + cpu;
 577         if (__pcpu_sigp_relax(pcpu->address, SIGP_STOP_AND_STORE_STATUS,
 578                               0, NULL) != SIGP_CC_ORDER_CODE_ACCEPTED)
 579                 return -EIO;
 580         if (!MACHINE_HAS_VX)
 581                 return 0;
 582         vx_sa = __pa(pcpu->lowcore->vector_save_area_addr);
 583         __pcpu_sigp_relax(pcpu->address, SIGP_STORE_ADDITIONAL_STATUS,
 584                           vx_sa, NULL);
 585         return 0;
 586 }
 587
 588 #endif /* CONFIG_CRASH_DUMP */
 589
 590 /*
 591  * Collect CPU state of the previous, crashed system.
 592  * There are four cases:
 593  * 1) standard zfcp dump
 594  *    condition: OLDMEM_BASE == NULL && ipl_info.type == IPL_TYPE_FCP_DUMP
 595  *    The state for all CPUs except the boot CPU needs to be collected
 596  *    with sigp stop-and-store-status. The boot CPU state is located in
 597  *    the absolute lowcore of the memory stored in the HSA. The zcore code
 598  *    will allocate the save area and copy the boot CPU state from the HSA.
 599  * 2) stand-alone kdump for SCSI (zfcp dump with swapped memory)
 600  *    condition: OLDMEM_BASE != NULL && ipl_info.type == IPL_TYPE_FCP_DUMP
 601  *    The state for all CPUs except the boot CPU needs to be collected
 602  *    with sigp stop-and-store-status. The firmware or the boot-loader
 603  *    stored the registers of the boot CPU in the absolute lowcore in the
 604  *    memory of the old system.
 605  * 3) kdump and the old kernel did not store the CPU state,
 606  *    or stand-alone kdump for DASD
 607  *    condition: OLDMEM_BASE != NULL && !is_kdump_kernel()
 608  *    The state for all CPUs except the boot CPU needs to be collected
 609  *    with sigp stop-and-store-status. The kexec code or the boot-loader
 610  *    stored the registers of the boot CPU in the memory of the old system.
 611  * 4) kdump and the old kernel stored the CPU state
 612  *    condition: OLDMEM_BASE != NULL && is_kdump_kernel()
 613  *    The state of all CPUs is stored in ELF sections in the memory of the
 614  *    old system. The ELF sections are picked up by the crash_dump code
 615  *    via elfcorehdr_addr.
 616  */
 617 void __init smp_save_dump_cpus(void)
 618 {
 619 #ifdef CONFIG_CRASH_DUMP
 620         int addr, cpu, boot_cpu_addr, max_cpu_addr;
 621         struct save_area_ext *sa_ext;
 622         bool is_boot_cpu;
 623
 624         if (is_kdump_kernel())
 625                 /* Previous system stored the CPU states. Nothing to do. */
 626                 return;
 627         if (!(OLDMEM_BASE || ipl_info.type == IPL_TYPE_FCP_DUMP))
 628                 /* No previous system present, normal boot. */
 629                 return;
 630         /* Set multi-threading state to the previous system. */
 631         pcpu_set_smt(sclp.mtid_prev);
 632         max_cpu_addr = SCLP_MAX_CORES << sclp.mtid_prev;
 633         for (cpu = 0, addr = 0; addr <= max_cpu_addr; addr++) {
 634                 if (__pcpu_sigp_relax(addr, SIGP_SENSE, 0, NULL) ==
 635                     SIGP_CC_NOT_OPERATIONAL)
 636                         continue;
 637                 cpu += 1;
 638         }
 639         dump_save_areas.areas = (void *)memblock_alloc(sizeof(void *) * cpu, 8);
 640         dump_save_areas.count = cpu;
 641         boot_cpu_addr = stap();
 642         for (cpu = 0, addr = 0; addr <= max_cpu_addr; addr++) {
 643                 if (__pcpu_sigp_relax(addr, SIGP_SENSE, 0, NULL) ==
 644                     SIGP_CC_NOT_OPERATIONAL)
 645                         continue;
 646                 sa_ext = (void *) memblock_alloc(sizeof(*sa_ext), 8);
 647                 dump_save_areas.areas[cpu] = sa_ext;
 648                 if (!sa_ext)
 649                         panic("could not allocate memory for save area\n");
 650                 is_boot_cpu = (addr == boot_cpu_addr);
 651                 cpu += 1;
 652                 if (is_boot_cpu && !OLDMEM_BASE)
 653                         /* Skip boot CPU for standard zfcp dump. */
 654                         continue;
 655                 /* Get state for this CPU. */
 656                 __smp_store_cpu_state(sa_ext, addr, is_boot_cpu);
 657         }
 658         diag308_reset();
 659         pcpu_set_smt(0);
 660 #endif /* CONFIG_CRASH_DUMP */
 661 }
 662
 663 void smp_cpu_set_polarization(int cpu, int val)
 664 {
 665         pcpu_devices[cpu].polarization = val;
 666 }
 667
 668 int smp_cpu_get_polarization(int cpu)
 669 {
 670         return pcpu_devices[cpu].polarization;
 671 }
 672
 673 static struct sclp_core_info *smp_get_core_info(void)
 674 {
 675         static int use_sigp_detection;
 676         struct sclp_core_info *info;
 677         int address;
 678
 679         info = kzalloc(sizeof(*info), GFP_KERNEL);
 680         if (info && (use_sigp_detection || sclp_get_core_info(info))) {
 681                 use_sigp_detection = 1;
 682                 for (address = 0;
 683                      address < (SCLP_MAX_CORES << smp_cpu_mt_shift);
 684                      address += (1U << smp_cpu_mt_shift)) {
 685                         if (__pcpu_sigp_relax(address, SIGP_SENSE, 0, NULL) ==
 686                             SIGP_CC_NOT_OPERATIONAL)
 687                                 continue;
 688                         info->core[info->configured].core_id =
 689                                 address >> smp_cpu_mt_shift;
 690                         info->configured++;
 691                 }
 692                 info->combined = info->configured;
 693         }
 694         return info;
 695 }
 696
 697 static int smp_add_present_cpu(int cpu);
 698
 699 static int __smp_rescan_cpus(struct sclp_core_info *info, int sysfs_add)
 700 {
 701         struct pcpu *pcpu;
 702         cpumask_t avail;
 703         int cpu, nr, i, j;
 704         u16 address;
 705
 706         nr = 0;
 707         cpumask_xor(&avail, cpu_possible_mask, cpu_present_mask);
 708         cpu = cpumask_first(&avail);
 709         for (i = 0; (i < info->combined) && (cpu < nr_cpu_ids); i++) {
 710                 if (sclp.has_core_type && info->core[i].type != boot_core_type)
 711                         continue;
 712                 address = info->core[i].core_id << smp_cpu_mt_shift;
 713                 for (j = 0; j <= smp_cpu_mtid; j++) {
 714                         if (pcpu_find_address(cpu_present_mask, address + j))
 715                                 continue;
 716                         pcpu = pcpu_devices + cpu;
 717                         pcpu->address = address + j;
 718                         pcpu->state =
 719                                 (cpu >= info->configured*(smp_cpu_mtid + 1)) ?
 720                                 CPU_STATE_STANDBY : CPU_STATE_CONFIGURED;
 721                         smp_cpu_set_polarization(cpu, POLARIZATION_UNKNOWN);
 722                         set_cpu_present(cpu, true);
 723                         if (sysfs_add && smp_add_present_cpu(cpu) != 0)
 724                                 set_cpu_present(cpu, false);
 725                         else
 726                                 nr++;
 727                         cpu = cpumask_next(cpu, &avail);
 728                         if (cpu >= nr_cpu_ids)
 729                                 break;
 730                 }
 731         }
 732         return nr;
 733 }
 734
 735 static void __init smp_detect_cpus(void)
 736 {
 737         unsigned int cpu, mtid, c_cpus, s_cpus;
 738         struct sclp_core_info *info;
 739         u16 address;
 740
 741         /* Get CPU information */
 742         info = smp_get_core_info();
 743         if (!info)
 744                 panic("smp_detect_cpus failed to allocate memory\n");
 745
 746         /* Find boot CPU type */
 747         if (sclp.has_core_type) {
 748                 address = stap();
 749                 for (cpu = 0; cpu < info->combined; cpu++)
 750                         if (info->core[cpu].core_id == address) {
 751                                 /* The boot cpu dictates the cpu type. */
 752                                 boot_core_type = info->core[cpu].type;
 753                                 break;
 754                         }
 755                 if (cpu >= info->combined)
 756                         panic("Could not find boot CPU type");
 757         }
 758
 759         /* Set multi-threading state for the current system */
 760         mtid = boot_core_type ? sclp.mtid : sclp.mtid_cp;
 761         mtid = (mtid < smp_max_threads) ? mtid : smp_max_threads - 1;
 762         pcpu_set_smt(mtid);
 763
 764         /* Print number of CPUs */
 765         c_cpus = s_cpus = 0;
 766         for (cpu = 0; cpu < info->combined; cpu++) {
 767                 if (sclp.has_core_type &&
 768                     info->core[cpu].type != boot_core_type)
 769                         continue;
 770                 if (cpu < info->configured)
 771                         c_cpus += smp_cpu_mtid + 1;
 772                 else
 773                         s_cpus += smp_cpu_mtid + 1;
 774         }
 775         pr_info("%d configured CPUs, %d standby CPUs\n", c_cpus, s_cpus);
 776
 777         /* Add CPUs present at boot */
 778         get_online_cpus();
 779         __smp_rescan_cpus(info, 0);
 780         put_online_cpus();
 781         kfree(info);
 782 }
 783
 784 /*
 785  *      Activate a secondary processor.
 786  */
 787 static void smp_start_secondary(void *cpuvoid)
 788 {
 789         S390_lowcore.last_update_clock = get_tod_clock();
 790         S390_lowcore.restart_stack = (unsigned long) restart_stack;
 791         S390_lowcore.restart_fn = (unsigned long) do_restart;
 792         S390_lowcore.restart_data = 0;
 793         S390_lowcore.restart_source = -1UL;
 794         restore_access_regs(S390_lowcore.access_regs_save_area);
 795         __ctl_load(S390_lowcore.cregs_save_area, 0, 15);
 796         __load_psw_mask(PSW_KERNEL_BITS | PSW_MASK_DAT);
 797         cpu_init();
 798         preempt_disable();
 799         init_cpu_timer();
 800         vtime_init();
 801         pfault_init();
 802         notify_cpu_starting(smp_processor_id());
 803         set_cpu_online(smp_processor_id(), true);
 804         inc_irq_stat(CPU_RST);
 805         local_irq_enable();
 806         cpu_startup_entry(CPUHP_ONLINE);
 807 }
 808
 809 /* Upping and downing of CPUs */
 810 int __cpu_up(unsigned int cpu, struct task_struct *tidle)
 811 {
 812         struct pcpu *pcpu;
 813         int base, i, rc;
 814
 815         pcpu = pcpu_devices + cpu;
 816         if (pcpu->state != CPU_STATE_CONFIGURED)
 817                 return -EIO;
 818         base = cpu - (cpu % (smp_cpu_mtid + 1));
 819         for (i = 0; i <= smp_cpu_mtid; i++) {
 820                 if (base + i < nr_cpu_ids)
 821                         if (cpu_online(base + i))
 822                                 break;
 823         }
 824         /*
 825          * If this is the first CPU of the core to get online
 826          * do an initial CPU reset.
 827          */
 828         if (i > smp_cpu_mtid &&
 829             pcpu_sigp_retry(pcpu_devices + base, SIGP_INITIAL_CPU_RESET, 0) !=
 830             SIGP_CC_ORDER_CODE_ACCEPTED)
 831                 return -EIO;
 832
 833         rc = pcpu_alloc_lowcore(pcpu, cpu);
 834         if (rc)
 835                 return rc;
 836         pcpu_prepare_secondary(pcpu, cpu);
 837         pcpu_attach_task(pcpu, tidle);
 838         pcpu_start_fn(pcpu, smp_start_secondary, NULL);
 839         /* Wait until cpu puts itself in the online & active maps */
 840         while (!cpu_online(cpu) || !cpu_active(cpu))
 841                 cpu_relax();
 842         return 0;
 843 }
 844
 845 static unsigned int setup_possible_cpus __initdata;
 846
 847 static int __init _setup_possible_cpus(char *s)
 848 {
 849         get_option(&s, &setup_possible_cpus);
 850         return 0;
 851 }
 852 early_param("possible_cpus", _setup_possible_cpus);
 853
 854 #ifdef CONFIG_HOTPLUG_CPU
 855
 856 int __cpu_disable(void)
 857 {
 858         unsigned long cregs[16];
 859
 860         /* Handle possible pending IPIs */
 861         smp_handle_ext_call();
 862         set_cpu_online(smp_processor_id(), false);
 863         /* Disable pseudo page faults on this cpu. */
 864         pfault_fini();
 865         /* Disable interrupt sources via control register. */
 866         __ctl_store(cregs, 0, 15);
 867         cregs[0]  &= ~0x0000ee70UL;     /* disable all external interrupts */
 868         cregs[6]  &= ~0xff000000UL;     /* disable all I/O interrupts */
 869         cregs[14] &= ~0x1f000000UL;     /* disable most machine checks */
 870         __ctl_load(cregs, 0, 15);
 871         clear_cpu_flag(CIF_NOHZ_DELAY);
 872         return 0;
 873 }
 874
 875 void __cpu_die(unsigned int cpu)
 876 {
 877         struct pcpu *pcpu;
 878
 879         /* Wait until target cpu is down */
 880         pcpu = pcpu_devices + cpu;
 881         while (!pcpu_stopped(pcpu))
 882                 cpu_relax();
 883         pcpu_free_lowcore(pcpu);
 884         atomic_dec(&init_mm.context.attach_count);
 885         cpumask_clear_cpu(cpu, mm_cpumask(&init_mm));
 886         if (MACHINE_HAS_TLB_LC)
 887                 cpumask_clear_cpu(cpu, &init_mm.context.cpu_attach_mask);
 888 }
 889
 890 void __noreturn cpu_die(void)
 891 {
 892         idle_task_exit();
 893         pcpu_sigp_retry(pcpu_devices + smp_processor_id(), SIGP_STOP, 0);
 894         for (;;) ;
 895 }
 896
 897 #endif /* CONFIG_HOTPLUG_CPU */
 898
 899 void __init smp_fill_possible_mask(void)
 900 {
 901         unsigned int possible, sclp_max, cpu;
 902
 903         sclp_max = max(sclp.mtid, sclp.mtid_cp) + 1;
 904         sclp_max = min(smp_max_threads, sclp_max);
 905         sclp_max = sclp.max_cores * sclp_max ?: nr_cpu_ids;
 906         possible = setup_possible_cpus ?: nr_cpu_ids;
 907         possible = min(possible, sclp_max);
 908         for (cpu = 0; cpu < possible && cpu < nr_cpu_ids; cpu++)
 909                 set_cpu_possible(cpu, true);
 910 }
 911
 912 void __init smp_prepare_cpus(unsigned int max_cpus)
 913 {
 914         /* request the 0x1201 emergency signal external interrupt */
 915         if (register_external_irq(EXT_IRQ_EMERGENCY_SIG, do_ext_call_interrupt))
 916                 panic("Couldn't request external interrupt 0x1201");
 917         /* request the 0x1202 external call external interrupt */
 918         if (register_external_irq(EXT_IRQ_EXTERNAL_CALL, do_ext_call_interrupt))
 919                 panic("Couldn't request external interrupt 0x1202");
 920         smp_detect_cpus();
 921 }
 922
 923 void __init smp_prepare_boot_cpu(void)
 924 {
 925         struct pcpu *pcpu = pcpu_devices;
 926
 927         pcpu->state = CPU_STATE_CONFIGURED;
 928         pcpu->address = stap();
 929         pcpu->lowcore = (struct _lowcore *)(unsigned long) store_prefix();
 930         S390_lowcore.percpu_offset = __per_cpu_offset[0];
 931         smp_cpu_set_polarization(0, POLARIZATION_UNKNOWN);
 932         set_cpu_present(0, true);
 933         set_cpu_online(0, true);
 934 }
 935
 936 void __init smp_cpus_done(unsigned int max_cpus)
 937 {
 938 }
 939
 940 void __init smp_setup_processor_id(void)
 941 {
 942         S390_lowcore.cpu_nr = 0;
 943         S390_lowcore.spinlock_lockval = arch_spin_lockval(0);
 944 }
 945
 946 /*
 947  * the frequency of the profiling timer can be changed
 948  * by writing a multiplier value into /proc/profile.
 949  *
 950  * usually you want to run this on all CPUs ;)
 951  */
 952 int setup_profiling_timer(unsigned int multiplier)
 953 {
 954         return 0;
 955 }
 956
 957 #ifdef CONFIG_HOTPLUG_CPU
 958 static ssize_t cpu_configure_show(struct device *dev,
 959                                   struct device_attribute *attr, char *buf)
 960 {
 961         ssize_t count;
 962
 963         mutex_lock(&smp_cpu_state_mutex);
 964         count = sprintf(buf, "%d\n", pcpu_devices[dev->id].state);
 965         mutex_unlock(&smp_cpu_state_mutex);
 966         return count;
 967 }
 968
 969 static ssize_t cpu_configure_store(struct device *dev,
 970                                    struct device_attribute *attr,
 971                                    const char *buf, size_t count)
 972 {
 973         struct pcpu *pcpu;
 974         int cpu, val, rc, i;
 975         char delim;
 976
 977         if (sscanf(buf, "%d %c", &val, &delim) != 1)
 978                 return -EINVAL;
 979         if (val != 0 && val != 1)
 980                 return -EINVAL;
 981         get_online_cpus();
 982         mutex_lock(&smp_cpu_state_mutex);
 983         rc = -EBUSY;
 984         /* disallow configuration changes of online cpus and cpu 0 */
 985         cpu = dev->id;
 986         cpu -= cpu % (smp_cpu_mtid + 1);
 987         if (cpu == 0)
 988                 goto out;
 989         for (i = 0; i <= smp_cpu_mtid; i++)
 990                 if (cpu_online(cpu + i))
 991                         goto out;
 992         pcpu = pcpu_devices + cpu;
 993         rc = 0;
 994         switch (val) {
 995         case 0:
 996                 if (pcpu->state != CPU_STATE_CONFIGURED)
 997                         break;
 998                 rc = sclp_core_deconfigure(pcpu->address >> smp_cpu_mt_shift);
 999                 if (rc)
1000                         break;
1001                 for (i = 0; i <= smp_cpu_mtid; i++) {
1002                         if (cpu + i >= nr_cpu_ids || !cpu_present(cpu + i))
1003                                 continue;
1004                         pcpu[i].state = CPU_STATE_STANDBY;
1005                         smp_cpu_set_polarization(cpu + i,
1006                                                  POLARIZATION_UNKNOWN);
1007                 }
1008                 topology_expect_change();
1009                 break;
1010         case 1:
1011                 if (pcpu->state != CPU_STATE_STANDBY)
1012                         break;
1013                 rc = sclp_core_configure(pcpu->address >> smp_cpu_mt_shift);
1014                 if (rc)
1015                         break;
1016                 for (i = 0; i <= smp_cpu_mtid; i++) {
1017                         if (cpu + i >= nr_cpu_ids || !cpu_present(cpu + i))
1018                                 continue;
1019                         pcpu[i].state = CPU_STATE_CONFIGURED;
1020                         smp_cpu_set_polarization(cpu + i,
1021                                                  POLARIZATION_UNKNOWN);
1022                 }
1023                 topology_expect_change();
1024                 break;
1025         default:
1026                 break;
1027         }
1028 out:
1029         mutex_unlock(&smp_cpu_state_mutex);
1030         put_online_cpus();
1031         return rc ? rc : count;
1032 }
1033 static DEVICE_ATTR(configure, 0644, cpu_configure_show, cpu_configure_store);
1034 #endif /* CONFIG_HOTPLUG_CPU */
1035
1036 static ssize_t show_cpu_address(struct device *dev,
1037                                 struct device_attribute *attr, char *buf)
1038 {
1039         return sprintf(buf, "%d\n", pcpu_devices[dev->id].address);
1040 }
1041 static DEVICE_ATTR(address, 0444, show_cpu_address, NULL);
1042
1043 static struct attribute *cpu_common_attrs[] = {
1044 #ifdef CONFIG_HOTPLUG_CPU
1045         &dev_attr_configure.attr,
1046 #endif
1047         &dev_attr_address.attr,
1048         NULL,
1049 };
1050
1051 static struct attribute_group cpu_common_attr_group = {
1052         .attrs = cpu_common_attrs,
1053 };
1054
1055 static struct attribute *cpu_online_attrs[] = {
1056         &dev_attr_idle_count.attr,
1057         &dev_attr_idle_time_us.attr,
1058         NULL,
1059 };
1060
1061 static struct attribute_group cpu_online_attr_group = {
1062         .attrs = cpu_online_attrs,
1063 };
1064
1065 static int smp_cpu_notify(struct notifier_block *self, unsigned long action,
1066                           void *hcpu)
1067 {
1068         unsigned int cpu = (unsigned int)(long)hcpu;
1069         struct device *s = &per_cpu(cpu_device, cpu)->dev;
1070         int err = 0;
1071
1072         switch (action & ~CPU_TASKS_FROZEN) {
1073         case CPU_ONLINE:
1074                 err = sysfs_create_group(&s->kobj, &cpu_online_attr_group);
1075                 break;
1076         case CPU_DEAD:
1077                 sysfs_remove_group(&s->kobj, &cpu_online_attr_group);
1078                 break;
1079         }
1080         return notifier_from_errno(err);
1081 }
1082
1083 static int smp_add_present_cpu(int cpu)
1084 {
1085         struct device *s;
1086         struct cpu *c;
1087         int rc;
1088
1089         c = kzalloc(sizeof(*c), GFP_KERNEL);
1090         if (!c)
1091                 return -ENOMEM;
1092         per_cpu(cpu_device, cpu) = c;
1093         s = &c->dev;
1094         c->hotpluggable = 1;
1095         rc = register_cpu(c, cpu);
1096         if (rc)
1097                 goto out;
1098         rc = sysfs_create_group(&s->kobj, &cpu_common_attr_group);
1099         if (rc)
1100                 goto out_cpu;
1101         if (cpu_online(cpu)) {
1102                 rc = sysfs_create_group(&s->kobj, &cpu_online_attr_group);
1103                 if (rc)
1104                         goto out_online;
1105         }
1106         rc = topology_cpu_init(c);
1107         if (rc)
1108                 goto out_topology;
1109         return 0;
1110
1111 out_topology:
1112         if (cpu_online(cpu))
1113                 sysfs_remove_group(&s->kobj, &cpu_online_attr_group);
1114 out_online:
1115         sysfs_remove_group(&s->kobj, &cpu_common_attr_group);
1116 out_cpu:
1117 #ifdef CONFIG_HOTPLUG_CPU
1118         unregister_cpu(c);
1119 #endif
1120 out:
1121         return rc;
1122 }
1123
1124 #ifdef CONFIG_HOTPLUG_CPU
1125
1126 int __ref smp_rescan_cpus(void)
1127 {
1128         struct sclp_core_info *info;
1129         int nr;
1130
1131         info = smp_get_core_info();
1132         if (!info)
1133                 return -ENOMEM;
1134         get_online_cpus();
1135         mutex_lock(&smp_cpu_state_mutex);
1136         nr = __smp_rescan_cpus(info, 1);
1137         mutex_unlock(&smp_cpu_state_mutex);
1138         put_online_cpus();
1139         kfree(info);
1140         if (nr)
1141                 topology_schedule_update();
1142         return 0;
1143 }
1144
1145 static ssize_t __ref rescan_store(struct device *dev,
1146                                   struct device_attribute *attr,
1147                                   const char *buf,
1148                                   size_t count)
1149 {
1150         int rc;
1151
1152         rc = smp_rescan_cpus();
1153         return rc ? rc : count;
1154 }
1155 static DEVICE_ATTR(rescan, 0200, NULL, rescan_store);
1156 #endif /* CONFIG_HOTPLUG_CPU */
1157
1158 static int __init s390_smp_init(void)
1159 {
1160         int cpu, rc = 0;
1161
1162 #ifdef CONFIG_HOTPLUG_CPU
1163         rc = device_create_file(cpu_subsys.dev_root, &dev_attr_rescan);
1164         if (rc)
1165                 return rc;
1166 #endif
1167         cpu_notifier_register_begin();
1168         for_each_present_cpu(cpu) {
1169                 rc = smp_add_present_cpu(cpu);
1170                 if (rc)
1171                         goto out;
1172         }
1173
1174         __hotcpu_notifier(smp_cpu_notify, 0);
1175
1176 out:
1177         cpu_notifier_register_done();
1178         return rc;
1179 }
1180 subsys_initcall(s390_smp_init);