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         unsigned long ec_clk;           /* sigp timestamp for ec_xxx */
  70         signed char state;              /* physical cpu state */
  71         signed char polarization;       /* physical polarization */
  72         u16 address;                    /* physical cpu address */
  73 };
  74
  75 static u8 boot_core_type;
  76 static struct pcpu pcpu_devices[NR_CPUS];
  77
  78 unsigned int smp_cpu_mt_shift;
  79 EXPORT_SYMBOL(smp_cpu_mt_shift);
  80
  81 unsigned int smp_cpu_mtid;
  82 EXPORT_SYMBOL(smp_cpu_mtid);
  83
  84 #ifdef CONFIG_CRASH_DUMP
  85 __vector128 __initdata boot_cpu_vector_save_area[__NUM_VXRS];
  86 #endif
  87
  88 static unsigned int smp_max_threads __initdata = -1U;
  89
  90 static int __init early_nosmt(char *s)
  91 {
  92         smp_max_threads = 1;
  93         return 0;
  94 }
  95 early_param("nosmt", early_nosmt);
  96
  97 static int __init early_smt(char *s)
  98 {
  99         get_option(&s, &smp_max_threads);
 100         return 0;
 101 }
 102 early_param("smt", early_smt);
 103
 104 /*
 105  * The smp_cpu_state_mutex must be held when changing the state or polarization
 106  * member of a pcpu data structure within the pcpu_devices arreay.
 107  */
 108 DEFINE_MUTEX(smp_cpu_state_mutex);
 109
 110 /*
 111  * Signal processor helper functions.
 112  */
 113 static inline int __pcpu_sigp_relax(u16 addr, u8 order, unsigned long parm)
 114 {
 115         int cc;
 116
 117         while (1) {
 118                 cc = __pcpu_sigp(addr, order, parm, NULL);
 119                 if (cc != SIGP_CC_BUSY)
 120                         return cc;
 121                 cpu_relax();
 122         }
 123 }
 124
 125 static int pcpu_sigp_retry(struct pcpu *pcpu, u8 order, u32 parm)
 126 {
 127         int cc, retry;
 128
 129         for (retry = 0; ; retry++) {
 130                 cc = __pcpu_sigp(pcpu->address, order, parm, NULL);
 131                 if (cc != SIGP_CC_BUSY)
 132                         break;
 133                 if (retry >= 3)
 134                         udelay(10);
 135         }
 136         return cc;
 137 }
 138
 139 static inline int pcpu_stopped(struct pcpu *pcpu)
 140 {
 141         u32 uninitialized_var(status);
 142
 143         if (__pcpu_sigp(pcpu->address, SIGP_SENSE,
 144                         0, &status) != SIGP_CC_STATUS_STORED)
 145                 return 0;
 146         return !!(status & (SIGP_STATUS_CHECK_STOP|SIGP_STATUS_STOPPED));
 147 }
 148
 149 static inline int pcpu_running(struct pcpu *pcpu)
 150 {
 151         if (__pcpu_sigp(pcpu->address, SIGP_SENSE_RUNNING,
 152                         0, NULL) != SIGP_CC_STATUS_STORED)
 153                 return 1;
 154         /* Status stored condition code is equivalent to cpu not running. */
 155         return 0;
 156 }
 157
 158 /*
 159  * Find struct pcpu by cpu address.
 160  */
 161 static struct pcpu *pcpu_find_address(const struct cpumask *mask, u16 address)
 162 {
 163         int cpu;
 164
 165         for_each_cpu(cpu, mask)
 166                 if (pcpu_devices[cpu].address == address)
 167                         return pcpu_devices + cpu;
 168         return NULL;
 169 }
 170
 171 static void pcpu_ec_call(struct pcpu *pcpu, int ec_bit)
 172 {
 173         int order;
 174
 175         if (test_and_set_bit(ec_bit, &pcpu->ec_mask))
 176                 return;
 177         order = pcpu_running(pcpu) ? SIGP_EXTERNAL_CALL : SIGP_EMERGENCY_SIGNAL;
 178         pcpu->ec_clk = get_tod_clock_fast();
 179         pcpu_sigp_retry(pcpu, order, 0);
 180 }
 181
 182 #define ASYNC_FRAME_OFFSET (ASYNC_SIZE - STACK_FRAME_OVERHEAD - __PT_SIZE)
 183 #define PANIC_FRAME_OFFSET (PAGE_SIZE - STACK_FRAME_OVERHEAD - __PT_SIZE)
 184
 185 static int pcpu_alloc_lowcore(struct pcpu *pcpu, int cpu)
 186 {
 187         unsigned long async_stack, panic_stack;
 188         struct lowcore *lc;
 189
 190         if (pcpu != &pcpu_devices[0]) {
 191                 pcpu->lowcore = (struct lowcore *)
 192                         __get_free_pages(GFP_KERNEL | GFP_DMA, LC_ORDER);
 193                 async_stack = __get_free_pages(GFP_KERNEL, ASYNC_ORDER);
 194                 panic_stack = __get_free_page(GFP_KERNEL);
 195                 if (!pcpu->lowcore || !panic_stack || !async_stack)
 196                         goto out;
 197         } else {
 198                 async_stack = pcpu->lowcore->async_stack - ASYNC_FRAME_OFFSET;
 199                 panic_stack = pcpu->lowcore->panic_stack - PANIC_FRAME_OFFSET;
 200         }
 201         lc = pcpu->lowcore;
 202         memcpy(lc, &S390_lowcore, 512);
 203         memset((char *) lc + 512, 0, sizeof(*lc) - 512);
 204         lc->async_stack = async_stack + ASYNC_FRAME_OFFSET;
 205         lc->panic_stack = panic_stack + PANIC_FRAME_OFFSET;
 206         lc->cpu_nr = cpu;
 207         lc->spinlock_lockval = arch_spin_lockval(cpu);
 208         lc->br_r1_trampoline = 0x07f1;  /* br %r1 */
 209         if (MACHINE_HAS_VX)
 210                 lc->vector_save_area_addr =
 211                         (unsigned long) &lc->vector_save_area;
 212         if (vdso_alloc_per_cpu(lc))
 213                 goto out;
 214         lowcore_ptr[cpu] = lc;
 215         pcpu_sigp_retry(pcpu, SIGP_SET_PREFIX, (u32)(unsigned long) lc);
 216         return 0;
 217 out:
 218         if (pcpu != &pcpu_devices[0]) {
 219                 free_page(panic_stack);
 220                 free_pages(async_stack, ASYNC_ORDER);
 221                 free_pages((unsigned long) pcpu->lowcore, LC_ORDER);
 222         }
 223         return -ENOMEM;
 224 }
 225
 226 #ifdef CONFIG_HOTPLUG_CPU
 227
 228 static void pcpu_free_lowcore(struct pcpu *pcpu)
 229 {
 230         pcpu_sigp_retry(pcpu, SIGP_SET_PREFIX, 0);
 231         lowcore_ptr[pcpu - pcpu_devices] = NULL;
 232         vdso_free_per_cpu(pcpu->lowcore);
 233         if (pcpu == &pcpu_devices[0])
 234                 return;
 235         free_page(pcpu->lowcore->panic_stack-PANIC_FRAME_OFFSET);
 236         free_pages(pcpu->lowcore->async_stack-ASYNC_FRAME_OFFSET, ASYNC_ORDER);
 237         free_pages((unsigned long) pcpu->lowcore, LC_ORDER);
 238 }
 239
 240 #endif /* CONFIG_HOTPLUG_CPU */
 241
 242 static void pcpu_prepare_secondary(struct pcpu *pcpu, int cpu)
 243 {
 244         struct lowcore *lc = pcpu->lowcore;
 245
 246         cpumask_set_cpu(cpu, &init_mm.context.cpu_attach_mask);
 247         cpumask_set_cpu(cpu, mm_cpumask(&init_mm));
 248         lc->cpu_nr = cpu;
 249         lc->spinlock_lockval = arch_spin_lockval(cpu);
 250         lc->percpu_offset = __per_cpu_offset[cpu];
 251         lc->kernel_asce = S390_lowcore.kernel_asce;
 252         lc->machine_flags = S390_lowcore.machine_flags;
 253         lc->user_timer = lc->system_timer = lc->steal_timer = 0;
 254         __ctl_store(lc->cregs_save_area, 0, 15);
 255         save_access_regs((unsigned int *) lc->access_regs_save_area);
 256         memcpy(lc->stfle_fac_list, S390_lowcore.stfle_fac_list,
 257                sizeof(lc->stfle_fac_list));
 258         memcpy(lc->alt_stfle_fac_list, S390_lowcore.alt_stfle_fac_list,
 259                sizeof(lc->alt_stfle_fac_list));
 260 }
 261
 262 static void pcpu_attach_task(struct pcpu *pcpu, struct task_struct *tsk)
 263 {
 264         struct lowcore *lc = pcpu->lowcore;
 265         struct thread_info *ti = task_thread_info(tsk);
 266
 267         lc->kernel_stack = (unsigned long) task_stack_page(tsk)
 268                 + THREAD_SIZE - STACK_FRAME_OVERHEAD - sizeof(struct pt_regs);
 269         lc->thread_info = (unsigned long) task_thread_info(tsk);
 270         lc->current_task = (unsigned long) tsk;
 271         lc->lpp = LPP_MAGIC;
 272         lc->current_pid = tsk->pid;
 273         lc->user_timer = ti->user_timer;
 274         lc->system_timer = ti->system_timer;
 275         lc->steal_timer = 0;
 276 }
 277
 278 static void pcpu_start_fn(struct pcpu *pcpu, void (*func)(void *), void *data)
 279 {
 280         struct lowcore *lc = pcpu->lowcore;
 281
 282         lc->restart_stack = lc->kernel_stack;
 283         lc->restart_fn = (unsigned long) func;
 284         lc->restart_data = (unsigned long) data;
 285         lc->restart_source = -1UL;
 286         pcpu_sigp_retry(pcpu, SIGP_RESTART, 0);
 287 }
 288
 289 /*
 290  * Call function via PSW restart on pcpu and stop the current cpu.
 291  */
 292 static void pcpu_delegate(struct pcpu *pcpu, void (*func)(void *),
 293                           void *data, unsigned long stack)
 294 {
 295         struct lowcore *lc = lowcore_ptr[pcpu - pcpu_devices];
 296         unsigned long source_cpu = stap();
 297
 298         __load_psw_mask(PSW_KERNEL_BITS);
 299         if (pcpu->address == source_cpu)
 300                 func(data);     /* should not return */
 301         /* Stop target cpu (if func returns this stops the current cpu). */
 302         pcpu_sigp_retry(pcpu, SIGP_STOP, 0);
 303         /* Restart func on the target cpu and stop the current cpu. */
 304         mem_assign_absolute(lc->restart_stack, stack);
 305         mem_assign_absolute(lc->restart_fn, (unsigned long) func);
 306         mem_assign_absolute(lc->restart_data, (unsigned long) data);
 307         mem_assign_absolute(lc->restart_source, source_cpu);
 308         __bpon();
 309         asm volatile(
 310                 "0:     sigp    0,%0,%2 # sigp restart to target cpu\n"
 311                 "       brc     2,0b    # busy, try again\n"
 312                 "1:     sigp    0,%1,%3 # sigp stop to current cpu\n"
 313                 "       brc     2,1b    # busy, try again\n"
 314                 : : "d" (pcpu->address), "d" (source_cpu),
 315                     "K" (SIGP_RESTART), "K" (SIGP_STOP)
 316                 : "0", "1", "cc");
 317         for (;;) ;
 318 }
 319
 320 /*
 321  * Enable additional logical cpus for multi-threading.
 322  */
 323 static int pcpu_set_smt(unsigned int mtid)
 324 {
 325         int cc;
 326
 327         if (smp_cpu_mtid == mtid)
 328                 return 0;
 329         cc = __pcpu_sigp(0, SIGP_SET_MULTI_THREADING, mtid, NULL);
 330         if (cc == 0) {
 331                 smp_cpu_mtid = mtid;
 332                 smp_cpu_mt_shift = 0;
 333                 while (smp_cpu_mtid >= (1U << smp_cpu_mt_shift))
 334                         smp_cpu_mt_shift++;
 335                 pcpu_devices[0].address = stap();
 336         }
 337         return cc;
 338 }
 339
 340 /*
 341  * Call function on an online CPU.
 342  */
 343 void smp_call_online_cpu(void (*func)(void *), void *data)
 344 {
 345         struct pcpu *pcpu;
 346
 347         /* Use the current cpu if it is online. */
 348         pcpu = pcpu_find_address(cpu_online_mask, stap());
 349         if (!pcpu)
 350                 /* Use the first online cpu. */
 351                 pcpu = pcpu_devices + cpumask_first(cpu_online_mask);
 352         pcpu_delegate(pcpu, func, data, (unsigned long) restart_stack);
 353 }
 354
 355 /*
 356  * Call function on the ipl CPU.
 357  */
 358 void smp_call_ipl_cpu(void (*func)(void *), void *data)
 359 {
 360         struct lowcore *lc = pcpu_devices->lowcore;
 361
 362         if (pcpu_devices[0].address == stap())
 363                 lc = &S390_lowcore;
 364
 365         pcpu_delegate(&pcpu_devices[0], func, data,
 366                       lc->panic_stack - PANIC_FRAME_OFFSET + PAGE_SIZE);
 367 }
 368
 369 int smp_find_processor_id(u16 address)
 370 {
 371         int cpu;
 372
 373         for_each_present_cpu(cpu)
 374                 if (pcpu_devices[cpu].address == address)
 375                         return cpu;
 376         return -1;
 377 }
 378
 379 int smp_vcpu_scheduled(int cpu)
 380 {
 381         return pcpu_running(pcpu_devices + cpu);
 382 }
 383
 384 void smp_yield_cpu(int cpu)
 385 {
 386         if (MACHINE_HAS_DIAG9C) {
 387                 diag_stat_inc_norecursion(DIAG_STAT_X09C);
 388                 asm volatile("diag %0,0,0x9c"
 389                              : : "d" (pcpu_devices[cpu].address));
 390         } else if (MACHINE_HAS_DIAG44) {
 391                 diag_stat_inc_norecursion(DIAG_STAT_X044);
 392                 asm volatile("diag 0,0,0x44");
 393         }
 394 }
 395
 396 /*
 397  * Send cpus emergency shutdown signal. This gives the cpus the
 398  * opportunity to complete outstanding interrupts.
 399  */
 400 static void smp_emergency_stop(cpumask_t *cpumask)
 401 {
 402         u64 end;
 403         int cpu;
 404
 405         end = get_tod_clock() + (1000000UL << 12);
 406         for_each_cpu(cpu, cpumask) {
 407                 struct pcpu *pcpu = pcpu_devices + cpu;
 408                 set_bit(ec_stop_cpu, &pcpu->ec_mask);
 409                 while (__pcpu_sigp(pcpu->address, SIGP_EMERGENCY_SIGNAL,
 410                                    0, NULL) == SIGP_CC_BUSY &&
 411                        get_tod_clock() < end)
 412                         cpu_relax();
 413         }
 414         while (get_tod_clock() < end) {
 415                 for_each_cpu(cpu, cpumask)
 416                         if (pcpu_stopped(pcpu_devices + cpu))
 417                                 cpumask_clear_cpu(cpu, cpumask);
 418                 if (cpumask_empty(cpumask))
 419                         break;
 420                 cpu_relax();
 421         }
 422 }
 423
 424 /*
 425  * Stop all cpus but the current one.
 426  */
 427 void smp_send_stop(void)
 428 {
 429         cpumask_t cpumask;
 430         int cpu;
 431
 432         /* Disable all interrupts/machine checks */
 433         __load_psw_mask(PSW_KERNEL_BITS | PSW_MASK_DAT);
 434         trace_hardirqs_off();
 435
 436         debug_set_critical();
 437         cpumask_copy(&cpumask, cpu_online_mask);
 438         cpumask_clear_cpu(smp_processor_id(), &cpumask);
 439
 440         if (oops_in_progress)
 441                 smp_emergency_stop(&cpumask);
 442
 443         /* stop all processors */
 444         for_each_cpu(cpu, &cpumask) {
 445                 struct pcpu *pcpu = pcpu_devices + cpu;
 446                 pcpu_sigp_retry(pcpu, SIGP_STOP, 0);
 447                 while (!pcpu_stopped(pcpu))
 448                         cpu_relax();
 449         }
 450 }
 451
 452 /*
 453  * This is the main routine where commands issued by other
 454  * cpus are handled.
 455  */
 456 static void smp_handle_ext_call(void)
 457 {
 458         unsigned long bits;
 459
 460         /* handle bit signal external calls */
 461         bits = xchg(&pcpu_devices[smp_processor_id()].ec_mask, 0);
 462         if (test_bit(ec_stop_cpu, &bits))
 463                 smp_stop_cpu();
 464         if (test_bit(ec_schedule, &bits))
 465                 scheduler_ipi();
 466         if (test_bit(ec_call_function_single, &bits))
 467                 generic_smp_call_function_single_interrupt();
 468 }
 469
 470 static void do_ext_call_interrupt(struct ext_code ext_code,
 471                                   unsigned int param32, unsigned long param64)
 472 {
 473         inc_irq_stat(ext_code.code == 0x1202 ? IRQEXT_EXC : IRQEXT_EMS);
 474         smp_handle_ext_call();
 475 }
 476
 477 void arch_send_call_function_ipi_mask(const struct cpumask *mask)
 478 {
 479         int cpu;
 480
 481         for_each_cpu(cpu, mask)
 482                 pcpu_ec_call(pcpu_devices + cpu, ec_call_function_single);
 483 }
 484
 485 void arch_send_call_function_single_ipi(int cpu)
 486 {
 487         pcpu_ec_call(pcpu_devices + cpu, ec_call_function_single);
 488 }
 489
 490 /*
 491  * this function sends a 'reschedule' IPI to another CPU.
 492  * it goes straight through and wastes no time serializing
 493  * anything. Worst case is that we lose a reschedule ...
 494  */
 495 void smp_send_reschedule(int cpu)
 496 {
 497         pcpu_ec_call(pcpu_devices + cpu, ec_schedule);
 498 }
 499
 500 /*
 501  * parameter area for the set/clear control bit callbacks
 502  */
 503 struct ec_creg_mask_parms {
 504         unsigned long orval;
 505         unsigned long andval;
 506         int cr;
 507 };
 508
 509 /*
 510  * callback for setting/clearing control bits
 511  */
 512 static void smp_ctl_bit_callback(void *info)
 513 {
 514         struct ec_creg_mask_parms *pp = info;
 515         unsigned long cregs[16];
 516
 517         __ctl_store(cregs, 0, 15);
 518         cregs[pp->cr] = (cregs[pp->cr] & pp->andval) | pp->orval;
 519         __ctl_load(cregs, 0, 15);
 520 }
 521
 522 /*
 523  * Set a bit in a control register of all cpus
 524  */
 525 void smp_ctl_set_bit(int cr, int bit)
 526 {
 527         struct ec_creg_mask_parms parms = { 1UL << bit, -1UL, cr };
 528
 529         on_each_cpu(smp_ctl_bit_callback, &parms, 1);
 530 }
 531 EXPORT_SYMBOL(smp_ctl_set_bit);
 532
 533 /*
 534  * Clear a bit in a control register of all cpus
 535  */
 536 void smp_ctl_clear_bit(int cr, int bit)
 537 {
 538         struct ec_creg_mask_parms parms = { 0, ~(1UL << bit), cr };
 539
 540         on_each_cpu(smp_ctl_bit_callback, &parms, 1);
 541 }
 542 EXPORT_SYMBOL(smp_ctl_clear_bit);
 543
 544 #ifdef CONFIG_CRASH_DUMP
 545
 546 int smp_store_status(int cpu)
 547 {
 548         struct pcpu *pcpu = pcpu_devices + cpu;
 549         unsigned long pa;
 550
 551         pa = __pa(&pcpu->lowcore->floating_pt_save_area);
 552         if (__pcpu_sigp_relax(pcpu->address, SIGP_STORE_STATUS_AT_ADDRESS,
 553                               pa) != SIGP_CC_ORDER_CODE_ACCEPTED)
 554                 return -EIO;
 555         if (!MACHINE_HAS_VX)
 556                 return 0;
 557         pa = __pa(pcpu->lowcore->vector_save_area_addr);
 558         if (__pcpu_sigp_relax(pcpu->address, SIGP_STORE_ADDITIONAL_STATUS,
 559                               pa) != SIGP_CC_ORDER_CODE_ACCEPTED)
 560                 return -EIO;
 561         return 0;
 562 }
 563
 564 /*
 565  * Collect CPU state of the previous, crashed system.
 566  * There are four cases:
 567  * 1) standard zfcp dump
 568  *    condition: OLDMEM_BASE == NULL && ipl_info.type == IPL_TYPE_FCP_DUMP
 569  *    The state for all CPUs except the boot CPU needs to be collected
 570  *    with sigp stop-and-store-status. The boot CPU state is located in
 571  *    the absolute lowcore of the memory stored in the HSA. The zcore code
 572  *    will copy the boot CPU state from the HSA.
 573  * 2) stand-alone kdump for SCSI (zfcp dump with swapped memory)
 574  *    condition: OLDMEM_BASE != NULL && ipl_info.type == IPL_TYPE_FCP_DUMP
 575  *    The state for all CPUs except the boot CPU needs to be collected
 576  *    with sigp stop-and-store-status. The firmware or the boot-loader
 577  *    stored the registers of the boot CPU in the absolute lowcore in the
 578  *    memory of the old system.
 579  * 3) kdump and the old kernel did not store the CPU state,
 580  *    or stand-alone kdump for DASD
 581  *    condition: OLDMEM_BASE != NULL && !is_kdump_kernel()
 582  *    The state for all CPUs except the boot CPU needs to be collected
 583  *    with sigp stop-and-store-status. The kexec code or the boot-loader
 584  *    stored the registers of the boot CPU in the memory of the old system.
 585  * 4) kdump and the old kernel stored the CPU state
 586  *    condition: OLDMEM_BASE != NULL && is_kdump_kernel()
 587  *    This case does not exist for s390 anymore, setup_arch explicitly
 588  *    deactivates the elfcorehdr= kernel parameter
 589  */
 590 static __init void smp_save_cpu_vxrs(struct save_area *sa, u16 addr,
 591                                      bool is_boot_cpu, unsigned long page)
 592 {
 593         __vector128 *vxrs = (__vector128 *) page;
 594
 595         if (is_boot_cpu)
 596                 vxrs = boot_cpu_vector_save_area;
 597         else
 598                 __pcpu_sigp_relax(addr, SIGP_STORE_ADDITIONAL_STATUS, page);
 599         save_area_add_vxrs(sa, vxrs);
 600 }
 601
 602 static __init void smp_save_cpu_regs(struct save_area *sa, u16 addr,
 603                                      bool is_boot_cpu, unsigned long page)
 604 {
 605         void *regs = (void *) page;
 606
 607         if (is_boot_cpu)
 608                 copy_oldmem_kernel(regs, (void *) __LC_FPREGS_SAVE_AREA, 512);
 609         else
 610                 __pcpu_sigp_relax(addr, SIGP_STORE_STATUS_AT_ADDRESS, page);
 611         save_area_add_regs(sa, regs);
 612 }
 613
 614 void __init smp_save_dump_cpus(void)
 615 {
 616         int addr, boot_cpu_addr, max_cpu_addr;
 617         struct save_area *sa;
 618         unsigned long page;
 619         bool is_boot_cpu;
 620
 621         if (!(OLDMEM_BASE || ipl_info.type == IPL_TYPE_FCP_DUMP))
 622                 /* No previous system present, normal boot. */
 623                 return;
 624         /* Allocate a page as dumping area for the store status sigps */
 625         page = memblock_alloc_base(PAGE_SIZE, PAGE_SIZE, 1UL << 31);
 626         /* Set multi-threading state to the previous system. */
 627         pcpu_set_smt(sclp.mtid_prev);
 628         boot_cpu_addr = stap();
 629         max_cpu_addr = SCLP_MAX_CORES << sclp.mtid_prev;
 630         for (addr = 0; addr <= max_cpu_addr; addr++) {
 631                 if (__pcpu_sigp_relax(addr, SIGP_SENSE, 0) ==
 632                     SIGP_CC_NOT_OPERATIONAL)
 633                         continue;
 634                 is_boot_cpu = (addr == boot_cpu_addr);
 635                 /* Allocate save area */
 636                 sa = save_area_alloc(is_boot_cpu);
 637                 if (!sa)
 638                         panic("could not allocate memory for save area\n");
 639                 if (MACHINE_HAS_VX)
 640                         /* Get the vector registers */
 641                         smp_save_cpu_vxrs(sa, addr, is_boot_cpu, page);
 642                 /*
 643                  * For a zfcp dump OLDMEM_BASE == NULL and the registers
 644                  * of the boot CPU are stored in the HSA. To retrieve
 645                  * these registers an SCLP request is required which is
 646                  * done by drivers/s390/char/zcore.c:init_cpu_info()
 647                  */
 648                 if (!is_boot_cpu || OLDMEM_BASE)
 649                         /* Get the CPU registers */
 650                         smp_save_cpu_regs(sa, addr, is_boot_cpu, page);
 651         }
 652         memblock_free(page, PAGE_SIZE);
 653         diag308_reset();
 654         pcpu_set_smt(0);
 655 }
 656 #endif /* CONFIG_CRASH_DUMP */
 657
 658 void smp_cpu_set_polarization(int cpu, int val)
 659 {
 660         pcpu_devices[cpu].polarization = val;
 661 }
 662
 663 int smp_cpu_get_polarization(int cpu)
 664 {
 665         return pcpu_devices[cpu].polarization;
 666 }
 667
 668 static struct sclp_core_info *smp_get_core_info(void)
 669 {
 670         static int use_sigp_detection;
 671         struct sclp_core_info *info;
 672         int address;
 673
 674         info = kzalloc(sizeof(*info), GFP_KERNEL);
 675         if (info && (use_sigp_detection || sclp_get_core_info(info))) {
 676                 use_sigp_detection = 1;
 677                 for (address = 0;
 678                      address < (SCLP_MAX_CORES << smp_cpu_mt_shift);
 679                      address += (1U << smp_cpu_mt_shift)) {
 680                         if (__pcpu_sigp_relax(address, SIGP_SENSE, 0) ==
 681                             SIGP_CC_NOT_OPERATIONAL)
 682                                 continue;
 683                         info->core[info->configured].core_id =
 684                                 address >> smp_cpu_mt_shift;
 685                         info->configured++;
 686                 }
 687                 info->combined = info->configured;
 688         }
 689         return info;
 690 }
 691
 692 static int smp_add_present_cpu(int cpu);
 693
 694 static int smp_add_core(struct sclp_core_entry *core, cpumask_t *avail,
 695                         bool configured, bool early)
 696 {
 697         struct pcpu *pcpu;
 698         int cpu, nr, i;
 699         u16 address;
 700
 701         nr = 0;
 702         if (sclp.has_core_type && core->type != boot_core_type)
 703                 return nr;
 704         cpu = cpumask_first(avail);
 705         address = core->core_id << smp_cpu_mt_shift;
 706         for (i = 0; (i <= smp_cpu_mtid) && (cpu < nr_cpu_ids); i++) {
 707                 if (pcpu_find_address(cpu_present_mask, address + i))
 708                         continue;
 709                 pcpu = pcpu_devices + cpu;
 710                 pcpu->address = address + i;
 711                 if (configured)
 712                         pcpu->state = CPU_STATE_CONFIGURED;
 713                 else
 714                         pcpu->state = CPU_STATE_STANDBY;
 715                 smp_cpu_set_polarization(cpu, POLARIZATION_UNKNOWN);
 716                 set_cpu_present(cpu, true);
 717                 if (!early && smp_add_present_cpu(cpu) != 0)
 718                         set_cpu_present(cpu, false);
 719                 else
 720                         nr++;
 721                 cpumask_clear_cpu(cpu, avail);
 722                 cpu = cpumask_next(cpu, avail);
 723         }
 724         return nr;
 725 }
 726
 727 static int __smp_rescan_cpus(struct sclp_core_info *info, bool early)
 728 {
 729         struct sclp_core_entry *core;
 730         static cpumask_t avail;
 731         bool configured;
 732         u16 core_id;
 733         int nr, i;
 734
 735         nr = 0;
 736         cpumask_xor(&avail, cpu_possible_mask, cpu_present_mask);
 737         /*
 738          * Add IPL core first (which got logical CPU number 0) to make sure
 739          * that all SMT threads get subsequent logical CPU numbers.
 740          */
 741         if (early) {
 742                 core_id = pcpu_devices[0].address >> smp_cpu_mt_shift;
 743                 for (i = 0; i < info->configured; i++) {
 744                         core = &info->core[i];
 745                         if (core->core_id == core_id) {
 746                                 nr += smp_add_core(core, &avail, true, early);
 747                                 break;
 748                         }
 749                 }
 750         }
 751         for (i = 0; i < info->combined; i++) {
 752                 configured = i < info->configured;
 753                 nr += smp_add_core(&info->core[i], &avail, configured, early);
 754         }
 755         return nr;
 756 }
 757
 758 static void __init smp_detect_cpus(void)
 759 {
 760         unsigned int cpu, mtid, c_cpus, s_cpus;
 761         struct sclp_core_info *info;
 762         u16 address;
 763
 764         /* Get CPU information */
 765         info = smp_get_core_info();
 766         if (!info)
 767                 panic("smp_detect_cpus failed to allocate memory\n");
 768
 769         /* Find boot CPU type */
 770         if (sclp.has_core_type) {
 771                 address = stap();
 772                 for (cpu = 0; cpu < info->combined; cpu++)
 773                         if (info->core[cpu].core_id == address) {
 774                                 /* The boot cpu dictates the cpu type. */
 775                                 boot_core_type = info->core[cpu].type;
 776                                 break;
 777                         }
 778                 if (cpu >= info->combined)
 779                         panic("Could not find boot CPU type");
 780         }
 781
 782         /* Set multi-threading state for the current system */
 783         mtid = boot_core_type ? sclp.mtid : sclp.mtid_cp;
 784         mtid = (mtid < smp_max_threads) ? mtid : smp_max_threads - 1;
 785         pcpu_set_smt(mtid);
 786
 787         /* Print number of CPUs */
 788         c_cpus = s_cpus = 0;
 789         for (cpu = 0; cpu < info->combined; cpu++) {
 790                 if (sclp.has_core_type &&
 791                     info->core[cpu].type != boot_core_type)
 792                         continue;
 793                 if (cpu < info->configured)
 794                         c_cpus += smp_cpu_mtid + 1;
 795                 else
 796                         s_cpus += smp_cpu_mtid + 1;
 797         }
 798         pr_info("%d configured CPUs, %d standby CPUs\n", c_cpus, s_cpus);
 799
 800         /* Add CPUs present at boot */
 801         get_online_cpus();
 802         __smp_rescan_cpus(info, true);
 803         put_online_cpus();
 804         kfree(info);
 805 }
 806
 807 /*
 808  *      Activate a secondary processor.
 809  */
 810 static void smp_start_secondary(void *cpuvoid)
 811 {
 812         S390_lowcore.last_update_clock = get_tod_clock();
 813         S390_lowcore.restart_stack = (unsigned long) restart_stack;
 814         S390_lowcore.restart_fn = (unsigned long) do_restart;
 815         S390_lowcore.restart_data = 0;
 816         S390_lowcore.restart_source = -1UL;
 817         restore_access_regs(S390_lowcore.access_regs_save_area);
 818         __ctl_load(S390_lowcore.cregs_save_area, 0, 15);
 819         __load_psw_mask(PSW_KERNEL_BITS | PSW_MASK_DAT);
 820         cpu_init();
 821         preempt_disable();
 822         init_cpu_timer();
 823         vtime_init();
 824         pfault_init();
 825         notify_cpu_starting(smp_processor_id());
 826         set_cpu_online(smp_processor_id(), true);
 827         inc_irq_stat(CPU_RST);
 828         local_irq_enable();
 829         cpu_startup_entry(CPUHP_AP_ONLINE_IDLE);
 830 }
 831
 832 /* Upping and downing of CPUs */
 833 int __cpu_up(unsigned int cpu, struct task_struct *tidle)
 834 {
 835         struct pcpu *pcpu;
 836         int base, i, rc;
 837
 838         pcpu = pcpu_devices + cpu;
 839         if (pcpu->state != CPU_STATE_CONFIGURED)
 840                 return -EIO;
 841         base = cpu - (cpu % (smp_cpu_mtid + 1));
 842         for (i = 0; i <= smp_cpu_mtid; i++) {
 843                 if (base + i < nr_cpu_ids)
 844                         if (cpu_online(base + i))
 845                                 break;
 846         }
 847         /*
 848          * If this is the first CPU of the core to get online
 849          * do an initial CPU reset.
 850          */
 851         if (i > smp_cpu_mtid &&
 852             pcpu_sigp_retry(pcpu_devices + base, SIGP_INITIAL_CPU_RESET, 0) !=
 853             SIGP_CC_ORDER_CODE_ACCEPTED)
 854                 return -EIO;
 855
 856         rc = pcpu_alloc_lowcore(pcpu, cpu);
 857         if (rc)
 858                 return rc;
 859         pcpu_prepare_secondary(pcpu, cpu);
 860         pcpu_attach_task(pcpu, tidle);
 861         pcpu_start_fn(pcpu, smp_start_secondary, NULL);
 862         /* Wait until cpu puts itself in the online & active maps */
 863         while (!cpu_online(cpu))
 864                 cpu_relax();
 865         return 0;
 866 }
 867
 868 static unsigned int setup_possible_cpus __initdata;
 869
 870 static int __init _setup_possible_cpus(char *s)
 871 {
 872         get_option(&s, &setup_possible_cpus);
 873         return 0;
 874 }
 875 early_param("possible_cpus", _setup_possible_cpus);
 876
 877 #ifdef CONFIG_HOTPLUG_CPU
 878
 879 int __cpu_disable(void)
 880 {
 881         unsigned long cregs[16];
 882
 883         /* Handle possible pending IPIs */
 884         smp_handle_ext_call();
 885         set_cpu_online(smp_processor_id(), false);
 886         /* Disable pseudo page faults on this cpu. */
 887         pfault_fini();
 888         /* Disable interrupt sources via control register. */
 889         __ctl_store(cregs, 0, 15);
 890         cregs[0]  &= ~0x0000ee70UL;     /* disable all external interrupts */
 891         cregs[6]  &= ~0xff000000UL;     /* disable all I/O interrupts */
 892         cregs[14] &= ~0x1f000000UL;     /* disable most machine checks */
 893         __ctl_load(cregs, 0, 15);
 894         clear_cpu_flag(CIF_NOHZ_DELAY);
 895         return 0;
 896 }
 897
 898 void __cpu_die(unsigned int cpu)
 899 {
 900         struct pcpu *pcpu;
 901
 902         /* Wait until target cpu is down */
 903         pcpu = pcpu_devices + cpu;
 904         while (!pcpu_stopped(pcpu))
 905                 cpu_relax();
 906         pcpu_free_lowcore(pcpu);
 907         cpumask_clear_cpu(cpu, mm_cpumask(&init_mm));
 908         cpumask_clear_cpu(cpu, &init_mm.context.cpu_attach_mask);
 909 }
 910
 911 void __noreturn cpu_die(void)
 912 {
 913         idle_task_exit();
 914         __bpon();
 915         pcpu_sigp_retry(pcpu_devices + smp_processor_id(), SIGP_STOP, 0);
 916         for (;;) ;
 917 }
 918
 919 #endif /* CONFIG_HOTPLUG_CPU */
 920
 921 void __init smp_fill_possible_mask(void)
 922 {
 923         unsigned int possible, sclp_max, cpu;
 924
 925         sclp_max = max(sclp.mtid, sclp.mtid_cp) + 1;
 926         sclp_max = min(smp_max_threads, sclp_max);
 927         sclp_max = (sclp.max_cores * sclp_max) ?: nr_cpu_ids;
 928         possible = setup_possible_cpus ?: nr_cpu_ids;
 929         possible = min(possible, sclp_max);
 930         for (cpu = 0; cpu < possible && cpu < nr_cpu_ids; cpu++)
 931                 set_cpu_possible(cpu, true);
 932 }
 933
 934 void __init smp_prepare_cpus(unsigned int max_cpus)
 935 {
 936         /* request the 0x1201 emergency signal external interrupt */
 937         if (register_external_irq(EXT_IRQ_EMERGENCY_SIG, do_ext_call_interrupt))
 938                 panic("Couldn't request external interrupt 0x1201");
 939         /* request the 0x1202 external call external interrupt */
 940         if (register_external_irq(EXT_IRQ_EXTERNAL_CALL, do_ext_call_interrupt))
 941                 panic("Couldn't request external interrupt 0x1202");
 942         smp_detect_cpus();
 943 }
 944
 945 void __init smp_prepare_boot_cpu(void)
 946 {
 947         struct pcpu *pcpu = pcpu_devices;
 948
 949         pcpu->state = CPU_STATE_CONFIGURED;
 950         pcpu->address = stap();
 951         pcpu->lowcore = (struct lowcore *)(unsigned long) store_prefix();
 952         S390_lowcore.percpu_offset = __per_cpu_offset[0];
 953         smp_cpu_set_polarization(0, POLARIZATION_UNKNOWN);
 954         set_cpu_present(0, true);
 955         set_cpu_online(0, true);
 956 }
 957
 958 void __init smp_cpus_done(unsigned int max_cpus)
 959 {
 960 }
 961
 962 void __init smp_setup_processor_id(void)
 963 {
 964         S390_lowcore.cpu_nr = 0;
 965         S390_lowcore.spinlock_lockval = arch_spin_lockval(0);
 966 }
 967
 968 /*
 969  * the frequency of the profiling timer can be changed
 970  * by writing a multiplier value into /proc/profile.
 971  *
 972  * usually you want to run this on all CPUs ;)
 973  */
 974 int setup_profiling_timer(unsigned int multiplier)
 975 {
 976         return 0;
 977 }
 978
 979 #ifdef CONFIG_HOTPLUG_CPU
 980 static ssize_t cpu_configure_show(struct device *dev,
 981                                   struct device_attribute *attr, char *buf)
 982 {
 983         ssize_t count;
 984
 985         mutex_lock(&smp_cpu_state_mutex);
 986         count = sprintf(buf, "%d\n", pcpu_devices[dev->id].state);
 987         mutex_unlock(&smp_cpu_state_mutex);
 988         return count;
 989 }
 990
 991 static ssize_t cpu_configure_store(struct device *dev,
 992                                    struct device_attribute *attr,
 993                                    const char *buf, size_t count)
 994 {
 995         struct pcpu *pcpu;
 996         int cpu, val, rc, i;
 997         char delim;
 998
 999         if (sscanf(buf, "%d %c", &val, &delim) != 1)
1000                 return -EINVAL;
1001         if (val != 0 && val != 1)
1002                 return -EINVAL;
1003         get_online_cpus();
1004         mutex_lock(&smp_cpu_state_mutex);
1005         rc = -EBUSY;
1006         /* disallow configuration changes of online cpus and cpu 0 */
1007         cpu = dev->id;
1008         cpu -= cpu % (smp_cpu_mtid + 1);
1009         if (cpu == 0)
1010                 goto out;
1011         for (i = 0; i <= smp_cpu_mtid; i++)
1012                 if (cpu_online(cpu + i))
1013                         goto out;
1014         pcpu = pcpu_devices + cpu;
1015         rc = 0;
1016         switch (val) {
1017         case 0:
1018                 if (pcpu->state != CPU_STATE_CONFIGURED)
1019                         break;
1020                 rc = sclp_core_deconfigure(pcpu->address >> smp_cpu_mt_shift);
1021                 if (rc)
1022                         break;
1023                 for (i = 0; i <= smp_cpu_mtid; i++) {
1024                         if (cpu + i >= nr_cpu_ids || !cpu_present(cpu + i))
1025                                 continue;
1026                         pcpu[i].state = CPU_STATE_STANDBY;
1027                         smp_cpu_set_polarization(cpu + i,
1028                                                  POLARIZATION_UNKNOWN);
1029                 }
1030                 topology_expect_change();
1031                 break;
1032         case 1:
1033                 if (pcpu->state != CPU_STATE_STANDBY)
1034                         break;
1035                 rc = sclp_core_configure(pcpu->address >> smp_cpu_mt_shift);
1036                 if (rc)
1037                         break;
1038                 for (i = 0; i <= smp_cpu_mtid; i++) {
1039                         if (cpu + i >= nr_cpu_ids || !cpu_present(cpu + i))
1040                                 continue;
1041                         pcpu[i].state = CPU_STATE_CONFIGURED;
1042                         smp_cpu_set_polarization(cpu + i,
1043                                                  POLARIZATION_UNKNOWN);
1044                 }
1045                 topology_expect_change();
1046                 break;
1047         default:
1048                 break;
1049         }
1050 out:
1051         mutex_unlock(&smp_cpu_state_mutex);
1052         put_online_cpus();
1053         return rc ? rc : count;
1054 }
1055 static DEVICE_ATTR(configure, 0644, cpu_configure_show, cpu_configure_store);
1056 #endif /* CONFIG_HOTPLUG_CPU */
1057
1058 static ssize_t show_cpu_address(struct device *dev,
1059                                 struct device_attribute *attr, char *buf)
1060 {
1061         return sprintf(buf, "%d\n", pcpu_devices[dev->id].address);
1062 }
1063 static DEVICE_ATTR(address, 0444, show_cpu_address, NULL);
1064
1065 static struct attribute *cpu_common_attrs[] = {
1066 #ifdef CONFIG_HOTPLUG_CPU
1067         &dev_attr_configure.attr,
1068 #endif
1069         &dev_attr_address.attr,
1070         NULL,
1071 };
1072
1073 static struct attribute_group cpu_common_attr_group = {
1074         .attrs = cpu_common_attrs,
1075 };
1076
1077 static struct attribute *cpu_online_attrs[] = {
1078         &dev_attr_idle_count.attr,
1079         &dev_attr_idle_time_us.attr,
1080         NULL,
1081 };
1082
1083 static struct attribute_group cpu_online_attr_group = {
1084         .attrs = cpu_online_attrs,
1085 };
1086
1087 static int smp_cpu_notify(struct notifier_block *self, unsigned long action,
1088                           void *hcpu)
1089 {
1090         unsigned int cpu = (unsigned int)(long)hcpu;
1091         struct device *s = &per_cpu(cpu_device, cpu)->dev;
1092         int err = 0;
1093
1094         switch (action & ~CPU_TASKS_FROZEN) {
1095         case CPU_ONLINE:
1096                 err = sysfs_create_group(&s->kobj, &cpu_online_attr_group);
1097                 break;
1098         case CPU_DEAD:
1099                 sysfs_remove_group(&s->kobj, &cpu_online_attr_group);
1100                 break;
1101         }
1102         return notifier_from_errno(err);
1103 }
1104
1105 static int smp_add_present_cpu(int cpu)
1106 {
1107         struct device *s;
1108         struct cpu *c;
1109         int rc;
1110
1111         c = kzalloc(sizeof(*c), GFP_KERNEL);
1112         if (!c)
1113                 return -ENOMEM;
1114         per_cpu(cpu_device, cpu) = c;
1115         s = &c->dev;
1116         c->hotpluggable = 1;
1117         rc = register_cpu(c, cpu);
1118         if (rc)
1119                 goto out;
1120         rc = sysfs_create_group(&s->kobj, &cpu_common_attr_group);
1121         if (rc)
1122                 goto out_cpu;
1123         if (cpu_online(cpu)) {
1124                 rc = sysfs_create_group(&s->kobj, &cpu_online_attr_group);
1125                 if (rc)
1126                         goto out_online;
1127         }
1128         rc = topology_cpu_init(c);
1129         if (rc)
1130                 goto out_topology;
1131         return 0;
1132
1133 out_topology:
1134         if (cpu_online(cpu))
1135                 sysfs_remove_group(&s->kobj, &cpu_online_attr_group);
1136 out_online:
1137         sysfs_remove_group(&s->kobj, &cpu_common_attr_group);
1138 out_cpu:
1139 #ifdef CONFIG_HOTPLUG_CPU
1140         unregister_cpu(c);
1141 #endif
1142 out:
1143         return rc;
1144 }
1145
1146 #ifdef CONFIG_HOTPLUG_CPU
1147
1148 int __ref smp_rescan_cpus(void)
1149 {
1150         struct sclp_core_info *info;
1151         int nr;
1152
1153         info = smp_get_core_info();
1154         if (!info)
1155                 return -ENOMEM;
1156         get_online_cpus();
1157         mutex_lock(&smp_cpu_state_mutex);
1158         nr = __smp_rescan_cpus(info, false);
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(rescan, 0200, NULL, rescan_store);
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         cpu_notifier_register_begin();
1194         for_each_present_cpu(cpu) {
1195                 rc = smp_add_present_cpu(cpu);
1196                 if (rc)
1197                         goto out;
1198         }
1199
1200         __hotcpu_notifier(smp_cpu_notify, 0);
1201
1202 out:
1203         cpu_notifier_register_done();
1204         return rc;
1205 }
1206 subsys_initcall(s390_smp_init);