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         if (MACHINE_HAS_VX)
 209                 lc->vector_save_area_addr =
 210                         (unsigned long) &lc->vector_save_area;
 211         if (vdso_alloc_per_cpu(lc))
 212                 goto out;
 213         lowcore_ptr[cpu] = lc;
 214         pcpu_sigp_retry(pcpu, SIGP_SET_PREFIX, (u32)(unsigned long) lc);
 215         return 0;
 216 out:
 217         if (pcpu != &pcpu_devices[0]) {
 218                 free_page(panic_stack);
 219                 free_pages(async_stack, ASYNC_ORDER);
 220                 free_pages((unsigned long) pcpu->lowcore, LC_ORDER);
 221         }
 222         return -ENOMEM;
 223 }
 224
 225 #ifdef CONFIG_HOTPLUG_CPU
 226
 227 static void pcpu_free_lowcore(struct pcpu *pcpu)
 228 {
 229         pcpu_sigp_retry(pcpu, SIGP_SET_PREFIX, 0);
 230         lowcore_ptr[pcpu - pcpu_devices] = NULL;
 231         vdso_free_per_cpu(pcpu->lowcore);
 232         if (pcpu == &pcpu_devices[0])
 233                 return;
 234         free_page(pcpu->lowcore->panic_stack-PANIC_FRAME_OFFSET);
 235         free_pages(pcpu->lowcore->async_stack-ASYNC_FRAME_OFFSET, ASYNC_ORDER);
 236         free_pages((unsigned long) pcpu->lowcore, LC_ORDER);
 237 }
 238
 239 #endif /* CONFIG_HOTPLUG_CPU */
 240
 241 static void pcpu_prepare_secondary(struct pcpu *pcpu, int cpu)
 242 {
 243         struct lowcore *lc = pcpu->lowcore;
 244
 245         if (MACHINE_HAS_TLB_LC)
 246                 cpumask_set_cpu(cpu, &init_mm.context.cpu_attach_mask);
 247         cpumask_set_cpu(cpu, mm_cpumask(&init_mm));
 248         atomic_inc(&init_mm.context.attach_count);
 249         lc->cpu_nr = cpu;
 250         lc->spinlock_lockval = arch_spin_lockval(cpu);
 251         lc->percpu_offset = __per_cpu_offset[cpu];
 252         lc->kernel_asce = S390_lowcore.kernel_asce;
 253         lc->machine_flags = S390_lowcore.machine_flags;
 254         lc->user_timer = lc->system_timer = lc->steal_timer = 0;
 255         __ctl_store(lc->cregs_save_area, 0, 15);
 256         save_access_regs((unsigned int *) lc->access_regs_save_area);
 257         memcpy(lc->stfle_fac_list, S390_lowcore.stfle_fac_list,
 258                MAX_FACILITY_BIT/8);
 259 }
 260
 261 static void pcpu_attach_task(struct pcpu *pcpu, struct task_struct *tsk)
 262 {
 263         struct lowcore *lc = pcpu->lowcore;
 264         struct thread_info *ti = task_thread_info(tsk);
 265
 266         lc->kernel_stack = (unsigned long) task_stack_page(tsk)
 267                 + THREAD_SIZE - STACK_FRAME_OVERHEAD - sizeof(struct pt_regs);
 268         lc->thread_info = (unsigned long) task_thread_info(tsk);
 269         lc->current_task = (unsigned long) tsk;
 270         lc->lpp = LPP_MAGIC;
 271         lc->current_pid = tsk->pid;
 272         lc->user_timer = ti->user_timer;
 273         lc->system_timer = ti->system_timer;
 274         lc->steal_timer = 0;
 275 }
 276
 277 static void pcpu_start_fn(struct pcpu *pcpu, void (*func)(void *), void *data)
 278 {
 279         struct lowcore *lc = pcpu->lowcore;
 280
 281         lc->restart_stack = lc->kernel_stack;
 282         lc->restart_fn = (unsigned long) func;
 283         lc->restart_data = (unsigned long) data;
 284         lc->restart_source = -1UL;
 285         pcpu_sigp_retry(pcpu, SIGP_RESTART, 0);
 286 }
 287
 288 /*
 289  * Call function via PSW restart on pcpu and stop the current cpu.
 290  */
 291 static void pcpu_delegate(struct pcpu *pcpu, void (*func)(void *),
 292                           void *data, unsigned long stack)
 293 {
 294         struct lowcore *lc = lowcore_ptr[pcpu - pcpu_devices];
 295         unsigned long source_cpu = stap();
 296
 297         __load_psw_mask(PSW_KERNEL_BITS);
 298         if (pcpu->address == source_cpu)
 299                 func(data);     /* should not return */
 300         /* Stop target cpu (if func returns this stops the current cpu). */
 301         pcpu_sigp_retry(pcpu, SIGP_STOP, 0);
 302         /* Restart func on the target cpu and stop the current cpu. */
 303         mem_assign_absolute(lc->restart_stack, stack);
 304         mem_assign_absolute(lc->restart_fn, (unsigned long) func);
 305         mem_assign_absolute(lc->restart_data, (unsigned long) data);
 306         mem_assign_absolute(lc->restart_source, source_cpu);
 307         asm volatile(
 308                 "0:     sigp    0,%0,%2 # sigp restart to target cpu\n"
 309                 "       brc     2,0b    # busy, try again\n"
 310                 "1:     sigp    0,%1,%3 # sigp stop to current cpu\n"
 311                 "       brc     2,1b    # busy, try again\n"
 312                 : : "d" (pcpu->address), "d" (source_cpu),
 313                     "K" (SIGP_RESTART), "K" (SIGP_STOP)
 314                 : "0", "1", "cc");
 315         for (;;) ;
 316 }
 317
 318 /*
 319  * Enable additional logical cpus for multi-threading.
 320  */
 321 static int pcpu_set_smt(unsigned int mtid)
 322 {
 323         register unsigned long reg1 asm ("1") = (unsigned long) mtid;
 324         int cc;
 325
 326         if (smp_cpu_mtid == mtid)
 327                 return 0;
 328         asm volatile(
 329                 "       sigp    %1,0,%2 # sigp set multi-threading\n"
 330                 "       ipm     %0\n"
 331                 "       srl     %0,28\n"
 332                 : "=d" (cc) : "d" (reg1), "K" (SIGP_SET_MULTI_THREADING)
 333                 : "cc");
 334         if (cc == 0) {
 335                 smp_cpu_mtid = mtid;
 336                 smp_cpu_mt_shift = 0;
 337                 while (smp_cpu_mtid >= (1U << smp_cpu_mt_shift))
 338                         smp_cpu_mt_shift++;
 339                 pcpu_devices[0].address = stap();
 340         }
 341         return cc;
 342 }
 343
 344 /*
 345  * Call function on an online CPU.
 346  */
 347 void smp_call_online_cpu(void (*func)(void *), void *data)
 348 {
 349         struct pcpu *pcpu;
 350
 351         /* Use the current cpu if it is online. */
 352         pcpu = pcpu_find_address(cpu_online_mask, stap());
 353         if (!pcpu)
 354                 /* Use the first online cpu. */
 355                 pcpu = pcpu_devices + cpumask_first(cpu_online_mask);
 356         pcpu_delegate(pcpu, func, data, (unsigned long) restart_stack);
 357 }
 358
 359 /*
 360  * Call function on the ipl CPU.
 361  */
 362 void smp_call_ipl_cpu(void (*func)(void *), void *data)
 363 {
 364         pcpu_delegate(&pcpu_devices[0], func, data,
 365                       pcpu_devices->lowcore->panic_stack -
 366                       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         if (!page)
 627                 panic("could not allocate memory for save area\n");
 628         /* Set multi-threading state to the previous system. */
 629         pcpu_set_smt(sclp.mtid_prev);
 630         boot_cpu_addr = stap();
 631         max_cpu_addr = SCLP_MAX_CORES << sclp.mtid_prev;
 632         for (addr = 0; addr <= max_cpu_addr; addr++) {
 633                 if (__pcpu_sigp_relax(addr, SIGP_SENSE, 0) ==
 634                     SIGP_CC_NOT_OPERATIONAL)
 635                         continue;
 636                 is_boot_cpu = (addr == boot_cpu_addr);
 637                 /* Allocate save area */
 638                 sa = save_area_alloc(is_boot_cpu);
 639                 if (!sa)
 640                         panic("could not allocate memory for save area\n");
 641                 if (MACHINE_HAS_VX)
 642                         /* Get the vector registers */
 643                         smp_save_cpu_vxrs(sa, addr, is_boot_cpu, page);
 644                 /*
 645                  * For a zfcp dump OLDMEM_BASE == NULL and the registers
 646                  * of the boot CPU are stored in the HSA. To retrieve
 647                  * these registers an SCLP request is required which is
 648                  * done by drivers/s390/char/zcore.c:init_cpu_info()
 649                  */
 650                 if (!is_boot_cpu || OLDMEM_BASE)
 651                         /* Get the CPU registers */
 652                         smp_save_cpu_regs(sa, addr, is_boot_cpu, page);
 653         }
 654         memblock_free(page, PAGE_SIZE);
 655         diag308_reset();
 656         pcpu_set_smt(0);
 657 }
 658 #endif /* CONFIG_CRASH_DUMP */
 659
 660 void smp_cpu_set_polarization(int cpu, int val)
 661 {
 662         pcpu_devices[cpu].polarization = val;
 663 }
 664
 665 int smp_cpu_get_polarization(int cpu)
 666 {
 667         return pcpu_devices[cpu].polarization;
 668 }
 669
 670 static struct sclp_core_info *smp_get_core_info(void)
 671 {
 672         static int use_sigp_detection;
 673         struct sclp_core_info *info;
 674         int address;
 675
 676         info = kzalloc(sizeof(*info), GFP_KERNEL);
 677         if (info && (use_sigp_detection || sclp_get_core_info(info))) {
 678                 use_sigp_detection = 1;
 679                 for (address = 0;
 680                      address < (SCLP_MAX_CORES << smp_cpu_mt_shift);
 681                      address += (1U << smp_cpu_mt_shift)) {
 682                         if (__pcpu_sigp_relax(address, SIGP_SENSE, 0) ==
 683                             SIGP_CC_NOT_OPERATIONAL)
 684                                 continue;
 685                         info->core[info->configured].core_id =
 686                                 address >> smp_cpu_mt_shift;
 687                         info->configured++;
 688                 }
 689                 info->combined = info->configured;
 690         }
 691         return info;
 692 }
 693
 694 static int smp_add_present_cpu(int cpu);
 695
 696 static int __smp_rescan_cpus(struct sclp_core_info *info, int sysfs_add)
 697 {
 698         struct pcpu *pcpu;
 699         cpumask_t avail;
 700         int cpu, nr, i, j;
 701         u16 address;
 702
 703         nr = 0;
 704         cpumask_xor(&avail, cpu_possible_mask, cpu_present_mask);
 705         cpu = cpumask_first(&avail);
 706         for (i = 0; (i < info->combined) && (cpu < nr_cpu_ids); i++) {
 707                 if (sclp.has_core_type && info->core[i].type != boot_core_type)
 708                         continue;
 709                 address = info->core[i].core_id << smp_cpu_mt_shift;
 710                 for (j = 0; j <= smp_cpu_mtid; j++) {
 711                         if (pcpu_find_address(cpu_present_mask, address + j))
 712                                 continue;
 713                         pcpu = pcpu_devices + cpu;
 714                         pcpu->address = address + j;
 715                         pcpu->state =
 716                                 (cpu >= info->configured*(smp_cpu_mtid + 1)) ?
 717                                 CPU_STATE_STANDBY : CPU_STATE_CONFIGURED;
 718                         smp_cpu_set_polarization(cpu, POLARIZATION_UNKNOWN);
 719                         set_cpu_present(cpu, true);
 720                         if (sysfs_add && smp_add_present_cpu(cpu) != 0)
 721                                 set_cpu_present(cpu, false);
 722                         else
 723                                 nr++;
 724                         cpu = cpumask_next(cpu, &avail);
 725                         if (cpu >= nr_cpu_ids)
 726                                 break;
 727                 }
 728         }
 729         return nr;
 730 }
 731
 732 static void __init smp_detect_cpus(void)
 733 {
 734         unsigned int cpu, mtid, c_cpus, s_cpus;
 735         struct sclp_core_info *info;
 736         u16 address;
 737
 738         /* Get CPU information */
 739         info = smp_get_core_info();
 740         if (!info)
 741                 panic("smp_detect_cpus failed to allocate memory\n");
 742
 743         /* Find boot CPU type */
 744         if (sclp.has_core_type) {
 745                 address = stap();
 746                 for (cpu = 0; cpu < info->combined; cpu++)
 747                         if (info->core[cpu].core_id == address) {
 748                                 /* The boot cpu dictates the cpu type. */
 749                                 boot_core_type = info->core[cpu].type;
 750                                 break;
 751                         }
 752                 if (cpu >= info->combined)
 753                         panic("Could not find boot CPU type");
 754         }
 755
 756         /* Set multi-threading state for the current system */
 757         mtid = boot_core_type ? sclp.mtid : sclp.mtid_cp;
 758         mtid = (mtid < smp_max_threads) ? mtid : smp_max_threads - 1;
 759         pcpu_set_smt(mtid);
 760
 761         /* Print number of CPUs */
 762         c_cpus = s_cpus = 0;
 763         for (cpu = 0; cpu < info->combined; cpu++) {
 764                 if (sclp.has_core_type &&
 765                     info->core[cpu].type != boot_core_type)
 766                         continue;
 767                 if (cpu < info->configured)
 768                         c_cpus += smp_cpu_mtid + 1;
 769                 else
 770                         s_cpus += smp_cpu_mtid + 1;
 771         }
 772         pr_info("%d configured CPUs, %d standby CPUs\n", c_cpus, s_cpus);
 773
 774         /* Add CPUs present at boot */
 775         get_online_cpus();
 776         __smp_rescan_cpus(info, 0);
 777         put_online_cpus();
 778         kfree(info);
 779 }
 780
 781 /*
 782  *      Activate a secondary processor.
 783  */
 784 static void smp_start_secondary(void *cpuvoid)
 785 {
 786         S390_lowcore.last_update_clock = get_tod_clock();
 787         S390_lowcore.restart_stack = (unsigned long) restart_stack;
 788         S390_lowcore.restart_fn = (unsigned long) do_restart;
 789         S390_lowcore.restart_data = 0;
 790         S390_lowcore.restart_source = -1UL;
 791         restore_access_regs(S390_lowcore.access_regs_save_area);
 792         __ctl_load(S390_lowcore.cregs_save_area, 0, 15);
 793         __load_psw_mask(PSW_KERNEL_BITS | PSW_MASK_DAT);
 794         cpu_init();
 795         preempt_disable();
 796         init_cpu_timer();
 797         vtime_init();
 798         pfault_init();
 799         notify_cpu_starting(smp_processor_id());
 800         set_cpu_online(smp_processor_id(), true);
 801         inc_irq_stat(CPU_RST);
 802         local_irq_enable();
 803         cpu_startup_entry(CPUHP_ONLINE);
 804 }
 805
 806 /* Upping and downing of CPUs */
 807 int __cpu_up(unsigned int cpu, struct task_struct *tidle)
 808 {
 809         struct pcpu *pcpu;
 810         int base, i, rc;
 811
 812         pcpu = pcpu_devices + cpu;
 813         if (pcpu->state != CPU_STATE_CONFIGURED)
 814                 return -EIO;
 815         base = cpu - (cpu % (smp_cpu_mtid + 1));
 816         for (i = 0; i <= smp_cpu_mtid; i++) {
 817                 if (base + i < nr_cpu_ids)
 818                         if (cpu_online(base + i))
 819                                 break;
 820         }
 821         /*
 822          * If this is the first CPU of the core to get online
 823          * do an initial CPU reset.
 824          */
 825         if (i > smp_cpu_mtid &&
 826             pcpu_sigp_retry(pcpu_devices + base, SIGP_INITIAL_CPU_RESET, 0) !=
 827             SIGP_CC_ORDER_CODE_ACCEPTED)
 828                 return -EIO;
 829
 830         rc = pcpu_alloc_lowcore(pcpu, cpu);
 831         if (rc)
 832                 return rc;
 833         pcpu_prepare_secondary(pcpu, cpu);
 834         pcpu_attach_task(pcpu, tidle);
 835         pcpu_start_fn(pcpu, smp_start_secondary, NULL);
 836         /* Wait until cpu puts itself in the online & active maps */
 837         while (!cpu_online(cpu) || !cpu_active(cpu))
 838                 cpu_relax();
 839         return 0;
 840 }
 841
 842 static unsigned int setup_possible_cpus __initdata;
 843
 844 static int __init _setup_possible_cpus(char *s)
 845 {
 846         get_option(&s, &setup_possible_cpus);
 847         return 0;
 848 }
 849 early_param("possible_cpus", _setup_possible_cpus);
 850
 851 #ifdef CONFIG_HOTPLUG_CPU
 852
 853 int __cpu_disable(void)
 854 {
 855         unsigned long cregs[16];
 856
 857         /* Handle possible pending IPIs */
 858         smp_handle_ext_call();
 859         set_cpu_online(smp_processor_id(), false);
 860         /* Disable pseudo page faults on this cpu. */
 861         pfault_fini();
 862         /* Disable interrupt sources via control register. */
 863         __ctl_store(cregs, 0, 15);
 864         cregs[0]  &= ~0x0000ee70UL;     /* disable all external interrupts */
 865         cregs[6]  &= ~0xff000000UL;     /* disable all I/O interrupts */
 866         cregs[14] &= ~0x1f000000UL;     /* disable most machine checks */
 867         __ctl_load(cregs, 0, 15);
 868         clear_cpu_flag(CIF_NOHZ_DELAY);
 869         return 0;
 870 }
 871
 872 void __cpu_die(unsigned int cpu)
 873 {
 874         struct pcpu *pcpu;
 875
 876         /* Wait until target cpu is down */
 877         pcpu = pcpu_devices + cpu;
 878         while (!pcpu_stopped(pcpu))
 879                 cpu_relax();
 880         pcpu_free_lowcore(pcpu);
 881         atomic_dec(&init_mm.context.attach_count);
 882         cpumask_clear_cpu(cpu, mm_cpumask(&init_mm));
 883         if (MACHINE_HAS_TLB_LC)
 884                 cpumask_clear_cpu(cpu, &init_mm.context.cpu_attach_mask);
 885 }
 886
 887 void __noreturn cpu_die(void)
 888 {
 889         idle_task_exit();
 890         pcpu_sigp_retry(pcpu_devices + smp_processor_id(), SIGP_STOP, 0);
 891         for (;;) ;
 892 }
 893
 894 #endif /* CONFIG_HOTPLUG_CPU */
 895
 896 void __init smp_fill_possible_mask(void)
 897 {
 898         unsigned int possible, sclp_max, cpu;
 899
 900         sclp_max = max(sclp.mtid, sclp.mtid_cp) + 1;
 901         sclp_max = min(smp_max_threads, sclp_max);
 902         sclp_max = sclp.max_cores * sclp_max ?: nr_cpu_ids;
 903         possible = setup_possible_cpus ?: nr_cpu_ids;
 904         possible = min(possible, sclp_max);
 905         for (cpu = 0; cpu < possible && cpu < nr_cpu_ids; cpu++)
 906                 set_cpu_possible(cpu, true);
 907 }
 908
 909 void __init smp_prepare_cpus(unsigned int max_cpus)
 910 {
 911         /* request the 0x1201 emergency signal external interrupt */
 912         if (register_external_irq(EXT_IRQ_EMERGENCY_SIG, do_ext_call_interrupt))
 913                 panic("Couldn't request external interrupt 0x1201");
 914         /* request the 0x1202 external call external interrupt */
 915         if (register_external_irq(EXT_IRQ_EXTERNAL_CALL, do_ext_call_interrupt))
 916                 panic("Couldn't request external interrupt 0x1202");
 917         smp_detect_cpus();
 918 }
 919
 920 void __init smp_prepare_boot_cpu(void)
 921 {
 922         struct pcpu *pcpu = pcpu_devices;
 923
 924         pcpu->state = CPU_STATE_CONFIGURED;
 925         pcpu->address = stap();
 926         pcpu->lowcore = (struct lowcore *)(unsigned long) store_prefix();
 927         S390_lowcore.percpu_offset = __per_cpu_offset[0];
 928         smp_cpu_set_polarization(0, POLARIZATION_UNKNOWN);
 929         set_cpu_present(0, true);
 930         set_cpu_online(0, true);
 931 }
 932
 933 void __init smp_cpus_done(unsigned int max_cpus)
 934 {
 935 }
 936
 937 void __init smp_setup_processor_id(void)
 938 {
 939         S390_lowcore.cpu_nr = 0;
 940         S390_lowcore.spinlock_lockval = arch_spin_lockval(0);
 941 }
 942
 943 /*
 944  * the frequency of the profiling timer can be changed
 945  * by writing a multiplier value into /proc/profile.
 946  *
 947  * usually you want to run this on all CPUs ;)
 948  */
 949 int setup_profiling_timer(unsigned int multiplier)
 950 {
 951         return 0;
 952 }
 953
 954 #ifdef CONFIG_HOTPLUG_CPU
 955 static ssize_t cpu_configure_show(struct device *dev,
 956                                   struct device_attribute *attr, char *buf)
 957 {
 958         ssize_t count;
 959
 960         mutex_lock(&smp_cpu_state_mutex);
 961         count = sprintf(buf, "%d\n", pcpu_devices[dev->id].state);
 962         mutex_unlock(&smp_cpu_state_mutex);
 963         return count;
 964 }
 965
 966 static ssize_t cpu_configure_store(struct device *dev,
 967                                    struct device_attribute *attr,
 968                                    const char *buf, size_t count)
 969 {
 970         struct pcpu *pcpu;
 971         int cpu, val, rc, i;
 972         char delim;
 973
 974         if (sscanf(buf, "%d %c", &val, &delim) != 1)
 975                 return -EINVAL;
 976         if (val != 0 && val != 1)
 977                 return -EINVAL;
 978         get_online_cpus();
 979         mutex_lock(&smp_cpu_state_mutex);
 980         rc = -EBUSY;
 981         /* disallow configuration changes of online cpus and cpu 0 */
 982         cpu = dev->id;
 983         cpu -= cpu % (smp_cpu_mtid + 1);
 984         if (cpu == 0)
 985                 goto out;
 986         for (i = 0; i <= smp_cpu_mtid; i++)
 987                 if (cpu_online(cpu + i))
 988                         goto out;
 989         pcpu = pcpu_devices + cpu;
 990         rc = 0;
 991         switch (val) {
 992         case 0:
 993                 if (pcpu->state != CPU_STATE_CONFIGURED)
 994                         break;
 995                 rc = sclp_core_deconfigure(pcpu->address >> smp_cpu_mt_shift);
 996                 if (rc)
 997                         break;
 998                 for (i = 0; i <= smp_cpu_mtid; i++) {
 999                         if (cpu + i >= nr_cpu_ids || !cpu_present(cpu + i))
1000                                 continue;
1001                         pcpu[i].state = CPU_STATE_STANDBY;
1002                         smp_cpu_set_polarization(cpu + i,
1003                                                  POLARIZATION_UNKNOWN);
1004                 }
1005                 topology_expect_change();
1006                 break;
1007         case 1:
1008                 if (pcpu->state != CPU_STATE_STANDBY)
1009                         break;
1010                 rc = sclp_core_configure(pcpu->address >> smp_cpu_mt_shift);
1011                 if (rc)
1012                         break;
1013                 for (i = 0; i <= smp_cpu_mtid; i++) {
1014                         if (cpu + i >= nr_cpu_ids || !cpu_present(cpu + i))
1015                                 continue;
1016                         pcpu[i].state = CPU_STATE_CONFIGURED;
1017                         smp_cpu_set_polarization(cpu + i,
1018                                                  POLARIZATION_UNKNOWN);
1019                 }
1020                 topology_expect_change();
1021                 break;
1022         default:
1023                 break;
1024         }
1025 out:
1026         mutex_unlock(&smp_cpu_state_mutex);
1027         put_online_cpus();
1028         return rc ? rc : count;
1029 }
1030 static DEVICE_ATTR(configure, 0644, cpu_configure_show, cpu_configure_store);
1031 #endif /* CONFIG_HOTPLUG_CPU */
1032
1033 static ssize_t show_cpu_address(struct device *dev,
1034                                 struct device_attribute *attr, char *buf)
1035 {
1036         return sprintf(buf, "%d\n", pcpu_devices[dev->id].address);
1037 }
1038 static DEVICE_ATTR(address, 0444, show_cpu_address, NULL);
1039
1040 static struct attribute *cpu_common_attrs[] = {
1041 #ifdef CONFIG_HOTPLUG_CPU
1042         &dev_attr_configure.attr,
1043 #endif
1044         &dev_attr_address.attr,
1045         NULL,
1046 };
1047
1048 static struct attribute_group cpu_common_attr_group = {
1049         .attrs = cpu_common_attrs,
1050 };
1051
1052 static struct attribute *cpu_online_attrs[] = {
1053         &dev_attr_idle_count.attr,
1054         &dev_attr_idle_time_us.attr,
1055         NULL,
1056 };
1057
1058 static struct attribute_group cpu_online_attr_group = {
1059         .attrs = cpu_online_attrs,
1060 };
1061
1062 static int smp_cpu_notify(struct notifier_block *self, unsigned long action,
1063                           void *hcpu)
1064 {
1065         unsigned int cpu = (unsigned int)(long)hcpu;
1066         struct device *s = &per_cpu(cpu_device, cpu)->dev;
1067         int err = 0;
1068
1069         switch (action & ~CPU_TASKS_FROZEN) {
1070         case CPU_ONLINE:
1071                 err = sysfs_create_group(&s->kobj, &cpu_online_attr_group);
1072                 break;
1073         case CPU_DEAD:
1074                 sysfs_remove_group(&s->kobj, &cpu_online_attr_group);
1075                 break;
1076         }
1077         return notifier_from_errno(err);
1078 }
1079
1080 static int smp_add_present_cpu(int cpu)
1081 {
1082         struct device *s;
1083         struct cpu *c;
1084         int rc;
1085
1086         c = kzalloc(sizeof(*c), GFP_KERNEL);
1087         if (!c)
1088                 return -ENOMEM;
1089         per_cpu(cpu_device, cpu) = c;
1090         s = &c->dev;
1091         c->hotpluggable = 1;
1092         rc = register_cpu(c, cpu);
1093         if (rc)
1094                 goto out;
1095         rc = sysfs_create_group(&s->kobj, &cpu_common_attr_group);
1096         if (rc)
1097                 goto out_cpu;
1098         if (cpu_online(cpu)) {
1099                 rc = sysfs_create_group(&s->kobj, &cpu_online_attr_group);
1100                 if (rc)
1101                         goto out_online;
1102         }
1103         rc = topology_cpu_init(c);
1104         if (rc)
1105                 goto out_topology;
1106         return 0;
1107
1108 out_topology:
1109         if (cpu_online(cpu))
1110                 sysfs_remove_group(&s->kobj, &cpu_online_attr_group);
1111 out_online:
1112         sysfs_remove_group(&s->kobj, &cpu_common_attr_group);
1113 out_cpu:
1114 #ifdef CONFIG_HOTPLUG_CPU
1115         unregister_cpu(c);
1116 #endif
1117 out:
1118         return rc;
1119 }
1120
1121 #ifdef CONFIG_HOTPLUG_CPU
1122
1123 int __ref smp_rescan_cpus(void)
1124 {
1125         struct sclp_core_info *info;
1126         int nr;
1127
1128         info = smp_get_core_info();
1129         if (!info)
1130                 return -ENOMEM;
1131         get_online_cpus();
1132         mutex_lock(&smp_cpu_state_mutex);
1133         nr = __smp_rescan_cpus(info, 1);
1134         mutex_unlock(&smp_cpu_state_mutex);
1135         put_online_cpus();
1136         kfree(info);
1137         if (nr)
1138                 topology_schedule_update();
1139         return 0;
1140 }
1141
1142 static ssize_t __ref rescan_store(struct device *dev,
1143                                   struct device_attribute *attr,
1144                                   const char *buf,
1145                                   size_t count)
1146 {
1147         int rc;
1148
1149         rc = smp_rescan_cpus();
1150         return rc ? rc : count;
1151 }
1152 static DEVICE_ATTR(rescan, 0200, NULL, rescan_store);
1153 #endif /* CONFIG_HOTPLUG_CPU */
1154
1155 static int __init s390_smp_init(void)
1156 {
1157         int cpu, rc = 0;
1158
1159 #ifdef CONFIG_HOTPLUG_CPU
1160         rc = device_create_file(cpu_subsys.dev_root, &dev_attr_rescan);
1161         if (rc)
1162                 return rc;
1163 #endif
1164         cpu_notifier_register_begin();
1165         for_each_present_cpu(cpu) {
1166                 rc = smp_add_present_cpu(cpu);
1167                 if (rc)
1168                         goto out;
1169         }
1170
1171         __hotcpu_notifier(smp_cpu_notify, 0);
1172
1173 out:
1174         cpu_notifier_register_done();
1175         return rc;
1176 }
1177 subsys_initcall(s390_smp_init);