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         cpumask_set_cpu(cpu, &init_mm.context.cpu_attach_mask);
 246         cpumask_set_cpu(cpu, mm_cpumask(&init_mm));
 247         lc->cpu_nr = cpu;
 248         lc->spinlock_lockval = arch_spin_lockval(cpu);
 249         lc->percpu_offset = __per_cpu_offset[cpu];
 250         lc->kernel_asce = S390_lowcore.kernel_asce;
 251         lc->machine_flags = S390_lowcore.machine_flags;
 252         lc->user_timer = lc->system_timer = lc->steal_timer = 0;
 253         __ctl_store(lc->cregs_save_area, 0, 15);
 254         save_access_regs((unsigned int *) lc->access_regs_save_area);
 255         memcpy(lc->stfle_fac_list, S390_lowcore.stfle_fac_list,
 256                MAX_FACILITY_BIT/8);
 257 }
 258
 259 static void pcpu_attach_task(struct pcpu *pcpu, struct task_struct *tsk)
 260 {
 261         struct lowcore *lc = pcpu->lowcore;
 262         struct thread_info *ti = task_thread_info(tsk);
 263
 264         lc->kernel_stack = (unsigned long) task_stack_page(tsk)
 265                 + THREAD_SIZE - STACK_FRAME_OVERHEAD - sizeof(struct pt_regs);
 266         lc->thread_info = (unsigned long) task_thread_info(tsk);
 267         lc->current_task = (unsigned long) tsk;
 268         lc->lpp = LPP_MAGIC;
 269         lc->current_pid = tsk->pid;
 270         lc->user_timer = ti->user_timer;
 271         lc->system_timer = ti->system_timer;
 272         lc->steal_timer = 0;
 273 }
 274
 275 static void pcpu_start_fn(struct pcpu *pcpu, void (*func)(void *), void *data)
 276 {
 277         struct lowcore *lc = pcpu->lowcore;
 278
 279         lc->restart_stack = lc->kernel_stack;
 280         lc->restart_fn = (unsigned long) func;
 281         lc->restart_data = (unsigned long) data;
 282         lc->restart_source = -1UL;
 283         pcpu_sigp_retry(pcpu, SIGP_RESTART, 0);
 284 }
 285
 286 /*
 287  * Call function via PSW restart on pcpu and stop the current cpu.
 288  */
 289 static void pcpu_delegate(struct pcpu *pcpu, void (*func)(void *),
 290                           void *data, unsigned long stack)
 291 {
 292         struct lowcore *lc = lowcore_ptr[pcpu - pcpu_devices];
 293         unsigned long source_cpu = stap();
 294
 295         __load_psw_mask(PSW_KERNEL_BITS);
 296         if (pcpu->address == source_cpu)
 297                 func(data);     /* should not return */
 298         /* Stop target cpu (if func returns this stops the current cpu). */
 299         pcpu_sigp_retry(pcpu, SIGP_STOP, 0);
 300         /* Restart func on the target cpu and stop the current cpu. */
 301         mem_assign_absolute(lc->restart_stack, stack);
 302         mem_assign_absolute(lc->restart_fn, (unsigned long) func);
 303         mem_assign_absolute(lc->restart_data, (unsigned long) data);
 304         mem_assign_absolute(lc->restart_source, source_cpu);
 305         asm volatile(
 306                 "0:     sigp    0,%0,%2 # sigp restart to target cpu\n"
 307                 "       brc     2,0b    # busy, try again\n"
 308                 "1:     sigp    0,%1,%3 # sigp stop to current cpu\n"
 309                 "       brc     2,1b    # busy, try again\n"
 310                 : : "d" (pcpu->address), "d" (source_cpu),
 311                     "K" (SIGP_RESTART), "K" (SIGP_STOP)
 312                 : "0", "1", "cc");
 313         for (;;) ;
 314 }
 315
 316 /*
 317  * Enable additional logical cpus for multi-threading.
 318  */
 319 static int pcpu_set_smt(unsigned int mtid)
 320 {
 321         int cc;
 322
 323         if (smp_cpu_mtid == mtid)
 324                 return 0;
 325         cc = __pcpu_sigp(0, SIGP_SET_MULTI_THREADING, mtid, NULL);
 326         if (cc == 0) {
 327                 smp_cpu_mtid = mtid;
 328                 smp_cpu_mt_shift = 0;
 329                 while (smp_cpu_mtid >= (1U << smp_cpu_mt_shift))
 330                         smp_cpu_mt_shift++;
 331                 pcpu_devices[0].address = stap();
 332         }
 333         return cc;
 334 }
 335
 336 /*
 337  * Call function on an online CPU.
 338  */
 339 void smp_call_online_cpu(void (*func)(void *), void *data)
 340 {
 341         struct pcpu *pcpu;
 342
 343         /* Use the current cpu if it is online. */
 344         pcpu = pcpu_find_address(cpu_online_mask, stap());
 345         if (!pcpu)
 346                 /* Use the first online cpu. */
 347                 pcpu = pcpu_devices + cpumask_first(cpu_online_mask);
 348         pcpu_delegate(pcpu, func, data, (unsigned long) restart_stack);
 349 }
 350
 351 /*
 352  * Call function on the ipl CPU.
 353  */
 354 void smp_call_ipl_cpu(void (*func)(void *), void *data)
 355 {
 356         pcpu_delegate(&pcpu_devices[0], func, data,
 357                       pcpu_devices->lowcore->panic_stack -
 358                       PANIC_FRAME_OFFSET + PAGE_SIZE);
 359 }
 360
 361 int smp_find_processor_id(u16 address)
 362 {
 363         int cpu;
 364
 365         for_each_present_cpu(cpu)
 366                 if (pcpu_devices[cpu].address == address)
 367                         return cpu;
 368         return -1;
 369 }
 370
 371 bool arch_vcpu_is_preempted(int cpu)
 372 {
 373         if (test_cpu_flag_of(CIF_ENABLED_WAIT, cpu))
 374                 return false;
 375         if (pcpu_running(pcpu_devices + cpu))
 376                 return false;
 377         return true;
 378 }
 379 EXPORT_SYMBOL(arch_vcpu_is_preempted);
 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 int smp_store_status(int cpu)
 544 {
 545         struct pcpu *pcpu = pcpu_devices + cpu;
 546         unsigned long pa;
 547
 548         pa = __pa(&pcpu->lowcore->floating_pt_save_area);
 549         if (__pcpu_sigp_relax(pcpu->address, SIGP_STORE_STATUS_AT_ADDRESS,
 550                               pa) != SIGP_CC_ORDER_CODE_ACCEPTED)
 551                 return -EIO;
 552         if (!MACHINE_HAS_VX)
 553                 return 0;
 554         pa = __pa(pcpu->lowcore->vector_save_area_addr);
 555         if (__pcpu_sigp_relax(pcpu->address, SIGP_STORE_ADDITIONAL_STATUS,
 556                               pa) != SIGP_CC_ORDER_CODE_ACCEPTED)
 557                 return -EIO;
 558         return 0;
 559 }
 560
 561 /*
 562  * Collect CPU state of the previous, crashed system.
 563  * There are four cases:
 564  * 1) standard zfcp dump
 565  *    condition: OLDMEM_BASE == NULL && ipl_info.type == IPL_TYPE_FCP_DUMP
 566  *    The state for all CPUs except the boot CPU needs to be collected
 567  *    with sigp stop-and-store-status. The boot CPU state is located in
 568  *    the absolute lowcore of the memory stored in the HSA. The zcore code
 569  *    will copy the boot CPU state from the HSA.
 570  * 2) stand-alone kdump for SCSI (zfcp dump with swapped memory)
 571  *    condition: OLDMEM_BASE != NULL && ipl_info.type == IPL_TYPE_FCP_DUMP
 572  *    The state for all CPUs except the boot CPU needs to be collected
 573  *    with sigp stop-and-store-status. The firmware or the boot-loader
 574  *    stored the registers of the boot CPU in the absolute lowcore in the
 575  *    memory of the old system.
 576  * 3) kdump and the old kernel did not store the CPU state,
 577  *    or stand-alone kdump for DASD
 578  *    condition: OLDMEM_BASE != NULL && !is_kdump_kernel()
 579  *    The state for all CPUs except the boot CPU needs to be collected
 580  *    with sigp stop-and-store-status. The kexec code or the boot-loader
 581  *    stored the registers of the boot CPU in the memory of the old system.
 582  * 4) kdump and the old kernel stored the CPU state
 583  *    condition: OLDMEM_BASE != NULL && is_kdump_kernel()
 584  *    This case does not exist for s390 anymore, setup_arch explicitly
 585  *    deactivates the elfcorehdr= kernel parameter
 586  */
 587 static __init void smp_save_cpu_vxrs(struct save_area *sa, u16 addr,
 588                                      bool is_boot_cpu, unsigned long page)
 589 {
 590         __vector128 *vxrs = (__vector128 *) page;
 591
 592         if (is_boot_cpu)
 593                 vxrs = boot_cpu_vector_save_area;
 594         else
 595                 __pcpu_sigp_relax(addr, SIGP_STORE_ADDITIONAL_STATUS, page);
 596         save_area_add_vxrs(sa, vxrs);
 597 }
 598
 599 static __init void smp_save_cpu_regs(struct save_area *sa, u16 addr,
 600                                      bool is_boot_cpu, unsigned long page)
 601 {
 602         void *regs = (void *) page;
 603
 604         if (is_boot_cpu)
 605                 copy_oldmem_kernel(regs, (void *) __LC_FPREGS_SAVE_AREA, 512);
 606         else
 607                 __pcpu_sigp_relax(addr, SIGP_STORE_STATUS_AT_ADDRESS, page);
 608         save_area_add_regs(sa, regs);
 609 }
 610
 611 void __init smp_save_dump_cpus(void)
 612 {
 613         int addr, boot_cpu_addr, max_cpu_addr;
 614         struct save_area *sa;
 615         unsigned long page;
 616         bool is_boot_cpu;
 617
 618         if (!(OLDMEM_BASE || ipl_info.type == IPL_TYPE_FCP_DUMP))
 619                 /* No previous system present, normal boot. */
 620                 return;
 621         /* Allocate a page as dumping area for the store status sigps */
 622         page = memblock_alloc_base(PAGE_SIZE, PAGE_SIZE, 1UL << 31);
 623         /* Set multi-threading state to the previous system. */
 624         pcpu_set_smt(sclp.mtid_prev);
 625         boot_cpu_addr = stap();
 626         max_cpu_addr = SCLP_MAX_CORES << sclp.mtid_prev;
 627         for (addr = 0; addr <= max_cpu_addr; addr++) {
 628                 if (__pcpu_sigp_relax(addr, SIGP_SENSE, 0) ==
 629                     SIGP_CC_NOT_OPERATIONAL)
 630                         continue;
 631                 is_boot_cpu = (addr == boot_cpu_addr);
 632                 /* Allocate save area */
 633                 sa = save_area_alloc(is_boot_cpu);
 634                 if (!sa)
 635                         panic("could not allocate memory for save area\n");
 636                 if (MACHINE_HAS_VX)
 637                         /* Get the vector registers */
 638                         smp_save_cpu_vxrs(sa, addr, is_boot_cpu, page);
 639                 /*
 640                  * For a zfcp dump OLDMEM_BASE == NULL and the registers
 641                  * of the boot CPU are stored in the HSA. To retrieve
 642                  * these registers an SCLP request is required which is
 643                  * done by drivers/s390/char/zcore.c:init_cpu_info()
 644                  */
 645                 if (!is_boot_cpu || OLDMEM_BASE)
 646                         /* Get the CPU registers */
 647                         smp_save_cpu_regs(sa, addr, is_boot_cpu, page);
 648         }
 649         memblock_free(page, PAGE_SIZE);
 650         diag308_reset();
 651         pcpu_set_smt(0);
 652 }
 653 #endif /* CONFIG_CRASH_DUMP */
 654
 655 void smp_cpu_set_polarization(int cpu, int val)
 656 {
 657         pcpu_devices[cpu].polarization = val;
 658 }
 659
 660 int smp_cpu_get_polarization(int cpu)
 661 {
 662         return pcpu_devices[cpu].polarization;
 663 }
 664
 665 static struct sclp_core_info *smp_get_core_info(void)
 666 {
 667         static int use_sigp_detection;
 668         struct sclp_core_info *info;
 669         int address;
 670
 671         info = kzalloc(sizeof(*info), GFP_KERNEL);
 672         if (info && (use_sigp_detection || sclp_get_core_info(info))) {
 673                 use_sigp_detection = 1;
 674                 for (address = 0;
 675                      address < (SCLP_MAX_CORES << smp_cpu_mt_shift);
 676                      address += (1U << smp_cpu_mt_shift)) {
 677                         if (__pcpu_sigp_relax(address, SIGP_SENSE, 0) ==
 678                             SIGP_CC_NOT_OPERATIONAL)
 679                                 continue;
 680                         info->core[info->configured].core_id =
 681                                 address >> smp_cpu_mt_shift;
 682                         info->configured++;
 683                 }
 684                 info->combined = info->configured;
 685         }
 686         return info;
 687 }
 688
 689 static int smp_add_present_cpu(int cpu);
 690
 691 static int __smp_rescan_cpus(struct sclp_core_info *info, int sysfs_add)
 692 {
 693         struct pcpu *pcpu;
 694         cpumask_t avail;
 695         int cpu, nr, i, j;
 696         u16 address;
 697
 698         nr = 0;
 699         cpumask_xor(&avail, cpu_possible_mask, cpu_present_mask);
 700         cpu = cpumask_first(&avail);
 701         for (i = 0; (i < info->combined) && (cpu < nr_cpu_ids); i++) {
 702                 if (sclp.has_core_type && info->core[i].type != boot_core_type)
 703                         continue;
 704                 address = info->core[i].core_id << smp_cpu_mt_shift;
 705                 for (j = 0; j <= smp_cpu_mtid; j++) {
 706                         if (pcpu_find_address(cpu_present_mask, address + j))
 707                                 continue;
 708                         pcpu = pcpu_devices + cpu;
 709                         pcpu->address = address + j;
 710                         pcpu->state =
 711                                 (cpu >= info->configured*(smp_cpu_mtid + 1)) ?
 712                                 CPU_STATE_STANDBY : CPU_STATE_CONFIGURED;
 713                         smp_cpu_set_polarization(cpu, POLARIZATION_UNKNOWN);
 714                         set_cpu_present(cpu, true);
 715                         if (sysfs_add && smp_add_present_cpu(cpu) != 0)
 716                                 set_cpu_present(cpu, false);
 717                         else
 718                                 nr++;
 719                         cpu = cpumask_next(cpu, &avail);
 720                         if (cpu >= nr_cpu_ids)
 721                                 break;
 722                 }
 723         }
 724         return nr;
 725 }
 726
 727 static void __init smp_detect_cpus(void)
 728 {
 729         unsigned int cpu, mtid, c_cpus, s_cpus;
 730         struct sclp_core_info *info;
 731         u16 address;
 732
 733         /* Get CPU information */
 734         info = smp_get_core_info();
 735         if (!info)
 736                 panic("smp_detect_cpus failed to allocate memory\n");
 737
 738         /* Find boot CPU type */
 739         if (sclp.has_core_type) {
 740                 address = stap();
 741                 for (cpu = 0; cpu < info->combined; cpu++)
 742                         if (info->core[cpu].core_id == address) {
 743                                 /* The boot cpu dictates the cpu type. */
 744                                 boot_core_type = info->core[cpu].type;
 745                                 break;
 746                         }
 747                 if (cpu >= info->combined)
 748                         panic("Could not find boot CPU type");
 749         }
 750
 751         /* Set multi-threading state for the current system */
 752         mtid = boot_core_type ? sclp.mtid : sclp.mtid_cp;
 753         mtid = (mtid < smp_max_threads) ? mtid : smp_max_threads - 1;
 754         pcpu_set_smt(mtid);
 755
 756         /* Print number of CPUs */
 757         c_cpus = s_cpus = 0;
 758         for (cpu = 0; cpu < info->combined; cpu++) {
 759                 if (sclp.has_core_type &&
 760                     info->core[cpu].type != boot_core_type)
 761                         continue;
 762                 if (cpu < info->configured)
 763                         c_cpus += smp_cpu_mtid + 1;
 764                 else
 765                         s_cpus += smp_cpu_mtid + 1;
 766         }
 767         pr_info("%d configured CPUs, %d standby CPUs\n", c_cpus, s_cpus);
 768
 769         /* Add CPUs present at boot */
 770         get_online_cpus();
 771         __smp_rescan_cpus(info, 0);
 772         put_online_cpus();
 773         kfree(info);
 774 }
 775
 776 /*
 777  *      Activate a secondary processor.
 778  */
 779 static void smp_start_secondary(void *cpuvoid)
 780 {
 781         S390_lowcore.last_update_clock = get_tod_clock();
 782         S390_lowcore.restart_stack = (unsigned long) restart_stack;
 783         S390_lowcore.restart_fn = (unsigned long) do_restart;
 784         S390_lowcore.restart_data = 0;
 785         S390_lowcore.restart_source = -1UL;
 786         restore_access_regs(S390_lowcore.access_regs_save_area);
 787         __ctl_load(S390_lowcore.cregs_save_area, 0, 15);
 788         __load_psw_mask(PSW_KERNEL_BITS | PSW_MASK_DAT);
 789         cpu_init();
 790         preempt_disable();
 791         init_cpu_timer();
 792         vtime_init();
 793         pfault_init();
 794         notify_cpu_starting(smp_processor_id());
 795         set_cpu_online(smp_processor_id(), true);
 796         inc_irq_stat(CPU_RST);
 797         local_irq_enable();
 798         cpu_startup_entry(CPUHP_AP_ONLINE_IDLE);
 799 }
 800
 801 /* Upping and downing of CPUs */
 802 int __cpu_up(unsigned int cpu, struct task_struct *tidle)
 803 {
 804         struct pcpu *pcpu;
 805         int base, i, rc;
 806
 807         pcpu = pcpu_devices + cpu;
 808         if (pcpu->state != CPU_STATE_CONFIGURED)
 809                 return -EIO;
 810         base = cpu - (cpu % (smp_cpu_mtid + 1));
 811         for (i = 0; i <= smp_cpu_mtid; i++) {
 812                 if (base + i < nr_cpu_ids)
 813                         if (cpu_online(base + i))
 814                                 break;
 815         }
 816         /*
 817          * If this is the first CPU of the core to get online
 818          * do an initial CPU reset.
 819          */
 820         if (i > smp_cpu_mtid &&
 821             pcpu_sigp_retry(pcpu_devices + base, SIGP_INITIAL_CPU_RESET, 0) !=
 822             SIGP_CC_ORDER_CODE_ACCEPTED)
 823                 return -EIO;
 824
 825         rc = pcpu_alloc_lowcore(pcpu, cpu);
 826         if (rc)
 827                 return rc;
 828         pcpu_prepare_secondary(pcpu, cpu);
 829         pcpu_attach_task(pcpu, tidle);
 830         pcpu_start_fn(pcpu, smp_start_secondary, NULL);
 831         /* Wait until cpu puts itself in the online & active maps */
 832         while (!cpu_online(cpu))
 833                 cpu_relax();
 834         return 0;
 835 }
 836
 837 static unsigned int setup_possible_cpus __initdata;
 838
 839 static int __init _setup_possible_cpus(char *s)
 840 {
 841         get_option(&s, &setup_possible_cpus);
 842         return 0;
 843 }
 844 early_param("possible_cpus", _setup_possible_cpus);
 845
 846 #ifdef CONFIG_HOTPLUG_CPU
 847
 848 int __cpu_disable(void)
 849 {
 850         unsigned long cregs[16];
 851
 852         /* Handle possible pending IPIs */
 853         smp_handle_ext_call();
 854         set_cpu_online(smp_processor_id(), false);
 855         /* Disable pseudo page faults on this cpu. */
 856         pfault_fini();
 857         /* Disable interrupt sources via control register. */
 858         __ctl_store(cregs, 0, 15);
 859         cregs[0]  &= ~0x0000ee70UL;     /* disable all external interrupts */
 860         cregs[6]  &= ~0xff000000UL;     /* disable all I/O interrupts */
 861         cregs[14] &= ~0x1f000000UL;     /* disable most machine checks */
 862         __ctl_load(cregs, 0, 15);
 863         clear_cpu_flag(CIF_NOHZ_DELAY);
 864         return 0;
 865 }
 866
 867 void __cpu_die(unsigned int cpu)
 868 {
 869         struct pcpu *pcpu;
 870
 871         /* Wait until target cpu is down */
 872         pcpu = pcpu_devices + cpu;
 873         while (!pcpu_stopped(pcpu))
 874                 cpu_relax();
 875         pcpu_free_lowcore(pcpu);
 876         cpumask_clear_cpu(cpu, mm_cpumask(&init_mm));
 877         cpumask_clear_cpu(cpu, &init_mm.context.cpu_attach_mask);
 878 }
 879
 880 void __noreturn cpu_die(void)
 881 {
 882         idle_task_exit();
 883         pcpu_sigp_retry(pcpu_devices + smp_processor_id(), SIGP_STOP, 0);
 884         for (;;) ;
 885 }
 886
 887 #endif /* CONFIG_HOTPLUG_CPU */
 888
 889 void __init smp_fill_possible_mask(void)
 890 {
 891         unsigned int possible, sclp_max, cpu;
 892
 893         sclp_max = max(sclp.mtid, sclp.mtid_cp) + 1;
 894         sclp_max = min(smp_max_threads, sclp_max);
 895         sclp_max = (sclp.max_cores * sclp_max) ?: nr_cpu_ids;
 896         possible = setup_possible_cpus ?: nr_cpu_ids;
 897         possible = min(possible, sclp_max);
 898         for (cpu = 0; cpu < possible && cpu < nr_cpu_ids; cpu++)
 899                 set_cpu_possible(cpu, true);
 900 }
 901
 902 void __init smp_prepare_cpus(unsigned int max_cpus)
 903 {
 904         /* request the 0x1201 emergency signal external interrupt */
 905         if (register_external_irq(EXT_IRQ_EMERGENCY_SIG, do_ext_call_interrupt))
 906                 panic("Couldn't request external interrupt 0x1201");
 907         /* request the 0x1202 external call external interrupt */
 908         if (register_external_irq(EXT_IRQ_EXTERNAL_CALL, do_ext_call_interrupt))
 909                 panic("Couldn't request external interrupt 0x1202");
 910         smp_detect_cpus();
 911 }
 912
 913 void __init smp_prepare_boot_cpu(void)
 914 {
 915         struct pcpu *pcpu = pcpu_devices;
 916
 917         pcpu->state = CPU_STATE_CONFIGURED;
 918         pcpu->address = stap();
 919         pcpu->lowcore = (struct lowcore *)(unsigned long) store_prefix();
 920         S390_lowcore.percpu_offset = __per_cpu_offset[0];
 921         smp_cpu_set_polarization(0, POLARIZATION_UNKNOWN);
 922         set_cpu_present(0, true);
 923         set_cpu_online(0, true);
 924 }
 925
 926 void __init smp_cpus_done(unsigned int max_cpus)
 927 {
 928 }
 929
 930 void __init smp_setup_processor_id(void)
 931 {
 932         S390_lowcore.cpu_nr = 0;
 933         S390_lowcore.spinlock_lockval = arch_spin_lockval(0);
 934 }
 935
 936 /*
 937  * the frequency of the profiling timer can be changed
 938  * by writing a multiplier value into /proc/profile.
 939  *
 940  * usually you want to run this on all CPUs ;)
 941  */
 942 int setup_profiling_timer(unsigned int multiplier)
 943 {
 944         return 0;
 945 }
 946
 947 #ifdef CONFIG_HOTPLUG_CPU
 948 static ssize_t cpu_configure_show(struct device *dev,
 949                                   struct device_attribute *attr, char *buf)
 950 {
 951         ssize_t count;
 952
 953         mutex_lock(&smp_cpu_state_mutex);
 954         count = sprintf(buf, "%d\n", pcpu_devices[dev->id].state);
 955         mutex_unlock(&smp_cpu_state_mutex);
 956         return count;
 957 }
 958
 959 static ssize_t cpu_configure_store(struct device *dev,
 960                                    struct device_attribute *attr,
 961                                    const char *buf, size_t count)
 962 {
 963         struct pcpu *pcpu;
 964         int cpu, val, rc, i;
 965         char delim;
 966
 967         if (sscanf(buf, "%d %c", &val, &delim) != 1)
 968                 return -EINVAL;
 969         if (val != 0 && val != 1)
 970                 return -EINVAL;
 971         get_online_cpus();
 972         mutex_lock(&smp_cpu_state_mutex);
 973         rc = -EBUSY;
 974         /* disallow configuration changes of online cpus and cpu 0 */
 975         cpu = dev->id;
 976         cpu -= cpu % (smp_cpu_mtid + 1);
 977         if (cpu == 0)
 978                 goto out;
 979         for (i = 0; i <= smp_cpu_mtid; i++)
 980                 if (cpu_online(cpu + i))
 981                         goto out;
 982         pcpu = pcpu_devices + cpu;
 983         rc = 0;
 984         switch (val) {
 985         case 0:
 986                 if (pcpu->state != CPU_STATE_CONFIGURED)
 987                         break;
 988                 rc = sclp_core_deconfigure(pcpu->address >> smp_cpu_mt_shift);
 989                 if (rc)
 990                         break;
 991                 for (i = 0; i <= smp_cpu_mtid; i++) {
 992                         if (cpu + i >= nr_cpu_ids || !cpu_present(cpu + i))
 993                                 continue;
 994                         pcpu[i].state = CPU_STATE_STANDBY;
 995                         smp_cpu_set_polarization(cpu + i,
 996                                                  POLARIZATION_UNKNOWN);
 997                 }
 998                 topology_expect_change();
 999                 break;
1000         case 1:
1001                 if (pcpu->state != CPU_STATE_STANDBY)
1002                         break;
1003                 rc = sclp_core_configure(pcpu->address >> smp_cpu_mt_shift);
1004                 if (rc)
1005                         break;
1006                 for (i = 0; i <= smp_cpu_mtid; i++) {
1007                         if (cpu + i >= nr_cpu_ids || !cpu_present(cpu + i))
1008                                 continue;
1009                         pcpu[i].state = CPU_STATE_CONFIGURED;
1010                         smp_cpu_set_polarization(cpu + i,
1011                                                  POLARIZATION_UNKNOWN);
1012                 }
1013                 topology_expect_change();
1014                 break;
1015         default:
1016                 break;
1017         }
1018 out:
1019         mutex_unlock(&smp_cpu_state_mutex);
1020         put_online_cpus();
1021         return rc ? rc : count;
1022 }
1023 static DEVICE_ATTR(configure, 0644, cpu_configure_show, cpu_configure_store);
1024 #endif /* CONFIG_HOTPLUG_CPU */
1025
1026 static ssize_t show_cpu_address(struct device *dev,
1027                                 struct device_attribute *attr, char *buf)
1028 {
1029         return sprintf(buf, "%d\n", pcpu_devices[dev->id].address);
1030 }
1031 static DEVICE_ATTR(address, 0444, show_cpu_address, NULL);
1032
1033 static struct attribute *cpu_common_attrs[] = {
1034 #ifdef CONFIG_HOTPLUG_CPU
1035         &dev_attr_configure.attr,
1036 #endif
1037         &dev_attr_address.attr,
1038         NULL,
1039 };
1040
1041 static struct attribute_group cpu_common_attr_group = {
1042         .attrs = cpu_common_attrs,
1043 };
1044
1045 static struct attribute *cpu_online_attrs[] = {
1046         &dev_attr_idle_count.attr,
1047         &dev_attr_idle_time_us.attr,
1048         NULL,
1049 };
1050
1051 static struct attribute_group cpu_online_attr_group = {
1052         .attrs = cpu_online_attrs,
1053 };
1054
1055 static int smp_cpu_online(unsigned int cpu)
1056 {
1057         struct device *s = &per_cpu(cpu_device, cpu)->dev;
1058
1059         return sysfs_create_group(&s->kobj, &cpu_online_attr_group);
1060 }
1061 static int smp_cpu_pre_down(unsigned int cpu)
1062 {
1063         struct device *s = &per_cpu(cpu_device, cpu)->dev;
1064
1065         sysfs_remove_group(&s->kobj, &cpu_online_attr_group);
1066         return 0;
1067 }
1068
1069 static int smp_add_present_cpu(int cpu)
1070 {
1071         struct device *s;
1072         struct cpu *c;
1073         int rc;
1074
1075         c = kzalloc(sizeof(*c), GFP_KERNEL);
1076         if (!c)
1077                 return -ENOMEM;
1078         per_cpu(cpu_device, cpu) = c;
1079         s = &c->dev;
1080         c->hotpluggable = 1;
1081         rc = register_cpu(c, cpu);
1082         if (rc)
1083                 goto out;
1084         rc = sysfs_create_group(&s->kobj, &cpu_common_attr_group);
1085         if (rc)
1086                 goto out_cpu;
1087         rc = topology_cpu_init(c);
1088         if (rc)
1089                 goto out_topology;
1090         return 0;
1091
1092 out_topology:
1093         sysfs_remove_group(&s->kobj, &cpu_common_attr_group);
1094 out_cpu:
1095 #ifdef CONFIG_HOTPLUG_CPU
1096         unregister_cpu(c);
1097 #endif
1098 out:
1099         return rc;
1100 }
1101
1102 #ifdef CONFIG_HOTPLUG_CPU
1103
1104 int __ref smp_rescan_cpus(void)
1105 {
1106         struct sclp_core_info *info;
1107         int nr;
1108
1109         info = smp_get_core_info();
1110         if (!info)
1111                 return -ENOMEM;
1112         get_online_cpus();
1113         mutex_lock(&smp_cpu_state_mutex);
1114         nr = __smp_rescan_cpus(info, 1);
1115         mutex_unlock(&smp_cpu_state_mutex);
1116         put_online_cpus();
1117         kfree(info);
1118         if (nr)
1119                 topology_schedule_update();
1120         return 0;
1121 }
1122
1123 static ssize_t __ref rescan_store(struct device *dev,
1124                                   struct device_attribute *attr,
1125                                   const char *buf,
1126                                   size_t count)
1127 {
1128         int rc;
1129
1130         rc = smp_rescan_cpus();
1131         return rc ? rc : count;
1132 }
1133 static DEVICE_ATTR(rescan, 0200, NULL, rescan_store);
1134 #endif /* CONFIG_HOTPLUG_CPU */
1135
1136 static int __init s390_smp_init(void)
1137 {
1138         int cpu, rc = 0;
1139
1140 #ifdef CONFIG_HOTPLUG_CPU
1141         rc = device_create_file(cpu_subsys.dev_root, &dev_attr_rescan);
1142         if (rc)
1143                 return rc;
1144 #endif
1145         for_each_present_cpu(cpu) {
1146                 rc = smp_add_present_cpu(cpu);
1147                 if (rc)
1148                         goto out;
1149         }
1150
1151         rc = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "s390/smp:online",
1152                                smp_cpu_online, smp_cpu_pre_down);
1153 out:
1154         return rc;
1155 }
1156 subsys_initcall(s390_smp_init);