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/bootmem.h>
  23 #include <linux/export.h>
  24 #include <linux/init.h>
  25 #include <linux/mm.h>
  26 #include <linux/err.h>
  27 #include <linux/spinlock.h>
  28 #include <linux/kernel_stat.h>
  29 #include <linux/kmemleak.h>
  30 #include <linux/delay.h>
  31 #include <linux/interrupt.h>
  32 #include <linux/irqflags.h>
  33 #include <linux/cpu.h>
  34 #include <linux/slab.h>
  35 #include <linux/sched/hotplug.h>
  36 #include <linux/sched/task_stack.h>
  37 #include <linux/crash_dump.h>
  38 #include <linux/memblock.h>
  39 #include <asm/asm-offsets.h>
  40 #include <asm/diag.h>
  41 #include <asm/switch_to.h>
  42 #include <asm/facility.h>
  43 #include <asm/ipl.h>
  44 #include <asm/setup.h>
  45 #include <asm/irq.h>
  46 #include <asm/tlbflush.h>
  47 #include <asm/vtimer.h>
  48 #include <asm/lowcore.h>
  49 #include <asm/sclp.h>
  50 #include <asm/vdso.h>
  51 #include <asm/debug.h>
  52 #include <asm/os_info.h>
  53 #include <asm/sigp.h>
  54 #include <asm/idle.h>
  55 #include <asm/nmi.h>
  56 #include "entry.h"
  57
  58 enum {
  59         ec_schedule = 0,
  60         ec_call_function_single,
  61         ec_stop_cpu,
  62 };
  63
  64 enum {
  65         CPU_STATE_STANDBY,
  66         CPU_STATE_CONFIGURED,
  67 };
  68
  69 static DEFINE_PER_CPU(struct cpu *, cpu_device);
  70
  71 struct pcpu {
  72         struct lowcore *lowcore;        /* lowcore page(s) for the cpu */
  73         unsigned long ec_mask;          /* bit mask for ec_xxx functions */
  74         unsigned long ec_clk;           /* sigp timestamp for ec_xxx */
  75         signed char state;              /* physical cpu state */
  76         signed char polarization;       /* physical polarization */
  77         u16 address;                    /* physical cpu address */
  78 };
  79
  80 static u8 boot_core_type;
  81 static struct pcpu pcpu_devices[NR_CPUS];
  82
  83 static struct kmem_cache *pcpu_mcesa_cache;
  84
  85 unsigned int smp_cpu_mt_shift;
  86 EXPORT_SYMBOL(smp_cpu_mt_shift);
  87
  88 unsigned int smp_cpu_mtid;
  89 EXPORT_SYMBOL(smp_cpu_mtid);
  90
  91 #ifdef CONFIG_CRASH_DUMP
  92 __vector128 __initdata boot_cpu_vector_save_area[__NUM_VXRS];
  93 #endif
  94
  95 static unsigned int smp_max_threads __initdata = -1U;
  96
  97 static int __init early_nosmt(char *s)
  98 {
  99         smp_max_threads = 1;
 100         return 0;
 101 }
 102 early_param("nosmt", early_nosmt);
 103
 104 static int __init early_smt(char *s)
 105 {
 106         get_option(&s, &smp_max_threads);
 107         return 0;
 108 }
 109 early_param("smt", early_smt);
 110
 111 /*
 112  * The smp_cpu_state_mutex must be held when changing the state or polarization
 113  * member of a pcpu data structure within the pcpu_devices arreay.
 114  */
 115 DEFINE_MUTEX(smp_cpu_state_mutex);
 116
 117 /*
 118  * Signal processor helper functions.
 119  */
 120 static inline int __pcpu_sigp_relax(u16 addr, u8 order, unsigned long parm)
 121 {
 122         int cc;
 123
 124         while (1) {
 125                 cc = __pcpu_sigp(addr, order, parm, NULL);
 126                 if (cc != SIGP_CC_BUSY)
 127                         return cc;
 128                 cpu_relax();
 129         }
 130 }
 131
 132 static int pcpu_sigp_retry(struct pcpu *pcpu, u8 order, u32 parm)
 133 {
 134         int cc, retry;
 135
 136         for (retry = 0; ; retry++) {
 137                 cc = __pcpu_sigp(pcpu->address, order, parm, NULL);
 138                 if (cc != SIGP_CC_BUSY)
 139                         break;
 140                 if (retry >= 3)
 141                         udelay(10);
 142         }
 143         return cc;
 144 }
 145
 146 static inline int pcpu_stopped(struct pcpu *pcpu)
 147 {
 148         u32 uninitialized_var(status);
 149
 150         if (__pcpu_sigp(pcpu->address, SIGP_SENSE,
 151                         0, &status) != SIGP_CC_STATUS_STORED)
 152                 return 0;
 153         return !!(status & (SIGP_STATUS_CHECK_STOP|SIGP_STATUS_STOPPED));
 154 }
 155
 156 static inline int pcpu_running(struct pcpu *pcpu)
 157 {
 158         if (__pcpu_sigp(pcpu->address, SIGP_SENSE_RUNNING,
 159                         0, NULL) != SIGP_CC_STATUS_STORED)
 160                 return 1;
 161         /* Status stored condition code is equivalent to cpu not running. */
 162         return 0;
 163 }
 164
 165 /*
 166  * Find struct pcpu by cpu address.
 167  */
 168 static struct pcpu *pcpu_find_address(const struct cpumask *mask, u16 address)
 169 {
 170         int cpu;
 171
 172         for_each_cpu(cpu, mask)
 173                 if (pcpu_devices[cpu].address == address)
 174                         return pcpu_devices + cpu;
 175         return NULL;
 176 }
 177
 178 static void pcpu_ec_call(struct pcpu *pcpu, int ec_bit)
 179 {
 180         int order;
 181
 182         if (test_and_set_bit(ec_bit, &pcpu->ec_mask))
 183                 return;
 184         order = pcpu_running(pcpu) ? SIGP_EXTERNAL_CALL : SIGP_EMERGENCY_SIGNAL;
 185         pcpu->ec_clk = get_tod_clock_fast();
 186         pcpu_sigp_retry(pcpu, order, 0);
 187 }
 188
 189 #define ASYNC_FRAME_OFFSET (ASYNC_SIZE - STACK_FRAME_OVERHEAD - __PT_SIZE)
 190 #define PANIC_FRAME_OFFSET (PAGE_SIZE - STACK_FRAME_OVERHEAD - __PT_SIZE)
 191
 192 static int pcpu_alloc_lowcore(struct pcpu *pcpu, int cpu)
 193 {
 194         unsigned long async_stack, panic_stack;
 195         unsigned long mcesa_origin, mcesa_bits;
 196         struct lowcore *lc;
 197
 198         mcesa_origin = mcesa_bits = 0;
 199         if (pcpu != &pcpu_devices[0]) {
 200                 pcpu->lowcore = (struct lowcore *)
 201                         __get_free_pages(GFP_KERNEL | GFP_DMA, LC_ORDER);
 202                 async_stack = __get_free_pages(GFP_KERNEL, ASYNC_ORDER);
 203                 panic_stack = __get_free_page(GFP_KERNEL);
 204                 if (!pcpu->lowcore || !panic_stack || !async_stack)
 205                         goto out;
 206                 if (MACHINE_HAS_VX || MACHINE_HAS_GS) {
 207                         mcesa_origin = (unsigned long)
 208                                 kmem_cache_alloc(pcpu_mcesa_cache, GFP_KERNEL);
 209                         if (!mcesa_origin)
 210                                 goto out;
 211                         /* The pointer is stored with mcesa_bits ORed in */
 212                         kmemleak_not_leak((void *) mcesa_origin);
 213                         mcesa_bits = MACHINE_HAS_GS ? 11 : 0;
 214                 }
 215         } else {
 216                 async_stack = pcpu->lowcore->async_stack - ASYNC_FRAME_OFFSET;
 217                 panic_stack = pcpu->lowcore->panic_stack - PANIC_FRAME_OFFSET;
 218                 mcesa_origin = pcpu->lowcore->mcesad & MCESA_ORIGIN_MASK;
 219                 mcesa_bits = pcpu->lowcore->mcesad & MCESA_LC_MASK;
 220         }
 221         lc = pcpu->lowcore;
 222         memcpy(lc, &S390_lowcore, 512);
 223         memset((char *) lc + 512, 0, sizeof(*lc) - 512);
 224         lc->async_stack = async_stack + ASYNC_FRAME_OFFSET;
 225         lc->panic_stack = panic_stack + PANIC_FRAME_OFFSET;
 226         lc->mcesad = mcesa_origin | mcesa_bits;
 227         lc->cpu_nr = cpu;
 228         lc->spinlock_lockval = arch_spin_lockval(cpu);
 229         if (vdso_alloc_per_cpu(lc))
 230                 goto out;
 231         lowcore_ptr[cpu] = lc;
 232         pcpu_sigp_retry(pcpu, SIGP_SET_PREFIX, (u32)(unsigned long) lc);
 233         return 0;
 234 out:
 235         if (pcpu != &pcpu_devices[0]) {
 236                 if (mcesa_origin)
 237                         kmem_cache_free(pcpu_mcesa_cache,
 238                                         (void *) mcesa_origin);
 239                 free_page(panic_stack);
 240                 free_pages(async_stack, ASYNC_ORDER);
 241                 free_pages((unsigned long) pcpu->lowcore, LC_ORDER);
 242         }
 243         return -ENOMEM;
 244 }
 245
 246 #ifdef CONFIG_HOTPLUG_CPU
 247
 248 static void pcpu_free_lowcore(struct pcpu *pcpu)
 249 {
 250         unsigned long mcesa_origin;
 251
 252         pcpu_sigp_retry(pcpu, SIGP_SET_PREFIX, 0);
 253         lowcore_ptr[pcpu - pcpu_devices] = NULL;
 254         vdso_free_per_cpu(pcpu->lowcore);
 255         if (pcpu == &pcpu_devices[0])
 256                 return;
 257         if (MACHINE_HAS_VX || MACHINE_HAS_GS) {
 258                 mcesa_origin = pcpu->lowcore->mcesad & MCESA_ORIGIN_MASK;
 259                 kmem_cache_free(pcpu_mcesa_cache, (void *) mcesa_origin);
 260         }
 261         free_page(pcpu->lowcore->panic_stack-PANIC_FRAME_OFFSET);
 262         free_pages(pcpu->lowcore->async_stack-ASYNC_FRAME_OFFSET, ASYNC_ORDER);
 263         free_pages((unsigned long) pcpu->lowcore, LC_ORDER);
 264 }
 265
 266 #endif /* CONFIG_HOTPLUG_CPU */
 267
 268 static void pcpu_prepare_secondary(struct pcpu *pcpu, int cpu)
 269 {
 270         struct lowcore *lc = pcpu->lowcore;
 271
 272         cpumask_set_cpu(cpu, &init_mm.context.cpu_attach_mask);
 273         cpumask_set_cpu(cpu, mm_cpumask(&init_mm));
 274         lc->cpu_nr = cpu;
 275         lc->spinlock_lockval = arch_spin_lockval(cpu);
 276         lc->percpu_offset = __per_cpu_offset[cpu];
 277         lc->kernel_asce = S390_lowcore.kernel_asce;
 278         lc->machine_flags = S390_lowcore.machine_flags;
 279         lc->user_timer = lc->system_timer = lc->steal_timer = 0;
 280         __ctl_store(lc->cregs_save_area, 0, 15);
 281         save_access_regs((unsigned int *) lc->access_regs_save_area);
 282         memcpy(lc->stfle_fac_list, S390_lowcore.stfle_fac_list,
 283                MAX_FACILITY_BIT/8);
 284 }
 285
 286 static void pcpu_attach_task(struct pcpu *pcpu, struct task_struct *tsk)
 287 {
 288         struct lowcore *lc = pcpu->lowcore;
 289
 290         lc->kernel_stack = (unsigned long) task_stack_page(tsk)
 291                 + THREAD_SIZE - STACK_FRAME_OVERHEAD - sizeof(struct pt_regs);
 292         lc->current_task = (unsigned long) tsk;
 293         lc->lpp = LPP_MAGIC;
 294         lc->current_pid = tsk->pid;
 295         lc->user_timer = tsk->thread.user_timer;
 296         lc->guest_timer = tsk->thread.guest_timer;
 297         lc->system_timer = tsk->thread.system_timer;
 298         lc->hardirq_timer = tsk->thread.hardirq_timer;
 299         lc->softirq_timer = tsk->thread.softirq_timer;
 300         lc->steal_timer = 0;
 301 }
 302
 303 static void pcpu_start_fn(struct pcpu *pcpu, void (*func)(void *), void *data)
 304 {
 305         struct lowcore *lc = pcpu->lowcore;
 306
 307         lc->restart_stack = lc->kernel_stack;
 308         lc->restart_fn = (unsigned long) func;
 309         lc->restart_data = (unsigned long) data;
 310         lc->restart_source = -1UL;
 311         pcpu_sigp_retry(pcpu, SIGP_RESTART, 0);
 312 }
 313
 314 /*
 315  * Call function via PSW restart on pcpu and stop the current cpu.
 316  */
 317 static void pcpu_delegate(struct pcpu *pcpu, void (*func)(void *),
 318                           void *data, unsigned long stack)
 319 {
 320         struct lowcore *lc = lowcore_ptr[pcpu - pcpu_devices];
 321         unsigned long source_cpu = stap();
 322
 323         __load_psw_mask(PSW_KERNEL_BITS);
 324         if (pcpu->address == source_cpu)
 325                 func(data);     /* should not return */
 326         /* Stop target cpu (if func returns this stops the current cpu). */
 327         pcpu_sigp_retry(pcpu, SIGP_STOP, 0);
 328         /* Restart func on the target cpu and stop the current cpu. */
 329         mem_assign_absolute(lc->restart_stack, stack);
 330         mem_assign_absolute(lc->restart_fn, (unsigned long) func);
 331         mem_assign_absolute(lc->restart_data, (unsigned long) data);
 332         mem_assign_absolute(lc->restart_source, source_cpu);
 333         asm volatile(
 334                 "0:     sigp    0,%0,%2 # sigp restart to target cpu\n"
 335                 "       brc     2,0b    # busy, try again\n"
 336                 "1:     sigp    0,%1,%3 # sigp stop to current cpu\n"
 337                 "       brc     2,1b    # busy, try again\n"
 338                 : : "d" (pcpu->address), "d" (source_cpu),
 339                     "K" (SIGP_RESTART), "K" (SIGP_STOP)
 340                 : "0", "1", "cc");
 341         for (;;) ;
 342 }
 343
 344 /*
 345  * Enable additional logical cpus for multi-threading.
 346  */
 347 static int pcpu_set_smt(unsigned int mtid)
 348 {
 349         int cc;
 350
 351         if (smp_cpu_mtid == mtid)
 352                 return 0;
 353         cc = __pcpu_sigp(0, SIGP_SET_MULTI_THREADING, mtid, NULL);
 354         if (cc == 0) {
 355                 smp_cpu_mtid = mtid;
 356                 smp_cpu_mt_shift = 0;
 357                 while (smp_cpu_mtid >= (1U << smp_cpu_mt_shift))
 358                         smp_cpu_mt_shift++;
 359                 pcpu_devices[0].address = stap();
 360         }
 361         return cc;
 362 }
 363
 364 /*
 365  * Call function on an online CPU.
 366  */
 367 void smp_call_online_cpu(void (*func)(void *), void *data)
 368 {
 369         struct pcpu *pcpu;
 370
 371         /* Use the current cpu if it is online. */
 372         pcpu = pcpu_find_address(cpu_online_mask, stap());
 373         if (!pcpu)
 374                 /* Use the first online cpu. */
 375                 pcpu = pcpu_devices + cpumask_first(cpu_online_mask);
 376         pcpu_delegate(pcpu, func, data, (unsigned long) restart_stack);
 377 }
 378
 379 /*
 380  * Call function on the ipl CPU.
 381  */
 382 void smp_call_ipl_cpu(void (*func)(void *), void *data)
 383 {
 384         pcpu_delegate(&pcpu_devices[0], func, data,
 385                       pcpu_devices->lowcore->panic_stack -
 386                       PANIC_FRAME_OFFSET + PAGE_SIZE);
 387 }
 388
 389 int smp_find_processor_id(u16 address)
 390 {
 391         int cpu;
 392
 393         for_each_present_cpu(cpu)
 394                 if (pcpu_devices[cpu].address == address)
 395                         return cpu;
 396         return -1;
 397 }
 398
 399 bool arch_vcpu_is_preempted(int cpu)
 400 {
 401         if (test_cpu_flag_of(CIF_ENABLED_WAIT, cpu))
 402                 return false;
 403         if (pcpu_running(pcpu_devices + cpu))
 404                 return false;
 405         return true;
 406 }
 407 EXPORT_SYMBOL(arch_vcpu_is_preempted);
 408
 409 void smp_yield_cpu(int cpu)
 410 {
 411         if (MACHINE_HAS_DIAG9C) {
 412                 diag_stat_inc_norecursion(DIAG_STAT_X09C);
 413                 asm volatile("diag %0,0,0x9c"
 414                              : : "d" (pcpu_devices[cpu].address));
 415         } else if (MACHINE_HAS_DIAG44) {
 416                 diag_stat_inc_norecursion(DIAG_STAT_X044);
 417                 asm volatile("diag 0,0,0x44");
 418         }
 419 }
 420
 421 /*
 422  * Send cpus emergency shutdown signal. This gives the cpus the
 423  * opportunity to complete outstanding interrupts.
 424  */
 425 static void smp_emergency_stop(cpumask_t *cpumask)
 426 {
 427         u64 end;
 428         int cpu;
 429
 430         end = get_tod_clock() + (1000000UL << 12);
 431         for_each_cpu(cpu, cpumask) {
 432                 struct pcpu *pcpu = pcpu_devices + cpu;
 433                 set_bit(ec_stop_cpu, &pcpu->ec_mask);
 434                 while (__pcpu_sigp(pcpu->address, SIGP_EMERGENCY_SIGNAL,
 435                                    0, NULL) == SIGP_CC_BUSY &&
 436                        get_tod_clock() < end)
 437                         cpu_relax();
 438         }
 439         while (get_tod_clock() < end) {
 440                 for_each_cpu(cpu, cpumask)
 441                         if (pcpu_stopped(pcpu_devices + cpu))
 442                                 cpumask_clear_cpu(cpu, cpumask);
 443                 if (cpumask_empty(cpumask))
 444                         break;
 445                 cpu_relax();
 446         }
 447 }
 448
 449 /*
 450  * Stop all cpus but the current one.
 451  */
 452 void smp_send_stop(void)
 453 {
 454         cpumask_t cpumask;
 455         int cpu;
 456
 457         /* Disable all interrupts/machine checks */
 458         __load_psw_mask(PSW_KERNEL_BITS | PSW_MASK_DAT);
 459         trace_hardirqs_off();
 460
 461         debug_set_critical();
 462         cpumask_copy(&cpumask, cpu_online_mask);
 463         cpumask_clear_cpu(smp_processor_id(), &cpumask);
 464
 465         if (oops_in_progress)
 466                 smp_emergency_stop(&cpumask);
 467
 468         /* stop all processors */
 469         for_each_cpu(cpu, &cpumask) {
 470                 struct pcpu *pcpu = pcpu_devices + cpu;
 471                 pcpu_sigp_retry(pcpu, SIGP_STOP, 0);
 472                 while (!pcpu_stopped(pcpu))
 473                         cpu_relax();
 474         }
 475 }
 476
 477 /*
 478  * This is the main routine where commands issued by other
 479  * cpus are handled.
 480  */
 481 static void smp_handle_ext_call(void)
 482 {
 483         unsigned long bits;
 484
 485         /* handle bit signal external calls */
 486         bits = xchg(&pcpu_devices[smp_processor_id()].ec_mask, 0);
 487         if (test_bit(ec_stop_cpu, &bits))
 488                 smp_stop_cpu();
 489         if (test_bit(ec_schedule, &bits))
 490                 scheduler_ipi();
 491         if (test_bit(ec_call_function_single, &bits))
 492                 generic_smp_call_function_single_interrupt();
 493 }
 494
 495 static void do_ext_call_interrupt(struct ext_code ext_code,
 496                                   unsigned int param32, unsigned long param64)
 497 {
 498         inc_irq_stat(ext_code.code == 0x1202 ? IRQEXT_EXC : IRQEXT_EMS);
 499         smp_handle_ext_call();
 500 }
 501
 502 void arch_send_call_function_ipi_mask(const struct cpumask *mask)
 503 {
 504         int cpu;
 505
 506         for_each_cpu(cpu, mask)
 507                 pcpu_ec_call(pcpu_devices + cpu, ec_call_function_single);
 508 }
 509
 510 void arch_send_call_function_single_ipi(int cpu)
 511 {
 512         pcpu_ec_call(pcpu_devices + cpu, ec_call_function_single);
 513 }
 514
 515 /*
 516  * this function sends a 'reschedule' IPI to another CPU.
 517  * it goes straight through and wastes no time serializing
 518  * anything. Worst case is that we lose a reschedule ...
 519  */
 520 void smp_send_reschedule(int cpu)
 521 {
 522         pcpu_ec_call(pcpu_devices + cpu, ec_schedule);
 523 }
 524
 525 /*
 526  * parameter area for the set/clear control bit callbacks
 527  */
 528 struct ec_creg_mask_parms {
 529         unsigned long orval;
 530         unsigned long andval;
 531         int cr;
 532 };
 533
 534 /*
 535  * callback for setting/clearing control bits
 536  */
 537 static void smp_ctl_bit_callback(void *info)
 538 {
 539         struct ec_creg_mask_parms *pp = info;
 540         unsigned long cregs[16];
 541
 542         __ctl_store(cregs, 0, 15);
 543         cregs[pp->cr] = (cregs[pp->cr] & pp->andval) | pp->orval;
 544         __ctl_load(cregs, 0, 15);
 545 }
 546
 547 /*
 548  * Set a bit in a control register of all cpus
 549  */
 550 void smp_ctl_set_bit(int cr, int bit)
 551 {
 552         struct ec_creg_mask_parms parms = { 1UL << bit, -1UL, cr };
 553
 554         on_each_cpu(smp_ctl_bit_callback, &parms, 1);
 555 }
 556 EXPORT_SYMBOL(smp_ctl_set_bit);
 557
 558 /*
 559  * Clear a bit in a control register of all cpus
 560  */
 561 void smp_ctl_clear_bit(int cr, int bit)
 562 {
 563         struct ec_creg_mask_parms parms = { 0, ~(1UL << bit), cr };
 564
 565         on_each_cpu(smp_ctl_bit_callback, &parms, 1);
 566 }
 567 EXPORT_SYMBOL(smp_ctl_clear_bit);
 568
 569 #ifdef CONFIG_CRASH_DUMP
 570
 571 int smp_store_status(int cpu)
 572 {
 573         struct pcpu *pcpu = pcpu_devices + cpu;
 574         unsigned long pa;
 575
 576         pa = __pa(&pcpu->lowcore->floating_pt_save_area);
 577         if (__pcpu_sigp_relax(pcpu->address, SIGP_STORE_STATUS_AT_ADDRESS,
 578                               pa) != SIGP_CC_ORDER_CODE_ACCEPTED)
 579                 return -EIO;
 580         if (!MACHINE_HAS_VX && !MACHINE_HAS_GS)
 581                 return 0;
 582         pa = __pa(pcpu->lowcore->mcesad & MCESA_ORIGIN_MASK);
 583         if (MACHINE_HAS_GS)
 584                 pa |= pcpu->lowcore->mcesad & MCESA_LC_MASK;
 585         if (__pcpu_sigp_relax(pcpu->address, SIGP_STORE_ADDITIONAL_STATUS,
 586                               pa) != SIGP_CC_ORDER_CODE_ACCEPTED)
 587                 return -EIO;
 588         return 0;
 589 }
 590
 591 /*
 592  * Collect CPU state of the previous, crashed system.
 593  * There are four cases:
 594  * 1) standard zfcp dump
 595  *    condition: OLDMEM_BASE == NULL && ipl_info.type == IPL_TYPE_FCP_DUMP
 596  *    The state for all CPUs except the boot CPU needs to be collected
 597  *    with sigp stop-and-store-status. The boot CPU state is located in
 598  *    the absolute lowcore of the memory stored in the HSA. The zcore code
 599  *    will copy the boot CPU state from the HSA.
 600  * 2) stand-alone kdump for SCSI (zfcp dump with swapped memory)
 601  *    condition: OLDMEM_BASE != NULL && ipl_info.type == IPL_TYPE_FCP_DUMP
 602  *    The state for all CPUs except the boot CPU needs to be collected
 603  *    with sigp stop-and-store-status. The firmware or the boot-loader
 604  *    stored the registers of the boot CPU in the absolute lowcore in the
 605  *    memory of the old system.
 606  * 3) kdump and the old kernel did not store the CPU state,
 607  *    or stand-alone kdump for DASD
 608  *    condition: OLDMEM_BASE != NULL && !is_kdump_kernel()
 609  *    The state for all CPUs except the boot CPU needs to be collected
 610  *    with sigp stop-and-store-status. The kexec code or the boot-loader
 611  *    stored the registers of the boot CPU in the memory of the old system.
 612  * 4) kdump and the old kernel stored the CPU state
 613  *    condition: OLDMEM_BASE != NULL && is_kdump_kernel()
 614  *    This case does not exist for s390 anymore, setup_arch explicitly
 615  *    deactivates the elfcorehdr= kernel parameter
 616  */
 617 static __init void smp_save_cpu_vxrs(struct save_area *sa, u16 addr,
 618                                      bool is_boot_cpu, unsigned long page)
 619 {
 620         __vector128 *vxrs = (__vector128 *) page;
 621
 622         if (is_boot_cpu)
 623                 vxrs = boot_cpu_vector_save_area;
 624         else
 625                 __pcpu_sigp_relax(addr, SIGP_STORE_ADDITIONAL_STATUS, page);
 626         save_area_add_vxrs(sa, vxrs);
 627 }
 628
 629 static __init void smp_save_cpu_regs(struct save_area *sa, u16 addr,
 630                                      bool is_boot_cpu, unsigned long page)
 631 {
 632         void *regs = (void *) page;
 633
 634         if (is_boot_cpu)
 635                 copy_oldmem_kernel(regs, (void *) __LC_FPREGS_SAVE_AREA, 512);
 636         else
 637                 __pcpu_sigp_relax(addr, SIGP_STORE_STATUS_AT_ADDRESS, page);
 638         save_area_add_regs(sa, regs);
 639 }
 640
 641 void __init smp_save_dump_cpus(void)
 642 {
 643         int addr, boot_cpu_addr, max_cpu_addr;
 644         struct save_area *sa;
 645         unsigned long page;
 646         bool is_boot_cpu;
 647
 648         if (!(OLDMEM_BASE || ipl_info.type == IPL_TYPE_FCP_DUMP))
 649                 /* No previous system present, normal boot. */
 650                 return;
 651         /* Allocate a page as dumping area for the store status sigps */
 652         page = memblock_alloc_base(PAGE_SIZE, PAGE_SIZE, 1UL << 31);
 653         /* Set multi-threading state to the previous system. */
 654         pcpu_set_smt(sclp.mtid_prev);
 655         boot_cpu_addr = stap();
 656         max_cpu_addr = SCLP_MAX_CORES << sclp.mtid_prev;
 657         for (addr = 0; addr <= max_cpu_addr; addr++) {
 658                 if (__pcpu_sigp_relax(addr, SIGP_SENSE, 0) ==
 659                     SIGP_CC_NOT_OPERATIONAL)
 660                         continue;
 661                 is_boot_cpu = (addr == boot_cpu_addr);
 662                 /* Allocate save area */
 663                 sa = save_area_alloc(is_boot_cpu);
 664                 if (!sa)
 665                         panic("could not allocate memory for save area\n");
 666                 if (MACHINE_HAS_VX)
 667                         /* Get the vector registers */
 668                         smp_save_cpu_vxrs(sa, addr, is_boot_cpu, page);
 669                 /*
 670                  * For a zfcp dump OLDMEM_BASE == NULL and the registers
 671                  * of the boot CPU are stored in the HSA. To retrieve
 672                  * these registers an SCLP request is required which is
 673                  * done by drivers/s390/char/zcore.c:init_cpu_info()
 674                  */
 675                 if (!is_boot_cpu || OLDMEM_BASE)
 676                         /* Get the CPU registers */
 677                         smp_save_cpu_regs(sa, addr, is_boot_cpu, page);
 678         }
 679         memblock_free(page, PAGE_SIZE);
 680         diag308_reset();
 681         pcpu_set_smt(0);
 682 }
 683 #endif /* CONFIG_CRASH_DUMP */
 684
 685 void smp_cpu_set_polarization(int cpu, int val)
 686 {
 687         pcpu_devices[cpu].polarization = val;
 688 }
 689
 690 int smp_cpu_get_polarization(int cpu)
 691 {
 692         return pcpu_devices[cpu].polarization;
 693 }
 694
 695 static void __ref smp_get_core_info(struct sclp_core_info *info, int early)
 696 {
 697         static int use_sigp_detection;
 698         int address;
 699
 700         if (use_sigp_detection || sclp_get_core_info(info, early)) {
 701                 use_sigp_detection = 1;
 702                 for (address = 0;
 703                      address < (SCLP_MAX_CORES << smp_cpu_mt_shift);
 704                      address += (1U << smp_cpu_mt_shift)) {
 705                         if (__pcpu_sigp_relax(address, SIGP_SENSE, 0) ==
 706                             SIGP_CC_NOT_OPERATIONAL)
 707                                 continue;
 708                         info->core[info->configured].core_id =
 709                                 address >> smp_cpu_mt_shift;
 710                         info->configured++;
 711                 }
 712                 info->combined = info->configured;
 713         }
 714 }
 715
 716 static int smp_add_present_cpu(int cpu);
 717
 718 static int __smp_rescan_cpus(struct sclp_core_info *info, int sysfs_add)
 719 {
 720         struct pcpu *pcpu;
 721         cpumask_t avail;
 722         int cpu, nr, i, j;
 723         u16 address;
 724
 725         nr = 0;
 726         cpumask_xor(&avail, cpu_possible_mask, cpu_present_mask);
 727         cpu = cpumask_first(&avail);
 728         for (i = 0; (i < info->combined) && (cpu < nr_cpu_ids); i++) {
 729                 if (sclp.has_core_type && info->core[i].type != boot_core_type)
 730                         continue;
 731                 address = info->core[i].core_id << smp_cpu_mt_shift;
 732                 for (j = 0; j <= smp_cpu_mtid; j++) {
 733                         if (pcpu_find_address(cpu_present_mask, address + j))
 734                                 continue;
 735                         pcpu = pcpu_devices + cpu;
 736                         pcpu->address = address + j;
 737                         pcpu->state =
 738                                 (cpu >= info->configured*(smp_cpu_mtid + 1)) ?
 739                                 CPU_STATE_STANDBY : CPU_STATE_CONFIGURED;
 740                         smp_cpu_set_polarization(cpu, POLARIZATION_UNKNOWN);
 741                         set_cpu_present(cpu, true);
 742                         if (sysfs_add && smp_add_present_cpu(cpu) != 0)
 743                                 set_cpu_present(cpu, false);
 744                         else
 745                                 nr++;
 746                         cpu = cpumask_next(cpu, &avail);
 747                         if (cpu >= nr_cpu_ids)
 748                                 break;
 749                 }
 750         }
 751         return nr;
 752 }
 753
 754 void __init smp_detect_cpus(void)
 755 {
 756         unsigned int cpu, mtid, c_cpus, s_cpus;
 757         struct sclp_core_info *info;
 758         u16 address;
 759
 760         /* Get CPU information */
 761         info = memblock_virt_alloc(sizeof(*info), 8);
 762         smp_get_core_info(info, 1);
 763         /* Find boot CPU type */
 764         if (sclp.has_core_type) {
 765                 address = stap();
 766                 for (cpu = 0; cpu < info->combined; cpu++)
 767                         if (info->core[cpu].core_id == address) {
 768                                 /* The boot cpu dictates the cpu type. */
 769                                 boot_core_type = info->core[cpu].type;
 770                                 break;
 771                         }
 772                 if (cpu >= info->combined)
 773                         panic("Could not find boot CPU type");
 774         }
 775
 776         /* Set multi-threading state for the current system */
 777         mtid = boot_core_type ? sclp.mtid : sclp.mtid_cp;
 778         mtid = (mtid < smp_max_threads) ? mtid : smp_max_threads - 1;
 779         pcpu_set_smt(mtid);
 780
 781         /* Print number of CPUs */
 782         c_cpus = s_cpus = 0;
 783         for (cpu = 0; cpu < info->combined; cpu++) {
 784                 if (sclp.has_core_type &&
 785                     info->core[cpu].type != boot_core_type)
 786                         continue;
 787                 if (cpu < info->configured)
 788                         c_cpus += smp_cpu_mtid + 1;
 789                 else
 790                         s_cpus += smp_cpu_mtid + 1;
 791         }
 792         pr_info("%d configured CPUs, %d standby CPUs\n", c_cpus, s_cpus);
 793
 794         /* Add CPUs present at boot */
 795         get_online_cpus();
 796         __smp_rescan_cpus(info, 0);
 797         put_online_cpus();
 798         memblock_free_early((unsigned long)info, sizeof(*info));
 799 }
 800
 801 /*
 802  *      Activate a secondary processor.
 803  */
 804 static void smp_start_secondary(void *cpuvoid)
 805 {
 806         S390_lowcore.last_update_clock = get_tod_clock();
 807         S390_lowcore.restart_stack = (unsigned long) restart_stack;
 808         S390_lowcore.restart_fn = (unsigned long) do_restart;
 809         S390_lowcore.restart_data = 0;
 810         S390_lowcore.restart_source = -1UL;
 811         restore_access_regs(S390_lowcore.access_regs_save_area);
 812         __ctl_load(S390_lowcore.cregs_save_area, 0, 15);
 813         __load_psw_mask(PSW_KERNEL_BITS | PSW_MASK_DAT);
 814         cpu_init();
 815         preempt_disable();
 816         init_cpu_timer();
 817         vtime_init();
 818         pfault_init();
 819         notify_cpu_starting(smp_processor_id());
 820         set_cpu_online(smp_processor_id(), true);
 821         inc_irq_stat(CPU_RST);
 822         local_irq_enable();
 823         cpu_startup_entry(CPUHP_AP_ONLINE_IDLE);
 824 }
 825
 826 /* Upping and downing of CPUs */
 827 int __cpu_up(unsigned int cpu, struct task_struct *tidle)
 828 {
 829         struct pcpu *pcpu;
 830         int base, i, rc;
 831
 832         pcpu = pcpu_devices + cpu;
 833         if (pcpu->state != CPU_STATE_CONFIGURED)
 834                 return -EIO;
 835         base = smp_get_base_cpu(cpu);
 836         for (i = 0; i <= smp_cpu_mtid; i++) {
 837                 if (base + i < nr_cpu_ids)
 838                         if (cpu_online(base + i))
 839                                 break;
 840         }
 841         /*
 842          * If this is the first CPU of the core to get online
 843          * do an initial CPU reset.
 844          */
 845         if (i > smp_cpu_mtid &&
 846             pcpu_sigp_retry(pcpu_devices + base, SIGP_INITIAL_CPU_RESET, 0) !=
 847             SIGP_CC_ORDER_CODE_ACCEPTED)
 848                 return -EIO;
 849
 850         rc = pcpu_alloc_lowcore(pcpu, cpu);
 851         if (rc)
 852                 return rc;
 853         pcpu_prepare_secondary(pcpu, cpu);
 854         pcpu_attach_task(pcpu, tidle);
 855         pcpu_start_fn(pcpu, smp_start_secondary, NULL);
 856         /* Wait until cpu puts itself in the online & active maps */
 857         while (!cpu_online(cpu))
 858                 cpu_relax();
 859         return 0;
 860 }
 861
 862 static unsigned int setup_possible_cpus __initdata;
 863
 864 static int __init _setup_possible_cpus(char *s)
 865 {
 866         get_option(&s, &setup_possible_cpus);
 867         return 0;
 868 }
 869 early_param("possible_cpus", _setup_possible_cpus);
 870
 871 #ifdef CONFIG_HOTPLUG_CPU
 872
 873 int __cpu_disable(void)
 874 {
 875         unsigned long cregs[16];
 876
 877         /* Handle possible pending IPIs */
 878         smp_handle_ext_call();
 879         set_cpu_online(smp_processor_id(), false);
 880         /* Disable pseudo page faults on this cpu. */
 881         pfault_fini();
 882         /* Disable interrupt sources via control register. */
 883         __ctl_store(cregs, 0, 15);
 884         cregs[0]  &= ~0x0000ee70UL;     /* disable all external interrupts */
 885         cregs[6]  &= ~0xff000000UL;     /* disable all I/O interrupts */
 886         cregs[14] &= ~0x1f000000UL;     /* disable most machine checks */
 887         __ctl_load(cregs, 0, 15);
 888         clear_cpu_flag(CIF_NOHZ_DELAY);
 889         return 0;
 890 }
 891
 892 void __cpu_die(unsigned int cpu)
 893 {
 894         struct pcpu *pcpu;
 895
 896         /* Wait until target cpu is down */
 897         pcpu = pcpu_devices + cpu;
 898         while (!pcpu_stopped(pcpu))
 899                 cpu_relax();
 900         pcpu_free_lowcore(pcpu);
 901         cpumask_clear_cpu(cpu, mm_cpumask(&init_mm));
 902         cpumask_clear_cpu(cpu, &init_mm.context.cpu_attach_mask);
 903 }
 904
 905 void __noreturn cpu_die(void)
 906 {
 907         idle_task_exit();
 908         pcpu_sigp_retry(pcpu_devices + smp_processor_id(), SIGP_STOP, 0);
 909         for (;;) ;
 910 }
 911
 912 #endif /* CONFIG_HOTPLUG_CPU */
 913
 914 void __init smp_fill_possible_mask(void)
 915 {
 916         unsigned int possible, sclp_max, cpu;
 917
 918         sclp_max = max(sclp.mtid, sclp.mtid_cp) + 1;
 919         sclp_max = min(smp_max_threads, sclp_max);
 920         sclp_max = (sclp.max_cores * sclp_max) ?: nr_cpu_ids;
 921         possible = setup_possible_cpus ?: nr_cpu_ids;
 922         possible = min(possible, sclp_max);
 923         for (cpu = 0; cpu < possible && cpu < nr_cpu_ids; cpu++)
 924                 set_cpu_possible(cpu, true);
 925 }
 926
 927 void __init smp_prepare_cpus(unsigned int max_cpus)
 928 {
 929         unsigned long size;
 930
 931         /* request the 0x1201 emergency signal external interrupt */
 932         if (register_external_irq(EXT_IRQ_EMERGENCY_SIG, do_ext_call_interrupt))
 933                 panic("Couldn't request external interrupt 0x1201");
 934         /* request the 0x1202 external call external interrupt */
 935         if (register_external_irq(EXT_IRQ_EXTERNAL_CALL, do_ext_call_interrupt))
 936                 panic("Couldn't request external interrupt 0x1202");
 937         /* create slab cache for the machine-check-extended-save-areas */
 938         if (MACHINE_HAS_VX || MACHINE_HAS_GS) {
 939                 size = 1UL << (MACHINE_HAS_GS ? 11 : 10);
 940                 pcpu_mcesa_cache = kmem_cache_create("nmi_save_areas",
 941                                                      size, size, 0, NULL);
 942                 if (!pcpu_mcesa_cache)
 943                         panic("Couldn't create nmi save area cache");
 944         }
 945 }
 946
 947 void __init smp_prepare_boot_cpu(void)
 948 {
 949         struct pcpu *pcpu = pcpu_devices;
 950
 951         WARN_ON(!cpu_present(0) || !cpu_online(0));
 952         pcpu->state = CPU_STATE_CONFIGURED;
 953         pcpu->lowcore = (struct lowcore *)(unsigned long) store_prefix();
 954         S390_lowcore.percpu_offset = __per_cpu_offset[0];
 955         smp_cpu_set_polarization(0, POLARIZATION_UNKNOWN);
 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         pcpu_devices[0].address = stap();
 965         S390_lowcore.cpu_nr = 0;
 966         S390_lowcore.spinlock_lockval = arch_spin_lockval(0);
 967 }
 968
 969 /*
 970  * the frequency of the profiling timer can be changed
 971  * by writing a multiplier value into /proc/profile.
 972  *
 973  * usually you want to run this on all CPUs ;)
 974  */
 975 int setup_profiling_timer(unsigned int multiplier)
 976 {
 977         return 0;
 978 }
 979
 980 #ifdef CONFIG_HOTPLUG_CPU
 981 static ssize_t cpu_configure_show(struct device *dev,
 982                                   struct device_attribute *attr, char *buf)
 983 {
 984         ssize_t count;
 985
 986         mutex_lock(&smp_cpu_state_mutex);
 987         count = sprintf(buf, "%d\n", pcpu_devices[dev->id].state);
 988         mutex_unlock(&smp_cpu_state_mutex);
 989         return count;
 990 }
 991
 992 static ssize_t cpu_configure_store(struct device *dev,
 993                                    struct device_attribute *attr,
 994                                    const char *buf, size_t count)
 995 {
 996         struct pcpu *pcpu;
 997         int cpu, val, rc, i;
 998         char delim;
 999
1000         if (sscanf(buf, "%d %c", &val, &delim) != 1)
1001                 return -EINVAL;
1002         if (val != 0 && val != 1)
1003                 return -EINVAL;
1004         get_online_cpus();
1005         mutex_lock(&smp_cpu_state_mutex);
1006         rc = -EBUSY;
1007         /* disallow configuration changes of online cpus and cpu 0 */
1008         cpu = dev->id;
1009         cpu = smp_get_base_cpu(cpu);
1010         if (cpu == 0)
1011                 goto out;
1012         for (i = 0; i <= smp_cpu_mtid; i++)
1013                 if (cpu_online(cpu + i))
1014                         goto out;
1015         pcpu = pcpu_devices + cpu;
1016         rc = 0;
1017         switch (val) {
1018         case 0:
1019                 if (pcpu->state != CPU_STATE_CONFIGURED)
1020                         break;
1021                 rc = sclp_core_deconfigure(pcpu->address >> smp_cpu_mt_shift);
1022                 if (rc)
1023                         break;
1024                 for (i = 0; i <= smp_cpu_mtid; i++) {
1025                         if (cpu + i >= nr_cpu_ids || !cpu_present(cpu + i))
1026                                 continue;
1027                         pcpu[i].state = CPU_STATE_STANDBY;
1028                         smp_cpu_set_polarization(cpu + i,
1029                                                  POLARIZATION_UNKNOWN);
1030                 }
1031                 topology_expect_change();
1032                 break;
1033         case 1:
1034                 if (pcpu->state != CPU_STATE_STANDBY)
1035                         break;
1036                 rc = sclp_core_configure(pcpu->address >> smp_cpu_mt_shift);
1037                 if (rc)
1038                         break;
1039                 for (i = 0; i <= smp_cpu_mtid; i++) {
1040                         if (cpu + i >= nr_cpu_ids || !cpu_present(cpu + i))
1041                                 continue;
1042                         pcpu[i].state = CPU_STATE_CONFIGURED;
1043                         smp_cpu_set_polarization(cpu + i,
1044                                                  POLARIZATION_UNKNOWN);
1045                 }
1046                 topology_expect_change();
1047                 break;
1048         default:
1049                 break;
1050         }
1051 out:
1052         mutex_unlock(&smp_cpu_state_mutex);
1053         put_online_cpus();
1054         return rc ? rc : count;
1055 }
1056 static DEVICE_ATTR(configure, 0644, cpu_configure_show, cpu_configure_store);
1057 #endif /* CONFIG_HOTPLUG_CPU */
1058
1059 static ssize_t show_cpu_address(struct device *dev,
1060                                 struct device_attribute *attr, char *buf)
1061 {
1062         return sprintf(buf, "%d\n", pcpu_devices[dev->id].address);
1063 }
1064 static DEVICE_ATTR(address, 0444, show_cpu_address, NULL);
1065
1066 static struct attribute *cpu_common_attrs[] = {
1067 #ifdef CONFIG_HOTPLUG_CPU
1068         &dev_attr_configure.attr,
1069 #endif
1070         &dev_attr_address.attr,
1071         NULL,
1072 };
1073
1074 static struct attribute_group cpu_common_attr_group = {
1075         .attrs = cpu_common_attrs,
1076 };
1077
1078 static struct attribute *cpu_online_attrs[] = {
1079         &dev_attr_idle_count.attr,
1080         &dev_attr_idle_time_us.attr,
1081         NULL,
1082 };
1083
1084 static struct attribute_group cpu_online_attr_group = {
1085         .attrs = cpu_online_attrs,
1086 };
1087
1088 static int smp_cpu_online(unsigned int cpu)
1089 {
1090         struct device *s = &per_cpu(cpu_device, cpu)->dev;
1091
1092         return sysfs_create_group(&s->kobj, &cpu_online_attr_group);
1093 }
1094 static int smp_cpu_pre_down(unsigned int cpu)
1095 {
1096         struct device *s = &per_cpu(cpu_device, cpu)->dev;
1097
1098         sysfs_remove_group(&s->kobj, &cpu_online_attr_group);
1099         return 0;
1100 }
1101
1102 static int smp_add_present_cpu(int cpu)
1103 {
1104         struct device *s;
1105         struct cpu *c;
1106         int rc;
1107
1108         c = kzalloc(sizeof(*c), GFP_KERNEL);
1109         if (!c)
1110                 return -ENOMEM;
1111         per_cpu(cpu_device, cpu) = c;
1112         s = &c->dev;
1113         c->hotpluggable = 1;
1114         rc = register_cpu(c, cpu);
1115         if (rc)
1116                 goto out;
1117         rc = sysfs_create_group(&s->kobj, &cpu_common_attr_group);
1118         if (rc)
1119                 goto out_cpu;
1120         rc = topology_cpu_init(c);
1121         if (rc)
1122                 goto out_topology;
1123         return 0;
1124
1125 out_topology:
1126         sysfs_remove_group(&s->kobj, &cpu_common_attr_group);
1127 out_cpu:
1128 #ifdef CONFIG_HOTPLUG_CPU
1129         unregister_cpu(c);
1130 #endif
1131 out:
1132         return rc;
1133 }
1134
1135 #ifdef CONFIG_HOTPLUG_CPU
1136
1137 int __ref smp_rescan_cpus(void)
1138 {
1139         struct sclp_core_info *info;
1140         int nr;
1141
1142         info = kzalloc(sizeof(*info), GFP_KERNEL);
1143         if (!info)
1144                 return -ENOMEM;
1145         smp_get_core_info(info, 0);
1146         get_online_cpus();
1147         mutex_lock(&smp_cpu_state_mutex);
1148         nr = __smp_rescan_cpus(info, 1);
1149         mutex_unlock(&smp_cpu_state_mutex);
1150         put_online_cpus();
1151         kfree(info);
1152         if (nr)
1153                 topology_schedule_update();
1154         return 0;
1155 }
1156
1157 static ssize_t __ref rescan_store(struct device *dev,
1158                                   struct device_attribute *attr,
1159                                   const char *buf,
1160                                   size_t count)
1161 {
1162         int rc;
1163
1164         rc = smp_rescan_cpus();
1165         return rc ? rc : count;
1166 }
1167 static DEVICE_ATTR(rescan, 0200, NULL, rescan_store);
1168 #endif /* CONFIG_HOTPLUG_CPU */
1169
1170 static int __init s390_smp_init(void)
1171 {
1172         int cpu, rc = 0;
1173
1174 #ifdef CONFIG_HOTPLUG_CPU
1175         rc = device_create_file(cpu_subsys.dev_root, &dev_attr_rescan);
1176         if (rc)
1177                 return rc;
1178 #endif
1179         for_each_present_cpu(cpu) {
1180                 rc = smp_add_present_cpu(cpu);
1181                 if (rc)
1182                         goto out;
1183         }
1184
1185         rc = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "s390/smp:online",
1186                                smp_cpu_online, smp_cpu_pre_down);
1187         rc = rc <= 0 ? rc : 0;
1188 out:
1189         return rc;
1190 }
1191 subsys_initcall(s390_smp_init);