1 /* 2 * arch/s390/kernel/smp.c 3 * 4 * Copyright IBM Corp. 1999, 2009 5 * Author(s): Denis Joseph Barrow (djbarrow@de.ibm.com,barrow_dj@yahoo.com), 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 * We work with logical cpu numbering everywhere we can. The only 14 * functions using the real cpu address (got from STAP) are the sigp 15 * functions. For all other functions we use the identity mapping. 16 * That means that cpu_number_map[i] == i for every cpu. cpu_number_map is 17 * used e.g. to find the idle task belonging to a logical cpu. Every array 18 * in the kernel is sorted by the logical cpu number and not by the physical 19 * one which is causing all the confusion with __cpu_logical_map and 20 * cpu_number_map in other architectures. 21 */ 22 23 #define KMSG_COMPONENT "cpu" 24 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt 25 26 #include <linux/workqueue.h> 27 #include <linux/module.h> 28 #include <linux/init.h> 29 #include <linux/mm.h> 30 #include <linux/err.h> 31 #include <linux/spinlock.h> 32 #include <linux/kernel_stat.h> 33 #include <linux/delay.h> 34 #include <linux/cache.h> 35 #include <linux/interrupt.h> 36 #include <linux/irqflags.h> 37 #include <linux/cpu.h> 38 #include <linux/timex.h> 39 #include <linux/bootmem.h> 40 #include <linux/slab.h> 41 #include <asm/asm-offsets.h> 42 #include <asm/ipl.h> 43 #include <asm/setup.h> 44 #include <asm/sigp.h> 45 #include <asm/pgalloc.h> 46 #include <asm/irq.h> 47 #include <asm/cpcmd.h> 48 #include <asm/tlbflush.h> 49 #include <asm/timer.h> 50 #include <asm/lowcore.h> 51 #include <asm/sclp.h> 52 #include <asm/cputime.h> 53 #include <asm/vdso.h> 54 #include <asm/cpu.h> 55 #include "entry.h" 56 57 /* logical cpu to cpu address */ 58 unsigned short __cpu_logical_map[NR_CPUS]; 59 60 static struct task_struct *current_set[NR_CPUS]; 61 62 static u8 smp_cpu_type; 63 static int smp_use_sigp_detection; 64 65 enum s390_cpu_state { 66 CPU_STATE_STANDBY, 67 CPU_STATE_CONFIGURED, 68 }; 69 70 DEFINE_MUTEX(smp_cpu_state_mutex); 71 int smp_cpu_polarization[NR_CPUS]; 72 static int smp_cpu_state[NR_CPUS]; 73 static int cpu_management; 74 75 static DEFINE_PER_CPU(struct cpu, cpu_devices); 76 77 static void smp_ext_bitcall(int, int); 78 79 static int raw_cpu_stopped(int cpu) 80 { 81 u32 status; 82 83 switch (raw_sigp_ps(&status, 0, cpu, sigp_sense)) { 84 case sigp_status_stored: 85 /* Check for stopped and check stop state */ 86 if (status & 0x50) 87 return 1; 88 break; 89 default: 90 break; 91 } 92 return 0; 93 } 94 95 static inline int cpu_stopped(int cpu) 96 { 97 return raw_cpu_stopped(cpu_logical_map(cpu)); 98 } 99 100 void smp_switch_to_ipl_cpu(void (*func)(void *), void *data) 101 { 102 struct _lowcore *lc, *current_lc; 103 struct stack_frame *sf; 104 struct pt_regs *regs; 105 unsigned long sp; 106 107 if (smp_processor_id() == 0) 108 func(data); 109 __load_psw_mask(PSW_BASE_BITS | PSW_DEFAULT_KEY); 110 /* Disable lowcore protection */ 111 __ctl_clear_bit(0, 28); 112 current_lc = lowcore_ptr[smp_processor_id()]; 113 lc = lowcore_ptr[0]; 114 if (!lc) 115 lc = current_lc; 116 lc->restart_psw.mask = PSW_BASE_BITS | PSW_DEFAULT_KEY; 117 lc->restart_psw.addr = PSW_ADDR_AMODE | (unsigned long) smp_restart_cpu; 118 if (!cpu_online(0)) 119 smp_switch_to_cpu(func, data, 0, stap(), __cpu_logical_map[0]); 120 while (sigp(0, sigp_stop_and_store_status) == sigp_busy) 121 cpu_relax(); 122 sp = lc->panic_stack; 123 sp -= sizeof(struct pt_regs); 124 regs = (struct pt_regs *) sp; 125 memcpy(®s->gprs, ¤t_lc->gpregs_save_area, sizeof(regs->gprs)); 126 regs->psw = lc->psw_save_area; 127 sp -= STACK_FRAME_OVERHEAD; 128 sf = (struct stack_frame *) sp; 129 sf->back_chain = regs->gprs[15]; 130 smp_switch_to_cpu(func, data, sp, stap(), __cpu_logical_map[0]); 131 } 132 133 void smp_send_stop(void) 134 { 135 int cpu, rc; 136 137 /* Disable all interrupts/machine checks */ 138 __load_psw_mask(psw_kernel_bits & ~PSW_MASK_MCHECK); 139 trace_hardirqs_off(); 140 141 /* stop all processors */ 142 for_each_online_cpu(cpu) { 143 if (cpu == smp_processor_id()) 144 continue; 145 do { 146 rc = sigp(cpu, sigp_stop); 147 } while (rc == sigp_busy); 148 149 while (!cpu_stopped(cpu)) 150 cpu_relax(); 151 } 152 } 153 154 /* 155 * This is the main routine where commands issued by other 156 * cpus are handled. 157 */ 158 159 static void do_ext_call_interrupt(unsigned int ext_int_code, 160 unsigned int param32, unsigned long param64) 161 { 162 unsigned long bits; 163 164 kstat_cpu(smp_processor_id()).irqs[EXTINT_IPI]++; 165 /* 166 * handle bit signal external calls 167 */ 168 bits = xchg(&S390_lowcore.ext_call_fast, 0); 169 170 if (test_bit(ec_schedule, &bits)) 171 scheduler_ipi(); 172 173 if (test_bit(ec_call_function, &bits)) 174 generic_smp_call_function_interrupt(); 175 176 if (test_bit(ec_call_function_single, &bits)) 177 generic_smp_call_function_single_interrupt(); 178 } 179 180 /* 181 * Send an external call sigp to another cpu and return without waiting 182 * for its completion. 183 */ 184 static void smp_ext_bitcall(int cpu, int sig) 185 { 186 /* 187 * Set signaling bit in lowcore of target cpu and kick it 188 */ 189 set_bit(sig, (unsigned long *) &lowcore_ptr[cpu]->ext_call_fast); 190 while (sigp(cpu, sigp_emergency_signal) == sigp_busy) 191 udelay(10); 192 } 193 194 void arch_send_call_function_ipi_mask(const struct cpumask *mask) 195 { 196 int cpu; 197 198 for_each_cpu(cpu, mask) 199 smp_ext_bitcall(cpu, ec_call_function); 200 } 201 202 void arch_send_call_function_single_ipi(int cpu) 203 { 204 smp_ext_bitcall(cpu, ec_call_function_single); 205 } 206 207 #ifndef CONFIG_64BIT 208 /* 209 * this function sends a 'purge tlb' signal to another CPU. 210 */ 211 static void smp_ptlb_callback(void *info) 212 { 213 __tlb_flush_local(); 214 } 215 216 void smp_ptlb_all(void) 217 { 218 on_each_cpu(smp_ptlb_callback, NULL, 1); 219 } 220 EXPORT_SYMBOL(smp_ptlb_all); 221 #endif /* ! CONFIG_64BIT */ 222 223 /* 224 * this function sends a 'reschedule' IPI to another CPU. 225 * it goes straight through and wastes no time serializing 226 * anything. Worst case is that we lose a reschedule ... 227 */ 228 void smp_send_reschedule(int cpu) 229 { 230 smp_ext_bitcall(cpu, ec_schedule); 231 } 232 233 /* 234 * parameter area for the set/clear control bit callbacks 235 */ 236 struct ec_creg_mask_parms { 237 unsigned long orvals[16]; 238 unsigned long andvals[16]; 239 }; 240 241 /* 242 * callback for setting/clearing control bits 243 */ 244 static void smp_ctl_bit_callback(void *info) 245 { 246 struct ec_creg_mask_parms *pp = info; 247 unsigned long cregs[16]; 248 int i; 249 250 __ctl_store(cregs, 0, 15); 251 for (i = 0; i <= 15; i++) 252 cregs[i] = (cregs[i] & pp->andvals[i]) | pp->orvals[i]; 253 __ctl_load(cregs, 0, 15); 254 } 255 256 /* 257 * Set a bit in a control register of all cpus 258 */ 259 void smp_ctl_set_bit(int cr, int bit) 260 { 261 struct ec_creg_mask_parms parms; 262 263 memset(&parms.orvals, 0, sizeof(parms.orvals)); 264 memset(&parms.andvals, 0xff, sizeof(parms.andvals)); 265 parms.orvals[cr] = 1 << bit; 266 on_each_cpu(smp_ctl_bit_callback, &parms, 1); 267 } 268 EXPORT_SYMBOL(smp_ctl_set_bit); 269 270 /* 271 * Clear a bit in a control register of all cpus 272 */ 273 void smp_ctl_clear_bit(int cr, int bit) 274 { 275 struct ec_creg_mask_parms parms; 276 277 memset(&parms.orvals, 0, sizeof(parms.orvals)); 278 memset(&parms.andvals, 0xff, sizeof(parms.andvals)); 279 parms.andvals[cr] = ~(1L << bit); 280 on_each_cpu(smp_ctl_bit_callback, &parms, 1); 281 } 282 EXPORT_SYMBOL(smp_ctl_clear_bit); 283 284 #ifdef CONFIG_ZFCPDUMP 285 286 static void __init smp_get_save_area(unsigned int cpu, unsigned int phy_cpu) 287 { 288 if (ipl_info.type != IPL_TYPE_FCP_DUMP) 289 return; 290 if (cpu >= NR_CPUS) { 291 pr_warning("CPU %i exceeds the maximum %i and is excluded from " 292 "the dump\n", cpu, NR_CPUS - 1); 293 return; 294 } 295 zfcpdump_save_areas[cpu] = kmalloc(sizeof(struct save_area), GFP_KERNEL); 296 while (raw_sigp(phy_cpu, sigp_stop_and_store_status) == sigp_busy) 297 cpu_relax(); 298 memcpy_real(zfcpdump_save_areas[cpu], 299 (void *)(unsigned long) store_prefix() + SAVE_AREA_BASE, 300 sizeof(struct save_area)); 301 } 302 303 struct save_area *zfcpdump_save_areas[NR_CPUS + 1]; 304 EXPORT_SYMBOL_GPL(zfcpdump_save_areas); 305 306 #else 307 308 static inline void smp_get_save_area(unsigned int cpu, unsigned int phy_cpu) { } 309 310 #endif /* CONFIG_ZFCPDUMP */ 311 312 static int cpu_known(int cpu_id) 313 { 314 int cpu; 315 316 for_each_present_cpu(cpu) { 317 if (__cpu_logical_map[cpu] == cpu_id) 318 return 1; 319 } 320 return 0; 321 } 322 323 static int smp_rescan_cpus_sigp(cpumask_t avail) 324 { 325 int cpu_id, logical_cpu; 326 327 logical_cpu = cpumask_first(&avail); 328 if (logical_cpu >= nr_cpu_ids) 329 return 0; 330 for (cpu_id = 0; cpu_id <= MAX_CPU_ADDRESS; cpu_id++) { 331 if (cpu_known(cpu_id)) 332 continue; 333 __cpu_logical_map[logical_cpu] = cpu_id; 334 smp_cpu_polarization[logical_cpu] = POLARIZATION_UNKNWN; 335 if (!cpu_stopped(logical_cpu)) 336 continue; 337 set_cpu_present(logical_cpu, true); 338 smp_cpu_state[logical_cpu] = CPU_STATE_CONFIGURED; 339 logical_cpu = cpumask_next(logical_cpu, &avail); 340 if (logical_cpu >= nr_cpu_ids) 341 break; 342 } 343 return 0; 344 } 345 346 static int smp_rescan_cpus_sclp(cpumask_t avail) 347 { 348 struct sclp_cpu_info *info; 349 int cpu_id, logical_cpu, cpu; 350 int rc; 351 352 logical_cpu = cpumask_first(&avail); 353 if (logical_cpu >= nr_cpu_ids) 354 return 0; 355 info = kmalloc(sizeof(*info), GFP_KERNEL); 356 if (!info) 357 return -ENOMEM; 358 rc = sclp_get_cpu_info(info); 359 if (rc) 360 goto out; 361 for (cpu = 0; cpu < info->combined; cpu++) { 362 if (info->has_cpu_type && info->cpu[cpu].type != smp_cpu_type) 363 continue; 364 cpu_id = info->cpu[cpu].address; 365 if (cpu_known(cpu_id)) 366 continue; 367 __cpu_logical_map[logical_cpu] = cpu_id; 368 smp_cpu_polarization[logical_cpu] = POLARIZATION_UNKNWN; 369 set_cpu_present(logical_cpu, true); 370 if (cpu >= info->configured) 371 smp_cpu_state[logical_cpu] = CPU_STATE_STANDBY; 372 else 373 smp_cpu_state[logical_cpu] = CPU_STATE_CONFIGURED; 374 logical_cpu = cpumask_next(logical_cpu, &avail); 375 if (logical_cpu >= nr_cpu_ids) 376 break; 377 } 378 out: 379 kfree(info); 380 return rc; 381 } 382 383 static int __smp_rescan_cpus(void) 384 { 385 cpumask_t avail; 386 387 cpumask_xor(&avail, cpu_possible_mask, cpu_present_mask); 388 if (smp_use_sigp_detection) 389 return smp_rescan_cpus_sigp(avail); 390 else 391 return smp_rescan_cpus_sclp(avail); 392 } 393 394 static void __init smp_detect_cpus(void) 395 { 396 unsigned int cpu, c_cpus, s_cpus; 397 struct sclp_cpu_info *info; 398 u16 boot_cpu_addr, cpu_addr; 399 400 c_cpus = 1; 401 s_cpus = 0; 402 boot_cpu_addr = __cpu_logical_map[0]; 403 info = kmalloc(sizeof(*info), GFP_KERNEL); 404 if (!info) 405 panic("smp_detect_cpus failed to allocate memory\n"); 406 /* Use sigp detection algorithm if sclp doesn't work. */ 407 if (sclp_get_cpu_info(info)) { 408 smp_use_sigp_detection = 1; 409 for (cpu = 0; cpu <= MAX_CPU_ADDRESS; cpu++) { 410 if (cpu == boot_cpu_addr) 411 continue; 412 if (!raw_cpu_stopped(cpu)) 413 continue; 414 smp_get_save_area(c_cpus, cpu); 415 c_cpus++; 416 } 417 goto out; 418 } 419 420 if (info->has_cpu_type) { 421 for (cpu = 0; cpu < info->combined; cpu++) { 422 if (info->cpu[cpu].address == boot_cpu_addr) { 423 smp_cpu_type = info->cpu[cpu].type; 424 break; 425 } 426 } 427 } 428 429 for (cpu = 0; cpu < info->combined; cpu++) { 430 if (info->has_cpu_type && info->cpu[cpu].type != smp_cpu_type) 431 continue; 432 cpu_addr = info->cpu[cpu].address; 433 if (cpu_addr == boot_cpu_addr) 434 continue; 435 if (!raw_cpu_stopped(cpu_addr)) { 436 s_cpus++; 437 continue; 438 } 439 smp_get_save_area(c_cpus, cpu_addr); 440 c_cpus++; 441 } 442 out: 443 kfree(info); 444 pr_info("%d configured CPUs, %d standby CPUs\n", c_cpus, s_cpus); 445 get_online_cpus(); 446 __smp_rescan_cpus(); 447 put_online_cpus(); 448 } 449 450 /* 451 * Activate a secondary processor. 452 */ 453 int __cpuinit start_secondary(void *cpuvoid) 454 { 455 /* Setup the cpu */ 456 cpu_init(); 457 preempt_disable(); 458 /* Enable TOD clock interrupts on the secondary cpu. */ 459 init_cpu_timer(); 460 /* Enable cpu timer interrupts on the secondary cpu. */ 461 init_cpu_vtimer(); 462 /* Enable pfault pseudo page faults on this cpu. */ 463 pfault_init(); 464 465 /* call cpu notifiers */ 466 notify_cpu_starting(smp_processor_id()); 467 /* Mark this cpu as online */ 468 ipi_call_lock(); 469 set_cpu_online(smp_processor_id(), true); 470 ipi_call_unlock(); 471 /* Switch on interrupts */ 472 local_irq_enable(); 473 /* cpu_idle will call schedule for us */ 474 cpu_idle(); 475 return 0; 476 } 477 478 struct create_idle { 479 struct work_struct work; 480 struct task_struct *idle; 481 struct completion done; 482 int cpu; 483 }; 484 485 static void __cpuinit smp_fork_idle(struct work_struct *work) 486 { 487 struct create_idle *c_idle; 488 489 c_idle = container_of(work, struct create_idle, work); 490 c_idle->idle = fork_idle(c_idle->cpu); 491 complete(&c_idle->done); 492 } 493 494 static int __cpuinit smp_alloc_lowcore(int cpu) 495 { 496 unsigned long async_stack, panic_stack; 497 struct _lowcore *lowcore; 498 499 lowcore = (void *) __get_free_pages(GFP_KERNEL | GFP_DMA, LC_ORDER); 500 if (!lowcore) 501 return -ENOMEM; 502 async_stack = __get_free_pages(GFP_KERNEL, ASYNC_ORDER); 503 panic_stack = __get_free_page(GFP_KERNEL); 504 if (!panic_stack || !async_stack) 505 goto out; 506 memcpy(lowcore, &S390_lowcore, 512); 507 memset((char *)lowcore + 512, 0, sizeof(*lowcore) - 512); 508 lowcore->async_stack = async_stack + ASYNC_SIZE; 509 lowcore->panic_stack = panic_stack + PAGE_SIZE; 510 511 #ifndef CONFIG_64BIT 512 if (MACHINE_HAS_IEEE) { 513 unsigned long save_area; 514 515 save_area = get_zeroed_page(GFP_KERNEL); 516 if (!save_area) 517 goto out; 518 lowcore->extended_save_area_addr = (u32) save_area; 519 } 520 #else 521 if (vdso_alloc_per_cpu(cpu, lowcore)) 522 goto out; 523 #endif 524 lowcore_ptr[cpu] = lowcore; 525 return 0; 526 527 out: 528 free_page(panic_stack); 529 free_pages(async_stack, ASYNC_ORDER); 530 free_pages((unsigned long) lowcore, LC_ORDER); 531 return -ENOMEM; 532 } 533 534 static void smp_free_lowcore(int cpu) 535 { 536 struct _lowcore *lowcore; 537 538 lowcore = lowcore_ptr[cpu]; 539 #ifndef CONFIG_64BIT 540 if (MACHINE_HAS_IEEE) 541 free_page((unsigned long) lowcore->extended_save_area_addr); 542 #else 543 vdso_free_per_cpu(cpu, lowcore); 544 #endif 545 free_page(lowcore->panic_stack - PAGE_SIZE); 546 free_pages(lowcore->async_stack - ASYNC_SIZE, ASYNC_ORDER); 547 free_pages((unsigned long) lowcore, LC_ORDER); 548 lowcore_ptr[cpu] = NULL; 549 } 550 551 /* Upping and downing of CPUs */ 552 int __cpuinit __cpu_up(unsigned int cpu) 553 { 554 struct _lowcore *cpu_lowcore; 555 struct create_idle c_idle; 556 struct task_struct *idle; 557 struct stack_frame *sf; 558 u32 lowcore; 559 int ccode; 560 561 if (smp_cpu_state[cpu] != CPU_STATE_CONFIGURED) 562 return -EIO; 563 idle = current_set[cpu]; 564 if (!idle) { 565 c_idle.done = COMPLETION_INITIALIZER_ONSTACK(c_idle.done); 566 INIT_WORK_ONSTACK(&c_idle.work, smp_fork_idle); 567 c_idle.cpu = cpu; 568 schedule_work(&c_idle.work); 569 wait_for_completion(&c_idle.done); 570 if (IS_ERR(c_idle.idle)) 571 return PTR_ERR(c_idle.idle); 572 idle = c_idle.idle; 573 current_set[cpu] = c_idle.idle; 574 } 575 init_idle(idle, cpu); 576 if (smp_alloc_lowcore(cpu)) 577 return -ENOMEM; 578 do { 579 ccode = sigp(cpu, sigp_initial_cpu_reset); 580 if (ccode == sigp_busy) 581 udelay(10); 582 if (ccode == sigp_not_operational) 583 goto err_out; 584 } while (ccode == sigp_busy); 585 586 lowcore = (u32)(unsigned long)lowcore_ptr[cpu]; 587 while (sigp_p(lowcore, cpu, sigp_set_prefix) == sigp_busy) 588 udelay(10); 589 590 cpu_lowcore = lowcore_ptr[cpu]; 591 cpu_lowcore->kernel_stack = (unsigned long) 592 task_stack_page(idle) + THREAD_SIZE; 593 cpu_lowcore->thread_info = (unsigned long) task_thread_info(idle); 594 sf = (struct stack_frame *) (cpu_lowcore->kernel_stack 595 - sizeof(struct pt_regs) 596 - sizeof(struct stack_frame)); 597 memset(sf, 0, sizeof(struct stack_frame)); 598 sf->gprs[9] = (unsigned long) sf; 599 cpu_lowcore->save_area[15] = (unsigned long) sf; 600 __ctl_store(cpu_lowcore->cregs_save_area, 0, 15); 601 atomic_inc(&init_mm.context.attach_count); 602 asm volatile( 603 " stam 0,15,0(%0)" 604 : : "a" (&cpu_lowcore->access_regs_save_area) : "memory"); 605 cpu_lowcore->percpu_offset = __per_cpu_offset[cpu]; 606 cpu_lowcore->current_task = (unsigned long) idle; 607 cpu_lowcore->cpu_nr = cpu; 608 cpu_lowcore->kernel_asce = S390_lowcore.kernel_asce; 609 cpu_lowcore->machine_flags = S390_lowcore.machine_flags; 610 cpu_lowcore->ftrace_func = S390_lowcore.ftrace_func; 611 memcpy(cpu_lowcore->stfle_fac_list, S390_lowcore.stfle_fac_list, 612 MAX_FACILITY_BIT/8); 613 eieio(); 614 615 while (sigp(cpu, sigp_restart) == sigp_busy) 616 udelay(10); 617 618 while (!cpu_online(cpu)) 619 cpu_relax(); 620 return 0; 621 622 err_out: 623 smp_free_lowcore(cpu); 624 return -EIO; 625 } 626 627 static int __init setup_possible_cpus(char *s) 628 { 629 int pcpus, cpu; 630 631 pcpus = simple_strtoul(s, NULL, 0); 632 init_cpu_possible(cpumask_of(0)); 633 for (cpu = 1; cpu < pcpus && cpu < nr_cpu_ids; cpu++) 634 set_cpu_possible(cpu, true); 635 return 0; 636 } 637 early_param("possible_cpus", setup_possible_cpus); 638 639 #ifdef CONFIG_HOTPLUG_CPU 640 641 int __cpu_disable(void) 642 { 643 struct ec_creg_mask_parms cr_parms; 644 int cpu = smp_processor_id(); 645 646 set_cpu_online(cpu, false); 647 648 /* Disable pfault pseudo page faults on this cpu. */ 649 pfault_fini(); 650 651 memset(&cr_parms.orvals, 0, sizeof(cr_parms.orvals)); 652 memset(&cr_parms.andvals, 0xff, sizeof(cr_parms.andvals)); 653 654 /* disable all external interrupts */ 655 cr_parms.orvals[0] = 0; 656 cr_parms.andvals[0] = ~(1 << 15 | 1 << 14 | 1 << 13 | 1 << 11 | 657 1 << 10 | 1 << 9 | 1 << 6 | 1 << 4); 658 /* disable all I/O interrupts */ 659 cr_parms.orvals[6] = 0; 660 cr_parms.andvals[6] = ~(1 << 31 | 1 << 30 | 1 << 29 | 1 << 28 | 661 1 << 27 | 1 << 26 | 1 << 25 | 1 << 24); 662 /* disable most machine checks */ 663 cr_parms.orvals[14] = 0; 664 cr_parms.andvals[14] = ~(1 << 28 | 1 << 27 | 1 << 26 | 665 1 << 25 | 1 << 24); 666 667 smp_ctl_bit_callback(&cr_parms); 668 669 return 0; 670 } 671 672 void __cpu_die(unsigned int cpu) 673 { 674 /* Wait until target cpu is down */ 675 while (!cpu_stopped(cpu)) 676 cpu_relax(); 677 while (sigp_p(0, cpu, sigp_set_prefix) == sigp_busy) 678 udelay(10); 679 smp_free_lowcore(cpu); 680 atomic_dec(&init_mm.context.attach_count); 681 } 682 683 void __noreturn cpu_die(void) 684 { 685 idle_task_exit(); 686 while (sigp(smp_processor_id(), sigp_stop) == sigp_busy) 687 cpu_relax(); 688 for (;;); 689 } 690 691 #endif /* CONFIG_HOTPLUG_CPU */ 692 693 void __init smp_prepare_cpus(unsigned int max_cpus) 694 { 695 #ifndef CONFIG_64BIT 696 unsigned long save_area = 0; 697 #endif 698 unsigned long async_stack, panic_stack; 699 struct _lowcore *lowcore; 700 701 smp_detect_cpus(); 702 703 /* request the 0x1201 emergency signal external interrupt */ 704 if (register_external_interrupt(0x1201, do_ext_call_interrupt) != 0) 705 panic("Couldn't request external interrupt 0x1201"); 706 707 /* Reallocate current lowcore, but keep its contents. */ 708 lowcore = (void *) __get_free_pages(GFP_KERNEL | GFP_DMA, LC_ORDER); 709 panic_stack = __get_free_page(GFP_KERNEL); 710 async_stack = __get_free_pages(GFP_KERNEL, ASYNC_ORDER); 711 BUG_ON(!lowcore || !panic_stack || !async_stack); 712 #ifndef CONFIG_64BIT 713 if (MACHINE_HAS_IEEE) 714 save_area = get_zeroed_page(GFP_KERNEL); 715 #endif 716 local_irq_disable(); 717 local_mcck_disable(); 718 lowcore_ptr[smp_processor_id()] = lowcore; 719 *lowcore = S390_lowcore; 720 lowcore->panic_stack = panic_stack + PAGE_SIZE; 721 lowcore->async_stack = async_stack + ASYNC_SIZE; 722 #ifndef CONFIG_64BIT 723 if (MACHINE_HAS_IEEE) 724 lowcore->extended_save_area_addr = (u32) save_area; 725 #endif 726 set_prefix((u32)(unsigned long) lowcore); 727 local_mcck_enable(); 728 local_irq_enable(); 729 #ifdef CONFIG_64BIT 730 if (vdso_alloc_per_cpu(smp_processor_id(), &S390_lowcore)) 731 BUG(); 732 #endif 733 } 734 735 void __init smp_prepare_boot_cpu(void) 736 { 737 BUG_ON(smp_processor_id() != 0); 738 739 current_thread_info()->cpu = 0; 740 set_cpu_present(0, true); 741 set_cpu_online(0, true); 742 S390_lowcore.percpu_offset = __per_cpu_offset[0]; 743 current_set[0] = current; 744 smp_cpu_state[0] = CPU_STATE_CONFIGURED; 745 smp_cpu_polarization[0] = POLARIZATION_UNKNWN; 746 } 747 748 void __init smp_cpus_done(unsigned int max_cpus) 749 { 750 } 751 752 void __init smp_setup_processor_id(void) 753 { 754 S390_lowcore.cpu_nr = 0; 755 __cpu_logical_map[0] = stap(); 756 } 757 758 /* 759 * the frequency of the profiling timer can be changed 760 * by writing a multiplier value into /proc/profile. 761 * 762 * usually you want to run this on all CPUs ;) 763 */ 764 int setup_profiling_timer(unsigned int multiplier) 765 { 766 return 0; 767 } 768 769 #ifdef CONFIG_HOTPLUG_CPU 770 static ssize_t cpu_configure_show(struct sys_device *dev, 771 struct sysdev_attribute *attr, char *buf) 772 { 773 ssize_t count; 774 775 mutex_lock(&smp_cpu_state_mutex); 776 count = sprintf(buf, "%d\n", smp_cpu_state[dev->id]); 777 mutex_unlock(&smp_cpu_state_mutex); 778 return count; 779 } 780 781 static ssize_t cpu_configure_store(struct sys_device *dev, 782 struct sysdev_attribute *attr, 783 const char *buf, size_t count) 784 { 785 int cpu = dev->id; 786 int val, rc; 787 char delim; 788 789 if (sscanf(buf, "%d %c", &val, &delim) != 1) 790 return -EINVAL; 791 if (val != 0 && val != 1) 792 return -EINVAL; 793 794 get_online_cpus(); 795 mutex_lock(&smp_cpu_state_mutex); 796 rc = -EBUSY; 797 /* disallow configuration changes of online cpus and cpu 0 */ 798 if (cpu_online(cpu) || cpu == 0) 799 goto out; 800 rc = 0; 801 switch (val) { 802 case 0: 803 if (smp_cpu_state[cpu] == CPU_STATE_CONFIGURED) { 804 rc = sclp_cpu_deconfigure(__cpu_logical_map[cpu]); 805 if (!rc) { 806 smp_cpu_state[cpu] = CPU_STATE_STANDBY; 807 smp_cpu_polarization[cpu] = POLARIZATION_UNKNWN; 808 } 809 } 810 break; 811 case 1: 812 if (smp_cpu_state[cpu] == CPU_STATE_STANDBY) { 813 rc = sclp_cpu_configure(__cpu_logical_map[cpu]); 814 if (!rc) { 815 smp_cpu_state[cpu] = CPU_STATE_CONFIGURED; 816 smp_cpu_polarization[cpu] = POLARIZATION_UNKNWN; 817 } 818 } 819 break; 820 default: 821 break; 822 } 823 out: 824 mutex_unlock(&smp_cpu_state_mutex); 825 put_online_cpus(); 826 return rc ? rc : count; 827 } 828 static SYSDEV_ATTR(configure, 0644, cpu_configure_show, cpu_configure_store); 829 #endif /* CONFIG_HOTPLUG_CPU */ 830 831 static ssize_t cpu_polarization_show(struct sys_device *dev, 832 struct sysdev_attribute *attr, char *buf) 833 { 834 int cpu = dev->id; 835 ssize_t count; 836 837 mutex_lock(&smp_cpu_state_mutex); 838 switch (smp_cpu_polarization[cpu]) { 839 case POLARIZATION_HRZ: 840 count = sprintf(buf, "horizontal\n"); 841 break; 842 case POLARIZATION_VL: 843 count = sprintf(buf, "vertical:low\n"); 844 break; 845 case POLARIZATION_VM: 846 count = sprintf(buf, "vertical:medium\n"); 847 break; 848 case POLARIZATION_VH: 849 count = sprintf(buf, "vertical:high\n"); 850 break; 851 default: 852 count = sprintf(buf, "unknown\n"); 853 break; 854 } 855 mutex_unlock(&smp_cpu_state_mutex); 856 return count; 857 } 858 static SYSDEV_ATTR(polarization, 0444, cpu_polarization_show, NULL); 859 860 static ssize_t show_cpu_address(struct sys_device *dev, 861 struct sysdev_attribute *attr, char *buf) 862 { 863 return sprintf(buf, "%d\n", __cpu_logical_map[dev->id]); 864 } 865 static SYSDEV_ATTR(address, 0444, show_cpu_address, NULL); 866 867 868 static struct attribute *cpu_common_attrs[] = { 869 #ifdef CONFIG_HOTPLUG_CPU 870 &attr_configure.attr, 871 #endif 872 &attr_address.attr, 873 &attr_polarization.attr, 874 NULL, 875 }; 876 877 static struct attribute_group cpu_common_attr_group = { 878 .attrs = cpu_common_attrs, 879 }; 880 881 static ssize_t show_capability(struct sys_device *dev, 882 struct sysdev_attribute *attr, char *buf) 883 { 884 unsigned int capability; 885 int rc; 886 887 rc = get_cpu_capability(&capability); 888 if (rc) 889 return rc; 890 return sprintf(buf, "%u\n", capability); 891 } 892 static SYSDEV_ATTR(capability, 0444, show_capability, NULL); 893 894 static ssize_t show_idle_count(struct sys_device *dev, 895 struct sysdev_attribute *attr, char *buf) 896 { 897 struct s390_idle_data *idle; 898 unsigned long long idle_count; 899 unsigned int sequence; 900 901 idle = &per_cpu(s390_idle, dev->id); 902 repeat: 903 sequence = idle->sequence; 904 smp_rmb(); 905 if (sequence & 1) 906 goto repeat; 907 idle_count = idle->idle_count; 908 if (idle->idle_enter) 909 idle_count++; 910 smp_rmb(); 911 if (idle->sequence != sequence) 912 goto repeat; 913 return sprintf(buf, "%llu\n", idle_count); 914 } 915 static SYSDEV_ATTR(idle_count, 0444, show_idle_count, NULL); 916 917 static ssize_t show_idle_time(struct sys_device *dev, 918 struct sysdev_attribute *attr, char *buf) 919 { 920 struct s390_idle_data *idle; 921 unsigned long long now, idle_time, idle_enter; 922 unsigned int sequence; 923 924 idle = &per_cpu(s390_idle, dev->id); 925 now = get_clock(); 926 repeat: 927 sequence = idle->sequence; 928 smp_rmb(); 929 if (sequence & 1) 930 goto repeat; 931 idle_time = idle->idle_time; 932 idle_enter = idle->idle_enter; 933 if (idle_enter != 0ULL && idle_enter < now) 934 idle_time += now - idle_enter; 935 smp_rmb(); 936 if (idle->sequence != sequence) 937 goto repeat; 938 return sprintf(buf, "%llu\n", idle_time >> 12); 939 } 940 static SYSDEV_ATTR(idle_time_us, 0444, show_idle_time, NULL); 941 942 static struct attribute *cpu_online_attrs[] = { 943 &attr_capability.attr, 944 &attr_idle_count.attr, 945 &attr_idle_time_us.attr, 946 NULL, 947 }; 948 949 static struct attribute_group cpu_online_attr_group = { 950 .attrs = cpu_online_attrs, 951 }; 952 953 static int __cpuinit smp_cpu_notify(struct notifier_block *self, 954 unsigned long action, void *hcpu) 955 { 956 unsigned int cpu = (unsigned int)(long)hcpu; 957 struct cpu *c = &per_cpu(cpu_devices, cpu); 958 struct sys_device *s = &c->sysdev; 959 struct s390_idle_data *idle; 960 int err = 0; 961 962 switch (action) { 963 case CPU_ONLINE: 964 case CPU_ONLINE_FROZEN: 965 idle = &per_cpu(s390_idle, cpu); 966 memset(idle, 0, sizeof(struct s390_idle_data)); 967 err = sysfs_create_group(&s->kobj, &cpu_online_attr_group); 968 break; 969 case CPU_DEAD: 970 case CPU_DEAD_FROZEN: 971 sysfs_remove_group(&s->kobj, &cpu_online_attr_group); 972 break; 973 } 974 return notifier_from_errno(err); 975 } 976 977 static struct notifier_block __cpuinitdata smp_cpu_nb = { 978 .notifier_call = smp_cpu_notify, 979 }; 980 981 static int __devinit smp_add_present_cpu(int cpu) 982 { 983 struct cpu *c = &per_cpu(cpu_devices, cpu); 984 struct sys_device *s = &c->sysdev; 985 int rc; 986 987 c->hotpluggable = 1; 988 rc = register_cpu(c, cpu); 989 if (rc) 990 goto out; 991 rc = sysfs_create_group(&s->kobj, &cpu_common_attr_group); 992 if (rc) 993 goto out_cpu; 994 if (!cpu_online(cpu)) 995 goto out; 996 rc = sysfs_create_group(&s->kobj, &cpu_online_attr_group); 997 if (!rc) 998 return 0; 999 sysfs_remove_group(&s->kobj, &cpu_common_attr_group); 1000 out_cpu: 1001 #ifdef CONFIG_HOTPLUG_CPU 1002 unregister_cpu(c); 1003 #endif 1004 out: 1005 return rc; 1006 } 1007 1008 #ifdef CONFIG_HOTPLUG_CPU 1009 1010 int __ref smp_rescan_cpus(void) 1011 { 1012 cpumask_t newcpus; 1013 int cpu; 1014 int rc; 1015 1016 get_online_cpus(); 1017 mutex_lock(&smp_cpu_state_mutex); 1018 cpumask_copy(&newcpus, cpu_present_mask); 1019 rc = __smp_rescan_cpus(); 1020 if (rc) 1021 goto out; 1022 cpumask_andnot(&newcpus, cpu_present_mask, &newcpus); 1023 for_each_cpu(cpu, &newcpus) { 1024 rc = smp_add_present_cpu(cpu); 1025 if (rc) 1026 set_cpu_present(cpu, false); 1027 } 1028 rc = 0; 1029 out: 1030 mutex_unlock(&smp_cpu_state_mutex); 1031 put_online_cpus(); 1032 if (!cpumask_empty(&newcpus)) 1033 topology_schedule_update(); 1034 return rc; 1035 } 1036 1037 static ssize_t __ref rescan_store(struct sysdev_class *class, 1038 struct sysdev_class_attribute *attr, 1039 const char *buf, 1040 size_t count) 1041 { 1042 int rc; 1043 1044 rc = smp_rescan_cpus(); 1045 return rc ? rc : count; 1046 } 1047 static SYSDEV_CLASS_ATTR(rescan, 0200, NULL, rescan_store); 1048 #endif /* CONFIG_HOTPLUG_CPU */ 1049 1050 static ssize_t dispatching_show(struct sysdev_class *class, 1051 struct sysdev_class_attribute *attr, 1052 char *buf) 1053 { 1054 ssize_t count; 1055 1056 mutex_lock(&smp_cpu_state_mutex); 1057 count = sprintf(buf, "%d\n", cpu_management); 1058 mutex_unlock(&smp_cpu_state_mutex); 1059 return count; 1060 } 1061 1062 static ssize_t dispatching_store(struct sysdev_class *dev, 1063 struct sysdev_class_attribute *attr, 1064 const char *buf, 1065 size_t count) 1066 { 1067 int val, rc; 1068 char delim; 1069 1070 if (sscanf(buf, "%d %c", &val, &delim) != 1) 1071 return -EINVAL; 1072 if (val != 0 && val != 1) 1073 return -EINVAL; 1074 rc = 0; 1075 get_online_cpus(); 1076 mutex_lock(&smp_cpu_state_mutex); 1077 if (cpu_management == val) 1078 goto out; 1079 rc = topology_set_cpu_management(val); 1080 if (!rc) 1081 cpu_management = val; 1082 out: 1083 mutex_unlock(&smp_cpu_state_mutex); 1084 put_online_cpus(); 1085 return rc ? rc : count; 1086 } 1087 static SYSDEV_CLASS_ATTR(dispatching, 0644, dispatching_show, 1088 dispatching_store); 1089 1090 static int __init topology_init(void) 1091 { 1092 int cpu; 1093 int rc; 1094 1095 register_cpu_notifier(&smp_cpu_nb); 1096 1097 #ifdef CONFIG_HOTPLUG_CPU 1098 rc = sysdev_class_create_file(&cpu_sysdev_class, &attr_rescan); 1099 if (rc) 1100 return rc; 1101 #endif 1102 rc = sysdev_class_create_file(&cpu_sysdev_class, &attr_dispatching); 1103 if (rc) 1104 return rc; 1105 for_each_present_cpu(cpu) { 1106 rc = smp_add_present_cpu(cpu); 1107 if (rc) 1108 return rc; 1109 } 1110 return 0; 1111 } 1112 subsys_initcall(topology_init); 1113