1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * SMP related functions 4 * 5 * Copyright IBM Corp. 1999, 2012 6 * Author(s): Denis Joseph Barrow, 7 * Martin Schwidefsky <schwidefsky@de.ibm.com>, 8 * Heiko Carstens <heiko.carstens@de.ibm.com>, 9 * 10 * based on other smp stuff by 11 * (c) 1995 Alan Cox, CymruNET Ltd <alan@cymru.net> 12 * (c) 1998 Ingo Molnar 13 * 14 * The code outside of smp.c uses logical cpu numbers, only smp.c does 15 * the translation of logical to physical cpu ids. All new code that 16 * operates on physical cpu numbers needs to go into smp.c. 17 */ 18 19 #define KMSG_COMPONENT "cpu" 20 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt 21 22 #include <linux/workqueue.h> 23 #include <linux/memblock.h> 24 #include <linux/export.h> 25 #include <linux/init.h> 26 #include <linux/mm.h> 27 #include <linux/err.h> 28 #include <linux/spinlock.h> 29 #include <linux/kernel_stat.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/kprobes.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 <asm/stacktrace.h> 57 #include <asm/topology.h> 58 #include "entry.h" 59 60 enum { 61 ec_schedule = 0, 62 ec_call_function_single, 63 ec_stop_cpu, 64 }; 65 66 enum { 67 CPU_STATE_STANDBY, 68 CPU_STATE_CONFIGURED, 69 }; 70 71 static DEFINE_PER_CPU(struct cpu *, cpu_device); 72 73 struct pcpu { 74 struct lowcore *lowcore; /* lowcore page(s) for the cpu */ 75 unsigned long ec_mask; /* bit mask for ec_xxx functions */ 76 unsigned long ec_clk; /* sigp timestamp for ec_xxx */ 77 signed char state; /* physical cpu state */ 78 signed char polarization; /* physical polarization */ 79 u16 address; /* physical cpu address */ 80 }; 81 82 static u8 boot_core_type; 83 static struct pcpu pcpu_devices[NR_CPUS]; 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 static int pcpu_alloc_lowcore(struct pcpu *pcpu, int cpu) 190 { 191 unsigned long async_stack, nodat_stack; 192 struct lowcore *lc; 193 194 if (pcpu != &pcpu_devices[0]) { 195 pcpu->lowcore = (struct lowcore *) 196 __get_free_pages(GFP_KERNEL | GFP_DMA, LC_ORDER); 197 nodat_stack = __get_free_pages(GFP_KERNEL, THREAD_SIZE_ORDER); 198 if (!pcpu->lowcore || !nodat_stack) 199 goto out; 200 } else { 201 nodat_stack = pcpu->lowcore->nodat_stack - STACK_INIT_OFFSET; 202 } 203 async_stack = stack_alloc(); 204 if (!async_stack) 205 goto out; 206 lc = pcpu->lowcore; 207 memcpy(lc, &S390_lowcore, 512); 208 memset((char *) lc + 512, 0, sizeof(*lc) - 512); 209 lc->async_stack = async_stack + STACK_INIT_OFFSET; 210 lc->nodat_stack = nodat_stack + STACK_INIT_OFFSET; 211 lc->cpu_nr = cpu; 212 lc->spinlock_lockval = arch_spin_lockval(cpu); 213 lc->spinlock_index = 0; 214 lc->br_r1_trampoline = 0x07f1; /* br %r1 */ 215 if (nmi_alloc_per_cpu(lc)) 216 goto out_async; 217 if (vdso_alloc_per_cpu(lc)) 218 goto out_mcesa; 219 lowcore_ptr[cpu] = lc; 220 pcpu_sigp_retry(pcpu, SIGP_SET_PREFIX, (u32)(unsigned long) lc); 221 return 0; 222 223 out_mcesa: 224 nmi_free_per_cpu(lc); 225 out_async: 226 stack_free(async_stack); 227 out: 228 if (pcpu != &pcpu_devices[0]) { 229 free_pages(nodat_stack, THREAD_SIZE_ORDER); 230 free_pages((unsigned long) pcpu->lowcore, LC_ORDER); 231 } 232 return -ENOMEM; 233 } 234 235 static void pcpu_free_lowcore(struct pcpu *pcpu) 236 { 237 unsigned long async_stack, nodat_stack, lowcore; 238 239 nodat_stack = pcpu->lowcore->nodat_stack - STACK_INIT_OFFSET; 240 async_stack = pcpu->lowcore->async_stack - STACK_INIT_OFFSET; 241 lowcore = (unsigned long) pcpu->lowcore; 242 243 pcpu_sigp_retry(pcpu, SIGP_SET_PREFIX, 0); 244 lowcore_ptr[pcpu - pcpu_devices] = NULL; 245 vdso_free_per_cpu(pcpu->lowcore); 246 nmi_free_per_cpu(pcpu->lowcore); 247 stack_free(async_stack); 248 if (pcpu == &pcpu_devices[0]) 249 return; 250 free_pages(nodat_stack, THREAD_SIZE_ORDER); 251 free_pages(lowcore, LC_ORDER); 252 } 253 254 static void pcpu_prepare_secondary(struct pcpu *pcpu, int cpu) 255 { 256 struct lowcore *lc = pcpu->lowcore; 257 258 cpumask_set_cpu(cpu, &init_mm.context.cpu_attach_mask); 259 cpumask_set_cpu(cpu, mm_cpumask(&init_mm)); 260 lc->cpu_nr = cpu; 261 lc->spinlock_lockval = arch_spin_lockval(cpu); 262 lc->spinlock_index = 0; 263 lc->percpu_offset = __per_cpu_offset[cpu]; 264 lc->kernel_asce = S390_lowcore.kernel_asce; 265 lc->user_asce = S390_lowcore.kernel_asce; 266 lc->machine_flags = S390_lowcore.machine_flags; 267 lc->user_timer = lc->system_timer = 268 lc->steal_timer = lc->avg_steal_timer = 0; 269 __ctl_store(lc->cregs_save_area, 0, 15); 270 lc->cregs_save_area[1] = lc->kernel_asce; 271 lc->cregs_save_area[7] = lc->vdso_asce; 272 save_access_regs((unsigned int *) lc->access_regs_save_area); 273 memcpy(lc->stfle_fac_list, S390_lowcore.stfle_fac_list, 274 sizeof(lc->stfle_fac_list)); 275 memcpy(lc->alt_stfle_fac_list, S390_lowcore.alt_stfle_fac_list, 276 sizeof(lc->alt_stfle_fac_list)); 277 arch_spin_lock_setup(cpu); 278 } 279 280 static void pcpu_attach_task(struct pcpu *pcpu, struct task_struct *tsk) 281 { 282 struct lowcore *lc = pcpu->lowcore; 283 284 lc->kernel_stack = (unsigned long) task_stack_page(tsk) 285 + THREAD_SIZE - STACK_FRAME_OVERHEAD - sizeof(struct pt_regs); 286 lc->current_task = (unsigned long) tsk; 287 lc->lpp = LPP_MAGIC; 288 lc->current_pid = tsk->pid; 289 lc->user_timer = tsk->thread.user_timer; 290 lc->guest_timer = tsk->thread.guest_timer; 291 lc->system_timer = tsk->thread.system_timer; 292 lc->hardirq_timer = tsk->thread.hardirq_timer; 293 lc->softirq_timer = tsk->thread.softirq_timer; 294 lc->steal_timer = 0; 295 } 296 297 static void pcpu_start_fn(struct pcpu *pcpu, void (*func)(void *), void *data) 298 { 299 struct lowcore *lc = pcpu->lowcore; 300 301 lc->restart_stack = lc->nodat_stack; 302 lc->restart_fn = (unsigned long) func; 303 lc->restart_data = (unsigned long) data; 304 lc->restart_source = -1UL; 305 pcpu_sigp_retry(pcpu, SIGP_RESTART, 0); 306 } 307 308 /* 309 * Call function via PSW restart on pcpu and stop the current cpu. 310 */ 311 static void __pcpu_delegate(void (*func)(void*), void *data) 312 { 313 func(data); /* should not return */ 314 } 315 316 static void __no_sanitize_address pcpu_delegate(struct pcpu *pcpu, 317 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 | PSW_MASK_DAT); 324 if (pcpu->address == source_cpu) 325 CALL_ON_STACK(__pcpu_delegate, stack, 2, func, data); 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 __bpon(); 334 asm volatile( 335 "0: sigp 0,%0,%2 # sigp restart to target cpu\n" 336 " brc 2,0b # busy, try again\n" 337 "1: sigp 0,%1,%3 # sigp stop to current cpu\n" 338 " brc 2,1b # busy, try again\n" 339 : : "d" (pcpu->address), "d" (source_cpu), 340 "K" (SIGP_RESTART), "K" (SIGP_STOP) 341 : "0", "1", "cc"); 342 for (;;) ; 343 } 344 345 /* 346 * Enable additional logical cpus for multi-threading. 347 */ 348 static int pcpu_set_smt(unsigned int mtid) 349 { 350 int cc; 351 352 if (smp_cpu_mtid == mtid) 353 return 0; 354 cc = __pcpu_sigp(0, SIGP_SET_MULTI_THREADING, mtid, NULL); 355 if (cc == 0) { 356 smp_cpu_mtid = mtid; 357 smp_cpu_mt_shift = 0; 358 while (smp_cpu_mtid >= (1U << smp_cpu_mt_shift)) 359 smp_cpu_mt_shift++; 360 pcpu_devices[0].address = stap(); 361 } 362 return cc; 363 } 364 365 /* 366 * Call function on an online CPU. 367 */ 368 void smp_call_online_cpu(void (*func)(void *), void *data) 369 { 370 struct pcpu *pcpu; 371 372 /* Use the current cpu if it is online. */ 373 pcpu = pcpu_find_address(cpu_online_mask, stap()); 374 if (!pcpu) 375 /* Use the first online cpu. */ 376 pcpu = pcpu_devices + cpumask_first(cpu_online_mask); 377 pcpu_delegate(pcpu, func, data, (unsigned long) restart_stack); 378 } 379 380 /* 381 * Call function on the ipl CPU. 382 */ 383 void smp_call_ipl_cpu(void (*func)(void *), void *data) 384 { 385 struct lowcore *lc = pcpu_devices->lowcore; 386 387 if (pcpu_devices[0].address == stap()) 388 lc = &S390_lowcore; 389 390 pcpu_delegate(&pcpu_devices[0], func, data, 391 lc->nodat_stack); 392 } 393 394 int smp_find_processor_id(u16 address) 395 { 396 int cpu; 397 398 for_each_present_cpu(cpu) 399 if (pcpu_devices[cpu].address == address) 400 return cpu; 401 return -1; 402 } 403 404 bool arch_vcpu_is_preempted(int cpu) 405 { 406 if (test_cpu_flag_of(CIF_ENABLED_WAIT, cpu)) 407 return false; 408 if (pcpu_running(pcpu_devices + cpu)) 409 return false; 410 return true; 411 } 412 EXPORT_SYMBOL(arch_vcpu_is_preempted); 413 414 void smp_yield_cpu(int cpu) 415 { 416 if (MACHINE_HAS_DIAG9C) { 417 diag_stat_inc_norecursion(DIAG_STAT_X09C); 418 asm volatile("diag %0,0,0x9c" 419 : : "d" (pcpu_devices[cpu].address)); 420 } else if (MACHINE_HAS_DIAG44 && !smp_cpu_mtid) { 421 diag_stat_inc_norecursion(DIAG_STAT_X044); 422 asm volatile("diag 0,0,0x44"); 423 } 424 } 425 426 /* 427 * Send cpus emergency shutdown signal. This gives the cpus the 428 * opportunity to complete outstanding interrupts. 429 */ 430 void notrace smp_emergency_stop(void) 431 { 432 cpumask_t cpumask; 433 u64 end; 434 int cpu; 435 436 cpumask_copy(&cpumask, cpu_online_mask); 437 cpumask_clear_cpu(smp_processor_id(), &cpumask); 438 439 end = get_tod_clock() + (1000000UL << 12); 440 for_each_cpu(cpu, &cpumask) { 441 struct pcpu *pcpu = pcpu_devices + cpu; 442 set_bit(ec_stop_cpu, &pcpu->ec_mask); 443 while (__pcpu_sigp(pcpu->address, SIGP_EMERGENCY_SIGNAL, 444 0, NULL) == SIGP_CC_BUSY && 445 get_tod_clock() < end) 446 cpu_relax(); 447 } 448 while (get_tod_clock() < end) { 449 for_each_cpu(cpu, &cpumask) 450 if (pcpu_stopped(pcpu_devices + cpu)) 451 cpumask_clear_cpu(cpu, &cpumask); 452 if (cpumask_empty(&cpumask)) 453 break; 454 cpu_relax(); 455 } 456 } 457 NOKPROBE_SYMBOL(smp_emergency_stop); 458 459 /* 460 * Stop all cpus but the current one. 461 */ 462 void smp_send_stop(void) 463 { 464 int cpu; 465 466 /* Disable all interrupts/machine checks */ 467 __load_psw_mask(PSW_KERNEL_BITS | PSW_MASK_DAT); 468 trace_hardirqs_off(); 469 470 debug_set_critical(); 471 472 if (oops_in_progress) 473 smp_emergency_stop(); 474 475 /* stop all processors */ 476 for_each_online_cpu(cpu) { 477 if (cpu == smp_processor_id()) 478 continue; 479 pcpu_sigp_retry(pcpu_devices + cpu, SIGP_STOP, 0); 480 while (!pcpu_stopped(pcpu_devices + cpu)) 481 cpu_relax(); 482 } 483 } 484 485 /* 486 * This is the main routine where commands issued by other 487 * cpus are handled. 488 */ 489 static void smp_handle_ext_call(void) 490 { 491 unsigned long bits; 492 493 /* handle bit signal external calls */ 494 bits = xchg(&pcpu_devices[smp_processor_id()].ec_mask, 0); 495 if (test_bit(ec_stop_cpu, &bits)) 496 smp_stop_cpu(); 497 if (test_bit(ec_schedule, &bits)) 498 scheduler_ipi(); 499 if (test_bit(ec_call_function_single, &bits)) 500 generic_smp_call_function_single_interrupt(); 501 } 502 503 static void do_ext_call_interrupt(struct ext_code ext_code, 504 unsigned int param32, unsigned long param64) 505 { 506 inc_irq_stat(ext_code.code == 0x1202 ? IRQEXT_EXC : IRQEXT_EMS); 507 smp_handle_ext_call(); 508 } 509 510 void arch_send_call_function_ipi_mask(const struct cpumask *mask) 511 { 512 int cpu; 513 514 for_each_cpu(cpu, mask) 515 pcpu_ec_call(pcpu_devices + cpu, ec_call_function_single); 516 } 517 518 void arch_send_call_function_single_ipi(int cpu) 519 { 520 pcpu_ec_call(pcpu_devices + cpu, ec_call_function_single); 521 } 522 523 /* 524 * this function sends a 'reschedule' IPI to another CPU. 525 * it goes straight through and wastes no time serializing 526 * anything. Worst case is that we lose a reschedule ... 527 */ 528 void smp_send_reschedule(int cpu) 529 { 530 pcpu_ec_call(pcpu_devices + cpu, ec_schedule); 531 } 532 533 /* 534 * parameter area for the set/clear control bit callbacks 535 */ 536 struct ec_creg_mask_parms { 537 unsigned long orval; 538 unsigned long andval; 539 int cr; 540 }; 541 542 /* 543 * callback for setting/clearing control bits 544 */ 545 static void smp_ctl_bit_callback(void *info) 546 { 547 struct ec_creg_mask_parms *pp = info; 548 unsigned long cregs[16]; 549 550 __ctl_store(cregs, 0, 15); 551 cregs[pp->cr] = (cregs[pp->cr] & pp->andval) | pp->orval; 552 __ctl_load(cregs, 0, 15); 553 } 554 555 /* 556 * Set a bit in a control register of all cpus 557 */ 558 void smp_ctl_set_bit(int cr, int bit) 559 { 560 struct ec_creg_mask_parms parms = { 1UL << bit, -1UL, cr }; 561 562 on_each_cpu(smp_ctl_bit_callback, &parms, 1); 563 } 564 EXPORT_SYMBOL(smp_ctl_set_bit); 565 566 /* 567 * Clear a bit in a control register of all cpus 568 */ 569 void smp_ctl_clear_bit(int cr, int bit) 570 { 571 struct ec_creg_mask_parms parms = { 0, ~(1UL << bit), cr }; 572 573 on_each_cpu(smp_ctl_bit_callback, &parms, 1); 574 } 575 EXPORT_SYMBOL(smp_ctl_clear_bit); 576 577 #ifdef CONFIG_CRASH_DUMP 578 579 int smp_store_status(int cpu) 580 { 581 struct pcpu *pcpu = pcpu_devices + cpu; 582 unsigned long pa; 583 584 pa = __pa(&pcpu->lowcore->floating_pt_save_area); 585 if (__pcpu_sigp_relax(pcpu->address, SIGP_STORE_STATUS_AT_ADDRESS, 586 pa) != SIGP_CC_ORDER_CODE_ACCEPTED) 587 return -EIO; 588 if (!MACHINE_HAS_VX && !MACHINE_HAS_GS) 589 return 0; 590 pa = __pa(pcpu->lowcore->mcesad & MCESA_ORIGIN_MASK); 591 if (MACHINE_HAS_GS) 592 pa |= pcpu->lowcore->mcesad & MCESA_LC_MASK; 593 if (__pcpu_sigp_relax(pcpu->address, SIGP_STORE_ADDITIONAL_STATUS, 594 pa) != SIGP_CC_ORDER_CODE_ACCEPTED) 595 return -EIO; 596 return 0; 597 } 598 599 /* 600 * Collect CPU state of the previous, crashed system. 601 * There are four cases: 602 * 1) standard zfcp dump 603 * condition: OLDMEM_BASE == NULL && ipl_info.type == IPL_TYPE_FCP_DUMP 604 * The state for all CPUs except the boot CPU needs to be collected 605 * with sigp stop-and-store-status. The boot CPU state is located in 606 * the absolute lowcore of the memory stored in the HSA. The zcore code 607 * will copy the boot CPU state from the HSA. 608 * 2) stand-alone kdump for SCSI (zfcp dump with swapped memory) 609 * condition: OLDMEM_BASE != NULL && ipl_info.type == IPL_TYPE_FCP_DUMP 610 * The state for all CPUs except the boot CPU needs to be collected 611 * with sigp stop-and-store-status. The firmware or the boot-loader 612 * stored the registers of the boot CPU in the absolute lowcore in the 613 * memory of the old system. 614 * 3) kdump and the old kernel did not store the CPU state, 615 * or stand-alone kdump for DASD 616 * condition: OLDMEM_BASE != NULL && !is_kdump_kernel() 617 * The state for all CPUs except the boot CPU needs to be collected 618 * with sigp stop-and-store-status. The kexec code or the boot-loader 619 * stored the registers of the boot CPU in the memory of the old system. 620 * 4) kdump and the old kernel stored the CPU state 621 * condition: OLDMEM_BASE != NULL && is_kdump_kernel() 622 * This case does not exist for s390 anymore, setup_arch explicitly 623 * deactivates the elfcorehdr= kernel parameter 624 */ 625 static __init void smp_save_cpu_vxrs(struct save_area *sa, u16 addr, 626 bool is_boot_cpu, unsigned long page) 627 { 628 __vector128 *vxrs = (__vector128 *) page; 629 630 if (is_boot_cpu) 631 vxrs = boot_cpu_vector_save_area; 632 else 633 __pcpu_sigp_relax(addr, SIGP_STORE_ADDITIONAL_STATUS, page); 634 save_area_add_vxrs(sa, vxrs); 635 } 636 637 static __init void smp_save_cpu_regs(struct save_area *sa, u16 addr, 638 bool is_boot_cpu, unsigned long page) 639 { 640 void *regs = (void *) page; 641 642 if (is_boot_cpu) 643 copy_oldmem_kernel(regs, (void *) __LC_FPREGS_SAVE_AREA, 512); 644 else 645 __pcpu_sigp_relax(addr, SIGP_STORE_STATUS_AT_ADDRESS, page); 646 save_area_add_regs(sa, regs); 647 } 648 649 void __init smp_save_dump_cpus(void) 650 { 651 int addr, boot_cpu_addr, max_cpu_addr; 652 struct save_area *sa; 653 unsigned long page; 654 bool is_boot_cpu; 655 656 if (!(OLDMEM_BASE || ipl_info.type == IPL_TYPE_FCP_DUMP)) 657 /* No previous system present, normal boot. */ 658 return; 659 /* Allocate a page as dumping area for the store status sigps */ 660 page = memblock_phys_alloc_range(PAGE_SIZE, PAGE_SIZE, 0, 1UL << 31); 661 if (!page) 662 panic("ERROR: Failed to allocate %lx bytes below %lx\n", 663 PAGE_SIZE, 1UL << 31); 664 665 /* Set multi-threading state to the previous system. */ 666 pcpu_set_smt(sclp.mtid_prev); 667 boot_cpu_addr = stap(); 668 max_cpu_addr = SCLP_MAX_CORES << sclp.mtid_prev; 669 for (addr = 0; addr <= max_cpu_addr; addr++) { 670 if (__pcpu_sigp_relax(addr, SIGP_SENSE, 0) == 671 SIGP_CC_NOT_OPERATIONAL) 672 continue; 673 is_boot_cpu = (addr == boot_cpu_addr); 674 /* Allocate save area */ 675 sa = save_area_alloc(is_boot_cpu); 676 if (!sa) 677 panic("could not allocate memory for save area\n"); 678 if (MACHINE_HAS_VX) 679 /* Get the vector registers */ 680 smp_save_cpu_vxrs(sa, addr, is_boot_cpu, page); 681 /* 682 * For a zfcp dump OLDMEM_BASE == NULL and the registers 683 * of the boot CPU are stored in the HSA. To retrieve 684 * these registers an SCLP request is required which is 685 * done by drivers/s390/char/zcore.c:init_cpu_info() 686 */ 687 if (!is_boot_cpu || OLDMEM_BASE) 688 /* Get the CPU registers */ 689 smp_save_cpu_regs(sa, addr, is_boot_cpu, page); 690 } 691 memblock_free(page, PAGE_SIZE); 692 diag_dma_ops.diag308_reset(); 693 pcpu_set_smt(0); 694 } 695 #endif /* CONFIG_CRASH_DUMP */ 696 697 void smp_cpu_set_polarization(int cpu, int val) 698 { 699 pcpu_devices[cpu].polarization = val; 700 } 701 702 int smp_cpu_get_polarization(int cpu) 703 { 704 return pcpu_devices[cpu].polarization; 705 } 706 707 static void __ref smp_get_core_info(struct sclp_core_info *info, int early) 708 { 709 static int use_sigp_detection; 710 int address; 711 712 if (use_sigp_detection || sclp_get_core_info(info, early)) { 713 use_sigp_detection = 1; 714 for (address = 0; 715 address < (SCLP_MAX_CORES << smp_cpu_mt_shift); 716 address += (1U << smp_cpu_mt_shift)) { 717 if (__pcpu_sigp_relax(address, SIGP_SENSE, 0) == 718 SIGP_CC_NOT_OPERATIONAL) 719 continue; 720 info->core[info->configured].core_id = 721 address >> smp_cpu_mt_shift; 722 info->configured++; 723 } 724 info->combined = info->configured; 725 } 726 } 727 728 static int smp_add_present_cpu(int cpu); 729 730 static int smp_add_core(struct sclp_core_entry *core, cpumask_t *avail, 731 bool configured, bool early) 732 { 733 struct pcpu *pcpu; 734 int cpu, nr, i; 735 u16 address; 736 737 nr = 0; 738 if (sclp.has_core_type && core->type != boot_core_type) 739 return nr; 740 cpu = cpumask_first(avail); 741 address = core->core_id << smp_cpu_mt_shift; 742 for (i = 0; (i <= smp_cpu_mtid) && (cpu < nr_cpu_ids); i++) { 743 if (pcpu_find_address(cpu_present_mask, address + i)) 744 continue; 745 pcpu = pcpu_devices + cpu; 746 pcpu->address = address + i; 747 if (configured) 748 pcpu->state = CPU_STATE_CONFIGURED; 749 else 750 pcpu->state = CPU_STATE_STANDBY; 751 smp_cpu_set_polarization(cpu, POLARIZATION_UNKNOWN); 752 set_cpu_present(cpu, true); 753 if (!early && smp_add_present_cpu(cpu) != 0) 754 set_cpu_present(cpu, false); 755 else 756 nr++; 757 cpumask_clear_cpu(cpu, avail); 758 cpu = cpumask_next(cpu, avail); 759 } 760 return nr; 761 } 762 763 static int __smp_rescan_cpus(struct sclp_core_info *info, bool early) 764 { 765 struct sclp_core_entry *core; 766 cpumask_t avail; 767 bool configured; 768 u16 core_id; 769 int nr, i; 770 771 nr = 0; 772 cpumask_xor(&avail, cpu_possible_mask, cpu_present_mask); 773 /* 774 * Add IPL core first (which got logical CPU number 0) to make sure 775 * that all SMT threads get subsequent logical CPU numbers. 776 */ 777 if (early) { 778 core_id = pcpu_devices[0].address >> smp_cpu_mt_shift; 779 for (i = 0; i < info->configured; i++) { 780 core = &info->core[i]; 781 if (core->core_id == core_id) { 782 nr += smp_add_core(core, &avail, true, early); 783 break; 784 } 785 } 786 } 787 for (i = 0; i < info->combined; i++) { 788 configured = i < info->configured; 789 nr += smp_add_core(&info->core[i], &avail, configured, early); 790 } 791 return nr; 792 } 793 794 void __init smp_detect_cpus(void) 795 { 796 unsigned int cpu, mtid, c_cpus, s_cpus; 797 struct sclp_core_info *info; 798 u16 address; 799 800 /* Get CPU information */ 801 info = memblock_alloc(sizeof(*info), 8); 802 if (!info) 803 panic("%s: Failed to allocate %zu bytes align=0x%x\n", 804 __func__, sizeof(*info), 8); 805 smp_get_core_info(info, 1); 806 /* Find boot CPU type */ 807 if (sclp.has_core_type) { 808 address = stap(); 809 for (cpu = 0; cpu < info->combined; cpu++) 810 if (info->core[cpu].core_id == address) { 811 /* The boot cpu dictates the cpu type. */ 812 boot_core_type = info->core[cpu].type; 813 break; 814 } 815 if (cpu >= info->combined) 816 panic("Could not find boot CPU type"); 817 } 818 819 /* Set multi-threading state for the current system */ 820 mtid = boot_core_type ? sclp.mtid : sclp.mtid_cp; 821 mtid = (mtid < smp_max_threads) ? mtid : smp_max_threads - 1; 822 pcpu_set_smt(mtid); 823 824 /* Print number of CPUs */ 825 c_cpus = s_cpus = 0; 826 for (cpu = 0; cpu < info->combined; cpu++) { 827 if (sclp.has_core_type && 828 info->core[cpu].type != boot_core_type) 829 continue; 830 if (cpu < info->configured) 831 c_cpus += smp_cpu_mtid + 1; 832 else 833 s_cpus += smp_cpu_mtid + 1; 834 } 835 pr_info("%d configured CPUs, %d standby CPUs\n", c_cpus, s_cpus); 836 837 /* Add CPUs present at boot */ 838 get_online_cpus(); 839 __smp_rescan_cpus(info, true); 840 put_online_cpus(); 841 memblock_free_early((unsigned long)info, sizeof(*info)); 842 } 843 844 static void smp_init_secondary(void) 845 { 846 int cpu = smp_processor_id(); 847 848 S390_lowcore.last_update_clock = get_tod_clock(); 849 restore_access_regs(S390_lowcore.access_regs_save_area); 850 set_cpu_flag(CIF_ASCE_PRIMARY); 851 set_cpu_flag(CIF_ASCE_SECONDARY); 852 cpu_init(); 853 preempt_disable(); 854 init_cpu_timer(); 855 vtime_init(); 856 pfault_init(); 857 notify_cpu_starting(smp_processor_id()); 858 if (topology_cpu_dedicated(cpu)) 859 set_cpu_flag(CIF_DEDICATED_CPU); 860 else 861 clear_cpu_flag(CIF_DEDICATED_CPU); 862 set_cpu_online(smp_processor_id(), true); 863 inc_irq_stat(CPU_RST); 864 local_irq_enable(); 865 cpu_startup_entry(CPUHP_AP_ONLINE_IDLE); 866 } 867 868 /* 869 * Activate a secondary processor. 870 */ 871 static void __no_sanitize_address smp_start_secondary(void *cpuvoid) 872 { 873 S390_lowcore.restart_stack = (unsigned long) restart_stack; 874 S390_lowcore.restart_fn = (unsigned long) do_restart; 875 S390_lowcore.restart_data = 0; 876 S390_lowcore.restart_source = -1UL; 877 __ctl_load(S390_lowcore.cregs_save_area, 0, 15); 878 __load_psw_mask(PSW_KERNEL_BITS | PSW_MASK_DAT); 879 CALL_ON_STACK_NORETURN(smp_init_secondary, S390_lowcore.kernel_stack); 880 } 881 882 /* Upping and downing of CPUs */ 883 int __cpu_up(unsigned int cpu, struct task_struct *tidle) 884 { 885 struct pcpu *pcpu; 886 int base, i, rc; 887 888 pcpu = pcpu_devices + cpu; 889 if (pcpu->state != CPU_STATE_CONFIGURED) 890 return -EIO; 891 base = smp_get_base_cpu(cpu); 892 for (i = 0; i <= smp_cpu_mtid; i++) { 893 if (base + i < nr_cpu_ids) 894 if (cpu_online(base + i)) 895 break; 896 } 897 /* 898 * If this is the first CPU of the core to get online 899 * do an initial CPU reset. 900 */ 901 if (i > smp_cpu_mtid && 902 pcpu_sigp_retry(pcpu_devices + base, SIGP_INITIAL_CPU_RESET, 0) != 903 SIGP_CC_ORDER_CODE_ACCEPTED) 904 return -EIO; 905 906 rc = pcpu_alloc_lowcore(pcpu, cpu); 907 if (rc) 908 return rc; 909 pcpu_prepare_secondary(pcpu, cpu); 910 pcpu_attach_task(pcpu, tidle); 911 pcpu_start_fn(pcpu, smp_start_secondary, NULL); 912 /* Wait until cpu puts itself in the online & active maps */ 913 while (!cpu_online(cpu)) 914 cpu_relax(); 915 return 0; 916 } 917 918 static unsigned int setup_possible_cpus __initdata; 919 920 static int __init _setup_possible_cpus(char *s) 921 { 922 get_option(&s, &setup_possible_cpus); 923 return 0; 924 } 925 early_param("possible_cpus", _setup_possible_cpus); 926 927 int __cpu_disable(void) 928 { 929 unsigned long cregs[16]; 930 931 /* Handle possible pending IPIs */ 932 smp_handle_ext_call(); 933 set_cpu_online(smp_processor_id(), false); 934 /* Disable pseudo page faults on this cpu. */ 935 pfault_fini(); 936 /* Disable interrupt sources via control register. */ 937 __ctl_store(cregs, 0, 15); 938 cregs[0] &= ~0x0000ee70UL; /* disable all external interrupts */ 939 cregs[6] &= ~0xff000000UL; /* disable all I/O interrupts */ 940 cregs[14] &= ~0x1f000000UL; /* disable most machine checks */ 941 __ctl_load(cregs, 0, 15); 942 clear_cpu_flag(CIF_NOHZ_DELAY); 943 return 0; 944 } 945 946 void __cpu_die(unsigned int cpu) 947 { 948 struct pcpu *pcpu; 949 950 /* Wait until target cpu is down */ 951 pcpu = pcpu_devices + cpu; 952 while (!pcpu_stopped(pcpu)) 953 cpu_relax(); 954 pcpu_free_lowcore(pcpu); 955 cpumask_clear_cpu(cpu, mm_cpumask(&init_mm)); 956 cpumask_clear_cpu(cpu, &init_mm.context.cpu_attach_mask); 957 } 958 959 void __noreturn cpu_die(void) 960 { 961 idle_task_exit(); 962 __bpon(); 963 pcpu_sigp_retry(pcpu_devices + smp_processor_id(), SIGP_STOP, 0); 964 for (;;) ; 965 } 966 967 void __init smp_fill_possible_mask(void) 968 { 969 unsigned int possible, sclp_max, cpu; 970 971 sclp_max = max(sclp.mtid, sclp.mtid_cp) + 1; 972 sclp_max = min(smp_max_threads, sclp_max); 973 sclp_max = (sclp.max_cores * sclp_max) ?: nr_cpu_ids; 974 possible = setup_possible_cpus ?: nr_cpu_ids; 975 possible = min(possible, sclp_max); 976 for (cpu = 0; cpu < possible && cpu < nr_cpu_ids; cpu++) 977 set_cpu_possible(cpu, true); 978 } 979 980 void __init smp_prepare_cpus(unsigned int max_cpus) 981 { 982 /* request the 0x1201 emergency signal external interrupt */ 983 if (register_external_irq(EXT_IRQ_EMERGENCY_SIG, do_ext_call_interrupt)) 984 panic("Couldn't request external interrupt 0x1201"); 985 /* request the 0x1202 external call external interrupt */ 986 if (register_external_irq(EXT_IRQ_EXTERNAL_CALL, do_ext_call_interrupt)) 987 panic("Couldn't request external interrupt 0x1202"); 988 } 989 990 void __init smp_prepare_boot_cpu(void) 991 { 992 struct pcpu *pcpu = pcpu_devices; 993 994 WARN_ON(!cpu_present(0) || !cpu_online(0)); 995 pcpu->state = CPU_STATE_CONFIGURED; 996 pcpu->lowcore = (struct lowcore *)(unsigned long) store_prefix(); 997 S390_lowcore.percpu_offset = __per_cpu_offset[0]; 998 smp_cpu_set_polarization(0, POLARIZATION_UNKNOWN); 999 } 1000 1001 void __init smp_cpus_done(unsigned int max_cpus) 1002 { 1003 } 1004 1005 void __init smp_setup_processor_id(void) 1006 { 1007 pcpu_devices[0].address = stap(); 1008 S390_lowcore.cpu_nr = 0; 1009 S390_lowcore.spinlock_lockval = arch_spin_lockval(0); 1010 S390_lowcore.spinlock_index = 0; 1011 } 1012 1013 /* 1014 * the frequency of the profiling timer can be changed 1015 * by writing a multiplier value into /proc/profile. 1016 * 1017 * usually you want to run this on all CPUs ;) 1018 */ 1019 int setup_profiling_timer(unsigned int multiplier) 1020 { 1021 return 0; 1022 } 1023 1024 static ssize_t cpu_configure_show(struct device *dev, 1025 struct device_attribute *attr, char *buf) 1026 { 1027 ssize_t count; 1028 1029 mutex_lock(&smp_cpu_state_mutex); 1030 count = sprintf(buf, "%d\n", pcpu_devices[dev->id].state); 1031 mutex_unlock(&smp_cpu_state_mutex); 1032 return count; 1033 } 1034 1035 static ssize_t cpu_configure_store(struct device *dev, 1036 struct device_attribute *attr, 1037 const char *buf, size_t count) 1038 { 1039 struct pcpu *pcpu; 1040 int cpu, val, rc, i; 1041 char delim; 1042 1043 if (sscanf(buf, "%d %c", &val, &delim) != 1) 1044 return -EINVAL; 1045 if (val != 0 && val != 1) 1046 return -EINVAL; 1047 get_online_cpus(); 1048 mutex_lock(&smp_cpu_state_mutex); 1049 rc = -EBUSY; 1050 /* disallow configuration changes of online cpus and cpu 0 */ 1051 cpu = dev->id; 1052 cpu = smp_get_base_cpu(cpu); 1053 if (cpu == 0) 1054 goto out; 1055 for (i = 0; i <= smp_cpu_mtid; i++) 1056 if (cpu_online(cpu + i)) 1057 goto out; 1058 pcpu = pcpu_devices + cpu; 1059 rc = 0; 1060 switch (val) { 1061 case 0: 1062 if (pcpu->state != CPU_STATE_CONFIGURED) 1063 break; 1064 rc = sclp_core_deconfigure(pcpu->address >> smp_cpu_mt_shift); 1065 if (rc) 1066 break; 1067 for (i = 0; i <= smp_cpu_mtid; i++) { 1068 if (cpu + i >= nr_cpu_ids || !cpu_present(cpu + i)) 1069 continue; 1070 pcpu[i].state = CPU_STATE_STANDBY; 1071 smp_cpu_set_polarization(cpu + i, 1072 POLARIZATION_UNKNOWN); 1073 } 1074 topology_expect_change(); 1075 break; 1076 case 1: 1077 if (pcpu->state != CPU_STATE_STANDBY) 1078 break; 1079 rc = sclp_core_configure(pcpu->address >> smp_cpu_mt_shift); 1080 if (rc) 1081 break; 1082 for (i = 0; i <= smp_cpu_mtid; i++) { 1083 if (cpu + i >= nr_cpu_ids || !cpu_present(cpu + i)) 1084 continue; 1085 pcpu[i].state = CPU_STATE_CONFIGURED; 1086 smp_cpu_set_polarization(cpu + i, 1087 POLARIZATION_UNKNOWN); 1088 } 1089 topology_expect_change(); 1090 break; 1091 default: 1092 break; 1093 } 1094 out: 1095 mutex_unlock(&smp_cpu_state_mutex); 1096 put_online_cpus(); 1097 return rc ? rc : count; 1098 } 1099 static DEVICE_ATTR(configure, 0644, cpu_configure_show, cpu_configure_store); 1100 1101 static ssize_t show_cpu_address(struct device *dev, 1102 struct device_attribute *attr, char *buf) 1103 { 1104 return sprintf(buf, "%d\n", pcpu_devices[dev->id].address); 1105 } 1106 static DEVICE_ATTR(address, 0444, show_cpu_address, NULL); 1107 1108 static struct attribute *cpu_common_attrs[] = { 1109 &dev_attr_configure.attr, 1110 &dev_attr_address.attr, 1111 NULL, 1112 }; 1113 1114 static struct attribute_group cpu_common_attr_group = { 1115 .attrs = cpu_common_attrs, 1116 }; 1117 1118 static struct attribute *cpu_online_attrs[] = { 1119 &dev_attr_idle_count.attr, 1120 &dev_attr_idle_time_us.attr, 1121 NULL, 1122 }; 1123 1124 static struct attribute_group cpu_online_attr_group = { 1125 .attrs = cpu_online_attrs, 1126 }; 1127 1128 static int smp_cpu_online(unsigned int cpu) 1129 { 1130 struct device *s = &per_cpu(cpu_device, cpu)->dev; 1131 1132 return sysfs_create_group(&s->kobj, &cpu_online_attr_group); 1133 } 1134 static int smp_cpu_pre_down(unsigned int cpu) 1135 { 1136 struct device *s = &per_cpu(cpu_device, cpu)->dev; 1137 1138 sysfs_remove_group(&s->kobj, &cpu_online_attr_group); 1139 return 0; 1140 } 1141 1142 static int smp_add_present_cpu(int cpu) 1143 { 1144 struct device *s; 1145 struct cpu *c; 1146 int rc; 1147 1148 c = kzalloc(sizeof(*c), GFP_KERNEL); 1149 if (!c) 1150 return -ENOMEM; 1151 per_cpu(cpu_device, cpu) = c; 1152 s = &c->dev; 1153 c->hotpluggable = 1; 1154 rc = register_cpu(c, cpu); 1155 if (rc) 1156 goto out; 1157 rc = sysfs_create_group(&s->kobj, &cpu_common_attr_group); 1158 if (rc) 1159 goto out_cpu; 1160 rc = topology_cpu_init(c); 1161 if (rc) 1162 goto out_topology; 1163 return 0; 1164 1165 out_topology: 1166 sysfs_remove_group(&s->kobj, &cpu_common_attr_group); 1167 out_cpu: 1168 unregister_cpu(c); 1169 out: 1170 return rc; 1171 } 1172 1173 int __ref smp_rescan_cpus(void) 1174 { 1175 struct sclp_core_info *info; 1176 int nr; 1177 1178 info = kzalloc(sizeof(*info), GFP_KERNEL); 1179 if (!info) 1180 return -ENOMEM; 1181 smp_get_core_info(info, 0); 1182 get_online_cpus(); 1183 mutex_lock(&smp_cpu_state_mutex); 1184 nr = __smp_rescan_cpus(info, false); 1185 mutex_unlock(&smp_cpu_state_mutex); 1186 put_online_cpus(); 1187 kfree(info); 1188 if (nr) 1189 topology_schedule_update(); 1190 return 0; 1191 } 1192 1193 static ssize_t __ref rescan_store(struct device *dev, 1194 struct device_attribute *attr, 1195 const char *buf, 1196 size_t count) 1197 { 1198 int rc; 1199 1200 rc = lock_device_hotplug_sysfs(); 1201 if (rc) 1202 return rc; 1203 rc = smp_rescan_cpus(); 1204 unlock_device_hotplug(); 1205 return rc ? rc : count; 1206 } 1207 static DEVICE_ATTR_WO(rescan); 1208 1209 static int __init s390_smp_init(void) 1210 { 1211 int cpu, rc = 0; 1212 1213 rc = device_create_file(cpu_subsys.dev_root, &dev_attr_rescan); 1214 if (rc) 1215 return rc; 1216 for_each_present_cpu(cpu) { 1217 rc = smp_add_present_cpu(cpu); 1218 if (rc) 1219 goto out; 1220 } 1221 1222 rc = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "s390/smp:online", 1223 smp_cpu_online, smp_cpu_pre_down); 1224 rc = rc <= 0 ? rc : 0; 1225 out: 1226 return rc; 1227 } 1228 subsys_initcall(s390_smp_init); 1229