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/bootmem.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/kmemleak.h> 31 #include <linux/delay.h> 32 #include <linux/interrupt.h> 33 #include <linux/irqflags.h> 34 #include <linux/cpu.h> 35 #include <linux/slab.h> 36 #include <linux/sched/hotplug.h> 37 #include <linux/sched/task_stack.h> 38 #include <linux/crash_dump.h> 39 #include <linux/memblock.h> 40 #include <linux/kprobes.h> 41 #include <asm/asm-offsets.h> 42 #include <asm/diag.h> 43 #include <asm/switch_to.h> 44 #include <asm/facility.h> 45 #include <asm/ipl.h> 46 #include <asm/setup.h> 47 #include <asm/irq.h> 48 #include <asm/tlbflush.h> 49 #include <asm/vtimer.h> 50 #include <asm/lowcore.h> 51 #include <asm/sclp.h> 52 #include <asm/vdso.h> 53 #include <asm/debug.h> 54 #include <asm/os_info.h> 55 #include <asm/sigp.h> 56 #include <asm/idle.h> 57 #include <asm/nmi.h> 58 #include <asm/topology.h> 59 #include "entry.h" 60 61 enum { 62 ec_schedule = 0, 63 ec_call_function_single, 64 ec_stop_cpu, 65 }; 66 67 enum { 68 CPU_STATE_STANDBY, 69 CPU_STATE_CONFIGURED, 70 }; 71 72 static DEFINE_PER_CPU(struct cpu *, cpu_device); 73 74 struct pcpu { 75 struct lowcore *lowcore; /* lowcore page(s) for the cpu */ 76 unsigned long ec_mask; /* bit mask for ec_xxx functions */ 77 unsigned long ec_clk; /* sigp timestamp for ec_xxx */ 78 signed char state; /* physical cpu state */ 79 signed char polarization; /* physical polarization */ 80 u16 address; /* physical cpu address */ 81 }; 82 83 static u8 boot_core_type; 84 static struct pcpu pcpu_devices[NR_CPUS]; 85 86 unsigned int smp_cpu_mt_shift; 87 EXPORT_SYMBOL(smp_cpu_mt_shift); 88 89 unsigned int smp_cpu_mtid; 90 EXPORT_SYMBOL(smp_cpu_mtid); 91 92 #ifdef CONFIG_CRASH_DUMP 93 __vector128 __initdata boot_cpu_vector_save_area[__NUM_VXRS]; 94 #endif 95 96 static unsigned int smp_max_threads __initdata = -1U; 97 98 static int __init early_nosmt(char *s) 99 { 100 smp_max_threads = 1; 101 return 0; 102 } 103 early_param("nosmt", early_nosmt); 104 105 static int __init early_smt(char *s) 106 { 107 get_option(&s, &smp_max_threads); 108 return 0; 109 } 110 early_param("smt", early_smt); 111 112 /* 113 * The smp_cpu_state_mutex must be held when changing the state or polarization 114 * member of a pcpu data structure within the pcpu_devices arreay. 115 */ 116 DEFINE_MUTEX(smp_cpu_state_mutex); 117 118 /* 119 * Signal processor helper functions. 120 */ 121 static inline int __pcpu_sigp_relax(u16 addr, u8 order, unsigned long parm) 122 { 123 int cc; 124 125 while (1) { 126 cc = __pcpu_sigp(addr, order, parm, NULL); 127 if (cc != SIGP_CC_BUSY) 128 return cc; 129 cpu_relax(); 130 } 131 } 132 133 static int pcpu_sigp_retry(struct pcpu *pcpu, u8 order, u32 parm) 134 { 135 int cc, retry; 136 137 for (retry = 0; ; retry++) { 138 cc = __pcpu_sigp(pcpu->address, order, parm, NULL); 139 if (cc != SIGP_CC_BUSY) 140 break; 141 if (retry >= 3) 142 udelay(10); 143 } 144 return cc; 145 } 146 147 static inline int pcpu_stopped(struct pcpu *pcpu) 148 { 149 u32 uninitialized_var(status); 150 151 if (__pcpu_sigp(pcpu->address, SIGP_SENSE, 152 0, &status) != SIGP_CC_STATUS_STORED) 153 return 0; 154 return !!(status & (SIGP_STATUS_CHECK_STOP|SIGP_STATUS_STOPPED)); 155 } 156 157 static inline int pcpu_running(struct pcpu *pcpu) 158 { 159 if (__pcpu_sigp(pcpu->address, SIGP_SENSE_RUNNING, 160 0, NULL) != SIGP_CC_STATUS_STORED) 161 return 1; 162 /* Status stored condition code is equivalent to cpu not running. */ 163 return 0; 164 } 165 166 /* 167 * Find struct pcpu by cpu address. 168 */ 169 static struct pcpu *pcpu_find_address(const struct cpumask *mask, u16 address) 170 { 171 int cpu; 172 173 for_each_cpu(cpu, mask) 174 if (pcpu_devices[cpu].address == address) 175 return pcpu_devices + cpu; 176 return NULL; 177 } 178 179 static void pcpu_ec_call(struct pcpu *pcpu, int ec_bit) 180 { 181 int order; 182 183 if (test_and_set_bit(ec_bit, &pcpu->ec_mask)) 184 return; 185 order = pcpu_running(pcpu) ? SIGP_EXTERNAL_CALL : SIGP_EMERGENCY_SIGNAL; 186 pcpu->ec_clk = get_tod_clock_fast(); 187 pcpu_sigp_retry(pcpu, order, 0); 188 } 189 190 #define ASYNC_FRAME_OFFSET (ASYNC_SIZE - STACK_FRAME_OVERHEAD - __PT_SIZE) 191 #define PANIC_FRAME_OFFSET (PAGE_SIZE - STACK_FRAME_OVERHEAD - __PT_SIZE) 192 193 static int pcpu_alloc_lowcore(struct pcpu *pcpu, int cpu) 194 { 195 unsigned long async_stack, panic_stack; 196 struct lowcore *lc; 197 198 if (pcpu != &pcpu_devices[0]) { 199 pcpu->lowcore = (struct lowcore *) 200 __get_free_pages(GFP_KERNEL | GFP_DMA, LC_ORDER); 201 async_stack = __get_free_pages(GFP_KERNEL, ASYNC_ORDER); 202 panic_stack = __get_free_page(GFP_KERNEL); 203 if (!pcpu->lowcore || !panic_stack || !async_stack) 204 goto out; 205 } else { 206 async_stack = pcpu->lowcore->async_stack - ASYNC_FRAME_OFFSET; 207 panic_stack = pcpu->lowcore->panic_stack - PANIC_FRAME_OFFSET; 208 } 209 lc = pcpu->lowcore; 210 memcpy(lc, &S390_lowcore, 512); 211 memset((char *) lc + 512, 0, sizeof(*lc) - 512); 212 lc->async_stack = async_stack + ASYNC_FRAME_OFFSET; 213 lc->panic_stack = panic_stack + PANIC_FRAME_OFFSET; 214 lc->cpu_nr = cpu; 215 lc->spinlock_lockval = arch_spin_lockval(cpu); 216 lc->spinlock_index = 0; 217 lc->br_r1_trampoline = 0x07f1; /* br %r1 */ 218 if (nmi_alloc_per_cpu(lc)) 219 goto out; 220 if (vdso_alloc_per_cpu(lc)) 221 goto out_mcesa; 222 lowcore_ptr[cpu] = lc; 223 pcpu_sigp_retry(pcpu, SIGP_SET_PREFIX, (u32)(unsigned long) lc); 224 return 0; 225 226 out_mcesa: 227 nmi_free_per_cpu(lc); 228 out: 229 if (pcpu != &pcpu_devices[0]) { 230 free_page(panic_stack); 231 free_pages(async_stack, ASYNC_ORDER); 232 free_pages((unsigned long) pcpu->lowcore, LC_ORDER); 233 } 234 return -ENOMEM; 235 } 236 237 #ifdef CONFIG_HOTPLUG_CPU 238 239 static void pcpu_free_lowcore(struct pcpu *pcpu) 240 { 241 pcpu_sigp_retry(pcpu, SIGP_SET_PREFIX, 0); 242 lowcore_ptr[pcpu - pcpu_devices] = NULL; 243 vdso_free_per_cpu(pcpu->lowcore); 244 nmi_free_per_cpu(pcpu->lowcore); 245 if (pcpu == &pcpu_devices[0]) 246 return; 247 free_page(pcpu->lowcore->panic_stack-PANIC_FRAME_OFFSET); 248 free_pages(pcpu->lowcore->async_stack-ASYNC_FRAME_OFFSET, ASYNC_ORDER); 249 free_pages((unsigned long) pcpu->lowcore, LC_ORDER); 250 } 251 252 #endif /* CONFIG_HOTPLUG_CPU */ 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->machine_flags = S390_lowcore.machine_flags; 266 lc->user_timer = lc->system_timer = lc->steal_timer = 0; 267 __ctl_store(lc->cregs_save_area, 0, 15); 268 save_access_regs((unsigned int *) lc->access_regs_save_area); 269 memcpy(lc->stfle_fac_list, S390_lowcore.stfle_fac_list, 270 sizeof(lc->stfle_fac_list)); 271 memcpy(lc->alt_stfle_fac_list, S390_lowcore.alt_stfle_fac_list, 272 sizeof(lc->alt_stfle_fac_list)); 273 arch_spin_lock_setup(cpu); 274 } 275 276 static void pcpu_attach_task(struct pcpu *pcpu, struct task_struct *tsk) 277 { 278 struct lowcore *lc = pcpu->lowcore; 279 280 lc->kernel_stack = (unsigned long) task_stack_page(tsk) 281 + THREAD_SIZE - STACK_FRAME_OVERHEAD - sizeof(struct pt_regs); 282 lc->current_task = (unsigned long) tsk; 283 lc->lpp = LPP_MAGIC; 284 lc->current_pid = tsk->pid; 285 lc->user_timer = tsk->thread.user_timer; 286 lc->guest_timer = tsk->thread.guest_timer; 287 lc->system_timer = tsk->thread.system_timer; 288 lc->hardirq_timer = tsk->thread.hardirq_timer; 289 lc->softirq_timer = tsk->thread.softirq_timer; 290 lc->steal_timer = 0; 291 } 292 293 static void pcpu_start_fn(struct pcpu *pcpu, void (*func)(void *), void *data) 294 { 295 struct lowcore *lc = pcpu->lowcore; 296 297 lc->restart_stack = lc->kernel_stack; 298 lc->restart_fn = (unsigned long) func; 299 lc->restart_data = (unsigned long) data; 300 lc->restart_source = -1UL; 301 pcpu_sigp_retry(pcpu, SIGP_RESTART, 0); 302 } 303 304 /* 305 * Call function via PSW restart on pcpu and stop the current cpu. 306 */ 307 static void pcpu_delegate(struct pcpu *pcpu, void (*func)(void *), 308 void *data, unsigned long stack) 309 { 310 struct lowcore *lc = lowcore_ptr[pcpu - pcpu_devices]; 311 unsigned long source_cpu = stap(); 312 313 __load_psw_mask(PSW_KERNEL_BITS); 314 if (pcpu->address == source_cpu) 315 func(data); /* should not return */ 316 /* Stop target cpu (if func returns this stops the current cpu). */ 317 pcpu_sigp_retry(pcpu, SIGP_STOP, 0); 318 /* Restart func on the target cpu and stop the current cpu. */ 319 mem_assign_absolute(lc->restart_stack, stack); 320 mem_assign_absolute(lc->restart_fn, (unsigned long) func); 321 mem_assign_absolute(lc->restart_data, (unsigned long) data); 322 mem_assign_absolute(lc->restart_source, source_cpu); 323 __bpon(); 324 asm volatile( 325 "0: sigp 0,%0,%2 # sigp restart to target cpu\n" 326 " brc 2,0b # busy, try again\n" 327 "1: sigp 0,%1,%3 # sigp stop to current cpu\n" 328 " brc 2,1b # busy, try again\n" 329 : : "d" (pcpu->address), "d" (source_cpu), 330 "K" (SIGP_RESTART), "K" (SIGP_STOP) 331 : "0", "1", "cc"); 332 for (;;) ; 333 } 334 335 /* 336 * Enable additional logical cpus for multi-threading. 337 */ 338 static int pcpu_set_smt(unsigned int mtid) 339 { 340 int cc; 341 342 if (smp_cpu_mtid == mtid) 343 return 0; 344 cc = __pcpu_sigp(0, SIGP_SET_MULTI_THREADING, mtid, NULL); 345 if (cc == 0) { 346 smp_cpu_mtid = mtid; 347 smp_cpu_mt_shift = 0; 348 while (smp_cpu_mtid >= (1U << smp_cpu_mt_shift)) 349 smp_cpu_mt_shift++; 350 pcpu_devices[0].address = stap(); 351 } 352 return cc; 353 } 354 355 /* 356 * Call function on an online CPU. 357 */ 358 void smp_call_online_cpu(void (*func)(void *), void *data) 359 { 360 struct pcpu *pcpu; 361 362 /* Use the current cpu if it is online. */ 363 pcpu = pcpu_find_address(cpu_online_mask, stap()); 364 if (!pcpu) 365 /* Use the first online cpu. */ 366 pcpu = pcpu_devices + cpumask_first(cpu_online_mask); 367 pcpu_delegate(pcpu, func, data, (unsigned long) restart_stack); 368 } 369 370 /* 371 * Call function on the ipl CPU. 372 */ 373 void smp_call_ipl_cpu(void (*func)(void *), void *data) 374 { 375 pcpu_delegate(&pcpu_devices[0], func, data, 376 pcpu_devices->lowcore->panic_stack - 377 PANIC_FRAME_OFFSET + PAGE_SIZE); 378 } 379 380 int smp_find_processor_id(u16 address) 381 { 382 int cpu; 383 384 for_each_present_cpu(cpu) 385 if (pcpu_devices[cpu].address == address) 386 return cpu; 387 return -1; 388 } 389 390 bool arch_vcpu_is_preempted(int cpu) 391 { 392 if (test_cpu_flag_of(CIF_ENABLED_WAIT, cpu)) 393 return false; 394 if (pcpu_running(pcpu_devices + cpu)) 395 return false; 396 return true; 397 } 398 EXPORT_SYMBOL(arch_vcpu_is_preempted); 399 400 void smp_yield_cpu(int cpu) 401 { 402 if (MACHINE_HAS_DIAG9C) { 403 diag_stat_inc_norecursion(DIAG_STAT_X09C); 404 asm volatile("diag %0,0,0x9c" 405 : : "d" (pcpu_devices[cpu].address)); 406 } else if (MACHINE_HAS_DIAG44) { 407 diag_stat_inc_norecursion(DIAG_STAT_X044); 408 asm volatile("diag 0,0,0x44"); 409 } 410 } 411 412 /* 413 * Send cpus emergency shutdown signal. This gives the cpus the 414 * opportunity to complete outstanding interrupts. 415 */ 416 void notrace smp_emergency_stop(void) 417 { 418 cpumask_t cpumask; 419 u64 end; 420 int cpu; 421 422 cpumask_copy(&cpumask, cpu_online_mask); 423 cpumask_clear_cpu(smp_processor_id(), &cpumask); 424 425 end = get_tod_clock() + (1000000UL << 12); 426 for_each_cpu(cpu, &cpumask) { 427 struct pcpu *pcpu = pcpu_devices + cpu; 428 set_bit(ec_stop_cpu, &pcpu->ec_mask); 429 while (__pcpu_sigp(pcpu->address, SIGP_EMERGENCY_SIGNAL, 430 0, NULL) == SIGP_CC_BUSY && 431 get_tod_clock() < end) 432 cpu_relax(); 433 } 434 while (get_tod_clock() < end) { 435 for_each_cpu(cpu, &cpumask) 436 if (pcpu_stopped(pcpu_devices + cpu)) 437 cpumask_clear_cpu(cpu, &cpumask); 438 if (cpumask_empty(&cpumask)) 439 break; 440 cpu_relax(); 441 } 442 } 443 NOKPROBE_SYMBOL(smp_emergency_stop); 444 445 /* 446 * Stop all cpus but the current one. 447 */ 448 void smp_send_stop(void) 449 { 450 int cpu; 451 452 /* Disable all interrupts/machine checks */ 453 __load_psw_mask(PSW_KERNEL_BITS | PSW_MASK_DAT); 454 trace_hardirqs_off(); 455 456 debug_set_critical(); 457 458 if (oops_in_progress) 459 smp_emergency_stop(); 460 461 /* stop all processors */ 462 for_each_online_cpu(cpu) { 463 if (cpu == smp_processor_id()) 464 continue; 465 pcpu_sigp_retry(pcpu_devices + cpu, SIGP_STOP, 0); 466 while (!pcpu_stopped(pcpu_devices + cpu)) 467 cpu_relax(); 468 } 469 } 470 471 /* 472 * This is the main routine where commands issued by other 473 * cpus are handled. 474 */ 475 static void smp_handle_ext_call(void) 476 { 477 unsigned long bits; 478 479 /* handle bit signal external calls */ 480 bits = xchg(&pcpu_devices[smp_processor_id()].ec_mask, 0); 481 if (test_bit(ec_stop_cpu, &bits)) 482 smp_stop_cpu(); 483 if (test_bit(ec_schedule, &bits)) 484 scheduler_ipi(); 485 if (test_bit(ec_call_function_single, &bits)) 486 generic_smp_call_function_single_interrupt(); 487 } 488 489 static void do_ext_call_interrupt(struct ext_code ext_code, 490 unsigned int param32, unsigned long param64) 491 { 492 inc_irq_stat(ext_code.code == 0x1202 ? IRQEXT_EXC : IRQEXT_EMS); 493 smp_handle_ext_call(); 494 } 495 496 void arch_send_call_function_ipi_mask(const struct cpumask *mask) 497 { 498 int cpu; 499 500 for_each_cpu(cpu, mask) 501 pcpu_ec_call(pcpu_devices + cpu, ec_call_function_single); 502 } 503 504 void arch_send_call_function_single_ipi(int cpu) 505 { 506 pcpu_ec_call(pcpu_devices + cpu, ec_call_function_single); 507 } 508 509 /* 510 * this function sends a 'reschedule' IPI to another CPU. 511 * it goes straight through and wastes no time serializing 512 * anything. Worst case is that we lose a reschedule ... 513 */ 514 void smp_send_reschedule(int cpu) 515 { 516 pcpu_ec_call(pcpu_devices + cpu, ec_schedule); 517 } 518 519 /* 520 * parameter area for the set/clear control bit callbacks 521 */ 522 struct ec_creg_mask_parms { 523 unsigned long orval; 524 unsigned long andval; 525 int cr; 526 }; 527 528 /* 529 * callback for setting/clearing control bits 530 */ 531 static void smp_ctl_bit_callback(void *info) 532 { 533 struct ec_creg_mask_parms *pp = info; 534 unsigned long cregs[16]; 535 536 __ctl_store(cregs, 0, 15); 537 cregs[pp->cr] = (cregs[pp->cr] & pp->andval) | pp->orval; 538 __ctl_load(cregs, 0, 15); 539 } 540 541 /* 542 * Set a bit in a control register of all cpus 543 */ 544 void smp_ctl_set_bit(int cr, int bit) 545 { 546 struct ec_creg_mask_parms parms = { 1UL << bit, -1UL, cr }; 547 548 on_each_cpu(smp_ctl_bit_callback, &parms, 1); 549 } 550 EXPORT_SYMBOL(smp_ctl_set_bit); 551 552 /* 553 * Clear a bit in a control register of all cpus 554 */ 555 void smp_ctl_clear_bit(int cr, int bit) 556 { 557 struct ec_creg_mask_parms parms = { 0, ~(1UL << bit), cr }; 558 559 on_each_cpu(smp_ctl_bit_callback, &parms, 1); 560 } 561 EXPORT_SYMBOL(smp_ctl_clear_bit); 562 563 #ifdef CONFIG_CRASH_DUMP 564 565 int smp_store_status(int cpu) 566 { 567 struct pcpu *pcpu = pcpu_devices + cpu; 568 unsigned long pa; 569 570 pa = __pa(&pcpu->lowcore->floating_pt_save_area); 571 if (__pcpu_sigp_relax(pcpu->address, SIGP_STORE_STATUS_AT_ADDRESS, 572 pa) != SIGP_CC_ORDER_CODE_ACCEPTED) 573 return -EIO; 574 if (!MACHINE_HAS_VX && !MACHINE_HAS_GS) 575 return 0; 576 pa = __pa(pcpu->lowcore->mcesad & MCESA_ORIGIN_MASK); 577 if (MACHINE_HAS_GS) 578 pa |= pcpu->lowcore->mcesad & MCESA_LC_MASK; 579 if (__pcpu_sigp_relax(pcpu->address, SIGP_STORE_ADDITIONAL_STATUS, 580 pa) != SIGP_CC_ORDER_CODE_ACCEPTED) 581 return -EIO; 582 return 0; 583 } 584 585 /* 586 * Collect CPU state of the previous, crashed system. 587 * There are four cases: 588 * 1) standard zfcp dump 589 * condition: OLDMEM_BASE == NULL && ipl_info.type == IPL_TYPE_FCP_DUMP 590 * The state for all CPUs except the boot CPU needs to be collected 591 * with sigp stop-and-store-status. The boot CPU state is located in 592 * the absolute lowcore of the memory stored in the HSA. The zcore code 593 * will copy the boot CPU state from the HSA. 594 * 2) stand-alone kdump for SCSI (zfcp dump with swapped memory) 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 firmware or the boot-loader 598 * stored the registers of the boot CPU in the absolute lowcore in the 599 * memory of the old system. 600 * 3) kdump and the old kernel did not store the CPU state, 601 * or stand-alone kdump for DASD 602 * condition: OLDMEM_BASE != NULL && !is_kdump_kernel() 603 * The state for all CPUs except the boot CPU needs to be collected 604 * with sigp stop-and-store-status. The kexec code or the boot-loader 605 * stored the registers of the boot CPU in the memory of the old system. 606 * 4) kdump and the old kernel stored the CPU state 607 * condition: OLDMEM_BASE != NULL && is_kdump_kernel() 608 * This case does not exist for s390 anymore, setup_arch explicitly 609 * deactivates the elfcorehdr= kernel parameter 610 */ 611 static __init void smp_save_cpu_vxrs(struct save_area *sa, u16 addr, 612 bool is_boot_cpu, unsigned long page) 613 { 614 __vector128 *vxrs = (__vector128 *) page; 615 616 if (is_boot_cpu) 617 vxrs = boot_cpu_vector_save_area; 618 else 619 __pcpu_sigp_relax(addr, SIGP_STORE_ADDITIONAL_STATUS, page); 620 save_area_add_vxrs(sa, vxrs); 621 } 622 623 static __init void smp_save_cpu_regs(struct save_area *sa, u16 addr, 624 bool is_boot_cpu, unsigned long page) 625 { 626 void *regs = (void *) page; 627 628 if (is_boot_cpu) 629 copy_oldmem_kernel(regs, (void *) __LC_FPREGS_SAVE_AREA, 512); 630 else 631 __pcpu_sigp_relax(addr, SIGP_STORE_STATUS_AT_ADDRESS, page); 632 save_area_add_regs(sa, regs); 633 } 634 635 void __init smp_save_dump_cpus(void) 636 { 637 int addr, boot_cpu_addr, max_cpu_addr; 638 struct save_area *sa; 639 unsigned long page; 640 bool is_boot_cpu; 641 642 if (!(OLDMEM_BASE || ipl_info.type == IPL_TYPE_FCP_DUMP)) 643 /* No previous system present, normal boot. */ 644 return; 645 /* Allocate a page as dumping area for the store status sigps */ 646 page = memblock_alloc_base(PAGE_SIZE, PAGE_SIZE, 1UL << 31); 647 /* Set multi-threading state to the previous system. */ 648 pcpu_set_smt(sclp.mtid_prev); 649 boot_cpu_addr = stap(); 650 max_cpu_addr = SCLP_MAX_CORES << sclp.mtid_prev; 651 for (addr = 0; addr <= max_cpu_addr; addr++) { 652 if (__pcpu_sigp_relax(addr, SIGP_SENSE, 0) == 653 SIGP_CC_NOT_OPERATIONAL) 654 continue; 655 is_boot_cpu = (addr == boot_cpu_addr); 656 /* Allocate save area */ 657 sa = save_area_alloc(is_boot_cpu); 658 if (!sa) 659 panic("could not allocate memory for save area\n"); 660 if (MACHINE_HAS_VX) 661 /* Get the vector registers */ 662 smp_save_cpu_vxrs(sa, addr, is_boot_cpu, page); 663 /* 664 * For a zfcp dump OLDMEM_BASE == NULL and the registers 665 * of the boot CPU are stored in the HSA. To retrieve 666 * these registers an SCLP request is required which is 667 * done by drivers/s390/char/zcore.c:init_cpu_info() 668 */ 669 if (!is_boot_cpu || OLDMEM_BASE) 670 /* Get the CPU registers */ 671 smp_save_cpu_regs(sa, addr, is_boot_cpu, page); 672 } 673 memblock_free(page, PAGE_SIZE); 674 diag308_reset(); 675 pcpu_set_smt(0); 676 } 677 #endif /* CONFIG_CRASH_DUMP */ 678 679 void smp_cpu_set_polarization(int cpu, int val) 680 { 681 pcpu_devices[cpu].polarization = val; 682 } 683 684 int smp_cpu_get_polarization(int cpu) 685 { 686 return pcpu_devices[cpu].polarization; 687 } 688 689 static void __ref smp_get_core_info(struct sclp_core_info *info, int early) 690 { 691 static int use_sigp_detection; 692 int address; 693 694 if (use_sigp_detection || sclp_get_core_info(info, early)) { 695 use_sigp_detection = 1; 696 for (address = 0; 697 address < (SCLP_MAX_CORES << smp_cpu_mt_shift); 698 address += (1U << smp_cpu_mt_shift)) { 699 if (__pcpu_sigp_relax(address, SIGP_SENSE, 0) == 700 SIGP_CC_NOT_OPERATIONAL) 701 continue; 702 info->core[info->configured].core_id = 703 address >> smp_cpu_mt_shift; 704 info->configured++; 705 } 706 info->combined = info->configured; 707 } 708 } 709 710 static int smp_add_present_cpu(int cpu); 711 712 static int __smp_rescan_cpus(struct sclp_core_info *info, int sysfs_add) 713 { 714 struct pcpu *pcpu; 715 cpumask_t avail; 716 int cpu, nr, i, j; 717 u16 address; 718 719 nr = 0; 720 cpumask_xor(&avail, cpu_possible_mask, cpu_present_mask); 721 cpu = cpumask_first(&avail); 722 for (i = 0; (i < info->combined) && (cpu < nr_cpu_ids); i++) { 723 if (sclp.has_core_type && info->core[i].type != boot_core_type) 724 continue; 725 address = info->core[i].core_id << smp_cpu_mt_shift; 726 for (j = 0; j <= smp_cpu_mtid; j++) { 727 if (pcpu_find_address(cpu_present_mask, address + j)) 728 continue; 729 pcpu = pcpu_devices + cpu; 730 pcpu->address = address + j; 731 pcpu->state = 732 (cpu >= info->configured*(smp_cpu_mtid + 1)) ? 733 CPU_STATE_STANDBY : CPU_STATE_CONFIGURED; 734 smp_cpu_set_polarization(cpu, POLARIZATION_UNKNOWN); 735 set_cpu_present(cpu, true); 736 if (sysfs_add && smp_add_present_cpu(cpu) != 0) 737 set_cpu_present(cpu, false); 738 else 739 nr++; 740 cpu = cpumask_next(cpu, &avail); 741 if (cpu >= nr_cpu_ids) 742 break; 743 } 744 } 745 return nr; 746 } 747 748 void __init smp_detect_cpus(void) 749 { 750 unsigned int cpu, mtid, c_cpus, s_cpus; 751 struct sclp_core_info *info; 752 u16 address; 753 754 /* Get CPU information */ 755 info = memblock_virt_alloc(sizeof(*info), 8); 756 smp_get_core_info(info, 1); 757 /* Find boot CPU type */ 758 if (sclp.has_core_type) { 759 address = stap(); 760 for (cpu = 0; cpu < info->combined; cpu++) 761 if (info->core[cpu].core_id == address) { 762 /* The boot cpu dictates the cpu type. */ 763 boot_core_type = info->core[cpu].type; 764 break; 765 } 766 if (cpu >= info->combined) 767 panic("Could not find boot CPU type"); 768 } 769 770 /* Set multi-threading state for the current system */ 771 mtid = boot_core_type ? sclp.mtid : sclp.mtid_cp; 772 mtid = (mtid < smp_max_threads) ? mtid : smp_max_threads - 1; 773 pcpu_set_smt(mtid); 774 775 /* Print number of CPUs */ 776 c_cpus = s_cpus = 0; 777 for (cpu = 0; cpu < info->combined; cpu++) { 778 if (sclp.has_core_type && 779 info->core[cpu].type != boot_core_type) 780 continue; 781 if (cpu < info->configured) 782 c_cpus += smp_cpu_mtid + 1; 783 else 784 s_cpus += smp_cpu_mtid + 1; 785 } 786 pr_info("%d configured CPUs, %d standby CPUs\n", c_cpus, s_cpus); 787 788 /* Add CPUs present at boot */ 789 get_online_cpus(); 790 __smp_rescan_cpus(info, 0); 791 put_online_cpus(); 792 memblock_free_early((unsigned long)info, sizeof(*info)); 793 } 794 795 /* 796 * Activate a secondary processor. 797 */ 798 static void smp_start_secondary(void *cpuvoid) 799 { 800 int cpu = smp_processor_id(); 801 802 S390_lowcore.last_update_clock = get_tod_clock(); 803 S390_lowcore.restart_stack = (unsigned long) restart_stack; 804 S390_lowcore.restart_fn = (unsigned long) do_restart; 805 S390_lowcore.restart_data = 0; 806 S390_lowcore.restart_source = -1UL; 807 restore_access_regs(S390_lowcore.access_regs_save_area); 808 __ctl_load(S390_lowcore.cregs_save_area, 0, 15); 809 __load_psw_mask(PSW_KERNEL_BITS | PSW_MASK_DAT); 810 cpu_init(); 811 preempt_disable(); 812 init_cpu_timer(); 813 vtime_init(); 814 pfault_init(); 815 notify_cpu_starting(cpu); 816 if (topology_cpu_dedicated(cpu)) 817 set_cpu_flag(CIF_DEDICATED_CPU); 818 else 819 clear_cpu_flag(CIF_DEDICATED_CPU); 820 set_cpu_online(cpu, 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 __bpon(); 909 pcpu_sigp_retry(pcpu_devices + smp_processor_id(), SIGP_STOP, 0); 910 for (;;) ; 911 } 912 913 #endif /* CONFIG_HOTPLUG_CPU */ 914 915 void __init smp_fill_possible_mask(void) 916 { 917 unsigned int possible, sclp_max, cpu; 918 919 sclp_max = max(sclp.mtid, sclp.mtid_cp) + 1; 920 sclp_max = min(smp_max_threads, sclp_max); 921 sclp_max = (sclp.max_cores * sclp_max) ?: nr_cpu_ids; 922 possible = setup_possible_cpus ?: nr_cpu_ids; 923 possible = min(possible, sclp_max); 924 for (cpu = 0; cpu < possible && cpu < nr_cpu_ids; cpu++) 925 set_cpu_possible(cpu, true); 926 } 927 928 void __init smp_prepare_cpus(unsigned int max_cpus) 929 { 930 /* request the 0x1201 emergency signal external interrupt */ 931 if (register_external_irq(EXT_IRQ_EMERGENCY_SIG, do_ext_call_interrupt)) 932 panic("Couldn't request external interrupt 0x1201"); 933 /* request the 0x1202 external call external interrupt */ 934 if (register_external_irq(EXT_IRQ_EXTERNAL_CALL, do_ext_call_interrupt)) 935 panic("Couldn't request external interrupt 0x1202"); 936 } 937 938 void __init smp_prepare_boot_cpu(void) 939 { 940 struct pcpu *pcpu = pcpu_devices; 941 942 WARN_ON(!cpu_present(0) || !cpu_online(0)); 943 pcpu->state = CPU_STATE_CONFIGURED; 944 pcpu->lowcore = (struct lowcore *)(unsigned long) store_prefix(); 945 S390_lowcore.percpu_offset = __per_cpu_offset[0]; 946 smp_cpu_set_polarization(0, POLARIZATION_UNKNOWN); 947 } 948 949 void __init smp_cpus_done(unsigned int max_cpus) 950 { 951 } 952 953 void __init smp_setup_processor_id(void) 954 { 955 pcpu_devices[0].address = stap(); 956 S390_lowcore.cpu_nr = 0; 957 S390_lowcore.spinlock_lockval = arch_spin_lockval(0); 958 S390_lowcore.spinlock_index = 0; 959 } 960 961 /* 962 * the frequency of the profiling timer can be changed 963 * by writing a multiplier value into /proc/profile. 964 * 965 * usually you want to run this on all CPUs ;) 966 */ 967 int setup_profiling_timer(unsigned int multiplier) 968 { 969 return 0; 970 } 971 972 #ifdef CONFIG_HOTPLUG_CPU 973 static ssize_t cpu_configure_show(struct device *dev, 974 struct device_attribute *attr, char *buf) 975 { 976 ssize_t count; 977 978 mutex_lock(&smp_cpu_state_mutex); 979 count = sprintf(buf, "%d\n", pcpu_devices[dev->id].state); 980 mutex_unlock(&smp_cpu_state_mutex); 981 return count; 982 } 983 984 static ssize_t cpu_configure_store(struct device *dev, 985 struct device_attribute *attr, 986 const char *buf, size_t count) 987 { 988 struct pcpu *pcpu; 989 int cpu, val, rc, i; 990 char delim; 991 992 if (sscanf(buf, "%d %c", &val, &delim) != 1) 993 return -EINVAL; 994 if (val != 0 && val != 1) 995 return -EINVAL; 996 get_online_cpus(); 997 mutex_lock(&smp_cpu_state_mutex); 998 rc = -EBUSY; 999 /* disallow configuration changes of online cpus and cpu 0 */ 1000 cpu = dev->id; 1001 cpu = smp_get_base_cpu(cpu); 1002 if (cpu == 0) 1003 goto out; 1004 for (i = 0; i <= smp_cpu_mtid; i++) 1005 if (cpu_online(cpu + i)) 1006 goto out; 1007 pcpu = pcpu_devices + cpu; 1008 rc = 0; 1009 switch (val) { 1010 case 0: 1011 if (pcpu->state != CPU_STATE_CONFIGURED) 1012 break; 1013 rc = sclp_core_deconfigure(pcpu->address >> smp_cpu_mt_shift); 1014 if (rc) 1015 break; 1016 for (i = 0; i <= smp_cpu_mtid; i++) { 1017 if (cpu + i >= nr_cpu_ids || !cpu_present(cpu + i)) 1018 continue; 1019 pcpu[i].state = CPU_STATE_STANDBY; 1020 smp_cpu_set_polarization(cpu + i, 1021 POLARIZATION_UNKNOWN); 1022 } 1023 topology_expect_change(); 1024 break; 1025 case 1: 1026 if (pcpu->state != CPU_STATE_STANDBY) 1027 break; 1028 rc = sclp_core_configure(pcpu->address >> smp_cpu_mt_shift); 1029 if (rc) 1030 break; 1031 for (i = 0; i <= smp_cpu_mtid; i++) { 1032 if (cpu + i >= nr_cpu_ids || !cpu_present(cpu + i)) 1033 continue; 1034 pcpu[i].state = CPU_STATE_CONFIGURED; 1035 smp_cpu_set_polarization(cpu + i, 1036 POLARIZATION_UNKNOWN); 1037 } 1038 topology_expect_change(); 1039 break; 1040 default: 1041 break; 1042 } 1043 out: 1044 mutex_unlock(&smp_cpu_state_mutex); 1045 put_online_cpus(); 1046 return rc ? rc : count; 1047 } 1048 static DEVICE_ATTR(configure, 0644, cpu_configure_show, cpu_configure_store); 1049 #endif /* CONFIG_HOTPLUG_CPU */ 1050 1051 static ssize_t show_cpu_address(struct device *dev, 1052 struct device_attribute *attr, char *buf) 1053 { 1054 return sprintf(buf, "%d\n", pcpu_devices[dev->id].address); 1055 } 1056 static DEVICE_ATTR(address, 0444, show_cpu_address, NULL); 1057 1058 static struct attribute *cpu_common_attrs[] = { 1059 #ifdef CONFIG_HOTPLUG_CPU 1060 &dev_attr_configure.attr, 1061 #endif 1062 &dev_attr_address.attr, 1063 NULL, 1064 }; 1065 1066 static struct attribute_group cpu_common_attr_group = { 1067 .attrs = cpu_common_attrs, 1068 }; 1069 1070 static struct attribute *cpu_online_attrs[] = { 1071 &dev_attr_idle_count.attr, 1072 &dev_attr_idle_time_us.attr, 1073 NULL, 1074 }; 1075 1076 static struct attribute_group cpu_online_attr_group = { 1077 .attrs = cpu_online_attrs, 1078 }; 1079 1080 static int smp_cpu_online(unsigned int cpu) 1081 { 1082 struct device *s = &per_cpu(cpu_device, cpu)->dev; 1083 1084 return sysfs_create_group(&s->kobj, &cpu_online_attr_group); 1085 } 1086 static int smp_cpu_pre_down(unsigned int cpu) 1087 { 1088 struct device *s = &per_cpu(cpu_device, cpu)->dev; 1089 1090 sysfs_remove_group(&s->kobj, &cpu_online_attr_group); 1091 return 0; 1092 } 1093 1094 static int smp_add_present_cpu(int cpu) 1095 { 1096 struct device *s; 1097 struct cpu *c; 1098 int rc; 1099 1100 c = kzalloc(sizeof(*c), GFP_KERNEL); 1101 if (!c) 1102 return -ENOMEM; 1103 per_cpu(cpu_device, cpu) = c; 1104 s = &c->dev; 1105 c->hotpluggable = 1; 1106 rc = register_cpu(c, cpu); 1107 if (rc) 1108 goto out; 1109 rc = sysfs_create_group(&s->kobj, &cpu_common_attr_group); 1110 if (rc) 1111 goto out_cpu; 1112 rc = topology_cpu_init(c); 1113 if (rc) 1114 goto out_topology; 1115 return 0; 1116 1117 out_topology: 1118 sysfs_remove_group(&s->kobj, &cpu_common_attr_group); 1119 out_cpu: 1120 #ifdef CONFIG_HOTPLUG_CPU 1121 unregister_cpu(c); 1122 #endif 1123 out: 1124 return rc; 1125 } 1126 1127 #ifdef CONFIG_HOTPLUG_CPU 1128 1129 int __ref smp_rescan_cpus(void) 1130 { 1131 struct sclp_core_info *info; 1132 int nr; 1133 1134 info = kzalloc(sizeof(*info), GFP_KERNEL); 1135 if (!info) 1136 return -ENOMEM; 1137 smp_get_core_info(info, 0); 1138 get_online_cpus(); 1139 mutex_lock(&smp_cpu_state_mutex); 1140 nr = __smp_rescan_cpus(info, 1); 1141 mutex_unlock(&smp_cpu_state_mutex); 1142 put_online_cpus(); 1143 kfree(info); 1144 if (nr) 1145 topology_schedule_update(); 1146 return 0; 1147 } 1148 1149 static ssize_t __ref rescan_store(struct device *dev, 1150 struct device_attribute *attr, 1151 const char *buf, 1152 size_t count) 1153 { 1154 int rc; 1155 1156 rc = smp_rescan_cpus(); 1157 return rc ? rc : count; 1158 } 1159 static DEVICE_ATTR_WO(rescan); 1160 #endif /* CONFIG_HOTPLUG_CPU */ 1161 1162 static int __init s390_smp_init(void) 1163 { 1164 int cpu, rc = 0; 1165 1166 #ifdef CONFIG_HOTPLUG_CPU 1167 rc = device_create_file(cpu_subsys.dev_root, &dev_attr_rescan); 1168 if (rc) 1169 return rc; 1170 #endif 1171 for_each_present_cpu(cpu) { 1172 rc = smp_add_present_cpu(cpu); 1173 if (rc) 1174 goto out; 1175 } 1176 1177 rc = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "s390/smp:online", 1178 smp_cpu_online, smp_cpu_pre_down); 1179 rc = rc <= 0 ? rc : 0; 1180 out: 1181 return rc; 1182 } 1183 subsys_initcall(s390_smp_init); 1184