Linux Audio

Check our new training course

Loading...
v6.8
   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 *
   9 *  based on other smp stuff by
  10 *    (c) 1995 Alan Cox, CymruNET Ltd  <alan@cymru.net>
  11 *    (c) 1998 Ingo Molnar
  12 *
  13 * The code outside of smp.c uses logical cpu numbers, only smp.c does
  14 * the translation of logical to physical cpu ids. All new code that
  15 * operates on physical cpu numbers needs to go into smp.c.
 
 
 
 
 
  16 */
  17
  18#define KMSG_COMPONENT "cpu"
  19#define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
  20
  21#include <linux/workqueue.h>
  22#include <linux/memblock.h>
  23#include <linux/export.h>
  24#include <linux/init.h>
  25#include <linux/mm.h>
  26#include <linux/err.h>
  27#include <linux/spinlock.h>
  28#include <linux/kernel_stat.h>
  29#include <linux/delay.h>
 
  30#include <linux/interrupt.h>
  31#include <linux/irqflags.h>
  32#include <linux/irq_work.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/ctlreg.h>
  41#include <asm/pfault.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/abs_lowcore.h>
  51#include <asm/sclp.h>
  52#include <asm/debug.h>
  53#include <asm/os_info.h>
  54#include <asm/sigp.h>
  55#include <asm/idle.h>
  56#include <asm/nmi.h>
  57#include <asm/stacktrace.h>
  58#include <asm/topology.h>
  59#include <asm/vdso.h>
  60#include <asm/maccess.h>
  61#include "entry.h"
  62
  63enum {
  64	ec_schedule = 0,
  65	ec_call_function_single,
  66	ec_stop_cpu,
  67	ec_mcck_pending,
  68	ec_irq_work,
  69};
  70
  71enum {
  72	CPU_STATE_STANDBY,
  73	CPU_STATE_CONFIGURED,
  74};
  75
  76static DEFINE_PER_CPU(struct cpu *, cpu_device);
 
  77
  78struct pcpu {
  79	unsigned long ec_mask;		/* bit mask for ec_xxx functions */
  80	unsigned long ec_clk;		/* sigp timestamp for ec_xxx */
  81	signed char state;		/* physical cpu state */
  82	signed char polarization;	/* physical polarization */
  83	u16 address;			/* physical cpu address */
  84};
  85
  86static u8 boot_core_type;
  87static struct pcpu pcpu_devices[NR_CPUS];
  88
  89unsigned int smp_cpu_mt_shift;
  90EXPORT_SYMBOL(smp_cpu_mt_shift);
  91
  92unsigned int smp_cpu_mtid;
  93EXPORT_SYMBOL(smp_cpu_mtid);
  94
  95#ifdef CONFIG_CRASH_DUMP
  96__vector128 __initdata boot_cpu_vector_save_area[__NUM_VXRS];
  97#endif
  98
  99static unsigned int smp_max_threads __initdata = -1U;
 100cpumask_t cpu_setup_mask;
 101
 102static int __init early_nosmt(char *s)
 103{
 104	smp_max_threads = 1;
 
 
 
 
 
 
 
 
 
 
 105	return 0;
 106}
 107early_param("nosmt", early_nosmt);
 108
 109static int __init early_smt(char *s)
 110{
 111	get_option(&s, &smp_max_threads);
 112	return 0;
 113}
 114early_param("smt", early_smt);
 115
 116/*
 117 * The smp_cpu_state_mutex must be held when changing the state or polarization
 118 * member of a pcpu data structure within the pcpu_devices array.
 119 */
 120DEFINE_MUTEX(smp_cpu_state_mutex);
 121
 122/*
 123 * Signal processor helper functions.
 124 */
 125static inline int __pcpu_sigp_relax(u16 addr, u8 order, unsigned long parm)
 126{
 127	int cc;
 128
 129	while (1) {
 130		cc = __pcpu_sigp(addr, order, parm, NULL);
 131		if (cc != SIGP_CC_BUSY)
 132			return cc;
 
 
 
 
 
 
 
 
 
 
 
 
 
 133		cpu_relax();
 134	}
 
 
 
 
 
 
 
 
 135}
 136
 137static int pcpu_sigp_retry(struct pcpu *pcpu, u8 order, u32 parm)
 138{
 139	int cc, retry;
 140
 141	for (retry = 0; ; retry++) {
 142		cc = __pcpu_sigp(pcpu->address, order, parm, NULL);
 143		if (cc != SIGP_CC_BUSY)
 144			break;
 145		if (retry >= 3)
 146			udelay(10);
 147	}
 148	return cc;
 149}
 150
 151static inline int pcpu_stopped(struct pcpu *pcpu)
 152{
 153	u32 status;
 154
 155	if (__pcpu_sigp(pcpu->address, SIGP_SENSE,
 156			0, &status) != SIGP_CC_STATUS_STORED)
 157		return 0;
 158	return !!(status & (SIGP_STATUS_CHECK_STOP|SIGP_STATUS_STOPPED));
 159}
 
 
 160
 161static inline int pcpu_running(struct pcpu *pcpu)
 162{
 163	if (__pcpu_sigp(pcpu->address, SIGP_SENSE_RUNNING,
 164			0, NULL) != SIGP_CC_STATUS_STORED)
 165		return 1;
 166	/* Status stored condition code is equivalent to cpu not running. */
 167	return 0;
 168}
 169
 170/*
 171 * Find struct pcpu by cpu address.
 
 172 */
 173static struct pcpu *pcpu_find_address(const struct cpumask *mask, u16 address)
 174{
 175	int cpu;
 176
 177	for_each_cpu(cpu, mask)
 178		if (pcpu_devices[cpu].address == address)
 179			return pcpu_devices + cpu;
 180	return NULL;
 181}
 182
 183static void pcpu_ec_call(struct pcpu *pcpu, int ec_bit)
 
 184{
 185	int order;
 186
 187	if (test_and_set_bit(ec_bit, &pcpu->ec_mask))
 188		return;
 189	order = pcpu_running(pcpu) ? SIGP_EXTERNAL_CALL : SIGP_EMERGENCY_SIGNAL;
 190	pcpu->ec_clk = get_tod_clock_fast();
 191	pcpu_sigp_retry(pcpu, order, 0);
 192}
 193
 194static int pcpu_alloc_lowcore(struct pcpu *pcpu, int cpu)
 195{
 196	unsigned long async_stack, nodat_stack, mcck_stack;
 197	struct lowcore *lc;
 198
 199	lc = (struct lowcore *) __get_free_pages(GFP_KERNEL | GFP_DMA, LC_ORDER);
 200	nodat_stack = __get_free_pages(GFP_KERNEL, THREAD_SIZE_ORDER);
 201	async_stack = stack_alloc();
 202	mcck_stack = stack_alloc();
 203	if (!lc || !nodat_stack || !async_stack || !mcck_stack)
 204		goto out;
 205	memcpy(lc, &S390_lowcore, 512);
 206	memset((char *) lc + 512, 0, sizeof(*lc) - 512);
 207	lc->async_stack = async_stack + STACK_INIT_OFFSET;
 208	lc->nodat_stack = nodat_stack + STACK_INIT_OFFSET;
 209	lc->mcck_stack = mcck_stack + STACK_INIT_OFFSET;
 210	lc->cpu_nr = cpu;
 211	lc->spinlock_lockval = arch_spin_lockval(cpu);
 212	lc->spinlock_index = 0;
 213	lc->return_lpswe = gen_lpswe(__LC_RETURN_PSW);
 214	lc->return_mcck_lpswe = gen_lpswe(__LC_RETURN_MCCK_PSW);
 215	lc->preempt_count = PREEMPT_DISABLED;
 216	if (nmi_alloc_mcesa(&lc->mcesad))
 217		goto out;
 218	if (abs_lowcore_map(cpu, lc, true))
 219		goto out_mcesa;
 220	lowcore_ptr[cpu] = lc;
 221	pcpu_sigp_retry(pcpu, SIGP_SET_PREFIX, __pa(lc));
 222	return 0;
 223
 224out_mcesa:
 225	nmi_free_mcesa(&lc->mcesad);
 226out:
 227	stack_free(mcck_stack);
 228	stack_free(async_stack);
 229	free_pages(nodat_stack, THREAD_SIZE_ORDER);
 230	free_pages((unsigned long) lc, LC_ORDER);
 231	return -ENOMEM;
 232}
 233
 234static void pcpu_free_lowcore(struct pcpu *pcpu)
 
 
 
 
 235{
 236	unsigned long async_stack, nodat_stack, mcck_stack;
 237	struct lowcore *lc;
 238	int cpu;
 239
 240	cpu = pcpu - pcpu_devices;
 241	lc = lowcore_ptr[cpu];
 242	nodat_stack = lc->nodat_stack - STACK_INIT_OFFSET;
 243	async_stack = lc->async_stack - STACK_INIT_OFFSET;
 244	mcck_stack = lc->mcck_stack - STACK_INIT_OFFSET;
 245	pcpu_sigp_retry(pcpu, SIGP_SET_PREFIX, 0);
 246	lowcore_ptr[cpu] = NULL;
 247	abs_lowcore_unmap(cpu);
 248	nmi_free_mcesa(&lc->mcesad);
 249	stack_free(async_stack);
 250	stack_free(mcck_stack);
 251	free_pages(nodat_stack, THREAD_SIZE_ORDER);
 252	free_pages((unsigned long) lc, LC_ORDER);
 253}
 254
 255static void pcpu_prepare_secondary(struct pcpu *pcpu, int cpu)
 256{
 257	struct lowcore *lc, *abs_lc;
 258
 259	lc = lowcore_ptr[cpu];
 260	cpumask_set_cpu(cpu, &init_mm.context.cpu_attach_mask);
 261	cpumask_set_cpu(cpu, mm_cpumask(&init_mm));
 262	lc->cpu_nr = cpu;
 263	lc->restart_flags = RESTART_FLAG_CTLREGS;
 264	lc->spinlock_lockval = arch_spin_lockval(cpu);
 265	lc->spinlock_index = 0;
 266	lc->percpu_offset = __per_cpu_offset[cpu];
 267	lc->kernel_asce = S390_lowcore.kernel_asce;
 268	lc->user_asce = s390_invalid_asce;
 269	lc->machine_flags = S390_lowcore.machine_flags;
 270	lc->user_timer = lc->system_timer =
 271		lc->steal_timer = lc->avg_steal_timer = 0;
 272	abs_lc = get_abs_lowcore();
 273	memcpy(lc->cregs_save_area, abs_lc->cregs_save_area, sizeof(lc->cregs_save_area));
 274	put_abs_lowcore(abs_lc);
 275	lc->cregs_save_area[1] = lc->kernel_asce;
 276	lc->cregs_save_area[7] = lc->user_asce;
 277	save_access_regs((unsigned int *) lc->access_regs_save_area);
 278	arch_spin_lock_setup(cpu);
 279}
 280
 281static void pcpu_attach_task(struct pcpu *pcpu, struct task_struct *tsk)
 282{
 283	struct lowcore *lc;
 284	int cpu;
 285
 286	cpu = pcpu - pcpu_devices;
 287	lc = lowcore_ptr[cpu];
 288	lc->kernel_stack = (unsigned long)task_stack_page(tsk) + STACK_INIT_OFFSET;
 289	lc->current_task = (unsigned long)tsk;
 290	lc->lpp = LPP_MAGIC;
 291	lc->current_pid = tsk->pid;
 292	lc->user_timer = tsk->thread.user_timer;
 293	lc->guest_timer = tsk->thread.guest_timer;
 294	lc->system_timer = tsk->thread.system_timer;
 295	lc->hardirq_timer = tsk->thread.hardirq_timer;
 296	lc->softirq_timer = tsk->thread.softirq_timer;
 297	lc->steal_timer = 0;
 298}
 299
 300static void pcpu_start_fn(struct pcpu *pcpu, void (*func)(void *), void *data)
 301{
 302	struct lowcore *lc;
 303	int cpu;
 304
 305	cpu = pcpu - pcpu_devices;
 306	lc = lowcore_ptr[cpu];
 307	lc->restart_stack = lc->kernel_stack;
 308	lc->restart_fn = (unsigned long) func;
 309	lc->restart_data = (unsigned long) data;
 310	lc->restart_source = -1U;
 311	pcpu_sigp_retry(pcpu, SIGP_RESTART, 0);
 312}
 313
 314typedef void (pcpu_delegate_fn)(void *);
 315
 316/*
 317 * Call function via PSW restart on pcpu and stop the current cpu.
 318 */
 319static void __pcpu_delegate(pcpu_delegate_fn *func, void *data)
 320{
 321	func(data);	/* should not return */
 322}
 323
 324static void pcpu_delegate(struct pcpu *pcpu,
 325			  pcpu_delegate_fn *func,
 326			  void *data, unsigned long stack)
 327{
 328	struct lowcore *lc, *abs_lc;
 329	unsigned int source_cpu;
 330
 331	lc = lowcore_ptr[pcpu - pcpu_devices];
 332	source_cpu = stap();
 333
 334	if (pcpu->address == source_cpu) {
 335		call_on_stack(2, stack, void, __pcpu_delegate,
 336			      pcpu_delegate_fn *, func, void *, data);
 337	}
 338	/* Stop target cpu (if func returns this stops the current cpu). */
 339	pcpu_sigp_retry(pcpu, SIGP_STOP, 0);
 340	pcpu_sigp_retry(pcpu, SIGP_CPU_RESET, 0);
 341	/* Restart func on the target cpu and stop the current cpu. */
 342	if (lc) {
 343		lc->restart_stack = stack;
 344		lc->restart_fn = (unsigned long)func;
 345		lc->restart_data = (unsigned long)data;
 346		lc->restart_source = source_cpu;
 347	} else {
 348		abs_lc = get_abs_lowcore();
 349		abs_lc->restart_stack = stack;
 350		abs_lc->restart_fn = (unsigned long)func;
 351		abs_lc->restart_data = (unsigned long)data;
 352		abs_lc->restart_source = source_cpu;
 353		put_abs_lowcore(abs_lc);
 354	}
 355	asm volatile(
 356		"0:	sigp	0,%0,%2	# sigp restart to target cpu\n"
 357		"	brc	2,0b	# busy, try again\n"
 358		"1:	sigp	0,%1,%3	# sigp stop to current cpu\n"
 359		"	brc	2,1b	# busy, try again\n"
 360		: : "d" (pcpu->address), "d" (source_cpu),
 361		    "K" (SIGP_RESTART), "K" (SIGP_STOP)
 362		: "0", "1", "cc");
 363	for (;;) ;
 364}
 
 
 365
 366/*
 367 * Enable additional logical cpus for multi-threading.
 
 
 368 */
 369static int pcpu_set_smt(unsigned int mtid)
 370{
 371	int cc;
 372
 373	if (smp_cpu_mtid == mtid)
 374		return 0;
 375	cc = __pcpu_sigp(0, SIGP_SET_MULTI_THREADING, mtid, NULL);
 376	if (cc == 0) {
 377		smp_cpu_mtid = mtid;
 378		smp_cpu_mt_shift = 0;
 379		while (smp_cpu_mtid >= (1U << smp_cpu_mt_shift))
 380			smp_cpu_mt_shift++;
 381		pcpu_devices[0].address = stap();
 382	}
 383	return cc;
 384}
 385
 386/*
 387 * Call function on an online CPU.
 388 */
 389void smp_call_online_cpu(void (*func)(void *), void *data)
 390{
 391	struct pcpu *pcpu;
 392
 393	/* Use the current cpu if it is online. */
 394	pcpu = pcpu_find_address(cpu_online_mask, stap());
 395	if (!pcpu)
 396		/* Use the first online cpu. */
 397		pcpu = pcpu_devices + cpumask_first(cpu_online_mask);
 398	pcpu_delegate(pcpu, func, data, (unsigned long) restart_stack);
 399}
 400
 401/*
 402 * Call function on the ipl CPU.
 403 */
 404void smp_call_ipl_cpu(void (*func)(void *), void *data)
 405{
 406	struct lowcore *lc = lowcore_ptr[0];
 407
 408	if (pcpu_devices[0].address == stap())
 409		lc = &S390_lowcore;
 410
 411	pcpu_delegate(&pcpu_devices[0], func, data,
 412		      lc->nodat_stack);
 413}
 414
 415int smp_find_processor_id(u16 address)
 416{
 417	int cpu;
 418
 419	for_each_present_cpu(cpu)
 420		if (pcpu_devices[cpu].address == address)
 421			return cpu;
 422	return -1;
 423}
 424
 425void schedule_mcck_handler(void)
 426{
 427	pcpu_ec_call(pcpu_devices + smp_processor_id(), ec_mcck_pending);
 428}
 429
 430bool notrace arch_vcpu_is_preempted(int cpu)
 431{
 432	if (test_cpu_flag_of(CIF_ENABLED_WAIT, cpu))
 433		return false;
 434	if (pcpu_running(pcpu_devices + cpu))
 435		return false;
 436	return true;
 437}
 438EXPORT_SYMBOL(arch_vcpu_is_preempted);
 439
 440void notrace smp_yield_cpu(int cpu)
 441{
 442	if (!MACHINE_HAS_DIAG9C)
 443		return;
 444	diag_stat_inc_norecursion(DIAG_STAT_X09C);
 445	asm volatile("diag %0,0,0x9c"
 446		     : : "d" (pcpu_devices[cpu].address));
 
 
 
 447}
 448EXPORT_SYMBOL_GPL(smp_yield_cpu);
 449
 450/*
 451 * Send cpus emergency shutdown signal. This gives the cpus the
 452 * opportunity to complete outstanding interrupts.
 453 */
 454void notrace smp_emergency_stop(void)
 455{
 456	static arch_spinlock_t lock = __ARCH_SPIN_LOCK_UNLOCKED;
 457	static cpumask_t cpumask;
 458	u64 end;
 459	int cpu;
 460
 461	arch_spin_lock(&lock);
 462	cpumask_copy(&cpumask, cpu_online_mask);
 463	cpumask_clear_cpu(smp_processor_id(), &cpumask);
 464
 465	end = get_tod_clock() + (1000000UL << 12);
 466	for_each_cpu(cpu, &cpumask) {
 467		struct pcpu *pcpu = pcpu_devices + cpu;
 468		set_bit(ec_stop_cpu, &pcpu->ec_mask);
 469		while (__pcpu_sigp(pcpu->address, SIGP_EMERGENCY_SIGNAL,
 470				   0, NULL) == SIGP_CC_BUSY &&
 471		       get_tod_clock() < end)
 472			cpu_relax();
 473	}
 474	while (get_tod_clock() < end) {
 475		for_each_cpu(cpu, &cpumask)
 476			if (pcpu_stopped(pcpu_devices + cpu))
 477				cpumask_clear_cpu(cpu, &cpumask);
 478		if (cpumask_empty(&cpumask))
 479			break;
 480		cpu_relax();
 481	}
 482	arch_spin_unlock(&lock);
 483}
 484NOKPROBE_SYMBOL(smp_emergency_stop);
 485
 486/*
 487 * Stop all cpus but the current one.
 488 */
 489void smp_send_stop(void)
 490{
 491	int cpu;
 492
 493	/* Disable all interrupts/machine checks */
 494	__load_psw_mask(PSW_KERNEL_BITS);
 495	trace_hardirqs_off();
 496
 497	debug_set_critical();
 
 498
 499	if (oops_in_progress)
 500		smp_emergency_stop();
 501
 502	/* stop all processors */
 503	for_each_online_cpu(cpu) {
 504		if (cpu == smp_processor_id())
 505			continue;
 506		pcpu_sigp_retry(pcpu_devices + cpu, SIGP_STOP, 0);
 507		while (!pcpu_stopped(pcpu_devices + cpu))
 508			cpu_relax();
 
 509	}
 
 
 
 
 
 
 510}
 511
 512/*
 513 * This is the main routine where commands issued by other
 514 * cpus are handled.
 515 */
 516static void smp_handle_ext_call(void)
 517{
 518	unsigned long bits;
 519
 520	/* handle bit signal external calls */
 521	bits = xchg(&pcpu_devices[smp_processor_id()].ec_mask, 0);
 522	if (test_bit(ec_stop_cpu, &bits))
 523		smp_stop_cpu();
 524	if (test_bit(ec_schedule, &bits))
 525		scheduler_ipi();
 526	if (test_bit(ec_call_function_single, &bits))
 527		generic_smp_call_function_single_interrupt();
 528	if (test_bit(ec_mcck_pending, &bits))
 529		s390_handle_mcck();
 530	if (test_bit(ec_irq_work, &bits))
 531		irq_work_run();
 532}
 533
 534static void do_ext_call_interrupt(struct ext_code ext_code,
 535				  unsigned int param32, unsigned long param64)
 536{
 537	inc_irq_stat(ext_code.code == 0x1202 ? IRQEXT_EXC : IRQEXT_EMS);
 538	smp_handle_ext_call();
 539}
 540
 541void arch_send_call_function_ipi_mask(const struct cpumask *mask)
 542{
 543	int cpu;
 544
 545	for_each_cpu(cpu, mask)
 546		pcpu_ec_call(pcpu_devices + cpu, ec_call_function_single);
 547}
 548
 549void arch_send_call_function_single_ipi(int cpu)
 550{
 551	pcpu_ec_call(pcpu_devices + cpu, ec_call_function_single);
 552}
 553
 554/*
 555 * this function sends a 'reschedule' IPI to another CPU.
 556 * it goes straight through and wastes no time serializing
 557 * anything. Worst case is that we lose a reschedule ...
 558 */
 559void arch_smp_send_reschedule(int cpu)
 560{
 561	pcpu_ec_call(pcpu_devices + cpu, ec_schedule);
 562}
 563
 564#ifdef CONFIG_IRQ_WORK
 565void arch_irq_work_raise(void)
 566{
 567	pcpu_ec_call(pcpu_devices + smp_processor_id(), ec_irq_work);
 568}
 569#endif
 570
 571#ifdef CONFIG_CRASH_DUMP
 572
 573int smp_store_status(int cpu)
 574{
 575	struct lowcore *lc;
 576	struct pcpu *pcpu;
 577	unsigned long pa;
 578
 579	pcpu = pcpu_devices + cpu;
 580	lc = lowcore_ptr[cpu];
 581	pa = __pa(&lc->floating_pt_save_area);
 582	if (__pcpu_sigp_relax(pcpu->address, SIGP_STORE_STATUS_AT_ADDRESS,
 583			      pa) != SIGP_CC_ORDER_CODE_ACCEPTED)
 584		return -EIO;
 585	if (!cpu_has_vx() && !MACHINE_HAS_GS)
 586		return 0;
 587	pa = lc->mcesad & MCESA_ORIGIN_MASK;
 588	if (MACHINE_HAS_GS)
 589		pa |= lc->mcesad & MCESA_LC_MASK;
 590	if (__pcpu_sigp_relax(pcpu->address, SIGP_STORE_ADDITIONAL_STATUS,
 591			      pa) != SIGP_CC_ORDER_CODE_ACCEPTED)
 592		return -EIO;
 
 
 
 
 
 
 
 593	return 0;
 594}
 595
 596/*
 597 * Collect CPU state of the previous, crashed system.
 598 * There are four cases:
 599 * 1) standard zfcp/nvme dump
 600 *    condition: OLDMEM_BASE == NULL && is_ipl_type_dump() == true
 601 *    The state for all CPUs except the boot CPU needs to be collected
 602 *    with sigp stop-and-store-status. The boot CPU state is located in
 603 *    the absolute lowcore of the memory stored in the HSA. The zcore code
 604 *    will copy the boot CPU state from the HSA.
 605 * 2) stand-alone kdump for SCSI/NVMe (zfcp/nvme dump with swapped memory)
 606 *    condition: OLDMEM_BASE != NULL && is_ipl_type_dump() == true
 607 *    The state for all CPUs except the boot CPU needs to be collected
 608 *    with sigp stop-and-store-status. The firmware or the boot-loader
 609 *    stored the registers of the boot CPU in the absolute lowcore in the
 610 *    memory of the old system.
 611 * 3) kdump and the old kernel did not store the CPU state,
 612 *    or stand-alone kdump for DASD
 613 *    condition: OLDMEM_BASE != NULL && !is_kdump_kernel()
 614 *    The state for all CPUs except the boot CPU needs to be collected
 615 *    with sigp stop-and-store-status. The kexec code or the boot-loader
 616 *    stored the registers of the boot CPU in the memory of the old system.
 617 * 4) kdump and the old kernel stored the CPU state
 618 *    condition: OLDMEM_BASE != NULL && is_kdump_kernel()
 619 *    This case does not exist for s390 anymore, setup_arch explicitly
 620 *    deactivates the elfcorehdr= kernel parameter
 621 */
 622static bool dump_available(void)
 623{
 624	return oldmem_data.start || is_ipl_type_dump();
 625}
 626
 627void __init smp_save_dump_ipl_cpu(void)
 628{
 629	struct save_area *sa;
 630	void *regs;
 631
 632	if (!dump_available())
 633		return;
 634	sa = save_area_alloc(true);
 635	regs = memblock_alloc(512, 8);
 636	if (!sa || !regs)
 637		panic("could not allocate memory for boot CPU save area\n");
 638	copy_oldmem_kernel(regs, __LC_FPREGS_SAVE_AREA, 512);
 639	save_area_add_regs(sa, regs);
 640	memblock_free(regs, 512);
 641	if (cpu_has_vx())
 642		save_area_add_vxrs(sa, boot_cpu_vector_save_area);
 643}
 644
 645void __init smp_save_dump_secondary_cpus(void)
 646{
 647	int addr, boot_cpu_addr, max_cpu_addr;
 648	struct save_area *sa;
 649	void *page;
 650
 651	if (!dump_available())
 652		return;
 653	/* Allocate a page as dumping area for the store status sigps */
 654	page = memblock_alloc_low(PAGE_SIZE, PAGE_SIZE);
 655	if (!page)
 656		panic("ERROR: Failed to allocate %lx bytes below %lx\n",
 657		      PAGE_SIZE, 1UL << 31);
 658
 659	/* Set multi-threading state to the previous system. */
 660	pcpu_set_smt(sclp.mtid_prev);
 661	boot_cpu_addr = stap();
 662	max_cpu_addr = SCLP_MAX_CORES << sclp.mtid_prev;
 663	for (addr = 0; addr <= max_cpu_addr; addr++) {
 664		if (addr == boot_cpu_addr)
 665			continue;
 666		if (__pcpu_sigp_relax(addr, SIGP_SENSE, 0) ==
 667		    SIGP_CC_NOT_OPERATIONAL)
 668			continue;
 669		sa = save_area_alloc(false);
 670		if (!sa)
 671			panic("could not allocate memory for save area\n");
 672		__pcpu_sigp_relax(addr, SIGP_STORE_STATUS_AT_ADDRESS, __pa(page));
 673		save_area_add_regs(sa, page);
 674		if (cpu_has_vx()) {
 675			__pcpu_sigp_relax(addr, SIGP_STORE_ADDITIONAL_STATUS, __pa(page));
 676			save_area_add_vxrs(sa, page);
 677		}
 
 678	}
 679	memblock_free(page, PAGE_SIZE);
 680	diag_amode31_ops.diag308_reset();
 681	pcpu_set_smt(0);
 682}
 683#endif /* CONFIG_CRASH_DUMP */
 684
 685void smp_cpu_set_polarization(int cpu, int val)
 686{
 687	pcpu_devices[cpu].polarization = val;
 688}
 689
 690int smp_cpu_get_polarization(int cpu)
 691{
 692	return pcpu_devices[cpu].polarization;
 693}
 694
 695int smp_cpu_get_cpu_address(int cpu)
 696{
 697	return pcpu_devices[cpu].address;
 
 
 698}
 699
 700static void __ref smp_get_core_info(struct sclp_core_info *info, int early)
 701{
 702	static int use_sigp_detection;
 703	int address;
 
 704
 705	if (use_sigp_detection || sclp_get_core_info(info, early)) {
 706		use_sigp_detection = 1;
 707		for (address = 0;
 708		     address < (SCLP_MAX_CORES << smp_cpu_mt_shift);
 709		     address += (1U << smp_cpu_mt_shift)) {
 710			if (__pcpu_sigp_relax(address, SIGP_SENSE, 0) ==
 711			    SIGP_CC_NOT_OPERATIONAL)
 
 
 
 
 
 
 712				continue;
 713			info->core[info->configured].core_id =
 714				address >> smp_cpu_mt_shift;
 715			info->configured++;
 716		}
 717		info->combined = info->configured;
 718	}
 719}
 720
 721static int smp_add_present_cpu(int cpu);
 
 
 
 
 
 
 
 722
 723static int smp_add_core(struct sclp_core_entry *core, cpumask_t *avail,
 724			bool configured, bool early)
 725{
 726	struct pcpu *pcpu;
 727	int cpu, nr, i;
 728	u16 address;
 729
 730	nr = 0;
 731	if (sclp.has_core_type && core->type != boot_core_type)
 732		return nr;
 733	cpu = cpumask_first(avail);
 734	address = core->core_id << smp_cpu_mt_shift;
 735	for (i = 0; (i <= smp_cpu_mtid) && (cpu < nr_cpu_ids); i++) {
 736		if (pcpu_find_address(cpu_present_mask, address + i))
 737			continue;
 738		pcpu = pcpu_devices + cpu;
 739		pcpu->address = address + i;
 740		if (configured)
 741			pcpu->state = CPU_STATE_CONFIGURED;
 742		else
 743			pcpu->state = CPU_STATE_STANDBY;
 744		smp_cpu_set_polarization(cpu, POLARIZATION_UNKNOWN);
 745		set_cpu_present(cpu, true);
 746		if (!early && smp_add_present_cpu(cpu) != 0)
 747			set_cpu_present(cpu, false);
 748		else
 749			nr++;
 750		cpumask_clear_cpu(cpu, avail);
 751		cpu = cpumask_next(cpu, avail);
 752	}
 753	return nr;
 
 
 
 
 
 754}
 755
 756static int __smp_rescan_cpus(struct sclp_core_info *info, bool early)
 
 
 
 757{
 758	struct sclp_core_entry *core;
 759	static cpumask_t avail;
 760	bool configured;
 761	u16 core_id;
 762	int nr, i;
 763
 764	cpus_read_lock();
 765	mutex_lock(&smp_cpu_state_mutex);
 766	nr = 0;
 767	cpumask_xor(&avail, cpu_possible_mask, cpu_present_mask);
 
 
 
 
 
 768	/*
 769	 * Add IPL core first (which got logical CPU number 0) to make sure
 770	 * that all SMT threads get subsequent logical CPU numbers.
 771	 */
 772	if (early) {
 773		core_id = pcpu_devices[0].address >> smp_cpu_mt_shift;
 774		for (i = 0; i < info->configured; i++) {
 775			core = &info->core[i];
 776			if (core->core_id == core_id) {
 777				nr += smp_add_core(core, &avail, true, early);
 778				break;
 779			}
 780		}
 781	}
 782	for (i = 0; i < info->combined; i++) {
 783		configured = i < info->configured;
 784		nr += smp_add_core(&info->core[i], &avail, configured, early);
 785	}
 786	mutex_unlock(&smp_cpu_state_mutex);
 787	cpus_read_unlock();
 788	return nr;
 789}
 790
 791void __init smp_detect_cpus(void)
 
 
 
 
 
 
 
 792{
 793	unsigned int cpu, mtid, c_cpus, s_cpus;
 794	struct sclp_core_info *info;
 795	u16 address;
 796
 797	/* Get CPU information */
 798	info = memblock_alloc(sizeof(*info), 8);
 799	if (!info)
 800		panic("%s: Failed to allocate %zu bytes align=0x%x\n",
 801		      __func__, sizeof(*info), 8);
 802	smp_get_core_info(info, 1);
 803	/* Find boot CPU type */
 804	if (sclp.has_core_type) {
 805		address = stap();
 806		for (cpu = 0; cpu < info->combined; cpu++)
 807			if (info->core[cpu].core_id == address) {
 808				/* The boot cpu dictates the cpu type. */
 809				boot_core_type = info->core[cpu].type;
 810				break;
 811			}
 812		if (cpu >= info->combined)
 813			panic("Could not find boot CPU type");
 814	}
 815
 816	/* Set multi-threading state for the current system */
 817	mtid = boot_core_type ? sclp.mtid : sclp.mtid_cp;
 818	mtid = (mtid < smp_max_threads) ? mtid : smp_max_threads - 1;
 819	pcpu_set_smt(mtid);
 820
 821	/* Print number of CPUs */
 822	c_cpus = s_cpus = 0;
 823	for (cpu = 0; cpu < info->combined; cpu++) {
 824		if (sclp.has_core_type &&
 825		    info->core[cpu].type != boot_core_type)
 826			continue;
 827		if (cpu < info->configured)
 828			c_cpus += smp_cpu_mtid + 1;
 829		else
 830			s_cpus += smp_cpu_mtid + 1;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 831	}
 832	pr_info("%d configured CPUs, %d standby CPUs\n", c_cpus, s_cpus);
 
 
 
 
 
 833
 834	/* Add CPUs present at boot */
 835	__smp_rescan_cpus(info, true);
 836	memblock_free(info, sizeof(*info));
 
 
 837}
 838
 839/*
 840 *	Activate a secondary processor.
 841 */
 842static void smp_start_secondary(void *cpuvoid)
 843{
 844	int cpu = raw_smp_processor_id();
 845
 846	S390_lowcore.last_update_clock = get_tod_clock();
 847	S390_lowcore.restart_stack = (unsigned long)restart_stack;
 848	S390_lowcore.restart_fn = (unsigned long)do_restart;
 849	S390_lowcore.restart_data = 0;
 850	S390_lowcore.restart_source = -1U;
 851	S390_lowcore.restart_flags = 0;
 852	restore_access_regs(S390_lowcore.access_regs_save_area);
 853	cpu_init();
 854	rcutree_report_cpu_starting(cpu);
 855	init_cpu_timer();
 856	vtime_init();
 857	vdso_getcpu_init();
 858	pfault_init();
 859	cpumask_set_cpu(cpu, &cpu_setup_mask);
 860	update_cpu_masks();
 861	notify_cpu_starting(cpu);
 862	if (topology_cpu_dedicated(cpu))
 863		set_cpu_flag(CIF_DEDICATED_CPU);
 864	else
 865		clear_cpu_flag(CIF_DEDICATED_CPU);
 866	set_cpu_online(cpu, true);
 867	inc_irq_stat(CPU_RST);
 868	local_irq_enable();
 869	cpu_startup_entry(CPUHP_AP_ONLINE_IDLE);
 870}
 871
 872/* Upping and downing of CPUs */
 873int __cpu_up(unsigned int cpu, struct task_struct *tidle)
 874{
 875	struct pcpu *pcpu = pcpu_devices + cpu;
 876	int rc;
 
 
 
 
 877
 878	if (pcpu->state != CPU_STATE_CONFIGURED)
 879		return -EIO;
 880	if (pcpu_sigp_retry(pcpu, SIGP_INITIAL_CPU_RESET, 0) !=
 881	    SIGP_CC_ORDER_CODE_ACCEPTED)
 882		return -EIO;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 883
 884	rc = pcpu_alloc_lowcore(pcpu, cpu);
 885	if (rc)
 886		return rc;
 887	/*
 888	 * Make sure global control register contents do not change
 889	 * until new CPU has initialized control registers.
 890	 */
 891	system_ctlreg_lock();
 892	pcpu_prepare_secondary(pcpu, cpu);
 893	pcpu_attach_task(pcpu, tidle);
 894	pcpu_start_fn(pcpu, smp_start_secondary, NULL);
 895	/* Wait until cpu puts itself in the online & active maps */
 896	while (!cpu_online(cpu))
 897		cpu_relax();
 898	system_ctlreg_unlock();
 899	return 0;
 900}
 901
 902static unsigned int setup_possible_cpus __initdata;
 
 
 
 903
 904static int __init _setup_possible_cpus(char *s)
 905{
 906	get_option(&s, &setup_possible_cpus);
 
 
 
 
 
 907	return 0;
 908}
 909early_param("possible_cpus", _setup_possible_cpus);
 
 
 910
 911int __cpu_disable(void)
 912{
 913	struct ctlreg cregs[16];
 914	int cpu;
 915
 916	/* Handle possible pending IPIs */
 917	smp_handle_ext_call();
 918	cpu = smp_processor_id();
 919	set_cpu_online(cpu, false);
 920	cpumask_clear_cpu(cpu, &cpu_setup_mask);
 921	update_cpu_masks();
 922	/* Disable pseudo page faults on this cpu. */
 923	pfault_fini();
 924	/* Disable interrupt sources via control register. */
 925	__local_ctl_store(0, 15, cregs);
 926	cregs[0].val  &= ~0x0000ee70UL;	/* disable all external interrupts */
 927	cregs[6].val  &= ~0xff000000UL;	/* disable all I/O interrupts */
 928	cregs[14].val &= ~0x1f000000UL;	/* disable most machine checks */
 929	__local_ctl_load(0, 15, cregs);
 930	clear_cpu_flag(CIF_NOHZ_DELAY);
 
 
 
 
 
 
 
 
 
 
 
 
 
 931	return 0;
 932}
 933
 934void __cpu_die(unsigned int cpu)
 935{
 936	struct pcpu *pcpu;
 937
 938	/* Wait until target cpu is down */
 939	pcpu = pcpu_devices + cpu;
 940	while (!pcpu_stopped(pcpu))
 941		cpu_relax();
 942	pcpu_free_lowcore(pcpu);
 943	cpumask_clear_cpu(cpu, mm_cpumask(&init_mm));
 944	cpumask_clear_cpu(cpu, &init_mm.context.cpu_attach_mask);
 
 945}
 946
 947void __noreturn cpu_die(void)
 948{
 949	idle_task_exit();
 950	pcpu_sigp_retry(pcpu_devices + smp_processor_id(), SIGP_STOP, 0);
 951	for (;;) ;
 
 952}
 953
 954void __init smp_fill_possible_mask(void)
 955{
 956	unsigned int possible, sclp_max, cpu;
 957
 958	sclp_max = max(sclp.mtid, sclp.mtid_cp) + 1;
 959	sclp_max = min(smp_max_threads, sclp_max);
 960	sclp_max = (sclp.max_cores * sclp_max) ?: nr_cpu_ids;
 961	possible = setup_possible_cpus ?: nr_cpu_ids;
 962	possible = min(possible, sclp_max);
 963	for (cpu = 0; cpu < possible && cpu < nr_cpu_ids; cpu++)
 964		set_cpu_possible(cpu, true);
 965}
 966
 967void __init smp_prepare_cpus(unsigned int max_cpus)
 968{
 969	if (register_external_irq(EXT_IRQ_EMERGENCY_SIG, do_ext_call_interrupt))
 
 
 
 
 
 
 
 
 
 970		panic("Couldn't request external interrupt 0x1201");
 971	system_ctl_set_bit(0, 14);
 972	if (register_external_irq(EXT_IRQ_EXTERNAL_CALL, do_ext_call_interrupt))
 973		panic("Couldn't request external interrupt 0x1202");
 974	system_ctl_set_bit(0, 13);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 975}
 976
 977void __init smp_prepare_boot_cpu(void)
 978{
 979	struct pcpu *pcpu = pcpu_devices;
 980
 981	WARN_ON(!cpu_present(0) || !cpu_online(0));
 982	pcpu->state = CPU_STATE_CONFIGURED;
 
 983	S390_lowcore.percpu_offset = __per_cpu_offset[0];
 984	smp_cpu_set_polarization(0, POLARIZATION_UNKNOWN);
 
 
 
 
 
 
 985}
 986
 987void __init smp_setup_processor_id(void)
 988{
 989	pcpu_devices[0].address = stap();
 990	S390_lowcore.cpu_nr = 0;
 991	S390_lowcore.spinlock_lockval = arch_spin_lockval(0);
 992	S390_lowcore.spinlock_index = 0;
 993}
 994
 995/*
 996 * the frequency of the profiling timer can be changed
 997 * by writing a multiplier value into /proc/profile.
 998 *
 999 * usually you want to run this on all CPUs ;)
1000 */
1001int setup_profiling_timer(unsigned int multiplier)
1002{
1003	return 0;
1004}
1005
1006static ssize_t cpu_configure_show(struct device *dev,
1007				  struct device_attribute *attr, char *buf)
 
1008{
1009	ssize_t count;
1010
1011	mutex_lock(&smp_cpu_state_mutex);
1012	count = sprintf(buf, "%d\n", pcpu_devices[dev->id].state);
1013	mutex_unlock(&smp_cpu_state_mutex);
1014	return count;
1015}
1016
1017static ssize_t cpu_configure_store(struct device *dev,
1018				   struct device_attribute *attr,
1019				   const char *buf, size_t count)
1020{
1021	struct pcpu *pcpu;
1022	int cpu, val, rc, i;
1023	char delim;
1024
1025	if (sscanf(buf, "%d %c", &val, &delim) != 1)
1026		return -EINVAL;
1027	if (val != 0 && val != 1)
1028		return -EINVAL;
1029	cpus_read_lock();
 
1030	mutex_lock(&smp_cpu_state_mutex);
1031	rc = -EBUSY;
1032	/* disallow configuration changes of online cpus */
1033	cpu = dev->id;
1034	cpu = smp_get_base_cpu(cpu);
1035	for (i = 0; i <= smp_cpu_mtid; i++)
1036		if (cpu_online(cpu + i))
1037			goto out;
1038	pcpu = pcpu_devices + cpu;
1039	rc = 0;
1040	switch (val) {
1041	case 0:
1042		if (pcpu->state != CPU_STATE_CONFIGURED)
1043			break;
1044		rc = sclp_core_deconfigure(pcpu->address >> smp_cpu_mt_shift);
1045		if (rc)
1046			break;
1047		for (i = 0; i <= smp_cpu_mtid; i++) {
1048			if (cpu + i >= nr_cpu_ids || !cpu_present(cpu + i))
1049				continue;
1050			pcpu[i].state = CPU_STATE_STANDBY;
1051			smp_cpu_set_polarization(cpu + i,
1052						 POLARIZATION_UNKNOWN);
1053		}
1054		topology_expect_change();
1055		break;
1056	case 1:
1057		if (pcpu->state != CPU_STATE_STANDBY)
1058			break;
1059		rc = sclp_core_configure(pcpu->address >> smp_cpu_mt_shift);
1060		if (rc)
1061			break;
1062		for (i = 0; i <= smp_cpu_mtid; i++) {
1063			if (cpu + i >= nr_cpu_ids || !cpu_present(cpu + i))
1064				continue;
1065			pcpu[i].state = CPU_STATE_CONFIGURED;
1066			smp_cpu_set_polarization(cpu + i,
1067						 POLARIZATION_UNKNOWN);
1068		}
1069		topology_expect_change();
1070		break;
1071	default:
1072		break;
1073	}
1074out:
1075	mutex_unlock(&smp_cpu_state_mutex);
1076	cpus_read_unlock();
1077	return rc ? rc : count;
1078}
1079static DEVICE_ATTR(configure, 0644, cpu_configure_show, cpu_configure_store);
 
1080
1081static ssize_t show_cpu_address(struct device *dev,
1082				struct device_attribute *attr, char *buf)
1083{
1084	return sprintf(buf, "%d\n", pcpu_devices[dev->id].address);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1085}
1086static DEVICE_ATTR(address, 0444, show_cpu_address, NULL);
 
 
 
 
 
 
 
 
1087
1088static struct attribute *cpu_common_attrs[] = {
1089	&dev_attr_configure.attr,
1090	&dev_attr_address.attr,
 
 
 
1091	NULL,
1092};
1093
1094static struct attribute_group cpu_common_attr_group = {
1095	.attrs = cpu_common_attrs,
1096};
1097
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1098static struct attribute *cpu_online_attrs[] = {
1099	&dev_attr_idle_count.attr,
1100	&dev_attr_idle_time_us.attr,
 
1101	NULL,
1102};
1103
1104static struct attribute_group cpu_online_attr_group = {
1105	.attrs = cpu_online_attrs,
1106};
1107
1108static int smp_cpu_online(unsigned int cpu)
 
1109{
1110	struct device *s = &per_cpu(cpu_device, cpu)->dev;
1111
1112	return sysfs_create_group(&s->kobj, &cpu_online_attr_group);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1113}
1114
1115static int smp_cpu_pre_down(unsigned int cpu)
1116{
1117	struct device *s = &per_cpu(cpu_device, cpu)->dev;
1118
1119	sysfs_remove_group(&s->kobj, &cpu_online_attr_group);
1120	return 0;
1121}
1122
1123static int smp_add_present_cpu(int cpu)
1124{
1125	struct device *s;
1126	struct cpu *c;
1127	int rc;
1128
1129	c = kzalloc(sizeof(*c), GFP_KERNEL);
1130	if (!c)
1131		return -ENOMEM;
1132	per_cpu(cpu_device, cpu) = c;
1133	s = &c->dev;
1134	c->hotpluggable = !!cpu;
1135	rc = register_cpu(c, cpu);
1136	if (rc)
1137		goto out;
1138	rc = sysfs_create_group(&s->kobj, &cpu_common_attr_group);
1139	if (rc)
1140		goto out_cpu;
1141	rc = topology_cpu_init(c);
1142	if (rc)
1143		goto out_topology;
1144	return 0;
1145
1146out_topology:
1147	sysfs_remove_group(&s->kobj, &cpu_common_attr_group);
1148out_cpu:
 
1149	unregister_cpu(c);
 
1150out:
1151	return rc;
1152}
1153
 
 
1154int __ref smp_rescan_cpus(void)
1155{
1156	struct sclp_core_info *info;
1157	int nr;
 
1158
1159	info = kzalloc(sizeof(*info), GFP_KERNEL);
1160	if (!info)
1161		return -ENOMEM;
1162	smp_get_core_info(info, 0);
1163	nr = __smp_rescan_cpus(info, false);
1164	kfree(info);
1165	if (nr)
 
 
 
 
 
 
 
 
 
 
1166		topology_schedule_update();
1167	return 0;
1168}
1169
1170static ssize_t __ref rescan_store(struct device *dev,
1171				  struct device_attribute *attr,
1172				  const char *buf,
1173				  size_t count)
1174{
1175	int rc;
1176
1177	rc = lock_device_hotplug_sysfs();
1178	if (rc)
1179		return rc;
1180	rc = smp_rescan_cpus();
1181	unlock_device_hotplug();
1182	return rc ? rc : count;
1183}
1184static DEVICE_ATTR_WO(rescan);
 
1185
1186static int __init s390_smp_init(void)
 
 
1187{
1188	struct device *dev_root;
1189	int cpu, rc = 0;
1190
1191	dev_root = bus_get_dev_root(&cpu_subsys);
1192	if (dev_root) {
1193		rc = device_create_file(dev_root, &dev_attr_rescan);
1194		put_device(dev_root);
1195		if (rc)
1196			return rc;
1197	}
 
 
 
 
 
 
1198
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1199	for_each_present_cpu(cpu) {
1200		rc = smp_add_present_cpu(cpu);
1201		if (rc)
1202			goto out;
1203	}
1204
1205	rc = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "s390/smp:online",
1206			       smp_cpu_online, smp_cpu_pre_down);
1207	rc = rc <= 0 ? rc : 0;
1208out:
1209	return rc;
1210}
1211subsys_initcall(s390_smp_init);
v3.1
 
   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 */
  58unsigned short __cpu_logical_map[NR_CPUS];
 
 
 
 
 
  59
  60static struct task_struct *current_set[NR_CPUS];
 
 
 
  61
  62static u8 smp_cpu_type;
  63static int smp_use_sigp_detection;
  64
  65enum s390_cpu_state {
  66	CPU_STATE_STANDBY,
  67	CPU_STATE_CONFIGURED,
 
 
 
  68};
  69
  70DEFINE_MUTEX(smp_cpu_state_mutex);
  71int smp_cpu_polarization[NR_CPUS];
  72static int smp_cpu_state[NR_CPUS];
  73static int cpu_management;
 
  74
  75static DEFINE_PER_CPU(struct cpu, cpu_devices);
 
  76
  77static void smp_ext_bitcall(int, int);
 
 
  78
  79static 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
  95static inline int cpu_stopped(int cpu)
  96{
  97	return raw_cpu_stopped(cpu_logical_map(cpu));
 
  98}
 
  99
 100void 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(&regs->gprs, &current_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
 133void 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
 159static 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 */
 184static 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
 194void 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
 202void 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 */
 211static void smp_ptlb_callback(void *info)
 212{
 213	__tlb_flush_local();
 214}
 215
 216void smp_ptlb_all(void)
 
 
 217{
 218	on_each_cpu(smp_ptlb_callback, NULL, 1);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 219}
 220EXPORT_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 */
 228void 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 */
 236struct 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 */
 244static 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 */
 259void 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] = 1UL << bit;
 266	on_each_cpu(smp_ctl_bit_callback, &parms, 1);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 267}
 268EXPORT_SYMBOL(smp_ctl_set_bit);
 269
 270/*
 271 * Clear a bit in a control register of all cpus
 272 */
 273void 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] = ~(1UL << bit);
 280	on_each_cpu(smp_ctl_bit_callback, &parms, 1);
 281}
 282EXPORT_SYMBOL(smp_ctl_clear_bit);
 283
 284#ifdef CONFIG_ZFCPDUMP
 
 285
 286static 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
 303struct save_area *zfcpdump_save_areas[NR_CPUS + 1];
 304EXPORT_SYMBOL_GPL(zfcpdump_save_areas);
 
 
 
 
 
 305
 306#else
 
 
 
 
 
 
 
 
 
 
 
 
 307
 308static inline void smp_get_save_area(unsigned int cpu, unsigned int phy_cpu) { }
 
 
 
 
 
 309
 310#endif /* CONFIG_ZFCPDUMP */
 
 
 311
 312static 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
 323static 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
 346static 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	}
 378out:
 379	kfree(info);
 380	return rc;
 
 
 
 
 
 
 381}
 382
 383static 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
 394static 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	}
 442out:
 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 */
 453int __cpuinit start_secondary(void *cpuvoid)
 454{
 455	cpu_init();
 456	preempt_disable();
 457	init_cpu_timer();
 458	init_cpu_vtimer();
 459	pfault_init();
 460
 461	notify_cpu_starting(smp_processor_id());
 462	ipi_call_lock();
 463	set_cpu_online(smp_processor_id(), true);
 464	ipi_call_unlock();
 465	__ctl_clear_bit(0, 28); /* Disable lowcore protection */
 466	S390_lowcore.restart_psw.mask = PSW_BASE_BITS | PSW_DEFAULT_KEY;
 467	S390_lowcore.restart_psw.addr =
 468		PSW_ADDR_AMODE | (unsigned long) psw_restart_int_handler;
 469	__ctl_set_bit(0, 28); /* Enable lowcore protection */
 470	/*
 471	 * Wait until the cpu which brought this one up marked it
 472	 * active before enabling interrupts.
 473	 */
 474	while (!cpumask_test_cpu(smp_processor_id(), cpu_active_mask))
 475		cpu_relax();
 476	local_irq_enable();
 477	/* cpu_idle will call schedule for us */
 478	cpu_idle();
 479	return 0;
 
 
 
 
 
 
 
 
 
 
 
 480}
 481
 482struct create_idle {
 483	struct work_struct work;
 484	struct task_struct *idle;
 485	struct completion done;
 486	int cpu;
 487};
 488
 489static void __cpuinit smp_fork_idle(struct work_struct *work)
 490{
 491	struct create_idle *c_idle;
 
 
 492
 493	c_idle = container_of(work, struct create_idle, work);
 494	c_idle->idle = fork_idle(c_idle->cpu);
 495	complete(&c_idle->done);
 496}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 497
 498static int __cpuinit smp_alloc_lowcore(int cpu)
 499{
 500	unsigned long async_stack, panic_stack;
 501	struct _lowcore *lowcore;
 502
 503	lowcore = (void *) __get_free_pages(GFP_KERNEL | GFP_DMA, LC_ORDER);
 504	if (!lowcore)
 505		return -ENOMEM;
 506	async_stack = __get_free_pages(GFP_KERNEL, ASYNC_ORDER);
 507	panic_stack = __get_free_page(GFP_KERNEL);
 508	if (!panic_stack || !async_stack)
 509		goto out;
 510	memcpy(lowcore, &S390_lowcore, 512);
 511	memset((char *)lowcore + 512, 0, sizeof(*lowcore) - 512);
 512	lowcore->async_stack = async_stack + ASYNC_SIZE;
 513	lowcore->panic_stack = panic_stack + PAGE_SIZE;
 514	lowcore->restart_psw.mask = PSW_BASE_BITS | PSW_DEFAULT_KEY;
 515	lowcore->restart_psw.addr =
 516		PSW_ADDR_AMODE | (unsigned long) restart_int_handler;
 517	if (user_mode != HOME_SPACE_MODE)
 518		lowcore->restart_psw.mask |= PSW_ASC_HOME;
 519#ifndef CONFIG_64BIT
 520	if (MACHINE_HAS_IEEE) {
 521		unsigned long save_area;
 522
 523		save_area = get_zeroed_page(GFP_KERNEL);
 524		if (!save_area)
 525			goto out;
 526		lowcore->extended_save_area_addr = (u32) save_area;
 527	}
 528#else
 529	if (vdso_alloc_per_cpu(cpu, lowcore))
 530		goto out;
 531#endif
 532	lowcore_ptr[cpu] = lowcore;
 533	return 0;
 534
 535out:
 536	free_page(panic_stack);
 537	free_pages(async_stack, ASYNC_ORDER);
 538	free_pages((unsigned long) lowcore, LC_ORDER);
 539	return -ENOMEM;
 540}
 541
 542static void smp_free_lowcore(int cpu)
 
 
 
 543{
 544	struct _lowcore *lowcore;
 545
 546	lowcore = lowcore_ptr[cpu];
 547#ifndef CONFIG_64BIT
 548	if (MACHINE_HAS_IEEE)
 549		free_page((unsigned long) lowcore->extended_save_area_addr);
 550#else
 551	vdso_free_per_cpu(cpu, lowcore);
 552#endif
 553	free_page(lowcore->panic_stack - PAGE_SIZE);
 554	free_pages(lowcore->async_stack - ASYNC_SIZE, ASYNC_ORDER);
 555	free_pages((unsigned long) lowcore, LC_ORDER);
 556	lowcore_ptr[cpu] = NULL;
 
 
 
 
 
 
 
 
 
 
 
 
 
 557}
 558
 559/* Upping and downing of CPUs */
 560int __cpuinit __cpu_up(unsigned int cpu)
 561{
 562	struct _lowcore *cpu_lowcore;
 563	struct create_idle c_idle;
 564	struct task_struct *idle;
 565	struct stack_frame *sf;
 566	u32 lowcore;
 567	int ccode;
 568
 569	if (smp_cpu_state[cpu] != CPU_STATE_CONFIGURED)
 
 
 
 570		return -EIO;
 571	idle = current_set[cpu];
 572	if (!idle) {
 573		c_idle.done = COMPLETION_INITIALIZER_ONSTACK(c_idle.done);
 574		INIT_WORK_ONSTACK(&c_idle.work, smp_fork_idle);
 575		c_idle.cpu = cpu;
 576		schedule_work(&c_idle.work);
 577		wait_for_completion(&c_idle.done);
 578		if (IS_ERR(c_idle.idle))
 579			return PTR_ERR(c_idle.idle);
 580		idle = c_idle.idle;
 581		current_set[cpu] = c_idle.idle;
 582	}
 583	init_idle(idle, cpu);
 584	if (smp_alloc_lowcore(cpu))
 585		return -ENOMEM;
 586	do {
 587		ccode = sigp(cpu, sigp_initial_cpu_reset);
 588		if (ccode == sigp_busy)
 589			udelay(10);
 590		if (ccode == sigp_not_operational)
 591			goto err_out;
 592	} while (ccode == sigp_busy);
 593
 594	lowcore = (u32)(unsigned long)lowcore_ptr[cpu];
 595	while (sigp_p(lowcore, cpu, sigp_set_prefix) == sigp_busy)
 596		udelay(10);
 597
 598	cpu_lowcore = lowcore_ptr[cpu];
 599	cpu_lowcore->kernel_stack = (unsigned long)
 600		task_stack_page(idle) + THREAD_SIZE;
 601	cpu_lowcore->thread_info = (unsigned long) task_thread_info(idle);
 602	sf = (struct stack_frame *) (cpu_lowcore->kernel_stack
 603				     - sizeof(struct pt_regs)
 604				     - sizeof(struct stack_frame));
 605	memset(sf, 0, sizeof(struct stack_frame));
 606	sf->gprs[9] = (unsigned long) sf;
 607	cpu_lowcore->save_area[15] = (unsigned long) sf;
 608	__ctl_store(cpu_lowcore->cregs_save_area, 0, 15);
 609	atomic_inc(&init_mm.context.attach_count);
 610	asm volatile(
 611		"	stam	0,15,0(%0)"
 612		: : "a" (&cpu_lowcore->access_regs_save_area) : "memory");
 613	cpu_lowcore->percpu_offset = __per_cpu_offset[cpu];
 614	cpu_lowcore->current_task = (unsigned long) idle;
 615	cpu_lowcore->cpu_nr = cpu;
 616	cpu_lowcore->kernel_asce = S390_lowcore.kernel_asce;
 617	cpu_lowcore->machine_flags = S390_lowcore.machine_flags;
 618	cpu_lowcore->ftrace_func = S390_lowcore.ftrace_func;
 619	memcpy(cpu_lowcore->stfle_fac_list, S390_lowcore.stfle_fac_list,
 620	       MAX_FACILITY_BIT/8);
 621	eieio();
 622
 623	while (sigp(cpu, sigp_restart) == sigp_busy)
 624		udelay(10);
 625
 
 
 
 
 
 
 
 
 
 
 
 
 626	while (!cpu_online(cpu))
 627		cpu_relax();
 
 628	return 0;
 
 629
 630err_out:
 631	smp_free_lowcore(cpu);
 632	return -EIO;
 633}
 634
 635static int __init setup_possible_cpus(char *s)
 636{
 637	int pcpus, cpu;
 638
 639	pcpus = simple_strtoul(s, NULL, 0);
 640	init_cpu_possible(cpumask_of(0));
 641	for (cpu = 1; cpu < pcpus && cpu < nr_cpu_ids; cpu++)
 642		set_cpu_possible(cpu, true);
 643	return 0;
 644}
 645early_param("possible_cpus", setup_possible_cpus);
 646
 647#ifdef CONFIG_HOTPLUG_CPU
 648
 649int __cpu_disable(void)
 650{
 651	struct ec_creg_mask_parms cr_parms;
 652	int cpu = smp_processor_id();
 653
 
 
 
 654	set_cpu_online(cpu, false);
 655
 656	/* Disable pfault pseudo page faults on this cpu. */
 
 657	pfault_fini();
 658
 659	memset(&cr_parms.orvals, 0, sizeof(cr_parms.orvals));
 660	memset(&cr_parms.andvals, 0xff, sizeof(cr_parms.andvals));
 661
 662	/* disable all external interrupts */
 663	cr_parms.orvals[0] = 0;
 664	cr_parms.andvals[0] = ~(1 << 15 | 1 << 14 | 1 << 13 | 1 << 11 |
 665				1 << 10 | 1 <<	9 | 1 <<  6 | 1 <<  5 |
 666				1 <<  4);
 667	/* disable all I/O interrupts */
 668	cr_parms.orvals[6] = 0;
 669	cr_parms.andvals[6] = ~(1 << 31 | 1 << 30 | 1 << 29 | 1 << 28 |
 670				1 << 27 | 1 << 26 | 1 << 25 | 1 << 24);
 671	/* disable most machine checks */
 672	cr_parms.orvals[14] = 0;
 673	cr_parms.andvals[14] = ~(1 << 28 | 1 << 27 | 1 << 26 |
 674				 1 << 25 | 1 << 24);
 675
 676	smp_ctl_bit_callback(&cr_parms);
 677
 678	return 0;
 679}
 680
 681void __cpu_die(unsigned int cpu)
 682{
 
 
 683	/* Wait until target cpu is down */
 684	while (!cpu_stopped(cpu))
 
 685		cpu_relax();
 686	while (sigp_p(0, cpu, sigp_set_prefix) == sigp_busy)
 687		udelay(10);
 688	smp_free_lowcore(cpu);
 689	atomic_dec(&init_mm.context.attach_count);
 690}
 691
 692void __noreturn cpu_die(void)
 693{
 694	idle_task_exit();
 695	while (sigp(smp_processor_id(), sigp_stop) == sigp_busy)
 696		cpu_relax();
 697	for (;;);
 698}
 699
 700#endif /* CONFIG_HOTPLUG_CPU */
 
 
 
 
 
 
 
 
 
 
 
 701
 702void __init smp_prepare_cpus(unsigned int max_cpus)
 703{
 704#ifndef CONFIG_64BIT
 705	unsigned long save_area = 0;
 706#endif
 707	unsigned long async_stack, panic_stack;
 708	struct _lowcore *lowcore;
 709
 710	smp_detect_cpus();
 711
 712	/* request the 0x1201 emergency signal external interrupt */
 713	if (register_external_interrupt(0x1201, do_ext_call_interrupt) != 0)
 714		panic("Couldn't request external interrupt 0x1201");
 715
 716	/* Reallocate current lowcore, but keep its contents. */
 717	lowcore = (void *) __get_free_pages(GFP_KERNEL | GFP_DMA, LC_ORDER);
 718	panic_stack = __get_free_page(GFP_KERNEL);
 719	async_stack = __get_free_pages(GFP_KERNEL, ASYNC_ORDER);
 720	BUG_ON(!lowcore || !panic_stack || !async_stack);
 721#ifndef CONFIG_64BIT
 722	if (MACHINE_HAS_IEEE)
 723		save_area = get_zeroed_page(GFP_KERNEL);
 724#endif
 725	local_irq_disable();
 726	local_mcck_disable();
 727	lowcore_ptr[smp_processor_id()] = lowcore;
 728	*lowcore = S390_lowcore;
 729	lowcore->panic_stack = panic_stack + PAGE_SIZE;
 730	lowcore->async_stack = async_stack + ASYNC_SIZE;
 731#ifndef CONFIG_64BIT
 732	if (MACHINE_HAS_IEEE)
 733		lowcore->extended_save_area_addr = (u32) save_area;
 734#endif
 735	set_prefix((u32)(unsigned long) lowcore);
 736	local_mcck_enable();
 737	local_irq_enable();
 738#ifdef CONFIG_64BIT
 739	if (vdso_alloc_per_cpu(smp_processor_id(), &S390_lowcore))
 740		BUG();
 741#endif
 742}
 743
 744void __init smp_prepare_boot_cpu(void)
 745{
 746	BUG_ON(smp_processor_id() != 0);
 747
 748	current_thread_info()->cpu = 0;
 749	set_cpu_present(0, true);
 750	set_cpu_online(0, true);
 751	S390_lowcore.percpu_offset = __per_cpu_offset[0];
 752	current_set[0] = current;
 753	smp_cpu_state[0] = CPU_STATE_CONFIGURED;
 754	smp_cpu_polarization[0] = POLARIZATION_UNKNWN;
 755}
 756
 757void __init smp_cpus_done(unsigned int max_cpus)
 758{
 759}
 760
 761void __init smp_setup_processor_id(void)
 762{
 
 763	S390_lowcore.cpu_nr = 0;
 764	__cpu_logical_map[0] = stap();
 
 765}
 766
 767/*
 768 * the frequency of the profiling timer can be changed
 769 * by writing a multiplier value into /proc/profile.
 770 *
 771 * usually you want to run this on all CPUs ;)
 772 */
 773int setup_profiling_timer(unsigned int multiplier)
 774{
 775	return 0;
 776}
 777
 778#ifdef CONFIG_HOTPLUG_CPU
 779static ssize_t cpu_configure_show(struct sys_device *dev,
 780				struct sysdev_attribute *attr, char *buf)
 781{
 782	ssize_t count;
 783
 784	mutex_lock(&smp_cpu_state_mutex);
 785	count = sprintf(buf, "%d\n", smp_cpu_state[dev->id]);
 786	mutex_unlock(&smp_cpu_state_mutex);
 787	return count;
 788}
 789
 790static ssize_t cpu_configure_store(struct sys_device *dev,
 791				  struct sysdev_attribute *attr,
 792				  const char *buf, size_t count)
 793{
 794	int cpu = dev->id;
 795	int val, rc;
 796	char delim;
 797
 798	if (sscanf(buf, "%d %c", &val, &delim) != 1)
 799		return -EINVAL;
 800	if (val != 0 && val != 1)
 801		return -EINVAL;
 802
 803	get_online_cpus();
 804	mutex_lock(&smp_cpu_state_mutex);
 805	rc = -EBUSY;
 806	/* disallow configuration changes of online cpus and cpu 0 */
 807	if (cpu_online(cpu) || cpu == 0)
 808		goto out;
 
 
 
 
 809	rc = 0;
 810	switch (val) {
 811	case 0:
 812		if (smp_cpu_state[cpu] == CPU_STATE_CONFIGURED) {
 813			rc = sclp_cpu_deconfigure(__cpu_logical_map[cpu]);
 814			if (!rc) {
 815				smp_cpu_state[cpu] = CPU_STATE_STANDBY;
 816				smp_cpu_polarization[cpu] = POLARIZATION_UNKNWN;
 817			}
 
 
 
 
 
 818		}
 
 819		break;
 820	case 1:
 821		if (smp_cpu_state[cpu] == CPU_STATE_STANDBY) {
 822			rc = sclp_cpu_configure(__cpu_logical_map[cpu]);
 823			if (!rc) {
 824				smp_cpu_state[cpu] = CPU_STATE_CONFIGURED;
 825				smp_cpu_polarization[cpu] = POLARIZATION_UNKNWN;
 826			}
 
 
 
 
 
 827		}
 
 828		break;
 829	default:
 830		break;
 831	}
 832out:
 833	mutex_unlock(&smp_cpu_state_mutex);
 834	put_online_cpus();
 835	return rc ? rc : count;
 836}
 837static SYSDEV_ATTR(configure, 0644, cpu_configure_show, cpu_configure_store);
 838#endif /* CONFIG_HOTPLUG_CPU */
 839
 840static ssize_t cpu_polarization_show(struct sys_device *dev,
 841				     struct sysdev_attribute *attr, char *buf)
 842{
 843	int cpu = dev->id;
 844	ssize_t count;
 845
 846	mutex_lock(&smp_cpu_state_mutex);
 847	switch (smp_cpu_polarization[cpu]) {
 848	case POLARIZATION_HRZ:
 849		count = sprintf(buf, "horizontal\n");
 850		break;
 851	case POLARIZATION_VL:
 852		count = sprintf(buf, "vertical:low\n");
 853		break;
 854	case POLARIZATION_VM:
 855		count = sprintf(buf, "vertical:medium\n");
 856		break;
 857	case POLARIZATION_VH:
 858		count = sprintf(buf, "vertical:high\n");
 859		break;
 860	default:
 861		count = sprintf(buf, "unknown\n");
 862		break;
 863	}
 864	mutex_unlock(&smp_cpu_state_mutex);
 865	return count;
 866}
 867static SYSDEV_ATTR(polarization, 0444, cpu_polarization_show, NULL);
 868
 869static ssize_t show_cpu_address(struct sys_device *dev,
 870				struct sysdev_attribute *attr, char *buf)
 871{
 872	return sprintf(buf, "%d\n", __cpu_logical_map[dev->id]);
 873}
 874static SYSDEV_ATTR(address, 0444, show_cpu_address, NULL);
 875
 876
 877static struct attribute *cpu_common_attrs[] = {
 878#ifdef CONFIG_HOTPLUG_CPU
 879	&attr_configure.attr,
 880#endif
 881	&attr_address.attr,
 882	&attr_polarization.attr,
 883	NULL,
 884};
 885
 886static struct attribute_group cpu_common_attr_group = {
 887	.attrs = cpu_common_attrs,
 888};
 889
 890static ssize_t show_capability(struct sys_device *dev,
 891				struct sysdev_attribute *attr, char *buf)
 892{
 893	unsigned int capability;
 894	int rc;
 895
 896	rc = get_cpu_capability(&capability);
 897	if (rc)
 898		return rc;
 899	return sprintf(buf, "%u\n", capability);
 900}
 901static SYSDEV_ATTR(capability, 0444, show_capability, NULL);
 902
 903static ssize_t show_idle_count(struct sys_device *dev,
 904				struct sysdev_attribute *attr, char *buf)
 905{
 906	struct s390_idle_data *idle;
 907	unsigned long long idle_count;
 908	unsigned int sequence;
 909
 910	idle = &per_cpu(s390_idle, dev->id);
 911repeat:
 912	sequence = idle->sequence;
 913	smp_rmb();
 914	if (sequence & 1)
 915		goto repeat;
 916	idle_count = idle->idle_count;
 917	if (idle->idle_enter)
 918		idle_count++;
 919	smp_rmb();
 920	if (idle->sequence != sequence)
 921		goto repeat;
 922	return sprintf(buf, "%llu\n", idle_count);
 923}
 924static SYSDEV_ATTR(idle_count, 0444, show_idle_count, NULL);
 925
 926static ssize_t show_idle_time(struct sys_device *dev,
 927				struct sysdev_attribute *attr, char *buf)
 928{
 929	struct s390_idle_data *idle;
 930	unsigned long long now, idle_time, idle_enter;
 931	unsigned int sequence;
 932
 933	idle = &per_cpu(s390_idle, dev->id);
 934	now = get_clock();
 935repeat:
 936	sequence = idle->sequence;
 937	smp_rmb();
 938	if (sequence & 1)
 939		goto repeat;
 940	idle_time = idle->idle_time;
 941	idle_enter = idle->idle_enter;
 942	if (idle_enter != 0ULL && idle_enter < now)
 943		idle_time += now - idle_enter;
 944	smp_rmb();
 945	if (idle->sequence != sequence)
 946		goto repeat;
 947	return sprintf(buf, "%llu\n", idle_time >> 12);
 948}
 949static SYSDEV_ATTR(idle_time_us, 0444, show_idle_time, NULL);
 950
 951static struct attribute *cpu_online_attrs[] = {
 952	&attr_capability.attr,
 953	&attr_idle_count.attr,
 954	&attr_idle_time_us.attr,
 955	NULL,
 956};
 957
 958static struct attribute_group cpu_online_attr_group = {
 959	.attrs = cpu_online_attrs,
 960};
 961
 962static int __cpuinit smp_cpu_notify(struct notifier_block *self,
 963				    unsigned long action, void *hcpu)
 964{
 965	unsigned int cpu = (unsigned int)(long)hcpu;
 966	struct cpu *c = &per_cpu(cpu_devices, cpu);
 967	struct sys_device *s = &c->sysdev;
 968	struct s390_idle_data *idle;
 969	int err = 0;
 970
 971	switch (action) {
 972	case CPU_ONLINE:
 973	case CPU_ONLINE_FROZEN:
 974		idle = &per_cpu(s390_idle, cpu);
 975		memset(idle, 0, sizeof(struct s390_idle_data));
 976		err = sysfs_create_group(&s->kobj, &cpu_online_attr_group);
 977		break;
 978	case CPU_DEAD:
 979	case CPU_DEAD_FROZEN:
 980		sysfs_remove_group(&s->kobj, &cpu_online_attr_group);
 981		break;
 982	}
 983	return notifier_from_errno(err);
 984}
 985
 986static struct notifier_block __cpuinitdata smp_cpu_nb = {
 987	.notifier_call = smp_cpu_notify,
 988};
 
 
 
 
 989
 990static int __devinit smp_add_present_cpu(int cpu)
 991{
 992	struct cpu *c = &per_cpu(cpu_devices, cpu);
 993	struct sys_device *s = &c->sysdev;
 994	int rc;
 995
 996	c->hotpluggable = 1;
 
 
 
 
 
 997	rc = register_cpu(c, cpu);
 998	if (rc)
 999		goto out;
1000	rc = sysfs_create_group(&s->kobj, &cpu_common_attr_group);
1001	if (rc)
1002		goto out_cpu;
1003	if (!cpu_online(cpu))
1004		goto out;
1005	rc = sysfs_create_group(&s->kobj, &cpu_online_attr_group);
1006	if (!rc)
1007		return 0;
 
1008	sysfs_remove_group(&s->kobj, &cpu_common_attr_group);
1009out_cpu:
1010#ifdef CONFIG_HOTPLUG_CPU
1011	unregister_cpu(c);
1012#endif
1013out:
1014	return rc;
1015}
1016
1017#ifdef CONFIG_HOTPLUG_CPU
1018
1019int __ref smp_rescan_cpus(void)
1020{
1021	cpumask_t newcpus;
1022	int cpu;
1023	int rc;
1024
1025	get_online_cpus();
1026	mutex_lock(&smp_cpu_state_mutex);
1027	cpumask_copy(&newcpus, cpu_present_mask);
1028	rc = __smp_rescan_cpus();
1029	if (rc)
1030		goto out;
1031	cpumask_andnot(&newcpus, cpu_present_mask, &newcpus);
1032	for_each_cpu(cpu, &newcpus) {
1033		rc = smp_add_present_cpu(cpu);
1034		if (rc)
1035			set_cpu_present(cpu, false);
1036	}
1037	rc = 0;
1038out:
1039	mutex_unlock(&smp_cpu_state_mutex);
1040	put_online_cpus();
1041	if (!cpumask_empty(&newcpus))
1042		topology_schedule_update();
1043	return rc;
1044}
1045
1046static ssize_t __ref rescan_store(struct sysdev_class *class,
1047				  struct sysdev_class_attribute *attr,
1048				  const char *buf,
1049				  size_t count)
1050{
1051	int rc;
1052
 
 
 
1053	rc = smp_rescan_cpus();
 
1054	return rc ? rc : count;
1055}
1056static SYSDEV_CLASS_ATTR(rescan, 0200, NULL, rescan_store);
1057#endif /* CONFIG_HOTPLUG_CPU */
1058
1059static ssize_t dispatching_show(struct sysdev_class *class,
1060				struct sysdev_class_attribute *attr,
1061				char *buf)
1062{
1063	ssize_t count;
 
1064
1065	mutex_lock(&smp_cpu_state_mutex);
1066	count = sprintf(buf, "%d\n", cpu_management);
1067	mutex_unlock(&smp_cpu_state_mutex);
1068	return count;
1069}
1070
1071static ssize_t dispatching_store(struct sysdev_class *dev,
1072				 struct sysdev_class_attribute *attr,
1073				 const char *buf,
1074				 size_t count)
1075{
1076	int val, rc;
1077	char delim;
1078
1079	if (sscanf(buf, "%d %c", &val, &delim) != 1)
1080		return -EINVAL;
1081	if (val != 0 && val != 1)
1082		return -EINVAL;
1083	rc = 0;
1084	get_online_cpus();
1085	mutex_lock(&smp_cpu_state_mutex);
1086	if (cpu_management == val)
1087		goto out;
1088	rc = topology_set_cpu_management(val);
1089	if (!rc)
1090		cpu_management = val;
1091out:
1092	mutex_unlock(&smp_cpu_state_mutex);
1093	put_online_cpus();
1094	return rc ? rc : count;
1095}
1096static SYSDEV_CLASS_ATTR(dispatching, 0644, dispatching_show,
1097			 dispatching_store);
1098
1099static int __init topology_init(void)
1100{
1101	int cpu;
1102	int rc;
1103
1104	register_cpu_notifier(&smp_cpu_nb);
1105
1106#ifdef CONFIG_HOTPLUG_CPU
1107	rc = sysdev_class_create_file(&cpu_sysdev_class, &attr_rescan);
1108	if (rc)
1109		return rc;
1110#endif
1111	rc = sysdev_class_create_file(&cpu_sysdev_class, &attr_dispatching);
1112	if (rc)
1113		return rc;
1114	for_each_present_cpu(cpu) {
1115		rc = smp_add_present_cpu(cpu);
1116		if (rc)
1117			return rc;
1118	}
1119	return 0;
 
 
 
 
 
1120}
1121subsys_initcall(topology_init);