Linux Audio

Check our new training course

Loading...
v6.13.7
   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/access-regs.h>
  40#include <asm/asm-offsets.h>
  41#include <asm/ctlreg.h>
  42#include <asm/pfault.h>
  43#include <asm/diag.h>
  44#include <asm/facility.h>
  45#include <asm/fpu.h>
  46#include <asm/ipl.h>
  47#include <asm/setup.h>
  48#include <asm/irq.h>
  49#include <asm/tlbflush.h>
  50#include <asm/vtimer.h>
  51#include <asm/abs_lowcore.h>
  52#include <asm/sclp.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/stacktrace.h>
  59#include <asm/topology.h>
  60#include <asm/vdso.h>
  61#include <asm/maccess.h>
  62#include "entry.h"
  63
  64enum {
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
  65	ec_schedule = 0,
 
  66	ec_call_function_single,
  67	ec_stop_cpu,
  68	ec_mcck_pending,
  69	ec_irq_work,
  70};
  71
  72enum {
  73	CPU_STATE_STANDBY,
  74	CPU_STATE_CONFIGURED,
  75};
  76
  77static u8 boot_core_type;
  78DEFINE_PER_CPU(struct pcpu, pcpu_devices);
  79/*
  80 * Pointer to the pcpu area of the boot CPU. This is required when a restart
  81 * interrupt is triggered on an offline CPU. For that case accessing percpu
  82 * data with the common primitives does not work, since the percpu offset is
  83 * stored in a non existent lowcore.
  84 */
  85static struct pcpu *ipl_pcpu;
  86
  87unsigned int smp_cpu_mt_shift;
  88EXPORT_SYMBOL(smp_cpu_mt_shift);
  89
  90unsigned int smp_cpu_mtid;
  91EXPORT_SYMBOL(smp_cpu_mtid);
  92
  93#ifdef CONFIG_CRASH_DUMP
  94__vector128 __initdata boot_cpu_vector_save_area[__NUM_VXRS];
  95#endif
  96
  97static unsigned int smp_max_threads __initdata = -1U;
  98cpumask_t cpu_setup_mask;
  99
 100static int __init early_nosmt(char *s)
 101{
 102	smp_max_threads = 1;
 103	return 0;
 104}
 105early_param("nosmt", early_nosmt);
 106
 107static int __init early_smt(char *s)
 108{
 109	get_option(&s, &smp_max_threads);
 110	return 0;
 111}
 112early_param("smt", early_smt);
 113
 114/*
 115 * The smp_cpu_state_mutex must be held when changing the state or polarization
 116 * member of a pcpu data structure within the pcpu_devices array.
 117 */
 118DEFINE_MUTEX(smp_cpu_state_mutex);
 119
 120/*
 121 * Signal processor helper functions.
 122 */
 123static inline int __pcpu_sigp_relax(u16 addr, u8 order, unsigned long parm)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 124{
 125	int cc;
 126
 127	while (1) {
 128		cc = __pcpu_sigp(addr, order, parm, NULL);
 129		if (cc != SIGP_CC_BUSY)
 130			return cc;
 131		cpu_relax();
 132	}
 133}
 134
 135static int pcpu_sigp_retry(struct pcpu *pcpu, u8 order, u32 parm)
 136{
 137	int cc, retry;
 138
 139	for (retry = 0; ; retry++) {
 140		cc = __pcpu_sigp(pcpu->address, order, parm, NULL);
 141		if (cc != SIGP_CC_BUSY)
 142			break;
 143		if (retry >= 3)
 144			udelay(10);
 145	}
 146	return cc;
 147}
 148
 149static inline int pcpu_stopped(struct pcpu *pcpu)
 150{
 151	u32 status;
 152
 153	if (__pcpu_sigp(pcpu->address, SIGP_SENSE,
 154			0, &status) != SIGP_CC_STATUS_STORED)
 155		return 0;
 156	return !!(status & (SIGP_STATUS_CHECK_STOP|SIGP_STATUS_STOPPED));
 
 157}
 158
 159static inline int pcpu_running(struct pcpu *pcpu)
 160{
 161	if (__pcpu_sigp(pcpu->address, SIGP_SENSE_RUNNING,
 162			0, NULL) != SIGP_CC_STATUS_STORED)
 163		return 1;
 164	/* Status stored condition code is equivalent to cpu not running. */
 165	return 0;
 166}
 167
 168/*
 169 * Find struct pcpu by cpu address.
 170 */
 171static struct pcpu *pcpu_find_address(const struct cpumask *mask, u16 address)
 172{
 173	int cpu;
 174
 175	for_each_cpu(cpu, mask)
 176		if (per_cpu(pcpu_devices, cpu).address == address)
 177			return &per_cpu(pcpu_devices, cpu);
 178	return NULL;
 179}
 180
 181static void pcpu_ec_call(struct pcpu *pcpu, int ec_bit)
 182{
 183	int order;
 184
 185	if (test_and_set_bit(ec_bit, &pcpu->ec_mask))
 186		return;
 187	order = pcpu_running(pcpu) ? SIGP_EXTERNAL_CALL : SIGP_EMERGENCY_SIGNAL;
 188	pcpu->ec_clk = get_tod_clock_fast();
 189	pcpu_sigp_retry(pcpu, order, 0);
 190}
 191
 192static int pcpu_alloc_lowcore(struct pcpu *pcpu, int cpu)
 193{
 194	unsigned long async_stack, nodat_stack, mcck_stack;
 195	struct lowcore *lc;
 196
 197	lc = (struct lowcore *) __get_free_pages(GFP_KERNEL | GFP_DMA, LC_ORDER);
 198	nodat_stack = __get_free_pages(GFP_KERNEL, THREAD_SIZE_ORDER);
 199	async_stack = stack_alloc();
 200	mcck_stack = stack_alloc();
 201	if (!lc || !nodat_stack || !async_stack || !mcck_stack)
 202		goto out;
 203	memcpy(lc, get_lowcore(), 512);
 
 
 
 204	memset((char *) lc + 512, 0, sizeof(*lc) - 512);
 205	lc->async_stack = async_stack + STACK_INIT_OFFSET;
 206	lc->nodat_stack = nodat_stack + STACK_INIT_OFFSET;
 207	lc->mcck_stack = mcck_stack + STACK_INIT_OFFSET;
 208	lc->cpu_nr = cpu;
 209	lc->spinlock_lockval = arch_spin_lockval(cpu);
 210	lc->spinlock_index = 0;
 211	lc->return_lpswe = gen_lpswe(__LC_RETURN_PSW);
 212	lc->return_mcck_lpswe = gen_lpswe(__LC_RETURN_MCCK_PSW);
 213	lc->preempt_count = PREEMPT_DISABLED;
 214	if (nmi_alloc_mcesa(&lc->mcesad))
 
 
 215		goto out;
 216	if (abs_lowcore_map(cpu, lc, true))
 217		goto out_mcesa;
 218	lowcore_ptr[cpu] = lc;
 219	pcpu_sigp_retry(pcpu, SIGP_SET_PREFIX, __pa(lc));
 220	return 0;
 221
 222out_mcesa:
 223	nmi_free_mcesa(&lc->mcesad);
 224out:
 225	stack_free(mcck_stack);
 226	stack_free(async_stack);
 227	free_pages(nodat_stack, THREAD_SIZE_ORDER);
 228	free_pages((unsigned long) lc, LC_ORDER);
 
 229	return -ENOMEM;
 230}
 231
 232static void pcpu_free_lowcore(struct pcpu *pcpu, int cpu)
 
 
 233{
 234	unsigned long async_stack, nodat_stack, mcck_stack;
 235	struct lowcore *lc;
 
 
 
 236
 237	lc = lowcore_ptr[cpu];
 238	nodat_stack = lc->nodat_stack - STACK_INIT_OFFSET;
 239	async_stack = lc->async_stack - STACK_INIT_OFFSET;
 240	mcck_stack = lc->mcck_stack - STACK_INIT_OFFSET;
 241	pcpu_sigp_retry(pcpu, SIGP_SET_PREFIX, 0);
 242	lowcore_ptr[cpu] = NULL;
 243	abs_lowcore_unmap(cpu);
 244	nmi_free_mcesa(&lc->mcesad);
 245	stack_free(async_stack);
 246	stack_free(mcck_stack);
 247	free_pages(nodat_stack, THREAD_SIZE_ORDER);
 248	free_pages((unsigned long) lc, LC_ORDER);
 249}
 250
 
 
 251static void pcpu_prepare_secondary(struct pcpu *pcpu, int cpu)
 252{
 253	struct lowcore *lc, *abs_lc;
 254
 255	lc = lowcore_ptr[cpu];
 256	cpumask_set_cpu(cpu, &init_mm.context.cpu_attach_mask);
 257	cpumask_set_cpu(cpu, mm_cpumask(&init_mm));
 258	lc->cpu_nr = cpu;
 259	lc->pcpu = (unsigned long)pcpu;
 260	lc->restart_flags = RESTART_FLAG_CTLREGS;
 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 = get_lowcore()->kernel_asce;
 265	lc->user_asce = s390_invalid_asce;
 266	lc->machine_flags = get_lowcore()->machine_flags;
 267	lc->user_timer = lc->system_timer =
 268		lc->steal_timer = lc->avg_steal_timer = 0;
 269	abs_lc = get_abs_lowcore();
 270	memcpy(lc->cregs_save_area, abs_lc->cregs_save_area, sizeof(lc->cregs_save_area));
 271	put_abs_lowcore(abs_lc);
 272	lc->cregs_save_area[1] = lc->kernel_asce;
 273	lc->cregs_save_area[7] = lc->user_asce;
 274	save_access_regs((unsigned int *) lc->access_regs_save_area);
 275	arch_spin_lock_setup(cpu);
 
 276}
 277
 278static void pcpu_attach_task(int cpu, struct task_struct *tsk)
 279{
 280	struct lowcore *lc;
 
 281
 282	lc = lowcore_ptr[cpu];
 283	lc->kernel_stack = (unsigned long)task_stack_page(tsk) + STACK_INIT_OFFSET;
 284	lc->current_task = (unsigned long)tsk;
 285	lc->lpp = LPP_MAGIC;
 286	lc->current_pid = tsk->pid;
 287	lc->user_timer = tsk->thread.user_timer;
 288	lc->guest_timer = tsk->thread.guest_timer;
 289	lc->system_timer = tsk->thread.system_timer;
 290	lc->hardirq_timer = tsk->thread.hardirq_timer;
 291	lc->softirq_timer = tsk->thread.softirq_timer;
 292	lc->steal_timer = 0;
 293}
 294
 295static void pcpu_start_fn(int cpu, void (*func)(void *), void *data)
 296{
 297	struct lowcore *lc;
 298
 299	lc = lowcore_ptr[cpu];
 300	lc->restart_stack = lc->kernel_stack;
 301	lc->restart_fn = (unsigned long) func;
 302	lc->restart_data = (unsigned long) data;
 303	lc->restart_source = -1U;
 304	pcpu_sigp_retry(per_cpu_ptr(&pcpu_devices, cpu), SIGP_RESTART, 0);
 305}
 306
 307typedef void (pcpu_delegate_fn)(void *);
 308
 309/*
 310 * Call function via PSW restart on pcpu and stop the current cpu.
 311 */
 312static void __pcpu_delegate(pcpu_delegate_fn *func, void *data)
 313{
 314	func(data);	/* should not return */
 315}
 316
 317static void pcpu_delegate(struct pcpu *pcpu, int cpu,
 318			  pcpu_delegate_fn *func,
 319			  void *data, unsigned long stack)
 320{
 321	struct lowcore *lc, *abs_lc;
 322	unsigned int source_cpu;
 323
 324	lc = lowcore_ptr[cpu];
 325	source_cpu = stap();
 326
 327	if (pcpu->address == source_cpu) {
 328		call_on_stack(2, stack, void, __pcpu_delegate,
 329			      pcpu_delegate_fn *, func, void *, data);
 330	}
 
 331	/* Stop target cpu (if func returns this stops the current cpu). */
 332	pcpu_sigp_retry(pcpu, SIGP_STOP, 0);
 333	pcpu_sigp_retry(pcpu, SIGP_CPU_RESET, 0);
 334	/* Restart func on the target cpu and stop the current cpu. */
 335	if (lc) {
 336		lc->restart_stack = stack;
 337		lc->restart_fn = (unsigned long)func;
 338		lc->restart_data = (unsigned long)data;
 339		lc->restart_source = source_cpu;
 340	} else {
 341		abs_lc = get_abs_lowcore();
 342		abs_lc->restart_stack = stack;
 343		abs_lc->restart_fn = (unsigned long)func;
 344		abs_lc->restart_data = (unsigned long)data;
 345		abs_lc->restart_source = source_cpu;
 346		put_abs_lowcore(abs_lc);
 347	}
 348	asm volatile(
 349		"0:	sigp	0,%0,%2	# sigp restart to target cpu\n"
 350		"	brc	2,0b	# busy, try again\n"
 351		"1:	sigp	0,%1,%3	# sigp stop to current cpu\n"
 352		"	brc	2,1b	# busy, try again\n"
 353		: : "d" (pcpu->address), "d" (source_cpu),
 354		    "K" (SIGP_RESTART), "K" (SIGP_STOP)
 355		: "0", "1", "cc");
 356	for (;;) ;
 357}
 358
 359/*
 360 * Enable additional logical cpus for multi-threading.
 361 */
 362static int pcpu_set_smt(unsigned int mtid)
 363{
 364	int cc;
 365
 366	if (smp_cpu_mtid == mtid)
 367		return 0;
 368	cc = __pcpu_sigp(0, SIGP_SET_MULTI_THREADING, mtid, NULL);
 369	if (cc == 0) {
 370		smp_cpu_mtid = mtid;
 371		smp_cpu_mt_shift = 0;
 372		while (smp_cpu_mtid >= (1U << smp_cpu_mt_shift))
 373			smp_cpu_mt_shift++;
 374		per_cpu(pcpu_devices, 0).address = stap();
 375	}
 376	return cc;
 377}
 378
 379/*
 380 * Call function on the ipl CPU.
 381 */
 382void smp_call_ipl_cpu(void (*func)(void *), void *data)
 383{
 384	struct lowcore *lc = lowcore_ptr[0];
 385
 386	if (ipl_pcpu->address == stap())
 387		lc = get_lowcore();
 388
 389	pcpu_delegate(ipl_pcpu, 0, func, data, lc->nodat_stack);
 390}
 391
 392int smp_find_processor_id(u16 address)
 393{
 394	int cpu;
 395
 396	for_each_present_cpu(cpu)
 397		if (per_cpu(pcpu_devices, cpu).address == address)
 398			return cpu;
 399	return -1;
 400}
 401
 402void schedule_mcck_handler(void)
 403{
 404	pcpu_ec_call(this_cpu_ptr(&pcpu_devices), ec_mcck_pending);
 405}
 406
 407bool notrace arch_vcpu_is_preempted(int cpu)
 408{
 409	if (test_cpu_flag_of(CIF_ENABLED_WAIT, cpu))
 410		return false;
 411	if (pcpu_running(per_cpu_ptr(&pcpu_devices, cpu)))
 412		return false;
 413	return true;
 414}
 415EXPORT_SYMBOL(arch_vcpu_is_preempted);
 416
 417void notrace smp_yield_cpu(int cpu)
 418{
 419	if (!MACHINE_HAS_DIAG9C)
 420		return;
 421	diag_stat_inc_norecursion(DIAG_STAT_X09C);
 422	asm volatile("diag %0,0,0x9c"
 423		     : : "d" (per_cpu(pcpu_devices, cpu).address));
 424}
 425EXPORT_SYMBOL_GPL(smp_yield_cpu);
 426
 427/*
 428 * Send cpus emergency shutdown signal. This gives the cpus the
 429 * opportunity to complete outstanding interrupts.
 430 */
 431void notrace smp_emergency_stop(void)
 432{
 433	static arch_spinlock_t lock = __ARCH_SPIN_LOCK_UNLOCKED;
 434	static cpumask_t cpumask;
 435	u64 end;
 436	int cpu;
 437
 438	arch_spin_lock(&lock);
 439	cpumask_copy(&cpumask, cpu_online_mask);
 440	cpumask_clear_cpu(smp_processor_id(), &cpumask);
 441
 442	end = get_tod_clock() + (1000000UL << 12);
 443	for_each_cpu(cpu, &cpumask) {
 444		struct pcpu *pcpu = per_cpu_ptr(&pcpu_devices, cpu);
 445		set_bit(ec_stop_cpu, &pcpu->ec_mask);
 446		while (__pcpu_sigp(pcpu->address, SIGP_EMERGENCY_SIGNAL,
 447				   0, NULL) == SIGP_CC_BUSY &&
 448		       get_tod_clock() < end)
 449			cpu_relax();
 450	}
 451	while (get_tod_clock() < end) {
 452		for_each_cpu(cpu, &cpumask)
 453			if (pcpu_stopped(per_cpu_ptr(&pcpu_devices, cpu)))
 454				cpumask_clear_cpu(cpu, &cpumask);
 455		if (cpumask_empty(&cpumask))
 456			break;
 457		cpu_relax();
 458	}
 459	arch_spin_unlock(&lock);
 460}
 461NOKPROBE_SYMBOL(smp_emergency_stop);
 462
 463/*
 464 * Stop all cpus but the current one.
 465 */
 466void smp_send_stop(void)
 467{
 468	struct pcpu *pcpu;
 469	int cpu;
 470
 471	/* Disable all interrupts/machine checks */
 472	__load_psw_mask(PSW_KERNEL_BITS);
 473	trace_hardirqs_off();
 474
 475	debug_set_critical();
 
 
 476
 477	if (oops_in_progress)
 478		smp_emergency_stop();
 479
 480	/* stop all processors */
 481	for_each_online_cpu(cpu) {
 482		if (cpu == smp_processor_id())
 483			continue;
 484		pcpu = per_cpu_ptr(&pcpu_devices, cpu);
 485		pcpu_sigp_retry(pcpu, SIGP_STOP, 0);
 486		while (!pcpu_stopped(pcpu))
 487			cpu_relax();
 488	}
 489}
 490
 491/*
 
 
 
 
 
 
 
 
 
 492 * This is the main routine where commands issued by other
 493 * cpus are handled.
 494 */
 495static void smp_handle_ext_call(void)
 
 496{
 497	unsigned long bits;
 
 
 
 
 
 
 
 
 
 
 
 498
 499	/* handle bit signal external calls */
 500	bits = this_cpu_xchg(pcpu_devices.ec_mask, 0);
 501	if (test_bit(ec_stop_cpu, &bits))
 502		smp_stop_cpu();
 
 503	if (test_bit(ec_schedule, &bits))
 504		scheduler_ipi();
 
 
 
 
 505	if (test_bit(ec_call_function_single, &bits))
 506		generic_smp_call_function_single_interrupt();
 507	if (test_bit(ec_mcck_pending, &bits))
 508		s390_handle_mcck();
 509	if (test_bit(ec_irq_work, &bits))
 510		irq_work_run();
 511}
 512
 513static void do_ext_call_interrupt(struct ext_code ext_code,
 514				  unsigned int param32, unsigned long param64)
 515{
 516	inc_irq_stat(ext_code.code == 0x1202 ? IRQEXT_EXC : IRQEXT_EMS);
 517	smp_handle_ext_call();
 518}
 519
 520void arch_send_call_function_ipi_mask(const struct cpumask *mask)
 521{
 522	int cpu;
 523
 524	for_each_cpu(cpu, mask)
 525		pcpu_ec_call(per_cpu_ptr(&pcpu_devices, cpu), ec_call_function_single);
 526}
 527
 528void arch_send_call_function_single_ipi(int cpu)
 529{
 530	pcpu_ec_call(per_cpu_ptr(&pcpu_devices, cpu), ec_call_function_single);
 531}
 532
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 533/*
 534 * this function sends a 'reschedule' IPI to another CPU.
 535 * it goes straight through and wastes no time serializing
 536 * anything. Worst case is that we lose a reschedule ...
 537 */
 538void arch_smp_send_reschedule(int cpu)
 539{
 540	pcpu_ec_call(per_cpu_ptr(&pcpu_devices, cpu), ec_schedule);
 541}
 542
 543#ifdef CONFIG_IRQ_WORK
 544void arch_irq_work_raise(void)
 
 
 
 
 
 
 
 
 
 
 
 545{
 546	pcpu_ec_call(this_cpu_ptr(&pcpu_devices), ec_irq_work);
 547}
 548#endif
 549
 550#ifdef CONFIG_CRASH_DUMP
 
 
 
 551
 552int smp_store_status(int cpu)
 
 
 
 553{
 554	struct lowcore *lc;
 555	struct pcpu *pcpu;
 556	unsigned long pa;
 557
 558	pcpu = per_cpu_ptr(&pcpu_devices, cpu);
 559	lc = lowcore_ptr[cpu];
 560	pa = __pa(&lc->floating_pt_save_area);
 561	if (__pcpu_sigp_relax(pcpu->address, SIGP_STORE_STATUS_AT_ADDRESS,
 562			      pa) != SIGP_CC_ORDER_CODE_ACCEPTED)
 563		return -EIO;
 564	if (!cpu_has_vx() && !MACHINE_HAS_GS)
 565		return 0;
 566	pa = lc->mcesad & MCESA_ORIGIN_MASK;
 567	if (MACHINE_HAS_GS)
 568		pa |= lc->mcesad & MCESA_LC_MASK;
 569	if (__pcpu_sigp_relax(pcpu->address, SIGP_STORE_ADDITIONAL_STATUS,
 570			      pa) != SIGP_CC_ORDER_CODE_ACCEPTED)
 571		return -EIO;
 572	return 0;
 573}
 
 574
 575/*
 576 * Collect CPU state of the previous, crashed system.
 577 * There are three cases:
 578 * 1) standard zfcp/nvme dump
 579 *    condition: OLDMEM_BASE == NULL && is_ipl_type_dump() == true
 580 *    The state for all CPUs except the boot CPU needs to be collected
 581 *    with sigp stop-and-store-status. The boot CPU state is located in
 582 *    the absolute lowcore of the memory stored in the HSA. The zcore code
 583 *    will copy the boot CPU state from the HSA.
 584 * 2) stand-alone kdump for SCSI/NVMe (zfcp/nvme dump with swapped memory)
 585 *    condition: OLDMEM_BASE != NULL && is_ipl_type_dump() == true
 586 *    The state for all CPUs except the boot CPU needs to be collected
 587 *    with sigp stop-and-store-status. The firmware or the boot-loader
 588 *    stored the registers of the boot CPU in the absolute lowcore in the
 589 *    memory of the old system.
 590 * 3) kdump or stand-alone kdump for DASD
 591 *    condition: OLDMEM_BASE != NULL && is_ipl_type_dump() == false
 592 *    The state for all CPUs except the boot CPU needs to be collected
 593 *    with sigp stop-and-store-status. The kexec code or the boot-loader
 594 *    stored the registers of the boot CPU in the memory of the old system.
 595 *
 596 * Note that the legacy kdump mode where the old kernel stored the CPU states
 597 * does no longer exist: setup_arch() explicitly deactivates the elfcorehdr=
 598 * kernel parameter. The is_kdump_kernel() implementation on s390 is independent
 599 * of the elfcorehdr= parameter.
 600 */
 601static bool dump_available(void)
 602{
 603	return oldmem_data.start || is_ipl_type_dump();
 
 
 604}
 
 605
 606void __init smp_save_dump_ipl_cpu(void)
 
 
 
 
 
 607{
 608	struct save_area *sa;
 609	void *regs;
 610
 611	if (!dump_available())
 612		return;
 613	sa = save_area_alloc(true);
 614	regs = memblock_alloc(512, 8);
 615	if (!sa || !regs)
 616		panic("could not allocate memory for boot CPU save area\n");
 617	copy_oldmem_kernel(regs, __LC_FPREGS_SAVE_AREA, 512);
 618	save_area_add_regs(sa, regs);
 619	memblock_free(regs, 512);
 620	if (cpu_has_vx())
 621		save_area_add_vxrs(sa, boot_cpu_vector_save_area);
 622}
 623
 624void __init smp_save_dump_secondary_cpus(void)
 625{
 626	int addr, boot_cpu_addr, max_cpu_addr;
 627	struct save_area *sa;
 628	void *page;
 629
 630	if (!dump_available())
 631		return;
 632	/* Allocate a page as dumping area for the store status sigps */
 633	page = memblock_alloc_low(PAGE_SIZE, PAGE_SIZE);
 634	if (!page)
 635		panic("ERROR: Failed to allocate %lx bytes below %lx\n",
 636		      PAGE_SIZE, 1UL << 31);
 637
 638	/* Set multi-threading state to the previous system. */
 639	pcpu_set_smt(sclp.mtid_prev);
 640	boot_cpu_addr = stap();
 641	max_cpu_addr = SCLP_MAX_CORES << sclp.mtid_prev;
 642	for (addr = 0; addr <= max_cpu_addr; addr++) {
 643		if (addr == boot_cpu_addr)
 644			continue;
 645		if (__pcpu_sigp_relax(addr, SIGP_SENSE, 0) ==
 646		    SIGP_CC_NOT_OPERATIONAL)
 647			continue;
 648		sa = save_area_alloc(false);
 649		if (!sa)
 650			panic("could not allocate memory for save area\n");
 651		__pcpu_sigp_relax(addr, SIGP_STORE_STATUS_AT_ADDRESS, __pa(page));
 652		save_area_add_regs(sa, page);
 653		if (cpu_has_vx()) {
 654			__pcpu_sigp_relax(addr, SIGP_STORE_ADDITIONAL_STATUS, __pa(page));
 655			save_area_add_vxrs(sa, page);
 656		}
 657	}
 658	memblock_free(page, PAGE_SIZE);
 659	diag_amode31_ops.diag308_reset();
 660	pcpu_set_smt(0);
 661}
 662#endif /* CONFIG_CRASH_DUMP */
 663
 664void smp_cpu_set_polarization(int cpu, int val)
 665{
 666	per_cpu(pcpu_devices, cpu).polarization = val;
 667}
 668
 669int smp_cpu_get_polarization(int cpu)
 670{
 671	return per_cpu(pcpu_devices, cpu).polarization;
 
 672}
 673
 674void smp_cpu_set_capacity(int cpu, unsigned long val)
 675{
 676	per_cpu(pcpu_devices, cpu).capacity = val;
 677}
 678
 679unsigned long smp_cpu_get_capacity(int cpu)
 680{
 681	return per_cpu(pcpu_devices, cpu).capacity;
 
 
 682}
 683
 684void smp_set_core_capacity(int cpu, unsigned long val)
 685{
 686	int i;
 687
 688	cpu = smp_get_base_cpu(cpu);
 689	for (i = cpu; (i <= cpu + smp_cpu_mtid) && (i < nr_cpu_ids); i++)
 690		smp_cpu_set_capacity(i, val);
 691}
 692
 693int smp_cpu_get_cpu_address(int cpu)
 694{
 695	return per_cpu(pcpu_devices, cpu).address;
 696}
 697
 698static void __ref smp_get_core_info(struct sclp_core_info *info, int early)
 699{
 700	static int use_sigp_detection;
 
 701	int address;
 702
 703	if (use_sigp_detection || sclp_get_core_info(info, early)) {
 
 704		use_sigp_detection = 1;
 705		for (address = 0;
 706		     address < (SCLP_MAX_CORES << smp_cpu_mt_shift);
 707		     address += (1U << smp_cpu_mt_shift)) {
 708			if (__pcpu_sigp_relax(address, SIGP_SENSE, 0) ==
 709			    SIGP_CC_NOT_OPERATIONAL)
 710				continue;
 711			info->core[info->configured].core_id =
 712				address >> smp_cpu_mt_shift;
 713			info->configured++;
 714		}
 715		info->combined = info->configured;
 716	}
 
 717}
 718
 719static int smp_add_core(struct sclp_core_entry *core, cpumask_t *avail,
 720			bool configured, bool early)
 
 
 721{
 722	struct pcpu *pcpu;
 
 723	int cpu, nr, i;
 724	u16 address;
 725
 726	nr = 0;
 727	if (sclp.has_core_type && core->type != boot_core_type)
 728		return nr;
 729	cpu = cpumask_first(avail);
 730	address = core->core_id << smp_cpu_mt_shift;
 731	for (i = 0; (i <= smp_cpu_mtid) && (cpu < nr_cpu_ids); i++) {
 732		if (pcpu_find_address(cpu_present_mask, address + i))
 733			continue;
 734		pcpu = per_cpu_ptr(&pcpu_devices, cpu);
 735		pcpu->address = address + i;
 736		if (configured)
 737			pcpu->state = CPU_STATE_CONFIGURED;
 738		else
 739			pcpu->state = CPU_STATE_STANDBY;
 740		smp_cpu_set_polarization(cpu, POLARIZATION_UNKNOWN);
 741		smp_cpu_set_capacity(cpu, CPU_CAPACITY_HIGH);
 742		set_cpu_present(cpu, true);
 743		if (!early && arch_register_cpu(cpu))
 744			set_cpu_present(cpu, false);
 745		else
 746			nr++;
 747		cpumask_clear_cpu(cpu, avail);
 748		cpu = cpumask_next(cpu, avail);
 749	}
 750	return nr;
 751}
 752
 753static int __smp_rescan_cpus(struct sclp_core_info *info, bool early)
 754{
 755	struct sclp_core_entry *core;
 756	static cpumask_t avail;
 757	bool configured;
 758	u16 core_id;
 759	int nr, i;
 760
 761	cpus_read_lock();
 762	mutex_lock(&smp_cpu_state_mutex);
 763	nr = 0;
 764	cpumask_xor(&avail, cpu_possible_mask, cpu_present_mask);
 765	/*
 766	 * Add IPL core first (which got logical CPU number 0) to make sure
 767	 * that all SMT threads get subsequent logical CPU numbers.
 768	 */
 769	if (early) {
 770		core_id = per_cpu(pcpu_devices, 0).address >> smp_cpu_mt_shift;
 771		for (i = 0; i < info->configured; i++) {
 772			core = &info->core[i];
 773			if (core->core_id == core_id) {
 774				nr += smp_add_core(core, &avail, true, early);
 775				break;
 776			}
 777		}
 778	}
 779	for (i = 0; i < info->combined; i++) {
 780		configured = i < info->configured;
 781		nr += smp_add_core(&info->core[i], &avail, configured, early);
 782	}
 783	mutex_unlock(&smp_cpu_state_mutex);
 784	cpus_read_unlock();
 785	return nr;
 786}
 787
 788void __init smp_detect_cpus(void)
 789{
 790	unsigned int cpu, mtid, c_cpus, s_cpus;
 791	struct sclp_core_info *info;
 792	u16 address;
 793
 794	/* Get CPU information */
 795	info = memblock_alloc(sizeof(*info), 8);
 796	if (!info)
 797		panic("%s: Failed to allocate %zu bytes align=0x%x\n",
 798		      __func__, sizeof(*info), 8);
 799	smp_get_core_info(info, 1);
 800	/* Find boot CPU type */
 801	if (sclp.has_core_type) {
 802		address = stap();
 803		for (cpu = 0; cpu < info->combined; cpu++)
 804			if (info->core[cpu].core_id == address) {
 805				/* The boot cpu dictates the cpu type. */
 806				boot_core_type = info->core[cpu].type;
 807				break;
 808			}
 809		if (cpu >= info->combined)
 810			panic("Could not find boot CPU type");
 811	}
 812
 813	/* Set multi-threading state for the current system */
 814	mtid = boot_core_type ? sclp.mtid : sclp.mtid_cp;
 815	mtid = (mtid < smp_max_threads) ? mtid : smp_max_threads - 1;
 816	pcpu_set_smt(mtid);
 817
 818	/* Print number of CPUs */
 819	c_cpus = s_cpus = 0;
 820	for (cpu = 0; cpu < info->combined; cpu++) {
 821		if (sclp.has_core_type &&
 822		    info->core[cpu].type != boot_core_type)
 823			continue;
 824		if (cpu < info->configured)
 825			c_cpus += smp_cpu_mtid + 1;
 826		else
 827			s_cpus += smp_cpu_mtid + 1;
 
 828	}
 829	pr_info("%d configured CPUs, %d standby CPUs\n", c_cpus, s_cpus);
 830	memblock_free(info, sizeof(*info));
 
 
 
 831}
 832
 833/*
 834 *	Activate a secondary processor.
 835 */
 836static void smp_start_secondary(void *cpuvoid)
 837{
 838	struct lowcore *lc = get_lowcore();
 839	int cpu = raw_smp_processor_id();
 840
 841	lc->last_update_clock = get_tod_clock();
 842	lc->restart_stack = (unsigned long)restart_stack;
 843	lc->restart_fn = (unsigned long)do_restart;
 844	lc->restart_data = 0;
 845	lc->restart_source = -1U;
 846	lc->restart_flags = 0;
 847	restore_access_regs(lc->access_regs_save_area);
 848	cpu_init();
 849	rcutree_report_cpu_starting(cpu);
 850	init_cpu_timer();
 851	vtime_init();
 852	vdso_getcpu_init();
 853	pfault_init();
 854	cpumask_set_cpu(cpu, &cpu_setup_mask);
 855	update_cpu_masks();
 856	notify_cpu_starting(cpu);
 857	if (topology_cpu_dedicated(cpu))
 858		set_cpu_flag(CIF_DEDICATED_CPU);
 859	else
 860		clear_cpu_flag(CIF_DEDICATED_CPU);
 861	set_cpu_online(cpu, true);
 862	inc_irq_stat(CPU_RST);
 863	local_irq_enable();
 864	cpu_startup_entry(CPUHP_AP_ONLINE_IDLE);
 
 865}
 866
 867/* Upping and downing of CPUs */
 868int __cpu_up(unsigned int cpu, struct task_struct *tidle)
 869{
 870	struct pcpu *pcpu = per_cpu_ptr(&pcpu_devices, cpu);
 871	int rc;
 872
 
 873	if (pcpu->state != CPU_STATE_CONFIGURED)
 874		return -EIO;
 875	if (pcpu_sigp_retry(pcpu, SIGP_INITIAL_CPU_RESET, 0) !=
 876	    SIGP_CC_ORDER_CODE_ACCEPTED)
 877		return -EIO;
 878
 879	rc = pcpu_alloc_lowcore(pcpu, cpu);
 880	if (rc)
 881		return rc;
 882	/*
 883	 * Make sure global control register contents do not change
 884	 * until new CPU has initialized control registers.
 885	 */
 886	system_ctlreg_lock();
 887	pcpu_prepare_secondary(pcpu, cpu);
 888	pcpu_attach_task(cpu, tidle);
 889	pcpu_start_fn(cpu, smp_start_secondary, NULL);
 890	/* Wait until cpu puts itself in the online & active maps */
 891	while (!cpu_online(cpu))
 892		cpu_relax();
 893	system_ctlreg_unlock();
 894	return 0;
 895}
 896
 897static unsigned int setup_possible_cpus __initdata;
 898
 899static int __init _setup_possible_cpus(char *s)
 900{
 901	get_option(&s, &setup_possible_cpus);
 
 
 
 
 
 
 902	return 0;
 903}
 904early_param("possible_cpus", _setup_possible_cpus);
 
 
 905
 906int __cpu_disable(void)
 907{
 908	struct ctlreg cregs[16];
 909	int cpu;
 910
 911	/* Handle possible pending IPIs */
 912	smp_handle_ext_call();
 913	cpu = smp_processor_id();
 914	set_cpu_online(cpu, false);
 915	cpumask_clear_cpu(cpu, &cpu_setup_mask);
 916	update_cpu_masks();
 917	/* Disable pseudo page faults on this cpu. */
 918	pfault_fini();
 919	/* Disable interrupt sources via control register. */
 920	__local_ctl_store(0, 15, cregs);
 921	cregs[0].val  &= ~0x0000ee70UL;	/* disable all external interrupts */
 922	cregs[6].val  &= ~0xff000000UL;	/* disable all I/O interrupts */
 923	cregs[14].val &= ~0x1f000000UL;	/* disable most machine checks */
 924	__local_ctl_load(0, 15, cregs);
 925	clear_cpu_flag(CIF_NOHZ_DELAY);
 926	return 0;
 927}
 928
 929void __cpu_die(unsigned int cpu)
 930{
 931	struct pcpu *pcpu;
 932
 933	/* Wait until target cpu is down */
 934	pcpu = per_cpu_ptr(&pcpu_devices, cpu);
 935	while (!pcpu_stopped(pcpu))
 936		cpu_relax();
 937	pcpu_free_lowcore(pcpu, cpu);
 938	cpumask_clear_cpu(cpu, mm_cpumask(&init_mm));
 939	cpumask_clear_cpu(cpu, &init_mm.context.cpu_attach_mask);
 940	pcpu->flags = 0;
 941}
 942
 943void __noreturn cpu_die(void)
 944{
 945	idle_task_exit();
 946	pcpu_sigp_retry(this_cpu_ptr(&pcpu_devices), SIGP_STOP, 0);
 947	for (;;) ;
 948}
 949
 950void __init smp_fill_possible_mask(void)
 951{
 952	unsigned int possible, sclp_max, cpu;
 953
 954	sclp_max = max(sclp.mtid, sclp.mtid_cp) + 1;
 955	sclp_max = min(smp_max_threads, sclp_max);
 956	sclp_max = (sclp.max_cores * sclp_max) ?: nr_cpu_ids;
 957	possible = setup_possible_cpus ?: nr_cpu_ids;
 958	possible = min(possible, sclp_max);
 959	for (cpu = 0; cpu < possible && cpu < nr_cpu_ids; cpu++)
 960		set_cpu_possible(cpu, true);
 961}
 962
 963void __init smp_prepare_cpus(unsigned int max_cpus)
 964{
 965	if (register_external_irq(EXT_IRQ_EMERGENCY_SIG, do_ext_call_interrupt))
 
 966		panic("Couldn't request external interrupt 0x1201");
 967	system_ctl_set_bit(0, 14);
 968	if (register_external_irq(EXT_IRQ_EXTERNAL_CALL, do_ext_call_interrupt))
 969		panic("Couldn't request external interrupt 0x1202");
 970	system_ctl_set_bit(0, 13);
 971	smp_rescan_cpus(true);
 972}
 973
 974void __init smp_prepare_boot_cpu(void)
 975{
 976	struct lowcore *lc = get_lowcore();
 977
 978	WARN_ON(!cpu_present(0) || !cpu_online(0));
 979	lc->percpu_offset = __per_cpu_offset[0];
 980	ipl_pcpu = per_cpu_ptr(&pcpu_devices, 0);
 981	ipl_pcpu->state = CPU_STATE_CONFIGURED;
 982	lc->pcpu = (unsigned long)ipl_pcpu;
 983	smp_cpu_set_polarization(0, POLARIZATION_UNKNOWN);
 984	smp_cpu_set_capacity(0, CPU_CAPACITY_HIGH);
 
 
 
 985}
 986
 987void __init smp_setup_processor_id(void)
 988{
 989	struct lowcore *lc = get_lowcore();
 990
 991	lc->cpu_nr = 0;
 992	per_cpu(pcpu_devices, 0).address = stap();
 993	lc->spinlock_lockval = arch_spin_lockval(0);
 994	lc->spinlock_index = 0;
 995}
 996
 997/*
 998 * the frequency of the profiling timer can be changed
 999 * by writing a multiplier value into /proc/profile.
1000 *
1001 * usually you want to run this on all CPUs ;)
1002 */
1003int setup_profiling_timer(unsigned int multiplier)
1004{
1005	return 0;
1006}
1007
 
1008static ssize_t cpu_configure_show(struct device *dev,
1009				  struct device_attribute *attr, char *buf)
1010{
1011	ssize_t count;
1012
1013	mutex_lock(&smp_cpu_state_mutex);
1014	count = sysfs_emit(buf, "%d\n", per_cpu(pcpu_devices, dev->id).state);
1015	mutex_unlock(&smp_cpu_state_mutex);
1016	return count;
1017}
1018
1019static ssize_t cpu_configure_store(struct device *dev,
1020				   struct device_attribute *attr,
1021				   const char *buf, size_t count)
1022{
1023	struct pcpu *pcpu;
1024	int cpu, val, rc, i;
1025	char delim;
1026
1027	if (sscanf(buf, "%d %c", &val, &delim) != 1)
1028		return -EINVAL;
1029	if (val != 0 && val != 1)
1030		return -EINVAL;
1031	cpus_read_lock();
1032	mutex_lock(&smp_cpu_state_mutex);
1033	rc = -EBUSY;
1034	/* disallow configuration changes of online cpus */
1035	cpu = dev->id;
1036	cpu = smp_get_base_cpu(cpu);
1037	for (i = 0; i <= smp_cpu_mtid; i++)
1038		if (cpu_online(cpu + i))
1039			goto out;
1040	pcpu = per_cpu_ptr(&pcpu_devices, cpu);
1041	rc = 0;
1042	switch (val) {
1043	case 0:
1044		if (pcpu->state != CPU_STATE_CONFIGURED)
1045			break;
1046		rc = sclp_core_deconfigure(pcpu->address >> smp_cpu_mt_shift);
1047		if (rc)
1048			break;
1049		for (i = 0; i <= smp_cpu_mtid; i++) {
1050			if (cpu + i >= nr_cpu_ids || !cpu_present(cpu + i))
1051				continue;
1052			per_cpu(pcpu_devices, cpu + i).state = CPU_STATE_STANDBY;
1053			smp_cpu_set_polarization(cpu + i,
1054						 POLARIZATION_UNKNOWN);
1055		}
1056		topology_expect_change();
1057		break;
1058	case 1:
1059		if (pcpu->state != CPU_STATE_STANDBY)
1060			break;
1061		rc = sclp_core_configure(pcpu->address >> smp_cpu_mt_shift);
1062		if (rc)
1063			break;
1064		for (i = 0; i <= smp_cpu_mtid; i++) {
1065			if (cpu + i >= nr_cpu_ids || !cpu_present(cpu + i))
1066				continue;
1067			per_cpu(pcpu_devices, cpu + i).state = CPU_STATE_CONFIGURED;
1068			smp_cpu_set_polarization(cpu + i,
1069						 POLARIZATION_UNKNOWN);
1070		}
1071		topology_expect_change();
1072		break;
1073	default:
1074		break;
1075	}
1076out:
1077	mutex_unlock(&smp_cpu_state_mutex);
1078	cpus_read_unlock();
1079	return rc ? rc : count;
1080}
1081static DEVICE_ATTR(configure, 0644, cpu_configure_show, cpu_configure_store);
 
1082
1083static ssize_t show_cpu_address(struct device *dev,
1084				struct device_attribute *attr, char *buf)
1085{
1086	return sysfs_emit(buf, "%d\n", per_cpu(pcpu_devices, dev->id).address);
1087}
1088static DEVICE_ATTR(address, 0444, show_cpu_address, NULL);
1089
1090static struct attribute *cpu_common_attrs[] = {
 
1091	&dev_attr_configure.attr,
 
1092	&dev_attr_address.attr,
1093	NULL,
1094};
1095
1096static struct attribute_group cpu_common_attr_group = {
1097	.attrs = cpu_common_attrs,
1098};
1099
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1100static struct attribute *cpu_online_attrs[] = {
1101	&dev_attr_idle_count.attr,
1102	&dev_attr_idle_time_us.attr,
1103	NULL,
1104};
1105
1106static struct attribute_group cpu_online_attr_group = {
1107	.attrs = cpu_online_attrs,
1108};
1109
1110static int smp_cpu_online(unsigned int cpu)
1111{
1112	struct cpu *c = per_cpu_ptr(&cpu_devices, cpu);
1113
1114	return sysfs_create_group(&c->dev.kobj, &cpu_online_attr_group);
1115}
1116
1117static int smp_cpu_pre_down(unsigned int cpu)
1118{
1119	struct cpu *c = per_cpu_ptr(&cpu_devices, cpu);
1120
1121	sysfs_remove_group(&c->dev.kobj, &cpu_online_attr_group);
1122	return 0;
 
 
 
 
 
 
 
 
 
 
 
 
1123}
1124
1125bool arch_cpu_is_hotpluggable(int cpu)
1126{
1127	return !!cpu;
1128}
1129
1130int arch_register_cpu(int cpu)
1131{
1132	struct cpu *c = per_cpu_ptr(&cpu_devices, cpu);
 
1133	int rc;
1134
1135	c->hotpluggable = arch_cpu_is_hotpluggable(cpu);
1136	rc = register_cpu(c, cpu);
1137	if (rc)
1138		goto out;
1139	rc = sysfs_create_group(&c->dev.kobj, &cpu_common_attr_group);
1140	if (rc)
1141		goto out_cpu;
 
 
 
 
 
1142	rc = topology_cpu_init(c);
1143	if (rc)
1144		goto out_topology;
1145	return 0;
1146
1147out_topology:
1148	sysfs_remove_group(&c->dev.kobj, &cpu_common_attr_group);
 
 
 
1149out_cpu:
 
1150	unregister_cpu(c);
 
1151out:
1152	return rc;
1153}
1154
1155int __ref smp_rescan_cpus(bool early)
 
 
1156{
1157	struct sclp_core_info *info;
1158	int nr;
1159
1160	info = kzalloc(sizeof(*info), GFP_KERNEL);
1161	if (!info)
1162		return -ENOMEM;
1163	smp_get_core_info(info, 0);
1164	nr = __smp_rescan_cpus(info, early);
 
 
 
1165	kfree(info);
1166	if (nr)
1167		topology_schedule_update();
1168	return 0;
1169}
1170
1171static ssize_t __ref rescan_store(struct device *dev,
1172				  struct device_attribute *attr,
1173				  const char *buf,
1174				  size_t count)
1175{
1176	int rc;
1177
1178	rc = lock_device_hotplug_sysfs();
1179	if (rc)
1180		return rc;
1181	rc = smp_rescan_cpus(false);
1182	unlock_device_hotplug();
1183	return rc ? rc : count;
1184}
1185static DEVICE_ATTR_WO(rescan);
 
1186
1187static int __init s390_smp_init(void)
1188{
1189	struct device *dev_root;
1190	int rc;
1191
1192	dev_root = bus_get_dev_root(&cpu_subsys);
1193	if (dev_root) {
1194		rc = device_create_file(dev_root, &dev_attr_rescan);
1195		put_device(dev_root);
 
 
 
 
1196		if (rc)
1197			return rc;
1198	}
1199	rc = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "s390/smp:online",
1200			       smp_cpu_online, smp_cpu_pre_down);
1201	rc = rc <= 0 ? rc : 0;
1202	return rc;
1203}
1204subsys_initcall(s390_smp_init);
v3.5.6
 
   1/*
   2 *  SMP related functions
   3 *
   4 *    Copyright IBM Corp. 1999,2012
   5 *    Author(s): Denis Joseph Barrow,
   6 *		 Martin Schwidefsky <schwidefsky@de.ibm.com>,
   7 *		 Heiko Carstens <heiko.carstens@de.ibm.com>,
   8 *
   9 *  based on other smp stuff by
  10 *    (c) 1995 Alan Cox, CymruNET Ltd  <alan@cymru.net>
  11 *    (c) 1998 Ingo Molnar
  12 *
  13 * The code outside of smp.c uses logical cpu numbers, only smp.c does
  14 * the translation of logical to physical cpu ids. All new code that
  15 * operates on physical cpu numbers needs to go into smp.c.
  16 */
  17
  18#define KMSG_COMPONENT "cpu"
  19#define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
  20
  21#include <linux/workqueue.h>
  22#include <linux/module.h>
 
  23#include <linux/init.h>
  24#include <linux/mm.h>
  25#include <linux/err.h>
  26#include <linux/spinlock.h>
  27#include <linux/kernel_stat.h>
  28#include <linux/delay.h>
  29#include <linux/interrupt.h>
  30#include <linux/irqflags.h>
 
  31#include <linux/cpu.h>
  32#include <linux/slab.h>
 
 
  33#include <linux/crash_dump.h>
 
 
  34#include <asm/asm-offsets.h>
  35#include <asm/switch_to.h>
 
 
  36#include <asm/facility.h>
 
  37#include <asm/ipl.h>
  38#include <asm/setup.h>
  39#include <asm/irq.h>
  40#include <asm/tlbflush.h>
  41#include <asm/timer.h>
  42#include <asm/lowcore.h>
  43#include <asm/sclp.h>
  44#include <asm/vdso.h>
  45#include <asm/debug.h>
  46#include <asm/os_info.h>
 
 
 
 
 
 
 
  47#include "entry.h"
  48
  49enum {
  50	sigp_sense = 1,
  51	sigp_external_call = 2,
  52	sigp_emergency_signal = 3,
  53	sigp_start = 4,
  54	sigp_stop = 5,
  55	sigp_restart = 6,
  56	sigp_stop_and_store_status = 9,
  57	sigp_initial_cpu_reset = 11,
  58	sigp_cpu_reset = 12,
  59	sigp_set_prefix = 13,
  60	sigp_store_status_at_address = 14,
  61	sigp_store_extended_status_at_address = 15,
  62	sigp_set_architecture = 18,
  63	sigp_conditional_emergency_signal = 19,
  64	sigp_sense_running = 21,
  65};
  66
  67enum {
  68	sigp_order_code_accepted = 0,
  69	sigp_status_stored = 1,
  70	sigp_busy = 2,
  71	sigp_not_operational = 3,
  72};
  73
  74enum {
  75	ec_schedule = 0,
  76	ec_call_function,
  77	ec_call_function_single,
  78	ec_stop_cpu,
 
 
  79};
  80
  81enum {
  82	CPU_STATE_STANDBY,
  83	CPU_STATE_CONFIGURED,
  84};
  85
  86struct pcpu {
  87	struct cpu cpu;
  88	struct _lowcore *lowcore;	/* lowcore page(s) for the cpu */
  89	unsigned long async_stack;	/* async stack for the cpu */
  90	unsigned long panic_stack;	/* panic stack for the cpu */
  91	unsigned long ec_mask;		/* bit mask for ec_xxx functions */
  92	int state;			/* physical cpu state */
  93	u32 status;			/* last status received via sigp */
  94	u16 address;			/* physical cpu address */
  95};
 
 
 
 
 
 
 
 
 
 
 
 
  96
  97static u8 boot_cpu_type;
  98static u16 boot_cpu_address;
  99static struct pcpu pcpu_devices[NR_CPUS];
 
 
 
 100
 
 
 
 
 
 
 
 
 
 
 
 101DEFINE_MUTEX(smp_cpu_state_mutex);
 102
 103/*
 104 * Signal processor helper functions.
 105 */
 106static inline int __pcpu_sigp(u16 addr, u8 order, u32 parm, u32 *status)
 107{
 108	register unsigned int reg1 asm ("1") = parm;
 109	int cc;
 110
 111	asm volatile(
 112		"	sigp	%1,%2,0(%3)\n"
 113		"	ipm	%0\n"
 114		"	srl	%0,28\n"
 115		: "=d" (cc), "+d" (reg1) : "d" (addr), "a" (order) : "cc");
 116	if (status && cc == 1)
 117		*status = reg1;
 118	return cc;
 119}
 120
 121static inline int __pcpu_sigp_relax(u16 addr, u8 order, u32 parm, u32 *status)
 122{
 123	int cc;
 124
 125	while (1) {
 126		cc = __pcpu_sigp(addr, order, parm, status);
 127		if (cc != sigp_busy)
 128			return cc;
 129		cpu_relax();
 130	}
 131}
 132
 133static 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, &pcpu->status);
 139		if (cc != sigp_busy)
 140			break;
 141		if (retry >= 3)
 142			udelay(10);
 143	}
 144	return cc;
 145}
 146
 147static inline int pcpu_stopped(struct pcpu *pcpu)
 148{
 149	if (__pcpu_sigp(pcpu->address, sigp_sense,
 150			0, &pcpu->status) != sigp_status_stored)
 
 
 151		return 0;
 152	/* Check for stopped and check stop state */
 153	return !!(pcpu->status & 0x50);
 154}
 155
 156static inline int pcpu_running(struct pcpu *pcpu)
 157{
 158	if (__pcpu_sigp(pcpu->address, sigp_sense_running,
 159			0, &pcpu->status) != sigp_status_stored)
 160		return 1;
 161	/* Check for running status */
 162	return !(pcpu->status & 0x400);
 163}
 164
 165/*
 166 * Find struct pcpu by cpu address.
 167 */
 168static struct pcpu *pcpu_find_address(const struct cpumask *mask, int address)
 169{
 170	int cpu;
 171
 172	for_each_cpu(cpu, mask)
 173		if (pcpu_devices[cpu].address == address)
 174			return pcpu_devices + cpu;
 175	return NULL;
 176}
 177
 178static void pcpu_ec_call(struct pcpu *pcpu, int ec_bit)
 179{
 180	int order;
 181
 182	set_bit(ec_bit, &pcpu->ec_mask);
 183	order = pcpu_running(pcpu) ?
 184		sigp_external_call : sigp_emergency_signal;
 
 185	pcpu_sigp_retry(pcpu, order, 0);
 186}
 187
 188static int __cpuinit pcpu_alloc_lowcore(struct pcpu *pcpu, int cpu)
 189{
 190	struct _lowcore *lc;
 
 191
 192	if (pcpu != &pcpu_devices[0]) {
 193		pcpu->lowcore =	(struct _lowcore *)
 194			__get_free_pages(GFP_KERNEL | GFP_DMA, LC_ORDER);
 195		pcpu->async_stack = __get_free_pages(GFP_KERNEL, ASYNC_ORDER);
 196		pcpu->panic_stack = __get_free_page(GFP_KERNEL);
 197		if (!pcpu->lowcore || !pcpu->panic_stack || !pcpu->async_stack)
 198			goto out;
 199	}
 200	lc = pcpu->lowcore;
 201	memcpy(lc, &S390_lowcore, 512);
 202	memset((char *) lc + 512, 0, sizeof(*lc) - 512);
 203	lc->async_stack = pcpu->async_stack + ASYNC_SIZE;
 204	lc->panic_stack = pcpu->panic_stack + PAGE_SIZE;
 
 205	lc->cpu_nr = cpu;
 206#ifndef CONFIG_64BIT
 207	if (MACHINE_HAS_IEEE) {
 208		lc->extended_save_area_addr = get_zeroed_page(GFP_KERNEL);
 209		if (!lc->extended_save_area_addr)
 210			goto out;
 211	}
 212#else
 213	if (vdso_alloc_per_cpu(lc))
 214		goto out;
 215#endif
 
 216	lowcore_ptr[cpu] = lc;
 217	pcpu_sigp_retry(pcpu, sigp_set_prefix, (u32)(unsigned long) lc);
 218	return 0;
 
 
 
 219out:
 220	if (pcpu != &pcpu_devices[0]) {
 221		free_page(pcpu->panic_stack);
 222		free_pages(pcpu->async_stack, ASYNC_ORDER);
 223		free_pages((unsigned long) pcpu->lowcore, LC_ORDER);
 224	}
 225	return -ENOMEM;
 226}
 227
 228#ifdef CONFIG_HOTPLUG_CPU
 229
 230static void pcpu_free_lowcore(struct pcpu *pcpu)
 231{
 232	pcpu_sigp_retry(pcpu, sigp_set_prefix, 0);
 233	lowcore_ptr[pcpu - pcpu_devices] = NULL;
 234#ifndef CONFIG_64BIT
 235	if (MACHINE_HAS_IEEE) {
 236		struct _lowcore *lc = pcpu->lowcore;
 237
 238		free_page((unsigned long) lc->extended_save_area_addr);
 239		lc->extended_save_area_addr = 0;
 240	}
 241#else
 242	vdso_free_per_cpu(pcpu->lowcore);
 243#endif
 244	if (pcpu != &pcpu_devices[0]) {
 245		free_page(pcpu->panic_stack);
 246		free_pages(pcpu->async_stack, ASYNC_ORDER);
 247		free_pages((unsigned long) pcpu->lowcore, LC_ORDER);
 248	}
 
 249}
 250
 251#endif /* CONFIG_HOTPLUG_CPU */
 252
 253static void pcpu_prepare_secondary(struct pcpu *pcpu, int cpu)
 254{
 255	struct _lowcore *lc = pcpu->lowcore;
 256
 257	atomic_inc(&init_mm.context.attach_count);
 
 
 258	lc->cpu_nr = cpu;
 
 
 
 
 259	lc->percpu_offset = __per_cpu_offset[cpu];
 260	lc->kernel_asce = S390_lowcore.kernel_asce;
 261	lc->machine_flags = S390_lowcore.machine_flags;
 262	lc->ftrace_func = S390_lowcore.ftrace_func;
 263	lc->user_timer = lc->system_timer = lc->steal_timer = 0;
 264	__ctl_store(lc->cregs_save_area, 0, 15);
 
 
 
 
 
 265	save_access_regs((unsigned int *) lc->access_regs_save_area);
 266	memcpy(lc->stfle_fac_list, S390_lowcore.stfle_fac_list,
 267	       MAX_FACILITY_BIT/8);
 268}
 269
 270static void pcpu_attach_task(struct pcpu *pcpu, struct task_struct *tsk)
 271{
 272	struct _lowcore *lc = pcpu->lowcore;
 273	struct thread_info *ti = task_thread_info(tsk);
 274
 275	lc->kernel_stack = (unsigned long) task_stack_page(tsk) + THREAD_SIZE;
 276	lc->thread_info = (unsigned long) task_thread_info(tsk);
 277	lc->current_task = (unsigned long) tsk;
 278	lc->user_timer = ti->user_timer;
 279	lc->system_timer = ti->system_timer;
 
 
 
 
 
 280	lc->steal_timer = 0;
 281}
 282
 283static void pcpu_start_fn(struct pcpu *pcpu, void (*func)(void *), void *data)
 284{
 285	struct _lowcore *lc = pcpu->lowcore;
 286
 
 287	lc->restart_stack = lc->kernel_stack;
 288	lc->restart_fn = (unsigned long) func;
 289	lc->restart_data = (unsigned long) data;
 290	lc->restart_source = -1UL;
 291	pcpu_sigp_retry(pcpu, sigp_restart, 0);
 292}
 293
 
 
 294/*
 295 * Call function via PSW restart on pcpu and stop the current cpu.
 296 */
 297static void pcpu_delegate(struct pcpu *pcpu, void (*func)(void *),
 
 
 
 
 
 
 298			  void *data, unsigned long stack)
 299{
 300	struct _lowcore *lc = lowcore_ptr[pcpu - pcpu_devices];
 301	struct {
 302		unsigned long	stack;
 303		void		*func;
 304		void		*data;
 305		unsigned long	source;
 306	} restart = { stack, func, data, stap() };
 307
 308	__load_psw_mask(psw_kernel_bits);
 309	if (pcpu->address == restart.source)
 310		func(data);	/* should not return */
 311	/* Stop target cpu (if func returns this stops the current cpu). */
 312	pcpu_sigp_retry(pcpu, sigp_stop, 0);
 
 313	/* Restart func on the target cpu and stop the current cpu. */
 314	memcpy_absolute(&lc->restart_stack, &restart, sizeof(restart));
 
 
 
 
 
 
 
 
 
 
 
 
 315	asm volatile(
 316		"0:	sigp	0,%0,6	# sigp restart to target cpu\n"
 317		"	brc	2,0b	# busy, try again\n"
 318		"1:	sigp	0,%1,5	# sigp stop to current cpu\n"
 319		"	brc	2,1b	# busy, try again\n"
 320		: : "d" (pcpu->address), "d" (restart.source) : "0", "1", "cc");
 
 
 321	for (;;) ;
 322}
 323
 324/*
 325 * Call function on an online CPU.
 326 */
 327void smp_call_online_cpu(void (*func)(void *), void *data)
 328{
 329	struct pcpu *pcpu;
 330
 331	/* Use the current cpu if it is online. */
 332	pcpu = pcpu_find_address(cpu_online_mask, stap());
 333	if (!pcpu)
 334		/* Use the first online cpu. */
 335		pcpu = pcpu_devices + cpumask_first(cpu_online_mask);
 336	pcpu_delegate(pcpu, func, data, (unsigned long) restart_stack);
 
 
 
 
 
 337}
 338
 339/*
 340 * Call function on the ipl CPU.
 341 */
 342void smp_call_ipl_cpu(void (*func)(void *), void *data)
 343{
 344	pcpu_delegate(&pcpu_devices[0], func, data,
 345		      pcpu_devices->panic_stack + PAGE_SIZE);
 
 
 
 
 346}
 347
 348int smp_find_processor_id(u16 address)
 349{
 350	int cpu;
 351
 352	for_each_present_cpu(cpu)
 353		if (pcpu_devices[cpu].address == address)
 354			return cpu;
 355	return -1;
 356}
 357
 358int smp_vcpu_scheduled(int cpu)
 359{
 360	return pcpu_running(pcpu_devices + cpu);
 361}
 362
 363void smp_yield(void)
 364{
 365	if (MACHINE_HAS_DIAG44)
 366		asm volatile("diag 0,0,0x44");
 
 
 
 367}
 
 368
 369void smp_yield_cpu(int cpu)
 370{
 371	if (MACHINE_HAS_DIAG9C)
 372		asm volatile("diag %0,0,0x9c"
 373			     : : "d" (pcpu_devices[cpu].address));
 374	else if (MACHINE_HAS_DIAG44)
 375		asm volatile("diag 0,0,0x44");
 376}
 
 377
 378/*
 379 * Send cpus emergency shutdown signal. This gives the cpus the
 380 * opportunity to complete outstanding interrupts.
 381 */
 382void smp_emergency_stop(cpumask_t *cpumask)
 383{
 
 
 384	u64 end;
 385	int cpu;
 386
 387	end = get_clock() + (1000000UL << 12);
 388	for_each_cpu(cpu, cpumask) {
 389		struct pcpu *pcpu = pcpu_devices + cpu;
 
 
 
 
 390		set_bit(ec_stop_cpu, &pcpu->ec_mask);
 391		while (__pcpu_sigp(pcpu->address, sigp_emergency_signal,
 392				   0, NULL) == sigp_busy &&
 393		       get_clock() < end)
 394			cpu_relax();
 395	}
 396	while (get_clock() < end) {
 397		for_each_cpu(cpu, cpumask)
 398			if (pcpu_stopped(pcpu_devices + cpu))
 399				cpumask_clear_cpu(cpu, cpumask);
 400		if (cpumask_empty(cpumask))
 401			break;
 402		cpu_relax();
 403	}
 
 404}
 
 405
 406/*
 407 * Stop all cpus but the current one.
 408 */
 409void smp_send_stop(void)
 410{
 411	cpumask_t cpumask;
 412	int cpu;
 413
 414	/* Disable all interrupts/machine checks */
 415	__load_psw_mask(psw_kernel_bits | PSW_MASK_DAT);
 416	trace_hardirqs_off();
 417
 418	debug_set_critical();
 419	cpumask_copy(&cpumask, cpu_online_mask);
 420	cpumask_clear_cpu(smp_processor_id(), &cpumask);
 421
 422	if (oops_in_progress)
 423		smp_emergency_stop(&cpumask);
 424
 425	/* stop all processors */
 426	for_each_cpu(cpu, &cpumask) {
 427		struct pcpu *pcpu = pcpu_devices + cpu;
 428		pcpu_sigp_retry(pcpu, sigp_stop, 0);
 
 
 429		while (!pcpu_stopped(pcpu))
 430			cpu_relax();
 431	}
 432}
 433
 434/*
 435 * Stop the current cpu.
 436 */
 437void smp_stop_cpu(void)
 438{
 439	pcpu_sigp_retry(pcpu_devices + smp_processor_id(), sigp_stop, 0);
 440	for (;;) ;
 441}
 442
 443/*
 444 * This is the main routine where commands issued by other
 445 * cpus are handled.
 446 */
 447static void do_ext_call_interrupt(struct ext_code ext_code,
 448				  unsigned int param32, unsigned long param64)
 449{
 450	unsigned long bits;
 451	int cpu;
 452
 453	cpu = smp_processor_id();
 454	if (ext_code.code == 0x1202)
 455		kstat_cpu(cpu).irqs[EXTINT_EXC]++;
 456	else
 457		kstat_cpu(cpu).irqs[EXTINT_EMS]++;
 458	/*
 459	 * handle bit signal external calls
 460	 */
 461	bits = xchg(&pcpu_devices[cpu].ec_mask, 0);
 462
 
 
 463	if (test_bit(ec_stop_cpu, &bits))
 464		smp_stop_cpu();
 465
 466	if (test_bit(ec_schedule, &bits))
 467		scheduler_ipi();
 468
 469	if (test_bit(ec_call_function, &bits))
 470		generic_smp_call_function_interrupt();
 471
 472	if (test_bit(ec_call_function_single, &bits))
 473		generic_smp_call_function_single_interrupt();
 
 
 
 
 
 474
 
 
 
 
 
 475}
 476
 477void arch_send_call_function_ipi_mask(const struct cpumask *mask)
 478{
 479	int cpu;
 480
 481	for_each_cpu(cpu, mask)
 482		pcpu_ec_call(pcpu_devices + cpu, ec_call_function);
 483}
 484
 485void arch_send_call_function_single_ipi(int cpu)
 486{
 487	pcpu_ec_call(pcpu_devices + cpu, ec_call_function_single);
 488}
 489
 490#ifndef CONFIG_64BIT
 491/*
 492 * this function sends a 'purge tlb' signal to another CPU.
 493 */
 494static void smp_ptlb_callback(void *info)
 495{
 496	__tlb_flush_local();
 497}
 498
 499void smp_ptlb_all(void)
 500{
 501	on_each_cpu(smp_ptlb_callback, NULL, 1);
 502}
 503EXPORT_SYMBOL(smp_ptlb_all);
 504#endif /* ! CONFIG_64BIT */
 505
 506/*
 507 * this function sends a 'reschedule' IPI to another CPU.
 508 * it goes straight through and wastes no time serializing
 509 * anything. Worst case is that we lose a reschedule ...
 510 */
 511void smp_send_reschedule(int cpu)
 512{
 513	pcpu_ec_call(pcpu_devices + cpu, ec_schedule);
 514}
 515
 516/*
 517 * parameter area for the set/clear control bit callbacks
 518 */
 519struct ec_creg_mask_parms {
 520	unsigned long orval;
 521	unsigned long andval;
 522	int cr;
 523};
 524
 525/*
 526 * callback for setting/clearing control bits
 527 */
 528static void smp_ctl_bit_callback(void *info)
 529{
 530	struct ec_creg_mask_parms *pp = info;
 531	unsigned long cregs[16];
 
 532
 533	__ctl_store(cregs, 0, 15);
 534	cregs[pp->cr] = (cregs[pp->cr] & pp->andval) | pp->orval;
 535	__ctl_load(cregs, 0, 15);
 536}
 537
 538/*
 539 * Set a bit in a control register of all cpus
 540 */
 541void smp_ctl_set_bit(int cr, int bit)
 542{
 543	struct ec_creg_mask_parms parms = { 1UL << bit, -1UL, cr };
 
 
 544
 545	on_each_cpu(smp_ctl_bit_callback, &parms, 1);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 546}
 547EXPORT_SYMBOL(smp_ctl_set_bit);
 548
 549/*
 550 * Clear a bit in a control register of all cpus
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 551 */
 552void smp_ctl_clear_bit(int cr, int bit)
 553{
 554	struct ec_creg_mask_parms parms = { 0, ~(1UL << bit), cr };
 555
 556	on_each_cpu(smp_ctl_bit_callback, &parms, 1);
 557}
 558EXPORT_SYMBOL(smp_ctl_clear_bit);
 559
 560#if defined(CONFIG_ZFCPDUMP) || defined(CONFIG_CRASH_DUMP)
 561
 562struct save_area *zfcpdump_save_areas[NR_CPUS + 1];
 563EXPORT_SYMBOL_GPL(zfcpdump_save_areas);
 564
 565static void __init smp_get_save_area(int cpu, u16 address)
 566{
 567	void *lc = pcpu_devices[0].lowcore;
 568	struct save_area *save_area;
 569
 570	if (is_kdump_kernel())
 571		return;
 572	if (!OLDMEM_BASE && (address == boot_cpu_address ||
 573			     ipl_info.type != IPL_TYPE_FCP_DUMP))
 574		return;
 575	if (cpu >= NR_CPUS) {
 576		pr_warning("CPU %i exceeds the maximum %i and is excluded "
 577			   "from the dump\n", cpu, NR_CPUS - 1);
 
 
 
 
 
 
 
 
 
 
 
 
 578		return;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 579	}
 580	save_area = kmalloc(sizeof(struct save_area), GFP_KERNEL);
 581	if (!save_area)
 582		panic("could not allocate memory for save area\n");
 583	zfcpdump_save_areas[cpu] = save_area;
 584#ifdef CONFIG_CRASH_DUMP
 585	if (address == boot_cpu_address) {
 586		/* Copy the registers of the boot cpu. */
 587		copy_oldmem_page(1, (void *) save_area, sizeof(*save_area),
 588				 SAVE_AREA_BASE - PAGE_SIZE, 0);
 589		return;
 590	}
 591#endif
 592	/* Get the registers of a non-boot cpu. */
 593	__pcpu_sigp_relax(address, sigp_stop_and_store_status, 0, NULL);
 594	memcpy_real(save_area, lc + SAVE_AREA_BASE, sizeof(*save_area));
 595}
 596
 597int smp_store_status(int cpu)
 598{
 599	struct pcpu *pcpu;
 
 600
 601	pcpu = pcpu_devices + cpu;
 602	if (__pcpu_sigp_relax(pcpu->address, sigp_stop_and_store_status,
 603			      0, NULL) != sigp_order_code_accepted)
 604		return -EIO;
 605	return 0;
 606}
 607
 608#else /* CONFIG_ZFCPDUMP || CONFIG_CRASH_DUMP */
 
 
 609
 610static inline void smp_get_save_area(int cpu, u16 address) { }
 
 
 
 611
 612#endif /* CONFIG_ZFCPDUMP || CONFIG_CRASH_DUMP */
 
 
 
 613
 614static struct sclp_cpu_info *smp_get_cpu_info(void)
 615{
 616	static int use_sigp_detection;
 617	struct sclp_cpu_info *info;
 618	int address;
 619
 620	info = kzalloc(sizeof(*info), GFP_KERNEL);
 621	if (info && (use_sigp_detection || sclp_get_cpu_info(info))) {
 622		use_sigp_detection = 1;
 623		for (address = 0; address <= MAX_CPU_ADDRESS; address++) {
 624			if (__pcpu_sigp_relax(address, sigp_sense, 0, NULL) ==
 625			    sigp_not_operational)
 
 
 626				continue;
 627			info->cpu[info->configured].address = address;
 
 628			info->configured++;
 629		}
 630		info->combined = info->configured;
 631	}
 632	return info;
 633}
 634
 635static int __devinit smp_add_present_cpu(int cpu);
 636
 637static int __devinit __smp_rescan_cpus(struct sclp_cpu_info *info,
 638				       int sysfs_add)
 639{
 640	struct pcpu *pcpu;
 641	cpumask_t avail;
 642	int cpu, nr, i;
 
 643
 644	nr = 0;
 645	cpumask_xor(&avail, cpu_possible_mask, cpu_present_mask);
 646	cpu = cpumask_first(&avail);
 647	for (i = 0; (i < info->combined) && (cpu < nr_cpu_ids); i++) {
 648		if (info->has_cpu_type && info->cpu[i].type != boot_cpu_type)
 
 
 649			continue;
 650		if (pcpu_find_address(cpu_present_mask, info->cpu[i].address))
 651			continue;
 652		pcpu = pcpu_devices + cpu;
 653		pcpu->address = info->cpu[i].address;
 654		pcpu->state = (cpu >= info->configured) ?
 655			CPU_STATE_STANDBY : CPU_STATE_CONFIGURED;
 656		cpu_set_polarization(cpu, POLARIZATION_UNKNOWN);
 
 657		set_cpu_present(cpu, true);
 658		if (sysfs_add && smp_add_present_cpu(cpu) != 0)
 659			set_cpu_present(cpu, false);
 660		else
 661			nr++;
 662		cpu = cpumask_next(cpu, &avail);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 663	}
 
 
 664	return nr;
 665}
 666
 667static void __init smp_detect_cpus(void)
 668{
 669	unsigned int cpu, c_cpus, s_cpus;
 670	struct sclp_cpu_info *info;
 
 671
 672	info = smp_get_cpu_info();
 
 673	if (!info)
 674		panic("smp_detect_cpus failed to allocate memory\n");
 675	if (info->has_cpu_type) {
 676		for (cpu = 0; cpu < info->combined; cpu++) {
 677			if (info->cpu[cpu].address != boot_cpu_address)
 678				continue;
 679			/* The boot cpu dictates the cpu type. */
 680			boot_cpu_type = info->cpu[cpu].type;
 681			break;
 682		}
 
 
 
 
 
 683	}
 
 
 
 
 
 
 
 684	c_cpus = s_cpus = 0;
 685	for (cpu = 0; cpu < info->combined; cpu++) {
 686		if (info->has_cpu_type && info->cpu[cpu].type != boot_cpu_type)
 
 687			continue;
 688		if (cpu < info->configured) {
 689			smp_get_save_area(c_cpus, info->cpu[cpu].address);
 690			c_cpus++;
 691		} else
 692			s_cpus++;
 693	}
 694	pr_info("%d configured CPUs, %d standby CPUs\n", c_cpus, s_cpus);
 695	get_online_cpus();
 696	__smp_rescan_cpus(info, 0);
 697	put_online_cpus();
 698	kfree(info);
 699}
 700
 701/*
 702 *	Activate a secondary processor.
 703 */
 704static void __cpuinit smp_start_secondary(void *cpuvoid)
 705{
 706	S390_lowcore.last_update_clock = get_clock();
 707	S390_lowcore.restart_stack = (unsigned long) restart_stack;
 708	S390_lowcore.restart_fn = (unsigned long) do_restart;
 709	S390_lowcore.restart_data = 0;
 710	S390_lowcore.restart_source = -1UL;
 711	restore_access_regs(S390_lowcore.access_regs_save_area);
 712	__ctl_load(S390_lowcore.cregs_save_area, 0, 15);
 713	__load_psw_mask(psw_kernel_bits | PSW_MASK_DAT);
 
 
 714	cpu_init();
 715	preempt_disable();
 716	init_cpu_timer();
 717	init_cpu_vtimer();
 
 718	pfault_init();
 719	notify_cpu_starting(smp_processor_id());
 720	ipi_call_lock();
 721	set_cpu_online(smp_processor_id(), true);
 722	ipi_call_unlock();
 
 
 
 
 
 723	local_irq_enable();
 724	/* cpu_idle will call schedule for us */
 725	cpu_idle();
 726}
 727
 728/* Upping and downing of CPUs */
 729int __cpuinit __cpu_up(unsigned int cpu, struct task_struct *tidle)
 730{
 731	struct pcpu *pcpu;
 732	int rc;
 733
 734	pcpu = pcpu_devices + cpu;
 735	if (pcpu->state != CPU_STATE_CONFIGURED)
 736		return -EIO;
 737	if (pcpu_sigp_retry(pcpu, sigp_initial_cpu_reset, 0) !=
 738	    sigp_order_code_accepted)
 739		return -EIO;
 740
 741	rc = pcpu_alloc_lowcore(pcpu, cpu);
 742	if (rc)
 743		return rc;
 
 
 
 
 
 744	pcpu_prepare_secondary(pcpu, cpu);
 745	pcpu_attach_task(pcpu, tidle);
 746	pcpu_start_fn(pcpu, smp_start_secondary, NULL);
 
 747	while (!cpu_online(cpu))
 748		cpu_relax();
 
 749	return 0;
 750}
 751
 752static int __init setup_possible_cpus(char *s)
 
 
 753{
 754	int max, cpu;
 755
 756	if (kstrtoint(s, 0, &max) < 0)
 757		return 0;
 758	init_cpu_possible(cpumask_of(0));
 759	for (cpu = 1; cpu < max && cpu < nr_cpu_ids; cpu++)
 760		set_cpu_possible(cpu, true);
 761	return 0;
 762}
 763early_param("possible_cpus", setup_possible_cpus);
 764
 765#ifdef CONFIG_HOTPLUG_CPU
 766
 767int __cpu_disable(void)
 768{
 769	unsigned long cregs[16];
 
 770
 771	set_cpu_online(smp_processor_id(), false);
 
 
 
 
 
 772	/* Disable pseudo page faults on this cpu. */
 773	pfault_fini();
 774	/* Disable interrupt sources via control register. */
 775	__ctl_store(cregs, 0, 15);
 776	cregs[0]  &= ~0x0000ee70UL;	/* disable all external interrupts */
 777	cregs[6]  &= ~0xff000000UL;	/* disable all I/O interrupts */
 778	cregs[14] &= ~0x1f000000UL;	/* disable most machine checks */
 779	__ctl_load(cregs, 0, 15);
 
 780	return 0;
 781}
 782
 783void __cpu_die(unsigned int cpu)
 784{
 785	struct pcpu *pcpu;
 786
 787	/* Wait until target cpu is down */
 788	pcpu = pcpu_devices + cpu;
 789	while (!pcpu_stopped(pcpu))
 790		cpu_relax();
 791	pcpu_free_lowcore(pcpu);
 792	atomic_dec(&init_mm.context.attach_count);
 
 
 793}
 794
 795void __noreturn cpu_die(void)
 796{
 797	idle_task_exit();
 798	pcpu_sigp_retry(pcpu_devices + smp_processor_id(), sigp_stop, 0);
 799	for (;;) ;
 800}
 801
 802#endif /* CONFIG_HOTPLUG_CPU */
 
 
 
 
 
 
 
 
 
 
 
 803
 804void __init smp_prepare_cpus(unsigned int max_cpus)
 805{
 806	/* request the 0x1201 emergency signal external interrupt */
 807	if (register_external_interrupt(0x1201, do_ext_call_interrupt) != 0)
 808		panic("Couldn't request external interrupt 0x1201");
 809	/* request the 0x1202 external call external interrupt */
 810	if (register_external_interrupt(0x1202, do_ext_call_interrupt) != 0)
 811		panic("Couldn't request external interrupt 0x1202");
 812	smp_detect_cpus();
 
 813}
 814
 815void __init smp_prepare_boot_cpu(void)
 816{
 817	struct pcpu *pcpu = pcpu_devices;
 818
 819	boot_cpu_address = stap();
 820	pcpu->state = CPU_STATE_CONFIGURED;
 821	pcpu->address = boot_cpu_address;
 822	pcpu->lowcore = (struct _lowcore *)(unsigned long) store_prefix();
 823	pcpu->async_stack = S390_lowcore.async_stack - ASYNC_SIZE;
 824	pcpu->panic_stack = S390_lowcore.panic_stack - PAGE_SIZE;
 825	S390_lowcore.percpu_offset = __per_cpu_offset[0];
 826	cpu_set_polarization(0, POLARIZATION_UNKNOWN);
 827	set_cpu_present(0, true);
 828	set_cpu_online(0, true);
 829}
 830
 831void __init smp_cpus_done(unsigned int max_cpus)
 832{
 833}
 834
 835void __init smp_setup_processor_id(void)
 836{
 837	S390_lowcore.cpu_nr = 0;
 
 838}
 839
 840/*
 841 * the frequency of the profiling timer can be changed
 842 * by writing a multiplier value into /proc/profile.
 843 *
 844 * usually you want to run this on all CPUs ;)
 845 */
 846int setup_profiling_timer(unsigned int multiplier)
 847{
 848	return 0;
 849}
 850
 851#ifdef CONFIG_HOTPLUG_CPU
 852static ssize_t cpu_configure_show(struct device *dev,
 853				  struct device_attribute *attr, char *buf)
 854{
 855	ssize_t count;
 856
 857	mutex_lock(&smp_cpu_state_mutex);
 858	count = sprintf(buf, "%d\n", pcpu_devices[dev->id].state);
 859	mutex_unlock(&smp_cpu_state_mutex);
 860	return count;
 861}
 862
 863static ssize_t cpu_configure_store(struct device *dev,
 864				   struct device_attribute *attr,
 865				   const char *buf, size_t count)
 866{
 867	struct pcpu *pcpu;
 868	int cpu, val, rc;
 869	char delim;
 870
 871	if (sscanf(buf, "%d %c", &val, &delim) != 1)
 872		return -EINVAL;
 873	if (val != 0 && val != 1)
 874		return -EINVAL;
 875	get_online_cpus();
 876	mutex_lock(&smp_cpu_state_mutex);
 877	rc = -EBUSY;
 878	/* disallow configuration changes of online cpus and cpu 0 */
 879	cpu = dev->id;
 880	if (cpu_online(cpu) || cpu == 0)
 881		goto out;
 882	pcpu = pcpu_devices + cpu;
 
 
 883	rc = 0;
 884	switch (val) {
 885	case 0:
 886		if (pcpu->state != CPU_STATE_CONFIGURED)
 887			break;
 888		rc = sclp_cpu_deconfigure(pcpu->address);
 889		if (rc)
 890			break;
 891		pcpu->state = CPU_STATE_STANDBY;
 892		cpu_set_polarization(cpu, POLARIZATION_UNKNOWN);
 
 
 
 
 
 893		topology_expect_change();
 894		break;
 895	case 1:
 896		if (pcpu->state != CPU_STATE_STANDBY)
 897			break;
 898		rc = sclp_cpu_configure(pcpu->address);
 899		if (rc)
 900			break;
 901		pcpu->state = CPU_STATE_CONFIGURED;
 902		cpu_set_polarization(cpu, POLARIZATION_UNKNOWN);
 
 
 
 
 
 903		topology_expect_change();
 904		break;
 905	default:
 906		break;
 907	}
 908out:
 909	mutex_unlock(&smp_cpu_state_mutex);
 910	put_online_cpus();
 911	return rc ? rc : count;
 912}
 913static DEVICE_ATTR(configure, 0644, cpu_configure_show, cpu_configure_store);
 914#endif /* CONFIG_HOTPLUG_CPU */
 915
 916static ssize_t show_cpu_address(struct device *dev,
 917				struct device_attribute *attr, char *buf)
 918{
 919	return sprintf(buf, "%d\n", pcpu_devices[dev->id].address);
 920}
 921static DEVICE_ATTR(address, 0444, show_cpu_address, NULL);
 922
 923static struct attribute *cpu_common_attrs[] = {
 924#ifdef CONFIG_HOTPLUG_CPU
 925	&dev_attr_configure.attr,
 926#endif
 927	&dev_attr_address.attr,
 928	NULL,
 929};
 930
 931static struct attribute_group cpu_common_attr_group = {
 932	.attrs = cpu_common_attrs,
 933};
 934
 935static ssize_t show_idle_count(struct device *dev,
 936				struct device_attribute *attr, char *buf)
 937{
 938	struct s390_idle_data *idle = &per_cpu(s390_idle, dev->id);
 939	unsigned long long idle_count;
 940	unsigned int sequence;
 941
 942	do {
 943		sequence = ACCESS_ONCE(idle->sequence);
 944		idle_count = ACCESS_ONCE(idle->idle_count);
 945		if (ACCESS_ONCE(idle->idle_enter))
 946			idle_count++;
 947	} while ((sequence & 1) || (idle->sequence != sequence));
 948	return sprintf(buf, "%llu\n", idle_count);
 949}
 950static DEVICE_ATTR(idle_count, 0444, show_idle_count, NULL);
 951
 952static ssize_t show_idle_time(struct device *dev,
 953				struct device_attribute *attr, char *buf)
 954{
 955	struct s390_idle_data *idle = &per_cpu(s390_idle, dev->id);
 956	unsigned long long now, idle_time, idle_enter, idle_exit;
 957	unsigned int sequence;
 958
 959	do {
 960		now = get_clock();
 961		sequence = ACCESS_ONCE(idle->sequence);
 962		idle_time = ACCESS_ONCE(idle->idle_time);
 963		idle_enter = ACCESS_ONCE(idle->idle_enter);
 964		idle_exit = ACCESS_ONCE(idle->idle_exit);
 965	} while ((sequence & 1) || (idle->sequence != sequence));
 966	idle_time += idle_enter ? ((idle_exit ? : now) - idle_enter) : 0;
 967	return sprintf(buf, "%llu\n", idle_time >> 12);
 968}
 969static DEVICE_ATTR(idle_time_us, 0444, show_idle_time, NULL);
 970
 971static struct attribute *cpu_online_attrs[] = {
 972	&dev_attr_idle_count.attr,
 973	&dev_attr_idle_time_us.attr,
 974	NULL,
 975};
 976
 977static struct attribute_group cpu_online_attr_group = {
 978	.attrs = cpu_online_attrs,
 979};
 980
 981static int __cpuinit smp_cpu_notify(struct notifier_block *self,
 982				    unsigned long action, void *hcpu)
 
 
 
 
 
 
 983{
 984	unsigned int cpu = (unsigned int)(long)hcpu;
 985	struct cpu *c = &pcpu_devices[cpu].cpu;
 986	struct device *s = &c->dev;
 987	int err = 0;
 988
 989	switch (action) {
 990	case CPU_ONLINE:
 991	case CPU_ONLINE_FROZEN:
 992		err = sysfs_create_group(&s->kobj, &cpu_online_attr_group);
 993		break;
 994	case CPU_DEAD:
 995	case CPU_DEAD_FROZEN:
 996		sysfs_remove_group(&s->kobj, &cpu_online_attr_group);
 997		break;
 998	}
 999	return notifier_from_errno(err);
1000}
1001
1002static struct notifier_block __cpuinitdata smp_cpu_nb = {
1003	.notifier_call = smp_cpu_notify,
1004};
 
1005
1006static int __devinit smp_add_present_cpu(int cpu)
1007{
1008	struct cpu *c = &pcpu_devices[cpu].cpu;
1009	struct device *s = &c->dev;
1010	int rc;
1011
1012	c->hotpluggable = 1;
1013	rc = register_cpu(c, cpu);
1014	if (rc)
1015		goto out;
1016	rc = sysfs_create_group(&s->kobj, &cpu_common_attr_group);
1017	if (rc)
1018		goto out_cpu;
1019	if (cpu_online(cpu)) {
1020		rc = sysfs_create_group(&s->kobj, &cpu_online_attr_group);
1021		if (rc)
1022			goto out_online;
1023	}
1024	rc = topology_cpu_init(c);
1025	if (rc)
1026		goto out_topology;
1027	return 0;
1028
1029out_topology:
1030	if (cpu_online(cpu))
1031		sysfs_remove_group(&s->kobj, &cpu_online_attr_group);
1032out_online:
1033	sysfs_remove_group(&s->kobj, &cpu_common_attr_group);
1034out_cpu:
1035#ifdef CONFIG_HOTPLUG_CPU
1036	unregister_cpu(c);
1037#endif
1038out:
1039	return rc;
1040}
1041
1042#ifdef CONFIG_HOTPLUG_CPU
1043
1044int __ref smp_rescan_cpus(void)
1045{
1046	struct sclp_cpu_info *info;
1047	int nr;
1048
1049	info = smp_get_cpu_info();
1050	if (!info)
1051		return -ENOMEM;
1052	get_online_cpus();
1053	mutex_lock(&smp_cpu_state_mutex);
1054	nr = __smp_rescan_cpus(info, 1);
1055	mutex_unlock(&smp_cpu_state_mutex);
1056	put_online_cpus();
1057	kfree(info);
1058	if (nr)
1059		topology_schedule_update();
1060	return 0;
1061}
1062
1063static ssize_t __ref rescan_store(struct device *dev,
1064				  struct device_attribute *attr,
1065				  const char *buf,
1066				  size_t count)
1067{
1068	int rc;
1069
1070	rc = smp_rescan_cpus();
 
 
 
 
1071	return rc ? rc : count;
1072}
1073static DEVICE_ATTR(rescan, 0200, NULL, rescan_store);
1074#endif /* CONFIG_HOTPLUG_CPU */
1075
1076static int __init s390_smp_init(void)
1077{
1078	int cpu, rc;
 
1079
1080	register_cpu_notifier(&smp_cpu_nb);
1081#ifdef CONFIG_HOTPLUG_CPU
1082	rc = device_create_file(cpu_subsys.dev_root, &dev_attr_rescan);
1083	if (rc)
1084		return rc;
1085#endif
1086	for_each_present_cpu(cpu) {
1087		rc = smp_add_present_cpu(cpu);
1088		if (rc)
1089			return rc;
1090	}
1091	return 0;
 
 
 
1092}
1093subsys_initcall(s390_smp_init);