1 // SPDX-License-Identifier: GPL-2.0
3 * SMP related functions
5 * Copyright IBM Corp. 1999, 2012
6 * Author(s): Denis Joseph Barrow,
7 * Martin Schwidefsky <schwidefsky@de.ibm.com>,
8 * Heiko Carstens <heiko.carstens@de.ibm.com>,
10 * based on other smp stuff by
11 * (c) 1995 Alan Cox, CymruNET Ltd <alan@cymru.net>
12 * (c) 1998 Ingo Molnar
14 * The code outside of smp.c uses logical cpu numbers, only smp.c does
15 * the translation of logical to physical cpu ids. All new code that
16 * operates on physical cpu numbers needs to go into smp.c.
19 #define KMSG_COMPONENT "cpu"
20 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
22 #include <linux/workqueue.h>
23 #include <linux/bootmem.h>
24 #include <linux/export.h>
25 #include <linux/init.h>
27 #include <linux/err.h>
28 #include <linux/spinlock.h>
29 #include <linux/kernel_stat.h>
30 #include <linux/kmemleak.h>
31 #include <linux/delay.h>
32 #include <linux/interrupt.h>
33 #include <linux/irqflags.h>
34 #include <linux/cpu.h>
35 #include <linux/slab.h>
36 #include <linux/sched/hotplug.h>
37 #include <linux/sched/task_stack.h>
38 #include <linux/crash_dump.h>
39 #include <linux/memblock.h>
40 #include <linux/kprobes.h>
41 #include <asm/asm-offsets.h>
43 #include <asm/switch_to.h>
44 #include <asm/facility.h>
46 #include <asm/setup.h>
48 #include <asm/tlbflush.h>
49 #include <asm/vtimer.h>
50 #include <asm/lowcore.h>
53 #include <asm/debug.h>
54 #include <asm/os_info.h>
62 ec_call_function_single,
71 static DEFINE_PER_CPU(struct cpu *, cpu_device);
74 struct lowcore *lowcore; /* lowcore page(s) for the cpu */
75 unsigned long ec_mask; /* bit mask for ec_xxx functions */
76 unsigned long ec_clk; /* sigp timestamp for ec_xxx */
77 signed char state; /* physical cpu state */
78 signed char polarization; /* physical polarization */
79 u16 address; /* physical cpu address */
82 static u8 boot_core_type;
83 static struct pcpu pcpu_devices[NR_CPUS];
85 unsigned int smp_cpu_mt_shift;
86 EXPORT_SYMBOL(smp_cpu_mt_shift);
88 unsigned int smp_cpu_mtid;
89 EXPORT_SYMBOL(smp_cpu_mtid);
91 #ifdef CONFIG_CRASH_DUMP
92 __vector128 __initdata boot_cpu_vector_save_area[__NUM_VXRS];
95 static unsigned int smp_max_threads __initdata = -1U;
97 static int __init early_nosmt(char *s)
102 early_param("nosmt", early_nosmt);
104 static int __init early_smt(char *s)
106 get_option(&s, &smp_max_threads);
109 early_param("smt", early_smt);
112 * The smp_cpu_state_mutex must be held when changing the state or polarization
113 * member of a pcpu data structure within the pcpu_devices arreay.
115 DEFINE_MUTEX(smp_cpu_state_mutex);
118 * Signal processor helper functions.
120 static inline int __pcpu_sigp_relax(u16 addr, u8 order, unsigned long parm)
125 cc = __pcpu_sigp(addr, order, parm, NULL);
126 if (cc != SIGP_CC_BUSY)
132 static int pcpu_sigp_retry(struct pcpu *pcpu, u8 order, u32 parm)
136 for (retry = 0; ; retry++) {
137 cc = __pcpu_sigp(pcpu->address, order, parm, NULL);
138 if (cc != SIGP_CC_BUSY)
146 static inline int pcpu_stopped(struct pcpu *pcpu)
148 u32 uninitialized_var(status);
150 if (__pcpu_sigp(pcpu->address, SIGP_SENSE,
151 0, &status) != SIGP_CC_STATUS_STORED)
153 return !!(status & (SIGP_STATUS_CHECK_STOP|SIGP_STATUS_STOPPED));
156 static inline int pcpu_running(struct pcpu *pcpu)
158 if (__pcpu_sigp(pcpu->address, SIGP_SENSE_RUNNING,
159 0, NULL) != SIGP_CC_STATUS_STORED)
161 /* Status stored condition code is equivalent to cpu not running. */
166 * Find struct pcpu by cpu address.
168 static struct pcpu *pcpu_find_address(const struct cpumask *mask, u16 address)
172 for_each_cpu(cpu, mask)
173 if (pcpu_devices[cpu].address == address)
174 return pcpu_devices + cpu;
178 static void pcpu_ec_call(struct pcpu *pcpu, int ec_bit)
182 if (test_and_set_bit(ec_bit, &pcpu->ec_mask))
184 order = pcpu_running(pcpu) ? SIGP_EXTERNAL_CALL : SIGP_EMERGENCY_SIGNAL;
185 pcpu->ec_clk = get_tod_clock_fast();
186 pcpu_sigp_retry(pcpu, order, 0);
189 #define ASYNC_FRAME_OFFSET (ASYNC_SIZE - STACK_FRAME_OVERHEAD - __PT_SIZE)
190 #define PANIC_FRAME_OFFSET (PAGE_SIZE - STACK_FRAME_OVERHEAD - __PT_SIZE)
192 static int pcpu_alloc_lowcore(struct pcpu *pcpu, int cpu)
194 unsigned long async_stack, panic_stack;
197 if (pcpu != &pcpu_devices[0]) {
198 pcpu->lowcore = (struct lowcore *)
199 __get_free_pages(GFP_KERNEL | GFP_DMA, LC_ORDER);
200 async_stack = __get_free_pages(GFP_KERNEL, ASYNC_ORDER);
201 panic_stack = __get_free_page(GFP_KERNEL);
202 if (!pcpu->lowcore || !panic_stack || !async_stack)
205 async_stack = pcpu->lowcore->async_stack - ASYNC_FRAME_OFFSET;
206 panic_stack = pcpu->lowcore->panic_stack - PANIC_FRAME_OFFSET;
209 memcpy(lc, &S390_lowcore, 512);
210 memset((char *) lc + 512, 0, sizeof(*lc) - 512);
211 lc->async_stack = async_stack + ASYNC_FRAME_OFFSET;
212 lc->panic_stack = panic_stack + PANIC_FRAME_OFFSET;
214 lc->spinlock_lockval = arch_spin_lockval(cpu);
215 lc->spinlock_index = 0;
216 if (nmi_alloc_per_cpu(lc))
218 if (vdso_alloc_per_cpu(lc))
220 lowcore_ptr[cpu] = lc;
221 pcpu_sigp_retry(pcpu, SIGP_SET_PREFIX, (u32)(unsigned long) lc);
225 nmi_free_per_cpu(lc);
227 if (pcpu != &pcpu_devices[0]) {
228 free_page(panic_stack);
229 free_pages(async_stack, ASYNC_ORDER);
230 free_pages((unsigned long) pcpu->lowcore, LC_ORDER);
235 #ifdef CONFIG_HOTPLUG_CPU
237 static void pcpu_free_lowcore(struct pcpu *pcpu)
239 pcpu_sigp_retry(pcpu, SIGP_SET_PREFIX, 0);
240 lowcore_ptr[pcpu - pcpu_devices] = NULL;
241 vdso_free_per_cpu(pcpu->lowcore);
242 nmi_free_per_cpu(pcpu->lowcore);
243 if (pcpu == &pcpu_devices[0])
245 free_page(pcpu->lowcore->panic_stack-PANIC_FRAME_OFFSET);
246 free_pages(pcpu->lowcore->async_stack-ASYNC_FRAME_OFFSET, ASYNC_ORDER);
247 free_pages((unsigned long) pcpu->lowcore, LC_ORDER);
250 #endif /* CONFIG_HOTPLUG_CPU */
252 static void pcpu_prepare_secondary(struct pcpu *pcpu, int cpu)
254 struct lowcore *lc = pcpu->lowcore;
256 cpumask_set_cpu(cpu, &init_mm.context.cpu_attach_mask);
257 cpumask_set_cpu(cpu, mm_cpumask(&init_mm));
259 lc->spinlock_lockval = arch_spin_lockval(cpu);
260 lc->spinlock_index = 0;
261 lc->percpu_offset = __per_cpu_offset[cpu];
262 lc->kernel_asce = S390_lowcore.kernel_asce;
263 lc->machine_flags = S390_lowcore.machine_flags;
264 lc->user_timer = lc->system_timer = lc->steal_timer = 0;
265 __ctl_store(lc->cregs_save_area, 0, 15);
266 save_access_regs((unsigned int *) lc->access_regs_save_area);
267 memcpy(lc->stfle_fac_list, S390_lowcore.stfle_fac_list,
269 arch_spin_lock_setup(cpu);
272 static void pcpu_attach_task(struct pcpu *pcpu, struct task_struct *tsk)
274 struct lowcore *lc = pcpu->lowcore;
276 lc->kernel_stack = (unsigned long) task_stack_page(tsk)
277 + THREAD_SIZE - STACK_FRAME_OVERHEAD - sizeof(struct pt_regs);
278 lc->current_task = (unsigned long) tsk;
280 lc->current_pid = tsk->pid;
281 lc->user_timer = tsk->thread.user_timer;
282 lc->guest_timer = tsk->thread.guest_timer;
283 lc->system_timer = tsk->thread.system_timer;
284 lc->hardirq_timer = tsk->thread.hardirq_timer;
285 lc->softirq_timer = tsk->thread.softirq_timer;
289 static void pcpu_start_fn(struct pcpu *pcpu, void (*func)(void *), void *data)
291 struct lowcore *lc = pcpu->lowcore;
293 lc->restart_stack = lc->kernel_stack;
294 lc->restart_fn = (unsigned long) func;
295 lc->restart_data = (unsigned long) data;
296 lc->restart_source = -1UL;
297 pcpu_sigp_retry(pcpu, SIGP_RESTART, 0);
301 * Call function via PSW restart on pcpu and stop the current cpu.
303 static void pcpu_delegate(struct pcpu *pcpu, void (*func)(void *),
304 void *data, unsigned long stack)
306 struct lowcore *lc = lowcore_ptr[pcpu - pcpu_devices];
307 unsigned long source_cpu = stap();
309 __load_psw_mask(PSW_KERNEL_BITS);
310 if (pcpu->address == source_cpu)
311 func(data); /* should not return */
312 /* Stop target cpu (if func returns this stops the current cpu). */
313 pcpu_sigp_retry(pcpu, SIGP_STOP, 0);
314 /* Restart func on the target cpu and stop the current cpu. */
315 mem_assign_absolute(lc->restart_stack, stack);
316 mem_assign_absolute(lc->restart_fn, (unsigned long) func);
317 mem_assign_absolute(lc->restart_data, (unsigned long) data);
318 mem_assign_absolute(lc->restart_source, source_cpu);
320 "0: sigp 0,%0,%2 # sigp restart to target cpu\n"
321 " brc 2,0b # busy, try again\n"
322 "1: sigp 0,%1,%3 # sigp stop to current cpu\n"
323 " brc 2,1b # busy, try again\n"
324 : : "d" (pcpu->address), "d" (source_cpu),
325 "K" (SIGP_RESTART), "K" (SIGP_STOP)
331 * Enable additional logical cpus for multi-threading.
333 static int pcpu_set_smt(unsigned int mtid)
337 if (smp_cpu_mtid == mtid)
339 cc = __pcpu_sigp(0, SIGP_SET_MULTI_THREADING, mtid, NULL);
342 smp_cpu_mt_shift = 0;
343 while (smp_cpu_mtid >= (1U << smp_cpu_mt_shift))
345 pcpu_devices[0].address = stap();
351 * Call function on an online CPU.
353 void smp_call_online_cpu(void (*func)(void *), void *data)
357 /* Use the current cpu if it is online. */
358 pcpu = pcpu_find_address(cpu_online_mask, stap());
360 /* Use the first online cpu. */
361 pcpu = pcpu_devices + cpumask_first(cpu_online_mask);
362 pcpu_delegate(pcpu, func, data, (unsigned long) restart_stack);
366 * Call function on the ipl CPU.
368 void smp_call_ipl_cpu(void (*func)(void *), void *data)
370 pcpu_delegate(&pcpu_devices[0], func, data,
371 pcpu_devices->lowcore->panic_stack -
372 PANIC_FRAME_OFFSET + PAGE_SIZE);
375 int smp_find_processor_id(u16 address)
379 for_each_present_cpu(cpu)
380 if (pcpu_devices[cpu].address == address)
385 bool arch_vcpu_is_preempted(int cpu)
387 if (test_cpu_flag_of(CIF_ENABLED_WAIT, cpu))
389 if (pcpu_running(pcpu_devices + cpu))
393 EXPORT_SYMBOL(arch_vcpu_is_preempted);
395 void smp_yield_cpu(int cpu)
397 if (MACHINE_HAS_DIAG9C) {
398 diag_stat_inc_norecursion(DIAG_STAT_X09C);
399 asm volatile("diag %0,0,0x9c"
400 : : "d" (pcpu_devices[cpu].address));
401 } else if (MACHINE_HAS_DIAG44) {
402 diag_stat_inc_norecursion(DIAG_STAT_X044);
403 asm volatile("diag 0,0,0x44");
408 * Send cpus emergency shutdown signal. This gives the cpus the
409 * opportunity to complete outstanding interrupts.
411 void notrace smp_emergency_stop(void)
417 cpumask_copy(&cpumask, cpu_online_mask);
418 cpumask_clear_cpu(smp_processor_id(), &cpumask);
420 end = get_tod_clock() + (1000000UL << 12);
421 for_each_cpu(cpu, &cpumask) {
422 struct pcpu *pcpu = pcpu_devices + cpu;
423 set_bit(ec_stop_cpu, &pcpu->ec_mask);
424 while (__pcpu_sigp(pcpu->address, SIGP_EMERGENCY_SIGNAL,
425 0, NULL) == SIGP_CC_BUSY &&
426 get_tod_clock() < end)
429 while (get_tod_clock() < end) {
430 for_each_cpu(cpu, &cpumask)
431 if (pcpu_stopped(pcpu_devices + cpu))
432 cpumask_clear_cpu(cpu, &cpumask);
433 if (cpumask_empty(&cpumask))
438 NOKPROBE_SYMBOL(smp_emergency_stop);
441 * Stop all cpus but the current one.
443 void smp_send_stop(void)
447 /* Disable all interrupts/machine checks */
448 __load_psw_mask(PSW_KERNEL_BITS | PSW_MASK_DAT);
449 trace_hardirqs_off();
451 debug_set_critical();
453 if (oops_in_progress)
454 smp_emergency_stop();
456 /* stop all processors */
457 for_each_online_cpu(cpu) {
458 if (cpu == smp_processor_id())
460 pcpu_sigp_retry(pcpu_devices + cpu, SIGP_STOP, 0);
461 while (!pcpu_stopped(pcpu_devices + cpu))
467 * This is the main routine where commands issued by other
470 static void smp_handle_ext_call(void)
474 /* handle bit signal external calls */
475 bits = xchg(&pcpu_devices[smp_processor_id()].ec_mask, 0);
476 if (test_bit(ec_stop_cpu, &bits))
478 if (test_bit(ec_schedule, &bits))
480 if (test_bit(ec_call_function_single, &bits))
481 generic_smp_call_function_single_interrupt();
484 static void do_ext_call_interrupt(struct ext_code ext_code,
485 unsigned int param32, unsigned long param64)
487 inc_irq_stat(ext_code.code == 0x1202 ? IRQEXT_EXC : IRQEXT_EMS);
488 smp_handle_ext_call();
491 void arch_send_call_function_ipi_mask(const struct cpumask *mask)
495 for_each_cpu(cpu, mask)
496 pcpu_ec_call(pcpu_devices + cpu, ec_call_function_single);
499 void arch_send_call_function_single_ipi(int cpu)
501 pcpu_ec_call(pcpu_devices + cpu, ec_call_function_single);
505 * this function sends a 'reschedule' IPI to another CPU.
506 * it goes straight through and wastes no time serializing
507 * anything. Worst case is that we lose a reschedule ...
509 void smp_send_reschedule(int cpu)
511 pcpu_ec_call(pcpu_devices + cpu, ec_schedule);
515 * parameter area for the set/clear control bit callbacks
517 struct ec_creg_mask_parms {
519 unsigned long andval;
524 * callback for setting/clearing control bits
526 static void smp_ctl_bit_callback(void *info)
528 struct ec_creg_mask_parms *pp = info;
529 unsigned long cregs[16];
531 __ctl_store(cregs, 0, 15);
532 cregs[pp->cr] = (cregs[pp->cr] & pp->andval) | pp->orval;
533 __ctl_load(cregs, 0, 15);
537 * Set a bit in a control register of all cpus
539 void smp_ctl_set_bit(int cr, int bit)
541 struct ec_creg_mask_parms parms = { 1UL << bit, -1UL, cr };
543 on_each_cpu(smp_ctl_bit_callback, &parms, 1);
545 EXPORT_SYMBOL(smp_ctl_set_bit);
548 * Clear a bit in a control register of all cpus
550 void smp_ctl_clear_bit(int cr, int bit)
552 struct ec_creg_mask_parms parms = { 0, ~(1UL << bit), cr };
554 on_each_cpu(smp_ctl_bit_callback, &parms, 1);
556 EXPORT_SYMBOL(smp_ctl_clear_bit);
558 #ifdef CONFIG_CRASH_DUMP
560 int smp_store_status(int cpu)
562 struct pcpu *pcpu = pcpu_devices + cpu;
565 pa = __pa(&pcpu->lowcore->floating_pt_save_area);
566 if (__pcpu_sigp_relax(pcpu->address, SIGP_STORE_STATUS_AT_ADDRESS,
567 pa) != SIGP_CC_ORDER_CODE_ACCEPTED)
569 if (!MACHINE_HAS_VX && !MACHINE_HAS_GS)
571 pa = __pa(pcpu->lowcore->mcesad & MCESA_ORIGIN_MASK);
573 pa |= pcpu->lowcore->mcesad & MCESA_LC_MASK;
574 if (__pcpu_sigp_relax(pcpu->address, SIGP_STORE_ADDITIONAL_STATUS,
575 pa) != SIGP_CC_ORDER_CODE_ACCEPTED)
581 * Collect CPU state of the previous, crashed system.
582 * There are four cases:
583 * 1) standard zfcp dump
584 * condition: OLDMEM_BASE == NULL && ipl_info.type == IPL_TYPE_FCP_DUMP
585 * The state for all CPUs except the boot CPU needs to be collected
586 * with sigp stop-and-store-status. The boot CPU state is located in
587 * the absolute lowcore of the memory stored in the HSA. The zcore code
588 * will copy the boot CPU state from the HSA.
589 * 2) stand-alone kdump for SCSI (zfcp dump with swapped memory)
590 * condition: OLDMEM_BASE != NULL && ipl_info.type == IPL_TYPE_FCP_DUMP
591 * The state for all CPUs except the boot CPU needs to be collected
592 * with sigp stop-and-store-status. The firmware or the boot-loader
593 * stored the registers of the boot CPU in the absolute lowcore in the
594 * memory of the old system.
595 * 3) kdump and the old kernel did not store the CPU state,
596 * or stand-alone kdump for DASD
597 * condition: OLDMEM_BASE != NULL && !is_kdump_kernel()
598 * The state for all CPUs except the boot CPU needs to be collected
599 * with sigp stop-and-store-status. The kexec code or the boot-loader
600 * stored the registers of the boot CPU in the memory of the old system.
601 * 4) kdump and the old kernel stored the CPU state
602 * condition: OLDMEM_BASE != NULL && is_kdump_kernel()
603 * This case does not exist for s390 anymore, setup_arch explicitly
604 * deactivates the elfcorehdr= kernel parameter
606 static __init void smp_save_cpu_vxrs(struct save_area *sa, u16 addr,
607 bool is_boot_cpu, unsigned long page)
609 __vector128 *vxrs = (__vector128 *) page;
612 vxrs = boot_cpu_vector_save_area;
614 __pcpu_sigp_relax(addr, SIGP_STORE_ADDITIONAL_STATUS, page);
615 save_area_add_vxrs(sa, vxrs);
618 static __init void smp_save_cpu_regs(struct save_area *sa, u16 addr,
619 bool is_boot_cpu, unsigned long page)
621 void *regs = (void *) page;
624 copy_oldmem_kernel(regs, (void *) __LC_FPREGS_SAVE_AREA, 512);
626 __pcpu_sigp_relax(addr, SIGP_STORE_STATUS_AT_ADDRESS, page);
627 save_area_add_regs(sa, regs);
630 void __init smp_save_dump_cpus(void)
632 int addr, boot_cpu_addr, max_cpu_addr;
633 struct save_area *sa;
637 if (!(OLDMEM_BASE || ipl_info.type == IPL_TYPE_FCP_DUMP))
638 /* No previous system present, normal boot. */
640 /* Allocate a page as dumping area for the store status sigps */
641 page = memblock_alloc_base(PAGE_SIZE, PAGE_SIZE, 1UL << 31);
642 /* Set multi-threading state to the previous system. */
643 pcpu_set_smt(sclp.mtid_prev);
644 boot_cpu_addr = stap();
645 max_cpu_addr = SCLP_MAX_CORES << sclp.mtid_prev;
646 for (addr = 0; addr <= max_cpu_addr; addr++) {
647 if (__pcpu_sigp_relax(addr, SIGP_SENSE, 0) ==
648 SIGP_CC_NOT_OPERATIONAL)
650 is_boot_cpu = (addr == boot_cpu_addr);
651 /* Allocate save area */
652 sa = save_area_alloc(is_boot_cpu);
654 panic("could not allocate memory for save area\n");
656 /* Get the vector registers */
657 smp_save_cpu_vxrs(sa, addr, is_boot_cpu, page);
659 * For a zfcp dump OLDMEM_BASE == NULL and the registers
660 * of the boot CPU are stored in the HSA. To retrieve
661 * these registers an SCLP request is required which is
662 * done by drivers/s390/char/zcore.c:init_cpu_info()
664 if (!is_boot_cpu || OLDMEM_BASE)
665 /* Get the CPU registers */
666 smp_save_cpu_regs(sa, addr, is_boot_cpu, page);
668 memblock_free(page, PAGE_SIZE);
672 #endif /* CONFIG_CRASH_DUMP */
674 void smp_cpu_set_polarization(int cpu, int val)
676 pcpu_devices[cpu].polarization = val;
679 int smp_cpu_get_polarization(int cpu)
681 return pcpu_devices[cpu].polarization;
684 static void __ref smp_get_core_info(struct sclp_core_info *info, int early)
686 static int use_sigp_detection;
689 if (use_sigp_detection || sclp_get_core_info(info, early)) {
690 use_sigp_detection = 1;
692 address < (SCLP_MAX_CORES << smp_cpu_mt_shift);
693 address += (1U << smp_cpu_mt_shift)) {
694 if (__pcpu_sigp_relax(address, SIGP_SENSE, 0) ==
695 SIGP_CC_NOT_OPERATIONAL)
697 info->core[info->configured].core_id =
698 address >> smp_cpu_mt_shift;
701 info->combined = info->configured;
705 static int smp_add_present_cpu(int cpu);
707 static int __smp_rescan_cpus(struct sclp_core_info *info, int sysfs_add)
715 cpumask_xor(&avail, cpu_possible_mask, cpu_present_mask);
716 cpu = cpumask_first(&avail);
717 for (i = 0; (i < info->combined) && (cpu < nr_cpu_ids); i++) {
718 if (sclp.has_core_type && info->core[i].type != boot_core_type)
720 address = info->core[i].core_id << smp_cpu_mt_shift;
721 for (j = 0; j <= smp_cpu_mtid; j++) {
722 if (pcpu_find_address(cpu_present_mask, address + j))
724 pcpu = pcpu_devices + cpu;
725 pcpu->address = address + j;
727 (cpu >= info->configured*(smp_cpu_mtid + 1)) ?
728 CPU_STATE_STANDBY : CPU_STATE_CONFIGURED;
729 smp_cpu_set_polarization(cpu, POLARIZATION_UNKNOWN);
730 set_cpu_present(cpu, true);
731 if (sysfs_add && smp_add_present_cpu(cpu) != 0)
732 set_cpu_present(cpu, false);
735 cpu = cpumask_next(cpu, &avail);
736 if (cpu >= nr_cpu_ids)
743 void __init smp_detect_cpus(void)
745 unsigned int cpu, mtid, c_cpus, s_cpus;
746 struct sclp_core_info *info;
749 /* Get CPU information */
750 info = memblock_virt_alloc(sizeof(*info), 8);
751 smp_get_core_info(info, 1);
752 /* Find boot CPU type */
753 if (sclp.has_core_type) {
755 for (cpu = 0; cpu < info->combined; cpu++)
756 if (info->core[cpu].core_id == address) {
757 /* The boot cpu dictates the cpu type. */
758 boot_core_type = info->core[cpu].type;
761 if (cpu >= info->combined)
762 panic("Could not find boot CPU type");
765 /* Set multi-threading state for the current system */
766 mtid = boot_core_type ? sclp.mtid : sclp.mtid_cp;
767 mtid = (mtid < smp_max_threads) ? mtid : smp_max_threads - 1;
770 /* Print number of CPUs */
772 for (cpu = 0; cpu < info->combined; cpu++) {
773 if (sclp.has_core_type &&
774 info->core[cpu].type != boot_core_type)
776 if (cpu < info->configured)
777 c_cpus += smp_cpu_mtid + 1;
779 s_cpus += smp_cpu_mtid + 1;
781 pr_info("%d configured CPUs, %d standby CPUs\n", c_cpus, s_cpus);
783 /* Add CPUs present at boot */
785 __smp_rescan_cpus(info, 0);
787 memblock_free_early((unsigned long)info, sizeof(*info));
791 * Activate a secondary processor.
793 static void smp_start_secondary(void *cpuvoid)
795 int cpu = smp_processor_id();
797 S390_lowcore.last_update_clock = get_tod_clock();
798 S390_lowcore.restart_stack = (unsigned long) restart_stack;
799 S390_lowcore.restart_fn = (unsigned long) do_restart;
800 S390_lowcore.restart_data = 0;
801 S390_lowcore.restart_source = -1UL;
802 restore_access_regs(S390_lowcore.access_regs_save_area);
803 __ctl_load(S390_lowcore.cregs_save_area, 0, 15);
804 __load_psw_mask(PSW_KERNEL_BITS | PSW_MASK_DAT);
810 notify_cpu_starting(cpu);
811 if (topology_cpu_dedicated(cpu))
812 set_cpu_flag(CIF_DEDICATED_CPU);
814 clear_cpu_flag(CIF_DEDICATED_CPU);
815 set_cpu_online(cpu, true);
816 inc_irq_stat(CPU_RST);
818 cpu_startup_entry(CPUHP_AP_ONLINE_IDLE);
821 /* Upping and downing of CPUs */
822 int __cpu_up(unsigned int cpu, struct task_struct *tidle)
827 pcpu = pcpu_devices + cpu;
828 if (pcpu->state != CPU_STATE_CONFIGURED)
830 base = smp_get_base_cpu(cpu);
831 for (i = 0; i <= smp_cpu_mtid; i++) {
832 if (base + i < nr_cpu_ids)
833 if (cpu_online(base + i))
837 * If this is the first CPU of the core to get online
838 * do an initial CPU reset.
840 if (i > smp_cpu_mtid &&
841 pcpu_sigp_retry(pcpu_devices + base, SIGP_INITIAL_CPU_RESET, 0) !=
842 SIGP_CC_ORDER_CODE_ACCEPTED)
845 rc = pcpu_alloc_lowcore(pcpu, cpu);
848 pcpu_prepare_secondary(pcpu, cpu);
849 pcpu_attach_task(pcpu, tidle);
850 pcpu_start_fn(pcpu, smp_start_secondary, NULL);
851 /* Wait until cpu puts itself in the online & active maps */
852 while (!cpu_online(cpu))
857 static unsigned int setup_possible_cpus __initdata;
859 static int __init _setup_possible_cpus(char *s)
861 get_option(&s, &setup_possible_cpus);
864 early_param("possible_cpus", _setup_possible_cpus);
866 #ifdef CONFIG_HOTPLUG_CPU
868 int __cpu_disable(void)
870 unsigned long cregs[16];
872 /* Handle possible pending IPIs */
873 smp_handle_ext_call();
874 set_cpu_online(smp_processor_id(), false);
875 /* Disable pseudo page faults on this cpu. */
877 /* Disable interrupt sources via control register. */
878 __ctl_store(cregs, 0, 15);
879 cregs[0] &= ~0x0000ee70UL; /* disable all external interrupts */
880 cregs[6] &= ~0xff000000UL; /* disable all I/O interrupts */
881 cregs[14] &= ~0x1f000000UL; /* disable most machine checks */
882 __ctl_load(cregs, 0, 15);
883 clear_cpu_flag(CIF_NOHZ_DELAY);
887 void __cpu_die(unsigned int cpu)
891 /* Wait until target cpu is down */
892 pcpu = pcpu_devices + cpu;
893 while (!pcpu_stopped(pcpu))
895 pcpu_free_lowcore(pcpu);
896 cpumask_clear_cpu(cpu, mm_cpumask(&init_mm));
897 cpumask_clear_cpu(cpu, &init_mm.context.cpu_attach_mask);
900 void __noreturn cpu_die(void)
903 pcpu_sigp_retry(pcpu_devices + smp_processor_id(), SIGP_STOP, 0);
907 #endif /* CONFIG_HOTPLUG_CPU */
909 void __init smp_fill_possible_mask(void)
911 unsigned int possible, sclp_max, cpu;
913 sclp_max = max(sclp.mtid, sclp.mtid_cp) + 1;
914 sclp_max = min(smp_max_threads, sclp_max);
915 sclp_max = (sclp.max_cores * sclp_max) ?: nr_cpu_ids;
916 possible = setup_possible_cpus ?: nr_cpu_ids;
917 possible = min(possible, sclp_max);
918 for (cpu = 0; cpu < possible && cpu < nr_cpu_ids; cpu++)
919 set_cpu_possible(cpu, true);
922 void __init smp_prepare_cpus(unsigned int max_cpus)
924 /* request the 0x1201 emergency signal external interrupt */
925 if (register_external_irq(EXT_IRQ_EMERGENCY_SIG, do_ext_call_interrupt))
926 panic("Couldn't request external interrupt 0x1201");
927 /* request the 0x1202 external call external interrupt */
928 if (register_external_irq(EXT_IRQ_EXTERNAL_CALL, do_ext_call_interrupt))
929 panic("Couldn't request external interrupt 0x1202");
932 void __init smp_prepare_boot_cpu(void)
934 struct pcpu *pcpu = pcpu_devices;
936 WARN_ON(!cpu_present(0) || !cpu_online(0));
937 pcpu->state = CPU_STATE_CONFIGURED;
938 pcpu->lowcore = (struct lowcore *)(unsigned long) store_prefix();
939 S390_lowcore.percpu_offset = __per_cpu_offset[0];
940 smp_cpu_set_polarization(0, POLARIZATION_UNKNOWN);
943 void __init smp_cpus_done(unsigned int max_cpus)
947 void __init smp_setup_processor_id(void)
949 pcpu_devices[0].address = stap();
950 S390_lowcore.cpu_nr = 0;
951 S390_lowcore.spinlock_lockval = arch_spin_lockval(0);
952 S390_lowcore.spinlock_index = 0;
956 * the frequency of the profiling timer can be changed
957 * by writing a multiplier value into /proc/profile.
959 * usually you want to run this on all CPUs ;)
961 int setup_profiling_timer(unsigned int multiplier)
966 #ifdef CONFIG_HOTPLUG_CPU
967 static ssize_t cpu_configure_show(struct device *dev,
968 struct device_attribute *attr, char *buf)
972 mutex_lock(&smp_cpu_state_mutex);
973 count = sprintf(buf, "%d\n", pcpu_devices[dev->id].state);
974 mutex_unlock(&smp_cpu_state_mutex);
978 static ssize_t cpu_configure_store(struct device *dev,
979 struct device_attribute *attr,
980 const char *buf, size_t count)
986 if (sscanf(buf, "%d %c", &val, &delim) != 1)
988 if (val != 0 && val != 1)
991 mutex_lock(&smp_cpu_state_mutex);
993 /* disallow configuration changes of online cpus and cpu 0 */
995 cpu = smp_get_base_cpu(cpu);
998 for (i = 0; i <= smp_cpu_mtid; i++)
999 if (cpu_online(cpu + i))
1001 pcpu = pcpu_devices + cpu;
1005 if (pcpu->state != CPU_STATE_CONFIGURED)
1007 rc = sclp_core_deconfigure(pcpu->address >> smp_cpu_mt_shift);
1010 for (i = 0; i <= smp_cpu_mtid; i++) {
1011 if (cpu + i >= nr_cpu_ids || !cpu_present(cpu + i))
1013 pcpu[i].state = CPU_STATE_STANDBY;
1014 smp_cpu_set_polarization(cpu + i,
1015 POLARIZATION_UNKNOWN);
1017 topology_expect_change();
1020 if (pcpu->state != CPU_STATE_STANDBY)
1022 rc = sclp_core_configure(pcpu->address >> smp_cpu_mt_shift);
1025 for (i = 0; i <= smp_cpu_mtid; i++) {
1026 if (cpu + i >= nr_cpu_ids || !cpu_present(cpu + i))
1028 pcpu[i].state = CPU_STATE_CONFIGURED;
1029 smp_cpu_set_polarization(cpu + i,
1030 POLARIZATION_UNKNOWN);
1032 topology_expect_change();
1038 mutex_unlock(&smp_cpu_state_mutex);
1040 return rc ? rc : count;
1042 static DEVICE_ATTR(configure, 0644, cpu_configure_show, cpu_configure_store);
1043 #endif /* CONFIG_HOTPLUG_CPU */
1045 static ssize_t show_cpu_address(struct device *dev,
1046 struct device_attribute *attr, char *buf)
1048 return sprintf(buf, "%d\n", pcpu_devices[dev->id].address);
1050 static DEVICE_ATTR(address, 0444, show_cpu_address, NULL);
1052 static struct attribute *cpu_common_attrs[] = {
1053 #ifdef CONFIG_HOTPLUG_CPU
1054 &dev_attr_configure.attr,
1056 &dev_attr_address.attr,
1060 static struct attribute_group cpu_common_attr_group = {
1061 .attrs = cpu_common_attrs,
1064 static struct attribute *cpu_online_attrs[] = {
1065 &dev_attr_idle_count.attr,
1066 &dev_attr_idle_time_us.attr,
1070 static struct attribute_group cpu_online_attr_group = {
1071 .attrs = cpu_online_attrs,
1074 static int smp_cpu_online(unsigned int cpu)
1076 struct device *s = &per_cpu(cpu_device, cpu)->dev;
1078 return sysfs_create_group(&s->kobj, &cpu_online_attr_group);
1080 static int smp_cpu_pre_down(unsigned int cpu)
1082 struct device *s = &per_cpu(cpu_device, cpu)->dev;
1084 sysfs_remove_group(&s->kobj, &cpu_online_attr_group);
1088 static int smp_add_present_cpu(int cpu)
1094 c = kzalloc(sizeof(*c), GFP_KERNEL);
1097 per_cpu(cpu_device, cpu) = c;
1099 c->hotpluggable = 1;
1100 rc = register_cpu(c, cpu);
1103 rc = sysfs_create_group(&s->kobj, &cpu_common_attr_group);
1106 rc = topology_cpu_init(c);
1112 sysfs_remove_group(&s->kobj, &cpu_common_attr_group);
1114 #ifdef CONFIG_HOTPLUG_CPU
1121 #ifdef CONFIG_HOTPLUG_CPU
1123 int __ref smp_rescan_cpus(void)
1125 struct sclp_core_info *info;
1128 info = kzalloc(sizeof(*info), GFP_KERNEL);
1131 smp_get_core_info(info, 0);
1133 mutex_lock(&smp_cpu_state_mutex);
1134 nr = __smp_rescan_cpus(info, 1);
1135 mutex_unlock(&smp_cpu_state_mutex);
1139 topology_schedule_update();
1143 static ssize_t __ref rescan_store(struct device *dev,
1144 struct device_attribute *attr,
1150 rc = smp_rescan_cpus();
1151 return rc ? rc : count;
1153 static DEVICE_ATTR(rescan, 0200, NULL, rescan_store);
1154 #endif /* CONFIG_HOTPLUG_CPU */
1156 static int __init s390_smp_init(void)
1160 #ifdef CONFIG_HOTPLUG_CPU
1161 rc = device_create_file(cpu_subsys.dev_root, &dev_attr_rescan);
1165 for_each_present_cpu(cpu) {
1166 rc = smp_add_present_cpu(cpu);
1171 rc = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "s390/smp:online",
1172 smp_cpu_online, smp_cpu_pre_down);
1173 rc = rc <= 0 ? rc : 0;
1177 subsys_initcall(s390_smp_init);