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/memblock.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/delay.h>
31 #include <linux/interrupt.h>
32 #include <linux/irqflags.h>
33 #include <linux/cpu.h>
34 #include <linux/slab.h>
35 #include <linux/sched/hotplug.h>
36 #include <linux/sched/task_stack.h>
37 #include <linux/crash_dump.h>
38 #include <linux/kprobes.h>
39 #include <asm/asm-offsets.h>
41 #include <asm/switch_to.h>
42 #include <asm/facility.h>
44 #include <asm/setup.h>
46 #include <asm/tlbflush.h>
47 #include <asm/vtimer.h>
48 #include <asm/lowcore.h>
51 #include <asm/debug.h>
52 #include <asm/os_info.h>
56 #include <asm/topology.h>
61 ec_call_function_single,
70 static DEFINE_PER_CPU(struct cpu *, cpu_device);
73 struct lowcore *lowcore; /* lowcore page(s) for the cpu */
74 unsigned long ec_mask; /* bit mask for ec_xxx functions */
75 unsigned long ec_clk; /* sigp timestamp for ec_xxx */
76 signed char state; /* physical cpu state */
77 signed char polarization; /* physical polarization */
78 u16 address; /* physical cpu address */
81 static u8 boot_core_type;
82 static struct pcpu pcpu_devices[NR_CPUS];
84 unsigned int smp_cpu_mt_shift;
85 EXPORT_SYMBOL(smp_cpu_mt_shift);
87 unsigned int smp_cpu_mtid;
88 EXPORT_SYMBOL(smp_cpu_mtid);
90 #ifdef CONFIG_CRASH_DUMP
91 __vector128 __initdata boot_cpu_vector_save_area[__NUM_VXRS];
94 static unsigned int smp_max_threads __initdata = -1U;
96 static int __init early_nosmt(char *s)
101 early_param("nosmt", early_nosmt);
103 static int __init early_smt(char *s)
105 get_option(&s, &smp_max_threads);
108 early_param("smt", early_smt);
111 * The smp_cpu_state_mutex must be held when changing the state or polarization
112 * member of a pcpu data structure within the pcpu_devices arreay.
114 DEFINE_MUTEX(smp_cpu_state_mutex);
117 * Signal processor helper functions.
119 static inline int __pcpu_sigp_relax(u16 addr, u8 order, unsigned long parm)
124 cc = __pcpu_sigp(addr, order, parm, NULL);
125 if (cc != SIGP_CC_BUSY)
131 static int pcpu_sigp_retry(struct pcpu *pcpu, u8 order, u32 parm)
135 for (retry = 0; ; retry++) {
136 cc = __pcpu_sigp(pcpu->address, order, parm, NULL);
137 if (cc != SIGP_CC_BUSY)
145 static inline int pcpu_stopped(struct pcpu *pcpu)
147 u32 uninitialized_var(status);
149 if (__pcpu_sigp(pcpu->address, SIGP_SENSE,
150 0, &status) != SIGP_CC_STATUS_STORED)
152 return !!(status & (SIGP_STATUS_CHECK_STOP|SIGP_STATUS_STOPPED));
155 static inline int pcpu_running(struct pcpu *pcpu)
157 if (__pcpu_sigp(pcpu->address, SIGP_SENSE_RUNNING,
158 0, NULL) != SIGP_CC_STATUS_STORED)
160 /* Status stored condition code is equivalent to cpu not running. */
165 * Find struct pcpu by cpu address.
167 static struct pcpu *pcpu_find_address(const struct cpumask *mask, u16 address)
171 for_each_cpu(cpu, mask)
172 if (pcpu_devices[cpu].address == address)
173 return pcpu_devices + cpu;
177 static void pcpu_ec_call(struct pcpu *pcpu, int ec_bit)
181 if (test_and_set_bit(ec_bit, &pcpu->ec_mask))
183 order = pcpu_running(pcpu) ? SIGP_EXTERNAL_CALL : SIGP_EMERGENCY_SIGNAL;
184 pcpu->ec_clk = get_tod_clock_fast();
185 pcpu_sigp_retry(pcpu, order, 0);
188 static int pcpu_alloc_lowcore(struct pcpu *pcpu, int cpu)
190 unsigned long async_stack, nodat_stack;
193 if (pcpu != &pcpu_devices[0]) {
194 pcpu->lowcore = (struct lowcore *)
195 __get_free_pages(GFP_KERNEL | GFP_DMA, LC_ORDER);
196 nodat_stack = __get_free_pages(GFP_KERNEL, THREAD_SIZE_ORDER);
197 if (!pcpu->lowcore || !nodat_stack)
200 nodat_stack = pcpu->lowcore->nodat_stack - STACK_INIT_OFFSET;
202 async_stack = stack_alloc();
206 memcpy(lc, &S390_lowcore, 512);
207 memset((char *) lc + 512, 0, sizeof(*lc) - 512);
208 lc->async_stack = async_stack + STACK_INIT_OFFSET;
209 lc->nodat_stack = nodat_stack + STACK_INIT_OFFSET;
211 lc->spinlock_lockval = arch_spin_lockval(cpu);
212 lc->spinlock_index = 0;
213 lc->br_r1_trampoline = 0x07f1; /* br %r1 */
214 if (nmi_alloc_per_cpu(lc))
216 if (vdso_alloc_per_cpu(lc))
218 lowcore_ptr[cpu] = lc;
219 pcpu_sigp_retry(pcpu, SIGP_SET_PREFIX, (u32)(unsigned long) lc);
223 nmi_free_per_cpu(lc);
225 stack_free(async_stack);
227 if (pcpu != &pcpu_devices[0]) {
228 free_pages(nodat_stack, THREAD_SIZE_ORDER);
229 free_pages((unsigned long) pcpu->lowcore, LC_ORDER);
234 #ifdef CONFIG_HOTPLUG_CPU
236 static void pcpu_free_lowcore(struct pcpu *pcpu)
238 unsigned long async_stack, nodat_stack, lowcore;
240 nodat_stack = pcpu->lowcore->nodat_stack - STACK_INIT_OFFSET;
241 async_stack = pcpu->lowcore->async_stack - STACK_INIT_OFFSET;
242 lowcore = (unsigned long) pcpu->lowcore;
244 pcpu_sigp_retry(pcpu, SIGP_SET_PREFIX, 0);
245 lowcore_ptr[pcpu - pcpu_devices] = NULL;
246 vdso_free_per_cpu(pcpu->lowcore);
247 nmi_free_per_cpu(pcpu->lowcore);
248 stack_free(async_stack);
249 if (pcpu == &pcpu_devices[0])
251 free_pages(nodat_stack, THREAD_SIZE_ORDER);
252 free_pages(lowcore, LC_ORDER);
255 #endif /* CONFIG_HOTPLUG_CPU */
257 static void pcpu_prepare_secondary(struct pcpu *pcpu, int cpu)
259 struct lowcore *lc = pcpu->lowcore;
261 cpumask_set_cpu(cpu, &init_mm.context.cpu_attach_mask);
262 cpumask_set_cpu(cpu, mm_cpumask(&init_mm));
264 lc->spinlock_lockval = arch_spin_lockval(cpu);
265 lc->spinlock_index = 0;
266 lc->percpu_offset = __per_cpu_offset[cpu];
267 lc->kernel_asce = S390_lowcore.kernel_asce;
268 lc->machine_flags = S390_lowcore.machine_flags;
269 lc->user_timer = lc->system_timer = lc->steal_timer = 0;
270 __ctl_store(lc->cregs_save_area, 0, 15);
271 save_access_regs((unsigned int *) lc->access_regs_save_area);
272 memcpy(lc->stfle_fac_list, S390_lowcore.stfle_fac_list,
273 sizeof(lc->stfle_fac_list));
274 memcpy(lc->alt_stfle_fac_list, S390_lowcore.alt_stfle_fac_list,
275 sizeof(lc->alt_stfle_fac_list));
276 arch_spin_lock_setup(cpu);
279 static void pcpu_attach_task(struct pcpu *pcpu, struct task_struct *tsk)
281 struct lowcore *lc = pcpu->lowcore;
283 lc->kernel_stack = (unsigned long) task_stack_page(tsk)
284 + THREAD_SIZE - STACK_FRAME_OVERHEAD - sizeof(struct pt_regs);
285 lc->current_task = (unsigned long) tsk;
287 lc->current_pid = tsk->pid;
288 lc->user_timer = tsk->thread.user_timer;
289 lc->guest_timer = tsk->thread.guest_timer;
290 lc->system_timer = tsk->thread.system_timer;
291 lc->hardirq_timer = tsk->thread.hardirq_timer;
292 lc->softirq_timer = tsk->thread.softirq_timer;
296 static void pcpu_start_fn(struct pcpu *pcpu, void (*func)(void *), void *data)
298 struct lowcore *lc = pcpu->lowcore;
300 lc->restart_stack = lc->nodat_stack;
301 lc->restart_fn = (unsigned long) func;
302 lc->restart_data = (unsigned long) data;
303 lc->restart_source = -1UL;
304 pcpu_sigp_retry(pcpu, SIGP_RESTART, 0);
308 * Call function via PSW restart on pcpu and stop the current cpu.
310 static void __pcpu_delegate(void (*func)(void*), void *data)
312 func(data); /* should not return */
315 static void __no_sanitize_address pcpu_delegate(struct pcpu *pcpu,
316 void (*func)(void *),
317 void *data, unsigned long stack)
319 struct lowcore *lc = lowcore_ptr[pcpu - pcpu_devices];
320 unsigned long source_cpu = stap();
322 __load_psw_mask(PSW_KERNEL_BITS | PSW_MASK_DAT);
323 if (pcpu->address == source_cpu)
324 CALL_ON_STACK(__pcpu_delegate, stack, 2, func, data);
325 /* Stop target cpu (if func returns this stops the current cpu). */
326 pcpu_sigp_retry(pcpu, SIGP_STOP, 0);
327 /* Restart func on the target cpu and stop the current cpu. */
328 mem_assign_absolute(lc->restart_stack, stack);
329 mem_assign_absolute(lc->restart_fn, (unsigned long) func);
330 mem_assign_absolute(lc->restart_data, (unsigned long) data);
331 mem_assign_absolute(lc->restart_source, source_cpu);
334 "0: sigp 0,%0,%2 # sigp restart to target cpu\n"
335 " brc 2,0b # busy, try again\n"
336 "1: sigp 0,%1,%3 # sigp stop to current cpu\n"
337 " brc 2,1b # busy, try again\n"
338 : : "d" (pcpu->address), "d" (source_cpu),
339 "K" (SIGP_RESTART), "K" (SIGP_STOP)
345 * Enable additional logical cpus for multi-threading.
347 static int pcpu_set_smt(unsigned int mtid)
351 if (smp_cpu_mtid == mtid)
353 cc = __pcpu_sigp(0, SIGP_SET_MULTI_THREADING, mtid, NULL);
356 smp_cpu_mt_shift = 0;
357 while (smp_cpu_mtid >= (1U << smp_cpu_mt_shift))
359 pcpu_devices[0].address = stap();
365 * Call function on an online CPU.
367 void smp_call_online_cpu(void (*func)(void *), void *data)
371 /* Use the current cpu if it is online. */
372 pcpu = pcpu_find_address(cpu_online_mask, stap());
374 /* Use the first online cpu. */
375 pcpu = pcpu_devices + cpumask_first(cpu_online_mask);
376 pcpu_delegate(pcpu, func, data, (unsigned long) restart_stack);
380 * Call function on the ipl CPU.
382 void smp_call_ipl_cpu(void (*func)(void *), void *data)
384 struct lowcore *lc = pcpu_devices->lowcore;
386 if (pcpu_devices[0].address == stap())
389 pcpu_delegate(&pcpu_devices[0], func, data,
393 int smp_find_processor_id(u16 address)
397 for_each_present_cpu(cpu)
398 if (pcpu_devices[cpu].address == address)
403 bool arch_vcpu_is_preempted(int cpu)
405 if (test_cpu_flag_of(CIF_ENABLED_WAIT, cpu))
407 if (pcpu_running(pcpu_devices + cpu))
411 EXPORT_SYMBOL(arch_vcpu_is_preempted);
413 void smp_yield_cpu(int cpu)
415 if (MACHINE_HAS_DIAG9C) {
416 diag_stat_inc_norecursion(DIAG_STAT_X09C);
417 asm volatile("diag %0,0,0x9c"
418 : : "d" (pcpu_devices[cpu].address));
419 } else if (MACHINE_HAS_DIAG44) {
420 diag_stat_inc_norecursion(DIAG_STAT_X044);
421 asm volatile("diag 0,0,0x44");
426 * Send cpus emergency shutdown signal. This gives the cpus the
427 * opportunity to complete outstanding interrupts.
429 void notrace smp_emergency_stop(void)
435 cpumask_copy(&cpumask, cpu_online_mask);
436 cpumask_clear_cpu(smp_processor_id(), &cpumask);
438 end = get_tod_clock() + (1000000UL << 12);
439 for_each_cpu(cpu, &cpumask) {
440 struct pcpu *pcpu = pcpu_devices + cpu;
441 set_bit(ec_stop_cpu, &pcpu->ec_mask);
442 while (__pcpu_sigp(pcpu->address, SIGP_EMERGENCY_SIGNAL,
443 0, NULL) == SIGP_CC_BUSY &&
444 get_tod_clock() < end)
447 while (get_tod_clock() < end) {
448 for_each_cpu(cpu, &cpumask)
449 if (pcpu_stopped(pcpu_devices + cpu))
450 cpumask_clear_cpu(cpu, &cpumask);
451 if (cpumask_empty(&cpumask))
456 NOKPROBE_SYMBOL(smp_emergency_stop);
459 * Stop all cpus but the current one.
461 void smp_send_stop(void)
465 /* Disable all interrupts/machine checks */
466 __load_psw_mask(PSW_KERNEL_BITS | PSW_MASK_DAT);
467 trace_hardirqs_off();
469 debug_set_critical();
471 if (oops_in_progress)
472 smp_emergency_stop();
474 /* stop all processors */
475 for_each_online_cpu(cpu) {
476 if (cpu == smp_processor_id())
478 pcpu_sigp_retry(pcpu_devices + cpu, SIGP_STOP, 0);
479 while (!pcpu_stopped(pcpu_devices + cpu))
485 * This is the main routine where commands issued by other
488 static void smp_handle_ext_call(void)
492 /* handle bit signal external calls */
493 bits = xchg(&pcpu_devices[smp_processor_id()].ec_mask, 0);
494 if (test_bit(ec_stop_cpu, &bits))
496 if (test_bit(ec_schedule, &bits))
498 if (test_bit(ec_call_function_single, &bits))
499 generic_smp_call_function_single_interrupt();
502 static void do_ext_call_interrupt(struct ext_code ext_code,
503 unsigned int param32, unsigned long param64)
505 inc_irq_stat(ext_code.code == 0x1202 ? IRQEXT_EXC : IRQEXT_EMS);
506 smp_handle_ext_call();
509 void arch_send_call_function_ipi_mask(const struct cpumask *mask)
513 for_each_cpu(cpu, mask)
514 pcpu_ec_call(pcpu_devices + cpu, ec_call_function_single);
517 void arch_send_call_function_single_ipi(int cpu)
519 pcpu_ec_call(pcpu_devices + cpu, ec_call_function_single);
523 * this function sends a 'reschedule' IPI to another CPU.
524 * it goes straight through and wastes no time serializing
525 * anything. Worst case is that we lose a reschedule ...
527 void smp_send_reschedule(int cpu)
529 pcpu_ec_call(pcpu_devices + cpu, ec_schedule);
533 * parameter area for the set/clear control bit callbacks
535 struct ec_creg_mask_parms {
537 unsigned long andval;
542 * callback for setting/clearing control bits
544 static void smp_ctl_bit_callback(void *info)
546 struct ec_creg_mask_parms *pp = info;
547 unsigned long cregs[16];
549 __ctl_store(cregs, 0, 15);
550 cregs[pp->cr] = (cregs[pp->cr] & pp->andval) | pp->orval;
551 __ctl_load(cregs, 0, 15);
555 * Set a bit in a control register of all cpus
557 void smp_ctl_set_bit(int cr, int bit)
559 struct ec_creg_mask_parms parms = { 1UL << bit, -1UL, cr };
561 on_each_cpu(smp_ctl_bit_callback, &parms, 1);
563 EXPORT_SYMBOL(smp_ctl_set_bit);
566 * Clear a bit in a control register of all cpus
568 void smp_ctl_clear_bit(int cr, int bit)
570 struct ec_creg_mask_parms parms = { 0, ~(1UL << bit), cr };
572 on_each_cpu(smp_ctl_bit_callback, &parms, 1);
574 EXPORT_SYMBOL(smp_ctl_clear_bit);
576 #ifdef CONFIG_CRASH_DUMP
578 int smp_store_status(int cpu)
580 struct pcpu *pcpu = pcpu_devices + cpu;
583 pa = __pa(&pcpu->lowcore->floating_pt_save_area);
584 if (__pcpu_sigp_relax(pcpu->address, SIGP_STORE_STATUS_AT_ADDRESS,
585 pa) != SIGP_CC_ORDER_CODE_ACCEPTED)
587 if (!MACHINE_HAS_VX && !MACHINE_HAS_GS)
589 pa = __pa(pcpu->lowcore->mcesad & MCESA_ORIGIN_MASK);
591 pa |= pcpu->lowcore->mcesad & MCESA_LC_MASK;
592 if (__pcpu_sigp_relax(pcpu->address, SIGP_STORE_ADDITIONAL_STATUS,
593 pa) != SIGP_CC_ORDER_CODE_ACCEPTED)
599 * Collect CPU state of the previous, crashed system.
600 * There are four cases:
601 * 1) standard zfcp dump
602 * condition: OLDMEM_BASE == NULL && ipl_info.type == IPL_TYPE_FCP_DUMP
603 * The state for all CPUs except the boot CPU needs to be collected
604 * with sigp stop-and-store-status. The boot CPU state is located in
605 * the absolute lowcore of the memory stored in the HSA. The zcore code
606 * will copy the boot CPU state from the HSA.
607 * 2) stand-alone kdump for SCSI (zfcp dump with swapped memory)
608 * condition: OLDMEM_BASE != NULL && ipl_info.type == IPL_TYPE_FCP_DUMP
609 * The state for all CPUs except the boot CPU needs to be collected
610 * with sigp stop-and-store-status. The firmware or the boot-loader
611 * stored the registers of the boot CPU in the absolute lowcore in the
612 * memory of the old system.
613 * 3) kdump and the old kernel did not store the CPU state,
614 * or stand-alone kdump for DASD
615 * condition: OLDMEM_BASE != NULL && !is_kdump_kernel()
616 * The state for all CPUs except the boot CPU needs to be collected
617 * with sigp stop-and-store-status. The kexec code or the boot-loader
618 * stored the registers of the boot CPU in the memory of the old system.
619 * 4) kdump and the old kernel stored the CPU state
620 * condition: OLDMEM_BASE != NULL && is_kdump_kernel()
621 * This case does not exist for s390 anymore, setup_arch explicitly
622 * deactivates the elfcorehdr= kernel parameter
624 static __init void smp_save_cpu_vxrs(struct save_area *sa, u16 addr,
625 bool is_boot_cpu, unsigned long page)
627 __vector128 *vxrs = (__vector128 *) page;
630 vxrs = boot_cpu_vector_save_area;
632 __pcpu_sigp_relax(addr, SIGP_STORE_ADDITIONAL_STATUS, page);
633 save_area_add_vxrs(sa, vxrs);
636 static __init void smp_save_cpu_regs(struct save_area *sa, u16 addr,
637 bool is_boot_cpu, unsigned long page)
639 void *regs = (void *) page;
642 copy_oldmem_kernel(regs, (void *) __LC_FPREGS_SAVE_AREA, 512);
644 __pcpu_sigp_relax(addr, SIGP_STORE_STATUS_AT_ADDRESS, page);
645 save_area_add_regs(sa, regs);
648 void __init smp_save_dump_cpus(void)
650 int addr, boot_cpu_addr, max_cpu_addr;
651 struct save_area *sa;
655 if (!(OLDMEM_BASE || ipl_info.type == IPL_TYPE_FCP_DUMP))
656 /* No previous system present, normal boot. */
658 /* Allocate a page as dumping area for the store status sigps */
659 page = memblock_alloc_base(PAGE_SIZE, PAGE_SIZE, 1UL << 31);
660 /* Set multi-threading state to the previous system. */
661 pcpu_set_smt(sclp.mtid_prev);
662 boot_cpu_addr = stap();
663 max_cpu_addr = SCLP_MAX_CORES << sclp.mtid_prev;
664 for (addr = 0; addr <= max_cpu_addr; addr++) {
665 if (__pcpu_sigp_relax(addr, SIGP_SENSE, 0) ==
666 SIGP_CC_NOT_OPERATIONAL)
668 is_boot_cpu = (addr == boot_cpu_addr);
669 /* Allocate save area */
670 sa = save_area_alloc(is_boot_cpu);
672 panic("could not allocate memory for save area\n");
674 /* Get the vector registers */
675 smp_save_cpu_vxrs(sa, addr, is_boot_cpu, page);
677 * For a zfcp dump OLDMEM_BASE == NULL and the registers
678 * of the boot CPU are stored in the HSA. To retrieve
679 * these registers an SCLP request is required which is
680 * done by drivers/s390/char/zcore.c:init_cpu_info()
682 if (!is_boot_cpu || OLDMEM_BASE)
683 /* Get the CPU registers */
684 smp_save_cpu_regs(sa, addr, is_boot_cpu, page);
686 memblock_free(page, PAGE_SIZE);
690 #endif /* CONFIG_CRASH_DUMP */
692 void smp_cpu_set_polarization(int cpu, int val)
694 pcpu_devices[cpu].polarization = val;
697 int smp_cpu_get_polarization(int cpu)
699 return pcpu_devices[cpu].polarization;
702 static void __ref smp_get_core_info(struct sclp_core_info *info, int early)
704 static int use_sigp_detection;
707 if (use_sigp_detection || sclp_get_core_info(info, early)) {
708 use_sigp_detection = 1;
710 address < (SCLP_MAX_CORES << smp_cpu_mt_shift);
711 address += (1U << smp_cpu_mt_shift)) {
712 if (__pcpu_sigp_relax(address, SIGP_SENSE, 0) ==
713 SIGP_CC_NOT_OPERATIONAL)
715 info->core[info->configured].core_id =
716 address >> smp_cpu_mt_shift;
719 info->combined = info->configured;
723 static int smp_add_present_cpu(int cpu);
725 static int __smp_rescan_cpus(struct sclp_core_info *info, int sysfs_add)
733 cpumask_xor(&avail, cpu_possible_mask, cpu_present_mask);
734 cpu = cpumask_first(&avail);
735 for (i = 0; (i < info->combined) && (cpu < nr_cpu_ids); i++) {
736 if (sclp.has_core_type && info->core[i].type != boot_core_type)
738 address = info->core[i].core_id << smp_cpu_mt_shift;
739 for (j = 0; j <= smp_cpu_mtid; j++) {
740 if (pcpu_find_address(cpu_present_mask, address + j))
742 pcpu = pcpu_devices + cpu;
743 pcpu->address = address + j;
745 (cpu >= info->configured*(smp_cpu_mtid + 1)) ?
746 CPU_STATE_STANDBY : CPU_STATE_CONFIGURED;
747 smp_cpu_set_polarization(cpu, POLARIZATION_UNKNOWN);
748 set_cpu_present(cpu, true);
749 if (sysfs_add && smp_add_present_cpu(cpu) != 0)
750 set_cpu_present(cpu, false);
753 cpu = cpumask_next(cpu, &avail);
754 if (cpu >= nr_cpu_ids)
761 void __init smp_detect_cpus(void)
763 unsigned int cpu, mtid, c_cpus, s_cpus;
764 struct sclp_core_info *info;
767 /* Get CPU information */
768 info = memblock_alloc(sizeof(*info), 8);
769 smp_get_core_info(info, 1);
770 /* Find boot CPU type */
771 if (sclp.has_core_type) {
773 for (cpu = 0; cpu < info->combined; cpu++)
774 if (info->core[cpu].core_id == address) {
775 /* The boot cpu dictates the cpu type. */
776 boot_core_type = info->core[cpu].type;
779 if (cpu >= info->combined)
780 panic("Could not find boot CPU type");
783 /* Set multi-threading state for the current system */
784 mtid = boot_core_type ? sclp.mtid : sclp.mtid_cp;
785 mtid = (mtid < smp_max_threads) ? mtid : smp_max_threads - 1;
788 /* Print number of CPUs */
790 for (cpu = 0; cpu < info->combined; cpu++) {
791 if (sclp.has_core_type &&
792 info->core[cpu].type != boot_core_type)
794 if (cpu < info->configured)
795 c_cpus += smp_cpu_mtid + 1;
797 s_cpus += smp_cpu_mtid + 1;
799 pr_info("%d configured CPUs, %d standby CPUs\n", c_cpus, s_cpus);
801 /* Add CPUs present at boot */
803 __smp_rescan_cpus(info, 0);
805 memblock_free_early((unsigned long)info, sizeof(*info));
808 static void smp_init_secondary(void)
810 int cpu = smp_processor_id();
812 S390_lowcore.last_update_clock = get_tod_clock();
813 restore_access_regs(S390_lowcore.access_regs_save_area);
819 notify_cpu_starting(smp_processor_id());
820 if (topology_cpu_dedicated(cpu))
821 set_cpu_flag(CIF_DEDICATED_CPU);
823 clear_cpu_flag(CIF_DEDICATED_CPU);
824 set_cpu_online(smp_processor_id(), true);
825 inc_irq_stat(CPU_RST);
827 cpu_startup_entry(CPUHP_AP_ONLINE_IDLE);
831 * Activate a secondary processor.
833 static void __no_sanitize_address smp_start_secondary(void *cpuvoid)
835 S390_lowcore.restart_stack = (unsigned long) restart_stack;
836 S390_lowcore.restart_fn = (unsigned long) do_restart;
837 S390_lowcore.restart_data = 0;
838 S390_lowcore.restart_source = -1UL;
839 __ctl_load(S390_lowcore.cregs_save_area, 0, 15);
840 __load_psw_mask(PSW_KERNEL_BITS | PSW_MASK_DAT);
841 CALL_ON_STACK(smp_init_secondary, S390_lowcore.kernel_stack, 0);
844 /* Upping and downing of CPUs */
845 int __cpu_up(unsigned int cpu, struct task_struct *tidle)
850 pcpu = pcpu_devices + cpu;
851 if (pcpu->state != CPU_STATE_CONFIGURED)
853 base = smp_get_base_cpu(cpu);
854 for (i = 0; i <= smp_cpu_mtid; i++) {
855 if (base + i < nr_cpu_ids)
856 if (cpu_online(base + i))
860 * If this is the first CPU of the core to get online
861 * do an initial CPU reset.
863 if (i > smp_cpu_mtid &&
864 pcpu_sigp_retry(pcpu_devices + base, SIGP_INITIAL_CPU_RESET, 0) !=
865 SIGP_CC_ORDER_CODE_ACCEPTED)
868 rc = pcpu_alloc_lowcore(pcpu, cpu);
871 pcpu_prepare_secondary(pcpu, cpu);
872 pcpu_attach_task(pcpu, tidle);
873 pcpu_start_fn(pcpu, smp_start_secondary, NULL);
874 /* Wait until cpu puts itself in the online & active maps */
875 while (!cpu_online(cpu))
880 static unsigned int setup_possible_cpus __initdata;
882 static int __init _setup_possible_cpus(char *s)
884 get_option(&s, &setup_possible_cpus);
887 early_param("possible_cpus", _setup_possible_cpus);
889 #ifdef CONFIG_HOTPLUG_CPU
891 int __cpu_disable(void)
893 unsigned long cregs[16];
895 /* Handle possible pending IPIs */
896 smp_handle_ext_call();
897 set_cpu_online(smp_processor_id(), false);
898 /* Disable pseudo page faults on this cpu. */
900 /* Disable interrupt sources via control register. */
901 __ctl_store(cregs, 0, 15);
902 cregs[0] &= ~0x0000ee70UL; /* disable all external interrupts */
903 cregs[6] &= ~0xff000000UL; /* disable all I/O interrupts */
904 cregs[14] &= ~0x1f000000UL; /* disable most machine checks */
905 __ctl_load(cregs, 0, 15);
906 clear_cpu_flag(CIF_NOHZ_DELAY);
910 void __cpu_die(unsigned int cpu)
914 /* Wait until target cpu is down */
915 pcpu = pcpu_devices + cpu;
916 while (!pcpu_stopped(pcpu))
918 pcpu_free_lowcore(pcpu);
919 cpumask_clear_cpu(cpu, mm_cpumask(&init_mm));
920 cpumask_clear_cpu(cpu, &init_mm.context.cpu_attach_mask);
923 void __noreturn cpu_die(void)
927 pcpu_sigp_retry(pcpu_devices + smp_processor_id(), SIGP_STOP, 0);
931 #endif /* CONFIG_HOTPLUG_CPU */
933 void __init smp_fill_possible_mask(void)
935 unsigned int possible, sclp_max, cpu;
937 sclp_max = max(sclp.mtid, sclp.mtid_cp) + 1;
938 sclp_max = min(smp_max_threads, sclp_max);
939 sclp_max = (sclp.max_cores * sclp_max) ?: nr_cpu_ids;
940 possible = setup_possible_cpus ?: nr_cpu_ids;
941 possible = min(possible, sclp_max);
942 for (cpu = 0; cpu < possible && cpu < nr_cpu_ids; cpu++)
943 set_cpu_possible(cpu, true);
946 void __init smp_prepare_cpus(unsigned int max_cpus)
948 /* request the 0x1201 emergency signal external interrupt */
949 if (register_external_irq(EXT_IRQ_EMERGENCY_SIG, do_ext_call_interrupt))
950 panic("Couldn't request external interrupt 0x1201");
951 /* request the 0x1202 external call external interrupt */
952 if (register_external_irq(EXT_IRQ_EXTERNAL_CALL, do_ext_call_interrupt))
953 panic("Couldn't request external interrupt 0x1202");
956 void __init smp_prepare_boot_cpu(void)
958 struct pcpu *pcpu = pcpu_devices;
960 WARN_ON(!cpu_present(0) || !cpu_online(0));
961 pcpu->state = CPU_STATE_CONFIGURED;
962 pcpu->lowcore = (struct lowcore *)(unsigned long) store_prefix();
963 S390_lowcore.percpu_offset = __per_cpu_offset[0];
964 smp_cpu_set_polarization(0, POLARIZATION_UNKNOWN);
967 void __init smp_cpus_done(unsigned int max_cpus)
971 void __init smp_setup_processor_id(void)
973 pcpu_devices[0].address = stap();
974 S390_lowcore.cpu_nr = 0;
975 S390_lowcore.spinlock_lockval = arch_spin_lockval(0);
976 S390_lowcore.spinlock_index = 0;
980 * the frequency of the profiling timer can be changed
981 * by writing a multiplier value into /proc/profile.
983 * usually you want to run this on all CPUs ;)
985 int setup_profiling_timer(unsigned int multiplier)
990 #ifdef CONFIG_HOTPLUG_CPU
991 static ssize_t cpu_configure_show(struct device *dev,
992 struct device_attribute *attr, char *buf)
996 mutex_lock(&smp_cpu_state_mutex);
997 count = sprintf(buf, "%d\n", pcpu_devices[dev->id].state);
998 mutex_unlock(&smp_cpu_state_mutex);
1002 static ssize_t cpu_configure_store(struct device *dev,
1003 struct device_attribute *attr,
1004 const char *buf, size_t count)
1007 int cpu, val, rc, i;
1010 if (sscanf(buf, "%d %c", &val, &delim) != 1)
1012 if (val != 0 && val != 1)
1015 mutex_lock(&smp_cpu_state_mutex);
1017 /* disallow configuration changes of online cpus and cpu 0 */
1019 cpu = smp_get_base_cpu(cpu);
1022 for (i = 0; i <= smp_cpu_mtid; i++)
1023 if (cpu_online(cpu + i))
1025 pcpu = pcpu_devices + cpu;
1029 if (pcpu->state != CPU_STATE_CONFIGURED)
1031 rc = sclp_core_deconfigure(pcpu->address >> smp_cpu_mt_shift);
1034 for (i = 0; i <= smp_cpu_mtid; i++) {
1035 if (cpu + i >= nr_cpu_ids || !cpu_present(cpu + i))
1037 pcpu[i].state = CPU_STATE_STANDBY;
1038 smp_cpu_set_polarization(cpu + i,
1039 POLARIZATION_UNKNOWN);
1041 topology_expect_change();
1044 if (pcpu->state != CPU_STATE_STANDBY)
1046 rc = sclp_core_configure(pcpu->address >> smp_cpu_mt_shift);
1049 for (i = 0; i <= smp_cpu_mtid; i++) {
1050 if (cpu + i >= nr_cpu_ids || !cpu_present(cpu + i))
1052 pcpu[i].state = CPU_STATE_CONFIGURED;
1053 smp_cpu_set_polarization(cpu + i,
1054 POLARIZATION_UNKNOWN);
1056 topology_expect_change();
1062 mutex_unlock(&smp_cpu_state_mutex);
1064 return rc ? rc : count;
1066 static DEVICE_ATTR(configure, 0644, cpu_configure_show, cpu_configure_store);
1067 #endif /* CONFIG_HOTPLUG_CPU */
1069 static ssize_t show_cpu_address(struct device *dev,
1070 struct device_attribute *attr, char *buf)
1072 return sprintf(buf, "%d\n", pcpu_devices[dev->id].address);
1074 static DEVICE_ATTR(address, 0444, show_cpu_address, NULL);
1076 static struct attribute *cpu_common_attrs[] = {
1077 #ifdef CONFIG_HOTPLUG_CPU
1078 &dev_attr_configure.attr,
1080 &dev_attr_address.attr,
1084 static struct attribute_group cpu_common_attr_group = {
1085 .attrs = cpu_common_attrs,
1088 static struct attribute *cpu_online_attrs[] = {
1089 &dev_attr_idle_count.attr,
1090 &dev_attr_idle_time_us.attr,
1094 static struct attribute_group cpu_online_attr_group = {
1095 .attrs = cpu_online_attrs,
1098 static int smp_cpu_online(unsigned int cpu)
1100 struct device *s = &per_cpu(cpu_device, cpu)->dev;
1102 return sysfs_create_group(&s->kobj, &cpu_online_attr_group);
1104 static int smp_cpu_pre_down(unsigned int cpu)
1106 struct device *s = &per_cpu(cpu_device, cpu)->dev;
1108 sysfs_remove_group(&s->kobj, &cpu_online_attr_group);
1112 static int smp_add_present_cpu(int cpu)
1118 c = kzalloc(sizeof(*c), GFP_KERNEL);
1121 per_cpu(cpu_device, cpu) = c;
1123 c->hotpluggable = 1;
1124 rc = register_cpu(c, cpu);
1127 rc = sysfs_create_group(&s->kobj, &cpu_common_attr_group);
1130 rc = topology_cpu_init(c);
1136 sysfs_remove_group(&s->kobj, &cpu_common_attr_group);
1138 #ifdef CONFIG_HOTPLUG_CPU
1145 #ifdef CONFIG_HOTPLUG_CPU
1147 int __ref smp_rescan_cpus(void)
1149 struct sclp_core_info *info;
1152 info = kzalloc(sizeof(*info), GFP_KERNEL);
1155 smp_get_core_info(info, 0);
1157 mutex_lock(&smp_cpu_state_mutex);
1158 nr = __smp_rescan_cpus(info, 1);
1159 mutex_unlock(&smp_cpu_state_mutex);
1163 topology_schedule_update();
1167 static ssize_t __ref rescan_store(struct device *dev,
1168 struct device_attribute *attr,
1174 rc = lock_device_hotplug_sysfs();
1177 rc = smp_rescan_cpus();
1178 unlock_device_hotplug();
1179 return rc ? rc : count;
1181 static DEVICE_ATTR_WO(rescan);
1182 #endif /* CONFIG_HOTPLUG_CPU */
1184 static int __init s390_smp_init(void)
1188 #ifdef CONFIG_HOTPLUG_CPU
1189 rc = device_create_file(cpu_subsys.dev_root, &dev_attr_rescan);
1193 for_each_present_cpu(cpu) {
1194 rc = smp_add_present_cpu(cpu);
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;
1205 subsys_initcall(s390_smp_init);