2 * SMP related functions
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>,
9 * based on other smp stuff by
10 * (c) 1995 Alan Cox, CymruNET Ltd <alan@cymru.net>
11 * (c) 1998 Ingo Molnar
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.
18 #define KMSG_COMPONENT "cpu"
19 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
21 #include <linux/workqueue.h>
22 #include <linux/module.h>
23 #include <linux/init.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>
38 #include <asm/setup.h>
40 #include <asm/tlbflush.h>
41 #include <asm/vtimer.h>
42 #include <asm/lowcore.h>
45 #include <asm/debug.h>
46 #include <asm/os_info.h>
53 ec_call_function_single,
62 static DEFINE_PER_CPU(struct cpu *, cpu_device);
65 struct _lowcore *lowcore; /* lowcore page(s) for the cpu */
66 unsigned long ec_mask; /* bit mask for ec_xxx functions */
67 signed char state; /* physical cpu state */
68 signed char polarization; /* physical polarization */
69 u16 address; /* physical cpu address */
72 static u8 boot_cpu_type;
73 static struct pcpu pcpu_devices[NR_CPUS];
75 unsigned int smp_cpu_mt_shift;
76 EXPORT_SYMBOL(smp_cpu_mt_shift);
78 unsigned int smp_cpu_mtid;
79 EXPORT_SYMBOL(smp_cpu_mtid);
81 static unsigned int smp_max_threads __initdata = -1U;
83 static int __init early_nosmt(char *s)
88 early_param("nosmt", early_nosmt);
90 static int __init early_smt(char *s)
92 get_option(&s, &smp_max_threads);
95 early_param("smt", early_smt);
98 * The smp_cpu_state_mutex must be held when changing the state or polarization
99 * member of a pcpu data structure within the pcpu_devices arreay.
101 DEFINE_MUTEX(smp_cpu_state_mutex);
104 * Signal processor helper functions.
106 static inline int __pcpu_sigp_relax(u16 addr, u8 order, unsigned long parm,
112 cc = __pcpu_sigp(addr, order, parm, NULL);
113 if (cc != SIGP_CC_BUSY)
119 static int pcpu_sigp_retry(struct pcpu *pcpu, u8 order, u32 parm)
123 for (retry = 0; ; retry++) {
124 cc = __pcpu_sigp(pcpu->address, order, parm, NULL);
125 if (cc != SIGP_CC_BUSY)
133 static inline int pcpu_stopped(struct pcpu *pcpu)
135 u32 uninitialized_var(status);
137 if (__pcpu_sigp(pcpu->address, SIGP_SENSE,
138 0, &status) != SIGP_CC_STATUS_STORED)
140 return !!(status & (SIGP_STATUS_CHECK_STOP|SIGP_STATUS_STOPPED));
143 static inline int pcpu_running(struct pcpu *pcpu)
145 if (__pcpu_sigp(pcpu->address, SIGP_SENSE_RUNNING,
146 0, NULL) != SIGP_CC_STATUS_STORED)
148 /* Status stored condition code is equivalent to cpu not running. */
153 * Find struct pcpu by cpu address.
155 static struct pcpu *pcpu_find_address(const struct cpumask *mask, u16 address)
159 for_each_cpu(cpu, mask)
160 if (pcpu_devices[cpu].address == address)
161 return pcpu_devices + cpu;
165 static void pcpu_ec_call(struct pcpu *pcpu, int ec_bit)
169 if (test_and_set_bit(ec_bit, &pcpu->ec_mask))
171 order = pcpu_running(pcpu) ? SIGP_EXTERNAL_CALL : SIGP_EMERGENCY_SIGNAL;
172 pcpu_sigp_retry(pcpu, order, 0);
175 #define ASYNC_FRAME_OFFSET (ASYNC_SIZE - STACK_FRAME_OVERHEAD - __PT_SIZE)
176 #define PANIC_FRAME_OFFSET (PAGE_SIZE - STACK_FRAME_OVERHEAD - __PT_SIZE)
178 static int pcpu_alloc_lowcore(struct pcpu *pcpu, int cpu)
180 unsigned long async_stack, panic_stack;
183 if (pcpu != &pcpu_devices[0]) {
184 pcpu->lowcore = (struct _lowcore *)
185 __get_free_pages(GFP_KERNEL | GFP_DMA, LC_ORDER);
186 async_stack = __get_free_pages(GFP_KERNEL, ASYNC_ORDER);
187 panic_stack = __get_free_page(GFP_KERNEL);
188 if (!pcpu->lowcore || !panic_stack || !async_stack)
191 async_stack = pcpu->lowcore->async_stack - ASYNC_FRAME_OFFSET;
192 panic_stack = pcpu->lowcore->panic_stack - PANIC_FRAME_OFFSET;
195 memcpy(lc, &S390_lowcore, 512);
196 memset((char *) lc + 512, 0, sizeof(*lc) - 512);
197 lc->async_stack = async_stack + ASYNC_FRAME_OFFSET;
198 lc->panic_stack = panic_stack + PANIC_FRAME_OFFSET;
200 lc->spinlock_lockval = arch_spin_lockval(cpu);
202 if (MACHINE_HAS_IEEE) {
203 lc->extended_save_area_addr = get_zeroed_page(GFP_KERNEL);
204 if (!lc->extended_save_area_addr)
209 lc->vector_save_area_addr =
210 (unsigned long) &lc->vector_save_area;
211 if (vdso_alloc_per_cpu(lc))
214 lowcore_ptr[cpu] = lc;
215 pcpu_sigp_retry(pcpu, SIGP_SET_PREFIX, (u32)(unsigned long) lc);
218 if (pcpu != &pcpu_devices[0]) {
219 free_page(panic_stack);
220 free_pages(async_stack, ASYNC_ORDER);
221 free_pages((unsigned long) pcpu->lowcore, LC_ORDER);
226 #ifdef CONFIG_HOTPLUG_CPU
228 static void pcpu_free_lowcore(struct pcpu *pcpu)
230 pcpu_sigp_retry(pcpu, SIGP_SET_PREFIX, 0);
231 lowcore_ptr[pcpu - pcpu_devices] = NULL;
233 if (MACHINE_HAS_IEEE) {
234 struct _lowcore *lc = pcpu->lowcore;
236 free_page((unsigned long) lc->extended_save_area_addr);
237 lc->extended_save_area_addr = 0;
240 vdso_free_per_cpu(pcpu->lowcore);
242 if (pcpu == &pcpu_devices[0])
244 free_page(pcpu->lowcore->panic_stack-PANIC_FRAME_OFFSET);
245 free_pages(pcpu->lowcore->async_stack-ASYNC_FRAME_OFFSET, ASYNC_ORDER);
246 free_pages((unsigned long) pcpu->lowcore, LC_ORDER);
249 #endif /* CONFIG_HOTPLUG_CPU */
251 static void pcpu_prepare_secondary(struct pcpu *pcpu, int cpu)
253 struct _lowcore *lc = pcpu->lowcore;
255 if (MACHINE_HAS_TLB_LC)
256 cpumask_set_cpu(cpu, &init_mm.context.cpu_attach_mask);
257 cpumask_set_cpu(cpu, mm_cpumask(&init_mm));
258 atomic_inc(&init_mm.context.attach_count);
260 lc->spinlock_lockval = arch_spin_lockval(cpu);
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,
271 static void pcpu_attach_task(struct pcpu *pcpu, struct task_struct *tsk)
273 struct _lowcore *lc = pcpu->lowcore;
274 struct thread_info *ti = task_thread_info(tsk);
276 lc->kernel_stack = (unsigned long) task_stack_page(tsk)
277 + THREAD_SIZE - STACK_FRAME_OVERHEAD - sizeof(struct pt_regs);
278 lc->thread_info = (unsigned long) task_thread_info(tsk);
279 lc->current_task = (unsigned long) tsk;
280 lc->user_timer = ti->user_timer;
281 lc->system_timer = ti->system_timer;
285 static void pcpu_start_fn(struct pcpu *pcpu, void (*func)(void *), void *data)
287 struct _lowcore *lc = pcpu->lowcore;
289 lc->restart_stack = lc->kernel_stack;
290 lc->restart_fn = (unsigned long) func;
291 lc->restart_data = (unsigned long) data;
292 lc->restart_source = -1UL;
293 pcpu_sigp_retry(pcpu, SIGP_RESTART, 0);
297 * Call function via PSW restart on pcpu and stop the current cpu.
299 static void pcpu_delegate(struct pcpu *pcpu, void (*func)(void *),
300 void *data, unsigned long stack)
302 struct _lowcore *lc = lowcore_ptr[pcpu - pcpu_devices];
303 unsigned long source_cpu = stap();
305 __load_psw_mask(PSW_KERNEL_BITS);
306 if (pcpu->address == source_cpu)
307 func(data); /* should not return */
308 /* Stop target cpu (if func returns this stops the current cpu). */
309 pcpu_sigp_retry(pcpu, SIGP_STOP, 0);
310 /* Restart func on the target cpu and stop the current cpu. */
311 mem_assign_absolute(lc->restart_stack, stack);
312 mem_assign_absolute(lc->restart_fn, (unsigned long) func);
313 mem_assign_absolute(lc->restart_data, (unsigned long) data);
314 mem_assign_absolute(lc->restart_source, source_cpu);
316 "0: sigp 0,%0,%2 # sigp restart to target cpu\n"
317 " brc 2,0b # busy, try again\n"
318 "1: sigp 0,%1,%3 # sigp stop to current cpu\n"
319 " brc 2,1b # busy, try again\n"
320 : : "d" (pcpu->address), "d" (source_cpu),
321 "K" (SIGP_RESTART), "K" (SIGP_STOP)
327 * Enable additional logical cpus for multi-threading.
329 static int pcpu_set_smt(unsigned int mtid)
331 register unsigned long reg1 asm ("1") = (unsigned long) mtid;
334 if (smp_cpu_mtid == mtid)
337 " sigp %1,0,%2 # sigp set multi-threading\n"
340 : "=d" (cc) : "d" (reg1), "K" (SIGP_SET_MULTI_THREADING)
344 smp_cpu_mt_shift = 0;
345 while (smp_cpu_mtid >= (1U << smp_cpu_mt_shift))
347 pcpu_devices[0].address = stap();
353 * Call function on an online CPU.
355 void smp_call_online_cpu(void (*func)(void *), void *data)
359 /* Use the current cpu if it is online. */
360 pcpu = pcpu_find_address(cpu_online_mask, stap());
362 /* Use the first online cpu. */
363 pcpu = pcpu_devices + cpumask_first(cpu_online_mask);
364 pcpu_delegate(pcpu, func, data, (unsigned long) restart_stack);
368 * Call function on the ipl CPU.
370 void smp_call_ipl_cpu(void (*func)(void *), void *data)
372 pcpu_delegate(&pcpu_devices[0], func, data,
373 pcpu_devices->lowcore->panic_stack -
374 PANIC_FRAME_OFFSET + PAGE_SIZE);
377 int smp_find_processor_id(u16 address)
381 for_each_present_cpu(cpu)
382 if (pcpu_devices[cpu].address == address)
387 int smp_vcpu_scheduled(int cpu)
389 return pcpu_running(pcpu_devices + cpu);
392 void smp_yield_cpu(int cpu)
394 if (MACHINE_HAS_DIAG9C)
395 asm volatile("diag %0,0,0x9c"
396 : : "d" (pcpu_devices[cpu].address));
397 else if (MACHINE_HAS_DIAG44)
398 asm volatile("diag 0,0,0x44");
402 * Send cpus emergency shutdown signal. This gives the cpus the
403 * opportunity to complete outstanding interrupts.
405 static void smp_emergency_stop(cpumask_t *cpumask)
410 end = get_tod_clock() + (1000000UL << 12);
411 for_each_cpu(cpu, cpumask) {
412 struct pcpu *pcpu = pcpu_devices + cpu;
413 set_bit(ec_stop_cpu, &pcpu->ec_mask);
414 while (__pcpu_sigp(pcpu->address, SIGP_EMERGENCY_SIGNAL,
415 0, NULL) == SIGP_CC_BUSY &&
416 get_tod_clock() < end)
419 while (get_tod_clock() < end) {
420 for_each_cpu(cpu, cpumask)
421 if (pcpu_stopped(pcpu_devices + cpu))
422 cpumask_clear_cpu(cpu, cpumask);
423 if (cpumask_empty(cpumask))
430 * Stop all cpus but the current one.
432 void smp_send_stop(void)
437 /* Disable all interrupts/machine checks */
438 __load_psw_mask(PSW_KERNEL_BITS | PSW_MASK_DAT);
439 trace_hardirqs_off();
441 debug_set_critical();
442 cpumask_copy(&cpumask, cpu_online_mask);
443 cpumask_clear_cpu(smp_processor_id(), &cpumask);
445 if (oops_in_progress)
446 smp_emergency_stop(&cpumask);
448 /* stop all processors */
449 for_each_cpu(cpu, &cpumask) {
450 struct pcpu *pcpu = pcpu_devices + cpu;
451 pcpu_sigp_retry(pcpu, SIGP_STOP, 0);
452 while (!pcpu_stopped(pcpu))
458 * This is the main routine where commands issued by other
461 static void smp_handle_ext_call(void)
465 /* handle bit signal external calls */
466 bits = xchg(&pcpu_devices[smp_processor_id()].ec_mask, 0);
467 if (test_bit(ec_stop_cpu, &bits))
469 if (test_bit(ec_schedule, &bits))
471 if (test_bit(ec_call_function_single, &bits))
472 generic_smp_call_function_single_interrupt();
475 static void do_ext_call_interrupt(struct ext_code ext_code,
476 unsigned int param32, unsigned long param64)
478 inc_irq_stat(ext_code.code == 0x1202 ? IRQEXT_EXC : IRQEXT_EMS);
479 smp_handle_ext_call();
482 void arch_send_call_function_ipi_mask(const struct cpumask *mask)
486 for_each_cpu(cpu, mask)
487 pcpu_ec_call(pcpu_devices + cpu, ec_call_function_single);
490 void arch_send_call_function_single_ipi(int cpu)
492 pcpu_ec_call(pcpu_devices + cpu, ec_call_function_single);
497 * this function sends a 'purge tlb' signal to another CPU.
499 static void smp_ptlb_callback(void *info)
504 void smp_ptlb_all(void)
506 on_each_cpu(smp_ptlb_callback, NULL, 1);
508 EXPORT_SYMBOL(smp_ptlb_all);
509 #endif /* ! CONFIG_64BIT */
512 * this function sends a 'reschedule' IPI to another CPU.
513 * it goes straight through and wastes no time serializing
514 * anything. Worst case is that we lose a reschedule ...
516 void smp_send_reschedule(int cpu)
518 pcpu_ec_call(pcpu_devices + cpu, ec_schedule);
522 * parameter area for the set/clear control bit callbacks
524 struct ec_creg_mask_parms {
526 unsigned long andval;
531 * callback for setting/clearing control bits
533 static void smp_ctl_bit_callback(void *info)
535 struct ec_creg_mask_parms *pp = info;
536 unsigned long cregs[16];
538 __ctl_store(cregs, 0, 15);
539 cregs[pp->cr] = (cregs[pp->cr] & pp->andval) | pp->orval;
540 __ctl_load(cregs, 0, 15);
544 * Set a bit in a control register of all cpus
546 void smp_ctl_set_bit(int cr, int bit)
548 struct ec_creg_mask_parms parms = { 1UL << bit, -1UL, cr };
550 on_each_cpu(smp_ctl_bit_callback, &parms, 1);
552 EXPORT_SYMBOL(smp_ctl_set_bit);
555 * Clear a bit in a control register of all cpus
557 void smp_ctl_clear_bit(int cr, int bit)
559 struct ec_creg_mask_parms parms = { 0, ~(1UL << bit), cr };
561 on_each_cpu(smp_ctl_bit_callback, &parms, 1);
563 EXPORT_SYMBOL(smp_ctl_clear_bit);
565 #ifdef CONFIG_CRASH_DUMP
567 static inline void __smp_store_cpu_state(int cpu, u16 address, int is_boot_cpu)
569 void *lc = pcpu_devices[0].lowcore;
570 struct save_area_ext *sa_ext;
573 sa_ext = dump_save_area_create(cpu);
575 panic("could not allocate memory for save area\n");
577 /* Copy the registers of the boot CPU. */
578 copy_oldmem_page(1, (void *) &sa_ext->sa, sizeof(sa_ext->sa),
579 SAVE_AREA_BASE - PAGE_SIZE, 0);
581 save_vx_regs_safe(sa_ext->vx_regs);
584 /* Get the registers of a non-boot cpu. */
585 __pcpu_sigp_relax(address, SIGP_STOP_AND_STORE_STATUS, 0, NULL);
586 memcpy_real(&sa_ext->sa, lc + SAVE_AREA_BASE, sizeof(sa_ext->sa));
589 /* Get the VX registers */
590 vx_sa = __get_free_page(GFP_KERNEL);
592 panic("could not allocate memory for VX save area\n");
593 __pcpu_sigp_relax(address, SIGP_STORE_ADDITIONAL_STATUS, vx_sa, NULL);
594 memcpy(sa_ext->vx_regs, (void *) vx_sa, sizeof(sa_ext->vx_regs));
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 allocate the save area and 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 * The state of all CPUs is stored in ELF sections in the memory of the
622 * old system. The ELF sections are picked up by the crash_dump code
623 * via elfcorehdr_addr.
625 static void __init smp_store_cpu_states(struct sclp_cpu_info *info)
627 unsigned int cpu, address, i, j;
630 if (is_kdump_kernel())
631 /* Previous system stored the CPU states. Nothing to do. */
633 if (!(OLDMEM_BASE || ipl_info.type == IPL_TYPE_FCP_DUMP))
634 /* No previous system present, normal boot. */
636 /* Set multi-threading state to the previous system. */
637 pcpu_set_smt(sclp_get_mtid_prev());
638 /* Collect CPU states. */
640 for (i = 0; i < info->configured; i++) {
641 /* Skip CPUs with different CPU type. */
642 if (info->has_cpu_type && info->cpu[i].type != boot_cpu_type)
644 for (j = 0; j <= smp_cpu_mtid; j++, cpu++) {
645 address = (info->cpu[i].core_id << smp_cpu_mt_shift) + j;
646 is_boot_cpu = (address == pcpu_devices[0].address);
647 if (is_boot_cpu && !OLDMEM_BASE)
648 /* Skip boot CPU for standard zfcp dump. */
650 /* Get state for this CPu. */
651 __smp_store_cpu_state(cpu, address, is_boot_cpu);
656 int smp_store_status(int cpu)
661 pcpu = pcpu_devices + cpu;
662 if (__pcpu_sigp_relax(pcpu->address, SIGP_STOP_AND_STORE_STATUS,
663 0, NULL) != SIGP_CC_ORDER_CODE_ACCEPTED)
667 vx_sa = __pa(pcpu->lowcore->vector_save_area_addr);
668 __pcpu_sigp_relax(pcpu->address, SIGP_STORE_ADDITIONAL_STATUS,
673 #endif /* CONFIG_CRASH_DUMP */
675 void smp_cpu_set_polarization(int cpu, int val)
677 pcpu_devices[cpu].polarization = val;
680 int smp_cpu_get_polarization(int cpu)
682 return pcpu_devices[cpu].polarization;
685 static struct sclp_cpu_info *smp_get_cpu_info(void)
687 static int use_sigp_detection;
688 struct sclp_cpu_info *info;
691 info = kzalloc(sizeof(*info), GFP_KERNEL);
692 if (info && (use_sigp_detection || sclp_get_cpu_info(info))) {
693 use_sigp_detection = 1;
694 for (address = 0; address <= MAX_CPU_ADDRESS;
695 address += (1U << smp_cpu_mt_shift)) {
696 if (__pcpu_sigp_relax(address, SIGP_SENSE, 0, NULL) ==
697 SIGP_CC_NOT_OPERATIONAL)
699 info->cpu[info->configured].core_id =
700 address >> smp_cpu_mt_shift;
703 info->combined = info->configured;
708 static int smp_add_present_cpu(int cpu);
710 static int __smp_rescan_cpus(struct sclp_cpu_info *info, int sysfs_add)
718 cpumask_xor(&avail, cpu_possible_mask, cpu_present_mask);
719 cpu = cpumask_first(&avail);
720 for (i = 0; (i < info->combined) && (cpu < nr_cpu_ids); i++) {
721 if (info->has_cpu_type && info->cpu[i].type != boot_cpu_type)
723 address = info->cpu[i].core_id << smp_cpu_mt_shift;
724 for (j = 0; j <= smp_cpu_mtid; j++) {
725 if (pcpu_find_address(cpu_present_mask, address + j))
727 pcpu = pcpu_devices + cpu;
728 pcpu->address = address + j;
730 (cpu >= info->configured*(smp_cpu_mtid + 1)) ?
731 CPU_STATE_STANDBY : CPU_STATE_CONFIGURED;
732 smp_cpu_set_polarization(cpu, POLARIZATION_UNKNOWN);
733 set_cpu_present(cpu, true);
734 if (sysfs_add && smp_add_present_cpu(cpu) != 0)
735 set_cpu_present(cpu, false);
738 cpu = cpumask_next(cpu, &avail);
739 if (cpu >= nr_cpu_ids)
746 static void __init smp_detect_cpus(void)
748 unsigned int cpu, mtid, c_cpus, s_cpus;
749 struct sclp_cpu_info *info;
752 /* Get CPU information */
753 info = smp_get_cpu_info();
755 panic("smp_detect_cpus failed to allocate memory\n");
757 /* Find boot CPU type */
758 if (info->has_cpu_type) {
760 for (cpu = 0; cpu < info->combined; cpu++)
761 if (info->cpu[cpu].core_id == address) {
762 /* The boot cpu dictates the cpu type. */
763 boot_cpu_type = info->cpu[cpu].type;
766 if (cpu >= info->combined)
767 panic("Could not find boot CPU type");
770 #ifdef CONFIG_CRASH_DUMP
771 /* Collect CPU state of previous system */
772 smp_store_cpu_states(info);
775 /* Set multi-threading state for the current system */
776 mtid = sclp_get_mtid(boot_cpu_type);
777 mtid = (mtid < smp_max_threads) ? mtid : smp_max_threads - 1;
780 /* Print number of CPUs */
782 for (cpu = 0; cpu < info->combined; cpu++) {
783 if (info->has_cpu_type && info->cpu[cpu].type != boot_cpu_type)
785 if (cpu < info->configured)
786 c_cpus += smp_cpu_mtid + 1;
788 s_cpus += smp_cpu_mtid + 1;
790 pr_info("%d configured CPUs, %d standby CPUs\n", c_cpus, s_cpus);
792 /* Add CPUs present at boot */
794 __smp_rescan_cpus(info, 0);
800 * Activate a secondary processor.
802 static void smp_start_secondary(void *cpuvoid)
804 S390_lowcore.last_update_clock = get_tod_clock();
805 S390_lowcore.restart_stack = (unsigned long) restart_stack;
806 S390_lowcore.restart_fn = (unsigned long) do_restart;
807 S390_lowcore.restart_data = 0;
808 S390_lowcore.restart_source = -1UL;
809 restore_access_regs(S390_lowcore.access_regs_save_area);
810 __ctl_load(S390_lowcore.cregs_save_area, 0, 15);
811 __load_psw_mask(PSW_KERNEL_BITS | PSW_MASK_DAT);
817 notify_cpu_starting(smp_processor_id());
818 set_cpu_online(smp_processor_id(), true);
819 inc_irq_stat(CPU_RST);
821 cpu_startup_entry(CPUHP_ONLINE);
824 /* Upping and downing of CPUs */
825 int __cpu_up(unsigned int cpu, struct task_struct *tidle)
830 pcpu = pcpu_devices + cpu;
831 if (pcpu->state != CPU_STATE_CONFIGURED)
833 base = cpu - (cpu % (smp_cpu_mtid + 1));
834 for (i = 0; i <= smp_cpu_mtid; i++) {
835 if (base + i < nr_cpu_ids)
836 if (cpu_online(base + i))
840 * If this is the first CPU of the core to get online
841 * do an initial CPU reset.
843 if (i > smp_cpu_mtid &&
844 pcpu_sigp_retry(pcpu_devices + base, SIGP_INITIAL_CPU_RESET, 0) !=
845 SIGP_CC_ORDER_CODE_ACCEPTED)
848 rc = pcpu_alloc_lowcore(pcpu, cpu);
851 pcpu_prepare_secondary(pcpu, cpu);
852 pcpu_attach_task(pcpu, tidle);
853 pcpu_start_fn(pcpu, smp_start_secondary, NULL);
854 while (!cpu_online(cpu))
859 static unsigned int setup_possible_cpus __initdata;
861 static int __init _setup_possible_cpus(char *s)
863 get_option(&s, &setup_possible_cpus);
866 early_param("possible_cpus", _setup_possible_cpus);
868 #ifdef CONFIG_HOTPLUG_CPU
870 int __cpu_disable(void)
872 unsigned long cregs[16];
874 /* Handle possible pending IPIs */
875 smp_handle_ext_call();
876 set_cpu_online(smp_processor_id(), false);
877 /* Disable pseudo page faults on this cpu. */
879 /* Disable interrupt sources via control register. */
880 __ctl_store(cregs, 0, 15);
881 cregs[0] &= ~0x0000ee70UL; /* disable all external interrupts */
882 cregs[6] &= ~0xff000000UL; /* disable all I/O interrupts */
883 cregs[14] &= ~0x1f000000UL; /* disable most machine checks */
884 __ctl_load(cregs, 0, 15);
885 clear_cpu_flag(CIF_NOHZ_DELAY);
889 void __cpu_die(unsigned int cpu)
893 /* Wait until target cpu is down */
894 pcpu = pcpu_devices + cpu;
895 while (!pcpu_stopped(pcpu))
897 pcpu_free_lowcore(pcpu);
898 atomic_dec(&init_mm.context.attach_count);
899 cpumask_clear_cpu(cpu, mm_cpumask(&init_mm));
900 if (MACHINE_HAS_TLB_LC)
901 cpumask_clear_cpu(cpu, &init_mm.context.cpu_attach_mask);
904 void __noreturn cpu_die(void)
907 pcpu_sigp_retry(pcpu_devices + smp_processor_id(), SIGP_STOP, 0);
911 #endif /* CONFIG_HOTPLUG_CPU */
913 void __init smp_fill_possible_mask(void)
915 unsigned int possible, sclp, cpu;
917 sclp = min(smp_max_threads, sclp_get_mtid_max() + 1);
918 sclp = sclp_get_max_cpu()*sclp ?: nr_cpu_ids;
919 possible = setup_possible_cpus ?: nr_cpu_ids;
920 possible = min(possible, sclp);
921 for (cpu = 0; cpu < possible && cpu < nr_cpu_ids; cpu++)
922 set_cpu_possible(cpu, true);
925 void __init smp_prepare_cpus(unsigned int max_cpus)
927 /* request the 0x1201 emergency signal external interrupt */
928 if (register_external_irq(EXT_IRQ_EMERGENCY_SIG, do_ext_call_interrupt))
929 panic("Couldn't request external interrupt 0x1201");
930 /* request the 0x1202 external call external interrupt */
931 if (register_external_irq(EXT_IRQ_EXTERNAL_CALL, do_ext_call_interrupt))
932 panic("Couldn't request external interrupt 0x1202");
936 void __init smp_prepare_boot_cpu(void)
938 struct pcpu *pcpu = pcpu_devices;
940 pcpu->state = CPU_STATE_CONFIGURED;
941 pcpu->address = stap();
942 pcpu->lowcore = (struct _lowcore *)(unsigned long) store_prefix();
943 S390_lowcore.percpu_offset = __per_cpu_offset[0];
944 smp_cpu_set_polarization(0, POLARIZATION_UNKNOWN);
945 set_cpu_present(0, true);
946 set_cpu_online(0, true);
949 void __init smp_cpus_done(unsigned int max_cpus)
953 void __init smp_setup_processor_id(void)
955 S390_lowcore.cpu_nr = 0;
956 S390_lowcore.spinlock_lockval = arch_spin_lockval(0);
960 * the frequency of the profiling timer can be changed
961 * by writing a multiplier value into /proc/profile.
963 * usually you want to run this on all CPUs ;)
965 int setup_profiling_timer(unsigned int multiplier)
970 #ifdef CONFIG_HOTPLUG_CPU
971 static ssize_t cpu_configure_show(struct device *dev,
972 struct device_attribute *attr, char *buf)
976 mutex_lock(&smp_cpu_state_mutex);
977 count = sprintf(buf, "%d\n", pcpu_devices[dev->id].state);
978 mutex_unlock(&smp_cpu_state_mutex);
982 static ssize_t cpu_configure_store(struct device *dev,
983 struct device_attribute *attr,
984 const char *buf, size_t count)
990 if (sscanf(buf, "%d %c", &val, &delim) != 1)
992 if (val != 0 && val != 1)
995 mutex_lock(&smp_cpu_state_mutex);
997 /* disallow configuration changes of online cpus and cpu 0 */
999 cpu -= cpu % (smp_cpu_mtid + 1);
1002 for (i = 0; i <= smp_cpu_mtid; i++)
1003 if (cpu_online(cpu + i))
1005 pcpu = pcpu_devices + cpu;
1009 if (pcpu->state != CPU_STATE_CONFIGURED)
1011 rc = sclp_cpu_deconfigure(pcpu->address >> smp_cpu_mt_shift);
1014 for (i = 0; i <= smp_cpu_mtid; i++) {
1015 if (cpu + i >= nr_cpu_ids || !cpu_present(cpu + i))
1017 pcpu[i].state = CPU_STATE_STANDBY;
1018 smp_cpu_set_polarization(cpu + i,
1019 POLARIZATION_UNKNOWN);
1021 topology_expect_change();
1024 if (pcpu->state != CPU_STATE_STANDBY)
1026 rc = sclp_cpu_configure(pcpu->address >> smp_cpu_mt_shift);
1029 for (i = 0; i <= smp_cpu_mtid; i++) {
1030 if (cpu + i >= nr_cpu_ids || !cpu_present(cpu + i))
1032 pcpu[i].state = CPU_STATE_CONFIGURED;
1033 smp_cpu_set_polarization(cpu + i,
1034 POLARIZATION_UNKNOWN);
1036 topology_expect_change();
1042 mutex_unlock(&smp_cpu_state_mutex);
1044 return rc ? rc : count;
1046 static DEVICE_ATTR(configure, 0644, cpu_configure_show, cpu_configure_store);
1047 #endif /* CONFIG_HOTPLUG_CPU */
1049 static ssize_t show_cpu_address(struct device *dev,
1050 struct device_attribute *attr, char *buf)
1052 return sprintf(buf, "%d\n", pcpu_devices[dev->id].address);
1054 static DEVICE_ATTR(address, 0444, show_cpu_address, NULL);
1056 static struct attribute *cpu_common_attrs[] = {
1057 #ifdef CONFIG_HOTPLUG_CPU
1058 &dev_attr_configure.attr,
1060 &dev_attr_address.attr,
1064 static struct attribute_group cpu_common_attr_group = {
1065 .attrs = cpu_common_attrs,
1068 static struct attribute *cpu_online_attrs[] = {
1069 &dev_attr_idle_count.attr,
1070 &dev_attr_idle_time_us.attr,
1074 static struct attribute_group cpu_online_attr_group = {
1075 .attrs = cpu_online_attrs,
1078 static int smp_cpu_notify(struct notifier_block *self, unsigned long action,
1081 unsigned int cpu = (unsigned int)(long)hcpu;
1082 struct device *s = &per_cpu(cpu_device, cpu)->dev;
1085 switch (action & ~CPU_TASKS_FROZEN) {
1087 err = sysfs_create_group(&s->kobj, &cpu_online_attr_group);
1090 sysfs_remove_group(&s->kobj, &cpu_online_attr_group);
1093 return notifier_from_errno(err);
1096 static int smp_add_present_cpu(int cpu)
1102 c = kzalloc(sizeof(*c), GFP_KERNEL);
1105 per_cpu(cpu_device, cpu) = c;
1107 c->hotpluggable = 1;
1108 rc = register_cpu(c, cpu);
1111 rc = sysfs_create_group(&s->kobj, &cpu_common_attr_group);
1114 if (cpu_online(cpu)) {
1115 rc = sysfs_create_group(&s->kobj, &cpu_online_attr_group);
1119 rc = topology_cpu_init(c);
1125 if (cpu_online(cpu))
1126 sysfs_remove_group(&s->kobj, &cpu_online_attr_group);
1128 sysfs_remove_group(&s->kobj, &cpu_common_attr_group);
1130 #ifdef CONFIG_HOTPLUG_CPU
1137 #ifdef CONFIG_HOTPLUG_CPU
1139 int __ref smp_rescan_cpus(void)
1141 struct sclp_cpu_info *info;
1144 info = smp_get_cpu_info();
1148 mutex_lock(&smp_cpu_state_mutex);
1149 nr = __smp_rescan_cpus(info, 1);
1150 mutex_unlock(&smp_cpu_state_mutex);
1154 topology_schedule_update();
1158 static ssize_t __ref rescan_store(struct device *dev,
1159 struct device_attribute *attr,
1165 rc = smp_rescan_cpus();
1166 return rc ? rc : count;
1168 static DEVICE_ATTR(rescan, 0200, NULL, rescan_store);
1169 #endif /* CONFIG_HOTPLUG_CPU */
1171 static int __init s390_smp_init(void)
1175 #ifdef CONFIG_HOTPLUG_CPU
1176 rc = device_create_file(cpu_subsys.dev_root, &dev_attr_rescan);
1180 cpu_notifier_register_begin();
1181 for_each_present_cpu(cpu) {
1182 rc = smp_add_present_cpu(cpu);
1187 __hotcpu_notifier(smp_cpu_notify, 0);
1190 cpu_notifier_register_done();
1193 subsys_initcall(s390_smp_init);