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/bootmem.h>
23 #include <linux/module.h>
24 #include <linux/init.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/cpu.h>
33 #include <linux/slab.h>
34 #include <linux/crash_dump.h>
35 #include <linux/memblock.h>
36 #include <asm/asm-offsets.h>
38 #include <asm/switch_to.h>
39 #include <asm/facility.h>
41 #include <asm/setup.h>
43 #include <asm/tlbflush.h>
44 #include <asm/vtimer.h>
45 #include <asm/lowcore.h>
48 #include <asm/debug.h>
49 #include <asm/os_info.h>
56 ec_call_function_single,
65 static DEFINE_PER_CPU(struct cpu *, cpu_device);
68 struct lowcore *lowcore; /* lowcore page(s) for the cpu */
69 unsigned long ec_mask; /* bit mask for ec_xxx functions */
70 unsigned long ec_clk; /* sigp timestamp for ec_xxx */
71 signed char state; /* physical cpu state */
72 signed char polarization; /* physical polarization */
73 u16 address; /* physical cpu address */
76 static u8 boot_core_type;
77 static struct pcpu pcpu_devices[NR_CPUS];
79 unsigned int smp_cpu_mt_shift;
80 EXPORT_SYMBOL(smp_cpu_mt_shift);
82 unsigned int smp_cpu_mtid;
83 EXPORT_SYMBOL(smp_cpu_mtid);
85 #ifdef CONFIG_CRASH_DUMP
86 __vector128 __initdata boot_cpu_vector_save_area[__NUM_VXRS];
89 static unsigned int smp_max_threads __initdata = -1U;
91 static int __init early_nosmt(char *s)
96 early_param("nosmt", early_nosmt);
98 static int __init early_smt(char *s)
100 get_option(&s, &smp_max_threads);
103 early_param("smt", early_smt);
106 * The smp_cpu_state_mutex must be held when changing the state or polarization
107 * member of a pcpu data structure within the pcpu_devices arreay.
109 DEFINE_MUTEX(smp_cpu_state_mutex);
112 * Signal processor helper functions.
114 static inline int __pcpu_sigp_relax(u16 addr, u8 order, unsigned long parm)
119 cc = __pcpu_sigp(addr, order, parm, NULL);
120 if (cc != SIGP_CC_BUSY)
126 static int pcpu_sigp_retry(struct pcpu *pcpu, u8 order, u32 parm)
130 for (retry = 0; ; retry++) {
131 cc = __pcpu_sigp(pcpu->address, order, parm, NULL);
132 if (cc != SIGP_CC_BUSY)
140 static inline int pcpu_stopped(struct pcpu *pcpu)
142 u32 uninitialized_var(status);
144 if (__pcpu_sigp(pcpu->address, SIGP_SENSE,
145 0, &status) != SIGP_CC_STATUS_STORED)
147 return !!(status & (SIGP_STATUS_CHECK_STOP|SIGP_STATUS_STOPPED));
150 static inline int pcpu_running(struct pcpu *pcpu)
152 if (__pcpu_sigp(pcpu->address, SIGP_SENSE_RUNNING,
153 0, NULL) != SIGP_CC_STATUS_STORED)
155 /* Status stored condition code is equivalent to cpu not running. */
160 * Find struct pcpu by cpu address.
162 static struct pcpu *pcpu_find_address(const struct cpumask *mask, u16 address)
166 for_each_cpu(cpu, mask)
167 if (pcpu_devices[cpu].address == address)
168 return pcpu_devices + cpu;
172 static void pcpu_ec_call(struct pcpu *pcpu, int ec_bit)
176 if (test_and_set_bit(ec_bit, &pcpu->ec_mask))
178 order = pcpu_running(pcpu) ? SIGP_EXTERNAL_CALL : SIGP_EMERGENCY_SIGNAL;
179 pcpu->ec_clk = get_tod_clock_fast();
180 pcpu_sigp_retry(pcpu, order, 0);
183 #define ASYNC_FRAME_OFFSET (ASYNC_SIZE - STACK_FRAME_OVERHEAD - __PT_SIZE)
184 #define PANIC_FRAME_OFFSET (PAGE_SIZE - STACK_FRAME_OVERHEAD - __PT_SIZE)
186 static int pcpu_alloc_lowcore(struct pcpu *pcpu, int cpu)
188 unsigned long async_stack, panic_stack;
191 if (pcpu != &pcpu_devices[0]) {
192 pcpu->lowcore = (struct lowcore *)
193 __get_free_pages(GFP_KERNEL | GFP_DMA, LC_ORDER);
194 async_stack = __get_free_pages(GFP_KERNEL, ASYNC_ORDER);
195 panic_stack = __get_free_page(GFP_KERNEL);
196 if (!pcpu->lowcore || !panic_stack || !async_stack)
199 async_stack = pcpu->lowcore->async_stack - ASYNC_FRAME_OFFSET;
200 panic_stack = pcpu->lowcore->panic_stack - PANIC_FRAME_OFFSET;
203 memcpy(lc, &S390_lowcore, 512);
204 memset((char *) lc + 512, 0, sizeof(*lc) - 512);
205 lc->async_stack = async_stack + ASYNC_FRAME_OFFSET;
206 lc->panic_stack = panic_stack + PANIC_FRAME_OFFSET;
208 lc->spinlock_lockval = arch_spin_lockval(cpu);
210 lc->vector_save_area_addr =
211 (unsigned long) &lc->vector_save_area;
212 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;
232 vdso_free_per_cpu(pcpu->lowcore);
233 if (pcpu == &pcpu_devices[0])
235 free_page(pcpu->lowcore->panic_stack-PANIC_FRAME_OFFSET);
236 free_pages(pcpu->lowcore->async_stack-ASYNC_FRAME_OFFSET, ASYNC_ORDER);
237 free_pages((unsigned long) pcpu->lowcore, LC_ORDER);
240 #endif /* CONFIG_HOTPLUG_CPU */
242 static void pcpu_prepare_secondary(struct pcpu *pcpu, int cpu)
244 struct lowcore *lc = pcpu->lowcore;
246 cpumask_set_cpu(cpu, &init_mm.context.cpu_attach_mask);
247 cpumask_set_cpu(cpu, mm_cpumask(&init_mm));
249 lc->spinlock_lockval = arch_spin_lockval(cpu);
250 lc->percpu_offset = __per_cpu_offset[cpu];
251 lc->kernel_asce = S390_lowcore.kernel_asce;
252 lc->machine_flags = S390_lowcore.machine_flags;
253 lc->user_timer = lc->system_timer = lc->steal_timer = 0;
254 __ctl_store(lc->cregs_save_area, 0, 15);
255 save_access_regs((unsigned int *) lc->access_regs_save_area);
256 memcpy(lc->stfle_fac_list, S390_lowcore.stfle_fac_list,
260 static void pcpu_attach_task(struct pcpu *pcpu, struct task_struct *tsk)
262 struct lowcore *lc = pcpu->lowcore;
264 lc->kernel_stack = (unsigned long) task_stack_page(tsk)
265 + THREAD_SIZE - STACK_FRAME_OVERHEAD - sizeof(struct pt_regs);
266 lc->current_task = (unsigned long) tsk;
268 lc->current_pid = tsk->pid;
269 lc->user_timer = tsk->thread.user_timer;
270 lc->system_timer = tsk->thread.system_timer;
274 static void pcpu_start_fn(struct pcpu *pcpu, void (*func)(void *), void *data)
276 struct lowcore *lc = pcpu->lowcore;
278 lc->restart_stack = lc->kernel_stack;
279 lc->restart_fn = (unsigned long) func;
280 lc->restart_data = (unsigned long) data;
281 lc->restart_source = -1UL;
282 pcpu_sigp_retry(pcpu, SIGP_RESTART, 0);
286 * Call function via PSW restart on pcpu and stop the current cpu.
288 static void pcpu_delegate(struct pcpu *pcpu, void (*func)(void *),
289 void *data, unsigned long stack)
291 struct lowcore *lc = lowcore_ptr[pcpu - pcpu_devices];
292 unsigned long source_cpu = stap();
294 __load_psw_mask(PSW_KERNEL_BITS);
295 if (pcpu->address == source_cpu)
296 func(data); /* should not return */
297 /* Stop target cpu (if func returns this stops the current cpu). */
298 pcpu_sigp_retry(pcpu, SIGP_STOP, 0);
299 /* Restart func on the target cpu and stop the current cpu. */
300 mem_assign_absolute(lc->restart_stack, stack);
301 mem_assign_absolute(lc->restart_fn, (unsigned long) func);
302 mem_assign_absolute(lc->restart_data, (unsigned long) data);
303 mem_assign_absolute(lc->restart_source, source_cpu);
305 "0: sigp 0,%0,%2 # sigp restart to target cpu\n"
306 " brc 2,0b # busy, try again\n"
307 "1: sigp 0,%1,%3 # sigp stop to current cpu\n"
308 " brc 2,1b # busy, try again\n"
309 : : "d" (pcpu->address), "d" (source_cpu),
310 "K" (SIGP_RESTART), "K" (SIGP_STOP)
316 * Enable additional logical cpus for multi-threading.
318 static int pcpu_set_smt(unsigned int mtid)
322 if (smp_cpu_mtid == mtid)
324 cc = __pcpu_sigp(0, SIGP_SET_MULTI_THREADING, mtid, NULL);
327 smp_cpu_mt_shift = 0;
328 while (smp_cpu_mtid >= (1U << smp_cpu_mt_shift))
330 pcpu_devices[0].address = stap();
336 * Call function on an online CPU.
338 void smp_call_online_cpu(void (*func)(void *), void *data)
342 /* Use the current cpu if it is online. */
343 pcpu = pcpu_find_address(cpu_online_mask, stap());
345 /* Use the first online cpu. */
346 pcpu = pcpu_devices + cpumask_first(cpu_online_mask);
347 pcpu_delegate(pcpu, func, data, (unsigned long) restart_stack);
351 * Call function on the ipl CPU.
353 void smp_call_ipl_cpu(void (*func)(void *), void *data)
355 pcpu_delegate(&pcpu_devices[0], func, data,
356 pcpu_devices->lowcore->panic_stack -
357 PANIC_FRAME_OFFSET + PAGE_SIZE);
360 int smp_find_processor_id(u16 address)
364 for_each_present_cpu(cpu)
365 if (pcpu_devices[cpu].address == address)
370 bool arch_vcpu_is_preempted(int cpu)
372 if (test_cpu_flag_of(CIF_ENABLED_WAIT, cpu))
374 if (pcpu_running(pcpu_devices + cpu))
378 EXPORT_SYMBOL(arch_vcpu_is_preempted);
380 void smp_yield_cpu(int cpu)
382 if (MACHINE_HAS_DIAG9C) {
383 diag_stat_inc_norecursion(DIAG_STAT_X09C);
384 asm volatile("diag %0,0,0x9c"
385 : : "d" (pcpu_devices[cpu].address));
386 } else if (MACHINE_HAS_DIAG44) {
387 diag_stat_inc_norecursion(DIAG_STAT_X044);
388 asm volatile("diag 0,0,0x44");
393 * Send cpus emergency shutdown signal. This gives the cpus the
394 * opportunity to complete outstanding interrupts.
396 static void smp_emergency_stop(cpumask_t *cpumask)
401 end = get_tod_clock() + (1000000UL << 12);
402 for_each_cpu(cpu, cpumask) {
403 struct pcpu *pcpu = pcpu_devices + cpu;
404 set_bit(ec_stop_cpu, &pcpu->ec_mask);
405 while (__pcpu_sigp(pcpu->address, SIGP_EMERGENCY_SIGNAL,
406 0, NULL) == SIGP_CC_BUSY &&
407 get_tod_clock() < end)
410 while (get_tod_clock() < end) {
411 for_each_cpu(cpu, cpumask)
412 if (pcpu_stopped(pcpu_devices + cpu))
413 cpumask_clear_cpu(cpu, cpumask);
414 if (cpumask_empty(cpumask))
421 * Stop all cpus but the current one.
423 void smp_send_stop(void)
428 /* Disable all interrupts/machine checks */
429 __load_psw_mask(PSW_KERNEL_BITS | PSW_MASK_DAT);
430 trace_hardirqs_off();
432 debug_set_critical();
433 cpumask_copy(&cpumask, cpu_online_mask);
434 cpumask_clear_cpu(smp_processor_id(), &cpumask);
436 if (oops_in_progress)
437 smp_emergency_stop(&cpumask);
439 /* stop all processors */
440 for_each_cpu(cpu, &cpumask) {
441 struct pcpu *pcpu = pcpu_devices + cpu;
442 pcpu_sigp_retry(pcpu, SIGP_STOP, 0);
443 while (!pcpu_stopped(pcpu))
449 * This is the main routine where commands issued by other
452 static void smp_handle_ext_call(void)
456 /* handle bit signal external calls */
457 bits = xchg(&pcpu_devices[smp_processor_id()].ec_mask, 0);
458 if (test_bit(ec_stop_cpu, &bits))
460 if (test_bit(ec_schedule, &bits))
462 if (test_bit(ec_call_function_single, &bits))
463 generic_smp_call_function_single_interrupt();
466 static void do_ext_call_interrupt(struct ext_code ext_code,
467 unsigned int param32, unsigned long param64)
469 inc_irq_stat(ext_code.code == 0x1202 ? IRQEXT_EXC : IRQEXT_EMS);
470 smp_handle_ext_call();
473 void arch_send_call_function_ipi_mask(const struct cpumask *mask)
477 for_each_cpu(cpu, mask)
478 pcpu_ec_call(pcpu_devices + cpu, ec_call_function_single);
481 void arch_send_call_function_single_ipi(int cpu)
483 pcpu_ec_call(pcpu_devices + cpu, ec_call_function_single);
487 * this function sends a 'reschedule' IPI to another CPU.
488 * it goes straight through and wastes no time serializing
489 * anything. Worst case is that we lose a reschedule ...
491 void smp_send_reschedule(int cpu)
493 pcpu_ec_call(pcpu_devices + cpu, ec_schedule);
497 * parameter area for the set/clear control bit callbacks
499 struct ec_creg_mask_parms {
501 unsigned long andval;
506 * callback for setting/clearing control bits
508 static void smp_ctl_bit_callback(void *info)
510 struct ec_creg_mask_parms *pp = info;
511 unsigned long cregs[16];
513 __ctl_store(cregs, 0, 15);
514 cregs[pp->cr] = (cregs[pp->cr] & pp->andval) | pp->orval;
515 __ctl_load(cregs, 0, 15);
519 * Set a bit in a control register of all cpus
521 void smp_ctl_set_bit(int cr, int bit)
523 struct ec_creg_mask_parms parms = { 1UL << bit, -1UL, cr };
525 on_each_cpu(smp_ctl_bit_callback, &parms, 1);
527 EXPORT_SYMBOL(smp_ctl_set_bit);
530 * Clear a bit in a control register of all cpus
532 void smp_ctl_clear_bit(int cr, int bit)
534 struct ec_creg_mask_parms parms = { 0, ~(1UL << bit), cr };
536 on_each_cpu(smp_ctl_bit_callback, &parms, 1);
538 EXPORT_SYMBOL(smp_ctl_clear_bit);
540 #ifdef CONFIG_CRASH_DUMP
542 int smp_store_status(int cpu)
544 struct pcpu *pcpu = pcpu_devices + cpu;
547 pa = __pa(&pcpu->lowcore->floating_pt_save_area);
548 if (__pcpu_sigp_relax(pcpu->address, SIGP_STORE_STATUS_AT_ADDRESS,
549 pa) != SIGP_CC_ORDER_CODE_ACCEPTED)
553 pa = __pa(pcpu->lowcore->vector_save_area_addr);
554 if (__pcpu_sigp_relax(pcpu->address, SIGP_STORE_ADDITIONAL_STATUS,
555 pa) != SIGP_CC_ORDER_CODE_ACCEPTED)
561 * Collect CPU state of the previous, crashed system.
562 * There are four cases:
563 * 1) standard zfcp dump
564 * condition: OLDMEM_BASE == NULL && ipl_info.type == IPL_TYPE_FCP_DUMP
565 * The state for all CPUs except the boot CPU needs to be collected
566 * with sigp stop-and-store-status. The boot CPU state is located in
567 * the absolute lowcore of the memory stored in the HSA. The zcore code
568 * will copy the boot CPU state from the HSA.
569 * 2) stand-alone kdump for SCSI (zfcp dump with swapped memory)
570 * condition: OLDMEM_BASE != NULL && ipl_info.type == IPL_TYPE_FCP_DUMP
571 * The state for all CPUs except the boot CPU needs to be collected
572 * with sigp stop-and-store-status. The firmware or the boot-loader
573 * stored the registers of the boot CPU in the absolute lowcore in the
574 * memory of the old system.
575 * 3) kdump and the old kernel did not store the CPU state,
576 * or stand-alone kdump for DASD
577 * condition: OLDMEM_BASE != NULL && !is_kdump_kernel()
578 * The state for all CPUs except the boot CPU needs to be collected
579 * with sigp stop-and-store-status. The kexec code or the boot-loader
580 * stored the registers of the boot CPU in the memory of the old system.
581 * 4) kdump and the old kernel stored the CPU state
582 * condition: OLDMEM_BASE != NULL && is_kdump_kernel()
583 * This case does not exist for s390 anymore, setup_arch explicitly
584 * deactivates the elfcorehdr= kernel parameter
586 static __init void smp_save_cpu_vxrs(struct save_area *sa, u16 addr,
587 bool is_boot_cpu, unsigned long page)
589 __vector128 *vxrs = (__vector128 *) page;
592 vxrs = boot_cpu_vector_save_area;
594 __pcpu_sigp_relax(addr, SIGP_STORE_ADDITIONAL_STATUS, page);
595 save_area_add_vxrs(sa, vxrs);
598 static __init void smp_save_cpu_regs(struct save_area *sa, u16 addr,
599 bool is_boot_cpu, unsigned long page)
601 void *regs = (void *) page;
604 copy_oldmem_kernel(regs, (void *) __LC_FPREGS_SAVE_AREA, 512);
606 __pcpu_sigp_relax(addr, SIGP_STORE_STATUS_AT_ADDRESS, page);
607 save_area_add_regs(sa, regs);
610 void __init smp_save_dump_cpus(void)
612 int addr, boot_cpu_addr, max_cpu_addr;
613 struct save_area *sa;
617 if (!(OLDMEM_BASE || ipl_info.type == IPL_TYPE_FCP_DUMP))
618 /* No previous system present, normal boot. */
620 /* Allocate a page as dumping area for the store status sigps */
621 page = memblock_alloc_base(PAGE_SIZE, PAGE_SIZE, 1UL << 31);
622 /* Set multi-threading state to the previous system. */
623 pcpu_set_smt(sclp.mtid_prev);
624 boot_cpu_addr = stap();
625 max_cpu_addr = SCLP_MAX_CORES << sclp.mtid_prev;
626 for (addr = 0; addr <= max_cpu_addr; addr++) {
627 if (__pcpu_sigp_relax(addr, SIGP_SENSE, 0) ==
628 SIGP_CC_NOT_OPERATIONAL)
630 is_boot_cpu = (addr == boot_cpu_addr);
631 /* Allocate save area */
632 sa = save_area_alloc(is_boot_cpu);
634 panic("could not allocate memory for save area\n");
636 /* Get the vector registers */
637 smp_save_cpu_vxrs(sa, addr, is_boot_cpu, page);
639 * For a zfcp dump OLDMEM_BASE == NULL and the registers
640 * of the boot CPU are stored in the HSA. To retrieve
641 * these registers an SCLP request is required which is
642 * done by drivers/s390/char/zcore.c:init_cpu_info()
644 if (!is_boot_cpu || OLDMEM_BASE)
645 /* Get the CPU registers */
646 smp_save_cpu_regs(sa, addr, is_boot_cpu, page);
648 memblock_free(page, PAGE_SIZE);
652 #endif /* CONFIG_CRASH_DUMP */
654 void smp_cpu_set_polarization(int cpu, int val)
656 pcpu_devices[cpu].polarization = val;
659 int smp_cpu_get_polarization(int cpu)
661 return pcpu_devices[cpu].polarization;
664 static void __ref smp_get_core_info(struct sclp_core_info *info, int early)
666 static int use_sigp_detection;
669 if (use_sigp_detection || sclp_get_core_info(info, early)) {
670 use_sigp_detection = 1;
672 address < (SCLP_MAX_CORES << smp_cpu_mt_shift);
673 address += (1U << smp_cpu_mt_shift)) {
674 if (__pcpu_sigp_relax(address, SIGP_SENSE, 0) ==
675 SIGP_CC_NOT_OPERATIONAL)
677 info->core[info->configured].core_id =
678 address >> smp_cpu_mt_shift;
681 info->combined = info->configured;
685 static int smp_add_present_cpu(int cpu);
687 static int __smp_rescan_cpus(struct sclp_core_info *info, int sysfs_add)
695 cpumask_xor(&avail, cpu_possible_mask, cpu_present_mask);
696 cpu = cpumask_first(&avail);
697 for (i = 0; (i < info->combined) && (cpu < nr_cpu_ids); i++) {
698 if (sclp.has_core_type && info->core[i].type != boot_core_type)
700 address = info->core[i].core_id << smp_cpu_mt_shift;
701 for (j = 0; j <= smp_cpu_mtid; j++) {
702 if (pcpu_find_address(cpu_present_mask, address + j))
704 pcpu = pcpu_devices + cpu;
705 pcpu->address = address + j;
707 (cpu >= info->configured*(smp_cpu_mtid + 1)) ?
708 CPU_STATE_STANDBY : CPU_STATE_CONFIGURED;
709 smp_cpu_set_polarization(cpu, POLARIZATION_UNKNOWN);
710 set_cpu_present(cpu, true);
711 if (sysfs_add && smp_add_present_cpu(cpu) != 0)
712 set_cpu_present(cpu, false);
715 cpu = cpumask_next(cpu, &avail);
716 if (cpu >= nr_cpu_ids)
723 void __init smp_detect_cpus(void)
725 unsigned int cpu, mtid, c_cpus, s_cpus;
726 struct sclp_core_info *info;
729 /* Get CPU information */
730 info = memblock_virt_alloc(sizeof(*info), 8);
731 smp_get_core_info(info, 1);
732 /* Find boot CPU type */
733 if (sclp.has_core_type) {
735 for (cpu = 0; cpu < info->combined; cpu++)
736 if (info->core[cpu].core_id == address) {
737 /* The boot cpu dictates the cpu type. */
738 boot_core_type = info->core[cpu].type;
741 if (cpu >= info->combined)
742 panic("Could not find boot CPU type");
745 /* Set multi-threading state for the current system */
746 mtid = boot_core_type ? sclp.mtid : sclp.mtid_cp;
747 mtid = (mtid < smp_max_threads) ? mtid : smp_max_threads - 1;
750 /* Print number of CPUs */
752 for (cpu = 0; cpu < info->combined; cpu++) {
753 if (sclp.has_core_type &&
754 info->core[cpu].type != boot_core_type)
756 if (cpu < info->configured)
757 c_cpus += smp_cpu_mtid + 1;
759 s_cpus += smp_cpu_mtid + 1;
761 pr_info("%d configured CPUs, %d standby CPUs\n", c_cpus, s_cpus);
763 /* Add CPUs present at boot */
765 __smp_rescan_cpus(info, 0);
767 memblock_free_early((unsigned long)info, sizeof(*info));
771 * Activate a secondary processor.
773 static void smp_start_secondary(void *cpuvoid)
775 S390_lowcore.last_update_clock = get_tod_clock();
776 S390_lowcore.restart_stack = (unsigned long) restart_stack;
777 S390_lowcore.restart_fn = (unsigned long) do_restart;
778 S390_lowcore.restart_data = 0;
779 S390_lowcore.restart_source = -1UL;
780 restore_access_regs(S390_lowcore.access_regs_save_area);
781 __ctl_load(S390_lowcore.cregs_save_area, 0, 15);
782 __load_psw_mask(PSW_KERNEL_BITS | PSW_MASK_DAT);
788 notify_cpu_starting(smp_processor_id());
789 set_cpu_online(smp_processor_id(), true);
790 inc_irq_stat(CPU_RST);
792 cpu_startup_entry(CPUHP_AP_ONLINE_IDLE);
795 /* Upping and downing of CPUs */
796 int __cpu_up(unsigned int cpu, struct task_struct *tidle)
801 pcpu = pcpu_devices + cpu;
802 if (pcpu->state != CPU_STATE_CONFIGURED)
804 base = smp_get_base_cpu(cpu);
805 for (i = 0; i <= smp_cpu_mtid; i++) {
806 if (base + i < nr_cpu_ids)
807 if (cpu_online(base + i))
811 * If this is the first CPU of the core to get online
812 * do an initial CPU reset.
814 if (i > smp_cpu_mtid &&
815 pcpu_sigp_retry(pcpu_devices + base, SIGP_INITIAL_CPU_RESET, 0) !=
816 SIGP_CC_ORDER_CODE_ACCEPTED)
819 rc = pcpu_alloc_lowcore(pcpu, cpu);
822 pcpu_prepare_secondary(pcpu, cpu);
823 pcpu_attach_task(pcpu, tidle);
824 pcpu_start_fn(pcpu, smp_start_secondary, NULL);
825 /* Wait until cpu puts itself in the online & active maps */
826 while (!cpu_online(cpu))
831 static unsigned int setup_possible_cpus __initdata;
833 static int __init _setup_possible_cpus(char *s)
835 get_option(&s, &setup_possible_cpus);
838 early_param("possible_cpus", _setup_possible_cpus);
840 #ifdef CONFIG_HOTPLUG_CPU
842 int __cpu_disable(void)
844 unsigned long cregs[16];
846 /* Handle possible pending IPIs */
847 smp_handle_ext_call();
848 set_cpu_online(smp_processor_id(), false);
849 /* Disable pseudo page faults on this cpu. */
851 /* Disable interrupt sources via control register. */
852 __ctl_store(cregs, 0, 15);
853 cregs[0] &= ~0x0000ee70UL; /* disable all external interrupts */
854 cregs[6] &= ~0xff000000UL; /* disable all I/O interrupts */
855 cregs[14] &= ~0x1f000000UL; /* disable most machine checks */
856 __ctl_load(cregs, 0, 15);
857 clear_cpu_flag(CIF_NOHZ_DELAY);
861 void __cpu_die(unsigned int cpu)
865 /* Wait until target cpu is down */
866 pcpu = pcpu_devices + cpu;
867 while (!pcpu_stopped(pcpu))
869 pcpu_free_lowcore(pcpu);
870 cpumask_clear_cpu(cpu, mm_cpumask(&init_mm));
871 cpumask_clear_cpu(cpu, &init_mm.context.cpu_attach_mask);
874 void __noreturn cpu_die(void)
877 pcpu_sigp_retry(pcpu_devices + smp_processor_id(), SIGP_STOP, 0);
881 #endif /* CONFIG_HOTPLUG_CPU */
883 void __init smp_fill_possible_mask(void)
885 unsigned int possible, sclp_max, cpu;
887 sclp_max = max(sclp.mtid, sclp.mtid_cp) + 1;
888 sclp_max = min(smp_max_threads, sclp_max);
889 sclp_max = (sclp.max_cores * sclp_max) ?: nr_cpu_ids;
890 possible = setup_possible_cpus ?: nr_cpu_ids;
891 possible = min(possible, sclp_max);
892 for (cpu = 0; cpu < possible && cpu < nr_cpu_ids; cpu++)
893 set_cpu_possible(cpu, true);
896 void __init smp_prepare_cpus(unsigned int max_cpus)
898 /* request the 0x1201 emergency signal external interrupt */
899 if (register_external_irq(EXT_IRQ_EMERGENCY_SIG, do_ext_call_interrupt))
900 panic("Couldn't request external interrupt 0x1201");
901 /* request the 0x1202 external call external interrupt */
902 if (register_external_irq(EXT_IRQ_EXTERNAL_CALL, do_ext_call_interrupt))
903 panic("Couldn't request external interrupt 0x1202");
906 void __init smp_prepare_boot_cpu(void)
908 struct pcpu *pcpu = pcpu_devices;
910 pcpu->state = CPU_STATE_CONFIGURED;
911 pcpu->address = stap();
912 pcpu->lowcore = (struct lowcore *)(unsigned long) store_prefix();
913 S390_lowcore.percpu_offset = __per_cpu_offset[0];
914 smp_cpu_set_polarization(0, POLARIZATION_UNKNOWN);
915 set_cpu_present(0, true);
916 set_cpu_online(0, true);
919 void __init smp_cpus_done(unsigned int max_cpus)
923 void __init smp_setup_processor_id(void)
925 S390_lowcore.cpu_nr = 0;
926 S390_lowcore.spinlock_lockval = arch_spin_lockval(0);
930 * the frequency of the profiling timer can be changed
931 * by writing a multiplier value into /proc/profile.
933 * usually you want to run this on all CPUs ;)
935 int setup_profiling_timer(unsigned int multiplier)
940 #ifdef CONFIG_HOTPLUG_CPU
941 static ssize_t cpu_configure_show(struct device *dev,
942 struct device_attribute *attr, char *buf)
946 mutex_lock(&smp_cpu_state_mutex);
947 count = sprintf(buf, "%d\n", pcpu_devices[dev->id].state);
948 mutex_unlock(&smp_cpu_state_mutex);
952 static ssize_t cpu_configure_store(struct device *dev,
953 struct device_attribute *attr,
954 const char *buf, size_t count)
960 if (sscanf(buf, "%d %c", &val, &delim) != 1)
962 if (val != 0 && val != 1)
965 mutex_lock(&smp_cpu_state_mutex);
967 /* disallow configuration changes of online cpus and cpu 0 */
969 cpu = smp_get_base_cpu(cpu);
972 for (i = 0; i <= smp_cpu_mtid; i++)
973 if (cpu_online(cpu + i))
975 pcpu = pcpu_devices + cpu;
979 if (pcpu->state != CPU_STATE_CONFIGURED)
981 rc = sclp_core_deconfigure(pcpu->address >> smp_cpu_mt_shift);
984 for (i = 0; i <= smp_cpu_mtid; i++) {
985 if (cpu + i >= nr_cpu_ids || !cpu_present(cpu + i))
987 pcpu[i].state = CPU_STATE_STANDBY;
988 smp_cpu_set_polarization(cpu + i,
989 POLARIZATION_UNKNOWN);
991 topology_expect_change();
994 if (pcpu->state != CPU_STATE_STANDBY)
996 rc = sclp_core_configure(pcpu->address >> smp_cpu_mt_shift);
999 for (i = 0; i <= smp_cpu_mtid; i++) {
1000 if (cpu + i >= nr_cpu_ids || !cpu_present(cpu + i))
1002 pcpu[i].state = CPU_STATE_CONFIGURED;
1003 smp_cpu_set_polarization(cpu + i,
1004 POLARIZATION_UNKNOWN);
1006 topology_expect_change();
1012 mutex_unlock(&smp_cpu_state_mutex);
1014 return rc ? rc : count;
1016 static DEVICE_ATTR(configure, 0644, cpu_configure_show, cpu_configure_store);
1017 #endif /* CONFIG_HOTPLUG_CPU */
1019 static ssize_t show_cpu_address(struct device *dev,
1020 struct device_attribute *attr, char *buf)
1022 return sprintf(buf, "%d\n", pcpu_devices[dev->id].address);
1024 static DEVICE_ATTR(address, 0444, show_cpu_address, NULL);
1026 static struct attribute *cpu_common_attrs[] = {
1027 #ifdef CONFIG_HOTPLUG_CPU
1028 &dev_attr_configure.attr,
1030 &dev_attr_address.attr,
1034 static struct attribute_group cpu_common_attr_group = {
1035 .attrs = cpu_common_attrs,
1038 static struct attribute *cpu_online_attrs[] = {
1039 &dev_attr_idle_count.attr,
1040 &dev_attr_idle_time_us.attr,
1044 static struct attribute_group cpu_online_attr_group = {
1045 .attrs = cpu_online_attrs,
1048 static int smp_cpu_online(unsigned int cpu)
1050 struct device *s = &per_cpu(cpu_device, cpu)->dev;
1052 return sysfs_create_group(&s->kobj, &cpu_online_attr_group);
1054 static int smp_cpu_pre_down(unsigned int cpu)
1056 struct device *s = &per_cpu(cpu_device, cpu)->dev;
1058 sysfs_remove_group(&s->kobj, &cpu_online_attr_group);
1062 static int smp_add_present_cpu(int cpu)
1068 c = kzalloc(sizeof(*c), GFP_KERNEL);
1071 per_cpu(cpu_device, cpu) = c;
1073 c->hotpluggable = 1;
1074 rc = register_cpu(c, cpu);
1077 rc = sysfs_create_group(&s->kobj, &cpu_common_attr_group);
1080 rc = topology_cpu_init(c);
1086 sysfs_remove_group(&s->kobj, &cpu_common_attr_group);
1088 #ifdef CONFIG_HOTPLUG_CPU
1095 #ifdef CONFIG_HOTPLUG_CPU
1097 int __ref smp_rescan_cpus(void)
1099 struct sclp_core_info *info;
1102 info = kzalloc(sizeof(*info), GFP_KERNEL);
1105 smp_get_core_info(info, 0);
1107 mutex_lock(&smp_cpu_state_mutex);
1108 nr = __smp_rescan_cpus(info, 1);
1109 mutex_unlock(&smp_cpu_state_mutex);
1113 topology_schedule_update();
1117 static ssize_t __ref rescan_store(struct device *dev,
1118 struct device_attribute *attr,
1124 rc = smp_rescan_cpus();
1125 return rc ? rc : count;
1127 static DEVICE_ATTR(rescan, 0200, NULL, rescan_store);
1128 #endif /* CONFIG_HOTPLUG_CPU */
1130 static int __init s390_smp_init(void)
1134 #ifdef CONFIG_HOTPLUG_CPU
1135 rc = device_create_file(cpu_subsys.dev_root, &dev_attr_rescan);
1139 for_each_present_cpu(cpu) {
1140 rc = smp_add_present_cpu(cpu);
1145 rc = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "s390/smp:online",
1146 smp_cpu_online, smp_cpu_pre_down);
1150 subsys_initcall(s390_smp_init);