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 <linux/memblock.h>
35 #include <asm/asm-offsets.h>
37 #include <asm/switch_to.h>
38 #include <asm/facility.h>
40 #include <asm/setup.h>
42 #include <asm/tlbflush.h>
43 #include <asm/vtimer.h>
44 #include <asm/lowcore.h>
47 #include <asm/debug.h>
48 #include <asm/os_info.h>
55 ec_call_function_single,
64 static DEFINE_PER_CPU(struct cpu *, cpu_device);
67 struct lowcore *lowcore; /* lowcore page(s) for the cpu */
68 unsigned long ec_mask; /* bit mask for ec_xxx functions */
69 unsigned long ec_clk; /* sigp timestamp for ec_xxx */
70 signed char state; /* physical cpu state */
71 signed char polarization; /* physical polarization */
72 u16 address; /* physical cpu address */
75 static u8 boot_core_type;
76 static struct pcpu pcpu_devices[NR_CPUS];
78 unsigned int smp_cpu_mt_shift;
79 EXPORT_SYMBOL(smp_cpu_mt_shift);
81 unsigned int smp_cpu_mtid;
82 EXPORT_SYMBOL(smp_cpu_mtid);
84 #ifdef CONFIG_CRASH_DUMP
85 __vector128 __initdata boot_cpu_vector_save_area[__NUM_VXRS];
88 static unsigned int smp_max_threads __initdata = -1U;
90 static int __init early_nosmt(char *s)
95 early_param("nosmt", early_nosmt);
97 static int __init early_smt(char *s)
99 get_option(&s, &smp_max_threads);
102 early_param("smt", early_smt);
105 * The smp_cpu_state_mutex must be held when changing the state or polarization
106 * member of a pcpu data structure within the pcpu_devices arreay.
108 DEFINE_MUTEX(smp_cpu_state_mutex);
111 * Signal processor helper functions.
113 static inline int __pcpu_sigp_relax(u16 addr, u8 order, unsigned long parm)
118 cc = __pcpu_sigp(addr, order, parm, NULL);
119 if (cc != SIGP_CC_BUSY)
125 static int pcpu_sigp_retry(struct pcpu *pcpu, u8 order, u32 parm)
129 for (retry = 0; ; retry++) {
130 cc = __pcpu_sigp(pcpu->address, order, parm, NULL);
131 if (cc != SIGP_CC_BUSY)
139 static inline int pcpu_stopped(struct pcpu *pcpu)
141 u32 uninitialized_var(status);
143 if (__pcpu_sigp(pcpu->address, SIGP_SENSE,
144 0, &status) != SIGP_CC_STATUS_STORED)
146 return !!(status & (SIGP_STATUS_CHECK_STOP|SIGP_STATUS_STOPPED));
149 static inline int pcpu_running(struct pcpu *pcpu)
151 if (__pcpu_sigp(pcpu->address, SIGP_SENSE_RUNNING,
152 0, NULL) != SIGP_CC_STATUS_STORED)
154 /* Status stored condition code is equivalent to cpu not running. */
159 * Find struct pcpu by cpu address.
161 static struct pcpu *pcpu_find_address(const struct cpumask *mask, u16 address)
165 for_each_cpu(cpu, mask)
166 if (pcpu_devices[cpu].address == address)
167 return pcpu_devices + cpu;
171 static void pcpu_ec_call(struct pcpu *pcpu, int ec_bit)
175 if (test_and_set_bit(ec_bit, &pcpu->ec_mask))
177 order = pcpu_running(pcpu) ? SIGP_EXTERNAL_CALL : SIGP_EMERGENCY_SIGNAL;
178 pcpu->ec_clk = get_tod_clock_fast();
179 pcpu_sigp_retry(pcpu, order, 0);
182 #define ASYNC_FRAME_OFFSET (ASYNC_SIZE - STACK_FRAME_OVERHEAD - __PT_SIZE)
183 #define PANIC_FRAME_OFFSET (PAGE_SIZE - STACK_FRAME_OVERHEAD - __PT_SIZE)
185 static int pcpu_alloc_lowcore(struct pcpu *pcpu, int cpu)
187 unsigned long async_stack, panic_stack;
190 if (pcpu != &pcpu_devices[0]) {
191 pcpu->lowcore = (struct lowcore *)
192 __get_free_pages(GFP_KERNEL | GFP_DMA, LC_ORDER);
193 async_stack = __get_free_pages(GFP_KERNEL, ASYNC_ORDER);
194 panic_stack = __get_free_page(GFP_KERNEL);
195 if (!pcpu->lowcore || !panic_stack || !async_stack)
198 async_stack = pcpu->lowcore->async_stack - ASYNC_FRAME_OFFSET;
199 panic_stack = pcpu->lowcore->panic_stack - PANIC_FRAME_OFFSET;
202 memcpy(lc, &S390_lowcore, 512);
203 memset((char *) lc + 512, 0, sizeof(*lc) - 512);
204 lc->async_stack = async_stack + ASYNC_FRAME_OFFSET;
205 lc->panic_stack = panic_stack + PANIC_FRAME_OFFSET;
207 lc->spinlock_lockval = arch_spin_lockval(cpu);
208 lc->br_r1_trampoline = 0x07f1; /* br %r1 */
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,
257 sizeof(lc->stfle_fac_list));
258 memcpy(lc->alt_stfle_fac_list, S390_lowcore.alt_stfle_fac_list,
259 sizeof(lc->alt_stfle_fac_list));
262 static void pcpu_attach_task(struct pcpu *pcpu, struct task_struct *tsk)
264 struct lowcore *lc = pcpu->lowcore;
265 struct thread_info *ti = task_thread_info(tsk);
267 lc->kernel_stack = (unsigned long) task_stack_page(tsk)
268 + THREAD_SIZE - STACK_FRAME_OVERHEAD - sizeof(struct pt_regs);
269 lc->thread_info = (unsigned long) task_thread_info(tsk);
270 lc->current_task = (unsigned long) tsk;
272 lc->current_pid = tsk->pid;
273 lc->user_timer = ti->user_timer;
274 lc->system_timer = ti->system_timer;
278 static void pcpu_start_fn(struct pcpu *pcpu, void (*func)(void *), void *data)
280 struct lowcore *lc = pcpu->lowcore;
282 lc->restart_stack = lc->kernel_stack;
283 lc->restart_fn = (unsigned long) func;
284 lc->restart_data = (unsigned long) data;
285 lc->restart_source = -1UL;
286 pcpu_sigp_retry(pcpu, SIGP_RESTART, 0);
290 * Call function via PSW restart on pcpu and stop the current cpu.
292 static void pcpu_delegate(struct pcpu *pcpu, void (*func)(void *),
293 void *data, unsigned long stack)
295 struct lowcore *lc = lowcore_ptr[pcpu - pcpu_devices];
296 unsigned long source_cpu = stap();
298 __load_psw_mask(PSW_KERNEL_BITS);
299 if (pcpu->address == source_cpu)
300 func(data); /* should not return */
301 /* Stop target cpu (if func returns this stops the current cpu). */
302 pcpu_sigp_retry(pcpu, SIGP_STOP, 0);
303 /* Restart func on the target cpu and stop the current cpu. */
304 mem_assign_absolute(lc->restart_stack, stack);
305 mem_assign_absolute(lc->restart_fn, (unsigned long) func);
306 mem_assign_absolute(lc->restart_data, (unsigned long) data);
307 mem_assign_absolute(lc->restart_source, source_cpu);
310 "0: sigp 0,%0,%2 # sigp restart to target cpu\n"
311 " brc 2,0b # busy, try again\n"
312 "1: sigp 0,%1,%3 # sigp stop to current cpu\n"
313 " brc 2,1b # busy, try again\n"
314 : : "d" (pcpu->address), "d" (source_cpu),
315 "K" (SIGP_RESTART), "K" (SIGP_STOP)
321 * Enable additional logical cpus for multi-threading.
323 static int pcpu_set_smt(unsigned int mtid)
327 if (smp_cpu_mtid == mtid)
329 cc = __pcpu_sigp(0, SIGP_SET_MULTI_THREADING, mtid, NULL);
332 smp_cpu_mt_shift = 0;
333 while (smp_cpu_mtid >= (1U << smp_cpu_mt_shift))
335 pcpu_devices[0].address = stap();
341 * Call function on an online CPU.
343 void smp_call_online_cpu(void (*func)(void *), void *data)
347 /* Use the current cpu if it is online. */
348 pcpu = pcpu_find_address(cpu_online_mask, stap());
350 /* Use the first online cpu. */
351 pcpu = pcpu_devices + cpumask_first(cpu_online_mask);
352 pcpu_delegate(pcpu, func, data, (unsigned long) restart_stack);
356 * Call function on the ipl CPU.
358 void smp_call_ipl_cpu(void (*func)(void *), void *data)
360 struct lowcore *lc = pcpu_devices->lowcore;
362 if (pcpu_devices[0].address == stap())
365 pcpu_delegate(&pcpu_devices[0], func, data,
366 lc->panic_stack - PANIC_FRAME_OFFSET + PAGE_SIZE);
369 int smp_find_processor_id(u16 address)
373 for_each_present_cpu(cpu)
374 if (pcpu_devices[cpu].address == address)
379 int smp_vcpu_scheduled(int cpu)
381 return pcpu_running(pcpu_devices + cpu);
384 void smp_yield_cpu(int cpu)
386 if (MACHINE_HAS_DIAG9C) {
387 diag_stat_inc_norecursion(DIAG_STAT_X09C);
388 asm volatile("diag %0,0,0x9c"
389 : : "d" (pcpu_devices[cpu].address));
390 } else if (MACHINE_HAS_DIAG44) {
391 diag_stat_inc_norecursion(DIAG_STAT_X044);
392 asm volatile("diag 0,0,0x44");
397 * Send cpus emergency shutdown signal. This gives the cpus the
398 * opportunity to complete outstanding interrupts.
400 static void smp_emergency_stop(cpumask_t *cpumask)
405 end = get_tod_clock() + (1000000UL << 12);
406 for_each_cpu(cpu, cpumask) {
407 struct pcpu *pcpu = pcpu_devices + cpu;
408 set_bit(ec_stop_cpu, &pcpu->ec_mask);
409 while (__pcpu_sigp(pcpu->address, SIGP_EMERGENCY_SIGNAL,
410 0, NULL) == SIGP_CC_BUSY &&
411 get_tod_clock() < end)
414 while (get_tod_clock() < end) {
415 for_each_cpu(cpu, cpumask)
416 if (pcpu_stopped(pcpu_devices + cpu))
417 cpumask_clear_cpu(cpu, cpumask);
418 if (cpumask_empty(cpumask))
425 * Stop all cpus but the current one.
427 void smp_send_stop(void)
432 /* Disable all interrupts/machine checks */
433 __load_psw_mask(PSW_KERNEL_BITS | PSW_MASK_DAT);
434 trace_hardirqs_off();
436 debug_set_critical();
437 cpumask_copy(&cpumask, cpu_online_mask);
438 cpumask_clear_cpu(smp_processor_id(), &cpumask);
440 if (oops_in_progress)
441 smp_emergency_stop(&cpumask);
443 /* stop all processors */
444 for_each_cpu(cpu, &cpumask) {
445 struct pcpu *pcpu = pcpu_devices + cpu;
446 pcpu_sigp_retry(pcpu, SIGP_STOP, 0);
447 while (!pcpu_stopped(pcpu))
453 * This is the main routine where commands issued by other
456 static void smp_handle_ext_call(void)
460 /* handle bit signal external calls */
461 bits = xchg(&pcpu_devices[smp_processor_id()].ec_mask, 0);
462 if (test_bit(ec_stop_cpu, &bits))
464 if (test_bit(ec_schedule, &bits))
466 if (test_bit(ec_call_function_single, &bits))
467 generic_smp_call_function_single_interrupt();
470 static void do_ext_call_interrupt(struct ext_code ext_code,
471 unsigned int param32, unsigned long param64)
473 inc_irq_stat(ext_code.code == 0x1202 ? IRQEXT_EXC : IRQEXT_EMS);
474 smp_handle_ext_call();
477 void arch_send_call_function_ipi_mask(const struct cpumask *mask)
481 for_each_cpu(cpu, mask)
482 pcpu_ec_call(pcpu_devices + cpu, ec_call_function_single);
485 void arch_send_call_function_single_ipi(int cpu)
487 pcpu_ec_call(pcpu_devices + cpu, ec_call_function_single);
491 * this function sends a 'reschedule' IPI to another CPU.
492 * it goes straight through and wastes no time serializing
493 * anything. Worst case is that we lose a reschedule ...
495 void smp_send_reschedule(int cpu)
497 pcpu_ec_call(pcpu_devices + cpu, ec_schedule);
501 * parameter area for the set/clear control bit callbacks
503 struct ec_creg_mask_parms {
505 unsigned long andval;
510 * callback for setting/clearing control bits
512 static void smp_ctl_bit_callback(void *info)
514 struct ec_creg_mask_parms *pp = info;
515 unsigned long cregs[16];
517 __ctl_store(cregs, 0, 15);
518 cregs[pp->cr] = (cregs[pp->cr] & pp->andval) | pp->orval;
519 __ctl_load(cregs, 0, 15);
523 * Set a bit in a control register of all cpus
525 void smp_ctl_set_bit(int cr, int bit)
527 struct ec_creg_mask_parms parms = { 1UL << bit, -1UL, cr };
529 on_each_cpu(smp_ctl_bit_callback, &parms, 1);
531 EXPORT_SYMBOL(smp_ctl_set_bit);
534 * Clear a bit in a control register of all cpus
536 void smp_ctl_clear_bit(int cr, int bit)
538 struct ec_creg_mask_parms parms = { 0, ~(1UL << bit), cr };
540 on_each_cpu(smp_ctl_bit_callback, &parms, 1);
542 EXPORT_SYMBOL(smp_ctl_clear_bit);
544 #ifdef CONFIG_CRASH_DUMP
546 int smp_store_status(int cpu)
548 struct pcpu *pcpu = pcpu_devices + cpu;
551 pa = __pa(&pcpu->lowcore->floating_pt_save_area);
552 if (__pcpu_sigp_relax(pcpu->address, SIGP_STORE_STATUS_AT_ADDRESS,
553 pa) != SIGP_CC_ORDER_CODE_ACCEPTED)
557 pa = __pa(pcpu->lowcore->vector_save_area_addr);
558 if (__pcpu_sigp_relax(pcpu->address, SIGP_STORE_ADDITIONAL_STATUS,
559 pa) != SIGP_CC_ORDER_CODE_ACCEPTED)
565 * Collect CPU state of the previous, crashed system.
566 * There are four cases:
567 * 1) standard zfcp dump
568 * condition: OLDMEM_BASE == NULL && ipl_info.type == IPL_TYPE_FCP_DUMP
569 * The state for all CPUs except the boot CPU needs to be collected
570 * with sigp stop-and-store-status. The boot CPU state is located in
571 * the absolute lowcore of the memory stored in the HSA. The zcore code
572 * will copy the boot CPU state from the HSA.
573 * 2) stand-alone kdump for SCSI (zfcp dump with swapped memory)
574 * condition: OLDMEM_BASE != NULL && ipl_info.type == IPL_TYPE_FCP_DUMP
575 * The state for all CPUs except the boot CPU needs to be collected
576 * with sigp stop-and-store-status. The firmware or the boot-loader
577 * stored the registers of the boot CPU in the absolute lowcore in the
578 * memory of the old system.
579 * 3) kdump and the old kernel did not store the CPU state,
580 * or stand-alone kdump for DASD
581 * condition: OLDMEM_BASE != NULL && !is_kdump_kernel()
582 * The state for all CPUs except the boot CPU needs to be collected
583 * with sigp stop-and-store-status. The kexec code or the boot-loader
584 * stored the registers of the boot CPU in the memory of the old system.
585 * 4) kdump and the old kernel stored the CPU state
586 * condition: OLDMEM_BASE != NULL && is_kdump_kernel()
587 * This case does not exist for s390 anymore, setup_arch explicitly
588 * deactivates the elfcorehdr= kernel parameter
590 static __init void smp_save_cpu_vxrs(struct save_area *sa, u16 addr,
591 bool is_boot_cpu, unsigned long page)
593 __vector128 *vxrs = (__vector128 *) page;
596 vxrs = boot_cpu_vector_save_area;
598 __pcpu_sigp_relax(addr, SIGP_STORE_ADDITIONAL_STATUS, page);
599 save_area_add_vxrs(sa, vxrs);
602 static __init void smp_save_cpu_regs(struct save_area *sa, u16 addr,
603 bool is_boot_cpu, unsigned long page)
605 void *regs = (void *) page;
608 copy_oldmem_kernel(regs, (void *) __LC_FPREGS_SAVE_AREA, 512);
610 __pcpu_sigp_relax(addr, SIGP_STORE_STATUS_AT_ADDRESS, page);
611 save_area_add_regs(sa, regs);
614 void __init smp_save_dump_cpus(void)
616 int addr, boot_cpu_addr, max_cpu_addr;
617 struct save_area *sa;
621 if (!(OLDMEM_BASE || ipl_info.type == IPL_TYPE_FCP_DUMP))
622 /* No previous system present, normal boot. */
624 /* Allocate a page as dumping area for the store status sigps */
625 page = memblock_alloc_base(PAGE_SIZE, PAGE_SIZE, 1UL << 31);
626 /* Set multi-threading state to the previous system. */
627 pcpu_set_smt(sclp.mtid_prev);
628 boot_cpu_addr = stap();
629 max_cpu_addr = SCLP_MAX_CORES << sclp.mtid_prev;
630 for (addr = 0; addr <= max_cpu_addr; addr++) {
631 if (__pcpu_sigp_relax(addr, SIGP_SENSE, 0) ==
632 SIGP_CC_NOT_OPERATIONAL)
634 is_boot_cpu = (addr == boot_cpu_addr);
635 /* Allocate save area */
636 sa = save_area_alloc(is_boot_cpu);
638 panic("could not allocate memory for save area\n");
640 /* Get the vector registers */
641 smp_save_cpu_vxrs(sa, addr, is_boot_cpu, page);
643 * For a zfcp dump OLDMEM_BASE == NULL and the registers
644 * of the boot CPU are stored in the HSA. To retrieve
645 * these registers an SCLP request is required which is
646 * done by drivers/s390/char/zcore.c:init_cpu_info()
648 if (!is_boot_cpu || OLDMEM_BASE)
649 /* Get the CPU registers */
650 smp_save_cpu_regs(sa, addr, is_boot_cpu, page);
652 memblock_free(page, PAGE_SIZE);
656 #endif /* CONFIG_CRASH_DUMP */
658 void smp_cpu_set_polarization(int cpu, int val)
660 pcpu_devices[cpu].polarization = val;
663 int smp_cpu_get_polarization(int cpu)
665 return pcpu_devices[cpu].polarization;
668 static struct sclp_core_info *smp_get_core_info(void)
670 static int use_sigp_detection;
671 struct sclp_core_info *info;
674 info = kzalloc(sizeof(*info), GFP_KERNEL);
675 if (info && (use_sigp_detection || sclp_get_core_info(info))) {
676 use_sigp_detection = 1;
678 address < (SCLP_MAX_CORES << smp_cpu_mt_shift);
679 address += (1U << smp_cpu_mt_shift)) {
680 if (__pcpu_sigp_relax(address, SIGP_SENSE, 0) ==
681 SIGP_CC_NOT_OPERATIONAL)
683 info->core[info->configured].core_id =
684 address >> smp_cpu_mt_shift;
687 info->combined = info->configured;
692 static int smp_add_present_cpu(int cpu);
694 static int smp_add_core(struct sclp_core_entry *core, cpumask_t *avail,
695 bool configured, bool early)
702 if (sclp.has_core_type && core->type != boot_core_type)
704 cpu = cpumask_first(avail);
705 address = core->core_id << smp_cpu_mt_shift;
706 for (i = 0; (i <= smp_cpu_mtid) && (cpu < nr_cpu_ids); i++) {
707 if (pcpu_find_address(cpu_present_mask, address + i))
709 pcpu = pcpu_devices + cpu;
710 pcpu->address = address + i;
712 pcpu->state = CPU_STATE_CONFIGURED;
714 pcpu->state = CPU_STATE_STANDBY;
715 smp_cpu_set_polarization(cpu, POLARIZATION_UNKNOWN);
716 set_cpu_present(cpu, true);
717 if (!early && smp_add_present_cpu(cpu) != 0)
718 set_cpu_present(cpu, false);
721 cpumask_clear_cpu(cpu, avail);
722 cpu = cpumask_next(cpu, avail);
727 static int __smp_rescan_cpus(struct sclp_core_info *info, bool early)
729 struct sclp_core_entry *core;
730 static cpumask_t avail;
736 cpumask_xor(&avail, cpu_possible_mask, cpu_present_mask);
738 * Add IPL core first (which got logical CPU number 0) to make sure
739 * that all SMT threads get subsequent logical CPU numbers.
742 core_id = pcpu_devices[0].address >> smp_cpu_mt_shift;
743 for (i = 0; i < info->configured; i++) {
744 core = &info->core[i];
745 if (core->core_id == core_id) {
746 nr += smp_add_core(core, &avail, true, early);
751 for (i = 0; i < info->combined; i++) {
752 configured = i < info->configured;
753 nr += smp_add_core(&info->core[i], &avail, configured, early);
758 static void __init smp_detect_cpus(void)
760 unsigned int cpu, mtid, c_cpus, s_cpus;
761 struct sclp_core_info *info;
764 /* Get CPU information */
765 info = smp_get_core_info();
767 panic("smp_detect_cpus failed to allocate memory\n");
769 /* Find boot CPU type */
770 if (sclp.has_core_type) {
772 for (cpu = 0; cpu < info->combined; cpu++)
773 if (info->core[cpu].core_id == address) {
774 /* The boot cpu dictates the cpu type. */
775 boot_core_type = info->core[cpu].type;
778 if (cpu >= info->combined)
779 panic("Could not find boot CPU type");
782 /* Set multi-threading state for the current system */
783 mtid = boot_core_type ? sclp.mtid : sclp.mtid_cp;
784 mtid = (mtid < smp_max_threads) ? mtid : smp_max_threads - 1;
787 /* Print number of CPUs */
789 for (cpu = 0; cpu < info->combined; cpu++) {
790 if (sclp.has_core_type &&
791 info->core[cpu].type != boot_core_type)
793 if (cpu < info->configured)
794 c_cpus += smp_cpu_mtid + 1;
796 s_cpus += smp_cpu_mtid + 1;
798 pr_info("%d configured CPUs, %d standby CPUs\n", c_cpus, s_cpus);
800 /* Add CPUs present at boot */
802 __smp_rescan_cpus(info, true);
808 * Activate a secondary processor.
810 static void smp_start_secondary(void *cpuvoid)
812 S390_lowcore.last_update_clock = get_tod_clock();
813 S390_lowcore.restart_stack = (unsigned long) restart_stack;
814 S390_lowcore.restart_fn = (unsigned long) do_restart;
815 S390_lowcore.restart_data = 0;
816 S390_lowcore.restart_source = -1UL;
817 restore_access_regs(S390_lowcore.access_regs_save_area);
818 __ctl_load(S390_lowcore.cregs_save_area, 0, 15);
819 __load_psw_mask(PSW_KERNEL_BITS | PSW_MASK_DAT);
825 notify_cpu_starting(smp_processor_id());
826 set_cpu_online(smp_processor_id(), true);
827 inc_irq_stat(CPU_RST);
829 cpu_startup_entry(CPUHP_AP_ONLINE_IDLE);
832 /* Upping and downing of CPUs */
833 int __cpu_up(unsigned int cpu, struct task_struct *tidle)
838 pcpu = pcpu_devices + cpu;
839 if (pcpu->state != CPU_STATE_CONFIGURED)
841 base = cpu - (cpu % (smp_cpu_mtid + 1));
842 for (i = 0; i <= smp_cpu_mtid; i++) {
843 if (base + i < nr_cpu_ids)
844 if (cpu_online(base + i))
848 * If this is the first CPU of the core to get online
849 * do an initial CPU reset.
851 if (i > smp_cpu_mtid &&
852 pcpu_sigp_retry(pcpu_devices + base, SIGP_INITIAL_CPU_RESET, 0) !=
853 SIGP_CC_ORDER_CODE_ACCEPTED)
856 rc = pcpu_alloc_lowcore(pcpu, cpu);
859 pcpu_prepare_secondary(pcpu, cpu);
860 pcpu_attach_task(pcpu, tidle);
861 pcpu_start_fn(pcpu, smp_start_secondary, NULL);
862 /* Wait until cpu puts itself in the online & active maps */
863 while (!cpu_online(cpu))
868 static unsigned int setup_possible_cpus __initdata;
870 static int __init _setup_possible_cpus(char *s)
872 get_option(&s, &setup_possible_cpus);
875 early_param("possible_cpus", _setup_possible_cpus);
877 #ifdef CONFIG_HOTPLUG_CPU
879 int __cpu_disable(void)
881 unsigned long cregs[16];
883 /* Handle possible pending IPIs */
884 smp_handle_ext_call();
885 set_cpu_online(smp_processor_id(), false);
886 /* Disable pseudo page faults on this cpu. */
888 /* Disable interrupt sources via control register. */
889 __ctl_store(cregs, 0, 15);
890 cregs[0] &= ~0x0000ee70UL; /* disable all external interrupts */
891 cregs[6] &= ~0xff000000UL; /* disable all I/O interrupts */
892 cregs[14] &= ~0x1f000000UL; /* disable most machine checks */
893 __ctl_load(cregs, 0, 15);
894 clear_cpu_flag(CIF_NOHZ_DELAY);
898 void __cpu_die(unsigned int cpu)
902 /* Wait until target cpu is down */
903 pcpu = pcpu_devices + cpu;
904 while (!pcpu_stopped(pcpu))
906 pcpu_free_lowcore(pcpu);
907 cpumask_clear_cpu(cpu, mm_cpumask(&init_mm));
908 cpumask_clear_cpu(cpu, &init_mm.context.cpu_attach_mask);
911 void __noreturn cpu_die(void)
915 pcpu_sigp_retry(pcpu_devices + smp_processor_id(), SIGP_STOP, 0);
919 #endif /* CONFIG_HOTPLUG_CPU */
921 void __init smp_fill_possible_mask(void)
923 unsigned int possible, sclp_max, cpu;
925 sclp_max = max(sclp.mtid, sclp.mtid_cp) + 1;
926 sclp_max = min(smp_max_threads, sclp_max);
927 sclp_max = (sclp.max_cores * sclp_max) ?: nr_cpu_ids;
928 possible = setup_possible_cpus ?: nr_cpu_ids;
929 possible = min(possible, sclp_max);
930 for (cpu = 0; cpu < possible && cpu < nr_cpu_ids; cpu++)
931 set_cpu_possible(cpu, true);
934 void __init smp_prepare_cpus(unsigned int max_cpus)
936 /* request the 0x1201 emergency signal external interrupt */
937 if (register_external_irq(EXT_IRQ_EMERGENCY_SIG, do_ext_call_interrupt))
938 panic("Couldn't request external interrupt 0x1201");
939 /* request the 0x1202 external call external interrupt */
940 if (register_external_irq(EXT_IRQ_EXTERNAL_CALL, do_ext_call_interrupt))
941 panic("Couldn't request external interrupt 0x1202");
945 void __init smp_prepare_boot_cpu(void)
947 struct pcpu *pcpu = pcpu_devices;
949 pcpu->state = CPU_STATE_CONFIGURED;
950 pcpu->address = stap();
951 pcpu->lowcore = (struct lowcore *)(unsigned long) store_prefix();
952 S390_lowcore.percpu_offset = __per_cpu_offset[0];
953 smp_cpu_set_polarization(0, POLARIZATION_UNKNOWN);
954 set_cpu_present(0, true);
955 set_cpu_online(0, true);
958 void __init smp_cpus_done(unsigned int max_cpus)
962 void __init smp_setup_processor_id(void)
964 S390_lowcore.cpu_nr = 0;
965 S390_lowcore.spinlock_lockval = arch_spin_lockval(0);
969 * the frequency of the profiling timer can be changed
970 * by writing a multiplier value into /proc/profile.
972 * usually you want to run this on all CPUs ;)
974 int setup_profiling_timer(unsigned int multiplier)
979 #ifdef CONFIG_HOTPLUG_CPU
980 static ssize_t cpu_configure_show(struct device *dev,
981 struct device_attribute *attr, char *buf)
985 mutex_lock(&smp_cpu_state_mutex);
986 count = sprintf(buf, "%d\n", pcpu_devices[dev->id].state);
987 mutex_unlock(&smp_cpu_state_mutex);
991 static ssize_t cpu_configure_store(struct device *dev,
992 struct device_attribute *attr,
993 const char *buf, size_t count)
999 if (sscanf(buf, "%d %c", &val, &delim) != 1)
1001 if (val != 0 && val != 1)
1004 mutex_lock(&smp_cpu_state_mutex);
1006 /* disallow configuration changes of online cpus and cpu 0 */
1008 cpu -= cpu % (smp_cpu_mtid + 1);
1011 for (i = 0; i <= smp_cpu_mtid; i++)
1012 if (cpu_online(cpu + i))
1014 pcpu = pcpu_devices + cpu;
1018 if (pcpu->state != CPU_STATE_CONFIGURED)
1020 rc = sclp_core_deconfigure(pcpu->address >> smp_cpu_mt_shift);
1023 for (i = 0; i <= smp_cpu_mtid; i++) {
1024 if (cpu + i >= nr_cpu_ids || !cpu_present(cpu + i))
1026 pcpu[i].state = CPU_STATE_STANDBY;
1027 smp_cpu_set_polarization(cpu + i,
1028 POLARIZATION_UNKNOWN);
1030 topology_expect_change();
1033 if (pcpu->state != CPU_STATE_STANDBY)
1035 rc = sclp_core_configure(pcpu->address >> smp_cpu_mt_shift);
1038 for (i = 0; i <= smp_cpu_mtid; i++) {
1039 if (cpu + i >= nr_cpu_ids || !cpu_present(cpu + i))
1041 pcpu[i].state = CPU_STATE_CONFIGURED;
1042 smp_cpu_set_polarization(cpu + i,
1043 POLARIZATION_UNKNOWN);
1045 topology_expect_change();
1051 mutex_unlock(&smp_cpu_state_mutex);
1053 return rc ? rc : count;
1055 static DEVICE_ATTR(configure, 0644, cpu_configure_show, cpu_configure_store);
1056 #endif /* CONFIG_HOTPLUG_CPU */
1058 static ssize_t show_cpu_address(struct device *dev,
1059 struct device_attribute *attr, char *buf)
1061 return sprintf(buf, "%d\n", pcpu_devices[dev->id].address);
1063 static DEVICE_ATTR(address, 0444, show_cpu_address, NULL);
1065 static struct attribute *cpu_common_attrs[] = {
1066 #ifdef CONFIG_HOTPLUG_CPU
1067 &dev_attr_configure.attr,
1069 &dev_attr_address.attr,
1073 static struct attribute_group cpu_common_attr_group = {
1074 .attrs = cpu_common_attrs,
1077 static struct attribute *cpu_online_attrs[] = {
1078 &dev_attr_idle_count.attr,
1079 &dev_attr_idle_time_us.attr,
1083 static struct attribute_group cpu_online_attr_group = {
1084 .attrs = cpu_online_attrs,
1087 static int smp_cpu_notify(struct notifier_block *self, unsigned long action,
1090 unsigned int cpu = (unsigned int)(long)hcpu;
1091 struct device *s = &per_cpu(cpu_device, cpu)->dev;
1094 switch (action & ~CPU_TASKS_FROZEN) {
1096 err = sysfs_create_group(&s->kobj, &cpu_online_attr_group);
1099 sysfs_remove_group(&s->kobj, &cpu_online_attr_group);
1102 return notifier_from_errno(err);
1105 static int smp_add_present_cpu(int cpu)
1111 c = kzalloc(sizeof(*c), GFP_KERNEL);
1114 per_cpu(cpu_device, cpu) = c;
1116 c->hotpluggable = 1;
1117 rc = register_cpu(c, cpu);
1120 rc = sysfs_create_group(&s->kobj, &cpu_common_attr_group);
1123 if (cpu_online(cpu)) {
1124 rc = sysfs_create_group(&s->kobj, &cpu_online_attr_group);
1128 rc = topology_cpu_init(c);
1134 if (cpu_online(cpu))
1135 sysfs_remove_group(&s->kobj, &cpu_online_attr_group);
1137 sysfs_remove_group(&s->kobj, &cpu_common_attr_group);
1139 #ifdef CONFIG_HOTPLUG_CPU
1146 #ifdef CONFIG_HOTPLUG_CPU
1148 int __ref smp_rescan_cpus(void)
1150 struct sclp_core_info *info;
1153 info = smp_get_core_info();
1157 mutex_lock(&smp_cpu_state_mutex);
1158 nr = __smp_rescan_cpus(info, false);
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(rescan, 0200, NULL, rescan_store);
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 cpu_notifier_register_begin();
1194 for_each_present_cpu(cpu) {
1195 rc = smp_add_present_cpu(cpu);
1200 __hotcpu_notifier(smp_cpu_notify, 0);
1203 cpu_notifier_register_done();
1206 subsys_initcall(s390_smp_init);