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/irq_work.h>
34 #include <linux/cpu.h>
35 #include <linux/slab.h>
36 #include <linux/sched/hotplug.h>
37 #include <linux/sched/task_stack.h>
38 #include <linux/crash_dump.h>
39 #include <linux/kprobes.h>
40 #include <asm/asm-offsets.h>
42 #include <asm/switch_to.h>
43 #include <asm/facility.h>
45 #include <asm/setup.h>
47 #include <asm/tlbflush.h>
48 #include <asm/vtimer.h>
49 #include <asm/lowcore.h>
51 #include <asm/debug.h>
52 #include <asm/os_info.h>
56 #include <asm/stacktrace.h>
57 #include <asm/topology.h>
63 ec_call_function_single,
74 static DEFINE_PER_CPU(struct cpu *, cpu_device);
77 unsigned long ec_mask; /* bit mask for ec_xxx functions */
78 unsigned long ec_clk; /* sigp timestamp for ec_xxx */
79 signed char state; /* physical cpu state */
80 signed char polarization; /* physical polarization */
81 u16 address; /* physical cpu address */
84 static u8 boot_core_type;
85 static struct pcpu pcpu_devices[NR_CPUS];
87 unsigned int smp_cpu_mt_shift;
88 EXPORT_SYMBOL(smp_cpu_mt_shift);
90 unsigned int smp_cpu_mtid;
91 EXPORT_SYMBOL(smp_cpu_mtid);
93 #ifdef CONFIG_CRASH_DUMP
94 __vector128 __initdata boot_cpu_vector_save_area[__NUM_VXRS];
97 static unsigned int smp_max_threads __initdata = -1U;
99 static int __init early_nosmt(char *s)
104 early_param("nosmt", early_nosmt);
106 static int __init early_smt(char *s)
108 get_option(&s, &smp_max_threads);
111 early_param("smt", early_smt);
114 * The smp_cpu_state_mutex must be held when changing the state or polarization
115 * member of a pcpu data structure within the pcpu_devices arreay.
117 DEFINE_MUTEX(smp_cpu_state_mutex);
120 * Signal processor helper functions.
122 static inline int __pcpu_sigp_relax(u16 addr, u8 order, unsigned long parm)
127 cc = __pcpu_sigp(addr, order, parm, NULL);
128 if (cc != SIGP_CC_BUSY)
134 static int pcpu_sigp_retry(struct pcpu *pcpu, u8 order, u32 parm)
138 for (retry = 0; ; retry++) {
139 cc = __pcpu_sigp(pcpu->address, order, parm, NULL);
140 if (cc != SIGP_CC_BUSY)
148 static inline int pcpu_stopped(struct pcpu *pcpu)
152 if (__pcpu_sigp(pcpu->address, SIGP_SENSE,
153 0, &status) != SIGP_CC_STATUS_STORED)
155 return !!(status & (SIGP_STATUS_CHECK_STOP|SIGP_STATUS_STOPPED));
158 static inline int pcpu_running(struct pcpu *pcpu)
160 if (__pcpu_sigp(pcpu->address, SIGP_SENSE_RUNNING,
161 0, NULL) != SIGP_CC_STATUS_STORED)
163 /* Status stored condition code is equivalent to cpu not running. */
168 * Find struct pcpu by cpu address.
170 static struct pcpu *pcpu_find_address(const struct cpumask *mask, u16 address)
174 for_each_cpu(cpu, mask)
175 if (pcpu_devices[cpu].address == address)
176 return pcpu_devices + cpu;
180 static void pcpu_ec_call(struct pcpu *pcpu, int ec_bit)
184 if (test_and_set_bit(ec_bit, &pcpu->ec_mask))
186 order = pcpu_running(pcpu) ? SIGP_EXTERNAL_CALL : SIGP_EMERGENCY_SIGNAL;
187 pcpu->ec_clk = get_tod_clock_fast();
188 pcpu_sigp_retry(pcpu, order, 0);
191 static int pcpu_alloc_lowcore(struct pcpu *pcpu, int cpu)
193 unsigned long async_stack, nodat_stack, mcck_stack;
196 lc = (struct lowcore *) __get_free_pages(GFP_KERNEL | GFP_DMA, LC_ORDER);
197 nodat_stack = __get_free_pages(GFP_KERNEL, THREAD_SIZE_ORDER);
198 async_stack = stack_alloc();
199 mcck_stack = stack_alloc();
200 if (!lc || !nodat_stack || !async_stack || !mcck_stack)
202 memcpy(lc, &S390_lowcore, 512);
203 memset((char *) lc + 512, 0, sizeof(*lc) - 512);
204 lc->async_stack = async_stack + STACK_INIT_OFFSET;
205 lc->nodat_stack = nodat_stack + STACK_INIT_OFFSET;
206 lc->mcck_stack = mcck_stack + STACK_INIT_OFFSET;
208 lc->spinlock_lockval = arch_spin_lockval(cpu);
209 lc->spinlock_index = 0;
210 lc->br_r1_trampoline = 0x07f1; /* br %r1 */
211 lc->return_lpswe = gen_lpswe(__LC_RETURN_PSW);
212 lc->return_mcck_lpswe = gen_lpswe(__LC_RETURN_MCCK_PSW);
213 if (nmi_alloc_per_cpu(lc))
215 lowcore_ptr[cpu] = lc;
216 pcpu_sigp_retry(pcpu, SIGP_SET_PREFIX, (u32)(unsigned long) lc);
220 stack_free(mcck_stack);
221 stack_free(async_stack);
222 free_pages(nodat_stack, THREAD_SIZE_ORDER);
223 free_pages((unsigned long) lc, LC_ORDER);
227 static void pcpu_free_lowcore(struct pcpu *pcpu)
229 unsigned long async_stack, nodat_stack, mcck_stack;
233 cpu = pcpu - pcpu_devices;
234 lc = lowcore_ptr[cpu];
235 nodat_stack = lc->nodat_stack - STACK_INIT_OFFSET;
236 async_stack = lc->async_stack - STACK_INIT_OFFSET;
237 mcck_stack = lc->mcck_stack - STACK_INIT_OFFSET;
238 pcpu_sigp_retry(pcpu, SIGP_SET_PREFIX, 0);
239 lowcore_ptr[cpu] = NULL;
240 nmi_free_per_cpu(lc);
241 stack_free(async_stack);
242 stack_free(mcck_stack);
243 free_pages(nodat_stack, THREAD_SIZE_ORDER);
244 free_pages((unsigned long) lc, LC_ORDER);
247 static void pcpu_prepare_secondary(struct pcpu *pcpu, int cpu)
249 struct lowcore *lc = lowcore_ptr[cpu];
251 cpumask_set_cpu(cpu, &init_mm.context.cpu_attach_mask);
252 cpumask_set_cpu(cpu, mm_cpumask(&init_mm));
254 lc->spinlock_lockval = arch_spin_lockval(cpu);
255 lc->spinlock_index = 0;
256 lc->percpu_offset = __per_cpu_offset[cpu];
257 lc->kernel_asce = S390_lowcore.kernel_asce;
258 lc->user_asce = s390_invalid_asce;
259 lc->machine_flags = S390_lowcore.machine_flags;
260 lc->user_timer = lc->system_timer =
261 lc->steal_timer = lc->avg_steal_timer = 0;
262 __ctl_store(lc->cregs_save_area, 0, 15);
263 lc->cregs_save_area[1] = lc->kernel_asce;
264 lc->cregs_save_area[7] = lc->user_asce;
265 save_access_regs((unsigned int *) lc->access_regs_save_area);
266 arch_spin_lock_setup(cpu);
269 static void pcpu_attach_task(struct pcpu *pcpu, struct task_struct *tsk)
274 cpu = pcpu - pcpu_devices;
275 lc = lowcore_ptr[cpu];
276 lc->kernel_stack = (unsigned long) task_stack_page(tsk)
277 + THREAD_SIZE - STACK_FRAME_OVERHEAD - sizeof(struct pt_regs);
278 lc->current_task = (unsigned long) tsk;
280 lc->current_pid = tsk->pid;
281 lc->user_timer = tsk->thread.user_timer;
282 lc->guest_timer = tsk->thread.guest_timer;
283 lc->system_timer = tsk->thread.system_timer;
284 lc->hardirq_timer = tsk->thread.hardirq_timer;
285 lc->softirq_timer = tsk->thread.softirq_timer;
289 static void pcpu_start_fn(struct pcpu *pcpu, void (*func)(void *), void *data)
294 cpu = pcpu - pcpu_devices;
295 lc = lowcore_ptr[cpu];
296 lc->restart_stack = lc->nodat_stack;
297 lc->restart_fn = (unsigned long) func;
298 lc->restart_data = (unsigned long) data;
299 lc->restart_source = -1UL;
300 pcpu_sigp_retry(pcpu, SIGP_RESTART, 0);
304 * Call function via PSW restart on pcpu and stop the current cpu.
306 static void __pcpu_delegate(void (*func)(void*), void *data)
308 func(data); /* should not return */
311 static void __no_sanitize_address pcpu_delegate(struct pcpu *pcpu,
312 void (*func)(void *),
313 void *data, unsigned long stack)
315 struct lowcore *lc = lowcore_ptr[pcpu - pcpu_devices];
316 unsigned long source_cpu = stap();
318 __load_psw_mask(PSW_KERNEL_BITS | PSW_MASK_DAT);
319 if (pcpu->address == source_cpu)
320 CALL_ON_STACK(__pcpu_delegate, stack, 2, func, data);
321 /* Stop target cpu (if func returns this stops the current cpu). */
322 pcpu_sigp_retry(pcpu, SIGP_STOP, 0);
323 /* Restart func on the target cpu and stop the current cpu. */
324 mem_assign_absolute(lc->restart_stack, stack);
325 mem_assign_absolute(lc->restart_fn, (unsigned long) func);
326 mem_assign_absolute(lc->restart_data, (unsigned long) data);
327 mem_assign_absolute(lc->restart_source, source_cpu);
330 "0: sigp 0,%0,%2 # sigp restart to target cpu\n"
331 " brc 2,0b # busy, try again\n"
332 "1: sigp 0,%1,%3 # sigp stop to current cpu\n"
333 " brc 2,1b # busy, try again\n"
334 : : "d" (pcpu->address), "d" (source_cpu),
335 "K" (SIGP_RESTART), "K" (SIGP_STOP)
341 * Enable additional logical cpus for multi-threading.
343 static int pcpu_set_smt(unsigned int mtid)
347 if (smp_cpu_mtid == mtid)
349 cc = __pcpu_sigp(0, SIGP_SET_MULTI_THREADING, mtid, NULL);
352 smp_cpu_mt_shift = 0;
353 while (smp_cpu_mtid >= (1U << smp_cpu_mt_shift))
355 pcpu_devices[0].address = stap();
361 * Call function on an online CPU.
363 void smp_call_online_cpu(void (*func)(void *), void *data)
367 /* Use the current cpu if it is online. */
368 pcpu = pcpu_find_address(cpu_online_mask, stap());
370 /* Use the first online cpu. */
371 pcpu = pcpu_devices + cpumask_first(cpu_online_mask);
372 pcpu_delegate(pcpu, func, data, (unsigned long) restart_stack);
376 * Call function on the ipl CPU.
378 void smp_call_ipl_cpu(void (*func)(void *), void *data)
380 struct lowcore *lc = lowcore_ptr[0];
382 if (pcpu_devices[0].address == stap())
385 pcpu_delegate(&pcpu_devices[0], func, data,
389 int smp_find_processor_id(u16 address)
393 for_each_present_cpu(cpu)
394 if (pcpu_devices[cpu].address == address)
399 void schedule_mcck_handler(void)
401 pcpu_ec_call(pcpu_devices + smp_processor_id(), ec_mcck_pending);
404 bool notrace arch_vcpu_is_preempted(int cpu)
406 if (test_cpu_flag_of(CIF_ENABLED_WAIT, cpu))
408 if (pcpu_running(pcpu_devices + cpu))
412 EXPORT_SYMBOL(arch_vcpu_is_preempted);
414 void notrace smp_yield_cpu(int cpu)
416 if (!MACHINE_HAS_DIAG9C)
418 diag_stat_inc_norecursion(DIAG_STAT_X09C);
419 asm volatile("diag %0,0,0x9c"
420 : : "d" (pcpu_devices[cpu].address));
422 EXPORT_SYMBOL_GPL(smp_yield_cpu);
425 * Send cpus emergency shutdown signal. This gives the cpus the
426 * opportunity to complete outstanding interrupts.
428 void notrace smp_emergency_stop(void)
430 static arch_spinlock_t lock = __ARCH_SPIN_LOCK_UNLOCKED;
431 static cpumask_t cpumask;
435 arch_spin_lock(&lock);
436 cpumask_copy(&cpumask, cpu_online_mask);
437 cpumask_clear_cpu(smp_processor_id(), &cpumask);
439 end = get_tod_clock() + (1000000UL << 12);
440 for_each_cpu(cpu, &cpumask) {
441 struct pcpu *pcpu = pcpu_devices + cpu;
442 set_bit(ec_stop_cpu, &pcpu->ec_mask);
443 while (__pcpu_sigp(pcpu->address, SIGP_EMERGENCY_SIGNAL,
444 0, NULL) == SIGP_CC_BUSY &&
445 get_tod_clock() < end)
448 while (get_tod_clock() < end) {
449 for_each_cpu(cpu, &cpumask)
450 if (pcpu_stopped(pcpu_devices + cpu))
451 cpumask_clear_cpu(cpu, &cpumask);
452 if (cpumask_empty(&cpumask))
456 arch_spin_unlock(&lock);
458 NOKPROBE_SYMBOL(smp_emergency_stop);
461 * Stop all cpus but the current one.
463 void smp_send_stop(void)
467 /* Disable all interrupts/machine checks */
468 __load_psw_mask(PSW_KERNEL_BITS | PSW_MASK_DAT);
469 trace_hardirqs_off();
471 debug_set_critical();
473 if (oops_in_progress)
474 smp_emergency_stop();
476 /* stop all processors */
477 for_each_online_cpu(cpu) {
478 if (cpu == smp_processor_id())
480 pcpu_sigp_retry(pcpu_devices + cpu, SIGP_STOP, 0);
481 while (!pcpu_stopped(pcpu_devices + cpu))
487 * This is the main routine where commands issued by other
490 static void smp_handle_ext_call(void)
494 /* handle bit signal external calls */
495 bits = xchg(&pcpu_devices[smp_processor_id()].ec_mask, 0);
496 if (test_bit(ec_stop_cpu, &bits))
498 if (test_bit(ec_schedule, &bits))
500 if (test_bit(ec_call_function_single, &bits))
501 generic_smp_call_function_single_interrupt();
502 if (test_bit(ec_mcck_pending, &bits))
503 __s390_handle_mcck();
504 if (test_bit(ec_irq_work, &bits))
508 static void do_ext_call_interrupt(struct ext_code ext_code,
509 unsigned int param32, unsigned long param64)
511 inc_irq_stat(ext_code.code == 0x1202 ? IRQEXT_EXC : IRQEXT_EMS);
512 smp_handle_ext_call();
515 void arch_send_call_function_ipi_mask(const struct cpumask *mask)
519 for_each_cpu(cpu, mask)
520 pcpu_ec_call(pcpu_devices + cpu, ec_call_function_single);
523 void arch_send_call_function_single_ipi(int cpu)
525 pcpu_ec_call(pcpu_devices + cpu, ec_call_function_single);
529 * this function sends a 'reschedule' IPI to another CPU.
530 * it goes straight through and wastes no time serializing
531 * anything. Worst case is that we lose a reschedule ...
533 void smp_send_reschedule(int cpu)
535 pcpu_ec_call(pcpu_devices + cpu, ec_schedule);
538 #ifdef CONFIG_IRQ_WORK
539 void arch_irq_work_raise(void)
541 pcpu_ec_call(pcpu_devices + smp_processor_id(), ec_irq_work);
546 * parameter area for the set/clear control bit callbacks
548 struct ec_creg_mask_parms {
550 unsigned long andval;
555 * callback for setting/clearing control bits
557 static void smp_ctl_bit_callback(void *info)
559 struct ec_creg_mask_parms *pp = info;
560 unsigned long cregs[16];
562 __ctl_store(cregs, 0, 15);
563 cregs[pp->cr] = (cregs[pp->cr] & pp->andval) | pp->orval;
564 __ctl_load(cregs, 0, 15);
568 * Set a bit in a control register of all cpus
570 void smp_ctl_set_bit(int cr, int bit)
572 struct ec_creg_mask_parms parms = { 1UL << bit, -1UL, cr };
574 on_each_cpu(smp_ctl_bit_callback, &parms, 1);
576 EXPORT_SYMBOL(smp_ctl_set_bit);
579 * Clear a bit in a control register of all cpus
581 void smp_ctl_clear_bit(int cr, int bit)
583 struct ec_creg_mask_parms parms = { 0, ~(1UL << bit), cr };
585 on_each_cpu(smp_ctl_bit_callback, &parms, 1);
587 EXPORT_SYMBOL(smp_ctl_clear_bit);
589 #ifdef CONFIG_CRASH_DUMP
591 int smp_store_status(int cpu)
597 pcpu = pcpu_devices + cpu;
598 lc = lowcore_ptr[cpu];
599 pa = __pa(&lc->floating_pt_save_area);
600 if (__pcpu_sigp_relax(pcpu->address, SIGP_STORE_STATUS_AT_ADDRESS,
601 pa) != SIGP_CC_ORDER_CODE_ACCEPTED)
603 if (!MACHINE_HAS_VX && !MACHINE_HAS_GS)
605 pa = __pa(lc->mcesad & MCESA_ORIGIN_MASK);
607 pa |= lc->mcesad & MCESA_LC_MASK;
608 if (__pcpu_sigp_relax(pcpu->address, SIGP_STORE_ADDITIONAL_STATUS,
609 pa) != SIGP_CC_ORDER_CODE_ACCEPTED)
615 * Collect CPU state of the previous, crashed system.
616 * There are four cases:
617 * 1) standard zfcp/nvme dump
618 * condition: OLDMEM_BASE == NULL && is_ipl_type_dump() == true
619 * The state for all CPUs except the boot CPU needs to be collected
620 * with sigp stop-and-store-status. The boot CPU state is located in
621 * the absolute lowcore of the memory stored in the HSA. The zcore code
622 * will copy the boot CPU state from the HSA.
623 * 2) stand-alone kdump for SCSI/NVMe (zfcp/nvme dump with swapped memory)
624 * condition: OLDMEM_BASE != NULL && is_ipl_type_dump() == true
625 * The state for all CPUs except the boot CPU needs to be collected
626 * with sigp stop-and-store-status. The firmware or the boot-loader
627 * stored the registers of the boot CPU in the absolute lowcore in the
628 * memory of the old system.
629 * 3) kdump and the old kernel did not store the CPU state,
630 * or stand-alone kdump for DASD
631 * condition: OLDMEM_BASE != NULL && !is_kdump_kernel()
632 * The state for all CPUs except the boot CPU needs to be collected
633 * with sigp stop-and-store-status. The kexec code or the boot-loader
634 * stored the registers of the boot CPU in the memory of the old system.
635 * 4) kdump and the old kernel stored the CPU state
636 * condition: OLDMEM_BASE != NULL && is_kdump_kernel()
637 * This case does not exist for s390 anymore, setup_arch explicitly
638 * deactivates the elfcorehdr= kernel parameter
640 static __init void smp_save_cpu_vxrs(struct save_area *sa, u16 addr,
641 bool is_boot_cpu, unsigned long page)
643 __vector128 *vxrs = (__vector128 *) page;
646 vxrs = boot_cpu_vector_save_area;
648 __pcpu_sigp_relax(addr, SIGP_STORE_ADDITIONAL_STATUS, page);
649 save_area_add_vxrs(sa, vxrs);
652 static __init void smp_save_cpu_regs(struct save_area *sa, u16 addr,
653 bool is_boot_cpu, unsigned long page)
655 void *regs = (void *) page;
658 copy_oldmem_kernel(regs, (void *) __LC_FPREGS_SAVE_AREA, 512);
660 __pcpu_sigp_relax(addr, SIGP_STORE_STATUS_AT_ADDRESS, page);
661 save_area_add_regs(sa, regs);
664 void __init smp_save_dump_cpus(void)
666 int addr, boot_cpu_addr, max_cpu_addr;
667 struct save_area *sa;
671 if (!(OLDMEM_BASE || is_ipl_type_dump()))
672 /* No previous system present, normal boot. */
674 /* Allocate a page as dumping area for the store status sigps */
675 page = memblock_phys_alloc_range(PAGE_SIZE, PAGE_SIZE, 0, 1UL << 31);
677 panic("ERROR: Failed to allocate %lx bytes below %lx\n",
678 PAGE_SIZE, 1UL << 31);
680 /* Set multi-threading state to the previous system. */
681 pcpu_set_smt(sclp.mtid_prev);
682 boot_cpu_addr = stap();
683 max_cpu_addr = SCLP_MAX_CORES << sclp.mtid_prev;
684 for (addr = 0; addr <= max_cpu_addr; addr++) {
685 if (__pcpu_sigp_relax(addr, SIGP_SENSE, 0) ==
686 SIGP_CC_NOT_OPERATIONAL)
688 is_boot_cpu = (addr == boot_cpu_addr);
689 /* Allocate save area */
690 sa = save_area_alloc(is_boot_cpu);
692 panic("could not allocate memory for save area\n");
694 /* Get the vector registers */
695 smp_save_cpu_vxrs(sa, addr, is_boot_cpu, page);
697 * For a zfcp/nvme dump OLDMEM_BASE == NULL and the registers
698 * of the boot CPU are stored in the HSA. To retrieve
699 * these registers an SCLP request is required which is
700 * done by drivers/s390/char/zcore.c:init_cpu_info()
702 if (!is_boot_cpu || OLDMEM_BASE)
703 /* Get the CPU registers */
704 smp_save_cpu_regs(sa, addr, is_boot_cpu, page);
706 memblock_free(page, PAGE_SIZE);
707 diag_dma_ops.diag308_reset();
710 #endif /* CONFIG_CRASH_DUMP */
712 void smp_cpu_set_polarization(int cpu, int val)
714 pcpu_devices[cpu].polarization = val;
717 int smp_cpu_get_polarization(int cpu)
719 return pcpu_devices[cpu].polarization;
722 int smp_cpu_get_cpu_address(int cpu)
724 return pcpu_devices[cpu].address;
727 static void __ref smp_get_core_info(struct sclp_core_info *info, int early)
729 static int use_sigp_detection;
732 if (use_sigp_detection || sclp_get_core_info(info, early)) {
733 use_sigp_detection = 1;
735 address < (SCLP_MAX_CORES << smp_cpu_mt_shift);
736 address += (1U << smp_cpu_mt_shift)) {
737 if (__pcpu_sigp_relax(address, SIGP_SENSE, 0) ==
738 SIGP_CC_NOT_OPERATIONAL)
740 info->core[info->configured].core_id =
741 address >> smp_cpu_mt_shift;
744 info->combined = info->configured;
748 static int smp_add_present_cpu(int cpu);
750 static int smp_add_core(struct sclp_core_entry *core, cpumask_t *avail,
751 bool configured, bool early)
758 if (sclp.has_core_type && core->type != boot_core_type)
760 cpu = cpumask_first(avail);
761 address = core->core_id << smp_cpu_mt_shift;
762 for (i = 0; (i <= smp_cpu_mtid) && (cpu < nr_cpu_ids); i++) {
763 if (pcpu_find_address(cpu_present_mask, address + i))
765 pcpu = pcpu_devices + cpu;
766 pcpu->address = address + i;
768 pcpu->state = CPU_STATE_CONFIGURED;
770 pcpu->state = CPU_STATE_STANDBY;
771 smp_cpu_set_polarization(cpu, POLARIZATION_UNKNOWN);
772 set_cpu_present(cpu, true);
773 if (!early && smp_add_present_cpu(cpu) != 0)
774 set_cpu_present(cpu, false);
777 cpumask_clear_cpu(cpu, avail);
778 cpu = cpumask_next(cpu, avail);
783 static int __smp_rescan_cpus(struct sclp_core_info *info, bool early)
785 struct sclp_core_entry *core;
786 static cpumask_t avail;
792 mutex_lock(&smp_cpu_state_mutex);
794 cpumask_xor(&avail, cpu_possible_mask, cpu_present_mask);
796 * Add IPL core first (which got logical CPU number 0) to make sure
797 * that all SMT threads get subsequent logical CPU numbers.
800 core_id = pcpu_devices[0].address >> smp_cpu_mt_shift;
801 for (i = 0; i < info->configured; i++) {
802 core = &info->core[i];
803 if (core->core_id == core_id) {
804 nr += smp_add_core(core, &avail, true, early);
809 for (i = 0; i < info->combined; i++) {
810 configured = i < info->configured;
811 nr += smp_add_core(&info->core[i], &avail, configured, early);
813 mutex_unlock(&smp_cpu_state_mutex);
818 void __init smp_detect_cpus(void)
820 unsigned int cpu, mtid, c_cpus, s_cpus;
821 struct sclp_core_info *info;
824 /* Get CPU information */
825 info = memblock_alloc(sizeof(*info), 8);
827 panic("%s: Failed to allocate %zu bytes align=0x%x\n",
828 __func__, sizeof(*info), 8);
829 smp_get_core_info(info, 1);
830 /* Find boot CPU type */
831 if (sclp.has_core_type) {
833 for (cpu = 0; cpu < info->combined; cpu++)
834 if (info->core[cpu].core_id == address) {
835 /* The boot cpu dictates the cpu type. */
836 boot_core_type = info->core[cpu].type;
839 if (cpu >= info->combined)
840 panic("Could not find boot CPU type");
843 /* Set multi-threading state for the current system */
844 mtid = boot_core_type ? sclp.mtid : sclp.mtid_cp;
845 mtid = (mtid < smp_max_threads) ? mtid : smp_max_threads - 1;
848 /* Print number of CPUs */
850 for (cpu = 0; cpu < info->combined; cpu++) {
851 if (sclp.has_core_type &&
852 info->core[cpu].type != boot_core_type)
854 if (cpu < info->configured)
855 c_cpus += smp_cpu_mtid + 1;
857 s_cpus += smp_cpu_mtid + 1;
859 pr_info("%d configured CPUs, %d standby CPUs\n", c_cpus, s_cpus);
861 /* Add CPUs present at boot */
862 __smp_rescan_cpus(info, true);
863 memblock_free_early((unsigned long)info, sizeof(*info));
866 static void smp_init_secondary(void)
868 int cpu = raw_smp_processor_id();
870 S390_lowcore.last_update_clock = get_tod_clock();
871 restore_access_regs(S390_lowcore.access_regs_save_area);
873 rcu_cpu_starting(cpu);
878 notify_cpu_starting(cpu);
879 if (topology_cpu_dedicated(cpu))
880 set_cpu_flag(CIF_DEDICATED_CPU);
882 clear_cpu_flag(CIF_DEDICATED_CPU);
883 set_cpu_online(cpu, true);
885 inc_irq_stat(CPU_RST);
887 cpu_startup_entry(CPUHP_AP_ONLINE_IDLE);
891 * Activate a secondary processor.
893 static void __no_sanitize_address smp_start_secondary(void *cpuvoid)
895 S390_lowcore.restart_stack = (unsigned long) restart_stack;
896 S390_lowcore.restart_fn = (unsigned long) do_restart;
897 S390_lowcore.restart_data = 0;
898 S390_lowcore.restart_source = -1UL;
899 __ctl_load(S390_lowcore.cregs_save_area, 0, 15);
900 __load_psw_mask(PSW_KERNEL_BITS | PSW_MASK_DAT);
901 CALL_ON_STACK_NORETURN(smp_init_secondary, S390_lowcore.kernel_stack);
904 /* Upping and downing of CPUs */
905 int __cpu_up(unsigned int cpu, struct task_struct *tidle)
907 struct pcpu *pcpu = pcpu_devices + cpu;
910 if (pcpu->state != CPU_STATE_CONFIGURED)
912 if (pcpu_sigp_retry(pcpu, SIGP_INITIAL_CPU_RESET, 0) !=
913 SIGP_CC_ORDER_CODE_ACCEPTED)
916 rc = pcpu_alloc_lowcore(pcpu, cpu);
919 pcpu_prepare_secondary(pcpu, cpu);
920 pcpu_attach_task(pcpu, tidle);
921 pcpu_start_fn(pcpu, smp_start_secondary, NULL);
922 /* Wait until cpu puts itself in the online & active maps */
923 while (!cpu_online(cpu))
928 static unsigned int setup_possible_cpus __initdata;
930 static int __init _setup_possible_cpus(char *s)
932 get_option(&s, &setup_possible_cpus);
935 early_param("possible_cpus", _setup_possible_cpus);
937 int __cpu_disable(void)
939 unsigned long cregs[16];
941 /* Handle possible pending IPIs */
942 smp_handle_ext_call();
943 set_cpu_online(smp_processor_id(), false);
945 /* Disable pseudo page faults on this cpu. */
947 /* Disable interrupt sources via control register. */
948 __ctl_store(cregs, 0, 15);
949 cregs[0] &= ~0x0000ee70UL; /* disable all external interrupts */
950 cregs[6] &= ~0xff000000UL; /* disable all I/O interrupts */
951 cregs[14] &= ~0x1f000000UL; /* disable most machine checks */
952 __ctl_load(cregs, 0, 15);
953 clear_cpu_flag(CIF_NOHZ_DELAY);
957 void __cpu_die(unsigned int cpu)
961 /* Wait until target cpu is down */
962 pcpu = pcpu_devices + cpu;
963 while (!pcpu_stopped(pcpu))
965 pcpu_free_lowcore(pcpu);
966 cpumask_clear_cpu(cpu, mm_cpumask(&init_mm));
967 cpumask_clear_cpu(cpu, &init_mm.context.cpu_attach_mask);
970 void __noreturn cpu_die(void)
974 pcpu_sigp_retry(pcpu_devices + smp_processor_id(), SIGP_STOP, 0);
978 void __init smp_fill_possible_mask(void)
980 unsigned int possible, sclp_max, cpu;
982 sclp_max = max(sclp.mtid, sclp.mtid_cp) + 1;
983 sclp_max = min(smp_max_threads, sclp_max);
984 sclp_max = (sclp.max_cores * sclp_max) ?: nr_cpu_ids;
985 possible = setup_possible_cpus ?: nr_cpu_ids;
986 possible = min(possible, sclp_max);
987 for (cpu = 0; cpu < possible && cpu < nr_cpu_ids; cpu++)
988 set_cpu_possible(cpu, true);
991 void __init smp_prepare_cpus(unsigned int max_cpus)
993 /* request the 0x1201 emergency signal external interrupt */
994 if (register_external_irq(EXT_IRQ_EMERGENCY_SIG, do_ext_call_interrupt))
995 panic("Couldn't request external interrupt 0x1201");
996 /* request the 0x1202 external call external interrupt */
997 if (register_external_irq(EXT_IRQ_EXTERNAL_CALL, do_ext_call_interrupt))
998 panic("Couldn't request external interrupt 0x1202");
1001 void __init smp_prepare_boot_cpu(void)
1003 struct pcpu *pcpu = pcpu_devices;
1005 WARN_ON(!cpu_present(0) || !cpu_online(0));
1006 pcpu->state = CPU_STATE_CONFIGURED;
1007 S390_lowcore.percpu_offset = __per_cpu_offset[0];
1008 smp_cpu_set_polarization(0, POLARIZATION_UNKNOWN);
1011 void __init smp_setup_processor_id(void)
1013 pcpu_devices[0].address = stap();
1014 S390_lowcore.cpu_nr = 0;
1015 S390_lowcore.spinlock_lockval = arch_spin_lockval(0);
1016 S390_lowcore.spinlock_index = 0;
1020 * the frequency of the profiling timer can be changed
1021 * by writing a multiplier value into /proc/profile.
1023 * usually you want to run this on all CPUs ;)
1025 int setup_profiling_timer(unsigned int multiplier)
1030 static ssize_t cpu_configure_show(struct device *dev,
1031 struct device_attribute *attr, char *buf)
1035 mutex_lock(&smp_cpu_state_mutex);
1036 count = sprintf(buf, "%d\n", pcpu_devices[dev->id].state);
1037 mutex_unlock(&smp_cpu_state_mutex);
1041 static ssize_t cpu_configure_store(struct device *dev,
1042 struct device_attribute *attr,
1043 const char *buf, size_t count)
1046 int cpu, val, rc, i;
1049 if (sscanf(buf, "%d %c", &val, &delim) != 1)
1051 if (val != 0 && val != 1)
1054 mutex_lock(&smp_cpu_state_mutex);
1056 /* disallow configuration changes of online cpus and cpu 0 */
1058 cpu = smp_get_base_cpu(cpu);
1061 for (i = 0; i <= smp_cpu_mtid; i++)
1062 if (cpu_online(cpu + i))
1064 pcpu = pcpu_devices + cpu;
1068 if (pcpu->state != CPU_STATE_CONFIGURED)
1070 rc = sclp_core_deconfigure(pcpu->address >> smp_cpu_mt_shift);
1073 for (i = 0; i <= smp_cpu_mtid; i++) {
1074 if (cpu + i >= nr_cpu_ids || !cpu_present(cpu + i))
1076 pcpu[i].state = CPU_STATE_STANDBY;
1077 smp_cpu_set_polarization(cpu + i,
1078 POLARIZATION_UNKNOWN);
1080 topology_expect_change();
1083 if (pcpu->state != CPU_STATE_STANDBY)
1085 rc = sclp_core_configure(pcpu->address >> smp_cpu_mt_shift);
1088 for (i = 0; i <= smp_cpu_mtid; i++) {
1089 if (cpu + i >= nr_cpu_ids || !cpu_present(cpu + i))
1091 pcpu[i].state = CPU_STATE_CONFIGURED;
1092 smp_cpu_set_polarization(cpu + i,
1093 POLARIZATION_UNKNOWN);
1095 topology_expect_change();
1101 mutex_unlock(&smp_cpu_state_mutex);
1103 return rc ? rc : count;
1105 static DEVICE_ATTR(configure, 0644, cpu_configure_show, cpu_configure_store);
1107 static ssize_t show_cpu_address(struct device *dev,
1108 struct device_attribute *attr, char *buf)
1110 return sprintf(buf, "%d\n", pcpu_devices[dev->id].address);
1112 static DEVICE_ATTR(address, 0444, show_cpu_address, NULL);
1114 static struct attribute *cpu_common_attrs[] = {
1115 &dev_attr_configure.attr,
1116 &dev_attr_address.attr,
1120 static struct attribute_group cpu_common_attr_group = {
1121 .attrs = cpu_common_attrs,
1124 static struct attribute *cpu_online_attrs[] = {
1125 &dev_attr_idle_count.attr,
1126 &dev_attr_idle_time_us.attr,
1130 static struct attribute_group cpu_online_attr_group = {
1131 .attrs = cpu_online_attrs,
1134 static int smp_cpu_online(unsigned int cpu)
1136 struct device *s = &per_cpu(cpu_device, cpu)->dev;
1138 return sysfs_create_group(&s->kobj, &cpu_online_attr_group);
1141 static int smp_cpu_pre_down(unsigned int cpu)
1143 struct device *s = &per_cpu(cpu_device, cpu)->dev;
1145 sysfs_remove_group(&s->kobj, &cpu_online_attr_group);
1149 static int smp_add_present_cpu(int cpu)
1155 c = kzalloc(sizeof(*c), GFP_KERNEL);
1158 per_cpu(cpu_device, cpu) = c;
1160 c->hotpluggable = 1;
1161 rc = register_cpu(c, cpu);
1164 rc = sysfs_create_group(&s->kobj, &cpu_common_attr_group);
1167 rc = topology_cpu_init(c);
1173 sysfs_remove_group(&s->kobj, &cpu_common_attr_group);
1180 int __ref smp_rescan_cpus(void)
1182 struct sclp_core_info *info;
1185 info = kzalloc(sizeof(*info), GFP_KERNEL);
1188 smp_get_core_info(info, 0);
1189 nr = __smp_rescan_cpus(info, false);
1192 topology_schedule_update();
1196 static ssize_t __ref rescan_store(struct device *dev,
1197 struct device_attribute *attr,
1203 rc = lock_device_hotplug_sysfs();
1206 rc = smp_rescan_cpus();
1207 unlock_device_hotplug();
1208 return rc ? rc : count;
1210 static DEVICE_ATTR_WO(rescan);
1212 static int __init s390_smp_init(void)
1216 rc = device_create_file(cpu_subsys.dev_root, &dev_attr_rescan);
1219 for_each_present_cpu(cpu) {
1220 rc = smp_add_present_cpu(cpu);
1225 rc = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "s390/smp:online",
1226 smp_cpu_online, smp_cpu_pre_down);
1227 rc = rc <= 0 ? rc : 0;
1231 subsys_initcall(s390_smp_init);
1233 static __always_inline void set_new_lowcore(struct lowcore *lc)
1235 union register_pair dst, src;
1238 src.even = (unsigned long) &S390_lowcore;
1239 src.odd = sizeof(S390_lowcore);
1240 dst.even = (unsigned long) lc;
1241 dst.odd = sizeof(*lc);
1242 pfx = (unsigned long) lc;
1245 " mvcl %[dst],%[src]\n"
1247 : [dst] "+&d" (dst.pair), [src] "+&d" (src.pair)
1252 static int __init smp_reinit_ipl_cpu(void)
1254 unsigned long async_stack, nodat_stack, mcck_stack;
1255 struct lowcore *lc, *lc_ipl;
1256 unsigned long flags;
1258 lc_ipl = lowcore_ptr[0];
1259 lc = (struct lowcore *) __get_free_pages(GFP_KERNEL | GFP_DMA, LC_ORDER);
1260 nodat_stack = __get_free_pages(GFP_KERNEL, THREAD_SIZE_ORDER);
1261 async_stack = stack_alloc();
1262 mcck_stack = stack_alloc();
1263 if (!lc || !nodat_stack || !async_stack || !mcck_stack)
1264 panic("Couldn't allocate memory");
1266 local_irq_save(flags);
1267 local_mcck_disable();
1268 set_new_lowcore(lc);
1269 S390_lowcore.nodat_stack = nodat_stack + STACK_INIT_OFFSET;
1270 S390_lowcore.async_stack = async_stack + STACK_INIT_OFFSET;
1271 S390_lowcore.mcck_stack = mcck_stack + STACK_INIT_OFFSET;
1272 lowcore_ptr[0] = lc;
1273 local_mcck_enable();
1274 local_irq_restore(flags);
1276 free_pages(lc_ipl->async_stack - STACK_INIT_OFFSET, THREAD_SIZE_ORDER);
1277 memblock_free_late(lc_ipl->mcck_stack - STACK_INIT_OFFSET, THREAD_SIZE);
1278 memblock_free_late((unsigned long) lc_ipl, sizeof(*lc_ipl));
1282 early_initcall(smp_reinit_ipl_cpu);