1 // SPDX-License-Identifier: GPL-2.0
3 * SMP related functions
5 * Copyright IBM Corp. 1999, 2012
6 * Author(s): Denis Joseph Barrow,
7 * Martin Schwidefsky <schwidefsky@de.ibm.com>,
8 * Heiko Carstens <heiko.carstens@de.ibm.com>,
10 * based on other smp stuff by
11 * (c) 1995 Alan Cox, CymruNET Ltd <alan@cymru.net>
12 * (c) 1998 Ingo Molnar
14 * The code outside of smp.c uses logical cpu numbers, only smp.c does
15 * the translation of logical to physical cpu ids. All new code that
16 * operates on physical cpu numbers needs to go into smp.c.
19 #define KMSG_COMPONENT "cpu"
20 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
22 #include <linux/workqueue.h>
23 #include <linux/memblock.h>
24 #include <linux/export.h>
25 #include <linux/init.h>
27 #include <linux/err.h>
28 #include <linux/spinlock.h>
29 #include <linux/kernel_stat.h>
30 #include <linux/delay.h>
31 #include <linux/interrupt.h>
32 #include <linux/irqflags.h>
33 #include <linux/cpu.h>
34 #include <linux/slab.h>
35 #include <linux/sched/hotplug.h>
36 #include <linux/sched/task_stack.h>
37 #include <linux/crash_dump.h>
38 #include <linux/kprobes.h>
39 #include <asm/asm-offsets.h>
41 #include <asm/switch_to.h>
42 #include <asm/facility.h>
44 #include <asm/setup.h>
46 #include <asm/tlbflush.h>
47 #include <asm/vtimer.h>
48 #include <asm/lowcore.h>
51 #include <asm/debug.h>
52 #include <asm/os_info.h>
56 #include <asm/stacktrace.h>
57 #include <asm/topology.h>
62 ec_call_function_single
,
71 static DEFINE_PER_CPU(struct cpu
*, cpu_device
);
74 struct lowcore
*lowcore
; /* lowcore page(s) for the cpu */
75 unsigned long ec_mask
; /* bit mask for ec_xxx functions */
76 unsigned long ec_clk
; /* sigp timestamp for ec_xxx */
77 signed char state
; /* physical cpu state */
78 signed char polarization
; /* physical polarization */
79 u16 address
; /* physical cpu address */
82 static u8 boot_core_type
;
83 static struct pcpu pcpu_devices
[NR_CPUS
];
85 unsigned int smp_cpu_mt_shift
;
86 EXPORT_SYMBOL(smp_cpu_mt_shift
);
88 unsigned int smp_cpu_mtid
;
89 EXPORT_SYMBOL(smp_cpu_mtid
);
91 #ifdef CONFIG_CRASH_DUMP
92 __vector128 __initdata boot_cpu_vector_save_area
[__NUM_VXRS
];
95 static unsigned int smp_max_threads __initdata
= -1U;
97 static int __init
early_nosmt(char *s
)
102 early_param("nosmt", early_nosmt
);
104 static int __init
early_smt(char *s
)
106 get_option(&s
, &smp_max_threads
);
109 early_param("smt", early_smt
);
112 * The smp_cpu_state_mutex must be held when changing the state or polarization
113 * member of a pcpu data structure within the pcpu_devices arreay.
115 DEFINE_MUTEX(smp_cpu_state_mutex
);
118 * Signal processor helper functions.
120 static inline int __pcpu_sigp_relax(u16 addr
, u8 order
, unsigned long parm
)
125 cc
= __pcpu_sigp(addr
, order
, parm
, NULL
);
126 if (cc
!= SIGP_CC_BUSY
)
132 static int pcpu_sigp_retry(struct pcpu
*pcpu
, u8 order
, u32 parm
)
136 for (retry
= 0; ; retry
++) {
137 cc
= __pcpu_sigp(pcpu
->address
, order
, parm
, NULL
);
138 if (cc
!= SIGP_CC_BUSY
)
146 static inline int pcpu_stopped(struct pcpu
*pcpu
)
148 u32
uninitialized_var(status
);
150 if (__pcpu_sigp(pcpu
->address
, SIGP_SENSE
,
151 0, &status
) != SIGP_CC_STATUS_STORED
)
153 return !!(status
& (SIGP_STATUS_CHECK_STOP
|SIGP_STATUS_STOPPED
));
156 static inline int pcpu_running(struct pcpu
*pcpu
)
158 if (__pcpu_sigp(pcpu
->address
, SIGP_SENSE_RUNNING
,
159 0, NULL
) != SIGP_CC_STATUS_STORED
)
161 /* Status stored condition code is equivalent to cpu not running. */
166 * Find struct pcpu by cpu address.
168 static struct pcpu
*pcpu_find_address(const struct cpumask
*mask
, u16 address
)
172 for_each_cpu(cpu
, mask
)
173 if (pcpu_devices
[cpu
].address
== address
)
174 return pcpu_devices
+ cpu
;
178 static void pcpu_ec_call(struct pcpu
*pcpu
, int ec_bit
)
182 if (test_and_set_bit(ec_bit
, &pcpu
->ec_mask
))
184 order
= pcpu_running(pcpu
) ? SIGP_EXTERNAL_CALL
: SIGP_EMERGENCY_SIGNAL
;
185 pcpu
->ec_clk
= get_tod_clock_fast();
186 pcpu_sigp_retry(pcpu
, order
, 0);
189 static int pcpu_alloc_lowcore(struct pcpu
*pcpu
, int cpu
)
191 unsigned long async_stack
, nodat_stack
;
194 if (pcpu
!= &pcpu_devices
[0]) {
195 pcpu
->lowcore
= (struct lowcore
*)
196 __get_free_pages(GFP_KERNEL
| GFP_DMA
, LC_ORDER
);
197 nodat_stack
= __get_free_pages(GFP_KERNEL
, THREAD_SIZE_ORDER
);
198 if (!pcpu
->lowcore
|| !nodat_stack
)
201 nodat_stack
= pcpu
->lowcore
->nodat_stack
- STACK_INIT_OFFSET
;
203 async_stack
= stack_alloc();
207 memcpy(lc
, &S390_lowcore
, 512);
208 memset((char *) lc
+ 512, 0, sizeof(*lc
) - 512);
209 lc
->async_stack
= async_stack
+ STACK_INIT_OFFSET
;
210 lc
->nodat_stack
= nodat_stack
+ STACK_INIT_OFFSET
;
212 lc
->spinlock_lockval
= arch_spin_lockval(cpu
);
213 lc
->spinlock_index
= 0;
214 lc
->br_r1_trampoline
= 0x07f1; /* br %r1 */
215 if (nmi_alloc_per_cpu(lc
))
217 if (vdso_alloc_per_cpu(lc
))
219 lowcore_ptr
[cpu
] = lc
;
220 pcpu_sigp_retry(pcpu
, SIGP_SET_PREFIX
, (u32
)(unsigned long) lc
);
224 nmi_free_per_cpu(lc
);
226 stack_free(async_stack
);
228 if (pcpu
!= &pcpu_devices
[0]) {
229 free_pages(nodat_stack
, THREAD_SIZE_ORDER
);
230 free_pages((unsigned long) pcpu
->lowcore
, LC_ORDER
);
235 #ifdef CONFIG_HOTPLUG_CPU
237 static void pcpu_free_lowcore(struct pcpu
*pcpu
)
239 unsigned long async_stack
, nodat_stack
, lowcore
;
241 nodat_stack
= pcpu
->lowcore
->nodat_stack
- STACK_INIT_OFFSET
;
242 async_stack
= pcpu
->lowcore
->async_stack
- STACK_INIT_OFFSET
;
243 lowcore
= (unsigned long) pcpu
->lowcore
;
245 pcpu_sigp_retry(pcpu
, SIGP_SET_PREFIX
, 0);
246 lowcore_ptr
[pcpu
- pcpu_devices
] = NULL
;
247 vdso_free_per_cpu(pcpu
->lowcore
);
248 nmi_free_per_cpu(pcpu
->lowcore
);
249 stack_free(async_stack
);
250 if (pcpu
== &pcpu_devices
[0])
252 free_pages(nodat_stack
, THREAD_SIZE_ORDER
);
253 free_pages(lowcore
, LC_ORDER
);
256 #endif /* CONFIG_HOTPLUG_CPU */
258 static void pcpu_prepare_secondary(struct pcpu
*pcpu
, int cpu
)
260 struct lowcore
*lc
= pcpu
->lowcore
;
262 cpumask_set_cpu(cpu
, &init_mm
.context
.cpu_attach_mask
);
263 cpumask_set_cpu(cpu
, mm_cpumask(&init_mm
));
265 lc
->spinlock_lockval
= arch_spin_lockval(cpu
);
266 lc
->spinlock_index
= 0;
267 lc
->percpu_offset
= __per_cpu_offset
[cpu
];
268 lc
->kernel_asce
= S390_lowcore
.kernel_asce
;
269 lc
->machine_flags
= S390_lowcore
.machine_flags
;
270 lc
->user_timer
= lc
->system_timer
=
271 lc
->steal_timer
= lc
->avg_steal_timer
= 0;
272 __ctl_store(lc
->cregs_save_area
, 0, 15);
273 save_access_regs((unsigned int *) lc
->access_regs_save_area
);
274 memcpy(lc
->stfle_fac_list
, S390_lowcore
.stfle_fac_list
,
275 sizeof(lc
->stfle_fac_list
));
276 memcpy(lc
->alt_stfle_fac_list
, S390_lowcore
.alt_stfle_fac_list
,
277 sizeof(lc
->alt_stfle_fac_list
));
278 arch_spin_lock_setup(cpu
);
281 static void pcpu_attach_task(struct pcpu
*pcpu
, struct task_struct
*tsk
)
283 struct lowcore
*lc
= pcpu
->lowcore
;
285 lc
->kernel_stack
= (unsigned long) task_stack_page(tsk
)
286 + THREAD_SIZE
- STACK_FRAME_OVERHEAD
- sizeof(struct pt_regs
);
287 lc
->current_task
= (unsigned long) tsk
;
289 lc
->current_pid
= tsk
->pid
;
290 lc
->user_timer
= tsk
->thread
.user_timer
;
291 lc
->guest_timer
= tsk
->thread
.guest_timer
;
292 lc
->system_timer
= tsk
->thread
.system_timer
;
293 lc
->hardirq_timer
= tsk
->thread
.hardirq_timer
;
294 lc
->softirq_timer
= tsk
->thread
.softirq_timer
;
298 static void pcpu_start_fn(struct pcpu
*pcpu
, void (*func
)(void *), void *data
)
300 struct lowcore
*lc
= pcpu
->lowcore
;
302 lc
->restart_stack
= lc
->nodat_stack
;
303 lc
->restart_fn
= (unsigned long) func
;
304 lc
->restart_data
= (unsigned long) data
;
305 lc
->restart_source
= -1UL;
306 pcpu_sigp_retry(pcpu
, SIGP_RESTART
, 0);
310 * Call function via PSW restart on pcpu and stop the current cpu.
312 static void __pcpu_delegate(void (*func
)(void*), void *data
)
314 func(data
); /* should not return */
317 static void __no_sanitize_address
pcpu_delegate(struct pcpu
*pcpu
,
318 void (*func
)(void *),
319 void *data
, unsigned long stack
)
321 struct lowcore
*lc
= lowcore_ptr
[pcpu
- pcpu_devices
];
322 unsigned long source_cpu
= stap();
324 __load_psw_mask(PSW_KERNEL_BITS
| PSW_MASK_DAT
);
325 if (pcpu
->address
== source_cpu
)
326 CALL_ON_STACK(__pcpu_delegate
, stack
, 2, func
, data
);
327 /* Stop target cpu (if func returns this stops the current cpu). */
328 pcpu_sigp_retry(pcpu
, SIGP_STOP
, 0);
329 /* Restart func on the target cpu and stop the current cpu. */
330 mem_assign_absolute(lc
->restart_stack
, stack
);
331 mem_assign_absolute(lc
->restart_fn
, (unsigned long) func
);
332 mem_assign_absolute(lc
->restart_data
, (unsigned long) data
);
333 mem_assign_absolute(lc
->restart_source
, source_cpu
);
336 "0: sigp 0,%0,%2 # sigp restart to target cpu\n"
337 " brc 2,0b # busy, try again\n"
338 "1: sigp 0,%1,%3 # sigp stop to current cpu\n"
339 " brc 2,1b # busy, try again\n"
340 : : "d" (pcpu
->address
), "d" (source_cpu
),
341 "K" (SIGP_RESTART
), "K" (SIGP_STOP
)
347 * Enable additional logical cpus for multi-threading.
349 static int pcpu_set_smt(unsigned int mtid
)
353 if (smp_cpu_mtid
== mtid
)
355 cc
= __pcpu_sigp(0, SIGP_SET_MULTI_THREADING
, mtid
, NULL
);
358 smp_cpu_mt_shift
= 0;
359 while (smp_cpu_mtid
>= (1U << smp_cpu_mt_shift
))
361 pcpu_devices
[0].address
= stap();
367 * Call function on an online CPU.
369 void smp_call_online_cpu(void (*func
)(void *), void *data
)
373 /* Use the current cpu if it is online. */
374 pcpu
= pcpu_find_address(cpu_online_mask
, stap());
376 /* Use the first online cpu. */
377 pcpu
= pcpu_devices
+ cpumask_first(cpu_online_mask
);
378 pcpu_delegate(pcpu
, func
, data
, (unsigned long) restart_stack
);
382 * Call function on the ipl CPU.
384 void smp_call_ipl_cpu(void (*func
)(void *), void *data
)
386 struct lowcore
*lc
= pcpu_devices
->lowcore
;
388 if (pcpu_devices
[0].address
== stap())
391 pcpu_delegate(&pcpu_devices
[0], func
, data
,
395 int smp_find_processor_id(u16 address
)
399 for_each_present_cpu(cpu
)
400 if (pcpu_devices
[cpu
].address
== address
)
405 bool arch_vcpu_is_preempted(int cpu
)
407 if (test_cpu_flag_of(CIF_ENABLED_WAIT
, cpu
))
409 if (pcpu_running(pcpu_devices
+ cpu
))
413 EXPORT_SYMBOL(arch_vcpu_is_preempted
);
415 void smp_yield_cpu(int cpu
)
417 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
));
421 } else if (MACHINE_HAS_DIAG44
) {
422 diag_stat_inc_norecursion(DIAG_STAT_X044
);
423 asm volatile("diag 0,0,0x44");
428 * Send cpus emergency shutdown signal. This gives the cpus the
429 * opportunity to complete outstanding interrupts.
431 void notrace
smp_emergency_stop(void)
437 cpumask_copy(&cpumask
, cpu_online_mask
);
438 cpumask_clear_cpu(smp_processor_id(), &cpumask
);
440 end
= get_tod_clock() + (1000000UL << 12);
441 for_each_cpu(cpu
, &cpumask
) {
442 struct pcpu
*pcpu
= pcpu_devices
+ cpu
;
443 set_bit(ec_stop_cpu
, &pcpu
->ec_mask
);
444 while (__pcpu_sigp(pcpu
->address
, SIGP_EMERGENCY_SIGNAL
,
445 0, NULL
) == SIGP_CC_BUSY
&&
446 get_tod_clock() < end
)
449 while (get_tod_clock() < end
) {
450 for_each_cpu(cpu
, &cpumask
)
451 if (pcpu_stopped(pcpu_devices
+ cpu
))
452 cpumask_clear_cpu(cpu
, &cpumask
);
453 if (cpumask_empty(&cpumask
))
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();
504 static void do_ext_call_interrupt(struct ext_code ext_code
,
505 unsigned int param32
, unsigned long param64
)
507 inc_irq_stat(ext_code
.code
== 0x1202 ? IRQEXT_EXC
: IRQEXT_EMS
);
508 smp_handle_ext_call();
511 void arch_send_call_function_ipi_mask(const struct cpumask
*mask
)
515 for_each_cpu(cpu
, mask
)
516 pcpu_ec_call(pcpu_devices
+ cpu
, ec_call_function_single
);
519 void arch_send_call_function_single_ipi(int cpu
)
521 pcpu_ec_call(pcpu_devices
+ cpu
, ec_call_function_single
);
525 * this function sends a 'reschedule' IPI to another CPU.
526 * it goes straight through and wastes no time serializing
527 * anything. Worst case is that we lose a reschedule ...
529 void smp_send_reschedule(int cpu
)
531 pcpu_ec_call(pcpu_devices
+ cpu
, ec_schedule
);
535 * parameter area for the set/clear control bit callbacks
537 struct ec_creg_mask_parms
{
539 unsigned long andval
;
544 * callback for setting/clearing control bits
546 static void smp_ctl_bit_callback(void *info
)
548 struct ec_creg_mask_parms
*pp
= info
;
549 unsigned long cregs
[16];
551 __ctl_store(cregs
, 0, 15);
552 cregs
[pp
->cr
] = (cregs
[pp
->cr
] & pp
->andval
) | pp
->orval
;
553 __ctl_load(cregs
, 0, 15);
557 * Set a bit in a control register of all cpus
559 void smp_ctl_set_bit(int cr
, int bit
)
561 struct ec_creg_mask_parms parms
= { 1UL << bit
, -1UL, cr
};
563 on_each_cpu(smp_ctl_bit_callback
, &parms
, 1);
565 EXPORT_SYMBOL(smp_ctl_set_bit
);
568 * Clear a bit in a control register of all cpus
570 void smp_ctl_clear_bit(int cr
, int bit
)
572 struct ec_creg_mask_parms parms
= { 0, ~(1UL << bit
), cr
};
574 on_each_cpu(smp_ctl_bit_callback
, &parms
, 1);
576 EXPORT_SYMBOL(smp_ctl_clear_bit
);
578 #ifdef CONFIG_CRASH_DUMP
580 int smp_store_status(int cpu
)
582 struct pcpu
*pcpu
= pcpu_devices
+ cpu
;
585 pa
= __pa(&pcpu
->lowcore
->floating_pt_save_area
);
586 if (__pcpu_sigp_relax(pcpu
->address
, SIGP_STORE_STATUS_AT_ADDRESS
,
587 pa
) != SIGP_CC_ORDER_CODE_ACCEPTED
)
589 if (!MACHINE_HAS_VX
&& !MACHINE_HAS_GS
)
591 pa
= __pa(pcpu
->lowcore
->mcesad
& MCESA_ORIGIN_MASK
);
593 pa
|= pcpu
->lowcore
->mcesad
& MCESA_LC_MASK
;
594 if (__pcpu_sigp_relax(pcpu
->address
, SIGP_STORE_ADDITIONAL_STATUS
,
595 pa
) != SIGP_CC_ORDER_CODE_ACCEPTED
)
601 * Collect CPU state of the previous, crashed system.
602 * There are four cases:
603 * 1) standard zfcp dump
604 * condition: OLDMEM_BASE == NULL && ipl_info.type == IPL_TYPE_FCP_DUMP
605 * The state for all CPUs except the boot CPU needs to be collected
606 * with sigp stop-and-store-status. The boot CPU state is located in
607 * the absolute lowcore of the memory stored in the HSA. The zcore code
608 * will copy the boot CPU state from the HSA.
609 * 2) stand-alone kdump for SCSI (zfcp dump with swapped memory)
610 * condition: OLDMEM_BASE != NULL && ipl_info.type == IPL_TYPE_FCP_DUMP
611 * The state for all CPUs except the boot CPU needs to be collected
612 * with sigp stop-and-store-status. The firmware or the boot-loader
613 * stored the registers of the boot CPU in the absolute lowcore in the
614 * memory of the old system.
615 * 3) kdump and the old kernel did not store the CPU state,
616 * or stand-alone kdump for DASD
617 * condition: OLDMEM_BASE != NULL && !is_kdump_kernel()
618 * The state for all CPUs except the boot CPU needs to be collected
619 * with sigp stop-and-store-status. The kexec code or the boot-loader
620 * stored the registers of the boot CPU in the memory of the old system.
621 * 4) kdump and the old kernel stored the CPU state
622 * condition: OLDMEM_BASE != NULL && is_kdump_kernel()
623 * This case does not exist for s390 anymore, setup_arch explicitly
624 * deactivates the elfcorehdr= kernel parameter
626 static __init
void smp_save_cpu_vxrs(struct save_area
*sa
, u16 addr
,
627 bool is_boot_cpu
, unsigned long page
)
629 __vector128
*vxrs
= (__vector128
*) page
;
632 vxrs
= boot_cpu_vector_save_area
;
634 __pcpu_sigp_relax(addr
, SIGP_STORE_ADDITIONAL_STATUS
, page
);
635 save_area_add_vxrs(sa
, vxrs
);
638 static __init
void smp_save_cpu_regs(struct save_area
*sa
, u16 addr
,
639 bool is_boot_cpu
, unsigned long page
)
641 void *regs
= (void *) page
;
644 copy_oldmem_kernel(regs
, (void *) __LC_FPREGS_SAVE_AREA
, 512);
646 __pcpu_sigp_relax(addr
, SIGP_STORE_STATUS_AT_ADDRESS
, page
);
647 save_area_add_regs(sa
, regs
);
650 void __init
smp_save_dump_cpus(void)
652 int addr
, boot_cpu_addr
, max_cpu_addr
;
653 struct save_area
*sa
;
657 if (!(OLDMEM_BASE
|| ipl_info
.type
== IPL_TYPE_FCP_DUMP
))
658 /* No previous system present, normal boot. */
660 /* Allocate a page as dumping area for the store status sigps */
661 page
= memblock_phys_alloc_range(PAGE_SIZE
, PAGE_SIZE
, 0, 1UL << 31);
663 panic("ERROR: Failed to allocate %lx bytes below %lx\n",
664 PAGE_SIZE
, 1UL << 31);
666 /* Set multi-threading state to the previous system. */
667 pcpu_set_smt(sclp
.mtid_prev
);
668 boot_cpu_addr
= stap();
669 max_cpu_addr
= SCLP_MAX_CORES
<< sclp
.mtid_prev
;
670 for (addr
= 0; addr
<= max_cpu_addr
; addr
++) {
671 if (__pcpu_sigp_relax(addr
, SIGP_SENSE
, 0) ==
672 SIGP_CC_NOT_OPERATIONAL
)
674 is_boot_cpu
= (addr
== boot_cpu_addr
);
675 /* Allocate save area */
676 sa
= save_area_alloc(is_boot_cpu
);
678 panic("could not allocate memory for save area\n");
680 /* Get the vector registers */
681 smp_save_cpu_vxrs(sa
, addr
, is_boot_cpu
, page
);
683 * For a zfcp dump OLDMEM_BASE == NULL and the registers
684 * of the boot CPU are stored in the HSA. To retrieve
685 * these registers an SCLP request is required which is
686 * done by drivers/s390/char/zcore.c:init_cpu_info()
688 if (!is_boot_cpu
|| OLDMEM_BASE
)
689 /* Get the CPU registers */
690 smp_save_cpu_regs(sa
, addr
, is_boot_cpu
, page
);
692 memblock_free(page
, PAGE_SIZE
);
693 diag_dma_ops
.diag308_reset();
696 #endif /* CONFIG_CRASH_DUMP */
698 void smp_cpu_set_polarization(int cpu
, int val
)
700 pcpu_devices
[cpu
].polarization
= val
;
703 int smp_cpu_get_polarization(int cpu
)
705 return pcpu_devices
[cpu
].polarization
;
708 static void __ref
smp_get_core_info(struct sclp_core_info
*info
, int early
)
710 static int use_sigp_detection
;
713 if (use_sigp_detection
|| sclp_get_core_info(info
, early
)) {
714 use_sigp_detection
= 1;
716 address
< (SCLP_MAX_CORES
<< smp_cpu_mt_shift
);
717 address
+= (1U << smp_cpu_mt_shift
)) {
718 if (__pcpu_sigp_relax(address
, SIGP_SENSE
, 0) ==
719 SIGP_CC_NOT_OPERATIONAL
)
721 info
->core
[info
->configured
].core_id
=
722 address
>> smp_cpu_mt_shift
;
725 info
->combined
= info
->configured
;
729 static int smp_add_present_cpu(int cpu
);
731 static int __smp_rescan_cpus(struct sclp_core_info
*info
, int sysfs_add
)
739 cpumask_xor(&avail
, cpu_possible_mask
, cpu_present_mask
);
740 cpu
= cpumask_first(&avail
);
741 for (i
= 0; (i
< info
->combined
) && (cpu
< nr_cpu_ids
); i
++) {
742 if (sclp
.has_core_type
&& info
->core
[i
].type
!= boot_core_type
)
744 address
= info
->core
[i
].core_id
<< smp_cpu_mt_shift
;
745 for (j
= 0; j
<= smp_cpu_mtid
; j
++) {
746 if (pcpu_find_address(cpu_present_mask
, address
+ j
))
748 pcpu
= pcpu_devices
+ cpu
;
749 pcpu
->address
= address
+ j
;
751 (cpu
>= info
->configured
*(smp_cpu_mtid
+ 1)) ?
752 CPU_STATE_STANDBY
: CPU_STATE_CONFIGURED
;
753 smp_cpu_set_polarization(cpu
, POLARIZATION_UNKNOWN
);
754 set_cpu_present(cpu
, true);
755 if (sysfs_add
&& smp_add_present_cpu(cpu
) != 0)
756 set_cpu_present(cpu
, false);
759 cpu
= cpumask_next(cpu
, &avail
);
760 if (cpu
>= nr_cpu_ids
)
767 void __init
smp_detect_cpus(void)
769 unsigned int cpu
, mtid
, c_cpus
, s_cpus
;
770 struct sclp_core_info
*info
;
773 /* Get CPU information */
774 info
= memblock_alloc(sizeof(*info
), 8);
776 panic("%s: Failed to allocate %zu bytes align=0x%x\n",
777 __func__
, sizeof(*info
), 8);
778 smp_get_core_info(info
, 1);
779 /* Find boot CPU type */
780 if (sclp
.has_core_type
) {
782 for (cpu
= 0; cpu
< info
->combined
; cpu
++)
783 if (info
->core
[cpu
].core_id
== address
) {
784 /* The boot cpu dictates the cpu type. */
785 boot_core_type
= info
->core
[cpu
].type
;
788 if (cpu
>= info
->combined
)
789 panic("Could not find boot CPU type");
792 /* Set multi-threading state for the current system */
793 mtid
= boot_core_type
? sclp
.mtid
: sclp
.mtid_cp
;
794 mtid
= (mtid
< smp_max_threads
) ? mtid
: smp_max_threads
- 1;
797 /* Print number of CPUs */
799 for (cpu
= 0; cpu
< info
->combined
; cpu
++) {
800 if (sclp
.has_core_type
&&
801 info
->core
[cpu
].type
!= boot_core_type
)
803 if (cpu
< info
->configured
)
804 c_cpus
+= smp_cpu_mtid
+ 1;
806 s_cpus
+= smp_cpu_mtid
+ 1;
808 pr_info("%d configured CPUs, %d standby CPUs\n", c_cpus
, s_cpus
);
810 /* Add CPUs present at boot */
812 __smp_rescan_cpus(info
, 0);
814 memblock_free_early((unsigned long)info
, sizeof(*info
));
817 static void smp_init_secondary(void)
819 int cpu
= smp_processor_id();
821 S390_lowcore
.last_update_clock
= get_tod_clock();
822 restore_access_regs(S390_lowcore
.access_regs_save_area
);
828 notify_cpu_starting(smp_processor_id());
829 if (topology_cpu_dedicated(cpu
))
830 set_cpu_flag(CIF_DEDICATED_CPU
);
832 clear_cpu_flag(CIF_DEDICATED_CPU
);
833 set_cpu_online(smp_processor_id(), true);
834 inc_irq_stat(CPU_RST
);
836 cpu_startup_entry(CPUHP_AP_ONLINE_IDLE
);
840 * Activate a secondary processor.
842 static void __no_sanitize_address
smp_start_secondary(void *cpuvoid
)
844 S390_lowcore
.restart_stack
= (unsigned long) restart_stack
;
845 S390_lowcore
.restart_fn
= (unsigned long) do_restart
;
846 S390_lowcore
.restart_data
= 0;
847 S390_lowcore
.restart_source
= -1UL;
848 __ctl_load(S390_lowcore
.cregs_save_area
, 0, 15);
849 __load_psw_mask(PSW_KERNEL_BITS
| PSW_MASK_DAT
);
850 CALL_ON_STACK(smp_init_secondary
, S390_lowcore
.kernel_stack
, 0);
853 /* Upping and downing of CPUs */
854 int __cpu_up(unsigned int cpu
, struct task_struct
*tidle
)
859 pcpu
= pcpu_devices
+ cpu
;
860 if (pcpu
->state
!= CPU_STATE_CONFIGURED
)
862 base
= smp_get_base_cpu(cpu
);
863 for (i
= 0; i
<= smp_cpu_mtid
; i
++) {
864 if (base
+ i
< nr_cpu_ids
)
865 if (cpu_online(base
+ i
))
869 * If this is the first CPU of the core to get online
870 * do an initial CPU reset.
872 if (i
> smp_cpu_mtid
&&
873 pcpu_sigp_retry(pcpu_devices
+ base
, SIGP_INITIAL_CPU_RESET
, 0) !=
874 SIGP_CC_ORDER_CODE_ACCEPTED
)
877 rc
= pcpu_alloc_lowcore(pcpu
, cpu
);
880 pcpu_prepare_secondary(pcpu
, cpu
);
881 pcpu_attach_task(pcpu
, tidle
);
882 pcpu_start_fn(pcpu
, smp_start_secondary
, NULL
);
883 /* Wait until cpu puts itself in the online & active maps */
884 while (!cpu_online(cpu
))
889 static unsigned int setup_possible_cpus __initdata
;
891 static int __init
_setup_possible_cpus(char *s
)
893 get_option(&s
, &setup_possible_cpus
);
896 early_param("possible_cpus", _setup_possible_cpus
);
898 #ifdef CONFIG_HOTPLUG_CPU
900 int __cpu_disable(void)
902 unsigned long cregs
[16];
904 /* Handle possible pending IPIs */
905 smp_handle_ext_call();
906 set_cpu_online(smp_processor_id(), false);
907 /* Disable pseudo page faults on this cpu. */
909 /* Disable interrupt sources via control register. */
910 __ctl_store(cregs
, 0, 15);
911 cregs
[0] &= ~0x0000ee70UL
; /* disable all external interrupts */
912 cregs
[6] &= ~0xff000000UL
; /* disable all I/O interrupts */
913 cregs
[14] &= ~0x1f000000UL
; /* disable most machine checks */
914 __ctl_load(cregs
, 0, 15);
915 clear_cpu_flag(CIF_NOHZ_DELAY
);
919 void __cpu_die(unsigned int cpu
)
923 /* Wait until target cpu is down */
924 pcpu
= pcpu_devices
+ cpu
;
925 while (!pcpu_stopped(pcpu
))
927 pcpu_free_lowcore(pcpu
);
928 cpumask_clear_cpu(cpu
, mm_cpumask(&init_mm
));
929 cpumask_clear_cpu(cpu
, &init_mm
.context
.cpu_attach_mask
);
932 void __noreturn
cpu_die(void)
936 pcpu_sigp_retry(pcpu_devices
+ smp_processor_id(), SIGP_STOP
, 0);
940 #endif /* CONFIG_HOTPLUG_CPU */
942 void __init
smp_fill_possible_mask(void)
944 unsigned int possible
, sclp_max
, cpu
;
946 sclp_max
= max(sclp
.mtid
, sclp
.mtid_cp
) + 1;
947 sclp_max
= min(smp_max_threads
, sclp_max
);
948 sclp_max
= (sclp
.max_cores
* sclp_max
) ?: nr_cpu_ids
;
949 possible
= setup_possible_cpus
?: nr_cpu_ids
;
950 possible
= min(possible
, sclp_max
);
951 for (cpu
= 0; cpu
< possible
&& cpu
< nr_cpu_ids
; cpu
++)
952 set_cpu_possible(cpu
, true);
955 void __init
smp_prepare_cpus(unsigned int max_cpus
)
957 /* request the 0x1201 emergency signal external interrupt */
958 if (register_external_irq(EXT_IRQ_EMERGENCY_SIG
, do_ext_call_interrupt
))
959 panic("Couldn't request external interrupt 0x1201");
960 /* request the 0x1202 external call external interrupt */
961 if (register_external_irq(EXT_IRQ_EXTERNAL_CALL
, do_ext_call_interrupt
))
962 panic("Couldn't request external interrupt 0x1202");
965 void __init
smp_prepare_boot_cpu(void)
967 struct pcpu
*pcpu
= pcpu_devices
;
969 WARN_ON(!cpu_present(0) || !cpu_online(0));
970 pcpu
->state
= CPU_STATE_CONFIGURED
;
971 pcpu
->lowcore
= (struct lowcore
*)(unsigned long) store_prefix();
972 S390_lowcore
.percpu_offset
= __per_cpu_offset
[0];
973 smp_cpu_set_polarization(0, POLARIZATION_UNKNOWN
);
976 void __init
smp_cpus_done(unsigned int max_cpus
)
980 void __init
smp_setup_processor_id(void)
982 pcpu_devices
[0].address
= stap();
983 S390_lowcore
.cpu_nr
= 0;
984 S390_lowcore
.spinlock_lockval
= arch_spin_lockval(0);
985 S390_lowcore
.spinlock_index
= 0;
989 * the frequency of the profiling timer can be changed
990 * by writing a multiplier value into /proc/profile.
992 * usually you want to run this on all CPUs ;)
994 int setup_profiling_timer(unsigned int multiplier
)
999 #ifdef CONFIG_HOTPLUG_CPU
1000 static ssize_t
cpu_configure_show(struct device
*dev
,
1001 struct device_attribute
*attr
, char *buf
)
1005 mutex_lock(&smp_cpu_state_mutex
);
1006 count
= sprintf(buf
, "%d\n", pcpu_devices
[dev
->id
].state
);
1007 mutex_unlock(&smp_cpu_state_mutex
);
1011 static ssize_t
cpu_configure_store(struct device
*dev
,
1012 struct device_attribute
*attr
,
1013 const char *buf
, size_t count
)
1016 int cpu
, val
, rc
, i
;
1019 if (sscanf(buf
, "%d %c", &val
, &delim
) != 1)
1021 if (val
!= 0 && val
!= 1)
1024 mutex_lock(&smp_cpu_state_mutex
);
1026 /* disallow configuration changes of online cpus and cpu 0 */
1028 cpu
= smp_get_base_cpu(cpu
);
1031 for (i
= 0; i
<= smp_cpu_mtid
; i
++)
1032 if (cpu_online(cpu
+ i
))
1034 pcpu
= pcpu_devices
+ cpu
;
1038 if (pcpu
->state
!= CPU_STATE_CONFIGURED
)
1040 rc
= sclp_core_deconfigure(pcpu
->address
>> smp_cpu_mt_shift
);
1043 for (i
= 0; i
<= smp_cpu_mtid
; i
++) {
1044 if (cpu
+ i
>= nr_cpu_ids
|| !cpu_present(cpu
+ i
))
1046 pcpu
[i
].state
= CPU_STATE_STANDBY
;
1047 smp_cpu_set_polarization(cpu
+ i
,
1048 POLARIZATION_UNKNOWN
);
1050 topology_expect_change();
1053 if (pcpu
->state
!= CPU_STATE_STANDBY
)
1055 rc
= sclp_core_configure(pcpu
->address
>> smp_cpu_mt_shift
);
1058 for (i
= 0; i
<= smp_cpu_mtid
; i
++) {
1059 if (cpu
+ i
>= nr_cpu_ids
|| !cpu_present(cpu
+ i
))
1061 pcpu
[i
].state
= CPU_STATE_CONFIGURED
;
1062 smp_cpu_set_polarization(cpu
+ i
,
1063 POLARIZATION_UNKNOWN
);
1065 topology_expect_change();
1071 mutex_unlock(&smp_cpu_state_mutex
);
1073 return rc
? rc
: count
;
1075 static DEVICE_ATTR(configure
, 0644, cpu_configure_show
, cpu_configure_store
);
1076 #endif /* CONFIG_HOTPLUG_CPU */
1078 static ssize_t
show_cpu_address(struct device
*dev
,
1079 struct device_attribute
*attr
, char *buf
)
1081 return sprintf(buf
, "%d\n", pcpu_devices
[dev
->id
].address
);
1083 static DEVICE_ATTR(address
, 0444, show_cpu_address
, NULL
);
1085 static struct attribute
*cpu_common_attrs
[] = {
1086 #ifdef CONFIG_HOTPLUG_CPU
1087 &dev_attr_configure
.attr
,
1089 &dev_attr_address
.attr
,
1093 static struct attribute_group cpu_common_attr_group
= {
1094 .attrs
= cpu_common_attrs
,
1097 static struct attribute
*cpu_online_attrs
[] = {
1098 &dev_attr_idle_count
.attr
,
1099 &dev_attr_idle_time_us
.attr
,
1103 static struct attribute_group cpu_online_attr_group
= {
1104 .attrs
= cpu_online_attrs
,
1107 static int smp_cpu_online(unsigned int cpu
)
1109 struct device
*s
= &per_cpu(cpu_device
, cpu
)->dev
;
1111 return sysfs_create_group(&s
->kobj
, &cpu_online_attr_group
);
1113 static int smp_cpu_pre_down(unsigned int cpu
)
1115 struct device
*s
= &per_cpu(cpu_device
, cpu
)->dev
;
1117 sysfs_remove_group(&s
->kobj
, &cpu_online_attr_group
);
1121 static int smp_add_present_cpu(int cpu
)
1127 c
= kzalloc(sizeof(*c
), GFP_KERNEL
);
1130 per_cpu(cpu_device
, cpu
) = c
;
1132 c
->hotpluggable
= 1;
1133 rc
= register_cpu(c
, cpu
);
1136 rc
= sysfs_create_group(&s
->kobj
, &cpu_common_attr_group
);
1139 rc
= topology_cpu_init(c
);
1145 sysfs_remove_group(&s
->kobj
, &cpu_common_attr_group
);
1147 #ifdef CONFIG_HOTPLUG_CPU
1154 #ifdef CONFIG_HOTPLUG_CPU
1156 int __ref
smp_rescan_cpus(void)
1158 struct sclp_core_info
*info
;
1161 info
= kzalloc(sizeof(*info
), GFP_KERNEL
);
1164 smp_get_core_info(info
, 0);
1166 mutex_lock(&smp_cpu_state_mutex
);
1167 nr
= __smp_rescan_cpus(info
, 1);
1168 mutex_unlock(&smp_cpu_state_mutex
);
1172 topology_schedule_update();
1176 static ssize_t __ref
rescan_store(struct device
*dev
,
1177 struct device_attribute
*attr
,
1183 rc
= lock_device_hotplug_sysfs();
1186 rc
= smp_rescan_cpus();
1187 unlock_device_hotplug();
1188 return rc
? rc
: count
;
1190 static DEVICE_ATTR_WO(rescan
);
1191 #endif /* CONFIG_HOTPLUG_CPU */
1193 static int __init
s390_smp_init(void)
1197 #ifdef CONFIG_HOTPLUG_CPU
1198 rc
= device_create_file(cpu_subsys
.dev_root
, &dev_attr_rescan
);
1202 for_each_present_cpu(cpu
) {
1203 rc
= smp_add_present_cpu(cpu
);
1208 rc
= cpuhp_setup_state(CPUHP_AP_ONLINE_DYN
, "s390/smp:online",
1209 smp_cpu_online
, smp_cpu_pre_down
);
1210 rc
= rc
<= 0 ? rc
: 0;
1214 subsys_initcall(s390_smp_init
);