Loading...
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * SMP related functions
4 *
5 * Copyright IBM Corp. 1999, 2012
6 * Author(s): Denis Joseph Barrow,
7 * Martin Schwidefsky <schwidefsky@de.ibm.com>,
8 *
9 * based on other smp stuff by
10 * (c) 1995 Alan Cox, CymruNET Ltd <alan@cymru.net>
11 * (c) 1998 Ingo Molnar
12 *
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.
16 */
17
18#define KMSG_COMPONENT "cpu"
19#define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
20
21#include <linux/workqueue.h>
22#include <linux/memblock.h>
23#include <linux/export.h>
24#include <linux/init.h>
25#include <linux/mm.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/irq_work.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>
40#include <asm/diag.h>
41#include <asm/switch_to.h>
42#include <asm/facility.h>
43#include <asm/ipl.h>
44#include <asm/setup.h>
45#include <asm/irq.h>
46#include <asm/tlbflush.h>
47#include <asm/vtimer.h>
48#include <asm/abs_lowcore.h>
49#include <asm/sclp.h>
50#include <asm/debug.h>
51#include <asm/os_info.h>
52#include <asm/sigp.h>
53#include <asm/idle.h>
54#include <asm/nmi.h>
55#include <asm/stacktrace.h>
56#include <asm/topology.h>
57#include <asm/vdso.h>
58#include <asm/maccess.h>
59#include "entry.h"
60
61enum {
62 ec_schedule = 0,
63 ec_call_function_single,
64 ec_stop_cpu,
65 ec_mcck_pending,
66 ec_irq_work,
67};
68
69enum {
70 CPU_STATE_STANDBY,
71 CPU_STATE_CONFIGURED,
72};
73
74static DEFINE_PER_CPU(struct cpu *, cpu_device);
75
76struct pcpu {
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 */
82};
83
84static u8 boot_core_type;
85static struct pcpu pcpu_devices[NR_CPUS];
86
87unsigned int smp_cpu_mt_shift;
88EXPORT_SYMBOL(smp_cpu_mt_shift);
89
90unsigned int smp_cpu_mtid;
91EXPORT_SYMBOL(smp_cpu_mtid);
92
93#ifdef CONFIG_CRASH_DUMP
94__vector128 __initdata boot_cpu_vector_save_area[__NUM_VXRS];
95#endif
96
97static unsigned int smp_max_threads __initdata = -1U;
98cpumask_t cpu_setup_mask;
99
100static int __init early_nosmt(char *s)
101{
102 smp_max_threads = 1;
103 return 0;
104}
105early_param("nosmt", early_nosmt);
106
107static int __init early_smt(char *s)
108{
109 get_option(&s, &smp_max_threads);
110 return 0;
111}
112early_param("smt", early_smt);
113
114/*
115 * The smp_cpu_state_mutex must be held when changing the state or polarization
116 * member of a pcpu data structure within the pcpu_devices arreay.
117 */
118DEFINE_MUTEX(smp_cpu_state_mutex);
119
120/*
121 * Signal processor helper functions.
122 */
123static inline int __pcpu_sigp_relax(u16 addr, u8 order, unsigned long parm)
124{
125 int cc;
126
127 while (1) {
128 cc = __pcpu_sigp(addr, order, parm, NULL);
129 if (cc != SIGP_CC_BUSY)
130 return cc;
131 cpu_relax();
132 }
133}
134
135static int pcpu_sigp_retry(struct pcpu *pcpu, u8 order, u32 parm)
136{
137 int cc, retry;
138
139 for (retry = 0; ; retry++) {
140 cc = __pcpu_sigp(pcpu->address, order, parm, NULL);
141 if (cc != SIGP_CC_BUSY)
142 break;
143 if (retry >= 3)
144 udelay(10);
145 }
146 return cc;
147}
148
149static inline int pcpu_stopped(struct pcpu *pcpu)
150{
151 u32 status;
152
153 if (__pcpu_sigp(pcpu->address, SIGP_SENSE,
154 0, &status) != SIGP_CC_STATUS_STORED)
155 return 0;
156 return !!(status & (SIGP_STATUS_CHECK_STOP|SIGP_STATUS_STOPPED));
157}
158
159static inline int pcpu_running(struct pcpu *pcpu)
160{
161 if (__pcpu_sigp(pcpu->address, SIGP_SENSE_RUNNING,
162 0, NULL) != SIGP_CC_STATUS_STORED)
163 return 1;
164 /* Status stored condition code is equivalent to cpu not running. */
165 return 0;
166}
167
168/*
169 * Find struct pcpu by cpu address.
170 */
171static struct pcpu *pcpu_find_address(const struct cpumask *mask, u16 address)
172{
173 int cpu;
174
175 for_each_cpu(cpu, mask)
176 if (pcpu_devices[cpu].address == address)
177 return pcpu_devices + cpu;
178 return NULL;
179}
180
181static void pcpu_ec_call(struct pcpu *pcpu, int ec_bit)
182{
183 int order;
184
185 if (test_and_set_bit(ec_bit, &pcpu->ec_mask))
186 return;
187 order = pcpu_running(pcpu) ? SIGP_EXTERNAL_CALL : SIGP_EMERGENCY_SIGNAL;
188 pcpu->ec_clk = get_tod_clock_fast();
189 pcpu_sigp_retry(pcpu, order, 0);
190}
191
192static int pcpu_alloc_lowcore(struct pcpu *pcpu, int cpu)
193{
194 unsigned long async_stack, nodat_stack, mcck_stack;
195 struct lowcore *lc;
196
197 lc = (struct lowcore *) __get_free_pages(GFP_KERNEL | GFP_DMA, LC_ORDER);
198 nodat_stack = __get_free_pages(GFP_KERNEL, THREAD_SIZE_ORDER);
199 async_stack = stack_alloc();
200 mcck_stack = stack_alloc();
201 if (!lc || !nodat_stack || !async_stack || !mcck_stack)
202 goto out;
203 memcpy(lc, &S390_lowcore, 512);
204 memset((char *) lc + 512, 0, sizeof(*lc) - 512);
205 lc->async_stack = async_stack + STACK_INIT_OFFSET;
206 lc->nodat_stack = nodat_stack + STACK_INIT_OFFSET;
207 lc->mcck_stack = mcck_stack + STACK_INIT_OFFSET;
208 lc->cpu_nr = cpu;
209 lc->spinlock_lockval = arch_spin_lockval(cpu);
210 lc->spinlock_index = 0;
211 lc->return_lpswe = gen_lpswe(__LC_RETURN_PSW);
212 lc->return_mcck_lpswe = gen_lpswe(__LC_RETURN_MCCK_PSW);
213 lc->preempt_count = PREEMPT_DISABLED;
214 if (nmi_alloc_mcesa(&lc->mcesad))
215 goto out;
216 if (abs_lowcore_map(cpu, lc, true))
217 goto out_mcesa;
218 lowcore_ptr[cpu] = lc;
219 pcpu_sigp_retry(pcpu, SIGP_SET_PREFIX, __pa(lc));
220 return 0;
221
222out_mcesa:
223 nmi_free_mcesa(&lc->mcesad);
224out:
225 stack_free(mcck_stack);
226 stack_free(async_stack);
227 free_pages(nodat_stack, THREAD_SIZE_ORDER);
228 free_pages((unsigned long) lc, LC_ORDER);
229 return -ENOMEM;
230}
231
232static void pcpu_free_lowcore(struct pcpu *pcpu)
233{
234 unsigned long async_stack, nodat_stack, mcck_stack;
235 struct lowcore *lc;
236 int cpu;
237
238 cpu = pcpu - pcpu_devices;
239 lc = lowcore_ptr[cpu];
240 nodat_stack = lc->nodat_stack - STACK_INIT_OFFSET;
241 async_stack = lc->async_stack - STACK_INIT_OFFSET;
242 mcck_stack = lc->mcck_stack - STACK_INIT_OFFSET;
243 pcpu_sigp_retry(pcpu, SIGP_SET_PREFIX, 0);
244 lowcore_ptr[cpu] = NULL;
245 abs_lowcore_unmap(cpu);
246 nmi_free_mcesa(&lc->mcesad);
247 stack_free(async_stack);
248 stack_free(mcck_stack);
249 free_pages(nodat_stack, THREAD_SIZE_ORDER);
250 free_pages((unsigned long) lc, LC_ORDER);
251}
252
253static void pcpu_prepare_secondary(struct pcpu *pcpu, int cpu)
254{
255 struct lowcore *lc = lowcore_ptr[cpu];
256
257 cpumask_set_cpu(cpu, &init_mm.context.cpu_attach_mask);
258 cpumask_set_cpu(cpu, mm_cpumask(&init_mm));
259 lc->cpu_nr = cpu;
260 lc->restart_flags = RESTART_FLAG_CTLREGS;
261 lc->spinlock_lockval = arch_spin_lockval(cpu);
262 lc->spinlock_index = 0;
263 lc->percpu_offset = __per_cpu_offset[cpu];
264 lc->kernel_asce = S390_lowcore.kernel_asce;
265 lc->user_asce = s390_invalid_asce;
266 lc->machine_flags = S390_lowcore.machine_flags;
267 lc->user_timer = lc->system_timer =
268 lc->steal_timer = lc->avg_steal_timer = 0;
269 __ctl_store(lc->cregs_save_area, 0, 15);
270 lc->cregs_save_area[1] = lc->kernel_asce;
271 lc->cregs_save_area[7] = lc->user_asce;
272 save_access_regs((unsigned int *) lc->access_regs_save_area);
273 arch_spin_lock_setup(cpu);
274}
275
276static void pcpu_attach_task(struct pcpu *pcpu, struct task_struct *tsk)
277{
278 struct lowcore *lc;
279 int cpu;
280
281 cpu = pcpu - pcpu_devices;
282 lc = lowcore_ptr[cpu];
283 lc->kernel_stack = (unsigned long) task_stack_page(tsk)
284 + THREAD_SIZE - STACK_FRAME_OVERHEAD - sizeof(struct pt_regs);
285 lc->current_task = (unsigned long) tsk;
286 lc->lpp = LPP_MAGIC;
287 lc->current_pid = tsk->pid;
288 lc->user_timer = tsk->thread.user_timer;
289 lc->guest_timer = tsk->thread.guest_timer;
290 lc->system_timer = tsk->thread.system_timer;
291 lc->hardirq_timer = tsk->thread.hardirq_timer;
292 lc->softirq_timer = tsk->thread.softirq_timer;
293 lc->steal_timer = 0;
294}
295
296static void pcpu_start_fn(struct pcpu *pcpu, void (*func)(void *), void *data)
297{
298 struct lowcore *lc;
299 int cpu;
300
301 cpu = pcpu - pcpu_devices;
302 lc = lowcore_ptr[cpu];
303 lc->restart_stack = lc->kernel_stack;
304 lc->restart_fn = (unsigned long) func;
305 lc->restart_data = (unsigned long) data;
306 lc->restart_source = -1U;
307 pcpu_sigp_retry(pcpu, SIGP_RESTART, 0);
308}
309
310typedef void (pcpu_delegate_fn)(void *);
311
312/*
313 * Call function via PSW restart on pcpu and stop the current cpu.
314 */
315static void __pcpu_delegate(pcpu_delegate_fn *func, void *data)
316{
317 func(data); /* should not return */
318}
319
320static void pcpu_delegate(struct pcpu *pcpu,
321 pcpu_delegate_fn *func,
322 void *data, unsigned long stack)
323{
324 struct lowcore *lc, *abs_lc;
325 unsigned int source_cpu;
326 unsigned long flags;
327
328 lc = lowcore_ptr[pcpu - pcpu_devices];
329 source_cpu = stap();
330 __load_psw_mask(PSW_KERNEL_BITS | PSW_MASK_DAT);
331 if (pcpu->address == source_cpu) {
332 call_on_stack(2, stack, void, __pcpu_delegate,
333 pcpu_delegate_fn *, func, void *, data);
334 }
335 /* Stop target cpu (if func returns this stops the current cpu). */
336 pcpu_sigp_retry(pcpu, SIGP_STOP, 0);
337 /* Restart func on the target cpu and stop the current cpu. */
338 if (lc) {
339 lc->restart_stack = stack;
340 lc->restart_fn = (unsigned long)func;
341 lc->restart_data = (unsigned long)data;
342 lc->restart_source = source_cpu;
343 } else {
344 abs_lc = get_abs_lowcore(&flags);
345 abs_lc->restart_stack = stack;
346 abs_lc->restart_fn = (unsigned long)func;
347 abs_lc->restart_data = (unsigned long)data;
348 abs_lc->restart_source = source_cpu;
349 put_abs_lowcore(abs_lc, flags);
350 }
351 __bpon();
352 asm volatile(
353 "0: sigp 0,%0,%2 # sigp restart to target cpu\n"
354 " brc 2,0b # busy, try again\n"
355 "1: sigp 0,%1,%3 # sigp stop to current cpu\n"
356 " brc 2,1b # busy, try again\n"
357 : : "d" (pcpu->address), "d" (source_cpu),
358 "K" (SIGP_RESTART), "K" (SIGP_STOP)
359 : "0", "1", "cc");
360 for (;;) ;
361}
362
363/*
364 * Enable additional logical cpus for multi-threading.
365 */
366static int pcpu_set_smt(unsigned int mtid)
367{
368 int cc;
369
370 if (smp_cpu_mtid == mtid)
371 return 0;
372 cc = __pcpu_sigp(0, SIGP_SET_MULTI_THREADING, mtid, NULL);
373 if (cc == 0) {
374 smp_cpu_mtid = mtid;
375 smp_cpu_mt_shift = 0;
376 while (smp_cpu_mtid >= (1U << smp_cpu_mt_shift))
377 smp_cpu_mt_shift++;
378 pcpu_devices[0].address = stap();
379 }
380 return cc;
381}
382
383/*
384 * Call function on an online CPU.
385 */
386void smp_call_online_cpu(void (*func)(void *), void *data)
387{
388 struct pcpu *pcpu;
389
390 /* Use the current cpu if it is online. */
391 pcpu = pcpu_find_address(cpu_online_mask, stap());
392 if (!pcpu)
393 /* Use the first online cpu. */
394 pcpu = pcpu_devices + cpumask_first(cpu_online_mask);
395 pcpu_delegate(pcpu, func, data, (unsigned long) restart_stack);
396}
397
398/*
399 * Call function on the ipl CPU.
400 */
401void smp_call_ipl_cpu(void (*func)(void *), void *data)
402{
403 struct lowcore *lc = lowcore_ptr[0];
404
405 if (pcpu_devices[0].address == stap())
406 lc = &S390_lowcore;
407
408 pcpu_delegate(&pcpu_devices[0], func, data,
409 lc->nodat_stack);
410}
411
412int smp_find_processor_id(u16 address)
413{
414 int cpu;
415
416 for_each_present_cpu(cpu)
417 if (pcpu_devices[cpu].address == address)
418 return cpu;
419 return -1;
420}
421
422void schedule_mcck_handler(void)
423{
424 pcpu_ec_call(pcpu_devices + smp_processor_id(), ec_mcck_pending);
425}
426
427bool notrace arch_vcpu_is_preempted(int cpu)
428{
429 if (test_cpu_flag_of(CIF_ENABLED_WAIT, cpu))
430 return false;
431 if (pcpu_running(pcpu_devices + cpu))
432 return false;
433 return true;
434}
435EXPORT_SYMBOL(arch_vcpu_is_preempted);
436
437void notrace smp_yield_cpu(int cpu)
438{
439 if (!MACHINE_HAS_DIAG9C)
440 return;
441 diag_stat_inc_norecursion(DIAG_STAT_X09C);
442 asm volatile("diag %0,0,0x9c"
443 : : "d" (pcpu_devices[cpu].address));
444}
445EXPORT_SYMBOL_GPL(smp_yield_cpu);
446
447/*
448 * Send cpus emergency shutdown signal. This gives the cpus the
449 * opportunity to complete outstanding interrupts.
450 */
451void notrace smp_emergency_stop(void)
452{
453 static arch_spinlock_t lock = __ARCH_SPIN_LOCK_UNLOCKED;
454 static cpumask_t cpumask;
455 u64 end;
456 int cpu;
457
458 arch_spin_lock(&lock);
459 cpumask_copy(&cpumask, cpu_online_mask);
460 cpumask_clear_cpu(smp_processor_id(), &cpumask);
461
462 end = get_tod_clock() + (1000000UL << 12);
463 for_each_cpu(cpu, &cpumask) {
464 struct pcpu *pcpu = pcpu_devices + cpu;
465 set_bit(ec_stop_cpu, &pcpu->ec_mask);
466 while (__pcpu_sigp(pcpu->address, SIGP_EMERGENCY_SIGNAL,
467 0, NULL) == SIGP_CC_BUSY &&
468 get_tod_clock() < end)
469 cpu_relax();
470 }
471 while (get_tod_clock() < end) {
472 for_each_cpu(cpu, &cpumask)
473 if (pcpu_stopped(pcpu_devices + cpu))
474 cpumask_clear_cpu(cpu, &cpumask);
475 if (cpumask_empty(&cpumask))
476 break;
477 cpu_relax();
478 }
479 arch_spin_unlock(&lock);
480}
481NOKPROBE_SYMBOL(smp_emergency_stop);
482
483/*
484 * Stop all cpus but the current one.
485 */
486void smp_send_stop(void)
487{
488 int cpu;
489
490 /* Disable all interrupts/machine checks */
491 __load_psw_mask(PSW_KERNEL_BITS | PSW_MASK_DAT);
492 trace_hardirqs_off();
493
494 debug_set_critical();
495
496 if (oops_in_progress)
497 smp_emergency_stop();
498
499 /* stop all processors */
500 for_each_online_cpu(cpu) {
501 if (cpu == smp_processor_id())
502 continue;
503 pcpu_sigp_retry(pcpu_devices + cpu, SIGP_STOP, 0);
504 while (!pcpu_stopped(pcpu_devices + cpu))
505 cpu_relax();
506 }
507}
508
509/*
510 * This is the main routine where commands issued by other
511 * cpus are handled.
512 */
513static void smp_handle_ext_call(void)
514{
515 unsigned long bits;
516
517 /* handle bit signal external calls */
518 bits = xchg(&pcpu_devices[smp_processor_id()].ec_mask, 0);
519 if (test_bit(ec_stop_cpu, &bits))
520 smp_stop_cpu();
521 if (test_bit(ec_schedule, &bits))
522 scheduler_ipi();
523 if (test_bit(ec_call_function_single, &bits))
524 generic_smp_call_function_single_interrupt();
525 if (test_bit(ec_mcck_pending, &bits))
526 __s390_handle_mcck();
527 if (test_bit(ec_irq_work, &bits))
528 irq_work_run();
529}
530
531static void do_ext_call_interrupt(struct ext_code ext_code,
532 unsigned int param32, unsigned long param64)
533{
534 inc_irq_stat(ext_code.code == 0x1202 ? IRQEXT_EXC : IRQEXT_EMS);
535 smp_handle_ext_call();
536}
537
538void arch_send_call_function_ipi_mask(const struct cpumask *mask)
539{
540 int cpu;
541
542 for_each_cpu(cpu, mask)
543 pcpu_ec_call(pcpu_devices + cpu, ec_call_function_single);
544}
545
546void arch_send_call_function_single_ipi(int cpu)
547{
548 pcpu_ec_call(pcpu_devices + cpu, ec_call_function_single);
549}
550
551/*
552 * this function sends a 'reschedule' IPI to another CPU.
553 * it goes straight through and wastes no time serializing
554 * anything. Worst case is that we lose a reschedule ...
555 */
556void smp_send_reschedule(int cpu)
557{
558 pcpu_ec_call(pcpu_devices + cpu, ec_schedule);
559}
560
561#ifdef CONFIG_IRQ_WORK
562void arch_irq_work_raise(void)
563{
564 pcpu_ec_call(pcpu_devices + smp_processor_id(), ec_irq_work);
565}
566#endif
567
568/*
569 * parameter area for the set/clear control bit callbacks
570 */
571struct ec_creg_mask_parms {
572 unsigned long orval;
573 unsigned long andval;
574 int cr;
575};
576
577/*
578 * callback for setting/clearing control bits
579 */
580static void smp_ctl_bit_callback(void *info)
581{
582 struct ec_creg_mask_parms *pp = info;
583 unsigned long cregs[16];
584
585 __ctl_store(cregs, 0, 15);
586 cregs[pp->cr] = (cregs[pp->cr] & pp->andval) | pp->orval;
587 __ctl_load(cregs, 0, 15);
588}
589
590static DEFINE_SPINLOCK(ctl_lock);
591
592void smp_ctl_set_clear_bit(int cr, int bit, bool set)
593{
594 struct ec_creg_mask_parms parms = { .cr = cr, };
595 struct lowcore *abs_lc;
596 unsigned long flags;
597 u64 ctlreg;
598
599 if (set) {
600 parms.orval = 1UL << bit;
601 parms.andval = -1UL;
602 } else {
603 parms.orval = 0;
604 parms.andval = ~(1UL << bit);
605 }
606 spin_lock(&ctl_lock);
607 abs_lc = get_abs_lowcore(&flags);
608 ctlreg = abs_lc->cregs_save_area[cr];
609 ctlreg = (ctlreg & parms.andval) | parms.orval;
610 abs_lc->cregs_save_area[cr] = ctlreg;
611 put_abs_lowcore(abs_lc, flags);
612 spin_unlock(&ctl_lock);
613 on_each_cpu(smp_ctl_bit_callback, &parms, 1);
614}
615EXPORT_SYMBOL(smp_ctl_set_clear_bit);
616
617#ifdef CONFIG_CRASH_DUMP
618
619int smp_store_status(int cpu)
620{
621 struct lowcore *lc;
622 struct pcpu *pcpu;
623 unsigned long pa;
624
625 pcpu = pcpu_devices + cpu;
626 lc = lowcore_ptr[cpu];
627 pa = __pa(&lc->floating_pt_save_area);
628 if (__pcpu_sigp_relax(pcpu->address, SIGP_STORE_STATUS_AT_ADDRESS,
629 pa) != SIGP_CC_ORDER_CODE_ACCEPTED)
630 return -EIO;
631 if (!MACHINE_HAS_VX && !MACHINE_HAS_GS)
632 return 0;
633 pa = lc->mcesad & MCESA_ORIGIN_MASK;
634 if (MACHINE_HAS_GS)
635 pa |= lc->mcesad & MCESA_LC_MASK;
636 if (__pcpu_sigp_relax(pcpu->address, SIGP_STORE_ADDITIONAL_STATUS,
637 pa) != SIGP_CC_ORDER_CODE_ACCEPTED)
638 return -EIO;
639 return 0;
640}
641
642/*
643 * Collect CPU state of the previous, crashed system.
644 * There are four cases:
645 * 1) standard zfcp/nvme dump
646 * condition: OLDMEM_BASE == NULL && is_ipl_type_dump() == true
647 * The state for all CPUs except the boot CPU needs to be collected
648 * with sigp stop-and-store-status. The boot CPU state is located in
649 * the absolute lowcore of the memory stored in the HSA. The zcore code
650 * will copy the boot CPU state from the HSA.
651 * 2) stand-alone kdump for SCSI/NVMe (zfcp/nvme dump with swapped memory)
652 * condition: OLDMEM_BASE != NULL && is_ipl_type_dump() == true
653 * The state for all CPUs except the boot CPU needs to be collected
654 * with sigp stop-and-store-status. The firmware or the boot-loader
655 * stored the registers of the boot CPU in the absolute lowcore in the
656 * memory of the old system.
657 * 3) kdump and the old kernel did not store the CPU state,
658 * or stand-alone kdump for DASD
659 * condition: OLDMEM_BASE != NULL && !is_kdump_kernel()
660 * The state for all CPUs except the boot CPU needs to be collected
661 * with sigp stop-and-store-status. The kexec code or the boot-loader
662 * stored the registers of the boot CPU in the memory of the old system.
663 * 4) kdump and the old kernel stored the CPU state
664 * condition: OLDMEM_BASE != NULL && is_kdump_kernel()
665 * This case does not exist for s390 anymore, setup_arch explicitly
666 * deactivates the elfcorehdr= kernel parameter
667 */
668static bool dump_available(void)
669{
670 return oldmem_data.start || is_ipl_type_dump();
671}
672
673void __init smp_save_dump_ipl_cpu(void)
674{
675 struct save_area *sa;
676 void *regs;
677
678 if (!dump_available())
679 return;
680 sa = save_area_alloc(true);
681 regs = memblock_alloc(512, 8);
682 if (!sa || !regs)
683 panic("could not allocate memory for boot CPU save area\n");
684 copy_oldmem_kernel(regs, __LC_FPREGS_SAVE_AREA, 512);
685 save_area_add_regs(sa, regs);
686 memblock_free(regs, 512);
687 if (MACHINE_HAS_VX)
688 save_area_add_vxrs(sa, boot_cpu_vector_save_area);
689}
690
691void __init smp_save_dump_secondary_cpus(void)
692{
693 int addr, boot_cpu_addr, max_cpu_addr;
694 struct save_area *sa;
695 void *page;
696
697 if (!dump_available())
698 return;
699 /* Allocate a page as dumping area for the store status sigps */
700 page = memblock_alloc_low(PAGE_SIZE, PAGE_SIZE);
701 if (!page)
702 panic("ERROR: Failed to allocate %lx bytes below %lx\n",
703 PAGE_SIZE, 1UL << 31);
704
705 /* Set multi-threading state to the previous system. */
706 pcpu_set_smt(sclp.mtid_prev);
707 boot_cpu_addr = stap();
708 max_cpu_addr = SCLP_MAX_CORES << sclp.mtid_prev;
709 for (addr = 0; addr <= max_cpu_addr; addr++) {
710 if (addr == boot_cpu_addr)
711 continue;
712 if (__pcpu_sigp_relax(addr, SIGP_SENSE, 0) ==
713 SIGP_CC_NOT_OPERATIONAL)
714 continue;
715 sa = save_area_alloc(false);
716 if (!sa)
717 panic("could not allocate memory for save area\n");
718 __pcpu_sigp_relax(addr, SIGP_STORE_STATUS_AT_ADDRESS, __pa(page));
719 save_area_add_regs(sa, page);
720 if (MACHINE_HAS_VX) {
721 __pcpu_sigp_relax(addr, SIGP_STORE_ADDITIONAL_STATUS, __pa(page));
722 save_area_add_vxrs(sa, page);
723 }
724 }
725 memblock_free(page, PAGE_SIZE);
726 diag_amode31_ops.diag308_reset();
727 pcpu_set_smt(0);
728}
729#endif /* CONFIG_CRASH_DUMP */
730
731void smp_cpu_set_polarization(int cpu, int val)
732{
733 pcpu_devices[cpu].polarization = val;
734}
735
736int smp_cpu_get_polarization(int cpu)
737{
738 return pcpu_devices[cpu].polarization;
739}
740
741int smp_cpu_get_cpu_address(int cpu)
742{
743 return pcpu_devices[cpu].address;
744}
745
746static void __ref smp_get_core_info(struct sclp_core_info *info, int early)
747{
748 static int use_sigp_detection;
749 int address;
750
751 if (use_sigp_detection || sclp_get_core_info(info, early)) {
752 use_sigp_detection = 1;
753 for (address = 0;
754 address < (SCLP_MAX_CORES << smp_cpu_mt_shift);
755 address += (1U << smp_cpu_mt_shift)) {
756 if (__pcpu_sigp_relax(address, SIGP_SENSE, 0) ==
757 SIGP_CC_NOT_OPERATIONAL)
758 continue;
759 info->core[info->configured].core_id =
760 address >> smp_cpu_mt_shift;
761 info->configured++;
762 }
763 info->combined = info->configured;
764 }
765}
766
767static int smp_add_present_cpu(int cpu);
768
769static int smp_add_core(struct sclp_core_entry *core, cpumask_t *avail,
770 bool configured, bool early)
771{
772 struct pcpu *pcpu;
773 int cpu, nr, i;
774 u16 address;
775
776 nr = 0;
777 if (sclp.has_core_type && core->type != boot_core_type)
778 return nr;
779 cpu = cpumask_first(avail);
780 address = core->core_id << smp_cpu_mt_shift;
781 for (i = 0; (i <= smp_cpu_mtid) && (cpu < nr_cpu_ids); i++) {
782 if (pcpu_find_address(cpu_present_mask, address + i))
783 continue;
784 pcpu = pcpu_devices + cpu;
785 pcpu->address = address + i;
786 if (configured)
787 pcpu->state = CPU_STATE_CONFIGURED;
788 else
789 pcpu->state = CPU_STATE_STANDBY;
790 smp_cpu_set_polarization(cpu, POLARIZATION_UNKNOWN);
791 set_cpu_present(cpu, true);
792 if (!early && smp_add_present_cpu(cpu) != 0)
793 set_cpu_present(cpu, false);
794 else
795 nr++;
796 cpumask_clear_cpu(cpu, avail);
797 cpu = cpumask_next(cpu, avail);
798 }
799 return nr;
800}
801
802static int __smp_rescan_cpus(struct sclp_core_info *info, bool early)
803{
804 struct sclp_core_entry *core;
805 static cpumask_t avail;
806 bool configured;
807 u16 core_id;
808 int nr, i;
809
810 cpus_read_lock();
811 mutex_lock(&smp_cpu_state_mutex);
812 nr = 0;
813 cpumask_xor(&avail, cpu_possible_mask, cpu_present_mask);
814 /*
815 * Add IPL core first (which got logical CPU number 0) to make sure
816 * that all SMT threads get subsequent logical CPU numbers.
817 */
818 if (early) {
819 core_id = pcpu_devices[0].address >> smp_cpu_mt_shift;
820 for (i = 0; i < info->configured; i++) {
821 core = &info->core[i];
822 if (core->core_id == core_id) {
823 nr += smp_add_core(core, &avail, true, early);
824 break;
825 }
826 }
827 }
828 for (i = 0; i < info->combined; i++) {
829 configured = i < info->configured;
830 nr += smp_add_core(&info->core[i], &avail, configured, early);
831 }
832 mutex_unlock(&smp_cpu_state_mutex);
833 cpus_read_unlock();
834 return nr;
835}
836
837void __init smp_detect_cpus(void)
838{
839 unsigned int cpu, mtid, c_cpus, s_cpus;
840 struct sclp_core_info *info;
841 u16 address;
842
843 /* Get CPU information */
844 info = memblock_alloc(sizeof(*info), 8);
845 if (!info)
846 panic("%s: Failed to allocate %zu bytes align=0x%x\n",
847 __func__, sizeof(*info), 8);
848 smp_get_core_info(info, 1);
849 /* Find boot CPU type */
850 if (sclp.has_core_type) {
851 address = stap();
852 for (cpu = 0; cpu < info->combined; cpu++)
853 if (info->core[cpu].core_id == address) {
854 /* The boot cpu dictates the cpu type. */
855 boot_core_type = info->core[cpu].type;
856 break;
857 }
858 if (cpu >= info->combined)
859 panic("Could not find boot CPU type");
860 }
861
862 /* Set multi-threading state for the current system */
863 mtid = boot_core_type ? sclp.mtid : sclp.mtid_cp;
864 mtid = (mtid < smp_max_threads) ? mtid : smp_max_threads - 1;
865 pcpu_set_smt(mtid);
866
867 /* Print number of CPUs */
868 c_cpus = s_cpus = 0;
869 for (cpu = 0; cpu < info->combined; cpu++) {
870 if (sclp.has_core_type &&
871 info->core[cpu].type != boot_core_type)
872 continue;
873 if (cpu < info->configured)
874 c_cpus += smp_cpu_mtid + 1;
875 else
876 s_cpus += smp_cpu_mtid + 1;
877 }
878 pr_info("%d configured CPUs, %d standby CPUs\n", c_cpus, s_cpus);
879
880 /* Add CPUs present at boot */
881 __smp_rescan_cpus(info, true);
882 memblock_free(info, sizeof(*info));
883}
884
885/*
886 * Activate a secondary processor.
887 */
888static void smp_start_secondary(void *cpuvoid)
889{
890 int cpu = raw_smp_processor_id();
891
892 S390_lowcore.last_update_clock = get_tod_clock();
893 S390_lowcore.restart_stack = (unsigned long)restart_stack;
894 S390_lowcore.restart_fn = (unsigned long)do_restart;
895 S390_lowcore.restart_data = 0;
896 S390_lowcore.restart_source = -1U;
897 S390_lowcore.restart_flags = 0;
898 restore_access_regs(S390_lowcore.access_regs_save_area);
899 cpu_init();
900 rcu_cpu_starting(cpu);
901 init_cpu_timer();
902 vtime_init();
903 vdso_getcpu_init();
904 pfault_init();
905 cpumask_set_cpu(cpu, &cpu_setup_mask);
906 update_cpu_masks();
907 notify_cpu_starting(cpu);
908 if (topology_cpu_dedicated(cpu))
909 set_cpu_flag(CIF_DEDICATED_CPU);
910 else
911 clear_cpu_flag(CIF_DEDICATED_CPU);
912 set_cpu_online(cpu, true);
913 inc_irq_stat(CPU_RST);
914 local_irq_enable();
915 cpu_startup_entry(CPUHP_AP_ONLINE_IDLE);
916}
917
918/* Upping and downing of CPUs */
919int __cpu_up(unsigned int cpu, struct task_struct *tidle)
920{
921 struct pcpu *pcpu = pcpu_devices + cpu;
922 int rc;
923
924 if (pcpu->state != CPU_STATE_CONFIGURED)
925 return -EIO;
926 if (pcpu_sigp_retry(pcpu, SIGP_INITIAL_CPU_RESET, 0) !=
927 SIGP_CC_ORDER_CODE_ACCEPTED)
928 return -EIO;
929
930 rc = pcpu_alloc_lowcore(pcpu, cpu);
931 if (rc)
932 return rc;
933 pcpu_prepare_secondary(pcpu, cpu);
934 pcpu_attach_task(pcpu, tidle);
935 pcpu_start_fn(pcpu, smp_start_secondary, NULL);
936 /* Wait until cpu puts itself in the online & active maps */
937 while (!cpu_online(cpu))
938 cpu_relax();
939 return 0;
940}
941
942static unsigned int setup_possible_cpus __initdata;
943
944static int __init _setup_possible_cpus(char *s)
945{
946 get_option(&s, &setup_possible_cpus);
947 return 0;
948}
949early_param("possible_cpus", _setup_possible_cpus);
950
951int __cpu_disable(void)
952{
953 unsigned long cregs[16];
954 int cpu;
955
956 /* Handle possible pending IPIs */
957 smp_handle_ext_call();
958 cpu = smp_processor_id();
959 set_cpu_online(cpu, false);
960 cpumask_clear_cpu(cpu, &cpu_setup_mask);
961 update_cpu_masks();
962 /* Disable pseudo page faults on this cpu. */
963 pfault_fini();
964 /* Disable interrupt sources via control register. */
965 __ctl_store(cregs, 0, 15);
966 cregs[0] &= ~0x0000ee70UL; /* disable all external interrupts */
967 cregs[6] &= ~0xff000000UL; /* disable all I/O interrupts */
968 cregs[14] &= ~0x1f000000UL; /* disable most machine checks */
969 __ctl_load(cregs, 0, 15);
970 clear_cpu_flag(CIF_NOHZ_DELAY);
971 return 0;
972}
973
974void __cpu_die(unsigned int cpu)
975{
976 struct pcpu *pcpu;
977
978 /* Wait until target cpu is down */
979 pcpu = pcpu_devices + cpu;
980 while (!pcpu_stopped(pcpu))
981 cpu_relax();
982 pcpu_free_lowcore(pcpu);
983 cpumask_clear_cpu(cpu, mm_cpumask(&init_mm));
984 cpumask_clear_cpu(cpu, &init_mm.context.cpu_attach_mask);
985}
986
987void __noreturn cpu_die(void)
988{
989 idle_task_exit();
990 __bpon();
991 pcpu_sigp_retry(pcpu_devices + smp_processor_id(), SIGP_STOP, 0);
992 for (;;) ;
993}
994
995void __init smp_fill_possible_mask(void)
996{
997 unsigned int possible, sclp_max, cpu;
998
999 sclp_max = max(sclp.mtid, sclp.mtid_cp) + 1;
1000 sclp_max = min(smp_max_threads, sclp_max);
1001 sclp_max = (sclp.max_cores * sclp_max) ?: nr_cpu_ids;
1002 possible = setup_possible_cpus ?: nr_cpu_ids;
1003 possible = min(possible, sclp_max);
1004 for (cpu = 0; cpu < possible && cpu < nr_cpu_ids; cpu++)
1005 set_cpu_possible(cpu, true);
1006}
1007
1008void __init smp_prepare_cpus(unsigned int max_cpus)
1009{
1010 /* request the 0x1201 emergency signal external interrupt */
1011 if (register_external_irq(EXT_IRQ_EMERGENCY_SIG, do_ext_call_interrupt))
1012 panic("Couldn't request external interrupt 0x1201");
1013 /* request the 0x1202 external call external interrupt */
1014 if (register_external_irq(EXT_IRQ_EXTERNAL_CALL, do_ext_call_interrupt))
1015 panic("Couldn't request external interrupt 0x1202");
1016}
1017
1018void __init smp_prepare_boot_cpu(void)
1019{
1020 struct pcpu *pcpu = pcpu_devices;
1021
1022 WARN_ON(!cpu_present(0) || !cpu_online(0));
1023 pcpu->state = CPU_STATE_CONFIGURED;
1024 S390_lowcore.percpu_offset = __per_cpu_offset[0];
1025 smp_cpu_set_polarization(0, POLARIZATION_UNKNOWN);
1026}
1027
1028void __init smp_setup_processor_id(void)
1029{
1030 pcpu_devices[0].address = stap();
1031 S390_lowcore.cpu_nr = 0;
1032 S390_lowcore.spinlock_lockval = arch_spin_lockval(0);
1033 S390_lowcore.spinlock_index = 0;
1034}
1035
1036/*
1037 * the frequency of the profiling timer can be changed
1038 * by writing a multiplier value into /proc/profile.
1039 *
1040 * usually you want to run this on all CPUs ;)
1041 */
1042int setup_profiling_timer(unsigned int multiplier)
1043{
1044 return 0;
1045}
1046
1047static ssize_t cpu_configure_show(struct device *dev,
1048 struct device_attribute *attr, char *buf)
1049{
1050 ssize_t count;
1051
1052 mutex_lock(&smp_cpu_state_mutex);
1053 count = sprintf(buf, "%d\n", pcpu_devices[dev->id].state);
1054 mutex_unlock(&smp_cpu_state_mutex);
1055 return count;
1056}
1057
1058static ssize_t cpu_configure_store(struct device *dev,
1059 struct device_attribute *attr,
1060 const char *buf, size_t count)
1061{
1062 struct pcpu *pcpu;
1063 int cpu, val, rc, i;
1064 char delim;
1065
1066 if (sscanf(buf, "%d %c", &val, &delim) != 1)
1067 return -EINVAL;
1068 if (val != 0 && val != 1)
1069 return -EINVAL;
1070 cpus_read_lock();
1071 mutex_lock(&smp_cpu_state_mutex);
1072 rc = -EBUSY;
1073 /* disallow configuration changes of online cpus and cpu 0 */
1074 cpu = dev->id;
1075 cpu = smp_get_base_cpu(cpu);
1076 if (cpu == 0)
1077 goto out;
1078 for (i = 0; i <= smp_cpu_mtid; i++)
1079 if (cpu_online(cpu + i))
1080 goto out;
1081 pcpu = pcpu_devices + cpu;
1082 rc = 0;
1083 switch (val) {
1084 case 0:
1085 if (pcpu->state != CPU_STATE_CONFIGURED)
1086 break;
1087 rc = sclp_core_deconfigure(pcpu->address >> smp_cpu_mt_shift);
1088 if (rc)
1089 break;
1090 for (i = 0; i <= smp_cpu_mtid; i++) {
1091 if (cpu + i >= nr_cpu_ids || !cpu_present(cpu + i))
1092 continue;
1093 pcpu[i].state = CPU_STATE_STANDBY;
1094 smp_cpu_set_polarization(cpu + i,
1095 POLARIZATION_UNKNOWN);
1096 }
1097 topology_expect_change();
1098 break;
1099 case 1:
1100 if (pcpu->state != CPU_STATE_STANDBY)
1101 break;
1102 rc = sclp_core_configure(pcpu->address >> smp_cpu_mt_shift);
1103 if (rc)
1104 break;
1105 for (i = 0; i <= smp_cpu_mtid; i++) {
1106 if (cpu + i >= nr_cpu_ids || !cpu_present(cpu + i))
1107 continue;
1108 pcpu[i].state = CPU_STATE_CONFIGURED;
1109 smp_cpu_set_polarization(cpu + i,
1110 POLARIZATION_UNKNOWN);
1111 }
1112 topology_expect_change();
1113 break;
1114 default:
1115 break;
1116 }
1117out:
1118 mutex_unlock(&smp_cpu_state_mutex);
1119 cpus_read_unlock();
1120 return rc ? rc : count;
1121}
1122static DEVICE_ATTR(configure, 0644, cpu_configure_show, cpu_configure_store);
1123
1124static ssize_t show_cpu_address(struct device *dev,
1125 struct device_attribute *attr, char *buf)
1126{
1127 return sprintf(buf, "%d\n", pcpu_devices[dev->id].address);
1128}
1129static DEVICE_ATTR(address, 0444, show_cpu_address, NULL);
1130
1131static struct attribute *cpu_common_attrs[] = {
1132 &dev_attr_configure.attr,
1133 &dev_attr_address.attr,
1134 NULL,
1135};
1136
1137static struct attribute_group cpu_common_attr_group = {
1138 .attrs = cpu_common_attrs,
1139};
1140
1141static struct attribute *cpu_online_attrs[] = {
1142 &dev_attr_idle_count.attr,
1143 &dev_attr_idle_time_us.attr,
1144 NULL,
1145};
1146
1147static struct attribute_group cpu_online_attr_group = {
1148 .attrs = cpu_online_attrs,
1149};
1150
1151static int smp_cpu_online(unsigned int cpu)
1152{
1153 struct device *s = &per_cpu(cpu_device, cpu)->dev;
1154
1155 return sysfs_create_group(&s->kobj, &cpu_online_attr_group);
1156}
1157
1158static int smp_cpu_pre_down(unsigned int cpu)
1159{
1160 struct device *s = &per_cpu(cpu_device, cpu)->dev;
1161
1162 sysfs_remove_group(&s->kobj, &cpu_online_attr_group);
1163 return 0;
1164}
1165
1166static int smp_add_present_cpu(int cpu)
1167{
1168 struct device *s;
1169 struct cpu *c;
1170 int rc;
1171
1172 c = kzalloc(sizeof(*c), GFP_KERNEL);
1173 if (!c)
1174 return -ENOMEM;
1175 per_cpu(cpu_device, cpu) = c;
1176 s = &c->dev;
1177 c->hotpluggable = 1;
1178 rc = register_cpu(c, cpu);
1179 if (rc)
1180 goto out;
1181 rc = sysfs_create_group(&s->kobj, &cpu_common_attr_group);
1182 if (rc)
1183 goto out_cpu;
1184 rc = topology_cpu_init(c);
1185 if (rc)
1186 goto out_topology;
1187 return 0;
1188
1189out_topology:
1190 sysfs_remove_group(&s->kobj, &cpu_common_attr_group);
1191out_cpu:
1192 unregister_cpu(c);
1193out:
1194 return rc;
1195}
1196
1197int __ref smp_rescan_cpus(void)
1198{
1199 struct sclp_core_info *info;
1200 int nr;
1201
1202 info = kzalloc(sizeof(*info), GFP_KERNEL);
1203 if (!info)
1204 return -ENOMEM;
1205 smp_get_core_info(info, 0);
1206 nr = __smp_rescan_cpus(info, false);
1207 kfree(info);
1208 if (nr)
1209 topology_schedule_update();
1210 return 0;
1211}
1212
1213static ssize_t __ref rescan_store(struct device *dev,
1214 struct device_attribute *attr,
1215 const char *buf,
1216 size_t count)
1217{
1218 int rc;
1219
1220 rc = lock_device_hotplug_sysfs();
1221 if (rc)
1222 return rc;
1223 rc = smp_rescan_cpus();
1224 unlock_device_hotplug();
1225 return rc ? rc : count;
1226}
1227static DEVICE_ATTR_WO(rescan);
1228
1229static int __init s390_smp_init(void)
1230{
1231 int cpu, rc = 0;
1232
1233 rc = device_create_file(cpu_subsys.dev_root, &dev_attr_rescan);
1234 if (rc)
1235 return rc;
1236 for_each_present_cpu(cpu) {
1237 rc = smp_add_present_cpu(cpu);
1238 if (rc)
1239 goto out;
1240 }
1241
1242 rc = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "s390/smp:online",
1243 smp_cpu_online, smp_cpu_pre_down);
1244 rc = rc <= 0 ? rc : 0;
1245out:
1246 return rc;
1247}
1248subsys_initcall(s390_smp_init);
1249
1250static __always_inline void set_new_lowcore(struct lowcore *lc)
1251{
1252 union register_pair dst, src;
1253 u32 pfx;
1254
1255 src.even = (unsigned long) &S390_lowcore;
1256 src.odd = sizeof(S390_lowcore);
1257 dst.even = (unsigned long) lc;
1258 dst.odd = sizeof(*lc);
1259 pfx = __pa(lc);
1260
1261 asm volatile(
1262 " mvcl %[dst],%[src]\n"
1263 " spx %[pfx]\n"
1264 : [dst] "+&d" (dst.pair), [src] "+&d" (src.pair)
1265 : [pfx] "Q" (pfx)
1266 : "memory", "cc");
1267}
1268
1269int __init smp_reinit_ipl_cpu(void)
1270{
1271 unsigned long async_stack, nodat_stack, mcck_stack;
1272 struct lowcore *lc, *lc_ipl;
1273 unsigned long flags, cr0;
1274 u64 mcesad;
1275
1276 lc_ipl = lowcore_ptr[0];
1277 lc = (struct lowcore *) __get_free_pages(GFP_KERNEL | GFP_DMA, LC_ORDER);
1278 nodat_stack = __get_free_pages(GFP_KERNEL, THREAD_SIZE_ORDER);
1279 async_stack = stack_alloc();
1280 mcck_stack = stack_alloc();
1281 if (!lc || !nodat_stack || !async_stack || !mcck_stack || nmi_alloc_mcesa(&mcesad))
1282 panic("Couldn't allocate memory");
1283
1284 local_irq_save(flags);
1285 local_mcck_disable();
1286 set_new_lowcore(lc);
1287 S390_lowcore.nodat_stack = nodat_stack + STACK_INIT_OFFSET;
1288 S390_lowcore.async_stack = async_stack + STACK_INIT_OFFSET;
1289 S390_lowcore.mcck_stack = mcck_stack + STACK_INIT_OFFSET;
1290 __ctl_store(cr0, 0, 0);
1291 __ctl_clear_bit(0, 28); /* disable lowcore protection */
1292 S390_lowcore.mcesad = mcesad;
1293 __ctl_load(cr0, 0, 0);
1294 if (abs_lowcore_map(0, lc, false))
1295 panic("Couldn't remap absolute lowcore");
1296 lowcore_ptr[0] = lc;
1297 local_mcck_enable();
1298 local_irq_restore(flags);
1299
1300 free_pages(lc_ipl->async_stack - STACK_INIT_OFFSET, THREAD_SIZE_ORDER);
1301 memblock_free_late(__pa(lc_ipl->mcck_stack - STACK_INIT_OFFSET), THREAD_SIZE);
1302 memblock_free_late(__pa(lc_ipl), sizeof(*lc_ipl));
1303
1304 return 0;
1305}
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * SMP related functions
4 *
5 * Copyright IBM Corp. 1999, 2012
6 * Author(s): Denis Joseph Barrow,
7 * Martin Schwidefsky <schwidefsky@de.ibm.com>,
8 *
9 * based on other smp stuff by
10 * (c) 1995 Alan Cox, CymruNET Ltd <alan@cymru.net>
11 * (c) 1998 Ingo Molnar
12 *
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.
16 */
17
18#define KMSG_COMPONENT "cpu"
19#define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
20
21#include <linux/workqueue.h>
22#include <linux/memblock.h>
23#include <linux/export.h>
24#include <linux/init.h>
25#include <linux/mm.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/irq_work.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>
40#include <asm/ctlreg.h>
41#include <asm/pfault.h>
42#include <asm/diag.h>
43#include <asm/switch_to.h>
44#include <asm/facility.h>
45#include <asm/ipl.h>
46#include <asm/setup.h>
47#include <asm/irq.h>
48#include <asm/tlbflush.h>
49#include <asm/vtimer.h>
50#include <asm/abs_lowcore.h>
51#include <asm/sclp.h>
52#include <asm/debug.h>
53#include <asm/os_info.h>
54#include <asm/sigp.h>
55#include <asm/idle.h>
56#include <asm/nmi.h>
57#include <asm/stacktrace.h>
58#include <asm/topology.h>
59#include <asm/vdso.h>
60#include <asm/maccess.h>
61#include "entry.h"
62
63enum {
64 ec_schedule = 0,
65 ec_call_function_single,
66 ec_stop_cpu,
67 ec_mcck_pending,
68 ec_irq_work,
69};
70
71enum {
72 CPU_STATE_STANDBY,
73 CPU_STATE_CONFIGURED,
74};
75
76static DEFINE_PER_CPU(struct cpu *, cpu_device);
77
78struct pcpu {
79 unsigned long ec_mask; /* bit mask for ec_xxx functions */
80 unsigned long ec_clk; /* sigp timestamp for ec_xxx */
81 signed char state; /* physical cpu state */
82 signed char polarization; /* physical polarization */
83 u16 address; /* physical cpu address */
84};
85
86static u8 boot_core_type;
87static struct pcpu pcpu_devices[NR_CPUS];
88
89unsigned int smp_cpu_mt_shift;
90EXPORT_SYMBOL(smp_cpu_mt_shift);
91
92unsigned int smp_cpu_mtid;
93EXPORT_SYMBOL(smp_cpu_mtid);
94
95#ifdef CONFIG_CRASH_DUMP
96__vector128 __initdata boot_cpu_vector_save_area[__NUM_VXRS];
97#endif
98
99static unsigned int smp_max_threads __initdata = -1U;
100cpumask_t cpu_setup_mask;
101
102static int __init early_nosmt(char *s)
103{
104 smp_max_threads = 1;
105 return 0;
106}
107early_param("nosmt", early_nosmt);
108
109static int __init early_smt(char *s)
110{
111 get_option(&s, &smp_max_threads);
112 return 0;
113}
114early_param("smt", early_smt);
115
116/*
117 * The smp_cpu_state_mutex must be held when changing the state or polarization
118 * member of a pcpu data structure within the pcpu_devices array.
119 */
120DEFINE_MUTEX(smp_cpu_state_mutex);
121
122/*
123 * Signal processor helper functions.
124 */
125static inline int __pcpu_sigp_relax(u16 addr, u8 order, unsigned long parm)
126{
127 int cc;
128
129 while (1) {
130 cc = __pcpu_sigp(addr, order, parm, NULL);
131 if (cc != SIGP_CC_BUSY)
132 return cc;
133 cpu_relax();
134 }
135}
136
137static int pcpu_sigp_retry(struct pcpu *pcpu, u8 order, u32 parm)
138{
139 int cc, retry;
140
141 for (retry = 0; ; retry++) {
142 cc = __pcpu_sigp(pcpu->address, order, parm, NULL);
143 if (cc != SIGP_CC_BUSY)
144 break;
145 if (retry >= 3)
146 udelay(10);
147 }
148 return cc;
149}
150
151static inline int pcpu_stopped(struct pcpu *pcpu)
152{
153 u32 status;
154
155 if (__pcpu_sigp(pcpu->address, SIGP_SENSE,
156 0, &status) != SIGP_CC_STATUS_STORED)
157 return 0;
158 return !!(status & (SIGP_STATUS_CHECK_STOP|SIGP_STATUS_STOPPED));
159}
160
161static inline int pcpu_running(struct pcpu *pcpu)
162{
163 if (__pcpu_sigp(pcpu->address, SIGP_SENSE_RUNNING,
164 0, NULL) != SIGP_CC_STATUS_STORED)
165 return 1;
166 /* Status stored condition code is equivalent to cpu not running. */
167 return 0;
168}
169
170/*
171 * Find struct pcpu by cpu address.
172 */
173static struct pcpu *pcpu_find_address(const struct cpumask *mask, u16 address)
174{
175 int cpu;
176
177 for_each_cpu(cpu, mask)
178 if (pcpu_devices[cpu].address == address)
179 return pcpu_devices + cpu;
180 return NULL;
181}
182
183static void pcpu_ec_call(struct pcpu *pcpu, int ec_bit)
184{
185 int order;
186
187 if (test_and_set_bit(ec_bit, &pcpu->ec_mask))
188 return;
189 order = pcpu_running(pcpu) ? SIGP_EXTERNAL_CALL : SIGP_EMERGENCY_SIGNAL;
190 pcpu->ec_clk = get_tod_clock_fast();
191 pcpu_sigp_retry(pcpu, order, 0);
192}
193
194static int pcpu_alloc_lowcore(struct pcpu *pcpu, int cpu)
195{
196 unsigned long async_stack, nodat_stack, mcck_stack;
197 struct lowcore *lc;
198
199 lc = (struct lowcore *) __get_free_pages(GFP_KERNEL | GFP_DMA, LC_ORDER);
200 nodat_stack = __get_free_pages(GFP_KERNEL, THREAD_SIZE_ORDER);
201 async_stack = stack_alloc();
202 mcck_stack = stack_alloc();
203 if (!lc || !nodat_stack || !async_stack || !mcck_stack)
204 goto out;
205 memcpy(lc, &S390_lowcore, 512);
206 memset((char *) lc + 512, 0, sizeof(*lc) - 512);
207 lc->async_stack = async_stack + STACK_INIT_OFFSET;
208 lc->nodat_stack = nodat_stack + STACK_INIT_OFFSET;
209 lc->mcck_stack = mcck_stack + STACK_INIT_OFFSET;
210 lc->cpu_nr = cpu;
211 lc->spinlock_lockval = arch_spin_lockval(cpu);
212 lc->spinlock_index = 0;
213 lc->return_lpswe = gen_lpswe(__LC_RETURN_PSW);
214 lc->return_mcck_lpswe = gen_lpswe(__LC_RETURN_MCCK_PSW);
215 lc->preempt_count = PREEMPT_DISABLED;
216 if (nmi_alloc_mcesa(&lc->mcesad))
217 goto out;
218 if (abs_lowcore_map(cpu, lc, true))
219 goto out_mcesa;
220 lowcore_ptr[cpu] = lc;
221 pcpu_sigp_retry(pcpu, SIGP_SET_PREFIX, __pa(lc));
222 return 0;
223
224out_mcesa:
225 nmi_free_mcesa(&lc->mcesad);
226out:
227 stack_free(mcck_stack);
228 stack_free(async_stack);
229 free_pages(nodat_stack, THREAD_SIZE_ORDER);
230 free_pages((unsigned long) lc, LC_ORDER);
231 return -ENOMEM;
232}
233
234static void pcpu_free_lowcore(struct pcpu *pcpu)
235{
236 unsigned long async_stack, nodat_stack, mcck_stack;
237 struct lowcore *lc;
238 int cpu;
239
240 cpu = pcpu - pcpu_devices;
241 lc = lowcore_ptr[cpu];
242 nodat_stack = lc->nodat_stack - STACK_INIT_OFFSET;
243 async_stack = lc->async_stack - STACK_INIT_OFFSET;
244 mcck_stack = lc->mcck_stack - STACK_INIT_OFFSET;
245 pcpu_sigp_retry(pcpu, SIGP_SET_PREFIX, 0);
246 lowcore_ptr[cpu] = NULL;
247 abs_lowcore_unmap(cpu);
248 nmi_free_mcesa(&lc->mcesad);
249 stack_free(async_stack);
250 stack_free(mcck_stack);
251 free_pages(nodat_stack, THREAD_SIZE_ORDER);
252 free_pages((unsigned long) lc, LC_ORDER);
253}
254
255static void pcpu_prepare_secondary(struct pcpu *pcpu, int cpu)
256{
257 struct lowcore *lc, *abs_lc;
258
259 lc = lowcore_ptr[cpu];
260 cpumask_set_cpu(cpu, &init_mm.context.cpu_attach_mask);
261 cpumask_set_cpu(cpu, mm_cpumask(&init_mm));
262 lc->cpu_nr = cpu;
263 lc->restart_flags = RESTART_FLAG_CTLREGS;
264 lc->spinlock_lockval = arch_spin_lockval(cpu);
265 lc->spinlock_index = 0;
266 lc->percpu_offset = __per_cpu_offset[cpu];
267 lc->kernel_asce = S390_lowcore.kernel_asce;
268 lc->user_asce = s390_invalid_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 abs_lc = get_abs_lowcore();
273 memcpy(lc->cregs_save_area, abs_lc->cregs_save_area, sizeof(lc->cregs_save_area));
274 put_abs_lowcore(abs_lc);
275 lc->cregs_save_area[1] = lc->kernel_asce;
276 lc->cregs_save_area[7] = lc->user_asce;
277 save_access_regs((unsigned int *) lc->access_regs_save_area);
278 arch_spin_lock_setup(cpu);
279}
280
281static void pcpu_attach_task(struct pcpu *pcpu, struct task_struct *tsk)
282{
283 struct lowcore *lc;
284 int cpu;
285
286 cpu = pcpu - pcpu_devices;
287 lc = lowcore_ptr[cpu];
288 lc->kernel_stack = (unsigned long)task_stack_page(tsk) + STACK_INIT_OFFSET;
289 lc->current_task = (unsigned long)tsk;
290 lc->lpp = LPP_MAGIC;
291 lc->current_pid = tsk->pid;
292 lc->user_timer = tsk->thread.user_timer;
293 lc->guest_timer = tsk->thread.guest_timer;
294 lc->system_timer = tsk->thread.system_timer;
295 lc->hardirq_timer = tsk->thread.hardirq_timer;
296 lc->softirq_timer = tsk->thread.softirq_timer;
297 lc->steal_timer = 0;
298}
299
300static void pcpu_start_fn(struct pcpu *pcpu, void (*func)(void *), void *data)
301{
302 struct lowcore *lc;
303 int cpu;
304
305 cpu = pcpu - pcpu_devices;
306 lc = lowcore_ptr[cpu];
307 lc->restart_stack = lc->kernel_stack;
308 lc->restart_fn = (unsigned long) func;
309 lc->restart_data = (unsigned long) data;
310 lc->restart_source = -1U;
311 pcpu_sigp_retry(pcpu, SIGP_RESTART, 0);
312}
313
314typedef void (pcpu_delegate_fn)(void *);
315
316/*
317 * Call function via PSW restart on pcpu and stop the current cpu.
318 */
319static void __pcpu_delegate(pcpu_delegate_fn *func, void *data)
320{
321 func(data); /* should not return */
322}
323
324static void pcpu_delegate(struct pcpu *pcpu,
325 pcpu_delegate_fn *func,
326 void *data, unsigned long stack)
327{
328 struct lowcore *lc, *abs_lc;
329 unsigned int source_cpu;
330
331 lc = lowcore_ptr[pcpu - pcpu_devices];
332 source_cpu = stap();
333
334 if (pcpu->address == source_cpu) {
335 call_on_stack(2, stack, void, __pcpu_delegate,
336 pcpu_delegate_fn *, func, void *, data);
337 }
338 /* Stop target cpu (if func returns this stops the current cpu). */
339 pcpu_sigp_retry(pcpu, SIGP_STOP, 0);
340 pcpu_sigp_retry(pcpu, SIGP_CPU_RESET, 0);
341 /* Restart func on the target cpu and stop the current cpu. */
342 if (lc) {
343 lc->restart_stack = stack;
344 lc->restart_fn = (unsigned long)func;
345 lc->restart_data = (unsigned long)data;
346 lc->restart_source = source_cpu;
347 } else {
348 abs_lc = get_abs_lowcore();
349 abs_lc->restart_stack = stack;
350 abs_lc->restart_fn = (unsigned long)func;
351 abs_lc->restart_data = (unsigned long)data;
352 abs_lc->restart_source = source_cpu;
353 put_abs_lowcore(abs_lc);
354 }
355 asm volatile(
356 "0: sigp 0,%0,%2 # sigp restart to target cpu\n"
357 " brc 2,0b # busy, try again\n"
358 "1: sigp 0,%1,%3 # sigp stop to current cpu\n"
359 " brc 2,1b # busy, try again\n"
360 : : "d" (pcpu->address), "d" (source_cpu),
361 "K" (SIGP_RESTART), "K" (SIGP_STOP)
362 : "0", "1", "cc");
363 for (;;) ;
364}
365
366/*
367 * Enable additional logical cpus for multi-threading.
368 */
369static int pcpu_set_smt(unsigned int mtid)
370{
371 int cc;
372
373 if (smp_cpu_mtid == mtid)
374 return 0;
375 cc = __pcpu_sigp(0, SIGP_SET_MULTI_THREADING, mtid, NULL);
376 if (cc == 0) {
377 smp_cpu_mtid = mtid;
378 smp_cpu_mt_shift = 0;
379 while (smp_cpu_mtid >= (1U << smp_cpu_mt_shift))
380 smp_cpu_mt_shift++;
381 pcpu_devices[0].address = stap();
382 }
383 return cc;
384}
385
386/*
387 * Call function on an online CPU.
388 */
389void smp_call_online_cpu(void (*func)(void *), void *data)
390{
391 struct pcpu *pcpu;
392
393 /* Use the current cpu if it is online. */
394 pcpu = pcpu_find_address(cpu_online_mask, stap());
395 if (!pcpu)
396 /* Use the first online cpu. */
397 pcpu = pcpu_devices + cpumask_first(cpu_online_mask);
398 pcpu_delegate(pcpu, func, data, (unsigned long) restart_stack);
399}
400
401/*
402 * Call function on the ipl CPU.
403 */
404void smp_call_ipl_cpu(void (*func)(void *), void *data)
405{
406 struct lowcore *lc = lowcore_ptr[0];
407
408 if (pcpu_devices[0].address == stap())
409 lc = &S390_lowcore;
410
411 pcpu_delegate(&pcpu_devices[0], func, data,
412 lc->nodat_stack);
413}
414
415int smp_find_processor_id(u16 address)
416{
417 int cpu;
418
419 for_each_present_cpu(cpu)
420 if (pcpu_devices[cpu].address == address)
421 return cpu;
422 return -1;
423}
424
425void schedule_mcck_handler(void)
426{
427 pcpu_ec_call(pcpu_devices + smp_processor_id(), ec_mcck_pending);
428}
429
430bool notrace arch_vcpu_is_preempted(int cpu)
431{
432 if (test_cpu_flag_of(CIF_ENABLED_WAIT, cpu))
433 return false;
434 if (pcpu_running(pcpu_devices + cpu))
435 return false;
436 return true;
437}
438EXPORT_SYMBOL(arch_vcpu_is_preempted);
439
440void notrace smp_yield_cpu(int cpu)
441{
442 if (!MACHINE_HAS_DIAG9C)
443 return;
444 diag_stat_inc_norecursion(DIAG_STAT_X09C);
445 asm volatile("diag %0,0,0x9c"
446 : : "d" (pcpu_devices[cpu].address));
447}
448EXPORT_SYMBOL_GPL(smp_yield_cpu);
449
450/*
451 * Send cpus emergency shutdown signal. This gives the cpus the
452 * opportunity to complete outstanding interrupts.
453 */
454void notrace smp_emergency_stop(void)
455{
456 static arch_spinlock_t lock = __ARCH_SPIN_LOCK_UNLOCKED;
457 static cpumask_t cpumask;
458 u64 end;
459 int cpu;
460
461 arch_spin_lock(&lock);
462 cpumask_copy(&cpumask, cpu_online_mask);
463 cpumask_clear_cpu(smp_processor_id(), &cpumask);
464
465 end = get_tod_clock() + (1000000UL << 12);
466 for_each_cpu(cpu, &cpumask) {
467 struct pcpu *pcpu = pcpu_devices + cpu;
468 set_bit(ec_stop_cpu, &pcpu->ec_mask);
469 while (__pcpu_sigp(pcpu->address, SIGP_EMERGENCY_SIGNAL,
470 0, NULL) == SIGP_CC_BUSY &&
471 get_tod_clock() < end)
472 cpu_relax();
473 }
474 while (get_tod_clock() < end) {
475 for_each_cpu(cpu, &cpumask)
476 if (pcpu_stopped(pcpu_devices + cpu))
477 cpumask_clear_cpu(cpu, &cpumask);
478 if (cpumask_empty(&cpumask))
479 break;
480 cpu_relax();
481 }
482 arch_spin_unlock(&lock);
483}
484NOKPROBE_SYMBOL(smp_emergency_stop);
485
486/*
487 * Stop all cpus but the current one.
488 */
489void smp_send_stop(void)
490{
491 int cpu;
492
493 /* Disable all interrupts/machine checks */
494 __load_psw_mask(PSW_KERNEL_BITS);
495 trace_hardirqs_off();
496
497 debug_set_critical();
498
499 if (oops_in_progress)
500 smp_emergency_stop();
501
502 /* stop all processors */
503 for_each_online_cpu(cpu) {
504 if (cpu == smp_processor_id())
505 continue;
506 pcpu_sigp_retry(pcpu_devices + cpu, SIGP_STOP, 0);
507 while (!pcpu_stopped(pcpu_devices + cpu))
508 cpu_relax();
509 }
510}
511
512/*
513 * This is the main routine where commands issued by other
514 * cpus are handled.
515 */
516static void smp_handle_ext_call(void)
517{
518 unsigned long bits;
519
520 /* handle bit signal external calls */
521 bits = xchg(&pcpu_devices[smp_processor_id()].ec_mask, 0);
522 if (test_bit(ec_stop_cpu, &bits))
523 smp_stop_cpu();
524 if (test_bit(ec_schedule, &bits))
525 scheduler_ipi();
526 if (test_bit(ec_call_function_single, &bits))
527 generic_smp_call_function_single_interrupt();
528 if (test_bit(ec_mcck_pending, &bits))
529 s390_handle_mcck();
530 if (test_bit(ec_irq_work, &bits))
531 irq_work_run();
532}
533
534static void do_ext_call_interrupt(struct ext_code ext_code,
535 unsigned int param32, unsigned long param64)
536{
537 inc_irq_stat(ext_code.code == 0x1202 ? IRQEXT_EXC : IRQEXT_EMS);
538 smp_handle_ext_call();
539}
540
541void arch_send_call_function_ipi_mask(const struct cpumask *mask)
542{
543 int cpu;
544
545 for_each_cpu(cpu, mask)
546 pcpu_ec_call(pcpu_devices + cpu, ec_call_function_single);
547}
548
549void arch_send_call_function_single_ipi(int cpu)
550{
551 pcpu_ec_call(pcpu_devices + cpu, ec_call_function_single);
552}
553
554/*
555 * this function sends a 'reschedule' IPI to another CPU.
556 * it goes straight through and wastes no time serializing
557 * anything. Worst case is that we lose a reschedule ...
558 */
559void arch_smp_send_reschedule(int cpu)
560{
561 pcpu_ec_call(pcpu_devices + cpu, ec_schedule);
562}
563
564#ifdef CONFIG_IRQ_WORK
565void arch_irq_work_raise(void)
566{
567 pcpu_ec_call(pcpu_devices + smp_processor_id(), ec_irq_work);
568}
569#endif
570
571#ifdef CONFIG_CRASH_DUMP
572
573int smp_store_status(int cpu)
574{
575 struct lowcore *lc;
576 struct pcpu *pcpu;
577 unsigned long pa;
578
579 pcpu = pcpu_devices + cpu;
580 lc = lowcore_ptr[cpu];
581 pa = __pa(&lc->floating_pt_save_area);
582 if (__pcpu_sigp_relax(pcpu->address, SIGP_STORE_STATUS_AT_ADDRESS,
583 pa) != SIGP_CC_ORDER_CODE_ACCEPTED)
584 return -EIO;
585 if (!cpu_has_vx() && !MACHINE_HAS_GS)
586 return 0;
587 pa = lc->mcesad & MCESA_ORIGIN_MASK;
588 if (MACHINE_HAS_GS)
589 pa |= lc->mcesad & MCESA_LC_MASK;
590 if (__pcpu_sigp_relax(pcpu->address, SIGP_STORE_ADDITIONAL_STATUS,
591 pa) != SIGP_CC_ORDER_CODE_ACCEPTED)
592 return -EIO;
593 return 0;
594}
595
596/*
597 * Collect CPU state of the previous, crashed system.
598 * There are four cases:
599 * 1) standard zfcp/nvme dump
600 * condition: OLDMEM_BASE == NULL && is_ipl_type_dump() == true
601 * The state for all CPUs except the boot CPU needs to be collected
602 * with sigp stop-and-store-status. The boot CPU state is located in
603 * the absolute lowcore of the memory stored in the HSA. The zcore code
604 * will copy the boot CPU state from the HSA.
605 * 2) stand-alone kdump for SCSI/NVMe (zfcp/nvme dump with swapped memory)
606 * condition: OLDMEM_BASE != NULL && is_ipl_type_dump() == true
607 * The state for all CPUs except the boot CPU needs to be collected
608 * with sigp stop-and-store-status. The firmware or the boot-loader
609 * stored the registers of the boot CPU in the absolute lowcore in the
610 * memory of the old system.
611 * 3) kdump and the old kernel did not store the CPU state,
612 * or stand-alone kdump for DASD
613 * condition: OLDMEM_BASE != NULL && !is_kdump_kernel()
614 * The state for all CPUs except the boot CPU needs to be collected
615 * with sigp stop-and-store-status. The kexec code or the boot-loader
616 * stored the registers of the boot CPU in the memory of the old system.
617 * 4) kdump and the old kernel stored the CPU state
618 * condition: OLDMEM_BASE != NULL && is_kdump_kernel()
619 * This case does not exist for s390 anymore, setup_arch explicitly
620 * deactivates the elfcorehdr= kernel parameter
621 */
622static bool dump_available(void)
623{
624 return oldmem_data.start || is_ipl_type_dump();
625}
626
627void __init smp_save_dump_ipl_cpu(void)
628{
629 struct save_area *sa;
630 void *regs;
631
632 if (!dump_available())
633 return;
634 sa = save_area_alloc(true);
635 regs = memblock_alloc(512, 8);
636 if (!sa || !regs)
637 panic("could not allocate memory for boot CPU save area\n");
638 copy_oldmem_kernel(regs, __LC_FPREGS_SAVE_AREA, 512);
639 save_area_add_regs(sa, regs);
640 memblock_free(regs, 512);
641 if (cpu_has_vx())
642 save_area_add_vxrs(sa, boot_cpu_vector_save_area);
643}
644
645void __init smp_save_dump_secondary_cpus(void)
646{
647 int addr, boot_cpu_addr, max_cpu_addr;
648 struct save_area *sa;
649 void *page;
650
651 if (!dump_available())
652 return;
653 /* Allocate a page as dumping area for the store status sigps */
654 page = memblock_alloc_low(PAGE_SIZE, PAGE_SIZE);
655 if (!page)
656 panic("ERROR: Failed to allocate %lx bytes below %lx\n",
657 PAGE_SIZE, 1UL << 31);
658
659 /* Set multi-threading state to the previous system. */
660 pcpu_set_smt(sclp.mtid_prev);
661 boot_cpu_addr = stap();
662 max_cpu_addr = SCLP_MAX_CORES << sclp.mtid_prev;
663 for (addr = 0; addr <= max_cpu_addr; addr++) {
664 if (addr == boot_cpu_addr)
665 continue;
666 if (__pcpu_sigp_relax(addr, SIGP_SENSE, 0) ==
667 SIGP_CC_NOT_OPERATIONAL)
668 continue;
669 sa = save_area_alloc(false);
670 if (!sa)
671 panic("could not allocate memory for save area\n");
672 __pcpu_sigp_relax(addr, SIGP_STORE_STATUS_AT_ADDRESS, __pa(page));
673 save_area_add_regs(sa, page);
674 if (cpu_has_vx()) {
675 __pcpu_sigp_relax(addr, SIGP_STORE_ADDITIONAL_STATUS, __pa(page));
676 save_area_add_vxrs(sa, page);
677 }
678 }
679 memblock_free(page, PAGE_SIZE);
680 diag_amode31_ops.diag308_reset();
681 pcpu_set_smt(0);
682}
683#endif /* CONFIG_CRASH_DUMP */
684
685void smp_cpu_set_polarization(int cpu, int val)
686{
687 pcpu_devices[cpu].polarization = val;
688}
689
690int smp_cpu_get_polarization(int cpu)
691{
692 return pcpu_devices[cpu].polarization;
693}
694
695int smp_cpu_get_cpu_address(int cpu)
696{
697 return pcpu_devices[cpu].address;
698}
699
700static void __ref smp_get_core_info(struct sclp_core_info *info, int early)
701{
702 static int use_sigp_detection;
703 int address;
704
705 if (use_sigp_detection || sclp_get_core_info(info, early)) {
706 use_sigp_detection = 1;
707 for (address = 0;
708 address < (SCLP_MAX_CORES << smp_cpu_mt_shift);
709 address += (1U << smp_cpu_mt_shift)) {
710 if (__pcpu_sigp_relax(address, SIGP_SENSE, 0) ==
711 SIGP_CC_NOT_OPERATIONAL)
712 continue;
713 info->core[info->configured].core_id =
714 address >> smp_cpu_mt_shift;
715 info->configured++;
716 }
717 info->combined = info->configured;
718 }
719}
720
721static int smp_add_present_cpu(int cpu);
722
723static int smp_add_core(struct sclp_core_entry *core, cpumask_t *avail,
724 bool configured, bool early)
725{
726 struct pcpu *pcpu;
727 int cpu, nr, i;
728 u16 address;
729
730 nr = 0;
731 if (sclp.has_core_type && core->type != boot_core_type)
732 return nr;
733 cpu = cpumask_first(avail);
734 address = core->core_id << smp_cpu_mt_shift;
735 for (i = 0; (i <= smp_cpu_mtid) && (cpu < nr_cpu_ids); i++) {
736 if (pcpu_find_address(cpu_present_mask, address + i))
737 continue;
738 pcpu = pcpu_devices + cpu;
739 pcpu->address = address + i;
740 if (configured)
741 pcpu->state = CPU_STATE_CONFIGURED;
742 else
743 pcpu->state = CPU_STATE_STANDBY;
744 smp_cpu_set_polarization(cpu, POLARIZATION_UNKNOWN);
745 set_cpu_present(cpu, true);
746 if (!early && smp_add_present_cpu(cpu) != 0)
747 set_cpu_present(cpu, false);
748 else
749 nr++;
750 cpumask_clear_cpu(cpu, avail);
751 cpu = cpumask_next(cpu, avail);
752 }
753 return nr;
754}
755
756static int __smp_rescan_cpus(struct sclp_core_info *info, bool early)
757{
758 struct sclp_core_entry *core;
759 static cpumask_t avail;
760 bool configured;
761 u16 core_id;
762 int nr, i;
763
764 cpus_read_lock();
765 mutex_lock(&smp_cpu_state_mutex);
766 nr = 0;
767 cpumask_xor(&avail, cpu_possible_mask, cpu_present_mask);
768 /*
769 * Add IPL core first (which got logical CPU number 0) to make sure
770 * that all SMT threads get subsequent logical CPU numbers.
771 */
772 if (early) {
773 core_id = pcpu_devices[0].address >> smp_cpu_mt_shift;
774 for (i = 0; i < info->configured; i++) {
775 core = &info->core[i];
776 if (core->core_id == core_id) {
777 nr += smp_add_core(core, &avail, true, early);
778 break;
779 }
780 }
781 }
782 for (i = 0; i < info->combined; i++) {
783 configured = i < info->configured;
784 nr += smp_add_core(&info->core[i], &avail, configured, early);
785 }
786 mutex_unlock(&smp_cpu_state_mutex);
787 cpus_read_unlock();
788 return nr;
789}
790
791void __init smp_detect_cpus(void)
792{
793 unsigned int cpu, mtid, c_cpus, s_cpus;
794 struct sclp_core_info *info;
795 u16 address;
796
797 /* Get CPU information */
798 info = memblock_alloc(sizeof(*info), 8);
799 if (!info)
800 panic("%s: Failed to allocate %zu bytes align=0x%x\n",
801 __func__, sizeof(*info), 8);
802 smp_get_core_info(info, 1);
803 /* Find boot CPU type */
804 if (sclp.has_core_type) {
805 address = stap();
806 for (cpu = 0; cpu < info->combined; cpu++)
807 if (info->core[cpu].core_id == address) {
808 /* The boot cpu dictates the cpu type. */
809 boot_core_type = info->core[cpu].type;
810 break;
811 }
812 if (cpu >= info->combined)
813 panic("Could not find boot CPU type");
814 }
815
816 /* Set multi-threading state for the current system */
817 mtid = boot_core_type ? sclp.mtid : sclp.mtid_cp;
818 mtid = (mtid < smp_max_threads) ? mtid : smp_max_threads - 1;
819 pcpu_set_smt(mtid);
820
821 /* Print number of CPUs */
822 c_cpus = s_cpus = 0;
823 for (cpu = 0; cpu < info->combined; cpu++) {
824 if (sclp.has_core_type &&
825 info->core[cpu].type != boot_core_type)
826 continue;
827 if (cpu < info->configured)
828 c_cpus += smp_cpu_mtid + 1;
829 else
830 s_cpus += smp_cpu_mtid + 1;
831 }
832 pr_info("%d configured CPUs, %d standby CPUs\n", c_cpus, s_cpus);
833
834 /* Add CPUs present at boot */
835 __smp_rescan_cpus(info, true);
836 memblock_free(info, sizeof(*info));
837}
838
839/*
840 * Activate a secondary processor.
841 */
842static void smp_start_secondary(void *cpuvoid)
843{
844 int cpu = raw_smp_processor_id();
845
846 S390_lowcore.last_update_clock = get_tod_clock();
847 S390_lowcore.restart_stack = (unsigned long)restart_stack;
848 S390_lowcore.restart_fn = (unsigned long)do_restart;
849 S390_lowcore.restart_data = 0;
850 S390_lowcore.restart_source = -1U;
851 S390_lowcore.restart_flags = 0;
852 restore_access_regs(S390_lowcore.access_regs_save_area);
853 cpu_init();
854 rcutree_report_cpu_starting(cpu);
855 init_cpu_timer();
856 vtime_init();
857 vdso_getcpu_init();
858 pfault_init();
859 cpumask_set_cpu(cpu, &cpu_setup_mask);
860 update_cpu_masks();
861 notify_cpu_starting(cpu);
862 if (topology_cpu_dedicated(cpu))
863 set_cpu_flag(CIF_DEDICATED_CPU);
864 else
865 clear_cpu_flag(CIF_DEDICATED_CPU);
866 set_cpu_online(cpu, true);
867 inc_irq_stat(CPU_RST);
868 local_irq_enable();
869 cpu_startup_entry(CPUHP_AP_ONLINE_IDLE);
870}
871
872/* Upping and downing of CPUs */
873int __cpu_up(unsigned int cpu, struct task_struct *tidle)
874{
875 struct pcpu *pcpu = pcpu_devices + cpu;
876 int rc;
877
878 if (pcpu->state != CPU_STATE_CONFIGURED)
879 return -EIO;
880 if (pcpu_sigp_retry(pcpu, SIGP_INITIAL_CPU_RESET, 0) !=
881 SIGP_CC_ORDER_CODE_ACCEPTED)
882 return -EIO;
883
884 rc = pcpu_alloc_lowcore(pcpu, cpu);
885 if (rc)
886 return rc;
887 /*
888 * Make sure global control register contents do not change
889 * until new CPU has initialized control registers.
890 */
891 system_ctlreg_lock();
892 pcpu_prepare_secondary(pcpu, cpu);
893 pcpu_attach_task(pcpu, tidle);
894 pcpu_start_fn(pcpu, smp_start_secondary, NULL);
895 /* Wait until cpu puts itself in the online & active maps */
896 while (!cpu_online(cpu))
897 cpu_relax();
898 system_ctlreg_unlock();
899 return 0;
900}
901
902static unsigned int setup_possible_cpus __initdata;
903
904static int __init _setup_possible_cpus(char *s)
905{
906 get_option(&s, &setup_possible_cpus);
907 return 0;
908}
909early_param("possible_cpus", _setup_possible_cpus);
910
911int __cpu_disable(void)
912{
913 struct ctlreg cregs[16];
914 int cpu;
915
916 /* Handle possible pending IPIs */
917 smp_handle_ext_call();
918 cpu = smp_processor_id();
919 set_cpu_online(cpu, false);
920 cpumask_clear_cpu(cpu, &cpu_setup_mask);
921 update_cpu_masks();
922 /* Disable pseudo page faults on this cpu. */
923 pfault_fini();
924 /* Disable interrupt sources via control register. */
925 __local_ctl_store(0, 15, cregs);
926 cregs[0].val &= ~0x0000ee70UL; /* disable all external interrupts */
927 cregs[6].val &= ~0xff000000UL; /* disable all I/O interrupts */
928 cregs[14].val &= ~0x1f000000UL; /* disable most machine checks */
929 __local_ctl_load(0, 15, cregs);
930 clear_cpu_flag(CIF_NOHZ_DELAY);
931 return 0;
932}
933
934void __cpu_die(unsigned int cpu)
935{
936 struct pcpu *pcpu;
937
938 /* Wait until target cpu is down */
939 pcpu = pcpu_devices + cpu;
940 while (!pcpu_stopped(pcpu))
941 cpu_relax();
942 pcpu_free_lowcore(pcpu);
943 cpumask_clear_cpu(cpu, mm_cpumask(&init_mm));
944 cpumask_clear_cpu(cpu, &init_mm.context.cpu_attach_mask);
945}
946
947void __noreturn cpu_die(void)
948{
949 idle_task_exit();
950 pcpu_sigp_retry(pcpu_devices + smp_processor_id(), SIGP_STOP, 0);
951 for (;;) ;
952}
953
954void __init smp_fill_possible_mask(void)
955{
956 unsigned int possible, sclp_max, cpu;
957
958 sclp_max = max(sclp.mtid, sclp.mtid_cp) + 1;
959 sclp_max = min(smp_max_threads, sclp_max);
960 sclp_max = (sclp.max_cores * sclp_max) ?: nr_cpu_ids;
961 possible = setup_possible_cpus ?: nr_cpu_ids;
962 possible = min(possible, sclp_max);
963 for (cpu = 0; cpu < possible && cpu < nr_cpu_ids; cpu++)
964 set_cpu_possible(cpu, true);
965}
966
967void __init smp_prepare_cpus(unsigned int max_cpus)
968{
969 if (register_external_irq(EXT_IRQ_EMERGENCY_SIG, do_ext_call_interrupt))
970 panic("Couldn't request external interrupt 0x1201");
971 system_ctl_set_bit(0, 14);
972 if (register_external_irq(EXT_IRQ_EXTERNAL_CALL, do_ext_call_interrupt))
973 panic("Couldn't request external interrupt 0x1202");
974 system_ctl_set_bit(0, 13);
975}
976
977void __init smp_prepare_boot_cpu(void)
978{
979 struct pcpu *pcpu = pcpu_devices;
980
981 WARN_ON(!cpu_present(0) || !cpu_online(0));
982 pcpu->state = CPU_STATE_CONFIGURED;
983 S390_lowcore.percpu_offset = __per_cpu_offset[0];
984 smp_cpu_set_polarization(0, POLARIZATION_UNKNOWN);
985}
986
987void __init smp_setup_processor_id(void)
988{
989 pcpu_devices[0].address = stap();
990 S390_lowcore.cpu_nr = 0;
991 S390_lowcore.spinlock_lockval = arch_spin_lockval(0);
992 S390_lowcore.spinlock_index = 0;
993}
994
995/*
996 * the frequency of the profiling timer can be changed
997 * by writing a multiplier value into /proc/profile.
998 *
999 * usually you want to run this on all CPUs ;)
1000 */
1001int setup_profiling_timer(unsigned int multiplier)
1002{
1003 return 0;
1004}
1005
1006static ssize_t cpu_configure_show(struct device *dev,
1007 struct device_attribute *attr, char *buf)
1008{
1009 ssize_t count;
1010
1011 mutex_lock(&smp_cpu_state_mutex);
1012 count = sprintf(buf, "%d\n", pcpu_devices[dev->id].state);
1013 mutex_unlock(&smp_cpu_state_mutex);
1014 return count;
1015}
1016
1017static ssize_t cpu_configure_store(struct device *dev,
1018 struct device_attribute *attr,
1019 const char *buf, size_t count)
1020{
1021 struct pcpu *pcpu;
1022 int cpu, val, rc, i;
1023 char delim;
1024
1025 if (sscanf(buf, "%d %c", &val, &delim) != 1)
1026 return -EINVAL;
1027 if (val != 0 && val != 1)
1028 return -EINVAL;
1029 cpus_read_lock();
1030 mutex_lock(&smp_cpu_state_mutex);
1031 rc = -EBUSY;
1032 /* disallow configuration changes of online cpus */
1033 cpu = dev->id;
1034 cpu = smp_get_base_cpu(cpu);
1035 for (i = 0; i <= smp_cpu_mtid; i++)
1036 if (cpu_online(cpu + i))
1037 goto out;
1038 pcpu = pcpu_devices + cpu;
1039 rc = 0;
1040 switch (val) {
1041 case 0:
1042 if (pcpu->state != CPU_STATE_CONFIGURED)
1043 break;
1044 rc = sclp_core_deconfigure(pcpu->address >> smp_cpu_mt_shift);
1045 if (rc)
1046 break;
1047 for (i = 0; i <= smp_cpu_mtid; i++) {
1048 if (cpu + i >= nr_cpu_ids || !cpu_present(cpu + i))
1049 continue;
1050 pcpu[i].state = CPU_STATE_STANDBY;
1051 smp_cpu_set_polarization(cpu + i,
1052 POLARIZATION_UNKNOWN);
1053 }
1054 topology_expect_change();
1055 break;
1056 case 1:
1057 if (pcpu->state != CPU_STATE_STANDBY)
1058 break;
1059 rc = sclp_core_configure(pcpu->address >> smp_cpu_mt_shift);
1060 if (rc)
1061 break;
1062 for (i = 0; i <= smp_cpu_mtid; i++) {
1063 if (cpu + i >= nr_cpu_ids || !cpu_present(cpu + i))
1064 continue;
1065 pcpu[i].state = CPU_STATE_CONFIGURED;
1066 smp_cpu_set_polarization(cpu + i,
1067 POLARIZATION_UNKNOWN);
1068 }
1069 topology_expect_change();
1070 break;
1071 default:
1072 break;
1073 }
1074out:
1075 mutex_unlock(&smp_cpu_state_mutex);
1076 cpus_read_unlock();
1077 return rc ? rc : count;
1078}
1079static DEVICE_ATTR(configure, 0644, cpu_configure_show, cpu_configure_store);
1080
1081static ssize_t show_cpu_address(struct device *dev,
1082 struct device_attribute *attr, char *buf)
1083{
1084 return sprintf(buf, "%d\n", pcpu_devices[dev->id].address);
1085}
1086static DEVICE_ATTR(address, 0444, show_cpu_address, NULL);
1087
1088static struct attribute *cpu_common_attrs[] = {
1089 &dev_attr_configure.attr,
1090 &dev_attr_address.attr,
1091 NULL,
1092};
1093
1094static struct attribute_group cpu_common_attr_group = {
1095 .attrs = cpu_common_attrs,
1096};
1097
1098static struct attribute *cpu_online_attrs[] = {
1099 &dev_attr_idle_count.attr,
1100 &dev_attr_idle_time_us.attr,
1101 NULL,
1102};
1103
1104static struct attribute_group cpu_online_attr_group = {
1105 .attrs = cpu_online_attrs,
1106};
1107
1108static int smp_cpu_online(unsigned int cpu)
1109{
1110 struct device *s = &per_cpu(cpu_device, cpu)->dev;
1111
1112 return sysfs_create_group(&s->kobj, &cpu_online_attr_group);
1113}
1114
1115static int smp_cpu_pre_down(unsigned int cpu)
1116{
1117 struct device *s = &per_cpu(cpu_device, cpu)->dev;
1118
1119 sysfs_remove_group(&s->kobj, &cpu_online_attr_group);
1120 return 0;
1121}
1122
1123static int smp_add_present_cpu(int cpu)
1124{
1125 struct device *s;
1126 struct cpu *c;
1127 int rc;
1128
1129 c = kzalloc(sizeof(*c), GFP_KERNEL);
1130 if (!c)
1131 return -ENOMEM;
1132 per_cpu(cpu_device, cpu) = c;
1133 s = &c->dev;
1134 c->hotpluggable = !!cpu;
1135 rc = register_cpu(c, cpu);
1136 if (rc)
1137 goto out;
1138 rc = sysfs_create_group(&s->kobj, &cpu_common_attr_group);
1139 if (rc)
1140 goto out_cpu;
1141 rc = topology_cpu_init(c);
1142 if (rc)
1143 goto out_topology;
1144 return 0;
1145
1146out_topology:
1147 sysfs_remove_group(&s->kobj, &cpu_common_attr_group);
1148out_cpu:
1149 unregister_cpu(c);
1150out:
1151 return rc;
1152}
1153
1154int __ref smp_rescan_cpus(void)
1155{
1156 struct sclp_core_info *info;
1157 int nr;
1158
1159 info = kzalloc(sizeof(*info), GFP_KERNEL);
1160 if (!info)
1161 return -ENOMEM;
1162 smp_get_core_info(info, 0);
1163 nr = __smp_rescan_cpus(info, false);
1164 kfree(info);
1165 if (nr)
1166 topology_schedule_update();
1167 return 0;
1168}
1169
1170static ssize_t __ref rescan_store(struct device *dev,
1171 struct device_attribute *attr,
1172 const char *buf,
1173 size_t count)
1174{
1175 int rc;
1176
1177 rc = lock_device_hotplug_sysfs();
1178 if (rc)
1179 return rc;
1180 rc = smp_rescan_cpus();
1181 unlock_device_hotplug();
1182 return rc ? rc : count;
1183}
1184static DEVICE_ATTR_WO(rescan);
1185
1186static int __init s390_smp_init(void)
1187{
1188 struct device *dev_root;
1189 int cpu, rc = 0;
1190
1191 dev_root = bus_get_dev_root(&cpu_subsys);
1192 if (dev_root) {
1193 rc = device_create_file(dev_root, &dev_attr_rescan);
1194 put_device(dev_root);
1195 if (rc)
1196 return rc;
1197 }
1198
1199 for_each_present_cpu(cpu) {
1200 rc = smp_add_present_cpu(cpu);
1201 if (rc)
1202 goto out;
1203 }
1204
1205 rc = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "s390/smp:online",
1206 smp_cpu_online, smp_cpu_pre_down);
1207 rc = rc <= 0 ? rc : 0;
1208out:
1209 return rc;
1210}
1211subsys_initcall(s390_smp_init);