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 * Heiko Carstens <heiko.carstens@de.ibm.com>,
9 *
10 * based on other smp stuff by
11 * (c) 1995 Alan Cox, CymruNET Ltd <alan@cymru.net>
12 * (c) 1998 Ingo Molnar
13 *
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.
17 */
18
19#define KMSG_COMPONENT "cpu"
20#define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
21
22#include <linux/workqueue.h>
23#include <linux/memblock.h>
24#include <linux/export.h>
25#include <linux/init.h>
26#include <linux/mm.h>
27#include <linux/err.h>
28#include <linux/spinlock.h>
29#include <linux/kernel_stat.h>
30#include <linux/delay.h>
31#include <linux/interrupt.h>
32#include <linux/irqflags.h>
33#include <linux/irq_work.h>
34#include <linux/cpu.h>
35#include <linux/slab.h>
36#include <linux/sched/hotplug.h>
37#include <linux/sched/task_stack.h>
38#include <linux/crash_dump.h>
39#include <linux/kprobes.h>
40#include <asm/asm-offsets.h>
41#include <asm/diag.h>
42#include <asm/switch_to.h>
43#include <asm/facility.h>
44#include <asm/ipl.h>
45#include <asm/setup.h>
46#include <asm/irq.h>
47#include <asm/tlbflush.h>
48#include <asm/vtimer.h>
49#include <asm/lowcore.h>
50#include <asm/sclp.h>
51#include <asm/debug.h>
52#include <asm/os_info.h>
53#include <asm/sigp.h>
54#include <asm/idle.h>
55#include <asm/nmi.h>
56#include <asm/stacktrace.h>
57#include <asm/topology.h>
58#include <asm/vdso.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->br_r1_trampoline = 0x07f1; /* br %r1 */
212 lc->return_lpswe = gen_lpswe(__LC_RETURN_PSW);
213 lc->return_mcck_lpswe = gen_lpswe(__LC_RETURN_MCCK_PSW);
214 lc->preempt_count = PREEMPT_DISABLED;
215 if (nmi_alloc_per_cpu(lc))
216 goto out;
217 lowcore_ptr[cpu] = lc;
218 pcpu_sigp_retry(pcpu, SIGP_SET_PREFIX, (u32)(unsigned long) lc);
219 return 0;
220
221out:
222 stack_free(mcck_stack);
223 stack_free(async_stack);
224 free_pages(nodat_stack, THREAD_SIZE_ORDER);
225 free_pages((unsigned long) lc, LC_ORDER);
226 return -ENOMEM;
227}
228
229static void pcpu_free_lowcore(struct pcpu *pcpu)
230{
231 unsigned long async_stack, nodat_stack, mcck_stack;
232 struct lowcore *lc;
233 int cpu;
234
235 cpu = pcpu - pcpu_devices;
236 lc = lowcore_ptr[cpu];
237 nodat_stack = lc->nodat_stack - STACK_INIT_OFFSET;
238 async_stack = lc->async_stack - STACK_INIT_OFFSET;
239 mcck_stack = lc->mcck_stack - STACK_INIT_OFFSET;
240 pcpu_sigp_retry(pcpu, SIGP_SET_PREFIX, 0);
241 lowcore_ptr[cpu] = NULL;
242 nmi_free_per_cpu(lc);
243 stack_free(async_stack);
244 stack_free(mcck_stack);
245 free_pages(nodat_stack, THREAD_SIZE_ORDER);
246 free_pages((unsigned long) lc, LC_ORDER);
247}
248
249static void pcpu_prepare_secondary(struct pcpu *pcpu, int cpu)
250{
251 struct lowcore *lc = lowcore_ptr[cpu];
252
253 cpumask_set_cpu(cpu, &init_mm.context.cpu_attach_mask);
254 cpumask_set_cpu(cpu, mm_cpumask(&init_mm));
255 lc->cpu_nr = cpu;
256 lc->restart_flags = RESTART_FLAG_CTLREGS;
257 lc->spinlock_lockval = arch_spin_lockval(cpu);
258 lc->spinlock_index = 0;
259 lc->percpu_offset = __per_cpu_offset[cpu];
260 lc->kernel_asce = S390_lowcore.kernel_asce;
261 lc->user_asce = s390_invalid_asce;
262 lc->machine_flags = S390_lowcore.machine_flags;
263 lc->user_timer = lc->system_timer =
264 lc->steal_timer = lc->avg_steal_timer = 0;
265 __ctl_store(lc->cregs_save_area, 0, 15);
266 lc->cregs_save_area[1] = lc->kernel_asce;
267 lc->cregs_save_area[7] = lc->user_asce;
268 save_access_regs((unsigned int *) lc->access_regs_save_area);
269 arch_spin_lock_setup(cpu);
270}
271
272static void pcpu_attach_task(struct pcpu *pcpu, struct task_struct *tsk)
273{
274 struct lowcore *lc;
275 int cpu;
276
277 cpu = pcpu - pcpu_devices;
278 lc = lowcore_ptr[cpu];
279 lc->kernel_stack = (unsigned long) task_stack_page(tsk)
280 + THREAD_SIZE - STACK_FRAME_OVERHEAD - sizeof(struct pt_regs);
281 lc->current_task = (unsigned long) tsk;
282 lc->lpp = LPP_MAGIC;
283 lc->current_pid = tsk->pid;
284 lc->user_timer = tsk->thread.user_timer;
285 lc->guest_timer = tsk->thread.guest_timer;
286 lc->system_timer = tsk->thread.system_timer;
287 lc->hardirq_timer = tsk->thread.hardirq_timer;
288 lc->softirq_timer = tsk->thread.softirq_timer;
289 lc->steal_timer = 0;
290}
291
292static void pcpu_start_fn(struct pcpu *pcpu, void (*func)(void *), void *data)
293{
294 struct lowcore *lc;
295 int cpu;
296
297 cpu = pcpu - pcpu_devices;
298 lc = lowcore_ptr[cpu];
299 lc->restart_stack = lc->nodat_stack;
300 lc->restart_fn = (unsigned long) func;
301 lc->restart_data = (unsigned long) data;
302 lc->restart_source = -1U;
303 pcpu_sigp_retry(pcpu, SIGP_RESTART, 0);
304}
305
306typedef void (pcpu_delegate_fn)(void *);
307
308/*
309 * Call function via PSW restart on pcpu and stop the current cpu.
310 */
311static void __pcpu_delegate(pcpu_delegate_fn *func, void *data)
312{
313 func(data); /* should not return */
314}
315
316static void pcpu_delegate(struct pcpu *pcpu,
317 pcpu_delegate_fn *func,
318 void *data, unsigned long stack)
319{
320 struct lowcore *lc = lowcore_ptr[pcpu - pcpu_devices];
321 unsigned int source_cpu = stap();
322
323 __load_psw_mask(PSW_KERNEL_BITS | PSW_MASK_DAT);
324 if (pcpu->address == source_cpu) {
325 call_on_stack(2, stack, void, __pcpu_delegate,
326 pcpu_delegate_fn *, func, void *, data);
327 }
328 /* Stop target cpu (if func returns this stops the current cpu). */
329 pcpu_sigp_retry(pcpu, SIGP_STOP, 0);
330 /* Restart func on the target cpu and stop the current cpu. */
331 mem_assign_absolute(lc->restart_stack, stack);
332 mem_assign_absolute(lc->restart_fn, (unsigned long) func);
333 mem_assign_absolute(lc->restart_data, (unsigned long) data);
334 mem_assign_absolute(lc->restart_source, source_cpu);
335 __bpon();
336 asm volatile(
337 "0: sigp 0,%0,%2 # sigp restart to target cpu\n"
338 " brc 2,0b # busy, try again\n"
339 "1: sigp 0,%1,%3 # sigp stop to current cpu\n"
340 " brc 2,1b # busy, try again\n"
341 : : "d" (pcpu->address), "d" (source_cpu),
342 "K" (SIGP_RESTART), "K" (SIGP_STOP)
343 : "0", "1", "cc");
344 for (;;) ;
345}
346
347/*
348 * Enable additional logical cpus for multi-threading.
349 */
350static int pcpu_set_smt(unsigned int mtid)
351{
352 int cc;
353
354 if (smp_cpu_mtid == mtid)
355 return 0;
356 cc = __pcpu_sigp(0, SIGP_SET_MULTI_THREADING, mtid, NULL);
357 if (cc == 0) {
358 smp_cpu_mtid = mtid;
359 smp_cpu_mt_shift = 0;
360 while (smp_cpu_mtid >= (1U << smp_cpu_mt_shift))
361 smp_cpu_mt_shift++;
362 pcpu_devices[0].address = stap();
363 }
364 return cc;
365}
366
367/*
368 * Call function on an online CPU.
369 */
370void smp_call_online_cpu(void (*func)(void *), void *data)
371{
372 struct pcpu *pcpu;
373
374 /* Use the current cpu if it is online. */
375 pcpu = pcpu_find_address(cpu_online_mask, stap());
376 if (!pcpu)
377 /* Use the first online cpu. */
378 pcpu = pcpu_devices + cpumask_first(cpu_online_mask);
379 pcpu_delegate(pcpu, func, data, (unsigned long) restart_stack);
380}
381
382/*
383 * Call function on the ipl CPU.
384 */
385void smp_call_ipl_cpu(void (*func)(void *), void *data)
386{
387 struct lowcore *lc = lowcore_ptr[0];
388
389 if (pcpu_devices[0].address == stap())
390 lc = &S390_lowcore;
391
392 pcpu_delegate(&pcpu_devices[0], func, data,
393 lc->nodat_stack);
394}
395
396int smp_find_processor_id(u16 address)
397{
398 int cpu;
399
400 for_each_present_cpu(cpu)
401 if (pcpu_devices[cpu].address == address)
402 return cpu;
403 return -1;
404}
405
406void schedule_mcck_handler(void)
407{
408 pcpu_ec_call(pcpu_devices + smp_processor_id(), ec_mcck_pending);
409}
410
411bool notrace arch_vcpu_is_preempted(int cpu)
412{
413 if (test_cpu_flag_of(CIF_ENABLED_WAIT, cpu))
414 return false;
415 if (pcpu_running(pcpu_devices + cpu))
416 return false;
417 return true;
418}
419EXPORT_SYMBOL(arch_vcpu_is_preempted);
420
421void notrace smp_yield_cpu(int cpu)
422{
423 if (!MACHINE_HAS_DIAG9C)
424 return;
425 diag_stat_inc_norecursion(DIAG_STAT_X09C);
426 asm volatile("diag %0,0,0x9c"
427 : : "d" (pcpu_devices[cpu].address));
428}
429EXPORT_SYMBOL_GPL(smp_yield_cpu);
430
431/*
432 * Send cpus emergency shutdown signal. This gives the cpus the
433 * opportunity to complete outstanding interrupts.
434 */
435void notrace smp_emergency_stop(void)
436{
437 static arch_spinlock_t lock = __ARCH_SPIN_LOCK_UNLOCKED;
438 static cpumask_t cpumask;
439 u64 end;
440 int cpu;
441
442 arch_spin_lock(&lock);
443 cpumask_copy(&cpumask, cpu_online_mask);
444 cpumask_clear_cpu(smp_processor_id(), &cpumask);
445
446 end = get_tod_clock() + (1000000UL << 12);
447 for_each_cpu(cpu, &cpumask) {
448 struct pcpu *pcpu = pcpu_devices + cpu;
449 set_bit(ec_stop_cpu, &pcpu->ec_mask);
450 while (__pcpu_sigp(pcpu->address, SIGP_EMERGENCY_SIGNAL,
451 0, NULL) == SIGP_CC_BUSY &&
452 get_tod_clock() < end)
453 cpu_relax();
454 }
455 while (get_tod_clock() < end) {
456 for_each_cpu(cpu, &cpumask)
457 if (pcpu_stopped(pcpu_devices + cpu))
458 cpumask_clear_cpu(cpu, &cpumask);
459 if (cpumask_empty(&cpumask))
460 break;
461 cpu_relax();
462 }
463 arch_spin_unlock(&lock);
464}
465NOKPROBE_SYMBOL(smp_emergency_stop);
466
467/*
468 * Stop all cpus but the current one.
469 */
470void smp_send_stop(void)
471{
472 int cpu;
473
474 /* Disable all interrupts/machine checks */
475 __load_psw_mask(PSW_KERNEL_BITS | PSW_MASK_DAT);
476 trace_hardirqs_off();
477
478 debug_set_critical();
479
480 if (oops_in_progress)
481 smp_emergency_stop();
482
483 /* stop all processors */
484 for_each_online_cpu(cpu) {
485 if (cpu == smp_processor_id())
486 continue;
487 pcpu_sigp_retry(pcpu_devices + cpu, SIGP_STOP, 0);
488 while (!pcpu_stopped(pcpu_devices + cpu))
489 cpu_relax();
490 }
491}
492
493/*
494 * This is the main routine where commands issued by other
495 * cpus are handled.
496 */
497static void smp_handle_ext_call(void)
498{
499 unsigned long bits;
500
501 /* handle bit signal external calls */
502 bits = xchg(&pcpu_devices[smp_processor_id()].ec_mask, 0);
503 if (test_bit(ec_stop_cpu, &bits))
504 smp_stop_cpu();
505 if (test_bit(ec_schedule, &bits))
506 scheduler_ipi();
507 if (test_bit(ec_call_function_single, &bits))
508 generic_smp_call_function_single_interrupt();
509 if (test_bit(ec_mcck_pending, &bits))
510 __s390_handle_mcck();
511 if (test_bit(ec_irq_work, &bits))
512 irq_work_run();
513}
514
515static void do_ext_call_interrupt(struct ext_code ext_code,
516 unsigned int param32, unsigned long param64)
517{
518 inc_irq_stat(ext_code.code == 0x1202 ? IRQEXT_EXC : IRQEXT_EMS);
519 smp_handle_ext_call();
520}
521
522void arch_send_call_function_ipi_mask(const struct cpumask *mask)
523{
524 int cpu;
525
526 for_each_cpu(cpu, mask)
527 pcpu_ec_call(pcpu_devices + cpu, ec_call_function_single);
528}
529
530void arch_send_call_function_single_ipi(int cpu)
531{
532 pcpu_ec_call(pcpu_devices + cpu, ec_call_function_single);
533}
534
535/*
536 * this function sends a 'reschedule' IPI to another CPU.
537 * it goes straight through and wastes no time serializing
538 * anything. Worst case is that we lose a reschedule ...
539 */
540void smp_send_reschedule(int cpu)
541{
542 pcpu_ec_call(pcpu_devices + cpu, ec_schedule);
543}
544
545#ifdef CONFIG_IRQ_WORK
546void arch_irq_work_raise(void)
547{
548 pcpu_ec_call(pcpu_devices + smp_processor_id(), ec_irq_work);
549}
550#endif
551
552/*
553 * parameter area for the set/clear control bit callbacks
554 */
555struct ec_creg_mask_parms {
556 unsigned long orval;
557 unsigned long andval;
558 int cr;
559};
560
561/*
562 * callback for setting/clearing control bits
563 */
564static void smp_ctl_bit_callback(void *info)
565{
566 struct ec_creg_mask_parms *pp = info;
567 unsigned long cregs[16];
568
569 __ctl_store(cregs, 0, 15);
570 cregs[pp->cr] = (cregs[pp->cr] & pp->andval) | pp->orval;
571 __ctl_load(cregs, 0, 15);
572}
573
574static DEFINE_SPINLOCK(ctl_lock);
575static unsigned long ctlreg;
576
577/*
578 * Set a bit in a control register of all cpus
579 */
580void smp_ctl_set_bit(int cr, int bit)
581{
582 struct ec_creg_mask_parms parms = { 1UL << bit, -1UL, cr };
583
584 spin_lock(&ctl_lock);
585 memcpy_absolute(&ctlreg, &S390_lowcore.cregs_save_area[cr], sizeof(ctlreg));
586 __set_bit(bit, &ctlreg);
587 memcpy_absolute(&S390_lowcore.cregs_save_area[cr], &ctlreg, sizeof(ctlreg));
588 spin_unlock(&ctl_lock);
589 on_each_cpu(smp_ctl_bit_callback, &parms, 1);
590}
591EXPORT_SYMBOL(smp_ctl_set_bit);
592
593/*
594 * Clear a bit in a control register of all cpus
595 */
596void smp_ctl_clear_bit(int cr, int bit)
597{
598 struct ec_creg_mask_parms parms = { 0, ~(1UL << bit), cr };
599
600 spin_lock(&ctl_lock);
601 memcpy_absolute(&ctlreg, &S390_lowcore.cregs_save_area[cr], sizeof(ctlreg));
602 __clear_bit(bit, &ctlreg);
603 memcpy_absolute(&S390_lowcore.cregs_save_area[cr], &ctlreg, sizeof(ctlreg));
604 spin_unlock(&ctl_lock);
605 on_each_cpu(smp_ctl_bit_callback, &parms, 1);
606}
607EXPORT_SYMBOL(smp_ctl_clear_bit);
608
609#ifdef CONFIG_CRASH_DUMP
610
611int smp_store_status(int cpu)
612{
613 struct lowcore *lc;
614 struct pcpu *pcpu;
615 unsigned long pa;
616
617 pcpu = pcpu_devices + cpu;
618 lc = lowcore_ptr[cpu];
619 pa = __pa(&lc->floating_pt_save_area);
620 if (__pcpu_sigp_relax(pcpu->address, SIGP_STORE_STATUS_AT_ADDRESS,
621 pa) != SIGP_CC_ORDER_CODE_ACCEPTED)
622 return -EIO;
623 if (!MACHINE_HAS_VX && !MACHINE_HAS_GS)
624 return 0;
625 pa = __pa(lc->mcesad & MCESA_ORIGIN_MASK);
626 if (MACHINE_HAS_GS)
627 pa |= lc->mcesad & MCESA_LC_MASK;
628 if (__pcpu_sigp_relax(pcpu->address, SIGP_STORE_ADDITIONAL_STATUS,
629 pa) != SIGP_CC_ORDER_CODE_ACCEPTED)
630 return -EIO;
631 return 0;
632}
633
634/*
635 * Collect CPU state of the previous, crashed system.
636 * There are four cases:
637 * 1) standard zfcp/nvme dump
638 * condition: OLDMEM_BASE == NULL && is_ipl_type_dump() == true
639 * The state for all CPUs except the boot CPU needs to be collected
640 * with sigp stop-and-store-status. The boot CPU state is located in
641 * the absolute lowcore of the memory stored in the HSA. The zcore code
642 * will copy the boot CPU state from the HSA.
643 * 2) stand-alone kdump for SCSI/NVMe (zfcp/nvme dump with swapped memory)
644 * condition: OLDMEM_BASE != NULL && is_ipl_type_dump() == true
645 * The state for all CPUs except the boot CPU needs to be collected
646 * with sigp stop-and-store-status. The firmware or the boot-loader
647 * stored the registers of the boot CPU in the absolute lowcore in the
648 * memory of the old system.
649 * 3) kdump and the old kernel did not store the CPU state,
650 * or stand-alone kdump for DASD
651 * condition: OLDMEM_BASE != NULL && !is_kdump_kernel()
652 * The state for all CPUs except the boot CPU needs to be collected
653 * with sigp stop-and-store-status. The kexec code or the boot-loader
654 * stored the registers of the boot CPU in the memory of the old system.
655 * 4) kdump and the old kernel stored the CPU state
656 * condition: OLDMEM_BASE != NULL && is_kdump_kernel()
657 * This case does not exist for s390 anymore, setup_arch explicitly
658 * deactivates the elfcorehdr= kernel parameter
659 */
660static __init void smp_save_cpu_vxrs(struct save_area *sa, u16 addr,
661 bool is_boot_cpu, unsigned long page)
662{
663 __vector128 *vxrs = (__vector128 *) page;
664
665 if (is_boot_cpu)
666 vxrs = boot_cpu_vector_save_area;
667 else
668 __pcpu_sigp_relax(addr, SIGP_STORE_ADDITIONAL_STATUS, page);
669 save_area_add_vxrs(sa, vxrs);
670}
671
672static __init void smp_save_cpu_regs(struct save_area *sa, u16 addr,
673 bool is_boot_cpu, unsigned long page)
674{
675 void *regs = (void *) page;
676
677 if (is_boot_cpu)
678 copy_oldmem_kernel(regs, (void *) __LC_FPREGS_SAVE_AREA, 512);
679 else
680 __pcpu_sigp_relax(addr, SIGP_STORE_STATUS_AT_ADDRESS, page);
681 save_area_add_regs(sa, regs);
682}
683
684void __init smp_save_dump_cpus(void)
685{
686 int addr, boot_cpu_addr, max_cpu_addr;
687 struct save_area *sa;
688 unsigned long page;
689 bool is_boot_cpu;
690
691 if (!(OLDMEM_BASE || is_ipl_type_dump()))
692 /* No previous system present, normal boot. */
693 return;
694 /* Allocate a page as dumping area for the store status sigps */
695 page = memblock_phys_alloc_range(PAGE_SIZE, PAGE_SIZE, 0, 1UL << 31);
696 if (!page)
697 panic("ERROR: Failed to allocate %lx bytes below %lx\n",
698 PAGE_SIZE, 1UL << 31);
699
700 /* Set multi-threading state to the previous system. */
701 pcpu_set_smt(sclp.mtid_prev);
702 boot_cpu_addr = stap();
703 max_cpu_addr = SCLP_MAX_CORES << sclp.mtid_prev;
704 for (addr = 0; addr <= max_cpu_addr; addr++) {
705 if (__pcpu_sigp_relax(addr, SIGP_SENSE, 0) ==
706 SIGP_CC_NOT_OPERATIONAL)
707 continue;
708 is_boot_cpu = (addr == boot_cpu_addr);
709 /* Allocate save area */
710 sa = save_area_alloc(is_boot_cpu);
711 if (!sa)
712 panic("could not allocate memory for save area\n");
713 if (MACHINE_HAS_VX)
714 /* Get the vector registers */
715 smp_save_cpu_vxrs(sa, addr, is_boot_cpu, page);
716 /*
717 * For a zfcp/nvme dump OLDMEM_BASE == NULL and the registers
718 * of the boot CPU are stored in the HSA. To retrieve
719 * these registers an SCLP request is required which is
720 * done by drivers/s390/char/zcore.c:init_cpu_info()
721 */
722 if (!is_boot_cpu || OLDMEM_BASE)
723 /* Get the CPU registers */
724 smp_save_cpu_regs(sa, addr, is_boot_cpu, page);
725 }
726 memblock_free(page, PAGE_SIZE);
727 diag_dma_ops.diag308_reset();
728 pcpu_set_smt(0);
729}
730#endif /* CONFIG_CRASH_DUMP */
731
732void smp_cpu_set_polarization(int cpu, int val)
733{
734 pcpu_devices[cpu].polarization = val;
735}
736
737int smp_cpu_get_polarization(int cpu)
738{
739 return pcpu_devices[cpu].polarization;
740}
741
742int smp_cpu_get_cpu_address(int cpu)
743{
744 return pcpu_devices[cpu].address;
745}
746
747static void __ref smp_get_core_info(struct sclp_core_info *info, int early)
748{
749 static int use_sigp_detection;
750 int address;
751
752 if (use_sigp_detection || sclp_get_core_info(info, early)) {
753 use_sigp_detection = 1;
754 for (address = 0;
755 address < (SCLP_MAX_CORES << smp_cpu_mt_shift);
756 address += (1U << smp_cpu_mt_shift)) {
757 if (__pcpu_sigp_relax(address, SIGP_SENSE, 0) ==
758 SIGP_CC_NOT_OPERATIONAL)
759 continue;
760 info->core[info->configured].core_id =
761 address >> smp_cpu_mt_shift;
762 info->configured++;
763 }
764 info->combined = info->configured;
765 }
766}
767
768static int smp_add_present_cpu(int cpu);
769
770static int smp_add_core(struct sclp_core_entry *core, cpumask_t *avail,
771 bool configured, bool early)
772{
773 struct pcpu *pcpu;
774 int cpu, nr, i;
775 u16 address;
776
777 nr = 0;
778 if (sclp.has_core_type && core->type != boot_core_type)
779 return nr;
780 cpu = cpumask_first(avail);
781 address = core->core_id << smp_cpu_mt_shift;
782 for (i = 0; (i <= smp_cpu_mtid) && (cpu < nr_cpu_ids); i++) {
783 if (pcpu_find_address(cpu_present_mask, address + i))
784 continue;
785 pcpu = pcpu_devices + cpu;
786 pcpu->address = address + i;
787 if (configured)
788 pcpu->state = CPU_STATE_CONFIGURED;
789 else
790 pcpu->state = CPU_STATE_STANDBY;
791 smp_cpu_set_polarization(cpu, POLARIZATION_UNKNOWN);
792 set_cpu_present(cpu, true);
793 if (!early && smp_add_present_cpu(cpu) != 0)
794 set_cpu_present(cpu, false);
795 else
796 nr++;
797 cpumask_clear_cpu(cpu, avail);
798 cpu = cpumask_next(cpu, avail);
799 }
800 return nr;
801}
802
803static int __smp_rescan_cpus(struct sclp_core_info *info, bool early)
804{
805 struct sclp_core_entry *core;
806 static cpumask_t avail;
807 bool configured;
808 u16 core_id;
809 int nr, i;
810
811 get_online_cpus();
812 mutex_lock(&smp_cpu_state_mutex);
813 nr = 0;
814 cpumask_xor(&avail, cpu_possible_mask, cpu_present_mask);
815 /*
816 * Add IPL core first (which got logical CPU number 0) to make sure
817 * that all SMT threads get subsequent logical CPU numbers.
818 */
819 if (early) {
820 core_id = pcpu_devices[0].address >> smp_cpu_mt_shift;
821 for (i = 0; i < info->configured; i++) {
822 core = &info->core[i];
823 if (core->core_id == core_id) {
824 nr += smp_add_core(core, &avail, true, early);
825 break;
826 }
827 }
828 }
829 for (i = 0; i < info->combined; i++) {
830 configured = i < info->configured;
831 nr += smp_add_core(&info->core[i], &avail, configured, early);
832 }
833 mutex_unlock(&smp_cpu_state_mutex);
834 put_online_cpus();
835 return nr;
836}
837
838void __init smp_detect_cpus(void)
839{
840 unsigned int cpu, mtid, c_cpus, s_cpus;
841 struct sclp_core_info *info;
842 u16 address;
843
844 /* Get CPU information */
845 info = memblock_alloc(sizeof(*info), 8);
846 if (!info)
847 panic("%s: Failed to allocate %zu bytes align=0x%x\n",
848 __func__, sizeof(*info), 8);
849 smp_get_core_info(info, 1);
850 /* Find boot CPU type */
851 if (sclp.has_core_type) {
852 address = stap();
853 for (cpu = 0; cpu < info->combined; cpu++)
854 if (info->core[cpu].core_id == address) {
855 /* The boot cpu dictates the cpu type. */
856 boot_core_type = info->core[cpu].type;
857 break;
858 }
859 if (cpu >= info->combined)
860 panic("Could not find boot CPU type");
861 }
862
863 /* Set multi-threading state for the current system */
864 mtid = boot_core_type ? sclp.mtid : sclp.mtid_cp;
865 mtid = (mtid < smp_max_threads) ? mtid : smp_max_threads - 1;
866 pcpu_set_smt(mtid);
867
868 /* Print number of CPUs */
869 c_cpus = s_cpus = 0;
870 for (cpu = 0; cpu < info->combined; cpu++) {
871 if (sclp.has_core_type &&
872 info->core[cpu].type != boot_core_type)
873 continue;
874 if (cpu < info->configured)
875 c_cpus += smp_cpu_mtid + 1;
876 else
877 s_cpus += smp_cpu_mtid + 1;
878 }
879 pr_info("%d configured CPUs, %d standby CPUs\n", c_cpus, s_cpus);
880
881 /* Add CPUs present at boot */
882 __smp_rescan_cpus(info, true);
883 memblock_free_early((unsigned long)info, sizeof(*info));
884}
885
886static void smp_init_secondary(void)
887{
888 int cpu = raw_smp_processor_id();
889
890 S390_lowcore.last_update_clock = get_tod_clock();
891 restore_access_regs(S390_lowcore.access_regs_save_area);
892 cpu_init();
893 rcu_cpu_starting(cpu);
894 init_cpu_timer();
895 vtime_init();
896 vdso_getcpu_init();
897 pfault_init();
898 cpumask_set_cpu(cpu, &cpu_setup_mask);
899 update_cpu_masks();
900 notify_cpu_starting(cpu);
901 if (topology_cpu_dedicated(cpu))
902 set_cpu_flag(CIF_DEDICATED_CPU);
903 else
904 clear_cpu_flag(CIF_DEDICATED_CPU);
905 set_cpu_online(cpu, true);
906 inc_irq_stat(CPU_RST);
907 local_irq_enable();
908 cpu_startup_entry(CPUHP_AP_ONLINE_IDLE);
909}
910
911/*
912 * Activate a secondary processor.
913 */
914static void smp_start_secondary(void *cpuvoid)
915{
916 S390_lowcore.restart_stack = (unsigned long) restart_stack;
917 S390_lowcore.restart_fn = (unsigned long) do_restart;
918 S390_lowcore.restart_data = 0;
919 S390_lowcore.restart_source = -1U;
920 S390_lowcore.restart_flags = 0;
921 call_on_stack_noreturn(smp_init_secondary, S390_lowcore.kernel_stack);
922}
923
924/* Upping and downing of CPUs */
925int __cpu_up(unsigned int cpu, struct task_struct *tidle)
926{
927 struct pcpu *pcpu = pcpu_devices + cpu;
928 int rc;
929
930 if (pcpu->state != CPU_STATE_CONFIGURED)
931 return -EIO;
932 if (pcpu_sigp_retry(pcpu, SIGP_INITIAL_CPU_RESET, 0) !=
933 SIGP_CC_ORDER_CODE_ACCEPTED)
934 return -EIO;
935
936 rc = pcpu_alloc_lowcore(pcpu, cpu);
937 if (rc)
938 return rc;
939 pcpu_prepare_secondary(pcpu, cpu);
940 pcpu_attach_task(pcpu, tidle);
941 pcpu_start_fn(pcpu, smp_start_secondary, NULL);
942 /* Wait until cpu puts itself in the online & active maps */
943 while (!cpu_online(cpu))
944 cpu_relax();
945 return 0;
946}
947
948static unsigned int setup_possible_cpus __initdata;
949
950static int __init _setup_possible_cpus(char *s)
951{
952 get_option(&s, &setup_possible_cpus);
953 return 0;
954}
955early_param("possible_cpus", _setup_possible_cpus);
956
957int __cpu_disable(void)
958{
959 unsigned long cregs[16];
960 int cpu;
961
962 /* Handle possible pending IPIs */
963 smp_handle_ext_call();
964 cpu = smp_processor_id();
965 set_cpu_online(cpu, false);
966 cpumask_clear_cpu(cpu, &cpu_setup_mask);
967 update_cpu_masks();
968 /* Disable pseudo page faults on this cpu. */
969 pfault_fini();
970 /* Disable interrupt sources via control register. */
971 __ctl_store(cregs, 0, 15);
972 cregs[0] &= ~0x0000ee70UL; /* disable all external interrupts */
973 cregs[6] &= ~0xff000000UL; /* disable all I/O interrupts */
974 cregs[14] &= ~0x1f000000UL; /* disable most machine checks */
975 __ctl_load(cregs, 0, 15);
976 clear_cpu_flag(CIF_NOHZ_DELAY);
977 return 0;
978}
979
980void __cpu_die(unsigned int cpu)
981{
982 struct pcpu *pcpu;
983
984 /* Wait until target cpu is down */
985 pcpu = pcpu_devices + cpu;
986 while (!pcpu_stopped(pcpu))
987 cpu_relax();
988 pcpu_free_lowcore(pcpu);
989 cpumask_clear_cpu(cpu, mm_cpumask(&init_mm));
990 cpumask_clear_cpu(cpu, &init_mm.context.cpu_attach_mask);
991}
992
993void __noreturn cpu_die(void)
994{
995 idle_task_exit();
996 __bpon();
997 pcpu_sigp_retry(pcpu_devices + smp_processor_id(), SIGP_STOP, 0);
998 for (;;) ;
999}
1000
1001void __init smp_fill_possible_mask(void)
1002{
1003 unsigned int possible, sclp_max, cpu;
1004
1005 sclp_max = max(sclp.mtid, sclp.mtid_cp) + 1;
1006 sclp_max = min(smp_max_threads, sclp_max);
1007 sclp_max = (sclp.max_cores * sclp_max) ?: nr_cpu_ids;
1008 possible = setup_possible_cpus ?: nr_cpu_ids;
1009 possible = min(possible, sclp_max);
1010 for (cpu = 0; cpu < possible && cpu < nr_cpu_ids; cpu++)
1011 set_cpu_possible(cpu, true);
1012}
1013
1014void __init smp_prepare_cpus(unsigned int max_cpus)
1015{
1016 /* request the 0x1201 emergency signal external interrupt */
1017 if (register_external_irq(EXT_IRQ_EMERGENCY_SIG, do_ext_call_interrupt))
1018 panic("Couldn't request external interrupt 0x1201");
1019 /* request the 0x1202 external call external interrupt */
1020 if (register_external_irq(EXT_IRQ_EXTERNAL_CALL, do_ext_call_interrupt))
1021 panic("Couldn't request external interrupt 0x1202");
1022}
1023
1024void __init smp_prepare_boot_cpu(void)
1025{
1026 struct pcpu *pcpu = pcpu_devices;
1027
1028 WARN_ON(!cpu_present(0) || !cpu_online(0));
1029 pcpu->state = CPU_STATE_CONFIGURED;
1030 S390_lowcore.percpu_offset = __per_cpu_offset[0];
1031 smp_cpu_set_polarization(0, POLARIZATION_UNKNOWN);
1032}
1033
1034void __init smp_setup_processor_id(void)
1035{
1036 pcpu_devices[0].address = stap();
1037 S390_lowcore.cpu_nr = 0;
1038 S390_lowcore.spinlock_lockval = arch_spin_lockval(0);
1039 S390_lowcore.spinlock_index = 0;
1040}
1041
1042/*
1043 * the frequency of the profiling timer can be changed
1044 * by writing a multiplier value into /proc/profile.
1045 *
1046 * usually you want to run this on all CPUs ;)
1047 */
1048int setup_profiling_timer(unsigned int multiplier)
1049{
1050 return 0;
1051}
1052
1053static ssize_t cpu_configure_show(struct device *dev,
1054 struct device_attribute *attr, char *buf)
1055{
1056 ssize_t count;
1057
1058 mutex_lock(&smp_cpu_state_mutex);
1059 count = sprintf(buf, "%d\n", pcpu_devices[dev->id].state);
1060 mutex_unlock(&smp_cpu_state_mutex);
1061 return count;
1062}
1063
1064static ssize_t cpu_configure_store(struct device *dev,
1065 struct device_attribute *attr,
1066 const char *buf, size_t count)
1067{
1068 struct pcpu *pcpu;
1069 int cpu, val, rc, i;
1070 char delim;
1071
1072 if (sscanf(buf, "%d %c", &val, &delim) != 1)
1073 return -EINVAL;
1074 if (val != 0 && val != 1)
1075 return -EINVAL;
1076 get_online_cpus();
1077 mutex_lock(&smp_cpu_state_mutex);
1078 rc = -EBUSY;
1079 /* disallow configuration changes of online cpus and cpu 0 */
1080 cpu = dev->id;
1081 cpu = smp_get_base_cpu(cpu);
1082 if (cpu == 0)
1083 goto out;
1084 for (i = 0; i <= smp_cpu_mtid; i++)
1085 if (cpu_online(cpu + i))
1086 goto out;
1087 pcpu = pcpu_devices + cpu;
1088 rc = 0;
1089 switch (val) {
1090 case 0:
1091 if (pcpu->state != CPU_STATE_CONFIGURED)
1092 break;
1093 rc = sclp_core_deconfigure(pcpu->address >> smp_cpu_mt_shift);
1094 if (rc)
1095 break;
1096 for (i = 0; i <= smp_cpu_mtid; i++) {
1097 if (cpu + i >= nr_cpu_ids || !cpu_present(cpu + i))
1098 continue;
1099 pcpu[i].state = CPU_STATE_STANDBY;
1100 smp_cpu_set_polarization(cpu + i,
1101 POLARIZATION_UNKNOWN);
1102 }
1103 topology_expect_change();
1104 break;
1105 case 1:
1106 if (pcpu->state != CPU_STATE_STANDBY)
1107 break;
1108 rc = sclp_core_configure(pcpu->address >> smp_cpu_mt_shift);
1109 if (rc)
1110 break;
1111 for (i = 0; i <= smp_cpu_mtid; i++) {
1112 if (cpu + i >= nr_cpu_ids || !cpu_present(cpu + i))
1113 continue;
1114 pcpu[i].state = CPU_STATE_CONFIGURED;
1115 smp_cpu_set_polarization(cpu + i,
1116 POLARIZATION_UNKNOWN);
1117 }
1118 topology_expect_change();
1119 break;
1120 default:
1121 break;
1122 }
1123out:
1124 mutex_unlock(&smp_cpu_state_mutex);
1125 put_online_cpus();
1126 return rc ? rc : count;
1127}
1128static DEVICE_ATTR(configure, 0644, cpu_configure_show, cpu_configure_store);
1129
1130static ssize_t show_cpu_address(struct device *dev,
1131 struct device_attribute *attr, char *buf)
1132{
1133 return sprintf(buf, "%d\n", pcpu_devices[dev->id].address);
1134}
1135static DEVICE_ATTR(address, 0444, show_cpu_address, NULL);
1136
1137static struct attribute *cpu_common_attrs[] = {
1138 &dev_attr_configure.attr,
1139 &dev_attr_address.attr,
1140 NULL,
1141};
1142
1143static struct attribute_group cpu_common_attr_group = {
1144 .attrs = cpu_common_attrs,
1145};
1146
1147static struct attribute *cpu_online_attrs[] = {
1148 &dev_attr_idle_count.attr,
1149 &dev_attr_idle_time_us.attr,
1150 NULL,
1151};
1152
1153static struct attribute_group cpu_online_attr_group = {
1154 .attrs = cpu_online_attrs,
1155};
1156
1157static int smp_cpu_online(unsigned int cpu)
1158{
1159 struct device *s = &per_cpu(cpu_device, cpu)->dev;
1160
1161 return sysfs_create_group(&s->kobj, &cpu_online_attr_group);
1162}
1163
1164static int smp_cpu_pre_down(unsigned int cpu)
1165{
1166 struct device *s = &per_cpu(cpu_device, cpu)->dev;
1167
1168 sysfs_remove_group(&s->kobj, &cpu_online_attr_group);
1169 return 0;
1170}
1171
1172static int smp_add_present_cpu(int cpu)
1173{
1174 struct device *s;
1175 struct cpu *c;
1176 int rc;
1177
1178 c = kzalloc(sizeof(*c), GFP_KERNEL);
1179 if (!c)
1180 return -ENOMEM;
1181 per_cpu(cpu_device, cpu) = c;
1182 s = &c->dev;
1183 c->hotpluggable = 1;
1184 rc = register_cpu(c, cpu);
1185 if (rc)
1186 goto out;
1187 rc = sysfs_create_group(&s->kobj, &cpu_common_attr_group);
1188 if (rc)
1189 goto out_cpu;
1190 rc = topology_cpu_init(c);
1191 if (rc)
1192 goto out_topology;
1193 return 0;
1194
1195out_topology:
1196 sysfs_remove_group(&s->kobj, &cpu_common_attr_group);
1197out_cpu:
1198 unregister_cpu(c);
1199out:
1200 return rc;
1201}
1202
1203int __ref smp_rescan_cpus(void)
1204{
1205 struct sclp_core_info *info;
1206 int nr;
1207
1208 info = kzalloc(sizeof(*info), GFP_KERNEL);
1209 if (!info)
1210 return -ENOMEM;
1211 smp_get_core_info(info, 0);
1212 nr = __smp_rescan_cpus(info, false);
1213 kfree(info);
1214 if (nr)
1215 topology_schedule_update();
1216 return 0;
1217}
1218
1219static ssize_t __ref rescan_store(struct device *dev,
1220 struct device_attribute *attr,
1221 const char *buf,
1222 size_t count)
1223{
1224 int rc;
1225
1226 rc = lock_device_hotplug_sysfs();
1227 if (rc)
1228 return rc;
1229 rc = smp_rescan_cpus();
1230 unlock_device_hotplug();
1231 return rc ? rc : count;
1232}
1233static DEVICE_ATTR_WO(rescan);
1234
1235static int __init s390_smp_init(void)
1236{
1237 int cpu, rc = 0;
1238
1239 rc = device_create_file(cpu_subsys.dev_root, &dev_attr_rescan);
1240 if (rc)
1241 return rc;
1242 for_each_present_cpu(cpu) {
1243 rc = smp_add_present_cpu(cpu);
1244 if (rc)
1245 goto out;
1246 }
1247
1248 rc = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "s390/smp:online",
1249 smp_cpu_online, smp_cpu_pre_down);
1250 rc = rc <= 0 ? rc : 0;
1251out:
1252 return rc;
1253}
1254subsys_initcall(s390_smp_init);
1255
1256static __always_inline void set_new_lowcore(struct lowcore *lc)
1257{
1258 union register_pair dst, src;
1259 u32 pfx;
1260
1261 src.even = (unsigned long) &S390_lowcore;
1262 src.odd = sizeof(S390_lowcore);
1263 dst.even = (unsigned long) lc;
1264 dst.odd = sizeof(*lc);
1265 pfx = (unsigned long) lc;
1266
1267 asm volatile(
1268 " mvcl %[dst],%[src]\n"
1269 " spx %[pfx]\n"
1270 : [dst] "+&d" (dst.pair), [src] "+&d" (src.pair)
1271 : [pfx] "Q" (pfx)
1272 : "memory", "cc");
1273}
1274
1275static int __init smp_reinit_ipl_cpu(void)
1276{
1277 unsigned long async_stack, nodat_stack, mcck_stack;
1278 struct lowcore *lc, *lc_ipl;
1279 unsigned long flags;
1280
1281 lc_ipl = lowcore_ptr[0];
1282 lc = (struct lowcore *) __get_free_pages(GFP_KERNEL | GFP_DMA, LC_ORDER);
1283 nodat_stack = __get_free_pages(GFP_KERNEL, THREAD_SIZE_ORDER);
1284 async_stack = stack_alloc();
1285 mcck_stack = stack_alloc();
1286 if (!lc || !nodat_stack || !async_stack || !mcck_stack)
1287 panic("Couldn't allocate memory");
1288
1289 local_irq_save(flags);
1290 local_mcck_disable();
1291 set_new_lowcore(lc);
1292 S390_lowcore.nodat_stack = nodat_stack + STACK_INIT_OFFSET;
1293 S390_lowcore.async_stack = async_stack + STACK_INIT_OFFSET;
1294 S390_lowcore.mcck_stack = mcck_stack + STACK_INIT_OFFSET;
1295 lowcore_ptr[0] = lc;
1296 local_mcck_enable();
1297 local_irq_restore(flags);
1298
1299 free_pages(lc_ipl->async_stack - STACK_INIT_OFFSET, THREAD_SIZE_ORDER);
1300 memblock_free_late(lc_ipl->mcck_stack - STACK_INIT_OFFSET, THREAD_SIZE);
1301 memblock_free_late((unsigned long) lc_ipl, sizeof(*lc_ipl));
1302
1303 return 0;
1304}
1305early_initcall(smp_reinit_ipl_cpu);