Loading...
1/*
2 * arch/s390/kernel/smp.c
3 *
4 * Copyright IBM Corp. 1999, 2009
5 * Author(s): Denis Joseph Barrow (djbarrow@de.ibm.com,barrow_dj@yahoo.com),
6 * Martin Schwidefsky (schwidefsky@de.ibm.com)
7 * Heiko Carstens (heiko.carstens@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 * We work with logical cpu numbering everywhere we can. The only
14 * functions using the real cpu address (got from STAP) are the sigp
15 * functions. For all other functions we use the identity mapping.
16 * That means that cpu_number_map[i] == i for every cpu. cpu_number_map is
17 * used e.g. to find the idle task belonging to a logical cpu. Every array
18 * in the kernel is sorted by the logical cpu number and not by the physical
19 * one which is causing all the confusion with __cpu_logical_map and
20 * cpu_number_map in other architectures.
21 */
22
23#define KMSG_COMPONENT "cpu"
24#define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
25
26#include <linux/workqueue.h>
27#include <linux/module.h>
28#include <linux/init.h>
29#include <linux/mm.h>
30#include <linux/err.h>
31#include <linux/spinlock.h>
32#include <linux/kernel_stat.h>
33#include <linux/delay.h>
34#include <linux/cache.h>
35#include <linux/interrupt.h>
36#include <linux/irqflags.h>
37#include <linux/cpu.h>
38#include <linux/timex.h>
39#include <linux/bootmem.h>
40#include <linux/slab.h>
41#include <asm/asm-offsets.h>
42#include <asm/ipl.h>
43#include <asm/setup.h>
44#include <asm/sigp.h>
45#include <asm/pgalloc.h>
46#include <asm/irq.h>
47#include <asm/cpcmd.h>
48#include <asm/tlbflush.h>
49#include <asm/timer.h>
50#include <asm/lowcore.h>
51#include <asm/sclp.h>
52#include <asm/cputime.h>
53#include <asm/vdso.h>
54#include <asm/cpu.h>
55#include "entry.h"
56
57/* logical cpu to cpu address */
58unsigned short __cpu_logical_map[NR_CPUS];
59
60static struct task_struct *current_set[NR_CPUS];
61
62static u8 smp_cpu_type;
63static int smp_use_sigp_detection;
64
65enum s390_cpu_state {
66 CPU_STATE_STANDBY,
67 CPU_STATE_CONFIGURED,
68};
69
70DEFINE_MUTEX(smp_cpu_state_mutex);
71int smp_cpu_polarization[NR_CPUS];
72static int smp_cpu_state[NR_CPUS];
73static int cpu_management;
74
75static DEFINE_PER_CPU(struct cpu, cpu_devices);
76
77static void smp_ext_bitcall(int, int);
78
79static int raw_cpu_stopped(int cpu)
80{
81 u32 status;
82
83 switch (raw_sigp_ps(&status, 0, cpu, sigp_sense)) {
84 case sigp_status_stored:
85 /* Check for stopped and check stop state */
86 if (status & 0x50)
87 return 1;
88 break;
89 default:
90 break;
91 }
92 return 0;
93}
94
95static inline int cpu_stopped(int cpu)
96{
97 return raw_cpu_stopped(cpu_logical_map(cpu));
98}
99
100void smp_switch_to_ipl_cpu(void (*func)(void *), void *data)
101{
102 struct _lowcore *lc, *current_lc;
103 struct stack_frame *sf;
104 struct pt_regs *regs;
105 unsigned long sp;
106
107 if (smp_processor_id() == 0)
108 func(data);
109 __load_psw_mask(PSW_BASE_BITS | PSW_DEFAULT_KEY);
110 /* Disable lowcore protection */
111 __ctl_clear_bit(0, 28);
112 current_lc = lowcore_ptr[smp_processor_id()];
113 lc = lowcore_ptr[0];
114 if (!lc)
115 lc = current_lc;
116 lc->restart_psw.mask = PSW_BASE_BITS | PSW_DEFAULT_KEY;
117 lc->restart_psw.addr = PSW_ADDR_AMODE | (unsigned long) smp_restart_cpu;
118 if (!cpu_online(0))
119 smp_switch_to_cpu(func, data, 0, stap(), __cpu_logical_map[0]);
120 while (sigp(0, sigp_stop_and_store_status) == sigp_busy)
121 cpu_relax();
122 sp = lc->panic_stack;
123 sp -= sizeof(struct pt_regs);
124 regs = (struct pt_regs *) sp;
125 memcpy(®s->gprs, ¤t_lc->gpregs_save_area, sizeof(regs->gprs));
126 regs->psw = lc->psw_save_area;
127 sp -= STACK_FRAME_OVERHEAD;
128 sf = (struct stack_frame *) sp;
129 sf->back_chain = regs->gprs[15];
130 smp_switch_to_cpu(func, data, sp, stap(), __cpu_logical_map[0]);
131}
132
133void smp_send_stop(void)
134{
135 int cpu, rc;
136
137 /* Disable all interrupts/machine checks */
138 __load_psw_mask(psw_kernel_bits & ~PSW_MASK_MCHECK);
139 trace_hardirqs_off();
140
141 /* stop all processors */
142 for_each_online_cpu(cpu) {
143 if (cpu == smp_processor_id())
144 continue;
145 do {
146 rc = sigp(cpu, sigp_stop);
147 } while (rc == sigp_busy);
148
149 while (!cpu_stopped(cpu))
150 cpu_relax();
151 }
152}
153
154/*
155 * This is the main routine where commands issued by other
156 * cpus are handled.
157 */
158
159static void do_ext_call_interrupt(unsigned int ext_int_code,
160 unsigned int param32, unsigned long param64)
161{
162 unsigned long bits;
163
164 kstat_cpu(smp_processor_id()).irqs[EXTINT_IPI]++;
165 /*
166 * handle bit signal external calls
167 */
168 bits = xchg(&S390_lowcore.ext_call_fast, 0);
169
170 if (test_bit(ec_schedule, &bits))
171 scheduler_ipi();
172
173 if (test_bit(ec_call_function, &bits))
174 generic_smp_call_function_interrupt();
175
176 if (test_bit(ec_call_function_single, &bits))
177 generic_smp_call_function_single_interrupt();
178}
179
180/*
181 * Send an external call sigp to another cpu and return without waiting
182 * for its completion.
183 */
184static void smp_ext_bitcall(int cpu, int sig)
185{
186 /*
187 * Set signaling bit in lowcore of target cpu and kick it
188 */
189 set_bit(sig, (unsigned long *) &lowcore_ptr[cpu]->ext_call_fast);
190 while (sigp(cpu, sigp_emergency_signal) == sigp_busy)
191 udelay(10);
192}
193
194void arch_send_call_function_ipi_mask(const struct cpumask *mask)
195{
196 int cpu;
197
198 for_each_cpu(cpu, mask)
199 smp_ext_bitcall(cpu, ec_call_function);
200}
201
202void arch_send_call_function_single_ipi(int cpu)
203{
204 smp_ext_bitcall(cpu, ec_call_function_single);
205}
206
207#ifndef CONFIG_64BIT
208/*
209 * this function sends a 'purge tlb' signal to another CPU.
210 */
211static void smp_ptlb_callback(void *info)
212{
213 __tlb_flush_local();
214}
215
216void smp_ptlb_all(void)
217{
218 on_each_cpu(smp_ptlb_callback, NULL, 1);
219}
220EXPORT_SYMBOL(smp_ptlb_all);
221#endif /* ! CONFIG_64BIT */
222
223/*
224 * this function sends a 'reschedule' IPI to another CPU.
225 * it goes straight through and wastes no time serializing
226 * anything. Worst case is that we lose a reschedule ...
227 */
228void smp_send_reschedule(int cpu)
229{
230 smp_ext_bitcall(cpu, ec_schedule);
231}
232
233/*
234 * parameter area for the set/clear control bit callbacks
235 */
236struct ec_creg_mask_parms {
237 unsigned long orvals[16];
238 unsigned long andvals[16];
239};
240
241/*
242 * callback for setting/clearing control bits
243 */
244static void smp_ctl_bit_callback(void *info)
245{
246 struct ec_creg_mask_parms *pp = info;
247 unsigned long cregs[16];
248 int i;
249
250 __ctl_store(cregs, 0, 15);
251 for (i = 0; i <= 15; i++)
252 cregs[i] = (cregs[i] & pp->andvals[i]) | pp->orvals[i];
253 __ctl_load(cregs, 0, 15);
254}
255
256/*
257 * Set a bit in a control register of all cpus
258 */
259void smp_ctl_set_bit(int cr, int bit)
260{
261 struct ec_creg_mask_parms parms;
262
263 memset(&parms.orvals, 0, sizeof(parms.orvals));
264 memset(&parms.andvals, 0xff, sizeof(parms.andvals));
265 parms.orvals[cr] = 1UL << bit;
266 on_each_cpu(smp_ctl_bit_callback, &parms, 1);
267}
268EXPORT_SYMBOL(smp_ctl_set_bit);
269
270/*
271 * Clear a bit in a control register of all cpus
272 */
273void smp_ctl_clear_bit(int cr, int bit)
274{
275 struct ec_creg_mask_parms parms;
276
277 memset(&parms.orvals, 0, sizeof(parms.orvals));
278 memset(&parms.andvals, 0xff, sizeof(parms.andvals));
279 parms.andvals[cr] = ~(1UL << bit);
280 on_each_cpu(smp_ctl_bit_callback, &parms, 1);
281}
282EXPORT_SYMBOL(smp_ctl_clear_bit);
283
284#ifdef CONFIG_ZFCPDUMP
285
286static void __init smp_get_save_area(unsigned int cpu, unsigned int phy_cpu)
287{
288 if (ipl_info.type != IPL_TYPE_FCP_DUMP)
289 return;
290 if (cpu >= NR_CPUS) {
291 pr_warning("CPU %i exceeds the maximum %i and is excluded from "
292 "the dump\n", cpu, NR_CPUS - 1);
293 return;
294 }
295 zfcpdump_save_areas[cpu] = kmalloc(sizeof(struct save_area), GFP_KERNEL);
296 while (raw_sigp(phy_cpu, sigp_stop_and_store_status) == sigp_busy)
297 cpu_relax();
298 memcpy_real(zfcpdump_save_areas[cpu],
299 (void *)(unsigned long) store_prefix() + SAVE_AREA_BASE,
300 sizeof(struct save_area));
301}
302
303struct save_area *zfcpdump_save_areas[NR_CPUS + 1];
304EXPORT_SYMBOL_GPL(zfcpdump_save_areas);
305
306#else
307
308static inline void smp_get_save_area(unsigned int cpu, unsigned int phy_cpu) { }
309
310#endif /* CONFIG_ZFCPDUMP */
311
312static int cpu_known(int cpu_id)
313{
314 int cpu;
315
316 for_each_present_cpu(cpu) {
317 if (__cpu_logical_map[cpu] == cpu_id)
318 return 1;
319 }
320 return 0;
321}
322
323static int smp_rescan_cpus_sigp(cpumask_t avail)
324{
325 int cpu_id, logical_cpu;
326
327 logical_cpu = cpumask_first(&avail);
328 if (logical_cpu >= nr_cpu_ids)
329 return 0;
330 for (cpu_id = 0; cpu_id <= MAX_CPU_ADDRESS; cpu_id++) {
331 if (cpu_known(cpu_id))
332 continue;
333 __cpu_logical_map[logical_cpu] = cpu_id;
334 smp_cpu_polarization[logical_cpu] = POLARIZATION_UNKNWN;
335 if (!cpu_stopped(logical_cpu))
336 continue;
337 set_cpu_present(logical_cpu, true);
338 smp_cpu_state[logical_cpu] = CPU_STATE_CONFIGURED;
339 logical_cpu = cpumask_next(logical_cpu, &avail);
340 if (logical_cpu >= nr_cpu_ids)
341 break;
342 }
343 return 0;
344}
345
346static int smp_rescan_cpus_sclp(cpumask_t avail)
347{
348 struct sclp_cpu_info *info;
349 int cpu_id, logical_cpu, cpu;
350 int rc;
351
352 logical_cpu = cpumask_first(&avail);
353 if (logical_cpu >= nr_cpu_ids)
354 return 0;
355 info = kmalloc(sizeof(*info), GFP_KERNEL);
356 if (!info)
357 return -ENOMEM;
358 rc = sclp_get_cpu_info(info);
359 if (rc)
360 goto out;
361 for (cpu = 0; cpu < info->combined; cpu++) {
362 if (info->has_cpu_type && info->cpu[cpu].type != smp_cpu_type)
363 continue;
364 cpu_id = info->cpu[cpu].address;
365 if (cpu_known(cpu_id))
366 continue;
367 __cpu_logical_map[logical_cpu] = cpu_id;
368 smp_cpu_polarization[logical_cpu] = POLARIZATION_UNKNWN;
369 set_cpu_present(logical_cpu, true);
370 if (cpu >= info->configured)
371 smp_cpu_state[logical_cpu] = CPU_STATE_STANDBY;
372 else
373 smp_cpu_state[logical_cpu] = CPU_STATE_CONFIGURED;
374 logical_cpu = cpumask_next(logical_cpu, &avail);
375 if (logical_cpu >= nr_cpu_ids)
376 break;
377 }
378out:
379 kfree(info);
380 return rc;
381}
382
383static int __smp_rescan_cpus(void)
384{
385 cpumask_t avail;
386
387 cpumask_xor(&avail, cpu_possible_mask, cpu_present_mask);
388 if (smp_use_sigp_detection)
389 return smp_rescan_cpus_sigp(avail);
390 else
391 return smp_rescan_cpus_sclp(avail);
392}
393
394static void __init smp_detect_cpus(void)
395{
396 unsigned int cpu, c_cpus, s_cpus;
397 struct sclp_cpu_info *info;
398 u16 boot_cpu_addr, cpu_addr;
399
400 c_cpus = 1;
401 s_cpus = 0;
402 boot_cpu_addr = __cpu_logical_map[0];
403 info = kmalloc(sizeof(*info), GFP_KERNEL);
404 if (!info)
405 panic("smp_detect_cpus failed to allocate memory\n");
406 /* Use sigp detection algorithm if sclp doesn't work. */
407 if (sclp_get_cpu_info(info)) {
408 smp_use_sigp_detection = 1;
409 for (cpu = 0; cpu <= MAX_CPU_ADDRESS; cpu++) {
410 if (cpu == boot_cpu_addr)
411 continue;
412 if (!raw_cpu_stopped(cpu))
413 continue;
414 smp_get_save_area(c_cpus, cpu);
415 c_cpus++;
416 }
417 goto out;
418 }
419
420 if (info->has_cpu_type) {
421 for (cpu = 0; cpu < info->combined; cpu++) {
422 if (info->cpu[cpu].address == boot_cpu_addr) {
423 smp_cpu_type = info->cpu[cpu].type;
424 break;
425 }
426 }
427 }
428
429 for (cpu = 0; cpu < info->combined; cpu++) {
430 if (info->has_cpu_type && info->cpu[cpu].type != smp_cpu_type)
431 continue;
432 cpu_addr = info->cpu[cpu].address;
433 if (cpu_addr == boot_cpu_addr)
434 continue;
435 if (!raw_cpu_stopped(cpu_addr)) {
436 s_cpus++;
437 continue;
438 }
439 smp_get_save_area(c_cpus, cpu_addr);
440 c_cpus++;
441 }
442out:
443 kfree(info);
444 pr_info("%d configured CPUs, %d standby CPUs\n", c_cpus, s_cpus);
445 get_online_cpus();
446 __smp_rescan_cpus();
447 put_online_cpus();
448}
449
450/*
451 * Activate a secondary processor.
452 */
453int __cpuinit start_secondary(void *cpuvoid)
454{
455 cpu_init();
456 preempt_disable();
457 init_cpu_timer();
458 init_cpu_vtimer();
459 pfault_init();
460
461 notify_cpu_starting(smp_processor_id());
462 ipi_call_lock();
463 set_cpu_online(smp_processor_id(), true);
464 ipi_call_unlock();
465 __ctl_clear_bit(0, 28); /* Disable lowcore protection */
466 S390_lowcore.restart_psw.mask = PSW_BASE_BITS | PSW_DEFAULT_KEY;
467 S390_lowcore.restart_psw.addr =
468 PSW_ADDR_AMODE | (unsigned long) psw_restart_int_handler;
469 __ctl_set_bit(0, 28); /* Enable lowcore protection */
470 /*
471 * Wait until the cpu which brought this one up marked it
472 * active before enabling interrupts.
473 */
474 while (!cpumask_test_cpu(smp_processor_id(), cpu_active_mask))
475 cpu_relax();
476 local_irq_enable();
477 /* cpu_idle will call schedule for us */
478 cpu_idle();
479 return 0;
480}
481
482struct create_idle {
483 struct work_struct work;
484 struct task_struct *idle;
485 struct completion done;
486 int cpu;
487};
488
489static void __cpuinit smp_fork_idle(struct work_struct *work)
490{
491 struct create_idle *c_idle;
492
493 c_idle = container_of(work, struct create_idle, work);
494 c_idle->idle = fork_idle(c_idle->cpu);
495 complete(&c_idle->done);
496}
497
498static int __cpuinit smp_alloc_lowcore(int cpu)
499{
500 unsigned long async_stack, panic_stack;
501 struct _lowcore *lowcore;
502
503 lowcore = (void *) __get_free_pages(GFP_KERNEL | GFP_DMA, LC_ORDER);
504 if (!lowcore)
505 return -ENOMEM;
506 async_stack = __get_free_pages(GFP_KERNEL, ASYNC_ORDER);
507 panic_stack = __get_free_page(GFP_KERNEL);
508 if (!panic_stack || !async_stack)
509 goto out;
510 memcpy(lowcore, &S390_lowcore, 512);
511 memset((char *)lowcore + 512, 0, sizeof(*lowcore) - 512);
512 lowcore->async_stack = async_stack + ASYNC_SIZE;
513 lowcore->panic_stack = panic_stack + PAGE_SIZE;
514 lowcore->restart_psw.mask = PSW_BASE_BITS | PSW_DEFAULT_KEY;
515 lowcore->restart_psw.addr =
516 PSW_ADDR_AMODE | (unsigned long) restart_int_handler;
517 if (user_mode != HOME_SPACE_MODE)
518 lowcore->restart_psw.mask |= PSW_ASC_HOME;
519#ifndef CONFIG_64BIT
520 if (MACHINE_HAS_IEEE) {
521 unsigned long save_area;
522
523 save_area = get_zeroed_page(GFP_KERNEL);
524 if (!save_area)
525 goto out;
526 lowcore->extended_save_area_addr = (u32) save_area;
527 }
528#else
529 if (vdso_alloc_per_cpu(cpu, lowcore))
530 goto out;
531#endif
532 lowcore_ptr[cpu] = lowcore;
533 return 0;
534
535out:
536 free_page(panic_stack);
537 free_pages(async_stack, ASYNC_ORDER);
538 free_pages((unsigned long) lowcore, LC_ORDER);
539 return -ENOMEM;
540}
541
542static void smp_free_lowcore(int cpu)
543{
544 struct _lowcore *lowcore;
545
546 lowcore = lowcore_ptr[cpu];
547#ifndef CONFIG_64BIT
548 if (MACHINE_HAS_IEEE)
549 free_page((unsigned long) lowcore->extended_save_area_addr);
550#else
551 vdso_free_per_cpu(cpu, lowcore);
552#endif
553 free_page(lowcore->panic_stack - PAGE_SIZE);
554 free_pages(lowcore->async_stack - ASYNC_SIZE, ASYNC_ORDER);
555 free_pages((unsigned long) lowcore, LC_ORDER);
556 lowcore_ptr[cpu] = NULL;
557}
558
559/* Upping and downing of CPUs */
560int __cpuinit __cpu_up(unsigned int cpu)
561{
562 struct _lowcore *cpu_lowcore;
563 struct create_idle c_idle;
564 struct task_struct *idle;
565 struct stack_frame *sf;
566 u32 lowcore;
567 int ccode;
568
569 if (smp_cpu_state[cpu] != CPU_STATE_CONFIGURED)
570 return -EIO;
571 idle = current_set[cpu];
572 if (!idle) {
573 c_idle.done = COMPLETION_INITIALIZER_ONSTACK(c_idle.done);
574 INIT_WORK_ONSTACK(&c_idle.work, smp_fork_idle);
575 c_idle.cpu = cpu;
576 schedule_work(&c_idle.work);
577 wait_for_completion(&c_idle.done);
578 if (IS_ERR(c_idle.idle))
579 return PTR_ERR(c_idle.idle);
580 idle = c_idle.idle;
581 current_set[cpu] = c_idle.idle;
582 }
583 init_idle(idle, cpu);
584 if (smp_alloc_lowcore(cpu))
585 return -ENOMEM;
586 do {
587 ccode = sigp(cpu, sigp_initial_cpu_reset);
588 if (ccode == sigp_busy)
589 udelay(10);
590 if (ccode == sigp_not_operational)
591 goto err_out;
592 } while (ccode == sigp_busy);
593
594 lowcore = (u32)(unsigned long)lowcore_ptr[cpu];
595 while (sigp_p(lowcore, cpu, sigp_set_prefix) == sigp_busy)
596 udelay(10);
597
598 cpu_lowcore = lowcore_ptr[cpu];
599 cpu_lowcore->kernel_stack = (unsigned long)
600 task_stack_page(idle) + THREAD_SIZE;
601 cpu_lowcore->thread_info = (unsigned long) task_thread_info(idle);
602 sf = (struct stack_frame *) (cpu_lowcore->kernel_stack
603 - sizeof(struct pt_regs)
604 - sizeof(struct stack_frame));
605 memset(sf, 0, sizeof(struct stack_frame));
606 sf->gprs[9] = (unsigned long) sf;
607 cpu_lowcore->save_area[15] = (unsigned long) sf;
608 __ctl_store(cpu_lowcore->cregs_save_area, 0, 15);
609 atomic_inc(&init_mm.context.attach_count);
610 asm volatile(
611 " stam 0,15,0(%0)"
612 : : "a" (&cpu_lowcore->access_regs_save_area) : "memory");
613 cpu_lowcore->percpu_offset = __per_cpu_offset[cpu];
614 cpu_lowcore->current_task = (unsigned long) idle;
615 cpu_lowcore->cpu_nr = cpu;
616 cpu_lowcore->kernel_asce = S390_lowcore.kernel_asce;
617 cpu_lowcore->machine_flags = S390_lowcore.machine_flags;
618 cpu_lowcore->ftrace_func = S390_lowcore.ftrace_func;
619 memcpy(cpu_lowcore->stfle_fac_list, S390_lowcore.stfle_fac_list,
620 MAX_FACILITY_BIT/8);
621 eieio();
622
623 while (sigp(cpu, sigp_restart) == sigp_busy)
624 udelay(10);
625
626 while (!cpu_online(cpu))
627 cpu_relax();
628 return 0;
629
630err_out:
631 smp_free_lowcore(cpu);
632 return -EIO;
633}
634
635static int __init setup_possible_cpus(char *s)
636{
637 int pcpus, cpu;
638
639 pcpus = simple_strtoul(s, NULL, 0);
640 init_cpu_possible(cpumask_of(0));
641 for (cpu = 1; cpu < pcpus && cpu < nr_cpu_ids; cpu++)
642 set_cpu_possible(cpu, true);
643 return 0;
644}
645early_param("possible_cpus", setup_possible_cpus);
646
647#ifdef CONFIG_HOTPLUG_CPU
648
649int __cpu_disable(void)
650{
651 struct ec_creg_mask_parms cr_parms;
652 int cpu = smp_processor_id();
653
654 set_cpu_online(cpu, false);
655
656 /* Disable pfault pseudo page faults on this cpu. */
657 pfault_fini();
658
659 memset(&cr_parms.orvals, 0, sizeof(cr_parms.orvals));
660 memset(&cr_parms.andvals, 0xff, sizeof(cr_parms.andvals));
661
662 /* disable all external interrupts */
663 cr_parms.orvals[0] = 0;
664 cr_parms.andvals[0] = ~(1 << 15 | 1 << 14 | 1 << 13 | 1 << 11 |
665 1 << 10 | 1 << 9 | 1 << 6 | 1 << 5 |
666 1 << 4);
667 /* disable all I/O interrupts */
668 cr_parms.orvals[6] = 0;
669 cr_parms.andvals[6] = ~(1 << 31 | 1 << 30 | 1 << 29 | 1 << 28 |
670 1 << 27 | 1 << 26 | 1 << 25 | 1 << 24);
671 /* disable most machine checks */
672 cr_parms.orvals[14] = 0;
673 cr_parms.andvals[14] = ~(1 << 28 | 1 << 27 | 1 << 26 |
674 1 << 25 | 1 << 24);
675
676 smp_ctl_bit_callback(&cr_parms);
677
678 return 0;
679}
680
681void __cpu_die(unsigned int cpu)
682{
683 /* Wait until target cpu is down */
684 while (!cpu_stopped(cpu))
685 cpu_relax();
686 while (sigp_p(0, cpu, sigp_set_prefix) == sigp_busy)
687 udelay(10);
688 smp_free_lowcore(cpu);
689 atomic_dec(&init_mm.context.attach_count);
690}
691
692void __noreturn cpu_die(void)
693{
694 idle_task_exit();
695 while (sigp(smp_processor_id(), sigp_stop) == sigp_busy)
696 cpu_relax();
697 for (;;);
698}
699
700#endif /* CONFIG_HOTPLUG_CPU */
701
702void __init smp_prepare_cpus(unsigned int max_cpus)
703{
704#ifndef CONFIG_64BIT
705 unsigned long save_area = 0;
706#endif
707 unsigned long async_stack, panic_stack;
708 struct _lowcore *lowcore;
709
710 smp_detect_cpus();
711
712 /* request the 0x1201 emergency signal external interrupt */
713 if (register_external_interrupt(0x1201, do_ext_call_interrupt) != 0)
714 panic("Couldn't request external interrupt 0x1201");
715
716 /* Reallocate current lowcore, but keep its contents. */
717 lowcore = (void *) __get_free_pages(GFP_KERNEL | GFP_DMA, LC_ORDER);
718 panic_stack = __get_free_page(GFP_KERNEL);
719 async_stack = __get_free_pages(GFP_KERNEL, ASYNC_ORDER);
720 BUG_ON(!lowcore || !panic_stack || !async_stack);
721#ifndef CONFIG_64BIT
722 if (MACHINE_HAS_IEEE)
723 save_area = get_zeroed_page(GFP_KERNEL);
724#endif
725 local_irq_disable();
726 local_mcck_disable();
727 lowcore_ptr[smp_processor_id()] = lowcore;
728 *lowcore = S390_lowcore;
729 lowcore->panic_stack = panic_stack + PAGE_SIZE;
730 lowcore->async_stack = async_stack + ASYNC_SIZE;
731#ifndef CONFIG_64BIT
732 if (MACHINE_HAS_IEEE)
733 lowcore->extended_save_area_addr = (u32) save_area;
734#endif
735 set_prefix((u32)(unsigned long) lowcore);
736 local_mcck_enable();
737 local_irq_enable();
738#ifdef CONFIG_64BIT
739 if (vdso_alloc_per_cpu(smp_processor_id(), &S390_lowcore))
740 BUG();
741#endif
742}
743
744void __init smp_prepare_boot_cpu(void)
745{
746 BUG_ON(smp_processor_id() != 0);
747
748 current_thread_info()->cpu = 0;
749 set_cpu_present(0, true);
750 set_cpu_online(0, true);
751 S390_lowcore.percpu_offset = __per_cpu_offset[0];
752 current_set[0] = current;
753 smp_cpu_state[0] = CPU_STATE_CONFIGURED;
754 smp_cpu_polarization[0] = POLARIZATION_UNKNWN;
755}
756
757void __init smp_cpus_done(unsigned int max_cpus)
758{
759}
760
761void __init smp_setup_processor_id(void)
762{
763 S390_lowcore.cpu_nr = 0;
764 __cpu_logical_map[0] = stap();
765}
766
767/*
768 * the frequency of the profiling timer can be changed
769 * by writing a multiplier value into /proc/profile.
770 *
771 * usually you want to run this on all CPUs ;)
772 */
773int setup_profiling_timer(unsigned int multiplier)
774{
775 return 0;
776}
777
778#ifdef CONFIG_HOTPLUG_CPU
779static ssize_t cpu_configure_show(struct sys_device *dev,
780 struct sysdev_attribute *attr, char *buf)
781{
782 ssize_t count;
783
784 mutex_lock(&smp_cpu_state_mutex);
785 count = sprintf(buf, "%d\n", smp_cpu_state[dev->id]);
786 mutex_unlock(&smp_cpu_state_mutex);
787 return count;
788}
789
790static ssize_t cpu_configure_store(struct sys_device *dev,
791 struct sysdev_attribute *attr,
792 const char *buf, size_t count)
793{
794 int cpu = dev->id;
795 int val, rc;
796 char delim;
797
798 if (sscanf(buf, "%d %c", &val, &delim) != 1)
799 return -EINVAL;
800 if (val != 0 && val != 1)
801 return -EINVAL;
802
803 get_online_cpus();
804 mutex_lock(&smp_cpu_state_mutex);
805 rc = -EBUSY;
806 /* disallow configuration changes of online cpus and cpu 0 */
807 if (cpu_online(cpu) || cpu == 0)
808 goto out;
809 rc = 0;
810 switch (val) {
811 case 0:
812 if (smp_cpu_state[cpu] == CPU_STATE_CONFIGURED) {
813 rc = sclp_cpu_deconfigure(__cpu_logical_map[cpu]);
814 if (!rc) {
815 smp_cpu_state[cpu] = CPU_STATE_STANDBY;
816 smp_cpu_polarization[cpu] = POLARIZATION_UNKNWN;
817 }
818 }
819 break;
820 case 1:
821 if (smp_cpu_state[cpu] == CPU_STATE_STANDBY) {
822 rc = sclp_cpu_configure(__cpu_logical_map[cpu]);
823 if (!rc) {
824 smp_cpu_state[cpu] = CPU_STATE_CONFIGURED;
825 smp_cpu_polarization[cpu] = POLARIZATION_UNKNWN;
826 }
827 }
828 break;
829 default:
830 break;
831 }
832out:
833 mutex_unlock(&smp_cpu_state_mutex);
834 put_online_cpus();
835 return rc ? rc : count;
836}
837static SYSDEV_ATTR(configure, 0644, cpu_configure_show, cpu_configure_store);
838#endif /* CONFIG_HOTPLUG_CPU */
839
840static ssize_t cpu_polarization_show(struct sys_device *dev,
841 struct sysdev_attribute *attr, char *buf)
842{
843 int cpu = dev->id;
844 ssize_t count;
845
846 mutex_lock(&smp_cpu_state_mutex);
847 switch (smp_cpu_polarization[cpu]) {
848 case POLARIZATION_HRZ:
849 count = sprintf(buf, "horizontal\n");
850 break;
851 case POLARIZATION_VL:
852 count = sprintf(buf, "vertical:low\n");
853 break;
854 case POLARIZATION_VM:
855 count = sprintf(buf, "vertical:medium\n");
856 break;
857 case POLARIZATION_VH:
858 count = sprintf(buf, "vertical:high\n");
859 break;
860 default:
861 count = sprintf(buf, "unknown\n");
862 break;
863 }
864 mutex_unlock(&smp_cpu_state_mutex);
865 return count;
866}
867static SYSDEV_ATTR(polarization, 0444, cpu_polarization_show, NULL);
868
869static ssize_t show_cpu_address(struct sys_device *dev,
870 struct sysdev_attribute *attr, char *buf)
871{
872 return sprintf(buf, "%d\n", __cpu_logical_map[dev->id]);
873}
874static SYSDEV_ATTR(address, 0444, show_cpu_address, NULL);
875
876
877static struct attribute *cpu_common_attrs[] = {
878#ifdef CONFIG_HOTPLUG_CPU
879 &attr_configure.attr,
880#endif
881 &attr_address.attr,
882 &attr_polarization.attr,
883 NULL,
884};
885
886static struct attribute_group cpu_common_attr_group = {
887 .attrs = cpu_common_attrs,
888};
889
890static ssize_t show_capability(struct sys_device *dev,
891 struct sysdev_attribute *attr, char *buf)
892{
893 unsigned int capability;
894 int rc;
895
896 rc = get_cpu_capability(&capability);
897 if (rc)
898 return rc;
899 return sprintf(buf, "%u\n", capability);
900}
901static SYSDEV_ATTR(capability, 0444, show_capability, NULL);
902
903static ssize_t show_idle_count(struct sys_device *dev,
904 struct sysdev_attribute *attr, char *buf)
905{
906 struct s390_idle_data *idle;
907 unsigned long long idle_count;
908 unsigned int sequence;
909
910 idle = &per_cpu(s390_idle, dev->id);
911repeat:
912 sequence = idle->sequence;
913 smp_rmb();
914 if (sequence & 1)
915 goto repeat;
916 idle_count = idle->idle_count;
917 if (idle->idle_enter)
918 idle_count++;
919 smp_rmb();
920 if (idle->sequence != sequence)
921 goto repeat;
922 return sprintf(buf, "%llu\n", idle_count);
923}
924static SYSDEV_ATTR(idle_count, 0444, show_idle_count, NULL);
925
926static ssize_t show_idle_time(struct sys_device *dev,
927 struct sysdev_attribute *attr, char *buf)
928{
929 struct s390_idle_data *idle;
930 unsigned long long now, idle_time, idle_enter;
931 unsigned int sequence;
932
933 idle = &per_cpu(s390_idle, dev->id);
934 now = get_clock();
935repeat:
936 sequence = idle->sequence;
937 smp_rmb();
938 if (sequence & 1)
939 goto repeat;
940 idle_time = idle->idle_time;
941 idle_enter = idle->idle_enter;
942 if (idle_enter != 0ULL && idle_enter < now)
943 idle_time += now - idle_enter;
944 smp_rmb();
945 if (idle->sequence != sequence)
946 goto repeat;
947 return sprintf(buf, "%llu\n", idle_time >> 12);
948}
949static SYSDEV_ATTR(idle_time_us, 0444, show_idle_time, NULL);
950
951static struct attribute *cpu_online_attrs[] = {
952 &attr_capability.attr,
953 &attr_idle_count.attr,
954 &attr_idle_time_us.attr,
955 NULL,
956};
957
958static struct attribute_group cpu_online_attr_group = {
959 .attrs = cpu_online_attrs,
960};
961
962static int __cpuinit smp_cpu_notify(struct notifier_block *self,
963 unsigned long action, void *hcpu)
964{
965 unsigned int cpu = (unsigned int)(long)hcpu;
966 struct cpu *c = &per_cpu(cpu_devices, cpu);
967 struct sys_device *s = &c->sysdev;
968 struct s390_idle_data *idle;
969 int err = 0;
970
971 switch (action) {
972 case CPU_ONLINE:
973 case CPU_ONLINE_FROZEN:
974 idle = &per_cpu(s390_idle, cpu);
975 memset(idle, 0, sizeof(struct s390_idle_data));
976 err = sysfs_create_group(&s->kobj, &cpu_online_attr_group);
977 break;
978 case CPU_DEAD:
979 case CPU_DEAD_FROZEN:
980 sysfs_remove_group(&s->kobj, &cpu_online_attr_group);
981 break;
982 }
983 return notifier_from_errno(err);
984}
985
986static struct notifier_block __cpuinitdata smp_cpu_nb = {
987 .notifier_call = smp_cpu_notify,
988};
989
990static int __devinit smp_add_present_cpu(int cpu)
991{
992 struct cpu *c = &per_cpu(cpu_devices, cpu);
993 struct sys_device *s = &c->sysdev;
994 int rc;
995
996 c->hotpluggable = 1;
997 rc = register_cpu(c, cpu);
998 if (rc)
999 goto out;
1000 rc = sysfs_create_group(&s->kobj, &cpu_common_attr_group);
1001 if (rc)
1002 goto out_cpu;
1003 if (!cpu_online(cpu))
1004 goto out;
1005 rc = sysfs_create_group(&s->kobj, &cpu_online_attr_group);
1006 if (!rc)
1007 return 0;
1008 sysfs_remove_group(&s->kobj, &cpu_common_attr_group);
1009out_cpu:
1010#ifdef CONFIG_HOTPLUG_CPU
1011 unregister_cpu(c);
1012#endif
1013out:
1014 return rc;
1015}
1016
1017#ifdef CONFIG_HOTPLUG_CPU
1018
1019int __ref smp_rescan_cpus(void)
1020{
1021 cpumask_t newcpus;
1022 int cpu;
1023 int rc;
1024
1025 get_online_cpus();
1026 mutex_lock(&smp_cpu_state_mutex);
1027 cpumask_copy(&newcpus, cpu_present_mask);
1028 rc = __smp_rescan_cpus();
1029 if (rc)
1030 goto out;
1031 cpumask_andnot(&newcpus, cpu_present_mask, &newcpus);
1032 for_each_cpu(cpu, &newcpus) {
1033 rc = smp_add_present_cpu(cpu);
1034 if (rc)
1035 set_cpu_present(cpu, false);
1036 }
1037 rc = 0;
1038out:
1039 mutex_unlock(&smp_cpu_state_mutex);
1040 put_online_cpus();
1041 if (!cpumask_empty(&newcpus))
1042 topology_schedule_update();
1043 return rc;
1044}
1045
1046static ssize_t __ref rescan_store(struct sysdev_class *class,
1047 struct sysdev_class_attribute *attr,
1048 const char *buf,
1049 size_t count)
1050{
1051 int rc;
1052
1053 rc = smp_rescan_cpus();
1054 return rc ? rc : count;
1055}
1056static SYSDEV_CLASS_ATTR(rescan, 0200, NULL, rescan_store);
1057#endif /* CONFIG_HOTPLUG_CPU */
1058
1059static ssize_t dispatching_show(struct sysdev_class *class,
1060 struct sysdev_class_attribute *attr,
1061 char *buf)
1062{
1063 ssize_t count;
1064
1065 mutex_lock(&smp_cpu_state_mutex);
1066 count = sprintf(buf, "%d\n", cpu_management);
1067 mutex_unlock(&smp_cpu_state_mutex);
1068 return count;
1069}
1070
1071static ssize_t dispatching_store(struct sysdev_class *dev,
1072 struct sysdev_class_attribute *attr,
1073 const char *buf,
1074 size_t count)
1075{
1076 int val, rc;
1077 char delim;
1078
1079 if (sscanf(buf, "%d %c", &val, &delim) != 1)
1080 return -EINVAL;
1081 if (val != 0 && val != 1)
1082 return -EINVAL;
1083 rc = 0;
1084 get_online_cpus();
1085 mutex_lock(&smp_cpu_state_mutex);
1086 if (cpu_management == val)
1087 goto out;
1088 rc = topology_set_cpu_management(val);
1089 if (!rc)
1090 cpu_management = val;
1091out:
1092 mutex_unlock(&smp_cpu_state_mutex);
1093 put_online_cpus();
1094 return rc ? rc : count;
1095}
1096static SYSDEV_CLASS_ATTR(dispatching, 0644, dispatching_show,
1097 dispatching_store);
1098
1099static int __init topology_init(void)
1100{
1101 int cpu;
1102 int rc;
1103
1104 register_cpu_notifier(&smp_cpu_nb);
1105
1106#ifdef CONFIG_HOTPLUG_CPU
1107 rc = sysdev_class_create_file(&cpu_sysdev_class, &attr_rescan);
1108 if (rc)
1109 return rc;
1110#endif
1111 rc = sysdev_class_create_file(&cpu_sysdev_class, &attr_dispatching);
1112 if (rc)
1113 return rc;
1114 for_each_present_cpu(cpu) {
1115 rc = smp_add_present_cpu(cpu);
1116 if (rc)
1117 return rc;
1118 }
1119 return 0;
1120}
1121subsys_initcall(topology_init);
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);