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/*
2 * SMP related functions
3 *
4 * Copyright IBM Corp. 1999, 2012
5 * Author(s): Denis Joseph Barrow,
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 * 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/module.h>
23#include <linux/init.h>
24#include <linux/mm.h>
25#include <linux/err.h>
26#include <linux/spinlock.h>
27#include <linux/kernel_stat.h>
28#include <linux/delay.h>
29#include <linux/interrupt.h>
30#include <linux/irqflags.h>
31#include <linux/cpu.h>
32#include <linux/slab.h>
33#include <linux/crash_dump.h>
34#include <linux/memblock.h>
35#include <asm/asm-offsets.h>
36#include <asm/diag.h>
37#include <asm/switch_to.h>
38#include <asm/facility.h>
39#include <asm/ipl.h>
40#include <asm/setup.h>
41#include <asm/irq.h>
42#include <asm/tlbflush.h>
43#include <asm/vtimer.h>
44#include <asm/lowcore.h>
45#include <asm/sclp.h>
46#include <asm/vdso.h>
47#include <asm/debug.h>
48#include <asm/os_info.h>
49#include <asm/sigp.h>
50#include <asm/idle.h>
51#include "entry.h"
52
53enum {
54 ec_schedule = 0,
55 ec_call_function_single,
56 ec_stop_cpu,
57};
58
59enum {
60 CPU_STATE_STANDBY,
61 CPU_STATE_CONFIGURED,
62};
63
64static DEFINE_PER_CPU(struct cpu *, cpu_device);
65
66struct pcpu {
67 struct lowcore *lowcore; /* lowcore page(s) for the cpu */
68 unsigned long ec_mask; /* bit mask for ec_xxx functions */
69 unsigned long ec_clk; /* sigp timestamp for ec_xxx */
70 signed char state; /* physical cpu state */
71 signed char polarization; /* physical polarization */
72 u16 address; /* physical cpu address */
73};
74
75static u8 boot_core_type;
76static struct pcpu pcpu_devices[NR_CPUS];
77
78unsigned int smp_cpu_mt_shift;
79EXPORT_SYMBOL(smp_cpu_mt_shift);
80
81unsigned int smp_cpu_mtid;
82EXPORT_SYMBOL(smp_cpu_mtid);
83
84#ifdef CONFIG_CRASH_DUMP
85__vector128 __initdata boot_cpu_vector_save_area[__NUM_VXRS];
86#endif
87
88static unsigned int smp_max_threads __initdata = -1U;
89
90static int __init early_nosmt(char *s)
91{
92 smp_max_threads = 1;
93 return 0;
94}
95early_param("nosmt", early_nosmt);
96
97static int __init early_smt(char *s)
98{
99 get_option(&s, &smp_max_threads);
100 return 0;
101}
102early_param("smt", early_smt);
103
104/*
105 * The smp_cpu_state_mutex must be held when changing the state or polarization
106 * member of a pcpu data structure within the pcpu_devices arreay.
107 */
108DEFINE_MUTEX(smp_cpu_state_mutex);
109
110/*
111 * Signal processor helper functions.
112 */
113static inline int __pcpu_sigp_relax(u16 addr, u8 order, unsigned long parm)
114{
115 int cc;
116
117 while (1) {
118 cc = __pcpu_sigp(addr, order, parm, NULL);
119 if (cc != SIGP_CC_BUSY)
120 return cc;
121 cpu_relax();
122 }
123}
124
125static int pcpu_sigp_retry(struct pcpu *pcpu, u8 order, u32 parm)
126{
127 int cc, retry;
128
129 for (retry = 0; ; retry++) {
130 cc = __pcpu_sigp(pcpu->address, order, parm, NULL);
131 if (cc != SIGP_CC_BUSY)
132 break;
133 if (retry >= 3)
134 udelay(10);
135 }
136 return cc;
137}
138
139static inline int pcpu_stopped(struct pcpu *pcpu)
140{
141 u32 uninitialized_var(status);
142
143 if (__pcpu_sigp(pcpu->address, SIGP_SENSE,
144 0, &status) != SIGP_CC_STATUS_STORED)
145 return 0;
146 return !!(status & (SIGP_STATUS_CHECK_STOP|SIGP_STATUS_STOPPED));
147}
148
149static inline int pcpu_running(struct pcpu *pcpu)
150{
151 if (__pcpu_sigp(pcpu->address, SIGP_SENSE_RUNNING,
152 0, NULL) != SIGP_CC_STATUS_STORED)
153 return 1;
154 /* Status stored condition code is equivalent to cpu not running. */
155 return 0;
156}
157
158/*
159 * Find struct pcpu by cpu address.
160 */
161static struct pcpu *pcpu_find_address(const struct cpumask *mask, u16 address)
162{
163 int cpu;
164
165 for_each_cpu(cpu, mask)
166 if (pcpu_devices[cpu].address == address)
167 return pcpu_devices + cpu;
168 return NULL;
169}
170
171static void pcpu_ec_call(struct pcpu *pcpu, int ec_bit)
172{
173 int order;
174
175 if (test_and_set_bit(ec_bit, &pcpu->ec_mask))
176 return;
177 order = pcpu_running(pcpu) ? SIGP_EXTERNAL_CALL : SIGP_EMERGENCY_SIGNAL;
178 pcpu->ec_clk = get_tod_clock_fast();
179 pcpu_sigp_retry(pcpu, order, 0);
180}
181
182#define ASYNC_FRAME_OFFSET (ASYNC_SIZE - STACK_FRAME_OVERHEAD - __PT_SIZE)
183#define PANIC_FRAME_OFFSET (PAGE_SIZE - STACK_FRAME_OVERHEAD - __PT_SIZE)
184
185static int pcpu_alloc_lowcore(struct pcpu *pcpu, int cpu)
186{
187 unsigned long async_stack, panic_stack;
188 struct lowcore *lc;
189
190 if (pcpu != &pcpu_devices[0]) {
191 pcpu->lowcore = (struct lowcore *)
192 __get_free_pages(GFP_KERNEL | GFP_DMA, LC_ORDER);
193 async_stack = __get_free_pages(GFP_KERNEL, ASYNC_ORDER);
194 panic_stack = __get_free_page(GFP_KERNEL);
195 if (!pcpu->lowcore || !panic_stack || !async_stack)
196 goto out;
197 } else {
198 async_stack = pcpu->lowcore->async_stack - ASYNC_FRAME_OFFSET;
199 panic_stack = pcpu->lowcore->panic_stack - PANIC_FRAME_OFFSET;
200 }
201 lc = pcpu->lowcore;
202 memcpy(lc, &S390_lowcore, 512);
203 memset((char *) lc + 512, 0, sizeof(*lc) - 512);
204 lc->async_stack = async_stack + ASYNC_FRAME_OFFSET;
205 lc->panic_stack = panic_stack + PANIC_FRAME_OFFSET;
206 lc->cpu_nr = cpu;
207 lc->spinlock_lockval = arch_spin_lockval(cpu);
208 if (MACHINE_HAS_VX)
209 lc->vector_save_area_addr =
210 (unsigned long) &lc->vector_save_area;
211 if (vdso_alloc_per_cpu(lc))
212 goto out;
213 lowcore_ptr[cpu] = lc;
214 pcpu_sigp_retry(pcpu, SIGP_SET_PREFIX, (u32)(unsigned long) lc);
215 return 0;
216out:
217 if (pcpu != &pcpu_devices[0]) {
218 free_page(panic_stack);
219 free_pages(async_stack, ASYNC_ORDER);
220 free_pages((unsigned long) pcpu->lowcore, LC_ORDER);
221 }
222 return -ENOMEM;
223}
224
225#ifdef CONFIG_HOTPLUG_CPU
226
227static void pcpu_free_lowcore(struct pcpu *pcpu)
228{
229 pcpu_sigp_retry(pcpu, SIGP_SET_PREFIX, 0);
230 lowcore_ptr[pcpu - pcpu_devices] = NULL;
231 vdso_free_per_cpu(pcpu->lowcore);
232 if (pcpu == &pcpu_devices[0])
233 return;
234 free_page(pcpu->lowcore->panic_stack-PANIC_FRAME_OFFSET);
235 free_pages(pcpu->lowcore->async_stack-ASYNC_FRAME_OFFSET, ASYNC_ORDER);
236 free_pages((unsigned long) pcpu->lowcore, LC_ORDER);
237}
238
239#endif /* CONFIG_HOTPLUG_CPU */
240
241static void pcpu_prepare_secondary(struct pcpu *pcpu, int cpu)
242{
243 struct lowcore *lc = pcpu->lowcore;
244
245 if (MACHINE_HAS_TLB_LC)
246 cpumask_set_cpu(cpu, &init_mm.context.cpu_attach_mask);
247 cpumask_set_cpu(cpu, mm_cpumask(&init_mm));
248 atomic_inc(&init_mm.context.attach_count);
249 lc->cpu_nr = cpu;
250 lc->spinlock_lockval = arch_spin_lockval(cpu);
251 lc->percpu_offset = __per_cpu_offset[cpu];
252 lc->kernel_asce = S390_lowcore.kernel_asce;
253 lc->machine_flags = S390_lowcore.machine_flags;
254 lc->user_timer = lc->system_timer = lc->steal_timer = 0;
255 __ctl_store(lc->cregs_save_area, 0, 15);
256 save_access_regs((unsigned int *) lc->access_regs_save_area);
257 memcpy(lc->stfle_fac_list, S390_lowcore.stfle_fac_list,
258 MAX_FACILITY_BIT/8);
259}
260
261static void pcpu_attach_task(struct pcpu *pcpu, struct task_struct *tsk)
262{
263 struct lowcore *lc = pcpu->lowcore;
264 struct thread_info *ti = task_thread_info(tsk);
265
266 lc->kernel_stack = (unsigned long) task_stack_page(tsk)
267 + THREAD_SIZE - STACK_FRAME_OVERHEAD - sizeof(struct pt_regs);
268 lc->thread_info = (unsigned long) task_thread_info(tsk);
269 lc->current_task = (unsigned long) tsk;
270 lc->lpp = LPP_MAGIC;
271 lc->current_pid = tsk->pid;
272 lc->user_timer = ti->user_timer;
273 lc->system_timer = ti->system_timer;
274 lc->steal_timer = 0;
275}
276
277static void pcpu_start_fn(struct pcpu *pcpu, void (*func)(void *), void *data)
278{
279 struct lowcore *lc = pcpu->lowcore;
280
281 lc->restart_stack = lc->kernel_stack;
282 lc->restart_fn = (unsigned long) func;
283 lc->restart_data = (unsigned long) data;
284 lc->restart_source = -1UL;
285 pcpu_sigp_retry(pcpu, SIGP_RESTART, 0);
286}
287
288/*
289 * Call function via PSW restart on pcpu and stop the current cpu.
290 */
291static void pcpu_delegate(struct pcpu *pcpu, void (*func)(void *),
292 void *data, unsigned long stack)
293{
294 struct lowcore *lc = lowcore_ptr[pcpu - pcpu_devices];
295 unsigned long source_cpu = stap();
296
297 __load_psw_mask(PSW_KERNEL_BITS);
298 if (pcpu->address == source_cpu)
299 func(data); /* should not return */
300 /* Stop target cpu (if func returns this stops the current cpu). */
301 pcpu_sigp_retry(pcpu, SIGP_STOP, 0);
302 /* Restart func on the target cpu and stop the current cpu. */
303 mem_assign_absolute(lc->restart_stack, stack);
304 mem_assign_absolute(lc->restart_fn, (unsigned long) func);
305 mem_assign_absolute(lc->restart_data, (unsigned long) data);
306 mem_assign_absolute(lc->restart_source, source_cpu);
307 asm volatile(
308 "0: sigp 0,%0,%2 # sigp restart to target cpu\n"
309 " brc 2,0b # busy, try again\n"
310 "1: sigp 0,%1,%3 # sigp stop to current cpu\n"
311 " brc 2,1b # busy, try again\n"
312 : : "d" (pcpu->address), "d" (source_cpu),
313 "K" (SIGP_RESTART), "K" (SIGP_STOP)
314 : "0", "1", "cc");
315 for (;;) ;
316}
317
318/*
319 * Enable additional logical cpus for multi-threading.
320 */
321static int pcpu_set_smt(unsigned int mtid)
322{
323 register unsigned long reg1 asm ("1") = (unsigned long) mtid;
324 int cc;
325
326 if (smp_cpu_mtid == mtid)
327 return 0;
328 asm volatile(
329 " sigp %1,0,%2 # sigp set multi-threading\n"
330 " ipm %0\n"
331 " srl %0,28\n"
332 : "=d" (cc) : "d" (reg1), "K" (SIGP_SET_MULTI_THREADING)
333 : "cc");
334 if (cc == 0) {
335 smp_cpu_mtid = mtid;
336 smp_cpu_mt_shift = 0;
337 while (smp_cpu_mtid >= (1U << smp_cpu_mt_shift))
338 smp_cpu_mt_shift++;
339 pcpu_devices[0].address = stap();
340 }
341 return cc;
342}
343
344/*
345 * Call function on an online CPU.
346 */
347void smp_call_online_cpu(void (*func)(void *), void *data)
348{
349 struct pcpu *pcpu;
350
351 /* Use the current cpu if it is online. */
352 pcpu = pcpu_find_address(cpu_online_mask, stap());
353 if (!pcpu)
354 /* Use the first online cpu. */
355 pcpu = pcpu_devices + cpumask_first(cpu_online_mask);
356 pcpu_delegate(pcpu, func, data, (unsigned long) restart_stack);
357}
358
359/*
360 * Call function on the ipl CPU.
361 */
362void smp_call_ipl_cpu(void (*func)(void *), void *data)
363{
364 pcpu_delegate(&pcpu_devices[0], func, data,
365 pcpu_devices->lowcore->panic_stack -
366 PANIC_FRAME_OFFSET + PAGE_SIZE);
367}
368
369int smp_find_processor_id(u16 address)
370{
371 int cpu;
372
373 for_each_present_cpu(cpu)
374 if (pcpu_devices[cpu].address == address)
375 return cpu;
376 return -1;
377}
378
379int smp_vcpu_scheduled(int cpu)
380{
381 return pcpu_running(pcpu_devices + cpu);
382}
383
384void smp_yield_cpu(int cpu)
385{
386 if (MACHINE_HAS_DIAG9C) {
387 diag_stat_inc_norecursion(DIAG_STAT_X09C);
388 asm volatile("diag %0,0,0x9c"
389 : : "d" (pcpu_devices[cpu].address));
390 } else if (MACHINE_HAS_DIAG44) {
391 diag_stat_inc_norecursion(DIAG_STAT_X044);
392 asm volatile("diag 0,0,0x44");
393 }
394}
395
396/*
397 * Send cpus emergency shutdown signal. This gives the cpus the
398 * opportunity to complete outstanding interrupts.
399 */
400static void smp_emergency_stop(cpumask_t *cpumask)
401{
402 u64 end;
403 int cpu;
404
405 end = get_tod_clock() + (1000000UL << 12);
406 for_each_cpu(cpu, cpumask) {
407 struct pcpu *pcpu = pcpu_devices + cpu;
408 set_bit(ec_stop_cpu, &pcpu->ec_mask);
409 while (__pcpu_sigp(pcpu->address, SIGP_EMERGENCY_SIGNAL,
410 0, NULL) == SIGP_CC_BUSY &&
411 get_tod_clock() < end)
412 cpu_relax();
413 }
414 while (get_tod_clock() < end) {
415 for_each_cpu(cpu, cpumask)
416 if (pcpu_stopped(pcpu_devices + cpu))
417 cpumask_clear_cpu(cpu, cpumask);
418 if (cpumask_empty(cpumask))
419 break;
420 cpu_relax();
421 }
422}
423
424/*
425 * Stop all cpus but the current one.
426 */
427void smp_send_stop(void)
428{
429 cpumask_t cpumask;
430 int cpu;
431
432 /* Disable all interrupts/machine checks */
433 __load_psw_mask(PSW_KERNEL_BITS | PSW_MASK_DAT);
434 trace_hardirqs_off();
435
436 debug_set_critical();
437 cpumask_copy(&cpumask, cpu_online_mask);
438 cpumask_clear_cpu(smp_processor_id(), &cpumask);
439
440 if (oops_in_progress)
441 smp_emergency_stop(&cpumask);
442
443 /* stop all processors */
444 for_each_cpu(cpu, &cpumask) {
445 struct pcpu *pcpu = pcpu_devices + cpu;
446 pcpu_sigp_retry(pcpu, SIGP_STOP, 0);
447 while (!pcpu_stopped(pcpu))
448 cpu_relax();
449 }
450}
451
452/*
453 * This is the main routine where commands issued by other
454 * cpus are handled.
455 */
456static void smp_handle_ext_call(void)
457{
458 unsigned long bits;
459
460 /* handle bit signal external calls */
461 bits = xchg(&pcpu_devices[smp_processor_id()].ec_mask, 0);
462 if (test_bit(ec_stop_cpu, &bits))
463 smp_stop_cpu();
464 if (test_bit(ec_schedule, &bits))
465 scheduler_ipi();
466 if (test_bit(ec_call_function_single, &bits))
467 generic_smp_call_function_single_interrupt();
468}
469
470static void do_ext_call_interrupt(struct ext_code ext_code,
471 unsigned int param32, unsigned long param64)
472{
473 inc_irq_stat(ext_code.code == 0x1202 ? IRQEXT_EXC : IRQEXT_EMS);
474 smp_handle_ext_call();
475}
476
477void arch_send_call_function_ipi_mask(const struct cpumask *mask)
478{
479 int cpu;
480
481 for_each_cpu(cpu, mask)
482 pcpu_ec_call(pcpu_devices + cpu, ec_call_function_single);
483}
484
485void arch_send_call_function_single_ipi(int cpu)
486{
487 pcpu_ec_call(pcpu_devices + cpu, ec_call_function_single);
488}
489
490/*
491 * this function sends a 'reschedule' IPI to another CPU.
492 * it goes straight through and wastes no time serializing
493 * anything. Worst case is that we lose a reschedule ...
494 */
495void smp_send_reschedule(int cpu)
496{
497 pcpu_ec_call(pcpu_devices + cpu, ec_schedule);
498}
499
500/*
501 * parameter area for the set/clear control bit callbacks
502 */
503struct ec_creg_mask_parms {
504 unsigned long orval;
505 unsigned long andval;
506 int cr;
507};
508
509/*
510 * callback for setting/clearing control bits
511 */
512static void smp_ctl_bit_callback(void *info)
513{
514 struct ec_creg_mask_parms *pp = info;
515 unsigned long cregs[16];
516
517 __ctl_store(cregs, 0, 15);
518 cregs[pp->cr] = (cregs[pp->cr] & pp->andval) | pp->orval;
519 __ctl_load(cregs, 0, 15);
520}
521
522/*
523 * Set a bit in a control register of all cpus
524 */
525void smp_ctl_set_bit(int cr, int bit)
526{
527 struct ec_creg_mask_parms parms = { 1UL << bit, -1UL, cr };
528
529 on_each_cpu(smp_ctl_bit_callback, &parms, 1);
530}
531EXPORT_SYMBOL(smp_ctl_set_bit);
532
533/*
534 * Clear a bit in a control register of all cpus
535 */
536void smp_ctl_clear_bit(int cr, int bit)
537{
538 struct ec_creg_mask_parms parms = { 0, ~(1UL << bit), cr };
539
540 on_each_cpu(smp_ctl_bit_callback, &parms, 1);
541}
542EXPORT_SYMBOL(smp_ctl_clear_bit);
543
544#ifdef CONFIG_CRASH_DUMP
545
546int smp_store_status(int cpu)
547{
548 struct pcpu *pcpu = pcpu_devices + cpu;
549 unsigned long pa;
550
551 pa = __pa(&pcpu->lowcore->floating_pt_save_area);
552 if (__pcpu_sigp_relax(pcpu->address, SIGP_STORE_STATUS_AT_ADDRESS,
553 pa) != SIGP_CC_ORDER_CODE_ACCEPTED)
554 return -EIO;
555 if (!MACHINE_HAS_VX)
556 return 0;
557 pa = __pa(pcpu->lowcore->vector_save_area_addr);
558 if (__pcpu_sigp_relax(pcpu->address, SIGP_STORE_ADDITIONAL_STATUS,
559 pa) != SIGP_CC_ORDER_CODE_ACCEPTED)
560 return -EIO;
561 return 0;
562}
563
564/*
565 * Collect CPU state of the previous, crashed system.
566 * There are four cases:
567 * 1) standard zfcp dump
568 * condition: OLDMEM_BASE == NULL && ipl_info.type == IPL_TYPE_FCP_DUMP
569 * The state for all CPUs except the boot CPU needs to be collected
570 * with sigp stop-and-store-status. The boot CPU state is located in
571 * the absolute lowcore of the memory stored in the HSA. The zcore code
572 * will copy the boot CPU state from the HSA.
573 * 2) stand-alone kdump for SCSI (zfcp dump with swapped memory)
574 * condition: OLDMEM_BASE != NULL && ipl_info.type == IPL_TYPE_FCP_DUMP
575 * The state for all CPUs except the boot CPU needs to be collected
576 * with sigp stop-and-store-status. The firmware or the boot-loader
577 * stored the registers of the boot CPU in the absolute lowcore in the
578 * memory of the old system.
579 * 3) kdump and the old kernel did not store the CPU state,
580 * or stand-alone kdump for DASD
581 * condition: OLDMEM_BASE != NULL && !is_kdump_kernel()
582 * The state for all CPUs except the boot CPU needs to be collected
583 * with sigp stop-and-store-status. The kexec code or the boot-loader
584 * stored the registers of the boot CPU in the memory of the old system.
585 * 4) kdump and the old kernel stored the CPU state
586 * condition: OLDMEM_BASE != NULL && is_kdump_kernel()
587 * This case does not exist for s390 anymore, setup_arch explicitly
588 * deactivates the elfcorehdr= kernel parameter
589 */
590static __init void smp_save_cpu_vxrs(struct save_area *sa, u16 addr,
591 bool is_boot_cpu, unsigned long page)
592{
593 __vector128 *vxrs = (__vector128 *) page;
594
595 if (is_boot_cpu)
596 vxrs = boot_cpu_vector_save_area;
597 else
598 __pcpu_sigp_relax(addr, SIGP_STORE_ADDITIONAL_STATUS, page);
599 save_area_add_vxrs(sa, vxrs);
600}
601
602static __init void smp_save_cpu_regs(struct save_area *sa, u16 addr,
603 bool is_boot_cpu, unsigned long page)
604{
605 void *regs = (void *) page;
606
607 if (is_boot_cpu)
608 copy_oldmem_kernel(regs, (void *) __LC_FPREGS_SAVE_AREA, 512);
609 else
610 __pcpu_sigp_relax(addr, SIGP_STORE_STATUS_AT_ADDRESS, page);
611 save_area_add_regs(sa, regs);
612}
613
614void __init smp_save_dump_cpus(void)
615{
616 int addr, boot_cpu_addr, max_cpu_addr;
617 struct save_area *sa;
618 unsigned long page;
619 bool is_boot_cpu;
620
621 if (!(OLDMEM_BASE || ipl_info.type == IPL_TYPE_FCP_DUMP))
622 /* No previous system present, normal boot. */
623 return;
624 /* Allocate a page as dumping area for the store status sigps */
625 page = memblock_alloc_base(PAGE_SIZE, PAGE_SIZE, 1UL << 31);
626 /* Set multi-threading state to the previous system. */
627 pcpu_set_smt(sclp.mtid_prev);
628 boot_cpu_addr = stap();
629 max_cpu_addr = SCLP_MAX_CORES << sclp.mtid_prev;
630 for (addr = 0; addr <= max_cpu_addr; addr++) {
631 if (__pcpu_sigp_relax(addr, SIGP_SENSE, 0) ==
632 SIGP_CC_NOT_OPERATIONAL)
633 continue;
634 is_boot_cpu = (addr == boot_cpu_addr);
635 /* Allocate save area */
636 sa = save_area_alloc(is_boot_cpu);
637 if (!sa)
638 panic("could not allocate memory for save area\n");
639 if (MACHINE_HAS_VX)
640 /* Get the vector registers */
641 smp_save_cpu_vxrs(sa, addr, is_boot_cpu, page);
642 /*
643 * For a zfcp dump OLDMEM_BASE == NULL and the registers
644 * of the boot CPU are stored in the HSA. To retrieve
645 * these registers an SCLP request is required which is
646 * done by drivers/s390/char/zcore.c:init_cpu_info()
647 */
648 if (!is_boot_cpu || OLDMEM_BASE)
649 /* Get the CPU registers */
650 smp_save_cpu_regs(sa, addr, is_boot_cpu, page);
651 }
652 memblock_free(page, PAGE_SIZE);
653 diag308_reset();
654 pcpu_set_smt(0);
655}
656#endif /* CONFIG_CRASH_DUMP */
657
658void smp_cpu_set_polarization(int cpu, int val)
659{
660 pcpu_devices[cpu].polarization = val;
661}
662
663int smp_cpu_get_polarization(int cpu)
664{
665 return pcpu_devices[cpu].polarization;
666}
667
668static struct sclp_core_info *smp_get_core_info(void)
669{
670 static int use_sigp_detection;
671 struct sclp_core_info *info;
672 int address;
673
674 info = kzalloc(sizeof(*info), GFP_KERNEL);
675 if (info && (use_sigp_detection || sclp_get_core_info(info))) {
676 use_sigp_detection = 1;
677 for (address = 0;
678 address < (SCLP_MAX_CORES << smp_cpu_mt_shift);
679 address += (1U << smp_cpu_mt_shift)) {
680 if (__pcpu_sigp_relax(address, SIGP_SENSE, 0) ==
681 SIGP_CC_NOT_OPERATIONAL)
682 continue;
683 info->core[info->configured].core_id =
684 address >> smp_cpu_mt_shift;
685 info->configured++;
686 }
687 info->combined = info->configured;
688 }
689 return info;
690}
691
692static int smp_add_present_cpu(int cpu);
693
694static int __smp_rescan_cpus(struct sclp_core_info *info, int sysfs_add)
695{
696 struct pcpu *pcpu;
697 cpumask_t avail;
698 int cpu, nr, i, j;
699 u16 address;
700
701 nr = 0;
702 cpumask_xor(&avail, cpu_possible_mask, cpu_present_mask);
703 cpu = cpumask_first(&avail);
704 for (i = 0; (i < info->combined) && (cpu < nr_cpu_ids); i++) {
705 if (sclp.has_core_type && info->core[i].type != boot_core_type)
706 continue;
707 address = info->core[i].core_id << smp_cpu_mt_shift;
708 for (j = 0; j <= smp_cpu_mtid; j++) {
709 if (pcpu_find_address(cpu_present_mask, address + j))
710 continue;
711 pcpu = pcpu_devices + cpu;
712 pcpu->address = address + j;
713 pcpu->state =
714 (cpu >= info->configured*(smp_cpu_mtid + 1)) ?
715 CPU_STATE_STANDBY : CPU_STATE_CONFIGURED;
716 smp_cpu_set_polarization(cpu, POLARIZATION_UNKNOWN);
717 set_cpu_present(cpu, true);
718 if (sysfs_add && smp_add_present_cpu(cpu) != 0)
719 set_cpu_present(cpu, false);
720 else
721 nr++;
722 cpu = cpumask_next(cpu, &avail);
723 if (cpu >= nr_cpu_ids)
724 break;
725 }
726 }
727 return nr;
728}
729
730static void __init smp_detect_cpus(void)
731{
732 unsigned int cpu, mtid, c_cpus, s_cpus;
733 struct sclp_core_info *info;
734 u16 address;
735
736 /* Get CPU information */
737 info = smp_get_core_info();
738 if (!info)
739 panic("smp_detect_cpus failed to allocate memory\n");
740
741 /* Find boot CPU type */
742 if (sclp.has_core_type) {
743 address = stap();
744 for (cpu = 0; cpu < info->combined; cpu++)
745 if (info->core[cpu].core_id == address) {
746 /* The boot cpu dictates the cpu type. */
747 boot_core_type = info->core[cpu].type;
748 break;
749 }
750 if (cpu >= info->combined)
751 panic("Could not find boot CPU type");
752 }
753
754 /* Set multi-threading state for the current system */
755 mtid = boot_core_type ? sclp.mtid : sclp.mtid_cp;
756 mtid = (mtid < smp_max_threads) ? mtid : smp_max_threads - 1;
757 pcpu_set_smt(mtid);
758
759 /* Print number of CPUs */
760 c_cpus = s_cpus = 0;
761 for (cpu = 0; cpu < info->combined; cpu++) {
762 if (sclp.has_core_type &&
763 info->core[cpu].type != boot_core_type)
764 continue;
765 if (cpu < info->configured)
766 c_cpus += smp_cpu_mtid + 1;
767 else
768 s_cpus += smp_cpu_mtid + 1;
769 }
770 pr_info("%d configured CPUs, %d standby CPUs\n", c_cpus, s_cpus);
771
772 /* Add CPUs present at boot */
773 get_online_cpus();
774 __smp_rescan_cpus(info, 0);
775 put_online_cpus();
776 kfree(info);
777}
778
779/*
780 * Activate a secondary processor.
781 */
782static void smp_start_secondary(void *cpuvoid)
783{
784 S390_lowcore.last_update_clock = get_tod_clock();
785 S390_lowcore.restart_stack = (unsigned long) restart_stack;
786 S390_lowcore.restart_fn = (unsigned long) do_restart;
787 S390_lowcore.restart_data = 0;
788 S390_lowcore.restart_source = -1UL;
789 restore_access_regs(S390_lowcore.access_regs_save_area);
790 __ctl_load(S390_lowcore.cregs_save_area, 0, 15);
791 __load_psw_mask(PSW_KERNEL_BITS | PSW_MASK_DAT);
792 cpu_init();
793 preempt_disable();
794 init_cpu_timer();
795 vtime_init();
796 pfault_init();
797 notify_cpu_starting(smp_processor_id());
798 set_cpu_online(smp_processor_id(), true);
799 inc_irq_stat(CPU_RST);
800 local_irq_enable();
801 cpu_startup_entry(CPUHP_AP_ONLINE_IDLE);
802}
803
804/* Upping and downing of CPUs */
805int __cpu_up(unsigned int cpu, struct task_struct *tidle)
806{
807 struct pcpu *pcpu;
808 int base, i, rc;
809
810 pcpu = pcpu_devices + cpu;
811 if (pcpu->state != CPU_STATE_CONFIGURED)
812 return -EIO;
813 base = cpu - (cpu % (smp_cpu_mtid + 1));
814 for (i = 0; i <= smp_cpu_mtid; i++) {
815 if (base + i < nr_cpu_ids)
816 if (cpu_online(base + i))
817 break;
818 }
819 /*
820 * If this is the first CPU of the core to get online
821 * do an initial CPU reset.
822 */
823 if (i > smp_cpu_mtid &&
824 pcpu_sigp_retry(pcpu_devices + base, SIGP_INITIAL_CPU_RESET, 0) !=
825 SIGP_CC_ORDER_CODE_ACCEPTED)
826 return -EIO;
827
828 rc = pcpu_alloc_lowcore(pcpu, cpu);
829 if (rc)
830 return rc;
831 pcpu_prepare_secondary(pcpu, cpu);
832 pcpu_attach_task(pcpu, tidle);
833 pcpu_start_fn(pcpu, smp_start_secondary, NULL);
834 /* Wait until cpu puts itself in the online & active maps */
835 while (!cpu_online(cpu) || !cpu_active(cpu))
836 cpu_relax();
837 return 0;
838}
839
840static unsigned int setup_possible_cpus __initdata;
841
842static int __init _setup_possible_cpus(char *s)
843{
844 get_option(&s, &setup_possible_cpus);
845 return 0;
846}
847early_param("possible_cpus", _setup_possible_cpus);
848
849#ifdef CONFIG_HOTPLUG_CPU
850
851int __cpu_disable(void)
852{
853 unsigned long cregs[16];
854
855 /* Handle possible pending IPIs */
856 smp_handle_ext_call();
857 set_cpu_online(smp_processor_id(), false);
858 /* Disable pseudo page faults on this cpu. */
859 pfault_fini();
860 /* Disable interrupt sources via control register. */
861 __ctl_store(cregs, 0, 15);
862 cregs[0] &= ~0x0000ee70UL; /* disable all external interrupts */
863 cregs[6] &= ~0xff000000UL; /* disable all I/O interrupts */
864 cregs[14] &= ~0x1f000000UL; /* disable most machine checks */
865 __ctl_load(cregs, 0, 15);
866 clear_cpu_flag(CIF_NOHZ_DELAY);
867 return 0;
868}
869
870void __cpu_die(unsigned int cpu)
871{
872 struct pcpu *pcpu;
873
874 /* Wait until target cpu is down */
875 pcpu = pcpu_devices + cpu;
876 while (!pcpu_stopped(pcpu))
877 cpu_relax();
878 pcpu_free_lowcore(pcpu);
879 atomic_dec(&init_mm.context.attach_count);
880 cpumask_clear_cpu(cpu, mm_cpumask(&init_mm));
881 if (MACHINE_HAS_TLB_LC)
882 cpumask_clear_cpu(cpu, &init_mm.context.cpu_attach_mask);
883}
884
885void __noreturn cpu_die(void)
886{
887 idle_task_exit();
888 pcpu_sigp_retry(pcpu_devices + smp_processor_id(), SIGP_STOP, 0);
889 for (;;) ;
890}
891
892#endif /* CONFIG_HOTPLUG_CPU */
893
894void __init smp_fill_possible_mask(void)
895{
896 unsigned int possible, sclp_max, cpu;
897
898 sclp_max = max(sclp.mtid, sclp.mtid_cp) + 1;
899 sclp_max = min(smp_max_threads, sclp_max);
900 sclp_max = sclp.max_cores * sclp_max ?: nr_cpu_ids;
901 possible = setup_possible_cpus ?: nr_cpu_ids;
902 possible = min(possible, sclp_max);
903 for (cpu = 0; cpu < possible && cpu < nr_cpu_ids; cpu++)
904 set_cpu_possible(cpu, true);
905}
906
907void __init smp_prepare_cpus(unsigned int max_cpus)
908{
909 /* request the 0x1201 emergency signal external interrupt */
910 if (register_external_irq(EXT_IRQ_EMERGENCY_SIG, do_ext_call_interrupt))
911 panic("Couldn't request external interrupt 0x1201");
912 /* request the 0x1202 external call external interrupt */
913 if (register_external_irq(EXT_IRQ_EXTERNAL_CALL, do_ext_call_interrupt))
914 panic("Couldn't request external interrupt 0x1202");
915 smp_detect_cpus();
916}
917
918void __init smp_prepare_boot_cpu(void)
919{
920 struct pcpu *pcpu = pcpu_devices;
921
922 pcpu->state = CPU_STATE_CONFIGURED;
923 pcpu->address = stap();
924 pcpu->lowcore = (struct lowcore *)(unsigned long) store_prefix();
925 S390_lowcore.percpu_offset = __per_cpu_offset[0];
926 smp_cpu_set_polarization(0, POLARIZATION_UNKNOWN);
927 set_cpu_present(0, true);
928 set_cpu_online(0, true);
929}
930
931void __init smp_cpus_done(unsigned int max_cpus)
932{
933}
934
935void __init smp_setup_processor_id(void)
936{
937 S390_lowcore.cpu_nr = 0;
938 S390_lowcore.spinlock_lockval = arch_spin_lockval(0);
939}
940
941/*
942 * the frequency of the profiling timer can be changed
943 * by writing a multiplier value into /proc/profile.
944 *
945 * usually you want to run this on all CPUs ;)
946 */
947int setup_profiling_timer(unsigned int multiplier)
948{
949 return 0;
950}
951
952#ifdef CONFIG_HOTPLUG_CPU
953static ssize_t cpu_configure_show(struct device *dev,
954 struct device_attribute *attr, char *buf)
955{
956 ssize_t count;
957
958 mutex_lock(&smp_cpu_state_mutex);
959 count = sprintf(buf, "%d\n", pcpu_devices[dev->id].state);
960 mutex_unlock(&smp_cpu_state_mutex);
961 return count;
962}
963
964static ssize_t cpu_configure_store(struct device *dev,
965 struct device_attribute *attr,
966 const char *buf, size_t count)
967{
968 struct pcpu *pcpu;
969 int cpu, val, rc, i;
970 char delim;
971
972 if (sscanf(buf, "%d %c", &val, &delim) != 1)
973 return -EINVAL;
974 if (val != 0 && val != 1)
975 return -EINVAL;
976 get_online_cpus();
977 mutex_lock(&smp_cpu_state_mutex);
978 rc = -EBUSY;
979 /* disallow configuration changes of online cpus and cpu 0 */
980 cpu = dev->id;
981 cpu -= cpu % (smp_cpu_mtid + 1);
982 if (cpu == 0)
983 goto out;
984 for (i = 0; i <= smp_cpu_mtid; i++)
985 if (cpu_online(cpu + i))
986 goto out;
987 pcpu = pcpu_devices + cpu;
988 rc = 0;
989 switch (val) {
990 case 0:
991 if (pcpu->state != CPU_STATE_CONFIGURED)
992 break;
993 rc = sclp_core_deconfigure(pcpu->address >> smp_cpu_mt_shift);
994 if (rc)
995 break;
996 for (i = 0; i <= smp_cpu_mtid; i++) {
997 if (cpu + i >= nr_cpu_ids || !cpu_present(cpu + i))
998 continue;
999 pcpu[i].state = CPU_STATE_STANDBY;
1000 smp_cpu_set_polarization(cpu + i,
1001 POLARIZATION_UNKNOWN);
1002 }
1003 topology_expect_change();
1004 break;
1005 case 1:
1006 if (pcpu->state != CPU_STATE_STANDBY)
1007 break;
1008 rc = sclp_core_configure(pcpu->address >> smp_cpu_mt_shift);
1009 if (rc)
1010 break;
1011 for (i = 0; i <= smp_cpu_mtid; i++) {
1012 if (cpu + i >= nr_cpu_ids || !cpu_present(cpu + i))
1013 continue;
1014 pcpu[i].state = CPU_STATE_CONFIGURED;
1015 smp_cpu_set_polarization(cpu + i,
1016 POLARIZATION_UNKNOWN);
1017 }
1018 topology_expect_change();
1019 break;
1020 default:
1021 break;
1022 }
1023out:
1024 mutex_unlock(&smp_cpu_state_mutex);
1025 put_online_cpus();
1026 return rc ? rc : count;
1027}
1028static DEVICE_ATTR(configure, 0644, cpu_configure_show, cpu_configure_store);
1029#endif /* CONFIG_HOTPLUG_CPU */
1030
1031static ssize_t show_cpu_address(struct device *dev,
1032 struct device_attribute *attr, char *buf)
1033{
1034 return sprintf(buf, "%d\n", pcpu_devices[dev->id].address);
1035}
1036static DEVICE_ATTR(address, 0444, show_cpu_address, NULL);
1037
1038static struct attribute *cpu_common_attrs[] = {
1039#ifdef CONFIG_HOTPLUG_CPU
1040 &dev_attr_configure.attr,
1041#endif
1042 &dev_attr_address.attr,
1043 NULL,
1044};
1045
1046static struct attribute_group cpu_common_attr_group = {
1047 .attrs = cpu_common_attrs,
1048};
1049
1050static struct attribute *cpu_online_attrs[] = {
1051 &dev_attr_idle_count.attr,
1052 &dev_attr_idle_time_us.attr,
1053 NULL,
1054};
1055
1056static struct attribute_group cpu_online_attr_group = {
1057 .attrs = cpu_online_attrs,
1058};
1059
1060static int smp_cpu_notify(struct notifier_block *self, unsigned long action,
1061 void *hcpu)
1062{
1063 unsigned int cpu = (unsigned int)(long)hcpu;
1064 struct device *s = &per_cpu(cpu_device, cpu)->dev;
1065 int err = 0;
1066
1067 switch (action & ~CPU_TASKS_FROZEN) {
1068 case CPU_ONLINE:
1069 err = sysfs_create_group(&s->kobj, &cpu_online_attr_group);
1070 break;
1071 case CPU_DEAD:
1072 sysfs_remove_group(&s->kobj, &cpu_online_attr_group);
1073 break;
1074 }
1075 return notifier_from_errno(err);
1076}
1077
1078static int smp_add_present_cpu(int cpu)
1079{
1080 struct device *s;
1081 struct cpu *c;
1082 int rc;
1083
1084 c = kzalloc(sizeof(*c), GFP_KERNEL);
1085 if (!c)
1086 return -ENOMEM;
1087 per_cpu(cpu_device, cpu) = c;
1088 s = &c->dev;
1089 c->hotpluggable = 1;
1090 rc = register_cpu(c, cpu);
1091 if (rc)
1092 goto out;
1093 rc = sysfs_create_group(&s->kobj, &cpu_common_attr_group);
1094 if (rc)
1095 goto out_cpu;
1096 if (cpu_online(cpu)) {
1097 rc = sysfs_create_group(&s->kobj, &cpu_online_attr_group);
1098 if (rc)
1099 goto out_online;
1100 }
1101 rc = topology_cpu_init(c);
1102 if (rc)
1103 goto out_topology;
1104 return 0;
1105
1106out_topology:
1107 if (cpu_online(cpu))
1108 sysfs_remove_group(&s->kobj, &cpu_online_attr_group);
1109out_online:
1110 sysfs_remove_group(&s->kobj, &cpu_common_attr_group);
1111out_cpu:
1112#ifdef CONFIG_HOTPLUG_CPU
1113 unregister_cpu(c);
1114#endif
1115out:
1116 return rc;
1117}
1118
1119#ifdef CONFIG_HOTPLUG_CPU
1120
1121int __ref smp_rescan_cpus(void)
1122{
1123 struct sclp_core_info *info;
1124 int nr;
1125
1126 info = smp_get_core_info();
1127 if (!info)
1128 return -ENOMEM;
1129 get_online_cpus();
1130 mutex_lock(&smp_cpu_state_mutex);
1131 nr = __smp_rescan_cpus(info, 1);
1132 mutex_unlock(&smp_cpu_state_mutex);
1133 put_online_cpus();
1134 kfree(info);
1135 if (nr)
1136 topology_schedule_update();
1137 return 0;
1138}
1139
1140static ssize_t __ref rescan_store(struct device *dev,
1141 struct device_attribute *attr,
1142 const char *buf,
1143 size_t count)
1144{
1145 int rc;
1146
1147 rc = smp_rescan_cpus();
1148 return rc ? rc : count;
1149}
1150static DEVICE_ATTR(rescan, 0200, NULL, rescan_store);
1151#endif /* CONFIG_HOTPLUG_CPU */
1152
1153static int __init s390_smp_init(void)
1154{
1155 int cpu, rc = 0;
1156
1157#ifdef CONFIG_HOTPLUG_CPU
1158 rc = device_create_file(cpu_subsys.dev_root, &dev_attr_rescan);
1159 if (rc)
1160 return rc;
1161#endif
1162 cpu_notifier_register_begin();
1163 for_each_present_cpu(cpu) {
1164 rc = smp_add_present_cpu(cpu);
1165 if (rc)
1166 goto out;
1167 }
1168
1169 __hotcpu_notifier(smp_cpu_notify, 0);
1170
1171out:
1172 cpu_notifier_register_done();
1173 return rc;
1174}
1175subsys_initcall(s390_smp_init);