Loading...
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * SMP related functions
4 *
5 * Copyright IBM Corp. 1999, 2012
6 * Author(s): Denis Joseph Barrow,
7 * Martin Schwidefsky <schwidefsky@de.ibm.com>,
8 * 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/bootmem.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/cpu.h>
34#include <linux/slab.h>
35#include <linux/sched/hotplug.h>
36#include <linux/sched/task_stack.h>
37#include <linux/crash_dump.h>
38#include <linux/memblock.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/vdso.h>
52#include <asm/debug.h>
53#include <asm/os_info.h>
54#include <asm/sigp.h>
55#include <asm/idle.h>
56#include <asm/nmi.h>
57#include <asm/topology.h>
58#include "entry.h"
59
60enum {
61 ec_schedule = 0,
62 ec_call_function_single,
63 ec_stop_cpu,
64};
65
66enum {
67 CPU_STATE_STANDBY,
68 CPU_STATE_CONFIGURED,
69};
70
71static DEFINE_PER_CPU(struct cpu *, cpu_device);
72
73struct pcpu {
74 struct lowcore *lowcore; /* lowcore page(s) for the cpu */
75 unsigned long ec_mask; /* bit mask for ec_xxx functions */
76 unsigned long ec_clk; /* sigp timestamp for ec_xxx */
77 signed char state; /* physical cpu state */
78 signed char polarization; /* physical polarization */
79 u16 address; /* physical cpu address */
80};
81
82static u8 boot_core_type;
83static struct pcpu pcpu_devices[NR_CPUS];
84
85unsigned int smp_cpu_mt_shift;
86EXPORT_SYMBOL(smp_cpu_mt_shift);
87
88unsigned int smp_cpu_mtid;
89EXPORT_SYMBOL(smp_cpu_mtid);
90
91#ifdef CONFIG_CRASH_DUMP
92__vector128 __initdata boot_cpu_vector_save_area[__NUM_VXRS];
93#endif
94
95static unsigned int smp_max_threads __initdata = -1U;
96
97static int __init early_nosmt(char *s)
98{
99 smp_max_threads = 1;
100 return 0;
101}
102early_param("nosmt", early_nosmt);
103
104static int __init early_smt(char *s)
105{
106 get_option(&s, &smp_max_threads);
107 return 0;
108}
109early_param("smt", early_smt);
110
111/*
112 * The smp_cpu_state_mutex must be held when changing the state or polarization
113 * member of a pcpu data structure within the pcpu_devices arreay.
114 */
115DEFINE_MUTEX(smp_cpu_state_mutex);
116
117/*
118 * Signal processor helper functions.
119 */
120static inline int __pcpu_sigp_relax(u16 addr, u8 order, unsigned long parm)
121{
122 int cc;
123
124 while (1) {
125 cc = __pcpu_sigp(addr, order, parm, NULL);
126 if (cc != SIGP_CC_BUSY)
127 return cc;
128 cpu_relax();
129 }
130}
131
132static int pcpu_sigp_retry(struct pcpu *pcpu, u8 order, u32 parm)
133{
134 int cc, retry;
135
136 for (retry = 0; ; retry++) {
137 cc = __pcpu_sigp(pcpu->address, order, parm, NULL);
138 if (cc != SIGP_CC_BUSY)
139 break;
140 if (retry >= 3)
141 udelay(10);
142 }
143 return cc;
144}
145
146static inline int pcpu_stopped(struct pcpu *pcpu)
147{
148 u32 uninitialized_var(status);
149
150 if (__pcpu_sigp(pcpu->address, SIGP_SENSE,
151 0, &status) != SIGP_CC_STATUS_STORED)
152 return 0;
153 return !!(status & (SIGP_STATUS_CHECK_STOP|SIGP_STATUS_STOPPED));
154}
155
156static inline int pcpu_running(struct pcpu *pcpu)
157{
158 if (__pcpu_sigp(pcpu->address, SIGP_SENSE_RUNNING,
159 0, NULL) != SIGP_CC_STATUS_STORED)
160 return 1;
161 /* Status stored condition code is equivalent to cpu not running. */
162 return 0;
163}
164
165/*
166 * Find struct pcpu by cpu address.
167 */
168static struct pcpu *pcpu_find_address(const struct cpumask *mask, u16 address)
169{
170 int cpu;
171
172 for_each_cpu(cpu, mask)
173 if (pcpu_devices[cpu].address == address)
174 return pcpu_devices + cpu;
175 return NULL;
176}
177
178static void pcpu_ec_call(struct pcpu *pcpu, int ec_bit)
179{
180 int order;
181
182 if (test_and_set_bit(ec_bit, &pcpu->ec_mask))
183 return;
184 order = pcpu_running(pcpu) ? SIGP_EXTERNAL_CALL : SIGP_EMERGENCY_SIGNAL;
185 pcpu->ec_clk = get_tod_clock_fast();
186 pcpu_sigp_retry(pcpu, order, 0);
187}
188
189#define ASYNC_FRAME_OFFSET (ASYNC_SIZE - STACK_FRAME_OVERHEAD - __PT_SIZE)
190#define PANIC_FRAME_OFFSET (PAGE_SIZE - STACK_FRAME_OVERHEAD - __PT_SIZE)
191
192static int pcpu_alloc_lowcore(struct pcpu *pcpu, int cpu)
193{
194 unsigned long async_stack, panic_stack;
195 struct lowcore *lc;
196
197 if (pcpu != &pcpu_devices[0]) {
198 pcpu->lowcore = (struct lowcore *)
199 __get_free_pages(GFP_KERNEL | GFP_DMA, LC_ORDER);
200 async_stack = __get_free_pages(GFP_KERNEL, ASYNC_ORDER);
201 panic_stack = __get_free_page(GFP_KERNEL);
202 if (!pcpu->lowcore || !panic_stack || !async_stack)
203 goto out;
204 } else {
205 async_stack = pcpu->lowcore->async_stack - ASYNC_FRAME_OFFSET;
206 panic_stack = pcpu->lowcore->panic_stack - PANIC_FRAME_OFFSET;
207 }
208 lc = pcpu->lowcore;
209 memcpy(lc, &S390_lowcore, 512);
210 memset((char *) lc + 512, 0, sizeof(*lc) - 512);
211 lc->async_stack = async_stack + ASYNC_FRAME_OFFSET;
212 lc->panic_stack = panic_stack + PANIC_FRAME_OFFSET;
213 lc->cpu_nr = cpu;
214 lc->spinlock_lockval = arch_spin_lockval(cpu);
215 lc->spinlock_index = 0;
216 lc->br_r1_trampoline = 0x07f1; /* br %r1 */
217 if (nmi_alloc_per_cpu(lc))
218 goto out;
219 if (vdso_alloc_per_cpu(lc))
220 goto out_mcesa;
221 lowcore_ptr[cpu] = lc;
222 pcpu_sigp_retry(pcpu, SIGP_SET_PREFIX, (u32)(unsigned long) lc);
223 return 0;
224
225out_mcesa:
226 nmi_free_per_cpu(lc);
227out:
228 if (pcpu != &pcpu_devices[0]) {
229 free_page(panic_stack);
230 free_pages(async_stack, ASYNC_ORDER);
231 free_pages((unsigned long) pcpu->lowcore, LC_ORDER);
232 }
233 return -ENOMEM;
234}
235
236#ifdef CONFIG_HOTPLUG_CPU
237
238static void pcpu_free_lowcore(struct pcpu *pcpu)
239{
240 pcpu_sigp_retry(pcpu, SIGP_SET_PREFIX, 0);
241 lowcore_ptr[pcpu - pcpu_devices] = NULL;
242 vdso_free_per_cpu(pcpu->lowcore);
243 nmi_free_per_cpu(pcpu->lowcore);
244 if (pcpu == &pcpu_devices[0])
245 return;
246 free_page(pcpu->lowcore->panic_stack-PANIC_FRAME_OFFSET);
247 free_pages(pcpu->lowcore->async_stack-ASYNC_FRAME_OFFSET, ASYNC_ORDER);
248 free_pages((unsigned long) pcpu->lowcore, LC_ORDER);
249}
250
251#endif /* CONFIG_HOTPLUG_CPU */
252
253static void pcpu_prepare_secondary(struct pcpu *pcpu, int cpu)
254{
255 struct lowcore *lc = pcpu->lowcore;
256
257 cpumask_set_cpu(cpu, &init_mm.context.cpu_attach_mask);
258 cpumask_set_cpu(cpu, mm_cpumask(&init_mm));
259 lc->cpu_nr = cpu;
260 lc->spinlock_lockval = arch_spin_lockval(cpu);
261 lc->spinlock_index = 0;
262 lc->percpu_offset = __per_cpu_offset[cpu];
263 lc->kernel_asce = S390_lowcore.kernel_asce;
264 lc->machine_flags = S390_lowcore.machine_flags;
265 lc->user_timer = lc->system_timer = lc->steal_timer = 0;
266 __ctl_store(lc->cregs_save_area, 0, 15);
267 save_access_regs((unsigned int *) lc->access_regs_save_area);
268 memcpy(lc->stfle_fac_list, S390_lowcore.stfle_fac_list,
269 sizeof(lc->stfle_fac_list));
270 memcpy(lc->alt_stfle_fac_list, S390_lowcore.alt_stfle_fac_list,
271 sizeof(lc->alt_stfle_fac_list));
272 arch_spin_lock_setup(cpu);
273}
274
275static void pcpu_attach_task(struct pcpu *pcpu, struct task_struct *tsk)
276{
277 struct lowcore *lc = pcpu->lowcore;
278
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 = pcpu->lowcore;
295
296 lc->restart_stack = lc->kernel_stack;
297 lc->restart_fn = (unsigned long) func;
298 lc->restart_data = (unsigned long) data;
299 lc->restart_source = -1UL;
300 pcpu_sigp_retry(pcpu, SIGP_RESTART, 0);
301}
302
303/*
304 * Call function via PSW restart on pcpu and stop the current cpu.
305 */
306static void pcpu_delegate(struct pcpu *pcpu, void (*func)(void *),
307 void *data, unsigned long stack)
308{
309 struct lowcore *lc = lowcore_ptr[pcpu - pcpu_devices];
310 unsigned long source_cpu = stap();
311
312 __load_psw_mask(PSW_KERNEL_BITS);
313 if (pcpu->address == source_cpu)
314 func(data); /* should not return */
315 /* Stop target cpu (if func returns this stops the current cpu). */
316 pcpu_sigp_retry(pcpu, SIGP_STOP, 0);
317 /* Restart func on the target cpu and stop the current cpu. */
318 mem_assign_absolute(lc->restart_stack, stack);
319 mem_assign_absolute(lc->restart_fn, (unsigned long) func);
320 mem_assign_absolute(lc->restart_data, (unsigned long) data);
321 mem_assign_absolute(lc->restart_source, source_cpu);
322 __bpon();
323 asm volatile(
324 "0: sigp 0,%0,%2 # sigp restart to target cpu\n"
325 " brc 2,0b # busy, try again\n"
326 "1: sigp 0,%1,%3 # sigp stop to current cpu\n"
327 " brc 2,1b # busy, try again\n"
328 : : "d" (pcpu->address), "d" (source_cpu),
329 "K" (SIGP_RESTART), "K" (SIGP_STOP)
330 : "0", "1", "cc");
331 for (;;) ;
332}
333
334/*
335 * Enable additional logical cpus for multi-threading.
336 */
337static int pcpu_set_smt(unsigned int mtid)
338{
339 int cc;
340
341 if (smp_cpu_mtid == mtid)
342 return 0;
343 cc = __pcpu_sigp(0, SIGP_SET_MULTI_THREADING, mtid, NULL);
344 if (cc == 0) {
345 smp_cpu_mtid = mtid;
346 smp_cpu_mt_shift = 0;
347 while (smp_cpu_mtid >= (1U << smp_cpu_mt_shift))
348 smp_cpu_mt_shift++;
349 pcpu_devices[0].address = stap();
350 }
351 return cc;
352}
353
354/*
355 * Call function on an online CPU.
356 */
357void smp_call_online_cpu(void (*func)(void *), void *data)
358{
359 struct pcpu *pcpu;
360
361 /* Use the current cpu if it is online. */
362 pcpu = pcpu_find_address(cpu_online_mask, stap());
363 if (!pcpu)
364 /* Use the first online cpu. */
365 pcpu = pcpu_devices + cpumask_first(cpu_online_mask);
366 pcpu_delegate(pcpu, func, data, (unsigned long) restart_stack);
367}
368
369/*
370 * Call function on the ipl CPU.
371 */
372void smp_call_ipl_cpu(void (*func)(void *), void *data)
373{
374 pcpu_delegate(&pcpu_devices[0], func, data,
375 pcpu_devices->lowcore->panic_stack -
376 PANIC_FRAME_OFFSET + PAGE_SIZE);
377}
378
379int smp_find_processor_id(u16 address)
380{
381 int cpu;
382
383 for_each_present_cpu(cpu)
384 if (pcpu_devices[cpu].address == address)
385 return cpu;
386 return -1;
387}
388
389bool arch_vcpu_is_preempted(int cpu)
390{
391 if (test_cpu_flag_of(CIF_ENABLED_WAIT, cpu))
392 return false;
393 if (pcpu_running(pcpu_devices + cpu))
394 return false;
395 return true;
396}
397EXPORT_SYMBOL(arch_vcpu_is_preempted);
398
399void smp_yield_cpu(int cpu)
400{
401 if (MACHINE_HAS_DIAG9C) {
402 diag_stat_inc_norecursion(DIAG_STAT_X09C);
403 asm volatile("diag %0,0,0x9c"
404 : : "d" (pcpu_devices[cpu].address));
405 } else if (MACHINE_HAS_DIAG44) {
406 diag_stat_inc_norecursion(DIAG_STAT_X044);
407 asm volatile("diag 0,0,0x44");
408 }
409}
410
411/*
412 * Send cpus emergency shutdown signal. This gives the cpus the
413 * opportunity to complete outstanding interrupts.
414 */
415void notrace smp_emergency_stop(void)
416{
417 cpumask_t cpumask;
418 u64 end;
419 int cpu;
420
421 cpumask_copy(&cpumask, cpu_online_mask);
422 cpumask_clear_cpu(smp_processor_id(), &cpumask);
423
424 end = get_tod_clock() + (1000000UL << 12);
425 for_each_cpu(cpu, &cpumask) {
426 struct pcpu *pcpu = pcpu_devices + cpu;
427 set_bit(ec_stop_cpu, &pcpu->ec_mask);
428 while (__pcpu_sigp(pcpu->address, SIGP_EMERGENCY_SIGNAL,
429 0, NULL) == SIGP_CC_BUSY &&
430 get_tod_clock() < end)
431 cpu_relax();
432 }
433 while (get_tod_clock() < end) {
434 for_each_cpu(cpu, &cpumask)
435 if (pcpu_stopped(pcpu_devices + cpu))
436 cpumask_clear_cpu(cpu, &cpumask);
437 if (cpumask_empty(&cpumask))
438 break;
439 cpu_relax();
440 }
441}
442NOKPROBE_SYMBOL(smp_emergency_stop);
443
444/*
445 * Stop all cpus but the current one.
446 */
447void smp_send_stop(void)
448{
449 int cpu;
450
451 /* Disable all interrupts/machine checks */
452 __load_psw_mask(PSW_KERNEL_BITS | PSW_MASK_DAT);
453 trace_hardirqs_off();
454
455 debug_set_critical();
456
457 if (oops_in_progress)
458 smp_emergency_stop();
459
460 /* stop all processors */
461 for_each_online_cpu(cpu) {
462 if (cpu == smp_processor_id())
463 continue;
464 pcpu_sigp_retry(pcpu_devices + cpu, SIGP_STOP, 0);
465 while (!pcpu_stopped(pcpu_devices + cpu))
466 cpu_relax();
467 }
468}
469
470/*
471 * This is the main routine where commands issued by other
472 * cpus are handled.
473 */
474static void smp_handle_ext_call(void)
475{
476 unsigned long bits;
477
478 /* handle bit signal external calls */
479 bits = xchg(&pcpu_devices[smp_processor_id()].ec_mask, 0);
480 if (test_bit(ec_stop_cpu, &bits))
481 smp_stop_cpu();
482 if (test_bit(ec_schedule, &bits))
483 scheduler_ipi();
484 if (test_bit(ec_call_function_single, &bits))
485 generic_smp_call_function_single_interrupt();
486}
487
488static void do_ext_call_interrupt(struct ext_code ext_code,
489 unsigned int param32, unsigned long param64)
490{
491 inc_irq_stat(ext_code.code == 0x1202 ? IRQEXT_EXC : IRQEXT_EMS);
492 smp_handle_ext_call();
493}
494
495void arch_send_call_function_ipi_mask(const struct cpumask *mask)
496{
497 int cpu;
498
499 for_each_cpu(cpu, mask)
500 pcpu_ec_call(pcpu_devices + cpu, ec_call_function_single);
501}
502
503void arch_send_call_function_single_ipi(int cpu)
504{
505 pcpu_ec_call(pcpu_devices + cpu, ec_call_function_single);
506}
507
508/*
509 * this function sends a 'reschedule' IPI to another CPU.
510 * it goes straight through and wastes no time serializing
511 * anything. Worst case is that we lose a reschedule ...
512 */
513void smp_send_reschedule(int cpu)
514{
515 pcpu_ec_call(pcpu_devices + cpu, ec_schedule);
516}
517
518/*
519 * parameter area for the set/clear control bit callbacks
520 */
521struct ec_creg_mask_parms {
522 unsigned long orval;
523 unsigned long andval;
524 int cr;
525};
526
527/*
528 * callback for setting/clearing control bits
529 */
530static void smp_ctl_bit_callback(void *info)
531{
532 struct ec_creg_mask_parms *pp = info;
533 unsigned long cregs[16];
534
535 __ctl_store(cregs, 0, 15);
536 cregs[pp->cr] = (cregs[pp->cr] & pp->andval) | pp->orval;
537 __ctl_load(cregs, 0, 15);
538}
539
540/*
541 * Set a bit in a control register of all cpus
542 */
543void smp_ctl_set_bit(int cr, int bit)
544{
545 struct ec_creg_mask_parms parms = { 1UL << bit, -1UL, cr };
546
547 on_each_cpu(smp_ctl_bit_callback, &parms, 1);
548}
549EXPORT_SYMBOL(smp_ctl_set_bit);
550
551/*
552 * Clear a bit in a control register of all cpus
553 */
554void smp_ctl_clear_bit(int cr, int bit)
555{
556 struct ec_creg_mask_parms parms = { 0, ~(1UL << bit), cr };
557
558 on_each_cpu(smp_ctl_bit_callback, &parms, 1);
559}
560EXPORT_SYMBOL(smp_ctl_clear_bit);
561
562#ifdef CONFIG_CRASH_DUMP
563
564int smp_store_status(int cpu)
565{
566 struct pcpu *pcpu = pcpu_devices + cpu;
567 unsigned long pa;
568
569 pa = __pa(&pcpu->lowcore->floating_pt_save_area);
570 if (__pcpu_sigp_relax(pcpu->address, SIGP_STORE_STATUS_AT_ADDRESS,
571 pa) != SIGP_CC_ORDER_CODE_ACCEPTED)
572 return -EIO;
573 if (!MACHINE_HAS_VX && !MACHINE_HAS_GS)
574 return 0;
575 pa = __pa(pcpu->lowcore->mcesad & MCESA_ORIGIN_MASK);
576 if (MACHINE_HAS_GS)
577 pa |= pcpu->lowcore->mcesad & MCESA_LC_MASK;
578 if (__pcpu_sigp_relax(pcpu->address, SIGP_STORE_ADDITIONAL_STATUS,
579 pa) != SIGP_CC_ORDER_CODE_ACCEPTED)
580 return -EIO;
581 return 0;
582}
583
584/*
585 * Collect CPU state of the previous, crashed system.
586 * There are four cases:
587 * 1) standard zfcp dump
588 * condition: OLDMEM_BASE == NULL && ipl_info.type == IPL_TYPE_FCP_DUMP
589 * The state for all CPUs except the boot CPU needs to be collected
590 * with sigp stop-and-store-status. The boot CPU state is located in
591 * the absolute lowcore of the memory stored in the HSA. The zcore code
592 * will copy the boot CPU state from the HSA.
593 * 2) stand-alone kdump for SCSI (zfcp dump with swapped memory)
594 * condition: OLDMEM_BASE != NULL && ipl_info.type == IPL_TYPE_FCP_DUMP
595 * The state for all CPUs except the boot CPU needs to be collected
596 * with sigp stop-and-store-status. The firmware or the boot-loader
597 * stored the registers of the boot CPU in the absolute lowcore in the
598 * memory of the old system.
599 * 3) kdump and the old kernel did not store the CPU state,
600 * or stand-alone kdump for DASD
601 * condition: OLDMEM_BASE != NULL && !is_kdump_kernel()
602 * The state for all CPUs except the boot CPU needs to be collected
603 * with sigp stop-and-store-status. The kexec code or the boot-loader
604 * stored the registers of the boot CPU in the memory of the old system.
605 * 4) kdump and the old kernel stored the CPU state
606 * condition: OLDMEM_BASE != NULL && is_kdump_kernel()
607 * This case does not exist for s390 anymore, setup_arch explicitly
608 * deactivates the elfcorehdr= kernel parameter
609 */
610static __init void smp_save_cpu_vxrs(struct save_area *sa, u16 addr,
611 bool is_boot_cpu, unsigned long page)
612{
613 __vector128 *vxrs = (__vector128 *) page;
614
615 if (is_boot_cpu)
616 vxrs = boot_cpu_vector_save_area;
617 else
618 __pcpu_sigp_relax(addr, SIGP_STORE_ADDITIONAL_STATUS, page);
619 save_area_add_vxrs(sa, vxrs);
620}
621
622static __init void smp_save_cpu_regs(struct save_area *sa, u16 addr,
623 bool is_boot_cpu, unsigned long page)
624{
625 void *regs = (void *) page;
626
627 if (is_boot_cpu)
628 copy_oldmem_kernel(regs, (void *) __LC_FPREGS_SAVE_AREA, 512);
629 else
630 __pcpu_sigp_relax(addr, SIGP_STORE_STATUS_AT_ADDRESS, page);
631 save_area_add_regs(sa, regs);
632}
633
634void __init smp_save_dump_cpus(void)
635{
636 int addr, boot_cpu_addr, max_cpu_addr;
637 struct save_area *sa;
638 unsigned long page;
639 bool is_boot_cpu;
640
641 if (!(OLDMEM_BASE || ipl_info.type == IPL_TYPE_FCP_DUMP))
642 /* No previous system present, normal boot. */
643 return;
644 /* Allocate a page as dumping area for the store status sigps */
645 page = memblock_alloc_base(PAGE_SIZE, PAGE_SIZE, 1UL << 31);
646 /* Set multi-threading state to the previous system. */
647 pcpu_set_smt(sclp.mtid_prev);
648 boot_cpu_addr = stap();
649 max_cpu_addr = SCLP_MAX_CORES << sclp.mtid_prev;
650 for (addr = 0; addr <= max_cpu_addr; addr++) {
651 if (__pcpu_sigp_relax(addr, SIGP_SENSE, 0) ==
652 SIGP_CC_NOT_OPERATIONAL)
653 continue;
654 is_boot_cpu = (addr == boot_cpu_addr);
655 /* Allocate save area */
656 sa = save_area_alloc(is_boot_cpu);
657 if (!sa)
658 panic("could not allocate memory for save area\n");
659 if (MACHINE_HAS_VX)
660 /* Get the vector registers */
661 smp_save_cpu_vxrs(sa, addr, is_boot_cpu, page);
662 /*
663 * For a zfcp dump OLDMEM_BASE == NULL and the registers
664 * of the boot CPU are stored in the HSA. To retrieve
665 * these registers an SCLP request is required which is
666 * done by drivers/s390/char/zcore.c:init_cpu_info()
667 */
668 if (!is_boot_cpu || OLDMEM_BASE)
669 /* Get the CPU registers */
670 smp_save_cpu_regs(sa, addr, is_boot_cpu, page);
671 }
672 memblock_free(page, PAGE_SIZE);
673 diag308_reset();
674 pcpu_set_smt(0);
675}
676#endif /* CONFIG_CRASH_DUMP */
677
678void smp_cpu_set_polarization(int cpu, int val)
679{
680 pcpu_devices[cpu].polarization = val;
681}
682
683int smp_cpu_get_polarization(int cpu)
684{
685 return pcpu_devices[cpu].polarization;
686}
687
688static void __ref smp_get_core_info(struct sclp_core_info *info, int early)
689{
690 static int use_sigp_detection;
691 int address;
692
693 if (use_sigp_detection || sclp_get_core_info(info, early)) {
694 use_sigp_detection = 1;
695 for (address = 0;
696 address < (SCLP_MAX_CORES << smp_cpu_mt_shift);
697 address += (1U << smp_cpu_mt_shift)) {
698 if (__pcpu_sigp_relax(address, SIGP_SENSE, 0) ==
699 SIGP_CC_NOT_OPERATIONAL)
700 continue;
701 info->core[info->configured].core_id =
702 address >> smp_cpu_mt_shift;
703 info->configured++;
704 }
705 info->combined = info->configured;
706 }
707}
708
709static int smp_add_present_cpu(int cpu);
710
711static int __smp_rescan_cpus(struct sclp_core_info *info, int sysfs_add)
712{
713 struct pcpu *pcpu;
714 cpumask_t avail;
715 int cpu, nr, i, j;
716 u16 address;
717
718 nr = 0;
719 cpumask_xor(&avail, cpu_possible_mask, cpu_present_mask);
720 cpu = cpumask_first(&avail);
721 for (i = 0; (i < info->combined) && (cpu < nr_cpu_ids); i++) {
722 if (sclp.has_core_type && info->core[i].type != boot_core_type)
723 continue;
724 address = info->core[i].core_id << smp_cpu_mt_shift;
725 for (j = 0; j <= smp_cpu_mtid; j++) {
726 if (pcpu_find_address(cpu_present_mask, address + j))
727 continue;
728 pcpu = pcpu_devices + cpu;
729 pcpu->address = address + j;
730 pcpu->state =
731 (cpu >= info->configured*(smp_cpu_mtid + 1)) ?
732 CPU_STATE_STANDBY : CPU_STATE_CONFIGURED;
733 smp_cpu_set_polarization(cpu, POLARIZATION_UNKNOWN);
734 set_cpu_present(cpu, true);
735 if (sysfs_add && smp_add_present_cpu(cpu) != 0)
736 set_cpu_present(cpu, false);
737 else
738 nr++;
739 cpu = cpumask_next(cpu, &avail);
740 if (cpu >= nr_cpu_ids)
741 break;
742 }
743 }
744 return nr;
745}
746
747void __init smp_detect_cpus(void)
748{
749 unsigned int cpu, mtid, c_cpus, s_cpus;
750 struct sclp_core_info *info;
751 u16 address;
752
753 /* Get CPU information */
754 info = memblock_virt_alloc(sizeof(*info), 8);
755 smp_get_core_info(info, 1);
756 /* Find boot CPU type */
757 if (sclp.has_core_type) {
758 address = stap();
759 for (cpu = 0; cpu < info->combined; cpu++)
760 if (info->core[cpu].core_id == address) {
761 /* The boot cpu dictates the cpu type. */
762 boot_core_type = info->core[cpu].type;
763 break;
764 }
765 if (cpu >= info->combined)
766 panic("Could not find boot CPU type");
767 }
768
769 /* Set multi-threading state for the current system */
770 mtid = boot_core_type ? sclp.mtid : sclp.mtid_cp;
771 mtid = (mtid < smp_max_threads) ? mtid : smp_max_threads - 1;
772 pcpu_set_smt(mtid);
773
774 /* Print number of CPUs */
775 c_cpus = s_cpus = 0;
776 for (cpu = 0; cpu < info->combined; cpu++) {
777 if (sclp.has_core_type &&
778 info->core[cpu].type != boot_core_type)
779 continue;
780 if (cpu < info->configured)
781 c_cpus += smp_cpu_mtid + 1;
782 else
783 s_cpus += smp_cpu_mtid + 1;
784 }
785 pr_info("%d configured CPUs, %d standby CPUs\n", c_cpus, s_cpus);
786
787 /* Add CPUs present at boot */
788 get_online_cpus();
789 __smp_rescan_cpus(info, 0);
790 put_online_cpus();
791 memblock_free_early((unsigned long)info, sizeof(*info));
792}
793
794/*
795 * Activate a secondary processor.
796 */
797static void smp_start_secondary(void *cpuvoid)
798{
799 int cpu = smp_processor_id();
800
801 S390_lowcore.last_update_clock = get_tod_clock();
802 S390_lowcore.restart_stack = (unsigned long) restart_stack;
803 S390_lowcore.restart_fn = (unsigned long) do_restart;
804 S390_lowcore.restart_data = 0;
805 S390_lowcore.restart_source = -1UL;
806 restore_access_regs(S390_lowcore.access_regs_save_area);
807 __ctl_load(S390_lowcore.cregs_save_area, 0, 15);
808 __load_psw_mask(PSW_KERNEL_BITS | PSW_MASK_DAT);
809 cpu_init();
810 preempt_disable();
811 init_cpu_timer();
812 vtime_init();
813 pfault_init();
814 notify_cpu_starting(cpu);
815 if (topology_cpu_dedicated(cpu))
816 set_cpu_flag(CIF_DEDICATED_CPU);
817 else
818 clear_cpu_flag(CIF_DEDICATED_CPU);
819 set_cpu_online(cpu, true);
820 inc_irq_stat(CPU_RST);
821 local_irq_enable();
822 cpu_startup_entry(CPUHP_AP_ONLINE_IDLE);
823}
824
825/* Upping and downing of CPUs */
826int __cpu_up(unsigned int cpu, struct task_struct *tidle)
827{
828 struct pcpu *pcpu;
829 int base, i, rc;
830
831 pcpu = pcpu_devices + cpu;
832 if (pcpu->state != CPU_STATE_CONFIGURED)
833 return -EIO;
834 base = smp_get_base_cpu(cpu);
835 for (i = 0; i <= smp_cpu_mtid; i++) {
836 if (base + i < nr_cpu_ids)
837 if (cpu_online(base + i))
838 break;
839 }
840 /*
841 * If this is the first CPU of the core to get online
842 * do an initial CPU reset.
843 */
844 if (i > smp_cpu_mtid &&
845 pcpu_sigp_retry(pcpu_devices + base, SIGP_INITIAL_CPU_RESET, 0) !=
846 SIGP_CC_ORDER_CODE_ACCEPTED)
847 return -EIO;
848
849 rc = pcpu_alloc_lowcore(pcpu, cpu);
850 if (rc)
851 return rc;
852 pcpu_prepare_secondary(pcpu, cpu);
853 pcpu_attach_task(pcpu, tidle);
854 pcpu_start_fn(pcpu, smp_start_secondary, NULL);
855 /* Wait until cpu puts itself in the online & active maps */
856 while (!cpu_online(cpu))
857 cpu_relax();
858 return 0;
859}
860
861static unsigned int setup_possible_cpus __initdata;
862
863static int __init _setup_possible_cpus(char *s)
864{
865 get_option(&s, &setup_possible_cpus);
866 return 0;
867}
868early_param("possible_cpus", _setup_possible_cpus);
869
870#ifdef CONFIG_HOTPLUG_CPU
871
872int __cpu_disable(void)
873{
874 unsigned long cregs[16];
875
876 /* Handle possible pending IPIs */
877 smp_handle_ext_call();
878 set_cpu_online(smp_processor_id(), false);
879 /* Disable pseudo page faults on this cpu. */
880 pfault_fini();
881 /* Disable interrupt sources via control register. */
882 __ctl_store(cregs, 0, 15);
883 cregs[0] &= ~0x0000ee70UL; /* disable all external interrupts */
884 cregs[6] &= ~0xff000000UL; /* disable all I/O interrupts */
885 cregs[14] &= ~0x1f000000UL; /* disable most machine checks */
886 __ctl_load(cregs, 0, 15);
887 clear_cpu_flag(CIF_NOHZ_DELAY);
888 return 0;
889}
890
891void __cpu_die(unsigned int cpu)
892{
893 struct pcpu *pcpu;
894
895 /* Wait until target cpu is down */
896 pcpu = pcpu_devices + cpu;
897 while (!pcpu_stopped(pcpu))
898 cpu_relax();
899 pcpu_free_lowcore(pcpu);
900 cpumask_clear_cpu(cpu, mm_cpumask(&init_mm));
901 cpumask_clear_cpu(cpu, &init_mm.context.cpu_attach_mask);
902}
903
904void __noreturn cpu_die(void)
905{
906 idle_task_exit();
907 __bpon();
908 pcpu_sigp_retry(pcpu_devices + smp_processor_id(), SIGP_STOP, 0);
909 for (;;) ;
910}
911
912#endif /* CONFIG_HOTPLUG_CPU */
913
914void __init smp_fill_possible_mask(void)
915{
916 unsigned int possible, sclp_max, cpu;
917
918 sclp_max = max(sclp.mtid, sclp.mtid_cp) + 1;
919 sclp_max = min(smp_max_threads, sclp_max);
920 sclp_max = (sclp.max_cores * sclp_max) ?: nr_cpu_ids;
921 possible = setup_possible_cpus ?: nr_cpu_ids;
922 possible = min(possible, sclp_max);
923 for (cpu = 0; cpu < possible && cpu < nr_cpu_ids; cpu++)
924 set_cpu_possible(cpu, true);
925}
926
927void __init smp_prepare_cpus(unsigned int max_cpus)
928{
929 /* request the 0x1201 emergency signal external interrupt */
930 if (register_external_irq(EXT_IRQ_EMERGENCY_SIG, do_ext_call_interrupt))
931 panic("Couldn't request external interrupt 0x1201");
932 /* request the 0x1202 external call external interrupt */
933 if (register_external_irq(EXT_IRQ_EXTERNAL_CALL, do_ext_call_interrupt))
934 panic("Couldn't request external interrupt 0x1202");
935}
936
937void __init smp_prepare_boot_cpu(void)
938{
939 struct pcpu *pcpu = pcpu_devices;
940
941 WARN_ON(!cpu_present(0) || !cpu_online(0));
942 pcpu->state = CPU_STATE_CONFIGURED;
943 pcpu->lowcore = (struct lowcore *)(unsigned long) store_prefix();
944 S390_lowcore.percpu_offset = __per_cpu_offset[0];
945 smp_cpu_set_polarization(0, POLARIZATION_UNKNOWN);
946}
947
948void __init smp_cpus_done(unsigned int max_cpus)
949{
950}
951
952void __init smp_setup_processor_id(void)
953{
954 pcpu_devices[0].address = stap();
955 S390_lowcore.cpu_nr = 0;
956 S390_lowcore.spinlock_lockval = arch_spin_lockval(0);
957 S390_lowcore.spinlock_index = 0;
958}
959
960/*
961 * the frequency of the profiling timer can be changed
962 * by writing a multiplier value into /proc/profile.
963 *
964 * usually you want to run this on all CPUs ;)
965 */
966int setup_profiling_timer(unsigned int multiplier)
967{
968 return 0;
969}
970
971#ifdef CONFIG_HOTPLUG_CPU
972static ssize_t cpu_configure_show(struct device *dev,
973 struct device_attribute *attr, char *buf)
974{
975 ssize_t count;
976
977 mutex_lock(&smp_cpu_state_mutex);
978 count = sprintf(buf, "%d\n", pcpu_devices[dev->id].state);
979 mutex_unlock(&smp_cpu_state_mutex);
980 return count;
981}
982
983static ssize_t cpu_configure_store(struct device *dev,
984 struct device_attribute *attr,
985 const char *buf, size_t count)
986{
987 struct pcpu *pcpu;
988 int cpu, val, rc, i;
989 char delim;
990
991 if (sscanf(buf, "%d %c", &val, &delim) != 1)
992 return -EINVAL;
993 if (val != 0 && val != 1)
994 return -EINVAL;
995 get_online_cpus();
996 mutex_lock(&smp_cpu_state_mutex);
997 rc = -EBUSY;
998 /* disallow configuration changes of online cpus and cpu 0 */
999 cpu = dev->id;
1000 cpu = smp_get_base_cpu(cpu);
1001 if (cpu == 0)
1002 goto out;
1003 for (i = 0; i <= smp_cpu_mtid; i++)
1004 if (cpu_online(cpu + i))
1005 goto out;
1006 pcpu = pcpu_devices + cpu;
1007 rc = 0;
1008 switch (val) {
1009 case 0:
1010 if (pcpu->state != CPU_STATE_CONFIGURED)
1011 break;
1012 rc = sclp_core_deconfigure(pcpu->address >> smp_cpu_mt_shift);
1013 if (rc)
1014 break;
1015 for (i = 0; i <= smp_cpu_mtid; i++) {
1016 if (cpu + i >= nr_cpu_ids || !cpu_present(cpu + i))
1017 continue;
1018 pcpu[i].state = CPU_STATE_STANDBY;
1019 smp_cpu_set_polarization(cpu + i,
1020 POLARIZATION_UNKNOWN);
1021 }
1022 topology_expect_change();
1023 break;
1024 case 1:
1025 if (pcpu->state != CPU_STATE_STANDBY)
1026 break;
1027 rc = sclp_core_configure(pcpu->address >> smp_cpu_mt_shift);
1028 if (rc)
1029 break;
1030 for (i = 0; i <= smp_cpu_mtid; i++) {
1031 if (cpu + i >= nr_cpu_ids || !cpu_present(cpu + i))
1032 continue;
1033 pcpu[i].state = CPU_STATE_CONFIGURED;
1034 smp_cpu_set_polarization(cpu + i,
1035 POLARIZATION_UNKNOWN);
1036 }
1037 topology_expect_change();
1038 break;
1039 default:
1040 break;
1041 }
1042out:
1043 mutex_unlock(&smp_cpu_state_mutex);
1044 put_online_cpus();
1045 return rc ? rc : count;
1046}
1047static DEVICE_ATTR(configure, 0644, cpu_configure_show, cpu_configure_store);
1048#endif /* CONFIG_HOTPLUG_CPU */
1049
1050static ssize_t show_cpu_address(struct device *dev,
1051 struct device_attribute *attr, char *buf)
1052{
1053 return sprintf(buf, "%d\n", pcpu_devices[dev->id].address);
1054}
1055static DEVICE_ATTR(address, 0444, show_cpu_address, NULL);
1056
1057static struct attribute *cpu_common_attrs[] = {
1058#ifdef CONFIG_HOTPLUG_CPU
1059 &dev_attr_configure.attr,
1060#endif
1061 &dev_attr_address.attr,
1062 NULL,
1063};
1064
1065static struct attribute_group cpu_common_attr_group = {
1066 .attrs = cpu_common_attrs,
1067};
1068
1069static struct attribute *cpu_online_attrs[] = {
1070 &dev_attr_idle_count.attr,
1071 &dev_attr_idle_time_us.attr,
1072 NULL,
1073};
1074
1075static struct attribute_group cpu_online_attr_group = {
1076 .attrs = cpu_online_attrs,
1077};
1078
1079static int smp_cpu_online(unsigned int cpu)
1080{
1081 struct device *s = &per_cpu(cpu_device, cpu)->dev;
1082
1083 return sysfs_create_group(&s->kobj, &cpu_online_attr_group);
1084}
1085static int smp_cpu_pre_down(unsigned int cpu)
1086{
1087 struct device *s = &per_cpu(cpu_device, cpu)->dev;
1088
1089 sysfs_remove_group(&s->kobj, &cpu_online_attr_group);
1090 return 0;
1091}
1092
1093static int smp_add_present_cpu(int cpu)
1094{
1095 struct device *s;
1096 struct cpu *c;
1097 int rc;
1098
1099 c = kzalloc(sizeof(*c), GFP_KERNEL);
1100 if (!c)
1101 return -ENOMEM;
1102 per_cpu(cpu_device, cpu) = c;
1103 s = &c->dev;
1104 c->hotpluggable = 1;
1105 rc = register_cpu(c, cpu);
1106 if (rc)
1107 goto out;
1108 rc = sysfs_create_group(&s->kobj, &cpu_common_attr_group);
1109 if (rc)
1110 goto out_cpu;
1111 rc = topology_cpu_init(c);
1112 if (rc)
1113 goto out_topology;
1114 return 0;
1115
1116out_topology:
1117 sysfs_remove_group(&s->kobj, &cpu_common_attr_group);
1118out_cpu:
1119#ifdef CONFIG_HOTPLUG_CPU
1120 unregister_cpu(c);
1121#endif
1122out:
1123 return rc;
1124}
1125
1126#ifdef CONFIG_HOTPLUG_CPU
1127
1128int __ref smp_rescan_cpus(void)
1129{
1130 struct sclp_core_info *info;
1131 int nr;
1132
1133 info = kzalloc(sizeof(*info), GFP_KERNEL);
1134 if (!info)
1135 return -ENOMEM;
1136 smp_get_core_info(info, 0);
1137 get_online_cpus();
1138 mutex_lock(&smp_cpu_state_mutex);
1139 nr = __smp_rescan_cpus(info, 1);
1140 mutex_unlock(&smp_cpu_state_mutex);
1141 put_online_cpus();
1142 kfree(info);
1143 if (nr)
1144 topology_schedule_update();
1145 return 0;
1146}
1147
1148static ssize_t __ref rescan_store(struct device *dev,
1149 struct device_attribute *attr,
1150 const char *buf,
1151 size_t count)
1152{
1153 int rc;
1154
1155 rc = smp_rescan_cpus();
1156 return rc ? rc : count;
1157}
1158static DEVICE_ATTR_WO(rescan);
1159#endif /* CONFIG_HOTPLUG_CPU */
1160
1161static int __init s390_smp_init(void)
1162{
1163 int cpu, rc = 0;
1164
1165#ifdef CONFIG_HOTPLUG_CPU
1166 rc = device_create_file(cpu_subsys.dev_root, &dev_attr_rescan);
1167 if (rc)
1168 return rc;
1169#endif
1170 for_each_present_cpu(cpu) {
1171 rc = smp_add_present_cpu(cpu);
1172 if (rc)
1173 goto out;
1174 }
1175
1176 rc = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "s390/smp:online",
1177 smp_cpu_online, smp_cpu_pre_down);
1178 rc = rc <= 0 ? rc : 0;
1179out:
1180 return rc;
1181}
1182subsys_initcall(s390_smp_init);
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);