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