6db8526a602d7c4b50a01d967c50023e1341cc02
[linux-2.6.git] / arch / s390 / kernel / smp.c
1 /*
2  *  SMP related functions
3  *
4  *    Copyright IBM Corp. 1999,2012
5  *    Author(s): Denis Joseph Barrow,
6  *               Martin Schwidefsky <schwidefsky@de.ibm.com>,
7  *               Heiko Carstens <heiko.carstens@de.ibm.com>,
8  *
9  *  based on other smp stuff by
10  *    (c) 1995 Alan Cox, CymruNET Ltd  <alan@cymru.net>
11  *    (c) 1998 Ingo Molnar
12  *
13  * The code outside of smp.c uses logical cpu numbers, only smp.c does
14  * the translation of logical to physical cpu ids. All new code that
15  * operates on physical cpu numbers needs to go into smp.c.
16  */
17
18 #define KMSG_COMPONENT "cpu"
19 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
20
21 #include <linux/workqueue.h>
22 #include <linux/module.h>
23 #include <linux/init.h>
24 #include <linux/mm.h>
25 #include <linux/err.h>
26 #include <linux/spinlock.h>
27 #include <linux/kernel_stat.h>
28 #include <linux/delay.h>
29 #include <linux/interrupt.h>
30 #include <linux/irqflags.h>
31 #include <linux/cpu.h>
32 #include <linux/slab.h>
33 #include <linux/crash_dump.h>
34 #include <asm/asm-offsets.h>
35 #include <asm/ipl.h>
36 #include <asm/setup.h>
37 #include <asm/irq.h>
38 #include <asm/tlbflush.h>
39 #include <asm/timer.h>
40 #include <asm/lowcore.h>
41 #include <asm/sclp.h>
42 #include <asm/vdso.h>
43 #include "entry.h"
44
45 enum {
46         sigp_sense = 1,
47         sigp_external_call = 2,
48         sigp_emergency_signal = 3,
49         sigp_start = 4,
50         sigp_stop = 5,
51         sigp_restart = 6,
52         sigp_stop_and_store_status = 9,
53         sigp_initial_cpu_reset = 11,
54         sigp_cpu_reset = 12,
55         sigp_set_prefix = 13,
56         sigp_store_status_at_address = 14,
57         sigp_store_extended_status_at_address = 15,
58         sigp_set_architecture = 18,
59         sigp_conditional_emergency_signal = 19,
60         sigp_sense_running = 21,
61 };
62
63 enum {
64         sigp_order_code_accepted = 0,
65         sigp_status_stored = 1,
66         sigp_busy = 2,
67         sigp_not_operational = 3,
68 };
69
70 enum {
71         ec_schedule = 0,
72         ec_call_function,
73         ec_call_function_single,
74         ec_stop_cpu,
75 };
76
77 enum {
78         CPU_STATE_STANDBY,
79         CPU_STATE_CONFIGURED,
80 };
81
82 struct pcpu {
83         struct cpu cpu;
84         struct task_struct *idle;       /* idle process for the cpu */
85         struct _lowcore *lowcore;       /* lowcore page(s) for the cpu */
86         unsigned long async_stack;      /* async stack for the cpu */
87         unsigned long panic_stack;      /* panic stack for the cpu */
88         unsigned long ec_mask;          /* bit mask for ec_xxx functions */
89         int state;                      /* physical cpu state */
90         u32 status;                     /* last status received via sigp */
91         u16 address;                    /* physical cpu address */
92 };
93
94 static u8 boot_cpu_type;
95 static u16 boot_cpu_address;
96 static struct pcpu pcpu_devices[NR_CPUS];
97
98 DEFINE_MUTEX(smp_cpu_state_mutex);
99
100 /*
101  * Signal processor helper functions.
102  */
103 static inline int __pcpu_sigp(u16 addr, u8 order, u32 parm, u32 *status)
104 {
105         register unsigned int reg1 asm ("1") = parm;
106         int cc;
107
108         asm volatile(
109                 "       sigp    %1,%2,0(%3)\n"
110                 "       ipm     %0\n"
111                 "       srl     %0,28\n"
112                 : "=d" (cc), "+d" (reg1) : "d" (addr), "a" (order) : "cc");
113         if (status && cc == 1)
114                 *status = reg1;
115         return cc;
116 }
117
118 static inline int __pcpu_sigp_relax(u16 addr, u8 order, u32 parm, u32 *status)
119 {
120         int cc;
121
122         while (1) {
123                 cc = __pcpu_sigp(addr, order, parm, status);
124                 if (cc != sigp_busy)
125                         return cc;
126                 cpu_relax();
127         }
128 }
129
130 static int pcpu_sigp_retry(struct pcpu *pcpu, u8 order, u32 parm)
131 {
132         int cc, retry;
133
134         for (retry = 0; ; retry++) {
135                 cc = __pcpu_sigp(pcpu->address, order, parm, &pcpu->status);
136                 if (cc != sigp_busy)
137                         break;
138                 if (retry >= 3)
139                         udelay(10);
140         }
141         return cc;
142 }
143
144 static inline int pcpu_stopped(struct pcpu *pcpu)
145 {
146         if (__pcpu_sigp(pcpu->address, sigp_sense,
147                         0, &pcpu->status) != sigp_status_stored)
148                 return 0;
149         /* Check for stopped and check stop state */
150         return !!(pcpu->status & 0x50);
151 }
152
153 static inline int pcpu_running(struct pcpu *pcpu)
154 {
155         if (__pcpu_sigp(pcpu->address, sigp_sense_running,
156                         0, &pcpu->status) != sigp_status_stored)
157                 return 1;
158         /* Check for running status */
159         return !(pcpu->status & 0x400);
160 }
161
162 /*
163  * Find struct pcpu by cpu address.
164  */
165 static struct pcpu *pcpu_find_address(const struct cpumask *mask, int address)
166 {
167         int cpu;
168
169         for_each_cpu(cpu, mask)
170                 if (pcpu_devices[cpu].address == address)
171                         return pcpu_devices + cpu;
172         return NULL;
173 }
174
175 static void pcpu_ec_call(struct pcpu *pcpu, int ec_bit)
176 {
177         int order;
178
179         set_bit(ec_bit, &pcpu->ec_mask);
180         order = pcpu_running(pcpu) ?
181                 sigp_external_call : sigp_emergency_signal;
182         pcpu_sigp_retry(pcpu, order, 0);
183 }
184
185 static int __cpuinit pcpu_alloc_lowcore(struct pcpu *pcpu, int cpu)
186 {
187         struct _lowcore *lc;
188
189         if (pcpu != &pcpu_devices[0]) {
190                 pcpu->lowcore = (struct _lowcore *)
191                         __get_free_pages(GFP_KERNEL | GFP_DMA, LC_ORDER);
192                 pcpu->async_stack = __get_free_pages(GFP_KERNEL, ASYNC_ORDER);
193                 pcpu->panic_stack = __get_free_page(GFP_KERNEL);
194                 if (!pcpu->lowcore || !pcpu->panic_stack || !pcpu->async_stack)
195                         goto out;
196         }
197         lc = pcpu->lowcore;
198         memcpy(lc, &S390_lowcore, 512);
199         memset((char *) lc + 512, 0, sizeof(*lc) - 512);
200         lc->async_stack = pcpu->async_stack + ASYNC_SIZE;
201         lc->panic_stack = pcpu->panic_stack + PAGE_SIZE;
202         lc->cpu_nr = cpu;
203 #ifndef CONFIG_64BIT
204         if (MACHINE_HAS_IEEE) {
205                 lc->extended_save_area_addr = get_zeroed_page(GFP_KERNEL);
206                 if (!lc->extended_save_area_addr)
207                         goto out;
208         }
209 #else
210         if (vdso_alloc_per_cpu(lc))
211                 goto out;
212 #endif
213         lowcore_ptr[cpu] = lc;
214         pcpu_sigp_retry(pcpu, sigp_set_prefix, (u32)(unsigned long) lc);
215         return 0;
216 out:
217         if (pcpu != &pcpu_devices[0]) {
218                 free_page(pcpu->panic_stack);
219                 free_pages(pcpu->async_stack, ASYNC_ORDER);
220                 free_pages((unsigned long) pcpu->lowcore, LC_ORDER);
221         }
222         return -ENOMEM;
223 }
224
225 static void pcpu_free_lowcore(struct pcpu *pcpu)
226 {
227         pcpu_sigp_retry(pcpu, sigp_set_prefix, 0);
228         lowcore_ptr[pcpu - pcpu_devices] = NULL;
229 #ifndef CONFIG_64BIT
230         if (MACHINE_HAS_IEEE) {
231                 struct _lowcore *lc = pcpu->lowcore;
232
233                 free_page((unsigned long) lc->extended_save_area_addr);
234                 lc->extended_save_area_addr = 0;
235         }
236 #else
237         vdso_free_per_cpu(pcpu->lowcore);
238 #endif
239         if (pcpu != &pcpu_devices[0]) {
240                 free_page(pcpu->panic_stack);
241                 free_pages(pcpu->async_stack, ASYNC_ORDER);
242                 free_pages((unsigned long) pcpu->lowcore, LC_ORDER);
243         }
244 }
245
246 static void pcpu_prepare_secondary(struct pcpu *pcpu, int cpu)
247 {
248         struct _lowcore *lc = pcpu->lowcore;
249
250         atomic_inc(&init_mm.context.attach_count);
251         lc->cpu_nr = cpu;
252         lc->percpu_offset = __per_cpu_offset[cpu];
253         lc->kernel_asce = S390_lowcore.kernel_asce;
254         lc->machine_flags = S390_lowcore.machine_flags;
255         lc->ftrace_func = S390_lowcore.ftrace_func;
256         lc->user_timer = lc->system_timer = lc->steal_timer = 0;
257         __ctl_store(lc->cregs_save_area, 0, 15);
258         save_access_regs((unsigned int *) lc->access_regs_save_area);
259         memcpy(lc->stfle_fac_list, S390_lowcore.stfle_fac_list,
260                MAX_FACILITY_BIT/8);
261 }
262
263 static void pcpu_attach_task(struct pcpu *pcpu, struct task_struct *tsk)
264 {
265         struct _lowcore *lc = pcpu->lowcore;
266         struct thread_info *ti = task_thread_info(tsk);
267
268         lc->kernel_stack = (unsigned long) task_stack_page(tsk) + THREAD_SIZE;
269         lc->thread_info = (unsigned long) task_thread_info(tsk);
270         lc->current_task = (unsigned long) tsk;
271         lc->user_timer = ti->user_timer;
272         lc->system_timer = ti->system_timer;
273         lc->steal_timer = 0;
274 }
275
276 static void pcpu_start_fn(struct pcpu *pcpu, void (*func)(void *), void *data)
277 {
278         struct _lowcore *lc = pcpu->lowcore;
279
280         lc->restart_stack = lc->kernel_stack;
281         lc->restart_fn = (unsigned long) func;
282         lc->restart_data = (unsigned long) data;
283         lc->restart_source = -1UL;
284         pcpu_sigp_retry(pcpu, sigp_restart, 0);
285 }
286
287 /*
288  * Call function via PSW restart on pcpu and stop the current cpu.
289  */
290 static void pcpu_delegate(struct pcpu *pcpu, void (*func)(void *),
291                           void *data, unsigned long stack)
292 {
293         struct _lowcore *lc = pcpu->lowcore;
294         unsigned short this_cpu;
295
296         __load_psw_mask(psw_kernel_bits);
297         this_cpu = stap();
298         if (pcpu->address == this_cpu)
299                 func(data);     /* should not return */
300         /* Stop target cpu (if func returns this stops the current cpu). */
301         pcpu_sigp_retry(pcpu, sigp_stop, 0);
302         /* Restart func on the target cpu and stop the current cpu. */
303         lc->restart_stack = stack;
304         lc->restart_fn = (unsigned long) func;
305         lc->restart_data = (unsigned long) data;
306         lc->restart_source = (unsigned long) this_cpu;
307         asm volatile(
308                 "0:     sigp    0,%0,6  # sigp restart to target cpu\n"
309                 "       brc     2,0b    # busy, try again\n"
310                 "1:     sigp    0,%1,5  # sigp stop to current cpu\n"
311                 "       brc     2,1b    # busy, try again\n"
312                 : : "d" (pcpu->address), "d" (this_cpu) : "0", "1", "cc");
313         for (;;) ;
314 }
315
316 /*
317  * Call function on an online CPU.
318  */
319 void smp_call_online_cpu(void (*func)(void *), void *data)
320 {
321         struct pcpu *pcpu;
322
323         /* Use the current cpu if it is online. */
324         pcpu = pcpu_find_address(cpu_online_mask, stap());
325         if (!pcpu)
326                 /* Use the first online cpu. */
327                 pcpu = pcpu_devices + cpumask_first(cpu_online_mask);
328         pcpu_delegate(pcpu, func, data, (unsigned long) restart_stack);
329 }
330
331 /*
332  * Call function on the ipl CPU.
333  */
334 void smp_call_ipl_cpu(void (*func)(void *), void *data)
335 {
336         pcpu_delegate(&pcpu_devices[0], func, data, pcpu_devices->panic_stack);
337 }
338
339 int smp_find_processor_id(u16 address)
340 {
341         int cpu;
342
343         for_each_present_cpu(cpu)
344                 if (pcpu_devices[cpu].address == address)
345                         return cpu;
346         return -1;
347 }
348
349 int smp_vcpu_scheduled(int cpu)
350 {
351         return pcpu_running(pcpu_devices + cpu);
352 }
353
354 void smp_yield(void)
355 {
356         if (MACHINE_HAS_DIAG44)
357                 asm volatile("diag 0,0,0x44");
358 }
359
360 void smp_yield_cpu(int cpu)
361 {
362         if (MACHINE_HAS_DIAG9C)
363                 asm volatile("diag %0,0,0x9c"
364                              : : "d" (pcpu_devices[cpu].address));
365         else if (MACHINE_HAS_DIAG44)
366                 asm volatile("diag 0,0,0x44");
367 }
368
369 /*
370  * Send cpus emergency shutdown signal. This gives the cpus the
371  * opportunity to complete outstanding interrupts.
372  */
373 void smp_emergency_stop(cpumask_t *cpumask)
374 {
375         u64 end;
376         int cpu;
377
378         end = get_clock() + (1000000UL << 12);
379         for_each_cpu(cpu, cpumask) {
380                 struct pcpu *pcpu = pcpu_devices + cpu;
381                 set_bit(ec_stop_cpu, &pcpu->ec_mask);
382                 while (__pcpu_sigp(pcpu->address, sigp_emergency_signal,
383                                    0, NULL) == sigp_busy &&
384                        get_clock() < end)
385                         cpu_relax();
386         }
387         while (get_clock() < end) {
388                 for_each_cpu(cpu, cpumask)
389                         if (pcpu_stopped(pcpu_devices + cpu))
390                                 cpumask_clear_cpu(cpu, cpumask);
391                 if (cpumask_empty(cpumask))
392                         break;
393                 cpu_relax();
394         }
395 }
396
397 /*
398  * Stop all cpus but the current one.
399  */
400 void smp_send_stop(void)
401 {
402         cpumask_t cpumask;
403         int cpu;
404
405         /* Disable all interrupts/machine checks */
406         __load_psw_mask(psw_kernel_bits | PSW_MASK_DAT);
407         trace_hardirqs_off();
408
409         cpumask_copy(&cpumask, cpu_online_mask);
410         cpumask_clear_cpu(smp_processor_id(), &cpumask);
411
412         if (oops_in_progress)
413                 smp_emergency_stop(&cpumask);
414
415         /* stop all processors */
416         for_each_cpu(cpu, &cpumask) {
417                 struct pcpu *pcpu = pcpu_devices + cpu;
418                 pcpu_sigp_retry(pcpu, sigp_stop, 0);
419                 while (!pcpu_stopped(pcpu))
420                         cpu_relax();
421         }
422 }
423
424 /*
425  * Stop the current cpu.
426  */
427 void smp_stop_cpu(void)
428 {
429         pcpu_sigp_retry(pcpu_devices + smp_processor_id(), sigp_stop, 0);
430         for (;;) ;
431 }
432
433 /*
434  * This is the main routine where commands issued by other
435  * cpus are handled.
436  */
437 static void do_ext_call_interrupt(unsigned int ext_int_code,
438                                   unsigned int param32, unsigned long param64)
439 {
440         unsigned long bits;
441         int cpu;
442
443         cpu = smp_processor_id();
444         if ((ext_int_code & 0xffff) == 0x1202)
445                 kstat_cpu(cpu).irqs[EXTINT_EXC]++;
446         else
447                 kstat_cpu(cpu).irqs[EXTINT_EMS]++;
448         /*
449          * handle bit signal external calls
450          */
451         bits = xchg(&pcpu_devices[cpu].ec_mask, 0);
452
453         if (test_bit(ec_stop_cpu, &bits))
454                 smp_stop_cpu();
455
456         if (test_bit(ec_schedule, &bits))
457                 scheduler_ipi();
458
459         if (test_bit(ec_call_function, &bits))
460                 generic_smp_call_function_interrupt();
461
462         if (test_bit(ec_call_function_single, &bits))
463                 generic_smp_call_function_single_interrupt();
464
465 }
466
467 void arch_send_call_function_ipi_mask(const struct cpumask *mask)
468 {
469         int cpu;
470
471         for_each_cpu(cpu, mask)
472                 pcpu_ec_call(pcpu_devices + cpu, ec_call_function);
473 }
474
475 void arch_send_call_function_single_ipi(int cpu)
476 {
477         pcpu_ec_call(pcpu_devices + cpu, ec_call_function_single);
478 }
479
480 #ifndef CONFIG_64BIT
481 /*
482  * this function sends a 'purge tlb' signal to another CPU.
483  */
484 static void smp_ptlb_callback(void *info)
485 {
486         __tlb_flush_local();
487 }
488
489 void smp_ptlb_all(void)
490 {
491         on_each_cpu(smp_ptlb_callback, NULL, 1);
492 }
493 EXPORT_SYMBOL(smp_ptlb_all);
494 #endif /* ! CONFIG_64BIT */
495
496 /*
497  * this function sends a 'reschedule' IPI to another CPU.
498  * it goes straight through and wastes no time serializing
499  * anything. Worst case is that we lose a reschedule ...
500  */
501 void smp_send_reschedule(int cpu)
502 {
503         pcpu_ec_call(pcpu_devices + cpu, ec_schedule);
504 }
505
506 /*
507  * parameter area for the set/clear control bit callbacks
508  */
509 struct ec_creg_mask_parms {
510         unsigned long orval;
511         unsigned long andval;
512         int cr;
513 };
514
515 /*
516  * callback for setting/clearing control bits
517  */
518 static void smp_ctl_bit_callback(void *info)
519 {
520         struct ec_creg_mask_parms *pp = info;
521         unsigned long cregs[16];
522
523         __ctl_store(cregs, 0, 15);
524         cregs[pp->cr] = (cregs[pp->cr] & pp->andval) | pp->orval;
525         __ctl_load(cregs, 0, 15);
526 }
527
528 /*
529  * Set a bit in a control register of all cpus
530  */
531 void smp_ctl_set_bit(int cr, int bit)
532 {
533         struct ec_creg_mask_parms parms = { 1UL << bit, -1UL, cr };
534
535         on_each_cpu(smp_ctl_bit_callback, &parms, 1);
536 }
537 EXPORT_SYMBOL(smp_ctl_set_bit);
538
539 /*
540  * Clear a bit in a control register of all cpus
541  */
542 void smp_ctl_clear_bit(int cr, int bit)
543 {
544         struct ec_creg_mask_parms parms = { 0, ~(1UL << bit), cr };
545
546         on_each_cpu(smp_ctl_bit_callback, &parms, 1);
547 }
548 EXPORT_SYMBOL(smp_ctl_clear_bit);
549
550 #if defined(CONFIG_ZFCPDUMP) || defined(CONFIG_CRASH_DUMP)
551
552 struct save_area *zfcpdump_save_areas[NR_CPUS + 1];
553 EXPORT_SYMBOL_GPL(zfcpdump_save_areas);
554
555 static void __init smp_get_save_area(int cpu, u16 address)
556 {
557         void *lc = pcpu_devices[0].lowcore;
558         struct save_area *save_area;
559
560         if (is_kdump_kernel())
561                 return;
562         if (!OLDMEM_BASE && (address == boot_cpu_address ||
563                              ipl_info.type != IPL_TYPE_FCP_DUMP))
564                 return;
565         if (cpu >= NR_CPUS) {
566                 pr_warning("CPU %i exceeds the maximum %i and is excluded "
567                            "from the dump\n", cpu, NR_CPUS - 1);
568                 return;
569         }
570         save_area = kmalloc(sizeof(struct save_area), GFP_KERNEL);
571         if (!save_area)
572                 panic("could not allocate memory for save area\n");
573         zfcpdump_save_areas[cpu] = save_area;
574 #ifdef CONFIG_CRASH_DUMP
575         if (address == boot_cpu_address) {
576                 /* Copy the registers of the boot cpu. */
577                 copy_oldmem_page(1, (void *) save_area, sizeof(*save_area),
578                                  SAVE_AREA_BASE - PAGE_SIZE, 0);
579                 return;
580         }
581 #endif
582         /* Get the registers of a non-boot cpu. */
583         __pcpu_sigp_relax(address, sigp_stop_and_store_status, 0, NULL);
584         memcpy_real(save_area, lc + SAVE_AREA_BASE, sizeof(*save_area));
585 }
586
587 int smp_store_status(int cpu)
588 {
589         struct pcpu *pcpu;
590
591         pcpu = pcpu_devices + cpu;
592         if (__pcpu_sigp_relax(pcpu->address, sigp_stop_and_store_status,
593                               0, NULL) != sigp_order_code_accepted)
594                 return -EIO;
595         return 0;
596 }
597
598 #else /* CONFIG_ZFCPDUMP || CONFIG_CRASH_DUMP */
599
600 static inline void smp_get_save_area(int cpu, u16 address) { }
601
602 #endif /* CONFIG_ZFCPDUMP || CONFIG_CRASH_DUMP */
603
604 static struct sclp_cpu_info *smp_get_cpu_info(void)
605 {
606         static int use_sigp_detection;
607         struct sclp_cpu_info *info;
608         int address;
609
610         info = kzalloc(sizeof(*info), GFP_KERNEL);
611         if (info && (use_sigp_detection || sclp_get_cpu_info(info))) {
612                 use_sigp_detection = 1;
613                 for (address = 0; address <= MAX_CPU_ADDRESS; address++) {
614                         if (__pcpu_sigp_relax(address, sigp_sense, 0, NULL) ==
615                             sigp_not_operational)
616                                 continue;
617                         info->cpu[info->configured].address = address;
618                         info->configured++;
619                 }
620                 info->combined = info->configured;
621         }
622         return info;
623 }
624
625 static int __devinit smp_add_present_cpu(int cpu);
626
627 static int __devinit __smp_rescan_cpus(struct sclp_cpu_info *info,
628                                        int sysfs_add)
629 {
630         struct pcpu *pcpu;
631         cpumask_t avail;
632         int cpu, nr, i;
633
634         nr = 0;
635         cpumask_xor(&avail, cpu_possible_mask, cpu_present_mask);
636         cpu = cpumask_first(&avail);
637         for (i = 0; (i < info->combined) && (cpu < nr_cpu_ids); i++) {
638                 if (info->has_cpu_type && info->cpu[i].type != boot_cpu_type)
639                         continue;
640                 if (pcpu_find_address(cpu_present_mask, info->cpu[i].address))
641                         continue;
642                 pcpu = pcpu_devices + cpu;
643                 pcpu->address = info->cpu[i].address;
644                 pcpu->state = (cpu >= info->configured) ?
645                         CPU_STATE_STANDBY : CPU_STATE_CONFIGURED;
646                 cpu_set_polarization(cpu, POLARIZATION_UNKNOWN);
647                 set_cpu_present(cpu, true);
648                 if (sysfs_add && smp_add_present_cpu(cpu) != 0)
649                         set_cpu_present(cpu, false);
650                 else
651                         nr++;
652                 cpu = cpumask_next(cpu, &avail);
653         }
654         return nr;
655 }
656
657 static void __init smp_detect_cpus(void)
658 {
659         unsigned int cpu, c_cpus, s_cpus;
660         struct sclp_cpu_info *info;
661
662         info = smp_get_cpu_info();
663         if (!info)
664                 panic("smp_detect_cpus failed to allocate memory\n");
665         if (info->has_cpu_type) {
666                 for (cpu = 0; cpu < info->combined; cpu++) {
667                         if (info->cpu[cpu].address != boot_cpu_address)
668                                 continue;
669                         /* The boot cpu dictates the cpu type. */
670                         boot_cpu_type = info->cpu[cpu].type;
671                         break;
672                 }
673         }
674         c_cpus = s_cpus = 0;
675         for (cpu = 0; cpu < info->combined; cpu++) {
676                 if (info->has_cpu_type && info->cpu[cpu].type != boot_cpu_type)
677                         continue;
678                 if (cpu < info->configured) {
679                         smp_get_save_area(c_cpus, info->cpu[cpu].address);
680                         c_cpus++;
681                 } else
682                         s_cpus++;
683         }
684         pr_info("%d configured CPUs, %d standby CPUs\n", c_cpus, s_cpus);
685         get_online_cpus();
686         __smp_rescan_cpus(info, 0);
687         put_online_cpus();
688         kfree(info);
689 }
690
691 /*
692  *      Activate a secondary processor.
693  */
694 static void __cpuinit smp_start_secondary(void *cpuvoid)
695 {
696         S390_lowcore.last_update_clock = get_clock();
697         S390_lowcore.restart_stack = (unsigned long) restart_stack;
698         S390_lowcore.restart_fn = (unsigned long) do_restart;
699         S390_lowcore.restart_data = 0;
700         S390_lowcore.restart_source = -1UL;
701         restore_access_regs(S390_lowcore.access_regs_save_area);
702         __ctl_load(S390_lowcore.cregs_save_area, 0, 15);
703         __load_psw_mask(psw_kernel_bits | PSW_MASK_DAT);
704         cpu_init();
705         preempt_disable();
706         init_cpu_timer();
707         init_cpu_vtimer();
708         pfault_init();
709         notify_cpu_starting(smp_processor_id());
710         ipi_call_lock();
711         set_cpu_online(smp_processor_id(), true);
712         ipi_call_unlock();
713         /*
714          * Wait until the cpu which brought this one up marked it
715          * active before enabling interrupts.
716          */
717         while (!cpumask_test_cpu(smp_processor_id(), cpu_active_mask))
718                 cpu_relax();
719         local_irq_enable();
720         /* cpu_idle will call schedule for us */
721         cpu_idle();
722 }
723
724 struct create_idle {
725         struct work_struct work;
726         struct task_struct *idle;
727         struct completion done;
728         int cpu;
729 };
730
731 static void __cpuinit smp_fork_idle(struct work_struct *work)
732 {
733         struct create_idle *c_idle;
734
735         c_idle = container_of(work, struct create_idle, work);
736         c_idle->idle = fork_idle(c_idle->cpu);
737         complete(&c_idle->done);
738 }
739
740 /* Upping and downing of CPUs */
741 int __cpuinit __cpu_up(unsigned int cpu)
742 {
743         struct create_idle c_idle;
744         struct pcpu *pcpu;
745         int rc;
746
747         pcpu = pcpu_devices + cpu;
748         if (pcpu->state != CPU_STATE_CONFIGURED)
749                 return -EIO;
750         if (pcpu_sigp_retry(pcpu, sigp_initial_cpu_reset, 0) !=
751             sigp_order_code_accepted)
752                 return -EIO;
753         if (!pcpu->idle) {
754                 c_idle.done = COMPLETION_INITIALIZER_ONSTACK(c_idle.done);
755                 INIT_WORK_ONSTACK(&c_idle.work, smp_fork_idle);
756                 c_idle.cpu = cpu;
757                 schedule_work(&c_idle.work);
758                 wait_for_completion(&c_idle.done);
759                 if (IS_ERR(c_idle.idle))
760                         return PTR_ERR(c_idle.idle);
761                 pcpu->idle = c_idle.idle;
762         }
763         init_idle(pcpu->idle, cpu);
764         rc = pcpu_alloc_lowcore(pcpu, cpu);
765         if (rc)
766                 return rc;
767         pcpu_prepare_secondary(pcpu, cpu);
768         pcpu_attach_task(pcpu, pcpu->idle);
769         pcpu_start_fn(pcpu, smp_start_secondary, NULL);
770         while (!cpu_online(cpu))
771                 cpu_relax();
772         return 0;
773 }
774
775 static int __init setup_possible_cpus(char *s)
776 {
777         int max, cpu;
778
779         if (kstrtoint(s, 0, &max) < 0)
780                 return 0;
781         init_cpu_possible(cpumask_of(0));
782         for (cpu = 1; cpu < max && cpu < nr_cpu_ids; cpu++)
783                 set_cpu_possible(cpu, true);
784         return 0;
785 }
786 early_param("possible_cpus", setup_possible_cpus);
787
788 #ifdef CONFIG_HOTPLUG_CPU
789
790 int __cpu_disable(void)
791 {
792         unsigned long cregs[16];
793
794         set_cpu_online(smp_processor_id(), false);
795         /* Disable pseudo page faults on this cpu. */
796         pfault_fini();
797         /* Disable interrupt sources via control register. */
798         __ctl_store(cregs, 0, 15);
799         cregs[0]  &= ~0x0000ee70UL;     /* disable all external interrupts */
800         cregs[6]  &= ~0xff000000UL;     /* disable all I/O interrupts */
801         cregs[14] &= ~0x1f000000UL;     /* disable most machine checks */
802         __ctl_load(cregs, 0, 15);
803         return 0;
804 }
805
806 void __cpu_die(unsigned int cpu)
807 {
808         struct pcpu *pcpu;
809
810         /* Wait until target cpu is down */
811         pcpu = pcpu_devices + cpu;
812         while (!pcpu_stopped(pcpu))
813                 cpu_relax();
814         pcpu_free_lowcore(pcpu);
815         atomic_dec(&init_mm.context.attach_count);
816 }
817
818 void __noreturn cpu_die(void)
819 {
820         idle_task_exit();
821         pcpu_sigp_retry(pcpu_devices + smp_processor_id(), sigp_stop, 0);
822         for (;;) ;
823 }
824
825 #endif /* CONFIG_HOTPLUG_CPU */
826
827 void __init smp_prepare_cpus(unsigned int max_cpus)
828 {
829         /* request the 0x1201 emergency signal external interrupt */
830         if (register_external_interrupt(0x1201, do_ext_call_interrupt) != 0)
831                 panic("Couldn't request external interrupt 0x1201");
832         /* request the 0x1202 external call external interrupt */
833         if (register_external_interrupt(0x1202, do_ext_call_interrupt) != 0)
834                 panic("Couldn't request external interrupt 0x1202");
835         smp_detect_cpus();
836 }
837
838 void __init smp_prepare_boot_cpu(void)
839 {
840         struct pcpu *pcpu = pcpu_devices;
841
842         boot_cpu_address = stap();
843         pcpu->idle = current;
844         pcpu->state = CPU_STATE_CONFIGURED;
845         pcpu->address = boot_cpu_address;
846         pcpu->lowcore = (struct _lowcore *)(unsigned long) store_prefix();
847         pcpu->async_stack = S390_lowcore.async_stack - ASYNC_SIZE;
848         pcpu->panic_stack = S390_lowcore.panic_stack - PAGE_SIZE;
849         S390_lowcore.percpu_offset = __per_cpu_offset[0];
850         cpu_set_polarization(0, POLARIZATION_UNKNOWN);
851         set_cpu_present(0, true);
852         set_cpu_online(0, true);
853 }
854
855 void __init smp_cpus_done(unsigned int max_cpus)
856 {
857 }
858
859 void __init smp_setup_processor_id(void)
860 {
861         S390_lowcore.cpu_nr = 0;
862 }
863
864 /*
865  * the frequency of the profiling timer can be changed
866  * by writing a multiplier value into /proc/profile.
867  *
868  * usually you want to run this on all CPUs ;)
869  */
870 int setup_profiling_timer(unsigned int multiplier)
871 {
872         return 0;
873 }
874
875 #ifdef CONFIG_HOTPLUG_CPU
876 static ssize_t cpu_configure_show(struct device *dev,
877                                   struct device_attribute *attr, char *buf)
878 {
879         ssize_t count;
880
881         mutex_lock(&smp_cpu_state_mutex);
882         count = sprintf(buf, "%d\n", pcpu_devices[dev->id].state);
883         mutex_unlock(&smp_cpu_state_mutex);
884         return count;
885 }
886
887 static ssize_t cpu_configure_store(struct device *dev,
888                                    struct device_attribute *attr,
889                                    const char *buf, size_t count)
890 {
891         struct pcpu *pcpu;
892         int cpu, val, rc;
893         char delim;
894
895         if (sscanf(buf, "%d %c", &val, &delim) != 1)
896                 return -EINVAL;
897         if (val != 0 && val != 1)
898                 return -EINVAL;
899         get_online_cpus();
900         mutex_lock(&smp_cpu_state_mutex);
901         rc = -EBUSY;
902         /* disallow configuration changes of online cpus and cpu 0 */
903         cpu = dev->id;
904         if (cpu_online(cpu) || cpu == 0)
905                 goto out;
906         pcpu = pcpu_devices + cpu;
907         rc = 0;
908         switch (val) {
909         case 0:
910                 if (pcpu->state != CPU_STATE_CONFIGURED)
911                         break;
912                 rc = sclp_cpu_deconfigure(pcpu->address);
913                 if (rc)
914                         break;
915                 pcpu->state = CPU_STATE_STANDBY;
916                 cpu_set_polarization(cpu, POLARIZATION_UNKNOWN);
917                 topology_expect_change();
918                 break;
919         case 1:
920                 if (pcpu->state != CPU_STATE_STANDBY)
921                         break;
922                 rc = sclp_cpu_configure(pcpu->address);
923                 if (rc)
924                         break;
925                 pcpu->state = CPU_STATE_CONFIGURED;
926                 cpu_set_polarization(cpu, POLARIZATION_UNKNOWN);
927                 topology_expect_change();
928                 break;
929         default:
930                 break;
931         }
932 out:
933         mutex_unlock(&smp_cpu_state_mutex);
934         put_online_cpus();
935         return rc ? rc : count;
936 }
937 static DEVICE_ATTR(configure, 0644, cpu_configure_show, cpu_configure_store);
938 #endif /* CONFIG_HOTPLUG_CPU */
939
940 static ssize_t show_cpu_address(struct device *dev,
941                                 struct device_attribute *attr, char *buf)
942 {
943         return sprintf(buf, "%d\n", pcpu_devices[dev->id].address);
944 }
945 static DEVICE_ATTR(address, 0444, show_cpu_address, NULL);
946
947 static struct attribute *cpu_common_attrs[] = {
948 #ifdef CONFIG_HOTPLUG_CPU
949         &dev_attr_configure.attr,
950 #endif
951         &dev_attr_address.attr,
952         NULL,
953 };
954
955 static struct attribute_group cpu_common_attr_group = {
956         .attrs = cpu_common_attrs,
957 };
958
959 static ssize_t show_capability(struct device *dev,
960                                 struct device_attribute *attr, char *buf)
961 {
962         unsigned int capability;
963         int rc;
964
965         rc = get_cpu_capability(&capability);
966         if (rc)
967                 return rc;
968         return sprintf(buf, "%u\n", capability);
969 }
970 static DEVICE_ATTR(capability, 0444, show_capability, NULL);
971
972 static ssize_t show_idle_count(struct device *dev,
973                                 struct device_attribute *attr, char *buf)
974 {
975         struct s390_idle_data *idle;
976         unsigned long long idle_count;
977         unsigned int sequence;
978
979         idle = &per_cpu(s390_idle, dev->id);
980 repeat:
981         sequence = idle->sequence;
982         smp_rmb();
983         if (sequence & 1)
984                 goto repeat;
985         idle_count = idle->idle_count;
986         if (idle->idle_enter)
987                 idle_count++;
988         smp_rmb();
989         if (idle->sequence != sequence)
990                 goto repeat;
991         return sprintf(buf, "%llu\n", idle_count);
992 }
993 static DEVICE_ATTR(idle_count, 0444, show_idle_count, NULL);
994
995 static ssize_t show_idle_time(struct device *dev,
996                                 struct device_attribute *attr, char *buf)
997 {
998         struct s390_idle_data *idle;
999         unsigned long long now, idle_time, idle_enter;
1000         unsigned int sequence;
1001
1002         idle = &per_cpu(s390_idle, dev->id);
1003         now = get_clock();
1004 repeat:
1005         sequence = idle->sequence;
1006         smp_rmb();
1007         if (sequence & 1)
1008                 goto repeat;
1009         idle_time = idle->idle_time;
1010         idle_enter = idle->idle_enter;
1011         if (idle_enter != 0ULL && idle_enter < now)
1012                 idle_time += now - idle_enter;
1013         smp_rmb();
1014         if (idle->sequence != sequence)
1015                 goto repeat;
1016         return sprintf(buf, "%llu\n", idle_time >> 12);
1017 }
1018 static DEVICE_ATTR(idle_time_us, 0444, show_idle_time, NULL);
1019
1020 static struct attribute *cpu_online_attrs[] = {
1021         &dev_attr_capability.attr,
1022         &dev_attr_idle_count.attr,
1023         &dev_attr_idle_time_us.attr,
1024         NULL,
1025 };
1026
1027 static struct attribute_group cpu_online_attr_group = {
1028         .attrs = cpu_online_attrs,
1029 };
1030
1031 static int __cpuinit smp_cpu_notify(struct notifier_block *self,
1032                                     unsigned long action, void *hcpu)
1033 {
1034         unsigned int cpu = (unsigned int)(long)hcpu;
1035         struct cpu *c = &pcpu_devices[cpu].cpu;
1036         struct device *s = &c->dev;
1037         struct s390_idle_data *idle;
1038         int err = 0;
1039
1040         switch (action) {
1041         case CPU_ONLINE:
1042         case CPU_ONLINE_FROZEN:
1043                 idle = &per_cpu(s390_idle, cpu);
1044                 memset(idle, 0, sizeof(struct s390_idle_data));
1045                 err = sysfs_create_group(&s->kobj, &cpu_online_attr_group);
1046                 break;
1047         case CPU_DEAD:
1048         case CPU_DEAD_FROZEN:
1049                 sysfs_remove_group(&s->kobj, &cpu_online_attr_group);
1050                 break;
1051         }
1052         return notifier_from_errno(err);
1053 }
1054
1055 static struct notifier_block __cpuinitdata smp_cpu_nb = {
1056         .notifier_call = smp_cpu_notify,
1057 };
1058
1059 static int __devinit smp_add_present_cpu(int cpu)
1060 {
1061         struct cpu *c = &pcpu_devices[cpu].cpu;
1062         struct device *s = &c->dev;
1063         int rc;
1064
1065         c->hotpluggable = 1;
1066         rc = register_cpu(c, cpu);
1067         if (rc)
1068                 goto out;
1069         rc = sysfs_create_group(&s->kobj, &cpu_common_attr_group);
1070         if (rc)
1071                 goto out_cpu;
1072         if (cpu_online(cpu)) {
1073                 rc = sysfs_create_group(&s->kobj, &cpu_online_attr_group);
1074                 if (rc)
1075                         goto out_online;
1076         }
1077         rc = topology_cpu_init(c);
1078         if (rc)
1079                 goto out_topology;
1080         return 0;
1081
1082 out_topology:
1083         if (cpu_online(cpu))
1084                 sysfs_remove_group(&s->kobj, &cpu_online_attr_group);
1085 out_online:
1086         sysfs_remove_group(&s->kobj, &cpu_common_attr_group);
1087 out_cpu:
1088 #ifdef CONFIG_HOTPLUG_CPU
1089         unregister_cpu(c);
1090 #endif
1091 out:
1092         return rc;
1093 }
1094
1095 #ifdef CONFIG_HOTPLUG_CPU
1096
1097 int __ref smp_rescan_cpus(void)
1098 {
1099         struct sclp_cpu_info *info;
1100         int nr;
1101
1102         info = smp_get_cpu_info();
1103         if (!info)
1104                 return -ENOMEM;
1105         get_online_cpus();
1106         mutex_lock(&smp_cpu_state_mutex);
1107         nr = __smp_rescan_cpus(info, 1);
1108         mutex_unlock(&smp_cpu_state_mutex);
1109         put_online_cpus();
1110         kfree(info);
1111         if (nr)
1112                 topology_schedule_update();
1113         return 0;
1114 }
1115
1116 static ssize_t __ref rescan_store(struct device *dev,
1117                                   struct device_attribute *attr,
1118                                   const char *buf,
1119                                   size_t count)
1120 {
1121         int rc;
1122
1123         rc = smp_rescan_cpus();
1124         return rc ? rc : count;
1125 }
1126 static DEVICE_ATTR(rescan, 0200, NULL, rescan_store);
1127 #endif /* CONFIG_HOTPLUG_CPU */
1128
1129 static int __init s390_smp_init(void)
1130 {
1131         int cpu, rc;
1132
1133         register_cpu_notifier(&smp_cpu_nb);
1134 #ifdef CONFIG_HOTPLUG_CPU
1135         rc = device_create_file(cpu_subsys.dev_root, &dev_attr_rescan);
1136         if (rc)
1137                 return rc;
1138 #endif
1139         for_each_present_cpu(cpu) {
1140                 rc = smp_add_present_cpu(cpu);
1141                 if (rc)
1142                         return rc;
1143         }
1144         return 0;
1145 }
1146 subsys_initcall(s390_smp_init);