[S390] kernel: Shutdown Actions Interface
[safe/jmp/linux-2.6] / arch / s390 / kernel / smp.c
1 /*
2  *  arch/s390/kernel/smp.c
3  *
4  *    Copyright IBM Corp. 1999,2007
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 #include <linux/module.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/cache.h>
31 #include <linux/interrupt.h>
32 #include <linux/cpu.h>
33 #include <linux/timex.h>
34 #include <linux/bootmem.h>
35 #include <asm/ipl.h>
36 #include <asm/setup.h>
37 #include <asm/sigp.h>
38 #include <asm/pgalloc.h>
39 #include <asm/irq.h>
40 #include <asm/s390_ext.h>
41 #include <asm/cpcmd.h>
42 #include <asm/tlbflush.h>
43 #include <asm/timer.h>
44 #include <asm/lowcore.h>
45 #include <asm/sclp.h>
46 #include <asm/cpu.h>
47
48 /*
49  * An array with a pointer the lowcore of every CPU.
50  */
51 struct _lowcore *lowcore_ptr[NR_CPUS];
52 EXPORT_SYMBOL(lowcore_ptr);
53
54 cpumask_t cpu_online_map = CPU_MASK_NONE;
55 EXPORT_SYMBOL(cpu_online_map);
56
57 cpumask_t cpu_possible_map = CPU_MASK_ALL;
58 EXPORT_SYMBOL(cpu_possible_map);
59
60 static struct task_struct *current_set[NR_CPUS];
61
62 static u8 smp_cpu_type;
63 static int smp_use_sigp_detection;
64
65 enum s390_cpu_state {
66         CPU_STATE_STANDBY,
67         CPU_STATE_CONFIGURED,
68 };
69
70 #ifdef CONFIG_HOTPLUG_CPU
71 static DEFINE_MUTEX(smp_cpu_state_mutex);
72 #endif
73 static int smp_cpu_state[NR_CPUS];
74
75 static DEFINE_PER_CPU(struct cpu, cpu_devices);
76 DEFINE_PER_CPU(struct s390_idle_data, s390_idle);
77
78 static void smp_ext_bitcall(int, ec_bit_sig);
79
80 /*
81  * Structure and data for __smp_call_function_map(). This is designed to
82  * minimise static memory requirements. It also looks cleaner.
83  */
84 static DEFINE_SPINLOCK(call_lock);
85
86 struct call_data_struct {
87         void (*func) (void *info);
88         void *info;
89         cpumask_t started;
90         cpumask_t finished;
91         int wait;
92 };
93
94 static struct call_data_struct *call_data;
95
96 /*
97  * 'Call function' interrupt callback
98  */
99 static void do_call_function(void)
100 {
101         void (*func) (void *info) = call_data->func;
102         void *info = call_data->info;
103         int wait = call_data->wait;
104
105         cpu_set(smp_processor_id(), call_data->started);
106         (*func)(info);
107         if (wait)
108                 cpu_set(smp_processor_id(), call_data->finished);;
109 }
110
111 static void __smp_call_function_map(void (*func) (void *info), void *info,
112                                     int nonatomic, int wait, cpumask_t map)
113 {
114         struct call_data_struct data;
115         int cpu, local = 0;
116
117         /*
118          * Can deadlock when interrupts are disabled or if in wrong context.
119          */
120         WARN_ON(irqs_disabled() || in_irq());
121
122         /*
123          * Check for local function call. We have to have the same call order
124          * as in on_each_cpu() because of machine_restart_smp().
125          */
126         if (cpu_isset(smp_processor_id(), map)) {
127                 local = 1;
128                 cpu_clear(smp_processor_id(), map);
129         }
130
131         cpus_and(map, map, cpu_online_map);
132         if (cpus_empty(map))
133                 goto out;
134
135         data.func = func;
136         data.info = info;
137         data.started = CPU_MASK_NONE;
138         data.wait = wait;
139         if (wait)
140                 data.finished = CPU_MASK_NONE;
141
142         spin_lock(&call_lock);
143         call_data = &data;
144
145         for_each_cpu_mask(cpu, map)
146                 smp_ext_bitcall(cpu, ec_call_function);
147
148         /* Wait for response */
149         while (!cpus_equal(map, data.started))
150                 cpu_relax();
151         if (wait)
152                 while (!cpus_equal(map, data.finished))
153                         cpu_relax();
154         spin_unlock(&call_lock);
155 out:
156         if (local) {
157                 local_irq_disable();
158                 func(info);
159                 local_irq_enable();
160         }
161 }
162
163 /*
164  * smp_call_function:
165  * @func: the function to run; this must be fast and non-blocking
166  * @info: an arbitrary pointer to pass to the function
167  * @nonatomic: unused
168  * @wait: if true, wait (atomically) until function has completed on other CPUs
169  *
170  * Run a function on all other CPUs.
171  *
172  * You must not call this function with disabled interrupts, from a
173  * hardware interrupt handler or from a bottom half.
174  */
175 int smp_call_function(void (*func) (void *info), void *info, int nonatomic,
176                       int wait)
177 {
178         cpumask_t map;
179
180         preempt_disable();
181         map = cpu_online_map;
182         cpu_clear(smp_processor_id(), map);
183         __smp_call_function_map(func, info, nonatomic, wait, map);
184         preempt_enable();
185         return 0;
186 }
187 EXPORT_SYMBOL(smp_call_function);
188
189 /*
190  * smp_call_function_single:
191  * @cpu: the CPU where func should run
192  * @func: the function to run; this must be fast and non-blocking
193  * @info: an arbitrary pointer to pass to the function
194  * @nonatomic: unused
195  * @wait: if true, wait (atomically) until function has completed on other CPUs
196  *
197  * Run a function on one processor.
198  *
199  * You must not call this function with disabled interrupts, from a
200  * hardware interrupt handler or from a bottom half.
201  */
202 int smp_call_function_single(int cpu, void (*func) (void *info), void *info,
203                              int nonatomic, int wait)
204 {
205         preempt_disable();
206         __smp_call_function_map(func, info, nonatomic, wait,
207                                 cpumask_of_cpu(cpu));
208         preempt_enable();
209         return 0;
210 }
211 EXPORT_SYMBOL(smp_call_function_single);
212
213 void smp_send_stop(void)
214 {
215         int cpu, rc;
216
217         /* Disable all interrupts/machine checks */
218         __load_psw_mask(psw_kernel_bits & ~PSW_MASK_MCHECK);
219
220         /* write magic number to zero page (absolute 0) */
221         lowcore_ptr[smp_processor_id()]->panic_magic = __PANIC_MAGIC;
222
223         /* stop all processors */
224         for_each_online_cpu(cpu) {
225                 if (cpu == smp_processor_id())
226                         continue;
227                 do {
228                         rc = signal_processor(cpu, sigp_stop);
229                 } while (rc == sigp_busy);
230
231                 while (!smp_cpu_not_running(cpu))
232                         cpu_relax();
233         }
234 }
235
236 /*
237  * This is the main routine where commands issued by other
238  * cpus are handled.
239  */
240
241 static void do_ext_call_interrupt(__u16 code)
242 {
243         unsigned long bits;
244
245         /*
246          * handle bit signal external calls
247          *
248          * For the ec_schedule signal we have to do nothing. All the work
249          * is done automatically when we return from the interrupt.
250          */
251         bits = xchg(&S390_lowcore.ext_call_fast, 0);
252
253         if (test_bit(ec_call_function, &bits))
254                 do_call_function();
255 }
256
257 /*
258  * Send an external call sigp to another cpu and return without waiting
259  * for its completion.
260  */
261 static void smp_ext_bitcall(int cpu, ec_bit_sig sig)
262 {
263         /*
264          * Set signaling bit in lowcore of target cpu and kick it
265          */
266         set_bit(sig, (unsigned long *) &lowcore_ptr[cpu]->ext_call_fast);
267         while (signal_processor(cpu, sigp_emergency_signal) == sigp_busy)
268                 udelay(10);
269 }
270
271 #ifndef CONFIG_64BIT
272 /*
273  * this function sends a 'purge tlb' signal to another CPU.
274  */
275 void smp_ptlb_callback(void *info)
276 {
277         __tlb_flush_local();
278 }
279
280 void smp_ptlb_all(void)
281 {
282         on_each_cpu(smp_ptlb_callback, NULL, 0, 1);
283 }
284 EXPORT_SYMBOL(smp_ptlb_all);
285 #endif /* ! CONFIG_64BIT */
286
287 /*
288  * this function sends a 'reschedule' IPI to another CPU.
289  * it goes straight through and wastes no time serializing
290  * anything. Worst case is that we lose a reschedule ...
291  */
292 void smp_send_reschedule(int cpu)
293 {
294         smp_ext_bitcall(cpu, ec_schedule);
295 }
296
297 /*
298  * parameter area for the set/clear control bit callbacks
299  */
300 struct ec_creg_mask_parms {
301         unsigned long orvals[16];
302         unsigned long andvals[16];
303 };
304
305 /*
306  * callback for setting/clearing control bits
307  */
308 static void smp_ctl_bit_callback(void *info)
309 {
310         struct ec_creg_mask_parms *pp = info;
311         unsigned long cregs[16];
312         int i;
313
314         __ctl_store(cregs, 0, 15);
315         for (i = 0; i <= 15; i++)
316                 cregs[i] = (cregs[i] & pp->andvals[i]) | pp->orvals[i];
317         __ctl_load(cregs, 0, 15);
318 }
319
320 /*
321  * Set a bit in a control register of all cpus
322  */
323 void smp_ctl_set_bit(int cr, int bit)
324 {
325         struct ec_creg_mask_parms parms;
326
327         memset(&parms.orvals, 0, sizeof(parms.orvals));
328         memset(&parms.andvals, 0xff, sizeof(parms.andvals));
329         parms.orvals[cr] = 1 << bit;
330         on_each_cpu(smp_ctl_bit_callback, &parms, 0, 1);
331 }
332 EXPORT_SYMBOL(smp_ctl_set_bit);
333
334 /*
335  * Clear a bit in a control register of all cpus
336  */
337 void smp_ctl_clear_bit(int cr, int bit)
338 {
339         struct ec_creg_mask_parms parms;
340
341         memset(&parms.orvals, 0, sizeof(parms.orvals));
342         memset(&parms.andvals, 0xff, sizeof(parms.andvals));
343         parms.andvals[cr] = ~(1L << bit);
344         on_each_cpu(smp_ctl_bit_callback, &parms, 0, 1);
345 }
346 EXPORT_SYMBOL(smp_ctl_clear_bit);
347
348 /*
349  * In early ipl state a temp. logically cpu number is needed, so the sigp
350  * functions can be used to sense other cpus. Since NR_CPUS is >= 2 on
351  * CONFIG_SMP and the ipl cpu is logical cpu 0, it must be 1.
352  */
353 #define CPU_INIT_NO     1
354
355 #if defined(CONFIG_ZFCPDUMP) || defined(CONFIG_ZFCPDUMP_MODULE)
356
357 /*
358  * zfcpdump_prefix_array holds prefix registers for the following scenario:
359  * 64 bit zfcpdump kernel and 31 bit kernel which is to be dumped. We have to
360  * save its prefix registers, since they get lost, when switching from 31 bit
361  * to 64 bit.
362  */
363 unsigned int zfcpdump_prefix_array[NR_CPUS + 1] \
364         __attribute__((__section__(".data")));
365
366 static void __init smp_get_save_area(unsigned int cpu, unsigned int phy_cpu)
367 {
368         if (ipl_info.type != IPL_TYPE_FCP_DUMP)
369                 return;
370         if (cpu >= NR_CPUS) {
371                 printk(KERN_WARNING "Registers for cpu %i not saved since dump "
372                        "kernel was compiled with NR_CPUS=%i\n", cpu, NR_CPUS);
373                 return;
374         }
375         zfcpdump_save_areas[cpu] = kmalloc(sizeof(union save_area), GFP_KERNEL);
376         __cpu_logical_map[CPU_INIT_NO] = (__u16) phy_cpu;
377         while (signal_processor(CPU_INIT_NO, sigp_stop_and_store_status) ==
378                sigp_busy)
379                 cpu_relax();
380         memcpy(zfcpdump_save_areas[cpu],
381                (void *)(unsigned long) store_prefix() + SAVE_AREA_BASE,
382                SAVE_AREA_SIZE);
383 #ifdef CONFIG_64BIT
384         /* copy original prefix register */
385         zfcpdump_save_areas[cpu]->s390x.pref_reg = zfcpdump_prefix_array[cpu];
386 #endif
387 }
388
389 union save_area *zfcpdump_save_areas[NR_CPUS + 1];
390 EXPORT_SYMBOL_GPL(zfcpdump_save_areas);
391
392 #else
393
394 static inline void smp_get_save_area(unsigned int cpu, unsigned int phy_cpu) { }
395
396 #endif /* CONFIG_ZFCPDUMP || CONFIG_ZFCPDUMP_MODULE */
397
398 static int cpu_stopped(int cpu)
399 {
400         __u32 status;
401
402         /* Check for stopped state */
403         if (signal_processor_ps(&status, 0, cpu, sigp_sense) ==
404             sigp_status_stored) {
405                 if (status & 0x40)
406                         return 1;
407         }
408         return 0;
409 }
410
411 static int cpu_known(int cpu_id)
412 {
413         int cpu;
414
415         for_each_present_cpu(cpu) {
416                 if (__cpu_logical_map[cpu] == cpu_id)
417                         return 1;
418         }
419         return 0;
420 }
421
422 static int smp_rescan_cpus_sigp(cpumask_t avail)
423 {
424         int cpu_id, logical_cpu;
425
426         logical_cpu = first_cpu(avail);
427         if (logical_cpu == NR_CPUS)
428                 return 0;
429         for (cpu_id = 0; cpu_id <= 65535; cpu_id++) {
430                 if (cpu_known(cpu_id))
431                         continue;
432                 __cpu_logical_map[logical_cpu] = cpu_id;
433                 if (!cpu_stopped(logical_cpu))
434                         continue;
435                 cpu_set(logical_cpu, cpu_present_map);
436                 smp_cpu_state[logical_cpu] = CPU_STATE_CONFIGURED;
437                 logical_cpu = next_cpu(logical_cpu, avail);
438                 if (logical_cpu == NR_CPUS)
439                         break;
440         }
441         return 0;
442 }
443
444 static int smp_rescan_cpus_sclp(cpumask_t avail)
445 {
446         struct sclp_cpu_info *info;
447         int cpu_id, logical_cpu, cpu;
448         int rc;
449
450         logical_cpu = first_cpu(avail);
451         if (logical_cpu == NR_CPUS)
452                 return 0;
453         info = kmalloc(sizeof(*info), GFP_KERNEL);
454         if (!info)
455                 return -ENOMEM;
456         rc = sclp_get_cpu_info(info);
457         if (rc)
458                 goto out;
459         for (cpu = 0; cpu < info->combined; cpu++) {
460                 if (info->has_cpu_type && info->cpu[cpu].type != smp_cpu_type)
461                         continue;
462                 cpu_id = info->cpu[cpu].address;
463                 if (cpu_known(cpu_id))
464                         continue;
465                 __cpu_logical_map[logical_cpu] = cpu_id;
466                 cpu_set(logical_cpu, cpu_present_map);
467                 if (cpu >= info->configured)
468                         smp_cpu_state[logical_cpu] = CPU_STATE_STANDBY;
469                 else
470                         smp_cpu_state[logical_cpu] = CPU_STATE_CONFIGURED;
471                 logical_cpu = next_cpu(logical_cpu, avail);
472                 if (logical_cpu == NR_CPUS)
473                         break;
474         }
475 out:
476         kfree(info);
477         return rc;
478 }
479
480 static int smp_rescan_cpus(void)
481 {
482         cpumask_t avail;
483
484         cpus_xor(avail, cpu_possible_map, cpu_present_map);
485         if (smp_use_sigp_detection)
486                 return smp_rescan_cpus_sigp(avail);
487         else
488                 return smp_rescan_cpus_sclp(avail);
489 }
490
491 static void __init smp_detect_cpus(void)
492 {
493         unsigned int cpu, c_cpus, s_cpus;
494         struct sclp_cpu_info *info;
495         u16 boot_cpu_addr, cpu_addr;
496
497         c_cpus = 1;
498         s_cpus = 0;
499         boot_cpu_addr = S390_lowcore.cpu_data.cpu_addr;
500         info = kmalloc(sizeof(*info), GFP_KERNEL);
501         if (!info)
502                 panic("smp_detect_cpus failed to allocate memory\n");
503         /* Use sigp detection algorithm if sclp doesn't work. */
504         if (sclp_get_cpu_info(info)) {
505                 smp_use_sigp_detection = 1;
506                 for (cpu = 0; cpu <= 65535; cpu++) {
507                         if (cpu == boot_cpu_addr)
508                                 continue;
509                         __cpu_logical_map[CPU_INIT_NO] = cpu;
510                         if (!cpu_stopped(CPU_INIT_NO))
511                                 continue;
512                         smp_get_save_area(c_cpus, cpu);
513                         c_cpus++;
514                 }
515                 goto out;
516         }
517
518         if (info->has_cpu_type) {
519                 for (cpu = 0; cpu < info->combined; cpu++) {
520                         if (info->cpu[cpu].address == boot_cpu_addr) {
521                                 smp_cpu_type = info->cpu[cpu].type;
522                                 break;
523                         }
524                 }
525         }
526
527         for (cpu = 0; cpu < info->combined; cpu++) {
528                 if (info->has_cpu_type && info->cpu[cpu].type != smp_cpu_type)
529                         continue;
530                 cpu_addr = info->cpu[cpu].address;
531                 if (cpu_addr == boot_cpu_addr)
532                         continue;
533                 __cpu_logical_map[CPU_INIT_NO] = cpu_addr;
534                 if (!cpu_stopped(CPU_INIT_NO)) {
535                         s_cpus++;
536                         continue;
537                 }
538                 smp_get_save_area(c_cpus, cpu_addr);
539                 c_cpus++;
540         }
541 out:
542         kfree(info);
543         printk(KERN_INFO "CPUs: %d configured, %d standby\n", c_cpus, s_cpus);
544         lock_cpu_hotplug();
545         smp_rescan_cpus();
546         unlock_cpu_hotplug();
547 }
548
549 /*
550  *      Activate a secondary processor.
551  */
552 int __cpuinit start_secondary(void *cpuvoid)
553 {
554         /* Setup the cpu */
555         cpu_init();
556         preempt_disable();
557         /* Enable TOD clock interrupts on the secondary cpu. */
558         init_cpu_timer();
559 #ifdef CONFIG_VIRT_TIMER
560         /* Enable cpu timer interrupts on the secondary cpu. */
561         init_cpu_vtimer();
562 #endif
563         /* Enable pfault pseudo page faults on this cpu. */
564         pfault_init();
565
566         /* Mark this cpu as online */
567         cpu_set(smp_processor_id(), cpu_online_map);
568         /* Switch on interrupts */
569         local_irq_enable();
570         /* Print info about this processor */
571         print_cpu_info(&S390_lowcore.cpu_data);
572         /* cpu_idle will call schedule for us */
573         cpu_idle();
574         return 0;
575 }
576
577 static void __init smp_create_idle(unsigned int cpu)
578 {
579         struct task_struct *p;
580
581         /*
582          *  don't care about the psw and regs settings since we'll never
583          *  reschedule the forked task.
584          */
585         p = fork_idle(cpu);
586         if (IS_ERR(p))
587                 panic("failed fork for CPU %u: %li", cpu, PTR_ERR(p));
588         current_set[cpu] = p;
589         spin_lock_init(&(&per_cpu(s390_idle, cpu))->lock);
590 }
591
592 /* Upping and downing of CPUs */
593 int __cpu_up(unsigned int cpu)
594 {
595         struct task_struct *idle;
596         struct _lowcore *cpu_lowcore;
597         struct stack_frame *sf;
598         sigp_ccode ccode;
599
600         if (smp_cpu_state[cpu] != CPU_STATE_CONFIGURED)
601                 return -EIO;
602
603         ccode = signal_processor_p((__u32)(unsigned long)(lowcore_ptr[cpu]),
604                                    cpu, sigp_set_prefix);
605         if (ccode) {
606                 printk("sigp_set_prefix failed for cpu %d "
607                        "with condition code %d\n",
608                        (int) cpu, (int) ccode);
609                 return -EIO;
610         }
611
612         idle = current_set[cpu];
613         cpu_lowcore = lowcore_ptr[cpu];
614         cpu_lowcore->kernel_stack = (unsigned long)
615                 task_stack_page(idle) + THREAD_SIZE;
616         sf = (struct stack_frame *) (cpu_lowcore->kernel_stack
617                                      - sizeof(struct pt_regs)
618                                      - sizeof(struct stack_frame));
619         memset(sf, 0, sizeof(struct stack_frame));
620         sf->gprs[9] = (unsigned long) sf;
621         cpu_lowcore->save_area[15] = (unsigned long) sf;
622         __ctl_store(cpu_lowcore->cregs_save_area[0], 0, 15);
623         asm volatile(
624                 "       stam    0,15,0(%0)"
625                 : : "a" (&cpu_lowcore->access_regs_save_area) : "memory");
626         cpu_lowcore->percpu_offset = __per_cpu_offset[cpu];
627         cpu_lowcore->current_task = (unsigned long) idle;
628         cpu_lowcore->cpu_data.cpu_nr = cpu;
629         eieio();
630
631         while (signal_processor(cpu, sigp_restart) == sigp_busy)
632                 udelay(10);
633
634         while (!cpu_online(cpu))
635                 cpu_relax();
636         return 0;
637 }
638
639 static int __init setup_possible_cpus(char *s)
640 {
641         int pcpus, cpu;
642
643         pcpus = simple_strtoul(s, NULL, 0);
644         cpu_possible_map = cpumask_of_cpu(0);
645         for (cpu = 1; cpu < pcpus && cpu < NR_CPUS; cpu++)
646                 cpu_set(cpu, cpu_possible_map);
647         return 0;
648 }
649 early_param("possible_cpus", setup_possible_cpus);
650
651 #ifdef CONFIG_HOTPLUG_CPU
652
653 int __cpu_disable(void)
654 {
655         struct ec_creg_mask_parms cr_parms;
656         int cpu = smp_processor_id();
657
658         cpu_clear(cpu, cpu_online_map);
659
660         /* Disable pfault pseudo page faults on this cpu. */
661         pfault_fini();
662
663         memset(&cr_parms.orvals, 0, sizeof(cr_parms.orvals));
664         memset(&cr_parms.andvals, 0xff, sizeof(cr_parms.andvals));
665
666         /* disable all external interrupts */
667         cr_parms.orvals[0] = 0;
668         cr_parms.andvals[0] = ~(1 << 15 | 1 << 14 | 1 << 13 | 1 << 12 |
669                                 1 << 11 | 1 << 10 | 1 <<  6 | 1 <<  4);
670         /* disable all I/O interrupts */
671         cr_parms.orvals[6] = 0;
672         cr_parms.andvals[6] = ~(1 << 31 | 1 << 30 | 1 << 29 | 1 << 28 |
673                                 1 << 27 | 1 << 26 | 1 << 25 | 1 << 24);
674         /* disable most machine checks */
675         cr_parms.orvals[14] = 0;
676         cr_parms.andvals[14] = ~(1 << 28 | 1 << 27 | 1 << 26 |
677                                  1 << 25 | 1 << 24);
678
679         smp_ctl_bit_callback(&cr_parms);
680
681         return 0;
682 }
683
684 void __cpu_die(unsigned int cpu)
685 {
686         /* Wait until target cpu is down */
687         while (!smp_cpu_not_running(cpu))
688                 cpu_relax();
689         printk(KERN_INFO "Processor %d spun down\n", cpu);
690 }
691
692 void cpu_die(void)
693 {
694         idle_task_exit();
695         signal_processor(smp_processor_id(), sigp_stop);
696         BUG();
697         for (;;);
698 }
699
700 #endif /* CONFIG_HOTPLUG_CPU */
701
702 /*
703  *      Cycle through the processors and setup structures.
704  */
705
706 void __init smp_prepare_cpus(unsigned int max_cpus)
707 {
708         unsigned long stack;
709         unsigned int cpu;
710         int i;
711
712         smp_detect_cpus();
713
714         /* request the 0x1201 emergency signal external interrupt */
715         if (register_external_interrupt(0x1201, do_ext_call_interrupt) != 0)
716                 panic("Couldn't request external interrupt 0x1201");
717         memset(lowcore_ptr, 0, sizeof(lowcore_ptr));
718         /*
719          *  Initialize prefix pages and stacks for all possible cpus
720          */
721         print_cpu_info(&S390_lowcore.cpu_data);
722
723         for_each_possible_cpu(i) {
724                 lowcore_ptr[i] = (struct _lowcore *)
725                         __get_free_pages(GFP_KERNEL | GFP_DMA,
726                                          sizeof(void*) == 8 ? 1 : 0);
727                 stack = __get_free_pages(GFP_KERNEL, ASYNC_ORDER);
728                 if (!lowcore_ptr[i] || !stack)
729                         panic("smp_boot_cpus failed to allocate memory\n");
730
731                 *(lowcore_ptr[i]) = S390_lowcore;
732                 lowcore_ptr[i]->async_stack = stack + ASYNC_SIZE;
733                 stack = __get_free_pages(GFP_KERNEL, 0);
734                 if (!stack)
735                         panic("smp_boot_cpus failed to allocate memory\n");
736                 lowcore_ptr[i]->panic_stack = stack + PAGE_SIZE;
737 #ifndef CONFIG_64BIT
738                 if (MACHINE_HAS_IEEE) {
739                         lowcore_ptr[i]->extended_save_area_addr =
740                                 (__u32) __get_free_pages(GFP_KERNEL, 0);
741                         if (!lowcore_ptr[i]->extended_save_area_addr)
742                                 panic("smp_boot_cpus failed to "
743                                       "allocate memory\n");
744                 }
745 #endif
746         }
747 #ifndef CONFIG_64BIT
748         if (MACHINE_HAS_IEEE)
749                 ctl_set_bit(14, 29); /* enable extended save area */
750 #endif
751         set_prefix((u32)(unsigned long) lowcore_ptr[smp_processor_id()]);
752
753         for_each_possible_cpu(cpu)
754                 if (cpu != smp_processor_id())
755                         smp_create_idle(cpu);
756 }
757
758 void __init smp_prepare_boot_cpu(void)
759 {
760         BUG_ON(smp_processor_id() != 0);
761
762         current_thread_info()->cpu = 0;
763         cpu_set(0, cpu_present_map);
764         cpu_set(0, cpu_online_map);
765         S390_lowcore.percpu_offset = __per_cpu_offset[0];
766         current_set[0] = current;
767         smp_cpu_state[0] = CPU_STATE_CONFIGURED;
768         spin_lock_init(&(&__get_cpu_var(s390_idle))->lock);
769 }
770
771 void __init smp_cpus_done(unsigned int max_cpus)
772 {
773 }
774
775 /*
776  * the frequency of the profiling timer can be changed
777  * by writing a multiplier value into /proc/profile.
778  *
779  * usually you want to run this on all CPUs ;)
780  */
781 int setup_profiling_timer(unsigned int multiplier)
782 {
783         return 0;
784 }
785
786 #ifdef CONFIG_HOTPLUG_CPU
787 static ssize_t cpu_configure_show(struct sys_device *dev, char *buf)
788 {
789         ssize_t count;
790
791         mutex_lock(&smp_cpu_state_mutex);
792         count = sprintf(buf, "%d\n", smp_cpu_state[dev->id]);
793         mutex_unlock(&smp_cpu_state_mutex);
794         return count;
795 }
796
797 static ssize_t cpu_configure_store(struct sys_device *dev, const char *buf,
798                                    size_t count)
799 {
800         int cpu = dev->id;
801         int val, rc;
802         char delim;
803
804         if (sscanf(buf, "%d %c", &val, &delim) != 1)
805                 return -EINVAL;
806         if (val != 0 && val != 1)
807                 return -EINVAL;
808
809         mutex_lock(&smp_cpu_state_mutex);
810         lock_cpu_hotplug();
811         rc = -EBUSY;
812         if (cpu_online(cpu))
813                 goto out;
814         rc = 0;
815         switch (val) {
816         case 0:
817                 if (smp_cpu_state[cpu] == CPU_STATE_CONFIGURED) {
818                         rc = sclp_cpu_deconfigure(__cpu_logical_map[cpu]);
819                         if (!rc)
820                                 smp_cpu_state[cpu] = CPU_STATE_STANDBY;
821                 }
822                 break;
823         case 1:
824                 if (smp_cpu_state[cpu] == CPU_STATE_STANDBY) {
825                         rc = sclp_cpu_configure(__cpu_logical_map[cpu]);
826                         if (!rc)
827                                 smp_cpu_state[cpu] = CPU_STATE_CONFIGURED;
828                 }
829                 break;
830         default:
831                 break;
832         }
833 out:
834         unlock_cpu_hotplug();
835         mutex_unlock(&smp_cpu_state_mutex);
836         return rc ? rc : count;
837 }
838 static SYSDEV_ATTR(configure, 0644, cpu_configure_show, cpu_configure_store);
839 #endif /* CONFIG_HOTPLUG_CPU */
840
841 static ssize_t show_cpu_address(struct sys_device *dev, char *buf)
842 {
843         return sprintf(buf, "%d\n", __cpu_logical_map[dev->id]);
844 }
845 static SYSDEV_ATTR(address, 0444, show_cpu_address, NULL);
846
847
848 static struct attribute *cpu_common_attrs[] = {
849 #ifdef CONFIG_HOTPLUG_CPU
850         &attr_configure.attr,
851 #endif
852         &attr_address.attr,
853         NULL,
854 };
855
856 static struct attribute_group cpu_common_attr_group = {
857         .attrs = cpu_common_attrs,
858 };
859
860 static ssize_t show_capability(struct sys_device *dev, char *buf)
861 {
862         unsigned int capability;
863         int rc;
864
865         rc = get_cpu_capability(&capability);
866         if (rc)
867                 return rc;
868         return sprintf(buf, "%u\n", capability);
869 }
870 static SYSDEV_ATTR(capability, 0444, show_capability, NULL);
871
872 static ssize_t show_idle_count(struct sys_device *dev, char *buf)
873 {
874         struct s390_idle_data *idle;
875         unsigned long long idle_count;
876
877         idle = &per_cpu(s390_idle, dev->id);
878         spin_lock_irq(&idle->lock);
879         idle_count = idle->idle_count;
880         spin_unlock_irq(&idle->lock);
881         return sprintf(buf, "%llu\n", idle_count);
882 }
883 static SYSDEV_ATTR(idle_count, 0444, show_idle_count, NULL);
884
885 static ssize_t show_idle_time(struct sys_device *dev, char *buf)
886 {
887         struct s390_idle_data *idle;
888         unsigned long long new_time;
889
890         idle = &per_cpu(s390_idle, dev->id);
891         spin_lock_irq(&idle->lock);
892         if (idle->in_idle) {
893                 new_time = get_clock();
894                 idle->idle_time += new_time - idle->idle_enter;
895                 idle->idle_enter = new_time;
896         }
897         new_time = idle->idle_time;
898         spin_unlock_irq(&idle->lock);
899         return sprintf(buf, "%llu\n", new_time >> 12);
900 }
901 static SYSDEV_ATTR(idle_time_us, 0444, show_idle_time, NULL);
902
903 static struct attribute *cpu_online_attrs[] = {
904         &attr_capability.attr,
905         &attr_idle_count.attr,
906         &attr_idle_time_us.attr,
907         NULL,
908 };
909
910 static struct attribute_group cpu_online_attr_group = {
911         .attrs = cpu_online_attrs,
912 };
913
914 static int __cpuinit smp_cpu_notify(struct notifier_block *self,
915                                     unsigned long action, void *hcpu)
916 {
917         unsigned int cpu = (unsigned int)(long)hcpu;
918         struct cpu *c = &per_cpu(cpu_devices, cpu);
919         struct sys_device *s = &c->sysdev;
920         struct s390_idle_data *idle;
921
922         switch (action) {
923         case CPU_ONLINE:
924         case CPU_ONLINE_FROZEN:
925                 idle = &per_cpu(s390_idle, cpu);
926                 spin_lock_irq(&idle->lock);
927                 idle->idle_enter = 0;
928                 idle->idle_time = 0;
929                 idle->idle_count = 0;
930                 spin_unlock_irq(&idle->lock);
931                 if (sysfs_create_group(&s->kobj, &cpu_online_attr_group))
932                         return NOTIFY_BAD;
933                 break;
934         case CPU_DEAD:
935         case CPU_DEAD_FROZEN:
936                 sysfs_remove_group(&s->kobj, &cpu_online_attr_group);
937                 break;
938         }
939         return NOTIFY_OK;
940 }
941
942 static struct notifier_block __cpuinitdata smp_cpu_nb = {
943         .notifier_call = smp_cpu_notify,
944 };
945
946 static int smp_add_present_cpu(int cpu)
947 {
948         struct cpu *c = &per_cpu(cpu_devices, cpu);
949         struct sys_device *s = &c->sysdev;
950         int rc;
951
952         c->hotpluggable = 1;
953         rc = register_cpu(c, cpu);
954         if (rc)
955                 goto out;
956         rc = sysfs_create_group(&s->kobj, &cpu_common_attr_group);
957         if (rc)
958                 goto out_cpu;
959         if (!cpu_online(cpu))
960                 goto out;
961         rc = sysfs_create_group(&s->kobj, &cpu_online_attr_group);
962         if (!rc)
963                 return 0;
964         sysfs_remove_group(&s->kobj, &cpu_common_attr_group);
965 out_cpu:
966 #ifdef CONFIG_HOTPLUG_CPU
967         unregister_cpu(c);
968 #endif
969 out:
970         return rc;
971 }
972
973 #ifdef CONFIG_HOTPLUG_CPU
974 static ssize_t rescan_store(struct sys_device *dev, const char *buf,
975                             size_t count)
976 {
977         cpumask_t newcpus;
978         int cpu;
979         int rc;
980
981         mutex_lock(&smp_cpu_state_mutex);
982         lock_cpu_hotplug();
983         newcpus = cpu_present_map;
984         rc = smp_rescan_cpus();
985         if (rc)
986                 goto out;
987         cpus_andnot(newcpus, cpu_present_map, newcpus);
988         for_each_cpu_mask(cpu, newcpus) {
989                 rc = smp_add_present_cpu(cpu);
990                 if (rc)
991                         cpu_clear(cpu, cpu_present_map);
992         }
993         rc = 0;
994 out:
995         unlock_cpu_hotplug();
996         mutex_unlock(&smp_cpu_state_mutex);
997         return rc ? rc : count;
998 }
999 static SYSDEV_ATTR(rescan, 0200, NULL, rescan_store);
1000 #endif /* CONFIG_HOTPLUG_CPU */
1001
1002 static int __init topology_init(void)
1003 {
1004         int cpu;
1005         int rc;
1006
1007         register_cpu_notifier(&smp_cpu_nb);
1008
1009 #ifdef CONFIG_HOTPLUG_CPU
1010         rc = sysfs_create_file(&cpu_sysdev_class.kset.kobj,
1011                                &attr_rescan.attr);
1012         if (rc)
1013                 return rc;
1014 #endif
1015         for_each_present_cpu(cpu) {
1016                 rc = smp_add_present_cpu(cpu);
1017                 if (rc)
1018                         return rc;
1019         }
1020         return 0;
1021 }
1022 subsys_initcall(topology_init);