Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/teigland/dlm
[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 #define KMSG_COMPONENT "cpu"
24 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
25
26 #include <linux/module.h>
27 #include <linux/init.h>
28 #include <linux/mm.h>
29 #include <linux/err.h>
30 #include <linux/spinlock.h>
31 #include <linux/kernel_stat.h>
32 #include <linux/delay.h>
33 #include <linux/cache.h>
34 #include <linux/interrupt.h>
35 #include <linux/cpu.h>
36 #include <linux/timex.h>
37 #include <linux/bootmem.h>
38 #include <asm/ipl.h>
39 #include <asm/setup.h>
40 #include <asm/sigp.h>
41 #include <asm/pgalloc.h>
42 #include <asm/irq.h>
43 #include <asm/s390_ext.h>
44 #include <asm/cpcmd.h>
45 #include <asm/tlbflush.h>
46 #include <asm/timer.h>
47 #include <asm/lowcore.h>
48 #include <asm/sclp.h>
49 #include <asm/cpu.h>
50 #include <asm/vdso.h>
51 #include "entry.h"
52
53 /*
54  * An array with a pointer the lowcore of every CPU.
55  */
56 struct _lowcore *lowcore_ptr[NR_CPUS];
57 EXPORT_SYMBOL(lowcore_ptr);
58
59 static struct task_struct *current_set[NR_CPUS];
60
61 static u8 smp_cpu_type;
62 static int smp_use_sigp_detection;
63
64 enum s390_cpu_state {
65         CPU_STATE_STANDBY,
66         CPU_STATE_CONFIGURED,
67 };
68
69 DEFINE_MUTEX(smp_cpu_state_mutex);
70 int smp_cpu_polarization[NR_CPUS];
71 static int smp_cpu_state[NR_CPUS];
72 static int cpu_management;
73
74 static DEFINE_PER_CPU(struct cpu, cpu_devices);
75
76 static void smp_ext_bitcall(int, ec_bit_sig);
77
78 void smp_send_stop(void)
79 {
80         int cpu, rc;
81
82         /* Disable all interrupts/machine checks */
83         __load_psw_mask(psw_kernel_bits & ~PSW_MASK_MCHECK);
84
85         /* write magic number to zero page (absolute 0) */
86         lowcore_ptr[smp_processor_id()]->panic_magic = __PANIC_MAGIC;
87
88         /* stop all processors */
89         for_each_online_cpu(cpu) {
90                 if (cpu == smp_processor_id())
91                         continue;
92                 do {
93                         rc = signal_processor(cpu, sigp_stop);
94                 } while (rc == sigp_busy);
95
96                 while (!smp_cpu_not_running(cpu))
97                         cpu_relax();
98         }
99 }
100
101 /*
102  * This is the main routine where commands issued by other
103  * cpus are handled.
104  */
105
106 static void do_ext_call_interrupt(__u16 code)
107 {
108         unsigned long bits;
109
110         /*
111          * handle bit signal external calls
112          *
113          * For the ec_schedule signal we have to do nothing. All the work
114          * is done automatically when we return from the interrupt.
115          */
116         bits = xchg(&S390_lowcore.ext_call_fast, 0);
117
118         if (test_bit(ec_call_function, &bits))
119                 generic_smp_call_function_interrupt();
120
121         if (test_bit(ec_call_function_single, &bits))
122                 generic_smp_call_function_single_interrupt();
123 }
124
125 /*
126  * Send an external call sigp to another cpu and return without waiting
127  * for its completion.
128  */
129 static void smp_ext_bitcall(int cpu, ec_bit_sig sig)
130 {
131         /*
132          * Set signaling bit in lowcore of target cpu and kick it
133          */
134         set_bit(sig, (unsigned long *) &lowcore_ptr[cpu]->ext_call_fast);
135         while (signal_processor(cpu, sigp_emergency_signal) == sigp_busy)
136                 udelay(10);
137 }
138
139 void arch_send_call_function_ipi(cpumask_t mask)
140 {
141         int cpu;
142
143         for_each_cpu_mask(cpu, mask)
144                 smp_ext_bitcall(cpu, ec_call_function);
145 }
146
147 void arch_send_call_function_single_ipi(int cpu)
148 {
149         smp_ext_bitcall(cpu, ec_call_function_single);
150 }
151
152 #ifndef CONFIG_64BIT
153 /*
154  * this function sends a 'purge tlb' signal to another CPU.
155  */
156 static void smp_ptlb_callback(void *info)
157 {
158         __tlb_flush_local();
159 }
160
161 void smp_ptlb_all(void)
162 {
163         on_each_cpu(smp_ptlb_callback, NULL, 1);
164 }
165 EXPORT_SYMBOL(smp_ptlb_all);
166 #endif /* ! CONFIG_64BIT */
167
168 /*
169  * this function sends a 'reschedule' IPI to another CPU.
170  * it goes straight through and wastes no time serializing
171  * anything. Worst case is that we lose a reschedule ...
172  */
173 void smp_send_reschedule(int cpu)
174 {
175         smp_ext_bitcall(cpu, ec_schedule);
176 }
177
178 /*
179  * parameter area for the set/clear control bit callbacks
180  */
181 struct ec_creg_mask_parms {
182         unsigned long orvals[16];
183         unsigned long andvals[16];
184 };
185
186 /*
187  * callback for setting/clearing control bits
188  */
189 static void smp_ctl_bit_callback(void *info)
190 {
191         struct ec_creg_mask_parms *pp = info;
192         unsigned long cregs[16];
193         int i;
194
195         __ctl_store(cregs, 0, 15);
196         for (i = 0; i <= 15; i++)
197                 cregs[i] = (cregs[i] & pp->andvals[i]) | pp->orvals[i];
198         __ctl_load(cregs, 0, 15);
199 }
200
201 /*
202  * Set a bit in a control register of all cpus
203  */
204 void smp_ctl_set_bit(int cr, int bit)
205 {
206         struct ec_creg_mask_parms parms;
207
208         memset(&parms.orvals, 0, sizeof(parms.orvals));
209         memset(&parms.andvals, 0xff, sizeof(parms.andvals));
210         parms.orvals[cr] = 1 << bit;
211         on_each_cpu(smp_ctl_bit_callback, &parms, 1);
212 }
213 EXPORT_SYMBOL(smp_ctl_set_bit);
214
215 /*
216  * Clear a bit in a control register of all cpus
217  */
218 void smp_ctl_clear_bit(int cr, int bit)
219 {
220         struct ec_creg_mask_parms parms;
221
222         memset(&parms.orvals, 0, sizeof(parms.orvals));
223         memset(&parms.andvals, 0xff, sizeof(parms.andvals));
224         parms.andvals[cr] = ~(1L << bit);
225         on_each_cpu(smp_ctl_bit_callback, &parms, 1);
226 }
227 EXPORT_SYMBOL(smp_ctl_clear_bit);
228
229 /*
230  * In early ipl state a temp. logically cpu number is needed, so the sigp
231  * functions can be used to sense other cpus. Since NR_CPUS is >= 2 on
232  * CONFIG_SMP and the ipl cpu is logical cpu 0, it must be 1.
233  */
234 #define CPU_INIT_NO     1
235
236 #if defined(CONFIG_ZFCPDUMP) || defined(CONFIG_ZFCPDUMP_MODULE)
237
238 /*
239  * zfcpdump_prefix_array holds prefix registers for the following scenario:
240  * 64 bit zfcpdump kernel and 31 bit kernel which is to be dumped. We have to
241  * save its prefix registers, since they get lost, when switching from 31 bit
242  * to 64 bit.
243  */
244 unsigned int zfcpdump_prefix_array[NR_CPUS + 1] \
245         __attribute__((__section__(".data")));
246
247 static void __init smp_get_save_area(unsigned int cpu, unsigned int phy_cpu)
248 {
249         if (ipl_info.type != IPL_TYPE_FCP_DUMP)
250                 return;
251         if (cpu >= NR_CPUS) {
252                 pr_warning("CPU %i exceeds the maximum %i and is excluded from "
253                            "the dump\n", cpu, NR_CPUS - 1);
254                 return;
255         }
256         zfcpdump_save_areas[cpu] = kmalloc(sizeof(union save_area), GFP_KERNEL);
257         __cpu_logical_map[CPU_INIT_NO] = (__u16) phy_cpu;
258         while (signal_processor(CPU_INIT_NO, sigp_stop_and_store_status) ==
259                sigp_busy)
260                 cpu_relax();
261         memcpy(zfcpdump_save_areas[cpu],
262                (void *)(unsigned long) store_prefix() + SAVE_AREA_BASE,
263                SAVE_AREA_SIZE);
264 #ifdef CONFIG_64BIT
265         /* copy original prefix register */
266         zfcpdump_save_areas[cpu]->s390x.pref_reg = zfcpdump_prefix_array[cpu];
267 #endif
268 }
269
270 union save_area *zfcpdump_save_areas[NR_CPUS + 1];
271 EXPORT_SYMBOL_GPL(zfcpdump_save_areas);
272
273 #else
274
275 static inline void smp_get_save_area(unsigned int cpu, unsigned int phy_cpu) { }
276
277 #endif /* CONFIG_ZFCPDUMP || CONFIG_ZFCPDUMP_MODULE */
278
279 static int cpu_stopped(int cpu)
280 {
281         __u32 status;
282
283         /* Check for stopped state */
284         if (signal_processor_ps(&status, 0, cpu, sigp_sense) ==
285             sigp_status_stored) {
286                 if (status & 0x40)
287                         return 1;
288         }
289         return 0;
290 }
291
292 static int cpu_known(int cpu_id)
293 {
294         int cpu;
295
296         for_each_present_cpu(cpu) {
297                 if (__cpu_logical_map[cpu] == cpu_id)
298                         return 1;
299         }
300         return 0;
301 }
302
303 static int smp_rescan_cpus_sigp(cpumask_t avail)
304 {
305         int cpu_id, logical_cpu;
306
307         logical_cpu = first_cpu(avail);
308         if (logical_cpu == NR_CPUS)
309                 return 0;
310         for (cpu_id = 0; cpu_id <= 65535; cpu_id++) {
311                 if (cpu_known(cpu_id))
312                         continue;
313                 __cpu_logical_map[logical_cpu] = cpu_id;
314                 smp_cpu_polarization[logical_cpu] = POLARIZATION_UNKNWN;
315                 if (!cpu_stopped(logical_cpu))
316                         continue;
317                 cpu_set(logical_cpu, cpu_present_map);
318                 smp_cpu_state[logical_cpu] = CPU_STATE_CONFIGURED;
319                 logical_cpu = next_cpu(logical_cpu, avail);
320                 if (logical_cpu == NR_CPUS)
321                         break;
322         }
323         return 0;
324 }
325
326 static int smp_rescan_cpus_sclp(cpumask_t avail)
327 {
328         struct sclp_cpu_info *info;
329         int cpu_id, logical_cpu, cpu;
330         int rc;
331
332         logical_cpu = first_cpu(avail);
333         if (logical_cpu == NR_CPUS)
334                 return 0;
335         info = kmalloc(sizeof(*info), GFP_KERNEL);
336         if (!info)
337                 return -ENOMEM;
338         rc = sclp_get_cpu_info(info);
339         if (rc)
340                 goto out;
341         for (cpu = 0; cpu < info->combined; cpu++) {
342                 if (info->has_cpu_type && info->cpu[cpu].type != smp_cpu_type)
343                         continue;
344                 cpu_id = info->cpu[cpu].address;
345                 if (cpu_known(cpu_id))
346                         continue;
347                 __cpu_logical_map[logical_cpu] = cpu_id;
348                 smp_cpu_polarization[logical_cpu] = POLARIZATION_UNKNWN;
349                 cpu_set(logical_cpu, cpu_present_map);
350                 if (cpu >= info->configured)
351                         smp_cpu_state[logical_cpu] = CPU_STATE_STANDBY;
352                 else
353                         smp_cpu_state[logical_cpu] = CPU_STATE_CONFIGURED;
354                 logical_cpu = next_cpu(logical_cpu, avail);
355                 if (logical_cpu == NR_CPUS)
356                         break;
357         }
358 out:
359         kfree(info);
360         return rc;
361 }
362
363 static int __smp_rescan_cpus(void)
364 {
365         cpumask_t avail;
366
367         cpus_xor(avail, cpu_possible_map, cpu_present_map);
368         if (smp_use_sigp_detection)
369                 return smp_rescan_cpus_sigp(avail);
370         else
371                 return smp_rescan_cpus_sclp(avail);
372 }
373
374 static void __init smp_detect_cpus(void)
375 {
376         unsigned int cpu, c_cpus, s_cpus;
377         struct sclp_cpu_info *info;
378         u16 boot_cpu_addr, cpu_addr;
379
380         c_cpus = 1;
381         s_cpus = 0;
382         boot_cpu_addr = S390_lowcore.cpu_data.cpu_addr;
383         info = kmalloc(sizeof(*info), GFP_KERNEL);
384         if (!info)
385                 panic("smp_detect_cpus failed to allocate memory\n");
386         /* Use sigp detection algorithm if sclp doesn't work. */
387         if (sclp_get_cpu_info(info)) {
388                 smp_use_sigp_detection = 1;
389                 for (cpu = 0; cpu <= 65535; cpu++) {
390                         if (cpu == boot_cpu_addr)
391                                 continue;
392                         __cpu_logical_map[CPU_INIT_NO] = cpu;
393                         if (!cpu_stopped(CPU_INIT_NO))
394                                 continue;
395                         smp_get_save_area(c_cpus, cpu);
396                         c_cpus++;
397                 }
398                 goto out;
399         }
400
401         if (info->has_cpu_type) {
402                 for (cpu = 0; cpu < info->combined; cpu++) {
403                         if (info->cpu[cpu].address == boot_cpu_addr) {
404                                 smp_cpu_type = info->cpu[cpu].type;
405                                 break;
406                         }
407                 }
408         }
409
410         for (cpu = 0; cpu < info->combined; cpu++) {
411                 if (info->has_cpu_type && info->cpu[cpu].type != smp_cpu_type)
412                         continue;
413                 cpu_addr = info->cpu[cpu].address;
414                 if (cpu_addr == boot_cpu_addr)
415                         continue;
416                 __cpu_logical_map[CPU_INIT_NO] = cpu_addr;
417                 if (!cpu_stopped(CPU_INIT_NO)) {
418                         s_cpus++;
419                         continue;
420                 }
421                 smp_get_save_area(c_cpus, cpu_addr);
422                 c_cpus++;
423         }
424 out:
425         kfree(info);
426         pr_info("%d configured CPUs, %d standby CPUs\n", c_cpus, s_cpus);
427         get_online_cpus();
428         __smp_rescan_cpus();
429         put_online_cpus();
430 }
431
432 /*
433  *      Activate a secondary processor.
434  */
435 int __cpuinit start_secondary(void *cpuvoid)
436 {
437         /* Setup the cpu */
438         cpu_init();
439         preempt_disable();
440         /* Enable TOD clock interrupts on the secondary cpu. */
441         init_cpu_timer();
442         /* Enable cpu timer interrupts on the secondary cpu. */
443         init_cpu_vtimer();
444         /* Enable pfault pseudo page faults on this cpu. */
445         pfault_init();
446
447         /* call cpu notifiers */
448         notify_cpu_starting(smp_processor_id());
449         /* Mark this cpu as online */
450         ipi_call_lock();
451         cpu_set(smp_processor_id(), cpu_online_map);
452         ipi_call_unlock();
453         /* Switch on interrupts */
454         local_irq_enable();
455         /* Print info about this processor */
456         print_cpu_info(&S390_lowcore.cpu_data);
457         /* cpu_idle will call schedule for us */
458         cpu_idle();
459         return 0;
460 }
461
462 static void __init smp_create_idle(unsigned int cpu)
463 {
464         struct task_struct *p;
465
466         /*
467          *  don't care about the psw and regs settings since we'll never
468          *  reschedule the forked task.
469          */
470         p = fork_idle(cpu);
471         if (IS_ERR(p))
472                 panic("failed fork for CPU %u: %li", cpu, PTR_ERR(p));
473         current_set[cpu] = p;
474 }
475
476 static int __cpuinit smp_alloc_lowcore(int cpu)
477 {
478         unsigned long async_stack, panic_stack;
479         struct _lowcore *lowcore;
480         int lc_order;
481
482         lc_order = sizeof(long) == 8 ? 1 : 0;
483         lowcore = (void *) __get_free_pages(GFP_KERNEL | GFP_DMA, lc_order);
484         if (!lowcore)
485                 return -ENOMEM;
486         async_stack = __get_free_pages(GFP_KERNEL, ASYNC_ORDER);
487         panic_stack = __get_free_page(GFP_KERNEL);
488         if (!panic_stack || !async_stack)
489                 goto out;
490         memcpy(lowcore, &S390_lowcore, 512);
491         memset((char *)lowcore + 512, 0, sizeof(*lowcore) - 512);
492         lowcore->async_stack = async_stack + ASYNC_SIZE;
493         lowcore->panic_stack = panic_stack + PAGE_SIZE;
494
495 #ifndef CONFIG_64BIT
496         if (MACHINE_HAS_IEEE) {
497                 unsigned long save_area;
498
499                 save_area = get_zeroed_page(GFP_KERNEL);
500                 if (!save_area)
501                         goto out;
502                 lowcore->extended_save_area_addr = (u32) save_area;
503         }
504 #else
505         if (vdso_alloc_per_cpu(cpu, lowcore))
506                 goto out;
507 #endif
508         lowcore_ptr[cpu] = lowcore;
509         return 0;
510
511 out:
512         free_page(panic_stack);
513         free_pages(async_stack, ASYNC_ORDER);
514         free_pages((unsigned long) lowcore, lc_order);
515         return -ENOMEM;
516 }
517
518 #ifdef CONFIG_HOTPLUG_CPU
519 static void smp_free_lowcore(int cpu)
520 {
521         struct _lowcore *lowcore;
522         int lc_order;
523
524         lc_order = sizeof(long) == 8 ? 1 : 0;
525         lowcore = lowcore_ptr[cpu];
526 #ifndef CONFIG_64BIT
527         if (MACHINE_HAS_IEEE)
528                 free_page((unsigned long) lowcore->extended_save_area_addr);
529 #else
530         vdso_free_per_cpu(cpu, lowcore);
531 #endif
532         free_page(lowcore->panic_stack - PAGE_SIZE);
533         free_pages(lowcore->async_stack - ASYNC_SIZE, ASYNC_ORDER);
534         free_pages((unsigned long) lowcore, lc_order);
535         lowcore_ptr[cpu] = NULL;
536 }
537 #endif /* CONFIG_HOTPLUG_CPU */
538
539 /* Upping and downing of CPUs */
540 int __cpuinit __cpu_up(unsigned int cpu)
541 {
542         struct task_struct *idle;
543         struct _lowcore *cpu_lowcore;
544         struct stack_frame *sf;
545         sigp_ccode ccode;
546
547         if (smp_cpu_state[cpu] != CPU_STATE_CONFIGURED)
548                 return -EIO;
549         if (smp_alloc_lowcore(cpu))
550                 return -ENOMEM;
551
552         ccode = signal_processor_p((__u32)(unsigned long)(lowcore_ptr[cpu]),
553                                    cpu, sigp_set_prefix);
554         if (ccode)
555                 return -EIO;
556
557         idle = current_set[cpu];
558         cpu_lowcore = lowcore_ptr[cpu];
559         cpu_lowcore->kernel_stack = (unsigned long)
560                 task_stack_page(idle) + THREAD_SIZE;
561         cpu_lowcore->thread_info = (unsigned long) task_thread_info(idle);
562         sf = (struct stack_frame *) (cpu_lowcore->kernel_stack
563                                      - sizeof(struct pt_regs)
564                                      - sizeof(struct stack_frame));
565         memset(sf, 0, sizeof(struct stack_frame));
566         sf->gprs[9] = (unsigned long) sf;
567         cpu_lowcore->save_area[15] = (unsigned long) sf;
568         __ctl_store(cpu_lowcore->cregs_save_area, 0, 15);
569         asm volatile(
570                 "       stam    0,15,0(%0)"
571                 : : "a" (&cpu_lowcore->access_regs_save_area) : "memory");
572         cpu_lowcore->percpu_offset = __per_cpu_offset[cpu];
573         cpu_lowcore->current_task = (unsigned long) idle;
574         cpu_lowcore->cpu_data.cpu_nr = cpu;
575         cpu_lowcore->kernel_asce = S390_lowcore.kernel_asce;
576         cpu_lowcore->ipl_device = S390_lowcore.ipl_device;
577         eieio();
578
579         while (signal_processor(cpu, sigp_restart) == sigp_busy)
580                 udelay(10);
581
582         while (!cpu_online(cpu))
583                 cpu_relax();
584         return 0;
585 }
586
587 static int __init setup_possible_cpus(char *s)
588 {
589         int pcpus, cpu;
590
591         pcpus = simple_strtoul(s, NULL, 0);
592         cpu_possible_map = cpumask_of_cpu(0);
593         for (cpu = 1; cpu < pcpus && cpu < NR_CPUS; cpu++)
594                 cpu_set(cpu, cpu_possible_map);
595         return 0;
596 }
597 early_param("possible_cpus", setup_possible_cpus);
598
599 #ifdef CONFIG_HOTPLUG_CPU
600
601 int __cpu_disable(void)
602 {
603         struct ec_creg_mask_parms cr_parms;
604         int cpu = smp_processor_id();
605
606         cpu_clear(cpu, cpu_online_map);
607
608         /* Disable pfault pseudo page faults on this cpu. */
609         pfault_fini();
610
611         memset(&cr_parms.orvals, 0, sizeof(cr_parms.orvals));
612         memset(&cr_parms.andvals, 0xff, sizeof(cr_parms.andvals));
613
614         /* disable all external interrupts */
615         cr_parms.orvals[0] = 0;
616         cr_parms.andvals[0] = ~(1 << 15 | 1 << 14 | 1 << 13 | 1 << 12 |
617                                 1 << 11 | 1 << 10 | 1 <<  6 | 1 <<  4);
618         /* disable all I/O interrupts */
619         cr_parms.orvals[6] = 0;
620         cr_parms.andvals[6] = ~(1 << 31 | 1 << 30 | 1 << 29 | 1 << 28 |
621                                 1 << 27 | 1 << 26 | 1 << 25 | 1 << 24);
622         /* disable most machine checks */
623         cr_parms.orvals[14] = 0;
624         cr_parms.andvals[14] = ~(1 << 28 | 1 << 27 | 1 << 26 |
625                                  1 << 25 | 1 << 24);
626
627         smp_ctl_bit_callback(&cr_parms);
628
629         return 0;
630 }
631
632 void __cpu_die(unsigned int cpu)
633 {
634         /* Wait until target cpu is down */
635         while (!smp_cpu_not_running(cpu))
636                 cpu_relax();
637         smp_free_lowcore(cpu);
638         pr_info("Processor %d stopped\n", cpu);
639 }
640
641 void cpu_die(void)
642 {
643         idle_task_exit();
644         signal_processor(smp_processor_id(), sigp_stop);
645         BUG();
646         for (;;);
647 }
648
649 #endif /* CONFIG_HOTPLUG_CPU */
650
651 void __init smp_prepare_cpus(unsigned int max_cpus)
652 {
653 #ifndef CONFIG_64BIT
654         unsigned long save_area = 0;
655 #endif
656         unsigned long async_stack, panic_stack;
657         struct _lowcore *lowcore;
658         unsigned int cpu;
659         int lc_order;
660
661         smp_detect_cpus();
662
663         /* request the 0x1201 emergency signal external interrupt */
664         if (register_external_interrupt(0x1201, do_ext_call_interrupt) != 0)
665                 panic("Couldn't request external interrupt 0x1201");
666         print_cpu_info(&S390_lowcore.cpu_data);
667
668         /* Reallocate current lowcore, but keep its contents. */
669         lc_order = sizeof(long) == 8 ? 1 : 0;
670         lowcore = (void *) __get_free_pages(GFP_KERNEL | GFP_DMA, lc_order);
671         panic_stack = __get_free_page(GFP_KERNEL);
672         async_stack = __get_free_pages(GFP_KERNEL, ASYNC_ORDER);
673         BUG_ON(!lowcore || !panic_stack || !async_stack);
674 #ifndef CONFIG_64BIT
675         if (MACHINE_HAS_IEEE)
676                 save_area = get_zeroed_page(GFP_KERNEL);
677 #endif
678         local_irq_disable();
679         local_mcck_disable();
680         lowcore_ptr[smp_processor_id()] = lowcore;
681         *lowcore = S390_lowcore;
682         lowcore->panic_stack = panic_stack + PAGE_SIZE;
683         lowcore->async_stack = async_stack + ASYNC_SIZE;
684 #ifndef CONFIG_64BIT
685         if (MACHINE_HAS_IEEE)
686                 lowcore->extended_save_area_addr = (u32) save_area;
687 #else
688         BUG_ON(vdso_alloc_per_cpu(smp_processor_id(), lowcore));
689 #endif
690         set_prefix((u32)(unsigned long) lowcore);
691         local_mcck_enable();
692         local_irq_enable();
693         for_each_possible_cpu(cpu)
694                 if (cpu != smp_processor_id())
695                         smp_create_idle(cpu);
696 }
697
698 void __init smp_prepare_boot_cpu(void)
699 {
700         BUG_ON(smp_processor_id() != 0);
701
702         current_thread_info()->cpu = 0;
703         cpu_set(0, cpu_present_map);
704         cpu_set(0, cpu_online_map);
705         S390_lowcore.percpu_offset = __per_cpu_offset[0];
706         current_set[0] = current;
707         smp_cpu_state[0] = CPU_STATE_CONFIGURED;
708         smp_cpu_polarization[0] = POLARIZATION_UNKNWN;
709 }
710
711 void __init smp_cpus_done(unsigned int max_cpus)
712 {
713 }
714
715 /*
716  * the frequency of the profiling timer can be changed
717  * by writing a multiplier value into /proc/profile.
718  *
719  * usually you want to run this on all CPUs ;)
720  */
721 int setup_profiling_timer(unsigned int multiplier)
722 {
723         return 0;
724 }
725
726 #ifdef CONFIG_HOTPLUG_CPU
727 static ssize_t cpu_configure_show(struct sys_device *dev,
728                                 struct sysdev_attribute *attr, char *buf)
729 {
730         ssize_t count;
731
732         mutex_lock(&smp_cpu_state_mutex);
733         count = sprintf(buf, "%d\n", smp_cpu_state[dev->id]);
734         mutex_unlock(&smp_cpu_state_mutex);
735         return count;
736 }
737
738 static ssize_t cpu_configure_store(struct sys_device *dev,
739                                   struct sysdev_attribute *attr,
740                                   const char *buf, size_t count)
741 {
742         int cpu = dev->id;
743         int val, rc;
744         char delim;
745
746         if (sscanf(buf, "%d %c", &val, &delim) != 1)
747                 return -EINVAL;
748         if (val != 0 && val != 1)
749                 return -EINVAL;
750
751         get_online_cpus();
752         mutex_lock(&smp_cpu_state_mutex);
753         rc = -EBUSY;
754         if (cpu_online(cpu))
755                 goto out;
756         rc = 0;
757         switch (val) {
758         case 0:
759                 if (smp_cpu_state[cpu] == CPU_STATE_CONFIGURED) {
760                         rc = sclp_cpu_deconfigure(__cpu_logical_map[cpu]);
761                         if (!rc) {
762                                 smp_cpu_state[cpu] = CPU_STATE_STANDBY;
763                                 smp_cpu_polarization[cpu] = POLARIZATION_UNKNWN;
764                         }
765                 }
766                 break;
767         case 1:
768                 if (smp_cpu_state[cpu] == CPU_STATE_STANDBY) {
769                         rc = sclp_cpu_configure(__cpu_logical_map[cpu]);
770                         if (!rc) {
771                                 smp_cpu_state[cpu] = CPU_STATE_CONFIGURED;
772                                 smp_cpu_polarization[cpu] = POLARIZATION_UNKNWN;
773                         }
774                 }
775                 break;
776         default:
777                 break;
778         }
779 out:
780         mutex_unlock(&smp_cpu_state_mutex);
781         put_online_cpus();
782         return rc ? rc : count;
783 }
784 static SYSDEV_ATTR(configure, 0644, cpu_configure_show, cpu_configure_store);
785 #endif /* CONFIG_HOTPLUG_CPU */
786
787 static ssize_t cpu_polarization_show(struct sys_device *dev,
788                                      struct sysdev_attribute *attr, char *buf)
789 {
790         int cpu = dev->id;
791         ssize_t count;
792
793         mutex_lock(&smp_cpu_state_mutex);
794         switch (smp_cpu_polarization[cpu]) {
795         case POLARIZATION_HRZ:
796                 count = sprintf(buf, "horizontal\n");
797                 break;
798         case POLARIZATION_VL:
799                 count = sprintf(buf, "vertical:low\n");
800                 break;
801         case POLARIZATION_VM:
802                 count = sprintf(buf, "vertical:medium\n");
803                 break;
804         case POLARIZATION_VH:
805                 count = sprintf(buf, "vertical:high\n");
806                 break;
807         default:
808                 count = sprintf(buf, "unknown\n");
809                 break;
810         }
811         mutex_unlock(&smp_cpu_state_mutex);
812         return count;
813 }
814 static SYSDEV_ATTR(polarization, 0444, cpu_polarization_show, NULL);
815
816 static ssize_t show_cpu_address(struct sys_device *dev,
817                                 struct sysdev_attribute *attr, char *buf)
818 {
819         return sprintf(buf, "%d\n", __cpu_logical_map[dev->id]);
820 }
821 static SYSDEV_ATTR(address, 0444, show_cpu_address, NULL);
822
823
824 static struct attribute *cpu_common_attrs[] = {
825 #ifdef CONFIG_HOTPLUG_CPU
826         &attr_configure.attr,
827 #endif
828         &attr_address.attr,
829         &attr_polarization.attr,
830         NULL,
831 };
832
833 static struct attribute_group cpu_common_attr_group = {
834         .attrs = cpu_common_attrs,
835 };
836
837 static ssize_t show_capability(struct sys_device *dev,
838                                 struct sysdev_attribute *attr, char *buf)
839 {
840         unsigned int capability;
841         int rc;
842
843         rc = get_cpu_capability(&capability);
844         if (rc)
845                 return rc;
846         return sprintf(buf, "%u\n", capability);
847 }
848 static SYSDEV_ATTR(capability, 0444, show_capability, NULL);
849
850 static ssize_t show_idle_count(struct sys_device *dev,
851                                 struct sysdev_attribute *attr, char *buf)
852 {
853         struct s390_idle_data *idle;
854         unsigned long long idle_count;
855
856         idle = &per_cpu(s390_idle, dev->id);
857         spin_lock(&idle->lock);
858         idle_count = idle->idle_count;
859         if (idle->idle_enter)
860                 idle_count++;
861         spin_unlock(&idle->lock);
862         return sprintf(buf, "%llu\n", idle_count);
863 }
864 static SYSDEV_ATTR(idle_count, 0444, show_idle_count, NULL);
865
866 static ssize_t show_idle_time(struct sys_device *dev,
867                                 struct sysdev_attribute *attr, char *buf)
868 {
869         struct s390_idle_data *idle;
870         unsigned long long now, idle_time, idle_enter;
871
872         idle = &per_cpu(s390_idle, dev->id);
873         spin_lock(&idle->lock);
874         now = get_clock();
875         idle_time = idle->idle_time;
876         idle_enter = idle->idle_enter;
877         if (idle_enter != 0ULL && idle_enter < now)
878                 idle_time += now - idle_enter;
879         spin_unlock(&idle->lock);
880         return sprintf(buf, "%llu\n", idle_time >> 12);
881 }
882 static SYSDEV_ATTR(idle_time_us, 0444, show_idle_time, NULL);
883
884 static struct attribute *cpu_online_attrs[] = {
885         &attr_capability.attr,
886         &attr_idle_count.attr,
887         &attr_idle_time_us.attr,
888         NULL,
889 };
890
891 static struct attribute_group cpu_online_attr_group = {
892         .attrs = cpu_online_attrs,
893 };
894
895 static int __cpuinit smp_cpu_notify(struct notifier_block *self,
896                                     unsigned long action, void *hcpu)
897 {
898         unsigned int cpu = (unsigned int)(long)hcpu;
899         struct cpu *c = &per_cpu(cpu_devices, cpu);
900         struct sys_device *s = &c->sysdev;
901         struct s390_idle_data *idle;
902
903         switch (action) {
904         case CPU_ONLINE:
905         case CPU_ONLINE_FROZEN:
906                 idle = &per_cpu(s390_idle, cpu);
907                 spin_lock_irq(&idle->lock);
908                 idle->idle_enter = 0;
909                 idle->idle_time = 0;
910                 idle->idle_count = 0;
911                 spin_unlock_irq(&idle->lock);
912                 if (sysfs_create_group(&s->kobj, &cpu_online_attr_group))
913                         return NOTIFY_BAD;
914                 break;
915         case CPU_DEAD:
916         case CPU_DEAD_FROZEN:
917                 sysfs_remove_group(&s->kobj, &cpu_online_attr_group);
918                 break;
919         }
920         return NOTIFY_OK;
921 }
922
923 static struct notifier_block __cpuinitdata smp_cpu_nb = {
924         .notifier_call = smp_cpu_notify,
925 };
926
927 static int __devinit smp_add_present_cpu(int cpu)
928 {
929         struct cpu *c = &per_cpu(cpu_devices, cpu);
930         struct sys_device *s = &c->sysdev;
931         int rc;
932
933         c->hotpluggable = 1;
934         rc = register_cpu(c, cpu);
935         if (rc)
936                 goto out;
937         rc = sysfs_create_group(&s->kobj, &cpu_common_attr_group);
938         if (rc)
939                 goto out_cpu;
940         if (!cpu_online(cpu))
941                 goto out;
942         rc = sysfs_create_group(&s->kobj, &cpu_online_attr_group);
943         if (!rc)
944                 return 0;
945         sysfs_remove_group(&s->kobj, &cpu_common_attr_group);
946 out_cpu:
947 #ifdef CONFIG_HOTPLUG_CPU
948         unregister_cpu(c);
949 #endif
950 out:
951         return rc;
952 }
953
954 #ifdef CONFIG_HOTPLUG_CPU
955
956 int __ref smp_rescan_cpus(void)
957 {
958         cpumask_t newcpus;
959         int cpu;
960         int rc;
961
962         get_online_cpus();
963         mutex_lock(&smp_cpu_state_mutex);
964         newcpus = cpu_present_map;
965         rc = __smp_rescan_cpus();
966         if (rc)
967                 goto out;
968         cpus_andnot(newcpus, cpu_present_map, newcpus);
969         for_each_cpu_mask(cpu, newcpus) {
970                 rc = smp_add_present_cpu(cpu);
971                 if (rc)
972                         cpu_clear(cpu, cpu_present_map);
973         }
974         rc = 0;
975 out:
976         mutex_unlock(&smp_cpu_state_mutex);
977         put_online_cpus();
978         if (!cpus_empty(newcpus))
979                 topology_schedule_update();
980         return rc;
981 }
982
983 static ssize_t __ref rescan_store(struct sysdev_class *class, const char *buf,
984                                   size_t count)
985 {
986         int rc;
987
988         rc = smp_rescan_cpus();
989         return rc ? rc : count;
990 }
991 static SYSDEV_CLASS_ATTR(rescan, 0200, NULL, rescan_store);
992 #endif /* CONFIG_HOTPLUG_CPU */
993
994 static ssize_t dispatching_show(struct sysdev_class *class, char *buf)
995 {
996         ssize_t count;
997
998         mutex_lock(&smp_cpu_state_mutex);
999         count = sprintf(buf, "%d\n", cpu_management);
1000         mutex_unlock(&smp_cpu_state_mutex);
1001         return count;
1002 }
1003
1004 static ssize_t dispatching_store(struct sysdev_class *dev, const char *buf,
1005                                  size_t count)
1006 {
1007         int val, rc;
1008         char delim;
1009
1010         if (sscanf(buf, "%d %c", &val, &delim) != 1)
1011                 return -EINVAL;
1012         if (val != 0 && val != 1)
1013                 return -EINVAL;
1014         rc = 0;
1015         get_online_cpus();
1016         mutex_lock(&smp_cpu_state_mutex);
1017         if (cpu_management == val)
1018                 goto out;
1019         rc = topology_set_cpu_management(val);
1020         if (!rc)
1021                 cpu_management = val;
1022 out:
1023         mutex_unlock(&smp_cpu_state_mutex);
1024         put_online_cpus();
1025         return rc ? rc : count;
1026 }
1027 static SYSDEV_CLASS_ATTR(dispatching, 0644, dispatching_show,
1028                          dispatching_store);
1029
1030 static int __init topology_init(void)
1031 {
1032         int cpu;
1033         int rc;
1034
1035         register_cpu_notifier(&smp_cpu_nb);
1036
1037 #ifdef CONFIG_HOTPLUG_CPU
1038         rc = sysdev_class_create_file(&cpu_sysdev_class, &attr_rescan);
1039         if (rc)
1040                 return rc;
1041 #endif
1042         rc = sysdev_class_create_file(&cpu_sysdev_class, &attr_dispatching);
1043         if (rc)
1044                 return rc;
1045         for_each_present_cpu(cpu) {
1046                 rc = smp_add_present_cpu(cpu);
1047                 if (rc)
1048                         return rc;
1049         }
1050         return 0;
1051 }
1052 subsys_initcall(topology_init);