eebce7fdc97ccfb838b9c8ce695e66691b9ab1c3
[linux-2.6.git] / arch / s390 / kernel / smp.c
1 /*
2  *  arch/s390/kernel/smp.c
3  *
4  *    Copyright IBM Corp. 1999, 2009
5  *    Author(s): Denis Joseph Barrow (djbarrow@de.ibm.com,barrow_dj@yahoo.com),
6  *               Martin Schwidefsky (schwidefsky@de.ibm.com)
7  *               Heiko Carstens (heiko.carstens@de.ibm.com)
8  *
9  *  based on other smp stuff by
10  *    (c) 1995 Alan Cox, CymruNET Ltd  <alan@cymru.net>
11  *    (c) 1998 Ingo Molnar
12  *
13  * We work with logical cpu numbering everywhere we can. The only
14  * functions using the real cpu address (got from STAP) are the sigp
15  * functions. For all other functions we use the identity mapping.
16  * That means that cpu_number_map[i] == i for every cpu. cpu_number_map is
17  * used e.g. to find the idle task belonging to a logical cpu. Every array
18  * in the kernel is sorted by the logical cpu number and not by the physical
19  * one which is causing all the confusion with __cpu_logical_map and
20  * cpu_number_map in other architectures.
21  */
22
23 #define KMSG_COMPONENT "cpu"
24 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
25
26 #include <linux/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/irqflags.h>
36 #include <linux/cpu.h>
37 #include <linux/timex.h>
38 #include <linux/bootmem.h>
39 #include <asm/ipl.h>
40 #include <asm/setup.h>
41 #include <asm/sigp.h>
42 #include <asm/pgalloc.h>
43 #include <asm/irq.h>
44 #include <asm/s390_ext.h>
45 #include <asm/cpcmd.h>
46 #include <asm/tlbflush.h>
47 #include <asm/timer.h>
48 #include <asm/lowcore.h>
49 #include <asm/sclp.h>
50 #include <asm/cputime.h>
51 #include <asm/vdso.h>
52 #include <asm/cpu.h>
53 #include "entry.h"
54
55 static struct task_struct *current_set[NR_CPUS];
56
57 static u8 smp_cpu_type;
58 static int smp_use_sigp_detection;
59
60 enum s390_cpu_state {
61         CPU_STATE_STANDBY,
62         CPU_STATE_CONFIGURED,
63 };
64
65 DEFINE_MUTEX(smp_cpu_state_mutex);
66 int smp_cpu_polarization[NR_CPUS];
67 static int smp_cpu_state[NR_CPUS];
68 static int cpu_management;
69
70 static DEFINE_PER_CPU(struct cpu, cpu_devices);
71
72 static void smp_ext_bitcall(int, ec_bit_sig);
73
74 static int cpu_stopped(int cpu)
75 {
76         __u32 status;
77
78         switch (signal_processor_ps(&status, 0, cpu, sigp_sense)) {
79         case sigp_status_stored:
80                 /* Check for stopped and check stop state */
81                 if (status & 0x50)
82                         return 1;
83                 break;
84         default:
85                 break;
86         }
87         return 0;
88 }
89
90 void smp_send_stop(void)
91 {
92         int cpu, rc;
93
94         /* Disable all interrupts/machine checks */
95         __load_psw_mask(psw_kernel_bits & ~PSW_MASK_MCHECK);
96         trace_hardirqs_off();
97
98         /* stop all processors */
99         for_each_online_cpu(cpu) {
100                 if (cpu == smp_processor_id())
101                         continue;
102                 do {
103                         rc = signal_processor(cpu, sigp_stop);
104                 } while (rc == sigp_busy);
105
106                 while (!cpu_stopped(cpu))
107                         cpu_relax();
108         }
109 }
110
111 /*
112  * This is the main routine where commands issued by other
113  * cpus are handled.
114  */
115
116 static void do_ext_call_interrupt(__u16 code)
117 {
118         unsigned long bits;
119
120         /*
121          * handle bit signal external calls
122          *
123          * For the ec_schedule signal we have to do nothing. All the work
124          * is done automatically when we return from the interrupt.
125          */
126         bits = xchg(&S390_lowcore.ext_call_fast, 0);
127
128         if (test_bit(ec_call_function, &bits))
129                 generic_smp_call_function_interrupt();
130
131         if (test_bit(ec_call_function_single, &bits))
132                 generic_smp_call_function_single_interrupt();
133 }
134
135 /*
136  * Send an external call sigp to another cpu and return without waiting
137  * for its completion.
138  */
139 static void smp_ext_bitcall(int cpu, ec_bit_sig sig)
140 {
141         /*
142          * Set signaling bit in lowcore of target cpu and kick it
143          */
144         set_bit(sig, (unsigned long *) &lowcore_ptr[cpu]->ext_call_fast);
145         while (signal_processor(cpu, sigp_emergency_signal) == sigp_busy)
146                 udelay(10);
147 }
148
149 void arch_send_call_function_ipi_mask(const struct cpumask *mask)
150 {
151         int cpu;
152
153         for_each_cpu(cpu, mask)
154                 smp_ext_bitcall(cpu, ec_call_function);
155 }
156
157 void arch_send_call_function_single_ipi(int cpu)
158 {
159         smp_ext_bitcall(cpu, ec_call_function_single);
160 }
161
162 #ifndef CONFIG_64BIT
163 /*
164  * this function sends a 'purge tlb' signal to another CPU.
165  */
166 static void smp_ptlb_callback(void *info)
167 {
168         __tlb_flush_local();
169 }
170
171 void smp_ptlb_all(void)
172 {
173         on_each_cpu(smp_ptlb_callback, NULL, 1);
174 }
175 EXPORT_SYMBOL(smp_ptlb_all);
176 #endif /* ! CONFIG_64BIT */
177
178 /*
179  * this function sends a 'reschedule' IPI to another CPU.
180  * it goes straight through and wastes no time serializing
181  * anything. Worst case is that we lose a reschedule ...
182  */
183 void smp_send_reschedule(int cpu)
184 {
185         smp_ext_bitcall(cpu, ec_schedule);
186 }
187
188 /*
189  * parameter area for the set/clear control bit callbacks
190  */
191 struct ec_creg_mask_parms {
192         unsigned long orvals[16];
193         unsigned long andvals[16];
194 };
195
196 /*
197  * callback for setting/clearing control bits
198  */
199 static void smp_ctl_bit_callback(void *info)
200 {
201         struct ec_creg_mask_parms *pp = info;
202         unsigned long cregs[16];
203         int i;
204
205         __ctl_store(cregs, 0, 15);
206         for (i = 0; i <= 15; i++)
207                 cregs[i] = (cregs[i] & pp->andvals[i]) | pp->orvals[i];
208         __ctl_load(cregs, 0, 15);
209 }
210
211 /*
212  * Set a bit in a control register of all cpus
213  */
214 void smp_ctl_set_bit(int cr, int bit)
215 {
216         struct ec_creg_mask_parms parms;
217
218         memset(&parms.orvals, 0, sizeof(parms.orvals));
219         memset(&parms.andvals, 0xff, sizeof(parms.andvals));
220         parms.orvals[cr] = 1 << bit;
221         on_each_cpu(smp_ctl_bit_callback, &parms, 1);
222 }
223 EXPORT_SYMBOL(smp_ctl_set_bit);
224
225 /*
226  * Clear a bit in a control register of all cpus
227  */
228 void smp_ctl_clear_bit(int cr, int bit)
229 {
230         struct ec_creg_mask_parms parms;
231
232         memset(&parms.orvals, 0, sizeof(parms.orvals));
233         memset(&parms.andvals, 0xff, sizeof(parms.andvals));
234         parms.andvals[cr] = ~(1L << bit);
235         on_each_cpu(smp_ctl_bit_callback, &parms, 1);
236 }
237 EXPORT_SYMBOL(smp_ctl_clear_bit);
238
239 /*
240  * In early ipl state a temp. logically cpu number is needed, so the sigp
241  * functions can be used to sense other cpus. Since NR_CPUS is >= 2 on
242  * CONFIG_SMP and the ipl cpu is logical cpu 0, it must be 1.
243  */
244 #define CPU_INIT_NO     1
245
246 #ifdef CONFIG_ZFCPDUMP
247
248 /*
249  * zfcpdump_prefix_array holds prefix registers for the following scenario:
250  * 64 bit zfcpdump kernel and 31 bit kernel which is to be dumped. We have to
251  * save its prefix registers, since they get lost, when switching from 31 bit
252  * to 64 bit.
253  */
254 unsigned int zfcpdump_prefix_array[NR_CPUS + 1] \
255         __attribute__((__section__(".data")));
256
257 static void __init smp_get_save_area(unsigned int cpu, unsigned int phy_cpu)
258 {
259         if (ipl_info.type != IPL_TYPE_FCP_DUMP)
260                 return;
261         if (cpu >= NR_CPUS) {
262                 pr_warning("CPU %i exceeds the maximum %i and is excluded from "
263                            "the dump\n", cpu, NR_CPUS - 1);
264                 return;
265         }
266         zfcpdump_save_areas[cpu] = kmalloc(sizeof(union save_area), GFP_KERNEL);
267         __cpu_logical_map[CPU_INIT_NO] = (__u16) phy_cpu;
268         while (signal_processor(CPU_INIT_NO, sigp_stop_and_store_status) ==
269                sigp_busy)
270                 cpu_relax();
271         memcpy(zfcpdump_save_areas[cpu],
272                (void *)(unsigned long) store_prefix() + SAVE_AREA_BASE,
273                SAVE_AREA_SIZE);
274 #ifdef CONFIG_64BIT
275         /* copy original prefix register */
276         zfcpdump_save_areas[cpu]->s390x.pref_reg = zfcpdump_prefix_array[cpu];
277 #endif
278 }
279
280 union save_area *zfcpdump_save_areas[NR_CPUS + 1];
281 EXPORT_SYMBOL_GPL(zfcpdump_save_areas);
282
283 #else
284
285 static inline void smp_get_save_area(unsigned int cpu, unsigned int phy_cpu) { }
286
287 #endif /* CONFIG_ZFCPDUMP */
288
289 static int cpu_known(int cpu_id)
290 {
291         int cpu;
292
293         for_each_present_cpu(cpu) {
294                 if (__cpu_logical_map[cpu] == cpu_id)
295                         return 1;
296         }
297         return 0;
298 }
299
300 static int smp_rescan_cpus_sigp(cpumask_t avail)
301 {
302         int cpu_id, logical_cpu;
303
304         logical_cpu = cpumask_first(&avail);
305         if (logical_cpu >= nr_cpu_ids)
306                 return 0;
307         for (cpu_id = 0; cpu_id <= MAX_CPU_ADDRESS; cpu_id++) {
308                 if (cpu_known(cpu_id))
309                         continue;
310                 __cpu_logical_map[logical_cpu] = cpu_id;
311                 smp_cpu_polarization[logical_cpu] = POLARIZATION_UNKNWN;
312                 if (!cpu_stopped(logical_cpu))
313                         continue;
314                 cpu_set(logical_cpu, cpu_present_map);
315                 smp_cpu_state[logical_cpu] = CPU_STATE_CONFIGURED;
316                 logical_cpu = cpumask_next(logical_cpu, &avail);
317                 if (logical_cpu >= nr_cpu_ids)
318                         break;
319         }
320         return 0;
321 }
322
323 static int smp_rescan_cpus_sclp(cpumask_t avail)
324 {
325         struct sclp_cpu_info *info;
326         int cpu_id, logical_cpu, cpu;
327         int rc;
328
329         logical_cpu = cpumask_first(&avail);
330         if (logical_cpu >= nr_cpu_ids)
331                 return 0;
332         info = kmalloc(sizeof(*info), GFP_KERNEL);
333         if (!info)
334                 return -ENOMEM;
335         rc = sclp_get_cpu_info(info);
336         if (rc)
337                 goto out;
338         for (cpu = 0; cpu < info->combined; cpu++) {
339                 if (info->has_cpu_type && info->cpu[cpu].type != smp_cpu_type)
340                         continue;
341                 cpu_id = info->cpu[cpu].address;
342                 if (cpu_known(cpu_id))
343                         continue;
344                 __cpu_logical_map[logical_cpu] = cpu_id;
345                 smp_cpu_polarization[logical_cpu] = POLARIZATION_UNKNWN;
346                 cpu_set(logical_cpu, cpu_present_map);
347                 if (cpu >= info->configured)
348                         smp_cpu_state[logical_cpu] = CPU_STATE_STANDBY;
349                 else
350                         smp_cpu_state[logical_cpu] = CPU_STATE_CONFIGURED;
351                 logical_cpu = cpumask_next(logical_cpu, &avail);
352                 if (logical_cpu >= nr_cpu_ids)
353                         break;
354         }
355 out:
356         kfree(info);
357         return rc;
358 }
359
360 static int __smp_rescan_cpus(void)
361 {
362         cpumask_t avail;
363
364         cpus_xor(avail, cpu_possible_map, cpu_present_map);
365         if (smp_use_sigp_detection)
366                 return smp_rescan_cpus_sigp(avail);
367         else
368                 return smp_rescan_cpus_sclp(avail);
369 }
370
371 static void __init smp_detect_cpus(void)
372 {
373         unsigned int cpu, c_cpus, s_cpus;
374         struct sclp_cpu_info *info;
375         u16 boot_cpu_addr, cpu_addr;
376
377         c_cpus = 1;
378         s_cpus = 0;
379         boot_cpu_addr = __cpu_logical_map[0];
380         info = kmalloc(sizeof(*info), GFP_KERNEL);
381         if (!info)
382                 panic("smp_detect_cpus failed to allocate memory\n");
383         /* Use sigp detection algorithm if sclp doesn't work. */
384         if (sclp_get_cpu_info(info)) {
385                 smp_use_sigp_detection = 1;
386                 for (cpu = 0; cpu <= MAX_CPU_ADDRESS; cpu++) {
387                         if (cpu == boot_cpu_addr)
388                                 continue;
389                         __cpu_logical_map[CPU_INIT_NO] = cpu;
390                         if (!cpu_stopped(CPU_INIT_NO))
391                                 continue;
392                         smp_get_save_area(c_cpus, cpu);
393                         c_cpus++;
394                 }
395                 goto out;
396         }
397
398         if (info->has_cpu_type) {
399                 for (cpu = 0; cpu < info->combined; cpu++) {
400                         if (info->cpu[cpu].address == boot_cpu_addr) {
401                                 smp_cpu_type = info->cpu[cpu].type;
402                                 break;
403                         }
404                 }
405         }
406
407         for (cpu = 0; cpu < info->combined; cpu++) {
408                 if (info->has_cpu_type && info->cpu[cpu].type != smp_cpu_type)
409                         continue;
410                 cpu_addr = info->cpu[cpu].address;
411                 if (cpu_addr == boot_cpu_addr)
412                         continue;
413                 __cpu_logical_map[CPU_INIT_NO] = cpu_addr;
414                 if (!cpu_stopped(CPU_INIT_NO)) {
415                         s_cpus++;
416                         continue;
417                 }
418                 smp_get_save_area(c_cpus, cpu_addr);
419                 c_cpus++;
420         }
421 out:
422         kfree(info);
423         pr_info("%d configured CPUs, %d standby CPUs\n", c_cpus, s_cpus);
424         get_online_cpus();
425         __smp_rescan_cpus();
426         put_online_cpus();
427 }
428
429 /*
430  *      Activate a secondary processor.
431  */
432 int __cpuinit start_secondary(void *cpuvoid)
433 {
434         /* Setup the cpu */
435         cpu_init();
436         preempt_disable();
437         /* Enable TOD clock interrupts on the secondary cpu. */
438         init_cpu_timer();
439         /* Enable cpu timer interrupts on the secondary cpu. */
440         init_cpu_vtimer();
441         /* Enable pfault pseudo page faults on this cpu. */
442         pfault_init();
443
444         /* call cpu notifiers */
445         notify_cpu_starting(smp_processor_id());
446         /* Mark this cpu as online */
447         ipi_call_lock();
448         cpu_set(smp_processor_id(), cpu_online_map);
449         ipi_call_unlock();
450         /* Switch on interrupts */
451         local_irq_enable();
452         /* Print info about this processor */
453         print_cpu_info();
454         /* cpu_idle will call schedule for us */
455         cpu_idle();
456         return 0;
457 }
458
459 static void __init smp_create_idle(unsigned int cpu)
460 {
461         struct task_struct *p;
462
463         /*
464          *  don't care about the psw and regs settings since we'll never
465          *  reschedule the forked task.
466          */
467         p = fork_idle(cpu);
468         if (IS_ERR(p))
469                 panic("failed fork for CPU %u: %li", cpu, PTR_ERR(p));
470         current_set[cpu] = p;
471 }
472
473 static int __cpuinit smp_alloc_lowcore(int cpu)
474 {
475         unsigned long async_stack, panic_stack;
476         struct _lowcore *lowcore;
477
478         lowcore = (void *) __get_free_pages(GFP_KERNEL | GFP_DMA, LC_ORDER);
479         if (!lowcore)
480                 return -ENOMEM;
481         async_stack = __get_free_pages(GFP_KERNEL, ASYNC_ORDER);
482         panic_stack = __get_free_page(GFP_KERNEL);
483         if (!panic_stack || !async_stack)
484                 goto out;
485         memcpy(lowcore, &S390_lowcore, 512);
486         memset((char *)lowcore + 512, 0, sizeof(*lowcore) - 512);
487         lowcore->async_stack = async_stack + ASYNC_SIZE;
488         lowcore->panic_stack = panic_stack + PAGE_SIZE;
489
490 #ifndef CONFIG_64BIT
491         if (MACHINE_HAS_IEEE) {
492                 unsigned long save_area;
493
494                 save_area = get_zeroed_page(GFP_KERNEL);
495                 if (!save_area)
496                         goto out;
497                 lowcore->extended_save_area_addr = (u32) save_area;
498         }
499 #else
500         if (vdso_alloc_per_cpu(cpu, lowcore))
501                 goto out;
502 #endif
503         lowcore_ptr[cpu] = lowcore;
504         return 0;
505
506 out:
507         free_page(panic_stack);
508         free_pages(async_stack, ASYNC_ORDER);
509         free_pages((unsigned long) lowcore, LC_ORDER);
510         return -ENOMEM;
511 }
512
513 static void smp_free_lowcore(int cpu)
514 {
515         struct _lowcore *lowcore;
516
517         lowcore = lowcore_ptr[cpu];
518 #ifndef CONFIG_64BIT
519         if (MACHINE_HAS_IEEE)
520                 free_page((unsigned long) lowcore->extended_save_area_addr);
521 #else
522         vdso_free_per_cpu(cpu, lowcore);
523 #endif
524         free_page(lowcore->panic_stack - PAGE_SIZE);
525         free_pages(lowcore->async_stack - ASYNC_SIZE, ASYNC_ORDER);
526         free_pages((unsigned long) lowcore, LC_ORDER);
527         lowcore_ptr[cpu] = NULL;
528 }
529
530 /* Upping and downing of CPUs */
531 int __cpuinit __cpu_up(unsigned int cpu)
532 {
533         struct task_struct *idle;
534         struct _lowcore *cpu_lowcore;
535         struct stack_frame *sf;
536         sigp_ccode ccode;
537         u32 lowcore;
538
539         if (smp_cpu_state[cpu] != CPU_STATE_CONFIGURED)
540                 return -EIO;
541         if (smp_alloc_lowcore(cpu))
542                 return -ENOMEM;
543         do {
544                 ccode = signal_processor(cpu, sigp_initial_cpu_reset);
545                 if (ccode == sigp_busy)
546                         udelay(10);
547                 if (ccode == sigp_not_operational)
548                         goto err_out;
549         } while (ccode == sigp_busy);
550
551         lowcore = (u32)(unsigned long)lowcore_ptr[cpu];
552         while (signal_processor_p(lowcore, cpu, sigp_set_prefix) == sigp_busy)
553                 udelay(10);
554
555         idle = current_set[cpu];
556         cpu_lowcore = lowcore_ptr[cpu];
557         cpu_lowcore->kernel_stack = (unsigned long)
558                 task_stack_page(idle) + THREAD_SIZE;
559         cpu_lowcore->thread_info = (unsigned long) task_thread_info(idle);
560         sf = (struct stack_frame *) (cpu_lowcore->kernel_stack
561                                      - sizeof(struct pt_regs)
562                                      - sizeof(struct stack_frame));
563         memset(sf, 0, sizeof(struct stack_frame));
564         sf->gprs[9] = (unsigned long) sf;
565         cpu_lowcore->save_area[15] = (unsigned long) sf;
566         __ctl_store(cpu_lowcore->cregs_save_area, 0, 15);
567         asm volatile(
568                 "       stam    0,15,0(%0)"
569                 : : "a" (&cpu_lowcore->access_regs_save_area) : "memory");
570         cpu_lowcore->percpu_offset = __per_cpu_offset[cpu];
571         cpu_lowcore->current_task = (unsigned long) idle;
572         cpu_lowcore->cpu_nr = cpu;
573         cpu_lowcore->kernel_asce = S390_lowcore.kernel_asce;
574         cpu_lowcore->machine_flags = S390_lowcore.machine_flags;
575         cpu_lowcore->ftrace_func = S390_lowcore.ftrace_func;
576         eieio();
577
578         while (signal_processor(cpu, sigp_restart) == sigp_busy)
579                 udelay(10);
580
581         while (!cpu_online(cpu))
582                 cpu_relax();
583         return 0;
584
585 err_out:
586         smp_free_lowcore(cpu);
587         return -EIO;
588 }
589
590 static int __init setup_possible_cpus(char *s)
591 {
592         int pcpus, cpu;
593
594         pcpus = simple_strtoul(s, NULL, 0);
595         init_cpu_possible(cpumask_of(0));
596         for (cpu = 1; cpu < pcpus && cpu < nr_cpu_ids; cpu++)
597                 set_cpu_possible(cpu, true);
598         return 0;
599 }
600 early_param("possible_cpus", setup_possible_cpus);
601
602 #ifdef CONFIG_HOTPLUG_CPU
603
604 int __cpu_disable(void)
605 {
606         struct ec_creg_mask_parms cr_parms;
607         int cpu = smp_processor_id();
608
609         cpu_clear(cpu, cpu_online_map);
610
611         /* Disable pfault pseudo page faults on this cpu. */
612         pfault_fini();
613
614         memset(&cr_parms.orvals, 0, sizeof(cr_parms.orvals));
615         memset(&cr_parms.andvals, 0xff, sizeof(cr_parms.andvals));
616
617         /* disable all external interrupts */
618         cr_parms.orvals[0] = 0;
619         cr_parms.andvals[0] = ~(1 << 15 | 1 << 14 | 1 << 13 | 1 << 12 |
620                                 1 << 11 | 1 << 10 | 1 <<  6 | 1 <<  4);
621         /* disable all I/O interrupts */
622         cr_parms.orvals[6] = 0;
623         cr_parms.andvals[6] = ~(1 << 31 | 1 << 30 | 1 << 29 | 1 << 28 |
624                                 1 << 27 | 1 << 26 | 1 << 25 | 1 << 24);
625         /* disable most machine checks */
626         cr_parms.orvals[14] = 0;
627         cr_parms.andvals[14] = ~(1 << 28 | 1 << 27 | 1 << 26 |
628                                  1 << 25 | 1 << 24);
629
630         smp_ctl_bit_callback(&cr_parms);
631
632         return 0;
633 }
634
635 void __cpu_die(unsigned int cpu)
636 {
637         /* Wait until target cpu is down */
638         while (!cpu_stopped(cpu))
639                 cpu_relax();
640         while (signal_processor_p(0, cpu, sigp_set_prefix) == sigp_busy)
641                 udelay(10);
642         smp_free_lowcore(cpu);
643         pr_info("Processor %d stopped\n", cpu);
644 }
645
646 void cpu_die(void)
647 {
648         idle_task_exit();
649         while (signal_processor(smp_processor_id(), sigp_stop) == sigp_busy)
650                 cpu_relax();
651         for (;;);
652 }
653
654 #endif /* CONFIG_HOTPLUG_CPU */
655
656 void __init smp_prepare_cpus(unsigned int max_cpus)
657 {
658 #ifndef CONFIG_64BIT
659         unsigned long save_area = 0;
660 #endif
661         unsigned long async_stack, panic_stack;
662         struct _lowcore *lowcore;
663         unsigned int cpu;
664
665         smp_detect_cpus();
666
667         /* request the 0x1201 emergency signal external interrupt */
668         if (register_external_interrupt(0x1201, do_ext_call_interrupt) != 0)
669                 panic("Couldn't request external interrupt 0x1201");
670         print_cpu_info();
671
672         /* Reallocate current lowcore, but keep its contents. */
673         lowcore = (void *) __get_free_pages(GFP_KERNEL | GFP_DMA, LC_ORDER);
674         panic_stack = __get_free_page(GFP_KERNEL);
675         async_stack = __get_free_pages(GFP_KERNEL, ASYNC_ORDER);
676         BUG_ON(!lowcore || !panic_stack || !async_stack);
677 #ifndef CONFIG_64BIT
678         if (MACHINE_HAS_IEEE)
679                 save_area = get_zeroed_page(GFP_KERNEL);
680 #endif
681         local_irq_disable();
682         local_mcck_disable();
683         lowcore_ptr[smp_processor_id()] = lowcore;
684         *lowcore = S390_lowcore;
685         lowcore->panic_stack = panic_stack + PAGE_SIZE;
686         lowcore->async_stack = async_stack + ASYNC_SIZE;
687 #ifndef CONFIG_64BIT
688         if (MACHINE_HAS_IEEE)
689                 lowcore->extended_save_area_addr = (u32) save_area;
690 #endif
691         set_prefix((u32)(unsigned long) lowcore);
692         local_mcck_enable();
693         local_irq_enable();
694 #ifdef CONFIG_64BIT
695         if (vdso_alloc_per_cpu(smp_processor_id(), &S390_lowcore))
696                 BUG();
697 #endif
698         for_each_possible_cpu(cpu)
699                 if (cpu != smp_processor_id())
700                         smp_create_idle(cpu);
701 }
702
703 void __init smp_prepare_boot_cpu(void)
704 {
705         BUG_ON(smp_processor_id() != 0);
706
707         current_thread_info()->cpu = 0;
708         cpu_set(0, cpu_present_map);
709         cpu_set(0, cpu_online_map);
710         S390_lowcore.percpu_offset = __per_cpu_offset[0];
711         current_set[0] = current;
712         smp_cpu_state[0] = CPU_STATE_CONFIGURED;
713         smp_cpu_polarization[0] = POLARIZATION_UNKNWN;
714 }
715
716 void __init smp_cpus_done(unsigned int max_cpus)
717 {
718 }
719
720 void __init smp_setup_processor_id(void)
721 {
722         S390_lowcore.cpu_nr = 0;
723         __cpu_logical_map[0] = stap();
724 }
725
726 /*
727  * the frequency of the profiling timer can be changed
728  * by writing a multiplier value into /proc/profile.
729  *
730  * usually you want to run this on all CPUs ;)
731  */
732 int setup_profiling_timer(unsigned int multiplier)
733 {
734         return 0;
735 }
736
737 #ifdef CONFIG_HOTPLUG_CPU
738 static ssize_t cpu_configure_show(struct sys_device *dev,
739                                 struct sysdev_attribute *attr, char *buf)
740 {
741         ssize_t count;
742
743         mutex_lock(&smp_cpu_state_mutex);
744         count = sprintf(buf, "%d\n", smp_cpu_state[dev->id]);
745         mutex_unlock(&smp_cpu_state_mutex);
746         return count;
747 }
748
749 static ssize_t cpu_configure_store(struct sys_device *dev,
750                                   struct sysdev_attribute *attr,
751                                   const char *buf, size_t count)
752 {
753         int cpu = dev->id;
754         int val, rc;
755         char delim;
756
757         if (sscanf(buf, "%d %c", &val, &delim) != 1)
758                 return -EINVAL;
759         if (val != 0 && val != 1)
760                 return -EINVAL;
761
762         get_online_cpus();
763         mutex_lock(&smp_cpu_state_mutex);
764         rc = -EBUSY;
765         if (cpu_online(cpu))
766                 goto out;
767         rc = 0;
768         switch (val) {
769         case 0:
770                 if (smp_cpu_state[cpu] == CPU_STATE_CONFIGURED) {
771                         rc = sclp_cpu_deconfigure(__cpu_logical_map[cpu]);
772                         if (!rc) {
773                                 smp_cpu_state[cpu] = CPU_STATE_STANDBY;
774                                 smp_cpu_polarization[cpu] = POLARIZATION_UNKNWN;
775                         }
776                 }
777                 break;
778         case 1:
779                 if (smp_cpu_state[cpu] == CPU_STATE_STANDBY) {
780                         rc = sclp_cpu_configure(__cpu_logical_map[cpu]);
781                         if (!rc) {
782                                 smp_cpu_state[cpu] = CPU_STATE_CONFIGURED;
783                                 smp_cpu_polarization[cpu] = POLARIZATION_UNKNWN;
784                         }
785                 }
786                 break;
787         default:
788                 break;
789         }
790 out:
791         mutex_unlock(&smp_cpu_state_mutex);
792         put_online_cpus();
793         return rc ? rc : count;
794 }
795 static SYSDEV_ATTR(configure, 0644, cpu_configure_show, cpu_configure_store);
796 #endif /* CONFIG_HOTPLUG_CPU */
797
798 static ssize_t cpu_polarization_show(struct sys_device *dev,
799                                      struct sysdev_attribute *attr, char *buf)
800 {
801         int cpu = dev->id;
802         ssize_t count;
803
804         mutex_lock(&smp_cpu_state_mutex);
805         switch (smp_cpu_polarization[cpu]) {
806         case POLARIZATION_HRZ:
807                 count = sprintf(buf, "horizontal\n");
808                 break;
809         case POLARIZATION_VL:
810                 count = sprintf(buf, "vertical:low\n");
811                 break;
812         case POLARIZATION_VM:
813                 count = sprintf(buf, "vertical:medium\n");
814                 break;
815         case POLARIZATION_VH:
816                 count = sprintf(buf, "vertical:high\n");
817                 break;
818         default:
819                 count = sprintf(buf, "unknown\n");
820                 break;
821         }
822         mutex_unlock(&smp_cpu_state_mutex);
823         return count;
824 }
825 static SYSDEV_ATTR(polarization, 0444, cpu_polarization_show, NULL);
826
827 static ssize_t show_cpu_address(struct sys_device *dev,
828                                 struct sysdev_attribute *attr, char *buf)
829 {
830         return sprintf(buf, "%d\n", __cpu_logical_map[dev->id]);
831 }
832 static SYSDEV_ATTR(address, 0444, show_cpu_address, NULL);
833
834
835 static struct attribute *cpu_common_attrs[] = {
836 #ifdef CONFIG_HOTPLUG_CPU
837         &attr_configure.attr,
838 #endif
839         &attr_address.attr,
840         &attr_polarization.attr,
841         NULL,
842 };
843
844 static struct attribute_group cpu_common_attr_group = {
845         .attrs = cpu_common_attrs,
846 };
847
848 static ssize_t show_capability(struct sys_device *dev,
849                                 struct sysdev_attribute *attr, char *buf)
850 {
851         unsigned int capability;
852         int rc;
853
854         rc = get_cpu_capability(&capability);
855         if (rc)
856                 return rc;
857         return sprintf(buf, "%u\n", capability);
858 }
859 static SYSDEV_ATTR(capability, 0444, show_capability, NULL);
860
861 static ssize_t show_idle_count(struct sys_device *dev,
862                                 struct sysdev_attribute *attr, char *buf)
863 {
864         struct s390_idle_data *idle;
865         unsigned long long idle_count;
866         unsigned int sequence;
867
868         idle = &per_cpu(s390_idle, dev->id);
869 repeat:
870         sequence = idle->sequence;
871         smp_rmb();
872         if (sequence & 1)
873                 goto repeat;
874         idle_count = idle->idle_count;
875         if (idle->idle_enter)
876                 idle_count++;
877         smp_rmb();
878         if (idle->sequence != sequence)
879                 goto repeat;
880         return sprintf(buf, "%llu\n", idle_count);
881 }
882 static SYSDEV_ATTR(idle_count, 0444, show_idle_count, NULL);
883
884 static ssize_t show_idle_time(struct sys_device *dev,
885                                 struct sysdev_attribute *attr, char *buf)
886 {
887         struct s390_idle_data *idle;
888         unsigned long long now, idle_time, idle_enter;
889         unsigned int sequence;
890
891         idle = &per_cpu(s390_idle, dev->id);
892         now = get_clock();
893 repeat:
894         sequence = idle->sequence;
895         smp_rmb();
896         if (sequence & 1)
897                 goto repeat;
898         idle_time = idle->idle_time;
899         idle_enter = idle->idle_enter;
900         if (idle_enter != 0ULL && idle_enter < now)
901                 idle_time += now - idle_enter;
902         smp_rmb();
903         if (idle->sequence != sequence)
904                 goto repeat;
905         return sprintf(buf, "%llu\n", idle_time >> 12);
906 }
907 static SYSDEV_ATTR(idle_time_us, 0444, show_idle_time, NULL);
908
909 static struct attribute *cpu_online_attrs[] = {
910         &attr_capability.attr,
911         &attr_idle_count.attr,
912         &attr_idle_time_us.attr,
913         NULL,
914 };
915
916 static struct attribute_group cpu_online_attr_group = {
917         .attrs = cpu_online_attrs,
918 };
919
920 static int __cpuinit smp_cpu_notify(struct notifier_block *self,
921                                     unsigned long action, void *hcpu)
922 {
923         unsigned int cpu = (unsigned int)(long)hcpu;
924         struct cpu *c = &per_cpu(cpu_devices, cpu);
925         struct sys_device *s = &c->sysdev;
926         struct s390_idle_data *idle;
927
928         switch (action) {
929         case CPU_ONLINE:
930         case CPU_ONLINE_FROZEN:
931                 idle = &per_cpu(s390_idle, cpu);
932                 memset(idle, 0, sizeof(struct s390_idle_data));
933                 if (sysfs_create_group(&s->kobj, &cpu_online_attr_group))
934                         return NOTIFY_BAD;
935                 break;
936         case CPU_DEAD:
937         case CPU_DEAD_FROZEN:
938                 sysfs_remove_group(&s->kobj, &cpu_online_attr_group);
939                 break;
940         }
941         return NOTIFY_OK;
942 }
943
944 static struct notifier_block __cpuinitdata smp_cpu_nb = {
945         .notifier_call = smp_cpu_notify,
946 };
947
948 static int __devinit smp_add_present_cpu(int cpu)
949 {
950         struct cpu *c = &per_cpu(cpu_devices, cpu);
951         struct sys_device *s = &c->sysdev;
952         int rc;
953
954         c->hotpluggable = 1;
955         rc = register_cpu(c, cpu);
956         if (rc)
957                 goto out;
958         rc = sysfs_create_group(&s->kobj, &cpu_common_attr_group);
959         if (rc)
960                 goto out_cpu;
961         if (!cpu_online(cpu))
962                 goto out;
963         rc = sysfs_create_group(&s->kobj, &cpu_online_attr_group);
964         if (!rc)
965                 return 0;
966         sysfs_remove_group(&s->kobj, &cpu_common_attr_group);
967 out_cpu:
968 #ifdef CONFIG_HOTPLUG_CPU
969         unregister_cpu(c);
970 #endif
971 out:
972         return rc;
973 }
974
975 #ifdef CONFIG_HOTPLUG_CPU
976
977 int __ref smp_rescan_cpus(void)
978 {
979         cpumask_t newcpus;
980         int cpu;
981         int rc;
982
983         get_online_cpus();
984         mutex_lock(&smp_cpu_state_mutex);
985         newcpus = cpu_present_map;
986         rc = __smp_rescan_cpus();
987         if (rc)
988                 goto out;
989         cpus_andnot(newcpus, cpu_present_map, newcpus);
990         for_each_cpu_mask(cpu, newcpus) {
991                 rc = smp_add_present_cpu(cpu);
992                 if (rc)
993                         cpu_clear(cpu, cpu_present_map);
994         }
995         rc = 0;
996 out:
997         mutex_unlock(&smp_cpu_state_mutex);
998         put_online_cpus();
999         if (!cpus_empty(newcpus))
1000                 topology_schedule_update();
1001         return rc;
1002 }
1003
1004 static ssize_t __ref rescan_store(struct sysdev_class *class, const char *buf,
1005                                   size_t count)
1006 {
1007         int rc;
1008
1009         rc = smp_rescan_cpus();
1010         return rc ? rc : count;
1011 }
1012 static SYSDEV_CLASS_ATTR(rescan, 0200, NULL, rescan_store);
1013 #endif /* CONFIG_HOTPLUG_CPU */
1014
1015 static ssize_t dispatching_show(struct sysdev_class *class, char *buf)
1016 {
1017         ssize_t count;
1018
1019         mutex_lock(&smp_cpu_state_mutex);
1020         count = sprintf(buf, "%d\n", cpu_management);
1021         mutex_unlock(&smp_cpu_state_mutex);
1022         return count;
1023 }
1024
1025 static ssize_t dispatching_store(struct sysdev_class *dev, const char *buf,
1026                                  size_t count)
1027 {
1028         int val, rc;
1029         char delim;
1030
1031         if (sscanf(buf, "%d %c", &val, &delim) != 1)
1032                 return -EINVAL;
1033         if (val != 0 && val != 1)
1034                 return -EINVAL;
1035         rc = 0;
1036         get_online_cpus();
1037         mutex_lock(&smp_cpu_state_mutex);
1038         if (cpu_management == val)
1039                 goto out;
1040         rc = topology_set_cpu_management(val);
1041         if (!rc)
1042                 cpu_management = val;
1043 out:
1044         mutex_unlock(&smp_cpu_state_mutex);
1045         put_online_cpus();
1046         return rc ? rc : count;
1047 }
1048 static SYSDEV_CLASS_ATTR(dispatching, 0644, dispatching_show,
1049                          dispatching_store);
1050
1051 static int __init topology_init(void)
1052 {
1053         int cpu;
1054         int rc;
1055
1056         register_cpu_notifier(&smp_cpu_nb);
1057
1058 #ifdef CONFIG_HOTPLUG_CPU
1059         rc = sysdev_class_create_file(&cpu_sysdev_class, &attr_rescan);
1060         if (rc)
1061                 return rc;
1062 #endif
1063         rc = sysdev_class_create_file(&cpu_sysdev_class, &attr_dispatching);
1064         if (rc)
1065                 return rc;
1066         for_each_present_cpu(cpu) {
1067                 rc = smp_add_present_cpu(cpu);
1068                 if (rc)
1069                         return rc;
1070         }
1071         return 0;
1072 }
1073 subsys_initcall(topology_init);