[S390] kdump backend code
[linux-3.10.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/workqueue.h>
27 #include <linux/module.h>
28 #include <linux/init.h>
29 #include <linux/mm.h>
30 #include <linux/err.h>
31 #include <linux/spinlock.h>
32 #include <linux/kernel_stat.h>
33 #include <linux/delay.h>
34 #include <linux/cache.h>
35 #include <linux/interrupt.h>
36 #include <linux/irqflags.h>
37 #include <linux/cpu.h>
38 #include <linux/timex.h>
39 #include <linux/bootmem.h>
40 #include <linux/slab.h>
41 #include <linux/crash_dump.h>
42 #include <asm/asm-offsets.h>
43 #include <asm/ipl.h>
44 #include <asm/setup.h>
45 #include <asm/sigp.h>
46 #include <asm/pgalloc.h>
47 #include <asm/irq.h>
48 #include <asm/cpcmd.h>
49 #include <asm/tlbflush.h>
50 #include <asm/timer.h>
51 #include <asm/lowcore.h>
52 #include <asm/sclp.h>
53 #include <asm/cputime.h>
54 #include <asm/vdso.h>
55 #include <asm/cpu.h>
56 #include "entry.h"
57
58 /* logical cpu to cpu address */
59 unsigned short __cpu_logical_map[NR_CPUS];
60
61 static struct task_struct *current_set[NR_CPUS];
62
63 static u8 smp_cpu_type;
64 static int smp_use_sigp_detection;
65
66 enum s390_cpu_state {
67         CPU_STATE_STANDBY,
68         CPU_STATE_CONFIGURED,
69 };
70
71 DEFINE_MUTEX(smp_cpu_state_mutex);
72 int smp_cpu_polarization[NR_CPUS];
73 static int smp_cpu_state[NR_CPUS];
74 static int cpu_management;
75
76 static DEFINE_PER_CPU(struct cpu, cpu_devices);
77
78 static void smp_ext_bitcall(int, int);
79
80 static int raw_cpu_stopped(int cpu)
81 {
82         u32 status;
83
84         switch (raw_sigp_ps(&status, 0, cpu, sigp_sense)) {
85         case sigp_status_stored:
86                 /* Check for stopped and check stop state */
87                 if (status & 0x50)
88                         return 1;
89                 break;
90         default:
91                 break;
92         }
93         return 0;
94 }
95
96 static inline int cpu_stopped(int cpu)
97 {
98         return raw_cpu_stopped(cpu_logical_map(cpu));
99 }
100
101 /*
102  * Ensure that PSW restart is done on an online CPU
103  */
104 void smp_restart_with_online_cpu(void)
105 {
106         int cpu;
107
108         for_each_online_cpu(cpu) {
109                 if (stap() == __cpu_logical_map[cpu]) {
110                         /* We are online: Enable DAT again and return */
111                         __load_psw_mask(psw_kernel_bits & ~PSW_MASK_MCHECK);
112                         return;
113                 }
114         }
115         /* We are not online: Do PSW restart on an online CPU */
116         while (sigp(cpu, sigp_restart) == sigp_busy)
117                 cpu_relax();
118         /* And stop ourself */
119         while (raw_sigp(stap(), sigp_stop) == sigp_busy)
120                 cpu_relax();
121         for (;;);
122 }
123
124 void smp_switch_to_ipl_cpu(void (*func)(void *), void *data)
125 {
126         struct _lowcore *lc, *current_lc;
127         struct stack_frame *sf;
128         struct pt_regs *regs;
129         unsigned long sp;
130
131         if (smp_processor_id() == 0)
132                 func(data);
133         __load_psw_mask(PSW_BASE_BITS | PSW_DEFAULT_KEY);
134         /* Disable lowcore protection */
135         __ctl_clear_bit(0, 28);
136         current_lc = lowcore_ptr[smp_processor_id()];
137         lc = lowcore_ptr[0];
138         if (!lc)
139                 lc = current_lc;
140         lc->restart_psw.mask = PSW_BASE_BITS | PSW_DEFAULT_KEY;
141         lc->restart_psw.addr = PSW_ADDR_AMODE | (unsigned long) smp_restart_cpu;
142         if (!cpu_online(0))
143                 smp_switch_to_cpu(func, data, 0, stap(), __cpu_logical_map[0]);
144         while (sigp(0, sigp_stop_and_store_status) == sigp_busy)
145                 cpu_relax();
146         sp = lc->panic_stack;
147         sp -= sizeof(struct pt_regs);
148         regs = (struct pt_regs *) sp;
149         memcpy(&regs->gprs, &current_lc->gpregs_save_area, sizeof(regs->gprs));
150         regs->psw = lc->psw_save_area;
151         sp -= STACK_FRAME_OVERHEAD;
152         sf = (struct stack_frame *) sp;
153         sf->back_chain = regs->gprs[15];
154         smp_switch_to_cpu(func, data, sp, stap(), __cpu_logical_map[0]);
155 }
156
157 void smp_send_stop(void)
158 {
159         int cpu, rc;
160
161         /* Disable all interrupts/machine checks */
162         __load_psw_mask(psw_kernel_bits & ~PSW_MASK_MCHECK);
163         trace_hardirqs_off();
164
165         /* stop all processors */
166         for_each_online_cpu(cpu) {
167                 if (cpu == smp_processor_id())
168                         continue;
169                 do {
170                         rc = sigp(cpu, sigp_stop);
171                 } while (rc == sigp_busy);
172
173                 while (!cpu_stopped(cpu))
174                         cpu_relax();
175         }
176 }
177
178 /*
179  * This is the main routine where commands issued by other
180  * cpus are handled.
181  */
182
183 static void do_ext_call_interrupt(unsigned int ext_int_code,
184                                   unsigned int param32, unsigned long param64)
185 {
186         unsigned long bits;
187
188         kstat_cpu(smp_processor_id()).irqs[EXTINT_IPI]++;
189         /*
190          * handle bit signal external calls
191          */
192         bits = xchg(&S390_lowcore.ext_call_fast, 0);
193
194         if (test_bit(ec_schedule, &bits))
195                 scheduler_ipi();
196
197         if (test_bit(ec_call_function, &bits))
198                 generic_smp_call_function_interrupt();
199
200         if (test_bit(ec_call_function_single, &bits))
201                 generic_smp_call_function_single_interrupt();
202 }
203
204 /*
205  * Send an external call sigp to another cpu and return without waiting
206  * for its completion.
207  */
208 static void smp_ext_bitcall(int cpu, int sig)
209 {
210         /*
211          * Set signaling bit in lowcore of target cpu and kick it
212          */
213         set_bit(sig, (unsigned long *) &lowcore_ptr[cpu]->ext_call_fast);
214         while (sigp(cpu, sigp_emergency_signal) == sigp_busy)
215                 udelay(10);
216 }
217
218 void arch_send_call_function_ipi_mask(const struct cpumask *mask)
219 {
220         int cpu;
221
222         for_each_cpu(cpu, mask)
223                 smp_ext_bitcall(cpu, ec_call_function);
224 }
225
226 void arch_send_call_function_single_ipi(int cpu)
227 {
228         smp_ext_bitcall(cpu, ec_call_function_single);
229 }
230
231 #ifndef CONFIG_64BIT
232 /*
233  * this function sends a 'purge tlb' signal to another CPU.
234  */
235 static void smp_ptlb_callback(void *info)
236 {
237         __tlb_flush_local();
238 }
239
240 void smp_ptlb_all(void)
241 {
242         on_each_cpu(smp_ptlb_callback, NULL, 1);
243 }
244 EXPORT_SYMBOL(smp_ptlb_all);
245 #endif /* ! CONFIG_64BIT */
246
247 /*
248  * this function sends a 'reschedule' IPI to another CPU.
249  * it goes straight through and wastes no time serializing
250  * anything. Worst case is that we lose a reschedule ...
251  */
252 void smp_send_reschedule(int cpu)
253 {
254         smp_ext_bitcall(cpu, ec_schedule);
255 }
256
257 /*
258  * parameter area for the set/clear control bit callbacks
259  */
260 struct ec_creg_mask_parms {
261         unsigned long orvals[16];
262         unsigned long andvals[16];
263 };
264
265 /*
266  * callback for setting/clearing control bits
267  */
268 static void smp_ctl_bit_callback(void *info)
269 {
270         struct ec_creg_mask_parms *pp = info;
271         unsigned long cregs[16];
272         int i;
273
274         __ctl_store(cregs, 0, 15);
275         for (i = 0; i <= 15; i++)
276                 cregs[i] = (cregs[i] & pp->andvals[i]) | pp->orvals[i];
277         __ctl_load(cregs, 0, 15);
278 }
279
280 /*
281  * Set a bit in a control register of all cpus
282  */
283 void smp_ctl_set_bit(int cr, int bit)
284 {
285         struct ec_creg_mask_parms parms;
286
287         memset(&parms.orvals, 0, sizeof(parms.orvals));
288         memset(&parms.andvals, 0xff, sizeof(parms.andvals));
289         parms.orvals[cr] = 1UL << bit;
290         on_each_cpu(smp_ctl_bit_callback, &parms, 1);
291 }
292 EXPORT_SYMBOL(smp_ctl_set_bit);
293
294 /*
295  * Clear a bit in a control register of all cpus
296  */
297 void smp_ctl_clear_bit(int cr, int bit)
298 {
299         struct ec_creg_mask_parms parms;
300
301         memset(&parms.orvals, 0, sizeof(parms.orvals));
302         memset(&parms.andvals, 0xff, sizeof(parms.andvals));
303         parms.andvals[cr] = ~(1UL << bit);
304         on_each_cpu(smp_ctl_bit_callback, &parms, 1);
305 }
306 EXPORT_SYMBOL(smp_ctl_clear_bit);
307
308 #if defined(CONFIG_ZFCPDUMP) || defined(CONFIG_CRASH_DUMP)
309
310 static void __init smp_get_save_area(unsigned int cpu, unsigned int phy_cpu)
311 {
312         if (ipl_info.type != IPL_TYPE_FCP_DUMP && !OLDMEM_BASE)
313                 return;
314         if (is_kdump_kernel())
315                 return;
316         if (cpu >= NR_CPUS) {
317                 pr_warning("CPU %i exceeds the maximum %i and is excluded from "
318                            "the dump\n", cpu, NR_CPUS - 1);
319                 return;
320         }
321         zfcpdump_save_areas[cpu] = kmalloc(sizeof(struct save_area), GFP_KERNEL);
322         while (raw_sigp(phy_cpu, sigp_stop_and_store_status) == sigp_busy)
323                 cpu_relax();
324         memcpy_real(zfcpdump_save_areas[cpu],
325                     (void *)(unsigned long) store_prefix() + SAVE_AREA_BASE,
326                     sizeof(struct save_area));
327 }
328
329 struct save_area *zfcpdump_save_areas[NR_CPUS + 1];
330 EXPORT_SYMBOL_GPL(zfcpdump_save_areas);
331
332 #else
333
334 static inline void smp_get_save_area(unsigned int cpu, unsigned int phy_cpu) { }
335
336 #endif /* CONFIG_ZFCPDUMP */
337
338 static int cpu_known(int cpu_id)
339 {
340         int cpu;
341
342         for_each_present_cpu(cpu) {
343                 if (__cpu_logical_map[cpu] == cpu_id)
344                         return 1;
345         }
346         return 0;
347 }
348
349 static int smp_rescan_cpus_sigp(cpumask_t avail)
350 {
351         int cpu_id, logical_cpu;
352
353         logical_cpu = cpumask_first(&avail);
354         if (logical_cpu >= nr_cpu_ids)
355                 return 0;
356         for (cpu_id = 0; cpu_id <= MAX_CPU_ADDRESS; cpu_id++) {
357                 if (cpu_known(cpu_id))
358                         continue;
359                 __cpu_logical_map[logical_cpu] = cpu_id;
360                 smp_cpu_polarization[logical_cpu] = POLARIZATION_UNKNWN;
361                 if (!cpu_stopped(logical_cpu))
362                         continue;
363                 set_cpu_present(logical_cpu, true);
364                 smp_cpu_state[logical_cpu] = CPU_STATE_CONFIGURED;
365                 logical_cpu = cpumask_next(logical_cpu, &avail);
366                 if (logical_cpu >= nr_cpu_ids)
367                         break;
368         }
369         return 0;
370 }
371
372 static int smp_rescan_cpus_sclp(cpumask_t avail)
373 {
374         struct sclp_cpu_info *info;
375         int cpu_id, logical_cpu, cpu;
376         int rc;
377
378         logical_cpu = cpumask_first(&avail);
379         if (logical_cpu >= nr_cpu_ids)
380                 return 0;
381         info = kmalloc(sizeof(*info), GFP_KERNEL);
382         if (!info)
383                 return -ENOMEM;
384         rc = sclp_get_cpu_info(info);
385         if (rc)
386                 goto out;
387         for (cpu = 0; cpu < info->combined; cpu++) {
388                 if (info->has_cpu_type && info->cpu[cpu].type != smp_cpu_type)
389                         continue;
390                 cpu_id = info->cpu[cpu].address;
391                 if (cpu_known(cpu_id))
392                         continue;
393                 __cpu_logical_map[logical_cpu] = cpu_id;
394                 smp_cpu_polarization[logical_cpu] = POLARIZATION_UNKNWN;
395                 set_cpu_present(logical_cpu, true);
396                 if (cpu >= info->configured)
397                         smp_cpu_state[logical_cpu] = CPU_STATE_STANDBY;
398                 else
399                         smp_cpu_state[logical_cpu] = CPU_STATE_CONFIGURED;
400                 logical_cpu = cpumask_next(logical_cpu, &avail);
401                 if (logical_cpu >= nr_cpu_ids)
402                         break;
403         }
404 out:
405         kfree(info);
406         return rc;
407 }
408
409 static int __smp_rescan_cpus(void)
410 {
411         cpumask_t avail;
412
413         cpumask_xor(&avail, cpu_possible_mask, cpu_present_mask);
414         if (smp_use_sigp_detection)
415                 return smp_rescan_cpus_sigp(avail);
416         else
417                 return smp_rescan_cpus_sclp(avail);
418 }
419
420 static void __init smp_detect_cpus(void)
421 {
422         unsigned int cpu, c_cpus, s_cpus;
423         struct sclp_cpu_info *info;
424         u16 boot_cpu_addr, cpu_addr;
425
426         c_cpus = 1;
427         s_cpus = 0;
428         boot_cpu_addr = __cpu_logical_map[0];
429         info = kmalloc(sizeof(*info), GFP_KERNEL);
430         if (!info)
431                 panic("smp_detect_cpus failed to allocate memory\n");
432 #ifdef CONFIG_CRASH_DUMP
433         if (OLDMEM_BASE && !is_kdump_kernel()) {
434                 struct save_area *save_area;
435
436                 save_area = kmalloc(sizeof(*save_area), GFP_KERNEL);
437                 if (!save_area)
438                         panic("could not allocate memory for save area\n");
439                 copy_oldmem_page(1, (void *) save_area, sizeof(*save_area),
440                                  0x200, 0);
441                 zfcpdump_save_areas[0] = save_area;
442         }
443 #endif
444         /* Use sigp detection algorithm if sclp doesn't work. */
445         if (sclp_get_cpu_info(info)) {
446                 smp_use_sigp_detection = 1;
447                 for (cpu = 0; cpu <= MAX_CPU_ADDRESS; cpu++) {
448                         if (cpu == boot_cpu_addr)
449                                 continue;
450                         if (!raw_cpu_stopped(cpu))
451                                 continue;
452                         smp_get_save_area(c_cpus, cpu);
453                         c_cpus++;
454                 }
455                 goto out;
456         }
457
458         if (info->has_cpu_type) {
459                 for (cpu = 0; cpu < info->combined; cpu++) {
460                         if (info->cpu[cpu].address == boot_cpu_addr) {
461                                 smp_cpu_type = info->cpu[cpu].type;
462                                 break;
463                         }
464                 }
465         }
466
467         for (cpu = 0; cpu < info->combined; cpu++) {
468                 if (info->has_cpu_type && info->cpu[cpu].type != smp_cpu_type)
469                         continue;
470                 cpu_addr = info->cpu[cpu].address;
471                 if (cpu_addr == boot_cpu_addr)
472                         continue;
473                 if (!raw_cpu_stopped(cpu_addr)) {
474                         s_cpus++;
475                         continue;
476                 }
477                 smp_get_save_area(c_cpus, cpu_addr);
478                 c_cpus++;
479         }
480 out:
481         kfree(info);
482         pr_info("%d configured CPUs, %d standby CPUs\n", c_cpus, s_cpus);
483         get_online_cpus();
484         __smp_rescan_cpus();
485         put_online_cpus();
486 }
487
488 /*
489  *      Activate a secondary processor.
490  */
491 int __cpuinit start_secondary(void *cpuvoid)
492 {
493         cpu_init();
494         preempt_disable();
495         init_cpu_timer();
496         init_cpu_vtimer();
497         pfault_init();
498
499         notify_cpu_starting(smp_processor_id());
500         ipi_call_lock();
501         set_cpu_online(smp_processor_id(), true);
502         ipi_call_unlock();
503         __ctl_clear_bit(0, 28); /* Disable lowcore protection */
504         S390_lowcore.restart_psw.mask = PSW_BASE_BITS | PSW_DEFAULT_KEY;
505         S390_lowcore.restart_psw.addr =
506                 PSW_ADDR_AMODE | (unsigned long) psw_restart_int_handler;
507         __ctl_set_bit(0, 28); /* Enable lowcore protection */
508         /*
509          * Wait until the cpu which brought this one up marked it
510          * active before enabling interrupts.
511          */
512         while (!cpumask_test_cpu(smp_processor_id(), cpu_active_mask))
513                 cpu_relax();
514         local_irq_enable();
515         /* cpu_idle will call schedule for us */
516         cpu_idle();
517         return 0;
518 }
519
520 struct create_idle {
521         struct work_struct work;
522         struct task_struct *idle;
523         struct completion done;
524         int cpu;
525 };
526
527 static void __cpuinit smp_fork_idle(struct work_struct *work)
528 {
529         struct create_idle *c_idle;
530
531         c_idle = container_of(work, struct create_idle, work);
532         c_idle->idle = fork_idle(c_idle->cpu);
533         complete(&c_idle->done);
534 }
535
536 static int __cpuinit smp_alloc_lowcore(int cpu)
537 {
538         unsigned long async_stack, panic_stack;
539         struct _lowcore *lowcore;
540
541         lowcore = (void *) __get_free_pages(GFP_KERNEL | GFP_DMA, LC_ORDER);
542         if (!lowcore)
543                 return -ENOMEM;
544         async_stack = __get_free_pages(GFP_KERNEL, ASYNC_ORDER);
545         panic_stack = __get_free_page(GFP_KERNEL);
546         if (!panic_stack || !async_stack)
547                 goto out;
548         memcpy(lowcore, &S390_lowcore, 512);
549         memset((char *)lowcore + 512, 0, sizeof(*lowcore) - 512);
550         lowcore->async_stack = async_stack + ASYNC_SIZE;
551         lowcore->panic_stack = panic_stack + PAGE_SIZE;
552         lowcore->restart_psw.mask = PSW_BASE_BITS | PSW_DEFAULT_KEY;
553         lowcore->restart_psw.addr =
554                 PSW_ADDR_AMODE | (unsigned long) restart_int_handler;
555         if (user_mode != HOME_SPACE_MODE)
556                 lowcore->restart_psw.mask |= PSW_ASC_HOME;
557 #ifndef CONFIG_64BIT
558         if (MACHINE_HAS_IEEE) {
559                 unsigned long save_area;
560
561                 save_area = get_zeroed_page(GFP_KERNEL);
562                 if (!save_area)
563                         goto out;
564                 lowcore->extended_save_area_addr = (u32) save_area;
565         }
566 #else
567         if (vdso_alloc_per_cpu(cpu, lowcore))
568                 goto out;
569 #endif
570         lowcore_ptr[cpu] = lowcore;
571         return 0;
572
573 out:
574         free_page(panic_stack);
575         free_pages(async_stack, ASYNC_ORDER);
576         free_pages((unsigned long) lowcore, LC_ORDER);
577         return -ENOMEM;
578 }
579
580 static void smp_free_lowcore(int cpu)
581 {
582         struct _lowcore *lowcore;
583
584         lowcore = lowcore_ptr[cpu];
585 #ifndef CONFIG_64BIT
586         if (MACHINE_HAS_IEEE)
587                 free_page((unsigned long) lowcore->extended_save_area_addr);
588 #else
589         vdso_free_per_cpu(cpu, lowcore);
590 #endif
591         free_page(lowcore->panic_stack - PAGE_SIZE);
592         free_pages(lowcore->async_stack - ASYNC_SIZE, ASYNC_ORDER);
593         free_pages((unsigned long) lowcore, LC_ORDER);
594         lowcore_ptr[cpu] = NULL;
595 }
596
597 /* Upping and downing of CPUs */
598 int __cpuinit __cpu_up(unsigned int cpu)
599 {
600         struct _lowcore *cpu_lowcore;
601         struct create_idle c_idle;
602         struct task_struct *idle;
603         struct stack_frame *sf;
604         u32 lowcore;
605         int ccode;
606
607         if (smp_cpu_state[cpu] != CPU_STATE_CONFIGURED)
608                 return -EIO;
609         idle = current_set[cpu];
610         if (!idle) {
611                 c_idle.done = COMPLETION_INITIALIZER_ONSTACK(c_idle.done);
612                 INIT_WORK_ONSTACK(&c_idle.work, smp_fork_idle);
613                 c_idle.cpu = cpu;
614                 schedule_work(&c_idle.work);
615                 wait_for_completion(&c_idle.done);
616                 if (IS_ERR(c_idle.idle))
617                         return PTR_ERR(c_idle.idle);
618                 idle = c_idle.idle;
619                 current_set[cpu] = c_idle.idle;
620         }
621         init_idle(idle, cpu);
622         if (smp_alloc_lowcore(cpu))
623                 return -ENOMEM;
624         do {
625                 ccode = sigp(cpu, sigp_initial_cpu_reset);
626                 if (ccode == sigp_busy)
627                         udelay(10);
628                 if (ccode == sigp_not_operational)
629                         goto err_out;
630         } while (ccode == sigp_busy);
631
632         lowcore = (u32)(unsigned long)lowcore_ptr[cpu];
633         while (sigp_p(lowcore, cpu, sigp_set_prefix) == sigp_busy)
634                 udelay(10);
635
636         cpu_lowcore = lowcore_ptr[cpu];
637         cpu_lowcore->kernel_stack = (unsigned long)
638                 task_stack_page(idle) + THREAD_SIZE;
639         cpu_lowcore->thread_info = (unsigned long) task_thread_info(idle);
640         sf = (struct stack_frame *) (cpu_lowcore->kernel_stack
641                                      - sizeof(struct pt_regs)
642                                      - sizeof(struct stack_frame));
643         memset(sf, 0, sizeof(struct stack_frame));
644         sf->gprs[9] = (unsigned long) sf;
645         cpu_lowcore->save_area[15] = (unsigned long) sf;
646         __ctl_store(cpu_lowcore->cregs_save_area, 0, 15);
647         atomic_inc(&init_mm.context.attach_count);
648         asm volatile(
649                 "       stam    0,15,0(%0)"
650                 : : "a" (&cpu_lowcore->access_regs_save_area) : "memory");
651         cpu_lowcore->percpu_offset = __per_cpu_offset[cpu];
652         cpu_lowcore->current_task = (unsigned long) idle;
653         cpu_lowcore->cpu_nr = cpu;
654         cpu_lowcore->kernel_asce = S390_lowcore.kernel_asce;
655         cpu_lowcore->machine_flags = S390_lowcore.machine_flags;
656         cpu_lowcore->ftrace_func = S390_lowcore.ftrace_func;
657         memcpy(cpu_lowcore->stfle_fac_list, S390_lowcore.stfle_fac_list,
658                MAX_FACILITY_BIT/8);
659         eieio();
660
661         while (sigp(cpu, sigp_restart) == sigp_busy)
662                 udelay(10);
663
664         while (!cpu_online(cpu))
665                 cpu_relax();
666         return 0;
667
668 err_out:
669         smp_free_lowcore(cpu);
670         return -EIO;
671 }
672
673 static int __init setup_possible_cpus(char *s)
674 {
675         int pcpus, cpu;
676
677         pcpus = simple_strtoul(s, NULL, 0);
678         init_cpu_possible(cpumask_of(0));
679         for (cpu = 1; cpu < pcpus && cpu < nr_cpu_ids; cpu++)
680                 set_cpu_possible(cpu, true);
681         return 0;
682 }
683 early_param("possible_cpus", setup_possible_cpus);
684
685 #ifdef CONFIG_HOTPLUG_CPU
686
687 int __cpu_disable(void)
688 {
689         struct ec_creg_mask_parms cr_parms;
690         int cpu = smp_processor_id();
691
692         set_cpu_online(cpu, false);
693
694         /* Disable pfault pseudo page faults on this cpu. */
695         pfault_fini();
696
697         memset(&cr_parms.orvals, 0, sizeof(cr_parms.orvals));
698         memset(&cr_parms.andvals, 0xff, sizeof(cr_parms.andvals));
699
700         /* disable all external interrupts */
701         cr_parms.orvals[0] = 0;
702         cr_parms.andvals[0] = ~(1 << 15 | 1 << 14 | 1 << 13 | 1 << 11 |
703                                 1 << 10 | 1 <<  9 | 1 <<  6 | 1 <<  5 |
704                                 1 <<  4);
705         /* disable all I/O interrupts */
706         cr_parms.orvals[6] = 0;
707         cr_parms.andvals[6] = ~(1 << 31 | 1 << 30 | 1 << 29 | 1 << 28 |
708                                 1 << 27 | 1 << 26 | 1 << 25 | 1 << 24);
709         /* disable most machine checks */
710         cr_parms.orvals[14] = 0;
711         cr_parms.andvals[14] = ~(1 << 28 | 1 << 27 | 1 << 26 |
712                                  1 << 25 | 1 << 24);
713
714         smp_ctl_bit_callback(&cr_parms);
715
716         return 0;
717 }
718
719 void __cpu_die(unsigned int cpu)
720 {
721         /* Wait until target cpu is down */
722         while (!cpu_stopped(cpu))
723                 cpu_relax();
724         while (sigp_p(0, cpu, sigp_set_prefix) == sigp_busy)
725                 udelay(10);
726         smp_free_lowcore(cpu);
727         atomic_dec(&init_mm.context.attach_count);
728 }
729
730 void __noreturn cpu_die(void)
731 {
732         idle_task_exit();
733         while (sigp(smp_processor_id(), sigp_stop) == sigp_busy)
734                 cpu_relax();
735         for (;;);
736 }
737
738 #endif /* CONFIG_HOTPLUG_CPU */
739
740 void __init smp_prepare_cpus(unsigned int max_cpus)
741 {
742 #ifndef CONFIG_64BIT
743         unsigned long save_area = 0;
744 #endif
745         unsigned long async_stack, panic_stack;
746         struct _lowcore *lowcore;
747
748         smp_detect_cpus();
749
750         /* request the 0x1201 emergency signal external interrupt */
751         if (register_external_interrupt(0x1201, do_ext_call_interrupt) != 0)
752                 panic("Couldn't request external interrupt 0x1201");
753
754         /* Reallocate current lowcore, but keep its contents. */
755         lowcore = (void *) __get_free_pages(GFP_KERNEL | GFP_DMA, LC_ORDER);
756         panic_stack = __get_free_page(GFP_KERNEL);
757         async_stack = __get_free_pages(GFP_KERNEL, ASYNC_ORDER);
758         BUG_ON(!lowcore || !panic_stack || !async_stack);
759 #ifndef CONFIG_64BIT
760         if (MACHINE_HAS_IEEE)
761                 save_area = get_zeroed_page(GFP_KERNEL);
762 #endif
763         local_irq_disable();
764         local_mcck_disable();
765         lowcore_ptr[smp_processor_id()] = lowcore;
766         *lowcore = S390_lowcore;
767         lowcore->panic_stack = panic_stack + PAGE_SIZE;
768         lowcore->async_stack = async_stack + ASYNC_SIZE;
769 #ifndef CONFIG_64BIT
770         if (MACHINE_HAS_IEEE)
771                 lowcore->extended_save_area_addr = (u32) save_area;
772 #endif
773         set_prefix((u32)(unsigned long) lowcore);
774         local_mcck_enable();
775         local_irq_enable();
776 #ifdef CONFIG_64BIT
777         if (vdso_alloc_per_cpu(smp_processor_id(), &S390_lowcore))
778                 BUG();
779 #endif
780 }
781
782 void __init smp_prepare_boot_cpu(void)
783 {
784         BUG_ON(smp_processor_id() != 0);
785
786         current_thread_info()->cpu = 0;
787         set_cpu_present(0, true);
788         set_cpu_online(0, true);
789         S390_lowcore.percpu_offset = __per_cpu_offset[0];
790         current_set[0] = current;
791         smp_cpu_state[0] = CPU_STATE_CONFIGURED;
792         smp_cpu_polarization[0] = POLARIZATION_UNKNWN;
793 }
794
795 void __init smp_cpus_done(unsigned int max_cpus)
796 {
797 }
798
799 void __init smp_setup_processor_id(void)
800 {
801         S390_lowcore.cpu_nr = 0;
802         __cpu_logical_map[0] = stap();
803 }
804
805 /*
806  * the frequency of the profiling timer can be changed
807  * by writing a multiplier value into /proc/profile.
808  *
809  * usually you want to run this on all CPUs ;)
810  */
811 int setup_profiling_timer(unsigned int multiplier)
812 {
813         return 0;
814 }
815
816 #ifdef CONFIG_HOTPLUG_CPU
817 static ssize_t cpu_configure_show(struct sys_device *dev,
818                                 struct sysdev_attribute *attr, char *buf)
819 {
820         ssize_t count;
821
822         mutex_lock(&smp_cpu_state_mutex);
823         count = sprintf(buf, "%d\n", smp_cpu_state[dev->id]);
824         mutex_unlock(&smp_cpu_state_mutex);
825         return count;
826 }
827
828 static ssize_t cpu_configure_store(struct sys_device *dev,
829                                   struct sysdev_attribute *attr,
830                                   const char *buf, size_t count)
831 {
832         int cpu = dev->id;
833         int val, rc;
834         char delim;
835
836         if (sscanf(buf, "%d %c", &val, &delim) != 1)
837                 return -EINVAL;
838         if (val != 0 && val != 1)
839                 return -EINVAL;
840
841         get_online_cpus();
842         mutex_lock(&smp_cpu_state_mutex);
843         rc = -EBUSY;
844         /* disallow configuration changes of online cpus and cpu 0 */
845         if (cpu_online(cpu) || cpu == 0)
846                 goto out;
847         rc = 0;
848         switch (val) {
849         case 0:
850                 if (smp_cpu_state[cpu] == CPU_STATE_CONFIGURED) {
851                         rc = sclp_cpu_deconfigure(__cpu_logical_map[cpu]);
852                         if (!rc) {
853                                 smp_cpu_state[cpu] = CPU_STATE_STANDBY;
854                                 smp_cpu_polarization[cpu] = POLARIZATION_UNKNWN;
855                         }
856                 }
857                 break;
858         case 1:
859                 if (smp_cpu_state[cpu] == CPU_STATE_STANDBY) {
860                         rc = sclp_cpu_configure(__cpu_logical_map[cpu]);
861                         if (!rc) {
862                                 smp_cpu_state[cpu] = CPU_STATE_CONFIGURED;
863                                 smp_cpu_polarization[cpu] = POLARIZATION_UNKNWN;
864                         }
865                 }
866                 break;
867         default:
868                 break;
869         }
870 out:
871         mutex_unlock(&smp_cpu_state_mutex);
872         put_online_cpus();
873         return rc ? rc : count;
874 }
875 static SYSDEV_ATTR(configure, 0644, cpu_configure_show, cpu_configure_store);
876 #endif /* CONFIG_HOTPLUG_CPU */
877
878 static ssize_t cpu_polarization_show(struct sys_device *dev,
879                                      struct sysdev_attribute *attr, char *buf)
880 {
881         int cpu = dev->id;
882         ssize_t count;
883
884         mutex_lock(&smp_cpu_state_mutex);
885         switch (smp_cpu_polarization[cpu]) {
886         case POLARIZATION_HRZ:
887                 count = sprintf(buf, "horizontal\n");
888                 break;
889         case POLARIZATION_VL:
890                 count = sprintf(buf, "vertical:low\n");
891                 break;
892         case POLARIZATION_VM:
893                 count = sprintf(buf, "vertical:medium\n");
894                 break;
895         case POLARIZATION_VH:
896                 count = sprintf(buf, "vertical:high\n");
897                 break;
898         default:
899                 count = sprintf(buf, "unknown\n");
900                 break;
901         }
902         mutex_unlock(&smp_cpu_state_mutex);
903         return count;
904 }
905 static SYSDEV_ATTR(polarization, 0444, cpu_polarization_show, NULL);
906
907 static ssize_t show_cpu_address(struct sys_device *dev,
908                                 struct sysdev_attribute *attr, char *buf)
909 {
910         return sprintf(buf, "%d\n", __cpu_logical_map[dev->id]);
911 }
912 static SYSDEV_ATTR(address, 0444, show_cpu_address, NULL);
913
914
915 static struct attribute *cpu_common_attrs[] = {
916 #ifdef CONFIG_HOTPLUG_CPU
917         &attr_configure.attr,
918 #endif
919         &attr_address.attr,
920         &attr_polarization.attr,
921         NULL,
922 };
923
924 static struct attribute_group cpu_common_attr_group = {
925         .attrs = cpu_common_attrs,
926 };
927
928 static ssize_t show_capability(struct sys_device *dev,
929                                 struct sysdev_attribute *attr, char *buf)
930 {
931         unsigned int capability;
932         int rc;
933
934         rc = get_cpu_capability(&capability);
935         if (rc)
936                 return rc;
937         return sprintf(buf, "%u\n", capability);
938 }
939 static SYSDEV_ATTR(capability, 0444, show_capability, NULL);
940
941 static ssize_t show_idle_count(struct sys_device *dev,
942                                 struct sysdev_attribute *attr, char *buf)
943 {
944         struct s390_idle_data *idle;
945         unsigned long long idle_count;
946         unsigned int sequence;
947
948         idle = &per_cpu(s390_idle, dev->id);
949 repeat:
950         sequence = idle->sequence;
951         smp_rmb();
952         if (sequence & 1)
953                 goto repeat;
954         idle_count = idle->idle_count;
955         if (idle->idle_enter)
956                 idle_count++;
957         smp_rmb();
958         if (idle->sequence != sequence)
959                 goto repeat;
960         return sprintf(buf, "%llu\n", idle_count);
961 }
962 static SYSDEV_ATTR(idle_count, 0444, show_idle_count, NULL);
963
964 static ssize_t show_idle_time(struct sys_device *dev,
965                                 struct sysdev_attribute *attr, char *buf)
966 {
967         struct s390_idle_data *idle;
968         unsigned long long now, idle_time, idle_enter;
969         unsigned int sequence;
970
971         idle = &per_cpu(s390_idle, dev->id);
972         now = get_clock();
973 repeat:
974         sequence = idle->sequence;
975         smp_rmb();
976         if (sequence & 1)
977                 goto repeat;
978         idle_time = idle->idle_time;
979         idle_enter = idle->idle_enter;
980         if (idle_enter != 0ULL && idle_enter < now)
981                 idle_time += now - idle_enter;
982         smp_rmb();
983         if (idle->sequence != sequence)
984                 goto repeat;
985         return sprintf(buf, "%llu\n", idle_time >> 12);
986 }
987 static SYSDEV_ATTR(idle_time_us, 0444, show_idle_time, NULL);
988
989 static struct attribute *cpu_online_attrs[] = {
990         &attr_capability.attr,
991         &attr_idle_count.attr,
992         &attr_idle_time_us.attr,
993         NULL,
994 };
995
996 static struct attribute_group cpu_online_attr_group = {
997         .attrs = cpu_online_attrs,
998 };
999
1000 static int __cpuinit smp_cpu_notify(struct notifier_block *self,
1001                                     unsigned long action, void *hcpu)
1002 {
1003         unsigned int cpu = (unsigned int)(long)hcpu;
1004         struct cpu *c = &per_cpu(cpu_devices, cpu);
1005         struct sys_device *s = &c->sysdev;
1006         struct s390_idle_data *idle;
1007         int err = 0;
1008
1009         switch (action) {
1010         case CPU_ONLINE:
1011         case CPU_ONLINE_FROZEN:
1012                 idle = &per_cpu(s390_idle, cpu);
1013                 memset(idle, 0, sizeof(struct s390_idle_data));
1014                 err = sysfs_create_group(&s->kobj, &cpu_online_attr_group);
1015                 break;
1016         case CPU_DEAD:
1017         case CPU_DEAD_FROZEN:
1018                 sysfs_remove_group(&s->kobj, &cpu_online_attr_group);
1019                 break;
1020         }
1021         return notifier_from_errno(err);
1022 }
1023
1024 static struct notifier_block __cpuinitdata smp_cpu_nb = {
1025         .notifier_call = smp_cpu_notify,
1026 };
1027
1028 static int __devinit smp_add_present_cpu(int cpu)
1029 {
1030         struct cpu *c = &per_cpu(cpu_devices, cpu);
1031         struct sys_device *s = &c->sysdev;
1032         int rc;
1033
1034         c->hotpluggable = 1;
1035         rc = register_cpu(c, cpu);
1036         if (rc)
1037                 goto out;
1038         rc = sysfs_create_group(&s->kobj, &cpu_common_attr_group);
1039         if (rc)
1040                 goto out_cpu;
1041         if (!cpu_online(cpu))
1042                 goto out;
1043         rc = sysfs_create_group(&s->kobj, &cpu_online_attr_group);
1044         if (!rc)
1045                 return 0;
1046         sysfs_remove_group(&s->kobj, &cpu_common_attr_group);
1047 out_cpu:
1048 #ifdef CONFIG_HOTPLUG_CPU
1049         unregister_cpu(c);
1050 #endif
1051 out:
1052         return rc;
1053 }
1054
1055 #ifdef CONFIG_HOTPLUG_CPU
1056
1057 int __ref smp_rescan_cpus(void)
1058 {
1059         cpumask_t newcpus;
1060         int cpu;
1061         int rc;
1062
1063         get_online_cpus();
1064         mutex_lock(&smp_cpu_state_mutex);
1065         cpumask_copy(&newcpus, cpu_present_mask);
1066         rc = __smp_rescan_cpus();
1067         if (rc)
1068                 goto out;
1069         cpumask_andnot(&newcpus, cpu_present_mask, &newcpus);
1070         for_each_cpu(cpu, &newcpus) {
1071                 rc = smp_add_present_cpu(cpu);
1072                 if (rc)
1073                         set_cpu_present(cpu, false);
1074         }
1075         rc = 0;
1076 out:
1077         mutex_unlock(&smp_cpu_state_mutex);
1078         put_online_cpus();
1079         if (!cpumask_empty(&newcpus))
1080                 topology_schedule_update();
1081         return rc;
1082 }
1083
1084 static ssize_t __ref rescan_store(struct sysdev_class *class,
1085                                   struct sysdev_class_attribute *attr,
1086                                   const char *buf,
1087                                   size_t count)
1088 {
1089         int rc;
1090
1091         rc = smp_rescan_cpus();
1092         return rc ? rc : count;
1093 }
1094 static SYSDEV_CLASS_ATTR(rescan, 0200, NULL, rescan_store);
1095 #endif /* CONFIG_HOTPLUG_CPU */
1096
1097 static ssize_t dispatching_show(struct sysdev_class *class,
1098                                 struct sysdev_class_attribute *attr,
1099                                 char *buf)
1100 {
1101         ssize_t count;
1102
1103         mutex_lock(&smp_cpu_state_mutex);
1104         count = sprintf(buf, "%d\n", cpu_management);
1105         mutex_unlock(&smp_cpu_state_mutex);
1106         return count;
1107 }
1108
1109 static ssize_t dispatching_store(struct sysdev_class *dev,
1110                                  struct sysdev_class_attribute *attr,
1111                                  const char *buf,
1112                                  size_t count)
1113 {
1114         int val, rc;
1115         char delim;
1116
1117         if (sscanf(buf, "%d %c", &val, &delim) != 1)
1118                 return -EINVAL;
1119         if (val != 0 && val != 1)
1120                 return -EINVAL;
1121         rc = 0;
1122         get_online_cpus();
1123         mutex_lock(&smp_cpu_state_mutex);
1124         if (cpu_management == val)
1125                 goto out;
1126         rc = topology_set_cpu_management(val);
1127         if (!rc)
1128                 cpu_management = val;
1129 out:
1130         mutex_unlock(&smp_cpu_state_mutex);
1131         put_online_cpus();
1132         return rc ? rc : count;
1133 }
1134 static SYSDEV_CLASS_ATTR(dispatching, 0644, dispatching_show,
1135                          dispatching_store);
1136
1137 static int __init topology_init(void)
1138 {
1139         int cpu;
1140         int rc;
1141
1142         register_cpu_notifier(&smp_cpu_nb);
1143
1144 #ifdef CONFIG_HOTPLUG_CPU
1145         rc = sysdev_class_create_file(&cpu_sysdev_class, &attr_rescan);
1146         if (rc)
1147                 return rc;
1148 #endif
1149         rc = sysdev_class_create_file(&cpu_sysdev_class, &attr_dispatching);
1150         if (rc)
1151                 return rc;
1152         for_each_present_cpu(cpu) {
1153                 rc = smp_add_present_cpu(cpu);
1154                 if (rc)
1155                         return rc;
1156         }
1157         return 0;
1158 }
1159 subsys_initcall(topology_init);