Loading...
1/*
2 * SMP related functions
3 *
4 * Copyright IBM Corp. 1999, 2012
5 * Author(s): Denis Joseph Barrow,
6 * Martin Schwidefsky <schwidefsky@de.ibm.com>,
7 * Heiko Carstens <heiko.carstens@de.ibm.com>,
8 *
9 * based on other smp stuff by
10 * (c) 1995 Alan Cox, CymruNET Ltd <alan@cymru.net>
11 * (c) 1998 Ingo Molnar
12 *
13 * The code outside of smp.c uses logical cpu numbers, only smp.c does
14 * the translation of logical to physical cpu ids. All new code that
15 * operates on physical cpu numbers needs to go into smp.c.
16 */
17
18#define KMSG_COMPONENT "cpu"
19#define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
20
21#include <linux/workqueue.h>
22#include <linux/bootmem.h>
23#include <linux/module.h>
24#include <linux/init.h>
25#include <linux/mm.h>
26#include <linux/err.h>
27#include <linux/spinlock.h>
28#include <linux/kernel_stat.h>
29#include <linux/delay.h>
30#include <linux/interrupt.h>
31#include <linux/irqflags.h>
32#include <linux/cpu.h>
33#include <linux/slab.h>
34#include <linux/crash_dump.h>
35#include <linux/memblock.h>
36#include <asm/asm-offsets.h>
37#include <asm/diag.h>
38#include <asm/switch_to.h>
39#include <asm/facility.h>
40#include <asm/ipl.h>
41#include <asm/setup.h>
42#include <asm/irq.h>
43#include <asm/tlbflush.h>
44#include <asm/vtimer.h>
45#include <asm/lowcore.h>
46#include <asm/sclp.h>
47#include <asm/vdso.h>
48#include <asm/debug.h>
49#include <asm/os_info.h>
50#include <asm/sigp.h>
51#include <asm/idle.h>
52#include "entry.h"
53
54enum {
55 ec_schedule = 0,
56 ec_call_function_single,
57 ec_stop_cpu,
58};
59
60enum {
61 CPU_STATE_STANDBY,
62 CPU_STATE_CONFIGURED,
63};
64
65static DEFINE_PER_CPU(struct cpu *, cpu_device);
66
67struct pcpu {
68 struct lowcore *lowcore; /* lowcore page(s) for the cpu */
69 unsigned long ec_mask; /* bit mask for ec_xxx functions */
70 unsigned long ec_clk; /* sigp timestamp for ec_xxx */
71 signed char state; /* physical cpu state */
72 signed char polarization; /* physical polarization */
73 u16 address; /* physical cpu address */
74};
75
76static u8 boot_core_type;
77static struct pcpu pcpu_devices[NR_CPUS];
78
79unsigned int smp_cpu_mt_shift;
80EXPORT_SYMBOL(smp_cpu_mt_shift);
81
82unsigned int smp_cpu_mtid;
83EXPORT_SYMBOL(smp_cpu_mtid);
84
85#ifdef CONFIG_CRASH_DUMP
86__vector128 __initdata boot_cpu_vector_save_area[__NUM_VXRS];
87#endif
88
89static unsigned int smp_max_threads __initdata = -1U;
90
91static int __init early_nosmt(char *s)
92{
93 smp_max_threads = 1;
94 return 0;
95}
96early_param("nosmt", early_nosmt);
97
98static int __init early_smt(char *s)
99{
100 get_option(&s, &smp_max_threads);
101 return 0;
102}
103early_param("smt", early_smt);
104
105/*
106 * The smp_cpu_state_mutex must be held when changing the state or polarization
107 * member of a pcpu data structure within the pcpu_devices arreay.
108 */
109DEFINE_MUTEX(smp_cpu_state_mutex);
110
111/*
112 * Signal processor helper functions.
113 */
114static inline int __pcpu_sigp_relax(u16 addr, u8 order, unsigned long parm)
115{
116 int cc;
117
118 while (1) {
119 cc = __pcpu_sigp(addr, order, parm, NULL);
120 if (cc != SIGP_CC_BUSY)
121 return cc;
122 cpu_relax();
123 }
124}
125
126static int pcpu_sigp_retry(struct pcpu *pcpu, u8 order, u32 parm)
127{
128 int cc, retry;
129
130 for (retry = 0; ; retry++) {
131 cc = __pcpu_sigp(pcpu->address, order, parm, NULL);
132 if (cc != SIGP_CC_BUSY)
133 break;
134 if (retry >= 3)
135 udelay(10);
136 }
137 return cc;
138}
139
140static inline int pcpu_stopped(struct pcpu *pcpu)
141{
142 u32 uninitialized_var(status);
143
144 if (__pcpu_sigp(pcpu->address, SIGP_SENSE,
145 0, &status) != SIGP_CC_STATUS_STORED)
146 return 0;
147 return !!(status & (SIGP_STATUS_CHECK_STOP|SIGP_STATUS_STOPPED));
148}
149
150static inline int pcpu_running(struct pcpu *pcpu)
151{
152 if (__pcpu_sigp(pcpu->address, SIGP_SENSE_RUNNING,
153 0, NULL) != SIGP_CC_STATUS_STORED)
154 return 1;
155 /* Status stored condition code is equivalent to cpu not running. */
156 return 0;
157}
158
159/*
160 * Find struct pcpu by cpu address.
161 */
162static struct pcpu *pcpu_find_address(const struct cpumask *mask, u16 address)
163{
164 int cpu;
165
166 for_each_cpu(cpu, mask)
167 if (pcpu_devices[cpu].address == address)
168 return pcpu_devices + cpu;
169 return NULL;
170}
171
172static void pcpu_ec_call(struct pcpu *pcpu, int ec_bit)
173{
174 int order;
175
176 if (test_and_set_bit(ec_bit, &pcpu->ec_mask))
177 return;
178 order = pcpu_running(pcpu) ? SIGP_EXTERNAL_CALL : SIGP_EMERGENCY_SIGNAL;
179 pcpu->ec_clk = get_tod_clock_fast();
180 pcpu_sigp_retry(pcpu, order, 0);
181}
182
183#define ASYNC_FRAME_OFFSET (ASYNC_SIZE - STACK_FRAME_OVERHEAD - __PT_SIZE)
184#define PANIC_FRAME_OFFSET (PAGE_SIZE - STACK_FRAME_OVERHEAD - __PT_SIZE)
185
186static int pcpu_alloc_lowcore(struct pcpu *pcpu, int cpu)
187{
188 unsigned long async_stack, panic_stack;
189 struct lowcore *lc;
190
191 if (pcpu != &pcpu_devices[0]) {
192 pcpu->lowcore = (struct lowcore *)
193 __get_free_pages(GFP_KERNEL | GFP_DMA, LC_ORDER);
194 async_stack = __get_free_pages(GFP_KERNEL, ASYNC_ORDER);
195 panic_stack = __get_free_page(GFP_KERNEL);
196 if (!pcpu->lowcore || !panic_stack || !async_stack)
197 goto out;
198 } else {
199 async_stack = pcpu->lowcore->async_stack - ASYNC_FRAME_OFFSET;
200 panic_stack = pcpu->lowcore->panic_stack - PANIC_FRAME_OFFSET;
201 }
202 lc = pcpu->lowcore;
203 memcpy(lc, &S390_lowcore, 512);
204 memset((char *) lc + 512, 0, sizeof(*lc) - 512);
205 lc->async_stack = async_stack + ASYNC_FRAME_OFFSET;
206 lc->panic_stack = panic_stack + PANIC_FRAME_OFFSET;
207 lc->cpu_nr = cpu;
208 lc->spinlock_lockval = arch_spin_lockval(cpu);
209 if (MACHINE_HAS_VX)
210 lc->vector_save_area_addr =
211 (unsigned long) &lc->vector_save_area;
212 if (vdso_alloc_per_cpu(lc))
213 goto out;
214 lowcore_ptr[cpu] = lc;
215 pcpu_sigp_retry(pcpu, SIGP_SET_PREFIX, (u32)(unsigned long) lc);
216 return 0;
217out:
218 if (pcpu != &pcpu_devices[0]) {
219 free_page(panic_stack);
220 free_pages(async_stack, ASYNC_ORDER);
221 free_pages((unsigned long) pcpu->lowcore, LC_ORDER);
222 }
223 return -ENOMEM;
224}
225
226#ifdef CONFIG_HOTPLUG_CPU
227
228static void pcpu_free_lowcore(struct pcpu *pcpu)
229{
230 pcpu_sigp_retry(pcpu, SIGP_SET_PREFIX, 0);
231 lowcore_ptr[pcpu - pcpu_devices] = NULL;
232 vdso_free_per_cpu(pcpu->lowcore);
233 if (pcpu == &pcpu_devices[0])
234 return;
235 free_page(pcpu->lowcore->panic_stack-PANIC_FRAME_OFFSET);
236 free_pages(pcpu->lowcore->async_stack-ASYNC_FRAME_OFFSET, ASYNC_ORDER);
237 free_pages((unsigned long) pcpu->lowcore, LC_ORDER);
238}
239
240#endif /* CONFIG_HOTPLUG_CPU */
241
242static void pcpu_prepare_secondary(struct pcpu *pcpu, int cpu)
243{
244 struct lowcore *lc = pcpu->lowcore;
245
246 cpumask_set_cpu(cpu, &init_mm.context.cpu_attach_mask);
247 cpumask_set_cpu(cpu, mm_cpumask(&init_mm));
248 lc->cpu_nr = cpu;
249 lc->spinlock_lockval = arch_spin_lockval(cpu);
250 lc->percpu_offset = __per_cpu_offset[cpu];
251 lc->kernel_asce = S390_lowcore.kernel_asce;
252 lc->machine_flags = S390_lowcore.machine_flags;
253 lc->user_timer = lc->system_timer = lc->steal_timer = 0;
254 __ctl_store(lc->cregs_save_area, 0, 15);
255 save_access_regs((unsigned int *) lc->access_regs_save_area);
256 memcpy(lc->stfle_fac_list, S390_lowcore.stfle_fac_list,
257 MAX_FACILITY_BIT/8);
258}
259
260static void pcpu_attach_task(struct pcpu *pcpu, struct task_struct *tsk)
261{
262 struct lowcore *lc = pcpu->lowcore;
263
264 lc->kernel_stack = (unsigned long) task_stack_page(tsk)
265 + THREAD_SIZE - STACK_FRAME_OVERHEAD - sizeof(struct pt_regs);
266 lc->current_task = (unsigned long) tsk;
267 lc->lpp = LPP_MAGIC;
268 lc->current_pid = tsk->pid;
269 lc->user_timer = tsk->thread.user_timer;
270 lc->system_timer = tsk->thread.system_timer;
271 lc->steal_timer = 0;
272}
273
274static void pcpu_start_fn(struct pcpu *pcpu, void (*func)(void *), void *data)
275{
276 struct lowcore *lc = pcpu->lowcore;
277
278 lc->restart_stack = lc->kernel_stack;
279 lc->restart_fn = (unsigned long) func;
280 lc->restart_data = (unsigned long) data;
281 lc->restart_source = -1UL;
282 pcpu_sigp_retry(pcpu, SIGP_RESTART, 0);
283}
284
285/*
286 * Call function via PSW restart on pcpu and stop the current cpu.
287 */
288static void pcpu_delegate(struct pcpu *pcpu, void (*func)(void *),
289 void *data, unsigned long stack)
290{
291 struct lowcore *lc = lowcore_ptr[pcpu - pcpu_devices];
292 unsigned long source_cpu = stap();
293
294 __load_psw_mask(PSW_KERNEL_BITS);
295 if (pcpu->address == source_cpu)
296 func(data); /* should not return */
297 /* Stop target cpu (if func returns this stops the current cpu). */
298 pcpu_sigp_retry(pcpu, SIGP_STOP, 0);
299 /* Restart func on the target cpu and stop the current cpu. */
300 mem_assign_absolute(lc->restart_stack, stack);
301 mem_assign_absolute(lc->restart_fn, (unsigned long) func);
302 mem_assign_absolute(lc->restart_data, (unsigned long) data);
303 mem_assign_absolute(lc->restart_source, source_cpu);
304 asm volatile(
305 "0: sigp 0,%0,%2 # sigp restart to target cpu\n"
306 " brc 2,0b # busy, try again\n"
307 "1: sigp 0,%1,%3 # sigp stop to current cpu\n"
308 " brc 2,1b # busy, try again\n"
309 : : "d" (pcpu->address), "d" (source_cpu),
310 "K" (SIGP_RESTART), "K" (SIGP_STOP)
311 : "0", "1", "cc");
312 for (;;) ;
313}
314
315/*
316 * Enable additional logical cpus for multi-threading.
317 */
318static int pcpu_set_smt(unsigned int mtid)
319{
320 int cc;
321
322 if (smp_cpu_mtid == mtid)
323 return 0;
324 cc = __pcpu_sigp(0, SIGP_SET_MULTI_THREADING, mtid, NULL);
325 if (cc == 0) {
326 smp_cpu_mtid = mtid;
327 smp_cpu_mt_shift = 0;
328 while (smp_cpu_mtid >= (1U << smp_cpu_mt_shift))
329 smp_cpu_mt_shift++;
330 pcpu_devices[0].address = stap();
331 }
332 return cc;
333}
334
335/*
336 * Call function on an online CPU.
337 */
338void smp_call_online_cpu(void (*func)(void *), void *data)
339{
340 struct pcpu *pcpu;
341
342 /* Use the current cpu if it is online. */
343 pcpu = pcpu_find_address(cpu_online_mask, stap());
344 if (!pcpu)
345 /* Use the first online cpu. */
346 pcpu = pcpu_devices + cpumask_first(cpu_online_mask);
347 pcpu_delegate(pcpu, func, data, (unsigned long) restart_stack);
348}
349
350/*
351 * Call function on the ipl CPU.
352 */
353void smp_call_ipl_cpu(void (*func)(void *), void *data)
354{
355 pcpu_delegate(&pcpu_devices[0], func, data,
356 pcpu_devices->lowcore->panic_stack -
357 PANIC_FRAME_OFFSET + PAGE_SIZE);
358}
359
360int smp_find_processor_id(u16 address)
361{
362 int cpu;
363
364 for_each_present_cpu(cpu)
365 if (pcpu_devices[cpu].address == address)
366 return cpu;
367 return -1;
368}
369
370bool arch_vcpu_is_preempted(int cpu)
371{
372 if (test_cpu_flag_of(CIF_ENABLED_WAIT, cpu))
373 return false;
374 if (pcpu_running(pcpu_devices + cpu))
375 return false;
376 return true;
377}
378EXPORT_SYMBOL(arch_vcpu_is_preempted);
379
380void smp_yield_cpu(int cpu)
381{
382 if (MACHINE_HAS_DIAG9C) {
383 diag_stat_inc_norecursion(DIAG_STAT_X09C);
384 asm volatile("diag %0,0,0x9c"
385 : : "d" (pcpu_devices[cpu].address));
386 } else if (MACHINE_HAS_DIAG44) {
387 diag_stat_inc_norecursion(DIAG_STAT_X044);
388 asm volatile("diag 0,0,0x44");
389 }
390}
391
392/*
393 * Send cpus emergency shutdown signal. This gives the cpus the
394 * opportunity to complete outstanding interrupts.
395 */
396static void smp_emergency_stop(cpumask_t *cpumask)
397{
398 u64 end;
399 int cpu;
400
401 end = get_tod_clock() + (1000000UL << 12);
402 for_each_cpu(cpu, cpumask) {
403 struct pcpu *pcpu = pcpu_devices + cpu;
404 set_bit(ec_stop_cpu, &pcpu->ec_mask);
405 while (__pcpu_sigp(pcpu->address, SIGP_EMERGENCY_SIGNAL,
406 0, NULL) == SIGP_CC_BUSY &&
407 get_tod_clock() < end)
408 cpu_relax();
409 }
410 while (get_tod_clock() < end) {
411 for_each_cpu(cpu, cpumask)
412 if (pcpu_stopped(pcpu_devices + cpu))
413 cpumask_clear_cpu(cpu, cpumask);
414 if (cpumask_empty(cpumask))
415 break;
416 cpu_relax();
417 }
418}
419
420/*
421 * Stop all cpus but the current one.
422 */
423void smp_send_stop(void)
424{
425 cpumask_t cpumask;
426 int cpu;
427
428 /* Disable all interrupts/machine checks */
429 __load_psw_mask(PSW_KERNEL_BITS | PSW_MASK_DAT);
430 trace_hardirqs_off();
431
432 debug_set_critical();
433 cpumask_copy(&cpumask, cpu_online_mask);
434 cpumask_clear_cpu(smp_processor_id(), &cpumask);
435
436 if (oops_in_progress)
437 smp_emergency_stop(&cpumask);
438
439 /* stop all processors */
440 for_each_cpu(cpu, &cpumask) {
441 struct pcpu *pcpu = pcpu_devices + cpu;
442 pcpu_sigp_retry(pcpu, SIGP_STOP, 0);
443 while (!pcpu_stopped(pcpu))
444 cpu_relax();
445 }
446}
447
448/*
449 * This is the main routine where commands issued by other
450 * cpus are handled.
451 */
452static void smp_handle_ext_call(void)
453{
454 unsigned long bits;
455
456 /* handle bit signal external calls */
457 bits = xchg(&pcpu_devices[smp_processor_id()].ec_mask, 0);
458 if (test_bit(ec_stop_cpu, &bits))
459 smp_stop_cpu();
460 if (test_bit(ec_schedule, &bits))
461 scheduler_ipi();
462 if (test_bit(ec_call_function_single, &bits))
463 generic_smp_call_function_single_interrupt();
464}
465
466static void do_ext_call_interrupt(struct ext_code ext_code,
467 unsigned int param32, unsigned long param64)
468{
469 inc_irq_stat(ext_code.code == 0x1202 ? IRQEXT_EXC : IRQEXT_EMS);
470 smp_handle_ext_call();
471}
472
473void arch_send_call_function_ipi_mask(const struct cpumask *mask)
474{
475 int cpu;
476
477 for_each_cpu(cpu, mask)
478 pcpu_ec_call(pcpu_devices + cpu, ec_call_function_single);
479}
480
481void arch_send_call_function_single_ipi(int cpu)
482{
483 pcpu_ec_call(pcpu_devices + cpu, ec_call_function_single);
484}
485
486/*
487 * this function sends a 'reschedule' IPI to another CPU.
488 * it goes straight through and wastes no time serializing
489 * anything. Worst case is that we lose a reschedule ...
490 */
491void smp_send_reschedule(int cpu)
492{
493 pcpu_ec_call(pcpu_devices + cpu, ec_schedule);
494}
495
496/*
497 * parameter area for the set/clear control bit callbacks
498 */
499struct ec_creg_mask_parms {
500 unsigned long orval;
501 unsigned long andval;
502 int cr;
503};
504
505/*
506 * callback for setting/clearing control bits
507 */
508static void smp_ctl_bit_callback(void *info)
509{
510 struct ec_creg_mask_parms *pp = info;
511 unsigned long cregs[16];
512
513 __ctl_store(cregs, 0, 15);
514 cregs[pp->cr] = (cregs[pp->cr] & pp->andval) | pp->orval;
515 __ctl_load(cregs, 0, 15);
516}
517
518/*
519 * Set a bit in a control register of all cpus
520 */
521void smp_ctl_set_bit(int cr, int bit)
522{
523 struct ec_creg_mask_parms parms = { 1UL << bit, -1UL, cr };
524
525 on_each_cpu(smp_ctl_bit_callback, &parms, 1);
526}
527EXPORT_SYMBOL(smp_ctl_set_bit);
528
529/*
530 * Clear a bit in a control register of all cpus
531 */
532void smp_ctl_clear_bit(int cr, int bit)
533{
534 struct ec_creg_mask_parms parms = { 0, ~(1UL << bit), cr };
535
536 on_each_cpu(smp_ctl_bit_callback, &parms, 1);
537}
538EXPORT_SYMBOL(smp_ctl_clear_bit);
539
540#ifdef CONFIG_CRASH_DUMP
541
542int smp_store_status(int cpu)
543{
544 struct pcpu *pcpu = pcpu_devices + cpu;
545 unsigned long pa;
546
547 pa = __pa(&pcpu->lowcore->floating_pt_save_area);
548 if (__pcpu_sigp_relax(pcpu->address, SIGP_STORE_STATUS_AT_ADDRESS,
549 pa) != SIGP_CC_ORDER_CODE_ACCEPTED)
550 return -EIO;
551 if (!MACHINE_HAS_VX)
552 return 0;
553 pa = __pa(pcpu->lowcore->vector_save_area_addr);
554 if (__pcpu_sigp_relax(pcpu->address, SIGP_STORE_ADDITIONAL_STATUS,
555 pa) != SIGP_CC_ORDER_CODE_ACCEPTED)
556 return -EIO;
557 return 0;
558}
559
560/*
561 * Collect CPU state of the previous, crashed system.
562 * There are four cases:
563 * 1) standard zfcp dump
564 * condition: OLDMEM_BASE == NULL && ipl_info.type == IPL_TYPE_FCP_DUMP
565 * The state for all CPUs except the boot CPU needs to be collected
566 * with sigp stop-and-store-status. The boot CPU state is located in
567 * the absolute lowcore of the memory stored in the HSA. The zcore code
568 * will copy the boot CPU state from the HSA.
569 * 2) stand-alone kdump for SCSI (zfcp dump with swapped memory)
570 * condition: OLDMEM_BASE != NULL && ipl_info.type == IPL_TYPE_FCP_DUMP
571 * The state for all CPUs except the boot CPU needs to be collected
572 * with sigp stop-and-store-status. The firmware or the boot-loader
573 * stored the registers of the boot CPU in the absolute lowcore in the
574 * memory of the old system.
575 * 3) kdump and the old kernel did not store the CPU state,
576 * or stand-alone kdump for DASD
577 * condition: OLDMEM_BASE != NULL && !is_kdump_kernel()
578 * The state for all CPUs except the boot CPU needs to be collected
579 * with sigp stop-and-store-status. The kexec code or the boot-loader
580 * stored the registers of the boot CPU in the memory of the old system.
581 * 4) kdump and the old kernel stored the CPU state
582 * condition: OLDMEM_BASE != NULL && is_kdump_kernel()
583 * This case does not exist for s390 anymore, setup_arch explicitly
584 * deactivates the elfcorehdr= kernel parameter
585 */
586static __init void smp_save_cpu_vxrs(struct save_area *sa, u16 addr,
587 bool is_boot_cpu, unsigned long page)
588{
589 __vector128 *vxrs = (__vector128 *) page;
590
591 if (is_boot_cpu)
592 vxrs = boot_cpu_vector_save_area;
593 else
594 __pcpu_sigp_relax(addr, SIGP_STORE_ADDITIONAL_STATUS, page);
595 save_area_add_vxrs(sa, vxrs);
596}
597
598static __init void smp_save_cpu_regs(struct save_area *sa, u16 addr,
599 bool is_boot_cpu, unsigned long page)
600{
601 void *regs = (void *) page;
602
603 if (is_boot_cpu)
604 copy_oldmem_kernel(regs, (void *) __LC_FPREGS_SAVE_AREA, 512);
605 else
606 __pcpu_sigp_relax(addr, SIGP_STORE_STATUS_AT_ADDRESS, page);
607 save_area_add_regs(sa, regs);
608}
609
610void __init smp_save_dump_cpus(void)
611{
612 int addr, boot_cpu_addr, max_cpu_addr;
613 struct save_area *sa;
614 unsigned long page;
615 bool is_boot_cpu;
616
617 if (!(OLDMEM_BASE || ipl_info.type == IPL_TYPE_FCP_DUMP))
618 /* No previous system present, normal boot. */
619 return;
620 /* Allocate a page as dumping area for the store status sigps */
621 page = memblock_alloc_base(PAGE_SIZE, PAGE_SIZE, 1UL << 31);
622 /* Set multi-threading state to the previous system. */
623 pcpu_set_smt(sclp.mtid_prev);
624 boot_cpu_addr = stap();
625 max_cpu_addr = SCLP_MAX_CORES << sclp.mtid_prev;
626 for (addr = 0; addr <= max_cpu_addr; addr++) {
627 if (__pcpu_sigp_relax(addr, SIGP_SENSE, 0) ==
628 SIGP_CC_NOT_OPERATIONAL)
629 continue;
630 is_boot_cpu = (addr == boot_cpu_addr);
631 /* Allocate save area */
632 sa = save_area_alloc(is_boot_cpu);
633 if (!sa)
634 panic("could not allocate memory for save area\n");
635 if (MACHINE_HAS_VX)
636 /* Get the vector registers */
637 smp_save_cpu_vxrs(sa, addr, is_boot_cpu, page);
638 /*
639 * For a zfcp dump OLDMEM_BASE == NULL and the registers
640 * of the boot CPU are stored in the HSA. To retrieve
641 * these registers an SCLP request is required which is
642 * done by drivers/s390/char/zcore.c:init_cpu_info()
643 */
644 if (!is_boot_cpu || OLDMEM_BASE)
645 /* Get the CPU registers */
646 smp_save_cpu_regs(sa, addr, is_boot_cpu, page);
647 }
648 memblock_free(page, PAGE_SIZE);
649 diag308_reset();
650 pcpu_set_smt(0);
651}
652#endif /* CONFIG_CRASH_DUMP */
653
654void smp_cpu_set_polarization(int cpu, int val)
655{
656 pcpu_devices[cpu].polarization = val;
657}
658
659int smp_cpu_get_polarization(int cpu)
660{
661 return pcpu_devices[cpu].polarization;
662}
663
664static void __ref smp_get_core_info(struct sclp_core_info *info, int early)
665{
666 static int use_sigp_detection;
667 int address;
668
669 if (use_sigp_detection || sclp_get_core_info(info, early)) {
670 use_sigp_detection = 1;
671 for (address = 0;
672 address < (SCLP_MAX_CORES << smp_cpu_mt_shift);
673 address += (1U << smp_cpu_mt_shift)) {
674 if (__pcpu_sigp_relax(address, SIGP_SENSE, 0) ==
675 SIGP_CC_NOT_OPERATIONAL)
676 continue;
677 info->core[info->configured].core_id =
678 address >> smp_cpu_mt_shift;
679 info->configured++;
680 }
681 info->combined = info->configured;
682 }
683}
684
685static int smp_add_present_cpu(int cpu);
686
687static int __smp_rescan_cpus(struct sclp_core_info *info, int sysfs_add)
688{
689 struct pcpu *pcpu;
690 cpumask_t avail;
691 int cpu, nr, i, j;
692 u16 address;
693
694 nr = 0;
695 cpumask_xor(&avail, cpu_possible_mask, cpu_present_mask);
696 cpu = cpumask_first(&avail);
697 for (i = 0; (i < info->combined) && (cpu < nr_cpu_ids); i++) {
698 if (sclp.has_core_type && info->core[i].type != boot_core_type)
699 continue;
700 address = info->core[i].core_id << smp_cpu_mt_shift;
701 for (j = 0; j <= smp_cpu_mtid; j++) {
702 if (pcpu_find_address(cpu_present_mask, address + j))
703 continue;
704 pcpu = pcpu_devices + cpu;
705 pcpu->address = address + j;
706 pcpu->state =
707 (cpu >= info->configured*(smp_cpu_mtid + 1)) ?
708 CPU_STATE_STANDBY : CPU_STATE_CONFIGURED;
709 smp_cpu_set_polarization(cpu, POLARIZATION_UNKNOWN);
710 set_cpu_present(cpu, true);
711 if (sysfs_add && smp_add_present_cpu(cpu) != 0)
712 set_cpu_present(cpu, false);
713 else
714 nr++;
715 cpu = cpumask_next(cpu, &avail);
716 if (cpu >= nr_cpu_ids)
717 break;
718 }
719 }
720 return nr;
721}
722
723void __init smp_detect_cpus(void)
724{
725 unsigned int cpu, mtid, c_cpus, s_cpus;
726 struct sclp_core_info *info;
727 u16 address;
728
729 /* Get CPU information */
730 info = memblock_virt_alloc(sizeof(*info), 8);
731 smp_get_core_info(info, 1);
732 /* Find boot CPU type */
733 if (sclp.has_core_type) {
734 address = stap();
735 for (cpu = 0; cpu < info->combined; cpu++)
736 if (info->core[cpu].core_id == address) {
737 /* The boot cpu dictates the cpu type. */
738 boot_core_type = info->core[cpu].type;
739 break;
740 }
741 if (cpu >= info->combined)
742 panic("Could not find boot CPU type");
743 }
744
745 /* Set multi-threading state for the current system */
746 mtid = boot_core_type ? sclp.mtid : sclp.mtid_cp;
747 mtid = (mtid < smp_max_threads) ? mtid : smp_max_threads - 1;
748 pcpu_set_smt(mtid);
749
750 /* Print number of CPUs */
751 c_cpus = s_cpus = 0;
752 for (cpu = 0; cpu < info->combined; cpu++) {
753 if (sclp.has_core_type &&
754 info->core[cpu].type != boot_core_type)
755 continue;
756 if (cpu < info->configured)
757 c_cpus += smp_cpu_mtid + 1;
758 else
759 s_cpus += smp_cpu_mtid + 1;
760 }
761 pr_info("%d configured CPUs, %d standby CPUs\n", c_cpus, s_cpus);
762
763 /* Add CPUs present at boot */
764 get_online_cpus();
765 __smp_rescan_cpus(info, 0);
766 put_online_cpus();
767 memblock_free_early((unsigned long)info, sizeof(*info));
768}
769
770/*
771 * Activate a secondary processor.
772 */
773static void smp_start_secondary(void *cpuvoid)
774{
775 S390_lowcore.last_update_clock = get_tod_clock();
776 S390_lowcore.restart_stack = (unsigned long) restart_stack;
777 S390_lowcore.restart_fn = (unsigned long) do_restart;
778 S390_lowcore.restart_data = 0;
779 S390_lowcore.restart_source = -1UL;
780 restore_access_regs(S390_lowcore.access_regs_save_area);
781 __ctl_load(S390_lowcore.cregs_save_area, 0, 15);
782 __load_psw_mask(PSW_KERNEL_BITS | PSW_MASK_DAT);
783 cpu_init();
784 preempt_disable();
785 init_cpu_timer();
786 vtime_init();
787 pfault_init();
788 notify_cpu_starting(smp_processor_id());
789 set_cpu_online(smp_processor_id(), true);
790 inc_irq_stat(CPU_RST);
791 local_irq_enable();
792 cpu_startup_entry(CPUHP_AP_ONLINE_IDLE);
793}
794
795/* Upping and downing of CPUs */
796int __cpu_up(unsigned int cpu, struct task_struct *tidle)
797{
798 struct pcpu *pcpu;
799 int base, i, rc;
800
801 pcpu = pcpu_devices + cpu;
802 if (pcpu->state != CPU_STATE_CONFIGURED)
803 return -EIO;
804 base = smp_get_base_cpu(cpu);
805 for (i = 0; i <= smp_cpu_mtid; i++) {
806 if (base + i < nr_cpu_ids)
807 if (cpu_online(base + i))
808 break;
809 }
810 /*
811 * If this is the first CPU of the core to get online
812 * do an initial CPU reset.
813 */
814 if (i > smp_cpu_mtid &&
815 pcpu_sigp_retry(pcpu_devices + base, SIGP_INITIAL_CPU_RESET, 0) !=
816 SIGP_CC_ORDER_CODE_ACCEPTED)
817 return -EIO;
818
819 rc = pcpu_alloc_lowcore(pcpu, cpu);
820 if (rc)
821 return rc;
822 pcpu_prepare_secondary(pcpu, cpu);
823 pcpu_attach_task(pcpu, tidle);
824 pcpu_start_fn(pcpu, smp_start_secondary, NULL);
825 /* Wait until cpu puts itself in the online & active maps */
826 while (!cpu_online(cpu))
827 cpu_relax();
828 return 0;
829}
830
831static unsigned int setup_possible_cpus __initdata;
832
833static int __init _setup_possible_cpus(char *s)
834{
835 get_option(&s, &setup_possible_cpus);
836 return 0;
837}
838early_param("possible_cpus", _setup_possible_cpus);
839
840#ifdef CONFIG_HOTPLUG_CPU
841
842int __cpu_disable(void)
843{
844 unsigned long cregs[16];
845
846 /* Handle possible pending IPIs */
847 smp_handle_ext_call();
848 set_cpu_online(smp_processor_id(), false);
849 /* Disable pseudo page faults on this cpu. */
850 pfault_fini();
851 /* Disable interrupt sources via control register. */
852 __ctl_store(cregs, 0, 15);
853 cregs[0] &= ~0x0000ee70UL; /* disable all external interrupts */
854 cregs[6] &= ~0xff000000UL; /* disable all I/O interrupts */
855 cregs[14] &= ~0x1f000000UL; /* disable most machine checks */
856 __ctl_load(cregs, 0, 15);
857 clear_cpu_flag(CIF_NOHZ_DELAY);
858 return 0;
859}
860
861void __cpu_die(unsigned int cpu)
862{
863 struct pcpu *pcpu;
864
865 /* Wait until target cpu is down */
866 pcpu = pcpu_devices + cpu;
867 while (!pcpu_stopped(pcpu))
868 cpu_relax();
869 pcpu_free_lowcore(pcpu);
870 cpumask_clear_cpu(cpu, mm_cpumask(&init_mm));
871 cpumask_clear_cpu(cpu, &init_mm.context.cpu_attach_mask);
872}
873
874void __noreturn cpu_die(void)
875{
876 idle_task_exit();
877 pcpu_sigp_retry(pcpu_devices + smp_processor_id(), SIGP_STOP, 0);
878 for (;;) ;
879}
880
881#endif /* CONFIG_HOTPLUG_CPU */
882
883void __init smp_fill_possible_mask(void)
884{
885 unsigned int possible, sclp_max, cpu;
886
887 sclp_max = max(sclp.mtid, sclp.mtid_cp) + 1;
888 sclp_max = min(smp_max_threads, sclp_max);
889 sclp_max = (sclp.max_cores * sclp_max) ?: nr_cpu_ids;
890 possible = setup_possible_cpus ?: nr_cpu_ids;
891 possible = min(possible, sclp_max);
892 for (cpu = 0; cpu < possible && cpu < nr_cpu_ids; cpu++)
893 set_cpu_possible(cpu, true);
894}
895
896void __init smp_prepare_cpus(unsigned int max_cpus)
897{
898 /* request the 0x1201 emergency signal external interrupt */
899 if (register_external_irq(EXT_IRQ_EMERGENCY_SIG, do_ext_call_interrupt))
900 panic("Couldn't request external interrupt 0x1201");
901 /* request the 0x1202 external call external interrupt */
902 if (register_external_irq(EXT_IRQ_EXTERNAL_CALL, do_ext_call_interrupt))
903 panic("Couldn't request external interrupt 0x1202");
904}
905
906void __init smp_prepare_boot_cpu(void)
907{
908 struct pcpu *pcpu = pcpu_devices;
909
910 pcpu->state = CPU_STATE_CONFIGURED;
911 pcpu->address = stap();
912 pcpu->lowcore = (struct lowcore *)(unsigned long) store_prefix();
913 S390_lowcore.percpu_offset = __per_cpu_offset[0];
914 smp_cpu_set_polarization(0, POLARIZATION_UNKNOWN);
915 set_cpu_present(0, true);
916 set_cpu_online(0, true);
917}
918
919void __init smp_cpus_done(unsigned int max_cpus)
920{
921}
922
923void __init smp_setup_processor_id(void)
924{
925 S390_lowcore.cpu_nr = 0;
926 S390_lowcore.spinlock_lockval = arch_spin_lockval(0);
927}
928
929/*
930 * the frequency of the profiling timer can be changed
931 * by writing a multiplier value into /proc/profile.
932 *
933 * usually you want to run this on all CPUs ;)
934 */
935int setup_profiling_timer(unsigned int multiplier)
936{
937 return 0;
938}
939
940#ifdef CONFIG_HOTPLUG_CPU
941static ssize_t cpu_configure_show(struct device *dev,
942 struct device_attribute *attr, char *buf)
943{
944 ssize_t count;
945
946 mutex_lock(&smp_cpu_state_mutex);
947 count = sprintf(buf, "%d\n", pcpu_devices[dev->id].state);
948 mutex_unlock(&smp_cpu_state_mutex);
949 return count;
950}
951
952static ssize_t cpu_configure_store(struct device *dev,
953 struct device_attribute *attr,
954 const char *buf, size_t count)
955{
956 struct pcpu *pcpu;
957 int cpu, val, rc, i;
958 char delim;
959
960 if (sscanf(buf, "%d %c", &val, &delim) != 1)
961 return -EINVAL;
962 if (val != 0 && val != 1)
963 return -EINVAL;
964 get_online_cpus();
965 mutex_lock(&smp_cpu_state_mutex);
966 rc = -EBUSY;
967 /* disallow configuration changes of online cpus and cpu 0 */
968 cpu = dev->id;
969 cpu = smp_get_base_cpu(cpu);
970 if (cpu == 0)
971 goto out;
972 for (i = 0; i <= smp_cpu_mtid; i++)
973 if (cpu_online(cpu + i))
974 goto out;
975 pcpu = pcpu_devices + cpu;
976 rc = 0;
977 switch (val) {
978 case 0:
979 if (pcpu->state != CPU_STATE_CONFIGURED)
980 break;
981 rc = sclp_core_deconfigure(pcpu->address >> smp_cpu_mt_shift);
982 if (rc)
983 break;
984 for (i = 0; i <= smp_cpu_mtid; i++) {
985 if (cpu + i >= nr_cpu_ids || !cpu_present(cpu + i))
986 continue;
987 pcpu[i].state = CPU_STATE_STANDBY;
988 smp_cpu_set_polarization(cpu + i,
989 POLARIZATION_UNKNOWN);
990 }
991 topology_expect_change();
992 break;
993 case 1:
994 if (pcpu->state != CPU_STATE_STANDBY)
995 break;
996 rc = sclp_core_configure(pcpu->address >> smp_cpu_mt_shift);
997 if (rc)
998 break;
999 for (i = 0; i <= smp_cpu_mtid; i++) {
1000 if (cpu + i >= nr_cpu_ids || !cpu_present(cpu + i))
1001 continue;
1002 pcpu[i].state = CPU_STATE_CONFIGURED;
1003 smp_cpu_set_polarization(cpu + i,
1004 POLARIZATION_UNKNOWN);
1005 }
1006 topology_expect_change();
1007 break;
1008 default:
1009 break;
1010 }
1011out:
1012 mutex_unlock(&smp_cpu_state_mutex);
1013 put_online_cpus();
1014 return rc ? rc : count;
1015}
1016static DEVICE_ATTR(configure, 0644, cpu_configure_show, cpu_configure_store);
1017#endif /* CONFIG_HOTPLUG_CPU */
1018
1019static ssize_t show_cpu_address(struct device *dev,
1020 struct device_attribute *attr, char *buf)
1021{
1022 return sprintf(buf, "%d\n", pcpu_devices[dev->id].address);
1023}
1024static DEVICE_ATTR(address, 0444, show_cpu_address, NULL);
1025
1026static struct attribute *cpu_common_attrs[] = {
1027#ifdef CONFIG_HOTPLUG_CPU
1028 &dev_attr_configure.attr,
1029#endif
1030 &dev_attr_address.attr,
1031 NULL,
1032};
1033
1034static struct attribute_group cpu_common_attr_group = {
1035 .attrs = cpu_common_attrs,
1036};
1037
1038static struct attribute *cpu_online_attrs[] = {
1039 &dev_attr_idle_count.attr,
1040 &dev_attr_idle_time_us.attr,
1041 NULL,
1042};
1043
1044static struct attribute_group cpu_online_attr_group = {
1045 .attrs = cpu_online_attrs,
1046};
1047
1048static int smp_cpu_online(unsigned int cpu)
1049{
1050 struct device *s = &per_cpu(cpu_device, cpu)->dev;
1051
1052 return sysfs_create_group(&s->kobj, &cpu_online_attr_group);
1053}
1054static int smp_cpu_pre_down(unsigned int cpu)
1055{
1056 struct device *s = &per_cpu(cpu_device, cpu)->dev;
1057
1058 sysfs_remove_group(&s->kobj, &cpu_online_attr_group);
1059 return 0;
1060}
1061
1062static int smp_add_present_cpu(int cpu)
1063{
1064 struct device *s;
1065 struct cpu *c;
1066 int rc;
1067
1068 c = kzalloc(sizeof(*c), GFP_KERNEL);
1069 if (!c)
1070 return -ENOMEM;
1071 per_cpu(cpu_device, cpu) = c;
1072 s = &c->dev;
1073 c->hotpluggable = 1;
1074 rc = register_cpu(c, cpu);
1075 if (rc)
1076 goto out;
1077 rc = sysfs_create_group(&s->kobj, &cpu_common_attr_group);
1078 if (rc)
1079 goto out_cpu;
1080 rc = topology_cpu_init(c);
1081 if (rc)
1082 goto out_topology;
1083 return 0;
1084
1085out_topology:
1086 sysfs_remove_group(&s->kobj, &cpu_common_attr_group);
1087out_cpu:
1088#ifdef CONFIG_HOTPLUG_CPU
1089 unregister_cpu(c);
1090#endif
1091out:
1092 return rc;
1093}
1094
1095#ifdef CONFIG_HOTPLUG_CPU
1096
1097int __ref smp_rescan_cpus(void)
1098{
1099 struct sclp_core_info *info;
1100 int nr;
1101
1102 info = kzalloc(sizeof(*info), GFP_KERNEL);
1103 if (!info)
1104 return -ENOMEM;
1105 smp_get_core_info(info, 0);
1106 get_online_cpus();
1107 mutex_lock(&smp_cpu_state_mutex);
1108 nr = __smp_rescan_cpus(info, 1);
1109 mutex_unlock(&smp_cpu_state_mutex);
1110 put_online_cpus();
1111 kfree(info);
1112 if (nr)
1113 topology_schedule_update();
1114 return 0;
1115}
1116
1117static ssize_t __ref rescan_store(struct device *dev,
1118 struct device_attribute *attr,
1119 const char *buf,
1120 size_t count)
1121{
1122 int rc;
1123
1124 rc = smp_rescan_cpus();
1125 return rc ? rc : count;
1126}
1127static DEVICE_ATTR(rescan, 0200, NULL, rescan_store);
1128#endif /* CONFIG_HOTPLUG_CPU */
1129
1130static int __init s390_smp_init(void)
1131{
1132 int cpu, rc = 0;
1133
1134#ifdef CONFIG_HOTPLUG_CPU
1135 rc = device_create_file(cpu_subsys.dev_root, &dev_attr_rescan);
1136 if (rc)
1137 return rc;
1138#endif
1139 for_each_present_cpu(cpu) {
1140 rc = smp_add_present_cpu(cpu);
1141 if (rc)
1142 goto out;
1143 }
1144
1145 rc = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "s390/smp:online",
1146 smp_cpu_online, smp_cpu_pre_down);
1147out:
1148 return rc;
1149}
1150subsys_initcall(s390_smp_init);
1/*
2 * SMP related functions
3 *
4 * Copyright IBM Corp. 1999, 2012
5 * Author(s): Denis Joseph Barrow,
6 * Martin Schwidefsky <schwidefsky@de.ibm.com>,
7 * Heiko Carstens <heiko.carstens@de.ibm.com>,
8 *
9 * based on other smp stuff by
10 * (c) 1995 Alan Cox, CymruNET Ltd <alan@cymru.net>
11 * (c) 1998 Ingo Molnar
12 *
13 * The code outside of smp.c uses logical cpu numbers, only smp.c does
14 * the translation of logical to physical cpu ids. All new code that
15 * operates on physical cpu numbers needs to go into smp.c.
16 */
17
18#define KMSG_COMPONENT "cpu"
19#define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
20
21#include <linux/workqueue.h>
22#include <linux/module.h>
23#include <linux/init.h>
24#include <linux/mm.h>
25#include <linux/err.h>
26#include <linux/spinlock.h>
27#include <linux/kernel_stat.h>
28#include <linux/delay.h>
29#include <linux/interrupt.h>
30#include <linux/irqflags.h>
31#include <linux/cpu.h>
32#include <linux/slab.h>
33#include <linux/crash_dump.h>
34#include <asm/asm-offsets.h>
35#include <asm/switch_to.h>
36#include <asm/facility.h>
37#include <asm/ipl.h>
38#include <asm/setup.h>
39#include <asm/irq.h>
40#include <asm/tlbflush.h>
41#include <asm/vtimer.h>
42#include <asm/lowcore.h>
43#include <asm/sclp.h>
44#include <asm/vdso.h>
45#include <asm/debug.h>
46#include <asm/os_info.h>
47#include <asm/sigp.h>
48#include "entry.h"
49
50enum {
51 ec_schedule = 0,
52 ec_call_function_single,
53 ec_stop_cpu,
54};
55
56enum {
57 CPU_STATE_STANDBY,
58 CPU_STATE_CONFIGURED,
59};
60
61struct pcpu {
62 struct cpu *cpu;
63 struct _lowcore *lowcore; /* lowcore page(s) for the cpu */
64 unsigned long async_stack; /* async stack for the cpu */
65 unsigned long panic_stack; /* panic stack for the cpu */
66 unsigned long ec_mask; /* bit mask for ec_xxx functions */
67 int state; /* physical cpu state */
68 int polarization; /* physical polarization */
69 u16 address; /* physical cpu address */
70};
71
72static u8 boot_cpu_type;
73static u16 boot_cpu_address;
74static struct pcpu pcpu_devices[NR_CPUS];
75
76/*
77 * The smp_cpu_state_mutex must be held when changing the state or polarization
78 * member of a pcpu data structure within the pcpu_devices arreay.
79 */
80DEFINE_MUTEX(smp_cpu_state_mutex);
81
82/*
83 * Signal processor helper functions.
84 */
85static inline int __pcpu_sigp_relax(u16 addr, u8 order, u32 parm, u32 *status)
86{
87 int cc;
88
89 while (1) {
90 cc = __pcpu_sigp(addr, order, parm, NULL);
91 if (cc != SIGP_CC_BUSY)
92 return cc;
93 cpu_relax();
94 }
95}
96
97static int pcpu_sigp_retry(struct pcpu *pcpu, u8 order, u32 parm)
98{
99 int cc, retry;
100
101 for (retry = 0; ; retry++) {
102 cc = __pcpu_sigp(pcpu->address, order, parm, NULL);
103 if (cc != SIGP_CC_BUSY)
104 break;
105 if (retry >= 3)
106 udelay(10);
107 }
108 return cc;
109}
110
111static inline int pcpu_stopped(struct pcpu *pcpu)
112{
113 u32 uninitialized_var(status);
114
115 if (__pcpu_sigp(pcpu->address, SIGP_SENSE,
116 0, &status) != SIGP_CC_STATUS_STORED)
117 return 0;
118 return !!(status & (SIGP_STATUS_CHECK_STOP|SIGP_STATUS_STOPPED));
119}
120
121static inline int pcpu_running(struct pcpu *pcpu)
122{
123 if (__pcpu_sigp(pcpu->address, SIGP_SENSE_RUNNING,
124 0, NULL) != SIGP_CC_STATUS_STORED)
125 return 1;
126 /* Status stored condition code is equivalent to cpu not running. */
127 return 0;
128}
129
130/*
131 * Find struct pcpu by cpu address.
132 */
133static struct pcpu *pcpu_find_address(const struct cpumask *mask, int address)
134{
135 int cpu;
136
137 for_each_cpu(cpu, mask)
138 if (pcpu_devices[cpu].address == address)
139 return pcpu_devices + cpu;
140 return NULL;
141}
142
143static void pcpu_ec_call(struct pcpu *pcpu, int ec_bit)
144{
145 int order;
146
147 if (test_and_set_bit(ec_bit, &pcpu->ec_mask))
148 return;
149 order = pcpu_running(pcpu) ? SIGP_EXTERNAL_CALL : SIGP_EMERGENCY_SIGNAL;
150 pcpu_sigp_retry(pcpu, order, 0);
151}
152
153static int pcpu_alloc_lowcore(struct pcpu *pcpu, int cpu)
154{
155 struct _lowcore *lc;
156
157 if (pcpu != &pcpu_devices[0]) {
158 pcpu->lowcore = (struct _lowcore *)
159 __get_free_pages(GFP_KERNEL | GFP_DMA, LC_ORDER);
160 pcpu->async_stack = __get_free_pages(GFP_KERNEL, ASYNC_ORDER);
161 pcpu->panic_stack = __get_free_page(GFP_KERNEL);
162 if (!pcpu->lowcore || !pcpu->panic_stack || !pcpu->async_stack)
163 goto out;
164 }
165 lc = pcpu->lowcore;
166 memcpy(lc, &S390_lowcore, 512);
167 memset((char *) lc + 512, 0, sizeof(*lc) - 512);
168 lc->async_stack = pcpu->async_stack + ASYNC_SIZE
169 - STACK_FRAME_OVERHEAD - sizeof(struct pt_regs);
170 lc->panic_stack = pcpu->panic_stack + PAGE_SIZE
171 - STACK_FRAME_OVERHEAD - sizeof(struct pt_regs);
172 lc->cpu_nr = cpu;
173#ifndef CONFIG_64BIT
174 if (MACHINE_HAS_IEEE) {
175 lc->extended_save_area_addr = get_zeroed_page(GFP_KERNEL);
176 if (!lc->extended_save_area_addr)
177 goto out;
178 }
179#else
180 if (vdso_alloc_per_cpu(lc))
181 goto out;
182#endif
183 lowcore_ptr[cpu] = lc;
184 pcpu_sigp_retry(pcpu, SIGP_SET_PREFIX, (u32)(unsigned long) lc);
185 return 0;
186out:
187 if (pcpu != &pcpu_devices[0]) {
188 free_page(pcpu->panic_stack);
189 free_pages(pcpu->async_stack, ASYNC_ORDER);
190 free_pages((unsigned long) pcpu->lowcore, LC_ORDER);
191 }
192 return -ENOMEM;
193}
194
195#ifdef CONFIG_HOTPLUG_CPU
196
197static void pcpu_free_lowcore(struct pcpu *pcpu)
198{
199 pcpu_sigp_retry(pcpu, SIGP_SET_PREFIX, 0);
200 lowcore_ptr[pcpu - pcpu_devices] = NULL;
201#ifndef CONFIG_64BIT
202 if (MACHINE_HAS_IEEE) {
203 struct _lowcore *lc = pcpu->lowcore;
204
205 free_page((unsigned long) lc->extended_save_area_addr);
206 lc->extended_save_area_addr = 0;
207 }
208#else
209 vdso_free_per_cpu(pcpu->lowcore);
210#endif
211 if (pcpu != &pcpu_devices[0]) {
212 free_page(pcpu->panic_stack);
213 free_pages(pcpu->async_stack, ASYNC_ORDER);
214 free_pages((unsigned long) pcpu->lowcore, LC_ORDER);
215 }
216}
217
218#endif /* CONFIG_HOTPLUG_CPU */
219
220static void pcpu_prepare_secondary(struct pcpu *pcpu, int cpu)
221{
222 struct _lowcore *lc = pcpu->lowcore;
223
224 if (MACHINE_HAS_TLB_LC)
225 cpumask_set_cpu(cpu, &init_mm.context.cpu_attach_mask);
226 cpumask_set_cpu(cpu, mm_cpumask(&init_mm));
227 atomic_inc(&init_mm.context.attach_count);
228 lc->cpu_nr = cpu;
229 lc->percpu_offset = __per_cpu_offset[cpu];
230 lc->kernel_asce = S390_lowcore.kernel_asce;
231 lc->machine_flags = S390_lowcore.machine_flags;
232 lc->ftrace_func = S390_lowcore.ftrace_func;
233 lc->user_timer = lc->system_timer = lc->steal_timer = 0;
234 __ctl_store(lc->cregs_save_area, 0, 15);
235 save_access_regs((unsigned int *) lc->access_regs_save_area);
236 memcpy(lc->stfle_fac_list, S390_lowcore.stfle_fac_list,
237 MAX_FACILITY_BIT/8);
238}
239
240static void pcpu_attach_task(struct pcpu *pcpu, struct task_struct *tsk)
241{
242 struct _lowcore *lc = pcpu->lowcore;
243 struct thread_info *ti = task_thread_info(tsk);
244
245 lc->kernel_stack = (unsigned long) task_stack_page(tsk)
246 + THREAD_SIZE - STACK_FRAME_OVERHEAD - sizeof(struct pt_regs);
247 lc->thread_info = (unsigned long) task_thread_info(tsk);
248 lc->current_task = (unsigned long) tsk;
249 lc->user_timer = ti->user_timer;
250 lc->system_timer = ti->system_timer;
251 lc->steal_timer = 0;
252}
253
254static void pcpu_start_fn(struct pcpu *pcpu, void (*func)(void *), void *data)
255{
256 struct _lowcore *lc = pcpu->lowcore;
257
258 lc->restart_stack = lc->kernel_stack;
259 lc->restart_fn = (unsigned long) func;
260 lc->restart_data = (unsigned long) data;
261 lc->restart_source = -1UL;
262 pcpu_sigp_retry(pcpu, SIGP_RESTART, 0);
263}
264
265/*
266 * Call function via PSW restart on pcpu and stop the current cpu.
267 */
268static void pcpu_delegate(struct pcpu *pcpu, void (*func)(void *),
269 void *data, unsigned long stack)
270{
271 struct _lowcore *lc = lowcore_ptr[pcpu - pcpu_devices];
272 unsigned long source_cpu = stap();
273
274 __load_psw_mask(PSW_KERNEL_BITS);
275 if (pcpu->address == source_cpu)
276 func(data); /* should not return */
277 /* Stop target cpu (if func returns this stops the current cpu). */
278 pcpu_sigp_retry(pcpu, SIGP_STOP, 0);
279 /* Restart func on the target cpu and stop the current cpu. */
280 mem_assign_absolute(lc->restart_stack, stack);
281 mem_assign_absolute(lc->restart_fn, (unsigned long) func);
282 mem_assign_absolute(lc->restart_data, (unsigned long) data);
283 mem_assign_absolute(lc->restart_source, source_cpu);
284 asm volatile(
285 "0: sigp 0,%0,%2 # sigp restart to target cpu\n"
286 " brc 2,0b # busy, try again\n"
287 "1: sigp 0,%1,%3 # sigp stop to current cpu\n"
288 " brc 2,1b # busy, try again\n"
289 : : "d" (pcpu->address), "d" (source_cpu),
290 "K" (SIGP_RESTART), "K" (SIGP_STOP)
291 : "0", "1", "cc");
292 for (;;) ;
293}
294
295/*
296 * Call function on an online CPU.
297 */
298void smp_call_online_cpu(void (*func)(void *), void *data)
299{
300 struct pcpu *pcpu;
301
302 /* Use the current cpu if it is online. */
303 pcpu = pcpu_find_address(cpu_online_mask, stap());
304 if (!pcpu)
305 /* Use the first online cpu. */
306 pcpu = pcpu_devices + cpumask_first(cpu_online_mask);
307 pcpu_delegate(pcpu, func, data, (unsigned long) restart_stack);
308}
309
310/*
311 * Call function on the ipl CPU.
312 */
313void smp_call_ipl_cpu(void (*func)(void *), void *data)
314{
315 pcpu_delegate(&pcpu_devices[0], func, data,
316 pcpu_devices->panic_stack + PAGE_SIZE);
317}
318
319int smp_find_processor_id(u16 address)
320{
321 int cpu;
322
323 for_each_present_cpu(cpu)
324 if (pcpu_devices[cpu].address == address)
325 return cpu;
326 return -1;
327}
328
329int smp_vcpu_scheduled(int cpu)
330{
331 return pcpu_running(pcpu_devices + cpu);
332}
333
334void smp_yield(void)
335{
336 if (MACHINE_HAS_DIAG44)
337 asm volatile("diag 0,0,0x44");
338}
339
340void smp_yield_cpu(int cpu)
341{
342 if (MACHINE_HAS_DIAG9C)
343 asm volatile("diag %0,0,0x9c"
344 : : "d" (pcpu_devices[cpu].address));
345 else if (MACHINE_HAS_DIAG44)
346 asm volatile("diag 0,0,0x44");
347}
348
349/*
350 * Send cpus emergency shutdown signal. This gives the cpus the
351 * opportunity to complete outstanding interrupts.
352 */
353static void smp_emergency_stop(cpumask_t *cpumask)
354{
355 u64 end;
356 int cpu;
357
358 end = get_tod_clock() + (1000000UL << 12);
359 for_each_cpu(cpu, cpumask) {
360 struct pcpu *pcpu = pcpu_devices + cpu;
361 set_bit(ec_stop_cpu, &pcpu->ec_mask);
362 while (__pcpu_sigp(pcpu->address, SIGP_EMERGENCY_SIGNAL,
363 0, NULL) == SIGP_CC_BUSY &&
364 get_tod_clock() < end)
365 cpu_relax();
366 }
367 while (get_tod_clock() < end) {
368 for_each_cpu(cpu, cpumask)
369 if (pcpu_stopped(pcpu_devices + cpu))
370 cpumask_clear_cpu(cpu, cpumask);
371 if (cpumask_empty(cpumask))
372 break;
373 cpu_relax();
374 }
375}
376
377/*
378 * Stop all cpus but the current one.
379 */
380void smp_send_stop(void)
381{
382 cpumask_t cpumask;
383 int cpu;
384
385 /* Disable all interrupts/machine checks */
386 __load_psw_mask(PSW_KERNEL_BITS | PSW_MASK_DAT);
387 trace_hardirqs_off();
388
389 debug_set_critical();
390 cpumask_copy(&cpumask, cpu_online_mask);
391 cpumask_clear_cpu(smp_processor_id(), &cpumask);
392
393 if (oops_in_progress)
394 smp_emergency_stop(&cpumask);
395
396 /* stop all processors */
397 for_each_cpu(cpu, &cpumask) {
398 struct pcpu *pcpu = pcpu_devices + cpu;
399 pcpu_sigp_retry(pcpu, SIGP_STOP, 0);
400 while (!pcpu_stopped(pcpu))
401 cpu_relax();
402 }
403}
404
405/*
406 * Stop the current cpu.
407 */
408void smp_stop_cpu(void)
409{
410 pcpu_sigp_retry(pcpu_devices + smp_processor_id(), SIGP_STOP, 0);
411 for (;;) ;
412}
413
414/*
415 * This is the main routine where commands issued by other
416 * cpus are handled.
417 */
418static void smp_handle_ext_call(void)
419{
420 unsigned long bits;
421
422 /* handle bit signal external calls */
423 bits = xchg(&pcpu_devices[smp_processor_id()].ec_mask, 0);
424 if (test_bit(ec_stop_cpu, &bits))
425 smp_stop_cpu();
426 if (test_bit(ec_schedule, &bits))
427 scheduler_ipi();
428 if (test_bit(ec_call_function_single, &bits))
429 generic_smp_call_function_single_interrupt();
430}
431
432static void do_ext_call_interrupt(struct ext_code ext_code,
433 unsigned int param32, unsigned long param64)
434{
435 inc_irq_stat(ext_code.code == 0x1202 ? IRQEXT_EXC : IRQEXT_EMS);
436 smp_handle_ext_call();
437}
438
439void arch_send_call_function_ipi_mask(const struct cpumask *mask)
440{
441 int cpu;
442
443 for_each_cpu(cpu, mask)
444 pcpu_ec_call(pcpu_devices + cpu, ec_call_function_single);
445}
446
447void arch_send_call_function_single_ipi(int cpu)
448{
449 pcpu_ec_call(pcpu_devices + cpu, ec_call_function_single);
450}
451
452#ifndef CONFIG_64BIT
453/*
454 * this function sends a 'purge tlb' signal to another CPU.
455 */
456static void smp_ptlb_callback(void *info)
457{
458 __tlb_flush_local();
459}
460
461void smp_ptlb_all(void)
462{
463 on_each_cpu(smp_ptlb_callback, NULL, 1);
464}
465EXPORT_SYMBOL(smp_ptlb_all);
466#endif /* ! CONFIG_64BIT */
467
468/*
469 * this function sends a 'reschedule' IPI to another CPU.
470 * it goes straight through and wastes no time serializing
471 * anything. Worst case is that we lose a reschedule ...
472 */
473void smp_send_reschedule(int cpu)
474{
475 pcpu_ec_call(pcpu_devices + cpu, ec_schedule);
476}
477
478/*
479 * parameter area for the set/clear control bit callbacks
480 */
481struct ec_creg_mask_parms {
482 unsigned long orval;
483 unsigned long andval;
484 int cr;
485};
486
487/*
488 * callback for setting/clearing control bits
489 */
490static void smp_ctl_bit_callback(void *info)
491{
492 struct ec_creg_mask_parms *pp = info;
493 unsigned long cregs[16];
494
495 __ctl_store(cregs, 0, 15);
496 cregs[pp->cr] = (cregs[pp->cr] & pp->andval) | pp->orval;
497 __ctl_load(cregs, 0, 15);
498}
499
500/*
501 * Set a bit in a control register of all cpus
502 */
503void smp_ctl_set_bit(int cr, int bit)
504{
505 struct ec_creg_mask_parms parms = { 1UL << bit, -1UL, cr };
506
507 on_each_cpu(smp_ctl_bit_callback, &parms, 1);
508}
509EXPORT_SYMBOL(smp_ctl_set_bit);
510
511/*
512 * Clear a bit in a control register of all cpus
513 */
514void smp_ctl_clear_bit(int cr, int bit)
515{
516 struct ec_creg_mask_parms parms = { 0, ~(1UL << bit), cr };
517
518 on_each_cpu(smp_ctl_bit_callback, &parms, 1);
519}
520EXPORT_SYMBOL(smp_ctl_clear_bit);
521
522#if defined(CONFIG_ZFCPDUMP) || defined(CONFIG_CRASH_DUMP)
523
524static void __init smp_get_save_area(int cpu, u16 address)
525{
526 void *lc = pcpu_devices[0].lowcore;
527 struct save_area *save_area;
528
529 if (is_kdump_kernel())
530 return;
531 if (!OLDMEM_BASE && (address == boot_cpu_address ||
532 ipl_info.type != IPL_TYPE_FCP_DUMP))
533 return;
534 save_area = dump_save_area_create(cpu);
535 if (!save_area)
536 panic("could not allocate memory for save area\n");
537#ifdef CONFIG_CRASH_DUMP
538 if (address == boot_cpu_address) {
539 /* Copy the registers of the boot cpu. */
540 copy_oldmem_page(1, (void *) save_area, sizeof(*save_area),
541 SAVE_AREA_BASE - PAGE_SIZE, 0);
542 return;
543 }
544#endif
545 /* Get the registers of a non-boot cpu. */
546 __pcpu_sigp_relax(address, SIGP_STOP_AND_STORE_STATUS, 0, NULL);
547 memcpy_real(save_area, lc + SAVE_AREA_BASE, sizeof(*save_area));
548}
549
550int smp_store_status(int cpu)
551{
552 struct pcpu *pcpu;
553
554 pcpu = pcpu_devices + cpu;
555 if (__pcpu_sigp_relax(pcpu->address, SIGP_STOP_AND_STORE_STATUS,
556 0, NULL) != SIGP_CC_ORDER_CODE_ACCEPTED)
557 return -EIO;
558 return 0;
559}
560
561#else /* CONFIG_ZFCPDUMP || CONFIG_CRASH_DUMP */
562
563static inline void smp_get_save_area(int cpu, u16 address) { }
564
565#endif /* CONFIG_ZFCPDUMP || CONFIG_CRASH_DUMP */
566
567void smp_cpu_set_polarization(int cpu, int val)
568{
569 pcpu_devices[cpu].polarization = val;
570}
571
572int smp_cpu_get_polarization(int cpu)
573{
574 return pcpu_devices[cpu].polarization;
575}
576
577static struct sclp_cpu_info *smp_get_cpu_info(void)
578{
579 static int use_sigp_detection;
580 struct sclp_cpu_info *info;
581 int address;
582
583 info = kzalloc(sizeof(*info), GFP_KERNEL);
584 if (info && (use_sigp_detection || sclp_get_cpu_info(info))) {
585 use_sigp_detection = 1;
586 for (address = 0; address <= MAX_CPU_ADDRESS; address++) {
587 if (__pcpu_sigp_relax(address, SIGP_SENSE, 0, NULL) ==
588 SIGP_CC_NOT_OPERATIONAL)
589 continue;
590 info->cpu[info->configured].address = address;
591 info->configured++;
592 }
593 info->combined = info->configured;
594 }
595 return info;
596}
597
598static int smp_add_present_cpu(int cpu);
599
600static int __smp_rescan_cpus(struct sclp_cpu_info *info, int sysfs_add)
601{
602 struct pcpu *pcpu;
603 cpumask_t avail;
604 int cpu, nr, i;
605
606 nr = 0;
607 cpumask_xor(&avail, cpu_possible_mask, cpu_present_mask);
608 cpu = cpumask_first(&avail);
609 for (i = 0; (i < info->combined) && (cpu < nr_cpu_ids); i++) {
610 if (info->has_cpu_type && info->cpu[i].type != boot_cpu_type)
611 continue;
612 if (pcpu_find_address(cpu_present_mask, info->cpu[i].address))
613 continue;
614 pcpu = pcpu_devices + cpu;
615 pcpu->address = info->cpu[i].address;
616 pcpu->state = (i >= info->configured) ?
617 CPU_STATE_STANDBY : CPU_STATE_CONFIGURED;
618 smp_cpu_set_polarization(cpu, POLARIZATION_UNKNOWN);
619 set_cpu_present(cpu, true);
620 if (sysfs_add && smp_add_present_cpu(cpu) != 0)
621 set_cpu_present(cpu, false);
622 else
623 nr++;
624 cpu = cpumask_next(cpu, &avail);
625 }
626 return nr;
627}
628
629static void __init smp_detect_cpus(void)
630{
631 unsigned int cpu, c_cpus, s_cpus;
632 struct sclp_cpu_info *info;
633
634 info = smp_get_cpu_info();
635 if (!info)
636 panic("smp_detect_cpus failed to allocate memory\n");
637 if (info->has_cpu_type) {
638 for (cpu = 0; cpu < info->combined; cpu++) {
639 if (info->cpu[cpu].address != boot_cpu_address)
640 continue;
641 /* The boot cpu dictates the cpu type. */
642 boot_cpu_type = info->cpu[cpu].type;
643 break;
644 }
645 }
646 c_cpus = s_cpus = 0;
647 for (cpu = 0; cpu < info->combined; cpu++) {
648 if (info->has_cpu_type && info->cpu[cpu].type != boot_cpu_type)
649 continue;
650 if (cpu < info->configured) {
651 smp_get_save_area(c_cpus, info->cpu[cpu].address);
652 c_cpus++;
653 } else
654 s_cpus++;
655 }
656 pr_info("%d configured CPUs, %d standby CPUs\n", c_cpus, s_cpus);
657 get_online_cpus();
658 __smp_rescan_cpus(info, 0);
659 put_online_cpus();
660 kfree(info);
661}
662
663/*
664 * Activate a secondary processor.
665 */
666static void smp_start_secondary(void *cpuvoid)
667{
668 S390_lowcore.last_update_clock = get_tod_clock();
669 S390_lowcore.restart_stack = (unsigned long) restart_stack;
670 S390_lowcore.restart_fn = (unsigned long) do_restart;
671 S390_lowcore.restart_data = 0;
672 S390_lowcore.restart_source = -1UL;
673 restore_access_regs(S390_lowcore.access_regs_save_area);
674 __ctl_load(S390_lowcore.cregs_save_area, 0, 15);
675 __load_psw_mask(PSW_KERNEL_BITS | PSW_MASK_DAT);
676 cpu_init();
677 preempt_disable();
678 init_cpu_timer();
679 init_cpu_vtimer();
680 pfault_init();
681 notify_cpu_starting(smp_processor_id());
682 set_cpu_online(smp_processor_id(), true);
683 inc_irq_stat(CPU_RST);
684 local_irq_enable();
685 cpu_startup_entry(CPUHP_ONLINE);
686}
687
688/* Upping and downing of CPUs */
689int __cpu_up(unsigned int cpu, struct task_struct *tidle)
690{
691 struct pcpu *pcpu;
692 int rc;
693
694 pcpu = pcpu_devices + cpu;
695 if (pcpu->state != CPU_STATE_CONFIGURED)
696 return -EIO;
697 if (pcpu_sigp_retry(pcpu, SIGP_INITIAL_CPU_RESET, 0) !=
698 SIGP_CC_ORDER_CODE_ACCEPTED)
699 return -EIO;
700
701 rc = pcpu_alloc_lowcore(pcpu, cpu);
702 if (rc)
703 return rc;
704 pcpu_prepare_secondary(pcpu, cpu);
705 pcpu_attach_task(pcpu, tidle);
706 pcpu_start_fn(pcpu, smp_start_secondary, NULL);
707 while (!cpu_online(cpu))
708 cpu_relax();
709 return 0;
710}
711
712static unsigned int setup_possible_cpus __initdata;
713
714static int __init _setup_possible_cpus(char *s)
715{
716 get_option(&s, &setup_possible_cpus);
717 return 0;
718}
719early_param("possible_cpus", _setup_possible_cpus);
720
721#ifdef CONFIG_HOTPLUG_CPU
722
723int __cpu_disable(void)
724{
725 unsigned long cregs[16];
726
727 /* Handle possible pending IPIs */
728 smp_handle_ext_call();
729 set_cpu_online(smp_processor_id(), false);
730 /* Disable pseudo page faults on this cpu. */
731 pfault_fini();
732 /* Disable interrupt sources via control register. */
733 __ctl_store(cregs, 0, 15);
734 cregs[0] &= ~0x0000ee70UL; /* disable all external interrupts */
735 cregs[6] &= ~0xff000000UL; /* disable all I/O interrupts */
736 cregs[14] &= ~0x1f000000UL; /* disable most machine checks */
737 __ctl_load(cregs, 0, 15);
738 return 0;
739}
740
741void __cpu_die(unsigned int cpu)
742{
743 struct pcpu *pcpu;
744
745 /* Wait until target cpu is down */
746 pcpu = pcpu_devices + cpu;
747 while (!pcpu_stopped(pcpu))
748 cpu_relax();
749 pcpu_free_lowcore(pcpu);
750 atomic_dec(&init_mm.context.attach_count);
751 cpumask_clear_cpu(cpu, mm_cpumask(&init_mm));
752 if (MACHINE_HAS_TLB_LC)
753 cpumask_clear_cpu(cpu, &init_mm.context.cpu_attach_mask);
754}
755
756void __noreturn cpu_die(void)
757{
758 idle_task_exit();
759 pcpu_sigp_retry(pcpu_devices + smp_processor_id(), SIGP_STOP, 0);
760 for (;;) ;
761}
762
763#endif /* CONFIG_HOTPLUG_CPU */
764
765void __init smp_fill_possible_mask(void)
766{
767 unsigned int possible, sclp, cpu;
768
769 sclp = sclp_get_max_cpu() ?: nr_cpu_ids;
770 possible = setup_possible_cpus ?: nr_cpu_ids;
771 possible = min(possible, sclp);
772 for (cpu = 0; cpu < possible && cpu < nr_cpu_ids; cpu++)
773 set_cpu_possible(cpu, true);
774}
775
776void __init smp_prepare_cpus(unsigned int max_cpus)
777{
778 /* request the 0x1201 emergency signal external interrupt */
779 if (register_external_irq(EXT_IRQ_EMERGENCY_SIG, do_ext_call_interrupt))
780 panic("Couldn't request external interrupt 0x1201");
781 /* request the 0x1202 external call external interrupt */
782 if (register_external_irq(EXT_IRQ_EXTERNAL_CALL, do_ext_call_interrupt))
783 panic("Couldn't request external interrupt 0x1202");
784 smp_detect_cpus();
785}
786
787void __init smp_prepare_boot_cpu(void)
788{
789 struct pcpu *pcpu = pcpu_devices;
790
791 boot_cpu_address = stap();
792 pcpu->state = CPU_STATE_CONFIGURED;
793 pcpu->address = boot_cpu_address;
794 pcpu->lowcore = (struct _lowcore *)(unsigned long) store_prefix();
795 pcpu->async_stack = S390_lowcore.async_stack - ASYNC_SIZE
796 + STACK_FRAME_OVERHEAD + sizeof(struct pt_regs);
797 pcpu->panic_stack = S390_lowcore.panic_stack - PAGE_SIZE
798 + STACK_FRAME_OVERHEAD + sizeof(struct pt_regs);
799 S390_lowcore.percpu_offset = __per_cpu_offset[0];
800 smp_cpu_set_polarization(0, POLARIZATION_UNKNOWN);
801 set_cpu_present(0, true);
802 set_cpu_online(0, true);
803}
804
805void __init smp_cpus_done(unsigned int max_cpus)
806{
807}
808
809void __init smp_setup_processor_id(void)
810{
811 S390_lowcore.cpu_nr = 0;
812}
813
814/*
815 * the frequency of the profiling timer can be changed
816 * by writing a multiplier value into /proc/profile.
817 *
818 * usually you want to run this on all CPUs ;)
819 */
820int setup_profiling_timer(unsigned int multiplier)
821{
822 return 0;
823}
824
825#ifdef CONFIG_HOTPLUG_CPU
826static ssize_t cpu_configure_show(struct device *dev,
827 struct device_attribute *attr, char *buf)
828{
829 ssize_t count;
830
831 mutex_lock(&smp_cpu_state_mutex);
832 count = sprintf(buf, "%d\n", pcpu_devices[dev->id].state);
833 mutex_unlock(&smp_cpu_state_mutex);
834 return count;
835}
836
837static ssize_t cpu_configure_store(struct device *dev,
838 struct device_attribute *attr,
839 const char *buf, size_t count)
840{
841 struct pcpu *pcpu;
842 int cpu, val, rc;
843 char delim;
844
845 if (sscanf(buf, "%d %c", &val, &delim) != 1)
846 return -EINVAL;
847 if (val != 0 && val != 1)
848 return -EINVAL;
849 get_online_cpus();
850 mutex_lock(&smp_cpu_state_mutex);
851 rc = -EBUSY;
852 /* disallow configuration changes of online cpus and cpu 0 */
853 cpu = dev->id;
854 if (cpu_online(cpu) || cpu == 0)
855 goto out;
856 pcpu = pcpu_devices + cpu;
857 rc = 0;
858 switch (val) {
859 case 0:
860 if (pcpu->state != CPU_STATE_CONFIGURED)
861 break;
862 rc = sclp_cpu_deconfigure(pcpu->address);
863 if (rc)
864 break;
865 pcpu->state = CPU_STATE_STANDBY;
866 smp_cpu_set_polarization(cpu, POLARIZATION_UNKNOWN);
867 topology_expect_change();
868 break;
869 case 1:
870 if (pcpu->state != CPU_STATE_STANDBY)
871 break;
872 rc = sclp_cpu_configure(pcpu->address);
873 if (rc)
874 break;
875 pcpu->state = CPU_STATE_CONFIGURED;
876 smp_cpu_set_polarization(cpu, POLARIZATION_UNKNOWN);
877 topology_expect_change();
878 break;
879 default:
880 break;
881 }
882out:
883 mutex_unlock(&smp_cpu_state_mutex);
884 put_online_cpus();
885 return rc ? rc : count;
886}
887static DEVICE_ATTR(configure, 0644, cpu_configure_show, cpu_configure_store);
888#endif /* CONFIG_HOTPLUG_CPU */
889
890static ssize_t show_cpu_address(struct device *dev,
891 struct device_attribute *attr, char *buf)
892{
893 return sprintf(buf, "%d\n", pcpu_devices[dev->id].address);
894}
895static DEVICE_ATTR(address, 0444, show_cpu_address, NULL);
896
897static struct attribute *cpu_common_attrs[] = {
898#ifdef CONFIG_HOTPLUG_CPU
899 &dev_attr_configure.attr,
900#endif
901 &dev_attr_address.attr,
902 NULL,
903};
904
905static struct attribute_group cpu_common_attr_group = {
906 .attrs = cpu_common_attrs,
907};
908
909static ssize_t show_idle_count(struct device *dev,
910 struct device_attribute *attr, char *buf)
911{
912 struct s390_idle_data *idle = &per_cpu(s390_idle, dev->id);
913 unsigned long long idle_count;
914 unsigned int sequence;
915
916 do {
917 sequence = ACCESS_ONCE(idle->sequence);
918 idle_count = ACCESS_ONCE(idle->idle_count);
919 if (ACCESS_ONCE(idle->clock_idle_enter))
920 idle_count++;
921 } while ((sequence & 1) || (ACCESS_ONCE(idle->sequence) != sequence));
922 return sprintf(buf, "%llu\n", idle_count);
923}
924static DEVICE_ATTR(idle_count, 0444, show_idle_count, NULL);
925
926static ssize_t show_idle_time(struct device *dev,
927 struct device_attribute *attr, char *buf)
928{
929 struct s390_idle_data *idle = &per_cpu(s390_idle, dev->id);
930 unsigned long long now, idle_time, idle_enter, idle_exit;
931 unsigned int sequence;
932
933 do {
934 now = get_tod_clock();
935 sequence = ACCESS_ONCE(idle->sequence);
936 idle_time = ACCESS_ONCE(idle->idle_time);
937 idle_enter = ACCESS_ONCE(idle->clock_idle_enter);
938 idle_exit = ACCESS_ONCE(idle->clock_idle_exit);
939 } while ((sequence & 1) || (ACCESS_ONCE(idle->sequence) != sequence));
940 idle_time += idle_enter ? ((idle_exit ? : now) - idle_enter) : 0;
941 return sprintf(buf, "%llu\n", idle_time >> 12);
942}
943static DEVICE_ATTR(idle_time_us, 0444, show_idle_time, NULL);
944
945static struct attribute *cpu_online_attrs[] = {
946 &dev_attr_idle_count.attr,
947 &dev_attr_idle_time_us.attr,
948 NULL,
949};
950
951static struct attribute_group cpu_online_attr_group = {
952 .attrs = cpu_online_attrs,
953};
954
955static int smp_cpu_notify(struct notifier_block *self, unsigned long action,
956 void *hcpu)
957{
958 unsigned int cpu = (unsigned int)(long)hcpu;
959 struct cpu *c = pcpu_devices[cpu].cpu;
960 struct device *s = &c->dev;
961 int err = 0;
962
963 switch (action & ~CPU_TASKS_FROZEN) {
964 case CPU_ONLINE:
965 err = sysfs_create_group(&s->kobj, &cpu_online_attr_group);
966 break;
967 case CPU_DEAD:
968 sysfs_remove_group(&s->kobj, &cpu_online_attr_group);
969 break;
970 }
971 return notifier_from_errno(err);
972}
973
974static int smp_add_present_cpu(int cpu)
975{
976 struct device *s;
977 struct cpu *c;
978 int rc;
979
980 c = kzalloc(sizeof(*c), GFP_KERNEL);
981 if (!c)
982 return -ENOMEM;
983 pcpu_devices[cpu].cpu = c;
984 s = &c->dev;
985 c->hotpluggable = 1;
986 rc = register_cpu(c, cpu);
987 if (rc)
988 goto out;
989 rc = sysfs_create_group(&s->kobj, &cpu_common_attr_group);
990 if (rc)
991 goto out_cpu;
992 if (cpu_online(cpu)) {
993 rc = sysfs_create_group(&s->kobj, &cpu_online_attr_group);
994 if (rc)
995 goto out_online;
996 }
997 rc = topology_cpu_init(c);
998 if (rc)
999 goto out_topology;
1000 return 0;
1001
1002out_topology:
1003 if (cpu_online(cpu))
1004 sysfs_remove_group(&s->kobj, &cpu_online_attr_group);
1005out_online:
1006 sysfs_remove_group(&s->kobj, &cpu_common_attr_group);
1007out_cpu:
1008#ifdef CONFIG_HOTPLUG_CPU
1009 unregister_cpu(c);
1010#endif
1011out:
1012 return rc;
1013}
1014
1015#ifdef CONFIG_HOTPLUG_CPU
1016
1017int __ref smp_rescan_cpus(void)
1018{
1019 struct sclp_cpu_info *info;
1020 int nr;
1021
1022 info = smp_get_cpu_info();
1023 if (!info)
1024 return -ENOMEM;
1025 get_online_cpus();
1026 mutex_lock(&smp_cpu_state_mutex);
1027 nr = __smp_rescan_cpus(info, 1);
1028 mutex_unlock(&smp_cpu_state_mutex);
1029 put_online_cpus();
1030 kfree(info);
1031 if (nr)
1032 topology_schedule_update();
1033 return 0;
1034}
1035
1036static ssize_t __ref rescan_store(struct device *dev,
1037 struct device_attribute *attr,
1038 const char *buf,
1039 size_t count)
1040{
1041 int rc;
1042
1043 rc = smp_rescan_cpus();
1044 return rc ? rc : count;
1045}
1046static DEVICE_ATTR(rescan, 0200, NULL, rescan_store);
1047#endif /* CONFIG_HOTPLUG_CPU */
1048
1049static int __init s390_smp_init(void)
1050{
1051 int cpu, rc = 0;
1052
1053#ifdef CONFIG_HOTPLUG_CPU
1054 rc = device_create_file(cpu_subsys.dev_root, &dev_attr_rescan);
1055 if (rc)
1056 return rc;
1057#endif
1058 cpu_notifier_register_begin();
1059 for_each_present_cpu(cpu) {
1060 rc = smp_add_present_cpu(cpu);
1061 if (rc)
1062 goto out;
1063 }
1064
1065 __hotcpu_notifier(smp_cpu_notify, 0);
1066
1067out:
1068 cpu_notifier_register_done();
1069 return rc;
1070}
1071subsys_initcall(s390_smp_init);