projects/linux·blob master · refs

1155 lines · 29.5 KiB

1// SPDX-License-Identifier: GPL-2.0
2/*
3 * SMP related functions
4 *
5 * Copyright IBM Corp. 1999, 2012
6 * Author(s): Denis Joseph Barrow,
7 * Martin Schwidefsky <[email protected]>,
8 *
9 * based on other smp stuff by
10 * (c) 1995 Alan Cox, CymruNET Ltd <[email protected]>
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 pr_fmt(fmt) "cpu: " fmt
19
20#include <linux/cpufeature.h>
21#include <linux/workqueue.h>
22#include <linux/memblock.h>
23#include <linux/export.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/irq_work.h>
33#include <linux/cpu.h>
34#include <linux/slab.h>
35#include <linux/sched/hotplug.h>
36#include <linux/sched/task_stack.h>
37#include <linux/crash_dump.h>
38#include <linux/kprobes.h>
39#include <asm/access-regs.h>
40#include <asm/asm-offsets.h>
41#include <asm/machine.h>
42#include <asm/ctlreg.h>
43#include <asm/pfault.h>
44#include <asm/diag.h>
45#include <asm/facility.h>
46#include <asm/fpu.h>
47#include <asm/ipl.h>
48#include <asm/setup.h>
49#include <asm/irq.h>
50#include <asm/tlbflush.h>
51#include <asm/vtime.h>
52#include <asm/abs_lowcore.h>
53#include <asm/sclp.h>
54#include <asm/debug.h>
55#include <asm/os_info.h>
56#include <asm/sigp.h>
57#include <asm/nmi.h>
58#include <asm/stacktrace.h>
59#include <asm/topology.h>
60#include <asm/vdso.h>
61#include <asm/maccess.h>
62#include "entry.h"
63
64enum {
65 ec_schedule = 0,
66 ec_call_function_single,
67 ec_stop_cpu,
68 ec_mcck_pending,
69 ec_irq_work,
70};
71
72enum {
73 CPU_STATE_STANDBY,
74 CPU_STATE_CONFIGURED,
75};
76
77static u8 boot_core_type;
78DEFINE_PER_CPU(struct pcpu, pcpu_devices);
79/*
80 * Pointer to the pcpu area of the boot CPU. This is required when a restart
81 * interrupt is triggered on an offline CPU. For that case accessing percpu
82 * data with the common primitives does not work, since the percpu offset is
83 * stored in a non existent lowcore.
84 */
85static struct pcpu *ipl_pcpu;
86
87unsigned int smp_cpu_mt_shift;
88EXPORT_SYMBOL(smp_cpu_mt_shift);
89
90unsigned int smp_cpu_mtid;
91EXPORT_SYMBOL(smp_cpu_mtid);
92
93#ifdef CONFIG_CRASH_DUMP
94__vector128 __initdata boot_cpu_vector_save_area[__NUM_VXRS];
95#endif
96
97static unsigned int smp_max_threads __initdata = -1U;
98cpumask_t cpu_setup_mask;
99
100static int __init early_smt(char *s)
101{
102 get_option(&s, &smp_max_threads);
103 return 0;
104}
105early_param("smt", early_smt);
106
107/*
108 * The smp_cpu_state_mutex must be held when changing the state or polarization
109 * member of a pcpu data structure within the pcpu_devices array.
110 */
111DEFINE_MUTEX(smp_cpu_state_mutex);
112
113/*
114 * Signal processor helper functions.
115 */
116static inline int __pcpu_sigp_relax(u16 addr, u8 order, unsigned long parm)
117{
118 int cc;
119
120 while (1) {
121 cc = __pcpu_sigp(addr, order, parm, NULL);
122 if (cc != SIGP_CC_BUSY)
123 return cc;
124 cpu_relax();
125 }
126}
127
128static int pcpu_sigp_retry(struct pcpu *pcpu, u8 order, u32 parm)
129{
130 int cc, retry;
131
132 for (retry = 0; ; retry++) {
133 cc = __pcpu_sigp(pcpu->address, order, parm, NULL);
134 if (cc != SIGP_CC_BUSY)
135 break;
136 if (retry >= 3)
137 udelay(10);
138 }
139 return cc;
140}
141
142static inline int pcpu_stopped(struct pcpu *pcpu)
143{
144 u32 status;
145
146 if (__pcpu_sigp(pcpu->address, SIGP_SENSE,
147 0, &status) != SIGP_CC_STATUS_STORED)
148 return 0;
149 return !!(status & (SIGP_STATUS_CHECK_STOP|SIGP_STATUS_STOPPED));
150}
151
152static inline int pcpu_running(struct pcpu *pcpu)
153{
154 if (__pcpu_sigp(pcpu->address, SIGP_SENSE_RUNNING,
155 0, NULL) != SIGP_CC_STATUS_STORED)
156 return 1;
157 /* Status stored condition code is equivalent to cpu not running. */
158 return 0;
159}
160
161/*
162 * Find struct pcpu by cpu address.
163 */
164static struct pcpu *pcpu_find_address(const struct cpumask *mask, u16 address)
165{
166 int cpu;
167
168 for_each_cpu(cpu, mask)
169 if (per_cpu(pcpu_devices, cpu).address == address)
170 return &per_cpu(pcpu_devices, cpu);
171 return NULL;
172}
173
174static void pcpu_ec_call(struct pcpu *pcpu, int ec_bit)
175{
176 if (test_and_set_bit(ec_bit, &pcpu->ec_mask))
177 return;
178 pcpu->ec_clk = get_tod_clock_fast();
179 pcpu_sigp_retry(pcpu, SIGP_EXTERNAL_CALL, 0);
180}
181
182static int pcpu_alloc_lowcore(struct pcpu *pcpu, int cpu)
183{
184 unsigned long async_stack, nodat_stack, mcck_stack;
185 struct lowcore *lc;
186
187 lc = (struct lowcore *) __get_free_pages(GFP_KERNEL | GFP_DMA, LC_ORDER);
188 nodat_stack = __get_free_pages(GFP_KERNEL, THREAD_SIZE_ORDER);
189 async_stack = stack_alloc();
190 mcck_stack = stack_alloc();
191 if (!lc || !nodat_stack || !async_stack || !mcck_stack)
192 goto out;
193 memcpy(lc, get_lowcore(), 512);
194 memset((char *) lc + 512, 0, sizeof(*lc) - 512);
195 lc->async_stack = async_stack + STACK_INIT_OFFSET;
196 lc->nodat_stack = nodat_stack + STACK_INIT_OFFSET;
197 lc->mcck_stack = mcck_stack + STACK_INIT_OFFSET;
198 lc->cpu_nr = cpu;
199 lc->spinlock_lockval = arch_spin_lockval(cpu);
200 lc->spinlock_index = 0;
201 lc->return_lpswe = gen_lpswe(__LC_RETURN_PSW);
202 lc->return_mcck_lpswe = gen_lpswe(__LC_RETURN_MCCK_PSW);
203 lc->preempt_count = PREEMPT_DISABLED;
204 if (nmi_alloc_mcesa(&lc->mcesad))
205 goto out;
206 if (abs_lowcore_map(cpu, lc, true))
207 goto out_mcesa;
208 lowcore_ptr[cpu] = lc;
209 pcpu_sigp_retry(pcpu, SIGP_SET_PREFIX, __pa(lc));
210 return 0;
211
212out_mcesa:
213 nmi_free_mcesa(&lc->mcesad);
214out:
215 stack_free(mcck_stack);
216 stack_free(async_stack);
217 free_pages(nodat_stack, THREAD_SIZE_ORDER);
218 free_pages((unsigned long) lc, LC_ORDER);
219 return -ENOMEM;
220}
221
222static void pcpu_free_lowcore(struct pcpu *pcpu, int cpu)
223{
224 unsigned long async_stack, nodat_stack, mcck_stack;
225 struct lowcore *lc;
226
227 lc = lowcore_ptr[cpu];
228 nodat_stack = lc->nodat_stack - STACK_INIT_OFFSET;
229 async_stack = lc->async_stack - STACK_INIT_OFFSET;
230 mcck_stack = lc->mcck_stack - STACK_INIT_OFFSET;
231 pcpu_sigp_retry(pcpu, SIGP_SET_PREFIX, 0);
232 lowcore_ptr[cpu] = NULL;
233 abs_lowcore_unmap(cpu);
234 nmi_free_mcesa(&lc->mcesad);
235 stack_free(async_stack);
236 stack_free(mcck_stack);
237 free_pages(nodat_stack, THREAD_SIZE_ORDER);
238 free_pages((unsigned long) lc, LC_ORDER);
239}
240
241static void pcpu_prepare_secondary(struct pcpu *pcpu, int cpu)
242{
243 struct lowcore *lc, *abs_lc;
244
245 lc = lowcore_ptr[cpu];
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->pcpu = (unsigned long)pcpu;
250 lc->restart_flags = RESTART_FLAG_CTLREGS;
251 lc->spinlock_lockval = arch_spin_lockval(cpu);
252 lc->spinlock_index = 0;
253 lc->percpu_offset = __per_cpu_offset[cpu];
254 lc->kernel_asce = get_lowcore()->kernel_asce;
255 lc->user_asce = s390_invalid_asce;
256 lc->user_timer = lc->system_timer =
257 lc->steal_timer = lc->avg_steal_timer = 0;
258 abs_lc = get_abs_lowcore();
259 memcpy(lc->cregs_save_area, abs_lc->cregs_save_area, sizeof(lc->cregs_save_area));
260 put_abs_lowcore(abs_lc);
261 lc->cregs_save_area[1] = lc->user_asce;
262 lc->cregs_save_area[7] = lc->user_asce;
263 save_access_regs((unsigned int *) lc->access_regs_save_area);
264 arch_spin_lock_setup(cpu);
265}
266
267static void pcpu_attach_task(int cpu, struct task_struct *tsk)
268{
269 struct lowcore *lc;
270
271 lc = lowcore_ptr[cpu];
272 lc->kernel_stack = (unsigned long)task_stack_page(tsk) + STACK_INIT_OFFSET;
273 lc->current_task = (unsigned long)tsk;
274 lc->lpp = LPP_MAGIC;
275 lc->current_pid = tsk->pid;
276 lc->user_timer = tsk->thread.user_timer;
277 lc->guest_timer = tsk->thread.guest_timer;
278 lc->system_timer = tsk->thread.system_timer;
279 lc->hardirq_timer = tsk->thread.hardirq_timer;
280 lc->softirq_timer = tsk->thread.softirq_timer;
281 lc->steal_timer = 0;
282#ifdef CONFIG_STACKPROTECTOR
283 lc->stack_canary = tsk->stack_canary;
284#endif
285}
286
287static void pcpu_start_fn(int cpu, void (*func)(void *), void *data)
288{
289 struct lowcore *lc;
290
291 lc = lowcore_ptr[cpu];
292 lc->restart_stack = lc->kernel_stack;
293 lc->restart_fn = (unsigned long) func;
294 lc->restart_data = (unsigned long) data;
295 lc->restart_source = -1U;
296 pcpu_sigp_retry(per_cpu_ptr(&pcpu_devices, cpu), SIGP_RESTART, 0);
297}
298
299typedef void (pcpu_delegate_fn)(void *);
300
301/*
302 * Call function via PSW restart on pcpu and stop the current cpu.
303 */
304static void __pcpu_delegate(pcpu_delegate_fn *func, void *data)
305{
306 func(data); /* should not return */
307}
308
309static void __noreturn pcpu_delegate(struct pcpu *pcpu, int cpu,
310 pcpu_delegate_fn *func,
311 void *data, unsigned long stack)
312{
313 struct lowcore *lc, *abs_lc;
314 unsigned int source_cpu;
315
316 lc = lowcore_ptr[cpu];
317 source_cpu = stap();
318
319 if (pcpu->address == source_cpu) {
320 call_on_stack(2, stack, void, __pcpu_delegate,
321 pcpu_delegate_fn *, func, void *, data);
322 }
323 /* Stop target cpu (if func returns this stops the current cpu). */
324 pcpu_sigp_retry(pcpu, SIGP_STOP, 0);
325 pcpu_sigp_retry(pcpu, SIGP_CPU_RESET, 0);
326 /* Restart func on the target cpu and stop the current cpu. */
327 if (lc) {
328 lc->restart_stack = stack;
329 lc->restart_fn = (unsigned long)func;
330 lc->restart_data = (unsigned long)data;
331 lc->restart_source = source_cpu;
332 } else {
333 abs_lc = get_abs_lowcore();
334 abs_lc->restart_stack = stack;
335 abs_lc->restart_fn = (unsigned long)func;
336 abs_lc->restart_data = (unsigned long)data;
337 abs_lc->restart_source = source_cpu;
338 put_abs_lowcore(abs_lc);
339 }
340 asm volatile(
341 "0: sigp 0,%0,%2 # sigp restart to target cpu\n"
342 " brc 2,0b # busy, try again\n"
343 "1: sigp 0,%1,%3 # sigp stop to current cpu\n"
344 " brc 2,1b # busy, try again"
345 : : "d" (pcpu->address), "d" (source_cpu),
346 "K" (SIGP_RESTART), "K" (SIGP_STOP)
347 : "0", "1", "cc");
348 for (;;) ;
349}
350
351/*
352 * Enable additional logical cpus for multi-threading.
353 */
354static int pcpu_set_smt(unsigned int mtid)
355{
356 int cc;
357
358 if (smp_cpu_mtid == mtid)
359 return 0;
360 cc = __pcpu_sigp(0, SIGP_SET_MULTI_THREADING, mtid, NULL);
361 if (cc == 0) {
362 smp_cpu_mtid = mtid;
363 smp_cpu_mt_shift = 0;
364 while (smp_cpu_mtid >= (1U << smp_cpu_mt_shift))
365 smp_cpu_mt_shift++;
366 per_cpu(pcpu_devices, 0).address = stap();
367 }
368 return cc;
369}
370
371/*
372 * Call function on the ipl CPU.
373 */
374void __noreturn smp_call_ipl_cpu(void (*func)(void *), void *data)
375{
376 struct lowcore *lc = lowcore_ptr[0];
377
378 if (ipl_pcpu->address == stap())
379 lc = get_lowcore();
380
381 pcpu_delegate(ipl_pcpu, 0, func, data, lc->nodat_stack);
382}
383
384int smp_find_processor_id(u16 address)
385{
386 int cpu;
387
388 for_each_present_cpu(cpu)
389 if (per_cpu(pcpu_devices, cpu).address == address)
390 return cpu;
391 return -1;
392}
393
394void schedule_mcck_handler(void)
395{
396 pcpu_ec_call(this_cpu_ptr(&pcpu_devices), ec_mcck_pending);
397}
398
399bool notrace arch_vcpu_is_preempted(int cpu)
400{
401 if (test_cpu_flag_of(CIF_ENABLED_WAIT, cpu))
402 return false;
403 if (pcpu_running(per_cpu_ptr(&pcpu_devices, cpu)))
404 return false;
405 return true;
406}
407EXPORT_SYMBOL(arch_vcpu_is_preempted);
408
409void notrace smp_yield_cpu(int cpu)
410{
411 if (!machine_has_diag9c())
412 return;
413 diag_stat_inc_norecursion(DIAG_STAT_X09C);
414 asm volatile("diag %0,0,0x9c"
415 : : "d" (per_cpu(pcpu_devices, cpu).address));
416}
417EXPORT_SYMBOL_GPL(smp_yield_cpu);
418
419/*
420 * Send cpus emergency shutdown signal. This gives the cpus the
421 * opportunity to complete outstanding interrupts.
422 */
423void notrace smp_emergency_stop(void)
424{
425 static arch_spinlock_t lock = __ARCH_SPIN_LOCK_UNLOCKED;
426 static cpumask_t cpumask;
427 u64 end;
428 int cpu;
429
430 arch_spin_lock(&lock);
431 cpumask_copy(&cpumask, cpu_online_mask);
432 cpumask_clear_cpu(smp_processor_id(), &cpumask);
433
434 end = get_tod_clock_monotonic() + (1000000UL << 12);
435 for_each_cpu(cpu, &cpumask) {
436 struct pcpu *pcpu = per_cpu_ptr(&pcpu_devices, cpu);
437 set_bit(ec_stop_cpu, &pcpu->ec_mask);
438 while (__pcpu_sigp(pcpu->address, SIGP_EMERGENCY_SIGNAL,
439 0, NULL) == SIGP_CC_BUSY &&
440 get_tod_clock_monotonic() < end)
441 cpu_relax();
442 }
443 while (get_tod_clock_monotonic() < end) {
444 for_each_cpu(cpu, &cpumask)
445 if (pcpu_stopped(per_cpu_ptr(&pcpu_devices, cpu)))
446 cpumask_clear_cpu(cpu, &cpumask);
447 if (cpumask_empty(&cpumask))
448 break;
449 cpu_relax();
450 }
451 arch_spin_unlock(&lock);
452}
453NOKPROBE_SYMBOL(smp_emergency_stop);
454
455/*
456 * Stop all cpus but the current one.
457 */
458void smp_send_stop(void)
459{
460 struct pcpu *pcpu;
461 int cpu;
462
463 /* Disable all interrupts/machine checks */
464 __load_psw_mask(PSW_KERNEL_BITS);
465 trace_hardirqs_off();
466
467 debug_set_critical();
468
469 if (oops_in_progress)
470 smp_emergency_stop();
471
472 /* stop all processors */
473 for_each_online_cpu(cpu) {
474 if (cpu == smp_processor_id())
475 continue;
476 pcpu = per_cpu_ptr(&pcpu_devices, cpu);
477 pcpu_sigp_retry(pcpu, SIGP_STOP, 0);
478 while (!pcpu_stopped(pcpu))
479 cpu_relax();
480 }
481}
482
483/*
484 * This is the main routine where commands issued by other
485 * cpus are handled.
486 */
487static void smp_handle_ext_call(void)
488{
489 unsigned long bits;
490
491 /* handle bit signal external calls */
492 bits = this_cpu_xchg(pcpu_devices.ec_mask, 0);
493 if (test_bit(ec_stop_cpu, &bits))
494 smp_stop_cpu();
495 if (test_bit(ec_schedule, &bits))
496 scheduler_ipi();
497 if (test_bit(ec_call_function_single, &bits))
498 generic_smp_call_function_single_interrupt();
499 if (test_bit(ec_mcck_pending, &bits))
500 s390_handle_mcck();
501 if (test_bit(ec_irq_work, &bits))
502 irq_work_run();
503}
504
505static void do_ext_call_interrupt(struct ext_code ext_code,
506 unsigned int param32, unsigned long param64)
507{
508 inc_irq_stat(ext_code.code == 0x1202 ? IRQEXT_EXC : IRQEXT_EMS);
509 smp_handle_ext_call();
510}
511
512void arch_send_call_function_ipi_mask(const struct cpumask *mask)
513{
514 int cpu;
515
516 for_each_cpu(cpu, mask)
517 pcpu_ec_call(per_cpu_ptr(&pcpu_devices, cpu), ec_call_function_single);
518}
519
520void arch_send_call_function_single_ipi(int cpu)
521{
522 pcpu_ec_call(per_cpu_ptr(&pcpu_devices, cpu), ec_call_function_single);
523}
524
525/*
526 * this function sends a 'reschedule' IPI to another CPU.
527 * it goes straight through and wastes no time serializing
528 * anything. Worst case is that we lose a reschedule ...
529 */
530void arch_smp_send_reschedule(int cpu)
531{
532 pcpu_ec_call(per_cpu_ptr(&pcpu_devices, cpu), ec_schedule);
533}
534
535#ifdef CONFIG_IRQ_WORK
536void arch_irq_work_raise(void)
537{
538 pcpu_ec_call(this_cpu_ptr(&pcpu_devices), ec_irq_work);
539}
540#endif
541
542#ifdef CONFIG_CRASH_DUMP
543
544int smp_store_status(int cpu)
545{
546 struct lowcore *lc;
547 struct pcpu *pcpu;
548 unsigned long pa;
549
550 pcpu = per_cpu_ptr(&pcpu_devices, cpu);
551 lc = lowcore_ptr[cpu];
552 pa = __pa(&lc->floating_pt_save_area);
553 if (__pcpu_sigp_relax(pcpu->address, SIGP_STORE_STATUS_AT_ADDRESS,
554 pa) != SIGP_CC_ORDER_CODE_ACCEPTED)
555 return -EIO;
556 if (!cpu_has_vx() && !cpu_has_gs())
557 return 0;
558 pa = lc->mcesad & MCESA_ORIGIN_MASK;
559 if (cpu_has_gs())
560 pa |= lc->mcesad & MCESA_LC_MASK;
561 if (__pcpu_sigp_relax(pcpu->address, SIGP_STORE_ADDITIONAL_STATUS,
562 pa) != SIGP_CC_ORDER_CODE_ACCEPTED)
563 return -EIO;
564 return 0;
565}
566
567/*
568 * Collect CPU state of the previous, crashed system.
569 * There are three cases:
570 * 1) standard zfcp/nvme dump
571 * condition: OLDMEM_BASE == NULL && is_ipl_type_dump() == true
572 * The state for all CPUs except the boot CPU needs to be collected
573 * with sigp stop-and-store-status. The boot CPU state is located in
574 * the absolute lowcore of the memory stored in the HSA. The zcore code
575 * will copy the boot CPU state from the HSA.
576 * 2) stand-alone kdump for SCSI/NVMe (zfcp/nvme dump with swapped memory)
577 * condition: OLDMEM_BASE != NULL && is_ipl_type_dump() == true
578 * The state for all CPUs except the boot CPU needs to be collected
579 * with sigp stop-and-store-status. The firmware or the boot-loader
580 * stored the registers of the boot CPU in the absolute lowcore in the
581 * memory of the old system.
582 * 3) kdump or stand-alone kdump for DASD
583 * condition: OLDMEM_BASE != NULL && is_ipl_type_dump() == false
584 * The state for all CPUs except the boot CPU needs to be collected
585 * with sigp stop-and-store-status. The kexec code or the boot-loader
586 * stored the registers of the boot CPU in the memory of the old system.
587 *
588 * Note that the legacy kdump mode where the old kernel stored the CPU states
589 * does no longer exist: setup_arch() explicitly deactivates the elfcorehdr=
590 * kernel parameter. The is_kdump_kernel() implementation on s390 is independent
591 * of the elfcorehdr= parameter.
592 */
593static bool dump_available(void)
594{
595 return oldmem_data.start || is_ipl_type_dump();
596}
597
598void __init smp_save_dump_ipl_cpu(void)
599{
600 struct save_area *sa;
601 void *regs;
602
603 if (!dump_available())
604 return;
605 sa = save_area_alloc(true);
606 regs = memblock_alloc_or_panic(512, 8);
607 copy_oldmem_kernel(regs, __LC_FPREGS_SAVE_AREA, 512);
608 save_area_add_regs(sa, regs);
609 memblock_free(regs, 512);
610 if (cpu_has_vx())
611 save_area_add_vxrs(sa, boot_cpu_vector_save_area);
612}
613
614void __init smp_save_dump_secondary_cpus(void)
615{
616 int addr, boot_cpu_addr, max_cpu_addr;
617 struct save_area *sa;
618 void *page;
619
620 if (!dump_available())
621 return;
622 /* Allocate a page as dumping area for the store status sigps */
623 page = memblock_alloc_low(PAGE_SIZE, PAGE_SIZE);
624 if (!page)
625 panic("ERROR: Failed to allocate %lx bytes below %lx\n",
626 PAGE_SIZE, 1UL << 31);
627
628 /* Set multi-threading state to the previous system. */
629 pcpu_set_smt(sclp.mtid_prev);
630 boot_cpu_addr = stap();
631 max_cpu_addr = SCLP_MAX_CORES << sclp.mtid_prev;
632 for (addr = 0; addr <= max_cpu_addr; addr++) {
633 if (addr == boot_cpu_addr)
634 continue;
635 if (__pcpu_sigp_relax(addr, SIGP_SENSE, 0) ==
636 SIGP_CC_NOT_OPERATIONAL)
637 continue;
638 sa = save_area_alloc(false);
639 __pcpu_sigp_relax(addr, SIGP_STORE_STATUS_AT_ADDRESS, __pa(page));
640 save_area_add_regs(sa, page);
641 if (cpu_has_vx()) {
642 __pcpu_sigp_relax(addr, SIGP_STORE_ADDITIONAL_STATUS, __pa(page));
643 save_area_add_vxrs(sa, page);
644 }
645 }
646 memblock_free(page, PAGE_SIZE);
647 diag_amode31_ops.diag308_reset();
648 pcpu_set_smt(0);
649}
650#endif /* CONFIG_CRASH_DUMP */
651
652void smp_cpu_set_polarization(int cpu, int val)
653{
654 per_cpu(pcpu_devices, cpu).polarization = val;
655}
656
657int smp_cpu_get_polarization(int cpu)
658{
659 return per_cpu(pcpu_devices, cpu).polarization;
660}
661
662void smp_set_core_capacity(int cpu, unsigned long val)
663{
664 int i;
665
666 cpu = smp_get_base_cpu(cpu);
667 for (i = cpu; (i <= cpu + smp_cpu_mtid) && (i < nr_cpu_ids); i++)
668 topology_set_cpu_scale(i, val);
669}
670
671int smp_cpu_get_cpu_address(int cpu)
672{
673 return per_cpu(pcpu_devices, cpu).address;
674}
675
676static void __ref smp_get_core_info(struct sclp_core_info *info, int early)
677{
678 static int use_sigp_detection;
679 int address;
680
681 if (use_sigp_detection || sclp_get_core_info(info, early)) {
682 use_sigp_detection = 1;
683 for (address = 0;
684 address < (SCLP_MAX_CORES << smp_cpu_mt_shift);
685 address += (1U << smp_cpu_mt_shift)) {
686 if (__pcpu_sigp_relax(address, SIGP_SENSE, 0) ==
687 SIGP_CC_NOT_OPERATIONAL)
688 continue;
689 info->core[info->configured].core_id =
690 address >> smp_cpu_mt_shift;
691 info->core[info->configured].type = boot_core_type;
692 info->configured++;
693 }
694 info->combined = info->configured;
695 }
696}
697
698static int smp_add_core(struct sclp_core_entry *core, cpumask_t *avail,
699 bool configured, bool early)
700{
701 struct pcpu *pcpu;
702 int cpu, nr, i;
703 u16 address;
704
705 nr = 0;
706 if (sclp.has_core_type && core->type != boot_core_type)
707 return nr;
708 cpu = cpumask_first(avail);
709 address = core->core_id << smp_cpu_mt_shift;
710 for (i = 0; (i <= smp_cpu_mtid) && (cpu < nr_cpu_ids); i++) {
711 if (pcpu_find_address(cpu_present_mask, address + i))
712 continue;
713 pcpu = per_cpu_ptr(&pcpu_devices, cpu);
714 pcpu->address = address + i;
715 if (configured)
716 pcpu->state = CPU_STATE_CONFIGURED;
717 else
718 pcpu->state = CPU_STATE_STANDBY;
719 smp_cpu_set_polarization(cpu, POLARIZATION_UNKNOWN);
720 topology_set_cpu_scale(cpu, CPU_CAPACITY_HIGH);
721 set_cpu_present(cpu, true);
722 if (!early && arch_register_cpu(cpu))
723 set_cpu_present(cpu, false);
724 else
725 nr++;
726 cpumask_clear_cpu(cpu, avail);
727 cpu = cpumask_next(cpu, avail);
728 }
729 return nr;
730}
731
732static int __smp_rescan_cpus(struct sclp_core_info *info, bool early)
733{
734 struct sclp_core_entry *core;
735 static cpumask_t avail;
736 bool configured;
737 u16 core_id;
738 int nr, i;
739
740 cpus_read_lock();
741 mutex_lock(&smp_cpu_state_mutex);
742 nr = 0;
743 cpumask_xor(&avail, cpu_possible_mask, cpu_present_mask);
744 /*
745 * Add IPL core first (which got logical CPU number 0) to make sure
746 * that all SMT threads get subsequent logical CPU numbers.
747 */
748 if (early) {
749 core_id = per_cpu(pcpu_devices, 0).address >> smp_cpu_mt_shift;
750 for (i = 0; i < info->configured; i++) {
751 core = &info->core[i];
752 if (core->core_id == core_id) {
753 nr += smp_add_core(core, &avail, true, early);
754 break;
755 }
756 }
757 }
758 for (i = 0; i < info->combined; i++) {
759 configured = i < info->configured;
760 nr += smp_add_core(&info->core[i], &avail, configured, early);
761 }
762 mutex_unlock(&smp_cpu_state_mutex);
763 cpus_read_unlock();
764 return nr;
765}
766
767void __init smp_detect_cpus(void)
768{
769 unsigned int cpu, mtid, c_cpus, s_cpus;
770 struct sclp_core_info *info;
771 u16 address;
772
773 /* Get CPU information */
774 info = memblock_alloc_or_panic(sizeof(*info), 8);
775 smp_get_core_info(info, 1);
776 /* Find boot CPU type */
777 if (sclp.has_core_type) {
778 address = stap();
779 for (cpu = 0; cpu < info->combined; cpu++)
780 if (info->core[cpu].core_id == address) {
781 /* The boot cpu dictates the cpu type. */
782 boot_core_type = info->core[cpu].type;
783 break;
784 }
785 if (cpu >= info->combined)
786 panic("Could not find boot CPU type");
787 }
788
789 /* Set multi-threading state for the current system */
790 mtid = boot_core_type ? sclp.mtid : sclp.mtid_cp;
791 mtid = (mtid < smp_max_threads) ? mtid : smp_max_threads - 1;
792 pcpu_set_smt(mtid);
793 cpu_smt_set_num_threads(smp_cpu_mtid + 1, smp_cpu_mtid + 1);
794
795 /* Print number of CPUs */
796 c_cpus = s_cpus = 0;
797 for (cpu = 0; cpu < info->combined; cpu++) {
798 if (sclp.has_core_type &&
799 info->core[cpu].type != boot_core_type)
800 continue;
801 if (cpu < info->configured)
802 c_cpus += smp_cpu_mtid + 1;
803 else
804 s_cpus += smp_cpu_mtid + 1;
805 }
806 pr_info("%d configured CPUs, %d standby CPUs\n", c_cpus, s_cpus);
807 memblock_free(info, sizeof(*info));
808}
809
810/*
811 * Activate a secondary processor.
812 */
813static void smp_start_secondary(void *cpuvoid)
814{
815 struct lowcore *lc = get_lowcore();
816 int cpu = raw_smp_processor_id();
817
818 lc->last_update_clock = get_tod_clock();
819 lc->restart_stack = (unsigned long)restart_stack;
820 lc->restart_fn = (unsigned long)do_restart;
821 lc->restart_data = 0;
822 lc->restart_source = -1U;
823 lc->restart_flags = 0;
824 restore_access_regs(lc->access_regs_save_area);
825 cpu_init();
826 rcutree_report_cpu_starting(cpu);
827 init_cpu_timer();
828 vtime_init();
829 vdso_getcpu_init();
830 pfault_init();
831 cpumask_set_cpu(cpu, &cpu_setup_mask);
832 update_cpu_masks();
833 notify_cpu_starting(cpu);
834 if (topology_cpu_dedicated(cpu))
835 set_cpu_flag(CIF_DEDICATED_CPU);
836 else
837 clear_cpu_flag(CIF_DEDICATED_CPU);
838 set_cpu_online(cpu, true);
839 inc_irq_stat(CPU_RST);
840 local_irq_enable();
841 cpu_startup_entry(CPUHP_AP_ONLINE_IDLE);
842}
843
844/* Upping and downing of CPUs */
845int __cpu_up(unsigned int cpu, struct task_struct *tidle)
846{
847 struct pcpu *pcpu = per_cpu_ptr(&pcpu_devices, cpu);
848 int rc;
849
850 if (pcpu->state != CPU_STATE_CONFIGURED)
851 return -EIO;
852 if (pcpu_sigp_retry(pcpu, SIGP_INITIAL_CPU_RESET, 0) !=
853 SIGP_CC_ORDER_CODE_ACCEPTED)
854 return -EIO;
855
856 rc = pcpu_alloc_lowcore(pcpu, cpu);
857 if (rc)
858 return rc;
859 /*
860 * Make sure global control register contents do not change
861 * until new CPU has initialized control registers.
862 */
863 system_ctlreg_lock();
864 pcpu_prepare_secondary(pcpu, cpu);
865 pcpu_attach_task(cpu, tidle);
866 pcpu_start_fn(cpu, smp_start_secondary, NULL);
867 /* Wait until cpu puts itself in the online & active maps */
868 while (!cpu_online(cpu))
869 cpu_relax();
870 system_ctlreg_unlock();
871 return 0;
872}
873
874static unsigned int setup_possible_cpus __initdata;
875
876static int __init _setup_possible_cpus(char *s)
877{
878 get_option(&s, &setup_possible_cpus);
879 return 0;
880}
881early_param("possible_cpus", _setup_possible_cpus);
882
883int __cpu_disable(void)
884{
885 struct ctlreg cregs[16];
886 int cpu;
887
888 /* Handle possible pending IPIs */
889 smp_handle_ext_call();
890 cpu = smp_processor_id();
891 set_cpu_online(cpu, false);
892 cpumask_clear_cpu(cpu, &cpu_setup_mask);
893 update_cpu_masks();
894 /* Disable pseudo page faults on this cpu. */
895 pfault_fini();
896 /* Disable interrupt sources via control register. */
897 __local_ctl_store(0, 15, cregs);
898 cregs[0].val &= ~0x0000ee70UL; /* disable all external interrupts */
899 cregs[6].val &= ~0xff000000UL; /* disable all I/O interrupts */
900 cregs[14].val &= ~0x1f000000UL; /* disable most machine checks */
901 __local_ctl_load(0, 15, cregs);
902 return 0;
903}
904
905void __cpu_die(unsigned int cpu)
906{
907 struct pcpu *pcpu;
908
909 /* Wait until target cpu is down */
910 pcpu = per_cpu_ptr(&pcpu_devices, cpu);
911 while (!pcpu_stopped(pcpu))
912 cpu_relax();
913 pcpu_free_lowcore(pcpu, cpu);
914 cpumask_clear_cpu(cpu, mm_cpumask(&init_mm));
915 cpumask_clear_cpu(cpu, &init_mm.context.cpu_attach_mask);
916 pcpu->flags = 0;
917}
918
919void __noreturn cpu_die(void)
920{
921 idle_task_exit();
922 pcpu_sigp_retry(this_cpu_ptr(&pcpu_devices), SIGP_STOP, 0);
923 for (;;) ;
924}
925
926void __init smp_fill_possible_mask(void)
927{
928 unsigned int possible, sclp_max, cpu;
929
930 sclp_max = max(sclp.mtid, sclp.mtid_cp) + 1;
931 sclp_max = min(smp_max_threads, sclp_max);
932 sclp_max = (sclp.max_cores * sclp_max) ?: nr_cpu_ids;
933 possible = setup_possible_cpus ?: nr_cpu_ids;
934 possible = min(possible, sclp_max);
935 for (cpu = 0; cpu < possible && cpu < nr_cpu_ids; cpu++)
936 set_cpu_possible(cpu, true);
937}
938
939void __init smp_prepare_cpus(unsigned int max_cpus)
940{
941 if (register_external_irq(EXT_IRQ_EMERGENCY_SIG, do_ext_call_interrupt))
942 panic("Couldn't request external interrupt 0x1201");
943 system_ctl_set_bit(0, 14);
944 if (register_external_irq(EXT_IRQ_EXTERNAL_CALL, do_ext_call_interrupt))
945 panic("Couldn't request external interrupt 0x1202");
946 system_ctl_set_bit(0, 13);
947 smp_rescan_cpus(true);
948}
949
950void __init smp_prepare_boot_cpu(void)
951{
952 struct lowcore *lc = get_lowcore();
953
954 WARN_ON(!cpu_present(0) || !cpu_online(0));
955 lc->percpu_offset = __per_cpu_offset[0];
956 ipl_pcpu = per_cpu_ptr(&pcpu_devices, 0);
957 ipl_pcpu->state = CPU_STATE_CONFIGURED;
958 lc->pcpu = (unsigned long)ipl_pcpu;
959 smp_cpu_set_polarization(0, POLARIZATION_UNKNOWN);
960 topology_set_cpu_scale(0, CPU_CAPACITY_HIGH);
961}
962
963void __init smp_setup_processor_id(void)
964{
965 struct lowcore *lc = get_lowcore();
966
967 lc->cpu_nr = 0;
968 per_cpu(pcpu_devices, 0).address = stap();
969 lc->spinlock_lockval = arch_spin_lockval(0);
970 lc->spinlock_index = 0;
971}
972
973/*
974 * the frequency of the profiling timer can be changed
975 * by writing a multiplier value into /proc/profile.
976 *
977 * usually you want to run this on all CPUs ;)
978 */
979int setup_profiling_timer(unsigned int multiplier)
980{
981 return 0;
982}
983
984static ssize_t cpu_configure_show(struct device *dev,
985 struct device_attribute *attr, char *buf)
986{
987 ssize_t count;
988
989 mutex_lock(&smp_cpu_state_mutex);
990 count = sysfs_emit(buf, "%d\n", per_cpu(pcpu_devices, dev->id).state);
991 mutex_unlock(&smp_cpu_state_mutex);
992 return count;
993}
994
995static ssize_t cpu_configure_store(struct device *dev,
996 struct device_attribute *attr,
997 const char *buf, size_t count)
998{
999 struct pcpu *pcpu;
1000 int cpu, val, rc, i;
1001 char delim;
1002
1003 if (sscanf(buf, "%d %c", &val, &delim) != 1)
1004 return -EINVAL;
1005 if (val != 0 && val != 1)
1006 return -EINVAL;
1007 cpus_read_lock();
1008 mutex_lock(&smp_cpu_state_mutex);
1009 rc = -EBUSY;
1010 /* disallow configuration changes of online cpus */
1011 cpu = dev->id;
1012 cpu = smp_get_base_cpu(cpu);
1013 for (i = 0; i <= smp_cpu_mtid; i++)
1014 if (cpu_online(cpu + i))
1015 goto out;
1016 pcpu = per_cpu_ptr(&pcpu_devices, cpu);
1017 rc = 0;
1018 switch (val) {
1019 case 0:
1020 if (pcpu->state != CPU_STATE_CONFIGURED)
1021 break;
1022 rc = sclp_core_deconfigure(pcpu->address >> smp_cpu_mt_shift);
1023 if (rc)
1024 break;
1025 for (i = 0; i <= smp_cpu_mtid; i++) {
1026 if (cpu + i >= nr_cpu_ids || !cpu_present(cpu + i))
1027 continue;
1028 per_cpu(pcpu_devices, cpu + i).state = CPU_STATE_STANDBY;
1029 smp_cpu_set_polarization(cpu + i,
1030 POLARIZATION_UNKNOWN);
1031 set_cpu_enabled(cpu + i, false);
1032 }
1033 topology_expect_change();
1034 break;
1035 case 1:
1036 if (pcpu->state != CPU_STATE_STANDBY)
1037 break;
1038 rc = sclp_core_configure(pcpu->address >> smp_cpu_mt_shift);
1039 if (rc)
1040 break;
1041 for (i = 0; i <= smp_cpu_mtid; i++) {
1042 if (cpu + i >= nr_cpu_ids || !cpu_present(cpu + i))
1043 continue;
1044 per_cpu(pcpu_devices, cpu + i).state = CPU_STATE_CONFIGURED;
1045 smp_cpu_set_polarization(cpu + i,
1046 POLARIZATION_UNKNOWN);
1047 set_cpu_enabled(cpu + i, true);
1048 }
1049 topology_expect_change();
1050 break;
1051 default:
1052 break;
1053 }
1054out:
1055 mutex_unlock(&smp_cpu_state_mutex);
1056 cpus_read_unlock();
1057 return rc ? rc : count;
1058}
1059static DEVICE_ATTR(configure, 0644, cpu_configure_show, cpu_configure_store);
1060
1061static ssize_t show_cpu_address(struct device *dev,
1062 struct device_attribute *attr, char *buf)
1063{
1064 return sysfs_emit(buf, "%d\n", per_cpu(pcpu_devices, dev->id).address);
1065}
1066static DEVICE_ATTR(address, 0444, show_cpu_address, NULL);
1067
1068static struct attribute *cpu_common_attrs[] = {
1069 &dev_attr_configure.attr,
1070 &dev_attr_address.attr,
1071 NULL,
1072};
1073
1074static struct attribute_group cpu_common_attr_group = {
1075 .attrs = cpu_common_attrs,
1076};
1077
1078bool arch_cpu_is_hotpluggable(int cpu)
1079{
1080 return !!cpu;
1081}
1082
1083int arch_register_cpu(int cpu)
1084{
1085 struct pcpu *pcpu = per_cpu_ptr(&pcpu_devices, cpu);
1086 struct cpu *c = per_cpu_ptr(&cpu_devices, cpu);
1087 int rc;
1088
1089 c->hotpluggable = arch_cpu_is_hotpluggable(cpu);
1090 rc = register_cpu(c, cpu);
1091 if (rc)
1092 goto out;
1093 rc = sysfs_create_group(&c->dev.kobj, &cpu_common_attr_group);
1094 if (rc)
1095 goto out_cpu;
1096 rc = topology_cpu_init(c);
1097 if (rc)
1098 goto out_topology;
1099 if (pcpu->state != CPU_STATE_CONFIGURED)
1100 set_cpu_enabled(cpu, false);
1101 return 0;
1102
1103out_topology:
1104 sysfs_remove_group(&c->dev.kobj, &cpu_common_attr_group);
1105out_cpu:
1106 unregister_cpu(c);
1107out:
1108 return rc;
1109}
1110
1111int __ref smp_rescan_cpus(bool early)
1112{
1113 struct sclp_core_info *info;
1114 int nr;
1115
1116 info = kzalloc_obj(*info);
1117 if (!info)
1118 return -ENOMEM;
1119 smp_get_core_info(info, 0);
1120 nr = __smp_rescan_cpus(info, early);
1121 kfree(info);
1122 if (nr && !early)
1123 topology_schedule_update();
1124 return 0;
1125}
1126
1127static ssize_t __ref rescan_store(struct device *dev,
1128 struct device_attribute *attr,
1129 const char *buf,
1130 size_t count)
1131{
1132 int rc;
1133
1134 rc = lock_device_hotplug_sysfs();
1135 if (rc)
1136 return rc;
1137 rc = smp_rescan_cpus(false);
1138 unlock_device_hotplug();
1139 return rc ? rc : count;
1140}
1141static DEVICE_ATTR_WO(rescan);
1142
1143static int __init s390_smp_init(void)
1144{
1145 struct device *dev_root;
1146 int rc = 0;
1147
1148 dev_root = bus_get_dev_root(&cpu_subsys);
1149 if (dev_root) {
1150 rc = device_create_file(dev_root, &dev_attr_rescan);
1151 put_device(dev_root);
1152 }
1153 return rc;
1154}
1155subsys_initcall(s390_smp_init);