Loading...
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 <schwidefsky@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/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/ctlreg.h>
42#include <asm/pfault.h>
43#include <asm/diag.h>
44#include <asm/facility.h>
45#include <asm/fpu.h>
46#include <asm/ipl.h>
47#include <asm/setup.h>
48#include <asm/irq.h>
49#include <asm/tlbflush.h>
50#include <asm/vtimer.h>
51#include <asm/abs_lowcore.h>
52#include <asm/sclp.h>
53#include <asm/debug.h>
54#include <asm/os_info.h>
55#include <asm/sigp.h>
56#include <asm/idle.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_nosmt(char *s)
101{
102 smp_max_threads = 1;
103 return 0;
104}
105early_param("nosmt", early_nosmt);
106
107static int __init early_smt(char *s)
108{
109 get_option(&s, &smp_max_threads);
110 return 0;
111}
112early_param("smt", early_smt);
113
114/*
115 * The smp_cpu_state_mutex must be held when changing the state or polarization
116 * member of a pcpu data structure within the pcpu_devices array.
117 */
118DEFINE_MUTEX(smp_cpu_state_mutex);
119
120/*
121 * Signal processor helper functions.
122 */
123static inline int __pcpu_sigp_relax(u16 addr, u8 order, unsigned long parm)
124{
125 int cc;
126
127 while (1) {
128 cc = __pcpu_sigp(addr, order, parm, NULL);
129 if (cc != SIGP_CC_BUSY)
130 return cc;
131 cpu_relax();
132 }
133}
134
135static int pcpu_sigp_retry(struct pcpu *pcpu, u8 order, u32 parm)
136{
137 int cc, retry;
138
139 for (retry = 0; ; retry++) {
140 cc = __pcpu_sigp(pcpu->address, order, parm, NULL);
141 if (cc != SIGP_CC_BUSY)
142 break;
143 if (retry >= 3)
144 udelay(10);
145 }
146 return cc;
147}
148
149static inline int pcpu_stopped(struct pcpu *pcpu)
150{
151 u32 status;
152
153 if (__pcpu_sigp(pcpu->address, SIGP_SENSE,
154 0, &status) != SIGP_CC_STATUS_STORED)
155 return 0;
156 return !!(status & (SIGP_STATUS_CHECK_STOP|SIGP_STATUS_STOPPED));
157}
158
159static inline int pcpu_running(struct pcpu *pcpu)
160{
161 if (__pcpu_sigp(pcpu->address, SIGP_SENSE_RUNNING,
162 0, NULL) != SIGP_CC_STATUS_STORED)
163 return 1;
164 /* Status stored condition code is equivalent to cpu not running. */
165 return 0;
166}
167
168/*
169 * Find struct pcpu by cpu address.
170 */
171static struct pcpu *pcpu_find_address(const struct cpumask *mask, u16 address)
172{
173 int cpu;
174
175 for_each_cpu(cpu, mask)
176 if (per_cpu(pcpu_devices, cpu).address == address)
177 return &per_cpu(pcpu_devices, cpu);
178 return NULL;
179}
180
181static void pcpu_ec_call(struct pcpu *pcpu, int ec_bit)
182{
183 int order;
184
185 if (test_and_set_bit(ec_bit, &pcpu->ec_mask))
186 return;
187 order = pcpu_running(pcpu) ? SIGP_EXTERNAL_CALL : SIGP_EMERGENCY_SIGNAL;
188 pcpu->ec_clk = get_tod_clock_fast();
189 pcpu_sigp_retry(pcpu, order, 0);
190}
191
192static int pcpu_alloc_lowcore(struct pcpu *pcpu, int cpu)
193{
194 unsigned long async_stack, nodat_stack, mcck_stack;
195 struct lowcore *lc;
196
197 lc = (struct lowcore *) __get_free_pages(GFP_KERNEL | GFP_DMA, LC_ORDER);
198 nodat_stack = __get_free_pages(GFP_KERNEL, THREAD_SIZE_ORDER);
199 async_stack = stack_alloc();
200 mcck_stack = stack_alloc();
201 if (!lc || !nodat_stack || !async_stack || !mcck_stack)
202 goto out;
203 memcpy(lc, get_lowcore(), 512);
204 memset((char *) lc + 512, 0, sizeof(*lc) - 512);
205 lc->async_stack = async_stack + STACK_INIT_OFFSET;
206 lc->nodat_stack = nodat_stack + STACK_INIT_OFFSET;
207 lc->mcck_stack = mcck_stack + STACK_INIT_OFFSET;
208 lc->cpu_nr = cpu;
209 lc->spinlock_lockval = arch_spin_lockval(cpu);
210 lc->spinlock_index = 0;
211 lc->return_lpswe = gen_lpswe(__LC_RETURN_PSW);
212 lc->return_mcck_lpswe = gen_lpswe(__LC_RETURN_MCCK_PSW);
213 lc->preempt_count = PREEMPT_DISABLED;
214 if (nmi_alloc_mcesa(&lc->mcesad))
215 goto out;
216 if (abs_lowcore_map(cpu, lc, true))
217 goto out_mcesa;
218 lowcore_ptr[cpu] = lc;
219 pcpu_sigp_retry(pcpu, SIGP_SET_PREFIX, __pa(lc));
220 return 0;
221
222out_mcesa:
223 nmi_free_mcesa(&lc->mcesad);
224out:
225 stack_free(mcck_stack);
226 stack_free(async_stack);
227 free_pages(nodat_stack, THREAD_SIZE_ORDER);
228 free_pages((unsigned long) lc, LC_ORDER);
229 return -ENOMEM;
230}
231
232static void pcpu_free_lowcore(struct pcpu *pcpu, int cpu)
233{
234 unsigned long async_stack, nodat_stack, mcck_stack;
235 struct lowcore *lc;
236
237 lc = lowcore_ptr[cpu];
238 nodat_stack = lc->nodat_stack - STACK_INIT_OFFSET;
239 async_stack = lc->async_stack - STACK_INIT_OFFSET;
240 mcck_stack = lc->mcck_stack - STACK_INIT_OFFSET;
241 pcpu_sigp_retry(pcpu, SIGP_SET_PREFIX, 0);
242 lowcore_ptr[cpu] = NULL;
243 abs_lowcore_unmap(cpu);
244 nmi_free_mcesa(&lc->mcesad);
245 stack_free(async_stack);
246 stack_free(mcck_stack);
247 free_pages(nodat_stack, THREAD_SIZE_ORDER);
248 free_pages((unsigned long) lc, LC_ORDER);
249}
250
251static void pcpu_prepare_secondary(struct pcpu *pcpu, int cpu)
252{
253 struct lowcore *lc, *abs_lc;
254
255 lc = lowcore_ptr[cpu];
256 cpumask_set_cpu(cpu, &init_mm.context.cpu_attach_mask);
257 cpumask_set_cpu(cpu, mm_cpumask(&init_mm));
258 lc->cpu_nr = cpu;
259 lc->pcpu = (unsigned long)pcpu;
260 lc->restart_flags = RESTART_FLAG_CTLREGS;
261 lc->spinlock_lockval = arch_spin_lockval(cpu);
262 lc->spinlock_index = 0;
263 lc->percpu_offset = __per_cpu_offset[cpu];
264 lc->kernel_asce = get_lowcore()->kernel_asce;
265 lc->user_asce = s390_invalid_asce;
266 lc->machine_flags = get_lowcore()->machine_flags;
267 lc->user_timer = lc->system_timer =
268 lc->steal_timer = lc->avg_steal_timer = 0;
269 abs_lc = get_abs_lowcore();
270 memcpy(lc->cregs_save_area, abs_lc->cregs_save_area, sizeof(lc->cregs_save_area));
271 put_abs_lowcore(abs_lc);
272 lc->cregs_save_area[1] = lc->kernel_asce;
273 lc->cregs_save_area[7] = lc->user_asce;
274 save_access_regs((unsigned int *) lc->access_regs_save_area);
275 arch_spin_lock_setup(cpu);
276}
277
278static void pcpu_attach_task(int cpu, struct task_struct *tsk)
279{
280 struct lowcore *lc;
281
282 lc = lowcore_ptr[cpu];
283 lc->kernel_stack = (unsigned long)task_stack_page(tsk) + STACK_INIT_OFFSET;
284 lc->current_task = (unsigned long)tsk;
285 lc->lpp = LPP_MAGIC;
286 lc->current_pid = tsk->pid;
287 lc->user_timer = tsk->thread.user_timer;
288 lc->guest_timer = tsk->thread.guest_timer;
289 lc->system_timer = tsk->thread.system_timer;
290 lc->hardirq_timer = tsk->thread.hardirq_timer;
291 lc->softirq_timer = tsk->thread.softirq_timer;
292 lc->steal_timer = 0;
293}
294
295static void pcpu_start_fn(int cpu, void (*func)(void *), void *data)
296{
297 struct lowcore *lc;
298
299 lc = lowcore_ptr[cpu];
300 lc->restart_stack = lc->kernel_stack;
301 lc->restart_fn = (unsigned long) func;
302 lc->restart_data = (unsigned long) data;
303 lc->restart_source = -1U;
304 pcpu_sigp_retry(per_cpu_ptr(&pcpu_devices, cpu), SIGP_RESTART, 0);
305}
306
307typedef void (pcpu_delegate_fn)(void *);
308
309/*
310 * Call function via PSW restart on pcpu and stop the current cpu.
311 */
312static void __pcpu_delegate(pcpu_delegate_fn *func, void *data)
313{
314 func(data); /* should not return */
315}
316
317static void pcpu_delegate(struct pcpu *pcpu, int cpu,
318 pcpu_delegate_fn *func,
319 void *data, unsigned long stack)
320{
321 struct lowcore *lc, *abs_lc;
322 unsigned int source_cpu;
323
324 lc = lowcore_ptr[cpu];
325 source_cpu = stap();
326
327 if (pcpu->address == source_cpu) {
328 call_on_stack(2, stack, void, __pcpu_delegate,
329 pcpu_delegate_fn *, func, void *, data);
330 }
331 /* Stop target cpu (if func returns this stops the current cpu). */
332 pcpu_sigp_retry(pcpu, SIGP_STOP, 0);
333 pcpu_sigp_retry(pcpu, SIGP_CPU_RESET, 0);
334 /* Restart func on the target cpu and stop the current cpu. */
335 if (lc) {
336 lc->restart_stack = stack;
337 lc->restart_fn = (unsigned long)func;
338 lc->restart_data = (unsigned long)data;
339 lc->restart_source = source_cpu;
340 } else {
341 abs_lc = get_abs_lowcore();
342 abs_lc->restart_stack = stack;
343 abs_lc->restart_fn = (unsigned long)func;
344 abs_lc->restart_data = (unsigned long)data;
345 abs_lc->restart_source = source_cpu;
346 put_abs_lowcore(abs_lc);
347 }
348 asm volatile(
349 "0: sigp 0,%0,%2 # sigp restart to target cpu\n"
350 " brc 2,0b # busy, try again\n"
351 "1: sigp 0,%1,%3 # sigp stop to current cpu\n"
352 " brc 2,1b # busy, try again\n"
353 : : "d" (pcpu->address), "d" (source_cpu),
354 "K" (SIGP_RESTART), "K" (SIGP_STOP)
355 : "0", "1", "cc");
356 for (;;) ;
357}
358
359/*
360 * Enable additional logical cpus for multi-threading.
361 */
362static int pcpu_set_smt(unsigned int mtid)
363{
364 int cc;
365
366 if (smp_cpu_mtid == mtid)
367 return 0;
368 cc = __pcpu_sigp(0, SIGP_SET_MULTI_THREADING, mtid, NULL);
369 if (cc == 0) {
370 smp_cpu_mtid = mtid;
371 smp_cpu_mt_shift = 0;
372 while (smp_cpu_mtid >= (1U << smp_cpu_mt_shift))
373 smp_cpu_mt_shift++;
374 per_cpu(pcpu_devices, 0).address = stap();
375 }
376 return cc;
377}
378
379/*
380 * Call function on the ipl CPU.
381 */
382void smp_call_ipl_cpu(void (*func)(void *), void *data)
383{
384 struct lowcore *lc = lowcore_ptr[0];
385
386 if (ipl_pcpu->address == stap())
387 lc = get_lowcore();
388
389 pcpu_delegate(ipl_pcpu, 0, func, data, lc->nodat_stack);
390}
391
392int smp_find_processor_id(u16 address)
393{
394 int cpu;
395
396 for_each_present_cpu(cpu)
397 if (per_cpu(pcpu_devices, cpu).address == address)
398 return cpu;
399 return -1;
400}
401
402void schedule_mcck_handler(void)
403{
404 pcpu_ec_call(this_cpu_ptr(&pcpu_devices), ec_mcck_pending);
405}
406
407bool notrace arch_vcpu_is_preempted(int cpu)
408{
409 if (test_cpu_flag_of(CIF_ENABLED_WAIT, cpu))
410 return false;
411 if (pcpu_running(per_cpu_ptr(&pcpu_devices, cpu)))
412 return false;
413 return true;
414}
415EXPORT_SYMBOL(arch_vcpu_is_preempted);
416
417void notrace smp_yield_cpu(int cpu)
418{
419 if (!MACHINE_HAS_DIAG9C)
420 return;
421 diag_stat_inc_norecursion(DIAG_STAT_X09C);
422 asm volatile("diag %0,0,0x9c"
423 : : "d" (per_cpu(pcpu_devices, cpu).address));
424}
425EXPORT_SYMBOL_GPL(smp_yield_cpu);
426
427/*
428 * Send cpus emergency shutdown signal. This gives the cpus the
429 * opportunity to complete outstanding interrupts.
430 */
431void notrace smp_emergency_stop(void)
432{
433 static arch_spinlock_t lock = __ARCH_SPIN_LOCK_UNLOCKED;
434 static cpumask_t cpumask;
435 u64 end;
436 int cpu;
437
438 arch_spin_lock(&lock);
439 cpumask_copy(&cpumask, cpu_online_mask);
440 cpumask_clear_cpu(smp_processor_id(), &cpumask);
441
442 end = get_tod_clock() + (1000000UL << 12);
443 for_each_cpu(cpu, &cpumask) {
444 struct pcpu *pcpu = per_cpu_ptr(&pcpu_devices, cpu);
445 set_bit(ec_stop_cpu, &pcpu->ec_mask);
446 while (__pcpu_sigp(pcpu->address, SIGP_EMERGENCY_SIGNAL,
447 0, NULL) == SIGP_CC_BUSY &&
448 get_tod_clock() < end)
449 cpu_relax();
450 }
451 while (get_tod_clock() < end) {
452 for_each_cpu(cpu, &cpumask)
453 if (pcpu_stopped(per_cpu_ptr(&pcpu_devices, cpu)))
454 cpumask_clear_cpu(cpu, &cpumask);
455 if (cpumask_empty(&cpumask))
456 break;
457 cpu_relax();
458 }
459 arch_spin_unlock(&lock);
460}
461NOKPROBE_SYMBOL(smp_emergency_stop);
462
463/*
464 * Stop all cpus but the current one.
465 */
466void smp_send_stop(void)
467{
468 struct pcpu *pcpu;
469 int cpu;
470
471 /* Disable all interrupts/machine checks */
472 __load_psw_mask(PSW_KERNEL_BITS);
473 trace_hardirqs_off();
474
475 debug_set_critical();
476
477 if (oops_in_progress)
478 smp_emergency_stop();
479
480 /* stop all processors */
481 for_each_online_cpu(cpu) {
482 if (cpu == smp_processor_id())
483 continue;
484 pcpu = per_cpu_ptr(&pcpu_devices, cpu);
485 pcpu_sigp_retry(pcpu, SIGP_STOP, 0);
486 while (!pcpu_stopped(pcpu))
487 cpu_relax();
488 }
489}
490
491/*
492 * This is the main routine where commands issued by other
493 * cpus are handled.
494 */
495static void smp_handle_ext_call(void)
496{
497 unsigned long bits;
498
499 /* handle bit signal external calls */
500 bits = this_cpu_xchg(pcpu_devices.ec_mask, 0);
501 if (test_bit(ec_stop_cpu, &bits))
502 smp_stop_cpu();
503 if (test_bit(ec_schedule, &bits))
504 scheduler_ipi();
505 if (test_bit(ec_call_function_single, &bits))
506 generic_smp_call_function_single_interrupt();
507 if (test_bit(ec_mcck_pending, &bits))
508 s390_handle_mcck();
509 if (test_bit(ec_irq_work, &bits))
510 irq_work_run();
511}
512
513static void do_ext_call_interrupt(struct ext_code ext_code,
514 unsigned int param32, unsigned long param64)
515{
516 inc_irq_stat(ext_code.code == 0x1202 ? IRQEXT_EXC : IRQEXT_EMS);
517 smp_handle_ext_call();
518}
519
520void arch_send_call_function_ipi_mask(const struct cpumask *mask)
521{
522 int cpu;
523
524 for_each_cpu(cpu, mask)
525 pcpu_ec_call(per_cpu_ptr(&pcpu_devices, cpu), ec_call_function_single);
526}
527
528void arch_send_call_function_single_ipi(int cpu)
529{
530 pcpu_ec_call(per_cpu_ptr(&pcpu_devices, cpu), ec_call_function_single);
531}
532
533/*
534 * this function sends a 'reschedule' IPI to another CPU.
535 * it goes straight through and wastes no time serializing
536 * anything. Worst case is that we lose a reschedule ...
537 */
538void arch_smp_send_reschedule(int cpu)
539{
540 pcpu_ec_call(per_cpu_ptr(&pcpu_devices, cpu), ec_schedule);
541}
542
543#ifdef CONFIG_IRQ_WORK
544void arch_irq_work_raise(void)
545{
546 pcpu_ec_call(this_cpu_ptr(&pcpu_devices), ec_irq_work);
547}
548#endif
549
550#ifdef CONFIG_CRASH_DUMP
551
552int smp_store_status(int cpu)
553{
554 struct lowcore *lc;
555 struct pcpu *pcpu;
556 unsigned long pa;
557
558 pcpu = per_cpu_ptr(&pcpu_devices, cpu);
559 lc = lowcore_ptr[cpu];
560 pa = __pa(&lc->floating_pt_save_area);
561 if (__pcpu_sigp_relax(pcpu->address, SIGP_STORE_STATUS_AT_ADDRESS,
562 pa) != SIGP_CC_ORDER_CODE_ACCEPTED)
563 return -EIO;
564 if (!cpu_has_vx() && !MACHINE_HAS_GS)
565 return 0;
566 pa = lc->mcesad & MCESA_ORIGIN_MASK;
567 if (MACHINE_HAS_GS)
568 pa |= lc->mcesad & MCESA_LC_MASK;
569 if (__pcpu_sigp_relax(pcpu->address, SIGP_STORE_ADDITIONAL_STATUS,
570 pa) != SIGP_CC_ORDER_CODE_ACCEPTED)
571 return -EIO;
572 return 0;
573}
574
575/*
576 * Collect CPU state of the previous, crashed system.
577 * There are three cases:
578 * 1) standard zfcp/nvme dump
579 * condition: OLDMEM_BASE == NULL && is_ipl_type_dump() == true
580 * The state for all CPUs except the boot CPU needs to be collected
581 * with sigp stop-and-store-status. The boot CPU state is located in
582 * the absolute lowcore of the memory stored in the HSA. The zcore code
583 * will copy the boot CPU state from the HSA.
584 * 2) stand-alone kdump for SCSI/NVMe (zfcp/nvme dump with swapped memory)
585 * condition: OLDMEM_BASE != NULL && is_ipl_type_dump() == true
586 * The state for all CPUs except the boot CPU needs to be collected
587 * with sigp stop-and-store-status. The firmware or the boot-loader
588 * stored the registers of the boot CPU in the absolute lowcore in the
589 * memory of the old system.
590 * 3) kdump or stand-alone kdump for DASD
591 * condition: OLDMEM_BASE != NULL && is_ipl_type_dump() == false
592 * The state for all CPUs except the boot CPU needs to be collected
593 * with sigp stop-and-store-status. The kexec code or the boot-loader
594 * stored the registers of the boot CPU in the memory of the old system.
595 *
596 * Note that the legacy kdump mode where the old kernel stored the CPU states
597 * does no longer exist: setup_arch() explicitly deactivates the elfcorehdr=
598 * kernel parameter. The is_kdump_kernel() implementation on s390 is independent
599 * of the elfcorehdr= parameter.
600 */
601static bool dump_available(void)
602{
603 return oldmem_data.start || is_ipl_type_dump();
604}
605
606void __init smp_save_dump_ipl_cpu(void)
607{
608 struct save_area *sa;
609 void *regs;
610
611 if (!dump_available())
612 return;
613 sa = save_area_alloc(true);
614 regs = memblock_alloc(512, 8);
615 if (!sa || !regs)
616 panic("could not allocate memory for boot CPU save area\n");
617 copy_oldmem_kernel(regs, __LC_FPREGS_SAVE_AREA, 512);
618 save_area_add_regs(sa, regs);
619 memblock_free(regs, 512);
620 if (cpu_has_vx())
621 save_area_add_vxrs(sa, boot_cpu_vector_save_area);
622}
623
624void __init smp_save_dump_secondary_cpus(void)
625{
626 int addr, boot_cpu_addr, max_cpu_addr;
627 struct save_area *sa;
628 void *page;
629
630 if (!dump_available())
631 return;
632 /* Allocate a page as dumping area for the store status sigps */
633 page = memblock_alloc_low(PAGE_SIZE, PAGE_SIZE);
634 if (!page)
635 panic("ERROR: Failed to allocate %lx bytes below %lx\n",
636 PAGE_SIZE, 1UL << 31);
637
638 /* Set multi-threading state to the previous system. */
639 pcpu_set_smt(sclp.mtid_prev);
640 boot_cpu_addr = stap();
641 max_cpu_addr = SCLP_MAX_CORES << sclp.mtid_prev;
642 for (addr = 0; addr <= max_cpu_addr; addr++) {
643 if (addr == boot_cpu_addr)
644 continue;
645 if (__pcpu_sigp_relax(addr, SIGP_SENSE, 0) ==
646 SIGP_CC_NOT_OPERATIONAL)
647 continue;
648 sa = save_area_alloc(false);
649 if (!sa)
650 panic("could not allocate memory for save area\n");
651 __pcpu_sigp_relax(addr, SIGP_STORE_STATUS_AT_ADDRESS, __pa(page));
652 save_area_add_regs(sa, page);
653 if (cpu_has_vx()) {
654 __pcpu_sigp_relax(addr, SIGP_STORE_ADDITIONAL_STATUS, __pa(page));
655 save_area_add_vxrs(sa, page);
656 }
657 }
658 memblock_free(page, PAGE_SIZE);
659 diag_amode31_ops.diag308_reset();
660 pcpu_set_smt(0);
661}
662#endif /* CONFIG_CRASH_DUMP */
663
664void smp_cpu_set_polarization(int cpu, int val)
665{
666 per_cpu(pcpu_devices, cpu).polarization = val;
667}
668
669int smp_cpu_get_polarization(int cpu)
670{
671 return per_cpu(pcpu_devices, cpu).polarization;
672}
673
674void smp_cpu_set_capacity(int cpu, unsigned long val)
675{
676 per_cpu(pcpu_devices, cpu).capacity = val;
677}
678
679unsigned long smp_cpu_get_capacity(int cpu)
680{
681 return per_cpu(pcpu_devices, cpu).capacity;
682}
683
684void smp_set_core_capacity(int cpu, unsigned long val)
685{
686 int i;
687
688 cpu = smp_get_base_cpu(cpu);
689 for (i = cpu; (i <= cpu + smp_cpu_mtid) && (i < nr_cpu_ids); i++)
690 smp_cpu_set_capacity(i, val);
691}
692
693int smp_cpu_get_cpu_address(int cpu)
694{
695 return per_cpu(pcpu_devices, cpu).address;
696}
697
698static void __ref smp_get_core_info(struct sclp_core_info *info, int early)
699{
700 static int use_sigp_detection;
701 int address;
702
703 if (use_sigp_detection || sclp_get_core_info(info, early)) {
704 use_sigp_detection = 1;
705 for (address = 0;
706 address < (SCLP_MAX_CORES << smp_cpu_mt_shift);
707 address += (1U << smp_cpu_mt_shift)) {
708 if (__pcpu_sigp_relax(address, SIGP_SENSE, 0) ==
709 SIGP_CC_NOT_OPERATIONAL)
710 continue;
711 info->core[info->configured].core_id =
712 address >> smp_cpu_mt_shift;
713 info->configured++;
714 }
715 info->combined = info->configured;
716 }
717}
718
719static int smp_add_core(struct sclp_core_entry *core, cpumask_t *avail,
720 bool configured, bool early)
721{
722 struct pcpu *pcpu;
723 int cpu, nr, i;
724 u16 address;
725
726 nr = 0;
727 if (sclp.has_core_type && core->type != boot_core_type)
728 return nr;
729 cpu = cpumask_first(avail);
730 address = core->core_id << smp_cpu_mt_shift;
731 for (i = 0; (i <= smp_cpu_mtid) && (cpu < nr_cpu_ids); i++) {
732 if (pcpu_find_address(cpu_present_mask, address + i))
733 continue;
734 pcpu = per_cpu_ptr(&pcpu_devices, cpu);
735 pcpu->address = address + i;
736 if (configured)
737 pcpu->state = CPU_STATE_CONFIGURED;
738 else
739 pcpu->state = CPU_STATE_STANDBY;
740 smp_cpu_set_polarization(cpu, POLARIZATION_UNKNOWN);
741 smp_cpu_set_capacity(cpu, CPU_CAPACITY_HIGH);
742 set_cpu_present(cpu, true);
743 if (!early && arch_register_cpu(cpu))
744 set_cpu_present(cpu, false);
745 else
746 nr++;
747 cpumask_clear_cpu(cpu, avail);
748 cpu = cpumask_next(cpu, avail);
749 }
750 return nr;
751}
752
753static int __smp_rescan_cpus(struct sclp_core_info *info, bool early)
754{
755 struct sclp_core_entry *core;
756 static cpumask_t avail;
757 bool configured;
758 u16 core_id;
759 int nr, i;
760
761 cpus_read_lock();
762 mutex_lock(&smp_cpu_state_mutex);
763 nr = 0;
764 cpumask_xor(&avail, cpu_possible_mask, cpu_present_mask);
765 /*
766 * Add IPL core first (which got logical CPU number 0) to make sure
767 * that all SMT threads get subsequent logical CPU numbers.
768 */
769 if (early) {
770 core_id = per_cpu(pcpu_devices, 0).address >> smp_cpu_mt_shift;
771 for (i = 0; i < info->configured; i++) {
772 core = &info->core[i];
773 if (core->core_id == core_id) {
774 nr += smp_add_core(core, &avail, true, early);
775 break;
776 }
777 }
778 }
779 for (i = 0; i < info->combined; i++) {
780 configured = i < info->configured;
781 nr += smp_add_core(&info->core[i], &avail, configured, early);
782 }
783 mutex_unlock(&smp_cpu_state_mutex);
784 cpus_read_unlock();
785 return nr;
786}
787
788void __init smp_detect_cpus(void)
789{
790 unsigned int cpu, mtid, c_cpus, s_cpus;
791 struct sclp_core_info *info;
792 u16 address;
793
794 /* Get CPU information */
795 info = memblock_alloc(sizeof(*info), 8);
796 if (!info)
797 panic("%s: Failed to allocate %zu bytes align=0x%x\n",
798 __func__, sizeof(*info), 8);
799 smp_get_core_info(info, 1);
800 /* Find boot CPU type */
801 if (sclp.has_core_type) {
802 address = stap();
803 for (cpu = 0; cpu < info->combined; cpu++)
804 if (info->core[cpu].core_id == address) {
805 /* The boot cpu dictates the cpu type. */
806 boot_core_type = info->core[cpu].type;
807 break;
808 }
809 if (cpu >= info->combined)
810 panic("Could not find boot CPU type");
811 }
812
813 /* Set multi-threading state for the current system */
814 mtid = boot_core_type ? sclp.mtid : sclp.mtid_cp;
815 mtid = (mtid < smp_max_threads) ? mtid : smp_max_threads - 1;
816 pcpu_set_smt(mtid);
817
818 /* Print number of CPUs */
819 c_cpus = s_cpus = 0;
820 for (cpu = 0; cpu < info->combined; cpu++) {
821 if (sclp.has_core_type &&
822 info->core[cpu].type != boot_core_type)
823 continue;
824 if (cpu < info->configured)
825 c_cpus += smp_cpu_mtid + 1;
826 else
827 s_cpus += smp_cpu_mtid + 1;
828 }
829 pr_info("%d configured CPUs, %d standby CPUs\n", c_cpus, s_cpus);
830 memblock_free(info, sizeof(*info));
831}
832
833/*
834 * Activate a secondary processor.
835 */
836static void smp_start_secondary(void *cpuvoid)
837{
838 struct lowcore *lc = get_lowcore();
839 int cpu = raw_smp_processor_id();
840
841 lc->last_update_clock = get_tod_clock();
842 lc->restart_stack = (unsigned long)restart_stack;
843 lc->restart_fn = (unsigned long)do_restart;
844 lc->restart_data = 0;
845 lc->restart_source = -1U;
846 lc->restart_flags = 0;
847 restore_access_regs(lc->access_regs_save_area);
848 cpu_init();
849 rcutree_report_cpu_starting(cpu);
850 init_cpu_timer();
851 vtime_init();
852 vdso_getcpu_init();
853 pfault_init();
854 cpumask_set_cpu(cpu, &cpu_setup_mask);
855 update_cpu_masks();
856 notify_cpu_starting(cpu);
857 if (topology_cpu_dedicated(cpu))
858 set_cpu_flag(CIF_DEDICATED_CPU);
859 else
860 clear_cpu_flag(CIF_DEDICATED_CPU);
861 set_cpu_online(cpu, true);
862 inc_irq_stat(CPU_RST);
863 local_irq_enable();
864 cpu_startup_entry(CPUHP_AP_ONLINE_IDLE);
865}
866
867/* Upping and downing of CPUs */
868int __cpu_up(unsigned int cpu, struct task_struct *tidle)
869{
870 struct pcpu *pcpu = per_cpu_ptr(&pcpu_devices, cpu);
871 int rc;
872
873 if (pcpu->state != CPU_STATE_CONFIGURED)
874 return -EIO;
875 if (pcpu_sigp_retry(pcpu, SIGP_INITIAL_CPU_RESET, 0) !=
876 SIGP_CC_ORDER_CODE_ACCEPTED)
877 return -EIO;
878
879 rc = pcpu_alloc_lowcore(pcpu, cpu);
880 if (rc)
881 return rc;
882 /*
883 * Make sure global control register contents do not change
884 * until new CPU has initialized control registers.
885 */
886 system_ctlreg_lock();
887 pcpu_prepare_secondary(pcpu, cpu);
888 pcpu_attach_task(cpu, tidle);
889 pcpu_start_fn(cpu, smp_start_secondary, NULL);
890 /* Wait until cpu puts itself in the online & active maps */
891 while (!cpu_online(cpu))
892 cpu_relax();
893 system_ctlreg_unlock();
894 return 0;
895}
896
897static unsigned int setup_possible_cpus __initdata;
898
899static int __init _setup_possible_cpus(char *s)
900{
901 get_option(&s, &setup_possible_cpus);
902 return 0;
903}
904early_param("possible_cpus", _setup_possible_cpus);
905
906int __cpu_disable(void)
907{
908 struct ctlreg cregs[16];
909 int cpu;
910
911 /* Handle possible pending IPIs */
912 smp_handle_ext_call();
913 cpu = smp_processor_id();
914 set_cpu_online(cpu, false);
915 cpumask_clear_cpu(cpu, &cpu_setup_mask);
916 update_cpu_masks();
917 /* Disable pseudo page faults on this cpu. */
918 pfault_fini();
919 /* Disable interrupt sources via control register. */
920 __local_ctl_store(0, 15, cregs);
921 cregs[0].val &= ~0x0000ee70UL; /* disable all external interrupts */
922 cregs[6].val &= ~0xff000000UL; /* disable all I/O interrupts */
923 cregs[14].val &= ~0x1f000000UL; /* disable most machine checks */
924 __local_ctl_load(0, 15, cregs);
925 clear_cpu_flag(CIF_NOHZ_DELAY);
926 return 0;
927}
928
929void __cpu_die(unsigned int cpu)
930{
931 struct pcpu *pcpu;
932
933 /* Wait until target cpu is down */
934 pcpu = per_cpu_ptr(&pcpu_devices, cpu);
935 while (!pcpu_stopped(pcpu))
936 cpu_relax();
937 pcpu_free_lowcore(pcpu, cpu);
938 cpumask_clear_cpu(cpu, mm_cpumask(&init_mm));
939 cpumask_clear_cpu(cpu, &init_mm.context.cpu_attach_mask);
940 pcpu->flags = 0;
941}
942
943void __noreturn cpu_die(void)
944{
945 idle_task_exit();
946 pcpu_sigp_retry(this_cpu_ptr(&pcpu_devices), SIGP_STOP, 0);
947 for (;;) ;
948}
949
950void __init smp_fill_possible_mask(void)
951{
952 unsigned int possible, sclp_max, cpu;
953
954 sclp_max = max(sclp.mtid, sclp.mtid_cp) + 1;
955 sclp_max = min(smp_max_threads, sclp_max);
956 sclp_max = (sclp.max_cores * sclp_max) ?: nr_cpu_ids;
957 possible = setup_possible_cpus ?: nr_cpu_ids;
958 possible = min(possible, sclp_max);
959 for (cpu = 0; cpu < possible && cpu < nr_cpu_ids; cpu++)
960 set_cpu_possible(cpu, true);
961}
962
963void __init smp_prepare_cpus(unsigned int max_cpus)
964{
965 if (register_external_irq(EXT_IRQ_EMERGENCY_SIG, do_ext_call_interrupt))
966 panic("Couldn't request external interrupt 0x1201");
967 system_ctl_set_bit(0, 14);
968 if (register_external_irq(EXT_IRQ_EXTERNAL_CALL, do_ext_call_interrupt))
969 panic("Couldn't request external interrupt 0x1202");
970 system_ctl_set_bit(0, 13);
971 smp_rescan_cpus(true);
972}
973
974void __init smp_prepare_boot_cpu(void)
975{
976 struct lowcore *lc = get_lowcore();
977
978 WARN_ON(!cpu_present(0) || !cpu_online(0));
979 lc->percpu_offset = __per_cpu_offset[0];
980 ipl_pcpu = per_cpu_ptr(&pcpu_devices, 0);
981 ipl_pcpu->state = CPU_STATE_CONFIGURED;
982 lc->pcpu = (unsigned long)ipl_pcpu;
983 smp_cpu_set_polarization(0, POLARIZATION_UNKNOWN);
984 smp_cpu_set_capacity(0, CPU_CAPACITY_HIGH);
985}
986
987void __init smp_setup_processor_id(void)
988{
989 struct lowcore *lc = get_lowcore();
990
991 lc->cpu_nr = 0;
992 per_cpu(pcpu_devices, 0).address = stap();
993 lc->spinlock_lockval = arch_spin_lockval(0);
994 lc->spinlock_index = 0;
995}
996
997/*
998 * the frequency of the profiling timer can be changed
999 * by writing a multiplier value into /proc/profile.
1000 *
1001 * usually you want to run this on all CPUs ;)
1002 */
1003int setup_profiling_timer(unsigned int multiplier)
1004{
1005 return 0;
1006}
1007
1008static ssize_t cpu_configure_show(struct device *dev,
1009 struct device_attribute *attr, char *buf)
1010{
1011 ssize_t count;
1012
1013 mutex_lock(&smp_cpu_state_mutex);
1014 count = sysfs_emit(buf, "%d\n", per_cpu(pcpu_devices, dev->id).state);
1015 mutex_unlock(&smp_cpu_state_mutex);
1016 return count;
1017}
1018
1019static ssize_t cpu_configure_store(struct device *dev,
1020 struct device_attribute *attr,
1021 const char *buf, size_t count)
1022{
1023 struct pcpu *pcpu;
1024 int cpu, val, rc, i;
1025 char delim;
1026
1027 if (sscanf(buf, "%d %c", &val, &delim) != 1)
1028 return -EINVAL;
1029 if (val != 0 && val != 1)
1030 return -EINVAL;
1031 cpus_read_lock();
1032 mutex_lock(&smp_cpu_state_mutex);
1033 rc = -EBUSY;
1034 /* disallow configuration changes of online cpus */
1035 cpu = dev->id;
1036 cpu = smp_get_base_cpu(cpu);
1037 for (i = 0; i <= smp_cpu_mtid; i++)
1038 if (cpu_online(cpu + i))
1039 goto out;
1040 pcpu = per_cpu_ptr(&pcpu_devices, cpu);
1041 rc = 0;
1042 switch (val) {
1043 case 0:
1044 if (pcpu->state != CPU_STATE_CONFIGURED)
1045 break;
1046 rc = sclp_core_deconfigure(pcpu->address >> smp_cpu_mt_shift);
1047 if (rc)
1048 break;
1049 for (i = 0; i <= smp_cpu_mtid; i++) {
1050 if (cpu + i >= nr_cpu_ids || !cpu_present(cpu + i))
1051 continue;
1052 per_cpu(pcpu_devices, cpu + i).state = CPU_STATE_STANDBY;
1053 smp_cpu_set_polarization(cpu + i,
1054 POLARIZATION_UNKNOWN);
1055 }
1056 topology_expect_change();
1057 break;
1058 case 1:
1059 if (pcpu->state != CPU_STATE_STANDBY)
1060 break;
1061 rc = sclp_core_configure(pcpu->address >> smp_cpu_mt_shift);
1062 if (rc)
1063 break;
1064 for (i = 0; i <= smp_cpu_mtid; i++) {
1065 if (cpu + i >= nr_cpu_ids || !cpu_present(cpu + i))
1066 continue;
1067 per_cpu(pcpu_devices, cpu + i).state = CPU_STATE_CONFIGURED;
1068 smp_cpu_set_polarization(cpu + i,
1069 POLARIZATION_UNKNOWN);
1070 }
1071 topology_expect_change();
1072 break;
1073 default:
1074 break;
1075 }
1076out:
1077 mutex_unlock(&smp_cpu_state_mutex);
1078 cpus_read_unlock();
1079 return rc ? rc : count;
1080}
1081static DEVICE_ATTR(configure, 0644, cpu_configure_show, cpu_configure_store);
1082
1083static ssize_t show_cpu_address(struct device *dev,
1084 struct device_attribute *attr, char *buf)
1085{
1086 return sysfs_emit(buf, "%d\n", per_cpu(pcpu_devices, dev->id).address);
1087}
1088static DEVICE_ATTR(address, 0444, show_cpu_address, NULL);
1089
1090static struct attribute *cpu_common_attrs[] = {
1091 &dev_attr_configure.attr,
1092 &dev_attr_address.attr,
1093 NULL,
1094};
1095
1096static struct attribute_group cpu_common_attr_group = {
1097 .attrs = cpu_common_attrs,
1098};
1099
1100static struct attribute *cpu_online_attrs[] = {
1101 &dev_attr_idle_count.attr,
1102 &dev_attr_idle_time_us.attr,
1103 NULL,
1104};
1105
1106static struct attribute_group cpu_online_attr_group = {
1107 .attrs = cpu_online_attrs,
1108};
1109
1110static int smp_cpu_online(unsigned int cpu)
1111{
1112 struct cpu *c = per_cpu_ptr(&cpu_devices, cpu);
1113
1114 return sysfs_create_group(&c->dev.kobj, &cpu_online_attr_group);
1115}
1116
1117static int smp_cpu_pre_down(unsigned int cpu)
1118{
1119 struct cpu *c = per_cpu_ptr(&cpu_devices, cpu);
1120
1121 sysfs_remove_group(&c->dev.kobj, &cpu_online_attr_group);
1122 return 0;
1123}
1124
1125bool arch_cpu_is_hotpluggable(int cpu)
1126{
1127 return !!cpu;
1128}
1129
1130int arch_register_cpu(int cpu)
1131{
1132 struct cpu *c = per_cpu_ptr(&cpu_devices, cpu);
1133 int rc;
1134
1135 c->hotpluggable = arch_cpu_is_hotpluggable(cpu);
1136 rc = register_cpu(c, cpu);
1137 if (rc)
1138 goto out;
1139 rc = sysfs_create_group(&c->dev.kobj, &cpu_common_attr_group);
1140 if (rc)
1141 goto out_cpu;
1142 rc = topology_cpu_init(c);
1143 if (rc)
1144 goto out_topology;
1145 return 0;
1146
1147out_topology:
1148 sysfs_remove_group(&c->dev.kobj, &cpu_common_attr_group);
1149out_cpu:
1150 unregister_cpu(c);
1151out:
1152 return rc;
1153}
1154
1155int __ref smp_rescan_cpus(bool early)
1156{
1157 struct sclp_core_info *info;
1158 int nr;
1159
1160 info = kzalloc(sizeof(*info), GFP_KERNEL);
1161 if (!info)
1162 return -ENOMEM;
1163 smp_get_core_info(info, 0);
1164 nr = __smp_rescan_cpus(info, early);
1165 kfree(info);
1166 if (nr)
1167 topology_schedule_update();
1168 return 0;
1169}
1170
1171static ssize_t __ref rescan_store(struct device *dev,
1172 struct device_attribute *attr,
1173 const char *buf,
1174 size_t count)
1175{
1176 int rc;
1177
1178 rc = lock_device_hotplug_sysfs();
1179 if (rc)
1180 return rc;
1181 rc = smp_rescan_cpus(false);
1182 unlock_device_hotplug();
1183 return rc ? rc : count;
1184}
1185static DEVICE_ATTR_WO(rescan);
1186
1187static int __init s390_smp_init(void)
1188{
1189 struct device *dev_root;
1190 int rc;
1191
1192 dev_root = bus_get_dev_root(&cpu_subsys);
1193 if (dev_root) {
1194 rc = device_create_file(dev_root, &dev_attr_rescan);
1195 put_device(dev_root);
1196 if (rc)
1197 return rc;
1198 }
1199 rc = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "s390/smp:online",
1200 smp_cpu_online, smp_cpu_pre_down);
1201 rc = rc <= 0 ? rc : 0;
1202 return rc;
1203}
1204subsys_initcall(s390_smp_init);
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 <schwidefsky@de.ibm.com>,
8 * Heiko Carstens <heiko.carstens@de.ibm.com>,
9 *
10 * based on other smp stuff by
11 * (c) 1995 Alan Cox, CymruNET Ltd <alan@cymru.net>
12 * (c) 1998 Ingo Molnar
13 *
14 * The code outside of smp.c uses logical cpu numbers, only smp.c does
15 * the translation of logical to physical cpu ids. All new code that
16 * operates on physical cpu numbers needs to go into smp.c.
17 */
18
19#define KMSG_COMPONENT "cpu"
20#define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
21
22#include <linux/workqueue.h>
23#include <linux/memblock.h>
24#include <linux/export.h>
25#include <linux/init.h>
26#include <linux/mm.h>
27#include <linux/err.h>
28#include <linux/spinlock.h>
29#include <linux/kernel_stat.h>
30#include <linux/delay.h>
31#include <linux/interrupt.h>
32#include <linux/irqflags.h>
33#include <linux/irq_work.h>
34#include <linux/cpu.h>
35#include <linux/slab.h>
36#include <linux/sched/hotplug.h>
37#include <linux/sched/task_stack.h>
38#include <linux/crash_dump.h>
39#include <linux/kprobes.h>
40#include <asm/asm-offsets.h>
41#include <asm/diag.h>
42#include <asm/switch_to.h>
43#include <asm/facility.h>
44#include <asm/ipl.h>
45#include <asm/setup.h>
46#include <asm/irq.h>
47#include <asm/tlbflush.h>
48#include <asm/vtimer.h>
49#include <asm/lowcore.h>
50#include <asm/sclp.h>
51#include <asm/debug.h>
52#include <asm/os_info.h>
53#include <asm/sigp.h>
54#include <asm/idle.h>
55#include <asm/nmi.h>
56#include <asm/stacktrace.h>
57#include <asm/topology.h>
58#include <asm/vdso.h>
59#include "entry.h"
60
61enum {
62 ec_schedule = 0,
63 ec_call_function_single,
64 ec_stop_cpu,
65 ec_mcck_pending,
66 ec_irq_work,
67};
68
69enum {
70 CPU_STATE_STANDBY,
71 CPU_STATE_CONFIGURED,
72};
73
74static DEFINE_PER_CPU(struct cpu *, cpu_device);
75
76struct pcpu {
77 unsigned long ec_mask; /* bit mask for ec_xxx functions */
78 unsigned long ec_clk; /* sigp timestamp for ec_xxx */
79 signed char state; /* physical cpu state */
80 signed char polarization; /* physical polarization */
81 u16 address; /* physical cpu address */
82};
83
84static u8 boot_core_type;
85static struct pcpu pcpu_devices[NR_CPUS];
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_nosmt(char *s)
101{
102 smp_max_threads = 1;
103 return 0;
104}
105early_param("nosmt", early_nosmt);
106
107static int __init early_smt(char *s)
108{
109 get_option(&s, &smp_max_threads);
110 return 0;
111}
112early_param("smt", early_smt);
113
114/*
115 * The smp_cpu_state_mutex must be held when changing the state or polarization
116 * member of a pcpu data structure within the pcpu_devices arreay.
117 */
118DEFINE_MUTEX(smp_cpu_state_mutex);
119
120/*
121 * Signal processor helper functions.
122 */
123static inline int __pcpu_sigp_relax(u16 addr, u8 order, unsigned long parm)
124{
125 int cc;
126
127 while (1) {
128 cc = __pcpu_sigp(addr, order, parm, NULL);
129 if (cc != SIGP_CC_BUSY)
130 return cc;
131 cpu_relax();
132 }
133}
134
135static int pcpu_sigp_retry(struct pcpu *pcpu, u8 order, u32 parm)
136{
137 int cc, retry;
138
139 for (retry = 0; ; retry++) {
140 cc = __pcpu_sigp(pcpu->address, order, parm, NULL);
141 if (cc != SIGP_CC_BUSY)
142 break;
143 if (retry >= 3)
144 udelay(10);
145 }
146 return cc;
147}
148
149static inline int pcpu_stopped(struct pcpu *pcpu)
150{
151 u32 status;
152
153 if (__pcpu_sigp(pcpu->address, SIGP_SENSE,
154 0, &status) != SIGP_CC_STATUS_STORED)
155 return 0;
156 return !!(status & (SIGP_STATUS_CHECK_STOP|SIGP_STATUS_STOPPED));
157}
158
159static inline int pcpu_running(struct pcpu *pcpu)
160{
161 if (__pcpu_sigp(pcpu->address, SIGP_SENSE_RUNNING,
162 0, NULL) != SIGP_CC_STATUS_STORED)
163 return 1;
164 /* Status stored condition code is equivalent to cpu not running. */
165 return 0;
166}
167
168/*
169 * Find struct pcpu by cpu address.
170 */
171static struct pcpu *pcpu_find_address(const struct cpumask *mask, u16 address)
172{
173 int cpu;
174
175 for_each_cpu(cpu, mask)
176 if (pcpu_devices[cpu].address == address)
177 return pcpu_devices + cpu;
178 return NULL;
179}
180
181static void pcpu_ec_call(struct pcpu *pcpu, int ec_bit)
182{
183 int order;
184
185 if (test_and_set_bit(ec_bit, &pcpu->ec_mask))
186 return;
187 order = pcpu_running(pcpu) ? SIGP_EXTERNAL_CALL : SIGP_EMERGENCY_SIGNAL;
188 pcpu->ec_clk = get_tod_clock_fast();
189 pcpu_sigp_retry(pcpu, order, 0);
190}
191
192static int pcpu_alloc_lowcore(struct pcpu *pcpu, int cpu)
193{
194 unsigned long async_stack, nodat_stack, mcck_stack;
195 struct lowcore *lc;
196
197 lc = (struct lowcore *) __get_free_pages(GFP_KERNEL | GFP_DMA, LC_ORDER);
198 nodat_stack = __get_free_pages(GFP_KERNEL, THREAD_SIZE_ORDER);
199 async_stack = stack_alloc();
200 mcck_stack = stack_alloc();
201 if (!lc || !nodat_stack || !async_stack || !mcck_stack)
202 goto out;
203 memcpy(lc, &S390_lowcore, 512);
204 memset((char *) lc + 512, 0, sizeof(*lc) - 512);
205 lc->async_stack = async_stack + STACK_INIT_OFFSET;
206 lc->nodat_stack = nodat_stack + STACK_INIT_OFFSET;
207 lc->mcck_stack = mcck_stack + STACK_INIT_OFFSET;
208 lc->cpu_nr = cpu;
209 lc->spinlock_lockval = arch_spin_lockval(cpu);
210 lc->spinlock_index = 0;
211 lc->br_r1_trampoline = 0x07f1; /* br %r1 */
212 lc->return_lpswe = gen_lpswe(__LC_RETURN_PSW);
213 lc->return_mcck_lpswe = gen_lpswe(__LC_RETURN_MCCK_PSW);
214 lc->preempt_count = PREEMPT_DISABLED;
215 if (nmi_alloc_per_cpu(lc))
216 goto out;
217 lowcore_ptr[cpu] = lc;
218 pcpu_sigp_retry(pcpu, SIGP_SET_PREFIX, (u32)(unsigned long) lc);
219 return 0;
220
221out:
222 stack_free(mcck_stack);
223 stack_free(async_stack);
224 free_pages(nodat_stack, THREAD_SIZE_ORDER);
225 free_pages((unsigned long) lc, LC_ORDER);
226 return -ENOMEM;
227}
228
229static void pcpu_free_lowcore(struct pcpu *pcpu)
230{
231 unsigned long async_stack, nodat_stack, mcck_stack;
232 struct lowcore *lc;
233 int cpu;
234
235 cpu = pcpu - pcpu_devices;
236 lc = lowcore_ptr[cpu];
237 nodat_stack = lc->nodat_stack - STACK_INIT_OFFSET;
238 async_stack = lc->async_stack - STACK_INIT_OFFSET;
239 mcck_stack = lc->mcck_stack - STACK_INIT_OFFSET;
240 pcpu_sigp_retry(pcpu, SIGP_SET_PREFIX, 0);
241 lowcore_ptr[cpu] = NULL;
242 nmi_free_per_cpu(lc);
243 stack_free(async_stack);
244 stack_free(mcck_stack);
245 free_pages(nodat_stack, THREAD_SIZE_ORDER);
246 free_pages((unsigned long) lc, LC_ORDER);
247}
248
249static void pcpu_prepare_secondary(struct pcpu *pcpu, int cpu)
250{
251 struct lowcore *lc = lowcore_ptr[cpu];
252
253 cpumask_set_cpu(cpu, &init_mm.context.cpu_attach_mask);
254 cpumask_set_cpu(cpu, mm_cpumask(&init_mm));
255 lc->cpu_nr = cpu;
256 lc->restart_flags = RESTART_FLAG_CTLREGS;
257 lc->spinlock_lockval = arch_spin_lockval(cpu);
258 lc->spinlock_index = 0;
259 lc->percpu_offset = __per_cpu_offset[cpu];
260 lc->kernel_asce = S390_lowcore.kernel_asce;
261 lc->user_asce = s390_invalid_asce;
262 lc->machine_flags = S390_lowcore.machine_flags;
263 lc->user_timer = lc->system_timer =
264 lc->steal_timer = lc->avg_steal_timer = 0;
265 __ctl_store(lc->cregs_save_area, 0, 15);
266 lc->cregs_save_area[1] = lc->kernel_asce;
267 lc->cregs_save_area[7] = lc->user_asce;
268 save_access_regs((unsigned int *) lc->access_regs_save_area);
269 arch_spin_lock_setup(cpu);
270}
271
272static void pcpu_attach_task(struct pcpu *pcpu, struct task_struct *tsk)
273{
274 struct lowcore *lc;
275 int cpu;
276
277 cpu = pcpu - pcpu_devices;
278 lc = lowcore_ptr[cpu];
279 lc->kernel_stack = (unsigned long) task_stack_page(tsk)
280 + THREAD_SIZE - STACK_FRAME_OVERHEAD - sizeof(struct pt_regs);
281 lc->current_task = (unsigned long) tsk;
282 lc->lpp = LPP_MAGIC;
283 lc->current_pid = tsk->pid;
284 lc->user_timer = tsk->thread.user_timer;
285 lc->guest_timer = tsk->thread.guest_timer;
286 lc->system_timer = tsk->thread.system_timer;
287 lc->hardirq_timer = tsk->thread.hardirq_timer;
288 lc->softirq_timer = tsk->thread.softirq_timer;
289 lc->steal_timer = 0;
290}
291
292static void pcpu_start_fn(struct pcpu *pcpu, void (*func)(void *), void *data)
293{
294 struct lowcore *lc;
295 int cpu;
296
297 cpu = pcpu - pcpu_devices;
298 lc = lowcore_ptr[cpu];
299 lc->restart_stack = lc->nodat_stack;
300 lc->restart_fn = (unsigned long) func;
301 lc->restart_data = (unsigned long) data;
302 lc->restart_source = -1U;
303 pcpu_sigp_retry(pcpu, SIGP_RESTART, 0);
304}
305
306typedef void (pcpu_delegate_fn)(void *);
307
308/*
309 * Call function via PSW restart on pcpu and stop the current cpu.
310 */
311static void __pcpu_delegate(pcpu_delegate_fn *func, void *data)
312{
313 func(data); /* should not return */
314}
315
316static void pcpu_delegate(struct pcpu *pcpu,
317 pcpu_delegate_fn *func,
318 void *data, unsigned long stack)
319{
320 struct lowcore *lc = lowcore_ptr[pcpu - pcpu_devices];
321 unsigned int source_cpu = stap();
322
323 __load_psw_mask(PSW_KERNEL_BITS | PSW_MASK_DAT);
324 if (pcpu->address == source_cpu) {
325 call_on_stack(2, stack, void, __pcpu_delegate,
326 pcpu_delegate_fn *, func, void *, data);
327 }
328 /* Stop target cpu (if func returns this stops the current cpu). */
329 pcpu_sigp_retry(pcpu, SIGP_STOP, 0);
330 /* Restart func on the target cpu and stop the current cpu. */
331 mem_assign_absolute(lc->restart_stack, stack);
332 mem_assign_absolute(lc->restart_fn, (unsigned long) func);
333 mem_assign_absolute(lc->restart_data, (unsigned long) data);
334 mem_assign_absolute(lc->restart_source, source_cpu);
335 __bpon();
336 asm volatile(
337 "0: sigp 0,%0,%2 # sigp restart to target cpu\n"
338 " brc 2,0b # busy, try again\n"
339 "1: sigp 0,%1,%3 # sigp stop to current cpu\n"
340 " brc 2,1b # busy, try again\n"
341 : : "d" (pcpu->address), "d" (source_cpu),
342 "K" (SIGP_RESTART), "K" (SIGP_STOP)
343 : "0", "1", "cc");
344 for (;;) ;
345}
346
347/*
348 * Enable additional logical cpus for multi-threading.
349 */
350static int pcpu_set_smt(unsigned int mtid)
351{
352 int cc;
353
354 if (smp_cpu_mtid == mtid)
355 return 0;
356 cc = __pcpu_sigp(0, SIGP_SET_MULTI_THREADING, mtid, NULL);
357 if (cc == 0) {
358 smp_cpu_mtid = mtid;
359 smp_cpu_mt_shift = 0;
360 while (smp_cpu_mtid >= (1U << smp_cpu_mt_shift))
361 smp_cpu_mt_shift++;
362 pcpu_devices[0].address = stap();
363 }
364 return cc;
365}
366
367/*
368 * Call function on an online CPU.
369 */
370void smp_call_online_cpu(void (*func)(void *), void *data)
371{
372 struct pcpu *pcpu;
373
374 /* Use the current cpu if it is online. */
375 pcpu = pcpu_find_address(cpu_online_mask, stap());
376 if (!pcpu)
377 /* Use the first online cpu. */
378 pcpu = pcpu_devices + cpumask_first(cpu_online_mask);
379 pcpu_delegate(pcpu, func, data, (unsigned long) restart_stack);
380}
381
382/*
383 * Call function on the ipl CPU.
384 */
385void smp_call_ipl_cpu(void (*func)(void *), void *data)
386{
387 struct lowcore *lc = lowcore_ptr[0];
388
389 if (pcpu_devices[0].address == stap())
390 lc = &S390_lowcore;
391
392 pcpu_delegate(&pcpu_devices[0], func, data,
393 lc->nodat_stack);
394}
395
396int smp_find_processor_id(u16 address)
397{
398 int cpu;
399
400 for_each_present_cpu(cpu)
401 if (pcpu_devices[cpu].address == address)
402 return cpu;
403 return -1;
404}
405
406void schedule_mcck_handler(void)
407{
408 pcpu_ec_call(pcpu_devices + smp_processor_id(), ec_mcck_pending);
409}
410
411bool notrace arch_vcpu_is_preempted(int cpu)
412{
413 if (test_cpu_flag_of(CIF_ENABLED_WAIT, cpu))
414 return false;
415 if (pcpu_running(pcpu_devices + cpu))
416 return false;
417 return true;
418}
419EXPORT_SYMBOL(arch_vcpu_is_preempted);
420
421void notrace smp_yield_cpu(int cpu)
422{
423 if (!MACHINE_HAS_DIAG9C)
424 return;
425 diag_stat_inc_norecursion(DIAG_STAT_X09C);
426 asm volatile("diag %0,0,0x9c"
427 : : "d" (pcpu_devices[cpu].address));
428}
429EXPORT_SYMBOL_GPL(smp_yield_cpu);
430
431/*
432 * Send cpus emergency shutdown signal. This gives the cpus the
433 * opportunity to complete outstanding interrupts.
434 */
435void notrace smp_emergency_stop(void)
436{
437 static arch_spinlock_t lock = __ARCH_SPIN_LOCK_UNLOCKED;
438 static cpumask_t cpumask;
439 u64 end;
440 int cpu;
441
442 arch_spin_lock(&lock);
443 cpumask_copy(&cpumask, cpu_online_mask);
444 cpumask_clear_cpu(smp_processor_id(), &cpumask);
445
446 end = get_tod_clock() + (1000000UL << 12);
447 for_each_cpu(cpu, &cpumask) {
448 struct pcpu *pcpu = pcpu_devices + cpu;
449 set_bit(ec_stop_cpu, &pcpu->ec_mask);
450 while (__pcpu_sigp(pcpu->address, SIGP_EMERGENCY_SIGNAL,
451 0, NULL) == SIGP_CC_BUSY &&
452 get_tod_clock() < end)
453 cpu_relax();
454 }
455 while (get_tod_clock() < end) {
456 for_each_cpu(cpu, &cpumask)
457 if (pcpu_stopped(pcpu_devices + cpu))
458 cpumask_clear_cpu(cpu, &cpumask);
459 if (cpumask_empty(&cpumask))
460 break;
461 cpu_relax();
462 }
463 arch_spin_unlock(&lock);
464}
465NOKPROBE_SYMBOL(smp_emergency_stop);
466
467/*
468 * Stop all cpus but the current one.
469 */
470void smp_send_stop(void)
471{
472 int cpu;
473
474 /* Disable all interrupts/machine checks */
475 __load_psw_mask(PSW_KERNEL_BITS | PSW_MASK_DAT);
476 trace_hardirqs_off();
477
478 debug_set_critical();
479
480 if (oops_in_progress)
481 smp_emergency_stop();
482
483 /* stop all processors */
484 for_each_online_cpu(cpu) {
485 if (cpu == smp_processor_id())
486 continue;
487 pcpu_sigp_retry(pcpu_devices + cpu, SIGP_STOP, 0);
488 while (!pcpu_stopped(pcpu_devices + cpu))
489 cpu_relax();
490 }
491}
492
493/*
494 * This is the main routine where commands issued by other
495 * cpus are handled.
496 */
497static void smp_handle_ext_call(void)
498{
499 unsigned long bits;
500
501 /* handle bit signal external calls */
502 bits = xchg(&pcpu_devices[smp_processor_id()].ec_mask, 0);
503 if (test_bit(ec_stop_cpu, &bits))
504 smp_stop_cpu();
505 if (test_bit(ec_schedule, &bits))
506 scheduler_ipi();
507 if (test_bit(ec_call_function_single, &bits))
508 generic_smp_call_function_single_interrupt();
509 if (test_bit(ec_mcck_pending, &bits))
510 __s390_handle_mcck();
511 if (test_bit(ec_irq_work, &bits))
512 irq_work_run();
513}
514
515static void do_ext_call_interrupt(struct ext_code ext_code,
516 unsigned int param32, unsigned long param64)
517{
518 inc_irq_stat(ext_code.code == 0x1202 ? IRQEXT_EXC : IRQEXT_EMS);
519 smp_handle_ext_call();
520}
521
522void arch_send_call_function_ipi_mask(const struct cpumask *mask)
523{
524 int cpu;
525
526 for_each_cpu(cpu, mask)
527 pcpu_ec_call(pcpu_devices + cpu, ec_call_function_single);
528}
529
530void arch_send_call_function_single_ipi(int cpu)
531{
532 pcpu_ec_call(pcpu_devices + cpu, ec_call_function_single);
533}
534
535/*
536 * this function sends a 'reschedule' IPI to another CPU.
537 * it goes straight through and wastes no time serializing
538 * anything. Worst case is that we lose a reschedule ...
539 */
540void smp_send_reschedule(int cpu)
541{
542 pcpu_ec_call(pcpu_devices + cpu, ec_schedule);
543}
544
545#ifdef CONFIG_IRQ_WORK
546void arch_irq_work_raise(void)
547{
548 pcpu_ec_call(pcpu_devices + smp_processor_id(), ec_irq_work);
549}
550#endif
551
552/*
553 * parameter area for the set/clear control bit callbacks
554 */
555struct ec_creg_mask_parms {
556 unsigned long orval;
557 unsigned long andval;
558 int cr;
559};
560
561/*
562 * callback for setting/clearing control bits
563 */
564static void smp_ctl_bit_callback(void *info)
565{
566 struct ec_creg_mask_parms *pp = info;
567 unsigned long cregs[16];
568
569 __ctl_store(cregs, 0, 15);
570 cregs[pp->cr] = (cregs[pp->cr] & pp->andval) | pp->orval;
571 __ctl_load(cregs, 0, 15);
572}
573
574static DEFINE_SPINLOCK(ctl_lock);
575static unsigned long ctlreg;
576
577/*
578 * Set a bit in a control register of all cpus
579 */
580void smp_ctl_set_bit(int cr, int bit)
581{
582 struct ec_creg_mask_parms parms = { 1UL << bit, -1UL, cr };
583
584 spin_lock(&ctl_lock);
585 memcpy_absolute(&ctlreg, &S390_lowcore.cregs_save_area[cr], sizeof(ctlreg));
586 __set_bit(bit, &ctlreg);
587 memcpy_absolute(&S390_lowcore.cregs_save_area[cr], &ctlreg, sizeof(ctlreg));
588 spin_unlock(&ctl_lock);
589 on_each_cpu(smp_ctl_bit_callback, &parms, 1);
590}
591EXPORT_SYMBOL(smp_ctl_set_bit);
592
593/*
594 * Clear a bit in a control register of all cpus
595 */
596void smp_ctl_clear_bit(int cr, int bit)
597{
598 struct ec_creg_mask_parms parms = { 0, ~(1UL << bit), cr };
599
600 spin_lock(&ctl_lock);
601 memcpy_absolute(&ctlreg, &S390_lowcore.cregs_save_area[cr], sizeof(ctlreg));
602 __clear_bit(bit, &ctlreg);
603 memcpy_absolute(&S390_lowcore.cregs_save_area[cr], &ctlreg, sizeof(ctlreg));
604 spin_unlock(&ctl_lock);
605 on_each_cpu(smp_ctl_bit_callback, &parms, 1);
606}
607EXPORT_SYMBOL(smp_ctl_clear_bit);
608
609#ifdef CONFIG_CRASH_DUMP
610
611int smp_store_status(int cpu)
612{
613 struct lowcore *lc;
614 struct pcpu *pcpu;
615 unsigned long pa;
616
617 pcpu = pcpu_devices + cpu;
618 lc = lowcore_ptr[cpu];
619 pa = __pa(&lc->floating_pt_save_area);
620 if (__pcpu_sigp_relax(pcpu->address, SIGP_STORE_STATUS_AT_ADDRESS,
621 pa) != SIGP_CC_ORDER_CODE_ACCEPTED)
622 return -EIO;
623 if (!MACHINE_HAS_VX && !MACHINE_HAS_GS)
624 return 0;
625 pa = __pa(lc->mcesad & MCESA_ORIGIN_MASK);
626 if (MACHINE_HAS_GS)
627 pa |= lc->mcesad & MCESA_LC_MASK;
628 if (__pcpu_sigp_relax(pcpu->address, SIGP_STORE_ADDITIONAL_STATUS,
629 pa) != SIGP_CC_ORDER_CODE_ACCEPTED)
630 return -EIO;
631 return 0;
632}
633
634/*
635 * Collect CPU state of the previous, crashed system.
636 * There are four cases:
637 * 1) standard zfcp/nvme dump
638 * condition: OLDMEM_BASE == NULL && is_ipl_type_dump() == true
639 * The state for all CPUs except the boot CPU needs to be collected
640 * with sigp stop-and-store-status. The boot CPU state is located in
641 * the absolute lowcore of the memory stored in the HSA. The zcore code
642 * will copy the boot CPU state from the HSA.
643 * 2) stand-alone kdump for SCSI/NVMe (zfcp/nvme dump with swapped memory)
644 * condition: OLDMEM_BASE != NULL && is_ipl_type_dump() == true
645 * The state for all CPUs except the boot CPU needs to be collected
646 * with sigp stop-and-store-status. The firmware or the boot-loader
647 * stored the registers of the boot CPU in the absolute lowcore in the
648 * memory of the old system.
649 * 3) kdump and the old kernel did not store the CPU state,
650 * or stand-alone kdump for DASD
651 * condition: OLDMEM_BASE != NULL && !is_kdump_kernel()
652 * The state for all CPUs except the boot CPU needs to be collected
653 * with sigp stop-and-store-status. The kexec code or the boot-loader
654 * stored the registers of the boot CPU in the memory of the old system.
655 * 4) kdump and the old kernel stored the CPU state
656 * condition: OLDMEM_BASE != NULL && is_kdump_kernel()
657 * This case does not exist for s390 anymore, setup_arch explicitly
658 * deactivates the elfcorehdr= kernel parameter
659 */
660static __init void smp_save_cpu_vxrs(struct save_area *sa, u16 addr,
661 bool is_boot_cpu, unsigned long page)
662{
663 __vector128 *vxrs = (__vector128 *) page;
664
665 if (is_boot_cpu)
666 vxrs = boot_cpu_vector_save_area;
667 else
668 __pcpu_sigp_relax(addr, SIGP_STORE_ADDITIONAL_STATUS, page);
669 save_area_add_vxrs(sa, vxrs);
670}
671
672static __init void smp_save_cpu_regs(struct save_area *sa, u16 addr,
673 bool is_boot_cpu, unsigned long page)
674{
675 void *regs = (void *) page;
676
677 if (is_boot_cpu)
678 copy_oldmem_kernel(regs, (void *) __LC_FPREGS_SAVE_AREA, 512);
679 else
680 __pcpu_sigp_relax(addr, SIGP_STORE_STATUS_AT_ADDRESS, page);
681 save_area_add_regs(sa, regs);
682}
683
684void __init smp_save_dump_cpus(void)
685{
686 int addr, boot_cpu_addr, max_cpu_addr;
687 struct save_area *sa;
688 unsigned long page;
689 bool is_boot_cpu;
690
691 if (!(OLDMEM_BASE || is_ipl_type_dump()))
692 /* No previous system present, normal boot. */
693 return;
694 /* Allocate a page as dumping area for the store status sigps */
695 page = memblock_phys_alloc_range(PAGE_SIZE, PAGE_SIZE, 0, 1UL << 31);
696 if (!page)
697 panic("ERROR: Failed to allocate %lx bytes below %lx\n",
698 PAGE_SIZE, 1UL << 31);
699
700 /* Set multi-threading state to the previous system. */
701 pcpu_set_smt(sclp.mtid_prev);
702 boot_cpu_addr = stap();
703 max_cpu_addr = SCLP_MAX_CORES << sclp.mtid_prev;
704 for (addr = 0; addr <= max_cpu_addr; addr++) {
705 if (__pcpu_sigp_relax(addr, SIGP_SENSE, 0) ==
706 SIGP_CC_NOT_OPERATIONAL)
707 continue;
708 is_boot_cpu = (addr == boot_cpu_addr);
709 /* Allocate save area */
710 sa = save_area_alloc(is_boot_cpu);
711 if (!sa)
712 panic("could not allocate memory for save area\n");
713 if (MACHINE_HAS_VX)
714 /* Get the vector registers */
715 smp_save_cpu_vxrs(sa, addr, is_boot_cpu, page);
716 /*
717 * For a zfcp/nvme dump OLDMEM_BASE == NULL and the registers
718 * of the boot CPU are stored in the HSA. To retrieve
719 * these registers an SCLP request is required which is
720 * done by drivers/s390/char/zcore.c:init_cpu_info()
721 */
722 if (!is_boot_cpu || OLDMEM_BASE)
723 /* Get the CPU registers */
724 smp_save_cpu_regs(sa, addr, is_boot_cpu, page);
725 }
726 memblock_free(page, PAGE_SIZE);
727 diag_dma_ops.diag308_reset();
728 pcpu_set_smt(0);
729}
730#endif /* CONFIG_CRASH_DUMP */
731
732void smp_cpu_set_polarization(int cpu, int val)
733{
734 pcpu_devices[cpu].polarization = val;
735}
736
737int smp_cpu_get_polarization(int cpu)
738{
739 return pcpu_devices[cpu].polarization;
740}
741
742int smp_cpu_get_cpu_address(int cpu)
743{
744 return pcpu_devices[cpu].address;
745}
746
747static void __ref smp_get_core_info(struct sclp_core_info *info, int early)
748{
749 static int use_sigp_detection;
750 int address;
751
752 if (use_sigp_detection || sclp_get_core_info(info, early)) {
753 use_sigp_detection = 1;
754 for (address = 0;
755 address < (SCLP_MAX_CORES << smp_cpu_mt_shift);
756 address += (1U << smp_cpu_mt_shift)) {
757 if (__pcpu_sigp_relax(address, SIGP_SENSE, 0) ==
758 SIGP_CC_NOT_OPERATIONAL)
759 continue;
760 info->core[info->configured].core_id =
761 address >> smp_cpu_mt_shift;
762 info->configured++;
763 }
764 info->combined = info->configured;
765 }
766}
767
768static int smp_add_present_cpu(int cpu);
769
770static int smp_add_core(struct sclp_core_entry *core, cpumask_t *avail,
771 bool configured, bool early)
772{
773 struct pcpu *pcpu;
774 int cpu, nr, i;
775 u16 address;
776
777 nr = 0;
778 if (sclp.has_core_type && core->type != boot_core_type)
779 return nr;
780 cpu = cpumask_first(avail);
781 address = core->core_id << smp_cpu_mt_shift;
782 for (i = 0; (i <= smp_cpu_mtid) && (cpu < nr_cpu_ids); i++) {
783 if (pcpu_find_address(cpu_present_mask, address + i))
784 continue;
785 pcpu = pcpu_devices + cpu;
786 pcpu->address = address + i;
787 if (configured)
788 pcpu->state = CPU_STATE_CONFIGURED;
789 else
790 pcpu->state = CPU_STATE_STANDBY;
791 smp_cpu_set_polarization(cpu, POLARIZATION_UNKNOWN);
792 set_cpu_present(cpu, true);
793 if (!early && smp_add_present_cpu(cpu) != 0)
794 set_cpu_present(cpu, false);
795 else
796 nr++;
797 cpumask_clear_cpu(cpu, avail);
798 cpu = cpumask_next(cpu, avail);
799 }
800 return nr;
801}
802
803static int __smp_rescan_cpus(struct sclp_core_info *info, bool early)
804{
805 struct sclp_core_entry *core;
806 static cpumask_t avail;
807 bool configured;
808 u16 core_id;
809 int nr, i;
810
811 get_online_cpus();
812 mutex_lock(&smp_cpu_state_mutex);
813 nr = 0;
814 cpumask_xor(&avail, cpu_possible_mask, cpu_present_mask);
815 /*
816 * Add IPL core first (which got logical CPU number 0) to make sure
817 * that all SMT threads get subsequent logical CPU numbers.
818 */
819 if (early) {
820 core_id = pcpu_devices[0].address >> smp_cpu_mt_shift;
821 for (i = 0; i < info->configured; i++) {
822 core = &info->core[i];
823 if (core->core_id == core_id) {
824 nr += smp_add_core(core, &avail, true, early);
825 break;
826 }
827 }
828 }
829 for (i = 0; i < info->combined; i++) {
830 configured = i < info->configured;
831 nr += smp_add_core(&info->core[i], &avail, configured, early);
832 }
833 mutex_unlock(&smp_cpu_state_mutex);
834 put_online_cpus();
835 return nr;
836}
837
838void __init smp_detect_cpus(void)
839{
840 unsigned int cpu, mtid, c_cpus, s_cpus;
841 struct sclp_core_info *info;
842 u16 address;
843
844 /* Get CPU information */
845 info = memblock_alloc(sizeof(*info), 8);
846 if (!info)
847 panic("%s: Failed to allocate %zu bytes align=0x%x\n",
848 __func__, sizeof(*info), 8);
849 smp_get_core_info(info, 1);
850 /* Find boot CPU type */
851 if (sclp.has_core_type) {
852 address = stap();
853 for (cpu = 0; cpu < info->combined; cpu++)
854 if (info->core[cpu].core_id == address) {
855 /* The boot cpu dictates the cpu type. */
856 boot_core_type = info->core[cpu].type;
857 break;
858 }
859 if (cpu >= info->combined)
860 panic("Could not find boot CPU type");
861 }
862
863 /* Set multi-threading state for the current system */
864 mtid = boot_core_type ? sclp.mtid : sclp.mtid_cp;
865 mtid = (mtid < smp_max_threads) ? mtid : smp_max_threads - 1;
866 pcpu_set_smt(mtid);
867
868 /* Print number of CPUs */
869 c_cpus = s_cpus = 0;
870 for (cpu = 0; cpu < info->combined; cpu++) {
871 if (sclp.has_core_type &&
872 info->core[cpu].type != boot_core_type)
873 continue;
874 if (cpu < info->configured)
875 c_cpus += smp_cpu_mtid + 1;
876 else
877 s_cpus += smp_cpu_mtid + 1;
878 }
879 pr_info("%d configured CPUs, %d standby CPUs\n", c_cpus, s_cpus);
880
881 /* Add CPUs present at boot */
882 __smp_rescan_cpus(info, true);
883 memblock_free_early((unsigned long)info, sizeof(*info));
884}
885
886static void smp_init_secondary(void)
887{
888 int cpu = raw_smp_processor_id();
889
890 S390_lowcore.last_update_clock = get_tod_clock();
891 restore_access_regs(S390_lowcore.access_regs_save_area);
892 cpu_init();
893 rcu_cpu_starting(cpu);
894 init_cpu_timer();
895 vtime_init();
896 vdso_getcpu_init();
897 pfault_init();
898 cpumask_set_cpu(cpu, &cpu_setup_mask);
899 update_cpu_masks();
900 notify_cpu_starting(cpu);
901 if (topology_cpu_dedicated(cpu))
902 set_cpu_flag(CIF_DEDICATED_CPU);
903 else
904 clear_cpu_flag(CIF_DEDICATED_CPU);
905 set_cpu_online(cpu, true);
906 inc_irq_stat(CPU_RST);
907 local_irq_enable();
908 cpu_startup_entry(CPUHP_AP_ONLINE_IDLE);
909}
910
911/*
912 * Activate a secondary processor.
913 */
914static void smp_start_secondary(void *cpuvoid)
915{
916 S390_lowcore.restart_stack = (unsigned long) restart_stack;
917 S390_lowcore.restart_fn = (unsigned long) do_restart;
918 S390_lowcore.restart_data = 0;
919 S390_lowcore.restart_source = -1U;
920 S390_lowcore.restart_flags = 0;
921 call_on_stack_noreturn(smp_init_secondary, S390_lowcore.kernel_stack);
922}
923
924/* Upping and downing of CPUs */
925int __cpu_up(unsigned int cpu, struct task_struct *tidle)
926{
927 struct pcpu *pcpu = pcpu_devices + cpu;
928 int rc;
929
930 if (pcpu->state != CPU_STATE_CONFIGURED)
931 return -EIO;
932 if (pcpu_sigp_retry(pcpu, SIGP_INITIAL_CPU_RESET, 0) !=
933 SIGP_CC_ORDER_CODE_ACCEPTED)
934 return -EIO;
935
936 rc = pcpu_alloc_lowcore(pcpu, cpu);
937 if (rc)
938 return rc;
939 pcpu_prepare_secondary(pcpu, cpu);
940 pcpu_attach_task(pcpu, tidle);
941 pcpu_start_fn(pcpu, smp_start_secondary, NULL);
942 /* Wait until cpu puts itself in the online & active maps */
943 while (!cpu_online(cpu))
944 cpu_relax();
945 return 0;
946}
947
948static unsigned int setup_possible_cpus __initdata;
949
950static int __init _setup_possible_cpus(char *s)
951{
952 get_option(&s, &setup_possible_cpus);
953 return 0;
954}
955early_param("possible_cpus", _setup_possible_cpus);
956
957int __cpu_disable(void)
958{
959 unsigned long cregs[16];
960 int cpu;
961
962 /* Handle possible pending IPIs */
963 smp_handle_ext_call();
964 cpu = smp_processor_id();
965 set_cpu_online(cpu, false);
966 cpumask_clear_cpu(cpu, &cpu_setup_mask);
967 update_cpu_masks();
968 /* Disable pseudo page faults on this cpu. */
969 pfault_fini();
970 /* Disable interrupt sources via control register. */
971 __ctl_store(cregs, 0, 15);
972 cregs[0] &= ~0x0000ee70UL; /* disable all external interrupts */
973 cregs[6] &= ~0xff000000UL; /* disable all I/O interrupts */
974 cregs[14] &= ~0x1f000000UL; /* disable most machine checks */
975 __ctl_load(cregs, 0, 15);
976 clear_cpu_flag(CIF_NOHZ_DELAY);
977 return 0;
978}
979
980void __cpu_die(unsigned int cpu)
981{
982 struct pcpu *pcpu;
983
984 /* Wait until target cpu is down */
985 pcpu = pcpu_devices + cpu;
986 while (!pcpu_stopped(pcpu))
987 cpu_relax();
988 pcpu_free_lowcore(pcpu);
989 cpumask_clear_cpu(cpu, mm_cpumask(&init_mm));
990 cpumask_clear_cpu(cpu, &init_mm.context.cpu_attach_mask);
991}
992
993void __noreturn cpu_die(void)
994{
995 idle_task_exit();
996 __bpon();
997 pcpu_sigp_retry(pcpu_devices + smp_processor_id(), SIGP_STOP, 0);
998 for (;;) ;
999}
1000
1001void __init smp_fill_possible_mask(void)
1002{
1003 unsigned int possible, sclp_max, cpu;
1004
1005 sclp_max = max(sclp.mtid, sclp.mtid_cp) + 1;
1006 sclp_max = min(smp_max_threads, sclp_max);
1007 sclp_max = (sclp.max_cores * sclp_max) ?: nr_cpu_ids;
1008 possible = setup_possible_cpus ?: nr_cpu_ids;
1009 possible = min(possible, sclp_max);
1010 for (cpu = 0; cpu < possible && cpu < nr_cpu_ids; cpu++)
1011 set_cpu_possible(cpu, true);
1012}
1013
1014void __init smp_prepare_cpus(unsigned int max_cpus)
1015{
1016 /* request the 0x1201 emergency signal external interrupt */
1017 if (register_external_irq(EXT_IRQ_EMERGENCY_SIG, do_ext_call_interrupt))
1018 panic("Couldn't request external interrupt 0x1201");
1019 /* request the 0x1202 external call external interrupt */
1020 if (register_external_irq(EXT_IRQ_EXTERNAL_CALL, do_ext_call_interrupt))
1021 panic("Couldn't request external interrupt 0x1202");
1022}
1023
1024void __init smp_prepare_boot_cpu(void)
1025{
1026 struct pcpu *pcpu = pcpu_devices;
1027
1028 WARN_ON(!cpu_present(0) || !cpu_online(0));
1029 pcpu->state = CPU_STATE_CONFIGURED;
1030 S390_lowcore.percpu_offset = __per_cpu_offset[0];
1031 smp_cpu_set_polarization(0, POLARIZATION_UNKNOWN);
1032}
1033
1034void __init smp_setup_processor_id(void)
1035{
1036 pcpu_devices[0].address = stap();
1037 S390_lowcore.cpu_nr = 0;
1038 S390_lowcore.spinlock_lockval = arch_spin_lockval(0);
1039 S390_lowcore.spinlock_index = 0;
1040}
1041
1042/*
1043 * the frequency of the profiling timer can be changed
1044 * by writing a multiplier value into /proc/profile.
1045 *
1046 * usually you want to run this on all CPUs ;)
1047 */
1048int setup_profiling_timer(unsigned int multiplier)
1049{
1050 return 0;
1051}
1052
1053static ssize_t cpu_configure_show(struct device *dev,
1054 struct device_attribute *attr, char *buf)
1055{
1056 ssize_t count;
1057
1058 mutex_lock(&smp_cpu_state_mutex);
1059 count = sprintf(buf, "%d\n", pcpu_devices[dev->id].state);
1060 mutex_unlock(&smp_cpu_state_mutex);
1061 return count;
1062}
1063
1064static ssize_t cpu_configure_store(struct device *dev,
1065 struct device_attribute *attr,
1066 const char *buf, size_t count)
1067{
1068 struct pcpu *pcpu;
1069 int cpu, val, rc, i;
1070 char delim;
1071
1072 if (sscanf(buf, "%d %c", &val, &delim) != 1)
1073 return -EINVAL;
1074 if (val != 0 && val != 1)
1075 return -EINVAL;
1076 get_online_cpus();
1077 mutex_lock(&smp_cpu_state_mutex);
1078 rc = -EBUSY;
1079 /* disallow configuration changes of online cpus and cpu 0 */
1080 cpu = dev->id;
1081 cpu = smp_get_base_cpu(cpu);
1082 if (cpu == 0)
1083 goto out;
1084 for (i = 0; i <= smp_cpu_mtid; i++)
1085 if (cpu_online(cpu + i))
1086 goto out;
1087 pcpu = pcpu_devices + cpu;
1088 rc = 0;
1089 switch (val) {
1090 case 0:
1091 if (pcpu->state != CPU_STATE_CONFIGURED)
1092 break;
1093 rc = sclp_core_deconfigure(pcpu->address >> smp_cpu_mt_shift);
1094 if (rc)
1095 break;
1096 for (i = 0; i <= smp_cpu_mtid; i++) {
1097 if (cpu + i >= nr_cpu_ids || !cpu_present(cpu + i))
1098 continue;
1099 pcpu[i].state = CPU_STATE_STANDBY;
1100 smp_cpu_set_polarization(cpu + i,
1101 POLARIZATION_UNKNOWN);
1102 }
1103 topology_expect_change();
1104 break;
1105 case 1:
1106 if (pcpu->state != CPU_STATE_STANDBY)
1107 break;
1108 rc = sclp_core_configure(pcpu->address >> smp_cpu_mt_shift);
1109 if (rc)
1110 break;
1111 for (i = 0; i <= smp_cpu_mtid; i++) {
1112 if (cpu + i >= nr_cpu_ids || !cpu_present(cpu + i))
1113 continue;
1114 pcpu[i].state = CPU_STATE_CONFIGURED;
1115 smp_cpu_set_polarization(cpu + i,
1116 POLARIZATION_UNKNOWN);
1117 }
1118 topology_expect_change();
1119 break;
1120 default:
1121 break;
1122 }
1123out:
1124 mutex_unlock(&smp_cpu_state_mutex);
1125 put_online_cpus();
1126 return rc ? rc : count;
1127}
1128static DEVICE_ATTR(configure, 0644, cpu_configure_show, cpu_configure_store);
1129
1130static ssize_t show_cpu_address(struct device *dev,
1131 struct device_attribute *attr, char *buf)
1132{
1133 return sprintf(buf, "%d\n", pcpu_devices[dev->id].address);
1134}
1135static DEVICE_ATTR(address, 0444, show_cpu_address, NULL);
1136
1137static struct attribute *cpu_common_attrs[] = {
1138 &dev_attr_configure.attr,
1139 &dev_attr_address.attr,
1140 NULL,
1141};
1142
1143static struct attribute_group cpu_common_attr_group = {
1144 .attrs = cpu_common_attrs,
1145};
1146
1147static struct attribute *cpu_online_attrs[] = {
1148 &dev_attr_idle_count.attr,
1149 &dev_attr_idle_time_us.attr,
1150 NULL,
1151};
1152
1153static struct attribute_group cpu_online_attr_group = {
1154 .attrs = cpu_online_attrs,
1155};
1156
1157static int smp_cpu_online(unsigned int cpu)
1158{
1159 struct device *s = &per_cpu(cpu_device, cpu)->dev;
1160
1161 return sysfs_create_group(&s->kobj, &cpu_online_attr_group);
1162}
1163
1164static int smp_cpu_pre_down(unsigned int cpu)
1165{
1166 struct device *s = &per_cpu(cpu_device, cpu)->dev;
1167
1168 sysfs_remove_group(&s->kobj, &cpu_online_attr_group);
1169 return 0;
1170}
1171
1172static int smp_add_present_cpu(int cpu)
1173{
1174 struct device *s;
1175 struct cpu *c;
1176 int rc;
1177
1178 c = kzalloc(sizeof(*c), GFP_KERNEL);
1179 if (!c)
1180 return -ENOMEM;
1181 per_cpu(cpu_device, cpu) = c;
1182 s = &c->dev;
1183 c->hotpluggable = 1;
1184 rc = register_cpu(c, cpu);
1185 if (rc)
1186 goto out;
1187 rc = sysfs_create_group(&s->kobj, &cpu_common_attr_group);
1188 if (rc)
1189 goto out_cpu;
1190 rc = topology_cpu_init(c);
1191 if (rc)
1192 goto out_topology;
1193 return 0;
1194
1195out_topology:
1196 sysfs_remove_group(&s->kobj, &cpu_common_attr_group);
1197out_cpu:
1198 unregister_cpu(c);
1199out:
1200 return rc;
1201}
1202
1203int __ref smp_rescan_cpus(void)
1204{
1205 struct sclp_core_info *info;
1206 int nr;
1207
1208 info = kzalloc(sizeof(*info), GFP_KERNEL);
1209 if (!info)
1210 return -ENOMEM;
1211 smp_get_core_info(info, 0);
1212 nr = __smp_rescan_cpus(info, false);
1213 kfree(info);
1214 if (nr)
1215 topology_schedule_update();
1216 return 0;
1217}
1218
1219static ssize_t __ref rescan_store(struct device *dev,
1220 struct device_attribute *attr,
1221 const char *buf,
1222 size_t count)
1223{
1224 int rc;
1225
1226 rc = lock_device_hotplug_sysfs();
1227 if (rc)
1228 return rc;
1229 rc = smp_rescan_cpus();
1230 unlock_device_hotplug();
1231 return rc ? rc : count;
1232}
1233static DEVICE_ATTR_WO(rescan);
1234
1235static int __init s390_smp_init(void)
1236{
1237 int cpu, rc = 0;
1238
1239 rc = device_create_file(cpu_subsys.dev_root, &dev_attr_rescan);
1240 if (rc)
1241 return rc;
1242 for_each_present_cpu(cpu) {
1243 rc = smp_add_present_cpu(cpu);
1244 if (rc)
1245 goto out;
1246 }
1247
1248 rc = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "s390/smp:online",
1249 smp_cpu_online, smp_cpu_pre_down);
1250 rc = rc <= 0 ? rc : 0;
1251out:
1252 return rc;
1253}
1254subsys_initcall(s390_smp_init);
1255
1256static __always_inline void set_new_lowcore(struct lowcore *lc)
1257{
1258 union register_pair dst, src;
1259 u32 pfx;
1260
1261 src.even = (unsigned long) &S390_lowcore;
1262 src.odd = sizeof(S390_lowcore);
1263 dst.even = (unsigned long) lc;
1264 dst.odd = sizeof(*lc);
1265 pfx = (unsigned long) lc;
1266
1267 asm volatile(
1268 " mvcl %[dst],%[src]\n"
1269 " spx %[pfx]\n"
1270 : [dst] "+&d" (dst.pair), [src] "+&d" (src.pair)
1271 : [pfx] "Q" (pfx)
1272 : "memory", "cc");
1273}
1274
1275static int __init smp_reinit_ipl_cpu(void)
1276{
1277 unsigned long async_stack, nodat_stack, mcck_stack;
1278 struct lowcore *lc, *lc_ipl;
1279 unsigned long flags;
1280
1281 lc_ipl = lowcore_ptr[0];
1282 lc = (struct lowcore *) __get_free_pages(GFP_KERNEL | GFP_DMA, LC_ORDER);
1283 nodat_stack = __get_free_pages(GFP_KERNEL, THREAD_SIZE_ORDER);
1284 async_stack = stack_alloc();
1285 mcck_stack = stack_alloc();
1286 if (!lc || !nodat_stack || !async_stack || !mcck_stack)
1287 panic("Couldn't allocate memory");
1288
1289 local_irq_save(flags);
1290 local_mcck_disable();
1291 set_new_lowcore(lc);
1292 S390_lowcore.nodat_stack = nodat_stack + STACK_INIT_OFFSET;
1293 S390_lowcore.async_stack = async_stack + STACK_INIT_OFFSET;
1294 S390_lowcore.mcck_stack = mcck_stack + STACK_INIT_OFFSET;
1295 lowcore_ptr[0] = lc;
1296 local_mcck_enable();
1297 local_irq_restore(flags);
1298
1299 free_pages(lc_ipl->async_stack - STACK_INIT_OFFSET, THREAD_SIZE_ORDER);
1300 memblock_free_late(lc_ipl->mcck_stack - STACK_INIT_OFFSET, THREAD_SIZE);
1301 memblock_free_late((unsigned long) lc_ipl, sizeof(*lc_ipl));
1302
1303 return 0;
1304}
1305early_initcall(smp_reinit_ipl_cpu);