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
64 enum {
65 ec_schedule = 0,
66 ec_call_function_single,
67 ec_stop_cpu,
68 ec_mcck_pending,
69 ec_irq_work,
70 };
71
72 enum {
73 CPU_STATE_STANDBY,
74 CPU_STATE_CONFIGURED,
75 };
76
77 static u8 boot_core_type;
78 DEFINE_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 */
85 static struct pcpu *ipl_pcpu;
86
87 unsigned int smp_cpu_mt_shift;
88 EXPORT_SYMBOL(smp_cpu_mt_shift);
89
90 unsigned int smp_cpu_mtid;
91 EXPORT_SYMBOL(smp_cpu_mtid);
92
93 #ifdef CONFIG_CRASH_DUMP
94 __vector128 __initdata boot_cpu_vector_save_area[__NUM_VXRS];
95 #endif
96
97 static unsigned int smp_max_threads __initdata = -1U;
98 cpumask_t cpu_setup_mask;
99
early_nosmt(char * s)100 static int __init early_nosmt(char *s)
101 {
102 smp_max_threads = 1;
103 return 0;
104 }
105 early_param("nosmt", early_nosmt);
106
early_smt(char * s)107 static int __init early_smt(char *s)
108 {
109 get_option(&s, &smp_max_threads);
110 return 0;
111 }
112 early_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 */
118 DEFINE_MUTEX(smp_cpu_state_mutex);
119
120 /*
121 * Signal processor helper functions.
122 */
__pcpu_sigp_relax(u16 addr,u8 order,unsigned long parm)123 static 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
pcpu_sigp_retry(struct pcpu * pcpu,u8 order,u32 parm)135 static 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
pcpu_stopped(struct pcpu * pcpu)149 static 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
pcpu_running(struct pcpu * pcpu)159 static 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 */
pcpu_find_address(const struct cpumask * mask,u16 address)171 static 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
pcpu_ec_call(struct pcpu * pcpu,int ec_bit)181 static 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
pcpu_alloc_lowcore(struct pcpu * pcpu,int cpu)192 static 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
222 out_mcesa:
223 nmi_free_mcesa(&lc->mcesad);
224 out:
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
pcpu_free_lowcore(struct pcpu * pcpu,int cpu)232 static 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
pcpu_prepare_secondary(struct pcpu * pcpu,int cpu)251 static 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
pcpu_attach_task(int cpu,struct task_struct * tsk)278 static 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
pcpu_start_fn(int cpu,void (* func)(void *),void * data)295 static 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
307 typedef void (pcpu_delegate_fn)(void *);
308
309 /*
310 * Call function via PSW restart on pcpu and stop the current cpu.
311 */
__pcpu_delegate(pcpu_delegate_fn * func,void * data)312 static void __pcpu_delegate(pcpu_delegate_fn *func, void *data)
313 {
314 func(data); /* should not return */
315 }
316
pcpu_delegate(struct pcpu * pcpu,int cpu,pcpu_delegate_fn * func,void * data,unsigned long stack)317 static 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 */
pcpu_set_smt(unsigned int mtid)362 static 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 */
smp_call_ipl_cpu(void (* func)(void *),void * data)382 void 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
smp_find_processor_id(u16 address)392 int 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
schedule_mcck_handler(void)402 void schedule_mcck_handler(void)
403 {
404 pcpu_ec_call(this_cpu_ptr(&pcpu_devices), ec_mcck_pending);
405 }
406
arch_vcpu_is_preempted(int cpu)407 bool 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 }
415 EXPORT_SYMBOL(arch_vcpu_is_preempted);
416
smp_yield_cpu(int cpu)417 void 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 }
425 EXPORT_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 */
smp_emergency_stop(void)431 void 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 }
461 NOKPROBE_SYMBOL(smp_emergency_stop);
462
463 /*
464 * Stop all cpus but the current one.
465 */
smp_send_stop(void)466 void 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 */
smp_handle_ext_call(void)495 static 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
do_ext_call_interrupt(struct ext_code ext_code,unsigned int param32,unsigned long param64)513 static 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
arch_send_call_function_ipi_mask(const struct cpumask * mask)520 void 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
arch_send_call_function_single_ipi(int cpu)528 void 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 */
arch_smp_send_reschedule(int cpu)538 void 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
arch_irq_work_raise(void)544 void 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
smp_store_status(int cpu)552 int 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 four 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 and the old kernel did not store the CPU state,
591 * or stand-alone kdump for DASD
592 * condition: OLDMEM_BASE != NULL && !is_kdump_kernel()
593 * The state for all CPUs except the boot CPU needs to be collected
594 * with sigp stop-and-store-status. The kexec code or the boot-loader
595 * stored the registers of the boot CPU in the memory of the old system.
596 * 4) kdump and the old kernel stored the CPU state
597 * condition: OLDMEM_BASE != NULL && is_kdump_kernel()
598 * This case does not exist for s390 anymore, setup_arch explicitly
599 * deactivates the elfcorehdr= kernel parameter
600 */
dump_available(void)601 static bool dump_available(void)
602 {
603 return oldmem_data.start || is_ipl_type_dump();
604 }
605
smp_save_dump_ipl_cpu(void)606 void __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
smp_save_dump_secondary_cpus(void)624 void __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
smp_cpu_set_polarization(int cpu,int val)664 void smp_cpu_set_polarization(int cpu, int val)
665 {
666 per_cpu(pcpu_devices, cpu).polarization = val;
667 }
668
smp_cpu_get_polarization(int cpu)669 int smp_cpu_get_polarization(int cpu)
670 {
671 return per_cpu(pcpu_devices, cpu).polarization;
672 }
673
smp_cpu_get_cpu_address(int cpu)674 int smp_cpu_get_cpu_address(int cpu)
675 {
676 return per_cpu(pcpu_devices, cpu).address;
677 }
678
smp_get_core_info(struct sclp_core_info * info,int early)679 static void __ref smp_get_core_info(struct sclp_core_info *info, int early)
680 {
681 static int use_sigp_detection;
682 int address;
683
684 if (use_sigp_detection || sclp_get_core_info(info, early)) {
685 use_sigp_detection = 1;
686 for (address = 0;
687 address < (SCLP_MAX_CORES << smp_cpu_mt_shift);
688 address += (1U << smp_cpu_mt_shift)) {
689 if (__pcpu_sigp_relax(address, SIGP_SENSE, 0) ==
690 SIGP_CC_NOT_OPERATIONAL)
691 continue;
692 info->core[info->configured].core_id =
693 address >> smp_cpu_mt_shift;
694 info->configured++;
695 }
696 info->combined = info->configured;
697 }
698 }
699
smp_add_core(struct sclp_core_entry * core,cpumask_t * avail,bool configured,bool early)700 static int smp_add_core(struct sclp_core_entry *core, cpumask_t *avail,
701 bool configured, bool early)
702 {
703 struct pcpu *pcpu;
704 int cpu, nr, i;
705 u16 address;
706
707 nr = 0;
708 if (sclp.has_core_type && core->type != boot_core_type)
709 return nr;
710 cpu = cpumask_first(avail);
711 address = core->core_id << smp_cpu_mt_shift;
712 for (i = 0; (i <= smp_cpu_mtid) && (cpu < nr_cpu_ids); i++) {
713 if (pcpu_find_address(cpu_present_mask, address + i))
714 continue;
715 pcpu = per_cpu_ptr(&pcpu_devices, cpu);
716 pcpu->address = address + i;
717 if (configured)
718 pcpu->state = CPU_STATE_CONFIGURED;
719 else
720 pcpu->state = CPU_STATE_STANDBY;
721 smp_cpu_set_polarization(cpu, POLARIZATION_UNKNOWN);
722 set_cpu_present(cpu, true);
723 if (!early && arch_register_cpu(cpu))
724 set_cpu_present(cpu, false);
725 else
726 nr++;
727 cpumask_clear_cpu(cpu, avail);
728 cpu = cpumask_next(cpu, avail);
729 }
730 return nr;
731 }
732
__smp_rescan_cpus(struct sclp_core_info * info,bool early)733 static int __smp_rescan_cpus(struct sclp_core_info *info, bool early)
734 {
735 struct sclp_core_entry *core;
736 static cpumask_t avail;
737 bool configured;
738 u16 core_id;
739 int nr, i;
740
741 cpus_read_lock();
742 mutex_lock(&smp_cpu_state_mutex);
743 nr = 0;
744 cpumask_xor(&avail, cpu_possible_mask, cpu_present_mask);
745 /*
746 * Add IPL core first (which got logical CPU number 0) to make sure
747 * that all SMT threads get subsequent logical CPU numbers.
748 */
749 if (early) {
750 core_id = per_cpu(pcpu_devices, 0).address >> smp_cpu_mt_shift;
751 for (i = 0; i < info->configured; i++) {
752 core = &info->core[i];
753 if (core->core_id == core_id) {
754 nr += smp_add_core(core, &avail, true, early);
755 break;
756 }
757 }
758 }
759 for (i = 0; i < info->combined; i++) {
760 configured = i < info->configured;
761 nr += smp_add_core(&info->core[i], &avail, configured, early);
762 }
763 mutex_unlock(&smp_cpu_state_mutex);
764 cpus_read_unlock();
765 return nr;
766 }
767
smp_detect_cpus(void)768 void __init smp_detect_cpus(void)
769 {
770 unsigned int cpu, mtid, c_cpus, s_cpus;
771 struct sclp_core_info *info;
772 u16 address;
773
774 /* Get CPU information */
775 info = memblock_alloc(sizeof(*info), 8);
776 if (!info)
777 panic("%s: Failed to allocate %zu bytes align=0x%x\n",
778 __func__, sizeof(*info), 8);
779 smp_get_core_info(info, 1);
780 /* Find boot CPU type */
781 if (sclp.has_core_type) {
782 address = stap();
783 for (cpu = 0; cpu < info->combined; cpu++)
784 if (info->core[cpu].core_id == address) {
785 /* The boot cpu dictates the cpu type. */
786 boot_core_type = info->core[cpu].type;
787 break;
788 }
789 if (cpu >= info->combined)
790 panic("Could not find boot CPU type");
791 }
792
793 /* Set multi-threading state for the current system */
794 mtid = boot_core_type ? sclp.mtid : sclp.mtid_cp;
795 mtid = (mtid < smp_max_threads) ? mtid : smp_max_threads - 1;
796 pcpu_set_smt(mtid);
797
798 /* Print number of CPUs */
799 c_cpus = s_cpus = 0;
800 for (cpu = 0; cpu < info->combined; cpu++) {
801 if (sclp.has_core_type &&
802 info->core[cpu].type != boot_core_type)
803 continue;
804 if (cpu < info->configured)
805 c_cpus += smp_cpu_mtid + 1;
806 else
807 s_cpus += smp_cpu_mtid + 1;
808 }
809 pr_info("%d configured CPUs, %d standby CPUs\n", c_cpus, s_cpus);
810 memblock_free(info, sizeof(*info));
811 }
812
813 /*
814 * Activate a secondary processor.
815 */
smp_start_secondary(void * cpuvoid)816 static void smp_start_secondary(void *cpuvoid)
817 {
818 struct lowcore *lc = get_lowcore();
819 int cpu = raw_smp_processor_id();
820
821 lc->last_update_clock = get_tod_clock();
822 lc->restart_stack = (unsigned long)restart_stack;
823 lc->restart_fn = (unsigned long)do_restart;
824 lc->restart_data = 0;
825 lc->restart_source = -1U;
826 lc->restart_flags = 0;
827 restore_access_regs(lc->access_regs_save_area);
828 cpu_init();
829 rcutree_report_cpu_starting(cpu);
830 init_cpu_timer();
831 vtime_init();
832 vdso_getcpu_init();
833 pfault_init();
834 cpumask_set_cpu(cpu, &cpu_setup_mask);
835 update_cpu_masks();
836 notify_cpu_starting(cpu);
837 if (topology_cpu_dedicated(cpu))
838 set_cpu_flag(CIF_DEDICATED_CPU);
839 else
840 clear_cpu_flag(CIF_DEDICATED_CPU);
841 set_cpu_online(cpu, true);
842 inc_irq_stat(CPU_RST);
843 local_irq_enable();
844 cpu_startup_entry(CPUHP_AP_ONLINE_IDLE);
845 }
846
847 /* Upping and downing of CPUs */
__cpu_up(unsigned int cpu,struct task_struct * tidle)848 int __cpu_up(unsigned int cpu, struct task_struct *tidle)
849 {
850 struct pcpu *pcpu = per_cpu_ptr(&pcpu_devices, cpu);
851 int rc;
852
853 if (pcpu->state != CPU_STATE_CONFIGURED)
854 return -EIO;
855 if (pcpu_sigp_retry(pcpu, SIGP_INITIAL_CPU_RESET, 0) !=
856 SIGP_CC_ORDER_CODE_ACCEPTED)
857 return -EIO;
858
859 rc = pcpu_alloc_lowcore(pcpu, cpu);
860 if (rc)
861 return rc;
862 /*
863 * Make sure global control register contents do not change
864 * until new CPU has initialized control registers.
865 */
866 system_ctlreg_lock();
867 pcpu_prepare_secondary(pcpu, cpu);
868 pcpu_attach_task(cpu, tidle);
869 pcpu_start_fn(cpu, smp_start_secondary, NULL);
870 /* Wait until cpu puts itself in the online & active maps */
871 while (!cpu_online(cpu))
872 cpu_relax();
873 system_ctlreg_unlock();
874 return 0;
875 }
876
877 static unsigned int setup_possible_cpus __initdata;
878
_setup_possible_cpus(char * s)879 static int __init _setup_possible_cpus(char *s)
880 {
881 get_option(&s, &setup_possible_cpus);
882 return 0;
883 }
884 early_param("possible_cpus", _setup_possible_cpus);
885
__cpu_disable(void)886 int __cpu_disable(void)
887 {
888 struct ctlreg cregs[16];
889 int cpu;
890
891 /* Handle possible pending IPIs */
892 smp_handle_ext_call();
893 cpu = smp_processor_id();
894 set_cpu_online(cpu, false);
895 cpumask_clear_cpu(cpu, &cpu_setup_mask);
896 update_cpu_masks();
897 /* Disable pseudo page faults on this cpu. */
898 pfault_fini();
899 /* Disable interrupt sources via control register. */
900 __local_ctl_store(0, 15, cregs);
901 cregs[0].val &= ~0x0000ee70UL; /* disable all external interrupts */
902 cregs[6].val &= ~0xff000000UL; /* disable all I/O interrupts */
903 cregs[14].val &= ~0x1f000000UL; /* disable most machine checks */
904 __local_ctl_load(0, 15, cregs);
905 clear_cpu_flag(CIF_NOHZ_DELAY);
906 return 0;
907 }
908
__cpu_die(unsigned int cpu)909 void __cpu_die(unsigned int cpu)
910 {
911 struct pcpu *pcpu;
912
913 /* Wait until target cpu is down */
914 pcpu = per_cpu_ptr(&pcpu_devices, cpu);
915 while (!pcpu_stopped(pcpu))
916 cpu_relax();
917 pcpu_free_lowcore(pcpu, cpu);
918 cpumask_clear_cpu(cpu, mm_cpumask(&init_mm));
919 cpumask_clear_cpu(cpu, &init_mm.context.cpu_attach_mask);
920 pcpu->flags = 0;
921 }
922
cpu_die(void)923 void __noreturn cpu_die(void)
924 {
925 idle_task_exit();
926 pcpu_sigp_retry(this_cpu_ptr(&pcpu_devices), SIGP_STOP, 0);
927 for (;;) ;
928 }
929
smp_fill_possible_mask(void)930 void __init smp_fill_possible_mask(void)
931 {
932 unsigned int possible, sclp_max, cpu;
933
934 sclp_max = max(sclp.mtid, sclp.mtid_cp) + 1;
935 sclp_max = min(smp_max_threads, sclp_max);
936 sclp_max = (sclp.max_cores * sclp_max) ?: nr_cpu_ids;
937 possible = setup_possible_cpus ?: nr_cpu_ids;
938 possible = min(possible, sclp_max);
939 for (cpu = 0; cpu < possible && cpu < nr_cpu_ids; cpu++)
940 set_cpu_possible(cpu, true);
941 }
942
smp_prepare_cpus(unsigned int max_cpus)943 void __init smp_prepare_cpus(unsigned int max_cpus)
944 {
945 if (register_external_irq(EXT_IRQ_EMERGENCY_SIG, do_ext_call_interrupt))
946 panic("Couldn't request external interrupt 0x1201");
947 system_ctl_set_bit(0, 14);
948 if (register_external_irq(EXT_IRQ_EXTERNAL_CALL, do_ext_call_interrupt))
949 panic("Couldn't request external interrupt 0x1202");
950 system_ctl_set_bit(0, 13);
951 smp_rescan_cpus(true);
952 }
953
smp_prepare_boot_cpu(void)954 void __init smp_prepare_boot_cpu(void)
955 {
956 struct lowcore *lc = get_lowcore();
957
958 WARN_ON(!cpu_present(0) || !cpu_online(0));
959 lc->percpu_offset = __per_cpu_offset[0];
960 ipl_pcpu = per_cpu_ptr(&pcpu_devices, 0);
961 ipl_pcpu->state = CPU_STATE_CONFIGURED;
962 lc->pcpu = (unsigned long)ipl_pcpu;
963 smp_cpu_set_polarization(0, POLARIZATION_UNKNOWN);
964 }
965
smp_setup_processor_id(void)966 void __init smp_setup_processor_id(void)
967 {
968 struct lowcore *lc = get_lowcore();
969
970 lc->cpu_nr = 0;
971 per_cpu(pcpu_devices, 0).address = stap();
972 lc->spinlock_lockval = arch_spin_lockval(0);
973 lc->spinlock_index = 0;
974 }
975
976 /*
977 * the frequency of the profiling timer can be changed
978 * by writing a multiplier value into /proc/profile.
979 *
980 * usually you want to run this on all CPUs ;)
981 */
setup_profiling_timer(unsigned int multiplier)982 int setup_profiling_timer(unsigned int multiplier)
983 {
984 return 0;
985 }
986
cpu_configure_show(struct device * dev,struct device_attribute * attr,char * buf)987 static ssize_t cpu_configure_show(struct device *dev,
988 struct device_attribute *attr, char *buf)
989 {
990 ssize_t count;
991
992 mutex_lock(&smp_cpu_state_mutex);
993 count = sprintf(buf, "%d\n", per_cpu(pcpu_devices, dev->id).state);
994 mutex_unlock(&smp_cpu_state_mutex);
995 return count;
996 }
997
cpu_configure_store(struct device * dev,struct device_attribute * attr,const char * buf,size_t count)998 static ssize_t cpu_configure_store(struct device *dev,
999 struct device_attribute *attr,
1000 const char *buf, size_t count)
1001 {
1002 struct pcpu *pcpu;
1003 int cpu, val, rc, i;
1004 char delim;
1005
1006 if (sscanf(buf, "%d %c", &val, &delim) != 1)
1007 return -EINVAL;
1008 if (val != 0 && val != 1)
1009 return -EINVAL;
1010 cpus_read_lock();
1011 mutex_lock(&smp_cpu_state_mutex);
1012 rc = -EBUSY;
1013 /* disallow configuration changes of online cpus */
1014 cpu = dev->id;
1015 cpu = smp_get_base_cpu(cpu);
1016 for (i = 0; i <= smp_cpu_mtid; i++)
1017 if (cpu_online(cpu + i))
1018 goto out;
1019 pcpu = per_cpu_ptr(&pcpu_devices, cpu);
1020 rc = 0;
1021 switch (val) {
1022 case 0:
1023 if (pcpu->state != CPU_STATE_CONFIGURED)
1024 break;
1025 rc = sclp_core_deconfigure(pcpu->address >> smp_cpu_mt_shift);
1026 if (rc)
1027 break;
1028 for (i = 0; i <= smp_cpu_mtid; i++) {
1029 if (cpu + i >= nr_cpu_ids || !cpu_present(cpu + i))
1030 continue;
1031 per_cpu(pcpu_devices, cpu + i).state = CPU_STATE_STANDBY;
1032 smp_cpu_set_polarization(cpu + i,
1033 POLARIZATION_UNKNOWN);
1034 }
1035 topology_expect_change();
1036 break;
1037 case 1:
1038 if (pcpu->state != CPU_STATE_STANDBY)
1039 break;
1040 rc = sclp_core_configure(pcpu->address >> smp_cpu_mt_shift);
1041 if (rc)
1042 break;
1043 for (i = 0; i <= smp_cpu_mtid; i++) {
1044 if (cpu + i >= nr_cpu_ids || !cpu_present(cpu + i))
1045 continue;
1046 per_cpu(pcpu_devices, cpu + i).state = CPU_STATE_CONFIGURED;
1047 smp_cpu_set_polarization(cpu + i,
1048 POLARIZATION_UNKNOWN);
1049 }
1050 topology_expect_change();
1051 break;
1052 default:
1053 break;
1054 }
1055 out:
1056 mutex_unlock(&smp_cpu_state_mutex);
1057 cpus_read_unlock();
1058 return rc ? rc : count;
1059 }
1060 static DEVICE_ATTR(configure, 0644, cpu_configure_show, cpu_configure_store);
1061
show_cpu_address(struct device * dev,struct device_attribute * attr,char * buf)1062 static ssize_t show_cpu_address(struct device *dev,
1063 struct device_attribute *attr, char *buf)
1064 {
1065 return sprintf(buf, "%d\n", per_cpu(pcpu_devices, dev->id).address);
1066 }
1067 static DEVICE_ATTR(address, 0444, show_cpu_address, NULL);
1068
1069 static struct attribute *cpu_common_attrs[] = {
1070 &dev_attr_configure.attr,
1071 &dev_attr_address.attr,
1072 NULL,
1073 };
1074
1075 static struct attribute_group cpu_common_attr_group = {
1076 .attrs = cpu_common_attrs,
1077 };
1078
1079 static struct attribute *cpu_online_attrs[] = {
1080 &dev_attr_idle_count.attr,
1081 &dev_attr_idle_time_us.attr,
1082 NULL,
1083 };
1084
1085 static struct attribute_group cpu_online_attr_group = {
1086 .attrs = cpu_online_attrs,
1087 };
1088
smp_cpu_online(unsigned int cpu)1089 static int smp_cpu_online(unsigned int cpu)
1090 {
1091 struct cpu *c = per_cpu_ptr(&cpu_devices, cpu);
1092
1093 return sysfs_create_group(&c->dev.kobj, &cpu_online_attr_group);
1094 }
1095
smp_cpu_pre_down(unsigned int cpu)1096 static int smp_cpu_pre_down(unsigned int cpu)
1097 {
1098 struct cpu *c = per_cpu_ptr(&cpu_devices, cpu);
1099
1100 sysfs_remove_group(&c->dev.kobj, &cpu_online_attr_group);
1101 return 0;
1102 }
1103
arch_cpu_is_hotpluggable(int cpu)1104 bool arch_cpu_is_hotpluggable(int cpu)
1105 {
1106 return !!cpu;
1107 }
1108
arch_register_cpu(int cpu)1109 int arch_register_cpu(int cpu)
1110 {
1111 struct cpu *c = per_cpu_ptr(&cpu_devices, cpu);
1112 int rc;
1113
1114 c->hotpluggable = arch_cpu_is_hotpluggable(cpu);
1115 rc = register_cpu(c, cpu);
1116 if (rc)
1117 goto out;
1118 rc = sysfs_create_group(&c->dev.kobj, &cpu_common_attr_group);
1119 if (rc)
1120 goto out_cpu;
1121 rc = topology_cpu_init(c);
1122 if (rc)
1123 goto out_topology;
1124 return 0;
1125
1126 out_topology:
1127 sysfs_remove_group(&c->dev.kobj, &cpu_common_attr_group);
1128 out_cpu:
1129 unregister_cpu(c);
1130 out:
1131 return rc;
1132 }
1133
smp_rescan_cpus(bool early)1134 int __ref smp_rescan_cpus(bool early)
1135 {
1136 struct sclp_core_info *info;
1137 int nr;
1138
1139 info = kzalloc(sizeof(*info), GFP_KERNEL);
1140 if (!info)
1141 return -ENOMEM;
1142 smp_get_core_info(info, 0);
1143 nr = __smp_rescan_cpus(info, early);
1144 kfree(info);
1145 if (nr)
1146 topology_schedule_update();
1147 return 0;
1148 }
1149
rescan_store(struct device * dev,struct device_attribute * attr,const char * buf,size_t count)1150 static ssize_t __ref rescan_store(struct device *dev,
1151 struct device_attribute *attr,
1152 const char *buf,
1153 size_t count)
1154 {
1155 int rc;
1156
1157 rc = lock_device_hotplug_sysfs();
1158 if (rc)
1159 return rc;
1160 rc = smp_rescan_cpus(false);
1161 unlock_device_hotplug();
1162 return rc ? rc : count;
1163 }
1164 static DEVICE_ATTR_WO(rescan);
1165
s390_smp_init(void)1166 static int __init s390_smp_init(void)
1167 {
1168 struct device *dev_root;
1169 int rc;
1170
1171 dev_root = bus_get_dev_root(&cpu_subsys);
1172 if (dev_root) {
1173 rc = device_create_file(dev_root, &dev_attr_rescan);
1174 put_device(dev_root);
1175 if (rc)
1176 return rc;
1177 }
1178 rc = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "s390/smp:online",
1179 smp_cpu_online, smp_cpu_pre_down);
1180 rc = rc <= 0 ? rc : 0;
1181 return rc;
1182 }
1183 subsys_initcall(s390_smp_init);
1184