xref: /openbmc/linux/arch/ia64/kernel/smpboot.c (revision 2208f39c)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * SMP boot-related support
4  *
5  * Copyright (C) 1998-2003, 2005 Hewlett-Packard Co
6  *	David Mosberger-Tang <davidm@hpl.hp.com>
7  * Copyright (C) 2001, 2004-2005 Intel Corp
8  * 	Rohit Seth <rohit.seth@intel.com>
9  * 	Suresh Siddha <suresh.b.siddha@intel.com>
10  * 	Gordon Jin <gordon.jin@intel.com>
11  *	Ashok Raj  <ashok.raj@intel.com>
12  *
13  * 01/05/16 Rohit Seth <rohit.seth@intel.com>	Moved SMP booting functions from smp.c to here.
14  * 01/04/27 David Mosberger <davidm@hpl.hp.com>	Added ITC synching code.
15  * 02/07/31 David Mosberger <davidm@hpl.hp.com>	Switch over to hotplug-CPU boot-sequence.
16  *						smp_boot_cpus()/smp_commence() is replaced by
17  *						smp_prepare_cpus()/__cpu_up()/smp_cpus_done().
18  * 04/06/21 Ashok Raj		<ashok.raj@intel.com> Added CPU Hotplug Support
19  * 04/12/26 Jin Gordon <gordon.jin@intel.com>
20  * 04/12/26 Rohit Seth <rohit.seth@intel.com>
21  *						Add multi-threading and multi-core detection
22  * 05/01/30 Suresh Siddha <suresh.b.siddha@intel.com>
23  *						Setup cpu_sibling_map and cpu_core_map
24  */
25 
26 #include <linux/module.h>
27 #include <linux/acpi.h>
28 #include <linux/memblock.h>
29 #include <linux/cpu.h>
30 #include <linux/delay.h>
31 #include <linux/init.h>
32 #include <linux/interrupt.h>
33 #include <linux/irq.h>
34 #include <linux/kernel.h>
35 #include <linux/kernel_stat.h>
36 #include <linux/mm.h>
37 #include <linux/notifier.h>
38 #include <linux/smp.h>
39 #include <linux/spinlock.h>
40 #include <linux/efi.h>
41 #include <linux/percpu.h>
42 #include <linux/bitops.h>
43 
44 #include <linux/atomic.h>
45 #include <asm/cache.h>
46 #include <asm/current.h>
47 #include <asm/delay.h>
48 #include <asm/io.h>
49 #include <asm/irq.h>
50 #include <asm/mca.h>
51 #include <asm/page.h>
52 #include <asm/processor.h>
53 #include <asm/ptrace.h>
54 #include <asm/sal.h>
55 #include <asm/tlbflush.h>
56 #include <asm/unistd.h>
57 
58 #define SMP_DEBUG 0
59 
60 #if SMP_DEBUG
61 #define Dprintk(x...)  printk(x)
62 #else
63 #define Dprintk(x...)
64 #endif
65 
66 #ifdef CONFIG_HOTPLUG_CPU
67 #ifdef CONFIG_PERMIT_BSP_REMOVE
68 #define bsp_remove_ok	1
69 #else
70 #define bsp_remove_ok	0
71 #endif
72 
73 /*
74  * Global array allocated for NR_CPUS at boot time
75  */
76 struct sal_to_os_boot sal_boot_rendez_state[NR_CPUS];
77 
78 /*
79  * start_ap in head.S uses this to store current booting cpu
80  * info.
81  */
82 struct sal_to_os_boot *sal_state_for_booting_cpu = &sal_boot_rendez_state[0];
83 
84 #define set_brendez_area(x) (sal_state_for_booting_cpu = &sal_boot_rendez_state[(x)]);
85 
86 #else
87 #define set_brendez_area(x)
88 #endif
89 
90 
91 /*
92  * ITC synchronization related stuff:
93  */
94 #define MASTER	(0)
95 #define SLAVE	(SMP_CACHE_BYTES/8)
96 
97 #define NUM_ROUNDS	64	/* magic value */
98 #define NUM_ITERS	5	/* likewise */
99 
100 static DEFINE_SPINLOCK(itc_sync_lock);
101 static volatile unsigned long go[SLAVE + 1];
102 
103 #define DEBUG_ITC_SYNC	0
104 
105 extern void start_ap (void);
106 extern unsigned long ia64_iobase;
107 
108 struct task_struct *task_for_booting_cpu;
109 
110 /*
111  * State for each CPU
112  */
113 DEFINE_PER_CPU(int, cpu_state);
114 
115 cpumask_t cpu_core_map[NR_CPUS] __cacheline_aligned;
116 EXPORT_SYMBOL(cpu_core_map);
117 DEFINE_PER_CPU_SHARED_ALIGNED(cpumask_t, cpu_sibling_map);
118 EXPORT_PER_CPU_SYMBOL(cpu_sibling_map);
119 
120 int smp_num_siblings = 1;
121 
122 /* which logical CPU number maps to which CPU (physical APIC ID) */
123 volatile int ia64_cpu_to_sapicid[NR_CPUS];
124 EXPORT_SYMBOL(ia64_cpu_to_sapicid);
125 
126 static cpumask_t cpu_callin_map;
127 
128 struct smp_boot_data smp_boot_data __initdata;
129 
130 unsigned long ap_wakeup_vector = -1; /* External Int use to wakeup APs */
131 
132 char __initdata no_int_routing;
133 
134 unsigned char smp_int_redirect; /* are INT and IPI redirectable by the chipset? */
135 
136 #ifdef CONFIG_FORCE_CPEI_RETARGET
137 #define CPEI_OVERRIDE_DEFAULT	(1)
138 #else
139 #define CPEI_OVERRIDE_DEFAULT	(0)
140 #endif
141 
142 unsigned int force_cpei_retarget = CPEI_OVERRIDE_DEFAULT;
143 
144 static int __init
145 cmdl_force_cpei(char *str)
146 {
147 	int value=0;
148 
149 	get_option (&str, &value);
150 	force_cpei_retarget = value;
151 
152 	return 1;
153 }
154 
155 __setup("force_cpei=", cmdl_force_cpei);
156 
157 static int __init
158 nointroute (char *str)
159 {
160 	no_int_routing = 1;
161 	printk ("no_int_routing on\n");
162 	return 1;
163 }
164 
165 __setup("nointroute", nointroute);
166 
167 static void fix_b0_for_bsp(void)
168 {
169 #ifdef CONFIG_HOTPLUG_CPU
170 	int cpuid;
171 	static int fix_bsp_b0 = 1;
172 
173 	cpuid = smp_processor_id();
174 
175 	/*
176 	 * Cache the b0 value on the first AP that comes up
177 	 */
178 	if (!(fix_bsp_b0 && cpuid))
179 		return;
180 
181 	sal_boot_rendez_state[0].br[0] = sal_boot_rendez_state[cpuid].br[0];
182 	printk ("Fixed BSP b0 value from CPU %d\n", cpuid);
183 
184 	fix_bsp_b0 = 0;
185 #endif
186 }
187 
188 void
189 sync_master (void *arg)
190 {
191 	unsigned long flags, i;
192 
193 	go[MASTER] = 0;
194 
195 	local_irq_save(flags);
196 	{
197 		for (i = 0; i < NUM_ROUNDS*NUM_ITERS; ++i) {
198 			while (!go[MASTER])
199 				cpu_relax();
200 			go[MASTER] = 0;
201 			go[SLAVE] = ia64_get_itc();
202 		}
203 	}
204 	local_irq_restore(flags);
205 }
206 
207 /*
208  * Return the number of cycles by which our itc differs from the itc on the master
209  * (time-keeper) CPU.  A positive number indicates our itc is ahead of the master,
210  * negative that it is behind.
211  */
212 static inline long
213 get_delta (long *rt, long *master)
214 {
215 	unsigned long best_t0 = 0, best_t1 = ~0UL, best_tm = 0;
216 	unsigned long tcenter, t0, t1, tm;
217 	long i;
218 
219 	for (i = 0; i < NUM_ITERS; ++i) {
220 		t0 = ia64_get_itc();
221 		go[MASTER] = 1;
222 		while (!(tm = go[SLAVE]))
223 			cpu_relax();
224 		go[SLAVE] = 0;
225 		t1 = ia64_get_itc();
226 
227 		if (t1 - t0 < best_t1 - best_t0)
228 			best_t0 = t0, best_t1 = t1, best_tm = tm;
229 	}
230 
231 	*rt = best_t1 - best_t0;
232 	*master = best_tm - best_t0;
233 
234 	/* average best_t0 and best_t1 without overflow: */
235 	tcenter = (best_t0/2 + best_t1/2);
236 	if (best_t0 % 2 + best_t1 % 2 == 2)
237 		++tcenter;
238 	return tcenter - best_tm;
239 }
240 
241 /*
242  * Synchronize ar.itc of the current (slave) CPU with the ar.itc of the MASTER CPU
243  * (normally the time-keeper CPU).  We use a closed loop to eliminate the possibility of
244  * unaccounted-for errors (such as getting a machine check in the middle of a calibration
245  * step).  The basic idea is for the slave to ask the master what itc value it has and to
246  * read its own itc before and after the master responds.  Each iteration gives us three
247  * timestamps:
248  *
249  *	slave		master
250  *
251  *	t0 ---\
252  *             ---\
253  *		   --->
254  *			tm
255  *		   /---
256  *	       /---
257  *	t1 <---
258  *
259  *
260  * The goal is to adjust the slave's ar.itc such that tm falls exactly half-way between t0
261  * and t1.  If we achieve this, the clocks are synchronized provided the interconnect
262  * between the slave and the master is symmetric.  Even if the interconnect were
263  * asymmetric, we would still know that the synchronization error is smaller than the
264  * roundtrip latency (t0 - t1).
265  *
266  * When the interconnect is quiet and symmetric, this lets us synchronize the itc to
267  * within one or two cycles.  However, we can only *guarantee* that the synchronization is
268  * accurate to within a round-trip time, which is typically in the range of several
269  * hundred cycles (e.g., ~500 cycles).  In practice, this means that the itc's are usually
270  * almost perfectly synchronized, but we shouldn't assume that the accuracy is much better
271  * than half a micro second or so.
272  */
273 void
274 ia64_sync_itc (unsigned int master)
275 {
276 	long i, delta, adj, adjust_latency = 0, done = 0;
277 	unsigned long flags, rt, master_time_stamp, bound;
278 #if DEBUG_ITC_SYNC
279 	struct {
280 		long rt;	/* roundtrip time */
281 		long master;	/* master's timestamp */
282 		long diff;	/* difference between midpoint and master's timestamp */
283 		long lat;	/* estimate of itc adjustment latency */
284 	} t[NUM_ROUNDS];
285 #endif
286 
287 	/*
288 	 * Make sure local timer ticks are disabled while we sync.  If
289 	 * they were enabled, we'd have to worry about nasty issues
290 	 * like setting the ITC ahead of (or a long time before) the
291 	 * next scheduled tick.
292 	 */
293 	BUG_ON((ia64_get_itv() & (1 << 16)) == 0);
294 
295 	go[MASTER] = 1;
296 
297 	if (smp_call_function_single(master, sync_master, NULL, 0) < 0) {
298 		printk(KERN_ERR "sync_itc: failed to get attention of CPU %u!\n", master);
299 		return;
300 	}
301 
302 	while (go[MASTER])
303 		cpu_relax();	/* wait for master to be ready */
304 
305 	spin_lock_irqsave(&itc_sync_lock, flags);
306 	{
307 		for (i = 0; i < NUM_ROUNDS; ++i) {
308 			delta = get_delta(&rt, &master_time_stamp);
309 			if (delta == 0) {
310 				done = 1;	/* let's lock on to this... */
311 				bound = rt;
312 			}
313 
314 			if (!done) {
315 				if (i > 0) {
316 					adjust_latency += -delta;
317 					adj = -delta + adjust_latency/4;
318 				} else
319 					adj = -delta;
320 
321 				ia64_set_itc(ia64_get_itc() + adj);
322 			}
323 #if DEBUG_ITC_SYNC
324 			t[i].rt = rt;
325 			t[i].master = master_time_stamp;
326 			t[i].diff = delta;
327 			t[i].lat = adjust_latency/4;
328 #endif
329 		}
330 	}
331 	spin_unlock_irqrestore(&itc_sync_lock, flags);
332 
333 #if DEBUG_ITC_SYNC
334 	for (i = 0; i < NUM_ROUNDS; ++i)
335 		printk("rt=%5ld master=%5ld diff=%5ld adjlat=%5ld\n",
336 		       t[i].rt, t[i].master, t[i].diff, t[i].lat);
337 #endif
338 
339 	printk(KERN_INFO "CPU %d: synchronized ITC with CPU %u (last diff %ld cycles, "
340 	       "maxerr %lu cycles)\n", smp_processor_id(), master, delta, rt);
341 }
342 
343 /*
344  * Ideally sets up per-cpu profiling hooks.  Doesn't do much now...
345  */
346 static inline void smp_setup_percpu_timer(void)
347 {
348 }
349 
350 static void
351 smp_callin (void)
352 {
353 	int cpuid, phys_id, itc_master;
354 	struct cpuinfo_ia64 *last_cpuinfo, *this_cpuinfo;
355 	extern void ia64_init_itm(void);
356 	extern volatile int time_keeper_id;
357 
358 	cpuid = smp_processor_id();
359 	phys_id = hard_smp_processor_id();
360 	itc_master = time_keeper_id;
361 
362 	if (cpu_online(cpuid)) {
363 		printk(KERN_ERR "huh, phys CPU#0x%x, CPU#0x%x already present??\n",
364 		       phys_id, cpuid);
365 		BUG();
366 	}
367 
368 	fix_b0_for_bsp();
369 
370 	/*
371 	 * numa_node_id() works after this.
372 	 */
373 	set_numa_node(cpu_to_node_map[cpuid]);
374 	set_numa_mem(local_memory_node(cpu_to_node_map[cpuid]));
375 
376 	spin_lock(&vector_lock);
377 	/* Setup the per cpu irq handling data structures */
378 	__setup_vector_irq(cpuid);
379 	notify_cpu_starting(cpuid);
380 	set_cpu_online(cpuid, true);
381 	per_cpu(cpu_state, cpuid) = CPU_ONLINE;
382 	spin_unlock(&vector_lock);
383 
384 	smp_setup_percpu_timer();
385 
386 	ia64_mca_cmc_vector_setup();	/* Setup vector on AP */
387 
388 	local_irq_enable();
389 
390 	if (!(sal_platform_features & IA64_SAL_PLATFORM_FEATURE_ITC_DRIFT)) {
391 		/*
392 		 * Synchronize the ITC with the BP.  Need to do this after irqs are
393 		 * enabled because ia64_sync_itc() calls smp_call_function_single(), which
394 		 * calls spin_unlock_bh(), which calls spin_unlock_bh(), which calls
395 		 * local_bh_enable(), which bugs out if irqs are not enabled...
396 		 */
397 		Dprintk("Going to syncup ITC with ITC Master.\n");
398 		ia64_sync_itc(itc_master);
399 	}
400 
401 	/*
402 	 * Get our bogomips.
403 	 */
404 	ia64_init_itm();
405 
406 	/*
407 	 * Delay calibration can be skipped if new processor is identical to the
408 	 * previous processor.
409 	 */
410 	last_cpuinfo = cpu_data(cpuid - 1);
411 	this_cpuinfo = local_cpu_data;
412 	if (last_cpuinfo->itc_freq != this_cpuinfo->itc_freq ||
413 	    last_cpuinfo->proc_freq != this_cpuinfo->proc_freq ||
414 	    last_cpuinfo->features != this_cpuinfo->features ||
415 	    last_cpuinfo->revision != this_cpuinfo->revision ||
416 	    last_cpuinfo->family != this_cpuinfo->family ||
417 	    last_cpuinfo->archrev != this_cpuinfo->archrev ||
418 	    last_cpuinfo->model != this_cpuinfo->model)
419 		calibrate_delay();
420 	local_cpu_data->loops_per_jiffy = loops_per_jiffy;
421 
422 	/*
423 	 * Allow the master to continue.
424 	 */
425 	cpumask_set_cpu(cpuid, &cpu_callin_map);
426 	Dprintk("Stack on CPU %d at about %p\n",cpuid, &cpuid);
427 }
428 
429 
430 /*
431  * Activate a secondary processor.  head.S calls this.
432  */
433 int
434 start_secondary (void *unused)
435 {
436 	/* Early console may use I/O ports */
437 	ia64_set_kr(IA64_KR_IO_BASE, __pa(ia64_iobase));
438 #ifndef CONFIG_PRINTK_TIME
439 	Dprintk("start_secondary: starting CPU 0x%x\n", hard_smp_processor_id());
440 #endif
441 	efi_map_pal_code();
442 	cpu_init();
443 	preempt_disable();
444 	smp_callin();
445 
446 	cpu_startup_entry(CPUHP_AP_ONLINE_IDLE);
447 	return 0;
448 }
449 
450 static int
451 do_boot_cpu (int sapicid, int cpu, struct task_struct *idle)
452 {
453 	int timeout;
454 
455 	task_for_booting_cpu = idle;
456 	Dprintk("Sending wakeup vector %lu to AP 0x%x/0x%x.\n", ap_wakeup_vector, cpu, sapicid);
457 
458 	set_brendez_area(cpu);
459 	ia64_send_ipi(cpu, ap_wakeup_vector, IA64_IPI_DM_INT, 0);
460 
461 	/*
462 	 * Wait 10s total for the AP to start
463 	 */
464 	Dprintk("Waiting on callin_map ...");
465 	for (timeout = 0; timeout < 100000; timeout++) {
466 		if (cpumask_test_cpu(cpu, &cpu_callin_map))
467 			break;  /* It has booted */
468 		barrier(); /* Make sure we re-read cpu_callin_map */
469 		udelay(100);
470 	}
471 	Dprintk("\n");
472 
473 	if (!cpumask_test_cpu(cpu, &cpu_callin_map)) {
474 		printk(KERN_ERR "Processor 0x%x/0x%x is stuck.\n", cpu, sapicid);
475 		ia64_cpu_to_sapicid[cpu] = -1;
476 		set_cpu_online(cpu, false);  /* was set in smp_callin() */
477 		return -EINVAL;
478 	}
479 	return 0;
480 }
481 
482 static int __init
483 decay (char *str)
484 {
485 	int ticks;
486 	get_option (&str, &ticks);
487 	return 1;
488 }
489 
490 __setup("decay=", decay);
491 
492 /*
493  * Initialize the logical CPU number to SAPICID mapping
494  */
495 void __init
496 smp_build_cpu_map (void)
497 {
498 	int sapicid, cpu, i;
499 	int boot_cpu_id = hard_smp_processor_id();
500 
501 	for (cpu = 0; cpu < NR_CPUS; cpu++) {
502 		ia64_cpu_to_sapicid[cpu] = -1;
503 	}
504 
505 	ia64_cpu_to_sapicid[0] = boot_cpu_id;
506 	init_cpu_present(cpumask_of(0));
507 	set_cpu_possible(0, true);
508 	for (cpu = 1, i = 0; i < smp_boot_data.cpu_count; i++) {
509 		sapicid = smp_boot_data.cpu_phys_id[i];
510 		if (sapicid == boot_cpu_id)
511 			continue;
512 		set_cpu_present(cpu, true);
513 		set_cpu_possible(cpu, true);
514 		ia64_cpu_to_sapicid[cpu] = sapicid;
515 		cpu++;
516 	}
517 }
518 
519 /*
520  * Cycle through the APs sending Wakeup IPIs to boot each.
521  */
522 void __init
523 smp_prepare_cpus (unsigned int max_cpus)
524 {
525 	int boot_cpu_id = hard_smp_processor_id();
526 
527 	/*
528 	 * Initialize the per-CPU profiling counter/multiplier
529 	 */
530 
531 	smp_setup_percpu_timer();
532 
533 	cpumask_set_cpu(0, &cpu_callin_map);
534 
535 	local_cpu_data->loops_per_jiffy = loops_per_jiffy;
536 	ia64_cpu_to_sapicid[0] = boot_cpu_id;
537 
538 	printk(KERN_INFO "Boot processor id 0x%x/0x%x\n", 0, boot_cpu_id);
539 
540 	current_thread_info()->cpu = 0;
541 
542 	/*
543 	 * If SMP should be disabled, then really disable it!
544 	 */
545 	if (!max_cpus) {
546 		printk(KERN_INFO "SMP mode deactivated.\n");
547 		init_cpu_online(cpumask_of(0));
548 		init_cpu_present(cpumask_of(0));
549 		init_cpu_possible(cpumask_of(0));
550 		return;
551 	}
552 }
553 
554 void smp_prepare_boot_cpu(void)
555 {
556 	set_cpu_online(smp_processor_id(), true);
557 	cpumask_set_cpu(smp_processor_id(), &cpu_callin_map);
558 	set_numa_node(cpu_to_node_map[smp_processor_id()]);
559 	per_cpu(cpu_state, smp_processor_id()) = CPU_ONLINE;
560 }
561 
562 #ifdef CONFIG_HOTPLUG_CPU
563 static inline void
564 clear_cpu_sibling_map(int cpu)
565 {
566 	int i;
567 
568 	for_each_cpu(i, &per_cpu(cpu_sibling_map, cpu))
569 		cpumask_clear_cpu(cpu, &per_cpu(cpu_sibling_map, i));
570 	for_each_cpu(i, &cpu_core_map[cpu])
571 		cpumask_clear_cpu(cpu, &cpu_core_map[i]);
572 
573 	per_cpu(cpu_sibling_map, cpu) = cpu_core_map[cpu] = CPU_MASK_NONE;
574 }
575 
576 static void
577 remove_siblinginfo(int cpu)
578 {
579 	int last = 0;
580 
581 	if (cpu_data(cpu)->threads_per_core == 1 &&
582 	    cpu_data(cpu)->cores_per_socket == 1) {
583 		cpumask_clear_cpu(cpu, &cpu_core_map[cpu]);
584 		cpumask_clear_cpu(cpu, &per_cpu(cpu_sibling_map, cpu));
585 		return;
586 	}
587 
588 	last = (cpumask_weight(&cpu_core_map[cpu]) == 1 ? 1 : 0);
589 
590 	/* remove it from all sibling map's */
591 	clear_cpu_sibling_map(cpu);
592 }
593 
594 extern void fixup_irqs(void);
595 
596 int migrate_platform_irqs(unsigned int cpu)
597 {
598 	int new_cpei_cpu;
599 	struct irq_data *data = NULL;
600 	const struct cpumask *mask;
601 	int 		retval = 0;
602 
603 	/*
604 	 * dont permit CPEI target to removed.
605 	 */
606 	if (cpe_vector > 0 && is_cpu_cpei_target(cpu)) {
607 		printk ("CPU (%d) is CPEI Target\n", cpu);
608 		if (can_cpei_retarget()) {
609 			/*
610 			 * Now re-target the CPEI to a different processor
611 			 */
612 			new_cpei_cpu = cpumask_any(cpu_online_mask);
613 			mask = cpumask_of(new_cpei_cpu);
614 			set_cpei_target_cpu(new_cpei_cpu);
615 			data = irq_get_irq_data(ia64_cpe_irq);
616 			/*
617 			 * Switch for now, immediately, we need to do fake intr
618 			 * as other interrupts, but need to study CPEI behaviour with
619 			 * polling before making changes.
620 			 */
621 			if (data && data->chip) {
622 				data->chip->irq_disable(data);
623 				data->chip->irq_set_affinity(data, mask, false);
624 				data->chip->irq_enable(data);
625 				printk ("Re-targeting CPEI to cpu %d\n", new_cpei_cpu);
626 			}
627 		}
628 		if (!data) {
629 			printk ("Unable to retarget CPEI, offline cpu [%d] failed\n", cpu);
630 			retval = -EBUSY;
631 		}
632 	}
633 	return retval;
634 }
635 
636 /* must be called with cpucontrol mutex held */
637 int __cpu_disable(void)
638 {
639 	int cpu = smp_processor_id();
640 
641 	/*
642 	 * dont permit boot processor for now
643 	 */
644 	if (cpu == 0 && !bsp_remove_ok) {
645 		printk ("Your platform does not support removal of BSP\n");
646 		return (-EBUSY);
647 	}
648 
649 	set_cpu_online(cpu, false);
650 
651 	if (migrate_platform_irqs(cpu)) {
652 		set_cpu_online(cpu, true);
653 		return -EBUSY;
654 	}
655 
656 	remove_siblinginfo(cpu);
657 	fixup_irqs();
658 	local_flush_tlb_all();
659 	cpumask_clear_cpu(cpu, &cpu_callin_map);
660 	return 0;
661 }
662 
663 void __cpu_die(unsigned int cpu)
664 {
665 	unsigned int i;
666 
667 	for (i = 0; i < 100; i++) {
668 		/* They ack this in play_dead by setting CPU_DEAD */
669 		if (per_cpu(cpu_state, cpu) == CPU_DEAD)
670 		{
671 			printk ("CPU %d is now offline\n", cpu);
672 			return;
673 		}
674 		msleep(100);
675 	}
676  	printk(KERN_ERR "CPU %u didn't die...\n", cpu);
677 }
678 #endif /* CONFIG_HOTPLUG_CPU */
679 
680 void
681 smp_cpus_done (unsigned int dummy)
682 {
683 	int cpu;
684 	unsigned long bogosum = 0;
685 
686 	/*
687 	 * Allow the user to impress friends.
688 	 */
689 
690 	for_each_online_cpu(cpu) {
691 		bogosum += cpu_data(cpu)->loops_per_jiffy;
692 	}
693 
694 	printk(KERN_INFO "Total of %d processors activated (%lu.%02lu BogoMIPS).\n",
695 	       (int)num_online_cpus(), bogosum/(500000/HZ), (bogosum/(5000/HZ))%100);
696 }
697 
698 static inline void set_cpu_sibling_map(int cpu)
699 {
700 	int i;
701 
702 	for_each_online_cpu(i) {
703 		if ((cpu_data(cpu)->socket_id == cpu_data(i)->socket_id)) {
704 			cpumask_set_cpu(i, &cpu_core_map[cpu]);
705 			cpumask_set_cpu(cpu, &cpu_core_map[i]);
706 			if (cpu_data(cpu)->core_id == cpu_data(i)->core_id) {
707 				cpumask_set_cpu(i,
708 						&per_cpu(cpu_sibling_map, cpu));
709 				cpumask_set_cpu(cpu,
710 						&per_cpu(cpu_sibling_map, i));
711 			}
712 		}
713 	}
714 }
715 
716 int
717 __cpu_up(unsigned int cpu, struct task_struct *tidle)
718 {
719 	int ret;
720 	int sapicid;
721 
722 	sapicid = ia64_cpu_to_sapicid[cpu];
723 	if (sapicid == -1)
724 		return -EINVAL;
725 
726 	/*
727 	 * Already booted cpu? not valid anymore since we dont
728 	 * do idle loop tightspin anymore.
729 	 */
730 	if (cpumask_test_cpu(cpu, &cpu_callin_map))
731 		return -EINVAL;
732 
733 	per_cpu(cpu_state, cpu) = CPU_UP_PREPARE;
734 	/* Processor goes to start_secondary(), sets online flag */
735 	ret = do_boot_cpu(sapicid, cpu, tidle);
736 	if (ret < 0)
737 		return ret;
738 
739 	if (cpu_data(cpu)->threads_per_core == 1 &&
740 	    cpu_data(cpu)->cores_per_socket == 1) {
741 		cpumask_set_cpu(cpu, &per_cpu(cpu_sibling_map, cpu));
742 		cpumask_set_cpu(cpu, &cpu_core_map[cpu]);
743 		return 0;
744 	}
745 
746 	set_cpu_sibling_map(cpu);
747 
748 	return 0;
749 }
750 
751 /*
752  * Assume that CPUs have been discovered by some platform-dependent interface.  For
753  * SoftSDV/Lion, that would be ACPI.
754  *
755  * Setup of the IPI irq handler is done in irq.c:init_IRQ_SMP().
756  */
757 void __init
758 init_smp_config(void)
759 {
760 	struct fptr {
761 		unsigned long fp;
762 		unsigned long gp;
763 	} *ap_startup;
764 	long sal_ret;
765 
766 	/* Tell SAL where to drop the APs.  */
767 	ap_startup = (struct fptr *) start_ap;
768 	sal_ret = ia64_sal_set_vectors(SAL_VECTOR_OS_BOOT_RENDEZ,
769 				       ia64_tpa(ap_startup->fp), ia64_tpa(ap_startup->gp), 0, 0, 0, 0);
770 	if (sal_ret < 0)
771 		printk(KERN_ERR "SMP: Can't set SAL AP Boot Rendezvous: %s\n",
772 		       ia64_sal_strerror(sal_ret));
773 }
774 
775 /*
776  * identify_siblings(cpu) gets called from identify_cpu. This populates the
777  * information related to logical execution units in per_cpu_data structure.
778  */
779 void identify_siblings(struct cpuinfo_ia64 *c)
780 {
781 	long status;
782 	u16 pltid;
783 	pal_logical_to_physical_t info;
784 
785 	status = ia64_pal_logical_to_phys(-1, &info);
786 	if (status != PAL_STATUS_SUCCESS) {
787 		if (status != PAL_STATUS_UNIMPLEMENTED) {
788 			printk(KERN_ERR
789 				"ia64_pal_logical_to_phys failed with %ld\n",
790 				status);
791 			return;
792 		}
793 
794 		info.overview_ppid = 0;
795 		info.overview_cpp  = 1;
796 		info.overview_tpc  = 1;
797 	}
798 
799 	status = ia64_sal_physical_id_info(&pltid);
800 	if (status != PAL_STATUS_SUCCESS) {
801 		if (status != PAL_STATUS_UNIMPLEMENTED)
802 			printk(KERN_ERR
803 				"ia64_sal_pltid failed with %ld\n",
804 				status);
805 		return;
806 	}
807 
808 	c->socket_id =  (pltid << 8) | info.overview_ppid;
809 
810 	if (info.overview_cpp == 1 && info.overview_tpc == 1)
811 		return;
812 
813 	c->cores_per_socket = info.overview_cpp;
814 	c->threads_per_core = info.overview_tpc;
815 	c->num_log = info.overview_num_log;
816 
817 	c->core_id = info.log1_cid;
818 	c->thread_id = info.log1_tid;
819 }
820 
821 /*
822  * returns non zero, if multi-threading is enabled
823  * on at least one physical package. Due to hotplug cpu
824  * and (maxcpus=), all threads may not necessarily be enabled
825  * even though the processor supports multi-threading.
826  */
827 int is_multithreading_enabled(void)
828 {
829 	int i, j;
830 
831 	for_each_present_cpu(i) {
832 		for_each_present_cpu(j) {
833 			if (j == i)
834 				continue;
835 			if ((cpu_data(j)->socket_id == cpu_data(i)->socket_id)) {
836 				if (cpu_data(j)->core_id == cpu_data(i)->core_id)
837 					return 1;
838 			}
839 		}
840 	}
841 	return 0;
842 }
843 EXPORT_SYMBOL_GPL(is_multithreading_enabled);
844