xref: /openbmc/linux/arch/ia64/kernel/smpboot.c (revision 6f4eaea2)
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/efi.h>
49 #include <asm/io.h>
50 #include <asm/irq.h>
51 #include <asm/mca.h>
52 #include <asm/page.h>
53 #include <asm/processor.h>
54 #include <asm/ptrace.h>
55 #include <asm/sal.h>
56 #include <asm/tlbflush.h>
57 #include <asm/unistd.h>
58 
59 #define SMP_DEBUG 0
60 
61 #if SMP_DEBUG
62 #define Dprintk(x...)  printk(x)
63 #else
64 #define Dprintk(x...)
65 #endif
66 
67 #ifdef CONFIG_HOTPLUG_CPU
68 #ifdef CONFIG_PERMIT_BSP_REMOVE
69 #define bsp_remove_ok	1
70 #else
71 #define bsp_remove_ok	0
72 #endif
73 
74 /*
75  * Global array allocated for NR_CPUS at boot time
76  */
77 struct sal_to_os_boot sal_boot_rendez_state[NR_CPUS];
78 
79 /*
80  * start_ap in head.S uses this to store current booting cpu
81  * info.
82  */
83 struct sal_to_os_boot *sal_state_for_booting_cpu = &sal_boot_rendez_state[0];
84 
85 #define set_brendez_area(x) (sal_state_for_booting_cpu = &sal_boot_rendez_state[(x)]);
86 
87 #else
88 #define set_brendez_area(x)
89 #endif
90 
91 
92 /*
93  * ITC synchronization related stuff:
94  */
95 #define MASTER	(0)
96 #define SLAVE	(SMP_CACHE_BYTES/8)
97 
98 #define NUM_ROUNDS	64	/* magic value */
99 #define NUM_ITERS	5	/* likewise */
100 
101 static DEFINE_SPINLOCK(itc_sync_lock);
102 static volatile unsigned long go[SLAVE + 1];
103 
104 #define DEBUG_ITC_SYNC	0
105 
106 extern void start_ap (void);
107 extern unsigned long ia64_iobase;
108 
109 struct task_struct *task_for_booting_cpu;
110 
111 /*
112  * State for each CPU
113  */
114 DEFINE_PER_CPU(int, cpu_state);
115 
116 cpumask_t cpu_core_map[NR_CPUS] __cacheline_aligned;
117 EXPORT_SYMBOL(cpu_core_map);
118 DEFINE_PER_CPU_SHARED_ALIGNED(cpumask_t, cpu_sibling_map);
119 EXPORT_PER_CPU_SYMBOL(cpu_sibling_map);
120 
121 int smp_num_siblings = 1;
122 
123 /* which logical CPU number maps to which CPU (physical APIC ID) */
124 volatile int ia64_cpu_to_sapicid[NR_CPUS];
125 EXPORT_SYMBOL(ia64_cpu_to_sapicid);
126 
127 static cpumask_t cpu_callin_map;
128 
129 struct smp_boot_data smp_boot_data __initdata;
130 
131 unsigned long ap_wakeup_vector = -1; /* External Int use to wakeup APs */
132 
133 char __initdata no_int_routing;
134 
135 unsigned char smp_int_redirect; /* are INT and IPI redirectable by the chipset? */
136 
137 #ifdef CONFIG_FORCE_CPEI_RETARGET
138 #define CPEI_OVERRIDE_DEFAULT	(1)
139 #else
140 #define CPEI_OVERRIDE_DEFAULT	(0)
141 #endif
142 
143 unsigned int force_cpei_retarget = CPEI_OVERRIDE_DEFAULT;
144 
145 static int __init
146 cmdl_force_cpei(char *str)
147 {
148 	int value=0;
149 
150 	get_option (&str, &value);
151 	force_cpei_retarget = value;
152 
153 	return 1;
154 }
155 
156 __setup("force_cpei=", cmdl_force_cpei);
157 
158 static int __init
159 nointroute (char *str)
160 {
161 	no_int_routing = 1;
162 	printk ("no_int_routing on\n");
163 	return 1;
164 }
165 
166 __setup("nointroute", nointroute);
167 
168 static void fix_b0_for_bsp(void)
169 {
170 #ifdef CONFIG_HOTPLUG_CPU
171 	int cpuid;
172 	static int fix_bsp_b0 = 1;
173 
174 	cpuid = smp_processor_id();
175 
176 	/*
177 	 * Cache the b0 value on the first AP that comes up
178 	 */
179 	if (!(fix_bsp_b0 && cpuid))
180 		return;
181 
182 	sal_boot_rendez_state[0].br[0] = sal_boot_rendez_state[cpuid].br[0];
183 	printk ("Fixed BSP b0 value from CPU %d\n", cpuid);
184 
185 	fix_bsp_b0 = 0;
186 #endif
187 }
188 
189 void
190 sync_master (void *arg)
191 {
192 	unsigned long flags, i;
193 
194 	go[MASTER] = 0;
195 
196 	local_irq_save(flags);
197 	{
198 		for (i = 0; i < NUM_ROUNDS*NUM_ITERS; ++i) {
199 			while (!go[MASTER])
200 				cpu_relax();
201 			go[MASTER] = 0;
202 			go[SLAVE] = ia64_get_itc();
203 		}
204 	}
205 	local_irq_restore(flags);
206 }
207 
208 /*
209  * Return the number of cycles by which our itc differs from the itc on the master
210  * (time-keeper) CPU.  A positive number indicates our itc is ahead of the master,
211  * negative that it is behind.
212  */
213 static inline long
214 get_delta (long *rt, long *master)
215 {
216 	unsigned long best_t0 = 0, best_t1 = ~0UL, best_tm = 0;
217 	unsigned long tcenter, t0, t1, tm;
218 	long i;
219 
220 	for (i = 0; i < NUM_ITERS; ++i) {
221 		t0 = ia64_get_itc();
222 		go[MASTER] = 1;
223 		while (!(tm = go[SLAVE]))
224 			cpu_relax();
225 		go[SLAVE] = 0;
226 		t1 = ia64_get_itc();
227 
228 		if (t1 - t0 < best_t1 - best_t0)
229 			best_t0 = t0, best_t1 = t1, best_tm = tm;
230 	}
231 
232 	*rt = best_t1 - best_t0;
233 	*master = best_tm - best_t0;
234 
235 	/* average best_t0 and best_t1 without overflow: */
236 	tcenter = (best_t0/2 + best_t1/2);
237 	if (best_t0 % 2 + best_t1 % 2 == 2)
238 		++tcenter;
239 	return tcenter - best_tm;
240 }
241 
242 /*
243  * Synchronize ar.itc of the current (slave) CPU with the ar.itc of the MASTER CPU
244  * (normally the time-keeper CPU).  We use a closed loop to eliminate the possibility of
245  * unaccounted-for errors (such as getting a machine check in the middle of a calibration
246  * step).  The basic idea is for the slave to ask the master what itc value it has and to
247  * read its own itc before and after the master responds.  Each iteration gives us three
248  * timestamps:
249  *
250  *	slave		master
251  *
252  *	t0 ---\
253  *             ---\
254  *		   --->
255  *			tm
256  *		   /---
257  *	       /---
258  *	t1 <---
259  *
260  *
261  * The goal is to adjust the slave's ar.itc such that tm falls exactly half-way between t0
262  * and t1.  If we achieve this, the clocks are synchronized provided the interconnect
263  * between the slave and the master is symmetric.  Even if the interconnect were
264  * asymmetric, we would still know that the synchronization error is smaller than the
265  * roundtrip latency (t0 - t1).
266  *
267  * When the interconnect is quiet and symmetric, this lets us synchronize the itc to
268  * within one or two cycles.  However, we can only *guarantee* that the synchronization is
269  * accurate to within a round-trip time, which is typically in the range of several
270  * hundred cycles (e.g., ~500 cycles).  In practice, this means that the itc's are usually
271  * almost perfectly synchronized, but we shouldn't assume that the accuracy is much better
272  * than half a micro second or so.
273  */
274 void
275 ia64_sync_itc (unsigned int master)
276 {
277 	long i, delta, adj, adjust_latency = 0, done = 0;
278 	unsigned long flags, rt, master_time_stamp, bound;
279 #if DEBUG_ITC_SYNC
280 	struct {
281 		long rt;	/* roundtrip time */
282 		long master;	/* master's timestamp */
283 		long diff;	/* difference between midpoint and master's timestamp */
284 		long lat;	/* estimate of itc adjustment latency */
285 	} t[NUM_ROUNDS];
286 #endif
287 
288 	/*
289 	 * Make sure local timer ticks are disabled while we sync.  If
290 	 * they were enabled, we'd have to worry about nasty issues
291 	 * like setting the ITC ahead of (or a long time before) the
292 	 * next scheduled tick.
293 	 */
294 	BUG_ON((ia64_get_itv() & (1 << 16)) == 0);
295 
296 	go[MASTER] = 1;
297 
298 	if (smp_call_function_single(master, sync_master, NULL, 0) < 0) {
299 		printk(KERN_ERR "sync_itc: failed to get attention of CPU %u!\n", master);
300 		return;
301 	}
302 
303 	while (go[MASTER])
304 		cpu_relax();	/* wait for master to be ready */
305 
306 	spin_lock_irqsave(&itc_sync_lock, flags);
307 	{
308 		for (i = 0; i < NUM_ROUNDS; ++i) {
309 			delta = get_delta(&rt, &master_time_stamp);
310 			if (delta == 0) {
311 				done = 1;	/* let's lock on to this... */
312 				bound = rt;
313 			}
314 
315 			if (!done) {
316 				if (i > 0) {
317 					adjust_latency += -delta;
318 					adj = -delta + adjust_latency/4;
319 				} else
320 					adj = -delta;
321 
322 				ia64_set_itc(ia64_get_itc() + adj);
323 			}
324 #if DEBUG_ITC_SYNC
325 			t[i].rt = rt;
326 			t[i].master = master_time_stamp;
327 			t[i].diff = delta;
328 			t[i].lat = adjust_latency/4;
329 #endif
330 		}
331 	}
332 	spin_unlock_irqrestore(&itc_sync_lock, flags);
333 
334 #if DEBUG_ITC_SYNC
335 	for (i = 0; i < NUM_ROUNDS; ++i)
336 		printk("rt=%5ld master=%5ld diff=%5ld adjlat=%5ld\n",
337 		       t[i].rt, t[i].master, t[i].diff, t[i].lat);
338 #endif
339 
340 	printk(KERN_INFO "CPU %d: synchronized ITC with CPU %u (last diff %ld cycles, "
341 	       "maxerr %lu cycles)\n", smp_processor_id(), master, delta, rt);
342 }
343 
344 /*
345  * Ideally sets up per-cpu profiling hooks.  Doesn't do much now...
346  */
347 static inline void smp_setup_percpu_timer(void)
348 {
349 }
350 
351 static void
352 smp_callin (void)
353 {
354 	int cpuid, phys_id, itc_master;
355 	struct cpuinfo_ia64 *last_cpuinfo, *this_cpuinfo;
356 	extern void ia64_init_itm(void);
357 	extern volatile int time_keeper_id;
358 
359 	cpuid = smp_processor_id();
360 	phys_id = hard_smp_processor_id();
361 	itc_master = time_keeper_id;
362 
363 	if (cpu_online(cpuid)) {
364 		printk(KERN_ERR "huh, phys CPU#0x%x, CPU#0x%x already present??\n",
365 		       phys_id, cpuid);
366 		BUG();
367 	}
368 
369 	fix_b0_for_bsp();
370 
371 	/*
372 	 * numa_node_id() works after this.
373 	 */
374 	set_numa_node(cpu_to_node_map[cpuid]);
375 	set_numa_mem(local_memory_node(cpu_to_node_map[cpuid]));
376 
377 	spin_lock(&vector_lock);
378 	/* Setup the per cpu irq handling data structures */
379 	__setup_vector_irq(cpuid);
380 	notify_cpu_starting(cpuid);
381 	set_cpu_online(cpuid, true);
382 	per_cpu(cpu_state, cpuid) = CPU_ONLINE;
383 	spin_unlock(&vector_lock);
384 
385 	smp_setup_percpu_timer();
386 
387 	ia64_mca_cmc_vector_setup();	/* Setup vector on AP */
388 
389 	local_irq_enable();
390 
391 	if (!(sal_platform_features & IA64_SAL_PLATFORM_FEATURE_ITC_DRIFT)) {
392 		/*
393 		 * Synchronize the ITC with the BP.  Need to do this after irqs are
394 		 * enabled because ia64_sync_itc() calls smp_call_function_single(), which
395 		 * calls spin_unlock_bh(), which calls spin_unlock_bh(), which calls
396 		 * local_bh_enable(), which bugs out if irqs are not enabled...
397 		 */
398 		Dprintk("Going to syncup ITC with ITC Master.\n");
399 		ia64_sync_itc(itc_master);
400 	}
401 
402 	/*
403 	 * Get our bogomips.
404 	 */
405 	ia64_init_itm();
406 
407 	/*
408 	 * Delay calibration can be skipped if new processor is identical to the
409 	 * previous processor.
410 	 */
411 	last_cpuinfo = cpu_data(cpuid - 1);
412 	this_cpuinfo = local_cpu_data;
413 	if (last_cpuinfo->itc_freq != this_cpuinfo->itc_freq ||
414 	    last_cpuinfo->proc_freq != this_cpuinfo->proc_freq ||
415 	    last_cpuinfo->features != this_cpuinfo->features ||
416 	    last_cpuinfo->revision != this_cpuinfo->revision ||
417 	    last_cpuinfo->family != this_cpuinfo->family ||
418 	    last_cpuinfo->archrev != this_cpuinfo->archrev ||
419 	    last_cpuinfo->model != this_cpuinfo->model)
420 		calibrate_delay();
421 	local_cpu_data->loops_per_jiffy = loops_per_jiffy;
422 
423 	/*
424 	 * Allow the master to continue.
425 	 */
426 	cpumask_set_cpu(cpuid, &cpu_callin_map);
427 	Dprintk("Stack on CPU %d at about %p\n",cpuid, &cpuid);
428 }
429 
430 
431 /*
432  * Activate a secondary processor.  head.S calls this.
433  */
434 int
435 start_secondary (void *unused)
436 {
437 	/* Early console may use I/O ports */
438 	ia64_set_kr(IA64_KR_IO_BASE, __pa(ia64_iobase));
439 #ifndef CONFIG_PRINTK_TIME
440 	Dprintk("start_secondary: starting CPU 0x%x\n", hard_smp_processor_id());
441 #endif
442 	efi_map_pal_code();
443 	cpu_init();
444 	preempt_disable();
445 	smp_callin();
446 
447 	cpu_startup_entry(CPUHP_AP_ONLINE_IDLE);
448 	return 0;
449 }
450 
451 static int
452 do_boot_cpu (int sapicid, int cpu, struct task_struct *idle)
453 {
454 	int timeout;
455 
456 	task_for_booting_cpu = idle;
457 	Dprintk("Sending wakeup vector %lu to AP 0x%x/0x%x.\n", ap_wakeup_vector, cpu, sapicid);
458 
459 	set_brendez_area(cpu);
460 	ia64_send_ipi(cpu, ap_wakeup_vector, IA64_IPI_DM_INT, 0);
461 
462 	/*
463 	 * Wait 10s total for the AP to start
464 	 */
465 	Dprintk("Waiting on callin_map ...");
466 	for (timeout = 0; timeout < 100000; timeout++) {
467 		if (cpumask_test_cpu(cpu, &cpu_callin_map))
468 			break;  /* It has booted */
469 		barrier(); /* Make sure we re-read cpu_callin_map */
470 		udelay(100);
471 	}
472 	Dprintk("\n");
473 
474 	if (!cpumask_test_cpu(cpu, &cpu_callin_map)) {
475 		printk(KERN_ERR "Processor 0x%x/0x%x is stuck.\n", cpu, sapicid);
476 		ia64_cpu_to_sapicid[cpu] = -1;
477 		set_cpu_online(cpu, false);  /* was set in smp_callin() */
478 		return -EINVAL;
479 	}
480 	return 0;
481 }
482 
483 static int __init
484 decay (char *str)
485 {
486 	int ticks;
487 	get_option (&str, &ticks);
488 	return 1;
489 }
490 
491 __setup("decay=", decay);
492 
493 /*
494  * Initialize the logical CPU number to SAPICID mapping
495  */
496 void __init
497 smp_build_cpu_map (void)
498 {
499 	int sapicid, cpu, i;
500 	int boot_cpu_id = hard_smp_processor_id();
501 
502 	for (cpu = 0; cpu < NR_CPUS; cpu++) {
503 		ia64_cpu_to_sapicid[cpu] = -1;
504 	}
505 
506 	ia64_cpu_to_sapicid[0] = boot_cpu_id;
507 	init_cpu_present(cpumask_of(0));
508 	set_cpu_possible(0, true);
509 	for (cpu = 1, i = 0; i < smp_boot_data.cpu_count; i++) {
510 		sapicid = smp_boot_data.cpu_phys_id[i];
511 		if (sapicid == boot_cpu_id)
512 			continue;
513 		set_cpu_present(cpu, true);
514 		set_cpu_possible(cpu, true);
515 		ia64_cpu_to_sapicid[cpu] = sapicid;
516 		cpu++;
517 	}
518 }
519 
520 /*
521  * Cycle through the APs sending Wakeup IPIs to boot each.
522  */
523 void __init
524 smp_prepare_cpus (unsigned int max_cpus)
525 {
526 	int boot_cpu_id = hard_smp_processor_id();
527 
528 	/*
529 	 * Initialize the per-CPU profiling counter/multiplier
530 	 */
531 
532 	smp_setup_percpu_timer();
533 
534 	cpumask_set_cpu(0, &cpu_callin_map);
535 
536 	local_cpu_data->loops_per_jiffy = loops_per_jiffy;
537 	ia64_cpu_to_sapicid[0] = boot_cpu_id;
538 
539 	printk(KERN_INFO "Boot processor id 0x%x/0x%x\n", 0, boot_cpu_id);
540 
541 	current_thread_info()->cpu = 0;
542 
543 	/*
544 	 * If SMP should be disabled, then really disable it!
545 	 */
546 	if (!max_cpus) {
547 		printk(KERN_INFO "SMP mode deactivated.\n");
548 		init_cpu_online(cpumask_of(0));
549 		init_cpu_present(cpumask_of(0));
550 		init_cpu_possible(cpumask_of(0));
551 		return;
552 	}
553 }
554 
555 void smp_prepare_boot_cpu(void)
556 {
557 	set_cpu_online(smp_processor_id(), true);
558 	cpumask_set_cpu(smp_processor_id(), &cpu_callin_map);
559 	set_numa_node(cpu_to_node_map[smp_processor_id()]);
560 	per_cpu(cpu_state, smp_processor_id()) = CPU_ONLINE;
561 }
562 
563 #ifdef CONFIG_HOTPLUG_CPU
564 static inline void
565 clear_cpu_sibling_map(int cpu)
566 {
567 	int i;
568 
569 	for_each_cpu(i, &per_cpu(cpu_sibling_map, cpu))
570 		cpumask_clear_cpu(cpu, &per_cpu(cpu_sibling_map, i));
571 	for_each_cpu(i, &cpu_core_map[cpu])
572 		cpumask_clear_cpu(cpu, &cpu_core_map[i]);
573 
574 	per_cpu(cpu_sibling_map, cpu) = cpu_core_map[cpu] = CPU_MASK_NONE;
575 }
576 
577 static void
578 remove_siblinginfo(int cpu)
579 {
580 	int last = 0;
581 
582 	if (cpu_data(cpu)->threads_per_core == 1 &&
583 	    cpu_data(cpu)->cores_per_socket == 1) {
584 		cpumask_clear_cpu(cpu, &cpu_core_map[cpu]);
585 		cpumask_clear_cpu(cpu, &per_cpu(cpu_sibling_map, cpu));
586 		return;
587 	}
588 
589 	last = (cpumask_weight(&cpu_core_map[cpu]) == 1 ? 1 : 0);
590 
591 	/* remove it from all sibling map's */
592 	clear_cpu_sibling_map(cpu);
593 }
594 
595 extern void fixup_irqs(void);
596 
597 int migrate_platform_irqs(unsigned int cpu)
598 {
599 	int new_cpei_cpu;
600 	struct irq_data *data = NULL;
601 	const struct cpumask *mask;
602 	int 		retval = 0;
603 
604 	/*
605 	 * dont permit CPEI target to removed.
606 	 */
607 	if (cpe_vector > 0 && is_cpu_cpei_target(cpu)) {
608 		printk ("CPU (%d) is CPEI Target\n", cpu);
609 		if (can_cpei_retarget()) {
610 			/*
611 			 * Now re-target the CPEI to a different processor
612 			 */
613 			new_cpei_cpu = cpumask_any(cpu_online_mask);
614 			mask = cpumask_of(new_cpei_cpu);
615 			set_cpei_target_cpu(new_cpei_cpu);
616 			data = irq_get_irq_data(ia64_cpe_irq);
617 			/*
618 			 * Switch for now, immediately, we need to do fake intr
619 			 * as other interrupts, but need to study CPEI behaviour with
620 			 * polling before making changes.
621 			 */
622 			if (data && data->chip) {
623 				data->chip->irq_disable(data);
624 				data->chip->irq_set_affinity(data, mask, false);
625 				data->chip->irq_enable(data);
626 				printk ("Re-targeting CPEI to cpu %d\n", new_cpei_cpu);
627 			}
628 		}
629 		if (!data) {
630 			printk ("Unable to retarget CPEI, offline cpu [%d] failed\n", cpu);
631 			retval = -EBUSY;
632 		}
633 	}
634 	return retval;
635 }
636 
637 /* must be called with cpucontrol mutex held */
638 int __cpu_disable(void)
639 {
640 	int cpu = smp_processor_id();
641 
642 	/*
643 	 * dont permit boot processor for now
644 	 */
645 	if (cpu == 0 && !bsp_remove_ok) {
646 		printk ("Your platform does not support removal of BSP\n");
647 		return (-EBUSY);
648 	}
649 
650 	set_cpu_online(cpu, false);
651 
652 	if (migrate_platform_irqs(cpu)) {
653 		set_cpu_online(cpu, true);
654 		return -EBUSY;
655 	}
656 
657 	remove_siblinginfo(cpu);
658 	fixup_irqs();
659 	local_flush_tlb_all();
660 	cpumask_clear_cpu(cpu, &cpu_callin_map);
661 	return 0;
662 }
663 
664 void __cpu_die(unsigned int cpu)
665 {
666 	unsigned int i;
667 
668 	for (i = 0; i < 100; i++) {
669 		/* They ack this in play_dead by setting CPU_DEAD */
670 		if (per_cpu(cpu_state, cpu) == CPU_DEAD)
671 		{
672 			printk ("CPU %d is now offline\n", cpu);
673 			return;
674 		}
675 		msleep(100);
676 	}
677  	printk(KERN_ERR "CPU %u didn't die...\n", cpu);
678 }
679 #endif /* CONFIG_HOTPLUG_CPU */
680 
681 void
682 smp_cpus_done (unsigned int dummy)
683 {
684 	int cpu;
685 	unsigned long bogosum = 0;
686 
687 	/*
688 	 * Allow the user to impress friends.
689 	 */
690 
691 	for_each_online_cpu(cpu) {
692 		bogosum += cpu_data(cpu)->loops_per_jiffy;
693 	}
694 
695 	printk(KERN_INFO "Total of %d processors activated (%lu.%02lu BogoMIPS).\n",
696 	       (int)num_online_cpus(), bogosum/(500000/HZ), (bogosum/(5000/HZ))%100);
697 }
698 
699 static inline void set_cpu_sibling_map(int cpu)
700 {
701 	int i;
702 
703 	for_each_online_cpu(i) {
704 		if ((cpu_data(cpu)->socket_id == cpu_data(i)->socket_id)) {
705 			cpumask_set_cpu(i, &cpu_core_map[cpu]);
706 			cpumask_set_cpu(cpu, &cpu_core_map[i]);
707 			if (cpu_data(cpu)->core_id == cpu_data(i)->core_id) {
708 				cpumask_set_cpu(i,
709 						&per_cpu(cpu_sibling_map, cpu));
710 				cpumask_set_cpu(cpu,
711 						&per_cpu(cpu_sibling_map, i));
712 			}
713 		}
714 	}
715 }
716 
717 int
718 __cpu_up(unsigned int cpu, struct task_struct *tidle)
719 {
720 	int ret;
721 	int sapicid;
722 
723 	sapicid = ia64_cpu_to_sapicid[cpu];
724 	if (sapicid == -1)
725 		return -EINVAL;
726 
727 	/*
728 	 * Already booted cpu? not valid anymore since we dont
729 	 * do idle loop tightspin anymore.
730 	 */
731 	if (cpumask_test_cpu(cpu, &cpu_callin_map))
732 		return -EINVAL;
733 
734 	per_cpu(cpu_state, cpu) = CPU_UP_PREPARE;
735 	/* Processor goes to start_secondary(), sets online flag */
736 	ret = do_boot_cpu(sapicid, cpu, tidle);
737 	if (ret < 0)
738 		return ret;
739 
740 	if (cpu_data(cpu)->threads_per_core == 1 &&
741 	    cpu_data(cpu)->cores_per_socket == 1) {
742 		cpumask_set_cpu(cpu, &per_cpu(cpu_sibling_map, cpu));
743 		cpumask_set_cpu(cpu, &cpu_core_map[cpu]);
744 		return 0;
745 	}
746 
747 	set_cpu_sibling_map(cpu);
748 
749 	return 0;
750 }
751 
752 /*
753  * Assume that CPUs have been discovered by some platform-dependent interface.  For
754  * SoftSDV/Lion, that would be ACPI.
755  *
756  * Setup of the IPI irq handler is done in irq.c:init_IRQ_SMP().
757  */
758 void __init
759 init_smp_config(void)
760 {
761 	struct fptr {
762 		unsigned long fp;
763 		unsigned long gp;
764 	} *ap_startup;
765 	long sal_ret;
766 
767 	/* Tell SAL where to drop the APs.  */
768 	ap_startup = (struct fptr *) start_ap;
769 	sal_ret = ia64_sal_set_vectors(SAL_VECTOR_OS_BOOT_RENDEZ,
770 				       ia64_tpa(ap_startup->fp), ia64_tpa(ap_startup->gp), 0, 0, 0, 0);
771 	if (sal_ret < 0)
772 		printk(KERN_ERR "SMP: Can't set SAL AP Boot Rendezvous: %s\n",
773 		       ia64_sal_strerror(sal_ret));
774 }
775 
776 /*
777  * identify_siblings(cpu) gets called from identify_cpu. This populates the
778  * information related to logical execution units in per_cpu_data structure.
779  */
780 void identify_siblings(struct cpuinfo_ia64 *c)
781 {
782 	long status;
783 	u16 pltid;
784 	pal_logical_to_physical_t info;
785 
786 	status = ia64_pal_logical_to_phys(-1, &info);
787 	if (status != PAL_STATUS_SUCCESS) {
788 		if (status != PAL_STATUS_UNIMPLEMENTED) {
789 			printk(KERN_ERR
790 				"ia64_pal_logical_to_phys failed with %ld\n",
791 				status);
792 			return;
793 		}
794 
795 		info.overview_ppid = 0;
796 		info.overview_cpp  = 1;
797 		info.overview_tpc  = 1;
798 	}
799 
800 	status = ia64_sal_physical_id_info(&pltid);
801 	if (status != PAL_STATUS_SUCCESS) {
802 		if (status != PAL_STATUS_UNIMPLEMENTED)
803 			printk(KERN_ERR
804 				"ia64_sal_pltid failed with %ld\n",
805 				status);
806 		return;
807 	}
808 
809 	c->socket_id =  (pltid << 8) | info.overview_ppid;
810 
811 	if (info.overview_cpp == 1 && info.overview_tpc == 1)
812 		return;
813 
814 	c->cores_per_socket = info.overview_cpp;
815 	c->threads_per_core = info.overview_tpc;
816 	c->num_log = info.overview_num_log;
817 
818 	c->core_id = info.log1_cid;
819 	c->thread_id = info.log1_tid;
820 }
821 
822 /*
823  * returns non zero, if multi-threading is enabled
824  * on at least one physical package. Due to hotplug cpu
825  * and (maxcpus=), all threads may not necessarily be enabled
826  * even though the processor supports multi-threading.
827  */
828 int is_multithreading_enabled(void)
829 {
830 	int i, j;
831 
832 	for_each_present_cpu(i) {
833 		for_each_present_cpu(j) {
834 			if (j == i)
835 				continue;
836 			if ((cpu_data(j)->socket_id == cpu_data(i)->socket_id)) {
837 				if (cpu_data(j)->core_id == cpu_data(i)->core_id)
838 					return 1;
839 			}
840 		}
841 	}
842 	return 0;
843 }
844 EXPORT_SYMBOL_GPL(is_multithreading_enabled);
845