xref: /openbmc/linux/arch/ia64/kernel/smpboot.c (revision 8c0b9ee8)
1 /*
2  * SMP boot-related support
3  *
4  * Copyright (C) 1998-2003, 2005 Hewlett-Packard Co
5  *	David Mosberger-Tang <davidm@hpl.hp.com>
6  * Copyright (C) 2001, 2004-2005 Intel Corp
7  * 	Rohit Seth <rohit.seth@intel.com>
8  * 	Suresh Siddha <suresh.b.siddha@intel.com>
9  * 	Gordon Jin <gordon.jin@intel.com>
10  *	Ashok Raj  <ashok.raj@intel.com>
11  *
12  * 01/05/16 Rohit Seth <rohit.seth@intel.com>	Moved SMP booting functions from smp.c to here.
13  * 01/04/27 David Mosberger <davidm@hpl.hp.com>	Added ITC synching code.
14  * 02/07/31 David Mosberger <davidm@hpl.hp.com>	Switch over to hotplug-CPU boot-sequence.
15  *						smp_boot_cpus()/smp_commence() is replaced by
16  *						smp_prepare_cpus()/__cpu_up()/smp_cpus_done().
17  * 04/06/21 Ashok Raj		<ashok.raj@intel.com> Added CPU Hotplug Support
18  * 04/12/26 Jin Gordon <gordon.jin@intel.com>
19  * 04/12/26 Rohit Seth <rohit.seth@intel.com>
20  *						Add multi-threading and multi-core detection
21  * 05/01/30 Suresh Siddha <suresh.b.siddha@intel.com>
22  *						Setup cpu_sibling_map and cpu_core_map
23  */
24 
25 #include <linux/module.h>
26 #include <linux/acpi.h>
27 #include <linux/bootmem.h>
28 #include <linux/cpu.h>
29 #include <linux/delay.h>
30 #include <linux/init.h>
31 #include <linux/interrupt.h>
32 #include <linux/irq.h>
33 #include <linux/kernel.h>
34 #include <linux/kernel_stat.h>
35 #include <linux/mm.h>
36 #include <linux/notifier.h>
37 #include <linux/smp.h>
38 #include <linux/spinlock.h>
39 #include <linux/efi.h>
40 #include <linux/percpu.h>
41 #include <linux/bitops.h>
42 
43 #include <linux/atomic.h>
44 #include <asm/cache.h>
45 #include <asm/current.h>
46 #include <asm/delay.h>
47 #include <asm/io.h>
48 #include <asm/irq.h>
49 #include <asm/machvec.h>
50 #include <asm/mca.h>
51 #include <asm/page.h>
52 #include <asm/paravirt.h>
53 #include <asm/pgalloc.h>
54 #include <asm/pgtable.h>
55 #include <asm/processor.h>
56 #include <asm/ptrace.h>
57 #include <asm/sal.h>
58 #include <asm/tlbflush.h>
59 #include <asm/unistd.h>
60 #include <asm/sn/arch.h>
61 
62 #define SMP_DEBUG 0
63 
64 #if SMP_DEBUG
65 #define Dprintk(x...)  printk(x)
66 #else
67 #define Dprintk(x...)
68 #endif
69 
70 #ifdef CONFIG_HOTPLUG_CPU
71 #ifdef CONFIG_PERMIT_BSP_REMOVE
72 #define bsp_remove_ok	1
73 #else
74 #define bsp_remove_ok	0
75 #endif
76 
77 /*
78  * Global array allocated for NR_CPUS at boot time
79  */
80 struct sal_to_os_boot sal_boot_rendez_state[NR_CPUS];
81 
82 /*
83  * start_ap in head.S uses this to store current booting cpu
84  * info.
85  */
86 struct sal_to_os_boot *sal_state_for_booting_cpu = &sal_boot_rendez_state[0];
87 
88 #define set_brendez_area(x) (sal_state_for_booting_cpu = &sal_boot_rendez_state[(x)]);
89 
90 #else
91 #define set_brendez_area(x)
92 #endif
93 
94 
95 /*
96  * ITC synchronization related stuff:
97  */
98 #define MASTER	(0)
99 #define SLAVE	(SMP_CACHE_BYTES/8)
100 
101 #define NUM_ROUNDS	64	/* magic value */
102 #define NUM_ITERS	5	/* likewise */
103 
104 static DEFINE_SPINLOCK(itc_sync_lock);
105 static volatile unsigned long go[SLAVE + 1];
106 
107 #define DEBUG_ITC_SYNC	0
108 
109 extern void start_ap (void);
110 extern unsigned long ia64_iobase;
111 
112 struct task_struct *task_for_booting_cpu;
113 
114 /*
115  * State for each CPU
116  */
117 DEFINE_PER_CPU(int, cpu_state);
118 
119 cpumask_t cpu_core_map[NR_CPUS] __cacheline_aligned;
120 EXPORT_SYMBOL(cpu_core_map);
121 DEFINE_PER_CPU_SHARED_ALIGNED(cpumask_t, cpu_sibling_map);
122 EXPORT_PER_CPU_SYMBOL(cpu_sibling_map);
123 
124 int smp_num_siblings = 1;
125 
126 /* which logical CPU number maps to which CPU (physical APIC ID) */
127 volatile int ia64_cpu_to_sapicid[NR_CPUS];
128 EXPORT_SYMBOL(ia64_cpu_to_sapicid);
129 
130 static volatile cpumask_t cpu_callin_map;
131 
132 struct smp_boot_data smp_boot_data __initdata;
133 
134 unsigned long ap_wakeup_vector = -1; /* External Int use to wakeup APs */
135 
136 char __initdata no_int_routing;
137 
138 unsigned char smp_int_redirect; /* are INT and IPI redirectable by the chipset? */
139 
140 #ifdef CONFIG_FORCE_CPEI_RETARGET
141 #define CPEI_OVERRIDE_DEFAULT	(1)
142 #else
143 #define CPEI_OVERRIDE_DEFAULT	(0)
144 #endif
145 
146 unsigned int force_cpei_retarget = CPEI_OVERRIDE_DEFAULT;
147 
148 static int __init
149 cmdl_force_cpei(char *str)
150 {
151 	int value=0;
152 
153 	get_option (&str, &value);
154 	force_cpei_retarget = value;
155 
156 	return 1;
157 }
158 
159 __setup("force_cpei=", cmdl_force_cpei);
160 
161 static int __init
162 nointroute (char *str)
163 {
164 	no_int_routing = 1;
165 	printk ("no_int_routing on\n");
166 	return 1;
167 }
168 
169 __setup("nointroute", nointroute);
170 
171 static void fix_b0_for_bsp(void)
172 {
173 #ifdef CONFIG_HOTPLUG_CPU
174 	int cpuid;
175 	static int fix_bsp_b0 = 1;
176 
177 	cpuid = smp_processor_id();
178 
179 	/*
180 	 * Cache the b0 value on the first AP that comes up
181 	 */
182 	if (!(fix_bsp_b0 && cpuid))
183 		return;
184 
185 	sal_boot_rendez_state[0].br[0] = sal_boot_rendez_state[cpuid].br[0];
186 	printk ("Fixed BSP b0 value from CPU %d\n", cpuid);
187 
188 	fix_bsp_b0 = 0;
189 #endif
190 }
191 
192 void
193 sync_master (void *arg)
194 {
195 	unsigned long flags, i;
196 
197 	go[MASTER] = 0;
198 
199 	local_irq_save(flags);
200 	{
201 		for (i = 0; i < NUM_ROUNDS*NUM_ITERS; ++i) {
202 			while (!go[MASTER])
203 				cpu_relax();
204 			go[MASTER] = 0;
205 			go[SLAVE] = ia64_get_itc();
206 		}
207 	}
208 	local_irq_restore(flags);
209 }
210 
211 /*
212  * Return the number of cycles by which our itc differs from the itc on the master
213  * (time-keeper) CPU.  A positive number indicates our itc is ahead of the master,
214  * negative that it is behind.
215  */
216 static inline long
217 get_delta (long *rt, long *master)
218 {
219 	unsigned long best_t0 = 0, best_t1 = ~0UL, best_tm = 0;
220 	unsigned long tcenter, t0, t1, tm;
221 	long i;
222 
223 	for (i = 0; i < NUM_ITERS; ++i) {
224 		t0 = ia64_get_itc();
225 		go[MASTER] = 1;
226 		while (!(tm = go[SLAVE]))
227 			cpu_relax();
228 		go[SLAVE] = 0;
229 		t1 = ia64_get_itc();
230 
231 		if (t1 - t0 < best_t1 - best_t0)
232 			best_t0 = t0, best_t1 = t1, best_tm = tm;
233 	}
234 
235 	*rt = best_t1 - best_t0;
236 	*master = best_tm - best_t0;
237 
238 	/* average best_t0 and best_t1 without overflow: */
239 	tcenter = (best_t0/2 + best_t1/2);
240 	if (best_t0 % 2 + best_t1 % 2 == 2)
241 		++tcenter;
242 	return tcenter - best_tm;
243 }
244 
245 /*
246  * Synchronize ar.itc of the current (slave) CPU with the ar.itc of the MASTER CPU
247  * (normally the time-keeper CPU).  We use a closed loop to eliminate the possibility of
248  * unaccounted-for errors (such as getting a machine check in the middle of a calibration
249  * step).  The basic idea is for the slave to ask the master what itc value it has and to
250  * read its own itc before and after the master responds.  Each iteration gives us three
251  * timestamps:
252  *
253  *	slave		master
254  *
255  *	t0 ---\
256  *             ---\
257  *		   --->
258  *			tm
259  *		   /---
260  *	       /---
261  *	t1 <---
262  *
263  *
264  * The goal is to adjust the slave's ar.itc such that tm falls exactly half-way between t0
265  * and t1.  If we achieve this, the clocks are synchronized provided the interconnect
266  * between the slave and the master is symmetric.  Even if the interconnect were
267  * asymmetric, we would still know that the synchronization error is smaller than the
268  * roundtrip latency (t0 - t1).
269  *
270  * When the interconnect is quiet and symmetric, this lets us synchronize the itc to
271  * within one or two cycles.  However, we can only *guarantee* that the synchronization is
272  * accurate to within a round-trip time, which is typically in the range of several
273  * hundred cycles (e.g., ~500 cycles).  In practice, this means that the itc's are usually
274  * almost perfectly synchronized, but we shouldn't assume that the accuracy is much better
275  * than half a micro second or so.
276  */
277 void
278 ia64_sync_itc (unsigned int master)
279 {
280 	long i, delta, adj, adjust_latency = 0, done = 0;
281 	unsigned long flags, rt, master_time_stamp, bound;
282 #if DEBUG_ITC_SYNC
283 	struct {
284 		long rt;	/* roundtrip time */
285 		long master;	/* master's timestamp */
286 		long diff;	/* difference between midpoint and master's timestamp */
287 		long lat;	/* estimate of itc adjustment latency */
288 	} t[NUM_ROUNDS];
289 #endif
290 
291 	/*
292 	 * Make sure local timer ticks are disabled while we sync.  If
293 	 * they were enabled, we'd have to worry about nasty issues
294 	 * like setting the ITC ahead of (or a long time before) the
295 	 * next scheduled tick.
296 	 */
297 	BUG_ON((ia64_get_itv() & (1 << 16)) == 0);
298 
299 	go[MASTER] = 1;
300 
301 	if (smp_call_function_single(master, sync_master, NULL, 0) < 0) {
302 		printk(KERN_ERR "sync_itc: failed to get attention of CPU %u!\n", master);
303 		return;
304 	}
305 
306 	while (go[MASTER])
307 		cpu_relax();	/* wait for master to be ready */
308 
309 	spin_lock_irqsave(&itc_sync_lock, flags);
310 	{
311 		for (i = 0; i < NUM_ROUNDS; ++i) {
312 			delta = get_delta(&rt, &master_time_stamp);
313 			if (delta == 0) {
314 				done = 1;	/* let's lock on to this... */
315 				bound = rt;
316 			}
317 
318 			if (!done) {
319 				if (i > 0) {
320 					adjust_latency += -delta;
321 					adj = -delta + adjust_latency/4;
322 				} else
323 					adj = -delta;
324 
325 				ia64_set_itc(ia64_get_itc() + adj);
326 			}
327 #if DEBUG_ITC_SYNC
328 			t[i].rt = rt;
329 			t[i].master = master_time_stamp;
330 			t[i].diff = delta;
331 			t[i].lat = adjust_latency/4;
332 #endif
333 		}
334 	}
335 	spin_unlock_irqrestore(&itc_sync_lock, flags);
336 
337 #if DEBUG_ITC_SYNC
338 	for (i = 0; i < NUM_ROUNDS; ++i)
339 		printk("rt=%5ld master=%5ld diff=%5ld adjlat=%5ld\n",
340 		       t[i].rt, t[i].master, t[i].diff, t[i].lat);
341 #endif
342 
343 	printk(KERN_INFO "CPU %d: synchronized ITC with CPU %u (last diff %ld cycles, "
344 	       "maxerr %lu cycles)\n", smp_processor_id(), master, delta, rt);
345 }
346 
347 /*
348  * Ideally sets up per-cpu profiling hooks.  Doesn't do much now...
349  */
350 static inline void smp_setup_percpu_timer(void)
351 {
352 }
353 
354 static void
355 smp_callin (void)
356 {
357 	int cpuid, phys_id, itc_master;
358 	struct cpuinfo_ia64 *last_cpuinfo, *this_cpuinfo;
359 	extern void ia64_init_itm(void);
360 	extern volatile int time_keeper_id;
361 
362 #ifdef CONFIG_PERFMON
363 	extern void pfm_init_percpu(void);
364 #endif
365 
366 	cpuid = smp_processor_id();
367 	phys_id = hard_smp_processor_id();
368 	itc_master = time_keeper_id;
369 
370 	if (cpu_online(cpuid)) {
371 		printk(KERN_ERR "huh, phys CPU#0x%x, CPU#0x%x already present??\n",
372 		       phys_id, cpuid);
373 		BUG();
374 	}
375 
376 	fix_b0_for_bsp();
377 
378 	/*
379 	 * numa_node_id() works after this.
380 	 */
381 	set_numa_node(cpu_to_node_map[cpuid]);
382 	set_numa_mem(local_memory_node(cpu_to_node_map[cpuid]));
383 
384 	spin_lock(&vector_lock);
385 	/* Setup the per cpu irq handling data structures */
386 	__setup_vector_irq(cpuid);
387 	notify_cpu_starting(cpuid);
388 	set_cpu_online(cpuid, true);
389 	per_cpu(cpu_state, cpuid) = CPU_ONLINE;
390 	spin_unlock(&vector_lock);
391 
392 	smp_setup_percpu_timer();
393 
394 	ia64_mca_cmc_vector_setup();	/* Setup vector on AP */
395 
396 #ifdef CONFIG_PERFMON
397 	pfm_init_percpu();
398 #endif
399 
400 	local_irq_enable();
401 
402 	if (!(sal_platform_features & IA64_SAL_PLATFORM_FEATURE_ITC_DRIFT)) {
403 		/*
404 		 * Synchronize the ITC with the BP.  Need to do this after irqs are
405 		 * enabled because ia64_sync_itc() calls smp_call_function_single(), which
406 		 * calls spin_unlock_bh(), which calls spin_unlock_bh(), which calls
407 		 * local_bh_enable(), which bugs out if irqs are not enabled...
408 		 */
409 		Dprintk("Going to syncup ITC with ITC Master.\n");
410 		ia64_sync_itc(itc_master);
411 	}
412 
413 	/*
414 	 * Get our bogomips.
415 	 */
416 	ia64_init_itm();
417 
418 	/*
419 	 * Delay calibration can be skipped if new processor is identical to the
420 	 * previous processor.
421 	 */
422 	last_cpuinfo = cpu_data(cpuid - 1);
423 	this_cpuinfo = local_cpu_data;
424 	if (last_cpuinfo->itc_freq != this_cpuinfo->itc_freq ||
425 	    last_cpuinfo->proc_freq != this_cpuinfo->proc_freq ||
426 	    last_cpuinfo->features != this_cpuinfo->features ||
427 	    last_cpuinfo->revision != this_cpuinfo->revision ||
428 	    last_cpuinfo->family != this_cpuinfo->family ||
429 	    last_cpuinfo->archrev != this_cpuinfo->archrev ||
430 	    last_cpuinfo->model != this_cpuinfo->model)
431 		calibrate_delay();
432 	local_cpu_data->loops_per_jiffy = loops_per_jiffy;
433 
434 	/*
435 	 * Allow the master to continue.
436 	 */
437 	cpu_set(cpuid, cpu_callin_map);
438 	Dprintk("Stack on CPU %d at about %p\n",cpuid, &cpuid);
439 }
440 
441 
442 /*
443  * Activate a secondary processor.  head.S calls this.
444  */
445 int
446 start_secondary (void *unused)
447 {
448 	/* Early console may use I/O ports */
449 	ia64_set_kr(IA64_KR_IO_BASE, __pa(ia64_iobase));
450 #ifndef CONFIG_PRINTK_TIME
451 	Dprintk("start_secondary: starting CPU 0x%x\n", hard_smp_processor_id());
452 #endif
453 	efi_map_pal_code();
454 	cpu_init();
455 	preempt_disable();
456 	smp_callin();
457 
458 	cpu_startup_entry(CPUHP_ONLINE);
459 	return 0;
460 }
461 
462 static int
463 do_boot_cpu (int sapicid, int cpu, struct task_struct *idle)
464 {
465 	int timeout;
466 
467 	task_for_booting_cpu = idle;
468 	Dprintk("Sending wakeup vector %lu to AP 0x%x/0x%x.\n", ap_wakeup_vector, cpu, sapicid);
469 
470 	set_brendez_area(cpu);
471 	platform_send_ipi(cpu, ap_wakeup_vector, IA64_IPI_DM_INT, 0);
472 
473 	/*
474 	 * Wait 10s total for the AP to start
475 	 */
476 	Dprintk("Waiting on callin_map ...");
477 	for (timeout = 0; timeout < 100000; timeout++) {
478 		if (cpu_isset(cpu, cpu_callin_map))
479 			break;  /* It has booted */
480 		udelay(100);
481 	}
482 	Dprintk("\n");
483 
484 	if (!cpu_isset(cpu, cpu_callin_map)) {
485 		printk(KERN_ERR "Processor 0x%x/0x%x is stuck.\n", cpu, sapicid);
486 		ia64_cpu_to_sapicid[cpu] = -1;
487 		set_cpu_online(cpu, false);  /* was set in smp_callin() */
488 		return -EINVAL;
489 	}
490 	return 0;
491 }
492 
493 static int __init
494 decay (char *str)
495 {
496 	int ticks;
497 	get_option (&str, &ticks);
498 	return 1;
499 }
500 
501 __setup("decay=", decay);
502 
503 /*
504  * Initialize the logical CPU number to SAPICID mapping
505  */
506 void __init
507 smp_build_cpu_map (void)
508 {
509 	int sapicid, cpu, i;
510 	int boot_cpu_id = hard_smp_processor_id();
511 
512 	for (cpu = 0; cpu < NR_CPUS; cpu++) {
513 		ia64_cpu_to_sapicid[cpu] = -1;
514 	}
515 
516 	ia64_cpu_to_sapicid[0] = boot_cpu_id;
517 	init_cpu_present(cpumask_of(0));
518 	set_cpu_possible(0, true);
519 	for (cpu = 1, i = 0; i < smp_boot_data.cpu_count; i++) {
520 		sapicid = smp_boot_data.cpu_phys_id[i];
521 		if (sapicid == boot_cpu_id)
522 			continue;
523 		set_cpu_present(cpu, true);
524 		set_cpu_possible(cpu, true);
525 		ia64_cpu_to_sapicid[cpu] = sapicid;
526 		cpu++;
527 	}
528 }
529 
530 /*
531  * Cycle through the APs sending Wakeup IPIs to boot each.
532  */
533 void __init
534 smp_prepare_cpus (unsigned int max_cpus)
535 {
536 	int boot_cpu_id = hard_smp_processor_id();
537 
538 	/*
539 	 * Initialize the per-CPU profiling counter/multiplier
540 	 */
541 
542 	smp_setup_percpu_timer();
543 
544 	cpu_set(0, cpu_callin_map);
545 
546 	local_cpu_data->loops_per_jiffy = loops_per_jiffy;
547 	ia64_cpu_to_sapicid[0] = boot_cpu_id;
548 
549 	printk(KERN_INFO "Boot processor id 0x%x/0x%x\n", 0, boot_cpu_id);
550 
551 	current_thread_info()->cpu = 0;
552 
553 	/*
554 	 * If SMP should be disabled, then really disable it!
555 	 */
556 	if (!max_cpus) {
557 		printk(KERN_INFO "SMP mode deactivated.\n");
558 		init_cpu_online(cpumask_of(0));
559 		init_cpu_present(cpumask_of(0));
560 		init_cpu_possible(cpumask_of(0));
561 		return;
562 	}
563 }
564 
565 void smp_prepare_boot_cpu(void)
566 {
567 	set_cpu_online(smp_processor_id(), true);
568 	cpu_set(smp_processor_id(), cpu_callin_map);
569 	set_numa_node(cpu_to_node_map[smp_processor_id()]);
570 	per_cpu(cpu_state, smp_processor_id()) = CPU_ONLINE;
571 	paravirt_post_smp_prepare_boot_cpu();
572 }
573 
574 #ifdef CONFIG_HOTPLUG_CPU
575 static inline void
576 clear_cpu_sibling_map(int cpu)
577 {
578 	int i;
579 
580 	for_each_cpu_mask(i, per_cpu(cpu_sibling_map, cpu))
581 		cpu_clear(cpu, per_cpu(cpu_sibling_map, i));
582 	for_each_cpu_mask(i, cpu_core_map[cpu])
583 		cpu_clear(cpu, cpu_core_map[i]);
584 
585 	per_cpu(cpu_sibling_map, cpu) = cpu_core_map[cpu] = CPU_MASK_NONE;
586 }
587 
588 static void
589 remove_siblinginfo(int cpu)
590 {
591 	int last = 0;
592 
593 	if (cpu_data(cpu)->threads_per_core == 1 &&
594 	    cpu_data(cpu)->cores_per_socket == 1) {
595 		cpu_clear(cpu, cpu_core_map[cpu]);
596 		cpu_clear(cpu, per_cpu(cpu_sibling_map, cpu));
597 		return;
598 	}
599 
600 	last = (cpus_weight(cpu_core_map[cpu]) == 1 ? 1 : 0);
601 
602 	/* remove it from all sibling map's */
603 	clear_cpu_sibling_map(cpu);
604 }
605 
606 extern void fixup_irqs(void);
607 
608 int migrate_platform_irqs(unsigned int cpu)
609 {
610 	int new_cpei_cpu;
611 	struct irq_data *data = NULL;
612 	const struct cpumask *mask;
613 	int 		retval = 0;
614 
615 	/*
616 	 * dont permit CPEI target to removed.
617 	 */
618 	if (cpe_vector > 0 && is_cpu_cpei_target(cpu)) {
619 		printk ("CPU (%d) is CPEI Target\n", cpu);
620 		if (can_cpei_retarget()) {
621 			/*
622 			 * Now re-target the CPEI to a different processor
623 			 */
624 			new_cpei_cpu = cpumask_any(cpu_online_mask);
625 			mask = cpumask_of(new_cpei_cpu);
626 			set_cpei_target_cpu(new_cpei_cpu);
627 			data = irq_get_irq_data(ia64_cpe_irq);
628 			/*
629 			 * Switch for now, immediately, we need to do fake intr
630 			 * as other interrupts, but need to study CPEI behaviour with
631 			 * polling before making changes.
632 			 */
633 			if (data && data->chip) {
634 				data->chip->irq_disable(data);
635 				data->chip->irq_set_affinity(data, mask, false);
636 				data->chip->irq_enable(data);
637 				printk ("Re-targeting CPEI to cpu %d\n", new_cpei_cpu);
638 			}
639 		}
640 		if (!data) {
641 			printk ("Unable to retarget CPEI, offline cpu [%d] failed\n", cpu);
642 			retval = -EBUSY;
643 		}
644 	}
645 	return retval;
646 }
647 
648 /* must be called with cpucontrol mutex held */
649 int __cpu_disable(void)
650 {
651 	int cpu = smp_processor_id();
652 
653 	/*
654 	 * dont permit boot processor for now
655 	 */
656 	if (cpu == 0 && !bsp_remove_ok) {
657 		printk ("Your platform does not support removal of BSP\n");
658 		return (-EBUSY);
659 	}
660 
661 	if (ia64_platform_is("sn2")) {
662 		if (!sn_cpu_disable_allowed(cpu))
663 			return -EBUSY;
664 	}
665 
666 	set_cpu_online(cpu, false);
667 
668 	if (migrate_platform_irqs(cpu)) {
669 		set_cpu_online(cpu, true);
670 		return -EBUSY;
671 	}
672 
673 	remove_siblinginfo(cpu);
674 	fixup_irqs();
675 	local_flush_tlb_all();
676 	cpu_clear(cpu, cpu_callin_map);
677 	return 0;
678 }
679 
680 void __cpu_die(unsigned int cpu)
681 {
682 	unsigned int i;
683 
684 	for (i = 0; i < 100; i++) {
685 		/* They ack this in play_dead by setting CPU_DEAD */
686 		if (per_cpu(cpu_state, cpu) == CPU_DEAD)
687 		{
688 			printk ("CPU %d is now offline\n", cpu);
689 			return;
690 		}
691 		msleep(100);
692 	}
693  	printk(KERN_ERR "CPU %u didn't die...\n", cpu);
694 }
695 #endif /* CONFIG_HOTPLUG_CPU */
696 
697 void
698 smp_cpus_done (unsigned int dummy)
699 {
700 	int cpu;
701 	unsigned long bogosum = 0;
702 
703 	/*
704 	 * Allow the user to impress friends.
705 	 */
706 
707 	for_each_online_cpu(cpu) {
708 		bogosum += cpu_data(cpu)->loops_per_jiffy;
709 	}
710 
711 	printk(KERN_INFO "Total of %d processors activated (%lu.%02lu BogoMIPS).\n",
712 	       (int)num_online_cpus(), bogosum/(500000/HZ), (bogosum/(5000/HZ))%100);
713 }
714 
715 static inline void set_cpu_sibling_map(int cpu)
716 {
717 	int i;
718 
719 	for_each_online_cpu(i) {
720 		if ((cpu_data(cpu)->socket_id == cpu_data(i)->socket_id)) {
721 			cpu_set(i, cpu_core_map[cpu]);
722 			cpu_set(cpu, cpu_core_map[i]);
723 			if (cpu_data(cpu)->core_id == cpu_data(i)->core_id) {
724 				cpu_set(i, per_cpu(cpu_sibling_map, cpu));
725 				cpu_set(cpu, per_cpu(cpu_sibling_map, i));
726 			}
727 		}
728 	}
729 }
730 
731 int
732 __cpu_up(unsigned int cpu, struct task_struct *tidle)
733 {
734 	int ret;
735 	int sapicid;
736 
737 	sapicid = ia64_cpu_to_sapicid[cpu];
738 	if (sapicid == -1)
739 		return -EINVAL;
740 
741 	/*
742 	 * Already booted cpu? not valid anymore since we dont
743 	 * do idle loop tightspin anymore.
744 	 */
745 	if (cpu_isset(cpu, cpu_callin_map))
746 		return -EINVAL;
747 
748 	per_cpu(cpu_state, cpu) = CPU_UP_PREPARE;
749 	/* Processor goes to start_secondary(), sets online flag */
750 	ret = do_boot_cpu(sapicid, cpu, tidle);
751 	if (ret < 0)
752 		return ret;
753 
754 	if (cpu_data(cpu)->threads_per_core == 1 &&
755 	    cpu_data(cpu)->cores_per_socket == 1) {
756 		cpu_set(cpu, per_cpu(cpu_sibling_map, cpu));
757 		cpu_set(cpu, cpu_core_map[cpu]);
758 		return 0;
759 	}
760 
761 	set_cpu_sibling_map(cpu);
762 
763 	return 0;
764 }
765 
766 /*
767  * Assume that CPUs have been discovered by some platform-dependent interface.  For
768  * SoftSDV/Lion, that would be ACPI.
769  *
770  * Setup of the IPI irq handler is done in irq.c:init_IRQ_SMP().
771  */
772 void __init
773 init_smp_config(void)
774 {
775 	struct fptr {
776 		unsigned long fp;
777 		unsigned long gp;
778 	} *ap_startup;
779 	long sal_ret;
780 
781 	/* Tell SAL where to drop the APs.  */
782 	ap_startup = (struct fptr *) start_ap;
783 	sal_ret = ia64_sal_set_vectors(SAL_VECTOR_OS_BOOT_RENDEZ,
784 				       ia64_tpa(ap_startup->fp), ia64_tpa(ap_startup->gp), 0, 0, 0, 0);
785 	if (sal_ret < 0)
786 		printk(KERN_ERR "SMP: Can't set SAL AP Boot Rendezvous: %s\n",
787 		       ia64_sal_strerror(sal_ret));
788 }
789 
790 /*
791  * identify_siblings(cpu) gets called from identify_cpu. This populates the
792  * information related to logical execution units in per_cpu_data structure.
793  */
794 void identify_siblings(struct cpuinfo_ia64 *c)
795 {
796 	long status;
797 	u16 pltid;
798 	pal_logical_to_physical_t info;
799 
800 	status = ia64_pal_logical_to_phys(-1, &info);
801 	if (status != PAL_STATUS_SUCCESS) {
802 		if (status != PAL_STATUS_UNIMPLEMENTED) {
803 			printk(KERN_ERR
804 				"ia64_pal_logical_to_phys failed with %ld\n",
805 				status);
806 			return;
807 		}
808 
809 		info.overview_ppid = 0;
810 		info.overview_cpp  = 1;
811 		info.overview_tpc  = 1;
812 	}
813 
814 	status = ia64_sal_physical_id_info(&pltid);
815 	if (status != PAL_STATUS_SUCCESS) {
816 		if (status != PAL_STATUS_UNIMPLEMENTED)
817 			printk(KERN_ERR
818 				"ia64_sal_pltid failed with %ld\n",
819 				status);
820 		return;
821 	}
822 
823 	c->socket_id =  (pltid << 8) | info.overview_ppid;
824 
825 	if (info.overview_cpp == 1 && info.overview_tpc == 1)
826 		return;
827 
828 	c->cores_per_socket = info.overview_cpp;
829 	c->threads_per_core = info.overview_tpc;
830 	c->num_log = info.overview_num_log;
831 
832 	c->core_id = info.log1_cid;
833 	c->thread_id = info.log1_tid;
834 }
835 
836 /*
837  * returns non zero, if multi-threading is enabled
838  * on at least one physical package. Due to hotplug cpu
839  * and (maxcpus=), all threads may not necessarily be enabled
840  * even though the processor supports multi-threading.
841  */
842 int is_multithreading_enabled(void)
843 {
844 	int i, j;
845 
846 	for_each_present_cpu(i) {
847 		for_each_present_cpu(j) {
848 			if (j == i)
849 				continue;
850 			if ((cpu_data(j)->socket_id == cpu_data(i)->socket_id)) {
851 				if (cpu_data(j)->core_id == cpu_data(i)->core_id)
852 					return 1;
853 			}
854 		}
855 	}
856 	return 0;
857 }
858 EXPORT_SYMBOL_GPL(is_multithreading_enabled);
859